diff options
117 files changed, 2656 insertions, 707 deletions
diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index 2b789c4ec..a4bb62013 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -72,6 +72,9 @@ go_library( "//pkg/abi", "//pkg/binary", "//pkg/bits", + "//pkg/usermem", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/abi/linux/futex.go b/pkg/abi/linux/futex.go index 08bfde3b5..8138088a6 100644 --- a/pkg/abi/linux/futex.go +++ b/pkg/abi/linux/futex.go @@ -60,3 +60,21 @@ const ( FUTEX_WAITERS = 0x80000000 FUTEX_OWNER_DIED = 0x40000000 ) + +// FUTEX_BITSET_MATCH_ANY has all bits set. +const FUTEX_BITSET_MATCH_ANY = 0xffffffff + +// ROBUST_LIST_LIMIT protects against a deliberately circular list. +const ROBUST_LIST_LIMIT = 2048 + +// RobustListHead corresponds to Linux's struct robust_list_head. +// +// +marshal +type RobustListHead struct { + List uint64 + FutexOffset uint64 + ListOpPending uint64 +} + +// SizeOfRobustListHead is the size of a RobustListHead struct. +var SizeOfRobustListHead = (*RobustListHead)(nil).SizeBytes() diff --git a/pkg/abi/linux/netdevice.go b/pkg/abi/linux/netdevice.go index 7866352b4..0faf015c7 100644 --- a/pkg/abi/linux/netdevice.go +++ b/pkg/abi/linux/netdevice.go @@ -22,6 +22,8 @@ const ( ) // IFReq is an interface request. +// +// +marshal type IFReq struct { // IFName is an encoded name, normally null-terminated. This should be // accessed via the Name and SetName functions. @@ -79,6 +81,8 @@ type IFMap struct { // IFConf is used to return a list of interfaces and their addresses. See // netdevice(7) and struct ifconf for more detail on its use. +// +// +marshal type IFConf struct { Len int32 _ [4]byte // Pad to sizeof(struct ifconf). diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 46d8b0b42..a91f9f018 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -14,6 +14,14 @@ package linux +import ( + "io" + + "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" +) + // This file contains structures required to support netfilter, specifically // the iptables tool. @@ -76,6 +84,8 @@ const ( // IPTEntry is an iptable rule. It corresponds to struct ipt_entry in // include/uapi/linux/netfilter_ipv4/ip_tables.h. +// +// +marshal type IPTEntry struct { // IP is used to filter packets based on the IP header. IP IPTIP @@ -112,21 +122,41 @@ type IPTEntry struct { // SizeOfIPTEntry is the size of an IPTEntry. const SizeOfIPTEntry = 112 -// KernelIPTEntry is identical to IPTEntry, but includes the Elems field. This -// struct marshaled via the binary package to write an IPTEntry to userspace. +// KernelIPTEntry is identical to IPTEntry, but includes the Elems field. +// KernelIPTEntry itself is not Marshallable but it implements some methods of +// marshal.Marshallable that help in other implementations of Marshallable. type KernelIPTEntry struct { - IPTEntry + Entry IPTEntry // Elems holds the data for all this rule's matches followed by the // target. It is variable length -- users have to iterate over any // matches and use TargetOffset and NextOffset to make sense of the // data. - Elems []byte + Elems primitive.ByteSlice +} + +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (ke *KernelIPTEntry) SizeBytes() int { + return ke.Entry.SizeBytes() + ke.Elems.SizeBytes() +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (ke *KernelIPTEntry) MarshalBytes(dst []byte) { + ke.Entry.MarshalBytes(dst) + ke.Elems.MarshalBytes(dst[ke.Entry.SizeBytes():]) +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (ke *KernelIPTEntry) UnmarshalBytes(src []byte) { + ke.Entry.UnmarshalBytes(src) + ke.Elems.UnmarshalBytes(src[ke.Entry.SizeBytes():]) } // IPTIP contains information for matching a packet's IP header. // It corresponds to struct ipt_ip in // include/uapi/linux/netfilter_ipv4/ip_tables.h. +// +// +marshal type IPTIP struct { // Src is the source IP address. Src InetAddr @@ -189,6 +219,8 @@ const SizeOfIPTIP = 84 // XTCounters holds packet and byte counts for a rule. It corresponds to struct // xt_counters in include/uapi/linux/netfilter/x_tables.h. +// +// +marshal type XTCounters struct { // Pcnt is the packet count. Pcnt uint64 @@ -321,6 +353,8 @@ const SizeOfXTRedirectTarget = 56 // IPTGetinfo is the argument for the IPT_SO_GET_INFO sockopt. It corresponds // to struct ipt_getinfo in include/uapi/linux/netfilter_ipv4/ip_tables.h. +// +// +marshal type IPTGetinfo struct { Name TableName ValidHooks uint32 @@ -336,6 +370,8 @@ const SizeOfIPTGetinfo = 84 // IPTGetEntries is the argument for the IPT_SO_GET_ENTRIES sockopt. It // corresponds to struct ipt_get_entries in // include/uapi/linux/netfilter_ipv4/ip_tables.h. +// +// +marshal type IPTGetEntries struct { Name TableName Size uint32 @@ -350,13 +386,103 @@ type IPTGetEntries struct { const SizeOfIPTGetEntries = 40 // KernelIPTGetEntries is identical to IPTGetEntries, but includes the -// Entrytable field. This struct marshaled via the binary package to write an -// KernelIPTGetEntries to userspace. +// Entrytable field. This has been manually made marshal.Marshallable since it +// is dynamically sized. type KernelIPTGetEntries struct { IPTGetEntries Entrytable []KernelIPTEntry } +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (ke *KernelIPTGetEntries) SizeBytes() int { + res := ke.IPTGetEntries.SizeBytes() + for _, entry := range ke.Entrytable { + res += entry.SizeBytes() + } + return res +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (ke *KernelIPTGetEntries) MarshalBytes(dst []byte) { + ke.IPTGetEntries.MarshalBytes(dst) + marshalledUntil := ke.IPTGetEntries.SizeBytes() + for i := 0; i < len(ke.Entrytable); i++ { + ke.Entrytable[i].MarshalBytes(dst[marshalledUntil:]) + marshalledUntil += ke.Entrytable[i].SizeBytes() + } +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (ke *KernelIPTGetEntries) UnmarshalBytes(src []byte) { + ke.IPTGetEntries.UnmarshalBytes(src) + unmarshalledUntil := ke.IPTGetEntries.SizeBytes() + for i := 0; i < len(ke.Entrytable); i++ { + ke.Entrytable[i].UnmarshalBytes(src[unmarshalledUntil:]) + unmarshalledUntil += ke.Entrytable[i].SizeBytes() + } +} + +// Packed implements marshal.Marshallable.Packed. +func (ke *KernelIPTGetEntries) Packed() bool { + // KernelIPTGetEntries isn't packed because the ke.Entrytable contains an + // indirection to the actual data we want to marshal (the slice data + // pointer), and the memory for KernelIPTGetEntries contains the slice + // header which we don't want to marshal. + return false +} + +// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. +func (ke *KernelIPTGetEntries) MarshalUnsafe(dst []byte) { + // Fall back to safe Marshal because the type in not packed. + ke.MarshalBytes(dst) +} + +// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. +func (ke *KernelIPTGetEntries) UnmarshalUnsafe(src []byte) { + // Fall back to safe Unmarshal because the type in not packed. + ke.UnmarshalBytes(src) +} + +// CopyIn implements marshal.Marshallable.CopyIn. +func (ke *KernelIPTGetEntries) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { + buf := task.CopyScratchBuffer(ke.SizeBytes()) // escapes: okay. + length, err := task.CopyInBytes(addr, buf) // escapes: okay. + // Unmarshal unconditionally. If we had a short copy-in, this results in a + // partially unmarshalled struct. + ke.UnmarshalBytes(buf) // escapes: fallback. + return length, err +} + +// CopyOut implements marshal.Marshallable.CopyOut. +func (ke *KernelIPTGetEntries) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) { + // Type KernelIPTGetEntries doesn't have a packed layout in memory, fall + // back to MarshalBytes. + return task.CopyOutBytes(addr, ke.marshalAll(task)) +} + +// CopyOutN implements marshal.Marshallable.CopyOutN. +func (ke *KernelIPTGetEntries) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) { + // Type KernelIPTGetEntries doesn't have a packed layout in memory, fall + // back to MarshalBytes. + return task.CopyOutBytes(addr, ke.marshalAll(task)[:limit]) +} + +func (ke *KernelIPTGetEntries) marshalAll(task marshal.Task) []byte { + buf := task.CopyScratchBuffer(ke.SizeBytes()) + ke.MarshalBytes(buf) + return buf +} + +// WriteTo implements io.WriterTo.WriteTo. +func (ke *KernelIPTGetEntries) WriteTo(w io.Writer) (int64, error) { + buf := make([]byte, ke.SizeBytes()) + ke.MarshalBytes(buf) + length, err := w.Write(buf) + return int64(length), err +} + +var _ marshal.Marshallable = (*KernelIPTGetEntries)(nil) + // IPTReplace is the argument for the IPT_SO_SET_REPLACE sockopt. It // corresponds to struct ipt_replace in // include/uapi/linux/netfilter_ipv4/ip_tables.h. @@ -374,12 +500,6 @@ type IPTReplace struct { // Entries [0]IPTEntry } -// KernelIPTReplace is identical to IPTReplace, but includes the Entries field. -type KernelIPTReplace struct { - IPTReplace - Entries [0]IPTEntry -} - // SizeOfIPTReplace is the size of an IPTReplace. const SizeOfIPTReplace = 96 @@ -392,6 +512,8 @@ func (en ExtensionName) String() string { } // TableName holds the name of a netfilter table. +// +// +marshal type TableName [XT_TABLE_MAXNAMELEN]byte // String implements fmt.Stringer. diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index 95337c168..c24a8216e 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -234,6 +234,8 @@ const ( const SockAddrMax = 128 // InetAddr is struct in_addr, from uapi/linux/in.h. +// +// +marshal type InetAddr [4]byte // SockAddrInet is struct sockaddr_in, from uapi/linux/in.h. @@ -303,6 +305,8 @@ func (s *SockAddrUnix) implementsSockAddr() {} func (s *SockAddrNetlink) implementsSockAddr() {} // Linger is struct linger, from include/linux/socket.h. +// +// +marshal type Linger struct { OnOff int32 Linger int32 @@ -317,6 +321,8 @@ const SizeOfLinger = 8 // the end of this struct or within existing unusued space, so its size grows // over time. The current iteration is based on linux v4.17. New versions are // always backwards compatible. +// +// +marshal type TCPInfo struct { State uint8 CaState uint8 @@ -414,6 +420,8 @@ var SizeOfControlMessageHeader = int(binary.Size(ControlMessageHeader{})) // A ControlMessageCredentials is an SCM_CREDENTIALS socket control message. // // ControlMessageCredentials represents struct ucred from linux/socket.h. +// +// +marshal type ControlMessageCredentials struct { PID int32 UID uint32 diff --git a/pkg/p9/messages.go b/pkg/p9/messages.go index 57b89ad7d..2cb59f934 100644 --- a/pkg/p9/messages.go +++ b/pkg/p9/messages.go @@ -2506,7 +2506,7 @@ type msgFactory struct { var msgRegistry registry type registry struct { - factories [math.MaxUint8]msgFactory + factories [math.MaxUint8 + 1]msgFactory // largestFixedSize is computed so that given some message size M, you can // compute the maximum payload size (e.g. for Twrite, Rread) with diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index daba8b172..fd95eb2d2 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -28,7 +28,14 @@ import ( ) // Registers represents the CPU registers for this architecture. -type Registers = linux.PtraceRegs +// +// +stateify savable +type Registers struct { + linux.PtraceRegs + + // TPIDR_EL0 is the EL0 Read/Write Software Thread ID Register. + TPIDR_EL0 uint64 +} const ( // SyscallWidth is the width of insturctions. @@ -101,9 +108,6 @@ type State struct { // Our floating point state. aarch64FPState `state:"wait"` - // TLS pointer - TPValue uint64 - // FeatureSet is a pointer to the currently active feature set. FeatureSet *cpuid.FeatureSet @@ -157,7 +161,6 @@ func (s *State) Fork() State { return State{ Regs: s.Regs, aarch64FPState: s.aarch64FPState.fork(), - TPValue: s.TPValue, FeatureSet: s.FeatureSet, OrigR0: s.OrigR0, } @@ -241,18 +244,18 @@ func (s *State) ptraceGetRegs() Registers { return s.Regs } -var registersSize = (*Registers)(nil).SizeBytes() +var ptraceRegistersSize = (*linux.PtraceRegs)(nil).SizeBytes() // PtraceSetRegs implements Context.PtraceSetRegs. func (s *State) PtraceSetRegs(src io.Reader) (int, error) { var regs Registers - buf := make([]byte, registersSize) + buf := make([]byte, ptraceRegistersSize) if _, err := io.ReadFull(src, buf); err != nil { return 0, err } regs.UnmarshalUnsafe(buf) s.Regs = regs - return registersSize, nil + return ptraceRegistersSize, nil } // PtraceGetFPRegs implements Context.PtraceGetFPRegs. @@ -278,7 +281,7 @@ const ( func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < registersSize { + if maxlen < ptraceRegistersSize { return 0, syserror.EFAULT } return s.PtraceGetRegs(dst) @@ -291,7 +294,7 @@ func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < registersSize { + if maxlen < ptraceRegistersSize { return 0, syserror.EFAULT } return s.PtraceSetRegs(src) diff --git a/pkg/sentry/arch/arch_amd64.go b/pkg/sentry/arch/arch_amd64.go index 3b3a0a272..1c3e3c14c 100644 --- a/pkg/sentry/arch/arch_amd64.go +++ b/pkg/sentry/arch/arch_amd64.go @@ -300,7 +300,7 @@ func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { // PTRACE_PEEKUSER and PTRACE_POKEUSER are only effective on regs and // u_debugreg, returning 0 or silently no-oping for other fields // respectively. - if addr < uintptr(registersSize) { + if addr < uintptr(ptraceRegistersSize) { regs := c.ptraceGetRegs() buf := make([]byte, regs.SizeBytes()) regs.MarshalUnsafe(buf) @@ -315,7 +315,7 @@ func (c *context64) PtracePokeUser(addr, data uintptr) error { if addr&7 != 0 || addr >= userStructSize { return syscall.EIO } - if addr < uintptr(registersSize) { + if addr < uintptr(ptraceRegistersSize) { regs := c.ptraceGetRegs() buf := make([]byte, regs.SizeBytes()) regs.MarshalUnsafe(buf) diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index ada7ac7b8..cabbf60e0 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -142,7 +142,7 @@ func (c *context64) SetStack(value uintptr) { // TLS returns the current TLS pointer. func (c *context64) TLS() uintptr { - return uintptr(c.TPValue) + return uintptr(c.Regs.TPIDR_EL0) } // SetTLS sets the current TLS pointer. Returns false if value is invalid. @@ -151,7 +151,7 @@ func (c *context64) SetTLS(value uintptr) bool { return false } - c.TPValue = uint64(value) + c.Regs.TPIDR_EL0 = uint64(value) return true } diff --git a/pkg/sentry/arch/arch_x86.go b/pkg/sentry/arch/arch_x86.go index dc458b37f..b9405b320 100644 --- a/pkg/sentry/arch/arch_x86.go +++ b/pkg/sentry/arch/arch_x86.go @@ -31,7 +31,11 @@ import ( ) // Registers represents the CPU registers for this architecture. -type Registers = linux.PtraceRegs +// +// +stateify savable +type Registers struct { + linux.PtraceRegs +} // System-related constants for x86. const ( @@ -311,12 +315,12 @@ func (s *State) ptraceGetRegs() Registers { return regs } -var registersSize = (*Registers)(nil).SizeBytes() +var ptraceRegistersSize = (*linux.PtraceRegs)(nil).SizeBytes() // PtraceSetRegs implements Context.PtraceSetRegs. func (s *State) PtraceSetRegs(src io.Reader) (int, error) { var regs Registers - buf := make([]byte, registersSize) + buf := make([]byte, ptraceRegistersSize) if _, err := io.ReadFull(src, buf); err != nil { return 0, err } @@ -374,7 +378,7 @@ func (s *State) PtraceSetRegs(src io.Reader) (int, error) { } regs.Eflags = (s.Regs.Eflags &^ eflagsPtraceMutable) | (regs.Eflags & eflagsPtraceMutable) s.Regs = regs - return registersSize, nil + return ptraceRegistersSize, nil } // isUserSegmentSelector returns true if the given segment selector specifies a @@ -543,7 +547,7 @@ const ( func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < registersSize { + if maxlen < ptraceRegistersSize { return 0, syserror.EFAULT } return s.PtraceGetRegs(dst) @@ -563,7 +567,7 @@ func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) { switch regset { case _NT_PRSTATUS: - if maxlen < registersSize { + if maxlen < ptraceRegistersSize { return 0, syserror.EFAULT } return s.PtraceSetRegs(src) diff --git a/pkg/sentry/fs/file_operations.go b/pkg/sentry/fs/file_operations.go index beba0f771..f5537411e 100644 --- a/pkg/sentry/fs/file_operations.go +++ b/pkg/sentry/fs/file_operations.go @@ -160,6 +160,7 @@ type FileOperations interface { // refer. // // Preconditions: The AddressSpace (if any) that io refers to is activated. + // Must only be called from a task goroutine. Ioctl(ctx context.Context, file *File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) } diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index 2018b978a..a91cae3ef 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -132,7 +132,7 @@ func (sfd *slaveFileDescription) Write(ctx context.Context, src usermem.IOSequen return sfd.inode.t.ld.outputQueueWrite(ctx, src) } -// Ioctl implements vfs.FileDescripionImpl.Ioctl. +// Ioctl implements vfs.FileDescriptionImpl.Ioctl. func (sfd *slaveFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { switch cmd := args[1].Uint(); cmd { case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD index 3e00c2abb..737007748 100644 --- a/pkg/sentry/fsimpl/fuse/BUILD +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -6,13 +6,17 @@ go_library( name = "fuse", srcs = [ "dev.go", + "fusefs.go", ], visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", + "//pkg/log", "//pkg/sentry/fsimpl/devtmpfs", + "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel", + "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/usermem", diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index dc33268af..c9e12a94f 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -51,6 +51,9 @@ type DeviceFD struct { vfs.DentryMetadataFileDescriptionImpl vfs.NoLockFD + // mounted specifies whether a FUSE filesystem was mounted using the DeviceFD. + mounted bool + // TODO(gvisor.dev/issue/2987): Add all the data structures needed to enqueue // and deque requests, control synchronization and establish communication // between the FUSE kernel module and the /dev/fuse character device. @@ -61,26 +64,51 @@ func (fd *DeviceFD) Release() {} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *DeviceFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } // Read implements vfs.FileDescriptionImpl.Read. func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *DeviceFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } // Write implements vfs.FileDescriptionImpl.Write. func (fd *DeviceFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } // Seek implements vfs.FileDescriptionImpl.Seek. func (fd *DeviceFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. + if !fd.mounted { + return 0, syserror.EPERM + } + return 0, syserror.ENOSYS } diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go new file mode 100644 index 000000000..f7775fb9b --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -0,0 +1,200 @@ +// 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 fuse implements fusefs. +package fuse + +import ( + "strconv" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" + "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 = "fuse" + +// FilesystemType implements vfs.FilesystemType. +type FilesystemType struct{} + +type filesystemOptions struct { + // userID specifies the numeric uid of the mount owner. + // This option should not be specified by the filesystem owner. + // It is set by libfuse (or, if libfuse is not used, must be set + // by the filesystem itself). For more information, see man page + // for fuse(8) + userID uint32 + + // groupID specifies the numeric gid of the mount owner. + // This option should not be specified by the filesystem owner. + // It is set by libfuse (or, if libfuse is not used, must be set + // by the filesystem itself). For more information, see man page + // for fuse(8) + groupID uint32 + + // rootMode specifies the the file mode of the filesystem's root. + rootMode linux.FileMode +} + +// filesystem implements vfs.FilesystemImpl. +type filesystem struct { + kernfs.Filesystem + devMinor uint32 + + // fuseFD is the FD returned when opening /dev/fuse. It is used for communication + // between the FUSE server daemon and the sentry fusefs. + fuseFD *DeviceFD + + // opts is the options the fusefs is initialized with. + opts filesystemOptions +} + +// Name implements vfs.FilesystemType.Name. +func (FilesystemType) Name() string { + return Name +} + +// 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) { + devMinor, err := vfsObj.GetAnonBlockDevMinor() + if err != nil { + return nil, nil, err + } + + var fsopts filesystemOptions + mopts := vfs.GenericParseMountOptions(opts.Data) + deviceDescriptorStr, ok := mopts["fd"] + if !ok { + log.Warningf("%s.GetFilesystem: communication file descriptor N (obtained by opening /dev/fuse) must be specified as 'fd=N'", fsType.Name()) + return nil, nil, syserror.EINVAL + } + delete(mopts, "fd") + + deviceDescriptor, err := strconv.ParseInt(deviceDescriptorStr, 10 /* base */, 32 /* bitSize */) + if err != nil { + return nil, nil, err + } + + kernelTask := kernel.TaskFromContext(ctx) + if kernelTask == nil { + log.Warningf("%s.GetFilesystem: couldn't get kernel task from context", fsType.Name()) + return nil, nil, syserror.EINVAL + } + fuseFd := kernelTask.GetFileVFS2(int32(deviceDescriptor)) + + // Parse and set all the other supported FUSE mount options. + // TODO: Expand the supported mount options. + if userIDStr, ok := mopts["user_id"]; ok { + delete(mopts, "user_id") + userID, err := strconv.ParseUint(userIDStr, 10, 32) + if err != nil { + log.Warningf("%s.GetFilesystem: invalid user_id: user_id=%s", fsType.Name(), userIDStr) + return nil, nil, syserror.EINVAL + } + fsopts.userID = uint32(userID) + } + + if groupIDStr, ok := mopts["group_id"]; ok { + delete(mopts, "group_id") + groupID, err := strconv.ParseUint(groupIDStr, 10, 32) + if err != nil { + log.Warningf("%s.GetFilesystem: invalid group_id: group_id=%s", fsType.Name(), groupIDStr) + return nil, nil, syserror.EINVAL + } + fsopts.groupID = uint32(groupID) + } + + rootMode := linux.FileMode(0777) + modeStr, ok := mopts["rootmode"] + if ok { + delete(mopts, "rootmode") + mode, err := strconv.ParseUint(modeStr, 8, 32) + if err != nil { + log.Warningf("%s.GetFilesystem: invalid mode: %q", fsType.Name(), modeStr) + return nil, nil, syserror.EINVAL + } + rootMode = linux.FileMode(mode) + } + fsopts.rootMode = rootMode + + // Check for unparsed options. + if len(mopts) != 0 { + log.Warningf("%s.GetFilesystem: unknown options: %v", fsType.Name(), mopts) + return nil, nil, syserror.EINVAL + } + + // Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to + // mount a FUSE filesystem. + fuseFD := fuseFd.Impl().