diff options
Diffstat (limited to 'pkg/sentry')
538 files changed, 27349 insertions, 42357 deletions
diff --git a/pkg/sentry/BUILD b/pkg/sentry/BUILD deleted file mode 100644 index e759dc36f..000000000 --- a/pkg/sentry/BUILD +++ /dev/null @@ -1,14 +0,0 @@ -package(licenses = ["notice"]) - -# The "internal" package_group should be used as much as possible by packages -# that should remain Sentry-internal (i.e. not be exposed directly to command -# line tooling or APIs). -package_group( - name = "internal", - packages = [ - "//pkg/sentry/...", - "//runsc/...", - # Code generated by go_marshal relies on go_marshal libraries. - "//tools/go_marshal/...", - ], -) diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD deleted file mode 100644 index e27f21e5e..000000000 --- a/pkg/sentry/arch/BUILD +++ /dev/null @@ -1,48 +0,0 @@ -load("//tools:defs.bzl", "go_library", "proto_library") - -package(licenses = ["notice"]) - -go_library( - name = "arch", - srcs = [ - "aligned.go", - "arch.go", - "arch_aarch64.go", - "arch_amd64.go", - "arch_amd64.s", - "arch_arm64.go", - "arch_state_aarch64.go", - "arch_state_x86.go", - "arch_x86.go", - "arch_x86_impl.go", - "auxv.go", - "signal.go", - "signal_act.go", - "signal_amd64.go", - "signal_arm64.go", - "signal_info.go", - "signal_stack.go", - "stack.go", - "syscalls_amd64.go", - "syscalls_arm64.go", - ], - visibility = ["//:sandbox"], - deps = [ - ":registers_go_proto", - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/cpuid", - "//pkg/log", - "//pkg/sentry/limits", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -proto_library( - name = "registers", - srcs = ["registers.proto"], - visibility = ["//visibility:public"], -) diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index c29e1b841..c29e1b841 100644..100755 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go diff --git a/pkg/sentry/arch/arch_aarch64_state_autogen.go b/pkg/sentry/arch/arch_aarch64_state_autogen.go new file mode 100755 index 000000000..9c6dfdf2e --- /dev/null +++ b/pkg/sentry/arch/arch_aarch64_state_autogen.go @@ -0,0 +1,6 @@ +// automatically generated by stateify. + +// +build arm64 +// +build arm64 + +package arch diff --git a/pkg/sentry/arch/arch_amd64_state_autogen.go b/pkg/sentry/arch/arch_amd64_state_autogen.go new file mode 100755 index 000000000..73c523c90 --- /dev/null +++ b/pkg/sentry/arch/arch_amd64_state_autogen.go @@ -0,0 +1,28 @@ +// automatically generated by stateify. + +// +build amd64 +// +build amd64 +// +build amd64 + +package arch + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *context64) beforeSave() {} +func (x *context64) save(m state.Map) { + x.beforeSave() + m.Save("State", &x.State) + m.Save("sigFPState", &x.sigFPState) +} + +func (x *context64) afterLoad() {} +func (x *context64) load(m state.Map) { + m.Load("State", &x.State) + m.Load("sigFPState", &x.sigFPState) +} + +func init() { + state.Register("pkg/sentry/arch.context64", (*context64)(nil), state.Fns{Save: (*context64).save, Load: (*context64).load}) +} diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index db99c5acb..db99c5acb 100644..100755 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go diff --git a/pkg/sentry/arch/arch_arm64_state_autogen.go b/pkg/sentry/arch/arch_arm64_state_autogen.go new file mode 100755 index 000000000..49f2e3d67 --- /dev/null +++ b/pkg/sentry/arch/arch_arm64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build arm64 + +package arch diff --git a/pkg/sentry/arch/arch_impl_state_autogen.go b/pkg/sentry/arch/arch_impl_state_autogen.go new file mode 100755 index 000000000..8b567801f --- /dev/null +++ b/pkg/sentry/arch/arch_impl_state_autogen.go @@ -0,0 +1,29 @@ +// automatically generated by stateify. + +// +build amd64 i386 + +package arch + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *State) beforeSave() {} +func (x *State) save(m state.Map) { + x.beforeSave() + var Regs syscallPtraceRegs = x.saveRegs() + m.SaveValue("Regs", Regs) + m.Save("x86FPState", &x.x86FPState) + m.Save("FeatureSet", &x.FeatureSet) +} + +func (x *State) load(m state.Map) { + m.LoadWait("x86FPState", &x.x86FPState) + m.Load("FeatureSet", &x.FeatureSet) + m.LoadValue("Regs", new(syscallPtraceRegs), func(y interface{}) { x.loadRegs(y.(syscallPtraceRegs)) }) + m.AfterLoad(x.afterLoad) +} + +func init() { + state.Register("pkg/sentry/arch.State", (*State)(nil), state.Fns{Save: (*State).save, Load: (*State).load}) +} diff --git a/pkg/sentry/arch/arch_state_aarch64.go b/pkg/sentry/arch/arch_state_aarch64.go index 0136a85ad..0136a85ad 100644..100755 --- a/pkg/sentry/arch/arch_state_aarch64.go +++ b/pkg/sentry/arch/arch_state_aarch64.go diff --git a/pkg/sentry/arch/arch_state_autogen.go b/pkg/sentry/arch/arch_state_autogen.go new file mode 100755 index 000000000..a06d96c71 --- /dev/null +++ b/pkg/sentry/arch/arch_state_autogen.go @@ -0,0 +1,166 @@ +// automatically generated by stateify. + +// +build amd64 i386 +// +build amd64 i386 +// +build i386 amd64 arm64 + +package arch + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *MmapLayout) beforeSave() {} +func (x *MmapLayout) save(m state.Map) { + x.beforeSave() + m.Save("MinAddr", &x.MinAddr) + m.Save("MaxAddr", &x.MaxAddr) + m.Save("BottomUpBase", &x.BottomUpBase) + m.Save("TopDownBase", &x.TopDownBase) + m.Save("DefaultDirection", &x.DefaultDirection) + m.Save("MaxStackRand", &x.MaxStackRand) +} + +func (x *MmapLayout) afterLoad() {} +func (x *MmapLayout) load(m state.Map) { + m.Load("MinAddr", &x.MinAddr) + m.Load("MaxAddr", &x.MaxAddr) + m.Load("BottomUpBase", &x.BottomUpBase) + m.Load("TopDownBase", &x.TopDownBase) + m.Load("DefaultDirection", &x.DefaultDirection) + m.Load("MaxStackRand", &x.MaxStackRand) +} + +func (x *syscallPtraceRegs) beforeSave() {} +func (x *syscallPtraceRegs) save(m state.Map) { + x.beforeSave() + m.Save("R15", &x.R15) + m.Save("R14", &x.R14) + m.Save("R13", &x.R13) + m.Save("R12", &x.R12) + m.Save("Rbp", &x.Rbp) + m.Save("Rbx", &x.Rbx) + m.Save("R11", &x.R11) + m.Save("R10", &x.R10) + m.Save("R9", &x.R9) + m.Save("R8", &x.R8) + m.Save("Rax", &x.Rax) + m.Save("Rcx", &x.Rcx) + m.Save("Rdx", &x.Rdx) + m.Save("Rsi", &x.Rsi) + m.Save("Rdi", &x.Rdi) + m.Save("Orig_rax", &x.Orig_rax) + m.Save("Rip", &x.Rip) + m.Save("Cs", &x.Cs) + m.Save("Eflags", &x.Eflags) + m.Save("Rsp", &x.Rsp) + m.Save("Ss", &x.Ss) + m.Save("Fs_base", &x.Fs_base) + m.Save("Gs_base", &x.Gs_base) + m.Save("Ds", &x.Ds) + m.Save("Es", &x.Es) + m.Save("Fs", &x.Fs) + m.Save("Gs", &x.Gs) +} + +func (x *syscallPtraceRegs) afterLoad() {} +func (x *syscallPtraceRegs) load(m state.Map) { + m.Load("R15", &x.R15) + m.Load("R14", &x.R14) + m.Load("R13", &x.R13) + m.Load("R12", &x.R12) + m.Load("Rbp", &x.Rbp) + m.Load("Rbx", &x.Rbx) + m.Load("R11", &x.R11) + m.Load("R10", &x.R10) + m.Load("R9", &x.R9) + m.Load("R8", &x.R8) + m.Load("Rax", &x.Rax) + m.Load("Rcx", &x.Rcx) + m.Load("Rdx", &x.Rdx) + m.Load("Rsi", &x.Rsi) + m.Load("Rdi", &x.Rdi) + m.Load("Orig_rax", &x.Orig_rax) + m.Load("Rip", &x.Rip) + m.Load("Cs", &x.Cs) + m.Load("Eflags", &x.Eflags) + m.Load("Rsp", &x.Rsp) + m.Load("Ss", &x.Ss) + m.Load("Fs_base", &x.Fs_base) + m.Load("Gs_base", &x.Gs_base) + m.Load("Ds", &x.Ds) + m.Load("Es", &x.Es) + m.Load("Fs", &x.Fs) + m.Load("Gs", &x.Gs) +} + +func (x *AuxEntry) beforeSave() {} +func (x *AuxEntry) save(m state.Map) { + x.beforeSave() + m.Save("Key", &x.Key) + m.Save("Value", &x.Value) +} + +func (x *AuxEntry) afterLoad() {} +func (x *AuxEntry) load(m state.Map) { + m.Load("Key", &x.Key) + m.Load("Value", &x.Value) +} + +func (x *SignalAct) beforeSave() {} +func (x *SignalAct) save(m state.Map) { + x.beforeSave() + m.Save("Handler", &x.Handler) + m.Save("Flags", &x.Flags) + m.Save("Restorer", &x.Restorer) + m.Save("Mask", &x.Mask) +} + +func (x *SignalAct) afterLoad() {} +func (x *SignalAct) load(m state.Map) { + m.Load("Handler", &x.Handler) + m.Load("Flags", &x.Flags) + m.Load("Restorer", &x.Restorer) + m.Load("Mask", &x.Mask) +} + +func (x *SignalStack) beforeSave() {} +func (x *SignalStack) save(m state.Map) { + x.beforeSave() + m.Save("Addr", &x.Addr) + m.Save("Flags", &x.Flags) + m.Save("Size", &x.Size) +} + +func (x *SignalStack) afterLoad() {} +func (x *SignalStack) load(m state.Map) { + m.Load("Addr", &x.Addr) + m.Load("Flags", &x.Flags) + m.Load("Size", &x.Size) +} + +func (x *SignalInfo) beforeSave() {} +func (x *SignalInfo) save(m state.Map) { + x.beforeSave() + m.Save("Signo", &x.Signo) + m.Save("Errno", &x.Errno) + m.Save("Code", &x.Code) + m.Save("Fields", &x.Fields) +} + +func (x *SignalInfo) afterLoad() {} +func (x *SignalInfo) load(m state.Map) { + m.Load("Signo", &x.Signo) + m.Load("Errno", &x.Errno) + m.Load("Code", &x.Code) + m.Load("Fields", &x.Fields) +} + +func init() { + state.Register("pkg/sentry/arch.MmapLayout", (*MmapLayout)(nil), state.Fns{Save: (*MmapLayout).save, Load: (*MmapLayout).load}) + state.Register("pkg/sentry/arch.syscallPtraceRegs", (*syscallPtraceRegs)(nil), state.Fns{Save: (*syscallPtraceRegs).save, Load: (*syscallPtraceRegs).load}) + state.Register("pkg/sentry/arch.AuxEntry", (*AuxEntry)(nil), state.Fns{Save: (*AuxEntry).save, Load: (*AuxEntry).load}) + state.Register("pkg/sentry/arch.SignalAct", (*SignalAct)(nil), state.Fns{Save: (*SignalAct).save, Load: (*SignalAct).load}) + state.Register("pkg/sentry/arch.SignalStack", (*SignalStack)(nil), state.Fns{Save: (*SignalStack).save, Load: (*SignalStack).load}) + state.Register("pkg/sentry/arch.SignalInfo", (*SignalInfo)(nil), state.Fns{Save: (*SignalInfo).save, Load: (*SignalInfo).load}) +} diff --git a/pkg/sentry/arch/arch_x86_impl.go b/pkg/sentry/arch/arch_x86_impl.go index 04ac283c6..04ac283c6 100644..100755 --- a/pkg/sentry/arch/arch_x86_impl.go +++ b/pkg/sentry/arch/arch_x86_impl.go diff --git a/pkg/sentry/arch/registers.proto b/pkg/sentry/arch/registers.proto deleted file mode 100644 index 60c027aab..000000000 --- a/pkg/sentry/arch/registers.proto +++ /dev/null @@ -1,92 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -syntax = "proto3"; - -package gvisor; - -message AMD64Registers { - uint64 rax = 1; - uint64 rbx = 2; - uint64 rcx = 3; - uint64 rdx = 4; - uint64 rsi = 5; - uint64 rdi = 6; - uint64 rsp = 7; - uint64 rbp = 8; - - uint64 r8 = 9; - uint64 r9 = 10; - uint64 r10 = 11; - uint64 r11 = 12; - uint64 r12 = 13; - uint64 r13 = 14; - uint64 r14 = 15; - uint64 r15 = 16; - - uint64 rip = 17; - uint64 rflags = 18; - uint64 orig_rax = 19; - uint64 cs = 20; - uint64 ds = 21; - uint64 es = 22; - uint64 fs = 23; - uint64 gs = 24; - uint64 ss = 25; - uint64 fs_base = 26; - uint64 gs_base = 27; -} - -message ARM64Registers { - uint64 r0 = 1; - uint64 r1 = 2; - uint64 r2 = 3; - uint64 r3 = 4; - uint64 r4 = 5; - uint64 r5 = 6; - uint64 r6 = 7; - uint64 r7 = 8; - uint64 r8 = 9; - uint64 r9 = 10; - uint64 r10 = 11; - uint64 r11 = 12; - uint64 r12 = 13; - uint64 r13 = 14; - uint64 r14 = 15; - uint64 r15 = 16; - uint64 r16 = 17; - uint64 r17 = 18; - uint64 r18 = 19; - uint64 r19 = 20; - uint64 r20 = 21; - uint64 r21 = 22; - uint64 r22 = 23; - uint64 r23 = 24; - uint64 r24 = 25; - uint64 r25 = 26; - uint64 r26 = 27; - uint64 r27 = 28; - uint64 r28 = 29; - uint64 r29 = 30; - uint64 r30 = 31; - uint64 sp = 32; - uint64 pc = 33; - uint64 pstate = 34; -} -message Registers { - oneof arch { - AMD64Registers amd64 = 1; - ARM64Registers arm64 = 2; - } -} diff --git a/pkg/sentry/arch/registers_go_proto/registers.pb.go b/pkg/sentry/arch/registers_go_proto/registers.pb.go new file mode 100755 index 000000000..c4e9584b5 --- /dev/null +++ b/pkg/sentry/arch/registers_go_proto/registers.pb.go @@ -0,0 +1,697 @@ +// Code generated by protoc-gen-go. DO NOT EDIT. +// source: pkg/sentry/arch/registers.proto + +package gvisor + +import ( + fmt "fmt" + proto "github.com/golang/protobuf/proto" + math "math" +) + +// Reference imports to suppress errors if they are not otherwise used. +var _ = proto.Marshal +var _ = fmt.Errorf +var _ = math.Inf + +// This is a compile-time assertion to ensure that this generated file +// is compatible with the proto package it is being compiled against. +// A compilation error at this line likely means your copy of the +// proto package needs to be updated. +const _ = proto.ProtoPackageIsVersion3 // please upgrade the proto package + +type AMD64Registers struct { + Rax uint64 `protobuf:"varint,1,opt,name=rax,proto3" json:"rax,omitempty"` + Rbx uint64 `protobuf:"varint,2,opt,name=rbx,proto3" json:"rbx,omitempty"` + Rcx uint64 `protobuf:"varint,3,opt,name=rcx,proto3" json:"rcx,omitempty"` + Rdx uint64 `protobuf:"varint,4,opt,name=rdx,proto3" json:"rdx,omitempty"` + Rsi uint64 `protobuf:"varint,5,opt,name=rsi,proto3" json:"rsi,omitempty"` + Rdi uint64 `protobuf:"varint,6,opt,name=rdi,proto3" json:"rdi,omitempty"` + Rsp uint64 `protobuf:"varint,7,opt,name=rsp,proto3" json:"rsp,omitempty"` + Rbp uint64 `protobuf:"varint,8,opt,name=rbp,proto3" json:"rbp,omitempty"` + R8 uint64 `protobuf:"varint,9,opt,name=r8,proto3" json:"r8,omitempty"` + R9 uint64 `protobuf:"varint,10,opt,name=r9,proto3" json:"r9,omitempty"` + R10 uint64 `protobuf:"varint,11,opt,name=r10,proto3" json:"r10,omitempty"` + R11 uint64 `protobuf:"varint,12,opt,name=r11,proto3" json:"r11,omitempty"` + R12 uint64 `protobuf:"varint,13,opt,name=r12,proto3" json:"r12,omitempty"` + R13 uint64 `protobuf:"varint,14,opt,name=r13,proto3" json:"r13,omitempty"` + R14 uint64 `protobuf:"varint,15,opt,name=r14,proto3" json:"r14,omitempty"` + R15 uint64 `protobuf:"varint,16,opt,name=r15,proto3" json:"r15,omitempty"` + Rip uint64 `protobuf:"varint,17,opt,name=rip,proto3" json:"rip,omitempty"` + Rflags uint64 `protobuf:"varint,18,opt,name=rflags,proto3" json:"rflags,omitempty"` + OrigRax uint64 `protobuf:"varint,19,opt,name=orig_rax,json=origRax,proto3" json:"orig_rax,omitempty"` + Cs uint64 `protobuf:"varint,20,opt,name=cs,proto3" json:"cs,omitempty"` + Ds uint64 `protobuf:"varint,21,opt,name=ds,proto3" json:"ds,omitempty"` + Es uint64 `protobuf:"varint,22,opt,name=es,proto3" json:"es,omitempty"` + Fs uint64 `protobuf:"varint,23,opt,name=fs,proto3" json:"fs,omitempty"` + Gs uint64 `protobuf:"varint,24,opt,name=gs,proto3" json:"gs,omitempty"` + Ss uint64 `protobuf:"varint,25,opt,name=ss,proto3" json:"ss,omitempty"` + FsBase uint64 `protobuf:"varint,26,opt,name=fs_base,json=fsBase,proto3" json:"fs_base,omitempty"` + GsBase uint64 `protobuf:"varint,27,opt,name=gs_base,json=gsBase,proto3" json:"gs_base,omitempty"` + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *AMD64Registers) Reset() { *m = AMD64Registers{} } +func (m *AMD64Registers) String() string { return proto.CompactTextString(m) } +func (*AMD64Registers) ProtoMessage() {} +func (*AMD64Registers) Descriptor() ([]byte, []int) { + return fileDescriptor_082b7510610e0457, []int{0} +} + +func (m *AMD64Registers) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_AMD64Registers.Unmarshal(m, b) +} +func (m *AMD64Registers) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_AMD64Registers.Marshal(b, m, deterministic) +} +func (m *AMD64Registers) XXX_Merge(src proto.Message) { + xxx_messageInfo_AMD64Registers.Merge(m, src) +} +func (m *AMD64Registers) XXX_Size() int { + return xxx_messageInfo_AMD64Registers.Size(m) +} +func (m *AMD64Registers) XXX_DiscardUnknown() { + xxx_messageInfo_AMD64Registers.DiscardUnknown(m) +} + +var xxx_messageInfo_AMD64Registers proto.InternalMessageInfo + +func (m *AMD64Registers) GetRax() uint64 { + if m != nil { + return m.Rax + } + return 0 +} + +func (m *AMD64Registers) GetRbx() uint64 { + if m != nil { + return m.Rbx + } + return 0 +} + +func (m *AMD64Registers) GetRcx() uint64 { + if m != nil { + return m.Rcx + } + return 0 +} + +func (m *AMD64Registers) GetRdx() uint64 { + if m != nil { + return m.Rdx + } + return 0 +} + +func (m *AMD64Registers) GetRsi() uint64 { + if m != nil { + return m.Rsi + } + return 0 +} + +func (m *AMD64Registers) GetRdi() uint64 { + if m != nil { + return m.Rdi + } + return 0 +} + +func (m *AMD64Registers) GetRsp() uint64 { + if m != nil { + return m.Rsp + } + return 0 +} + +func (m *AMD64Registers) GetRbp() uint64 { + if m != nil { + return m.Rbp + } + return 0 +} + +func (m *AMD64Registers) GetR8() uint64 { + if m != nil { + return m.R8 + } + return 0 +} + +func (m *AMD64Registers) GetR9() uint64 { + if m != nil { + return m.R9 + } + return 0 +} + +func (m *AMD64Registers) GetR10() uint64 { + if m != nil { + return m.R10 + } + return 0 +} + +func (m *AMD64Registers) GetR11() uint64 { + if m != nil { + return m.R11 + } + return 0 +} + +func (m *AMD64Registers) GetR12() uint64 { + if m != nil { + return m.R12 + } + return 0 +} + +func (m *AMD64Registers) GetR13() uint64 { + if m != nil { + return m.R13 + } + return 0 +} + +func (m *AMD64Registers) GetR14() uint64 { + if m != nil { + return m.R14 + } + return 0 +} + +func (m *AMD64Registers) GetR15() uint64 { + if m != nil { + return m.R15 + } + return 0 +} + +func (m *AMD64Registers) GetRip() uint64 { + if m != nil { + return m.Rip + } + return 0 +} + +func (m *AMD64Registers) GetRflags() uint64 { + if m != nil { + return m.Rflags + } + return 0 +} + +func (m *AMD64Registers) GetOrigRax() uint64 { + if m != nil { + return m.OrigRax + } + return 0 +} + +func (m *AMD64Registers) GetCs() uint64 { + if m != nil { + return m.Cs + } + return 0 +} + +func (m *AMD64Registers) GetDs() uint64 { + if m != nil { + return m.Ds + } + return 0 +} + +func (m *AMD64Registers) GetEs() uint64 { + if m != nil { + return m.Es + } + return 0 +} + +func (m *AMD64Registers) GetFs() uint64 { + if m != nil { + return m.Fs + } + return 0 +} + +func (m *AMD64Registers) GetGs() uint64 { + if m != nil { + return m.Gs + } + return 0 +} + +func (m *AMD64Registers) GetSs() uint64 { + if m != nil { + return m.Ss + } + return 0 +} + +func (m *AMD64Registers) GetFsBase() uint64 { + if m != nil { + return m.FsBase + } + return 0 +} + +func (m *AMD64Registers) GetGsBase() uint64 { + if m != nil { + return m.GsBase + } + return 0 +} + +type ARM64Registers struct { + R0 uint64 `protobuf:"varint,1,opt,name=r0,proto3" json:"r0,omitempty"` + R1 uint64 `protobuf:"varint,2,opt,name=r1,proto3" json:"r1,omitempty"` + R2 uint64 `protobuf:"varint,3,opt,name=r2,proto3" json:"r2,omitempty"` + R3 uint64 `protobuf:"varint,4,opt,name=r3,proto3" json:"r3,omitempty"` + R4 uint64 `protobuf:"varint,5,opt,name=r4,proto3" json:"r4,omitempty"` + R5 uint64 `protobuf:"varint,6,opt,name=r5,proto3" json:"r5,omitempty"` + R6 uint64 `protobuf:"varint,7,opt,name=r6,proto3" json:"r6,omitempty"` + R7 uint64 `protobuf:"varint,8,opt,name=r7,proto3" json:"r7,omitempty"` + R8 uint64 `protobuf:"varint,9,opt,name=r8,proto3" json:"r8,omitempty"` + R9 uint64 `protobuf:"varint,10,opt,name=r9,proto3" json:"r9,omitempty"` + R10 uint64 `protobuf:"varint,11,opt,name=r10,proto3" json:"r10,omitempty"` + R11 uint64 `protobuf:"varint,12,opt,name=r11,proto3" json:"r11,omitempty"` + R12 uint64 `protobuf:"varint,13,opt,name=r12,proto3" json:"r12,omitempty"` + R13 uint64 `protobuf:"varint,14,opt,name=r13,proto3" json:"r13,omitempty"` + R14 uint64 `protobuf:"varint,15,opt,name=r14,proto3" json:"r14,omitempty"` + R15 uint64 `protobuf:"varint,16,opt,name=r15,proto3" json:"r15,omitempty"` + R16 uint64 `protobuf:"varint,17,opt,name=r16,proto3" json:"r16,omitempty"` + R17 uint64 `protobuf:"varint,18,opt,name=r17,proto3" json:"r17,omitempty"` + R18 uint64 `protobuf:"varint,19,opt,name=r18,proto3" json:"r18,omitempty"` + R19 uint64 `protobuf:"varint,20,opt,name=r19,proto3" json:"r19,omitempty"` + R20 uint64 `protobuf:"varint,21,opt,name=r20,proto3" json:"r20,omitempty"` + R21 uint64 `protobuf:"varint,22,opt,name=r21,proto3" json:"r21,omitempty"` + R22 uint64 `protobuf:"varint,23,opt,name=r22,proto3" json:"r22,omitempty"` + R23 uint64 `protobuf:"varint,24,opt,name=r23,proto3" json:"r23,omitempty"` + R24 uint64 `protobuf:"varint,25,opt,name=r24,proto3" json:"r24,omitempty"` + R25 uint64 `protobuf:"varint,26,opt,name=r25,proto3" json:"r25,omitempty"` + R26 uint64 `protobuf:"varint,27,opt,name=r26,proto3" json:"r26,omitempty"` + R27 uint64 `protobuf:"varint,28,opt,name=r27,proto3" json:"r27,omitempty"` + R28 uint64 `protobuf:"varint,29,opt,name=r28,proto3" json:"r28,omitempty"` + R29 uint64 `protobuf:"varint,30,opt,name=r29,proto3" json:"r29,omitempty"` + R30 uint64 `protobuf:"varint,31,opt,name=r30,proto3" json:"r30,omitempty"` + Sp uint64 `protobuf:"varint,32,opt,name=sp,proto3" json:"sp,omitempty"` + Pc uint64 `protobuf:"varint,33,opt,name=pc,proto3" json:"pc,omitempty"` + Pstate uint64 `protobuf:"varint,34,opt,name=pstate,proto3" json:"pstate,omitempty"` + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *ARM64Registers) Reset() { *m = ARM64Registers{} } +func (m *ARM64Registers) String() string { return proto.CompactTextString(m) } +func (*ARM64Registers) ProtoMessage() {} +func (*ARM64Registers) Descriptor() ([]byte, []int) { + return fileDescriptor_082b7510610e0457, []int{1} +} + +func (m *ARM64Registers) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_ARM64Registers.Unmarshal(m, b) +} +func (m *ARM64Registers) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_ARM64Registers.Marshal(b, m, deterministic) +} +func (m *ARM64Registers) XXX_Merge(src proto.Message) { + xxx_messageInfo_ARM64Registers.Merge(m, src) +} +func (m *ARM64Registers) XXX_Size() int { + return xxx_messageInfo_ARM64Registers.Size(m) +} +func (m *ARM64Registers) XXX_DiscardUnknown() { + xxx_messageInfo_ARM64Registers.DiscardUnknown(m) +} + +var xxx_messageInfo_ARM64Registers proto.InternalMessageInfo + +func (m *ARM64Registers) GetR0() uint64 { + if m != nil { + return m.R0 + } + return 0 +} + +func (m *ARM64Registers) GetR1() uint64 { + if m != nil { + return m.R1 + } + return 0 +} + +func (m *ARM64Registers) GetR2() uint64 { + if m != nil { + return m.R2 + } + return 0 +} + +func (m *ARM64Registers) GetR3() uint64 { + if m != nil { + return m.R3 + } + return 0 +} + +func (m *ARM64Registers) GetR4() uint64 { + if m != nil { + return m.R4 + } + return 0 +} + +func (m *ARM64Registers) GetR5() uint64 { + if m != nil { + return m.R5 + } + return 0 +} + +func (m *ARM64Registers) GetR6() uint64 { + if m != nil { + return m.R6 + } + return 0 +} + +func (m *ARM64Registers) GetR7() uint64 { + if m != nil { + return m.R7 + } + return 0 +} + +func (m *ARM64Registers) GetR8() uint64 { + if m != nil { + return m.R8 + } + return 0 +} + +func (m *ARM64Registers) GetR9() uint64 { + if m != nil { + return m.R9 + } + return 0 +} + +func (m *ARM64Registers) GetR10() uint64 { + if m != nil { + return m.R10 + } + return 0 +} + +func (m *ARM64Registers) GetR11() uint64 { + if m != nil { + return m.R11 + } + return 0 +} + +func (m *ARM64Registers) GetR12() uint64 { + if m != nil { + return m.R12 + } + return 0 +} + +func (m *ARM64Registers) GetR13() uint64 { + if m != nil { + return m.R13 + } + return 0 +} + +func (m *ARM64Registers) GetR14() uint64 { + if m != nil { + return m.R14 + } + return 0 +} + +func (m *ARM64Registers) GetR15() uint64 { + if m != nil { + return m.R15 + } + return 0 +} + +func (m *ARM64Registers) GetR16() uint64 { + if m != nil { + return m.R16 + } + return 0 +} + +func (m *ARM64Registers) GetR17() uint64 { + if m != nil { + return m.R17 + } + return 0 +} + +func (m *ARM64Registers) GetR18() uint64 { + if m != nil { + return m.R18 + } + return 0 +} + +func (m *ARM64Registers) GetR19() uint64 { + if m != nil { + return m.R19 + } + return 0 +} + +func (m *ARM64Registers) GetR20() uint64 { + if m != nil { + return m.R20 + } + return 0 +} + +func (m *ARM64Registers) GetR21() uint64 { + if m != nil { + return m.R21 + } + return 0 +} + +func (m *ARM64Registers) GetR22() uint64 { + if m != nil { + return m.R22 + } + return 0 +} + +func (m *ARM64Registers) GetR23() uint64 { + if m != nil { + return m.R23 + } + return 0 +} + +func (m *ARM64Registers) GetR24() uint64 { + if m != nil { + return m.R24 + } + return 0 +} + +func (m *ARM64Registers) GetR25() uint64 { + if m != nil { + return m.R25 + } + return 0 +} + +func (m *ARM64Registers) GetR26() uint64 { + if m != nil { + return m.R26 + } + return 0 +} + +func (m *ARM64Registers) GetR27() uint64 { + if m != nil { + return m.R27 + } + return 0 +} + +func (m *ARM64Registers) GetR28() uint64 { + if m != nil { + return m.R28 + } + return 0 +} + +func (m *ARM64Registers) GetR29() uint64 { + if m != nil { + return m.R29 + } + return 0 +} + +func (m *ARM64Registers) GetR30() uint64 { + if m != nil { + return m.R30 + } + return 0 +} + +func (m *ARM64Registers) GetSp() uint64 { + if m != nil { + return m.Sp + } + return 0 +} + +func (m *ARM64Registers) GetPc() uint64 { + if m != nil { + return m.Pc + } + return 0 +} + +func (m *ARM64Registers) GetPstate() uint64 { + if m != nil { + return m.Pstate + } + return 0 +} + +type Registers struct { + // Types that are valid to be assigned to Arch: + // *Registers_Amd64 + // *Registers_Arm64 + Arch isRegisters_Arch `protobuf_oneof:"arch"` + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *Registers) Reset() { *m = Registers{} } +func (m *Registers) String() string { return proto.CompactTextString(m) } +func (*Registers) ProtoMessage() {} +func (*Registers) Descriptor() ([]byte, []int) { + return fileDescriptor_082b7510610e0457, []int{2} +} + +func (m *Registers) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_Registers.Unmarshal(m, b) +} +func (m *Registers) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_Registers.Marshal(b, m, deterministic) +} +func (m *Registers) XXX_Merge(src proto.Message) { + xxx_messageInfo_Registers.Merge(m, src) +} +func (m *Registers) XXX_Size() int { + return xxx_messageInfo_Registers.Size(m) +} +func (m *Registers) XXX_DiscardUnknown() { + xxx_messageInfo_Registers.DiscardUnknown(m) +} + +var xxx_messageInfo_Registers proto.InternalMessageInfo + +type isRegisters_Arch interface { + isRegisters_Arch() +} + +type Registers_Amd64 struct { + Amd64 *AMD64Registers `protobuf:"bytes,1,opt,name=amd64,proto3,oneof"` +} + +type Registers_Arm64 struct { + Arm64 *ARM64Registers `protobuf:"bytes,2,opt,name=arm64,proto3,oneof"` +} + +func (*Registers_Amd64) isRegisters_Arch() {} + +func (*Registers_Arm64) isRegisters_Arch() {} + +func (m *Registers) GetArch() isRegisters_Arch { + if m != nil { + return m.Arch + } + return nil +} + +func (m *Registers) GetAmd64() *AMD64Registers { + if x, ok := m.GetArch().(*Registers_Amd64); ok { + return x.Amd64 + } + return nil +} + +func (m *Registers) GetArm64() *ARM64Registers { + if x, ok := m.GetArch().(*Registers_Arm64); ok { + return x.Arm64 + } + return nil +} + +// XXX_OneofWrappers is for the internal use of the proto package. +func (*Registers) XXX_OneofWrappers() []interface{} { + return []interface{}{ + (*Registers_Amd64)(nil), + (*Registers_Arm64)(nil), + } +} + +func init() { + proto.RegisterType((*AMD64Registers)(nil), "gvisor.AMD64Registers") + proto.RegisterType((*ARM64Registers)(nil), "gvisor.ARM64Registers") + proto.RegisterType((*Registers)(nil), "gvisor.Registers") +} + +func init() { proto.RegisterFile("pkg/sentry/arch/registers.proto", fileDescriptor_082b7510610e0457) } + +var fileDescriptor_082b7510610e0457 = []byte{ + // 544 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xcc, 0x94, 0x4b, 0x72, 0xd3, 0x40, + 0x10, 0x86, 0xc9, 0xc4, 0xb1, 0x93, 0x09, 0x98, 0x20, 0xc0, 0xe9, 0xf0, 0x4a, 0xf0, 0x8a, 0x95, + 0xad, 0x97, 0x65, 0x7b, 0x49, 0x8a, 0x05, 0x9b, 0x6c, 0x7c, 0x81, 0x94, 0x2c, 0xcb, 0x42, 0x05, + 0xc1, 0x53, 0xd3, 0x2e, 0x4a, 0xac, 0xb9, 0x29, 0x27, 0xa1, 0xfa, 0x31, 0x90, 0xdc, 0x80, 0x5d, + 0x7f, 0xdf, 0xb4, 0x4a, 0x23, 0xfd, 0x3d, 0x63, 0x2f, 0xdd, 0xd7, 0x66, 0x8a, 0xf5, 0xf7, 0xbd, + 0xff, 0x39, 0x2d, 0x7d, 0xf5, 0x65, 0xea, 0xeb, 0xa6, 0xc5, 0x7d, 0xed, 0x71, 0xe2, 0xfc, 0x6e, + 0xbf, 0x8b, 0xfa, 0xcd, 0x8f, 0x16, 0x77, 0x7e, 0xfc, 0xab, 0x67, 0x87, 0x1f, 0x6f, 0x3e, 0x15, + 0xf9, 0x2a, 0x34, 0x44, 0x67, 0xf6, 0xd0, 0x97, 0x1d, 0x1c, 0x5c, 0x1d, 0x7c, 0xe8, 0xad, 0xa8, + 0x64, 0xb3, 0xee, 0xc0, 0xa8, 0x59, 0x8b, 0xa9, 0x3a, 0x38, 0x54, 0x53, 0x89, 0xd9, 0x74, 0xd0, + 0x53, 0xb3, 0x11, 0x83, 0x2d, 0x1c, 0xa9, 0xc1, 0x56, 0x7a, 0x5a, 0xe8, 0x87, 0x1e, 0x31, 0xe8, + 0x60, 0x10, 0x7a, 0x9c, 0xbc, 0xcb, 0xc1, 0x71, 0x78, 0x97, 0x8b, 0x86, 0xd6, 0xf8, 0x05, 0x9c, + 0xb0, 0x30, 0x7e, 0xc1, 0xbc, 0x04, 0xab, 0xbc, 0xe4, 0x27, 0x92, 0x18, 0x4e, 0xf5, 0x89, 0x24, + 0x16, 0x93, 0xc0, 0xe3, 0x60, 0x12, 0x31, 0x29, 0x3c, 0x09, 0x26, 0x15, 0x93, 0xc1, 0x30, 0x98, + 0x4c, 0x4c, 0x0e, 0x4f, 0x83, 0xc9, 0xc5, 0xcc, 0xe0, 0x2c, 0x98, 0x19, 0x9b, 0xd6, 0xc1, 0x33, + 0x35, 0xad, 0x8b, 0x46, 0xb6, 0xef, 0xb7, 0xdf, 0xca, 0x06, 0x21, 0x62, 0xa9, 0x14, 0x5d, 0xd8, + 0xe3, 0x9d, 0x6f, 0x9b, 0x5b, 0xfa, 0x95, 0xcf, 0x79, 0x65, 0x40, 0xbc, 0x2a, 0x3b, 0xfa, 0x80, + 0x0a, 0xe1, 0x85, 0x7c, 0x40, 0x85, 0xc4, 0x1b, 0x84, 0x97, 0xc2, 0x1b, 0xe6, 0x1a, 0x61, 0x24, + 0x5c, 0x33, 0x6f, 0x11, 0xce, 0x85, 0xb7, 0xcc, 0x0d, 0x02, 0x08, 0x37, 0xcc, 0x88, 0x70, 0x21, + 0x8c, 0x18, 0x9d, 0xdb, 0xc1, 0x16, 0x6f, 0xd7, 0x25, 0xd6, 0xf0, 0x4a, 0xf6, 0xb4, 0xc5, 0xeb, + 0x12, 0x6b, 0x5a, 0x68, 0x74, 0xe1, 0xb5, 0x2c, 0x34, 0xbc, 0x30, 0xfe, 0x4d, 0x53, 0xb0, 0xba, + 0xb9, 0x3f, 0x05, 0xf4, 0x97, 0x63, 0x1d, 0x02, 0xe3, 0x63, 0xe6, 0x44, 0x47, 0xc0, 0xf8, 0x84, + 0x39, 0xd5, 0x01, 0x30, 0x3e, 0x65, 0xce, 0x34, 0x7e, 0xe3, 0x33, 0xe6, 0x5c, 0xc3, 0x37, 0x3e, + 0x67, 0x9e, 0x69, 0xf4, 0xc6, 0xcf, 0x98, 0x0b, 0x0d, 0xde, 0xf8, 0x82, 0x79, 0xae, 0xb1, 0x1b, + 0x3f, 0xff, 0xef, 0x52, 0x4f, 0x8a, 0xbf, 0xa9, 0x27, 0x85, 0x98, 0xb9, 0x46, 0x4e, 0xa5, 0x98, + 0x85, 0x46, 0x4d, 0xa5, 0x98, 0xa5, 0xe6, 0x4c, 0x25, 0x9b, 0x34, 0xd6, 0xa4, 0xa9, 0x14, 0x93, + 0x68, 0xd6, 0x54, 0x8a, 0x49, 0x35, 0x6d, 0x2a, 0xc5, 0x64, 0x9a, 0x37, 0x95, 0x62, 0x72, 0x4d, + 0x9c, 0x4a, 0x31, 0x33, 0x8d, 0x9b, 0x4a, 0x31, 0x85, 0xe6, 0x4c, 0xa5, 0x98, 0x39, 0xbc, 0x09, + 0x46, 0xf6, 0x9c, 0x2e, 0xe0, 0x6d, 0x30, 0xb2, 0xe7, 0x74, 0x09, 0xef, 0x82, 0x91, 0x3d, 0x67, + 0x31, 0x5c, 0xaa, 0xc9, 0x78, 0x12, 0xd0, 0xc1, 0x95, 0x8e, 0x1b, 0x9f, 0x4f, 0x57, 0xc1, 0x7b, + 0x61, 0x57, 0xd1, 0x89, 0x70, 0xb8, 0x2f, 0xf7, 0x35, 0x8c, 0x65, 0xc8, 0x84, 0xc6, 0x68, 0x4f, + 0xfe, 0x8d, 0xd7, 0xc4, 0x1e, 0x95, 0x77, 0x9b, 0x22, 0xe7, 0x09, 0x3b, 0x4d, 0x47, 0x13, 0xb9, + 0x8f, 0x26, 0x0f, 0xef, 0xa2, 0xcf, 0x8f, 0x56, 0xd2, 0xc6, 0xfd, 0xfe, 0xae, 0xc8, 0x79, 0x02, + 0xef, 0xf7, 0x3f, 0x98, 0x5a, 0xee, 0xa7, 0xb6, 0xeb, 0xbe, 0xed, 0xd1, 0xbd, 0xb7, 0xee, 0xf3, + 0x75, 0x97, 0xfd, 0x09, 0x00, 0x00, 0xff, 0xff, 0x79, 0x30, 0x5f, 0x13, 0x11, 0x05, 0x00, 0x00, +} diff --git a/pkg/sentry/arch/signal.go b/pkg/sentry/arch/signal.go index 8b03d0187..8b03d0187 100644..100755 --- a/pkg/sentry/arch/signal.go +++ b/pkg/sentry/arch/signal.go diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go index 0c1db4b13..0c1db4b13 100644..100755 --- a/pkg/sentry/arch/signal_arm64.go +++ b/pkg/sentry/arch/signal_arm64.go diff --git a/pkg/sentry/arch/syscalls_arm64.go b/pkg/sentry/arch/syscalls_arm64.go index dc13b6124..dc13b6124 100644..100755 --- a/pkg/sentry/arch/syscalls_arm64.go +++ b/pkg/sentry/arch/syscalls_arm64.go diff --git a/pkg/sentry/contexttest/BUILD b/pkg/sentry/contexttest/BUILD deleted file mode 100644 index 6f4c86684..000000000 --- a/pkg/sentry/contexttest/BUILD +++ /dev/null @@ -1,21 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "contexttest", - testonly = 1, - srcs = ["contexttest.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/memutil", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/platform/ptrace", - "//pkg/sentry/uniqueid", - ], -) diff --git a/pkg/sentry/contexttest/contexttest.go b/pkg/sentry/contexttest/contexttest.go deleted file mode 100644 index 031fc64ec..000000000 --- a/pkg/sentry/contexttest/contexttest.go +++ /dev/null @@ -1,188 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package contexttest builds a test context.Context. -package contexttest - -import ( - "os" - "sync/atomic" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/memutil" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/platform/ptrace" - "gvisor.dev/gvisor/pkg/sentry/uniqueid" -) - -// Context returns a Context that may be used in tests. Uses ptrace as the -// platform.Platform. -// -// Note that some filesystems may require a minimal kernel for testing, which -// this test context does not provide. For such tests, see kernel/contexttest. -func Context(tb testing.TB) context.Context { - const memfileName = "contexttest-memory" - memfd, err := memutil.CreateMemFD(memfileName, 0) - if err != nil { - tb.Fatalf("error creating application memory file: %v", err) - } - memfile := os.NewFile(uintptr(memfd), memfileName) - mf, err := pgalloc.NewMemoryFile(memfile, pgalloc.MemoryFileOpts{}) - if err != nil { - memfile.Close() - tb.Fatalf("error creating pgalloc.MemoryFile: %v", err) - } - p, err := ptrace.New() - if err != nil { - tb.Fatal(err) - } - // Test usage of context.Background is fine. - return &TestContext{ - Context: context.Background(), - l: limits.NewLimitSet(), - mf: mf, - platform: p, - creds: auth.NewAnonymousCredentials(), - otherValues: make(map[interface{}]interface{}), - } -} - -// TestContext represents a context with minimal functionality suitable for -// running tests. -type TestContext struct { - context.Context - l *limits.LimitSet - mf *pgalloc.MemoryFile - platform platform.Platform - creds *auth.Credentials - otherValues map[interface{}]interface{} -} - -// globalUniqueID tracks incremental unique identifiers for tests. -var globalUniqueID uint64 - -// globalUniqueIDProvider implements unix.UniqueIDProvider. -type globalUniqueIDProvider struct{} - -// UniqueID implements unix.UniqueIDProvider.UniqueID. -func (*globalUniqueIDProvider) UniqueID() uint64 { - return atomic.AddUint64(&globalUniqueID, 1) -} - -// lastInotifyCookie is a monotonically increasing counter for generating unique -// inotify cookies. Must be accessed using atomic ops. -var lastInotifyCookie uint32 - -// hostClock implements ktime.Clock. -type hostClock struct { - ktime.WallRateClock - ktime.NoClockEvents -} - -// Now implements ktime.Clock.Now. -func (hostClock) Now() ktime.Time { - return ktime.FromNanoseconds(time.Now().UnixNano()) -} - -// RegisterValue registers additional values with this test context. Useful for -// providing values from external packages that contexttest can't depend on. -func (t *TestContext) RegisterValue(key, value interface{}) { - t.otherValues[key] = value -} - -// Value implements context.Context. -func (t *TestContext) Value(key interface{}) interface{} { - switch key { - case auth.CtxCredentials: - return t.creds - case limits.CtxLimits: - return t.l - case pgalloc.CtxMemoryFile: - return t.mf - case pgalloc.CtxMemoryFileProvider: - return t - case platform.CtxPlatform: - return t.platform - case uniqueid.CtxGlobalUniqueID: - return (*globalUniqueIDProvider).UniqueID(nil) - case uniqueid.CtxGlobalUniqueIDProvider: - return &globalUniqueIDProvider{} - case uniqueid.CtxInotifyCookie: - return atomic.AddUint32(&lastInotifyCookie, 1) - case ktime.CtxRealtimeClock: - return hostClock{} - default: - if val, ok := t.otherValues[key]; ok { - return val - } - return t.Context.Value(key) - } -} - -// MemoryFile implements pgalloc.MemoryFileProvider.MemoryFile. -func (t *TestContext) MemoryFile() *pgalloc.MemoryFile { - return t.mf -} - -// RootContext returns a Context that may be used in tests that need root -// credentials. Uses ptrace as the platform.Platform. -func RootContext(tb testing.TB) context.Context { - return WithCreds(Context(tb), auth.NewRootCredentials(auth.NewRootUserNamespace())) -} - -// WithCreds returns a copy of ctx carrying creds. -func WithCreds(ctx context.Context, creds *auth.Credentials) context.Context { - return &authContext{ctx, creds} -} - -type authContext struct { - context.Context - creds *auth.Credentials -} - -// Value implements context.Context. -func (ac *authContext) Value(key interface{}) interface{} { - switch key { - case auth.CtxCredentials: - return ac.creds - default: - return ac.Context.Value(key) - } -} - -// WithLimitSet returns a copy of ctx carrying l. -func WithLimitSet(ctx context.Context, l *limits.LimitSet) context.Context { - return limitContext{ctx, l} -} - -type limitContext struct { - context.Context - l *limits.LimitSet -} - -// Value implements context.Context. -func (lc limitContext) Value(key interface{}) interface{} { - switch key { - case limits.CtxLimits: - return lc.l - default: - return lc.Context.Value(key) - } -} diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD deleted file mode 100644 index d16d78aa5..000000000 --- a/pkg/sentry/control/BUILD +++ /dev/null @@ -1,52 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "control", - srcs = [ - "control.go", - "logging.go", - "pprof.go", - "proc.go", - "state.go", - ], - visibility = [ - "//:sandbox", - ], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fd", - "//pkg/fspath", - "//pkg/log", - "//pkg/sentry/fs", - "//pkg/sentry/fs/host", - "//pkg/sentry/fsbridge", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/state", - "//pkg/sentry/strace", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/sentry/watchdog", - "//pkg/sync", - "//pkg/syserror", - "//pkg/tcpip/link/sniffer", - "//pkg/urpc", - ], -) - -go_test( - name = "control_test", - size = "small", - srcs = ["proc_test.go"], - library = ":control", - deps = [ - "//pkg/log", - "//pkg/sentry/kernel/time", - "//pkg/sentry/usage", - ], -) diff --git a/pkg/sentry/control/control_state_autogen.go b/pkg/sentry/control/control_state_autogen.go new file mode 100755 index 000000000..bd5797221 --- /dev/null +++ b/pkg/sentry/control/control_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package control diff --git a/pkg/sentry/control/proc_test.go b/pkg/sentry/control/proc_test.go deleted file mode 100644 index 0a88459b2..000000000 --- a/pkg/sentry/control/proc_test.go +++ /dev/null @@ -1,166 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package control - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/log" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usage" -) - -func init() { - log.SetLevel(log.Debug) -} - -// Tests that ProcessData.Table() prints with the correct format. -func TestProcessListTable(t *testing.T) { - testCases := []struct { - pl []*Process - expected string - }{ - { - pl: []*Process{}, - expected: "UID PID PPID C TTY STIME TIME CMD", - }, - { - pl: []*Process{ - { - UID: 0, - PID: 0, - PPID: 0, - C: 0, - TTY: "?", - STime: "0", - Time: "0", - Cmd: "zero", - }, - { - UID: 1, - PID: 1, - PPID: 1, - C: 1, - TTY: "pts/4", - STime: "1", - Time: "1", - Cmd: "one", - }, - }, - expected: `UID PID PPID C TTY STIME TIME CMD -0 0 0 0 ? 0 0 zero -1 1 1 1 pts/4 1 1 one`, - }, - } - - for _, tc := range testCases { - output := ProcessListToTable(tc.pl) - - if tc.expected != output { - t.Errorf("PrintTable(%v): got:\n%s\nwant:\n%s", tc.pl, output, tc.expected) - } - } -} - -func TestProcessListJSON(t *testing.T) { - testCases := []struct { - pl []*Process - expected string - }{ - { - pl: []*Process{}, - expected: "[]", - }, - { - pl: []*Process{ - { - UID: 0, - PID: 0, - PPID: 0, - C: 0, - STime: "0", - Time: "0", - Cmd: "zero", - }, - { - UID: 1, - PID: 1, - PPID: 1, - C: 1, - STime: "1", - Time: "1", - Cmd: "one", - }, - }, - expected: "[0,1]", - }, - } - - for _, tc := range testCases { - output, err := PrintPIDsJSON(tc.pl) - if err != nil { - t.Errorf("failed to generate JSON: %v", err) - } - - if tc.expected != output { - t.Errorf("PrintJSON(%v): got:\n%s\nwant:\n%s", tc.pl, output, tc.expected) - } - } -} - -func TestPercentCPU(t *testing.T) { - testCases := []struct { - stats usage.CPUStats - startTime ktime.Time - now ktime.Time - expected int32 - }{ - { - // Verify that 100% use is capped at 99. - stats: usage.CPUStats{UserTime: 1e9, SysTime: 1e9}, - startTime: ktime.FromNanoseconds(7e9), - now: ktime.FromNanoseconds(9e9), - expected: 99, - }, - { - // Verify that if usage > lifetime, we get at most 99% - // usage. - stats: usage.CPUStats{UserTime: 2e9, SysTime: 2e9}, - startTime: ktime.FromNanoseconds(7e9), - now: ktime.FromNanoseconds(9e9), - expected: 99, - }, - { - // Verify that 50% usage is reported correctly. - stats: usage.CPUStats{UserTime: 1e9, SysTime: 1e9}, - startTime: ktime.FromNanoseconds(12e9), - now: ktime.FromNanoseconds(16e9), - expected: 50, - }, - { - // Verify that 0% usage is reported correctly. - stats: usage.CPUStats{UserTime: 0, SysTime: 0}, - startTime: ktime.FromNanoseconds(12e9), - now: ktime.FromNanoseconds(14e9), - expected: 0, - }, - } - - for _, tc := range testCases { - if pcpu := percentCPU(tc.stats, tc.startTime, tc.now); pcpu != tc.expected { - t.Errorf("percentCPU(%v, %v, %v): got %d, want %d", tc.stats, tc.startTime, tc.now, pcpu, tc.expected) - } - } -} diff --git a/pkg/sentry/device/BUILD b/pkg/sentry/device/BUILD deleted file mode 100644 index e403cbd8b..000000000 --- a/pkg/sentry/device/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "device", - srcs = ["device.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/sync", - ], -) - -go_test( - name = "device_test", - size = "small", - srcs = ["device_test.go"], - library = ":device", -) diff --git a/pkg/sentry/device/device_state_autogen.go b/pkg/sentry/device/device_state_autogen.go new file mode 100755 index 000000000..dd41a5659 --- /dev/null +++ b/pkg/sentry/device/device_state_autogen.go @@ -0,0 +1,52 @@ +// automatically generated by stateify. + +package device + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Registry) beforeSave() {} +func (x *Registry) save(m state.Map) { + x.beforeSave() + m.Save("lastAnonDeviceMinor", &x.lastAnonDeviceMinor) + m.Save("devices", &x.devices) +} + +func (x *Registry) afterLoad() {} +func (x *Registry) load(m state.Map) { + m.Load("lastAnonDeviceMinor", &x.lastAnonDeviceMinor) + m.Load("devices", &x.devices) +} + +func (x *ID) beforeSave() {} +func (x *ID) save(m state.Map) { + x.beforeSave() + m.Save("Major", &x.Major) + m.Save("Minor", &x.Minor) +} + +func (x *ID) afterLoad() {} +func (x *ID) load(m state.Map) { + m.Load("Major", &x.Major) + m.Load("Minor", &x.Minor) +} + +func (x *Device) beforeSave() {} +func (x *Device) save(m state.Map) { + x.beforeSave() + m.Save("ID", &x.ID) + m.Save("last", &x.last) +} + +func (x *Device) afterLoad() {} +func (x *Device) load(m state.Map) { + m.Load("ID", &x.ID) + m.Load("last", &x.last) +} + +func init() { + state.Register("pkg/sentry/device.Registry", (*Registry)(nil), state.Fns{Save: (*Registry).save, Load: (*Registry).load}) + state.Register("pkg/sentry/device.ID", (*ID)(nil), state.Fns{Save: (*ID).save, Load: (*ID).load}) + state.Register("pkg/sentry/device.Device", (*Device)(nil), state.Fns{Save: (*Device).save, Load: (*Device).load}) +} diff --git a/pkg/sentry/device/device_test.go b/pkg/sentry/device/device_test.go deleted file mode 100644 index e3f51ce4f..000000000 --- a/pkg/sentry/device/device_test.go +++ /dev/null @@ -1,59 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package device - -import ( - "testing" -) - -func TestMultiDevice(t *testing.T) { - device := &MultiDevice{} - - // Check that Load fails to install virtual inodes that are - // uninitialized. - if device.Load(MultiDeviceKey{}, 0) { - t.Fatalf("got load of invalid virtual inode 0, want unsuccessful") - } - - inode := device.Map(MultiDeviceKey{}) - - // Assert that the same raw device and inode map to - // a consistent virtual inode. - if i := device.Map(MultiDeviceKey{}); i != inode { - t.Fatalf("got inode %d, want %d in %s", i, inode, device) - } - - // Assert that a new inode or new device does not conflict. - if i := device.Map(MultiDeviceKey{Device: 0, Inode: 1}); i == inode { - t.Fatalf("got reused inode %d, want new distinct inode in %s", i, device) - } - last := device.Map(MultiDeviceKey{Device: 1, Inode: 0}) - if last == inode { - t.Fatalf("got reused inode %d, want new distinct inode in %s", last, device) - } - - // Virtual is the virtual inode we want to load. - virtual := last + 1 - - // Assert that we can load a virtual inode at a new place. - if !device.Load(MultiDeviceKey{Device: 0, Inode: 2}, virtual) { - t.Fatalf("got load of virtual inode %d failed, want success in %s", virtual, device) - } - - // Assert that the next inode skips over the loaded one. - if i := device.Map(MultiDeviceKey{Device: 0, Inode: 3}); i != virtual+1 { - t.Fatalf("got inode %d, want %d in %s", i, virtual+1, device) - } -} diff --git a/pkg/sentry/devices/memdev/BUILD b/pkg/sentry/devices/memdev/BUILD deleted file mode 100644 index abe58f818..000000000 --- a/pkg/sentry/devices/memdev/BUILD +++ /dev/null @@ -1,28 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "memdev", - srcs = [ - "full.go", - "memdev.go", - "null.go", - "random.go", - "zero.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/rand", - "//pkg/safemem", - "//pkg/sentry/fsimpl/devtmpfs", - "//pkg/sentry/memmap", - "//pkg/sentry/mm", - "//pkg/sentry/pgalloc", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/devices/memdev/full.go b/pkg/sentry/devices/memdev/full.go deleted file mode 100644 index c7e197691..000000000 --- a/pkg/sentry/devices/memdev/full.go +++ /dev/null @@ -1,75 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memdev - -import ( - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -const fullDevMinor = 7 - -// fullDevice implements vfs.Device for /dev/full. -type fullDevice struct{} - -// Open implements vfs.Device.Open. -func (fullDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &fullFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ - UseDentryMetadata: true, - }); err != nil { - return nil, err - } - return &fd.vfsfd, nil -} - -// fullFD implements vfs.FileDescriptionImpl for /dev/full. -type fullFD struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl - vfs.DentryMetadataFileDescriptionImpl -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *fullFD) Release() { - // noop -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *fullFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return dst.ZeroOut(ctx, dst.NumBytes()) -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *fullFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return dst.ZeroOut(ctx, dst.NumBytes()) -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *fullFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.ENOSPC -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *fullFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.ENOSPC -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *fullFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - return 0, nil -} diff --git a/pkg/sentry/devices/memdev/memdev.go b/pkg/sentry/devices/memdev/memdev.go deleted file mode 100644 index 5759900c4..000000000 --- a/pkg/sentry/devices/memdev/memdev.go +++ /dev/null @@ -1,59 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package memdev implements "mem" character devices, as implemented in Linux -// by drivers/char/mem.c and drivers/char/random.c. -package memdev - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/devtmpfs" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// Register registers all devices implemented by this package in vfsObj. -func Register(vfsObj *vfs.VirtualFilesystem) error { - for minor, dev := range map[uint32]vfs.Device{ - nullDevMinor: nullDevice{}, - zeroDevMinor: zeroDevice{}, - fullDevMinor: fullDevice{}, - randomDevMinor: randomDevice{}, - urandomDevMinor: randomDevice{}, - } { - if err := vfsObj.RegisterDevice(vfs.CharDevice, linux.MEM_MAJOR, minor, dev, &vfs.RegisterDeviceOptions{ - GroupName: "mem", - }); err != nil { - return err - } - } - return nil -} - -// CreateDevtmpfsFiles creates device special files in dev representing all -// devices implemented by this package. -func CreateDevtmpfsFiles(ctx context.Context, dev *devtmpfs.Accessor) error { - for minor, name := range map[uint32]string{ - nullDevMinor: "null", - zeroDevMinor: "zero", - fullDevMinor: "full", - randomDevMinor: "random", - urandomDevMinor: "urandom", - } { - if err := dev.CreateDeviceFile(ctx, name, vfs.CharDevice, linux.MEM_MAJOR, minor, 0666 /* mode */); err != nil { - return err - } - } - return nil -} diff --git a/pkg/sentry/devices/memdev/null.go b/pkg/sentry/devices/memdev/null.go deleted file mode 100644 index 33d060d02..000000000 --- a/pkg/sentry/devices/memdev/null.go +++ /dev/null @@ -1,76 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memdev - -import ( - "io" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/usermem" -) - -const nullDevMinor = 3 - -// nullDevice implements vfs.Device for /dev/null. -type nullDevice struct{} - -// Open implements vfs.Device.Open. -func (nullDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &nullFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ - UseDentryMetadata: true, - }); err != nil { - return nil, err - } - return &fd.vfsfd, nil -} - -// nullFD implements vfs.FileDescriptionImpl for /dev/null. -type nullFD struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl - vfs.DentryMetadataFileDescriptionImpl -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *nullFD) Release() { - // noop -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *nullFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return 0, io.EOF -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *nullFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return 0, io.EOF -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *nullFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return src.NumBytes(), nil -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *nullFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return src.NumBytes(), nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *nullFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - return 0, nil -} diff --git a/pkg/sentry/devices/memdev/random.go b/pkg/sentry/devices/memdev/random.go deleted file mode 100644 index acfa23149..000000000 --- a/pkg/sentry/devices/memdev/random.go +++ /dev/null @@ -1,92 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memdev - -import ( - "sync/atomic" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/rand" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/usermem" -) - -const ( - randomDevMinor = 8 - urandomDevMinor = 9 -) - -// randomDevice implements vfs.Device for /dev/random and /dev/urandom. -type randomDevice struct{} - -// Open implements vfs.Device.Open. -func (randomDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &randomFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ - UseDentryMetadata: true, - }); err != nil { - return nil, err - } - return &fd.vfsfd, nil -} - -// randomFD implements vfs.FileDescriptionImpl for /dev/random. -type randomFD struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl - vfs.DentryMetadataFileDescriptionImpl - - // off is the "file offset". off is accessed using atomic memory - // operations. - off int64 -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *randomFD) Release() { - // noop -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *randomFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return dst.CopyOutFrom(ctx, safemem.FromIOReader{rand.Reader}) -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *randomFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - n, err := dst.CopyOutFrom(ctx, safemem.FromIOReader{rand.Reader}) - atomic.AddInt64(&fd.off, n) - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *randomFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - // In Linux, this mixes the written bytes into the entropy pool; we just - // throw them away. - return src.NumBytes(), nil -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *randomFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - atomic.AddInt64(&fd.off, src.NumBytes()) - return src.NumBytes(), nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *randomFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - // Linux: drivers/char/random.c:random_fops.llseek == urandom_fops.llseek - // == noop_llseek - return atomic.LoadInt64(&fd.off), nil -} diff --git a/pkg/sentry/devices/memdev/zero.go b/pkg/sentry/devices/memdev/zero.go deleted file mode 100644 index 3b1372b9e..000000000 --- a/pkg/sentry/devices/memdev/zero.go +++ /dev/null @@ -1,88 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memdev - -import ( - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/usermem" -) - -const zeroDevMinor = 5 - -// zeroDevice implements vfs.Device for /dev/zero. -type zeroDevice struct{} - -// Open implements vfs.Device.Open. -func (zeroDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &zeroFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{ - UseDentryMetadata: true, - }); err != nil { - return nil, err - } - return &fd.vfsfd, nil -} - -// zeroFD implements vfs.FileDescriptionImpl for /dev/zero. -type zeroFD struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl - vfs.DentryMetadataFileDescriptionImpl -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *zeroFD) Release() { - // noop -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *zeroFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return dst.ZeroOut(ctx, dst.NumBytes()) -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *zeroFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return dst.ZeroOut(ctx, dst.NumBytes()) -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *zeroFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return src.NumBytes(), nil -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *zeroFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return src.NumBytes(), nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *zeroFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - return 0, nil -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *zeroFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - m, err := mm.NewSharedAnonMappable(opts.Length, pgalloc.MemoryFileProviderFromContext(ctx)) - if err != nil { - return err - } - opts.MappingIdentity = m - opts.Mappable = m - return nil -} diff --git a/pkg/sentry/fs/BUILD b/pkg/sentry/fs/BUILD deleted file mode 100644 index ea85ab33c..000000000 --- a/pkg/sentry/fs/BUILD +++ /dev/null @@ -1,135 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_library( - name = "fs", - srcs = [ - "attr.go", - "context.go", - "copy_up.go", - "dentry.go", - "dirent.go", - "dirent_cache.go", - "dirent_cache_limiter.go", - "dirent_list.go", - "dirent_state.go", - "event_list.go", - "file.go", - "file_operations.go", - "file_overlay.go", - "file_state.go", - "filesystems.go", - "flags.go", - "fs.go", - "inode.go", - "inode_inotify.go", - "inode_operations.go", - "inode_overlay.go", - "inotify.go", - "inotify_event.go", - "inotify_watch.go", - "mock.go", - "mount.go", - "mount_overlay.go", - "mounts.go", - "offset.go", - "overlay.go", - "path.go", - "restore.go", - "save.go", - "seek.go", - "splice.go", - "sync.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/amutex", - "//pkg/context", - "//pkg/log", - "//pkg/metric", - "//pkg/p9", - "//pkg/refs", - "//pkg/secio", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs/lock", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/memmap", - "//pkg/sentry/platform", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/uniqueid", - "//pkg/sentry/usage", - "//pkg/state", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_template_instance( - name = "dirent_list", - out = "dirent_list.go", - package = "fs", - prefix = "dirent", - template = "//pkg/ilist:generic_list", - types = { - "Linker": "*Dirent", - "Element": "*Dirent", - }, -) - -go_template_instance( - name = "event_list", - out = "event_list.go", - package = "fs", - prefix = "event", - template = "//pkg/ilist:generic_list", - types = { - "Linker": "*Event", - "Element": "*Event", - }, -) - -go_test( - name = "fs_x_test", - size = "small", - srcs = [ - "copy_up_test.go", - "file_overlay_test.go", - "inode_overlay_test.go", - "mounts_test.go", - ], - deps = [ - ":fs", - "//pkg/context", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/fs/tmpfs", - "//pkg/sentry/kernel/contexttest", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -go_test( - name = "fs_test", - size = "small", - srcs = [ - "dirent_cache_test.go", - "dirent_refs_test.go", - "mount_test.go", - "path_test.go", - ], - library = ":fs", - deps = [ - "//pkg/context", - "//pkg/sentry/contexttest", - ], -) diff --git a/pkg/sentry/fs/README.md b/pkg/sentry/fs/README.md deleted file mode 100644 index db4a1b730..000000000 --- a/pkg/sentry/fs/README.md +++ /dev/null @@ -1,229 +0,0 @@ -This package provides an implementation of the Linux virtual filesystem. - -[TOC] - -## Overview - -- An `fs.Dirent` caches an `fs.Inode` in memory at a path in the VFS, giving - the `fs.Inode` a relative position with respect to other `fs.Inode`s. - -- If an `fs.Dirent` is referenced by two file descriptors, then those file - descriptors are coherent with each other: they depend on the same - `fs.Inode`. - -- A mount point is an `fs.Dirent` for which `fs.Dirent.mounted` is true. It - exposes the root of a mounted filesystem. - -- The `fs.Inode` produced by a registered filesystem on mount(2) owns an - `fs.MountedFilesystem` from which other `fs.Inode`s will be looked up. For a - remote filesystem, the `fs.MountedFilesystem` owns the connection to that - remote filesystem. - -- In general: - -``` -fs.Inode <------------------------------ -| | -| | -produced by | -exactly one | -| responsible for the -| virtual identity of -v | -fs.MountedFilesystem ------------------- -``` - -Glossary: - -- VFS: virtual filesystem. - -- inode: a virtual file object holding a cached view of a file on a backing - filesystem (includes metadata and page caches). - -- superblock: the virtual state of a mounted filesystem (e.g. the virtual - inode number set). - -- mount namespace: a view of the mounts under a root (during path traversal, - the VFS makes visible/follows the mount point that is in the current task's - mount namespace). - -## Save and restore - -An application's hard dependencies on filesystem state can be broken down into -two categories: - -- The state necessary to execute a traversal on or view the *virtual* - filesystem hierarchy, regardless of what files an application has open. - -- The state necessary to represent open files. - -The first is always necessary to save and restore. An application may never have -any open file descriptors, but across save and restore it should see a coherent -view of any mount namespace. NOTE(b/63601033): Currently only one "initial" -mount namespace is supported. - -The second is so that system calls across save and restore are coherent with -each other (e.g. so that unintended re-reads or overwrites do not occur). - -Specifically this state is: - -- An `fs.MountManager` containing mount points. - -- A `kernel.FDTable` containing pointers to open files. - -Anything else managed by the VFS that can be easily loaded into memory from a -filesystem is synced back to those filesystems and is not saved. Examples are -pages in page caches used for optimizations (i.e. readahead and writeback), and -directory entries used to accelerate path lookups. - -### Mount points - -Saving and restoring a mount point means saving and restoring: - -- The root of the mounted filesystem. - -- Mount flags, which control how the VFS interacts with the mounted - filesystem. - -- Any relevant metadata about the mounted filesystem. - -- All `fs.Inode`s referenced by the application that reside under the mount - point. - -`fs.MountedFilesystem` is metadata about a filesystem that is mounted. It is -referenced by every `fs.Inode` loaded into memory under the mount point -including the `fs.Inode` of the mount point itself. The `fs.MountedFilesystem` -maps file objects on the filesystem to a virtualized `fs.Inode` number and vice -versa. - -To restore all `fs.Inode`s under a given mount point, each `fs.Inode` leverages -its dependency on an `fs.MountedFilesystem`. Since the `fs.MountedFilesystem` -knows how an `fs.Inode` maps to a file object on a backing filesystem, this -mapping can be trivially consulted by each `fs.Inode` when the `fs.Inode` is -restored. - -In detail, a mount point is saved in two steps: - -- First, after the kernel is paused but before state.Save, we walk all mount - namespaces and install a mapping from `fs.Inode` numbers to file paths - relative to the root of the mounted filesystem in each - `fs.MountedFilesystem`. This is subsequently called the set of `fs.Inode` - mappings. - -- Second, during state.Save, each `fs.MountedFilesystem` decides whether to - save the set of `fs.Inode` mappings. In-memory filesystems, like tmpfs, have - no need to save a set of `fs.Inode` mappings, since the `fs.Inode`s can be - entirely encoded in state file. Each `fs.MountedFilesystem` also optionally - saves the device name from when the filesystem was originally mounted. Each - `fs.Inode` saves its virtual identifier and a reference to a - `fs.MountedFilesystem`. - -A mount point is restored in two steps: - -- First, before state.Load, all mount configurations are stored in a global - `fs.RestoreEnvironment`. This tells us what mount points the user wants to - restore and how to re-establish pointers to backing filesystems. - -- Second, during state.Load, each `fs.MountedFilesystem` optionally searches - for a mount in the `fs.RestoreEnvironment` that matches its saved device - name. The `fs.MountedFilesystem` then reestablishes a pointer to the root of - the mounted filesystem. For example, the mount specification provides the - network connection for a mounted remote filesystem client to communicate - with its remote file server. The `fs.MountedFilesystem` also trivially loads - its set of `fs.Inode` mappings. When an `fs.Inode` is encountered, the - `fs.Inode` loads its virtual identifier and its reference a - `fs.MountedFilesystem`. It uses the `fs.MountedFilesystem` to obtain the - root of the mounted filesystem and the `fs.Inode` mappings to obtain the - relative file path to its data. With these, the `fs.Inode` re-establishes a - pointer to its file object. - -A mount point can trivially restore its `fs.Inode`s in parallel since -`fs.Inode`s have a restore dependency on their `fs.MountedFilesystem` and not on -each other. - -### Open files - -An `fs.File` references the following filesystem objects: - -```go -fs.File -> fs.Dirent -> fs.Inode -> fs.MountedFilesystem -``` - -The `fs.Inode` is restored using its `fs.MountedFilesystem`. The -[Mount points](#mount-points) section above describes how this happens in -detail. The `fs.Dirent` restores its pointer to an `fs.Inode`, pointers to -parent and children `fs.Dirents`, and the basename of the file. - -Otherwise an `fs.File` restores flags, an offset, and a unique identifier (only -used internally). - -It may use the `fs.Inode`, which it indirectly holds a reference on through the -`fs.Dirent`, to reestablish an open file handle on the backing filesystem (e.g. -to continue reading and writing). - -## Overlay - -The overlay implementation in the fs package takes Linux overlayfs as a frame of -reference but corrects for several POSIX consistency errors. - -In Linux overlayfs, the `struct inode` used for reading and writing to the same -file may be different. This is because the `struct inode` is dissociated with -the process of copying up the file from the upper to the lower directory. Since -flock(2) and fcntl(2) locks, inotify(7) watches, page caches, and a file's -identity are all stored directly or indirectly off the `struct inode`, these -properties of the `struct inode` may be stale after the first modification. This -can lead to file locking bugs, missed inotify events, and inconsistent data in -shared memory mappings of files, to name a few problems. - -The fs package maintains a single `fs.Inode` to represent a directory entry in -an overlay and defines operations on this `fs.Inode` which synchronize with the -copy up process. This achieves several things: - -+ File locks, inotify watches, and the identity of the file need not be copied - at all. - -+ Memory mappings of files coordinate with the copy up process so that if a - file in the lower directory is memory mapped, all references to it are - invalidated, forcing the application to re-fault on memory mappings of the - file under the upper directory. - -The `fs.Inode` holds metadata about files in the upper and/or lower directories -via an `fs.overlayEntry`. The `fs.overlayEntry` implements the `fs.Mappable` -interface. It multiplexes between upper and lower directory memory mappings and -stores a copy of memory references so they can be transferred to the upper -directory `fs.Mappable` when the file is copied up. - -The lower filesystem in an overlay may contain another (nested) overlay, but the -upper filesystem may not contain another overlay. In other words, nested -overlays form a tree structure that only allows branching in the lower -filesystem. - -Caching decisions in the overlay are delegated to the upper filesystem, meaning -that the Keep and Revalidate methods on the overlay return the same values as -the upper filesystem. A small wrinkle is that the lower filesystem is not -allowed to return `true` from Revalidate, as the overlay can not reload inodes -from the lower filesystem. A lower filesystem that does return `true` from -Revalidate will trigger a panic. - -The `fs.Inode` also holds a reference to a `fs.MountedFilesystem` that -normalizes across the mounted filesystem state of the upper and lower -directories. - -When a file is copied from the lower to the upper directory, attempts to -interact with the file block until the copy completes. All copying synchronizes -with rename(2). - -## Future Work - -### Overlay - -When a file is copied from a lower directory to an upper directory, several -locks are taken: the global renamuMu and the copyMu of the `fs.Inode` being -copied. This blocks operations on the file, including fault handling of memory -mappings. Performance could be improved by copying files into a temporary -directory that resides on the same filesystem as the upper directory and doing -an atomic rename, holding locks only during the rename operation. - -Additionally files are copied up synchronously. For large files, this causes a -noticeable latency. Performance could be improved by pipelining copies at -non-overlapping file offsets. diff --git a/pkg/sentry/fs/anon/BUILD b/pkg/sentry/fs/anon/BUILD deleted file mode 100644 index aedcecfa1..000000000 --- a/pkg/sentry/fs/anon/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "anon", - srcs = [ - "anon.go", - "device.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fs/anon/anon_state_autogen.go b/pkg/sentry/fs/anon/anon_state_autogen.go new file mode 100755 index 000000000..b2b1a466e --- /dev/null +++ b/pkg/sentry/fs/anon/anon_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package anon diff --git a/pkg/sentry/fs/copy_up_test.go b/pkg/sentry/fs/copy_up_test.go deleted file mode 100644 index 91792d9fe..000000000 --- a/pkg/sentry/fs/copy_up_test.go +++ /dev/null @@ -1,183 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs_test - -import ( - "bytes" - "crypto/rand" - "fmt" - "io" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/fs" - _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/usermem" -) - -const ( - // origFileSize is the original file size. This many bytes should be - // copied up before the test file is modified. - origFileSize = 4096 - - // truncatedFileSize is the size to truncate all test files. - truncateFileSize = 10 -) - -// TestConcurrentCopyUp is a copy up stress test for an overlay. -// -// It creates a 64-level deep directory tree in the lower filesystem and -// populates the last subdirectory with 64 files containing random content: -// -// /lower -// /sudir0/.../subdir63/ -// /file0 -// ... -// /file63 -// -// The files are truncated concurrently by 4 goroutines per file. -// These goroutines contend with copying up all parent 64 subdirectories -// as well as the final file content. -// -// At the end of the test, we assert that the files respect the new truncated -// size and contain the content we expect. -func TestConcurrentCopyUp(t *testing.T) { - ctx := contexttest.Context(t) - files := makeOverlayTestFiles(t) - - var wg sync.WaitGroup - for _, file := range files { - for i := 0; i < 4; i++ { - wg.Add(1) - go func(o *overlayTestFile) { - if err := o.File.Dirent.Inode.Truncate(ctx, o.File.Dirent, truncateFileSize); err != nil { - t.Fatalf("failed to copy up: %v", err) - } - wg.Done() - }(file) - } - } - wg.Wait() - - for _, file := range files { - got := make([]byte, origFileSize) - n, err := file.File.Readv(ctx, usermem.BytesIOSequence(got)) - if int(n) != truncateFileSize { - t.Fatalf("read %d bytes from file, want %d", n, truncateFileSize) - } - if err != nil && err != io.EOF { - t.Fatalf("read got error %v, want nil", err) - } - if !bytes.Equal(got[:n], file.content[:truncateFileSize]) { - t.Fatalf("file content is %v, want %v", got[:n], file.content[:truncateFileSize]) - } - } -} - -type overlayTestFile struct { - File *fs.File - name string - content []byte -} - -func makeOverlayTestFiles(t *testing.T) []*overlayTestFile { - ctx := contexttest.Context(t) - - // Create a lower tmpfs mount. - fsys, _ := fs.FindFilesystem("tmpfs") - lower, err := fsys.Mount(contexttest.Context(t), "", fs.MountSourceFlags{}, "", nil) - if err != nil { - t.Fatalf("failed to mount tmpfs: %v", err) - } - lowerRoot := fs.NewDirent(ctx, lower, "") - - // Make a deep set of subdirectories that everyone shares. - next := lowerRoot - for i := 0; i < 64; i++ { - name := fmt.Sprintf("subdir%d", i) - err := next.CreateDirectory(ctx, lowerRoot, name, fs.FilePermsFromMode(0777)) - if err != nil { - t.Fatalf("failed to create dir %q: %v", name, err) - } - next, err = next.Walk(ctx, lowerRoot, name) - if err != nil { - t.Fatalf("failed to walk to %q: %v", name, err) - } - } - - // Make a bunch of files in the last directory. - var files []*overlayTestFile - for i := 0; i < 64; i++ { - name := fmt.Sprintf("file%d", i) - f, err := next.Create(ctx, next, name, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - t.Fatalf("failed to create file %q: %v", name, err) - } - defer f.DecRef() - - relname, _ := f.Dirent.FullName(lowerRoot) - - o := &overlayTestFile{ - name: relname, - content: make([]byte, origFileSize), - } - - if _, err := rand.Read(o.content); err != nil { - t.Fatalf("failed to read from /dev/urandom: %v", err) - } - - if _, err := f.Writev(ctx, usermem.BytesIOSequence(o.content)); err != nil { - t.Fatalf("failed to write content to file %q: %v", name, err) - } - - files = append(files, o) - } - - // Create an empty upper tmpfs mount which we will copy up into. - upper, err := fsys.Mount(ctx, "", fs.MountSourceFlags{}, "", nil) - if err != nil { - t.Fatalf("failed to mount tmpfs: %v", err) - } - - // Construct an overlay root. - overlay, err := fs.NewOverlayRoot(ctx, upper, lower, fs.MountSourceFlags{}) - if err != nil { - t.Fatalf("failed to construct overlay root: %v", err) - } - - // Create a MountNamespace to traverse the file system. - mns, err := fs.NewMountNamespace(ctx, overlay) - if err != nil { - t.Fatalf("failed to construct mount manager: %v", err) - } - - // Walk to all of the files in the overlay, open them readable. - for _, f := range files { - maxTraversals := uint(0) - d, err := mns.FindInode(ctx, mns.Root(), mns.Root(), f.name, &maxTraversals) - if err != nil { - t.Fatalf("failed to find %q: %v", f.name, err) - } - defer d.DecRef() - - f.File, err = d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("failed to open file %q readable: %v", f.name, err) - } - } - - return files -} diff --git a/pkg/sentry/fs/dev/BUILD b/pkg/sentry/fs/dev/BUILD deleted file mode 100644 index 9379a4d7b..000000000 --- a/pkg/sentry/fs/dev/BUILD +++ /dev/null @@ -1,40 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "dev", - srcs = [ - "dev.go", - "device.go", - "fs.go", - "full.go", - "net_tun.go", - "null.go", - "random.go", - "tty.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/rand", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/fs/tmpfs", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/memmap", - "//pkg/sentry/mm", - "//pkg/sentry/pgalloc", - "//pkg/sentry/socket/netstack", - "//pkg/syserror", - "//pkg/tcpip/link/tun", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/fs/dev/dev_state_autogen.go b/pkg/sentry/fs/dev/dev_state_autogen.go new file mode 100755 index 000000000..272f02672 --- /dev/null +++ b/pkg/sentry/fs/dev/dev_state_autogen.go @@ -0,0 +1,154 @@ +// automatically generated by stateify. + +package dev + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *fullDevice) beforeSave() {} +func (x *fullDevice) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *fullDevice) afterLoad() {} +func (x *fullDevice) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *fullFileOperations) beforeSave() {} +func (x *fullFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *fullFileOperations) afterLoad() {} +func (x *fullFileOperations) load(m state.Map) { +} + +func (x *netTunInodeOperations) beforeSave() {} +func (x *netTunInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *netTunInodeOperations) afterLoad() {} +func (x *netTunInodeOperations) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *netTunFileOperations) beforeSave() {} +func (x *netTunFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("device", &x.device) +} + +func (x *netTunFileOperations) afterLoad() {} +func (x *netTunFileOperations) load(m state.Map) { + m.Load("device", &x.device) +} + +func (x *nullDevice) beforeSave() {} +func (x *nullDevice) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *nullDevice) afterLoad() {} +func (x *nullDevice) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *nullFileOperations) beforeSave() {} +func (x *nullFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *nullFileOperations) afterLoad() {} +func (x *nullFileOperations) load(m state.Map) { +} + +func (x *zeroDevice) beforeSave() {} +func (x *zeroDevice) save(m state.Map) { + x.beforeSave() + m.Save("nullDevice", &x.nullDevice) +} + +func (x *zeroDevice) afterLoad() {} +func (x *zeroDevice) load(m state.Map) { + m.Load("nullDevice", &x.nullDevice) +} + +func (x *zeroFileOperations) beforeSave() {} +func (x *zeroFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *zeroFileOperations) afterLoad() {} +func (x *zeroFileOperations) load(m state.Map) { +} + +func (x *randomDevice) beforeSave() {} +func (x *randomDevice) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *randomDevice) afterLoad() {} +func (x *randomDevice) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *randomFileOperations) beforeSave() {} +func (x *randomFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *randomFileOperations) afterLoad() {} +func (x *randomFileOperations) load(m state.Map) { +} + +func (x *ttyInodeOperations) beforeSave() {} +func (x *ttyInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *ttyInodeOperations) afterLoad() {} +func (x *ttyInodeOperations) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *ttyFileOperations) beforeSave() {} +func (x *ttyFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *ttyFileOperations) afterLoad() {} +func (x *ttyFileOperations) load(m state.Map) { +} + +func init() { + state.Register("pkg/sentry/fs/dev.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("pkg/sentry/fs/dev.fullDevice", (*fullDevice)(nil), state.Fns{Save: (*fullDevice).save, Load: (*fullDevice).load}) + state.Register("pkg/sentry/fs/dev.fullFileOperations", (*fullFileOperations)(nil), state.Fns{Save: (*fullFileOperations).save, Load: (*fullFileOperations).load}) + state.Register("pkg/sentry/fs/dev.netTunInodeOperations", (*netTunInodeOperations)(nil), state.Fns{Save: (*netTunInodeOperations).save, Load: (*netTunInodeOperations).load}) + state.Register("pkg/sentry/fs/dev.netTunFileOperations", (*netTunFileOperations)(nil), state.Fns{Save: (*netTunFileOperations).save, Load: (*netTunFileOperations).load}) + state.Register("pkg/sentry/fs/dev.nullDevice", (*nullDevice)(nil), state.Fns{Save: (*nullDevice).save, Load: (*nullDevice).load}) + state.Register("pkg/sentry/fs/dev.nullFileOperations", (*nullFileOperations)(nil), state.Fns{Save: (*nullFileOperations).save, Load: (*nullFileOperations).load}) + state.Register("pkg/sentry/fs/dev.zeroDevice", (*zeroDevice)(nil), state.Fns{Save: (*zeroDevice).save, Load: (*zeroDevice).load}) + state.Register("pkg/sentry/fs/dev.zeroFileOperations", (*zeroFileOperations)(nil), state.Fns{Save: (*zeroFileOperations).save, Load: (*zeroFileOperations).load}) + state.Register("pkg/sentry/fs/dev.randomDevice", (*randomDevice)(nil), state.Fns{Save: (*randomDevice).save, Load: (*randomDevice).load}) + state.Register("pkg/sentry/fs/dev.randomFileOperations", (*randomFileOperations)(nil), state.Fns{Save: (*randomFileOperations).save, Load: (*randomFileOperations).load}) + state.Register("pkg/sentry/fs/dev.ttyInodeOperations", (*ttyInodeOperations)(nil), state.Fns{Save: (*ttyInodeOperations).save, Load: (*ttyInodeOperations).load}) + state.Register("pkg/sentry/fs/dev.ttyFileOperations", (*ttyFileOperations)(nil), state.Fns{Save: (*ttyFileOperations).save, Load: (*ttyFileOperations).load}) +} diff --git a/pkg/sentry/fs/dev/net_tun.go b/pkg/sentry/fs/dev/net_tun.go index dc7ad075a..dc7ad075a 100644..100755 --- a/pkg/sentry/fs/dev/net_tun.go +++ b/pkg/sentry/fs/dev/net_tun.go diff --git a/pkg/sentry/fs/dirent_cache_test.go b/pkg/sentry/fs/dirent_cache_test.go deleted file mode 100644 index 395c879f5..000000000 --- a/pkg/sentry/fs/dirent_cache_test.go +++ /dev/null @@ -1,247 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs - -import ( - "testing" -) - -func TestDirentCache(t *testing.T) { - const maxSize = 5 - - c := NewDirentCache(maxSize) - - // Size starts at 0. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // Create a Dirent d. - d := NewNegativeDirent("") - - // c does not contain d. - if got, want := c.contains(d), false; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Add d to the cache. - c.Add(d) - - // Size is now 1. - if got, want := c.Size(), uint64(1); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c contains d. - if got, want := c.contains(d), true; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Add maxSize-1 more elements. d should be oldest element. - for i := 0; i < maxSize-1; i++ { - c.Add(NewNegativeDirent("")) - } - - // Size is maxSize. - if got, want := c.Size(), uint64(maxSize); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c contains d. - if got, want := c.contains(d), true; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // "Bump" d to the front by re-adding it. - c.Add(d) - - // Size is maxSize. - if got, want := c.Size(), uint64(maxSize); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c contains d. - if got, want := c.contains(d), true; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Add maxSize-1 more elements. d should again be oldest element. - for i := 0; i < maxSize-1; i++ { - c.Add(NewNegativeDirent("")) - } - - // Size is maxSize. - if got, want := c.Size(), uint64(maxSize); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c contains d. - if got, want := c.contains(d), true; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Add one more element, which will bump d from the cache. - c.Add(NewNegativeDirent("")) - - // Size is maxSize. - if got, want := c.Size(), uint64(maxSize); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c does not contain d. - if got, want := c.contains(d), false; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Invalidating causes size to be 0 and list to be empty. - c.Invalidate() - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - if got, want := c.list.Empty(), true; got != want { - t.Errorf("c.list.Empty() got %v, want %v", got, want) - } - - // Fill cache with maxSize dirents. - for i := 0; i < maxSize; i++ { - c.Add(NewNegativeDirent("")) - } -} - -func TestDirentCacheLimiter(t *testing.T) { - const ( - globalMaxSize = 5 - maxSize = 3 - ) - - limit := NewDirentCacheLimiter(globalMaxSize) - c1 := NewDirentCache(maxSize) - c1.limit = limit - c2 := NewDirentCache(maxSize) - c2.limit = limit - - // Create a Dirent d. - d := NewNegativeDirent("") - - // Add d to the cache. - c1.Add(d) - if got, want := c1.Size(), uint64(1); got != want { - t.Errorf("c1.Size() got %v, want %v", got, want) - } - - // Add maxSize-1 more elements. d should be oldest element. - for i := 0; i < maxSize-1; i++ { - c1.Add(NewNegativeDirent("")) - } - if got, want := c1.Size(), uint64(maxSize); got != want { - t.Errorf("c1.Size() got %v, want %v", got, want) - } - - // Check that d is still there. - if got, want := c1.contains(d), true; got != want { - t.Errorf("c1.contains(d) got %v want %v", got, want) - } - - // Fill up the other cache, it will start dropping old entries from the cache - // when the global limit is reached. - for i := 0; i < maxSize; i++ { - c2.Add(NewNegativeDirent("")) - } - - // Check is what's remaining from global max. - if got, want := c2.Size(), globalMaxSize-maxSize; int(got) != want { - t.Errorf("c2.Size() got %v, want %v", got, want) - } - - // Check that d was not dropped. - if got, want := c1.contains(d), true; got != want { - t.Errorf("c1.contains(d) got %v want %v", got, want) - } - - // Add an entry that will eventually be dropped. Check is done later... - drop := NewNegativeDirent("") - c1.Add(drop) - - // Check that d is bumped to front even when global limit is reached. - c1.Add(d) - if got, want := c1.contains(d), true; got != want { - t.Errorf("c1.contains(d) got %v want %v", got, want) - } - - // Add 2 more element and check that: - // - d is still in the list: to verify that d was bumped - // - d2/d3 are in the list: older entries are dropped when global limit is - // reached. - // - drop is not in the list: indeed older elements are dropped. - d2 := NewNegativeDirent("") - c1.Add(d2) - d3 := NewNegativeDirent("") - c1.Add(d3) - if got, want := c1.contains(d), true; got != want { - t.Errorf("c1.contains(d) got %v want %v", got, want) - } - if got, want := c1.contains(d2), true; got != want { - t.Errorf("c1.contains(d2) got %v want %v", got, want) - } - if got, want := c1.contains(d3), true; got != want { - t.Errorf("c1.contains(d3) got %v want %v", got, want) - } - if got, want := c1.contains(drop), false; got != want { - t.Errorf("c1.contains(drop) got %v want %v", got, want) - } - - // Drop all entries from one cache. The other will be allowed to grow. - c1.Invalidate() - c2.Add(NewNegativeDirent("")) - if got, want := c2.Size(), uint64(maxSize); got != want { - t.Errorf("c2.Size() got %v, want %v", got, want) - } -} - -// TestNilDirentCache tests that a nil cache supports all cache operations, but -// treats them as noop. -func TestNilDirentCache(t *testing.T) { - // Create a nil cache. - var c *DirentCache - - // Size is zero. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // Call Add. - c.Add(NewNegativeDirent("")) - - // Size is zero. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // Call Remove. - c.Remove(NewNegativeDirent("")) - - // Size is zero. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // Call Invalidate. - c.Invalidate() - - // Size is zero. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } -} diff --git a/pkg/sentry/fs/dirent_list.go b/pkg/sentry/fs/dirent_list.go new file mode 100755 index 000000000..acdce100c --- /dev/null +++ b/pkg/sentry/fs/dirent_list.go @@ -0,0 +1,186 @@ +package fs + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type direntElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (direntElementMapper) linkerFor(elem *Dirent) *Dirent { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type direntList struct { + head *Dirent + tail *Dirent +} + +// Reset resets list l to the empty state. +func (l *direntList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *direntList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *direntList) Front() *Dirent { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *direntList) Back() *Dirent { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *direntList) PushFront(e *Dirent) { + linker := direntElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + direntElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *direntList) PushBack(e *Dirent) { + linker := direntElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + direntElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *direntList) PushBackList(m *direntList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + direntElementMapper{}.linkerFor(l.tail).SetNext(m.head) + direntElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *direntList) InsertAfter(b, e *Dirent) { + bLinker := direntElementMapper{}.linkerFor(b) + eLinker := direntElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + direntElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *direntList) InsertBefore(a, e *Dirent) { + aLinker := direntElementMapper{}.linkerFor(a) + eLinker := direntElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + direntElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *direntList) Remove(e *Dirent) { + linker := direntElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + direntElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + direntElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type direntEntry struct { + next *Dirent + prev *Dirent +} + +// Next returns the entry that follows e in the list. +func (e *direntEntry) Next() *Dirent { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *direntEntry) Prev() *Dirent { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *direntEntry) SetNext(elem *Dirent) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *direntEntry) SetPrev(elem *Dirent) { + e.prev = elem +} diff --git a/pkg/sentry/fs/dirent_refs_test.go b/pkg/sentry/fs/dirent_refs_test.go deleted file mode 100644 index 98d69c6f2..000000000 --- a/pkg/sentry/fs/dirent_refs_test.go +++ /dev/null @@ -1,418 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs - -import ( - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" -) - -func newMockDirInode(ctx context.Context, cache *DirentCache) *Inode { - return NewMockInode(ctx, NewMockMountSource(cache), StableAttr{Type: Directory}) -} - -func TestWalkPositive(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - root := NewDirent(ctx, newMockDirInode(ctx, nil), "root") - - if got := root.ReadRefs(); got != 1 { - t.Fatalf("root has a ref count of %d, want %d", got, 1) - } - - name := "d" - d, err := root.walk(ctx, root, name, false) - if err != nil { - t.Fatalf("root.walk(root, %q) got %v, want nil", name, err) - } - - if got := root.ReadRefs(); got != 2 { - t.Fatalf("root has a ref count of %d, want %d", got, 2) - } - - if got := d.ReadRefs(); got != 1 { - t.Fatalf("child name = %q has a ref count of %d, want %d", d.name, got, 1) - } - - d.DecRef() - - if got := root.ReadRefs(); got != 1 { - t.Fatalf("root has a ref count of %d, want %d", got, 1) - } - - if got := d.ReadRefs(); got != 0 { - t.Fatalf("child name = %q has a ref count of %d, want %d", d.name, got, 0) - } - - root.flush() - - if got := len(root.children); got != 0 { - t.Fatalf("root has %d children, want %d", got, 0) - } -} - -func TestWalkNegative(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - root := NewDirent(ctx, NewEmptyDir(ctx, nil), "root") - mn := root.Inode.InodeOperations.(*mockInodeOperationsLookupNegative) - - if got := root.ReadRefs(); got != 1 { - t.Fatalf("root has a ref count of %d, want %d", got, 1) - } - - name := "d" - for i := 0; i < 100; i++ { - _, err := root.walk(ctx, root, name, false) - if err != syscall.ENOENT { - t.Fatalf("root.walk(root, %q) got %v, want %v", name, err, syscall.ENOENT) - } - } - - if got := root.ReadRefs(); got != 1 { - t.Fatalf("root has a ref count of %d, want %d", got, 1) - } - - if got := len(root.children); got != 1 { - t.Fatalf("root has %d children, want %d", got, 1) - } - - w, ok := root.children[name] - if !ok { - t.Fatalf("root wants child at %q", name) - } - - child := w.Get() - if child == nil { - t.Fatalf("root wants to resolve weak reference") - } - - if !child.(*Dirent).IsNegative() { - t.Fatalf("root found positive child at %q, want negative", name) - } - - if got := child.(*Dirent).ReadRefs(); got != 2 { - t.Fatalf("child has a ref count of %d, want %d", got, 2) - } - - child.DecRef() - - if got := child.(*Dirent).ReadRefs(); got != 1 { - t.Fatalf("child has a ref count of %d, want %d", got, 1) - } - - if got := len(root.children); got != 1 { - t.Fatalf("root has %d children, want %d", got, 1) - } - - root.DecRef() - - if got := root.ReadRefs(); got != 0 { - t.Fatalf("root has a ref count of %d, want %d", got, 0) - } - - AsyncBarrier() - - if got := mn.releaseCalled; got != true { - t.Fatalf("root.Close was called %v, want true", got) - } -} - -type mockInodeOperationsLookupNegative struct { - *MockInodeOperations - releaseCalled bool -} - -func NewEmptyDir(ctx context.Context, cache *DirentCache) *Inode { - m := NewMockMountSource(cache) - return NewInode(ctx, &mockInodeOperationsLookupNegative{ - MockInodeOperations: NewMockInodeOperations(ctx), - }, m, StableAttr{Type: Directory}) -} - -func (m *mockInodeOperationsLookupNegative) Lookup(ctx context.Context, dir *Inode, p string) (*Dirent, error) { - return NewNegativeDirent(p), nil -} - -func (m *mockInodeOperationsLookupNegative) Release(context.Context) { - m.releaseCalled = true -} - -func TestHashNegativeToPositive(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - root := NewDirent(ctx, NewEmptyDir(ctx, nil), "root") - - name := "d" - _, err := root.walk(ctx, root, name, false) - if err != syscall.ENOENT { - t.Fatalf("root.walk(root, %q) got %v, want %v", name, err, syscall.ENOENT) - } - - if got := root.exists(ctx, root, name); got != false { - t.Fatalf("got %q exists, want does not exist", name) - } - - f, err := root.Create(ctx, root, name, FileFlags{}, FilePermissions{}) - if err != nil { - t.Fatalf("root.Create(%q, _), got error %v, want nil", name, err) - } - d := f.Dirent - - if d.IsNegative() { - t.Fatalf("got negative Dirent, want positive") - } - - if got := d.ReadRefs(); got != 1 { - t.Fatalf("child %q has a ref count of %d, want %d", name, got, 1) - } - - if got := root.ReadRefs(); got != 2 { - t.Fatalf("root has a ref count of %d, want %d", got, 2) - } - - if got := len(root.children); got != 1 { - t.Fatalf("got %d children, want %d", got, 1) - } - - w, ok := root.children[name] - if !ok { - t.Fatalf("failed to find weak reference to %q", name) - } - - child := w.Get() - if child == nil { - t.Fatalf("want to resolve weak reference") - } - - if child.(*Dirent) != d { - t.Fatalf("got foreign child") - } -} - -func TestRevalidate(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - for _, test := range []struct { - // desc is the test's description. - desc string - - // Whether to make negative Dirents. - makeNegative bool - }{ - { - desc: "Revalidate negative Dirent", - makeNegative: true, - }, - { - desc: "Revalidate positive Dirent", - makeNegative: false, - }, - } { - t.Run(test.desc, func(t *testing.T) { - ctx := contexttest.Context(t) - root := NewDirent(ctx, NewMockInodeRevalidate(ctx, test.makeNegative), "root") - - name := "d" - d1, err := root.walk(ctx, root, name, false) - if !test.makeNegative && err != nil { - t.Fatalf("root.walk(root, %q) got %v, want nil", name, err) - } - d2, err := root.walk(ctx, root, name, false) - if !test.makeNegative && err != nil { - t.Fatalf("root.walk(root, %q) got %v, want nil", name, err) - } - if !test.makeNegative && d1 == d2 { - t.Fatalf("revalidating walk got same *Dirent, want different") - } - if got := len(root.children); got != 1 { - t.Errorf("revalidating walk got %d children, want %d", got, 1) - } - }) - } -} - -type MockInodeOperationsRevalidate struct { - *MockInodeOperations - makeNegative bool -} - -func NewMockInodeRevalidate(ctx context.Context, makeNegative bool) *Inode { - mn := NewMockInodeOperations(ctx) - m := NewMockMountSource(nil) - m.MountSourceOperations.(*MockMountSourceOps).revalidate = true - return NewInode(ctx, &MockInodeOperationsRevalidate{MockInodeOperations: mn, makeNegative: makeNegative}, m, StableAttr{Type: Directory}) -} - -func (m *MockInodeOperationsRevalidate) Lookup(ctx context.Context, dir *Inode, p string) (*Dirent, error) { - if !m.makeNegative { - return m.MockInodeOperations.Lookup(ctx, dir, p) - } - return NewNegativeDirent(p), nil -} - -func TestCreateExtraRefs(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - for _, test := range []struct { - // desc is the test's description. - desc string - - // root is the Dirent to create from. - root *Dirent - - // expected references on walked Dirent. - refs int64 - }{ - { - desc: "Create caching", - root: NewDirent(ctx, NewEmptyDir(ctx, NewDirentCache(1)), "root"), - refs: 2, - }, - { - desc: "Create not caching", - root: NewDirent(ctx, NewEmptyDir(ctx, nil), "root"), - refs: 1, - }, - } { - t.Run(test.desc, func(t *testing.T) { - name := "d" - f, err := test.root.Create(ctx, test.root, name, FileFlags{}, FilePermissions{}) - if err != nil { - t.Fatalf("root.Create(root, %q) failed: %v", name, err) - } - d := f.Dirent - - if got := d.ReadRefs(); got != test.refs { - t.Errorf("dirent has a ref count of %d, want %d", got, test.refs) - } - }) - } -} - -func TestRemoveExtraRefs(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - for _, test := range []struct { - // desc is the test's description. - desc string - - // root is the Dirent to make and remove from. - root *Dirent - }{ - { - desc: "Remove caching", - root: NewDirent(ctx, NewEmptyDir(ctx, NewDirentCache(1)), "root"), - }, - { - desc: "Remove not caching", - root: NewDirent(ctx, NewEmptyDir(ctx, nil), "root"), - }, - } { - t.Run(test.desc, func(t *testing.T) { - name := "d" - f, err := test.root.Create(ctx, test.root, name, FileFlags{}, FilePermissions{}) - if err != nil { - t.Fatalf("root.Create(%q, _) failed: %v", name, err) - } - d := f.Dirent - - if err := test.root.Remove(contexttest.Context(t), test.root, name, false /* dirPath */); err != nil { - t.Fatalf("root.Remove(root, %q) failed: %v", name, err) - } - - if got := d.ReadRefs(); got != 1 { - t.Fatalf("dirent has a ref count of %d, want %d", got, 1) - } - - d.DecRef() - - test.root.flush() - - if got := len(test.root.children); got != 0 { - t.Errorf("root has %d children, want %d", got, 0) - } - }) - } -} - -func TestRenameExtraRefs(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - for _, test := range []struct { - // desc is the test's description. - desc string - - // cache of extra Dirent references, may be nil. - cache *DirentCache - }{ - { - desc: "Rename no caching", - cache: nil, - }, - { - desc: "Rename caching", - cache: NewDirentCache(5), - }, - } { - t.Run(test.desc, func(t *testing.T) { - ctx := contexttest.Context(t) - - dirAttr := StableAttr{Type: Directory} - - oldParent := NewDirent(ctx, NewMockInode(ctx, NewMockMountSource(test.cache), dirAttr), "old_parent") - newParent := NewDirent(ctx, NewMockInode(ctx, NewMockMountSource(test.cache), dirAttr), "new_parent") - - renamed, err := oldParent.Walk(ctx, oldParent, "old_child") - if err != nil { - t.Fatalf("Walk(oldParent, %q) got error %v, want nil", "old_child", err) - } - replaced, err := newParent.Walk(ctx, oldParent, "new_child") - if err != nil { - t.Fatalf("Walk(newParent, %q) got error %v, want nil", "new_child", err) - } - - if err := Rename(contexttest.RootContext(t), oldParent /*root */, oldParent, "old_child", newParent, "new_child"); err != nil { - t.Fatalf("Rename got error %v, want nil", err) - } - - oldParent.flush() - newParent.flush() - - // Expect to have only active references. - if got := renamed.ReadRefs(); got != 1 { - t.Errorf("renamed has ref count %d, want only active references %d", got, 1) - } - if got := replaced.ReadRefs(); got != 1 { - t.Errorf("replaced has ref count %d, want only active references %d", got, 1) - } - }) - } -} diff --git a/pkg/sentry/fs/event_list.go b/pkg/sentry/fs/event_list.go new file mode 100755 index 000000000..0274f41a2 --- /dev/null +++ b/pkg/sentry/fs/event_list.go @@ -0,0 +1,186 @@ +package fs + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type eventElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (eventElementMapper) linkerFor(elem *Event) *Event { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type eventList struct { + head *Event + tail *Event +} + +// Reset resets list l to the empty state. +func (l *eventList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *eventList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *eventList) Front() *Event { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *eventList) Back() *Event { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *eventList) PushFront(e *Event) { + linker := eventElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + eventElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *eventList) PushBack(e *Event) { + linker := eventElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + eventElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *eventList) PushBackList(m *eventList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + eventElementMapper{}.linkerFor(l.tail).SetNext(m.head) + eventElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *eventList) InsertAfter(b, e *Event) { + bLinker := eventElementMapper{}.linkerFor(b) + eLinker := eventElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + eventElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *eventList) InsertBefore(a, e *Event) { + aLinker := eventElementMapper{}.linkerFor(a) + eLinker := eventElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + eventElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *eventList) Remove(e *Event) { + linker := eventElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + eventElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + eventElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type eventEntry struct { + next *Event + prev *Event +} + +// Next returns the entry that follows e in the list. +func (e *eventEntry) Next() *Event { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *eventEntry) Prev() *Event { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *eventEntry) SetNext(elem *Event) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *eventEntry) SetPrev(elem *Event) { + e.prev = elem +} diff --git a/pkg/sentry/fs/fdpipe/BUILD b/pkg/sentry/fs/fdpipe/BUILD deleted file mode 100644 index 1d09e983c..000000000 --- a/pkg/sentry/fs/fdpipe/BUILD +++ /dev/null @@ -1,48 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "fdpipe", - srcs = [ - "pipe.go", - "pipe_opener.go", - "pipe_state.go", - ], - imports = ["gvisor.dev/gvisor/pkg/sentry/fs"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/fd", - "//pkg/fdnotifier", - "//pkg/log", - "//pkg/safemem", - "//pkg/secio", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "fdpipe_test", - size = "small", - srcs = [ - "pipe_opener_test.go", - "pipe_test.go", - ], - library = ":fdpipe", - deps = [ - "//pkg/context", - "//pkg/fd", - "//pkg/fdnotifier", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/syserror", - "//pkg/usermem", - "@com_github_google_uuid//:go_default_library", - ], -) diff --git a/pkg/sentry/fs/fdpipe/fdpipe_state_autogen.go b/pkg/sentry/fs/fdpipe/fdpipe_state_autogen.go new file mode 100755 index 000000000..9ed7a3d41 --- /dev/null +++ b/pkg/sentry/fs/fdpipe/fdpipe_state_autogen.go @@ -0,0 +1,27 @@ +// automatically generated by stateify. + +package fdpipe + +import ( + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/state" +) + +func (x *pipeOperations) save(m state.Map) { + x.beforeSave() + var flags fs.FileFlags = x.saveFlags() + m.SaveValue("flags", flags) + m.Save("opener", &x.opener) + m.Save("readAheadBuffer", &x.readAheadBuffer) +} + +func (x *pipeOperations) load(m state.Map) { + m.LoadWait("opener", &x.opener) + m.Load("readAheadBuffer", &x.readAheadBuffer) + m.LoadValue("flags", new(fs.FileFlags), func(y interface{}) { x.loadFlags(y.(fs.FileFlags)) }) + m.AfterLoad(x.afterLoad) +} + +func init() { + state.Register("pkg/sentry/fs/fdpipe.pipeOperations", (*pipeOperations)(nil), state.Fns{Save: (*pipeOperations).save, Load: (*pipeOperations).load}) +} diff --git a/pkg/sentry/fs/fdpipe/pipe_opener_test.go b/pkg/sentry/fs/fdpipe/pipe_opener_test.go deleted file mode 100644 index e556da48a..000000000 --- a/pkg/sentry/fs/fdpipe/pipe_opener_test.go +++ /dev/null @@ -1,523 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fdpipe - -import ( - "bytes" - "fmt" - "io" - "os" - "path" - "syscall" - "testing" - "time" - - "github.com/google/uuid" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -type hostOpener struct { - name string -} - -func (h *hostOpener) NonBlockingOpen(_ context.Context, p fs.PermMask) (*fd.FD, error) { - var flags int - switch { - case p.Read && p.Write: - flags = syscall.O_RDWR - case p.Write: - flags = syscall.O_WRONLY - case p.Read: - flags = syscall.O_RDONLY - default: - return nil, syscall.EINVAL - } - f, err := syscall.Open(h.name, flags|syscall.O_NONBLOCK, 0666) - if err != nil { - return nil, err - } - return fd.New(f), nil -} - -func pipename() string { - return fmt.Sprintf(path.Join(os.TempDir(), "test-named-pipe-%s"), uuid.New()) -} - -func mkpipe(name string) error { - return syscall.Mknod(name, syscall.S_IFIFO|0666, 0) -} - -func TestTryOpen(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // makePipe is true if the test case should create the pipe. - makePipe bool - - // flags are the fs.FileFlags used to open the pipe. - flags fs.FileFlags - - // expectFile is true if a fs.File is expected. - expectFile bool - - // err is the expected error - err error - }{ - { - desc: "FileFlags lacking Read and Write are invalid", - makePipe: false, - flags: fs.FileFlags{}, /* bogus */ - expectFile: false, - err: syscall.EINVAL, - }, - { - desc: "NonBlocking Read only error returns immediately", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Read: true, NonBlocking: true}, - expectFile: false, - err: syscall.ENOENT, - }, - { - desc: "NonBlocking Read only success returns immediately", - makePipe: true, - flags: fs.FileFlags{Read: true, NonBlocking: true}, - expectFile: true, - err: nil, - }, - { - desc: "NonBlocking Write only error returns immediately", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Write: true, NonBlocking: true}, - expectFile: false, - err: syscall.ENOENT, - }, - { - desc: "NonBlocking Write only no reader error returns immediately", - makePipe: true, - flags: fs.FileFlags{Write: true, NonBlocking: true}, - expectFile: false, - err: syscall.ENXIO, - }, - { - desc: "ReadWrite error returns immediately", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Read: true, Write: true}, - expectFile: false, - err: syscall.ENOENT, - }, - { - desc: "ReadWrite returns immediately", - makePipe: true, - flags: fs.FileFlags{Read: true, Write: true}, - expectFile: true, - err: nil, - }, - { - desc: "Blocking Write only returns open error", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Write: true}, - expectFile: false, - err: syscall.ENOENT, /* from bogus perms */ - }, - { - desc: "Blocking Read only returns open error", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Read: true}, - expectFile: false, - err: syscall.ENOENT, - }, - { - desc: "Blocking Write only returns with syserror.ErrWouldBlock", - makePipe: true, - flags: fs.FileFlags{Write: true}, - expectFile: false, - err: syserror.ErrWouldBlock, - }, - { - desc: "Blocking Read only returns with syserror.ErrWouldBlock", - makePipe: true, - flags: fs.FileFlags{Read: true}, - expectFile: false, - err: syserror.ErrWouldBlock, - }, - } { - name := pipename() - if test.makePipe { - // Create the pipe. We do this per-test case to keep tests independent. - if err := mkpipe(name); err != nil { - t.Errorf("%s: failed to make host pipe: %v", test.desc, err) - continue - } - defer syscall.Unlink(name) - } - - // Use a host opener to keep things simple. - opener := &hostOpener{name: name} - - pipeOpenState := &pipeOpenState{} - ctx := contexttest.Context(t) - pipeOps, err := pipeOpenState.TryOpen(ctx, opener, test.flags) - if unwrapError(err) != test.err { - t.Errorf("%s: got error %v, want %v", test.desc, err, test.err) - if pipeOps != nil { - // Cleanup the state of the pipe, and remove the fd from the - // fdnotifier. Sadly this needed to maintain the correctness - // of other tests because the fdnotifier is global. - pipeOps.Release() - } - continue - } - if (pipeOps != nil) != test.expectFile { - t.Errorf("%s: got non-nil file %v, want %v", test.desc, pipeOps != nil, test.expectFile) - } - if pipeOps != nil { - // Same as above. - pipeOps.Release() - } - } -} - -func TestPipeOpenUnblocksEventually(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // partnerIsReader is true if the goroutine opening the same pipe as the test case - // should open the pipe read only. Otherwise write only. This also means that the - // test case will open the pipe in the opposite way. - partnerIsReader bool - - // partnerIsBlocking is true if the goroutine opening the same pipe as the test case - // should do so without the O_NONBLOCK flag, otherwise opens the pipe with O_NONBLOCK - // until ENXIO is not returned. - partnerIsBlocking bool - }{ - { - desc: "Blocking Read with blocking writer partner opens eventually", - partnerIsReader: false, - partnerIsBlocking: true, - }, - { - desc: "Blocking Write with blocking reader partner opens eventually", - partnerIsReader: true, - partnerIsBlocking: true, - }, - { - desc: "Blocking Read with non-blocking writer partner opens eventually", - partnerIsReader: false, - partnerIsBlocking: false, - }, - { - desc: "Blocking Write with non-blocking reader partner opens eventually", - partnerIsReader: true, - partnerIsBlocking: false, - }, - } { - // Create the pipe. We do this per-test case to keep tests independent. - name := pipename() - if err := mkpipe(name); err != nil { - t.Errorf("%s: failed to make host pipe: %v", test.desc, err) - continue - } - defer syscall.Unlink(name) - - // Spawn the partner. - type fderr struct { - fd int - err error - } - errch := make(chan fderr, 1) - go func() { - var flags int - if test.partnerIsReader { - flags = syscall.O_RDONLY - } else { - flags = syscall.O_WRONLY - } - if test.partnerIsBlocking { - fd, err := syscall.Open(name, flags, 0666) - errch <- fderr{fd: fd, err: err} - } else { - var fd int - err := error(syscall.ENXIO) - for err == syscall.ENXIO { - fd, err = syscall.Open(name, flags|syscall.O_NONBLOCK, 0666) - time.Sleep(1 * time.Second) - } - errch <- fderr{fd: fd, err: err} - } - }() - - // Setup file flags for either a read only or write only open. - flags := fs.FileFlags{ - Read: !test.partnerIsReader, - Write: test.partnerIsReader, - } - - // Open the pipe in a blocking way, which should succeed eventually. - opener := &hostOpener{name: name} - ctx := contexttest.Context(t) - pipeOps, err := Open(ctx, opener, flags) - if pipeOps != nil { - // Same as TestTryOpen. - pipeOps.Release() - } - - // Check that the partner opened the file successfully. - e := <-errch - if e.err != nil { - t.Errorf("%s: partner got error %v, wanted nil", test.desc, e.err) - continue - } - // If so, then close the partner fd to avoid leaking an fd. - syscall.Close(e.fd) - - // Check that our blocking open was successful. - if err != nil { - t.Errorf("%s: blocking open got error %v, wanted nil", test.desc, err) - continue - } - if pipeOps == nil { - t.Errorf("%s: blocking open got nil file, wanted non-nil", test.desc) - continue - } - } -} - -func TestCopiedReadAheadBuffer(t *testing.T) { - // Create the pipe. - name := pipename() - if err := mkpipe(name); err != nil { - t.Fatalf("failed to make host pipe: %v", err) - } - defer syscall.Unlink(name) - - // We're taking advantage of the fact that pipes opened read only always return - // success, but internally they are not deemed "opened" until we're sure that - // another writer comes along. This means we can open the same pipe write only - // with no problems + write to it, given that opener.Open already tried to open - // the pipe RDONLY and succeeded, which we know happened if TryOpen returns - // syserror.ErrwouldBlock. - // - // This simulates the open(RDONLY) <-> open(WRONLY)+write race we care about, but - // does not cause our test to be racy (which would be terrible). - opener := &hostOpener{name: name} - pipeOpenState := &pipeOpenState{} - ctx := contexttest.Context(t) - pipeOps, err := pipeOpenState.TryOpen(ctx, opener, fs.FileFlags{Read: true}) - if pipeOps != nil { - pipeOps.Release() - t.Fatalf("open(%s, %o) got file, want nil", name, syscall.O_RDONLY) - } - if err != syserror.ErrWouldBlock { - t.Fatalf("open(%s, %o) got error %v, want %v", name, syscall.O_RDONLY, err, syserror.ErrWouldBlock) - } - - // Then open the same pipe write only and write some bytes to it. The next - // time we try to open the pipe read only again via the pipeOpenState, we should - // succeed and buffer some of the bytes written. - fd, err := syscall.Open(name, syscall.O_WRONLY, 0666) - if err != nil { - t.Fatalf("open(%s, %o) got error %v, want nil", name, syscall.O_WRONLY, err) - } - defer syscall.Close(fd) - - data := []byte("hello") - if n, err := syscall.Write(fd, data); n != len(data) || err != nil { - t.Fatalf("write(%v) got (%d, %v), want (%d, nil)", data, n, err, len(data)) - } - - // Try the read again, knowing that it should succeed this time. - pipeOps, err = pipeOpenState.TryOpen(ctx, opener, fs.FileFlags{Read: true}) - if pipeOps == nil { - t.Fatalf("open(%s, %o) got nil file, want not nil", name, syscall.O_RDONLY) - } - defer pipeOps.Release() - - if err != nil { - t.Fatalf("open(%s, %o) got error %v, want nil", name, syscall.O_RDONLY, err) - } - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - }) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, pipeOps) - - // Check that the file we opened points to a pipe with a non-empty read ahead buffer. - bufsize := len(pipeOps.readAheadBuffer) - if bufsize != 1 { - t.Fatalf("read ahead buffer got %d bytes, want %d", bufsize, 1) - } - - // Now for the final test, try to read everything in, expecting to get back all of - // the bytes that were written at once. Note that in the wild there is no atomic - // read size so expecting to get all bytes from a single writer when there are - // multiple readers is a bad expectation. - buf := make([]byte, len(data)) - ioseq := usermem.BytesIOSequence(buf) - n, err := pipeOps.Read(ctx, file, ioseq, 0) - if err != nil { - t.Fatalf("read request got error %v, want nil", err) - } - if n != int64(len(data)) { - t.Fatalf("read request got %d bytes, want %d", n, len(data)) - } - if !bytes.Equal(buf, data) { - t.Errorf("read request got bytes [%v], want [%v]", buf, data) - } -} - -func TestPipeHangup(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // flags control how we open our end of the pipe and must be read - // only or write only. They also dicate how a coordinating partner - // fd is opened, which is their inverse (read only -> write only, etc). - flags fs.FileFlags - - // hangupSelf if true causes the test case to close our end of the pipe - // and causes hangup errors to be asserted on our coordinating partner's - // fd. If hangupSelf is false, then our partner's fd is closed and the - // hangup errors are expected on our end of the pipe. - hangupSelf bool - }{ - { - desc: "Read only gets hangup error", - flags: fs.FileFlags{Read: true}, - }, - { - desc: "Write only gets hangup error", - flags: fs.FileFlags{Write: true}, - }, - { - desc: "Read only generates hangup error", - flags: fs.FileFlags{Read: true}, - hangupSelf: true, - }, - { - desc: "Write only generates hangup error", - flags: fs.FileFlags{Write: true}, - hangupSelf: true, - }, - } { - if test.flags.Read == test.flags.Write { - t.Errorf("%s: test requires a single reader or writer", test.desc) - continue - } - - // Create the pipe. We do this per-test case to keep tests independent. - name := pipename() - if err := mkpipe(name); err != nil { - t.Errorf("%s: failed to make host pipe: %v", test.desc, err) - continue - } - defer syscall.Unlink(name) - - // Fire off a partner routine which tries to open the same pipe blocking, - // which will synchronize with us. The channel allows us to get back the - // fd once we expect this partner routine to succeed, so we can manifest - // hangup events more directly. - fdchan := make(chan int, 1) - go func() { - // Be explicit about the flags to protect the test from - // misconfiguration. - var flags int - if test.flags.Read { - flags = syscall.O_WRONLY - } else { - flags = syscall.O_RDONLY - } - fd, err := syscall.Open(name, flags, 0666) - if err != nil { - t.Logf("Open(%q, %o, 0666) partner failed: %v", name, flags, err) - } - fdchan <- fd - }() - - // Open our end in a blocking way to ensure that we coordinate. - opener := &hostOpener{name: name} - ctx := contexttest.Context(t) - pipeOps, err := Open(ctx, opener, test.flags) - if err != nil { - t.Errorf("%s: Open got error %v, want nil", test.desc, err) - continue - } - // Don't defer file.DecRef here because that causes the hangup we're - // trying to test for. - - // Expect the partner routine to have coordinated with us and get back - // its open fd. - f := <-fdchan - if f < 0 { - t.Errorf("%s: partner routine got fd %d, want > 0", test.desc, f) - pipeOps.Release() - continue - } - - if test.hangupSelf { - // Hangup self and assert that our partner got the expected hangup - // error. - pipeOps.Release() - - if test.flags.Read { - // Partner is writer. - assertWriterHungup(t, test.desc, fd.NewReadWriter(f)) - } else { - // Partner is reader. - assertReaderHungup(t, test.desc, fd.NewReadWriter(f)) - } - } else { - // Hangup our partner and expect us to get the hangup error. - syscall.Close(f) - defer pipeOps.Release() - - if test.flags.Read { - assertReaderHungup(t, test.desc, pipeOps.(*pipeOperations).file) - } else { - assertWriterHungup(t, test.desc, pipeOps.(*pipeOperations).file) - } - } - } -} - -func assertReaderHungup(t *testing.T, desc string, reader io.Reader) bool { - // Drain the pipe completely, it might have crap in it, but expect EOF eventually. - var err error - for err == nil { - _, err = reader.Read(make([]byte, 10)) - } - if err != io.EOF { - t.Errorf("%s: read from self after hangup got error %v, want %v", desc, err, io.EOF) - return false - } - return true -} - -func assertWriterHungup(t *testing.T, desc string, writer io.Writer) bool { - if _, err := writer.Write([]byte("hello")); unwrapError(err) != syscall.EPIPE { - t.Errorf("%s: write to self after hangup got error %v, want %v", desc, err, syscall.EPIPE) - return false - } - return true -} diff --git a/pkg/sentry/fs/fdpipe/pipe_test.go b/pkg/sentry/fs/fdpipe/pipe_test.go deleted file mode 100644 index 5aff0cc95..000000000 --- a/pkg/sentry/fs/fdpipe/pipe_test.go +++ /dev/null @@ -1,505 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fdpipe - -import ( - "bytes" - "io" - "os" - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -func singlePipeFD() (int, error) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - return -1, err - } - syscall.Close(fds[1]) - return fds[0], nil -} - -func singleDirFD() (int, error) { - return syscall.Open(os.TempDir(), syscall.O_RDONLY, 0666) -} - -func mockPipeDirent(t *testing.T) *fs.Dirent { - ctx := contexttest.Context(t) - node := fs.NewMockInodeOperations(ctx) - node.UAttr = fs.UnstableAttr{ - Perms: fs.FilePermissions{ - User: fs.PermMask{Read: true, Write: true}, - }, - } - inode := fs.NewInode(ctx, node, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - BlockSize: usermem.PageSize, - }) - return fs.NewDirent(ctx, inode, "") -} - -func TestNewPipe(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // getfd generates the fd to pass to newPipeOperations. - getfd func() (int, error) - - // flags are the fs.FileFlags passed to newPipeOperations. - flags fs.FileFlags - - // readAheadBuffer is the buffer passed to newPipeOperations. - readAheadBuffer []byte - - // err is the expected error. - err error - }{ - { - desc: "Cannot make new pipe from bad fd", - getfd: func() (int, error) { return -1, nil }, - err: syscall.EINVAL, - }, - { - desc: "Cannot make new pipe from non-pipe fd", - getfd: singleDirFD, - err: syscall.EINVAL, - }, - { - desc: "Can make new pipe from pipe fd", - getfd: singlePipeFD, - flags: fs.FileFlags{Read: true}, - readAheadBuffer: []byte("hello"), - }, - } { - gfd, err := test.getfd() - if err != nil { - t.Errorf("%s: getfd got (%d, %v), want (fd, nil)", test.desc, gfd, err) - continue - } - f := fd.New(gfd) - - p, err := newPipeOperations(contexttest.Context(t), nil, test.flags, f, test.readAheadBuffer) - if p != nil { - // This is necessary to remove the fd from the global fd notifier. - defer p.Release() - } else { - // If there is no p to DecRef on, because newPipeOperations failed, then the - // file still needs to be closed. - defer f.Close() - } - - if err != test.err { - t.Errorf("%s: got error %v, want %v", test.desc, err, test.err) - continue - } - // Check the state of the pipe given that it was successfully opened. - if err == nil { - if p == nil { - t.Errorf("%s: got nil pipe and nil error, want (pipe, nil)", test.desc) - continue - } - if flags := p.flags; test.flags != flags { - t.Errorf("%s: got file flags %s, want %s", test.desc, flags, test.flags) - continue - } - if len(test.readAheadBuffer) != len(p.readAheadBuffer) { - t.Errorf("%s: got read ahead buffer length %d, want %d", test.desc, len(p.readAheadBuffer), len(test.readAheadBuffer)) - continue - } - fileFlags, _, errno := syscall.Syscall(syscall.SYS_FCNTL, uintptr(p.file.FD()), syscall.F_GETFL, 0) - if errno != 0 { - t.Errorf("%s: failed to get file flags for fd %d, got %v, want 0", test.desc, p.file.FD(), errno) - continue - } - if fileFlags&syscall.O_NONBLOCK == 0 { - t.Errorf("%s: pipe is blocking, expected non-blocking", test.desc) - continue - } - if !fdnotifier.HasFD(int32(f.FD())) { - t.Errorf("%s: pipe fd %d is not registered for events", test.desc, f.FD) - } - } - } -} - -func TestPipeDestruction(t *testing.T) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Fatalf("failed to create pipes: got %v, want nil", err) - } - f := fd.New(fds[0]) - - // We don't care about the other end, just use the read end. - syscall.Close(fds[1]) - - // Test the read end, but it doesn't really matter which. - p, err := newPipeOperations(contexttest.Context(t), nil, fs.FileFlags{Read: true}, f, nil) - if err != nil { - f.Close() - t.Fatalf("newPipeOperations got error %v, want nil", err) - } - // Drop our only reference, which should trigger the destructor. - p.Release() - - if fdnotifier.HasFD(int32(fds[0])) { - t.Fatalf("after DecRef fdnotifier has fd %d, want no longer registered", fds[0]) - } - if p.file != nil { - t.Errorf("after DecRef got file, want nil") - } -} - -type Seek struct{} - -type ReadDir struct{} - -type Writev struct { - Src usermem.IOSequence -} - -type Readv struct { - Dst usermem.IOSequence -} - -type Fsync struct{} - -func TestPipeRequest(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // request to execute. - context interface{} - - // flags determines whether to use the read or write end - // of the pipe, for this test it can only be Read or Write. - flags fs.FileFlags - - // keepOpenPartner if false closes the other end of the pipe, - // otherwise this is delayed until the end of the test. - keepOpenPartner bool - - // expected error - err error - }{ - { - desc: "ReadDir on pipe returns ENOTDIR", - context: &ReadDir{}, - err: syscall.ENOTDIR, - }, - { - desc: "Fsync on pipe returns EINVAL", - context: &Fsync{}, - err: syscall.EINVAL, - }, - { - desc: "Seek on pipe returns ESPIPE", - context: &Seek{}, - err: syscall.ESPIPE, - }, - { - desc: "Readv on pipe from empty buffer returns nil", - context: &Readv{Dst: usermem.BytesIOSequence(nil)}, - flags: fs.FileFlags{Read: true}, - }, - { - desc: "Readv on pipe from non-empty buffer and closed partner returns EOF", - context: &Readv{Dst: usermem.BytesIOSequence(make([]byte, 10))}, - flags: fs.FileFlags{Read: true}, - err: io.EOF, - }, - { - desc: "Readv on pipe from non-empty buffer and open partner returns EWOULDBLOCK", - context: &Readv{Dst: usermem.BytesIOSequence(make([]byte, 10))}, - flags: fs.FileFlags{Read: true}, - keepOpenPartner: true, - err: syserror.ErrWouldBlock, - }, - { - desc: "Writev on pipe from empty buffer returns nil", - context: &Writev{Src: usermem.BytesIOSequence(nil)}, - flags: fs.FileFlags{Write: true}, - }, - { - desc: "Writev on pipe from non-empty buffer and closed partner returns EPIPE", - context: &Writev{Src: usermem.BytesIOSequence([]byte("hello"))}, - flags: fs.FileFlags{Write: true}, - err: syscall.EPIPE, - }, - { - desc: "Writev on pipe from non-empty buffer and open partner succeeds", - context: &Writev{Src: usermem.BytesIOSequence([]byte("hello"))}, - flags: fs.FileFlags{Write: true}, - keepOpenPartner: true, - }, - } { - if test.flags.Read && test.flags.Write { - panic("both read and write not supported for this test") - } - - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Errorf("%s: failed to create pipes: got %v, want nil", test.desc, err) - continue - } - - // Configure the fd and partner fd based on the file flags. - testFd, partnerFd := fds[0], fds[1] - if test.flags.Write { - testFd, partnerFd = fds[1], fds[0] - } - - // Configure closing the fds. - if test.keepOpenPartner { - defer syscall.Close(partnerFd) - } else { - syscall.Close(partnerFd) - } - - // Create the pipe. - ctx := contexttest.Context(t) - p, err := newPipeOperations(ctx, nil, test.flags, fd.New(testFd), nil) - if err != nil { - t.Fatalf("%s: newPipeOperations got error %v, want nil", test.desc, err) - } - defer p.Release() - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{Type: fs.Pipe}) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) - - // Issue request via the appropriate function. - switch c := test.context.(type) { - case *Seek: - _, err = p.Seek(ctx, file, 0, 0) - case *ReadDir: - _, err = p.Readdir(ctx, file, nil) - case *Readv: - _, err = p.Read(ctx, file, c.Dst, 0) - case *Writev: - _, err = p.Write(ctx, file, c.Src, 0) - case *Fsync: - err = p.Fsync(ctx, file, 0, fs.FileMaxOffset, fs.SyncAll) - default: - t.Errorf("%s: unknown request type %T", test.desc, test.context) - } - - if unwrapError(err) != test.err { - t.Errorf("%s: got error %v, want %v", test.desc, err, test.err) - } - } -} - -func TestPipeReadAheadBuffer(t *testing.T) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Fatalf("failed to create pipes: got %v, want nil", err) - } - rfile := fd.New(fds[0]) - - // Eventually close the write end, which is not wrapped in a pipe object. - defer syscall.Close(fds[1]) - - // Write some bytes to this end. - data := []byte("world") - if n, err := syscall.Write(fds[1], data); n != len(data) || err != nil { - rfile.Close() - t.Fatalf("write to pipe got (%d, %v), want (%d, nil)", n, err, len(data)) - } - // Close the write end immediately, we don't care about it. - - buffered := []byte("hello ") - ctx := contexttest.Context(t) - p, err := newPipeOperations(ctx, nil, fs.FileFlags{Read: true}, rfile, buffered) - if err != nil { - rfile.Close() - t.Fatalf("newPipeOperations got error %v, want nil", err) - } - defer p.Release() - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - }) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) - - // In total we expect to read data + buffered. - total := append(buffered, data...) - - buf := make([]byte, len(total)) - iov := usermem.BytesIOSequence(buf) - n, err := p.Read(contexttest.Context(t), file, iov, 0) - if err != nil { - t.Fatalf("read request got error %v, want nil", err) - } - if n != int64(len(total)) { - t.Fatalf("read request got %d bytes, want %d", n, len(total)) - } - if !bytes.Equal(buf, total) { - t.Errorf("read request got bytes [%v], want [%v]", buf, total) - } -} - -// This is very important for pipes in general because they can return -// EWOULDBLOCK and for those that block they must continue until they have read -// all of the data (and report it as such). -func TestPipeReadsAccumulate(t *testing.T) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Fatalf("failed to create pipes: got %v, want nil", err) - } - rfile := fd.New(fds[0]) - - // Eventually close the write end, it doesn't depend on a pipe object. - defer syscall.Close(fds[1]) - - // Get a new read only pipe reference. - ctx := contexttest.Context(t) - p, err := newPipeOperations(ctx, nil, fs.FileFlags{Read: true}, rfile, nil) - if err != nil { - rfile.Close() - t.Fatalf("newPipeOperations got error %v, want nil", err) - } - // Don't forget to remove the fd from the fd notifier. Otherwise other tests will - // likely be borked, because it's global :( - defer p.Release() - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - }) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) - - // Write some some bytes to the pipe. - data := []byte("some message") - if n, err := syscall.Write(fds[1], data); n != len(data) || err != nil { - t.Fatalf("write to pipe got (%d, %v), want (%d, nil)", n, err, len(data)) - } - - // Construct a segment vec that is a bit more than we have written so we - // trigger an EWOULDBLOCK. - wantBytes := len(data) + 1 - readBuffer := make([]byte, wantBytes) - iov := usermem.BytesIOSequence(readBuffer) - n, err := p.Read(ctx, file, iov, 0) - total := n - iov = iov.DropFirst64(n) - if err != syserror.ErrWouldBlock { - t.Fatalf("Readv got error %v, want %v", err, syserror.ErrWouldBlock) - } - - // Write a few more bytes to allow us to read more/accumulate. - extra := []byte("extra") - if n, err := syscall.Write(fds[1], extra); n != len(extra) || err != nil { - t.Fatalf("write to pipe got (%d, %v), want (%d, nil)", n, err, len(extra)) - } - - // This time, using the same request, we should not block. - n, err = p.Read(ctx, file, iov, 0) - total += n - if err != nil { - t.Fatalf("Readv got error %v, want nil", err) - } - - // Assert that the result we got back is cumulative. - if total != int64(wantBytes) { - t.Fatalf("Readv sequence got %d bytes, want %d", total, wantBytes) - } - - if want := append(data, extra[0]); !bytes.Equal(readBuffer, want) { - t.Errorf("Readv sequence got %v, want %v", readBuffer, want) - } -} - -// Same as TestReadsAccumulate. -func TestPipeWritesAccumulate(t *testing.T) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Fatalf("failed to create pipes: got %v, want nil", err) - } - wfile := fd.New(fds[1]) - - // Eventually close the read end, it doesn't depend on a pipe object. - defer syscall.Close(fds[0]) - - // Get a new write only pipe reference. - ctx := contexttest.Context(t) - p, err := newPipeOperations(ctx, nil, fs.FileFlags{Write: true}, wfile, nil) - if err != nil { - wfile.Close() - t.Fatalf("newPipeOperations got error %v, want nil", err) - } - // Don't forget to remove the fd from the fd notifier. Otherwise other tests - // will likely be borked, because it's global :( - defer p.Release() - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - }) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) - - pipeSize, _, errno := syscall.Syscall(syscall.SYS_FCNTL, uintptr(wfile.FD()), syscall.F_GETPIPE_SZ, 0) - if errno != 0 { - t.Fatalf("fcntl(F_GETPIPE_SZ) failed: %v", errno) - } - t.Logf("Pipe buffer size: %d", pipeSize) - - // Construct a segment vec that is larger than the pipe size to trigger an - // EWOULDBLOCK. - wantBytes := int(pipeSize) * 2 - writeBuffer := make([]byte, wantBytes) - for i := 0; i < wantBytes; i++ { - writeBuffer[i] = 'a' - } - iov := usermem.BytesIOSequence(writeBuffer) - n, err := p.Write(ctx, file, iov, 0) - if err != syserror.ErrWouldBlock { - t.Fatalf("Writev got error %v, want %v", err, syserror.ErrWouldBlock) - } - if n != int64(pipeSize) { - t.Fatalf("Writev partial write, got: %v, want %v", n, pipeSize) - } - total := n - iov = iov.DropFirst64(n) - - // Read the entire pipe buf size to make space for the second half. - readBuffer := make([]byte, n) - if n, err := syscall.Read(fds[0], readBuffer); n != len(readBuffer) || err != nil { - t.Fatalf("write to pipe got (%d, %v), want (%d, nil)", n, err, len(readBuffer)) - } - if !bytes.Equal(readBuffer, writeBuffer[:len(readBuffer)]) { - t.Fatalf("wrong data read from pipe, got: %v, want: %v", readBuffer, writeBuffer) - } - - // This time we should not block. - n, err = p.Write(ctx, file, iov, 0) - if err != nil { - t.Fatalf("Writev got error %v, want nil", err) - } - if n != int64(pipeSize) { - t.Fatalf("Writev partial write, got: %v, want %v", n, pipeSize) - } - total += n - - // Assert that the result we got back is cumulative. - if total != int64(wantBytes) { - t.Fatalf("Writev sequence got %d bytes, want %d", total, wantBytes) - } -} diff --git a/pkg/sentry/fs/file_overlay_test.go b/pkg/sentry/fs/file_overlay_test.go deleted file mode 100644 index 1971cc680..000000000 --- a/pkg/sentry/fs/file_overlay_test.go +++ /dev/null @@ -1,192 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs_test - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" -) - -func TestReaddir(t *testing.T) { - ctx := contexttest.Context(t) - ctx = &rootContext{ - Context: ctx, - root: fs.NewDirent(ctx, newTestRamfsDir(ctx, nil, nil), "root"), - } - for _, test := range []struct { - // Test description. - desc string - - // Lookup parameters. - dir *fs.Inode - - // Want from lookup. - err error - names []string - }{ - { - desc: "no upper, lower has entries", - dir: fs.NewTestOverlayDir(ctx, - nil, /* upper */ - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - {name: "b"}, - }, nil), /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "b"}, - }, - { - desc: "upper has entries, no lower", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - {name: "b"}, - }, nil), /* upper */ - nil, /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "b"}, - }, - { - desc: "upper and lower, entries combine", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - {name: "b"}, - }, nil), /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "b"}, - }, - { - desc: "upper and lower, entries combine, none are masked", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - }, []string{"b"}), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - {name: "c"}, - }, nil), /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "c"}, - }, - { - desc: "upper and lower, entries combine, upper masks some of lower", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - }, []string{"b"}), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - {name: "b"}, /* will be masked */ - {name: "c"}, - }, nil), /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "c"}, - }, - } { - t.Run(test.desc, func(t *testing.T) { - openDir, err := test.dir.GetFile(ctx, fs.NewDirent(ctx, test.dir, "stub"), fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("GetFile got error %v, want nil", err) - } - stubSerializer := &fs.CollectEntriesSerializer{} - err = openDir.Readdir(ctx, stubSerializer) - if err != test.err { - t.Fatalf("Readdir got error %v, want nil", err) - } - if err != nil { - return - } - if !reflect.DeepEqual(stubSerializer.Order, test.names) { - t.Errorf("Readdir got names %v, want %v", stubSerializer.Order, test.names) - } - }) - } -} - -func TestReaddirRevalidation(t *testing.T) { - ctx := contexttest.Context(t) - ctx = &rootContext{ - Context: ctx, - root: fs.NewDirent(ctx, newTestRamfsDir(ctx, nil, nil), "root"), - } - - // Create an overlay with two directories, each with one file. - upper := newTestRamfsDir(ctx, []dirContent{{name: "a"}}, nil) - lower := newTestRamfsDir(ctx, []dirContent{{name: "b"}}, nil) - overlay := fs.NewTestOverlayDir(ctx, upper, lower, true /* revalidate */) - - // Get a handle to the dirent in the upper filesystem so that we can - // modify it without going through the dirent. - upperDir := upper.InodeOperations.(*dir).InodeOperations.(*ramfs.Dir) - - // Check that overlay returns the files from both upper and lower. - openDir, err := overlay.GetFile(ctx, fs.NewDirent(ctx, overlay, "stub"), fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("GetFile got error %v, want nil", err) - } - ser := &fs.CollectEntriesSerializer{} - if err := openDir.Readdir(ctx, ser); err != nil { - t.Fatalf("Readdir got error %v, want nil", err) - } - got, want := ser.Order, []string{".", "..", "a", "b"} - if !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got names %v, want %v", got, want) - } - - // Remove "a" from the upper and add "c". - if err := upperDir.Remove(ctx, upper, "a"); err != nil { - t.Fatalf("error removing child: %v", err) - } - upperDir.AddChild(ctx, "c", fs.NewInode(ctx, fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermissions{}, 0), - upper.MountSource, fs.StableAttr{Type: fs.RegularFile})) - - // Seek to beginning of the directory and do the readdir again. - if _, err := openDir.Seek(ctx, fs.SeekSet, 0); err != nil { - t.Fatalf("error seeking to beginning of dir: %v", err) - } - ser = &fs.CollectEntriesSerializer{} - if err := openDir.Readdir(ctx, ser); err != nil { - t.Fatalf("Readdir got error %v, want nil", err) - } - - // Readdir should return the updated children. - got, want = ser.Order, []string{".", "..", "b", "c"} - if !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got names %v, want %v", got, want) - } -} - -type rootContext struct { - context.Context - root *fs.Dirent -} - -// Value implements context.Context. -func (r *rootContext) Value(key interface{}) interface{} { - switch key { - case fs.CtxRoot: - r.root.IncRef() - return r.root - default: - return r.Context.Value(key) - } -} diff --git a/pkg/sentry/fs/filetest/BUILD b/pkg/sentry/fs/filetest/BUILD deleted file mode 100644 index a8000e010..000000000 --- a/pkg/sentry/fs/filetest/BUILD +++ /dev/null @@ -1,19 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "filetest", - testonly = 1, - srcs = ["filetest.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/fs/filetest/filetest.go b/pkg/sentry/fs/filetest/filetest.go deleted file mode 100644 index 8049538f2..000000000 --- a/pkg/sentry/fs/filetest/filetest.go +++ /dev/null @@ -1,61 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package filetest provides a test implementation of an fs.File. -package filetest - -import ( - "fmt" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/anon" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -// TestFileOperations is an implementation of the File interface. It provides all -// required methods. -type TestFileOperations struct { - fsutil.FileNoopRelease `state:"nosave"` - fsutil.FilePipeSeek `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FileNoFsync `state:"nosave"` - fsutil.FileNoopFlush `state:"nosave"` - fsutil.FileNoMMap `state:"nosave"` - fsutil.FileNoIoctl `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - waiter.AlwaysReady `state:"nosave"` -} - -// NewTestFile creates and initializes a new test file. -func NewTestFile(tb testing.TB) *fs.File { - ctx := contexttest.Context(tb) - dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "test") - return fs.NewFile(ctx, dirent, fs.FileFlags{}, &TestFileOperations{}) -} - -// Read just fails the request. -func (*TestFileOperations) Read(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { - return 0, fmt.Errorf("Readv not implemented") -} - -// Write just fails the request. -func (*TestFileOperations) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { - return 0, fmt.Errorf("Writev not implemented") -} diff --git a/pkg/sentry/fs/fs_state_autogen.go b/pkg/sentry/fs/fs_state_autogen.go new file mode 100755 index 000000000..f6d6286c2 --- /dev/null +++ b/pkg/sentry/fs/fs_state_autogen.go @@ -0,0 +1,638 @@ +// automatically generated by stateify. + +package fs + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *StableAttr) beforeSave() {} +func (x *StableAttr) save(m state.Map) { + x.beforeSave() + m.Save("Type", &x.Type) + m.Save("DeviceID", &x.DeviceID) + m.Save("InodeID", &x.InodeID) + m.Save("BlockSize", &x.BlockSize) + m.Save("DeviceFileMajor", &x.DeviceFileMajor) + m.Save("DeviceFileMinor", &x.DeviceFileMinor) +} + +func (x *StableAttr) afterLoad() {} +func (x *StableAttr) load(m state.Map) { + m.Load("Type", &x.Type) + m.Load("DeviceID", &x.DeviceID) + m.Load("InodeID", &x.InodeID) + m.Load("BlockSize", &x.BlockSize) + m.Load("DeviceFileMajor", &x.DeviceFileMajor) + m.Load("DeviceFileMinor", &x.DeviceFileMinor) +} + +func (x *UnstableAttr) beforeSave() {} +func (x *UnstableAttr) save(m state.Map) { + x.beforeSave() + m.Save("Size", &x.Size) + m.Save("Usage", &x.Usage) + m.Save("Perms", &x.Perms) + m.Save("Owner", &x.Owner) + m.Save("AccessTime", &x.AccessTime) + m.Save("ModificationTime", &x.ModificationTime) + m.Save("StatusChangeTime", &x.StatusChangeTime) + m.Save("Links", &x.Links) +} + +func (x *UnstableAttr) afterLoad() {} +func (x *UnstableAttr) load(m state.Map) { + m.Load("Size", &x.Size) + m.Load("Usage", &x.Usage) + m.Load("Perms", &x.Perms) + m.Load("Owner", &x.Owner) + m.Load("AccessTime", &x.AccessTime) + m.Load("ModificationTime", &x.ModificationTime) + m.Load("StatusChangeTime", &x.StatusChangeTime) + m.Load("Links", &x.Links) +} + +func (x *AttrMask) beforeSave() {} +func (x *AttrMask) save(m state.Map) { + x.beforeSave() + m.Save("Type", &x.Type) + m.Save("DeviceID", &x.DeviceID) + m.Save("InodeID", &x.InodeID) + m.Save("BlockSize", &x.BlockSize) + m.Save("Size", &x.Size) + m.Save("Usage", &x.Usage) + m.Save("Perms", &x.Perms) + m.Save("UID", &x.UID) + m.Save("GID", &x.GID) + m.Save("AccessTime", &x.AccessTime) + m.Save("ModificationTime", &x.ModificationTime) + m.Save("StatusChangeTime", &x.StatusChangeTime) + m.Save("Links", &x.Links) +} + +func (x *AttrMask) afterLoad() {} +func (x *AttrMask) load(m state.Map) { + m.Load("Type", &x.Type) + m.Load("DeviceID", &x.DeviceID) + m.Load("InodeID", &x.InodeID) + m.Load("BlockSize", &x.BlockSize) + m.Load("Size", &x.Size) + m.Load("Usage", &x.Usage) + m.Load("Perms", &x.Perms) + m.Load("UID", &x.UID) + m.Load("GID", &x.GID) + m.Load("AccessTime", &x.AccessTime) + m.Load("ModificationTime", &x.ModificationTime) + m.Load("StatusChangeTime", &x.StatusChangeTime) + m.Load("Links", &x.Links) +} + +func (x *PermMask) beforeSave() {} +func (x *PermMask) save(m state.Map) { + x.beforeSave() + m.Save("Read", &x.Read) + m.Save("Write", &x.Write) + m.Save("Execute", &x.Execute) +} + +func (x *PermMask) afterLoad() {} +func (x *PermMask) load(m state.Map) { + m.Load("Read", &x.Read) + m.Load("Write", &x.Write) + m.Load("Execute", &x.Execute) +} + +func (x *FilePermissions) beforeSave() {} +func (x *FilePermissions) save(m state.Map) { + x.beforeSave() + m.Save("User", &x.User) + m.Save("Group", &x.Group) + m.Save("Other", &x.Other) + m.Save("Sticky", &x.Sticky) + m.Save("SetUID", &x.SetUID) + m.Save("SetGID", &x.SetGID) +} + +func (x *FilePermissions) afterLoad() {} +func (x *FilePermissions) load(m state.Map) { + m.Load("User", &x.User) + m.Load("Group", &x.Group) + m.Load("Other", &x.Other) + m.Load("Sticky", &x.Sticky) + m.Load("SetUID", &x.SetUID) + m.Load("SetGID", &x.SetGID) +} + +func (x *FileOwner) beforeSave() {} +func (x *FileOwner) save(m state.Map) { + x.beforeSave() + m.Save("UID", &x.UID) + m.Save("GID", &x.GID) +} + +func (x *FileOwner) afterLoad() {} +func (x *FileOwner) load(m state.Map) { + m.Load("UID", &x.UID) + m.Load("GID", &x.GID) +} + +func (x *DentAttr) beforeSave() {} +func (x *DentAttr) save(m state.Map) { + x.beforeSave() + m.Save("Type", &x.Type) + m.Save("InodeID", &x.InodeID) +} + +func (x *DentAttr) afterLoad() {} +func (x *DentAttr) load(m state.Map) { + m.Load("Type", &x.Type) + m.Load("InodeID", &x.InodeID) +} + +func (x *SortedDentryMap) beforeSave() {} +func (x *SortedDentryMap) save(m state.Map) { + x.beforeSave() + m.Save("names", &x.names) + m.Save("entries", &x.entries) +} + +func (x *SortedDentryMap) afterLoad() {} +func (x *SortedDentryMap) load(m state.Map) { + m.Load("names", &x.names) + m.Load("entries", &x.entries) +} + +func (x *Dirent) save(m state.Map) { + x.beforeSave() + var children map[string]*Dirent = x.saveChildren() + m.SaveValue("children", children) + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("userVisible", &x.userVisible) + m.Save("Inode", &x.Inode) + m.Save("name", &x.name) + m.Save("parent", &x.parent) + m.Save("deleted", &x.deleted) + m.Save("mounted", &x.mounted) +} + +func (x *Dirent) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("userVisible", &x.userVisible) + m.Load("Inode", &x.Inode) + m.Load("name", &x.name) + m.Load("parent", &x.parent) + m.Load("deleted", &x.deleted) + m.Load("mounted", &x.mounted) + m.LoadValue("children", new(map[string]*Dirent), func(y interface{}) { x.loadChildren(y.(map[string]*Dirent)) }) + m.AfterLoad(x.afterLoad) +} + +func (x *DirentCache) beforeSave() {} +func (x *DirentCache) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.currentSize) { + m.Failf("currentSize is %v, expected zero", x.currentSize) + } + if !state.IsZeroValue(x.list) { + m.Failf("list is %v, expected zero", x.list) + } + m.Save("maxSize", &x.maxSize) + m.Save("limit", &x.limit) +} + +func (x *DirentCache) afterLoad() {} +func (x *DirentCache) load(m state.Map) { + m.Load("maxSize", &x.maxSize) + m.Load("limit", &x.limit) +} + +func (x *DirentCacheLimiter) beforeSave() {} +func (x *DirentCacheLimiter) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.count) { + m.Failf("count is %v, expected zero", x.count) + } + m.Save("max", &x.max) +} + +func (x *DirentCacheLimiter) afterLoad() {} +func (x *DirentCacheLimiter) load(m state.Map) { + m.Load("max", &x.max) +} + +func (x *direntList) beforeSave() {} +func (x *direntList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *direntList) afterLoad() {} +func (x *direntList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *direntEntry) beforeSave() {} +func (x *direntEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *direntEntry) afterLoad() {} +func (x *direntEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *eventList) beforeSave() {} +func (x *eventList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *eventList) afterLoad() {} +func (x *eventList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *eventEntry) beforeSave() {} +func (x *eventEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *eventEntry) afterLoad() {} +func (x *eventEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *File) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("UniqueID", &x.UniqueID) + m.Save("Dirent", &x.Dirent) + m.Save("flags", &x.flags) + m.Save("async", &x.async) + m.Save("FileOperations", &x.FileOperations) + m.Save("offset", &x.offset) +} + +func (x *File) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("UniqueID", &x.UniqueID) + m.Load("Dirent", &x.Dirent) + m.Load("flags", &x.flags) + m.Load("async", &x.async) + m.LoadWait("FileOperations", &x.FileOperations) + m.Load("offset", &x.offset) + m.AfterLoad(x.afterLoad) +} + +func (x *overlayFileOperations) beforeSave() {} +func (x *overlayFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("upper", &x.upper) + m.Save("lower", &x.lower) + m.Save("dirCursor", &x.dirCursor) +} + +func (x *overlayFileOperations) afterLoad() {} +func (x *overlayFileOperations) load(m state.Map) { + m.Load("upper", &x.upper) + m.Load("lower", &x.lower) + m.Load("dirCursor", &x.dirCursor) +} + +func (x *overlayMappingIdentity) beforeSave() {} +func (x *overlayMappingIdentity) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("id", &x.id) + m.Save("overlayFile", &x.overlayFile) +} + +func (x *overlayMappingIdentity) afterLoad() {} +func (x *overlayMappingIdentity) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("id", &x.id) + m.Load("overlayFile", &x.overlayFile) +} + +func (x *MountSourceFlags) beforeSave() {} +func (x *MountSourceFlags) save(m state.Map) { + x.beforeSave() + m.Save("ReadOnly", &x.ReadOnly) + m.Save("NoAtime", &x.NoAtime) + m.Save("ForcePageCache", &x.ForcePageCache) + m.Save("NoExec", &x.NoExec) +} + +func (x *MountSourceFlags) afterLoad() {} +func (x *MountSourceFlags) load(m state.Map) { + m.Load("ReadOnly", &x.ReadOnly) + m.Load("NoAtime", &x.NoAtime) + m.Load("ForcePageCache", &x.ForcePageCache) + m.Load("NoExec", &x.NoExec) +} + +func (x *FileFlags) beforeSave() {} +func (x *FileFlags) save(m state.Map) { + x.beforeSave() + m.Save("Direct", &x.Direct) + m.Save("NonBlocking", &x.NonBlocking) + m.Save("DSync", &x.DSync) + m.Save("Sync", &x.Sync) + m.Save("Append", &x.Append) + m.Save("Read", &x.Read) + m.Save("Write", &x.Write) + m.Save("Pread", &x.Pread) + m.Save("Pwrite", &x.Pwrite) + m.Save("Directory", &x.Directory) + m.Save("Async", &x.Async) + m.Save("LargeFile", &x.LargeFile) + m.Save("NonSeekable", &x.NonSeekable) + m.Save("Truncate", &x.Truncate) +} + +func (x *FileFlags) afterLoad() {} +func (x *FileFlags) load(m state.Map) { + m.Load("Direct", &x.Direct) + m.Load("NonBlocking", &x.NonBlocking) + m.Load("DSync", &x.DSync) + m.Load("Sync", &x.Sync) + m.Load("Append", &x.Append) + m.Load("Read", &x.Read) + m.Load("Write", &x.Write) + m.Load("Pread", &x.Pread) + m.Load("Pwrite", &x.Pwrite) + m.Load("Directory", &x.Directory) + m.Load("Async", &x.Async) + m.Load("LargeFile", &x.LargeFile) + m.Load("NonSeekable", &x.NonSeekable) + m.Load("Truncate", &x.Truncate) +} + +func (x *Inode) beforeSave() {} +func (x *Inode) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("InodeOperations", &x.InodeOperations) + m.Save("StableAttr", &x.StableAttr) + m.Save("LockCtx", &x.LockCtx) + m.Save("Watches", &x.Watches) + m.Save("MountSource", &x.MountSource) + m.Save("overlay", &x.overlay) +} + +func (x *Inode) afterLoad() {} +func (x *Inode) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("InodeOperations", &x.InodeOperations) + m.Load("StableAttr", &x.StableAttr) + m.Load("LockCtx", &x.LockCtx) + m.Load("Watches", &x.Watches) + m.Load("MountSource", &x.MountSource) + m.Load("overlay", &x.overlay) +} + +func (x *LockCtx) beforeSave() {} +func (x *LockCtx) save(m state.Map) { + x.beforeSave() + m.Save("Posix", &x.Posix) + m.Save("BSD", &x.BSD) +} + +func (x *LockCtx) afterLoad() {} +func (x *LockCtx) load(m state.Map) { + m.Load("Posix", &x.Posix) + m.Load("BSD", &x.BSD) +} + +func (x *Watches) beforeSave() {} +func (x *Watches) save(m state.Map) { + x.beforeSave() + m.Save("ws", &x.ws) + m.Save("unlinked", &x.unlinked) +} + +func (x *Watches) afterLoad() {} +func (x *Watches) load(m state.Map) { + m.Load("ws", &x.ws) + m.Load("unlinked", &x.unlinked) +} + +func (x *Inotify) beforeSave() {} +func (x *Inotify) save(m state.Map) { + x.beforeSave() + m.Save("id", &x.id) + m.Save("events", &x.events) + m.Save("scratch", &x.scratch) + m.Save("nextWatch", &x.nextWatch) + m.Save("watches", &x.watches) +} + +func (x *Inotify) afterLoad() {} +func (x *Inotify) load(m state.Map) { + m.Load("id", &x.id) + m.Load("events", &x.events) + m.Load("scratch", &x.scratch) + m.Load("nextWatch", &x.nextWatch) + m.Load("watches", &x.watches) +} + +func (x *Event) beforeSave() {} +func (x *Event) save(m state.Map) { + x.beforeSave() + m.Save("eventEntry", &x.eventEntry) + m.Save("wd", &x.wd) + m.Save("mask", &x.mask) + m.Save("cookie", &x.cookie) + m.Save("len", &x.len) + m.Save("name", &x.name) +} + +func (x *Event) afterLoad() {} +func (x *Event) load(m state.Map) { + m.Load("eventEntry", &x.eventEntry) + m.Load("wd", &x.wd) + m.Load("mask", &x.mask) + m.Load("cookie", &x.cookie) + m.Load("len", &x.len) + m.Load("name", &x.name) +} + +func (x *Watch) beforeSave() {} +func (x *Watch) save(m state.Map) { + x.beforeSave() + m.Save("owner", &x.owner) + m.Save("wd", &x.wd) + m.Save("target", &x.target) + m.Save("unpinned", &x.unpinned) + m.Save("mask", &x.mask) + m.Save("pins", &x.pins) +} + +func (x *Watch) afterLoad() {} +func (x *Watch) load(m state.Map) { + m.Load("owner", &x.owner) + m.Load("wd", &x.wd) + m.Load("target", &x.target) + m.Load("unpinned", &x.unpinned) + m.Load("mask", &x.mask) + m.Load("pins", &x.pins) +} + +func (x *MountSource) beforeSave() {} +func (x *MountSource) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("MountSourceOperations", &x.MountSourceOperations) + m.Save("FilesystemType", &x.FilesystemType) + m.Save("Flags", &x.Flags) + m.Save("fscache", &x.fscache) + m.Save("direntRefs", &x.direntRefs) +} + +func (x *MountSource) afterLoad() {} +func (x *MountSource) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("MountSourceOperations", &x.MountSourceOperations) + m.Load("FilesystemType", &x.FilesystemType) + m.Load("Flags", &x.Flags) + m.Load("fscache", &x.fscache) + m.Load("direntRefs", &x.direntRefs) +} + +func (x *SimpleMountSourceOperations) beforeSave() {} +func (x *SimpleMountSourceOperations) save(m state.Map) { + x.beforeSave() + m.Save("keep", &x.keep) + m.Save("revalidate", &x.revalidate) + m.Save("cacheReaddir", &x.cacheReaddir) +} + +func (x *SimpleMountSourceOperations) afterLoad() {} +func (x *SimpleMountSourceOperations) load(m state.Map) { + m.Load("keep", &x.keep) + m.Load("revalidate", &x.revalidate) + m.Load("cacheReaddir", &x.cacheReaddir) +} + +func (x *overlayMountSourceOperations) beforeSave() {} +func (x *overlayMountSourceOperations) save(m state.Map) { + x.beforeSave() + m.Save("upper", &x.upper) + m.Save("lower", &x.lower) +} + +func (x *overlayMountSourceOperations) afterLoad() {} +func (x *overlayMountSourceOperations) load(m state.Map) { + m.Load("upper", &x.upper) + m.Load("lower", &x.lower) +} + +func (x *overlayFilesystem) beforeSave() {} +func (x *overlayFilesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *overlayFilesystem) afterLoad() {} +func (x *overlayFilesystem) load(m state.Map) { +} + +func (x *Mount) beforeSave() {} +func (x *Mount) save(m state.Map) { + x.beforeSave() + m.Save("ID", &x.ID) + m.Save("ParentID", &x.ParentID) + m.Save("root", &x.root) + m.Save("previous", &x.previous) +} + +func (x *Mount) afterLoad() {} +func (x *Mount) load(m state.Map) { + m.Load("ID", &x.ID) + m.Load("ParentID", &x.ParentID) + m.Load("root", &x.root) + m.Load("previous", &x.previous) +} + +func (x *MountNamespace) beforeSave() {} +func (x *MountNamespace) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("userns", &x.userns) + m.Save("root", &x.root) + m.Save("mounts", &x.mounts) + m.Save("mountID", &x.mountID) +} + +func (x *MountNamespace) afterLoad() {} +func (x *MountNamespace) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("userns", &x.userns) + m.Load("root", &x.root) + m.Load("mounts", &x.mounts) + m.Load("mountID", &x.mountID) +} + +func (x *overlayEntry) beforeSave() {} +func (x *overlayEntry) save(m state.Map) { + x.beforeSave() + m.Save("lowerExists", &x.lowerExists) + m.Save("lower", &x.lower) + m.Save("mappings", &x.mappings) + m.Save("upper", &x.upper) + m.Save("dirCache", &x.dirCache) +} + +func (x *overlayEntry) afterLoad() {} +func (x *overlayEntry) load(m state.Map) { + m.Load("lowerExists", &x.lowerExists) + m.Load("lower", &x.lower) + m.Load("mappings", &x.mappings) + m.Load("upper", &x.upper) + m.Load("dirCache", &x.dirCache) +} + +func init() { + state.Register("pkg/sentry/fs.StableAttr", (*StableAttr)(nil), state.Fns{Save: (*StableAttr).save, Load: (*StableAttr).load}) + state.Register("pkg/sentry/fs.UnstableAttr", (*UnstableAttr)(nil), state.Fns{Save: (*UnstableAttr).save, Load: (*UnstableAttr).load}) + state.Register("pkg/sentry/fs.AttrMask", (*AttrMask)(nil), state.Fns{Save: (*AttrMask).save, Load: (*AttrMask).load}) + state.Register("pkg/sentry/fs.PermMask", (*PermMask)(nil), state.Fns{Save: (*PermMask).save, Load: (*PermMask).load}) + state.Register("pkg/sentry/fs.FilePermissions", (*FilePermissions)(nil), state.Fns{Save: (*FilePermissions).save, Load: (*FilePermissions).load}) + state.Register("pkg/sentry/fs.FileOwner", (*FileOwner)(nil), state.Fns{Save: (*FileOwner).save, Load: (*FileOwner).load}) + state.Register("pkg/sentry/fs.DentAttr", (*DentAttr)(nil), state.Fns{Save: (*DentAttr).save, Load: (*DentAttr).load}) + state.Register("pkg/sentry/fs.SortedDentryMap", (*SortedDentryMap)(nil), state.Fns{Save: (*SortedDentryMap).save, Load: (*SortedDentryMap).load}) + state.Register("pkg/sentry/fs.Dirent", (*Dirent)(nil), state.Fns{Save: (*Dirent).save, Load: (*Dirent).load}) + state.Register("pkg/sentry/fs.DirentCache", (*DirentCache)(nil), state.Fns{Save: (*DirentCache).save, Load: (*DirentCache).load}) + state.Register("pkg/sentry/fs.DirentCacheLimiter", (*DirentCacheLimiter)(nil), state.Fns{Save: (*DirentCacheLimiter).save, Load: (*DirentCacheLimiter).load}) + state.Register("pkg/sentry/fs.direntList", (*direntList)(nil), state.Fns{Save: (*direntList).save, Load: (*direntList).load}) + state.Register("pkg/sentry/fs.direntEntry", (*direntEntry)(nil), state.Fns{Save: (*direntEntry).save, Load: (*direntEntry).load}) + state.Register("pkg/sentry/fs.eventList", (*eventList)(nil), state.Fns{Save: (*eventList).save, Load: (*eventList).load}) + state.Register("pkg/sentry/fs.eventEntry", (*eventEntry)(nil), state.Fns{Save: (*eventEntry).save, Load: (*eventEntry).load}) + state.Register("pkg/sentry/fs.File", (*File)(nil), state.Fns{Save: (*File).save, Load: (*File).load}) + state.Register("pkg/sentry/fs.overlayFileOperations", (*overlayFileOperations)(nil), state.Fns{Save: (*overlayFileOperations).save, Load: (*overlayFileOperations).load}) + state.Register("pkg/sentry/fs.overlayMappingIdentity", (*overlayMappingIdentity)(nil), state.Fns{Save: (*overlayMappingIdentity).save, Load: (*overlayMappingIdentity).load}) + state.Register("pkg/sentry/fs.MountSourceFlags", (*MountSourceFlags)(nil), state.Fns{Save: (*MountSourceFlags).save, Load: (*MountSourceFlags).load}) + state.Register("pkg/sentry/fs.FileFlags", (*FileFlags)(nil), state.Fns{Save: (*FileFlags).save, Load: (*FileFlags).load}) + state.Register("pkg/sentry/fs.Inode", (*Inode)(nil), state.Fns{Save: (*Inode).save, Load: (*Inode).load}) + state.Register("pkg/sentry/fs.LockCtx", (*LockCtx)(nil), state.Fns{Save: (*LockCtx).save, Load: (*LockCtx).load}) + state.Register("pkg/sentry/fs.Watches", (*Watches)(nil), state.Fns{Save: (*Watches).save, Load: (*Watches).load}) + state.Register("pkg/sentry/fs.Inotify", (*Inotify)(nil), state.Fns{Save: (*Inotify).save, Load: (*Inotify).load}) + state.Register("pkg/sentry/fs.Event", (*Event)(nil), state.Fns{Save: (*Event).save, Load: (*Event).load}) + state.Register("pkg/sentry/fs.Watch", (*Watch)(nil), state.Fns{Save: (*Watch).save, Load: (*Watch).load}) + state.Register("pkg/sentry/fs.MountSource", (*MountSource)(nil), state.Fns{Save: (*MountSource).save, Load: (*MountSource).load}) + state.Register("pkg/sentry/fs.SimpleMountSourceOperations", (*SimpleMountSourceOperations)(nil), state.Fns{Save: (*SimpleMountSourceOperations).save, Load: (*SimpleMountSourceOperations).load}) + state.Register("pkg/sentry/fs.overlayMountSourceOperations", (*overlayMountSourceOperations)(nil), state.Fns{Save: (*overlayMountSourceOperations).save, Load: (*overlayMountSourceOperations).load}) + state.Register("pkg/sentry/fs.overlayFilesystem", (*overlayFilesystem)(nil), state.Fns{Save: (*overlayFilesystem).save, Load: (*overlayFilesystem).load}) + state.Register("pkg/sentry/fs.Mount", (*Mount)(nil), state.Fns{Save: (*Mount).save, Load: (*Mount).load}) + state.Register("pkg/sentry/fs.MountNamespace", (*MountNamespace)(nil), state.Fns{Save: (*MountNamespace).save, Load: (*MountNamespace).load}) + state.Register("pkg/sentry/fs.overlayEntry", (*overlayEntry)(nil), state.Fns{Save: (*overlayEntry).save, Load: (*overlayEntry).load}) +} diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD deleted file mode 100644 index 789369220..000000000 --- a/pkg/sentry/fs/fsutil/BUILD +++ /dev/null @@ -1,118 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "dirty_set_impl", - out = "dirty_set_impl.go", - imports = { - "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", - }, - package = "fsutil", - prefix = "Dirty", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "memmap.MappableRange", - "Value": "DirtyInfo", - "Functions": "dirtySetFunctions", - }, -) - -go_template_instance( - name = "frame_ref_set_impl", - out = "frame_ref_set_impl.go", - imports = { - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", - }, - package = "fsutil", - prefix = "FrameRef", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "platform.FileRange", - "Value": "uint64", - "Functions": "FrameRefSetFunctions", - }, -) - -go_template_instance( - name = "file_range_set_impl", - out = "file_range_set_impl.go", - imports = { - "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", - }, - package = "fsutil", - prefix = "FileRange", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "memmap.MappableRange", - "Value": "uint64", - "Functions": "FileRangeSetFunctions", - }, -) - -go_library( - name = "fsutil", - srcs = [ - "dirty_set.go", - "dirty_set_impl.go", - "file.go", - "file_range_set.go", - "file_range_set_impl.go", - "frame_ref_set.go", - "frame_ref_set_impl.go", - "fsutil.go", - "host_file_mapper.go", - "host_file_mapper_state.go", - "host_file_mapper_unsafe.go", - "host_mappable.go", - "inode.go", - "inode_cached.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/log", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/state", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "fsutil_test", - size = "small", - srcs = [ - "dirty_set_test.go", - "inode_cached_test.go", - ], - library = ":fsutil", - deps = [ - "//pkg/context", - "//pkg/safemem", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fs/fsutil/README.md b/pkg/sentry/fs/fsutil/README.md deleted file mode 100644 index 8be367334..000000000 --- a/pkg/sentry/fs/fsutil/README.md +++ /dev/null @@ -1,207 +0,0 @@ -This package provides utilities for implementing virtual filesystem objects. - -[TOC] - -## Page cache - -`CachingInodeOperations` implements a page cache for files that cannot use the -host page cache. Normally these are files that store their data in a remote -filesystem. This also applies to files that are accessed on a platform that does -not support directly memory mapping host file descriptors (e.g. the ptrace -platform). - -An `CachingInodeOperations` buffers regions of a single file into memory. It is -owned by an `fs.Inode`, the in-memory representation of a file (all open file -descriptors are backed by an `fs.Inode`). The `fs.Inode` provides operations for -reading memory into an `CachingInodeOperations`, to represent the contents of -the file in-memory, and for writing memory out, to relieve memory pressure on -the kernel and to synchronize in-memory changes to filesystems. - -An `CachingInodeOperations` enables readable and/or writable memory access to -file content. Files can be mapped shared or private, see mmap(2). When a file is -mapped shared, changes to the file via write(2) and truncate(2) are reflected in -the shared memory region. Conversely, when the shared memory region is modified, -changes to the file are visible via read(2). Multiple shared mappings of the -same file are coherent with each other. This is consistent with Linux. - -When a file is mapped private, updates to the mapped memory are not visible to -other memory mappings. Updates to the mapped memory are also not reflected in -the file content as seen by read(2). If the file is changed after a private -mapping is created, for instance by write(2), the change to the file may or may -not be reflected in the private mapping. This is consistent with Linux. - -An `CachingInodeOperations` keeps track of ranges of memory that were modified -(or "dirtied"). When the file is explicitly synced via fsync(2), only the dirty -ranges are written out to the filesystem. Any error returned indicates a failure -to write all dirty memory of an `CachingInodeOperations` to the filesystem. In -this case the filesystem may be in an inconsistent state. The same operation can -be performed on the shared memory itself using msync(2). If neither fsync(2) nor -msync(2) is performed, then the dirty memory is written out in accordance with -the `CachingInodeOperations` eviction strategy (see below) and there is no -guarantee that memory will be written out successfully in full. - -### Memory allocation and eviction - -An `CachingInodeOperations` implements the following allocation and eviction -strategy: - -- Memory is allocated and brought up to date with the contents of a file when - a region of mapped memory is accessed (or "faulted on"). - -- Dirty memory is written out to filesystems when an fsync(2) or msync(2) - operation is performed on a memory mapped file, for all memory mapped files - when saved, and/or when there are no longer any memory mappings of a range - of a file, see munmap(2). As the latter implies, in the absence of a panic - or SIGKILL, dirty memory is written out for all memory mapped files when an - application exits. - -- Memory is freed when there are no longer any memory mappings of a range of a - file (e.g. when an application exits). This behavior is consistent with - Linux for shared memory that has been locked via mlock(2). - -Notably, memory is not allocated for read(2) or write(2) operations. This means -that reads and writes to the file are only accelerated by an -`CachingInodeOperations` if the file being read or written has been memory -mapped *and* if the shared memory has been accessed at the region being read or -written. This diverges from Linux which buffers memory into a page cache on -read(2) proactively (i.e. readahead) and delays writing it out to filesystems on -write(2) (i.e. writeback). The absence of these optimizations is not visible to -applications beyond less than optimal performance when repeatedly reading and/or -writing to same region of a file. See [Future Work](#future-work) for plans to -implement these optimizations. - -Additionally, memory held by `CachingInodeOperationss` is currently unbounded in -size. An `CachingInodeOperations` does not write out dirty memory and free it -under system memory pressure. This can cause pathological memory usage. - -When memory is written back, an `CachingInodeOperations` may write regions of -shared memory that were never modified. This is due to the strategy of -minimizing page faults (see below) and handling only a subset of memory write -faults. In the absence of an application or sentry crash, it is guaranteed that -if a region of shared memory was written to, it is written back to a filesystem. - -### Life of a shared memory mapping - -A file is memory mapped via mmap(2). For example, if `A` is an address, an -application may execute: - -``` -mmap(A, 0x1000, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); -``` - -This creates a shared mapping of fd that reflects 4k of the contents of fd -starting at offset 0, accessible at address `A`. This in turn creates a virtual -memory area region ("vma") which indicates that [`A`, `A`+0x1000) is now a valid -address range for this application to access. - -At this point, memory has not been allocated in the file's -`CachingInodeOperations`. It is also the case that the address range [`A`, -`A`+0x1000) has not been mapped on the host on behalf of the application. If the -application then tries to modify 8 bytes of the shared memory: - -``` -char buffer[] = "aaaaaaaa"; -memcpy(A, buffer, 8); -``` - -The host then sends a `SIGSEGV` to the sentry because the address range [`A`, -`A`+8) is not mapped on the host. The `SIGSEGV` indicates that the memory was -accessed writable. The sentry looks up the vma associated with [`A`, `A`+8), -finds the file that was mapped and its `CachingInodeOperations`. It then calls -`CachingInodeOperations.Translate` which allocates memory to back [`A`, `A`+8). -It may choose to allocate more memory (i.e. do "readahead") to minimize -subsequent faults. - -Memory that is allocated comes from a host tmpfs file (see -`pgalloc.MemoryFile`). The host tmpfs file memory is brought up to date with the -contents of the mapped file on its filesystem. The region of the host tmpfs file -that reflects the mapped file is then mapped into the host address space of the -application so that subsequent memory accesses do not repeatedly generate a -`SIGSEGV`. - -The range that was allocated, including any extra memory allocation to minimize -faults, is marked dirty due to the write fault. This overcounts dirty memory if -the extra memory allocated is never modified. - -To make the scenario more interesting, imagine that this application spawns -another process and maps the same file in the exact same way: - -``` -mmap(A, 0x1000, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); -``` - -Imagine that this process then tries to modify the file again but with only 4 -bytes: - -``` -char buffer[] = "bbbb"; -memcpy(A, buffer, 4); -``` - -Since the first process has already mapped and accessed the same region of the -file writable, `CachingInodeOperations.Translate` is called but returns the -memory that has already been allocated rather than allocating new memory. The -address range [`A`, `A`+0x1000) reflects the same cached view of the file as the -first process sees. For example, reading 8 bytes from the file from either -process via read(2) starting at offset 0 returns a consistent "bbbbaaaa". - -When this process no longer needs the shared memory, it may do: - -``` -munmap(A, 0x1000); -``` - -At this point, the modified memory cached by the `CachingInodeOperations` is not -written back to the file because it is still in use by the first process that -mapped it. When the first process also does: - -``` -munmap(A, 0x1000); -``` - -Then the last memory mapping of the file at the range [0, 0x1000) is gone. The -file's `CachingInodeOperations` then starts writing back memory marked dirty to -the file on its filesystem. Once writing completes, regardless of whether it was -successful, the `CachingInodeOperations` frees the memory cached at the range -[0, 0x1000). - -Subsequent read(2) or write(2) operations on the file go directly to the -filesystem since there no longer exists memory for it in its -`CachingInodeOperations`. - -## Future Work - -### Page cache - -The sentry does not yet implement the readahead and writeback optimizations for -read(2) and write(2) respectively. To do so, on read(2) and/or write(2) the -sentry must ensure that memory is allocated in a page cache to read or write -into. However, the sentry cannot boundlessly allocate memory. If it did, the -host would eventually OOM-kill the sentry+application process. This means that -the sentry must implement a page cache memory allocation strategy that is -bounded by a global user or container imposed limit. When this limit is -approached, the sentry must decide from which page cache memory should be freed -so that it can allocate more memory. If it makes a poor decision, the sentry may -end up freeing and re-allocating memory to back regions of files that are -frequently used, nullifying the optimization (and in some cases causing worse -performance due to the overhead of memory allocation and general management). -This is a form of "cache thrashing". - -In Linux, much research has been done to select and implement a lightweight but -optimal page cache eviction algorithm. Linux makes use of hardware page bits to -keep track of whether memory has been accessed. The sentry does not have direct -access to hardware. Implementing a similarly lightweight and optimal page cache -eviction algorithm will need to either introduce a kernel interface to obtain -these page bits or find a suitable alternative proxy for access events. - -In Linux, readahead happens by default but is not always ideal. For instance, -for files that are not read sequentially, it would be more ideal to simply read -from only those regions of the file rather than to optimistically cache some -number of bytes ahead of the read (up to 2MB in Linux) if the bytes cached won't -be accessed. Linux implements the fadvise64(2) system call for applications to -specify that a range of a file will not be accessed sequentially. The advice bit -FADV_RANDOM turns off the readahead optimization for the given range in the -given file. However fadvise64 is rarely used by applications so Linux implements -a readahead backoff strategy if reads are not sequential. To ensure that -application performance is not degraded, the sentry must implement a similar -backoff strategy. diff --git a/pkg/sentry/fs/fsutil/dirty_set_impl.go b/pkg/sentry/fs/fsutil/dirty_set_impl.go new file mode 100755 index 000000000..2510b81b3 --- /dev/null +++ b/pkg/sentry/fs/fsutil/dirty_set_impl.go @@ -0,0 +1,1274 @@ +package fsutil + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/sentry/memmap" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + DirtyminDegree = 3 + + DirtymaxDegree = 2 * DirtyminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type DirtySet struct { + root Dirtynode `state:".(*DirtySegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *DirtySet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *DirtySet) IsEmptyRange(r __generics_imported0.MappableRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *DirtySet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *DirtySet) SpanRange(r __generics_imported0.MappableRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *DirtySet) FirstSegment() DirtyIterator { + if s.root.nrSegments == 0 { + return DirtyIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *DirtySet) LastSegment() DirtyIterator { + if s.root.nrSegments == 0 { + return DirtyIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *DirtySet) FirstGap() DirtyGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return DirtyGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *DirtySet) LastGap() DirtyGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return DirtyGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *DirtySet) Find(key uint64) (DirtyIterator, DirtyGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return DirtyIterator{n, i}, DirtyGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return DirtyIterator{}, DirtyGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *DirtySet) FindSegment(key uint64) DirtyIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *DirtySet) LowerBoundSegment(min uint64) DirtyIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *DirtySet) UpperBoundSegment(max uint64) DirtyIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *DirtySet) FindGap(key uint64) DirtyGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *DirtySet) LowerBoundGap(min uint64) DirtyGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *DirtySet) UpperBoundGap(max uint64) DirtyGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *DirtySet) Add(r __generics_imported0.MappableRange, val DirtyInfo) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *DirtySet) AddWithoutMerging(r __generics_imported0.MappableRange, val DirtyInfo) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *DirtySet) Insert(gap DirtyGapIterator, r __generics_imported0.MappableRange, val DirtyInfo) DirtyIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (dirtySetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (dirtySetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (dirtySetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *DirtySet) InsertWithoutMerging(gap DirtyGapIterator, r __generics_imported0.MappableRange, val DirtyInfo) DirtyIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *DirtySet) InsertWithoutMergingUnchecked(gap DirtyGapIterator, r __generics_imported0.MappableRange, val DirtyInfo) DirtyIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return DirtyIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *DirtySet) Remove(seg DirtyIterator) DirtyGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + dirtySetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(DirtyGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *DirtySet) RemoveAll() { + s.root = Dirtynode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *DirtySet) RemoveRange(r __generics_imported0.MappableRange) DirtyGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *DirtySet) Merge(first, second DirtyIterator) DirtyIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *DirtySet) MergeUnchecked(first, second DirtyIterator) DirtyIterator { + if first.End() == second.Start() { + if mval, ok := (dirtySetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return DirtyIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *DirtySet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *DirtySet) MergeRange(r __generics_imported0.MappableRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *DirtySet) MergeAdjacent(r __generics_imported0.MappableRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *DirtySet) Split(seg DirtyIterator, split uint64) (DirtyIterator, DirtyIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *DirtySet) SplitUnchecked(seg DirtyIterator, split uint64) (DirtyIterator, DirtyIterator) { + val1, val2 := (dirtySetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.MappableRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *DirtySet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *DirtySet) Isolate(seg DirtyIterator, r __generics_imported0.MappableRange) DirtyIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *DirtySet) ApplyContiguous(r __generics_imported0.MappableRange, fn func(seg DirtyIterator)) DirtyGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return DirtyGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return DirtyGapIterator{} + } + } +} + +// +stateify savable +type Dirtynode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *Dirtynode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [DirtymaxDegree - 1]__generics_imported0.MappableRange + values [DirtymaxDegree - 1]DirtyInfo + children [DirtymaxDegree]*Dirtynode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Dirtynode) firstSegment() DirtyIterator { + for n.hasChildren { + n = n.children[0] + } + return DirtyIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Dirtynode) lastSegment() DirtyIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return DirtyIterator{n, n.nrSegments - 1} +} + +func (n *Dirtynode) prevSibling() *Dirtynode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *Dirtynode) nextSibling() *Dirtynode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *Dirtynode) rebalanceBeforeInsert(gap DirtyGapIterator) DirtyGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < DirtymaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &Dirtynode{ + nrSegments: DirtyminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &Dirtynode{ + nrSegments: DirtyminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:DirtyminDegree-1], n.keys[:DirtyminDegree-1]) + copy(left.values[:DirtyminDegree-1], n.values[:DirtyminDegree-1]) + copy(right.keys[:DirtyminDegree-1], n.keys[DirtyminDegree:]) + copy(right.values[:DirtyminDegree-1], n.values[DirtyminDegree:]) + n.keys[0], n.values[0] = n.keys[DirtyminDegree-1], n.values[DirtyminDegree-1] + DirtyzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:DirtyminDegree], n.children[:DirtyminDegree]) + copy(right.children[:DirtyminDegree], n.children[DirtyminDegree:]) + DirtyzeroNodeSlice(n.children[2:]) + for i := 0; i < DirtyminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < DirtyminDegree { + return DirtyGapIterator{left, gap.index} + } + return DirtyGapIterator{right, gap.index - DirtyminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[DirtyminDegree-1], n.values[DirtyminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &Dirtynode{ + nrSegments: DirtyminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:DirtyminDegree-1], n.keys[DirtyminDegree:]) + copy(sibling.values[:DirtyminDegree-1], n.values[DirtyminDegree:]) + DirtyzeroValueSlice(n.values[DirtyminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:DirtyminDegree], n.children[DirtyminDegree:]) + DirtyzeroNodeSlice(n.children[DirtyminDegree:]) + for i := 0; i < DirtyminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = DirtyminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < DirtyminDegree { + return gap + } + return DirtyGapIterator{sibling, gap.index - DirtyminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *Dirtynode) rebalanceAfterRemove(gap DirtyGapIterator) DirtyGapIterator { + for { + if n.nrSegments >= DirtyminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= DirtyminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + dirtySetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return DirtyGapIterator{n, 0} + } + if gap.node == n { + return DirtyGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= DirtyminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + dirtySetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return DirtyGapIterator{n, n.nrSegments} + } + return DirtyGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return DirtyGapIterator{p, gap.index} + } + if gap.node == right { + return DirtyGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *Dirtynode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = DirtyGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + dirtySetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type DirtyIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *Dirtynode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg DirtyIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg DirtyIterator) Range() __generics_imported0.MappableRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg DirtyIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg DirtyIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg DirtyIterator) SetRangeUnchecked(r __generics_imported0.MappableRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg DirtyIterator) SetRange(r __generics_imported0.MappableRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg DirtyIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg DirtyIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg DirtyIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg DirtyIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg DirtyIterator) Value() DirtyInfo { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg DirtyIterator) ValuePtr() *DirtyInfo { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg DirtyIterator) SetValue(val DirtyInfo) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg DirtyIterator) PrevSegment() DirtyIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return DirtyIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return DirtyIterator{} + } + return DirtysegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg DirtyIterator) NextSegment() DirtyIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return DirtyIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return DirtyIterator{} + } + return DirtysegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg DirtyIterator) PrevGap() DirtyGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return DirtyGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg DirtyIterator) NextGap() DirtyGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return DirtyGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg DirtyIterator) PrevNonEmpty() (DirtyIterator, DirtyGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return DirtyIterator{}, gap + } + return gap.PrevSegment(), DirtyGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg DirtyIterator) NextNonEmpty() (DirtyIterator, DirtyGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return DirtyIterator{}, gap + } + return gap.NextSegment(), DirtyGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type DirtyGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *Dirtynode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap DirtyGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap DirtyGapIterator) Range() __generics_imported0.MappableRange { + return __generics_imported0.MappableRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap DirtyGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return dirtySetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap DirtyGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return dirtySetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap DirtyGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap DirtyGapIterator) PrevSegment() DirtyIterator { + return DirtysegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap DirtyGapIterator) NextSegment() DirtyIterator { + return DirtysegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap DirtyGapIterator) PrevGap() DirtyGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return DirtyGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap DirtyGapIterator) NextGap() DirtyGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return DirtyGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func DirtysegmentBeforePosition(n *Dirtynode, i int) DirtyIterator { + for i == 0 { + if n.parent == nil { + return DirtyIterator{} + } + n, i = n.parent, n.parentIndex + } + return DirtyIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func DirtysegmentAfterPosition(n *Dirtynode, i int) DirtyIterator { + for i == n.nrSegments { + if n.parent == nil { + return DirtyIterator{} + } + n, i = n.parent, n.parentIndex + } + return DirtyIterator{n, i} +} + +func DirtyzeroValueSlice(slice []DirtyInfo) { + + for i := range slice { + dirtySetFunctions{}.ClearValue(&slice[i]) + } +} + +func DirtyzeroNodeSlice(slice []*Dirtynode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *DirtySet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *Dirtynode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *Dirtynode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type DirtySegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []DirtyInfo +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *DirtySet) ExportSortedSlices() *DirtySegmentDataSlices { + var sds DirtySegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *DirtySet) ImportSortedSlices(sds *DirtySegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.MappableRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *DirtySet) saveRoot() *DirtySegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *DirtySet) loadRoot(sds *DirtySegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/fs/fsutil/dirty_set_test.go b/pkg/sentry/fs/fsutil/dirty_set_test.go deleted file mode 100644 index e3579c23c..000000000 --- a/pkg/sentry/fs/fsutil/dirty_set_test.go +++ /dev/null @@ -1,38 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fsutil - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/usermem" -) - -func TestDirtySet(t *testing.T) { - var set DirtySet - set.MarkDirty(memmap.MappableRange{0, 2 * usermem.PageSize}) - set.KeepDirty(memmap.MappableRange{usermem.PageSize, 2 * usermem.PageSize}) - set.MarkClean(memmap.MappableRange{0, 2 * usermem.PageSize}) - want := &DirtySegmentDataSlices{ - Start: []uint64{usermem.PageSize}, - End: []uint64{2 * usermem.PageSize}, - Values: []DirtyInfo{{Keep: true}}, - } - if got := set.ExportSortedSlices(); !reflect.DeepEqual(got, want) { - t.Errorf("set:\n\tgot %v,\n\twant %v", got, want) - } -} diff --git a/pkg/sentry/fs/fsutil/file_range_set_impl.go b/pkg/sentry/fs/fsutil/file_range_set_impl.go new file mode 100755 index 000000000..01e7a2401 --- /dev/null +++ b/pkg/sentry/fs/fsutil/file_range_set_impl.go @@ -0,0 +1,1274 @@ +package fsutil + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/sentry/memmap" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + FileRangeminDegree = 3 + + FileRangemaxDegree = 2 * FileRangeminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type FileRangeSet struct { + root FileRangenode `state:".(*FileRangeSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *FileRangeSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *FileRangeSet) IsEmptyRange(r __generics_imported0.MappableRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *FileRangeSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *FileRangeSet) SpanRange(r __generics_imported0.MappableRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *FileRangeSet) FirstSegment() FileRangeIterator { + if s.root.nrSegments == 0 { + return FileRangeIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *FileRangeSet) LastSegment() FileRangeIterator { + if s.root.nrSegments == 0 { + return FileRangeIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *FileRangeSet) FirstGap() FileRangeGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return FileRangeGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *FileRangeSet) LastGap() FileRangeGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return FileRangeGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *FileRangeSet) Find(key uint64) (FileRangeIterator, FileRangeGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return FileRangeIterator{n, i}, FileRangeGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return FileRangeIterator{}, FileRangeGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *FileRangeSet) FindSegment(key uint64) FileRangeIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *FileRangeSet) LowerBoundSegment(min uint64) FileRangeIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *FileRangeSet) UpperBoundSegment(max uint64) FileRangeIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *FileRangeSet) FindGap(key uint64) FileRangeGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *FileRangeSet) LowerBoundGap(min uint64) FileRangeGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *FileRangeSet) UpperBoundGap(max uint64) FileRangeGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *FileRangeSet) Add(r __generics_imported0.MappableRange, val uint64) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *FileRangeSet) AddWithoutMerging(r __generics_imported0.MappableRange, val uint64) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *FileRangeSet) Insert(gap FileRangeGapIterator, r __generics_imported0.MappableRange, val uint64) FileRangeIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (FileRangeSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (FileRangeSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (FileRangeSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *FileRangeSet) InsertWithoutMerging(gap FileRangeGapIterator, r __generics_imported0.MappableRange, val uint64) FileRangeIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *FileRangeSet) InsertWithoutMergingUnchecked(gap FileRangeGapIterator, r __generics_imported0.MappableRange, val uint64) FileRangeIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return FileRangeIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *FileRangeSet) Remove(seg FileRangeIterator) FileRangeGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + FileRangeSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(FileRangeGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *FileRangeSet) RemoveAll() { + s.root = FileRangenode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *FileRangeSet) RemoveRange(r __generics_imported0.MappableRange) FileRangeGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *FileRangeSet) Merge(first, second FileRangeIterator) FileRangeIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *FileRangeSet) MergeUnchecked(first, second FileRangeIterator) FileRangeIterator { + if first.End() == second.Start() { + if mval, ok := (FileRangeSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return FileRangeIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *FileRangeSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *FileRangeSet) MergeRange(r __generics_imported0.MappableRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *FileRangeSet) MergeAdjacent(r __generics_imported0.MappableRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *FileRangeSet) Split(seg FileRangeIterator, split uint64) (FileRangeIterator, FileRangeIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *FileRangeSet) SplitUnchecked(seg FileRangeIterator, split uint64) (FileRangeIterator, FileRangeIterator) { + val1, val2 := (FileRangeSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.MappableRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *FileRangeSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *FileRangeSet) Isolate(seg FileRangeIterator, r __generics_imported0.MappableRange) FileRangeIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *FileRangeSet) ApplyContiguous(r __generics_imported0.MappableRange, fn func(seg FileRangeIterator)) FileRangeGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return FileRangeGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return FileRangeGapIterator{} + } + } +} + +// +stateify savable +type FileRangenode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *FileRangenode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [FileRangemaxDegree - 1]__generics_imported0.MappableRange + values [FileRangemaxDegree - 1]uint64 + children [FileRangemaxDegree]*FileRangenode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *FileRangenode) firstSegment() FileRangeIterator { + for n.hasChildren { + n = n.children[0] + } + return FileRangeIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *FileRangenode) lastSegment() FileRangeIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return FileRangeIterator{n, n.nrSegments - 1} +} + +func (n *FileRangenode) prevSibling() *FileRangenode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *FileRangenode) nextSibling() *FileRangenode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *FileRangenode) rebalanceBeforeInsert(gap FileRangeGapIterator) FileRangeGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < FileRangemaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &FileRangenode{ + nrSegments: FileRangeminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &FileRangenode{ + nrSegments: FileRangeminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:FileRangeminDegree-1], n.keys[:FileRangeminDegree-1]) + copy(left.values[:FileRangeminDegree-1], n.values[:FileRangeminDegree-1]) + copy(right.keys[:FileRangeminDegree-1], n.keys[FileRangeminDegree:]) + copy(right.values[:FileRangeminDegree-1], n.values[FileRangeminDegree:]) + n.keys[0], n.values[0] = n.keys[FileRangeminDegree-1], n.values[FileRangeminDegree-1] + FileRangezeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:FileRangeminDegree], n.children[:FileRangeminDegree]) + copy(right.children[:FileRangeminDegree], n.children[FileRangeminDegree:]) + FileRangezeroNodeSlice(n.children[2:]) + for i := 0; i < FileRangeminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < FileRangeminDegree { + return FileRangeGapIterator{left, gap.index} + } + return FileRangeGapIterator{right, gap.index - FileRangeminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[FileRangeminDegree-1], n.values[FileRangeminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &FileRangenode{ + nrSegments: FileRangeminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:FileRangeminDegree-1], n.keys[FileRangeminDegree:]) + copy(sibling.values[:FileRangeminDegree-1], n.values[FileRangeminDegree:]) + FileRangezeroValueSlice(n.values[FileRangeminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:FileRangeminDegree], n.children[FileRangeminDegree:]) + FileRangezeroNodeSlice(n.children[FileRangeminDegree:]) + for i := 0; i < FileRangeminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = FileRangeminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < FileRangeminDegree { + return gap + } + return FileRangeGapIterator{sibling, gap.index - FileRangeminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *FileRangenode) rebalanceAfterRemove(gap FileRangeGapIterator) FileRangeGapIterator { + for { + if n.nrSegments >= FileRangeminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= FileRangeminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + FileRangeSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return FileRangeGapIterator{n, 0} + } + if gap.node == n { + return FileRangeGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= FileRangeminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + FileRangeSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return FileRangeGapIterator{n, n.nrSegments} + } + return FileRangeGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return FileRangeGapIterator{p, gap.index} + } + if gap.node == right { + return FileRangeGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *FileRangenode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = FileRangeGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + FileRangeSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type FileRangeIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *FileRangenode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg FileRangeIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg FileRangeIterator) Range() __generics_imported0.MappableRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg FileRangeIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg FileRangeIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg FileRangeIterator) SetRangeUnchecked(r __generics_imported0.MappableRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg FileRangeIterator) SetRange(r __generics_imported0.MappableRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg FileRangeIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg FileRangeIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg FileRangeIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg FileRangeIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg FileRangeIterator) Value() uint64 { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg FileRangeIterator) ValuePtr() *uint64 { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg FileRangeIterator) SetValue(val uint64) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg FileRangeIterator) PrevSegment() FileRangeIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return FileRangeIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return FileRangeIterator{} + } + return FileRangesegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg FileRangeIterator) NextSegment() FileRangeIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return FileRangeIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return FileRangeIterator{} + } + return FileRangesegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg FileRangeIterator) PrevGap() FileRangeGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return FileRangeGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg FileRangeIterator) NextGap() FileRangeGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return FileRangeGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg FileRangeIterator) PrevNonEmpty() (FileRangeIterator, FileRangeGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return FileRangeIterator{}, gap + } + return gap.PrevSegment(), FileRangeGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg FileRangeIterator) NextNonEmpty() (FileRangeIterator, FileRangeGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return FileRangeIterator{}, gap + } + return gap.NextSegment(), FileRangeGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type FileRangeGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *FileRangenode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap FileRangeGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap FileRangeGapIterator) Range() __generics_imported0.MappableRange { + return __generics_imported0.MappableRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap FileRangeGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return FileRangeSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap FileRangeGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return FileRangeSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap FileRangeGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap FileRangeGapIterator) PrevSegment() FileRangeIterator { + return FileRangesegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap FileRangeGapIterator) NextSegment() FileRangeIterator { + return FileRangesegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap FileRangeGapIterator) PrevGap() FileRangeGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return FileRangeGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap FileRangeGapIterator) NextGap() FileRangeGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return FileRangeGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func FileRangesegmentBeforePosition(n *FileRangenode, i int) FileRangeIterator { + for i == 0 { + if n.parent == nil { + return FileRangeIterator{} + } + n, i = n.parent, n.parentIndex + } + return FileRangeIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func FileRangesegmentAfterPosition(n *FileRangenode, i int) FileRangeIterator { + for i == n.nrSegments { + if n.parent == nil { + return FileRangeIterator{} + } + n, i = n.parent, n.parentIndex + } + return FileRangeIterator{n, i} +} + +func FileRangezeroValueSlice(slice []uint64) { + + for i := range slice { + FileRangeSetFunctions{}.ClearValue(&slice[i]) + } +} + +func FileRangezeroNodeSlice(slice []*FileRangenode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *FileRangeSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *FileRangenode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *FileRangenode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type FileRangeSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []uint64 +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *FileRangeSet) ExportSortedSlices() *FileRangeSegmentDataSlices { + var sds FileRangeSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *FileRangeSet) ImportSortedSlices(sds *FileRangeSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.MappableRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *FileRangeSet) saveRoot() *FileRangeSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *FileRangeSet) loadRoot(sds *FileRangeSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/fs/fsutil/frame_ref_set_impl.go b/pkg/sentry/fs/fsutil/frame_ref_set_impl.go new file mode 100755 index 000000000..88695dbd1 --- /dev/null +++ b/pkg/sentry/fs/fsutil/frame_ref_set_impl.go @@ -0,0 +1,1274 @@ +package fsutil + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/sentry/platform" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + FrameRefminDegree = 3 + + FrameRefmaxDegree = 2 * FrameRefminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type FrameRefSet struct { + root FrameRefnode `state:".(*FrameRefSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *FrameRefSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *FrameRefSet) IsEmptyRange(r __generics_imported0.FileRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *FrameRefSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *FrameRefSet) SpanRange(r __generics_imported0.FileRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *FrameRefSet) FirstSegment() FrameRefIterator { + if s.root.nrSegments == 0 { + return FrameRefIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *FrameRefSet) LastSegment() FrameRefIterator { + if s.root.nrSegments == 0 { + return FrameRefIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *FrameRefSet) FirstGap() FrameRefGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return FrameRefGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *FrameRefSet) LastGap() FrameRefGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return FrameRefGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *FrameRefSet) Find(key uint64) (FrameRefIterator, FrameRefGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return FrameRefIterator{n, i}, FrameRefGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return FrameRefIterator{}, FrameRefGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *FrameRefSet) FindSegment(key uint64) FrameRefIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *FrameRefSet) LowerBoundSegment(min uint64) FrameRefIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *FrameRefSet) UpperBoundSegment(max uint64) FrameRefIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *FrameRefSet) FindGap(key uint64) FrameRefGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *FrameRefSet) LowerBoundGap(min uint64) FrameRefGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *FrameRefSet) UpperBoundGap(max uint64) FrameRefGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *FrameRefSet) Add(r __generics_imported0.FileRange, val uint64) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *FrameRefSet) AddWithoutMerging(r __generics_imported0.FileRange, val uint64) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *FrameRefSet) Insert(gap FrameRefGapIterator, r __generics_imported0.FileRange, val uint64) FrameRefIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (FrameRefSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (FrameRefSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (FrameRefSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *FrameRefSet) InsertWithoutMerging(gap FrameRefGapIterator, r __generics_imported0.FileRange, val uint64) FrameRefIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *FrameRefSet) InsertWithoutMergingUnchecked(gap FrameRefGapIterator, r __generics_imported0.FileRange, val uint64) FrameRefIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return FrameRefIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *FrameRefSet) Remove(seg FrameRefIterator) FrameRefGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + FrameRefSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(FrameRefGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *FrameRefSet) RemoveAll() { + s.root = FrameRefnode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *FrameRefSet) RemoveRange(r __generics_imported0.FileRange) FrameRefGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *FrameRefSet) Merge(first, second FrameRefIterator) FrameRefIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *FrameRefSet) MergeUnchecked(first, second FrameRefIterator) FrameRefIterator { + if first.End() == second.Start() { + if mval, ok := (FrameRefSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return FrameRefIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *FrameRefSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *FrameRefSet) MergeRange(r __generics_imported0.FileRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *FrameRefSet) MergeAdjacent(r __generics_imported0.FileRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *FrameRefSet) Split(seg FrameRefIterator, split uint64) (FrameRefIterator, FrameRefIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *FrameRefSet) SplitUnchecked(seg FrameRefIterator, split uint64) (FrameRefIterator, FrameRefIterator) { + val1, val2 := (FrameRefSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.FileRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *FrameRefSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *FrameRefSet) Isolate(seg FrameRefIterator, r __generics_imported0.FileRange) FrameRefIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *FrameRefSet) ApplyContiguous(r __generics_imported0.FileRange, fn func(seg FrameRefIterator)) FrameRefGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return FrameRefGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return FrameRefGapIterator{} + } + } +} + +// +stateify savable +type FrameRefnode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *FrameRefnode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [FrameRefmaxDegree - 1]__generics_imported0.FileRange + values [FrameRefmaxDegree - 1]uint64 + children [FrameRefmaxDegree]*FrameRefnode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *FrameRefnode) firstSegment() FrameRefIterator { + for n.hasChildren { + n = n.children[0] + } + return FrameRefIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *FrameRefnode) lastSegment() FrameRefIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return FrameRefIterator{n, n.nrSegments - 1} +} + +func (n *FrameRefnode) prevSibling() *FrameRefnode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *FrameRefnode) nextSibling() *FrameRefnode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *FrameRefnode) rebalanceBeforeInsert(gap FrameRefGapIterator) FrameRefGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < FrameRefmaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &FrameRefnode{ + nrSegments: FrameRefminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &FrameRefnode{ + nrSegments: FrameRefminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:FrameRefminDegree-1], n.keys[:FrameRefminDegree-1]) + copy(left.values[:FrameRefminDegree-1], n.values[:FrameRefminDegree-1]) + copy(right.keys[:FrameRefminDegree-1], n.keys[FrameRefminDegree:]) + copy(right.values[:FrameRefminDegree-1], n.values[FrameRefminDegree:]) + n.keys[0], n.values[0] = n.keys[FrameRefminDegree-1], n.values[FrameRefminDegree-1] + FrameRefzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:FrameRefminDegree], n.children[:FrameRefminDegree]) + copy(right.children[:FrameRefminDegree], n.children[FrameRefminDegree:]) + FrameRefzeroNodeSlice(n.children[2:]) + for i := 0; i < FrameRefminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < FrameRefminDegree { + return FrameRefGapIterator{left, gap.index} + } + return FrameRefGapIterator{right, gap.index - FrameRefminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[FrameRefminDegree-1], n.values[FrameRefminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &FrameRefnode{ + nrSegments: FrameRefminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:FrameRefminDegree-1], n.keys[FrameRefminDegree:]) + copy(sibling.values[:FrameRefminDegree-1], n.values[FrameRefminDegree:]) + FrameRefzeroValueSlice(n.values[FrameRefminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:FrameRefminDegree], n.children[FrameRefminDegree:]) + FrameRefzeroNodeSlice(n.children[FrameRefminDegree:]) + for i := 0; i < FrameRefminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = FrameRefminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < FrameRefminDegree { + return gap + } + return FrameRefGapIterator{sibling, gap.index - FrameRefminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *FrameRefnode) rebalanceAfterRemove(gap FrameRefGapIterator) FrameRefGapIterator { + for { + if n.nrSegments >= FrameRefminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= FrameRefminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + FrameRefSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return FrameRefGapIterator{n, 0} + } + if gap.node == n { + return FrameRefGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= FrameRefminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + FrameRefSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return FrameRefGapIterator{n, n.nrSegments} + } + return FrameRefGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return FrameRefGapIterator{p, gap.index} + } + if gap.node == right { + return FrameRefGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *FrameRefnode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = FrameRefGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + FrameRefSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type FrameRefIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *FrameRefnode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg FrameRefIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg FrameRefIterator) Range() __generics_imported0.FileRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg FrameRefIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg FrameRefIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg FrameRefIterator) SetRangeUnchecked(r __generics_imported0.FileRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg FrameRefIterator) SetRange(r __generics_imported0.FileRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg FrameRefIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg FrameRefIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg FrameRefIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg FrameRefIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg FrameRefIterator) Value() uint64 { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg FrameRefIterator) ValuePtr() *uint64 { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg FrameRefIterator) SetValue(val uint64) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg FrameRefIterator) PrevSegment() FrameRefIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return FrameRefIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return FrameRefIterator{} + } + return FrameRefsegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg FrameRefIterator) NextSegment() FrameRefIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return FrameRefIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return FrameRefIterator{} + } + return FrameRefsegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg FrameRefIterator) PrevGap() FrameRefGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return FrameRefGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg FrameRefIterator) NextGap() FrameRefGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return FrameRefGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg FrameRefIterator) PrevNonEmpty() (FrameRefIterator, FrameRefGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return FrameRefIterator{}, gap + } + return gap.PrevSegment(), FrameRefGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg FrameRefIterator) NextNonEmpty() (FrameRefIterator, FrameRefGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return FrameRefIterator{}, gap + } + return gap.NextSegment(), FrameRefGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type FrameRefGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *FrameRefnode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap FrameRefGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap FrameRefGapIterator) Range() __generics_imported0.FileRange { + return __generics_imported0.FileRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap FrameRefGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return FrameRefSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap FrameRefGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return FrameRefSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap FrameRefGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap FrameRefGapIterator) PrevSegment() FrameRefIterator { + return FrameRefsegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap FrameRefGapIterator) NextSegment() FrameRefIterator { + return FrameRefsegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap FrameRefGapIterator) PrevGap() FrameRefGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return FrameRefGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap FrameRefGapIterator) NextGap() FrameRefGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return FrameRefGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func FrameRefsegmentBeforePosition(n *FrameRefnode, i int) FrameRefIterator { + for i == 0 { + if n.parent == nil { + return FrameRefIterator{} + } + n, i = n.parent, n.parentIndex + } + return FrameRefIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func FrameRefsegmentAfterPosition(n *FrameRefnode, i int) FrameRefIterator { + for i == n.nrSegments { + if n.parent == nil { + return FrameRefIterator{} + } + n, i = n.parent, n.parentIndex + } + return FrameRefIterator{n, i} +} + +func FrameRefzeroValueSlice(slice []uint64) { + + for i := range slice { + FrameRefSetFunctions{}.ClearValue(&slice[i]) + } +} + +func FrameRefzeroNodeSlice(slice []*FrameRefnode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *FrameRefSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *FrameRefnode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *FrameRefnode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type FrameRefSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []uint64 +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *FrameRefSet) ExportSortedSlices() *FrameRefSegmentDataSlices { + var sds FrameRefSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *FrameRefSet) ImportSortedSlices(sds *FrameRefSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.FileRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *FrameRefSet) saveRoot() *FrameRefSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *FrameRefSet) loadRoot(sds *FrameRefSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/fs/fsutil/fsutil_impl_state_autogen.go b/pkg/sentry/fs/fsutil/fsutil_impl_state_autogen.go new file mode 100755 index 000000000..a0baca0c5 --- /dev/null +++ b/pkg/sentry/fs/fsutil/fsutil_impl_state_autogen.go @@ -0,0 +1,169 @@ +// automatically generated by stateify. + +package fsutil + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *DirtySet) beforeSave() {} +func (x *DirtySet) save(m state.Map) { + x.beforeSave() + var root *DirtySegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *DirtySet) afterLoad() {} +func (x *DirtySet) load(m state.Map) { + m.LoadValue("root", new(*DirtySegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*DirtySegmentDataSlices)) }) +} + +func (x *Dirtynode) beforeSave() {} +func (x *Dirtynode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *Dirtynode) afterLoad() {} +func (x *Dirtynode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *DirtySegmentDataSlices) beforeSave() {} +func (x *DirtySegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *DirtySegmentDataSlices) afterLoad() {} +func (x *DirtySegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *FileRangeSet) beforeSave() {} +func (x *FileRangeSet) save(m state.Map) { + x.beforeSave() + var root *FileRangeSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *FileRangeSet) afterLoad() {} +func (x *FileRangeSet) load(m state.Map) { + m.LoadValue("root", new(*FileRangeSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*FileRangeSegmentDataSlices)) }) +} + +func (x *FileRangenode) beforeSave() {} +func (x *FileRangenode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *FileRangenode) afterLoad() {} +func (x *FileRangenode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *FileRangeSegmentDataSlices) beforeSave() {} +func (x *FileRangeSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *FileRangeSegmentDataSlices) afterLoad() {} +func (x *FileRangeSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *FrameRefSet) beforeSave() {} +func (x *FrameRefSet) save(m state.Map) { + x.beforeSave() + var root *FrameRefSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *FrameRefSet) afterLoad() {} +func (x *FrameRefSet) load(m state.Map) { + m.LoadValue("root", new(*FrameRefSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*FrameRefSegmentDataSlices)) }) +} + +func (x *FrameRefnode) beforeSave() {} +func (x *FrameRefnode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *FrameRefnode) afterLoad() {} +func (x *FrameRefnode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *FrameRefSegmentDataSlices) beforeSave() {} +func (x *FrameRefSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *FrameRefSegmentDataSlices) afterLoad() {} +func (x *FrameRefSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func init() { + state.Register("pkg/sentry/fs/fsutil.DirtySet", (*DirtySet)(nil), state.Fns{Save: (*DirtySet).save, Load: (*DirtySet).load}) + state.Register("pkg/sentry/fs/fsutil.Dirtynode", (*Dirtynode)(nil), state.Fns{Save: (*Dirtynode).save, Load: (*Dirtynode).load}) + state.Register("pkg/sentry/fs/fsutil.DirtySegmentDataSlices", (*DirtySegmentDataSlices)(nil), state.Fns{Save: (*DirtySegmentDataSlices).save, Load: (*DirtySegmentDataSlices).load}) + state.Register("pkg/sentry/fs/fsutil.FileRangeSet", (*FileRangeSet)(nil), state.Fns{Save: (*FileRangeSet).save, Load: (*FileRangeSet).load}) + state.Register("pkg/sentry/fs/fsutil.FileRangenode", (*FileRangenode)(nil), state.Fns{Save: (*FileRangenode).save, Load: (*FileRangenode).load}) + state.Register("pkg/sentry/fs/fsutil.FileRangeSegmentDataSlices", (*FileRangeSegmentDataSlices)(nil), state.Fns{Save: (*FileRangeSegmentDataSlices).save, Load: (*FileRangeSegmentDataSlices).load}) + state.Register("pkg/sentry/fs/fsutil.FrameRefSet", (*FrameRefSet)(nil), state.Fns{Save: (*FrameRefSet).save, Load: (*FrameRefSet).load}) + state.Register("pkg/sentry/fs/fsutil.FrameRefnode", (*FrameRefnode)(nil), state.Fns{Save: (*FrameRefnode).save, Load: (*FrameRefnode).load}) + state.Register("pkg/sentry/fs/fsutil.FrameRefSegmentDataSlices", (*FrameRefSegmentDataSlices)(nil), state.Fns{Save: (*FrameRefSegmentDataSlices).save, Load: (*FrameRefSegmentDataSlices).load}) +} diff --git a/pkg/sentry/fs/fsutil/fsutil_state_autogen.go b/pkg/sentry/fs/fsutil/fsutil_state_autogen.go new file mode 100755 index 000000000..80b93ad25 --- /dev/null +++ b/pkg/sentry/fs/fsutil/fsutil_state_autogen.go @@ -0,0 +1,204 @@ +// automatically generated by stateify. + +package fsutil + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *DirtyInfo) beforeSave() {} +func (x *DirtyInfo) save(m state.Map) { + x.beforeSave() + m.Save("Keep", &x.Keep) +} + +func (x *DirtyInfo) afterLoad() {} +func (x *DirtyInfo) load(m state.Map) { + m.Load("Keep", &x.Keep) +} + +func (x *StaticDirFileOperations) beforeSave() {} +func (x *StaticDirFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("dentryMap", &x.dentryMap) + m.Save("dirCursor", &x.dirCursor) +} + +func (x *StaticDirFileOperations) afterLoad() {} +func (x *StaticDirFileOperations) load(m state.Map) { + m.Load("dentryMap", &x.dentryMap) + m.Load("dirCursor", &x.dirCursor) +} + +func (x *NoReadWriteFile) beforeSave() {} +func (x *NoReadWriteFile) save(m state.Map) { + x.beforeSave() +} + +func (x *NoReadWriteFile) afterLoad() {} +func (x *NoReadWriteFile) load(m state.Map) { +} + +func (x *FileStaticContentReader) beforeSave() {} +func (x *FileStaticContentReader) save(m state.Map) { + x.beforeSave() + m.Save("content", &x.content) +} + +func (x *FileStaticContentReader) afterLoad() {} +func (x *FileStaticContentReader) load(m state.Map) { + m.Load("content", &x.content) +} + +func (x *HostFileMapper) beforeSave() {} +func (x *HostFileMapper) save(m state.Map) { + x.beforeSave() + m.Save("refs", &x.refs) +} + +func (x *HostFileMapper) load(m state.Map) { + m.Load("refs", &x.refs) + m.AfterLoad(x.afterLoad) +} + +func (x *HostMappable) beforeSave() {} +func (x *HostMappable) save(m state.Map) { + x.beforeSave() + m.Save("hostFileMapper", &x.hostFileMapper) + m.Save("backingFile", &x.backingFile) + m.Save("mappings", &x.mappings) +} + +func (x *HostMappable) afterLoad() {} +func (x *HostMappable) load(m state.Map) { + m.Load("hostFileMapper", &x.hostFileMapper) + m.Load("backingFile", &x.backingFile) + m.Load("mappings", &x.mappings) +} + +func (x *SimpleFileInode) beforeSave() {} +func (x *SimpleFileInode) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *SimpleFileInode) afterLoad() {} +func (x *SimpleFileInode) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *NoReadWriteFileInode) beforeSave() {} +func (x *NoReadWriteFileInode) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *NoReadWriteFileInode) afterLoad() {} +func (x *NoReadWriteFileInode) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *InodeSimpleAttributes) beforeSave() {} +func (x *InodeSimpleAttributes) save(m state.Map) { + x.beforeSave() + m.Save("fsType", &x.fsType) + m.Save("unstable", &x.unstable) +} + +func (x *InodeSimpleAttributes) afterLoad() {} +func (x *InodeSimpleAttributes) load(m state.Map) { + m.Load("fsType", &x.fsType) + m.Load("unstable", &x.unstable) +} + +func (x *InodeSimpleExtendedAttributes) beforeSave() {} +func (x *InodeSimpleExtendedAttributes) save(m state.Map) { + x.beforeSave() + m.Save("xattrs", &x.xattrs) +} + +func (x *InodeSimpleExtendedAttributes) afterLoad() {} +func (x *InodeSimpleExtendedAttributes) load(m state.Map) { + m.Load("xattrs", &x.xattrs) +} + +func (x *staticFile) beforeSave() {} +func (x *staticFile) save(m state.Map) { + x.beforeSave() + m.Save("FileStaticContentReader", &x.FileStaticContentReader) +} + +func (x *staticFile) afterLoad() {} +func (x *staticFile) load(m state.Map) { + m.Load("FileStaticContentReader", &x.FileStaticContentReader) +} + +func (x *InodeStaticFileGetter) beforeSave() {} +func (x *InodeStaticFileGetter) save(m state.Map) { + x.beforeSave() + m.Save("Contents", &x.Contents) +} + +func (x *InodeStaticFileGetter) afterLoad() {} +func (x *InodeStaticFileGetter) load(m state.Map) { + m.Load("Contents", &x.Contents) +} + +func (x *CachingInodeOperations) beforeSave() {} +func (x *CachingInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("backingFile", &x.backingFile) + m.Save("mfp", &x.mfp) + m.Save("opts", &x.opts) + m.Save("attr", &x.attr) + m.Save("dirtyAttr", &x.dirtyAttr) + m.Save("mappings", &x.mappings) + m.Save("cache", &x.cache) + m.Save("dirty", &x.dirty) + m.Save("hostFileMapper", &x.hostFileMapper) + m.Save("refs", &x.refs) +} + +func (x *CachingInodeOperations) afterLoad() {} +func (x *CachingInodeOperations) load(m state.Map) { + m.Load("backingFile", &x.backingFile) + m.Load("mfp", &x.mfp) + m.Load("opts", &x.opts) + m.Load("attr", &x.attr) + m.Load("dirtyAttr", &x.dirtyAttr) + m.Load("mappings", &x.mappings) + m.Load("cache", &x.cache) + m.Load("dirty", &x.dirty) + m.Load("hostFileMapper", &x.hostFileMapper) + m.Load("refs", &x.refs) +} + +func (x *CachingInodeOperationsOptions) beforeSave() {} +func (x *CachingInodeOperationsOptions) save(m state.Map) { + x.beforeSave() + m.Save("ForcePageCache", &x.ForcePageCache) + m.Save("LimitHostFDTranslation", &x.LimitHostFDTranslation) +} + +func (x *CachingInodeOperationsOptions) afterLoad() {} +func (x *CachingInodeOperationsOptions) load(m state.Map) { + m.Load("ForcePageCache", &x.ForcePageCache) + m.Load("LimitHostFDTranslation", &x.LimitHostFDTranslation) +} + +func init() { + state.Register("pkg/sentry/fs/fsutil.DirtyInfo", (*DirtyInfo)(nil), state.Fns{Save: (*DirtyInfo).save, Load: (*DirtyInfo).load}) + state.Register("pkg/sentry/fs/fsutil.StaticDirFileOperations", (*StaticDirFileOperations)(nil), state.Fns{Save: (*StaticDirFileOperations).save, Load: (*StaticDirFileOperations).load}) + state.Register("pkg/sentry/fs/fsutil.NoReadWriteFile", (*NoReadWriteFile)(nil), state.Fns{Save: (*NoReadWriteFile).save, Load: (*NoReadWriteFile).load}) + state.Register("pkg/sentry/fs/fsutil.FileStaticContentReader", (*FileStaticContentReader)(nil), state.Fns{Save: (*FileStaticContentReader).save, Load: (*FileStaticContentReader).load}) + state.Register("pkg/sentry/fs/fsutil.HostFileMapper", (*HostFileMapper)(nil), state.Fns{Save: (*HostFileMapper).save, Load: (*HostFileMapper).load}) + state.Register("pkg/sentry/fs/fsutil.HostMappable", (*HostMappable)(nil), state.Fns{Save: (*HostMappable).save, Load: (*HostMappable).load}) + state.Register("pkg/sentry/fs/fsutil.SimpleFileInode", (*SimpleFileInode)(nil), state.Fns{Save: (*SimpleFileInode).save, Load: (*SimpleFileInode).load}) + state.Register("pkg/sentry/fs/fsutil.NoReadWriteFileInode", (*NoReadWriteFileInode)(nil), state.Fns{Save: (*NoReadWriteFileInode).save, Load: (*NoReadWriteFileInode).load}) + state.Register("pkg/sentry/fs/fsutil.InodeSimpleAttributes", (*InodeSimpleAttributes)(nil), state.Fns{Save: (*InodeSimpleAttributes).save, Load: (*InodeSimpleAttributes).load}) + state.Register("pkg/sentry/fs/fsutil.InodeSimpleExtendedAttributes", (*InodeSimpleExtendedAttributes)(nil), state.Fns{Save: (*InodeSimpleExtendedAttributes).save, Load: (*InodeSimpleExtendedAttributes).load}) + state.Register("pkg/sentry/fs/fsutil.staticFile", (*staticFile)(nil), state.Fns{Save: (*staticFile).save, Load: (*staticFile).load}) + state.Register("pkg/sentry/fs/fsutil.InodeStaticFileGetter", (*InodeStaticFileGetter)(nil), state.Fns{Save: (*InodeStaticFileGetter).save, Load: (*InodeStaticFileGetter).load}) + state.Register("pkg/sentry/fs/fsutil.CachingInodeOperations", (*CachingInodeOperations)(nil), state.Fns{Save: (*CachingInodeOperations).save, Load: (*CachingInodeOperations).load}) + state.Register("pkg/sentry/fs/fsutil.CachingInodeOperationsOptions", (*CachingInodeOperationsOptions)(nil), state.Fns{Save: (*CachingInodeOperationsOptions).save, Load: (*CachingInodeOperationsOptions).load}) +} diff --git a/pkg/sentry/fs/fsutil/inode_cached_test.go b/pkg/sentry/fs/fsutil/inode_cached_test.go deleted file mode 100644 index 1547584c5..000000000 --- a/pkg/sentry/fs/fsutil/inode_cached_test.go +++ /dev/null @@ -1,389 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fsutil - -import ( - "bytes" - "io" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -type noopBackingFile struct{} - -func (noopBackingFile) ReadToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offset uint64) (uint64, error) { - return dsts.NumBytes(), nil -} - -func (noopBackingFile) WriteFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) { - return srcs.NumBytes(), nil -} - -func (noopBackingFile) SetMaskedAttributes(context.Context, fs.AttrMask, fs.UnstableAttr, bool) error { - return nil -} - -func (noopBackingFile) Sync(context.Context) error { - return nil -} - -func (noopBackingFile) FD() int { - return -1 -} - -func (noopBackingFile) Allocate(ctx context.Context, offset int64, length int64) error { - return nil -} - -func TestSetPermissions(t *testing.T) { - ctx := contexttest.Context(t) - - uattr := fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Perms: fs.FilePermsFromMode(0444), - }) - iops := NewCachingInodeOperations(ctx, noopBackingFile{}, uattr, CachingInodeOperationsOptions{}) - defer iops.Release() - - perms := fs.FilePermsFromMode(0777) - if !iops.SetPermissions(ctx, nil, perms) { - t.Fatalf("SetPermissions failed, want success") - } - - // Did permissions change? - if iops.attr.Perms != perms { - t.Fatalf("got perms +%v, want +%v", iops.attr.Perms, perms) - } - - // Did status change time change? - if !iops.dirtyAttr.StatusChangeTime { - t.Fatalf("got status change time not dirty, want dirty") - } - if iops.attr.StatusChangeTime.Equal(uattr.StatusChangeTime) { - t.Fatalf("got status change time unchanged") - } -} - -func TestSetTimestamps(t *testing.T) { - ctx := contexttest.Context(t) - for _, test := range []struct { - desc string - ts fs.TimeSpec - wantChanged fs.AttrMask - }{ - { - desc: "noop", - ts: fs.TimeSpec{ - ATimeOmit: true, - MTimeOmit: true, - }, - wantChanged: fs.AttrMask{}, - }, - { - desc: "access time only", - ts: fs.TimeSpec{ - ATime: ktime.NowFromContext(ctx), - MTimeOmit: true, - }, - wantChanged: fs.AttrMask{ - AccessTime: true, - }, - }, - { - desc: "modification time only", - ts: fs.TimeSpec{ - ATimeOmit: true, - MTime: ktime.NowFromContext(ctx), - }, - wantChanged: fs.AttrMask{ - ModificationTime: true, - }, - }, - { - desc: "access and modification time", - ts: fs.TimeSpec{ - ATime: ktime.NowFromContext(ctx), - MTime: ktime.NowFromContext(ctx), - }, - wantChanged: fs.AttrMask{ - AccessTime: true, - ModificationTime: true, - }, - }, - { - desc: "system time access and modification time", - ts: fs.TimeSpec{ - ATimeSetSystemTime: true, - MTimeSetSystemTime: true, - }, - wantChanged: fs.AttrMask{ - AccessTime: true, - ModificationTime: true, - }, - }, - } { - t.Run(test.desc, func(t *testing.T) { - ctx := contexttest.Context(t) - - epoch := ktime.ZeroTime - uattr := fs.UnstableAttr{ - AccessTime: epoch, - ModificationTime: epoch, - StatusChangeTime: epoch, - } - iops := NewCachingInodeOperations(ctx, noopBackingFile{}, uattr, CachingInodeOperationsOptions{}) - defer iops.Release() - - if err := iops.SetTimestamps(ctx, nil, test.ts); err != nil { - t.Fatalf("SetTimestamps got error %v, want nil", err) - } - if test.wantChanged.AccessTime { - if !iops.attr.AccessTime.After(uattr.AccessTime) { - t.Fatalf("diritied access time did not advance, want %v > %v", iops.attr.AccessTime, uattr.AccessTime) - } - if !iops.dirtyAttr.StatusChangeTime { - t.Fatalf("dirty access time requires dirty status change time") - } - if !iops.attr.StatusChangeTime.After(uattr.StatusChangeTime) { - t.Fatalf("dirtied status change time did not advance") - } - } - if test.wantChanged.ModificationTime { - if !iops.attr.ModificationTime.After(uattr.ModificationTime) { - t.Fatalf("diritied modification time did not advance") - } - if !iops.dirtyAttr.StatusChangeTime { - t.Fatalf("dirty modification time requires dirty status change time") - } - if !iops.attr.StatusChangeTime.After(uattr.StatusChangeTime) { - t.Fatalf("dirtied status change time did not advance") - } - } - }) - } -} - -func TestTruncate(t *testing.T) { - ctx := contexttest.Context(t) - - uattr := fs.UnstableAttr{ - Size: 0, - } - iops := NewCachingInodeOperations(ctx, noopBackingFile{}, uattr, CachingInodeOperationsOptions{}) - defer iops.Release() - - if err := iops.Truncate(ctx, nil, uattr.Size); err != nil { - t.Fatalf("Truncate got error %v, want nil", err) - } - var size int64 = 4096 - if err := iops.Truncate(ctx, nil, size); err != nil { - t.Fatalf("Truncate got error %v, want nil", err) - } - if iops.attr.Size != size { - t.Fatalf("Truncate got %d, want %d", iops.attr.Size, size) - } - if !iops.dirtyAttr.ModificationTime || !iops.dirtyAttr.StatusChangeTime { - t.Fatalf("Truncate did not dirty modification and status change time") - } - if !iops.attr.ModificationTime.After(uattr.ModificationTime) { - t.Fatalf("dirtied modification time did not change") - } - if !iops.attr.StatusChangeTime.After(uattr.StatusChangeTime) { - t.Fatalf("dirtied status change time did not change") - } -} - -type sliceBackingFile struct { - data []byte -} - -func newSliceBackingFile(data []byte) *sliceBackingFile { - return &sliceBackingFile{data} -} - -func (f *sliceBackingFile) ReadToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offset uint64) (uint64, error) { - r := safemem.BlockSeqReader{safemem.BlockSeqOf(safemem.BlockFromSafeSlice(f.data)).DropFirst64(offset)} - return r.ReadToBlocks(dsts) -} - -func (f *sliceBackingFile) WriteFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) { - w := safemem.BlockSeqWriter{safemem.BlockSeqOf(safemem.BlockFromSafeSlice(f.data)).DropFirst64(offset)} - return w.WriteFromBlocks(srcs) -} - -func (*sliceBackingFile) SetMaskedAttributes(context.Context, fs.AttrMask, fs.UnstableAttr, bool) error { - return nil -} - -func (*sliceBackingFile) Sync(context.Context) error { - return nil -} - -func (*sliceBackingFile) FD() int { - return -1 -} - -func (f *sliceBackingFile) Allocate(ctx context.Context, offset int64, length int64) error { - return syserror.EOPNOTSUPP -} - -type noopMappingSpace struct{} - -// Invalidate implements memmap.MappingSpace.Invalidate. -func (noopMappingSpace) Invalidate(ar usermem.AddrRange, opts memmap.InvalidateOpts) { -} - -func anonInode(ctx context.Context) *fs.Inode { - return fs.NewInode(ctx, &SimpleFileInode{ - InodeSimpleAttributes: NewInodeSimpleAttributes(ctx, fs.FileOwnerFromContext(ctx), fs.FilePermissions{ - User: fs.PermMask{Read: true, Write: true}, - }, 0), - }, fs.NewPseudoMountSource(ctx), fs.StableAttr{ - Type: fs.Anonymous, - BlockSize: usermem.PageSize, - }) -} - -func pagesOf(bs ...byte) []byte { - buf := make([]byte, 0, len(bs)*usermem.PageSize) - for _, b := range bs { - buf = append(buf, bytes.Repeat([]byte{b}, usermem.PageSize)...) - } - return buf -} - -func TestRead(t *testing.T) { - ctx := contexttest.Context(t) - - // Construct a 3-page file. - buf := pagesOf('a', 'b', 'c') - file := fs.NewFile(ctx, fs.NewDirent(ctx, anonInode(ctx), "anon"), fs.FileFlags{}, nil) - uattr := fs.UnstableAttr{ - Size: int64(len(buf)), - } - iops := NewCachingInodeOperations(ctx, newSliceBackingFile(buf), uattr, CachingInodeOperationsOptions{}) - defer iops.Release() - - // Expect the cache to be initially empty. - if cached := iops.cache.Span(); cached != 0 { - t.Errorf("Span got %d, want 0", cached) - } - - // Create a memory mapping of the second page (as CachingInodeOperations - // expects to only cache mapped pages), then call Translate to force it to - // be cached. - var ms noopMappingSpace - ar := usermem.AddrRange{usermem.PageSize, 2 * usermem.PageSize} - if err := iops.AddMapping(ctx, ms, ar, usermem.PageSize, true); err != nil { - t.Fatalf("AddMapping got %v, want nil", err) - } - mr := memmap.MappableRange{usermem.PageSize, 2 * usermem.PageSize} - if _, err := iops.Translate(ctx, mr, mr, usermem.Read); err != nil { - t.Fatalf("Translate got %v, want nil", err) - } - if cached := iops.cache.Span(); cached != usermem.PageSize { - t.Errorf("SpanRange got %d, want %d", cached, usermem.PageSize) - } - - // Try to read 4 pages. The first and third pages should be read directly - // from the "file", the second page should be read from the cache, and only - // 3 pages (the size of the file) should be readable. - rbuf := make([]byte, 4*usermem.PageSize) - dst := usermem.BytesIOSequence(rbuf) - n, err := iops.Read(ctx, file, dst, 0) - if n != 3*usermem.PageSize || (err != nil && err != io.EOF) { - t.Fatalf("Read got (%d, %v), want (%d, nil or EOF)", n, err, 3*usermem.PageSize) - } - rbuf = rbuf[:3*usermem.PageSize] - - // Did we get the bytes we expect? - if !bytes.Equal(rbuf, buf) { - t.Errorf("Read back bytes %v, want %v", rbuf, buf) - } - - // Delete the memory mapping before iops.Release(). The cached page will - // either be evicted by ctx's pgalloc.MemoryFile, or dropped by - // iops.Release(). - iops.RemoveMapping(ctx, ms, ar, usermem.PageSize, true) -} - -func TestWrite(t *testing.T) { - ctx := contexttest.Context(t) - - // Construct a 4-page file. - buf := pagesOf('a', 'b', 'c', 'd') - orig := append([]byte(nil), buf...) - inode := anonInode(ctx) - uattr := fs.UnstableAttr{ - Size: int64(len(buf)), - } - iops := NewCachingInodeOperations(ctx, newSliceBackingFile(buf), uattr, CachingInodeOperationsOptions{}) - defer iops.Release() - - // Expect the cache to be initially empty. - if cached := iops.cache.Span(); cached != 0 { - t.Errorf("Span got %d, want 0", cached) - } - - // Create a memory mapping of the second and third pages (as - // CachingInodeOperations expects to only cache mapped pages), then call - // Translate to force them to be cached. - var ms noopMappingSpace - ar := usermem.AddrRange{usermem.PageSize, 3 * usermem.PageSize} - if err := iops.AddMapping(ctx, ms, ar, usermem.PageSize, true); err != nil { - t.Fatalf("AddMapping got %v, want nil", err) - } - defer iops.RemoveMapping(ctx, ms, ar, usermem.PageSize, true) - mr := memmap.MappableRange{usermem.PageSize, 3 * usermem.PageSize} - if _, err := iops.Translate(ctx, mr, mr, usermem.Read); err != nil { - t.Fatalf("Translate got %v, want nil", err) - } - if cached := iops.cache.Span(); cached != 2*usermem.PageSize { - t.Errorf("SpanRange got %d, want %d", cached, 2*usermem.PageSize) - } - - // Write to the first 2 pages. - wbuf := pagesOf('e', 'f') - src := usermem.BytesIOSequence(wbuf) - n, err := iops.Write(ctx, src, 0) - if n != 2*usermem.PageSize || err != nil { - t.Fatalf("Write got (%d, %v), want (%d, nil)", n, err, 2*usermem.PageSize) - } - - // The first page should have been written directly, since it was not cached. - want := append([]byte(nil), orig...) - copy(want, pagesOf('e')) - if !bytes.Equal(buf, want) { - t.Errorf("File contents are %v, want %v", buf, want) - } - - // Sync back to the "backing file". - if err := iops.WriteOut(ctx, inode); err != nil { - t.Errorf("Sync got %v, want nil", err) - } - - // Now the second page should have been written as well. - copy(want[usermem.PageSize:], pagesOf('f')) - if !bytes.Equal(buf, want) { - t.Errorf("File contents are %v, want %v", buf, want) - } -} diff --git a/pkg/sentry/fs/g3doc/inotify.md b/pkg/sentry/fs/g3doc/inotify.md deleted file mode 100644 index 85063d4e6..000000000 --- a/pkg/sentry/fs/g3doc/inotify.md +++ /dev/null @@ -1,122 +0,0 @@ -# Inotify - -Inotify implements the like-named filesystem event notification system for the -sentry, see `inotify(7)`. - -## Architecture - -For the most part, the sentry implementation of inotify mirrors the Linux -architecture. Inotify instances (i.e. the fd returned by inotify_init(2)) are -backed by a pseudo-filesystem. Events are generated from various places in the -sentry, including the [syscall layer][syscall_dir], the [vfs layer][dirent] and -the [process fd table][fd_table]. Watches are stored in inodes and generated -events are queued to the inotify instance owning the watches for delivery to the -user. - -## Objects - -Here is a brief description of the existing and new objects involved in the -sentry inotify mechanism, and how they interact: - -### [`fs.Inotify`][inotify] - -- An inotify instances, created by inotify_init(2)/inotify_init1(2). -- The inotify fd has a `fs.Dirent`, supports filesystem syscalls to read - events. -- Has multiple `fs.Watch`es, with at most one watch per target inode, per - inotify instance. -- Has an instance `id` which is globally unique. This is *not* the fd number - for this instance, since the fd can be duped. This `id` is not externally - visible. - -### [`fs.Watch`][watch] - -- An inotify watch, created/deleted by - inotify_add_watch(2)/inotify_rm_watch(2). -- Owned by an `fs.Inotify` instance, each watch keeps a pointer to the - `owner`. -- Associated with a single `fs.Inode`, which is the watch `target`. While the - watch is active, it indirectly pins `target` to memory. See the "Reference - Model" section for a detailed explanation. -- Filesystem operations on `target` generate `fs.Event`s. - -### [`fs.Event`][event] - -- A simple struct encapsulating all the fields for an inotify event. -- Generated by `fs.Watch`es and forwarded to the watches' `owner`s. -- Serialized to the user during read(2) syscalls on the associated - `fs.Inotify`'s fd. - -### [`fs.Dirent`][dirent] - -- Many inotify events are generated inside dirent methods. Events are - generated in the dirent methods rather than `fs.Inode` methods because some - events carry the name of the subject node, and node names are generally - unavailable in an `fs.Inode`. -- Dirents do not directly contain state for any watches. Instead, they forward - notifications to the underlying `fs.Inode`. - -### [`fs.Inode`][inode] - -- Interacts with inotify through `fs.Watch`es. -- Inodes contain a map of all active `fs.Watch`es on them. -- An `fs.Inotify` instance can have at most one `fs.Watch` per inode. - `fs.Watch`es on an inode are indexed by their `owner`'s `id`. -- All inotify logic is encapsulated in the [`Watches`][inode_watches] struct - in an inode. Logically, `Watches` is the set of inotify watches on the - inode. - -## Reference Model - -The sentry inotify implementation has a complex reference model. An inotify -watch observes a single inode. For efficient lookup, the state for a watch is -stored directly on the target inode. This state needs to be persistent for the -lifetime of watch. Unlike usual filesystem metadata, the watch state has no -"on-disk" representation, so they cannot be reconstructed by the filesystem if -the inode is flushed from memory. This effectively means we need to keep any -inodes with actives watches pinned to memory. - -We can't just hold an extra ref on the inode to pin it to memory because some -filesystems (such as gofer-based filesystems) don't have persistent inodes. In -such a filesystem, if we just pin the inode, nothing prevents the enclosing -dirent from being GCed. Once the dirent is GCed, the pinned inode is -unreachable -- these filesystems generate a new inode by re-reading the node -state on the next walk. Incidentally, hardlinks also don't work on these -filesystems for this reason. - -To prevent the above scenario, when a new watch is added on an inode, we *pin* -the dirent we used to reach the inode. Note that due to hardlinks, this dirent -may not be the only dirent pointing to the inode. Attempting to set an inotify -watch via multiple hardlinks to the same file results in the same watch being -returned for both links. However, for each new dirent we use to reach the same -inode, we add a new pin. We need a new pin for each new dirent used to reach the -inode because we have no guarantees about the deletion order of the different -links to the inode. - -## Lock Ordering - -There are 4 locks related to the inotify implementation: - -- `Inotify.mu`: the inotify instance lock. -- `Inotify.evMu`: the inotify event queue lock. -- `Watch.mu`: the watch lock, used to protect pins. -- `fs.Watches.mu`: the inode watch set mu, used to protect the collection of - watches on the inode. - -The correct lock ordering for inotify code is: - -`Inotify.mu` -> `fs.Watches.mu` -> `Watch.mu` -> `Inotify.evMu`. - -We need a distinct lock for the event queue because by the time a goroutine -attempts to queue a new event, it is already holding `fs.Watches.mu`. If we used -`Inotify.mu` to also protect the event queue, this would violate the above lock -ordering. - -[dirent]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/dirent.go -[event]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify_event.go -[fd_table]: https://github.com/google/gvisor/blob/master/pkg/sentry/kernel/fd_table.go -[inode]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inode.go -[inode_watches]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inode_inotify.go -[inotify]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify.go -[syscall_dir]: https://github.com/google/gvisor/blob/master/pkg/sentry/syscalls/linux/ -[watch]: https://github.com/google/gvisor/blob/master/pkg/sentry/fs/inotify_watch.go diff --git a/pkg/sentry/fs/gofer/BUILD b/pkg/sentry/fs/gofer/BUILD deleted file mode 100644 index fea135eea..000000000 --- a/pkg/sentry/fs/gofer/BUILD +++ /dev/null @@ -1,67 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "gofer", - srcs = [ - "attr.go", - "cache_policy.go", - "context_file.go", - "device.go", - "fifo.go", - "file.go", - "file_state.go", - "fs.go", - "handles.go", - "inode.go", - "inode_state.go", - "path.go", - "session.go", - "session_state.go", - "socket.go", - "util.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fd", - "//pkg/log", - "//pkg/metric", - "//pkg/p9", - "//pkg/refs", - "//pkg/safemem", - "//pkg/secio", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fdpipe", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/host", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/pipe", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/socket/unix/transport", - "//pkg/sync", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/unet", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "gofer_test", - size = "small", - srcs = ["gofer_test.go"], - library = ":gofer", - deps = [ - "//pkg/context", - "//pkg/p9", - "//pkg/p9/p9test", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - ], -) diff --git a/pkg/sentry/fs/gofer/fifo.go b/pkg/sentry/fs/gofer/fifo.go index 456557058..456557058 100644..100755 --- a/pkg/sentry/fs/gofer/fifo.go +++ b/pkg/sentry/fs/gofer/fifo.go diff --git a/pkg/sentry/fs/gofer/gofer_state_autogen.go b/pkg/sentry/fs/gofer/gofer_state_autogen.go new file mode 100755 index 000000000..7db9211b4 --- /dev/null +++ b/pkg/sentry/fs/gofer/gofer_state_autogen.go @@ -0,0 +1,145 @@ +// automatically generated by stateify. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *fifo) beforeSave() {} +func (x *fifo) save(m state.Map) { + x.beforeSave() + m.Save("InodeOperations", &x.InodeOperations) + m.Save("fileIops", &x.fileIops) +} + +func (x *fifo) afterLoad() {} +func (x *fifo) load(m state.Map) { + m.Load("InodeOperations", &x.InodeOperations) + m.Load("fileIops", &x.fileIops) +} + +func (x *fileOperations) beforeSave() {} +func (x *fileOperations) save(m state.Map) { + x.beforeSave() + m.Save("inodeOperations", &x.inodeOperations) + m.Save("dirCursor", &x.dirCursor) + m.Save("flags", &x.flags) +} + +func (x *fileOperations) load(m state.Map) { + m.LoadWait("inodeOperations", &x.inodeOperations) + m.Load("dirCursor", &x.dirCursor) + m.LoadWait("flags", &x.flags) + m.AfterLoad(x.afterLoad) +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *inodeOperations) beforeSave() {} +func (x *inodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("fileState", &x.fileState) + m.Save("cachingInodeOps", &x.cachingInodeOps) +} + +func (x *inodeOperations) afterLoad() {} +func (x *inodeOperations) load(m state.Map) { + m.LoadWait("fileState", &x.fileState) + m.Load("cachingInodeOps", &x.cachingInodeOps) +} + +func (x *inodeFileState) save(m state.Map) { + x.beforeSave() + var loading struct{} = x.saveLoading() + m.SaveValue("loading", loading) + m.Save("s", &x.s) + m.Save("sattr", &x.sattr) + m.Save("savedUAttr", &x.savedUAttr) + m.Save("hostMappable", &x.hostMappable) +} + +func (x *inodeFileState) load(m state.Map) { + m.LoadWait("s", &x.s) + m.LoadWait("sattr", &x.sattr) + m.Load("savedUAttr", &x.savedUAttr) + m.Load("hostMappable", &x.hostMappable) + m.LoadValue("loading", new(struct{}), func(y interface{}) { x.loadLoading(y.(struct{})) }) + m.AfterLoad(x.afterLoad) +} + +func (x *overrideInfo) beforeSave() {} +func (x *overrideInfo) save(m state.Map) { + x.beforeSave() + m.Save("dirent", &x.dirent) + m.Save("endpoint", &x.endpoint) + m.Save("inode", &x.inode) +} + +func (x *overrideInfo) afterLoad() {} +func (x *overrideInfo) load(m state.Map) { + m.Load("dirent", &x.dirent) + m.Load("endpoint", &x.endpoint) + m.Load("inode", &x.inode) +} + +func (x *overrideMaps) beforeSave() {} +func (x *overrideMaps) save(m state.Map) { + x.beforeSave() + m.Save("pathMap", &x.pathMap) +} + +func (x *overrideMaps) afterLoad() {} +func (x *overrideMaps) load(m state.Map) { + m.Load("pathMap", &x.pathMap) +} + +func (x *session) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("msize", &x.msize) + m.Save("version", &x.version) + m.Save("cachePolicy", &x.cachePolicy) + m.Save("aname", &x.aname) + m.Save("superBlockFlags", &x.superBlockFlags) + m.Save("limitHostFDTranslation", &x.limitHostFDTranslation) + m.Save("overlayfsStaleRead", &x.overlayfsStaleRead) + m.Save("connID", &x.connID) + m.Save("inodeMappings", &x.inodeMappings) + m.Save("mounter", &x.mounter) + m.Save("overrides", &x.overrides) +} + +func (x *session) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.LoadWait("msize", &x.msize) + m.LoadWait("version", &x.version) + m.LoadWait("cachePolicy", &x.cachePolicy) + m.LoadWait("aname", &x.aname) + m.LoadWait("superBlockFlags", &x.superBlockFlags) + m.Load("limitHostFDTranslation", &x.limitHostFDTranslation) + m.Load("overlayfsStaleRead", &x.overlayfsStaleRead) + m.LoadWait("connID", &x.connID) + m.LoadWait("inodeMappings", &x.inodeMappings) + m.LoadWait("mounter", &x.mounter) + m.LoadWait("overrides", &x.overrides) + m.AfterLoad(x.afterLoad) +} + +func init() { + state.Register("pkg/sentry/fs/gofer.fifo", (*fifo)(nil), state.Fns{Save: (*fifo).save, Load: (*fifo).load}) + state.Register("pkg/sentry/fs/gofer.fileOperations", (*fileOperations)(nil), state.Fns{Save: (*fileOperations).save, Load: (*fileOperations).load}) + state.Register("pkg/sentry/fs/gofer.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("pkg/sentry/fs/gofer.inodeOperations", (*inodeOperations)(nil), state.Fns{Save: (*inodeOperations).save, Load: (*inodeOperations).load}) + state.Register("pkg/sentry/fs/gofer.inodeFileState", (*inodeFileState)(nil), state.Fns{Save: (*inodeFileState).save, Load: (*inodeFileState).load}) + state.Register("pkg/sentry/fs/gofer.overrideInfo", (*overrideInfo)(nil), state.Fns{Save: (*overrideInfo).save, Load: (*overrideInfo).load}) + state.Register("pkg/sentry/fs/gofer.overrideMaps", (*overrideMaps)(nil), state.Fns{Save: (*overrideMaps).save, Load: (*overrideMaps).load}) + state.Register("pkg/sentry/fs/gofer.session", (*session)(nil), state.Fns{Save: (*session).save, Load: (*session).load}) +} diff --git a/pkg/sentry/fs/gofer/gofer_test.go b/pkg/sentry/fs/gofer/gofer_test.go deleted file mode 100644 index 2df2fe889..000000000 --- a/pkg/sentry/fs/gofer/gofer_test.go +++ /dev/null @@ -1,310 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "fmt" - "syscall" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/p9/p9test" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// rootTest runs a test with a p9 mock and an fs.InodeOperations created from -// the attached root directory. The root file will be closed and client -// disconnected, but additional files must be closed manually. -func rootTest(t *testing.T, name string, cp cachePolicy, fn func(context.Context, *p9test.Harness, *p9test.Mock, *fs.Inode)) { - t.Run(name, func(t *testing.T) { - h, c := p9test.NewHarness(t) - defer h.Finish() - - // Create a new root. Note that we pass an empty, but non-nil - // map here. This allows tests to extend the root children - // dynamically. - root := h.NewDirectory(map[string]p9test.Generator{})(nil) - - // Return this as the root. - h.Attacher.EXPECT().Attach().Return(root, nil).Times(1) - - // ... and open via the client. - rootFile, err := c.Attach("/") - if err != nil { - t.Fatalf("unable to attach: %v", err) - } - defer rootFile.Close() - - // Wrap an a session. - s := &session{ - mounter: fs.RootOwner, - cachePolicy: cp, - client: c, - } - - // ... and an INode, with only the mode being explicitly valid for now. - ctx := contexttest.Context(t) - sattr, rootInodeOperations := newInodeOperations(ctx, s, contextFile{ - file: rootFile, - }, root.QID, p9.AttrMaskAll(), root.Attr) - m := fs.NewMountSource(ctx, s, &filesystem{}, fs.MountSourceFlags{}) - rootInode := fs.NewInode(ctx, rootInodeOperations, m, sattr) - - // Ensure that the cache is fully invalidated, so that any - // close actions actually take place before the full harness is - // torn down. - defer func() { - m.FlushDirentRefs() - - // Wait for all resources to be released, otherwise the - // operations may fail after we close the rootFile. - fs.AsyncBarrier() - }() - - // Execute the test. - fn(ctx, h, root, rootInode) - }) -} - -func TestLookup(t *testing.T) { - type lookupTest struct { - // Name of the test. - name string - - // Expected return value. - want error - } - - tests := []lookupTest{ - { - name: "mock Walk passes (function succeeds)", - want: nil, - }, - { - name: "mock Walk fails (function fails)", - want: syscall.ENOENT, - }, - } - - const file = "file" // The walked target file. - - for _, test := range tests { - rootTest(t, test.name, cacheNone, func(ctx context.Context, h *p9test.Harness, rootFile *p9test.Mock, rootInode *fs.Inode) { - // Setup the appropriate result. - rootFile.WalkCallback = func() error { - return test.want - } - if test.want == nil { - // Set the contents of the root. We expect a - // normal file generator for ppp above. This is - // overriden by setting WalkErr in the mock. - rootFile.AddChild(file, h.NewFile()) - } - - // Call function. - dirent, err := rootInode.Lookup(ctx, file) - - // Unwrap the InodeOperations. - var newInodeOperations fs.InodeOperations - if dirent != nil { - if dirent.IsNegative() { - err = syscall.ENOENT - } else { - newInodeOperations = dirent.Inode.InodeOperations - } - } - - // Check return values. - if err != test.want { - t.Errorf("Lookup got err %v, want %v", err, test.want) - } - if err == nil && newInodeOperations == nil { - t.Errorf("Lookup got non-nil err and non-nil node, wanted at least one non-nil") - } - }) - } -} - -func TestRevalidation(t *testing.T) { - type revalidationTest struct { - cachePolicy cachePolicy - - // Whether dirent should be reloaded before any modifications. - preModificationWantReload bool - - // Whether dirent should be reloaded after updating an unstable - // attribute on the remote fs. - postModificationWantReload bool - - // Whether dirent unstable attributes should be updated after - // updating an attribute on the remote fs. - postModificationWantUpdatedAttrs bool - - // Whether dirent should be reloaded after the remote has - // removed the file. - postRemovalWantReload bool - } - - tests := []revalidationTest{ - { - // Policy cacheNone causes Revalidate to always return - // true. - cachePolicy: cacheNone, - preModificationWantReload: true, - postModificationWantReload: true, - postModificationWantUpdatedAttrs: true, - postRemovalWantReload: true, - }, - { - // Policy cacheAll causes Revalidate to always return - // false. - cachePolicy: cacheAll, - preModificationWantReload: false, - postModificationWantReload: false, - postModificationWantUpdatedAttrs: false, - postRemovalWantReload: false, - }, - { - // Policy cacheAllWritethrough causes Revalidate to - // always return false. - cachePolicy: cacheAllWritethrough, - preModificationWantReload: false, - postModificationWantReload: false, - postModificationWantUpdatedAttrs: false, - postRemovalWantReload: false, - }, - { - // Policy cacheRemoteRevalidating causes Revalidate to - // return update cached unstable attrs, and returns - // true only when the remote inode itself has been - // removed or replaced. - cachePolicy: cacheRemoteRevalidating, - preModificationWantReload: false, - postModificationWantReload: false, - postModificationWantUpdatedAttrs: true, - postRemovalWantReload: true, - }, - } - - const file = "file" // The file walked below. - - for _, test := range tests { - name := fmt.Sprintf("cachepolicy=%s", test.cachePolicy) - rootTest(t, name, test.cachePolicy, func(ctx context.Context, h *p9test.Harness, rootFile *p9test.Mock, rootInode *fs.Inode) { - // Wrap in a dirent object. - rootDir := fs.NewDirent(ctx, rootInode, "root") - - // Create a mock file a child of the root. We save when - // this is generated, so that when the time changed, we - // can update the original entry. - var origMocks []*p9test.Mock - rootFile.AddChild(file, func(parent *p9test.Mock) *p9test.Mock { - // Regular a regular file that has a consistent - // path number. This might be used by - // validation so we don't change it. - m := h.NewMock(parent, 0, p9.Attr{ - Mode: p9.ModeRegular, - }) - origMocks = append(origMocks, m) - return m - }) - - // Do the walk. - dirent, err := rootDir.Walk(ctx, rootDir, file) - if err != nil { - t.Fatalf("Lookup failed: %v", err) - } - - // We must release the dirent, of the test will fail - // with a reference leak. This is tracked by p9test. - defer dirent.DecRef() - - // Walk again. Depending on the cache policy, we may - // get a new dirent. - newDirent, err := rootDir.Walk(ctx, rootDir, file) - if err != nil { - t.Fatalf("Lookup failed: %v", err) - } - if test.preModificationWantReload && dirent == newDirent { - t.Errorf("Lookup with cachePolicy=%s got old dirent %+v, wanted a new dirent", test.cachePolicy, dirent) - } - if !test.preModificationWantReload && dirent != newDirent { - t.Errorf("Lookup with cachePolicy=%s got new dirent %+v, wanted old dirent %+v", test.cachePolicy, newDirent, dirent) - } - newDirent.DecRef() // See above. - - // Modify the underlying mocked file's modification - // time for the next walk that occurs. - nowSeconds := time.Now().Unix() - rootFile.AddChild(file, func(parent *p9test.Mock) *p9test.Mock { - // Ensure that the path is the same as above, - // but we change only the modification time of - // the file. - return h.NewMock(parent, 0, p9.Attr{ - Mode: p9.ModeRegular, - MTimeSeconds: uint64(nowSeconds), - }) - }) - - // We also modify the original time, so that GetAttr - // behaves as expected for the caching case. - for _, m := range origMocks { - m.Attr.MTimeSeconds = uint64(nowSeconds) - } - - // Walk again. Depending on the cache policy, we may - // get a new dirent. - newDirent, err = rootDir.Walk(ctx, rootDir, file) - if err != nil { - t.Fatalf("Lookup failed: %v", err) - } - if test.postModificationWantReload && dirent == newDirent { - t.Errorf("Lookup with cachePolicy=%s got old dirent, wanted a new dirent", test.cachePolicy) - } - if !test.postModificationWantReload && dirent != newDirent { - t.Errorf("Lookup with cachePolicy=%s got new dirent, wanted old dirent", test.cachePolicy) - } - uattrs, err := newDirent.Inode.UnstableAttr(ctx) - if err != nil { - t.Fatalf("Error getting unstable attrs: %v", err) - } - gotModTimeSeconds := uattrs.ModificationTime.Seconds() - if test.postModificationWantUpdatedAttrs && gotModTimeSeconds != nowSeconds { - t.Fatalf("Lookup with cachePolicy=%s got new modification time %v, wanted %v", test.cachePolicy, gotModTimeSeconds, nowSeconds) - } - newDirent.DecRef() // See above. - - // Remove the file from the remote fs, subsequent walks - // should now fail to find anything. - rootFile.RemoveChild(file) - - // Walk again. Depending on the cache policy, we may - // get ENOENT. - newDirent, err = rootDir.Walk(ctx, rootDir, file) - if test.postRemovalWantReload && err == nil { - t.Errorf("Lookup with cachePolicy=%s got nil error, wanted ENOENT", test.cachePolicy) - } - if !test.postRemovalWantReload && (err != nil || dirent != newDirent) { - t.Errorf("Lookup with cachePolicy=%s got new dirent and error %v, wanted old dirent and nil error", test.cachePolicy, err) - } - if err == nil { - newDirent.DecRef() // See above. - } - }) - } -} diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD deleted file mode 100644 index 011625c80..000000000 --- a/pkg/sentry/fs/host/BUILD +++ /dev/null @@ -1,83 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "host", - srcs = [ - "control.go", - "descriptor.go", - "descriptor_state.go", - "device.go", - "file.go", - "host.go", - "inode.go", - "inode_state.go", - "ioctl_unsafe.go", - "socket.go", - "socket_iovec.go", - "socket_state.go", - "socket_unsafe.go", - "tty.go", - "util.go", - "util_amd64_unsafe.go", - "util_arm64_unsafe.go", - "util_unsafe.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fd", - "//pkg/fdnotifier", - "//pkg/log", - "//pkg/refs", - "//pkg/safemem", - "//pkg/secio", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/socket/control", - "//pkg/sentry/socket/unix", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/unimpl", - "//pkg/sentry/uniqueid", - "//pkg/sync", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/unet", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "host_test", - size = "small", - srcs = [ - "descriptor_test.go", - "inode_test.go", - "socket_test.go", - "wait_test.go", - ], - library = ":host", - deps = [ - "//pkg/fd", - "//pkg/fdnotifier", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket", - "//pkg/sentry/socket/unix/transport", - "//pkg/syserr", - "//pkg/tcpip", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/fs/host/descriptor_test.go b/pkg/sentry/fs/host/descriptor_test.go deleted file mode 100644 index d8e4605b6..000000000 --- a/pkg/sentry/fs/host/descriptor_test.go +++ /dev/null @@ -1,78 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "io/ioutil" - "path/filepath" - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestDescriptorRelease(t *testing.T) { - for _, tc := range []struct { - name string - saveable bool - wouldBlock bool - }{ - {name: "all false"}, - {name: "saveable", saveable: true}, - {name: "wouldBlock", wouldBlock: true}, - } { - t.Run(tc.name, func(t *testing.T) { - dir, err := ioutil.TempDir("", "descriptor_test") - if err != nil { - t.Fatal("ioutil.TempDir() failed:", err) - } - - fd, err := syscall.Open(filepath.Join(dir, "file"), syscall.O_RDWR|syscall.O_CREAT, 0666) - if err != nil { - t.Fatal("failed to open temp file:", err) - } - - // FD ownership is transferred to the descritor. - queue := &waiter.Queue{} - d, err := newDescriptor(fd, tc.saveable, tc.wouldBlock, queue) - if err != nil { - syscall.Close(fd) - t.Fatalf("newDescriptor(%d, %t, %t, queue) failed, err: %v", fd, tc.saveable, tc.wouldBlock, err) - } - if tc.saveable { - if d.origFD < 0 { - t.Errorf("saveable descriptor must preserve origFD, desc: %+v", d) - } - } - if tc.wouldBlock { - if !fdnotifier.HasFD(int32(d.value)) { - t.Errorf("FD not registered with notifier, desc: %+v", d) - } - } - - oldVal := d.value - d.Release() - if d.value != -1 { - t.Errorf("d.value want: -1, got: %d", d.value) - } - if tc.wouldBlock { - if fdnotifier.HasFD(int32(oldVal)) { - t.Errorf("FD not unregistered with notifier, desc: %+v", d) - } - } - }) - } -} diff --git a/pkg/sentry/fs/host/host.go b/pkg/sentry/fs/host/host.go index 081ba1dd8..081ba1dd8 100644..100755 --- a/pkg/sentry/fs/host/host.go +++ b/pkg/sentry/fs/host/host.go diff --git a/pkg/sentry/fs/host/host_amd64_unsafe_state_autogen.go b/pkg/sentry/fs/host/host_amd64_unsafe_state_autogen.go new file mode 100755 index 000000000..488cbdfcf --- /dev/null +++ b/pkg/sentry/fs/host/host_amd64_unsafe_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build amd64 + +package host diff --git a/pkg/sentry/fs/host/host_arm64_unsafe_state_autogen.go b/pkg/sentry/fs/host/host_arm64_unsafe_state_autogen.go new file mode 100755 index 000000000..7371b44db --- /dev/null +++ b/pkg/sentry/fs/host/host_arm64_unsafe_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build arm64 + +package host diff --git a/pkg/sentry/fs/host/host_state_autogen.go b/pkg/sentry/fs/host/host_state_autogen.go new file mode 100755 index 000000000..8351ece3b --- /dev/null +++ b/pkg/sentry/fs/host/host_state_autogen.go @@ -0,0 +1,117 @@ +// automatically generated by stateify. + +package host + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *descriptor) save(m state.Map) { + x.beforeSave() + m.Save("origFD", &x.origFD) + m.Save("wouldBlock", &x.wouldBlock) +} + +func (x *descriptor) load(m state.Map) { + m.Load("origFD", &x.origFD) + m.Load("wouldBlock", &x.wouldBlock) + m.AfterLoad(x.afterLoad) +} + +func (x *fileOperations) beforeSave() {} +func (x *fileOperations) save(m state.Map) { + x.beforeSave() + m.Save("iops", &x.iops) + m.Save("dirCursor", &x.dirCursor) +} + +func (x *fileOperations) afterLoad() {} +func (x *fileOperations) load(m state.Map) { + m.LoadWait("iops", &x.iops) + m.Load("dirCursor", &x.dirCursor) +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *inodeOperations) beforeSave() {} +func (x *inodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("fileState", &x.fileState) + m.Save("cachingInodeOps", &x.cachingInodeOps) +} + +func (x *inodeOperations) afterLoad() {} +func (x *inodeOperations) load(m state.Map) { + m.LoadWait("fileState", &x.fileState) + m.Load("cachingInodeOps", &x.cachingInodeOps) +} + +func (x *inodeFileState) beforeSave() {} +func (x *inodeFileState) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.queue) { + m.Failf("queue is %v, expected zero", x.queue) + } + m.Save("descriptor", &x.descriptor) + m.Save("sattr", &x.sattr) + m.Save("savedUAttr", &x.savedUAttr) +} + +func (x *inodeFileState) load(m state.Map) { + m.LoadWait("descriptor", &x.descriptor) + m.LoadWait("sattr", &x.sattr) + m.Load("savedUAttr", &x.savedUAttr) + m.AfterLoad(x.afterLoad) +} + +func (x *ConnectedEndpoint) save(m state.Map) { + x.beforeSave() + m.Save("ref", &x.ref) + m.Save("queue", &x.queue) + m.Save("path", &x.path) + m.Save("srfd", &x.srfd) + m.Save("stype", &x.stype) +} + +func (x *ConnectedEndpoint) load(m state.Map) { + m.Load("ref", &x.ref) + m.Load("queue", &x.queue) + m.Load("path", &x.path) + m.LoadWait("srfd", &x.srfd) + m.Load("stype", &x.stype) + m.AfterLoad(x.afterLoad) +} + +func (x *TTYFileOperations) beforeSave() {} +func (x *TTYFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("fileOperations", &x.fileOperations) + m.Save("session", &x.session) + m.Save("fgProcessGroup", &x.fgProcessGroup) + m.Save("termios", &x.termios) +} + +func (x *TTYFileOperations) afterLoad() {} +func (x *TTYFileOperations) load(m state.Map) { + m.Load("fileOperations", &x.fileOperations) + m.Load("session", &x.session) + m.Load("fgProcessGroup", &x.fgProcessGroup) + m.Load("termios", &x.termios) +} + +func init() { + state.Register("pkg/sentry/fs/host.descriptor", (*descriptor)(nil), state.Fns{Save: (*descriptor).save, Load: (*descriptor).load}) + state.Register("pkg/sentry/fs/host.fileOperations", (*fileOperations)(nil), state.Fns{Save: (*fileOperations).save, Load: (*fileOperations).load}) + state.Register("pkg/sentry/fs/host.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("pkg/sentry/fs/host.inodeOperations", (*inodeOperations)(nil), state.Fns{Save: (*inodeOperations).save, Load: (*inodeOperations).load}) + state.Register("pkg/sentry/fs/host.inodeFileState", (*inodeFileState)(nil), state.Fns{Save: (*inodeFileState).save, Load: (*inodeFileState).load}) + state.Register("pkg/sentry/fs/host.ConnectedEndpoint", (*ConnectedEndpoint)(nil), state.Fns{Save: (*ConnectedEndpoint).save, Load: (*ConnectedEndpoint).load}) + state.Register("pkg/sentry/fs/host.TTYFileOperations", (*TTYFileOperations)(nil), state.Fns{Save: (*TTYFileOperations).save, Load: (*TTYFileOperations).load}) +} diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go deleted file mode 100644 index 4c374681c..000000000 --- a/pkg/sentry/fs/host/inode_test.go +++ /dev/null @@ -1,46 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// TestCloseFD verifies fds will be closed. -func TestCloseFD(t *testing.T) { - var p [2]int - if err := syscall.Pipe(p[0:]); err != nil { - t.Fatalf("Failed to create pipe %v", err) - } - defer syscall.Close(p[0]) - defer syscall.Close(p[1]) - - // Use the write-end because we will detect if it's closed on the read end. - ctx := contexttest.Context(t) - file, err := NewFile(ctx, p[1], fs.RootOwner) - if err != nil { - t.Fatalf("Failed to create File: %v", err) - } - file.DecRef() - - s := make([]byte, 10) - if c, err := syscall.Read(p[0], s); c != 0 || err != nil { - t.Errorf("want 0, nil (EOF) from read end, got %v, %v", c, err) - } -} diff --git a/pkg/sentry/fs/host/socket_test.go b/pkg/sentry/fs/host/socket_test.go deleted file mode 100644 index eb4afe520..000000000 --- a/pkg/sentry/fs/host/socket_test.go +++ /dev/null @@ -1,246 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "reflect" - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -var ( - // Make sure that ConnectedEndpoint implements transport.ConnectedEndpoint. - _ = transport.ConnectedEndpoint(new(ConnectedEndpoint)) - - // Make sure that ConnectedEndpoint implements transport.Receiver. - _ = transport.Receiver(new(ConnectedEndpoint)) -) - -func getFl(fd int) (uint32, error) { - fl, _, err := syscall.RawSyscall(syscall.SYS_FCNTL, uintptr(fd), syscall.F_GETFL, 0) - if err == 0 { - return uint32(fl), nil - } - return 0, err -} - -func TestSocketIsBlocking(t *testing.T) { - // Using socketpair here because it's already connected. - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("host socket creation failed: %v", err) - } - - fl, err := getFl(pair[0]) - if err != nil { - t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[0], err) - } - if fl&syscall.O_NONBLOCK == syscall.O_NONBLOCK { - t.Fatalf("Expected socket %v to be blocking", pair[0]) - } - if fl, err = getFl(pair[1]); err != nil { - t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[1], err) - } - if fl&syscall.O_NONBLOCK == syscall.O_NONBLOCK { - t.Fatalf("Expected socket %v to be blocking", pair[1]) - } - sock, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) failed => %v", pair[0], err) - } - defer sock.DecRef() - // Test that the socket now is non-blocking. - if fl, err = getFl(pair[0]); err != nil { - t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[0], err) - } - if fl&syscall.O_NONBLOCK != syscall.O_NONBLOCK { - t.Errorf("Expected socket %v to have become non-blocking", pair[0]) - } - if fl, err = getFl(pair[1]); err != nil { - t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[1], err) - } - if fl&syscall.O_NONBLOCK == syscall.O_NONBLOCK { - t.Errorf("Did not expect socket %v to become non-blocking", pair[1]) - } -} - -func TestSocketWritev(t *testing.T) { - // Using socketpair here because it's already connected. - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("host socket creation failed: %v", err) - } - socket, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[0], err) - } - defer socket.DecRef() - buf := []byte("hello world\n") - n, err := socket.Writev(contexttest.Context(t), usermem.BytesIOSequence(buf)) - if err != nil { - t.Fatalf("socket writev failed: %v", err) - } - - if n != int64(len(buf)) { - t.Fatalf("socket writev wrote incorrect bytes: %d", n) - } -} - -func TestSocketWritevLen0(t *testing.T) { - // Using socketpair here because it's already connected. - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("host socket creation failed: %v", err) - } - socket, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[0], err) - } - defer socket.DecRef() - n, err := socket.Writev(contexttest.Context(t), usermem.BytesIOSequence(nil)) - if err != nil { - t.Fatalf("socket writev failed: %v", err) - } - - if n != 0 { - t.Fatalf("socket writev wrote incorrect bytes: %d", n) - } -} - -func TestSocketSendMsgLen0(t *testing.T) { - // Using socketpair here because it's already connected. - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("host socket creation failed: %v", err) - } - sfile, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[0], err) - } - defer sfile.DecRef() - - s := sfile.FileOperations.(socket.Socket) - n, terr := s.SendMsg(nil, usermem.BytesIOSequence(nil), []byte{}, 0, false, ktime.Time{}, socket.ControlMessages{}) - if n != 0 { - t.Fatalf("socket sendmsg() failed: %v wrote: %d", terr, n) - } - - if terr != nil { - t.Fatalf("socket sendmsg() failed: %v", terr) - } -} - -func TestListen(t *testing.T) { - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) => %v", err) - } - sfile1, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[0], err) - } - defer sfile1.DecRef() - socket1 := sfile1.FileOperations.(socket.Socket) - - sfile2, err := newSocket(contexttest.Context(t), pair[1], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[1], err) - } - defer sfile2.DecRef() - socket2 := sfile2.FileOperations.(socket.Socket) - - // Socketpairs can not be listened to. - if err := socket1.Listen(nil, 64); err != syserr.ErrInvalidEndpointState { - t.Fatalf("socket1.Listen(nil, 64) => %v, want syserr.ErrInvalidEndpointState", err) - } - if err := socket2.Listen(nil, 64); err != syserr.ErrInvalidEndpointState { - t.Fatalf("socket2.Listen(nil, 64) => %v, want syserr.ErrInvalidEndpointState", err) - } - - // Create a Unix socket, do not bind it. - sock, err := syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) => %v", err) - } - sfile3, err := newSocket(contexttest.Context(t), sock, false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", sock, err) - } - defer sfile3.DecRef() - socket3 := sfile3.FileOperations.(socket.Socket) - - // This socket is not bound so we can't listen on it. - if err := socket3.Listen(nil, 64); err != syserr.ErrInvalidEndpointState { - t.Fatalf("socket3.Listen(nil, 64) => %v, want syserr.ErrInvalidEndpointState", err) - } -} - -func TestPasscred(t *testing.T) { - e := ConnectedEndpoint{} - if got, want := e.Passcred(), false; got != want { - t.Errorf("Got %#v.Passcred() = %t, want = %t", e, got, want) - } -} - -func TestGetLocalAddress(t *testing.T) { - e := ConnectedEndpoint{path: "foo"} - want := tcpip.FullAddress{Addr: tcpip.Address("foo")} - if got, err := e.GetLocalAddress(); err != nil || got != want { - t.Errorf("Got %#v.GetLocalAddress() = %#v, %v, want = %#v, %v", e, got, err, want, nil) - } -} - -func TestQueuedSize(t *testing.T) { - e := ConnectedEndpoint{} - tests := []struct { - name string - f func() int64 - }{ - {"SendQueuedSize", e.SendQueuedSize}, - {"RecvQueuedSize", e.RecvQueuedSize}, - } - - for _, test := range tests { - if got, want := test.f(), int64(-1); got != want { - t.Errorf("Got %#v.%s() = %d, want = %d", e, test.name, got, want) - } - } -} - -func TestRelease(t *testing.T) { - f, err := syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM|syscall.SOCK_NONBLOCK|syscall.SOCK_CLOEXEC, 0) - if err != nil { - t.Fatal("Creating socket:", err) - } - c := &ConnectedEndpoint{queue: &waiter.Queue{}, file: fd.New(f)} - want := &ConnectedEndpoint{queue: c.queue} - want.ref.DecRef() - fdnotifier.AddFD(int32(c.file.FD()), nil) - c.Release() - if !reflect.DeepEqual(c, want) { - t.Errorf("got = %#v, want = %#v", c, want) - } -} diff --git a/pkg/sentry/fs/host/util_amd64_unsafe.go b/pkg/sentry/fs/host/util_amd64_unsafe.go index 66da6e9f5..66da6e9f5 100644..100755 --- a/pkg/sentry/fs/host/util_amd64_unsafe.go +++ b/pkg/sentry/fs/host/util_amd64_unsafe.go diff --git a/pkg/sentry/fs/host/util_arm64_unsafe.go b/pkg/sentry/fs/host/util_arm64_unsafe.go index e8cb94aeb..e8cb94aeb 100644..100755 --- a/pkg/sentry/fs/host/util_arm64_unsafe.go +++ b/pkg/sentry/fs/host/util_arm64_unsafe.go diff --git a/pkg/sentry/fs/host/wait_test.go b/pkg/sentry/fs/host/wait_test.go deleted file mode 100644 index d49c3a635..000000000 --- a/pkg/sentry/fs/host/wait_test.go +++ /dev/null @@ -1,70 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "syscall" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestWait(t *testing.T) { - var fds [2]int - err := syscall.Pipe(fds[:]) - if err != nil { - t.Fatalf("Unable to create pipe: %v", err) - } - - defer syscall.Close(fds[1]) - - ctx := contexttest.Context(t) - file, err := NewFile(ctx, fds[0], fs.RootOwner) - if err != nil { - syscall.Close(fds[0]) - t.Fatalf("NewFile failed: %v", err) - } - - defer file.DecRef() - - r := file.Readiness(waiter.EventIn) - if r != 0 { - t.Fatalf("File is ready for read when it shouldn't be.") - } - - e, ch := waiter.NewChannelEntry(nil) - file.EventRegister(&e, waiter.EventIn) - defer file.EventUnregister(&e) - - // Check that there are no notifications yet. - if len(ch) != 0 { - t.Fatalf("Channel is non-empty") - } - - // Write to the pipe, so it should be writable now. - syscall.Write(fds[1], []byte{1}) - - // Check that we get a notification. We need to yield the current thread - // so that the fdnotifier can deliver notifications, so we use a - // 1-second timeout instead of just checking the length of the channel. - select { - case <-ch: - case <-time.After(1 * time.Second): - t.Fatalf("Channel not notified") - } -} diff --git a/pkg/sentry/fs/inode_overlay_test.go b/pkg/sentry/fs/inode_overlay_test.go deleted file mode 100644 index 389c219d6..000000000 --- a/pkg/sentry/fs/inode_overlay_test.go +++ /dev/null @@ -1,470 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs_test - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/syserror" -) - -func TestLookup(t *testing.T) { - ctx := contexttest.Context(t) - for _, test := range []struct { - // Test description. - desc string - - // Lookup parameters. - dir *fs.Inode - name string - - // Want from lookup. - found bool - hasUpper bool - hasLower bool - }{ - { - desc: "no upper, lower has name", - dir: fs.NewTestOverlayDir(ctx, - nil, /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: false, - hasLower: true, - }, - { - desc: "no lower, upper has name", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* upper */ - nil, /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: true, - hasLower: false, - }, - { - desc: "upper and lower, only lower has name", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "b", - dir: false, - }, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: false, - hasLower: true, - }, - { - desc: "upper and lower, only upper has name", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "b", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: true, - hasLower: false, - }, - { - desc: "upper and lower, both have file", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: true, - hasLower: false, - }, - { - desc: "upper and lower, both have directory", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: true, - }, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: true, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: true, - hasLower: true, - }, - { - desc: "upper and lower, upper negative masks lower file", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, nil, []string{"a"}), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: false, - hasUpper: false, - hasLower: false, - }, - { - desc: "upper and lower, upper negative does not mask lower file", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, nil, []string{"b"}), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: false, - hasLower: true, - }, - } { - t.Run(test.desc, func(t *testing.T) { - dirent, err := test.dir.Lookup(ctx, test.name) - if test.found && (err == syserror.ENOENT || dirent.IsNegative()) { - t.Fatalf("lookup %q expected to find positive dirent, got dirent %v err %v", test.name, dirent, err) - } - if !test.found { - if err != syserror.ENOENT && !dirent.IsNegative() { - t.Errorf("lookup %q expected to return ENOENT or negative dirent, got dirent %v err %v", test.name, dirent, err) - } - // Nothing more to check. - return - } - if hasUpper := dirent.Inode.TestHasUpperFS(); hasUpper != test.hasUpper { - t.Fatalf("lookup got upper filesystem %v, want %v", hasUpper, test.hasUpper) - } - if hasLower := dirent.Inode.TestHasLowerFS(); hasLower != test.hasLower { - t.Errorf("lookup got lower filesystem %v, want %v", hasLower, test.hasLower) - } - }) - } -} - -func TestLookupRevalidation(t *testing.T) { - // File name used in the tests. - fileName := "foofile" - ctx := contexttest.Context(t) - for _, tc := range []struct { - // Test description. - desc string - - // Upper and lower fs for the overlay. - upper *fs.Inode - lower *fs.Inode - - // Whether the upper requires revalidation. - revalidate bool - - // Whether we should get the same dirent on second lookup. - wantSame bool - }{ - { - desc: "file from upper with no revalidation", - upper: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - lower: newTestRamfsDir(ctx, nil, nil), - revalidate: false, - wantSame: true, - }, - { - desc: "file from upper with revalidation", - upper: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - lower: newTestRamfsDir(ctx, nil, nil), - revalidate: true, - wantSame: false, - }, - { - desc: "file from lower with no revalidation", - upper: newTestRamfsDir(ctx, nil, nil), - lower: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - revalidate: false, - wantSame: true, - }, - { - desc: "file from lower with revalidation", - upper: newTestRamfsDir(ctx, nil, nil), - lower: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - revalidate: true, - // The file does not exist in the upper, so we do not - // need to revalidate it. - wantSame: true, - }, - { - desc: "file from upper and lower with no revalidation", - upper: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - lower: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - revalidate: false, - wantSame: true, - }, - { - desc: "file from upper and lower with revalidation", - upper: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - lower: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - revalidate: true, - wantSame: false, - }, - } { - t.Run(tc.desc, func(t *testing.T) { - root := fs.NewDirent(ctx, newTestRamfsDir(ctx, nil, nil), "root") - ctx = &rootContext{ - Context: ctx, - root: root, - } - overlay := fs.NewDirent(ctx, fs.NewTestOverlayDir(ctx, tc.upper, tc.lower, tc.revalidate), "overlay") - // Lookup the file twice through the overlay. - first, err := overlay.Walk(ctx, root, fileName) - if err != nil { - t.Fatalf("overlay.Walk(%q) failed: %v", fileName, err) - } - second, err := overlay.Walk(ctx, root, fileName) - if err != nil { - t.Fatalf("overlay.Walk(%q) failed: %v", fileName, err) - } - - if tc.wantSame && first != second { - t.Errorf("dirent lookup got different dirents, wanted same\nfirst=%+v\nsecond=%+v", first, second) - } else if !tc.wantSame && first == second { - t.Errorf("dirent lookup got the same dirent, wanted different: %+v", first) - } - }) - } -} - -func TestCacheFlush(t *testing.T) { - ctx := contexttest.Context(t) - - // Upper and lower each have a file. - upperFileName := "file-from-upper" - lowerFileName := "file-from-lower" - upper := newTestRamfsDir(ctx, []dirContent{{name: upperFileName}}, nil) - lower := newTestRamfsDir(ctx, []dirContent{{name: lowerFileName}}, nil) - - overlay := fs.NewTestOverlayDir(ctx, upper, lower, true /* revalidate */) - - mns, err := fs.NewMountNamespace(ctx, overlay) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - root := mns.Root() - defer root.DecRef() - - ctx = &rootContext{ - Context: ctx, - root: root, - } - - for _, fileName := range []string{upperFileName, lowerFileName} { - // Walk to the file. - maxTraversals := uint(0) - dirent, err := mns.FindInode(ctx, root, nil, fileName, &maxTraversals) - if err != nil { - t.Fatalf("FindInode(%q) failed: %v", fileName, err) - } - - // Get a file from the dirent. - file, err := dirent.Inode.GetFile(ctx, dirent, fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("GetFile() failed: %v", err) - } - - // The dirent should have 3 refs, one from us, one from the - // file, and one from the dirent cache. - // dirent cache. - if got, want := dirent.ReadRefs(), 3; int(got) != want { - t.Errorf("dirent.ReadRefs() got %d want %d", got, want) - } - - // Drop the file reference. - file.DecRef() - - // Dirent should have 2 refs left. - if got, want := dirent.ReadRefs(), 2; int(got) != want { - t.Errorf("dirent.ReadRefs() got %d want %d", got, want) - } - - // Flush the dirent cache. - mns.FlushMountSourceRefs() - - // Dirent should have 1 ref left from the dirent cache. - if got, want := dirent.ReadRefs(), 1; int(got) != want { - t.Errorf("dirent.ReadRefs() got %d want %d", got, want) - } - - // Drop our ref. - dirent.DecRef() - - // We should be back to zero refs. - if got, want := dirent.ReadRefs(), 0; int(got) != want { - t.Errorf("dirent.ReadRefs() got %d want %d", got, want) - } - } - -} - -type dir struct { - fs.InodeOperations - - // List of negative child names. - negative []string - - // ReaddirCalled records whether Readdir was called on a file - // corresponding to this inode. - ReaddirCalled bool -} - -// GetXattr implements InodeOperations.GetXattr. -func (d *dir) GetXattr(_ context.Context, _ *fs.Inode, name string, _ uint64) (string, error) { - for _, n := range d.negative { - if name == fs.XattrOverlayWhiteout(n) { - return "y", nil - } - } - return "", syserror.ENOATTR -} - -// GetFile implements InodeOperations.GetFile. -func (d *dir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { - file, err := d.InodeOperations.GetFile(ctx, dirent, flags) - if err != nil { - return nil, err - } - defer file.DecRef() - // Wrap the file's FileOperations in a dirFile. - fops := &dirFile{ - FileOperations: file.FileOperations, - inode: d, - } - return fs.NewFile(ctx, dirent, flags, fops), nil -} - -type dirContent struct { - name string - dir bool -} - -type dirFile struct { - fs.FileOperations - inode *dir -} - -type inode struct { - fsutil.InodeGenericChecker `state:"nosave"` - fsutil.InodeNoExtendedAttributes `state:"nosave"` - fsutil.InodeNoopRelease `state:"nosave"` - fsutil.InodeNoopWriteOut `state:"nosave"` - fsutil.InodeNotAllocatable `state:"nosave"` - fsutil.InodeNotDirectory `state:"nosave"` - fsutil.InodeNotMappable `state:"nosave"` - fsutil.InodeNotSocket `state:"nosave"` - fsutil.InodeNotSymlink `state:"nosave"` - fsutil.InodeNotTruncatable `state:"nosave"` - fsutil.InodeNotVirtual `state:"nosave"` - - fsutil.InodeSimpleAttributes - fsutil.InodeStaticFileGetter -} - -// Readdir implements fs.FileOperations.Readdir. It sets the ReaddirCalled -// field on the inode. -func (f *dirFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { - f.inode.ReaddirCalled = true - return f.FileOperations.Readdir(ctx, file, ser) -} - -func newTestRamfsInode(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - inode := fs.NewInode(ctx, &inode{ - InodeStaticFileGetter: fsutil.InodeStaticFileGetter{ - Contents: []byte("foobar"), - }, - }, msrc, fs.StableAttr{Type: fs.RegularFile}) - return inode -} - -func newTestRamfsDir(ctx context.Context, contains []dirContent, negative []string) *fs.Inode { - msrc := fs.NewPseudoMountSource(ctx) - contents := make(map[string]*fs.Inode) - for _, c := range contains { - if c.dir { - contents[c.name] = newTestRamfsDir(ctx, nil, nil) - } else { - contents[c.name] = newTestRamfsInode(ctx, msrc) - } - } - dops := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermissions{ - User: fs.PermMask{Read: true, Execute: true}, - }) - return fs.NewInode(ctx, &dir{ - InodeOperations: dops, - negative: negative, - }, msrc, fs.StableAttr{Type: fs.Directory}) -} diff --git a/pkg/sentry/fs/lock/BUILD b/pkg/sentry/fs/lock/BUILD deleted file mode 100644 index ae3331737..000000000 --- a/pkg/sentry/fs/lock/BUILD +++ /dev/null @@ -1,58 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "lock_range", - out = "lock_range.go", - package = "lock", - prefix = "Lock", - template = "//pkg/segment:generic_range", - types = { - "T": "uint64", - }, -) - -go_template_instance( - name = "lock_set", - out = "lock_set.go", - consts = { - "minDegree": "3", - }, - package = "lock", - prefix = "Lock", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "LockRange", - "Value": "Lock", - "Functions": "lockSetFunctions", - }, -) - -go_library( - name = "lock", - srcs = [ - "lock.go", - "lock_range.go", - "lock_set.go", - "lock_set_functions.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/log", - "//pkg/sync", - "//pkg/waiter", - ], -) - -go_test( - name = "lock_test", - size = "small", - srcs = [ - "lock_range_test.go", - "lock_test.go", - ], - library = ":lock", -) diff --git a/pkg/sentry/fs/lock/lock_range.go b/pkg/sentry/fs/lock/lock_range.go new file mode 100755 index 000000000..7a6f77640 --- /dev/null +++ b/pkg/sentry/fs/lock/lock_range.go @@ -0,0 +1,62 @@ +package lock + +// A Range represents a contiguous range of T. +// +// +stateify savable +type LockRange struct { + // Start is the inclusive start of the range. + Start uint64 + + // End is the exclusive end of the range. + End uint64 +} + +// WellFormed returns true if r.Start <= r.End. All other methods on a Range +// require that the Range is well-formed. +func (r LockRange) WellFormed() bool { + return r.Start <= r.End +} + +// Length returns the length of the range. +func (r LockRange) Length() uint64 { + return r.End - r.Start +} + +// Contains returns true if r contains x. +func (r LockRange) Contains(x uint64) bool { + return r.Start <= x && x < r.End +} + +// Overlaps returns true if r and r2 overlap. +func (r LockRange) Overlaps(r2 LockRange) bool { + return r.Start < r2.End && r2.Start < r.End +} + +// IsSupersetOf returns true if r is a superset of r2; that is, the range r2 is +// contained within r. +func (r LockRange) IsSupersetOf(r2 LockRange) bool { + return r.Start <= r2.Start && r.End >= r2.End +} + +// Intersect returns a range consisting of the intersection between r and r2. +// If r and r2 do not overlap, Intersect returns a range with unspecified +// bounds, but for which Length() == 0. +func (r LockRange) Intersect(r2 LockRange) LockRange { + if r.Start < r2.Start { + r.Start = r2.Start + } + if r.End > r2.End { + r.End = r2.End + } + if r.End < r.Start { + r.End = r.Start + } + return r +} + +// CanSplitAt returns true if it is legal to split a segment spanning the range +// r at x; that is, splitting at x would produce two ranges, both of which have +// non-zero length. +func (r LockRange) CanSplitAt(x uint64) bool { + return r.Contains(x) && r.Start < x +} diff --git a/pkg/sentry/fs/lock/lock_range_test.go b/pkg/sentry/fs/lock/lock_range_test.go deleted file mode 100644 index 6221199d1..000000000 --- a/pkg/sentry/fs/lock/lock_range_test.go +++ /dev/null @@ -1,136 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package lock - -import ( - "syscall" - "testing" -) - -func TestComputeRange(t *testing.T) { - tests := []struct { - // Description of test. - name string - - // Requested start of the lock range. - start int64 - - // Requested length of the lock range, - // can be negative :( - length int64 - - // Pre-computed file offset based on whence. - // Will be added to start. - offset int64 - - // Expected error. - err error - - // If error is nil, the expected LockRange. - LockRange - }{ - { - name: "offset, start, and length all zero", - LockRange: LockRange{Start: 0, End: LockEOF}, - }, - { - name: "zero offset, zero start, positive length", - start: 0, - length: 4096, - offset: 0, - LockRange: LockRange{Start: 0, End: 4096}, - }, - { - name: "zero offset, negative start", - start: -4096, - offset: 0, - err: syscall.EINVAL, - }, - { - name: "large offset, negative start, positive length", - start: -2048, - length: 2048, - offset: 4096, - LockRange: LockRange{Start: 2048, End: 4096}, - }, - { - name: "large offset, negative start, zero length", - start: -2048, - length: 0, - offset: 4096, - LockRange: LockRange{Start: 2048, End: LockEOF}, - }, - { - name: "zero offset, zero start, negative length", - start: 0, - length: -4096, - offset: 0, - err: syscall.EINVAL, - }, - { - name: "large offset, zero start, negative length", - start: 0, - length: -4096, - offset: 4096, - LockRange: LockRange{Start: 0, End: 4096}, - }, - { - name: "offset, start, and length equal, length is negative", - start: 1024, - length: -1024, - offset: 1024, - LockRange: LockRange{Start: 1024, End: 2048}, - }, - { - name: "offset, start, and length equal, start is negative", - start: -1024, - length: 1024, - offset: 1024, - LockRange: LockRange{Start: 0, End: 1024}, - }, - { - name: "offset, start, and length equal, offset is negative", - start: 1024, - length: 1024, - offset: -1024, - LockRange: LockRange{Start: 0, End: 1024}, - }, - { - name: "offset, start, and length equal, all negative", - start: -1024, - length: -1024, - offset: -1024, - err: syscall.EINVAL, - }, - { - name: "offset, start, and length equal, all positive", - start: 1024, - length: 1024, - offset: 1024, - LockRange: LockRange{Start: 2048, End: 3072}, - }, - } - - for _, test := range tests { - rng, err := ComputeRange(test.start, test.length, test.offset) - if err != test.err { - t.Errorf("%s: lockRange(%d, %d, %d) got error %v, want %v", test.name, test.start, test.length, test.offset, err, test.err) - continue - } - if err == nil && rng != test.LockRange { - t.Errorf("%s: lockRange(%d, %d, %d) got LockRange %v, want %v", test.name, test.start, test.length, test.offset, rng, test.LockRange) - } - } -} diff --git a/pkg/sentry/fs/lock/lock_set.go b/pkg/sentry/fs/lock/lock_set.go new file mode 100755 index 000000000..2343ca0b4 --- /dev/null +++ b/pkg/sentry/fs/lock/lock_set.go @@ -0,0 +1,1270 @@ +package lock + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + LockminDegree = 3 + + LockmaxDegree = 2 * LockminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type LockSet struct { + root Locknode `state:".(*LockSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *LockSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *LockSet) IsEmptyRange(r LockRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *LockSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *LockSet) SpanRange(r LockRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *LockSet) FirstSegment() LockIterator { + if s.root.nrSegments == 0 { + return LockIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *LockSet) LastSegment() LockIterator { + if s.root.nrSegments == 0 { + return LockIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *LockSet) FirstGap() LockGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return LockGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *LockSet) LastGap() LockGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return LockGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *LockSet) Find(key uint64) (LockIterator, LockGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return LockIterator{n, i}, LockGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return LockIterator{}, LockGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *LockSet) FindSegment(key uint64) LockIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *LockSet) LowerBoundSegment(min uint64) LockIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *LockSet) UpperBoundSegment(max uint64) LockIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *LockSet) FindGap(key uint64) LockGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *LockSet) LowerBoundGap(min uint64) LockGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *LockSet) UpperBoundGap(max uint64) LockGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *LockSet) Add(r LockRange, val Lock) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *LockSet) AddWithoutMerging(r LockRange, val Lock) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *LockSet) Insert(gap LockGapIterator, r LockRange, val Lock) LockIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (lockSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (lockSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (lockSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *LockSet) InsertWithoutMerging(gap LockGapIterator, r LockRange, val Lock) LockIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *LockSet) InsertWithoutMergingUnchecked(gap LockGapIterator, r LockRange, val Lock) LockIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return LockIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *LockSet) Remove(seg LockIterator) LockGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + lockSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(LockGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *LockSet) RemoveAll() { + s.root = Locknode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *LockSet) RemoveRange(r LockRange) LockGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *LockSet) Merge(first, second LockIterator) LockIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *LockSet) MergeUnchecked(first, second LockIterator) LockIterator { + if first.End() == second.Start() { + if mval, ok := (lockSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return LockIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *LockSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *LockSet) MergeRange(r LockRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *LockSet) MergeAdjacent(r LockRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *LockSet) Split(seg LockIterator, split uint64) (LockIterator, LockIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *LockSet) SplitUnchecked(seg LockIterator, split uint64) (LockIterator, LockIterator) { + val1, val2 := (lockSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), LockRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *LockSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *LockSet) Isolate(seg LockIterator, r LockRange) LockIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *LockSet) ApplyContiguous(r LockRange, fn func(seg LockIterator)) LockGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return LockGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return LockGapIterator{} + } + } +} + +// +stateify savable +type Locknode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *Locknode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [LockmaxDegree - 1]LockRange + values [LockmaxDegree - 1]Lock + children [LockmaxDegree]*Locknode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Locknode) firstSegment() LockIterator { + for n.hasChildren { + n = n.children[0] + } + return LockIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Locknode) lastSegment() LockIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return LockIterator{n, n.nrSegments - 1} +} + +func (n *Locknode) prevSibling() *Locknode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *Locknode) nextSibling() *Locknode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *Locknode) rebalanceBeforeInsert(gap LockGapIterator) LockGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < LockmaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &Locknode{ + nrSegments: LockminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &Locknode{ + nrSegments: LockminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:LockminDegree-1], n.keys[:LockminDegree-1]) + copy(left.values[:LockminDegree-1], n.values[:LockminDegree-1]) + copy(right.keys[:LockminDegree-1], n.keys[LockminDegree:]) + copy(right.values[:LockminDegree-1], n.values[LockminDegree:]) + n.keys[0], n.values[0] = n.keys[LockminDegree-1], n.values[LockminDegree-1] + LockzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:LockminDegree], n.children[:LockminDegree]) + copy(right.children[:LockminDegree], n.children[LockminDegree:]) + LockzeroNodeSlice(n.children[2:]) + for i := 0; i < LockminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < LockminDegree { + return LockGapIterator{left, gap.index} + } + return LockGapIterator{right, gap.index - LockminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[LockminDegree-1], n.values[LockminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &Locknode{ + nrSegments: LockminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:LockminDegree-1], n.keys[LockminDegree:]) + copy(sibling.values[:LockminDegree-1], n.values[LockminDegree:]) + LockzeroValueSlice(n.values[LockminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:LockminDegree], n.children[LockminDegree:]) + LockzeroNodeSlice(n.children[LockminDegree:]) + for i := 0; i < LockminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = LockminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < LockminDegree { + return gap + } + return LockGapIterator{sibling, gap.index - LockminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *Locknode) rebalanceAfterRemove(gap LockGapIterator) LockGapIterator { + for { + if n.nrSegments >= LockminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= LockminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + lockSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return LockGapIterator{n, 0} + } + if gap.node == n { + return LockGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= LockminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + lockSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return LockGapIterator{n, n.nrSegments} + } + return LockGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return LockGapIterator{p, gap.index} + } + if gap.node == right { + return LockGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *Locknode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = LockGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + lockSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type LockIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *Locknode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg LockIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg LockIterator) Range() LockRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg LockIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg LockIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg LockIterator) SetRangeUnchecked(r LockRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg LockIterator) SetRange(r LockRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg LockIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg LockIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg LockIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg LockIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg LockIterator) Value() Lock { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg LockIterator) ValuePtr() *Lock { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg LockIterator) SetValue(val Lock) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg LockIterator) PrevSegment() LockIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return LockIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return LockIterator{} + } + return LocksegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg LockIterator) NextSegment() LockIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return LockIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return LockIterator{} + } + return LocksegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg LockIterator) PrevGap() LockGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return LockGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg LockIterator) NextGap() LockGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return LockGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg LockIterator) PrevNonEmpty() (LockIterator, LockGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return LockIterator{}, gap + } + return gap.PrevSegment(), LockGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg LockIterator) NextNonEmpty() (LockIterator, LockGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return LockIterator{}, gap + } + return gap.NextSegment(), LockGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type LockGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *Locknode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap LockGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap LockGapIterator) Range() LockRange { + return LockRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap LockGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return lockSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap LockGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return lockSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap LockGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap LockGapIterator) PrevSegment() LockIterator { + return LocksegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap LockGapIterator) NextSegment() LockIterator { + return LocksegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap LockGapIterator) PrevGap() LockGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return LockGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap LockGapIterator) NextGap() LockGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return LockGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func LocksegmentBeforePosition(n *Locknode, i int) LockIterator { + for i == 0 { + if n.parent == nil { + return LockIterator{} + } + n, i = n.parent, n.parentIndex + } + return LockIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func LocksegmentAfterPosition(n *Locknode, i int) LockIterator { + for i == n.nrSegments { + if n.parent == nil { + return LockIterator{} + } + n, i = n.parent, n.parentIndex + } + return LockIterator{n, i} +} + +func LockzeroValueSlice(slice []Lock) { + + for i := range slice { + lockSetFunctions{}.ClearValue(&slice[i]) + } +} + +func LockzeroNodeSlice(slice []*Locknode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *LockSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *Locknode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *Locknode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type LockSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []Lock +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *LockSet) ExportSortedSlices() *LockSegmentDataSlices { + var sds LockSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *LockSet) ImportSortedSlices(sds *LockSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := LockRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *LockSet) saveRoot() *LockSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *LockSet) loadRoot(sds *LockSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/fs/lock/lock_state_autogen.go b/pkg/sentry/fs/lock/lock_state_autogen.go new file mode 100755 index 000000000..aabf3d570 --- /dev/null +++ b/pkg/sentry/fs/lock/lock_state_autogen.go @@ -0,0 +1,108 @@ +// automatically generated by stateify. + +package lock + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Lock) beforeSave() {} +func (x *Lock) save(m state.Map) { + x.beforeSave() + m.Save("Readers", &x.Readers) + m.Save("HasWriter", &x.HasWriter) + m.Save("Writer", &x.Writer) +} + +func (x *Lock) afterLoad() {} +func (x *Lock) load(m state.Map) { + m.Load("Readers", &x.Readers) + m.Load("HasWriter", &x.HasWriter) + m.Load("Writer", &x.Writer) +} + +func (x *Locks) beforeSave() {} +func (x *Locks) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.blockedQueue) { + m.Failf("blockedQueue is %v, expected zero", x.blockedQueue) + } + m.Save("locks", &x.locks) +} + +func (x *Locks) afterLoad() {} +func (x *Locks) load(m state.Map) { + m.Load("locks", &x.locks) +} + +func (x *LockRange) beforeSave() {} +func (x *LockRange) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) +} + +func (x *LockRange) afterLoad() {} +func (x *LockRange) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) +} + +func (x *LockSet) beforeSave() {} +func (x *LockSet) save(m state.Map) { + x.beforeSave() + var root *LockSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *LockSet) afterLoad() {} +func (x *LockSet) load(m state.Map) { + m.LoadValue("root", new(*LockSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*LockSegmentDataSlices)) }) +} + +func (x *Locknode) beforeSave() {} +func (x *Locknode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *Locknode) afterLoad() {} +func (x *Locknode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *LockSegmentDataSlices) beforeSave() {} +func (x *LockSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *LockSegmentDataSlices) afterLoad() {} +func (x *LockSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func init() { + state.Register("pkg/sentry/fs/lock.Lock", (*Lock)(nil), state.Fns{Save: (*Lock).save, Load: (*Lock).load}) + state.Register("pkg/sentry/fs/lock.Locks", (*Locks)(nil), state.Fns{Save: (*Locks).save, Load: (*Locks).load}) + state.Register("pkg/sentry/fs/lock.LockRange", (*LockRange)(nil), state.Fns{Save: (*LockRange).save, Load: (*LockRange).load}) + state.Register("pkg/sentry/fs/lock.LockSet", (*LockSet)(nil), state.Fns{Save: (*LockSet).save, Load: (*LockSet).load}) + state.Register("pkg/sentry/fs/lock.Locknode", (*Locknode)(nil), state.Fns{Save: (*Locknode).save, Load: (*Locknode).load}) + state.Register("pkg/sentry/fs/lock.LockSegmentDataSlices", (*LockSegmentDataSlices)(nil), state.Fns{Save: (*LockSegmentDataSlices).save, Load: (*LockSegmentDataSlices).load}) +} diff --git a/pkg/sentry/fs/lock/lock_test.go b/pkg/sentry/fs/lock/lock_test.go deleted file mode 100644 index ba002aeb7..000000000 --- a/pkg/sentry/fs/lock/lock_test.go +++ /dev/null @@ -1,1059 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package lock - -import ( - "reflect" - "testing" -) - -type entry struct { - Lock - LockRange -} - -func equals(e0, e1 []entry) bool { - if len(e0) != len(e1) { - return false - } - for i := range e0 { - for k := range e0[i].Lock.Readers { - if !e1[i].Lock.Readers[k] { - return false - } - } - for k := range e1[i].Lock.Readers { - if !e0[i].Lock.Readers[k] { - return false - } - } - if !reflect.DeepEqual(e0[i].LockRange, e1[i].LockRange) { - return false - } - if e0[i].Lock.HasWriter != e1[i].Lock.HasWriter { - return false - } - if e0[i].Lock.Writer != e1[i].Lock.Writer { - return false - } - } - return true -} - -// fill a LockSet with consecutive region locks. Will panic if -// LockRanges are not consecutive. -func fill(entries []entry) LockSet { - l := LockSet{} - for _, e := range entries { - gap := l.FindGap(e.LockRange.Start) - if !gap.Ok() { - panic("cannot insert into existing segment") - } - l.Insert(gap, e.LockRange, e.Lock) - } - return l -} - -func TestCanLockEmpty(t *testing.T) { - l := LockSet{} - - // Expect to be able to take any locks given that the set is empty. - eof := l.FirstGap().End() - r := LockRange{0, eof} - if !l.canLock(1, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 1) - } - if !l.canLock(2, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 2) - } - if !l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 1) - } - if !l.canLock(2, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 2) - } -} - -func TestCanLock(t *testing.T) { - // + -------------- + ---------- + -------------- + --------- + - // | Readers 1 & 2 | Readers 1 | Readers 1 & 3 | Writer 1 | - // + ------------- + ---------- + -------------- + --------- + - // 0 1024 2048 3072 4096 - l := fill([]entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true}}, - LockRange: LockRange{1024, 2048}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 3: true}}, - LockRange: LockRange{2048, 3072}, - }, - { - Lock: Lock{HasWriter: true, Writer: 1}, - LockRange: LockRange{3072, 4096}, - }, - }) - - // Now that we have a mildly interesting layout, try some checks on different - // ranges, uids, and lock types. - // - // Expect to be able to extend the read lock, despite the writer lock, because - // the writer has the same uid as the requested read lock. - r := LockRange{0, 8192} - if !l.canLock(1, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 1) - } - // Expect to *not* be able to extend the read lock since there is an overlapping - // writer region locked by someone other than the uid. - if l.canLock(2, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got true, want false", ReadLock, r, 2) - } - // Expect to be able to extend the read lock if there are only other readers in - // the way. - r = LockRange{64, 3072} - if !l.canLock(2, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 2) - } - // Expect to be able to set a read lock beyond the range of any existing locks. - r = LockRange{4096, 10240} - if !l.canLock(2, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 2) - } - - // Expect to not be able to take a write lock with other readers in the way. - r = LockRange{0, 8192} - if l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got true, want false", WriteLock, r, 1) - } - // Expect to be able to extend the write lock for the same uid. - r = LockRange{3072, 8192} - if !l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 1) - } - // Expect to not be able to overlap a write lock for two different uids. - if l.canLock(2, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got true, want false", WriteLock, r, 2) - } - // Expect to be able to set a write lock that is beyond the range of any - // existing locks. - r = LockRange{8192, 10240} - if !l.canLock(2, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 2) - } - // Expect to be able to upgrade a read lock (any portion of it). - r = LockRange{1024, 2048} - if !l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 1) - } - r = LockRange{1080, 2000} - if !l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 1) - } -} - -func TestSetLock(t *testing.T) { - tests := []struct { - // description of test. - name string - - // LockSet entries to pre-fill. - before []entry - - // Description of region to lock: - // - // start is the file offset of the lock. - start uint64 - // end is the end file offset of the lock. - end uint64 - // uid of lock attempter. - uid UniqueID - // lock type requested. - lockType LockType - - // success is true if taking the above - // lock should succeed. - success bool - - // Expected layout of the set after locking - // if success is true. - after []entry - }{ - { - name: "set zero length ReadLock on empty set", - start: 0, - end: 0, - uid: 0, - lockType: ReadLock, - success: true, - }, - { - name: "set zero length WriteLock on empty set", - start: 0, - end: 0, - uid: 0, - lockType: WriteLock, - success: true, - }, - { - name: "set ReadLock on empty set", - start: 0, - end: LockEOF, - uid: 0, - lockType: ReadLock, - success: true, - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - }, - { - name: "set WriteLock on empty set", - start: 0, - end: LockEOF, - uid: 0, - lockType: WriteLock, - success: true, - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - }, - { - name: "set ReadLock on WriteLock same uid", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 0, - lockType: ReadLock, - success: true, - // + ----------- + --------------------------- + - // | Readers 0 | Writer 0 | - // + ----------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 4096}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "set WriteLock on ReadLock same uid", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 0, - lockType: WriteLock, - success: true, - // + ----------- + --------------------------- + - // | Writer 0 | Readers 0 | - // + ----------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "set ReadLock on WriteLock different uid", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 1, - lockType: ReadLock, - success: false, - }, - { - name: "set WriteLock on ReadLock different uid", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 1, - lockType: WriteLock, - success: false, - }, - { - name: "split ReadLock for overlapping lock at start 0", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 1, - lockType: ReadLock, - success: true, - // + -------------- + --------------------------- + - // | Readers 0 & 1 | Readers 0 | - // + -------------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "split ReadLock for overlapping lock at non-zero start", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 4096, - end: 8192, - uid: 1, - lockType: ReadLock, - success: true, - // + ---------- + -------------- + ----------- + - // | Readers 0 | Readers 0 & 1 | Readers 0 | - // + ---------- + -------------- + ----------- + - // 0 4096 8192 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{4096, 8192}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{8192, LockEOF}, - }, - }, - }, - { - name: "fill front gap with ReadLock", - // + --------- + ---------------------------- + - // | gap | Readers 0 | - // + --------- + ---------------------------- + - // 0 1024 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, LockEOF}, - }, - }, - start: 0, - end: 8192, - uid: 0, - lockType: ReadLock, - success: true, - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - }, - { - name: "fill end gap with ReadLock", - // + ---------------------------- + - // | Readers 0 | - // + ---------------------------- + - // 0 4096 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 4096}, - }, - }, - start: 1024, - end: LockEOF, - uid: 0, - lockType: ReadLock, - success: true, - // Note that this is not merged after lock does a Split. This is - // fine because the two locks will still *behave* as one. In other - // words we can fragment any lock all we want and semantically it - // makes no difference. - // - // + ----------- + --------------------------- + - // | Readers 0 | Readers 0 | - // + ----------- + --------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, LockEOF}, - }, - }, - }, - { - name: "fill gap with ReadLock and split", - // + --------- + ---------------------------- + - // | gap | Readers 0 | - // + --------- + ---------------------------- + - // 0 1024 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 1, - lockType: ReadLock, - success: true, - // + --------- + ------------- + ------------- + - // | Reader 1 | Readers 0 & 1 | Reader 0 | - // + ----------+ ------------- + ------------- + - // 0 1024 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{1024, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "upgrade ReadLock to WriteLock for single uid fill gap", - // + ------------- + --------- + --- + ------------- + - // | Readers 0 & 1 | Readers 0 | gap | Readers 0 & 2 | - // + ------------- + --------- + --- + ------------- + - // 0 1024 2048 4096 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, 2048}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 1024, - end: 4096, - uid: 0, - lockType: WriteLock, - success: true, - // + ------------- + -------- + ------------- + - // | Readers 0 & 1 | Writer 0 | Readers 0 & 2 | - // + ------------- + -------- + ------------- + - // 0 1024 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{1024, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "upgrade ReadLock to WriteLock for single uid keep gap", - // + ------------- + --------- + --- + ------------- + - // | Readers 0 & 1 | Readers 0 | gap | Readers 0 & 2 | - // + ------------- + --------- + --- + ------------- + - // 0 1024 2048 4096 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, 2048}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 1024, - end: 3072, - uid: 0, - lockType: WriteLock, - success: true, - // + ------------- + -------- + --- + ------------- + - // | Readers 0 & 1 | Writer 0 | gap | Readers 0 & 2 | - // + ------------- + -------- + --- + ------------- + - // 0 1024 3072 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{1024, 3072}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "fail to upgrade ReadLock to WriteLock with conflicting Reader", - // + ------------- + --------- + - // | Readers 0 & 1 | Readers 0 | - // + ------------- + --------- + - // 0 1024 2048 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, 2048}, - }, - }, - start: 0, - end: 2048, - uid: 0, - lockType: WriteLock, - success: false, - }, - { - name: "take WriteLock on whole file if all uids are the same", - // + ------------- + --------- + --------- + ---------- + - // | Writer 0 | Readers 0 | Readers 0 | Readers 0 | - // + ------------- + --------- + --------- + ---------- + - // 0 1024 2048 4096 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, 2048}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{2048, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 0, - end: LockEOF, - uid: 0, - lockType: WriteLock, - success: true, - // We do not manually merge locks. Semantically a fragmented lock - // held by the same uid will behave as one lock so it makes no difference. - // - // + ------------- + ---------------------------- + - // | Writer 0 | Writer 0 | - // + ------------- + ---------------------------- + - // 0 1024 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{1024, LockEOF}, - }, - }, - }, - } - - for _, test := range tests { - l := fill(test.before) - - r := LockRange{Start: test.start, End: test.end} - success := l.lock(test.uid, test.lockType, r) - var got []entry - for seg := l.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - got = append(got, entry{ - Lock: seg.Value(), - LockRange: seg.Range(), - }) - } - - if success != test.success { - t.Errorf("%s: setlock(%v, %+v, %d, %d) got success %v, want %v", test.name, test.before, r, test.uid, test.lockType, success, test.success) - continue - } - - if success { - if !equals(got, test.after) { - t.Errorf("%s: got set %+v, want %+v", test.name, got, test.after) - } - } - } -} - -func TestUnlock(t *testing.T) { - tests := []struct { - // description of test. - name string - - // LockSet entries to pre-fill. - before []entry - - // Description of region to unlock: - // - // start is the file start of the lock. - start uint64 - // end is the end file start of the lock. - end uint64 - // uid of lock holder. - uid UniqueID - - // Expected layout of the set after unlocking. - after []entry - }{ - { - name: "unlock zero length on empty set", - start: 0, - end: 0, - uid: 0, - }, - { - name: "unlock on empty set (no-op)", - start: 0, - end: LockEOF, - uid: 0, - }, - { - name: "unlock uid not locked (no-op)", - // + --------------------------- + - // | Readers 1 & 2 | - // + --------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 1024, - end: 4096, - uid: 0, - // + --------------------------- + - // | Readers 1 & 2 | - // + --------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - }, - { - name: "unlock ReadLock over entire file", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: LockEOF, - uid: 0, - }, - { - name: "unlock WriteLock over entire file", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: LockEOF, - uid: 0, - }, - { - name: "unlock partial ReadLock (start)", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 0, - // + ------ + --------------------------- + - // | gap | Readers 0 | - // +------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock partial WriteLock (start)", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 0, - // + ------ + --------------------------- + - // | gap | Writer 0 | - // +------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock partial ReadLock (end)", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 4096, - end: LockEOF, - uid: 0, - // + --------------------------- + - // | Readers 0 | - // +---------------------------- + - // 0 4096 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 4096}, - }, - }, - }, - { - name: "unlock partial WriteLock (end)", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 4096, - end: LockEOF, - uid: 0, - // + --------------------------- + - // | Writer 0 | - // +---------------------------- + - // 0 4096 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 4096}, - }, - }, - }, - { - name: "unlock for single uid", - // + ------------- + --------- + ------------------- + - // | Readers 0 & 1 | Writer 0 | Readers 0 & 1 & 2 | - // + ------------- + --------- + ------------------- + - // 0 1024 4096 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{1024, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 0, - end: LockEOF, - uid: 0, - // + --------- + --- + --------------- + - // | Readers 1 | gap | Readers 1 & 2 | - // + --------- + --- + --------------- + - // 0 1024 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock subsection locked", - // + ------------------------------- + - // | Readers 0 & 1 & 2 | - // + ------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true, 2: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 1024, - end: 4096, - uid: 0, - // + ----------------- + ------------- + ----------------- + - // | Readers 0 & 1 & 2 | Readers 1 & 2 | Readers 0 & 1 & 2 | - // + ----------------- + ------------- + ----------------- + - // 0 1024 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true, 2: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{1024, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock mid-gap to increase gap", - // + --------- + ----- + ------------------- + - // | Writer 0 | gap | Readers 0 & 1 | - // + --------- + ----- + ------------------- + - // 0 1024 4096 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 8, - end: 2048, - uid: 0, - // + --------- + ----- + ------------------- + - // | Writer 0 | gap | Readers 0 & 1 | - // + --------- + ----- + ------------------- + - // 0 8 4096 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 8}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock split region on uid mid-gap", - // + --------- + ----- + ------------------- + - // | Writer 0 | gap | Readers 0 & 1 | - // + --------- + ----- + ------------------- + - // 0 1024 4096 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 2048, - end: 8192, - uid: 0, - // + --------- + ----- + --------- + ------------- + - // | Writer 0 | gap | Readers 1 | Readers 0 & 1 | - // + --------- + ----- + --------- + ------------- + - // 0 1024 4096 8192 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true}}, - LockRange: LockRange{4096, 8192}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{8192, LockEOF}, - }, - }, - }, - } - - for _, test := range tests { - l := fill(test.before) - - r := LockRange{Start: test.start, End: test.end} - l.unlock(test.uid, r) - var got []entry - for seg := l.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - got = append(got, entry{ - Lock: seg.Value(), - LockRange: seg.Range(), - }) - } - if !equals(got, test.after) { - t.Errorf("%s: got set %+v, want %+v", test.name, got, test.after) - } - } -} diff --git a/pkg/sentry/fs/mount_test.go b/pkg/sentry/fs/mount_test.go deleted file mode 100644 index a3d10770b..000000000 --- a/pkg/sentry/fs/mount_test.go +++ /dev/null @@ -1,272 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs - -import ( - "fmt" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/contexttest" -) - -// cacheReallyContains iterates through the dirent cache to determine whether -// it contains the given dirent. -func cacheReallyContains(cache *DirentCache, d *Dirent) bool { - for i := cache.list.Front(); i != nil; i = i.Next() { - if i == d { - return true - } - } - return false -} - -func mountPathsAre(root *Dirent, got []*Mount, want ...string) error { - gotPaths := make(map[string]struct{}, len(got)) - gotStr := make([]string, len(got)) - for i, g := range got { - if groot := g.Root(); groot != nil { - name, _ := groot.FullName(root) - groot.DecRef() - gotStr[i] = name - gotPaths[name] = struct{}{} - } - } - if len(got) != len(want) { - return fmt.Errorf("mount paths are different, got: %q, want: %q", gotStr, want) - } - for _, w := range want { - if _, ok := gotPaths[w]; !ok { - return fmt.Errorf("no mount with path %q found", w) - } - } - return nil -} - -// TestMountSourceOnlyCachedOnce tests that a Dirent that is mounted over only ends -// up in a single Dirent Cache. NOTE(b/63848693): Having a dirent in multiple -// caches causes major consistency issues. -func TestMountSourceOnlyCachedOnce(t *testing.T) { - ctx := contexttest.Context(t) - - rootCache := NewDirentCache(100) - rootInode := NewMockInode(ctx, NewMockMountSource(rootCache), StableAttr{ - Type: Directory, - }) - mm, err := NewMountNamespace(ctx, rootInode) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - rootDirent := mm.Root() - defer rootDirent.DecRef() - - // Get a child of the root which we will mount over. Note that the - // MockInodeOperations causes Walk to always succeed. - child, err := rootDirent.Walk(ctx, rootDirent, "child") - if err != nil { - t.Fatalf("failed to walk to child dirent: %v", err) - } - child.maybeExtendReference() // Cache. - - // Ensure that the root cache contains the child. - if !cacheReallyContains(rootCache, child) { - t.Errorf("wanted rootCache to contain child dirent, but it did not") - } - - // Create a new cache and inode, and mount it over child. - submountCache := NewDirentCache(100) - submountInode := NewMockInode(ctx, NewMockMountSource(submountCache), StableAttr{ - Type: Directory, - }) - if err := mm.Mount(ctx, child, submountInode); err != nil { - t.Fatalf("failed to mount over child: %v", err) - } - - // Walk to the child again. - child2, err := rootDirent.Walk(ctx, rootDirent, "child") - if err != nil { - t.Fatalf("failed to walk to child dirent: %v", err) - } - - // Should have a different Dirent than before. - if child == child2 { - t.Fatalf("expected %v not equal to %v, but they are the same", child, child2) - } - - // Neither of the caches should no contain the child. - if cacheReallyContains(rootCache, child) { - t.Errorf("wanted rootCache not to contain child dirent, but it did") - } - if cacheReallyContains(submountCache, child) { - t.Errorf("wanted submountCache not to contain child dirent, but it did") - } -} - -func TestAllMountsUnder(t *testing.T) { - ctx := contexttest.Context(t) - - rootCache := NewDirentCache(100) - rootInode := NewMockInode(ctx, NewMockMountSource(rootCache), StableAttr{ - Type: Directory, - }) - mm, err := NewMountNamespace(ctx, rootInode) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - rootDirent := mm.Root() - defer rootDirent.DecRef() - - // Add mounts at the following paths: - paths := []string{ - "/foo", - "/foo/bar", - "/foo/bar/baz", - "/foo/qux", - "/waldo", - } - - var maxTraversals uint - for _, p := range paths { - maxTraversals = 0 - d, err := mm.FindLink(ctx, rootDirent, nil, p, &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", p, err) - } - - submountInode := NewMockInode(ctx, NewMockMountSource(nil), StableAttr{ - Type: Directory, - }) - if err := mm.Mount(ctx, d, submountInode); err != nil { - t.Fatalf("could not mount at %q: %v", p, err) - } - d.DecRef() - } - - // mm root should contain all submounts (and does not include the root mount). - rootMnt := mm.FindMount(rootDirent) - submounts := mm.AllMountsUnder(rootMnt) - allPaths := append(paths, "/") - if err := mountPathsAre(rootDirent, submounts, allPaths...); err != nil { - t.Error(err) - } - - // Each mount should have a unique ID. - foundIDs := make(map[uint64]struct{}) - for _, m := range submounts { - if _, ok := foundIDs[m.ID]; ok { - t.Errorf("got multiple mounts with id %d", m.ID) - } - foundIDs[m.ID] = struct{}{} - } - - // Root mount should have no parent. - if p := rootMnt.ParentID; p != invalidMountID { - t.Errorf("root.Parent got %v wanted nil", p) - } - - // Check that "foo" mount has 3 children. - maxTraversals = 0 - d, err := mm.FindLink(ctx, rootDirent, nil, "/foo", &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", "/foo", err) - } - defer d.DecRef() - submounts = mm.AllMountsUnder(mm.FindMount(d)) - if err := mountPathsAre(rootDirent, submounts, "/foo", "/foo/bar", "/foo/qux", "/foo/bar/baz"); err != nil { - t.Error(err) - } - - // "waldo" mount should have no children. - maxTraversals = 0 - waldo, err := mm.FindLink(ctx, rootDirent, nil, "/waldo", &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", "/waldo", err) - } - defer waldo.DecRef() - submounts = mm.AllMountsUnder(mm.FindMount(waldo)) - if err := mountPathsAre(rootDirent, submounts, "/waldo"); err != nil { - t.Error(err) - } -} - -func TestUnmount(t *testing.T) { - ctx := contexttest.Context(t) - - rootCache := NewDirentCache(100) - rootInode := NewMockInode(ctx, NewMockMountSource(rootCache), StableAttr{ - Type: Directory, - }) - mm, err := NewMountNamespace(ctx, rootInode) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - rootDirent := mm.Root() - defer rootDirent.DecRef() - - // Add mounts at the following paths: - paths := []string{ - "/foo", - "/foo/bar", - "/foo/bar/goo", - "/foo/bar/goo/abc", - "/foo/abc", - "/foo/def", - "/waldo", - "/wally", - } - - var maxTraversals uint - for _, p := range paths { - maxTraversals = 0 - d, err := mm.FindLink(ctx, rootDirent, nil, p, &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", p, err) - } - - submountInode := NewMockInode(ctx, NewMockMountSource(nil), StableAttr{ - Type: Directory, - }) - if err := mm.Mount(ctx, d, submountInode); err != nil { - t.Fatalf("could not mount at %q: %v", p, err) - } - d.DecRef() - } - - allPaths := make([]string, len(paths)+1) - allPaths[0] = "/" - copy(allPaths[1:], paths) - - rootMnt := mm.FindMount(rootDirent) - for i := len(paths) - 1; i >= 0; i-- { - maxTraversals = 0 - p := paths[i] - d, err := mm.FindLink(ctx, rootDirent, nil, p, &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", p, err) - } - - if err := mm.Unmount(ctx, d, false); err != nil { - t.Fatalf("could not unmount at %q: %v", p, err) - } - d.DecRef() - - // Remove the path that has been unmounted and the check that the remaining - // mounts are still there. - allPaths = allPaths[:len(allPaths)-1] - submounts := mm.AllMountsUnder(rootMnt) - if err := mountPathsAre(rootDirent, submounts, allPaths...); err != nil { - t.Error(err) - } - } -} diff --git a/pkg/sentry/fs/mounts_test.go b/pkg/sentry/fs/mounts_test.go deleted file mode 100644 index a69b41468..000000000 --- a/pkg/sentry/fs/mounts_test.go +++ /dev/null @@ -1,105 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs_test - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" -) - -// Creates a new MountNamespace with filesystem: -// / (root dir) -// |-foo (dir) -// |-bar (file) -func createMountNamespace(ctx context.Context) (*fs.MountNamespace, error) { - perms := fs.FilePermsFromMode(0777) - m := fs.NewPseudoMountSource(ctx) - - barFile := fsutil.NewSimpleFileInode(ctx, fs.RootOwner, perms, 0) - fooDir := ramfs.NewDir(ctx, map[string]*fs.Inode{ - "bar": fs.NewInode(ctx, barFile, m, fs.StableAttr{Type: fs.RegularFile}), - }, fs.RootOwner, perms) - rootDir := ramfs.NewDir(ctx, map[string]*fs.Inode{ - "foo": fs.NewInode(ctx, fooDir, m, fs.StableAttr{Type: fs.Directory}), - }, fs.RootOwner, perms) - - return fs.NewMountNamespace(ctx, fs.NewInode(ctx, rootDir, m, fs.StableAttr{Type: fs.Directory})) -} - -func TestFindLink(t *testing.T) { - ctx := contexttest.Context(t) - mm, err := createMountNamespace(ctx) - if err != nil { - t.Fatalf("createMountNamespace failed: %v", err) - } - - root := mm.Root() - defer root.DecRef() - foo, err := root.Walk(ctx, root, "foo") - if err != nil { - t.Fatalf("Error walking to foo: %v", err) - } - - // Positive cases. - for _, tc := range []struct { - findPath string - wd *fs.Dirent - wantPath string - }{ - {".", root, "/"}, - {".", foo, "/foo"}, - {"..", foo, "/"}, - {"../../..", foo, "/"}, - {"///foo", foo, "/foo"}, - {"/foo", foo, "/foo"}, - {"/foo/bar", foo, "/foo/bar"}, - {"/foo/.///./bar", foo, "/foo/bar"}, - {"/foo///bar", foo, "/foo/bar"}, - {"/foo/../foo/bar", foo, "/foo/bar"}, - {"foo/bar", root, "/foo/bar"}, - {"foo////bar", root, "/foo/bar"}, - {"bar", foo, "/foo/bar"}, - } { - wdPath, _ := tc.wd.FullName(root) - maxTraversals := uint(0) - if d, err := mm.FindLink(ctx, root, tc.wd, tc.findPath, &maxTraversals); err != nil { - t.Errorf("FindLink(%q, wd=%q) failed: %v", tc.findPath, wdPath, err) - } else if got, _ := d.FullName(root); got != tc.wantPath { - t.Errorf("FindLink(%q, wd=%q) got dirent %q, want %q", tc.findPath, wdPath, got, tc.wantPath) - } - } - - // Negative cases. - for _, tc := range []struct { - findPath string - wd *fs.Dirent - }{ - {"bar", root}, - {"/bar", root}, - {"/foo/../../bar", root}, - {"foo", foo}, - } { - wdPath, _ := tc.wd.FullName(root) - maxTraversals := uint(0) - if _, err := mm.FindLink(ctx, root, tc.wd, tc.findPath, &maxTraversals); err == nil { - t.Errorf("FindLink(%q, wd=%q) did not return error", tc.findPath, wdPath) - } - } -} diff --git a/pkg/sentry/fs/path_test.go b/pkg/sentry/fs/path_test.go deleted file mode 100644 index e6f57ebba..000000000 --- a/pkg/sentry/fs/path_test.go +++ /dev/null @@ -1,289 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs - -import ( - "testing" -) - -// TestSplitLast tests variants of path splitting. -func TestSplitLast(t *testing.T) { - cases := []struct { - path string - dir string - file string - }{ - {path: "/", dir: "/", file: "."}, - {path: "/.", dir: "/", file: "."}, - {path: "/./", dir: "/", file: "."}, - {path: "/./.", dir: "/.", file: "."}, - {path: "/././", dir: "/.", file: "."}, - {path: "/./..", dir: "/.", file: ".."}, - {path: "/./../", dir: "/.", file: ".."}, - {path: "/..", dir: "/", file: ".."}, - {path: "/../", dir: "/", file: ".."}, - {path: "/../.", dir: "/..", file: "."}, - {path: "/.././", dir: "/..", file: "."}, - {path: "/../..", dir: "/..", file: ".."}, - {path: "/../../", dir: "/..", file: ".."}, - - {path: "", dir: ".", file: "."}, - {path: ".", dir: ".", file: "."}, - {path: "./", dir: ".", file: "."}, - {path: "./.", dir: ".", file: "."}, - {path: "././", dir: ".", file: "."}, - {path: "./..", dir: ".", file: ".."}, - {path: "./../", dir: ".", file: ".."}, - {path: "..", dir: ".", file: ".."}, - {path: "../", dir: ".", file: ".."}, - {path: "../.", dir: "..", file: "."}, - {path: ".././", dir: "..", file: "."}, - {path: "../..", dir: "..", file: ".."}, - {path: "../../", dir: "..", file: ".."}, - - {path: "/foo", dir: "/", file: "foo"}, - {path: "/foo/", dir: "/", file: "foo"}, - {path: "/foo/.", dir: "/foo", file: "."}, - {path: "/foo/./", dir: "/foo", file: "."}, - {path: "/foo/./.", dir: "/foo/.", file: "."}, - {path: "/foo/./..", dir: "/foo/.", file: ".."}, - {path: "/foo/..", dir: "/foo", file: ".."}, - {path: "/foo/../", dir: "/foo", file: ".."}, - {path: "/foo/../.", dir: "/foo/..", file: "."}, - {path: "/foo/../..", dir: "/foo/..", file: ".."}, - - {path: "/foo/bar", dir: "/foo", file: "bar"}, - {path: "/foo/bar/", dir: "/foo", file: "bar"}, - {path: "/foo/bar/.", dir: "/foo/bar", file: "."}, - {path: "/foo/bar/./", dir: "/foo/bar", file: "."}, - {path: "/foo/bar/./.", dir: "/foo/bar/.", file: "."}, - {path: "/foo/bar/./..", dir: "/foo/bar/.", file: ".."}, - {path: "/foo/bar/..", dir: "/foo/bar", file: ".."}, - {path: "/foo/bar/../", dir: "/foo/bar", file: ".."}, - {path: "/foo/bar/../.", dir: "/foo/bar/..", file: "."}, - {path: "/foo/bar/../..", dir: "/foo/bar/..", file: ".."}, - - {path: "foo", dir: ".", file: "foo"}, - {path: "foo", dir: ".", file: "foo"}, - {path: "foo/", dir: ".", file: "foo"}, - {path: "foo/.", dir: "foo", file: "."}, - {path: "foo/./", dir: "foo", file: "."}, - {path: "foo/./.", dir: "foo/.", file: "."}, - {path: "foo/./..", dir: "foo/.", file: ".."}, - {path: "foo/..", dir: "foo", file: ".."}, - {path: "foo/../", dir: "foo", file: ".."}, - {path: "foo/../.", dir: "foo/..", file: "."}, - {path: "foo/../..", dir: "foo/..", file: ".."}, - {path: "foo/", dir: ".", file: "foo"}, - {path: "foo/.", dir: "foo", file: "."}, - - {path: "foo/bar", dir: "foo", file: "bar"}, - {path: "foo/bar/", dir: "foo", file: "bar"}, - {path: "foo/bar/.", dir: "foo/bar", file: "."}, - {path: "foo/bar/./", dir: "foo/bar", file: "."}, - {path: "foo/bar/./.", dir: "foo/bar/.", file: "."}, - {path: "foo/bar/./..", dir: "foo/bar/.", file: ".."}, - {path: "foo/bar/..", dir: "foo/bar", file: ".."}, - {path: "foo/bar/../", dir: "foo/bar", file: ".."}, - {path: "foo/bar/../.", dir: "foo/bar/..", file: "."}, - {path: "foo/bar/../..", dir: "foo/bar/..", file: ".."}, - {path: "foo/bar/", dir: "foo", file: "bar"}, - {path: "foo/bar/.", dir: "foo/bar", file: "."}, - } - - for _, c := range cases { - dir, file := SplitLast(c.path) - if dir != c.dir || file != c.file { - t.Errorf("SplitLast(%q) got (%q, %q), expected (%q, %q)", c.path, dir, file, c.dir, c.file) - } - } -} - -// TestSplitFirst tests variants of path splitting. -func TestSplitFirst(t *testing.T) { - cases := []struct { - path string - first string - remainder string - }{ - {path: "/", first: "/", remainder: ""}, - {path: "/.", first: "/", remainder: "."}, - {path: "///.", first: "/", remainder: "//."}, - {path: "/.///", first: "/", remainder: "."}, - {path: "/./.", first: "/", remainder: "./."}, - {path: "/././", first: "/", remainder: "./."}, - {path: "/./..", first: "/", remainder: "./.."}, - {path: "/./../", first: "/", remainder: "./.."}, - {path: "/..", first: "/", remainder: ".."}, - {path: "/../", first: "/", remainder: ".."}, - {path: "/../.", first: "/", remainder: "../."}, - {path: "/.././", first: "/", remainder: "../."}, - {path: "/../..", first: "/", remainder: "../.."}, - {path: "/../../", first: "/", remainder: "../.."}, - - {path: "", first: ".", remainder: ""}, - {path: ".", first: ".", remainder: ""}, - {path: "./", first: ".", remainder: ""}, - {path: ".///", first: ".", remainder: ""}, - {path: "./.", first: ".", remainder: "."}, - {path: "././", first: ".", remainder: "."}, - {path: "./..", first: ".", remainder: ".."}, - {path: "./../", first: ".", remainder: ".."}, - {path: "..", first: "..", remainder: ""}, - {path: "../", first: "..", remainder: ""}, - {path: "../.", first: "..", remainder: "."}, - {path: ".././", first: "..", remainder: "."}, - {path: "../..", first: "..", remainder: ".."}, - {path: "../../", first: "..", remainder: ".."}, - - {path: "/foo", first: "/", remainder: "foo"}, - {path: "/foo/", first: "/", remainder: "foo"}, - {path: "/foo///", first: "/", remainder: "foo"}, - {path: "/foo/.", first: "/", remainder: "foo/."}, - {path: "/foo/./", first: "/", remainder: "foo/."}, - {path: "/foo/./.", first: "/", remainder: "foo/./."}, - {path: "/foo/./..", first: "/", remainder: "foo/./.."}, - {path: "/foo/..", first: "/", remainder: "foo/.."}, - {path: "/foo/../", first: "/", remainder: "foo/.."}, - {path: "/foo/../.", first: "/", remainder: "foo/../."}, - {path: "/foo/../..", first: "/", remainder: "foo/../.."}, - - {path: "/foo/bar", first: "/", remainder: "foo/bar"}, - {path: "///foo/bar", first: "/", remainder: "//foo/bar"}, - {path: "/foo///bar", first: "/", remainder: "foo///bar"}, - {path: "/foo/bar/.", first: "/", remainder: "foo/bar/."}, - {path: "/foo/bar/./", first: "/", remainder: "foo/bar/."}, - {path: "/foo/bar/./.", first: "/", remainder: "foo/bar/./."}, - {path: "/foo/bar/./..", first: "/", remainder: "foo/bar/./.."}, - {path: "/foo/bar/..", first: "/", remainder: "foo/bar/.."}, - {path: "/foo/bar/../", first: "/", remainder: "foo/bar/.."}, - {path: "/foo/bar/../.", first: "/", remainder: "foo/bar/../."}, - {path: "/foo/bar/../..", first: "/", remainder: "foo/bar/../.."}, - - {path: "foo", first: "foo", remainder: ""}, - {path: "foo", first: "foo", remainder: ""}, - {path: "foo/", first: "foo", remainder: ""}, - {path: "foo///", first: "foo", remainder: ""}, - {path: "foo/.", first: "foo", remainder: "."}, - {path: "foo/./", first: "foo", remainder: "."}, - {path: "foo/./.", first: "foo", remainder: "./."}, - {path: "foo/./..", first: "foo", remainder: "./.."}, - {path: "foo/..", first: "foo", remainder: ".."}, - {path: "foo/../", first: "foo", remainder: ".."}, - {path: "foo/../.", first: "foo", remainder: "../."}, - {path: "foo/../..", first: "foo", remainder: "../.."}, - {path: "foo/", first: "foo", remainder: ""}, - {path: "foo/.", first: "foo", remainder: "."}, - - {path: "foo/bar", first: "foo", remainder: "bar"}, - {path: "foo///bar", first: "foo", remainder: "bar"}, - {path: "foo/bar/", first: "foo", remainder: "bar"}, - {path: "foo/bar/.", first: "foo", remainder: "bar/."}, - {path: "foo/bar/./", first: "foo", remainder: "bar/."}, - {path: "foo/bar/./.", first: "foo", remainder: "bar/./."}, - {path: "foo/bar/./..", first: "foo", remainder: "bar/./.."}, - {path: "foo/bar/..", first: "foo", remainder: "bar/.."}, - {path: "foo/bar/../", first: "foo", remainder: "bar/.."}, - {path: "foo/bar/../.", first: "foo", remainder: "bar/../."}, - {path: "foo/bar/../..", first: "foo", remainder: "bar/../.."}, - {path: "foo/bar/", first: "foo", remainder: "bar"}, - {path: "foo/bar/.", first: "foo", remainder: "bar/."}, - } - - for _, c := range cases { - first, remainder := SplitFirst(c.path) - if first != c.first || remainder != c.remainder { - t.Errorf("SplitFirst(%q) got (%q, %q), expected (%q, %q)", c.path, first, remainder, c.first, c.remainder) - } - } -} - -// TestIsSubpath tests the IsSubpath method. -func TestIsSubpath(t *testing.T) { - tcs := []struct { - // Two absolute paths. - pathA string - pathB string - - // Whether pathA is a subpath of pathB. - wantIsSubpath bool - - // Relative path from pathA to pathB. Only checked if - // wantIsSubpath is true. - wantRelpath string - }{ - { - pathA: "/foo/bar/baz", - pathB: "/foo", - wantIsSubpath: true, - wantRelpath: "bar/baz", - }, - { - pathA: "/foo", - pathB: "/foo/bar/baz", - wantIsSubpath: false, - }, - { - pathA: "/foo", - pathB: "/foo", - wantIsSubpath: false, - }, - { - pathA: "/foobar", - pathB: "/foo", - wantIsSubpath: false, - }, - { - pathA: "/foo", - pathB: "/foobar", - wantIsSubpath: false, - }, - { - pathA: "/foo", - pathB: "/foobar", - wantIsSubpath: false, - }, - { - pathA: "/", - pathB: "/foo", - wantIsSubpath: false, - }, - { - pathA: "/foo", - pathB: "/", - wantIsSubpath: true, - wantRelpath: "foo", - }, - { - pathA: "/foo/bar/../bar", - pathB: "/foo", - wantIsSubpath: true, - wantRelpath: "bar", - }, - { - pathA: "/foo/bar", - pathB: "/foo/../foo", - wantIsSubpath: true, - wantRelpath: "bar", - }, - } - - for _, tc := range tcs { - gotRelpath, gotIsSubpath := IsSubpath(tc.pathA, tc.pathB) - if gotRelpath != tc.wantRelpath || gotIsSubpath != tc.wantIsSubpath { - t.Errorf("IsSubpath(%q, %q) got %q %t, want %q %t", tc.pathA, tc.pathB, gotRelpath, gotIsSubpath, tc.wantRelpath, tc.wantIsSubpath) - } - } -} diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD deleted file mode 100644 index 77c2c5c0e..000000000 --- a/pkg/sentry/fs/proc/BUILD +++ /dev/null @@ -1,72 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "proc", - srcs = [ - "cgroup.go", - "cpuinfo.go", - "exec_args.go", - "fds.go", - "filesystems.go", - "fs.go", - "inode.go", - "loadavg.go", - "meminfo.go", - "mounts.go", - "net.go", - "proc.go", - "stat.go", - "sys.go", - "sys_net.go", - "sys_net_state.go", - "task.go", - "uid_gid_map.go", - "uptime.go", - "version.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/log", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/proc/device", - "//pkg/sentry/fs/proc/seqfile", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/fsbridge", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/mm", - "//pkg/sentry/socket", - "//pkg/sentry/socket/unix", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sync", - "//pkg/syserror", - "//pkg/tcpip/header", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "proc_test", - size = "small", - srcs = [ - "net_test.go", - "sys_net_test.go", - ], - library = ":proc", - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/inet", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fs/proc/README.md b/pkg/sentry/fs/proc/README.md deleted file mode 100644 index 6667a0916..000000000 --- a/pkg/sentry/fs/proc/README.md +++ /dev/null @@ -1,336 +0,0 @@ -This document tracks what is implemented in procfs. Refer to -Documentation/filesystems/proc.txt in the Linux project for information about -procfs generally. - -**NOTE**: This document is not guaranteed to be up to date. If you find an -inconsistency, please file a bug. - -[TOC] - -## Kernel data - -The following files are implemented: - -<!-- mdformat off(don't wrap the table) --> - -| File /proc/ | Content | -| :------------------------ | :---------------------------------------------------- | -| [cpuinfo](#cpuinfo) | Info about the CPU | -| [filesystems](#filesystems) | Supported filesystems | -| [loadavg](#loadavg) | Load average of last 1, 5 & 15 minutes | -| [meminfo](#meminfo) | Overall memory info | -| [stat](#stat) | Overall kernel statistics | -| [sys](#sys) | Change parameters within the kernel | -| [uptime](#uptime) | Wall clock since boot, combined idle time of all cpus | -| [version](#version) | Kernel version | - -<!-- mdformat on --> - -### cpuinfo - -```bash -$ cat /proc/cpuinfo -processor : 0 -vendor_id : GenuineIntel -cpu family : 6 -model : 45 -model name : unknown -stepping : unknown -cpu MHz : 1234.588 -fpu : yes -fpu_exception : yes -cpuid level : 13 -wp : yes -flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx xsaveopt -bogomips : 1234.59 -clflush size : 64 -cache_alignment : 64 -address sizes : 46 bits physical, 48 bits virtual -power management: - -... -``` - -Notable divergences: - -Field name | Notes -:--------------- | :--------------------------------------- -model name | Always unknown -stepping | Always unknown -fpu | Always yes -fpu_exception | Always yes -wp | Always yes -bogomips | Bogus value (matches cpu MHz) -clflush size | Always 64 -cache_alignment | Always 64 -address sizes | Always 46 bits physical, 48 bits virtual -power management | Always blank - -Otherwise fields are derived from the sentry configuration. - -### filesystems - -```bash -$ cat /proc/filesystems -nodev 9p -nodev devpts -nodev devtmpfs -nodev proc -nodev sysfs -nodev tmpfs -``` - -### loadavg - -```bash -$ cat /proc/loadavg -0.00 0.00 0.00 0/0 0 -``` - -Column | Notes -:------------------------------------ | :---------- -CPU.IO utilization in last 1 minute | Always zero -CPU.IO utilization in last 5 minutes | Always zero -CPU.IO utilization in last 10 minutes | Always zero -Num currently running processes | Always zero -Total num processes | Always zero - -TODO(b/62345059): Populate the columns with accurate statistics. - -### meminfo - -```bash -$ cat /proc/meminfo -MemTotal: 2097152 kB -MemFree: 2083540 kB -MemAvailable: 2083540 kB -Buffers: 0 kB -Cached: 4428 kB -SwapCache: 0 kB -Active: 10812 kB -Inactive: 2216 kB -Active(anon): 8600 kB -Inactive(anon): 0 kB -Active(file): 2212 kB -Inactive(file): 2216 kB -Unevictable: 0 kB -Mlocked: 0 kB -SwapTotal: 0 kB -SwapFree: 0 kB -Dirty: 0 kB -Writeback: 0 kB -AnonPages: 8600 kB -Mapped: 4428 kB -Shmem: 0 kB - -``` - -Notable divergences: - -Field name | Notes -:---------------- | :----------------------------------------------------- -Buffers | Always zero, no block devices -SwapCache | Always zero, no swap -Inactive(anon) | Always zero, see SwapCache -Unevictable | Always zero TODO(b/31823263) -Mlocked | Always zero TODO(b/31823263) -SwapTotal | Always zero, no swap -SwapFree | Always zero, no swap -Dirty | Always zero TODO(b/31823263) -Writeback | Always zero TODO(b/31823263) -MemAvailable | Uses the same value as MemFree since there is no swap. -Slab | Missing -SReclaimable | Missing -SUnreclaim | Missing -KernelStack | Missing -PageTables | Missing -NFS_Unstable | Missing -Bounce | Missing -WritebackTmp | Missing -CommitLimit | Missing -Committed_AS | Missing -VmallocTotal | Missing -VmallocUsed | Missing -VmallocChunk | Missing -HardwareCorrupted | Missing -AnonHugePages | Missing -ShmemHugePages | Missing -ShmemPmdMapped | Missing -HugePages_Total | Missing -HugePages_Free | Missing -HugePages_Rsvd | Missing -HugePages_Surp | Missing -Hugepagesize | Missing -DirectMap4k | Missing -DirectMap2M | Missing -DirectMap1G | Missing - -### stat - -```bash -$ cat /proc/stat -cpu 0 0 0 0 0 0 0 0 0 0 -cpu0 0 0 0 0 0 0 0 0 0 0 -cpu1 0 0 0 0 0 0 0 0 0 0 -cpu2 0 0 0 0 0 0 0 0 0 0 -cpu3 0 0 0 0 0 0 0 0 0 0 -cpu4 0 0 0 0 0 0 0 0 0 0 -cpu5 0 0 0 0 0 0 0 0 0 0 -cpu6 0 0 0 0 0 0 0 0 0 0 -cpu7 0 0 0 0 0 0 0 0 0 0 -intr 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -ctxt 0 -btime 1504040968 -processes 0 -procs_running 0 -procs_blokkcked 0 -softirq 0 0 0 0 0 0 0 0 0 0 0 -``` - -All fields except for `btime` are always zero. - -TODO(b/37226836): Populate with accurate fields. - -### sys - -```bash -$ ls /proc/sys -kernel vm -``` - -Directory | Notes -:-------- | :---------------------------- -abi | Missing -debug | Missing -dev | Missing -fs | Missing -kernel | Contains hostname (only) -net | Missing -user | Missing -vm | Contains mmap_min_addr (only) - -### uptime - -```bash -$ cat /proc/uptime -3204.62 0.00 -``` - -Column | Notes -:------------------------------- | :---------------------------- -Total num seconds system running | Time since procfs was mounted -Number of seconds idle | Always zero - -### version - -```bash -$ cat /proc/version -Linux version 4.4 #1 SMP Sun Jan 10 15:06:54 PST 2016 -``` - -## Process-specific data - -The following files are implemented: - -File /proc/PID | Content -:---------------------- | :--------------------------------------------------- -[auxv](#auxv) | Copy of auxiliary vector for the process -[cmdline](#cmdline) | Command line arguments -[comm](#comm) | Command name associated with the process -[environ](#environ) | Process environment -[exe](#exe) | Symlink to the process's executable -[fd](#fd) | Directory containing links to open file descriptors -[fdinfo](#fdinfo) | Information associated with open file descriptors -[gid_map](#gid_map) | Mappings for group IDs inside the user namespace -[io](#io) | IO statistics -[maps](#maps) | Memory mappings (anon, executables, library files) -[mounts](#mounts) | Mounted filesystems -[mountinfo](#mountinfo) | Information about mounts -[ns](#ns) | Directory containing info about supported namespaces -[stat](#stat) | Process statistics -[statm](#statm) | Process memory statistics -[status](#status) | Process status in human readable format -[task](#task) | Directory containing info about running threads -[uid_map](#uid_map) | Mappings for user IDs inside the user namespace - -### auxv - -TODO - -### cmdline - -TODO - -### comm - -TODO - -### environment - -TODO - -### exe - -TODO - -### fd - -TODO - -### fdinfo - -TODO - -### gid_map - -TODO - -### io - -Only has data for rchar, wchar, syscr, and syscw. - -TODO: add more detail. - -### maps - -TODO - -### mounts - -TODO - -### mountinfo - -TODO - -### ns - -TODO - -### stat - -Only has data for pid, comm, state, ppid, utime, stime, cutime, cstime, -num_threads, and exit_signal. - -TODO: add more detail. - -### statm - -Only has data for vss and rss. - -TODO: add more detail. - -### status - -Contains data for Name, State, Tgid, Pid, Ppid, TracerPid, FDSize, VmSize, -VmRSS, Threads, CapInh, CapPrm, CapEff, CapBnd, Seccomp. - -TODO: add more detail. - -### task - -TODO - -### uid_map - -TODO diff --git a/pkg/sentry/fs/proc/device/BUILD b/pkg/sentry/fs/proc/device/BUILD deleted file mode 100644 index 52c9aa93d..000000000 --- a/pkg/sentry/fs/proc/device/BUILD +++ /dev/null @@ -1,10 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "device", - srcs = ["device.go"], - visibility = ["//pkg/sentry:internal"], - deps = ["//pkg/sentry/device"], -) diff --git a/pkg/sentry/fs/proc/device/device_state_autogen.go b/pkg/sentry/fs/proc/device/device_state_autogen.go new file mode 100755 index 000000000..4a5e3cc88 --- /dev/null +++ b/pkg/sentry/fs/proc/device/device_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package device diff --git a/pkg/sentry/fs/proc/net_test.go b/pkg/sentry/fs/proc/net_test.go deleted file mode 100644 index f18681405..000000000 --- a/pkg/sentry/fs/proc/net_test.go +++ /dev/null @@ -1,74 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/inet" -) - -func newIPv6TestStack() *inet.TestStack { - s := inet.NewTestStack() - s.SupportsIPv6Flag = true - return s -} - -func TestIfinet6NoAddresses(t *testing.T) { - n := &ifinet6{s: newIPv6TestStack()} - if got := n.contents(); got != nil { - t.Errorf("Got n.contents() = %v, want = %v", got, nil) - } -} - -func TestIfinet6(t *testing.T) { - s := newIPv6TestStack() - s.InterfacesMap[1] = inet.Interface{Name: "eth0"} - s.InterfaceAddrsMap[1] = []inet.InterfaceAddr{ - { - Family: linux.AF_INET6, - PrefixLen: 128, - Addr: []byte("\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f"), - }, - } - s.InterfacesMap[2] = inet.Interface{Name: "eth1"} - s.InterfaceAddrsMap[2] = []inet.InterfaceAddr{ - { - Family: linux.AF_INET6, - PrefixLen: 128, - Addr: []byte("\x10\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f"), - }, - } - want := map[string]struct{}{ - "000102030405060708090a0b0c0d0e0f 01 80 00 00 eth0\n": {}, - "101112131415161718191a1b1c1d1e1f 02 80 00 00 eth1\n": {}, - } - - n := &ifinet6{s: s} - contents := n.contents() - if len(contents) != len(want) { - t.Errorf("Got len(n.contents()) = %d, want = %d", len(contents), len(want)) - } - got := map[string]struct{}{} - for _, l := range contents { - got[l] = struct{}{} - } - - if !reflect.DeepEqual(got, want) { - t.Errorf("Got n.contents() = %v, want = %v", got, want) - } -} diff --git a/pkg/sentry/fs/proc/proc_state_autogen.go b/pkg/sentry/fs/proc/proc_state_autogen.go new file mode 100755 index 000000000..45f2b0a40 --- /dev/null +++ b/pkg/sentry/fs/proc/proc_state_autogen.go @@ -0,0 +1,743 @@ +// automatically generated by stateify. + +package proc + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *execArgInode) beforeSave() {} +func (x *execArgInode) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("arg", &x.arg) + m.Save("t", &x.t) +} + +func (x *execArgInode) afterLoad() {} +func (x *execArgInode) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("arg", &x.arg) + m.Load("t", &x.t) +} + +func (x *execArgFile) beforeSave() {} +func (x *execArgFile) save(m state.Map) { + x.beforeSave() + m.Save("arg", &x.arg) + m.Save("t", &x.t) +} + +func (x *execArgFile) afterLoad() {} +func (x *execArgFile) load(m state.Map) { + m.Load("arg", &x.arg) + m.Load("t", &x.t) +} + +func (x *fdDir) beforeSave() {} +func (x *fdDir) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("t", &x.t) +} + +func (x *fdDir) afterLoad() {} +func (x *fdDir) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("t", &x.t) +} + +func (x *fdDirFile) beforeSave() {} +func (x *fdDirFile) save(m state.Map) { + x.beforeSave() + m.Save("isInfoFile", &x.isInfoFile) + m.Save("t", &x.t) +} + +func (x *fdDirFile) afterLoad() {} +func (x *fdDirFile) load(m state.Map) { + m.Load("isInfoFile", &x.isInfoFile) + m.Load("t", &x.t) +} + +func (x *fdInfoDir) beforeSave() {} +func (x *fdInfoDir) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("t", &x.t) +} + +func (x *fdInfoDir) afterLoad() {} +func (x *fdInfoDir) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("t", &x.t) +} + +func (x *filesystemsData) beforeSave() {} +func (x *filesystemsData) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystemsData) afterLoad() {} +func (x *filesystemsData) load(m state.Map) { +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *taskOwnedInodeOps) beforeSave() {} +func (x *taskOwnedInodeOps) save(m state.Map) { + x.beforeSave() + m.Save("InodeOperations", &x.InodeOperations) + m.Save("t", &x.t) +} + +func (x *taskOwnedInodeOps) afterLoad() {} +func (x *taskOwnedInodeOps) load(m state.Map) { + m.Load("InodeOperations", &x.InodeOperations) + m.Load("t", &x.t) +} + +func (x *staticFileInodeOps) beforeSave() {} +func (x *staticFileInodeOps) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeStaticFileGetter", &x.InodeStaticFileGetter) +} + +func (x *staticFileInodeOps) afterLoad() {} +func (x *staticFileInodeOps) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeStaticFileGetter", &x.InodeStaticFileGetter) +} + +func (x *loadavgData) beforeSave() {} +func (x *loadavgData) save(m state.Map) { + x.beforeSave() +} + +func (x *loadavgData) afterLoad() {} +func (x *loadavgData) load(m state.Map) { +} + +func (x *meminfoData) beforeSave() {} +func (x *meminfoData) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *meminfoData) afterLoad() {} +func (x *meminfoData) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *mountInfoFile) beforeSave() {} +func (x *mountInfoFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *mountInfoFile) afterLoad() {} +func (x *mountInfoFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *mountsFile) beforeSave() {} +func (x *mountsFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *mountsFile) afterLoad() {} +func (x *mountsFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *ifinet6) beforeSave() {} +func (x *ifinet6) save(m state.Map) { + x.beforeSave() + m.Save("s", &x.s) +} + +func (x *ifinet6) afterLoad() {} +func (x *ifinet6) load(m state.Map) { + m.Load("s", &x.s) +} + +func (x *netDev) beforeSave() {} +func (x *netDev) save(m state.Map) { + x.beforeSave() + m.Save("s", &x.s) +} + +func (x *netDev) afterLoad() {} +func (x *netDev) load(m state.Map) { + m.Load("s", &x.s) +} + +func (x *netSnmp) beforeSave() {} +func (x *netSnmp) save(m state.Map) { + x.beforeSave() + m.Save("s", &x.s) +} + +func (x *netSnmp) afterLoad() {} +func (x *netSnmp) load(m state.Map) { + m.Load("s", &x.s) +} + +func (x *netRoute) beforeSave() {} +func (x *netRoute) save(m state.Map) { + x.beforeSave() + m.Save("s", &x.s) +} + +func (x *netRoute) afterLoad() {} +func (x *netRoute) load(m state.Map) { + m.Load("s", &x.s) +} + +func (x *netUnix) beforeSave() {} +func (x *netUnix) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *netUnix) afterLoad() {} +func (x *netUnix) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *netTCP) beforeSave() {} +func (x *netTCP) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *netTCP) afterLoad() {} +func (x *netTCP) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *netTCP6) beforeSave() {} +func (x *netTCP6) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *netTCP6) afterLoad() {} +func (x *netTCP6) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *netUDP) beforeSave() {} +func (x *netUDP) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *netUDP) afterLoad() {} +func (x *netUDP) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *proc) beforeSave() {} +func (x *proc) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("k", &x.k) + m.Save("pidns", &x.pidns) + m.Save("cgroupControllers", &x.cgroupControllers) +} + +func (x *proc) afterLoad() {} +func (x *proc) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("k", &x.k) + m.Load("pidns", &x.pidns) + m.Load("cgroupControllers", &x.cgroupControllers) +} + +func (x *self) beforeSave() {} +func (x *self) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) + m.Save("pidns", &x.pidns) +} + +func (x *self) afterLoad() {} +func (x *self) load(m state.Map) { + m.Load("Symlink", &x.Symlink) + m.Load("pidns", &x.pidns) +} + +func (x *threadSelf) beforeSave() {} +func (x *threadSelf) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) + m.Save("pidns", &x.pidns) +} + +func (x *threadSelf) afterLoad() {} +func (x *threadSelf) load(m state.Map) { + m.Load("Symlink", &x.Symlink) + m.Load("pidns", &x.pidns) +} + +func (x *rootProcFile) beforeSave() {} +func (x *rootProcFile) save(m state.Map) { + x.beforeSave() + m.Save("iops", &x.iops) +} + +func (x *rootProcFile) afterLoad() {} +func (x *rootProcFile) load(m state.Map) { + m.Load("iops", &x.iops) +} + +func (x *statData) beforeSave() {} +func (x *statData) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *statData) afterLoad() {} +func (x *statData) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *mmapMinAddrData) beforeSave() {} +func (x *mmapMinAddrData) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *mmapMinAddrData) afterLoad() {} +func (x *mmapMinAddrData) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *overcommitMemory) beforeSave() {} +func (x *overcommitMemory) save(m state.Map) { + x.beforeSave() +} + +func (x *overcommitMemory) afterLoad() {} +func (x *overcommitMemory) load(m state.Map) { +} + +func (x *hostname) beforeSave() {} +func (x *hostname) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) +} + +func (x *hostname) afterLoad() {} +func (x *hostname) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) +} + +func (x *hostnameFile) beforeSave() {} +func (x *hostnameFile) save(m state.Map) { + x.beforeSave() +} + +func (x *hostnameFile) afterLoad() {} +func (x *hostnameFile) load(m state.Map) { +} + +func (x *tcpMemInode) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("dir", &x.dir) + m.Save("s", &x.s) + m.Save("size", &x.size) +} + +func (x *tcpMemInode) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("dir", &x.dir) + m.LoadWait("s", &x.s) + m.Load("size", &x.size) + m.AfterLoad(x.afterLoad) +} + +func (x *tcpMemFile) beforeSave() {} +func (x *tcpMemFile) save(m state.Map) { + x.beforeSave() + m.Save("tcpMemInode", &x.tcpMemInode) +} + +func (x *tcpMemFile) afterLoad() {} +func (x *tcpMemFile) load(m state.Map) { + m.Load("tcpMemInode", &x.tcpMemInode) +} + +func (x *tcpSack) beforeSave() {} +func (x *tcpSack) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("stack", &x.stack) + m.Save("enabled", &x.enabled) +} + +func (x *tcpSack) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.LoadWait("stack", &x.stack) + m.Load("enabled", &x.enabled) + m.AfterLoad(x.afterLoad) +} + +func (x *tcpSackFile) beforeSave() {} +func (x *tcpSackFile) save(m state.Map) { + x.beforeSave() + m.Save("tcpSack", &x.tcpSack) + m.Save("stack", &x.stack) +} + +func (x *tcpSackFile) afterLoad() {} +func (x *tcpSackFile) load(m state.Map) { + m.Load("tcpSack", &x.tcpSack) + m.LoadWait("stack", &x.stack) +} + +func (x *taskDir) beforeSave() {} +func (x *taskDir) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("t", &x.t) + m.Save("pidns", &x.pidns) +} + +func (x *taskDir) afterLoad() {} +func (x *taskDir) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("t", &x.t) + m.Load("pidns", &x.pidns) +} + +func (x *subtasks) beforeSave() {} +func (x *subtasks) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("t", &x.t) + m.Save("p", &x.p) +} + +func (x *subtasks) afterLoad() {} +func (x *subtasks) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("t", &x.t) + m.Load("p", &x.p) +} + +func (x *subtasksFile) beforeSave() {} +func (x *subtasksFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) + m.Save("pidns", &x.pidns) +} + +func (x *subtasksFile) afterLoad() {} +func (x *subtasksFile) load(m state.Map) { + m.Load("t", &x.t) + m.Load("pidns", &x.pidns) +} + +func (x *exe) beforeSave() {} +func (x *exe) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) + m.Save("t", &x.t) +} + +func (x *exe) afterLoad() {} +func (x *exe) load(m state.Map) { + m.Load("Symlink", &x.Symlink) + m.Load("t", &x.t) +} + +func (x *namespaceSymlink) beforeSave() {} +func (x *namespaceSymlink) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) + m.Save("t", &x.t) +} + +func (x *namespaceSymlink) afterLoad() {} +func (x *namespaceSymlink) load(m state.Map) { + m.Load("Symlink", &x.Symlink) + m.Load("t", &x.t) +} + +func (x *mapsData) beforeSave() {} +func (x *mapsData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *mapsData) afterLoad() {} +func (x *mapsData) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *smapsData) beforeSave() {} +func (x *smapsData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *smapsData) afterLoad() {} +func (x *smapsData) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *taskStatData) beforeSave() {} +func (x *taskStatData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) + m.Save("tgstats", &x.tgstats) + m.Save("pidns", &x.pidns) +} + +func (x *taskStatData) afterLoad() {} +func (x *taskStatData) load(m state.Map) { + m.Load("t", &x.t) + m.Load("tgstats", &x.tgstats) + m.Load("pidns", &x.pidns) +} + +func (x *statmData) beforeSave() {} +func (x *statmData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *statmData) afterLoad() {} +func (x *statmData) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *statusData) beforeSave() {} +func (x *statusData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) + m.Save("pidns", &x.pidns) +} + +func (x *statusData) afterLoad() {} +func (x *statusData) load(m state.Map) { + m.Load("t", &x.t) + m.Load("pidns", &x.pidns) +} + +func (x *ioData) beforeSave() {} +func (x *ioData) save(m state.Map) { + x.beforeSave() + m.Save("ioUsage", &x.ioUsage) +} + +func (x *ioData) afterLoad() {} +func (x *ioData) load(m state.Map) { + m.Load("ioUsage", &x.ioUsage) +} + +func (x *comm) beforeSave() {} +func (x *comm) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("t", &x.t) +} + +func (x *comm) afterLoad() {} +func (x *comm) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("t", &x.t) +} + +func (x *commFile) beforeSave() {} +func (x *commFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *commFile) afterLoad() {} +func (x *commFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *auxvec) beforeSave() {} +func (x *auxvec) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("t", &x.t) +} + +func (x *auxvec) afterLoad() {} +func (x *auxvec) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("t", &x.t) +} + +func (x *auxvecFile) beforeSave() {} +func (x *auxvecFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *auxvecFile) afterLoad() {} +func (x *auxvecFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *oomScoreAdj) beforeSave() {} +func (x *oomScoreAdj) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("t", &x.t) +} + +func (x *oomScoreAdj) afterLoad() {} +func (x *oomScoreAdj) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("t", &x.t) +} + +func (x *oomScoreAdjFile) beforeSave() {} +func (x *oomScoreAdjFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *oomScoreAdjFile) afterLoad() {} +func (x *oomScoreAdjFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *idMapInodeOperations) beforeSave() {} +func (x *idMapInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("t", &x.t) + m.Save("gids", &x.gids) +} + +func (x *idMapInodeOperations) afterLoad() {} +func (x *idMapInodeOperations) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("t", &x.t) + m.Load("gids", &x.gids) +} + +func (x *idMapFileOperations) beforeSave() {} +func (x *idMapFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("iops", &x.iops) +} + +func (x *idMapFileOperations) afterLoad() {} +func (x *idMapFileOperations) load(m state.Map) { + m.Load("iops", &x.iops) +} + +func (x *uptime) beforeSave() {} +func (x *uptime) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("startTime", &x.startTime) +} + +func (x *uptime) afterLoad() {} +func (x *uptime) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("startTime", &x.startTime) +} + +func (x *uptimeFile) beforeSave() {} +func (x *uptimeFile) save(m state.Map) { + x.beforeSave() + m.Save("startTime", &x.startTime) +} + +func (x *uptimeFile) afterLoad() {} +func (x *uptimeFile) load(m state.Map) { + m.Load("startTime", &x.startTime) +} + +func (x *versionData) beforeSave() {} +func (x *versionData) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *versionData) afterLoad() {} +func (x *versionData) load(m state.Map) { + m.Load("k", &x.k) +} + +func init() { + state.Register("pkg/sentry/fs/proc.execArgInode", (*execArgInode)(nil), state.Fns{Save: (*execArgInode).save, Load: (*execArgInode).load}) + state.Register("pkg/sentry/fs/proc.execArgFile", (*execArgFile)(nil), state.Fns{Save: (*execArgFile).save, Load: (*execArgFile).load}) + state.Register("pkg/sentry/fs/proc.fdDir", (*fdDir)(nil), state.Fns{Save: (*fdDir).save, Load: (*fdDir).load}) + state.Register("pkg/sentry/fs/proc.fdDirFile", (*fdDirFile)(nil), state.Fns{Save: (*fdDirFile).save, Load: (*fdDirFile).load}) + state.Register("pkg/sentry/fs/proc.fdInfoDir", (*fdInfoDir)(nil), state.Fns{Save: (*fdInfoDir).save, Load: (*fdInfoDir).load}) + state.Register("pkg/sentry/fs/proc.filesystemsData", (*filesystemsData)(nil), state.Fns{Save: (*filesystemsData).save, Load: (*filesystemsData).load}) + state.Register("pkg/sentry/fs/proc.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("pkg/sentry/fs/proc.taskOwnedInodeOps", (*taskOwnedInodeOps)(nil), state.Fns{Save: (*taskOwnedInodeOps).save, Load: (*taskOwnedInodeOps).load}) + state.Register("pkg/sentry/fs/proc.staticFileInodeOps", (*staticFileInodeOps)(nil), state.Fns{Save: (*staticFileInodeOps).save, Load: (*staticFileInodeOps).load}) + state.Register("pkg/sentry/fs/proc.loadavgData", (*loadavgData)(nil), state.Fns{Save: (*loadavgData).save, Load: (*loadavgData).load}) + state.Register("pkg/sentry/fs/proc.meminfoData", (*meminfoData)(nil), state.Fns{Save: (*meminfoData).save, Load: (*meminfoData).load}) + state.Register("pkg/sentry/fs/proc.mountInfoFile", (*mountInfoFile)(nil), state.Fns{Save: (*mountInfoFile).save, Load: (*mountInfoFile).load}) + state.Register("pkg/sentry/fs/proc.mountsFile", (*mountsFile)(nil), state.Fns{Save: (*mountsFile).save, Load: (*mountsFile).load}) + state.Register("pkg/sentry/fs/proc.ifinet6", (*ifinet6)(nil), state.Fns{Save: (*ifinet6).save, Load: (*ifinet6).load}) + state.Register("pkg/sentry/fs/proc.netDev", (*netDev)(nil), state.Fns{Save: (*netDev).save, Load: (*netDev).load}) + state.Register("pkg/sentry/fs/proc.netSnmp", (*netSnmp)(nil), state.Fns{Save: (*netSnmp).save, Load: (*netSnmp).load}) + state.Register("pkg/sentry/fs/proc.netRoute", (*netRoute)(nil), state.Fns{Save: (*netRoute).save, Load: (*netRoute).load}) + state.Register("pkg/sentry/fs/proc.netUnix", (*netUnix)(nil), state.Fns{Save: (*netUnix).save, Load: (*netUnix).load}) + state.Register("pkg/sentry/fs/proc.netTCP", (*netTCP)(nil), state.Fns{Save: (*netTCP).save, Load: (*netTCP).load}) + state.Register("pkg/sentry/fs/proc.netTCP6", (*netTCP6)(nil), state.Fns{Save: (*netTCP6).save, Load: (*netTCP6).load}) + state.Register("pkg/sentry/fs/proc.netUDP", (*netUDP)(nil), state.Fns{Save: (*netUDP).save, Load: (*netUDP).load}) + state.Register("pkg/sentry/fs/proc.proc", (*proc)(nil), state.Fns{Save: (*proc).save, Load: (*proc).load}) + state.Register("pkg/sentry/fs/proc.self", (*self)(nil), state.Fns{Save: (*self).save, Load: (*self).load}) + state.Register("pkg/sentry/fs/proc.threadSelf", (*threadSelf)(nil), state.Fns{Save: (*threadSelf).save, Load: (*threadSelf).load}) + state.Register("pkg/sentry/fs/proc.rootProcFile", (*rootProcFile)(nil), state.Fns{Save: (*rootProcFile).save, Load: (*rootProcFile).load}) + state.Register("pkg/sentry/fs/proc.statData", (*statData)(nil), state.Fns{Save: (*statData).save, Load: (*statData).load}) + state.Register("pkg/sentry/fs/proc.mmapMinAddrData", (*mmapMinAddrData)(nil), state.Fns{Save: (*mmapMinAddrData).save, Load: (*mmapMinAddrData).load}) + state.Register("pkg/sentry/fs/proc.overcommitMemory", (*overcommitMemory)(nil), state.Fns{Save: (*overcommitMemory).save, Load: (*overcommitMemory).load}) + state.Register("pkg/sentry/fs/proc.hostname", (*hostname)(nil), state.Fns{Save: (*hostname).save, Load: (*hostname).load}) + state.Register("pkg/sentry/fs/proc.hostnameFile", (*hostnameFile)(nil), state.Fns{Save: (*hostnameFile).save, Load: (*hostnameFile).load}) + state.Register("pkg/sentry/fs/proc.tcpMemInode", (*tcpMemInode)(nil), state.Fns{Save: (*tcpMemInode).save, Load: (*tcpMemInode).load}) + state.Register("pkg/sentry/fs/proc.tcpMemFile", (*tcpMemFile)(nil), state.Fns{Save: (*tcpMemFile).save, Load: (*tcpMemFile).load}) + state.Register("pkg/sentry/fs/proc.tcpSack", (*tcpSack)(nil), state.Fns{Save: (*tcpSack).save, Load: (*tcpSack).load}) + state.Register("pkg/sentry/fs/proc.tcpSackFile", (*tcpSackFile)(nil), state.Fns{Save: (*tcpSackFile).save, Load: (*tcpSackFile).load}) + state.Register("pkg/sentry/fs/proc.taskDir", (*taskDir)(nil), state.Fns{Save: (*taskDir).save, Load: (*taskDir).load}) + state.Register("pkg/sentry/fs/proc.subtasks", (*subtasks)(nil), state.Fns{Save: (*subtasks).save, Load: (*subtasks).load}) + state.Register("pkg/sentry/fs/proc.subtasksFile", (*subtasksFile)(nil), state.Fns{Save: (*subtasksFile).save, Load: (*subtasksFile).load}) + state.Register("pkg/sentry/fs/proc.exe", (*exe)(nil), state.Fns{Save: (*exe).save, Load: (*exe).load}) + state.Register("pkg/sentry/fs/proc.namespaceSymlink", (*namespaceSymlink)(nil), state.Fns{Save: (*namespaceSymlink).save, Load: (*namespaceSymlink).load}) + state.Register("pkg/sentry/fs/proc.mapsData", (*mapsData)(nil), state.Fns{Save: (*mapsData).save, Load: (*mapsData).load}) + state.Register("pkg/sentry/fs/proc.smapsData", (*smapsData)(nil), state.Fns{Save: (*smapsData).save, Load: (*smapsData).load}) + state.Register("pkg/sentry/fs/proc.taskStatData", (*taskStatData)(nil), state.Fns{Save: (*taskStatData).save, Load: (*taskStatData).load}) + state.Register("pkg/sentry/fs/proc.statmData", (*statmData)(nil), state.Fns{Save: (*statmData).save, Load: (*statmData).load}) + state.Register("pkg/sentry/fs/proc.statusData", (*statusData)(nil), state.Fns{Save: (*statusData).save, Load: (*statusData).load}) + state.Register("pkg/sentry/fs/proc.ioData", (*ioData)(nil), state.Fns{Save: (*ioData).save, Load: (*ioData).load}) + state.Register("pkg/sentry/fs/proc.comm", (*comm)(nil), state.Fns{Save: (*comm).save, Load: (*comm).load}) + state.Register("pkg/sentry/fs/proc.commFile", (*commFile)(nil), state.Fns{Save: (*commFile).save, Load: (*commFile).load}) + state.Register("pkg/sentry/fs/proc.auxvec", (*auxvec)(nil), state.Fns{Save: (*auxvec).save, Load: (*auxvec).load}) + state.Register("pkg/sentry/fs/proc.auxvecFile", (*auxvecFile)(nil), state.Fns{Save: (*auxvecFile).save, Load: (*auxvecFile).load}) + state.Register("pkg/sentry/fs/proc.oomScoreAdj", (*oomScoreAdj)(nil), state.Fns{Save: (*oomScoreAdj).save, Load: (*oomScoreAdj).load}) + state.Register("pkg/sentry/fs/proc.oomScoreAdjFile", (*oomScoreAdjFile)(nil), state.Fns{Save: (*oomScoreAdjFile).save, Load: (*oomScoreAdjFile).load}) + state.Register("pkg/sentry/fs/proc.idMapInodeOperations", (*idMapInodeOperations)(nil), state.Fns{Save: (*idMapInodeOperations).save, Load: (*idMapInodeOperations).load}) + state.Register("pkg/sentry/fs/proc.idMapFileOperations", (*idMapFileOperations)(nil), state.Fns{Save: (*idMapFileOperations).save, Load: (*idMapFileOperations).load}) + state.Register("pkg/sentry/fs/proc.uptime", (*uptime)(nil), state.Fns{Save: (*uptime).save, Load: (*uptime).load}) + state.Register("pkg/sentry/fs/proc.uptimeFile", (*uptimeFile)(nil), state.Fns{Save: (*uptimeFile).save, Load: (*uptimeFile).load}) + state.Register("pkg/sentry/fs/proc.versionData", (*versionData)(nil), state.Fns{Save: (*versionData).save, Load: (*versionData).load}) +} diff --git a/pkg/sentry/fs/proc/seqfile/BUILD b/pkg/sentry/fs/proc/seqfile/BUILD deleted file mode 100644 index 21338d912..000000000 --- a/pkg/sentry/fs/proc/seqfile/BUILD +++ /dev/null @@ -1,35 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "seqfile", - srcs = ["seqfile.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/proc/device", - "//pkg/sentry/kernel/time", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "seqfile_test", - size = "small", - srcs = ["seqfile_test.go"], - library = ":seqfile", - deps = [ - "//pkg/context", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/ramfs", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fs/proc/seqfile/seqfile_state_autogen.go b/pkg/sentry/fs/proc/seqfile/seqfile_state_autogen.go new file mode 100755 index 000000000..cfd3a40b4 --- /dev/null +++ b/pkg/sentry/fs/proc/seqfile/seqfile_state_autogen.go @@ -0,0 +1,58 @@ +// automatically generated by stateify. + +package seqfile + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *SeqData) beforeSave() {} +func (x *SeqData) save(m state.Map) { + x.beforeSave() + m.Save("Buf", &x.Buf) + m.Save("Handle", &x.Handle) +} + +func (x *SeqData) afterLoad() {} +func (x *SeqData) load(m state.Map) { + m.Load("Buf", &x.Buf) + m.Load("Handle", &x.Handle) +} + +func (x *SeqFile) beforeSave() {} +func (x *SeqFile) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("SeqSource", &x.SeqSource) + m.Save("source", &x.source) + m.Save("generation", &x.generation) + m.Save("lastRead", &x.lastRead) +} + +func (x *SeqFile) afterLoad() {} +func (x *SeqFile) load(m state.Map) { + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("SeqSource", &x.SeqSource) + m.Load("source", &x.source) + m.Load("generation", &x.generation) + m.Load("lastRead", &x.lastRead) +} + +func (x *seqFileOperations) beforeSave() {} +func (x *seqFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("seqFile", &x.seqFile) +} + +func (x *seqFileOperations) afterLoad() {} +func (x *seqFileOperations) load(m state.Map) { + m.Load("seqFile", &x.seqFile) +} + +func init() { + state.Register("pkg/sentry/fs/proc/seqfile.SeqData", (*SeqData)(nil), state.Fns{Save: (*SeqData).save, Load: (*SeqData).load}) + state.Register("pkg/sentry/fs/proc/seqfile.SeqFile", (*SeqFile)(nil), state.Fns{Save: (*SeqFile).save, Load: (*SeqFile).load}) + state.Register("pkg/sentry/fs/proc/seqfile.seqFileOperations", (*seqFileOperations)(nil), state.Fns{Save: (*seqFileOperations).save, Load: (*seqFileOperations).load}) +} diff --git a/pkg/sentry/fs/proc/seqfile/seqfile_test.go b/pkg/sentry/fs/proc/seqfile/seqfile_test.go deleted file mode 100644 index 98e394569..000000000 --- a/pkg/sentry/fs/proc/seqfile/seqfile_test.go +++ /dev/null @@ -1,279 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package seqfile - -import ( - "bytes" - "fmt" - "io" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/usermem" -) - -type seqTest struct { - actual []SeqData - update bool -} - -func (s *seqTest) Init() { - var sq []SeqData - // Create some SeqData. - for i := 0; i < 10; i++ { - var b []byte - for j := 0; j < 10; j++ { - b = append(b, byte(i)) - } - sq = append(sq, SeqData{ - Buf: b, - Handle: &testHandle{i: i}, - }) - } - s.actual = sq -} - -// NeedsUpdate reports whether we need to update the data we've previously read. -func (s *seqTest) NeedsUpdate(int64) bool { - return s.update -} - -// ReadSeqFiledata returns a slice of SeqData which contains elements -// greater than the handle. -func (s *seqTest) ReadSeqFileData(ctx context.Context, handle SeqHandle) ([]SeqData, int64) { - if handle == nil { - return s.actual, 0 - } - h := *handle.(*testHandle) - var ret []SeqData - for _, b := range s.actual { - // We want the next one. - h2 := *b.Handle.(*testHandle) - if h2.i > h.i { - ret = append(ret, b) - } - } - return ret, 0 -} - -// Flatten a slice of slices into one slice. -func flatten(buf ...[]byte) []byte { - var flat []byte - for _, b := range buf { - flat = append(flat, b...) - } - return flat -} - -type testHandle struct { - i int -} - -type testTable struct { - offset int64 - readBufferSize int - expectedData []byte - expectedError error -} - -func runTableTests(ctx context.Context, table []testTable, dirent *fs.Dirent) error { - for _, tt := range table { - file, err := dirent.Inode.InodeOperations.GetFile(ctx, dirent, fs.FileFlags{Read: true}) - if err != nil { - return fmt.Errorf("GetFile returned error: %v", err) - } - - data := make([]byte, tt.readBufferSize) - resultLen, err := file.Preadv(ctx, usermem.BytesIOSequence(data), tt.offset) - if err != tt.expectedError { - return fmt.Errorf("t.Preadv(len: %v, offset: %v) (error) => %v expected %v", tt.readBufferSize, tt.offset, err, tt.expectedError) - } - expectedLen := int64(len(tt.expectedData)) - if resultLen != expectedLen { - // We make this just an error so we wall through and print the data below. - return fmt.Errorf("t.Preadv(len: %v, offset: %v) (size) => %v expected %v", tt.readBufferSize, tt.offset, resultLen, expectedLen) - } - if !bytes.Equal(data[:expectedLen], tt.expectedData) { - return fmt.Errorf("t.Preadv(len: %v, offset: %v) (data) => %v expected %v", tt.readBufferSize, tt.offset, data[:expectedLen], tt.expectedData) - } - } - return nil -} - -func TestSeqFile(t *testing.T) { - testSource := &seqTest{} - testSource.Init() - - // Create a file that can be R/W. - ctx := contexttest.Context(t) - m := fs.NewPseudoMountSource(ctx) - contents := map[string]*fs.Inode{ - "foo": NewSeqFileInode(ctx, testSource, m), - } - root := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0777)) - - // How about opening it? - inode := fs.NewInode(ctx, root, m, fs.StableAttr{Type: fs.Directory}) - dirent2, err := root.Lookup(ctx, inode, "foo") - if err != nil { - t.Fatalf("failed to walk to foo for n2: %v", err) - } - n2 := dirent2.Inode.InodeOperations - file2, err := n2.GetFile(ctx, dirent2, fs.FileFlags{Read: true, Write: true}) - if err != nil { - t.Fatalf("GetFile returned error: %v", err) - } - - // Writing? - if _, err := file2.Writev(ctx, usermem.BytesIOSequence([]byte("test"))); err == nil { - t.Fatalf("managed to write to n2: %v", err) - } - - // How about reading? - dirent3, err := root.Lookup(ctx, inode, "foo") - if err != nil { - t.Fatalf("failed to walk to foo: %v", err) - } - n3 := dirent3.Inode.InodeOperations - if n2 != n3 { - t.Error("got n2 != n3, want same") - } - - testSource.update = true - - table := []testTable{ - // Read past the end. - {100, 4, []byte{}, io.EOF}, - {110, 4, []byte{}, io.EOF}, - {200, 4, []byte{}, io.EOF}, - // Read a truncated first line. - {0, 4, testSource.actual[0].Buf[:4], nil}, - // Read the whole first line. - {0, 10, testSource.actual[0].Buf, nil}, - // Read the whole first line + 5 bytes of second line. - {0, 15, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf[:5]), nil}, - // First 4 bytes of the second line. - {10, 4, testSource.actual[1].Buf[:4], nil}, - // Read the two first lines. - {0, 20, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf), nil}, - // Read three lines. - {0, 30, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf, testSource.actual[2].Buf), nil}, - // Read everything, but use a bigger buffer than necessary. - {0, 150, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf, testSource.actual[2].Buf, testSource.actual[3].Buf, testSource.actual[4].Buf, testSource.actual[5].Buf, testSource.actual[6].Buf, testSource.actual[7].Buf, testSource.actual[8].Buf, testSource.actual[9].Buf), nil}, - // Read the last 3 bytes. - {97, 10, testSource.actual[9].Buf[7:], nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed with testSource.update = %v : %v", testSource.update, err) - } - - // Disable updates and do it again. - testSource.update = false - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed with testSource.update = %v: %v", testSource.update, err) - } -} - -// Test that we behave correctly when the file is updated. -func TestSeqFileFileUpdated(t *testing.T) { - testSource := &seqTest{} - testSource.Init() - testSource.update = true - - // Create a file that can be R/W. - ctx := contexttest.Context(t) - m := fs.NewPseudoMountSource(ctx) - contents := map[string]*fs.Inode{ - "foo": NewSeqFileInode(ctx, testSource, m), - } - root := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0777)) - - // How about opening it? - inode := fs.NewInode(ctx, root, m, fs.StableAttr{Type: fs.Directory}) - dirent2, err := root.Lookup(ctx, inode, "foo") - if err != nil { - t.Fatalf("failed to walk to foo for dirent2: %v", err) - } - - table := []testTable{ - {0, 16, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf[:6]), nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed: %v", err) - } - // Delete the first entry. - cut := testSource.actual[0].Buf - testSource.actual = testSource.actual[1:] - - table = []testTable{ - // Try reading buffer 0 with an offset. This will not delete the old data. - {1, 5, cut[1:6], nil}, - // Reset our file by reading at offset 0. - {0, 10, testSource.actual[0].Buf, nil}, - {16, 14, flatten(testSource.actual[1].Buf[6:], testSource.actual[2].Buf), nil}, - // Read the same data a second time. - {16, 14, flatten(testSource.actual[1].Buf[6:], testSource.actual[2].Buf), nil}, - // Read the following two lines. - {30, 20, flatten(testSource.actual[3].Buf, testSource.actual[4].Buf), nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after removing first entry: %v", err) - } - - // Add a new duplicate line in the middle (6666...) - after := testSource.actual[5:] - testSource.actual = testSource.actual[:4] - // Note the list must be sorted. - testSource.actual = append(testSource.actual, after[0]) - testSource.actual = append(testSource.actual, after...) - - table = []testTable{ - {50, 20, flatten(testSource.actual[4].Buf, testSource.actual[5].Buf), nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after adding middle entry: %v", err) - } - // This will be used in a later test. - oldTestData := testSource.actual - - // Delete everything. - testSource.actual = testSource.actual[:0] - table = []testTable{ - {20, 20, []byte{}, io.EOF}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after removing all entries: %v", err) - } - // Restore some of the data. - testSource.actual = oldTestData[:1] - table = []testTable{ - {6, 20, testSource.actual[0].Buf[6:], nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after adding first entry back: %v", err) - } - - // Re-extend the data - testSource.actual = oldTestData - table = []testTable{ - {30, 20, flatten(testSource.actual[3].Buf, testSource.actual[4].Buf), nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after extending testSource: %v", err) - } -} diff --git a/pkg/sentry/fs/proc/sys_net_test.go b/pkg/sentry/fs/proc/sys_net_test.go deleted file mode 100644 index 355e83d47..000000000 --- a/pkg/sentry/fs/proc/sys_net_test.go +++ /dev/null @@ -1,125 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/usermem" -) - -func TestQuerySendBufferSize(t *testing.T) { - ctx := context.Background() - s := inet.NewTestStack() - s.TCPSendBufSize = inet.TCPBufferSize{100, 200, 300} - tmi := &tcpMemInode{s: s, dir: tcpWMem} - tmf := &tcpMemFile{tcpMemInode: tmi} - - buf := make([]byte, 100) - dst := usermem.BytesIOSequence(buf) - n, err := tmf.Read(ctx, nil, dst, 0) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - if got, want := string(buf[:n]), "100\t200\t300\n"; got != want { - t.Fatalf("Bad string: got %v, want %v", got, want) - } -} - -func TestQueryRecvBufferSize(t *testing.T) { - ctx := context.Background() - s := inet.NewTestStack() - s.TCPRecvBufSize = inet.TCPBufferSize{100, 200, 300} - tmi := &tcpMemInode{s: s, dir: tcpRMem} - tmf := &tcpMemFile{tcpMemInode: tmi} - - buf := make([]byte, 100) - dst := usermem.BytesIOSequence(buf) - n, err := tmf.Read(ctx, nil, dst, 0) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - if got, want := string(buf[:n]), "100\t200\t300\n"; got != want { - t.Fatalf("Bad string: got %v, want %v", got, want) - } -} - -var cases = []struct { - str string - initial inet.TCPBufferSize - final inet.TCPBufferSize -}{ - { - str: "", - initial: inet.TCPBufferSize{1, 2, 3}, - final: inet.TCPBufferSize{1, 2, 3}, - }, - { - str: "100\n", - initial: inet.TCPBufferSize{1, 100, 200}, - final: inet.TCPBufferSize{100, 100, 200}, - }, - { - str: "100 200 300\n", - initial: inet.TCPBufferSize{1, 2, 3}, - final: inet.TCPBufferSize{100, 200, 300}, - }, -} - -func TestConfigureSendBufferSize(t *testing.T) { - ctx := context.Background() - s := inet.NewTestStack() - for _, c := range cases { - s.TCPSendBufSize = c.initial - tmi := &tcpMemInode{s: s, dir: tcpWMem} - tmf := &tcpMemFile{tcpMemInode: tmi} - - // Write the values. - src := usermem.BytesIOSequence([]byte(c.str)) - if n, err := tmf.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { - t.Errorf("Write, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) - } - - // Read the values from the stack and check them. - if s.TCPSendBufSize != c.final { - t.Errorf("TCPSendBufferSize, case = %q: got %v, wanted %v", c.str, s.TCPSendBufSize, c.final) - } - } -} - -func TestConfigureRecvBufferSize(t *testing.T) { - ctx := context.Background() - s := inet.NewTestStack() - for _, c := range cases { - s.TCPRecvBufSize = c.initial - tmi := &tcpMemInode{s: s, dir: tcpRMem} - tmf := &tcpMemFile{tcpMemInode: tmi} - - // Write the values. - src := usermem.BytesIOSequence([]byte(c.str)) - if n, err := tmf.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { - t.Errorf("Write, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) - } - - // Read the values from the stack and check them. - if s.TCPRecvBufSize != c.final { - t.Errorf("TCPRecvBufferSize, case = %q: got %v, wanted %v", c.str, s.TCPRecvBufSize, c.final) - } - } -} diff --git a/pkg/sentry/fs/ramfs/BUILD b/pkg/sentry/fs/ramfs/BUILD deleted file mode 100644 index 8ca823fb3..000000000 --- a/pkg/sentry/fs/ramfs/BUILD +++ /dev/null @@ -1,37 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "ramfs", - srcs = [ - "dir.go", - "socket.go", - "symlink.go", - "tree.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/socket/unix/transport", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "ramfs_test", - size = "small", - srcs = ["tree_test.go"], - library = ":ramfs", - deps = [ - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - ], -) diff --git a/pkg/sentry/fs/ramfs/ramfs_state_autogen.go b/pkg/sentry/fs/ramfs/ramfs_state_autogen.go new file mode 100755 index 000000000..0a001e0b6 --- /dev/null +++ b/pkg/sentry/fs/ramfs/ramfs_state_autogen.go @@ -0,0 +1,94 @@ +// automatically generated by stateify. + +package ramfs + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Dir) beforeSave() {} +func (x *Dir) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("children", &x.children) + m.Save("dentryMap", &x.dentryMap) +} + +func (x *Dir) afterLoad() {} +func (x *Dir) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("children", &x.children) + m.Load("dentryMap", &x.dentryMap) +} + +func (x *dirFileOperations) beforeSave() {} +func (x *dirFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("dirCursor", &x.dirCursor) + m.Save("dir", &x.dir) +} + +func (x *dirFileOperations) afterLoad() {} +func (x *dirFileOperations) load(m state.Map) { + m.Load("dirCursor", &x.dirCursor) + m.Load("dir", &x.dir) +} + +func (x *Socket) beforeSave() {} +func (x *Socket) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("ep", &x.ep) +} + +func (x *Socket) afterLoad() {} +func (x *Socket) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("ep", &x.ep) +} + +func (x *socketFileOperations) beforeSave() {} +func (x *socketFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *socketFileOperations) afterLoad() {} +func (x *socketFileOperations) load(m state.Map) { +} + +func (x *Symlink) beforeSave() {} +func (x *Symlink) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("Target", &x.Target) +} + +func (x *Symlink) afterLoad() {} +func (x *Symlink) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("Target", &x.Target) +} + +func (x *symlinkFileOperations) beforeSave() {} +func (x *symlinkFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *symlinkFileOperations) afterLoad() {} +func (x *symlinkFileOperations) load(m state.Map) { +} + +func init() { + state.Register("pkg/sentry/fs/ramfs.Dir", (*Dir)(nil), state.Fns{Save: (*Dir).save, Load: (*Dir).load}) + state.Register("pkg/sentry/fs/ramfs.dirFileOperations", (*dirFileOperations)(nil), state.Fns{Save: (*dirFileOperations).save, Load: (*dirFileOperations).load}) + state.Register("pkg/sentry/fs/ramfs.Socket", (*Socket)(nil), state.Fns{Save: (*Socket).save, Load: (*Socket).load}) + state.Register("pkg/sentry/fs/ramfs.socketFileOperations", (*socketFileOperations)(nil), state.Fns{Save: (*socketFileOperations).save, Load: (*socketFileOperations).load}) + state.Register("pkg/sentry/fs/ramfs.Symlink", (*Symlink)(nil), state.Fns{Save: (*Symlink).save, Load: (*Symlink).load}) + state.Register("pkg/sentry/fs/ramfs.symlinkFileOperations", (*symlinkFileOperations)(nil), state.Fns{Save: (*symlinkFileOperations).save, Load: (*symlinkFileOperations).load}) +} diff --git a/pkg/sentry/fs/ramfs/tree_test.go b/pkg/sentry/fs/ramfs/tree_test.go deleted file mode 100644 index a6ed8b2c5..000000000 --- a/pkg/sentry/fs/ramfs/tree_test.go +++ /dev/null @@ -1,80 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ramfs - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -func TestMakeDirectoryTree(t *testing.T) { - - for _, test := range []struct { - name string - subdirs []string - }{ - { - name: "abs paths", - subdirs: []string{ - "/tmp", - "/tmp/a/b", - "/tmp/a/c/d", - "/tmp/c", - "/proc", - "/dev/a/b", - "/tmp", - }, - }, - { - name: "rel paths", - subdirs: []string{ - "tmp", - "tmp/a/b", - "tmp/a/c/d", - "tmp/c", - "proc", - "dev/a/b", - "tmp", - }, - }, - } { - ctx := contexttest.Context(t) - mount := fs.NewPseudoMountSource(ctx) - tree, err := MakeDirectoryTree(ctx, mount, test.subdirs) - if err != nil { - t.Errorf("%s: failed to make ramfs tree, got error %v, want nil", test.name, err) - continue - } - - // Expect to be able to find each of the paths. - mm, err := fs.NewMountNamespace(ctx, tree) - if err != nil { - t.Errorf("%s: failed to create mount manager: %v", test.name, err) - continue - } - root := mm.Root() - defer mm.DecRef() - - for _, p := range test.subdirs { - maxTraversals := uint(0) - if _, err := mm.FindInode(ctx, root, nil, p, &maxTraversals); err != nil { - t.Errorf("%s: failed to find node %s: %v", test.name, p, err) - break - } - } - } -} diff --git a/pkg/sentry/fs/sys/BUILD b/pkg/sentry/fs/sys/BUILD deleted file mode 100644 index f2e8b9932..000000000 --- a/pkg/sentry/fs/sys/BUILD +++ /dev/null @@ -1,24 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "sys", - srcs = [ - "device.go", - "devices.go", - "fs.go", - "sys.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/kernel", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fs/sys/sys_state_autogen.go b/pkg/sentry/fs/sys/sys_state_autogen.go new file mode 100755 index 000000000..733c504b1 --- /dev/null +++ b/pkg/sentry/fs/sys/sys_state_autogen.go @@ -0,0 +1,34 @@ +// automatically generated by stateify. + +package sys + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *cpunum) beforeSave() {} +func (x *cpunum) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeStaticFileGetter", &x.InodeStaticFileGetter) +} + +func (x *cpunum) afterLoad() {} +func (x *cpunum) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeStaticFileGetter", &x.InodeStaticFileGetter) +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func init() { + state.Register("pkg/sentry/fs/sys.cpunum", (*cpunum)(nil), state.Fns{Save: (*cpunum).save, Load: (*cpunum).load}) + state.Register("pkg/sentry/fs/sys.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) +} diff --git a/pkg/sentry/fs/timerfd/BUILD b/pkg/sentry/fs/timerfd/BUILD deleted file mode 100644 index d16cdb4df..000000000 --- a/pkg/sentry/fs/timerfd/BUILD +++ /dev/null @@ -1,19 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "timerfd", - srcs = ["timerfd.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel/time", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/fs/timerfd/timerfd_state_autogen.go b/pkg/sentry/fs/timerfd/timerfd_state_autogen.go new file mode 100755 index 000000000..b1335d3c7 --- /dev/null +++ b/pkg/sentry/fs/timerfd/timerfd_state_autogen.go @@ -0,0 +1,27 @@ +// automatically generated by stateify. + +package timerfd + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *TimerOperations) beforeSave() {} +func (x *TimerOperations) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.events) { + m.Failf("events is %v, expected zero", x.events) + } + m.Save("timer", &x.timer) + m.Save("val", &x.val) +} + +func (x *TimerOperations) afterLoad() {} +func (x *TimerOperations) load(m state.Map) { + m.Load("timer", &x.timer) + m.Load("val", &x.val) +} + +func init() { + state.Register("pkg/sentry/fs/timerfd.TimerOperations", (*TimerOperations)(nil), state.Fns{Save: (*TimerOperations).save, Load: (*TimerOperations).load}) +} diff --git a/pkg/sentry/fs/tmpfs/BUILD b/pkg/sentry/fs/tmpfs/BUILD deleted file mode 100644 index aa7199014..000000000 --- a/pkg/sentry/fs/tmpfs/BUILD +++ /dev/null @@ -1,50 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "tmpfs", - srcs = [ - "device.go", - "file_regular.go", - "fs.go", - "inode_file.go", - "tmpfs.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/metric", - "//pkg/safemem", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/pipe", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "tmpfs_test", - size = "small", - srcs = ["file_test.go"], - library = ":tmpfs", - deps = [ - "//pkg/context", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/contexttest", - "//pkg/sentry/usage", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fs/tmpfs/file_test.go b/pkg/sentry/fs/tmpfs/file_test.go deleted file mode 100644 index aaba35502..000000000 --- a/pkg/sentry/fs/tmpfs/file_test.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "bytes" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/usermem" -) - -func newFileInode(ctx context.Context) *fs.Inode { - m := fs.NewCachingMountSource(ctx, &Filesystem{}, fs.MountSourceFlags{}) - iops := NewInMemoryFile(ctx, usage.Tmpfs, fs.WithCurrentTime(ctx, fs.UnstableAttr{})) - return fs.NewInode(ctx, iops, m, fs.StableAttr{ - DeviceID: tmpfsDevice.DeviceID(), - InodeID: tmpfsDevice.NextIno(), - BlockSize: usermem.PageSize, - Type: fs.RegularFile, - }) -} - -func newFile(ctx context.Context) *fs.File { - inode := newFileInode(ctx) - f, _ := inode.GetFile(ctx, fs.NewDirent(ctx, inode, "stub"), fs.FileFlags{Read: true, Write: true}) - return f -} - -// Allocate once, write twice. -func TestGrow(t *testing.T) { - ctx := contexttest.Context(t) - f := newFile(ctx) - defer f.DecRef() - - abuf := bytes.Repeat([]byte{'a'}, 68) - n, err := f.Pwritev(ctx, usermem.BytesIOSequence(abuf), 0) - if n != int64(len(abuf)) || err != nil { - t.Fatalf("Pwritev got (%d, %v) want (%d, nil)", n, err, len(abuf)) - } - - bbuf := bytes.Repeat([]byte{'b'}, 856) - n, err = f.Pwritev(ctx, usermem.BytesIOSequence(bbuf), 68) - if n != int64(len(bbuf)) || err != nil { - t.Fatalf("Pwritev got (%d, %v) want (%d, nil)", n, err, len(bbuf)) - } - - rbuf := make([]byte, len(abuf)+len(bbuf)) - n, err = f.Preadv(ctx, usermem.BytesIOSequence(rbuf), 0) - if n != int64(len(rbuf)) || err != nil { - t.Fatalf("Preadv got (%d, %v) want (%d, nil)", n, err, len(rbuf)) - } - - if want := append(abuf, bbuf...); !bytes.Equal(rbuf, want) { - t.Fatalf("Read %v, want %v", rbuf, want) - } -} diff --git a/pkg/sentry/fs/tmpfs/tmpfs_state_autogen.go b/pkg/sentry/fs/tmpfs/tmpfs_state_autogen.go new file mode 100755 index 000000000..e4d2584fd --- /dev/null +++ b/pkg/sentry/fs/tmpfs/tmpfs_state_autogen.go @@ -0,0 +1,108 @@ +// automatically generated by stateify. + +package tmpfs + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *regularFileOperations) beforeSave() {} +func (x *regularFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("iops", &x.iops) +} + +func (x *regularFileOperations) afterLoad() {} +func (x *regularFileOperations) load(m state.Map) { + m.Load("iops", &x.iops) +} + +func (x *Filesystem) beforeSave() {} +func (x *Filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *Filesystem) afterLoad() {} +func (x *Filesystem) load(m state.Map) { +} + +func (x *fileInodeOperations) beforeSave() {} +func (x *fileInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("kernel", &x.kernel) + m.Save("memUsage", &x.memUsage) + m.Save("attr", &x.attr) + m.Save("mappings", &x.mappings) + m.Save("writableMappingPages", &x.writableMappingPages) + m.Save("data", &x.data) + m.Save("seals", &x.seals) +} + +func (x *fileInodeOperations) afterLoad() {} +func (x *fileInodeOperations) load(m state.Map) { + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("kernel", &x.kernel) + m.Load("memUsage", &x.memUsage) + m.Load("attr", &x.attr) + m.Load("mappings", &x.mappings) + m.Load("writableMappingPages", &x.writableMappingPages) + m.Load("data", &x.data) + m.Load("seals", &x.seals) +} + +func (x *Dir) beforeSave() {} +func (x *Dir) save(m state.Map) { + x.beforeSave() + m.Save("ramfsDir", &x.ramfsDir) + m.Save("kernel", &x.kernel) +} + +func (x *Dir) load(m state.Map) { + m.Load("ramfsDir", &x.ramfsDir) + m.Load("kernel", &x.kernel) + m.AfterLoad(x.afterLoad) +} + +func (x *Symlink) beforeSave() {} +func (x *Symlink) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) +} + +func (x *Symlink) afterLoad() {} +func (x *Symlink) load(m state.Map) { + m.Load("Symlink", &x.Symlink) +} + +func (x *Socket) beforeSave() {} +func (x *Socket) save(m state.Map) { + x.beforeSave() + m.Save("Socket", &x.Socket) +} + +func (x *Socket) afterLoad() {} +func (x *Socket) load(m state.Map) { + m.Load("Socket", &x.Socket) +} + +func (x *Fifo) beforeSave() {} +func (x *Fifo) save(m state.Map) { + x.beforeSave() + m.Save("InodeOperations", &x.InodeOperations) +} + +func (x *Fifo) afterLoad() {} +func (x *Fifo) load(m state.Map) { + m.Load("InodeOperations", &x.InodeOperations) +} + +func init() { + state.Register("pkg/sentry/fs/tmpfs.regularFileOperations", (*regularFileOperations)(nil), state.Fns{Save: (*regularFileOperations).save, Load: (*regularFileOperations).load}) + state.Register("pkg/sentry/fs/tmpfs.Filesystem", (*Filesystem)(nil), state.Fns{Save: (*Filesystem).save, Load: (*Filesystem).load}) + state.Register("pkg/sentry/fs/tmpfs.fileInodeOperations", (*fileInodeOperations)(nil), state.Fns{Save: (*fileInodeOperations).save, Load: (*fileInodeOperations).load}) + state.Register("pkg/sentry/fs/tmpfs.Dir", (*Dir)(nil), state.Fns{Save: (*Dir).save, Load: (*Dir).load}) + state.Register("pkg/sentry/fs/tmpfs.Symlink", (*Symlink)(nil), state.Fns{Save: (*Symlink).save, Load: (*Symlink).load}) + state.Register("pkg/sentry/fs/tmpfs.Socket", (*Socket)(nil), state.Fns{Save: (*Socket).save, Load: (*Socket).load}) + state.Register("pkg/sentry/fs/tmpfs.Fifo", (*Fifo)(nil), state.Fns{Save: (*Fifo).save, Load: (*Fifo).load}) +} diff --git a/pkg/sentry/fs/tty/BUILD b/pkg/sentry/fs/tty/BUILD deleted file mode 100644 index 5cb0e0417..000000000 --- a/pkg/sentry/fs/tty/BUILD +++ /dev/null @@ -1,47 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "tty", - srcs = [ - "dir.go", - "fs.go", - "line_discipline.go", - "master.go", - "queue.go", - "slave.go", - "terminal.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/refs", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/unimpl", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "tty_test", - size = "small", - srcs = ["tty_test.go"], - library = ":tty", - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/contexttest", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fs/tty/tty_state_autogen.go b/pkg/sentry/fs/tty/tty_state_autogen.go new file mode 100755 index 000000000..9963096dd --- /dev/null +++ b/pkg/sentry/fs/tty/tty_state_autogen.go @@ -0,0 +1,210 @@ +// automatically generated by stateify. + +package tty + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *dirInodeOperations) beforeSave() {} +func (x *dirInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("msrc", &x.msrc) + m.Save("master", &x.master) + m.Save("slaves", &x.slaves) + m.Save("dentryMap", &x.dentryMap) + m.Save("next", &x.next) +} + +func (x *dirInodeOperations) afterLoad() {} +func (x *dirInodeOperations) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("msrc", &x.msrc) + m.Load("master", &x.master) + m.Load("slaves", &x.slaves) + m.Load("dentryMap", &x.dentryMap) + m.Load("next", &x.next) +} + +func (x *dirFileOperations) beforeSave() {} +func (x *dirFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("di", &x.di) + m.Save("dirCursor", &x.dirCursor) +} + +func (x *dirFileOperations) afterLoad() {} +func (x *dirFileOperations) load(m state.Map) { + m.Load("di", &x.di) + m.Load("dirCursor", &x.dirCursor) +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *superOperations) beforeSave() {} +func (x *superOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *superOperations) afterLoad() {} +func (x *superOperations) load(m state.Map) { +} + +func (x *lineDiscipline) beforeSave() {} +func (x *lineDiscipline) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.masterWaiter) { + m.Failf("masterWaiter is %v, expected zero", x.masterWaiter) + } + if !state.IsZeroValue(x.slaveWaiter) { + m.Failf("slaveWaiter is %v, expected zero", x.slaveWaiter) + } + m.Save("size", &x.size) + m.Save("inQueue", &x.inQueue) + m.Save("outQueue", &x.outQueue) + m.Save("termios", &x.termios) + m.Save("column", &x.column) +} + +func (x *lineDiscipline) afterLoad() {} +func (x *lineDiscipline) load(m state.Map) { + m.Load("size", &x.size) + m.Load("inQueue", &x.inQueue) + m.Load("outQueue", &x.outQueue) + m.Load("termios", &x.termios) + m.Load("column", &x.column) +} + +func (x *outputQueueTransformer) beforeSave() {} +func (x *outputQueueTransformer) save(m state.Map) { + x.beforeSave() +} + +func (x *outputQueueTransformer) afterLoad() {} +func (x *outputQueueTransformer) load(m state.Map) { +} + +func (x *inputQueueTransformer) beforeSave() {} +func (x *inputQueueTransformer) save(m state.Map) { + x.beforeSave() +} + +func (x *inputQueueTransformer) afterLoad() {} +func (x *inputQueueTransformer) load(m state.Map) { +} + +func (x *masterInodeOperations) beforeSave() {} +func (x *masterInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("d", &x.d) +} + +func (x *masterInodeOperations) afterLoad() {} +func (x *masterInodeOperations) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("d", &x.d) +} + +func (x *masterFileOperations) beforeSave() {} +func (x *masterFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("d", &x.d) + m.Save("t", &x.t) +} + +func (x *masterFileOperations) afterLoad() {} +func (x *masterFileOperations) load(m state.Map) { + m.Load("d", &x.d) + m.Load("t", &x.t) +} + +func (x *queue) beforeSave() {} +func (x *queue) save(m state.Map) { + x.beforeSave() + m.Save("readBuf", &x.readBuf) + m.Save("waitBuf", &x.waitBuf) + m.Save("waitBufLen", &x.waitBufLen) + m.Save("readable", &x.readable) + m.Save("transformer", &x.transformer) +} + +func (x *queue) afterLoad() {} +func (x *queue) load(m state.Map) { + m.Load("readBuf", &x.readBuf) + m.Load("waitBuf", &x.waitBuf) + m.Load("waitBufLen", &x.waitBufLen) + m.Load("readable", &x.readable) + m.Load("transformer", &x.transformer) +} + +func (x *slaveInodeOperations) beforeSave() {} +func (x *slaveInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("d", &x.d) + m.Save("t", &x.t) +} + +func (x *slaveInodeOperations) afterLoad() {} +func (x *slaveInodeOperations) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("d", &x.d) + m.Load("t", &x.t) +} + +func (x *slaveFileOperations) beforeSave() {} +func (x *slaveFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("si", &x.si) +} + +func (x *slaveFileOperations) afterLoad() {} +func (x *slaveFileOperations) load(m state.Map) { + m.Load("si", &x.si) +} + +func (x *Terminal) beforeSave() {} +func (x *Terminal) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("n", &x.n) + m.Save("d", &x.d) + m.Save("ld", &x.ld) + m.Save("masterKTTY", &x.masterKTTY) + m.Save("slaveKTTY", &x.slaveKTTY) +} + +func (x *Terminal) afterLoad() {} +func (x *Terminal) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("n", &x.n) + m.Load("d", &x.d) + m.Load("ld", &x.ld) + m.Load("masterKTTY", &x.masterKTTY) + m.Load("slaveKTTY", &x.slaveKTTY) +} + +func init() { + state.Register("pkg/sentry/fs/tty.dirInodeOperations", (*dirInodeOperations)(nil), state.Fns{Save: (*dirInodeOperations).save, Load: (*dirInodeOperations).load}) + state.Register("pkg/sentry/fs/tty.dirFileOperations", (*dirFileOperations)(nil), state.Fns{Save: (*dirFileOperations).save, Load: (*dirFileOperations).load}) + state.Register("pkg/sentry/fs/tty.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("pkg/sentry/fs/tty.superOperations", (*superOperations)(nil), state.Fns{Save: (*superOperations).save, Load: (*superOperations).load}) + state.Register("pkg/sentry/fs/tty.lineDiscipline", (*lineDiscipline)(nil), state.Fns{Save: (*lineDiscipline).save, Load: (*lineDiscipline).load}) + state.Register("pkg/sentry/fs/tty.outputQueueTransformer", (*outputQueueTransformer)(nil), state.Fns{Save: (*outputQueueTransformer).save, Load: (*outputQueueTransformer).load}) + state.Register("pkg/sentry/fs/tty.inputQueueTransformer", (*inputQueueTransformer)(nil), state.Fns{Save: (*inputQueueTransformer).save, Load: (*inputQueueTransformer).load}) + state.Register("pkg/sentry/fs/tty.masterInodeOperations", (*masterInodeOperations)(nil), state.Fns{Save: (*masterInodeOperations).save, Load: (*masterInodeOperations).load}) + state.Register("pkg/sentry/fs/tty.masterFileOperations", (*masterFileOperations)(nil), state.Fns{Save: (*masterFileOperations).save, Load: (*masterFileOperations).load}) + state.Register("pkg/sentry/fs/tty.queue", (*queue)(nil), state.Fns{Save: (*queue).save, Load: (*queue).load}) + state.Register("pkg/sentry/fs/tty.slaveInodeOperations", (*slaveInodeOperations)(nil), state.Fns{Save: (*slaveInodeOperations).save, Load: (*slaveInodeOperations).load}) + state.Register("pkg/sentry/fs/tty.slaveFileOperations", (*slaveFileOperations)(nil), state.Fns{Save: (*slaveFileOperations).save, Load: (*slaveFileOperations).load}) + state.Register("pkg/sentry/fs/tty.Terminal", (*Terminal)(nil), state.Fns{Save: (*Terminal).save, Load: (*Terminal).load}) +} diff --git a/pkg/sentry/fs/tty/tty_test.go b/pkg/sentry/fs/tty/tty_test.go deleted file mode 100644 index 2cbc05678..000000000 --- a/pkg/sentry/fs/tty/tty_test.go +++ /dev/null @@ -1,56 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tty - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/usermem" -) - -func TestSimpleMasterToSlave(t *testing.T) { - ld := newLineDiscipline(linux.DefaultSlaveTermios) - ctx := contexttest.Context(t) - inBytes := []byte("hello, tty\n") - src := usermem.BytesIOSequence(inBytes) - outBytes := make([]byte, 32) - dst := usermem.BytesIOSequence(outBytes) - - // Write to the input queue. - nw, err := ld.inputQueueWrite(ctx, src) - if err != nil { - t.Fatalf("error writing to input queue: %v", err) - } - if nw != int64(len(inBytes)) { - t.Fatalf("wrote wrong length: got %d, want %d", nw, len(inBytes)) - } - - // Read from the input queue. - nr, err := ld.inputQueueRead(ctx, dst) - if err != nil { - t.Fatalf("error reading from input queue: %v", err) - } - if nr != int64(len(inBytes)) { - t.Fatalf("read wrong length: got %d, want %d", nr, len(inBytes)) - } - - outStr := string(outBytes[:nr]) - inStr := string(inBytes) - if outStr != inStr { - t.Fatalf("written and read strings do not match: got %q, want %q", outStr, inStr) - } -} diff --git a/pkg/sentry/fsbridge/BUILD b/pkg/sentry/fsbridge/BUILD deleted file mode 100644 index 6c798f0bd..000000000 --- a/pkg/sentry/fsbridge/BUILD +++ /dev/null @@ -1,24 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "fsbridge", - srcs = [ - "bridge.go", - "fs.go", - "vfs.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/memmap", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fsbridge/bridge.go b/pkg/sentry/fsbridge/bridge.go index 8e7590721..8e7590721 100644..100755 --- a/pkg/sentry/fsbridge/bridge.go +++ b/pkg/sentry/fsbridge/bridge.go diff --git a/pkg/sentry/fsbridge/fs.go b/pkg/sentry/fsbridge/fs.go index 093ce1fb3..093ce1fb3 100644..100755 --- a/pkg/sentry/fsbridge/fs.go +++ b/pkg/sentry/fsbridge/fs.go diff --git a/pkg/sentry/fsbridge/fsbridge_state_autogen.go b/pkg/sentry/fsbridge/fsbridge_state_autogen.go new file mode 100755 index 000000000..51b57d859 --- /dev/null +++ b/pkg/sentry/fsbridge/fsbridge_state_autogen.go @@ -0,0 +1,66 @@ +// automatically generated by stateify. + +package fsbridge + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *fsFile) beforeSave() {} +func (x *fsFile) save(m state.Map) { + x.beforeSave() + m.Save("file", &x.file) +} + +func (x *fsFile) afterLoad() {} +func (x *fsFile) load(m state.Map) { + m.Load("file", &x.file) +} + +func (x *fsLookup) beforeSave() {} +func (x *fsLookup) save(m state.Map) { + x.beforeSave() + m.Save("mntns", &x.mntns) + m.Save("root", &x.root) + m.Save("workingDir", &x.workingDir) +} + +func (x *fsLookup) afterLoad() {} +func (x *fsLookup) load(m state.Map) { + m.Load("mntns", &x.mntns) + m.Load("root", &x.root) + m.Load("workingDir", &x.workingDir) +} + +func (x *vfsFile) beforeSave() {} +func (x *vfsFile) save(m state.Map) { + x.beforeSave() + m.Save("file", &x.file) +} + +func (x *vfsFile) afterLoad() {} +func (x *vfsFile) load(m state.Map) { + m.Load("file", &x.file) +} + +func (x *vfsLookup) beforeSave() {} +func (x *vfsLookup) save(m state.Map) { + x.beforeSave() + m.Save("mntns", &x.mntns) + m.Save("root", &x.root) + m.Save("workingDir", &x.workingDir) +} + +func (x *vfsLookup) afterLoad() {} +func (x *vfsLookup) load(m state.Map) { + m.Load("mntns", &x.mntns) + m.Load("root", &x.root) + m.Load("workingDir", &x.workingDir) +} + +func init() { + state.Register("pkg/sentry/fsbridge.fsFile", (*fsFile)(nil), state.Fns{Save: (*fsFile).save, Load: (*fsFile).load}) + state.Register("pkg/sentry/fsbridge.fsLookup", (*fsLookup)(nil), state.Fns{Save: (*fsLookup).save, Load: (*fsLookup).load}) + state.Register("pkg/sentry/fsbridge.vfsFile", (*vfsFile)(nil), state.Fns{Save: (*vfsFile).save, Load: (*vfsFile).load}) + state.Register("pkg/sentry/fsbridge.vfsLookup", (*vfsLookup)(nil), state.Fns{Save: (*vfsLookup).save, Load: (*vfsLookup).load}) +} diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go index 79b808359..79b808359 100644..100755 --- a/pkg/sentry/fsbridge/vfs.go +++ b/pkg/sentry/fsbridge/vfs.go diff --git a/pkg/sentry/fsimpl/devtmpfs/BUILD b/pkg/sentry/fsimpl/devtmpfs/BUILD deleted file mode 100644 index aa0c2ad8c..000000000 --- a/pkg/sentry/fsimpl/devtmpfs/BUILD +++ /dev/null @@ -1,33 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -licenses(["notice"]) - -go_library( - name = "devtmpfs", - srcs = ["devtmpfs.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/fsimpl/tmpfs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/sync", - ], -) - -go_test( - name = "devtmpfs_test", - size = "small", - srcs = ["devtmpfs_test.go"], - library = ":devtmpfs", - deps = [ - "//pkg/abi/linux", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/tmpfs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - ], -) diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go deleted file mode 100644 index abd4f24e7..000000000 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ /dev/null @@ -1,191 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package devtmpfs provides an implementation of /dev based on tmpfs, -// analogous to Linux's devtmpfs. -package devtmpfs - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" -) - -// Name is the default filesystem name. -const Name = "devtmpfs" - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct { - initOnce sync.Once - initErr error - - // fs is the tmpfs filesystem that backs all mounts of this FilesystemType. - // root is fs' root. fs and root are immutable. - fs *vfs.Filesystem - root *vfs.Dentry -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fst *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fst.initOnce.Do(func() { - fs, root, err := tmpfs.FilesystemType{}.GetFilesystem(ctx, vfsObj, creds, "" /* source */, vfs.GetFilesystemOptions{ - Data: "mode=0755", // opts from drivers/base/devtmpfs.c:devtmpfs_init() - }) - if err != nil { - fst.initErr = err - return - } - fst.fs = fs - fst.root = root - }) - if fst.initErr != nil { - return nil, nil, fst.initErr - } - fst.fs.IncRef() - fst.root.IncRef() - return fst.fs, fst.root, nil -} - -// Accessor allows devices to create device special files in devtmpfs. -type Accessor struct { - vfsObj *vfs.VirtualFilesystem - mntns *vfs.MountNamespace - root vfs.VirtualDentry - creds *auth.Credentials -} - -// NewAccessor returns an Accessor that supports creation of device special -// files in the devtmpfs instance registered with name fsTypeName in vfsObj. -func NewAccessor(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, fsTypeName string) (*Accessor, error) { - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "devtmpfs" /* source */, fsTypeName, &vfs.GetFilesystemOptions{}) - if err != nil { - return nil, err - } - return &Accessor{ - vfsObj: vfsObj, - mntns: mntns, - root: mntns.Root(), - creds: creds, - }, nil -} - -// Release must be called when a is no longer in use. -func (a *Accessor) Release() { - a.root.DecRef() - a.mntns.DecRef() -} - -// accessorContext implements context.Context by extending an existing -// context.Context with an Accessor's values for VFS-relevant state. -type accessorContext struct { - context.Context - a *Accessor -} - -func (a *Accessor) wrapContext(ctx context.Context) *accessorContext { - return &accessorContext{ - Context: ctx, - a: a, - } -} - -// Value implements context.Context.Value. -func (ac *accessorContext) Value(key interface{}) interface{} { - switch key { - case vfs.CtxMountNamespace: - ac.a.mntns.IncRef() - return ac.a.mntns - case vfs.CtxRoot: - ac.a.root.IncRef() - return ac.a.root - default: - return ac.Context.Value(key) - } -} - -func (a *Accessor) pathOperationAt(pathname string) *vfs.PathOperation { - return &vfs.PathOperation{ - Root: a.root, - Start: a.root, - Path: fspath.Parse(pathname), - } -} - -// CreateDeviceFile creates a device special file at the given pathname in the -// devtmpfs instance accessed by the Accessor. -func (a *Accessor) CreateDeviceFile(ctx context.Context, pathname string, kind vfs.DeviceKind, major, minor uint32, perms uint16) error { - mode := (linux.FileMode)(perms) - switch kind { - case vfs.BlockDevice: - mode |= linux.S_IFBLK - case vfs.CharDevice: - mode |= linux.S_IFCHR - default: - panic(fmt.Sprintf("invalid vfs.DeviceKind: %v", kind)) - } - // NOTE: Linux's devtmpfs refuses to automatically delete files it didn't - // create, which it recognizes by storing a pointer to the kdevtmpfs struct - // thread in struct inode::i_private. Accessor doesn't yet support deletion - // of files at all, and probably won't as long as we don't need to support - // kernel modules, so this is moot for now. - return a.vfsObj.MknodAt(a.wrapContext(ctx), a.creds, a.pathOperationAt(pathname), &vfs.MknodOptions{ - Mode: mode, - DevMajor: major, - DevMinor: minor, - }) -} - -// UserspaceInit creates symbolic links and mount points in the devtmpfs -// instance accessed by the Accessor that are created by userspace in Linux. It -// does not create mounts. -func (a *Accessor) UserspaceInit(ctx context.Context) error { - actx := a.wrapContext(ctx) - - // systemd: src/shared/dev-setup.c:dev_setup() - for _, symlink := range []struct { - source string - target string - }{ - // /proc/kcore is not implemented. - {source: "fd", target: "/proc/self/fd"}, - {source: "stdin", target: "/proc/self/fd/0"}, - {source: "stdout", target: "/proc/self/fd/1"}, - {source: "stderr", target: "/proc/self/fd/2"}, - } { - if err := a.vfsObj.SymlinkAt(actx, a.creds, a.pathOperationAt(symlink.source), symlink.target); err != nil { - return fmt.Errorf("failed to create symlink %q => %q: %v", symlink.source, symlink.target, err) - } - } - - // systemd: src/core/mount-setup.c:mount_table - for _, dir := range []string{ - "shm", - "pts", - } { - if err := a.vfsObj.MkdirAt(actx, a.creds, a.pathOperationAt(dir), &vfs.MkdirOptions{ - // systemd: src/core/mount-setup.c:mount_one() - Mode: 0755, - }); err != nil { - return fmt.Errorf("failed to create directory %q: %v", dir, err) - } - } - - return nil -} diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go deleted file mode 100644 index b6d52c015..000000000 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go +++ /dev/null @@ -1,122 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package devtmpfs - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func TestDevtmpfs(t *testing.T) { - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - // Register tmpfs just so that we can have a root filesystem that isn't - // devtmpfs. - vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - vfsObj.MustRegisterFilesystemType("devtmpfs", &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - - // Create a test mount namespace with devtmpfs mounted at "/dev". - const devPath = "/dev" - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "tmpfs" /* source */, "tmpfs" /* fsTypeName */, &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("failed to create tmpfs root mount: %v", err) - } - defer mntns.DecRef() - root := mntns.Root() - defer root.DecRef() - devpop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(devPath), - } - if err := vfsObj.MkdirAt(ctx, creds, &devpop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - t.Fatalf("failed to create mount point: %v", err) - } - if err := vfsObj.MountAt(ctx, creds, "devtmpfs" /* source */, &devpop, "devtmpfs" /* fsTypeName */, &vfs.MountOptions{}); err != nil { - t.Fatalf("failed to mount devtmpfs: %v", err) - } - - a, err := NewAccessor(ctx, vfsObj, creds, "devtmpfs") - if err != nil { - t.Fatalf("failed to create devtmpfs.Accessor: %v", err) - } - defer a.Release() - - // Create "userspace-initialized" files using a devtmpfs.Accessor. - if err := a.UserspaceInit(ctx); err != nil { - t.Fatalf("failed to userspace-initialize devtmpfs: %v", err) - } - // Created files should be visible in the test mount namespace. - abspath := devPath + "/fd" - target, err := vfsObj.ReadlinkAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(abspath), - }) - if want := "/proc/self/fd"; err != nil || target != want { - t.Fatalf("readlink(%q): got (%q, %v), wanted (%q, nil)", abspath, target, err, want) - } - - // Create a dummy device special file using a devtmpfs.Accessor. - const ( - pathInDev = "dummy" - kind = vfs.CharDevice - major = 12 - minor = 34 - perms = 0600 - wantMode = linux.S_IFCHR | perms - ) - if err := a.CreateDeviceFile(ctx, pathInDev, kind, major, minor, perms); err != nil { - t.Fatalf("failed to create device file: %v", err) - } - // The device special file should be visible in the test mount namespace. - abspath = devPath + "/" + pathInDev - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(abspath), - }, &vfs.StatOptions{ - Mask: linux.STATX_TYPE | linux.STATX_MODE, - }) - if err != nil { - t.Fatalf("failed to stat device file at %q: %v", abspath, err) - } - if stat.Mode != wantMode { - t.Errorf("device file mode: got %v, wanted %v", stat.Mode, wantMode) - } - if stat.RdevMajor != major { - t.Errorf("major device number: got %v, wanted %v", stat.RdevMajor, major) - } - if stat.RdevMinor != minor { - t.Errorf("minor device number: got %v, wanted %v", stat.RdevMinor, minor) - } -} diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD deleted file mode 100644 index 6f78f478f..000000000 --- a/pkg/sentry/fsimpl/ext/BUILD +++ /dev/null @@ -1,88 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "dirent_list", - out = "dirent_list.go", - package = "ext", - prefix = "dirent", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*dirent", - "Linker": "*dirent", - }, -) - -go_library( - name = "ext", - srcs = [ - "block_map_file.go", - "dentry.go", - "directory.go", - "dirent_list.go", - "ext.go", - "extent_file.go", - "file_description.go", - "filesystem.go", - "inode.go", - "regular_file.go", - "symlink.go", - "utils.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/fd", - "//pkg/fspath", - "//pkg/log", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fs", - "//pkg/sentry/fsimpl/ext/disklayout", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/memmap", - "//pkg/sentry/syscalls/linux", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "ext_test", - size = "small", - srcs = [ - "block_map_test.go", - "ext_test.go", - "extent_test.go", - ], - data = [ - "//pkg/sentry/fsimpl/ext:assets/bigfile.txt", - "//pkg/sentry/fsimpl/ext:assets/file.txt", - "//pkg/sentry/fsimpl/ext:assets/tiny.ext2", - "//pkg/sentry/fsimpl/ext:assets/tiny.ext3", - "//pkg/sentry/fsimpl/ext:assets/tiny.ext4", - ], - library = ":ext", - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/ext/disklayout", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - "//runsc/testutil", - "@com_github_google_go-cmp//cmp:go_default_library", - "@com_github_google_go-cmp//cmp/cmpopts:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/ext/README.md b/pkg/sentry/fsimpl/ext/README.md deleted file mode 100644 index af00cfda8..000000000 --- a/pkg/sentry/fsimpl/ext/README.md +++ /dev/null @@ -1,117 +0,0 @@ -## EXT(2/3/4) File System - -This is a filesystem driver which supports ext2, ext3 and ext4 filesystems. -Linux has specialized drivers for each variant but none which supports all. This -library takes advantage of ext's backward compatibility and understands the -internal organization of on-disk structures to support all variants. - -This driver implementation diverges from the Linux implementations in being more -forgiving about versioning. For instance, if a filesystem contains both extent -based inodes and classical block map based inodes, this driver will not complain -and interpret them both correctly. While in Linux this would be an issue. This -blurs the line between the three ext fs variants. - -Ext2 is considered deprecated as of Red Hat Enterprise Linux 7, and ext3 has -been superseded by ext4 by large performance gains. Thus it is recommended to -upgrade older filesystem images to ext4 using e2fsprogs for better performance. - -### Read Only - -This driver currently only allows read only operations. A lot of the design -decisions are based on this feature. There are plans to implement write (the -process for which is documented in the future work section). - -### Performance - -One of the biggest wins about this driver is that it directly talks to the -underlying block device (or whatever persistent storage is being used), instead -of making expensive RPCs to a gofer. - -Another advantage is that ext fs supports fast concurrent reads. Currently the -device is represented using a `io.ReaderAt` which allows for concurrent reads. -All reads are directly passed to the device driver which intelligently serves -the read requests in the optimal order. There is no congestion due to locking -while reading in the filesystem level. - -Reads are optimized further in the way file data is transferred over to user -memory. Ext fs directly copies over file data from disk into user memory with no -additional allocations on the way. We can only get faster by preloading file -data into memory (see future work section). - -The internal structures used to represent files, inodes and file descriptors use -a lot of inheritance. With the level of indirection that an interface adds with -an internal pointer, it can quickly fragment a structure across memory. As this -runs along side a full blown kernel (which is memory intensive), having a -fragmented struct might hurt performance. Hence these internal structures, -though interfaced, are tightly packed in memory using the same inheritance -pattern that pkg/sentry/vfs uses. The pkg/sentry/fsimpl/ext/disklayout package -makes an execption to this pattern for reasons documented in the package. - -### Security - -This driver also intends to help sandbox the container better by reducing the -surface of the host kernel that the application touches. It prevents the -application from exploiting vulnerabilities in the host filesystem driver. All -`io.ReaderAt.ReadAt()` calls are translated to `pread(2)` which are directly -passed to the device driver in the kernel. Hence this reduces the surface for -attack. - -The application can not affect any host filesystems other than the one passed -via block device by the user. - -### Future Work - -#### Write - -To support write operations we would need to modify the block device underneath. -Currently, the driver does not modify the device at all, not even for updating -the access times for reads. Modifying the filesystem incorrectly can corrupt it -and render it unreadable for other correct ext(x) drivers. Hence caution must be -maintained while modifying metadata structures. - -Ext4 specifically is built for performance and has added a lot of complexity as -to how metadata structures are modified. For instance, files that are organized -via an extent tree which must be balanced and file data blocks must be placed in -the same extent as much as possible to increase locality. Such properties must -be maintained while modifying the tree. - -Ext filesystems boast a lot about locality, which plays a big role in them being -performant. The block allocation algorithm in Linux does a good job in keeping -related data together. This behavior must be maintained as much as possible, -else we might end up degrading the filesystem performance over time. - -Ext4 also supports a wide variety of features which are specialized for varying -use cases. Implementing all of them can get difficult very quickly. - -Ext(x) checksums all its metadata structures to check for corruption, so -modification of any metadata struct must correspond with re-checksumming the -struct. Linux filesystem drivers also order on-disk updates intelligently to not -corrupt the filesystem and also remain performant. The in-memory metadata -structures must be kept in sync with what is on disk. - -There is also replication of some important structures across the filesystem. -All replicas must be updated when their original copy is updated. There is also -provisioning for snapshotting which must be kept in mind, although it should not -affect this implementation unless we allow users to create filesystem snapshots. - -Ext4 also introduced journaling (jbd2). The journal must be updated -appropriately. - -#### Performance - -To improve performance we should implement a buffer cache, and optionally, read -ahead for small files. While doing so we must also keep in mind the memory usage -and have a reasonable cap on how much file data we want to hold in memory. - -#### Features - -Our current implementation will work with most ext4 filesystems for readonly -purposed. However, the following features are not supported yet: - -- Journal -- Snapshotting -- Extended Attributes -- Hash Tree Directories -- Meta Block Groups -- Multiple Mount Protection -- Bigalloc diff --git a/pkg/sentry/fsimpl/ext/assets/README.md b/pkg/sentry/fsimpl/ext/assets/README.md deleted file mode 100644 index 6f1e81b3a..000000000 --- a/pkg/sentry/fsimpl/ext/assets/README.md +++ /dev/null @@ -1,36 +0,0 @@ -### Tiny Ext(2/3/4) Images - -The images are of size 64Kb which supports 64 1k blocks and 16 inodes. This is -the smallest size mkfs.ext(2/3/4) works with. - -These images were generated using the following commands. - -```bash -fallocate -l 64K tiny.ext$VERSION -mkfs.ext$VERSION -j tiny.ext$VERSION -``` - -where `VERSION` is `2`, `3` or `4`. - -You can mount it using: - -```bash -sudo mount -o loop tiny.ext$VERSION $MOUNTPOINT -``` - -`file.txt`, `bigfile.txt` and `symlink.txt` were added to this image by just -mounting it and copying (while preserving links) those files to the mountpoint -directory using: - -```bash -sudo cp -P {file.txt,symlink.txt,bigfile.txt} $MOUNTPOINT -``` - -The files in this directory mirror the contents and organisation of the files -stored in the image. - -You can umount the filesystem using: - -```bash -sudo umount $MOUNTPOINT -``` diff --git a/pkg/sentry/fsimpl/ext/assets/bigfile.txt b/pkg/sentry/fsimpl/ext/assets/bigfile.txt deleted file mode 100644 index 3857cf516..000000000 --- a/pkg/sentry/fsimpl/ext/assets/bigfile.txt +++ /dev/null @@ -1,41 +0,0 @@ -Lorem ipsum dolor sit amet, consectetur adipiscing elit. Phasellus faucibus eleifend orci, ut ornare nibh faucibus eu. Cras at condimentum massa. Nullam luctus, elit non porttitor congue, sapien diam feugiat sapien, sed eleifend nulla mauris non arcu. Sed lacinia mauris magna, eu mollis libero varius sit amet. Donec mollis, quam convallis commodo posuere, dolor nisi placerat nisi, in faucibus augue mi eu lorem. In pharetra consectetur faucibus. Ut euismod ex efficitur egestas tincidunt. Maecenas condimentum ut ante in rutrum. Vivamus sed arcu tempor, faucibus turpis et, lacinia diam. - -Sed in lacus vel nisl interdum bibendum in sed justo. Nunc tellus risus, molestie vitae arcu sed, molestie tempus ligula. Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Nunc risus neque, volutpat et ante non, ullamcorper condimentum ante. Aliquam sed metus in urna condimentum convallis. Vivamus ut libero mauris. Proin mollis posuere consequat. Vestibulum placerat mollis est et pulvinar. - -Donec rutrum odio ac diam pharetra, id fermentum magna cursus. Pellentesque in dapibus elit, et condimentum orci. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. Suspendisse euismod dapibus est, id vestibulum mauris. Nulla facilisi. Nulla cursus gravida nisi. Phasellus vestibulum rutrum lectus, a dignissim mauris hendrerit vitae. In at elementum mauris. Integer vel efficitur velit. Nullam fringilla sapien mi, quis luctus neque efficitur ac. Aenean nec quam dapibus nunc commodo pharetra. Proin sapien mi, fermentum aliquet vulputate non, aliquet porttitor diam. Quisque lacinia, urna et finibus fermentum, nunc lacus vehicula ex, sed congue metus lectus ac quam. Aliquam erat volutpat. Suspendisse sodales, dolor ut tincidunt finibus, augue erat varius tellus, a interdum erat sem at nunc. Vestibulum cursus iaculis sapien, vitae feugiat dui auctor quis. - -Pellentesque nec maximus nulla, eu blandit diam. Maecenas quis arcu ornare, congue ante at, vehicula ipsum. Praesent feugiat mauris rutrum sem fermentum, nec luctus ipsum placerat. Pellentesque placerat ipsum at dignissim fringilla. Vivamus et posuere sem, eget hendrerit felis. Aenean vulputate, augue vel mollis feugiat, justo ipsum mollis dolor, eu mollis elit neque ut ipsum. Orci varius natoque penatibus et magnis dis parturient montes, nascetur ridiculus mus. Fusce bibendum sem quam, vulputate laoreet mi dapibus imperdiet. Sed a purus non nibh pretium aliquet. Integer eget luctus augue, vitae tincidunt magna. Ut eros enim, egestas eu nulla et, lobortis egestas arcu. Cras id ipsum ac justo lacinia rutrum. Vivamus lectus leo, ultricies sed justo at, pellentesque feugiat magna. Ut sollicitudin neque elit, vel ornare mauris commodo id. - -Duis dapibus orci et sapien finibus finibus. Mauris eleifend, lacus at vestibulum maximus, quam ligula pharetra erat, sit amet dapibus neque elit vitae neque. In bibendum sollicitudin erat, eget ultricies tortor malesuada at. Sed sit amet orci turpis. Donec feugiat ligula nibh, molestie tincidunt lectus elementum id. Donec volutpat maximus nibh, in vulputate felis posuere eu. Cras tincidunt ullamcorper lacus. Phasellus porta lorem auctor, congue magna a, commodo elit. - -Etiam auctor mi quis elit sodales, eu pulvinar arcu condimentum. Aenean imperdiet risus et dapibus tincidunt. Nullam tincidunt dictum dui, sed commodo urna rutrum id. Ut mollis libero vel elit laoreet bibendum. Quisque arcu arcu, tincidunt at ultricies id, vulputate nec metus. In tristique posuere quam sit amet volutpat. Vivamus scelerisque et nunc at dapibus. Fusce finibus libero ut ligula pretium rhoncus. Mauris non elit in arcu finibus imperdiet. Pellentesque nec massa odio. Proin rutrum mauris non sagittis efficitur. Aliquam auctor quam at dignissim faucibus. Ut eget ligula in magna posuere ultricies vitae sit amet turpis. Duis maximus odio nulla. Donec gravida sem tristique tempus scelerisque. - -Interdum et malesuada fames ac ante ipsum primis in faucibus. Fusce pharetra magna vulputate aliquet tempus. Duis id hendrerit arcu. Quisque ut ex elit. Integer velit orci, venenatis ut sapien ac, placerat porttitor dui. Interdum et malesuada fames ac ante ipsum primis in faucibus. Nunc hendrerit cursus diam, hendrerit finibus ipsum scelerisque ut. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. - -Nulla non euismod neque. Phasellus vel sapien eu metus pulvinar rhoncus. Suspendisse eu mollis tellus, quis vestibulum tortor. Maecenas interdum dolor sed nulla fermentum maximus. Donec imperdiet ullamcorper condimentum. Nam quis nibh ante. Praesent quis tellus ut tortor pulvinar blandit sit amet ut sapien. Vestibulum est orci, pellentesque vitae tristique sit amet, tristique non felis. - -Vivamus sodales pellentesque varius. Sed vel tempus ligula. Nulla tristique nisl vel dui facilisis, ac sodales augue hendrerit. Proin augue nisi, vestibulum quis augue nec, sagittis tincidunt velit. Vestibulum euismod, nulla nec sodales faucibus, urna sapien vulputate magna, id varius metus sapien ut neque. Duis in mollis urna, in scelerisque enim. Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Nunc condimentum dictum turpis, et egestas neque dapibus eget. Quisque fringilla, dui eu venenatis eleifend, erat nibh lacinia urna, at lacinia lacus sapien eu dui. Duis eu erat ut mi lacinia convallis a sed ex. - -Fusce elit metus, tincidunt nec eleifend a, hendrerit nec ligula. Duis placerat finibus sollicitudin. In euismod porta tellus, in luctus justo bibendum bibendum. Maecenas at magna eleifend lectus tincidunt suscipit ut a ligula. Nulla tempor accumsan felis, fermentum dapibus est eleifend vitae. Mauris urna sem, fringilla at ultricies non, ultrices in arcu. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. Nam vehicula nunc at laoreet imperdiet. Nunc tristique ut risus id aliquet. Integer eleifend massa orci. - -Vestibulum sed ante sollicitudin nisi fringilla bibendum nec vel quam. Sed pretium augue eu ligula congue pulvinar. Donec vitae magna tincidunt, pharetra lacus id, convallis nulla. Cras viverra nisl nisl, varius convallis leo vulputate nec. Morbi at consequat dui, sed aliquet metus. Sed suscipit fermentum mollis. Maecenas nec mi sodales, tincidunt purus in, tristique mauris. Orci varius natoque penatibus et magnis dis parturient montes, nascetur ridiculus mus. Donec interdum mi in velit efficitur, quis ultrices ex imperdiet. Sed vestibulum, magna ut tristique pretium, mi ipsum placerat tellus, non tempor enim augue et ex. Pellentesque eget felis quis ante sodales viverra ac sed lacus. Donec suscipit tempus massa, eget laoreet massa molestie at. - -Aenean fringilla dui non aliquet consectetur. Fusce cursus quam nec orci hendrerit faucibus. Donec consequat suscipit enim, non volutpat lectus auctor interdum. Proin lorem purus, maximus vel orci vitae, suscipit egestas turpis. Donec risus urna, congue a sem eu, aliquet placerat odio. Morbi gravida tristique turpis, quis efficitur enim. Nunc interdum gravida ipsum vel facilisis. Nunc congue finibus sollicitudin. Quisque euismod aliquet lectus et tincidunt. Curabitur ultrices sem ut mi fringilla fermentum. Morbi pretium, nisi sit amet dapibus congue, dolor enim consectetur risus, a interdum ligula odio sed odio. Quisque facilisis, mi at suscipit gravida, nunc sapien cursus justo, ut luctus odio nulla quis leo. Integer condimentum lobortis mauris, non egestas tellus lobortis sit amet. - -In sollicitudin velit ac ante vehicula, vitae varius tortor mollis. In hac habitasse platea dictumst. Quisque et orci lorem. Integer malesuada fringilla luctus. Pellentesque malesuada, mi non lobortis porttitor, ante ligula vulputate ante, nec dictum risus eros sit amet sapien. Nulla aliquam lorem libero, ac varius nulla tristique eget. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. Ut pellentesque mauris orci, vel consequat mi varius a. Ut sit amet elit vulputate, lacinia metus non, fermentum nisl. Pellentesque eu nisi sed quam egestas blandit. Duis sit amet lobortis dolor. Donec consectetur sem interdum, tristique elit sit amet, sodales lacus. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. Fusce id aliquam augue. Sed pretium congue risus vitae lacinia. Vestibulum non vulputate risus, ut malesuada justo. - -Sed odio elit, consectetur ac mauris quis, consequat commodo libero. Fusce sodales velit vulputate pulvinar fermentum. Donec iaculis nec nisl eget faucibus. Mauris at dictum velit. Donec fermentum lectus eu viverra volutpat. Aliquam consequat facilisis lorem, cursus consequat dui bibendum ullamcorper. Pellentesque nulla magna, imperdiet at magna et, cursus egestas enim. Nullam semper molestie lectus sit amet semper. Duis eget tincidunt est. Integer id neque risus. Integer ultricies hendrerit vestibulum. Donec blandit blandit sagittis. Nunc consectetur vitae nisi consectetur volutpat. - -Nulla id lorem fermentum, efficitur magna a, hendrerit dui. Vivamus sagittis orci gravida, bibendum quam eget, molestie est. Phasellus nec enim tincidunt, volutpat sapien non, laoreet diam. Nulla posuere enim nec porttitor lobortis. Donec auctor odio ut orci eleifend, ut eleifend purus convallis. Interdum et malesuada fames ac ante ipsum primis in faucibus. Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut hendrerit, purus eget viverra tincidunt, sem magna imperdiet libero, et aliquam turpis neque vitae elit. Maecenas semper varius iaculis. Cras non lorem quis quam bibendum eleifend in et libero. Curabitur at purus mauris. Lorem ipsum dolor sit amet, consectetur adipiscing elit. Vivamus porta diam sed elit eleifend gravida. - -Nulla facilisi. Ut ultricies diam vel diam consectetur, vel porta augue molestie. Fusce interdum sapien et metus facilisis pellentesque. Nulla convallis sem at nunc vehicula facilisis. Nam ac rutrum purus. Nunc bibendum, dolor sit amet tempus ullamcorper, lorem leo tempor sem, id fringilla nunc augue scelerisque augue. Nullam sit amet rutrum nisl. Vestibulum ante ipsum primis in faucibus orci luctus et ultrices posuere cubilia Curae; Donec sed mauris gravida eros vehicula sagittis at eget orci. Cras elementum, eros at accumsan bibendum, libero neque blandit purus, vitae vestibulum libero massa ac nibh. Integer at placerat nulla. Mauris eu eleifend orci. Aliquam consequat ligula vitae erat porta lobortis. Duis fermentum elit ac aliquet ornare. - -Mauris eget cursus tellus, eget sodales purus. Aliquam malesuada, augue id vulputate finibus, nisi ex bibendum nisl, sit amet laoreet quam urna a dolor. Nullam ultricies, sapien eu laoreet consequat, erat eros dignissim diam, ultrices sodales lectus mauris et leo. Morbi lacinia eu ante at tempus. Sed iaculis finibus magna malesuada efficitur. Donec faucibus erat sit amet elementum feugiat. Praesent a placerat nisi. Etiam lacinia gravida diam, et sollicitudin sapien tincidunt ut. - -Maecenas felis quam, tincidunt vitae venenatis scelerisque, viverra vitae odio. Phasellus enim neque, ultricies suscipit malesuada sit amet, vehicula sit amet purus. Nulla placerat sit amet dui vel tincidunt. Nam quis neque vel magna commodo egestas. Vestibulum sagittis rutrum lorem ut congue. Maecenas vel ultrices tellus. Donec efficitur, urna ac consequat iaculis, lorem felis pharetra eros, eget faucibus orci lectus sit amet arcu. - -Ut a tempus nisi. Nulla facilisi. Praesent vulputate maximus mi et dapibus. Sed sit amet libero ac augue hendrerit efficitur in a sapien. Mauris placerat velit sit amet tellus sollicitudin faucibus. Donec egestas a magna ac suscipit. Duis enim sapien, mollis sed egestas et, vestibulum vel leo. - -Proin quis dapibus dui. Donec eu tincidunt nunc. Vivamus eget purus consectetur, maximus ante vitae, tincidunt elit. Aenean mattis dolor a gravida aliquam. Praesent quis tellus id sem maximus vulputate nec sed nulla. Lorem ipsum dolor sit amet, consectetur adipiscing elit. Curabitur metus nulla, volutpat volutpat est eu, hendrerit congue erat. Aliquam sollicitudin augue ante. Sed sollicitudin, magna eu consequat elementum, mi augue ullamcorper felis, molestie imperdiet erat metus iaculis est. Proin ac tortor nisi. Pellentesque quis nisi risus. Integer enim sapien, tincidunt quis tortor id, accumsan venenatis mi. Nulla facilisi. - -Cras pretium sit amet quam congue maximus. Morbi lacus libero, imperdiet commodo massa sed, scelerisque placerat libero. Cras nisl nisi, consectetur sed bibendum eu, venenatis at enim. Proin sodales justo at quam aliquam, a consectetur mi ornare. Donec porta ac est sit amet efficitur. Suspendisse vestibulum tortor id neque imperdiet, id lacinia risus vehicula. Phasellus ac eleifend purus. Mauris vel gravida ante. Aliquam vitae lobortis risus. Sed vehicula consectetur tincidunt. Nam et justo vitae purus molestie consequat. Pellentesque ipsum ex, convallis quis blandit non, gravida et urna. Donec diam ligula amet. diff --git a/pkg/sentry/fsimpl/ext/assets/file.txt b/pkg/sentry/fsimpl/ext/assets/file.txt deleted file mode 100644 index 980a0d5f1..000000000 --- a/pkg/sentry/fsimpl/ext/assets/file.txt +++ /dev/null @@ -1 +0,0 @@ -Hello World! diff --git a/pkg/sentry/fsimpl/ext/assets/symlink.txt b/pkg/sentry/fsimpl/ext/assets/symlink.txt deleted file mode 120000 index 4c330738c..000000000 --- a/pkg/sentry/fsimpl/ext/assets/symlink.txt +++ /dev/null @@ -1 +0,0 @@ -file.txt