(*DeviceFD) + fuseFD.mounted = true + + fs := &filesystem{ + devMinor: devMinor, + fuseFD: fuseFD, + opts: fsopts, + } + + fs.VFSFilesystem().Init(vfsObj, &fsType, fs) + + // TODO: dispatch a FUSE_INIT request to the FUSE daemon server before + // returning. Mount will not block on this dispatched request. + + // root is the fusefs root directory. + root := fs.newInode(creds, fsopts.rootMode) + + return fs.VFSFilesystem(), root.VFSDentry(), nil +} + +// Release implements vfs.FilesystemImpl.Release. +func (fs *filesystem) Release() { + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) + fs.Filesystem.Release() +} + +// Inode implements kernfs.Inode. +type Inode struct { + kernfs.InodeAttrs + kernfs.InodeNoDynamicLookup + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + kernfs.OrderedChildren + + locks vfs.FileLocks + + dentry kernfs.Dentry +} + +func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *kernfs.Dentry { + i := &Inode{} + i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755) + i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + i.dentry.Init(i) + + return &i.dentry +} + +// Open implements kernfs.Inode.Open. +func (i *Inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts) + if err != nil { + return nil, err + } + return fd.VFSFileDescription(), nil +} diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index cd5f5049e..00e3c99cd 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -150,11 +150,9 @@ afterSymlink: return nil, err } if d != d.parent && !d.cachedMetadataAuthoritative() { - _, attrMask, attr, err := d.parent.file.getAttr(ctx, dentryAttrMask()) - if err != nil { + if err := d.parent.updateFromGetattr(ctx); err != nil { return nil, err } - d.parent.updateFromP9Attrs(attrMask, &attr) } rp.Advance() return d.parent, nil @@ -209,17 +207,28 @@ func (fs *filesystem) getChildLocked(ctx context.Context, vfsObj *vfs.VirtualFil // Preconditions: As for getChildLocked. !parent.isSynthetic(). func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, child *dentry, ds **[]*dentry) (*dentry, error) { + if child != nil { + // Need to lock child.metadataMu because we might be updating child + // metadata. We need to hold the lock *before* getting metadata from the + // server and release it after updating local metadata. + child.metadataMu.Lock() + } qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) if err != nil && err != syserror.ENOENT { + if child != nil { + child.metadataMu.Unlock() + } return nil, err } if child != nil { if !file.isNil() && inoFromPath(qid.Path) == child.ino { // The file at this path hasn't changed. Just update cached metadata. file.close(ctx) - child.updateFromP9Attrs(attrMask, &attr) + child.updateFromP9AttrsLocked(attrMask, &attr) + child.metadataMu.Unlock() return child, nil } + child.metadataMu.Unlock() if file.isNil() && child.isSynthetic() { // We have a synthetic file, and no remote file has arisen to // replace it. @@ -1325,7 +1334,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.renameMuRUnlockAndCheckCaching(&ds) return err } - if err := d.setStat(ctx, rp.Credentials(), &opts.Stat, rp.Mount()); err != nil { + if err := d.setStat(ctx, rp.Credentials(), &opts, rp.Mount()); err != nil { fs.renameMuRUnlockAndCheckCaching(&ds) return err } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index b74d489a0..e20de84b5 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -785,8 +785,8 @@ func (d *dentry) cachedMetadataAuthoritative() bool { // 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() +// Precondition: d.metadataMu must be locked. +func (d *dentry) updateFromP9AttrsLocked(mask p9.AttrMask, attr *p9.Attr) { if mask.Mode { if got, want := uint32(attr.Mode.FileType()), d.fileType(); got != want { d.metadataMu.Unlock() @@ -822,7 +822,6 @@ func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { if mask.Size { d.updateFileSizeLocked(attr.Size) } - d.metadataMu.Unlock() } // Preconditions: !d.isSynthetic() @@ -834,6 +833,10 @@ func (d *dentry) updateFromGetattr(ctx context.Context) error { file p9file handleMuRLocked bool ) + // d.metadataMu must be locked *before* we getAttr so that we do not end up + // updating stale attributes in d.updateFromP9AttrsLocked(). + d.metadataMu.Lock() + defer d.metadataMu.Unlock() d.handleMu.RLock() if !d.handle.file.isNil() { file = d.handle.file @@ -849,7 +852,7 @@ func (d *dentry) updateFromGetattr(ctx context.Context) error { if err != nil { return err } - d.updateFromP9Attrs(attrMask, &attr) + d.updateFromP9AttrsLocked(attrMask, &attr) return nil } @@ -885,7 +888,8 @@ func (d *dentry) statTo(stat *linux.Statx) { stat.DevMinor = d.fs.devMinor } -func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { +func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs.SetStatOptions, mnt *vfs.Mount) error { + stat := &opts.Stat if stat.Mask == 0 { return nil } @@ -893,7 +897,7 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin 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 { + if err := vfs.CheckSetStat(ctx, creds, opts, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err } if err := mnt.CheckBeginWrite(); err != nil { @@ -934,6 +938,17 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin } if !d.isSynthetic() { if stat.Mask != 0 { + if stat.Mask&linux.STATX_SIZE != 0 { + // Check whether to allow a truncate request to be made. + switch d.mode & linux.S_IFMT { + case linux.S_IFREG: + // Allow. + case linux.S_IFDIR: + return syserror.EISDIR + default: + return syserror.EINVAL + } + } if err := d.file.setAttr(ctx, p9.SetAttrMask{ Permissions: stat.Mask&linux.STATX_MODE != 0, UID: stat.Mask&linux.STATX_UID != 0, @@ -1495,7 +1510,7 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - if err := fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, fd.vfsfd.Mount()); err != nil { + if err := fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts, fd.vfsfd.Mount()); err != nil { return err } if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 1a88cb657..c894f2ca0 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -373,7 +373,7 @@ func (i *inode) fstat(fs *filesystem) (linux.Statx, error) { // SetStat implements kernfs.Inode. func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { - s := opts.Stat + s := &opts.Stat m := s.Mask if m == 0 { @@ -386,7 +386,7 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre if err := syscall.Fstat(i.hostFD, &hostStat); err != nil { return err } - if err := vfs.CheckSetStat(ctx, creds, &s, linux.FileMode(hostStat.Mode&linux.PermissionsMask), auth.KUID(hostStat.Uid), auth.KGID(hostStat.Gid)); err != nil { + if err := vfs.CheckSetStat(ctx, creds, &opts, linux.FileMode(hostStat.Mode), auth.KUID(hostStat.Uid), auth.KGID(hostStat.Gid)); err != nil { return err } @@ -396,6 +396,9 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre } } if m&linux.STATX_SIZE != 0 { + if hostStat.Mode&linux.S_IFMT != linux.S_IFREG { + return syserror.EINVAL + } if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil { return err } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 3f0aea73a..1d37ccb98 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -112,7 +112,7 @@ func (fd *GenericDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence return fd.DirectoryFileDescriptionDefaultImpl.PWrite(ctx, src, offset, opts) } -// Release implements vfs.FileDecriptionImpl.Release. +// Release implements vfs.FileDescriptionImpl.Release. func (fd *GenericDirectoryFD) Release() {} func (fd *GenericDirectoryFD) filesystem() *vfs.Filesystem { @@ -123,7 +123,7 @@ func (fd *GenericDirectoryFD) inode() Inode { return fd.vfsfd.VirtualDentry().Dentry().Impl().(*Dentry).inode } -// IterDirents implements vfs.FileDecriptionImpl.IterDirents. IterDirents holds +// IterDirents implements vfs.FileDescriptionImpl.IterDirents. IterDirents holds // o.mu when calling cb. func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { fd.mu.Lock() @@ -198,7 +198,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent return err } -// Seek implements vfs.FileDecriptionImpl.Seek. +// Seek implements vfs.FileDescriptionImpl.Seek. func (fd *GenericDirectoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { fd.mu.Lock() defer fd.mu.Unlock() diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 2ab3f1761..579e627f0 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -267,7 +267,7 @@ func (a *InodeAttrs) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *aut 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 { + if err := vfs.CheckSetStat(ctx, creds, &opts, a.Mode(), auth.KUID(atomic.LoadUint32(&a.uid)), auth.KGID(atomic.LoadUint32(&a.gid))); err != nil { return err } diff --git a/pkg/sentry/fsimpl/overlay/filesystem.go b/pkg/sentry/fsimpl/overlay/filesystem.go index ff82e1f20..6b705e955 100644 --- a/pkg/sentry/fsimpl/overlay/filesystem.go +++ b/pkg/sentry/fsimpl/overlay/filesystem.go @@ -1104,7 +1104,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts } mode := linux.FileMode(atomic.LoadUint32(&d.mode)) - if err := vfs.CheckSetStat(ctx, rp.Credentials(), &opts.Stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, rp.Credentials(), &opts, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err } mnt := rp.Mount() diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index a3c1f7a8d..c0749e711 100644 --- a/pkg/sentry/fsimpl/overlay/non_directory.go +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -151,7 +151,7 @@ func (fd *nonDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux func (fd *nonDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { d := fd.dentry() mode := linux.FileMode(atomic.LoadUint32(&d.mode)) - if err := vfs.CheckSetStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { + if err := vfs.CheckSetStat(ctx, auth.CredentialsFromContext(ctx), &opts, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err } mnt := fd.vfsfd.Mount() @@ -176,7 +176,7 @@ func (fd *nonDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) return nil } -// StatFS implements vfs.FileDesciptionImpl.StatFS. +// StatFS implements vfs.FileDescriptionImpl.StatFS. func (fd *nonDirectoryFD) StatFS(ctx context.Context) (linux.Statfs, error) { return fd.filesystem().statFS(ctx) } diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index dad4db1a7..79c2725f3 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -128,7 +128,7 @@ func (fd *subtasksFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallbac return fd.GenericDirectoryFD.IterDirents(ctx, cb) } -// Seek implements vfs.FileDecriptionImpl.Seek. +// Seek implements vfs.FileDescriptionImpl.Seek. func (fd *subtasksFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { if fd.task.ExitState() >= kernel.TaskExitZombie { return 0, syserror.ENOENT diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index a0f20c2d4..ef210a69b 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -649,7 +649,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.mu.RUnlock() return err } - if err := d.inode.setStat(ctx, rp.Credentials(), &opts.Stat); err != nil { + if err := d.inode.setStat(ctx, rp.Credentials(), &opts); err != nil { fs.mu.RUnlock() return err } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index d7f4f0779..2545d88e9 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -452,7 +452,8 @@ func (i *inode) statTo(stat *linux.Statx) { } } -func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx) error { +func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, opts *vfs.SetStatOptions) error { + stat := &opts.Stat if stat.Mask == 0 { return nil } @@ -460,7 +461,7 @@ func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linu 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 { + if err := vfs.CheckSetStat(ctx, creds, opts, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { return err } i.mu.Lock() @@ -695,7 +696,7 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) d := fd.dentry() - if err := d.inode.setStat(ctx, creds, &opts.Stat); err != nil { + if err := d.inode.setStat(ctx, creds, &opts); err != nil { return err } diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 25fe1921b..f6886a758 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -132,6 +132,7 @@ go_library( "task_stop.go", "task_syscall.go", "task_usermem.go", + "task_work.go", "thread_group.go", "threads.go", "timekeeper.go", diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go index 732e66da4..bcc1b29a8 100644 --- a/pkg/sentry/kernel/futex/futex.go +++ b/pkg/sentry/kernel/futex/futex.go @@ -717,10 +717,10 @@ func (m *Manager) lockPILocked(w *Waiter, t Target, addr usermem.Addr, tid uint3 } } -// UnlockPI unlock the futex following the Priority-inheritance futex -// rules. The address provided must contain the caller's TID. If there are -// waiters, TID of the next waiter (FIFO) is set to the given address, and the -// waiter woken up. If there are no waiters, 0 is set to the address. +// UnlockPI unlocks the futex following the Priority-inheritance futex rules. +// The address provided must contain the caller's TID. If there are waiters, +// TID of the next waiter (FIFO) is set to the given address, and the waiter +// woken up. If there are no waiters, 0 is set to the address. func (m *Manager) UnlockPI(t Target, addr usermem.Addr, tid uint32, private bool) error { k, err := getKey(t, addr, private) if err != nil { diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 240cd6fe0..15dae0f5b 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -1469,6 +1469,11 @@ func (k *Kernel) NowMonotonic() int64 { return now } +// AfterFunc implements tcpip.Clock.AfterFunc. +func (k *Kernel) AfterFunc(d time.Duration, f func()) tcpip.Timer { + return ktime.TcpipAfterFunc(k.realtimeClock, d, f) +} + // SetMemoryFile sets Kernel.mf. SetMemoryFile must be called before Init or // LoadFrom. func (k *Kernel) SetMemoryFile(mf *pgalloc.MemoryFile) { diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index f48247c94..c4db05bd8 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -68,6 +68,21 @@ type Task struct { // runState is exclusive to the task goroutine. runState taskRunState + // taskWorkCount represents the current size of the task work queue. It is + // used to avoid acquiring taskWorkMu when the queue is empty. + // + // Must accessed with atomic memory operations. + taskWorkCount int32 + + // taskWorkMu protects taskWork. + taskWorkMu sync.Mutex `state:"nosave"` + + // taskWork is a queue of work to be executed before resuming user execution. + // It is similar to the task_work mechanism in Linux. + // + // taskWork is exclusive to the task goroutine. + taskWork []TaskWorker + // haveSyscallReturn is true if tc.Arch().Return() represents a value // returned by a syscall (or set by ptrace after a syscall). // @@ -550,6 +565,10 @@ type Task struct { // futexWaiter is exclusive to the task goroutine. futexWaiter *futex.Waiter `state:"nosave"` + // robustList is a pointer to the head of the tasks's robust futex + // list. + robustList usermem.Addr + // startTime is the real time at which the task started. It is set when // a Task is created or invokes execve(2). // diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 9b69f3cbe..7803b98d0 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -207,6 +207,9 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { return flags.CloseOnExec }) + // Handle the robust futex list. + t.exitRobustList() + // NOTE(b/30815691): We currently do not implement privileged // executables (set-user/group-ID bits and file capabilities). This // allows us to unconditionally enable user dumpability on the new mm. diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index c4ade6e8e..231ac548a 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -253,6 +253,9 @@ func (*runExitMain) execute(t *Task) taskRunState { } } + // Handle the robust futex list. + t.exitRobustList() + // Deactivate the address space and update max RSS before releasing the // task's MM. t.Deactivate() diff --git a/pkg/sentry/kernel/task_futex.go b/pkg/sentry/kernel/task_futex.go index a53e77c9f..4b535c949 100644 --- a/pkg/sentry/kernel/task_futex.go +++ b/pkg/sentry/kernel/task_futex.go @@ -15,6 +15,7 @@ package kernel import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/usermem" ) @@ -52,3 +53,127 @@ func (t *Task) LoadUint32(addr usermem.Addr) (uint32, error) { func (t *Task) GetSharedKey(addr usermem.Addr) (futex.Key, error) { return t.MemoryManager().GetSharedFutexKey(t, addr) } + +// GetRobustList sets the robust futex list for the task. +func (t *Task) GetRobustList() usermem.Addr { + t.mu.Lock() + addr := t.robustList + t.mu.Unlock() + return addr +} + +// SetRobustList sets the robust futex list for the task. +func (t *Task) SetRobustList(addr usermem.Addr) { + t.mu.Lock() + t.robustList = addr + t.mu.Unlock() +} + +// exitRobustList walks the robust futex list, marking locks dead and notifying +// wakers. It corresponds to Linux's exit_robust_list(). Following Linux, +// errors are silently ignored. +func (t *Task) exitRobustList() { + t.mu.Lock() + addr := t.robustList + t.robustList = 0 + t.mu.Unlock() + + if addr == 0 { + return + } + + var rl linux.RobustListHead + if _, err := rl.CopyIn(t, usermem.Addr(addr)); err != nil { + return + } + + next := rl.List + done := 0 + var pendingLockAddr usermem.Addr + if rl.ListOpPending != 0 { + pendingLockAddr = usermem.Addr(rl.ListOpPending + rl.FutexOffset) + } + + // Wake up normal elements. + for usermem.Addr(next) != addr { + // We traverse to the next element of the list before we + // actually wake anything. This prevents the race where waking + // this futex causes a modification of the list. + thisLockAddr := usermem.Addr(next + rl.FutexOffset) + + // Try to decode the next element in the list before waking the + // current futex. But don't check the error until after we've + // woken the current futex. Linux does it in this order too + _, nextErr := t.CopyIn(usermem.Addr(next), &next) + + // Wakeup the current futex if it's not pending. + if thisLockAddr != pendingLockAddr { + t.wakeRobustListOne(thisLockAddr) + } + + // If there was an error copying the next futex, we must bail. + if nextErr != nil { + break + } + + // This is a user structure, so it could be a massive list, or + // even contain a loop if they are trying to mess with us. We + // cap traversal to prevent that. + done++ + if done >= linux.ROBUST_LIST_LIMIT { + break + } + } + + // Is there a pending entry to wake? + if pendingLockAddr != 0 { + t.wakeRobustListOne(pendingLockAddr) + } +} + +// wakeRobustListOne wakes a single futex from the robust list. +func (t *Task) wakeRobustListOne(addr usermem.Addr) { + // Bit 0 in address signals PI futex. + pi := addr&1 == 1 + addr = addr &^ 1 + + // Load the futex. + f, err := t.LoadUint32(addr) + if err != nil { + // Can't read this single value? Ignore the problem. + // We can wake the other futexes in the list. + return + } + + tid := uint32(t.ThreadID()) + for { + // Is this held by someone else? + if f&linux.FUTEX_TID_MASK != tid { + return + } + + // This thread is dying and it's holding this futex. We need to + // set the owner died bit and wake up any waiters. + newF := (f & linux.FUTEX_WAITERS) | linux.FUTEX_OWNER_DIED + if curF, err := t.CompareAndSwapUint32(addr, f, newF); err != nil { + return + } else if curF != f { + // Futex changed out from under us. Try again... + f = curF + continue + } + + // Wake waiters if there are any. + if f&linux.FUTEX_WAITERS != 0 { + private := f&linux.FUTEX_PRIVATE_FLAG != 0 + if pi { + t.Futex().UnlockPI(t, addr, tid, private) + return + } + t.Futex().Wake(t, addr, private, linux.FUTEX_BITSET_MATCH_ANY, 1) + } + + // Done. + return + } +} diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index d654dd997..7d4f44caf 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -167,7 +167,22 @@ func (app *runApp) execute(t *Task) taskRunState { return (*runInterrupt)(nil) } - // We're about to switch to the application again. If there's still a + // Execute any task work callbacks before returning to user space. + if atomic.LoadInt32(&t.taskWorkCount) > 0 { + t.taskWorkMu.Lock() + queue := t.taskWork + t.taskWork = nil + atomic.StoreInt32(&t.taskWorkCount, 0) + t.taskWorkMu.Unlock() + + // Do not hold taskWorkMu while executing task work, which may register + // more work. + for _, work := range queue { + work.TaskWork(t) + } + } + + // We're about to switch to the application again. If there's still an // unhandled SyscallRestartErrno that wasn't translated to an EINTR, // restart the syscall that was interrupted. If there's a saved signal // mask, restore it. (Note that restoring the saved signal mask may unblock diff --git a/pkg/sentry/kernel/task_work.go b/pkg/sentry/kernel/task_work.go new file mode 100644 index 000000000..dda5a433a --- /dev/null +++ b/pkg/sentry/kernel/task_work.go @@ -0,0 +1,38 @@ +// 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 kernel + +import "sync/atomic" + +// TaskWorker is a deferred task. +// +// This must be savable. +type TaskWorker interface { + // TaskWork will be executed prior to returning to user space. Note that + // TaskWork may call RegisterWork again, but this will not be executed until + // the next return to user space, unlike in Linux. This effectively allows + // registration of indefinite user return hooks, but not by default. + TaskWork(t *Task) +} + +// RegisterWork can be used to register additional task work that will be +// performed prior to returning to user space. See TaskWorker.TaskWork for +// semantics regarding registration. +func (t *Task) RegisterWork(work TaskWorker) { + t.taskWorkMu.Lock() + defer t.taskWorkMu.Unlock() + atomic.AddInt32(&t.taskWorkCount, 1) + t.taskWork = append(t.taskWork, work) +} diff --git a/pkg/sentry/kernel/time/BUILD b/pkg/sentry/kernel/time/BUILD index 7ba7dc50c..2817aa3ba 100644 --- a/pkg/sentry/kernel/time/BUILD +++ b/pkg/sentry/kernel/time/BUILD @@ -6,6 +6,7 @@ go_library( name = "time", srcs = [ "context.go", + "tcpip.go", "time.go", ], visibility = ["//pkg/sentry:internal"], diff --git a/pkg/sentry/kernel/time/tcpip.go b/pkg/sentry/kernel/time/tcpip.go new file mode 100644 index 000000000..c4474c0cf --- /dev/null +++ b/pkg/sentry/kernel/time/tcpip.go @@ -0,0 +1,131 @@ +// 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 time + +import ( + "sync" + "time" +) + +// TcpipAfterFunc waits for duration to elapse according to clock then runs fn. +// The timer is started immediately and will fire exactly once. +func TcpipAfterFunc(clock Clock, duration time.Duration, fn func()) *TcpipTimer { + timer := &TcpipTimer{ + clock: clock, + } + timer.notifier = functionNotifier{ + fn: func() { + // tcpip.Timer.Stop() explicitly states that the function is called in a + // separate goroutine that Stop() does not synchronize with. + // Timer.Destroy() synchronizes with calls to TimerListener.Notify(). + // This is semantically meaningful because, in the former case, it's + // legal to call tcpip.Timer.Stop() while holding locks that may also be + // taken by the function, but this isn't so in the latter case. Most + // immediately, Timer calls TimerListener.Notify() while holding + // Timer.mu. A deadlock occurs without spawning a goroutine: + // T1: (Timer expires) + // => Timer.Tick() <- Timer.mu.Lock() called + // => TimerListener.Notify() + // => Timer.Stop() + // => Timer.Destroy() <- Timer.mu.Lock() called, deadlock! + // + // Spawning a goroutine avoids the deadlock: + // T1: (Timer expires) + // => Timer.Tick() <- Timer.mu.Lock() called + // => TimerListener.Notify() <- Launches T2 + // T2: + // => Timer.Stop() + // => Timer.Destroy() <- Timer.mu.Lock() called, blocks + // T1: + // => (returns) <- Timer.mu.Unlock() called + // T2: + // => (continues) <- No deadlock! + go func() { + timer.Stop() + fn() + }() + }, + } + timer.Reset(duration) + return timer +} + +// TcpipTimer is a resettable timer with variable duration expirations. +// Implements tcpip.Timer, which does not define a Destroy method; instead, all +// resources are released after timer expiration and calls to Timer.Stop. +// +// Must be created by AfterFunc. +type TcpipTimer struct { + // clock is the time source. clock is immutable. + clock Clock + + // notifier is called when the Timer expires. notifier is immutable. + notifier functionNotifier + + // mu protects t. + mu sync.Mutex + + // t stores the latest running Timer. This is replaced whenever Reset is + // called since Timer cannot be restarted once it has been Destroyed by Stop. + // + // This field is nil iff Stop has been called. + t *Timer +} + +// Stop implements tcpip.Timer.Stop. +func (r *TcpipTimer) Stop() bool { + r.mu.Lock() + defer r.mu.Unlock() + + if r.t == nil { + return false + } + _, lastSetting := r.t.Swap(Setting{}) + r.t.Destroy() + r.t = nil + return lastSetting.Enabled +} + +// Reset implements tcpip.Timer.Reset. +func (r *TcpipTimer) Reset(d time.Duration) { + r.mu.Lock() + defer r.mu.Unlock() + + if r.t == nil { + r.t = NewTimer(r.clock, &r.notifier) + } + + r.t.Swap(Setting{ + Enabled: true, + Period: 0, + Next: r.clock.Now().Add(d), + }) +} + +// functionNotifier is a TimerListener that runs a function. +// +// functionNotifier cannot be saved or loaded. +type functionNotifier struct { + fn func() +} + +// Notify implements ktime.TimerListener.Notify. +func (f *functionNotifier) Notify(uint64, Setting) (Setting, bool) { + f.fn() + return Setting{}, false +} + +// Destroy implements ktime.TimerListener.Destroy. +func (f *functionNotifier) Destroy() {} diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index ccacaea6b..fca3a5478 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -58,7 +58,13 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { regs.Pstate &= ^uint64(UserFlagsClear) regs.Pstate |= UserFlagsSet + + SetTLS(regs.TPIDR_EL0) + kernelExitToEl0() + + regs.TPIDR_EL0 = GetTLS() + vector = c.vecCode // Perform the switch. diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD index c40c6d673..c0fd3425b 100644 --- a/pkg/sentry/socket/BUILD +++ b/pkg/sentry/socket/BUILD @@ -20,5 +20,6 @@ go_library( "//pkg/syserr", "//pkg/tcpip", "//pkg/usermem", + "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index ff81ea6e6..e76e498de 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -40,6 +40,8 @@ go_library( "//pkg/tcpip/stack", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index a92aed2c9..532a1ea5d 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -36,6 +36,8 @@ import ( "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) const ( @@ -319,7 +321,7 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { } // GetSockOpt implements socket.Socket.GetSockOpt. -func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { if outLen < 0 { return nil, syserr.ErrInvalidArgument } @@ -364,7 +366,8 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr if err != nil { return nil, syserr.FromError(err) } - return opt, nil + optP := primitive.ByteSlice(opt) + return &optP, nil } // SetSockOpt implements socket.Socket.SetSockOpt. @@ -708,6 +711,6 @@ func (p *socketProvider) Pair(t *kernel.Task, stype linux.SockType, protocol int func init() { for _, family := range []int{syscall.AF_INET, syscall.AF_INET6} { socket.RegisterProvider(family, &socketProvider{family}) - socket.RegisterProviderVFS2(family, &socketProviderVFS2{}) + socket.RegisterProviderVFS2(family, &socketProviderVFS2{family}) } } diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 1243143ea..a9f0604ae 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -66,7 +66,7 @@ func nflog(format string, args ...interface{}) { func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) { // Read in the struct and table name. var info linux.IPTGetinfo - if _, err := t.CopyIn(outPtr, &info); err != nil { + if _, err := info.CopyIn(t, outPtr); err != nil { return linux.IPTGetinfo{}, syserr.FromError(err) } @@ -84,7 +84,7 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen int) (linux.KernelIPTGetEntries, *syserr.Error) { // Read in the struct and table name. var userEntries linux.IPTGetEntries - if _, err := t.CopyIn(outPtr, &userEntries); err != nil { + if _, err := userEntries.CopyIn(t, outPtr); err != nil { nflog("couldn't copy in entries %q", userEntries.Name) return linux.KernelIPTGetEntries{}, syserr.FromError(err) } @@ -145,7 +145,7 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin // Each rule corresponds to an entry. entry := linux.KernelIPTEntry{ - IPTEntry: linux.IPTEntry{ + Entry: linux.IPTEntry{ IP: linux.IPTIP{ Protocol: uint16(rule.Filter.Protocol), }, @@ -153,20 +153,20 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin TargetOffset: linux.SizeOfIPTEntry, }, } - copy(entry.IPTEntry.IP.Dst[:], rule.Filter.Dst) - copy(entry.IPTEntry.IP.DstMask[:], rule.Filter.DstMask) - copy(entry.IPTEntry.IP.Src[:], rule.Filter.Src) - copy(entry.IPTEntry.IP.SrcMask[:], rule.Filter.SrcMask) - copy(entry.IPTEntry.IP.OutputInterface[:], rule.Filter.OutputInterface) - copy(entry.IPTEntry.IP.OutputInterfaceMask[:], rule.Filter.OutputInterfaceMask) + copy(entry.Entry.IP.Dst[:], rule.Filter.Dst) + copy(entry.Entry.IP.DstMask[:], rule.Filter.DstMask) + copy(entry.Entry.IP.Src[:], rule.Filter.Src) + copy(entry.Entry.IP.SrcMask[:], rule.Filter.SrcMask) + copy(entry.Entry.IP.OutputInterface[:], rule.Filter.OutputInterface) + copy(entry.Entry.IP.OutputInterfaceMask[:], rule.Filter.OutputInterfaceMask) if rule.Filter.DstInvert { - entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_DSTIP + entry.Entry.IP.InverseFlags |= linux.IPT_INV_DSTIP } if rule.Filter.SrcInvert { - entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_SRCIP + entry.Entry.IP.InverseFlags |= linux.IPT_INV_SRCIP } if rule.Filter.OutputInterfaceInvert { - entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_VIA_OUT + entry.Entry.IP.InverseFlags |= linux.IPT_INV_VIA_OUT } for _, matcher := range rule.Matchers { @@ -178,8 +178,8 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin panic(fmt.Sprintf("matcher %T is not 64-bit aligned", matcher)) } entry.Elems = append(entry.Elems, serialized...) - entry.NextOffset += uint16(len(serialized)) - entry.TargetOffset += uint16(len(serialized)) + entry.Entry.NextOffset += uint16(len(serialized)) + entry.Entry.TargetOffset += uint16(len(serialized)) } // Serialize and append the target. @@ -188,11 +188,11 @@ func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (lin panic(fmt.Sprintf("target %T is not 64-bit aligned", rule.Target)) } entry.Elems = append(entry.Elems, serialized...) - entry.NextOffset += uint16(len(serialized)) + entry.Entry.NextOffset += uint16(len(serialized)) nflog("convert to binary: adding entry: %+v", entry) - entries.Size += uint32(entry.NextOffset) + entries.Size += uint32(entry.Entry.NextOffset) entries.Entrytable = append(entries.Entrytable, entry) info.NumEntries++ } diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index d5ca3ac56..0546801bf 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -36,6 +36,8 @@ go_library( "//pkg/tcpip", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 81f34c5a2..98ca7add0 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -38,6 +38,8 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) const sizeOfInt32 int = 4 @@ -330,7 +332,7 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { } // GetSockOpt implements socket.Socket.GetSockOpt. -func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { switch level { case linux.SOL_SOCKET: switch name { @@ -340,24 +342,26 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr } s.mu.Lock() defer s.mu.Unlock() - return int32(s.sendBufferSize), nil + sendBufferSizeP := primitive.Int32(s.sendBufferSize) + return &sendBufferSizeP, nil case linux.SO_RCVBUF: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } // We don't have limit on receiving size. - return int32(math.MaxInt32), nil + recvBufferSizeP := primitive.Int32(math.MaxInt32) + return &recvBufferSizeP, nil case linux.SO_PASSCRED: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var passcred int32 + var passcred primitive.Int32 if s.Passcred() { passcred = 1 } - return passcred, nil + return &passcred, nil default: socket.GetSockOptEmitUnimplementedEvent(t, name) diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index ea6ebd0e2..1fb777a6c 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -51,6 +51,8 @@ go_library( "//pkg/tcpip/transport/udp", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 964ec8414..44b3fff46 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -62,6 +62,8 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) func mustCreateMetric(name, description string) *tcpip.StatCounter { @@ -910,7 +912,7 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error { // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { // TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is // implemented specifically for netstack.SocketOperations rather than // commonEndpoint. commonEndpoint should be extended to support socket @@ -920,25 +922,25 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - val := int32(0) + val := primitive.Int32(0) s.readMu.Lock() defer s.readMu.Unlock() if s.sockOptTimestamp { val = 1 } - return val, nil + return &val, nil } if level == linux.SOL_TCP && name == linux.TCP_INQ { if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - val := int32(0) + val := primitive.Int32(0) s.readMu.Lock() defer s.readMu.Unlock() if s.sockOptInq { val = 1 } - return val, nil + return &val, nil } if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { @@ -956,7 +958,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us if err != nil { return nil, err } - return info, nil + return &info, nil case linux.IPT_SO_GET_ENTRIES: if outLen < linux.SizeOfIPTGetEntries { @@ -971,7 +973,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us if err != nil { return nil, err } - return entries, nil + return &entries, nil } } @@ -981,7 +983,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us // GetSockOpt can be used to implement the linux syscall getsockopt(2) for // sockets backed by a commonEndpoint. -func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) { +func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (marshal.Marshallable, *syserr.Error) { switch level { case linux.SOL_SOCKET: return getSockOptSocket(t, s, ep, family, skType, name, outLen) @@ -1014,7 +1016,7 @@ func boolToInt32(v bool) int32 { } // getSockOptSocket implements GetSockOpt when level is SOL_SOCKET. -func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) { +func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (marshal.Marshallable, *syserr.Error) { // TODO(b/124056281): Stop rejecting short optLen values in getsockopt. switch name { case linux.SO_ERROR: @@ -1025,9 +1027,12 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam // Get the last error and convert it. err := ep.GetSockOpt(tcpip.ErrorOption{}) if err == nil { - return int32(0), nil + optP := primitive.Int32(0) + return &optP, nil } - return int32(syserr.TranslateNetstackError(err).ToLinux().Number()), nil + + optP := primitive.Int32(syserr.TranslateNetstackError(err).ToLinux().Number()) + return &optP, nil case linux.SO_PEERCRED: if family != linux.AF_UNIX || outLen < syscall.SizeofUcred { @@ -1035,11 +1040,12 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam } tcred := t.Credentials() - return syscall.Ucred{ - Pid: int32(t.ThreadGroup().ID()), - Uid: uint32(tcred.EffectiveKUID.In(tcred.UserNamespace).OrOverflow()), - Gid: uint32(tcred.EffectiveKGID.In(tcred.UserNamespace).OrOverflow()), - }, nil + creds := linux.ControlMessageCredentials{ + PID: int32(t.ThreadGroup().ID()), + UID: uint32(tcred.EffectiveKUID.In(tcred.UserNamespace).OrOverflow()), + GID: uint32(tcred.EffectiveKGID.In(tcred.UserNamespace).OrOverflow()), + } + return &creds, nil case linux.SO_PASSCRED: if outLen < sizeOfInt32 { @@ -1050,7 +1056,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_SNDBUF: if outLen < sizeOfInt32 { @@ -1066,7 +1074,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam size = math.MaxInt32 } - return int32(size), nil + sizeP := primitive.Int32(size) + return &sizeP, nil case linux.SO_RCVBUF: if outLen < sizeOfInt32 { @@ -1082,7 +1091,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam size = math.MaxInt32 } - return int32(size), nil + sizeP := primitive.Int32(size) + return &sizeP, nil case linux.SO_REUSEADDR: if outLen < sizeOfInt32 { @@ -1093,7 +1103,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_REUSEPORT: if outLen < sizeOfInt32 { @@ -1104,7 +1115,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_BINDTODEVICE: var v tcpip.BindToDeviceOption @@ -1112,7 +1125,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.TranslateNetstackError(err) } if v == 0 { - return []byte{}, nil + var b primitive.ByteSlice + return &b, nil } if outLen < linux.IFNAMSIZ { return nil, syserr.ErrInvalidArgument @@ -1127,7 +1141,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam // interface was removed. return nil, syserr.ErrUnknownDevice } - return append([]byte(nic.Name), 0), nil + + name := primitive.ByteSlice(append([]byte(nic.Name), 0)) + return &name, nil case linux.SO_BROADCAST: if outLen < sizeOfInt32 { @@ -1138,7 +1154,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_KEEPALIVE: if outLen < sizeOfInt32 { @@ -1149,13 +1167,17 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.SO_LINGER: if outLen < linux.SizeOfLinger { return nil, syserr.ErrInvalidArgument } - return linux.Linger{}, nil + + linger := linux.Linger{} + return &linger, nil case linux.SO_SNDTIMEO: // TODO(igudger): Linux allows shorter lengths for partial results. @@ -1163,7 +1185,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - return linux.NsecToTimeval(s.SendTimeout()), nil + sendTimeout := linux.NsecToTimeval(s.SendTimeout()) + return &sendTimeout, nil case linux.SO_RCVTIMEO: // TODO(igudger): Linux allows shorter lengths for partial results. @@ -1171,7 +1194,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - return linux.NsecToTimeval(s.RecvTimeout()), nil + recvTimeout := linux.NsecToTimeval(s.RecvTimeout()) + return &recvTimeout, nil case linux.SO_OOBINLINE: if outLen < sizeOfInt32 { @@ -1183,7 +1207,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.TranslateNetstackError(err) } - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.SO_NO_CHECK: if outLen < sizeOfInt32 { @@ -1194,7 +1219,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil default: socket.GetSockOptEmitUnimplementedEvent(t, name) @@ -1203,7 +1229,7 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam } // getSockOptTCP implements GetSockOpt when level is SOL_TCP. -func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interface{}, *syserr.Error) { +func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (marshal.Marshallable, *syserr.Error) { switch name { case linux.TCP_NODELAY: if outLen < sizeOfInt32 { @@ -1214,7 +1240,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(!v), nil + + vP := primitive.Int32(boolToInt32(!v)) + return &vP, nil case linux.TCP_CORK: if outLen < sizeOfInt32 { @@ -1225,7 +1253,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.TCP_QUICKACK: if outLen < sizeOfInt32 { @@ -1236,7 +1266,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.TCP_MAXSEG: if outLen < sizeOfInt32 { @@ -1247,8 +1279,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.TCP_KEEPIDLE: if outLen < sizeOfInt32 { @@ -1259,8 +1291,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err := ep.GetSockOpt(&v); err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(time.Duration(v) / time.Second), nil + keepAliveIdle := primitive.Int32(time.Duration(v) / time.Second) + return &keepAliveIdle, nil case linux.TCP_KEEPINTVL: if outLen < sizeOfInt32 { @@ -1271,8 +1303,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err := ep.GetSockOpt(&v); err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(time.Duration(v) / time.Second), nil + keepAliveInterval := primitive.Int32(time.Duration(v) / time.Second) + return &keepAliveInterval, nil case linux.TCP_KEEPCNT: if outLen < sizeOfInt32 { @@ -1283,8 +1315,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.TCP_USER_TIMEOUT: if outLen < sizeOfInt32 { @@ -1295,8 +1327,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err := ep.GetSockOpt(&v); err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(time.Duration(v) / time.Millisecond), nil + tcpUserTimeout := primitive.Int32(time.Duration(v) / time.Millisecond) + return &tcpUserTimeout, nil case linux.TCP_INFO: var v tcpip.TCPInfoOption @@ -1309,12 +1341,13 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa info := linux.TCPInfo{} // Linux truncates the output binary to outLen. - ib := binary.Marshal(nil, usermem.ByteOrder, &info) - if len(ib) > outLen { - ib = ib[:outLen] + buf := t.CopyScratchBuffer(info.SizeBytes()) + info.MarshalUnsafe(buf) + if len(buf) > outLen { + buf = buf[:outLen] } - - return ib, nil + bufP := primitive.ByteSlice(buf) + return &bufP, nil case linux.TCP_CC_INFO, linux.TCP_NOTSENT_LOWAT, @@ -1344,7 +1377,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa } b := make([]byte, toCopy) copy(b, v) - return b, nil + + bP := primitive.ByteSlice(b) + return &bP, nil case linux.TCP_LINGER2: if outLen < sizeOfInt32 { @@ -1356,7 +1391,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return nil, syserr.TranslateNetstackError(err) } - return int32(time.Duration(v) / time.Second), nil + lingerTimeout := primitive.Int32(time.Duration(v) / time.Second) + return &lingerTimeout, nil case linux.TCP_DEFER_ACCEPT: if outLen < sizeOfInt32 { @@ -1368,7 +1404,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return nil, syserr.TranslateNetstackError(err) } - return int32(time.Duration(v) / time.Second), nil + tcpDeferAccept := primitive.Int32(time.Duration(v) / time.Second) + return &tcpDeferAccept, nil case linux.TCP_SYNCNT: if outLen < sizeOfInt32 { @@ -1379,8 +1416,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.TCP_WINDOW_CLAMP: if outLen < sizeOfInt32 { @@ -1391,8 +1428,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil default: emitUnimplementedEventTCP(t, name) } @@ -1400,7 +1437,7 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa } // getSockOptIPv6 implements GetSockOpt when level is SOL_IPV6. -func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interface{}, *syserr.Error) { +func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (marshal.Marshallable, *syserr.Error) { switch name { case linux.IPV6_V6ONLY: if outLen < sizeOfInt32 { @@ -1411,7 +1448,9 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1419,21 +1458,24 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf case linux.IPV6_TCLASS: // Length handling for parity with Linux. if outLen == 0 { - return make([]byte, 0), nil + var b primitive.ByteSlice + return &b, nil } v, err := ep.GetSockOptInt(tcpip.IPv6TrafficClassOption) if err != nil { return nil, syserr.TranslateNetstackError(err) } - uintv := uint32(v) + uintv := primitive.Uint32(v) // Linux truncates the output binary to outLen. - ib := binary.Marshal(nil, usermem.ByteOrder, &uintv) + ib := t.CopyScratchBuffer(uintv.SizeBytes()) + uintv.MarshalUnsafe(ib) // Handle cases where outLen is lesser than sizeOfInt32. if len(ib) > outLen { ib = ib[:outLen] } - return ib, nil + ibP := primitive.ByteSlice(ib) + return &ibP, nil case linux.IPV6_RECVTCLASS: if outLen < sizeOfInt32 { @@ -1444,7 +1486,9 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil default: emitUnimplementedEventIPv6(t, name) @@ -1453,7 +1497,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf } // getSockOptIP implements GetSockOpt when level is SOL_IP. -func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family int) (interface{}, *syserr.Error) { +func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family int) (marshal.Marshallable, *syserr.Error) { switch name { case linux.IP_TTL: if outLen < sizeOfInt32 { @@ -1466,11 +1510,12 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } // Fill in the default value, if needed. - if v == 0 { - v = DefaultTTL + vP := primitive.Int32(v) + if vP == 0 { + vP = DefaultTTL } - return int32(v), nil + return &vP, nil case linux.IP_MULTICAST_TTL: if outLen < sizeOfInt32 { @@ -1482,7 +1527,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in return nil, syserr.TranslateNetstackError(err) } - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.IP_MULTICAST_IF: if outLen < len(linux.InetAddr{}) { @@ -1496,7 +1542,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in a, _ := ConvertAddress(linux.AF_INET, tcpip.FullAddress{Addr: v.InterfaceAddr}) - return a.