\ No newline at end of file diff --git a/pkg/sentry/fsimpl/ext/assets/tiny.ext2 b/pkg/sentry/fsimpl/ext/assets/tiny.ext2 Binary files differdeleted file mode 100644 index 381ade9bf..000000000 --- a/pkg/sentry/fsimpl/ext/assets/tiny.ext2 +++ /dev/null diff --git a/pkg/sentry/fsimpl/ext/assets/tiny.ext3 b/pkg/sentry/fsimpl/ext/assets/tiny.ext3 Binary files differdeleted file mode 100644 index 0e97a324c..000000000 --- a/pkg/sentry/fsimpl/ext/assets/tiny.ext3 +++ /dev/null diff --git a/pkg/sentry/fsimpl/ext/assets/tiny.ext4 b/pkg/sentry/fsimpl/ext/assets/tiny.ext4 Binary files differdeleted file mode 100644 index a6859736d..000000000 --- a/pkg/sentry/fsimpl/ext/assets/tiny.ext4 +++ /dev/null diff --git a/pkg/sentry/fsimpl/ext/benchmark/BUILD b/pkg/sentry/fsimpl/ext/benchmark/BUILD deleted file mode 100644 index 6c5a559fd..000000000 --- a/pkg/sentry/fsimpl/ext/benchmark/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools:defs.bzl", "go_test") - -package(licenses = ["notice"]) - -go_test( - name = "benchmark_test", - size = "small", - srcs = ["benchmark_test.go"], - deps = [ - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/ext", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - ], -) diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go deleted file mode 100644 index 89caee3df..000000000 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ /dev/null @@ -1,206 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// These benchmarks emulate memfs benchmarks. Ext4 images must be created -// before this benchmark is run using the `make_deep_ext4.sh` script at -// /tmp/image-{depth}.ext4 for all the depths tested below. -// -// The benchmark itself cannot run the script because the script requires -// sudo privileges to create the file system images. -package benchmark_test - -import ( - "fmt" - "os" - "runtime" - "strings" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -var depths = []int{1, 2, 3, 8, 64, 100} - -const filename = "file.txt" - -// setUp opens imagePath as an ext Filesystem and returns all necessary -// elements required to run tests. If error is nil, it also returns a tear -// down function which must be called after the test is run for clean up. -func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesystem, *vfs.VirtualDentry, func(), error) { - f, err := os.Open(imagePath) - if err != nil { - return nil, nil, nil, nil, err - } - - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - return nil, nil, nil, nil, err - } - vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, imagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}) - if err != nil { - f.Close() - return nil, nil, nil, nil, err - } - - root := mntns.Root() - - tearDown := func() { - root.DecRef() - - if err := f.Close(); err != nil { - b.Fatalf("tearDown failed: %v", err) - } - } - return ctx, vfsObj, &root, tearDown, nil -} - -// mount mounts extfs at the path operation passed. Returns a tear down -// function which must be called after the test is run for clean up. -func mount(b *testing.B, imagePath string, vfsfs *vfs.VirtualFilesystem, pop *vfs.PathOperation) func() { - b.Helper() - - f, err := os.Open(imagePath) - if err != nil { - b.Fatalf("could not open image at %s: %v", imagePath, err) - } - - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - if err := vfsfs.MountAt(ctx, creds, imagePath, pop, "extfs", &vfs.MountOptions{ - GetFilesystemOptions: vfs.GetFilesystemOptions{ - InternalData: int(f.Fd()), - }, - }); err != nil { - b.Fatalf("failed to mount tmpfs submount: %v", err) - } - return func() { - if err := f.Close(); err != nil { - b.Fatalf("tearDown failed: %v", err) - } - } -} - -// BenchmarkVFS2Ext4fsStat emulates BenchmarkVFS2MemfsStat. -func BenchmarkVFS2Ext4fsStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx, vfsfs, root, tearDown, err := setUp(b, fmt.Sprintf("/tmp/image-%d.ext4", depth)) - if err != nil { - b.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - creds := auth.CredentialsFromContext(ctx) - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - for i := 1; i <= depth; i++ { - filePathBuilder.WriteString(fmt.Sprintf("%d", i)) - filePathBuilder.WriteByte('/') - } - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsfs.StatAt(ctx, creds, &vfs.PathOperation{ - Root: *root, - Start: *root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. - if stat.Size > 0 { - b.Fatalf("got wrong file size (%d)", stat.Size) - } - } - }) - } -} - -// BenchmarkVFS2ExtfsMountStat emulates BenchmarkVFS2MemfsMountStat. -func BenchmarkVFS2ExtfsMountStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - // Create root extfs with depth 1 so we can mount extfs again at /1/. - ctx, vfsfs, root, tearDown, err := setUp(b, fmt.Sprintf("/tmp/image-%d.ext4", 1)) - if err != nil { - b.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - creds := auth.CredentialsFromContext(ctx) - mountPointName := "/1/" - pop := vfs.PathOperation{ - Root: *root, - Start: *root, - Path: fspath.Parse(mountPointName), - } - - // Save the mount point for later use. - mountPoint, err := vfsfs.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to mount point: %v", err) - } - defer mountPoint.DecRef() - - // Create extfs submount. - mountTearDown := mount(b, fmt.Sprintf("/tmp/image-%d.ext4", depth), vfsfs, &pop) - defer mountTearDown() - - var filePathBuilder strings.Builder - filePathBuilder.WriteString(mountPointName) - for i := 1; i <= depth; i++ { - filePathBuilder.WriteString(fmt.Sprintf("%d", i)) - filePathBuilder.WriteByte('/') - } - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsfs.StatAt(ctx, creds, &vfs.PathOperation{ - Root: *root, - Start: *root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. touch(1) always creates files of size 0 (empty). - if stat.Size > 0 { - b.Fatalf("got wrong file size (%d)", stat.Size) - } - } - }) - } -} diff --git a/pkg/sentry/fsimpl/ext/benchmark/make_deep_ext4.sh b/pkg/sentry/fsimpl/ext/benchmark/make_deep_ext4.sh deleted file mode 100755 index d0910da1f..000000000 --- a/pkg/sentry/fsimpl/ext/benchmark/make_deep_ext4.sh +++ /dev/null @@ -1,72 +0,0 @@ -#!/bin/bash - -# Copyright 2019 The gVisor Authors. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -# This script creates an ext4 image with $1 depth of directories and a file in -# the inner most directory. The created file is at path /1/2/.../depth/file.txt. -# The ext4 image is written to $2. The image is temporarily mounted at -# /tmp/mountpoint. This script must be run with sudo privileges. - -# Usage: -# sudo bash make_deep_ext4.sh {depth} {output path} - -# Check positional arguments. -if [ "$#" -ne 2 ]; then - echo "Usage: sudo bash make_deep_ext4.sh {depth} {output path}" - exit 1 -fi - -# Make sure depth is a non-negative number. -if ! [[ "$1" =~ ^[0-9]+$ ]]; then - echo "Depth must be a non-negative number." - exit 1 -fi - -# Create a 1 MB filesystem image at the requested output path. -rm -f $2 -fallocate -l 1M $2 -if [ $? -ne 0 ]; then - echo "fallocate failed" - exit $? -fi - -# Convert that blank into an ext4 image. -mkfs.ext4 -j $2 -if [ $? -ne 0 ]; then - echo "mkfs.ext4 failed" - exit $? -fi - -# Mount the image. -MOUNTPOINT=/tmp/mountpoint -mkdir -p $MOUNTPOINT -mount -o loop $2 $MOUNTPOINT -if [ $? -ne 0 ]; then - echo "mount failed" - exit $? -fi - -# Create nested directories and the file. -if [ "$1" -eq 0 ]; then - FILEPATH=$MOUNTPOINT/file.txt -else - FILEPATH=$MOUNTPOINT/$(seq -s '/' 1 $1)/file.txt -fi -mkdir -p $(dirname $FILEPATH) || exit -touch $FILEPATH - -# Clean up. -umount $MOUNTPOINT -rm -rf $MOUNTPOINT diff --git a/pkg/sentry/fsimpl/ext/block_map_file.go b/pkg/sentry/fsimpl/ext/block_map_file.go deleted file mode 100644 index a2d8c3ad6..000000000 --- a/pkg/sentry/fsimpl/ext/block_map_file.go +++ /dev/null @@ -1,200 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - "math" - - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/syserror" -) - -const ( - // numDirectBlks is the number of direct blocks in ext block map inodes. - numDirectBlks = 12 -) - -// blockMapFile is a type of regular file which uses direct/indirect block -// addressing to store file data. This was deprecated in ext4. -type blockMapFile struct { - regFile regularFile - - // directBlks are the direct blocks numbers. The physical blocks pointed by - // these holds file data. Contains file blocks 0 to 11. - directBlks [numDirectBlks]uint32 - - // indirectBlk is the physical block which contains (blkSize/4) direct block - // numbers (as uint32 integers). - indirectBlk uint32 - - // doubleIndirectBlk is the physical block which contains (blkSize/4) indirect - // block numbers (as uint32 integers). - doubleIndirectBlk uint32 - - // tripleIndirectBlk is the physical block which contains (blkSize/4) doubly - // indirect block numbers (as uint32 integers). - tripleIndirectBlk uint32 - - // coverage at (i)th index indicates the amount of file data a node at - // height (i) covers. Height 0 is the direct block. - coverage [4]uint64 -} - -// Compiles only if blockMapFile implements io.ReaderAt. -var _ io.ReaderAt = (*blockMapFile)(nil) - -// newBlockMapFile is the blockMapFile constructor. It initializes the file to -// physical blocks map with (at most) the first 12 (direct) blocks. -func newBlockMapFile(regFile regularFile) (*blockMapFile, error) { - file := &blockMapFile{regFile: regFile} - file.regFile.impl = file - - for i := uint(0); i < 4; i++ { - file.coverage[i] = getCoverage(regFile.inode.blkSize, i) - } - - blkMap := regFile.inode.diskInode.Data() - binary.Unmarshal(blkMap[:numDirectBlks*4], binary.LittleEndian, &file.directBlks) - binary.Unmarshal(blkMap[numDirectBlks*4:(numDirectBlks+1)*4], binary.LittleEndian, &file.indirectBlk) - binary.Unmarshal(blkMap[(numDirectBlks+1)*4:(numDirectBlks+2)*4], binary.LittleEndian, &file.doubleIndirectBlk) - binary.Unmarshal(blkMap[(numDirectBlks+2)*4:(numDirectBlks+3)*4], binary.LittleEndian, &file.tripleIndirectBlk) - return file, nil -} - -// ReadAt implements io.ReaderAt.ReadAt. -func (f *blockMapFile) ReadAt(dst []byte, off int64) (int, error) { - if len(dst) == 0 { - return 0, nil - } - - if off < 0 { - return 0, syserror.EINVAL - } - - offset := uint64(off) - size := f.regFile.inode.diskInode.Size() - if offset >= size { - return 0, io.EOF - } - - // dirBlksEnd is the file offset until which direct blocks cover file data. - // Direct blocks cover 0 <= file offset < dirBlksEnd. - dirBlksEnd := numDirectBlks * f.coverage[0] - - // indirBlkEnd is the file offset until which the indirect block covers file - // data. The indirect block covers dirBlksEnd <= file offset < indirBlkEnd. - indirBlkEnd := dirBlksEnd + f.coverage[1] - - // doubIndirBlkEnd is the file offset until which the double indirect block - // covers file data. The double indirect block covers the range - // indirBlkEnd <= file offset < doubIndirBlkEnd. - doubIndirBlkEnd := indirBlkEnd + f.coverage[2] - - read := 0 - toRead := len(dst) - if uint64(toRead)+offset > size { - toRead = int(size - offset) - } - for read < toRead { - var err error - var curR int - - // Figure out which block to delegate the read to. - switch { - case offset < dirBlksEnd: - // Direct block. - curR, err = f.read(f.directBlks[offset/f.regFile.inode.blkSize], offset%f.regFile.inode.blkSize, 0, dst[read:]) - case offset < indirBlkEnd: - // Indirect block. - curR, err = f.read(f.indirectBlk, offset-dirBlksEnd, 1, dst[read:]) - case offset < doubIndirBlkEnd: - // Doubly indirect block. - curR, err = f.read(f.doubleIndirectBlk, offset-indirBlkEnd, 2, dst[read:]) - default: - // Triply indirect block. - curR, err = f.read(f.tripleIndirectBlk, offset-doubIndirBlkEnd, 3, dst[read:]) - } - - read += curR - offset += uint64(curR) - if err != nil { - return read, err - } - } - - if read < len(dst) { - return read, io.EOF - } - return read, nil -} - -// read is the recursive step of the ReadAt function. It relies on knowing the -// current node's location on disk (curPhyBlk) and its height in the block map -// tree. A height of 0 shows that the current node is actually holding file -// data. relFileOff tells the offset from which we need to start to reading -// under the current node. It is completely relative to the current node. -func (f *blockMapFile) read(curPhyBlk uint32, relFileOff uint64, height uint, dst []byte) (int, error) { - curPhyBlkOff := int64(curPhyBlk) * int64(f.regFile.inode.blkSize) - if height == 0 { - toRead := int(f.regFile.inode.blkSize - relFileOff) - if len(dst) < toRead { - toRead = len(dst) - } - - n, _ := f.regFile.inode.fs.dev.ReadAt(dst[:toRead], curPhyBlkOff+int64(relFileOff)) - if n < toRead { - return n, syserror.EIO - } - return n, nil - } - - childCov := f.coverage[height-1] - startIdx := relFileOff / childCov - endIdx := f.regFile.inode.blkSize / 4 // This is exclusive. - wantEndIdx := (relFileOff + uint64(len(dst))) / childCov - wantEndIdx++ // Make this exclusive. - if wantEndIdx < endIdx { - endIdx = wantEndIdx - } - - read := 0 - curChildOff := relFileOff % childCov - for i := startIdx; i < endIdx; i++ { - var childPhyBlk uint32 - err := readFromDisk(f.regFile.inode.fs.dev, curPhyBlkOff+int64(i*4), &childPhyBlk) - if err != nil { - return read, err - } - - n, err := f.read(childPhyBlk, curChildOff, height-1, dst[read:]) - read += n - if err != nil { - return read, err - } - - curChildOff = 0 - } - - return read, nil -} - -// getCoverage returns the number of bytes a node at the given height covers. -// Height 0 is the file data block itself. Height 1 is the indirect block. -// -// Formula: blkSize * ((blkSize / 4)^height) -func getCoverage(blkSize uint64, height uint) uint64 { - return blkSize * uint64(math.Pow(float64(blkSize/4), float64(height))) -} diff --git a/pkg/sentry/fsimpl/ext/block_map_test.go b/pkg/sentry/fsimpl/ext/block_map_test.go deleted file mode 100644 index 181727ef7..000000000 --- a/pkg/sentry/fsimpl/ext/block_map_test.go +++ /dev/null @@ -1,159 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "bytes" - "math/rand" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" -) - -// These consts are for mocking the block map tree. -const ( - mockBMBlkSize = uint32(16) - mockBMDiskSize = 2500 -) - -// TestBlockMapReader stress tests block map reader functionality. It performs -// random length reads from all possible positions in the block map structure. -func TestBlockMapReader(t *testing.T) { - mockBMFile, want := blockMapSetUp(t) - n := len(want) - - for from := 0; from < n; from++ { - got := make([]byte, n-from) - - if read, err := mockBMFile.ReadAt(got, int64(from)); err != nil { - t.Fatalf("file read operation from offset %d to %d only read %d bytes: %v", from, n, read, err) - } - - if diff := cmp.Diff(got, want[from:]); diff != "" { - t.Fatalf("file data from offset %d to %d mismatched (-want +got):\n%s", from, n, diff) - } - } -} - -// blkNumGen is a number generator which gives block numbers for building the -// block map file on disk. It gives unique numbers in a random order which -// facilitates in creating an extremely fragmented filesystem. -type blkNumGen struct { - nums []uint32 -} - -// newBlkNumGen is the blkNumGen constructor. -func newBlkNumGen() *blkNumGen { - blkNums := &blkNumGen{} - lim := mockBMDiskSize / mockBMBlkSize - blkNums.nums = make([]uint32, lim) - for i := uint32(0); i < lim; i++ { - blkNums.nums[i] = i - } - - rand.Shuffle(int(lim), func(i, j int) { - blkNums.nums[i], blkNums.nums[j] = blkNums.nums[j], blkNums.nums[i] - }) - return blkNums -} - -// next returns the next random block number. -func (n *blkNumGen) next() uint32 { - ret := n.nums[0] - n.nums = n.nums[1:] - return ret -} - -// blockMapSetUp creates a mock disk and a block map file. It initializes the -// block map file with 12 direct block, 1 indirect block, 1 double indirect -// block and 1 triple indirect block (basically fill it till the rim). It -// initializes the disk to reflect the inode. Also returns the file data that -// the inode covers and that is written to disk. -func blockMapSetUp(t *testing.T) (*blockMapFile, []byte) { - mockDisk := make([]byte, mockBMDiskSize) - regFile := regularFile{ - inode: inode{ - fs: &filesystem{ - dev: bytes.NewReader(mockDisk), - }, - diskInode: &disklayout.InodeNew{ - InodeOld: disklayout.InodeOld{ - SizeLo: getMockBMFileFize(), - }, - }, - blkSize: uint64(mockBMBlkSize), - }, - } - - var fileData []byte - blkNums := newBlkNumGen() - var data []byte - - // Write the direct blocks. - for i := 0; i < numDirectBlks; i++ { - curBlkNum := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, curBlkNum) - fileData = append(fileData, writeFileDataToBlock(mockDisk, curBlkNum, 0, blkNums)...) - } - - // Write to indirect block. - indirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, indirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, indirectBlk, 1, blkNums)...) - - // Write to indirect block. - doublyIndirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, doublyIndirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, doublyIndirectBlk, 2, blkNums)...) - - // Write to indirect block. - triplyIndirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, triplyIndirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, triplyIndirectBlk, 3, blkNums)...) - - copy(regFile.inode.diskInode.Data(), data) - - mockFile, err := newBlockMapFile(regFile) - if err != nil { - t.Fatalf("newBlockMapFile failed: %v", err) - } - return mockFile, fileData -} - -// writeFileDataToBlock writes random bytes to the block on disk. -func writeFileDataToBlock(disk []byte, blkNum uint32, height uint, blkNums *blkNumGen) []byte { - if height == 0 { - start := blkNum * mockBMBlkSize - end := start + mockBMBlkSize - rand.Read(disk[start:end]) - return disk[start:end] - } - - var fileData []byte - for off := blkNum * mockBMBlkSize; off < (blkNum+1)*mockBMBlkSize; off += 4 { - curBlkNum := blkNums.next() - copy(disk[off:off+4], binary.Marshal(nil, binary.LittleEndian, curBlkNum)) - fileData = append(fileData, writeFileDataToBlock(disk, curBlkNum, height-1, blkNums)...) - } - return fileData -} - -// getMockBMFileFize gets the size of the mock block map file which is used for -// testing. -func getMockBMFileFize() uint32 { - return uint32(numDirectBlks*getCoverage(uint64(mockBMBlkSize), 0) + getCoverage(uint64(mockBMBlkSize), 1) + getCoverage(uint64(mockBMBlkSize), 2) + getCoverage(uint64(mockBMBlkSize), 3)) -} diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go deleted file mode 100644 index a080cb189..000000000 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ /dev/null @@ -1,58 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// dentry implements vfs.DentryImpl. -type dentry struct { - vfsd vfs.Dentry - - // inode is the inode represented by this dentry. Multiple Dentries may - // share a single non-directory Inode (with hard links). inode is - // immutable. - inode *inode -} - -// Compiles only if dentry implements vfs.DentryImpl. -var _ vfs.DentryImpl = (*dentry)(nil) - -// newDentry is the dentry constructor. -func newDentry(in *inode) *dentry { - d := &dentry{ - inode: in, - } - d.vfsd.Init(d) - return d -} - -// IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef() { - d.inode.incRef() -} - -// TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef() bool { - return d.inode.tryIncRef() -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - // FIXME(b/134676337): filesystem.mu may not be locked as required by - // inode.decRef(). - d.inode.decRef() -} diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go deleted file mode 100644 index bd6ede995..000000000 --- a/pkg/sentry/fsimpl/ext/directory.go +++ /dev/null @@ -1,307 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" -) - -// directory represents a directory inode. It holds the childList in memory. -type directory struct { - inode inode - - // mu serializes the changes to childList. - // Lock Order (outermost locks must be taken first): - // directory.mu - // filesystem.mu - mu sync.Mutex - - // childList is a list containing (1) child dirents and (2) fake dirents - // (with diskDirent == nil) that represent the iteration position of - // directoryFDs. childList is used to support directoryFD.IterDirents() - // efficiently. childList is protected by mu. - childList direntList - - // childMap maps the child's filename to the dirent structure stored in - // childList. This adds some data replication but helps in faster path - // traversal. For consistency, key == childMap[key].diskDirent.FileName(). - // Immutable. - childMap map[string]*dirent -} - -// newDirectroy is the directory constructor. -func newDirectroy(inode inode, newDirent bool) (*directory, error) { - file := &directory{inode: inode, childMap: make(map[string]*dirent)} - file.inode.impl = file - - // Initialize childList by reading dirents from the underlying file. - if inode.diskInode.Flags().Index { - // TODO(b/134676337): Support hash tree directories. Currently only the '.' - // and '..' entries are read in. - - // Users cannot navigate this hash tree directory yet. - log.Warningf("hash tree directory being used which is unsupported") - return file, nil - } - - // The dirents are organized in a linear array in the file data. - // Extract the file data and decode the dirents. - regFile, err := newRegularFile(inode) - if err != nil { - return nil, err - } - - // buf is used as scratch space for reading in dirents from disk and - // unmarshalling them into dirent structs. - buf := make([]byte, disklayout.DirentSize) - size := inode.diskInode.Size() - for off, inc := uint64(0), uint64(0); off < size; off += inc { - toRead := size - off - if toRead > disklayout.DirentSize { - toRead = disklayout.DirentSize - } - if n, err := regFile.impl.ReadAt(buf[:toRead], int64(off)); uint64(n) < toRead { - return nil, err - } - - var curDirent dirent - if newDirent { - curDirent.diskDirent = &disklayout.DirentNew{} - } else { - curDirent.diskDirent = &disklayout.DirentOld{} - } - binary.Unmarshal(buf, binary.LittleEndian, curDirent.diskDirent) - - if curDirent.diskDirent.Inode() != 0 && len(curDirent.diskDirent.FileName()) != 0 { - // Inode number and name length fields being set to 0 is used to indicate - // an unused dirent. - file.childList.PushBack(&curDirent) - file.childMap[curDirent.diskDirent.FileName()] = &curDirent - } - - // The next dirent is placed exactly after this dirent record on disk. - inc = uint64(curDirent.diskDirent.RecordSize()) - } - - return file, nil -} - -func (i *inode) isDir() bool { - _, ok := i.impl.(*directory) - return ok -} - -// dirent is the directory.childList node. -type dirent struct { - diskDirent disklayout.Dirent - - // direntEntry links dirents into their parent directory.childList. - direntEntry -} - -// directoryFD represents a directory file description. It implements -// vfs.FileDescriptionImpl. -type directoryFD struct { - fileDescription - vfs.DirectoryFileDescriptionDefaultImpl - - // Protected by directory.mu. - iter *dirent - off int64 -} - -// Compiles only if directoryFD implements vfs.FileDescriptionImpl. -var _ vfs.FileDescriptionImpl = (*directoryFD)(nil) - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { - if fd.iter == nil { - return - } - - dir := fd.inode().impl.(*directory) - dir.mu.Lock() - dir.childList.Remove(fd.iter) - dir.mu.Unlock() - fd.iter = nil -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - extfs := fd.filesystem() - dir := fd.inode().impl.(*directory) - - dir.mu.Lock() - defer dir.mu.Unlock() - - // Ensure that fd.iter exists and is not linked into dir.childList. - var child *dirent - if fd.iter == nil { - // Start iteration at the beginning of dir. - child = dir.childList.Front() - fd.iter = &dirent{} - } else { - // Continue iteration from where we left off. - child = fd.iter.Next() - dir.childList.Remove(fd.iter) - } - for ; child != nil; child = child.Next() { - // Skip other directoryFD iterators. - if child.diskDirent != nil { - childType, ok := child.diskDirent.FileType() - if !ok { - // We will need to read the inode off disk. Do not increment - // ref count here because this inode is not being added to the - // dentry tree. - extfs.mu.Lock() - childInode, err := extfs.getOrCreateInodeLocked(child.diskDirent.Inode()) - extfs.mu.Unlock() - if err != nil { - // Usage of the file description after the error is - // undefined. This implementation would continue reading - // from the next dirent. - fd.off++ - dir.childList.InsertAfter(child, fd.iter) - return err - } - childType = fs.ToInodeType(childInode.diskInode.Mode().FileType()) - } - - if err := cb.Handle(vfs.Dirent{ - Name: child.diskDirent.FileName(), - Type: fs.ToDirentType(childType), - Ino: uint64(child.diskDirent.Inode()), - NextOff: fd.off + 1, - }); err != nil { - dir.childList.InsertBefore(child, fd.iter) - return err - } - fd.off++ - } - } - dir.childList.PushBack(fd.iter) - return nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - if whence != linux.SEEK_SET && whence != linux.SEEK_CUR { - return 0, syserror.EINVAL - } - - dir := fd.inode().impl.(*directory) - - dir.mu.Lock() - defer dir.mu.Unlock() - - // Find resulting offset. - if whence == linux.SEEK_CUR { - offset += fd.off - } - - if offset < 0 { - // lseek(2) specifies that EINVAL should be returned if the resulting offset - // is negative. - return 0, syserror.EINVAL - } - - n := int64(len(dir.childMap)) - realWantOff := offset - if realWantOff > n { - realWantOff = n - } - realCurOff := fd.off - if realCurOff > n { - realCurOff = n - } - - // Ensure that fd.iter exists and is linked into dir.childList so we can - // intelligently seek from the optimal position. - if fd.iter == nil { - fd.iter = &dirent{} - dir.childList.PushFront(fd.iter) - } - - // Guess that iterating from the current position is optimal. - child := fd.iter - diff := realWantOff - realCurOff // Shows direction and magnitude of travel. - - // See if starting from the beginning or end is better. - abDiff := diff - if diff < 0 { - abDiff = -diff - } - if abDiff > realWantOff { - // Starting from the beginning is best. - child = dir.childList.Front() - diff = realWantOff - } else if abDiff > (n - realWantOff) { - // Starting from the end is best. - child = dir.childList.Back() - // (n - 1) because the last non-nil dirent represents the (n-1)th offset. - diff = realWantOff - (n - 1) - } - - for child != nil { - // Skip other directoryFD iterators. - if child.diskDirent != nil { - if diff == 0 { - if child != fd.iter { - dir.childList.Remove(fd.iter) - dir.childList.InsertBefore(child, fd.iter) - } - - fd.off = offset - return offset, nil - } - - if diff < 0 { - diff++ - child = child.Prev() - } else { - diff-- - child = child.Next() - } - continue - } - - if diff < 0 { - child = child.Prev() - } else { - child = child.Next() - } - } - - // Reaching here indicates that the offset is beyond the end of the childList. - dir.childList.Remove(fd.iter) - dir.childList.PushBack(fd.iter) - fd.off = offset - return offset, nil -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *directoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - // mmap(2) specifies that EACCESS should be returned for non-regular file fds. - return syserror.EACCES -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/BUILD b/pkg/sentry/fsimpl/ext/disklayout/BUILD deleted file mode 100644 index 9bd9c76c0..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/BUILD +++ /dev/null @@ -1,47 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "disklayout", - srcs = [ - "block_group.go", - "block_group_32.go", - "block_group_64.go", - "dirent.go", - "dirent_new.go", - "dirent_old.go", - "disklayout.go", - "extent.go", - "inode.go", - "inode_new.go", - "inode_old.go", - "superblock.go", - "superblock_32.go", - "superblock_64.go", - "superblock_old.go", - "test_utils.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - ], -) - -go_test( - name = "disklayout_test", - size = "small", - srcs = [ - "block_group_test.go", - "dirent_test.go", - "extent_test.go", - "inode_test.go", - "superblock_test.go", - ], - library = ":disklayout", - deps = ["//pkg/sentry/kernel/time"], -) diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group.go b/pkg/sentry/fsimpl/ext/disklayout/block_group.go deleted file mode 100644 index ad6f4fef8..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/block_group.go +++ /dev/null @@ -1,137 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// BlockGroup represents a Linux ext block group descriptor. An ext file system -// is split into a series of block groups. This provides an access layer to -// information needed to access and use a block group. -// -// Location: -// - The block group descriptor table is always placed in the blocks -// immediately after the block containing the superblock. -// - The 1st block group descriptor in the original table is in the -// (sb.FirstDataBlock() + 1)th block. -// - See SuperBlock docs to see where the block group descriptor table is -// replicated. -// - sb.BgDescSize() must be used as the block group descriptor entry size -// while reading the table from disk. -// -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html#block-group-descriptors. -type BlockGroup interface { - // InodeTable returns the absolute block number of the block containing the - // inode table. This points to an array of Inode structs. Inode tables are - // statically allocated at mkfs time. The superblock records the number of - // inodes per group (length of this table) and the size of each inode struct. - InodeTable() uint64 - - // BlockBitmap returns the absolute block number of the block containing the - // block bitmap. This bitmap tracks the usage of data blocks within this block - // group and has its own checksum. - BlockBitmap() uint64 - - // InodeBitmap returns the absolute block number of the block containing the - // inode bitmap. This bitmap tracks the usage of this group's inode table - // entries and has its own checksum. - InodeBitmap() uint64 - - // ExclusionBitmap returns the absolute block number of the snapshot exclusion - // bitmap. - ExclusionBitmap() uint64 - - // FreeBlocksCount returns the number of free blocks in the group. - FreeBlocksCount() uint32 - - // FreeInodesCount returns the number of free inodes in the group. - FreeInodesCount() uint32 - - // DirectoryCount returns the number of inodes that represent directories - // under this block group. - DirectoryCount() uint32 - - // UnusedInodeCount returns the number of unused inodes beyond the last used - // inode in this group's inode table. As a result, we needn’t scan past the - // (InodesPerGroup - UnusedInodeCount())th entry in the inode table. - UnusedInodeCount() uint32 - - // BlockBitmapChecksum returns the block bitmap checksum. This is calculated - // using crc32c(FS UUID + group number + entire bitmap). - BlockBitmapChecksum() uint32 - - // InodeBitmapChecksum returns the inode bitmap checksum. This is calculated - // using crc32c(FS UUID + group number + entire bitmap). - InodeBitmapChecksum() uint32 - - // Checksum returns this block group's checksum. - // - // If SbMetadataCsum feature is set: - // - checksum is crc32c(FS UUID + group number + group descriptor - // structure) & 0xFFFF. - // - // If SbGdtCsum feature is set: - // - checksum is crc16(FS UUID + group number + group descriptor - // structure). - // - // SbMetadataCsum and SbGdtCsum should not be both set. - // If they are, Linux warns and asks to run fsck. - Checksum() uint16 - - // Flags returns BGFlags which represents the block group flags. - Flags() BGFlags -} - -// These are the different block group flags. -const ( - // BgInodeUninit indicates that inode table and bitmap are not initialized. - BgInodeUninit uint16 = 0x1 - - // BgBlockUninit indicates that block bitmap is not initialized. - BgBlockUninit uint16 = 0x2 - - // BgInodeZeroed indicates that inode table is zeroed. - BgInodeZeroed uint16 = 0x4 -) - -// BGFlags represents all the different combinations of block group flags. -type BGFlags struct { - InodeUninit bool - BlockUninit bool - InodeZeroed bool -} - -// ToInt converts a BGFlags struct back to its 16-bit representation. -func (f BGFlags) ToInt() uint16 { - var res uint16 - - if f.InodeUninit { - res |= BgInodeUninit - } - if f.BlockUninit { - res |= BgBlockUninit - } - if f.InodeZeroed { - res |= BgInodeZeroed - } - - return res -} - -// BGFlagsFromInt converts the 16-bit flag representation to a BGFlags struct. -func BGFlagsFromInt(flags uint16) BGFlags { - return BGFlags{ - InodeUninit: flags&BgInodeUninit > 0, - BlockUninit: flags&BgBlockUninit > 0, - InodeZeroed: flags&BgInodeZeroed > 0, - } -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group_32.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_32.go deleted file mode 100644 index 3e16c76db..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/block_group_32.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// BlockGroup32Bit emulates the first half of struct ext4_group_desc in -// fs/ext4/ext4.h. It is the block group descriptor struct for ext2, ext3 and -// 32-bit ext4 filesystems. It implements BlockGroup interface. -type BlockGroup32Bit struct { - BlockBitmapLo uint32 - InodeBitmapLo uint32 - InodeTableLo uint32 - FreeBlocksCountLo uint16 - FreeInodesCountLo uint16 - UsedDirsCountLo uint16 - FlagsRaw uint16 - ExcludeBitmapLo uint32 - BlockBitmapChecksumLo uint16 - InodeBitmapChecksumLo uint16 - ItableUnusedLo uint16 - ChecksumRaw uint16 -} - -// Compiles only if BlockGroup32Bit implements BlockGroup. -var _ BlockGroup = (*BlockGroup32Bit)(nil) - -// InodeTable implements BlockGroup.InodeTable. -func (bg *BlockGroup32Bit) InodeTable() uint64 { return uint64(bg.InodeTableLo) } - -// BlockBitmap implements BlockGroup.BlockBitmap. -func (bg *BlockGroup32Bit) BlockBitmap() uint64 { return uint64(bg.BlockBitmapLo) } - -// InodeBitmap implements BlockGroup.InodeBitmap. -func (bg *BlockGroup32Bit) InodeBitmap() uint64 { return uint64(bg.InodeBitmapLo) } - -// ExclusionBitmap implements BlockGroup.ExclusionBitmap. -func (bg *BlockGroup32Bit) ExclusionBitmap() uint64 { return uint64(bg.ExcludeBitmapLo) } - -// FreeBlocksCount implements BlockGroup.FreeBlocksCount. -func (bg *BlockGroup32Bit) FreeBlocksCount() uint32 { return uint32(bg.FreeBlocksCountLo) } - -// FreeInodesCount implements BlockGroup.FreeInodesCount. -func (bg *BlockGroup32Bit) FreeInodesCount() uint32 { return uint32(bg.FreeInodesCountLo) } - -// DirectoryCount implements BlockGroup.DirectoryCount. -func (bg *BlockGroup32Bit) DirectoryCount() uint32 { return uint32(bg.UsedDirsCountLo) } - -// UnusedInodeCount implements BlockGroup.UnusedInodeCount. -func (bg *BlockGroup32Bit) UnusedInodeCount() uint32 { return uint32(bg.ItableUnusedLo) } - -// BlockBitmapChecksum implements BlockGroup.BlockBitmapChecksum. -func (bg *BlockGroup32Bit) BlockBitmapChecksum() uint32 { return uint32(bg.BlockBitmapChecksumLo) } - -// InodeBitmapChecksum implements BlockGroup.InodeBitmapChecksum. -func (bg *BlockGroup32Bit) InodeBitmapChecksum() uint32 { return uint32(bg.InodeBitmapChecksumLo) } - -// Checksum implements BlockGroup.Checksum. -func (bg *BlockGroup32Bit) Checksum() uint16 { return bg.ChecksumRaw } - -// Flags implements BlockGroup.Flags. -func (bg *BlockGroup32Bit) Flags() BGFlags { return BGFlagsFromInt(bg.FlagsRaw) } diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group_64.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_64.go deleted file mode 100644 index 9a809197a..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/block_group_64.go +++ /dev/null @@ -1,93 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// BlockGroup64Bit emulates struct ext4_group_desc in fs/ext4/ext4.h. -// It is the block group descriptor struct for 64-bit ext4 filesystems. -// It implements BlockGroup interface. It is an extension of the 32-bit -// version of BlockGroup. -type BlockGroup64Bit struct { - // We embed the 32-bit struct here because 64-bit version is just an extension - // of the 32-bit version. - BlockGroup32Bit - - // 64-bit specific fields. - BlockBitmapHi uint32 - InodeBitmapHi uint32 - InodeTableHi uint32 - FreeBlocksCountHi uint16 - FreeInodesCountHi uint16 - UsedDirsCountHi uint16 - ItableUnusedHi uint16 - ExcludeBitmapHi uint32 - BlockBitmapChecksumHi uint16 - InodeBitmapChecksumHi uint16 - _ uint32 // Padding to 64 bytes. -} - -// Compiles only if BlockGroup64Bit implements BlockGroup. -var _ BlockGroup = (*BlockGroup64Bit)(nil) - -// Methods to override. Checksum() and Flags() are not overridden. - -// InodeTable implements BlockGroup.InodeTable. -func (bg *BlockGroup64Bit) InodeTable() uint64 { - return (uint64(bg.InodeTableHi) << 32) | uint64(bg.InodeTableLo) -} - -// BlockBitmap implements BlockGroup.BlockBitmap. -func (bg *BlockGroup64Bit) BlockBitmap() uint64 { - return (uint64(bg.BlockBitmapHi) << 32) | uint64(bg.BlockBitmapLo) -} - -// InodeBitmap implements BlockGroup.InodeBitmap. -func (bg *BlockGroup64Bit) InodeBitmap() uint64 { - return (uint64(bg.InodeBitmapHi) << 32) | uint64(bg.InodeBitmapLo) -} - -// ExclusionBitmap implements BlockGroup.ExclusionBitmap. -func (bg *BlockGroup64Bit) ExclusionBitmap() uint64 { - return (uint64(bg.ExcludeBitmapHi) << 32) | uint64(bg.ExcludeBitmapLo) -} - -// FreeBlocksCount implements BlockGroup.FreeBlocksCount. -func (bg *BlockGroup64Bit) FreeBlocksCount() uint32 { - return (uint32(bg.FreeBlocksCountHi) << 16) | uint32(bg.FreeBlocksCountLo) -} - -// FreeInodesCount implements BlockGroup.FreeInodesCount. -func (bg *BlockGroup64Bit) FreeInodesCount() uint32 { - return (uint32(bg.FreeInodesCountHi) << 16) | uint32(bg.FreeInodesCountLo) -} - -// DirectoryCount implements BlockGroup.DirectoryCount. -func (bg *BlockGroup64Bit) DirectoryCount() uint32 { - return (uint32(bg.UsedDirsCountHi) << 16) | uint32(bg.UsedDirsCountLo) -} - -// UnusedInodeCount implements BlockGroup.UnusedInodeCount. -func (bg *BlockGroup64Bit) UnusedInodeCount() uint32 { - return (uint32(bg.ItableUnusedHi) << 16) | uint32(bg.ItableUnusedLo) -} - -// BlockBitmapChecksum implements BlockGroup.BlockBitmapChecksum. -func (bg *BlockGroup64Bit) BlockBitmapChecksum() uint32 { - return (uint32(bg.BlockBitmapChecksumHi) << 16) | uint32(bg.BlockBitmapChecksumLo) -} - -// InodeBitmapChecksum implements BlockGroup.InodeBitmapChecksum. -func (bg *BlockGroup64Bit) InodeBitmapChecksum() uint32 { - return (uint32(bg.InodeBitmapChecksumHi) << 16) | uint32(bg.InodeBitmapChecksumLo) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group_test.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_test.go deleted file mode 100644 index 0ef4294c0..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/block_group_test.go +++ /dev/null @@ -1,26 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "testing" -) - -// TestBlockGroupSize tests that the block group descriptor structs are of the -// correct size. -func TestBlockGroupSize(t *testing.T) { - assertSize(t, BlockGroup32Bit{}, 32) - assertSize(t, BlockGroup64Bit{}, 64) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/dirent.go b/pkg/sentry/fsimpl/ext/disklayout/dirent.go deleted file mode 100644 index 417b6cf65..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -const ( - // MaxFileName is the maximum length of an ext fs file's name. - MaxFileName = 255 - - // DirentSize is the size of ext dirent structures. - DirentSize = 263 -) - -var ( - // inodeTypeByFileType maps ext4 file types to vfs inode types. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#ftype. - inodeTypeByFileType = map[uint8]fs.InodeType{ - 0: fs.Anonymous, - 1: fs.RegularFile, - 2: fs.Directory, - 3: fs.CharacterDevice, - 4: fs.BlockDevice, - 5: fs.Pipe, - 6: fs.Socket, - 7: fs.Symlink, - } -) - -// The Dirent interface should be implemented by structs representing ext -// directory entries. These are for the linear classical directories which -// just store a list of dirent structs. A directory is a series of data blocks -// where is each data block contains a linear array of dirents. The last entry -// of the block has a record size that takes it to the end of the block. The -// end of the directory is when you read dirInode.Size() bytes from the blocks. -// -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#linear-classic-directories. -type Dirent interface { - // Inode returns the absolute inode number of the underlying inode. - // Inode number 0 signifies an unused dirent. - Inode() uint32 - - // RecordSize returns the record length of this dirent on disk. The next - // dirent in the dirent list should be read after these many bytes from - // the current dirent. Must be a multiple of 4. - RecordSize() uint16 - - // FileName returns the name of the file. Can be at most 255 is length. - FileName() string - - // FileType returns the inode type of the underlying inode. This is a - // performance hack so that we do not have to read the underlying inode struct - // to know the type of inode. This will only work when the SbDirentFileType - // feature is set. If not, the second returned value will be false indicating - // that user code has to use the inode mode to extract the file type. - FileType() (fs.InodeType, bool) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/dirent_new.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_new.go deleted file mode 100644 index 29ae4a5c2..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent_new.go +++ /dev/null @@ -1,61 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// DirentNew represents the ext4 directory entry struct. This emulates Linux's -// ext4_dir_entry_2 struct. The FileName can not be more than 255 bytes so we -// only need 8 bits to store the NameLength. As a result, NameLength has been -// shortened and the other 8 bits are used to encode the file type. Use the -// FileTypeRaw field only if the SbDirentFileType feature is set. -// -// Note: This struct can be of variable size on disk. The one described below -// is of maximum size and the FileName beyond NameLength bytes might contain -// garbage. -type DirentNew struct { - InodeNumber uint32 - RecordLength uint16 - NameLength uint8 - FileTypeRaw uint8 - FileNameRaw [MaxFileName]byte -} - -// Compiles only if DirentNew implements Dirent. -var _ Dirent = (*DirentNew)(nil) - -// Inode implements Dirent.Inode. -func (d *DirentNew) Inode() uint32 { return d.InodeNumber } - -// RecordSize implements Dirent.RecordSize. -func (d *DirentNew) RecordSize() uint16 { return d.RecordLength } - -// FileName implements Dirent.FileName. -func (d *DirentNew) FileName() string { - return string(d.FileNameRaw[:d.NameLength]) -} - -// FileType implements Dirent.FileType. -func (d *DirentNew) FileType() (fs.InodeType, bool) { - if inodeType, ok := inodeTypeByFileType[d.FileTypeRaw]; ok { - return inodeType, true - } - - panic(fmt.Sprintf("unknown file type %v", d.FileTypeRaw)) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/dirent_old.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_old.go deleted file mode 100644 index 6fff12a6e..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent_old.go +++ /dev/null @@ -1,49 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import "gvisor.dev/gvisor/pkg/sentry/fs" - -// DirentOld represents the old directory entry struct which does not contain -// the file type. This emulates Linux's ext4_dir_entry struct. -// -// Note: This struct can be of variable size on disk. The one described below -// is of maximum size and the FileName beyond NameLength bytes might contain -// garbage. -type DirentOld struct { - InodeNumber uint32 - RecordLength uint16 - NameLength uint16 - FileNameRaw [MaxFileName]byte -} - -// Compiles only if DirentOld implements Dirent. -var _ Dirent = (*DirentOld)(nil) - -// Inode implements Dirent.Inode. -func (d *DirentOld) Inode() uint32 { return d.InodeNumber } - -// RecordSize implements Dirent.RecordSize. -func (d *DirentOld) RecordSize() uint16 { return d.RecordLength } - -// FileName implements Dirent.FileName. -func (d *DirentOld) FileName() string { - return string(d.FileNameRaw[:d.NameLength]) -} - -// FileType implements Dirent.FileType. -func (d *DirentOld) FileType() (fs.InodeType, bool) { - return fs.Anonymous, false -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/dirent_test.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_test.go deleted file mode 100644 index 934919f8a..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent_test.go +++ /dev/null @@ -1,26 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "testing" -) - -// TestDirentSize tests that the dirent structs are of the correct -// size. -func TestDirentSize(t *testing.T) { - assertSize(t, DirentOld{}, uintptr(DirentSize)) - assertSize(t, DirentNew{}, uintptr(DirentSize)) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/disklayout.go b/pkg/sentry/fsimpl/ext/disklayout/disklayout.go deleted file mode 100644 index bdf4e2132..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/disklayout.go +++ /dev/null @@ -1,50 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package disklayout provides Linux ext file system's disk level structures -// which can be directly read into from the underlying device. Structs aim to -// emulate structures `exactly` how they are layed out on disk. -// -// This library aims to be compatible with all ext(2/3/4) systems so it -// provides a generic interface for all major structures and various -// implementations (for different versions). The user code is responsible for -// using appropriate implementations based on the underlying device. -// -// Interfacing all major structures here serves a few purposes: -// - Abstracts away the complexity of the underlying structure from client -// code. The client only has to figure out versioning on set up and then -// can use these as black boxes and pass it higher up the stack. -// - Having pointer receivers forces the user to use pointers to these -// heavy structs. Hence, prevents the client code from unintentionally -// copying these by value while passing the interface around. -// - Version-based implementation selection is resolved on set up hence -// avoiding per call overhead of choosing implementation. -// - All interface methods are pretty light weight (do not take in any -// parameters by design). Passing pointer arguments to interface methods -// can lead to heap allocation as the compiler won't be able to perform -// escape analysis on an unknown implementation at compile time. -// -// Notes: -// - All fields in these structs are exported because binary.Read would -// panic otherwise. -// - All structures on disk are in little-endian order. Only jbd2 (journal) -// structures are in big-endian order. -// - All OS dependent fields in these structures will be interpretted using -// the Linux version of that field. -// - The suffix `Lo` in field names stands for lower bits of that field. -// - The suffix `Hi` in field names stands for upper bits of that field. -// - The suffix `Raw` has been added to indicate that the field is not split -// into Lo and Hi fields and also to resolve name collision with the -// respective interface. -package disklayout diff --git a/pkg/sentry/fsimpl/ext/disklayout/extent.go b/pkg/sentry/fsimpl/ext/disklayout/extent.go deleted file mode 100644 index 4110649ab..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/extent.go +++ /dev/null @@ -1,143 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// Extents were introduced in ext4 and provide huge performance gains in terms -// data locality and reduced metadata block usage. Extents are organized in -// extent trees. The root node is contained in inode.BlocksRaw. -// -// Terminology: -// - Physical Block: -// Filesystem data block which is addressed normally wrt the entire -// filesystem (addressed with 48 bits). -// -// - File Block: -// Data block containing *only* file data and addressed wrt to the file -// with only 32 bits. The (i)th file block contains file data from -// byte (i * sb.BlockSize()) to ((i+1) * sb.BlockSize()). - -const ( - // ExtentHeaderSize is the size of the header of an extent tree node. - ExtentHeaderSize = 12 - - // ExtentEntrySize is the size of an entry in an extent tree node. - // This size is the same for both leaf and internal nodes. - ExtentEntrySize = 12 - - // ExtentMagic is the magic number which must be present in the header. - ExtentMagic = 0xf30a -) - -// ExtentEntryPair couples an in-memory ExtendNode with the ExtentEntry that -// points to it. We want to cache these structs in memory to avoid repeated -// disk reads. -// -// Note: This struct itself does not represent an on-disk struct. -type ExtentEntryPair struct { - // Entry points to the child node on disk. - Entry ExtentEntry - // Node points to child node in memory. Is nil if the current node is a leaf. - Node *ExtentNode -} - -// ExtentNode represents an extent tree node. For internal nodes, all Entries -// will be ExtendIdxs. For leaf nodes, they will all be Extents. -// -// Note: This struct itself does not represent an on-disk struct. -type ExtentNode struct { - Header ExtentHeader - Entries []ExtentEntryPair -} - -// ExtentEntry represents an extent tree node entry. The entry can either be -// an ExtentIdx or Extent itself. This exists to simplify navigation logic. -type ExtentEntry interface { - // FileBlock returns the first file block number covered by this entry. - FileBlock() uint32 - - // PhysicalBlock returns the child physical block that this entry points to. - PhysicalBlock() uint64 -} - -// ExtentHeader emulates the ext4_extent_header struct in ext4. Each extent -// tree node begins with this and is followed by `NumEntries` number of: -// - Extent if `Depth` == 0 -// - ExtentIdx otherwise -type ExtentHeader struct { - // Magic in the extent magic number, must be 0xf30a. - Magic uint16 - - // NumEntries indicates the number of valid entries following the header. - NumEntries uint16 - - // MaxEntries that could follow the header. Used while adding entries. - MaxEntries uint16 - - // Height represents the distance of this node from the farthest leaf. Please - // note that Linux incorrectly calls this `Depth` (which means the distance - // of the node from the root). - Height uint16 - _ uint32 -} - -// ExtentIdx emulates the ext4_extent_idx struct in ext4. Only present in -// internal nodes. Sorted in ascending order based on FirstFileBlock since -// Linux does a binary search on this. This points to a block containing the -// child node. -type ExtentIdx struct { - FirstFileBlock uint32 - ChildBlockLo uint32 - ChildBlockHi uint16 - _ uint16 -} - -// Compiles only if ExtentIdx implements ExtentEntry. -var _ ExtentEntry = (*ExtentIdx)(nil) - -// FileBlock implements ExtentEntry.FileBlock. -func (ei *ExtentIdx) FileBlock() uint32 { - return ei.FirstFileBlock -} - -// PhysicalBlock implements ExtentEntry.PhysicalBlock. It returns the -// physical block number of the child block. -func (ei *ExtentIdx) PhysicalBlock() uint64 { - return (uint64(ei.ChildBlockHi) << 32) | uint64(ei.ChildBlockLo) -} - -// Extent represents the ext4_extent struct in ext4. Only present in leaf -// nodes. Sorted in ascending order based on FirstFileBlock since Linux does a -// binary search on this. This points to an array of data blocks containing the -// file data. It covers `Length` data blocks starting from `StartBlock`. -type Extent struct { - FirstFileBlock uint32 - Length uint16 - StartBlockHi uint16 - StartBlockLo uint32 -} - -// Compiles only if Extent implements ExtentEntry. -var _ ExtentEntry = (*Extent)(nil) - -// FileBlock implements ExtentEntry.FileBlock. -func (e *Extent) FileBlock() uint32 { - return e.FirstFileBlock -} - -// PhysicalBlock implements ExtentEntry.PhysicalBlock. It returns the -// physical block number of the first data block this extent covers. -func (e *Extent) PhysicalBlock() uint64 { - return (uint64(e.StartBlockHi) << 32) | uint64(e.StartBlockLo) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/extent_test.go b/pkg/sentry/fsimpl/ext/disklayout/extent_test.go deleted file mode 100644 index 8762b90db..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/extent_test.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "testing" -) - -// TestExtentSize tests that the extent structs are of the correct -// size. -func TestExtentSize(t *testing.T) { - assertSize(t, ExtentHeader{}, ExtentHeaderSize) - assertSize(t, ExtentIdx{}, ExtentEntrySize) - assertSize(t, Extent{}, ExtentEntrySize) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode.go b/pkg/sentry/fsimpl/ext/disklayout/inode.go deleted file mode 100644 index 88ae913f5..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/inode.go +++ /dev/null @@ -1,274 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" -) - -// Special inodes. See https://www.kernel.org/doc/html/latest/filesystems/ext4/overview.html#special-inodes. -const ( - // RootDirInode is the inode number of the root directory inode. - RootDirInode = 2 -) - -// The Inode interface must be implemented by structs representing ext inodes. -// The inode stores all the metadata pertaining to the file (except for the -// file name which is held by the directory entry). It does NOT expose all -// fields and should be extended if need be. -// -// Some file systems (e.g. FAT) use the directory entry to store all this -// information. Ext file systems do not so that they can support hard links. -// However, ext4 cheats a little bit and duplicates the file type in the -// directory entry for performance gains. -// -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#index-nodes. -type Inode interface { - // Mode returns the linux file mode which is majorly used to extract - // information like: - // - File permissions (read/write/execute by user/group/others). - // - Sticky, set UID and GID bits. - // - File type. - // - // Masks to extract this information are provided in pkg/abi/linux/file.go. - Mode() linux.FileMode - - // UID returns the owner UID. - UID() auth.KUID - - // GID returns the owner GID. - GID() auth.KGID - - // Size returns the size of the file in bytes. - Size() uint64 - - // InodeSize returns the size of this inode struct in bytes. - // In ext2 and ext3, the inode struct and inode disk record size was fixed at - // 128 bytes. Ext4 makes it possible for the inode struct to be bigger. - // However, accessing any field beyond the 128 bytes marker must be verified - // using this method. - InodeSize() uint16 - - // AccessTime returns the last access time. Shows when the file was last read. - // - // If InExtendedAttr is set, then this should NOT be used because the - // underlying field is used to store the extended attribute value checksum. - AccessTime() time.Time - - // ChangeTime returns the last change time. Shows when the file meta data - // (like permissions) was last changed. - // - // If InExtendedAttr is set, then this should NOT be used because the - // underlying field is used to store the lower 32 bits of the attribute - // value’s reference count. - ChangeTime() time.Time - - // ModificationTime returns the last modification time. Shows when the file - // content was last modified. - // - // If InExtendedAttr is set, then this should NOT be used because - // the underlying field contains the number of the inode that owns the - // extended attribute. - ModificationTime() time.Time - - // DeletionTime returns the deletion time. Inodes are marked as deleted by - // writing to the underlying field. FS tools can restore files until they are - // actually overwritten. - DeletionTime() time.Time - - // LinksCount returns the number of hard links to this inode. - // - // Normally there is an upper limit on the number of hard links: - // - ext2/ext3 = 32,000 - // - ext4 = 65,000 - // - // This implies that an ext4 directory cannot have more than 64,998 - // subdirectories because each subdirectory will have a hard link to the - // directory via the `..` entry. The directory has hard link via the `.` entry - // of its own. And finally the inode is initiated with 1 hard link (itself). - // - // The underlying value is reset to 1 if all the following hold: - // - Inode is a directory. - // - SbDirNlink is enabled. - // - Number of hard links is incremented past 64,999. - // Hard link value of 1 for a directory would indicate that the number of hard - // links is unknown because a directory can have minimum 2 hard links (itself - // and `.` entry). - LinksCount() uint16 - - // Flags returns InodeFlags which represents the inode flags. - Flags() InodeFlags - - // Data returns the underlying inode.i_block array as a slice so it's - // modifiable. This field is special and is used to store various kinds of - // things depending on the filesystem version and inode type. The underlying - // field name in Linux is a little misleading. - // - In ext2/ext3, it contains the block map. - // - In ext4, it contains the extent tree root node. - // - For inline files, it contains the file contents. - // - For symlinks, it contains the link path (if it fits here). - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#the-contents-of-inode-i-block. - Data() []byte -} - -// Inode flags. This is not comprehensive and flags which were not used in -// the Linux kernel have been excluded. -const ( - // InSync indicates that all writes to the file must be synchronous. - InSync = 0x8 - - // InImmutable indicates that this file is immutable. - InImmutable = 0x10 - - // InAppend indicates that this file can only be appended to. - InAppend = 0x20 - - // InNoDump indicates that teh dump(1) utility should not dump this file. - InNoDump = 0x40 - - // InNoAccessTime indicates that the access time of this inode must not be - // updated. - InNoAccessTime = 0x80 - - // InIndex indicates that this directory has hashed indexes. - InIndex = 0x1000 - - // InJournalData indicates that file data must always be written through a - // journal device. - InJournalData = 0x4000 - - // InDirSync indicates that all the directory entiry data must be written - // synchronously. - InDirSync = 0x10000 - - // InTopDir indicates that this inode is at the top of the directory hierarchy. - InTopDir = 0x20000 - - // InHugeFile indicates that this is a huge file. - InHugeFile = 0x40000 - - // InExtents indicates that this inode uses extents. - InExtents = 0x80000 - - // InExtendedAttr indicates that this inode stores a large extended attribute - // value in its data blocks. - InExtendedAttr = 0x200000 - - // InInline indicates that this inode has inline data. - InInline = 0x10000000 - - // InReserved indicates that this inode is reserved for the ext4 library. - InReserved = 0x80000000 -) - -// InodeFlags represents all possible combinations of inode flags. It aims to -// cover the bit masks and provide a more user-friendly interface. -type InodeFlags struct { - Sync bool - Immutable bool - Append bool - NoDump bool - NoAccessTime bool - Index bool - JournalData bool - DirSync bool - TopDir bool - HugeFile bool - Extents bool - ExtendedAttr bool - Inline bool - Reserved bool -} - -// ToInt converts inode flags back to its 32-bit rep. -func (f InodeFlags) ToInt() uint32 { - var res uint32 - - if f.Sync { - res |= InSync - } - if f.Immutable { - res |= InImmutable - } - if f.Append { - res |= InAppend - } - if f.NoDump { - res |= InNoDump - } - if f.NoAccessTime { - res |= InNoAccessTime - } - if f.Index { - res |= InIndex - } - if f.JournalData { - res |= InJournalData - } - if f.DirSync { - res |= InDirSync - } - if f.TopDir { - res |= InTopDir - } - if f.HugeFile { - res |= InHugeFile - } - if f.Extents { - res |= InExtents - } - if f.ExtendedAttr { - res |= InExtendedAttr - } - if f.Inline { - res |= InInline - } - if f.Reserved { - res |= InReserved - } - - return res -} - -// InodeFlagsFromInt converts the integer representation of inode flags to -// a InodeFlags struct. -func InodeFlagsFromInt(f uint32) InodeFlags { - return InodeFlags{ - Sync: f&InSync > 0, - Immutable: f&InImmutable > 0, - Append: f&InAppend > 0, - NoDump: f&InNoDump > 0, - NoAccessTime: f&InNoAccessTime > 0, - Index: f&InIndex > 0, - JournalData: f&InJournalData > 0, - DirSync: f&InDirSync > 0, - TopDir: f&InTopDir > 0, - HugeFile: f&InHugeFile > 0, - Extents: f&InExtents > 0, - ExtendedAttr: f&InExtendedAttr > 0, - Inline: f&InInline > 0, - Reserved: f&InReserved > 0, - } -} - -// These masks define how users can view/modify inode flags. The rest of the -// flags are for internal kernel usage only. -const ( - InUserReadFlagMask = 0x4BDFFF - InUserWriteFlagMask = 0x4B80FF -) diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode_new.go b/pkg/sentry/fsimpl/ext/disklayout/inode_new.go deleted file mode 100644 index 8f9f574ce..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/inode_new.go +++ /dev/null @@ -1,96 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import "gvisor.dev/gvisor/pkg/sentry/kernel/time" - -// InodeNew represents ext4 inode structure which can be bigger than -// OldInodeSize. The actual size of this struct should be determined using -// inode.ExtraInodeSize. Accessing any field here should be verified with the -// actual size. The extra space between the end of the inode struct and end of -// the inode record can be used to store extended attr. -// -// If the TimeExtra fields are in scope, the lower 2 bits of those are used -// to extend their counter part to be 34 bits wide; the rest (upper) 30 bits -// are used to provide nanoscond precision. Hence, these timestamps will now -// overflow in May 2446. -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#inode-timestamps. -type InodeNew struct { - InodeOld - - ExtraInodeSize uint16 - ChecksumHi uint16 - ChangeTimeExtra uint32 - ModificationTimeExtra uint32 - AccessTimeExtra uint32 - CreationTime uint32 - CreationTimeExtra uint32 - VersionHi uint32 - ProjectID uint32 -} - -// Compiles only if InodeNew implements Inode. -var _ Inode = (*InodeNew)(nil) - -// fromExtraTime decodes the extra time and constructs the kernel time struct -// with nanosecond precision. -func fromExtraTime(lo int32, extra uint32) time.Time { - // See description above InodeNew for format. - seconds := (int64(extra&0x3) << 32) + int64(lo) - nanoseconds := int64(extra >> 2) - return time.FromUnix(seconds, nanoseconds) -} - -// Only override methods which change due to ext4 specific fields. - -// Size implements Inode.Size. -func (in *InodeNew) Size() uint64 { - return (uint64(in.SizeHi) << 32) | uint64(in.SizeLo) -} - -// InodeSize implements Inode.InodeSize. -func (in *InodeNew) InodeSize() uint16 { - return OldInodeSize + in.ExtraInodeSize -} - -// ChangeTime implements Inode.ChangeTime. -func (in *InodeNew) ChangeTime() time.Time { - // Apply new timestamp logic if inode.ChangeTimeExtra is in scope. - if in.ExtraInodeSize >= 8 { - return fromExtraTime(in.ChangeTimeRaw, in.ChangeTimeExtra) - } - - return in.InodeOld.ChangeTime() -} - -// ModificationTime implements Inode.ModificationTime. -func (in *InodeNew) ModificationTime() time.Time { - // Apply new timestamp logic if inode.ModificationTimeExtra is in scope. - if in.ExtraInodeSize >= 12 { - return fromExtraTime(in.ModificationTimeRaw, in.ModificationTimeExtra) - } - - return in.InodeOld.ModificationTime() -} - -// AccessTime implements Inode.AccessTime. -func (in *InodeNew) AccessTime() time.Time { - // Apply new timestamp logic if inode.AccessTimeExtra is in scope. - if in.ExtraInodeSize >= 16 { - return fromExtraTime(in.AccessTimeRaw, in.AccessTimeExtra) - } - - return in.InodeOld.AccessTime() -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode_old.go b/pkg/sentry/fsimpl/ext/disklayout/inode_old.go deleted file mode 100644 index db25b11b6..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/inode_old.go +++ /dev/null @@ -1,117 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" -) - -const ( - // OldInodeSize is the inode size in ext2/ext3. - OldInodeSize = 128 -) - -// InodeOld implements Inode interface. It emulates ext2/ext3 inode struct. -// Inode struct size and record size are both 128 bytes for this. -// -// All fields representing time are in seconds since the epoch. Which means that -// they will overflow in January 2038. -type InodeOld struct { - ModeRaw uint16 - UIDLo uint16 - SizeLo uint32 - - // The time fields are signed integers because they could be negative to - // represent time before the epoch. - AccessTimeRaw int32 - ChangeTimeRaw int32 - ModificationTimeRaw int32 - DeletionTimeRaw int32 - - GIDLo uint16 - LinksCountRaw uint16 - BlocksCountLo uint32 - FlagsRaw uint32 - VersionLo uint32 // This is OS dependent. - DataRaw [60]byte - Generation uint32 - FileACLLo uint32 - SizeHi uint32 - ObsoFaddr uint32 - - // OS dependent fields have been inlined here. - BlocksCountHi uint16 - FileACLHi uint16 - UIDHi uint16 - GIDHi uint16 - ChecksumLo uint16 - _ uint16 -} - -// Compiles only if InodeOld implements Inode. -var _ Inode = (*InodeOld)(nil) - -// Mode implements Inode.Mode. -func (in *InodeOld) Mode() linux.FileMode { return linux.FileMode(in.ModeRaw) } - -// UID implements Inode.UID. -func (in *InodeOld) UID() auth.KUID { - return auth.KUID((uint32(in.UIDHi) << 16) | uint32(in.UIDLo)) -} - -// GID implements Inode.GID. -func (in *InodeOld) GID() auth.KGID { - return auth.KGID((uint32(in.GIDHi) << 16) | uint32(in.GIDLo)) -} - -// Size implements Inode.Size. -func (in *InodeOld) Size() uint64 { - // In ext2/ext3, in.SizeHi did not exist, it was instead named in.DirACL. - return uint64(in.SizeLo) -} - -// InodeSize implements Inode.InodeSize. -func (in *InodeOld) InodeSize() uint16 { return OldInodeSize } - -// AccessTime implements Inode.AccessTime. -func (in *InodeOld) AccessTime() time.Time { - return time.FromUnix(int64(in.AccessTimeRaw), 0) -} - -// ChangeTime implements Inode.ChangeTime. -func (in *InodeOld) ChangeTime() time.Time { - return time.FromUnix(int64(in.ChangeTimeRaw), 0) -} - -// ModificationTime implements Inode.ModificationTime. -func (in *InodeOld) ModificationTime() time.Time { - return time.FromUnix(int64(in.ModificationTimeRaw), 0) -} - -// DeletionTime implements Inode.DeletionTime. -func (in *InodeOld) DeletionTime() time.Time { - return time.FromUnix(int64(in.DeletionTimeRaw), 0) -} - -// LinksCount implements Inode.LinksCount. -func (in *InodeOld) LinksCount() uint16 { return in.LinksCountRaw } - -// Flags implements Inode.Flags. -func (in *InodeOld) Flags() InodeFlags { return InodeFlagsFromInt(in.FlagsRaw) } - -// Data implements Inode.Data. -func (in *InodeOld) Data() []byte { return in.DataRaw[:] } diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode_test.go b/pkg/sentry/fsimpl/ext/disklayout/inode_test.go deleted file mode 100644 index dd03ee50e..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/inode_test.go +++ /dev/null @@ -1,222 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "fmt" - "strconv" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/kernel/time" -) - -// TestInodeSize tests that the inode structs are of the correct size. -func TestInodeSize(t *testing.T) { - assertSize(t, InodeOld{}, OldInodeSize) - - // This was updated from 156 bytes to 160 bytes in Oct 2015. - assertSize(t, InodeNew{}, 160) -} - -// TestTimestampSeconds tests that the seconds part of [a/c/m] timestamps in -// ext4 inode structs are decoded correctly. -// -// These tests are derived from the table under https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#inode-timestamps. -func TestTimestampSeconds(t *testing.T) { - type timestampTest struct { - // msbSet tells if the most significant bit of InodeOld.[X]TimeRaw is set. - // If this is set then the 32-bit time is negative. - msbSet bool - - // lowerBound tells if we should take the lowest possible value of - // InodeOld.[X]TimeRaw while satisfying test.msbSet condition. If set to - // false it tells to take the highest possible value. - lowerBound bool - - // extraBits is InodeNew.[X]TimeExtra. - extraBits uint32 - - // want is the kernel time struct that is expected. - want time.Time - } - - tests := []timestampTest{ - // 1901-12-13 - { - msbSet: true, - lowerBound: true, - extraBits: 0, - want: time.FromUnix(int64(-0x80000000), 0), - }, - - // 1969-12-31 - { - msbSet: true, - lowerBound: false, - extraBits: 0, - want: time.FromUnix(int64(-1), 0), - }, - - // 1970-01-01 - { - msbSet: false, - lowerBound: true, - extraBits: 0, - want: time.FromUnix(int64(0), 0), - }, - - // 2038-01-19 - { - msbSet: false, - lowerBound: false, - extraBits: 0, - want: time.FromUnix(int64(0x7fffffff), 0), - }, - - // 2038-01-19 - { - msbSet: true, - lowerBound: true, - extraBits: 1, - want: time.FromUnix(int64(0x80000000), 0), - }, - - // 2106-02-07 - { - msbSet: true, - lowerBound: false, - extraBits: 1, - want: time.FromUnix(int64(0xffffffff), 0), - }, - - // 2106-02-07 - { - msbSet: false, - lowerBound: true, - extraBits: 1, - want: time.FromUnix(int64(0x100000000), 0), - }, - - // 2174-02-25 - { - msbSet: false, - lowerBound: false, - extraBits: 1, - want: time.FromUnix(int64(0x17fffffff), 0), - }, - - // 2174-02-25 - { - msbSet: true, - lowerBound: true, - extraBits: 2, - want: time.FromUnix(int64(0x180000000), 0), - }, - - // 2242-03-16 - { - msbSet: true, - lowerBound: false, - extraBits: 2, - want: time.FromUnix(int64(0x1ffffffff), 0), - }, - - // 2242-03-16 - { - msbSet: false, - lowerBound: true, - extraBits: 2, - want: time.FromUnix(int64(0x200000000), 0), - }, - - // 2310-04-04 - { - msbSet: false, - lowerBound: false, - extraBits: 2, - want: time.FromUnix(int64(0x27fffffff), 0), - }, - - // 2310-04-04 - { - msbSet: true, - lowerBound: true, - extraBits: 3, - want: time.FromUnix(int64(0x280000000), 0), - }, - - // 2378-04-22 - { - msbSet: true, - lowerBound: false, - extraBits: 3, - want: time.FromUnix(int64(0x2ffffffff), 0), - }, - - // 2378-04-22 - { - msbSet: false, - lowerBound: true, - extraBits: 3, - want: time.FromUnix(int64(0x300000000), 0), - }, - - // 2446-05-10 - { - msbSet: false, - lowerBound: false, - extraBits: 3, - want: time.FromUnix(int64(0x37fffffff), 0), - }, - } - - lowerMSB0 := int32(0) // binary: 00000000 00000000 00000000 00000000 - upperMSB0 := int32(0x7fffffff) // binary: 01111111 11111111 11111111 11111111 - lowerMSB1 := int32(-0x80000000) // binary: 10000000 00000000 00000000 00000000 - upperMSB1 := int32(-1) // binary: 11111111 11111111 11111111 11111111 - - get32BitTime := func(test timestampTest) int32 { - if test.msbSet { - if test.lowerBound { - return lowerMSB1 - } - - return upperMSB1 - } - - if test.lowerBound { - return lowerMSB0 - } - - return upperMSB0 - } - - getTestName := func(test timestampTest) string { - return fmt.Sprintf( - "Tests time decoding with epoch bits 0b%s and 32-bit raw time: MSB set=%t, lower bound=%t", - strconv.FormatInt(int64(test.extraBits), 2), - test.msbSet, - test.lowerBound, - ) - } - - for _, test := range tests { - t.Run(getTestName(test), func(t *testing.T) { - if got := fromExtraTime(get32BitTime(test), test.extraBits); got != test.want { - t.Errorf("Expected: %v, Got: %v", test.want, got) - } - }) - } -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock.go b/pkg/sentry/fsimpl/ext/disklayout/superblock.go deleted file mode 100644 index 8bb327006..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock.go +++ /dev/null @@ -1,471 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -const ( - // SbOffset is the absolute offset at which the superblock is placed. - SbOffset = 1024 -) - -// SuperBlock should be implemented by structs representing the ext superblock. -// The superblock holds a lot of information about the enclosing filesystem. -// This interface aims to provide access methods to important information held -// by the superblock. It does NOT expose all fields of the superblock, only the -// ones necessary. This can be expanded when need be. -// -// Location and replication: -// - The superblock is located at offset 1024 in block group 0. -// - Redundant copies of the superblock and group descriptors are kept in -// all groups if SbSparse feature flag is NOT set. If it is set, the -// replicas only exist in groups whose group number is either 0 or a -// power of 3, 5, or 7. -// - There is also a sparse superblock feature v2 in which there are just -// two replicas saved in the block groups pointed by sb.s_backup_bgs. -// -// Replicas should eventually be updated if the superblock is updated. -// -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html#super-block. -type SuperBlock interface { - // InodesCount returns the total number of inodes in this filesystem. - InodesCount() uint32 - - // BlocksCount returns the total number of data blocks in this filesystem. - BlocksCount() uint64 - - // FreeBlocksCount returns the number of free blocks in this filesystem. - FreeBlocksCount() uint64 - - // FreeInodesCount returns the number of free inodes in this filesystem. - FreeInodesCount() uint32 - - // MountCount returns the number of mounts since the last fsck. - MountCount() uint16 - - // MaxMountCount returns the number of mounts allowed beyond which a fsck is - // needed. - MaxMountCount() uint16 - - // FirstDataBlock returns the absolute block number of the first data block, - // which contains the super block itself. - // - // If the filesystem has 1kb data blocks then this should return 1. For all - // other configurations, this typically returns 0. - FirstDataBlock() uint32 - - // BlockSize returns the size of one data block in this filesystem. - // This can be calculated by 2^(10 + sb.s_log_block_size). This ensures that - // the smallest block size is 1kb. - BlockSize() uint64 - - // BlocksPerGroup returns the number of data blocks in a block group. - BlocksPerGroup() uint32 - - // ClusterSize returns block cluster size (set during mkfs time by admin). - // This can be calculated by 2^(10 + sb.s_log_cluster_size). This ensures that - // the smallest cluster size is 1kb. - // - // sb.s_log_cluster_size must equal sb.s_log_block_size if bigalloc feature - // is NOT set and consequently BlockSize() = ClusterSize() in that case. - ClusterSize() uint64 - - // ClustersPerGroup returns: - // - number of clusters per group if bigalloc is enabled. - // - BlocksPerGroup() otherwise. - ClustersPerGroup() uint32 - - // InodeSize returns the size of the inode disk record size in bytes. Use this - // to iterate over inode arrays on disk. - // - // In ext2 and ext3: - // - Each inode had a disk record of 128 bytes. - // - The inode struct size was fixed at 128 bytes. - // - // In ext4 its possible to allocate larger on-disk inodes: - // - Inode disk record size = sb.s_inode_size (function return value). - // = 256 (default) - // - Inode struct size = 128 + inode.i_extra_isize. - // = 128 + 32 = 160 (default) - InodeSize() uint16 - - // InodesPerGroup returns the number of inodes in a block group. - InodesPerGroup() uint32 - - // BgDescSize returns the size of the block group descriptor struct. - // - // In ext2, ext3, ext4 (without 64-bit feature), the block group descriptor - // is only 32 bytes long. - // In ext4 with 64-bit feature, the block group descriptor expands to AT LEAST - // 64 bytes. It might be bigger than that. - BgDescSize() uint16 - - // CompatibleFeatures returns the CompatFeatures struct which holds all the - // compatible features this fs supports. - CompatibleFeatures() CompatFeatures - - // IncompatibleFeatures returns the CompatFeatures struct which holds all the - // incompatible features this fs supports. - IncompatibleFeatures() IncompatFeatures - - // ReadOnlyCompatibleFeatures returns the CompatFeatures struct which holds all the - // readonly compatible features this fs supports. - ReadOnlyCompatibleFeatures() RoCompatFeatures - - // Magic() returns the magic signature which must be 0xef53. - Magic() uint16 - - // Revision returns the superblock revision. Superblock struct fields from - // offset 0x54 till 0x150 should only be used if superblock has DynamicRev. - Revision() SbRevision -} - -// SbRevision is the type for superblock revisions. -type SbRevision uint32 - -// Super block revisions. -const ( - // OldRev is the good old (original) format. - OldRev SbRevision = 0 - - // DynamicRev is v2 format w/ dynamic inode sizes. - DynamicRev SbRevision = 1 -) - -// Superblock compatible features. -// This is not exhaustive, unused features are not listed. -const ( - // SbDirPrealloc indicates directory preallocation. - SbDirPrealloc = 0x1 - - // SbHasJournal indicates the presence of a journal. jbd2 should only work - // with this being set. - SbHasJournal = 0x4 - - // SbExtAttr indicates extended attributes support. - SbExtAttr = 0x8 - - // SbResizeInode indicates that the fs has reserved GDT blocks (right after - // group descriptors) for fs expansion. - SbResizeInode = 0x10 - - // SbDirIndex indicates that the fs has directory indices. - SbDirIndex = 0x20 - - // SbSparseV2 stands for Sparse superblock version 2. - SbSparseV2 = 0x200 -) - -// CompatFeatures represents a superblock's compatible feature set. If the -// kernel does not understand any of these feature, it can still read/write -// to this fs. -type CompatFeatures struct { - DirPrealloc bool - HasJournal bool - ExtAttr bool - ResizeInode bool - DirIndex bool - SparseV2 bool -} - -// ToInt converts superblock compatible features back to its 32-bit rep. -func (f CompatFeatures) ToInt() uint32 { - var res uint32 - - if f.DirPrealloc { - res |= SbDirPrealloc - } - if f.HasJournal { - res |= SbHasJournal - } - if f.ExtAttr { - res |= SbExtAttr - } - if f.ResizeInode { - res |= SbResizeInode - } - if f.DirIndex { - res |= SbDirIndex - } - if f.SparseV2 { - res |= SbSparseV2 - } - - return res -} - -// CompatFeaturesFromInt converts the integer representation of superblock -// compatible features to CompatFeatures struct. -func CompatFeaturesFromInt(f uint32) CompatFeatures { - return CompatFeatures{ - DirPrealloc: f&SbDirPrealloc > 0, - HasJournal: f&SbHasJournal > 0, - ExtAttr: f&SbExtAttr > 0, - ResizeInode: f&SbResizeInode > 0, - DirIndex: f&SbDirIndex > 0, - SparseV2: f&SbSparseV2 > 0, - } -} - -// Superblock incompatible features. -// This is not exhaustive, unused features are not listed. -const ( - // SbDirentFileType indicates that directory entries record the file type. - // We should use struct DirentNew for dirents then. - SbDirentFileType = 0x2 - - // SbRecovery indicates that the filesystem needs recovery. - SbRecovery = 0x4 - - // SbJournalDev indicates that the filesystem has a separate journal device. - SbJournalDev = 0x8 - - // SbMetaBG indicates that the filesystem is using Meta block groups. Moves - // the group descriptors from the congested first block group into the first - // group of each metablock group to increase the maximum block groups limit - // and hence support much larger filesystems. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/overview.html#meta-block-groups. - SbMetaBG = 0x10 - - // SbExtents indicates that the filesystem uses extents. Must be set in ext4 - // filesystems. - SbExtents = 0x40 - - // SbIs64Bit indicates that this filesystem addresses blocks with 64-bits. - // Hence can support 2^64 data blocks. - SbIs64Bit = 0x80 - - // SbMMP indicates that this filesystem has multiple mount protection. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html#multiple-mount-protection. - SbMMP = 0x100 - - // SbFlexBg indicates that this filesystem has flexible block groups. Several - // block groups are tied into one logical block group so that all the metadata - // for the block groups (bitmaps and inode tables) are close together for - // faster loading. Consequently, large files will be continuous on disk. - // However, this does not affect the placement of redundant superblocks and - // group descriptors. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/overview.html#flexible-block-groups. - SbFlexBg = 0x200 - - // SbLargeDir shows that large directory enabled. Directory htree can be 3 - // levels deep. Directory htrees are allowed to be 2 levels deep otherwise. - SbLargeDir = 0x4000 - - // SbInlineData allows inline data in inodes for really small files. - SbInlineData = 0x8000 - - // SbEncrypted indicates that this fs contains encrypted inodes. - SbEncrypted = 0x10000 -) - -// IncompatFeatures represents a superblock's incompatible feature set. If the -// kernel does not understand any of these feature, it should refuse to mount. -type IncompatFeatures struct { - DirentFileType bool - Recovery bool - JournalDev bool - MetaBG bool - Extents bool - Is64Bit bool - MMP bool - FlexBg bool - LargeDir bool - InlineData bool - Encrypted bool -} - -// ToInt converts superblock incompatible features back to its 32-bit rep. -func (f IncompatFeatures) ToInt() uint32 { - var res uint32 - - if f.DirentFileType { - res |= SbDirentFileType - } - if f.Recovery { - res |= SbRecovery - } - if f.JournalDev { - res |= SbJournalDev - } - if f.MetaBG { - res |= SbMetaBG - } - if f.Extents { - res |= SbExtents - } - if f.Is64Bit { - res |= SbIs64Bit - } - if f.MMP { - res |= SbMMP - } - if f.FlexBg { - res |= SbFlexBg - } - if f.LargeDir { - res |= SbLargeDir - } - if f.InlineData { - res |= SbInlineData - } - if f.Encrypted { - res |= SbEncrypted - } - - return res -} - -// IncompatFeaturesFromInt converts the integer representation of superblock -// incompatible features to IncompatFeatures struct. -func IncompatFeaturesFromInt(f uint32) IncompatFeatures { - return IncompatFeatures{ - DirentFileType: f&SbDirentFileType > 0, - Recovery: f&SbRecovery > 0, - JournalDev: f&SbJournalDev > 0, - MetaBG: f&SbMetaBG > 0, - Extents: f&SbExtents > 0, - Is64Bit: f&SbIs64Bit > 0, - MMP: f&SbMMP > 0, - FlexBg: f&SbFlexBg > 0, - LargeDir: f&SbLargeDir > 0, - InlineData: f&SbInlineData > 0, - Encrypted: f&SbEncrypted > 0, - } -} - -// Superblock readonly compatible features. -// This is not exhaustive, unused features are not listed. -const ( - // SbSparse indicates sparse superblocks. Only groups with number either 0 or - // a power of 3, 5, or 7 will have redundant copies of the superblock and - // block descriptors. - SbSparse = 0x1 - - // SbLargeFile indicates that this fs has been used to store a file >= 2GiB. - SbLargeFile = 0x2 - - // SbHugeFile indicates that this fs contains files whose sizes are - // represented in units of logicals blocks, not 512-byte sectors. - SbHugeFile = 0x8 - - // SbGdtCsum indicates that group descriptors have checksums. - SbGdtCsum = 0x10 - - // SbDirNlink indicates that the new subdirectory limit is 64,999. Ext3 has a - // 32,000 subdirectory limit. - SbDirNlink = 0x20 - - // SbExtraIsize indicates that large inodes exist on this filesystem. - SbExtraIsize = 0x40 - - // SbHasSnapshot indicates the existence of a snapshot. - SbHasSnapshot = 0x80 - - // SbQuota enables usage tracking for all quota types. - SbQuota = 0x100 - - // SbBigalloc maps to the bigalloc feature. When set, the minimum allocation - // unit becomes a cluster rather than a data block. Then block bitmaps track - // clusters, not data blocks. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/overview.html#bigalloc. - SbBigalloc = 0x200 - - // SbMetadataCsum indicates that the fs supports metadata checksumming. - SbMetadataCsum = 0x400 - - // SbReadOnly marks this filesystem as readonly. Should refuse to mount in - // read/write mode. - SbReadOnly = 0x1000 -) - -// RoCompatFeatures represents a superblock's readonly compatible feature set. -// If the kernel does not understand any of these feature, it can still mount -// readonly. But if the user wants to mount read/write, the kernel should -// refuse to mount. -type RoCompatFeatures struct { - Sparse bool - LargeFile bool - HugeFile bool - GdtCsum bool - DirNlink bool - ExtraIsize bool - HasSnapshot bool - Quota bool - Bigalloc bool - MetadataCsum bool - ReadOnly bool -} - -// ToInt converts superblock readonly compatible features to its 32-bit rep. -func (f RoCompatFeatures) ToInt() uint32 { - var res uint32 - - if f.Sparse { - res |= SbSparse - } - if f.LargeFile { - res |= SbLargeFile - } - if f.HugeFile { - res |= SbHugeFile - } - if f.GdtCsum { - res |= SbGdtCsum - } - if f.DirNlink { - res |= SbDirNlink - } - if f.ExtraIsize { - res |= SbExtraIsize - } - if f.HasSnapshot { - res |= SbHasSnapshot - } - if f.Quota { - res |= SbQuota - } - if f.Bigalloc { - res |= SbBigalloc - } - if f.MetadataCsum { - res |= SbMetadataCsum - } - if f.ReadOnly { - res |= SbReadOnly - } - - return res -} - -// RoCompatFeaturesFromInt converts the integer representation of superblock -// readonly compatible features to RoCompatFeatures struct. -func RoCompatFeaturesFromInt(f uint32) RoCompatFeatures { - return RoCompatFeatures{ - Sparse: f&SbSparse > 0, - LargeFile: f&SbLargeFile > 0, - HugeFile: f&SbHugeFile > 0, - GdtCsum: f&SbGdtCsum > 0, - DirNlink: f&SbDirNlink > 0, - ExtraIsize: f&SbExtraIsize > 0, - HasSnapshot: f&SbHasSnapshot > 0, - Quota: f&SbQuota > 0, - Bigalloc: f&SbBigalloc > 0, - MetadataCsum: f&SbMetadataCsum > 0, - ReadOnly: f&SbReadOnly > 0, - } -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock_32.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_32.go deleted file mode 100644 index 53e515fd3..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock_32.go +++ /dev/null @@ -1,76 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// SuperBlock32Bit implements SuperBlock and represents the 32-bit version of -// the ext4_super_block struct in fs/ext4/ext4.h. Should be used only if -// RevLevel = DynamicRev and 64-bit feature is disabled. -type SuperBlock32Bit struct { - // We embed the old superblock struct here because the 32-bit version is just - // an extension of the old version. - SuperBlockOld - - FirstInode uint32 - InodeSizeRaw uint16 - BlockGroupNumber uint16 - FeatureCompat uint32 - FeatureIncompat uint32 - FeatureRoCompat uint32 - UUID [16]byte - VolumeName [16]byte - LastMounted [64]byte - AlgoUsageBitmap uint32 - PreallocBlocks uint8 - PreallocDirBlocks uint8 - ReservedGdtBlocks uint16 - JournalUUID [16]byte - JournalInum uint32 - JournalDev uint32 - LastOrphan uint32 - HashSeed [4]uint32 - DefaultHashVersion uint8 - JnlBackupType uint8 - BgDescSizeRaw uint16 - DefaultMountOpts uint32 - FirstMetaBg uint32 - MkfsTime uint32 - JnlBlocks [17]uint32 -} - -// Compiles only if SuperBlock32Bit implements SuperBlock. -var _ SuperBlock = (*SuperBlock32Bit)(nil) - -// Only override methods which change based on the additional fields above. -// Not overriding SuperBlock.BgDescSize because it would still return 32 here. - -// InodeSize implements SuperBlock.InodeSize. -func (sb *SuperBlock32Bit) InodeSize() uint16 { - return sb.InodeSizeRaw -} - -// CompatibleFeatures implements SuperBlock.CompatibleFeatures. -func (sb *SuperBlock32Bit) CompatibleFeatures() CompatFeatures { - return CompatFeaturesFromInt(sb.FeatureCompat) -} - -// IncompatibleFeatures implements SuperBlock.IncompatibleFeatures. -func (sb *SuperBlock32Bit) IncompatibleFeatures() IncompatFeatures { - return IncompatFeaturesFromInt(sb.FeatureIncompat) -} - -// ReadOnlyCompatibleFeatures implements SuperBlock.ReadOnlyCompatibleFeatures. -func (sb *SuperBlock32Bit) ReadOnlyCompatibleFeatures() RoCompatFeatures { - return RoCompatFeaturesFromInt(sb.FeatureRoCompat) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock_64.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_64.go deleted file mode 100644 index 7c1053fb4..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock_64.go +++ /dev/null @@ -1,95 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// SuperBlock64Bit implements SuperBlock and represents the 64-bit version of -// the ext4_super_block struct in fs/ext4/ext4.h. This sums up to be exactly -// 1024 bytes (smallest possible block size) and hence the superblock always -// fits in no more than one data block. Should only be used when the 64-bit -// feature is set. -type SuperBlock64Bit struct { - // We embed the 32-bit struct here because 64-bit version is just an extension - // of the 32-bit version. - SuperBlock32Bit - - BlocksCountHi uint32 - ReservedBlocksCountHi uint32 - FreeBlocksCountHi uint32 - MinInodeSize uint16 - WantInodeSize uint16 - Flags uint32 - RaidStride uint16 - MmpInterval uint16 - MmpBlock uint64 - RaidStripeWidth uint32 - LogGroupsPerFlex uint8 - ChecksumType uint8 - _ uint16 - KbytesWritten uint64 - SnapshotInum uint32 - SnapshotID uint32 - SnapshotRsrvBlocksCount uint64 - SnapshotList uint32 - ErrorCount uint32 - FirstErrorTime uint32 - FirstErrorInode uint32 - FirstErrorBlock uint64 - FirstErrorFunction [32]byte - FirstErrorLine uint32 - LastErrorTime uint32 - LastErrorInode uint32 - LastErrorLine uint32 - LastErrorBlock uint64 - LastErrorFunction [32]byte - MountOpts [64]byte - UserQuotaInum uint32 - GroupQuotaInum uint32 - OverheadBlocks uint32 - BackupBgs [2]uint32 - EncryptAlgos [4]uint8 - EncryptPwSalt [16]uint8 - LostFoundInode uint32 - ProjectQuotaInode uint32 - ChecksumSeed uint32 - WtimeHi uint8 - MtimeHi uint8 - MkfsTimeHi uint8 - LastCheckHi uint8 - FirstErrorTimeHi uint8 - LastErrorTimeHi uint8 - _ [2]uint8 - Encoding uint16 - EncodingFlags uint16 - _ [95]uint32 - Checksum uint32 -} - -// Compiles only if SuperBlock64Bit implements SuperBlock. -var _ SuperBlock = (*SuperBlock64Bit)(nil) - -// Only override methods which change based on the 64-bit feature. - -// BlocksCount implements SuperBlock.BlocksCount. -func (sb *SuperBlock64Bit) BlocksCount() uint64 { - return (uint64(sb.BlocksCountHi) << 32) | uint64(sb.BlocksCountLo) -} - -// FreeBlocksCount implements SuperBlock.FreeBlocksCount. -func (sb *SuperBlock64Bit) FreeBlocksCount() uint64 { - return (uint64(sb.FreeBlocksCountHi) << 32) | uint64(sb.FreeBlocksCountLo) -} - -// BgDescSize implements SuperBlock.BgDescSize. -func (sb *SuperBlock64Bit) BgDescSize() uint16 { return sb.BgDescSizeRaw } diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock_old.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_old.go deleted file mode 100644 index 9221e0251..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock_old.go +++ /dev/null @@ -1,105 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// SuperBlockOld implements SuperBlock and represents the old version of the -// superblock struct. Should be used only if RevLevel = OldRev. -type SuperBlockOld struct { - InodesCountRaw uint32 - BlocksCountLo uint32 - ReservedBlocksCount uint32 - FreeBlocksCountLo uint32 - FreeInodesCountRaw uint32 - FirstDataBlockRaw uint32 - LogBlockSize uint32 - LogClusterSize uint32 - BlocksPerGroupRaw uint32 - ClustersPerGroupRaw uint32 - InodesPerGroupRaw uint32 - Mtime uint32 - Wtime uint32 - MountCountRaw uint16 - MaxMountCountRaw uint16 - MagicRaw uint16 - State uint16 - Errors uint16 - MinorRevLevel uint16 - LastCheck uint32 - CheckInterval uint32 - CreatorOS uint32 - RevLevel uint32 - DefResUID uint16 - DefResGID uint16 -} - -// Compiles only if SuperBlockOld implements SuperBlock. -var _ SuperBlock = (*SuperBlockOld)(nil) - -// InodesCount implements SuperBlock.InodesCount. -func (sb *SuperBlockOld) InodesCount() uint32 { return sb.InodesCountRaw } - -// BlocksCount implements SuperBlock.BlocksCount. -func (sb *SuperBlockOld) BlocksCount() uint64 { return uint64(sb.BlocksCountLo) } - -// FreeBlocksCount implements SuperBlock.FreeBlocksCount. -func (sb *SuperBlockOld) FreeBlocksCount() uint64 { return uint64(sb.FreeBlocksCountLo) } - -// FreeInodesCount implements SuperBlock.FreeInodesCount. -func (sb *SuperBlockOld) FreeInodesCount() uint32 { return sb.FreeInodesCountRaw } - -// MountCount implements SuperBlock.MountCount. -func (sb *SuperBlockOld) MountCount() uint16 { return sb.MountCountRaw } - -// MaxMountCount implements SuperBlock.MaxMountCount. -func (sb *SuperBlockOld) MaxMountCount() uint16 { return sb.MaxMountCountRaw } - -// FirstDataBlock implements SuperBlock.FirstDataBlock. -func (sb *SuperBlockOld) FirstDataBlock() uint32 { return sb.FirstDataBlockRaw } - -// BlockSize implements SuperBlock.BlockSize. -func (sb *SuperBlockOld) BlockSize() uint64 { return 1 << (10 + sb.LogBlockSize) } - -// BlocksPerGroup implements SuperBlock.BlocksPerGroup. -func (sb *SuperBlockOld) BlocksPerGroup() uint32 { return sb.BlocksPerGroupRaw } - -// ClusterSize implements SuperBlock.ClusterSize. -func (sb *SuperBlockOld) ClusterSize() uint64 { return 1 << (10 + sb.LogClusterSize) } - -// ClustersPerGroup implements SuperBlock.ClustersPerGroup. -func (sb *SuperBlockOld) ClustersPerGroup() uint32 { return sb.ClustersPerGroupRaw } - -// InodeSize implements SuperBlock.InodeSize. -func (sb *SuperBlockOld) InodeSize() uint16 { return OldInodeSize } - -// InodesPerGroup implements SuperBlock.InodesPerGroup. -func (sb *SuperBlockOld) InodesPerGroup() uint32 { return sb.InodesPerGroupRaw } - -// BgDescSize implements SuperBlock.BgDescSize. -func (sb *SuperBlockOld) BgDescSize() uint16 { return 32 } - -// CompatibleFeatures implements SuperBlock.CompatibleFeatures. -func (sb *SuperBlockOld) CompatibleFeatures() CompatFeatures { return CompatFeatures{} } - -// IncompatibleFeatures implements SuperBlock.IncompatibleFeatures. -func (sb *SuperBlockOld) IncompatibleFeatures() IncompatFeatures { return IncompatFeatures{} } - -// ReadOnlyCompatibleFeatures implements SuperBlock.ReadOnlyCompatibleFeatures. -func (sb *SuperBlockOld) ReadOnlyCompatibleFeatures() RoCompatFeatures { return RoCompatFeatures{} } - -// Magic implements SuperBlock.Magic. -func (sb *SuperBlockOld) Magic() uint16 { return sb.MagicRaw } - -// Revision implements SuperBlock.Revision. -func (sb *SuperBlockOld) Revision() SbRevision { return SbRevision(sb.RevLevel) } diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock_test.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_test.go deleted file mode 100644 index 463b5ba21..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock_test.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "testing" -) - -// TestSuperBlockSize tests that the superblock structs are of the correct -// size. -func TestSuperBlockSize(t *testing.T) { - assertSize(t, SuperBlockOld{}, 84) - assertSize(t, SuperBlock32Bit{}, 336) - assertSize(t, SuperBlock64Bit{}, 1024) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/test_utils.go b/pkg/sentry/fsimpl/ext/disklayout/test_utils.go deleted file mode 100644 index 9c63f04c0..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/test_utils.go +++ /dev/null @@ -1,30 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/binary" -) - -func assertSize(t *testing.T, v interface{}, want uintptr) { - t.Helper() - - if got := binary.Size(v); got != want { - t.Errorf("struct %s should be exactly %d bytes but is %d bytes", reflect.TypeOf(v).Name(), want, got) - } -} diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go deleted file mode 100644 index 373d23b74..000000000 --- a/pkg/sentry/fsimpl/ext/ext.go +++ /dev/null @@ -1,135 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package ext implements readonly ext(2/3/4) filesystems. -package ext - -import ( - "errors" - "fmt" - "io" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// Compiles only if FilesystemType implements vfs.FilesystemType. -var _ vfs.FilesystemType = (*FilesystemType)(nil) - -// getDeviceFd returns an io.ReaderAt to the underlying device. -// Currently there are two ways of mounting an ext(2/3/4) fs: -// 1. Specify a mount with our internal special MountType in the OCI spec. -// 2. Expose the device to the container and mount it from application layer. -func getDeviceFd(source string, opts vfs.GetFilesystemOptions) (io.ReaderAt, error) { - if opts.InternalData == nil { - // User mount call. - // TODO(b/134676337): Open the device specified by `source` and return that. - panic("unimplemented") - } - - // GetFilesystem call originated from within the sentry. - devFd, ok := opts.InternalData.(int) - if !ok { - return nil, errors.New("internal data for ext fs must be an int containing the file descriptor to device") - } - - if devFd < 0 { - return nil, fmt.Errorf("ext device file descriptor is not valid: %d", devFd) - } - - // The fd.ReadWriter returned from fd.NewReadWriter() does not take ownership - // of the file descriptor and hence will not close it when it is garbage - // collected. - return fd.NewReadWriter(devFd), nil -} - -// isCompatible checks if the superblock has feature sets which are compatible. -// We only need to check the superblock incompatible feature set since we are -// mounting readonly. We will also need to check readonly compatible feature -// set when mounting for read/write. -func isCompatible(sb disklayout.SuperBlock) bool { - // Please note that what is being checked is limited based on the fact that we - // are mounting readonly and that we are not journaling. When mounting - // read/write or with a journal, this must be reevaluated. - incompatFeatures := sb.IncompatibleFeatures() - if incompatFeatures.MetaBG { - log.Warningf("ext fs: meta block groups are not supported") - return false - } - if incompatFeatures.MMP { - log.Warningf("ext fs: multiple mount protection is not supported") - return false - } - if incompatFeatures.Encrypted { - log.Warningf("ext fs: encrypted inodes not supported") - return false - } - if incompatFeatures.InlineData { - log.Warningf("ext fs: inline files not supported") - return false - } - return true -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - // TODO(b/134676337): Ensure that the user is mounting readonly. If not, - // EACCESS should be returned according to mount(2). Filesystem independent - // flags (like readonly) are currently not available in pkg/sentry/vfs. - - dev, err := getDeviceFd(source, opts) - if err != nil { - return nil, nil, err - } - - fs := filesystem{dev: dev, inodeCache: make(map[uint32]*inode)} - fs.vfsfs.Init(vfsObj, &fs) - fs.sb, err = readSuperBlock(dev) - if err != nil { - return nil, nil, err - } - - if fs.sb.Magic() != linux.EXT_SUPER_MAGIC { - // mount(2) specifies that EINVAL should be returned if the superblock is - // invalid. - return nil, nil, syserror.EINVAL - } - - // Refuse to mount if the filesystem is incompatible. - if !isCompatible(fs.sb) { - return nil, nil, syserror.EINVAL - } - - fs.bgs, err = readBlockGroups(dev, fs.sb) - if err != nil { - return nil, nil, err - } - - rootInode, err := fs.getOrCreateInodeLocked(disklayout.RootDirInode) - if err != nil { - return nil, nil, err - } - rootInode.incRef() - - return &fs.vfsfs, &newDentry(rootInode).vfsd, nil -} diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go deleted file mode 100644 index 29bb73765..000000000 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ /dev/null @@ -1,922 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "fmt" - "io" - "os" - "path" - "sort" - "testing" - - "github.com/google/go-cmp/cmp" - "github.com/google/go-cmp/cmp/cmpopts" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" - - "gvisor.dev/gvisor/runsc/testutil" -) - -const ( - assetsDir = "pkg/sentry/fsimpl/ext/assets" -) - -var ( - ext2ImagePath = path.Join(assetsDir, "tiny.ext2") - ext3ImagePath = path.Join(assetsDir, "tiny.ext3") - ext4ImagePath = path.Join(assetsDir, "tiny.ext4") -) - -// setUp opens imagePath as an ext Filesystem and returns all necessary -// elements required to run tests. If error is non-nil, it also returns a tear -// down function which must be called after the test is run for clean up. -func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesystem, *vfs.VirtualDentry, func(), error) { - localImagePath, err := testutil.FindFile(imagePath) - if err != nil { - return nil, nil, nil, nil, fmt.Errorf("failed to open local image at path %s: %v", imagePath, err) - } - - f, err := os.Open(localImagePath) - if err != nil { - return nil, nil, nil, nil, err - } - - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, localImagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}) - if err != nil { - f.Close() - return nil, nil, nil, nil, err - } - - root := mntns.Root() - - tearDown := func() { - root.DecRef() - - if err := f.Close(); err != nil { - t.Fatalf("tearDown failed: %v", err) - } - } - return ctx, vfsObj, &root, tearDown, nil -} - -// TODO(b/134676337): Test vfs.FilesystemImpl.ReadlinkAt and -// vfs.FilesystemImpl.StatFSAt which are not implemented in -// vfs.VirtualFilesystem yet. - -// TestSeek tests vfs.FileDescriptionImpl.Seek functionality. -func TestSeek(t *testing.T) { - type seekTest struct { - name string - image string - path string - } - - tests := []seekTest{ - { - name: "ext4 root dir seek", - image: ext4ImagePath, - path: "/", - }, - { - name: "ext3 root dir seek", - image: ext3ImagePath, - path: "/", - }, - { - name: "ext2 root dir seek", - image: ext2ImagePath, - path: "/", - }, - { - name: "ext4 reg file seek", - image: ext4ImagePath, - path: "/file.txt", - }, - { - name: "ext3 reg file seek", - image: ext3ImagePath, - path: "/file.txt", - }, - { - name: "ext2 reg file seek", - image: ext2ImagePath, - path: "/file.txt", - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - ctx, vfsfs, root, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fd, err := vfsfs.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt failed: %v", err) - } - - if n, err := fd.Seek(ctx, 0, linux.SEEK_SET); n != 0 || err != nil { - t.Errorf("expected seek position 0, got %d and error %v", n, err) - } - - stat, err := fd.Stat(ctx, vfs.StatOptions{}) - if err != nil { - t.Errorf("fd.stat failed for file %s in image %s: %v", test.path, test.image, err) - } - - // We should be able to seek beyond the end of file. - size := int64(stat.Size) - if n, err := fd.Seek(ctx, size, linux.SEEK_SET); n != size || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", size, n, err) - } - - // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Seek(ctx, -1, linux.SEEK_SET); err != syserror.EINVAL { - t.Errorf("expected error EINVAL but got %v", err) - } - - if n, err := fd.Seek(ctx, 3, linux.SEEK_CUR); n != size+3 || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", size+3, n, err) - } - - // Make sure negative offsets work with SEEK_CUR. - if n, err := fd.Seek(ctx, -2, linux.SEEK_CUR); n != size+1 || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", size+1, n, err) - } - - // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Seek(ctx, -(size + 2), linux.SEEK_CUR); err != syserror.EINVAL { - t.Errorf("expected error EINVAL but got %v", err) - } - - // Make sure SEEK_END works with regular files. - if _, ok := fd.Impl().(*regularFileFD); ok { - // Seek back to 0. - if n, err := fd.Seek(ctx, -size, linux.SEEK_END); n != 0 || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", 0, n, err) - } - - // Seek forward beyond EOF. - if n, err := fd.Seek(ctx, 1, linux.SEEK_END); n != size+1 || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", size+1, n, err) - } - - // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Seek(ctx, -(size + 1), linux.SEEK_END); err != syserror.EINVAL { - t.Errorf("expected error EINVAL but got %v", err) - } - } - }) - } -} - -// TestStatAt tests filesystem.StatAt functionality. -func TestStatAt(t *testing.T) { - type statAtTest struct { - name string - image string - path string - want linux.Statx - } - - tests := []statAtTest{ - { - name: "ext4 statx small file", - image: ext4ImagePath, - path: "/file.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13, - }, - }, - { - name: "ext3 statx small file", - image: ext3ImagePath, - path: "/file.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13, - }, - }, - { - name: "ext2 statx small file", - image: ext2ImagePath, - path: "/file.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13, - }, - }, - { - name: "ext4 statx big file", - image: ext4ImagePath, - path: "/bigfile.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13042, - }, - }, - { - name: "ext3 statx big file", - image: ext3ImagePath, - path: "/bigfile.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13042, - }, - }, - { - name: "ext2 statx big file", - image: ext2ImagePath, - path: "/bigfile.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13042, - }, - }, - { - name: "ext4 statx symlink file", - image: ext4ImagePath, - path: "/symlink.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0777 | linux.ModeSymlink, - Size: 8, - }, - }, - { - name: "ext3 statx symlink file", - image: ext3ImagePath, - path: "/symlink.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0777 | linux.ModeSymlink, - Size: 8, - }, - }, - { - name: "ext2 statx symlink file", - image: ext2ImagePath, - path: "/symlink.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0777 | linux.ModeSymlink, - Size: 8, - }, - }, - } - - // Ignore the fields that are not supported by filesystem.StatAt yet and - // those which are likely to change as the image does. - ignoredFields := map[string]bool{ - "Attributes": true, - "AttributesMask": true, - "Atime": true, - "Blocks": true, - "Btime": true, - "Ctime": true, - "DevMajor": true, - "DevMinor": true, - "Ino": true, - "Mask": true, - "Mtime": true, - "RdevMajor": true, - "RdevMinor": true, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - ctx, vfsfs, root, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - got, err := vfsfs.StatAt(ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, - &vfs.StatOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.StatAt failed for file %s in image %s: %v", test.path, test.image, err) - } - - cmpIgnoreFields := cmp.FilterPath(func(p cmp.Path) bool { - _, ok := ignoredFields[p.String()] - return ok - }, cmp.Ignore()) - if diff := cmp.Diff(got, test.want, cmpIgnoreFields, cmpopts.IgnoreUnexported(linux.Statx{})); diff != "" { - t.Errorf("stat mismatch (-want +got):\n%s", diff) - } - }) - } -} - -// TestRead tests the read functionality for vfs file descriptions. -func TestRead(t *testing.T) { - type readTest struct { - name string - image string - absPath string - } - - tests := []readTest{ - { - name: "ext4 read small file", - image: ext4ImagePath, - absPath: "/file.txt", - }, - { - name: "ext3 read small file", - image: ext3ImagePath, - absPath: "/file.txt", - }, - { - name: "ext2 read small file", - image: ext2ImagePath, - absPath: "/file.txt", - }, - { - name: "ext4 read big file", - image: ext4ImagePath, - absPath: "/bigfile.txt", - }, - { - name: "ext3 read big file", - image: ext3ImagePath, - absPath: "/bigfile.txt", - }, - { - name: "ext2 read big file", - image: ext2ImagePath, - absPath: "/bigfile.txt", - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - ctx, vfsfs, root, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fd, err := vfsfs.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.absPath)}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt failed: %v", err) - } - - // Get a local file descriptor and compare its functionality with a vfs file - // description for the same file. - localFile, err := testutil.FindFile(path.Join(assetsDir, test.absPath)) - if err != nil { - t.Fatalf("testutil.FindFile failed for %s: %v", test.absPath, err) - } - - f, err := os.Open(localFile) - if err != nil { - t.Fatalf("os.Open failed for %s: %v", localFile, err) - } - defer f.Close() - - // Read the entire file by reading one byte repeatedly. Doing this stress - // tests the underlying file reader implementation. - got := make([]byte, 1) - want := make([]byte, 1) - for { - n, err := f.Read(want) - fd.Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}) - - if diff := cmp.Diff(got, want); diff != "" { - t.Errorf("file data mismatch (-want +got):\n%s", diff) - } - - // Make sure there is no more file data left after getting EOF. - if n == 0 || err == io.EOF { - if n, _ := fd.Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}); n != 0 { - t.Errorf("extra unexpected file data in file %s in image %s", test.absPath, test.image) - } - - break - } - - if err != nil { - t.Fatalf("read failed: %v", err) - } - } - }) - } -} - -// iterDirentsCb is a simple callback which just keeps adding the dirents to an -// internal list. Implements vfs.IterDirentsCallback. -type iterDirentsCb struct { - dirents []vfs.Dirent -} - -// Compiles only if iterDirentCb implements vfs.IterDirentsCallback. -var _ vfs.IterDirentsCallback = (*iterDirentsCb)(nil) - -// newIterDirentsCb is the iterDirent -func newIterDirentCb() *iterDirentsCb { - return &iterDirentsCb{dirents: make([]vfs.Dirent, 0)} -} - -// Handle implements vfs.IterDirentsCallback.Handle. -func (cb *iterDirentsCb) Handle(dirent vfs.Dirent) error { - cb.dirents = append(cb.dirents, dirent) - return nil -} - -// TestIterDirents tests the FileDescriptionImpl.IterDirents functionality. -func TestIterDirents(t *testing.T) { - type iterDirentTest struct { - name string - image string - path string - want []vfs.Dirent - } - - wantDirents := []vfs.Dirent{ - { - Name: ".", - Type: linux.DT_DIR, - }, - { - Name: "..", - Type: linux.DT_DIR, - }, - { - Name: "lost+found", - Type: linux.DT_DIR, - }, - { - Name: "file.txt", - Type: linux.DT_REG, - }, - { - Name: "bigfile.txt", - Type: linux.DT_REG, - }, - { - Name: "symlink.txt", - Type: linux.DT_LNK, - }, - } - tests := []iterDirentTest{ - { - name: "ext4 root dir iteration", - image: ext4ImagePath, - path: "/", - want: wantDirents, - }, - { - name: "ext3 root dir iteration", - image: ext3ImagePath, - path: "/", - want: wantDirents, - }, - { - name: "ext2 root dir iteration", - image: ext2ImagePath, - path: "/", - want: wantDirents, - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - ctx, vfsfs, root, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fd, err := vfsfs.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt failed: %v", err) - } - - cb := &iterDirentsCb{} - if err = fd.IterDirents(ctx, cb); err != nil { - t.Fatalf("dir fd.IterDirents() failed: %v", err) - } - - sort.Slice(cb.dirents, func(i int, j int) bool { return cb.dirents[i].Name < cb.dirents[j].Name }) - sort.Slice(test.want, func(i int, j int) bool { return test.want[i].Name < test.want[j].Name }) - - // Ignore the inode number and offset of dirents because those are likely to - // change as the underlying image changes. - cmpIgnoreFields := cmp.FilterPath(func(p cmp.Path) bool { - return p.String() == "Ino" || p.String() == "NextOff" - }, cmp.Ignore()) - if diff := cmp.Diff(cb.dirents, test.want, cmpIgnoreFields); diff != "" { - t.Errorf("dirents mismatch (-want +got):\n%s", diff) - } - }) - } -} - -// TestRootDir tests that the root directory inode is correctly initialized and -// returned from setUp. -func TestRootDir(t *testing.T) { - type inodeProps struct { - Mode linux.FileMode - UID auth.KUID - GID auth.KGID - Size uint64 - InodeSize uint16 - Links uint16 - Flags disklayout.InodeFlags - } - - type rootDirTest struct { - name string - image string - wantInode inodeProps - } - - tests := []rootDirTest{ - { - name: "ext4 root dir", - image: ext4ImagePath, - wantInode: inodeProps{ - Mode: linux.ModeDirectory | 0755, - Size: 0x400, - InodeSize: 0x80, - Links: 3, - Flags: disklayout.InodeFlags{Extents: true}, - }, - }, - { - name: "ext3 root dir", - image: ext3ImagePath, - wantInode: inodeProps{ - Mode: linux.ModeDirectory | 0755, - Size: 0x400, - InodeSize: 0x80, - Links: 3, - }, - }, - { - name: "ext2 root dir", - image: ext2ImagePath, - wantInode: inodeProps{ - Mode: linux.ModeDirectory | 0755, - Size: 0x400, - InodeSize: 0x80, - Links: 3, - }, - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - _, _, vd, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - d, ok := vd.Dentry().Impl().(*dentry) - if !ok { - t.Fatalf("ext dentry of incorrect type: %T", vd.Dentry().Impl()) - } - - // Offload inode contents into local structs for comparison. - gotInode := inodeProps{ - Mode: d.inode.diskInode.Mode(), - UID: d.inode.diskInode.UID(), - GID: d.inode.diskInode.GID(), - Size: d.inode.diskInode.Size(), - InodeSize: d.inode.diskInode.InodeSize(), - Links: d.inode.diskInode.LinksCount(), - Flags: d.inode.diskInode.Flags(), - } - - if diff := cmp.Diff(gotInode, test.wantInode); diff != "" { - t.Errorf("inode mismatch (-want +got):\n%s", diff) - } - }) - } -} - -// TestFilesystemInit tests that the filesystem superblock and block group -// descriptors are correctly read in and initialized. -func TestFilesystemInit(t *testing.T) { - // sb only contains the immutable properties of the superblock. - type sb struct { - InodesCount uint32 - BlocksCount uint64 - MaxMountCount uint16 - FirstDataBlock uint32 - BlockSize uint64 - BlocksPerGroup uint32 - ClusterSize uint64 - ClustersPerGroup uint32 - InodeSize uint16 - InodesPerGroup uint32 - BgDescSize uint16 - Magic uint16 - Revision disklayout.SbRevision - CompatFeatures disklayout.CompatFeatures - IncompatFeatures disklayout.IncompatFeatures - RoCompatFeatures disklayout.RoCompatFeatures - } - - // bg only contains the immutable properties of the block group descriptor. - type bg struct { - InodeTable uint64 - BlockBitmap uint64 - InodeBitmap uint64 - ExclusionBitmap uint64 - Flags disklayout.BGFlags - } - - type fsInitTest struct { - name string - image string - wantSb sb - wantBgs []bg - } - - tests := []fsInitTest{ - { - name: "ext4 filesystem init", - image: ext4ImagePath, - wantSb: sb{ - InodesCount: 0x10, - BlocksCount: 0x40, - MaxMountCount: 0xffff, - FirstDataBlock: 0x1, - BlockSize: 0x400, - BlocksPerGroup: 0x2000, - ClusterSize: 0x400, - ClustersPerGroup: 0x2000, - InodeSize: 0x80, - InodesPerGroup: 0x10, - BgDescSize: 0x40, - Magic: linux.EXT_SUPER_MAGIC, - Revision: disklayout.DynamicRev, - CompatFeatures: disklayout.CompatFeatures{ - ExtAttr: true, - ResizeInode: true, - DirIndex: true, - }, - IncompatFeatures: disklayout.IncompatFeatures{ - DirentFileType: true, - Extents: true, - Is64Bit: true, - FlexBg: true, - }, - RoCompatFeatures: disklayout.RoCompatFeatures{ - Sparse: true, - LargeFile: true, - HugeFile: true, - DirNlink: true, - ExtraIsize: true, - MetadataCsum: true, - }, - }, - wantBgs: []bg{ - { - InodeTable: 0x23, - BlockBitmap: 0x3, - InodeBitmap: 0x13, - Flags: disklayout.BGFlags{ - InodeZeroed: true, - }, - }, - }, - }, - { - name: "ext3 filesystem init", - image: ext3ImagePath, - wantSb: sb{ - InodesCount: 0x10, - BlocksCount: 0x40, - MaxMountCount: 0xffff, - FirstDataBlock: 0x1, - BlockSize: 0x400, - BlocksPerGroup: 0x2000, - ClusterSize: 0x400, - ClustersPerGroup: 0x2000, - InodeSize: 0x80, - InodesPerGroup: 0x10, - BgDescSize: 0x20, - Magic: linux.EXT_SUPER_MAGIC, - Revision: disklayout.DynamicRev, - CompatFeatures: disklayout.CompatFeatures{ - ExtAttr: true, - ResizeInode: true, - DirIndex: true, - }, - IncompatFeatures: disklayout.IncompatFeatures{ - DirentFileType: true, - }, - RoCompatFeatures: disklayout.RoCompatFeatures{ - Sparse: true, - LargeFile: true, - }, - }, - wantBgs: []bg{ - { - InodeTable: 0x5, - BlockBitmap: 0x3, - InodeBitmap: 0x4, - Flags: disklayout.BGFlags{ - InodeZeroed: true, - }, - }, - }, - }, - { - name: "ext2 filesystem init", - image: ext2ImagePath, - wantSb: sb{ - InodesCount: 0x10, - BlocksCount: 0x40, - MaxMountCount: 0xffff, - FirstDataBlock: 0x1, - BlockSize: 0x400, - BlocksPerGroup: 0x2000, - ClusterSize: 0x400, - ClustersPerGroup: 0x2000, - InodeSize: 0x80, - InodesPerGroup: 0x10, - BgDescSize: 0x20, - Magic: linux.EXT_SUPER_MAGIC, - Revision: disklayout.DynamicRev, - CompatFeatures: disklayout.CompatFeatures{ - ExtAttr: true, - ResizeInode: true, - DirIndex: true, - }, - IncompatFeatures: disklayout.IncompatFeatures{ - DirentFileType: true, - }, - RoCompatFeatures: disklayout.RoCompatFeatures{ - Sparse: true, - LargeFile: true, - }, - }, - wantBgs: []bg{ - { - InodeTable: 0x5, - BlockBitmap: 0x3, - InodeBitmap: 0x4, - Flags: disklayout.BGFlags{ - InodeZeroed: true, - }, - }, - }, - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - _, _, vd, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fs, ok := vd.Mount().Filesystem().Impl().(*filesystem) - if !ok { - t.Fatalf("ext filesystem of incorrect type: %T", vd.Mount().Filesystem().Impl()) - } - - // Offload superblock and block group descriptors contents into - // local structs for comparison. - totalFreeInodes := uint32(0) - totalFreeBlocks := uint64(0) - gotSb := sb{ - InodesCount: fs.sb.InodesCount(), - BlocksCount: fs.sb.BlocksCount(), - MaxMountCount: fs.sb.MaxMountCount(), - FirstDataBlock: fs.sb.FirstDataBlock(), - BlockSize: fs.sb.BlockSize(), - BlocksPerGroup: fs.sb.BlocksPerGroup(), - ClusterSize: fs.sb.ClusterSize(), - ClustersPerGroup: fs.sb.ClustersPerGroup(), - InodeSize: fs.sb.InodeSize(), - InodesPerGroup: fs.sb.InodesPerGroup(), - BgDescSize: fs.sb.BgDescSize(), - Magic: fs.sb.Magic(), - Revision: fs.sb.Revision(), - CompatFeatures: fs.sb.CompatibleFeatures(), - IncompatFeatures: fs.sb.IncompatibleFeatures(), - RoCompatFeatures: fs.sb.ReadOnlyCompatibleFeatures(), - } - gotNumBgs := len(fs.bgs) - gotBgs := make([]bg, gotNumBgs) - for i := 0; i < gotNumBgs; i++ { - gotBgs[i].InodeTable = fs.bgs[i].InodeTable() - gotBgs[i].BlockBitmap = fs.bgs[i].BlockBitmap() - gotBgs[i].InodeBitmap = fs.bgs[i].InodeBitmap() - gotBgs[i].ExclusionBitmap = fs.bgs[i].ExclusionBitmap() - gotBgs[i].Flags = fs.bgs[i].Flags() - - totalFreeInodes += fs.bgs[i].FreeInodesCount() - totalFreeBlocks += uint64(fs.bgs[i].FreeBlocksCount()) - } - - if diff := cmp.Diff(gotSb, test.wantSb); diff != "" { - t.Errorf("superblock mismatch (-want +got):\n%s", diff) - } - - if diff := cmp.Diff(gotBgs, test.wantBgs); diff != "" { - t.Errorf("block group descriptors mismatch (-want +got):\n%s", diff) - } - - if diff := cmp.Diff(totalFreeInodes, fs.sb.FreeInodesCount()); diff != "" { - t.Errorf("total free inodes mismatch (-want +got):\n%s", diff) - } - - if diff := cmp.Diff(totalFreeBlocks, fs.sb.FreeBlocksCount()); diff != "" { - t.Errorf("total free blocks mismatch (-want +got):\n%s", diff) - } - }) - } -} diff --git a/pkg/sentry/fsimpl/ext/extent_file.go b/pkg/sentry/fsimpl/ext/extent_file.go deleted file mode 100644 index 11dcc0346..000000000 --- a/pkg/sentry/fsimpl/ext/extent_file.go +++ /dev/null @@ -1,237 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - "sort" - - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/syserror" -) - -// extentFile is a type of regular file which uses extents to store file data. -type extentFile struct { - regFile regularFile - - // root is the root extent node. This lives in the 60 byte diskInode.Data(). - // Immutable. - root disklayout.ExtentNode -} - -// Compiles only if extentFile implements io.ReaderAt. -var _ io.ReaderAt = (*extentFile)(nil) - -// newExtentFile is the extent file constructor. It reads the entire extent -// tree into memory. -// TODO(b/134676337): Build extent tree on demand to reduce memory usage. -func newExtentFile(regFile regularFile) (*extentFile, error) { - file := &extentFile{regFile: regFile} - file.regFile.impl = file - err := file.buildExtTree() - if err != nil { - return nil, err - } - return file, nil -} - -// buildExtTree builds the extent tree by reading it from disk by doing -// running a simple DFS. It first reads the root node from the inode struct in -// memory. Then it recursively builds the rest of the tree by reading it off -// disk. -// -// Precondition: inode flag InExtents must be set. -func (f *extentFile) buildExtTree() error { - rootNodeData := f.regFile.inode.diskInode.Data() - - binary.Unmarshal(rootNodeData[:disklayout.ExtentHeaderSize], binary.LittleEndian, &f.root.Header) - - // Root node can not have more than 4 entries: 60 bytes = 1 header + 4 entries. - if f.root.Header.NumEntries > 4 { - // read(2) specifies that EINVAL should be returned if the file is unsuitable - // for reading. - return syserror.EINVAL - } - - f.root.Entries = make([]disklayout.ExtentEntryPair, f.root.Header.NumEntries) - for i, off := uint16(0), disklayout.ExtentEntrySize; i < f.root.Header.NumEntries; i, off = i+1, off+disklayout.ExtentEntrySize { - var curEntry disklayout.ExtentEntry - if f.root.Header.Height == 0 { - // Leaf node. - curEntry = &disklayout.Extent{} - } else { - // Internal node. - curEntry = &disklayout.ExtentIdx{} - } - binary.Unmarshal(rootNodeData[off:off+disklayout.ExtentEntrySize], binary.LittleEndian, curEntry) - f.root.Entries[i].Entry = curEntry - } - - // If this node is internal, perform DFS. - if f.root.Header.Height > 0 { - for i := uint16(0); i < f.root.Header.NumEntries; i++ { - var err error - if f.root.Entries[i].Node, err = f.buildExtTreeFromDisk(f.root.Entries[i].Entry); err != nil { - return err - } - } - } - - return nil -} - -// buildExtTreeFromDisk reads the extent tree nodes from disk and recursively -// builds the tree. Performs a simple DFS. It returns the ExtentNode pointed to -// by the ExtentEntry. -func (f *extentFile) buildExtTreeFromDisk(entry disklayout.ExtentEntry) (*disklayout.ExtentNode, error) { - var header disklayout.ExtentHeader - off := entry.PhysicalBlock() * f.regFile.inode.blkSize - err := readFromDisk(f.regFile.inode.fs.dev, int64(off), &header) - if err != nil { - return nil, err - } - - entries := make([]disklayout.ExtentEntryPair, header.NumEntries) - for i, off := uint16(0), off+disklayout.ExtentEntrySize; i < header.NumEntries; i, off = i+1, off+disklayout.ExtentEntrySize { - var curEntry disklayout.ExtentEntry - if header.Height == 0 { - // Leaf node. - curEntry = &disklayout.Extent{} - } else { - // Internal node. - curEntry = &disklayout.ExtentIdx{} - } - - err := readFromDisk(f.regFile.inode.fs.dev, int64(off), curEntry) - if err != nil { - return nil, err - } - entries[i].Entry = curEntry - } - - // If this node is internal, perform DFS. - if header.Height > 0 { - for i := uint16(0); i < header.NumEntries; i++ { - var err error - entries[i].Node, err = f.buildExtTreeFromDisk(entries[i].Entry) - if err != nil { - return nil, err - } - } - } - - return &disklayout.ExtentNode{header, entries}, nil -} - -// ReadAt implements io.ReaderAt.ReadAt. -func (f *extentFile) ReadAt(dst []byte, off int64) (int, error) { - if len(dst) == 0 { - return 0, nil - } - - if off < 0 { - return 0, syserror.EINVAL - } - - if uint64(off) >= f.regFile.inode.diskInode.Size() { - return 0, io.EOF - } - - n, err := f.read(&f.root, uint64(off), dst) - if n < len(dst) && err == nil { - err = io.EOF - } - return n, err -} - -// read is the recursive step of extentFile.ReadAt which traverses the extent -// tree from the node passed and reads file data. -func (f *extentFile) read(node *disklayout.ExtentNode, off uint64, dst []byte) (int, error) { - // Perform a binary search for the node covering bytes starting at r.fileOff. - // A highly fragmented filesystem can have upto 340 entries and so linear - // search should be avoided. Finds the first entry which does not cover the - // file block we want and subtracts 1 to get the desired index. - fileBlk := uint32(off / f.regFile.inode.blkSize) - n := len(node.Entries) - found := sort.Search(n, func(i int) bool { - return node.Entries[i].Entry.FileBlock() > fileBlk - }) - 1 - - // We should be in this recursive step only if the data we want exists under - // the current node. - if found < 0 { - panic("searching for a file block in an extent entry which does not cover it") - } - - read := 0 - toRead := len(dst) - var curR int - var err error - for i := found; i < n && read < toRead; i++ { - if node.Header.Height == 0 { - curR, err = f.readFromExtent(node.Entries[i].Entry.(*disklayout.Extent), off, dst[read:]) - } else { - curR, err = f.read(node.Entries[i].Node, off, dst[read:]) - } - - read += curR - off += uint64(curR) - if err != nil { - return read, err - } - } - - return read, nil -} - -// readFromExtent reads file data from the extent. It takes advantage of the -// sequential nature of extents and reads file data from multiple blocks in one -// call. -// -// A non-nil error indicates that this is a partial read and there is probably -// more to read from this extent. The caller should propagate the error upward -// and not move to the next extent in the tree. -// -// A subsequent call to extentReader.Read should continue reading from where we -// left off as expected. -func (f *extentFile) readFromExtent(ex *disklayout.Extent, off uint64, dst []byte) (int, error) { - curFileBlk := uint32(off / f.regFile.inode.blkSize) - exFirstFileBlk := ex.FileBlock() - exLastFileBlk := exFirstFileBlk + uint32(ex.Length) // This is exclusive. - - // We should be in this recursive step only if the data we want exists under - // the current extent. - if curFileBlk < exFirstFileBlk || exLastFileBlk <= curFileBlk { - panic("searching for a file block in an extent which does not cover it") - } - - curPhyBlk := uint64(curFileBlk-exFirstFileBlk) + ex.PhysicalBlock() - readStart := curPhyBlk*f.regFile.inode.blkSize + (off % f.regFile.inode.blkSize) - - endPhyBlk := ex.PhysicalBlock() + uint64(ex.Length) - extentEnd := endPhyBlk * f.regFile.inode.blkSize // This is exclusive. - - toRead := int(extentEnd - readStart) - if len(dst) < toRead { - toRead = len(dst) - } - - n, _ := f.regFile.inode.fs.dev.ReadAt(dst[:toRead], int64(readStart)) - if n < toRead { - return n, syserror.EIO - } - return n, nil -} diff --git a/pkg/sentry/fsimpl/ext/extent_test.go b/pkg/sentry/fsimpl/ext/extent_test.go deleted file mode 100644 index a2382daa3..000000000 --- a/pkg/sentry/fsimpl/ext/extent_test.go +++ /dev/null @@ -1,267 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "bytes" - "math/rand" - "testing" - - "github.com/google/go-cmp/cmp" - "github.com/google/go-cmp/cmp/cmpopts" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" -) - -const ( - // mockExtentBlkSize is the mock block size used for testing. - // No block has more than 1 header + 4 entries. - mockExtentBlkSize = uint64(64) -) - -// The tree described below looks like: -// -// 0.{Head}[Idx][Idx] -// / \ -// / \ -// 1.{Head}[Ext][Ext] 2.{Head}[Idx] -// / | \ -// [Phy] [Phy, Phy] 3.{Head}[Ext] -// | -// [Phy, Phy, Phy] -// -// Legend: -// - Head = ExtentHeader -// - Idx = ExtentIdx -// - Ext = Extent -// - Phy = Physical Block -// -// Please note that ext4 might not construct extent trees looking like this. -// This is purely for testing the tree traversal logic. -var ( - node3 = &disklayout.ExtentNode{ - Header: disklayout.ExtentHeader{ - Magic: disklayout.ExtentMagic, - NumEntries: 1, - MaxEntries: 4, - Height: 0, - }, - Entries: []disklayout.ExtentEntryPair{ - { - Entry: &disklayout.Extent{ - FirstFileBlock: 3, - Length: 3, - StartBlockLo: 6, - }, - Node: nil, - }, - }, - } - - node2 = &disklayout.ExtentNode{ - Header: disklayout.ExtentHeader{ - Magic: disklayout.ExtentMagic, - NumEntries: 1, - MaxEntries: 4, - Height: 1, - }, - Entries: []disklayout.ExtentEntryPair{ - { - Entry: &disklayout.ExtentIdx{ - FirstFileBlock: 3, - ChildBlockLo: 2, - }, - Node: node3, - }, - }, - } - - node1 = &disklayout.ExtentNode{ - Header: disklayout.ExtentHeader{ - Magic: disklayout.ExtentMagic, - NumEntries: 2, - MaxEntries: 4, - Height: 0, - }, - Entries: []disklayout.ExtentEntryPair{ - { - Entry: &disklayout.Extent{ - FirstFileBlock: 0, - Length: 1, - StartBlockLo: 3, - }, - Node: nil, - }, - { - Entry: &disklayout.Extent{ - FirstFileBlock: 1, - Length: 2, - StartBlockLo: 4, - }, - Node: nil, - }, - }, - } - - node0 = &disklayout.ExtentNode{ - Header: disklayout.ExtentHeader{ - Magic: disklayout.ExtentMagic, - NumEntries: 2, - MaxEntries: 4, - Height: 2, - }, - Entries: []disklayout.ExtentEntryPair{ - { - Entry: &disklayout.ExtentIdx{ - FirstFileBlock: 0, - ChildBlockLo: 0, - }, - Node: node1, - }, - { - Entry: &disklayout.ExtentIdx{ - FirstFileBlock: 3, - ChildBlockLo: 1, - }, - Node: node2, - }, - }, - } -) - -// TestExtentReader stress tests extentReader functionality. It performs random -// length reads from all possible positions in the extent tree. -func TestExtentReader(t *testing.T) { - mockExtentFile, want := extentTreeSetUp(t, node0) - n := len(want) - - for from := 0; from < n; from++ { - got := make([]byte, n-from) - - if read, err := mockExtentFile.ReadAt(got, int64(from)); err != nil { - t.Fatalf("file read operation from offset %d to %d only read %d bytes: %v", from, n, read, err) - } - - if diff := cmp.Diff(got, want[from:]); diff != "" { - t.Fatalf("file data from offset %d to %d mismatched (-want +got):\n%s", from, n, diff) - } - } -} - -// TestBuildExtentTree tests the extent tree building logic. -func TestBuildExtentTree(t *testing.T) { - mockExtentFile, _ := extentTreeSetUp(t, node0) - - opt := cmpopts.IgnoreUnexported(disklayout.ExtentIdx{}, disklayout.ExtentHeader{}) - if diff := cmp.Diff(&mockExtentFile.root, node0, opt); diff != "" { - t.Errorf("extent tree mismatch (-want +got):\n%s", diff) - } -} - -// extentTreeSetUp writes the passed extent tree to a mock disk as an extent -// tree. It also constucts a mock extent file with the same tree built in it. -// It also writes random data file data and returns it. -func extentTreeSetUp(t *testing.T, root *disklayout.ExtentNode) (*extentFile, []byte) { - t.Helper() - - mockDisk := make([]byte, mockExtentBlkSize*10) - mockExtentFile := &extentFile{ - regFile: regularFile{ - inode: inode{ - fs: &filesystem{ - dev: bytes.NewReader(mockDisk), - }, - diskInode: &disklayout.InodeNew{ - InodeOld: disklayout.InodeOld{ - SizeLo: uint32(mockExtentBlkSize) * getNumPhyBlks(root), - }, - }, - blkSize: mockExtentBlkSize, - }, - }, - } - - fileData := writeTree(&mockExtentFile.regFile.inode, mockDisk, node0, mockExtentBlkSize) - - if err := mockExtentFile.buildExtTree(); err != nil { - t.Fatalf("inode.buildExtTree failed: %v", err) - } - return mockExtentFile, fileData -} - -// writeTree writes the tree represented by `root` to the inode and disk. It -// also writes random file data on disk. -func writeTree(in *inode, disk []byte, root *disklayout.ExtentNode, mockExtentBlkSize uint64) []byte { - rootData := binary.Marshal(nil, binary.LittleEndian, root.Header) - for _, ep := range root.Entries { - rootData = binary.Marshal(rootData, binary.LittleEndian, ep.Entry) - } - - copy(in.diskInode.Data(), rootData) - - var fileData []byte - for _, ep := range root.Entries { - if root.Header.Height == 0 { - fileData = append(fileData, writeFileDataToExtent(disk, ep.Entry.(*disklayout.Extent))...) - } else { - fileData = append(fileData, writeTreeToDisk(disk, ep)...) - } - } - return fileData -} - -// writeTreeToDisk is the recursive step for writeTree which writes the tree -// on the disk only. Also writes random file data on disk. -func writeTreeToDisk(disk []byte, curNode disklayout.ExtentEntryPair) []byte { - nodeData := binary.Marshal(nil, binary.LittleEndian, curNode.Node.Header) - for _, ep := range curNode.Node.Entries { - nodeData = binary.Marshal(nodeData, binary.LittleEndian, ep.Entry) - } - - copy(disk[curNode.Entry.PhysicalBlock()*mockExtentBlkSize:], nodeData) - - var fileData []byte - for _, ep := range curNode.Node.Entries { - if curNode.Node.Header.Height == 0 { - fileData = append(fileData, writeFileDataToExtent(disk, ep.Entry.(*disklayout.Extent))...) - } else { - fileData = append(fileData, writeTreeToDisk(disk, ep)...) - } - } - return fileData -} - -// writeFileDataToExtent writes random bytes to the blocks on disk that the -// passed extent points to. -func writeFileDataToExtent(disk []byte, ex *disklayout.Extent) []byte { - phyExStartBlk := ex.PhysicalBlock() - phyExStartOff := phyExStartBlk * mockExtentBlkSize - phyExEndOff := phyExStartOff + uint64(ex.Length)*mockExtentBlkSize - rand.Read(disk[phyExStartOff:phyExEndOff]) - return disk[phyExStartOff:phyExEndOff] -} - -// getNumPhyBlks returns the number of physical blocks covered under the node. -func getNumPhyBlks(node *disklayout.ExtentNode) uint32 { - var res uint32 - for _, ep := range node.Entries { - if node.Header.Height == 0 { - res += uint32(ep.Entry.(*disklayout.Extent).Length) - } else { - res += getNumPhyBlks(ep.Node) - } - } - return res -} diff --git a/pkg/sentry/fsimpl/ext/file_description.go b/pkg/sentry/fsimpl/ext/file_description.go deleted file mode 100644 index 92f7da40d..000000000 --- a/pkg/sentry/fsimpl/ext/file_description.go +++ /dev/null @@ -1,64 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// fileDescription is embedded by ext implementations of -// vfs.FileDescriptionImpl. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl -} - -func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) -} - -func (fd *fileDescription) inode() *inode { - return fd.vfsfd.Dentry().Impl().(*dentry).inode -} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - var stat linux.Statx - fd.inode().statTo(&stat) - return stat, nil -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - if opts.Stat.Mask == 0 { - return nil - } - return syserror.EPERM -} - -// SetStat implements vfs.FileDescriptionImpl.StatFS. -func (fd *fileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { - var stat linux.Statfs - fd.filesystem().statTo(&stat) - return stat, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *fileDescription) Sync(ctx context.Context) error { - return nil -} diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go deleted file mode 100644 index 8497be615..000000000 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ /dev/null @@ -1,507 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "errors" - "io" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" -) - -var ( - // errResolveDirent indicates that the vfs.ResolvingPath.Component() does - // not exist on the dentry tree but does exist on disk. So it has to be read in - // using the in-memory dirent and added to the dentry tree. Usually indicates - // the need to lock filesystem.mu for writing. - errResolveDirent = errors.New("resolve path component using dirent") -) - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - vfsfs vfs.Filesystem - - // mu serializes changes to the Dentry tree. - mu sync.RWMutex - - // dev represents the underlying fs device. It does not require protection - // because io.ReaderAt permits concurrent read calls to it. It translates to - // the pread syscall which passes on the read request directly to the device - // driver. Device drivers are intelligent in serving multiple concurrent read - // requests in the optimal order (taking locality into consideration). - dev io.ReaderAt - - // inodeCache maps absolute inode numbers to the corresponding Inode struct. - // Inodes should be removed from this once their reference count hits 0. - // - // Protected by mu because most additions (see IterDirents) and all removals - // from this corresponds to a change in the dentry tree. - inodeCache map[uint32]*inode - - // sb represents the filesystem superblock. Immutable after initialization. - sb disklayout.SuperBlock - - // bgs represents all the block group descriptors for the filesystem. - // Immutable after initialization. - bgs []disklayout.BlockGroup -} - -// Compiles only if filesystem implements vfs.FilesystemImpl. -var _ vfs.FilesystemImpl = (*filesystem)(nil) - -// stepLocked resolves rp.Component() in parent directory vfsd. The write -// parameter passed tells if the caller has acquired filesystem.mu for writing -// or not. If set to true, an existing inode on disk can be added to the dentry -// tree if not present already. -// -// stepLocked is loosely analogous to fs/namei.c:walk_component(). -// -// Preconditions: -// - filesystem.mu must be locked (for writing if write param is true). -// - !rp.Done(). -// - inode == vfsd.Impl().(*Dentry).inode. -func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write bool) (*vfs.Dentry, *inode, error) { - if !inode.isDir() { - return nil, nil, syserror.ENOTDIR - } - if err := inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, nil, err - } - - for { - nextVFSD, err := rp.ResolveComponent(vfsd) - if err != nil { - return nil, nil, err - } - if nextVFSD == nil { - // Since the Dentry tree is not the sole source of truth for extfs, if it's - // not in the Dentry tree, it might need to be pulled from disk. - childDirent, ok := inode.impl.(*directory).childMap[rp.Component()] - if !ok { - // The underlying inode does not exist on disk. - return nil, nil, syserror.ENOENT - } - - if !write { - // filesystem.mu must be held for writing to add to the dentry tree. - return nil, nil, errResolveDirent - } - - // Create and add the component's dirent to the dentry tree. - fs := rp.Mount().Filesystem().Impl().(*filesystem) - childInode, err := fs.getOrCreateInodeLocked(childDirent.diskDirent.Inode()) - if err != nil { - return nil, nil, err - } - // incRef because this is being added to the dentry tree. - childInode.incRef() - child := newDentry(childInode) - vfsd.InsertChild(&child.vfsd, rp.Component()) - - // Continue as usual now that nextVFSD is not nil. - nextVFSD = &child.vfsd - } - nextInode := nextVFSD.Impl().(*dentry).inode - if nextInode.isSymlink() && rp.ShouldFollowSymlink() { - if err := rp.HandleSymlink(inode.impl.(*symlink).target); err != nil { - return nil, nil, err - } - continue - } - rp.Advance() - return nextVFSD, nextInode, nil - } -} - -// walkLocked resolves rp to an existing file. The write parameter -// passed tells if the caller has acquired filesystem.mu for writing or not. -// If set to true, additions can be made to the dentry tree while walking. -// If errResolveDirent is returned, the walk needs to be continued with an -// upgraded filesystem.mu. -// -// walkLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). -// -// Preconditions: -// - filesystem.mu must be locked (for writing if write param is true). -func walkLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { - vfsd := rp.Start() - inode := vfsd.Impl().(*dentry).inode - for !rp.Done() { - var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode, write) - if err != nil { - return nil, nil, err - } - } - if rp.MustBeDir() && !inode.isDir() { - return nil, nil, syserror.ENOTDIR - } - return vfsd, inode, nil -} - -// walkParentLocked resolves all but the last path component of rp to an -// existing directory. It does not check that the returned directory is -// searchable by the provider of rp. The write parameter passed tells if the -// caller has acquired filesystem.mu for writing or not. If set to true, -// additions can be made to the dentry tree while walking. -// If errResolveDirent is returned, the walk needs to be continued with an -// upgraded filesystem.mu. -// -// walkParentLocked is loosely analogous to Linux's fs/namei.c:path_parentat(). -// -// Preconditions: -// - filesystem.mu must be locked (for writing if write param is true). -// - !rp.Done(). -func walkParentLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { - vfsd := rp.Start() - inode := vfsd.Impl().(*dentry).inode - for !rp.Final() { - var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode, write) - if err != nil { - return nil, nil, err - } - } - if !inode.isDir() { - return nil, nil, syserror.ENOTDIR - } - return vfsd, inode, nil -} - -// walk resolves rp to an existing file. If parent is set to true, it resolves -// the rp till the parent of the last component which should be an existing -// directory. If parent is false then resolves rp entirely. Attemps to resolve -// the path as far as it can with a read lock and upgrades the lock if needed. -func (fs *filesystem) walk(rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *inode, error) { - var ( - vfsd *vfs.Dentry - inode *inode - err error - ) - - // Try walking with the hopes that all dentries have already been pulled out - // of disk. This reduces congestion (allows concurrent walks). - fs.mu.RLock() - if parent { - vfsd, inode, err = walkParentLocked(rp, false) - } else { - vfsd, inode, err = walkLocked(rp, false) - } - fs.mu.RUnlock() - - if err == errResolveDirent { - // Upgrade lock and continue walking. Lock upgrading in the middle of the - // walk is fine as this is a read only filesystem. - fs.mu.Lock() - if parent { - vfsd, inode, err = walkParentLocked(rp, true) - } else { - vfsd, inode, err = walkLocked(rp, true) - } - fs.mu.Unlock() - } - - return vfsd, inode, err -} - -// getOrCreateInodeLocked gets the inode corresponding to the inode number passed in. -// It creates a new one with the given inode number if one does not exist. -// The caller must increment the ref count if adding this to the dentry tree. -// -// Precondition: must be holding fs.mu for writing. -func (fs *filesystem) getOrCreateInodeLocked(inodeNum uint32) (*inode, error) { - if in, ok := fs.inodeCache[inodeNum]; ok { - return in, nil - } - - in, err := newInode(fs, inodeNum) - if err != nil { - return nil, err - } - - fs.inodeCache[inodeNum] = in - return in, nil -} - -// statTo writes the statfs fields to the output parameter. -func (fs *filesystem) statTo(stat *linux.Statfs) { - stat.Type = uint64(fs.sb.Magic()) - stat.BlockSize = int64(fs.sb.BlockSize()) - stat.Blocks = fs.sb.BlocksCount() - stat.BlocksFree = fs.sb.FreeBlocksCount() - stat.BlocksAvailable = fs.sb.FreeBlocksCount() - stat.Files = uint64(fs.sb.InodesCount()) - stat.FilesFree = uint64(fs.sb.FreeInodesCount()) - stat.NameLength = disklayout.MaxFileName - stat.FragmentSize = int64(fs.sb.BlockSize()) - // TODO(b/134676337): Set Statfs.Flags and Statfs.FSID. -} - -// AccessAt implements vfs.Filesystem.Impl.AccessAt. -func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - _, inode, err := fs.walk(rp, false) - if err != nil { - return err - } - return inode.checkPermissions(rp.Credentials(), ats) -} - -// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. -func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - vfsd, inode, err := fs.walk(rp, false) - if err != nil { - return nil, err - } - - if opts.CheckSearchable { - if !inode.isDir() { - return nil, syserror.ENOTDIR - } - if err := inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - } - - inode.incRef() - return vfsd, nil -} - -// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. -func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - vfsd, inode, err := fs.walk(rp, true) - if err != nil { - return nil, err - } - inode.incRef() - return vfsd, nil -} - -// OpenAt implements vfs.FilesystemImpl.OpenAt. -func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - vfsd, inode, err := fs.walk(rp, false) - if err != nil { - return nil, err - } - - // EROFS is returned if write access is needed. - if vfs.MayWriteFileWithOpenFlags(opts.Flags) || opts.Flags&(linux.O_CREAT|linux.O_EXCL|linux.O_TMPFILE) != 0 { - return nil, syserror.EROFS - } - return inode.open(rp, vfsd, &opts) -} - -// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. -func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - _, inode, err := fs.walk(rp, false) - if err != nil { - return "", err - } - symlink, ok := inode.impl.(*symlink) - if !ok { - return "", syserror.EINVAL - } - return symlink.target, nil -} - -// StatAt implements vfs.FilesystemImpl.StatAt. -func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - _, inode, err := fs.walk(rp, false) - if err != nil { - return linux.Statx{}, err - } - var stat linux.Statx - inode.statTo(&stat) - return stat, nil -} - -// StatFSAt implements vfs.FilesystemImpl.StatFSAt. -func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - if _, _, err := fs.walk(rp, false); err != nil { - return linux.Statfs{}, err - } - - var stat linux.Statfs - fs.statTo(&stat) - return stat, nil -} - -// Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() {} - -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *filesystem) Sync(ctx context.Context) error { - // This is a readonly filesystem for now. - return nil -} - -// The vfs.FilesystemImpl functions below return EROFS because their respective -// man pages say that EROFS must be returned if the path resolves to a file on -// this read-only filesystem. - -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - if rp.Done() { - return syserror.EEXIST - } - - if _, _, err := fs.walk(rp, true); err != nil { - return err - } - - return syserror.EROFS -} - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - if rp.Done() { - return syserror.EEXIST - } - - if _, _, err := fs.walk(rp, true); err != nil { - return err - } - - return syserror.EROFS -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - if rp.Done() { - return syserror.EEXIST - } - - _, _, err := fs.walk(rp, true) - if err != nil { - return err - } - - return syserror.EROFS -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { - if rp.Done() { - return syserror.ENOENT - } - - _, _, err := fs.walk(rp, false) - if err != nil { - return err - } - - return syserror.EROFS -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - _, inode, err := fs.walk(rp, false) - if err != nil { - return err - } - - if !inode.isDir() { - return syserror.ENOTDIR - } - - return syserror.EROFS -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - _, _, err := fs.walk(rp, false) - if err != nil { - return err - } - - return syserror.EROFS -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - if rp.Done() { - return syserror.EEXIST - } - - _, _, err := fs.walk(rp, true) - if err != nil { - return err - } - - return syserror.EROFS -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - _, inode, err := fs.walk(rp, false) - if err != nil { - return err - } - - if inode.isDir() { - return syserror.EISDIR - } - - return syserror.EROFS -} - -// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { - _, _, err := fs.walk(rp, false) - if err != nil { - return nil, err - } - return nil, syserror.ENOTSUP -} - -// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { - _, _, err := fs.walk(rp, false) - if err != nil { - return "", err - } - return "", syserror.ENOTSUP -} - -// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. -func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - _, _, err := fs.walk(rp, false) - if err != nil { - return err - } - return syserror.ENOTSUP -} - -// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. -func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - _, _, err := fs.walk(rp, false) - if err != nil { - return err - } - return syserror.ENOTSUP -} - -// PrependPath implements vfs.FilesystemImpl.PrependPath. -func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) -} diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go deleted file mode 100644 index a39a37318..000000000 --- a/pkg/sentry/fsimpl/ext/inode.go +++ /dev/null @@ -1,220 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "fmt" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// inode represents an ext inode. -// -// inode uses the same inheritance pattern that pkg/sentry/vfs structures use. -// This has been done to increase memory locality. -// -// Implementations: -// inode -- -// |-- dir -// |-- symlink -// |-- regular-- -// |-- extent file -// |-- block map file -type inode struct { - // refs is a reference count. refs is accessed using atomic memory operations. - refs int64 - - // fs is the containing filesystem. - fs *filesystem - - // inodeNum is the inode number of this inode on disk. This is used to - // identify inodes within the ext filesystem. - inodeNum uint32 - - // blkSize is the fs data block size. Same as filesystem.sb.BlockSize(). - blkSize uint64 - - // diskInode gives us access to the inode struct on disk. Immutable. - diskInode disklayout.Inode - - // This is immutable. The first field of the implementations must have inode - // as the first field to ensure temporality. - impl interface{} -} - -// incRef increments the inode ref count. -func (in *inode) incRef() { - atomic.AddInt64(&in.refs, 1) -} - -// tryIncRef tries to increment the ref count. Returns true if successful. -func (in *inode) tryIncRef() bool { - for { - refs := atomic.LoadInt64(&in.refs) - if refs == 0 { - return false - } - if atomic.CompareAndSwapInt64(&in.refs, refs, refs+1) { - return true - } - } -} - -// decRef decrements the inode ref count and releases the inode resources if -// the ref count hits 0. -// -// Precondition: Must have locked filesystem.mu. -func (in *inode) decRef() { - if refs := atomic.AddInt64(&in.refs, -1); refs == 0 { - delete(in.fs.inodeCache, in.inodeNum) - } else if refs < 0 { - panic("ext.inode.decRef() called without holding a reference") - } -} - -// newInode is the inode constructor. Reads the inode off disk. Identifies -// inodes based on the absolute inode number on disk. -func newInode(fs *filesystem, inodeNum uint32) (*inode, error) { - if inodeNum == 0 { - panic("inode number 0 on ext filesystems is not possible") - } - - inodeRecordSize := fs.sb.InodeSize() - var diskInode disklayout.Inode - if inodeRecordSize == disklayout.OldInodeSize { - diskInode = &disklayout.InodeOld{} - } else { - diskInode = &disklayout.InodeNew{} - } - - // Calculate where the inode is actually placed. - inodesPerGrp := fs.sb.InodesPerGroup() - blkSize := fs.sb.BlockSize() - inodeTableOff := fs.bgs[getBGNum(inodeNum, inodesPerGrp)].InodeTable() * blkSize - inodeOff := inodeTableOff + uint64(uint32(inodeRecordSize)*getBGOff(inodeNum, inodesPerGrp)) - - if err := readFromDisk(fs.dev, int64(inodeOff), diskInode); err != nil { - return nil, err - } - - // Build the inode based on its type. - inode := inode{ - fs: fs, - inodeNum: inodeNum, - blkSize: blkSize, - diskInode: diskInode, - } - - switch diskInode.Mode().FileType() { - case linux.ModeSymlink: - f, err := newSymlink(inode) - if err != nil { - return nil, err - } - return &f.inode, nil - case linux.ModeRegular: - f, err := newRegularFile(inode) - if err != nil { - return nil, err - } - return &f.inode, nil - case linux.ModeDirectory: - f, err := newDirectroy(inode, fs.sb.IncompatibleFeatures().DirentFileType) - if err != nil { - return nil, err - } - return &f.inode, nil - default: - // TODO(b/134676337): Return appropriate errors for sockets, pipes and devices. - return nil, syserror.EINVAL - } -} - -// open creates and returns a file description for the dentry passed in. -func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(opts) - if err := in.checkPermissions(rp.Credentials(), ats); err != nil { - return nil, err - } - mnt := rp.Mount() - switch in.impl.(type) { - case *regularFile: - var fd regularFileFD - if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case *directory: - // Can't open directories writably. This check is not necessary for a read - // only filesystem but will be required when write is implemented. - if ats&vfs.MayWrite != 0 { - return nil, syserror.EISDIR - } - var fd directoryFD - if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case *symlink: - if opts.Flags&linux.O_PATH == 0 { - // Can't open symlinks without O_PATH. - return nil, syserror.ELOOP - } - var fd symlinkFD - fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) - return &fd.vfsfd, nil - default: - panic(fmt.Sprintf("unknown inode type: %T", in.impl)) - } -} - -func (in *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { - return vfs.GenericCheckPermissions(creds, ats, in.diskInode.Mode(), in.diskInode.UID(), in.diskInode.GID()) -} - -// statTo writes the statx fields to the output parameter. -func (in *inode) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | - linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_SIZE | - linux.STATX_ATIME | linux.STATX_CTIME | linux.STATX_MTIME - stat.Blksize = uint32(in.blkSize) - stat.Mode = uint16(in.diskInode.Mode()) - stat.Nlink = uint32(in.diskInode.LinksCount()) - stat.UID = uint32(in.diskInode.UID()) - stat.GID = uint32(in.diskInode.GID()) - stat.Ino = uint64(in.inodeNum) - stat.Size = in.diskInode.Size() - stat.Atime = in.diskInode.AccessTime().StatxTimestamp() - stat.Ctime = in.diskInode.ChangeTime().StatxTimestamp() - stat.Mtime = in.diskInode.ModificationTime().StatxTimestamp() - // TODO(b/134676337): Set stat.Blocks which is the number of 512 byte blocks - // (including metadata blocks) required to represent this file. -} - -// getBGNum returns the block group number that a given inode belongs to. -func getBGNum(inodeNum uint32, inodesPerGrp uint32) uint32 { - return (inodeNum - 1) / inodesPerGrp -} - -// getBGOff returns the offset at which the given inode lives in the block -// group's inode table, i.e. the index of the inode in the inode table. -func getBGOff(inodeNum uint32, inodesPerGrp uint32) uint32 { - return (inodeNum - 1) % inodesPerGrp -} diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go deleted file mode 100644 index 30135ddb0..000000000 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ /dev/null @@ -1,159 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// regularFile represents a regular file's inode. This too follows the -// inheritance pattern prevelant in the vfs layer described in -// pkg/sentry/vfs/README.md. -type regularFile struct { - inode inode - - // This is immutable. The first field of fileReader implementations must be - // regularFile to ensure temporality. - // io.ReaderAt is more strict than io.Reader in the sense that a partial read - // is always accompanied by an error. If a read spans past the end of file, a - // partial read (within file range) is done and io.EOF is returned. - impl io.ReaderAt -} - -// newRegularFile is the regularFile constructor. It figures out what kind of -// file this is and initializes the fileReader. -func newRegularFile(inode inode) (*regularFile, error) { - regFile := regularFile{ - inode: inode, - } - - inodeFlags := inode.diskInode.Flags() - - if inodeFlags.Extents { - file, err := newExtentFile(regFile) - if err != nil { - return nil, err - } - - file.regFile.inode.impl = &file.regFile - return &file.regFile, nil - } - - file, err := newBlockMapFile(regFile) - if err != nil { - return nil, err - } - file.regFile.inode.impl = &file.regFile - return &file.regFile, nil -} - -func (in *inode) isRegular() bool { - _, ok := in.impl.(*regularFile) - return ok -} - -// directoryFD represents a directory file description. It implements -// vfs.FileDescriptionImpl. -type regularFileFD struct { - fileDescription - - // off is the file offset. off is accessed using atomic memory operations. - off int64 - - // offMu serializes operations that may mutate off. - offMu sync.Mutex -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() {} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - safeReader := safemem.FromIOReaderAt{ - ReaderAt: fd.inode().impl.(*regularFile).impl, - Offset: offset, - } - - // Copies data from disk directly into usermem without any intermediate - // allocations (if dst is converted into BlockSeq such that it does not need - // safe copying). - return dst.CopyOutFrom(ctx, safeReader) -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.offMu.Lock() - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - // write(2) specifies that EBADF must be returned if the fd is not open for - // writing. - return 0, syserror.EBADF -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.offMu.Lock() - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *regularFileFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - return syserror.ENOTDIR -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.offMu.Lock() - defer fd.offMu.Unlock() - switch whence { - case linux.SEEK_SET: - // Use offset as specified. - case linux.SEEK_CUR: - offset += fd.off - case linux.SEEK_END: - offset += int64(fd.inode().diskInode.Size()) - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - // TODO(b/134676337): Implement mmap(2). - return syserror.ENODEV -} diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go deleted file mode 100644 index 1447a4dc1..000000000 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ /dev/null @@ -1,111 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// symlink represents a symlink inode. -type symlink struct { - inode inode - target string // immutable -} - -// newSymlink is the symlink constructor. It reads out the symlink target from -// the inode (however it might have been stored). -func newSymlink(inode inode) (*symlink, error) { - var file *symlink - var link []byte - - // If the symlink target is lesser than 60 bytes, its stores in inode.Data(). - // Otherwise either extents or block maps will be used to store the link. - size := inode.diskInode.Size() - if size < 60 { - link = inode.diskInode.Data()[:size] - } else { - // Create a regular file out of this inode and read out the target. - regFile, err := newRegularFile(inode) - if err != nil { - return nil, err - } - - link = make([]byte, size) - if n, err := regFile.impl.ReadAt(link, 0); uint64(n) < size { - return nil, err - } - } - - file = &symlink{inode: inode, target: string(link)} - file.inode.impl = file - return file, nil -} - -func (in *inode) isSymlink() bool { - _, ok := in.impl.(*symlink) - return ok -} - -// symlinkFD represents a symlink file description and implements implements -// vfs.FileDescriptionImpl. which may only be used if open options contains -// O_PATH. For this reason most of the functions return EBADF. -type symlinkFD struct { - fileDescription -} - -// Compiles only if symlinkFD implements vfs.FileDescriptionImpl. -var _ vfs.FileDescriptionImpl = (*symlinkFD)(nil) - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *symlinkFD) Release() {} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *symlinkFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return 0, syserror.EBADF -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *symlinkFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return 0, syserror.EBADF -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *symlinkFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.EBADF -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *symlinkFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.EBADF -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *symlinkFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - return syserror.ENOTDIR -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *symlinkFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - return 0, syserror.EBADF -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *symlinkFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - return syserror.EBADF -} diff --git a/pkg/sentry/fsimpl/ext/utils.go b/pkg/sentry/fsimpl/ext/utils.go deleted file mode 100644 index d8b728f8c..000000000 --- a/pkg/sentry/fsimpl/ext/utils.go +++ /dev/null @@ -1,94 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/syserror" -) - -// readFromDisk performs a binary read from disk into the given struct from -// the absolute offset provided. -func readFromDisk(dev io.ReaderAt, abOff int64, v interface{}) error { - n := binary.Size(v) - buf := make([]byte, n) - if read, _ := dev.ReadAt(buf, abOff); read < int(n) { - return syserror.EIO - } - - binary.Unmarshal(buf, binary.LittleEndian, v) - return nil -} - -// readSuperBlock reads the SuperBlock from block group 0 in the underlying -// device. There are three versions of the superblock. This function identifies -// and returns the correct version. -func readSuperBlock(dev io.ReaderAt) (disklayout.SuperBlock, error) { - var sb disklayout.SuperBlock = &disklayout.SuperBlockOld{} - if err := readFromDisk(dev, disklayout.SbOffset, sb); err != nil { - return nil, err - } - if sb.Revision() == disklayout.OldRev { - return sb, nil - } - - sb = &disklayout.SuperBlock32Bit{} - if err := readFromDisk(dev, disklayout.SbOffset, sb); err != nil { - return nil, err - } - if !sb.IncompatibleFeatures().Is64Bit { - return sb, nil - } - - sb = &disklayout.SuperBlock64Bit{} - if err := readFromDisk(dev, disklayout.SbOffset, sb); err != nil { - return nil, err - } - return sb, nil -} - -// blockGroupsCount returns the number of block groups in the ext fs. -func blockGroupsCount(sb disklayout.SuperBlock) uint64 { - blocksCount := sb.BlocksCount() - blocksPerGroup := uint64(sb.BlocksPerGroup()) - - // Round up the result. float64 can compromise precision so do it manually. - return (blocksCount + blocksPerGroup - 1) / blocksPerGroup -} - -// readBlockGroups reads the block group descriptor table from block group 0 in -// the underlying device. -func readBlockGroups(dev io.ReaderAt, sb disklayout.SuperBlock) ([]disklayout.BlockGroup, error) { - bgCount := blockGroupsCount(sb) - bgdSize := uint64(sb.BgDescSize()) - is64Bit := sb.IncompatibleFeatures().Is64Bit - bgds := make([]disklayout.BlockGroup, bgCount) - - for i, off := uint64(0), uint64(sb.FirstDataBlock()+1)*sb.BlockSize(); i < bgCount; i, off = i+1, off+bgdSize { - if is64Bit { - bgds[i] = &disklayout.BlockGroup64Bit{} - } else { - bgds[i] = &disklayout.BlockGroup32Bit{} - } - - if err := readFromDisk(dev, int64(off), bgds[i]); err != nil { - return nil, err - } - } - return bgds, nil -} diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD deleted file mode 100644 index 4ba76a1e8..000000000 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ /dev/null @@ -1,55 +0,0 @@ -load("//tools:defs.bzl", "go_library") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -licenses(["notice"]) - -go_template_instance( - name = "dentry_list", - out = "dentry_list.go", - package = "gofer", - prefix = "dentry", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*dentry", - "Linker": "*dentry", - }, -) - -go_library( - name = "gofer", - srcs = [ - "dentry_list.go", - "directory.go", - "filesystem.go", - "gofer.go", - "handle.go", - "handle_unsafe.go", - "p9file.go", - "pagemath.go", - "regular_file.go", - "special_file.go", - "symlink.go", - "time.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fd", - "//pkg/fspath", - "//pkg/log", - "//pkg/p9", - "//pkg/safemem", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/unet", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go deleted file mode 100644 index 5dbfc6250..000000000 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ /dev/null @@ -1,194 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "sync" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -func (d *dentry) isDir() bool { - return d.fileType() == linux.S_IFDIR -} - -// Preconditions: d.dirMu must be locked. d.isDir(). fs.opts.interop != -// InteropModeShared. -func (d *dentry) cacheNegativeChildLocked(name string) { - if d.negativeChildren == nil { - d.negativeChildren = make(map[string]struct{}) - } - d.negativeChildren[name] = struct{}{} -} - -type directoryFD struct { - fileDescription - vfs.DirectoryFileDescriptionDefaultImpl - - mu sync.Mutex - off int64 - dirents []vfs.Dirent -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - fd.mu.Lock() - defer fd.mu.Unlock() - - if fd.dirents == nil { - ds, err := fd.dentry().getDirents(ctx) - if err != nil { - return err - } - fd.dirents = ds - } - - for fd.off < int64(len(fd.dirents)) { - if err := cb.Handle(fd.dirents[fd.off]); err != nil { - return err - } - fd.off++ - } - return nil -} - -// Preconditions: d.isDir(). There exists at least one directoryFD representing d. -func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { - // 9P2000.L's readdir does not specify behavior in the presence of - // concurrent mutation of an iterated directory, so implementations may - // duplicate or omit entries in this case, which violates POSIX semantics. - // Thus we read all directory entries while holding d.dirMu to exclude - // directory mutations. (Note that it is impossible for the client to - // exclude concurrent mutation from other remote filesystem users. Since - // there is no way to detect if the server has incorrectly omitted - // directory entries, we simply assume that the server is well-behaved - // under InteropModeShared.) This is inconsistent with Linux (which appears - // to assume that directory fids have the correct semantics, and translates - // struct file_operations::readdir calls directly to readdir RPCs), but is - // consistent with VFS1. - // - // NOTE(b/135560623): In particular, some gofer implementations may not - // retain state between calls to Readdir, so may not provide a coherent - // directory stream across in the presence of mutation. - - d.fs.renameMu.RLock() - defer d.fs.renameMu.RUnlock() - d.dirMu.Lock() - defer d.dirMu.Unlock() - if d.dirents != nil { - return d.dirents, nil - } - - // It's not clear if 9P2000.L's readdir is expected to return "." and "..", - // so we generate them here. - parent := d.vfsd.ParentOrSelf().Impl().(*dentry) - dirents := []vfs.Dirent{ - { - Name: ".", - Type: linux.DT_DIR, - Ino: d.ino, - NextOff: 1, - }, - { - Name: "..", - Type: uint8(atomic.LoadUint32(&parent.mode) >> 12), - Ino: parent.ino, - NextOff: 2, - }, - } - off := uint64(0) - const count = 64 * 1024 // for consistency with the vfs1 client - d.handleMu.RLock() - defer d.handleMu.RUnlock() - if !d.handleReadable { - // This should not be possible because a readable handle should have - // been opened when the calling directoryFD was opened. - panic("gofer.dentry.getDirents called without a readable handle") - } - for { - p9ds, err := d.handle.file.readdir(ctx, off, count) - if err != nil { - return nil, err - } - if len(p9ds) == 0 { - // Cache dirents for future directoryFDs if permitted. - if d.fs.opts.interop != InteropModeShared { - d.dirents = dirents - } - return dirents, nil - } - for _, p9d := range p9ds { - if p9d.Name == "." || p9d.Name == ".." { - continue - } - dirent := vfs.Dirent{ - Name: p9d.Name, - Ino: p9d.QID.Path, - NextOff: int64(len(dirents) + 1), - } - // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or - // DMSOCKET. - switch p9d.Type { - case p9.TypeSymlink: - dirent.Type = linux.DT_LNK - case p9.TypeDir: - dirent.Type = linux.DT_DIR - default: - dirent.Type = linux.DT_REG - } - dirents = append(dirents, dirent) - } - off = p9ds[len(p9ds)-1].Offset - } -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.mu.Lock() - defer fd.mu.Unlock() - - switch whence { - case linux.SEEK_SET: - if offset < 0 { - return 0, syserror.EINVAL - } - if offset == 0 { - // Ensure that the next call to fd.IterDirents() calls - // fd.dentry().getDirents(). - fd.dirents = nil - } - fd.off = offset - return fd.off, nil - case linux.SEEK_CUR: - offset += fd.off - if offset < 0 { - return 0, syserror.EINVAL - } - // Don't clear fd.dirents in this case, even if offset == 0. - fd.off = offset - return fd.off, nil - default: - return 0, syserror.EINVAL - } -} diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go deleted file mode 100644 index 1e43df9ec..000000000 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ /dev/null @@ -1,1115 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "sync" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *filesystem) Sync(ctx context.Context) error { - // Snapshot current dentries and special files. - fs.syncMu.Lock() - ds := make([]*dentry, 0, len(fs.dentries)) - for d := range fs.dentries { - ds = append(ds, d) - } - sffds := make([]*specialFileFD, 0, len(fs.specialFileFDs)) - for sffd := range fs.specialFileFDs { - sffds = append(sffds, sffd) - } - fs.syncMu.Unlock() - - // Return the first error we encounter, but sync everything we can - // regardless. - var retErr error - - // Sync regular files. - for _, d := range ds { - if !d.TryIncRef() { - continue - } - err := d.syncSharedHandle(ctx) - d.DecRef() - if err != nil && retErr == nil { - retErr = err - } - } - - // Sync special files, which may be writable but do not use dentry shared - // handles (so they won't be synced by the above). - for _, sffd := range sffds { - if !sffd.vfsfd.TryIncRef() { - continue - } - err := sffd.Sync(ctx) - sffd.vfsfd.DecRef() - if err != nil && retErr == nil { - retErr = err - } - } - - return retErr -} - -// maxFilenameLen is the maximum length of a filename. This is dictated by 9P's -// encoding of strings, which uses 2 bytes for the length prefix. -const maxFilenameLen = (1 << 16) - 1 - -// dentrySlicePool is a pool of *[]*dentry used to store dentries for which -// dentry.checkCachingLocked() must be called. The pool holds pointers to -// slices because Go lacks generics, so sync.Pool operates on interface{}, so -// every call to (what should be) sync.Pool<[]*dentry>.Put() allocates a copy -// of the slice header on the heap. -var dentrySlicePool = sync.Pool{ - New: func() interface{} { - ds := make([]*dentry, 0, 4) // arbitrary non-zero initial capacity - return &ds - }, -} - -func appendDentry(ds *[]*dentry, d *dentry) *[]*dentry { - if ds == nil { - ds = dentrySlicePool.Get().(*[]*dentry) - } - *ds = append(*ds, d) - return ds -} - -// Preconditions: ds != nil. -func putDentrySlice(ds *[]*dentry) { - // Allow dentries to be GC'd. - for i := range *ds { - (*ds)[i] = nil - } - *ds = (*ds)[:0] - dentrySlicePool.Put(ds) -} - -// stepLocked resolves rp.Component() to an existing file, starting from the -// given directory. -// -// Dentries which may become cached as a result of the traversal are appended -// to *ds. -// -// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. -// !rp.Done(). If fs.opts.interop == InteropModeShared, then d's cached -// metadata must be up to date. -func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { - if !d.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } -afterSymlink: - name := rp.Component() - if name == "." { - rp.Advance() - return d, nil - } - if name == ".." { - parentVFSD, err := rp.ResolveParent(&d.vfsd) - if err != nil { - return nil, err - } - parent := parentVFSD.Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // We must assume that parentVFSD is correct, because if d has been - // moved elsewhere in the remote filesystem so that its parent has - // changed, we have no way of determining its new parent's location - // in the filesystem. Get updated metadata for parentVFSD. - _, attrMask, attr, err := parent.file.getAttr(ctx, dentryAttrMask()) - if err != nil { - return nil, err - } - parent.updateFromP9Attrs(attrMask, &attr) - } - rp.Advance() - return parent, nil - } - childVFSD, err := rp.ResolveChild(&d.vfsd, name) - if err != nil { - return nil, err - } - // FIXME(jamieliu): Linux performs revalidation before mount lookup - // (fs/namei.c:lookup_fast() => __d_lookup_rcu(), d_revalidate(), - // __follow_mount_rcu()). - child, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, childVFSD, ds) - if err != nil { - return nil, err - } - if child == nil { - return nil, syserror.ENOENT - } - if child.isSymlink() && rp.ShouldFollowSymlink() { - target, err := child.readlink(ctx, rp.Mount()) - if err != nil { - return nil, err - } - if err := rp.HandleSymlink(target); err != nil { - return nil, err - } - goto afterSymlink // don't check the current directory again - } - rp.Advance() - return child, nil -} - -// revalidateChildLocked must be called after a call to parent.vfsd.Child(name) -// or vfs.ResolvingPath.ResolveChild(name) returns childVFSD (which may be -// nil) to verify that the returned child (or lack thereof) is correct. If no file -// exists at name, revalidateChildLocked returns (nil, nil). -// -// Preconditions: fs.renameMu must be locked. parent.dirMu must be locked. -// parent.isDir(). name is not "." or "..". -// -// Postconditions: If revalidateChildLocked returns a non-nil dentry, its -// cached metadata is up to date. -func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, childVFSD *vfs.Dentry, ds **[]*dentry) (*dentry, error) { - if childVFSD != nil && fs.opts.interop != InteropModeShared { - // We have a cached dentry that is assumed to be correct. - return childVFSD.Impl().(*dentry), nil - } - // We either don't have a cached dentry or need to verify that it's still - // correct, either of which requires a remote lookup. Check if this name is - // valid before performing the lookup. - if len(name) > maxFilenameLen { - return nil, syserror.ENAMETOOLONG - } - // Check if we've already cached this lookup with a negative result. - if _, ok := parent.negativeChildren[name]; ok { - return nil, nil - } - // Perform the remote lookup. - qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) - if err != nil && err != syserror.ENOENT { - return nil, err - } - if childVFSD != nil { - child := childVFSD.Impl().(*dentry) - if !file.isNil() && qid.Path == child.ino { - // The file at this path hasn't changed. Just update cached - // metadata. - file.close(ctx) - child.updateFromP9Attrs(attrMask, &attr) - return child, nil - } - // The file at this path has changed or no longer exists. Remove - // the stale dentry from the tree, and re-evaluate its caching - // status (i.e. if it has 0 references, drop it). - vfsObj.ForceDeleteDentry(childVFSD) - *ds = appendDentry(*ds, child) - childVFSD = nil - } - if file.isNil() { - // No file exists at this path now. Cache the negative lookup if - // allowed. - if fs.opts.interop != InteropModeShared { - parent.cacheNegativeChildLocked(name) - } - return nil, nil - } - // Create a new dentry representing the file. - child, err := fs.newDentry(ctx, file, qid, attrMask, &attr) - if err != nil { - file.close(ctx) - return nil, err - } - parent.IncRef() // reference held by child on its parent - parent.vfsd.InsertChild(&child.vfsd, name) - // For now, child has 0 references, so our caller should call - // child.checkCachingLocked(). - *ds = appendDentry(*ds, child) - return child, nil -} - -// walkParentDirLocked resolves all but the last path component of rp to an -// existing directory, starting from the given directory (which is usually -// rp.Start().Impl().(*dentry)). It does not check that the returned directory -// is searchable by the provider of rp. -// -// Preconditions: fs.renameMu must be locked. !rp.Done(). If fs.opts.interop == -// InteropModeShared, then d's cached metadata must be up to date. -func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { - for !rp.Final() { - d.dirMu.Lock() - next, err := fs.stepLocked(ctx, rp, d, ds) - d.dirMu.Unlock() - if err != nil { - return nil, err - } - d = next - } - if !d.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// resolveLocked resolves rp to an existing file. -// -// Preconditions: fs.renameMu must be locked. -func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) { - d := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for rp.Start() as required by fs.stepLocked(). - if err := d.updateFromGetattr(ctx); err != nil { - return nil, err - } - } - for !rp.Done() { - d.dirMu.Lock() - next, err := fs.stepLocked(ctx, rp, d, ds) - d.dirMu.Unlock() - if err != nil { - return nil, err - } - d = next - } - if rp.MustBeDir() && !d.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// doCreateAt checks that creating a file at rp is permitted, then invokes -// create to do so. -// -// Preconditions: !rp.Done(). For the final path component in rp, -// !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for start as required by - // fs.walkParentDirLocked(). - if err := start.updateFromGetattr(ctx); err != nil { - return err - } - } - parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) - if err != nil { - return err - } - if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - if parent.isDeleted() { - return syserror.ENOENT - } - name := rp.Component() - if name == "." || name == ".." { - return syserror.EEXIST - } - if len(name) > maxFilenameLen { - return syserror.ENAMETOOLONG - } - if !dir && rp.MustBeDir() { - return syserror.ENOENT - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - parent.dirMu.Lock() - defer parent.dirMu.Unlock() - if fs.opts.interop == InteropModeShared { - // The existence of a dentry at name would be inconclusive because the - // file it represents may have been deleted from the remote filesystem, - // so we would need to make an RPC to revalidate the dentry. Just - // attempt the file creation RPC instead. If a file does exist, the RPC - // will fail with EEXIST like we would have. If the RPC succeeds, and a - // stale dentry exists, the dentry will fail revalidation next time - // it's used. - return create(parent, name) - } - if parent.vfsd.Child(name) != nil { - return syserror.EEXIST - } - // No cached dentry exists; however, there might still be an existing file - // at name. As above, we attempt the file creation RPC anyway. - if err := create(parent, name); err != nil { - return err - } - parent.touchCMtime(ctx) - delete(parent.negativeChildren, name) - parent.dirents = nil - return nil -} - -// Preconditions: !rp.Done(). -func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for start as required by - // fs.walkParentDirLocked(). - if err := start.updateFromGetattr(ctx); err != nil { - return err - } - } - parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) - if err != nil { - return err - } - if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - - name := rp.Component() - if dir { - if name == "." { - return syserror.EINVAL - } - if name == ".." { - return syserror.ENOTEMPTY - } - } else { - if name == "." || name == ".." { - return syserror.EISDIR - } - } - vfsObj := rp.VirtualFilesystem() - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - parent.dirMu.Lock() - defer parent.dirMu.Unlock() - childVFSD := parent.vfsd.Child(name) - var child *dentry - // We only need a dentry representing the file at name if it can be a mount - // point. If childVFSD is nil, then it can't be a mount point. If childVFSD - // is non-nil but stale, the actual file can't be a mount point either; we - // detect this case by just speculatively calling PrepareDeleteDentry and - // only revalidating the dentry if that fails (indicating that the existing - // dentry is a mount point). - if childVFSD != nil { - child = childVFSD.Impl().(*dentry) - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, childVFSD, &ds) - if err != nil { - return err - } - if child != nil { - childVFSD = &child.vfsd - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - return err - } - } else { - childVFSD = nil - } - } - } else if _, ok := parent.negativeChildren[name]; ok { - return syserror.ENOENT - } - flags := uint32(0) - if dir { - if child != nil && !child.isDir() { - return syserror.ENOTDIR - } - flags = linux.AT_REMOVEDIR - } else { - if child != nil && child.isDir() { - return syserror.EISDIR - } - if rp.MustBeDir() { - return syserror.ENOTDIR - } - } - err = parent.file.unlinkAt(ctx, name, flags) - if err != nil { - if childVFSD != nil { - vfsObj.AbortDeleteDentry(childVFSD) - } - return err - } - if fs.opts.interop != InteropModeShared { - parent.touchCMtime(ctx) - if dir { - parent.decLinks() - } - parent.cacheNegativeChildLocked(name) - parent.dirents = nil - } - if child != nil { - child.setDeleted() - vfsObj.CommitDeleteDentry(childVFSD) - ds = appendDentry(ds, child) - } - return nil -} - -// renameMuRUnlockAndCheckCaching calls fs.renameMu.RUnlock(), then calls -// dentry.checkCachingLocked on all dentries in *ds with fs.renameMu locked for -// writing. -// -// ds is a pointer-to-pointer since defer evaluates its arguments immediately, -// but dentry slices are allocated lazily, and it's much easier to say "defer -// fs.renameMuRUnlockAndCheckCaching(&ds)" than "defer func() { -// fs.renameMuRUnlockAndCheckCaching(ds) }()" to work around this. -func (fs *filesystem) renameMuRUnlockAndCheckCaching(ds **[]*dentry) { - fs.renameMu.RUnlock() - if *ds == nil { - return - } - if len(**ds) != 0 { - fs.renameMu.Lock() - for _, d := range **ds { - d.checkCachingLocked() - } - fs.renameMu.Unlock() - } - putDentrySlice(*ds) -} - -func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { - if *ds == nil { - fs.renameMu.Unlock() - return - } - for _, d := range **ds { - d.checkCachingLocked() - } - fs.renameMu.Unlock() - putDentrySlice(*ds) -} - -// AccessAt implements vfs.Filesystem.Impl.AccessAt. -func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return err - } - return d.checkPermissions(creds, ats) -} - -// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. -func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return nil, err - } - if opts.CheckSearchable { - if !d.