(*linux.SockAddrInet).Addr, nil + return &a.(*linux.SockAddrInet).Addr, nil case linux.IP_MULTICAST_LOOP: if outLen < sizeOfInt32 { @@ -1507,21 +1553,26 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.IP_TOS: // Length handling for parity with Linux. if outLen == 0 { - return []byte(nil), nil + var b primitive.ByteSlice + return &b, nil } v, err := ep.GetSockOptInt(tcpip.IPv4TOSOption) if err != nil { return nil, syserr.TranslateNetstackError(err) } if outLen < sizeOfInt32 { - return uint8(v), nil + vP := primitive.Uint8(v) + return &vP, nil } - return int32(v), nil + vP := primitive.Int32(v) + return &vP, nil case linux.IP_RECVTOS: if outLen < sizeOfInt32 { @@ -1532,7 +1583,9 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil case linux.IP_PKTINFO: if outLen < sizeOfInt32 { @@ -1543,7 +1596,9 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - return boolToInt32(v), nil + + vP := primitive.Int32(boolToInt32(v)) + return &vP, nil default: emitUnimplementedEventIP(t, name) @@ -2780,6 +2835,11 @@ func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, } func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + panic("ioctl(2) may only be called from a task goroutine") + } + // SIOCGSTAMP is implemented by netstack rather than all commonEndpoint // sockets. // TODO(b/78348848): Add a commonEndpoint method to support SIOCGSTAMP. @@ -2792,9 +2852,7 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy } tv := linux.NsecToTimeval(s.timestampNS) - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &tv, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := tv.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCINQ: @@ -2813,9 +2871,8 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy } // Copy result to userspace. - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ - AddressSpaceActive: true, - }) + vP := primitive.Int32(v) + _, err := vP.CopyOut(t, args[2].Pointer()) return 0, err } @@ -2824,6 +2881,11 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy // Ioctl performs a socket ioctl. func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + panic("ioctl(2) may only be called from a task goroutine") + } + switch arg := int(args[1].Int()); arg { case linux.SIOCGIFFLAGS, linux.SIOCGIFADDR, @@ -2840,37 +2902,28 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc linux.SIOCETHTOOL: var ifr linux.IFReq - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifr, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { + if _, err := ifr.CopyIn(t, args[2].Pointer()); err != nil { return 0, err } if err := interfaceIoctl(ctx, io, arg, &ifr); err != nil { return 0, err.ToError() } - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &ifr, usermem.IOOpts{ - AddressSpaceActive: true, - }) + _, err := ifr.CopyOut(t, args[2].Pointer()) return 0, err case linux.SIOCGIFCONF: // Return a list of interface addresses or the buffer size // necessary to hold the list. var ifc linux.IFConf - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifc, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { + if _, err := ifc.CopyIn(t, args[2].Pointer()); err != nil { return 0, err } - if err := ifconfIoctl(ctx, io, &ifc); err != nil { + if err := ifconfIoctl(ctx, t, io, &ifc); err != nil { return 0, err } - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), ifc, usermem.IOOpts{ - AddressSpaceActive: true, - }) - + _, err := ifc.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCINQ: @@ -2883,9 +2936,8 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc v = math.MaxInt32 } // Copy result to userspace. - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ - AddressSpaceActive: true, - }) + vP := primitive.Int32(v) + _, err := vP.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCOUTQ: @@ -2899,9 +2951,8 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc } // Copy result to userspace. - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ - AddressSpaceActive: true, - }) + vP := primitive.Int32(v) + _, err := vP.CopyOut(t, args[2].Pointer()) return 0, err case linux.SIOCGIFMEM, linux.SIOCGIFPFLAGS, linux.SIOCGMIIPHY, linux.SIOCGMIIREG: @@ -3050,7 +3101,7 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe } // ifconfIoctl populates a struct ifconf for the SIOCGIFCONF ioctl. -func ifconfIoctl(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error { +func ifconfIoctl(ctx context.Context, t *kernel.Task, io usermem.IO, ifc *linux.IFConf) error { // If Ptr is NULL, return the necessary buffer size via Len. // Otherwise, write up to Len bytes starting at Ptr containing ifreq // structs. @@ -3087,9 +3138,7 @@ func ifconfIoctl(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error { // Copy the ifr to userspace. dst := uintptr(ifc.Ptr) + uintptr(ifc.Len) ifc.Len += int32(linux.SizeOfIFReq) - if _, err := usermem.CopyObjectOut(ctx, io, usermem.Addr(dst), ifr, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { + if _, err := ifr.CopyOut(t, usermem.Addr(dst)); err != nil { return err } } diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index d65a89316..a9025b0ec 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -31,6 +31,8 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // SocketVFS2 encapsulates all the state needed to represent a network stack @@ -200,7 +202,7 @@ func (s *SocketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.Syscal // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { // TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is // implemented specifically for netstack.SocketVFS2 rather than // commonEndpoint. commonEndpoint should be extended to support socket @@ -210,25 +212,25 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem. if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - val := int32(0) + val := primitive.Int32(0) s.readMu.Lock() defer s.readMu.Unlock() if s.sockOptTimestamp { val = 1 } - return val, nil + return &val, nil } if level == linux.SOL_TCP && name == linux.TCP_INQ { if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - val := int32(0) + val := primitive.Int32(0) s.readMu.Lock() defer s.readMu.Unlock() if s.sockOptInq { val = 1 } - return val, nil + return &val, nil } if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { @@ -246,7 +248,7 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem. if err != nil { return nil, err } - return info, nil + return &info, nil case linux.IPT_SO_GET_ENTRIES: if outLen < linux.SizeOfIPTGetEntries { @@ -261,7 +263,7 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem. if err != nil { return nil, err } - return entries, nil + return &entries, nil } } diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index fcd7f9d7f..d112757fb 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -35,6 +35,7 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // ControlMessages represents the union of unix control messages and tcpip @@ -86,7 +87,7 @@ type SocketOps interface { Shutdown(t *kernel.Task, how int) *syserr.Error // GetSockOpt implements the getsockopt(2) linux syscall. - GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) + GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) // SetSockOpt implements the setsockopt(2) linux syscall. SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index cca5e70f1..061a689a9 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -35,5 +35,6 @@ go_library( "//pkg/tcpip", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 4bb2b6ff4..0482d33cf 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -40,6 +40,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // SocketOperations is a Unix socket. It is similar to a netstack socket, @@ -184,7 +185,7 @@ func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // a transport.Endpoint. -func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen) } diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index ff2149250..05c16fcfe 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -32,6 +32,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" + "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // SocketVFS2 implements socket.SocketVFS2 (and by extension, @@ -89,7 +90,7 @@ func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint3 // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // a transport.Endpoint. -func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { +func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen) } diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 217fcfef2..4a9b04fd0 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -99,5 +99,7 @@ go_library( "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index ea4f9b1a7..80c65164a 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -325,8 +325,8 @@ var AMD64 = &kernel.SyscallTable{ 270: syscalls.Supported("pselect", Pselect), 271: syscalls.Supported("ppoll", Ppoll), 272: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), - 273: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 274: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 273: syscalls.Supported("set_robust_list", SetRobustList), + 274: syscalls.Supported("get_robust_list", GetRobustList), 275: syscalls.Supported("splice", Splice), 276: syscalls.Supported("tee", Tee), 277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go index b68261f72..f04d78856 100644 --- a/pkg/sentry/syscalls/linux/sys_futex.go +++ b/pkg/sentry/syscalls/linux/sys_futex.go @@ -198,7 +198,7 @@ func Futex(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall switch cmd { case linux.FUTEX_WAIT: // WAIT uses a relative timeout. - mask = ^uint32(0) + mask = linux.FUTEX_BITSET_MATCH_ANY var timeoutDur time.Duration if !forever { timeoutDur = time.Duration(timespec.ToNsecCapped()) * time.Nanosecond @@ -286,3 +286,49 @@ func Futex(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, syserror.ENOSYS } } + +// SetRobustList implements linux syscall set_robust_list(2). +func SetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // Despite the syscall using the name 'pid' for this variable, it is + // very much a tid. + head := args[0].Pointer() + length := args[1].SizeT() + + if length != uint(linux.SizeOfRobustListHead) { + return 0, nil, syserror.EINVAL + } + t.SetRobustList(head) + return 0, nil, nil +} + +// GetRobustList implements linux syscall get_robust_list(2). +func GetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + // Despite the syscall using the name 'pid' for this variable, it is + // very much a tid. + tid := args[0].Int() + head := args[1].Pointer() + size := args[2].Pointer() + + if tid < 0 { + return 0, nil, syserror.EINVAL + } + + ot := t + if tid != 0 { + if ot = t.PIDNamespace().TaskWithID(kernel.ThreadID(tid)); ot == nil { + return 0, nil, syserror.ESRCH + } + } + + // Copy out head pointer. + if _, err := t.CopyOut(head, uint64(ot.GetRobustList())); err != nil { + return 0, nil, err + } + + // Copy out size, which is a constant. + if _, err := t.CopyOut(size, uint64(linux.SizeOfRobustListHead)); err != nil { + return 0, nil, err + } + + return 0, nil, nil +} diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 0760af77b..414fce8e3 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -29,6 +29,8 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // LINT.IfChange @@ -474,7 +476,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } if v != nil { - if _, err := t.CopyOut(optValAddr, v); err != nil { + if _, err := v.CopyOut(t, optValAddr); err != nil { return 0, nil, err } } @@ -484,7 +486,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // getSockOpt tries to handle common socket options, or dispatches to a specific // socket implementation. -func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr usermem.Addr, len int) (interface{}, *syserr.Error) { +func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr usermem.Addr, len int) (marshal.Marshallable, *syserr.Error) { if level == linux.SOL_SOCKET { switch name { case linux.SO_TYPE, linux.SO_DOMAIN, linux.SO_PROTOCOL: @@ -496,13 +498,16 @@ func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr use switch name { case linux.SO_TYPE: _, skType, _ := s.Type() - return int32(skType), nil + v := primitive.Int32(skType) + return &v, nil case linux.SO_DOMAIN: family, _, _ := s.Type() - return int32(family), nil + v := primitive.Int32(family) + return &v, nil case linux.SO_PROTOCOL: _, _, protocol := s.Type() - return int32(protocol), nil + v := primitive.Int32(protocol) + return &v, nil } } @@ -539,7 +544,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, syserror.EINVAL } buf := t.CopyScratchBuffer(int(optLen)) - if _, err := t.CopyIn(optValAddr, &buf); err != nil { + if _, err := t.CopyInBytes(optValAddr, buf); err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD index 0c740335b..64696b438 100644 --- a/pkg/sentry/syscalls/linux/vfs2/BUILD +++ b/pkg/sentry/syscalls/linux/vfs2/BUILD @@ -72,5 +72,7 @@ go_library( "//pkg/syserror", "//pkg/usermem", "//pkg/waiter", + "//tools/go_marshal/marshal", + "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go index adeaa39cc..ea337de7c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/mount.go +++ b/pkg/sentry/syscalls/linux/vfs2/mount.go @@ -77,8 +77,7 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Silently allow MS_NOSUID, since we don't implement set-id bits // anyway. - const unsupportedFlags = linux.MS_NODEV | - linux.MS_NODIRATIME | linux.MS_STRICTATIME + const unsupportedFlags = linux.MS_NODIRATIME | linux.MS_STRICTATIME // Linux just allows passing any flags to mount(2) - it won't fail when // unknown or unsupported flags are passed. Since we don't implement @@ -94,6 +93,12 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if flags&linux.MS_NOEXEC == linux.MS_NOEXEC { opts.Flags.NoExec = true } + if flags&linux.MS_NODEV == linux.MS_NODEV { + opts.Flags.NoDev = true + } + if flags&linux.MS_NOSUID == linux.MS_NOSUID { + opts.Flags.NoSUID = true + } if flags&linux.MS_RDONLY == linux.MS_RDONLY { opts.ReadOnly = true } diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 09ecfed26..6daedd173 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -178,6 +178,7 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc Mask: linux.STATX_SIZE, Size: uint64(length), }, + NeedWritePerm: true, }) return 0, nil, handleSetSizeError(t, err) } @@ -197,6 +198,10 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } defer file.DecRef() + if !file.IsWritable() { + return 0, nil, syserror.EINVAL + } + err := file.SetStat(t, vfs.SetStatOptions{ Stat: linux.Statx{ Mask: linux.STATX_SIZE, diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go index 10b668477..8096a8f9c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/socket.go +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -30,6 +30,8 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" + "gvisor.dev/gvisor/tools/go_marshal/marshal" + "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // minListenBacklog is the minimum reasonable backlog for listening sockets. @@ -477,7 +479,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } if v != nil { - if _, err := t.CopyOut(optValAddr, v); err != nil { + if _, err := v.CopyOut(t, optValAddr); err != nil { return 0, nil, err } } @@ -487,7 +489,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // getSockOpt tries to handle common socket options, or dispatches to a specific // socket implementation. -func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr usermem.Addr, len int) (interface{}, *syserr.Error) { +func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr usermem.Addr, len int) (marshal.Marshallable, *syserr.Error) { if level == linux.SOL_SOCKET { switch name { case linux.SO_TYPE, linux.SO_DOMAIN, linux.SO_PROTOCOL: @@ -499,13 +501,16 @@ func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr switch name { case linux.SO_TYPE: _, skType, _ := s.Type() - return int32(skType), nil + v := primitive.Int32(skType) + return &v, nil case linux.SO_DOMAIN: family, _, _ := s.Type() - return int32(family), nil + v := primitive.Int32(family) + return &v, nil case linux.SO_PROTOCOL: _, _, protocol := s.Type() - return int32(protocol), nil + v := primitive.Int32(protocol) + return &v, nil } } @@ -542,7 +547,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, syserror.EINVAL } buf := t.CopyScratchBuffer(int(optLen)) - if _, err := t.CopyIn(optValAddr, &buf); err != nil { + if _, err := t.CopyInBytes(optValAddr, buf); err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 945a364a7..63ab11f8c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -15,12 +15,15 @@ package vfs2 import ( + "io" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -110,16 +113,20 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Move data. var ( - n int64 - err error - inCh chan struct{} - outCh chan struct{} + n int64 + err error ) + dw := dualWaiter{ + inFile: inFile, + outFile: outFile, + } + defer dw.destroy() for { // If both input and output are pipes, delegate to the pipe - // implementation. Otherwise, exactly one end is a pipe, which we - // ensure is consistently ordered after the non-pipe FD's locks by - // passing the pipe FD as usermem.IO to the non-pipe end. + // implementation. Otherwise, exactly one end is a pipe, which + // we ensure is consistently ordered after the non-pipe FD's + // locks by passing the pipe FD as usermem.IO to the non-pipe + // end. switch { case inIsPipe && outIsPipe: n, err = pipe.Splice(t, outPipeFD, inPipeFD, count) @@ -137,38 +144,15 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } else { n, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{}) } + default: + panic("not possible") } + if n != 0 || err != syserror.ErrWouldBlock || nonBlock { break } - - // Note that the blocking behavior here is a bit different than the - // normal pattern. Because we need to have both data to read and data - // to write simultaneously, we actually explicitly block on both of - // these cases in turn before returning to the splice operation. - if inFile.Readiness(eventMaskRead)&eventMaskRead == 0 { - if inCh == nil { - inCh = make(chan struct{}, 1) - inW, _ := waiter.NewChannelEntry(inCh) - inFile.EventRegister(&inW, eventMaskRead) - defer inFile.EventUnregister(&inW) - continue // Need to refresh readiness. - } - if err = t.Block(inCh); err != nil { - break - } - } - if outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 { - if outCh == nil { - outCh = make(chan struct{}, 1) - outW, _ := waiter.NewChannelEntry(outCh) - outFile.EventRegister(&outW, eventMaskWrite) - defer outFile.EventUnregister(&outW) - continue // Need to refresh readiness. - } - if err = t.Block(outCh); err != nil { - break - } + if err = dw.waitForBoth(t); err != nil { + break } } @@ -247,45 +231,256 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo // Copy data. var ( - inCh chan struct{} - outCh chan struct{} + n int64 + err error ) + dw := dualWaiter{ + inFile: inFile, + outFile: outFile, + } + defer dw.destroy() for { - n, err := pipe.Tee(t, outPipeFD, inPipeFD, count) - if n != 0 { - return uintptr(n), nil, nil + n, err = pipe.Tee(t, outPipeFD, inPipeFD, count) + if n != 0 || err != syserror.ErrWouldBlock || nonBlock { + break + } + if err = dw.waitForBoth(t); err != nil { + break + } + } + if n == 0 { + return 0, nil, err + } + outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + return uintptr(n), nil, nil +} + +// Sendfile implements linux system call sendfile(2). +func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + outFD := args[0].Int() + inFD := args[1].Int() + offsetAddr := args[2].Pointer() + count := int64(args[3].SizeT()) + + inFile := t.GetFileVFS2(inFD) + if inFile == nil { + return 0, nil, syserror.EBADF + } + defer inFile.DecRef() + if !inFile.IsReadable() { + return 0, nil, syserror.EBADF + } + + outFile := t.GetFileVFS2(outFD) + if outFile == nil { + return 0, nil, syserror.EBADF + } + defer outFile.DecRef() + if !outFile.IsWritable() { + return 0, nil, syserror.EBADF + } + + // Verify that the outFile Append flag is not set. + if outFile.StatusFlags()&linux.O_APPEND != 0 { + return 0, nil, syserror.EINVAL + } + + // Verify that inFile is a regular file or block device. This is a + // requirement; the same check appears in Linux + // (fs/splice.c:splice_direct_to_actor). + if stat, err := inFile.Stat(t, vfs.StatOptions{Mask: linux.STATX_TYPE}); err != nil { + return 0, nil, err + } else if stat.Mask&linux.STATX_TYPE == 0 || + (stat.Mode&linux.S_IFMT != linux.S_IFREG && stat.Mode&linux.S_IFMT != linux.S_IFBLK) { + return 0, nil, syserror.EINVAL + } + + // Copy offset if it exists. + offset := int64(-1) + if offsetAddr != 0 { + if inFile.Options().DenyPRead { + return 0, nil, syserror.ESPIPE } - if err != syserror.ErrWouldBlock || nonBlock { + if _, err := t.CopyIn(offsetAddr, &offset); err != nil { return 0, nil, err } + if offset < 0 { + return 0, nil, syserror.EINVAL + } + if offset+count < 0 { + return 0, nil, syserror.EINVAL + } + } + + // Validate count. This must come after offset checks. + if count < 0 { + return 0, nil, syserror.EINVAL + } + if count == 0 { + return 0, nil, nil + } + if count > int64(kernel.MAX_RW_COUNT) { + count = int64(kernel.MAX_RW_COUNT) + } - // Note that the blocking behavior here is a bit different than the - // normal pattern. Because we need to have both data to read and data - // to write simultaneously, we actually explicitly block on both of - // these cases in turn before returning to the tee operation. - if inFile.Readiness(eventMaskRead)&eventMaskRead == 0 { - if inCh == nil { - inCh = make(chan struct{}, 1) - inW, _ := waiter.NewChannelEntry(inCh) - inFile.EventRegister(&inW, eventMaskRead) - defer inFile.EventUnregister(&inW) - continue // Need to refresh readiness. + // Copy data. + var ( + n int64 + err error + ) + dw := dualWaiter{ + inFile: inFile, + outFile: outFile, + } + defer dw.destroy() + outPipeFD, outIsPipe := outFile.Impl().