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - } - d.IncRef() - return &d.vfsd, nil -} - -// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. -func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for start as required by - // fs.walkParentDirLocked(). - if err := start.updateFromGetattr(ctx); err != nil { - return nil, err - } - } - d, err := fs.walkParentDirLocked(ctx, rp, start, &ds) - if err != nil { - return nil, err - } - d.IncRef() - return &d.vfsd, nil -} - -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, childName string) error { - if rp.Mount() != vd.Mount() { - return syserror.EXDEV - } - // 9P2000.L supports hard links, but we don't. - return syserror.EPERM - }) -} - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, name string) error { - creds := rp.Credentials() - if _, err := parent.file.mkdir(ctx, name, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)); err != nil { - return err - } - if fs.opts.interop != InteropModeShared { - parent.incLinks() - } - return nil - }) -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { - creds := rp.Credentials() - _, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) - return err - }) -} - -// OpenAt implements vfs.FilesystemImpl.OpenAt. -func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - // Reject O_TMPFILE, which is not supported; supporting it correctly in the - // presence of other remote filesystem users requires remote filesystem - // support, and it isn't clear that there's any way to implement this in - // 9P. - if opts.Flags&linux.O_TMPFILE != 0 { - return nil, syserror.EOPNOTSUPP - } - mayCreate := opts.Flags&linux.O_CREAT != 0 - mustCreate := opts.Flags&(linux.O_CREAT|linux.O_EXCL) == (linux.O_CREAT | linux.O_EXCL) - - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - - start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for start as required by fs.stepLocked(). - if err := start.updateFromGetattr(ctx); err != nil { - return nil, err - } - } - if rp.Done() { - return start.openLocked(ctx, rp, &opts) - } - -afterTrailingSymlink: - parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) - if err != nil { - return nil, err - } - // Check for search permission in the parent directory. - if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - // Determine whether or not we need to create a file. - parent.dirMu.Lock() - child, err := fs.stepLocked(ctx, rp, parent, &ds) - if err == syserror.ENOENT && mayCreate { - fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts) - parent.dirMu.Unlock() - return fd, err - } - if err != nil { - parent.dirMu.Unlock() - return nil, err - } - // Open existing child or follow symlink. - parent.dirMu.Unlock() - if mustCreate { - return nil, syserror.EEXIST - } - if child.isSymlink() && rp.ShouldFollowSymlink() { - target, err := child.readlink(ctx, rp.Mount()) - if err != nil { - return nil, err - } - if err := rp.HandleSymlink(target); err != nil { - return nil, err - } - start = parent - goto afterTrailingSymlink - } - return child.openLocked(ctx, rp, &opts) -} - -// Preconditions: fs.renameMu must be locked. -func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(opts) - if err := d.checkPermissions(rp.Credentials(), ats); err != nil { - return nil, err - } - mnt := rp.Mount() - filetype := d.fileType() - switch { - case filetype == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD: - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil { - return nil, err - } - fd := ®ularFileFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ - AllowDirectIO: true, - }); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case filetype == linux.S_IFDIR: - // Can't open directories with O_CREAT. - if opts.Flags&linux.O_CREAT != 0 { - return nil, syserror.EISDIR - } - // Can't open directories writably. - if ats&vfs.MayWrite != 0 { - return nil, syserror.EISDIR - } - if opts.Flags&linux.O_DIRECT != 0 { - return nil, syserror.EINVAL - } - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { - return nil, err - } - fd := &directoryFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case filetype == linux.S_IFLNK: - // Can't open symlinks without O_PATH (which is unimplemented). - return nil, syserror.ELOOP - default: - if opts.Flags&linux.O_DIRECT != 0 { - return nil, syserror.EINVAL - } - h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0) - if err != nil { - return nil, err - } - fd := &specialFileFD{ - handle: h, - } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - h.close(ctx) - return nil, err - } - return &fd.vfsfd, nil - } -} - -// Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. -func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { - if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { - return nil, err - } - if d.isDeleted() { - return nil, syserror.ENOENT - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return nil, err - } - defer mnt.EndWrite() - - // 9P2000.L's lcreate takes a fid representing the parent directory, and - // converts it into an open fid representing the created file, so we need - // to duplicate the directory fid first. - _, dirfile, err := d.file.walk(ctx, nil) - if err != nil { - return nil, err - } - creds := rp.Credentials() - name := rp.Component() - fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, (p9.OpenFlags)(opts.Flags), (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) - if err != nil { - dirfile.close(ctx) - return nil, err - } - // Then we need to walk to the file we just created to get a non-open fid - // representing it, and to get its metadata. This must use d.file since, as - // explained above, dirfile was invalidated by dirfile.Create(). - walkQID, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name) - if err != nil { - openFile.close(ctx) - if fdobj != nil { - fdobj.Close() - } - return nil, err - } - // Sanity-check that we walked to the file we created. - if createQID.Path != walkQID.Path { - // Probably due to concurrent remote filesystem mutation? - ctx.Warningf("gofer.dentry.createAndOpenChildLocked: created file has QID %v before walk, QID %v after (interop=%v)", createQID, walkQID, d.fs.opts.interop) - nonOpenFile.close(ctx) - openFile.close(ctx) - if fdobj != nil { - fdobj.Close() - } - return nil, syserror.EAGAIN - } - - // Construct the new dentry. - child, err := d.fs.newDentry(ctx, nonOpenFile, createQID, attrMask, &attr) - if err != nil { - nonOpenFile.close(ctx) - openFile.close(ctx) - if fdobj != nil { - fdobj.Close() - } - return nil, err - } - // Incorporate the fid that was opened by lcreate. - useRegularFileFD := child.fileType() == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD - if useRegularFileFD { - child.handleMu.Lock() - child.handle.file = openFile - if fdobj != nil { - child.handle.fd = int32(fdobj.Release()) - } - child.handleReadable = vfs.MayReadFileWithOpenFlags(opts.Flags) - child.handleWritable = vfs.MayWriteFileWithOpenFlags(opts.Flags) - child.handleMu.Unlock() - } - // Take a reference on the new dentry to be held by the new file - // description. (This reference also means that the new dentry is not - // eligible for caching yet, so we don't need to append to a dentry slice.) - child.refs = 1 - // Insert the dentry into the tree. - d.IncRef() // reference held by child on its parent d - d.vfsd.InsertChild(&child.vfsd, name) - if d.fs.opts.interop != InteropModeShared { - d.touchCMtime(ctx) - delete(d.negativeChildren, name) - d.dirents = nil - } - - // Finally, construct a file description representing the created file. - var childVFSFD *vfs.FileDescription - mnt.IncRef() - if useRegularFileFD { - fd := ®ularFileFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ - AllowDirectIO: true, - }); err != nil { - return nil, err - } - childVFSFD = &fd.vfsfd - } else { - fd := &specialFileFD{ - handle: handle{ - file: openFile, - fd: -1, - }, - } - if fdobj != nil { - fd.handle.fd = int32(fdobj.Release()) - } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - fd.handle.close(ctx) - return nil, err - } - childVFSFD = &fd.vfsfd - } - return childVFSFD, nil -} - -// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. -func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return "", err - } - if !d.isSymlink() { - return "", syserror.EINVAL - } - return d.readlink(ctx, rp.Mount()) -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { - if opts.Flags != 0 { - // Requires 9P support. - return syserror.EINVAL - } - - var ds *[]*dentry - fs.renameMu.Lock() - defer fs.renameMuUnlockAndCheckCaching(&ds) - newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) - if err != nil { - return err - } - newName := rp.Component() - if newName == "." || newName == ".." { - return syserror.EBUSY - } - mnt := rp.Mount() - if mnt != oldParentVD.Mount() { - return syserror.EXDEV - } - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - - oldParent := oldParentVD.Dentry().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - if err := oldParent.updateFromGetattr(ctx); err != nil { - return err - } - } - if err := oldParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - vfsObj := rp.VirtualFilesystem() - // We need a dentry representing the renamed file since, if it's a - // directory, we need to check for write permission on it. - oldParent.dirMu.Lock() - defer oldParent.dirMu.Unlock() - renamed, err := fs.revalidateChildLocked(ctx, vfsObj, oldParent, oldName, oldParent.vfsd.Child(oldName), &ds) - if err != nil { - return err - } - if renamed == nil { - return syserror.ENOENT - } - if renamed.isDir() { - if renamed == newParent || renamed.vfsd.IsAncestorOf(&newParent.vfsd) { - return syserror.EINVAL - } - if oldParent != newParent { - if err := renamed.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { - return err - } - } - } else { - if opts.MustBeDir || rp.MustBeDir() { - return syserror.ENOTDIR - } - } - - if oldParent != newParent { - if err := newParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - newParent.dirMu.Lock() - defer newParent.dirMu.Unlock() - } - if newParent.isDeleted() { - return syserror.ENOENT - } - replacedVFSD := newParent.vfsd.Child(newName) - var replaced *dentry - // This is similar to unlinkAt, except: - // - // - We revalidate the replaced dentry unconditionally for simplicity. - // - // - If rp.MustBeDir(), then we need a dentry representing the replaced - // file regardless to confirm that it's a directory. - if replacedVFSD != nil || rp.MustBeDir() { - replaced, err = fs.revalidateChildLocked(ctx, vfsObj, newParent, newName, replacedVFSD, &ds) - if err != nil { - return err - } - if replaced != nil { - if replaced.isDir() { - if !renamed.isDir() { - return syserror.EISDIR - } - } else { - if rp.MustBeDir() || renamed.isDir() { - return syserror.ENOTDIR - } - } - replacedVFSD = &replaced.vfsd - } else { - replacedVFSD = nil - } - } - - if oldParent == newParent && oldName == newName { - return nil - } - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { - return err - } - if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { - vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) - return err - } - if fs.opts.interop != InteropModeShared { - oldParent.cacheNegativeChildLocked(oldName) - oldParent.dirents = nil - delete(newParent.negativeChildren, newName) - newParent.dirents = nil - if renamed.isDir() { - oldParent.decLinks() - newParent.incLinks() - } - } - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, &newParent.vfsd, newName, replacedVFSD) - return nil -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - return fs.unlinkAt(ctx, rp, true /* dir */) -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return err - } - return d.setStat(ctx, rp.Credentials(), &opts.Stat, rp.Mount()) -} - -// StatAt implements vfs.FilesystemImpl.StatAt. -func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return linux.Statx{}, err - } - // Since walking updates metadata for all traversed dentries under - // InteropModeShared, including the returned one, we can return cached - // metadata here regardless of fs.opts.interop. - var stat linux.Statx - d.statTo(&stat) - return stat, nil -} - -// StatFSAt implements vfs.FilesystemImpl.StatFSAt. -func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return linux.Statfs{}, err - } - fsstat, err := d.file.statFS(ctx) - if err != nil { - return linux.Statfs{}, err - } - nameLen := uint64(fsstat.NameLength) - if nameLen > maxFilenameLen { - nameLen = maxFilenameLen - } - return linux.Statfs{ - // This is primarily for distinguishing a gofer file system in - // tests. Testing is important, so instead of defining - // something completely random, use a standard value. - Type: linux.V9FS_MAGIC, - BlockSize: int64(fsstat.BlockSize), - Blocks: fsstat.Blocks, - BlocksFree: fsstat.BlocksFree, - BlocksAvailable: fsstat.BlocksAvailable, - Files: fsstat.Files, - FilesFree: fsstat.FilesFree, - NameLength: nameLen, - }, nil -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { - creds := rp.Credentials() - _, err := parent.file.symlink(ctx, target, name, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) - return err - }) -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - return fs.unlinkAt(ctx, rp, false /* dir */) -} - -// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return nil, err - } - return d.listxattr(ctx) -} - -// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return "", err - } - return d.getxattr(ctx, name) -} - -// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. -func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return err - } - return d.setxattr(ctx, &opts) -} - -// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. -func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return err - } - return d.removexattr(ctx, name) -} - -// PrependPath implements vfs.FilesystemImpl.PrependPath. -func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { - fs.renameMu.RLock() - defer fs.renameMu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) -} diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go deleted file mode 100644 index cf276a417..000000000 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ /dev/null @@ -1,1180 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package gofer provides a filesystem implementation that is backed by a 9p -// server, interchangably referred to as "gofers" throughout this package. -// -// Lock order: -// regularFileFD/directoryFD.mu -// filesystem.renameMu -// dentry.dirMu -// filesystem.syncMu -// dentry.metadataMu -// *** "memmap.Mappable locks" below this point -// dentry.mapsMu -// *** "memmap.Mappable locks taken by Translate" below this point -// dentry.handleMu -// dentry.dataMu -// -// Locking dentry.dirMu in multiple dentries requires holding -// filesystem.renameMu for writing. -package gofer - -import ( - "fmt" - "strconv" - "sync" - "sync/atomic" - "syscall" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/unet" - "gvisor.dev/gvisor/pkg/usermem" -) - -// Name is the default filesystem name. -const Name = "9p" - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - vfsfs vfs.Filesystem - - // mfp is used to allocate memory that caches regular file contents. mfp is - // immutable. - mfp pgalloc.MemoryFileProvider - - // Immutable options. - opts filesystemOptions - - // client is the client used by this filesystem. client is immutable. - client *p9.Client - - // uid and gid are the effective KUID and KGID of the filesystem's creator, - // and are used as the owner and group for files that don't specify one. - // uid and gid are immutable. - uid auth.KUID - gid auth.KGID - - // renameMu serves two purposes: - // - // - It synchronizes path resolution with renaming initiated by this - // client. - // - // - It is held by path resolution to ensure that reachable dentries remain - // valid. A dentry is reachable by path resolution if it has a non-zero - // reference count (such that it is usable as vfs.ResolvingPath.Start() or - // is reachable from its children), or if it is a child dentry (such that - // it is reachable from its parent). - renameMu sync.RWMutex - - // cachedDentries contains all dentries with 0 references. (Due to race - // conditions, it may also contain dentries with non-zero references.) - // cachedDentriesLen is the number of dentries in cachedDentries. These - // fields are protected by renameMu. - cachedDentries dentryList - cachedDentriesLen uint64 - - // dentries contains all dentries in this filesystem. specialFileFDs - // contains all open specialFileFDs. These fields are protected by syncMu. - syncMu sync.Mutex - dentries map[*dentry]struct{} - specialFileFDs map[*specialFileFD]struct{} -} - -type filesystemOptions struct { - // "Standard" 9P options. - fd int - aname string - interop InteropMode // derived from the "cache" mount option - msize uint32 - version string - - // maxCachedDentries is the maximum number of dentries with 0 references - // retained by the client. - maxCachedDentries uint64 - - // If forcePageCache is true, host FDs may not be used for application - // memory mappings even if available; instead, the client must perform its - // own caching of regular file pages. This is primarily useful for testing. - forcePageCache bool - - // If limitHostFDTranslation is true, apply maxFillRange() constraints to - // host FD mappings returned by dentry.(memmap.Mappable).Translate(). This - // makes memory accounting behavior more consistent between cases where - // host FDs are / are not available, but may increase the frequency of - // sentry-handled page faults on files for which a host FD is available. - limitHostFDTranslation bool - - // If overlayfsStaleRead is true, O_RDONLY host FDs provided by the remote - // filesystem may not be coherent with writable host FDs opened later, so - // mappings of the former must be replaced by mappings of the latter. This - // is usually only the case when the remote filesystem is an overlayfs - // mount on Linux < 4.19. - overlayfsStaleRead bool - - // If regularFilesUseSpecialFileFD is true, application FDs representing - // regular files will use distinct file handles for each FD, in the same - // way that application FDs representing "special files" such as sockets - // do. Note that this disables client caching and mmap for regular files. - regularFilesUseSpecialFileFD bool -} - -// InteropMode controls the client's interaction with other remote filesystem -// users. -type InteropMode uint32 - -const ( - // InteropModeExclusive is appropriate when the filesystem client is the - // only user of the remote filesystem. - // - // - The client may cache arbitrary filesystem state (file data, metadata, - // filesystem structure, etc.). - // - // - Client changes to filesystem state may be sent to the remote - // filesystem asynchronously, except when server permission checks are - // necessary. - // - // - File timestamps are based on client clocks. This ensures that users of - // the client observe timestamps that are coherent with their own clocks - // and consistent with Linux's semantics. However, since it is not always - // possible for clients to set arbitrary atimes and mtimes, and never - // possible for clients to set arbitrary ctimes, file timestamp changes are - // stored in the client only and never sent to the remote filesystem. - InteropModeExclusive InteropMode = iota - - // InteropModeWritethrough is appropriate when there are read-only users of - // the remote filesystem that expect to observe changes made by the - // filesystem client. - // - // - The client may cache arbitrary filesystem state. - // - // - Client changes to filesystem state must be sent to the remote - // filesystem synchronously. - // - // - File timestamps are based on client clocks. As a corollary, access - // timestamp changes from other remote filesystem users will not be visible - // to the client. - InteropModeWritethrough - - // InteropModeShared is appropriate when there are users of the remote - // filesystem that may mutate its state other than the client. - // - // - The client must verify cached filesystem state before using it. - // - // - Client changes to filesystem state must be sent to the remote - // filesystem synchronously. - // - // - File timestamps are based on server clocks. This is necessary to - // ensure that timestamp changes are synchronized between remote filesystem - // users. - // - // Note that the correctness of InteropModeShared depends on the server - // correctly implementing 9P fids (i.e. each fid immutably represents a - // single filesystem object), even in the presence of remote filesystem - // mutations from other users. If this is violated, the behavior of the - // client is undefined. - InteropModeShared -) - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - mfp := pgalloc.MemoryFileProviderFromContext(ctx) - if mfp == nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: context does not provide a pgalloc.MemoryFileProvider") - return nil, nil, syserror.EINVAL - } - - mopts := vfs.GenericParseMountOptions(opts.Data) - var fsopts filesystemOptions - - // Check that the transport is "fd". - trans, ok := mopts["trans"] - if !ok { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: transport must be specified as 'trans=fd'") - return nil, nil, syserror.EINVAL - } - delete(mopts, "trans") - if trans != "fd" { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: unsupported transport: trans=%s", trans) - return nil, nil, syserror.EINVAL - } - - // Check that read and write FDs are provided and identical. - rfdstr, ok := mopts["rfdno"] - if !ok { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: read FD must be specified as 'rfdno=<file descriptor>") - return nil, nil, syserror.EINVAL - } - delete(mopts, "rfdno") - rfd, err := strconv.Atoi(rfdstr) - if err != nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid read FD: rfdno=%s", rfdstr) - return nil, nil, syserror.EINVAL - } - wfdstr, ok := mopts["wfdno"] - if !ok { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: write FD must be specified as 'wfdno=<file descriptor>") - return nil, nil, syserror.EINVAL - } - delete(mopts, "wfdno") - wfd, err := strconv.Atoi(wfdstr) - if err != nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid write FD: wfdno=%s", wfdstr) - return nil, nil, syserror.EINVAL - } - if rfd != wfd { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: read FD (%d) and write FD (%d) must be equal", rfd, wfd) - return nil, nil, syserror.EINVAL - } - fsopts.fd = rfd - - // Get the attach name. - fsopts.aname = "/" - if aname, ok := mopts["aname"]; ok { - delete(mopts, "aname") - fsopts.aname = aname - } - - // Parse the cache policy. For historical reasons, this defaults to the - // least generally-applicable option, InteropModeExclusive. - fsopts.interop = InteropModeExclusive - if cache, ok := mopts["cache"]; ok { - delete(mopts, "cache") - switch cache { - case "fscache": - fsopts.interop = InteropModeExclusive - case "fscache_writethrough": - fsopts.interop = InteropModeWritethrough - case "none": - fsopts.regularFilesUseSpecialFileFD = true - fallthrough - case "remote_revalidating": - fsopts.interop = InteropModeShared - default: - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid cache policy: cache=%s", cache) - return nil, nil, syserror.EINVAL - } - } - - // Parse the 9P message size. - fsopts.msize = 1024 * 1024 // 1M, tested to give good enough performance up to 64M - if msizestr, ok := mopts["msize"]; ok { - delete(mopts, "msize") - msize, err := strconv.ParseUint(msizestr, 10, 32) - if err != nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid message size: msize=%s", msizestr) - return nil, nil, syserror.EINVAL - } - fsopts.msize = uint32(msize) - } - - // Parse the 9P protocol version. - fsopts.version = p9.HighestVersionString() - if version, ok := mopts["version"]; ok { - delete(mopts, "version") - fsopts.version = version - } - - // Parse the dentry cache limit. - fsopts.maxCachedDentries = 1000 - if str, ok := mopts["dentry_cache_limit"]; ok { - delete(mopts, "dentry_cache_limit") - maxCachedDentries, err := strconv.ParseUint(str, 10, 64) - if err != nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid dentry cache limit: dentry_cache_limit=%s", str) - return nil, nil, syserror.EINVAL - } - fsopts.maxCachedDentries = maxCachedDentries - } - - // Handle simple flags. - if _, ok := mopts["force_page_cache"]; ok { - delete(mopts, "force_page_cache") - fsopts.forcePageCache = true - } - if _, ok := mopts["limit_host_fd_translation"]; ok { - delete(mopts, "limit_host_fd_translation") - fsopts.limitHostFDTranslation = true - } - if _, ok := mopts["overlayfs_stale_read"]; ok { - delete(mopts, "overlayfs_stale_read") - fsopts.overlayfsStaleRead = true - } - // fsopts.regularFilesUseSpecialFileFD can only be enabled by specifying - // "cache=none". - - // Check for unparsed options. - if len(mopts) != 0 { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: unknown options: %v", mopts) - return nil, nil, syserror.EINVAL - } - - // Establish a connection with the server. - conn, err := unet.NewSocket(fsopts.fd) - if err != nil { - return nil, nil, err - } - - // Perform version negotiation with the server. - ctx.UninterruptibleSleepStart(false) - client, err := p9.NewClient(conn, fsopts.msize, fsopts.version) - ctx.UninterruptibleSleepFinish(false) - if err != nil { - conn.Close() - return nil, nil, err - } - // Ownership of conn has been transferred to client. - - // Perform attach to obtain the filesystem root. - ctx.UninterruptibleSleepStart(false) - attached, err := client.Attach(fsopts.aname) - ctx.UninterruptibleSleepFinish(false) - if err != nil { - client.Close() - return nil, nil, err - } - attachFile := p9file{attached} - qid, attrMask, attr, err := attachFile.getAttr(ctx, dentryAttrMask()) - if err != nil { - attachFile.close(ctx) - client.Close() - return nil, nil, err - } - - // Construct the filesystem object. - fs := &filesystem{ - mfp: mfp, - opts: fsopts, - uid: creds.EffectiveKUID, - gid: creds.EffectiveKGID, - client: client, - dentries: make(map[*dentry]struct{}), - specialFileFDs: make(map[*specialFileFD]struct{}), - } - fs.vfsfs.Init(vfsObj, fs) - - // Construct the root dentry. - root, err := fs.newDentry(ctx, attachFile, qid, attrMask, &attr) - if err != nil { - attachFile.close(ctx) - fs.vfsfs.DecRef() - return nil, nil, err - } - // Set the root's reference count to 2. One reference is returned to the - // caller, and the other is deliberately leaked to prevent the root from - // being "cached" and subsequently evicted. Its resources will still be - // cleaned up by fs.Release(). - root.refs = 2 - - return &fs.vfsfs, &root.vfsd, nil -} - -// Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { - ctx := context.Background() - mf := fs.mfp.MemoryFile() - - fs.syncMu.Lock() - for d := range fs.dentries { - d.handleMu.Lock() - d.dataMu.Lock() - if d.handleWritable { - // Write dirty cached data to the remote file. - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt); err != nil { - log.Warningf("gofer.filesystem.Release: failed to flush dentry: %v", err) - } - // TODO(jamieliu): Do we need to flushf/fsync d? - } - // Discard cached pages. - d.cache.DropAll(mf) - d.dirty.RemoveAll() - d.dataMu.Unlock() - // Close the host fd if one exists. - if d.handle.fd >= 0 { - syscall.Close(int(d.handle.fd)) - d.handle.fd = -1 - } - d.handleMu.Unlock() - } - // There can't be any specialFileFDs still using fs, since each such - // FileDescription would hold a reference on a Mount holding a reference on - // fs. - fs.syncMu.Unlock() - - // Close the connection to the server. This implicitly clunks all fids. - fs.client.Close() -} - -// dentry implements vfs.DentryImpl. -type dentry struct { - vfsd vfs.Dentry - - // refs is the reference count. Each dentry holds a reference on its - // parent, even if disowned. refs is accessed using atomic memory - // operations. - refs int64 - - // fs is the owning filesystem. fs is immutable. - fs *filesystem - - // We don't support hard links, so each dentry maps 1:1 to an inode. - - // file is the unopened p9.File that backs this dentry. file is immutable. - file p9file - - // If deleted is non-zero, the file represented by this dentry has been - // deleted. deleted is accessed using atomic memory operations. - deleted uint32 - - // If cached is true, dentryEntry links dentry into - // filesystem.cachedDentries. cached and dentryEntry are protected by - // filesystem.renameMu. - cached bool - dentryEntry - - dirMu sync.Mutex - - // If this dentry represents a directory, and InteropModeShared is not in - // effect, negativeChildren is a set of child names in this directory that - // are known not to exist. negativeChildren is protected by dirMu. - negativeChildren map[string]struct{} - - // If this dentry represents a directory, InteropModeShared is not in - // effect, and dirents is not nil, it is a cache of all entries in the - // directory, in the order they were returned by the server. dirents is - // protected by dirMu. - dirents []vfs.Dirent - - // Cached metadata; protected by metadataMu and accessed using atomic - // memory operations unless otherwise specified. - metadataMu sync.Mutex - ino uint64 // immutable - mode uint32 // type is immutable, perms are mutable - uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic - gid uint32 // auth.KGID, but ... - blockSize uint32 // 0 if unknown - // Timestamps, all nsecs from the Unix epoch. - atime int64 - mtime int64 - ctime int64 - btime int64 - // File size, protected by both metadataMu and dataMu (i.e. both must be - // locked to mutate it). - size uint64 - - // nlink counts the number of hard links to this dentry. It's updated and - // accessed using atomic operations. It's not protected by metadataMu like the - // other metadata fields. - nlink uint32 - - mapsMu sync.Mutex - - // If this dentry represents a regular file, mappings tracks mappings of - // the file into memmap.MappingSpaces. mappings is protected by mapsMu. - mappings memmap.MappingSet - - // If this dentry represents a regular file or directory: - // - // - handle is the I/O handle used by all regularFileFDs/directoryFDs - // representing this dentry. - // - // - handleReadable is true if handle is readable. - // - // - handleWritable is true if handle is writable. - // - // Invariants: - // - // - If handleReadable == handleWritable == false, then handle.file == nil - // (i.e. there is no open handle). Conversely, if handleReadable || - // handleWritable == true, then handle.file != nil (i.e. there is an open - // handle). - // - // - handleReadable and handleWritable cannot transition from true to false - // (i.e. handles may not be downgraded). - // - // These fields are protected by handleMu. - handleMu sync.RWMutex - handle handle - handleReadable bool - handleWritable bool - - dataMu sync.RWMutex - - // If this dentry represents a regular file that is client-cached, cache - // maps offsets into the cached file to offsets into - // filesystem.mfp.MemoryFile() that store the file's data. cache is - // protected by dataMu. - cache fsutil.FileRangeSet - - // If this dentry represents a regular file that is client-cached, dirty - // tracks dirty segments in cache. dirty is protected by dataMu. - dirty fsutil.DirtySet - - // pf implements platform.File for mappings of handle.fd. - pf dentryPlatformFile - - // If this dentry represents a symbolic link, InteropModeShared is not in - // effect, and haveTarget is true, target is the symlink target. haveTarget - // and target are protected by dataMu. - haveTarget bool - target string -} - -// dentryAttrMask returns a p9.AttrMask enabling all attributes used by the -// gofer client. -func dentryAttrMask() p9.AttrMask { - return p9.AttrMask{ - Mode: true, - UID: true, - GID: true, - ATime: true, - MTime: true, - CTime: true, - Size: true, - BTime: true, - } -} - -// newDentry creates a new dentry representing the given file. The dentry -// initially has no references, but is not cached; it is the caller's -// responsibility to set the dentry's reference count and/or call -// dentry.checkCachingLocked() as appropriate. -func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, mask p9.AttrMask, attr *p9.Attr) (*dentry, error) { - if !mask.Mode { - ctx.Warningf("can't create gofer.dentry without file type") - return nil, syserror.EIO - } - if attr.Mode.FileType() == p9.ModeRegular && !mask.Size { - ctx.Warningf("can't create regular file gofer.dentry without file size") - return nil, syserror.EIO - } - - d := &dentry{ - fs: fs, - file: file, - ino: qid.Path, - mode: uint32(attr.Mode), - uid: uint32(fs.uid), - gid: uint32(fs.gid), - blockSize: usermem.PageSize, - handle: handle{ - fd: -1, - }, - } - d.pf.dentry = d - if mask.UID { - d.uid = uint32(attr.UID) - } - if mask.GID { - d.gid = uint32(attr.GID) - } - if mask.Size { - d.size = attr.Size - } - if attr.BlockSize != 0 { - d.blockSize = uint32(attr.BlockSize) - } - if mask.ATime { - d.atime = dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds) - } - if mask.MTime { - d.mtime = dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds) - } - if mask.CTime { - d.ctime = dentryTimestampFromP9(attr.CTimeSeconds, attr.CTimeNanoSeconds) - } - if mask.BTime { - d.btime = dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds) - } - if mask.NLink { - d.nlink = uint32(attr.NLink) - } - d.vfsd.Init(d) - - fs.syncMu.Lock() - fs.dentries[d] = struct{}{} - fs.syncMu.Unlock() - return d, nil -} - -// updateFromP9Attrs is called to update d's metadata after an update from the -// remote filesystem. -func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { - d.metadataMu.Lock() - if mask.Mode { - if got, want := uint32(attr.Mode.FileType()), d.fileType(); got != want { - d.metadataMu.Unlock() - panic(fmt.Sprintf("gofer.dentry file type changed from %#o to %#o", want, got)) - } - atomic.StoreUint32(&d.mode, uint32(attr.Mode)) - } - if mask.UID { - atomic.StoreUint32(&d.uid, uint32(attr.UID)) - } - if mask.GID { - atomic.StoreUint32(&d.gid, uint32(attr.GID)) - } - // There is no P9_GETATTR_* bit for I/O block size. - if attr.BlockSize != 0 { - atomic.StoreUint32(&d.blockSize, uint32(attr.BlockSize)) - } - if mask.ATime { - atomic.StoreInt64(&d.atime, dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds)) - } - if mask.MTime { - atomic.StoreInt64(&d.mtime, dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds)) - } - if mask.CTime { - atomic.StoreInt64(&d.ctime, dentryTimestampFromP9(attr.CTimeSeconds, attr.CTimeNanoSeconds)) - } - if mask.BTime { - atomic.StoreInt64(&d.btime, dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds)) - } - if mask.NLink { - atomic.StoreUint32(&d.nlink, uint32(attr.NLink)) - } - if mask.Size { - d.dataMu.Lock() - atomic.StoreUint64(&d.size, attr.Size) - d.dataMu.Unlock() - } - d.metadataMu.Unlock() -} - -func (d *dentry) updateFromGetattr(ctx context.Context) error { - // Use d.handle.file, which represents a 9P fid that has been opened, in - // preference to d.file, which represents a 9P fid that has not. This may - // be significantly more efficient in some implementations. - var ( - file p9file - handleMuRLocked bool - ) - d.handleMu.RLock() - if !d.handle.file.isNil() { - file = d.handle.file - handleMuRLocked = true - } else { - file = d.file - d.handleMu.RUnlock() - } - _, attrMask, attr, err := file.getAttr(ctx, dentryAttrMask()) - if handleMuRLocked { - d.handleMu.RUnlock() - } - if err != nil { - return err - } - d.updateFromP9Attrs(attrMask, &attr) - return nil -} - -func (d *dentry) fileType() uint32 { - return atomic.LoadUint32(&d.mode) & linux.S_IFMT -} - -func (d *dentry) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME - stat.Blksize = atomic.LoadUint32(&d.blockSize) - stat.Nlink = atomic.LoadUint32(&d.nlink) - stat.UID = atomic.LoadUint32(&d.uid) - stat.GID = atomic.LoadUint32(&d.gid) - stat.Mode = uint16(atomic.LoadUint32(&d.mode)) - stat.Ino = d.ino - stat.Size = atomic.LoadUint64(&d.size) - // This is consistent with regularFileFD.Seek(), which treats regular files - // as having no holes. - stat.Blocks = (stat.Size + 511) / 512 - stat.Atime = statxTimestampFromDentry(atomic.LoadInt64(&d.atime)) - stat.Btime = statxTimestampFromDentry(atomic.LoadInt64(&d.btime)) - stat.Ctime = statxTimestampFromDentry(atomic.LoadInt64(&d.ctime)) - stat.Mtime = statxTimestampFromDentry(atomic.LoadInt64(&d.mtime)) - // TODO(gvisor.dev/issue/1198): device number -} - -func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { - if stat.Mask == 0 { - return nil - } - if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_SIZE) != 0 { - return syserror.EPERM - } - mode := linux.FileMode(atomic.LoadUint32(&d.mode)) - if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { - return err - } - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - setLocalAtime := false - setLocalMtime := false - if d.fs.opts.interop != InteropModeShared { - // Timestamp updates will be handled locally. - setLocalAtime = stat.Mask&linux.STATX_ATIME != 0 - setLocalMtime = stat.Mask&linux.STATX_MTIME != 0 - stat.Mask &^= linux.STATX_ATIME | linux.STATX_MTIME - if !setLocalMtime && (stat.Mask&linux.STATX_SIZE != 0) { - // Truncate updates mtime. - setLocalMtime = true - stat.Mtime.Nsec = linux.UTIME_NOW - } - } - d.metadataMu.Lock() - defer d.metadataMu.Unlock() - if stat.Mask != 0 { - if err := d.file.setAttr(ctx, p9.SetAttrMask{ - Permissions: stat.Mask&linux.STATX_MODE != 0, - UID: stat.Mask&linux.STATX_UID != 0, - GID: stat.Mask&linux.STATX_GID != 0, - Size: stat.Mask&linux.STATX_SIZE != 0, - ATime: stat.Mask&linux.STATX_ATIME != 0, - MTime: stat.Mask&linux.STATX_MTIME != 0, - ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, - MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, - }, p9.SetAttr{ - Permissions: p9.FileMode(stat.Mode), - UID: p9.UID(stat.UID), - GID: p9.GID(stat.GID), - Size: stat.Size, - ATimeSeconds: uint64(stat.Atime.Sec), - ATimeNanoSeconds: uint64(stat.Atime.Nsec), - MTimeSeconds: uint64(stat.Mtime.Sec), - MTimeNanoSeconds: uint64(stat.Mtime.Nsec), - }); err != nil { - return err - } - } - if d.fs.opts.interop == InteropModeShared { - // There's no point to updating d's metadata in this case since it'll - // be overwritten by revalidation before the next time it's used - // anyway. (InteropModeShared inhibits client caching of regular file - // data, so there's no cache to truncate either.) - return nil - } - now, haveNow := nowFromContext(ctx) - if !haveNow { - ctx.Warningf("gofer.dentry.setStat: current time not available") - } - if stat.Mask&linux.STATX_MODE != 0 { - atomic.StoreUint32(&d.mode, d.fileType()|uint32(stat.Mode)) - } - if stat.Mask&linux.STATX_UID != 0 { - atomic.StoreUint32(&d.uid, stat.UID) - } - if stat.Mask&linux.STATX_GID != 0 { - atomic.StoreUint32(&d.gid, stat.GID) - } - if setLocalAtime { - if stat.Atime.Nsec == linux.UTIME_NOW { - if haveNow { - atomic.StoreInt64(&d.atime, now) - } - } else { - atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime)) - } - } - if setLocalMtime { - if stat.Mtime.Nsec == linux.UTIME_NOW { - if haveNow { - atomic.StoreInt64(&d.mtime, now) - } - } else { - atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime)) - } - } - if haveNow { - atomic.StoreInt64(&d.ctime, now) - } - if stat.Mask&linux.STATX_SIZE != 0 { - d.dataMu.Lock() - oldSize := d.size - d.size = stat.Size - // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings - // below. This allows concurrent calls to Read/Translate/etc. These - // functions synchronize with truncation by refusing to use cache - // contents beyond the new d.size. (We are still holding d.metadataMu, - // so we can't race with Write or another truncate.) - d.dataMu.Unlock() - if d.size < oldSize { - oldpgend := pageRoundUp(oldSize) - newpgend := pageRoundUp(d.size) - if oldpgend != newpgend { - d.mapsMu.Lock() - d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ - // Compare Linux's mm/truncate.c:truncate_setsize() => - // truncate_pagecache() => - // mm/memory.c:unmap_mapping_range(evencows=1). - InvalidatePrivate: true, - }) - d.mapsMu.Unlock() - } - // We are now guaranteed that there are no translations of - // truncated pages, and can remove them from the cache. Since - // truncated pages have been removed from the remote file, they - // should be dropped without being written back. - d.dataMu.Lock() - d.cache.Truncate(d.size, d.fs.mfp.MemoryFile()) - d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend}) - d.dataMu.Unlock() - } - } - return nil -} - -func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { - return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(atomic.LoadUint32(&d.mode)), auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) -} - -// IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef() { - // d.refs may be 0 if d.fs.renameMu is locked, which serializes against - // d.checkCachingLocked(). - atomic.AddInt64(&d.refs, 1) -} - -// TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef() bool { - for { - refs := atomic.LoadInt64(&d.refs) - if refs == 0 { - return false - } - if atomic.CompareAndSwapInt64(&d.refs, refs, refs+1) { - return true - } - } -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { - d.fs.renameMu.Lock() - d.checkCachingLocked() - d.fs.renameMu.Unlock() - } else if refs < 0 { - panic("gofer.dentry.DecRef() called without holding a reference") - } -} - -// checkCachingLocked should be called after d's reference count becomes 0 or it -// becomes disowned. -// -// Preconditions: d.fs.renameMu must be locked for writing. -func (d *dentry) checkCachingLocked() { - // Dentries with a non-zero reference count must be retained. (The only way - // to obtain a reference on a dentry with zero references is via path - // resolution, which requires renameMu, so if d.refs is zero then it will - // remain zero while we hold renameMu for writing.) - if atomic.LoadInt64(&d.refs) != 0 { - if d.cached { - d.fs.cachedDentries.Remove(d) - d.fs.cachedDentriesLen-- - d.cached = false - } - return - } - // Non-child dentries with zero references are no longer reachable by path - // resolution and should be dropped immediately. - if d.vfsd.Parent() == nil || d.vfsd.IsDisowned() { - if d.cached { - d.fs.cachedDentries.Remove(d) - d.fs.cachedDentriesLen-- - d.cached = false - } - d.destroyLocked() - return - } - // If d is already cached, just move it to the front of the LRU. - if d.cached { - d.fs.cachedDentries.Remove(d) - d.fs.cachedDentries.PushFront(d) - return - } - // Cache the dentry, then evict the least recently used cached dentry if - // the cache becomes over-full. - d.fs.cachedDentries.PushFront(d) - d.fs.cachedDentriesLen++ - d.cached = true - if d.fs.cachedDentriesLen > d.fs.opts.maxCachedDentries { - victim := d.fs.cachedDentries.Back() - d.fs.cachedDentries.Remove(victim) - d.fs.cachedDentriesLen-- - victim.cached = false - // victim.refs may have become non-zero from an earlier path - // resolution since it was inserted into fs.cachedDentries; see - // dentry.incRefLocked(). Either way, we brought - // fs.cachedDentriesLen back down to fs.opts.maxCachedDentries, so - // we don't loop. - if atomic.LoadInt64(&victim.refs) == 0 { - if victimParentVFSD := victim.vfsd.Parent(); victimParentVFSD != nil { - victimParent := victimParentVFSD.Impl().(*dentry) - victimParent.dirMu.Lock() - if !victim.vfsd.IsDisowned() { - // victim can't be a mount point (in any mount - // namespace), since VFS holds references on mount - // points. - d.fs.vfsfs.VirtualFilesystem().ForceDeleteDentry(&victim.vfsd) - // We're only deleting the dentry, not the file it - // represents, so we don't need to update - // victimParent.dirents etc. - } - victimParent.dirMu.Unlock() - } - victim.destroyLocked() - } - } -} - -// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is -// not a child dentry. -func (d *dentry) destroyLocked() { - ctx := context.Background() - d.handleMu.Lock() - if !d.handle.file.isNil() { - mf := d.fs.mfp.MemoryFile() - d.dataMu.Lock() - // Write dirty pages back to the remote filesystem. - if d.handleWritable { - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { - log.Warningf("gofer.dentry.DecRef: failed to write dirty data back: %v", err) - } - } - // Discard cached data. - d.cache.DropAll(mf) - d.dirty.RemoveAll() - d.dataMu.Unlock() - // Clunk open fids and close open host FDs. - d.handle.close(ctx) - } - d.handleMu.Unlock() - d.file.close(ctx) - // Remove d from the set of all dentries. - d.fs.syncMu.Lock() - delete(d.fs.dentries, d) - d.fs.syncMu.Unlock() - // Drop the reference held by d on its parent. - if parentVFSD := d.vfsd.Parent(); parentVFSD != nil { - parent := parentVFSD.Impl().(*dentry) - // This is parent.DecRef() without recursive locking of d.fs.renameMu. - if refs := atomic.AddInt64(&parent.refs, -1); refs == 0 { - parent.checkCachingLocked() - } else if refs < 0 { - panic("gofer.dentry.DecRef() called without holding a reference") - } - } -} - -func (d *dentry) isDeleted() bool { - return atomic.LoadUint32(&d.deleted) != 0 -} - -func (d *dentry) setDeleted() { - atomic.StoreUint32(&d.deleted, 1) -} - -func (d *dentry) listxattr(ctx context.Context) ([]string, error) { - return nil, syserror.ENOTSUP -} - -func (d *dentry) getxattr(ctx context.Context, name string) (string, error) { - // TODO(jamieliu): add vfs.GetxattrOptions.Size - return d.file.getXattr(ctx, name, linux.XATTR_SIZE_MAX) -} - -func (d *dentry) setxattr(ctx context.Context, opts *vfs.SetxattrOptions) error { - return d.file.setXattr(ctx, opts.Name, opts.Value, opts.Flags) -} - -func (d *dentry) removexattr(ctx context.Context, name string) error { - return syserror.ENOTSUP -} - -// Preconditions: d.isRegularFile() || d.isDirectory(). -func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { - // O_TRUNC unconditionally requires us to obtain a new handle (opened with - // O_TRUNC). - if !trunc { - d.handleMu.RLock() - if (!read || d.handleReadable) && (!write || d.handleWritable) { - // The current handle is sufficient. - d.handleMu.RUnlock() - return nil - } - d.handleMu.RUnlock() - } - - haveOldFD := false - d.handleMu.Lock() - if (read && !d.handleReadable) || (write && !d.handleWritable) || trunc { - // Get a new handle. - wantReadable := d.handleReadable || read - wantWritable := d.handleWritable || write - h, err := openHandle(ctx, d.file, wantReadable, wantWritable, trunc) - if err != nil { - d.handleMu.Unlock() - return err - } - if !d.handle.file.isNil() { - // Check that old and new handles are compatible: If the old handle - // includes a host file descriptor but the new one does not, or - // vice versa, old and new memory mappings may be incoherent. - haveOldFD = d.handle.fd >= 0 - haveNewFD := h.fd >= 0 - if haveOldFD != haveNewFD { - d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: can't change host FD availability from %v to %v across dentry handle upgrade", haveOldFD, haveNewFD) - h.close(ctx) - return syserror.EIO - } - if haveOldFD { - // We may have raced with callers of d.pf.FD() that are now - // using the old file descriptor, preventing us from safely - // closing it. We could handle this by invalidating existing - // memmap.Translations, but this is expensive. Instead, use - // dup3 to make the old file descriptor refer to the new file - // description, then close the new file descriptor (which is no - // longer needed). Racing callers may use the old or new file - // description, but this doesn't matter since they refer to the - // same file (unless d.fs.opts.overlayfsStaleRead is true, - // which we handle separately). - if err := syscall.Dup3(int(h.fd), int(d.handle.fd), 0); err != nil { - d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, d.handle.fd, err) - h.close(ctx) - return err - } - syscall.Close(int(h.fd)) - h.fd = d.handle.fd - if d.fs.opts.overlayfsStaleRead { - // Replace sentry mappings of the old FD with mappings of - // the new FD, since the two are not necessarily coherent. - if err := d.pf.hostFileMapper.RegenerateMappings(int(h.fd)); err != nil { - d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to replace sentry mappings of old FD with mappings of new FD: %v", err) - h.close(ctx) - return err - } - } - // Clunk the old fid before making the new handle visible (by - // unlocking d.handleMu). - d.handle.file.close(ctx) - } - } - // Switch to the new handle. - d.handle = h - d.handleReadable = wantReadable - d.handleWritable = wantWritable - } - d.handleMu.Unlock() - - if d.fs.opts.overlayfsStaleRead && haveOldFD { - // Invalidate application mappings that may be using the old FD; they - // will be replaced with mappings using the new FD after future calls - // to d.Translate(). This requires holding d.mapsMu, which precedes - // d.handleMu in the lock order. - d.mapsMu.Lock() - d.mappings.InvalidateAll(memmap.InvalidateOpts{}) - d.mapsMu.Unlock() - } - - return nil -} - -// incLinks increments link count. -// -// Preconditions: d.nlink != 0 && d.nlink < math.MaxUint32. -func (d *dentry) incLinks() { - v := atomic.AddUint32(&d.nlink, 1) - if v < 2 { - panic(fmt.Sprintf("dentry.nlink is invalid (was 0 or overflowed): %d", v)) - } -} - -// decLinks decrements link count. -// -// Preconditions: d.nlink > 1. -func (d *dentry) decLinks() { - v := atomic.AddUint32(&d.nlink, ^uint32(0)) - if v == 0 { - panic(fmt.Sprintf("dentry.nlink must be greater than 0: %d", v)) - } -} - -// fileDescription is embedded by gofer implementations of -// vfs.FileDescriptionImpl. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl -} - -func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) -} - -func (fd *fileDescription) dentry() *dentry { - return fd.vfsfd.Dentry().Impl().(*dentry) -} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - d := fd.dentry() - const validMask = uint32(linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME) - if d.fs.opts.interop == InteropModeShared && opts.Mask&(validMask) != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { - // TODO(jamieliu): Use specialFileFD.handle.file for the getattr if - // available? - if err := d.updateFromGetattr(ctx); err != nil { - return linux.Statx{}, err - } - } - var stat linux.Statx - d.statTo(&stat) - return stat, nil -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - return fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, fd.vfsfd.Mount()) -} - -// Listxattr implements vfs.FileDescriptionImpl.Listxattr. -func (fd *fileDescription) Listxattr(ctx context.Context) ([]string, error) { - return fd.dentry().listxattr(ctx) -} - -// Getxattr implements vfs.FileDescriptionImpl.Getxattr. -func (fd *fileDescription) Getxattr(ctx context.Context, name string) (string, error) { - return fd.dentry().getxattr(ctx, name) -} - -// Setxattr implements vfs.FileDescriptionImpl.Setxattr. -func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { - return fd.dentry().setxattr(ctx, &opts) -} - -// Removexattr implements vfs.FileDescriptionImpl.Removexattr. -func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { - return fd.dentry().removexattr(ctx, name) -} diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go deleted file mode 100644 index cfe66f797..000000000 --- a/pkg/sentry/fsimpl/gofer/handle.go +++ /dev/null @@ -1,135 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "syscall" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/safemem" -) - -// handle represents a remote "open file descriptor", consisting of an opened -// fid (p9.File) and optionally a host file descriptor. -type handle struct { - file p9file - fd int32 // -1 if unavailable -} - -// Preconditions: read || write. -func openHandle(ctx context.Context, file p9file, read, write, trunc bool) (handle, error) { - _, newfile, err := file.walk(ctx, nil) - if err != nil { - return handle{fd: -1}, err - } - var flags p9.OpenFlags - switch { - case read && !write: - flags = p9.ReadOnly - case !read && write: - flags = p9.WriteOnly - case read && write: - flags = p9.ReadWrite - } - if trunc { - flags |= p9.OpenTruncate - } - fdobj, _, _, err := newfile.open(ctx, flags) - if err != nil { - newfile.close(ctx) - return handle{fd: -1}, err - } - fd := int32(-1) - if fdobj != nil { - fd = int32(fdobj.Release()) - } - return handle{ - file: newfile, - fd: fd, - }, nil -} - -func (h *handle) close(ctx context.Context) { - h.file.close(ctx) - h.file = p9file{} - if h.fd >= 0 { - syscall.Close(int(h.fd)) - h.fd = -1 - } -} - -func (h *handle) readToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offset uint64) (uint64, error) { - if dsts.IsEmpty() { - return 0, nil - } - if h.fd >= 0 { - ctx.UninterruptibleSleepStart(false) - n, err := hostPreadv(h.fd, dsts, int64(offset)) - ctx.UninterruptibleSleepFinish(false) - return n, err - } - if dsts.NumBlocks() == 1 && !dsts.Head().NeedSafecopy() { - n, err := h.file.readAt(ctx, dsts.Head().ToSlice(), offset) - return uint64(n), err - } - // Buffer the read since p9.File.ReadAt() takes []byte. - buf := make([]byte, dsts.NumBytes()) - n, err := h.file.readAt(ctx, buf, offset) - if n == 0 { - return 0, err - } - if cp, cperr := safemem.CopySeq(dsts, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf[:n]))); cperr != nil { - return cp, cperr - } - return uint64(n), err -} - -func (h *handle) writeFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) { - if srcs.IsEmpty() { - return 0, nil - } - if h.fd >= 0 { - ctx.UninterruptibleSleepStart(false) - n, err := hostPwritev(h.fd, srcs, int64(offset)) - ctx.UninterruptibleSleepFinish(false) - return n, err - } - if srcs.NumBlocks() == 1 && !srcs.Head().NeedSafecopy() { - n, err := h.file.writeAt(ctx, srcs.Head().ToSlice(), offset) - return uint64(n), err - } - // Buffer the write since p9.File.WriteAt() takes []byte. - buf := make([]byte, srcs.NumBytes()) - cp, cperr := safemem.CopySeq(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), srcs) - if cp == 0 { - return 0, cperr - } - n, err := h.file.writeAt(ctx, buf[:cp], offset) - if err != nil { - return uint64(n), err - } - return cp, cperr -} - -func (h *handle) sync(ctx context.Context) error { - if h.fd >= 0 { - ctx.UninterruptibleSleepStart(false) - err := syscall.Fsync(int(h.fd)) - ctx.UninterruptibleSleepFinish(false) - return err - } - return h.file.fsync(ctx) -} diff --git a/pkg/sentry/fsimpl/gofer/handle_unsafe.go b/pkg/sentry/fsimpl/gofer/handle_unsafe.go deleted file mode 100644 index 19560ab26..000000000 --- a/pkg/sentry/fsimpl/gofer/handle_unsafe.go +++ /dev/null @@ -1,66 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "syscall" - "unsafe" - - "gvisor.dev/gvisor/pkg/safemem" -) - -// Preconditions: !dsts.IsEmpty(). -func hostPreadv(fd int32, dsts safemem.BlockSeq, off int64) (uint64, error) { - // No buffering is necessary regardless of safecopy; host syscalls will - // return EFAULT if appropriate, instead of raising SIGBUS. - if dsts.NumBlocks() == 1 { - // Use pread() instead of preadv() to avoid iovec allocation and - // copying. - dst := dsts.Head() - n, _, e := syscall.Syscall6(syscall.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil - } - iovs := safemem.IovecsFromBlockSeq(dsts) - n, _, e := syscall.Syscall6(syscall.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil -} - -// Preconditions: !srcs.IsEmpty(). -func hostPwritev(fd int32, srcs safemem.BlockSeq, off int64) (uint64, error) { - // No buffering is necessary regardless of safecopy; host syscalls will - // return EFAULT if appropriate, instead of raising SIGBUS. - if srcs.NumBlocks() == 1 { - // Use pwrite() instead of pwritev() to avoid iovec allocation and - // copying. - src := srcs.Head() - n, _, e := syscall.Syscall6(syscall.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil - } - iovs := safemem.IovecsFromBlockSeq(srcs) - n, _, e := syscall.Syscall6(syscall.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil -} diff --git a/pkg/sentry/fsimpl/gofer/p9file.go b/pkg/sentry/fsimpl/gofer/p9file.go deleted file mode 100644 index 755ac2985..000000000 --- a/pkg/sentry/fsimpl/gofer/p9file.go +++ /dev/null @@ -1,219 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/syserror" -) - -// p9file is a wrapper around p9.File that provides methods that are -// Context-aware. -type p9file struct { - file p9.File -} - -func (f p9file) isNil() bool { - return f.file == nil -} - -func (f p9file) walk(ctx context.Context, names []string) ([]p9.QID, p9file, error) { - ctx.UninterruptibleSleepStart(false) - qids, newfile, err := f.file.Walk(names) - ctx.UninterruptibleSleepFinish(false) - return qids, p9file{newfile}, err -} - -func (f p9file) walkGetAttr(ctx context.Context, names []string) ([]p9.QID, p9file, p9.AttrMask, p9.Attr, error) { - ctx.UninterruptibleSleepStart(false) - qids, newfile, attrMask, attr, err := f.file.WalkGetAttr(names) - ctx.UninterruptibleSleepFinish(false) - return qids, p9file{newfile}, attrMask, attr, err -} - -// walkGetAttrOne is a wrapper around p9.File.WalkGetAttr that takes a single -// path component and returns a single qid. -func (f p9file) walkGetAttrOne(ctx context.Context, name string) (p9.QID, p9file, p9.AttrMask, p9.Attr, error) { - ctx.UninterruptibleSleepStart(false) - qids, newfile, attrMask, attr, err := f.file.WalkGetAttr([]string{name}) - ctx.UninterruptibleSleepFinish(false) - if err != nil { - return p9.QID{}, p9file{}, p9.AttrMask{}, p9.Attr{}, err - } - if len(qids) != 1 { - ctx.Warningf("p9.File.WalkGetAttr returned %d qids (%v), wanted 1", len(qids), qids) - if newfile != nil { - p9file{newfile}.close(ctx) - } - return p9.QID{}, p9file{}, p9.AttrMask{}, p9.Attr{}, syserror.EIO - } - return qids[0], p9file{newfile}, attrMask, attr, nil -} - -func (f p9file) statFS(ctx context.Context) (p9.FSStat, error) { - ctx.UninterruptibleSleepStart(false) - fsstat, err := f.file.StatFS() - ctx.UninterruptibleSleepFinish(false) - return fsstat, err -} - -func (f p9file) getAttr(ctx context.Context, req p9.AttrMask) (p9.QID, p9.AttrMask, p9.Attr, error) { - ctx.UninterruptibleSleepStart(false) - qid, attrMask, attr, err := f.file.GetAttr(req) - ctx.UninterruptibleSleepFinish(false) - return qid, attrMask, attr, err -} - -func (f p9file) setAttr(ctx context.Context, valid p9.SetAttrMask, attr p9.SetAttr) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.SetAttr(valid, attr) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) getXattr(ctx context.Context, name string, size uint64) (string, error) { - ctx.UninterruptibleSleepStart(false) - val, err := f.file.GetXattr(name, size) - ctx.UninterruptibleSleepFinish(false) - return val, err -} - -func (f p9file) setXattr(ctx context.Context, name, value string, flags uint32) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.SetXattr(name, value, flags) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Allocate(mode, offset, length) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) close(ctx context.Context) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Close() - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) open(ctx context.Context, flags p9.OpenFlags) (*fd.FD, p9.QID, uint32, error) { - ctx.UninterruptibleSleepStart(false) - fdobj, qid, iounit, err := f.file.Open(flags) - ctx.UninterruptibleSleepFinish(false) - return fdobj, qid, iounit, err -} - -func (f p9file) readAt(ctx context.Context, p []byte, offset uint64) (int, error) { - ctx.UninterruptibleSleepStart(false) - n, err := f.file.ReadAt(p, offset) - ctx.UninterruptibleSleepFinish(false) - return n, err -} - -func (f p9file) writeAt(ctx context.Context, p []byte, offset uint64) (int, error) { - ctx.UninterruptibleSleepStart(false) - n, err := f.file.WriteAt(p, offset) - ctx.UninterruptibleSleepFinish(false) - return n, err -} - -func (f p9file) fsync(ctx context.Context) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.FSync() - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) create(ctx context.Context, name string, flags p9.OpenFlags, permissions p9.FileMode, uid p9.UID, gid p9.GID) (*fd.FD, p9file, p9.QID, uint32, error) { - ctx.UninterruptibleSleepStart(false) - fdobj, newfile, qid, iounit, err := f.file.Create(name, flags, permissions, uid, gid) - ctx.UninterruptibleSleepFinish(false) - return fdobj, p9file{newfile}, qid, iounit, err -} - -func (f p9file) mkdir(ctx context.Context, name string, permissions p9.FileMode, uid p9.UID, gid p9.GID) (p9.QID, error) { - ctx.UninterruptibleSleepStart(false) - qid, err := f.file.Mkdir(name, permissions, uid, gid) - ctx.UninterruptibleSleepFinish(false) - return qid, err -} - -func (f p9file) symlink(ctx context.Context, oldName string, newName string, uid p9.UID, gid p9.GID) (p9.QID, error) { - ctx.UninterruptibleSleepStart(false) - qid, err := f.file.Symlink(oldName, newName, uid, gid) - ctx.UninterruptibleSleepFinish(false) - return qid, err -} - -func (f p9file) link(ctx context.Context, target p9file, newName string) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Link(target.file, newName) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) mknod(ctx context.Context, name string, mode p9.FileMode, major uint32, minor uint32, uid p9.UID, gid p9.GID) (p9.QID, error) { - ctx.UninterruptibleSleepStart(false) - qid, err := f.file.Mknod(name, mode, major, minor, uid, gid) - ctx.UninterruptibleSleepFinish(false) - return qid, err -} - -func (f p9file) rename(ctx context.Context, newDir p9file, newName string) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Rename(newDir.file, newName) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) unlinkAt(ctx context.Context, name string, flags uint32) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.UnlinkAt(name, flags) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) readdir(ctx context.Context, offset uint64, count uint32) ([]p9.Dirent, error) { - ctx.UninterruptibleSleepStart(false) - dirents, err := f.file.Readdir(offset, count) - ctx.UninterruptibleSleepFinish(false) - return dirents, err -} - -func (f p9file) readlink(ctx context.Context) (string, error) { - ctx.UninterruptibleSleepStart(false) - target, err := f.file.Readlink() - ctx.UninterruptibleSleepFinish(false) - return target, err -} - -func (f p9file) flush(ctx context.Context) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Flush() - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) connect(ctx context.Context, flags p9.ConnectFlags) (*fd.FD, error) { - ctx.UninterruptibleSleepStart(false) - fdobj, err := f.file.Connect(flags) - ctx.UninterruptibleSleepFinish(false) - return fdobj, err -} diff --git a/pkg/sentry/fsimpl/gofer/pagemath.go b/pkg/sentry/fsimpl/gofer/pagemath.go deleted file mode 100644 index 847cb0784..000000000 --- a/pkg/sentry/fsimpl/gofer/pagemath.go +++ /dev/null @@ -1,31 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "gvisor.dev/gvisor/pkg/usermem" -) - -// This are equivalent to usermem.Addr.RoundDown/Up, but without the -// potentially truncating conversion to usermem.Addr. This is necessary because -// there is no way to define generic "PageRoundDown/Up" functions in Go. - -func pageRoundDown(x uint64) uint64 { - return x &^ (usermem.PageSize - 1) -} - -func pageRoundUp(x uint64) uint64 { - return pageRoundDown(x + usermem.PageSize - 1) -} diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go deleted file mode 100644 index 3593eb1d5..000000000 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ /dev/null @@ -1,877 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "fmt" - "io" - "math" - "sync" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -func (d *dentry) isRegularFile() bool { - return d.fileType() == linux.S_IFREG -} - -type regularFileFD struct { - fileDescription - - // off is the file offset. off is protected by mu. - mu sync.Mutex - off int64 -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { -} - -// OnClose implements vfs.FileDescriptionImpl.OnClose. -func (fd *regularFileFD) OnClose(ctx context.Context) error { - if !fd.vfsfd.IsWritable() { - return nil - } - // Skip flushing if writes may be buffered by the client, since (as with - // the VFS1 client) we don't flush buffered writes on close anyway. - d := fd.dentry() - if d.fs.opts.interop == InteropModeExclusive { - return nil - } - d.handleMu.RLock() - defer d.handleMu.RUnlock() - return d.handle.file.flush(ctx) -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if opts.Flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - // Check for reading at EOF before calling into MM (but not under - // InteropModeShared, which makes d.size unreliable). - d := fd.dentry() - if d.fs.opts.interop != InteropModeShared && uint64(offset) >= atomic.LoadUint64(&d.size) { - return 0, io.EOF - } - - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Lock d.metadataMu for the rest of the read to prevent d.size from - // changing. - d.metadataMu.Lock() - defer d.metadataMu.Unlock() - // Write dirty cached pages that will be touched by the read back to - // the remote file. - if err := d.writeback(ctx, offset, dst.NumBytes()); err != nil { - return 0, err - } - } - - rw := getDentryReadWriter(ctx, d, offset) - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Require the read to go to the remote file. - rw.direct = true - } - n, err := dst.CopyOutFrom(ctx, rw) - putDentryReadWriter(rw) - if d.fs.opts.interop != InteropModeShared { - // Compare Linux's mm/filemap.c:do_generic_file_read() => file_accessed(). - d.touchAtime(ctx, fd.vfsfd.Mount()) - } - return n, err -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - fd.mu.Lock() - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.off += n - fd.mu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if opts.Flags != 0 { - return 0, syserror.EOPNOTSUPP - } - limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) - if err != nil { - return 0, err - } - src = src.TakeFirst64(limit) - - d := fd.dentry() - d.metadataMu.Lock() - defer d.metadataMu.Unlock() - if d.fs.opts.interop != InteropModeShared { - // Compare Linux's mm/filemap.c:__generic_file_write_iter() => - // file_update_time(). This is d.touchCMtime(), but without locking - // d.metadataMu (recursively). - if now, ok := nowFromContext(ctx); ok { - atomic.StoreInt64(&d.mtime, now) - atomic.StoreInt64(&d.ctime, now) - } - } - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Write dirty cached pages that will be touched by the write back to - // the remote file. - if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { - return 0, err - } - // Remove touched pages from the cache. - pgstart := pageRoundDown(uint64(offset)) - pgend := pageRoundUp(uint64(offset + src.NumBytes())) - if pgend < pgstart { - return 0, syserror.EINVAL - } - mr := memmap.MappableRange{pgstart, pgend} - var freed []platform.FileRange - d.dataMu.Lock() - cseg := d.cache.LowerBoundSegment(mr.Start) - for cseg.Ok() && cseg.Start() < mr.End { - cseg = d.cache.Isolate(cseg, mr) - freed = append(freed, platform.FileRange{cseg.Value(), cseg.Value() + cseg.Range().Length()}) - cseg = d.cache.Remove(cseg).NextSegment() - } - d.dataMu.Unlock() - // Invalidate mappings of removed pages. - d.mapsMu.Lock() - d.mappings.Invalidate(mr, memmap.InvalidateOpts{}) - d.mapsMu.Unlock() - // Finally free pages removed from the cache. - mf := d.fs.mfp.MemoryFile() - for _, freedFR := range freed { - mf.DecRef(freedFR) - } - } - rw := getDentryReadWriter(ctx, d, offset) - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Require the write to go to the remote file. - rw.direct = true - } - n, err := src.CopyInTo(ctx, rw) - putDentryReadWriter(rw) - if n != 0 && fd.vfsfd.StatusFlags()&(linux.O_DSYNC|linux.O_SYNC) != 0 { - // Write dirty cached pages touched by the write back to the remote - // file. - if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { - return 0, err - } - // Request the remote filesystem to sync the remote file. - if err := d.handle.file.fsync(ctx); err != nil { - return 0, err - } - } - return n, err -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - fd.mu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n - fd.mu.Unlock() - return n, err -} - -type dentryReadWriter struct { - ctx context.Context - d *dentry - off uint64 - direct bool -} - -var dentryReadWriterPool = sync.Pool{ - New: func() interface{} { - return &dentryReadWriter{} - }, -} - -func getDentryReadWriter(ctx context.Context, d *dentry, offset int64) *dentryReadWriter { - rw := dentryReadWriterPool.Get().(*dentryReadWriter) - rw.ctx = ctx - rw.d = d - rw.off = uint64(offset) - rw.direct = false - return rw -} - -func putDentryReadWriter(rw *dentryReadWriter) { - rw.ctx = nil - rw.d = nil - dentryReadWriterPool.Put(rw) -} - -// ReadToBlocks implements safemem.Reader.ReadToBlocks. -func (rw *dentryReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { - if dsts.IsEmpty() { - return 0, nil - } - - // If we have a mmappable host FD (which must be used here to ensure - // coherence with memory-mapped I/O), or if InteropModeShared is in effect - // (which prevents us from caching file contents and makes dentry.size - // unreliable), or if the file was opened O_DIRECT, read directly from - // dentry.handle without locking dentry.dataMu. - rw.d.handleMu.RLock() - if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { - n, err := rw.d.handle.readToBlocksAt(rw.ctx, dsts, rw.off) - rw.d.handleMu.RUnlock() - rw.off += n - return n, err - } - - // Otherwise read from/through the cache. - mf := rw.d.fs.mfp.MemoryFile() - fillCache := mf.ShouldCacheEvictable() - var dataMuUnlock func() - if fillCache { - rw.d.dataMu.Lock() - dataMuUnlock = rw.d.dataMu.Unlock - } else { - rw.d.dataMu.RLock() - dataMuUnlock = rw.d.dataMu.RUnlock - } - - // Compute the range to read (limited by file size and overflow-checked). - if rw.off >= rw.d.size { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return 0, io.EOF - } - end := rw.d.size - if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { - end = rend - } - - var done uint64 - seg, gap := rw.d.cache.Find(rw.off) - for rw.off < end { - mr := memmap.MappableRange{rw.off, end} - switch { - case seg.Ok(): - // Get internal mappings from the cache. - ims, err := mf.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) - if err != nil { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, err - } - - // Copy from internal mappings. - n, err := safemem.CopySeq(dsts, ims) - done += n - rw.off += n - dsts = dsts.DropFirst64(n) - if err != nil { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, err - } - - // Continue. - seg, gap = seg.NextNonEmpty() - - case gap.Ok(): - gapMR := gap.Range().Intersect(mr) - if fillCache { - // Read into the cache, then re-enter the loop to read from the - // cache. - reqMR := memmap.MappableRange{ - Start: pageRoundDown(gapMR.Start), - End: pageRoundUp(gapMR.End), - } - optMR := gap.Range() - err := rw.d.cache.Fill(rw.ctx, reqMR, maxFillRange(reqMR, optMR), mf, usage.PageCache, rw.d.handle.readToBlocksAt) - mf.MarkEvictable(rw.d, pgalloc.EvictableRange{optMR.Start, optMR.End}) - seg, gap = rw.d.cache.Find(rw.off) - if !seg.Ok() { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, err - } - // err might have occurred in part of gap.Range() outside - // gapMR. Forget about it for now; if the error matters and - // persists, we'll run into it again in a later iteration of - // this loop. - } else { - // Read directly from the file. - gapDsts := dsts.TakeFirst64(gapMR.Length()) - n, err := rw.d.handle.readToBlocksAt(rw.ctx, gapDsts, gapMR.Start) - done += n - rw.off += n - dsts = dsts.DropFirst64(n) - // Partial reads are fine. But we must stop reading. - if n != gapDsts.NumBytes() || err != nil { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, err - } - - // Continue. - seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} - } - } - } - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, nil -} - -// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. -// -// Preconditions: rw.d.metadataMu must be locked. -func (rw *dentryReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - if srcs.IsEmpty() { - return 0, nil - } - - // If we have a mmappable host FD (which must be used here to ensure - // coherence with memory-mapped I/O), or if InteropModeShared is in effect - // (which prevents us from caching file contents), or if the file was - // opened with O_DIRECT, write directly to dentry.handle without locking - // dentry.dataMu. - rw.d.handleMu.RLock() - if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { - n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, srcs, rw.off) - rw.off += n - rw.d.dataMu.Lock() - if rw.off > rw.d.size { - atomic.StoreUint64(&rw.d.size, rw.off) - // The remote file's size will implicitly be extended to the correct - // value when we write back to it. - } - rw.d.dataMu.Unlock() - rw.d.handleMu.RUnlock() - return n, err - } - - // Otherwise write to/through the cache. - mf := rw.d.fs.mfp.MemoryFile() - rw.d.dataMu.Lock() - - // Compute the range to write (overflow-checked). - start := rw.off - end := rw.off + srcs.NumBytes() - if end <= rw.off { - end = math.MaxInt64 - } - - var ( - done uint64 - retErr error - ) - seg, gap := rw.d.cache.Find(rw.off) - for rw.off < end { - mr := memmap.MappableRange{rw.off, end} - switch { - case seg.Ok(): - // Get internal mappings from the cache. - segMR := seg.Range().Intersect(mr) - ims, err := mf.MapInternal(seg.FileRangeOf(segMR), usermem.Write) - if err != nil { - retErr = err - goto exitLoop - } - - // Copy to internal mappings. - n, err := safemem.CopySeq(ims, srcs) - done += n - rw.off += n - srcs = srcs.DropFirst64(n) - rw.d.dirty.MarkDirty(segMR) - if err != nil { - retErr = err - goto exitLoop - } - - // Continue. - seg, gap = seg.NextNonEmpty() - - case gap.Ok(): - // Write directly to the file. At present, we never fill the cache - // when writing, since doing so can convert small writes into - // inefficient read-modify-write cycles, and we have no mechanism - // for detecting or avoiding this. - gapMR := gap.Range().Intersect(mr) - gapSrcs := srcs.TakeFirst64(gapMR.Length()) - n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, gapSrcs, gapMR.Start) - done += n - rw.off += n - srcs = srcs.DropFirst64(n) - // Partial writes are fine. But we must stop writing. - if n != gapSrcs.NumBytes() || err != nil { - retErr = err - goto exitLoop - } - - // Continue. - seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} - } - } -exitLoop: - if rw.off > rw.d.size { - atomic.StoreUint64(&rw.d.size, rw.off) - // The remote file's size will implicitly be extended to the correct - // value when we write back to it. - } - // If InteropModeWritethrough is in effect, flush written data back to the - // remote filesystem. - if rw.d.fs.opts.interop == InteropModeWritethrough && done != 0 { - if err := fsutil.SyncDirty(rw.ctx, memmap.MappableRange{ - Start: start, - End: rw.off, - }, &rw.d.cache, &rw.d.dirty, rw.d.size, mf, rw.d.handle.writeFromBlocksAt); err != nil { - // We have no idea how many bytes were actually flushed. - rw.off = start - done = 0 - retErr = err - } - } - rw.d.dataMu.Unlock() - rw.d.handleMu.RUnlock() - return done, retErr -} - -func (d *dentry) writeback(ctx context.Context, offset, size int64) error { - if size == 0 { - return nil - } - d.handleMu.RLock() - defer d.handleMu.RUnlock() - d.dataMu.Lock() - defer d.dataMu.Unlock() - // Compute the range of valid bytes (overflow-checked). - if uint64(offset) >= d.size { - return nil - } - end := int64(d.size) - if rend := offset + size; rend > offset && rend < end { - end = rend - } - return fsutil.SyncDirty(ctx, memmap.MappableRange{ - Start: uint64(offset), - End: uint64(end), - }, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.mu.Lock() - defer fd.mu.Unlock() - switch whence { - case linux.SEEK_SET: - // Use offset as specified. - case linux.SEEK_CUR: - offset += fd.off - case linux.SEEK_END, linux.SEEK_DATA, linux.SEEK_HOLE: - // Ensure file size is up to date. - d := fd.dentry() - if fd.filesystem().opts.interop == InteropModeShared { - if err := d.updateFromGetattr(ctx); err != nil { - return 0, err - } - } - size := int64(atomic.LoadUint64(&d.size)) - // For SEEK_DATA and SEEK_HOLE, treat the file as a single contiguous - // block of data. - switch whence { - case linux.SEEK_END: - offset += size - case linux.SEEK_DATA: - if offset > size { - return 0, syserror.ENXIO - } - // Use offset as specified. - case linux.SEEK_HOLE: - if offset > size { - return 0, syserror.ENXIO - } - offset = size - } - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *regularFileFD) Sync(ctx context.Context) error { - return fd.dentry().syncSharedHandle(ctx) -} - -func (d *dentry) syncSharedHandle(ctx context.Context) error { - d.handleMu.RLock() - if !d.handleWritable { - d.handleMu.RUnlock() - return nil - } - d.dataMu.Lock() - // Write dirty cached data to the remote file. - err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) - d.dataMu.Unlock() - if err == nil { - // Sync the remote file. - err = d.handle.sync(ctx) - } - d.handleMu.RUnlock() - return err -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - d := fd.dentry() - switch d.fs.opts.interop { - case InteropModeExclusive: - // Any mapping is fine. - case InteropModeWritethrough: - // Shared writable mappings require a host FD, since otherwise we can't - // synchronously flush memory-mapped writes to the remote file. - if opts.Private || !opts.MaxPerms.Write { - break - } - fallthrough - case InteropModeShared: - // All mappings require a host FD to be coherent with other filesystem - // users. - if d.fs.opts.forcePageCache { - // Whether or not we have a host FD, we're not allowed to use it. - return syserror.ENODEV - } - d.handleMu.RLock() - haveFD := d.handle.fd >= 0 - d.handleMu.RUnlock() - if !haveFD { - return syserror.ENODEV - } - default: - panic(fmt.Sprintf("unknown InteropMode %v", d.fs.opts.interop)) - } - // After this point, d may be used as a memmap.Mappable. - d.pf.hostFileMapperInitOnce.Do(d.pf.hostFileMapper.Init) - return vfs.GenericConfigureMMap(&fd.vfsfd, d, opts) -} - -func (d *dentry) mayCachePages() bool { - if d.fs.opts.interop == InteropModeShared { - return false - } - if d.fs.opts.forcePageCache { - return true - } - d.handleMu.RLock() - haveFD := d.handle.fd >= 0 - d.handleMu.RUnlock() - return haveFD -} - -// AddMapping implements memmap.Mappable.AddMapping. -func (d *dentry) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { - d.mapsMu.Lock() - mapped := d.mappings.AddMapping(ms, ar, offset, writable) - // Do this unconditionally since whether we have a host FD can change - // across save/restore. - for _, r := range mapped { - d.pf.hostFileMapper.IncRefOn(r) - } - if d.mayCachePages() { - // d.Evict() will refuse to evict memory-mapped pages, so tell the - // MemoryFile to not bother trying. - mf := d.fs.mfp.MemoryFile() - for _, r := range mapped { - mf.MarkUnevictable(d, pgalloc.EvictableRange{r.Start, r.End}) - } - } - d.mapsMu.Unlock() - return nil -} - -// RemoveMapping implements memmap.Mappable.RemoveMapping. -func (d *dentry) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { - d.mapsMu.Lock() - unmapped := d.mappings.RemoveMapping(ms, ar, offset, writable) - for _, r := range unmapped { - d.pf.hostFileMapper.DecRefOn(r) - } - if d.mayCachePages() { - // Pages that are no longer referenced by any application memory - // mappings are now considered unused; allow MemoryFile to evict them - // when necessary. - mf := d.fs.mfp.MemoryFile() - d.dataMu.Lock() - for _, r := range unmapped { - // Since these pages are no longer mapped, they are no longer - // concurrently dirtyable by a writable memory mapping. - d.dirty.AllowClean(r) - mf.MarkEvictable(d, pgalloc.EvictableRange{r.Start, r.End}) - } - d.dataMu.Unlock() - } - d.mapsMu.Unlock() -} - -// CopyMapping implements memmap.Mappable.CopyMapping. -func (d *dentry) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { - return d.AddMapping(ctx, ms, dstAR, offset, writable) -} - -// Translate implements memmap.Mappable.Translate. -func (d *dentry) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { - d.handleMu.RLock() - if d.handle.fd >= 0 && !d.fs.opts.forcePageCache { - d.handleMu.RUnlock() - mr := optional - if d.fs.opts.limitHostFDTranslation { - mr = maxFillRange(required, optional) - } - return []memmap.Translation{ - { - Source: mr, - File: &d.pf, - Offset: mr.Start, - Perms: usermem.AnyAccess, - }, - }, nil - } - - d.dataMu.Lock() - - // Constrain translations to d.size (rounded up) to prevent translation to - // pages that may be concurrently truncated. - pgend := pageRoundUp(d.size) - var beyondEOF bool - if required.End > pgend { - if required.Start >= pgend { - d.dataMu.Unlock() - d.handleMu.RUnlock() - return nil, &memmap.BusError{io.EOF} - } - beyondEOF = true - required.End = pgend - } - if optional.End > pgend { - optional.End = pgend - } - - mf := d.fs.mfp.MemoryFile() - cerr := d.cache.Fill(ctx, required, maxFillRange(required, optional), mf, usage.PageCache, d.handle.readToBlocksAt) - - var ts []memmap.Translation - var translatedEnd uint64 - for seg := d.cache.FindSegment(required.Start); seg.Ok() && seg.Start() < required.End; seg, _ = seg.NextNonEmpty() { - segMR := seg.Range().Intersect(optional) - // TODO(jamieliu): Make Translations writable even if writability is - // not required if already kept-dirty by another writable translation. - perms := usermem.AccessType{ - Read: true, - Execute: true, - } - if at.Write { - // From this point forward, this memory can be dirtied through the - // mapping at any time. - d.dirty.KeepDirty(segMR) - perms.Write = true - } - ts = append(ts, memmap.Translation{ - Source: segMR, - File: mf, - Offset: seg.FileRangeOf(segMR).Start, - Perms: perms, - }) - translatedEnd = segMR.End - } - - d.dataMu.Unlock() - d.handleMu.RUnlock() - - // Don't return the error returned by c.cache.Fill if it occurred outside - // of required. - if translatedEnd < required.End && cerr != nil { - return ts, &memmap.BusError{cerr} - } - if beyondEOF { - return ts, &memmap.BusError{io.EOF} - } - return ts, nil -} - -func maxFillRange(required, optional memmap.MappableRange) memmap.MappableRange { - const maxReadahead = 64 << 10 // 64 KB, chosen arbitrarily - if required.Length() >= maxReadahead { - return required - } - if optional.Length() <= maxReadahead { - return optional - } - optional.Start = required.Start - if optional.Length() <= maxReadahead { - return optional - } - optional.End = optional.Start + maxReadahead - return optional -} - -// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. -func (d *dentry) InvalidateUnsavable(ctx context.Context) error { - // Whether we have a host fd (and consequently what platform.File is - // mapped) can change across save/restore, so invalidate all translations - // unconditionally. - d.mapsMu.Lock() - defer d.mapsMu.Unlock() - d.mappings.InvalidateAll(memmap.InvalidateOpts{}) - - // Write the cache's contents back to the remote file so that if we have a - // host fd after restore, the remote file's contents are coherent. - mf := d.fs.mfp.MemoryFile() - d.dataMu.Lock() - defer d.dataMu.Unlock() - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { - return err - } - - // Discard the cache so that it's not stored in saved state. This is safe - // because per InvalidateUnsavable invariants, no new translations can have - // been returned after we invalidated all existing translations above. - d.cache.DropAll(mf) - d.dirty.RemoveAll() - - return nil -} - -// Evict implements pgalloc.EvictableMemoryUser.Evict. -func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { - d.mapsMu.Lock() - defer d.mapsMu.Unlock() - d.dataMu.Lock() - defer d.dataMu.Unlock() - - mr := memmap.MappableRange{er.Start, er.End} - mf := d.fs.mfp.MemoryFile() - // Only allow pages that are no longer memory-mapped to be evicted. - for mgap := d.mappings.LowerBoundGap(mr.Start); mgap.Ok() && mgap.Start() < mr.End; mgap = mgap.NextGap() { - mgapMR := mgap.Range().Intersect(mr) - if mgapMR.Length() == 0 { - continue - } - if err := fsutil.SyncDirty(ctx, mgapMR, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { - log.Warningf("Failed to writeback cached data %v: %v", mgapMR, err) - } - d.cache.Drop(mgapMR, mf) - d.dirty.KeepClean(mgapMR) - } -} - -// dentryPlatformFile implements platform.File. It exists solely because dentry -// cannot implement both vfs.DentryImpl.IncRef and platform.File.IncRef. -// -// dentryPlatformFile is only used when a host FD representing the remote file -// is available (i.e. dentry.handle.fd >= 0), and that FD is used for -// application memory mappings (i.e. !filesystem.opts.forcePageCache). -type dentryPlatformFile struct { - *dentry - - // fdRefs counts references on platform.File offsets. fdRefs is protected - // by dentry.dataMu. - fdRefs fsutil.FrameRefSet - - // If this dentry represents a regular file, and handle.fd >= 0, - // hostFileMapper caches mappings of handle.fd. - hostFileMapper fsutil.HostFileMapper - - // hostFileMapperInitOnce is used to lazily initialize hostFileMapper. - hostFileMapperInitOnce sync.Once -} - -// IncRef implements platform.File.IncRef. -func (d *dentryPlatformFile) IncRef(fr platform.FileRange) { - d.dataMu.Lock() - seg, gap := d.fdRefs.Find(fr.Start) - for { - switch { - case seg.Ok() && seg.Start() < fr.End: - seg = d.fdRefs.Isolate(seg, fr) - seg.SetValue(seg.Value() + 1) - seg, gap = seg.NextNonEmpty() - case gap.Ok() && gap.Start() < fr.End: - newRange := gap.Range().Intersect(fr) - usage.MemoryAccounting.Inc(newRange.Length(), usage.Mapped) - seg, gap = d.fdRefs.InsertWithoutMerging(gap, newRange, 1).NextNonEmpty() - default: - d.fdRefs.MergeAdjacent(fr) - d.dataMu.Unlock() - return - } - } -} - -// DecRef implements platform.File.DecRef. -func (d *dentryPlatformFile) DecRef(fr platform.FileRange) { - d.dataMu.Lock() - seg := d.fdRefs.FindSegment(fr.Start) - - for seg.Ok() && seg.Start() < fr.End { - seg = d.fdRefs.Isolate(seg, fr) - if old := seg.Value(); old == 1 { - usage.MemoryAccounting.Dec(seg.Range().Length(), usage.Mapped) - seg = d.fdRefs.Remove(seg).NextSegment() - } else { - seg.SetValue(old - 1) - seg = seg.NextSegment() - } - } - d.fdRefs.MergeAdjacent(fr) - d.dataMu.Unlock() - -} - -// MapInternal implements platform.File.MapInternal. -func (d *dentryPlatformFile) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { - d.handleMu.RLock() - bs, err := d.hostFileMapper.MapInternal(fr, int(d.handle.fd), at.Write) - d.handleMu.RUnlock() - return bs, err -} - -// FD implements platform.File.FD. -func (d *dentryPlatformFile) FD() int { - d.handleMu.RLock() - fd := d.handle.fd - d.handleMu.RUnlock() - return int(fd) -} diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go deleted file mode 100644 index 274f7346f..000000000 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ /dev/null @@ -1,167 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "sync" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// specialFileFD implements vfs.FileDescriptionImpl for files other than -// regular files, directories, and symlinks: pipes, sockets, etc. It is also -// used for regular files when filesystemOptions.specialRegularFiles is in -// effect. specialFileFD differs from regularFileFD by using per-FD handles -// instead of shared per-dentry handles, and never buffering I/O. -type specialFileFD struct { - fileDescription - - // handle is immutable. - handle handle - - // off is the file offset. off is protected by mu. (POSIX 2.9.7 only - // requires operations using the file offset to be atomic for regular files - // and symlinks; however, since specialFileFD may be used for regular - // files, we apply this atomicity unconditionally.) - mu sync.Mutex - off int64 -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *specialFileFD) Release() { - fd.handle.close(context.Background()) - fs := fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) - fs.syncMu.Lock() - delete(fs.specialFileFDs, fd) - fs.syncMu.Unlock() -} - -// OnClose implements vfs.FileDescriptionImpl.OnClose. -func (fd *specialFileFD) OnClose(ctx context.Context) error { - if !fd.vfsfd.IsWritable() { - return nil - } - return fd.handle.file.flush(ctx) -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if opts.Flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - // Going through dst.CopyOutFrom() holds MM locks around file operations of - // unknown duration. For regularFileFD, doing so is necessary to support - // mmap due to lock ordering; MM locks precede dentry.dataMu. That doesn't - // hold here since specialFileFD doesn't client-cache data. Just buffer the - // read instead. - if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { - d.touchAtime(ctx, fd.vfsfd.Mount()) - } - buf := make([]byte, dst.NumBytes()) - n, err := fd.handle.readToBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) - if n == 0 { - return 0, err - } - if cp, cperr := dst.CopyOut(ctx, buf[:n]); cperr != nil { - return int64(cp), cperr - } - return int64(n), err -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - fd.mu.Lock() - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.off += n - fd.mu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if opts.Flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - if fd.dentry().fileType() == linux.S_IFREG { - limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) - if err != nil { - return 0, err - } - src = src.TakeFirst64(limit) - } - - // Do a buffered write. See rationale in PRead. - if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { - d.touchCMtime(ctx) - } - buf := make([]byte, src.NumBytes()) - // Don't do partial writes if we get a partial read from src. - if _, err := src.CopyIn(ctx, buf); err != nil { - return 0, err - } - n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) - return int64(n), err -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - fd.mu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n - fd.mu.Unlock() - return n, err -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.mu.Lock() - defer fd.mu.Unlock() - switch whence { - case linux.SEEK_SET: - // Use offset as given. - case linux.SEEK_CUR: - offset += fd.off - default: - // SEEK_END, SEEK_DATA, and SEEK_HOLE aren't supported since it's not - // clear that file size is even meaningful for these files. - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *specialFileFD) Sync(ctx context.Context) error { - if !fd.vfsfd.IsWritable() { - return nil - } - return fd.handle.sync(ctx) -} diff --git a/pkg/sentry/fsimpl/gofer/symlink.go b/pkg/sentry/fsimpl/gofer/symlink.go deleted file mode 100644 index adf43be60..000000000 --- a/pkg/sentry/fsimpl/gofer/symlink.go +++ /dev/null @@ -1,47 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func (d *dentry) isSymlink() bool { - return d.fileType() == linux.S_IFLNK -} - -// Precondition: d.isSymlink(). -func (d *dentry) readlink(ctx context.Context, mnt *vfs.Mount) (string, error) { - if d.fs.opts.interop != InteropModeShared { - d.touchAtime(ctx, mnt) - d.dataMu.Lock() - if d.haveTarget { - target := d.target - d.dataMu.Unlock() - return target, nil - } - } - target, err := d.file.readlink(ctx) - if d.fs.opts.interop != InteropModeShared { - if err == nil { - d.haveTarget = true - d.target = target - } - d.dataMu.Unlock() - } - return target, err -} diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go deleted file mode 100644 index 7598ec6a8..000000000 --- a/pkg/sentry/fsimpl/gofer/time.go +++ /dev/null @@ -1,75 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func dentryTimestampFromP9(s, ns uint64) int64 { - return int64(s*1e9 + ns) -} - -func dentryTimestampFromStatx(ts linux.StatxTimestamp) int64 { - return ts.Sec*1e9 + int64(ts.Nsec) -} - -func statxTimestampFromDentry(ns int64) linux.StatxTimestamp { - return linux.StatxTimestamp{ - Sec: ns / 1e9, - Nsec: uint32(ns % 1e9), - } -} - -func nowFromContext(ctx context.Context) (int64, bool) { - if clock := ktime.RealtimeClockFromContext(ctx); clock != nil { - return clock.Now().Nanoseconds(), true - } - return 0, false -} - -// Preconditions: fs.interop != InteropModeShared. -func (d *dentry) touchAtime(ctx context.Context, mnt *vfs.Mount) { - if err := mnt.CheckBeginWrite(); err != nil { - return - } - now, ok := nowFromContext(ctx) - if !ok { - mnt.EndWrite() - return - } - d.metadataMu.Lock() - atomic.StoreInt64(&d.atime, now) - d.metadataMu.Unlock() - mnt.EndWrite() -} - -// Preconditions: fs.interop != InteropModeShared. The caller has successfully -// called vfs.Mount.CheckBeginWrite(). -func (d *dentry) touchCMtime(ctx context.Context) { - now, ok := nowFromContext(ctx) - if !ok { - return - } - d.metadataMu.Lock() - atomic.StoreInt64(&d.mtime, now) - atomic.StoreInt64(&d.ctime, now) - d.metadataMu.Unlock() -} diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD deleted file mode 100644 index 82e1fb74b..000000000 --- a/pkg/sentry/fsimpl/host/BUILD +++ /dev/null @@ -1,34 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "host", - srcs = [ - "host.go", - "ioctl_unsafe.go", - "tty.go", - "util.go", - "util_unsafe.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fd", - "//pkg/log", - "//pkg/refs", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fsimpl/kernfs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/memmap", - "//pkg/sentry/unimpl", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "@org_golang_x_sys//unix:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go deleted file mode 100644 index a54985ef5..000000000 --- a/pkg/sentry/fsimpl/host/host.go +++ /dev/null @@ -1,616 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package host provides a filesystem implementation for host files imported as -// file descriptors. -package host - -import ( - "errors" - "fmt" - "math" - "syscall" - - "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - kernfs.Filesystem -} - -// NewMount returns a new disconnected mount in vfsObj that may be passed to ImportFD. -func NewMount(vfsObj *vfs.VirtualFilesystem) (*vfs.Mount, error) { - fs := &filesystem{} - fs.Init(vfsObj) - vfsfs := fs.VFSFilesystem() - // NewDisconnectedMount will take an additional reference on vfsfs. - defer vfsfs.DecRef() - return vfsObj.NewDisconnectedMount(vfsfs, nil, &vfs.MountOptions{}) -} - -// ImportFD sets up and returns a vfs.FileDescription from a donated fd. -func ImportFD(mnt *vfs.Mount, hostFD int, ownerUID auth.KUID, ownerGID auth.KGID, isTTY bool) (*vfs.FileDescription, error) { - fs, ok := mnt.Filesystem().Impl().(*kernfs.Filesystem) - if !ok { - return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl()) - } - - // Retrieve metadata. - var s syscall.Stat_t - if err := syscall.Fstat(hostFD, &s); err != nil { - return nil, err - } - - fileMode := linux.FileMode(s.Mode) - fileType := fileMode.FileType() - // Pipes, character devices, and sockets. - isStream := fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK - - i := &inode{ - hostFD: hostFD, - isStream: isStream, - isTTY: isTTY, - canMap: canMap(uint32(fileType)), - ino: fs.NextIno(), - mode: fileMode, - uid: ownerUID, - gid: ownerGID, - // For simplicity, set offset to 0. Technically, we should - // only set to 0 on files that are not seekable (sockets, pipes, etc.), - // and use the offset from the host fd otherwise. - offset: 0, - } - - // These files can't be memory mapped, assert this. - if i.isStream && i.canMap { - panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") - } - - d := &kernfs.Dentry{} - d.Init(i) - // i.open will take a reference on d. - defer d.DecRef() - - return i.open(d.VFSDentry(), mnt) -} - -// inode implements kernfs.Inode. -type inode struct { - kernfs.InodeNotDirectory - kernfs.InodeNotSymlink - - // When the reference count reaches zero, the host fd is closed. - refs.AtomicRefCount - - // hostFD contains the host fd that this file was originally created from, - // which must be available at time of restore. - // - // This field is initialized at creation time and is immutable. - hostFD int - - // isStream is true if the host fd points to a file representing a stream, - // e.g. a socket or a pipe. Such files are not seekable and can return - // EWOULDBLOCK for I/O operations. - // - // This field is initialized at creation time and is immutable. - isStream bool - - // isTTY is true if this file represents a TTY. - // - // This field is initialized at creation time and is immutable. - isTTY bool - - // canMap specifies whether we allow the file to be memory mapped. - // - // This field is initialized at creation time and is immutable. - canMap bool - - // ino is an inode number unique within this filesystem. - // - // This field is initialized at creation time and is immutable. - ino uint64 - - // TODO(gvisor.dev/issue/1672): protect mode, uid, and gid with mutex. - - // mode is the file mode of this inode. Note that this value may become out - // of date if the mode is changed on the host, e.g. with chmod. - mode linux.FileMode - - // uid and gid of the file owner. Note that these refer to the owner of the - // file created on import, not the fd on the host. - uid auth.KUID - gid auth.KGID - - // offsetMu protects offset. - offsetMu sync.Mutex - - // offset specifies the current file offset. - offset int64 -} - -// Note that these flags may become out of date, since they can be modified -// on the host, e.g. with fcntl. -func fileFlagsFromHostFD(fd int) (int, error) { - flags, err := unix.FcntlInt(uintptr(fd), syscall.F_GETFL, 0) - if err != nil { - log.Warningf("Failed to get file flags for donated FD %d: %v", fd, err) - return 0, err - } - // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. - flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND - return flags, nil -} - -// CheckPermissions implements kernfs.Inode. -func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - return vfs.GenericCheckPermissions(creds, ats, i.mode, i.uid, i.gid) -} - -// Mode implements kernfs.Inode. -func (i *inode) Mode() linux.FileMode { - return i.mode -} - -// Stat implements kernfs.Inode. -func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - if opts.Mask&linux.STATX__RESERVED != 0 { - return linux.Statx{}, syserror.EINVAL - } - if opts.Sync&linux.AT_STATX_SYNC_TYPE == linux.AT_STATX_SYNC_TYPE { - return linux.Statx{}, syserror.EINVAL - } - - // Limit our host call only to known flags. - mask := opts.Mask & linux.STATX_ALL - var s unix.Statx_t - err := unix.Statx(i.hostFD, "", int(unix.AT_EMPTY_PATH|opts.Sync), int(mask), &s) - // Fallback to fstat(2), if statx(2) is not supported on the host. - // - // TODO(b/151263641): Remove fallback. - if err == syserror.ENOSYS { - return i.fstat(opts) - } else if err != nil { - return linux.Statx{}, err - } - - ls := linux.Statx{Mask: mask} - // Unconditionally fill blksize, attributes, and device numbers, as indicated - // by /include/uapi/linux/stat.h. - // - // RdevMajor/RdevMinor are left as zero, so as not to expose host device - // numbers. - // - // TODO(gvisor.dev/issue/1672): Use kernfs-specific, internally defined - // device numbers. If we use the device number from the host, it may collide - // with another sentry-internal device number. We handle device/inode - // numbers without relying on the host to prevent collisions. - ls.Blksize = s.Blksize - ls.Attributes = s.Attributes - ls.AttributesMask = s.Attributes_mask - - if mask|linux.STATX_TYPE != 0 { - ls.Mode |= s.Mode & linux.S_IFMT - } - if mask|linux.STATX_MODE != 0 { - ls.Mode |= s.Mode &^ linux.S_IFMT - } - if mask|linux.STATX_NLINK != 0 { - ls.Nlink = s.Nlink - } - if mask|linux.STATX_ATIME != 0 { - ls.Atime = unixToLinuxStatxTimestamp(s.Atime) - } - if mask|linux.STATX_BTIME != 0 { - ls.Btime = unixToLinuxStatxTimestamp(s.Btime) - } - if mask|linux.STATX_CTIME != 0 { - ls.Ctime = unixToLinuxStatxTimestamp(s.Ctime) - } - if mask|linux.STATX_MTIME != 0 { - ls.Mtime = unixToLinuxStatxTimestamp(s.Mtime) - } - if mask|linux.STATX_SIZE != 0 { - ls.Size = s.Size - } - if mask|linux.STATX_BLOCKS != 0 { - ls.Blocks = s.Blocks - } - - // Use our own internal inode number and file owner. - if mask|linux.STATX_INO != 0 { - ls.Ino = i.ino - } - if mask|linux.STATX_UID != 0 { - ls.UID = uint32(i.uid) - } - if mask|linux.STATX_GID != 0 { - ls.GID = uint32(i.gid) - } - - return ls, nil -} - -// fstat is a best-effort fallback for inode.Stat() if the host does not -// support statx(2). -// -// We ignore the mask and sync flags in opts and simply supply -// STATX_BASIC_STATS, as fstat(2) itself does not allow the specification -// of a mask or sync flags. fstat(2) does not provide any metadata -// equivalent to Statx.Attributes, Statx.AttributesMask, or Statx.Btime, so -// those fields remain empty. -func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { - var s unix.Stat_t - if err := unix.Fstat(i.hostFD, &s); err != nil { - return linux.Statx{}, err - } - - // Note that rdev numbers are left as 0; do not expose host device numbers. - ls := linux.Statx{ - Mask: linux.STATX_BASIC_STATS, - Blksize: uint32(s.Blksize), - Nlink: uint32(s.Nlink), - Mode: uint16(s.Mode), - Size: uint64(s.Size), - Blocks: uint64(s.Blocks), - Atime: timespecToStatxTimestamp(s.Atim), - Ctime: timespecToStatxTimestamp(s.Ctim), - Mtime: timespecToStatxTimestamp(s.Mtim), - } - - // Use our own internal inode number and file owner. - // - // TODO(gvisor.dev/issue/1672): Use a kernfs-specific device number as well. - // If we use the device number from the host, it may collide with another - // sentry-internal device number. We handle device/inode numbers without - // relying on the host to prevent collisions. - ls.Ino = i.ino - ls.UID = uint32(i.uid) - ls.GID = uint32(i.gid) - - return ls, nil -} - -// SetStat implements kernfs.Inode. -func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { - s := opts.Stat - - m := s.Mask - if m == 0 { - return nil - } - if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 { - return syserror.EPERM - } - if err := vfs.CheckSetStat(ctx, creds, &s, i.Mode(), i.uid, i.gid); err != nil { - return err - } - - if m&linux.STATX_MODE != 0 { - if err := syscall.Fchmod(i.hostFD, uint32(s.Mode)); err != nil { - return err - } - i.mode = linux.FileMode(s.Mode) - } - if m&linux.STATX_SIZE != 0 { - if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil { - return err - } - } - if m&(linux.STATX_ATIME|linux.STATX_MTIME) != 0 { - ts := [2]syscall.Timespec{ - toTimespec(s.Atime, m&linux.STATX_ATIME == 0), - toTimespec(s.Mtime, m&linux.STATX_MTIME == 0), - } - if err := setTimestamps(i.hostFD, &ts); err != nil { - return err - } - } - return nil -} - -// DecRef implements kernfs.Inode. -func (i *inode) DecRef() { - i.AtomicRefCount.DecRefWithDestructor(i.Destroy) -} - -// Destroy implements kernfs.Inode. -func (i *inode) Destroy() { - if err := unix.Close(i.hostFD); err != nil { - log.Warningf("failed to close host fd %d: %v", i.hostFD, err) - } -} - -// Open implements kernfs.Inode. -func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - return i.open(vfsd, rp.Mount()) -} - -func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { - fileType := i.mode.FileType() - if fileType == syscall.S_IFSOCK { - if i.isTTY { - return nil, errors.New("cannot use host socket as TTY") - } - // TODO(gvisor.dev/issue/1672): support importing sockets. - return nil, errors.New("importing host sockets not supported") - } - - // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that - // we don't allow importing arbitrary file types without proper support. - var ( - vfsfd *vfs.FileDescription - fdImpl vfs.FileDescriptionImpl - ) - if i.isTTY { - fd := &ttyFD{ - fileDescription: fileDescription{inode: i}, - termios: linux.DefaultSlaveTermios, - } - vfsfd = &fd.vfsfd - fdImpl = fd - } else { - // For simplicity, set offset to 0. Technically, we should - // only set to 0 on files that are not seekable (sockets, pipes, etc.), - // and use the offset from the host fd otherwise. - fd := &fileDescription{inode: i} - vfsfd = &fd.vfsfd - fdImpl = fd - } - - flags, err := fileFlagsFromHostFD(i.hostFD) - if err != nil { - return nil, err - } - - if err := vfsfd.Init(fdImpl, uint32(flags), mnt, d, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return vfsfd, nil -} - -// fileDescription is embedded by host fd implementations of FileDescriptionImpl. -// -// TODO(gvisor.dev/issue/1672): Implement Waitable interface. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl - - // inode is vfsfd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode), but - // cached to reduce indirections and casting. fileDescription does not hold - // a reference on the inode through the inode field (since one is already - // held via the Dentry). - // - // inode is immutable after fileDescription creation. - inode *inode -} - -// SetStat implements vfs.FileDescriptionImpl. -func (f *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - creds := auth.CredentialsFromContext(ctx) - return f.inode.SetStat(ctx, nil, creds, opts) -} - -// Stat implements vfs.FileDescriptionImpl. -func (f *fileDescription) Stat(_ context.Context, opts vfs.StatOptions) (linux.Statx, error) { - return f.inode.Stat(nil, opts) -} - -// Release implements vfs.FileDescriptionImpl. -func (f *fileDescription) Release() { - // noop -} - -// PRead implements FileDescriptionImpl. -func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - i := f.inode - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if i.isStream { - return 0, syserror.ESPIPE - } - - return readFromHostFD(ctx, i.hostFD, dst, offset, opts.Flags) -} - -// Read implements FileDescriptionImpl. -func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - i := f.inode - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if i.isStream { - n, err := readFromHostFD(ctx, i.hostFD, dst, -1, opts.Flags) - if isBlockError(err) { - // If we got any data at all, return it as a "completed" partial read - // rather than retrying until complete. - if n != 0 { - err = nil - } else { - err = syserror.ErrWouldBlock - } - } - return n, err - } - // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. - i.offsetMu.Lock() - n, err := readFromHostFD(ctx, i.hostFD, dst, i.offset, opts.Flags) - i.offset += n - i.offsetMu.Unlock() - return n, err -} - -func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, offset int64, flags uint32) (int64, error) { - // TODO(gvisor.dev/issue/1672): Support select preadv2 flags. - if flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - var reader safemem.Reader - if offset == -1 { - reader = safemem.FromIOReader{fd.NewReadWriter(hostFD)} - } else { - reader = safemem.FromVecReaderFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Preadv(hostFD, srcs, offset) - return int64(n), err - }, - } - } - n, err := dst.CopyOutFrom(ctx, reader) - return int64(n), err -} - -// PWrite implements FileDescriptionImpl. -func (f *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - i := f.inode - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if i.isStream { - return 0, syserror.ESPIPE - } - - return writeToHostFD(ctx, i.hostFD, src, offset, opts.Flags) -} - -// Write implements FileDescriptionImpl. -func (f *fileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - i := f.inode - // TODO(b/34716638): Some char devices do support offsets, e.g. /dev/null. - if i.isStream { - n, err := writeToHostFD(ctx, i.hostFD, src, -1, opts.Flags) - if isBlockError(err) { - err = syserror.ErrWouldBlock - } - return n, err - } - // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. - // TODO(gvisor.dev/issue/1672): Write to end of file and update offset if O_APPEND is set on this file. - i.offsetMu.Lock() - n, err := writeToHostFD(ctx, i.hostFD, src, i.offset, opts.Flags) - i.offset += n - i.offsetMu.Unlock() - return n, err -} - -func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offset int64, flags uint32) (int64, error) { - // TODO(gvisor.dev/issue/1672): Support select pwritev2 flags. - if flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - var writer safemem.Writer - if offset == -1 { - writer = safemem.FromIOWriter{fd.NewReadWriter(hostFD)} - } else { - writer = safemem.FromVecWriterFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Pwritev(hostFD, srcs, offset) - return int64(n), err - }, - } - } - n, err := src.CopyInTo(ctx, writer) - return int64(n), err -} - -// Seek implements FileDescriptionImpl. -// -// Note that we do not support seeking on directories, since we do not even -// allow directory fds to be imported at all. -func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (int64, error) { - i := f.inode - // TODO(b/34716638): Some char devices do support seeking, e.g. /dev/null. - if i.isStream { - return 0, syserror.ESPIPE - } - - i.offsetMu.Lock() - defer i.offsetMu.Unlock() - - switch whence { - case linux.SEEK_SET: - if offset < 0 { - return i.offset, syserror.EINVAL - } - i.offset = offset - - case linux.SEEK_CUR: - // Check for overflow. Note that underflow cannot occur, since i.offset >= 0. - if offset > math.MaxInt64-i.offset { - return i.offset, syserror.EOVERFLOW - } - if i.offset+offset < 0 { - return i.offset, syserror.EINVAL - } - i.offset += offset - - case linux.SEEK_END: - var s syscall.Stat_t - if err := syscall.Fstat(i.hostFD, &s); err != nil { - return i.offset, err - } - size := s.Size - - // Check for overflow. Note that underflow cannot occur, since size >= 0. - if offset > math.MaxInt64-size { - return i.offset, syserror.EOVERFLOW - } - if size+offset < 0 { - return i.offset, syserror.EINVAL - } - i.offset = size + offset - - case linux.SEEK_DATA, linux.SEEK_HOLE: - // Modifying the offset in the host file table should not matter, since - // this is the only place where we use it. - // - // For reading and writing, we always rely on our internal offset. - n, err := unix.Seek(i.hostFD, offset, int(whence)) - if err != nil { - return i.offset, err - } - i.offset = n - - default: - // Invalid whence. - return i.offset, syserror.EINVAL - } - - return i.offset, nil -} - -// Sync implements FileDescriptionImpl. -func (f *fileDescription) Sync(context.Context) error { - // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData optimization, so we always sync everything. - return unix.Fsync(f.inode.hostFD) -} - -// ConfigureMMap implements FileDescriptionImpl. -func (f *fileDescription) ConfigureMMap(_ context.Context, opts *memmap.MMapOpts) error { - if !f.inode.canMap { - return syserror.ENODEV - } - // TODO(gvisor.dev/issue/1672): Implement ConfigureMMap and Mappable interface. - return syserror.ENODEV -} diff --git a/pkg/sentry/fsimpl/host/ioctl_unsafe.go b/pkg/sentry/fsimpl/host/ioctl_unsafe.go deleted file mode 100644 index 0983bf7d8..000000000 --- a/pkg/sentry/fsimpl/host/ioctl_unsafe.go +++ /dev/null @@ -1,56 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "syscall" - "unsafe" - - "gvisor.dev/gvisor/pkg/abi/linux" -) - -func ioctlGetTermios(fd int) (*linux.Termios, error) { - var t linux.Termios - _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TCGETS, uintptr(unsafe.Pointer(&t))) - if errno != 0 { - return nil, errno - } - return &t, nil -} - -func ioctlSetTermios(fd int, req uint64, t *linux.Termios) error { - _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), uintptr(req), uintptr(unsafe.Pointer(t))) - if errno != 0 { - return errno - } - return nil -} - -func ioctlGetWinsize(fd int) (*linux.Winsize, error) { - var w linux.Winsize - _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TIOCGWINSZ, uintptr(unsafe.Pointer(&w))) - if errno != 0 { - return nil, errno - } - return &w, nil -} - -func ioctlSetWinsize(fd int, w *linux.Winsize) error { - _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TIOCSWINSZ, uintptr(unsafe.Pointer(w))) - if errno != 0 { - return errno - } - return nil -} diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go deleted file mode 100644 index 8936afb06..000000000 --- a/pkg/sentry/fsimpl/host/tty.go +++ /dev/null @@ -1,379 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/unimpl" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// ttyFD implements vfs.FileDescriptionImpl for a host file descriptor -// that wraps a TTY FD. -type ttyFD struct { - fileDescription - - // mu protects the fields below. - mu sync.Mutex `state:"nosave"` - - // session is the session attached to this ttyFD. - session *kernel.Session - - // fgProcessGroup is the foreground process group that is currently - // connected to this TTY. - fgProcessGroup *kernel.ProcessGroup - - // termios contains the terminal attributes for this TTY. - termios linux.KernelTermios -} - -// InitForegroundProcessGroup sets the foreground process group and session for -// the TTY. This should only be called once, after the foreground process group -// has been created, but before it has started running. -func (t *ttyFD) InitForegroundProcessGroup(pg *kernel.ProcessGroup) { - t.mu.Lock() - defer t.mu.Unlock() - if t.fgProcessGroup != nil { - panic("foreground process group is already set") - } - t.fgProcessGroup = pg - t.session = pg.Session() -} - -// ForegroundProcessGroup returns the foreground process for the TTY. -func (t *ttyFD) ForegroundProcessGroup() *kernel.ProcessGroup { - t.mu.Lock() - defer t.mu.Unlock() - return t.fgProcessGroup -} - -// Release implements fs.FileOperations.Release. -func (t *ttyFD) Release() { - t.mu.Lock() - t.fgProcessGroup = nil - t.mu.Unlock() - - t.fileDescription.Release() -} - -// PRead implements vfs.FileDescriptionImpl. -// -// Reading from a TTY is only allowed for foreground process groups. Background -// process groups will either get EIO or a SIGTTIN. -func (t *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - t.mu.Lock() - defer t.mu.Unlock() - - // Are we allowed to do the read? - // drivers/tty/n_tty.c:n_tty_read()=>job_control()=>tty_check_change(). - if err := t.checkChange(ctx, linux.SIGTTIN); err != nil { - return 0, err - } - - // Do the read. - return t.fileDescription.PRead(ctx, dst, offset, opts) -} - -// Read implements vfs.FileDescriptionImpl. -// -// Reading from a TTY is only allowed for foreground process groups. Background -// process groups will either get EIO or a SIGTTIN. -func (t *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - t.mu.Lock() - defer t.mu.Unlock() - - // Are we allowed to do the read? - // drivers/tty/n_tty.c:n_tty_read()=>job_control()=>tty_check_change(). - if err := t.checkChange(ctx, linux.SIGTTIN); err != nil { - return 0, err - } - - // Do the read. - return t.fileDescription.Read(ctx, dst, opts) -} - -// PWrite implements vfs.FileDescriptionImpl. -func (t *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - t.mu.Lock() - defer t.mu.Unlock() - - // Check whether TOSTOP is enabled. This corresponds to the check in - // drivers/tty/n_tty.c:n_tty_write(). - if t.termios.LEnabled(linux.TOSTOP) { - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - return 0, err - } - } - return t.fileDescription.PWrite(ctx, src, offset, opts) -} - -// Write implements vfs.FileDescriptionImpl. -func (t *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - t.mu.Lock() - defer t.mu.Unlock() - - // Check whether TOSTOP is enabled. This corresponds to the check in - // drivers/tty/n_tty.c:n_tty_write(). - if t.termios.LEnabled(linux.TOSTOP) { - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - return 0, err - } - } - return t.fileDescription.Write(ctx, src, opts) -} - -// Ioctl implements vfs.FileDescriptionImpl. -func (t *ttyFD) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { - // Ignore arg[0]. This is the real FD: - fd := t.inode.hostFD - ioctl := args[1].Uint64() - switch ioctl { - case linux.TCGETS: - termios, err := ioctlGetTermios(fd) - if err != nil { - return 0, err - } - _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), termios, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err - - case linux.TCSETS, linux.TCSETSW, linux.TCSETSF: - t.mu.Lock() - defer t.mu.Unlock() - - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - return 0, err - } - - var termios linux.Termios - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &termios, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - err := ioctlSetTermios(fd, ioctl, &termios) - if err == nil { - t.termios.FromTermios(termios) - } - return 0, err - - case linux.TIOCGPGRP: - // Args: pid_t *argp - // When successful, equivalent to *argp = tcgetpgrp(fd). - // Get the process group ID of the foreground process group on this - // terminal. - - pidns := kernel.PIDNamespaceFromContext(ctx) - if pidns == nil { - return 0, syserror.ENOTTY - } - - t.mu.Lock() - defer t.mu.Unlock() - - // Map the ProcessGroup into a ProcessGroupID in the task's PID namespace. - pgID := pidns.IDOfProcessGroup(t.fgProcessGroup) - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &pgID, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err - - case linux.TIOCSPGRP: - // Args: const pid_t *argp - // Equivalent to tcsetpgrp(fd, *argp). - // Set the foreground process group ID of this terminal. - - task := kernel.TaskFromContext(ctx) - if task == nil { - return 0, syserror.ENOTTY - } - - t.mu.Lock() - defer t.mu.Unlock() - - // Check that we are allowed to set the process group. - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - // drivers/tty/tty_io.c:tiocspgrp() converts -EIO from tty_check_change() - // to -ENOTTY. - if err == syserror.EIO { - return 0, syserror.ENOTTY - } - return 0, err - } - - // Check that calling task's process group is in the TTY session. - if task.ThreadGroup().Session() != t.session { - return 0, syserror.ENOTTY - } - - var pgID kernel.ProcessGroupID - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pgID, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - - // pgID must be non-negative. - if pgID < 0 { - return 0, syserror.EINVAL - } - - // Process group with pgID must exist in this PID namespace. - pidns := task.PIDNamespace() - pg := pidns.ProcessGroupWithID(pgID) - if pg == nil { - return 0, syserror.ESRCH - } - - // Check that new process group is in the TTY session. - if pg.Session() != t.session { - return 0, syserror.EPERM - } - - t.fgProcessGroup = pg - return 0, nil - - case linux.TIOCGWINSZ: - // Args: struct winsize *argp - // Get window size. - winsize, err := ioctlGetWinsize(fd) - if err != nil { - return 0, err - } - _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), winsize, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err - - case linux.TIOCSWINSZ: - // Args: const struct winsize *argp - // Set window size. - - // Unlike setting the termios, any process group (even background ones) can - // set the winsize. - - var winsize linux.Winsize - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &winsize, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - err := ioctlSetWinsize(fd, &winsize) - return 0, err - - // Unimplemented commands. - case linux.TIOCSETD, - linux.TIOCSBRK, - linux.TIOCCBRK, - linux.TCSBRK, - linux.TCSBRKP, - linux.TIOCSTI, - linux.TIOCCONS, - linux.FIONBIO, - linux.TIOCEXCL, - linux.TIOCNXCL, - linux.TIOCGEXCL, - linux.TIOCNOTTY, - linux.TIOCSCTTY, - linux.TIOCGSID, - linux.TIOCGETD, - linux.TIOCVHANGUP, - linux.TIOCGDEV, - linux.TIOCMGET, - linux.TIOCMSET, - linux.TIOCMBIC, - linux.TIOCMBIS, - linux.TIOCGICOUNT, - linux.TCFLSH, - linux.TIOCSSERIAL, - linux.TIOCGPTPEER: - - unimpl.EmitUnimplementedEvent(ctx) - fallthrough - default: - return 0, syserror.ENOTTY - } -} - -// checkChange checks that the process group is allowed to read, write, or -// change the state of the TTY. -// -// This corresponds to Linux drivers/tty/tty_io.c:tty_check_change(). The logic -// is a bit convoluted, but documented inline. -// -// Preconditions: t.mu must be held. -func (t *ttyFD) checkChange(ctx context.Context, sig linux.Signal) error { - task := kernel.TaskFromContext(ctx) - if task == nil { - // No task? Linux does not have an analog for this case, but - // tty_check_change is more of a blacklist of cases than a - // whitelist, and is surprisingly permissive. Allowing the - // change seems most appropriate. - return nil - } - - tg := task.ThreadGroup() - pg := tg.ProcessGroup() - - // If the session for the task is different than the session for the - // controlling TTY, then the change is allowed. Seems like a bad idea, - // but that's exactly what linux does. - if tg.Session() != t.fgProcessGroup.Session() { - return nil - } - - // If we are the foreground process group, then the change is allowed. - if pg == t.fgProcessGroup { - return nil - } - - // We are not the foreground process group. - - // Is the provided signal blocked or ignored? - if (task.SignalMask()&linux.SignalSetOf(sig) != 0) || tg.SignalHandlers().IsIgnored(sig) { - // If the signal is SIGTTIN, then we are attempting to read - // from the TTY. Don't send the signal and return EIO. - if sig == linux.SIGTTIN { - return syserror.EIO - } - - // Otherwise, we are writing or changing terminal state. This is allowed. - return nil - } - - // If the process group is an orphan, return EIO. - if pg.IsOrphan() { - return syserror.EIO - } - - // Otherwise, send the signal to the process group and return ERESTARTSYS. - // - // Note that Linux also unconditionally sets TIF_SIGPENDING on current, - // but this isn't necessary in gVisor because the rationale given in - // 040b6362d58f "tty: fix leakage of -ERESTARTSYS to userland" doesn't - // apply: the sentry will handle -ERESTARTSYS in - // kernel.runApp.execute() even if the kernel.Task isn't interrupted. - // - // Linux ignores the result of kill_pgrp(). - _ = pg.SendSignal(kernel.SignalInfoPriv(sig)) - return kernel.ERESTARTSYS -} diff --git a/pkg/sentry/fsimpl/host/util.go b/pkg/sentry/fsimpl/host/util.go deleted file mode 100644 index 2bc757b1a..000000000 --- a/pkg/sentry/fsimpl/host/util.go +++ /dev/null @@ -1,66 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "syscall" - - "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/syserror" -) - -func toTimespec(ts linux.StatxTimestamp, omit bool) syscall.Timespec { - if omit { - return syscall.Timespec{ - Sec: 0, - Nsec: unix.UTIME_OMIT, - } - } - return syscall.Timespec{ - Sec: ts.Sec, - Nsec: int64(ts.Nsec), - } -} - -func unixToLinuxStatxTimestamp(ts unix.StatxTimestamp) linux.StatxTimestamp { - return linux.StatxTimestamp{Sec: ts.Sec, Nsec: ts.Nsec} -} - -func timespecToStatxTimestamp(ts unix.Timespec) linux.StatxTimestamp { - return linux.StatxTimestamp{Sec: int64(ts.Sec), Nsec: uint32(ts.Nsec)} -} - -// wouldBlock returns true for file types that can return EWOULDBLOCK -// for blocking operations, e.g. pipes, character devices, and sockets. -func wouldBlock(fileType uint32) bool { - return fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK -} - -// canMap returns true if a file with fileType is allowed to be memory mapped. -// This is ported over from VFS1, but it's probably not the best way for us -// to check if a file can be memory mapped. -func canMap(fileType uint32) bool { - // TODO(gvisor.dev/issue/1672): Also allow "special files" to be mapped (see fs/host:canMap()). - // - // TODO(b/38213152): Some obscure character devices can be mapped. - return fileType == syscall.S_IFREG -} - -// isBlockError checks if an error is EAGAIN or EWOULDBLOCK. -// If so, they can be transformed into syserror.ErrWouldBlock. -func isBlockError(err error) bool { - return err == syserror.EAGAIN || err == syserror.EWOULDBLOCK -} diff --git a/pkg/sentry/fsimpl/host/util_unsafe.go b/pkg/sentry/fsimpl/host/util_unsafe.go deleted file mode 100644 index 5136ac844..000000000 --- a/pkg/sentry/fsimpl/host/util_unsafe.go +++ /dev/null @@ -1,34 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "syscall" - "unsafe" -) - -func setTimestamps(fd int, ts *[2]syscall.Timespec) error { - _, _, errno := syscall.Syscall6( - syscall.SYS_UTIMENSAT, - uintptr(fd), - 0, /* path */ - uintptr(unsafe.Pointer(ts)), - 0, /* flags */ - 0, 0) - if errno != 0 { - return errno - } - return nil -} diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD deleted file mode 100644 index e73f1f857..000000000 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ /dev/null @@ -1,61 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -licenses(["notice"]) - -go_template_instance( - name = "slot_list", - out = "slot_list.go", - package = "kernfs", - prefix = "slot", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*slot", - "Linker": "*slot", - }, -) - -go_library( - name = "kernfs", - srcs = [ - "dynamic_bytes_file.go", - "fd_impl_util.go", - "filesystem.go", - "inode_impl_util.go", - "kernfs.go", - "slot_list.go", - "symlink.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/log", - "//pkg/refs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/memmap", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -go_test( - name = "kernfs_test", - size = "small", - srcs = ["kernfs_test.go"], - deps = [ - ":kernfs", - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/testutil", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - "@com_github_google_go-cmp//cmp:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go deleted file mode 100644 index d8bddbafa..000000000 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ /dev/null @@ -1,133 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernfs - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// DynamicBytesFile implements kernfs.Inode and represents a read-only -// file whose contents are backed by a vfs.DynamicBytesSource. -// -// Must be instantiated with NewDynamicBytesFile or initialized with Init -// before first use. -// -// +stateify savable -type DynamicBytesFile struct { - InodeAttrs - InodeNoopRefCount - InodeNotDirectory - InodeNotSymlink - - data vfs.DynamicBytesSource -} - -var _ Inode = (*DynamicBytesFile)(nil) - -// Init initializes a dynamic bytes file. -func (f *DynamicBytesFile) Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) { - if perm&^linux.PermissionsMask != 0 { - panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) - } - f.InodeAttrs.Init(creds, ino, linux.ModeRegular|perm) - f.data = data -} - -// Open implements Inode.Open. -func (f *DynamicBytesFile) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &DynamicBytesFD{} - if err := fd.Init(rp.Mount(), vfsd, f.data, opts.Flags); err != nil { - return nil, err - } - return &fd.vfsfd, nil -} - -// SetStat implements Inode.SetStat. By default DynamicBytesFile doesn't allow -// inode attributes to be changed. Override SetStat() making it call -// f.InodeAttrs to allow it. -func (*DynamicBytesFile) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -// DynamicBytesFD implements vfs.FileDescriptionImpl for an FD backed by a -// DynamicBytesFile. -// -// Must be initialized with Init before first use. -// -// +stateify savable -type DynamicBytesFD struct { - vfs.FileDescriptionDefaultImpl - vfs.DynamicBytesFileDescriptionImpl - - vfsfd vfs.FileDescription - inode Inode -} - -// Init initializes a DynamicBytesFD. -func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, flags uint32) error { - if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { - return err - } - fd.inode = d.Impl().(*Dentry).inode - fd.SetDataSource(data) - return nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *DynamicBytesFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - return fd.DynamicBytesFileDescriptionImpl.Seek(ctx, offset, whence) -} - -// Read implmenets vfs.FileDescriptionImpl.Read. -func (fd *DynamicBytesFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return fd.DynamicBytesFileDescriptionImpl.Read(ctx, dst, opts) -} - -// PRead implmenets vfs.FileDescriptionImpl.PRead. -func (fd *DynamicBytesFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return fd.DynamicBytesFileDescriptionImpl.PRead(ctx, dst, offset, opts) -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *DynamicBytesFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return fd.DynamicBytesFileDescriptionImpl.Write(ctx, src, opts) -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *DynamicBytesFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return fd.DynamicBytesFileDescriptionImpl.PWrite(ctx, src, offset, opts) -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *DynamicBytesFD) Release() {} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *DynamicBytesFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - fs := fd.vfsfd.VirtualDentry().Mount().Filesystem() - return fd.inode.Stat(fs, opts) -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *DynamicBytesFD) SetStat(context.Context, vfs.SetStatOptions) error { - // DynamicBytesFiles are immutable. - return syserror.EPERM -} diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go deleted file mode 100644 index bfa786c88..000000000 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ /dev/null @@ -1,212 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// GenericDirectoryFD implements vfs.FileDescriptionImpl for a generic directory -// inode that uses OrderChildren to track child nodes. GenericDirectoryFD is not -// compatible with dynamic directories. -// -// Note that GenericDirectoryFD holds a lock over OrderedChildren while calling -// IterDirents callback. The IterDirents callback therefore cannot hash or -// unhash children, or recursively call IterDirents on the same underlying -// inode. -// -// Must be initialize with Init before first use. -type GenericDirectoryFD struct { - vfs.FileDescriptionDefaultImpl - vfs.DirectoryFileDescriptionDefaultImpl - - vfsfd vfs.FileDescription - children *OrderedChildren - off int64 -} - -// Init initializes a GenericDirectoryFD. -func (fd *GenericDirectoryFD) Init(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, opts *vfs.OpenOptions) error { - if vfs.AccessTypesForOpenFlags(opts)&vfs.MayWrite != 0 { - // Can't open directories for writing. - return syserror.EISDIR - } - if err := fd.vfsfd.Init(fd, opts.Flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { - return err - } - fd.children = children - return nil -} - -// VFSFileDescription returns a pointer to the vfs.FileDescription representing -// this object. -func (fd *GenericDirectoryFD) VFSFileDescription() *vfs.FileDescription { - return &fd.vfsfd -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *GenericDirectoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - return fd.FileDescriptionDefaultImpl.ConfigureMMap(ctx, opts) -} - -// Read implmenets vfs.FileDescriptionImpl.Read. -func (fd *GenericDirectoryFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return fd.DirectoryFileDescriptionDefaultImpl.Read(ctx, dst, opts) -} - -// PRead implmenets vfs.FileDescriptionImpl.PRead. -func (fd *GenericDirectoryFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return fd.DirectoryFileDescriptionDefaultImpl.PRead(ctx, dst, offset, opts) -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *GenericDirectoryFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return fd.DirectoryFileDescriptionDefaultImpl.Write(ctx, src, opts) -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *GenericDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return fd.DirectoryFileDescriptionDefaultImpl.PWrite(ctx, src, offset, opts) -} - -// Release implements vfs.FileDecriptionImpl.Release. -func (fd *GenericDirectoryFD) Release() {} - -func (fd *GenericDirectoryFD) filesystem() *vfs.Filesystem { - return fd.vfsfd.VirtualDentry().Mount().Filesystem() -} - -func (fd *GenericDirectoryFD) inode() Inode { - return fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode -} - -// IterDirents implements vfs.FileDecriptionImpl.IterDirents. IterDirents holds -// o.mu when calling cb. -func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - vfsFS := fd.filesystem() - fs := vfsFS.Impl().(*Filesystem) - vfsd := fd.vfsfd.VirtualDentry().Dentry() - - fs.mu.Lock() - defer fs.mu.Unlock() - - opts := vfs.StatOptions{Mask: linux.STATX_INO} - // Handle ".". - if fd.off == 0 { - stat, err := fd.inode().Stat(vfsFS, opts) - if err != nil { - return err - } - dirent := vfs.Dirent{ - Name: ".", - Type: linux.DT_DIR, - Ino: stat.Ino, - NextOff: 1, - } - if err := cb.Handle(dirent); err != nil { - return err - } - fd.off++ - } - - // Handle "..". - if fd.off == 1 { - parentInode := vfsd.ParentOrSelf().Impl().(*Dentry).inode - stat, err := parentInode.Stat(vfsFS, opts) - if err != nil { - return err - } - dirent := vfs.Dirent{ - Name: "..", - Type: linux.FileMode(stat.Mode).DirentType(), - Ino: stat.Ino, - NextOff: 2, - } - if err := cb.Handle(dirent); err != nil { - return err - } - fd.off++ - } - - // Handle static children. - fd.children.mu.RLock() - defer fd.children.mu.RUnlock() - // fd.off accounts for "." and "..", but fd.children do not track - // these. - childIdx := fd.off - 2 - for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() { - inode := it.Dentry.Impl().(*Dentry).inode - stat, err := inode.Stat(vfsFS, opts) - if err != nil { - return err - } - dirent := vfs.Dirent{ - Name: it.Name, - Type: linux.FileMode(stat.Mode).DirentType(), - Ino: stat.Ino, - NextOff: fd.off + 1, - } - if err := cb.Handle(dirent); err != nil { - return err - } - fd.off++ - } - - var err error - relOffset := fd.off - int64(len(fd.children.set)) - 2 - fd.off, err = fd.inode().IterDirents(ctx, cb, fd.off, relOffset) - return err -} - -// Seek implements vfs.FileDecriptionImpl.Seek. -func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fs := fd.filesystem().Impl().(*Filesystem) - fs.mu.Lock() - defer fs.mu.Unlock() - - switch whence { - case linux.SEEK_SET: - // Use offset as given. - case linux.SEEK_CUR: - offset += fd.off - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *GenericDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - fs := fd.filesystem() - inode := fd.inode() - return inode.Stat(fs, opts) -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *GenericDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - creds := auth.CredentialsFromContext(ctx) - inode := fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode - return inode.SetStat(ctx, fd.filesystem(), creds, opts) -} diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go deleted file mode 100644 index 31da8b511..000000000 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ /dev/null @@ -1,788 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernfs - -// This file implements vfs.FilesystemImpl for kernfs. - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// stepExistingLocked resolves rp.Component() in parent directory vfsd. -// -// stepExistingLocked is loosely analogous to fs/namei.c:walk_component(). -// -// Preconditions: Filesystem.mu must be locked for at least reading. !rp.Done(). -// -// Postcondition: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) (*vfs.Dentry, error) { - d := vfsd.Impl().(*Dentry) - if !d.isDir() { - return nil, syserror.ENOTDIR - } - // Directory searchable? - if err := d.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } -afterSymlink: - name := rp.Component() - // Revalidation must be skipped if name is "." or ".."; d or its parent - // respectively can't be expected to transition from invalidated back to - // valid, so detecting invalidation and retrying would loop forever. This - // is consistent with Linux: fs/namei.c:walk_component() => lookup_fast() - // calls d_revalidate(), but walk_component() => handle_dots() does not. - if name == "." { - rp.Advance() - return vfsd, nil - } - if name == ".." { - nextVFSD, err := rp.ResolveParent(vfsd) - if err != nil { - return nil, err - } - rp.Advance() - return nextVFSD, nil - } - d.dirMu.Lock() - nextVFSD, err := rp.ResolveChild(vfsd, name) - if err != nil { - d.dirMu.Unlock() - return nil, err - } - next, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, nextVFSD) - d.dirMu.Unlock() - if err != nil { - return nil, err - } - // Resolve any symlink at current path component. - if rp.ShouldFollowSymlink() && next.isSymlink() { - // TODO: VFS2 needs something extra for /proc/[pid]/fd/ "magic symlinks". - target, err := next.inode.Readlink(ctx) - if err != nil { - return nil, err - } - if err := rp.HandleSymlink(target); err != nil { - return nil, err - } - goto afterSymlink - - } - rp.Advance() - return &next.vfsd, nil -} - -// revalidateChildLocked must be called after a call to parent.vfsd.Child(name) -// or vfs.ResolvingPath.ResolveChild(name) returns childVFSD (which may be -// nil) to verify that the returned child (or lack thereof) is correct. -// -// Preconditions: Filesystem.mu must be locked for at least reading. -// parent.dirMu must be locked. parent.isDir(). name is not "." or "..". -// -// Postconditions: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *Dentry, name string, childVFSD *vfs.Dentry) (*Dentry, error) { - if childVFSD != nil { - // Cached dentry exists, revalidate. - child := childVFSD.Impl().(*Dentry) - if !child.inode.Valid(ctx) { - vfsObj.ForceDeleteDentry(childVFSD) - fs.deferDecRef(childVFSD) // Reference from Lookup. - childVFSD = nil - } - } - if childVFSD == nil { - // Dentry isn't cached; it either doesn't exist or failed - // revalidation. Attempt to resolve it via Lookup. - // - // FIXME(gvisor.dev/issue/1193): Inode.Lookup() should return - // *(kernfs.)Dentry, not *vfs.Dentry, since (kernfs.)Filesystem assumes - // that all dentries in the filesystem are (kernfs.)Dentry and performs - // vfs.DentryImpl casts accordingly. - var err error - childVFSD, err = parent.inode.Lookup(ctx, name) - if err != nil { - return nil, err - } - // Reference on childVFSD dropped by a corresponding Valid. - parent.insertChildLocked(name, childVFSD) - } - return childVFSD.Impl().(*Dentry), nil -} - -// walkExistingLocked resolves rp to an existing file. -// -// walkExistingLocked is loosely analogous to Linux's -// fs/namei.c:path_lookupat(). -// -// Preconditions: Filesystem.mu must be locked for at least reading. -// -// Postconditions: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { - vfsd := rp.Start() - for !rp.Done() { - var err error - vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd) - if err != nil { - return nil, nil, err - } - } - d := vfsd.Impl().(*Dentry) - if rp.MustBeDir() && !d.isDir() { - return nil, nil, syserror.ENOTDIR - } - return vfsd, d.inode, nil -} - -// walkParentDirLocked resolves all but the last path component of rp to an -// existing directory. It does not check that the returned directory is -// searchable by the provider of rp. -// -// walkParentDirLocked is loosely analogous to Linux's -// fs/namei.c:path_parentat(). -// -// Preconditions: Filesystem.mu must be locked for at least reading. !rp.Done(). -// -// Postconditions: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { - vfsd := rp.Start() - for !rp.Final() { - var err error - vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd) - if err != nil { - return nil, nil, err - } - } - d := vfsd.Impl().(*Dentry) - if !d.isDir() { - return nil, nil, syserror.ENOTDIR - } - return vfsd, d.inode, nil -} - -// checkCreateLocked checks that a file named rp.Component() may be created in -// directory parentVFSD, then returns rp.Component(). -// -// Preconditions: Filesystem.mu must be locked for at least reading. parentInode -// == parentVFSD.Impl().(*Dentry).Inode. isDir(parentInode) == true. -func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return "", err - } - pc := rp.Component() - if pc == "." || pc == ".." { - return "", syserror.EEXIST - } - childVFSD, err := rp.ResolveChild(parentVFSD, pc) - if err != nil { - return "", err - } - if childVFSD != nil { - return "", syserror.EEXIST - } - if parentVFSD.IsDisowned() { - return "", syserror.ENOENT - } - return pc, nil -} - -// checkDeleteLocked checks that the file represented by vfsd may be deleted. -// -// Preconditions: Filesystem.mu must be locked for at least reading. -func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { - parentVFSD := vfsd.Parent() - if parentVFSD == nil { - return syserror.EBUSY - } - if parentVFSD.IsDisowned() { - return syserror.ENOENT - } - if err := parentVFSD.Impl().(*Dentry).inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - return nil -} - -// Release implements vfs.FilesystemImpl.Release. -func (fs *Filesystem) Release() { -} - -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *Filesystem) Sync(ctx context.Context) error { - // All filesystem state is in-memory. - return nil -} - -// AccessAt implements vfs.Filesystem.Impl.AccessAt. -func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - defer fs.processDeferredDecRefs() - - _, inode, err := fs.walkExistingLocked(ctx, rp) - if err != nil { - return err - } - return inode.CheckPermissions(ctx, creds, ats) -} - -// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. -func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - fs.mu.RLock() - defer fs.processDeferredDecRefs() - defer fs.mu.RUnlock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) - if err != nil { - return nil, err - } - - if opts.CheckSearchable { - d := vfsd.Impl().(*Dentry) - if !d.isDir() { - return nil, syserror.ENOTDIR - } - if err := inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - } - vfsd.IncRef() // Ownership transferred to caller. - return vfsd, nil -} - -// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. -func (fs *Filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - fs.mu.RLock() - defer fs.processDeferredDecRefs() - defer fs.mu.RUnlock() - vfsd, _, err := fs.walkParentDirLocked(ctx, rp) - if err != nil { - return nil, err - } - vfsd.IncRef() // Ownership transferred to caller. - return vfsd, nil -} - -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - if rp.Done() { - return syserror.EEXIST - } - fs.mu.Lock() - defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() - if err != nil { - return err - } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) - if err != nil { - return err - } - if rp.Mount() != vd.Mount() { - return syserror.EXDEV - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - - d := vd.Dentry().Impl().(*Dentry) - if d.isDir() { - return syserror.EPERM - } - - child, err := parentInode.NewLink(ctx, pc, d.inode) - if err != nil { - return err - } - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) - return nil -} - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - if rp.Done() { - return syserror.EEXIST - } - fs.mu.Lock() - defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() - if err != nil { - return err - } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) - if err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - child, err := parentInode.NewDir(ctx, pc, opts) - if err != nil { - return err - } - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) - return nil -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - if rp.Done() { - return syserror.EEXIST - } - fs.mu.Lock() - defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() - if err != nil { - return err - } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) - if err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - new, err := parentInode.NewNode(ctx, pc, opts) - if err != nil { - return err - } - parentVFSD.Impl().(*Dentry).InsertChild(pc, new) - return nil -} - -// OpenAt implements vfs.FilesystemImpl.OpenAt. -func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - // Filter out flags that are not supported by kernfs. O_DIRECTORY and - // O_NOFOLLOW have no effect here (they're handled by VFS by setting - // appropriate bits in rp), but are returned by - // FileDescriptionImpl.StatusFlags(). - opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW - ats := vfs.AccessTypesForOpenFlags(&opts) - - // Do not create new file. - if opts.Flags&linux.O_CREAT == 0 { - fs.mu.RLock() - defer fs.processDeferredDecRefs() - defer fs.mu.RUnlock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) - if err != nil { - return nil, err - } - if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { - return nil, err - } - return inode.Open(rp, vfsd, opts) - } - - // May create new file. - mustCreate := opts.Flags&linux.O_EXCL != 0 - vfsd := rp.Start() - inode := vfsd.Impl().(*Dentry).inode - fs.mu.Lock() - defer fs.mu.Unlock() - if rp.Done() { - if rp.MustBeDir() { - return nil, syserror.EISDIR - } - if mustCreate { - return nil, syserror.EEXIST - } - if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { - return nil, err - } - return inode.Open(rp, vfsd, opts) - } -afterTrailingSymlink: - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() - if err != nil { - return nil, err - } - // Check for search permission in the parent directory. - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - // Reject attempts to open directories with O_CREAT. - if rp.MustBeDir() { - return nil, syserror.EISDIR - } - pc := rp.Component() - if pc == "." || pc == ".." { - return nil, syserror.EISDIR - } - // Determine whether or not we need to create a file. - childVFSD, err := rp.ResolveChild(parentVFSD, pc) - if err != nil { - return nil, err - } - if childVFSD == nil { - // Already checked for searchability above; now check for writability. - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { - return nil, err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return nil, err - } - defer rp.Mount().EndWrite() - // Create and open the child. - child, err := parentInode.NewFile(ctx, pc, opts) - if err != nil { - return nil, err - } - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) - return child.Impl().(*Dentry).inode.Open(rp, child, opts) - } - // Open existing file or follow symlink. - if mustCreate { - return nil, syserror.EEXIST - } - childDentry := childVFSD.Impl().(*Dentry) - childInode := childDentry.inode - if rp.ShouldFollowSymlink() { - if childDentry.isSymlink() { - target, err := childInode.Readlink(ctx) - if err != nil { - return nil, err - } - if err := rp.HandleSymlink(target); err != nil { - return nil, err - } - // rp.Final() may no longer be true since we now need to resolve the - // symlink target. - goto afterTrailingSymlink - } - } - if err := childInode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { - return nil, err - } - return childInode.Open(rp, childVFSD, opts) -} - -// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. -func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - fs.mu.RLock() - d, inode, err := fs.walkExistingLocked(ctx, rp) - fs.mu.RUnlock() - fs.processDeferredDecRefs() - if err != nil { - return "", err - } - if !d.Impl().(*Dentry).isSymlink() { - return "", syserror.EINVAL - } - return inode.Readlink(ctx) -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { - // Only RENAME_NOREPLACE is supported. - if opts.Flags&^linux.RENAME_NOREPLACE != 0 { - return syserror.EINVAL - } - noReplace := opts.Flags&linux.RENAME_NOREPLACE != 0 - - fs.mu.Lock() - defer fs.mu.Lock() - - // Resolve the destination directory first to verify that it's on this - // Mount. - dstDirVFSD, dstDirInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() - if err != nil { - return err - } - mnt := rp.Mount() - if mnt != oldParentVD.Mount() { - return syserror.EXDEV - } - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - - srcDirVFSD := oldParentVD.Dentry() - srcDir := srcDirVFSD.Impl().(*Dentry) - srcDir.dirMu.Lock() - src, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), srcDir, oldName, srcDirVFSD.Child(oldName)) - srcDir.dirMu.Unlock() - fs.processDeferredDecRefsLocked() - if err != nil { - return err - } - srcVFSD := &src.vfsd - - // Can we remove the src dentry? - if err := checkDeleteLocked(ctx, rp, srcVFSD); err != nil { - return err - } - - // Can we create the dst dentry? - var dstVFSD *vfs.Dentry - pc, err := checkCreateLocked(ctx, rp, dstDirVFSD, dstDirInode) - switch err { - case nil: - // Ok, continue with rename as replacement. - case syserror.EEXIST: - if noReplace { - // Won't overwrite existing node since RENAME_NOREPLACE was requested. - return syserror.EEXIST - } - dstVFSD, err = rp.ResolveChild(dstDirVFSD, pc) - if err != nil { - panic(fmt.Sprintf("Child %q for parent Dentry %+v disappeared inside atomic section?", pc, dstDirVFSD)) - } - default: - return err - } - - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - virtfs := rp.VirtualFilesystem() - - srcDirDentry := srcDirVFSD.Impl().(*Dentry) - dstDirDentry := dstDirVFSD.Impl().(*Dentry) - - // We can't deadlock here due to lock ordering because we're protected from - // concurrent renames by fs.mu held for writing. - srcDirDentry.dirMu.Lock() - defer srcDirDentry.dirMu.Unlock() - dstDirDentry.dirMu.Lock() - defer dstDirDentry.dirMu.Unlock() - - if err := virtfs.PrepareRenameDentry(mntns, srcVFSD, dstVFSD); err != nil { - return err - } - srcDirInode := srcDirDentry.inode - replaced, err := srcDirInode.Rename(ctx, srcVFSD.Name(), pc, srcVFSD, dstDirVFSD) - if err != nil { - virtfs.AbortRenameDentry(srcVFSD, dstVFSD) - return err - } - virtfs.CommitRenameReplaceDentry(srcVFSD, dstDirVFSD, pc, replaced) - return nil -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - fs.mu.Lock() - defer fs.mu.Unlock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) - fs.processDeferredDecRefsLocked() - if err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { - return err - } - if !vfsd.Impl().(*Dentry).isDir() { - return syserror.ENOTDIR - } - if inode.HasChildren() { - return syserror.ENOTEMPTY - } - virtfs := rp.VirtualFilesystem() - parentDentry := vfsd.Parent().Impl().(*Dentry) - parentDentry.dirMu.Lock() - defer parentDentry.dirMu.Unlock() - - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { - return err - } - if err := parentDentry.inode.RmDir(ctx, rp.Component(), vfsd); err != nil { - virtfs.AbortDeleteDentry(vfsd) - return err - } - virtfs.CommitDeleteDentry(vfsd) - return nil -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) - fs.mu.RUnlock() - fs.processDeferredDecRefs() - if err != nil { - return err - } - if opts.Stat.Mask == 0 { - return nil - } - return inode.SetStat(ctx, fs.VFSFilesystem(), rp.Credentials(), opts) -} - -// StatAt implements vfs.FilesystemImpl.StatAt. -func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) - fs.mu.RUnlock() - fs.processDeferredDecRefs() - if err != nil { - return linux.Statx{}, err - } - return inode.Stat(fs.VFSFilesystem(), opts) -} - -// StatFSAt implements vfs.FilesystemImpl.StatFSAt. -func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) - fs.mu.RUnlock() - fs.processDeferredDecRefs() - if err != nil { - return linux.Statfs{}, err - } - // TODO: actually implement statfs - return linux.Statfs{}, syserror.ENOSYS -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - if rp.Done() { - return syserror.EEXIST - } - fs.mu.Lock() - defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() - if err != nil { - return err - } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) - if err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - child, err := parentInode.NewSymlink(ctx, pc, target) - if err != nil { - return err - } - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) - return nil -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - fs.mu.Lock() - defer fs.mu.Unlock() - vfsd, _, err := fs.walkExistingLocked(ctx, rp) - fs.processDeferredDecRefsLocked() - if err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { - return err - } - if vfsd.Impl().(*Dentry).isDir() { - return syserror.EISDIR - } - virtfs := rp.VirtualFilesystem() - parentDentry := vfsd.Parent().Impl().(*Dentry) - parentDentry.dirMu.Lock() - defer parentDentry.dirMu.Unlock() - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { - return err - } - if err := parentDentry.inode.Unlink(ctx, rp.Component(), vfsd); err != nil { - virtfs.AbortDeleteDentry(vfsd) - return err - } - virtfs.CommitDeleteDentry(vfsd) - return nil -} - -// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { - fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) - fs.mu.RUnlock() - fs.processDeferredDecRefs() - if err != nil { - return nil, err - } - // kernfs currently does not support extended attributes. - return nil, syserror.ENOTSUP -} - -// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *Filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { - fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) - fs.mu.RUnlock() - fs.processDeferredDecRefs() - if err != nil { - return "", err - } - // kernfs currently does not support extended attributes. - return "", syserror.ENOTSUP -} - -// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. -func (fs *Filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) - fs.mu.RUnlock() - fs.processDeferredDecRefs() - if err != nil { - return err - } - // kernfs currently does not support extended attributes. - return syserror.ENOTSUP -} - -// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. -func (fs *Filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) - fs.mu.RUnlock() - fs.processDeferredDecRefs() - if err != nil { - return err - } - // kernfs currently does not support extended attributes. - return syserror.ENOTSUP -} - -// PrependPath implements vfs.FilesystemImpl.PrependPath. -func (fs *Filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) -} diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go deleted file mode 100644 index 5c84b10c9..000000000 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ /dev/null @@ -1,577 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernfs - -import ( - "fmt" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" -) - -// InodeNoopRefCount partially implements the Inode interface, specifically the -// inodeRefs sub interface. InodeNoopRefCount implements a simple reference -// count for inodes, performing no extra actions when references are obtained or -// released. This is suitable for simple file inodes that don't reference any -// resources. -type InodeNoopRefCount struct { -} - -// IncRef implements Inode.IncRef. -func (InodeNoopRefCount) IncRef() { -} - -// DecRef implements Inode.DecRef. -func (InodeNoopRefCount) DecRef() { -} - -// TryIncRef implements Inode.TryIncRef. -func (InodeNoopRefCount) TryIncRef() bool { - return true -} - -// Destroy implements Inode.Destroy. -func (InodeNoopRefCount) Destroy() { -} - -// InodeDirectoryNoNewChildren partially implements the Inode interface. -// InodeDirectoryNoNewChildren represents a directory inode which does not -// support creation of new children. -type InodeDirectoryNoNewChildren struct{} - -// NewFile implements Inode.NewFile. -func (InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -// NewDir implements Inode.NewDir. -func (InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -// NewLink implements Inode.NewLink. -func (InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -// NewSymlink implements Inode.NewSymlink. -func (InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -// NewNode implements Inode.NewNode. -func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -// InodeNotDirectory partially implements the Inode interface, specifically the -// inodeDirectory and inodeDynamicDirectory sub interfaces. Inodes that do not -// represent directories can embed this to provide no-op implementations for -// directory-related functions. -type InodeNotDirectory struct { -} - -// HasChildren implements Inode.HasChildren. -func (InodeNotDirectory) HasChildren() bool { - return false -} - -// NewFile implements Inode.NewFile. -func (InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { - panic("NewFile called on non-directory inode") -} - -// NewDir implements Inode.NewDir. -func (InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { - panic("NewDir called on non-directory inode") -} - -// NewLink implements Inode.NewLinkink. -func (InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { - panic("NewLink called on non-directory inode") -} - -// NewSymlink implements Inode.NewSymlink. -func (InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { - panic("NewSymlink called on non-directory inode") -} - -// NewNode implements Inode.NewNode. -func (InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { - panic("NewNode called on non-directory inode") -} - -// Unlink implements Inode.Unlink. -func (InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error { - panic("Unlink called on non-directory inode") -} - -// RmDir implements Inode.RmDir. -func (InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error { - panic("RmDir called on non-directory inode") -} - -// Rename implements Inode.Rename. -func (InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) { - panic("Rename called on non-directory inode") -} - -// Lookup implements Inode.Lookup. -func (InodeNotDirectory) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - panic("Lookup called on non-directory inode") -} - -// IterDirents implements Inode.IterDirents. -func (InodeNotDirectory) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) { - panic("IterDirents called on non-directory inode") -} - -// Valid implements Inode.Valid. -func (InodeNotDirectory) Valid(context.Context) bool { - return true -} - -// InodeNoDynamicLookup partially implements the Inode interface, specifically -// the inodeDynamicLookup sub interface. Directory inodes that do not support -// dymanic entries (i.e. entries that are not "hashed" into the -// vfs.Dentry.children) can embed this to provide no-op implementations for -// functions related to dynamic entries. -type InodeNoDynamicLookup struct{} - -// Lookup implements Inode.Lookup. -func (InodeNoDynamicLookup) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - return nil, syserror.ENOENT -} - -// IterDirents implements Inode.IterDirents. -func (InodeNoDynamicLookup) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { - return offset, nil -} - -// Valid implements Inode.Valid. -func (InodeNoDynamicLookup) Valid(ctx context.Context) bool { - return true -} - -// InodeNotSymlink partially implements the Inode interface, specifically the -// inodeSymlink sub interface. All inodes that are not symlinks may embed this -// to return the appropriate errors from symlink-related functions. -type InodeNotSymlink struct{} - -// Readlink implements Inode.Readlink. -func (InodeNotSymlink) Readlink(context.Context) (string, error) { - return "", syserror.EINVAL -} - -// InodeAttrs partially implements the Inode interface, specifically the -// inodeMetadata sub interface. InodeAttrs provides functionality related to -// inode attributes. -// -// Must be initialized by Init prior to first use. -type InodeAttrs struct { - ino uint64 - mode uint32 - uid uint32 - gid uint32 - nlink uint32 -} - -// Init initializes this InodeAttrs. -func (a *InodeAttrs) Init(creds *auth.Credentials, ino uint64, mode linux.FileMode) { - if mode.FileType() == 0 { - panic(fmt.Sprintf("No file type specified in 'mode' for InodeAttrs.Init(): mode=0%o", mode)) - } - - nlink := uint32(1) - if mode.FileType() == linux.ModeDirectory { - nlink = 2 - } - atomic.StoreUint64(&a.ino, ino) - atomic.StoreUint32(&a.mode, uint32(mode)) - atomic.StoreUint32(&a.uid, uint32(creds.EffectiveKUID)) - atomic.StoreUint32(&a.gid, uint32(creds.EffectiveKGID)) - atomic.StoreUint32(&a.nlink, nlink) -} - -// Mode implements Inode.Mode. -func (a *InodeAttrs) Mode() linux.FileMode { - return linux.FileMode(atomic.LoadUint32(&a.mode)) -} - -// Stat partially implements Inode.Stat. Note that this function doesn't provide -// all the stat fields, and the embedder should consider extending the result -// with filesystem-specific fields. -func (a *InodeAttrs) Stat(*vfs.Filesystem, vfs.StatOptions) (linux.Statx, error) { - var stat linux.Statx - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_NLINK - stat.Ino = atomic.LoadUint64(&a.ino) - stat.Mode = uint16(a.Mode()) - stat.UID = atomic.LoadUint32(&a.uid) - stat.GID = atomic.LoadUint32(&a.gid) - stat.Nlink = atomic.LoadUint32(&a.nlink) - - // TODO(gvisor.dev/issue/1193): Implement other stat fields like timestamps. - - return stat, nil -} - -// SetStat implements Inode.SetStat. -func (a *InodeAttrs) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { - if opts.Stat.Mask == 0 { - return nil - } - if opts.Stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID) != 0 { - return syserror.EPERM - } - if err := vfs.CheckSetStat(ctx, creds, &opts.Stat, a.Mode(), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { - return err - } - - stat := opts.Stat - if stat.Mask&linux.STATX_MODE != 0 { - for { - old := atomic.LoadUint32(&a.mode) - new := old | uint32(stat.Mode & ^uint16(linux.S_IFMT)) - if swapped := atomic.CompareAndSwapUint32(&a.mode, old, new); swapped { - break - } - } - } - - if stat.Mask&linux.STATX_UID != 0 { - atomic.StoreUint32(&a.uid, stat.UID) - } - if stat.Mask&linux.STATX_GID != 0 { - atomic.StoreUint32(&a.gid, stat.GID) - } - - // Note that not all fields are modifiable. For example, the file type and - // inode numbers are immutable after node creation. - - // TODO(gvisor.dev/issue/1193): Implement other stat fields like timestamps. - - return nil -} - -// CheckPermissions implements Inode.CheckPermissions. -func (a *InodeAttrs) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - return vfs.GenericCheckPermissions( - creds, - ats, - a.Mode(), - auth.KUID(atomic.LoadUint32(&a.uid)), - auth.KGID(atomic.LoadUint32(&a.gid)), - ) -} - -// IncLinks implements Inode.IncLinks. -func (a *InodeAttrs) IncLinks(n uint32) { - if atomic.AddUint32(&a.nlink, n) <= n { - panic("InodeLink.IncLinks called with no existing links") - } -} - -// DecLinks implements Inode.DecLinks. -func (a *InodeAttrs) DecLinks() { - if nlink := atomic.AddUint32(&a.nlink, ^uint32(0)); nlink == ^uint32(0) { - // Negative overflow - panic("Inode.DecLinks called at 0 links") - } -} - -type slot struct { - Name string - Dentry *vfs.Dentry - slotEntry -} - -// OrderedChildrenOptions contains initialization options for OrderedChildren. -type OrderedChildrenOptions struct { - // Writable indicates whether vfs.FilesystemImpl methods implemented by - // OrderedChildren may modify the tracked children. This applies to - // operations related to rename, unlink and rmdir. If an OrderedChildren is - // not writable, these operations all fail with EPERM. - Writable bool -} - -// OrderedChildren partially implements the Inode interface. OrderedChildren can -// be embedded in directory inodes to keep track of the children in the -// directory, and can then be used to implement a generic directory FD -- see -// GenericDirectoryFD. OrderedChildren is not compatible with dynamic -// directories. -// -// Must be initialize with Init before first use. -type OrderedChildren struct { - refs.AtomicRefCount - - // Can children be modified by user syscalls? It set to false, interface - // methods that would modify the children return EPERM. Immutable. - writable bool - - mu sync.RWMutex - order slotList - set map[string]*slot -} - -// Init initializes an OrderedChildren. -func (o *OrderedChildren) Init(opts OrderedChildrenOptions) { - o.writable = opts.Writable - o.set = make(map[string]*slot) -} - -// DecRef implements Inode.DecRef. -func (o *OrderedChildren) DecRef() { - o.AtomicRefCount.DecRefWithDestructor(o.Destroy) -} - -// Destroy cleans up resources referenced by this OrderedChildren. -func (o *OrderedChildren) Destroy() { - o.mu.Lock() - defer o.mu.Unlock() - o.order.Reset() - o.set = nil -} - -// Populate inserts children into this OrderedChildren, and d's dentry -// cache. Populate returns the number of directories inserted, which the caller -// may use to update the link count for the parent directory. -// -// Precondition: d.Impl() must be a kernfs Dentry. d must represent a directory -// inode. children must not contain any conflicting entries already in o. -func (o *OrderedChildren) Populate(d *Dentry, children map[string]*Dentry) uint32 { - var links uint32 - for name, child := range children { - if child.isDir() { - links++ - } - if err := o.Insert(name, child.VFSDentry()); err != nil { - panic(fmt.Sprintf("Collision when attempting to insert child %q (%+v) into %+v", name, child, d)) - } - d.InsertChild(name, child.VFSDentry()) - } - return links -} - -// HasChildren implements Inode.HasChildren. -func (o *OrderedChildren) HasChildren() bool { - o.mu.RLock() - defer o.mu.RUnlock() - return len(o.set) > 0 -} - -// Insert inserts child into o. This ignores the writability of o, as this is -// not part of the vfs.FilesystemImpl interface, and is a lower-level operation. -func (o *OrderedChildren) Insert(name string, child *vfs.Dentry) error { - o.mu.Lock() - defer o.mu.Unlock() - if _, ok := o.set[name]; ok { - return syserror.EEXIST - } - s := &slot{ - Name: name, - Dentry: child, - } - o.order.PushBack(s) - o.set[name] = s - return nil -} - -// Precondition: caller must hold o.mu for writing. -func (o *OrderedChildren) removeLocked(name string) { - if s, ok := o.set[name]; ok { - delete(o.set, name) - o.order.Remove(s) - } -} - -// Precondition: caller must hold o.mu for writing. -func (o *OrderedChildren) replaceChildLocked(name string, new *vfs.Dentry) *vfs.Dentry { - if s, ok := o.set[name]; ok { - // Existing slot with given name, simply replace the dentry. - var old *vfs.Dentry - old, s.Dentry = s.Dentry, new - return old - } - - // No existing slot with given name, create and hash new slot. - s := &slot{ - Name: name, - Dentry: new, - } - o.order.PushBack(s) - o.set[name] = s - return nil -} - -// Precondition: caller must hold o.mu for reading or writing. -func (o *OrderedChildren) checkExistingLocked(name string, child *vfs.Dentry) error { - s, ok := o.set[name] - if !ok { - return syserror.ENOENT - } - if s.Dentry != child { - panic(fmt.Sprintf("Dentry hashed into inode doesn't match what vfs thinks! OrderedChild: %+v, vfs: %+v", s.Dentry, child)) - } - return nil -} - -// Unlink implements Inode.Unlink. -func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *vfs.Dentry) error { - if !o.writable { - return syserror.EPERM - } - o.mu.Lock() - defer o.mu.Unlock() - if err := o.checkExistingLocked(name, child); err != nil { - return err - } - o.removeLocked(name) - return nil -} - -// Rmdir implements Inode.Rmdir. -func (o *OrderedChildren) RmDir(ctx context.Context, name string, child *vfs.Dentry) error { - // We're not responsible for checking that child is a directory, that it's - // empty, or updating any link counts; so this is the same as unlink. - return o.Unlink(ctx, name, child) -} - -type renameAcrossDifferentImplementationsError struct{} - -func (renameAcrossDifferentImplementationsError) Error() string { - return "rename across inodes with different implementations" -} - -// Rename implements Inode.Rename. -// -// Precondition: Rename may only be called across two directory inodes with -// identical implementations of Rename. Practically, this means filesystems that -// implement Rename by embedding OrderedChildren for any directory -// implementation must use OrderedChildren for all directory implementations -// that will support Rename. -// -// Postcondition: reference on any replaced dentry transferred to caller. -func (o *OrderedChildren) Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (*vfs.Dentry, error) { - dst, ok := dstDir.Impl().(*Dentry).inode.(interface{}).(*OrderedChildren) - if !ok { - return nil, renameAcrossDifferentImplementationsError{} - } - if !o.writable || !dst.writable { - return nil, syserror.EPERM - } - // Note: There's a potential deadlock below if concurrent calls to Rename - // refer to the same src and dst directories in reverse. We avoid any - // ordering issues because the caller is required to serialize concurrent - // calls to Rename in accordance with the interface declaration. - o.mu.Lock() - defer o.mu.Unlock() - if dst != o { - dst.mu.Lock() - defer dst.mu.Unlock() - } - if err := o.checkExistingLocked(oldname, child); err != nil { - return nil, err - } - replaced := dst.replaceChildLocked(newname, child) - return replaced, nil -} - -// nthLocked returns an iterator to the nth child tracked by this object. The -// iterator is valid until the caller releases o.mu. Returns nil if the -// requested index falls out of bounds. -// -// Preconditon: Caller must hold o.mu for reading. -func (o *OrderedChildren) nthLocked(i int64) *slot { - for it := o.order.Front(); it != nil && i >= 0; it = it.Next() { - if i == 0 { - return it - } - i-- - } - return nil -} - -// InodeSymlink partially implements Inode interface for symlinks. -type InodeSymlink struct { - InodeNotDirectory -} - -// Open implements Inode.Open. -func (InodeSymlink) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - return nil, syserror.ELOOP -} - -// StaticDirectory is a standard implementation of a directory with static -// contents. -// -// +stateify savable -type StaticDirectory struct { - InodeNotSymlink - InodeDirectoryNoNewChildren - InodeAttrs - InodeNoDynamicLookup - OrderedChildren -} - -var _ Inode = (*StaticDirectory)(nil) - -// NewStaticDir creates a new static directory and returns its dentry. -func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, children map[string]*Dentry) *Dentry { - inode := &StaticDirectory{} - inode.Init(creds, ino, perm) - - dentry := &Dentry{} - dentry.Init(inode) - - inode.OrderedChildren.Init(OrderedChildrenOptions{}) - links := inode.OrderedChildren.Populate(dentry, children) - inode.IncLinks(links) - - return dentry -} - -// Init initializes StaticDirectory. -func (s *StaticDirectory) Init(creds *auth.Credentials, ino uint64, perm linux.FileMode) { - if perm&^linux.PermissionsMask != 0 { - panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) - } - s.InodeAttrs.Init(creds, ino, linux.ModeDirectory|perm) -} - -// Open implements kernfs.Inode. -func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &s.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*StaticDirectory) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -// AlwaysValid partially implements kernfs.inodeDynamicLookup. -type AlwaysValid struct{} - -// Valid implements kernfs.inodeDynamicLookup. -func (*AlwaysValid) Valid(context.Context) bool { - return true -} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go deleted file mode 100644 index 794e38908..000000000 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ /dev/null @@ -1,423 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package kernfs provides the tools to implement inode-based filesystems. -// Kernfs has two main features: -// -// 1. The Inode interface, which maps VFS2's path-based filesystem operations to -// specific filesystem nodes. Kernfs uses the Inode interface to provide a -// blanket implementation for the vfs.FilesystemImpl. Kernfs also serves as -// the synchronization mechanism for all filesystem operations by holding a -// filesystem-wide lock across all operations. -// -// 2. Various utility types which provide generic implementations for various -// parts of the Inode and vfs.FileDescription interfaces. Client filesystems -// based on kernfs can embed the appropriate set of these to avoid having to -// reimplement common filesystem operations. See inode_impl_util.go and -// fd_impl_util.go. -// -// Reference Model: -// -// Kernfs dentries represents named pointers to inodes. Dentries and inode have -// independent lifetimes and reference counts. A child dentry unconditionally -// holds a reference on its parent directory's dentry. A dentry also holds a -// reference on the inode it points to. Multiple dentries can point to the same -// inode (for example, in the case of hardlinks). File descriptors hold a -// reference to the dentry they're opened on. -// -// Dentries are guaranteed to exist while holding Filesystem.mu for -// reading. Dropping dentries require holding Filesystem.mu for writing. To -// queue dentries for destruction from a read critical section, see -// Filesystem.deferDecRef. -// -// Lock ordering: -// -// kernfs.Filesystem.mu -// kernfs.Dentry.dirMu -// vfs.VirtualFilesystem.mountMu -// vfs.Dentry.mu -// kernfs.Filesystem.droppedDentriesMu -// (inode implementation locks, if any) -package kernfs - -import ( - "fmt" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" -) - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// Filesystem mostly implements vfs.FilesystemImpl for a generic in-memory -// filesystem. Concrete implementations are expected to embed this in their own -// Filesystem type. -type Filesystem struct { - vfsfs vfs.Filesystem - - droppedDentriesMu sync.Mutex - - // droppedDentries is a list of dentries waiting to be DecRef()ed. This is - // used to defer dentry destruction until mu can be acquired for - // writing. Protected by droppedDentriesMu. - droppedDentries []*vfs.Dentry - - // mu synchronizes the lifetime of Dentries on this filesystem. Holding it - // for reading guarantees continued existence of any resolved dentries, but - // the dentry tree may be modified. - // - // Kernfs dentries can only be DecRef()ed while holding mu for writing. For - // example: - // - // fs.mu.Lock() - // defer fs.mu.Unlock() - // ... - // dentry1.DecRef() - // defer dentry2.DecRef() // Ok, will run before Unlock. - // - // If discarding dentries in a read context, use Filesystem.deferDecRef. For - // example: - // - // fs.mu.RLock() - // fs.mu.processDeferredDecRefs() - // defer fs.mu.RUnlock() - // ... - // fs.deferDecRef(dentry) - mu sync.RWMutex - - // nextInoMinusOne is used to to allocate inode numbers on this - // filesystem. Must be accessed by atomic operations. - nextInoMinusOne uint64 -} - -// deferDecRef defers dropping a dentry ref until the next call to -// processDeferredDecRefs{,Locked}. See comment on Filesystem.mu. -// -// Precondition: d must not already be pending destruction. -func (fs *Filesystem) deferDecRef(d *vfs.Dentry) { - fs.droppedDentriesMu.Lock() - fs.droppedDentries = append(fs.droppedDentries, d) - fs.droppedDentriesMu.Unlock() -} - -// processDeferredDecRefs calls vfs.Dentry.DecRef on all dentries in the -// droppedDentries list. See comment on Filesystem.mu. -func (fs *Filesystem) processDeferredDecRefs() { - fs.mu.Lock() - fs.processDeferredDecRefsLocked() - fs.mu.Unlock() -} - -// Precondition: fs.mu must be held for writing. -func (fs *Filesystem) processDeferredDecRefsLocked() { - fs.droppedDentriesMu.Lock() - for _, d := range fs.droppedDentries { - d.DecRef() - } - fs.droppedDentries = fs.droppedDentries[:0] // Keep slice memory for reuse. - fs.droppedDentriesMu.Unlock() -} - -// Init initializes a kernfs filesystem. This should be called from during -// vfs.FilesystemType.NewFilesystem for the concrete filesystem embedding -// kernfs. -func (fs *Filesystem) Init(vfsObj *vfs.VirtualFilesystem) { - fs.vfsfs.Init(vfsObj, fs) -} - -// VFSFilesystem returns the generic vfs filesystem object. -func (fs *Filesystem) VFSFilesystem() *vfs.Filesystem { - return &fs.vfsfs -} - -// NextIno allocates a new inode number on this filesystem. -func (fs *Filesystem) NextIno() uint64 { - return atomic.AddUint64(&fs.nextInoMinusOne, 1) -} - -// These consts are used in the Dentry.flags field. -const ( - // Dentry points to a directory inode. - dflagsIsDir = 1 << iota - - // Dentry points to a symlink inode. - dflagsIsSymlink -) - -// Dentry implements vfs.DentryImpl. -// -// A kernfs dentry is similar to a dentry in a traditional filesystem: it's a -// named reference to an inode. A dentry generally lives as long as it's part of -// a mounted filesystem tree. Kernfs doesn't cache dentries once all references -// to them are removed. Dentries hold a single reference to the inode they point -// to, and child dentries hold a reference on their parent. -// -// Must be initialized by Init prior to first use. -type Dentry struct { - refs.AtomicRefCount - - vfsd vfs.Dentry - inode Inode - - // flags caches useful information about the dentry from the inode. See the - // dflags* consts above. Must be accessed by atomic ops. - flags uint32 - - // dirMu protects vfsd.children for directory dentries. - dirMu sync.Mutex -} - -// Init initializes this dentry. -// -// Precondition: Caller must hold a reference on inode. -// -// Postcondition: Caller's reference on inode is transferred to the dentry. -func (d *Dentry) Init(inode Inode) { - d.vfsd.Init(d) - d.inode = inode - ftype := inode.Mode().FileType() - if ftype == linux.ModeDirectory { - d.flags |= dflagsIsDir - } - if ftype == linux.ModeSymlink { - d.flags |= dflagsIsSymlink - } -} - -// VFSDentry returns the generic vfs dentry for this kernfs dentry. -func (d *Dentry) VFSDentry() *vfs.Dentry { - return &d.vfsd -} - -// isDir checks whether the dentry points to a directory inode. -func (d *Dentry) isDir() bool { - return atomic.LoadUint32(&d.flags)&dflagsIsDir != 0 -} - -// isSymlink checks whether the dentry points to a symlink inode. -func (d *Dentry) isSymlink() bool { - return atomic.LoadUint32(&d.flags)&dflagsIsSymlink != 0 -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *Dentry) DecRef() { - d.AtomicRefCount.DecRefWithDestructor(d.destroy) -} - -// Precondition: Dentry must be removed from VFS' dentry cache. -func (d *Dentry) destroy() { - d.inode.DecRef() // IncRef from Init. - d.inode = nil - if parent := d.vfsd.Parent(); parent != nil { - parent.DecRef() // IncRef from Dentry.InsertChild. - } -} - -// InsertChild inserts child into the vfs dentry cache with the given name under -// this dentry. This does not update the directory inode, so calling this on -// it's own isn't sufficient to insert a child into a directory. InsertChild -// updates the link count on d if required. -// -// Precondition: d must represent a directory inode. -func (d *Dentry) InsertChild(name string, child *vfs.Dentry) { - d.dirMu.Lock() - d.insertChildLocked(name, child) - d.dirMu.Unlock() -} - -// insertChildLocked is equivalent to InsertChild, with additional -// preconditions. -// -// Precondition: d.dirMu must be locked. -func (d *Dentry) insertChildLocked(name string, child *vfs.Dentry) { - if !d.isDir() { - panic(fmt.Sprintf("InsertChild called on non-directory Dentry: %+v.", d)) - } - vfsDentry := d.VFSDentry() - vfsDentry.IncRef() // DecRef in child's Dentry.destroy. - vfsDentry.InsertChild(child, name) -} - -// The Inode interface maps filesystem-level operations that operate on paths to -// equivalent operations on specific filesystem nodes. -// -// The interface methods are groups into logical categories as sub interfaces -// below. Generally, an implementation for each sub interface can be provided by -// embedding an appropriate type from inode_impl_utils.go. The sub interfaces -// are purely organizational. Methods declared directly in the main interface -// have no generic implementations, and should be explicitly provided by the -// client filesystem. -// -// Generally, implementations are not responsible for tasks that are common to -// all filesystems. These include: -// -// - Checking that dentries passed to methods are of the appropriate file type. -// - Checking permissions. -// - Updating link and reference counts. -// -// Specific responsibilities of implementations are documented below. -type Inode interface { - // Methods related to reference counting. A generic implementation is - // provided by InodeNoopRefCount. These methods are generally called by the - // equivalent Dentry methods. - inodeRefs - - // Methods related to node metadata. A generic implementation is provided by - // InodeAttrs. - inodeMetadata - - // Method for inodes that represent symlink. InodeNotSymlink provides a - // blanket implementation for all non-symlink inodes. - inodeSymlink - - // Method for inodes that represent directories. InodeNotDirectory provides - // a blanket implementation for all non-directory inodes. - inodeDirectory - - // Method for inodes that represent dynamic directories and their - // children. InodeNoDynamicLookup provides a blanket implementation for all - // non-dynamic-directory inodes. - inodeDynamicLookup - - // Open creates a file description for the filesystem object represented by - // this inode. The returned file description should hold a reference on the - // inode for its lifetime. - // - // Precondition: rp.Done(). vfsd.Impl() must be the kernfs Dentry containing - // the inode on which Open() is being called. - Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) -} - -type inodeRefs interface { - IncRef() - DecRef() - TryIncRef() bool - // Destroy is called when the inode reaches zero references. Destroy release - // all resources (references) on objects referenced by the inode, including - // any child dentries. - Destroy() -} - -type inodeMetadata interface { - // CheckPermissions checks that creds may access this inode for the - // requested access type, per the the rules of - // fs/namei.c:generic_permission(). - CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error - - // Mode returns the (struct stat)::st_mode value for this inode. This is - // separated from Stat for performance. - Mode() linux.FileMode - - // Stat returns the metadata for this inode. This corresponds to - // vfs.FilesystemImpl.StatAt. - Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) - - // SetStat updates the metadata for this inode. This corresponds to - // vfs.FilesystemImpl.SetStatAt. Implementations are responsible for checking - // if the operation can be performed (see vfs.CheckSetStat() for common - // checks). - SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error -} - -// Precondition: All methods in this interface may only be called on directory -// inodes. -type inodeDirectory interface { - // The New{File,Dir,Node,Symlink} methods below should return a new inode - // hashed into this inode. - // - // These inode constructors are inode-level operations rather than - // filesystem-level operations to allow client filesystems to mix different - // implementations based on the new node's location in the - // filesystem. - - // HasChildren returns true if the directory inode has any children. - HasChildren() bool - - // NewFile creates a new regular file inode. - NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) - - // NewDir creates a new directory inode. - NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) - - // NewLink creates a new hardlink to a specified inode in this - // directory. Implementations should create a new kernfs Dentry pointing to - // target, and update target's link count. - NewLink(ctx context.Context, name string, target Inode) (*vfs.Dentry, error) - - // NewSymlink creates a new symbolic link inode. - NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) - - // NewNode creates a new filesystem node for a mknod syscall. - NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) - - // Unlink removes a child dentry from this directory inode. - Unlink(ctx context.Context, name string, child *vfs.Dentry) error - - // RmDir removes an empty child directory from this directory - // inode. Implementations must update the parent directory's link count, - // if required. Implementations are not responsible for checking that child - // is a directory, checking for an empty directory. - RmDir(ctx context.Context, name string, child *vfs.Dentry) error - - // Rename is called on the source directory containing an inode being - // renamed. child should point to the resolved child in the source - // directory. If Rename replaces a dentry in the destination directory, it - // should return the replaced dentry or nil otherwise. - // - // Precondition: Caller must serialize concurrent calls to Rename. - Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (replaced *vfs.Dentry, err error) -} - -type inodeDynamicLookup interface { - // Lookup should return an appropriate dentry if name should resolve to a - // child of this dynamic directory inode. This gives the directory an - // opportunity on every lookup to resolve additional entries that aren't - // hashed into the directory. This is only called when the inode is a - // directory. If the inode is not a directory, or if the directory only - // contains a static set of children, the implementer can unconditionally - // return an appropriate error (ENOTDIR and ENOENT respectively). - // - // The child returned by Lookup will be hashed into the VFS dentry tree. Its - // lifetime can be controlled by the filesystem implementation with an - // appropriate implementation of Valid. - // - // Lookup returns the child with an extra reference and the caller owns this - // reference. - Lookup(ctx context.Context, name string) (*vfs.Dentry, error) - - // Valid should return true if this inode is still valid, or needs to - // be resolved again by a call to Lookup. - Valid(ctx context.Context) bool - - // IterDirents is used to iterate over dynamically created entries. It invokes - // cb on each entry in the directory represented by the FileDescription. - // 'offset' is the offset for the entire IterDirents call, which may include - // results from the caller. 'relOffset' is the offset inside the entries - // returned by this IterDirents invocation. In other words, - // 'offset+relOffset+1' is the value that should be set in vfs.Dirent.NextOff, - // while 'relOffset' is the place where iteration should start from. - IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (newOffset int64, err error) -} - -type inodeSymlink interface { - // Readlink resolves the target of a symbolic link. If an inode is not a - // symlink, the implementation should return EINVAL. - Readlink(ctx context.Context) (string, error) -} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go deleted file mode 100644 index fb0d25ad7..000000000 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ /dev/null @@ -1,320 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernfs_test - -import ( - "bytes" - "fmt" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -const defaultMode linux.FileMode = 01777 -const staticFileContent = "This is sample content for a static test file." - -// RootDentryFn is a generator function for creating the root dentry of a test -// filesystem. See newTestSystem. -type RootDentryFn func(*auth.Credentials, *filesystem) *kernfs.Dentry - -// newTestSystem sets up a minimal environment for running a test, including an -// instance of a test filesystem. Tests can control the contents of the -// filesystem by providing an appropriate rootFn, which should return a -// pre-populated root dentry. -func newTestSystem(t *testing.T, rootFn RootDentryFn) *testutil.System { - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - v := &vfs.VirtualFilesystem{} - if err := v.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mns, err := v.NewMountNamespace(ctx, creds, "", "testfs", &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("Failed to create testfs root mount: %v", err) - } - return testutil.NewSystem(ctx, t, v, mns) -} - -type fsType struct { - rootFn RootDentryFn -} - -type filesystem struct { - kernfs.Filesystem -} - -type file struct { - kernfs.DynamicBytesFile - content string -} - -func (fs *filesystem) newFile(creds *auth.Credentials, content string) *kernfs.Dentry { - f := &file{} - f.content = content - f.DynamicBytesFile.Init(creds, fs.NextIno(), f, 0777) - - d := &kernfs.Dentry{} - d.Init(f) - return d -} - -func (f *file) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%s", f.content) - return nil -} - -type attrs struct { - kernfs.InodeAttrs -} - -func (*attrs) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -type readonlyDir struct { - attrs - kernfs.InodeNotSymlink - kernfs.InodeNoDynamicLookup - kernfs.InodeDirectoryNoNewChildren - - kernfs.OrderedChildren - dentry kernfs.Dentry -} - -func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { - dir := &readonlyDir{} - dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) - dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - dir.dentry.Init(dir) - - dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) - - return &dir.dentry -} - -func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts); err != nil { - return nil, err - } - return fd.VFSFileDescription(), nil -} - -type dir struct { - attrs - kernfs.InodeNotSymlink - kernfs.InodeNoDynamicLookup - - fs *filesystem - dentry kernfs.Dentry - kernfs.OrderedChildren -} - -func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { - dir := &dir{} - dir.fs = fs - dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) - dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{Writable: true}) - dir.dentry.Init(dir) - - dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) - - return &dir.dentry -} - -func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { - creds := auth.CredentialsFromContext(ctx) - dir := d.fs.newDir(creds, opts.Mode, nil) - dirVFSD := dir.VFSDentry() - if err := d.OrderedChildren.Insert(name, dirVFSD); err != nil { - dir.DecRef() - return nil, err - } - d.IncLinks(1) - return dirVFSD, nil -} - -func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { - creds := auth.CredentialsFromContext(ctx) - f := d.fs.newFile(creds, "") - fVFSD := f.VFSDentry() - if err := d.OrderedChildren.Insert(name, fVFSD); err != nil { - f.DecRef() - return nil, err - } - return fVFSD, nil -} - -func (*dir) NewLink(context.Context, string, kernfs.Inode) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -func (*dir) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -func (fst *fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fs := &filesystem{} - fs.Init(vfsObj) - root := fst.rootFn(creds, fs) - return fs.VFSFilesystem(), root.VFSDentry(), nil -} - -// -------------------- Remainder of the file are test cases -------------------- - -func TestBasic(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - "file1": fs.newFile(creds, staticFileContent), - }) - }) - defer sys.Destroy() - sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef() -} - -func TestMkdirGetDentry(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - "dir1": fs.newDir(creds, 0755, nil), - }) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("dir1/a new directory") - if err := sys.VFS.MkdirAt(sys.Ctx, sys.Creds, pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { - t.Fatalf("MkdirAt for PathOperation %+v failed: %v", pop, err) - } - sys.GetDentryOrDie(pop).DecRef() -} - -func TestReadStaticFile(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - "file1": fs.newFile(creds, staticFileContent), - }) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("file1") - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }) - if err != nil { - t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) - } - defer fd.DecRef() - - content, err := sys.ReadToEnd(fd) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - if diff := cmp.Diff(staticFileContent, content); diff != "" { - t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) - } -} - -func TestCreateNewFileInStaticDir(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - "dir1": fs.newDir(creds, 0755, nil), - }) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("dir1/newfile") - opts := &vfs.OpenOptions{Flags: linux.O_CREAT | linux.O_EXCL, Mode: defaultMode} - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, opts) - if err != nil { - t.Fatalf("OpenAt(pop:%+v, opts:%+v) failed: %v", pop, opts, err) - } - - // Close the file. The file should persist. - fd.DecRef() - - fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }) - if err != nil { - t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) - } - fd.DecRef() -} - -func TestDirFDReadWrite(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, nil) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("/") - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }) - if err != nil { - t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) - } - defer fd.DecRef() - - // Read/Write should fail for directory FDs. - if _, err := fd.Read(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { - t.Fatalf("Read for directory FD failed with unexpected error: %v", err) - } - if _, err := fd.Write(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.WriteOptions{}); err != syserror.EBADF { - t.Fatalf("Write for directory FD failed with unexpected error: %v", err) - } -} - -func TestDirFDIterDirents(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - // Fill root with nodes backed by various inode implementations. - "dir1": fs.newReadonlyDir(creds, 0755, nil), - "dir2": fs.newDir(creds, 0755, map[string]*kernfs.Dentry{ - "dir3": fs.newDir(creds, 0755, nil), - }), - "file1": fs.newFile(creds, staticFileContent), - }) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("/") - sys.AssertAllDirentTypes(sys.ListDirents(pop), map[string]testutil.DirentType{ - "dir1": linux.DT_DIR, - "dir2": linux.DT_DIR, - "file1": linux.DT_REG, - }) -} diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go deleted file mode 100644 index 5918d3309..000000000 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ /dev/null @@ -1,61 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// StaticSymlink provides an Inode implementation for symlinks that point to -// a immutable target. -type StaticSymlink struct { - InodeAttrs - InodeNoopRefCount - InodeSymlink - - target string -} - -var _ Inode = (*StaticSymlink)(nil) - -// NewStaticSymlink creates a new symlink file pointing to 'target'. -func NewStaticSymlink(creds *auth.Credentials, ino uint64, target string) *Dentry { - inode := &StaticSymlink{} - inode.Init(creds, ino, target) - - d := &Dentry{} - d.Init(inode) - return d -} - -// Init initializes the instance. -func (s *StaticSymlink) Init(creds *auth.Credentials, ino uint64, target string) { - s.target = target - s.InodeAttrs.Init(creds, ino, linux.ModeSymlink|0777) -} - -// Readlink implements Inode. -func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { - return s.target, nil -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*StaticSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD deleted file mode 100644 index 8156984eb..000000000 --- a/pkg/sentry/fsimpl/proc/BUILD +++ /dev/null @@ -1,67 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -licenses(["notice"]) - -go_library( - name = "proc", - srcs = [ - "filesystem.go", - "subtasks.go", - "task.go", - "task_fds.go", - "task_files.go", - "task_net.go", - "tasks.go", - "tasks_files.go", - "tasks_sys.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/log", - "//pkg/refs", - "//pkg/safemem", - "//pkg/sentry/fs", - "//pkg/sentry/fsbridge", - "//pkg/sentry/fsimpl/kernfs", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/mm", - "//pkg/sentry/socket", - "//pkg/sentry/socket/unix", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/tcpip/header", - "//pkg/usermem", - ], -) - -go_test( - name = "proc_test", - size = "small", - srcs = [ - "tasks_sys_test.go", - "tasks_test.go", - ], - library = ":proc", - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/testutil", - "//pkg/sentry/fsimpl/tmpfs", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go deleted file mode 100644 index 5c19d5522..000000000 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ /dev/null @@ -1,95 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package proc implements a partial in-memory file system for procfs. -package proc - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// Name is the default filesystem name. -const Name = "proc" - -// FilesystemType is the factory class for procfs. -// -// +stateify savable -type FilesystemType struct{} - -var _ vfs.FilesystemType = (*FilesystemType)(nil) - -// GetFilesystem implements vfs.FilesystemType. -func (ft *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - k := kernel.KernelFromContext(ctx) - if k == nil { - return nil, nil, fmt.Errorf("procfs requires a kernel") - } - pidns := kernel.PIDNamespaceFromContext(ctx) - if pidns == nil { - return nil, nil, fmt.Errorf("procfs requires a PID namespace") - } - - procfs := &kernfs.Filesystem{} - procfs.VFSFilesystem().Init(vfsObj, procfs) - - var cgroups map[string]string - if opts.InternalData != nil { - data := opts.InternalData.(*InternalData) - cgroups = data.Cgroups - } - - _, dentry := newTasksInode(procfs, k, pidns, cgroups) - return procfs.VFSFilesystem(), dentry.VFSDentry(), nil -} - -// dynamicInode is an overfitted interface for common Inodes with -// dynamicByteSource types used in procfs. -type dynamicInode interface { - kernfs.Inode - vfs.DynamicBytesSource - - Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) -} - -func newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { - inode.Init(creds, ino, inode, perm) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -type staticFile struct { - kernfs.DynamicBytesFile - vfs.StaticData -} - -var _ dynamicInode = (*staticFile)(nil) - -func newStaticFile(data string) *staticFile { - return &staticFile{StaticData: vfs.StaticData{Data: data}} -} - -// InternalData contains internal data passed in to the procfs mount via -// vfs.GetFilesystemOptions.InternalData. -type InternalData struct { - Cgroups map[string]string -} diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go deleted file mode 100644 index a21313666..000000000 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ /dev/null @@ -1,135 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "sort" - "strconv" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// subtasksInode represents the inode for /proc/[pid]/task/ directory. -// -// +stateify savable -type subtasksInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeAttrs - kernfs.OrderedChildren - kernfs.AlwaysValid - - task *kernel.Task - pidns *kernel.PIDNamespace - inoGen InoGenerator - cgroupControllers map[string]string -} - -var _ kernfs.Inode = (*subtasksInode)(nil) - -func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenerator, cgroupControllers map[string]string) *kernfs.Dentry { - subInode := &subtasksInode{ - task: task, - pidns: pidns, - inoGen: inoGen, - cgroupControllers: cgroupControllers, - } - // Note: credentials are overridden by taskOwnedInode. - subInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - subInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - - inode := &taskOwnedInode{Inode: subInode, owner: task} - dentry := &kernfs.Dentry{} - dentry.Init(inode) - - return dentry -} - -// Lookup implements kernfs.inodeDynamicLookup. -func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - tid, err := strconv.ParseUint(name, 10, 32) - if err != nil { - return nil, syserror.ENOENT - } - - subTask := i.pidns.TaskWithID(kernel.ThreadID(tid)) - if subTask == nil { - return nil, syserror.ENOENT - } - if subTask.ThreadGroup() != i.task.ThreadGroup() { - return nil, syserror.ENOENT - } - - subTaskDentry := newTaskInode(i.inoGen, subTask, i.pidns, false, i.cgroupControllers) - return subTaskDentry.VFSDentry(), nil -} - -// IterDirents implements kernfs.inodeDynamicLookup. -func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { - tasks := i.task.ThreadGroup().MemberIDs(i.pidns) - if len(tasks) == 0 { - return offset, syserror.ENOENT - } - - tids := make([]int, 0, len(tasks)) - for _, tid := range tasks { - tids = append(tids, int(tid)) - } - - sort.Ints(tids) - for _, tid := range tids[relOffset:] { - dirent := vfs.Dirent{ - Name: strconv.FormatUint(uint64(tid), 10), - Type: linux.DT_DIR, - Ino: i.inoGen.NextIno(), - NextOff: offset + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - return offset, nil -} - -// Open implements kernfs.Inode. -func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// Stat implements kernfs.Inode. -func (i *subtasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.InodeAttrs.Stat(vsfs, opts) - if err != nil { - return linux.Statx{}, err - } - if opts.Mask&linux.STATX_NLINK != 0 { - stat.Nlink += uint32(i.task.ThreadGroup().Count()) - } - return stat, nil -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*subtasksInode) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go deleted file mode 100644 index aee2a4392..000000000 --- a/pkg/sentry/fsimpl/proc/task.go +++ /dev/null @@ -1,250 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// taskInode represents the inode for /proc/PID/ directory. -// -// +stateify savable -type taskInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeNoDynamicLookup - kernfs.InodeAttrs - kernfs.OrderedChildren - - task *kernel.Task -} - -var _ kernfs.Inode = (*taskInode)(nil) - -func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { - contents := map[string]*kernfs.Dentry{ - "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), - "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), - "comm": newComm(task, inoGen.NextIno(), 0444), - "environ": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), - "exe": newExeSymlink(task, inoGen.NextIno()), - "fd": newFDDirInode(task, inoGen), - "fdinfo": newFDInfoDirInode(task, inoGen), - "gid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: true}), - "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), - "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), - "mountinfo": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountInfoData{task: task}), - "mounts": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountsData{task: task}), - "net": newTaskNetDir(task, inoGen), - "ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{ - "net": newNamespaceSymlink(task, inoGen.NextIno(), "net"), - "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), - "user": newNamespaceSymlink(task, inoGen.NextIno(), "user"), - }), - "oom_score": newTaskOwnedFile(task, inoGen.NextIno(), 0444, newStaticFile("0\n")), - "oom_score_adj": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &oomScoreAdj{task: task}), - "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), - "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), - "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}), - "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}), - "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), - } - if isThreadGroup { - contents["task"] = newSubtasks(task, pidns, inoGen, cgroupControllers) - } - if len(cgroupControllers) > 0 { - contents["cgroup"] = newTaskOwnedFile(task, inoGen.NextIno(), 0444, newCgroupData(cgroupControllers)) - } - - taskInode := &taskInode{task: task} - // Note: credentials are overridden by taskOwnedInode. - taskInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - - inode := &taskOwnedInode{Inode: taskInode, owner: task} - dentry := &kernfs.Dentry{} - dentry.Init(inode) - - taskInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - links := taskInode.OrderedChildren.Populate(dentry, contents) - taskInode.IncLinks(links) - - return dentry -} - -// Valid implements kernfs.inodeDynamicLookup. This inode remains valid as long -// as the task is still running. When it's dead, another tasks with the same -// PID could replace it. -func (i *taskInode) Valid(ctx context.Context) bool { - return i.task.ExitState() != kernel.TaskExitDead -} - -// Open implements kernfs.Inode. -func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*taskInode) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -// taskOwnedInode implements kernfs.Inode and overrides inode owner with task -// effective user and group. -type taskOwnedInode struct { - kernfs.Inode - - // owner is the task that owns this inode. - owner *kernel.Task -} - -var _ kernfs.Inode = (*taskOwnedInode)(nil) - -func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { - // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, inode, perm) - - taskInode := &taskOwnedInode{Inode: inode, owner: task} - d := &kernfs.Dentry{} - d.Init(taskInode) - return d -} - -func newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, children map[string]*kernfs.Dentry) *kernfs.Dentry { - dir := &kernfs.StaticDirectory{} - - // Note: credentials are overridden by taskOwnedInode. - dir.Init(task.Credentials(), ino, perm) - - inode := &taskOwnedInode{Inode: dir, owner: task} - d := &kernfs.Dentry{} - d.Init(inode) - - dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - links := dir.OrderedChildren.Populate(d, children) - dir.IncLinks(links) - - return d -} - -// Stat implements kernfs.Inode. -func (i *taskOwnedInode) Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.Inode.Stat(fs, opts) - if err != nil { - return linux.Statx{}, err - } - if opts.Mask&(linux.STATX_UID|linux.STATX_GID) != 0 { - uid, gid := i.getOwner(linux.FileMode(stat.Mode)) - if opts.Mask&linux.STATX_UID != 0 { - stat.UID = uint32(uid) - } - if opts.Mask&linux.STATX_GID != 0 { - stat.GID = uint32(gid) - } - } - return stat, nil -} - -// CheckPermissions implements kernfs.Inode. -func (i *taskOwnedInode) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - mode := i.Mode() - uid, gid := i.getOwner(mode) - return vfs.GenericCheckPermissions(creds, ats, mode, uid, gid) -} - -func (i *taskOwnedInode) getOwner(mode linux.FileMode) (auth.KUID, auth.KGID) { - // By default, set the task owner as the file owner. - creds := i.owner.Credentials() - uid := creds.EffectiveKUID - gid := creds.EffectiveKGID - - // Linux doesn't apply dumpability adjustments to world readable/executable - // directories so that applications can stat /proc/PID to determine the - // effective UID of a process. See fs/proc/base.c:task_dump_owner. - if mode.FileType() == linux.ModeDirectory && mode.Permissions() == 0555 { - return uid, gid - } - - // If the task is not dumpable, then root (in the namespace preferred) - // owns the file. - m := getMM(i.owner) - if m == nil { - return auth.RootKUID, auth.RootKGID - } - if m.Dumpability() != mm.UserDumpable { - uid = auth.RootKUID - if kuid := creds.UserNamespace.MapToKUID(auth.RootUID); kuid.Ok() { - uid = kuid - } - gid = auth.RootKGID - if kgid := creds.UserNamespace.MapToKGID(auth.RootGID); kgid.Ok() { - gid = kgid - } - } - return uid, gid -} - -func newIO(t *kernel.Task, isThreadGroup bool) *ioData { - if isThreadGroup { - return &ioData{ioUsage: t.ThreadGroup()} - } - return &ioData{ioUsage: t} -} - -func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { - // Namespace symlinks should contain the namespace name and the inode number - // for the namespace instance, so for example user:[123456]. We currently fake - // the inode number by sticking the symlink inode in its place. - target := fmt.Sprintf("%s:[%d]", ns, ino) - - inode := &kernfs.StaticSymlink{} - // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, target) - - taskInode := &taskOwnedInode{Inode: inode, owner: task} - d := &kernfs.Dentry{} - d.Init(taskInode) - return d -} - -// newCgroupData creates inode that shows cgroup information. -// From man 7 cgroups: "For each cgroup hierarchy of which the process is a -// member, there is one entry containing three colon-separated fields: -// hierarchy-ID:controller-list:cgroup-path" -func newCgroupData(controllers map[string]string) dynamicInode { - var buf bytes.Buffer - - // The hierarchy ids must be positive integers (for cgroup v1), but the - // exact number does not matter, so long as they are unique. We can - // just use a counter, but since linux sorts this file in descending - // order, we must count down to preserve this behavior. - i := len(controllers) - for name, dir := range controllers { - fmt.Fprintf(&buf, "%d:%s:%s\n", i, name, dir) - i-- - } - return newStaticFile(buf.String()) -} diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go deleted file mode 100644 index 76bfc5307..000000000 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ /dev/null @@ -1,287 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - "sort" - "strconv" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -type fdDir struct { - inoGen InoGenerator - task *kernel.Task - - // When produceSymlinks is set, dirents produces for the FDs are reported - // as symlink. Otherwise, they are reported as regular files. - produceSymlink bool -} - -func (i *fdDir) lookup(name string) (*vfs.FileDescription, kernel.FDFlags, error) { - fd, err := strconv.ParseUint(name, 10, 64) - if err != nil { - return nil, kernel.FDFlags{}, syserror.ENOENT - } - - var ( - file *vfs.FileDescription - flags kernel.FDFlags - ) - i.task.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - file, flags = fdTable.GetVFS2(int32(fd)) - } - }) - if file == nil { - return nil, kernel.FDFlags{}, syserror.ENOENT - } - return file, flags, nil -} - -// IterDirents implements kernfs.inodeDynamicLookup. -func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, absOffset, relOffset int64) (int64, error) { - var fds []int32 - i.task.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.GetFDs() - } - }) - - offset := absOffset + relOffset - typ := uint8(linux.DT_REG) - if i.produceSymlink { - typ = linux.DT_LNK - } - - // Find the appropriate starting point. - idx := sort.Search(len(fds), func(i int) bool { return fds[i] >= int32(relOffset) }) - if idx >= len(fds) { - return offset, nil - } - for _, fd := range fds[idx:] { - dirent := vfs.Dirent{ - Name: strconv.FormatUint(uint64(fd), 10), - Type: typ, - Ino: i.inoGen.NextIno(), - NextOff: offset + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - return offset, nil -} - -// fdDirInode represents the inode for /proc/[pid]/fd directory. -// -// +stateify savable -type fdDirInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeAttrs - kernfs.OrderedChildren - kernfs.AlwaysValid - fdDir -} - -var _ kernfs.Inode = (*fdDirInode)(nil) - -func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { - inode := &fdDirInode{ - fdDir: fdDir{ - inoGen: inoGen, - task: task, - produceSymlink: true, - }, - } - inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - - dentry := &kernfs.Dentry{} - dentry.Init(inode) - inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - - return dentry -} - -// Lookup implements kernfs.inodeDynamicLookup. -func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - file, _, err := i.lookup(name) - if err != nil { - return nil, err - } - taskDentry := newFDSymlink(i.task.Credentials(), file, i.inoGen.NextIno()) - return taskDentry.VFSDentry(), nil -} - -// Open implements kernfs.Inode. -func (i *fdDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// CheckPermissions implements kernfs.Inode. -// -// This is to match Linux, which uses a special permission handler to guarantee -// that a process can still access /proc/self/fd after it has executed -// setuid. See fs/proc/fd.c:proc_fd_permission. -func (i *fdDirInode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - err := i.InodeAttrs.CheckPermissions(ctx, creds, ats) - if err == nil { - // Access granted, no extra check needed. - return nil - } - if t := kernel.TaskFromContext(ctx); t != nil { - // Allow access if the task trying to access it is in the thread group - // corresponding to this directory. - if i.task.ThreadGroup() == t.ThreadGroup() { - // Access granted (overridden). - return nil - } - } - return err -} - -// fdSymlink is an symlink for the /proc/[pid]/fd/[fd] file. -// -// +stateify savable -type fdSymlink struct { - refs.AtomicRefCount - kernfs.InodeAttrs - kernfs.InodeSymlink - - file *vfs.FileDescription -} - -var _ kernfs.Inode = (*fdSymlink)(nil) - -func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64) *kernfs.Dentry { - file.IncRef() - inode := &fdSymlink{file: file} - inode.Init(creds, ino, linux.ModeSymlink|0777) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { - root := vfs.RootFromContext(ctx) - defer root.DecRef() - - vfsObj := s.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem() - return vfsObj.PathnameWithDeleted(ctx, root, s.file.VirtualDentry()) -} - -func (s *fdSymlink) DecRef() { - s.AtomicRefCount.DecRefWithDestructor(func() { - s.Destroy() - }) -} - -func (s *fdSymlink) Destroy() { - s.file.DecRef() -} - -// fdInfoDirInode represents the inode for /proc/[pid]/fdinfo directory. -// -// +stateify savable -type fdInfoDirInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeAttrs - kernfs.OrderedChildren - kernfs.AlwaysValid - fdDir -} - -var _ kernfs.Inode = (*fdInfoDirInode)(nil) - -func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { - inode := &fdInfoDirInode{ - fdDir: fdDir{ - inoGen: inoGen, - task: task, - }, - } - inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - - dentry := &kernfs.Dentry{} - dentry.Init(inode) - inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - - return dentry -} - -// Lookup implements kernfs.inodeDynamicLookup. -func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - file, flags, err := i.lookup(name) - if err != nil { - return nil, err - } - - data := &fdInfoData{file: file, flags: flags} - dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data) - return dentry.VFSDentry(), nil -} - -// Open implements kernfs.Inode. -func (i *fdInfoDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// fdInfoData implements vfs.DynamicBytesSource for /proc/[pid]/fdinfo/[fd]. -// -// +stateify savable -type fdInfoData struct { - kernfs.DynamicBytesFile - refs.AtomicRefCount - - file *vfs.FileDescription - flags kernel.FDFlags -} - -var _ dynamicInode = (*fdInfoData)(nil) - -func (d *fdInfoData) DecRef() { - d.AtomicRefCount.DecRefWithDestructor(d.destroy) -} - -func (d *fdInfoData) destroy() { - d.file.DecRef() -} - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *fdInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/121266871): Include pos, locks, and other data. For now we only - // have flags. - // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt - flags := uint(d.file.StatusFlags()) | d.flags.ToLinuxFileFlags() - fmt.Fprintf(buf, "flags:\t0%o\n", flags) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go deleted file mode 100644 index 8c743df8d..000000000 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ /dev/null @@ -1,821 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - "io" - "sort" - "strings" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fsbridge" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// mm gets the kernel task's MemoryManager. No additional reference is taken on -// mm here. This is safe because MemoryManager.destroy is required to leave the -// MemoryManager in a state where it's still usable as a DynamicBytesSource. -func getMM(task *kernel.Task) *mm.MemoryManager { - var tmm *mm.MemoryManager - task.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - tmm = mm - } - }) - return tmm -} - -// getMMIncRef returns t's MemoryManager. If getMMIncRef succeeds, the -// MemoryManager's users count is incremented, and must be decremented by the -// caller when it is no longer in use. -func getMMIncRef(task *kernel.Task) (*mm.MemoryManager, error) { - if task.ExitState() == kernel.TaskExitDead { - return nil, syserror.ESRCH - } - var m *mm.MemoryManager - task.WithMuLocked(func(t *kernel.Task) { - m = t.MemoryManager() - }) - if m == nil || !m.IncUsers() { - return nil, io.EOF - } - return m, nil -} - -type bufferWriter struct { - buf *bytes.Buffer -} - -// WriteFromBlocks writes up to srcs.NumBytes() bytes from srcs and returns -// the number of bytes written. It may return a partial write without an -// error (i.e. (n, nil) where 0 < n < srcs.NumBytes()). It should not -// return a full write with an error (i.e. srcs.NumBytes(), err) where err -// != nil). -func (w *bufferWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - written := srcs.NumBytes() - for !srcs.IsEmpty() { - w.buf.Write(srcs.Head().ToSlice()) - srcs = srcs.Tail() - } - return written, nil -} - -// auxvData implements vfs.DynamicBytesSource for /proc/[pid]/auxv. -// -// +stateify savable -type auxvData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*auxvData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *auxvData) Generate(ctx context.Context, buf *bytes.Buffer) error { - m, err := getMMIncRef(d.task) - if err != nil { - return err - } - defer m.DecUsers(ctx) - - // Space for buffer with AT_NULL (0) terminator at the end. - auxv := m.Auxv() - buf.Grow((len(auxv) + 1) * 16) - for _, e := range auxv { - var tmp [8]byte - usermem.ByteOrder.PutUint64(tmp[:], e.Key) - buf.Write(tmp[:]) - - usermem.ByteOrder.PutUint64(tmp[:], uint64(e.Value)) - buf.Write(tmp[:]) - } - return nil -} - -// execArgType enumerates the types of exec arguments that are exposed through -// proc. -type execArgType int - -const ( - cmdlineDataArg execArgType = iota - environDataArg -) - -// cmdlineData implements vfs.DynamicBytesSource for /proc/[pid]/cmdline. -// -// +stateify savable -type cmdlineData struct { - kernfs.DynamicBytesFile - - task *kernel.Task - - // arg is the type of exec argument this file contains. - arg execArgType -} - -var _ dynamicInode = (*cmdlineData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *cmdlineData) Generate(ctx context.Context, buf *bytes.Buffer) error { - m, err := getMMIncRef(d.task) - if err != nil { - return err - } - defer m.DecUsers(ctx) - - // Figure out the bounds of the exec arg we are trying to read. - var ar usermem.AddrRange - switch d.arg { - case cmdlineDataArg: - ar = usermem.AddrRange{ - Start: m.ArgvStart(), - End: m.ArgvEnd(), - } - case environDataArg: - ar = usermem.AddrRange{ - Start: m.EnvvStart(), - End: m.EnvvEnd(), - } - default: - panic(fmt.Sprintf("unknown exec arg type %v", d.arg)) - } - if ar.Start == 0 || ar.End == 0 { - // Don't attempt to read before the start/end are set up. - return io.EOF - } - - // N.B. Technically this should be usermem.IOOpts.IgnorePermissions = true - // until Linux 4.9 (272ddc8b3735 "proc: don't use FOLL_FORCE for reading - // cmdline and environment"). - writer := &bufferWriter{buf: buf} - if n, err := m.CopyInTo(ctx, usermem.AddrRangeSeqOf(ar), writer, usermem.IOOpts{}); n == 0 || err != nil { - // Nothing to copy or something went wrong. - return err - } - - // On Linux, if the NULL byte at the end of the argument vector has been - // overwritten, it continues reading the environment vector as part of - // the argument vector. - if d.arg == cmdlineDataArg && buf.Bytes()[buf.Len()-1] != 0 { - if end := bytes.IndexByte(buf.Bytes(), 0); end != -1 { - // If we found a NULL character somewhere else in argv, truncate the - // return up to the NULL terminator (including it). - buf.Truncate(end) - return nil - } - - // There is no NULL terminator in the string, return into envp. - arEnvv := usermem.AddrRange{ - Start: m.EnvvStart(), - End: m.EnvvEnd(), - } - - // Upstream limits the returned amount to one page of slop. - // https://elixir.bootlin.com/linux/v4.20/source/fs/proc/base.c#L208 - // we'll return one page total between argv and envp because of the - // above page restrictions. - if buf.Len() >= usermem.PageSize { - // Returned at least one page already, nothing else to add. - return nil - } - remaining := usermem.PageSize - buf.Len() - if int(arEnvv.Length()) > remaining { - end, ok := arEnvv.Start.AddLength(uint64(remaining)) - if !ok { - return syserror.EFAULT - } - arEnvv.End = end - } - if _, err := m.CopyInTo(ctx, usermem.AddrRangeSeqOf(arEnvv), writer, usermem.IOOpts{}); err != nil { - return err - } - - // Linux will return envp up to and including the first NULL character, - // so find it. - if end := bytes.IndexByte(buf.Bytes()[ar.Length():], 0); end != -1 { - buf.Truncate(end) - } - } - - return nil -} - -// +stateify savable -type commInode struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -func newComm(task *kernel.Task, ino uint64, perm linux.FileMode) *kernfs.Dentry { - inode := &commInode{task: task} - inode.DynamicBytesFile.Init(task.Credentials(), ino, &commData{task: task}, perm) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -func (i *commInode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - // This file can always be read or written by members of the same thread - // group. See fs/proc/base.c:proc_tid_comm_permission. - // - // N.B. This check is currently a no-op as we don't yet support writing and - // this file is world-readable anyways. - t := kernel.TaskFromContext(ctx) - if t != nil && t.ThreadGroup() == i.task.ThreadGroup() && !ats.MayExec() { - return nil - } - - return i.DynamicBytesFile.CheckPermissions(ctx, creds, ats) -} - -// commData implements vfs.DynamicBytesSource for /proc/[pid]/comm. -// -// +stateify savable -type commData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*commData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *commData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(d.task.Name()) - buf.WriteString("\n") - return nil -} - -// idMapData implements vfs.DynamicBytesSource for /proc/[pid]/{gid_map|uid_map}. -// -// +stateify savable -type idMapData struct { - kernfs.DynamicBytesFile - - task *kernel.Task - gids bool -} - -var _ dynamicInode = (*idMapData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *idMapData) Generate(ctx context.Context, buf *bytes.Buffer) error { - var entries []auth.IDMapEntry - if d.gids { - entries = d.task.UserNamespace().GIDMap() - } else { - entries = d.task.UserNamespace().UIDMap() - } - for _, e := range entries { - fmt.Fprintf(buf, "%10d %10d %10d\n", e.FirstID, e.FirstParentID, e.Length) - } - return nil -} - -// mapsData implements vfs.DynamicBytesSource for /proc/[pid]/maps. -// -// +stateify savable -type mapsData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*mapsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *mapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - if mm := getMM(d.task); mm != nil { - mm.ReadMapsDataInto(ctx, buf) - } - return nil -} - -// smapsData implements vfs.DynamicBytesSource for /proc/[pid]/smaps. -// -// +stateify savable -type smapsData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*smapsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *smapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - if mm := getMM(d.task); mm != nil { - mm.ReadSmapsDataInto(ctx, buf) - } - return nil -} - -// +stateify savable -type taskStatData struct { - kernfs.DynamicBytesFile - - task *kernel.Task - - // If tgstats is true, accumulate fault stats (not implemented) and CPU - // time across all tasks in t's thread group. - tgstats bool - - // pidns is the PID namespace associated with the proc filesystem that - // includes the file using this statData. - pidns *kernel.PIDNamespace -} - -var _ dynamicInode = (*taskStatData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d ", s.pidns.IDOfTask(s.task)) - fmt.Fprintf(buf, "(%s) ", s.task.Name()) - fmt.Fprintf(buf, "%c ", s.task.StateStatus()[0]) - ppid := kernel.ThreadID(0) - if parent := s.task.Parent(); parent != nil { - ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) - } - fmt.Fprintf(buf, "%d ", ppid) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfProcessGroup(s.task.ThreadGroup().ProcessGroup())) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfSession(s.task.ThreadGroup().Session())) - fmt.Fprintf(buf, "0 0 " /* tty_nr tpgid */) - fmt.Fprintf(buf, "0 " /* flags */) - fmt.Fprintf(buf, "0 0 0 0 " /* minflt cminflt majflt cmajflt */) - var cputime usage.CPUStats - if s.tgstats { - cputime = s.task.ThreadGroup().CPUStats() - } else { - cputime = s.task.CPUStats() - } - fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - cputime = s.task.ThreadGroup().JoinedChildCPUStats() - fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - fmt.Fprintf(buf, "%d %d ", s.task.Priority(), s.task.Niceness()) - fmt.Fprintf(buf, "%d ", s.task.ThreadGroup().Count()) - - // itrealvalue. Since kernel 2.6.17, this field is no longer - // maintained, and is hard coded as 0. - fmt.Fprintf(buf, "0 ") - - // Start time is relative to boot time, expressed in clock ticks. - fmt.Fprintf(buf, "%d ", linux.ClockTFromDuration(s.task.StartTime().Sub(s.task.Kernel().Timekeeper().BootTime()))) - - var vss, rss uint64 - s.task.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - } - }) - fmt.Fprintf(buf, "%d %d ", vss, rss/usermem.PageSize) - - // rsslim. - fmt.Fprintf(buf, "%d ", s.task.ThreadGroup().Limits().Get(limits.Rss).Cur) - - fmt.Fprintf(buf, "0 0 0 0 0 " /* startcode endcode startstack kstkesp kstkeip */) - fmt.Fprintf(buf, "0 0 0 0 0 " /* signal blocked sigignore sigcatch wchan */) - fmt.Fprintf(buf, "0 0 " /* nswap cnswap */) - terminationSignal := linux.Signal(0) - if s.task == s.task.ThreadGroup().Leader() { - terminationSignal = s.task.ThreadGroup().TerminationSignal() - } - fmt.Fprintf(buf, "%d ", terminationSignal) - fmt.Fprintf(buf, "0 0 0 " /* processor rt_priority policy */) - fmt.Fprintf(buf, "0 0 0 " /* delayacct_blkio_ticks guest_time cguest_time */) - fmt.Fprintf(buf, "0 0 0 0 0 0 0 " /* start_data end_data start_brk arg_start arg_end env_start env_end */) - fmt.Fprintf(buf, "0\n" /* exit_code */) - - return nil -} - -// statmData implements vfs.DynamicBytesSource for /proc/[pid]/statm. -// -// +stateify savable -type statmData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*statmData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error { - var vss, rss uint64 - s.task.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - } - }) - - fmt.Fprintf(buf, "%d %d 0 0 0 0 0\n", vss/usermem.PageSize, rss/usermem.PageSize) - return nil -} - -// statusData implements vfs.DynamicBytesSource for /proc/[pid]/status. -// -// +stateify savable -type statusData struct { - kernfs.DynamicBytesFile - - task *kernel.Task - pidns *kernel.PIDNamespace -} - -var _ dynamicInode = (*statusData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "Name:\t%s\n", s.task.Name()) - fmt.Fprintf(buf, "State:\t%s\n", s.task.StateStatus()) - fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.task.ThreadGroup())) - fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.task)) - ppid := kernel.ThreadID(0) - if parent := s.task.Parent(); parent != nil { - ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) - } - fmt.Fprintf(buf, "PPid:\t%d\n", ppid) - tpid := kernel.ThreadID(0) - if tracer := s.task.Tracer(); tracer != nil { - tpid = s.pidns.IDOfTask(tracer) - } - fmt.Fprintf(buf, "TracerPid:\t%d\n", tpid) - var fds int - var vss, rss, data uint64 - s.task.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.Size() - } - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - data = mm.VirtualDataSize() - } - }) - fmt.Fprintf(buf, "FDSize:\t%d\n", fds) - fmt.Fprintf(buf, "VmSize:\t%d kB\n", vss>>10) - fmt.Fprintf(buf, "VmRSS:\t%d kB\n", rss>>10) - fmt.Fprintf(buf, "VmData:\t%d kB\n", data>>10) - fmt.Fprintf(buf, "Threads:\t%d\n", s.task.ThreadGroup().Count()) - creds := s.task.Credentials() - fmt.Fprintf(buf, "CapInh:\t%016x\n", creds.InheritableCaps) - fmt.Fprintf(buf, "CapPrm:\t%016x\n", creds.PermittedCaps) - fmt.Fprintf(buf, "CapEff:\t%016x\n", creds.EffectiveCaps) - fmt.Fprintf(buf, "CapBnd:\t%016x\n", creds.BoundingCaps) - fmt.Fprintf(buf, "Seccomp:\t%d\n", s.task.SeccompMode()) - // We unconditionally report a single NUMA node. See - // pkg/sentry/syscalls/linux/sys_mempolicy.go. - fmt.Fprintf(buf, "Mems_allowed:\t1\n") - fmt.Fprintf(buf, "Mems_allowed_list:\t0\n") - return nil -} - -// ioUsage is the /proc/[pid]/io and /proc/[pid]/task/[tid]/io data provider. -type ioUsage interface { - // IOUsage returns the io usage data. - IOUsage() *usage.IO -} - -// +stateify savable -type ioData struct { - kernfs.DynamicBytesFile - - ioUsage -} - -var _ dynamicInode = (*ioData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (i *ioData) Generate(ctx context.Context, buf *bytes.Buffer) error { - io := usage.IO{} - io.Accumulate(i.IOUsage()) - - fmt.Fprintf(buf, "char: %d\n", io.CharsRead) - fmt.Fprintf(buf, "wchar: %d\n", io.CharsWritten) - fmt.Fprintf(buf, "syscr: %d\n", io.ReadSyscalls) - fmt.Fprintf(buf, "syscw: %d\n", io.WriteSyscalls) - fmt.Fprintf(buf, "read_bytes: %d\n", io.BytesRead) - fmt.Fprintf(buf, "write_bytes: %d\n", io.BytesWritten) - fmt.Fprintf(buf, "cancelled_write_bytes: %d\n", io.BytesWriteCancelled) - return nil -} - -// oomScoreAdj is a stub of the /proc/<pid>/oom_score_adj file. -// -// +stateify savable -type oomScoreAdj struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ vfs.WritableDynamicBytesSource = (*oomScoreAdj)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (o *oomScoreAdj) Generate(ctx context.Context, buf *bytes.Buffer) error { - if o.task.ExitState() == kernel.TaskExitDead { - return syserror.ESRCH - } - fmt.Fprintf(buf, "%d\n", o.task.OOMScoreAdj()) - return nil -} - -// Write implements vfs.WritableDynamicBytesSource.Write. -func (o *oomScoreAdj) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - if src.NumBytes() == 0 { - return 0, nil - } - - // Limit input size so as not to impact performance if input size is large. - src = src.TakeFirst(usermem.PageSize - 1) - - var v int32 - n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) - if err != nil { - return 0, err - } - - if o.task.ExitState() == kernel.TaskExitDead { - return 0, syserror.ESRCH - } - if err := o.task.SetOOMScoreAdj(v); err != nil { - return 0, err - } - - return n, nil -} - -// exeSymlink is an symlink for the /proc/[pid]/exe file. -// -// +stateify savable -type exeSymlink struct { - kernfs.InodeAttrs - kernfs.InodeNoopRefCount - kernfs.InodeSymlink - - task *kernel.Task -} - -var _ kernfs.Inode = (*exeSymlink)(nil) - -func newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentry { - inode := &exeSymlink{task: task} - inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -// Readlink implements kernfs.Inode. -func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { - if !kernel.ContextCanTrace(ctx, s.task, false) { - return "", syserror.EACCES - } - - // Pull out the executable for /proc/[pid]/exe. - exec, err := s.executable() - if err != nil { - return "", err - } - defer exec.DecRef() - - return exec.PathnameWithDeleted(ctx), nil -} - -func (s *exeSymlink) executable() (file fsbridge.File, err error) { - s.task.WithMuLocked(func(t *kernel.Task) { - mm := t.MemoryManager() - if mm == nil { - // TODO(b/34851096): Check shouldn't allow Readlink once the - // Task is zombied. - err = syserror.EACCES - return - } - - // The MemoryManager may be destroyed, in which case - // MemoryManager.destroy will simply set the executable to nil - // (with locks held). - file = mm.Executable() - if file == nil { - err = syserror.ENOENT - } - }) - return -} - -// forEachMountSource runs f for the process root mount and each mount that is -// a descendant of the root. -func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { - var fsctx *kernel.FSContext - t.WithMuLocked(func(t *kernel.Task) { - fsctx = t.FSContext() - }) - if fsctx == nil { - // The task has been destroyed. Nothing to show here. - return - } - - // All mount points must be relative to the rootDir, and mounts outside - // will be excluded. - rootDir := fsctx.RootDirectory() - if rootDir == nil { - // The task has been destroyed. Nothing to show here. - return - } - defer rootDir.DecRef() - - mnt := t.MountNamespace().FindMount(rootDir) - if mnt == nil { - // Has it just been unmounted? - return - } - ms := t.MountNamespace().AllMountsUnder(mnt) - sort.Slice(ms, func(i, j int) bool { - return ms[i].ID < ms[j].ID - }) - for _, m := range ms { - mroot := m.Root() - if mroot == nil { - continue // No longer valid. - } - mountPath, desc := mroot.FullName(rootDir) - mroot.DecRef() - if !desc { - // MountSources that are not descendants of the chroot jail are ignored. - continue - } - fn(mountPath, m) - } -} - -// mountInfoData is used to implement /proc/[pid]/mountinfo. -// -// +stateify savable -type mountInfoData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*mountInfoData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (i *mountInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - forEachMount(i.task, func(mountPath string, m *fs.Mount) { - mroot := m.Root() - if mroot == nil { - return // No longer valid. - } - defer mroot.DecRef() - - // Format: - // 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue - // (1)(2)(3) (4) (5) (6) (7) (8) (9) (10) (11) - - // (1) MountSource ID. - fmt.Fprintf(buf, "%d ", m.ID) - - // (2) Parent ID (or this ID if there is no parent). - pID := m.ID - if !m.IsRoot() && !m.IsUndo() { - pID = m.ParentID - } - fmt.Fprintf(buf, "%d ", pID) - - // (3) Major:Minor device ID. We don't have a superblock, so we - // just use the root inode device number. - sa := mroot.Inode.StableAttr - fmt.Fprintf(buf, "%d:%d ", sa.DeviceFileMajor, sa.DeviceFileMinor) - - // (4) Root: the pathname of the directory in the filesystem - // which forms the root of this mount. - // - // NOTE(b/78135857): This will always be "/" until we implement - // bind mounts. - fmt.Fprintf(buf, "/ ") - - // (5) Mount point (relative to process root). - fmt.Fprintf(buf, "%s ", mountPath) - - // (6) Mount options. - flags := mroot.Inode.MountSource.Flags - opts := "rw" - if flags.ReadOnly { - opts = "ro" - } - if flags.NoAtime { - opts += ",noatime" - } - if flags.NoExec { - opts += ",noexec" - } - fmt.Fprintf(buf, "%s ", opts) - - // (7) Optional fields: zero or more fields of the form "tag[:value]". - // (8) Separator: the end of the optional fields is marked by a single hyphen. - fmt.Fprintf(buf, "- ") - - // (9) Filesystem type. - fmt.Fprintf(buf, "%s ", mroot.Inode.MountSource.FilesystemType) - - // (10) Mount source: filesystem-specific information or "none". - fmt.Fprintf(buf, "none ") - - // (11) Superblock options, and final newline. - fmt.Fprintf(buf, "%s\n", superBlockOpts(mountPath, mroot.Inode.MountSource)) - }) - return nil -} - -func superBlockOpts(mountPath string, msrc *fs.MountSource) string { - // gVisor doesn't (yet) have a concept of super block options, so we - // use the ro/rw bit from the mount flag. - opts := "rw" - if msrc.Flags.ReadOnly { - opts = "ro" - } - - // NOTE(b/147673608): If the mount is a cgroup, we also need to include - // the cgroup name in the options. For now we just read that from the - // path. - // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we - // should get this value from the cgroup itself, and not rely on the - // path. - if msrc.FilesystemType == "cgroup" { - splitPath := strings.Split(mountPath, "/") - cgroupType := splitPath[len(splitPath)-1] - opts += "," + cgroupType - } - return opts -} - -// mountsData is used to implement /proc/[pid]/mounts. -// -// +stateify savable -type mountsData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*mountInfoData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - forEachMount(i.task, func(mountPath string, m *fs.Mount) { - // Format: - // <special device or remote filesystem> <mount point> <filesystem type> <mount options> <needs dump> <fsck order> - // - // We use the filesystem name as the first field, since there - // is no real block device we can point to, and we also should - // not expose anything about the remote filesystem. - // - // Only ro/rw option is supported for now. - // - // The "needs dump"and fsck flags are always 0, which is allowed. - root := m.Root() - if root == nil { - return // No longer valid. - } - defer root.DecRef() - - flags := root.Inode.MountSource.Flags - opts := "rw" - if flags.ReadOnly { - opts = "ro" - } - fmt.Fprintf(buf, "%s %s %s %s %d %d\n", "none", mountPath, root.Inode.MountSource.FilesystemType, opts, 0, 0) - }) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go deleted file mode 100644 index 373a7b17d..000000000 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ /dev/null @@ -1,790 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - "io" - "reflect" - "time" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/socket/unix" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/usermem" -) - -func newTaskNetDir(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { - k := task.Kernel() - pidns := task.PIDNamespace() - root := auth.NewRootCredentials(pidns.UserNamespace()) - - var contents map[string]*kernfs.Dentry - if stack := task.NetworkNamespace().Stack(); stack != nil { - const ( - arp = "IP address HW type Flags HW address Mask Device\n" - netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n" - packet = "sk RefCnt Type Proto Iface R Rmem User Inode\n" - protocols = "protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n" - ptype = "Type Device Function\n" - upd6 = " sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n" - ) - psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)) - - // TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task - // network namespace. - contents = map[string]*kernfs.Dentry{ - "dev": newDentry(root, inoGen.NextIno(), 0444, &netDevData{stack: stack}), - "snmp": newDentry(root, inoGen.NextIno(), 0444, &netSnmpData{stack: stack}), - - // The following files are simple stubs until they are implemented in - // netstack, if the file contains a header the stub is just the header - // otherwise it is an empty file. - "arp": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(arp)), - "netlink": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(netlink)), - "netstat": newDentry(root, inoGen.NextIno(), 0444, &netStatData{}), - "packet": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(packet)), - "protocols": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(protocols)), - - // Linux sets psched values to: nsec per usec, psched tick in ns, 1000000, - // high res timer ticks per sec (ClockGetres returns 1ns resolution). - "psched": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(psched)), - "ptype": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(ptype)), - "route": newDentry(root, inoGen.NextIno(), 0444, &netRouteData{stack: stack}), - "tcp": newDentry(root, inoGen.NextIno(), 0444, &netTCPData{kernel: k}), - "udp": newDentry(root, inoGen.NextIno(), 0444, &netUDPData{kernel: k}), - "unix": newDentry(root, inoGen.NextIno(), 0444, &netUnixData{kernel: k}), - } - - if stack.SupportsIPv6() { - contents["if_inet6"] = newDentry(root, inoGen.NextIno(), 0444, &ifinet6{stack: stack}) - contents["ipv6_route"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")) - contents["tcp6"] = newDentry(root, inoGen.NextIno(), 0444, &netTCP6Data{kernel: k}) - contents["udp6"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile(upd6)) - } - } - - return newTaskOwnedDir(task, inoGen.NextIno(), 0555, contents) -} - -// ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. -// -// +stateify savable -type ifinet6 struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*ifinet6)(nil) - -func (n *ifinet6) contents() []string { - var lines []string - nics := n.stack.Interfaces() - for id, naddrs := range n.stack.InterfaceAddrs() { - nic, ok := nics[id] - if !ok { - // NIC was added after NICNames was called. We'll just ignore it. - continue - } - - for _, a := range naddrs { - // IPv6 only. - if a.Family != linux.AF_INET6 { - continue - } - - // Fields: - // IPv6 address displayed in 32 hexadecimal chars without colons - // Netlink device number (interface index) in hexadecimal (use nic id) - // Prefix length in hexadecimal - // Scope value (use 0) - // Interface flags - // Device name - lines = append(lines, fmt.Sprintf("%032x %02x %02x %02x %02x %8s\n", a.Addr, id, a.PrefixLen, 0, a.Flags, nic.Name)) - } - } - return lines -} - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *ifinet6) Generate(ctx context.Context, buf *bytes.Buffer) error { - for _, l := range n.contents() { - buf.WriteString(l) - } - return nil -} - -// netDevData implements vfs.DynamicBytesSource for /proc/net/dev. -// -// +stateify savable -type netDevData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netDevData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *netDevData) Generate(ctx context.Context, buf *bytes.Buffer) error { - interfaces := n.stack.Interfaces() - buf.WriteString("Inter-| Receive | Transmit\n") - buf.WriteString(" face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n") - - for _, i := range interfaces { - // Implements the same format as - // net/core/net-procfs.c:dev_seq_printf_stats. - var stats inet.StatDev - if err := n.stack.Statistics(&stats, i.Name); err != nil { - log.Warningf("Failed to retrieve interface statistics for %v: %v", i.Name, err) - continue - } - fmt.Fprintf( - buf, - "%6s: %7d %7d %4d %4d %4d %5d %10d %9d %8d %7d %4d %4d %4d %5d %7d %10d\n", - i.Name, - // Received - stats[0], // bytes - stats[1], // packets - stats[2], // errors - stats[3], // dropped - stats[4], // fifo - stats[5], // frame - stats[6], // compressed - stats[7], // multicast - // Transmitted - stats[8], // bytes - stats[9], // packets - stats[10], // errors - stats[11], // dropped - stats[12], // fifo - stats[13], // frame - stats[14], // compressed - stats[15], // multicast - ) - } - - return nil -} - -// netUnixData implements vfs.DynamicBytesSource for /proc/net/unix. -// -// +stateify savable -type netUnixData struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netUnixData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") - for _, se := range n.kernel.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) - continue - } - sfile := s.(*fs.File) - if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { - s.DecRef() - // Not a unix socket. - continue - } - sops := sfile.FileOperations.(*unix.SocketOperations) - - addr, err := sops.Endpoint().GetLocalAddress() - if err != nil { - log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err) - addr.Addr = "<unknown>" - } - - sockFlags := 0 - if ce, ok := sops.Endpoint().(transport.ConnectingEndpoint); ok { - if ce.Listening() { - // For unix domain sockets, linux reports a single flag - // value if the socket is listening, of __SO_ACCEPTCON. - sockFlags = linux.SO_ACCEPTCON - } - } - - // In the socket entry below, the value for the 'Num' field requires - // some consideration. Linux prints the address to the struct - // unix_sock representing a socket in the kernel, but may redact the - // value for unprivileged users depending on the kptr_restrict - // sysctl. - // - // One use for this field is to allow a privileged user to - // introspect into the kernel memory to determine information about - // a socket not available through procfs, such as the socket's peer. - // - // In gvisor, returning a pointer to our internal structures would - // be pointless, as it wouldn't match the memory layout for struct - // unix_sock, making introspection difficult. We could populate a - // struct unix_sock with the appropriate data, but even that - // requires consideration for which kernel version to emulate, as - // the definition of this struct changes over time. - // - // For now, we always redact this pointer. - fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d", - (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. - sfile.ReadRefs()-1, // RefCount, don't count our own ref. - 0, // Protocol, always 0 for UDS. - sockFlags, // Flags. - sops.Endpoint().Type(), // Type. - sops.State(), // State. - sfile.InodeID(), // Inode. - ) - - // Path - if len(addr.Addr) != 0 { - if addr.Addr[0] == 0 { - // Abstract path. - fmt.Fprintf(buf, " @%s", string(addr.Addr[1:])) - } else { - fmt.Fprintf(buf, " %s", string(addr.Addr)) - } - } - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - return nil -} - -func networkToHost16(n uint16) uint16 { - // n is in network byte order, so is big-endian. The most-significant byte - // should be stored in the lower address. - // - // We manually inline binary.BigEndian.Uint16() because Go does not support - // non-primitive consts, so binary.BigEndian is a (mutable) var, so calls to - // binary.BigEndian.Uint16() require a read of binary.BigEndian and an - // interface method call, defeating inlining. - buf := [2]byte{byte(n >> 8 & 0xff), byte(n & 0xff)} - return usermem.ByteOrder.Uint16(buf[:]) -} - -func writeInetAddr(w io.Writer, family int, i linux.SockAddr) { - switch family { - case linux.AF_INET: - var a linux.SockAddrInet - if i != nil { - a = *i.(*linux.SockAddrInet) - } - - // linux.SockAddrInet.Port is stored in the network byte order and is - // printed like a number in host byte order. Note that all numbers in host - // byte order are printed with the most-significant byte first when - // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. - port := networkToHost16(a.Port) - - // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order - // (i.e. most-significant byte in index 0). Linux represents this as a - // __be32 which is a typedef for an unsigned int, and is printed with - // %X. This means that for a little-endian machine, Linux prints the - // least-significant byte of the address first. To emulate this, we first - // invert the byte order for the address using usermem.ByteOrder.Uint32, - // which makes it have the equivalent encoding to a __be32 on a little - // endian machine. Note that this operation is a no-op on a big endian - // machine. Then similar to Linux, we format it with %X, which will print - // the most-significant byte of the __be32 address first, which is now - // actually the least-significant byte of the original address in - // linux.SockAddrInet.Addr on little endian machines, due to the conversion. - addr := usermem.ByteOrder.Uint32(a.Addr[:]) - - fmt.Fprintf(w, "%08X:%04X ", addr, port) - case linux.AF_INET6: - var a linux.SockAddrInet6 - if i != nil { - a = *i.(*linux.SockAddrInet6) - } - - port := networkToHost16(a.Port) - addr0 := usermem.ByteOrder.Uint32(a.Addr[0:4]) - addr1 := usermem.ByteOrder.Uint32(a.Addr[4:8]) - addr2 := usermem.ByteOrder.Uint32(a.Addr[8:12]) - addr3 := usermem.ByteOrder.Uint32(a.Addr[12:16]) - fmt.Fprintf(w, "%08X%08X%08X%08X:%04X ", addr0, addr1, addr2, addr3, port) - } -} - -func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, family int) error { - // t may be nil here if our caller is not part of a task goroutine. This can - // happen for example if we're here for "sentryctl cat". When t is nil, - // degrade gracefully and retrieve what we can. - t := kernel.TaskFromContext(ctx) - - for _, se := range k.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) - continue - } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) - if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) - } - if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { - s.DecRef() - // Not tcp4 sockets. - continue - } - - // Linux's documentation for the fields below can be found at - // https://www.kernel.org/doc/Documentation/networking/proc_net_tcp.txt. - // For Linux's implementation, see net/ipv4/tcp_ipv4.c:get_tcp4_sock(). - // Note that the header doesn't contain labels for all the fields. - - // Field: sl; entry number. - fmt.Fprintf(buf, "%4d: ", se.ID) - - // Field: local_adddress. - var localAddr linux.SockAddr - if t != nil { - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = local - } - } - writeInetAddr(buf, family, localAddr) - - // Field: rem_address. - var remoteAddr linux.SockAddr - if t != nil { - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = remote - } - } - writeInetAddr(buf, family, remoteAddr) - - // Field: state; socket state. - fmt.Fprintf(buf, "%02X ", sops.State()) - - // Field: tx_queue, rx_queue; number of packets in the transmit and - // receive queue. Unimplemented. - fmt.Fprintf(buf, "%08X:%08X ", 0, 0) - - // Field: tr, tm->when; timer active state and number of jiffies - // until timer expires. Unimplemented. - fmt.Fprintf(buf, "%02X:%08X ", 0, 0) - - // Field: retrnsmt; number of unrecovered RTO timeouts. - // Unimplemented. - fmt.Fprintf(buf, "%08X ", 0) - - // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) - fmt.Fprintf(buf, "%5d ", 0) - } else { - creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) - } - - // Field: timeout; number of unanswered 0-window probes. - // Unimplemented. - fmt.Fprintf(buf, "%8d ", 0) - - // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) - - // Field: refcount. Don't count the ref we obtain while deferencing - // the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) - - // Field: Socket struct address. Redacted due to the same reason as - // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. - fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) - - // Field: retransmit timeout. Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: predicted tick of soft clock (delayed ACK control data). - // Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: (ack.quick<<1)|ack.pingpong, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: sending congestion window, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: Slow start size threshold, -1 if threshold >= 0xFFFF. - // Unimplemented, report as large threshold. - fmt.Fprintf(buf, "%d", -1) - - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - - return nil -} - -// netTCPData implements vfs.DynamicBytesSource for /proc/net/tcp. -// -// +stateify savable -type netTCPData struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netTCPData)(nil) - -func (d *netTCPData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") - return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET) -} - -// netTCP6Data implements vfs.DynamicBytesSource for /proc/net/tcp6. -// -// +stateify savable -type netTCP6Data struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netTCP6Data)(nil) - -func (d *netTCP6Data) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n") - return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET6) -} - -// netUDPData implements vfs.DynamicBytesSource for /proc/net/udp. -// -// +stateify savable -type netUDPData struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netUDPData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // t may be nil here if our caller is not part of a task goroutine. This can - // happen for example if we're here for "sentryctl cat". When t is nil, - // degrade gracefully and retrieve what we can. - t := kernel.TaskFromContext(ctx) - - for _, se := range d.kernel.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) - continue - } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) - if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) - } - if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { - s.DecRef() - // Not udp4 socket. - continue - } - - // For Linux's implementation, see net/ipv4/udp.c:udp4_format_sock(). - - // Field: sl; entry number. - fmt.Fprintf(buf, "%5d: ", se.ID) - - // Field: local_adddress. - var localAddr linux.SockAddrInet - if t != nil { - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = *local.(*linux.SockAddrInet) - } - } - writeInetAddr(buf, linux.AF_INET, &localAddr) - - // Field: rem_address. - var remoteAddr linux.SockAddrInet - if t != nil { - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = *remote.(*linux.SockAddrInet) - } - } - writeInetAddr(buf, linux.AF_INET, &remoteAddr) - - // Field: state; socket state. - fmt.Fprintf(buf, "%02X ", sops.State()) - - // Field: tx_queue, rx_queue; number of packets in the transmit and - // receive queue. Unimplemented. - fmt.Fprintf(buf, "%08X:%08X ", 0, 0) - - // Field: tr, tm->when. Always 0 for UDP. - fmt.Fprintf(buf, "%02X:%08X ", 0, 0) - - // Field: retrnsmt. Always 0 for UDP. - fmt.Fprintf(buf, "%08X ", 0) - - // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) - fmt.Fprintf(buf, "%5d ", 0) - } else { - creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) - } - - // Field: timeout. Always 0 for UDP. - fmt.Fprintf(buf, "%8d ", 0) - - // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) - - // Field: ref; reference count on the socket inode. Don't count the ref - // we obtain while deferencing the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) - - // Field: Socket struct address. Redacted due to the same reason as - // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. - fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) - - // Field: drops; number of dropped packets. Unimplemented. - fmt.Fprintf(buf, "%d", 0) - - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - return nil -} - -// netSnmpData implements vfs.DynamicBytesSource for /proc/net/snmp. -// -// +stateify savable -type netSnmpData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netSnmpData)(nil) - -type snmpLine struct { - prefix string - header string -} - -var snmp = []snmpLine{ - { - prefix: "Ip", - header: "Forwarding DefaultTTL InReceives InHdrErrors InAddrErrors ForwDatagrams InUnknownProtos InDiscards InDelivers OutRequests OutDiscards OutNoRoutes ReasmTimeout ReasmReqds ReasmOKs ReasmFails FragOKs FragFails FragCreates", - }, - { - prefix: "Icmp", - header: "InMsgs InErrors InCsumErrors InDestUnreachs InTimeExcds InParmProbs InSrcQuenchs InRedirects InEchos InEchoReps InTimestamps InTimestampReps InAddrMasks InAddrMaskReps OutMsgs OutErrors OutDestUnreachs OutTimeExcds OutParmProbs OutSrcQuenchs OutRedirects OutEchos OutEchoReps OutTimestamps OutTimestampReps OutAddrMasks OutAddrMaskReps", - }, - { - prefix: "IcmpMsg", - }, - { - prefix: "Tcp", - header: "RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails EstabResets CurrEstab InSegs OutSegs RetransSegs InErrs OutRsts InCsumErrors", - }, - { - prefix: "Udp", - header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", - }, - { - prefix: "UdpLite", - header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", - }, -} - -func toSlice(a interface{}) []uint64 { - v := reflect.Indirect(reflect.ValueOf(a)) - return v.Slice(0, v.Len()).Interface().([]uint64) -} - -func sprintSlice(s []uint64) string { - if len(s) == 0 { - return "" - } - r := fmt.Sprint(s) - return r[1 : len(r)-1] // Remove "[]" introduced by fmt of slice. -} - -// Generate implements vfs.DynamicBytesSource. -func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error { - types := []interface{}{ - &inet.StatSNMPIP{}, - &inet.StatSNMPICMP{}, - nil, // TODO(gvisor.dev/issue/628): Support IcmpMsg stats. - &inet.StatSNMPTCP{}, - &inet.StatSNMPUDP{}, - &inet.StatSNMPUDPLite{}, - } - for i, stat := range types { - line := snmp[i] - if stat == nil { - fmt.Fprintf(buf, "%s:\n", line.prefix) - fmt.Fprintf(buf, "%s:\n", line.prefix) - continue - } - if err := d.stack.Statistics(stat, line.prefix); err != nil { - if err == syserror.EOPNOTSUPP { - log.Infof("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) - } else { - log.Warningf("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) - } - } - - fmt.Fprintf(buf, "%s: %s\n", line.prefix, line.header) - - if line.prefix == "Tcp" { - tcp := stat.(*inet.StatSNMPTCP) - // "Tcp" needs special processing because MaxConn is signed. RFC 2012. - fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) - } else { - fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) - } - } - return nil -} - -// netRouteData implements vfs.DynamicBytesSource for /proc/net/route. -// -// +stateify savable -type netRouteData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netRouteData)(nil) - -// Generate implements vfs.DynamicBytesSource. -// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. -func (d *netRouteData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%-127s\n", "Iface\tDestination\tGateway\tFlags\tRefCnt\tUse\tMetric\tMask\tMTU\tWindow\tIRTT") - - interfaces := d.stack.Interfaces() - for _, rt := range d.stack.RouteTable() { - // /proc/net/route only includes ipv4 routes. - if rt.Family != linux.AF_INET { - continue - } - - // /proc/net/route does not include broadcast or multicast routes. - if rt.Type == linux.RTN_BROADCAST || rt.Type == linux.RTN_MULTICAST { - continue - } - - iface, ok := interfaces[rt.OutputInterface] - if !ok || iface.Name == "lo" { - continue - } - - var ( - gw uint32 - prefix uint32 - flags = linux.RTF_UP - ) - if len(rt.GatewayAddr) == header.IPv4AddressSize { - flags |= linux.RTF_GATEWAY - gw = usermem.ByteOrder.Uint32(rt.GatewayAddr) - } - if len(rt.DstAddr) == header.IPv4AddressSize { - prefix = usermem.ByteOrder.Uint32(rt.DstAddr) - } - l := fmt.Sprintf( - "%s\t%08X\t%08X\t%04X\t%d\t%d\t%d\t%08X\t%d\t%d\t%d", - iface.Name, - prefix, - gw, - flags, - 0, // RefCnt. - 0, // Use. - 0, // Metric. - (uint32(1)<<rt.DstLen)-1, - 0, // MTU. - 0, // Window. - 0, // RTT. - ) - fmt.Fprintf(buf, "%-127s\n", l) - } - return nil -} - -// netStatData implements vfs.DynamicBytesSource for /proc/net/netstat. -// -// +stateify savable -type netStatData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netStatData)(nil) - -// Generate implements vfs.DynamicBytesSource. -// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. -func (d *netStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed " + - "EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps " + - "LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive " + - "PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost " + - "ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog " + - "TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser " + - "TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging " + - "TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo " + - "TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit " + - "TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans " + - "TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes " + - "TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail " + - "TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent " + - "TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose " + - "TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed " + - "TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld " + - "TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected " + - "TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback " + - "TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter " + - "TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail " + - "TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK " + - "TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail " + - "TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow " + - "TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets " + - "TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv " + - "TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect " + - "TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd " + - "TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq " + - "TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge " + - "TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess\n") - return nil -} diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go deleted file mode 100644 index 9f2ef8200..000000000 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ /dev/null @@ -1,257 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "sort" - "strconv" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -const ( - selfName = "self" - threadSelfName = "thread-self" -) - -// InoGenerator generates unique inode numbers for a given filesystem. -type InoGenerator interface { - NextIno() uint64 -} - -// tasksInode represents the inode for /proc/ directory. -// -// +stateify savable -type tasksInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeAttrs - kernfs.OrderedChildren - kernfs.AlwaysValid - - inoGen InoGenerator - pidns *kernel.PIDNamespace - - // '/proc/self' and '/proc/thread-self' have custom directory offsets in - // Linux. So handle them outside of OrderedChildren. - selfSymlink *vfs.Dentry - threadSelfSymlink *vfs.Dentry - - // cgroupControllers is a map of controller name to directory in the - // cgroup hierarchy. These controllers are immutable and will be listed - // in /proc/pid/cgroup if not nil. - cgroupControllers map[string]string -} - -var _ kernfs.Inode = (*tasksInode)(nil) - -func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { - root := auth.NewRootCredentials(pidns.UserNamespace()) - contents := map[string]*kernfs.Dentry{ - "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFileSetStat(cpuInfoData(k))), - "filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), - "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), - "sys": newSysDir(root, inoGen, k), - "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), - "net": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/net"), - "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), - "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), - "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), - } - - inode := &tasksInode{ - pidns: pidns, - inoGen: inoGen, - selfSymlink: newSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), - threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), - cgroupControllers: cgroupControllers, - } - inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) - - dentry := &kernfs.Dentry{} - dentry.Init(inode) - - inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - links := inode.OrderedChildren.Populate(dentry, contents) - inode.IncLinks(links) - - return inode, dentry -} - -// Lookup implements kernfs.inodeDynamicLookup. -func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - // Try to lookup a corresponding task. - tid, err := strconv.ParseUint(name, 10, 64) - if err != nil { - // If it failed to parse, check if it's one of the special handled files. - switch name { - case selfName: - return i.selfSymlink, nil - case threadSelfName: - return i.threadSelfSymlink, nil - } - return nil, syserror.ENOENT - } - - task := i.pidns.TaskWithID(kernel.ThreadID(tid)) - if task == nil { - return nil, syserror.ENOENT - } - - taskDentry := newTaskInode(i.inoGen, task, i.pidns, true, i.cgroupControllers) - return taskDentry.VFSDentry(), nil -} - -// IterDirents implements kernfs.inodeDynamicLookup. -func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, _ int64) (int64, error) { - // fs/proc/internal.h: #define FIRST_PROCESS_ENTRY 256 - const FIRST_PROCESS_ENTRY = 256 - - // Use maxTaskID to shortcut searches that will result in 0 entries. - const maxTaskID = kernel.TasksLimit + 1 - if offset >= maxTaskID { - return offset, nil - } - - // According to Linux (fs/proc/base.c:proc_pid_readdir()), process directories - // start at offset FIRST_PROCESS_ENTRY with '/proc/self', followed by - // '/proc/thread-self' and then '/proc/[pid]'. - if offset < FIRST_PROCESS_ENTRY { - offset = FIRST_PROCESS_ENTRY - } - - if offset == FIRST_PROCESS_ENTRY { - dirent := vfs.Dirent{ - Name: selfName, - Type: linux.DT_LNK, - Ino: i.inoGen.NextIno(), - NextOff: offset + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - if offset == FIRST_PROCESS_ENTRY+1 { - dirent := vfs.Dirent{ - Name: threadSelfName, - Type: linux.DT_LNK, - Ino: i.inoGen.NextIno(), - NextOff: offset + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - - // Collect all tasks that TGIDs are greater than the offset specified. Per - // Linux we only include in directory listings if it's the leader. But for - // whatever crazy reason, you can still walk to the given node. - var tids []int - startTid := offset - FIRST_PROCESS_ENTRY - 2 - for _, tg := range i.pidns.ThreadGroups() { - tid := i.pidns.IDOfThreadGroup(tg) - if int64(tid) < startTid { - continue - } - if leader := tg.Leader(); leader != nil { - tids = append(tids, int(tid)) - } - } - - if len(tids) == 0 { - return offset, nil - } - - sort.Ints(tids) - for _, tid := range tids { - dirent := vfs.Dirent{ - Name: strconv.FormatUint(uint64(tid), 10), - Type: linux.DT_DIR, - Ino: i.inoGen.NextIno(), - NextOff: FIRST_PROCESS_ENTRY + 2 + int64(tid) + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - return maxTaskID, nil -} - -// Open implements kernfs.Inode. -func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -func (i *tasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.InodeAttrs.Stat(vsfs, opts) - if err != nil { - return linux.Statx{}, err - } - - if opts.Mask&linux.STATX_NLINK != 0 { - // Add dynamic children to link count. - for _, tg := range i.pidns.ThreadGroups() { - if leader := tg.Leader(); leader != nil { - stat.Nlink++ - } - } - } - - return stat, nil -} - -// staticFileSetStat implements a special static file that allows inode -// attributes to be set. This is to support /proc files that are readonly, but -// allow attributes to be set. -type staticFileSetStat struct { - dynamicBytesFileSetAttr - vfs.StaticData -} - -var _ dynamicInode = (*staticFileSetStat)(nil) - -func newStaticFileSetStat(data string) *staticFileSetStat { - return &staticFileSetStat{StaticData: vfs.StaticData{Data: data}} -} - -func cpuInfoData(k *kernel.Kernel) string { - features := k.FeatureSet() - if features == nil { - // Kernel is always initialized with a FeatureSet. - panic("cpuinfo read with nil FeatureSet") - } - var buf bytes.Buffer - for i, max := uint(0), k.ApplicationCores(); i < max; i++ { - features.WriteCPUInfoTo(i, &buf) - } - return buf.String() -} - -func shmData(v uint64) dynamicInode { - return newStaticFile(strconv.FormatUint(v, 10)) -} diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go deleted file mode 100644 index 882c1981e..000000000 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ /dev/null @@ -1,370 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - "strconv" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -type selfSymlink struct { - kernfs.InodeAttrs - kernfs.InodeNoopRefCount - kernfs.InodeSymlink - - pidns *kernel.PIDNamespace -} - -var _ kernfs.Inode = (*selfSymlink)(nil) - -func newSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { - inode := &selfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|0777) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { - t := kernel.TaskFromContext(ctx) - if t == nil { - // Who is reading this link? - return "", syserror.EINVAL - } - tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) - if tgid == 0 { - return "", syserror.ENOENT - } - return strconv.FormatUint(uint64(tgid), 10), nil -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -type threadSelfSymlink struct { - kernfs.InodeAttrs - kernfs.InodeNoopRefCount - kernfs.InodeSymlink - - pidns *kernel.PIDNamespace -} - -var _ kernfs.Inode = (*threadSelfSymlink)(nil) - -func newThreadSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { - inode := &threadSelfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|0777) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { - t := kernel.TaskFromContext(ctx) - if t == nil { - // Who is reading this link? - return "", syserror.EINVAL - } - tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) - tid := s.pidns.IDOfTask(t) - if tid == 0 || tgid == 0 { - return "", syserror.ENOENT - } - return fmt.Sprintf("%d/task/%d", tgid, tid), nil -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -// dynamicBytesFileSetAttr implements a special file that allows inode -// attributes to be set. This is to support /proc files that are readonly, but -// allow attributes to be set. -type dynamicBytesFileSetAttr struct { - kernfs.DynamicBytesFile -} - -// SetStat implements Inode.SetStat. -func (d *dynamicBytesFileSetAttr) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { - return d.DynamicBytesFile.InodeAttrs.SetStat(ctx, fs, creds, opts) -} - -// cpuStats contains the breakdown of CPU time for /proc/stat. -type cpuStats struct { - // user is time spent in userspace tasks with non-positive niceness. - user uint64 - - // nice is time spent in userspace tasks with positive niceness. - nice uint64 - - // system is time spent in non-interrupt kernel context. - system uint64 - - // idle is time spent idle. - idle uint64 - - // ioWait is time spent waiting for IO. - ioWait uint64 - - // irq is time spent in interrupt context. - irq uint64 - - // softirq is time spent in software interrupt context. - softirq uint64 - - // steal is involuntary wait time. - steal uint64 - - // guest is time spent in guests with non-positive niceness. - guest uint64 - - // guestNice is time spent in guests with positive niceness. - guestNice uint64 -} - -// String implements fmt.Stringer. -func (c cpuStats) String() string { - return fmt.Sprintf("%d %d %d %d %d %d %d %d %d %d", c.user, c.nice, c.system, c.idle, c.ioWait, c.irq, c.softirq, c.steal, c.guest, c.guestNice) -} - -// statData implements vfs.DynamicBytesSource for /proc/stat. -// -// +stateify savable -type statData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*statData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*statData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/37226836): We currently export only zero CPU stats. We could - // at least provide some aggregate stats. - var cpu cpuStats - fmt.Fprintf(buf, "cpu %s\n", cpu) - - k := kernel.KernelFromContext(ctx) - for c, max := uint(0), k.ApplicationCores(); c < max; c++ { - fmt.Fprintf(buf, "cpu%d %s\n", c, cpu) - } - - // The total number of interrupts is dependent on the CPUs and PCI - // devices on the system. See arch_probe_nr_irqs. - // - // Since we don't report real interrupt stats, just choose an arbitrary - // value from a representative VM. - const numInterrupts = 256 - - // The Kernel doesn't handle real interrupts, so report all zeroes. - // TODO(b/37226836): We could count page faults as #PF. - fmt.Fprintf(buf, "intr 0") // total - for i := 0; i < numInterrupts; i++ { - fmt.Fprintf(buf, " 0") - } - fmt.Fprintf(buf, "\n") - - // Total number of context switches. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "ctxt 0\n") - - // CLOCK_REALTIME timestamp from boot, in seconds. - fmt.Fprintf(buf, "btime %d\n", k.Timekeeper().BootTime().Seconds()) - - // Total number of clones. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "processes 0\n") - - // Number of runnable tasks. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "procs_running 0\n") - - // Number of tasks waiting on IO. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "procs_blocked 0\n") - - // Number of each softirq handled. - fmt.Fprintf(buf, "softirq 0") // total - for i := 0; i < linux.NumSoftIRQ; i++ { - fmt.Fprintf(buf, " 0") - } - fmt.Fprintf(buf, "\n") - return nil -} - -// loadavgData backs /proc/loadavg. -// -// +stateify savable -type loadavgData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*loadavgData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/62345059): Include real data in fields. - // Column 1-3: CPU and IO utilization of the last 1, 5, and 10 minute periods. - // Column 4-5: currently running processes and the total number of processes. - // Column 6: the last process ID used. - fmt.Fprintf(buf, "%.2f %.2f %.2f %d/%d %d\n", 0.00, 0.00, 0.00, 0, 0, 0) - return nil -} - -// meminfoData implements vfs.DynamicBytesSource for /proc/meminfo. -// -// +stateify savable -type meminfoData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*meminfoData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - k := kernel.KernelFromContext(ctx) - mf := k.MemoryFile() - mf.UpdateUsage() - snapshot, totalUsage := usage.MemoryAccounting.Copy() - totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) - anon := snapshot.Anonymous + snapshot.Tmpfs - file := snapshot.PageCache + snapshot.Mapped - // We don't actually have active/inactive LRUs, so just make up numbers. - activeFile := (file / 2) &^ (usermem.PageSize - 1) - inactiveFile := file - activeFile - - fmt.Fprintf(buf, "MemTotal: %8d kB\n", totalSize/1024) - memFree := (totalSize - totalUsage) / 1024 - // We use MemFree as MemAvailable because we don't swap. - // TODO(rahat): When reclaim is implemented the value of MemAvailable - // should change. - fmt.Fprintf(buf, "MemFree: %8d kB\n", memFree) - fmt.Fprintf(buf, "MemAvailable: %8d kB\n", memFree) - fmt.Fprintf(buf, "Buffers: 0 kB\n") // memory usage by block devices - fmt.Fprintf(buf, "Cached: %8d kB\n", (file+snapshot.Tmpfs)/1024) - // Emulate a system with no swap, which disables inactivation of anon pages. - fmt.Fprintf(buf, "SwapCache: 0 kB\n") - fmt.Fprintf(buf, "Active: %8d kB\n", (anon+activeFile)/1024) - fmt.Fprintf(buf, "Inactive: %8d kB\n", inactiveFile/1024) - fmt.Fprintf(buf, "Active(anon): %8d kB\n", anon/1024) - fmt.Fprintf(buf, "Inactive(anon): 0 kB\n") - fmt.Fprintf(buf, "Active(file): %8d kB\n", activeFile/1024) - fmt.Fprintf(buf, "Inactive(file): %8d kB\n", inactiveFile/1024) - fmt.Fprintf(buf, "Unevictable: 0 kB\n") // TODO(b/31823263) - fmt.Fprintf(buf, "Mlocked: 0 kB\n") // TODO(b/31823263) - fmt.Fprintf(buf, "SwapTotal: 0 kB\n") - fmt.Fprintf(buf, "SwapFree: 0 kB\n") - fmt.Fprintf(buf, "Dirty: 0 kB\n") - fmt.Fprintf(buf, "Writeback: 0 kB\n") - fmt.Fprintf(buf, "AnonPages: %8d kB\n", anon/1024) - fmt.Fprintf(buf, "Mapped: %8d kB\n", file/1024) // doesn't count mapped tmpfs, which we don't know - fmt.Fprintf(buf, "Shmem: %8d kB\n", snapshot.Tmpfs/1024) - return nil -} - -// uptimeData implements vfs.DynamicBytesSource for /proc/uptime. -// -// +stateify savable -type uptimeData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*uptimeData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*uptimeData) Generate(ctx context.Context, buf *bytes.Buffer) error { - k := kernel.KernelFromContext(ctx) - now := time.NowFromContext(ctx) - - // Pretend that we've spent zero time sleeping (second number). - fmt.Fprintf(buf, "%.2f 0.00\n", now.Sub(k.Timekeeper().BootTime()).Seconds()) - return nil -} - -// versionData implements vfs.DynamicBytesSource for /proc/version. -// -// +stateify savable -type versionData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*versionData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { - k := kernel.KernelFromContext(ctx) - init := k.GlobalInit() - if init == nil { - // Attempted to read before the init Task is created. This can - // only occur during startup, which should never need to read - // this file. - panic("Attempted to read version before initial Task is available") - } - - // /proc/version takes the form: - // - // "SYSNAME version RELEASE (COMPILE_USER@COMPILE_HOST) - // (COMPILER_VERSION) VERSION" - // - // where: - // - SYSNAME, RELEASE, and VERSION are the same as returned by - // sys_utsname - // - COMPILE_USER is the user that build the kernel - // - COMPILE_HOST is the hostname of the machine on which the kernel - // was built - // - COMPILER_VERSION is the version reported by the building compiler - // - // Since we don't really want to expose build information to - // applications, those fields are omitted. - // - // FIXME(mpratt): Using Version from the init task SyscallTable - // disregards the different version a task may have (e.g., in a uts - // namespace). - ver := init.Leader().SyscallTable().Version - fmt.Fprintf(buf, "%s version %s %s\n", ver.Sysname, ver.Release, ver.Version) - return nil -} - -// filesystemsData backs /proc/filesystems. -// -// +stateify savable -type filesystemsData struct { - kernfs.DynamicBytesFile -} - -var _ dynamicInode = (*filesystemsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *filesystemsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - k := kernel.KernelFromContext(ctx) - k.VFS().GenerateProcFilesystems(buf) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go deleted file mode 100644 index 3d5dc463c..000000000 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ /dev/null @@ -1,211 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// newSysDir returns the dentry corresponding to /proc/sys directory. -func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "kernel": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "hostname": newDentry(root, inoGen.NextIno(), 0444, &hostnameData{}), - "shmall": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMALL)), - "shmmax": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMAX)), - "shmmni": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMNI)), - }), - "vm": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{}), - "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), - }), - "net": newSysNetDir(root, inoGen, k), - }) -} - -// newSysNetDir returns the dentry corresponding to /proc/sys/net directory. -func newSysNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { - var contents map[string]*kernfs.Dentry - - // TODO(gvisor.dev/issue/1833): Support for using the network stack in the - // network namespace of the calling process. - if stack := k.RootNetworkNamespace().Stack(); stack != nil { - contents = map[string]*kernfs.Dentry{ - "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "tcp_sack": newDentry(root, inoGen.NextIno(), 0644, &tcpSackData{stack: stack}), - - // The following files are simple stubs until they are implemented in - // netstack, most of these files are configuration related. We use the - // value closest to the actual netstack behavior or any empty file, all - // of these files will have mode 0444 (read-only for all users). - "ip_local_port_range": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("16000 65535")), - "ip_local_reserved_ports": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "ipfrag_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("30")), - "ip_nonlocal_bind": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "ip_no_pmtu_disc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - - // tcp_allowed_congestion_control tell the user what they are able to - // do as an unprivledged process so we leave it empty. - "tcp_allowed_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "tcp_available_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), - "tcp_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), - - // Many of the following stub files are features netstack doesn't - // support. The unsupported features return "0" to indicate they are - // disabled. - "tcp_base_mss": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1280")), - "tcp_dsack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_early_retrans": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fastopen": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fastopen_key": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "tcp_invalid_ratelimit": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_intvl": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_probes": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("7200")), - "tcp_mtu_probing": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_no_metrics_save": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_probe_interval": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_probe_threshold": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_retries1": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), - "tcp_retries2": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("15")), - "tcp_rfc1337": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_slow_start_after_idle": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_synack_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), - "tcp_syn_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), - "tcp_timestamps": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - }), - "core": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "default_qdisc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("pfifo_fast")), - "message_burst": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("10")), - "message_cost": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), - "optmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "rmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "rmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "somaxconn": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("128")), - "wmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "wmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - }), - } - } - - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "net": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents), - }) -} - -// mmapMinAddrData implements vfs.DynamicBytesSource for -// /proc/sys/vm/mmap_min_addr. -// -// +stateify savable -type mmapMinAddrData struct { - kernfs.DynamicBytesFile - - k *kernel.Kernel -} - -var _ dynamicInode = (*mmapMinAddrData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *mmapMinAddrData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d\n", d.k.Platform.MinUserAddress()) - return nil -} - -// hostnameData implements vfs.DynamicBytesSource for /proc/sys/kernel/hostname. -// -// +stateify savable -type hostnameData struct { - kernfs.DynamicBytesFile -} - -var _ dynamicInode = (*hostnameData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*hostnameData) Generate(ctx context.Context, buf *bytes.Buffer) error { - utsns := kernel.UTSNamespaceFromContext(ctx) - buf.WriteString(utsns.HostName()) - buf.WriteString("\n") - return nil -} - -// tcpSackData implements vfs.WritableDynamicBytesSource for -// /proc/sys/net/tcp_sack. -// -// +stateify savable -type tcpSackData struct { - kernfs.DynamicBytesFile - - stack inet.Stack `state:"wait"` - enabled *bool -} - -var _ vfs.WritableDynamicBytesSource = (*tcpSackData)(nil) - -// Generate implements vfs.DynamicBytesSource. -func (d *tcpSackData) Generate(ctx context.Context, buf *bytes.Buffer) error { - if d.enabled == nil { - sack, err := d.stack.TCPSACKEnabled() - if err != nil { - return err - } - d.enabled = &sack - } - - val := "0\n" - if *d.enabled { - // Technically, this is not quite compatible with Linux. Linux stores these - // as an integer, so if you write "2" into tcp_sack, you should get 2 back. - // Tough luck. - val = "1\n" - } - buf.WriteString(val) - return nil -} - -func (d *tcpSackData) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - if offset != 0 { - // No need to handle partial writes thus far. - return 0, syserror.EINVAL - } - if src.NumBytes() == 0 { - return 0, nil - } - - // Limit the amount of memory allocated. - src = src.TakeFirst(usermem.PageSize - 1) - - var v int32 - n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) - if err != nil { - return n, err - } - if d.enabled == nil { - d.enabled = new(bool) - } - *d.enabled = v != 0 - return n, d.stack.SetTCPSACKEnabled(*d.enabled) -} diff --git a/pkg/sentry/fsimpl/proc/tasks_sys_test.go b/pkg/sentry/fsimpl/proc/tasks_sys_test.go deleted file mode 100644 index be54897bb..000000000 --- a/pkg/sentry/fsimpl/proc/tasks_sys_test.go +++ /dev/null @@ -1,78 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/inet" -) - -func newIPv6TestStack() *inet.TestStack { - s := inet.NewTestStack() - s.SupportsIPv6Flag = true - return s -} - -func TestIfinet6NoAddresses(t *testing.T) { - n := &ifinet6{stack: newIPv6TestStack()} - var buf bytes.Buffer - n.Generate(contexttest.Context(t), &buf) - if buf.Len() > 0 { - t.Errorf("n.Generate() generated = %v, want = %v", buf.Bytes(), []byte{}) - } -} - -func TestIfinet6(t *testing.T) { - s := newIPv6TestStack() - s.InterfacesMap[1] = inet.Interface{Name: "eth0"} - s.InterfaceAddrsMap[1] = []inet.InterfaceAddr{ - { - Family: linux.AF_INET6, - PrefixLen: 128, - Addr: []byte("\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f"), - }, - } - s.InterfacesMap[2] = inet.Interface{Name: "eth1"} - s.InterfaceAddrsMap[2] = []inet.InterfaceAddr{ - { - Family: linux.AF_INET6, - PrefixLen: 128, - Addr: []byte("\x10\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f"), - }, - } - want := map[string]struct{}{ - "000102030405060708090a0b0c0d0e0f 01 80 00 00 eth0\n": {}, - "101112131415161718191a1b1c1d1e1f 02 80 00 00 eth1\n": {}, - } - - n := &ifinet6{stack: s} - contents := n.contents() - if len(contents) != len(want) { - t.Errorf("Got len(n.contents()) = %d, want = %d", len(contents), len(want)) - } - got := map[string]struct{}{} - for _, l := range contents { - got[l] = struct{}{} - } - - if !reflect.DeepEqual(got, want) { - t.Errorf("Got n.contents() = %v, want = %v", got, want) - } -} diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go deleted file mode 100644 index d0f97c137..000000000 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ /dev/null @@ -1,505 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "fmt" - "math" - "path" - "strconv" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -var ( - // Next offset 256 by convention. Adds 1 for the next offset. - selfLink = vfs.Dirent{Type: linux.DT_LNK, NextOff: 256 + 0 + 1} - threadSelfLink = vfs.Dirent{Type: linux.DT_LNK, NextOff: 256 + 1 + 1} - - // /proc/[pid] next offset starts at 256+2 (files above), then adds the - // PID, and adds 1 for the next offset. - proc1 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 1 + 1} - proc2 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 2 + 1} - proc3 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 3 + 1} -) - -var ( - tasksStaticFiles = map[string]testutil.DirentType{ - "cpuinfo": linux.DT_REG, - "filesystems": linux.DT_REG, - "loadavg": linux.DT_REG, - "meminfo": linux.DT_REG, - "mounts": linux.DT_LNK, - "net": linux.DT_LNK, - "self": linux.DT_LNK, - "stat": linux.DT_REG, - "sys": linux.DT_DIR, - "thread-self": linux.DT_LNK, - "uptime": linux.DT_REG, - "version": linux.DT_REG, - } - tasksStaticFilesNextOffs = map[string]int64{ - "self": selfLink.NextOff, - "thread-self": threadSelfLink.NextOff, - } - taskStaticFiles = map[string]testutil.DirentType{ - "auxv": linux.DT_REG, - "cgroup": linux.DT_REG, - "cmdline": linux.DT_REG, - "comm": linux.DT_REG, - "environ": linux.DT_REG, - "exe": linux.DT_LNK, - "fd": linux.DT_DIR, - "fdinfo": linux.DT_DIR, - "gid_map": linux.DT_REG, - "io": linux.DT_REG, - "maps": linux.DT_REG, - "mountinfo": linux.DT_REG, - "mounts": linux.DT_REG, - "net": linux.DT_DIR, - "ns": linux.DT_DIR, - "oom_score": linux.DT_REG, - "oom_score_adj": linux.DT_REG, - "smaps": linux.DT_REG, - "stat": linux.DT_REG, - "statm": linux.DT_REG, - "status": linux.DT_REG, - "task": linux.DT_DIR, - "uid_map": linux.DT_REG, - } -) - -func setup(t *testing.T) *testutil.System { - k, err := testutil.Boot() - if err != nil { - t.Fatalf("Error creating kernel: %v", err) - } - - ctx := k.SupervisorContext() - creds := auth.CredentialsFromContext(ctx) - - k.VFS().MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - - mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", tmpfs.Name, &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("NewMountNamespace(): %v", err) - } - pop := &vfs.PathOperation{ - Root: mntns.Root(), - Start: mntns.Root(), - Path: fspath.Parse("/proc"), - } - if err := k.VFS().MkdirAt(ctx, creds, pop, &vfs.MkdirOptions{Mode: 0777}); err != nil { - t.Fatalf("MkDir(/proc): %v", err) - } - - pop = &vfs.PathOperation{ - Root: mntns.Root(), - Start: mntns.Root(), - Path: fspath.Parse("/proc"), - } - mntOpts := &vfs.MountOptions{ - GetFilesystemOptions: vfs.GetFilesystemOptions{ - InternalData: &InternalData{ - Cgroups: map[string]string{ - "cpuset": "/foo/cpuset", - "memory": "/foo/memory", - }, - }, - }, - } - if err := k.VFS().MountAt(ctx, creds, "", pop, Name, mntOpts); err != nil { - t.Fatalf("MountAt(/proc): %v", err) - } - return testutil.NewSystem(ctx, t, k.VFS(), mntns) -} - -func TestTasksEmpty(t *testing.T) { - s := setup(t) - defer s.Destroy() - - collector := s.ListDirents(s.PathOpAtRoot("/proc")) - s.AssertAllDirentTypes(collector, tasksStaticFiles) - s.AssertDirentOffsets(collector, tasksStaticFilesNextOffs) -} - -func TestTasks(t *testing.T) { - s := setup(t) - defer s.Destroy() - - expectedDirents := make(map[string]testutil.DirentType) - for n, d := range tasksStaticFiles { - expectedDirents[n] = d - } - - k := kernel.KernelFromContext(s.Ctx) - var tasks []*kernel.Task - for i := 0; i < 5; i++ { - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root) - if err != nil { - t.Fatalf("CreateTask(): %v", err) - } - tasks = append(tasks, task) - expectedDirents[fmt.Sprintf("%d", i+1)] = linux.DT_DIR - } - - collector := s.ListDirents(s.PathOpAtRoot("/proc")) - s.AssertAllDirentTypes(collector, expectedDirents) - s.AssertDirentOffsets(collector, tasksStaticFilesNextOffs) - - lastPid := 0 - dirents := collector.OrderedDirents() - doneSkippingNonTaskDirs := false - for _, d := range dirents { - pid, err := strconv.Atoi(d.Name) - if err != nil { - if !doneSkippingNonTaskDirs { - // We haven't gotten to the task dirs yet. - continue - } - t.Fatalf("Invalid process directory %q", d.Name) - } - doneSkippingNonTaskDirs = true - if lastPid > pid { - t.Errorf("pids not in order: %v", dirents) - } - found := false - for _, t := range tasks { - if k.TaskSet().Root.IDOfTask(t) == kernel.ThreadID(pid) { - found = true - } - } - if !found { - t.Errorf("Additional task ID %d listed: %v", pid, tasks) - } - // Next offset starts at 256+2 ('self' and 'thread-self'), then adds the - // PID, and adds 1 for the next offset. - if want := int64(256 + 2 + pid + 1); d.NextOff != want { - t.Errorf("Wrong dirent offset want: %d got: %d: %+v", want, d.NextOff, d) - } - } - if !doneSkippingNonTaskDirs { - t.Fatalf("Never found any process directories.") - } - - // Test lookup. - for _, path := range []string{"/proc/1", "/proc/2"} { - fd, err := s.VFS.OpenAt( - s.Ctx, - s.Creds, - s.PathOpAtRoot(path), - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(%q) failed: %v", path, err) - } - defer fd.DecRef() - buf := make([]byte, 1) - bufIOSeq := usermem.BytesIOSequence(buf) - if _, err := fd.Read(s.Ctx, bufIOSeq, vfs.ReadOptions{}); err != syserror.EISDIR { - t.Errorf("wrong error reading directory: %v", err) - } - } - - if _, err := s.VFS.OpenAt( - s.Ctx, - s.Creds, - s.PathOpAtRoot("/proc/9999"), - &vfs.OpenOptions{}, - ); err != syserror.ENOENT { - t.Fatalf("wrong error from vfsfs.OpenAt(/proc/9999): %v", err) - } -} - -func TestTasksOffset(t *testing.T) { - s := setup(t) - defer s.Destroy() - - k := kernel.KernelFromContext(s.Ctx) - for i := 0; i < 3; i++ { - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - if _, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root); err != nil { - t.Fatalf("CreateTask(): %v", err) - } - } - - for _, tc := range []struct { - name string - offset int64 - wants map[string]vfs.Dirent - }{ - { - name: "small offset", - offset: 100, - wants: map[string]vfs.Dirent{ - "self": selfLink, - "thread-self": threadSelfLink, - "1": proc1, - "2": proc2, - "3": proc3, - }, - }, - { - name: "offset at start", - offset: 256, - wants: map[string]vfs.Dirent{ - "self": selfLink, - "thread-self": threadSelfLink, - "1": proc1, - "2": proc2, - "3": proc3, - }, - }, - { - name: "skip /proc/self", - offset: 257, - wants: map[string]vfs.Dirent{ - "thread-self": threadSelfLink, - "1": proc1, - "2": proc2, - "3": proc3, - }, - }, - { - name: "skip symlinks", - offset: 258, - wants: map[string]vfs.Dirent{ - "1": proc1, - "2": proc2, - "3": proc3, - }, - }, - { - name: "skip first process", - offset: 260, - wants: map[string]vfs.Dirent{ - "2": proc2, - "3": proc3, - }, - }, - { - name: "last process", - offset: 261, - wants: map[string]vfs.Dirent{ - "3": proc3, - }, - }, - { - name: "after last", - offset: 262, - wants: nil, - }, - { - name: "TaskLimit+1", - offset: kernel.TasksLimit + 1, - wants: nil, - }, - { - name: "max", - offset: math.MaxInt64, - wants: nil, - }, - } { - t.Run(tc.name, func(t *testing.T) { - s := s.WithSubtest(t) - fd, err := s.VFS.OpenAt( - s.Ctx, - s.Creds, - s.PathOpAtRoot("/proc"), - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) - } - defer fd.DecRef() - if _, err := fd.Seek(s.Ctx, tc.offset, linux.SEEK_SET); err != nil { - t.Fatalf("Seek(%d, SEEK_SET): %v", tc.offset, err) - } - - var collector testutil.DirentCollector - if err := fd.IterDirents(s.Ctx, &collector); err != nil { - t.Fatalf("IterDirent(): %v", err) - } - - expectedTypes := make(map[string]testutil.DirentType) - expectedOffsets := make(map[string]int64) - for name, want := range tc.wants { - expectedTypes[name] = want.Type - if want.NextOff != 0 { - expectedOffsets[name] = want.NextOff - } - } - - collector.SkipDotsChecks(true) // We seek()ed past the dots. - s.AssertAllDirentTypes(&collector, expectedTypes) - s.AssertDirentOffsets(&collector, expectedOffsets) - }) - } -} - -func TestTask(t *testing.T) { - s := setup(t) - defer s.Destroy() - - k := kernel.KernelFromContext(s.Ctx) - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - _, err := testutil.CreateTask(s.Ctx, "name", tc, s.MntNs, s.Root, s.Root) - if err != nil { - t.Fatalf("CreateTask(): %v", err) - } - - collector := s.ListDirents(s.PathOpAtRoot("/proc/1")) - s.AssertAllDirentTypes(collector, taskStaticFiles) -} - -func TestProcSelf(t *testing.T) { - s := setup(t) - defer s.Destroy() - - k := kernel.KernelFromContext(s.Ctx) - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, "name", tc, s.MntNs, s.Root, s.Root) - if err != nil { - t.Fatalf("CreateTask(): %v", err) - } - - collector := s.WithTemporaryContext(task).ListDirents(&vfs.PathOperation{ - Root: s.Root, - Start: s.Root, - Path: fspath.Parse("/proc/self/"), - FollowFinalSymlink: true, - }) - s.AssertAllDirentTypes(collector, taskStaticFiles) -} - -func iterateDir(ctx context.Context, t *testing.T, s *testutil.System, fd *vfs.FileDescription) { - t.Logf("Iterating: %s", fd.MappedName(ctx)) - - var collector testutil.DirentCollector - if err := fd.IterDirents(ctx, &collector); err != nil { - t.Fatalf("IterDirents(): %v", err) - } - if err := collector.Contains(".", linux.DT_DIR); err != nil { - t.Error(err.Error()) - } - if err := collector.Contains("..", linux.DT_DIR); err != nil { - t.Error(err.Error()) - } - - for _, d := range collector.Dirents() { - if d.Name == "." || d.Name == ".." { - continue - } - childPath := path.Join(fd.MappedName(ctx), d.Name) - if d.Type == linux.DT_LNK { - link, err := s.VFS.ReadlinkAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(childPath)}, - ) - if err != nil { - t.Errorf("vfsfs.ReadlinkAt(%v) failed: %v", childPath, err) - } else { - t.Logf("Skipping symlink: /proc%s => %s", childPath, link) - } - continue - } - - t.Logf("Opening: /proc%s", childPath) - child, err := s.VFS.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(childPath)}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Errorf("vfsfs.OpenAt(%v) failed: %v", childPath, err) - continue - } - defer child.DecRef() - stat, err := child.Stat(ctx, vfs.StatOptions{}) - if err != nil { - t.Errorf("Stat(%v) failed: %v", childPath, err) - } - if got := linux.FileMode(stat.Mode).DirentType(); got != d.Type { - t.Errorf("wrong file mode, stat: %v, dirent: %v", got, d.Type) - } - if d.Type == linux.DT_DIR { - // Found another dir, let's do it again! - iterateDir(ctx, t, s, child) - } - } -} - -// TestTree iterates all directories and stats every file. -func TestTree(t *testing.T) { - s := setup(t) - defer s.Destroy() - - k := kernel.KernelFromContext(s.Ctx) - - pop := &vfs.PathOperation{ - Root: s.Root, - Start: s.Root, - Path: fspath.Parse("test-file"), - } - opts := &vfs.OpenOptions{ - Flags: linux.O_RDONLY | linux.O_CREAT, - Mode: 0777, - } - file, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, opts) - if err != nil { - t.Fatalf("failed to create test file: %v", err) - } - defer file.DecRef() - - var tasks []*kernel.Task - for i := 0; i < 5; i++ { - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root) - if err != nil { - t.Fatalf("CreateTask(): %v", err) - } - // Add file to populate /proc/[pid]/fd and fdinfo directories. - task.FDTable().NewFDVFS2(task, 0, file, kernel.FDFlags{}) - tasks = append(tasks, task) - } - - ctx := tasks[0] - fd, err := s.VFS.OpenAt( - ctx, - auth.CredentialsFromContext(s.Ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse("/proc")}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(/proc) failed: %v", err) - } - iterateDir(ctx, t, s, fd) - fd.DecRef() -} diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD deleted file mode 100644 index a741e2bb6..000000000 --- a/pkg/sentry/fsimpl/sys/BUILD +++ /dev/null @@ -1,34 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -licenses(["notice"]) - -go_library( - name = "sys", - srcs = [ - "sys.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/fsimpl/kernfs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) - -go_test( - name = "sys_test", - srcs = ["sys_test.go"], - deps = [ - ":sys", - "//pkg/abi/linux", - "//pkg/sentry/fsimpl/testutil", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "@com_github_google_go-cmp//cmp:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go deleted file mode 100644 index 7abfd62f2..000000000 --- a/pkg/sentry/fsimpl/sys/sys.go +++ /dev/null @@ -1,129 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package sys implements sysfs. -package sys - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Name is the default filesystem name. -const Name = "sysfs" - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - kernfs.Filesystem -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fs := &filesystem{} - fs.Filesystem.Init(vfsObj) - k := kernel.KernelFromContext(ctx) - maxCPUCores := k.ApplicationCores() - defaultSysDirMode := linux.FileMode(0755) - - root := fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "block": fs.newDir(creds, defaultSysDirMode, nil), - "bus": fs.newDir(creds, defaultSysDirMode, nil), - "class": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "power_supply": fs.newDir(creds, defaultSysDirMode, nil), - }), - "dev": fs.newDir(creds, defaultSysDirMode, nil), - "devices": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "system": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "cpu": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "online": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - "possible": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - "present": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - }), - }), - }), - "firmware": fs.newDir(creds, defaultSysDirMode, nil), - "fs": fs.newDir(creds, defaultSysDirMode, nil), - "kernel": fs.newDir(creds, defaultSysDirMode, nil), - "module": fs.newDir(creds, defaultSysDirMode, nil), - "power": fs.newDir(creds, defaultSysDirMode, nil), - }) - return fs.VFSFilesystem(), root.VFSDentry(), nil -} - -// dir implements kernfs.Inode. -type dir struct { - kernfs.InodeAttrs - kernfs.InodeNoDynamicLookup - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - - kernfs.OrderedChildren - dentry kernfs.Dentry -} - -func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { - d := &dir{} - d.InodeAttrs.Init(creds, fs.NextIno(), linux.ModeDirectory|0755) - d.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - d.dentry.Init(d) - - d.IncLinks(d.OrderedChildren.Populate(&d.dentry, contents)) - - return &d.dentry -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -// Open implements kernfs.Inode.Open. -func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts); err != nil { - return nil, err - } - return fd.VFSFileDescription(), nil -} - -// cpuFile implements kernfs.Inode. -type cpuFile struct { - kernfs.DynamicBytesFile - maxCores uint -} - -// Generate implements vfs.DynamicBytesSource.Generate. -func (c *cpuFile) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "0-%d", c.maxCores-1) - return nil -} - -func (fs *filesystem) newCPUFile(creds *auth.Credentials, maxCores uint, mode linux.FileMode) *kernfs.Dentry { - c := &cpuFile{maxCores: maxCores} - c.DynamicBytesFile.Init(creds, fs.NextIno(), c, mode) - d := &kernfs.Dentry{} - d.Init(c) - return d -} diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go deleted file mode 100644 index 4b3602d47..000000000 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ /dev/null @@ -1,89 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package sys_test - -import ( - "fmt" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func newTestSystem(t *testing.T) *testutil.System { - k, err := testutil.Boot() - if err != nil { - t.Fatalf("Failed to create test kernel: %v", err) - } - ctx := k.SupervisorContext() - creds := auth.CredentialsFromContext(ctx) - k.VFS().MustRegisterFilesystemType(sys.Name, sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - - mns, err := k.VFS().NewMountNamespace(ctx, creds, "", sys.Name, &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("Failed to create new mount namespace: %v", err) - } - return testutil.NewSystem(ctx, t, k.VFS(), mns) -} - -func TestReadCPUFile(t *testing.T) { - s := newTestSystem(t) - defer s.Destroy() - k := kernel.KernelFromContext(s.Ctx) - maxCPUCores := k.ApplicationCores() - - expected := fmt.Sprintf("0-%d", maxCPUCores-1) - - for _, fname := range []string{"online", "possible", "present"} { - pop := s.PathOpAtRoot(fmt.Sprintf("devices/system/cpu/%s", fname)) - fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, &vfs.OpenOptions{}) - if err != nil { - t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) - } - defer fd.DecRef() - content, err := s.ReadToEnd(fd) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - if diff := cmp.Diff(expected, content); diff != "" { - t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) - } - } -} - -func TestSysRootContainsExpectedEntries(t *testing.T) { - s := newTestSystem(t) - defer s.Destroy() - pop := s.PathOpAtRoot("/") - s.AssertAllDirentTypes(s.ListDirents(pop), map[string]testutil.DirentType{ - "block": linux.DT_DIR, - "bus": linux.DT_DIR, - "class": linux.DT_DIR, - "dev": linux.DT_DIR, - "devices": linux.DT_DIR, - "firmware": linux.DT_DIR, - "fs": linux.DT_DIR, - "kernel": linux.DT_DIR, - "module": linux.DT_DIR, - "power": linux.DT_DIR, - }) -} diff --git a/pkg/sentry/fsimpl/testutil/BUILD b/pkg/sentry/fsimpl/testutil/BUILD deleted file mode 100644 index 0e4053a46..000000000 --- a/pkg/sentry/fsimpl/testutil/BUILD +++ /dev/null @@ -1,37 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "testutil", - testonly = 1, - srcs = [ - "kernel.go", - "testutil.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/cpuid", - "//pkg/fspath", - "//pkg/memutil", - "//pkg/sentry/fsbridge", - "//pkg/sentry/fsimpl/tmpfs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/sched", - "//pkg/sentry/limits", - "//pkg/sentry/loader", - "//pkg/sentry/mm", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/platform/kvm", - "//pkg/sentry/platform/ptrace", - "//pkg/sentry/time", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/usermem", - "@com_github_google_go-cmp//cmp:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go deleted file mode 100644 index c16a36cdb..000000000 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ /dev/null @@ -1,180 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package testutil - -import ( - "flag" - "fmt" - "os" - "runtime" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/cpuid" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/memutil" - "gvisor.dev/gvisor/pkg/sentry/fsbridge" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/sched" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/loader" - "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/time" - "gvisor.dev/gvisor/pkg/sentry/vfs" - - // Platforms are plugable. - _ "gvisor.dev/gvisor/pkg/sentry/platform/kvm" - _ "gvisor.dev/gvisor/pkg/sentry/platform/ptrace" -) - -var ( - platformFlag = flag.String("platform", "ptrace", "specify which platform to use") -) - -// Boot initializes a new bare bones kernel for test. -func Boot() (*kernel.Kernel, error) { - platformCtr, err := platform.Lookup(*platformFlag) - if err != nil { - return nil, fmt.Errorf("platform not found: %v", err) - } - deviceFile, err := platformCtr.OpenDevice() - if err != nil { - return nil, fmt.Errorf("creating platform: %v", err) - } - plat, err := platformCtr.New(deviceFile) - if err != nil { - return nil, fmt.Errorf("creating platform: %v", err) - } - - k := &kernel.Kernel{ - Platform: plat, - } - - mf, err := createMemoryFile() - if err != nil { - return nil, err - } - k.SetMemoryFile(mf) - - // Pass k as the platform since it is savable, unlike the actual platform. - vdso, err := loader.PrepareVDSO(nil, k) - if err != nil { - return nil, fmt.Errorf("creating vdso: %v", err) - } - - // Create timekeeper. - tk, err := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange()) - if err != nil { - return nil, fmt.Errorf("creating timekeeper: %v", err) - } - tk.SetClocks(time.NewCalibratedClocks()) - - creds := auth.NewRootCredentials(auth.NewRootUserNamespace()) - - // Initiate the Kernel object, which is required by the Context passed - // to createVFS in order to mount (among other things) procfs. - if err = k.Init(kernel.InitKernelArgs{ - ApplicationCores: uint(runtime.GOMAXPROCS(-1)), - FeatureSet: cpuid.HostFeatureSet(), - Timekeeper: tk, - RootUserNamespace: creds.UserNamespace, - Vdso: vdso, - RootUTSNamespace: kernel.NewUTSNamespace("hostname", "domain", creds.UserNamespace), - RootIPCNamespace: kernel.NewIPCNamespace(creds.UserNamespace), - RootAbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), - PIDNamespace: kernel.NewRootPIDNamespace(creds.UserNamespace), - }); err != nil { - return nil, fmt.Errorf("initializing kernel: %v", err) - } - - kernel.VFS2Enabled = true - - if err := k.VFS().Init(); err != nil { - return nil, fmt.Errorf("VFS init: %v", err) - } - k.VFS().MustRegisterFilesystemType(tmpfs.Name, &tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - AllowUserList: true, - }) - - ls, err := limits.NewLinuxLimitSet() - if err != nil { - return nil, err - } - tg := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) - k.TestOnly_SetGlobalInit(tg) - - return k, nil -} - -// CreateTask creates a new bare bones task for tests. -func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns *vfs.MountNamespace, root, cwd vfs.VirtualDentry) (*kernel.Task, error) { - k := kernel.KernelFromContext(ctx) - exe, err := newFakeExecutable(ctx, k.VFS(), auth.CredentialsFromContext(ctx), root) - if err != nil { - return nil, err - } - m := mm.NewMemoryManager(k, k, k.SleepForAddressSpaceActivation) - m.SetExecutable(fsbridge.NewVFSFile(exe)) - - config := &kernel.TaskConfig{ - Kernel: k, - ThreadGroup: tc, - TaskContext: &kernel.TaskContext{Name: name, MemoryManager: m}, - Credentials: auth.CredentialsFromContext(ctx), - NetworkNamespace: k.RootNetworkNamespace(), - AllowedCPUMask: sched.NewFullCPUSet(k.ApplicationCores()), - UTSNamespace: kernel.UTSNamespaceFromContext(ctx), - IPCNamespace: kernel.IPCNamespaceFromContext(ctx), - AbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), - MountNamespaceVFS2: mntns, - FSContext: kernel.NewFSContextVFS2(root, cwd, 0022), - FDTable: k.NewFDTable(), - } - return k.TaskSet().NewTask(config) -} - -func newFakeExecutable(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, root vfs.VirtualDentry) (*vfs.FileDescription, error) { - const name = "executable" - pop := &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(name), - } - opts := &vfs.OpenOptions{ - Flags: linux.O_RDONLY | linux.O_CREAT, - Mode: 0777, - } - return vfsObj.OpenAt(ctx, creds, pop, opts) -} - -func createMemoryFile() (*pgalloc.MemoryFile, error) { - const memfileName = "test-memory" - memfd, err := memutil.CreateMemFD(memfileName, 0) - if err != nil { - return nil, fmt.Errorf("error creating memfd: %v", err) - } - memfile := os.NewFile(uintptr(memfd), memfileName) - mf, err := pgalloc.NewMemoryFile(memfile, pgalloc.MemoryFileOpts{}) - if err != nil { - memfile.Close() - return nil, fmt.Errorf("error creating pgalloc.MemoryFile: %v", err) - } - return mf, nil -} diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go deleted file mode 100644 index e16808c63..000000000 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ /dev/null @@ -1,281 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package testutil provides common test utilities for kernfs-based -// filesystems. -package testutil - -import ( - "fmt" - "io" - "strings" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/usermem" -) - -// System represents the context for a single test. -// -// Test systems must be explicitly destroyed with System.Destroy. -type System struct { - t *testing.T - Ctx context.Context - Creds *auth.Credentials - VFS *vfs.VirtualFilesystem - Root vfs.VirtualDentry - MntNs *vfs.MountNamespace -} - -// NewSystem constructs a System. -// -// Precondition: Caller must hold a reference on MntNs, whose ownership -// is transferred to the new System. -func NewSystem(ctx context.Context, t *testing.T, v *vfs.VirtualFilesystem, mns *vfs.MountNamespace) *System { - s := &System{ - t: t, - Ctx: ctx, - Creds: auth.CredentialsFromContext(ctx), - VFS: v, - MntNs: mns, - Root: mns.Root(), - } - return s -} - -// WithSubtest creates a temporary test system with a new test harness, -// referencing all other resources from the original system. This is useful when -// a system is reused for multiple subtests, and the T needs to change for each -// case. Note that this is safe when test cases run in parallel, as all -// resources referenced by the system are immutable, or handle interior -// mutations in a thread-safe manner. -// -// The returned system must not outlive the original and should not be destroyed -// via System.Destroy. -func (s *System) WithSubtest(t *testing.T) *System { - return &System{ - t: t, - Ctx: s.Ctx, - Creds: s.Creds, - VFS: s.VFS, - MntNs: s.MntNs, - Root: s.Root, - } -} - -// WithTemporaryContext constructs a temporary test system with a new context -// ctx. The temporary system borrows all resources and references from the -// original system. The returned temporary system must not outlive the original -// system, and should not be destroyed via System.Destroy. -func (s *System) WithTemporaryContext(ctx context.Context) *System { - return &System{ - t: s.t, - Ctx: ctx, - Creds: s.Creds, - VFS: s.VFS, - MntNs: s.MntNs, - Root: s.Root, - } -} - -// Destroy release resources associated with a test system. -func (s *System) Destroy() { - s.Root.DecRef() - s.MntNs.DecRef() // Reference on MntNs passed to NewSystem. -} - -// ReadToEnd reads the contents of fd until EOF to a string. -func (s *System) ReadToEnd(fd *vfs.FileDescription) (string, error) { - buf := make([]byte, usermem.PageSize) - bufIOSeq := usermem.BytesIOSequence(buf) - opts := vfs.ReadOptions{} - - var content strings.Builder - for { - n, err := fd.Read(s.Ctx, bufIOSeq, opts) - if n == 0 || err != nil { - if err == io.EOF { - err = nil - } - return content.String(), err - } - content.Write(buf[:n]) - } -} - -// PathOpAtRoot constructs a PathOperation with the given path from -// the root of the filesystem. -func (s *System) PathOpAtRoot(path string) *vfs.PathOperation { - return &vfs.PathOperation{ - Root: s.Root, - Start: s.Root, - Path: fspath.Parse(path), - } -} - -// GetDentryOrDie attempts to resolve a dentry referred to by the -// provided path operation. If unsuccessful, the test fails. -func (s *System) GetDentryOrDie(pop *vfs.PathOperation) vfs.VirtualDentry { - vd, err := s.VFS.GetDentryAt(s.Ctx, s.Creds, pop, &vfs.GetDentryOptions{}) - if err != nil { - s.t.Fatalf("GetDentryAt(pop:%+v) failed: %v", pop, err) - } - return vd -} - -// DirentType is an alias for values for linux_dirent64.d_type. -type DirentType = uint8 - -// ListDirents lists the Dirents for a directory at pop. -func (s *System) ListDirents(pop *vfs.PathOperation) *DirentCollector { - fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, &vfs.OpenOptions{Flags: linux.O_RDONLY}) - if err != nil { - s.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) - } - defer fd.DecRef() - - collector := &DirentCollector{} - if err := fd.IterDirents(s.Ctx, collector); err != nil { - s.t.Fatalf("IterDirent failed: %v", err) - } - return collector -} - -// AssertAllDirentTypes verifies that the set of dirents in collector contains -// exactly the specified set of expected entries. AssertAllDirentTypes respects -// collector.skipDots, and implicitly checks for "." and ".." accordingly. -func (s *System) AssertAllDirentTypes(collector *DirentCollector, expected map[string]DirentType) { - // Also implicitly check for "." and "..", if enabled. - if !collector.skipDots { - expected["."] = linux.DT_DIR - expected[".."] = linux.DT_DIR - } - - dentryTypes := make(map[string]DirentType) - collector.mu.Lock() - for _, dirent := range collector.dirents { - dentryTypes[dirent.Name] = dirent.Type - } - collector.mu.Unlock() - if diff := cmp.Diff(expected, dentryTypes); diff != "" { - s.t.Fatalf("IterDirent had unexpected results:\n--- want\n+++ got\n%v", diff) - } -} - -// AssertDirentOffsets verifies that collector contains at least the entries -// specified in expected, with the given NextOff field. Entries specified in -// expected but missing from collector result in failure. Extra entries in -// collector are ignored. AssertDirentOffsets respects collector.skipDots, and -// implicitly checks for "." and ".." accordingly. -func (s *System) AssertDirentOffsets(collector *DirentCollector, expected map[string]int64) { - // Also implicitly check for "." and "..", if enabled. - if !collector.skipDots { - expected["."] = 1 - expected[".."] = 2 - } - - dentryNextOffs := make(map[string]int64) - collector.mu.Lock() - for _, dirent := range collector.dirents { - // Ignore extra entries in dentries that are not in expected. - if _, ok := expected[dirent.Name]; ok { - dentryNextOffs[dirent.Name] = dirent.NextOff - } - } - collector.mu.Unlock() - if diff := cmp.Diff(expected, dentryNextOffs); diff != "" { - s.t.Fatalf("IterDirent had unexpected results:\n--- want\n+++ got\n%v", diff) - } -} - -// DirentCollector provides an implementation for vfs.IterDirentsCallback for -// testing. It simply iterates to the end of a given directory FD and collects -// all dirents emitted by the callback. -type DirentCollector struct { - mu sync.Mutex - order []*vfs.Dirent - dirents map[string]*vfs.Dirent - // When the collector is used in various Assert* functions, should "." and - // ".." be implicitly checked? - skipDots bool -} - -// SkipDotsChecks enables or disables the implicit checks on "." and ".." when -// the collector is used in various Assert* functions. Note that "." and ".." -// are still collected if passed to d.Handle, so the caller should only disable -// the checks when they aren't expected. -func (d *DirentCollector) SkipDotsChecks(value bool) { - d.skipDots = value -} - -// Handle implements vfs.IterDirentsCallback.Handle. -func (d *DirentCollector) Handle(dirent vfs.Dirent) error { - d.mu.Lock() - if d.dirents == nil { - d.dirents = make(map[string]*vfs.Dirent) - } - d.order = append(d.order, &dirent) - d.dirents[dirent.Name] = &dirent - d.mu.Unlock() - return nil -} - -// Count returns the number of dirents currently in the collector. -func (d *DirentCollector) Count() int { - d.mu.Lock() - defer d.mu.Unlock() - return len(d.dirents) -} - -// Contains checks whether the collector has a dirent with the given name and -// type. -func (d *DirentCollector) Contains(name string, typ uint8) error { - d.mu.Lock() - defer d.mu.Unlock() - dirent, ok := d.dirents[name] - if !ok { - return fmt.Errorf("No dirent named %q found", name) - } - if dirent.Type != typ { - return fmt.Errorf("Dirent named %q found, but was expecting type %s, got: %+v", name, linux.DirentType.Parse(uint64(typ)), dirent) - } - return nil -} - -// Dirents returns all dirents discovered by this collector. -func (d *DirentCollector) Dirents() map[string]*vfs.Dirent { - d.mu.Lock() - dirents := make(map[string]*vfs.Dirent) - for n, d := range d.dirents { - dirents[n] = d - } - d.mu.Unlock() - return dirents -} - -// OrderedDirents returns an ordered list of dirents as discovered by this -// collector. -func (d *DirentCollector) OrderedDirents() []*vfs.Dirent { - d.mu.Lock() - dirents := make([]*vfs.Dirent, len(d.order)) - copy(dirents, d.order) - d.mu.Unlock() - return dirents -} diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD deleted file mode 100644 index 57abd5583..000000000 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ /dev/null @@ -1,98 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -licenses(["notice"]) - -go_template_instance( - name = "dentry_list", - out = "dentry_list.go", - package = "tmpfs", - prefix = "dentry", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*dentry", - "Linker": "*dentry", - }, -) - -go_library( - name = "tmpfs", - srcs = [ - "dentry_list.go", - "device_file.go", - "directory.go", - "filesystem.go", - "named_pipe.go", - "regular_file.go", - "symlink.go", - "tmpfs.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/amutex", - "//pkg/context", - "//pkg/fspath", - "//pkg/log", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/lock", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/pipe", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -go_test( - name = "benchmark_test", - size = "small", - srcs = ["benchmark_test.go"], - deps = [ - ":tmpfs", - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/refs", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/tmpfs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) - -go_test( - name = "tmpfs_test", - size = "small", - srcs = [ - "pipe_test.go", - "regular_file_test.go", - "stat_test.go", - ], - library = ":tmpfs", - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs/lock", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/contexttest", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go deleted file mode 100644 index 383133e44..000000000 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ /dev/null @@ -1,493 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package benchmark_test - -import ( - "fmt" - "runtime" - "strings" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Differences from stat_benchmark: -// -// - Syscall interception, CopyInPath, copyOutStat, and overlayfs overheads are -// not included. -// -// - *MountStat benchmarks use a tmpfs root mount and a tmpfs submount at /tmp. -// Non-MountStat benchmarks use a tmpfs root mount and no submounts. -// stat_benchmark uses a varying root mount, a tmpfs submount at /tmp, and a -// subdirectory /tmp/<top_dir> (assuming TEST_TMPDIR == "/tmp"). Thus -// stat_benchmark at depth 1 does a comparable amount of work to *MountStat -// benchmarks at depth 2, and non-MountStat benchmarks at depth 3. -var depths = []int{1, 2, 3, 8, 64, 100} - -const ( - mountPointName = "tmp" - filename = "gvisor_test_temp_0_1557494568" -) - -// This is copied from syscalls/linux/sys_file.go, with the dependency on -// kernel.Task stripped out. -func fileOpOn(ctx context.Context, mntns *fs.MountNamespace, root, wd *fs.Dirent, dirFD int32, path string, resolve bool, fn func(root *fs.Dirent, d *fs.Dirent) error) error { - var ( - d *fs.Dirent // The file. - rel *fs.Dirent // The relative directory for search (if required.) - err error - ) - - // Extract the working directory (maybe). - if len(path) > 0 && path[0] == '/' { - // Absolute path; rel can be nil. - } else if dirFD == linux.AT_FDCWD { - // Need to reference the working directory. - rel = wd - } else { - // Need to extract the given FD. - return syserror.EBADF - } - - // Lookup the node. - remainingTraversals := uint(linux.MaxSymlinkTraversals) - if resolve { - d, err = mntns.FindInode(ctx, root, rel, path, &remainingTraversals) - } else { - d, err = mntns.FindLink(ctx, root, rel, path, &remainingTraversals) - } - if err != nil { - return err - } - - err = fn(root, d) - d.DecRef() - return err -} - -func BenchmarkVFS1TmpfsStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - - // Create VFS. - tmpfsFS, ok := fs.FindFilesystem("tmpfs") - if !ok { - b.Fatalf("failed to find tmpfs filesystem type") - } - rootInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - mntns, err := fs.NewMountNamespace(ctx, rootInode) - if err != nil { - b.Fatalf("failed to create mount namespace: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - root := mntns.Root() - defer root.DecRef() - d := root - d.IncRef() - defer d.DecRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - next, err := d.Walk(ctx, root, name) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - d.DecRef() - d = next - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - file, err := d.Inode.Create(ctx, d, filename, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0644)) - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - file.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - dirPath := false - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - err := fileOpOn(ctx, mntns, root, root, linux.AT_FDCWD, filePath, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent) error { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR - } - uattr, err := d.Inode.UnstableAttr(ctx) - if err != nil { - return err - } - // Sanity check. - if uattr.Perms.User.Execute { - b.Fatalf("got wrong permissions (%0o)", uattr.Perms.LinuxMode()) - } - return nil - }) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS2MemfsStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - b.Fatalf("VFS init: %v", err) - } - vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - root := mntns.Root() - defer root.DecRef() - vd := root - vd.IncRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - pop := vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(name), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - nextVD, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - vd.DecRef() - vd = nextVD - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(filename), - FollowFinalSymlink: true, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: 0644, - }) - vd.DecRef() - vd = vfs.VirtualDentry{} - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - defer fd.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. - if stat.Mode&^linux.S_IFMT != 0644 { - b.Fatalf("got wrong permissions (%0o)", stat.Mode) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS1TmpfsMountStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - - // Create VFS. - tmpfsFS, ok := fs.FindFilesystem("tmpfs") - if !ok { - b.Fatalf("failed to find tmpfs filesystem type") - } - rootInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - mntns, err := fs.NewMountNamespace(ctx, rootInode) - if err != nil { - b.Fatalf("failed to create mount namespace: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create and mount the submount. - root := mntns.Root() - defer root.DecRef() - if err := root.Inode.CreateDirectory(ctx, root, mountPointName, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create mount point: %v", err) - } - mountPoint, err := root.Walk(ctx, root, mountPointName) - if err != nil { - b.Fatalf("failed to walk to mount point: %v", err) - } - defer mountPoint.DecRef() - submountInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs submount: %v", err) - } - if err := mntns.Mount(ctx, mountPoint, submountInode); err != nil { - b.Fatalf("failed to mount tmpfs submount: %v", err) - } - filePathBuilder.WriteString(mountPointName) - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - d, err := root.Walk(ctx, root, mountPointName) - if err != nil { - b.Fatalf("failed to walk to mount root: %v", err) - } - defer d.DecRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - next, err := d.Walk(ctx, root, name) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - d.DecRef() - d = next - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - file, err := d.Inode.Create(ctx, d, filename, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0644)) - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - file.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - dirPath := false - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - err := fileOpOn(ctx, mntns, root, root, linux.AT_FDCWD, filePath, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent) error { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR - } - uattr, err := d.Inode.UnstableAttr(ctx) - if err != nil { - return err - } - // Sanity check. - if uattr.Perms.User.Execute { - b.Fatalf("got wrong permissions (%0o)", uattr.Perms.LinuxMode()) - } - return nil - }) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS2MemfsMountStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - b.Fatalf("VFS init: %v", err) - } - vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create the mount point. - root := mntns.Root() - defer root.DecRef() - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(mountPointName), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create mount point: %v", err) - } - // Save the mount point for later use. - mountPoint, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to mount point: %v", err) - } - defer mountPoint.DecRef() - // Create and mount the submount. - if err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil { - b.Fatalf("failed to mount tmpfs submount: %v", err) - } - filePathBuilder.WriteString(mountPointName) - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - vd, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to mount root: %v", err) - } - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - pop := vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(name), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - nextVD, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - vd.DecRef() - vd = nextVD - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Verify that we didn't create any directories under the mount - // point (i.e. they were all created on the submount). - firstDirName := fmt.Sprintf("%d", depth) - if child := mountPoint.Dentry().Child(firstDirName); child != nil { - b.Fatalf("created directory %q under root mount, not submount", firstDirName) - } - - // Create the file that will be stat'd. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(filename), - FollowFinalSymlink: true, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: 0644, - }) - vd.DecRef() - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - fd.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. - if stat.Mode&^linux.S_IFMT != 0644 { - b.Fatalf("got wrong permissions (%0o)", stat.Mode) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func init() { - // Turn off reference leak checking for a fair comparison between vfs1 and - // vfs2. - refs.SetLeakMode(refs.NoLeakChecking) -} diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go deleted file mode 100644 index 83bf885ee..000000000 --- a/pkg/sentry/fsimpl/tmpfs/device_file.go +++ /dev/null @@ -1,49 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -type deviceFile struct { - inode inode - kind vfs.DeviceKind - major uint32 - minor uint32 -} - -func (fs *filesystem) newDeviceFile(creds *auth.Credentials, mode linux.FileMode, kind vfs.DeviceKind, major, minor uint32) *inode { - file := &deviceFile{ - kind: kind, - major: major, - minor: minor, - } - switch kind { - case vfs.BlockDevice: - mode |= linux.S_IFBLK - case vfs.CharDevice: - mode |= linux.S_IFCHR - default: - panic(fmt.Sprintf("invalid DeviceKind: %v", kind)) - } - file.inode.init(file, fs, creds, mode) - file.inode.nlink = 1 // from parent directory - return &file.inode -} diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go deleted file mode 100644 index 37c75ab64..000000000 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ /dev/null @@ -1,182 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -type directory struct { - inode inode - - // childList is a list containing (1) child Dentries and (2) fake Dentries - // (with inode == nil) that represent the iteration position of - // directoryFDs. childList is used to support directoryFD.IterDirents() - // efficiently. childList is protected by filesystem.mu. - childList dentryList -} - -func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *inode { - dir := &directory{} - dir.inode.init(dir, fs, creds, linux.S_IFDIR|mode) - dir.inode.nlink = 2 // from "." and parent directory or ".." for root - return &dir.inode -} - -type directoryFD struct { - fileDescription - vfs.DirectoryFileDescriptionDefaultImpl - - // Protected by filesystem.mu. - iter *dentry - off int64 -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { - if fd.iter != nil { - fs := fd.filesystem() - dir := fd.inode().impl.(*directory) - fs.mu.Lock() - dir.childList.Remove(fd.iter) - fs.mu.Unlock() - fd.iter = nil - } -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - fs := fd.filesystem() - vfsd := fd.vfsfd.VirtualDentry().Dentry() - - fs.mu.Lock() - defer fs.mu.Unlock() - - if fd.off == 0 { - if err := cb.Handle(vfs.Dirent{ - Name: ".", - Type: linux.DT_DIR, - Ino: vfsd.Impl().(*dentry).inode.ino, - NextOff: 1, - }); err != nil { - return err - } - fd.off++ - } - if fd.off == 1 { - parentInode := vfsd.ParentOrSelf().Impl().(*dentry).inode - if err := cb.Handle(vfs.Dirent{ - Name: "..", - Type: parentInode.direntType(), - Ino: parentInode.ino, - NextOff: 2, - }); err != nil { - return err - } - fd.off++ - } - - dir := vfsd.Impl().(*dentry).inode.impl.(*directory) - var child *dentry - if fd.iter == nil { - // Start iteration at the beginning of dir. - child = dir.childList.Front() - fd.iter = &dentry{} - } else { - // Continue iteration from where we left off. - child = fd.iter.Next() - dir.childList.Remove(fd.iter) - } - for child != nil { - // Skip other directoryFD iterators. - if child.inode != nil { - if err := cb.Handle(vfs.Dirent{ - Name: child.vfsd.Name(), - Type: child.inode.direntType(), - Ino: child.inode.ino, - NextOff: fd.off + 1, - }); err != nil { - dir.childList.InsertBefore(child, fd.iter) - return err - } - fd.off++ - } - child = child.Next() - } - dir.childList.PushBack(fd.iter) - return nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fs := fd.filesystem() - fs.mu.Lock() - defer fs.mu.Unlock() - - switch whence { - case linux.SEEK_SET: - // Use offset as given. - case linux.SEEK_CUR: - offset += fd.off - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - - // If the offset isn't changing (e.g. due to lseek(0, SEEK_CUR)), don't - // seek even if doing so might reposition the iterator due to concurrent - // mutation of the directory. Compare fs/libfs.c:dcache_dir_lseek(). - if fd.off == offset { - return offset, nil - } - - fd.off = offset - // Compensate for "." and "..". - remChildren := int64(0) - if offset >= 2 { - remChildren = offset - 2 - } - - dir := fd.inode().impl.(*directory) - - // Ensure that fd.iter exists and is not linked into dir.childList. - if fd.iter == nil { - fd.iter = &dentry{} - } else { - dir.childList.Remove(fd.iter) - } - // Insert fd.iter before the remChildren'th child, or at the end of the - // list if remChildren >= number of children. - child := dir.childList.Front() - for child != nil { - // Skip other directoryFD iterators. - if child.inode != nil { - if remChildren == 0 { - dir.childList.InsertBefore(child, fd.iter) - return offset, nil - } - remChildren-- - } - child = child.Next() - } - dir.childList.PushBack(fd.iter) - return offset, nil -} diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go deleted file mode 100644 index 12cc64385..000000000 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ /dev/null @@ -1,712 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *filesystem) Sync(ctx context.Context) error { - // All filesystem state is in-memory. - return nil -} - -// stepLocked resolves rp.Component() to an existing file, starting from the -// given directory. -// -// stepLocked is loosely analogous to fs/namei.c:walk_component(). -// -// Preconditions: filesystem.mu must be locked. !rp.Done(). -func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } -afterSymlink: - nextVFSD, err := rp.ResolveComponent(&d.vfsd) - if err != nil { - return nil, err - } - if nextVFSD == nil { - // Since the Dentry tree is the sole source of truth for tmpfs, if it's - // not in the Dentry tree, it doesn't exist. - return nil, syserror.ENOENT - } - next := nextVFSD.Impl().(*dentry) - if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO(gvisor.dev/issues/1197): Symlink traversals updates - // access time. - if err := rp.HandleSymlink(symlink.target); err != nil { - return nil, err - } - goto afterSymlink // don't check the current directory again - } - rp.Advance() - return next, nil -} - -// walkParentDirLocked resolves all but the last path component of rp to an -// existing directory, starting from the given directory (which is usually -// rp.Start().Impl().(*dentry)). It does not check that the returned directory -// is searchable by the provider of rp. -// -// walkParentDirLocked is loosely analogous to Linux's -// fs/namei.c:path_parentat(). -// -// Preconditions: filesystem.mu must be locked. !rp.Done(). -func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { - for !rp.Final() { - next, err := stepLocked(rp, d) - if err != nil { - return nil, err - } - d = next - } - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// resolveLocked resolves rp to an existing file. -// -// resolveLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). -// -// Preconditions: filesystem.mu must be locked. -func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { - d := rp.Start().Impl().(*dentry) - for !rp.Done() { - next, err := stepLocked(rp, d) - if err != nil { - return nil, err - } - d = next - } - if rp.MustBeDir() && !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// doCreateAt checks that creating a file at rp is permitted, then invokes -// create to do so. -// -// doCreateAt is loosely analogous to a conjunction of Linux's -// fs/namei.c:filename_create() and done_path_create(). -// -// Preconditions: !rp.Done(). For the final path component in rp, -// !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - name := rp.Component() - if name == "." || name == ".." { - return syserror.EEXIST - } - // Call parent.vfsd.Child() instead of stepLocked() or rp.ResolveChild(), - // because if the child exists we want to return EEXIST immediately instead - // of attempting symlink/mount traversal. - if parent.vfsd.Child(name) != nil { - return syserror.EEXIST - } - if !dir && rp.MustBeDir() { - return syserror.ENOENT - } - // In memfs, the only way to cause a dentry to be disowned is by removing - // it from the filesystem, so this check is equivalent to checking if - // parent has been removed. - if parent.vfsd.IsDisowned() { - return syserror.ENOENT - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - return create(parent, name) -} - -// AccessAt implements vfs.Filesystem.Impl.AccessAt. -func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return err - } - return d.inode.checkPermissions(creds, ats) -} - -// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. -func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return nil, err - } - if opts.CheckSearchable { - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - } - d.IncRef() - return &d.vfsd, nil -} - -// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. -func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return nil, err - } - d.IncRef() - return &d.vfsd, nil -} - -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - if rp.Mount() != vd.Mount() { - return syserror.EXDEV - } - d := vd.Dentry().Impl().(*dentry) - if d.inode.isDir() { - return syserror.EPERM - } - if d.inode.nlink == 0 { - return syserror.ENOENT - } - if d.inode.nlink == maxLinks { - return syserror.EMLINK - } - d.inode.incLinksLocked() - child := fs.newDentry(d.inode) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return fs.doCreateAt(rp, true /* dir */, func(parent *dentry, name string) error { - if parent.inode.nlink == maxLinks { - return syserror.EMLINK - } - parent.inode.incLinksLocked() // from child's ".." - child := fs.newDentry(fs.newDirectory(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - var childInode *inode - switch opts.Mode.FileType() { - case 0, linux.S_IFREG: - childInode = fs.newRegularFile(rp.Credentials(), opts.Mode) - case linux.S_IFIFO: - childInode = fs.newNamedPipe(rp.Credentials(), opts.Mode) - case linux.S_IFBLK: - childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.BlockDevice, opts.DevMajor, opts.DevMinor) - case linux.S_IFCHR: - childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor) - case linux.S_IFSOCK: - // Not yet supported. - return syserror.EPERM - default: - return syserror.EINVAL - } - child := fs.newDentry(childInode) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// OpenAt implements vfs.FilesystemImpl.OpenAt. -func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - if opts.Flags&linux.O_TMPFILE != 0 { - // Not yet supported. - return nil, syserror.EOPNOTSUPP - } - - // Handle O_CREAT and !O_CREAT separately, since in the latter case we - // don't need fs.mu for writing. - if opts.Flags&linux.O_CREAT == 0 { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return nil, err - } - return d.open(ctx, rp, &opts, false /* afterCreate */) - } - - mustCreate := opts.Flags&linux.O_EXCL != 0 - start := rp.Start().Impl().(*dentry) - fs.mu.Lock() - defer fs.mu.Unlock() - if rp.Done() { - // Reject attempts to open directories with O_CREAT. - if rp.MustBeDir() { - return nil, syserror.EISDIR - } - if mustCreate { - return nil, syserror.EEXIST - } - return start.open(ctx, rp, &opts, false /* afterCreate */) - } -afterTrailingSymlink: - parent, err := walkParentDirLocked(rp, start) - if err != nil { - return nil, err - } - // Check for search permission in the parent directory. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - // Reject attempts to open directories with O_CREAT. - if rp.MustBeDir() { - return nil, syserror.EISDIR - } - name := rp.Component() - if name == "." || name == ".." { - return nil, syserror.EISDIR - } - // Determine whether or not we need to create a file. - child, err := stepLocked(rp, parent) - if err == syserror.ENOENT { - // Already checked for searchability above; now check for writability. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { - return nil, err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return nil, err - } - defer rp.Mount().EndWrite() - // Create and open the child. - child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return child.open(ctx, rp, &opts, true) - } - if err != nil { - return nil, err - } - // Do we need to resolve a trailing symlink? - if !rp.Done() { - start = parent - goto afterTrailingSymlink - } - // Open existing file. - if mustCreate { - return nil, syserror.EEXIST - } - return child.open(ctx, rp, &opts, false) -} - -func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions, afterCreate bool) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(opts) - if !afterCreate { - if err := d.inode.checkPermissions(rp.Credentials(), ats); err != nil { - return nil, err - } - } - switch impl := d.inode.impl.(type) { - case *regularFile: - var fd regularFileFD - if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - if opts.Flags&linux.O_TRUNC != 0 { - if _, err := impl.truncate(0); err != nil { - return nil, err - } - } - return &fd.vfsfd, nil - case *directory: - // Can't open directories writably. - if ats&vfs.MayWrite != 0 { - return nil, syserror.EISDIR - } - var fd directoryFD - if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case *symlink: - // Can't open symlinks without O_PATH (which is unimplemented). - return nil, syserror.ELOOP - case *namedPipe: - return newNamedPipeFD(ctx, impl, rp, &d.vfsd, opts.Flags) - case *deviceFile: - return rp.VirtualFilesystem().OpenDeviceSpecialFile(ctx, rp.Mount(), &d.vfsd, impl.kind, impl.major, impl.minor, opts) - default: - panic(fmt.Sprintf("unknown inode type: %T", d.inode.impl)) - } -} - -// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. -func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return "", err - } - symlink, ok := d.inode.impl.(*symlink) - if !ok { - return "", syserror.EINVAL - } - return symlink.target, nil -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { - if opts.Flags != 0 { - // TODO(b/145974740): Support renameat2 flags. - return syserror.EINVAL - } - - // Resolve newParent first to verify that it's on this Mount. - fs.mu.Lock() - defer fs.mu.Unlock() - newParent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - newName := rp.Component() - if newName == "." || newName == ".." { - return syserror.EBUSY - } - mnt := rp.Mount() - if mnt != oldParentVD.Mount() { - return syserror.EXDEV - } - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - - oldParent := oldParentVD.Dentry().Impl().(*dentry) - if err := oldParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - // Call vfs.Dentry.Child() instead of stepLocked() or rp.ResolveChild(), - // because if the existing child is a symlink or mount point then we want - // to rename over it rather than follow it. - renamedVFSD := oldParent.vfsd.Child(oldName) - if renamedVFSD == nil { - return syserror.ENOENT - } - renamed := renamedVFSD.Impl().(*dentry) - if renamed.inode.isDir() { - if renamed == newParent || renamedVFSD.IsAncestorOf(&newParent.vfsd) { - return syserror.EINVAL - } - if oldParent != newParent { - // Writability is needed to change renamed's "..". - if err := renamed.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { - return err - } - } - } else { - if opts.MustBeDir || rp.MustBeDir() { - return syserror.ENOTDIR - } - } - - if err := newParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - replacedVFSD := newParent.vfsd.Child(newName) - var replaced *dentry - if replacedVFSD != nil { - replaced = replacedVFSD.Impl().(*dentry) - if replaced.inode.isDir() { - if !renamed.inode.isDir() { - return syserror.EISDIR - } - if replaced.vfsd.HasChildren() { - return syserror.ENOTEMPTY - } - } else { - if rp.MustBeDir() { - return syserror.ENOTDIR - } - if renamed.inode.isDir() { - return syserror.ENOTDIR - } - } - } else { - if renamed.inode.isDir() && newParent.inode.nlink == maxLinks { - return syserror.EMLINK - } - } - if newParent.vfsd.IsDisowned() { - return syserror.ENOENT - } - - // Linux places this check before some of those above; we do it here for - // simplicity, under the assumption that applications are not intentionally - // doing noop renames expecting them to succeed where non-noop renames - // would fail. - if renamedVFSD == replacedVFSD { - return nil - } - vfsObj := rp.VirtualFilesystem() - oldParentDir := oldParent.inode.impl.(*directory) - newParentDir := newParent.inode.impl.(*directory) - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := vfsObj.PrepareRenameDentry(mntns, renamedVFSD, replacedVFSD); err != nil { - return err - } - if replaced != nil { - newParentDir.childList.Remove(replaced) - if replaced.inode.isDir() { - newParent.inode.decLinksLocked() // from replaced's ".." - } - replaced.inode.decLinksLocked() - } - oldParentDir.childList.Remove(renamed) - newParentDir.childList.PushBack(renamed) - if renamed.inode.isDir() { - oldParent.inode.decLinksLocked() - newParent.inode.incLinksLocked() - } - // TODO(gvisor.dev/issues/1197): Update timestamps and parent directory - // sizes. - vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) - return nil -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - name := rp.Component() - if name == "." { - return syserror.EINVAL - } - if name == ".." { - return syserror.ENOTEMPTY - } - childVFSD := parent.vfsd.Child(name) - if childVFSD == nil { - return syserror.ENOENT - } - child := childVFSD.Impl().(*dentry) - if !child.inode.isDir() { - return syserror.ENOTDIR - } - if childVFSD.HasChildren() { - return syserror.ENOTEMPTY - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - vfsObj := rp.VirtualFilesystem() - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - return err - } - parent.inode.impl.(*directory).childList.Remove(child) - parent.inode.decLinksLocked() // from child's ".." - child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(childVFSD) - return nil -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return err - } - return d.inode.setStat(ctx, rp.Credentials(), &opts.Stat) -} - -// StatAt implements vfs.FilesystemImpl.StatAt. -func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return linux.Statx{}, err - } - var stat linux.Statx - d.inode.statTo(&stat) - return stat, nil -} - -// StatFSAt implements vfs.FilesystemImpl.StatFSAt. -func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return linux.Statfs{}, err - } - // TODO(gvisor.dev/issues/1197): Actually implement statfs. - return linux.Statfs{}, syserror.ENOSYS -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - name := rp.Component() - if name == "." || name == ".." { - return syserror.EISDIR - } - childVFSD := parent.vfsd.Child(name) - if childVFSD == nil { - return syserror.ENOENT - } - child := childVFSD.Impl().(*dentry) - if child.inode.isDir() { - return syserror.EISDIR - } - if rp.MustBeDir() { - return syserror.ENOTDIR - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - vfsObj := rp.VirtualFilesystem() - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - return err - } - parent.inode.impl.(*directory).childList.Remove(child) - child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(childVFSD) - return nil -} - -// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return nil, err - } - // TODO(b/127675828): support extended attributes - return nil, syserror.ENOTSUP -} - -// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return "", err - } - // TODO(b/127675828): support extended attributes - return "", syserror.ENOTSUP -} - -// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. -func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return err - } - // TODO(b/127675828): support extended attributes - return syserror.ENOTSUP -} - -// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. -func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return err - } - // TODO(b/127675828): support extended attributes - return syserror.ENOTSUP -} - -// PrependPath implements vfs.FilesystemImpl.PrependPath. -func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) -} diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go deleted file mode 100644 index 2c5c739df..000000000 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ /dev/null @@ -1,59 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/usermem" -) - -type namedPipe struct { - inode inode - - pipe *pipe.VFSPipe -} - -// Preconditions: -// * fs.mu must be locked. -// * rp.Mount().CheckBeginWrite() has been called successfully. -func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { - file := &namedPipe{pipe: pipe.NewVFSPipe(pipe.DefaultPipeSize, usermem.PageSize)} - file.inode.init(file, fs, creds, linux.S_IFIFO|mode) - file.inode.nlink = 1 // Only the parent has a link. - return &file.inode -} - -// namedPipeFD implements vfs.FileDescriptionImpl. Methods are implemented -// entirely via struct embedding. -type namedPipeFD struct { - fileDescription - - *pipe.VFSPipeFD -} - -func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { - var err error - var fd namedPipeFD - fd.VFSPipeFD, err = np.pipe.NewVFSPipeFD(ctx, vfsd, &fd.vfsfd, flags) - if err != nil { - return nil, err - } - fd.vfsfd.Init(&fd, flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}) - return &fd.vfsfd, nil -} diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go deleted file mode 100644 index 1614f2c39..000000000 --- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go +++ /dev/null @@ -1,238 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "bytes" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -const fileName = "mypipe" - -func TestSeparateFDs(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the read side. This is done in a concurrently because opening - // One end the pipe blocks until the other end is opened. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - rfdchan := make(chan *vfs.FileDescription) - go func() { - openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY} - rfd, _ := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - rfdchan <- rfd - }() - - // Open the write side. - openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY} - wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer wfd.DecRef() - - rfd, ok := <-rfdchan - if !ok { - t.Fatalf("failed to open pipe for reading %q", fileName) - } - defer rfd.DecRef() - - const msg = "vamos azul" - checkEmpty(ctx, t, rfd) - checkWrite(ctx, t, wfd, msg) - checkRead(ctx, t, rfd, msg) -} - -func TestNonblockingRead(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the read side as nonblocking. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_NONBLOCK} - rfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for reading %q: %v", fileName, err) - } - defer rfd.DecRef() - - // Open the write side. - openOpts = vfs.OpenOptions{Flags: linux.O_WRONLY} - wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer wfd.DecRef() - - const msg = "geh blau" - checkEmpty(ctx, t, rfd) - checkWrite(ctx, t, wfd, msg) - checkRead(ctx, t, rfd, msg) -} - -func TestNonblockingWriteError(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the write side as nonblocking, which should return ENXIO. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY | linux.O_NONBLOCK} - _, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != syserror.ENXIO { - t.Fatalf("expected ENXIO, but got error: %v", err) - } -} - -func TestSingleFD(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the pipe as readable and writable. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_RDWR} - fd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer fd.DecRef() - - const msg = "forza blu" - checkEmpty(ctx, t, fd) - checkWrite(ctx, t, fd, msg) - checkRead(ctx, t, fd, msg) -} - -// setup creates a VFS with a pipe in the root directory at path fileName. The -// returned VirtualDentry must be DecRef()'d be the caller. It calls t.Fatal -// upon failure. -func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesystem, vfs.VirtualDentry) { - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("failed to create tmpfs root mount: %v", err) - } - - // Create the pipe. - root := mntns.Root() - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - } - mknodOpts := vfs.MknodOptions{Mode: linux.ModeNamedPipe | 0644} - if err := vfsObj.MknodAt(ctx, creds, &pop, &mknodOpts); err != nil { - t.Fatalf("failed to create file %q: %v", fileName, err) - } - - // Sanity check: the file pipe exists and has the correct mode. - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - t.Fatalf("stat(%q) failed: %v", fileName, err) - } - if stat.Mode&^linux.S_IFMT != 0644 { - t.Errorf("got wrong permissions (%0o)", stat.Mode) - } - if stat.Mode&linux.S_IFMT != linux.ModeNamedPipe { - t.Errorf("got wrong file type (%0o)", stat.Mode) - } - - return ctx, creds, vfsObj, root -} - -// checkEmpty calls t.Fatal if the pipe in fd is not empty. -func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { - readData := make([]byte, 1) - dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) - if err != syserror.ErrWouldBlock { - t.Fatalf("expected ErrWouldBlock reading from empty pipe %q, but got: %v", fileName, err) - } - if bytesRead != 0 { - t.Fatalf("expected to read 0 bytes, but got %d", bytesRead) - } -} - -// checkWrite calls t.Fatal if it fails to write all of msg to fd. -func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { - writeData := []byte(msg) - src := usermem.BytesIOSequence(writeData) - bytesWritten, err := fd.Write(ctx, src, vfs.WriteOptions{}) - if err != nil { - t.Fatalf("error writing to pipe %q: %v", fileName, err) - } - if bytesWritten != int64(len(writeData)) { - t.Fatalf("expected to write %d bytes, but wrote %d", len(writeData), bytesWritten) - } -} - -// checkRead calls t.Fatal if it fails to read msg from fd. -func checkRead(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { - readData := make([]byte, len(msg)) - dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) - if err != nil { - t.Fatalf("error reading from pipe %q: %v", fileName, err) - } - if bytesRead != int64(len(msg)) { - t.Fatalf("expected to read %d bytes, but got %d", len(msg), bytesRead) - } - if !bytes.Equal(readData, []byte(msg)) { - t.Fatalf("expected to read %q from pipe, but got %q", msg, string(readData)) - } -} diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go deleted file mode 100644 index 26cd65605..000000000 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ /dev/null @@ -1,577 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "fmt" - "io" - "math" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// regularFile is a regular (=S_IFREG) tmpfs file. -type regularFile struct { - inode inode - - // memFile is a platform.File used to allocate pages to this regularFile. - memFile *pgalloc.MemoryFile - - // mapsMu protects mappings. - mapsMu sync.Mutex `state:"nosave"` - - // mappings tracks mappings of the file into memmap.MappingSpaces. - // - // Protected by mapsMu. - mappings memmap.MappingSet - - // writableMappingPages tracks how many pages of virtual memory are mapped - // as potentially writable from this file. If a page has multiple mappings, - // each mapping is counted separately. - // - // This counter is susceptible to overflow as we can potentially count - // mappings from many VMAs. We count pages rather than bytes to slightly - // mitigate this. - // - // Protected by mapsMu. - writableMappingPages uint64 - - // dataMu protects the fields below. - dataMu sync.RWMutex - - // data maps offsets into the file to offsets into memFile that store - // the file's data. - // - // Protected by dataMu. - data fsutil.FileRangeSet - - // seals represents file seals on this inode. - // - // Protected by dataMu. - seals uint32 - - // size is the size of data. - // - // Protected by both dataMu and inode.mu; reading it requires holding - // either mutex, while writing requires holding both AND using atomics. - // Readers that do not require consistency (like Stat) may read the - // value atomically without holding either lock. - size uint64 -} - -func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { - file := ®ularFile{ - memFile: fs.memFile, - } - file.inode.init(file, fs, creds, linux.S_IFREG|mode) - file.inode.nlink = 1 // from parent directory - return &file.inode -} - -// truncate grows or shrinks the file to the given size. It returns true if the -// file size was updated. -func (rf *regularFile) truncate(newSize uint64) (bool, error) { - rf.inode.mu.Lock() - defer rf.inode.mu.Unlock() - return rf.truncateLocked(newSize) -} - -// Preconditions: rf.inode.mu must be held. -func (rf *regularFile) truncateLocked(newSize uint64) (bool, error) { - oldSize := rf.size - if newSize == oldSize { - // Nothing to do. - return false, nil - } - - // Need to hold inode.mu and dataMu while modifying size. - rf.dataMu.Lock() - if newSize > oldSize { - // Can we grow the file? - if rf.seals&linux.F_SEAL_GROW != 0 { - rf.dataMu.Unlock() - return false, syserror.EPERM - } - // We only need to update the file size. - atomic.StoreUint64(&rf.size, newSize) - rf.dataMu.Unlock() - return true, nil - } - - // We are shrinking the file. First check if this is allowed. - if rf.seals&linux.F_SEAL_SHRINK != 0 { - rf.dataMu.Unlock() - return false, syserror.EPERM - } - - // Update the file size. - atomic.StoreUint64(&rf.size, newSize) - rf.dataMu.Unlock() - - // Invalidate past translations of truncated pages. - oldpgend := fs.OffsetPageEnd(int64(oldSize)) - newpgend := fs.OffsetPageEnd(int64(newSize)) - if newpgend < oldpgend { - rf.mapsMu.Lock() - rf.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ - // Compare Linux's mm/shmem.c:shmem_setattr() => - // mm/memory.c:unmap_mapping_range(evencows=1). - InvalidatePrivate: true, - }) - rf.mapsMu.Unlock() - } - - // We are now guaranteed that there are no translations of truncated pages, - // and can remove them. - rf.dataMu.Lock() - rf.data.Truncate(newSize, rf.memFile) - rf.dataMu.Unlock() - return true, nil -} - -// AddMapping implements memmap.Mappable.AddMapping. -func (rf *regularFile) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { - rf.mapsMu.Lock() - defer rf.mapsMu.Unlock() - rf.dataMu.RLock() - defer rf.dataMu.RUnlock() - - // Reject writable mapping if F_SEAL_WRITE is set. - if rf.seals&linux.F_SEAL_WRITE != 0 && writable { - return syserror.EPERM - } - - rf.mappings.AddMapping(ms, ar, offset, writable) - if writable { - pagesBefore := rf.writableMappingPages - - // ar is guaranteed to be page aligned per memmap.Mappable. - rf.writableMappingPages += uint64(ar.Length() / usermem.PageSize) - - if rf.writableMappingPages < pagesBefore { - panic(fmt.Sprintf("Overflow while mapping potentially writable pages pointing to a tmpfs file. Before %v, after %v", pagesBefore, rf.writableMappingPages)) - } - } - - return nil -} - -// RemoveMapping implements memmap.Mappable.RemoveMapping. -func (rf *regularFile) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { - rf.mapsMu.Lock() - defer rf.mapsMu.Unlock() - - rf.mappings.RemoveMapping(ms, ar, offset, writable) - - if writable { - pagesBefore := rf.writableMappingPages - - // ar is guaranteed to be page aligned per memmap.Mappable. - rf.writableMappingPages -= uint64(ar.Length() / usermem.PageSize) - - if rf.writableMappingPages > pagesBefore { - panic(fmt.Sprintf("Underflow while unmapping potentially writable pages pointing to a tmpfs file. Before %v, after %v", pagesBefore, rf.writableMappingPages)) - } - } -} - -// CopyMapping implements memmap.Mappable.CopyMapping. -func (rf *regularFile) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { - return rf.AddMapping(ctx, ms, dstAR, offset, writable) -} - -// Translate implements memmap.Mappable.Translate. -func (rf *regularFile) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { - rf.dataMu.Lock() - defer rf.dataMu.Unlock() - - // Constrain translations to f.attr.Size (rounded up) to prevent - // translation to pages that may be concurrently truncated. - pgend := fs.OffsetPageEnd(int64(rf.size)) - var beyondEOF bool - if required.End > pgend { - if required.Start >= pgend { - return nil, &memmap.BusError{io.EOF} - } - beyondEOF = true - required.End = pgend - } - if optional.End > pgend { - optional.End = pgend - } - - cerr := rf.data.Fill(ctx, required, optional, rf.memFile, usage.Tmpfs, func(_ context.Context, dsts safemem.BlockSeq, _ uint64) (uint64, error) { - // Newly-allocated pages are zeroed, so we don't need to do anything. - return dsts.NumBytes(), nil - }) - - var ts []memmap.Translation - var translatedEnd uint64 - for seg := rf.data.FindSegment(required.Start); seg.Ok() && seg.Start() < required.End; seg, _ = seg.NextNonEmpty() { - segMR := seg.Range().Intersect(optional) - ts = append(ts, memmap.Translation{ - Source: segMR, - File: rf.memFile, - Offset: seg.FileRangeOf(segMR).Start, - Perms: usermem.AnyAccess, - }) - translatedEnd = segMR.End - } - - // Don't return the error returned by f.data.Fill if it occurred outside of - // required. - if translatedEnd < required.End && cerr != nil { - return ts, &memmap.BusError{cerr} - } - if beyondEOF { - return ts, &memmap.BusError{io.EOF} - } - return ts, nil -} - -// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. -func (*regularFile) InvalidateUnsavable(context.Context) error { - return nil -} - -type regularFileFD struct { - fileDescription - - // off is the file offset. off is accessed using atomic memory operations. - // offMu serializes operations that may mutate off. - off int64 - offMu sync.Mutex -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { - // noop -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if dst.NumBytes() == 0 { - return 0, nil - } - f := fd.inode().impl.(*regularFile) - rw := getRegularFileReadWriter(f, offset) - n, err := dst.CopyOutFrom(ctx, rw) - putRegularFileReadWriter(rw) - return int64(n), err -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - fd.offMu.Lock() - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - srclen := src.NumBytes() - if srclen == 0 { - return 0, nil - } - f := fd.inode().impl.(*regularFile) - if end := offset + srclen; end < offset { - // Overflow. - return 0, syserror.EFBIG - } - - var err error - srclen, err = vfs.CheckLimit(ctx, offset, srclen) - if err != nil { - return 0, err - } - src = src.TakeFirst64(srclen) - - f.inode.mu.Lock() - rw := getRegularFileReadWriter(f, offset) - n, err := src.CopyInTo(ctx, rw) - f.inode.mu.Unlock() - putRegularFileReadWriter(rw) - return n, err -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - fd.offMu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.offMu.Lock() - defer fd.offMu.Unlock() - switch whence { - case linux.SEEK_SET: - // use offset as specified - case linux.SEEK_CUR: - offset += fd.off - case linux.SEEK_END: - offset += int64(atomic.LoadUint64(&fd.inode().impl.(*regularFile).size)) - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *regularFileFD) Sync(ctx context.Context) error { - return nil -} - -// LockBSD implements vfs.FileDescriptionImpl.LockBSD. -func (fd *regularFileFD) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error { - return fd.inode().lockBSD(uid, t, block) -} - -// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD. -func (fd *regularFileFD) UnlockBSD(ctx context.Context, uid lock.UniqueID) error { - fd.inode().unlockBSD(uid) - return nil -} - -// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (fd *regularFileFD) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error { - return fd.inode().lockPOSIX(uid, t, rng, block) -} - -// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error { - fd.inode().unlockPOSIX(uid, rng) - return nil -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - file := fd.inode().impl.(*regularFile) - return vfs.GenericConfigureMMap(&fd.vfsfd, file, opts) -} - -// regularFileReadWriter implements safemem.Reader and Safemem.Writer. -type regularFileReadWriter struct { - file *regularFile - - // Offset into the file to read/write at. Note that this may be - // different from the FD offset if PRead/PWrite is used. - off uint64 -} - -var regularFileReadWriterPool = sync.Pool{ - New: func() interface{} { - return ®ularFileReadWriter{} - }, -} - -func getRegularFileReadWriter(file *regularFile, offset int64) *regularFileReadWriter { - rw := regularFileReadWriterPool.Get().(*regularFileReadWriter) - rw.file = file - rw.off = uint64(offset) - return rw -} - -func putRegularFileReadWriter(rw *regularFileReadWriter) { - rw.file = nil - regularFileReadWriterPool.Put(rw) -} - -// ReadToBlocks implements safemem.Reader.ReadToBlocks. -func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { - rw.file.dataMu.RLock() - defer rw.file.dataMu.RUnlock() - size := rw.file.size - - // Compute the range to read (limited by file size and overflow-checked). - if rw.off >= size { - return 0, io.EOF - } - end := size - if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { - end = rend - } - - var done uint64 - seg, gap := rw.file.data.Find(uint64(rw.off)) - for rw.off < end { - mr := memmap.MappableRange{uint64(rw.off), uint64(end)} - switch { - case seg.Ok(): - // Get internal mappings. - ims, err := rw.file.memFile.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) - if err != nil { - return done, err - } - - // Copy from internal mappings. - n, err := safemem.CopySeq(dsts, ims) - done += n - rw.off += uint64(n) - dsts = dsts.DropFirst64(n) - if err != nil { - return done, err - } - - // Continue. - seg, gap = seg.NextNonEmpty() - - case gap.Ok(): - // Tmpfs holes are zero-filled. - gapmr := gap.Range().Intersect(mr) - dst := dsts.TakeFirst64(gapmr.Length()) - n, err := safemem.ZeroSeq(dst) - done += n - rw.off += uint64(n) - dsts = dsts.DropFirst64(n) - if err != nil { - return done, err - } - - // Continue. - seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} - } - } - return done, nil -} - -// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. -// -// Preconditions: inode.mu must be held. -func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - // Hold dataMu so we can modify size. - rw.file.dataMu.Lock() - defer rw.file.dataMu.Unlock() - - // Compute the range to write (overflow-checked). - end := rw.off + srcs.NumBytes() - if end <= rw.off { - end = math.MaxInt64 - } - - // Check if seals prevent either file growth or all writes. - switch { - case rw.file.seals&linux.F_SEAL_WRITE != 0: // Write sealed - return 0, syserror.EPERM - case end > rw.file.size && rw.file.seals&linux.F_SEAL_GROW != 0: // Grow sealed - // When growth is sealed, Linux effectively allows writes which would - // normally grow the file to partially succeed up to the current EOF, - // rounded down to the page boundary before the EOF. - // - // This happens because writes (and thus the growth check) for tmpfs - // files proceed page-by-page on Linux, and the final write to the page - // containing EOF fails, resulting in a partial write up to the start of - // that page. - // - // To emulate this behaviour, artifically truncate the write to the - // start of the page containing the current EOF. - // - // See Linux, mm/filemap.c:generic_perform_write() and - // mm/shmem.c:shmem_write_begin(). - if pgstart := uint64(usermem.Addr(rw.file.size).RoundDown()); end > pgstart { - end = pgstart - } - if end <= rw.off { - // Truncation would result in no data being written. - return 0, syserror.EPERM - } - } - - // Page-aligned mr for when we need to allocate memory. RoundUp can't - // overflow since end is an int64. - pgstartaddr := usermem.Addr(rw.off).RoundDown() - pgendaddr, _ := usermem.Addr(end).RoundUp() - pgMR := memmap.MappableRange{uint64(pgstartaddr), uint64(pgendaddr)} - - var ( - done uint64 - retErr error - ) - seg, gap := rw.file.data.Find(uint64(rw.off)) - for rw.off < end { - mr := memmap.MappableRange{uint64(rw.off), uint64(end)} - switch { - case seg.Ok(): - // Get internal mappings. - ims, err := rw.file.memFile.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Write) - if err != nil { - retErr = err - goto exitLoop - } - - // Copy to internal mappings. - n, err := safemem.CopySeq(ims, srcs) - done += n - rw.off += uint64(n) - srcs = srcs.DropFirst64(n) - if err != nil { - retErr = err - goto exitLoop - } - - // Continue. - seg, gap = seg.NextNonEmpty() - - case gap.Ok(): - // Allocate memory for the write. - gapMR := gap.Range().Intersect(pgMR) - fr, err := rw.file.memFile.Allocate(gapMR.Length(), usage.Tmpfs) - if err != nil { - retErr = err - goto exitLoop - } - - // Write to that memory as usual. - seg, gap = rw.file.data.Insert(gap, gapMR, fr.Start), fsutil.FileRangeGapIterator{} - } - } -exitLoop: - // If the write ends beyond the file's previous size, it causes the - // file to grow. - if rw.off > rw.file.size { - rw.file.size = rw.off - } - - return done, retErr -} diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go deleted file mode 100644 index 0399725cf..000000000 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ /dev/null @@ -1,496 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "bytes" - "fmt" - "io" - "sync/atomic" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// nextFileID is used to generate unique file names. -var nextFileID int64 - -// newTmpfsRoot creates a new tmpfs mount, and returns the root. If the error -// is not nil, then cleanup should be called when the root is no longer needed. -func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { - creds := auth.CredentialsFromContext(ctx) - - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) - } - - vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) - } - root := mntns.Root() - return vfsObj, root, func() { - root.DecRef() - mntns.DecRef() - }, nil -} - -// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If -// the returned err is not nil, then cleanup should be called when the FD is no -// longer needed. -func newFileFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - filename := fmt.Sprintf("tmpfs-test-file-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the file that will be write/read. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: linux.ModeRegular | mode, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create file %q: %v", filename, err) - } - - return fd, cleanup, nil -} - -// newDirFD is like newFileFD, but for directories. -func newDirFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - dirname := fmt.Sprintf("tmpfs-test-dir-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the dir. - if err := vfsObj.MkdirAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(dirname), - }, &vfs.MkdirOptions{ - Mode: linux.ModeDirectory | mode, - }); err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create directory %q: %v", dirname, err) - } - - // Open the dir and return it. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(dirname), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY | linux.O_DIRECTORY, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to open directory %q: %v", dirname, err) - } - - return fd, cleanup, nil -} - -// newPipeFD is like newFileFD, but for pipes. -func newPipeFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - pipename := fmt.Sprintf("tmpfs-test-pipe-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the pipe. - if err := vfsObj.MknodAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(pipename), - }, &vfs.MknodOptions{ - Mode: linux.ModeNamedPipe | mode, - }); err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create pipe %q: %v", pipename, err) - } - - // Open the pipe and return it. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(pipename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to open pipe %q: %v", pipename, err) - } - - return fd, cleanup, nil -} - -// Test that we can write some data to a file and read it back.` -func TestSimpleWriteRead(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - // Write. - data := []byte("foobarbaz") - n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) - if err != nil { - t.Fatalf("fd.Write failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) - } - if got, want := fd.Impl().(*regularFileFD).off, int64(len(data)); got != want { - t.Errorf("fd.Write left offset at %d, want %d", got, want) - } - - // Seek back to beginning. - if _, err := fd.Seek(ctx, 0, linux.SEEK_SET); err != nil { - t.Fatalf("fd.Seek failed: %v", err) - } - if got, want := fd.Impl().(*regularFileFD).off, int64(0); got != want { - t.Errorf("fd.Seek(0) left offset at %d, want %d", got, want) - } - - // Read. - buf := make([]byte, len(data)) - n, err = fd.Read(ctx, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) - if err != nil && err != io.EOF { - t.Fatalf("fd.Read failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.Read got short read length %d, want %d", n, len(data)) - } - if got, want := string(buf), string(data); got != want { - t.Errorf("Read got %q want %s", got, want) - } - if got, want := fd.Impl().(*regularFileFD).off, int64(len(data)); got != want { - t.Errorf("fd.Write left offset at %d, want %d", got, want) - } -} - -func TestPWrite(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - // Fill file with 1k 'a's. - data := bytes.Repeat([]byte{'a'}, 1000) - n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) - if err != nil { - t.Fatalf("fd.Write failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) - } - - // Write "gVisor is awesome" at various offsets. - buf := []byte("gVisor is awesome") - offsets := []int{0, 1, 2, 10, 20, 50, 100, len(data) - 100, len(data) - 1, len(data), len(data) + 1} - for _, offset := range offsets { - name := fmt.Sprintf("PWrite offset=%d", offset) - t.Run(name, func(t *testing.T) { - n, err := fd.PWrite(ctx, usermem.BytesIOSequence(buf), int64(offset), vfs.WriteOptions{}) - if err != nil { - t.Errorf("fd.PWrite got err %v want nil", err) - } - if n != int64(len(buf)) { - t.Errorf("fd.PWrite got %d bytes want %d", n, len(buf)) - } - - // Update data to reflect expected file contents. - if len(data) < offset+len(buf) { - data = append(data, make([]byte, (offset+len(buf))-len(data))...) - } - copy(data[offset:], buf) - - // Read the whole file and compare with data. - readBuf := make([]byte, len(data)) - n, err = fd.PRead(ctx, usermem.BytesIOSequence(readBuf), 0, vfs.ReadOptions{}) - if err != nil { - t.Fatalf("fd.PRead failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.PRead got short read length %d, want %d", n, len(data)) - } - if got, want := string(readBuf), string(data); got != want { - t.Errorf("PRead got %q want %s", got, want) - } - - }) - } -} - -func TestLocks(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - var ( - uid1 lock.UniqueID - uid2 lock.UniqueID - // Non-blocking. - block lock.Blocker - ) - - uid1 = 123 - uid2 = 456 - - if err := fd.Impl().LockBSD(ctx, uid1, lock.ReadLock, block); err != nil { - t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) - } - if err := fd.Impl().LockBSD(ctx, uid2, lock.ReadLock, block); err != nil { - t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) - } - if got, want := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block), syserror.ErrWouldBlock; got != want { - t.Fatalf("fd.Impl().LockBSD failed: got = %v, want = %v", got, want) - } - if err := fd.Impl().UnlockBSD(ctx, uid1); err != nil { - t.Fatalf("fd.Impl().UnlockBSD failed: err = %v", err) - } - if err := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block); err != nil { - t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) - } - - rng1 := lock.LockRange{0, 1} - rng2 := lock.LockRange{1, 2} - - if err := fd.Impl().LockPOSIX(ctx, uid1, lock.ReadLock, rng1, block); err != nil { - t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) - } - if err := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng2, block); err != nil { - t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) - } - if err := fd.Impl().LockPOSIX(ctx, uid1, lock.WriteLock, rng1, block); err != nil { - t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) - } - if got, want := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng1, block), syserror.ErrWouldBlock; got != want { - t.Fatalf("fd.Impl().LockPOSIX failed: got = %v, want = %v", got, want) - } - if err := fd.Impl().UnlockPOSIX(ctx, uid1, rng1); err != nil { - t.Fatalf("fd.Impl().UnlockPOSIX failed: err = %v", err) - } -} - -func TestPRead(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - // Write 100 sequences of 'gVisor is awesome'. - data := bytes.Repeat([]byte("gVisor is awsome"), 100) - n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) - if err != nil { - t.Fatalf("fd.Write failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) - } - - // Read various sizes from various offsets. - sizes := []int{0, 1, 2, 10, 20, 50, 100, 1000} - offsets := []int{0, 1, 2, 10, 20, 50, 100, 1000, len(data) - 100, len(data) - 1, len(data), len(data) + 1} - - for _, size := range sizes { - for _, offset := range offsets { - name := fmt.Sprintf("PRead offset=%d size=%d", offset, size) - t.Run(name, func(t *testing.T) { - var ( - wantRead []byte - wantErr error - ) - if offset < len(data) { - wantRead = data[offset:] - } else if size > 0 { - wantErr = io.EOF - } - if offset+size < len(data) { - wantRead = wantRead[:size] - } - buf := make([]byte, size) - n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), int64(offset), vfs.ReadOptions{}) - if err != wantErr { - t.Errorf("fd.PRead got err %v want %v", err, wantErr) - } - if n != int64(len(wantRead)) { - t.Errorf("fd.PRead got %d bytes want %d", n, len(wantRead)) - } - if got := string(buf[:n]); got != string(wantRead) { - t.Errorf("fd.PRead got %q want %q", got, string(wantRead)) - } - }) - } - } -} - -func TestTruncate(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - // Fill the file with some data. - data := bytes.Repeat([]byte("gVisor is awsome"), 100) - written, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) - if err != nil { - t.Fatalf("fd.Write failed: %v", err) - } - - // Size should be same as written. - sizeStatOpts := vfs.StatOptions{Mask: linux.STATX_SIZE} - stat, err := fd.Stat(ctx, sizeStatOpts) - if err != nil { - t.Fatalf("fd.Stat failed: %v", err) - } - if got, want := int64(stat.Size), written; got != want { - t.Errorf("fd.Stat got size %d, want %d", got, want) - } - - // Truncate down. - newSize := uint64(10) - if err := fd.SetStat(ctx, vfs.SetStatOptions{ - Stat: linux.Statx{ - Mask: linux.STATX_SIZE, - Size: newSize, - }, - }); err != nil { - t.Errorf("fd.Truncate failed: %v", err) - } - // Size should be updated. - statAfterTruncateDown, err := fd.Stat(ctx, sizeStatOpts) - if err != nil { - t.Fatalf("fd.Stat failed: %v", err) - } - if got, want := statAfterTruncateDown.Size, newSize; got != want { - t.Errorf("fd.Stat got size %d, want %d", got, want) - } - // We should only read newSize worth of data. - buf := make([]byte, 1000) - if n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0, vfs.ReadOptions{}); err != nil && err != io.EOF { - t.Fatalf("fd.PRead failed: %v", err) - } else if uint64(n) != newSize { - t.Errorf("fd.PRead got size %d, want %d", n, newSize) - } - // Mtime and Ctime should be bumped. - if got := statAfterTruncateDown.Mtime.ToNsec(); got <= stat.Mtime.ToNsec() { - t.Errorf("fd.Stat got Mtime %v, want > %v", got, stat.Mtime) - } - if got := statAfterTruncateDown.Ctime.ToNsec(); got <= stat.Ctime.ToNsec() { - t.Errorf("fd.Stat got Ctime %v, want > %v", got, stat.Ctime) - } - - // Truncate up. - newSize = 100 - if err := fd.SetStat(ctx, vfs.SetStatOptions{ - Stat: linux.Statx{ - Mask: linux.STATX_SIZE, - Size: newSize, - }, - }); err != nil { - t.Errorf("fd.Truncate failed: %v", err) - } - // Size should be updated. - statAfterTruncateUp, err := fd.Stat(ctx, sizeStatOpts) - if err != nil { - t.Fatalf("fd.Stat failed: %v", err) - } - if got, want := statAfterTruncateUp.Size, newSize; got != want { - t.Errorf("fd.Stat got size %d, want %d", got, want) - } - // We should read newSize worth of data. - buf = make([]byte, 1000) - if n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0, vfs.ReadOptions{}); err != nil && err != io.EOF { - t.Fatalf("fd.PRead failed: %v", err) - } else if uint64(n) != newSize { - t.Errorf("fd.PRead got size %d, want %d", n, newSize) - } - // Bytes should be null after 10, since we previously truncated to 10. - for i := uint64(10); i < newSize; i++ { - if buf[i] != 0 { - t.Errorf("fd.PRead got byte %d=%x, want 0", i, buf[i]) - break - } - } - // Mtime and Ctime should be bumped. - if got := statAfterTruncateUp.Mtime.ToNsec(); got <= statAfterTruncateDown.Mtime.ToNsec() { - t.Errorf("fd.Stat got Mtime %v, want > %v", got, statAfterTruncateDown.Mtime) - } - if got := statAfterTruncateUp.Ctime.ToNsec(); got <= statAfterTruncateDown.Ctime.ToNsec() { - t.Errorf("fd.Stat got Ctime %v, want > %v", got, stat.Ctime) - } - - // Truncate to the current size. - newSize = statAfterTruncateUp.Size - if err := fd.SetStat(ctx, vfs.SetStatOptions{ - Stat: linux.Statx{ - Mask: linux.STATX_SIZE, - Size: newSize, - }, - }); err != nil { - t.Errorf("fd.Truncate failed: %v", err) - } - statAfterTruncateNoop, err := fd.Stat(ctx, sizeStatOpts) - if err != nil { - t.Fatalf("fd.Stat failed: %v", err) - } - // Mtime and Ctime should not be bumped, since operation is a noop. - if got := statAfterTruncateNoop.Mtime.ToNsec(); got != statAfterTruncateUp.Mtime.ToNsec() { - t.Errorf("fd.Stat got Mtime %v, want %v", got, statAfterTruncateUp.Mtime) - } - if got := statAfterTruncateNoop.Ctime.ToNsec(); got != statAfterTruncateUp.Ctime.ToNsec() { - t.Errorf("fd.Stat got Ctime %v, want %v", got, statAfterTruncateUp.Ctime) - } -} diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go deleted file mode 100644 index ebe035dee..000000000 --- a/pkg/sentry/fsimpl/tmpfs/stat_test.go +++ /dev/null @@ -1,232 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "fmt" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func TestStatAfterCreate(t *testing.T) { - ctx := contexttest.Context(t) - mode := linux.FileMode(0644) - - // Run with different file types. - // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets. - for _, typ := range []string{"file", "dir", "pipe"} { - t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { - var ( - fd *vfs.FileDescription - cleanup func() - err error - ) - switch typ { - case "file": - fd, cleanup, err = newFileFD(ctx, mode) - case "dir": - fd, cleanup, err = newDirFD(ctx, mode) - case "pipe": - fd, cleanup, err = newPipeFD(ctx, mode) - default: - panic(fmt.Sprintf("unknown typ %q", typ)) - } - if err != nil { - t.Fatal(err) - } - defer cleanup() - - got, err := fd.Stat(ctx, vfs.StatOptions{}) - if err != nil { - t.Fatalf("Stat failed: %v", err) - } - - // Atime, Ctime, Mtime should all be current time (non-zero). - atime, ctime, mtime := got.Atime.ToNsec(), got.Ctime.ToNsec(), got.Mtime.ToNsec() - if atime != ctime || ctime != mtime { - t.Errorf("got atime=%d ctime=%d mtime=%d, wanted equal values", atime, ctime, mtime) - } - if atime == 0 { - t.Errorf("got atime=%d, want non-zero", atime) - } - - // Btime should be 0, as it is not set by tmpfs. - if btime := got.Btime.ToNsec(); btime != 0 { - t.Errorf("got btime %d, want 0", got.Btime.ToNsec()) - } - - // Size should be 0. - if got.Size != 0 { - t.Errorf("got size %d, want 0", got.Size) - } - - // Nlink should be 1 for files, 2 for dirs. - wantNlink := uint32(1) - if typ == "dir" { - wantNlink = 2 - } - if got.Nlink != wantNlink { - t.Errorf("got nlink %d, want %d", got.Nlink, wantNlink) - } - - // UID and GID are set from context creds. - creds := auth.CredentialsFromContext(ctx) - if got.UID != uint32(creds.EffectiveKUID) { - t.Errorf("got uid %d, want %d", got.UID, uint32(creds.EffectiveKUID)) - } - if got.GID != uint32(creds.EffectiveKGID) { - t.Errorf("got gid %d, want %d", got.GID, uint32(creds.EffectiveKGID)) - } - - // Mode. - wantMode := uint16(mode) - switch typ { - case "file": - wantMode |= linux.S_IFREG - case "dir": - wantMode |= linux.S_IFDIR - case "pipe": - wantMode |= linux.S_IFIFO - default: - panic(fmt.Sprintf("unknown typ %q", typ)) - } - - if got.Mode != wantMode { - t.Errorf("got mode %x, want %x", got.Mode, wantMode) - } - - // Ino. - if got.Ino == 0 { - t.Errorf("got ino %d, want not 0", got.Ino) - } - }) - } -} - -func TestSetStatAtime(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - allStatOptions := vfs.StatOptions{Mask: linux.STATX_ALL} - - // Get initial stat. - initialStat, err := fd.Stat(ctx, allStatOptions) - if err != nil { - t.Fatalf("Stat failed: %v", err) - } - - // Set atime, but without the mask. - if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: linux.Statx{ - Mask: 0, - Atime: linux.NsecToStatxTimestamp(100), - }}); err != nil { - t.Errorf("SetStat atime without mask failed: %v") - } - // Atime should be unchanged. - if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { - t.Errorf("Stat got error: %v", err) - } else if gotStat.Atime != initialStat.Atime { - t.Errorf("Stat got atime %d, want %d", gotStat.Atime, initialStat.Atime) - } - - // Set atime, this time included in the mask. - setStat := linux.Statx{ - Mask: linux.STATX_ATIME, - Atime: linux.NsecToStatxTimestamp(100), - } - if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { - t.Errorf("SetStat atime with mask failed: %v") - } - if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { - t.Errorf("Stat got error: %v", err) - } else if gotStat.Atime != setStat.Atime { - t.Errorf("Stat got atime %d, want %d", gotStat.Atime, setStat.Atime) - } -} - -func TestSetStat(t *testing.T) { - ctx := contexttest.Context(t) - mode := linux.FileMode(0644) - - // Run with different file types. - // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets. - for _, typ := range []string{"file", "dir", "pipe"} { - t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { - var ( - fd *vfs.FileDescription - cleanup func() - err error - ) - switch typ { - case "file": - fd, cleanup, err = newFileFD(ctx, mode) - case "dir": - fd, cleanup, err = newDirFD(ctx, mode) - case "pipe": - fd, cleanup, err = newPipeFD(ctx, mode) - default: - panic(fmt.Sprintf("unknown typ %q", typ)) - } - if err != nil { - t.Fatal(err) - } - defer cleanup() - - allStatOptions := vfs.StatOptions{Mask: linux.STATX_ALL} - - // Get initial stat. - initialStat, err := fd.Stat(ctx, allStatOptions) - if err != nil { - t.Fatalf("Stat failed: %v", err) - } - - // Set atime, but without the mask. - if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: linux.Statx{ - Mask: 0, - Atime: linux.NsecToStatxTimestamp(100), - }}); err != nil { - t.Errorf("SetStat atime without mask failed: %v") - } - // Atime should be unchanged. - if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { - t.Errorf("Stat got error: %v", err) - } else if gotStat.Atime != initialStat.Atime { - t.Errorf("Stat got atime %d, want %d", gotStat.Atime, initialStat.Atime) - } - - // Set atime, this time included in the mask. - setStat := linux.Statx{ - Mask: linux.STATX_ATIME, - Atime: linux.NsecToStatxTimestamp(100), - } - if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { - t.Errorf("SetStat atime with mask failed: %v") - } - if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { - t.Errorf("Stat got error: %v", err) - } else if gotStat.Atime != setStat.Atime { - t.Errorf("Stat got atime %d, want %d", gotStat.Atime, setStat.Atime) - } - }) - } -} diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go deleted file mode 100644 index 47e075ed4..000000000 --- a/pkg/sentry/fsimpl/tmpfs/symlink.go +++ /dev/null @@ -1,37 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" -) - -type symlink struct { - inode inode - target string // immutable -} - -func (fs *filesystem) newSymlink(creds *auth.Credentials, target string) *inode { - link := &symlink{ - target: target, - } - link.inode.init(link, fs, creds, linux.S_IFLNK|0777) - link.inode.nlink = 1 // from parent directory - return &link.inode -} - -// O_PATH is unimplemented, so there's no way to get a FileDescription -// representing a symlink yet. diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go deleted file mode 100644 index 2f9e6c876..000000000 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ /dev/null @@ -1,468 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package tmpfs provides a filesystem implementation that behaves like tmpfs: -// the Dentry tree is the sole source of truth for the state of the filesystem. -// -// Lock order: -// -// filesystem.mu -// inode.mu -// regularFileFD.offMu -// regularFile.mapsMu -// regularFile.dataMu -package tmpfs - -import ( - "fmt" - "math" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Name is the default filesystem name. -const Name = "tmpfs" - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - vfsfs vfs.Filesystem - - // memFile is used to allocate pages to for regular files. - memFile *pgalloc.MemoryFile - - // clock is a realtime clock used to set timestamps in file operations. - clock time.Clock - - // mu serializes changes to the Dentry tree. - mu sync.RWMutex - - nextInoMinusOne uint64 // accessed using atomic memory operations -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - memFileProvider := pgalloc.MemoryFileProviderFromContext(ctx) - if memFileProvider == nil { - panic("MemoryFileProviderFromContext returned nil") - } - clock := time.RealtimeClockFromContext(ctx) - fs := filesystem{ - memFile: memFileProvider.MemoryFile(), - clock: clock, - } - fs.vfsfs.Init(vfsObj, &fs) - root := fs.newDentry(fs.newDirectory(creds, 01777)) - return &fs.vfsfs, &root.vfsd, nil -} - -// Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { -} - -// dentry implements vfs.DentryImpl. -type dentry struct { - vfsd vfs.Dentry - - // inode is the inode represented by this dentry. Multiple Dentries may - // share a single non-directory inode (with hard links). inode is - // immutable. - inode *inode - - // tmpfs doesn't count references on dentries; because the dentry tree is - // the sole source of truth, it is by definition always consistent with the - // state of the filesystem. However, it does count references on inodes, - // because inode resources are released when all references are dropped. - // (tmpfs doesn't really have resources to release, but we implement - // reference counting because tmpfs regular files will.) - - // dentryEntry (ugh) links dentries into their parent directory.childList. - dentryEntry -} - -func (fs *filesystem) newDentry(inode *inode) *dentry { - d := &dentry{ - inode: inode, - } - d.vfsd.Init(d) - return d -} - -// IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef() { - d.inode.incRef() -} - -// TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef() bool { - return d.inode.tryIncRef() -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - d.inode.decRef() -} - -// inode represents a filesystem object. -type inode struct { - // clock is a realtime clock used to set timestamps in file operations. - clock time.Clock - - // refs is a reference count. refs is accessed using atomic memory - // operations. - // - // A reference is held on all inodes that are reachable in the filesystem - // tree. For non-directories (which may have multiple hard links), this - // means that a reference is dropped when nlink reaches 0. For directories, - // nlink never reaches 0 due to the "." entry; instead, - // filesystem.RmdirAt() drops the reference. - refs int64 - - // Inode metadata. Writing multiple fields atomically requires holding - // mu, othewise atomic operations can be used. - mu sync.Mutex - mode uint32 // file type and mode - nlink uint32 // protected by filesystem.mu instead of inode.mu - uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic - gid uint32 // auth.KGID, but ... - ino uint64 // immutable - - // Linux's tmpfs has no concept of btime. - atime int64 // nanoseconds - ctime int64 // nanoseconds - mtime int64 // nanoseconds - - // Only meaningful for device special files. - rdevMajor uint32 - rdevMinor uint32 - - // Advisory file locks, which lock at the inode level. - locks lock.FileLocks - - impl interface{} // immutable -} - -const maxLinks = math.MaxUint32 - -func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { - if mode.FileType() == 0 { - panic("file type is required in FileMode") - } - i.clock = fs.clock - i.refs = 1 - i.mode = uint32(mode) - i.uid = uint32(creds.EffectiveKUID) - i.gid = uint32(creds.EffectiveKGID) - i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) - // Tmpfs creation sets atime, ctime, and mtime to current time. - now := i.clock.Now().Nanoseconds() - i.atime = now - i.ctime = now - i.mtime = now - // i.nlink initialized by caller - i.impl = impl -} - -// incLinksLocked increments i's link count. -// -// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. -// i.nlink < maxLinks. -func (i *inode) incLinksLocked() { - if i.nlink == 0 { - panic("tmpfs.inode.incLinksLocked() called with no existing links") - } - if i.nlink == maxLinks { - panic("memfs.inode.incLinksLocked() called with maximum link count") - } - atomic.AddUint32(&i.nlink, 1) -} - -// decLinksLocked decrements i's link count. -// -// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. -func (i *inode) decLinksLocked() { - if i.nlink == 0 { - panic("tmpfs.inode.decLinksLocked() called with no existing links") - } - atomic.AddUint32(&i.nlink, ^uint32(0)) -} - -func (i *inode) incRef() { - if atomic.AddInt64(&i.refs, 1) <= 1 { - panic("tmpfs.inode.incRef() called without holding a reference") - } -} - -func (i *inode) tryIncRef() bool { - for { - refs := atomic.LoadInt64(&i.refs) - if refs == 0 { - return false - } - if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) { - return true - } - } -} - -func (i *inode) decRef() { - if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { - if regFile, ok := i.impl.(*regularFile); ok { - // Hold inode.mu and regFile.dataMu while mutating - // size. - i.mu.Lock() - regFile.dataMu.Lock() - regFile.data.DropAll(regFile.memFile) - atomic.StoreUint64(®File.size, 0) - regFile.dataMu.Unlock() - i.mu.Unlock() - } - } else if refs < 0 { - panic("tmpfs.inode.decRef() called without holding a reference") - } -} - -func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { - mode := linux.FileMode(atomic.LoadUint32(&i.mode)) - return vfs.GenericCheckPermissions(creds, ats, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))) -} - -// Go won't inline this function, and returning linux.Statx (which is quite -// big) means spending a lot of time in runtime.duffcopy(), so instead it's an -// output parameter. -// -// Note that Linux does not guarantee to return consistent data (in the case of -// a concurrent modification), so we do not require holding inode.mu. -func (i *inode) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | - linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_ATIME | - linux.STATX_BTIME | linux.STATX_CTIME | linux.STATX_MTIME - stat.Blksize = 1 // usermem.PageSize in tmpfs - stat.Nlink = atomic.LoadUint32(&i.nlink) - stat.UID = atomic.LoadUint32(&i.uid) - stat.GID = atomic.LoadUint32(&i.gid) - stat.Mode = uint16(atomic.LoadUint32(&i.mode)) - stat.Ino = i.ino - // Linux's tmpfs has no concept of btime, so zero-value is returned. - stat.Atime = linux.NsecToStatxTimestamp(i.atime) - stat.Ctime = linux.NsecToStatxTimestamp(i.ctime) - stat.Mtime = linux.NsecToStatxTimestamp(i.mtime) - // TODO(gvisor.dev/issues/1197): Device number. - switch impl := i.impl.(type) { - case *regularFile: - stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS - stat.Size = uint64(atomic.LoadUint64(&impl.size)) - // In tmpfs, this will be FileRangeSet.Span() / 512 (but also cached in - // a uint64 accessed using atomic memory operations to avoid taking - // locks). - stat.Blocks = allocatedBlocksForSize(stat.Size) - case *symlink: - stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS - stat.Size = uint64(len(impl.target)) - stat.Blocks = allocatedBlocksForSize(stat.Size) - case *deviceFile: - stat.RdevMajor = impl.major - stat.RdevMinor = impl.minor - case *directory, *namedPipe: - // Nothing to do. - default: - panic(fmt.Sprintf("unknown inode type: %T", i.impl)) - } -} - -func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx) error { - if stat.Mask == 0 { - return nil - } - if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME|linux.STATX_SIZE) != 0 { - return syserror.EPERM - } - mode := linux.FileMode(atomic.LoadUint32(&i.mode)) - if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { - return err - } - i.mu.Lock() - var ( - needsMtimeBump bool - needsCtimeBump bool - ) - mask := stat.Mask - if mask&linux.STATX_MODE != 0 { - ft := atomic.LoadUint32(&i.mode) & linux.S_IFMT - atomic.StoreUint32(&i.mode, ft|uint32(stat.Mode&^linux.S_IFMT)) - needsCtimeBump = true - } - if mask&linux.STATX_UID != 0 { - atomic.StoreUint32(&i.uid, stat.UID) - needsCtimeBump = true - } - if mask&linux.STATX_GID != 0 { - atomic.StoreUint32(&i.gid, stat.GID) - needsCtimeBump = true - } - if mask&linux.STATX_SIZE != 0 { - switch impl := i.impl.(type) { - case *regularFile: - updated, err := impl.truncateLocked(stat.Size) - if err != nil { - return err - } - if updated { - needsMtimeBump = true - needsCtimeBump = true - } - case *directory: - return syserror.EISDIR - default: - return syserror.EINVAL - } - } - if mask&linux.STATX_ATIME != 0 { - atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped()) - needsCtimeBump = true - } - if mask&linux.STATX_MTIME != 0 { - atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped()) - needsCtimeBump = true - // Ignore the mtime bump, since we just set it ourselves. - needsMtimeBump = false - } - if mask&linux.STATX_CTIME != 0 { - atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped()) - // Ignore the ctime bump, since we just set it ourselves. - needsCtimeBump = false - } - now := i.clock.Now().Nanoseconds() - if needsMtimeBump { - atomic.StoreInt64(&i.mtime, now) - } - if needsCtimeBump { - atomic.StoreInt64(&i.ctime, now) - } - i.mu.Unlock() - return nil -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) lockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { - switch i.impl.(type) { - case *regularFile: - return i.locks.LockBSD(uid, t, block) - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) unlockBSD(uid fslock.UniqueID) error { - switch i.impl.(type) { - case *regularFile: - i.locks.UnlockBSD(uid) - return nil - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) lockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { - switch i.impl.(type) { - case *regularFile: - return i.locks.LockPOSIX(uid, t, rng, block) - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) unlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) error { - switch i.impl.(type) { - case *regularFile: - i.locks.UnlockPOSIX(uid, rng) - return nil - } - return syserror.EBADF -} - -// allocatedBlocksForSize returns the number of 512B blocks needed to -// accommodate the given size in bytes, as appropriate for struct -// stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block -// size is independent of the "preferred block size for I/O", struct -// stat::st_blksize and struct statx::stx_blksize.) -func allocatedBlocksForSize(size uint64) uint64 { - return (size + 511) / 512 -} - -func (i *inode) direntType() uint8 { - switch impl := i.impl.(type) { - case *regularFile: - return linux.DT_REG - case *directory: - return linux.DT_DIR - case *symlink: - return linux.DT_LNK - case *deviceFile: - switch impl.kind { - case vfs.BlockDevice: - return linux.DT_BLK - case vfs.CharDevice: - return linux.DT_CHR - default: - panic(fmt.Sprintf("unknown vfs.DeviceKind: %v", impl.kind)) - } - default: - panic(fmt.Sprintf("unknown inode type: %T", i.impl)) - } -} - -func (i *inode) isDir() bool { - return linux.FileMode(i.mode).FileType() == linux.S_IFDIR -} - -// fileDescription is embedded by tmpfs implementations of -// vfs.FileDescriptionImpl. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl -} - -func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) -} - -func (fd *fileDescription) inode() *inode { - return fd.vfsfd.Dentry().Impl().(*dentry).inode -} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - var stat linux.Statx - fd.inode().statTo(&stat) - return stat, nil -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - creds := auth.CredentialsFromContext(ctx) - return fd.inode().setStat(ctx, creds, &opts.Stat) -} diff --git a/pkg/sentry/hostcpu/BUILD b/pkg/sentry/hostcpu/BUILD deleted file mode 100644 index e6933aa70..000000000 --- a/pkg/sentry/hostcpu/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "hostcpu", - srcs = [ - "getcpu_amd64.s", - "getcpu_arm64.s", - "hostcpu.go", - ], - visibility = ["//:sandbox"], -) - -go_test( - name = "hostcpu_test", - size = "small", - srcs = ["hostcpu_test.go"], - library = ":hostcpu", -) diff --git a/pkg/sentry/hostcpu/getcpu_arm64.s b/pkg/sentry/hostcpu/getcpu_arm64.s index caf9abb89..caf9abb89 100644..100755 --- a/pkg/sentry/hostcpu/getcpu_arm64.s +++ b/pkg/sentry/hostcpu/getcpu_arm64.s diff --git a/pkg/sentry/hostcpu/hostcpu_state_autogen.go b/pkg/sentry/hostcpu/hostcpu_state_autogen.go new file mode 100755 index 000000000..97d33d8bf --- /dev/null +++ b/pkg/sentry/hostcpu/hostcpu_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package hostcpu diff --git a/pkg/sentry/hostcpu/hostcpu_test.go b/pkg/sentry/hostcpu/hostcpu_test.go deleted file mode 100644 index 7d6885c9e..000000000 --- a/pkg/sentry/hostcpu/hostcpu_test.go +++ /dev/null @@ -1,52 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package hostcpu - -import ( - "fmt" - "testing" -) - -func TestMaxValueInLinuxBitmap(t *testing.T) { - for _, test := range []struct { - str string - max uint64 - }{ - {"0", 0}, - {"0\n", 0}, - {"0,2", 2}, - {"0-63", 63}, - {"0-3,8-11", 11}, - } { - t.Run(fmt.Sprintf("%q", test.str), func(t *testing.T) { - max, err := maxValueInLinuxBitmap(test.str) - if err != nil || max != test.max { - t.Errorf("maxValueInLinuxBitmap: got (%d, %v), wanted (%d, nil)", max, err, test.max) - } - }) - } -} - -func TestMaxValueInLinuxBitmapErrors(t *testing.T) { - for _, str := range []string{"", "\n"} { - t.Run(fmt.Sprintf("%q", str), func(t *testing.T) { - max, err := maxValueInLinuxBitmap(str) - if err == nil { - t.Errorf("maxValueInLinuxBitmap: got (%d, nil), wanted (_, error)", max) - } - t.Log(err) - }) - } -} diff --git a/pkg/sentry/hostmm/BUILD b/pkg/sentry/hostmm/BUILD deleted file mode 100644 index 61c78569d..000000000 --- a/pkg/sentry/hostmm/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "hostmm", - srcs = [ - "cgroup.go", - "hostmm.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/fd", - "//pkg/log", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/hostmm/hostmm_state_autogen.go b/pkg/sentry/hostmm/hostmm_state_autogen.go new file mode 100755 index 000000000..925c56e14 --- /dev/null +++ b/pkg/sentry/hostmm/hostmm_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package hostmm diff --git a/pkg/sentry/inet/BUILD b/pkg/sentry/inet/BUILD deleted file mode 100644 index 07bf39fed..000000000 --- a/pkg/sentry/inet/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package( - default_visibility = ["//:sandbox"], - licenses = ["notice"], -) - -go_library( - name = "inet", - srcs = [ - "context.go", - "inet.go", - "namespace.go", - "test_stack.go", - ], - deps = [ - "//pkg/context", - "//pkg/tcpip/stack", - ], -) diff --git a/pkg/sentry/inet/inet_state_autogen.go b/pkg/sentry/inet/inet_state_autogen.go new file mode 100755 index 000000000..d2985113b --- /dev/null +++ b/pkg/sentry/inet/inet_state_autogen.go @@ -0,0 +1,40 @@ +// automatically generated by stateify. + +package inet + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *TCPBufferSize) beforeSave() {} +func (x *TCPBufferSize) save(m state.Map) { + x.beforeSave() + m.Save("Min", &x.Min) + m.Save("Default", &x.Default) + m.Save("Max", &x.Max) +} + +func (x *TCPBufferSize) afterLoad() {} +func (x *TCPBufferSize) load(m state.Map) { + m.Load("Min", &x.Min) + m.Load("Default", &x.Default) + m.Load("Max", &x.Max) +} + +func (x *Namespace) beforeSave() {} +func (x *Namespace) save(m state.Map) { + x.beforeSave() + m.Save("creator", &x.creator) + m.Save("isRoot", &x.isRoot) +} + +func (x *Namespace) load(m state.Map) { + m.LoadWait("creator", &x.creator) + m.Load("isRoot", &x.isRoot) + m.AfterLoad(x.afterLoad) +} + +func init() { + state.Register("pkg/sentry/inet.TCPBufferSize", (*TCPBufferSize)(nil), state.Fns{Save: (*TCPBufferSize).save, Load: (*TCPBufferSize).load}) + state.Register("pkg/sentry/inet.Namespace", (*Namespace)(nil), state.Fns{Save: (*Namespace).save, Load: (*Namespace).load}) +} diff --git a/pkg/sentry/inet/namespace.go b/pkg/sentry/inet/namespace.go index 029af3025..029af3025 100644..100755 --- a/pkg/sentry/inet/namespace.go +++ b/pkg/sentry/inet/namespace.go diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD deleted file mode 100644 index beba29a09..000000000 --- a/pkg/sentry/kernel/BUILD +++ /dev/null @@ -1,234 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test", "proto_library") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "pending_signals_list", - out = "pending_signals_list.go", - package = "kernel", - prefix = "pendingSignal", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*pendingSignal", - "Linker": "*pendingSignal", - }, -) - -go_template_instance( - name = "process_group_list", - out = "process_group_list.go", - package = "kernel", - prefix = "processGroup", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*ProcessGroup", - "Linker": "*ProcessGroup", - }, -) - -go_template_instance( - name = "seqatomic_taskgoroutineschedinfo", - out = "seqatomic_taskgoroutineschedinfo_unsafe.go", - package = "kernel", - suffix = "TaskGoroutineSchedInfo", - template = "//pkg/sync:generic_seqatomic", - types = { - "Value": "TaskGoroutineSchedInfo", - }, -) - -go_template_instance( - name = "session_list", - out = "session_list.go", - package = "kernel", - prefix = "session", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*Session", - "Linker": "*Session", - }, -) - -go_template_instance( - name = "task_list", - out = "task_list.go", - package = "kernel", - prefix = "task", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*Task", - "Linker": "*Task", - }, -) - -go_template_instance( - name = "socket_list", - out = "socket_list.go", - package = "kernel", - prefix = "socket", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*SocketEntry", - "Linker": "*SocketEntry", - }, -) - -proto_library( - name = "uncaught_signal", - srcs = ["uncaught_signal.proto"], - visibility = ["//visibility:public"], - deps = ["//pkg/sentry/arch:registers_proto"], -) - -go_library( - name = "kernel", - srcs = [ - "abstract_socket_namespace.go", - "context.go", - "fd_table.go", - "fd_table_unsafe.go", - "fs_context.go", - "ipc_namespace.go", - "kernel.go", - "kernel_opts.go", - "kernel_state.go", - "pending_signals.go", - "pending_signals_list.go", - "pending_signals_state.go", - "posixtimer.go", - "process_group_list.go", - "ptrace.go", - "ptrace_amd64.go", - "ptrace_arm64.go", - "rseq.go", - "seccomp.go", - "seqatomic_taskgoroutineschedinfo_unsafe.go", - "session_list.go", - "sessions.go", - "signal.go", - "signal_handlers.go", - "socket_list.go", - "syscalls.go", - "syscalls_state.go", - "syslog.go", - "task.go", - "task_acct.go", - "task_block.go", - "task_clone.go", - "task_context.go", - "task_exec.go", - "task_exit.go", - "task_futex.go", - "task_identity.go", - "task_list.go", - "task_log.go", - "task_net.go", - "task_run.go", - "task_sched.go", - "task_signals.go", - "task_start.go", - "task_stop.go", - "task_syscall.go", - "task_usermem.go", - "thread_group.go", - "threads.go", - "timekeeper.go", - "timekeeper_state.go", - "tty.go", - "uts_namespace.go", - "vdso.go", - "version.go", - ], - imports = [ - "gvisor.dev/gvisor/pkg/bpf", - "gvisor.dev/gvisor/pkg/sentry/device", - "gvisor.dev/gvisor/pkg/tcpip", - ], - visibility = ["//:sandbox"], - deps = [ - ":uncaught_signal_go_proto", - "//pkg/abi", - "//pkg/abi/linux", - "//pkg/amutex", - "//pkg/binary", - "//pkg/bits", - "//pkg/bpf", - "//pkg/context", - "//pkg/cpuid", - "//pkg/eventchannel", - "//pkg/fspath", - "//pkg/log", - "//pkg/metric", - "//pkg/refs", - "//pkg/safemem", - "//pkg/secio", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/lock", - "//pkg/sentry/fs/timerfd", - "//pkg/sentry/fsbridge", - "//pkg/sentry/hostcpu", - "//pkg/sentry/inet", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/epoll", - "//pkg/sentry/kernel/futex", - "//pkg/sentry/kernel/sched", - "//pkg/sentry/kernel/semaphore", - "//pkg/sentry/kernel/shm", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/loader", - "//pkg/sentry/memmap", - "//pkg/sentry/mm", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/socket/netlink/port", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/time", - "//pkg/sentry/unimpl", - "//pkg/sentry/unimpl:unimplemented_syscall_go_proto", - "//pkg/sentry/uniqueid", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/state", - "//pkg/state/statefile", - "//pkg/sync", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/tcpip/stack", - "//pkg/usermem", - "//pkg/waiter", - "//tools/go_marshal/marshal", - ], -) - -go_test( - name = "kernel_test", - size = "small", - srcs = [ - "fd_table_test.go", - "table_test.go", - "task_test.go", - "timekeeper_test.go", - ], - library = ":kernel", - deps = [ - "//pkg/abi", - "//pkg/context", - "//pkg/sentry/arch", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/filetest", - "//pkg/sentry/kernel/sched", - "//pkg/sentry/limits", - "//pkg/sentry/pgalloc", - "//pkg/sentry/time", - "//pkg/sentry/usage", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/kernel/README.md b/pkg/sentry/kernel/README.md deleted file mode 100644 index 427311be8..000000000 --- a/pkg/sentry/kernel/README.md +++ /dev/null @@ -1,108 +0,0 @@ -This package contains: - -- A (partial) emulation of the "core Linux kernel", which governs task - execution and scheduling, system call dispatch, and signal handling. See - below for details. - -- The top-level interface for the sentry's Linux kernel emulation in general, - used by the `main` function of all versions of the sentry. This interface - revolves around the `Env` type (defined in `kernel.go`). - -# Background - -In Linux, each schedulable context is referred to interchangeably as a "task" or -"thread". Tasks can be divided into userspace and kernel tasks. In the sentry, -scheduling is managed by the Go runtime, so each schedulable context is a -goroutine; only "userspace" (application) contexts are referred to as tasks, and -represented by Task objects. (From this point forward, "task" refers to the -sentry's notion of a task unless otherwise specified.) - -At a high level, Linux application threads can be thought of as repeating a "run -loop": - -- Some amount of application code is executed in userspace. - -- A trap (explicit syscall invocation, hardware interrupt or exception, etc.) - causes control flow to switch to the kernel. - -- Some amount of kernel code is executed in kernelspace, e.g. to handle the - cause of the trap. - -- The kernel "returns from the trap" into application code. - -Analogously, each task in the sentry is associated with a *task goroutine* that -executes that task's run loop (`Task.run` in `task_run.go`). However, the -sentry's task run loop differs in structure in order to support saving execution -state to, and resuming execution from, checkpoints. - -While in kernelspace, a Linux thread can be descheduled (cease execution) in a -variety of ways: - -- It can yield or be preempted, becoming temporarily descheduled but still - runnable. At present, the sentry delegates scheduling of runnable threads to - the Go runtime. - -- It can exit, becoming permanently descheduled. The sentry's equivalent is - returning from `Task.run`, terminating the task goroutine. - -- It can enter interruptible sleep, a state in which it can be woken by a - caller-defined wakeup or the receipt of a signal. In the sentry, - interruptible sleep (which is ambiguously referred to as *blocking*) is - implemented by making all events that can end blocking (including signal - notifications) communicated via Go channels and using `select` to multiplex - wakeup sources; see `task_block.go`. - -- It can enter uninterruptible sleep, a state in which it can only be woken by - a caller-defined wakeup. Killable sleep is a closely related variant in - which the task can also be woken by SIGKILL. (These definitions also include - Linux's "group-stopped" (`TASK_STOPPED`) and "ptrace-stopped" - (`TASK_TRACED`) states.) - -To maximize compatibility with Linux, sentry checkpointing appears as a spurious -signal-delivery interrupt on all tasks; interrupted system calls return `EINTR` -or are automatically restarted as usual. However, these semantics require that -uninterruptible and killable sleeps do not appear to be interrupted. In other -words, the state of the task, including its progress through the interrupted -operation, must be preserved by checkpointing. For many such sleeps, the wakeup -condition is application-controlled, making it infeasible to wait for the sleep -to end before checkpointing. Instead, we must support checkpointing progress -through sleeping operations. - -# Implementation - -We break the task's control flow graph into *states*, delimited by: - -1. Points where uninterruptible and killable sleeps may occur. For example, - there exists a state boundary between signal dequeueing and signal delivery - because there may be an intervening ptrace signal-delivery-stop. - -2. Points where sleep-induced branches may "rejoin" normal execution. For - example, the syscall exit state exists because it can be reached immediately - following a synchronous syscall, or after a task that is sleeping in - `execve()` or `vfork()` resumes execution. - -3. Points containing large branches. This is strictly for organizational - purposes. For example, the state that processes interrupt-signaled - conditions is kept separate from the main "app" state to reduce the size of - the latter. - -4. `SyscallReinvoke`, which does not correspond to anything in Linux, and - exists solely to serve the autosave feature. - - - -States before which a stop may occur are represented as implementations of the -`taskRunState` interface named `run(state)`, allowing them to be saved and -restored. States that cannot be immediately preceded by a stop are simply `Task` -methods named `do(state)`. - -Conditions that can require task goroutines to cease execution for unknown -lengths of time are called *stops*. Stops are divided into *internal stops*, -which are stops whose start and end conditions are implemented within the -sentry, and *external stops*, which are stops whose start and end conditions are -not known to the sentry. Hence all uninterruptible and killable sleeps are -internal stops, and the existence of a pending checkpoint operation is an -external stop. Internal stops are reified into instances of the `TaskStop` type, -while external stops are merely counted. The task run loop alternates between -checking for stops and advancing the task's state. This allows checkpointing to -hold tasks in a stopped state while waiting for all tasks in the system to stop. diff --git a/pkg/sentry/kernel/auth/BUILD b/pkg/sentry/kernel/auth/BUILD deleted file mode 100644 index 2bc49483a..000000000 --- a/pkg/sentry/kernel/auth/BUILD +++ /dev/null @@ -1,69 +0,0 @@ -load("//tools:defs.bzl", "go_library") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "atomicptr_credentials", - out = "atomicptr_credentials_unsafe.go", - package = "auth", - suffix = "Credentials", - template = "//pkg/sync:generic_atomicptr", - types = { - "Value": "Credentials", - }, -) - -go_template_instance( - name = "id_map_range", - out = "id_map_range.go", - package = "auth", - prefix = "idMap", - template = "//pkg/segment:generic_range", - types = { - "T": "uint32", - }, -) - -go_template_instance( - name = "id_map_set", - out = "id_map_set.go", - consts = { - "minDegree": "3", - }, - package = "auth", - prefix = "idMap", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint32", - "Range": "idMapRange", - "Value": "uint32", - "Functions": "idMapFunctions", - }, -) - -go_library( - name = "auth", - srcs = [ - "atomicptr_credentials_unsafe.go", - "auth.go", - "capability_set.go", - "context.go", - "credentials.go", - "id.go", - "id_map.go", - "id_map_functions.go", - "id_map_range.go", - "id_map_set.go", - "user_namespace.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/bits", - "//pkg/context", - "//pkg/log", - "//pkg/sync", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/kernel/auth/atomicptr_credentials_unsafe.go b/pkg/sentry/kernel/auth/atomicptr_credentials_unsafe.go new file mode 100755 index 000000000..4535c958f --- /dev/null +++ b/pkg/sentry/kernel/auth/atomicptr_credentials_unsafe.go @@ -0,0 +1,37 @@ +package auth + +import ( + "sync/atomic" + "unsafe" +) + +// An AtomicPtr is a pointer to a value of type Value that can be atomically +// loaded and stored. The zero value of an AtomicPtr represents nil. +// +// Note that copying AtomicPtr by value performs a non-atomic read of the +// stored pointer, which is unsafe if Store() can be called concurrently; in +// this case, do `dst.Store(src.Load())` instead. +// +// +stateify savable +type AtomicPtrCredentials struct { + ptr unsafe.Pointer `state:".(*Credentials)"` +} + +func (p *AtomicPtrCredentials) savePtr() *Credentials { + return p.Load() +} + +func (p *AtomicPtrCredentials) loadPtr(v *Credentials) { + p.Store(v) +} + +// Load returns the value set by the most recent Store. It returns nil if there +// has been no previous call to Store. +func (p *AtomicPtrCredentials) Load() *Credentials { + return (*Credentials)(atomic.LoadPointer(&p.ptr)) +} + +// Store sets the value returned by Load to x. +func (p *AtomicPtrCredentials) Store(x *Credentials) { + atomic.StorePointer(&p.ptr, (unsafe.Pointer)(x)) +} diff --git a/pkg/sentry/kernel/auth/auth_state_autogen.go b/pkg/sentry/kernel/auth/auth_state_autogen.go new file mode 100755 index 000000000..09ca564b8 --- /dev/null +++ b/pkg/sentry/kernel/auth/auth_state_autogen.go @@ -0,0 +1,164 @@ +// automatically generated by stateify. + +package auth + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *AtomicPtrCredentials) beforeSave() {} +func (x *AtomicPtrCredentials) save(m state.Map) { + x.beforeSave() + var ptr *Credentials = x.savePtr() + m.SaveValue("ptr", ptr) +} + +func (x *AtomicPtrCredentials) afterLoad() {} +func (x *AtomicPtrCredentials) load(m state.Map) { + m.LoadValue("ptr", new(*Credentials), func(y interface{}) { x.loadPtr(y.