(*pipe.VFSPipeFD) + // Reading from input file should never block, since it is regular or + // block device. We only need to check if writing to the output file + // can block. + nonBlock := outFile.StatusFlags()&linux.O_NONBLOCK != 0 + if outIsPipe { + for n < count { + var spliceN int64 + if offset != -1 { + spliceN, err = inFile.PRead(t, outPipeFD.IOSequence(count), offset, vfs.ReadOptions{}) + offset += spliceN + } else { + spliceN, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{}) } - if err := t.Block(inCh); err != nil { - return 0, nil, err + n += spliceN + if err == syserror.ErrWouldBlock && !nonBlock { + err = dw.waitForBoth(t) + } + if err != nil { + break } } - if outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 { - if outCh == nil { - outCh = make(chan struct{}, 1) - outW, _ := waiter.NewChannelEntry(outCh) - outFile.EventRegister(&outW, eventMaskWrite) - defer outFile.EventUnregister(&outW) - continue // Need to refresh readiness. + } else { + // Read inFile to buffer, then write the contents to outFile. + buf := make([]byte, count) + for n < count { + var readN int64 + if offset != -1 { + readN, err = inFile.PRead(t, usermem.BytesIOSequence(buf), offset, vfs.ReadOptions{}) + offset += readN + } else { + readN, err = inFile.Read(t, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) + } + if readN == 0 && err == io.EOF { + // We reached the end of the file. Eat the + // error and exit the loop. + err = nil + break } - if err := t.Block(outCh); err != nil { - return 0, nil, err + n += readN + if err != nil { + break + } + + // Write all of the bytes that we read. This may need + // multiple write calls to complete. + wbuf := buf[:n] + for len(wbuf) > 0 { + var writeN int64 + writeN, err = outFile.Write(t, usermem.BytesIOSequence(wbuf), vfs.WriteOptions{}) + wbuf = wbuf[writeN:] + if err == syserror.ErrWouldBlock && !nonBlock { + err = dw.waitForOut(t) + } + if err != nil { + // We didn't complete the write. Only + // report the bytes that were actually + // written, and rewind the offset. + notWritten := int64(len(wbuf)) + n -= notWritten + if offset != -1 { + offset -= notWritten + } + break + } + } + if err == syserror.ErrWouldBlock && !nonBlock { + err = dw.waitForBoth(t) } + if err != nil { + break + } + } + } + + if offsetAddr != 0 { + // Copy out the new offset. + if _, err := t.CopyOut(offsetAddr, offset); err != nil { + return 0, nil, err + } + } + + if n == 0 { + return 0, nil, err + } + + inFile.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + return uintptr(n), nil, nil +} + +// dualWaiter is used to wait on one or both vfs.FileDescriptions. It is not +// thread-safe, and does not take a reference on the vfs.FileDescriptions. +// +// Users must call destroy() when finished. +type dualWaiter struct { + inFile *vfs.FileDescription + outFile *vfs.FileDescription + + inW waiter.Entry + inCh chan struct{} + outW waiter.Entry + outCh chan struct{} +} + +// waitForBoth waits for both dw.inFile and dw.outFile to be ready. +func (dw *dualWaiter) waitForBoth(t *kernel.Task) error { + if dw.inFile.Readiness(eventMaskRead)&eventMaskRead == 0 { + if dw.inCh == nil { + dw.inW, dw.inCh = waiter.NewChannelEntry(nil) + dw.inFile.EventRegister(&dw.inW, eventMaskRead) + // We might be ready now. Try again before blocking. + return nil + } + if err := t.Block(dw.inCh); err != nil { + return err + } + } + return dw.waitForOut(t) +} + +// waitForOut waits for dw.outfile to be read. +func (dw *dualWaiter) waitForOut(t *kernel.Task) error { + if dw.outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 { + if dw.outCh == nil { + dw.outW, dw.outCh = waiter.NewChannelEntry(nil) + dw.outFile.EventRegister(&dw.outW, eventMaskWrite) + // We might be ready now. Try again before blocking. + return nil } + if err := t.Block(dw.outCh); err != nil { + return err + } + } + return nil +} + +// destroy cleans up resources help by dw. No more calls to wait* can occur +// after destroy is called. +func (dw *dualWaiter) destroy() { + if dw.inCh != nil { + dw.inFile.EventUnregister(&dw.inW) + dw.inCh = nil + } + if dw.outCh != nil { + dw.outFile.EventUnregister(&dw.outW) + dw.outCh = nil } + dw.inFile = nil + dw.outFile = nil } diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 8f497ecc7..1b2cfad7d 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -44,7 +44,7 @@ func Override() { s.Table[23] = syscalls.Supported("select", Select) s.Table[32] = syscalls.Supported("dup", Dup) s.Table[33] = syscalls.Supported("dup2", Dup2) - delete(s.Table, 40) // sendfile + s.Table[40] = syscalls.Supported("sendfile", Sendfile) s.Table[41] = syscalls.Supported("socket", Socket) s.Table[42] = syscalls.Supported("connect", Connect) s.Table[43] = syscalls.Supported("accept", Accept) diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index f223aeda8..dfc8573fd 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -79,6 +79,17 @@ type MountFlags struct { // NoATime is equivalent to MS_NOATIME and indicates that the // filesystem should not update access time in-place. NoATime bool + + // NoDev is equivalent to MS_NODEV and indicates that the + // filesystem should not allow access to devices (special files). + // TODO(gVisor.dev/issue/3186): respect this flag in non FUSE + // filesystems. + NoDev bool + + // NoSUID is equivalent to MS_NOSUID and indicates that the + // filesystem should not honor set-user-ID and set-group-ID bits or + // file capabilities when executing programs. + NoSUID bool } // MountOptions contains options to VirtualFilesystem.MountAt(). @@ -153,6 +164,12 @@ type SetStatOptions struct { // == UTIME_OMIT (VFS users must unset the corresponding bit in Stat.Mask // instead). Stat linux.Statx + + // NeedWritePerm indicates that write permission on the file is needed for + // this operation. This is needed for truncate(2) (note that ftruncate(2) + // does not require the same check--instead, it checks that the fd is + // writable). + NeedWritePerm bool } // BoundEndpointOptions contains options to VirtualFilesystem.BoundEndpointAt() diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index 9cb050597..33389c1df 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -183,7 +183,8 @@ func MayWriteFileWithOpenFlags(flags uint32) bool { // CheckSetStat checks that creds has permission to change the metadata of a // file with the given permissions, UID, and GID as specified by stat, subject // to the rules of Linux's fs/attr.c:setattr_prepare(). -func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) error { +func CheckSetStat(ctx context.Context, creds *auth.Credentials, opts *SetStatOptions, mode linux.FileMode, kuid auth.KUID, kgid auth.KGID) error { + stat := &opts.Stat if stat.Mask&linux.STATX_SIZE != 0 { limit, err := CheckLimit(ctx, 0, int64(stat.Size)) if err != nil { @@ -215,6 +216,11 @@ func CheckSetStat(ctx context.Context, creds *auth.Credentials, stat *linux.Stat return syserror.EPERM } } + if opts.NeedWritePerm && !creds.HasCapability(linux.CAP_DAC_OVERRIDE) { + if err := GenericCheckPermissions(creds, MayWrite, mode, kuid, kgid); err != nil { + return err + } + } if stat.Mask&(linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME) != 0 { if !CanActAsOwner(creds, kuid) { if (stat.Mask&linux.STATX_ATIME != 0 && stat.Atime.Nsec != linux.UTIME_NOW) || diff --git a/pkg/tcpip/header/icmpv4.go b/pkg/tcpip/header/icmpv4.go index 7908c5744..1a631b31a 100644 --- a/pkg/tcpip/header/icmpv4.go +++ b/pkg/tcpip/header/icmpv4.go @@ -72,6 +72,7 @@ const ( // Values for ICMP code as defined in RFC 792. const ( ICMPv4TTLExceeded = 0 + ICMPv4HostUnreachable = 1 ICMPv4PortUnreachable = 3 ICMPv4FragmentationNeeded = 4 ) diff --git a/pkg/tcpip/header/icmpv6.go b/pkg/tcpip/header/icmpv6.go index c7ee2de57..a13b4b809 100644 --- a/pkg/tcpip/header/icmpv6.go +++ b/pkg/tcpip/header/icmpv6.go @@ -110,9 +110,16 @@ const ( ICMPv6RedirectMsg ICMPv6Type = 137 ) -// Values for ICMP code as defined in RFC 4443. +// Values for ICMP destination unreachable code as defined in RFC 4443 section +// 3.1. const ( - ICMPv6PortUnreachable = 4 + ICMPv6NetworkUnreachable = 0 + ICMPv6Prohibited = 1 + ICMPv6BeyondScope = 2 + ICMPv6AddressUnreachable = 3 + ICMPv6PortUnreachable = 4 + ICMPv6Policy = 5 + ICMPv6RejectRoute = 6 ) // Type is the ICMP type field. diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 1b67aa066..83e71cb8c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -129,6 +129,9 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { pkt.Data.TrimFront(header.ICMPv4MinimumSize) switch h.Code() { + case header.ICMPv4HostUnreachable: + e.handleControl(stack.ControlNoRoute, 0, pkt) + case header.ICMPv4PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 3b79749b5..ff1cb53dd 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -128,6 +128,8 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme } pkt.Data.TrimFront(header.ICMPv6DstUnreachableMinimumSize) switch header.ICMPv6(hdr).Code() { + case header.ICMPv6NetworkUnreachable: + e.handleControl(stack.ControlNetworkUnreachable, 0, pkt) case header.ICMPv6PortUnreachable: e.handleControl(stack.ControlPortUnreachable, 0, pkt) } diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index d39baf620..559a1c4dd 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -49,7 +49,8 @@ const ( type manipType int const ( - manipDstPrerouting manipType = iota + manipNone manipType = iota + manipDstPrerouting manipDstOutput ) @@ -113,13 +114,11 @@ type conn struct { // update the state of tcb. It is immutable. tcbHook Hook - // mu protects tcb. + // mu protects all mutable state. mu sync.Mutex `state:"nosave"` - // tcb is TCB control block. It is used to keep track of states // of tcp connection and is protected by mu. tcb tcpconntrack.TCB - // lastUsed is the last time the connection saw a relevant packet, and // is updated by each packet on the connection. It is protected by mu. lastUsed time.Time `state:".(unixTime)"` @@ -141,8 +140,26 @@ func (cn *conn) timedOut(now time.Time) bool { return now.Sub(cn.lastUsed) > defaultTimeout } +// update the connection tracking state. +// +// Precondition: ct.mu must be held. +func (ct *conn) updateLocked(tcpHeader header.TCP, hook Hook) { + // Update the state of tcb. tcb assumes it's always initialized on the + // client. However, we only need to know whether the connection is + // established or not, so the client/server distinction isn't important. + // TODO(gvisor.dev/issue/170): Add support in tcpconntrack to handle + // other tcp states. + if ct.tcb.IsEmpty() { + ct.tcb.Init(tcpHeader) + } else if hook == ct.tcbHook { + ct.tcb.UpdateStateOutbound(tcpHeader) + } else { + ct.tcb.UpdateStateInbound(tcpHeader) + } +} + // ConnTrack tracks all connections created for NAT rules. Most users are -// expected to only call handlePacket and createConnFor. +// expected to only call handlePacket, insertRedirectConn, and maybeInsertNoop. // // ConnTrack keeps all connections in a slice of buckets, each of which holds a // linked list of tuples. This gives us some desirable properties: @@ -248,8 +265,7 @@ func (ct *ConnTrack) connFor(pkt *PacketBuffer) (*conn, direction) { return nil, dirOriginal } -// createConnFor creates a new conn for pkt. -func (ct *ConnTrack) createConnFor(pkt *PacketBuffer, hook Hook, rt RedirectTarget) *conn { +func (ct *ConnTrack) insertRedirectConn(pkt *PacketBuffer, hook Hook, rt RedirectTarget) *conn { tid, err := packetToTupleID(pkt) if err != nil { return nil @@ -272,10 +288,15 @@ func (ct *ConnTrack) createConnFor(pkt *PacketBuffer, hook Hook, rt RedirectTarg manip = manipDstOutput } conn := newConn(tid, replyTID, manip, hook) + ct.insertConn(conn) + return conn +} +// insertConn inserts conn into the appropriate table bucket. +func (ct *ConnTrack) insertConn(conn *conn) { // Lock the buckets in the correct order. - tupleBucket := ct.bucket(tid) - replyBucket := ct.bucket(replyTID) + tupleBucket := ct.bucket(conn.original.tupleID) + replyBucket := ct.bucket(conn.reply.tupleID) ct.mu.RLock() defer ct.mu.RUnlock() if tupleBucket < replyBucket { @@ -289,22 +310,37 @@ func (ct *ConnTrack) createConnFor(pkt *PacketBuffer, hook Hook, rt RedirectTarg ct.buckets[tupleBucket].mu.Lock() } - // Add the tuple to the map. - ct.buckets[tupleBucket].tuples.PushFront(&conn.original) - ct.buckets[replyBucket].tuples.PushFront(&conn.reply) + // Now that we hold the locks, ensure the tuple hasn't been inserted by + // another thread. + alreadyInserted := false + for other := ct.buckets[tupleBucket].tuples.Front(); other != nil; other = other.Next() { + if other.tupleID == conn.original.tupleID { + alreadyInserted = true + break + } + } + + if !alreadyInserted { + // Add the tuple to the map. + ct.buckets[tupleBucket].tuples.PushFront(&conn.original) + ct.buckets[replyBucket].tuples.PushFront(&conn.reply) + } // Unlocking can happen in any order. ct.buckets[tupleBucket].mu.Unlock() if tupleBucket != replyBucket { ct.buckets[replyBucket].mu.Unlock() } - - return conn } // handlePacketPrerouting manipulates ports for packets in Prerouting hook. // TODO(gvisor.dev/issue/170): Change address for Prerouting hook. func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { + // If this is a noop entry, don't do anything. + if conn.manip == manipNone { + return + } + netHeader := header.IPv4(pkt.NetworkHeader) tcpHeader := header.TCP(pkt.TransportHeader) @@ -322,12 +358,22 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { netHeader.SetSourceAddress(conn.original.dstAddr) } + // TODO(gvisor.dev/issue/170): TCP checksums aren't usually validated + // on inbound packets, so we don't recalculate them. However, we should + // support cases when they are validated, e.g. when we can't offload + // receive checksumming. + netHeader.SetChecksum(0) netHeader.SetChecksum(^netHeader.CalculateChecksum()) } // handlePacketOutput manipulates ports for packets in Output hook. func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir direction) { + // If this is a noop entry, don't do anything. + if conn.manip == manipNone { + return + } + netHeader := header.IPv4(pkt.NetworkHeader) tcpHeader := header.TCP(pkt.TransportHeader) @@ -362,20 +408,31 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d } // handlePacket will manipulate the port and address of the packet if the -// connection exists. -func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) { +// connection exists. Returns whether, after the packet traverses the tables, +// it should create a new entry in the table. +func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Route) bool { if pkt.NatDone { - return + return false } if hook != Prerouting && hook != Output { - return + return false + } + + // TODO(gvisor.dev/issue/170): Support other transport protocols. + if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + return false } conn, dir := ct.connFor(pkt) + // Connection or Rule not found for the packet. if conn == nil { - // Connection not found for the packet or the packet is invalid. - return + return true + } + + tcpHeader := header.TCP(pkt.TransportHeader) + if tcpHeader == nil { + return false } switch hook { @@ -395,14 +452,39 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou // Mark the connection as having been used recently so it isn't reaped. conn.lastUsed = time.Now() // Update connection state. - if tcpHeader := header.TCP(pkt.TransportHeader); conn.tcb.IsEmpty() { - conn.tcb.Init(tcpHeader) - conn.tcbHook = hook - } else if hook == conn.tcbHook { - conn.tcb.UpdateStateOutbound(tcpHeader) - } else { - conn.tcb.UpdateStateInbound(tcpHeader) + conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + + return false +} + +// maybeInsertNoop tries to insert a no-op connection entry to keep connections +// from getting clobbered when replies arrive. It only inserts if there isn't +// already a connection for pkt. +// +// This should be called after traversing iptables rules only, to ensure that +// pkt.NatDone is set correctly. +func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) { + // If there were a rule applying to this packet, it would be marked + // with NatDone. + if pkt.NatDone { + return + } + + // We only track TCP connections. + if pkt.NetworkHeader == nil || header.IPv4(pkt.NetworkHeader).TransportProtocol() != header.TCPProtocolNumber { + return + } + + // This is the first packet we're seeing for the TCP connection. Insert + // the noop entry (an identity mapping) so that the response doesn't + // get NATed, breaking the connection. + tid, err := packetToTupleID(pkt) + if err != nil { + return } + conn := newConn(tid, tid.reply(), manipNone, hook) + conn.updateLocked(header.TCP(pkt.TransportHeader), hook) + ct.insertConn(conn) } // bucket gets the conntrack bucket for a tupleID. diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 5f647c5fe..cbbae4224 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -58,8 +58,7 @@ const reaperDelay = 5 * time.Second func DefaultTables() *IPTables { return &IPTables{ tables: [numTables]Table{ - // NAT table. - Table{ + natID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, @@ -68,22 +67,21 @@ func DefaultTables() *IPTables { Rule{Target: ErrorTarget{}}, }, BuiltinChains: [NumHooks]int{ - 0, // Prerouting. - 1, // Input. - HookUnset, // Forward. - 2, // Output. - 3, // Postrouting. + Prerouting: 0, + Input: 1, + Forward: HookUnset, + Output: 2, + Postrouting: 3, }, Underflows: [NumHooks]int{ - 0, // Prerouting. - 1, // Input. - HookUnset, // Forward. - 2, // Output. - 3, // Postrouting. + Prerouting: 0, + Input: 1, + Forward: HookUnset, + Output: 2, + Postrouting: 3, }, }, - // Mangle table. - Table{ + mangleID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, @@ -94,15 +92,14 @@ func DefaultTables() *IPTables { Output: 1, }, Underflows: [NumHooks]int{ - 0, // Prerouting. - HookUnset, // Input. - HookUnset, // Forward. - 1, // Output. - HookUnset, // Postrouting. + Prerouting: 0, + Input: HookUnset, + Forward: HookUnset, + Output: 1, + Postrouting: HookUnset, }, }, - // Filter table. - Table{ + filterID: Table{ Rules: []Rule{ Rule{Target: AcceptTarget{}}, Rule{Target: AcceptTarget{}}, @@ -110,27 +107,25 @@ func DefaultTables() *IPTables { Rule{Target: ErrorTarget{}}, }, BuiltinChains: [NumHooks]int{ - HookUnset, // Prerouting. - Input: 0, // Input. - Forward: 1, // Forward. - Output: 2, // Output. - HookUnset, // Postrouting. + Prerouting: HookUnset, + Input: 0, + Forward: 1, + Output: 2, + Postrouting: HookUnset, }, Underflows: [NumHooks]int{ - HookUnset, // Prerouting. - 0, // Input. - 1, // Forward. - 2, // Output. - HookUnset, // Postrouting. + Prerouting: HookUnset, + Input: 0, + Forward: 1, + Output: 2, + Postrouting: HookUnset, }, }, }, priorities: [NumHooks][]tableID{ - []tableID{mangleID, natID}, // Prerouting. - []tableID{natID, filterID}, // Input. - []tableID{}, // Forward. - []tableID{mangleID, natID, filterID}, // Output. - []tableID{}, // Postrouting. + Prerouting: []tableID{mangleID, natID}, + Input: []tableID{natID, filterID}, + Output: []tableID{mangleID, natID, filterID}, }, connections: ConnTrack{ seed: generateRandUint32(), @@ -145,18 +140,12 @@ func EmptyFilterTable() Table { return Table{ Rules: []Rule{}, BuiltinChains: [NumHooks]int{ - HookUnset, - 0, - 0, - 0, - HookUnset, + Prerouting: HookUnset, + Postrouting: HookUnset, }, Underflows: [NumHooks]int{ - HookUnset, - 0, - 0, - 0, - HookUnset, + Prerouting: HookUnset, + Postrouting: HookUnset, }, } } @@ -167,18 +156,10 @@ func EmptyNATTable() Table { return Table{ Rules: []Rule{}, BuiltinChains: [NumHooks]int{ - 0, - 0, - HookUnset, - 0, - 0, + Forward: HookUnset, }, Underflows: [NumHooks]int{ - 0, - 0, - HookUnset, - 0, - 0, + Forward: HookUnset, }, } } @@ -245,13 +226,18 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr // Packets are manipulated only if connection and matching // NAT rule exists. - it.connections.handlePacket(pkt, hook, gso, r) + shouldTrack := it.connections.handlePacket(pkt, hook, gso, r) // Go through each table containing the hook. it.mu.RLock() defer it.mu.RUnlock() priorities := it.priorities[hook] for _, tableID := range priorities { + // If handlePacket already NATed the packet, we don't need to + // check the NAT table. + if tableID == natID && pkt.NatDone { + continue + } table := it.tables[tableID] ruleIdx := table.BuiltinChains[hook] switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict { @@ -281,6 +267,20 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr } } + // If this connection should be tracked, try to add an entry for it. If + // traversing the nat table didn't end in adding an entry, + // maybeInsertNoop will add a no-op entry for the connection. This is + // needeed when establishing connections so that the SYN/ACK reply to an + // outgoing SYN is delivered to the correct endpoint rather than being + // redirected by a prerouting rule. + // + // From the iptables documentation: "If there is no rule, a `null' + // binding is created: this usually does not map the packet, but exists + // to ensure we don't map another stream over an existing one." + if shouldTrack { + it.connections.maybeInsertNoop(pkt, hook) + } + // Every table returned Accept. return true } diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index d43f60c67..dc88033c7 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -153,7 +153,7 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso // Set up conection for matching NAT rule. Only the first // packet of the connection comes here. Other packets will be // manipulated in connection tracking. - if conn := ct.createConnFor(pkt, hook, rt); conn != nil { + if conn := ct.insertRedirectConn(pkt, hook, rt); conn != nil { ct.handlePacket(pkt, hook, gso, r) } default: diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index e28c23d66..9dce11a97 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -469,7 +469,7 @@ type ndpState struct { rtrSolicit struct { // The timer used to send the next router solicitation message. - timer *time.Timer + timer tcpip.Timer // Used to let the Router Solicitation timer know that it has been stopped. // @@ -503,7 +503,7 @@ type ndpState struct { // to the DAD goroutine that DAD should stop. type dadState struct { // The DAD timer to send the next NS message, or resolve the address. - timer *time.Timer + timer tcpip.Timer // Used to let the DAD timer know that it has been stopped. // @@ -515,38 +515,38 @@ type dadState struct { // defaultRouterState holds data associated with a default router discovered by // a Router Advertisement (RA). type defaultRouterState struct { - // Timer to invalidate the default router. + // Job to invalidate the default router. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job } // onLinkPrefixState holds data associated with an on-link prefix discovered by // a Router Advertisement's Prefix Information option (PI) when the NDP // configurations was configured to do so. type onLinkPrefixState struct { - // Timer to invalidate the on-link prefix. + // Job to invalidate the on-link prefix. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job } // tempSLAACAddrState holds state associated with a temporary SLAAC address. type tempSLAACAddrState struct { - // Timer to deprecate the temporary SLAAC address. + // Job to deprecate the temporary SLAAC address. // // Must not be nil. - deprecationTimer *tcpip.CancellableTimer + deprecationJob *tcpip.Job - // Timer to invalidate the temporary SLAAC address. + // Job to invalidate the temporary SLAAC address. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job - // Timer to regenerate the temporary SLAAC address. + // Job to regenerate the temporary SLAAC address. // // Must not be nil. - regenTimer *tcpip.CancellableTimer + regenJob *tcpip.Job createdAt time.Time @@ -561,15 +561,15 @@ type tempSLAACAddrState struct { // slaacPrefixState holds state associated with a SLAAC prefix. type slaacPrefixState struct { - // Timer to deprecate the prefix. + // Job to deprecate the prefix. // // Must not be nil. - deprecationTimer *tcpip.CancellableTimer + deprecationJob *tcpip.Job - // Timer to invalidate the prefix. + // Job to invalidate the prefix. // // Must not be nil. - invalidationTimer *tcpip.CancellableTimer + invalidationJob *tcpip.Job // Nonzero only when the address is not valid forever. validUntil time.Time @@ -651,12 +651,12 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref } var done bool - var timer *time.Timer + var timer tcpip.Timer // We initially start a timer to fire immediately because some of the DAD work // cannot be done while holding the NIC's lock. This is effectively the same // as starting a goroutine but we use a timer that fires immediately so we can // reset it for the next DAD iteration. - timer = time.AfterFunc(0, func() { + timer = ndp.nic.stack.Clock().AfterFunc(0, func() { ndp.nic.mu.Lock() defer ndp.nic.mu.Unlock() @@ -871,9 +871,9 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { case ok && rl != 0: // This is an already discovered default router. Update - // the invalidation timer. - rtr.invalidationTimer.StopLocked() - rtr.invalidationTimer.Reset(rl) + // the invalidation job. + rtr.invalidationJob.Cancel() + rtr.invalidationJob.Schedule(rl) ndp.defaultRouters[ip] = rtr case ok && rl == 0: @@ -950,7 +950,7 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { return } - rtr.invalidationTimer.StopLocked() + rtr.invalidationJob.Cancel() delete(ndp.defaultRouters, ip) // Let the integrator know a discovered default router is invalidated. @@ -979,12 +979,12 @@ func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { } state := defaultRouterState{ - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { ndp.invalidateDefaultRouter(ip) }), } - state.invalidationTimer.Reset(rl) + state.invalidationJob.Schedule(rl) ndp.defaultRouters[ip] = state } @@ -1009,13 +1009,13 @@ func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) } state := onLinkPrefixState{ - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { ndp.invalidateOnLinkPrefix(prefix) }), } if l < header.NDPInfiniteLifetime { - state.invalidationTimer.Reset(l) + state.invalidationJob.Schedule(l) } ndp.onLinkPrefixes[prefix] = state @@ -1033,7 +1033,7 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { return } - s.invalidationTimer.StopLocked() + s.invalidationJob.Cancel() delete(ndp.onLinkPrefixes, prefix) // Let the integrator know a discovered on-link prefix is invalidated. @@ -1082,14 +1082,14 @@ func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformatio // This is an already discovered on-link prefix with a // new non-zero valid lifetime. // - // Update the invalidation timer. + // Update the invalidation job. - prefixState.invalidationTimer.StopLocked() + prefixState.invalidationJob.Cancel() if vl < header.NDPInfiniteLifetime { - // Prefix is valid for a finite lifetime, reset the timer to expire after + // Prefix is valid for a finite lifetime, schedule the job to execute after // the new valid lifetime. - prefixState.invalidationTimer.Reset(vl) + prefixState.invalidationJob.Schedule(vl) } ndp.onLinkPrefixes[prefix] = prefixState @@ -1154,7 +1154,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { } state := slaacPrefixState{ - deprecationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the deprecated SLAAC prefix %s", prefix)) @@ -1162,7 +1162,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { ndp.deprecateSLAACAddress(state.stableAddr.ref) }), - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the invalidated SLAAC prefix %s", prefix)) @@ -1184,19 +1184,19 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { if !ndp.generateSLAACAddr(prefix, &state) { // We were unable to generate an address for the prefix, we do not nothing - // further as there is no reason to maintain state or timers for a prefix we + // further as there is no reason to maintain state or jobs for a prefix we // do not have an address for. return } - // Setup the initial timers to deprecate and invalidate prefix. + // Setup the initial jobs to deprecate and invalidate prefix. if pl < header.NDPInfiniteLifetime && pl != 0 { - state.deprecationTimer.Reset(pl) + state.deprecationJob.Schedule(pl) } if vl < header.NDPInfiniteLifetime { - state.invalidationTimer.Reset(vl) + state.invalidationJob.Schedule(vl) state.validUntil = now.Add(vl) } @@ -1428,7 +1428,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla } state := tempSLAACAddrState{ - deprecationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to deprecate temporary address %s", prefix, generatedAddr)) @@ -1441,7 +1441,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.deprecateSLAACAddress(tempAddrState.ref) }), - invalidationTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to invalidate temporary address %s", prefix, generatedAddr)) @@ -1454,7 +1454,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, generatedAddr.Address, tempAddrState) }), - regenTimer: tcpip.NewCancellableTimer(&ndp.nic.mu, func() { + regenJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to regenerate temporary address after %s", prefix, generatedAddr)) @@ -1481,9 +1481,9 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ref: ref, } - state.deprecationTimer.Reset(pl) - state.invalidationTimer.Reset(vl) - state.regenTimer.Reset(pl - ndp.configs.RegenAdvanceDuration) + state.deprecationJob.Schedule(pl) + state.invalidationJob.Schedule(vl) + state.regenJob.Schedule(pl - ndp.configs.RegenAdvanceDuration) prefixState.generationAttempts++ prefixState.tempAddrs[generatedAddr.Address] = state @@ -1518,16 +1518,16 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat prefixState.stableAddr.ref.deprecated = false } - // If prefix was preferred for some finite lifetime before, stop the - // deprecation timer so it can be reset. - prefixState.deprecationTimer.StopLocked() + // If prefix was preferred for some finite lifetime before, cancel the + // deprecation job so it can be reset. + prefixState.deprecationJob.Cancel() now := time.Now() - // Reset the deprecation timer if prefix has a finite preferred lifetime. + // Schedule the deprecation job if prefix has a finite preferred lifetime. if pl < header.NDPInfiniteLifetime { if !deprecated { - prefixState.deprecationTimer.Reset(pl) + prefixState.deprecationJob.Schedule(pl) } prefixState.preferredUntil = now.Add(pl) } else { @@ -1546,9 +1546,9 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // 3) Otherwise, reset the valid lifetime of the prefix to 2 hours. if vl >= header.NDPInfiniteLifetime { - // Handle the infinite valid lifetime separately as we do not keep a timer - // in this case. - prefixState.invalidationTimer.StopLocked() + // Handle the infinite valid lifetime separately as we do not schedule a + // job in this case. + prefixState.invalidationJob.Cancel() prefixState.validUntil = time.Time{} } else { var effectiveVl time.Duration @@ -1569,8 +1569,8 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } if effectiveVl != 0 { - prefixState.invalidationTimer.StopLocked() - prefixState.invalidationTimer.Reset(effectiveVl) + prefixState.invalidationJob.Cancel() + prefixState.invalidationJob.Schedule(effectiveVl) prefixState.validUntil = now.Add(effectiveVl) } } @@ -1582,7 +1582,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // Note, we do not need to update the entries in the temporary address map - // after updating the timers because the timers are held as pointers. + // after updating the jobs because the jobs are held as pointers. var regenForAddr tcpip.Address allAddressesRegenerated := true for tempAddr, tempAddrState := range prefixState.tempAddrs { @@ -1596,14 +1596,14 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // If the address is no longer valid, invalidate it immediately. Otherwise, - // reset the invalidation timer. + // reset the invalidation job. newValidLifetime := validUntil.Sub(now) if newValidLifetime <= 0 { ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, tempAddr, tempAddrState) continue } - tempAddrState.invalidationTimer.StopLocked() - tempAddrState.invalidationTimer.Reset(newValidLifetime) + tempAddrState.invalidationJob.Cancel() + tempAddrState.invalidationJob.Schedule(newValidLifetime) // As per RFC 4941 section 3.3 step 4, the preferred lifetime of a temporary // address is the lower of the preferred lifetime of the stable address or @@ -1616,17 +1616,17 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } // If the address is no longer preferred, deprecate it immediately. - // Otherwise, reset the deprecation timer. + // Otherwise, schedule the deprecation job again. newPreferredLifetime := preferredUntil.Sub(now) - tempAddrState.deprecationTimer.StopLocked() + tempAddrState.deprecationJob.Cancel() if newPreferredLifetime <= 0 { ndp.deprecateSLAACAddress(tempAddrState.ref) } else { tempAddrState.ref.deprecated = false - tempAddrState.deprecationTimer.Reset(newPreferredLifetime) + tempAddrState.deprecationJob.Schedule(newPreferredLifetime) } - tempAddrState.regenTimer.StopLocked() + tempAddrState.regenJob.Cancel() if tempAddrState.regenerated { } else { allAddressesRegenerated = false @@ -1637,7 +1637,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // immediately after we finish iterating over the temporary addresses. regenForAddr = tempAddr } else { - tempAddrState.regenTimer.Reset(newPreferredLifetime - ndp.configs.RegenAdvanceDuration) + tempAddrState.regenJob.Schedule(newPreferredLifetime - ndp.configs.RegenAdvanceDuration) } } } @@ -1717,7 +1717,7 @@ func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPr ndp.cleanupSLAACPrefixResources(prefix, state) } -// cleanupSLAACPrefixResources cleansup a SLAAC prefix's timers and entry. +// cleanupSLAACPrefixResources cleans up a SLAAC prefix's jobs and entry. // // Panics if the SLAAC prefix is not known. // @@ -1729,8 +1729,8 @@ func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaa } state.stableAddr.ref = nil - state.deprecationTimer.StopLocked() - state.invalidationTimer.StopLocked() + state.deprecationJob.Cancel() + state.invalidationJob.Cancel() delete(ndp.slaacPrefixes, prefix) } @@ -1775,13 +1775,13 @@ func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWi } // cleanupTempSLAACAddrResourcesAndNotify cleans up a temporary SLAAC address's -// timers and entry. +// jobs and entry. // // The NIC that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResources(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) { - tempAddrState.deprecationTimer.StopLocked() - tempAddrState.invalidationTimer.StopLocked() - tempAddrState.regenTimer.StopLocked() + tempAddrState.deprecationJob.Cancel() + tempAddrState.invalidationJob.Cancel() + tempAddrState.regenJob.Cancel() delete(tempAddrs, tempAddr) } @@ -1860,7 +1860,7 @@ func (ndp *ndpState) startSolicitingRouters() { var done bool ndp.rtrSolicit.done = &done - ndp.rtrSolicit.timer = time.AfterFunc(delay, func() { + ndp.rtrSolicit.timer = ndp.nic.stack.Clock().AfterFunc(delay, func() { ndp.nic.mu.Lock() if done { // If we reach this point, it means that the RS timer fired after another diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 6f86abc98..644ba7c33 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -1254,7 +1254,7 @@ func TestRouterDiscovery(t *testing.T) { default: } - // Wait for lladdr2's router invalidation timer to fire. The lifetime + // Wait for lladdr2's router invalidation job to execute. The lifetime // of the router should have been updated to the most recent (smaller) // lifetime. // @@ -1271,7 +1271,7 @@ func TestRouterDiscovery(t *testing.T) { e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr2, 0)) expectRouterEvent(llAddr2, false) - // Wait for lladdr3's router invalidation timer to fire. The lifetime + // Wait for lladdr3's router invalidation job to execute. The lifetime // of the router should have been updated to the most recent (smaller) // lifetime. // @@ -1502,7 +1502,7 @@ func TestPrefixDiscovery(t *testing.T) { default: } - // Wait for prefix2's most recent invalidation timer plus some buffer to + // Wait for prefix2's most recent invalidation job plus some buffer to // expire. select { case e := <-ndpDisp.prefixC: @@ -2395,7 +2395,7 @@ func TestAutoGenTempAddrRegen(t *testing.T) { for _, addr := range tempAddrs { // Wait for a deprecation then invalidation event, or just an invalidation // event. We need to cover both cases but cannot deterministically hit both - // cases because the deprecation and invalidation timers could fire in any + // cases because the deprecation and invalidation jobs could execute in any // order. select { case e := <-ndpDisp.autoGenAddrC: @@ -2432,9 +2432,9 @@ func TestAutoGenTempAddrRegen(t *testing.T) { } } -// TestAutoGenTempAddrRegenTimerUpdates tests that a temporary address's -// regeneration timer gets updated when refreshing the address's lifetimes. -func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { +// TestAutoGenTempAddrRegenJobUpdates tests that a temporary address's +// regeneration job gets updated when refreshing the address's lifetimes. +func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { const ( nicID = 1 regenAfter = 2 * time.Second @@ -2533,7 +2533,7 @@ func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { // // A new temporary address should immediately be generated since the // regeneration time has already passed since the last address was generated - // - this regeneration does not depend on a timer. + // - this regeneration does not depend on a job. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100)) expectAutoGenAddrEvent(tempAddr2, newAddr) @@ -2559,11 +2559,11 @@ func TestAutoGenTempAddrRegenTimerUpdates(t *testing.T) { } // Set the maximum lifetimes for temporary addresses such that on the next - // RA, the regeneration timer gets reset. + // RA, the regeneration job gets scheduled again. // // The maximum lifetime is the sum of the minimum lifetimes for temporary // addresses + the time that has already passed since the last address was - // generated so that the regeneration timer is needed to generate the next + // generated so that the regeneration job is needed to generate the next // address. newLifetimes := newMinVLDuration + regenAfter + defaultAsyncNegativeEventTimeout ndpConfigs.MaxTempAddrValidLifetime = newLifetimes @@ -2993,9 +2993,9 @@ func TestAutoGenAddrDeprecateFromPI(t *testing.T) { expectPrimaryAddr(addr2) } -// TestAutoGenAddrTimerDeprecation tests that an address is properly deprecated +// TestAutoGenAddrJobDeprecation tests that an address is properly deprecated // when its preferred lifetime expires. -func TestAutoGenAddrTimerDeprecation(t *testing.T) { +func TestAutoGenAddrJobDeprecation(t *testing.T) { const nicID = 1 const newMinVL = 2 newMinVLDuration := newMinVL * time.Second @@ -3513,8 +3513,8 @@ func TestAutoGenAddrRemoval(t *testing.T) { } expectAutoGenAddrEvent(addr, invalidatedAddr) - // Wait for the original valid lifetime to make sure the original timer - // got stopped/cleaned up. + // Wait for the original valid lifetime to make sure the original job got + // cancelled/cleaned up. select { case <-ndpDisp.autoGenAddrC: t.Fatal("unexpectedly received an auto gen addr event") diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index cd4b7a449..9e1b2d25f 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -52,8 +52,11 @@ type TransportEndpointID struct { type ControlType int // The following are the allowed values for ControlType values. +// TODO(http://gvisor.dev/issue/3210): Support time exceeded messages. const ( - ControlPacketTooBig ControlType = iota + ControlNetworkUnreachable ControlType = iota + ControlNoRoute + ControlPacketTooBig ControlPortUnreachable ControlUnknown ) diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 2b7ece851..a6faa22c2 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -728,6 +728,11 @@ func New(opts Options) *Stack { return s } +// newJob returns a tcpip.Job using the Stack clock. +func (s *Stack) newJob(l sync.Locker, f func()) *tcpip.Job { + return tcpip.NewJob(s.clock, l, f) +} + // UniqueID returns a unique identifier. func (s *Stack) UniqueID() uint64 { return s.uniqueIDGenerator.UniqueID() @@ -801,9 +806,10 @@ func (s *Stack) SetTransportProtocolHandler(p tcpip.TransportProtocolNumber, h f } } -// NowNanoseconds implements tcpip.Clock.NowNanoseconds. -func (s *Stack) NowNanoseconds() int64 { - return s.clock.NowNanoseconds() +// Clock returns the Stack's clock for retrieving the current time and +// scheduling work. +func (s *Stack) Clock() tcpip.Clock { + return s.clock } // Stats returns a mutable copy of the current stats. diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index ff14a3b3c..21aafb0a2 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -192,7 +192,7 @@ func (e ErrSaveRejection) Error() string { return "save rejected due to unsupported networking state: " + e.Err.Error() } -// A Clock provides the current time. +// A Clock provides the current time and schedules work for execution. // // Times returned by a Clock should always be used for application-visible // time. Only monotonic times should be used for netstack internal timekeeping. @@ -203,6 +203,31 @@ type Clock interface { // NowMonotonic returns a monotonic time value. NowMonotonic() int64 + + // AfterFunc waits for the duration to elapse and then calls f in its own + // goroutine. It returns a Timer that can be used to cancel the call using + // its Stop method. + AfterFunc(d time.Duration, f func()) Timer +} + +// Timer represents a single event. A Timer must be created with +// Clock.AfterFunc. +type Timer interface { + // Stop prevents the Timer from firing. It returns true if the call stops the + // timer, false if the timer has already expired or been stopped. + // + // If Stop returns false, then the timer has already expired and the function + // f of Clock.AfterFunc(d, f) has been started in its own goroutine; Stop + // does not wait for f to complete before returning. If the caller needs to + // know whether f is completed, it must coordinate with f explicitly. + Stop() bool + + // Reset changes the timer to expire after duration d. + // + // Reset should be invoked only on stopped or expired timers. If the timer is + // known to have expired, Reset can be used directly. Otherwise, the caller + // must coordinate with the function f of Clock.AfterFunc(d, f). + Reset(d time.Duration) } // Address is a byte slice cast as a string that represents the address of a diff --git a/pkg/tcpip/time_unsafe.go b/pkg/tcpip/time_unsafe.go index 7f172f978..f32d58091 100644 --- a/pkg/tcpip/time_unsafe.go +++ b/pkg/tcpip/time_unsafe.go @@ -20,7 +20,7 @@ package tcpip import ( - _ "time" // Used with go:linkname. + "time" // Used with go:linkname. _ "unsafe" // Required for go:linkname. ) @@ -45,3 +45,31 @@ func (*StdClock) NowMonotonic() int64 { _, _, mono := now() return mono } + +// AfterFunc implements Clock.AfterFunc. +func (*StdClock) AfterFunc(d time.Duration, f func()) Timer { + return &stdTimer{ + t: time.AfterFunc(d, f), + } +} + +type stdTimer struct { + t *time.Timer +} + +var _ Timer = (*stdTimer)(nil) + +// Stop implements Timer.Stop. +func (st *stdTimer) Stop() bool { + return st.t.Stop() +} + +// Reset implements Timer.Reset. +func (st *stdTimer) Reset(d time.Duration) { + st.t.Reset(d) +} + +// NewStdTimer returns a Timer implemented with the time package. +func NewStdTimer(t *time.Timer) Timer { + return &stdTimer{t: t} +} diff --git a/pkg/tcpip/timer.go b/pkg/tcpip/timer.go index 5554c573f..f1dd7c310 100644 --- a/pkg/tcpip/timer.go +++ b/pkg/tcpip/timer.go @@ -20,50 +20,49 @@ import ( "gvisor.dev/gvisor/pkg/sync" ) -// cancellableTimerInstance is a specific instance of CancellableTimer. +// jobInstance is a specific instance of Job. // -// Different instances are created each time CancellableTimer is Reset so each -// timer has its own earlyReturn signal. This is to address a bug when a -// CancellableTimer is stopped and reset in quick succession resulting in a -// timer instance's earlyReturn signal being affected or seen by another timer -// instance. +// Different instances are created each time Job is scheduled so each timer has +// its own earlyReturn signal. This is to address a bug when a Job is stopped +// and reset in quick succession resulting in a timer instance's earlyReturn +// signal being affected or seen by another timer instance. // // Consider the following sceneario where timer instances share a common // earlyReturn signal (T1 creates, stops and resets a Cancellable timer under a // lock L; T2, T3, T4 and T5 are goroutines that handle the first (A), second // (B), third (C), and fourth (D) instance of the timer firing, respectively): // T1: Obtain L -// T1: Create a new CancellableTimer w/ lock L (create instance A) +// T1: Create a new Job w/ lock L (create instance A) // T2: instance A fires, blocked trying to obtain L. // T1: Attempt to stop instance A (set earlyReturn = true) -// T1: Reset timer (create instance B) +// T1: Schedule timer (create instance B) // T3: instance B fires, blocked trying to obtain L. // T1: Attempt to stop instance B (set earlyReturn = true) -// T1: Reset timer (create instance C) +// T1: Schedule timer (create instance C) // T4: instance C fires, blocked trying to obtain L. // T1: Attempt to stop instance C (set earlyReturn = true) -// T1: Reset timer (create instance D) +// T1: Schedule timer (create instance D) // T5: instance D fires, blocked trying to obtain L. // T1: Release L // -// Now that T1 has released L, any of the 4 timer instances can take L and check -// earlyReturn. If the timers simply check earlyReturn and then do nothing -// further, then instance D will never early return even though it was not -// requested to stop. If the timers reset earlyReturn before early returning, -// then all but one of the timers will do work when only one was expected to. -// If CancellableTimer resets earlyReturn when resetting, then all the timers +// Now that T1 has released L, any of the 4 timer instances can take L and +// check earlyReturn. If the timers simply check earlyReturn and then do +// nothing further, then instance D will never early return even though it was +// not requested to stop. If the timers reset earlyReturn before early +// returning, then all but one of the timers will do work when only one was +// expected to. If Job resets earlyReturn when resetting, then all the timers // will fire (again, when only one was expected to). // // To address the above concerns the simplest solution was to give each timer // its own earlyReturn signal. -type cancellableTimerInstance struct { - timer *time.Timer +type jobInstance struct { + timer Timer // Used to inform the timer to early return when it gets stopped while the // lock the timer tries to obtain when fired is held (T1 is a goroutine that // tries to cancel the timer and T2 is the goroutine that handles the timer // firing): - // T1: Obtain the lock, then call StopLocked() + // T1: Obtain the lock, then call Cancel() // T2: timer fires, and gets blocked on obtaining the lock // T1: Releases lock // T2: Obtains lock does unintended work @@ -74,29 +73,33 @@ type cancellableTimerInstance struct { earlyReturn *bool } -// stop stops the timer instance t from firing if it hasn't fired already. If it +// stop stops the job instance j from firing if it hasn't fired already. If it // has fired and is blocked at obtaining the lock, earlyReturn will be set to // true so that it will early return when it obtains the lock. -func (t *cancellableTimerInstance) stop() { - if t.timer != nil { - t.timer.Stop() - *t.earlyReturn = true +func (j *jobInstance) stop() { + if j.timer != nil { + j.timer.Stop() + *j.earlyReturn = true } } -// CancellableTimer is a timer that does some work and can be safely cancelled -// when it fires at the same time some "related work" is being done. +// Job represents some work that can be scheduled for execution. The work can +// be safely cancelled when it fires at the same time some "related work" is +// being done. // // The term "related work" is defined as some work that needs to be done while // holding some lock that the timer must also hold while doing some work. // -// Note, it is not safe to copy a CancellableTimer as its timer instance creates -// a closure over the address of the CancellableTimer. -type CancellableTimer struct { +// Note, it is not safe to copy a Job as its timer instance creates +// a closure over the address of the Job. +type Job struct { _ sync.NoCopy + // The clock used to schedule the backing timer + clock Clock + // The active instance of a cancellable timer. - instance cancellableTimerInstance + instance jobInstance // locker is the lock taken by the timer immediately after it fires and must // be held when attempting to stop the timer. @@ -113,59 +116,91 @@ type CancellableTimer struct { fn func() } -// StopLocked prevents the Timer from firing if it has not fired already. +// Cancel prevents the Job from executing if it has not executed already. // -// If the timer is blocked on obtaining the t.locker lock when StopLocked is -// called, it will early return instead of calling t.fn. +// Cancel requires appropriate locking to be in place for any resources managed +// by the Job. If the Job is blocked on obtaining the lock when Cancel is +// called, it will early return. // // Note, t will be modified. // -// t.locker MUST be locked. -func (t *CancellableTimer) StopLocked() { - t.instance.stop() +// j.locker MUST be locked. +func (j *Job) Cancel() { + j.instance.stop() // Nothing to do with the stopped instance anymore. - t.instance = cancellableTimerInstance{} + j.instance = jobInstance{} } -// Reset changes the timer to expire after duration d. +// Schedule schedules the Job for execution after duration d. This can be +// called on cancelled or completed Jobs to schedule them again. // -// Note, t will be modified. +// Schedule should be invoked only on unscheduled, cancelled, or completed +// Jobs. To be safe, callers should always call Cancel before calling Schedule. // -// Reset should only be called on stopped or expired timers. To be safe, callers -// should always call StopLocked before calling Reset. -func (t *CancellableTimer) Reset(d time.Duration) { +// Note, j will be modified. +func (j *Job) Schedule(d time.Duration) { // Create a new instance. earlyReturn := false // Capture the locker so that updating the timer does not cause a data race // when a timer fires and tries to obtain the lock (read the timer's locker). - locker := t.locker - t.instance = cancellableTimerInstance{ - timer: time.AfterFunc(d, func() { + locker := j.locker + j.instance = jobInstance{ + timer: j.clock.AfterFunc(d, func() { locker.Lock() defer locker.Unlock() if earlyReturn { // If we reach this point, it means that the timer fired while another - // goroutine called StopLocked while it had the lock. Simply return - // here and do nothing further. + // goroutine called Cancel while it had the lock. Simply return here + // and do nothing further. earlyReturn = false return } - t.fn() + j.fn() }), earlyReturn: &earlyReturn, } } -// NewCancellableTimer returns an unscheduled CancellableTimer with the given -// locker and fn. -// -// fn MUST NOT attempt to lock locker. -// -// Callers must call Reset to schedule the timer to fire. -func NewCancellableTimer(locker sync.Locker, fn func()) *CancellableTimer { - return &CancellableTimer{locker: locker, fn: fn} +// NewJob returns a new Job that can be used to schedule f to run in its own +// gorountine. l will be locked before calling f then unlocked after f returns. +// +// var clock tcpip.StdClock +// var mu sync.Mutex +// message := "foo" +// job := tcpip.NewJob(&clock, &mu, func() { +// fmt.Println(message) +// }) +// job.Schedule(time.Second) +// +// mu.Lock() +// message = "bar" +// mu.Unlock() +// +// // Output: bar +// +// f MUST NOT attempt to lock l. +// +// l MUST be locked prior to calling the returned job's Cancel(). +// +// var clock tcpip.StdClock +// var mu sync.Mutex +// message := "foo" +// job := tcpip.NewJob(&clock, &mu, func() { +// fmt.Println(message) +// }) +// job.Schedule(time.Second) +// +// mu.Lock() +// job.Cancel() +// mu.Unlock() +func NewJob(c Clock, l sync.Locker, f func()) *Job { + return &Job{ + clock: c, + locker: l, + fn: f, + } } diff --git a/pkg/tcpip/timer_test.go b/pkg/tcpip/timer_test.go index b4940e397..a82384c49 100644 --- a/pkg/tcpip/timer_test.go +++ b/pkg/tcpip/timer_test.go @@ -28,8 +28,8 @@ const ( longDuration = 1 * time.Second ) -func TestCancellableTimerReassignment(t *testing.T) { - var timer tcpip.CancellableTimer +func TestJobReschedule(t *testing.T) { + var clock tcpip.StdClock var wg sync.WaitGroup var lock sync.Mutex @@ -43,26 +43,27 @@ func TestCancellableTimerReassignment(t *testing.T) { // that has an active timer (even if it has been stopped as a stopped // timer may be blocked on a lock before it can check if it has been // stopped while another goroutine holds the same lock). - timer = *tcpip.NewCancellableTimer(&lock, func() { + job := tcpip.NewJob(&clock, &lock, func() { wg.Done() }) - timer.Reset(shortDuration) + job.Schedule(shortDuration) lock.Unlock() }() } wg.Wait() } -func TestCancellableTimerFire(t *testing.T) { +func TestJobExecution(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) - timer := tcpip.NewCancellableTimer(&lock, func() { + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -82,17 +83,18 @@ func TestCancellableTimerFire(t *testing.T) { func TestCancellableTimerResetFromLongDuration(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(middleDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(middleDuration) lock.Lock() - timer.StopLocked() + job.Cancel() lock.Unlock() - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -109,16 +111,17 @@ func TestCancellableTimerResetFromLongDuration(t *testing.T) { } } -func TestCancellableTimerResetFromShortDuration(t *testing.T) { +func TestJobRescheduleFromShortDuration(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() // Wait for timer to fire if it wasn't correctly stopped. @@ -128,7 +131,7 @@ func TestCancellableTimerResetFromShortDuration(t *testing.T) { case <-time.After(middleDuration): } - timer.Reset(shortDuration) + job.Schedule(shortDuration) // Wait for timer to fire. select { @@ -145,17 +148,18 @@ func TestCancellableTimerResetFromShortDuration(t *testing.T) { } } -func TestCancellableTimerImmediatelyStop(t *testing.T) { +func TestJobImmediatelyCancel(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) for i := 0; i < 1000; i++ { lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() } @@ -167,25 +171,26 @@ func TestCancellableTimerImmediatelyStop(t *testing.T) { } } -func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) { +func TestJobCancelledRescheduleWithoutLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) - timer.StopLocked() + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) + job.Cancel() lock.Unlock() for i := 0; i < 10; i++ { - timer.Reset(middleDuration) + job.Schedule(middleDuration) lock.Lock() // Sleep until the timer fires and gets blocked trying to take the lock. time.Sleep(middleDuration * 2) - timer.StopLocked() + job.Cancel() lock.Unlock() } @@ -201,17 +206,18 @@ func TestCancellableTimerStoppedResetWithoutLock(t *testing.T) { func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) for i := 0; i < 10; i++ { // Sleep until the timer fires and gets blocked trying to take the lock. time.Sleep(middleDuration) - timer.StopLocked() - timer.Reset(shortDuration) + job.Cancel() + job.Schedule(shortDuration) } lock.Unlock() @@ -230,18 +236,19 @@ func TestManyCancellableTimerResetAfterBlockedOnLock(t *testing.T) { } } -func TestManyCancellableTimerResetUnderLock(t *testing.T) { +func TestManyJobReschedulesUnderLock(t *testing.T) { t.Parallel() - ch := make(chan struct{}) + var clock tcpip.StdClock var lock sync.Mutex + ch := make(chan struct{}) lock.Lock() - timer := tcpip.NewCancellableTimer(&lock, func() { ch <- struct{}{} }) - timer.Reset(shortDuration) + job := tcpip.NewJob(&clock, &lock, func() { ch <- struct{}{} }) + job.Schedule(shortDuration) for i := 0; i < 10; i++ { - timer.StopLocked() - timer.Reset(shortDuration) + job.Cancel() + job.Schedule(shortDuration) } lock.Unlock() diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 678f4e016..4612be4e7 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -797,7 +797,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() - packet.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.Clock().NowNanoseconds() e.rcvMu.Unlock() e.stats.PacketsReceived.Increment() diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 8f167391f..0e46e6355 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -499,7 +499,7 @@ func (ep *endpoint) HandlePacket(nicID tcpip.NICID, localAddr tcpip.LinkAddress, combinedVV.Append(pkt.Data) packet.data = combinedVV } - packet.timestampNS = ep.stack.NowNanoseconds() + packet.timestampNS = ep.stack.Clock().NowNanoseconds() ep.rcvList.PushBack(&packet) ep.rcvBufSize += packet.data.Size() diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index c2e9fd29f..f85a68554 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -456,7 +456,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { defer e.mu.Unlock() // If a local address was specified, verify that it's valid. - if e.stack.CheckLocalAddress(addr.NIC, e.NetProto, addr.Addr) == 0 { + if len(addr.Addr) != 0 && e.stack.CheckLocalAddress(addr.NIC, e.NetProto, addr.Addr) == 0 { return tcpip.ErrBadLocalAddress } @@ -700,7 +700,7 @@ func (e *endpoint) HandlePacket(route *stack.Route, pkt *stack.PacketBuffer) { } combinedVV.Append(pkt.Data) packet.data = combinedVV - packet.timestampNS = e.stack.NowNanoseconds() + packet.timestampNS = e.stack.Clock().NowNanoseconds() e.rcvList.PushBack(packet) e.rcvBufSize += packet.data.Size() diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 81b740115..1798510bc 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -490,6 +490,9 @@ func (h *handshake) resolveRoute() *tcpip.Error { <-h.ep.undrain h.ep.mu.Lock() } + if n¬ifyError != 0 { + return h.ep.takeLastError() + } } // Wait for notification. @@ -616,6 +619,9 @@ func (h *handshake) execute() *tcpip.Error { <-h.ep.undrain h.ep.mu.Lock() } + if n¬ifyError != 0 { + return h.ep.takeLastError() + } case wakerForNewSegment: if err := h.processSegments(); err != nil { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 83dc10ed0..0f7487963 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -1209,6 +1209,14 @@ func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } +func (e *endpoint) takeLastError() *tcpip.Error { + e.lastErrorMu.Lock() + defer e.lastErrorMu.Unlock() + err := e.lastError + e.lastError = nil + return err +} + // Read reads data from the endpoint. func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { e.LockUser() @@ -1956,11 +1964,7 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { switch o := opt.(type) { case tcpip.ErrorOption: - e.lastErrorMu.Lock() - err := e.lastError - e.lastError = nil - e.lastErrorMu.Unlock() - return err + return e.takeLastError() case *tcpip.BindToDeviceOption: e.LockUser() @@ -2546,6 +2550,18 @@ func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.C e.sndBufMu.Unlock() e.notifyProtocolGoroutine(notifyMTUChanged) + + case stack.ControlNoRoute: + e.lastErrorMu.Lock() + e.lastError = tcpip.ErrNoRoute + e.lastErrorMu.Unlock() + e.notifyProtocolGoroutine(notifyError) + + case stack.ControlNetworkUnreachable: + e.lastErrorMu.Lock() + e.lastError = tcpip.ErrNetworkUnreachable + e.lastErrorMu.Unlock() + e.notifyProtocolGoroutine(notifyError) } } diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index a14643ae8..6e692da07 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -1451,7 +1451,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk packet.tos, _ = header.IPv6(pkt.NetworkHeader).TOS() } - packet.timestamp = e.stack.NowNanoseconds() + packet.timestamp = e.stack.Clock().NowNanoseconds() e.rcvMu.Unlock() diff --git a/pkg/test/dockerutil/exec.go b/pkg/test/dockerutil/exec.go index 921d1da9e..4c739c9e9 100644 --- a/pkg/test/dockerutil/exec.go +++ b/pkg/test/dockerutil/exec.go @@ -87,7 +87,6 @@ func (c *Container) doExec(ctx context.Context, r ExecOpts, args []string) (Proc execid: resp.ID, conn: hijack, }, nil - } func (c *Container) execConfig(r ExecOpts, cmd []string) types.ExecConfig { diff --git a/runsc/boot/vfs.go b/runsc/boot/vfs.go index 56f4ba15d..9a1ed8e9e 100644 --- a/runsc/boot/vfs.go +++ b/runsc/boot/vfs.go @@ -77,6 +77,10 @@ func registerFilesystems(k *kernel.Kernel) error { AllowUserMount: true, AllowUserList: true, }) + vfsObj.MustRegisterFilesystemType(fuse.Name, &fuse.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) // Setup files in devtmpfs. if err := memdev.Register(vfsObj); err != nil { @@ -119,6 +123,7 @@ func registerFilesystems(k *kernel.Kernel) error { return fmt.Errorf("creating fusedev devtmpfs files: %w", err) } } + return nil } diff --git a/runsc/fsgofer/BUILD b/runsc/fsgofer/BUILD index 1036b0630..05e3637f7 100644 --- a/runsc/fsgofer/BUILD +++ b/runsc/fsgofer/BUILD @@ -31,5 +31,6 @@ go_test( deps = [ "//pkg/log", "//pkg/p9", + "//pkg/test/testutil", ], ) diff --git a/runsc/fsgofer/fsgofer.go b/runsc/fsgofer/fsgofer.go index b7521bda7..82a46910e 100644 --- a/runsc/fsgofer/fsgofer.go +++ b/runsc/fsgofer/fsgofer.go @@ -132,7 +132,7 @@ func (a *attachPoint) Attach() (p9.File, error) { return nil, fmt.Errorf("attach point already attached, prefix: %s", a.prefix) } - f, err := openAnyFile(a.prefix, func(mode int) (*fd.FD, error) { + f, readable, err := openAnyFile(a.prefix, func(mode int) (*fd.FD, error) { return fd.Open(a.prefix, openFlags|mode, 0) }) if err != nil { @@ -144,7 +144,7 @@ func (a *attachPoint) Attach() (p9.File, error) { return nil, fmt.Errorf("unable to stat %q: %v", a.prefix, err) } - lf, err := newLocalFile(a, f, a.prefix, stat) + lf, err := newLocalFile(a, f, a.prefix, readable, stat) if err != nil { return nil, fmt.Errorf("unable to create localFile %q: %v", a.prefix, err) } @@ -212,6 +212,10 @@ type localFile struct { // opened with. file *fd.FD + // controlReadable tells whether 'file' was opened with read permissions + // during a walk. + controlReadable bool + // mode is the mode in which the file was opened. Set to invalidMode // if localFile isn't opened. mode p9.OpenFlags @@ -251,49 +255,57 @@ func reopenProcFd(f *fd.FD, mode int) (*fd.FD, error) { return fd.New(d), nil } -func openAnyFileFromParent(parent *localFile, name string) (*fd.FD, string, error) { +func openAnyFileFromParent(parent *localFile, name string) (*fd.FD, string, bool, error) { path := path.Join(parent.hostPath, name) - f, err := openAnyFile(path, func(mode int) (*fd.FD, error) { + f, readable, err := openAnyFile(path, func(mode int) (*fd.FD, error) { return fd.OpenAt(parent.file, name, openFlags|mode, 0) }) - return f, path, err + return f, path, readable, err } // openAnyFile attempts to open the file in O_RDONLY and if it fails fallsback // to O_PATH. 