(*Credentials)) }) +} + +func (x *Credentials) beforeSave() {} +func (x *Credentials) save(m state.Map) { + x.beforeSave() + m.Save("RealKUID", &x.RealKUID) + m.Save("EffectiveKUID", &x.EffectiveKUID) + m.Save("SavedKUID", &x.SavedKUID) + m.Save("RealKGID", &x.RealKGID) + m.Save("EffectiveKGID", &x.EffectiveKGID) + m.Save("SavedKGID", &x.SavedKGID) + m.Save("ExtraKGIDs", &x.ExtraKGIDs) + m.Save("PermittedCaps", &x.PermittedCaps) + m.Save("InheritableCaps", &x.InheritableCaps) + m.Save("EffectiveCaps", &x.EffectiveCaps) + m.Save("BoundingCaps", &x.BoundingCaps) + m.Save("KeepCaps", &x.KeepCaps) + m.Save("UserNamespace", &x.UserNamespace) +} + +func (x *Credentials) afterLoad() {} +func (x *Credentials) load(m state.Map) { + m.Load("RealKUID", &x.RealKUID) + m.Load("EffectiveKUID", &x.EffectiveKUID) + m.Load("SavedKUID", &x.SavedKUID) + m.Load("RealKGID", &x.RealKGID) + m.Load("EffectiveKGID", &x.EffectiveKGID) + m.Load("SavedKGID", &x.SavedKGID) + m.Load("ExtraKGIDs", &x.ExtraKGIDs) + m.Load("PermittedCaps", &x.PermittedCaps) + m.Load("InheritableCaps", &x.InheritableCaps) + m.Load("EffectiveCaps", &x.EffectiveCaps) + m.Load("BoundingCaps", &x.BoundingCaps) + m.Load("KeepCaps", &x.KeepCaps) + m.Load("UserNamespace", &x.UserNamespace) +} + +func (x *IDMapEntry) beforeSave() {} +func (x *IDMapEntry) save(m state.Map) { + x.beforeSave() + m.Save("FirstID", &x.FirstID) + m.Save("FirstParentID", &x.FirstParentID) + m.Save("Length", &x.Length) +} + +func (x *IDMapEntry) afterLoad() {} +func (x *IDMapEntry) load(m state.Map) { + m.Load("FirstID", &x.FirstID) + m.Load("FirstParentID", &x.FirstParentID) + m.Load("Length", &x.Length) +} + +func (x *idMapRange) beforeSave() {} +func (x *idMapRange) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) +} + +func (x *idMapRange) afterLoad() {} +func (x *idMapRange) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) +} + +func (x *idMapSet) beforeSave() {} +func (x *idMapSet) save(m state.Map) { + x.beforeSave() + var root *idMapSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *idMapSet) afterLoad() {} +func (x *idMapSet) load(m state.Map) { + m.LoadValue("root", new(*idMapSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*idMapSegmentDataSlices)) }) +} + +func (x *idMapnode) beforeSave() {} +func (x *idMapnode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *idMapnode) afterLoad() {} +func (x *idMapnode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *idMapSegmentDataSlices) beforeSave() {} +func (x *idMapSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *idMapSegmentDataSlices) afterLoad() {} +func (x *idMapSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *UserNamespace) beforeSave() {} +func (x *UserNamespace) save(m state.Map) { + x.beforeSave() + m.Save("parent", &x.parent) + m.Save("owner", &x.owner) + m.Save("uidMapFromParent", &x.uidMapFromParent) + m.Save("uidMapToParent", &x.uidMapToParent) + m.Save("gidMapFromParent", &x.gidMapFromParent) + m.Save("gidMapToParent", &x.gidMapToParent) +} + +func (x *UserNamespace) afterLoad() {} +func (x *UserNamespace) load(m state.Map) { + m.Load("parent", &x.parent) + m.Load("owner", &x.owner) + m.Load("uidMapFromParent", &x.uidMapFromParent) + m.Load("uidMapToParent", &x.uidMapToParent) + m.Load("gidMapFromParent", &x.gidMapFromParent) + m.Load("gidMapToParent", &x.gidMapToParent) +} + +func init() { + state.Register("pkg/sentry/kernel/auth.AtomicPtrCredentials", (*AtomicPtrCredentials)(nil), state.Fns{Save: (*AtomicPtrCredentials).save, Load: (*AtomicPtrCredentials).load}) + state.Register("pkg/sentry/kernel/auth.Credentials", (*Credentials)(nil), state.Fns{Save: (*Credentials).save, Load: (*Credentials).load}) + state.Register("pkg/sentry/kernel/auth.IDMapEntry", (*IDMapEntry)(nil), state.Fns{Save: (*IDMapEntry).save, Load: (*IDMapEntry).load}) + state.Register("pkg/sentry/kernel/auth.idMapRange", (*idMapRange)(nil), state.Fns{Save: (*idMapRange).save, Load: (*idMapRange).load}) + state.Register("pkg/sentry/kernel/auth.idMapSet", (*idMapSet)(nil), state.Fns{Save: (*idMapSet).save, Load: (*idMapSet).load}) + state.Register("pkg/sentry/kernel/auth.idMapnode", (*idMapnode)(nil), state.Fns{Save: (*idMapnode).save, Load: (*idMapnode).load}) + state.Register("pkg/sentry/kernel/auth.idMapSegmentDataSlices", (*idMapSegmentDataSlices)(nil), state.Fns{Save: (*idMapSegmentDataSlices).save, Load: (*idMapSegmentDataSlices).load}) + state.Register("pkg/sentry/kernel/auth.UserNamespace", (*UserNamespace)(nil), state.Fns{Save: (*UserNamespace).save, Load: (*UserNamespace).load}) +} diff --git a/pkg/sentry/kernel/auth/id_map_range.go b/pkg/sentry/kernel/auth/id_map_range.go new file mode 100755 index 000000000..833fa3518 --- /dev/null +++ b/pkg/sentry/kernel/auth/id_map_range.go @@ -0,0 +1,62 @@ +package auth + +// A Range represents a contiguous range of T. +// +// +stateify savable +type idMapRange struct { + // Start is the inclusive start of the range. + Start uint32 + + // End is the exclusive end of the range. + End uint32 +} + +// WellFormed returns true if r.Start <= r.End. All other methods on a Range +// require that the Range is well-formed. +func (r idMapRange) WellFormed() bool { + return r.Start <= r.End +} + +// Length returns the length of the range. +func (r idMapRange) Length() uint32 { + return r.End - r.Start +} + +// Contains returns true if r contains x. +func (r idMapRange) Contains(x uint32) bool { + return r.Start <= x && x < r.End +} + +// Overlaps returns true if r and r2 overlap. +func (r idMapRange) Overlaps(r2 idMapRange) bool { + return r.Start < r2.End && r2.Start < r.End +} + +// IsSupersetOf returns true if r is a superset of r2; that is, the range r2 is +// contained within r. +func (r idMapRange) IsSupersetOf(r2 idMapRange) bool { + return r.Start <= r2.Start && r.End >= r2.End +} + +// Intersect returns a range consisting of the intersection between r and r2. +// If r and r2 do not overlap, Intersect returns a range with unspecified +// bounds, but for which Length() == 0. +func (r idMapRange) Intersect(r2 idMapRange) idMapRange { + if r.Start < r2.Start { + r.Start = r2.Start + } + if r.End > r2.End { + r.End = r2.End + } + if r.End < r.Start { + r.End = r.Start + } + return r +} + +// CanSplitAt returns true if it is legal to split a segment spanning the range +// r at x; that is, splitting at x would produce two ranges, both of which have +// non-zero length. +func (r idMapRange) CanSplitAt(x uint32) bool { + return r.Contains(x) && r.Start < x +} diff --git a/pkg/sentry/kernel/auth/id_map_set.go b/pkg/sentry/kernel/auth/id_map_set.go new file mode 100755 index 000000000..73a17f281 --- /dev/null +++ b/pkg/sentry/kernel/auth/id_map_set.go @@ -0,0 +1,1270 @@ +package auth + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + idMapminDegree = 3 + + idMapmaxDegree = 2 * idMapminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type idMapSet struct { + root idMapnode `state:".(*idMapSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *idMapSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *idMapSet) IsEmptyRange(r idMapRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *idMapSet) Span() uint32 { + var sz uint32 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *idMapSet) SpanRange(r idMapRange) uint32 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint32 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *idMapSet) FirstSegment() idMapIterator { + if s.root.nrSegments == 0 { + return idMapIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *idMapSet) LastSegment() idMapIterator { + if s.root.nrSegments == 0 { + return idMapIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *idMapSet) FirstGap() idMapGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return idMapGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *idMapSet) LastGap() idMapGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return idMapGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *idMapSet) Find(key uint32) (idMapIterator, idMapGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return idMapIterator{n, i}, idMapGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return idMapIterator{}, idMapGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *idMapSet) FindSegment(key uint32) idMapIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *idMapSet) LowerBoundSegment(min uint32) idMapIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *idMapSet) UpperBoundSegment(max uint32) idMapIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *idMapSet) FindGap(key uint32) idMapGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *idMapSet) LowerBoundGap(min uint32) idMapGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *idMapSet) UpperBoundGap(max uint32) idMapGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *idMapSet) Add(r idMapRange, val uint32) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *idMapSet) AddWithoutMerging(r idMapRange, val uint32) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *idMapSet) Insert(gap idMapGapIterator, r idMapRange, val uint32) idMapIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (idMapFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (idMapFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (idMapFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *idMapSet) InsertWithoutMerging(gap idMapGapIterator, r idMapRange, val uint32) idMapIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *idMapSet) InsertWithoutMergingUnchecked(gap idMapGapIterator, r idMapRange, val uint32) idMapIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return idMapIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *idMapSet) Remove(seg idMapIterator) idMapGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + idMapFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(idMapGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *idMapSet) RemoveAll() { + s.root = idMapnode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *idMapSet) RemoveRange(r idMapRange) idMapGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *idMapSet) Merge(first, second idMapIterator) idMapIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *idMapSet) MergeUnchecked(first, second idMapIterator) idMapIterator { + if first.End() == second.Start() { + if mval, ok := (idMapFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return idMapIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *idMapSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *idMapSet) MergeRange(r idMapRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *idMapSet) MergeAdjacent(r idMapRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *idMapSet) Split(seg idMapIterator, split uint32) (idMapIterator, idMapIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *idMapSet) SplitUnchecked(seg idMapIterator, split uint32) (idMapIterator, idMapIterator) { + val1, val2 := (idMapFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), idMapRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *idMapSet) SplitAt(split uint32) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *idMapSet) Isolate(seg idMapIterator, r idMapRange) idMapIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *idMapSet) ApplyContiguous(r idMapRange, fn func(seg idMapIterator)) idMapGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return idMapGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return idMapGapIterator{} + } + } +} + +// +stateify savable +type idMapnode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *idMapnode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [idMapmaxDegree - 1]idMapRange + values [idMapmaxDegree - 1]uint32 + children [idMapmaxDegree]*idMapnode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *idMapnode) firstSegment() idMapIterator { + for n.hasChildren { + n = n.children[0] + } + return idMapIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *idMapnode) lastSegment() idMapIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return idMapIterator{n, n.nrSegments - 1} +} + +func (n *idMapnode) prevSibling() *idMapnode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *idMapnode) nextSibling() *idMapnode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *idMapnode) rebalanceBeforeInsert(gap idMapGapIterator) idMapGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < idMapmaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &idMapnode{ + nrSegments: idMapminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &idMapnode{ + nrSegments: idMapminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:idMapminDegree-1], n.keys[:idMapminDegree-1]) + copy(left.values[:idMapminDegree-1], n.values[:idMapminDegree-1]) + copy(right.keys[:idMapminDegree-1], n.keys[idMapminDegree:]) + copy(right.values[:idMapminDegree-1], n.values[idMapminDegree:]) + n.keys[0], n.values[0] = n.keys[idMapminDegree-1], n.values[idMapminDegree-1] + idMapzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:idMapminDegree], n.children[:idMapminDegree]) + copy(right.children[:idMapminDegree], n.children[idMapminDegree:]) + idMapzeroNodeSlice(n.children[2:]) + for i := 0; i < idMapminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < idMapminDegree { + return idMapGapIterator{left, gap.index} + } + return idMapGapIterator{right, gap.index - idMapminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[idMapminDegree-1], n.values[idMapminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &idMapnode{ + nrSegments: idMapminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:idMapminDegree-1], n.keys[idMapminDegree:]) + copy(sibling.values[:idMapminDegree-1], n.values[idMapminDegree:]) + idMapzeroValueSlice(n.values[idMapminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:idMapminDegree], n.children[idMapminDegree:]) + idMapzeroNodeSlice(n.children[idMapminDegree:]) + for i := 0; i < idMapminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = idMapminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < idMapminDegree { + return gap + } + return idMapGapIterator{sibling, gap.index - idMapminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *idMapnode) rebalanceAfterRemove(gap idMapGapIterator) idMapGapIterator { + for { + if n.nrSegments >= idMapminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= idMapminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + idMapFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return idMapGapIterator{n, 0} + } + if gap.node == n { + return idMapGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= idMapminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + idMapFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return idMapGapIterator{n, n.nrSegments} + } + return idMapGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return idMapGapIterator{p, gap.index} + } + if gap.node == right { + return idMapGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *idMapnode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = idMapGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + idMapFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type idMapIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *idMapnode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg idMapIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg idMapIterator) Range() idMapRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg idMapIterator) Start() uint32 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg idMapIterator) End() uint32 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg idMapIterator) SetRangeUnchecked(r idMapRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg idMapIterator) SetRange(r idMapRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg idMapIterator) SetStartUnchecked(start uint32) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg idMapIterator) SetStart(start uint32) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg idMapIterator) SetEndUnchecked(end uint32) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg idMapIterator) SetEnd(end uint32) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg idMapIterator) Value() uint32 { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg idMapIterator) ValuePtr() *uint32 { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg idMapIterator) SetValue(val uint32) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg idMapIterator) PrevSegment() idMapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return idMapIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return idMapIterator{} + } + return idMapsegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg idMapIterator) NextSegment() idMapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return idMapIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return idMapIterator{} + } + return idMapsegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg idMapIterator) PrevGap() idMapGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return idMapGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg idMapIterator) NextGap() idMapGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return idMapGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg idMapIterator) PrevNonEmpty() (idMapIterator, idMapGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return idMapIterator{}, gap + } + return gap.PrevSegment(), idMapGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg idMapIterator) NextNonEmpty() (idMapIterator, idMapGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return idMapIterator{}, gap + } + return gap.NextSegment(), idMapGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type idMapGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *idMapnode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap idMapGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap idMapGapIterator) Range() idMapRange { + return idMapRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap idMapGapIterator) Start() uint32 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return idMapFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap idMapGapIterator) End() uint32 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return idMapFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap idMapGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap idMapGapIterator) PrevSegment() idMapIterator { + return idMapsegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap idMapGapIterator) NextSegment() idMapIterator { + return idMapsegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap idMapGapIterator) PrevGap() idMapGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return idMapGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap idMapGapIterator) NextGap() idMapGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return idMapGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func idMapsegmentBeforePosition(n *idMapnode, i int) idMapIterator { + for i == 0 { + if n.parent == nil { + return idMapIterator{} + } + n, i = n.parent, n.parentIndex + } + return idMapIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func idMapsegmentAfterPosition(n *idMapnode, i int) idMapIterator { + for i == n.nrSegments { + if n.parent == nil { + return idMapIterator{} + } + n, i = n.parent, n.parentIndex + } + return idMapIterator{n, i} +} + +func idMapzeroValueSlice(slice []uint32) { + + for i := range slice { + idMapFunctions{}.ClearValue(&slice[i]) + } +} + +func idMapzeroNodeSlice(slice []*idMapnode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *idMapSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *idMapnode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *idMapnode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type idMapSegmentDataSlices struct { + Start []uint32 + End []uint32 + Values []uint32 +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *idMapSet) ExportSortedSlices() *idMapSegmentDataSlices { + var sds idMapSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *idMapSet) ImportSortedSlices(sds *idMapSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := idMapRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *idMapSet) saveRoot() *idMapSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *idMapSet) loadRoot(sds *idMapSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/kernel/contexttest/BUILD b/pkg/sentry/kernel/contexttest/BUILD deleted file mode 100644 index 9d26392c0..000000000 --- a/pkg/sentry/kernel/contexttest/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "contexttest", - testonly = 1, - srcs = ["contexttest.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/sentry/contexttest", - "//pkg/sentry/kernel", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - ], -) diff --git a/pkg/sentry/kernel/contexttest/contexttest.go b/pkg/sentry/kernel/contexttest/contexttest.go deleted file mode 100644 index 22c340e56..000000000 --- a/pkg/sentry/kernel/contexttest/contexttest.go +++ /dev/null @@ -1,40 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package contexttest provides a test context.Context which includes -// a dummy kernel pointing to a valid platform. -package contexttest - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" -) - -// Context returns a Context that may be used in tests. Uses ptrace as the -// platform.Platform, and provides a stub kernel that only serves to point to -// the platform. -func Context(tb testing.TB) context.Context { - ctx := contexttest.Context(tb) - k := &kernel.Kernel{ - Platform: platform.FromContext(ctx), - } - k.SetMemoryFile(pgalloc.MemoryFileFromContext(ctx)) - ctx.(*contexttest.TestContext).RegisterValue(kernel.CtxKernel, k) - return ctx -} diff --git a/pkg/sentry/kernel/epoll/BUILD b/pkg/sentry/kernel/epoll/BUILD deleted file mode 100644 index dedf0fa15..000000000 --- a/pkg/sentry/kernel/epoll/BUILD +++ /dev/null @@ -1,50 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "epoll_list", - out = "epoll_list.go", - package = "epoll", - prefix = "pollEntry", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*pollEntry", - "Linker": "*pollEntry", - }, -) - -go_library( - name = "epoll", - srcs = [ - "epoll.go", - "epoll_list.go", - "epoll_state.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/refs", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sync", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "epoll_test", - size = "small", - srcs = [ - "epoll_test.go", - ], - library = ":epoll", - deps = [ - "//pkg/sentry/contexttest", - "//pkg/sentry/fs/filetest", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/kernel/epoll/epoll_list.go b/pkg/sentry/kernel/epoll/epoll_list.go new file mode 100755 index 000000000..37f757fa8 --- /dev/null +++ b/pkg/sentry/kernel/epoll/epoll_list.go @@ -0,0 +1,186 @@ +package epoll + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type pollEntryElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (pollEntryElementMapper) linkerFor(elem *pollEntry) *pollEntry { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type pollEntryList struct { + head *pollEntry + tail *pollEntry +} + +// Reset resets list l to the empty state. +func (l *pollEntryList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *pollEntryList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *pollEntryList) Front() *pollEntry { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *pollEntryList) Back() *pollEntry { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *pollEntryList) PushFront(e *pollEntry) { + linker := pollEntryElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + pollEntryElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *pollEntryList) PushBack(e *pollEntry) { + linker := pollEntryElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + pollEntryElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *pollEntryList) PushBackList(m *pollEntryList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + pollEntryElementMapper{}.linkerFor(l.tail).SetNext(m.head) + pollEntryElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *pollEntryList) InsertAfter(b, e *pollEntry) { + bLinker := pollEntryElementMapper{}.linkerFor(b) + eLinker := pollEntryElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + pollEntryElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *pollEntryList) InsertBefore(a, e *pollEntry) { + aLinker := pollEntryElementMapper{}.linkerFor(a) + eLinker := pollEntryElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + pollEntryElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *pollEntryList) Remove(e *pollEntry) { + linker := pollEntryElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + pollEntryElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + pollEntryElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type pollEntryEntry struct { + next *pollEntry + prev *pollEntry +} + +// Next returns the entry that follows e in the list. +func (e *pollEntryEntry) Next() *pollEntry { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *pollEntryEntry) Prev() *pollEntry { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *pollEntryEntry) SetNext(elem *pollEntry) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *pollEntryEntry) SetPrev(elem *pollEntry) { + e.prev = elem +} diff --git a/pkg/sentry/kernel/epoll/epoll_state_autogen.go b/pkg/sentry/kernel/epoll/epoll_state_autogen.go new file mode 100755 index 000000000..afdea5bcf --- /dev/null +++ b/pkg/sentry/kernel/epoll/epoll_state_autogen.go @@ -0,0 +1,113 @@ +// automatically generated by stateify. + +package epoll + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *FileIdentifier) beforeSave() {} +func (x *FileIdentifier) save(m state.Map) { + x.beforeSave() + m.Save("File", &x.File) + m.Save("Fd", &x.Fd) +} + +func (x *FileIdentifier) afterLoad() {} +func (x *FileIdentifier) load(m state.Map) { + m.LoadWait("File", &x.File) + m.Load("Fd", &x.Fd) +} + +func (x *pollEntry) beforeSave() {} +func (x *pollEntry) save(m state.Map) { + x.beforeSave() + m.Save("pollEntryEntry", &x.pollEntryEntry) + m.Save("id", &x.id) + m.Save("userData", &x.userData) + m.Save("mask", &x.mask) + m.Save("flags", &x.flags) + m.Save("epoll", &x.epoll) +} + +func (x *pollEntry) load(m state.Map) { + m.Load("pollEntryEntry", &x.pollEntryEntry) + m.LoadWait("id", &x.id) + m.Load("userData", &x.userData) + m.Load("mask", &x.mask) + m.Load("flags", &x.flags) + m.Load("epoll", &x.epoll) + m.AfterLoad(x.afterLoad) +} + +func (x *EventPoll) beforeSave() {} +func (x *EventPoll) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.FilePipeSeek) { + m.Failf("FilePipeSeek is %v, expected zero", x.FilePipeSeek) + } + if !state.IsZeroValue(x.FileNotDirReaddir) { + m.Failf("FileNotDirReaddir is %v, expected zero", x.FileNotDirReaddir) + } + if !state.IsZeroValue(x.FileNoFsync) { + m.Failf("FileNoFsync is %v, expected zero", x.FileNoFsync) + } + if !state.IsZeroValue(x.FileNoopFlush) { + m.Failf("FileNoopFlush is %v, expected zero", x.FileNoopFlush) + } + if !state.IsZeroValue(x.FileNoIoctl) { + m.Failf("FileNoIoctl is %v, expected zero", x.FileNoIoctl) + } + if !state.IsZeroValue(x.FileNoMMap) { + m.Failf("FileNoMMap is %v, expected zero", x.FileNoMMap) + } + if !state.IsZeroValue(x.Queue) { + m.Failf("Queue is %v, expected zero", x.Queue) + } + m.Save("files", &x.files) + m.Save("readyList", &x.readyList) + m.Save("waitingList", &x.waitingList) + m.Save("disabledList", &x.disabledList) +} + +func (x *EventPoll) load(m state.Map) { + m.Load("files", &x.files) + m.Load("readyList", &x.readyList) + m.Load("waitingList", &x.waitingList) + m.Load("disabledList", &x.disabledList) + m.AfterLoad(x.afterLoad) +} + +func (x *pollEntryList) beforeSave() {} +func (x *pollEntryList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *pollEntryList) afterLoad() {} +func (x *pollEntryList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *pollEntryEntry) beforeSave() {} +func (x *pollEntryEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *pollEntryEntry) afterLoad() {} +func (x *pollEntryEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func init() { + state.Register("pkg/sentry/kernel/epoll.FileIdentifier", (*FileIdentifier)(nil), state.Fns{Save: (*FileIdentifier).save, Load: (*FileIdentifier).load}) + state.Register("pkg/sentry/kernel/epoll.pollEntry", (*pollEntry)(nil), state.Fns{Save: (*pollEntry).save, Load: (*pollEntry).load}) + state.Register("pkg/sentry/kernel/epoll.EventPoll", (*EventPoll)(nil), state.Fns{Save: (*EventPoll).save, Load: (*EventPoll).load}) + state.Register("pkg/sentry/kernel/epoll.pollEntryList", (*pollEntryList)(nil), state.Fns{Save: (*pollEntryList).save, Load: (*pollEntryList).load}) + state.Register("pkg/sentry/kernel/epoll.pollEntryEntry", (*pollEntryEntry)(nil), state.Fns{Save: (*pollEntryEntry).save, Load: (*pollEntryEntry).load}) +} diff --git a/pkg/sentry/kernel/epoll/epoll_test.go b/pkg/sentry/kernel/epoll/epoll_test.go deleted file mode 100644 index 22630e9c5..000000000 --- a/pkg/sentry/kernel/epoll/epoll_test.go +++ /dev/null @@ -1,54 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package epoll - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs/filetest" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestFileDestroyed(t *testing.T) { - f := filetest.NewTestFile(t) - id := FileIdentifier{f, 12} - - efile := NewEventPoll(contexttest.Context(t)) - e := efile.FileOperations.(*EventPoll) - if err := e.AddEntry(id, 0, waiter.EventIn, [2]int32{}); err != nil { - t.Fatalf("addEntry failed: %v", err) - } - - // Check that we get an event reported twice in a row. - evt := e.ReadEvents(1) - if len(evt) != 1 { - t.Fatalf("Unexpected number of ready events: want %v, got %v", 1, len(evt)) - } - - evt = e.ReadEvents(1) - if len(evt) != 1 { - t.Fatalf("Unexpected number of ready events: want %v, got %v", 1, len(evt)) - } - - // Destroy the file. Check that we get no more events. - f.DecRef() - - evt = e.ReadEvents(1) - if len(evt) != 0 { - t.Fatalf("Unexpected number of ready events: want %v, got %v", 0, len(evt)) - } - -} diff --git a/pkg/sentry/kernel/eventfd/BUILD b/pkg/sentry/kernel/eventfd/BUILD deleted file mode 100644 index 9983a32e5..000000000 --- a/pkg/sentry/kernel/eventfd/BUILD +++ /dev/null @@ -1,33 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "eventfd", - srcs = ["eventfd.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fdnotifier", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "eventfd_test", - size = "small", - srcs = ["eventfd_test.go"], - library = ":eventfd", - deps = [ - "//pkg/sentry/contexttest", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/kernel/eventfd/eventfd_state_autogen.go b/pkg/sentry/kernel/eventfd/eventfd_state_autogen.go new file mode 100755 index 000000000..9cf0ac817 --- /dev/null +++ b/pkg/sentry/kernel/eventfd/eventfd_state_autogen.go @@ -0,0 +1,29 @@ +// automatically generated by stateify. + +package eventfd + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *EventOperations) beforeSave() {} +func (x *EventOperations) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.wq) { + m.Failf("wq is %v, expected zero", x.wq) + } + m.Save("val", &x.val) + m.Save("semMode", &x.semMode) + m.Save("hostfd", &x.hostfd) +} + +func (x *EventOperations) afterLoad() {} +func (x *EventOperations) load(m state.Map) { + m.Load("val", &x.val) + m.Load("semMode", &x.semMode) + m.Load("hostfd", &x.hostfd) +} + +func init() { + state.Register("pkg/sentry/kernel/eventfd.EventOperations", (*EventOperations)(nil), state.Fns{Save: (*EventOperations).save, Load: (*EventOperations).load}) +} diff --git a/pkg/sentry/kernel/eventfd/eventfd_test.go b/pkg/sentry/kernel/eventfd/eventfd_test.go deleted file mode 100644 index 9b4892f74..000000000 --- a/pkg/sentry/kernel/eventfd/eventfd_test.go +++ /dev/null @@ -1,78 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package eventfd - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestEventfd(t *testing.T) { - initVals := []uint64{ - 0, - // Using a non-zero initial value verifies that writing to an - // eventfd signals when the eventfd's counter was already - // non-zero. - 343, - } - - for _, initVal := range initVals { - ctx := contexttest.Context(t) - - // Make a new event that is writable. - event := New(ctx, initVal, false) - - // Register a callback for a write event. - w, ch := waiter.NewChannelEntry(nil) - event.EventRegister(&w, waiter.EventIn) - defer event.EventUnregister(&w) - - data := []byte("00000124") - // Create and submit a write request. - n, err := event.Writev(ctx, usermem.BytesIOSequence(data)) - if err != nil { - t.Fatal(err) - } - if n != 8 { - t.Errorf("eventfd.write wrote %d bytes, not full int64", n) - } - - // Check if the callback fired due to the write event. - select { - case <-ch: - default: - t.Errorf("Didn't get notified of EventIn after write") - } - } -} - -func TestEventfdStat(t *testing.T) { - ctx := contexttest.Context(t) - - // Make a new event that is writable. - event := New(ctx, 0, false) - - // Create and submit an stat request. - uattr, err := event.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - t.Fatalf("eventfd stat request failed: %v", err) - } - if uattr.Size != 0 { - t.Fatal("EventFD size should be 0") - } -} diff --git a/pkg/sentry/kernel/fasync/BUILD b/pkg/sentry/kernel/fasync/BUILD deleted file mode 100644 index b9126e946..000000000 --- a/pkg/sentry/kernel/fasync/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "fasync", - srcs = ["fasync.go"], - visibility = ["//:sandbox"], - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/fs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sync", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/kernel/fasync/fasync_state_autogen.go b/pkg/sentry/kernel/fasync/fasync_state_autogen.go new file mode 100755 index 000000000..fdcd48f64 --- /dev/null +++ b/pkg/sentry/kernel/fasync/fasync_state_autogen.go @@ -0,0 +1,32 @@ +// automatically generated by stateify. + +package fasync + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *FileAsync) beforeSave() {} +func (x *FileAsync) save(m state.Map) { + x.beforeSave() + m.Save("e", &x.e) + m.Save("requester", &x.requester) + m.Save("registered", &x.registered) + m.Save("recipientPG", &x.recipientPG) + m.Save("recipientTG", &x.recipientTG) + m.Save("recipientT", &x.recipientT) +} + +func (x *FileAsync) afterLoad() {} +func (x *FileAsync) load(m state.Map) { + m.Load("e", &x.e) + m.Load("requester", &x.requester) + m.Load("registered", &x.registered) + m.Load("recipientPG", &x.recipientPG) + m.Load("recipientTG", &x.recipientTG) + m.Load("recipientT", &x.recipientT) +} + +func init() { + state.Register("pkg/sentry/kernel/fasync.FileAsync", (*FileAsync)(nil), state.Fns{Save: (*FileAsync).save, Load: (*FileAsync).load}) +} diff --git a/pkg/sentry/kernel/fd_table_test.go b/pkg/sentry/kernel/fd_table_test.go deleted file mode 100644 index 29f95a2c4..000000000 --- a/pkg/sentry/kernel/fd_table_test.go +++ /dev/null @@ -1,228 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernel - -import ( - "runtime" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/filetest" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sync" -) - -const ( - // maxFD is the maximum FD to try to create in the map. - // - // This number of open files has been seen in the wild. - maxFD = 2 * 1024 -) - -func runTest(t testing.TB, fn func(ctx context.Context, fdTable *FDTable, file *fs.File, limitSet *limits.LimitSet)) { - t.Helper() // Don't show in stacks. - - // Create the limits and context. - limitSet := limits.NewLimitSet() - limitSet.Set(limits.NumberOfFiles, limits.Limit{maxFD, maxFD}, true) - ctx := contexttest.WithLimitSet(contexttest.Context(t), limitSet) - - // Create a test file.; - file := filetest.NewTestFile(t) - - // Create the table. - fdTable := new(FDTable) - fdTable.init() - - // Run the test. - fn(ctx, fdTable, file, limitSet) -} - -// TestFDTableMany allocates maxFD FDs, i.e. maxes out the FDTable, until there -// is no room, then makes sure that NewFDAt works and also that if we remove -// one and add one that works too. -func TestFDTableMany(t *testing.T) { - runTest(t, func(ctx context.Context, fdTable *FDTable, file *fs.File, _ *limits.LimitSet) { - for i := 0; i < maxFD; i++ { - if _, err := fdTable.NewFDs(ctx, 0, []*fs.File{file}, FDFlags{}); err != nil { - t.Fatalf("Allocated %v FDs but wanted to allocate %v", i, maxFD) - } - } - - if _, err := fdTable.NewFDs(ctx, 0, []*fs.File{file}, FDFlags{}); err == nil { - t.Fatalf("fdTable.NewFDs(0, r) in full map: got nil, wanted error") - } - - if err := fdTable.NewFDAt(ctx, 1, file, FDFlags{}); err != nil { - t.Fatalf("fdTable.NewFDAt(1, r, FDFlags{}): got %v, wanted nil", err) - } - - i := int32(2) - fdTable.Remove(i) - if fds, err := fdTable.NewFDs(ctx, 0, []*fs.File{file}, FDFlags{}); err != nil || fds[0] != i { - t.Fatalf("Allocated %v FDs but wanted to allocate %v: %v", i, maxFD, err) - } - }) -} - -func TestFDTableOverLimit(t *testing.T) { - runTest(t, func(ctx context.Context, fdTable *FDTable, file *fs.File, _ *limits.LimitSet) { - if _, err := fdTable.NewFDs(ctx, maxFD, []*fs.File{file}, FDFlags{}); err == nil { - t.Fatalf("fdTable.NewFDs(maxFD, f): got nil, wanted error") - } - - if _, err := fdTable.NewFDs(ctx, maxFD-2, []*fs.File{file, file, file}, FDFlags{}); err == nil { - t.Fatalf("fdTable.NewFDs(maxFD-2, {f,f,f}): got nil, wanted error") - } - - if fds, err := fdTable.NewFDs(ctx, maxFD-3, []*fs.File{file, file, file}, FDFlags{}); err != nil { - t.Fatalf("fdTable.NewFDs(maxFD-3, {f,f,f}): got %v, wanted nil", err) - } else { - for _, fd := range fds { - fdTable.Remove(fd) - } - } - - if fds, err := fdTable.NewFDs(ctx, maxFD-1, []*fs.File{file}, FDFlags{}); err != nil || fds[0] != maxFD-1 { - t.Fatalf("fdTable.NewFDAt(1, r, FDFlags{}): got %v, wanted nil", err) - } - - if fds, err := fdTable.NewFDs(ctx, 0, []*fs.File{file}, FDFlags{}); err != nil { - t.Fatalf("Adding an FD to a resized map: got %v, want nil", err) - } else if len(fds) != 1 || fds[0] != 0 { - t.Fatalf("Added an FD to a resized map: got %v, want {1}", fds) - } - }) -} - -// TestFDTable does a set of simple tests to make sure simple adds, removes, -// GetRefs, and DecRefs work. The ordering is just weird enough that a -// table-driven approach seemed clumsy. -func TestFDTable(t *testing.T) { - runTest(t, func(ctx context.Context, fdTable *FDTable, file *fs.File, limitSet *limits.LimitSet) { - // Cap the limit at one. - limitSet.Set(limits.NumberOfFiles, limits.Limit{1, maxFD}, true) - - if _, err := fdTable.NewFDs(ctx, 0, []*fs.File{file}, FDFlags{}); err != nil { - t.Fatalf("Adding an FD to an empty 1-size map: got %v, want nil", err) - } - - if _, err := fdTable.NewFDs(ctx, 0, []*fs.File{file}, FDFlags{}); err == nil { - t.Fatalf("Adding an FD to a filled 1-size map: got nil, wanted an error") - } - - // Remove the previous limit. - limitSet.Set(limits.NumberOfFiles, limits.Limit{maxFD, maxFD}, true) - - if fds, err := fdTable.NewFDs(ctx, 0, []*fs.File{file}, FDFlags{}); err != nil { - t.Fatalf("Adding an FD to a resized map: got %v, want nil", err) - } else if len(fds) != 1 || fds[0] != 1 { - t.Fatalf("Added an FD to a resized map: got %v, want {1}", fds) - } - - if err := fdTable.NewFDAt(ctx, 1, file, FDFlags{}); err != nil { - t.Fatalf("Replacing FD 1 via fdTable.NewFDAt(1, r, FDFlags{}): got %v, wanted nil", err) - } - - if err := fdTable.NewFDAt(ctx, maxFD+1, file, FDFlags{}); err == nil { - t.Fatalf("Using an FD that was too large via fdTable.NewFDAt(%v, r, FDFlags{}): got nil, wanted an error", maxFD+1) - } - - if ref, _ := fdTable.Get(1); ref == nil { - t.Fatalf("fdTable.Get(1): got nil, wanted %v", file) - } - - if ref, _ := fdTable.Get(2); ref != nil { - t.Fatalf("fdTable.Get(2): got a %v, wanted nil", ref) - } - - ref, _ := fdTable.Remove(1) - if ref == nil { - t.Fatalf("fdTable.Remove(1) for an existing FD: failed, want success") - } - ref.DecRef() - - if ref, _ := fdTable.Remove(1); ref != nil { - t.Fatalf("r.Remove(1) for a removed FD: got success, want failure") - } - }) -} - -func TestDescriptorFlags(t *testing.T) { - runTest(t, func(ctx context.Context, fdTable *FDTable, file *fs.File, _ *limits.LimitSet) { - if err := fdTable.NewFDAt(ctx, 2, file, FDFlags{CloseOnExec: true}); err != nil { - t.Fatalf("fdTable.NewFDAt(2, r, FDFlags{}): got %v, wanted nil", err) - } - - newFile, flags := fdTable.Get(2) - if newFile == nil { - t.Fatalf("fdTable.Get(2): got a %v, wanted nil", newFile) - } - - if !flags.CloseOnExec { - t.Fatalf("new File flags %v don't match original %d\n", flags, 0) - } - }) -} - -func BenchmarkFDLookupAndDecRef(b *testing.B) { - b.StopTimer() // Setup. - - runTest(b, func(ctx context.Context, fdTable *FDTable, file *fs.File, _ *limits.LimitSet) { - fds, err := fdTable.NewFDs(ctx, 0, []*fs.File{file, file, file, file, file}, FDFlags{}) - if err != nil { - b.Fatalf("fdTable.NewFDs: got %v, wanted nil", err) - } - - b.StartTimer() // Benchmark. - for i := 0; i < b.N; i++ { - tf, _ := fdTable.Get(fds[i%len(fds)]) - tf.DecRef() - } - }) -} - -func BenchmarkFDLookupAndDecRefConcurrent(b *testing.B) { - b.StopTimer() // Setup. - - runTest(b, func(ctx context.Context, fdTable *FDTable, file *fs.File, _ *limits.LimitSet) { - fds, err := fdTable.NewFDs(ctx, 0, []*fs.File{file, file, file, file, file}, FDFlags{}) - if err != nil { - b.Fatalf("fdTable.NewFDs: got %v, wanted nil", err) - } - - concurrency := runtime.GOMAXPROCS(0) - if concurrency < 4 { - concurrency = 4 - } - each := b.N / concurrency - - b.StartTimer() // Benchmark. - var wg sync.WaitGroup - for i := 0; i < concurrency; i++ { - wg.Add(1) - go func() { - defer wg.Done() - for i := 0; i < each; i++ { - tf, _ := fdTable.Get(fds[i%len(fds)]) - tf.DecRef() - } - }() - } - wg.Wait() - }) -} diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD deleted file mode 100644 index c5021f2db..000000000 --- a/pkg/sentry/kernel/futex/BUILD +++ /dev/null @@ -1,57 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "atomicptr_bucket", - out = "atomicptr_bucket_unsafe.go", - package = "futex", - suffix = "Bucket", - template = "//pkg/sync:generic_atomicptr", - types = { - "Value": "bucket", - }, -) - -go_template_instance( - name = "waiter_list", - out = "waiter_list.go", - package = "futex", - prefix = "waiter", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*Waiter", - "Linker": "*Waiter", - }, -) - -go_library( - name = "futex", - srcs = [ - "atomicptr_bucket_unsafe.go", - "futex.go", - "waiter_list.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/log", - "//pkg/sentry/memmap", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -go_test( - name = "futex_test", - size = "small", - srcs = ["futex_test.go"], - library = ":futex", - deps = [ - "//pkg/sync", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/kernel/futex/atomicptr_bucket_unsafe.go b/pkg/sentry/kernel/futex/atomicptr_bucket_unsafe.go new file mode 100755 index 000000000..d3fdf09b0 --- /dev/null +++ b/pkg/sentry/kernel/futex/atomicptr_bucket_unsafe.go @@ -0,0 +1,37 @@ +package futex + +import ( + "sync/atomic" + "unsafe" +) + +// An AtomicPtr is a pointer to a value of type Value that can be atomically +// loaded and stored. The zero value of an AtomicPtr represents nil. +// +// Note that copying AtomicPtr by value performs a non-atomic read of the +// stored pointer, which is unsafe if Store() can be called concurrently; in +// this case, do `dst.Store(src.Load())` instead. +// +// +stateify savable +type AtomicPtrBucket struct { + ptr unsafe.Pointer `state:".(*bucket)"` +} + +func (p *AtomicPtrBucket) savePtr() *bucket { + return p.Load() +} + +func (p *AtomicPtrBucket) loadPtr(v *bucket) { + p.Store(v) +} + +// Load returns the value set by the most recent Store. It returns nil if there +// has been no previous call to Store. +func (p *AtomicPtrBucket) Load() *bucket { + return (*bucket)(atomic.LoadPointer(&p.ptr)) +} + +// Store sets the value returned by Load to x. +func (p *AtomicPtrBucket) Store(x *bucket) { + atomic.StorePointer(&p.ptr, (unsafe.Pointer)(x)) +} diff --git a/pkg/sentry/kernel/futex/futex_state_autogen.go b/pkg/sentry/kernel/futex/futex_state_autogen.go new file mode 100755 index 000000000..d5ed3466f --- /dev/null +++ b/pkg/sentry/kernel/futex/futex_state_autogen.go @@ -0,0 +1,79 @@ +// automatically generated by stateify. + +package futex + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *AtomicPtrBucket) beforeSave() {} +func (x *AtomicPtrBucket) save(m state.Map) { + x.beforeSave() + var ptr *bucket = x.savePtr() + m.SaveValue("ptr", ptr) +} + +func (x *AtomicPtrBucket) afterLoad() {} +func (x *AtomicPtrBucket) load(m state.Map) { + m.LoadValue("ptr", new(*bucket), func(y interface{}) { x.loadPtr(y.(*bucket)) }) +} + +func (x *bucket) beforeSave() {} +func (x *bucket) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.waiters) { + m.Failf("waiters is %v, expected zero", x.waiters) + } +} + +func (x *bucket) afterLoad() {} +func (x *bucket) load(m state.Map) { +} + +func (x *Manager) beforeSave() {} +func (x *Manager) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.privateBuckets) { + m.Failf("privateBuckets is %v, expected zero", x.privateBuckets) + } + m.Save("sharedBucket", &x.sharedBucket) +} + +func (x *Manager) afterLoad() {} +func (x *Manager) load(m state.Map) { + m.Load("sharedBucket", &x.sharedBucket) +} + +func (x *waiterList) beforeSave() {} +func (x *waiterList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *waiterList) afterLoad() {} +func (x *waiterList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *waiterEntry) beforeSave() {} +func (x *waiterEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *waiterEntry) afterLoad() {} +func (x *waiterEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func init() { + state.Register("pkg/sentry/kernel/futex.AtomicPtrBucket", (*AtomicPtrBucket)(nil), state.Fns{Save: (*AtomicPtrBucket).save, Load: (*AtomicPtrBucket).load}) + state.Register("pkg/sentry/kernel/futex.bucket", (*bucket)(nil), state.Fns{Save: (*bucket).save, Load: (*bucket).load}) + state.Register("pkg/sentry/kernel/futex.Manager", (*Manager)(nil), state.Fns{Save: (*Manager).save, Load: (*Manager).load}) + state.Register("pkg/sentry/kernel/futex.waiterList", (*waiterList)(nil), state.Fns{Save: (*waiterList).save, Load: (*waiterList).load}) + state.Register("pkg/sentry/kernel/futex.waiterEntry", (*waiterEntry)(nil), state.Fns{Save: (*waiterEntry).save, Load: (*waiterEntry).load}) +} diff --git a/pkg/sentry/kernel/futex/futex_test.go b/pkg/sentry/kernel/futex/futex_test.go deleted file mode 100644 index 7c5c7665b..000000000 --- a/pkg/sentry/kernel/futex/futex_test.go +++ /dev/null @@ -1,530 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package futex - -import ( - "math" - "runtime" - "sync/atomic" - "syscall" - "testing" - "unsafe" - - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/usermem" -) - -// testData implements the Target interface, and allows us to -// treat the address passed for futex operations as an index in -// a byte slice for testing simplicity. -type testData []byte - -const sizeofInt32 = 4 - -func newTestData(size uint) testData { - return make([]byte, size) -} - -func (t testData) SwapUint32(addr usermem.Addr, new uint32) (uint32, error) { - val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t[addr])), new) - return val, nil -} - -func (t testData) CompareAndSwapUint32(addr usermem.Addr, old, new uint32) (uint32, error) { - if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t[addr])), old, new) { - return old, nil - } - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil -} - -func (t testData) LoadUint32(addr usermem.Addr) (uint32, error) { - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil -} - -func (t testData) GetSharedKey(addr usermem.Addr) (Key, error) { - return Key{ - Kind: KindSharedMappable, - Offset: uint64(addr), - }, nil -} - -func futexKind(private bool) string { - if private { - return "private" - } - return "shared" -} - -func newPreparedTestWaiter(t *testing.T, m *Manager, ta Target, addr usermem.Addr, private bool, val uint32, bitmask uint32) *Waiter { - w := NewWaiter() - if err := m.WaitPrepare(w, ta, addr, private, val, bitmask); err != nil { - t.Fatalf("WaitPrepare failed: %v", err) - } - return w -} - -func TestFutexWake(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(sizeofInt32) - - // Start waiting for wakeup. - w := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w) - - // Perform a wakeup. - if n, err := m.Wake(d, 0, private, ^uint32(0), 1); err != nil || n != 1 { - t.Errorf("Wake: got (%d, %v), wanted (1, nil)", n, err) - } - - // Expect the waiter to have been woken. - if !w.woken() { - t.Error("waiter not woken") - } - }) - } -} - -func TestFutexWakeBitmask(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(sizeofInt32) - - // Start waiting for wakeup. - w := newPreparedTestWaiter(t, m, d, 0, private, 0, 0x0000ffff) - defer m.WaitComplete(w) - - // Perform a wakeup using the wrong bitmask. - if n, err := m.Wake(d, 0, private, 0xffff0000, 1); err != nil || n != 0 { - t.Errorf("Wake with non-matching bitmask: got (%d, %v), wanted (0, nil)", n, err) - } - - // Expect the waiter to still be waiting. - if w.woken() { - t.Error("waiter woken unexpectedly") - } - - // Perform a wakeup using the right bitmask. - if n, err := m.Wake(d, 0, private, 0x00000001, 1); err != nil || n != 1 { - t.Errorf("Wake with matching bitmask: got (%d, %v), wanted (1, nil)", n, err) - } - - // Expect that the waiter was woken. - if !w.woken() { - t.Error("waiter not woken") - } - }) - } -} - -func TestFutexWakeTwo(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(sizeofInt32) - - // Start three waiters waiting for wakeup. - var ws [3]*Waiter - for i := range ws { - ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) - } - - // Perform two wakeups. - const wakeups = 2 - if n, err := m.Wake(d, 0, private, ^uint32(0), 2); err != nil || n != wakeups { - t.Errorf("Wake: got (%d, %v), wanted (%d, nil)", n, err, wakeups) - } - - // Expect that exactly two waiters were woken. - // We don't get guarantees about exactly which two, - // (although we expect them to be w1 and w2). - awake := 0 - for i := range ws { - if ws[i].woken() { - awake++ - } - } - if awake != wakeups { - t.Errorf("got %d woken waiters, wanted %d", awake, wakeups) - } - }) - } -} - -func TestFutexWakeUnrelated(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(2 * sizeofInt32) - - // Start two waiters waiting for wakeup on different addresses. - w1 := newPreparedTestWaiter(t, m, d, 0*sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) - w2 := newPreparedTestWaiter(t, m, d, 1*sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) - - // Perform two wakeups on the second address. - if n, err := m.Wake(d, 1*sizeofInt32, private, ^uint32(0), 2); err != nil || n != 1 { - t.Errorf("Wake: got (%d, %v), wanted (1, nil)", n, err) - } - - // Expect that only the second waiter was woken. - if w1.woken() { - t.Error("w1 woken unexpectedly") - } - if !w2.woken() { - t.Error("w2 not woken") - } - }) - } -} - -func TestWakeOpEmpty(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(2 * sizeofInt32) - - // Perform wakeups with no waiters. - if n, err := m.WakeOp(d, 0, sizeofInt32, private, 10, 10, 0); err != nil || n != 0 { - t.Fatalf("WakeOp: got (%d, %v), wanted (0, nil)", n, err) - } - }) - } -} - -func TestWakeOpFirstNonEmpty(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(8) - - // Add two waiters on address 0. - w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) - w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) - - // Perform 10 wakeups on address 0. - if n, err := m.WakeOp(d, 0, sizeofInt32, private, 10, 0, 0); err != nil || n != 2 { - t.Errorf("WakeOp: got (%d, %v), wanted (2, nil)", n, err) - } - - // Expect that both waiters were woken. - if !w1.woken() { - t.Error("w1 not woken") - } - if !w2.woken() { - t.Error("w2 not woken") - } - }) - } -} - -func TestWakeOpSecondNonEmpty(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(8) - - // Add two waiters on address sizeofInt32. - w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) - w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) - - // Perform 10 wakeups on address sizeofInt32 (contingent on - // d.Op(0), which should succeed). - if n, err := m.WakeOp(d, 0, sizeofInt32, private, 0, 10, 0); err != nil || n != 2 { - t.Errorf("WakeOp: got (%d, %v), wanted (2, nil)", n, err) - } - - // Expect that both waiters were woken. - if !w1.woken() { - t.Error("w1 not woken") - } - if !w2.woken() { - t.Error("w2 not woken") - } - }) - } -} - -func TestWakeOpSecondNonEmptyFailingOp(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(8) - - // Add two waiters on address sizeofInt32. - w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) - w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) - - // Perform 10 wakeups on address sizeofInt32 (contingent on - // d.Op(1), which should fail). - if n, err := m.WakeOp(d, 0, sizeofInt32, private, 0, 10, 1); err != nil || n != 0 { - t.Errorf("WakeOp: got (%d, %v), wanted (0, nil)", n, err) - } - - // Expect that neither waiter was woken. - if w1.woken() { - t.Error("w1 woken unexpectedly") - } - if w2.woken() { - t.Error("w2 woken unexpectedly") - } - }) - } -} - -func TestWakeOpAllNonEmpty(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(8) - - // Add two waiters on address 0. - w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) - w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) - - // Add two waiters on address sizeofInt32. - w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w3) - w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w4) - - // Perform 10 wakeups on address 0 (unconditionally), and 10 - // wakeups on address sizeofInt32 (contingent on d.Op(0), which - // should succeed). - if n, err := m.WakeOp(d, 0, sizeofInt32, private, 10, 10, 0); err != nil || n != 4 { - t.Errorf("WakeOp: got (%d, %v), wanted (4, nil)", n, err) - } - - // Expect that all waiters were woken. - if !w1.woken() { - t.Error("w1 not woken") - } - if !w2.woken() { - t.Error("w2 not woken") - } - if !w3.woken() { - t.Error("w3 not woken") - } - if !w4.woken() { - t.Error("w4 not woken") - } - }) - } -} - -func TestWakeOpAllNonEmptyFailingOp(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(8) - - // Add two waiters on address 0. - w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) - w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) - - // Add two waiters on address sizeofInt32. - w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w3) - w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w4) - - // Perform 10 wakeups on address 0 (unconditionally), and 10 - // wakeups on address sizeofInt32 (contingent on d.Op(1), which - // should fail). - if n, err := m.WakeOp(d, 0, sizeofInt32, private, 10, 10, 1); err != nil || n != 2 { - t.Errorf("WakeOp: got (%d, %v), wanted (2, nil)", n, err) - } - - // Expect that only the first two waiters were woken. - if !w1.woken() { - t.Error("w1 not woken") - } - if !w2.woken() { - t.Error("w2 not woken") - } - if w3.woken() { - t.Error("w3 woken unexpectedly") - } - if w4.woken() { - t.Error("w4 woken unexpectedly") - } - }) - } -} - -func TestWakeOpSameAddress(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(8) - - // Add four waiters on address 0. - var ws [4]*Waiter - for i := range ws { - ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) - } - - // Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup - // on address 0 (contingent on d.Op(0), which should succeed). - const wakeups = 2 - if n, err := m.WakeOp(d, 0, 0, private, 1, 1, 0); err != nil || n != wakeups { - t.Errorf("WakeOp: got (%d, %v), wanted (%d, nil)", n, err, wakeups) - } - - // Expect that exactly two waiters were woken. - awake := 0 - for i := range ws { - if ws[i].woken() { - awake++ - } - } - if awake != wakeups { - t.Errorf("got %d woken waiters, wanted %d", awake, wakeups) - } - }) - } -} - -func TestWakeOpSameAddressFailingOp(t *testing.T) { - for _, private := range []bool{false, true} { - t.Run(futexKind(private), func(t *testing.T) { - m := NewManager() - d := newTestData(8) - - // Add four waiters on address 0. - var ws [4]*Waiter - for i := range ws { - ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) - } - - // Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup - // on address 0 (contingent on d.Op(1), which should fail). - const wakeups = 1 - if n, err := m.WakeOp(d, 0, 0, private, 1, 1, 1); err != nil || n != wakeups { - t.Errorf("WakeOp: got (%d, %v), wanted (%d, nil)", n, err, wakeups) - } - - // Expect that exactly one waiter was woken. - awake := 0 - for i := range ws { - if ws[i].woken() { - awake++ - } - } - if awake != wakeups { - t.Errorf("got %d woken waiters, wanted %d", awake, wakeups) - } - }) - } -} - -const ( - testMutexSize = sizeofInt32 - testMutexLocked uint32 = 1 - testMutexUnlocked uint32 = 0 -) - -// testMutex ties together a testData slice, an address, and a -// futex manager in order to implement the sync.Locker interface. -// Beyond being used as a Locker, this is a simple mechanism for -// changing the underlying values for simpler tests. -type testMutex struct { - a usermem.Addr - d testData - m *Manager -} - -func newTestMutex(addr usermem.Addr, d testData, m *Manager) *testMutex { - return &testMutex{a: addr, d: d, m: m} -} - -// Lock acquires the testMutex. -// This may wait for it to be available via the futex manager. -func (t *testMutex) Lock() { - for { - // Attempt to grab the lock. - if atomic.CompareAndSwapUint32( - (*uint32)(unsafe.Pointer(&t.d[t.a])), - testMutexUnlocked, - testMutexLocked) { - // Lock held. - return - } - - // Wait for it to be "not locked". - w := NewWaiter() - err := t.m.WaitPrepare(w, t.d, t.a, true, testMutexLocked, ^uint32(0)) - if err == syscall.EAGAIN { - continue - } - if err != nil { - // Should never happen. - panic("WaitPrepare returned unexpected error: " + err.Error()) - } - <-w.C - t.m.WaitComplete(w) - } -} - -// Unlock releases the testMutex. -// This will notify any waiters via the futex manager. -func (t *testMutex) Unlock() { - // Unlock. - atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d[t.a])), testMutexUnlocked) - - // Notify all waiters. - t.m.Wake(t.d, t.a, true, ^uint32(0), math.MaxInt32) -} - -// This function was shamelessly stolen from mutex_test.go. -func HammerMutex(l sync.Locker, loops int, cdone chan bool) { - for i := 0; i < loops; i++ { - l.Lock() - runtime.Gosched() - l.Unlock() - } - cdone <- true -} - -func TestMutexStress(t *testing.T) { - m := NewManager() - d := newTestData(testMutexSize) - tm := newTestMutex(0*testMutexSize, d, m) - c := make(chan bool) - - for i := 0; i < 10; i++ { - go HammerMutex(tm, 1000, c) - } - - for i := 0; i < 10; i++ { - <-c - } -} diff --git a/pkg/sentry/kernel/futex/waiter_list.go b/pkg/sentry/kernel/futex/waiter_list.go new file mode 100755 index 000000000..204eededf --- /dev/null +++ b/pkg/sentry/kernel/futex/waiter_list.go @@ -0,0 +1,186 @@ +package futex + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type waiterElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (waiterElementMapper) linkerFor(elem *Waiter) *Waiter { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type waiterList struct { + head *Waiter + tail *Waiter +} + +// Reset resets list l to the empty state. +func (l *waiterList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *waiterList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *waiterList) Front() *Waiter { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *waiterList) Back() *Waiter { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *waiterList) PushFront(e *Waiter) { + linker := waiterElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + waiterElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *waiterList) PushBack(e *Waiter) { + linker := waiterElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + waiterElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *waiterList) PushBackList(m *waiterList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + waiterElementMapper{}.linkerFor(l.tail).SetNext(m.head) + waiterElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *waiterList) InsertAfter(b, e *Waiter) { + bLinker := waiterElementMapper{}.linkerFor(b) + eLinker := waiterElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + waiterElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *waiterList) InsertBefore(a, e *Waiter) { + aLinker := waiterElementMapper{}.linkerFor(a) + eLinker := waiterElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + waiterElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *waiterList) Remove(e *Waiter) { + linker := waiterElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + waiterElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + waiterElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type waiterEntry struct { + next *Waiter + prev *Waiter +} + +// Next returns the entry that follows e in the list. +func (e *waiterEntry) Next() *Waiter { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *waiterEntry) Prev() *Waiter { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *waiterEntry) SetNext(elem *Waiter) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *waiterEntry) SetPrev(elem *Waiter) { + e.prev = elem +} diff --git a/pkg/sentry/kernel/g3doc/run_states.dot b/pkg/sentry/kernel/g3doc/run_states.dot deleted file mode 100644 index 7861fe1f5..000000000 --- a/pkg/sentry/kernel/g3doc/run_states.dot +++ /dev/null @@ -1,99 +0,0 @@ -digraph { - subgraph { - App; - } - subgraph { - Interrupt; - InterruptAfterSignalDeliveryStop; - } - subgraph { - Syscall; - SyscallAfterPtraceEventSeccomp; - SyscallEnter; - SyscallAfterSyscallEnterStop; - SyscallAfterSysemuStop; - SyscallInvoke; - SyscallAfterPtraceEventClone; - SyscallAfterExecStop; - SyscallAfterVforkStop; - SyscallReinvoke; - SyscallExit; - } - subgraph { - Vsyscall; - VsyscallAfterPtraceEventSeccomp; - VsyscallInvoke; - } - subgraph { - Exit; - ExitMain; // leave thread group, release resources, reparent children, kill PID namespace and wait if TGID 1 - ExitNotify; // signal parent/tracer, become waitable - ExitDone; // represented by t.runState == nil - } - - // Task exit - Exit -> ExitMain; - ExitMain -> ExitNotify; - ExitNotify -> ExitDone; - - // Execution of untrusted application code - App -> App; - - // Interrupts (usually signal delivery) - App -> Interrupt; - Interrupt -> Interrupt; // if other interrupt conditions may still apply - Interrupt -> Exit; // if killed - - // Syscalls - App -> Syscall; - Syscall -> SyscallEnter; - SyscallEnter -> SyscallInvoke; - SyscallInvoke -> SyscallExit; - SyscallExit -> App; - - // exit, exit_group - SyscallInvoke -> Exit; - - // execve - SyscallInvoke -> SyscallAfterExecStop; - SyscallAfterExecStop -> SyscallExit; - SyscallAfterExecStop -> App; // fatal signal pending - - // vfork - SyscallInvoke -> SyscallAfterVforkStop; - SyscallAfterVforkStop -> SyscallExit; - - // Vsyscalls - App -> Vsyscall; - Vsyscall -> VsyscallInvoke; - Vsyscall -> App; // fault while reading return address from stack - VsyscallInvoke -> App; - - // ptrace-specific branches - Interrupt -> InterruptAfterSignalDeliveryStop; - InterruptAfterSignalDeliveryStop -> Interrupt; - SyscallEnter -> SyscallAfterSyscallEnterStop; - SyscallAfterSyscallEnterStop -> SyscallInvoke; - SyscallAfterSyscallEnterStop -> SyscallExit; // skipped by tracer - SyscallAfterSyscallEnterStop -> App; // fatal signal pending - SyscallEnter -> SyscallAfterSysemuStop; - SyscallAfterSysemuStop -> SyscallExit; - SyscallAfterSysemuStop -> App; // fatal signal pending - SyscallInvoke -> SyscallAfterPtraceEventClone; - SyscallAfterPtraceEventClone -> SyscallExit; - SyscallAfterPtraceEventClone -> SyscallAfterVforkStop; - - // seccomp - Syscall -> App; // SECCOMP_RET_TRAP, SECCOMP_RET_ERRNO, SECCOMP_RET_KILL, SECCOMP_RET_TRACE without tracer - Syscall -> SyscallAfterPtraceEventSeccomp; // SECCOMP_RET_TRACE - SyscallAfterPtraceEventSeccomp -> SyscallEnter; - SyscallAfterPtraceEventSeccomp -> SyscallExit; // skipped by tracer - SyscallAfterPtraceEventSeccomp -> App; // fatal signal pending - Vsyscall -> VsyscallAfterPtraceEventSeccomp; - VsyscallAfterPtraceEventSeccomp -> VsyscallInvoke; - VsyscallAfterPtraceEventSeccomp -> App; - - // Autosave - SyscallInvoke -> SyscallReinvoke; - SyscallReinvoke -> SyscallInvoke; -} diff --git a/pkg/sentry/kernel/g3doc/run_states.png b/pkg/sentry/kernel/g3doc/run_states.png Binary files differdeleted file mode 100644 index b63b60f02..000000000 --- a/pkg/sentry/kernel/g3doc/run_states.png +++ /dev/null diff --git a/pkg/sentry/kernel/kernel_amd64_state_autogen.go b/pkg/sentry/kernel/kernel_amd64_state_autogen.go new file mode 100755 index 000000000..12de47ad0 --- /dev/null +++ b/pkg/sentry/kernel/kernel_amd64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build amd64 + +package kernel diff --git a/pkg/sentry/kernel/kernel_arm64_state_autogen.go b/pkg/sentry/kernel/kernel_arm64_state_autogen.go new file mode 100755 index 000000000..3c040d283 --- /dev/null +++ b/pkg/sentry/kernel/kernel_arm64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build arm64 + +package kernel diff --git a/pkg/sentry/kernel/kernel_opts.go b/pkg/sentry/kernel/kernel_opts.go index 2e66ec587..2e66ec587 100644..100755 --- a/pkg/sentry/kernel/kernel_opts.go +++ b/pkg/sentry/kernel/kernel_opts.go diff --git a/pkg/sentry/kernel/kernel_opts_state_autogen.go b/pkg/sentry/kernel/kernel_opts_state_autogen.go new file mode 100755 index 000000000..9ed7e27c9 --- /dev/null +++ b/pkg/sentry/kernel/kernel_opts_state_autogen.go @@ -0,0 +1,20 @@ +// automatically generated by stateify. + +package kernel + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *SpecialOpts) beforeSave() {} +func (x *SpecialOpts) save(m state.Map) { + x.beforeSave() +} + +func (x *SpecialOpts) afterLoad() {} +func (x *SpecialOpts) load(m state.Map) { +} + +func init() { + state.Register("pkg/sentry/kernel.SpecialOpts", (*SpecialOpts)(nil), state.Fns{Save: (*SpecialOpts).save, Load: (*SpecialOpts).load}) +} diff --git a/pkg/sentry/kernel/kernel_state_autogen.go b/pkg/sentry/kernel/kernel_state_autogen.go new file mode 100755 index 000000000..57a261086 --- /dev/null +++ b/pkg/sentry/kernel/kernel_state_autogen.go @@ -0,0 +1,1230 @@ +// automatically generated by stateify. + +package kernel + +import ( + "gvisor.dev/gvisor/pkg/bpf" + "gvisor.dev/gvisor/pkg/sentry/device" + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/tcpip" +) + +func (x *abstractEndpoint) beforeSave() {} +func (x *abstractEndpoint) save(m state.Map) { + x.beforeSave() + m.Save("ep", &x.ep) + m.Save("wr", &x.wr) + m.Save("name", &x.name) + m.Save("ns", &x.ns) +} + +func (x *abstractEndpoint) afterLoad() {} +func (x *abstractEndpoint) load(m state.Map) { + m.Load("ep", &x.ep) + m.Load("wr", &x.wr) + m.Load("name", &x.name) + m.Load("ns", &x.ns) +} + +func (x *AbstractSocketNamespace) beforeSave() {} +func (x *AbstractSocketNamespace) save(m state.Map) { + x.beforeSave() + m.Save("endpoints", &x.endpoints) +} + +func (x *AbstractSocketNamespace) afterLoad() {} +func (x *AbstractSocketNamespace) load(m state.Map) { + m.Load("endpoints", &x.endpoints) +} + +func (x *FDFlags) beforeSave() {} +func (x *FDFlags) save(m state.Map) { + x.beforeSave() + m.Save("CloseOnExec", &x.CloseOnExec) +} + +func (x *FDFlags) afterLoad() {} +func (x *FDFlags) load(m state.Map) { + m.Load("CloseOnExec", &x.CloseOnExec) +} + +func (x *descriptor) beforeSave() {} +func (x *descriptor) save(m state.Map) { + x.beforeSave() + m.Save("file", &x.file) + m.Save("fileVFS2", &x.fileVFS2) + m.Save("flags", &x.flags) +} + +func (x *descriptor) afterLoad() {} +func (x *descriptor) load(m state.Map) { + m.Load("file", &x.file) + m.Load("fileVFS2", &x.fileVFS2) + m.Load("flags", &x.flags) +} + +func (x *FDTable) beforeSave() {} +func (x *FDTable) save(m state.Map) { + x.beforeSave() + var descriptorTable map[int32]descriptor = x.saveDescriptorTable() + m.SaveValue("descriptorTable", descriptorTable) + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("k", &x.k) + m.Save("uid", &x.uid) + m.Save("next", &x.next) + m.Save("used", &x.used) +} + +func (x *FDTable) afterLoad() {} +func (x *FDTable) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("k", &x.k) + m.Load("uid", &x.uid) + m.Load("next", &x.next) + m.Load("used", &x.used) + m.LoadValue("descriptorTable", new(map[int32]descriptor), func(y interface{}) { x.loadDescriptorTable(y.(map[int32]descriptor)) }) +} + +func (x *FSContext) beforeSave() {} +func (x *FSContext) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("root", &x.root) + m.Save("rootVFS2", &x.rootVFS2) + m.Save("cwd", &x.cwd) + m.Save("cwdVFS2", &x.cwdVFS2) + m.Save("umask", &x.umask) +} + +func (x *FSContext) afterLoad() {} +func (x *FSContext) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("root", &x.root) + m.Load("rootVFS2", &x.rootVFS2) + m.Load("cwd", &x.cwd) + m.Load("cwdVFS2", &x.cwdVFS2) + m.Load("umask", &x.umask) +} + +func (x *IPCNamespace) beforeSave() {} +func (x *IPCNamespace) save(m state.Map) { + x.beforeSave() + m.Save("userNS", &x.userNS) + m.Save("semaphores", &x.semaphores) + m.Save("shms", &x.shms) +} + +func (x *IPCNamespace) afterLoad() {} +func (x *IPCNamespace) load(m state.Map) { + m.Load("userNS", &x.userNS) + m.Load("semaphores", &x.semaphores) + m.Load("shms", &x.shms) +} + +func (x *Kernel) beforeSave() {} +func (x *Kernel) save(m state.Map) { + x.beforeSave() + var danglingEndpoints []tcpip.Endpoint = x.saveDanglingEndpoints() + m.SaveValue("danglingEndpoints", danglingEndpoints) + var deviceRegistry *device.Registry = x.saveDeviceRegistry() + m.SaveValue("deviceRegistry", deviceRegistry) + m.Save("featureSet", &x.featureSet) + m.Save("timekeeper", &x.timekeeper) + m.Save("tasks", &x.tasks) + m.Save("rootUserNamespace", &x.rootUserNamespace) + m.Save("rootNetworkNamespace", &x.rootNetworkNamespace) + m.Save("applicationCores", &x.applicationCores) + m.Save("useHostCores", &x.useHostCores) + m.Save("extraAuxv", &x.extraAuxv) + m.Save("vdso", &x.vdso) + m.Save("rootUTSNamespace", &x.rootUTSNamespace) + m.Save("rootIPCNamespace", &x.rootIPCNamespace) + m.Save("rootAbstractSocketNamespace", &x.rootAbstractSocketNamespace) + m.Save("futexes", &x.futexes) + m.Save("globalInit", &x.globalInit) + m.Save("realtimeClock", &x.realtimeClock) + m.Save("monotonicClock", &x.monotonicClock) + m.Save("syslog", &x.syslog) + m.Save("runningTasks", &x.runningTasks) + m.Save("cpuClock", &x.cpuClock) + m.Save("cpuClockTickerDisabled", &x.cpuClockTickerDisabled) + m.Save("cpuClockTickerSetting", &x.cpuClockTickerSetting) + m.Save("fdMapUids", &x.fdMapUids) + m.Save("uniqueID", &x.uniqueID) + m.Save("nextInotifyCookie", &x.nextInotifyCookie) + m.Save("netlinkPorts", &x.netlinkPorts) + m.Save("sockets", &x.sockets) + m.Save("nextSocketEntry", &x.nextSocketEntry) + m.Save("DirentCacheLimiter", &x.DirentCacheLimiter) + m.Save("SpecialOpts", &x.SpecialOpts) + m.Save("vfs", &x.vfs) + m.Save("SleepForAddressSpaceActivation", &x.SleepForAddressSpaceActivation) +} + +func (x *Kernel) afterLoad() {} +func (x *Kernel) load(m state.Map) { + m.Load("featureSet", &x.featureSet) + m.Load("timekeeper", &x.timekeeper) + m.Load("tasks", &x.tasks) + m.Load("rootUserNamespace", &x.rootUserNamespace) + m.Load("rootNetworkNamespace", &x.rootNetworkNamespace) + m.Load("applicationCores", &x.applicationCores) + m.Load("useHostCores", &x.useHostCores) + m.Load("extraAuxv", &x.extraAuxv) + m.Load("vdso", &x.vdso) + m.Load("rootUTSNamespace", &x.rootUTSNamespace) + m.Load("rootIPCNamespace", &x.rootIPCNamespace) + m.Load("rootAbstractSocketNamespace", &x.rootAbstractSocketNamespace) + m.Load("futexes", &x.futexes) + m.Load("globalInit", &x.globalInit) + m.Load("realtimeClock", &x.realtimeClock) + m.Load("monotonicClock", &x.monotonicClock) + m.Load("syslog", &x.syslog) + m.Load("runningTasks", &x.runningTasks) + m.Load("cpuClock", &x.cpuClock) + m.Load("cpuClockTickerDisabled", &x.cpuClockTickerDisabled) + m.Load("cpuClockTickerSetting", &x.cpuClockTickerSetting) + m.Load("fdMapUids", &x.fdMapUids) + m.Load("uniqueID", &x.uniqueID) + m.Load("nextInotifyCookie", &x.nextInotifyCookie) + m.Load("netlinkPorts", &x.netlinkPorts) + m.Load("sockets", &x.sockets) + m.Load("nextSocketEntry", &x.nextSocketEntry) + m.Load("DirentCacheLimiter", &x.DirentCacheLimiter) + m.Load("SpecialOpts", &x.SpecialOpts) + m.Load("vfs", &x.vfs) + m.Load("SleepForAddressSpaceActivation", &x.SleepForAddressSpaceActivation) + m.LoadValue("danglingEndpoints", new([]tcpip.Endpoint), func(y interface{}) { x.loadDanglingEndpoints(y.([]tcpip.Endpoint)) }) + m.LoadValue("deviceRegistry", new(*device.Registry), func(y interface{}) { x.loadDeviceRegistry(y.(*device.Registry)) }) +} + +func (x *SocketEntry) beforeSave() {} +func (x *SocketEntry) save(m state.Map) { + x.beforeSave() + m.Save("socketEntry", &x.socketEntry) + m.Save("k", &x.k) + m.Save("Sock", &x.Sock) + m.Save("ID", &x.ID) +} + +func (x *SocketEntry) afterLoad() {} +func (x *SocketEntry) load(m state.Map) { + m.Load("socketEntry", &x.socketEntry) + m.Load("k", &x.k) + m.Load("Sock", &x.Sock) + m.Load("ID", &x.ID) +} + +func (x *pendingSignals) beforeSave() {} +func (x *pendingSignals) save(m state.Map) { + x.beforeSave() + var signals []savedPendingSignal = x.saveSignals() + m.SaveValue("signals", signals) +} + +func (x *pendingSignals) afterLoad() {} +func (x *pendingSignals) load(m state.Map) { + m.LoadValue("signals", new([]savedPendingSignal), func(y interface{}) { x.loadSignals(y.([]savedPendingSignal)) }) +} + +func (x *pendingSignalQueue) beforeSave() {} +func (x *pendingSignalQueue) save(m state.Map) { + x.beforeSave() + m.Save("pendingSignalList", &x.pendingSignalList) + m.Save("length", &x.length) +} + +func (x *pendingSignalQueue) afterLoad() {} +func (x *pendingSignalQueue) load(m state.Map) { + m.Load("pendingSignalList", &x.pendingSignalList) + m.Load("length", &x.length) +} + +func (x *pendingSignal) beforeSave() {} +func (x *pendingSignal) save(m state.Map) { + x.beforeSave() + m.Save("pendingSignalEntry", &x.pendingSignalEntry) + m.Save("SignalInfo", &x.SignalInfo) + m.Save("timer", &x.timer) +} + +func (x *pendingSignal) afterLoad() {} +func (x *pendingSignal) load(m state.Map) { + m.Load("pendingSignalEntry", &x.pendingSignalEntry) + m.Load("SignalInfo", &x.SignalInfo) + m.Load("timer", &x.timer) +} + +func (x *pendingSignalList) beforeSave() {} +func (x *pendingSignalList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *pendingSignalList) afterLoad() {} +func (x *pendingSignalList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *pendingSignalEntry) beforeSave() {} +func (x *pendingSignalEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *pendingSignalEntry) afterLoad() {} +func (x *pendingSignalEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *savedPendingSignal) beforeSave() {} +func (x *savedPendingSignal) save(m state.Map) { + x.beforeSave() + m.Save("si", &x.si) + m.Save("timer", &x.timer) +} + +func (x *savedPendingSignal) afterLoad() {} +func (x *savedPendingSignal) load(m state.Map) { + m.Load("si", &x.si) + m.Load("timer", &x.timer) +} + +func (x *IntervalTimer) beforeSave() {} +func (x *IntervalTimer) save(m state.Map) { + x.beforeSave() + m.Save("timer", &x.timer) + m.Save("target", &x.target) + m.Save("signo", &x.signo) + m.Save("id", &x.id) + m.Save("sigval", &x.sigval) + m.Save("group", &x.group) + m.Save("sigpending", &x.sigpending) + m.Save("sigorphan", &x.sigorphan) + m.Save("overrunCur", &x.overrunCur) + m.Save("overrunLast", &x.overrunLast) +} + +func (x *IntervalTimer) afterLoad() {} +func (x *IntervalTimer) load(m state.Map) { + m.Load("timer", &x.timer) + m.Load("target", &x.target) + m.Load("signo", &x.signo) + m.Load("id", &x.id) + m.Load("sigval", &x.sigval) + m.Load("group", &x.group) + m.Load("sigpending", &x.sigpending) + m.Load("sigorphan", &x.sigorphan) + m.Load("overrunCur", &x.overrunCur) + m.Load("overrunLast", &x.overrunLast) +} + +func (x *processGroupList) beforeSave() {} +func (x *processGroupList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *processGroupList) afterLoad() {} +func (x *processGroupList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *processGroupEntry) beforeSave() {} +func (x *processGroupEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *processGroupEntry) afterLoad() {} +func (x *processGroupEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *ptraceOptions) beforeSave() {} +func (x *ptraceOptions) save(m state.Map) { + x.beforeSave() + m.Save("ExitKill", &x.ExitKill) + m.Save("SysGood", &x.SysGood) + m.Save("TraceClone", &x.TraceClone) + m.Save("TraceExec", &x.TraceExec) + m.Save("TraceExit", &x.TraceExit) + m.Save("TraceFork", &x.TraceFork) + m.Save("TraceSeccomp", &x.TraceSeccomp) + m.Save("TraceVfork", &x.TraceVfork) + m.Save("TraceVforkDone", &x.TraceVforkDone) +} + +func (x *ptraceOptions) afterLoad() {} +func (x *ptraceOptions) load(m state.Map) { + m.Load("ExitKill", &x.ExitKill) + m.Load("SysGood", &x.SysGood) + m.Load("TraceClone", &x.TraceClone) + m.Load("TraceExec", &x.TraceExec) + m.Load("TraceExit", &x.TraceExit) + m.Load("TraceFork", &x.TraceFork) + m.Load("TraceSeccomp", &x.TraceSeccomp) + m.Load("TraceVfork", &x.TraceVfork) + m.Load("TraceVforkDone", &x.TraceVforkDone) +} + +func (x *ptraceStop) beforeSave() {} +func (x *ptraceStop) save(m state.Map) { + x.beforeSave() + m.Save("frozen", &x.frozen) + m.Save("listen", &x.listen) +} + +func (x *ptraceStop) afterLoad() {} +func (x *ptraceStop) load(m state.Map) { + m.Load("frozen", &x.frozen) + m.Load("listen", &x.listen) +} + +func (x *OldRSeqCriticalRegion) beforeSave() {} +func (x *OldRSeqCriticalRegion) save(m state.Map) { + x.beforeSave() + m.Save("CriticalSection", &x.CriticalSection) + m.Save("Restart", &x.Restart) +} + +func (x *OldRSeqCriticalRegion) afterLoad() {} +func (x *OldRSeqCriticalRegion) load(m state.Map) { + m.Load("CriticalSection", &x.CriticalSection) + m.Load("Restart", &x.Restart) +} + +func (x *sessionList) beforeSave() {} +func (x *sessionList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *sessionList) afterLoad() {} +func (x *sessionList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *sessionEntry) beforeSave() {} +func (x *sessionEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *sessionEntry) afterLoad() {} +func (x *sessionEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *Session) beforeSave() {} +func (x *Session) save(m state.Map) { + x.beforeSave() + m.Save("refs", &x.refs) + m.Save("leader", &x.leader) + m.Save("id", &x.id) + m.Save("foreground", &x.foreground) + m.Save("processGroups", &x.processGroups) + m.Save("sessionEntry", &x.sessionEntry) +} + +func (x *Session) afterLoad() {} +func (x *Session) load(m state.Map) { + m.Load("refs", &x.refs) + m.Load("leader", &x.leader) + m.Load("id", &x.id) + m.Load("foreground", &x.foreground) + m.Load("processGroups", &x.processGroups) + m.Load("sessionEntry", &x.sessionEntry) +} + +func (x *ProcessGroup) beforeSave() {} +func (x *ProcessGroup) save(m state.Map) { + x.beforeSave() + m.Save("refs", &x.refs) + m.Save("originator", &x.originator) + m.Save("id", &x.id) + m.Save("session", &x.session) + m.Save("ancestors", &x.ancestors) + m.Save("processGroupEntry", &x.processGroupEntry) +} + +func (x *ProcessGroup) afterLoad() {} +func (x *ProcessGroup) load(m state.Map) { + m.Load("refs", &x.refs) + m.Load("originator", &x.originator) + m.Load("id", &x.id) + m.Load("session", &x.session) + m.Load("ancestors", &x.ancestors) + m.Load("processGroupEntry", &x.processGroupEntry) +} + +func (x *SignalHandlers) beforeSave() {} +func (x *SignalHandlers) save(m state.Map) { + x.beforeSave() + m.Save("actions", &x.actions) +} + +func (x *SignalHandlers) afterLoad() {} +func (x *SignalHandlers) load(m state.Map) { + m.Load("actions", &x.actions) +} + +func (x *socketList) beforeSave() {} +func (x *socketList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *socketList) afterLoad() {} +func (x *socketList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *socketEntry) beforeSave() {} +func (x *socketEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *socketEntry) afterLoad() {} +func (x *socketEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *SyscallTable) beforeSave() {} +func (x *SyscallTable) save(m state.Map) { + x.beforeSave() + m.Save("OS", &x.OS) + m.Save("Arch", &x.Arch) +} + +func (x *SyscallTable) load(m state.Map) { + m.LoadWait("OS", &x.OS) + m.LoadWait("Arch", &x.Arch) + m.AfterLoad(x.afterLoad) +} + +func (x *syslog) beforeSave() {} +func (x *syslog) save(m state.Map) { + x.beforeSave() + m.Save("msg", &x.msg) +} + +func (x *syslog) afterLoad() {} +func (x *syslog) load(m state.Map) { + m.Load("msg", &x.msg) +} + +func (x *Task) beforeSave() {} +func (x *Task) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.signalQueue) { + m.Failf("signalQueue is %v, expected zero", x.signalQueue) + } + var ptraceTracer *Task = x.savePtraceTracer() + m.SaveValue("ptraceTracer", ptraceTracer) + var syscallFilters []bpf.Program = x.saveSyscallFilters() + m.SaveValue("syscallFilters", syscallFilters) + m.Save("taskNode", &x.taskNode) + m.Save("runState", &x.runState) + m.Save("haveSyscallReturn", &x.haveSyscallReturn) + m.Save("gosched", &x.gosched) + m.Save("yieldCount", &x.yieldCount) + m.Save("pendingSignals", &x.pendingSignals) + m.Save("signalMask", &x.signalMask) + m.Save("realSignalMask", &x.realSignalMask) + m.Save("haveSavedSignalMask", &x.haveSavedSignalMask) + m.Save("savedSignalMask", &x.savedSignalMask) + m.Save("signalStack", &x.signalStack) + m.Save("groupStopPending", &x.groupStopPending) + m.Save("groupStopAcknowledged", &x.groupStopAcknowledged) + m.Save("trapStopPending", &x.trapStopPending) + m.Save("trapNotifyPending", &x.trapNotifyPending) + m.Save("stop", &x.stop) + m.Save("exitStatus", &x.exitStatus) + m.Save("syscallRestartBlock", &x.syscallRestartBlock) + m.Save("k", &x.k) + m.Save("containerID", &x.containerID) + m.Save("tc", &x.tc) + m.Save("fsContext", &x.fsContext) + m.Save("fdTable", &x.fdTable) + m.Save("vforkParent", &x.vforkParent) + m.Save("exitState", &x.exitState) + m.Save("exitTracerNotified", &x.exitTracerNotified) + m.Save("exitTracerAcked", &x.exitTracerAcked) + m.Save("exitParentNotified", &x.exitParentNotified) + m.Save("exitParentAcked", &x.exitParentAcked) + m.Save("ptraceTracees", &x.ptraceTracees) + m.Save("ptraceSeized", &x.ptraceSeized) + m.Save("ptraceOpts", &x.ptraceOpts) + m.Save("ptraceSyscallMode", &x.ptraceSyscallMode) + m.Save("ptraceSinglestep", &x.ptraceSinglestep) + m.Save("ptraceCode", &x.ptraceCode) + m.Save("ptraceSiginfo", &x.ptraceSiginfo) + m.Save("ptraceEventMsg", &x.ptraceEventMsg) + m.Save("ioUsage", &x.ioUsage) + m.Save("creds", &x.creds) + m.Save("utsns", &x.utsns) + m.Save("ipcns", &x.ipcns) + m.Save("abstractSockets", &x.abstractSockets) + m.Save("mountNamespaceVFS2", &x.mountNamespaceVFS2) + m.Save("parentDeathSignal", &x.parentDeathSignal) + m.Save("cleartid", &x.cleartid) + m.Save("allowedCPUMask", &x.allowedCPUMask) + m.Save("cpu", &x.cpu) + m.Save("niceness", &x.niceness) + m.Save("numaPolicy", &x.numaPolicy) + m.Save("numaNodeMask", &x.numaNodeMask) + m.Save("netns", &x.netns) + m.Save("rseqCPU", &x.rseqCPU) + m.Save("oldRSeqCPUAddr", &x.oldRSeqCPUAddr) + m.Save("rseqAddr", &x.rseqAddr) + m.Save("rseqSignature", &x.rseqSignature) + m.Save("startTime", &x.startTime) +} + +func (x *Task) load(m state.Map) { + m.Load("taskNode", &x.taskNode) + m.Load("runState", &x.runState) + m.Load("haveSyscallReturn", &x.haveSyscallReturn) + m.Load("gosched", &x.gosched) + m.Load("yieldCount", &x.yieldCount) + m.Load("pendingSignals", &x.pendingSignals) + m.Load("signalMask", &x.signalMask) + m.Load("realSignalMask", &x.realSignalMask) + m.Load("haveSavedSignalMask", &x.haveSavedSignalMask) + m.Load("savedSignalMask", &x.savedSignalMask) + m.Load("signalStack", &x.signalStack) + m.Load("groupStopPending", &x.groupStopPending) + m.Load("groupStopAcknowledged", &x.groupStopAcknowledged) + m.Load("trapStopPending", &x.trapStopPending) + m.Load("trapNotifyPending", &x.trapNotifyPending) + m.Load("stop", &x.stop) + m.Load("exitStatus", &x.exitStatus) + m.Load("syscallRestartBlock", &x.syscallRestartBlock) + m.Load("k", &x.k) + m.Load("containerID", &x.containerID) + m.Load("tc", &x.tc) + m.Load("fsContext", &x.fsContext) + m.Load("fdTable", &x.fdTable) + m.Load("vforkParent", &x.vforkParent) + m.Load("exitState", &x.exitState) + m.Load("exitTracerNotified", &x.exitTracerNotified) + m.Load("exitTracerAcked", &x.exitTracerAcked) + m.Load("exitParentNotified", &x.exitParentNotified) + m.Load("exitParentAcked", &x.exitParentAcked) + m.Load("ptraceTracees", &x.ptraceTracees) + m.Load("ptraceSeized", &x.ptraceSeized) + m.Load("ptraceOpts", &x.ptraceOpts) + m.Load("ptraceSyscallMode", &x.ptraceSyscallMode) + m.Load("ptraceSinglestep", &x.ptraceSinglestep) + m.Load("ptraceCode", &x.ptraceCode) + m.Load("ptraceSiginfo", &x.ptraceSiginfo) + m.Load("ptraceEventMsg", &x.ptraceEventMsg) + m.Load("ioUsage", &x.ioUsage) + m.Load("creds", &x.creds) + m.Load("utsns", &x.utsns) + m.Load("ipcns", &x.ipcns) + m.Load("abstractSockets", &x.abstractSockets) + m.Load("mountNamespaceVFS2", &x.mountNamespaceVFS2) + m.Load("parentDeathSignal", &x.parentDeathSignal) + m.Load("cleartid", &x.cleartid) + m.Load("allowedCPUMask", &x.allowedCPUMask) + m.Load("cpu", &x.cpu) + m.Load("niceness", &x.niceness) + m.Load("numaPolicy", &x.numaPolicy) + m.Load("numaNodeMask", &x.numaNodeMask) + m.Load("netns", &x.netns) + m.Load("rseqCPU", &x.rseqCPU) + m.Load("oldRSeqCPUAddr", &x.oldRSeqCPUAddr) + m.Load("rseqAddr", &x.rseqAddr) + m.Load("rseqSignature", &x.rseqSignature) + m.Load("startTime", &x.startTime) + m.LoadValue("ptraceTracer", new(*Task), func(y interface{}) { x.loadPtraceTracer(y.(*Task)) }) + m.LoadValue("syscallFilters", new([]bpf.Program), func(y interface{}) { x.loadSyscallFilters(y.([]bpf.Program)) }) + m.AfterLoad(x.afterLoad) +} + +func (x *runSyscallAfterPtraceEventClone) beforeSave() {} +func (x *runSyscallAfterPtraceEventClone) save(m state.Map) { + x.beforeSave() + m.Save("vforkChild", &x.vforkChild) + m.Save("vforkChildTID", &x.vforkChildTID) +} + +func (x *runSyscallAfterPtraceEventClone) afterLoad() {} +func (x *runSyscallAfterPtraceEventClone) load(m state.Map) { + m.Load("vforkChild", &x.vforkChild) + m.Load("vforkChildTID", &x.vforkChildTID) +} + +func (x *runSyscallAfterVforkStop) beforeSave() {} +func (x *runSyscallAfterVforkStop) save(m state.Map) { + x.beforeSave() + m.Save("childTID", &x.childTID) +} + +func (x *runSyscallAfterVforkStop) afterLoad() {} +func (x *runSyscallAfterVforkStop) load(m state.Map) { + m.Load("childTID", &x.childTID) +} + +func (x *vforkStop) beforeSave() {} +func (x *vforkStop) save(m state.Map) { + x.beforeSave() +} + +func (x *vforkStop) afterLoad() {} +func (x *vforkStop) load(m state.Map) { +} + +func (x *TaskContext) beforeSave() {} +func (x *TaskContext) save(m state.Map) { + x.beforeSave() + m.Save("Name", &x.Name) + m.Save("Arch", &x.Arch) + m.Save("MemoryManager", &x.MemoryManager) + m.Save("fu", &x.fu) + m.Save("st", &x.st) +} + +func (x *TaskContext) afterLoad() {} +func (x *TaskContext) load(m state.Map) { + m.Load("Name", &x.Name) + m.Load("Arch", &x.Arch) + m.Load("MemoryManager", &x.MemoryManager) + m.Load("fu", &x.fu) + m.Load("st", &x.st) +} + +func (x *execStop) beforeSave() {} +func (x *execStop) save(m state.Map) { + x.beforeSave() +} + +func (x *execStop) afterLoad() {} +func (x *execStop) load(m state.Map) { +} + +func (x *runSyscallAfterExecStop) beforeSave() {} +func (x *runSyscallAfterExecStop) save(m state.Map) { + x.beforeSave() + m.Save("tc", &x.tc) +} + +func (x *runSyscallAfterExecStop) afterLoad() {} +func (x *runSyscallAfterExecStop) load(m state.Map) { + m.Load("tc", &x.tc) +} + +func (x *ExitStatus) beforeSave() {} +func (x *ExitStatus) save(m state.Map) { + x.beforeSave() + m.Save("Code", &x.Code) + m.Save("Signo", &x.Signo) +} + +func (x *ExitStatus) afterLoad() {} +func (x *ExitStatus) load(m state.Map) { + m.Load("Code", &x.Code) + m.Load("Signo", &x.Signo) +} + +func (x *runExit) beforeSave() {} +func (x *runExit) save(m state.Map) { + x.beforeSave() +} + +func (x *runExit) afterLoad() {} +func (x *runExit) load(m state.Map) { +} + +func (x *runExitMain) beforeSave() {} +func (x *runExitMain) save(m state.Map) { + x.beforeSave() +} + +func (x *runExitMain) afterLoad() {} +func (x *runExitMain) load(m state.Map) { +} + +func (x *runExitNotify) beforeSave() {} +func (x *runExitNotify) save(m state.Map) { + x.beforeSave() +} + +func (x *runExitNotify) afterLoad() {} +func (x *runExitNotify) load(m state.Map) { +} + +func (x *taskList) beforeSave() {} +func (x *taskList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *taskList) afterLoad() {} +func (x *taskList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *taskEntry) beforeSave() {} +func (x *taskEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *taskEntry) afterLoad() {} +func (x *taskEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *runApp) beforeSave() {} +func (x *runApp) save(m state.Map) { + x.beforeSave() +} + +func (x *runApp) afterLoad() {} +func (x *runApp) load(m state.Map) { +} + +func (x *TaskGoroutineSchedInfo) beforeSave() {} +func (x *TaskGoroutineSchedInfo) save(m state.Map) { + x.beforeSave() + m.Save("Timestamp", &x.Timestamp) + m.Save("State", &x.State) + m.Save("UserTicks", &x.UserTicks) + m.Save("SysTicks", &x.SysTicks) +} + +func (x *TaskGoroutineSchedInfo) afterLoad() {} +func (x *TaskGoroutineSchedInfo) load(m state.Map) { + m.Load("Timestamp", &x.Timestamp) + m.Load("State", &x.State) + m.Load("UserTicks", &x.UserTicks) + m.Load("SysTicks", &x.SysTicks) +} + +func (x *taskClock) beforeSave() {} +func (x *taskClock) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) + m.Save("includeSys", &x.includeSys) +} + +func (x *taskClock) afterLoad() {} +func (x *taskClock) load(m state.Map) { + m.Load("t", &x.t) + m.Load("includeSys", &x.includeSys) +} + +func (x *tgClock) beforeSave() {} +func (x *tgClock) save(m state.Map) { + x.beforeSave() + m.Save("tg", &x.tg) + m.Save("includeSys", &x.includeSys) +} + +func (x *tgClock) afterLoad() {} +func (x *tgClock) load(m state.Map) { + m.Load("tg", &x.tg) + m.Load("includeSys", &x.includeSys) +} + +func (x *groupStop) beforeSave() {} +func (x *groupStop) save(m state.Map) { + x.beforeSave() +} + +func (x *groupStop) afterLoad() {} +func (x *groupStop) load(m state.Map) { +} + +func (x *runInterrupt) beforeSave() {} +func (x *runInterrupt) save(m state.Map) { + x.beforeSave() +} + +func (x *runInterrupt) afterLoad() {} +func (x *runInterrupt) load(m state.Map) { +} + +func (x *runInterruptAfterSignalDeliveryStop) beforeSave() {} +func (x *runInterruptAfterSignalDeliveryStop) save(m state.Map) { + x.beforeSave() +} + +func (x *runInterruptAfterSignalDeliveryStop) afterLoad() {} +func (x *runInterruptAfterSignalDeliveryStop) load(m state.Map) { +} + +func (x *runSyscallAfterSyscallEnterStop) beforeSave() {} +func (x *runSyscallAfterSyscallEnterStop) save(m state.Map) { + x.beforeSave() +} + +func (x *runSyscallAfterSyscallEnterStop) afterLoad() {} +func (x *runSyscallAfterSyscallEnterStop) load(m state.Map) { +} + +func (x *runSyscallAfterSysemuStop) beforeSave() {} +func (x *runSyscallAfterSysemuStop) save(m state.Map) { + x.beforeSave() +} + +func (x *runSyscallAfterSysemuStop) afterLoad() {} +func (x *runSyscallAfterSysemuStop) load(m state.Map) { +} + +func (x *runSyscallReinvoke) beforeSave() {} +func (x *runSyscallReinvoke) save(m state.Map) { + x.beforeSave() +} + +func (x *runSyscallReinvoke) afterLoad() {} +func (x *runSyscallReinvoke) load(m state.Map) { +} + +func (x *runSyscallExit) beforeSave() {} +func (x *runSyscallExit) save(m state.Map) { + x.beforeSave() +} + +func (x *runSyscallExit) afterLoad() {} +func (x *runSyscallExit) load(m state.Map) { +} + +func (x *ThreadGroup) beforeSave() {} +func (x *ThreadGroup) save(m state.Map) { + x.beforeSave() + var oldRSeqCritical *OldRSeqCriticalRegion = x.saveOldRSeqCritical() + m.SaveValue("oldRSeqCritical", oldRSeqCritical) + m.Save("threadGroupNode", &x.threadGroupNode) + m.Save("signalHandlers", &x.signalHandlers) + m.Save("pendingSignals", &x.pendingSignals) + m.Save("groupStopDequeued", &x.groupStopDequeued) + m.Save("groupStopSignal", &x.groupStopSignal) + m.Save("groupStopPendingCount", &x.groupStopPendingCount) + m.Save("groupStopComplete", &x.groupStopComplete) + m.Save("groupStopWaitable", &x.groupStopWaitable) + m.Save("groupContNotify", &x.groupContNotify) + m.Save("groupContInterrupted", &x.groupContInterrupted) + m.Save("groupContWaitable", &x.groupContWaitable) + m.Save("exiting", &x.exiting) + m.Save("exitStatus", &x.exitStatus) + m.Save("terminationSignal", &x.terminationSignal) + m.Save("itimerRealTimer", &x.itimerRealTimer) + m.Save("itimerVirtSetting", &x.itimerVirtSetting) + m.Save("itimerProfSetting", &x.itimerProfSetting) + m.Save("rlimitCPUSoftSetting", &x.rlimitCPUSoftSetting) + m.Save("cpuTimersEnabled", &x.cpuTimersEnabled) + m.Save("timers", &x.timers) + m.Save("nextTimerID", &x.nextTimerID) + m.Save("exitedCPUStats", &x.exitedCPUStats) + m.Save("childCPUStats", &x.childCPUStats) + m.Save("ioUsage", &x.ioUsage) + m.Save("maxRSS", &x.maxRSS) + m.Save("childMaxRSS", &x.childMaxRSS) + m.Save("limits", &x.limits) + m.Save("processGroup", &x.processGroup) + m.Save("execed", &x.execed) + m.Save("mounts", &x.mounts) + m.Save("tty", &x.tty) + m.Save("oomScoreAdj", &x.oomScoreAdj) +} + +func (x *ThreadGroup) afterLoad() {} +func (x *ThreadGroup) load(m state.Map) { + m.Load("threadGroupNode", &x.threadGroupNode) + m.Load("signalHandlers", &x.signalHandlers) + m.Load("pendingSignals", &x.pendingSignals) + m.Load("groupStopDequeued", &x.groupStopDequeued) + m.Load("groupStopSignal", &x.groupStopSignal) + m.Load("groupStopPendingCount", &x.groupStopPendingCount) + m.Load("groupStopComplete", &x.groupStopComplete) + m.Load("groupStopWaitable", &x.groupStopWaitable) + m.Load("groupContNotify", &x.groupContNotify) + m.Load("groupContInterrupted", &x.groupContInterrupted) + m.Load("groupContWaitable", &x.groupContWaitable) + m.Load("exiting", &x.exiting) + m.Load("exitStatus", &x.exitStatus) + m.Load("terminationSignal", &x.terminationSignal) + m.Load("itimerRealTimer", &x.itimerRealTimer) + m.Load("itimerVirtSetting", &x.itimerVirtSetting) + m.Load("itimerProfSetting", &x.itimerProfSetting) + m.Load("rlimitCPUSoftSetting", &x.rlimitCPUSoftSetting) + m.Load("cpuTimersEnabled", &x.cpuTimersEnabled) + m.Load("timers", &x.timers) + m.Load("nextTimerID", &x.nextTimerID) + m.Load("exitedCPUStats", &x.exitedCPUStats) + m.Load("childCPUStats", &x.childCPUStats) + m.Load("ioUsage", &x.ioUsage) + m.Load("maxRSS", &x.maxRSS) + m.Load("childMaxRSS", &x.childMaxRSS) + m.Load("limits", &x.limits) + m.Load("processGroup", &x.processGroup) + m.Load("execed", &x.execed) + m.Load("mounts", &x.mounts) + m.Load("tty", &x.tty) + m.Load("oomScoreAdj", &x.oomScoreAdj) + m.LoadValue("oldRSeqCritical", new(*OldRSeqCriticalRegion), func(y interface{}) { x.loadOldRSeqCritical(y.(*OldRSeqCriticalRegion)) }) +} + +func (x *itimerRealListener) beforeSave() {} +func (x *itimerRealListener) save(m state.Map) { + x.beforeSave() + m.Save("tg", &x.tg) +} + +func (x *itimerRealListener) afterLoad() {} +func (x *itimerRealListener) load(m state.Map) { + m.Load("tg", &x.tg) +} + +func (x *TaskSet) beforeSave() {} +func (x *TaskSet) save(m state.Map) { + x.beforeSave() + m.Save("Root", &x.Root) + m.Save("sessions", &x.sessions) +} + +func (x *TaskSet) afterLoad() {} +func (x *TaskSet) load(m state.Map) { + m.Load("Root", &x.Root) + m.Load("sessions", &x.sessions) +} + +func (x *PIDNamespace) beforeSave() {} +func (x *PIDNamespace) save(m state.Map) { + x.beforeSave() + m.Save("owner", &x.owner) + m.Save("parent", &x.parent) + m.Save("userns", &x.userns) + m.Save("last", &x.last) + m.Save("tasks", &x.tasks) + m.Save("tids", &x.tids) + m.Save("tgids", &x.tgids) + m.Save("sessions", &x.sessions) + m.Save("sids", &x.sids) + m.Save("processGroups", &x.processGroups) + m.Save("pgids", &x.pgids) + m.Save("exiting", &x.exiting) +} + +func (x *PIDNamespace) afterLoad() {} +func (x *PIDNamespace) load(m state.Map) { + m.Load("owner", &x.owner) + m.Load("parent", &x.parent) + m.Load("userns", &x.userns) + m.Load("last", &x.last) + m.Load("tasks", &x.tasks) + m.Load("tids", &x.tids) + m.Load("tgids", &x.tgids) + m.Load("sessions", &x.sessions) + m.Load("sids", &x.sids) + m.Load("processGroups", &x.processGroups) + m.Load("pgids", &x.pgids) + m.Load("exiting", &x.exiting) +} + +func (x *threadGroupNode) beforeSave() {} +func (x *threadGroupNode) save(m state.Map) { + x.beforeSave() + m.Save("pidns", &x.pidns) + m.Save("leader", &x.leader) + m.Save("execing", &x.execing) + m.Save("tasks", &x.tasks) + m.Save("tasksCount", &x.tasksCount) + m.Save("liveTasks", &x.liveTasks) + m.Save("activeTasks", &x.activeTasks) +} + +func (x *threadGroupNode) afterLoad() {} +func (x *threadGroupNode) load(m state.Map) { + m.Load("pidns", &x.pidns) + m.Load("leader", &x.leader) + m.Load("execing", &x.execing) + m.Load("tasks", &x.tasks) + m.Load("tasksCount", &x.tasksCount) + m.Load("liveTasks", &x.liveTasks) + m.Load("activeTasks", &x.activeTasks) +} + +func (x *taskNode) beforeSave() {} +func (x *taskNode) save(m state.Map) { + x.beforeSave() + m.Save("tg", &x.tg) + m.Save("taskEntry", &x.taskEntry) + m.Save("parent", &x.parent) + m.Save("children", &x.children) + m.Save("childPIDNamespace", &x.childPIDNamespace) +} + +func (x *taskNode) afterLoad() {} +func (x *taskNode) load(m state.Map) { + m.LoadWait("tg", &x.tg) + m.Load("taskEntry", &x.taskEntry) + m.Load("parent", &x.parent) + m.Load("children", &x.children) + m.Load("childPIDNamespace", &x.childPIDNamespace) +} + +func (x *Timekeeper) save(m state.Map) { + x.beforeSave() + m.Save("bootTime", &x.bootTime) + m.Save("saveMonotonic", &x.saveMonotonic) + m.Save("saveRealtime", &x.saveRealtime) + m.Save("params", &x.params) +} + +func (x *Timekeeper) load(m state.Map) { + m.Load("bootTime", &x.bootTime) + m.Load("saveMonotonic", &x.saveMonotonic) + m.Load("saveRealtime", &x.saveRealtime) + m.Load("params", &x.params) + m.AfterLoad(x.afterLoad) +} + +func (x *timekeeperClock) beforeSave() {} +func (x *timekeeperClock) save(m state.Map) { + x.beforeSave() + m.Save("tk", &x.tk) + m.Save("c", &x.c) +} + +func (x *timekeeperClock) afterLoad() {} +func (x *timekeeperClock) load(m state.Map) { + m.Load("tk", &x.tk) + m.Load("c", &x.c) +} + +func (x *TTY) beforeSave() {} +func (x *TTY) save(m state.Map) { + x.beforeSave() + m.Save("Index", &x.Index) + m.Save("tg", &x.tg) +} + +func (x *TTY) afterLoad() {} +func (x *TTY) load(m state.Map) { + m.Load("Index", &x.Index) + m.Load("tg", &x.tg) +} + +func (x *UTSNamespace) beforeSave() {} +func (x *UTSNamespace) save(m state.Map) { + x.beforeSave() + m.Save("hostName", &x.hostName) + m.Save("domainName", &x.domainName) + m.Save("userns", &x.userns) +} + +func (x *UTSNamespace) afterLoad() {} +func (x *UTSNamespace) load(m state.Map) { + m.Load("hostName", &x.hostName) + m.Load("domainName", &x.domainName) + m.Load("userns", &x.userns) +} + +func (x *VDSOParamPage) beforeSave() {} +func (x *VDSOParamPage) save(m state.Map) { + x.beforeSave() + m.Save("mfp", &x.mfp) + m.Save("fr", &x.fr) + m.Save("seq", &x.seq) +} + +func (x *VDSOParamPage) afterLoad() {} +func (x *VDSOParamPage) load(m state.Map) { + m.Load("mfp", &x.mfp) + m.Load("fr", &x.fr) + m.Load("seq", &x.seq) +} + +func init() { + state.Register("pkg/sentry/kernel.abstractEndpoint", (*abstractEndpoint)(nil), state.Fns{Save: (*abstractEndpoint).save, Load: (*abstractEndpoint).load}) + state.Register("pkg/sentry/kernel.AbstractSocketNamespace", (*AbstractSocketNamespace)(nil), state.Fns{Save: (*AbstractSocketNamespace).save, Load: (*AbstractSocketNamespace).load}) + state.Register("pkg/sentry/kernel.FDFlags", (*FDFlags)(nil), state.Fns{Save: (*FDFlags).save, Load: (*FDFlags).load}) + state.Register("pkg/sentry/kernel.descriptor", (*descriptor)(nil), state.Fns{Save: (*descriptor).save, Load: (*descriptor).load}) + state.Register("pkg/sentry/kernel.FDTable", (*FDTable)(nil), state.Fns{Save: (*FDTable).save, Load: (*FDTable).load}) + state.Register("pkg/sentry/kernel.FSContext", (*FSContext)(nil), state.Fns{Save: (*FSContext).save, Load: (*FSContext).load}) + state.Register("pkg/sentry/kernel.IPCNamespace", (*IPCNamespace)(nil), state.Fns{Save: (*IPCNamespace).save, Load: (*IPCNamespace).load}) + state.Register("pkg/sentry/kernel.Kernel", (*Kernel)(nil), state.Fns{Save: (*Kernel).save, Load: (*Kernel).load}) + state.Register("pkg/sentry/kernel.SocketEntry", (*SocketEntry)(nil), state.Fns{Save: (*SocketEntry).save, Load: (*SocketEntry).load}) + state.Register("pkg/sentry/kernel.pendingSignals", (*pendingSignals)(nil), state.Fns{Save: (*pendingSignals).save, Load: (*pendingSignals).load}) + state.Register("pkg/sentry/kernel.pendingSignalQueue", (*pendingSignalQueue)(nil), state.Fns{Save: (*pendingSignalQueue).save, Load: (*pendingSignalQueue).load}) + state.Register("pkg/sentry/kernel.pendingSignal", (*pendingSignal)(nil), state.Fns{Save: (*pendingSignal).save, Load: (*pendingSignal).load}) + state.Register("pkg/sentry/kernel.pendingSignalList", (*pendingSignalList)(nil), state.Fns{Save: (*pendingSignalList).save, Load: (*pendingSignalList).load}) + state.Register("pkg/sentry/kernel.pendingSignalEntry", (*pendingSignalEntry)(nil), state.Fns{Save: (*pendingSignalEntry).save, Load: (*pendingSignalEntry).load}) + state.Register("pkg/sentry/kernel.savedPendingSignal", (*savedPendingSignal)(nil), state.Fns{Save: (*savedPendingSignal).save, Load: (*savedPendingSignal).load}) + state.Register("pkg/sentry/kernel.IntervalTimer", (*IntervalTimer)(nil), state.Fns{Save: (*IntervalTimer).save, Load: (*IntervalTimer).load}) + state.Register("pkg/sentry/kernel.processGroupList", (*processGroupList)(nil), state.Fns{Save: (*processGroupList).save, Load: (*processGroupList).load}) + state.Register("pkg/sentry/kernel.processGroupEntry", (*processGroupEntry)(nil), state.Fns{Save: (*processGroupEntry).save, Load: (*processGroupEntry).load}) + state.Register("pkg/sentry/kernel.ptraceOptions", (*ptraceOptions)(nil), state.Fns{Save: (*ptraceOptions).save, Load: (*ptraceOptions).load}) + state.Register("pkg/sentry/kernel.ptraceStop", (*ptraceStop)(nil), state.Fns{Save: (*ptraceStop).save, Load: (*ptraceStop).load}) + state.Register("pkg/sentry/kernel.OldRSeqCriticalRegion", (*OldRSeqCriticalRegion)(nil), state.Fns{Save: (*OldRSeqCriticalRegion).save, Load: (*OldRSeqCriticalRegion).load}) + state.Register("pkg/sentry/kernel.sessionList", (*sessionList)(nil), state.Fns{Save: (*sessionList).save, Load: (*sessionList).load}) + state.Register("pkg/sentry/kernel.sessionEntry", (*sessionEntry)(nil), state.Fns{Save: (*sessionEntry).save, Load: (*sessionEntry).load}) + state.Register("pkg/sentry/kernel.Session", (*Session)(nil), state.Fns{Save: (*Session).save, Load: (*Session).load}) + state.Register("pkg/sentry/kernel.ProcessGroup", (*ProcessGroup)(nil), state.Fns{Save: (*ProcessGroup).save, Load: (*ProcessGroup).load}) + state.Register("pkg/sentry/kernel.SignalHandlers", (*SignalHandlers)(nil), state.Fns{Save: (*SignalHandlers).save, Load: (*SignalHandlers).load}) + state.Register("pkg/sentry/kernel.socketList", (*socketList)(nil), state.Fns{Save: (*socketList).save, Load: (*socketList).load}) + state.Register("pkg/sentry/kernel.socketEntry", (*socketEntry)(nil), state.Fns{Save: (*socketEntry).save, Load: (*socketEntry).load}) + state.Register("pkg/sentry/kernel.SyscallTable", (*SyscallTable)(nil), state.Fns{Save: (*SyscallTable).save, Load: (*SyscallTable).load}) + state.Register("pkg/sentry/kernel.syslog", (*syslog)(nil), state.Fns{Save: (*syslog).save, Load: (*syslog).load}) + state.Register("pkg/sentry/kernel.Task", (*Task)(nil), state.Fns{Save: (*Task).save, Load: (*Task).load}) + state.Register("pkg/sentry/kernel.runSyscallAfterPtraceEventClone", (*runSyscallAfterPtraceEventClone)(nil), state.Fns{Save: (*runSyscallAfterPtraceEventClone).save, Load: (*runSyscallAfterPtraceEventClone).load}) + state.Register("pkg/sentry/kernel.runSyscallAfterVforkStop", (*runSyscallAfterVforkStop)(nil), state.Fns{Save: (*runSyscallAfterVforkStop).save, Load: (*runSyscallAfterVforkStop).load}) + state.Register("pkg/sentry/kernel.vforkStop", (*vforkStop)(nil), state.Fns{Save: (*vforkStop).save, Load: (*vforkStop).load}) + state.Register("pkg/sentry/kernel.TaskContext", (*TaskContext)(nil), state.Fns{Save: (*TaskContext).save, Load: (*TaskContext).load}) + state.Register("pkg/sentry/kernel.execStop", (*execStop)(nil), state.Fns{Save: (*execStop).save, Load: (*execStop).load}) + state.Register("pkg/sentry/kernel.runSyscallAfterExecStop", (*runSyscallAfterExecStop)(nil), state.Fns{Save: (*runSyscallAfterExecStop).save, Load: (*runSyscallAfterExecStop).load}) + state.Register("pkg/sentry/kernel.ExitStatus", (*ExitStatus)(nil), state.Fns{Save: (*ExitStatus).save, Load: (*ExitStatus).load}) + state.Register("pkg/sentry/kernel.runExit", (*runExit)(nil), state.Fns{Save: (*runExit).save, Load: (*runExit).load}) + state.Register("pkg/sentry/kernel.runExitMain", (*runExitMain)(nil), state.Fns{Save: (*runExitMain).save, Load: (*runExitMain).load}) + state.Register("pkg/sentry/kernel.runExitNotify", (*runExitNotify)(nil), state.Fns{Save: (*runExitNotify).save, Load: (*runExitNotify).load}) + state.Register("pkg/sentry/kernel.taskList", (*taskList)(nil), state.Fns{Save: (*taskList).save, Load: (*taskList).load}) + state.Register("pkg/sentry/kernel.taskEntry", (*taskEntry)(nil), state.Fns{Save: (*taskEntry).save, Load: (*taskEntry).load}) + state.Register("pkg/sentry/kernel.runApp", (*runApp)(nil), state.Fns{Save: (*runApp).save, Load: (*runApp).load}) + state.Register("pkg/sentry/kernel.TaskGoroutineSchedInfo", (*TaskGoroutineSchedInfo)(nil), state.Fns{Save: (*TaskGoroutineSchedInfo).save, Load: (*TaskGoroutineSchedInfo).load}) + state.Register("pkg/sentry/kernel.taskClock", (*taskClock)(nil), state.Fns{Save: (*taskClock).save, Load: (*taskClock).load}) + state.Register("pkg/sentry/kernel.tgClock", (*tgClock)(nil), state.Fns{Save: (*tgClock).save, Load: (*tgClock).load}) + state.Register("pkg/sentry/kernel.groupStop", (*groupStop)(nil), state.Fns{Save: (*groupStop).save, Load: (*groupStop).load}) + state.Register("pkg/sentry/kernel.runInterrupt", (*runInterrupt)(nil), state.Fns{Save: (*runInterrupt).save, Load: (*runInterrupt).load}) + state.Register("pkg/sentry/kernel.runInterruptAfterSignalDeliveryStop", (*runInterruptAfterSignalDeliveryStop)(nil), state.Fns{Save: (*runInterruptAfterSignalDeliveryStop).save, Load: (*runInterruptAfterSignalDeliveryStop).load}) + state.Register("pkg/sentry/kernel.runSyscallAfterSyscallEnterStop", (*runSyscallAfterSyscallEnterStop)(nil), state.Fns{Save: (*runSyscallAfterSyscallEnterStop).save, Load: (*runSyscallAfterSyscallEnterStop).load}) + state.Register("pkg/sentry/kernel.runSyscallAfterSysemuStop", (*runSyscallAfterSysemuStop)(nil), state.Fns{Save: (*runSyscallAfterSysemuStop).save, Load: (*runSyscallAfterSysemuStop).load}) + state.Register("pkg/sentry/kernel.runSyscallReinvoke", (*runSyscallReinvoke)(nil), state.Fns{Save: (*runSyscallReinvoke).save, Load: (*runSyscallReinvoke).load}) + state.Register("pkg/sentry/kernel.runSyscallExit", (*runSyscallExit)(nil), state.Fns{Save: (*runSyscallExit).save, Load: (*runSyscallExit).load}) + state.Register("pkg/sentry/kernel.ThreadGroup", (*ThreadGroup)(nil), state.Fns{Save: (*ThreadGroup).save, Load: (*ThreadGroup).load}) + state.Register("pkg/sentry/kernel.itimerRealListener", (*itimerRealListener)(nil), state.Fns{Save: (*itimerRealListener).save, Load: (*itimerRealListener).load}) + state.Register("pkg/sentry/kernel.TaskSet", (*TaskSet)(nil), state.Fns{Save: (*TaskSet).save, Load: (*TaskSet).load}) + state.Register("pkg/sentry/kernel.PIDNamespace", (*PIDNamespace)(nil), state.Fns{Save: (*PIDNamespace).save, Load: (*PIDNamespace).load}) + state.Register("pkg/sentry/kernel.threadGroupNode", (*threadGroupNode)(nil), state.Fns{Save: (*threadGroupNode).save, Load: (*threadGroupNode).load}) + state.Register("pkg/sentry/kernel.taskNode", (*taskNode)(nil), state.Fns{Save: (*taskNode).save, Load: (*taskNode).load}) + state.Register("pkg/sentry/kernel.Timekeeper", (*Timekeeper)(nil), state.Fns{Save: (*Timekeeper).save, Load: (*Timekeeper).load}) + state.Register("pkg/sentry/kernel.timekeeperClock", (*timekeeperClock)(nil), state.Fns{Save: (*timekeeperClock).save, Load: (*timekeeperClock).load}) + state.Register("pkg/sentry/kernel.TTY", (*TTY)(nil), state.Fns{Save: (*TTY).save, Load: (*TTY).load}) + state.Register("pkg/sentry/kernel.UTSNamespace", (*UTSNamespace)(nil), state.Fns{Save: (*UTSNamespace).save, Load: (*UTSNamespace).load}) + state.Register("pkg/sentry/kernel.VDSOParamPage", (*VDSOParamPage)(nil), state.Fns{Save: (*VDSOParamPage).save, Load: (*VDSOParamPage).load}) +} diff --git a/pkg/sentry/kernel/memevent/BUILD b/pkg/sentry/kernel/memevent/BUILD deleted file mode 100644 index 4486848d2..000000000 --- a/pkg/sentry/kernel/memevent/BUILD +++ /dev/null @@ -1,24 +0,0 @@ -load("//tools:defs.bzl", "go_library", "proto_library") - -package(licenses = ["notice"]) - -go_library( - name = "memevent", - srcs = ["memory_events.go"], - visibility = ["//:sandbox"], - deps = [ - ":memory_events_go_proto", - "//pkg/eventchannel", - "//pkg/log", - "//pkg/metric", - "//pkg/sentry/kernel", - "//pkg/sentry/usage", - "//pkg/sync", - ], -) - -proto_library( - name = "memory_events", - srcs = ["memory_events.proto"], - visibility = ["//visibility:public"], -) diff --git a/pkg/sentry/kernel/memevent/memevent_state_autogen.go b/pkg/sentry/kernel/memevent/memevent_state_autogen.go new file mode 100755 index 000000000..4a1679fa9 --- /dev/null +++ b/pkg/sentry/kernel/memevent/memevent_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package memevent diff --git a/pkg/sentry/kernel/memevent/memory_events.go b/pkg/sentry/kernel/memevent/memory_events.go index 200565bb8..200565bb8 100644..100755 --- a/pkg/sentry/kernel/memevent/memory_events.go +++ b/pkg/sentry/kernel/memevent/memory_events.go diff --git a/pkg/sentry/kernel/memevent/memory_events.proto b/pkg/sentry/kernel/memevent/memory_events.proto deleted file mode 100644 index bf8029ff5..000000000 --- a/pkg/sentry/kernel/memevent/memory_events.proto +++ /dev/null @@ -1,29 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -syntax = "proto3"; - -package gvisor; - -// MemoryUsageEvent describes the memory usage of the sandbox at a single -// instant in time. These messages are emitted periodically on the eventchannel. -message MemoryUsageEvent { - // The total memory usage of the sandboxed application in bytes, calculated - // using the 'fast' method. - uint64 total = 1; - - // Memory used to back memory-mapped regions for files in the application, in - // bytes. This corresponds to the usage.MemoryKind.Mapped memory type. - uint64 mapped = 2; -} diff --git a/pkg/sentry/kernel/memevent/memory_events_go_proto/memory_events.pb.go b/pkg/sentry/kernel/memevent/memory_events_go_proto/memory_events.pb.go new file mode 100755 index 000000000..f8b857fa9 --- /dev/null +++ b/pkg/sentry/kernel/memevent/memory_events_go_proto/memory_events.pb.go @@ -0,0 +1,88 @@ +// Code generated by protoc-gen-go. DO NOT EDIT. +// source: pkg/sentry/kernel/memevent/memory_events.proto + +package gvisor + +import ( + fmt "fmt" + proto "github.com/golang/protobuf/proto" + math "math" +) + +// Reference imports to suppress errors if they are not otherwise used. +var _ = proto.Marshal +var _ = fmt.Errorf +var _ = math.Inf + +// This is a compile-time assertion to ensure that this generated file +// is compatible with the proto package it is being compiled against. +// A compilation error at this line likely means your copy of the +// proto package needs to be updated. +const _ = proto.ProtoPackageIsVersion3 // please upgrade the proto package + +type MemoryUsageEvent struct { + Total uint64 `protobuf:"varint,1,opt,name=total,proto3" json:"total,omitempty"` + Mapped uint64 `protobuf:"varint,2,opt,name=mapped,proto3" json:"mapped,omitempty"` + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *MemoryUsageEvent) Reset() { *m = MemoryUsageEvent{} } +func (m *MemoryUsageEvent) String() string { return proto.CompactTextString(m) } +func (*MemoryUsageEvent) ProtoMessage() {} +func (*MemoryUsageEvent) Descriptor() ([]byte, []int) { + return fileDescriptor_cd85fc8d1130e4b0, []int{0} +} + +func (m *MemoryUsageEvent) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_MemoryUsageEvent.Unmarshal(m, b) +} +func (m *MemoryUsageEvent) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_MemoryUsageEvent.Marshal(b, m, deterministic) +} +func (m *MemoryUsageEvent) XXX_Merge(src proto.Message) { + xxx_messageInfo_MemoryUsageEvent.Merge(m, src) +} +func (m *MemoryUsageEvent) XXX_Size() int { + return xxx_messageInfo_MemoryUsageEvent.Size(m) +} +func (m *MemoryUsageEvent) XXX_DiscardUnknown() { + xxx_messageInfo_MemoryUsageEvent.DiscardUnknown(m) +} + +var xxx_messageInfo_MemoryUsageEvent proto.InternalMessageInfo + +func (m *MemoryUsageEvent) GetTotal() uint64 { + if m != nil { + return m.Total + } + return 0 +} + +func (m *MemoryUsageEvent) GetMapped() uint64 { + if m != nil { + return m.Mapped + } + return 0 +} + +func init() { + proto.RegisterType((*MemoryUsageEvent)(nil), "gvisor.MemoryUsageEvent") +} + +func init() { + proto.RegisterFile("pkg/sentry/kernel/memevent/memory_events.proto", fileDescriptor_cd85fc8d1130e4b0) +} + +var fileDescriptor_cd85fc8d1130e4b0 = []byte{ + // 128 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xe2, 0xd2, 0x2b, 0xc8, 0x4e, 0xd7, + 0x2f, 0x4e, 0xcd, 0x2b, 0x29, 0xaa, 0xd4, 0xcf, 0x4e, 0x2d, 0xca, 0x4b, 0xcd, 0xd1, 0xcf, 0x4d, + 0xcd, 0x4d, 0x2d, 0x4b, 0xcd, 0x2b, 0x01, 0x31, 0xf2, 0x8b, 0x2a, 0xe3, 0xc1, 0x9c, 0x62, 0xbd, + 0x82, 0xa2, 0xfc, 0x92, 0x7c, 0x21, 0xb6, 0xf4, 0xb2, 0xcc, 0xe2, 0xfc, 0x22, 0x25, 0x07, 0x2e, + 0x01, 0x5f, 0xb0, 0x74, 0x68, 0x71, 0x62, 0x7a, 0xaa, 0x2b, 0x48, 0x89, 0x90, 0x08, 0x17, 0x6b, + 0x49, 0x7e, 0x49, 0x62, 0x8e, 0x04, 0xa3, 0x02, 0xa3, 0x06, 0x4b, 0x10, 0x84, 0x23, 0x24, 0xc6, + 0xc5, 0x96, 0x9b, 0x58, 0x50, 0x90, 0x9a, 0x22, 0xc1, 0x04, 0x16, 0x86, 0xf2, 0x92, 0xd8, 0xc0, + 0x06, 0x1a, 0x03, 0x02, 0x00, 0x00, 0xff, 0xff, 0x99, 0x31, 0x2f, 0x9d, 0x82, 0x00, 0x00, 0x00, +} diff --git a/pkg/sentry/kernel/pending_signals_list.go b/pkg/sentry/kernel/pending_signals_list.go new file mode 100755 index 000000000..8eb40ac2c --- /dev/null +++ b/pkg/sentry/kernel/pending_signals_list.go @@ -0,0 +1,186 @@ +package kernel + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type pendingSignalElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (pendingSignalElementMapper) linkerFor(elem *pendingSignal) *pendingSignal { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type pendingSignalList struct { + head *pendingSignal + tail *pendingSignal +} + +// Reset resets list l to the empty state. +func (l *pendingSignalList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *pendingSignalList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *pendingSignalList) Front() *pendingSignal { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *pendingSignalList) Back() *pendingSignal { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *pendingSignalList) PushFront(e *pendingSignal) { + linker := pendingSignalElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + pendingSignalElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *pendingSignalList) PushBack(e *pendingSignal) { + linker := pendingSignalElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + pendingSignalElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *pendingSignalList) PushBackList(m *pendingSignalList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + pendingSignalElementMapper{}.linkerFor(l.tail).SetNext(m.head) + pendingSignalElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *pendingSignalList) InsertAfter(b, e *pendingSignal) { + bLinker := pendingSignalElementMapper{}.linkerFor(b) + eLinker := pendingSignalElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + pendingSignalElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *pendingSignalList) InsertBefore(a, e *pendingSignal) { + aLinker := pendingSignalElementMapper{}.linkerFor(a) + eLinker := pendingSignalElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + pendingSignalElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *pendingSignalList) Remove(e *pendingSignal) { + linker := pendingSignalElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + pendingSignalElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + pendingSignalElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type pendingSignalEntry struct { + next *pendingSignal + prev *pendingSignal +} + +// Next returns the entry that follows e in the list. +func (e *pendingSignalEntry) Next() *pendingSignal { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *pendingSignalEntry) Prev() *pendingSignal { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *pendingSignalEntry) SetNext(elem *pendingSignal) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *pendingSignalEntry) SetPrev(elem *pendingSignal) { + e.prev = elem +} diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD deleted file mode 100644 index f29dc0472..000000000 --- a/pkg/sentry/kernel/pipe/BUILD +++ /dev/null @@ -1,51 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "pipe", - srcs = [ - "device.go", - "node.go", - "pipe.go", - "pipe_util.go", - "reader.go", - "reader_writer.go", - "vfs.go", - "writer.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/amutex", - "//pkg/buffer", - "//pkg/context", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "pipe_test", - size = "small", - srcs = [ - "node_test.go", - "pipe_test.go", - ], - library = ":pipe", - deps = [ - "//pkg/context", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/kernel/pipe/node_test.go b/pkg/sentry/kernel/pipe/node_test.go deleted file mode 100644 index ab75a87ff..000000000 --- a/pkg/sentry/kernel/pipe/node_test.go +++ /dev/null @@ -1,320 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package pipe - -import ( - "testing" - "time" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -type sleeper struct { - context.Context - ch chan struct{} -} - -func newSleeperContext(t *testing.T) context.Context { - return &sleeper{ - Context: contexttest.Context(t), - ch: make(chan struct{}), - } -} - -func (s *sleeper) SleepStart() <-chan struct{} { - return s.ch -} - -func (s *sleeper) SleepFinish(bool) { -} - -func (s *sleeper) Cancel() { - s.ch <- struct{}{} -} - -func (s *sleeper) Interrupted() bool { - return len(s.ch) != 0 -} - -type openResult struct { - *fs.File - error -} - -var perms fs.FilePermissions = fs.FilePermissions{ - User: fs.PermMask{Read: true, Write: true}, -} - -func testOpenOrDie(ctx context.Context, t *testing.T, n fs.InodeOperations, flags fs.FileFlags, doneChan chan<- struct{}) (*fs.File, error) { - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{Type: fs.Pipe}) - d := fs.NewDirent(ctx, inode, "pipe") - file, err := n.GetFile(ctx, d, flags) - if err != nil { - t.Fatalf("open with flags %+v failed: %v", flags, err) - } - if doneChan != nil { - doneChan <- struct{}{} - } - return file, err -} - -func testOpen(ctx context.Context, t *testing.T, n fs.InodeOperations, flags fs.FileFlags, resChan chan<- openResult) (*fs.File, error) { - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{Type: fs.Pipe}) - d := fs.NewDirent(ctx, inode, "pipe") - file, err := n.GetFile(ctx, d, flags) - if resChan != nil { - resChan <- openResult{file, err} - } - return file, err -} - -func newNamedPipe(t *testing.T) *Pipe { - return NewPipe(true, DefaultPipeSize, usermem.PageSize) -} - -func newAnonPipe(t *testing.T) *Pipe { - return NewPipe(false, DefaultPipeSize, usermem.PageSize) -} - -// assertRecvBlocks ensures that a recv attempt on c blocks for at least -// blockDuration. This is useful for checking that a goroutine that is supposed -// to be executing a blocking operation is actually blocking. -func assertRecvBlocks(t *testing.T, c <-chan struct{}, blockDuration time.Duration, failMsg string) { - select { - case <-c: - t.Fatalf(failMsg) - case <-time.After(blockDuration): - // Ok, blocked for the required duration. - } -} - -func TestReadOpenBlocksForWriteOpen(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - rDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true}, rDone) - - // Verify that the open for read is blocking. - assertRecvBlocks(t, rDone, time.Millisecond*100, - "open for read not blocking with no writers") - - wDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Write: true}, wDone) - - <-wDone - <-rDone -} - -func TestWriteOpenBlocksForReadOpen(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - wDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Write: true}, wDone) - - // Verify that the open for write is blocking - assertRecvBlocks(t, wDone, time.Millisecond*100, - "open for write not blocking with no readers") - - rDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true}, rDone) - - <-rDone - <-wDone -} - -func TestMultipleWriteOpenDoesntCountAsReadOpen(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - rDone1 := make(chan struct{}) - rDone2 := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true}, rDone1) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true}, rDone2) - - assertRecvBlocks(t, rDone1, time.Millisecond*100, - "open for read didn't block with no writers") - assertRecvBlocks(t, rDone2, time.Millisecond*100, - "open for read didn't block with no writers") - - wDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Write: true}, wDone) - - <-wDone - <-rDone2 - <-rDone1 -} - -func TestClosedReaderBlocksWriteOpen(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - rFile, _ := testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true, NonBlocking: true}, nil) - rFile.DecRef() - - wDone := make(chan struct{}) - // This open for write should block because the reader is now gone. - go testOpenOrDie(ctx, t, f, fs.FileFlags{Write: true}, wDone) - assertRecvBlocks(t, wDone, time.Millisecond*100, - "open for write didn't block with no concurrent readers") - - // Open for read again. This should unblock the open for write. - rDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true}, rDone) - - <-rDone - <-wDone -} - -func TestReadWriteOpenNeverBlocks(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - rwDone := make(chan struct{}) - // Open for read-write never wait for a reader or writer, even if the - // nonblocking flag is not set. - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true, Write: true, NonBlocking: false}, rwDone) - <-rwDone -} - -func TestReadWriteOpenUnblocksReadOpen(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - rDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true}, rDone) - - rwDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true, Write: true}, rwDone) - - <-rwDone - <-rDone -} - -func TestReadWriteOpenUnblocksWriteOpen(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - wDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Write: true}, wDone) - - rwDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true, Write: true}, rwDone) - - <-rwDone - <-wDone -} - -func TestBlockedOpenIsCancellable(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - done := make(chan openResult) - go testOpen(ctx, t, f, fs.FileFlags{Read: true}, done) - select { - case <-done: - t.Fatalf("open for read didn't block with no writers") - case <-time.After(time.Millisecond * 100): - // Ok. - } - - ctx.(*sleeper).Cancel() - // If the cancel on the sleeper didn't work, the open for read would never - // return. - res := <-done - if res.error != syserror.ErrInterrupted { - t.Fatalf("Cancellation didn't cause GetFile to return fs.ErrInterrupted, got %v.", - res.error) - } -} - -func TestNonblockingReadOpenFileNoWriters(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - if _, err := testOpen(ctx, t, f, fs.FileFlags{Read: true, NonBlocking: true}, nil); err != nil { - t.Fatalf("Nonblocking open for read failed with error %v.", err) - } -} - -func TestNonblockingWriteOpenFileNoReaders(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - if _, err := testOpen(ctx, t, f, fs.FileFlags{Write: true, NonBlocking: true}, nil); err != syserror.ENXIO { - t.Fatalf("Nonblocking open for write failed unexpected error %v.", err) - } -} - -func TestNonBlockingReadOpenWithWriter(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - wDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Write: true}, wDone) - - // Open for write blocks since there are no readers yet. - assertRecvBlocks(t, wDone, time.Millisecond*100, - "Open for write didn't block with no reader.") - - if _, err := testOpen(ctx, t, f, fs.FileFlags{Read: true, NonBlocking: true}, nil); err != nil { - t.Fatalf("Nonblocking open for read failed with error %v.", err) - } - - // Open for write should now be unblocked. - <-wDone -} - -func TestNonBlockingWriteOpenWithReader(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newNamedPipe(t)) - - rDone := make(chan struct{}) - go testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true}, rDone) - - // Open for write blocked, since no reader yet. - assertRecvBlocks(t, rDone, time.Millisecond*100, - "Open for reader didn't block with no writer.") - - if _, err := testOpen(ctx, t, f, fs.FileFlags{Write: true, NonBlocking: true}, nil); err != nil { - t.Fatalf("Nonblocking open for write failed with error %v.", err) - } - - // Open for write should now be unblocked. - <-rDone -} - -func TestAnonReadOpen(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newAnonPipe(t)) - - if _, err := testOpen(ctx, t, f, fs.FileFlags{Read: true}, nil); err != nil { - t.Fatalf("open anon pipe for read failed: %v", err) - } -} - -func TestAnonWriteOpen(t *testing.T) { - ctx := newSleeperContext(t) - f := NewInodeOperations(ctx, perms, newAnonPipe(t)) - - if _, err := testOpen(ctx, t, f, fs.FileFlags{Write: true}, nil); err != nil { - t.Fatalf("open anon pipe for write failed: %v", err) - } -} diff --git a/pkg/sentry/kernel/pipe/pipe_state_autogen.go b/pkg/sentry/kernel/pipe/pipe_state_autogen.go new file mode 100755 index 000000000..b49ab46f9 --- /dev/null +++ b/pkg/sentry/kernel/pipe/pipe_state_autogen.go @@ -0,0 +1,84 @@ +// automatically generated by stateify. + +package pipe + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *inodeOperations) beforeSave() {} +func (x *inodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("p", &x.p) +} + +func (x *inodeOperations) afterLoad() {} +func (x *inodeOperations) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("p", &x.p) +} + +func (x *Pipe) beforeSave() {} +func (x *Pipe) save(m state.Map) { + x.beforeSave() + m.Save("isNamed", &x.isNamed) + m.Save("atomicIOBytes", &x.atomicIOBytes) + m.Save("readers", &x.readers) + m.Save("writers", &x.writers) + m.Save("view", &x.view) + m.Save("max", &x.max) + m.Save("hadWriter", &x.hadWriter) +} + +func (x *Pipe) afterLoad() {} +func (x *Pipe) load(m state.Map) { + m.Load("isNamed", &x.isNamed) + m.Load("atomicIOBytes", &x.atomicIOBytes) + m.Load("readers", &x.readers) + m.Load("writers", &x.writers) + m.Load("view", &x.view) + m.Load("max", &x.max) + m.Load("hadWriter", &x.hadWriter) +} + +func (x *Reader) beforeSave() {} +func (x *Reader) save(m state.Map) { + x.beforeSave() + m.Save("ReaderWriter", &x.ReaderWriter) +} + +func (x *Reader) afterLoad() {} +func (x *Reader) load(m state.Map) { + m.Load("ReaderWriter", &x.ReaderWriter) +} + +func (x *ReaderWriter) beforeSave() {} +func (x *ReaderWriter) save(m state.Map) { + x.beforeSave() + m.Save("Pipe", &x.Pipe) +} + +func (x *ReaderWriter) afterLoad() {} +func (x *ReaderWriter) load(m state.Map) { + m.Load("Pipe", &x.Pipe) +} + +func (x *Writer) beforeSave() {} +func (x *Writer) save(m state.Map) { + x.beforeSave() + m.Save("ReaderWriter", &x.ReaderWriter) +} + +func (x *Writer) afterLoad() {} +func (x *Writer) load(m state.Map) { + m.Load("ReaderWriter", &x.ReaderWriter) +} + +func init() { + state.Register("pkg/sentry/kernel/pipe.inodeOperations", (*inodeOperations)(nil), state.Fns{Save: (*inodeOperations).save, Load: (*inodeOperations).load}) + state.Register("pkg/sentry/kernel/pipe.Pipe", (*Pipe)(nil), state.Fns{Save: (*Pipe).save, Load: (*Pipe).load}) + state.Register("pkg/sentry/kernel/pipe.Reader", (*Reader)(nil), state.Fns{Save: (*Reader).save, Load: (*Reader).load}) + state.Register("pkg/sentry/kernel/pipe.ReaderWriter", (*ReaderWriter)(nil), state.Fns{Save: (*ReaderWriter).save, Load: (*ReaderWriter).load}) + state.Register("pkg/sentry/kernel/pipe.Writer", (*Writer)(nil), state.Fns{Save: (*Writer).save, Load: (*Writer).load}) +} diff --git a/pkg/sentry/kernel/pipe/pipe_test.go b/pkg/sentry/kernel/pipe/pipe_test.go deleted file mode 100644 index bda739dbe..000000000 --- a/pkg/sentry/kernel/pipe/pipe_test.go +++ /dev/null @@ -1,139 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package pipe - -import ( - "bytes" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestPipeRW(t *testing.T) { - ctx := contexttest.Context(t) - r, w := NewConnectedPipe(ctx, 65536, 4096) - defer r.DecRef() - defer w.DecRef() - - msg := []byte("here's some bytes") - wantN := int64(len(msg)) - n, err := w.Writev(ctx, usermem.BytesIOSequence(msg)) - if n != wantN || err != nil { - t.Fatalf("Writev: got (%d, %v), wanted (%d, nil)", n, err, wantN) - } - - buf := make([]byte, len(msg)) - n, err = r.Readv(ctx, usermem.BytesIOSequence(buf)) - if n != wantN || err != nil || !bytes.Equal(buf, msg) { - t.Fatalf("Readv: got (%d, %v) %q, wanted (%d, nil) %q", n, err, buf, wantN, msg) - } -} - -func TestPipeReadBlock(t *testing.T) { - ctx := contexttest.Context(t) - r, w := NewConnectedPipe(ctx, 65536, 4096) - defer r.DecRef() - defer w.DecRef() - - n, err := r.Readv(ctx, usermem.BytesIOSequence(make([]byte, 1))) - if n != 0 || err != syserror.ErrWouldBlock { - t.Fatalf("Readv: got (%d, %v), wanted (0, %v)", n, err, syserror.ErrWouldBlock) - } -} - -func TestPipeWriteBlock(t *testing.T) { - const atomicIOBytes = 2 - const capacity = MinimumPipeSize - - ctx := contexttest.Context(t) - r, w := NewConnectedPipe(ctx, capacity, atomicIOBytes) - defer r.DecRef() - defer w.DecRef() - - msg := make([]byte, capacity+1) - n, err := w.Writev(ctx, usermem.BytesIOSequence(msg)) - if wantN, wantErr := int64(capacity), syserror.ErrWouldBlock; n != wantN || err != wantErr { - t.Fatalf("Writev: got (%d, %v), wanted (%d, %v)", n, err, wantN, wantErr) - } -} - -func TestPipeWriteUntilEnd(t *testing.T) { - const atomicIOBytes = 2 - - ctx := contexttest.Context(t) - r, w := NewConnectedPipe(ctx, atomicIOBytes, atomicIOBytes) - defer r.DecRef() - defer w.DecRef() - - msg := []byte("here's some bytes") - - wDone := make(chan struct{}, 0) - rDone := make(chan struct{}, 0) - defer func() { - // Signal the reader to stop and wait until it does so. - close(wDone) - <-rDone - }() - - go func() { - defer close(rDone) - // Read from r until done is closed. - ctx := contexttest.Context(t) - buf := make([]byte, len(msg)+1) - dst := usermem.BytesIOSequence(buf) - e, ch := waiter.NewChannelEntry(nil) - r.EventRegister(&e, waiter.EventIn) - defer r.EventUnregister(&e) - for { - n, err := r.Readv(ctx, dst) - dst = dst.DropFirst64(n) - if err == syserror.ErrWouldBlock { - select { - case <-ch: - continue - case <-wDone: - // We expect to have 1 byte left in dst since len(buf) == - // len(msg)+1. - if dst.NumBytes() != 1 || !bytes.Equal(buf[:len(msg)], msg) { - t.Errorf("Reader: got %q (%d bytes remaining), wanted %q", buf, dst.NumBytes(), msg) - } - return - } - } - if err != nil { - t.Fatalf("Readv: got unexpected error %v", err) - } - } - }() - - src := usermem.BytesIOSequence(msg) - e, ch := waiter.NewChannelEntry(nil) - w.EventRegister(&e, waiter.EventOut) - defer w.EventUnregister(&e) - for src.NumBytes() != 0 { - n, err := w.Writev(ctx, src) - src = src.DropFirst64(n) - if err == syserror.ErrWouldBlock { - <-ch - continue - } - if err != nil { - t.Fatalf("Writev: got (%d, %v)", n, err) - } - } -} diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index 5a1d4fd57..5a1d4fd57 100644..100755 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index a5675bd70..a5675bd70 100644..100755 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go diff --git a/pkg/sentry/kernel/process_group_list.go b/pkg/sentry/kernel/process_group_list.go new file mode 100755 index 000000000..40c1a13a4 --- /dev/null +++ b/pkg/sentry/kernel/process_group_list.go @@ -0,0 +1,186 @@ +package kernel + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type processGroupElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (processGroupElementMapper) linkerFor(elem *ProcessGroup) *ProcessGroup { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type processGroupList struct { + head *ProcessGroup + tail *ProcessGroup +} + +// Reset resets list l to the empty state. +func (l *processGroupList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *processGroupList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *processGroupList) Front() *ProcessGroup { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *processGroupList) Back() *ProcessGroup { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *processGroupList) PushFront(e *ProcessGroup) { + linker := processGroupElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + processGroupElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *processGroupList) PushBack(e *ProcessGroup) { + linker := processGroupElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + processGroupElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *processGroupList) PushBackList(m *processGroupList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + processGroupElementMapper{}.linkerFor(l.tail).SetNext(m.head) + processGroupElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *processGroupList) InsertAfter(b, e *ProcessGroup) { + bLinker := processGroupElementMapper{}.linkerFor(b) + eLinker := processGroupElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + processGroupElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *processGroupList) InsertBefore(a, e *ProcessGroup) { + aLinker := processGroupElementMapper{}.linkerFor(a) + eLinker := processGroupElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + processGroupElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *processGroupList) Remove(e *ProcessGroup) { + linker := processGroupElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + processGroupElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + processGroupElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type processGroupEntry struct { + next *ProcessGroup + prev *ProcessGroup +} + +// Next returns the entry that follows e in the list. +func (e *processGroupEntry) Next() *ProcessGroup { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *processGroupEntry) Prev() *ProcessGroup { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *processGroupEntry) SetNext(elem *ProcessGroup) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *processGroupEntry) SetPrev(elem *ProcessGroup) { + e.prev = elem +} diff --git a/pkg/sentry/kernel/sched/BUILD b/pkg/sentry/kernel/sched/BUILD deleted file mode 100644 index 1b82e087b..000000000 --- a/pkg/sentry/kernel/sched/BUILD +++ /dev/null @@ -1,19 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "sched", - srcs = [ - "cpuset.go", - "sched.go", - ], - visibility = ["//pkg/sentry:internal"], -) - -go_test( - name = "sched_test", - size = "small", - srcs = ["cpuset_test.go"], - library = ":sched", -) diff --git a/pkg/sentry/kernel/sched/cpuset_test.go b/pkg/sentry/kernel/sched/cpuset_test.go deleted file mode 100644 index 3af9f1197..000000000 --- a/pkg/sentry/kernel/sched/cpuset_test.go +++ /dev/null @@ -1,44 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package sched - -import ( - "testing" -) - -func TestNumCPUs(t *testing.T) { - for i := uint(0); i < 1024; i++ { - c := NewCPUSet(i) - for j := uint(0); j < i; j++ { - c.Set(j) - } - n := c.NumCPUs() - if n != i { - t.Errorf("got wrong number of cpus %d, want %d", n, i) - } - } -} - -func TestClearAbove(t *testing.T) { - const n = 1024 - c := NewFullCPUSet(n) - for i := uint(0); i < n; i++ { - cpu := n - i - c.ClearAbove(cpu) - if got := c.NumCPUs(); got != cpu { - t.Errorf("iteration %d: got %d cpus, wanted %d", i, got, cpu) - } - } -} diff --git a/pkg/sentry/kernel/sched/sched_state_autogen.go b/pkg/sentry/kernel/sched/sched_state_autogen.go new file mode 100755 index 000000000..9705ca79d --- /dev/null +++ b/pkg/sentry/kernel/sched/sched_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package sched diff --git a/pkg/sentry/kernel/semaphore/BUILD b/pkg/sentry/kernel/semaphore/BUILD deleted file mode 100644 index 65e5427c1..000000000 --- a/pkg/sentry/kernel/semaphore/BUILD +++ /dev/null @@ -1,49 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "waiter_list", - out = "waiter_list.go", - package = "semaphore", - prefix = "waiter", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*waiter", - "Linker": "*waiter", - }, -) - -go_library( - name = "semaphore", - srcs = [ - "semaphore.go", - "waiter_list.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/log", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sync", - "//pkg/syserror", - ], -) - -go_test( - name = "semaphore_test", - size = "small", - srcs = ["semaphore_test.go"], - library = ":semaphore", - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/contexttest", - "//pkg/sentry/kernel/auth", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/kernel/semaphore/semaphore_state_autogen.go b/pkg/sentry/kernel/semaphore/semaphore_state_autogen.go new file mode 100755 index 000000000..db80a1490 --- /dev/null +++ b/pkg/sentry/kernel/semaphore/semaphore_state_autogen.go @@ -0,0 +1,117 @@ +// automatically generated by stateify. + +package semaphore + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Registry) beforeSave() {} +func (x *Registry) save(m state.Map) { + x.beforeSave() + m.Save("userNS", &x.userNS) + m.Save("semaphores", &x.semaphores) + m.Save("lastIDUsed", &x.lastIDUsed) +} + +func (x *Registry) afterLoad() {} +func (x *Registry) load(m state.Map) { + m.Load("userNS", &x.userNS) + m.Load("semaphores", &x.semaphores) + m.Load("lastIDUsed", &x.lastIDUsed) +} + +func (x *Set) beforeSave() {} +func (x *Set) save(m state.Map) { + x.beforeSave() + m.Save("registry", &x.registry) + m.Save("ID", &x.ID) + m.Save("key", &x.key) + m.Save("creator", &x.creator) + m.Save("owner", &x.owner) + m.Save("perms", &x.perms) + m.Save("opTime", &x.opTime) + m.Save("changeTime", &x.changeTime) + m.Save("sems", &x.sems) + m.Save("dead", &x.dead) +} + +func (x *Set) afterLoad() {} +func (x *Set) load(m state.Map) { + m.Load("registry", &x.registry) + m.Load("ID", &x.ID) + m.Load("key", &x.key) + m.Load("creator", &x.creator) + m.Load("owner", &x.owner) + m.Load("perms", &x.perms) + m.Load("opTime", &x.opTime) + m.Load("changeTime", &x.changeTime) + m.Load("sems", &x.sems) + m.Load("dead", &x.dead) +} + +func (x *sem) beforeSave() {} +func (x *sem) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.waiters) { + m.Failf("waiters is %v, expected zero", x.waiters) + } + m.Save("value", &x.value) + m.Save("pid", &x.pid) +} + +func (x *sem) afterLoad() {} +func (x *sem) load(m state.Map) { + m.Load("value", &x.value) + m.Load("pid", &x.pid) +} + +func (x *waiter) beforeSave() {} +func (x *waiter) save(m state.Map) { + x.beforeSave() + m.Save("waiterEntry", &x.waiterEntry) + m.Save("value", &x.value) + m.Save("ch", &x.ch) +} + +func (x *waiter) afterLoad() {} +func (x *waiter) load(m state.Map) { + m.Load("waiterEntry", &x.waiterEntry) + m.Load("value", &x.value) + m.Load("ch", &x.ch) +} + +func (x *waiterList) beforeSave() {} +func (x *waiterList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *waiterList) afterLoad() {} +func (x *waiterList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *waiterEntry) beforeSave() {} +func (x *waiterEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *waiterEntry) afterLoad() {} +func (x *waiterEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func init() { + state.Register("pkg/sentry/kernel/semaphore.Registry", (*Registry)(nil), state.Fns{Save: (*Registry).save, Load: (*Registry).load}) + state.Register("pkg/sentry/kernel/semaphore.Set", (*Set)(nil), state.Fns{Save: (*Set).save, Load: (*Set).load}) + state.Register("pkg/sentry/kernel/semaphore.sem", (*sem)(nil), state.Fns{Save: (*sem).save, Load: (*sem).load}) + state.Register("pkg/sentry/kernel/semaphore.waiter", (*waiter)(nil), state.Fns{Save: (*waiter).save, Load: (*waiter).load}) + state.Register("pkg/sentry/kernel/semaphore.waiterList", (*waiterList)(nil), state.Fns{Save: (*waiterList).save, Load: (*waiterList).load}) + state.Register("pkg/sentry/kernel/semaphore.waiterEntry", (*waiterEntry)(nil), state.Fns{Save: (*waiterEntry).save, Load: (*waiterEntry).load}) +} diff --git a/pkg/sentry/kernel/semaphore/semaphore_test.go b/pkg/sentry/kernel/semaphore/semaphore_test.go deleted file mode 100644 index e47acefdf..000000000 --- a/pkg/sentry/kernel/semaphore/semaphore_test.go +++ /dev/null @@ -1,172 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package semaphore - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/syserror" -) - -func executeOps(ctx context.Context, t *testing.T, set *Set, ops []linux.Sembuf, block bool) chan struct{} { - ch, _, err := set.executeOps(ctx, ops, 123) - if err != nil { - t.Fatalf("ExecuteOps(ops) failed, err: %v, ops: %+v", err, ops) - } - if block { - if ch == nil { - t.Fatalf("ExecuteOps(ops) got: nil, expected: !nil, ops: %+v", ops) - } - if signalled(ch) { - t.Fatalf("ExecuteOps(ops) channel should not have been signalled, ops: %+v", ops) - } - } else { - if ch != nil { - t.Fatalf("ExecuteOps(ops) got: %v, expected: nil, ops: %+v", ch, ops) - } - } - return ch -} - -func signalled(ch chan struct{}) bool { - select { - case <-ch: - return true - default: - return false - } -} - -func TestBasic(t *testing.T) { - ctx := contexttest.Context(t) - set := &Set{ID: 123, sems: make([]sem, 1)} - ops := []linux.Sembuf{ - {SemOp: 1}, - } - executeOps(ctx, t, set, ops, false) - - ops[0].SemOp = -1 - executeOps(ctx, t, set, ops, false) - - ops[0].SemOp = -1 - ch1 := executeOps(ctx, t, set, ops, true) - - ops[0].SemOp = 1 - executeOps(ctx, t, set, ops, false) - if !signalled(ch1) { - t.Fatalf("ExecuteOps(ops) channel should not have been signalled, ops: %+v", ops) - } -} - -func TestWaitForZero(t *testing.T) { - ctx := contexttest.Context(t) - set := &Set{ID: 123, sems: make([]sem, 1)} - ops := []linux.Sembuf{ - {SemOp: 0}, - } - executeOps(ctx, t, set, ops, false) - - ops[0].SemOp = -2 - ch1 := executeOps(ctx, t, set, ops, true) - - ops[0].SemOp = 0 - executeOps(ctx, t, set, ops, false) - - ops[0].SemOp = 1 - executeOps(ctx, t, set, ops, false) - - ops[0].SemOp = 0 - chZero1 := executeOps(ctx, t, set, ops, true) - - ops[0].SemOp = 0 - chZero2 := executeOps(ctx, t, set, ops, true) - - ops[0].SemOp = 1 - executeOps(ctx, t, set, ops, false) - if !signalled(ch1) { - t.Fatalf("ExecuteOps(ops) channel should have been signalled, ops: %+v, set: %+v", ops, set) - } - - ops[0].SemOp = -2 - executeOps(ctx, t, set, ops, false) - if !signalled(chZero1) { - t.Fatalf("ExecuteOps(ops) channel zero 1 should have been signalled, ops: %+v, set: %+v", ops, set) - } - if !signalled(chZero2) { - t.Fatalf("ExecuteOps(ops) channel zero 2 should have been signalled, ops: %+v, set: %+v", ops, set) - } -} - -func TestNoWait(t *testing.T) { - ctx := contexttest.Context(t) - set := &Set{ID: 123, sems: make([]sem, 1)} - ops := []linux.Sembuf{ - {SemOp: 1}, - } - executeOps(ctx, t, set, ops, false) - - ops[0].SemOp = -2 - ops[0].SemFlg = linux.IPC_NOWAIT - if _, _, err := set.executeOps(ctx, ops, 123); err != syserror.ErrWouldBlock { - t.Fatalf("ExecuteOps(ops) wrong result, got: %v, expected: %v", err, syserror.ErrWouldBlock) - } - - ops[0].SemOp = 0 - ops[0].SemFlg = linux.IPC_NOWAIT - if _, _, err := set.executeOps(ctx, ops, 123); err != syserror.ErrWouldBlock { - t.Fatalf("ExecuteOps(ops) wrong result, got: %v, expected: %v", err, syserror.ErrWouldBlock) - } -} - -func TestUnregister(t *testing.T) { - ctx := contexttest.Context(t) - r := NewRegistry(auth.NewRootUserNamespace()) - set, err := r.FindOrCreate(ctx, 123, 2, linux.FileMode(0x600), true, true, true) - if err != nil { - t.Fatalf("FindOrCreate() failed, err: %v", err) - } - if got := r.FindByID(set.ID); got.ID != set.ID { - t.Fatalf("FindById(%d) failed, got: %+v, expected: %+v", set.ID, got, set) - } - - ops := []linux.Sembuf{ - {SemOp: -1}, - } - chs := make([]chan struct{}, 0, 5) - for i := 0; i < 5; i++ { - ch := executeOps(ctx, t, set, ops, true) - chs = append(chs, ch) - } - - creds := auth.CredentialsFromContext(ctx) - if err := r.RemoveID(set.ID, creds); err != nil { - t.Fatalf("RemoveID(%d) failed, err: %v", set.ID, err) - } - if !set.dead { - t.Fatalf("set is not dead: %+v", set) - } - if got := r.FindByID(set.ID); got != nil { - t.Fatalf("FindById(%d) failed, got: %+v, expected: nil", set.ID, got) - } - for i, ch := range chs { - if !signalled(ch) { - t.Fatalf("channel %d should have been signalled", i) - } - } -} diff --git a/pkg/sentry/kernel/semaphore/waiter_list.go b/pkg/sentry/kernel/semaphore/waiter_list.go new file mode 100755 index 000000000..27120afe3 --- /dev/null +++ b/pkg/sentry/kernel/semaphore/waiter_list.go @@ -0,0 +1,186 @@ +package semaphore + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type waiterElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (waiterElementMapper) linkerFor(elem *waiter) *waiter { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type waiterList struct { + head *waiter + tail *waiter +} + +// Reset resets list l to the empty state. +func (l *waiterList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *waiterList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *waiterList) Front() *waiter { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *waiterList) Back() *waiter { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *waiterList) PushFront(e *waiter) { + linker := waiterElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + waiterElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *waiterList) PushBack(e *waiter) { + linker := waiterElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + waiterElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *waiterList) PushBackList(m *waiterList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + waiterElementMapper{}.linkerFor(l.tail).SetNext(m.head) + waiterElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *waiterList) InsertAfter(b, e *waiter) { + bLinker := waiterElementMapper{}.linkerFor(b) + eLinker := waiterElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + waiterElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *waiterList) InsertBefore(a, e *waiter) { + aLinker := waiterElementMapper{}.linkerFor(a) + eLinker := waiterElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + waiterElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *waiterList) Remove(e *waiter) { + linker := waiterElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + waiterElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + waiterElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type waiterEntry struct { + next *waiter + prev *waiter +} + +// Next returns the entry that follows e in the list. +func (e *waiterEntry) Next() *waiter { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *waiterEntry) Prev() *waiter { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *waiterEntry) SetNext(elem *waiter) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *waiterEntry) SetPrev(elem *waiter) { + e.prev = elem +} diff --git a/pkg/sentry/kernel/seqatomic_taskgoroutineschedinfo_unsafe.go b/pkg/sentry/kernel/seqatomic_taskgoroutineschedinfo_unsafe.go new file mode 100755 index 000000000..950645965 --- /dev/null +++ b/pkg/sentry/kernel/seqatomic_taskgoroutineschedinfo_unsafe.go @@ -0,0 +1,55 @@ +package kernel + +import ( + "fmt" + "reflect" + "strings" + "unsafe" + + "gvisor.dev/gvisor/pkg/sync" +) + +// SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race +// with any writer critical sections in sc. +func SeqAtomicLoadTaskGoroutineSchedInfo(sc *sync.SeqCount, ptr *TaskGoroutineSchedInfo) TaskGoroutineSchedInfo { + // This function doesn't use SeqAtomicTryLoad because doing so is + // measurably, significantly (~20%) slower; Go is awful at inlining. + var val TaskGoroutineSchedInfo + for { + epoch := sc.BeginRead() + if sync.RaceEnabled { + + sync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) + } else { + + val = *ptr + } + if sc.ReadOk(epoch) { + break + } + } + return val +} + +// SeqAtomicTryLoad returns a copy of *ptr while in a reader critical section +// in sc initiated by a call to sc.BeginRead() that returned epoch. If the read +// would race with a writer critical section, SeqAtomicTryLoad returns +// (unspecified, false). +func SeqAtomicTryLoadTaskGoroutineSchedInfo(sc *sync.SeqCount, epoch sync.SeqCountEpoch, ptr *TaskGoroutineSchedInfo) (TaskGoroutineSchedInfo, bool) { + var val TaskGoroutineSchedInfo + if sync.RaceEnabled { + sync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) + } else { + val = *ptr + } + return val, sc.ReadOk(epoch) +} + +func initTaskGoroutineSchedInfo() { + var val TaskGoroutineSchedInfo + typ := reflect.TypeOf(val) + name := typ.Name() + if ptrs := sync.PointersInType(typ, name); len(ptrs) != 0 { + panic(fmt.Sprintf("SeqAtomicLoad<%s> is invalid since values %s of type %s contain pointers:\n%s", typ, name, typ, strings.Join(ptrs, "\n"))) + } +} diff --git a/pkg/sentry/kernel/session_list.go b/pkg/sentry/kernel/session_list.go new file mode 100755 index 000000000..8174f413d --- /dev/null +++ b/pkg/sentry/kernel/session_list.go @@ -0,0 +1,186 @@ +package kernel + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type sessionElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (sessionElementMapper) linkerFor(elem *Session) *Session { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type sessionList struct { + head *Session + tail *Session +} + +// Reset resets list l to the empty state. +func (l *sessionList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *sessionList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *sessionList) Front() *Session { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *sessionList) Back() *Session { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *sessionList) PushFront(e *Session) { + linker := sessionElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + sessionElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *sessionList) PushBack(e *Session) { + linker := sessionElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + sessionElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *sessionList) PushBackList(m *sessionList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + sessionElementMapper{}.linkerFor(l.tail).SetNext(m.head) + sessionElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *sessionList) InsertAfter(b, e *Session) { + bLinker := sessionElementMapper{}.linkerFor(b) + eLinker := sessionElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + sessionElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *sessionList) InsertBefore(a, e *Session) { + aLinker := sessionElementMapper{}.linkerFor(a) + eLinker := sessionElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + sessionElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *sessionList) Remove(e *Session) { + linker := sessionElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + sessionElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + sessionElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type sessionEntry struct { + next *Session + prev *Session +} + +// Next returns the entry that follows e in the list. +func (e *sessionEntry) Next() *Session { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *sessionEntry) Prev() *Session { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *sessionEntry) SetNext(elem *Session) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *sessionEntry) SetPrev(elem *Session) { + e.prev = elem +} diff --git a/pkg/sentry/kernel/shm/BUILD b/pkg/sentry/kernel/shm/BUILD deleted file mode 100644 index bfd779837..000000000 --- a/pkg/sentry/kernel/shm/BUILD +++ /dev/null @@ -1,29 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "shm", - srcs = [ - "device.go", - "shm.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/log", - "//pkg/refs", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/usage", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/kernel/shm/shm_state_autogen.go b/pkg/sentry/kernel/shm/shm_state_autogen.go new file mode 100755 index 000000000..fa8f896f7 --- /dev/null +++ b/pkg/sentry/kernel/shm/shm_state_autogen.go @@ -0,0 +1,74 @@ +// automatically generated by stateify. + +package shm + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Registry) beforeSave() {} +func (x *Registry) save(m state.Map) { + x.beforeSave() + m.Save("userNS", &x.userNS) + m.Save("shms", &x.shms) + m.Save("keysToShms", &x.keysToShms) + m.Save("totalPages", &x.totalPages) + m.Save("lastIDUsed", &x.lastIDUsed) +} + +func (x *Registry) afterLoad() {} +func (x *Registry) load(m state.Map) { + m.Load("userNS", &x.userNS) + m.Load("shms", &x.shms) + m.Load("keysToShms", &x.keysToShms) + m.Load("totalPages", &x.totalPages) + m.Load("lastIDUsed", &x.lastIDUsed) +} + +func (x *Shm) beforeSave() {} +func (x *Shm) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("mfp", &x.mfp) + m.Save("registry", &x.registry) + m.Save("ID", &x.ID) + m.Save("creator", &x.creator) + m.Save("size", &x.size) + m.Save("effectiveSize", &x.effectiveSize) + m.Save("fr", &x.fr) + m.Save("key", &x.key) + m.Save("perms", &x.perms) + m.Save("owner", &x.owner) + m.Save("attachTime", &x.attachTime) + m.Save("detachTime", &x.detachTime) + m.Save("changeTime", &x.changeTime) + m.Save("creatorPID", &x.creatorPID) + m.Save("lastAttachDetachPID", &x.lastAttachDetachPID) + m.Save("pendingDestruction", &x.pendingDestruction) +} + +func (x *Shm) afterLoad() {} +func (x *Shm) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("mfp", &x.mfp) + m.Load("registry", &x.registry) + m.Load("ID", &x.ID) + m.Load("creator", &x.creator) + m.Load("size", &x.size) + m.Load("effectiveSize", &x.effectiveSize) + m.Load("fr", &x.fr) + m.Load("key", &x.key) + m.Load("perms", &x.perms) + m.Load("owner", &x.owner) + m.Load("attachTime", &x.attachTime) + m.Load("detachTime", &x.detachTime) + m.Load("changeTime", &x.changeTime) + m.Load("creatorPID", &x.creatorPID) + m.Load("lastAttachDetachPID", &x.lastAttachDetachPID) + m.Load("pendingDestruction", &x.pendingDestruction) +} + +func init() { + state.Register("pkg/sentry/kernel/shm.Registry", (*Registry)(nil), state.Fns{Save: (*Registry).save, Load: (*Registry).load}) + state.Register("pkg/sentry/kernel/shm.Shm", (*Shm)(nil), state.Fns{Save: (*Shm).save, Load: (*Shm).load}) +} diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD deleted file mode 100644 index 3eb78e91b..000000000 --- a/pkg/sentry/kernel/signalfd/BUILD +++ /dev/null @@ -1,22 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "signalfd", - srcs = ["signalfd.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go index 8243bb93e..8243bb93e 100644..100755 --- a/pkg/sentry/kernel/signalfd/signalfd.go +++ b/pkg/sentry/kernel/signalfd/signalfd.go diff --git a/pkg/sentry/kernel/signalfd/signalfd_state_autogen.go b/pkg/sentry/kernel/signalfd/signalfd_state_autogen.go new file mode 100755 index 000000000..2ab5b4702 --- /dev/null +++ b/pkg/sentry/kernel/signalfd/signalfd_state_autogen.go @@ -0,0 +1,24 @@ +// automatically generated by stateify. + +package signalfd + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *SignalOperations) beforeSave() {} +func (x *SignalOperations) save(m state.Map) { + x.beforeSave() + m.Save("target", &x.target) + m.Save("mask", &x.mask) +} + +func (x *SignalOperations) afterLoad() {} +func (x *SignalOperations) load(m state.Map) { + m.Load("target", &x.target) + m.Load("mask", &x.mask) +} + +func init() { + state.Register("pkg/sentry/kernel/signalfd.SignalOperations", (*SignalOperations)(nil), state.Fns{Save: (*SignalOperations).save, Load: (*SignalOperations).load}) +} diff --git a/pkg/sentry/kernel/socket_list.go b/pkg/sentry/kernel/socket_list.go new file mode 100755 index 000000000..ac93e2365 --- /dev/null +++ b/pkg/sentry/kernel/socket_list.go @@ -0,0 +1,186 @@ +package kernel + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type socketElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (socketElementMapper) linkerFor(elem *SocketEntry) *SocketEntry { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type socketList struct { + head *SocketEntry + tail *SocketEntry +} + +// Reset resets list l to the empty state. +func (l *socketList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *socketList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *socketList) Front() *SocketEntry { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *socketList) Back() *SocketEntry { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *socketList) PushFront(e *SocketEntry) { + linker := socketElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + socketElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *socketList) PushBack(e *SocketEntry) { + linker := socketElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + socketElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *socketList) PushBackList(m *socketList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + socketElementMapper{}.linkerFor(l.tail).SetNext(m.head) + socketElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *socketList) InsertAfter(b, e *SocketEntry) { + bLinker := socketElementMapper{}.linkerFor(b) + eLinker := socketElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + socketElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *socketList) InsertBefore(a, e *SocketEntry) { + aLinker := socketElementMapper{}.linkerFor(a) + eLinker := socketElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + socketElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *socketList) Remove(e *SocketEntry) { + linker := socketElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + socketElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + socketElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type socketEntry struct { + next *SocketEntry + prev *SocketEntry +} + +// Next returns the entry that follows e in the list. +func (e *socketEntry) Next() *SocketEntry { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *socketEntry) Prev() *SocketEntry { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *socketEntry) SetNext(elem *SocketEntry) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *socketEntry) SetPrev(elem *SocketEntry) { + e.prev = elem +} diff --git a/pkg/sentry/kernel/table_test.go b/pkg/sentry/kernel/table_test.go deleted file mode 100644 index 32cf47e05..000000000 --- a/pkg/sentry/kernel/table_test.go +++ /dev/null @@ -1,110 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernel - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/abi" - "gvisor.dev/gvisor/pkg/sentry/arch" -) - -const ( - maxTestSyscall = 1000 -) - -func createSyscallTable() *SyscallTable { - m := make(map[uintptr]Syscall) - for i := uintptr(0); i <= maxTestSyscall; i++ { - j := i - m[i] = Syscall{ - Fn: func(*Task, arch.SyscallArguments) (uintptr, *SyscallControl, error) { - return j, nil, nil - }, - } - } - - s := &SyscallTable{ - OS: abi.Linux, - Arch: arch.AMD64, - Table: m, - } - - RegisterSyscallTable(s) - return s -} - -func TestTable(t *testing.T) { - table := createSyscallTable() - defer func() { - // Cleanup registered tables to keep tests separate. - allSyscallTables = []*SyscallTable{} - }() - - // Go through all functions and check that they return the right value. - for i := uintptr(0); i < maxTestSyscall; i++ { - fn := table.Lookup(i) - if fn == nil { - t.Errorf("Syscall %v is set to nil", i) - continue - } - - v, _, _ := fn(nil, arch.SyscallArguments{}) - if v != i { - t.Errorf("Wrong return value for syscall %v: expected %v, got %v", i, i, v) - } - } - - // Check that values outside the range return nil. - for i := uintptr(maxTestSyscall + 1); i < maxTestSyscall+100; i++ { - fn := table.Lookup(i) - if fn != nil { - t.Errorf("Syscall %v is not nil: %v", i, fn) - continue - } - } -} - -func BenchmarkTableLookup(b *testing.B) { - table := createSyscallTable() - - b.ResetTimer() - - j := uintptr(0) - for i := 0; i < b.N; i++ { - table.Lookup(j) - j = (j + 1) % 310 - } - - b.StopTimer() - // Cleanup registered tables to keep tests separate. - allSyscallTables = []*SyscallTable{} -} - -func BenchmarkTableMapLookup(b *testing.B) { - table := createSyscallTable() - - b.ResetTimer() - - j := uintptr(0) - for i := 0; i < b.N; i++ { - table.mapLookup(j) - j = (j + 1) % 310 - } - - b.StopTimer() - // Cleanup registered tables to keep tests separate. - allSyscallTables = []*SyscallTable{} -} diff --git a/pkg/sentry/kernel/task_list.go b/pkg/sentry/kernel/task_list.go new file mode 100755 index 000000000..4dfcdbf2c --- /dev/null +++ b/pkg/sentry/kernel/task_list.go @@ -0,0 +1,186 @@ +package kernel + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type taskElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (taskElementMapper) linkerFor(elem *Task) *Task { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type taskList struct { + head *Task + tail *Task +} + +// Reset resets list l to the empty state. +func (l *taskList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *taskList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *taskList) Front() *Task { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *taskList) Back() *Task { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *taskList) PushFront(e *Task) { + linker := taskElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + taskElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *taskList) PushBack(e *Task) { + linker := taskElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + taskElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *taskList) PushBackList(m *taskList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + taskElementMapper{}.linkerFor(l.tail).SetNext(m.head) + taskElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *taskList) InsertAfter(b, e *Task) { + bLinker := taskElementMapper{}.linkerFor(b) + eLinker := taskElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + taskElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *taskList) InsertBefore(a, e *Task) { + aLinker := taskElementMapper{}.linkerFor(a) + eLinker := taskElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + taskElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *taskList) Remove(e *Task) { + linker := taskElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + taskElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + taskElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type taskEntry struct { + next *Task + prev *Task +} + +// Next returns the entry that follows e in the list. +func (e *taskEntry) Next() *Task { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *taskEntry) Prev() *Task { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *taskEntry) SetNext(elem *Task) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *taskEntry) SetPrev(elem *Task) { + e.prev = elem +} diff --git a/pkg/sentry/kernel/task_test.go b/pkg/sentry/kernel/task_test.go deleted file mode 100644 index cfcde9a7a..000000000 --- a/pkg/sentry/kernel/task_test.go +++ /dev/null @@ -1,69 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernel - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/kernel/sched" -) - -func TestTaskCPU(t *testing.T) { - for _, test := range []struct { - mask sched.CPUSet - tid ThreadID - cpu int32 - }{ - { - mask: []byte{0xff}, - tid: 1, - cpu: 0, - }, - { - mask: []byte{0xff}, - tid: 10, - cpu: 1, - }, - { - // more than 8 cpus. - mask: []byte{0xff, 0xff}, - tid: 10, - cpu: 9, - }, - { - // missing the first cpu. - mask: []byte{0xfe}, - tid: 1, - cpu: 1, - }, - { - mask: []byte{0xfe}, - tid: 10, - cpu: 3, - }, - { - // missing the fifth cpu. - mask: []byte{0xef}, - tid: 10, - cpu: 2, - }, - } { - assigned := assignCPU(test.mask, test.tid) - if test.cpu != assigned { - t.Errorf("assignCPU(%v, %v) got %v, want %v", test.mask, test.tid, assigned, test.cpu) - } - } - -} diff --git a/pkg/sentry/kernel/time/BUILD b/pkg/sentry/kernel/time/BUILD deleted file mode 100644 index 7ba7dc50c..000000000 --- a/pkg/sentry/kernel/time/BUILD +++ /dev/null @@ -1,19 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "time", - srcs = [ - "context.go", - "time.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sync", - "//pkg/syserror", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/kernel/time/time_state_autogen.go b/pkg/sentry/kernel/time/time_state_autogen.go new file mode 100755 index 000000000..ab6c6633d --- /dev/null +++ b/pkg/sentry/kernel/time/time_state_autogen.go @@ -0,0 +1,56 @@ +// automatically generated by stateify. + +package time + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Time) beforeSave() {} +func (x *Time) save(m state.Map) { + x.beforeSave() + m.Save("ns", &x.ns) +} + +func (x *Time) afterLoad() {} +func (x *Time) load(m state.Map) { + m.Load("ns", &x.ns) +} + +func (x *Setting) beforeSave() {} +func (x *Setting) save(m state.Map) { + x.beforeSave() + m.Save("Enabled", &x.Enabled) + m.Save("Next", &x.Next) + m.Save("Period", &x.Period) +} + +func (x *Setting) afterLoad() {} +func (x *Setting) load(m state.Map) { + m.Load("Enabled", &x.Enabled) + m.Load("Next", &x.Next) + m.Load("Period", &x.Period) +} + +func (x *Timer) beforeSave() {} +func (x *Timer) save(m state.Map) { + x.beforeSave() + m.Save("clock", &x.clock) + m.Save("listener", &x.listener) + m.Save("setting", &x.setting) + m.Save("paused", &x.paused) +} + +func (x *Timer) afterLoad() {} +func (x *Timer) load(m state.Map) { + m.Load("clock", &x.clock) + m.Load("listener", &x.listener) + m.Load("setting", &x.setting) + m.Load("paused", &x.paused) +} + +func init() { + state.Register("pkg/sentry/kernel/time.Time", (*Time)(nil), state.Fns{Save: (*Time).save, Load: (*Time).load}) + state.Register("pkg/sentry/kernel/time.Setting", (*Setting)(nil), state.Fns{Save: (*Setting).save, Load: (*Setting).load}) + state.Register("pkg/sentry/kernel/time.Timer", (*Timer)(nil), state.Fns{Save: (*Timer).save, Load: (*Timer).load}) +} diff --git a/pkg/sentry/kernel/timekeeper_test.go b/pkg/sentry/kernel/timekeeper_test.go deleted file mode 100644 index cf2f7ca72..000000000 --- a/pkg/sentry/kernel/timekeeper_test.go +++ /dev/null @@ -1,156 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernel - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - sentrytime "gvisor.dev/gvisor/pkg/sentry/time" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// mockClocks is a sentrytime.Clocks that simply returns the times in the -// struct. -type mockClocks struct { - monotonic int64 - realtime int64 -} - -// Update implements sentrytime.Clocks.Update. It does nothing. -func (*mockClocks) Update() (monotonicParams sentrytime.Parameters, monotonicOk bool, realtimeParam sentrytime.Parameters, realtimeOk bool) { - return -} - -// Update implements sentrytime.Clocks.GetTime. -func (c *mockClocks) GetTime(id sentrytime.ClockID) (int64, error) { - switch id { - case sentrytime.Monotonic: - return c.monotonic, nil - case sentrytime.Realtime: - return c.realtime, nil - default: - return 0, syserror.EINVAL - } -} - -// stateTestClocklessTimekeeper returns a test Timekeeper which has not had -// SetClocks called. -func stateTestClocklessTimekeeper(tb testing.TB) *Timekeeper { - ctx := contexttest.Context(tb) - mfp := pgalloc.MemoryFileProviderFromContext(ctx) - fr, err := mfp.MemoryFile().Allocate(usermem.PageSize, usage.Anonymous) - if err != nil { - tb.Fatalf("failed to allocate memory: %v", err) - } - return &Timekeeper{ - params: NewVDSOParamPage(mfp, fr), - } -} - -func stateTestTimekeeper(tb testing.TB) *Timekeeper { - t := stateTestClocklessTimekeeper(tb) - t.SetClocks(sentrytime.NewCalibratedClocks()) - return t -} - -// TestTimekeeperMonotonicZero tests that monotonic time starts at zero. -func TestTimekeeperMonotonicZero(t *testing.T) { - c := &mockClocks{ - monotonic: 100000, - } - - tk := stateTestClocklessTimekeeper(t) - tk.SetClocks(c) - defer tk.Destroy() - - now, err := tk.GetTime(sentrytime.Monotonic) - if err != nil { - t.Errorf("GetTime err got %v want nil", err) - } - if now != 0 { - t.Errorf("GetTime got %d want 0", now) - } - - c.monotonic += 10 - - now, err = tk.GetTime(sentrytime.Monotonic) - if err != nil { - t.Errorf("GetTime err got %v want nil", err) - } - if now != 10 { - t.Errorf("GetTime got %d want 10", now) - } -} - -// TestTimekeeperMonotonicJumpForward tests that monotonic time jumps forward -// after restore. -func TestTimekeeperMonotonicForward(t *testing.T) { - c := &mockClocks{ - monotonic: 900000, - realtime: 600000, - } - - tk := stateTestClocklessTimekeeper(t) - tk.restored = make(chan struct{}) - tk.saveMonotonic = 100000 - tk.saveRealtime = 400000 - tk.SetClocks(c) - defer tk.Destroy() - - // The monotonic clock should jump ahead by 200000 to 300000. - // - // The new system monotonic time (900000) is irrelevant to what the app - // sees. - now, err := tk.GetTime(sentrytime.Monotonic) - if err != nil { - t.Errorf("GetTime err got %v want nil", err) - } - if now != 300000 { - t.Errorf("GetTime got %d want 300000", now) - } -} - -// TestTimekeeperMonotonicJumpBackwards tests that monotonic time does not jump -// backwards when realtime goes backwards. -func TestTimekeeperMonotonicJumpBackwards(t *testing.T) { - c := &mockClocks{ - monotonic: 900000, - realtime: 400000, - } - - tk := stateTestClocklessTimekeeper(t) - tk.restored = make(chan struct{}) - tk.saveMonotonic = 100000 - tk.saveRealtime = 600000 - tk.SetClocks(c) - defer tk.Destroy() - - // The monotonic clock should remain at 100000. - // - // The new system monotonic time (900000) is irrelevant to what the app - // sees and we don't want to jump the monotonic clock backwards like - // realtime did. - now, err := tk.GetTime(sentrytime.Monotonic) - if err != nil { - t.Errorf("GetTime err got %v want nil", err) - } - if now != 100000 { - t.Errorf("GetTime got %d want 100000", now) - } -} diff --git a/pkg/sentry/kernel/tty.go b/pkg/sentry/kernel/tty.go index d0e0810e8..d0e0810e8 100644..100755 --- a/pkg/sentry/kernel/tty.go +++ b/pkg/sentry/kernel/tty.go diff --git a/pkg/sentry/kernel/uncaught_signal.proto b/pkg/sentry/kernel/uncaught_signal.proto deleted file mode 100644 index 0bdb062cb..000000000 --- a/pkg/sentry/kernel/uncaught_signal.proto +++ /dev/null @@ -1,37 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -syntax = "proto3"; - -package gvisor; - -import "pkg/sentry/arch/registers.proto"; - -message UncaughtSignal { - // Thread ID. - int32 tid = 1; - - // Process ID. - int32 pid = 2; - - // Registers at the time of the fault or signal. - Registers registers = 3; - - // Signal number. - int32 signal_number = 4; - - // The memory location which caused the fault (set if applicable, 0 - // otherwise). This will be set for SIGILL, SIGFPE, SIGSEGV, and SIGBUS. - uint64 fault_addr = 5; -} diff --git a/pkg/sentry/kernel/uncaught_signal_go_proto/uncaught_signal.pb.go b/pkg/sentry/kernel/uncaught_signal_go_proto/uncaught_signal.pb.go new file mode 100755 index 000000000..822e549ab --- /dev/null +++ b/pkg/sentry/kernel/uncaught_signal_go_proto/uncaught_signal.pb.go @@ -0,0 +1,119 @@ +// Code generated by protoc-gen-go. DO NOT EDIT. +// source: pkg/sentry/kernel/uncaught_signal.proto + +package gvisor + +import ( + fmt "fmt" + proto "github.com/golang/protobuf/proto" + registers_go_proto "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" + math "math" +) + +// Reference imports to suppress errors if they are not otherwise used. +var _ = proto.Marshal +var _ = fmt.Errorf +var _ = math.Inf + +// This is a compile-time assertion to ensure that this generated file +// is compatible with the proto package it is being compiled against. +// A compilation error at this line likely means your copy of the +// proto package needs to be updated. +const _ = proto.ProtoPackageIsVersion3 // please upgrade the proto package + +type UncaughtSignal struct { + Tid int32 `protobuf:"varint,1,opt,name=tid,proto3" json:"tid,omitempty"` + Pid int32 `protobuf:"varint,2,opt,name=pid,proto3" json:"pid,omitempty"` + Registers *registers_go_proto.Registers `protobuf:"bytes,3,opt,name=registers,proto3" json:"registers,omitempty"` + SignalNumber int32 `protobuf:"varint,4,opt,name=signal_number,json=signalNumber,proto3" json:"signal_number,omitempty"` + FaultAddr uint64 `protobuf:"varint,5,opt,name=fault_addr,json=faultAddr,proto3" json:"fault_addr,omitempty"` + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *UncaughtSignal) Reset() { *m = UncaughtSignal{} } +func (m *UncaughtSignal) String() string { return proto.CompactTextString(m) } +func (*UncaughtSignal) ProtoMessage() {} +func (*UncaughtSignal) Descriptor() ([]byte, []int) { + return fileDescriptor_5ca9e03e13704688, []int{0} +} + +func (m *UncaughtSignal) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_UncaughtSignal.Unmarshal(m, b) +} +func (m *UncaughtSignal) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_UncaughtSignal.Marshal(b, m, deterministic) +} +func (m *UncaughtSignal) XXX_Merge(src proto.Message) { + xxx_messageInfo_UncaughtSignal.Merge(m, src) +} +func (m *UncaughtSignal) XXX_Size() int { + return xxx_messageInfo_UncaughtSignal.Size(m) +} +func (m *UncaughtSignal) XXX_DiscardUnknown() { + xxx_messageInfo_UncaughtSignal.DiscardUnknown(m) +} + +var xxx_messageInfo_UncaughtSignal proto.InternalMessageInfo + +func (m *UncaughtSignal) GetTid() int32 { + if m != nil { + return m.Tid + } + return 0 +} + +func (m *UncaughtSignal) GetPid() int32 { + if m != nil { + return m.Pid + } + return 0 +} + +func (m *UncaughtSignal) GetRegisters() *registers_go_proto.Registers { + if m != nil { + return m.Registers + } + return nil +} + +func (m *UncaughtSignal) GetSignalNumber() int32 { + if m != nil { + return m.SignalNumber + } + return 0 +} + +func (m *UncaughtSignal) GetFaultAddr() uint64 { + if m != nil { + return m.FaultAddr + } + return 0 +} + +func init() { + proto.RegisterType((*UncaughtSignal)(nil), "gvisor.UncaughtSignal") +} + +func init() { + proto.RegisterFile("pkg/sentry/kernel/uncaught_signal.proto", fileDescriptor_5ca9e03e13704688) +} + +var fileDescriptor_5ca9e03e13704688 = []byte{ + // 210 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0x4c, 0x8e, 0x4d, 0x4a, 0xc6, 0x30, + 0x10, 0x86, 0x89, 0xfd, 0x81, 0xc6, 0x1f, 0x34, 0xab, 0x20, 0x88, 0x45, 0x17, 0x76, 0xd5, 0x80, + 0x9e, 0xc0, 0x0b, 0xb8, 0x88, 0xb8, 0x2e, 0x69, 0x13, 0xd3, 0xd0, 0x9a, 0x86, 0x49, 0x22, 0x78, + 0x24, 0x6f, 0x29, 0x4d, 0xd4, 0xef, 0xdb, 0x0d, 0xcf, 0xbc, 0xf3, 0xcc, 0x8b, 0x1f, 0xdc, 0xa2, + 0x99, 0x57, 0x36, 0xc0, 0x17, 0x5b, 0x14, 0x58, 0xb5, 0xb2, 0x68, 0x27, 0x11, 0xf5, 0x1c, 0x06, + 0x6f, 0xb4, 0x15, 0x6b, 0xef, 0x60, 0x0b, 0x1b, 0xa9, 0xf5, 0xa7, 0xf1, 0x1b, 0x5c, 0xdf, 0x1e, + 0x1d, 0x08, 0x98, 0x66, 0x06, 0x4a, 0x1b, 0x1f, 0x14, 0xf8, 0x1c, 0xbc, 0xfb, 0x46, 0xf8, 0xe2, + 0xed, 0x57, 0xf1, 0x9a, 0x0c, 0xe4, 0x12, 0x17, 0xc1, 0x48, 0x8a, 0x5a, 0xd4, 0x55, 0x7c, 0x1f, + 0x77, 0xe2, 0x8c, 0xa4, 0x27, 0x99, 0x38, 0x23, 0x09, 0xc3, 0xcd, 0xbf, 0x89, 0x16, 0x2d, 0xea, + 0x4e, 0x1f, 0xaf, 0xfa, 0xfc, 0xb3, 0xe7, 0x7f, 0x0b, 0x7e, 0xc8, 0x90, 0x7b, 0x7c, 0x9e, 0x0b, + 0x0e, 0x36, 0x7e, 0x8c, 0x0a, 0x68, 0x99, 0x64, 0x67, 0x19, 0xbe, 0x24, 0x46, 0x6e, 0x30, 0x7e, + 0x17, 0x71, 0x0d, 0x83, 0x90, 0x12, 0x68, 0xd5, 0xa2, 0xae, 0xe4, 0x4d, 0x22, 0xcf, 0x52, 0xc2, + 0x58, 0xa7, 0xca, 0x4f, 0x3f, 0x01, 0x00, 0x00, 0xff, 0xff, 0xfd, 0x62, 0x54, 0xdf, 0x06, 0x01, + 0x00, 0x00, +} diff --git a/pkg/sentry/limits/BUILD b/pkg/sentry/limits/BUILD deleted file mode 100644 index cf591c4c1..000000000 --- a/pkg/sentry/limits/BUILD +++ /dev/null @@ -1,27 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "limits", - srcs = [ - "context.go", - "limits.go", - "linux.go", - ], - visibility = ["//:sandbox"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sync", - ], -) - -go_test( - name = "limits_test", - size = "small", - srcs = [ - "limits_test.go", - ], - library = ":limits", -) diff --git a/pkg/sentry/limits/limits_state_autogen.go b/pkg/sentry/limits/limits_state_autogen.go new file mode 100755 index 000000000..aa42533a9 --- /dev/null +++ b/pkg/sentry/limits/limits_state_autogen.go @@ -0,0 +1,36 @@ +// automatically generated by stateify. + +package limits + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Limit) beforeSave() {} +func (x *Limit) save(m state.Map) { + x.beforeSave() + m.Save("Cur", &x.Cur) + m.Save("Max", &x.Max) +} + +func (x *Limit) afterLoad() {} +func (x *Limit) load(m state.Map) { + m.Load("Cur", &x.Cur) + m.Load("Max", &x.Max) +} + +func (x *LimitSet) beforeSave() {} +func (x *LimitSet) save(m state.Map) { + x.beforeSave() + m.Save("data", &x.data) +} + +func (x *LimitSet) afterLoad() {} +func (x *LimitSet) load(m state.Map) { + m.Load("data", &x.data) +} + +func init() { + state.Register("pkg/sentry/limits.Limit", (*Limit)(nil), state.Fns{Save: (*Limit).save, Load: (*Limit).load}) + state.Register("pkg/sentry/limits.LimitSet", (*LimitSet)(nil), state.Fns{Save: (*LimitSet).save, Load: (*LimitSet).load}) +} diff --git a/pkg/sentry/limits/limits_test.go b/pkg/sentry/limits/limits_test.go deleted file mode 100644 index 658a20f56..000000000 --- a/pkg/sentry/limits/limits_test.go +++ /dev/null @@ -1,43 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package limits - -import ( - "syscall" - "testing" -) - -func TestSet(t *testing.T) { - testCases := []struct { - limit Limit - privileged bool - expectedErr error - }{ - {limit: Limit{Cur: 50, Max: 50}, privileged: false, expectedErr: nil}, - {limit: Limit{Cur: 20, Max: 50}, privileged: false, expectedErr: nil}, - {limit: Limit{Cur: 20, Max: 60}, privileged: false, expectedErr: syscall.EPERM}, - {limit: Limit{Cur: 60, Max: 50}, privileged: false, expectedErr: syscall.EINVAL}, - {limit: Limit{Cur: 11, Max: 10}, privileged: false, expectedErr: syscall.EINVAL}, - {limit: Limit{Cur: 20, Max: 60}, privileged: true, expectedErr: nil}, - } - - ls := NewLimitSet() - for _, tc := range testCases { - if _, err := ls.Set(1, tc.limit, tc.privileged); err != tc.expectedErr { - t.Fatalf("Tried to set Limit to %+v and privilege %t: got %v, wanted %v", tc.limit, tc.privileged, err, tc.expectedErr) - } - } - -} diff --git a/pkg/sentry/loader/BUILD b/pkg/sentry/loader/BUILD deleted file mode 100644 index c6aa65f28..000000000 --- a/pkg/sentry/loader/BUILD +++ /dev/null @@ -1,50 +0,0 @@ -load("//tools:defs.bzl", "go_embed_data", "go_library") - -package(licenses = ["notice"]) - -go_embed_data( - name = "vdso_bin", - src = "//vdso:vdso.so", - package = "loader", - var = "vdsoBin", -) - -go_library( - name = "loader", - srcs = [ - "elf.go", - "interpreter.go", - "loader.go", - "vdso.go", - "vdso_state.go", - ":vdso_bin", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi", - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/cpuid", - "//pkg/log", - "//pkg/rand", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fsbridge", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/limits", - "//pkg/sentry/memmap", - "//pkg/sentry/mm", - "//pkg/sentry/pgalloc", - "//pkg/sentry/uniqueid", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/loader/loader_state_autogen.go b/pkg/sentry/loader/loader_state_autogen.go new file mode 100755 index 000000000..e28667944 --- /dev/null +++ b/pkg/sentry/loader/loader_state_autogen.go @@ -0,0 +1,57 @@ +// automatically generated by stateify. + +package loader + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *VDSO) beforeSave() {} +func (x *VDSO) save(m state.Map) { + x.beforeSave() + var phdrs []elfProgHeader = x.savePhdrs() + m.SaveValue("phdrs", phdrs) + m.Save("ParamPage", &x.ParamPage) + m.Save("vdso", &x.vdso) + m.Save("os", &x.os) + m.Save("arch", &x.arch) +} + +func (x *VDSO) afterLoad() {} +func (x *VDSO) load(m state.Map) { + m.Load("ParamPage", &x.ParamPage) + m.Load("vdso", &x.vdso) + m.Load("os", &x.os) + m.Load("arch", &x.arch) + m.LoadValue("phdrs", new([]elfProgHeader), func(y interface{}) { x.loadPhdrs(y.([]elfProgHeader)) }) +} + +func (x *elfProgHeader) beforeSave() {} +func (x *elfProgHeader) save(m state.Map) { + x.beforeSave() + m.Save("Type", &x.Type) + m.Save("Flags", &x.Flags) + m.Save("Off", &x.Off) + m.Save("Vaddr", &x.Vaddr) + m.Save("Paddr", &x.Paddr) + m.Save("Filesz", &x.Filesz) + m.Save("Memsz", &x.Memsz) + m.Save("Align", &x.Align) +} + +func (x *elfProgHeader) afterLoad() {} +func (x *elfProgHeader) load(m state.Map) { + m.Load("Type", &x.Type) + m.Load("Flags", &x.Flags) + m.Load("Off", &x.Off) + m.Load("Vaddr", &x.Vaddr) + m.Load("Paddr", &x.Paddr) + m.Load("Filesz", &x.Filesz) + m.Load("Memsz", &x.Memsz) + m.Load("Align", &x.Align) +} + +func init() { + state.Register("pkg/sentry/loader.VDSO", (*VDSO)(nil), state.Fns{Save: (*VDSO).save, Load: (*VDSO).load}) + state.Register("pkg/sentry/loader.elfProgHeader", (*elfProgHeader)(nil), state.Fns{Save: (*elfProgHeader).save, Load: (*elfProgHeader).load}) +} diff --git a/pkg/sentry/loader/vdso_bin.go b/pkg/sentry/loader/vdso_bin.go new file mode 100755 index 000000000..a5e414e21 --- /dev/null +++ b/pkg/sentry/loader/vdso_bin.go @@ -0,0 +1,5 @@ +// Generated by go_embed_data for //pkg/sentry/loader:vdso_bin. DO NOT EDIT. + +package loader + +var vdsoBin = []byte("ELF\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00>\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00@\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00@\x008\x00\x00@\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00p\xff\xff\xff\xff\xff\x00\x00p\xff\xff\xff\xff\xff\x83\x00\x00\x00\x00\x00\x00\x83\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00P\x00\x00\x00\x00\x00\x00Pp\xff\xff\xff\xff\xffPp\xff\xff\xff\xff\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00P\xe5td\x00\x00\x00@\x00\x00\x00\x00\x00\x00@p\xff\xff\xff\xff\xff@p\xff\xff\xff\xff\xff<\x00\x00\x00\x00\x00\x00\x00<\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\n\x00\x00\x00 \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00 \x00\x00p\xff\xff\xff\xff\xff\"\x00\x00\x00\x00\x00\x00\x00'\x00\x00\x00\x00 \x00pp\xff\xff\xff\xff\xff&\x00\x00\x00\x00\x00\x00\x00<\x00\x00\x00\x00 \x00\xa0p\xff\xff\xff\xff\xff_\x00\x00\x00\x00\x00\x00\x00P\x00\x00\x00\x00 \x000p\xff\xff\xff\xff\xff\n\x00\x00\x00\x00\x00\x00\x00^\x00\x00\x00\"\x00 \x00\x00p\xff\xff\xff\xff\xff\"\x00\x00\x00\x00\x00\x00\x00c\x00\x00\x00\"\x00 \x00pp\xff\xff\xff\xff\xff&\x00\x00\x00\x00\x00\x00\x00q\x00\x00\x00\"\x00 \x00\xa0p\xff\xff\xff\xff\xff_\x00\x00\x00\x00\x00\x00\x00~\x00\x00\x00\"\x00 \x000p\xff\xff\xff\xff\xff\n\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xf1\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00linux-vdso.so.1\x00LINUX_2.6\x00__vdso_time\x00__vdso_clock_gettime\x00__vdso_gettimeofday\x00__vdso_getcpu\x00time\x00clock_gettime\x00gettimeofday\x00getcpu\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xa1\xbf\xee
\x00\x00\x00 \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xf6u\xae\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00;<\x00\x00\x00\x00\x00\x000
\x00\x00X\x00\x00\x00`
\x00\x00p\x00\x00\x00\xc0
\x00\x00\x98\x00\x00\x00\xf0
\x00\x00\xb8\x00\x00\x00\x00\x00\x00\xd0\x00\x00\x00\xa0\x00\x00\xf0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00zR\x00x \x90\x00\x00\x00\x00\x00 \x00\x00\x00\xd0 \x00\x00&\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00$\x00\x00\x004\x00\x00\x00\xe8 \x00\x00_\x00\x00\x00\x00BAD0\x83\x8eTAB \x00\x00\x00\\\x00\x00\x00
\x00\x00\"\x00\x00\x00\x00AD \x83[A\x00\x00\x00\x00|\x00\x00\x000
\x00\x00\n\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00 \x00\x00\x00\x94\x00\x00\x00(
\x00\x00\xa0\x00\x00\x00\x00A\x83\x90AM \x00\x00\x00\xb4\x00\x00\x00\xa8
\x00\x00\xa3\x00\x00\x00\x00A\x83\x90AP\x00\x00\x00\x00\x00\x00\x00`p\xff\xff\xff\xff\xff \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00Pp\xff\xff\xff\xff\xff\n\x00\x00\x00\x00\x00\x00\x00\x85\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00 p\xff\xff\xff\xff\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00 \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xf0\xff\xffo\x00\x00\x00\x00\xd6p\xff\xff\xff\xff\xff\xfc\xff\xffo\x00\x00\x00\x00\xecp\xff\xff\xff\xff\xff\xfd\xff\xffo\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x83\xfft\x83\xfft \x85\xffuH\x89\xf7\xe9\xba\x00\x00\x00H\x89\xf7\xe9R\x00\x00\xb8\xe4\x00\x00\x00\xc3f. \x84\x00\x00\x00\x00\x00AVSH\x83\xecI\x89\xf6H\x85\xfft:H\x89\xfbH\x8d|$\xe8\x84\x00\x00\x00\x85\xc0u7H\x8bD$H\x89H\xb8\xcf\xf7S㥛\xc4 H\xf7l$H\x89\xd0H\xc1\xe8?H\xc1\xfaH\xc2H\x89S1\xc0M\x85\xf6tI\xc7\x00\x00\x00\x00H\x83\xc4[A^ÐSH\x83\xecH\x89\xfbH\x89\xe7\xe80\x00\x00\x00H\x8b$H\x85\xdbtH\x89H\x83\xc4[\xc3f. \x84\x00\x00\x00\x00\x00 @\x00\xb85\x00\x00H\x98Ð\x90\x90\x90\x90\x90SI\x89\xf8H\x8d
\xb5\xde\xff\xffH\x8b1f\x90Hc\xdeH\x83\xe3\xfeH\x8by(L\x8bY0L\x8bI8L\x8bQ@\xae\xe81H\x8b1H9\xdeu\xdcH\x85\xfftYH\xc1\xe2 \x89\xc0H \xd01\xc9L)\xd8HM\xc8H\xb8\x00\x00\x00\x00\x00ʚ;1\xd2I\xf7\xf2H\xf7\xe1H\xa4\xc2 I\xd1L\x89\xc8H\xc1\xe8 H\xb9SZ\x9b\xa0/\xb8D\x00H\xf7\xe1H\xc1\xea Hi\xc2\x00ʚ;I)\xc1I\x89M\x89H1\xc0[\xc31\xffL\x89Ƹ\xe4\x00\x00\x00[\xc3SI\x89\xf8H\x8d
\xde\xff\xffH\x8b1f\x90Hc\xdeH\x83\xe3\xfeH\x8byL\x8bYL\x8bIL\x8bQ \xae\xe81H\x8b1H9\xdeu\xdcH\x85\xfftYH\xc1\xe2 \x89\xc0H \xd01\xc9L)\xd8HM\xc8H\xb8\x00\x00\x00\x00\x00ʚ;1\xd2I\xf7\xf2H\xf7\xe1H\xa4\xc2 I\xd1L\x89\xc8H\xc1\xe8 H\xb9SZ\x9b\xa0/\xb8D\x00H\xf7\xe1H\xc1\xea Hi\xc2\x00ʚ;I)\xc1I\x89M\x89H1\xc0[ÿ\x00\x00\x00L\x89Ƹ\xe4\x00\x00\x00[\xc3\x00clang version 10.0.0 (https://github.com/llvm/llvm-project 407ac2eb5f136af5ddd213b8bcca176481ec5198)\x00\x00\x00\x00\x00\x00\x00 \x00\x00\x00\x00\x00\x00GNU\x00gold 1.11\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xf1\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00 \x00\x00\x00\x00\xf1\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00 \x00@p\xff\xff\xff\xff\xff\xa0\x00\x00\x00\x00\x00\x00\x009\x00\x00\x00\x00 \x00\xe0p\xff\xff\xff\xff\xff\xa3\x00\x00\x00\x00\x00\x00\x00]\x00\x00\x00\x00Pp\xff\xff\xff\xff\xff\x00\x00\x00\x00\x00\x00f\x00\x00\x00\x00\x00\xf1\xff\x00\x00p\xff\xff\xff\xff\xff\x00\x00\x00\x00\x00\x00\x00\x00s\x00\x00\x00\x00\x00\xf1\xff\x00\xf0o\xff\xff\xff\xff\xff\x00\x00\x00\x00\x00\x00\x00\x00{\x00\x00\x00\x00\xf1\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x85\x00\x00\x00\x00 \x00\x00p\xff\xff\xff\xff\xff\"\x00\x00\x00\x00\x00\x00\x00\x91\x00\x00\x00\x00 \x00pp\xff\xff\xff\xff\xff&\x00\x00\x00\x00\x00\x00\x00\xa6\x00\x00\x00\x00 \x00\xa0p\xff\xff\xff\xff\xff_\x00\x00\x00\x00\x00\x00\x00\xba\x00\x00\x00\x00 \x000p\xff\xff\xff\xff\xff\n\x00\x00\x00\x00\x00\x00\x00\xc8\x00\x00\x00\"\x00 \x00\x00p\xff\xff\xff\xff\xff\"\x00\x00\x00\x00\x00\x00\x00\xcd\x00\x00\x00\"\x00 \x00pp\xff\xff\xff\xff\xff&\x00\x00\x00\x00\x00\x00\x00\xdb\x00\x00\x00\"\x00 \x00\xa0p\xff\xff\xff\xff\xff_\x00\x00\x00\x00\x00\x00\x00\xe8\x00\x00\x00\"\x00 \x000p\xff\xff\xff\xff\xff\n\x00\x00\x00\x00\x00\x00\x00\x00vdso.cc\x00vdso_time.cc\x00_ZN4vdso13ClockRealtimeEP8timespec\x00_ZN4vdso14ClockMonotonicEP8timespec\x00_DYNAMIC\x00VDSO_PRELINK\x00_params\x00LINUX_2.6\x00__vdso_time\x00__vdso_clock_gettime\x00__vdso_gettimeofday\x00__vdso_getcpu\x00time\x00clock_gettime\x00gettimeofday\x00getcpu\x00\x00.text\x00.comment\x00.dynstr\x00.eh_frame_hdr\x00.gnu.version\x00.dynsym\x00.hash\x00.note\x00.eh_frame\x00.gnu.version_d\x00.dynamic\x00.shstrtab\x00.strtab\x00.symtab\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00;\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00 p\xff\xff\xff\xff\xff \x00\x00\x00\x00\x00\x00<\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x003\x00\x00\x00 \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00`p\xff\xff\xff\xff\xff`\x00\x00\x00\x00\x00\x00\xf0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00Pp\xff\xff\xff\xff\xffP\x00\x00\x00\x00\x00\x00\x85\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00&\x00\x00\x00\xff\xff\xffo\x00\x00\x00\x00\x00\x00\x00\xd6p\xff\xff\xff\xff\xff\xd6\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00Q\x00\x00\x00\xfd\xff\xffo\x00\x00\x00\x00\x00\x00\x00\xecp\xff\xff\xff\xff\xff\xec\x00\x00\x00\x00\x00\x008\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00@p\xff\xff\xff\xff\xff@\x00\x00\x00\x00\x00\x00<\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00G\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x80p\xff\xff\xff\xff\xff\x80\x00\x00\x00\x00\x00\x00\xd0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00`\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00Pp\xff\xff\xff\xff\xffP\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00pp\xff\xff\xff\xff\xffp\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x000\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x83\x00\x00\x00\x00\x00\x00f\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00A\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xec\x00\x00\x00\x00\x00\x00 \x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00{\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x98\x00\x00\x00\x00\x00\x00
\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00s\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xa0\x00\x00\x00\x00\x00\x00\xef\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00i\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x8f\x00\x00\x00\x00\x00\x00\x83\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00") diff --git a/pkg/sentry/memmap/BUILD b/pkg/sentry/memmap/BUILD deleted file mode 100644 index a98b66de1..000000000 --- a/pkg/sentry/memmap/BUILD +++ /dev/null @@ -1,55 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "mappable_range", - out = "mappable_range.go", - package = "memmap", - prefix = "Mappable", - template = "//pkg/segment:generic_range", - types = { - "T": "uint64", - }, -) - -go_template_instance( - name = "mapping_set_impl", - out = "mapping_set_impl.go", - package = "memmap", - prefix = "Mapping", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "MappableRange", - "Value": "MappingsOfRange", - "Functions": "mappingSetFunctions", - }, -) - -go_library( - name = "memmap", - srcs = [ - "mappable_range.go", - "mapping_set.go", - "mapping_set_impl.go", - "memmap.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/log", - "//pkg/sentry/platform", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -go_test( - name = "memmap_test", - size = "small", - srcs = ["mapping_set_test.go"], - library = ":memmap", - deps = ["//pkg/usermem"], -) diff --git a/pkg/sentry/memmap/mappable_range.go b/pkg/sentry/memmap/mappable_range.go new file mode 100755 index 000000000..6b6c2c685 --- /dev/null +++ b/pkg/sentry/memmap/mappable_range.go @@ -0,0 +1,62 @@ +package memmap + +// A Range represents a contiguous range of T. +// +// +stateify savable +type MappableRange struct { + // Start is the inclusive start of the range. + Start uint64 + + // End is the exclusive end of the range. + End uint64 +} + +// WellFormed returns true if r.Start <= r.End. All other methods on a Range +// require that the Range is well-formed. +func (r MappableRange) WellFormed() bool { + return r.Start <= r.End +} + +// Length returns the length of the range. +func (r MappableRange) Length() uint64 { + return r.End - r.Start +} + +// Contains returns true if r contains x. +func (r MappableRange) Contains(x uint64) bool { + return r.Start <= x && x < r.End +} + +// Overlaps returns true if r and r2 overlap. +func (r MappableRange) Overlaps(r2 MappableRange) bool { + return r.Start < r2.End && r2.Start < r.End +} + +// IsSupersetOf returns true if r is a superset of r2; that is, the range r2 is +// contained within r. +func (r MappableRange) IsSupersetOf(r2 MappableRange) bool { + return r.Start <= r2.Start && r.End >= r2.End +} + +// Intersect returns a range consisting of the intersection between r and r2. +// If r and r2 do not overlap, Intersect returns a range with unspecified +// bounds, but for which Length() == 0. +func (r MappableRange) Intersect(r2 MappableRange) MappableRange { + if r.Start < r2.Start { + r.Start = r2.Start + } + if r.End > r2.End { + r.End = r2.End + } + if r.End < r.Start { + r.End = r.Start + } + return r +} + +// CanSplitAt returns true if it is legal to split a segment spanning the range +// r at x; that is, splitting at x would produce two ranges, both of which have +// non-zero length. +func (r MappableRange) CanSplitAt(x uint64) bool { + return r.Contains(x) && r.Start < x +} diff --git a/pkg/sentry/memmap/mapping_set_impl.go b/pkg/sentry/memmap/mapping_set_impl.go new file mode 100755 index 000000000..e632f28a5 --- /dev/null +++ b/pkg/sentry/memmap/mapping_set_impl.go @@ -0,0 +1,1270 @@ +package memmap + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + MappingminDegree = 3 + + MappingmaxDegree = 2 * MappingminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type MappingSet struct { + root Mappingnode `state:".(*MappingSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *MappingSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *MappingSet) IsEmptyRange(r MappableRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *MappingSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *MappingSet) SpanRange(r MappableRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *MappingSet) FirstSegment() MappingIterator { + if s.root.nrSegments == 0 { + return MappingIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *MappingSet) LastSegment() MappingIterator { + if s.root.nrSegments == 0 { + return MappingIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *MappingSet) FirstGap() MappingGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return MappingGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *MappingSet) LastGap() MappingGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return MappingGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *MappingSet) Find(key uint64) (MappingIterator, MappingGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return MappingIterator{n, i}, MappingGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return MappingIterator{}, MappingGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *MappingSet) FindSegment(key uint64) MappingIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *MappingSet) LowerBoundSegment(min uint64) MappingIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *MappingSet) UpperBoundSegment(max uint64) MappingIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *MappingSet) FindGap(key uint64) MappingGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *MappingSet) LowerBoundGap(min uint64) MappingGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *MappingSet) UpperBoundGap(max uint64) MappingGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *MappingSet) Add(r MappableRange, val MappingsOfRange) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *MappingSet) AddWithoutMerging(r MappableRange, val MappingsOfRange) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *MappingSet) Insert(gap MappingGapIterator, r MappableRange, val MappingsOfRange) MappingIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (mappingSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (mappingSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (mappingSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *MappingSet) InsertWithoutMerging(gap MappingGapIterator, r MappableRange, val MappingsOfRange) MappingIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *MappingSet) InsertWithoutMergingUnchecked(gap MappingGapIterator, r MappableRange, val MappingsOfRange) MappingIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return MappingIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *MappingSet) Remove(seg MappingIterator) MappingGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + mappingSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(MappingGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *MappingSet) RemoveAll() { + s.root = Mappingnode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *MappingSet) RemoveRange(r MappableRange) MappingGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *MappingSet) Merge(first, second MappingIterator) MappingIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *MappingSet) MergeUnchecked(first, second MappingIterator) MappingIterator { + if first.End() == second.Start() { + if mval, ok := (mappingSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return MappingIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *MappingSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *MappingSet) MergeRange(r MappableRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *MappingSet) MergeAdjacent(r MappableRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *MappingSet) Split(seg MappingIterator, split uint64) (MappingIterator, MappingIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *MappingSet) SplitUnchecked(seg MappingIterator, split uint64) (MappingIterator, MappingIterator) { + val1, val2 := (mappingSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), MappableRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *MappingSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *MappingSet) Isolate(seg MappingIterator, r MappableRange) MappingIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *MappingSet) ApplyContiguous(r MappableRange, fn func(seg MappingIterator)) MappingGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return MappingGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return MappingGapIterator{} + } + } +} + +// +stateify savable +type Mappingnode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *Mappingnode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [MappingmaxDegree - 1]MappableRange + values [MappingmaxDegree - 1]MappingsOfRange + children [MappingmaxDegree]*Mappingnode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Mappingnode) firstSegment() MappingIterator { + for n.hasChildren { + n = n.children[0] + } + return MappingIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Mappingnode) lastSegment() MappingIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return MappingIterator{n, n.nrSegments - 1} +} + +func (n *Mappingnode) prevSibling() *Mappingnode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *Mappingnode) nextSibling() *Mappingnode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *Mappingnode) rebalanceBeforeInsert(gap MappingGapIterator) MappingGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < MappingmaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &Mappingnode{ + nrSegments: MappingminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &Mappingnode{ + nrSegments: MappingminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:MappingminDegree-1], n.keys[:MappingminDegree-1]) + copy(left.values[:MappingminDegree-1], n.values[:MappingminDegree-1]) + copy(right.keys[:MappingminDegree-1], n.keys[MappingminDegree:]) + copy(right.values[:MappingminDegree-1], n.values[MappingminDegree:]) + n.keys[0], n.values[0] = n.keys[MappingminDegree-1], n.values[MappingminDegree-1] + MappingzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:MappingminDegree], n.children[:MappingminDegree]) + copy(right.children[:MappingminDegree], n.children[MappingminDegree:]) + MappingzeroNodeSlice(n.children[2:]) + for i := 0; i < MappingminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < MappingminDegree { + return MappingGapIterator{left, gap.index} + } + return MappingGapIterator{right, gap.index - MappingminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[MappingminDegree-1], n.values[MappingminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &Mappingnode{ + nrSegments: MappingminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:MappingminDegree-1], n.keys[MappingminDegree:]) + copy(sibling.values[:MappingminDegree-1], n.values[MappingminDegree:]) + MappingzeroValueSlice(n.values[MappingminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:MappingminDegree], n.children[MappingminDegree:]) + MappingzeroNodeSlice(n.children[MappingminDegree:]) + for i := 0; i < MappingminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = MappingminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < MappingminDegree { + return gap + } + return MappingGapIterator{sibling, gap.index - MappingminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *Mappingnode) rebalanceAfterRemove(gap MappingGapIterator) MappingGapIterator { + for { + if n.nrSegments >= MappingminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= MappingminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + mappingSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return MappingGapIterator{n, 0} + } + if gap.node == n { + return MappingGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= MappingminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + mappingSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return MappingGapIterator{n, n.nrSegments} + } + return MappingGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return MappingGapIterator{p, gap.index} + } + if gap.node == right { + return MappingGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *Mappingnode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = MappingGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + mappingSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type MappingIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *Mappingnode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg MappingIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg MappingIterator) Range() MappableRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg MappingIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg MappingIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg MappingIterator) SetRangeUnchecked(r MappableRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg MappingIterator) SetRange(r MappableRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg MappingIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg MappingIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg MappingIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg MappingIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg MappingIterator) Value() MappingsOfRange { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg MappingIterator) ValuePtr() *MappingsOfRange { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg MappingIterator) SetValue(val MappingsOfRange) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg MappingIterator) PrevSegment() MappingIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return MappingIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return MappingIterator{} + } + return MappingsegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg MappingIterator) NextSegment() MappingIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return MappingIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return MappingIterator{} + } + return MappingsegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg MappingIterator) PrevGap() MappingGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return MappingGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg MappingIterator) NextGap() MappingGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return MappingGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg MappingIterator) PrevNonEmpty() (MappingIterator, MappingGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return MappingIterator{}, gap + } + return gap.PrevSegment(), MappingGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg MappingIterator) NextNonEmpty() (MappingIterator, MappingGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return MappingIterator{}, gap + } + return gap.NextSegment(), MappingGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type MappingGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *Mappingnode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap MappingGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap MappingGapIterator) Range() MappableRange { + return MappableRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap MappingGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return mappingSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap MappingGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return mappingSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap MappingGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap MappingGapIterator) PrevSegment() MappingIterator { + return MappingsegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap MappingGapIterator) NextSegment() MappingIterator { + return MappingsegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap MappingGapIterator) PrevGap() MappingGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return MappingGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap MappingGapIterator) NextGap() MappingGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return MappingGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func MappingsegmentBeforePosition(n *Mappingnode, i int) MappingIterator { + for i == 0 { + if n.parent == nil { + return MappingIterator{} + } + n, i = n.parent, n.parentIndex + } + return MappingIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func MappingsegmentAfterPosition(n *Mappingnode, i int) MappingIterator { + for i == n.nrSegments { + if n.parent == nil { + return MappingIterator{} + } + n, i = n.parent, n.parentIndex + } + return MappingIterator{n, i} +} + +func MappingzeroValueSlice(slice []MappingsOfRange) { + + for i := range slice { + mappingSetFunctions{}.ClearValue(&slice[i]) + } +} + +func MappingzeroNodeSlice(slice []*Mappingnode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *MappingSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *Mappingnode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *Mappingnode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type MappingSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []MappingsOfRange +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *MappingSet) ExportSortedSlices() *MappingSegmentDataSlices { + var sds MappingSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *MappingSet) ImportSortedSlices(sds *MappingSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := MappableRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *MappingSet) saveRoot() *MappingSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *MappingSet) loadRoot(sds *MappingSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/memmap/mapping_set_test.go b/pkg/sentry/memmap/mapping_set_test.go deleted file mode 100644 index d39efe38f..000000000 --- a/pkg/sentry/memmap/mapping_set_test.go +++ /dev/null @@ -1,260 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package memmap - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/usermem" -) - -type testMappingSpace struct { - // Ideally we'd store the full ranges that were invalidated, rather - // than individual calls to Invalidate, as they are an implementation - // detail, but this is the simplest way for now. - inv []usermem.AddrRange -} - -func (n *testMappingSpace) reset() { - n.inv = []usermem.AddrRange{} -} - -func (n *testMappingSpace) Invalidate(ar usermem.AddrRange, opts InvalidateOpts) { - n.inv = append(n.inv, ar) -} - -func TestAddRemoveMapping(t *testing.T) { - set := MappingSet{} - ms := &testMappingSpace{} - - mapped := set.AddMapping(ms, usermem.AddrRange{0x10000, 0x12000}, 0x1000, true) - if got, want := mapped, []MappableRange{{0x1000, 0x3000}}; !reflect.DeepEqual(got, want) { - t.Errorf("AddMapping: got %+v, wanted %+v", got, want) - } - - // Mappings (usermem.AddrRanges => memmap.MappableRange): - // [0x10000, 0x12000) => [0x1000, 0x3000) - t.Log(&set) - - mapped = set.AddMapping(ms, usermem.AddrRange{0x20000, 0x21000}, 0x2000, true) - if len(mapped) != 0 { - t.Errorf("AddMapping: got %+v, wanted []", mapped) - } - - // Mappings: - // [0x10000, 0x11000) => [0x1000, 0x2000) - // [0x11000, 0x12000) and [0x20000, 0x21000) => [0x2000, 0x3000) - t.Log(&set) - - mapped = set.AddMapping(ms, usermem.AddrRange{0x30000, 0x31000}, 0x4000, true) - if got, want := mapped, []MappableRange{{0x4000, 0x5000}}; !reflect.DeepEqual(got, want) { - t.Errorf("AddMapping: got %+v, wanted %+v", got, want) - } - - // Mappings: - // [0x10000, 0x11000) => [0x1000, 0x2000) - // [0x11000, 0x12000) and [0x20000, 0x21000) => [0x2000, 0x3000) - // [0x30000, 0x31000) => [0x4000, 0x5000) - t.Log(&set) - - mapped = set.AddMapping(ms, usermem.AddrRange{0x12000, 0x15000}, 0x3000, true) - if got, want := mapped, []MappableRange{{0x3000, 0x4000}, {0x5000, 0x6000}}; !reflect.DeepEqual(got, want) { - t.Errorf("AddMapping: got %+v, wanted %+v", got, want) - } - - // Mappings: - // [0x10000, 0x11000) => [0x1000, 0x2000) - // [0x11000, 0x12000) and [0x20000, 0x21000) => [0x2000, 0x3000) - // [0x12000, 0x13000) => [0x3000, 0x4000) - // [0x13000, 0x14000) and [0x30000, 0x31000) => [0x4000, 0x5000) - // [0x14000, 0x15000) => [0x5000, 0x6000) - t.Log(&set) - - unmapped := set.RemoveMapping(ms, usermem.AddrRange{0x10000, 0x11000}, 0x1000, true) - if got, want := unmapped, []MappableRange{{0x1000, 0x2000}}; !reflect.DeepEqual(got, want) { - t.Errorf("RemoveMapping: got %+v, wanted %+v", got, want) - } - - // Mappings: - // [0x11000, 0x12000) and [0x20000, 0x21000) => [0x2000, 0x3000) - // [0x12000, 0x13000) => [0x3000, 0x4000) - // [0x13000, 0x14000) and [0x30000, 0x31000) => [0x4000, 0x5000) - // [0x14000, 0x15000) => [0x5000, 0x6000) - t.Log(&set) - - unmapped = set.RemoveMapping(ms, usermem.AddrRange{0x20000, 0x21000}, 0x2000, true) - if len(unmapped) != 0 { - t.Errorf("RemoveMapping: got %+v, wanted []", unmapped) - } - - // Mappings: - // [0x11000, 0x13000) => [0x2000, 0x4000) - // [0x13000, 0x14000) and [0x30000, 0x31000) => [0x4000, 0x5000) - // [0x14000, 0x15000) => [0x5000, 0x6000) - t.Log(&set) - - unmapped = set.RemoveMapping(ms, usermem.AddrRange{0x11000, 0x15000}, 0x2000, true) - if got, want := unmapped, []MappableRange{{0x2000, 0x4000}, {0x5000, 0x6000}}; !reflect.DeepEqual(got, want) { - t.Errorf("RemoveMapping: got %+v, wanted %+v", got, want) - } - - // Mappings: - // [0x30000, 0x31000) => [0x4000, 0x5000) - t.Log(&set) - - unmapped = set.RemoveMapping(ms, usermem.AddrRange{0x30000, 0x31000}, 0x4000, true) - if got, want := unmapped, []MappableRange{{0x4000, 0x5000}}; !reflect.DeepEqual(got, want) { - t.Errorf("RemoveMapping: got %+v, wanted %+v", got, want) - } -} - -func TestInvalidateWholeMapping(t *testing.T) { - set := MappingSet{} - ms := &testMappingSpace{} - - set.AddMapping(ms, usermem.AddrRange{0x10000, 0x11000}, 0, true) - // Mappings: - // [0x10000, 0x11000) => [0, 0x1000) - t.Log(&set) - set.Invalidate(MappableRange{0, 0x1000}, InvalidateOpts{}) - if got, want := ms.inv, []usermem.AddrRange{{0x10000, 0x11000}}; !reflect.DeepEqual(got, want) { - t.Errorf("Invalidate: got %+v, wanted %+v", got, want) - } -} - -func TestInvalidatePartialMapping(t *testing.T) { - set := MappingSet{} - ms := &testMappingSpace{} - - set.AddMapping(ms, usermem.AddrRange{0x10000, 0x13000}, 0, true) - // Mappings: - // [0x10000, 0x13000) => [0, 0x3000) - t.Log(&set) - set.Invalidate(MappableRange{0x1000, 0x2000}, InvalidateOpts{}) - if got, want := ms.inv, []usermem.AddrRange{{0x11000, 0x12000}}; !reflect.DeepEqual(got, want) { - t.Errorf("Invalidate: got %+v, wanted %+v", got, want) - } -} - -func TestInvalidateMultipleMappings(t *testing.T) { - set := MappingSet{} - ms := &testMappingSpace{} - - set.AddMapping(ms, usermem.AddrRange{0x10000, 0x11000}, 0, true) - set.AddMapping(ms, usermem.AddrRange{0x20000, 0x21000}, 0x2000, true) - // Mappings: - // [0x10000, 0x11000) => [0, 0x1000) - // [0x12000, 0x13000) => [0x2000, 0x3000) - t.Log(&set) - set.Invalidate(MappableRange{0, 0x3000}, InvalidateOpts{}) - if got, want := ms.inv, []usermem.AddrRange{{0x10000, 0x11000}, {0x20000, 0x21000}}; !reflect.DeepEqual(got, want) { - t.Errorf("Invalidate: got %+v, wanted %+v", got, want) - } -} - -func TestInvalidateOverlappingMappings(t *testing.T) { - set := MappingSet{} - ms1 := &testMappingSpace{} - ms2 := &testMappingSpace{} - - set.AddMapping(ms1, usermem.AddrRange{0x10000, 0x12000}, 0, true) - set.AddMapping(ms2, usermem.AddrRange{0x20000, 0x22000}, 0x1000, true) - // Mappings: - // ms1:[0x10000, 0x12000) => [0, 0x2000) - // ms2:[0x11000, 0x13000) => [0x1000, 0x3000) - t.Log(&set) - set.Invalidate(MappableRange{0x1000, 0x2000}, InvalidateOpts{}) - if got, want := ms1.inv, []usermem.AddrRange{{0x11000, 0x12000}}; !reflect.DeepEqual(got, want) { - t.Errorf("Invalidate: ms1: got %+v, wanted %+v", got, want) - } - if got, want := ms2.inv, []usermem.AddrRange{{0x20000, 0x21000}}; !reflect.DeepEqual(got, want) { - t.Errorf("Invalidate: ms1: got %+v, wanted %+v", got, want) - } -} - -func TestMixedWritableMappings(t *testing.T) { - set := MappingSet{} - ms := &testMappingSpace{} - - mapped := set.AddMapping(ms, usermem.AddrRange{0x10000, 0x12000}, 0x1000, true) - if got, want := mapped, []MappableRange{{0x1000, 0x3000}}; !reflect.DeepEqual(got, want) { - t.Errorf("AddMapping: got %+v, wanted %+v", got, want) - } - - // Mappings: - // [0x10000, 0x12000) writable => [0x1000, 0x3000) - t.Log(&set) - - mapped = set.AddMapping(ms, usermem.AddrRange{0x20000, 0x22000}, 0x2000, false) - if got, want := mapped, []MappableRange{{0x3000, 0x4000}}; !reflect.DeepEqual(got, want) { - t.Errorf("AddMapping: got %+v, wanted %+v", got, want) - } - - // Mappings: - // [0x10000, 0x11000) writable => [0x1000, 0x2000) - // [0x11000, 0x12000) writable and [0x20000, 0x21000) readonly => [0x2000, 0x3000) - // [0x21000, 0x22000) readonly => [0x3000, 0x4000) - t.Log(&set) - - // Unmap should fail because we specified the readonly map address range, but - // asked to unmap a writable segment. - unmapped := set.RemoveMapping(ms, usermem.AddrRange{0x20000, 0x21000}, 0x2000, true) - if len(unmapped) != 0 { - t.Errorf("RemoveMapping: got %+v, wanted []", unmapped) - } - - // Readonly mapping removed, but writable mapping still exists in the range, - // so no mappable range fully unmapped. - unmapped = set.RemoveMapping(ms, usermem.AddrRange{0x20000, 0x21000}, 0x2000, false) - if len(unmapped) != 0 { - t.Errorf("RemoveMapping: got %+v, wanted []", unmapped) - } - - // Mappings: - // [0x10000, 0x12000) writable => [0x1000, 0x3000) - // [0x21000, 0x22000) readonly => [0x3000, 0x4000) - t.Log(&set) - - unmapped = set.RemoveMapping(ms, usermem.AddrRange{0x11000, 0x12000}, 0x2000, true) - if got, want := unmapped, []MappableRange{{0x2000, 0x3000}}; !reflect.DeepEqual(got, want) { - t.Errorf("RemoveMapping: got %+v, wanted %+v", got, want) - } - - // Mappings: - // [0x10000, 0x12000) writable => [0x1000, 0x3000) - // [0x21000, 0x22000) readonly => [0x3000, 0x4000) - t.Log(&set) - - // Unmap should fail since writable bit doesn't match. - unmapped = set.RemoveMapping(ms, usermem.AddrRange{0x10000, 0x12000}, 0x1000, false) - if len(unmapped) != 0 { - t.Errorf("RemoveMapping: got %+v, wanted []", unmapped) - } - - unmapped = set.RemoveMapping(ms, usermem.AddrRange{0x10000, 0x12000}, 0x1000, true) - if got, want := unmapped, []MappableRange{{0x1000, 0x2000}}; !reflect.DeepEqual(got, want) { - t.Errorf("RemoveMapping: got %+v, wanted %+v", got, want) - } - - // Mappings: - // [0x21000, 0x22000) readonly => [0x3000, 0x4000) - t.Log(&set) - - unmapped = set.RemoveMapping(ms, usermem.AddrRange{0x21000, 0x22000}, 0x3000, false) - if got, want := unmapped, []MappableRange{{0x3000, 0x4000}}; !reflect.DeepEqual(got, want) { - t.Errorf("RemoveMapping: got %+v, wanted %+v", got, want) - } -} diff --git a/pkg/sentry/memmap/memmap_impl_state_autogen.go b/pkg/sentry/memmap/memmap_impl_state_autogen.go new file mode 100755 index 000000000..b231fd9c3 --- /dev/null +++ b/pkg/sentry/memmap/memmap_impl_state_autogen.go @@ -0,0 +1,63 @@ +// automatically generated by stateify. + +package memmap + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *MappingSet) beforeSave() {} +func (x *MappingSet) save(m state.Map) { + x.beforeSave() + var root *MappingSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *MappingSet) afterLoad() {} +func (x *MappingSet) load(m state.Map) { + m.LoadValue("root", new(*MappingSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*MappingSegmentDataSlices)) }) +} + +func (x *Mappingnode) beforeSave() {} +func (x *Mappingnode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *Mappingnode) afterLoad() {} +func (x *Mappingnode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *MappingSegmentDataSlices) beforeSave() {} +func (x *MappingSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *MappingSegmentDataSlices) afterLoad() {} +func (x *MappingSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func init() { + state.Register("pkg/sentry/memmap.MappingSet", (*MappingSet)(nil), state.Fns{Save: (*MappingSet).save, Load: (*MappingSet).load}) + state.Register("pkg/sentry/memmap.Mappingnode", (*Mappingnode)(nil), state.Fns{Save: (*Mappingnode).save, Load: (*Mappingnode).load}) + state.Register("pkg/sentry/memmap.MappingSegmentDataSlices", (*MappingSegmentDataSlices)(nil), state.Fns{Save: (*MappingSegmentDataSlices).save, Load: (*MappingSegmentDataSlices).load}) +} diff --git a/pkg/sentry/memmap/memmap_state_autogen.go b/pkg/sentry/memmap/memmap_state_autogen.go new file mode 100755 index 000000000..2072dbad2 --- /dev/null +++ b/pkg/sentry/memmap/memmap_state_autogen.go @@ -0,0 +1,40 @@ +// automatically generated by stateify. + +package memmap + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *MappableRange) beforeSave() {} +func (x *MappableRange) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) +} + +func (x *MappableRange) afterLoad() {} +func (x *MappableRange) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) +} + +func (x *MappingOfRange) beforeSave() {} +func (x *MappingOfRange) save(m state.Map) { + x.beforeSave() + m.Save("MappingSpace", &x.MappingSpace) + m.Save("AddrRange", &x.AddrRange) + m.Save("Writable", &x.Writable) +} + +func (x *MappingOfRange) afterLoad() {} +func (x *MappingOfRange) load(m state.Map) { + m.Load("MappingSpace", &x.MappingSpace) + m.Load("AddrRange", &x.AddrRange) + m.Load("Writable", &x.Writable) +} + +func init() { + state.Register("pkg/sentry/memmap.MappableRange", (*MappableRange)(nil), state.Fns{Save: (*MappableRange).save, Load: (*MappableRange).load}) + state.Register("pkg/sentry/memmap.MappingOfRange", (*MappingOfRange)(nil), state.Fns{Save: (*MappingOfRange).save, Load: (*MappingOfRange).load}) +} diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD deleted file mode 100644 index 73591dab7..000000000 --- a/pkg/sentry/mm/BUILD +++ /dev/null @@ -1,141 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "file_refcount_set", - out = "file_refcount_set.go", - imports = { - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", - }, - package = "mm", - prefix = "fileRefcount", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "platform.FileRange", - "Value": "int32", - "Functions": "fileRefcountSetFunctions", - }, -) - -go_template_instance( - name = "vma_set", - out = "vma_set.go", - consts = { - "minDegree": "8", - }, - imports = { - "usermem": "gvisor.dev/gvisor/pkg/usermem", - }, - package = "mm", - prefix = "vma", - template = "//pkg/segment:generic_set", - types = { - "Key": "usermem.Addr", - "Range": "usermem.AddrRange", - "Value": "vma", - "Functions": "vmaSetFunctions", - }, -) - -go_template_instance( - name = "pma_set", - out = "pma_set.go", - consts = { - "minDegree": "8", - }, - imports = { - "usermem": "gvisor.dev/gvisor/pkg/usermem", - }, - package = "mm", - prefix = "pma", - template = "//pkg/segment:generic_set", - types = { - "Key": "usermem.Addr", - "Range": "usermem.AddrRange", - "Value": "pma", - "Functions": "pmaSetFunctions", - }, -) - -go_template_instance( - name = "io_list", - out = "io_list.go", - package = "mm", - prefix = "io", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*ioResult", - "Linker": "*ioResult", - }, -) - -go_library( - name = "mm", - srcs = [ - "address_space.go", - "aio_context.go", - "aio_context_state.go", - "debug.go", - "file_refcount_set.go", - "io.go", - "io_list.go", - "lifecycle.go", - "metadata.go", - "mm.go", - "pma.go", - "pma_set.go", - "procfs.go", - "save_restore.go", - "shm.go", - "special_mappable.go", - "syscalls.go", - "vma.go", - "vma_set.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/atomicbitops", - "//pkg/context", - "//pkg/log", - "//pkg/refs", - "//pkg/safecopy", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fs/proc/seqfile", - "//pkg/sentry/fsbridge", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/futex", - "//pkg/sentry/kernel/shm", - "//pkg/sentry/limits", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/usage", - "//pkg/sync", - "//pkg/syserror", - "//pkg/tcpip/buffer", - "//pkg/usermem", - ], -) - -go_test( - name = "mm_test", - size = "small", - srcs = ["mm_test.go"], - library = ":mm", - deps = [ - "//pkg/context", - "//pkg/sentry/arch", - "//pkg/sentry/contexttest", - "//pkg/sentry/limits", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/mm/README.md b/pkg/sentry/mm/README.md deleted file mode 100644 index f4d43d927..000000000 --- a/pkg/sentry/mm/README.md +++ /dev/null @@ -1,280 +0,0 @@ -This package provides an emulation of Linux semantics for application virtual -memory mappings. - -For completeness, this document also describes aspects of the memory management -subsystem defined outside this package. - -# Background - -We begin by describing semantics for virtual memory in Linux. - -A virtual address space is defined as a collection of mappings from virtual -addresses to physical memory. However, userspace applications do not configure -mappings to physical memory directly. Instead, applications configure memory -mappings from virtual addresses to offsets into a file using the `mmap` system -call.[^mmap-anon] For example, a call to: - - mmap( - /* addr = */ 0x400000, - /* length = */ 0x1000, - PROT_READ | PROT_WRITE, - MAP_SHARED, - /* fd = */ 3, - /* offset = */ 0); - -creates a mapping of length 0x1000 bytes, starting at virtual address (VA) -0x400000, to offset 0 in the file represented by file descriptor (FD) 3. Within -the Linux kernel, virtual memory mappings are represented by *virtual memory -areas* (VMAs). Supposing that FD 3 represents file /tmp/foo, the state of the -virtual memory subsystem after the `mmap` call may be depicted as: - - VMA: VA:0x400000 -> /tmp/foo:0x0 - -Establishing a virtual memory area does not necessarily establish a mapping to a -physical address, because Linux has not necessarily provisioned physical memory -to store the file's contents. Thus, if the application attempts to read the -contents of VA 0x400000, it may incur a *page fault*, a CPU exception that -forces the kernel to create such a mapping to service the read. - -For a file, doing so consists of several logical phases: - -1. The kernel allocates physical memory to store the contents of the required - part of the file, and copies file contents to the allocated memory. - Supposing that the kernel chooses the physical memory at physical address - (PA) 0x2fb000, the resulting state of the system is: - - VMA: VA:0x400000 -> /tmp/foo:0x0 - Filemap: /tmp/foo:0x0 -> PA:0x2fb000 - - (In Linux the state of the mapping from file offset to physical memory is - stored in `struct address_space`, but to avoid confusion with other notions - of address space we will refer to this system as filemap, named after Linux - kernel source file `mm/filemap.c`.) - -2. The kernel stores the effective mapping from virtual to physical address in - a *page table entry* (PTE) in the application's *page tables*, which are - used by the CPU's virtual memory hardware to perform address translation. - The resulting state of the system is: - - VMA: VA:0x400000 -> /tmp/foo:0x0 - Filemap: /tmp/foo:0x0 -> PA:0x2fb000 - PTE: VA:0x400000 -----------------> PA:0x2fb000 - - The PTE is required for the application to actually use the contents of the - mapped file as virtual memory. However, the PTE is derived from the VMA and - filemap state, both of which are independently mutable, such that mutations - to either will affect the PTE. For example: - - - The application may remove the VMA using the `munmap` system call. This - breaks the mapping from VA:0x400000 to /tmp/foo:0x0, and consequently - the mapping from VA:0x400000 to PA:0x2fb000. However, it does not - necessarily break the mapping from /tmp/foo:0x0 to PA:0x2fb000, so a - future mapping of the same file offset may reuse this physical memory. - - - The application may invalidate the file's contents by passing a length - of 0 to the `ftruncate` system call. This breaks the mapping from - /tmp/foo:0x0 to PA:0x2fb000, and consequently the mapping from - VA:0x400000 to PA:0x2fb000. However, it does not break the mapping from - VA:0x400000 to /tmp/foo:0x0, so future changes to the file's contents - may again be made visible at VA:0x400000 after another page fault - results in the allocation of a new physical address. - - Note that, in order to correctly break the mapping from VA:0x400000 to - PA:0x2fb000 in the latter case, filemap must also store a *reverse mapping* - from /tmp/foo:0x0 to VA:0x400000 so that it can locate and remove the PTE. - -[^mmap-anon]: Memory mappings to non-files are discussed in later sections. - -## Private Mappings - -The preceding example considered VMAs created using the `MAP_SHARED` flag, which -means that PTEs derived from the mapping should always use physical memory that -represents the current state of the mapped file.[^mmap-dev-zero] Applications -can alternatively pass the `MAP_PRIVATE` flag to create a *private mapping*. -Private mappings are *copy-on-write*. - -Suppose that the application instead created a private mapping in the previous -example. In Linux, the state of the system after a read page fault would be: - - VMA: VA:0x400000 -> /tmp/foo:0x0 (private) - Filemap: /tmp/foo:0x0 -> PA:0x2fb000 - PTE: VA:0x400000 -----------------> PA:0x2fb000 (read-only) - -Now suppose the application attempts to write to VA:0x400000. For a shared -mapping, the write would be propagated to PA:0x2fb000, and the kernel would be -responsible for ensuring that the write is later propagated to the mapped file. -For a private mapping, the write incurs another page fault since the PTE is -marked read-only. In response, the kernel allocates physical memory to store the -mapping's *private copy* of the file's contents, copies file contents to the -allocated memory, and changes the PTE to map to the private copy. Supposing that -the kernel chooses the physical memory at physical address (PA) 0x5ea000, the -resulting state of the system is: - - VMA: VA:0x400000 -> /tmp/foo:0x0 (private) - Filemap: /tmp/foo:0x0 -> PA:0x2fb000 - PTE: VA:0x400000 -----------------> PA:0x5ea000 - -Note that the filemap mapping from /tmp/foo:0x0 to PA:0x2fb000 may still exist, -but is now irrelevant to this mapping. - -[^mmap-dev-zero]: Modulo files with special mmap semantics such as `/dev/zero`. - -## Anonymous Mappings - -Instead of passing a file to the `mmap` system call, applications can instead -request an *anonymous* mapping by passing the `MAP_ANONYMOUS` flag. -Semantically, an anonymous mapping is essentially a mapping to an ephemeral file -initially filled with zero bytes. Practically speaking, this is how shared -anonymous mappings are implemented, but private anonymous mappings do not result -in the creation of an ephemeral file; since there would be no way to modify the -contents of the underlying file through a private mapping, all private anonymous -mappings use a single shared page filled with zero bytes until copy-on-write -occurs. - -# Virtual Memory in the Sentry - -The sentry implements application virtual memory atop a host kernel, introducing -an additional level of indirection to the above. - -Consider the same scenario as in the previous section. Since the sentry handles -application system calls, the effect of an application `mmap` system call is to -create a VMA in the sentry (as opposed to the host kernel): - - Sentry VMA: VA:0x400000 -> /tmp/foo:0x0 - -When the application first incurs a page fault on this address, the host kernel -delivers information about the page fault to the sentry in a platform-dependent -manner, and the sentry handles the fault: - -1. The sentry allocates memory to store the contents of the required part of - the file, and copies file contents to the allocated memory. However, since - the sentry is implemented atop a host kernel, it does not configure mappings - to physical memory directly. Instead, mappable "memory" in the sentry is - represented by a host file descriptor and offset, since (as noted in - "Background") this is the memory mapping primitive provided by the host - kernel. In general, memory is allocated from a temporary host file using the - `pgalloc` package. Supposing that the sentry allocates offset 0x3000 from - host file "memory-file", the resulting state is: - - Sentry VMA: VA:0x400000 -> /tmp/foo:0x0 - Sentry filemap: /tmp/foo:0x0 -> host:memory-file:0x3000 - -2. The sentry stores the effective mapping from virtual address to host file in - a host VMA by invoking the `mmap` system call: - - Sentry VMA: VA:0x400000 -> /tmp/foo:0x0 - Sentry filemap: /tmp/foo:0x0 -> host:memory-file:0x3000 - Host VMA: VA:0x400000 -----------------> host:memory-file:0x3000 - -3. The sentry returns control to the application, which immediately incurs the - page fault again.[^mmap-populate] However, since a host VMA now exists for - the faulting virtual address, the host kernel now handles the page fault as - described in "Background": - - Sentry VMA: VA:0x400000 -> /tmp/foo:0x0 - Sentry filemap: /tmp/foo:0x0 -> host:memory-file:0x3000 - Host VMA: VA:0x400000 -----------------> host:memory-file:0x3000 - Host filemap: host:memory-file:0x3000 -> PA:0x2fb000 - Host PTE: VA:0x400000 --------------------------------------------> PA:0x2fb000 - -Thus, from an implementation standpoint, host VMAs serve the same purpose in the -sentry that PTEs do in Linux. As in Linux, sentry VMA and filemap state is -independently mutable, and the desired state of host VMAs is derived from that -state. - -[^mmap-populate]: The sentry could force the host kernel to establish PTEs when - it creates the host VMA by passing the `MAP_POPULATE` flag to - the `mmap` system call, but usually does not. This is because, - to reduce the number of page faults that require handling by - the sentry and (correspondingly) the number of host `mmap` - system calls, the sentry usually creates host VMAs that are - much larger than the single faulting page. - -## Private Mappings - -The sentry implements private mappings consistently with Linux. Before -copy-on-write, the private mapping example given in the Background results in: - - Sentry VMA: VA:0x400000 -> /tmp/foo:0x0 (private) - Sentry filemap: /tmp/foo:0x0 -> host:memory-file:0x3000 - Host VMA: VA:0x400000 -----------------> host:memory-file:0x3000 (read-only) - Host filemap: host:memory-file:0x3000 -> PA:0x2fb000 - Host PTE: VA:0x400000 --------------------------------------------> PA:0x2fb000 (read-only) - -When the application attempts to write to this address, the host kernel delivers -information about the resulting page fault to the sentry. Analogous to Linux, -the sentry allocates memory to store the mapping's private copy of the file's -contents, copies file contents to the allocated memory, and changes the host VMA -to map to the private copy. Supposing that the sentry chooses the offset 0x4000 -in host file `memory-file` to store the private copy, the state of the system -after copy-on-write is: - - Sentry VMA: VA:0x400000 -> /tmp/foo:0x0 (private) - Sentry filemap: /tmp/foo:0x0 -> host:memory-file:0x3000 - Host VMA: VA:0x400000 -----------------> host:memory-file:0x4000 - Host filemap: host:memory-file:0x4000 -> PA:0x5ea000 - Host PTE: VA:0x400000 --------------------------------------------> PA:0x5ea000 - -However, this highlights an important difference between Linux and the sentry. -In Linux, page tables are concrete (architecture-dependent) data structures -owned by the kernel. Conversely, the sentry has the ability to create and -destroy host VMAs using host system calls, but it does not have direct access to -their state. Thus, as written, if the application invokes the `munmap` system -call to remove the sentry VMA, it is non-trivial for the sentry to determine -that it should deallocate `host:memory-file:0x4000`. This implies that the -sentry must retain information about the host VMAs that it has created. - -## Anonymous Mappings - -The sentry implements anonymous mappings consistently with Linux, except that -there is no shared zero page. - -# Implementation Constructs - -In Linux: - -- A virtual address space is represented by `struct mm_struct`. - -- VMAs are represented by `struct vm_area_struct`, stored in `struct - mm_struct::mmap`. - -- Mappings from file offsets to physical memory are stored in `struct - address_space`. - -- Reverse mappings from file offsets to virtual mappings are stored in `struct - address_space::i_mmap`. - -- Physical memory pages are represented by a pointer to `struct page` or an - index called a *page frame number* (PFN), represented by `pfn_t`. - -- PTEs are represented by architecture-dependent type `pte_t`, stored in a - table hierarchy rooted at `struct mm_struct::pgd`. - -In the sentry: - -- A virtual address space is represented by type [`mm.MemoryManager`][mm]. - -- Sentry VMAs are represented by type [`mm.vma`][mm], stored in - `mm.MemoryManager.vmas`. - -- Mappings from sentry file offsets to host file offsets are abstracted - through interface method [`memmap.Mappable.Translate`][memmap]. - -- Reverse mappings from sentry file offsets to virtual mappings are abstracted - through interface methods - [`memmap.Mappable.AddMapping` and `memmap.Mappable.RemoveMapping`][memmap]. - -- Host files that may be mapped into host VMAs are represented by type - [`platform.File`][platform]. - -- Host VMAs are represented in the sentry by type [`mm.pma`][mm] ("platform - mapping area"), stored in `mm.MemoryManager.pmas`. - -- Creation and destruction of host VMAs is abstracted through interface - methods - [`platform.AddressSpace.MapFile` and `platform.AddressSpace.Unmap`][platform]. - -[memmap]: https://github.com/google/gvisor/blob/master/pkg/sentry/memmap/memmap.go -[mm]: https://github.com/google/gvisor/blob/master/pkg/sentry/mm/mm.go -[pgalloc]: https://github.com/google/gvisor/blob/master/pkg/sentry/pgalloc/pgalloc.go -[platform]: https://github.com/google/gvisor/blob/master/pkg/sentry/platform/platform.go diff --git a/pkg/sentry/mm/file_refcount_set.go b/pkg/sentry/mm/file_refcount_set.go new file mode 100755 index 000000000..6b3081009 --- /dev/null +++ b/pkg/sentry/mm/file_refcount_set.go @@ -0,0 +1,1274 @@ +package mm + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/sentry/platform" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + fileRefcountminDegree = 3 + + fileRefcountmaxDegree = 2 * fileRefcountminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type fileRefcountSet struct { + root fileRefcountnode `state:".(*fileRefcountSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *fileRefcountSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *fileRefcountSet) IsEmptyRange(r __generics_imported0.FileRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *fileRefcountSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *fileRefcountSet) SpanRange(r __generics_imported0.FileRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *fileRefcountSet) FirstSegment() fileRefcountIterator { + if s.root.nrSegments == 0 { + return fileRefcountIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *fileRefcountSet) LastSegment() fileRefcountIterator { + if s.root.nrSegments == 0 { + return fileRefcountIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *fileRefcountSet) FirstGap() fileRefcountGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return fileRefcountGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *fileRefcountSet) LastGap() fileRefcountGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return fileRefcountGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *fileRefcountSet) Find(key uint64) (fileRefcountIterator, fileRefcountGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return fileRefcountIterator{n, i}, fileRefcountGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return fileRefcountIterator{}, fileRefcountGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *fileRefcountSet) FindSegment(key uint64) fileRefcountIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *fileRefcountSet) LowerBoundSegment(min uint64) fileRefcountIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *fileRefcountSet) UpperBoundSegment(max uint64) fileRefcountIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *fileRefcountSet) FindGap(key uint64) fileRefcountGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *fileRefcountSet) LowerBoundGap(min uint64) fileRefcountGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *fileRefcountSet) UpperBoundGap(max uint64) fileRefcountGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *fileRefcountSet) Add(r __generics_imported0.FileRange, val int32) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *fileRefcountSet) AddWithoutMerging(r __generics_imported0.FileRange, val int32) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *fileRefcountSet) Insert(gap fileRefcountGapIterator, r __generics_imported0.FileRange, val int32) fileRefcountIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (fileRefcountSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (fileRefcountSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (fileRefcountSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *fileRefcountSet) InsertWithoutMerging(gap fileRefcountGapIterator, r __generics_imported0.FileRange, val int32) fileRefcountIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *fileRefcountSet) InsertWithoutMergingUnchecked(gap fileRefcountGapIterator, r __generics_imported0.FileRange, val int32) fileRefcountIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return fileRefcountIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *fileRefcountSet) Remove(seg fileRefcountIterator) fileRefcountGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + fileRefcountSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(fileRefcountGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *fileRefcountSet) RemoveAll() { + s.root = fileRefcountnode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *fileRefcountSet) RemoveRange(r __generics_imported0.FileRange) fileRefcountGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *fileRefcountSet) Merge(first, second fileRefcountIterator) fileRefcountIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *fileRefcountSet) MergeUnchecked(first, second fileRefcountIterator) fileRefcountIterator { + if first.End() == second.Start() { + if mval, ok := (fileRefcountSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return fileRefcountIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *fileRefcountSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *fileRefcountSet) MergeRange(r __generics_imported0.FileRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *fileRefcountSet) MergeAdjacent(r __generics_imported0.FileRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *fileRefcountSet) Split(seg fileRefcountIterator, split uint64) (fileRefcountIterator, fileRefcountIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *fileRefcountSet) SplitUnchecked(seg fileRefcountIterator, split uint64) (fileRefcountIterator, fileRefcountIterator) { + val1, val2 := (fileRefcountSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.FileRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *fileRefcountSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *fileRefcountSet) Isolate(seg fileRefcountIterator, r __generics_imported0.FileRange) fileRefcountIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *fileRefcountSet) ApplyContiguous(r __generics_imported0.FileRange, fn func(seg fileRefcountIterator)) fileRefcountGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return fileRefcountGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return fileRefcountGapIterator{} + } + } +} + +// +stateify savable +type fileRefcountnode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *fileRefcountnode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [fileRefcountmaxDegree - 1]__generics_imported0.FileRange + values [fileRefcountmaxDegree - 1]int32 + children [fileRefcountmaxDegree]*fileRefcountnode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *fileRefcountnode) firstSegment() fileRefcountIterator { + for n.hasChildren { + n = n.children[0] + } + return fileRefcountIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *fileRefcountnode) lastSegment() fileRefcountIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return fileRefcountIterator{n, n.nrSegments - 1} +} + +func (n *fileRefcountnode) prevSibling() *fileRefcountnode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *fileRefcountnode) nextSibling() *fileRefcountnode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *fileRefcountnode) rebalanceBeforeInsert(gap fileRefcountGapIterator) fileRefcountGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < fileRefcountmaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &fileRefcountnode{ + nrSegments: fileRefcountminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &fileRefcountnode{ + nrSegments: fileRefcountminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:fileRefcountminDegree-1], n.keys[:fileRefcountminDegree-1]) + copy(left.values[:fileRefcountminDegree-1], n.values[:fileRefcountminDegree-1]) + copy(right.keys[:fileRefcountminDegree-1], n.keys[fileRefcountminDegree:]) + copy(right.values[:fileRefcountminDegree-1], n.values[fileRefcountminDegree:]) + n.keys[0], n.values[0] = n.keys[fileRefcountminDegree-1], n.values[fileRefcountminDegree-1] + fileRefcountzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:fileRefcountminDegree], n.children[:fileRefcountminDegree]) + copy(right.children[:fileRefcountminDegree], n.children[fileRefcountminDegree:]) + fileRefcountzeroNodeSlice(n.children[2:]) + for i := 0; i < fileRefcountminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < fileRefcountminDegree { + return fileRefcountGapIterator{left, gap.index} + } + return fileRefcountGapIterator{right, gap.index - fileRefcountminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[fileRefcountminDegree-1], n.values[fileRefcountminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &fileRefcountnode{ + nrSegments: fileRefcountminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:fileRefcountminDegree-1], n.keys[fileRefcountminDegree:]) + copy(sibling.values[:fileRefcountminDegree-1], n.values[fileRefcountminDegree:]) + fileRefcountzeroValueSlice(n.values[fileRefcountminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:fileRefcountminDegree], n.children[fileRefcountminDegree:]) + fileRefcountzeroNodeSlice(n.children[fileRefcountminDegree:]) + for i := 0; i < fileRefcountminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = fileRefcountminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < fileRefcountminDegree { + return gap + } + return fileRefcountGapIterator{sibling, gap.index - fileRefcountminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *fileRefcountnode) rebalanceAfterRemove(gap fileRefcountGapIterator) fileRefcountGapIterator { + for { + if n.nrSegments >= fileRefcountminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= fileRefcountminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + fileRefcountSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return fileRefcountGapIterator{n, 0} + } + if gap.node == n { + return fileRefcountGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= fileRefcountminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + fileRefcountSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return fileRefcountGapIterator{n, n.nrSegments} + } + return fileRefcountGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return fileRefcountGapIterator{p, gap.index} + } + if gap.node == right { + return fileRefcountGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *fileRefcountnode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = fileRefcountGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + fileRefcountSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type fileRefcountIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *fileRefcountnode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg fileRefcountIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg fileRefcountIterator) Range() __generics_imported0.FileRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg fileRefcountIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg fileRefcountIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg fileRefcountIterator) SetRangeUnchecked(r __generics_imported0.FileRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg fileRefcountIterator) SetRange(r __generics_imported0.FileRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg fileRefcountIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg fileRefcountIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg fileRefcountIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg fileRefcountIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg fileRefcountIterator) Value() int32 { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg fileRefcountIterator) ValuePtr() *int32 { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg fileRefcountIterator) SetValue(val int32) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg fileRefcountIterator) PrevSegment() fileRefcountIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return fileRefcountIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return fileRefcountIterator{} + } + return fileRefcountsegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg fileRefcountIterator) NextSegment() fileRefcountIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return fileRefcountIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return fileRefcountIterator{} + } + return fileRefcountsegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg fileRefcountIterator) PrevGap() fileRefcountGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return fileRefcountGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg fileRefcountIterator) NextGap() fileRefcountGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return fileRefcountGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg fileRefcountIterator) PrevNonEmpty() (fileRefcountIterator, fileRefcountGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return fileRefcountIterator{}, gap + } + return gap.PrevSegment(), fileRefcountGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg fileRefcountIterator) NextNonEmpty() (fileRefcountIterator, fileRefcountGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return fileRefcountIterator{}, gap + } + return gap.NextSegment(), fileRefcountGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type fileRefcountGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *fileRefcountnode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap fileRefcountGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap fileRefcountGapIterator) Range() __generics_imported0.FileRange { + return __generics_imported0.FileRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap fileRefcountGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return fileRefcountSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap fileRefcountGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return fileRefcountSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap fileRefcountGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap fileRefcountGapIterator) PrevSegment() fileRefcountIterator { + return fileRefcountsegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap fileRefcountGapIterator) NextSegment() fileRefcountIterator { + return fileRefcountsegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap fileRefcountGapIterator) PrevGap() fileRefcountGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return fileRefcountGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap fileRefcountGapIterator) NextGap() fileRefcountGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return fileRefcountGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func fileRefcountsegmentBeforePosition(n *fileRefcountnode, i int) fileRefcountIterator { + for i == 0 { + if n.parent == nil { + return fileRefcountIterator{} + } + n, i = n.parent, n.parentIndex + } + return fileRefcountIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func fileRefcountsegmentAfterPosition(n *fileRefcountnode, i int) fileRefcountIterator { + for i == n.nrSegments { + if n.parent == nil { + return fileRefcountIterator{} + } + n, i = n.parent, n.parentIndex + } + return fileRefcountIterator{n, i} +} + +func fileRefcountzeroValueSlice(slice []int32) { + + for i := range slice { + fileRefcountSetFunctions{}.ClearValue(&slice[i]) + } +} + +func fileRefcountzeroNodeSlice(slice []*fileRefcountnode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *fileRefcountSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *fileRefcountnode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *fileRefcountnode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type fileRefcountSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []int32 +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *fileRefcountSet) ExportSortedSlices() *fileRefcountSegmentDataSlices { + var sds fileRefcountSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *fileRefcountSet) ImportSortedSlices(sds *fileRefcountSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.FileRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *fileRefcountSet) saveRoot() *fileRefcountSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *fileRefcountSet) loadRoot(sds *fileRefcountSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/mm/io_list.go b/pkg/sentry/mm/io_list.go new file mode 100755 index 000000000..287e4305c --- /dev/null +++ b/pkg/sentry/mm/io_list.go @@ -0,0 +1,186 @@ +package mm + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type ioElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (ioElementMapper) linkerFor(elem *ioResult) *ioResult { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type ioList struct { + head *ioResult + tail *ioResult +} + +// Reset resets list l to the empty state. +func (l *ioList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *ioList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *ioList) Front() *ioResult { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *ioList) Back() *ioResult { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *ioList) PushFront(e *ioResult) { + linker := ioElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + ioElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *ioList) PushBack(e *ioResult) { + linker := ioElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + ioElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *ioList) PushBackList(m *ioList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + ioElementMapper{}.linkerFor(l.tail).SetNext(m.head) + ioElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *ioList) InsertAfter(b, e *ioResult) { + bLinker := ioElementMapper{}.linkerFor(b) + eLinker := ioElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + ioElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *ioList) InsertBefore(a, e *ioResult) { + aLinker := ioElementMapper{}.linkerFor(a) + eLinker := ioElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + ioElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *ioList) Remove(e *ioResult) { + linker := ioElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + ioElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + ioElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type ioEntry struct { + next *ioResult + prev *ioResult +} + +// Next returns the entry that follows e in the list. +func (e *ioEntry) Next() *ioResult { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *ioEntry) Prev() *ioResult { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *ioEntry) SetNext(elem *ioResult) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *ioEntry) SetPrev(elem *ioResult) { + e.prev = elem +} diff --git a/pkg/sentry/mm/mm_state_autogen.go b/pkg/sentry/mm/mm_state_autogen.go new file mode 100755 index 000000000..ef95c2836 --- /dev/null +++ b/pkg/sentry/mm/mm_state_autogen.go @@ -0,0 +1,396 @@ +// automatically generated by stateify. + +package mm + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *aioManager) beforeSave() {} +func (x *aioManager) save(m state.Map) { + x.beforeSave() + m.Save("contexts", &x.contexts) +} + +func (x *aioManager) afterLoad() {} +func (x *aioManager) load(m state.Map) { + m.Load("contexts", &x.contexts) +} + +func (x *ioResult) beforeSave() {} +func (x *ioResult) save(m state.Map) { + x.beforeSave() + m.Save("data", &x.data) + m.Save("ioEntry", &x.ioEntry) +} + +func (x *ioResult) afterLoad() {} +func (x *ioResult) load(m state.Map) { + m.Load("data", &x.data) + m.Load("ioEntry", &x.ioEntry) +} + +func (x *AIOContext) beforeSave() {} +func (x *AIOContext) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.dead) { + m.Failf("dead is %v, expected zero", x.dead) + } + m.Save("results", &x.results) + m.Save("maxOutstanding", &x.maxOutstanding) + m.Save("outstanding", &x.outstanding) +} + +func (x *AIOContext) load(m state.Map) { + m.Load("results", &x.results) + m.Load("maxOutstanding", &x.maxOutstanding) + m.Load("outstanding", &x.outstanding) + m.AfterLoad(x.afterLoad) +} + +func (x *aioMappable) beforeSave() {} +func (x *aioMappable) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("mfp", &x.mfp) + m.Save("fr", &x.fr) +} + +func (x *aioMappable) afterLoad() {} +func (x *aioMappable) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("mfp", &x.mfp) + m.Load("fr", &x.fr) +} + +func (x *fileRefcountSet) beforeSave() {} +func (x *fileRefcountSet) save(m state.Map) { + x.beforeSave() + var root *fileRefcountSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *fileRefcountSet) afterLoad() {} +func (x *fileRefcountSet) load(m state.Map) { + m.LoadValue("root", new(*fileRefcountSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*fileRefcountSegmentDataSlices)) }) +} + +func (x *fileRefcountnode) beforeSave() {} +func (x *fileRefcountnode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *fileRefcountnode) afterLoad() {} +func (x *fileRefcountnode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *fileRefcountSegmentDataSlices) beforeSave() {} +func (x *fileRefcountSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *fileRefcountSegmentDataSlices) afterLoad() {} +func (x *fileRefcountSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *ioList) beforeSave() {} +func (x *ioList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *ioList) afterLoad() {} +func (x *ioList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *ioEntry) beforeSave() {} +func (x *ioEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *ioEntry) afterLoad() {} +func (x *ioEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *MemoryManager) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.active) { + m.Failf("active is %v, expected zero", x.active) + } + if !state.IsZeroValue(x.captureInvalidations) { + m.Failf("captureInvalidations is %v, expected zero", x.captureInvalidations) + } + m.Save("p", &x.p) + m.Save("mfp", &x.mfp) + m.Save("layout", &x.layout) + m.Save("privateRefs", &x.privateRefs) + m.Save("users", &x.users) + m.Save("vmas", &x.vmas) + m.Save("brk", &x.brk) + m.Save("usageAS", &x.usageAS) + m.Save("lockedAS", &x.lockedAS) + m.Save("dataAS", &x.dataAS) + m.Save("defMLockMode", &x.defMLockMode) + m.Save("pmas", &x.pmas) + m.Save("curRSS", &x.curRSS) + m.Save("maxRSS", &x.maxRSS) + m.Save("argv", &x.argv) + m.Save("envv", &x.envv) + m.Save("auxv", &x.auxv) + m.Save("executable", &x.executable) + m.Save("dumpability", &x.dumpability) + m.Save("aioManager", &x.aioManager) + m.Save("sleepForActivation", &x.sleepForActivation) +} + +func (x *MemoryManager) load(m state.Map) { + m.Load("p", &x.p) + m.Load("mfp", &x.mfp) + m.Load("layout", &x.layout) + m.Load("privateRefs", &x.privateRefs) + m.Load("users", &x.users) + m.Load("vmas", &x.vmas) + m.Load("brk", &x.brk) + m.Load("usageAS", &x.usageAS) + m.Load("lockedAS", &x.lockedAS) + m.Load("dataAS", &x.dataAS) + m.Load("defMLockMode", &x.defMLockMode) + m.Load("pmas", &x.pmas) + m.Load("curRSS", &x.curRSS) + m.Load("maxRSS", &x.maxRSS) + m.Load("argv", &x.argv) + m.Load("envv", &x.envv) + m.Load("auxv", &x.auxv) + m.Load("executable", &x.executable) + m.Load("dumpability", &x.dumpability) + m.Load("aioManager", &x.aioManager) + m.Load("sleepForActivation", &x.sleepForActivation) + m.AfterLoad(x.afterLoad) +} + +func (x *vma) beforeSave() {} +func (x *vma) save(m state.Map) { + x.beforeSave() + var realPerms int = x.saveRealPerms() + m.SaveValue("realPerms", realPerms) + m.Save("mappable", &x.mappable) + m.Save("off", &x.off) + m.Save("dontfork", &x.dontfork) + m.Save("mlockMode", &x.mlockMode) + m.Save("numaPolicy", &x.numaPolicy) + m.Save("numaNodemask", &x.numaNodemask) + m.Save("id", &x.id) + m.Save("hint", &x.hint) +} + +func (x *vma) afterLoad() {} +func (x *vma) load(m state.Map) { + m.Load("mappable", &x.mappable) + m.Load("off", &x.off) + m.Load("dontfork", &x.dontfork) + m.Load("mlockMode", &x.mlockMode) + m.Load("numaPolicy", &x.numaPolicy) + m.Load("numaNodemask", &x.numaNodemask) + m.Load("id", &x.id) + m.Load("hint", &x.hint) + m.LoadValue("realPerms", new(int), func(y interface{}) { x.loadRealPerms(y.(int)) }) +} + +func (x *pma) beforeSave() {} +func (x *pma) save(m state.Map) { + x.beforeSave() + m.Save("off", &x.off) + m.Save("translatePerms", &x.translatePerms) + m.Save("effectivePerms", &x.effectivePerms) + m.Save("maxPerms", &x.maxPerms) + m.Save("needCOW", &x.needCOW) + m.Save("private", &x.private) +} + +func (x *pma) afterLoad() {} +func (x *pma) load(m state.Map) { + m.Load("off", &x.off) + m.Load("translatePerms", &x.translatePerms) + m.Load("effectivePerms", &x.effectivePerms) + m.Load("maxPerms", &x.maxPerms) + m.Load("needCOW", &x.needCOW) + m.Load("private", &x.private) +} + +func (x *privateRefs) beforeSave() {} +func (x *privateRefs) save(m state.Map) { + x.beforeSave() + m.Save("refs", &x.refs) +} + +func (x *privateRefs) afterLoad() {} +func (x *privateRefs) load(m state.Map) { + m.Load("refs", &x.refs) +} + +func (x *pmaSet) beforeSave() {} +func (x *pmaSet) save(m state.Map) { + x.beforeSave() + var root *pmaSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *pmaSet) afterLoad() {} +func (x *pmaSet) load(m state.Map) { + m.LoadValue("root", new(*pmaSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*pmaSegmentDataSlices)) }) +} + +func (x *pmanode) beforeSave() {} +func (x *pmanode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *pmanode) afterLoad() {} +func (x *pmanode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *pmaSegmentDataSlices) beforeSave() {} +func (x *pmaSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *pmaSegmentDataSlices) afterLoad() {} +func (x *pmaSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *SpecialMappable) beforeSave() {} +func (x *SpecialMappable) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("mfp", &x.mfp) + m.Save("fr", &x.fr) + m.Save("name", &x.name) +} + +func (x *SpecialMappable) afterLoad() {} +func (x *SpecialMappable) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("mfp", &x.mfp) + m.Load("fr", &x.fr) + m.Load("name", &x.name) +} + +func (x *vmaSet) beforeSave() {} +func (x *vmaSet) save(m state.Map) { + x.beforeSave() + var root *vmaSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *vmaSet) afterLoad() {} +func (x *vmaSet) load(m state.Map) { + m.LoadValue("root", new(*vmaSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*vmaSegmentDataSlices)) }) +} + +func (x *vmanode) beforeSave() {} +func (x *vmanode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *vmanode) afterLoad() {} +func (x *vmanode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *vmaSegmentDataSlices) beforeSave() {} +func (x *vmaSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *vmaSegmentDataSlices) afterLoad() {} +func (x *vmaSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func init() { + state.Register("pkg/sentry/mm.aioManager", (*aioManager)(nil), state.Fns{Save: (*aioManager).save, Load: (*aioManager).load}) + state.Register("pkg/sentry/mm.ioResult", (*ioResult)(nil), state.Fns{Save: (*ioResult).save, Load: (*ioResult).load}) + state.Register("pkg/sentry/mm.AIOContext", (*AIOContext)(nil), state.Fns{Save: (*AIOContext).save, Load: (*AIOContext).load}) + state.Register("pkg/sentry/mm.aioMappable", (*aioMappable)(nil), state.Fns{Save: (*aioMappable).save, Load: (*aioMappable).load}) + state.Register("pkg/sentry/mm.fileRefcountSet", (*fileRefcountSet)(nil), state.Fns{Save: (*fileRefcountSet).save, Load: (*fileRefcountSet).load}) + state.Register("pkg/sentry/mm.fileRefcountnode", (*fileRefcountnode)(nil), state.Fns{Save: (*fileRefcountnode).save, Load: (*fileRefcountnode).load}) + state.Register("pkg/sentry/mm.fileRefcountSegmentDataSlices", (*fileRefcountSegmentDataSlices)(nil), state.Fns{Save: (*fileRefcountSegmentDataSlices).save, Load: (*fileRefcountSegmentDataSlices).load}) + state.Register("pkg/sentry/mm.ioList", (*ioList)(nil), state.Fns{Save: (*ioList).save, Load: (*ioList).load}) + state.Register("pkg/sentry/mm.ioEntry", (*ioEntry)(nil), state.Fns{Save: (*ioEntry).save, Load: (*ioEntry).load}) + state.Register("pkg/sentry/mm.MemoryManager", (*MemoryManager)(nil), state.Fns{Save: (*MemoryManager).save, Load: (*MemoryManager).load}) + state.Register("pkg/sentry/mm.vma", (*vma)(nil), state.Fns{Save: (*vma).save, Load: (*vma).load}) + state.Register("pkg/sentry/mm.pma", (*pma)(nil), state.Fns{Save: (*pma).save, Load: (*pma).load}) + state.Register("pkg/sentry/mm.privateRefs", (*privateRefs)(nil), state.Fns{Save: (*privateRefs).save, Load: (*privateRefs).load}) + state.Register("pkg/sentry/mm.pmaSet", (*pmaSet)(nil), state.Fns{Save: (*pmaSet).save, Load: (*pmaSet).load}) + state.Register("pkg/sentry/mm.pmanode", (*pmanode)(nil), state.Fns{Save: (*pmanode).save, Load: (*pmanode).load}) + state.Register("pkg/sentry/mm.pmaSegmentDataSlices", (*pmaSegmentDataSlices)(nil), state.Fns{Save: (*pmaSegmentDataSlices).save, Load: (*pmaSegmentDataSlices).load}) + state.Register("pkg/sentry/mm.SpecialMappable", (*SpecialMappable)(nil), state.Fns{Save: (*SpecialMappable).save, Load: (*SpecialMappable).load}) + state.Register("pkg/sentry/mm.vmaSet", (*vmaSet)(nil), state.Fns{Save: (*vmaSet).save, Load: (*vmaSet).load}) + state.Register("pkg/sentry/mm.vmanode", (*vmanode)(nil), state.Fns{Save: (*vmanode).save, Load: (*vmanode).load}) + state.Register("pkg/sentry/mm.vmaSegmentDataSlices", (*vmaSegmentDataSlices)(nil), state.Fns{Save: (*vmaSegmentDataSlices).save, Load: (*vmaSegmentDataSlices).load}) +} diff --git a/pkg/sentry/mm/mm_test.go b/pkg/sentry/mm/mm_test.go deleted file mode 100644 index fdc308542..000000000 --- a/pkg/sentry/mm/mm_test.go +++ /dev/null @@ -1,230 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package mm - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -func testMemoryManager(ctx context.Context) *MemoryManager { - p := platform.FromContext(ctx) - mfp := pgalloc.MemoryFileProviderFromContext(ctx) - mm := NewMemoryManager(p, mfp, false) - mm.layout = arch.MmapLayout{ - MinAddr: p.MinUserAddress(), - MaxAddr: p.MaxUserAddress(), - BottomUpBase: p.MinUserAddress(), - TopDownBase: p.MaxUserAddress(), - } - return mm -} - -func (mm *MemoryManager) realUsageAS() uint64 { - return uint64(mm.vmas.Span()) -} - -func TestUsageASUpdates(t *testing.T) { - ctx := contexttest.Context(t) - mm := testMemoryManager(ctx) - defer mm.DecUsers(ctx) - - addr, err := mm.MMap(ctx, memmap.MMapOpts{ - Length: 2 * usermem.PageSize, - }) - if err != nil { - t.Fatalf("MMap got err %v want nil", err) - } - realUsage := mm.realUsageAS() - if mm.usageAS != realUsage { - t.Fatalf("usageAS believes %v bytes are mapped; %v bytes are actually mapped", mm.usageAS, realUsage) - } - - mm.MUnmap(ctx, addr, usermem.PageSize) - realUsage = mm.realUsageAS() - if mm.usageAS != realUsage { - t.Fatalf("usageAS believes %v bytes are mapped; %v bytes are actually mapped", mm.usageAS, realUsage) - } -} - -func (mm *MemoryManager) realDataAS() uint64 { - var sz uint64 - for seg := mm.vmas.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - vma := seg.Value() - if vma.isPrivateDataLocked() { - sz += uint64(seg.Range().Length()) - } - } - return sz -} - -func TestDataASUpdates(t *testing.T) { - ctx := contexttest.Context(t) - mm := testMemoryManager(ctx) - defer mm.DecUsers(ctx) - - addr, err := mm.MMap(ctx, memmap.MMapOpts{ - Length: 3 * usermem.PageSize, - Private: true, - Perms: usermem.Write, - MaxPerms: usermem.AnyAccess, - }) - if err != nil { - t.Fatalf("MMap got err %v want nil", err) - } - if mm.dataAS == 0 { - t.Fatalf("dataAS is 0, wanted not 0") - } - realDataAS := mm.realDataAS() - if mm.dataAS != realDataAS { - t.Fatalf("dataAS believes %v bytes are mapped; %v bytes are actually mapped", mm.dataAS, realDataAS) - } - - mm.MUnmap(ctx, addr, usermem.PageSize) - realDataAS = mm.realDataAS() - if mm.dataAS != realDataAS { - t.Fatalf("dataAS believes %v bytes are mapped; %v bytes are actually mapped", mm.dataAS, realDataAS) - } - - mm.MProtect(addr+usermem.PageSize, usermem.PageSize, usermem.Read, false) - realDataAS = mm.realDataAS() - if mm.dataAS != realDataAS { - t.Fatalf("dataAS believes %v bytes are mapped; %v bytes are actually mapped", mm.dataAS, realDataAS) - } - - mm.MRemap(ctx, addr+2*usermem.PageSize, usermem.PageSize, 2*usermem.PageSize, MRemapOpts{ - Move: MRemapMayMove, - }) - realDataAS = mm.realDataAS() - if mm.dataAS != realDataAS { - t.Fatalf("dataAS believes %v bytes are mapped; %v bytes are actually mapped", mm.dataAS, realDataAS) - } -} - -func TestBrkDataLimitUpdates(t *testing.T) { - limitSet := limits.NewLimitSet() - limitSet.Set(limits.Data, limits.Limit{}, true /* privileged */) // zero RLIMIT_DATA - - ctx := contexttest.WithLimitSet(contexttest.Context(t), limitSet) - mm := testMemoryManager(ctx) - defer mm.DecUsers(ctx) - - // Try to extend the brk by one page and expect doing so to fail. - oldBrk, _ := mm.Brk(ctx, 0) - if newBrk, _ := mm.Brk(ctx, oldBrk+usermem.PageSize); newBrk != oldBrk { - t.Errorf("brk() increased data segment above RLIMIT_DATA (old brk = %#x, new brk = %#x", oldBrk, newBrk) - } -} - -// TestIOAfterUnmap ensures that IO fails after unmap. -func TestIOAfterUnmap(t *testing.T) { - ctx := contexttest.Context(t) - mm := testMemoryManager(ctx) - defer mm.DecUsers(ctx) - - addr, err := mm.MMap(ctx, memmap.MMapOpts{ - Length: usermem.PageSize, - Private: true, - Perms: usermem.Read, - MaxPerms: usermem.AnyAccess, - }) - if err != nil { - t.Fatalf("MMap got err %v want nil", err) - } - - // IO works before munmap. - b := make([]byte, 1) - n, err := mm.CopyIn(ctx, addr, b, usermem.IOOpts{}) - if err != nil { - t.Errorf("CopyIn got err %v want nil", err) - } - if n != 1 { - t.Errorf("CopyIn got %d want 1", n) - } - - err = mm.MUnmap(ctx, addr, usermem.PageSize) - if err != nil { - t.Fatalf("MUnmap got err %v want nil", err) - } - - n, err = mm.CopyIn(ctx, addr, b, usermem.IOOpts{}) - if err != syserror.EFAULT { - t.Errorf("CopyIn got err %v want EFAULT", err) - } - if n != 0 { - t.Errorf("CopyIn got %d want 0", n) - } -} - -// TestIOAfterMProtect tests IO interaction with mprotect permissions. -func TestIOAfterMProtect(t *testing.T) { - ctx := contexttest.Context(t) - mm := testMemoryManager(ctx) - defer mm.DecUsers(ctx) - - addr, err := mm.MMap(ctx, memmap.MMapOpts{ - Length: usermem.PageSize, - Private: true, - Perms: usermem.ReadWrite, - MaxPerms: usermem.AnyAccess, - }) - if err != nil { - t.Fatalf("MMap got err %v want nil", err) - } - - // Writing works before mprotect. - b := make([]byte, 1) - n, err := mm.CopyOut(ctx, addr, b, usermem.IOOpts{}) - if err != nil { - t.Errorf("CopyOut got err %v want nil", err) - } - if n != 1 { - t.Errorf("CopyOut got %d want 1", n) - } - - err = mm.MProtect(addr, usermem.PageSize, usermem.Read, false) - if err != nil { - t.Errorf("MProtect got err %v want nil", err) - } - - // Without IgnorePermissions, CopyOut should no longer succeed. - n, err = mm.CopyOut(ctx, addr, b, usermem.IOOpts{}) - if err != syserror.EFAULT { - t.Errorf("CopyOut got err %v want EFAULT", err) - } - if n != 0 { - t.Errorf("CopyOut got %d want 0", n) - } - - // With IgnorePermissions, CopyOut should succeed despite mprotect. - n, err = mm.CopyOut(ctx, addr, b, usermem.IOOpts{ - IgnorePermissions: true, - }) - if err != nil { - t.Errorf("CopyOut got err %v want nil", err) - } - if n != 1 { - t.Errorf("CopyOut got %d want 1", n) - } -} diff --git a/pkg/sentry/mm/pma_set.go b/pkg/sentry/mm/pma_set.go new file mode 100755 index 000000000..8906e4edc --- /dev/null +++ b/pkg/sentry/mm/pma_set.go @@ -0,0 +1,1274 @@ +package mm + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/usermem" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + pmaminDegree = 8 + + pmamaxDegree = 2 * pmaminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type pmaSet struct { + root pmanode `state:".(*pmaSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *pmaSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *pmaSet) IsEmptyRange(r __generics_imported0.AddrRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *pmaSet) Span() __generics_imported0.Addr { + var sz __generics_imported0.Addr + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *pmaSet) SpanRange(r __generics_imported0.AddrRange) __generics_imported0.Addr { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz __generics_imported0.Addr + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *pmaSet) FirstSegment() pmaIterator { + if s.root.nrSegments == 0 { + return pmaIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *pmaSet) LastSegment() pmaIterator { + if s.root.nrSegments == 0 { + return pmaIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *pmaSet) FirstGap() pmaGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return pmaGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *pmaSet) LastGap() pmaGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return pmaGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *pmaSet) Find(key __generics_imported0.Addr) (pmaIterator, pmaGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return pmaIterator{n, i}, pmaGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return pmaIterator{}, pmaGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *pmaSet) FindSegment(key __generics_imported0.Addr) pmaIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *pmaSet) LowerBoundSegment(min __generics_imported0.Addr) pmaIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *pmaSet) UpperBoundSegment(max __generics_imported0.Addr) pmaIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *pmaSet) FindGap(key __generics_imported0.Addr) pmaGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *pmaSet) LowerBoundGap(min __generics_imported0.Addr) pmaGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *pmaSet) UpperBoundGap(max __generics_imported0.Addr) pmaGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *pmaSet) Add(r __generics_imported0.AddrRange, val pma) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *pmaSet) AddWithoutMerging(r __generics_imported0.AddrRange, val pma) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *pmaSet) Insert(gap pmaGapIterator, r __generics_imported0.AddrRange, val pma) pmaIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (pmaSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (pmaSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (pmaSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *pmaSet) InsertWithoutMerging(gap pmaGapIterator, r __generics_imported0.AddrRange, val pma) pmaIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *pmaSet) InsertWithoutMergingUnchecked(gap pmaGapIterator, r __generics_imported0.AddrRange, val pma) pmaIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return pmaIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *pmaSet) Remove(seg pmaIterator) pmaGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + pmaSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(pmaGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *pmaSet) RemoveAll() { + s.root = pmanode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *pmaSet) RemoveRange(r __generics_imported0.AddrRange) pmaGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *pmaSet) Merge(first, second pmaIterator) pmaIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *pmaSet) MergeUnchecked(first, second pmaIterator) pmaIterator { + if first.End() == second.Start() { + if mval, ok := (pmaSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return pmaIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *pmaSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *pmaSet) MergeRange(r __generics_imported0.AddrRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *pmaSet) MergeAdjacent(r __generics_imported0.AddrRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *pmaSet) Split(seg pmaIterator, split __generics_imported0.Addr) (pmaIterator, pmaIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *pmaSet) SplitUnchecked(seg pmaIterator, split __generics_imported0.Addr) (pmaIterator, pmaIterator) { + val1, val2 := (pmaSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.AddrRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *pmaSet) SplitAt(split __generics_imported0.Addr) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *pmaSet) Isolate(seg pmaIterator, r __generics_imported0.AddrRange) pmaIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *pmaSet) ApplyContiguous(r __generics_imported0.AddrRange, fn func(seg pmaIterator)) pmaGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return pmaGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return pmaGapIterator{} + } + } +} + +// +stateify savable +type pmanode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *pmanode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [pmamaxDegree - 1]__generics_imported0.AddrRange + values [pmamaxDegree - 1]pma + children [pmamaxDegree]*pmanode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *pmanode) firstSegment() pmaIterator { + for n.hasChildren { + n = n.children[0] + } + return pmaIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *pmanode) lastSegment() pmaIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return pmaIterator{n, n.nrSegments - 1} +} + +func (n *pmanode) prevSibling() *pmanode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *pmanode) nextSibling() *pmanode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *pmanode) rebalanceBeforeInsert(gap pmaGapIterator) pmaGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < pmamaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &pmanode{ + nrSegments: pmaminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &pmanode{ + nrSegments: pmaminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:pmaminDegree-1], n.keys[:pmaminDegree-1]) + copy(left.values[:pmaminDegree-1], n.values[:pmaminDegree-1]) + copy(right.keys[:pmaminDegree-1], n.keys[pmaminDegree:]) + copy(right.values[:pmaminDegree-1], n.values[pmaminDegree:]) + n.keys[0], n.values[0] = n.keys[pmaminDegree-1], n.values[pmaminDegree-1] + pmazeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:pmaminDegree], n.children[:pmaminDegree]) + copy(right.children[:pmaminDegree], n.children[pmaminDegree:]) + pmazeroNodeSlice(n.children[2:]) + for i := 0; i < pmaminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < pmaminDegree { + return pmaGapIterator{left, gap.index} + } + return pmaGapIterator{right, gap.index - pmaminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[pmaminDegree-1], n.values[pmaminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &pmanode{ + nrSegments: pmaminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:pmaminDegree-1], n.keys[pmaminDegree:]) + copy(sibling.values[:pmaminDegree-1], n.values[pmaminDegree:]) + pmazeroValueSlice(n.values[pmaminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:pmaminDegree], n.children[pmaminDegree:]) + pmazeroNodeSlice(n.children[pmaminDegree:]) + for i := 0; i < pmaminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = pmaminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < pmaminDegree { + return gap + } + return pmaGapIterator{sibling, gap.index - pmaminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *pmanode) rebalanceAfterRemove(gap pmaGapIterator) pmaGapIterator { + for { + if n.nrSegments >= pmaminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= pmaminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + pmaSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return pmaGapIterator{n, 0} + } + if gap.node == n { + return pmaGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= pmaminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + pmaSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return pmaGapIterator{n, n.nrSegments} + } + return pmaGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return pmaGapIterator{p, gap.index} + } + if gap.node == right { + return pmaGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *pmanode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = pmaGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + pmaSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type pmaIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *pmanode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg pmaIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg pmaIterator) Range() __generics_imported0.AddrRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg pmaIterator) Start() __generics_imported0.Addr { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg pmaIterator) End() __generics_imported0.Addr { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg pmaIterator) SetRangeUnchecked(r __generics_imported0.AddrRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg pmaIterator) SetRange(r __generics_imported0.AddrRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg pmaIterator) SetStartUnchecked(start __generics_imported0.Addr) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg pmaIterator) SetStart(start __generics_imported0.Addr) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg pmaIterator) SetEndUnchecked(end __generics_imported0.Addr) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg pmaIterator) SetEnd(end __generics_imported0.Addr) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg pmaIterator) Value() pma { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg pmaIterator) ValuePtr() *pma { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg pmaIterator) SetValue(val pma) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg pmaIterator) PrevSegment() pmaIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return pmaIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return pmaIterator{} + } + return pmasegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg pmaIterator) NextSegment() pmaIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return pmaIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return pmaIterator{} + } + return pmasegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg pmaIterator) PrevGap() pmaGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return pmaGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg pmaIterator) NextGap() pmaGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return pmaGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg pmaIterator) PrevNonEmpty() (pmaIterator, pmaGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return pmaIterator{}, gap + } + return gap.PrevSegment(), pmaGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg pmaIterator) NextNonEmpty() (pmaIterator, pmaGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return pmaIterator{}, gap + } + return gap.NextSegment(), pmaGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type pmaGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *pmanode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap pmaGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap pmaGapIterator) Range() __generics_imported0.AddrRange { + return __generics_imported0.AddrRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap pmaGapIterator) Start() __generics_imported0.Addr { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return pmaSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap pmaGapIterator) End() __generics_imported0.Addr { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return pmaSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap pmaGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap pmaGapIterator) PrevSegment() pmaIterator { + return pmasegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap pmaGapIterator) NextSegment() pmaIterator { + return pmasegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap pmaGapIterator) PrevGap() pmaGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return pmaGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap pmaGapIterator) NextGap() pmaGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return pmaGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func pmasegmentBeforePosition(n *pmanode, i int) pmaIterator { + for i == 0 { + if n.parent == nil { + return pmaIterator{} + } + n, i = n.parent, n.parentIndex + } + return pmaIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func pmasegmentAfterPosition(n *pmanode, i int) pmaIterator { + for i == n.nrSegments { + if n.parent == nil { + return pmaIterator{} + } + n, i = n.parent, n.parentIndex + } + return pmaIterator{n, i} +} + +func pmazeroValueSlice(slice []pma) { + + for i := range slice { + pmaSetFunctions{}.ClearValue(&slice[i]) + } +} + +func pmazeroNodeSlice(slice []*pmanode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *pmaSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *pmanode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *pmanode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type pmaSegmentDataSlices struct { + Start []__generics_imported0.Addr + End []__generics_imported0.Addr + Values []pma +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *pmaSet) ExportSortedSlices() *pmaSegmentDataSlices { + var sds pmaSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *pmaSet) ImportSortedSlices(sds *pmaSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.AddrRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *pmaSet) saveRoot() *pmaSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *pmaSet) loadRoot(sds *pmaSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/mm/vma_set.go b/pkg/sentry/mm/vma_set.go new file mode 100755 index 000000000..af6b1d317 --- /dev/null +++ b/pkg/sentry/mm/vma_set.go @@ -0,0 +1,1274 @@ +package mm + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/usermem" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + vmaminDegree = 8 + + vmamaxDegree = 2 * vmaminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type vmaSet struct { + root vmanode `state:".(*vmaSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *vmaSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *vmaSet) IsEmptyRange(r __generics_imported0.AddrRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *vmaSet) Span() __generics_imported0.Addr { + var sz __generics_imported0.Addr + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *vmaSet) SpanRange(r __generics_imported0.AddrRange) __generics_imported0.Addr { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz __generics_imported0.Addr + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *vmaSet) FirstSegment() vmaIterator { + if s.root.nrSegments == 0 { + return vmaIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *vmaSet) LastSegment() vmaIterator { + if s.root.nrSegments == 0 { + return vmaIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *vmaSet) FirstGap() vmaGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return vmaGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *vmaSet) LastGap() vmaGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return vmaGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *vmaSet) Find(key __generics_imported0.Addr) (vmaIterator, vmaGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return vmaIterator{n, i}, vmaGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return vmaIterator{}, vmaGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *vmaSet) FindSegment(key __generics_imported0.Addr) vmaIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *vmaSet) LowerBoundSegment(min __generics_imported0.Addr) vmaIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *vmaSet) UpperBoundSegment(max __generics_imported0.Addr) vmaIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *vmaSet) FindGap(key __generics_imported0.Addr) vmaGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *vmaSet) LowerBoundGap(min __generics_imported0.Addr) vmaGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *vmaSet) UpperBoundGap(max __generics_imported0.Addr) vmaGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *vmaSet) Add(r __generics_imported0.AddrRange, val vma) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *vmaSet) AddWithoutMerging(r __generics_imported0.AddrRange, val vma) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *vmaSet) Insert(gap vmaGapIterator, r __generics_imported0.AddrRange, val vma) vmaIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (vmaSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (vmaSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (vmaSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *vmaSet) InsertWithoutMerging(gap vmaGapIterator, r __generics_imported0.AddrRange, val vma) vmaIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *vmaSet) InsertWithoutMergingUnchecked(gap vmaGapIterator, r __generics_imported0.AddrRange, val vma) vmaIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return vmaIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *vmaSet) Remove(seg vmaIterator) vmaGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + vmaSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(vmaGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *vmaSet) RemoveAll() { + s.root = vmanode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *vmaSet) RemoveRange(r __generics_imported0.AddrRange) vmaGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *vmaSet) Merge(first, second vmaIterator) vmaIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *vmaSet) MergeUnchecked(first, second vmaIterator) vmaIterator { + if first.End() == second.Start() { + if mval, ok := (vmaSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return vmaIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *vmaSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *vmaSet) MergeRange(r __generics_imported0.AddrRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *vmaSet) MergeAdjacent(r __generics_imported0.AddrRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *vmaSet) Split(seg vmaIterator, split __generics_imported0.Addr) (vmaIterator, vmaIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *vmaSet) SplitUnchecked(seg vmaIterator, split __generics_imported0.Addr) (vmaIterator, vmaIterator) { + val1, val2 := (vmaSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.AddrRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *vmaSet) SplitAt(split __generics_imported0.Addr) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *vmaSet) Isolate(seg vmaIterator, r __generics_imported0.AddrRange) vmaIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *vmaSet) ApplyContiguous(r __generics_imported0.AddrRange, fn func(seg vmaIterator)) vmaGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return vmaGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return vmaGapIterator{} + } + } +} + +// +stateify savable +type vmanode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *vmanode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [vmamaxDegree - 1]__generics_imported0.AddrRange + values [vmamaxDegree - 1]vma + children [vmamaxDegree]*vmanode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *vmanode) firstSegment() vmaIterator { + for n.hasChildren { + n = n.children[0] + } + return vmaIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *vmanode) lastSegment() vmaIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return vmaIterator{n, n.nrSegments - 1} +} + +func (n *vmanode) prevSibling() *vmanode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *vmanode) nextSibling() *vmanode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *vmanode) rebalanceBeforeInsert(gap vmaGapIterator) vmaGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < vmamaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &vmanode{ + nrSegments: vmaminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &vmanode{ + nrSegments: vmaminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:vmaminDegree-1], n.keys[:vmaminDegree-1]) + copy(left.values[:vmaminDegree-1], n.values[:vmaminDegree-1]) + copy(right.keys[:vmaminDegree-1], n.keys[vmaminDegree:]) + copy(right.values[:vmaminDegree-1], n.values[vmaminDegree:]) + n.keys[0], n.values[0] = n.keys[vmaminDegree-1], n.values[vmaminDegree-1] + vmazeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:vmaminDegree], n.children[:vmaminDegree]) + copy(right.children[:vmaminDegree], n.children[vmaminDegree:]) + vmazeroNodeSlice(n.children[2:]) + for i := 0; i < vmaminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < vmaminDegree { + return vmaGapIterator{left, gap.index} + } + return vmaGapIterator{right, gap.index - vmaminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[vmaminDegree-1], n.values[vmaminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &vmanode{ + nrSegments: vmaminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:vmaminDegree-1], n.keys[vmaminDegree:]) + copy(sibling.values[:vmaminDegree-1], n.values[vmaminDegree:]) + vmazeroValueSlice(n.values[vmaminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:vmaminDegree], n.children[vmaminDegree:]) + vmazeroNodeSlice(n.children[vmaminDegree:]) + for i := 0; i < vmaminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = vmaminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < vmaminDegree { + return gap + } + return vmaGapIterator{sibling, gap.index - vmaminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *vmanode) rebalanceAfterRemove(gap vmaGapIterator) vmaGapIterator { + for { + if n.nrSegments >= vmaminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= vmaminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + vmaSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return vmaGapIterator{n, 0} + } + if gap.node == n { + return vmaGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= vmaminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + vmaSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return vmaGapIterator{n, n.nrSegments} + } + return vmaGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return vmaGapIterator{p, gap.index} + } + if gap.node == right { + return vmaGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *vmanode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = vmaGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + vmaSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type vmaIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *vmanode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg vmaIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg vmaIterator) Range() __generics_imported0.AddrRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg vmaIterator) Start() __generics_imported0.Addr { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg vmaIterator) End() __generics_imported0.Addr { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg vmaIterator) SetRangeUnchecked(r __generics_imported0.AddrRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg vmaIterator) SetRange(r __generics_imported0.AddrRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg vmaIterator) SetStartUnchecked(start __generics_imported0.Addr) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg vmaIterator) SetStart(start __generics_imported0.Addr) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg vmaIterator) SetEndUnchecked(end __generics_imported0.Addr) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg vmaIterator) SetEnd(end __generics_imported0.Addr) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg vmaIterator) Value() vma { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg vmaIterator) ValuePtr() *vma { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg vmaIterator) SetValue(val vma) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg vmaIterator) PrevSegment() vmaIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return vmaIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return vmaIterator{} + } + return vmasegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg vmaIterator) NextSegment() vmaIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return vmaIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return vmaIterator{} + } + return vmasegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg vmaIterator) PrevGap() vmaGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return vmaGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg vmaIterator) NextGap() vmaGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return vmaGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg vmaIterator) PrevNonEmpty() (vmaIterator, vmaGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return vmaIterator{}, gap + } + return gap.PrevSegment(), vmaGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg vmaIterator) NextNonEmpty() (vmaIterator, vmaGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return vmaIterator{}, gap + } + return gap.NextSegment(), vmaGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type vmaGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *vmanode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap vmaGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap vmaGapIterator) Range() __generics_imported0.AddrRange { + return __generics_imported0.AddrRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap vmaGapIterator) Start() __generics_imported0.Addr { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return vmaSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap vmaGapIterator) End() __generics_imported0.Addr { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return vmaSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap vmaGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap vmaGapIterator) PrevSegment() vmaIterator { + return vmasegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap vmaGapIterator) NextSegment() vmaIterator { + return vmasegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap vmaGapIterator) PrevGap() vmaGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return vmaGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap vmaGapIterator) NextGap() vmaGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return vmaGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func vmasegmentBeforePosition(n *vmanode, i int) vmaIterator { + for i == 0 { + if n.parent == nil { + return vmaIterator{} + } + n, i = n.parent, n.parentIndex + } + return vmaIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func vmasegmentAfterPosition(n *vmanode, i int) vmaIterator { + for i == n.nrSegments { + if n.parent == nil { + return vmaIterator{} + } + n, i = n.parent, n.parentIndex + } + return vmaIterator{n, i} +} + +func vmazeroValueSlice(slice []vma) { + + for i := range slice { + vmaSetFunctions{}.ClearValue(&slice[i]) + } +} + +func vmazeroNodeSlice(slice []*vmanode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *vmaSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *vmanode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *vmanode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type vmaSegmentDataSlices struct { + Start []__generics_imported0.Addr + End []__generics_imported0.Addr + Values []vma +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *vmaSet) ExportSortedSlices() *vmaSegmentDataSlices { + var sds vmaSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *vmaSet) ImportSortedSlices(sds *vmaSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.AddrRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *vmaSet) saveRoot() *vmaSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *vmaSet) loadRoot(sds *vmaSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/pgalloc/BUILD b/pkg/sentry/pgalloc/BUILD deleted file mode 100644 index 1eeb9f317..000000000 --- a/pkg/sentry/pgalloc/BUILD +++ /dev/null @@ -1,85 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "evictable_range", - out = "evictable_range.go", - package = "pgalloc", - prefix = "Evictable", - template = "//pkg/segment:generic_range", - types = { - "T": "uint64", - }, -) - -go_template_instance( - name = "evictable_range_set", - out = "evictable_range_set.go", - package = "pgalloc", - prefix = "evictableRange", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "EvictableRange", - "Value": "evictableRangeSetValue", - "Functions": "evictableRangeSetFunctions", - }, -) - -go_template_instance( - name = "usage_set", - out = "usage_set.go", - consts = { - "minDegree": "10", - }, - imports = { - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", - }, - package = "pgalloc", - prefix = "usage", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "platform.FileRange", - "Value": "usageInfo", - "Functions": "usageSetFunctions", - }, -) - -go_library( - name = "pgalloc", - srcs = [ - "context.go", - "evictable_range.go", - "evictable_range_set.go", - "pgalloc.go", - "pgalloc_unsafe.go", - "save_restore.go", - "usage_set.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/log", - "//pkg/memutil", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/hostmm", - "//pkg/sentry/platform", - "//pkg/sentry/usage", - "//pkg/state", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -go_test( - name = "pgalloc_test", - size = "small", - srcs = ["pgalloc_test.go"], - library = ":pgalloc", - deps = ["//pkg/usermem"], -) diff --git a/pkg/sentry/pgalloc/evictable_range.go b/pkg/sentry/pgalloc/evictable_range.go new file mode 100755 index 000000000..10ce2ff44 --- /dev/null +++ b/pkg/sentry/pgalloc/evictable_range.go @@ -0,0 +1,62 @@ +package pgalloc + +// A Range represents a contiguous range of T. +// +// +stateify savable +type EvictableRange struct { + // Start is the inclusive start of the range. + Start uint64 + + // End is the exclusive end of the range. + End uint64 +} + +// WellFormed returns true if r.Start <= r.End. All other methods on a Range +// require that the Range is well-formed. +func (r EvictableRange) WellFormed() bool { + return r.Start <= r.End +} + +// Length returns the length of the range. +func (r EvictableRange) Length() uint64 { + return r.End - r.Start +} + +// Contains returns true if r contains x. +func (r EvictableRange) Contains(x uint64) bool { + return r.Start <= x && x < r.End +} + +// Overlaps returns true if r and r2 overlap. +func (r EvictableRange) Overlaps(r2 EvictableRange) bool { + return r.Start < r2.End && r2.Start < r.End +} + +// IsSupersetOf returns true if r is a superset of r2; that is, the range r2 is +// contained within r. +func (r EvictableRange) IsSupersetOf(r2 EvictableRange) bool { + return r.Start <= r2.Start && r.End >= r2.End +} + +// Intersect returns a range consisting of the intersection between r and r2. +// If r and r2 do not overlap, Intersect returns a range with unspecified +// bounds, but for which Length() == 0. +func (r EvictableRange) Intersect(r2 EvictableRange) EvictableRange { + if r.Start < r2.Start { + r.Start = r2.Start + } + if r.End > r2.End { + r.End = r2.End + } + if r.End < r.Start { + r.End = r.Start + } + return r +} + +// CanSplitAt returns true if it is legal to split a segment spanning the range +// r at x; that is, splitting at x would produce two ranges, both of which have +// non-zero length. +func (r EvictableRange) CanSplitAt(x uint64) bool { + return r.Contains(x) && r.Start < x +} diff --git a/pkg/sentry/pgalloc/evictable_range_set.go b/pkg/sentry/pgalloc/evictable_range_set.go new file mode 100755 index 000000000..6fbd02434 --- /dev/null +++ b/pkg/sentry/pgalloc/evictable_range_set.go @@ -0,0 +1,1270 @@ +package pgalloc + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + evictableRangeminDegree = 3 + + evictableRangemaxDegree = 2 * evictableRangeminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type evictableRangeSet struct { + root evictableRangenode `state:".(*evictableRangeSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *evictableRangeSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *evictableRangeSet) IsEmptyRange(r EvictableRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *evictableRangeSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *evictableRangeSet) SpanRange(r EvictableRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *evictableRangeSet) FirstSegment() evictableRangeIterator { + if s.root.nrSegments == 0 { + return evictableRangeIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *evictableRangeSet) LastSegment() evictableRangeIterator { + if s.root.nrSegments == 0 { + return evictableRangeIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *evictableRangeSet) FirstGap() evictableRangeGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return evictableRangeGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *evictableRangeSet) LastGap() evictableRangeGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return evictableRangeGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *evictableRangeSet) Find(key uint64) (evictableRangeIterator, evictableRangeGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return evictableRangeIterator{n, i}, evictableRangeGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return evictableRangeIterator{}, evictableRangeGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *evictableRangeSet) FindSegment(key uint64) evictableRangeIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *evictableRangeSet) LowerBoundSegment(min uint64) evictableRangeIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *evictableRangeSet) UpperBoundSegment(max uint64) evictableRangeIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *evictableRangeSet) FindGap(key uint64) evictableRangeGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *evictableRangeSet) LowerBoundGap(min uint64) evictableRangeGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *evictableRangeSet) UpperBoundGap(max uint64) evictableRangeGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *evictableRangeSet) Add(r EvictableRange, val evictableRangeSetValue) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *evictableRangeSet) AddWithoutMerging(r EvictableRange, val evictableRangeSetValue) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *evictableRangeSet) Insert(gap evictableRangeGapIterator, r EvictableRange, val evictableRangeSetValue) evictableRangeIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (evictableRangeSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (evictableRangeSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (evictableRangeSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *evictableRangeSet) InsertWithoutMerging(gap evictableRangeGapIterator, r EvictableRange, val evictableRangeSetValue) evictableRangeIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *evictableRangeSet) InsertWithoutMergingUnchecked(gap evictableRangeGapIterator, r EvictableRange, val evictableRangeSetValue) evictableRangeIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return evictableRangeIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *evictableRangeSet) Remove(seg evictableRangeIterator) evictableRangeGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + evictableRangeSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(evictableRangeGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *evictableRangeSet) RemoveAll() { + s.root = evictableRangenode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *evictableRangeSet) RemoveRange(r EvictableRange) evictableRangeGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *evictableRangeSet) Merge(first, second evictableRangeIterator) evictableRangeIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *evictableRangeSet) MergeUnchecked(first, second evictableRangeIterator) evictableRangeIterator { + if first.End() == second.Start() { + if mval, ok := (evictableRangeSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return evictableRangeIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *evictableRangeSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *evictableRangeSet) MergeRange(r EvictableRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *evictableRangeSet) MergeAdjacent(r EvictableRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *evictableRangeSet) Split(seg evictableRangeIterator, split uint64) (evictableRangeIterator, evictableRangeIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *evictableRangeSet) SplitUnchecked(seg evictableRangeIterator, split uint64) (evictableRangeIterator, evictableRangeIterator) { + val1, val2 := (evictableRangeSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), EvictableRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *evictableRangeSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *evictableRangeSet) Isolate(seg evictableRangeIterator, r EvictableRange) evictableRangeIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *evictableRangeSet) ApplyContiguous(r EvictableRange, fn func(seg evictableRangeIterator)) evictableRangeGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return evictableRangeGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return evictableRangeGapIterator{} + } + } +} + +// +stateify savable +type evictableRangenode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *evictableRangenode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [evictableRangemaxDegree - 1]EvictableRange + values [evictableRangemaxDegree - 1]evictableRangeSetValue + children [evictableRangemaxDegree]*evictableRangenode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *evictableRangenode) firstSegment() evictableRangeIterator { + for n.hasChildren { + n = n.children[0] + } + return evictableRangeIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *evictableRangenode) lastSegment() evictableRangeIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return evictableRangeIterator{n, n.nrSegments - 1} +} + +func (n *evictableRangenode) prevSibling() *evictableRangenode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *evictableRangenode) nextSibling() *evictableRangenode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *evictableRangenode) rebalanceBeforeInsert(gap evictableRangeGapIterator) evictableRangeGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < evictableRangemaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &evictableRangenode{ + nrSegments: evictableRangeminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &evictableRangenode{ + nrSegments: evictableRangeminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:evictableRangeminDegree-1], n.keys[:evictableRangeminDegree-1]) + copy(left.values[:evictableRangeminDegree-1], n.values[:evictableRangeminDegree-1]) + copy(right.keys[:evictableRangeminDegree-1], n.keys[evictableRangeminDegree:]) + copy(right.values[:evictableRangeminDegree-1], n.values[evictableRangeminDegree:]) + n.keys[0], n.values[0] = n.keys[evictableRangeminDegree-1], n.values[evictableRangeminDegree-1] + evictableRangezeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:evictableRangeminDegree], n.children[:evictableRangeminDegree]) + copy(right.children[:evictableRangeminDegree], n.children[evictableRangeminDegree:]) + evictableRangezeroNodeSlice(n.children[2:]) + for i := 0; i < evictableRangeminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < evictableRangeminDegree { + return evictableRangeGapIterator{left, gap.index} + } + return evictableRangeGapIterator{right, gap.index - evictableRangeminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[evictableRangeminDegree-1], n.values[evictableRangeminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &evictableRangenode{ + nrSegments: evictableRangeminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:evictableRangeminDegree-1], n.keys[evictableRangeminDegree:]) + copy(sibling.values[:evictableRangeminDegree-1], n.values[evictableRangeminDegree:]) + evictableRangezeroValueSlice(n.values[evictableRangeminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:evictableRangeminDegree], n.children[evictableRangeminDegree:]) + evictableRangezeroNodeSlice(n.children[evictableRangeminDegree:]) + for i := 0; i < evictableRangeminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = evictableRangeminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < evictableRangeminDegree { + return gap + } + return evictableRangeGapIterator{sibling, gap.index - evictableRangeminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *evictableRangenode) rebalanceAfterRemove(gap evictableRangeGapIterator) evictableRangeGapIterator { + for { + if n.nrSegments >= evictableRangeminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= evictableRangeminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + evictableRangeSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return evictableRangeGapIterator{n, 0} + } + if gap.node == n { + return evictableRangeGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= evictableRangeminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + evictableRangeSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return evictableRangeGapIterator{n, n.nrSegments} + } + return evictableRangeGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return evictableRangeGapIterator{p, gap.index} + } + if gap.node == right { + return evictableRangeGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *evictableRangenode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = evictableRangeGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + evictableRangeSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type evictableRangeIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *evictableRangenode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg evictableRangeIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg evictableRangeIterator) Range() EvictableRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg evictableRangeIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg evictableRangeIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg evictableRangeIterator) SetRangeUnchecked(r EvictableRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg evictableRangeIterator) SetRange(r EvictableRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg evictableRangeIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg evictableRangeIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg evictableRangeIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg evictableRangeIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg evictableRangeIterator) Value() evictableRangeSetValue { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg evictableRangeIterator) ValuePtr() *evictableRangeSetValue { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg evictableRangeIterator) SetValue(val evictableRangeSetValue) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg evictableRangeIterator) PrevSegment() evictableRangeIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return evictableRangeIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return evictableRangeIterator{} + } + return evictableRangesegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg evictableRangeIterator) NextSegment() evictableRangeIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return evictableRangeIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return evictableRangeIterator{} + } + return evictableRangesegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg evictableRangeIterator) PrevGap() evictableRangeGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return evictableRangeGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg evictableRangeIterator) NextGap() evictableRangeGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return evictableRangeGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg evictableRangeIterator) PrevNonEmpty() (evictableRangeIterator, evictableRangeGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return evictableRangeIterator{}, gap + } + return gap.PrevSegment(), evictableRangeGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg evictableRangeIterator) NextNonEmpty() (evictableRangeIterator, evictableRangeGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return evictableRangeIterator{}, gap + } + return gap.NextSegment(), evictableRangeGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type evictableRangeGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *evictableRangenode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap evictableRangeGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap evictableRangeGapIterator) Range() EvictableRange { + return EvictableRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap evictableRangeGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return evictableRangeSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap evictableRangeGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return evictableRangeSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap evictableRangeGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap evictableRangeGapIterator) PrevSegment() evictableRangeIterator { + return evictableRangesegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap evictableRangeGapIterator) NextSegment() evictableRangeIterator { + return evictableRangesegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap evictableRangeGapIterator) PrevGap() evictableRangeGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return evictableRangeGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap evictableRangeGapIterator) NextGap() evictableRangeGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return evictableRangeGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func evictableRangesegmentBeforePosition(n *evictableRangenode, i int) evictableRangeIterator { + for i == 0 { + if n.parent == nil { + return evictableRangeIterator{} + } + n, i = n.parent, n.parentIndex + } + return evictableRangeIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func evictableRangesegmentAfterPosition(n *evictableRangenode, i int) evictableRangeIterator { + for i == n.nrSegments { + if n.parent == nil { + return evictableRangeIterator{} + } + n, i = n.parent, n.parentIndex + } + return evictableRangeIterator{n, i} +} + +func evictableRangezeroValueSlice(slice []evictableRangeSetValue) { + + for i := range slice { + evictableRangeSetFunctions{}.ClearValue(&slice[i]) + } +} + +func evictableRangezeroNodeSlice(slice []*evictableRangenode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *evictableRangeSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *evictableRangenode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *evictableRangenode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type evictableRangeSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []evictableRangeSetValue +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *evictableRangeSet) ExportSortedSlices() *evictableRangeSegmentDataSlices { + var sds evictableRangeSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *evictableRangeSet) ImportSortedSlices(sds *evictableRangeSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := EvictableRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *evictableRangeSet) saveRoot() *evictableRangeSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *evictableRangeSet) loadRoot(sds *evictableRangeSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/pgalloc/pgalloc_state_autogen.go b/pkg/sentry/pgalloc/pgalloc_state_autogen.go new file mode 100755 index 000000000..97e1c883b --- /dev/null +++ b/pkg/sentry/pgalloc/pgalloc_state_autogen.go @@ -0,0 +1,146 @@ +// automatically generated by stateify. + +package pgalloc + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *EvictableRange) beforeSave() {} +func (x *EvictableRange) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) +} + +func (x *EvictableRange) afterLoad() {} +func (x *EvictableRange) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) +} + +func (x *evictableRangeSet) beforeSave() {} +func (x *evictableRangeSet) save(m state.Map) { + x.beforeSave() + var root *evictableRangeSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *evictableRangeSet) afterLoad() {} +func (x *evictableRangeSet) load(m state.Map) { + m.LoadValue("root", new(*evictableRangeSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*evictableRangeSegmentDataSlices)) }) +} + +func (x *evictableRangenode) beforeSave() {} +func (x *evictableRangenode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *evictableRangenode) afterLoad() {} +func (x *evictableRangenode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *evictableRangeSegmentDataSlices) beforeSave() {} +func (x *evictableRangeSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *evictableRangeSegmentDataSlices) afterLoad() {} +func (x *evictableRangeSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *usageInfo) beforeSave() {} +func (x *usageInfo) save(m state.Map) { + x.beforeSave() + m.Save("kind", &x.kind) + m.Save("knownCommitted", &x.knownCommitted) + m.Save("refs", &x.refs) +} + +func (x *usageInfo) afterLoad() {} +func (x *usageInfo) load(m state.Map) { + m.Load("kind", &x.kind) + m.Load("knownCommitted", &x.knownCommitted) + m.Load("refs", &x.refs) +} + +func (x *usageSet) beforeSave() {} +func (x *usageSet) save(m state.Map) { + x.beforeSave() + var root *usageSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *usageSet) afterLoad() {} +func (x *usageSet) load(m state.Map) { + m.LoadValue("root", new(*usageSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*usageSegmentDataSlices)) }) +} + +func (x *usagenode) beforeSave() {} +func (x *usagenode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *usagenode) afterLoad() {} +func (x *usagenode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *usageSegmentDataSlices) beforeSave() {} +func (x *usageSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *usageSegmentDataSlices) afterLoad() {} +func (x *usageSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func init() { + state.Register("pkg/sentry/pgalloc.EvictableRange", (*EvictableRange)(nil), state.Fns{Save: (*EvictableRange).save, Load: (*EvictableRange).load}) + state.Register("pkg/sentry/pgalloc.evictableRangeSet", (*evictableRangeSet)(nil), state.Fns{Save: (*evictableRangeSet).save, Load: (*evictableRangeSet).load}) + state.Register("pkg/sentry/pgalloc.evictableRangenode", (*evictableRangenode)(nil), state.Fns{Save: (*evictableRangenode).save, Load: (*evictableRangenode).load}) + state.Register("pkg/sentry/pgalloc.evictableRangeSegmentDataSlices", (*evictableRangeSegmentDataSlices)(nil), state.Fns{Save: (*evictableRangeSegmentDataSlices).save, Load: (*evictableRangeSegmentDataSlices).load}) + state.Register("pkg/sentry/pgalloc.usageInfo", (*usageInfo)(nil), state.Fns{Save: (*usageInfo).save, Load: (*usageInfo).load}) + state.Register("pkg/sentry/pgalloc.usageSet", (*usageSet)(nil), state.Fns{Save: (*usageSet).save, Load: (*usageSet).load}) + state.Register("pkg/sentry/pgalloc.usagenode", (*usagenode)(nil), state.Fns{Save: (*usagenode).save, Load: (*usagenode).load}) + state.Register("pkg/sentry/pgalloc.usageSegmentDataSlices", (*usageSegmentDataSlices)(nil), state.Fns{Save: (*usageSegmentDataSlices).save, Load: (*usageSegmentDataSlices).load}) +} diff --git a/pkg/sentry/pgalloc/pgalloc_test.go b/pkg/sentry/pgalloc/pgalloc_test.go deleted file mode 100644 index 293f22c6b..000000000 --- a/pkg/sentry/pgalloc/pgalloc_test.go +++ /dev/null @@ -1,168 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package pgalloc - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/usermem" -) - -const ( - page = usermem.PageSize - hugepage = usermem.HugePageSize -) - -func TestFindUnallocatedRange(t *testing.T) { - for _, test := range []struct { - desc string - usage *usageSegmentDataSlices - start uint64 - length uint64 - alignment uint64 - unallocated uint64 - minUnallocated uint64 - }{ - { - desc: "Initial allocation succeeds", - usage: &usageSegmentDataSlices{}, - start: 0, - length: page, - alignment: page, - unallocated: 0, - minUnallocated: 0, - }, - { - desc: "Allocation begins at start of file", - usage: &usageSegmentDataSlices{ - Start: []uint64{page}, - End: []uint64{2 * page}, - Values: []usageInfo{{refs: 1}}, - }, - start: 0, - length: page, - alignment: page, - unallocated: 0, - minUnallocated: 0, - }, - { - desc: "In-use frames are not allocatable", - usage: &usageSegmentDataSlices{ - Start: []uint64{0, page}, - End: []uint64{page, 2 * page}, - Values: []usageInfo{{refs: 1}, {refs: 2}}, - }, - start: 0, - length: page, - alignment: page, - unallocated: 2 * page, - minUnallocated: 2 * page, - }, - { - desc: "Reclaimable frames are not allocatable", - usage: &usageSegmentDataSlices{ - Start: []uint64{0, page, 2 * page}, - End: []uint64{page, 2 * page, 3 * page}, - Values: []usageInfo{{refs: 1}, {refs: 0}, {refs: 1}}, - }, - start: 0, - length: page, - alignment: page, - unallocated: 3 * page, - minUnallocated: 3 * page, - }, - { - desc: "Gaps between in-use frames are allocatable", - usage: &usageSegmentDataSlices{ - Start: []uint64{0, 2 * page}, - End: []uint64{page, 3 * page}, - Values: []usageInfo{{refs: 1}, {refs: 1}}, - }, - start: 0, - length: page, - alignment: page, - unallocated: page, - minUnallocated: page, - }, - { - desc: "Inadequately-sized gaps are rejected", - usage: &usageSegmentDataSlices{ - Start: []uint64{0, 2 * page}, - End: []uint64{page, 3 * page}, - Values: []usageInfo{{refs: 1}, {refs: 1}}, - }, - start: 0, - length: 2 * page, - alignment: page, - unallocated: 3 * page, - minUnallocated: page, - }, - { - desc: "Hugepage alignment is honored", - usage: &usageSegmentDataSlices{ - Start: []uint64{0, hugepage + page}, - // Hugepage-sized gap here that shouldn't be allocated from - // since it's incorrectly aligned. - End: []uint64{page, hugepage + 2*page}, - Values: []usageInfo{{refs: 1}, {refs: 1}}, - }, - start: 0, - length: hugepage, - alignment: hugepage, - unallocated: 2 * hugepage, - minUnallocated: page, - }, - { - desc: "Pages before start ignored", - usage: &usageSegmentDataSlices{ - Start: []uint64{page, 3 * page}, - End: []uint64{2 * page, 4 * page}, - Values: []usageInfo{{refs: 1}, {refs: 2}}, - }, - start: page, - length: page, - alignment: page, - unallocated: 2 * page, - minUnallocated: 2 * page, - }, - { - desc: "start may be in the middle of segment", - usage: &usageSegmentDataSlices{ - Start: []uint64{0, 3 * page}, - End: []uint64{2 * page, 4 * page}, - Values: []usageInfo{{refs: 1}, {refs: 2}}, - }, - start: page, - length: page, - alignment: page, - unallocated: 2 * page, - minUnallocated: 2 * page, - }, - } { - t.Run(test.desc, func(t *testing.T) { - var usage usageSet - if err := usage.ImportSortedSlices(test.usage); err != nil { - t.Fatalf("Failed to initialize usage from %v: %v", test.usage, err) - } - unallocated, minUnallocated := findUnallocatedRange(&usage, test.start, test.length, test.alignment) - if unallocated != test.unallocated { - t.Errorf("findUnallocatedRange(%v, %x, %x, %x): got unallocated %x, wanted %x", test.usage, test.start, test.length, test.alignment, unallocated, test.unallocated) - } - if minUnallocated != test.minUnallocated { - t.Errorf("findUnallocatedRange(%v, %x, %x, %x): got minUnallocated %x, wanted %x", test.usage, test.start, test.length, test.alignment, minUnallocated, test.minUnallocated) - } - }) - } -} diff --git a/pkg/sentry/pgalloc/usage_set.go b/pkg/sentry/pgalloc/usage_set.go new file mode 100755 index 000000000..37b9235ca --- /dev/null +++ b/pkg/sentry/pgalloc/usage_set.go @@ -0,0 +1,1274 @@ +package pgalloc + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/sentry/platform" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + usageminDegree = 10 + + usagemaxDegree = 2 * usageminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type usageSet struct { + root usagenode `state:".(*usageSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *usageSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *usageSet) IsEmptyRange(r __generics_imported0.FileRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *usageSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *usageSet) SpanRange(r __generics_imported0.FileRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *usageSet) FirstSegment() usageIterator { + if s.root.nrSegments == 0 { + return usageIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *usageSet) LastSegment() usageIterator { + if s.root.nrSegments == 0 { + return usageIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *usageSet) FirstGap() usageGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return usageGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *usageSet) LastGap() usageGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return usageGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *usageSet) Find(key uint64) (usageIterator, usageGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return usageIterator{n, i}, usageGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return usageIterator{}, usageGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *usageSet) FindSegment(key uint64) usageIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *usageSet) LowerBoundSegment(min uint64) usageIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *usageSet) UpperBoundSegment(max uint64) usageIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *usageSet) FindGap(key uint64) usageGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *usageSet) LowerBoundGap(min uint64) usageGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *usageSet) UpperBoundGap(max uint64) usageGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *usageSet) Add(r __generics_imported0.FileRange, val usageInfo) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *usageSet) AddWithoutMerging(r __generics_imported0.FileRange, val usageInfo) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *usageSet) Insert(gap usageGapIterator, r __generics_imported0.FileRange, val usageInfo) usageIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (usageSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (usageSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (usageSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *usageSet) InsertWithoutMerging(gap usageGapIterator, r __generics_imported0.FileRange, val usageInfo) usageIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *usageSet) InsertWithoutMergingUnchecked(gap usageGapIterator, r __generics_imported0.FileRange, val usageInfo) usageIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return usageIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *usageSet) Remove(seg usageIterator) usageGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + usageSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(usageGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *usageSet) RemoveAll() { + s.root = usagenode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *usageSet) RemoveRange(r __generics_imported0.FileRange) usageGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *usageSet) Merge(first, second usageIterator) usageIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *usageSet) MergeUnchecked(first, second usageIterator) usageIterator { + if first.End() == second.Start() { + if mval, ok := (usageSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return usageIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *usageSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *usageSet) MergeRange(r __generics_imported0.FileRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *usageSet) MergeAdjacent(r __generics_imported0.FileRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *usageSet) Split(seg usageIterator, split uint64) (usageIterator, usageIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *usageSet) SplitUnchecked(seg usageIterator, split uint64) (usageIterator, usageIterator) { + val1, val2 := (usageSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.FileRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *usageSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *usageSet) Isolate(seg usageIterator, r __generics_imported0.FileRange) usageIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *usageSet) ApplyContiguous(r __generics_imported0.FileRange, fn func(seg usageIterator)) usageGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return usageGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return usageGapIterator{} + } + } +} + +// +stateify savable +type usagenode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *usagenode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [usagemaxDegree - 1]__generics_imported0.FileRange + values [usagemaxDegree - 1]usageInfo + children [usagemaxDegree]*usagenode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *usagenode) firstSegment() usageIterator { + for n.hasChildren { + n = n.children[0] + } + return usageIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *usagenode) lastSegment() usageIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return usageIterator{n, n.nrSegments - 1} +} + +func (n *usagenode) prevSibling() *usagenode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *usagenode) nextSibling() *usagenode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *usagenode) rebalanceBeforeInsert(gap usageGapIterator) usageGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < usagemaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &usagenode{ + nrSegments: usageminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &usagenode{ + nrSegments: usageminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:usageminDegree-1], n.keys[:usageminDegree-1]) + copy(left.values[:usageminDegree-1], n.values[:usageminDegree-1]) + copy(right.keys[:usageminDegree-1], n.keys[usageminDegree:]) + copy(right.values[:usageminDegree-1], n.values[usageminDegree:]) + n.keys[0], n.values[0] = n.keys[usageminDegree-1], n.values[usageminDegree-1] + usagezeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:usageminDegree], n.children[:usageminDegree]) + copy(right.children[:usageminDegree], n.children[usageminDegree:]) + usagezeroNodeSlice(n.children[2:]) + for i := 0; i < usageminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < usageminDegree { + return usageGapIterator{left, gap.index} + } + return usageGapIterator{right, gap.index - usageminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[usageminDegree-1], n.values[usageminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &usagenode{ + nrSegments: usageminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:usageminDegree-1], n.keys[usageminDegree:]) + copy(sibling.values[:usageminDegree-1], n.values[usageminDegree:]) + usagezeroValueSlice(n.values[usageminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:usageminDegree], n.children[usageminDegree:]) + usagezeroNodeSlice(n.children[usageminDegree:]) + for i := 0; i < usageminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = usageminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < usageminDegree { + return gap + } + return usageGapIterator{sibling, gap.index - usageminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *usagenode) rebalanceAfterRemove(gap usageGapIterator) usageGapIterator { + for { + if n.nrSegments >= usageminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= usageminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + usageSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return usageGapIterator{n, 0} + } + if gap.node == n { + return usageGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= usageminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + usageSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return usageGapIterator{n, n.nrSegments} + } + return usageGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return usageGapIterator{p, gap.index} + } + if gap.node == right { + return usageGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *usagenode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = usageGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + usageSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type usageIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *usagenode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg usageIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg usageIterator) Range() __generics_imported0.FileRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg usageIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg usageIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg usageIterator) SetRangeUnchecked(r __generics_imported0.FileRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg usageIterator) SetRange(r __generics_imported0.FileRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg usageIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg usageIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg usageIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg usageIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg usageIterator) Value() usageInfo { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg usageIterator) ValuePtr() *usageInfo { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg usageIterator) SetValue(val usageInfo) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg usageIterator) PrevSegment() usageIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return usageIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return usageIterator{} + } + return usagesegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg usageIterator) NextSegment() usageIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return usageIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return usageIterator{} + } + return usagesegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg usageIterator) PrevGap() usageGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return usageGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg usageIterator) NextGap() usageGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return usageGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg usageIterator) PrevNonEmpty() (usageIterator, usageGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return usageIterator{}, gap + } + return gap.PrevSegment(), usageGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg usageIterator) NextNonEmpty() (usageIterator, usageGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return usageIterator{}, gap + } + return gap.NextSegment(), usageGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type usageGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *usagenode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap usageGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap usageGapIterator) Range() __generics_imported0.FileRange { + return __generics_imported0.FileRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap usageGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return usageSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap usageGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return usageSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap usageGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap usageGapIterator) PrevSegment() usageIterator { + return usagesegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap usageGapIterator) NextSegment() usageIterator { + return usagesegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap usageGapIterator) PrevGap() usageGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return usageGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap usageGapIterator) NextGap() usageGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return usageGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func usagesegmentBeforePosition(n *usagenode, i int) usageIterator { + for i == 0 { + if n.parent == nil { + return usageIterator{} + } + n, i = n.parent, n.parentIndex + } + return usageIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func usagesegmentAfterPosition(n *usagenode, i int) usageIterator { + for i == n.nrSegments { + if n.parent == nil { + return usageIterator{} + } + n, i = n.parent, n.parentIndex + } + return usageIterator{n, i} +} + +func usagezeroValueSlice(slice []usageInfo) { + + for i := range slice { + usageSetFunctions{}.ClearValue(&slice[i]) + } +} + +func usagezeroNodeSlice(slice []*usagenode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *usageSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *usagenode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *usagenode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type usageSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []usageInfo +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *usageSet) ExportSortedSlices() *usageSegmentDataSlices { + var sds usageSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *usageSet) ImportSortedSlices(sds *usageSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.FileRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *usageSet) saveRoot() *usageSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *usageSet) loadRoot(sds *usageSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/platform/BUILD b/pkg/sentry/platform/BUILD deleted file mode 100644 index 453241eca..000000000 --- a/pkg/sentry/platform/BUILD +++ /dev/null @@ -1,39 +0,0 @@ -load("//tools:defs.bzl", "go_library") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "file_range", - out = "file_range.go", - package = "platform", - prefix = "File", - template = "//pkg/segment:generic_range", - types = { - "T": "uint64", - }, -) - -go_library( - name = "platform", - srcs = [ - "context.go", - "file_range.go", - "mmap_min_addr.go", - "platform.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/atomicbitops", - "//pkg/context", - "//pkg/log", - "//pkg/safecopy", - "//pkg/safemem", - "//pkg/seccomp", - "//pkg/sentry/arch", - "//pkg/sentry/usage", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/platform/file_range.go b/pkg/sentry/platform/file_range.go new file mode 100755 index 000000000..685d360e3 --- /dev/null +++ b/pkg/sentry/platform/file_range.go @@ -0,0 +1,62 @@ +package platform + +// A Range represents a contiguous range of T. +// +// +stateify savable +type FileRange struct { + // Start is the inclusive start of the range. + Start uint64 + + // End is the exclusive end of the range. + End uint64 +} + +// WellFormed returns true if r.Start <= r.End. All other methods on a Range +// require that the Range is well-formed. +func (r FileRange) WellFormed() bool { + return r.Start <= r.End +} + +// Length returns the length of the range. +func (r FileRange) Length() uint64 { + return r.End - r.Start +} + +// Contains returns true if r contains x. +func (r FileRange) Contains(x uint64) bool { + return r.Start <= x && x < r.End +} + +// Overlaps returns true if r and r2 overlap. +func (r FileRange) Overlaps(r2 FileRange) bool { + return r.Start < r2.End && r2.Start < r.End +} + +// IsSupersetOf returns true if r is a superset of r2; that is, the range r2 is +// contained within r. +func (r FileRange) IsSupersetOf(r2 FileRange) bool { + return r.Start <= r2.Start && r.End >= r2.End +} + +// Intersect returns a range consisting of the intersection between r and r2. +// If r and r2 do not overlap, Intersect returns a range with unspecified +// bounds, but for which Length() == 0. +func (r FileRange) Intersect(r2 FileRange) FileRange { + if r.Start < r2.Start { + r.Start = r2.Start + } + if r.End > r2.End { + r.End = r2.End + } + if r.End < r.Start { + r.End = r.Start + } + return r +} + +// CanSplitAt returns true if it is legal to split a segment spanning the range +// r at x; that is, splitting at x would produce two ranges, both of which have +// non-zero length. +func (r FileRange) CanSplitAt(x uint64) bool { + return r.Contains(x) && r.Start < x +} diff --git a/pkg/sentry/platform/interrupt/BUILD b/pkg/sentry/platform/interrupt/BUILD deleted file mode 100644 index 83b385f14..000000000 --- a/pkg/sentry/platform/interrupt/BUILD +++ /dev/null @@ -1,19 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "interrupt", - srcs = [ - "interrupt.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = ["//pkg/sync"], -) - -go_test( - name = "interrupt_test", - size = "small", - srcs = ["interrupt_test.go"], - library = ":interrupt", -) diff --git a/pkg/sentry/platform/interrupt/interrupt_state_autogen.go b/pkg/sentry/platform/interrupt/interrupt_state_autogen.go new file mode 100755 index 000000000..1336e5f01 --- /dev/null +++ b/pkg/sentry/platform/interrupt/interrupt_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package interrupt diff --git a/pkg/sentry/platform/interrupt/interrupt_test.go b/pkg/sentry/platform/interrupt/interrupt_test.go deleted file mode 100644 index 0ecdf6e7a..000000000 --- a/pkg/sentry/platform/interrupt/interrupt_test.go +++ /dev/null @@ -1,99 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package interrupt - -import ( - "testing" -) - -type countingReceiver struct { - interrupts int -} - -// NotifyInterrupt implements Receiver.NotifyInterrupt. -func (r *countingReceiver) NotifyInterrupt() { - r.interrupts++ -} - -func TestSingleInterruptBeforeEnable(t *testing.T) { - var ( - f Forwarder - r countingReceiver - ) - f.NotifyInterrupt() - // The interrupt should cause the first Enable to fail. - if f.Enable(&r) { - f.Disable() - t.Fatalf("Enable: got true, wanted false") - } - // The failing Enable "acknowledges" the interrupt, allowing future Enables - // to succeed. - if !f.Enable(&r) { - t.Fatalf("Enable: got false, wanted true") - } - f.Disable() -} - -func TestMultipleInterruptsBeforeEnable(t *testing.T) { - var ( - f Forwarder - r countingReceiver - ) - f.NotifyInterrupt() - f.NotifyInterrupt() - // The interrupts should cause the first Enable to fail. - if f.Enable(&r) { - f.Disable() - t.Fatalf("Enable: got true, wanted false") - } - // Interrupts are deduplicated while the Forwarder is disabled, so the - // failing Enable "acknowledges" all interrupts, allowing future Enables to - // succeed. - if !f.Enable(&r) { - t.Fatalf("Enable: got false, wanted true") - } - f.Disable() -} - -func TestSingleInterruptAfterEnable(t *testing.T) { - var ( - f Forwarder - r countingReceiver - ) - if !f.Enable(&r) { - t.Fatalf("Enable: got false, wanted true") - } - defer f.Disable() - f.NotifyInterrupt() - if r.interrupts != 1 { - t.Errorf("interrupts: got %d, wanted 1", r.interrupts) - } -} - -func TestMultipleInterruptsAfterEnable(t *testing.T) { - var ( - f Forwarder - r countingReceiver - ) - if !f.Enable(&r) { - t.Fatalf("Enable: got false, wanted true") - } - defer f.Disable() - f.NotifyInterrupt() - f.NotifyInterrupt() - if r.interrupts != 2 { - t.Errorf("interrupts: got %d, wanted 2", r.interrupts) - } -} diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD deleted file mode 100644 index 159f7eafd..000000000 --- a/pkg/sentry/platform/kvm/BUILD +++ /dev/null @@ -1,80 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "kvm", - srcs = [ - "address_space.go", - "allocator.go", - "bluepill.go", - "bluepill_amd64.go", - "bluepill_amd64.s", - "bluepill_amd64_unsafe.go", - "bluepill_arm64.go", - "bluepill_arm64.s", - "bluepill_arm64_unsafe.go", - "bluepill_fault.go", - "bluepill_unsafe.go", - "context.go", - "filters_amd64.go", - "filters_arm64.go", - "kvm.go", - "kvm_amd64.go", - "kvm_amd64_unsafe.go", - "kvm_arm64.go", - "kvm_arm64_unsafe.go", - "kvm_const.go", - "kvm_const_arm64.go", - "machine.go", - "machine_amd64.go", - "machine_amd64_unsafe.go", - "machine_arm64.go", - "machine_arm64_unsafe.go", - "machine_unsafe.go", - "physical_map.go", - "physical_map_amd64.go", - "physical_map_arm64.go", - "virtual_map.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/atomicbitops", - "//pkg/cpuid", - "//pkg/log", - "//pkg/procid", - "//pkg/safecopy", - "//pkg/seccomp", - "//pkg/sentry/arch", - "//pkg/sentry/platform", - "//pkg/sentry/platform/interrupt", - "//pkg/sentry/platform/ring0", - "//pkg/sentry/platform/ring0/pagetables", - "//pkg/sentry/time", - "//pkg/sync", - "//pkg/usermem", - ], -) - -go_test( - name = "kvm_test", - srcs = [ - "kvm_test.go", - "virtual_map_test.go", - ], - library = ":kvm", - tags = [ - "manual", - "nogotsan", - "requires-kvm", - ], - deps = [ - "//pkg/sentry/arch", - "//pkg/sentry/platform", - "//pkg/sentry/platform/kvm/testutil", - "//pkg/sentry/platform/ring0", - "//pkg/sentry/platform/ring0/pagetables", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.go b/pkg/sentry/platform/kvm/bluepill_arm64.go index 552341721..552341721 100644..100755 --- a/pkg/sentry/platform/kvm/bluepill_arm64.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64.go diff --git a/pkg/sentry/platform/kvm/bluepill_arm64.s b/pkg/sentry/platform/kvm/bluepill_arm64.s index 04efa0147..04efa0147 100644..100755 --- a/pkg/sentry/platform/kvm/bluepill_arm64.s +++ b/pkg/sentry/platform/kvm/bluepill_arm64.s diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index eb5ed574e..eb5ed574e 100644..100755 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go diff --git a/pkg/sentry/platform/kvm/filters_amd64.go b/pkg/sentry/platform/kvm/filters_amd64.go index 7d949f1dd..7d949f1dd 100644..100755 --- a/pkg/sentry/platform/kvm/filters_amd64.go +++ b/pkg/sentry/platform/kvm/filters_amd64.go diff --git a/pkg/sentry/platform/kvm/filters_arm64.go b/pkg/sentry/platform/kvm/filters_arm64.go index 9245d07c2..9245d07c2 100644..100755 --- a/pkg/sentry/platform/kvm/filters_arm64.go +++ b/pkg/sentry/platform/kvm/filters_arm64.go diff --git a/pkg/sentry/platform/kvm/kvm_amd64_state_autogen.go b/pkg/sentry/platform/kvm/kvm_amd64_state_autogen.go new file mode 100755 index 000000000..a69cbee8b --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_amd64_state_autogen.go @@ -0,0 +1,7 @@ +// automatically generated by stateify. + +// +build amd64 +// +build amd64 +// +build amd64 + +package kvm diff --git a/pkg/sentry/platform/kvm/kvm_amd64_unsafe_state_autogen.go b/pkg/sentry/platform/kvm/kvm_amd64_unsafe_state_autogen.go new file mode 100755 index 000000000..a69cbee8b --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_amd64_unsafe_state_autogen.go @@ -0,0 +1,7 @@ +// automatically generated by stateify. + +// +build amd64 +// +build amd64 +// +build amd64 + +package kvm diff --git a/pkg/sentry/platform/kvm/kvm_arm64.go b/pkg/sentry/platform/kvm/kvm_arm64.go index 79045651e..79045651e 100644..100755 --- a/pkg/sentry/platform/kvm/kvm_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_arm64.go diff --git a/pkg/sentry/platform/kvm/kvm_arm64_state_autogen.go b/pkg/sentry/platform/kvm/kvm_arm64_state_autogen.go new file mode 100755 index 000000000..90183b764 --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_arm64_state_autogen.go @@ -0,0 +1,7 @@ +// automatically generated by stateify. + +// +build arm64 +// +build arm64 +// +build arm64 + +package kvm diff --git a/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go b/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go index 6531bae1d..6531bae1d 100644..100755 --- a/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/kvm_arm64_unsafe.go diff --git a/pkg/sentry/platform/kvm/kvm_arm64_unsafe_state_autogen.go b/pkg/sentry/platform/kvm/kvm_arm64_unsafe_state_autogen.go new file mode 100755 index 000000000..90183b764 --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_arm64_unsafe_state_autogen.go @@ -0,0 +1,7 @@ +// automatically generated by stateify. + +// +build arm64 +// +build arm64 +// +build arm64 + +package kvm diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go index 5a74c6e36..5a74c6e36 100644..100755 --- a/pkg/sentry/platform/kvm/kvm_const_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go diff --git a/pkg/sentry/platform/kvm/kvm_state_autogen.go b/pkg/sentry/platform/kvm/kvm_state_autogen.go new file mode 100755 index 000000000..2325262dc --- /dev/null +++ b/pkg/sentry/platform/kvm/kvm_state_autogen.go @@ -0,0 +1,8 @@ +// automatically generated by stateify. + +// +build go1.12 +// +build !go1.15 +// +build go1.12 +// +build !go1.15 + +package kvm diff --git a/pkg/sentry/platform/kvm/kvm_test.go b/pkg/sentry/platform/kvm/kvm_test.go deleted file mode 100644 index c42752d50..000000000 --- a/pkg/sentry/platform/kvm/kvm_test.go +++ /dev/null @@ -1,533 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kvm - -import ( - "math/rand" - "reflect" - "sync/atomic" - "syscall" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/platform/kvm/testutil" - "gvisor.dev/gvisor/pkg/sentry/platform/ring0" - "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/usermem" -) - -var dummyFPState = (*byte)(arch.NewFloatingPointData()) - -type testHarness interface { - Errorf(format string, args ...interface{}) - Fatalf(format string, args ...interface{}) -} - -func kvmTest(t testHarness, setup func(*KVM), fn func(*vCPU) bool) { - // Create the machine. - deviceFile, err := OpenDevice() - if err != nil { - t.Fatalf("error opening device file: %v", err) - } - k, err := New(deviceFile) - if err != nil { - t.Fatalf("error creating KVM instance: %v", err) - } - defer k.machine.Destroy() - - // Call additional setup. - if setup != nil { - setup(k) - } - - var c *vCPU // For recovery. - defer func() { - redpill() - if c != nil { - k.machine.Put(c) - } - }() - for { - c = k.machine.Get() - if !fn(c) { - break - } - - // We put the vCPU here and clear the value so that the - // deferred recovery will not re-put it above. - k.machine.Put(c) - c = nil - } -} - -func bluepillTest(t testHarness, fn func(*vCPU)) { - kvmTest(t, nil, func(c *vCPU) bool { - bluepill(c) - fn(c) - return false - }) -} - -func TestKernelSyscall(t *testing.T) { - bluepillTest(t, func(c *vCPU) { - redpill() // Leave guest mode. - if got := atomic.LoadUint32(&c.state); got != vCPUUser { - t.Errorf("vCPU not in ready state: got %v", got) - } - }) -} - -func hostFault() { - defer func() { - recover() - }() - var foo *int - *foo = 0 -} - -func TestKernelFault(t *testing.T) { - hostFault() // Ensure recovery works. - bluepillTest(t, func(c *vCPU) { - hostFault() - if got := atomic.LoadUint32(&c.state); got != vCPUUser { - t.Errorf("vCPU not in ready state: got %v", got) - } - }) -} - -func TestKernelFloatingPoint(t *testing.T) { - bluepillTest(t, func(c *vCPU) { - if !testutil.FloatingPointWorks() { - t.Errorf("floating point does not work, and it should!") - } - }) -} - -func applicationTest(t testHarness, useHostMappings bool, target func(), fn func(*vCPU, *syscall.PtraceRegs, *pagetables.PageTables) bool) { - // Initialize registers & page tables. - var ( - regs syscall.PtraceRegs - pt *pagetables.PageTables - ) - testutil.SetTestTarget(®s, target) - - kvmTest(t, func(k *KVM) { - // Create new page tables. - as, _, err := k.NewAddressSpace(nil /* invalidator */) - if err != nil { - t.Fatalf("can't create new address space: %v", err) - } - pt = as.(*addressSpace).pageTables - - if useHostMappings { - // Apply the physical mappings to these page tables. - // (This is normally dangerous, since they point to - // physical pages that may not exist. This shouldn't be - // done for regular user code, but is fine for test - // purposes.) - applyPhysicalRegions(func(pr physicalRegion) bool { - pt.Map(usermem.Addr(pr.virtual), pr.length, pagetables.MapOpts{ - AccessType: usermem.AnyAccess, - User: true, - }, pr.physical) - return true // Keep iterating. - }) - } - }, func(c *vCPU) bool { - // Invoke the function with the extra data. - return fn(c, ®s, pt) - }) -} - -func TestApplicationSyscall(t *testing.T) { - applicationTest(t, true, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - FullRestore: true, - }, &si); err == platform.ErrContextInterrupt { - return true // Retry. - } else if err != nil { - t.Errorf("application syscall with full restore failed: %v", err) - } - return false - }) - applicationTest(t, true, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err == platform.ErrContextInterrupt { - return true // Retry. - } else if err != nil { - t.Errorf("application syscall with partial restore failed: %v", err) - } - return false - }) -} - -func TestApplicationFault(t *testing.T) { - applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - testutil.SetTouchTarget(regs, nil) // Cause fault. - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - FullRestore: true, - }, &si); err == platform.ErrContextInterrupt { - return true // Retry. - } else if err != platform.ErrContextSignal || si.Signo != int32(syscall.SIGSEGV) { - t.Errorf("application fault with full restore got (%v, %v), expected (%v, SIGSEGV)", err, si, platform.ErrContextSignal) - } - return false - }) - applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - testutil.SetTouchTarget(regs, nil) // Cause fault. - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err == platform.ErrContextInterrupt { - return true // Retry. - } else if err != platform.ErrContextSignal || si.Signo != int32(syscall.SIGSEGV) { - t.Errorf("application fault with partial restore got (%v, %v), expected (%v, SIGSEGV)", err, si, platform.ErrContextSignal) - } - return false - }) -} - -func TestRegistersSyscall(t *testing.T) { - applicationTest(t, true, testutil.TwiddleRegsSyscall, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - testutil.SetTestRegs(regs) // Fill values for all registers. - for { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err == platform.ErrContextInterrupt { - continue // Retry. - } else if err != nil { - t.Errorf("application register check with partial restore got unexpected error: %v", err) - } - if err := testutil.CheckTestRegs(regs, false); err != nil { - t.Errorf("application register check with partial restore failed: %v", err) - } - break // Done. - } - return false - }) -} - -func TestRegistersFault(t *testing.T) { - applicationTest(t, true, testutil.TwiddleRegsFault, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - testutil.SetTestRegs(regs) // Fill values for all registers. - for { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - FullRestore: true, - }, &si); err == platform.ErrContextInterrupt { - continue // Retry. - } else if err != platform.ErrContextSignal || si.Signo != int32(syscall.SIGSEGV) { - t.Errorf("application register check with full restore got unexpected error: %v", err) - } - if err := testutil.CheckTestRegs(regs, true); err != nil { - t.Errorf("application register check with full restore failed: %v", err) - } - break // Done. - } - return false - }) -} - -func TestSegments(t *testing.T) { - applicationTest(t, true, testutil.TwiddleSegments, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - testutil.SetTestSegments(regs) - for { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - FullRestore: true, - }, &si); err == platform.ErrContextInterrupt { - continue // Retry. - } else if err != nil { - t.Errorf("application segment check with full restore got unexpected error: %v", err) - } - if err := testutil.CheckTestSegments(regs); err != nil { - t.Errorf("application segment check with full restore failed: %v", err) - } - break // Done. - } - return false - }) -} - -func TestBounce(t *testing.T) { - applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - go func() { - time.Sleep(time.Millisecond) - c.BounceToKernel() - }() - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err != platform.ErrContextInterrupt { - t.Errorf("application partial restore: got %v, wanted %v", err, platform.ErrContextInterrupt) - } - return false - }) - applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - go func() { - time.Sleep(time.Millisecond) - c.BounceToKernel() - }() - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - FullRestore: true, - }, &si); err != platform.ErrContextInterrupt { - t.Errorf("application full restore: got %v, wanted %v", err, platform.ErrContextInterrupt) - } - return false - }) -} - -func TestBounceStress(t *testing.T) { - applicationTest(t, true, testutil.SpinLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - randomSleep := func() { - // O(hundreds of microseconds) is appropriate to ensure - // different overlaps and different schedules. - if n := rand.Intn(1000); n > 100 { - time.Sleep(time.Duration(n) * time.Microsecond) - } - } - for i := 0; i < 1000; i++ { - // Start an asynchronously executing goroutine that - // calls Bounce at pseudo-random point in time. - // This should wind up calling Bounce when the - // kernel is in various stages of the switch. - go func() { - randomSleep() - c.BounceToKernel() - }() - randomSleep() - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err != platform.ErrContextInterrupt { - t.Errorf("application partial restore: got %v, wanted %v", err, platform.ErrContextInterrupt) - } - c.unlock() - randomSleep() - c.lock() - } - return false - }) -} - -func TestInvalidate(t *testing.T) { - var data uintptr // Used below. - applicationTest(t, true, testutil.Touch, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - testutil.SetTouchTarget(regs, &data) // Read legitimate value. - for { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err == platform.ErrContextInterrupt { - continue // Retry. - } else if err != nil { - t.Errorf("application partial restore: got %v, wanted nil", err) - } - break // Done. - } - // Unmap the page containing data & invalidate. - pt.Unmap(usermem.Addr(reflect.ValueOf(&data).Pointer() & ^uintptr(usermem.PageSize-1)), usermem.PageSize) - for { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - Flush: true, - }, &si); err == platform.ErrContextInterrupt { - continue // Retry. - } else if err != platform.ErrContextSignal { - t.Errorf("application partial restore: got %v, wanted %v", err, platform.ErrContextSignal) - } - break // Success. - } - return false - }) -} - -// IsFault returns true iff the given signal represents a fault. -func IsFault(err error, si *arch.SignalInfo) bool { - return err == platform.ErrContextSignal && si.Signo == int32(syscall.SIGSEGV) -} - -func TestEmptyAddressSpace(t *testing.T) { - applicationTest(t, false, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err == platform.ErrContextInterrupt { - return true // Retry. - } else if !IsFault(err, &si) { - t.Errorf("first fault with partial restore failed got %v", err) - t.Logf("registers: %#v", ®s) - } - return false - }) - applicationTest(t, false, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - FullRestore: true, - }, &si); err == platform.ErrContextInterrupt { - return true // Retry. - } else if !IsFault(err, &si) { - t.Errorf("first fault with full restore failed got %v", err) - t.Logf("registers: %#v", ®s) - } - return false - }) -} - -func TestWrongVCPU(t *testing.T) { - kvmTest(t, nil, func(c1 *vCPU) bool { - kvmTest(t, nil, func(c2 *vCPU) bool { - // Basic test, one then the other. - bluepill(c1) - bluepill(c2) - if c2.switches == 0 { - // Don't allow the test to proceed if this fails. - t.Fatalf("wrong vCPU#2 switches: vCPU1=%+v,vCPU2=%+v", c1, c2) - } - - // Alternate vCPUs; we expect to need to trigger the - // wrong vCPU path on each switch. - for i := 0; i < 100; i++ { - bluepill(c1) - bluepill(c2) - } - if count := c1.switches; count < 90 { - t.Errorf("wrong vCPU#1 switches: vCPU1=%+v,vCPU2=%+v", c1, c2) - } - if count := c2.switches; count < 90 { - t.Errorf("wrong vCPU#2 switches: vCPU1=%+v,vCPU2=%+v", c1, c2) - } - return false - }) - return false - }) - kvmTest(t, nil, func(c1 *vCPU) bool { - kvmTest(t, nil, func(c2 *vCPU) bool { - bluepill(c1) - bluepill(c2) - return false - }) - return false - }) -} - -func BenchmarkApplicationSyscall(b *testing.B) { - var ( - i int // Iteration includes machine.Get() / machine.Put(). - a int // Count for ErrContextInterrupt. - ) - applicationTest(b, true, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err == platform.ErrContextInterrupt { - a++ - return true // Ignore. - } else if err != nil { - b.Fatalf("benchmark failed: %v", err) - } - i++ - return i < b.N - }) - if a != 0 { - b.Logf("ErrContextInterrupt occurred %d times (in %d iterations).", a, a+i) - } -} - -func BenchmarkKernelSyscall(b *testing.B) { - // Note that the target passed here is irrelevant, we never execute SwitchToUser. - applicationTest(b, true, testutil.Getpid, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - // iteration does not include machine.Get() / machine.Put(). - for i := 0; i < b.N; i++ { - testutil.Getpid() - } - return false - }) -} - -func BenchmarkWorldSwitchToUserRoundtrip(b *testing.B) { - // see BenchmarkApplicationSyscall. - var ( - i int - a int - ) - applicationTest(b, true, testutil.SyscallLoop, func(c *vCPU, regs *syscall.PtraceRegs, pt *pagetables.PageTables) bool { - var si arch.SignalInfo - if _, err := c.SwitchToUser(ring0.SwitchOpts{ - Registers: regs, - FloatingPointState: dummyFPState, - PageTables: pt, - }, &si); err == platform.ErrContextInterrupt { - a++ - return true // Ignore. - } else if err != nil { - b.Fatalf("benchmark failed: %v", err) - } - // This will intentionally cause the world switch. By executing - // a host syscall here, we force the transition between guest - // and host mode. - testutil.Getpid() - i++ - return i < b.N - }) - if a != 0 { - b.Logf("ErrContextInterrupt occurred %d times (in %d iterations).", a, a+i) - } -} diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index 09552837a..09552837a 100644..100755 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index b531f2f85..b531f2f85 100644..100755 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go diff --git a/pkg/sentry/platform/kvm/physical_map_amd64.go b/pkg/sentry/platform/kvm/physical_map_amd64.go index c5adfb577..c5adfb577 100644..100755 --- a/pkg/sentry/platform/kvm/physical_map_amd64.go +++ b/pkg/sentry/platform/kvm/physical_map_amd64.go diff --git a/pkg/sentry/platform/kvm/physical_map_arm64.go b/pkg/sentry/platform/kvm/physical_map_arm64.go index 4d8561453..4d8561453 100644..100755 --- a/pkg/sentry/platform/kvm/physical_map_arm64.go +++ b/pkg/sentry/platform/kvm/physical_map_arm64.go diff --git a/pkg/sentry/platform/kvm/testutil/BUILD b/pkg/sentry/platform/kvm/testutil/BUILD deleted file mode 100644 index f7605df8a..000000000 --- a/pkg/sentry/platform/kvm/testutil/BUILD +++ /dev/null @@ -1,16 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "testutil", - testonly = 1, - srcs = [ - "testutil.go", - "testutil_amd64.go", - "testutil_amd64.s", - "testutil_arm64.go", - "testutil_arm64.s", - ], - visibility = ["//pkg/sentry/platform/kvm:__pkg__"], -) diff --git a/pkg/sentry/platform/kvm/testutil/testutil.go b/pkg/sentry/platform/kvm/testutil/testutil.go deleted file mode 100644 index 5c1efa0fd..000000000 --- a/pkg/sentry/platform/kvm/testutil/testutil.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package testutil provides common assembly stubs for testing. -package testutil - -import ( - "fmt" - "strings" -) - -// Getpid executes a trivial system call. -func Getpid() - -// Touch touches the value in the first register. -func Touch() - -// SyscallLoop executes a syscall and loops. -func SyscallLoop() - -// SpinLoop spins on the CPU. -func SpinLoop() - -// HaltLoop immediately halts and loops. -func HaltLoop() - -// TwiddleRegsFault twiddles registers then faults. -func TwiddleRegsFault() - -// TwiddleRegsSyscall twiddles registers then executes a syscall. -func TwiddleRegsSyscall() - -// FloatingPointWorks is a floating point test. -// -// It returns true or false. -func FloatingPointWorks() bool - -// RegisterMismatchError is used for checking registers. -type RegisterMismatchError []string - -// Error returns a human-readable error. -func (r RegisterMismatchError) Error() string { - return strings.Join([]string(r), ";") -} - -// addRegisterMisatch allows simple chaining of register mismatches. -func addRegisterMismatch(err error, reg string, got, expected interface{}) error { - errStr := fmt.Sprintf("%s got %08x, expected %08x", reg, got, expected) - switch r := err.(type) { - case nil: - // Return a new register mismatch. - return RegisterMismatchError{errStr} - case RegisterMismatchError: - // Append the error. - r = append(r, errStr) - return r - default: - // Leave as is. - return err - } -} diff --git a/pkg/sentry/platform/kvm/testutil/testutil_amd64.go b/pkg/sentry/platform/kvm/testutil/testutil_amd64.go deleted file mode 100644 index 4c108abbf..000000000 --- a/pkg/sentry/platform/kvm/testutil/testutil_amd64.go +++ /dev/null @@ -1,138 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 - -package testutil - -import ( - "reflect" - "syscall" -) - -// TwiddleSegments reads segments into known registers. -func TwiddleSegments() - -// SetTestTarget sets the rip appropriately. -func SetTestTarget(regs *syscall.PtraceRegs, fn func()) { - regs.Rip = uint64(reflect.ValueOf(fn).Pointer()) -} - -// SetTouchTarget sets rax appropriately. -func SetTouchTarget(regs *syscall.PtraceRegs, target *uintptr) { - if target != nil { - regs.Rax = uint64(reflect.ValueOf(target).Pointer()) - } else { - regs.Rax = 0 - } -} - -// RewindSyscall rewinds a syscall RIP. -func RewindSyscall(regs *syscall.PtraceRegs) { - regs.Rip -= 2 -} - -// SetTestRegs initializes registers to known values. -func SetTestRegs(regs *syscall.PtraceRegs) { - regs.R15 = 0x15 - regs.R14 = 0x14 - regs.R13 = 0x13 - regs.R12 = 0x12 - regs.Rbp = 0xb9 - regs.Rbx = 0xb4 - regs.R11 = 0x11 - regs.R10 = 0x10 - regs.R9 = 0x09 - regs.R8 = 0x08 - regs.Rax = 0x44 - regs.Rcx = 0xc4 - regs.Rdx = 0xd4 - regs.Rsi = 0x51 - regs.Rdi = 0xd1 - regs.Rsp = 0x59 -} - -// CheckTestRegs checks that registers were twiddled per TwiddleRegs. -func CheckTestRegs(regs *syscall.PtraceRegs, full bool) (err error) { - if need := ^uint64(0x15); regs.R15 != need { - err = addRegisterMismatch(err, "R15", regs.R15, need) - } - if need := ^uint64(0x14); regs.R14 != need { - err = addRegisterMismatch(err, "R14", regs.R14, need) - } - if need := ^uint64(0x13); regs.R13 != need { - err = addRegisterMismatch(err, "R13", regs.R13, need) - } - if need := ^uint64(0x12); regs.R12 != need { - err = addRegisterMismatch(err, "R12", regs.R12, need) - } - if need := ^uint64(0xb9); regs.Rbp != need { - err = addRegisterMismatch(err, "Rbp", regs.Rbp, need) - } - if need := ^uint64(0xb4); regs.Rbx != need { - err = addRegisterMismatch(err, "Rbx", regs.Rbx, need) - } - if need := ^uint64(0x10); regs.R10 != need { - err = addRegisterMismatch(err, "R10", regs.R10, need) - } - if need := ^uint64(0x09); regs.R9 != need { - err = addRegisterMismatch(err, "R9", regs.R9, need) - } - if need := ^uint64(0x08); regs.R8 != need { - err = addRegisterMismatch(err, "R8", regs.R8, need) - } - if need := ^uint64(0x44); regs.Rax != need { - err = addRegisterMismatch(err, "Rax", regs.Rax, need) - } - if need := ^uint64(0xd4); regs.Rdx != need { - err = addRegisterMismatch(err, "Rdx", regs.Rdx, need) - } - if need := ^uint64(0x51); regs.Rsi != need { - err = addRegisterMismatch(err, "Rsi", regs.Rsi, need) - } - if need := ^uint64(0xd1); regs.Rdi != need { - err = addRegisterMismatch(err, "Rdi", regs.Rdi, need) - } - if need := ^uint64(0x59); regs.Rsp != need { - err = addRegisterMismatch(err, "Rsp", regs.Rsp, need) - } - // Rcx & R11 are ignored if !full is set. - if need := ^uint64(0x11); full && regs.R11 != need { - err = addRegisterMismatch(err, "R11", regs.R11, need) - } - if need := ^uint64(0xc4); full && regs.Rcx != need { - err = addRegisterMismatch(err, "Rcx", regs.Rcx, need) - } - return -} - -var fsData uint64 = 0x55 -var gsData uint64 = 0x85 - -// SetTestSegments initializes segments to known values. -func SetTestSegments(regs *syscall.PtraceRegs) { - regs.Fs_base = uint64(reflect.ValueOf(&fsData).Pointer()) - regs.Gs_base = uint64(reflect.ValueOf(&gsData).Pointer()) -} - -// CheckTestSegments checks that registers were twiddled per TwiddleSegments. -func CheckTestSegments(regs *syscall.PtraceRegs) (err error) { - if regs.Rax != fsData { - err = addRegisterMismatch(err, "Rax", regs.Rax, fsData) - } - if regs.Rbx != gsData { - err = addRegisterMismatch(err, "Rbx", regs.Rcx, gsData) - } - return -} diff --git a/pkg/sentry/platform/kvm/testutil/testutil_amd64.s b/pkg/sentry/platform/kvm/testutil/testutil_amd64.s deleted file mode 100644 index 491ec0c2a..000000000 --- a/pkg/sentry/platform/kvm/testutil/testutil_amd64.s +++ /dev/null @@ -1,98 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 - -// test_util_amd64.s provides AMD64 test functions. - -#include "funcdata.h" -#include "textflag.h" - -TEXT ·Getpid(SB),NOSPLIT,$0 - NO_LOCAL_POINTERS - MOVQ $39, AX // getpid - SYSCALL - RET - -TEXT ·Touch(SB),NOSPLIT,$0 -start: - MOVQ 0(AX), BX // deref AX - MOVQ $39, AX // getpid - SYSCALL - JMP start - -TEXT ·HaltLoop(SB),NOSPLIT,$0 -start: - HLT - JMP start - -TEXT ·SyscallLoop(SB),NOSPLIT,$0 -start: - SYSCALL - JMP start - -TEXT ·SpinLoop(SB),NOSPLIT,$0 -start: - JMP start - -TEXT ·FloatingPointWorks(SB),NOSPLIT,$0-8 - NO_LOCAL_POINTERS - MOVQ $1, AX - MOVQ AX, X0 - MOVQ $39, AX // getpid - SYSCALL - MOVQ X0, AX - CMPQ AX, $1 - SETEQ ret+0(FP) - RET - -#define TWIDDLE_REGS() \ - NOTQ R15; \ - NOTQ R14; \ - NOTQ R13; \ - NOTQ R12; \ - NOTQ BP; \ - NOTQ BX; \ - NOTQ R11; \ - NOTQ R10; \ - NOTQ R9; \ - NOTQ R8; \ - NOTQ AX; \ - NOTQ CX; \ - NOTQ DX; \ - NOTQ SI; \ - NOTQ DI; \ - NOTQ SP; - -TEXT ·TwiddleRegsSyscall(SB),NOSPLIT,$0 - TWIDDLE_REGS() - SYSCALL - RET // never reached - -TEXT ·TwiddleRegsFault(SB),NOSPLIT,$0 - TWIDDLE_REGS() - JMP AX // must fault - RET // never reached - -#define READ_FS() BYTE $0x64; BYTE $0x48; BYTE $0x8b; BYTE $0x00; -#define READ_GS() BYTE $0x65; BYTE $0x48; BYTE $0x8b; BYTE $0x00; - -TEXT ·TwiddleSegments(SB),NOSPLIT,$0 - MOVQ $0x0, AX - READ_GS() - MOVQ AX, BX - MOVQ $0x0, AX - READ_FS() - SYSCALL - RET // never reached diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go deleted file mode 100644 index 40b2e4acc..000000000 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go +++ /dev/null @@ -1,59 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -package testutil - -import ( - "fmt" - "reflect" - "syscall" -) - -// SetTestTarget sets the rip appropriately. -func SetTestTarget(regs *syscall.PtraceRegs, fn func()) { - regs.Pc = uint64(reflect.ValueOf(fn).Pointer()) -} - -// SetTouchTarget sets rax appropriately. -func SetTouchTarget(regs *syscall.PtraceRegs, target *uintptr) { - if target != nil { - regs.Regs[8] = uint64(reflect.ValueOf(target).Pointer()) - } else { - regs.Regs[8] = 0 - } -} - -// RewindSyscall rewinds a syscall RIP. -func RewindSyscall(regs *syscall.PtraceRegs) { - regs.Pc -= 4 -} - -// SetTestRegs initializes registers to known values. -func SetTestRegs(regs *syscall.PtraceRegs) { - for i := 0; i <= 30; i++ { - regs.Regs[i] = uint64(i) + 1 - } -} - -// CheckTestRegs checks that registers were twiddled per TwiddleRegs. -func CheckTestRegs(regs *syscall.PtraceRegs, full bool) (err error) { - for i := 0; i <= 30; i++ { - if need := ^uint64(i + 1); regs.Regs[i] != need { - err = addRegisterMismatch(err, fmt.Sprintf("R%d", i), regs.Regs[i], need) - } - } - return -} diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s deleted file mode 100644 index 0bebee852..000000000 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s +++ /dev/null @@ -1,106 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -// test_util_arm64.s provides ARM64 test functions. - -#include "funcdata.h" -#include "textflag.h" - -#define SYS_GETPID 172 - -// This function simulates the getpid syscall. -TEXT ·Getpid(SB),NOSPLIT,$0 - NO_LOCAL_POINTERS - MOVD $SYS_GETPID, R8 - SVC - RET - -TEXT ·Touch(SB),NOSPLIT,$0 -start: - MOVD 0(R8), R1 - MOVD $SYS_GETPID, R8 // getpid - SVC - B start - -TEXT ·HaltLoop(SB),NOSPLIT,$0 -start: - HLT - B start - -// This function simulates a loop of syscall. -TEXT ·SyscallLoop(SB),NOSPLIT,$0 -start: - SVC - B start - -TEXT ·SpinLoop(SB),NOSPLIT,$0 -start: - B start - -TEXT ·FloatingPointWorks(SB),NOSPLIT,$0-8 - NO_LOCAL_POINTERS - FMOVD $(9.9), F0 - MOVD $SYS_GETPID, R8 // getpid - SVC - FMOVD $(9.9), F1 - FCMPD F0, F1 - BNE isNaN - MOVD $1, R0 - MOVD R0, ret+0(FP) - RET -isNaN: - MOVD $0, ret+0(FP) - RET - -// MVN: bitwise logical NOT -// This case simulates an application that modified R0-R30. -#define TWIDDLE_REGS() \ - MVN R0, R0; \ - MVN R1, R1; \ - MVN R2, R2; \ - MVN R3, R3; \ - MVN R4, R4; \ - MVN R5, R5; \ - MVN R6, R6; \ - MVN R7, R7; \ - MVN R8, R8; \ - MVN R9, R9; \ - MVN R10, R10; \ - MVN R11, R11; \ - MVN R12, R12; \ - MVN R13, R13; \ - MVN R14, R14; \ - MVN R15, R15; \ - MVN R16, R16; \ - MVN R17, R17; \ - MVN R18_PLATFORM, R18_PLATFORM; \ - MVN R19, R19; \ - MVN R20, R20; \ - MVN R21, R21; \ - MVN R22, R22; \ - MVN R23, R23; \ - MVN R24, R24; \ - MVN R25, R25; \ - MVN R26, R26; \ - MVN R27, R27; \ - MVN g, g; \ - MVN R29, R29; \ - MVN R30, R30; - -TEXT ·TwiddleRegsSyscall(SB),NOSPLIT,$0 - TWIDDLE_REGS() - SVC - RET // never reached diff --git a/pkg/sentry/platform/kvm/virtual_map_test.go b/pkg/sentry/platform/kvm/virtual_map_test.go deleted file mode 100644 index 327e2be4f..000000000 --- a/pkg/sentry/platform/kvm/virtual_map_test.go +++ /dev/null @@ -1,93 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kvm - -import ( - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/usermem" -) - -type checker struct { - ok bool - accessType usermem.AccessType -} - -func (c *checker) Containing(addr uintptr) func(virtualRegion) { - c.ok = false // Reset for below calls. - return func(vr virtualRegion) { - if vr.virtual <= addr && addr < vr.virtual+vr.length { - c.ok = true - c.accessType = vr.accessType - } - } -} - -func TestParseMaps(t *testing.T) { - c := new(checker) - - // Simple test. - if err := applyVirtualRegions(c.Containing(0)); err != nil { - t.Fatalf("unexpected error: %v", err) - } - - // MMap a new page. - addr, _, errno := syscall.RawSyscall6( - syscall.SYS_MMAP, 0, usermem.PageSize, - syscall.PROT_READ|syscall.PROT_WRITE, - syscall.MAP_ANONYMOUS|syscall.MAP_PRIVATE, 0, 0) - if errno != 0 { - t.Fatalf("unexpected map error: %v", errno) - } - - // Re-parse maps. - if err := applyVirtualRegions(c.Containing(addr)); err != nil { - syscall.RawSyscall(syscall.SYS_MUNMAP, addr, usermem.PageSize, 0) - t.Fatalf("unexpected error: %v", err) - } - - // Assert that it now does contain the region. - if !c.ok { - syscall.RawSyscall(syscall.SYS_MUNMAP, addr, usermem.PageSize, 0) - t.Fatalf("updated map does not contain 0x%08x, expected true", addr) - } - - // Map the region as PROT_NONE. - newAddr, _, errno := syscall.RawSyscall6( - syscall.SYS_MMAP, addr, usermem.PageSize, - syscall.PROT_NONE, - syscall.MAP_ANONYMOUS|syscall.MAP_FIXED|syscall.MAP_PRIVATE, 0, 0) - if errno != 0 { - t.Fatalf("unexpected map error: %v", errno) - } - if newAddr != addr { - t.Fatalf("unable to remap address: got 0x%08x, wanted 0x%08x", newAddr, addr) - } - - // Re-parse maps. - if err := applyVirtualRegions(c.Containing(addr)); err != nil { - t.Fatalf("unexpected error: %v", err) - } - if !c.ok { - t.Fatalf("final map does not contain 0x%08x, expected true", addr) - } - if c.accessType.Read || c.accessType.Write || c.accessType.Execute { - t.Fatalf("final map has incorrect permissions for 0x%08x", addr) - } - - // Unmap the region. - syscall.RawSyscall(syscall.SYS_MUNMAP, addr, usermem.PageSize, 0) -} diff --git a/pkg/sentry/platform/platform_state_autogen.go b/pkg/sentry/platform/platform_state_autogen.go new file mode 100755 index 000000000..7597195ef --- /dev/null +++ b/pkg/sentry/platform/platform_state_autogen.go @@ -0,0 +1,24 @@ +// automatically generated by stateify. + +package platform + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *FileRange) beforeSave() {} +func (x *FileRange) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) +} + +func (x *FileRange) afterLoad() {} +func (x *FileRange) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) +} + +func init() { + state.Register("pkg/sentry/platform.FileRange", (*FileRange)(nil), state.Fns{Save: (*FileRange).save, Load: (*FileRange).load}) +} diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD deleted file mode 100644 index 30402c2df..000000000 --- a/pkg/sentry/platform/ptrace/BUILD +++ /dev/null @@ -1,39 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "ptrace", - srcs = [ - "filters.go", - "ptrace.go", - "ptrace_amd64.go", - "ptrace_arm64.go", - "ptrace_arm64_unsafe.go", - "ptrace_unsafe.go", - "stub_amd64.s", - "stub_arm64.s", - "stub_unsafe.go", - "subprocess.go", - "subprocess_amd64.go", - "subprocess_arm64.go", - "subprocess_linux.go", - "subprocess_linux_unsafe.go", - "subprocess_unsafe.go", - ], - visibility = ["//:sandbox"], - deps = [ - "//pkg/abi/linux", - "//pkg/log", - "//pkg/procid", - "//pkg/safecopy", - "//pkg/seccomp", - "//pkg/sentry/arch", - "//pkg/sentry/hostcpu", - "//pkg/sentry/platform", - "//pkg/sentry/platform/interrupt", - "//pkg/sync", - "//pkg/usermem", - "@org_golang_x_sys//unix:go_default_library", - ], -) diff --git a/pkg/sentry/platform/ptrace/ptrace_amd64_state_autogen.go b/pkg/sentry/platform/ptrace/ptrace_amd64_state_autogen.go new file mode 100755 index 000000000..f730ab393 --- /dev/null +++ b/pkg/sentry/platform/ptrace/ptrace_amd64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build amd64 + +package ptrace diff --git a/pkg/sentry/platform/ptrace/ptrace_arm64_state_autogen.go b/pkg/sentry/platform/ptrace/ptrace_arm64_state_autogen.go new file mode 100755 index 000000000..6239d1305 --- /dev/null +++ b/pkg/sentry/platform/ptrace/ptrace_arm64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build arm64 + +package ptrace diff --git a/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go b/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go index 32b8a6be9..32b8a6be9 100644..100755 --- a/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go +++ b/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe.go diff --git a/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe_state_autogen.go b/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe_state_autogen.go new file mode 100755 index 000000000..6239d1305 --- /dev/null +++ b/pkg/sentry/platform/ptrace/ptrace_arm64_unsafe_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build arm64 + +package ptrace diff --git a/pkg/sentry/platform/ptrace/ptrace_linux_state_autogen.go b/pkg/sentry/platform/ptrace/ptrace_linux_state_autogen.go new file mode 100755 index 000000000..9f90aef93 --- /dev/null +++ b/pkg/sentry/platform/ptrace/ptrace_linux_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build linux + +package ptrace diff --git a/pkg/sentry/platform/ptrace/ptrace_linux_unsafe_state_autogen.go b/pkg/sentry/platform/ptrace/ptrace_linux_unsafe_state_autogen.go new file mode 100755 index 000000000..45d94c547 --- /dev/null +++ b/pkg/sentry/platform/ptrace/ptrace_linux_unsafe_state_autogen.go @@ -0,0 +1,6 @@ +// automatically generated by stateify. + +// +build linux +// +build amd64 arm64 + +package ptrace diff --git a/pkg/sentry/platform/ptrace/ptrace_state_autogen.go b/pkg/sentry/platform/ptrace/ptrace_state_autogen.go new file mode 100755 index 000000000..4526fc387 --- /dev/null +++ b/pkg/sentry/platform/ptrace/ptrace_state_autogen.go @@ -0,0 +1,6 @@ +// automatically generated by stateify. + +// +build go1.12 +// +build !go1.15 + +package ptrace diff --git a/pkg/sentry/platform/ring0/BUILD b/pkg/sentry/platform/ring0/BUILD deleted file mode 100644 index 934b6fbcd..000000000 --- a/pkg/sentry/platform/ring0/BUILD +++ /dev/null @@ -1,83 +0,0 @@ -load("//tools:defs.bzl", "go_library") -load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") - -package(licenses = ["notice"]) - -go_template( - name = "defs_amd64", - srcs = [ - "defs.go", - "defs_amd64.go", - "offsets_amd64.go", - "x86.go", - ], - visibility = [":__subpackages__"], -) - -go_template( - name = "defs_arm64", - srcs = [ - "aarch64.go", - "defs.go", - "defs_arm64.go", - "offsets_arm64.go", - ], - visibility = [":__subpackages__"], -) - -go_template_instance( - name = "defs_impl_amd64", - out = "defs_impl_amd64.go", - package = "ring0", - template = ":defs_amd64", -) - -go_template_instance( - name = "defs_impl_arm64", - out = "defs_impl_arm64.go", - package = "ring0", - template = ":defs_arm64", -) - -genrule( - name = "entry_impl_amd64", - srcs = ["entry_amd64.s"], - outs = ["entry_impl_amd64.s"], - cmd = "(echo -e '// build +amd64\\n' && $(location //pkg/sentry/platform/ring0/gen_offsets) && cat $(SRCS)) > $@", - tools = ["//pkg/sentry/platform/ring0/gen_offsets"], -) - -genrule( - name = "entry_impl_arm64", - srcs = ["entry_arm64.s"], - outs = ["entry_impl_arm64.s"], - cmd = "(echo -e '// build +arm64\\n' && $(location //pkg/sentry/platform/ring0/gen_offsets) && cat $(SRCS)) > $@", - tools = ["//pkg/sentry/platform/ring0/gen_offsets"], -) - -go_library( - name = "ring0", - srcs = [ - "defs_impl_amd64.go", - "defs_impl_arm64.go", - "entry_amd64.go", - "entry_arm64.go", - "entry_impl_amd64.s", - "entry_impl_arm64.s", - "kernel.go", - "kernel_amd64.go", - "kernel_arm64.go", - "kernel_unsafe.go", - "lib_amd64.go", - "lib_amd64.s", - "lib_arm64.go", - "lib_arm64.s", - "ring0.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/cpuid", - "//pkg/sentry/platform/ring0/pagetables", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/platform/ring0/aarch64.go b/pkg/sentry/platform/ring0/aarch64.go deleted file mode 100644 index 8122ac6e2..000000000 --- a/pkg/sentry/platform/ring0/aarch64.go +++ /dev/null @@ -1,110 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -package ring0 - -// Useful bits. -const ( - _PGD_PGT_BASE = 0x1000 - _PGD_PGT_SIZE = 0x1000 - _PUD_PGT_BASE = 0x2000 - _PUD_PGT_SIZE = 0x1000 - _PMD_PGT_BASE = 0x3000 - _PMD_PGT_SIZE = 0x4000 - _PTE_PGT_BASE = 0x7000 - _PTE_PGT_SIZE = 0x1000 - - _PSR_D_BIT = 0x00000200 - _PSR_A_BIT = 0x00000100 - _PSR_I_BIT = 0x00000080 - _PSR_F_BIT = 0x00000040 -) - -const ( - // PSR bits - PSR_MODE_EL0t = 0x00000000 - PSR_MODE_EL1t = 0x00000004 - PSR_MODE_EL1h = 0x00000005 - PSR_MODE_MASK = 0x0000000f - - // KernelFlagsSet should always be set in the kernel. - KernelFlagsSet = PSR_MODE_EL1h - - // UserFlagsSet are always set in userspace. - UserFlagsSet = PSR_MODE_EL0t - - KernelFlagsClear = PSR_MODE_MASK - UserFlagsClear = PSR_MODE_MASK - - PsrDefaultSet = _PSR_D_BIT | _PSR_A_BIT | _PSR_I_BIT | _PSR_F_BIT -) - -// Vector is an exception vector. -type Vector uintptr - -// Exception vectors. -const ( - El1SyncInvalid = iota - El1IrqInvalid - El1FiqInvalid - El1ErrorInvalid - El1Sync - El1Irq - El1Fiq - El1Error - El0Sync - El0Irq - El0Fiq - El0Error - El0Sync_invalid - El0Irq_invalid - El0Fiq_invalid - El0Error_invalid - El1Sync_da - El1Sync_ia - El1Sync_sp_pc - El1Sync_undef - El1Sync_dbg - El1Sync_inv - El0Sync_svc - El0Sync_da - El0Sync_ia - El0Sync_fpsimd_acc - El0Sync_sve_acc - El0Sync_sys - El0Sync_sp_pc - El0Sync_undef - El0Sync_dbg - El0Sync_inv - _NR_INTERRUPTS -) - -// System call vectors. -const ( - Syscall Vector = El0Sync_svc - PageFault Vector = El0Sync_da - VirtualizationException Vector = El0Error -) - -// VirtualAddressBits returns the number bits available for virtual addresses. -func VirtualAddressBits() uint32 { - return 48 -} - -// PhysicalAddressBits returns the number of bits available for physical addresses. -func PhysicalAddressBits() uint32 { - return 40 -} diff --git a/pkg/sentry/platform/ring0/defs.go b/pkg/sentry/platform/ring0/defs.go deleted file mode 100644 index 86fd5ed58..000000000 --- a/pkg/sentry/platform/ring0/defs.go +++ /dev/null @@ -1,110 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ring0 - -import ( - "syscall" - - "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" -) - -// Kernel is a global kernel object. -// -// This contains global state, shared by multiple CPUs. -type Kernel struct { - KernelArchState -} - -// Hooks are hooks for kernel functions. -type Hooks interface { - // KernelSyscall is called for kernel system calls. - // - // Return from this call will restore registers and return to the kernel: the - // registers must be modified directly. - // - // If this function is not provided, a kernel exception results in halt. - // - // This must be go:nosplit, as this will be on the interrupt stack. - // Closures are permitted, as the pointer to the closure frame is not - // passed on the stack. - KernelSyscall() - - // KernelException handles an exception during kernel execution. - // - // Return from this call will restore registers and return to the kernel: the - // registers must be modified directly. - // - // If this function is not provided, a kernel exception results in halt. - // - // This must be go:nosplit, as this will be on the interrupt stack. - // Closures are permitted, as the pointer to the closure frame is not - // passed on the stack. - KernelException(Vector) -} - -// CPU is the per-CPU struct. -type CPU struct { - // self is a self reference. - // - // This is always guaranteed to be at offset zero. - self *CPU - - // kernel is reference to the kernel that this CPU was initialized - // with. This reference is kept for garbage collection purposes: CPU - // registers may refer to objects within the Kernel object that cannot - // be safely freed. - kernel *Kernel - - // CPUArchState is architecture-specific state. - CPUArchState - - // registers is a set of registers; these may be used on kernel system - // calls and exceptions via the Registers function. - registers syscall.PtraceRegs - - // hooks are kernel hooks. - hooks Hooks -} - -// Registers returns a modifiable-copy of the kernel registers. -// -// This is explicitly safe to call during KernelException and KernelSyscall. -// -//go:nosplit -func (c *CPU) Registers() *syscall.PtraceRegs { - return &c.registers -} - -// SwitchOpts are passed to the Switch function. -type SwitchOpts struct { - // Registers are the user register state. - Registers *syscall.PtraceRegs - - // FloatingPointState is a byte pointer where floating point state is - // saved and restored. - FloatingPointState *byte - - // PageTables are the application page tables. - PageTables *pagetables.PageTables - - // Flush indicates that a TLB flush should be forced on switch. - Flush bool - - // FullRestore indicates that an iret-based restore should be used. - FullRestore bool - - // SwitchArchOpts are architecture-specific options. - SwitchArchOpts -} diff --git a/pkg/sentry/platform/ring0/defs_amd64.go b/pkg/sentry/platform/ring0/defs_amd64.go deleted file mode 100644 index 9c6c2cf5c..000000000 --- a/pkg/sentry/platform/ring0/defs_amd64.go +++ /dev/null @@ -1,148 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 - -package ring0 - -import ( - "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/usermem" -) - -var ( - // UserspaceSize is the total size of userspace. - UserspaceSize = uintptr(1) << (VirtualAddressBits() - 1) - - // MaximumUserAddress is the largest possible user address. - MaximumUserAddress = (UserspaceSize - 1) & ^uintptr(usermem.PageSize-1) - - // KernelStartAddress is the starting kernel address. - KernelStartAddress = ^uintptr(0) - (UserspaceSize - 1) -) - -// Segment indices and Selectors. -const ( - // Index into GDT array. - _ = iota // Null descriptor first. - _ // Reserved (Linux is kernel 32). - segKcode // Kernel code (64-bit). - segKdata // Kernel data. - segUcode32 // User code (32-bit). - segUdata // User data. - segUcode64 // User code (64-bit). - segTss // Task segment descriptor. - segTssHi // Upper bits for TSS. - segLast // Last segment (terminal, not included). -) - -// Selectors. -const ( - Kcode Selector = segKcode << 3 - Kdata Selector = segKdata << 3 - Ucode32 Selector = (segUcode32 << 3) | 3 - Udata Selector = (segUdata << 3) | 3 - Ucode64 Selector = (segUcode64 << 3) | 3 - Tss Selector = segTss << 3 -) - -// Standard segments. -var ( - UserCodeSegment32 SegmentDescriptor - UserDataSegment SegmentDescriptor - UserCodeSegment64 SegmentDescriptor - KernelCodeSegment SegmentDescriptor - KernelDataSegment SegmentDescriptor -) - -// KernelOpts has initialization options for the kernel. -type KernelOpts struct { - // PageTables are the kernel pagetables; this must be provided. - PageTables *pagetables.PageTables -} - -// KernelArchState contains architecture-specific state. -type KernelArchState struct { - KernelOpts - - // globalIDT is our set of interrupt gates. - globalIDT idt64 -} - -// CPUArchState contains CPU-specific arch state. -type CPUArchState struct { - // stack is the stack used for interrupts on this CPU. - stack [256]byte - - // errorCode is the error code from the last exception. - errorCode uintptr - - // errorType indicates the type of error code here, it is always set - // along with the errorCode value above. - // - // It will either by 1, which indicates a user error, or 0 indicating a - // kernel error. If the error code below returns false (kernel error), - // then it cannot provide relevant information about the last - // exception. - errorType uintptr - - // gdt is the CPU's descriptor table. - gdt descriptorTable - - // tss is the CPU's task state. - tss TaskState64 -} - -// ErrorCode returns the last error code. -// -// The returned boolean indicates whether the error code corresponds to the -// last user error or not. If it does not, then fault information must be -// ignored. This is generally the result of a kernel fault while servicing a -// user fault. -// -//go:nosplit -func (c *CPU) ErrorCode() (value uintptr, user bool) { - return c.errorCode, c.errorType != 0 -} - -// ClearErrorCode resets the error code. -// -//go:nosplit -func (c *CPU) ClearErrorCode() { - c.errorCode = 0 // No code. - c.errorType = 1 // User mode. -} - -// SwitchArchOpts are embedded in SwitchOpts. -type SwitchArchOpts struct { - // UserPCID indicates that the application PCID to be used on switch, - // assuming that PCIDs are supported. - // - // Per pagetables_x86.go, a zero PCID implies a flush. - UserPCID uint16 - - // KernelPCID indicates that the kernel PCID to be used on return, - // assuming that PCIDs are supported. - // - // Per pagetables_x86.go, a zero PCID implies a flush. - KernelPCID uint16 -} - -func init() { - KernelCodeSegment.setCode64(0, 0, 0) - KernelDataSegment.setData(0, 0xffffffff, 0) - UserCodeSegment32.setCode64(0, 0, 3) - UserDataSegment.setData(0, 0xffffffff, 3) - UserCodeSegment64.setCode64(0, 0, 3) -} diff --git a/pkg/sentry/platform/ring0/defs_arm64.go b/pkg/sentry/platform/ring0/defs_arm64.go deleted file mode 100644 index 1583dda12..000000000 --- a/pkg/sentry/platform/ring0/defs_arm64.go +++ /dev/null @@ -1,137 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -package ring0 - -import ( - "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" - "gvisor.dev/gvisor/pkg/usermem" -) - -var ( - // UserspaceSize is the total size of userspace. - UserspaceSize = uintptr(1) << (VirtualAddressBits()) - - // MaximumUserAddress is the largest possible user address. - MaximumUserAddress = (UserspaceSize - 1) & ^uintptr(usermem.PageSize-1) - - // KernelStartAddress is the starting kernel address. - KernelStartAddress = ^uintptr(0) - (UserspaceSize - 1) -) - -// KernelOpts has initialization options for the kernel. -type KernelOpts struct { - // PageTables are the kernel pagetables; this must be provided. - PageTables *pagetables.PageTables -} - -// KernelArchState contains architecture-specific state. -type KernelArchState struct { - KernelOpts -} - -// CPUArchState contains CPU-specific arch state. -type CPUArchState struct { - // stack is the stack used for interrupts on this CPU. - stack [512]byte - - // errorCode is the error code from the last exception. - errorCode uintptr - - // errorType indicates the type of error code here, it is always set - // along with the errorCode value above. - // - // It will either by 1, which indicates a user error, or 0 indicating a - // kernel error. If the error code below returns false (kernel error), - // then it cannot provide relevant information about the last - // exception. - errorType uintptr - - // faultAddr is the value of far_el1. - faultAddr uintptr - - // ttbr0Kvm is the value of ttbr0_el1 for sentry. - ttbr0Kvm uintptr - - // ttbr0App is the value of ttbr0_el1 for applicaton. - ttbr0App uintptr - - // exception vector. - vecCode Vector - - // application context pointer. - appAddr uintptr - - // lazyVFP is the value of cpacr_el1. - lazyVFP uintptr -} - -// ErrorCode returns the last error code. -// -// The returned boolean indicates whether the error code corresponds to the -// last user error or not. If it does not, then fault information must be -// ignored. This is generally the result of a kernel fault while servicing a -// user fault. -// -//go:nosplit -func (c *CPU) ErrorCode() (value uintptr, user bool) { - return c.errorCode, c.errorType != 0 -} - -// ClearErrorCode resets the error code. -// -//go:nosplit -func (c *CPU) ClearErrorCode() { - c.errorCode = 0 // No code. - c.errorType = 1 // User mode. -} - -//go:nosplit -func (c *CPU) GetFaultAddr() (value uintptr) { - return c.faultAddr -} - -//go:nosplit -func (c *CPU) SetTtbr0Kvm(value uintptr) { - c.ttbr0Kvm = value -} - -//go:nosplit -func (c *CPU) SetTtbr0App(value uintptr) { - c.ttbr0App = value -} - -//go:nosplit -func (c *CPU) GetVector() (value Vector) { - return c.vecCode -} - -//go:nosplit -func (c *CPU) SetAppAddr(value uintptr) { - c.appAddr = value -} - -// SwitchArchOpts are embedded in SwitchOpts. -type SwitchArchOpts struct { - // UserASID indicates that the application ASID to be used on switch, - UserASID uint16 - - // KernelASID indicates that the kernel ASID to be used on return, - KernelASID uint16 -} - -func init() { -} diff --git a/pkg/sentry/platform/ring0/defs_impl_amd64.go b/pkg/sentry/platform/ring0/defs_impl_amd64.go new file mode 100755 index 000000000..178eabd85 --- /dev/null +++ b/pkg/sentry/platform/ring0/defs_impl_amd64.go @@ -0,0 +1,538 @@ +package ring0 + +import ( + "fmt" + "gvisor.dev/gvisor/pkg/cpuid" + "io" + "reflect" + "syscall" + + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Kernel is a global kernel object. +// +// This contains global state, shared by multiple CPUs. +type Kernel struct { + KernelArchState +} + +// Hooks are hooks for kernel functions. +type Hooks interface { + // KernelSyscall is called for kernel system calls. + // + // Return from this call will restore registers and return to the kernel: the + // registers must be modified directly. + // + // If this function is not provided, a kernel exception results in halt. + // + // This must be go:nosplit, as this will be on the interrupt stack. + // Closures are permitted, as the pointer to the closure frame is not + // passed on the stack. + KernelSyscall() + + // KernelException handles an exception during kernel execution. + // + // Return from this call will restore registers and return to the kernel: the + // registers must be modified directly. + // + // If this function is not provided, a kernel exception results in halt. + // + // This must be go:nosplit, as this will be on the interrupt stack. + // Closures are permitted, as the pointer to the closure frame is not + // passed on the stack. + KernelException(Vector) +} + +// CPU is the per-CPU struct. +type CPU struct { + // self is a self reference. + // + // This is always guaranteed to be at offset zero. + self *CPU + + // kernel is reference to the kernel that this CPU was initialized + // with. This reference is kept for garbage collection purposes: CPU + // registers may refer to objects within the Kernel object that cannot + // be safely freed. + kernel *Kernel + + // CPUArchState is architecture-specific state. + CPUArchState + + // registers is a set of registers; these may be used on kernel system + // calls and exceptions via the Registers function. + registers syscall.PtraceRegs + + // hooks are kernel hooks. + hooks Hooks +} + +// Registers returns a modifiable-copy of the kernel registers. +// +// This is explicitly safe to call during KernelException and KernelSyscall. +// +//go:nosplit +func (c *CPU) Registers() *syscall.PtraceRegs { + return &c.registers +} + +// SwitchOpts are passed to the Switch function. +type SwitchOpts struct { + // Registers are the user register state. + Registers *syscall.PtraceRegs + + // FloatingPointState is a byte pointer where floating point state is + // saved and restored. + FloatingPointState *byte + + // PageTables are the application page tables. + PageTables *pagetables.PageTables + + // Flush indicates that a TLB flush should be forced on switch. + Flush bool + + // FullRestore indicates that an iret-based restore should be used. + FullRestore bool + + // SwitchArchOpts are architecture-specific options. + SwitchArchOpts +} + +var ( + // UserspaceSize is the total size of userspace. + UserspaceSize = uintptr(1) << (VirtualAddressBits() - 1) + + // MaximumUserAddress is the largest possible user address. + MaximumUserAddress = (UserspaceSize - 1) & ^uintptr(usermem.PageSize-1) + + // KernelStartAddress is the starting kernel address. + KernelStartAddress = ^uintptr(0) - (UserspaceSize - 1) +) + +// Segment indices and Selectors. +const ( + // Index into GDT array. + _ = iota // Null descriptor first. + _ // Reserved (Linux is kernel 32). + segKcode // Kernel code (64-bit). + segKdata // Kernel data. + segUcode32 // User code (32-bit). + segUdata // User data. + segUcode64 // User code (64-bit). + segTss // Task segment descriptor. + segTssHi // Upper bits for TSS. + segLast // Last segment (terminal, not included). +) + +// Selectors. +const ( + Kcode Selector = segKcode << 3 + Kdata Selector = segKdata << 3 + Ucode32 Selector = (segUcode32 << 3) | 3 + Udata Selector = (segUdata << 3) | 3 + Ucode64 Selector = (segUcode64 << 3) | 3 + Tss Selector = segTss << 3 +) + +// Standard segments. +var ( + UserCodeSegment32 SegmentDescriptor + UserDataSegment SegmentDescriptor + UserCodeSegment64 SegmentDescriptor + KernelCodeSegment SegmentDescriptor + KernelDataSegment SegmentDescriptor +) + +// KernelOpts has initialization options for the kernel. +type KernelOpts struct { + // PageTables are the kernel pagetables; this must be provided. + PageTables *pagetables.PageTables +} + +// KernelArchState contains architecture-specific state. +type KernelArchState struct { + KernelOpts + + // globalIDT is our set of interrupt gates. + globalIDT idt64 +} + +// CPUArchState contains CPU-specific arch state. +type CPUArchState struct { + // stack is the stack used for interrupts on this CPU. + stack [256]byte + + // errorCode is the error code from the last exception. + errorCode uintptr + + // errorType indicates the type of error code here, it is always set + // along with the errorCode value above. + // + // It will either by 1, which indicates a user error, or 0 indicating a + // kernel error. If the error code below returns false (kernel error), + // then it cannot provide relevant information about the last + // exception. + errorType uintptr + + // gdt is the CPU's descriptor table. + gdt descriptorTable + + // tss is the CPU's task state. + tss TaskState64 +} + +// ErrorCode returns the last error code. +// +// The returned boolean indicates whether the error code corresponds to the +// last user error or not. If it does not, then fault information must be +// ignored. This is generally the result of a kernel fault while servicing a +// user fault. +// +//go:nosplit +func (c *CPU) ErrorCode() (value uintptr, user bool) { + return c.errorCode, c.errorType != 0 +} + +// ClearErrorCode resets the error code. +// +//go:nosplit +func (c *CPU) ClearErrorCode() { + c.errorCode = 0 + c.errorType = 1 +} + +// SwitchArchOpts are embedded in SwitchOpts. +type SwitchArchOpts struct { + // UserPCID indicates that the application PCID to be used on switch, + // assuming that PCIDs are supported. + // + // Per pagetables_x86.go, a zero PCID implies a flush. + UserPCID uint16 + + // KernelPCID indicates that the kernel PCID to be used on return, + // assuming that PCIDs are supported. + // + // Per pagetables_x86.go, a zero PCID implies a flush. + KernelPCID uint16 +} + +func init() { + KernelCodeSegment.setCode64(0, 0, 0) + KernelDataSegment.setData(0, 0xffffffff, 0) + UserCodeSegment32.setCode64(0, 0, 3) + UserDataSegment.setData(0, 0xffffffff, 3) + UserCodeSegment64.setCode64(0, 0, 3) +} + +// Emit prints architecture-specific offsets. +func Emit(w io.Writer) { + fmt.Fprintf(w, "// Automatically generated, do not edit.\n") + + c := &CPU{} + fmt.Fprintf(w, "\n// CPU offsets.\n") + fmt.Fprintf(w, "#define CPU_SELF 0x%02x\n", reflect.ValueOf(&c.self).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_REGISTERS 0x%02x\n", reflect.ValueOf(&c.registers).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_STACK_TOP 0x%02x\n", reflect.ValueOf(&c.stack[0]).Pointer()-reflect.ValueOf(c).Pointer()+uintptr(len(c.stack))) + fmt.Fprintf(w, "#define CPU_ERROR_CODE 0x%02x\n", reflect.ValueOf(&c.errorCode).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_ERROR_TYPE 0x%02x\n", reflect.ValueOf(&c.errorType).Pointer()-reflect.ValueOf(c).Pointer()) + + fmt.Fprintf(w, "\n// Bits.\n") + fmt.Fprintf(w, "#define _RFLAGS_IF 0x%02x\n", _RFLAGS_IF) + fmt.Fprintf(w, "#define _KERNEL_FLAGS 0x%02x\n", KernelFlagsSet) + + fmt.Fprintf(w, "\n// Vectors.\n") + fmt.Fprintf(w, "#define DivideByZero 0x%02x\n", DivideByZero) + fmt.Fprintf(w, "#define Debug 0x%02x\n", Debug) + fmt.Fprintf(w, "#define NMI 0x%02x\n", NMI) + fmt.Fprintf(w, "#define Breakpoint 0x%02x\n", Breakpoint) + fmt.Fprintf(w, "#define Overflow 0x%02x\n", Overflow) + fmt.Fprintf(w, "#define BoundRangeExceeded 0x%02x\n", BoundRangeExceeded) + fmt.Fprintf(w, "#define InvalidOpcode 0x%02x\n", InvalidOpcode) + fmt.Fprintf(w, "#define DeviceNotAvailable 0x%02x\n", DeviceNotAvailable) + fmt.Fprintf(w, "#define DoubleFault 0x%02x\n", DoubleFault) + fmt.Fprintf(w, "#define CoprocessorSegmentOverrun 0x%02x\n", CoprocessorSegmentOverrun) + fmt.Fprintf(w, "#define InvalidTSS 0x%02x\n", InvalidTSS) + fmt.Fprintf(w, "#define SegmentNotPresent 0x%02x\n", SegmentNotPresent) + fmt.Fprintf(w, "#define StackSegmentFault 0x%02x\n", StackSegmentFault) + fmt.Fprintf(w, "#define GeneralProtectionFault 0x%02x\n", GeneralProtectionFault) + fmt.Fprintf(w, "#define PageFault 0x%02x\n", PageFault) + fmt.Fprintf(w, "#define X87FloatingPointException 0x%02x\n", X87FloatingPointException) + fmt.Fprintf(w, "#define AlignmentCheck 0x%02x\n", AlignmentCheck) + fmt.Fprintf(w, "#define MachineCheck 0x%02x\n", MachineCheck) + fmt.Fprintf(w, "#define SIMDFloatingPointException 0x%02x\n", SIMDFloatingPointException) + fmt.Fprintf(w, "#define VirtualizationException 0x%02x\n", VirtualizationException) + fmt.Fprintf(w, "#define SecurityException 0x%02x\n", SecurityException) + fmt.Fprintf(w, "#define SyscallInt80 0x%02x\n", SyscallInt80) + fmt.Fprintf(w, "#define Syscall 0x%02x\n", Syscall) + + p := &syscall.PtraceRegs{} + fmt.Fprintf(w, "\n// Ptrace registers.\n") + fmt.Fprintf(w, "#define PTRACE_R15 0x%02x\n", reflect.ValueOf(&p.R15).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R14 0x%02x\n", reflect.ValueOf(&p.R14).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R13 0x%02x\n", reflect.ValueOf(&p.R13).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R12 0x%02x\n", reflect.ValueOf(&p.R12).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RBP 0x%02x\n", reflect.ValueOf(&p.Rbp).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RBX 0x%02x\n", reflect.ValueOf(&p.Rbx).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R11 0x%02x\n", reflect.ValueOf(&p.R11).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R10 0x%02x\n", reflect.ValueOf(&p.R10).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R9 0x%02x\n", reflect.ValueOf(&p.R9).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_R8 0x%02x\n", reflect.ValueOf(&p.R8).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RAX 0x%02x\n", reflect.ValueOf(&p.Rax).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RCX 0x%02x\n", reflect.ValueOf(&p.Rcx).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RDX 0x%02x\n", reflect.ValueOf(&p.Rdx).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RSI 0x%02x\n", reflect.ValueOf(&p.Rsi).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RDI 0x%02x\n", reflect.ValueOf(&p.Rdi).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_ORIGRAX 0x%02x\n", reflect.ValueOf(&p.Orig_rax).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RIP 0x%02x\n", reflect.ValueOf(&p.Rip).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_CS 0x%02x\n", reflect.ValueOf(&p.Cs).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_FLAGS 0x%02x\n", reflect.ValueOf(&p.Eflags).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_RSP 0x%02x\n", reflect.ValueOf(&p.Rsp).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_SS 0x%02x\n", reflect.ValueOf(&p.Ss).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_FS 0x%02x\n", reflect.ValueOf(&p.Fs_base).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_GS 0x%02x\n", reflect.ValueOf(&p.Gs_base).Pointer()-reflect.ValueOf(p).Pointer()) +} + +// Useful bits. +const ( + _CR0_PE = 1 << 0 + _CR0_ET = 1 << 4 + _CR0_AM = 1 << 18 + _CR0_PG = 1 << 31 + + _CR4_PSE = 1 << 4 + _CR4_PAE = 1 << 5 + _CR4_PGE = 1 << 7 + _CR4_OSFXSR = 1 << 9 + _CR4_OSXMMEXCPT = 1 << 10 + _CR4_FSGSBASE = 1 << 16 + _CR4_PCIDE = 1 << 17 + _CR4_OSXSAVE = 1 << 18 + _CR4_SMEP = 1 << 20 + + _RFLAGS_AC = 1 << 18 + _RFLAGS_NT = 1 << 14 + _RFLAGS_IOPL = 3 << 12 + _RFLAGS_DF = 1 << 10 + _RFLAGS_IF = 1 << 9 + _RFLAGS_STEP = 1 << 8 + _RFLAGS_RESERVED = 1 << 1 + + _EFER_SCE = 0x001 + _EFER_LME = 0x100 + _EFER_LMA = 0x400 + _EFER_NX = 0x800 + + _MSR_STAR = 0xc0000081 + _MSR_LSTAR = 0xc0000082 + _MSR_CSTAR = 0xc0000083 + _MSR_SYSCALL_MASK = 0xc0000084 + _MSR_PLATFORM_INFO = 0xce + _MSR_MISC_FEATURES = 0x140 + + _PLATFORM_INFO_CPUID_FAULT = 1 << 31 + + _MISC_FEATURE_CPUID_TRAP = 0x1 +) + +const ( + // KernelFlagsSet should always be set in the kernel. + KernelFlagsSet = _RFLAGS_RESERVED + + // UserFlagsSet are always set in userspace. + UserFlagsSet = _RFLAGS_RESERVED | _RFLAGS_IF + + // KernelFlagsClear should always be clear in the kernel. + KernelFlagsClear = _RFLAGS_STEP | _RFLAGS_IF | _RFLAGS_IOPL | _RFLAGS_AC | _RFLAGS_NT + + // UserFlagsClear are always cleared in userspace. + UserFlagsClear = _RFLAGS_NT | _RFLAGS_IOPL +) + +// Vector is an exception vector. +type Vector uintptr + +// Exception vectors. +const ( + DivideByZero Vector = iota + Debug + NMI + Breakpoint + Overflow + BoundRangeExceeded + InvalidOpcode + DeviceNotAvailable + DoubleFault + CoprocessorSegmentOverrun + InvalidTSS + SegmentNotPresent + StackSegmentFault + GeneralProtectionFault + PageFault + _ + X87FloatingPointException + AlignmentCheck + MachineCheck + SIMDFloatingPointException + VirtualizationException + SecurityException = 0x1e + SyscallInt80 = 0x80 + _NR_INTERRUPTS = SyscallInt80 + 1 +) + +// System call vectors. +const ( + Syscall Vector = _NR_INTERRUPTS +) + +// VirtualAddressBits returns the number bits available for virtual addresses. +// +// Note that sign-extension semantics apply to the highest order bit. +// +// FIXME(b/69382326): This should use the cpuid passed to Init. +func VirtualAddressBits() uint32 { + ax, _, _, _ := cpuid.HostID(0x80000008, 0) + return (ax >> 8) & 0xff +} + +// PhysicalAddressBits returns the number of bits available for physical addresses. +// +// FIXME(b/69382326): This should use the cpuid passed to Init. +func PhysicalAddressBits() uint32 { + ax, _, _, _ := cpuid.HostID(0x80000008, 0) + return ax & 0xff +} + +// Selector is a segment Selector. +type Selector uint16 + +// SegmentDescriptor is a segment descriptor. +type SegmentDescriptor struct { + bits [2]uint32 +} + +// descriptorTable is a collection of descriptors. +type descriptorTable [32]SegmentDescriptor + +// SegmentDescriptorFlags are typed flags within a descriptor. +type SegmentDescriptorFlags uint32 + +// SegmentDescriptorFlag declarations. +const ( + SegmentDescriptorAccess SegmentDescriptorFlags = 1 << 8 // Access bit (always set). + SegmentDescriptorWrite = 1 << 9 // Write permission. + SegmentDescriptorExpandDown = 1 << 10 // Grows down, not used. + SegmentDescriptorExecute = 1 << 11 // Execute permission. + SegmentDescriptorSystem = 1 << 12 // Zero => system, 1 => user code/data. + SegmentDescriptorPresent = 1 << 15 // Present. + SegmentDescriptorAVL = 1 << 20 // Available. + SegmentDescriptorLong = 1 << 21 // Long mode. + SegmentDescriptorDB = 1 << 22 // 16 or 32-bit. + SegmentDescriptorG = 1 << 23 // Granularity: page or byte. +) + +// Base returns the descriptor's base linear address. +func (d *SegmentDescriptor) Base() uint32 { + return d.bits[1]&0xFF000000 | (d.bits[1]&0x000000FF)<<16 | d.bits[0]>>16 +} + +// Limit returns the descriptor size. +func (d *SegmentDescriptor) Limit() uint32 { + l := d.bits[0]&0xFFFF | d.bits[1]&0xF0000 + if d.bits[1]&uint32(SegmentDescriptorG) != 0 { + l <<= 12 + l |= 0xFFF + } + return l +} + +// Flags returns descriptor flags. +func (d *SegmentDescriptor) Flags() SegmentDescriptorFlags { + return SegmentDescriptorFlags(d.bits[1] & 0x00F09F00) +} + +// DPL returns the descriptor privilege level. +func (d *SegmentDescriptor) DPL() int { + return int((d.bits[1] >> 13) & 3) +} + +func (d *SegmentDescriptor) setNull() { + d.bits[0] = 0 + d.bits[1] = 0 +} + +func (d *SegmentDescriptor) set(base, limit uint32, dpl int, flags SegmentDescriptorFlags) { + flags |= SegmentDescriptorPresent + if limit>>12 != 0 { + limit >>= 12 + flags |= SegmentDescriptorG + } + d.bits[0] = base<<16 | limit&0xFFFF + d.bits[1] = base&0xFF000000 | (base>>16)&0xFF | limit&0x000F0000 | uint32(flags) | uint32(dpl)<<13 +} + +func (d *SegmentDescriptor) setCode32(base, limit uint32, dpl int) { + d.set(base, limit, dpl, + SegmentDescriptorDB| + SegmentDescriptorExecute| + SegmentDescriptorSystem) +} + +func (d *SegmentDescriptor) setCode64(base, limit uint32, dpl int) { + d.set(base, limit, dpl, + SegmentDescriptorG| + SegmentDescriptorLong| + SegmentDescriptorExecute| + SegmentDescriptorSystem) +} + +func (d *SegmentDescriptor) setData(base, limit uint32, dpl int) { + d.set(base, limit, dpl, + SegmentDescriptorWrite| + SegmentDescriptorSystem) +} + +// setHi is only used for the TSS segment, which is magically 64-bits. +func (d *SegmentDescriptor) setHi(base uint32) { + d.bits[0] = base + d.bits[1] = 0 +} + +// Gate64 is a 64-bit task, trap, or interrupt gate. +type Gate64 struct { + bits [4]uint32 +} + +// idt64 is a 64-bit interrupt descriptor table. +type idt64 [_NR_INTERRUPTS]Gate64 + +func (g *Gate64) setInterrupt(cs Selector, rip uint64, dpl int, ist int) { + g.bits[0] = uint32(cs)<<16 | uint32(rip)&0xFFFF + g.bits[1] = uint32(rip)&0xFFFF0000 | SegmentDescriptorPresent | uint32(dpl)<<13 | 14<<8 | uint32(ist)&0x7 + g.bits[2] = uint32(rip >> 32) +} + +func (g *Gate64) setTrap(cs Selector, rip uint64, dpl int, ist int) { + g.setInterrupt(cs, rip, dpl, ist) + g.bits[1] |= 1 << 8 +} + +// TaskState64 is a 64-bit task state structure. +type TaskState64 struct { + _ uint32 + rsp0Lo, rsp0Hi uint32 + rsp1Lo, rsp1Hi uint32 + rsp2Lo, rsp2Hi uint32 + _ [2]uint32 + ist1Lo, ist1Hi uint32 + ist2Lo, ist2Hi uint32 + ist3Lo, ist3Hi uint32 + ist4Lo, ist4Hi uint32 + ist5Lo, ist5Hi uint32 + ist6Lo, ist6Hi uint32 + ist7Lo, ist7Hi uint32 + _ [2]uint32 + _ uint16 + ioPerm uint16 +} diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/defs_impl_arm64.go index 057fb5c69..23d30c1ef 100644..100755 --- a/pkg/sentry/platform/ring0/offsets_arm64.go +++ b/pkg/sentry/platform/ring0/defs_impl_arm64.go @@ -1,19 +1,3 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - package ring0 import ( @@ -21,8 +5,308 @@ import ( "io" "reflect" "syscall" + + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/usermem" +) + +// Useful bits. +const ( + _PGD_PGT_BASE = 0x1000 + _PGD_PGT_SIZE = 0x1000 + _PUD_PGT_BASE = 0x2000 + _PUD_PGT_SIZE = 0x1000 + _PMD_PGT_BASE = 0x3000 + _PMD_PGT_SIZE = 0x4000 + _PTE_PGT_BASE = 0x7000 + _PTE_PGT_SIZE = 0x1000 + + _PSR_D_BIT = 0x00000200 + _PSR_A_BIT = 0x00000100 + _PSR_I_BIT = 0x00000080 + _PSR_F_BIT = 0x00000040 +) + +const ( + // PSR bits + PSR_MODE_EL0t = 0x00000000 + PSR_MODE_EL1t = 0x00000004 + PSR_MODE_EL1h = 0x00000005 + PSR_MODE_MASK = 0x0000000f + + // KernelFlagsSet should always be set in the kernel. + KernelFlagsSet = PSR_MODE_EL1h + + // UserFlagsSet are always set in userspace. + UserFlagsSet = PSR_MODE_EL0t + + KernelFlagsClear = PSR_MODE_MASK + UserFlagsClear = PSR_MODE_MASK + + PsrDefaultSet = _PSR_D_BIT | _PSR_A_BIT | _PSR_I_BIT | _PSR_F_BIT ) +// Vector is an exception vector. +type Vector uintptr + +// Exception vectors. +const ( + El1SyncInvalid = iota + El1IrqInvalid + El1FiqInvalid + El1ErrorInvalid + El1Sync + El1Irq + El1Fiq + El1Error + El0Sync + El0Irq + El0Fiq + El0Error + El0Sync_invalid + El0Irq_invalid + El0Fiq_invalid + El0Error_invalid + El1Sync_da + El1Sync_ia + El1Sync_sp_pc + El1Sync_undef + El1Sync_dbg + El1Sync_inv + El0Sync_svc + El0Sync_da + El0Sync_ia + El0Sync_fpsimd_acc + El0Sync_sve_acc + El0Sync_sys + El0Sync_sp_pc + El0Sync_undef + El0Sync_dbg + El0Sync_inv + _NR_INTERRUPTS +) + +// System call vectors. +const ( + Syscall Vector = El0Sync_svc + PageFault Vector = El0Sync_da + VirtualizationException Vector = El0Error +) + +// VirtualAddressBits returns the number bits available for virtual addresses. +func VirtualAddressBits() uint32 { + return 48 +} + +// PhysicalAddressBits returns the number of bits available for physical addresses. +func PhysicalAddressBits() uint32 { + return 40 +} + +// Kernel is a global kernel object. +// +// This contains global state, shared by multiple CPUs. +type Kernel struct { + KernelArchState +} + +// Hooks are hooks for kernel functions. +type Hooks interface { + // KernelSyscall is called for kernel system calls. + // + // Return from this call will restore registers and return to the kernel: the + // registers must be modified directly. + // + // If this function is not provided, a kernel exception results in halt. + // + // This must be go:nosplit, as this will be on the interrupt stack. + // Closures are permitted, as the pointer to the closure frame is not + // passed on the stack. + KernelSyscall() + + // KernelException handles an exception during kernel execution. + // + // Return from this call will restore registers and return to the kernel: the + // registers must be modified directly. + // + // If this function is not provided, a kernel exception results in halt. + // + // This must be go:nosplit, as this will be on the interrupt stack. + // Closures are permitted, as the pointer to the closure frame is not + // passed on the stack. + KernelException(Vector) +} + +// CPU is the per-CPU struct. +type CPU struct { + // self is a self reference. + // + // This is always guaranteed to be at offset zero. + self *CPU + + // kernel is reference to the kernel that this CPU was initialized + // with. This reference is kept for garbage collection purposes: CPU + // registers may refer to objects within the Kernel object that cannot + // be safely freed. + kernel *Kernel + + // CPUArchState is architecture-specific state. + CPUArchState + + // registers is a set of registers; these may be used on kernel system + // calls and exceptions via the Registers function. + registers syscall.PtraceRegs + + // hooks are kernel hooks. + hooks Hooks +} + +// Registers returns a modifiable-copy of the kernel registers. +// +// This is explicitly safe to call during KernelException and KernelSyscall. +// +//go:nosplit +func (c *CPU) Registers() *syscall.PtraceRegs { + return &c.registers +} + +// SwitchOpts are passed to the Switch function. +type SwitchOpts struct { + // Registers are the user register state. + Registers *syscall.PtraceRegs + + // FloatingPointState is a byte pointer where floating point state is + // saved and restored. + FloatingPointState *byte + + // PageTables are the application page tables. + PageTables *pagetables.PageTables + + // Flush indicates that a TLB flush should be forced on switch. + Flush bool + + // FullRestore indicates that an iret-based restore should be used. + FullRestore bool + + // SwitchArchOpts are architecture-specific options. + SwitchArchOpts +} + +var ( + // UserspaceSize is the total size of userspace. + UserspaceSize = uintptr(1) << (VirtualAddressBits()) + + // MaximumUserAddress is the largest possible user address. + MaximumUserAddress = (UserspaceSize - 1) & ^uintptr(usermem.PageSize-1) + + // KernelStartAddress is the starting kernel address. + KernelStartAddress = ^uintptr(0) - (UserspaceSize - 1) +) + +// KernelOpts has initialization options for the kernel. +type KernelOpts struct { + // PageTables are the kernel pagetables; this must be provided. + PageTables *pagetables.PageTables +} + +// KernelArchState contains architecture-specific state. +type KernelArchState struct { + KernelOpts +} + +// CPUArchState contains CPU-specific arch state. +type CPUArchState struct { + // stack is the stack used for interrupts on this CPU. + stack [512]byte + + // errorCode is the error code from the last exception. + errorCode uintptr + + // errorType indicates the type of error code here, it is always set + // along with the errorCode value above. + // + // It will either by 1, which indicates a user error, or 0 indicating a + // kernel error. If the error code below returns false (kernel error), + // then it cannot provide relevant information about the last + // exception. + errorType uintptr + + // faultAddr is the value of far_el1. + faultAddr uintptr + + // ttbr0Kvm is the value of ttbr0_el1 for sentry. + ttbr0Kvm uintptr + + // ttbr0App is the value of ttbr0_el1 for applicaton. + ttbr0App uintptr + + // exception vector. + vecCode Vector + + // application context pointer. + appAddr uintptr + + // lazyVFP is the value of cpacr_el1. + lazyVFP uintptr +} + +// ErrorCode returns the last error code. +// +// The returned boolean indicates whether the error code corresponds to the +// last user error or not. If it does not, then fault information must be +// ignored. This is generally the result of a kernel fault while servicing a +// user fault. +// +//go:nosplit +func (c *CPU) ErrorCode() (value uintptr, user bool) { + return c.errorCode, c.errorType != 0 +} + +// ClearErrorCode resets the error code. +// +//go:nosplit +func (c *CPU) ClearErrorCode() { + c.errorCode = 0 + c.errorType = 1 +} + +//go:nosplit +func (c *CPU) GetFaultAddr() (value uintptr) { + return c.faultAddr +} + +//go:nosplit +func (c *CPU) SetTtbr0Kvm(value uintptr) { + c.ttbr0Kvm = value +} + +//go:nosplit +func (c *CPU) SetTtbr0App(value uintptr) { + c.ttbr0App = value +} + +//go:nosplit +func (c *CPU) GetVector() (value Vector) { + return c.vecCode +} + +//go:nosplit +func (c *CPU) SetAppAddr(value uintptr) { + c.appAddr = value +} + +// SwitchArchOpts are embedded in SwitchOpts. +type SwitchArchOpts struct { + // UserASID indicates that the application ASID to be used on switch, + UserASID uint16 + + // KernelASID indicates that the kernel ASID to be used on return, + KernelASID uint16 +} + +func init() { +} + // Emit prints architecture-specific offsets. func Emit(w io.Writer) { fmt.Fprintf(w, "// Automatically generated, do not edit.\n") diff --git a/pkg/sentry/platform/ring0/entry_arm64.go b/pkg/sentry/platform/ring0/entry_arm64.go index 62a93f3d6..62a93f3d6 100644..100755 --- a/pkg/sentry/platform/ring0/entry_arm64.go +++ b/pkg/sentry/platform/ring0/entry_arm64.go diff --git a/pkg/sentry/platform/ring0/entry_amd64.s b/pkg/sentry/platform/ring0/entry_impl_amd64.s index 02df38331..daba45f9d 100644..100755 --- a/pkg/sentry/platform/ring0/entry_amd64.s +++ b/pkg/sentry/platform/ring0/entry_impl_amd64.s @@ -1,3 +1,67 @@ +// build +amd64 + +// Automatically generated, do not edit. + +// CPU offsets. +#define CPU_SELF 0x00 +#define CPU_REGISTERS 0x288 +#define CPU_STACK_TOP 0x110 +#define CPU_ERROR_CODE 0x110 +#define CPU_ERROR_TYPE 0x118 + +// Bits. +#define _RFLAGS_IF 0x200 +#define _KERNEL_FLAGS 0x02 + +// Vectors. +#define DivideByZero 0x00 +#define Debug 0x01 +#define NMI 0x02 +#define Breakpoint 0x03 +#define Overflow 0x04 +#define BoundRangeExceeded 0x05 +#define InvalidOpcode 0x06 +#define DeviceNotAvailable 0x07 +#define DoubleFault 0x08 +#define CoprocessorSegmentOverrun 0x09 +#define InvalidTSS 0x0a +#define SegmentNotPresent 0x0b +#define StackSegmentFault 0x0c +#define GeneralProtectionFault 0x0d +#define PageFault 0x0e +#define X87FloatingPointException 0x10 +#define AlignmentCheck 0x11 +#define MachineCheck 0x12 +#define SIMDFloatingPointException 0x13 +#define VirtualizationException 0x14 +#define SecurityException 0x1e +#define SyscallInt80 0x80 +#define Syscall 0x81 + +// Ptrace registers. +#define PTRACE_R15 0x00 +#define PTRACE_R14 0x08 +#define PTRACE_R13 0x10 +#define PTRACE_R12 0x18 +#define PTRACE_RBP 0x20 +#define PTRACE_RBX 0x28 +#define PTRACE_R11 0x30 +#define PTRACE_R10 0x38 +#define PTRACE_R9 0x40 +#define PTRACE_R8 0x48 +#define PTRACE_RAX 0x50 +#define PTRACE_RCX 0x58 +#define PTRACE_RDX 0x60 +#define PTRACE_RSI 0x68 +#define PTRACE_RDI 0x70 +#define PTRACE_ORIGRAX 0x78 +#define PTRACE_RIP 0x80 +#define PTRACE_CS 0x88 +#define PTRACE_FLAGS 0x90 +#define PTRACE_RSP 0x98 +#define PTRACE_SS 0xa0 +#define PTRACE_FS 0xa8 +#define PTRACE_GS 0xb0 // Copyright 2018 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_impl_arm64.s index db6465663..ea202219a 100644..100755 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_impl_arm64.s @@ -1,3 +1,67 @@ +// build +arm64 + +// Automatically generated, do not edit. + +// CPU offsets. +#define CPU_SELF 0x00 +#define CPU_REGISTERS 0x288 +#define CPU_STACK_TOP 0x110 +#define CPU_ERROR_CODE 0x110 +#define CPU_ERROR_TYPE 0x118 + +// Bits. +#define _RFLAGS_IF 0x200 +#define _KERNEL_FLAGS 0x02 + +// Vectors. +#define DivideByZero 0x00 +#define Debug 0x01 +#define NMI 0x02 +#define Breakpoint 0x03 +#define Overflow 0x04 +#define BoundRangeExceeded 0x05 +#define InvalidOpcode 0x06 +#define DeviceNotAvailable 0x07 +#define DoubleFault 0x08 +#define CoprocessorSegmentOverrun 0x09 +#define InvalidTSS 0x0a +#define SegmentNotPresent 0x0b +#define StackSegmentFault 0x0c +#define GeneralProtectionFault 0x0d +#define PageFault 0x0e +#define X87FloatingPointException 0x10 +#define AlignmentCheck 0x11 +#define MachineCheck 0x12 +#define SIMDFloatingPointException 0x13 +#define VirtualizationException 0x14 +#define SecurityException 0x1e +#define SyscallInt80 0x80 +#define Syscall 0x81 + +// Ptrace registers. +#define PTRACE_R15 0x00 +#define PTRACE_R14 0x08 +#define PTRACE_R13 0x10 +#define PTRACE_R12 0x18 +#define PTRACE_RBP 0x20 +#define PTRACE_RBX 0x28 +#define PTRACE_R11 0x30 +#define PTRACE_R10 0x38 +#define PTRACE_R9 0x40 +#define PTRACE_R8 0x48 +#define PTRACE_RAX 0x50 +#define PTRACE_RCX 0x58 +#define PTRACE_RDX 0x60 +#define PTRACE_RSI 0x68 +#define PTRACE_RDI 0x70 +#define PTRACE_ORIGRAX 0x78 +#define PTRACE_RIP 0x80 +#define PTRACE_CS 0x88 +#define PTRACE_FLAGS 0x90 +#define PTRACE_RSP 0x98 +#define PTRACE_SS 0xa0 +#define PTRACE_FS 0xa8 +#define PTRACE_GS 0xb0 // Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); diff --git a/pkg/sentry/platform/ring0/gen_offsets/BUILD b/pkg/sentry/platform/ring0/gen_offsets/BUILD deleted file mode 100644 index 4cae10459..000000000 --- a/pkg/sentry/platform/ring0/gen_offsets/BUILD +++ /dev/null @@ -1,33 +0,0 @@ -load("//tools:defs.bzl", "go_binary") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "defs_impl_arm64", - out = "defs_impl_arm64.go", - package = "main", - template = "//pkg/sentry/platform/ring0:defs_arm64", -) - -go_template_instance( - name = "defs_impl_amd64", - out = "defs_impl_amd64.go", - package = "main", - template = "//pkg/sentry/platform/ring0:defs_amd64", -) - -go_binary( - name = "gen_offsets", - srcs = [ - "defs_impl_amd64.go", - "defs_impl_arm64.go", - "main.go", - ], - visibility = ["//pkg/sentry/platform/ring0:__pkg__"], - deps = [ - "//pkg/cpuid", - "//pkg/sentry/platform/ring0/pagetables", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/platform/ring0/gen_offsets/main.go b/pkg/sentry/platform/ring0/gen_offsets/main.go deleted file mode 100644 index a4927da2f..000000000 --- a/pkg/sentry/platform/ring0/gen_offsets/main.go +++ /dev/null @@ -1,24 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Binary gen_offsets is a helper for generating offset headers. -package main - -import ( - "os" -) - -func main() { - Emit(os.Stdout) -} diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index ccacaea6b..ccacaea6b 100644..100755 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index af075aae4..af075aae4 100644..100755 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s index 0e6a6235b..0e6a6235b 100644..100755 --- a/pkg/sentry/platform/ring0/lib_arm64.s +++ b/pkg/sentry/platform/ring0/lib_arm64.s diff --git a/pkg/sentry/platform/ring0/offsets_amd64.go b/pkg/sentry/platform/ring0/offsets_amd64.go deleted file mode 100644 index 85cc3fdad..000000000 --- a/pkg/sentry/platform/ring0/offsets_amd64.go +++ /dev/null @@ -1,92 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 - -package ring0 - -import ( - "fmt" - "io" - "reflect" - "syscall" -) - -// Emit prints architecture-specific offsets. -func Emit(w io.Writer) { - fmt.Fprintf(w, "// Automatically generated, do not edit.\n") - - c := &CPU{} - fmt.Fprintf(w, "\n// CPU offsets.\n") - fmt.Fprintf(w, "#define CPU_SELF 0x%02x\n", reflect.ValueOf(&c.self).Pointer()-reflect.ValueOf(c).Pointer()) - fmt.Fprintf(w, "#define CPU_REGISTERS 0x%02x\n", reflect.ValueOf(&c.registers).Pointer()-reflect.ValueOf(c).Pointer()) - fmt.Fprintf(w, "#define CPU_STACK_TOP 0x%02x\n", reflect.ValueOf(&c.stack[0]).Pointer()-reflect.ValueOf(c).Pointer()+uintptr(len(c.stack))) - fmt.Fprintf(w, "#define CPU_ERROR_CODE 0x%02x\n", reflect.ValueOf(&c.errorCode).Pointer()-reflect.ValueOf(c).Pointer()) - fmt.Fprintf(w, "#define CPU_ERROR_TYPE 0x%02x\n", reflect.ValueOf(&c.errorType).Pointer()-reflect.ValueOf(c).Pointer()) - - fmt.Fprintf(w, "\n// Bits.\n") - fmt.Fprintf(w, "#define _RFLAGS_IF 0x%02x\n", _RFLAGS_IF) - fmt.Fprintf(w, "#define _KERNEL_FLAGS 0x%02x\n", KernelFlagsSet) - - fmt.Fprintf(w, "\n// Vectors.\n") - fmt.Fprintf(w, "#define DivideByZero 0x%02x\n", DivideByZero) - fmt.Fprintf(w, "#define Debug 0x%02x\n", Debug) - fmt.Fprintf(w, "#define NMI 0x%02x\n", NMI) - fmt.Fprintf(w, "#define Breakpoint 0x%02x\n", Breakpoint) - fmt.Fprintf(w, "#define Overflow 0x%02x\n", Overflow) - fmt.Fprintf(w, "#define BoundRangeExceeded 0x%02x\n", BoundRangeExceeded) - fmt.Fprintf(w, "#define InvalidOpcode 0x%02x\n", InvalidOpcode) - fmt.Fprintf(w, "#define DeviceNotAvailable 0x%02x\n", DeviceNotAvailable) - fmt.Fprintf(w, "#define DoubleFault 0x%02x\n", DoubleFault) - fmt.Fprintf(w, "#define CoprocessorSegmentOverrun 0x%02x\n", CoprocessorSegmentOverrun) - fmt.Fprintf(w, "#define InvalidTSS 0x%02x\n", InvalidTSS) - fmt.Fprintf(w, "#define SegmentNotPresent 0x%02x\n", SegmentNotPresent) - fmt.Fprintf(w, "#define StackSegmentFault 0x%02x\n", StackSegmentFault) - fmt.Fprintf(w, "#define GeneralProtectionFault 0x%02x\n", GeneralProtectionFault) - fmt.Fprintf(w, "#define PageFault 0x%02x\n", PageFault) - fmt.Fprintf(w, "#define X87FloatingPointException 0x%02x\n", X87FloatingPointException) - fmt.Fprintf(w, "#define AlignmentCheck 0x%02x\n", AlignmentCheck) - fmt.Fprintf(w, "#define MachineCheck 0x%02x\n", MachineCheck) - fmt.Fprintf(w, "#define SIMDFloatingPointException 0x%02x\n", SIMDFloatingPointException) - fmt.Fprintf(w, "#define VirtualizationException 0x%02x\n", VirtualizationException) - fmt.Fprintf(w, "#define SecurityException 0x%02x\n", SecurityException) - fmt.Fprintf(w, "#define SyscallInt80 0x%02x\n", SyscallInt80) - fmt.Fprintf(w, "#define Syscall 0x%02x\n", Syscall) - - p := &syscall.PtraceRegs{} - fmt.Fprintf(w, "\n// Ptrace registers.\n") - fmt.Fprintf(w, "#define PTRACE_R15 0x%02x\n", reflect.ValueOf(&p.R15).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_R14 0x%02x\n", reflect.ValueOf(&p.R14).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_R13 0x%02x\n", reflect.ValueOf(&p.R13).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_R12 0x%02x\n", reflect.ValueOf(&p.R12).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RBP 0x%02x\n", reflect.ValueOf(&p.Rbp).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RBX 0x%02x\n", reflect.ValueOf(&p.Rbx).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_R11 0x%02x\n", reflect.ValueOf(&p.R11).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_R10 0x%02x\n", reflect.ValueOf(&p.R10).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_R9 0x%02x\n", reflect.ValueOf(&p.R9).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_R8 0x%02x\n", reflect.ValueOf(&p.R8).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RAX 0x%02x\n", reflect.ValueOf(&p.Rax).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RCX 0x%02x\n", reflect.ValueOf(&p.Rcx).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RDX 0x%02x\n", reflect.ValueOf(&p.Rdx).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RSI 0x%02x\n", reflect.ValueOf(&p.Rsi).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RDI 0x%02x\n", reflect.ValueOf(&p.Rdi).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_ORIGRAX 0x%02x\n", reflect.ValueOf(&p.Orig_rax).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RIP 0x%02x\n", reflect.ValueOf(&p.Rip).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_CS 0x%02x\n", reflect.ValueOf(&p.Cs).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_FLAGS 0x%02x\n", reflect.ValueOf(&p.Eflags).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_RSP 0x%02x\n", reflect.ValueOf(&p.Rsp).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_SS 0x%02x\n", reflect.ValueOf(&p.Ss).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_FS 0x%02x\n", reflect.ValueOf(&p.Fs_base).Pointer()-reflect.ValueOf(p).Pointer()) - fmt.Fprintf(w, "#define PTRACE_GS 0x%02x\n", reflect.ValueOf(&p.Gs_base).Pointer()-reflect.ValueOf(p).Pointer()) -} diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD deleted file mode 100644 index 581841555..000000000 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ /dev/null @@ -1,112 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test", "select_arch") -load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance") - -package(licenses = ["notice"]) - -go_template( - name = "generic_walker", - srcs = select_arch( - amd64 = ["walker_amd64.go"], - arm64 = ["walker_arm64.go"], - ), - opt_types = [ - "Visitor", - ], - visibility = [":__pkg__"], -) - -go_template_instance( - name = "walker_map", - out = "walker_map.go", - package = "pagetables", - prefix = "map", - template = ":generic_walker", - types = { - "Visitor": "mapVisitor", - }, -) - -go_template_instance( - name = "walker_unmap", - out = "walker_unmap.go", - package = "pagetables", - prefix = "unmap", - template = ":generic_walker", - types = { - "Visitor": "unmapVisitor", - }, -) - -go_template_instance( - name = "walker_lookup", - out = "walker_lookup.go", - package = "pagetables", - prefix = "lookup", - template = ":generic_walker", - types = { - "Visitor": "lookupVisitor", - }, -) - -go_template_instance( - name = "walker_empty", - out = "walker_empty.go", - package = "pagetables", - prefix = "empty", - template = ":generic_walker", - types = { - "Visitor": "emptyVisitor", - }, -) - -go_template_instance( - name = "walker_check", - out = "walker_check.go", - package = "pagetables", - prefix = "check", - template = ":generic_walker", - types = { - "Visitor": "checkVisitor", - }, -) - -go_library( - name = "pagetables", - srcs = [ - "allocator.go", - "allocator_unsafe.go", - "pagetables.go", - "pagetables_aarch64.go", - "pagetables_amd64.go", - "pagetables_arm64.go", - "pagetables_x86.go", - "pcids.go", - "walker_amd64.go", - "walker_arm64.go", - "walker_empty.go", - "walker_lookup.go", - "walker_map.go", - "walker_unmap.go", - ], - visibility = [ - "//pkg/sentry/platform/kvm:__subpackages__", - "//pkg/sentry/platform/ring0:__subpackages__", - ], - deps = [ - "//pkg/sync", - "//pkg/usermem", - ], -) - -go_test( - name = "pagetables_test", - size = "small", - srcs = [ - "pagetables_amd64_test.go", - "pagetables_arm64_test.go", - "pagetables_test.go", - "walker_check.go", - ], - library = ":pagetables", - deps = ["//pkg/usermem"], -) diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go index 78510ebed..78510ebed 100644..100755 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64_state_autogen.go b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64_state_autogen.go new file mode 100755 index 000000000..ae9d2b272 --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build arm64 + +package pagetables diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_state_autogen.go b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_state_autogen.go new file mode 100755 index 000000000..f48a8acd1 --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build amd64 + +package pagetables diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go deleted file mode 100644 index 54e8e554f..000000000 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go +++ /dev/null @@ -1,75 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 - -package pagetables - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/usermem" -) - -func Test2MAnd4K(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map a small page and a huge page. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42) - pt.Map(0x00007f0000000000, pmdSize, MapOpts{AccessType: usermem.Read}, pmdSize*47) - - checkMappings(t, pt, []mapping{ - {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}}, - {0x00007f0000000000, pmdSize, pmdSize * 47, MapOpts{AccessType: usermem.Read}}, - }) -} - -func Test1GAnd4K(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map a small page and a super page. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42) - pt.Map(0x00007f0000000000, pudSize, MapOpts{AccessType: usermem.Read}, pudSize*47) - - checkMappings(t, pt, []mapping{ - {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}}, - {0x00007f0000000000, pudSize, pudSize * 47, MapOpts{AccessType: usermem.Read}}, - }) -} - -func TestSplit1GPage(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map a super page and knock out the middle. - pt.Map(0x00007f0000000000, pudSize, MapOpts{AccessType: usermem.Read}, pudSize*42) - pt.Unmap(usermem.Addr(0x00007f0000000000+pteSize), pudSize-(2*pteSize)) - - checkMappings(t, pt, []mapping{ - {0x00007f0000000000, pteSize, pudSize * 42, MapOpts{AccessType: usermem.Read}}, - {0x00007f0000000000 + pudSize - pteSize, pteSize, pudSize*42 + pudSize - pteSize, MapOpts{AccessType: usermem.Read}}, - }) -} - -func TestSplit2MPage(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map a huge page and knock out the middle. - pt.Map(0x00007f0000000000, pmdSize, MapOpts{AccessType: usermem.Read}, pmdSize*42) - pt.Unmap(usermem.Addr(0x00007f0000000000+pteSize), pmdSize-(2*pteSize)) - - checkMappings(t, pt, []mapping{ - {0x00007f0000000000, pteSize, pmdSize * 42, MapOpts{AccessType: usermem.Read}}, - {0x00007f0000000000 + pmdSize - pteSize, pteSize, pmdSize*42 + pmdSize - pteSize, MapOpts{AccessType: usermem.Read}}, - }) -} diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go index 1a49f12a2..1a49f12a2 100644..100755 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_state_autogen.go b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_state_autogen.go new file mode 100755 index 000000000..ae9d2b272 --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build arm64 + +package pagetables diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go deleted file mode 100644 index 2f73d424f..000000000 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go +++ /dev/null @@ -1,80 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build arm64 - -package pagetables - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/usermem" -) - -func Test2MAnd4K(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map a small page and a huge page. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite, User: true}, pteSize*42) - pt.Map(0x0000ff0000000000, pmdSize, MapOpts{AccessType: usermem.Read, User: true}, pmdSize*47) - - pt.Map(0xffff000000400000, pteSize, MapOpts{AccessType: usermem.ReadWrite, User: false}, pteSize*42) - pt.Map(0xffffff0000000000, pmdSize, MapOpts{AccessType: usermem.Read, User: false}, pmdSize*47) - - checkMappings(t, pt, []mapping{ - {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite, User: true}}, - {0x0000ff0000000000, pmdSize, pmdSize * 47, MapOpts{AccessType: usermem.Read, User: true}}, - {0xffff000000400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite, User: false}}, - {0xffffff0000000000, pmdSize, pmdSize * 47, MapOpts{AccessType: usermem.Read, User: false}}, - }) -} - -func Test1GAnd4K(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map a small page and a super page. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite, User: true}, pteSize*42) - pt.Map(0x0000ff0000000000, pudSize, MapOpts{AccessType: usermem.Read, User: true}, pudSize*47) - - checkMappings(t, pt, []mapping{ - {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite, User: true}}, - {0x0000ff0000000000, pudSize, pudSize * 47, MapOpts{AccessType: usermem.Read, User: true}}, - }) -} - -func TestSplit1GPage(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map a super page and knock out the middle. - pt.Map(0x0000ff0000000000, pudSize, MapOpts{AccessType: usermem.Read, User: true}, pudSize*42) - pt.Unmap(usermem.Addr(0x0000ff0000000000+pteSize), pudSize-(2*pteSize)) - - checkMappings(t, pt, []mapping{ - {0x0000ff0000000000, pteSize, pudSize * 42, MapOpts{AccessType: usermem.Read, User: true}}, - {0x0000ff0000000000 + pudSize - pteSize, pteSize, pudSize*42 + pudSize - pteSize, MapOpts{AccessType: usermem.Read, User: true}}, - }) -} - -func TestSplit2MPage(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map a huge page and knock out the middle. - pt.Map(0x0000ff0000000000, pmdSize, MapOpts{AccessType: usermem.Read, User: true}, pmdSize*42) - pt.Unmap(usermem.Addr(0x0000ff0000000000+pteSize), pmdSize-(2*pteSize)) - - checkMappings(t, pt, []mapping{ - {0x0000ff0000000000, pteSize, pmdSize * 42, MapOpts{AccessType: usermem.Read, User: true}}, - {0x0000ff0000000000 + pmdSize - pteSize, pteSize, pmdSize*42 + pmdSize - pteSize, MapOpts{AccessType: usermem.Read, User: true}}, - }) -} diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_state_autogen.go b/pkg/sentry/platform/ring0/pagetables/pagetables_state_autogen.go new file mode 100755 index 000000000..52bab66fe --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build i386 amd64 + +package pagetables diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_test.go deleted file mode 100644 index 5c88d087d..000000000 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_test.go +++ /dev/null @@ -1,156 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package pagetables - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/usermem" -) - -type mapping struct { - start uintptr - length uintptr - addr uintptr - opts MapOpts -} - -type checkVisitor struct { - expected []mapping // Input. - current int // Temporary. - found []mapping // Output. - failed string // Output. -} - -func (v *checkVisitor) visit(start uintptr, pte *PTE, align uintptr) { - v.found = append(v.found, mapping{ - start: start, - length: align + 1, - addr: pte.Address(), - opts: pte.Opts(), - }) - if v.failed != "" { - // Don't keep looking for errors. - return - } - - if v.current >= len(v.expected) { - v.failed = "more mappings than expected" - } else if v.expected[v.current].start != start { - v.failed = "start didn't match expected" - } else if v.expected[v.current].length != (align + 1) { - v.failed = "end didn't match expected" - } else if v.expected[v.current].addr != pte.Address() { - v.failed = "address didn't match expected" - } else if v.expected[v.current].opts != pte.Opts() { - v.failed = "opts didn't match" - } - v.current++ -} - -func (*checkVisitor) requiresAlloc() bool { return false } - -func (*checkVisitor) requiresSplit() bool { return false } - -func checkMappings(t *testing.T, pt *PageTables, m []mapping) { - // Iterate over all the mappings. - w := checkWalker{ - pageTables: pt, - visitor: checkVisitor{ - expected: m, - }, - } - w.iterateRange(0, ^uintptr(0)) - - // Were we expected additional mappings? - if w.visitor.failed == "" && w.visitor.current != len(w.visitor.expected) { - w.visitor.failed = "insufficient mappings found" - } - - // Emit a meaningful error message on failure. - if w.visitor.failed != "" { - t.Errorf("%s; got %#v, wanted %#v", w.visitor.failed, w.visitor.found, w.visitor.expected) - } -} - -func TestUnmap(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map and unmap one entry. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42) - pt.Unmap(0x400000, pteSize) - - checkMappings(t, pt, nil) -} - -func TestReadOnly(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map one entry. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.Read}, pteSize*42) - - checkMappings(t, pt, []mapping{ - {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.Read}}, - }) -} - -func TestReadWrite(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map one entry. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42) - - checkMappings(t, pt, []mapping{ - {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}}, - }) -} - -func TestSerialEntries(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map two sequential entries. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42) - pt.Map(0x401000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*47) - - checkMappings(t, pt, []mapping{ - {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}}, - {0x401000, pteSize, pteSize * 47, MapOpts{AccessType: usermem.ReadWrite}}, - }) -} - -func TestSpanningEntries(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Span a pgd with two pages. - pt.Map(0x00007efffffff000, 2*pteSize, MapOpts{AccessType: usermem.Read}, pteSize*42) - - checkMappings(t, pt, []mapping{ - {0x00007efffffff000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.Read}}, - {0x00007f0000000000, pteSize, pteSize * 43, MapOpts{AccessType: usermem.Read}}, - }) -} - -func TestSparseEntries(t *testing.T) { - pt := New(NewRuntimeAllocator()) - - // Map two entries in different pgds. - pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42) - pt.Map(0x00007f0000000000, pteSize, MapOpts{AccessType: usermem.Read}, pteSize*47) - - checkMappings(t, pt, []mapping{ - {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}}, - {0x00007f0000000000, pteSize, pteSize * 47, MapOpts{AccessType: usermem.Read}}, - }) -} diff --git a/pkg/sentry/platform/ring0/pagetables/pcids.go b/pkg/sentry/platform/ring0/pagetables/pcids.go index 9206030bf..9206030bf 100644..100755 --- a/pkg/sentry/platform/ring0/pagetables/pcids.go +++ b/pkg/sentry/platform/ring0/pagetables/pcids.go diff --git a/pkg/sentry/platform/ring0/pagetables/walker_amd64.go b/pkg/sentry/platform/ring0/pagetables/walker_amd64.go index 8f9dacd93..8f9dacd93 100644..100755 --- a/pkg/sentry/platform/ring0/pagetables/walker_amd64.go +++ b/pkg/sentry/platform/ring0/pagetables/walker_amd64.go diff --git a/pkg/sentry/platform/ring0/pagetables/walker_arm64.go b/pkg/sentry/platform/ring0/pagetables/walker_arm64.go index c261d393a..c261d393a 100644..100755 --- a/pkg/sentry/platform/ring0/pagetables/walker_arm64.go +++ b/pkg/sentry/platform/ring0/pagetables/walker_arm64.go diff --git a/pkg/sentry/platform/ring0/pagetables/walker_empty.go b/pkg/sentry/platform/ring0/pagetables/walker_empty.go new file mode 100755 index 000000000..417784e17 --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/walker_empty.go @@ -0,0 +1,255 @@ +package pagetables + +// Walker walks page tables. +type emptyWalker struct { + // pageTables are the tables to walk. + pageTables *PageTables + + // Visitor is the set of arguments. + visitor emptyVisitor +} + +// iterateRange iterates over all appropriate levels of page tables for the given range. +// +// If requiresAlloc is true, then Set _must_ be called on all given PTEs. The +// exception is super pages. If a valid super page (huge or jumbo) cannot be +// installed, then the walk will continue to individual entries. +// +// This algorithm will attempt to maximize the use of super pages whenever +// possible. Whether a super page is provided will be clear through the range +// provided in the callback. +// +// Note that if requiresAlloc is true, then no gaps will be present. However, +// if alloc is not set, then the iteration will likely be full of gaps. +// +// Note that this function should generally be avoided in favor of Map, Unmap, +// etc. when not necessary. +// +// Precondition: start must be page-aligned. +// +// Precondition: start must be less than end. +// +// Precondition: If requiresAlloc is true, then start and end should not span +// non-canonical ranges. If they do, a panic will result. +// +//go:nosplit +func (w *emptyWalker) iterateRange(start, end uintptr) { + if start%pteSize != 0 { + panic("unaligned start") + } + if end < start { + panic("start > end") + } + if start < lowerTop { + if end <= lowerTop { + w.iterateRangeCanonical(start, end) + } else if end > lowerTop && end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + w.iterateRangeCanonical(upperBottom, end) + } + } else if start < upperBottom { + if end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(upperBottom, end) + } + } else { + w.iterateRangeCanonical(start, end) + } +} + +// next returns the next address quantized by the given size. +// +//go:nosplit +func emptynext(start uintptr, size uintptr) uintptr { + start &= ^(size - 1) + start += size + return start +} + +// iterateRangeCanonical walks a canonical range. +// +//go:nosplit +func (w *emptyWalker) iterateRangeCanonical(start, end uintptr) { + for pgdIndex := uint16((start & pgdMask) >> pgdShift); start < end && pgdIndex < entriesPerPage; pgdIndex++ { + var ( + pgdEntry = &w.pageTables.root[pgdIndex] + pudEntries *PTEs + ) + if !pgdEntry.Valid() { + if !w.visitor.requiresAlloc() { + + start = emptynext(start, pgdSize) + continue + } + + pudEntries = w.pageTables.Allocator.NewPTEs() + pgdEntry.setPageTable(w.pageTables, pudEntries) + } else { + pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address()) + } + + clearPUDEntries := uint16(0) + + for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ { + var ( + pudEntry = &pudEntries[pudIndex] + pmdEntries *PTEs + ) + if !pudEntry.Valid() { + if !w.visitor.requiresAlloc() { + + clearPUDEntries++ + start = emptynext(start, pudSize) + continue + } + + if start&(pudSize-1) == 0 && end-start >= pudSize { + pudEntry.SetSuper() + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + if pudEntry.Valid() { + start = emptynext(start, pudSize) + continue + } + } + + pmdEntries = w.pageTables.Allocator.NewPTEs() + pudEntry.setPageTable(w.pageTables, pmdEntries) + + } else if pudEntry.IsSuper() { + + if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < emptynext(start, pudSize)) { + + pmdEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pmdEntries[index].SetSuper() + pmdEntries[index].Set( + pudEntry.Address()+(pmdSize*uintptr(index)), + pudEntry.Opts()) + } + pudEntry.setPageTable(w.pageTables, pmdEntries) + } else { + + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + + if !pudEntry.Valid() { + clearPUDEntries++ + } + + start = emptynext(start, pudSize) + continue + } + } else { + pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address()) + } + + clearPMDEntries := uint16(0) + + for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ { + var ( + pmdEntry = &pmdEntries[pmdIndex] + pteEntries *PTEs + ) + if !pmdEntry.Valid() { + if !w.visitor.requiresAlloc() { + + clearPMDEntries++ + start = emptynext(start, pmdSize) + continue + } + + if start&(pmdSize-1) == 0 && end-start >= pmdSize { + pmdEntry.SetSuper() + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + if pmdEntry.Valid() { + start = emptynext(start, pmdSize) + continue + } + } + + pteEntries = w.pageTables.Allocator.NewPTEs() + pmdEntry.setPageTable(w.pageTables, pteEntries) + + } else if pmdEntry.IsSuper() { + + if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < emptynext(start, pmdSize)) { + + pteEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pteEntries[index].Set( + pmdEntry.Address()+(pteSize*uintptr(index)), + pmdEntry.Opts()) + } + pmdEntry.setPageTable(w.pageTables, pteEntries) + } else { + + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + + if !pmdEntry.Valid() { + clearPMDEntries++ + } + + start = emptynext(start, pmdSize) + continue + } + } else { + pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address()) + } + + clearPTEEntries := uint16(0) + + for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ { + var ( + pteEntry = &pteEntries[pteIndex] + ) + if !pteEntry.Valid() && !w.visitor.requiresAlloc() { + clearPTEEntries++ + start += pteSize + continue + } + + w.visitor.visit(uintptr(start), pteEntry, pteSize-1) + if !pteEntry.Valid() { + if w.visitor.requiresAlloc() { + panic("PTE not set after iteration with requiresAlloc!") + } + clearPTEEntries++ + } + + start += pteSize + continue + } + + if clearPTEEntries == entriesPerPage { + pmdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pteEntries) + clearPMDEntries++ + } + } + + if clearPMDEntries == entriesPerPage { + pudEntry.Clear() + w.pageTables.Allocator.FreePTEs(pmdEntries) + clearPUDEntries++ + } + } + + if clearPUDEntries == entriesPerPage { + pgdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pudEntries) + } + } +} diff --git a/pkg/sentry/platform/ring0/pagetables/walker_lookup.go b/pkg/sentry/platform/ring0/pagetables/walker_lookup.go new file mode 100755 index 000000000..906c9c50f --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/walker_lookup.go @@ -0,0 +1,255 @@ +package pagetables + +// Walker walks page tables. +type lookupWalker struct { + // pageTables are the tables to walk. + pageTables *PageTables + + // Visitor is the set of arguments. + visitor lookupVisitor +} + +// iterateRange iterates over all appropriate levels of page tables for the given range. +// +// If requiresAlloc is true, then Set _must_ be called on all given PTEs. The +// exception is super pages. If a valid super page (huge or jumbo) cannot be +// installed, then the walk will continue to individual entries. +// +// This algorithm will attempt to maximize the use of super pages whenever +// possible. Whether a super page is provided will be clear through the range +// provided in the callback. +// +// Note that if requiresAlloc is true, then no gaps will be present. However, +// if alloc is not set, then the iteration will likely be full of gaps. +// +// Note that this function should generally be avoided in favor of Map, Unmap, +// etc. when not necessary. +// +// Precondition: start must be page-aligned. +// +// Precondition: start must be less than end. +// +// Precondition: If requiresAlloc is true, then start and end should not span +// non-canonical ranges. If they do, a panic will result. +// +//go:nosplit +func (w *lookupWalker) iterateRange(start, end uintptr) { + if start%pteSize != 0 { + panic("unaligned start") + } + if end < start { + panic("start > end") + } + if start < lowerTop { + if end <= lowerTop { + w.iterateRangeCanonical(start, end) + } else if end > lowerTop && end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + w.iterateRangeCanonical(upperBottom, end) + } + } else if start < upperBottom { + if end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(upperBottom, end) + } + } else { + w.iterateRangeCanonical(start, end) + } +} + +// next returns the next address quantized by the given size. +// +//go:nosplit +func lookupnext(start uintptr, size uintptr) uintptr { + start &= ^(size - 1) + start += size + return start +} + +// iterateRangeCanonical walks a canonical range. +// +//go:nosplit +func (w *lookupWalker) iterateRangeCanonical(start, end uintptr) { + for pgdIndex := uint16((start & pgdMask) >> pgdShift); start < end && pgdIndex < entriesPerPage; pgdIndex++ { + var ( + pgdEntry = &w.pageTables.root[pgdIndex] + pudEntries *PTEs + ) + if !pgdEntry.Valid() { + if !w.visitor.requiresAlloc() { + + start = lookupnext(start, pgdSize) + continue + } + + pudEntries = w.pageTables.Allocator.NewPTEs() + pgdEntry.setPageTable(w.pageTables, pudEntries) + } else { + pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address()) + } + + clearPUDEntries := uint16(0) + + for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ { + var ( + pudEntry = &pudEntries[pudIndex] + pmdEntries *PTEs + ) + if !pudEntry.Valid() { + if !w.visitor.requiresAlloc() { + + clearPUDEntries++ + start = lookupnext(start, pudSize) + continue + } + + if start&(pudSize-1) == 0 && end-start >= pudSize { + pudEntry.SetSuper() + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + if pudEntry.Valid() { + start = lookupnext(start, pudSize) + continue + } + } + + pmdEntries = w.pageTables.Allocator.NewPTEs() + pudEntry.setPageTable(w.pageTables, pmdEntries) + + } else if pudEntry.IsSuper() { + + if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < lookupnext(start, pudSize)) { + + pmdEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pmdEntries[index].SetSuper() + pmdEntries[index].Set( + pudEntry.Address()+(pmdSize*uintptr(index)), + pudEntry.Opts()) + } + pudEntry.setPageTable(w.pageTables, pmdEntries) + } else { + + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + + if !pudEntry.Valid() { + clearPUDEntries++ + } + + start = lookupnext(start, pudSize) + continue + } + } else { + pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address()) + } + + clearPMDEntries := uint16(0) + + for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ { + var ( + pmdEntry = &pmdEntries[pmdIndex] + pteEntries *PTEs + ) + if !pmdEntry.Valid() { + if !w.visitor.requiresAlloc() { + + clearPMDEntries++ + start = lookupnext(start, pmdSize) + continue + } + + if start&(pmdSize-1) == 0 && end-start >= pmdSize { + pmdEntry.SetSuper() + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + if pmdEntry.Valid() { + start = lookupnext(start, pmdSize) + continue + } + } + + pteEntries = w.pageTables.Allocator.NewPTEs() + pmdEntry.setPageTable(w.pageTables, pteEntries) + + } else if pmdEntry.IsSuper() { + + if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < lookupnext(start, pmdSize)) { + + pteEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pteEntries[index].Set( + pmdEntry.Address()+(pteSize*uintptr(index)), + pmdEntry.Opts()) + } + pmdEntry.setPageTable(w.pageTables, pteEntries) + } else { + + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + + if !pmdEntry.Valid() { + clearPMDEntries++ + } + + start = lookupnext(start, pmdSize) + continue + } + } else { + pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address()) + } + + clearPTEEntries := uint16(0) + + for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ { + var ( + pteEntry = &pteEntries[pteIndex] + ) + if !pteEntry.Valid() && !w.visitor.requiresAlloc() { + clearPTEEntries++ + start += pteSize + continue + } + + w.visitor.visit(uintptr(start), pteEntry, pteSize-1) + if !pteEntry.Valid() { + if w.visitor.requiresAlloc() { + panic("PTE not set after iteration with requiresAlloc!") + } + clearPTEEntries++ + } + + start += pteSize + continue + } + + if clearPTEEntries == entriesPerPage { + pmdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pteEntries) + clearPMDEntries++ + } + } + + if clearPMDEntries == entriesPerPage { + pudEntry.Clear() + w.pageTables.Allocator.FreePTEs(pmdEntries) + clearPUDEntries++ + } + } + + if clearPUDEntries == entriesPerPage { + pgdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pudEntries) + } + } +} diff --git a/pkg/sentry/platform/ring0/pagetables/walker_map.go b/pkg/sentry/platform/ring0/pagetables/walker_map.go new file mode 100755 index 000000000..61ee3c825 --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/walker_map.go @@ -0,0 +1,255 @@ +package pagetables + +// Walker walks page tables. +type mapWalker struct { + // pageTables are the tables to walk. + pageTables *PageTables + + // Visitor is the set of arguments. + visitor mapVisitor +} + +// iterateRange iterates over all appropriate levels of page tables for the given range. +// +// If requiresAlloc is true, then Set _must_ be called on all given PTEs. The +// exception is super pages. If a valid super page (huge or jumbo) cannot be +// installed, then the walk will continue to individual entries. +// +// This algorithm will attempt to maximize the use of super pages whenever +// possible. Whether a super page is provided will be clear through the range +// provided in the callback. +// +// Note that if requiresAlloc is true, then no gaps will be present. However, +// if alloc is not set, then the iteration will likely be full of gaps. +// +// Note that this function should generally be avoided in favor of Map, Unmap, +// etc. when not necessary. +// +// Precondition: start must be page-aligned. +// +// Precondition: start must be less than end. +// +// Precondition: If requiresAlloc is true, then start and end should not span +// non-canonical ranges. If they do, a panic will result. +// +//go:nosplit +func (w *mapWalker) iterateRange(start, end uintptr) { + if start%pteSize != 0 { + panic("unaligned start") + } + if end < start { + panic("start > end") + } + if start < lowerTop { + if end <= lowerTop { + w.iterateRangeCanonical(start, end) + } else if end > lowerTop && end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + w.iterateRangeCanonical(upperBottom, end) + } + } else if start < upperBottom { + if end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(upperBottom, end) + } + } else { + w.iterateRangeCanonical(start, end) + } +} + +// next returns the next address quantized by the given size. +// +//go:nosplit +func mapnext(start uintptr, size uintptr) uintptr { + start &= ^(size - 1) + start += size + return start +} + +// iterateRangeCanonical walks a canonical range. +// +//go:nosplit +func (w *mapWalker) iterateRangeCanonical(start, end uintptr) { + for pgdIndex := uint16((start & pgdMask) >> pgdShift); start < end && pgdIndex < entriesPerPage; pgdIndex++ { + var ( + pgdEntry = &w.pageTables.root[pgdIndex] + pudEntries *PTEs + ) + if !pgdEntry.Valid() { + if !w.visitor.requiresAlloc() { + + start = mapnext(start, pgdSize) + continue + } + + pudEntries = w.pageTables.Allocator.NewPTEs() + pgdEntry.setPageTable(w.pageTables, pudEntries) + } else { + pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address()) + } + + clearPUDEntries := uint16(0) + + for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ { + var ( + pudEntry = &pudEntries[pudIndex] + pmdEntries *PTEs + ) + if !pudEntry.Valid() { + if !w.visitor.requiresAlloc() { + + clearPUDEntries++ + start = mapnext(start, pudSize) + continue + } + + if start&(pudSize-1) == 0 && end-start >= pudSize { + pudEntry.SetSuper() + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + if pudEntry.Valid() { + start = mapnext(start, pudSize) + continue + } + } + + pmdEntries = w.pageTables.Allocator.NewPTEs() + pudEntry.setPageTable(w.pageTables, pmdEntries) + + } else if pudEntry.IsSuper() { + + if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < mapnext(start, pudSize)) { + + pmdEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pmdEntries[index].SetSuper() + pmdEntries[index].Set( + pudEntry.Address()+(pmdSize*uintptr(index)), + pudEntry.Opts()) + } + pudEntry.setPageTable(w.pageTables, pmdEntries) + } else { + + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + + if !pudEntry.Valid() { + clearPUDEntries++ + } + + start = mapnext(start, pudSize) + continue + } + } else { + pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address()) + } + + clearPMDEntries := uint16(0) + + for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ { + var ( + pmdEntry = &pmdEntries[pmdIndex] + pteEntries *PTEs + ) + if !pmdEntry.Valid() { + if !w.visitor.requiresAlloc() { + + clearPMDEntries++ + start = mapnext(start, pmdSize) + continue + } + + if start&(pmdSize-1) == 0 && end-start >= pmdSize { + pmdEntry.SetSuper() + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + if pmdEntry.Valid() { + start = mapnext(start, pmdSize) + continue + } + } + + pteEntries = w.pageTables.Allocator.NewPTEs() + pmdEntry.setPageTable(w.pageTables, pteEntries) + + } else if pmdEntry.IsSuper() { + + if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < mapnext(start, pmdSize)) { + + pteEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pteEntries[index].Set( + pmdEntry.Address()+(pteSize*uintptr(index)), + pmdEntry.Opts()) + } + pmdEntry.setPageTable(w.pageTables, pteEntries) + } else { + + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + + if !pmdEntry.Valid() { + clearPMDEntries++ + } + + start = mapnext(start, pmdSize) + continue + } + } else { + pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address()) + } + + clearPTEEntries := uint16(0) + + for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ { + var ( + pteEntry = &pteEntries[pteIndex] + ) + if !pteEntry.Valid() && !w.visitor.requiresAlloc() { + clearPTEEntries++ + start += pteSize + continue + } + + w.visitor.visit(uintptr(start), pteEntry, pteSize-1) + if !pteEntry.Valid() { + if w.visitor.requiresAlloc() { + panic("PTE not set after iteration with requiresAlloc!") + } + clearPTEEntries++ + } + + start += pteSize + continue + } + + if clearPTEEntries == entriesPerPage { + pmdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pteEntries) + clearPMDEntries++ + } + } + + if clearPMDEntries == entriesPerPage { + pudEntry.Clear() + w.pageTables.Allocator.FreePTEs(pmdEntries) + clearPUDEntries++ + } + } + + if clearPUDEntries == entriesPerPage { + pgdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pudEntries) + } + } +} diff --git a/pkg/sentry/platform/ring0/pagetables/walker_unmap.go b/pkg/sentry/platform/ring0/pagetables/walker_unmap.go new file mode 100755 index 000000000..be2aa0ce4 --- /dev/null +++ b/pkg/sentry/platform/ring0/pagetables/walker_unmap.go @@ -0,0 +1,255 @@ +package pagetables + +// Walker walks page tables. +type unmapWalker struct { + // pageTables are the tables to walk. + pageTables *PageTables + + // Visitor is the set of arguments. + visitor unmapVisitor +} + +// iterateRange iterates over all appropriate levels of page tables for the given range. +// +// If requiresAlloc is true, then Set _must_ be called on all given PTEs. The +// exception is super pages. If a valid super page (huge or jumbo) cannot be +// installed, then the walk will continue to individual entries. +// +// This algorithm will attempt to maximize the use of super pages whenever +// possible. Whether a super page is provided will be clear through the range +// provided in the callback. +// +// Note that if requiresAlloc is true, then no gaps will be present. However, +// if alloc is not set, then the iteration will likely be full of gaps. +// +// Note that this function should generally be avoided in favor of Map, Unmap, +// etc. when not necessary. +// +// Precondition: start must be page-aligned. +// +// Precondition: start must be less than end. +// +// Precondition: If requiresAlloc is true, then start and end should not span +// non-canonical ranges. If they do, a panic will result. +// +//go:nosplit +func (w *unmapWalker) iterateRange(start, end uintptr) { + if start%pteSize != 0 { + panic("unaligned start") + } + if end < start { + panic("start > end") + } + if start < lowerTop { + if end <= lowerTop { + w.iterateRangeCanonical(start, end) + } else if end > lowerTop && end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(start, lowerTop) + w.iterateRangeCanonical(upperBottom, end) + } + } else if start < upperBottom { + if end <= upperBottom { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + } else { + if w.visitor.requiresAlloc() { + panic("alloc spans non-canonical range") + } + w.iterateRangeCanonical(upperBottom, end) + } + } else { + w.iterateRangeCanonical(start, end) + } +} + +// next returns the next address quantized by the given size. +// +//go:nosplit +func unmapnext(start uintptr, size uintptr) uintptr { + start &= ^(size - 1) + start += size + return start +} + +// iterateRangeCanonical walks a canonical range. +// +//go:nosplit +func (w *unmapWalker) iterateRangeCanonical(start, end uintptr) { + for pgdIndex := uint16((start & pgdMask) >> pgdShift); start < end && pgdIndex < entriesPerPage; pgdIndex++ { + var ( + pgdEntry = &w.pageTables.root[pgdIndex] + pudEntries *PTEs + ) + if !pgdEntry.Valid() { + if !w.visitor.requiresAlloc() { + + start = unmapnext(start, pgdSize) + continue + } + + pudEntries = w.pageTables.Allocator.NewPTEs() + pgdEntry.setPageTable(w.pageTables, pudEntries) + } else { + pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address()) + } + + clearPUDEntries := uint16(0) + + for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ { + var ( + pudEntry = &pudEntries[pudIndex] + pmdEntries *PTEs + ) + if !pudEntry.Valid() { + if !w.visitor.requiresAlloc() { + + clearPUDEntries++ + start = unmapnext(start, pudSize) + continue + } + + if start&(pudSize-1) == 0 && end-start >= pudSize { + pudEntry.SetSuper() + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + if pudEntry.Valid() { + start = unmapnext(start, pudSize) + continue + } + } + + pmdEntries = w.pageTables.Allocator.NewPTEs() + pudEntry.setPageTable(w.pageTables, pmdEntries) + + } else if pudEntry.IsSuper() { + + if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < unmapnext(start, pudSize)) { + + pmdEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pmdEntries[index].SetSuper() + pmdEntries[index].Set( + pudEntry.Address()+(pmdSize*uintptr(index)), + pudEntry.Opts()) + } + pudEntry.setPageTable(w.pageTables, pmdEntries) + } else { + + w.visitor.visit(uintptr(start), pudEntry, pudSize-1) + + if !pudEntry.Valid() { + clearPUDEntries++ + } + + start = unmapnext(start, pudSize) + continue + } + } else { + pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address()) + } + + clearPMDEntries := uint16(0) + + for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ { + var ( + pmdEntry = &pmdEntries[pmdIndex] + pteEntries *PTEs + ) + if !pmdEntry.Valid() { + if !w.visitor.requiresAlloc() { + + clearPMDEntries++ + start = unmapnext(start, pmdSize) + continue + } + + if start&(pmdSize-1) == 0 && end-start >= pmdSize { + pmdEntry.SetSuper() + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + if pmdEntry.Valid() { + start = unmapnext(start, pmdSize) + continue + } + } + + pteEntries = w.pageTables.Allocator.NewPTEs() + pmdEntry.setPageTable(w.pageTables, pteEntries) + + } else if pmdEntry.IsSuper() { + + if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < unmapnext(start, pmdSize)) { + + pteEntries = w.pageTables.Allocator.NewPTEs() + for index := uint16(0); index < entriesPerPage; index++ { + pteEntries[index].Set( + pmdEntry.Address()+(pteSize*uintptr(index)), + pmdEntry.Opts()) + } + pmdEntry.setPageTable(w.pageTables, pteEntries) + } else { + + w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1) + + if !pmdEntry.Valid() { + clearPMDEntries++ + } + + start = unmapnext(start, pmdSize) + continue + } + } else { + pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address()) + } + + clearPTEEntries := uint16(0) + + for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ { + var ( + pteEntry = &pteEntries[pteIndex] + ) + if !pteEntry.Valid() && !w.visitor.requiresAlloc() { + clearPTEEntries++ + start += pteSize + continue + } + + w.visitor.visit(uintptr(start), pteEntry, pteSize-1) + if !pteEntry.Valid() { + if w.visitor.requiresAlloc() { + panic("PTE not set after iteration with requiresAlloc!") + } + clearPTEEntries++ + } + + start += pteSize + continue + } + + if clearPTEEntries == entriesPerPage { + pmdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pteEntries) + clearPMDEntries++ + } + } + + if clearPMDEntries == entriesPerPage { + pudEntry.Clear() + w.pageTables.Allocator.FreePTEs(pmdEntries) + clearPUDEntries++ + } + } + + if clearPUDEntries == entriesPerPage { + pgdEntry.Clear() + w.pageTables.Allocator.FreePTEs(pudEntries) + } + } +} diff --git a/pkg/sentry/platform/ring0/ring0_amd64_state_autogen.go b/pkg/sentry/platform/ring0/ring0_amd64_state_autogen.go new file mode 100755 index 000000000..96cf5d331 --- /dev/null +++ b/pkg/sentry/platform/ring0/ring0_amd64_state_autogen.go @@ -0,0 +1,7 @@ +// automatically generated by stateify. + +// +build amd64 +// +build amd64 +// +build amd64 + +package ring0 diff --git a/pkg/sentry/platform/ring0/ring0_arm64_state_autogen.go b/pkg/sentry/platform/ring0/ring0_arm64_state_autogen.go new file mode 100755 index 000000000..7f2ab3537 --- /dev/null +++ b/pkg/sentry/platform/ring0/ring0_arm64_state_autogen.go @@ -0,0 +1,7 @@ +// automatically generated by stateify. + +// +build arm64 +// +build arm64 +// +build arm64 + +package ring0 diff --git a/pkg/sentry/platform/ring0/ring0_state_autogen.go b/pkg/sentry/platform/ring0/ring0_state_autogen.go new file mode 100755 index 000000000..327aba163 --- /dev/null +++ b/pkg/sentry/platform/ring0/ring0_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package ring0 diff --git a/pkg/sentry/platform/ring0/x86.go b/pkg/sentry/platform/ring0/x86.go deleted file mode 100644 index 5f80d64e8..000000000 --- a/pkg/sentry/platform/ring0/x86.go +++ /dev/null @@ -1,264 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build i386 amd64 - -package ring0 - -import ( - "gvisor.dev/gvisor/pkg/cpuid" -) - -// Useful bits. -const ( - _CR0_PE = 1 << 0 - _CR0_ET = 1 << 4 - _CR0_AM = 1 << 18 - _CR0_PG = 1 << 31 - - _CR4_PSE = 1 << 4 - _CR4_PAE = 1 << 5 - _CR4_PGE = 1 << 7 - _CR4_OSFXSR = 1 << 9 - _CR4_OSXMMEXCPT = 1 << 10 - _CR4_FSGSBASE = 1 << 16 - _CR4_PCIDE = 1 << 17 - _CR4_OSXSAVE = 1 << 18 - _CR4_SMEP = 1 << 20 - - _RFLAGS_AC = 1 << 18 - _RFLAGS_NT = 1 << 14 - _RFLAGS_IOPL = 3 << 12 - _RFLAGS_DF = 1 << 10 - _RFLAGS_IF = 1 << 9 - _RFLAGS_STEP = 1 << 8 - _RFLAGS_RESERVED = 1 << 1 - - _EFER_SCE = 0x001 - _EFER_LME = 0x100 - _EFER_LMA = 0x400 - _EFER_NX = 0x800 - - _MSR_STAR = 0xc0000081 - _MSR_LSTAR = 0xc0000082 - _MSR_CSTAR = 0xc0000083 - _MSR_SYSCALL_MASK = 0xc0000084 - _MSR_PLATFORM_INFO = 0xce - _MSR_MISC_FEATURES = 0x140 - - _PLATFORM_INFO_CPUID_FAULT = 1 << 31 - - _MISC_FEATURE_CPUID_TRAP = 0x1 -) - -const ( - // KernelFlagsSet should always be set in the kernel. - KernelFlagsSet = _RFLAGS_RESERVED - - // UserFlagsSet are always set in userspace. - UserFlagsSet = _RFLAGS_RESERVED | _RFLAGS_IF - - // KernelFlagsClear should always be clear in the kernel. - KernelFlagsClear = _RFLAGS_STEP | _RFLAGS_IF | _RFLAGS_IOPL | _RFLAGS_AC | _RFLAGS_NT - - // UserFlagsClear are always cleared in userspace. - UserFlagsClear = _RFLAGS_NT | _RFLAGS_IOPL -) - -// Vector is an exception vector. -type Vector uintptr - -// Exception vectors. -const ( - DivideByZero Vector = iota - Debug - NMI - Breakpoint - Overflow - BoundRangeExceeded - InvalidOpcode - DeviceNotAvailable - DoubleFault - CoprocessorSegmentOverrun - InvalidTSS - SegmentNotPresent - StackSegmentFault - GeneralProtectionFault - PageFault - _ - X87FloatingPointException - AlignmentCheck - MachineCheck - SIMDFloatingPointException - VirtualizationException - SecurityException = 0x1e - SyscallInt80 = 0x80 - _NR_INTERRUPTS = SyscallInt80 + 1 -) - -// System call vectors. -const ( - Syscall Vector = _NR_INTERRUPTS -) - -// VirtualAddressBits returns the number bits available for virtual addresses. -// -// Note that sign-extension semantics apply to the highest order bit. -// -// FIXME(b/69382326): This should use the cpuid passed to Init. -func VirtualAddressBits() uint32 { - ax, _, _, _ := cpuid.HostID(0x80000008, 0) - return (ax >> 8) & 0xff -} - -// PhysicalAddressBits returns the number of bits available for physical addresses. -// -// FIXME(b/69382326): This should use the cpuid passed to Init. -func PhysicalAddressBits() uint32 { - ax, _, _, _ := cpuid.HostID(0x80000008, 0) - return ax & 0xff -} - -// Selector is a segment Selector. -type Selector uint16 - -// SegmentDescriptor is a segment descriptor. -type SegmentDescriptor struct { - bits [2]uint32 -} - -// descriptorTable is a collection of descriptors. -type descriptorTable [32]SegmentDescriptor - -// SegmentDescriptorFlags are typed flags within a descriptor. -type SegmentDescriptorFlags uint32 - -// SegmentDescriptorFlag declarations. -const ( - SegmentDescriptorAccess SegmentDescriptorFlags = 1 << 8 // Access bit (always set). - SegmentDescriptorWrite = 1 << 9 // Write permission. - SegmentDescriptorExpandDown = 1 << 10 // Grows down, not used. - SegmentDescriptorExecute = 1 << 11 // Execute permission. - SegmentDescriptorSystem = 1 << 12 // Zero => system, 1 => user code/data. - SegmentDescriptorPresent = 1 << 15 // Present. - SegmentDescriptorAVL = 1 << 20 // Available. - SegmentDescriptorLong = 1 << 21 // Long mode. - SegmentDescriptorDB = 1 << 22 // 16 or 32-bit. - SegmentDescriptorG = 1 << 23 // Granularity: page or byte. -) - -// Base returns the descriptor's base linear address. -func (d *SegmentDescriptor) Base() uint32 { - return d.bits[1]&0xFF000000 | (d.bits[1]&0x000000FF)<<16 | d.bits[0]>>16 -} - -// Limit returns the descriptor size. -func (d *SegmentDescriptor) Limit() uint32 { - l := d.bits[0]&0xFFFF | d.bits[1]&0xF0000 - if d.bits[1]&uint32(SegmentDescriptorG) != 0 { - l <<= 12 - l |= 0xFFF - } - return l -} - -// Flags returns descriptor flags. -func (d *SegmentDescriptor) Flags() SegmentDescriptorFlags { - return SegmentDescriptorFlags(d.bits[1] & 0x00F09F00) -} - -// DPL returns the descriptor privilege level. -func (d *SegmentDescriptor) DPL() int { - return int((d.bits[1] >> 13) & 3) -} - -func (d *SegmentDescriptor) setNull() { - d.bits[0] = 0 - d.bits[1] = 0 -} - -func (d *SegmentDescriptor) set(base, limit uint32, dpl int, flags SegmentDescriptorFlags) { - flags |= SegmentDescriptorPresent - if limit>>12 != 0 { - limit >>= 12 - flags |= SegmentDescriptorG - } - d.bits[0] = base<<16 | limit&0xFFFF - d.bits[1] = base&0xFF000000 | (base>>16)&0xFF | limit&0x000F0000 | uint32(flags) | uint32(dpl)<<13 -} - -func (d *SegmentDescriptor) setCode32(base, limit uint32, dpl int) { - d.set(base, limit, dpl, - SegmentDescriptorDB| - SegmentDescriptorExecute| - SegmentDescriptorSystem) -} - -func (d *SegmentDescriptor) setCode64(base, limit uint32, dpl int) { - d.set(base, limit, dpl, - SegmentDescriptorG| - SegmentDescriptorLong| - SegmentDescriptorExecute| - SegmentDescriptorSystem) -} - -func (d *SegmentDescriptor) setData(base, limit uint32, dpl int) { - d.set(base, limit, dpl, - SegmentDescriptorWrite| - SegmentDescriptorSystem) -} - -// setHi is only used for the TSS segment, which is magically 64-bits. -func (d *SegmentDescriptor) setHi(base uint32) { - d.bits[0] = base - d.bits[1] = 0 -} - -// Gate64 is a 64-bit task, trap, or interrupt gate. -type Gate64 struct { - bits [4]uint32 -} - -// idt64 is a 64-bit interrupt descriptor table. -type idt64 [_NR_INTERRUPTS]Gate64 - -func (g *Gate64) setInterrupt(cs Selector, rip uint64, dpl int, ist int) { - g.bits[0] = uint32(cs)<<16 | uint32(rip)&0xFFFF - g.bits[1] = uint32(rip)&0xFFFF0000 | SegmentDescriptorPresent | uint32(dpl)<<13 | 14<<8 | uint32(ist)&0x7 - g.bits[2] = uint32(rip >> 32) -} - -func (g *Gate64) setTrap(cs Selector, rip uint64, dpl int, ist int) { - g.setInterrupt(cs, rip, dpl, ist) - g.bits[1] |= 1 << 8 -} - -// TaskState64 is a 64-bit task state structure. -type TaskState64 struct { - _ uint32 - rsp0Lo, rsp0Hi uint32 - rsp1Lo, rsp1Hi uint32 - rsp2Lo, rsp2Hi uint32 - _ [2]uint32 - ist1Lo, ist1Hi uint32 - ist2Lo, ist2Hi uint32 - ist3Lo, ist3Hi uint32 - ist4Lo, ist4Hi uint32 - ist5Lo, ist5Hi uint32 - ist6Lo, ist6Hi uint32 - ist7Lo, ist7Hi uint32 - _ [2]uint32 - _ uint16 - ioPerm uint16 -} diff --git a/pkg/sentry/sighandling/BUILD b/pkg/sentry/sighandling/BUILD deleted file mode 100644 index 6c38a3f44..000000000 --- a/pkg/sentry/sighandling/BUILD +++ /dev/null @@ -1,13 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "sighandling", - srcs = [ - "sighandling.go", - "sighandling_unsafe.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = ["//pkg/abi/linux"], -) diff --git a/pkg/sentry/sighandling/sighandling_state_autogen.go b/pkg/sentry/sighandling/sighandling_state_autogen.go new file mode 100755 index 000000000..da9d96382 --- /dev/null +++ b/pkg/sentry/sighandling/sighandling_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package sighandling diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD deleted file mode 100644 index 611fa22c3..000000000 --- a/pkg/sentry/socket/BUILD +++ /dev/null @@ -1,23 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "socket", - srcs = ["socket.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket/unix/transport", - "//pkg/syserr", - "//pkg/tcpip", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/socket/control/BUILD b/pkg/sentry/socket/control/BUILD deleted file mode 100644 index 4d42d29cb..000000000 --- a/pkg/sentry/socket/control/BUILD +++ /dev/null @@ -1,25 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "control", - srcs = ["control.go"], - imports = [ - "gvisor.dev/gvisor/pkg/sentry/fs", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/sentry/fs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/socket", - "//pkg/sentry/socket/unix/transport", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/socket/control/control_state_autogen.go b/pkg/sentry/socket/control/control_state_autogen.go new file mode 100755 index 000000000..8a37b04c0 --- /dev/null +++ b/pkg/sentry/socket/control/control_state_autogen.go @@ -0,0 +1,36 @@ +// automatically generated by stateify. + +package control + +import ( + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/state" +) + +func (x *RightsFiles) save(m state.Map) { + m.SaveValue("", ([]*fs.File)(*x)) +} + +func (x *RightsFiles) load(m state.Map) { + m.LoadValue("", new([]*fs.File), func(y interface{}) { *x = (RightsFiles)(y.([]*fs.File)) }) +} + +func (x *scmCredentials) beforeSave() {} +func (x *scmCredentials) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) + m.Save("kuid", &x.kuid) + m.Save("kgid", &x.kgid) +} + +func (x *scmCredentials) afterLoad() {} +func (x *scmCredentials) load(m state.Map) { + m.Load("t", &x.t) + m.Load("kuid", &x.kuid) + m.Load("kgid", &x.kgid) +} + +func init() { + state.Register("pkg/sentry/socket/control.RightsFiles", (*RightsFiles)(nil), state.Fns{Save: (*RightsFiles).save, Load: (*RightsFiles).load}) + state.Register("pkg/sentry/socket/control.scmCredentials", (*scmCredentials)(nil), state.Fns{Save: (*scmCredentials).save, Load: (*scmCredentials).load}) +} diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD deleted file mode 100644 index 023bad156..000000000 --- a/pkg/sentry/socket/hostinet/BUILD +++ /dev/null @@ -1,40 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "hostinet", - srcs = [ - "device.go", - "hostinet.go", - "save_restore.go", - "socket.go", - "socket_unsafe.go", - "sockopt_impl.go", - "stack.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/fdnotifier", - "//pkg/log", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket", - "//pkg/sentry/socket/control", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip/stack", - "//pkg/usermem", - "//pkg/waiter", - "@org_golang_x_sys//unix:go_default_library", - ], -) diff --git a/pkg/sentry/socket/hostinet/hostinet_impl_state_autogen.go b/pkg/sentry/socket/hostinet/hostinet_impl_state_autogen.go new file mode 100755 index 000000000..b0a59ba93 --- /dev/null +++ b/pkg/sentry/socket/hostinet/hostinet_impl_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package hostinet diff --git a/pkg/sentry/socket/hostinet/hostinet_state_autogen.go b/pkg/sentry/socket/hostinet/hostinet_state_autogen.go new file mode 100755 index 000000000..b0a59ba93 --- /dev/null +++ b/pkg/sentry/socket/hostinet/hostinet_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package hostinet diff --git a/pkg/sentry/socket/hostinet/sockopt_impl.go b/pkg/sentry/socket/hostinet/sockopt_impl.go index 8a783712e..8a783712e 100644..100755 --- a/pkg/sentry/socket/hostinet/sockopt_impl.go +++ b/pkg/sentry/socket/hostinet/sockopt_impl.go diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD deleted file mode 100644 index 721094bbf..000000000 --- a/pkg/sentry/socket/netfilter/BUILD +++ /dev/null @@ -1,29 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "netfilter", - srcs = [ - "extensions.go", - "netfilter.go", - "owner_matcher.go", - "targets.go", - "tcp_matcher.go", - "udp_matcher.go", - ], - # This target depends on netstack and should only be used by epsocket, - # which is allowed to depend on netstack. - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/log", - "//pkg/sentry/kernel", - "//pkg/syserr", - "//pkg/tcpip", - "//pkg/tcpip/header", - "//pkg/tcpip/stack", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/socket/netfilter/extensions.go b/pkg/sentry/socket/netfilter/extensions.go index 0336a32d8..0336a32d8 100644..100755 --- a/pkg/sentry/socket/netfilter/extensions.go +++ b/pkg/sentry/socket/netfilter/extensions.go diff --git a/pkg/sentry/socket/netfilter/netfilter_state_autogen.go b/pkg/sentry/socket/netfilter/netfilter_state_autogen.go new file mode 100755 index 000000000..6e95d89a4 --- /dev/null +++ b/pkg/sentry/socket/netfilter/netfilter_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package netfilter diff --git a/pkg/sentry/socket/netfilter/owner_matcher.go b/pkg/sentry/socket/netfilter/owner_matcher.go index 5949a7c29..5949a7c29 100644..100755 --- a/pkg/sentry/socket/netfilter/owner_matcher.go +++ b/pkg/sentry/socket/netfilter/owner_matcher.go diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go index c948de876..c948de876 100644..100755 --- a/pkg/sentry/socket/netfilter/targets.go +++ b/pkg/sentry/socket/netfilter/targets.go diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index ff1cfd8f6..ff1cfd8f6 100644..100755 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 3359418c1..3359418c1 100644..100755 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD deleted file mode 100644 index 1911cd9b8..000000000 --- a/pkg/sentry/socket/netlink/BUILD +++ /dev/null @@ -1,47 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "netlink", - srcs = [ - "message.go", - "provider.go", - "socket.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket", - "//pkg/sentry/socket/netlink/port", - "//pkg/sentry/socket/unix", - "//pkg/sentry/socket/unix/transport", - "//pkg/sync", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "netlink_test", - size = "small", - srcs = [ - "message_test.go", - ], - deps = [ - ":netlink", - "//pkg/abi/linux", - ], -) diff --git a/pkg/sentry/socket/netlink/message_test.go b/pkg/sentry/socket/netlink/message_test.go deleted file mode 100644 index ef13d9386..000000000 --- a/pkg/sentry/socket/netlink/message_test.go +++ /dev/null @@ -1,312 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package message_test - -import ( - "bytes" - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/socket/netlink" -) - -type dummyNetlinkMsg struct { - Foo uint16 -} - -func TestParseMessage(t *testing.T) { - tests := []struct { - desc string - input []byte - - header linux.NetlinkMessageHeader - dataMsg *dummyNetlinkMsg - restLen int - ok bool - }{ - { - desc: "valid", - input: []byte{ - 0x14, 0x00, 0x00, 0x00, // Length - 0x01, 0x00, // Type - 0x02, 0x00, // Flags - 0x03, 0x00, 0x00, 0x00, // Seq - 0x04, 0x00, 0x00, 0x00, // PortID - 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding - }, - header: linux.NetlinkMessageHeader{ - Length: 20, - Type: 1, - Flags: 2, - Seq: 3, - PortID: 4, - }, - dataMsg: &dummyNetlinkMsg{ - Foo: 0x3130, - }, - restLen: 0, - ok: true, - }, - { - desc: "valid with next message", - input: []byte{ - 0x14, 0x00, 0x00, 0x00, // Length - 0x01, 0x00, // Type - 0x02, 0x00, // Flags - 0x03, 0x00, 0x00, 0x00, // Seq - 0x04, 0x00, 0x00, 0x00, // PortID - 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding - 0xFF, // Next message (rest) - }, - header: linux.NetlinkMessageHeader{ - Length: 20, - Type: 1, - Flags: 2, - Seq: 3, - PortID: 4, - }, - dataMsg: &dummyNetlinkMsg{ - Foo: 0x3130, - }, - restLen: 1, - ok: true, - }, - { - desc: "valid for last message without padding", - input: []byte{ - 0x12, 0x00, 0x00, 0x00, // Length - 0x01, 0x00, // Type - 0x02, 0x00, // Flags - 0x03, 0x00, 0x00, 0x00, // Seq - 0x04, 0x00, 0x00, 0x00, // PortID - 0x30, 0x31, // Data message - }, - header: linux.NetlinkMessageHeader{ - Length: 18, - Type: 1, - Flags: 2, - Seq: 3, - PortID: 4, - }, - dataMsg: &dummyNetlinkMsg{ - Foo: 0x3130, - }, - restLen: 0, - ok: true, - }, - { - desc: "valid for last message not to be aligned", - input: []byte{ - 0x13, 0x00, 0x00, 0x00, // Length - 0x01, 0x00, // Type - 0x02, 0x00, // Flags - 0x03, 0x00, 0x00, 0x00, // Seq - 0x04, 0x00, 0x00, 0x00, // PortID - 0x30, 0x31, // Data message - 0x00, // Excessive 1 byte permitted at end - }, - header: linux.NetlinkMessageHeader{ - Length: 19, - Type: 1, - Flags: 2, - Seq: 3, - PortID: 4, - }, - dataMsg: &dummyNetlinkMsg{ - Foo: 0x3130, - }, - restLen: 0, - ok: true, - }, - { - desc: "header.Length too short", - input: []byte{ - 0x04, 0x00, 0x00, 0x00, // Length - 0x01, 0x00, // Type - 0x02, 0x00, // Flags - 0x03, 0x00, 0x00, 0x00, // Seq - 0x04, 0x00, 0x00, 0x00, // PortID - 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding - }, - ok: false, - }, - { - desc: "header.Length too long", - input: []byte{ - 0xFF, 0xFF, 0x00, 0x00, // Length - 0x01, 0x00, // Type - 0x02, 0x00, // Flags - 0x03, 0x00, 0x00, 0x00, // Seq - 0x04, 0x00, 0x00, 0x00, // PortID - 0x30, 0x31, 0x00, 0x00, // Data message with 2 bytes padding - }, - ok: false, - }, - { - desc: "header incomplete", - input: []byte{ - 0x04, 0x00, 0x00, 0x00, // Length - }, - ok: false, - }, - { - desc: "empty message", - input: []byte{}, - ok: false, - }, - } - for _, test := range tests { - msg, rest, ok := netlink.ParseMessage(test.input) - if ok != test.ok { - t.Errorf("%v: got ok = %v, want = %v", test.desc, ok, test.ok) - continue - } - if !test.ok { - continue - } - if !reflect.DeepEqual(msg.Header(), test.header) { - t.Errorf("%v: got hdr = %+v, want = %+v", test.desc, msg.Header(), test.header) - } - - dataMsg := &dummyNetlinkMsg{} - _, dataOk := msg.GetData(dataMsg) - if !dataOk { - t.Errorf("%v: GetData.ok = %v, want = true", test.desc, dataOk) - } else if !reflect.DeepEqual(dataMsg, test.dataMsg) { - t.Errorf("%v: GetData.msg = %+v, want = %+v", test.desc, dataMsg, test.dataMsg) - } - - if got, want := rest, test.input[len(test.input)-test.restLen:]; !bytes.Equal(got, want) { - t.Errorf("%v: got rest = %v, want = %v", test.desc, got, want) - } - } -} - -func TestAttrView(t *testing.T) { - tests := []struct { - desc string - input []byte - - // Outputs for ParseFirst. - hdr linux.NetlinkAttrHeader - value []byte - restLen int - ok bool - - // Outputs for Empty. - isEmpty bool - }{ - { - desc: "valid", - input: []byte{ - 0x06, 0x00, // Length - 0x01, 0x00, // Type - 0x30, 0x31, 0x00, 0x00, // Data with 2 bytes padding - }, - hdr: linux.NetlinkAttrHeader{ - Length: 6, - Type: 1, - }, - value: []byte{0x30, 0x31}, - restLen: 0, - ok: true, - isEmpty: false, - }, - { - desc: "at alignment", - input: []byte{ - 0x08, 0x00, // Length - 0x01, 0x00, // Type - 0x30, 0x31, 0x32, 0x33, // Data - }, - hdr: linux.NetlinkAttrHeader{ - Length: 8, - Type: 1, - }, - value: []byte{0x30, 0x31, 0x32, 0x33}, - restLen: 0, - ok: true, - isEmpty: false, - }, - { - desc: "at alignment with rest data", - input: []byte{ - 0x08, 0x00, // Length - 0x01, 0x00, // Type - 0x30, 0x31, 0x32, 0x33, // Data - 0xFF, 0xFE, // Rest data - }, - hdr: linux.NetlinkAttrHeader{ - Length: 8, - Type: 1, - }, - value: []byte{0x30, 0x31, 0x32, 0x33}, - restLen: 2, - ok: true, - isEmpty: false, - }, - { - desc: "hdr.Length too long", - input: []byte{ - 0xFF, 0x00, // Length - 0x01, 0x00, // Type - 0x30, 0x31, 0x32, 0x33, // Data - }, - ok: false, - isEmpty: false, - }, - { - desc: "hdr.Length too short", - input: []byte{ - 0x01, 0x00, // Length - 0x01, 0x00, // Type - 0x30, 0x31, 0x32, 0x33, // Data - }, - ok: false, - isEmpty: false, - }, - { - desc: "empty", - input: []byte{}, - ok: false, - isEmpty: true, - }, - } - for _, test := range tests { - attrs := netlink.AttrsView(test.input) - - // Test ParseFirst(). - hdr, value, rest, ok := attrs.ParseFirst() - if ok != test.ok { - t.Errorf("%v: got ok = %v, want = %v", test.desc, ok, test.ok) - } else if test.ok { - if !reflect.DeepEqual(hdr, test.hdr) { - t.Errorf("%v: got hdr = %+v, want = %+v", test.desc, hdr, test.hdr) - } - if !bytes.Equal(value, test.value) { - t.Errorf("%v: got value = %v, want = %v", test.desc, value, test.value) - } - if wantRest := test.input[len(test.input)-test.restLen:]; !bytes.Equal(rest, wantRest) { - t.Errorf("%v: got rest = %v, want = %v", test.desc, rest, wantRest) - } - } - - // Test Empty(). - if got, want := attrs.Empty(), test.isEmpty; got != want { - t.Errorf("%v: got empty = %v, want = %v", test.desc, got, want) - } - } -} diff --git a/pkg/sentry/socket/netlink/netlink_state_autogen.go b/pkg/sentry/socket/netlink/netlink_state_autogen.go new file mode 100755 index 000000000..792ac6774 --- /dev/null +++ b/pkg/sentry/socket/netlink/netlink_state_autogen.go @@ -0,0 +1,52 @@ +// automatically generated by stateify. + +package netlink + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Socket) beforeSave() {} +func (x *Socket) save(m state.Map) { + x.beforeSave() + m.Save("SendReceiveTimeout", &x.SendReceiveTimeout) + m.Save("ports", &x.ports) + m.Save("protocol", &x.protocol) + m.Save("skType", &x.skType) + m.Save("ep", &x.ep) + m.Save("connection", &x.connection) + m.Save("bound", &x.bound) + m.Save("portID", &x.portID) + m.Save("sendBufferSize", &x.sendBufferSize) + m.Save("passcred", &x.passcred) + m.Save("filter", &x.filter) +} + +func (x *Socket) afterLoad() {} +func (x *Socket) load(m state.Map) { + m.Load("SendReceiveTimeout", &x.SendReceiveTimeout) + m.Load("ports", &x.ports) + m.Load("protocol", &x.protocol) + m.Load("skType", &x.skType) + m.Load("ep", &x.ep) + m.Load("connection", &x.connection) + m.Load("bound", &x.bound) + m.Load("portID", &x.portID) + m.Load("sendBufferSize", &x.sendBufferSize) + m.Load("passcred", &x.passcred) + m.Load("filter", &x.filter) +} + +func (x *kernelSCM) beforeSave() {} +func (x *kernelSCM) save(m state.Map) { + x.beforeSave() +} + +func (x *kernelSCM) afterLoad() {} +func (x *kernelSCM) load(m state.Map) { +} + +func init() { + state.Register("pkg/sentry/socket/netlink.Socket", (*Socket)(nil), state.Fns{Save: (*Socket).save, Load: (*Socket).load}) + state.Register("pkg/sentry/socket/netlink.kernelSCM", (*kernelSCM)(nil), state.Fns{Save: (*kernelSCM).save, Load: (*kernelSCM).load}) +} diff --git a/pkg/sentry/socket/netlink/port/BUILD b/pkg/sentry/socket/netlink/port/BUILD deleted file mode 100644 index 3a22923d8..000000000 --- a/pkg/sentry/socket/netlink/port/BUILD +++ /dev/null @@ -1,16 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -go_library( - name = "port", - srcs = ["port.go"], - visibility = ["//pkg/sentry:internal"], - deps = ["//pkg/sync"], -) - -go_test( - name = "port_test", - srcs = ["port_test.go"], - library = ":port", -) diff --git a/pkg/sentry/socket/netlink/port/port_state_autogen.go b/pkg/sentry/socket/netlink/port/port_state_autogen.go new file mode 100755 index 000000000..c509cc7d5 --- /dev/null +++ b/pkg/sentry/socket/netlink/port/port_state_autogen.go @@ -0,0 +1,22 @@ +// automatically generated by stateify. + +package port + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Manager) beforeSave() {} +func (x *Manager) save(m state.Map) { + x.beforeSave() + m.Save("ports", &x.ports) +} + +func (x *Manager) afterLoad() {} +func (x *Manager) load(m state.Map) { + m.Load("ports", &x.ports) +} + +func init() { + state.Register("pkg/sentry/socket/netlink/port.Manager", (*Manager)(nil), state.Fns{Save: (*Manager).save, Load: (*Manager).load}) +} diff --git a/pkg/sentry/socket/netlink/port/port_test.go b/pkg/sentry/socket/netlink/port/port_test.go deleted file mode 100644 index 516f6cd6c..000000000 --- a/pkg/sentry/socket/netlink/port/port_test.go +++ /dev/null @@ -1,82 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package port - -import ( - "testing" -) - -func TestAllocateHint(t *testing.T) { - m := New() - - // We can get the hint port. - p, ok := m.Allocate(0, 1) - if !ok { - t.Errorf("m.Allocate got !ok want ok") - } - if p != 1 { - t.Errorf("m.Allocate(0, 1) got %d want 1", p) - } - - // Hint is taken. - p, ok = m.Allocate(0, 1) - if !ok { - t.Errorf("m.Allocate got !ok want ok") - } - if p == 1 { - t.Errorf("m.Allocate(0, 1) got 1 want anything else") - } - - // Hint is available for a different protocol. - p, ok = m.Allocate(1, 1) - if !ok { - t.Errorf("m.Allocate got !ok want ok") - } - if p != 1 { - t.Errorf("m.Allocate(1, 1) got %d want 1", p) - } - - m.Release(0, 1) - - // Hint is available again after release. - p, ok = m.Allocate(0, 1) - if !ok { - t.Errorf("m.Allocate got !ok want ok") - } - if p != 1 { - t.Errorf("m.Allocate(0, 1) got %d want 1", p) - } -} - -func TestAllocateExhausted(t *testing.T) { - m := New() - - // Fill all ports (0 is already reserved). - for i := int32(1); i < maxPorts; i++ { - p, ok := m.Allocate(0, i) - if !ok { - t.Fatalf("m.Allocate got !ok want ok") - } - if p != i { - t.Fatalf("m.Allocate(0, %d) got %d want %d", i, p, i) - } - } - - // Now no more can be allocated. - p, ok := m.Allocate(0, 1) - if ok { - t.Errorf("m.Allocate got %d, ok want !ok", p) - } -} diff --git a/pkg/sentry/socket/netlink/route/BUILD b/pkg/sentry/socket/netlink/route/BUILD deleted file mode 100644 index 93127398d..000000000 --- a/pkg/sentry/socket/netlink/route/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "route", - srcs = [ - "protocol.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/socket/netlink", - "//pkg/syserr", - ], -) diff --git a/pkg/sentry/socket/netlink/route/route_state_autogen.go b/pkg/sentry/socket/netlink/route/route_state_autogen.go new file mode 100755 index 000000000..bd10fe189 --- /dev/null +++ b/pkg/sentry/socket/netlink/route/route_state_autogen.go @@ -0,0 +1,20 @@ +// automatically generated by stateify. + +package route + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Protocol) beforeSave() {} +func (x *Protocol) save(m state.Map) { + x.beforeSave() +} + +func (x *Protocol) afterLoad() {} +func (x *Protocol) load(m state.Map) { +} + +func init() { + state.Register("pkg/sentry/socket/netlink/route.Protocol", (*Protocol)(nil), state.Fns{Save: (*Protocol).save, Load: (*Protocol).load}) +} diff --git a/pkg/sentry/socket/netlink/uevent/BUILD b/pkg/sentry/socket/netlink/uevent/BUILD deleted file mode 100644 index b6434923c..000000000 --- a/pkg/sentry/socket/netlink/uevent/BUILD +++ /dev/null @@ -1,16 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "uevent", - srcs = ["protocol.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/kernel", - "//pkg/sentry/socket/netlink", - "//pkg/syserr", - ], -) diff --git a/pkg/sentry/socket/netlink/uevent/protocol.go b/pkg/sentry/socket/netlink/uevent/protocol.go index 029ba21b5..029ba21b5 100644..100755 --- a/pkg/sentry/socket/netlink/uevent/protocol.go +++ b/pkg/sentry/socket/netlink/uevent/protocol.go diff --git a/pkg/sentry/socket/netlink/uevent/uevent_state_autogen.go b/pkg/sentry/socket/netlink/uevent/uevent_state_autogen.go new file mode 100755 index 000000000..b82dddf32 --- /dev/null +++ b/pkg/sentry/socket/netlink/uevent/uevent_state_autogen.go @@ -0,0 +1,20 @@ +// automatically generated by stateify. + +package uevent + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Protocol) beforeSave() {} +func (x *Protocol) save(m state.Map) { + x.beforeSave() +} + +func (x *Protocol) afterLoad() {} +func (x *Protocol) load(m state.Map) { +} + +func init() { + state.Register("pkg/sentry/socket/netlink/uevent.Protocol", (*Protocol)(nil), state.Fns{Save: (*Protocol).save, Load: (*Protocol).load}) +} diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD deleted file mode 100644 index cbf46b1e9..000000000 --- a/pkg/sentry/socket/netstack/BUILD +++ /dev/null @@ -1,49 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "netstack", - srcs = [ - "device.go", - "netstack.go", - "provider.go", - "save_restore.go", - "stack.go", - ], - visibility = [ - "//pkg/sentry:internal", - ], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/log", - "//pkg/metric", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket", - "//pkg/sentry/socket/netfilter", - "//pkg/sentry/unimpl", - "//pkg/sync", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/header", - "//pkg/tcpip/network/ipv4", - "//pkg/tcpip/network/ipv6", - "//pkg/tcpip/stack", - "//pkg/tcpip/transport/tcp", - "//pkg/tcpip/transport/udp", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/socket/netstack/device.go b/pkg/sentry/socket/netstack/device.go index fbeb89fb8..fbeb89fb8 100644..100755 --- a/pkg/sentry/socket/netstack/device.go +++ b/pkg/sentry/socket/netstack/device.go diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index f14c336b9..f14c336b9 100644..100755 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go diff --git a/pkg/sentry/socket/netstack/netstack_state_autogen.go b/pkg/sentry/socket/netstack/netstack_state_autogen.go new file mode 100755 index 000000000..9be6fe242 --- /dev/null +++ b/pkg/sentry/socket/netstack/netstack_state_autogen.go @@ -0,0 +1,58 @@ +// automatically generated by stateify. + +package netstack + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *SocketOperations) beforeSave() {} +func (x *SocketOperations) save(m state.Map) { + x.beforeSave() + m.Save("SendReceiveTimeout", &x.SendReceiveTimeout) + m.Save("Queue", &x.Queue) + m.Save("family", &x.family) + m.Save("Endpoint", &x.Endpoint) + m.Save("skType", &x.skType) + m.Save("protocol", &x.protocol) + m.Save("readViewHasData", &x.readViewHasData) + m.Save("readView", &x.readView) + m.Save("readCM", &x.readCM) + m.Save("sender", &x.sender) + m.Save("sockOptTimestamp", &x.sockOptTimestamp) + m.Save("timestampValid", &x.timestampValid) + m.Save("timestampNS", &x.timestampNS) + m.Save("sockOptInq", &x.sockOptInq) +} + +func (x *SocketOperations) afterLoad() {} +func (x *SocketOperations) load(m state.Map) { + m.Load("SendReceiveTimeout", &x.SendReceiveTimeout) + m.Load("Queue", &x.Queue) + m.Load("family", &x.family) + m.Load("Endpoint", &x.Endpoint) + m.Load("skType", &x.skType) + m.Load("protocol", &x.protocol) + m.Load("readViewHasData", &x.readViewHasData) + m.Load("readView", &x.readView) + m.Load("readCM", &x.readCM) + m.Load("sender", &x.sender) + m.Load("sockOptTimestamp", &x.sockOptTimestamp) + m.Load("timestampValid", &x.timestampValid) + m.Load("timestampNS", &x.timestampNS) + m.Load("sockOptInq", &x.sockOptInq) +} + +func (x *Stack) beforeSave() {} +func (x *Stack) save(m state.Map) { + x.beforeSave() +} + +func (x *Stack) load(m state.Map) { + m.AfterLoad(x.afterLoad) +} + +func init() { + state.Register("pkg/sentry/socket/netstack.SocketOperations", (*SocketOperations)(nil), state.Fns{Save: (*SocketOperations).save, Load: (*SocketOperations).load}) + state.Register("pkg/sentry/socket/netstack.Stack", (*Stack)(nil), state.Fns{Save: (*Stack).save, Load: (*Stack).load}) +} diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go index eb090e79b..eb090e79b 100644..100755 --- a/pkg/sentry/socket/netstack/provider.go +++ b/pkg/sentry/socket/netstack/provider.go diff --git a/pkg/sentry/socket/netstack/save_restore.go b/pkg/sentry/socket/netstack/save_restore.go index c7aaf722a..c7aaf722a 100644..100755 --- a/pkg/sentry/socket/netstack/save_restore.go +++ b/pkg/sentry/socket/netstack/save_restore.go diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index f5fa18136..f5fa18136 100644..100755 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go diff --git a/pkg/sentry/socket/socket_state_autogen.go b/pkg/sentry/socket/socket_state_autogen.go new file mode 100755 index 000000000..900c217c7 --- /dev/null +++ b/pkg/sentry/socket/socket_state_autogen.go @@ -0,0 +1,24 @@ +// automatically generated by stateify. + +package socket + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *SendReceiveTimeout) beforeSave() {} +func (x *SendReceiveTimeout) save(m state.Map) { + x.beforeSave() + m.Save("send", &x.send) + m.Save("recv", &x.recv) +} + +func (x *SendReceiveTimeout) afterLoad() {} +func (x *SendReceiveTimeout) load(m state.Map) { + m.Load("send", &x.send) + m.Load("recv", &x.recv) +} + +func init() { + state.Register("pkg/sentry/socket.SendReceiveTimeout", (*SendReceiveTimeout)(nil), state.Fns{Save: (*SendReceiveTimeout).save, Load: (*SendReceiveTimeout).load}) +} diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD deleted file mode 100644 index 08743deba..000000000 --- a/pkg/sentry/socket/unix/BUILD +++ /dev/null @@ -1,34 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "unix", - srcs = [ - "device.go", - "io.go", - "unix.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/refs", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket", - "//pkg/sentry/socket/control", - "//pkg/sentry/socket/netstack", - "//pkg/sentry/socket/unix/transport", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD deleted file mode 100644 index 74bcd6300..000000000 --- a/pkg/sentry/socket/unix/transport/BUILD +++ /dev/null @@ -1,40 +0,0 @@ -load("//tools:defs.bzl", "go_library") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "transport_message_list", - out = "transport_message_list.go", - package = "transport", - prefix = "message", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*message", - "Linker": "*message", - }, -) - -go_library( - name = "transport", - srcs = [ - "connectioned.go", - "connectioned_state.go", - "connectionless.go", - "queue.go", - "transport_message_list.go", - "unix.go", - ], - visibility = ["//:sandbox"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/ilist", - "//pkg/refs", - "//pkg/sync", - "//pkg/syserr", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/socket/unix/transport/transport_message_list.go b/pkg/sentry/socket/unix/transport/transport_message_list.go new file mode 100755 index 000000000..9edc731b4 --- /dev/null +++ b/pkg/sentry/socket/unix/transport/transport_message_list.go @@ -0,0 +1,186 @@ +package transport + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type messageElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (messageElementMapper) linkerFor(elem *message) *message { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type messageList struct { + head *message + tail *message +} + +// Reset resets list l to the empty state. +func (l *messageList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *messageList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *messageList) Front() *message { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *messageList) Back() *message { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *messageList) PushFront(e *message) { + linker := messageElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + messageElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *messageList) PushBack(e *message) { + linker := messageElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + messageElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *messageList) PushBackList(m *messageList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + messageElementMapper{}.linkerFor(l.tail).SetNext(m.head) + messageElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *messageList) InsertAfter(b, e *message) { + bLinker := messageElementMapper{}.linkerFor(b) + eLinker := messageElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + messageElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *messageList) InsertBefore(a, e *message) { + aLinker := messageElementMapper{}.linkerFor(a) + eLinker := messageElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + messageElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *messageList) Remove(e *message) { + linker := messageElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + messageElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + messageElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type messageEntry struct { + next *message + prev *message +} + +// Next returns the entry that follows e in the list. +func (e *messageEntry) Next() *message { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *messageEntry) Prev() *message { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *messageEntry) SetNext(elem *message) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *messageEntry) SetPrev(elem *message) { + e.prev = elem +} diff --git a/pkg/sentry/socket/unix/transport/transport_state_autogen.go b/pkg/sentry/socket/unix/transport/transport_state_autogen.go new file mode 100755 index 000000000..b47951498 --- /dev/null +++ b/pkg/sentry/socket/unix/transport/transport_state_autogen.go @@ -0,0 +1,193 @@ +// automatically generated by stateify. + +package transport + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *connectionedEndpoint) beforeSave() {} +func (x *connectionedEndpoint) save(m state.Map) { + x.beforeSave() + var acceptedChan []*connectionedEndpoint = x.saveAcceptedChan() + m.SaveValue("acceptedChan", acceptedChan) + m.Save("baseEndpoint", &x.baseEndpoint) + m.Save("id", &x.id) + m.Save("idGenerator", &x.idGenerator) + m.Save("stype", &x.stype) +} + +func (x *connectionedEndpoint) afterLoad() {} +func (x *connectionedEndpoint) load(m state.Map) { + m.Load("baseEndpoint", &x.baseEndpoint) + m.Load("id", &x.id) + m.Load("idGenerator", &x.idGenerator) + m.Load("stype", &x.stype) + m.LoadValue("acceptedChan", new([]*connectionedEndpoint), func(y interface{}) { x.loadAcceptedChan(y.([]*connectionedEndpoint)) }) +} + +func (x *connectionlessEndpoint) beforeSave() {} +func (x *connectionlessEndpoint) save(m state.Map) { + x.beforeSave() + m.Save("baseEndpoint", &x.baseEndpoint) +} + +func (x *connectionlessEndpoint) afterLoad() {} +func (x *connectionlessEndpoint) load(m state.Map) { + m.Load("baseEndpoint", &x.baseEndpoint) +} + +func (x *queue) beforeSave() {} +func (x *queue) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("ReaderQueue", &x.ReaderQueue) + m.Save("WriterQueue", &x.WriterQueue) + m.Save("closed", &x.closed) + m.Save("unread", &x.unread) + m.Save("used", &x.used) + m.Save("limit", &x.limit) + m.Save("dataList", &x.dataList) +} + +func (x *queue) afterLoad() {} +func (x *queue) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("ReaderQueue", &x.ReaderQueue) + m.Load("WriterQueue", &x.WriterQueue) + m.Load("closed", &x.closed) + m.Load("unread", &x.unread) + m.Load("used", &x.used) + m.Load("limit", &x.limit) + m.Load("dataList", &x.dataList) +} + +func (x *messageList) beforeSave() {} +func (x *messageList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *messageList) afterLoad() {} +func (x *messageList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *messageEntry) beforeSave() {} +func (x *messageEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *messageEntry) afterLoad() {} +func (x *messageEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *ControlMessages) beforeSave() {} +func (x *ControlMessages) save(m state.Map) { + x.beforeSave() + m.Save("Rights", &x.Rights) + m.Save("Credentials", &x.Credentials) +} + +func (x *ControlMessages) afterLoad() {} +func (x *ControlMessages) load(m state.Map) { + m.Load("Rights", &x.Rights) + m.Load("Credentials", &x.Credentials) +} + +func (x *message) beforeSave() {} +func (x *message) save(m state.Map) { + x.beforeSave() + m.Save("messageEntry", &x.messageEntry) + m.Save("Data", &x.Data) + m.Save("Control", &x.Control) + m.Save("Address", &x.Address) +} + +func (x *message) afterLoad() {} +func (x *message) load(m state.Map) { + m.Load("messageEntry", &x.messageEntry) + m.Load("Data", &x.Data) + m.Load("Control", &x.Control) + m.Load("Address", &x.Address) +} + +func (x *queueReceiver) beforeSave() {} +func (x *queueReceiver) save(m state.Map) { + x.beforeSave() + m.Save("readQueue", &x.readQueue) +} + +func (x *queueReceiver) afterLoad() {} +func (x *queueReceiver) load(m state.Map) { + m.Load("readQueue", &x.readQueue) +} + +func (x *streamQueueReceiver) beforeSave() {} +func (x *streamQueueReceiver) save(m state.Map) { + x.beforeSave() + m.Save("queueReceiver", &x.queueReceiver) + m.Save("buffer", &x.buffer) + m.Save("control", &x.control) + m.Save("addr", &x.addr) +} + +func (x *streamQueueReceiver) afterLoad() {} +func (x *streamQueueReceiver) load(m state.Map) { + m.Load("queueReceiver", &x.queueReceiver) + m.Load("buffer", &x.buffer) + m.Load("control", &x.control) + m.Load("addr", &x.addr) +} + +func (x *connectedEndpoint) beforeSave() {} +func (x *connectedEndpoint) save(m state.Map) { + x.beforeSave() + m.Save("endpoint", &x.endpoint) + m.Save("writeQueue", &x.writeQueue) +} + +func (x *connectedEndpoint) afterLoad() {} +func (x *connectedEndpoint) load(m state.Map) { + m.Load("endpoint", &x.endpoint) + m.Load("writeQueue", &x.writeQueue) +} + +func (x *baseEndpoint) beforeSave() {} +func (x *baseEndpoint) save(m state.Map) { + x.beforeSave() + m.Save("Queue", &x.Queue) + m.Save("passcred", &x.passcred) + m.Save("receiver", &x.receiver) + m.Save("connected", &x.connected) + m.Save("path", &x.path) +} + +func (x *baseEndpoint) afterLoad() {} +func (x *baseEndpoint) load(m state.Map) { + m.Load("Queue", &x.Queue) + m.Load("passcred", &x.passcred) + m.Load("receiver", &x.receiver) + m.Load("connected", &x.connected) + m.Load("path", &x.path) +} + +func init() { + state.Register("pkg/sentry/socket/unix/transport.connectionedEndpoint", (*connectionedEndpoint)(nil), state.Fns{Save: (*connectionedEndpoint).save, Load: (*connectionedEndpoint).load}) + state.Register("pkg/sentry/socket/unix/transport.connectionlessEndpoint", (*connectionlessEndpoint)(nil), state.Fns{Save: (*connectionlessEndpoint).save, Load: (*connectionlessEndpoint).load}) + state.Register("pkg/sentry/socket/unix/transport.queue", (*queue)(nil), state.Fns{Save: (*queue).save, Load: (*queue).load}) + state.Register("pkg/sentry/socket/unix/transport.messageList", (*messageList)(nil), state.Fns{Save: (*messageList).save, Load: (*messageList).load}) + state.Register("pkg/sentry/socket/unix/transport.messageEntry", (*messageEntry)(nil), state.Fns{Save: (*messageEntry).save, Load: (*messageEntry).load}) + state.Register("pkg/sentry/socket/unix/transport.ControlMessages", (*ControlMessages)(nil), state.Fns{Save: (*ControlMessages).save, Load: (*ControlMessages).load}) + state.Register("pkg/sentry/socket/unix/transport.message", (*message)(nil), state.Fns{Save: (*message).save, Load: (*message).load}) + state.Register("pkg/sentry/socket/unix/transport.queueReceiver", (*queueReceiver)(nil), state.Fns{Save: (*queueReceiver).save, Load: (*queueReceiver).load}) + state.Register("pkg/sentry/socket/unix/transport.streamQueueReceiver", (*streamQueueReceiver)(nil), state.Fns{Save: (*streamQueueReceiver).save, Load: (*streamQueueReceiver).load}) + state.Register("pkg/sentry/socket/unix/transport.connectedEndpoint", (*connectedEndpoint)(nil), state.Fns{Save: (*connectedEndpoint).save, Load: (*connectedEndpoint).load}) + state.Register("pkg/sentry/socket/unix/transport.baseEndpoint", (*baseEndpoint)(nil), state.Fns{Save: (*baseEndpoint).save, Load: (*baseEndpoint).load}) +} diff --git a/pkg/sentry/socket/unix/unix_state_autogen.go b/pkg/sentry/socket/unix/unix_state_autogen.go new file mode 100755 index 000000000..755373941 --- /dev/null +++ b/pkg/sentry/socket/unix/unix_state_autogen.go @@ -0,0 +1,28 @@ +// automatically generated by stateify. + +package unix + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *SocketOperations) beforeSave() {} +func (x *SocketOperations) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("SendReceiveTimeout", &x.SendReceiveTimeout) + m.Save("ep", &x.ep) + m.Save("stype", &x.stype) +} + +func (x *SocketOperations) afterLoad() {} +func (x *SocketOperations) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("SendReceiveTimeout", &x.SendReceiveTimeout) + m.Load("ep", &x.ep) + m.Load("stype", &x.stype) +} + +func init() { + state.Register("pkg/sentry/socket/unix.SocketOperations", (*SocketOperations)(nil), state.Fns{Save: (*SocketOperations).save, Load: (*SocketOperations).load}) +} diff --git a/pkg/sentry/state/BUILD b/pkg/sentry/state/BUILD deleted file mode 100644 index 0ea4aab8b..000000000 --- a/pkg/sentry/state/BUILD +++ /dev/null @@ -1,23 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "state", - srcs = [ - "state.go", - "state_metadata.go", - "state_unsafe.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/log", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/time", - "//pkg/sentry/watchdog", - "//pkg/state/statefile", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/state/state_state_autogen.go b/pkg/sentry/state/state_state_autogen.go new file mode 100755 index 000000000..6c2b29632 --- /dev/null +++ b/pkg/sentry/state/state_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package state diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD deleted file mode 100644 index 88d5db9fc..000000000 --- a/pkg/sentry/strace/BUILD +++ /dev/null @@ -1,45 +0,0 @@ -load("//tools:defs.bzl", "go_library", "proto_library") - -package(licenses = ["notice"]) - -go_library( - name = "strace", - srcs = [ - "capability.go", - "clone.go", - "epoll.go", - "futex.go", - "linux64_amd64.go", - "linux64_arm64.go", - "open.go", - "poll.go", - "ptrace.go", - "select.go", - "signal.go", - "socket.go", - "strace.go", - "syscalls.go", - ], - visibility = ["//:sandbox"], - deps = [ - ":strace_go_proto", - "//pkg/abi", - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/bits", - "//pkg/eventchannel", - "//pkg/seccomp", - "//pkg/sentry/arch", - "//pkg/sentry/kernel", - "//pkg/sentry/socket/netlink", - "//pkg/sentry/socket/netstack", - "//pkg/sentry/syscalls/linux", - "//pkg/usermem", - ], -) - -proto_library( - name = "strace", - srcs = ["strace.proto"], - visibility = ["//visibility:public"], -) diff --git a/pkg/sentry/strace/epoll.go b/pkg/sentry/strace/epoll.go index a6e48b836..a6e48b836 100644..100755 --- a/pkg/sentry/strace/epoll.go +++ b/pkg/sentry/strace/epoll.go diff --git a/pkg/sentry/strace/linux64_amd64.go b/pkg/sentry/strace/linux64_amd64.go index 71b92eaee..71b92eaee 100644..100755 --- a/pkg/sentry/strace/linux64_amd64.go +++ b/pkg/sentry/strace/linux64_amd64.go diff --git a/pkg/sentry/strace/linux64_arm64.go b/pkg/sentry/strace/linux64_arm64.go index bd7361a52..bd7361a52 100644..100755 --- a/pkg/sentry/strace/linux64_arm64.go +++ b/pkg/sentry/strace/linux64_arm64.go diff --git a/pkg/sentry/strace/select.go b/pkg/sentry/strace/select.go index 3a4c32aa0..3a4c32aa0 100644..100755 --- a/pkg/sentry/strace/select.go +++ b/pkg/sentry/strace/select.go diff --git a/pkg/sentry/strace/strace.proto b/pkg/sentry/strace/strace.proto deleted file mode 100644 index 906c52c51..000000000 --- a/pkg/sentry/strace/strace.proto +++ /dev/null @@ -1,49 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -syntax = "proto3"; - -package gvisor; - -message Strace { - // Process name that made the syscall. - string process = 1; - - // Syscall function name. - string function = 2; - - // List of syscall arguments formatted as strings. - repeated string args = 3; - - oneof info { - StraceEnter enter = 4; - StraceExit exit = 5; - } -} - -message StraceEnter {} - -message StraceExit { - // Return value formatted as string. - string return = 1; - - // Formatted error string in case syscall failed. - string error = 2; - - // Value of errno upon syscall exit. - int64 err_no = 3; // errno is a macro and gets expanded :-( - - // Time elapsed between syscall enter and exit. - int64 elapsed_ns = 4; -} diff --git a/pkg/sentry/strace/strace_amd64_state_autogen.go b/pkg/sentry/strace/strace_amd64_state_autogen.go new file mode 100755 index 000000000..c7d4b3eb4 --- /dev/null +++ b/pkg/sentry/strace/strace_amd64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build amd64 + +package strace diff --git a/pkg/sentry/strace/strace_arm64_state_autogen.go b/pkg/sentry/strace/strace_arm64_state_autogen.go new file mode 100755 index 000000000..9b8f66dc9 --- /dev/null +++ b/pkg/sentry/strace/strace_arm64_state_autogen.go @@ -0,0 +1,5 @@ +// automatically generated by stateify. + +// +build arm64 + +package strace diff --git a/pkg/sentry/strace/strace_go_proto/strace.pb.go b/pkg/sentry/strace/strace_go_proto/strace.pb.go new file mode 100755 index 000000000..ef45661bc --- /dev/null +++ b/pkg/sentry/strace/strace_go_proto/strace.pb.go @@ -0,0 +1,247 @@ +// Code generated by protoc-gen-go. DO NOT EDIT. +// source: pkg/sentry/strace/strace.proto + +package gvisor + +import ( + fmt "fmt" + proto "github.com/golang/protobuf/proto" + math "math" +) + +// Reference imports to suppress errors if they are not otherwise used. +var _ = proto.Marshal +var _ = fmt.Errorf +var _ = math.Inf + +// This is a compile-time assertion to ensure that this generated file +// is compatible with the proto package it is being compiled against. +// A compilation error at this line likely means your copy of the +// proto package needs to be updated. +const _ = proto.ProtoPackageIsVersion3 // please upgrade the proto package + +type Strace struct { + Process string `protobuf:"bytes,1,opt,name=process,proto3" json:"process,omitempty"` + Function string `protobuf:"bytes,2,opt,name=function,proto3" json:"function,omitempty"` + Args []string `protobuf:"bytes,3,rep,name=args,proto3" json:"args,omitempty"` + // Types that are valid to be assigned to Info: + // *Strace_Enter + // *Strace_Exit + Info isStrace_Info `protobuf_oneof:"info"` + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *Strace) Reset() { *m = Strace{} } +func (m *Strace) String() string { return proto.CompactTextString(m) } +func (*Strace) ProtoMessage() {} +func (*Strace) Descriptor() ([]byte, []int) { + return fileDescriptor_50c4b43677c82b5f, []int{0} +} + +func (m *Strace) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_Strace.Unmarshal(m, b) +} +func (m *Strace) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_Strace.Marshal(b, m, deterministic) +} +func (m *Strace) XXX_Merge(src proto.Message) { + xxx_messageInfo_Strace.Merge(m, src) +} +func (m *Strace) XXX_Size() int { + return xxx_messageInfo_Strace.Size(m) +} +func (m *Strace) XXX_DiscardUnknown() { + xxx_messageInfo_Strace.DiscardUnknown(m) +} + +var xxx_messageInfo_Strace proto.InternalMessageInfo + +func (m *Strace) GetProcess() string { + if m != nil { + return m.Process + } + return "" +} + +func (m *Strace) GetFunction() string { + if m != nil { + return m.Function + } + return "" +} + +func (m *Strace) GetArgs() []string { + if m != nil { + return m.Args + } + return nil +} + +type isStrace_Info interface { + isStrace_Info() +} + +type Strace_Enter struct { + Enter *StraceEnter `protobuf:"bytes,4,opt,name=enter,proto3,oneof"` +} + +type Strace_Exit struct { + Exit *StraceExit `protobuf:"bytes,5,opt,name=exit,proto3,oneof"` +} + +func (*Strace_Enter) isStrace_Info() {} + +func (*Strace_Exit) isStrace_Info() {} + +func (m *Strace) GetInfo() isStrace_Info { + if m != nil { + return m.Info + } + return nil +} + +func (m *Strace) GetEnter() *StraceEnter { + if x, ok := m.GetInfo().(*Strace_Enter); ok { + return x.Enter + } + return nil +} + +func (m *Strace) GetExit() *StraceExit { + if x, ok := m.GetInfo().(*Strace_Exit); ok { + return x.Exit + } + return nil +} + +// XXX_OneofWrappers is for the internal use of the proto package. +func (*Strace) XXX_OneofWrappers() []interface{} { + return []interface{}{ + (*Strace_Enter)(nil), + (*Strace_Exit)(nil), + } +} + +type StraceEnter struct { + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *StraceEnter) Reset() { *m = StraceEnter{} } +func (m *StraceEnter) String() string { return proto.CompactTextString(m) } +func (*StraceEnter) ProtoMessage() {} +func (*StraceEnter) Descriptor() ([]byte, []int) { + return fileDescriptor_50c4b43677c82b5f, []int{1} +} + +func (m *StraceEnter) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_StraceEnter.Unmarshal(m, b) +} +func (m *StraceEnter) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_StraceEnter.Marshal(b, m, deterministic) +} +func (m *StraceEnter) XXX_Merge(src proto.Message) { + xxx_messageInfo_StraceEnter.Merge(m, src) +} +func (m *StraceEnter) XXX_Size() int { + return xxx_messageInfo_StraceEnter.Size(m) +} +func (m *StraceEnter) XXX_DiscardUnknown() { + xxx_messageInfo_StraceEnter.DiscardUnknown(m) +} + +var xxx_messageInfo_StraceEnter proto.InternalMessageInfo + +type StraceExit struct { + Return string `protobuf:"bytes,1,opt,name=return,proto3" json:"return,omitempty"` + Error string `protobuf:"bytes,2,opt,name=error,proto3" json:"error,omitempty"` + ErrNo int64 `protobuf:"varint,3,opt,name=err_no,json=errNo,proto3" json:"err_no,omitempty"` + ElapsedNs int64 `protobuf:"varint,4,opt,name=elapsed_ns,json=elapsedNs,proto3" json:"elapsed_ns,omitempty"` + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *StraceExit) Reset() { *m = StraceExit{} } +func (m *StraceExit) String() string { return proto.CompactTextString(m) } +func (*StraceExit) ProtoMessage() {} +func (*StraceExit) Descriptor() ([]byte, []int) { + return fileDescriptor_50c4b43677c82b5f, []int{2} +} + +func (m *StraceExit) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_StraceExit.Unmarshal(m, b) +} +func (m *StraceExit) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_StraceExit.Marshal(b, m, deterministic) +} +func (m *StraceExit) XXX_Merge(src proto.Message) { + xxx_messageInfo_StraceExit.Merge(m, src) +} +func (m *StraceExit) XXX_Size() int { + return xxx_messageInfo_StraceExit.Size(m) +} +func (m *StraceExit) XXX_DiscardUnknown() { + xxx_messageInfo_StraceExit.DiscardUnknown(m) +} + +var xxx_messageInfo_StraceExit proto.InternalMessageInfo + +func (m *StraceExit) GetReturn() string { + if m != nil { + return m.Return + } + return "" +} + +func (m *StraceExit) GetError() string { + if m != nil { + return m.Error + } + return "" +} + +func (m *StraceExit) GetErrNo() int64 { + if m != nil { + return m.ErrNo + } + return 0 +} + +func (m *StraceExit) GetElapsedNs() int64 { + if m != nil { + return m.ElapsedNs + } + return 0 +} + +func init() { + proto.RegisterType((*Strace)(nil), "gvisor.Strace") + proto.RegisterType((*StraceEnter)(nil), "gvisor.StraceEnter") + proto.RegisterType((*StraceExit)(nil), "gvisor.StraceExit") +} + +func init() { proto.RegisterFile("pkg/sentry/strace/strace.proto", fileDescriptor_50c4b43677c82b5f) } + +var fileDescriptor_50c4b43677c82b5f = []byte{ + // 255 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0x5c, 0x90, 0xdd, 0x4a, 0xf4, 0x30, + 0x10, 0x86, 0xb7, 0x5f, 0xdb, 0x7c, 0x76, 0x16, 0x4f, 0xc6, 0x1f, 0x82, 0xa0, 0x94, 0x1e, 0x05, + 0x84, 0x2e, 0xe8, 0x1d, 0x08, 0xc2, 0x1e, 0xed, 0x41, 0xbc, 0x80, 0xa5, 0xd6, 0xd9, 0x12, 0x94, + 0x24, 0x4c, 0xb2, 0xb2, 0x5e, 0x96, 0x77, 0x28, 0xa6, 0xf1, 0x07, 0x8f, 0x92, 0x67, 0xde, 0x87, + 0x0c, 0x6f, 0xe0, 0xca, 0x3f, 0x4f, 0xab, 0x40, 0x36, 0xf2, 0xdb, 0x2a, 0x44, 0x1e, 0x46, 0xca, + 0x47, 0xef, 0xd9, 0x45, 0x87, 0x62, 0x7a, 0x35, 0xc1, 0x71, 0xf7, 0x5e, 0x80, 0x78, 0x48, 0x01, + 0x4a, 0xf8, 0xef, 0xd9, 0x8d, 0x14, 0x82, 0x2c, 0xda, 0x42, 0x35, 0xfa, 0x0b, 0xf1, 0x02, 0x8e, + 0x76, 0x7b, 0x3b, 0x46, 0xe3, 0xac, 0xfc, 0x97, 0xa2, 0x6f, 0x46, 0x84, 0x6a, 0xe0, 0x29, 0xc8, + 0xb2, 0x2d, 0x55, 0xa3, 0xd3, 0x1d, 0xaf, 0xa1, 0x26, 0x1b, 0x89, 0x65, 0xd5, 0x16, 0x6a, 0x79, + 0x73, 0xd2, 0xcf, 0xcb, 0xfa, 0x79, 0xd1, 0xfd, 0x67, 0xb4, 0x5e, 0xe8, 0xd9, 0x41, 0x05, 0x15, + 0x1d, 0x4c, 0x94, 0x75, 0x72, 0xf1, 0x8f, 0x7b, 0x30, 0x71, 0xbd, 0xd0, 0xc9, 0xb8, 0x13, 0x50, + 0x19, 0xbb, 0x73, 0xdd, 0x31, 0x2c, 0x7f, 0xbd, 0xd4, 0x79, 0x80, 0x1f, 0x19, 0xcf, 0x41, 0x30, + 0xc5, 0x3d, 0xdb, 0x5c, 0x22, 0x13, 0x9e, 0x42, 0x4d, 0xcc, 0x8e, 0x73, 0x81, 0x19, 0xf0, 0x0c, + 0x04, 0x31, 0x6f, 0xad, 0x93, 0x65, 0x5b, 0xa8, 0x32, 0x8d, 0x37, 0x0e, 0x2f, 0x01, 0xe8, 0x65, + 0xf0, 0x81, 0x9e, 0xb6, 0x36, 0xa4, 0x16, 0xa5, 0x6e, 0xf2, 0x64, 0x13, 0x1e, 0x45, 0xfa, 0xc3, + 0xdb, 0x8f, 0x00, 0x00, 0x00, 0xff, 0xff, 0x42, 0x9a, 0xbc, 0x81, 0x65, 0x01, 0x00, 0x00, +} diff --git a/pkg/sentry/strace/strace_state_autogen.go b/pkg/sentry/strace/strace_state_autogen.go new file mode 100755 index 000000000..33f6a7a54 --- /dev/null +++ b/pkg/sentry/strace/strace_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package strace diff --git a/pkg/sentry/syscalls/BUILD b/pkg/sentry/syscalls/BUILD deleted file mode 100644 index b8d1bd415..000000000 --- a/pkg/sentry/syscalls/BUILD +++ /dev/null @@ -1,21 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "syscalls", - srcs = [ - "epoll.go", - "syscalls.go", - ], - visibility = ["//:sandbox"], - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/arch", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/epoll", - "//pkg/sentry/kernel/time", - "//pkg/syserror", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD deleted file mode 100644 index 0d24fd3c4..000000000 --- a/pkg/sentry/syscalls/linux/BUILD +++ /dev/null @@ -1,104 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "linux", - srcs = [ - "error.go", - "flags.go", - "linux64.go", - "linux64_amd64.go", - "linux64_arm64.go", - "sigset.go", - "sys_aio.go", - "sys_capability.go", - "sys_clone_amd64.go", - "sys_clone_arm64.go", - "sys_epoll.go", - "sys_eventfd.go", - "sys_file.go", - "sys_futex.go", - "sys_getdents.go", - "sys_identity.go", - "sys_inotify.go", - "sys_lseek.go", - "sys_mempolicy.go", - "sys_mmap.go", - "sys_mount.go", - "sys_pipe.go", - "sys_poll.go", - "sys_prctl.go", - "sys_random.go", - "sys_read.go", - "sys_rlimit.go", - "sys_rseq.go", - "sys_rusage.go", - "sys_sched.go", - "sys_seccomp.go", - "sys_sem.go", - "sys_shm.go", - "sys_signal.go", - "sys_socket.go", - "sys_splice.go", - "sys_stat.go", - "sys_stat_amd64.go", - "sys_stat_arm64.go", - "sys_sync.go", - "sys_sysinfo.go", - "sys_syslog.go", - "sys_thread.go", - "sys_time.go", - "sys_timer.go", - "sys_timerfd.go", - "sys_tls.go", - "sys_utsname.go", - "sys_write.go", - "sys_xattr.go", - "timespec.go", - ], - visibility = ["//:sandbox"], - deps = [ - "//pkg/abi", - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/bpf", - "//pkg/context", - "//pkg/log", - "//pkg/metric", - "//pkg/rand", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/lock", - "//pkg/sentry/fs/timerfd", - "//pkg/sentry/fs/tmpfs", - "//pkg/sentry/fsbridge", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/epoll", - "//pkg/sentry/kernel/eventfd", - "//pkg/sentry/kernel/fasync", - "//pkg/sentry/kernel/pipe", - "//pkg/sentry/kernel/sched", - "//pkg/sentry/kernel/shm", - "//pkg/sentry/kernel/signalfd", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/loader", - "//pkg/sentry/memmap", - "//pkg/sentry/mm", - "//pkg/sentry/socket", - "//pkg/sentry/socket/control", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/syscalls", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 79066ad2a..79066ad2a 100644..100755 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index 7421619de..7421619de 100644..100755 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go diff --git a/pkg/sentry/syscalls/linux/linux_amd64_state_autogen.go b/pkg/sentry/syscalls/linux/linux_amd64_state_autogen.go new file mode 100755 index 000000000..a98193a5b --- /dev/null +++ b/pkg/sentry/syscalls/linux/linux_amd64_state_autogen.go @@ -0,0 +1,6 @@ +// automatically generated by stateify. + +// +build amd64 +// +build amd64 + +package linux diff --git a/pkg/sentry/syscalls/linux/linux_arm64_state_autogen.go b/pkg/sentry/syscalls/linux/linux_arm64_state_autogen.go new file mode 100755 index 000000000..b144adbda --- /dev/null +++ b/pkg/sentry/syscalls/linux/linux_arm64_state_autogen.go @@ -0,0 +1,6 @@ +// automatically generated by stateify. + +// +build arm64 +// +build arm64 + +package linux diff --git a/pkg/sentry/syscalls/linux/linux_state_autogen.go b/pkg/sentry/syscalls/linux/linux_state_autogen.go new file mode 100755 index 000000000..5107dd246 --- /dev/null +++ b/pkg/sentry/syscalls/linux/linux_state_autogen.go @@ -0,0 +1,83 @@ +// automatically generated by stateify. + +// +build amd64 +// +build amd64 arm64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *ioEvent) beforeSave() {} +func (x *ioEvent) save(m state.Map) { + x.beforeSave() + m.Save("Data", &x.Data) + m.Save("Obj", &x.Obj) + m.Save("Result", &x.Result) + m.Save("Result2", &x.Result2) +} + +func (x *ioEvent) afterLoad() {} +func (x *ioEvent) load(m state.Map) { + m.Load("Data", &x.Data) + m.Load("Obj", &x.Obj) + m.Load("Result", &x.Result) + m.Load("Result2", &x.Result2) +} + +func (x *futexWaitRestartBlock) beforeSave() {} +func (x *futexWaitRestartBlock) save(m state.Map) { + x.beforeSave() + m.Save("duration", &x.duration) + m.Save("addr", &x.addr) + m.Save("private", &x.private) + m.Save("val", &x.val) + m.Save("mask", &x.mask) +} + +func (x *futexWaitRestartBlock) afterLoad() {} +func (x *futexWaitRestartBlock) load(m state.Map) { + m.Load("duration", &x.duration) + m.Load("addr", &x.addr) + m.Load("private", &x.private) + m.Load("val", &x.val) + m.Load("mask", &x.mask) +} + +func (x *pollRestartBlock) beforeSave() {} +func (x *pollRestartBlock) save(m state.Map) { + x.beforeSave() + m.Save("pfdAddr", &x.pfdAddr) + m.Save("nfds", &x.nfds) + m.Save("timeout", &x.timeout) +} + +func (x *pollRestartBlock) afterLoad() {} +func (x *pollRestartBlock) load(m state.Map) { + m.Load("pfdAddr", &x.pfdAddr) + m.Load("nfds", &x.nfds) + m.Load("timeout", &x.timeout) +} + +func (x *clockNanosleepRestartBlock) beforeSave() {} +func (x *clockNanosleepRestartBlock) save(m state.Map) { + x.beforeSave() + m.Save("c", &x.c) + m.Save("duration", &x.duration) + m.Save("rem", &x.rem) +} + +func (x *clockNanosleepRestartBlock) afterLoad() {} +func (x *clockNanosleepRestartBlock) load(m state.Map) { + m.Load("c", &x.c) + m.Load("duration", &x.duration) + m.Load("rem", &x.rem) +} + +func init() { + state.Register("pkg/sentry/syscalls/linux.ioEvent", (*ioEvent)(nil), state.Fns{Save: (*ioEvent).save, Load: (*ioEvent).load}) + state.Register("pkg/sentry/syscalls/linux.futexWaitRestartBlock", (*futexWaitRestartBlock)(nil), state.Fns{Save: (*futexWaitRestartBlock).save, Load: (*futexWaitRestartBlock).load}) + state.Register("pkg/sentry/syscalls/linux.pollRestartBlock", (*pollRestartBlock)(nil), state.Fns{Save: (*pollRestartBlock).save, Load: (*pollRestartBlock).load}) + state.Register("pkg/sentry/syscalls/linux.clockNanosleepRestartBlock", (*clockNanosleepRestartBlock)(nil), state.Fns{Save: (*clockNanosleepRestartBlock).save, Load: (*clockNanosleepRestartBlock).load}) +} diff --git a/pkg/sentry/syscalls/linux/sys_clone_amd64.go b/pkg/sentry/syscalls/linux/sys_clone_amd64.go index dd43cf18d..dd43cf18d 100644..100755 --- a/pkg/sentry/syscalls/linux/sys_clone_amd64.go +++ b/pkg/sentry/syscalls/linux/sys_clone_amd64.go diff --git a/pkg/sentry/syscalls/linux/sys_clone_arm64.go b/pkg/sentry/syscalls/linux/sys_clone_arm64.go index cf68a8949..cf68a8949 100644..100755 --- a/pkg/sentry/syscalls/linux/sys_clone_arm64.go +++ b/pkg/sentry/syscalls/linux/sys_clone_arm64.go diff --git a/pkg/sentry/syscalls/linux/sys_rseq.go b/pkg/sentry/syscalls/linux/sys_rseq.go index 90db10ea6..90db10ea6 100644..100755 --- a/pkg/sentry/syscalls/linux/sys_rseq.go +++ b/pkg/sentry/syscalls/linux/sys_rseq.go diff --git a/pkg/sentry/syscalls/linux/sys_stat_amd64.go b/pkg/sentry/syscalls/linux/sys_stat_amd64.go index 0a04a6113..0a04a6113 100644..100755 --- a/pkg/sentry/syscalls/linux/sys_stat_amd64.go +++ b/pkg/sentry/syscalls/linux/sys_stat_amd64.go diff --git a/pkg/sentry/syscalls/linux/sys_stat_arm64.go b/pkg/sentry/syscalls/linux/sys_stat_arm64.go index 5a3b1bfad..5a3b1bfad 100644..100755 --- a/pkg/sentry/syscalls/linux/sys_stat_arm64.go +++ b/pkg/sentry/syscalls/linux/sys_stat_arm64.go diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 2de5e3422..2de5e3422 100644..100755 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD deleted file mode 100644 index 2eb210014..000000000 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ /dev/null @@ -1,53 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "vfs2", - srcs = [ - "epoll.go", - "epoll_unsafe.go", - "execve.go", - "fd.go", - "filesystem.go", - "fscontext.go", - "getdents.go", - "ioctl.go", - "linux64.go", - "linux64_override_amd64.go", - "linux64_override_arm64.go", - "mmap.go", - "path.go", - "poll.go", - "read_write.go", - "setstat.go", - "stat.go", - "stat_amd64.go", - "stat_arm64.go", - "sync.go", - "xattr.go", - ], - marshal = True, - visibility = ["//:sandbox"], - deps = [ - "//pkg/abi/linux", - "//pkg/bits", - "//pkg/fspath", - "//pkg/gohacks", - "//pkg/sentry/arch", - "//pkg/sentry/fsbridge", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/loader", - "//pkg/sentry/memmap", - "//pkg/sentry/syscalls", - "//pkg/sentry/syscalls/linux", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll.go b/pkg/sentry/syscalls/linux/vfs2/epoll.go index d6cb0e79a..d6cb0e79a 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/epoll.go +++ b/pkg/sentry/syscalls/linux/vfs2/epoll.go diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go b/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go index 825f325bf..825f325bf 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go +++ b/pkg/sentry/syscalls/linux/vfs2/epoll_unsafe.go diff --git a/pkg/sentry/syscalls/linux/vfs2/execve.go b/pkg/sentry/syscalls/linux/vfs2/execve.go index aef0078a8..aef0078a8 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/execve.go +++ b/pkg/sentry/syscalls/linux/vfs2/execve.go diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 3afcea665..3afcea665 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index a859095e2..a859095e2 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go diff --git a/pkg/sentry/syscalls/linux/vfs2/fscontext.go b/pkg/sentry/syscalls/linux/vfs2/fscontext.go index 317409a18..317409a18 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/fscontext.go +++ b/pkg/sentry/syscalls/linux/vfs2/fscontext.go diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go index a61cc5059..a61cc5059 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/getdents.go +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go index 5a2418da9..5a2418da9 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64.go b/pkg/sentry/syscalls/linux/vfs2/linux64.go index 19ee36081..19ee36081 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/linux64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64.go diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go index 7d220bc20..7d220bc20 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_amd64.go diff --git a/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go b/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go index a6b367468..a6b367468 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go +++ b/pkg/sentry/syscalls/linux/vfs2/linux64_override_arm64.go diff --git a/pkg/sentry/syscalls/linux/vfs2/mmap.go b/pkg/sentry/syscalls/linux/vfs2/mmap.go index 60a43f0a0..60a43f0a0 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/mmap.go +++ b/pkg/sentry/syscalls/linux/vfs2/mmap.go diff --git a/pkg/sentry/syscalls/linux/vfs2/path.go b/pkg/sentry/syscalls/linux/vfs2/path.go index 97da6c647..97da6c647 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/path.go +++ b/pkg/sentry/syscalls/linux/vfs2/path.go diff --git a/pkg/sentry/syscalls/linux/vfs2/poll.go b/pkg/sentry/syscalls/linux/vfs2/poll.go index dbf4882da..dbf4882da 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/poll.go +++ b/pkg/sentry/syscalls/linux/vfs2/poll.go diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index 35f6308d6..35f6308d6 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 136453ccc..136453ccc 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index fdfe49243..fdfe49243 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go diff --git a/pkg/sentry/syscalls/linux/vfs2/stat_amd64.go b/pkg/sentry/syscalls/linux/vfs2/stat_amd64.go index 2da538fc6..2da538fc6 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/stat_amd64.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat_amd64.go diff --git a/pkg/sentry/syscalls/linux/vfs2/stat_arm64.go b/pkg/sentry/syscalls/linux/vfs2/stat_arm64.go index 88b9c7627..88b9c7627 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/stat_arm64.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat_arm64.go diff --git a/pkg/sentry/syscalls/linux/vfs2/sync.go b/pkg/sentry/syscalls/linux/vfs2/sync.go index 365250b0b..365250b0b 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/sync.go +++ b/pkg/sentry/syscalls/linux/vfs2/sync.go diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2_abi_autogen_unsafe.go b/pkg/sentry/syscalls/linux/vfs2/vfs2_abi_autogen_unsafe.go new file mode 100755 index 000000000..fb2182415 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2_abi_autogen_unsafe.go @@ -0,0 +1,122 @@ +// Automatically generated marshal implementation. See tools/go_marshal. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/safecopy" + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "io" + "reflect" + "runtime" + "unsafe" +) + +// Marshallable types used by this file. +var _ marshal.Marshallable = (*sigSetWithSize)(nil) + +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (s *sigSetWithSize) SizeBytes() int { + return 16 +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (s *sigSetWithSize) MarshalBytes(dst []byte) { + usermem.ByteOrder.PutUint64(dst[:8], uint64(s.sigsetAddr)) + dst = dst[8:] + usermem.ByteOrder.PutUint64(dst[:8], uint64(s.sizeofSigset)) + dst = dst[8:] +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (s *sigSetWithSize) UnmarshalBytes(src []byte) { + s.sigsetAddr = uint64(usermem.ByteOrder.Uint64(src[:8])) + src = src[8:] + s.sizeofSigset = uint64(usermem.ByteOrder.Uint64(src[:8])) + src = src[8:] +} + +// Packed implements marshal.Marshallable.Packed. +func (s *sigSetWithSize) Packed() bool { + return true +} + +// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. +func (s *sigSetWithSize) MarshalUnsafe(dst []byte) { + safecopy.CopyIn(dst, unsafe.Pointer(s)) +} + +// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. +func (s *sigSetWithSize) UnmarshalUnsafe(src []byte) { + safecopy.CopyOut(unsafe.Pointer(s), src) +} + +// CopyOut implements marshal.Marshallable.CopyOut. +func (s *sigSetWithSize) CopyOut(task marshal.Task, addr usermem.Addr) error { + // Bypass escape analysis on s. The no-op arithmetic operation on the + // pointer makes the compiler think val doesn't depend on s. + // See src/runtime/stubs.go:noescape() in the golang toolchain. + ptr := unsafe.Pointer(s) + val := uintptr(ptr) + val = val^0 + + // Construct a slice backed by s's underlying memory. + var buf []byte + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf)) + hdr.Data = val + hdr.Len = s.SizeBytes() + hdr.Cap = s.SizeBytes() + + _, err := task.CopyOutBytes(addr, buf) + // Since we bypassed the compiler's escape analysis, indicate that s + // must live until after the CopyOutBytes. + runtime.KeepAlive(s) + return err +} + +// CopyIn implements marshal.Marshallable.CopyIn. +func (s *sigSetWithSize) CopyIn(task marshal.Task, addr usermem.Addr) error { + // Bypass escape analysis on s. The no-op arithmetic operation on the + // pointer makes the compiler think val doesn't depend on s. + // See src/runtime/stubs.go:noescape() in the golang toolchain. + ptr := unsafe.Pointer(s) + val := uintptr(ptr) + val = val^0 + + // Construct a slice backed by s's underlying memory. + var buf []byte + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf)) + hdr.Data = val + hdr.Len = s.SizeBytes() + hdr.Cap = s.SizeBytes() + + _, err := task.CopyInBytes(addr, buf) + // Since we bypassed the compiler's escape analysis, indicate that s + // must live until after the CopyInBytes. + runtime.KeepAlive(s) + return err +} + +// WriteTo implements io.WriterTo.WriteTo. +func (s *sigSetWithSize) WriteTo(w io.Writer) (int64, error) { + // Bypass escape analysis on s. The no-op arithmetic operation on the + // pointer makes the compiler think val doesn't depend on s. + // See src/runtime/stubs.go:noescape() in the golang toolchain. + ptr := unsafe.Pointer(s) + val := uintptr(ptr) + val = val^0 + + // Construct a slice backed by s's underlying memory. + var buf []byte + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&buf)) + hdr.Data = val + hdr.Len = s.SizeBytes() + hdr.Cap = s.SizeBytes() + + len, err := w.Write(buf) + // Since we bypassed the compiler's escape analysis, indicate that s + // must live until after the Write. + runtime.KeepAlive(s) + return int64(len), err +} + diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2_amd64_abi_autogen_unsafe.go b/pkg/sentry/syscalls/linux/vfs2/vfs2_amd64_abi_autogen_unsafe.go new file mode 100755 index 000000000..fc1b597de --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2_amd64_abi_autogen_unsafe.go @@ -0,0 +1,10 @@ +// Automatically generated marshal implementation. See tools/go_marshal. + +// +build amd64 +// +build amd64 + +package vfs2 + +import ( +) + diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2_amd64_state_autogen.go b/pkg/sentry/syscalls/linux/vfs2/vfs2_amd64_state_autogen.go new file mode 100755 index 000000000..b44f63872 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2_amd64_state_autogen.go @@ -0,0 +1,6 @@ +// automatically generated by stateify. + +// +build amd64 +// +build amd64 + +package vfs2 diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2_arm64_abi_autogen_unsafe.go b/pkg/sentry/syscalls/linux/vfs2/vfs2_arm64_abi_autogen_unsafe.go new file mode 100755 index 000000000..c2958877a --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2_arm64_abi_autogen_unsafe.go @@ -0,0 +1,10 @@ +// Automatically generated marshal implementation. See tools/go_marshal. + +// +build arm64 +// +build arm64 + +package vfs2 + +import ( +) + diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2_arm64_state_autogen.go b/pkg/sentry/syscalls/linux/vfs2/vfs2_arm64_state_autogen.go new file mode 100755 index 000000000..b61fa85e9 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2_arm64_state_autogen.go @@ -0,0 +1,6 @@ +// automatically generated by stateify. + +// +build arm64 +// +build arm64 + +package vfs2 diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2_state_autogen.go b/pkg/sentry/syscalls/linux/vfs2/vfs2_state_autogen.go new file mode 100755 index 000000000..570100331 --- /dev/null +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2_state_autogen.go @@ -0,0 +1,26 @@ +// automatically generated by stateify. + +package vfs2 + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *pollRestartBlock) beforeSave() {} +func (x *pollRestartBlock) save(m state.Map) { + x.beforeSave() + m.Save("pfdAddr", &x.pfdAddr) + m.Save("nfds", &x.nfds) + m.Save("timeout", &x.timeout) +} + +func (x *pollRestartBlock) afterLoad() {} +func (x *pollRestartBlock) load(m state.Map) { + m.Load("pfdAddr", &x.pfdAddr) + m.Load("nfds", &x.nfds) + m.Load("timeout", &x.timeout) +} + +func init() { + state.Register("pkg/sentry/syscalls/linux/vfs2.pollRestartBlock", (*pollRestartBlock)(nil), state.Fns{Save: (*pollRestartBlock).save, Load: (*pollRestartBlock).load}) +} diff --git a/pkg/sentry/syscalls/linux/vfs2/xattr.go b/pkg/sentry/syscalls/linux/vfs2/xattr.go index 89e9ff4d7..89e9ff4d7 100644..100755 --- a/pkg/sentry/syscalls/linux/vfs2/xattr.go +++ b/pkg/sentry/syscalls/linux/vfs2/xattr.go diff --git a/pkg/sentry/syscalls/syscalls_state_autogen.go b/pkg/sentry/syscalls/syscalls_state_autogen.go new file mode 100755 index 000000000..b577e39a3 --- /dev/null +++ b/pkg/sentry/syscalls/syscalls_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package syscalls diff --git a/pkg/sentry/time/BUILD b/pkg/sentry/time/BUILD deleted file mode 100644 index 04f81a35b..000000000 --- a/pkg/sentry/time/BUILD +++ /dev/null @@ -1,50 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "seqatomic_parameters", - out = "seqatomic_parameters_unsafe.go", - package = "time", - suffix = "Parameters", - template = "//pkg/sync:generic_seqatomic", - types = { - "Value": "Parameters", - }, -) - -go_library( - name = "time", - srcs = [ - "arith_arm64.go", - "calibrated_clock.go", - "clock_id.go", - "clocks.go", - "muldiv_amd64.s", - "muldiv_arm64.s", - "parameters.go", - "sampler.go", - "sampler_unsafe.go", - "seqatomic_parameters_unsafe.go", - "tsc_amd64.s", - "tsc_arm64.s", - ], - visibility = ["//:sandbox"], - deps = [ - "//pkg/log", - "//pkg/metric", - "//pkg/sync", - "//pkg/syserror", - ], -) - -go_test( - name = "time_test", - srcs = [ - "calibrated_clock_test.go", - "parameters_test.go", - "sampler_test.go", - ], - library = ":time", -) diff --git a/pkg/sentry/time/LICENSE b/pkg/sentry/time/LICENSE deleted file mode 100644 index 6a66aea5e..000000000 --- a/pkg/sentry/time/LICENSE +++ /dev/null @@ -1,27 +0,0 @@ -Copyright (c) 2009 The Go Authors. All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are -met: - - * Redistributions of source code must retain the above copyright -notice, this list of conditions and the following disclaimer. - * Redistributions in binary form must reproduce the above -copyright notice, this list of conditions and the following disclaimer -in the documentation and/or other materials provided with the -distribution. - * Neither the name of Google Inc. nor the names of its -contributors may be used to endorse or promote products derived from -this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/pkg/sentry/time/calibrated_clock_test.go b/pkg/sentry/time/calibrated_clock_test.go deleted file mode 100644 index d6622bfe2..000000000 --- a/pkg/sentry/time/calibrated_clock_test.go +++ /dev/null @@ -1,186 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package time - -import ( - "testing" - "time" -) - -// newTestCalibratedClock returns a CalibratedClock that collects samples from -// the given sample list and cycle counts from the given cycle list. -func newTestCalibratedClock(samples []sample, cycles []TSCValue) *CalibratedClock { - return &CalibratedClock{ - ref: newTestSampler(samples, cycles), - } -} - -func TestConstantFrequency(t *testing.T) { - // Perfectly constant frequency. - samples := []sample{ - {before: 100000, after: 100000 + defaultOverheadCycles, ref: 100}, - {before: 200000, after: 200000 + defaultOverheadCycles, ref: 200}, - {before: 300000, after: 300000 + defaultOverheadCycles, ref: 300}, - {before: 400000, after: 400000 + defaultOverheadCycles, ref: 400}, - {before: 500000, after: 500000 + defaultOverheadCycles, ref: 500}, - {before: 600000, after: 600000 + defaultOverheadCycles, ref: 600}, - {before: 700000, after: 700000 + defaultOverheadCycles, ref: 700}, - } - - c := newTestCalibratedClock(samples, nil) - - // Update from all samples. - for range samples { - c.Update() - } - - c.mu.RLock() - if !c.ready { - c.mu.RUnlock() - t.Fatalf("clock not ready") - } - // A bit after the last sample. - now, ok := c.params.ComputeTime(750000) - c.mu.RUnlock() - if !ok { - t.Fatalf("ComputeTime ok got %v want true", ok) - } - - t.Logf("now: %v", now) - - // Time should be between the current sample and where we'd expect the - // next sample. - if now < 700 || now > 800 { - t.Errorf("now got %v want > 700 && < 800", now) - } -} - -func TestErrorCorrection(t *testing.T) { - testCases := []struct { - name string - samples [5]sample - projectedTimeStart int64 - projectedTimeEnd int64 - }{ - // Initial calibration should be ~1MHz for each of these, and - // the reference clock changes in samples[2]. - { - name: "slow-down", - samples: [5]sample{ - {before: 1000000, after: 1000001, ref: ReferenceNS(1 * ApproxUpdateInterval.Nanoseconds())}, - {before: 2000000, after: 2000001, ref: ReferenceNS(2 * ApproxUpdateInterval.Nanoseconds())}, - // Reference clock has slowed down, causing 100ms of error. - {before: 3010000, after: 3010001, ref: ReferenceNS(3 * ApproxUpdateInterval.Nanoseconds())}, - {before: 4020000, after: 4020001, ref: ReferenceNS(4 * ApproxUpdateInterval.Nanoseconds())}, - {before: 5030000, after: 5030001, ref: ReferenceNS(5 * ApproxUpdateInterval.Nanoseconds())}, - }, - projectedTimeStart: 3005 * time.Millisecond.Nanoseconds(), - projectedTimeEnd: 3015 * time.Millisecond.Nanoseconds(), - }, - { - name: "speed-up", - samples: [5]sample{ - {before: 1000000, after: 1000001, ref: ReferenceNS(1 * ApproxUpdateInterval.Nanoseconds())}, - {before: 2000000, after: 2000001, ref: ReferenceNS(2 * ApproxUpdateInterval.Nanoseconds())}, - // Reference clock has sped up, causing 100ms of error. - {before: 2990000, after: 2990001, ref: ReferenceNS(3 * ApproxUpdateInterval.Nanoseconds())}, - {before: 3980000, after: 3980001, ref: ReferenceNS(4 * ApproxUpdateInterval.Nanoseconds())}, - {before: 4970000, after: 4970001, ref: ReferenceNS(5 * ApproxUpdateInterval.Nanoseconds())}, - }, - projectedTimeStart: 2985 * time.Millisecond.Nanoseconds(), - projectedTimeEnd: 2995 * time.Millisecond.Nanoseconds(), - }, - } - for _, tc := range testCases { - t.Run(tc.name, func(t *testing.T) { - c := newTestCalibratedClock(tc.samples[:], nil) - - // Initial calibration takes two updates. - _, ok := c.Update() - if ok { - t.Fatalf("Update ready too early") - } - - params, ok := c.Update() - if !ok { - t.Fatalf("Update not ready") - } - - // Initial calibration is ~1MHz. - hz := params.Frequency - if hz < 990000 || hz > 1010000 { - t.Fatalf("Frequency got %v want > 990kHz && < 1010kHz", hz) - } - - // Project time at the next update. Given the 1MHz - // calibration, it is expected to be ~3.1s/2.9s, not - // the actual 3s. - // - // N.B. the next update time is the "after" time above. - projected, ok := params.ComputeTime(tc.samples[2].after) - if !ok { - t.Fatalf("ComputeTime ok got %v want true", ok) - } - if projected < tc.projectedTimeStart || projected > tc.projectedTimeEnd { - t.Fatalf("ComputeTime(%v) got %v want > %v && < %v", tc.samples[2].after, projected, tc.projectedTimeStart, tc.projectedTimeEnd) - } - - // Update again to see the changed reference clock. - params, ok = c.Update() - if !ok { - t.Fatalf("Update not ready") - } - - // We now know that TSC = tc.samples[2].after -> 3s, - // but with the previous params indicated that TSC - // tc.samples[2].after -> 3.5s/2.5s. We can't allow the - // clock to go backwards, and having the clock jump - // forwards is undesirable. There should be a smooth - // transition that corrects the clock error over time. - // Check that the clock is continuous at TSC = - // tc.samples[2].after. - newProjected, ok := params.ComputeTime(tc.samples[2].after) - if !ok { - t.Fatalf("ComputeTime ok got %v want true", ok) - } - if newProjected != projected { - t.Errorf("Discontinuous time; ComputeTime(%v) got %v want %v", tc.samples[2].after, newProjected, projected) - } - - // As the reference clock stablizes, ensure that the clock error - // decreases. - initialErr := c.errorNS - t.Logf("initial error: %v ns", initialErr) - - _, ok = c.Update() - if !ok { - t.Fatalf("Update not ready") - } - if c.errorNS.Magnitude() > initialErr.Magnitude() { - t.Errorf("errorNS increased, got %v want |%v| <= |%v|", c.errorNS, c.errorNS, initialErr) - } - - _, ok = c.Update() - if !ok { - t.Fatalf("Update not ready") - } - if c.errorNS.Magnitude() > initialErr.Magnitude() { - t.Errorf("errorNS increased, got %v want |%v| <= |%v|", c.errorNS, c.errorNS, initialErr) - } - - t.Logf("final error: %v ns", c.errorNS) - }) - } -} diff --git a/pkg/sentry/time/parameters_test.go b/pkg/sentry/time/parameters_test.go deleted file mode 100644 index e1b9084ac..000000000 --- a/pkg/sentry/time/parameters_test.go +++ /dev/null @@ -1,486 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package time - -import ( - "math" - "testing" - "time" -) - -func TestParametersComputeTime(t *testing.T) { - testCases := []struct { - name string - params Parameters - now TSCValue - want int64 - }{ - { - // Now is the same as the base cycles. - name: "base-cycles", - params: Parameters{ - BaseCycles: 10000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - now: 10000, - want: 5000 * time.Millisecond.Nanoseconds(), - }, - { - // Now is the behind the base cycles. Time is frozen. - name: "backwards", - params: Parameters{ - BaseCycles: 10000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - now: 9000, - want: 5000 * time.Millisecond.Nanoseconds(), - }, - { - // Now is ahead of the base cycles. - name: "ahead", - params: Parameters{ - BaseCycles: 10000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - now: 15000, - want: 5500 * time.Millisecond.Nanoseconds(), - }, - } - for _, tc := range testCases { - t.Run(tc.name, func(t *testing.T) { - got, ok := tc.params.ComputeTime(tc.now) - if !ok { - t.Errorf("ComputeTime ok got %v want true", got) - } - if got != tc.want { - t.Errorf("ComputeTime got %+v want %+v", got, tc.want) - } - }) - } -} - -func TestParametersErrorAdjust(t *testing.T) { - testCases := []struct { - name string - oldParams Parameters - now TSCValue - newParams Parameters - want Parameters - errorNS ReferenceNS - wantErr bool - }{ - { - // newParams are perfectly continuous with oldParams - // and don't need adjustment. - name: "continuous", - oldParams: Parameters{ - BaseCycles: 0, - BaseRef: 0, - Frequency: 10000, - }, - now: 50000, - newParams: Parameters{ - BaseCycles: 50000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - want: Parameters{ - BaseCycles: 50000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - }, - { - // Same as "continuous", but with now ahead of - // newParams.BaseCycles. The result is the same as - // there is no error to correct. - name: "continuous-nowdiff", - oldParams: Parameters{ - BaseCycles: 0, - BaseRef: 0, - Frequency: 10000, - }, - now: 60000, - newParams: Parameters{ - BaseCycles: 50000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - want: Parameters{ - BaseCycles: 50000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - }, - { - // errorAdjust bails out if the TSC goes backwards. - name: "tsc-backwards", - oldParams: Parameters{ - BaseCycles: 10000, - BaseRef: ReferenceNS(1000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - now: 9000, - newParams: Parameters{ - BaseCycles: 9000, - BaseRef: ReferenceNS(1100 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - wantErr: true, - }, - { - // errorAdjust bails out if new params are from after now. - name: "params-after-now", - oldParams: Parameters{ - BaseCycles: 10000, - BaseRef: ReferenceNS(1000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - now: 11000, - newParams: Parameters{ - BaseCycles: 12000, - BaseRef: ReferenceNS(1200 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - wantErr: true, - }, - { - // Host clock sped up. - name: "speed-up", - oldParams: Parameters{ - BaseCycles: 0, - BaseRef: 0, - Frequency: 10000, - }, - now: 45000, - // Host frequency changed to 9000 immediately after - // oldParams was returned. - newParams: Parameters{ - BaseCycles: 45000, - // From oldParams, we think ref = 4.5s at cycles = 45000. - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 9000, - }, - want: Parameters{ - BaseCycles: 45000, - BaseRef: ReferenceNS(4500 * time.Millisecond.Nanoseconds()), - // We must decrease the new frequency by 50% to - // correct 0.5s of error in 1s - // (ApproxUpdateInterval). - Frequency: 4500, - }, - errorNS: ReferenceNS(-500 * time.Millisecond.Nanoseconds()), - }, - { - // Host clock sped up, with now ahead of newParams. - name: "speed-up-nowdiff", - oldParams: Parameters{ - BaseCycles: 0, - BaseRef: 0, - Frequency: 10000, - }, - now: 50000, - // Host frequency changed to 9000 immediately after - // oldParams was returned. - newParams: Parameters{ - BaseCycles: 45000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 9000, - }, - // nextRef = 6000ms - // nextCycles = 9000 * (6000ms - 5000ms) + 45000 - // nextCycles = 9000 * (1s) + 45000 - // nextCycles = 54000 - // f = (54000 - 50000) / 1s = 4000 - // - // ref = 5000ms - (50000 - 45000) / 4000 - // ref = 3.75s - want: Parameters{ - BaseCycles: 45000, - BaseRef: ReferenceNS(3750 * time.Millisecond.Nanoseconds()), - Frequency: 4000, - }, - // oldNow = 50000 * 10000 = 5s - // newNow = (50000 - 45000) / 9000 + 5s = 5.555s - errorNS: ReferenceNS((5000*time.Millisecond - 5555555555).Nanoseconds()), - }, - { - // Host clock sped up. The new parameters are so far - // ahead that the next update time already passed. - name: "speed-up-uncorrectable-baseref", - oldParams: Parameters{ - BaseCycles: 0, - BaseRef: 0, - Frequency: 10000, - }, - now: 50000, - // Host frequency changed to 5000 immediately after - // oldParams was returned. - newParams: Parameters{ - BaseCycles: 45000, - BaseRef: ReferenceNS(9000 * time.Millisecond.Nanoseconds()), - Frequency: 5000, - }, - // The next update should be at 10s, but newParams - // already passed 6s. Thus it is impossible to correct - // the clock by then. - wantErr: true, - }, - { - // Host clock sped up. The new parameters are moving so - // fast that the next update should be before now. - name: "speed-up-uncorrectable-frequency", - oldParams: Parameters{ - BaseCycles: 0, - BaseRef: 0, - Frequency: 10000, - }, - now: 55000, - // Host frequency changed to 7500 immediately after - // oldParams was returned. - newParams: Parameters{ - BaseCycles: 45000, - BaseRef: ReferenceNS(6000 * time.Millisecond.Nanoseconds()), - Frequency: 7500, - }, - // The next update should be at 6.5s, but newParams are - // so far ahead and fast that they reach 6.5s at cycle - // 48750, which before now! Thus it is impossible to - // correct the clock by then. - wantErr: true, - }, - { - // Host clock slowed down. - name: "slow-down", - oldParams: Parameters{ - BaseCycles: 0, - BaseRef: 0, - Frequency: 10000, - }, - now: 55000, - // Host frequency changed to 11000 immediately after - // oldParams was returned. - newParams: Parameters{ - BaseCycles: 55000, - // From oldParams, we think ref = 5.5s at cycles = 55000. - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 11000, - }, - want: Parameters{ - BaseCycles: 55000, - BaseRef: ReferenceNS(5500 * time.Millisecond.Nanoseconds()), - // We must increase the new frequency by 50% to - // correct 0.5s of error in 1s - // (ApproxUpdateInterval). - Frequency: 16500, - }, - errorNS: ReferenceNS(500 * time.Millisecond.Nanoseconds()), - }, - { - // Host clock slowed down, with now ahead of newParams. - name: "slow-down-nowdiff", - oldParams: Parameters{ - BaseCycles: 0, - BaseRef: 0, - Frequency: 10000, - }, - now: 60000, - // Host frequency changed to 11000 immediately after - // oldParams was returned. - newParams: Parameters{ - BaseCycles: 55000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 11000, - }, - // nextRef = 7000ms - // nextCycles = 11000 * (7000ms - 5000ms) + 55000 - // nextCycles = 11000 * (2000ms) + 55000 - // nextCycles = 77000 - // f = (77000 - 60000) / 1s = 17000 - // - // ref = 6000ms - (60000 - 55000) / 17000 - // ref = 5705882353ns - want: Parameters{ - BaseCycles: 55000, - BaseRef: ReferenceNS(5705882353), - Frequency: 17000, - }, - // oldNow = 60000 * 10000 = 6s - // newNow = (60000 - 55000) / 11000 + 5s = 5.4545s - errorNS: ReferenceNS((6*time.Second - 5454545454).Nanoseconds()), - }, - { - // Host time went backwards. - name: "time-backwards", - oldParams: Parameters{ - BaseCycles: 50000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - now: 60000, - newParams: Parameters{ - BaseCycles: 60000, - // From oldParams, we think ref = 6s at cycles = 60000. - BaseRef: ReferenceNS(4000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - want: Parameters{ - BaseCycles: 60000, - BaseRef: ReferenceNS(6000 * time.Millisecond.Nanoseconds()), - // We must increase the frequency by 200% to - // correct 2s of error in 1s - // (ApproxUpdateInterval). - Frequency: 30000, - }, - errorNS: ReferenceNS(2000 * time.Millisecond.Nanoseconds()), - }, - { - // Host time went backwards, with now ahead of newParams. - name: "time-backwards-nowdiff", - oldParams: Parameters{ - BaseCycles: 50000, - BaseRef: ReferenceNS(5000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - now: 65000, - // nextRef = 7500ms - // nextCycles = 10000 * (7500ms - 4000ms) + 60000 - // nextCycles = 10000 * (3500ms) + 60000 - // nextCycles = 95000 - // f = (95000 - 65000) / 1s = 30000 - // - // ref = 6500ms - (65000 - 60000) / 30000 - // ref = 6333333333ns - newParams: Parameters{ - BaseCycles: 60000, - BaseRef: ReferenceNS(4000 * time.Millisecond.Nanoseconds()), - Frequency: 10000, - }, - want: Parameters{ - BaseCycles: 60000, - BaseRef: ReferenceNS(6333333334), - Frequency: 30000, - }, - // oldNow = 65000 * 10000 = 6.5s - // newNow = (65000 - 60000) / 10000 + 4s = 4.5s - errorNS: ReferenceNS(2000 * time.Millisecond.Nanoseconds()), - }, - } - for _, tc := range testCases { - t.Run(tc.name, func(t *testing.T) { - got, errorNS, err := errorAdjust(tc.oldParams, tc.newParams, tc.now) - if err != nil && !tc.wantErr { - t.Errorf("err got %v want nil", err) - } else if err == nil && tc.wantErr { - t.Errorf("err got nil want non-nil") - } - - if got != tc.want { - t.Errorf("Parameters got %+v want %+v", got, tc.want) - } - if errorNS != tc.errorNS { - t.Errorf("errorNS got %v want %v", errorNS, tc.errorNS) - } - }) - } -} - -func testMuldiv(t *testing.T, v uint64) { - for i := uint64(1); i <= 1000000; i++ { - mult := uint64(1000000000) - div := i * mult - res, ok := muldiv64(v, mult, div) - if !ok { - t.Errorf("Result of %v * %v / %v ok got false want true", v, mult, div) - } - if want := v / i; res != want { - t.Errorf("Bad result of %v * %v / %v: got %v, want %v", v, mult, div, res, want) - } - } -} - -func TestMulDiv(t *testing.T) { - testMuldiv(t, math.MaxUint64) - for i := int64(-10); i <= 10; i++ { - testMuldiv(t, uint64(i)) - } -} - -func TestMulDivZero(t *testing.T) { - if r, ok := muldiv64(2, 4, 0); ok { - t.Errorf("muldiv64(2, 4, 0) got %d, ok want !ok", r) - } - - if r, ok := muldiv64(0, 0, 0); ok { - t.Errorf("muldiv64(0, 0, 0) got %d, ok want !ok", r) - } -} - -func TestMulDivOverflow(t *testing.T) { - testCases := []struct { - name string - val uint64 - mult uint64 - div uint64 - ok bool - ret uint64 - }{ - { - name: "2^62", - val: 1 << 63, - mult: 4, - div: 8, - ok: true, - ret: 1 << 62, - }, - { - name: "2^64-1", - val: 0xffffffffffffffff, - mult: 1, - div: 1, - ok: true, - ret: 0xffffffffffffffff, - }, - { - name: "2^64", - val: 1 << 63, - mult: 4, - div: 2, - ok: false, - }, - { - name: "2^125", - val: 1 << 63, - mult: 1 << 63, - div: 2, - ok: false, - }, - } - - for _, tc := range testCases { - t.Run(tc.name, func(t *testing.T) { - r, ok := muldiv64(tc.val, tc.mult, tc.div) - if ok != tc.ok { - t.Errorf("ok got %v want %v", ok, tc.ok) - } - if tc.ok && r != tc.ret { - t.Errorf("ret got %v want %v", r, tc.ret) - } - }) - } -} diff --git a/pkg/sentry/time/sampler_test.go b/pkg/sentry/time/sampler_test.go deleted file mode 100644 index 3e70a1134..000000000 --- a/pkg/sentry/time/sampler_test.go +++ /dev/null @@ -1,183 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package time - -import ( - "errors" - "testing" -) - -// errNoSamples is returned when testReferenceClocks runs out of samples. -var errNoSamples = errors.New("no samples available") - -// testReferenceClocks returns a preset list of samples and cycle counts. -type testReferenceClocks struct { - samples []sample - cycles []TSCValue -} - -// Sample implements referenceClocks.Sample, returning the next sample in the list. -func (t *testReferenceClocks) Sample(_ ClockID) (sample, error) { - if len(t.samples) == 0 { - return sample{}, errNoSamples - } - - s := t.samples[0] - if len(t.samples) == 1 { - t.samples = nil - } else { - t.samples = t.samples[1:] - } - - return s, nil -} - -// Cycles implements referenceClocks.Cycles, returning the next TSCValue in the list. -func (t *testReferenceClocks) Cycles() TSCValue { - if len(t.cycles) == 0 { - return 0 - } - - c := t.cycles[0] - if len(t.cycles) == 1 { - t.cycles = nil - } else { - t.cycles = t.cycles[1:] - } - - return c -} - -// newTestSampler returns a sampler that collects samples from -// the given sample list and cycle counts from the given cycle list. -func newTestSampler(samples []sample, cycles []TSCValue) *sampler { - return &sampler{ - clocks: &testReferenceClocks{ - samples: samples, - cycles: cycles, - }, - overhead: defaultOverheadCycles, - } -} - -// generateSamples generates n samples with the given overhead. -func generateSamples(n int, overhead TSCValue) []sample { - samples := []sample{{before: 1000000, after: 1000000 + overhead, ref: 100}} - for i := 0; i < n-1; i++ { - prev := samples[len(samples)-1] - samples = append(samples, sample{ - before: prev.before + 1000000, - after: prev.after + 1000000, - ref: prev.ref + 100, - }) - } - return samples -} - -// TestSample ensures that samples can be collected. -func TestSample(t *testing.T) { - testCases := []struct { - name string - samples []sample - err error - }{ - { - name: "basic", - samples: []sample{ - {before: 100000, after: 100000 + defaultOverheadCycles, ref: 100}, - }, - err: nil, - }, - { - // Sample with backwards TSC ignored. - // referenceClock should retry and get errNoSamples. - name: "backwards-tsc-ignored", - samples: []sample{ - {before: 100000, after: 90000, ref: 100}, - }, - err: errNoSamples, - }, - { - // Sample far above overhead skipped. - // referenceClock should retry and get errNoSamples. - name: "reject-overhead", - samples: []sample{ - {before: 100000, after: 100000 + 5*defaultOverheadCycles, ref: 100}, - }, - err: errNoSamples, - }, - { - // Maximum overhead allowed is bounded. - name: "over-max-overhead", - // Generate a bunch of samples. The reference clock - // needs a while to ramp up its expected overhead. - samples: generateSamples(100, 2*maxOverheadCycles), - err: errOverheadTooHigh, - }, - { - // Overhead at maximum overhead is allowed. - name: "max-overhead", - // Generate a bunch of samples. The reference clock - // needs a while to ramp up its expected overhead. - samples: generateSamples(100, maxOverheadCycles), - err: nil, - }, - } - for _, tc := range testCases { - t.Run(tc.name, func(t *testing.T) { - s := newTestSampler(tc.samples, nil) - err := s.Sample() - if err != tc.err { - t.Errorf("Sample err got %v want %v", err, tc.err) - } - }) - } -} - -// TestOutliersIgnored tests that referenceClock ignores samples with very high -// overhead. -func TestOutliersIgnored(t *testing.T) { - s := newTestSampler([]sample{ - {before: 100000, after: 100000 + defaultOverheadCycles, ref: 100}, - {before: 200000, after: 200000 + defaultOverheadCycles, ref: 200}, - {before: 300000, after: 300000 + defaultOverheadCycles, ref: 300}, - {before: 400000, after: 400000 + defaultOverheadCycles, ref: 400}, - {before: 500000, after: 500000 + 5*defaultOverheadCycles, ref: 500}, // Ignored - {before: 600000, after: 600000 + defaultOverheadCycles, ref: 600}, - {before: 700000, after: 700000 + defaultOverheadCycles, ref: 700}, - }, nil) - - // Collect 5 samples. - for i := 0; i < 5; i++ { - err := s.Sample() - if err != nil { - t.Fatalf("Unexpected error while sampling: %v", err) - } - } - - oldest, newest, ok := s.Range() - if !ok { - t.Fatalf("Range not ok") - } - - if oldest.ref != 100 { - t.Errorf("oldest.ref got %v want %v", oldest.ref, 100) - } - - // We skipped the high-overhead sample. - if newest.ref != 600 { - t.Errorf("newest.ref got %v want %v", newest.ref, 600) - } -} diff --git a/pkg/sentry/time/seqatomic_parameters_unsafe.go b/pkg/sentry/time/seqatomic_parameters_unsafe.go new file mode 100755 index 000000000..efd3ccae2 --- /dev/null +++ b/pkg/sentry/time/seqatomic_parameters_unsafe.go @@ -0,0 +1,55 @@ +package time + +import ( + "fmt" + "reflect" + "strings" + "unsafe" + + "gvisor.dev/gvisor/pkg/sync" +) + +// SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race +// with any writer critical sections in sc. +func SeqAtomicLoadParameters(sc *sync.SeqCount, ptr *Parameters) Parameters { + // This function doesn't use SeqAtomicTryLoad because doing so is + // measurably, significantly (~20%) slower; Go is awful at inlining. + var val Parameters + for { + epoch := sc.BeginRead() + if sync.RaceEnabled { + + sync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) + } else { + + val = *ptr + } + if sc.ReadOk(epoch) { + break + } + } + return val +} + +// SeqAtomicTryLoad returns a copy of *ptr while in a reader critical section +// in sc initiated by a call to sc.BeginRead() that returned epoch. If the read +// would race with a writer critical section, SeqAtomicTryLoad returns +// (unspecified, false). +func SeqAtomicTryLoadParameters(sc *sync.SeqCount, epoch sync.SeqCountEpoch, ptr *Parameters) (Parameters, bool) { + var val Parameters + if sync.RaceEnabled { + sync.Memmove(unsafe.Pointer(&val), unsafe.Pointer(ptr), unsafe.Sizeof(val)) + } else { + val = *ptr + } + return val, sc.ReadOk(epoch) +} + +func initParameters() { + var val Parameters + typ := reflect.TypeOf(val) + name := typ.Name() + if ptrs := sync.PointersInType(typ, name); len(ptrs) != 0 { + panic(fmt.Sprintf("SeqAtomicLoad<%s> is invalid since values %s of type %s contain pointers:\n%s", typ, name, typ, strings.Join(ptrs, "\n"))) + } +} diff --git a/pkg/sentry/time/time_arm64_state_autogen.go b/pkg/sentry/time/time_arm64_state_autogen.go new file mode 100755 index 000000000..2adc9c9e0 --- /dev/null +++ b/pkg/sentry/time/time_arm64_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package time diff --git a/pkg/sentry/time/time_state_autogen.go b/pkg/sentry/time/time_state_autogen.go new file mode 100755 index 000000000..2adc9c9e0 --- /dev/null +++ b/pkg/sentry/time/time_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package time diff --git a/pkg/sentry/unimpl/BUILD b/pkg/sentry/unimpl/BUILD deleted file mode 100644 index 5d4aa3a63..000000000 --- a/pkg/sentry/unimpl/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -load("//tools:defs.bzl", "go_library", "proto_library") - -package(licenses = ["notice"]) - -proto_library( - name = "unimplemented_syscall", - srcs = ["unimplemented_syscall.proto"], - visibility = ["//visibility:public"], - deps = ["//pkg/sentry/arch:registers_proto"], -) - -go_library( - name = "unimpl", - srcs = ["events.go"], - visibility = ["//:sandbox"], - deps = [ - "//pkg/context", - "//pkg/log", - ], -) diff --git a/pkg/sentry/unimpl/unimpl_state_autogen.go b/pkg/sentry/unimpl/unimpl_state_autogen.go new file mode 100755 index 000000000..b37d16f87 --- /dev/null +++ b/pkg/sentry/unimpl/unimpl_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package unimpl diff --git a/pkg/sentry/unimpl/unimplemented_syscall.proto b/pkg/sentry/unimpl/unimplemented_syscall.proto deleted file mode 100644 index 0d7a94be7..000000000 --- a/pkg/sentry/unimpl/unimplemented_syscall.proto +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -syntax = "proto3"; - -package gvisor; - -import "pkg/sentry/arch/registers.proto"; - -message UnimplementedSyscall { - // Task ID. - int32 tid = 1; - - // Registers at the time of the call. - Registers registers = 2; -} diff --git a/pkg/sentry/unimpl/unimplemented_syscall_go_proto/unimplemented_syscall.pb.go b/pkg/sentry/unimpl/unimplemented_syscall_go_proto/unimplemented_syscall.pb.go new file mode 100755 index 000000000..4dfb169cc --- /dev/null +++ b/pkg/sentry/unimpl/unimplemented_syscall_go_proto/unimplemented_syscall.pb.go @@ -0,0 +1,91 @@ +// Code generated by protoc-gen-go. DO NOT EDIT. +// source: pkg/sentry/unimpl/unimplemented_syscall.proto + +package gvisor + +import ( + fmt "fmt" + proto "github.com/golang/protobuf/proto" + registers_go_proto "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" + math "math" +) + +// Reference imports to suppress errors if they are not otherwise used. +var _ = proto.Marshal +var _ = fmt.Errorf +var _ = math.Inf + +// This is a compile-time assertion to ensure that this generated file +// is compatible with the proto package it is being compiled against. +// A compilation error at this line likely means your copy of the +// proto package needs to be updated. +const _ = proto.ProtoPackageIsVersion3 // please upgrade the proto package + +type UnimplementedSyscall struct { + Tid int32 `protobuf:"varint,1,opt,name=tid,proto3" json:"tid,omitempty"` + Registers *registers_go_proto.Registers `protobuf:"bytes,2,opt,name=registers,proto3" json:"registers,omitempty"` + XXX_NoUnkeyedLiteral struct{} `json:"-"` + XXX_unrecognized []byte `json:"-"` + XXX_sizecache int32 `json:"-"` +} + +func (m *UnimplementedSyscall) Reset() { *m = UnimplementedSyscall{} } +func (m *UnimplementedSyscall) String() string { return proto.CompactTextString(m) } +func (*UnimplementedSyscall) ProtoMessage() {} +func (*UnimplementedSyscall) Descriptor() ([]byte, []int) { + return fileDescriptor_ddc2fcd2bea3c75d, []int{0} +} + +func (m *UnimplementedSyscall) XXX_Unmarshal(b []byte) error { + return xxx_messageInfo_UnimplementedSyscall.Unmarshal(m, b) +} +func (m *UnimplementedSyscall) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + return xxx_messageInfo_UnimplementedSyscall.Marshal(b, m, deterministic) +} +func (m *UnimplementedSyscall) XXX_Merge(src proto.Message) { + xxx_messageInfo_UnimplementedSyscall.Merge(m, src) +} +func (m *UnimplementedSyscall) XXX_Size() int { + return xxx_messageInfo_UnimplementedSyscall.Size(m) +} +func (m *UnimplementedSyscall) XXX_DiscardUnknown() { + xxx_messageInfo_UnimplementedSyscall.DiscardUnknown(m) +} + +var xxx_messageInfo_UnimplementedSyscall proto.InternalMessageInfo + +func (m *UnimplementedSyscall) GetTid() int32 { + if m != nil { + return m.Tid + } + return 0 +} + +func (m *UnimplementedSyscall) GetRegisters() *registers_go_proto.Registers { + if m != nil { + return m.Registers + } + return nil +} + +func init() { + proto.RegisterType((*UnimplementedSyscall)(nil), "gvisor.UnimplementedSyscall") +} + +func init() { + proto.RegisterFile("pkg/sentry/unimpl/unimplemented_syscall.proto", fileDescriptor_ddc2fcd2bea3c75d) +} + +var fileDescriptor_ddc2fcd2bea3c75d = []byte{ + // 149 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xe2, 0xd2, 0x2d, 0xc8, 0x4e, 0xd7, + 0x2f, 0x4e, 0xcd, 0x2b, 0x29, 0xaa, 0xd4, 0x2f, 0xcd, 0xcb, 0xcc, 0x2d, 0xc8, 0x81, 0x52, 0xa9, + 0xb9, 0xa9, 0x79, 0x25, 0xa9, 0x29, 0xf1, 0xc5, 0x95, 0xc5, 0xc9, 0x89, 0x39, 0x39, 0x7a, 0x05, + 0x45, 0xf9, 0x25, 0xf9, 0x42, 0x6c, 0xe9, 0x65, 0x99, 0xc5, 0xf9, 0x45, 0x52, 0xf2, 0x48, 0xda, + 0x12, 0x8b, 0x92, 0x33, 0xf4, 0x8b, 0x52, 0xd3, 0x33, 0x8b, 0x4b, 0x52, 0x8b, 0x8a, 0x21, 0x0a, + 0x95, 0x22, 0xb9, 0x44, 0x42, 0x91, 0xcd, 0x09, 0x86, 0x18, 0x23, 0x24, 0xc0, 0xc5, 0x5c, 0x92, + 0x99, 0x22, 0xc1, 0xa8, 0xc0, 0xa8, 0xc1, 0x1a, 0x04, 0x62, 0x0a, 0xe9, 0x73, 0x71, 0xc2, 0x35, + 0x4b, 0x30, 0x29, 0x30, 0x6a, 0x70, 0x1b, 0x09, 0xea, 0x41, 0xac, 0xd1, 0x0b, 0x82, 0x49, 0x04, + 0x21, 0xd4, 0x24, 0xb1, 0x81, 0x6d, 0x30, 0x06, 0x04, 0x00, 0x00, 0xff, 0xff, 0x51, 0x4a, 0x47, + 0x79, 0xbb, 0x00, 0x00, 0x00, +} diff --git a/pkg/sentry/uniqueid/BUILD b/pkg/sentry/uniqueid/BUILD deleted file mode 100644 index 7467e6398..000000000 --- a/pkg/sentry/uniqueid/BUILD +++ /dev/null @@ -1,13 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "uniqueid", - srcs = ["context.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/context", - "//pkg/sentry/socket/unix/transport", - ], -) diff --git a/pkg/sentry/uniqueid/uniqueid_state_autogen.go b/pkg/sentry/uniqueid/uniqueid_state_autogen.go new file mode 100755 index 000000000..1890fdf46 --- /dev/null +++ b/pkg/sentry/uniqueid/uniqueid_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package uniqueid diff --git a/pkg/sentry/usage/BUILD b/pkg/sentry/usage/BUILD deleted file mode 100644 index 099315613..000000000 --- a/pkg/sentry/usage/BUILD +++ /dev/null @@ -1,22 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "usage", - srcs = [ - "cpu.go", - "io.go", - "memory.go", - "memory_unsafe.go", - "usage.go", - ], - visibility = [ - "//:sandbox", - ], - deps = [ - "//pkg/bits", - "//pkg/memutil", - "//pkg/sync", - ], -) diff --git a/pkg/sentry/usage/usage_state_autogen.go b/pkg/sentry/usage/usage_state_autogen.go new file mode 100755 index 000000000..42979ea25 --- /dev/null +++ b/pkg/sentry/usage/usage_state_autogen.go @@ -0,0 +1,50 @@ +// automatically generated by stateify. + +package usage + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *CPUStats) beforeSave() {} +func (x *CPUStats) save(m state.Map) { + x.beforeSave() + m.Save("UserTime", &x.UserTime) + m.Save("SysTime", &x.SysTime) + m.Save("VoluntarySwitches", &x.VoluntarySwitches) +} + +func (x *CPUStats) afterLoad() {} +func (x *CPUStats) load(m state.Map) { + m.Load("UserTime", &x.UserTime) + m.Load("SysTime", &x.SysTime) + m.Load("VoluntarySwitches", &x.VoluntarySwitches) +} + +func (x *IO) beforeSave() {} +func (x *IO) save(m state.Map) { + x.beforeSave() + m.Save("CharsRead", &x.CharsRead) + m.Save("CharsWritten", &x.CharsWritten) + m.Save("ReadSyscalls", &x.ReadSyscalls) + m.Save("WriteSyscalls", &x.WriteSyscalls) + m.Save("BytesRead", &x.BytesRead) + m.Save("BytesWritten", &x.BytesWritten) + m.Save("BytesWriteCancelled", &x.BytesWriteCancelled) +} + +func (x *IO) afterLoad() {} +func (x *IO) load(m state.Map) { + m.Load("CharsRead", &x.CharsRead) + m.Load("CharsWritten", &x.CharsWritten) + m.Load("ReadSyscalls", &x.ReadSyscalls) + m.Load("WriteSyscalls", &x.WriteSyscalls) + m.Load("BytesRead", &x.BytesRead) + m.Load("BytesWritten", &x.BytesWritten) + m.Load("BytesWriteCancelled", &x.BytesWriteCancelled) +} + +func init() { + state.Register("pkg/sentry/usage.CPUStats", (*CPUStats)(nil), state.Fns{Save: (*CPUStats).save, Load: (*CPUStats).load}) + state.Register("pkg/sentry/usage.IO", (*IO)(nil), state.Fns{Save: (*IO).save, Load: (*IO).load}) +} diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD deleted file mode 100644 index a2a06fc8f..000000000 --- a/pkg/sentry/vfs/BUILD +++ /dev/null @@ -1,80 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -licenses(["notice"]) - -go_template_instance( - name = "epoll_interest_list", - out = "epoll_interest_list.go", - package = "vfs", - prefix = "epollInterest", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*epollInterest", - "Linker": "*epollInterest", - }, -) - -go_library( - name = "vfs", - srcs = [ - "anonfs.go", - "context.go", - "debug.go", - "dentry.go", - "device.go", - "epoll.go", - "epoll_interest_list.go", - "file_description.go", - "file_description_impl_util.go", - "filesystem.go", - "filesystem_impl_util.go", - "filesystem_type.go", - "mount.go", - "mount_unsafe.go", - "options.go", - "pathname.go", - "permissions.go", - "resolving_path.go", - "vfs.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fd", - "//pkg/fspath", - "//pkg/gohacks", - "//pkg/log", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fs", - "//pkg/sentry/fs/lock", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/limits", - "//pkg/sentry/memmap", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - "@org_golang_x_sys//unix:go_default_library", - ], -) - -go_test( - name = "vfs_test", - size = "small", - srcs = [ - "file_description_impl_util_test.go", - "mount_test.go", - ], - library = ":vfs", - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/contexttest", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/vfs/README.md b/pkg/sentry/vfs/README.md deleted file mode 100644 index 9aa133bcb..000000000 --- a/pkg/sentry/vfs/README.md +++ /dev/null @@ -1,197 +0,0 @@ -# The gVisor Virtual Filesystem - -THIS PACKAGE IS CURRENTLY EXPERIMENTAL AND NOT READY OR ENABLED FOR PRODUCTION -USE. For the filesystem implementation currently used by gVisor, see the `fs` -package. - -## Implementation Notes - -### Reference Counting - -Filesystem, Dentry, Mount, MountNamespace, and FileDescription are all -reference-counted. Mount and MountNamespace are exclusively VFS-managed; when -their reference count reaches zero, VFS releases their resources. Filesystem and -FileDescription management is shared between VFS and filesystem implementations; -when their reference count reaches zero, VFS notifies the implementation by -calling `FilesystemImpl.Release()` or `FileDescriptionImpl.Release()` -respectively and then releases VFS-owned resources. Dentries are exclusively -managed by filesystem implementations; reference count changes are abstracted -through DentryImpl, which should release resources when reference count reaches -zero. - -Filesystem references are held by: - -- Mount: Each referenced Mount holds a reference on the mounted Filesystem. - -Dentry references are held by: - -- FileDescription: Each referenced FileDescription holds a reference on the - Dentry through which it was opened, via `FileDescription.vd.dentry`. - -- Mount: Each referenced Mount holds a reference on its mount point and on the - mounted filesystem root. The mount point is mutable (`mount(MS_MOVE)`). - -Mount references are held by: - -- FileDescription: Each referenced FileDescription holds a reference on the - Mount on which it was opened, via `FileDescription.vd.mount`. - -- Mount: Each referenced Mount holds a reference on its parent, which is the - mount containing its mount point. - -- VirtualFilesystem: A reference is held on each Mount that has not been - umounted. - -MountNamespace and FileDescription references are held by users of VFS. The -expectation is that each `kernel.Task` holds a reference on its corresponding -MountNamespace, and each file descriptor holds a reference on its represented -FileDescription. - -Notes: - -- Dentries do not hold a reference on their owning Filesystem. Instead, all - uses of a Dentry occur in the context of a Mount, which holds a reference on - the relevant Filesystem (see e.g. the VirtualDentry type). As a corollary, - when releasing references on both a Dentry and its corresponding Mount, the - Dentry's reference must be released first (because releasing the Mount's - reference may release the last reference on the Filesystem, whose state may - be required to release the Dentry reference). - -### The Inheritance Pattern - -Filesystem, Dentry, and FileDescription are all concepts featuring both state -that must be shared between VFS and filesystem implementations, and operations -that are implementation-defined. To facilitate this, each of these three -concepts follows the same pattern, shown below for Dentry: - -```go -// Dentry represents a node in a filesystem tree. -type Dentry struct { - // VFS-required dentry state. - parent *Dentry - // ... - - // impl is the DentryImpl associated with this Dentry. impl is immutable. - // This should be the last field in Dentry. - impl DentryImpl -} - -// Init must be called before first use of d. -func (d *Dentry) Init(impl DentryImpl) { - d.impl = impl -} - -// Impl returns the DentryImpl associated with d. -func (d *Dentry) Impl() DentryImpl { - return d.impl -} - -// DentryImpl contains implementation-specific details of a Dentry. -// Implementations of DentryImpl should contain their associated Dentry by -// value as their first field. -type DentryImpl interface { - // VFS-required implementation-defined dentry operations. - IncRef() - // ... -} -``` - -This construction, which is essentially a type-safe analogue to Linux's -`container_of` pattern, has the following properties: - -- VFS works almost exclusively with pointers to Dentry rather than DentryImpl - interface objects, such as in the type of `Dentry.parent`. This avoids - interface method calls (which are somewhat expensive to perform, and defeat - inlining and escape analysis), reduces the size of VFS types (since an - interface object is two pointers in size), and allows pointers to be loaded - and stored atomically using `sync/atomic`. Implementation-defined behavior - is accessed via `Dentry.impl` when required. - -- Filesystem implementations can access the implementation-defined state - associated with objects of VFS types by type-asserting or type-switching - (e.g. `Dentry.Impl().(*myDentry)`). Type assertions to a concrete type - require only an equality comparison of the interface object's type pointer - to a static constant, and are consequently very fast. - -- Filesystem implementations can access the VFS state associated with objects - of implementation-defined types directly. - -- VFS and implementation-defined state for a given type occupy the same - object, minimizing memory allocations and maximizing memory locality. `impl` - is the last field in `Dentry`, and `Dentry` is the first field in - `DentryImpl` implementations, for similar reasons: this tends to cause - fetching of the `Dentry.impl` interface object to also fetch `DentryImpl` - fields, either because they are in the same cache line or via next-line - prefetching. - -## Future Work - -- Most `mount(2)` features, and unmounting, are incomplete. - -- VFS1 filesystems are not directly compatible with VFS2. It may be possible - to implement shims that implement `vfs.FilesystemImpl` for - `fs.MountNamespace`, `vfs.DentryImpl` for `fs.Dirent`, and - `vfs.FileDescriptionImpl` for `fs.File`, which may be adequate for - filesystems that are not performance-critical (e.g. sysfs); however, it is - not clear that this will be less effort than simply porting the filesystems - in question. Practically speaking, the following filesystems will probably - need to be ported or made compatible through a shim to evaluate filesystem - performance on realistic workloads: - - - devfs/procfs/sysfs, which will realistically be necessary to execute - most applications. (Note that procfs and sysfs do not support hard - links, so they do not require the complexity of separate inode objects. - Also note that Linux's /dev is actually a variant of tmpfs called - devtmpfs.) - - - tmpfs. This should be relatively straightforward: copy/paste memfs, - store regular file contents in pgalloc-allocated memory instead of - `[]byte`, and add support for file timestamps. (In fact, it probably - makes more sense to convert memfs to tmpfs and not keep the former.) - - - A remote filesystem, either lisafs (if it is ready by the time that - other benchmarking prerequisites are) or v9fs (aka 9P, aka gofers). - - - epoll files. - - Filesystems that will need to be ported before switching to VFS2, but can - probably be skipped for early testing: - - - overlayfs, which is needed for (at least) synthetic mount points. - - - Support for host ttys. - - - timerfd files. - - Filesystems that can be probably dropped: - - - ashmem, which is far too incomplete to use. - - - binder, which is similarly far too incomplete to use. - - - whitelistfs, which we are already actively attempting to remove. - -- Save/restore. For instance, it is unclear if the current implementation of - the `state` package supports the inheritance pattern described above. - -- Many features that were previously implemented by VFS must now be - implemented by individual filesystems (though, in most cases, this should - consist of calls to hooks or libraries provided by `vfs` or other packages). - This includes, but is not necessarily limited to: - - - Block and character device special files - - - Inotify - - - File locking - - - `O_ASYNC` - -- Reference counts in the `vfs` package do not use the `refs` package since - `refs.AtomicRefCount` adds 64 bytes of overhead to each 8-byte reference - count, resulting in considerable cache bloat. 24 bytes of this overhead is - for weak reference support, which have poor performance and will not be used - by VFS2. The remaining 40 bytes is to store a descriptive string and stack - trace for reference leak checking; we can support reference leak checking - without incurring this space overhead by including the applicable - information directly in finalizers for applicable types. diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index a62e43589..a62e43589 100644..100755 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go diff --git a/pkg/sentry/vfs/context.go b/pkg/sentry/vfs/context.go index 82781e6d3..82781e6d3 100644..100755 --- a/pkg/sentry/vfs/context.go +++ b/pkg/sentry/vfs/context.go diff --git a/pkg/sentry/vfs/debug.go b/pkg/sentry/vfs/debug.go index 0ed20f249..0ed20f249 100644..100755 --- a/pkg/sentry/vfs/debug.go +++ b/pkg/sentry/vfs/debug.go diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index 35b208721..35b208721 100644..100755 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go index bda5576fa..bda5576fa 100644..100755 --- a/pkg/sentry/vfs/device.go +++ b/pkg/sentry/vfs/device.go diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 3da45d744..3da45d744 100644..100755 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go diff --git a/pkg/sentry/vfs/epoll_interest_list.go b/pkg/sentry/vfs/epoll_interest_list.go new file mode 100755 index 000000000..1bd41f400 --- /dev/null +++ b/pkg/sentry/vfs/epoll_interest_list.go @@ -0,0 +1,186 @@ +package vfs + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type epollInterestElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (epollInterestElementMapper) linkerFor(elem *epollInterest) *epollInterest { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type epollInterestList struct { + head *epollInterest + tail *epollInterest +} + +// Reset resets list l to the empty state. +func (l *epollInterestList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *epollInterestList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *epollInterestList) Front() *epollInterest { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *epollInterestList) Back() *epollInterest { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *epollInterestList) PushFront(e *epollInterest) { + linker := epollInterestElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + + if l.head != nil { + epollInterestElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *epollInterestList) PushBack(e *epollInterest) { + linker := epollInterestElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + + if l.tail != nil { + epollInterestElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *epollInterestList) PushBackList(m *epollInterestList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + epollInterestElementMapper{}.linkerFor(l.tail).SetNext(m.head) + epollInterestElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *epollInterestList) InsertAfter(b, e *epollInterest) { + bLinker := epollInterestElementMapper{}.linkerFor(b) + eLinker := epollInterestElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + epollInterestElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *epollInterestList) InsertBefore(a, e *epollInterest) { + aLinker := epollInterestElementMapper{}.linkerFor(a) + eLinker := epollInterestElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + epollInterestElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *epollInterestList) Remove(e *epollInterest) { + linker := epollInterestElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + epollInterestElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + epollInterestElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type epollInterestEntry struct { + next *epollInterest + prev *epollInterest +} + +// Next returns the entry that follows e in the list. +func (e *epollInterestEntry) Next() *epollInterest { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *epollInterestEntry) Prev() *epollInterest { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *epollInterestEntry) SetNext(elem *epollInterest) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *epollInterestEntry) SetPrev(elem *epollInterest) { + e.prev = elem +} diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 8ee549dc2..8ee549dc2 100644..100755 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index d45e602ce..d45e602ce 100644..100755 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go deleted file mode 100644 index 3a75d4d62..000000000 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ /dev/null @@ -1,223 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs - -import ( - "bytes" - "fmt" - "io" - "sync/atomic" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// fileDescription is the common fd struct which a filesystem implementation -// embeds in all of its file description implementations as required. -type fileDescription struct { - vfsfd FileDescription - FileDescriptionDefaultImpl -} - -// genCount contains the number of times its DynamicBytesSource.Generate() -// implementation has been called. -type genCount struct { - count uint64 // accessed using atomic memory ops -} - -// Generate implements DynamicBytesSource.Generate. -func (g *genCount) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d", atomic.AddUint64(&g.count, 1)) - return nil -} - -type storeData struct { - data string -} - -var _ WritableDynamicBytesSource = (*storeData)(nil) - -// Generate implements DynamicBytesSource. -func (d *storeData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(d.data) - return nil -} - -// Generate implements WritableDynamicBytesSource. -func (d *storeData) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - buf := make([]byte, src.NumBytes()) - n, err := src.CopyIn(ctx, buf) - if err != nil { - return 0, err - } - - d.data = string(buf[:n]) - return 0, nil -} - -// testFD is a read-only FileDescriptionImpl representing a regular file. -type testFD struct { - fileDescription - DynamicBytesFileDescriptionImpl - - data DynamicBytesSource -} - -func newTestFD(vfsObj *VirtualFilesystem, statusFlags uint32, data DynamicBytesSource) *FileDescription { - vd := vfsObj.NewAnonVirtualDentry("genCountFD") - defer vd.DecRef() - var fd testFD - fd.vfsfd.Init(&fd, statusFlags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{}) - fd.DynamicBytesFileDescriptionImpl.SetDataSource(data) - return &fd.vfsfd -} - -// Release implements FileDescriptionImpl.Release. -func (fd *testFD) Release() { -} - -// SetStatusFlags implements FileDescriptionImpl.SetStatusFlags. -// Stat implements FileDescriptionImpl.Stat. -func (fd *testFD) Stat(ctx context.Context, opts StatOptions) (linux.Statx, error) { - // Note that Statx.Mask == 0 in the return value. - return linux.Statx{}, nil -} - -// SetStat implements FileDescriptionImpl.SetStat. -func (fd *testFD) SetStat(ctx context.Context, opts SetStatOptions) error { - return syserror.EPERM -} - -func TestGenCountFD(t *testing.T) { - ctx := contexttest.Context(t) - - vfsObj := &VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - fd := newTestFD(vfsObj, linux.O_RDWR, &genCount{}) - defer fd.DecRef() - - // The first read causes Generate to be called to fill the FD's buffer. - buf := make([]byte, 2) - ioseq := usermem.BytesIOSequence(buf) - n, err := fd.Read(ctx, ioseq, ReadOptions{}) - if n != 1 || (err != nil && err != io.EOF) { - t.Fatalf("first Read: got (%d, %v), wanted (1, nil or EOF)", n, err) - } - if want := byte('1'); buf[0] != want { - t.Errorf("first Read: got byte %c, wanted %c", buf[0], want) - } - - // A second read without seeking is still at EOF. - n, err = fd.Read(ctx, ioseq, ReadOptions{}) - if n != 0 || err != io.EOF { - t.Fatalf("second Read: got (%d, %v), wanted (0, EOF)", n, err) - } - - // Seeking to the beginning of the file causes it to be regenerated. - n, err = fd.Seek(ctx, 0, linux.SEEK_SET) - if n != 0 || err != nil { - t.Fatalf("Seek: got (%d, %v), wanted (0, nil)", n, err) - } - n, err = fd.Read(ctx, ioseq, ReadOptions{}) - if n != 1 || (err != nil && err != io.EOF) { - t.Fatalf("Read after Seek: got (%d, %v), wanted (1, nil or EOF)", n, err) - } - if want := byte('2'); buf[0] != want { - t.Errorf("Read after Seek: got byte %c, wanted %c", buf[0], want) - } - - // PRead at the beginning of the file also causes it to be regenerated. - n, err = fd.PRead(ctx, ioseq, 0, ReadOptions{}) - if n != 1 || (err != nil && err != io.EOF) { - t.Fatalf("PRead: got (%d, %v), wanted (1, nil or EOF)", n, err) - } - if want := byte('3'); buf[0] != want { - t.Errorf("PRead: got byte %c, wanted %c", buf[0], want) - } - - // Write and PWrite fails. - if _, err := fd.Write(ctx, ioseq, WriteOptions{}); err != syserror.EINVAL { - t.Errorf("Write: got err %v, wanted %v", err, syserror.EINVAL) - } - if _, err := fd.PWrite(ctx, ioseq, 0, WriteOptions{}); err != syserror.EINVAL { - t.Errorf("Write: got err %v, wanted %v", err, syserror.EINVAL) - } -} - -func TestWritable(t *testing.T) { - ctx := contexttest.Context(t) - - vfsObj := &VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - fd := newTestFD(vfsObj, linux.O_RDWR, &storeData{data: "init"}) - defer fd.DecRef() - - buf := make([]byte, 10) - ioseq := usermem.BytesIOSequence(buf) - if n, err := fd.Read(ctx, ioseq, ReadOptions{}); n != 4 && err != io.EOF { - t.Fatalf("Read: got (%v, %v), wanted (4, EOF)", n, err) - } - if want := "init"; want == string(buf) { - t.Fatalf("Read: got %v, wanted %v", string(buf), want) - } - - // Test PWrite. - want := "write" - writeIOSeq := usermem.BytesIOSequence([]byte(want)) - if n, err := fd.PWrite(ctx, writeIOSeq, 0, WriteOptions{}); int(n) != len(want) && err != nil { - t.Errorf("PWrite: got err (%v, %v), wanted (%v, nil)", n, err, len(want)) - } - if n, err := fd.PRead(ctx, ioseq, 0, ReadOptions{}); int(n) != len(want) && err != io.EOF { - t.Fatalf("PRead: got (%v, %v), wanted (%v, EOF)", n, err, len(want)) - } - if want == string(buf) { - t.Fatalf("PRead: got %v, wanted %v", string(buf), want) - } - - // Test Seek to 0 followed by Write. - want = "write2" - writeIOSeq = usermem.BytesIOSequence([]byte(want)) - if n, err := fd.Seek(ctx, 0, linux.SEEK_SET); n != 0 && err != nil { - t.Errorf("Seek: got err (%v, %v), wanted (0, nil)", n, err) - } - if n, err := fd.Write(ctx, writeIOSeq, WriteOptions{}); int(n) != len(want) && err != nil { - t.Errorf("Write: got err (%v, %v), wanted (%v, nil)", n, err, len(want)) - } - if n, err := fd.PRead(ctx, ioseq, 0, ReadOptions{}); int(n) != len(want) && err != io.EOF { - t.Fatalf("PRead: got (%v, %v), wanted (%v, EOF)", n, err, len(want)) - } - if want == string(buf) { - t.Fatalf("PRead: got %v, wanted %v", string(buf), want) - } - - // Test failure if offset != 0. - if n, err := fd.Seek(ctx, 1, linux.SEEK_SET); n != 0 && err != nil { - t.Errorf("Seek: got err (%v, %v), wanted (0, nil)", n, err) - } - if n, err := fd.Write(ctx, writeIOSeq, WriteOptions{}); n != 0 && err != syserror.EINVAL { - t.Errorf("Write: got err (%v, %v), wanted (0, EINVAL)", n, err) - } - if n, err := fd.PWrite(ctx, writeIOSeq, 2, WriteOptions{}); n != 0 && err != syserror.EINVAL { - t.Errorf("PWrite: got err (%v, %v), wanted (0, EINVAL)", n, err) - } -} diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 332decce6..332decce6 100644..100755 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go diff --git a/pkg/sentry/vfs/filesystem_impl_util.go b/pkg/sentry/vfs/filesystem_impl_util.go index 7315a588e..7315a588e 100644..100755 --- a/pkg/sentry/vfs/filesystem_impl_util.go +++ b/pkg/sentry/vfs/filesystem_impl_util.go diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go index bb9cada81..bb9cada81 100644..100755 --- a/pkg/sentry/vfs/filesystem_type.go +++ b/pkg/sentry/vfs/filesystem_type.go diff --git a/pkg/sentry/vfs/lock/BUILD b/pkg/sentry/vfs/lock/BUILD deleted file mode 100644 index d9ab063b7..000000000 --- a/pkg/sentry/vfs/lock/BUILD +++ /dev/null @@ -1,13 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "lock", - srcs = ["lock.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/sentry/fs/lock", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/vfs/lock/lock.go b/pkg/sentry/vfs/lock/lock.go deleted file mode 100644 index 724dfe743..000000000 --- a/pkg/sentry/vfs/lock/lock.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package lock provides POSIX and BSD style file locking for VFS2 file -// implementations. -// -// The actual implementations can be found in the lock package under -// sentry/fs/lock. -package lock - -import ( - fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/syserror" -) - -// FileLocks supports POSIX and BSD style locks, which correspond to fcntl(2) -// and flock(2) respectively in Linux. It can be embedded into various file -// implementations for VFS2 that support locking. -// -// Note that in Linux these two types of locks are _not_ cooperative, because -// race and deadlock conditions make merging them prohibitive. We do the same -// and keep them oblivious to each other. -type FileLocks struct { - // bsd is a set of BSD-style advisory file wide locks, see flock(2). - bsd fslock.Locks - - // posix is a set of POSIX-style regional advisory locks, see fcntl(2). - posix fslock.Locks -} - -// LockBSD tries to acquire a BSD-style lock on the entire file. -func (fl *FileLocks) LockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { - if fl.bsd.LockRegion(uid, t, fslock.LockRange{0, fslock.LockEOF}, block) { - return nil - } - return syserror.ErrWouldBlock -} - -// UnlockBSD releases a BSD-style lock on the entire file. -// -// This operation is always successful, even if there did not exist a lock on -// the requested region held by uid in the first place. -func (fl *FileLocks) UnlockBSD(uid fslock.UniqueID) { - fl.bsd.UnlockRegion(uid, fslock.LockRange{0, fslock.LockEOF}) -} - -// LockPOSIX tries to acquire a POSIX-style lock on a file region. -func (fl *FileLocks) LockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { - if fl.posix.LockRegion(uid, t, rng, block) { - return nil - } - return syserror.ErrWouldBlock -} - -// UnlockPOSIX releases a POSIX-style lock on a file region. -// -// This operation is always successful, even if there did not exist a lock on -// the requested region held by uid in the first place. -func (fl *FileLocks) UnlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) { - fl.posix.UnlockRegion(uid, rng) -} diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 4b68cabda..4b68cabda 100644..100755 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go diff --git a/pkg/sentry/vfs/mount_test.go b/pkg/sentry/vfs/mount_test.go deleted file mode 100644 index 3b933468d..000000000 --- a/pkg/sentry/vfs/mount_test.go +++ /dev/null @@ -1,458 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package vfs - -import ( - "fmt" - "runtime" - "testing" - - "gvisor.dev/gvisor/pkg/sync" -) - -func TestMountTableLookupEmpty(t *testing.T) { - var mt mountTable - mt.Init() - - parent := &Mount{} - point := &Dentry{} - if m := mt.Lookup(parent, point); m != nil { - t.Errorf("empty mountTable lookup: got %p, wanted nil", m) - } -} - -func TestMountTableInsertLookup(t *testing.T) { - var mt mountTable - mt.Init() - - mount := &Mount{} - mount.storeKey(VirtualDentry{&Mount{}, &Dentry{}}) - mt.Insert(mount) - - if m := mt.Lookup(mount.parent(), mount.point()); m != mount { - t.Errorf("mountTable positive lookup: got %p, wanted %p", m, mount) - } - - otherParent := &Mount{} - if m := mt.Lookup(otherParent, mount.point()); m != nil { - t.Errorf("mountTable lookup with wrong mount parent: got %p, wanted nil", m) - } - otherPoint := &Dentry{} - if m := mt.Lookup(mount.parent(), otherPoint); m != nil { - t.Errorf("mountTable lookup with wrong mount point: got %p, wanted nil", m) - } -} - -// TODO: concurrent lookup/insertion/removal - -// must be powers of 2 -var benchNumMounts = []int{1 << 2, 1 << 5, 1 << 8} - -// For all of the following: -// -// - BenchmarkMountTableFoo tests usage pattern "Foo" for mountTable. -// -// - BenchmarkMountMapFoo tests usage pattern "Foo" for a -// sync.RWMutex-protected map. (Mutator benchmarks do not use a RWMutex, since -// mountTable also requires external synchronization between mutators.) -// -// - BenchmarkMountSyncMapFoo tests usage pattern "Foo" for a sync.Map. -// -// ParallelLookup is by far the most common and performance-sensitive operation -// for this application. NegativeLookup is also important, but less so (only -// relevant with multiple mount namespaces and significant differences in -// mounts between them). Insertion and removal are benchmarked for -// completeness. -const enableComparativeBenchmarks = false - -func newBenchMount() *Mount { - mount := &Mount{} - mount.storeKey(VirtualDentry{&Mount{}, &Dentry{}}) - return mount -} - -func BenchmarkMountTableParallelLookup(b *testing.B) { - for numG, maxG := 1, runtime.GOMAXPROCS(0); numG >= 0 && numG <= maxG; numG *= 2 { - for _, numMounts := range benchNumMounts { - desc := fmt.Sprintf("%dx%d", numG, numMounts) - b.Run(desc, func(b *testing.B) { - var mt mountTable - mt.Init() - keys := make([]VirtualDentry, 0, numMounts) - for i := 0; i < numMounts; i++ { - mount := newBenchMount() - mt.Insert(mount) - keys = append(keys, mount.loadKey()) - } - - var ready sync.WaitGroup - begin := make(chan struct{}) - var end sync.WaitGroup - for g := 0; g < numG; g++ { - ready.Add(1) - end.Add(1) - go func() { - defer end.Done() - ready.Done() - <-begin - for i := 0; i < b.N; i++ { - k := keys[i&(numMounts-1)] - m := mt.Lookup(k.mount, k.dentry) - if m == nil { - b.Fatalf("lookup failed") - } - if parent := m.parent(); parent != k.mount { - b.Fatalf("lookup returned mount with parent %p, wanted %p", parent, k.mount) - } - if point := m.point(); point != k.dentry { - b.Fatalf("lookup returned mount with point %p, wanted %p", point, k.dentry) - } - } - }() - } - - ready.Wait() - b.ResetTimer() - close(begin) - end.Wait() - }) - } - } -} - -func BenchmarkMountMapParallelLookup(b *testing.B) { - if !enableComparativeBenchmarks { - b.Skipf("comparative benchmarks are disabled") - } - - for numG, maxG := 1, runtime.GOMAXPROCS(0); numG >= 0 && numG <= maxG; numG *= 2 { - for _, numMounts := range benchNumMounts { - desc := fmt.Sprintf("%dx%d", numG, numMounts) - b.Run(desc, func(b *testing.B) { - var mu sync.RWMutex - ms := make(map[VirtualDentry]*Mount) - keys := make([]VirtualDentry, 0, numMounts) - for i := 0; i < numMounts; i++ { - mount := newBenchMount() - key := mount.loadKey() - ms[key] = mount - keys = append(keys, key) - } - - var ready sync.WaitGroup - begin := make(chan struct{}) - var end sync.WaitGroup - for g := 0; g < numG; g++ { - ready.Add(1) - end.Add(1) - go func() { - defer end.Done() - ready.Done() - <-begin - for i := 0; i < b.N; i++ { - k := keys[i&(numMounts-1)] - mu.RLock() - m := ms[k] - mu.RUnlock() - if m == nil { - b.Fatalf("lookup failed") - } - if parent := m.parent(); parent != k.mount { - b.Fatalf("lookup returned mount with parent %p, wanted %p", parent, k.mount) - } - if point := m.point(); point != k.dentry { - b.Fatalf("lookup returned mount with point %p, wanted %p", point, k.dentry) - } - } - }() - } - - ready.Wait() - b.ResetTimer() - close(begin) - end.Wait() - }) - } - } -} - -func BenchmarkMountSyncMapParallelLookup(b *testing.B) { - if !enableComparativeBenchmarks { - b.Skipf("comparative benchmarks are disabled") - } - - for numG, maxG := 1, runtime.GOMAXPROCS(0); numG >= 0 && numG <= maxG; numG *= 2 { - for _, numMounts := range benchNumMounts { - desc := fmt.Sprintf("%dx%d", numG, numMounts) - b.Run(desc, func(b *testing.B) { - var ms sync.Map - keys := make([]VirtualDentry, 0, numMounts) - for i := 0; i < numMounts; i++ { - mount := newBenchMount() - key := mount.loadKey() - ms.Store(key, mount) - keys = append(keys, key) - } - - var ready sync.WaitGroup - begin := make(chan struct{}) - var end sync.WaitGroup - for g := 0; g < numG; g++ { - ready.Add(1) - end.Add(1) - go func() { - defer end.Done() - ready.Done() - <-begin - for i := 0; i < b.N; i++ { - k := keys[i&(numMounts-1)] - mi, ok := ms.Load(k) - if !ok { - b.Fatalf("lookup failed") - } - m := mi.(*Mount) - if parent := m.parent(); parent != k.mount { - b.Fatalf("lookup returned mount with parent %p, wanted %p", parent, k.mount) - } - if point := m.point(); point != k.dentry { - b.Fatalf("lookup returned mount with point %p, wanted %p", point, k.dentry) - } - } - }() - } - - ready.Wait() - b.ResetTimer() - close(begin) - end.Wait() - }) - } - } -} - -func BenchmarkMountTableNegativeLookup(b *testing.B) { - for _, numMounts := range benchNumMounts { - desc := fmt.Sprintf("%d", numMounts) - b.Run(desc, func(b *testing.B) { - var mt mountTable - mt.Init() - for i := 0; i < numMounts; i++ { - mt.Insert(newBenchMount()) - } - negkeys := make([]VirtualDentry, 0, numMounts) - for i := 0; i < numMounts; i++ { - negkeys = append(negkeys, VirtualDentry{ - mount: &Mount{}, - dentry: &Dentry{}, - }) - } - - b.ResetTimer() - for i := 0; i < b.N; i++ { - k := negkeys[i&(numMounts-1)] - m := mt.Lookup(k.mount, k.dentry) - if m != nil { - b.Fatalf("lookup got %p, wanted nil", m) - } - } - }) - } -} - -func BenchmarkMountMapNegativeLookup(b *testing.B) { - if !enableComparativeBenchmarks { - b.Skipf("comparative benchmarks are disabled") - } - - for _, numMounts := range benchNumMounts { - desc := fmt.Sprintf("%d", numMounts) - b.Run(desc, func(b *testing.B) { - var mu sync.RWMutex - ms := make(map[VirtualDentry]*Mount) - for i := 0; i < numMounts; i++ { - mount := newBenchMount() - ms[mount.loadKey()] = mount - } - negkeys := make([]VirtualDentry, 0, numMounts) - for i := 0; i < numMounts; i++ { - negkeys = append(negkeys, VirtualDentry{ - mount: &Mount{}, - dentry: &Dentry{}, - }) - } - - b.ResetTimer() - for i := 0; i < b.N; i++ { - k := negkeys[i&(numMounts-1)] - mu.RLock() - m := ms[k] - mu.RUnlock() - if m != nil { - b.Fatalf("lookup got %p, wanted nil", m) - } - } - }) - } -} - -func BenchmarkMountSyncMapNegativeLookup(b *testing.B) { - if !enableComparativeBenchmarks { - b.Skipf("comparative benchmarks are disabled") - } - - for _, numMounts := range benchNumMounts { - desc := fmt.Sprintf("%d", numMounts) - b.Run(desc, func(b *testing.B) { - var ms sync.Map - for i := 0; i < numMounts; i++ { - mount := newBenchMount() - ms.Store(mount.loadKey(), mount) - } - negkeys := make([]VirtualDentry, 0, numMounts) - for i := 0; i < numMounts; i++ { - negkeys = append(negkeys, VirtualDentry{ - mount: &Mount{}, - dentry: &Dentry{}, - }) - } - - b.ResetTimer() - for i := 0; i < b.N; i++ { - k := negkeys[i&(numMounts-1)] - m, _ := ms.Load(k) - if m != nil { - b.Fatalf("lookup got %p, wanted nil", m) - } - } - }) - } -} - -func BenchmarkMountTableInsert(b *testing.B) { - // Preallocate Mounts so that allocation time isn't included in the - // benchmark. - mounts := make([]*Mount, 0, b.N) - for i := 0; i < b.N; i++ { - mounts = append(mounts, newBenchMount()) - } - - var mt mountTable - mt.Init() - b.ResetTimer() - for i := range mounts { - mt.Insert(mounts[i]) - } -} - -func BenchmarkMountMapInsert(b *testing.B) { - if !enableComparativeBenchmarks { - b.Skipf("comparative benchmarks are disabled") - } - - // Preallocate Mounts so that allocation time isn't included in the - // benchmark. - mounts := make([]*Mount, 0, b.N) - for i := 0; i < b.N; i++ { - mounts = append(mounts, newBenchMount()) - } - - ms := make(map[VirtualDentry]*Mount) - b.ResetTimer() - for i := range mounts { - mount := mounts[i] - ms[mount.loadKey()] = mount - } -} - -func BenchmarkMountSyncMapInsert(b *testing.B) { - if !enableComparativeBenchmarks { - b.Skipf("comparative benchmarks are disabled") - } - - // Preallocate Mounts so that allocation time isn't included in the - // benchmark. - mounts := make([]*Mount, 0, b.N) - for i := 0; i < b.N; i++ { - mounts = append(mounts, newBenchMount()) - } - - var ms sync.Map - b.ResetTimer() - for i := range mounts { - mount := mounts[i] - ms.Store(mount.loadKey(), mount) - } -} - -func BenchmarkMountTableRemove(b *testing.B) { - mounts := make([]*Mount, 0, b.N) - for i := 0; i < b.N; i++ { - mounts = append(mounts, newBenchMount()) - } - var mt mountTable - mt.Init() - for i := range mounts { - mt.Insert(mounts[i]) - } - - b.ResetTimer() - for i := range mounts { - mt.Remove(mounts[i]) - } -} - -func BenchmarkMountMapRemove(b *testing.B) { - if !enableComparativeBenchmarks { - b.Skipf("comparative benchmarks are disabled") - } - - mounts := make([]*Mount, 0, b.N) - for i := 0; i < b.N; i++ { - mounts = append(mounts, newBenchMount()) - } - ms := make(map[VirtualDentry]*Mount) - for i := range mounts { - mount := mounts[i] - ms[mount.loadKey()] = mount - } - - b.ResetTimer() - for i := range mounts { - mount := mounts[i] - delete(ms, mount.loadKey()) - } -} - -func BenchmarkMountSyncMapRemove(b *testing.B) { - if !enableComparativeBenchmarks { - b.Skipf("comparative benchmarks are disabled") - } - - mounts := make([]*Mount, 0, b.N) - for i := 0; i < b.N; i++ { - mounts = append(mounts, newBenchMount()) - } - var ms sync.Map - for i := range mounts { - mount := mounts[i] - ms.Store(mount.loadKey(), mount) - } - - b.ResetTimer() - for i := range mounts { - mount := mounts[i] - ms.Delete(mount.loadKey()) - } -} diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index bc7581698..bc7581698 100644..100755 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 3e90dc4ed..3e90dc4ed 100644..100755 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go index b318c681a..b318c681a 100644..100755 --- a/pkg/sentry/vfs/pathname.go +++ b/pkg/sentry/vfs/pathname.go diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index f9647f90e..f9647f90e 100644..100755 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 8f31495da..8f31495da 100644..100755 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 03d1fb943..03d1fb943 100644..100755 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go diff --git a/pkg/sentry/vfs/vfs_state_autogen.go b/pkg/sentry/vfs/vfs_state_autogen.go new file mode 100755 index 000000000..d0221f0ff --- /dev/null +++ b/pkg/sentry/vfs/vfs_state_autogen.go @@ -0,0 +1,221 @@ +// automatically generated by stateify. + +// +build go1.12 +// +build !go1.15 + +package vfs + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Dentry) beforeSave() {} +func (x *Dentry) save(m state.Map) { + x.beforeSave() + m.Save("parent", &x.parent) + m.Save("name", &x.name) + m.Save("flags", &x.flags) + m.Save("mounts", &x.mounts) + m.Save("children", &x.children) + m.Save("impl", &x.impl) +} + +func (x *Dentry) afterLoad() {} +func (x *Dentry) load(m state.Map) { + m.Load("parent", &x.parent) + m.Load("name", &x.name) + m.Load("flags", &x.flags) + m.Load("mounts", &x.mounts) + m.Load("children", &x.children) + m.Load("impl", &x.impl) +} + +func (x *registeredDevice) beforeSave() {} +func (x *registeredDevice) save(m state.Map) { + x.beforeSave() + m.Save("dev", &x.dev) + m.Save("opts", &x.opts) +} + +func (x *registeredDevice) afterLoad() {} +func (x *registeredDevice) load(m state.Map) { + m.Load("dev", &x.dev) + m.Load("opts", &x.opts) +} + +func (x *RegisterDeviceOptions) beforeSave() {} +func (x *RegisterDeviceOptions) save(m state.Map) { + x.beforeSave() + m.Save("GroupName", &x.GroupName) +} + +func (x *RegisterDeviceOptions) afterLoad() {} +func (x *RegisterDeviceOptions) load(m state.Map) { + m.Load("GroupName", &x.GroupName) +} + +func (x *epollInterestList) beforeSave() {} +func (x *epollInterestList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *epollInterestList) afterLoad() {} +func (x *epollInterestList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *epollInterestEntry) beforeSave() {} +func (x *epollInterestEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *epollInterestEntry) afterLoad() {} +func (x *epollInterestEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *Filesystem) beforeSave() {} +func (x *Filesystem) save(m state.Map) { + x.beforeSave() + m.Save("refs", &x.refs) + m.Save("vfs", &x.vfs) + m.Save("impl", &x.impl) +} + +func (x *Filesystem) afterLoad() {} +func (x *Filesystem) load(m state.Map) { + m.Load("refs", &x.refs) + m.Load("vfs", &x.vfs) + m.Load("impl", &x.impl) +} + +func (x *registeredFilesystemType) beforeSave() {} +func (x *registeredFilesystemType) save(m state.Map) { + x.beforeSave() + m.Save("fsType", &x.fsType) + m.Save("opts", &x.opts) +} + +func (x *registeredFilesystemType) afterLoad() {} +func (x *registeredFilesystemType) load(m state.Map) { + m.Load("fsType", &x.fsType) + m.Load("opts", &x.opts) +} + +func (x *Mount) beforeSave() {} +func (x *Mount) save(m state.Map) { + x.beforeSave() + m.Save("vfs", &x.vfs) + m.Save("fs", &x.fs) + m.Save("root", &x.root) + m.Save("ID", &x.ID) + m.Save("key", &x.key) + m.Save("ns", &x.ns) + m.Save("refs", &x.refs) + m.Save("children", &x.children) + m.Save("umounted", &x.umounted) + m.Save("flags", &x.flags) + m.Save("writers", &x.writers) +} + +func (x *Mount) afterLoad() {} +func (x *Mount) load(m state.Map) { + m.Load("vfs", &x.vfs) + m.Load("fs", &x.fs) + m.Load("root", &x.root) + m.Load("ID", &x.ID) + m.Load("key", &x.key) + m.Load("ns", &x.ns) + m.Load("refs", &x.refs) + m.Load("children", &x.children) + m.Load("umounted", &x.umounted) + m.Load("flags", &x.flags) + m.Load("writers", &x.writers) +} + +func (x *MountNamespace) beforeSave() {} +func (x *MountNamespace) save(m state.Map) { + x.beforeSave() + m.Save("root", &x.root) + m.Save("refs", &x.refs) + m.Save("mountpoints", &x.mountpoints) +} + +func (x *MountNamespace) afterLoad() {} +func (x *MountNamespace) load(m state.Map) { + m.Load("root", &x.root) + m.Load("refs", &x.refs) + m.Load("mountpoints", &x.mountpoints) +} + +func (x *mountTable) beforeSave() {} +func (x *mountTable) save(m state.Map) { + x.beforeSave() + m.Save("seed", &x.seed) + m.Save("size", &x.size) +} + +func (x *mountTable) afterLoad() {} +func (x *mountTable) load(m state.Map) { + m.Load("seed", &x.seed) + m.Load("size", &x.size) +} + +func (x *VirtualFilesystem) beforeSave() {} +func (x *VirtualFilesystem) save(m state.Map) { + x.beforeSave() + m.Save("mounts", &x.mounts) + m.Save("mountpoints", &x.mountpoints) + m.Save("anonMount", &x.anonMount) + m.Save("devices", &x.devices) + m.Save("anonBlockDevMinorNext", &x.anonBlockDevMinorNext) + m.Save("anonBlockDevMinor", &x.anonBlockDevMinor) + m.Save("fsTypes", &x.fsTypes) + m.Save("filesystems", &x.filesystems) +} + +func (x *VirtualFilesystem) afterLoad() {} +func (x *VirtualFilesystem) load(m state.Map) { + m.Load("mounts", &x.mounts) + m.Load("mountpoints", &x.mountpoints) + m.Load("anonMount", &x.anonMount) + m.Load("devices", &x.devices) + m.Load("anonBlockDevMinorNext", &x.anonBlockDevMinorNext) + m.Load("anonBlockDevMinor", &x.anonBlockDevMinor) + m.Load("fsTypes", &x.fsTypes) + m.Load("filesystems", &x.filesystems) +} + +func (x *VirtualDentry) beforeSave() {} +func (x *VirtualDentry) save(m state.Map) { + x.beforeSave() + m.Save("mount", &x.mount) + m.Save("dentry", &x.dentry) +} + +func (x *VirtualDentry) afterLoad() {} +func (x *VirtualDentry) load(m state.Map) { + m.Load("mount", &x.mount) + m.Load("dentry", &x.dentry) +} + +func init() { + state.Register("pkg/sentry/vfs.Dentry", (*Dentry)(nil), state.Fns{Save: (*Dentry).save, Load: (*Dentry).load}) + state.Register("pkg/sentry/vfs.registeredDevice", (*registeredDevice)(nil), state.Fns{Save: (*registeredDevice).save, Load: (*registeredDevice).load}) + state.Register("pkg/sentry/vfs.RegisterDeviceOptions", (*RegisterDeviceOptions)(nil), state.Fns{Save: (*RegisterDeviceOptions).save, Load: (*RegisterDeviceOptions).load}) + state.Register("pkg/sentry/vfs.epollInterestList", (*epollInterestList)(nil), state.Fns{Save: (*epollInterestList).save, Load: (*epollInterestList).load}) + state.Register("pkg/sentry/vfs.epollInterestEntry", (*epollInterestEntry)(nil), state.Fns{Save: (*epollInterestEntry).save, Load: (*epollInterestEntry).load}) + state.Register("pkg/sentry/vfs.Filesystem", (*Filesystem)(nil), state.Fns{Save: (*Filesystem).save, Load: (*Filesystem).load}) + state.Register("pkg/sentry/vfs.registeredFilesystemType", (*registeredFilesystemType)(nil), state.Fns{Save: (*registeredFilesystemType).save, Load: (*registeredFilesystemType).load}) + state.Register("pkg/sentry/vfs.Mount", (*Mount)(nil), state.Fns{Save: (*Mount).save, Load: (*Mount).load}) + state.Register("pkg/sentry/vfs.MountNamespace", (*MountNamespace)(nil), state.Fns{Save: (*MountNamespace).save, Load: (*MountNamespace).load}) + state.Register("pkg/sentry/vfs.mountTable", (*mountTable)(nil), state.Fns{Save: (*mountTable).save, Load: (*mountTable).load}) + state.Register("pkg/sentry/vfs.VirtualFilesystem", (*VirtualFilesystem)(nil), state.Fns{Save: (*VirtualFilesystem).save, Load: (*VirtualFilesystem).load}) + state.Register("pkg/sentry/vfs.VirtualDentry", (*VirtualDentry)(nil), state.Fns{Save: (*VirtualDentry).save, Load: (*VirtualDentry).load}) +} diff --git a/pkg/sentry/watchdog/BUILD b/pkg/sentry/watchdog/BUILD deleted file mode 100644 index 1c5a1c9b6..000000000 --- a/pkg/sentry/watchdog/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "watchdog", - srcs = ["watchdog.go"], - visibility = ["//:sandbox"], - deps = [ - "//pkg/abi/linux", - "//pkg/log", - "//pkg/metric", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/time", - "//pkg/sync", - ], -) diff --git a/pkg/sentry/watchdog/watchdog_state_autogen.go b/pkg/sentry/watchdog/watchdog_state_autogen.go new file mode 100755 index 000000000..bce0200e7 --- /dev/null +++ b/pkg/sentry/watchdog/watchdog_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package watchdog |