'path' is used for logging messages only. 'fn' is what does the // actual file open and is customizable by the caller. -func openAnyFile(path string, fn func(mode int) (*fd.FD, error)) (*fd.FD, error) { +func openAnyFile(path string, fn func(mode int) (*fd.FD, error)) (*fd.FD, bool, error) { // Attempt to open file in the following mode in order: // 1. RDONLY | NONBLOCK: for all files, directories, ro mounts, FIFOs. // Use non-blocking to prevent getting stuck inside open(2) for // FIFOs. This option has no effect on regular files. // 2. PATH: for symlinks, sockets. - modes := []int{syscall.O_RDONLY | syscall.O_NONBLOCK, unix.O_PATH} + options := []struct { + mode int + readable bool + }{ + { + mode: syscall.O_RDONLY | syscall.O_NONBLOCK, + readable: true, + }, + { + mode: unix.O_PATH, + readable: false, + }, + } var err error - var file *fd.FD - for i, mode := range modes { - file, err = fn(mode) + for i, option := range options { + var file *fd.FD + file, err = fn(option.mode) if err == nil { - // openat succeeded, we're done. - break + // Succeeded opening the file, we're done. + return file, option.readable, nil } switch e := extractErrno(err); e { case syscall.ENOENT: // File doesn't exist, no point in retrying. - return nil, e + return nil, false, e } - // openat failed. Try again with next mode, preserving 'err' in case this - // was the last attempt. - log.Debugf("Attempt %d to open file failed, mode: %#x, path: %q, err: %v", i, openFlags|mode, path, err) + // File failed to open. Try again with next mode, preserving 'err' in case + // this was the last attempt. + log.Debugf("Attempt %d to open file failed, mode: %#x, path: %q, err: %v", i, openFlags|option.mode, path, err) } - if err != nil { - // All attempts to open file have failed, return the last error. - log.Debugf("Failed to open file, path: %q, err: %v", path, err) - return nil, extractErrno(err) - } - - return file, nil + // All attempts to open file have failed, return the last error. + log.Debugf("Failed to open file, path: %q, err: %v", path, err) + return nil, false, extractErrno(err) } func getSupportedFileType(stat syscall.Stat_t, permitSocket bool) (fileType, error) { @@ -316,18 +328,19 @@ func getSupportedFileType(stat syscall.Stat_t, permitSocket bool) (fileType, err return ft, nil } -func newLocalFile(a *attachPoint, file *fd.FD, path string, stat syscall.Stat_t) (*localFile, error) { +func newLocalFile(a *attachPoint, file *fd.FD, path string, readable bool, stat syscall.Stat_t) (*localFile, error) { ft, err := getSupportedFileType(stat, a.conf.HostUDS) if err != nil { return nil, err } return &localFile{ - attachPoint: a, - hostPath: path, - file: file, - mode: invalidMode, - ft: ft, + attachPoint: a, + hostPath: path, + file: file, + mode: invalidMode, + ft: ft, + controlReadable: readable, }, nil } @@ -380,7 +393,7 @@ func (l *localFile) Open(flags p9.OpenFlags) (*fd.FD, p9.QID, uint32, error) { // Check if control file can be used or if a new open must be created. var newFile *fd.FD - if flags == p9.ReadOnly { + if flags == p9.ReadOnly && l.controlReadable { log.Debugf("Open reusing control file, flags: %v, %q", flags, l.hostPath) newFile = l.file } else { @@ -518,7 +531,7 @@ func (l *localFile) Mkdir(name string, perm p9.FileMode, uid p9.UID, gid p9.GID) func (l *localFile) Walk(names []string) ([]p9.QID, p9.File, error) { // Duplicate current file if 'names' is empty. if len(names) == 0 { - newFile, err := openAnyFile(l.hostPath, func(mode int) (*fd.FD, error) { + newFile, readable, err := openAnyFile(l.hostPath, func(mode int) (*fd.FD, error) { return reopenProcFd(l.file, openFlags|mode) }) if err != nil { @@ -532,10 +545,11 @@ func (l *localFile) Walk(names []string) ([]p9.QID, p9.File, error) { } c := &localFile{ - attachPoint: l.attachPoint, - hostPath: l.hostPath, - file: newFile, - mode: invalidMode, + attachPoint: l.attachPoint, + hostPath: l.hostPath, + file: newFile, + mode: invalidMode, + controlReadable: readable, } return []p9.QID{l.attachPoint.makeQID(stat)}, c, nil } @@ -543,7 +557,7 @@ func (l *localFile) Walk(names []string) ([]p9.QID, p9.File, error) { var qids []p9.QID last := l for _, name := range names { - f, path, err := openAnyFileFromParent(last, name) + f, path, readable, err := openAnyFileFromParent(last, name) if last != l { last.Close() } @@ -555,7 +569,7 @@ func (l *localFile) Walk(names []string) ([]p9.QID, p9.File, error) { f.Close() return nil, nil, extractErrno(err) } - c, err := newLocalFile(last.attachPoint, f, path, stat) + c, err := newLocalFile(last.attachPoint, f, path, readable, stat) if err != nil { f.Close() return nil, nil, extractErrno(err) diff --git a/runsc/fsgofer/fsgofer_test.go b/runsc/fsgofer/fsgofer_test.go index 05af7e397..5b37e6aa1 100644 --- a/runsc/fsgofer/fsgofer_test.go +++ b/runsc/fsgofer/fsgofer_test.go @@ -26,6 +26,19 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/p9" + "gvisor.dev/gvisor/pkg/test/testutil" +) + +var allOpenFlags = []p9.OpenFlags{p9.ReadOnly, p9.WriteOnly, p9.ReadWrite} + +var ( + allTypes = []fileType{regular, directory, symlink} + + // allConfs is set in init(). + allConfs []Config + + rwConfs = []Config{{ROMount: false}} + roConfs = []Config{{ROMount: true}} ) func init() { @@ -39,6 +52,13 @@ func init() { } } +func configTestName(config *Config) string { + if config.ROMount { + return "ROMount" + } + return "RWMount" +} + func assertPanic(t *testing.T, f func()) { defer func() { if r := recover(); r == nil { @@ -88,18 +108,6 @@ func testReadWrite(f p9.File, flags p9.OpenFlags, content []byte) error { return nil } -var allOpenFlags = []p9.OpenFlags{p9.ReadOnly, p9.WriteOnly, p9.ReadWrite} - -var ( - allTypes = []fileType{regular, directory, symlink} - - // allConfs is set in init() above. - allConfs []Config - - rwConfs = []Config{{ROMount: false}} - roConfs = []Config{{ROMount: true}} -) - type state struct { root *localFile file *localFile @@ -117,42 +125,46 @@ func runAll(t *testing.T, test func(*testing.T, state)) { func runCustom(t *testing.T, types []fileType, confs []Config, test func(*testing.T, state)) { for _, c := range confs { - t.Logf("Config: %+v", c) - for _, ft := range types { - t.Logf("File type: %v", ft) + name := fmt.Sprintf("%s/%v", configTestName(&c), ft) + t.Run(name, func(t *testing.T) { + path, name, err := setup(ft) + if err != nil { + t.Fatalf("%v", err) + } + defer os.RemoveAll(path) - path, name, err := setup(ft) - if err != nil { - t.Fatalf("%v", err) - } - defer os.RemoveAll(path) + a, err := NewAttachPoint(path, c) + if err != nil { + t.Fatalf("NewAttachPoint failed: %v", err) + } + root, err := a.Attach() + if err != nil { + t.Fatalf("Attach failed, err: %v", err) + } - a, err := NewAttachPoint(path, c) - if err != nil { - t.Fatalf("NewAttachPoint failed: %v", err) - } - root, err := a.Attach() - if err != nil { - t.Fatalf("Attach failed, err: %v", err) - } + _, file, err := root.Walk([]string{name}) + if err != nil { + root.Close() + t.Fatalf("root.Walk({%q}) failed, err: %v", "symlink", err) + } - _, file, err := root.Walk([]string{name}) - if err != nil { + st := state{ + root: root.(*localFile), + file: file.(*localFile), + conf: c, + ft: ft, + } + test(t, st) + file.Close() root.Close() - t.Fatalf("root.Walk({%q}) failed, err: %v", "symlink", err) - } - - st := state{root: root.(*localFile), file: file.(*localFile), conf: c, ft: ft} - test(t, st) - file.Close() - root.Close() + }) } } } func setup(ft fileType) (string, string, error) { - path, err := ioutil.TempDir("", "root-") + path, err := ioutil.TempDir(testutil.TmpDir(), "root-") if err != nil { return "", "", fmt.Errorf("ioutil.TempDir() failed, err: %v", err) } @@ -308,6 +320,32 @@ func TestUnopened(t *testing.T) { }) } +// TestOpenOPath is a regression test to ensure that a file that cannot be open +// for read is allowed to be open. This was happening because the control file +// was open with O_PATH, but Open() was not checking for it and allowing the +// control file to be reused. +func TestOpenOPath(t *testing.T) { + runCustom(t, []fileType{regular}, rwConfs, func(t *testing.T, s state) { + // Fist remove all permissions on the file. + if err := s.file.SetAttr(p9.SetAttrMask{Permissions: true}, p9.SetAttr{Permissions: p9.FileMode(0)}); err != nil { + t.Fatalf("SetAttr(): %v", err) + } + // Then walk to the file again to open a new control file. + filename := filepath.Base(s.file.hostPath) + _, newFile, err := s.root.Walk([]string{filename}) + if err != nil { + t.Fatalf("root.Walk(%q): %v", filename, err) + } + + if newFile.(*localFile).controlReadable { + t.Fatalf("control file didn't open with O_PATH: %+v", newFile) + } + if _, _, _, err := newFile.Open(p9.ReadOnly); err != syscall.EACCES { + t.Fatalf("Open() should have failed, got: %v, wanted: EACCES", err) + } + }) +} + func SetGetAttr(l *localFile, valid p9.SetAttrMask, attr p9.SetAttr) (p9.Attr, error) { if err := l.SetAttr(valid, attr); err != nil { return p9.Attr{}, err diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index f5ac79370..f303030aa 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -263,6 +263,13 @@ func TestNATPreRedirectTCPPort(t *testing.T) { singleTest(t, NATPreRedirectTCPPort{}) } +func TestNATPreRedirectTCPOutgoing(t *testing.T) { + singleTest(t, NATPreRedirectTCPOutgoing{}) +} + +func TestNATOutRedirectTCPIncoming(t *testing.T) { + singleTest(t, NATOutRedirectTCPIncoming{}) +} func TestNATOutRedirectUDPPort(t *testing.T) { singleTest(t, NATOutRedirectUDPPort{}) } diff --git a/test/iptables/nat.go b/test/iptables/nat.go index 8562b0820..149dec2bb 100644 --- a/test/iptables/nat.go +++ b/test/iptables/nat.go @@ -28,6 +28,8 @@ const ( func init() { RegisterTestCase(NATPreRedirectUDPPort{}) RegisterTestCase(NATPreRedirectTCPPort{}) + RegisterTestCase(NATPreRedirectTCPOutgoing{}) + RegisterTestCase(NATOutRedirectTCPIncoming{}) RegisterTestCase(NATOutRedirectUDPPort{}) RegisterTestCase(NATOutRedirectTCPPort{}) RegisterTestCase(NATDropUDP{}) @@ -91,6 +93,56 @@ func (NATPreRedirectTCPPort) LocalAction(ip net.IP) error { return connectTCP(ip, dropPort, sendloopDuration) } +// NATPreRedirectTCPOutgoing verifies that outgoing TCP connections aren't +// affected by PREROUTING connection tracking. +type NATPreRedirectTCPOutgoing struct{} + +// Name implements TestCase.Name. +func (NATPreRedirectTCPOutgoing) Name() string { + return "NATPreRedirectTCPOutgoing" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATPreRedirectTCPOutgoing) ContainerAction(ip net.IP) error { + // Redirect all incoming TCP traffic to a closed port. + if err := natTable("-A", "PREROUTING", "-p", "tcp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", dropPort)); err != nil { + return err + } + + // Establish a connection to the host process. + return connectTCP(ip, acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (NATPreRedirectTCPOutgoing) LocalAction(ip net.IP) error { + return listenTCP(acceptPort, sendloopDuration) +} + +// NATOutRedirectTCPIncoming verifies that incoming TCP connections aren't +// affected by OUTPUT connection tracking. +type NATOutRedirectTCPIncoming struct{} + +// Name implements TestCase.Name. +func (NATOutRedirectTCPIncoming) Name() string { + return "NATOutRedirectTCPIncoming" +} + +// ContainerAction implements TestCase.ContainerAction. +func (NATOutRedirectTCPIncoming) ContainerAction(ip net.IP) error { + // Redirect all outgoing TCP traffic to a closed port. + if err := natTable("-A", "OUTPUT", "-p", "tcp", "-j", "REDIRECT", "--to-ports", fmt.Sprintf("%d", dropPort)); err != nil { + return err + } + + // Establish a connection to the host process. + return listenTCP(acceptPort, sendloopDuration) +} + +// LocalAction implements TestCase.LocalAction. +func (NATOutRedirectTCPIncoming) LocalAction(ip net.IP) error { + return connectTCP(ip, acceptPort, sendloopDuration) +} + // NATOutRedirectUDPPort tests that packets are redirected to different port. type NATOutRedirectUDPPort struct{} diff --git a/test/packetimpact/runner/packetimpact_test.go b/test/packetimpact/runner/packetimpact_test.go index ff5f5c7f1..1a0221893 100644 --- a/test/packetimpact/runner/packetimpact_test.go +++ b/test/packetimpact/runner/packetimpact_test.go @@ -280,11 +280,13 @@ func TestOne(t *testing.T) { } // Because the Linux kernel receives the SYN-ACK but didn't send the SYN it - // will issue a RST. To prevent this IPtables can be used to filter out all + // will issue an RST. To prevent this IPtables can be used to filter out all // incoming packets. The raw socket that packetimpact tests use will still see // everything. - if logs, err := testbench.Exec(ctx, dockerutil.ExecOpts{}, "iptables", "-A", "INPUT", "-i", testNetDev, "-j", "DROP"); err != nil { - t.Fatalf("unable to Exec iptables on container %s: %s, logs from testbench:\n%s", testbench.Name, err, logs) + for _, bin := range []string{"iptables", "ip6tables"} { + if logs, err := testbench.Exec(ctx, dockerutil.ExecOpts{}, bin, "-A", "INPUT", "-i", testNetDev, "-p", "tcp", "-j", "DROP"); err != nil { + t.Fatalf("unable to Exec %s on container %s: %s, logs from testbench:\n%s", bin, testbench.Name, err, logs) + } } // FIXME(b/156449515): Some piece of the system has a race. The old diff --git a/test/packetimpact/testbench/connections.go b/test/packetimpact/testbench/connections.go index 5d9cec73e..87ce58c24 100644 --- a/test/packetimpact/testbench/connections.go +++ b/test/packetimpact/testbench/connections.go @@ -41,7 +41,8 @@ func portFromSockaddr(sa unix.Sockaddr) (uint16, error) { return 0, fmt.Errorf("sockaddr type %T does not contain port", sa) } -// pickPort makes a new socket and returns the socket FD and port. The domain should be AF_INET or AF_INET6. The caller must close the FD when done with +// pickPort makes a new socket and returns the socket FD and port. The domain +// should be AF_INET or AF_INET6. The caller must close the FD when done with // the port if there is no error. func pickPort(domain, typ int) (fd int, port uint16, err error) { fd, err = unix.Socket(domain, typ, 0) @@ -1061,3 +1062,58 @@ func (conn *UDPIPv6) Close() { func (conn *UDPIPv6) Drain() { conn.sniffer.Drain() } + +// TCPIPv6 maintains the state for all the layers in a TCP/IPv6 connection. +type TCPIPv6 Connection + +// NewTCPIPv6 creates a new TCPIPv6 connection with reasonable defaults. +func NewTCPIPv6(t *testing.T, outgoingTCP, incomingTCP TCP) TCPIPv6 { + etherState, err := newEtherState(Ether{}, Ether{}) + if err != nil { + t.Fatalf("can't make etherState: %s", err) + } + ipv6State, err := newIPv6State(IPv6{}, IPv6{}) + if err != nil { + t.Fatalf("can't make ipv6State: %s", err) + } + tcpState, err := newTCPState(unix.AF_INET6, outgoingTCP, incomingTCP) + if err != nil { + t.Fatalf("can't make tcpState: %s", err) + } + injector, err := NewInjector(t) + if err != nil { + t.Fatalf("can't make injector: %s", err) + } + sniffer, err := NewSniffer(t) + if err != nil { + t.Fatalf("can't make sniffer: %s", err) + } + + return TCPIPv6{ + layerStates: []layerState{etherState, ipv6State, tcpState}, + injector: injector, + sniffer: sniffer, + t: t, + } +} + +func (conn *TCPIPv6) SrcPort() uint16 { + state := conn.layerStates[2].(*tcpState) + return *state.out.SrcPort +} + +// ExpectData is a convenient method that expects a Layer and the Layer after +// it. If it doens't arrive in time, it returns nil. +func (conn *TCPIPv6) ExpectData(tcp *TCP, payload *Payload, timeout time.Duration) (Layers, error) { + expected := make([]Layer, len(conn.layerStates)) + expected[len(expected)-1] = tcp + if payload != nil { + expected = append(expected, payload) + } + return (*Connection)(conn).ExpectFrame(expected, timeout) +} + +// Close frees associated resources held by the TCPIPv6 connection. +func (conn *TCPIPv6) Close() { + (*Connection)(conn).Close() +} diff --git a/test/packetimpact/testbench/layers.go b/test/packetimpact/testbench/layers.go index 645f6c1a9..24aa46cce 100644 --- a/test/packetimpact/testbench/layers.go +++ b/test/packetimpact/testbench/layers.go @@ -805,7 +805,11 @@ func (l *ICMPv6) ToBytes() ([]byte, error) { // We need to search forward to find the IPv6 header. for prev := l.Prev(); prev != nil; prev = prev.Prev() { if ipv6, ok := prev.(*IPv6); ok { - h.SetChecksum(header.ICMPv6Checksum(h, *ipv6.SrcAddr, *ipv6.DstAddr, buffer.VectorisedView{})) + payload, err := payload(l) + if err != nil { + return nil, err + } + h.SetChecksum(header.ICMPv6Checksum(h, *ipv6.SrcAddr, *ipv6.DstAddr, payload)) break } } diff --git a/test/packetimpact/tests/BUILD b/test/packetimpact/tests/BUILD index 6a07889be..27905dcff 100644 --- a/test/packetimpact/tests/BUILD +++ b/test/packetimpact/tests/BUILD @@ -220,6 +220,16 @@ packetimpact_go_test( ) packetimpact_go_test( + name = "tcp_network_unreachable", + srcs = ["tcp_network_unreachable_test.go"], + deps = [ + "//pkg/tcpip/header", + "//test/packetimpact/testbench", + "@org_golang_x_sys//unix:go_default_library", + ], +) + +packetimpact_go_test( name = "tcp_cork_mss", srcs = ["tcp_cork_mss_test.go"], deps = [ diff --git a/test/packetimpact/tests/tcp_network_unreachable_test.go b/test/packetimpact/tests/tcp_network_unreachable_test.go new file mode 100644 index 000000000..868a08da8 --- /dev/null +++ b/test/packetimpact/tests/tcp_network_unreachable_test.go @@ -0,0 +1,139 @@ +// 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 tcp_synsent_reset_test + +import ( + "context" + "flag" + "net" + "syscall" + "testing" + "time" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/test/packetimpact/testbench" +) + +func init() { + testbench.RegisterFlags(flag.CommandLine) +} + +// TestTCPSynSentUnreachable verifies that TCP connections fail immediately when +// an ICMP destination unreachable message is sent in response to the inital +// SYN. +func TestTCPSynSentUnreachable(t *testing.T) { + // Create the DUT and connection. + dut := testbench.NewDUT(t) + defer dut.TearDown() + clientFD, clientPort := dut.CreateBoundSocket(unix.SOCK_STREAM|unix.SOCK_NONBLOCK, unix.IPPROTO_TCP, net.ParseIP(testbench.RemoteIPv4)) + port := uint16(9001) + conn := testbench.NewTCPIPv4(t, testbench.TCP{SrcPort: &port, DstPort: &clientPort}, testbench.TCP{SrcPort: &clientPort, DstPort: &port}) + defer conn.Close() + + // Bring the DUT to SYN-SENT state with a non-blocking connect. + ctx, cancel := context.WithTimeout(context.Background(), testbench.RPCTimeout) + defer cancel() + sa := unix.SockaddrInet4{Port: int(port)} + copy(sa.Addr[:], net.IP(net.ParseIP(testbench.LocalIPv4)).To4()) + if _, err := dut.ConnectWithErrno(ctx, clientFD, &sa); err != syscall.Errno(unix.EINPROGRESS) { + t.Errorf("expected connect to fail with EINPROGRESS, but got %v", err) + } + + // Get the SYN. + tcpLayers, err := conn.ExpectData(&testbench.TCP{Flags: testbench.Uint8(header.TCPFlagSyn)}, nil, time.Second) + if err != nil { + t.Fatalf("expected SYN: %s", err) + } + + // Send a host unreachable message. + rawConn := (*testbench.Connection)(&conn) + layers := rawConn.CreateFrame(nil) + layers = layers[:len(layers)-1] + const ipLayer = 1 + const tcpLayer = ipLayer + 1 + ip, ok := tcpLayers[ipLayer].(*testbench.IPv4) + if !ok { + t.Fatalf("expected %s to be IPv4", tcpLayers[ipLayer]) + } + tcp, ok := tcpLayers[tcpLayer].(*testbench.TCP) + if !ok { + t.Fatalf("expected %s to be TCP", tcpLayers[tcpLayer]) + } + var icmpv4 testbench.ICMPv4 = testbench.ICMPv4{Type: testbench.ICMPv4Type(header.ICMPv4DstUnreachable), Code: testbench.Uint8(header.ICMPv4HostUnreachable)} + layers = append(layers, &icmpv4, ip, tcp) + rawConn.SendFrameStateless(layers) + + if _, err = dut.ConnectWithErrno(ctx, clientFD, &sa); err != syscall.Errno(unix.EHOSTUNREACH) { + t.Errorf("expected connect to fail with EHOSTUNREACH, but got %v", err) + } +} + +// TestTCPSynSentUnreachable6 verifies that TCP connections fail immediately when +// an ICMP destination unreachable message is sent in response to the inital +// SYN. +func TestTCPSynSentUnreachable6(t *testing.T) { + // Create the DUT and connection. + dut := testbench.NewDUT(t) + defer dut.TearDown() + clientFD, clientPort := dut.CreateBoundSocket(unix.SOCK_STREAM|unix.SOCK_NONBLOCK, unix.IPPROTO_TCP, net.ParseIP(testbench.RemoteIPv6)) + conn := testbench.NewTCPIPv6(t, testbench.TCP{DstPort: &clientPort}, testbench.TCP{SrcPort: &clientPort}) + defer conn.Close() + + // Bring the DUT to SYN-SENT state with a non-blocking connect. + ctx, cancel := context.WithTimeout(context.Background(), testbench.RPCTimeout) + defer cancel() + sa := unix.SockaddrInet6{ + Port: int(conn.SrcPort()), + ZoneId: uint32(testbench.RemoteInterfaceID), + } + copy(sa.Addr[:], net.IP(net.ParseIP(testbench.LocalIPv6)).To16()) + if _, err := dut.ConnectWithErrno(ctx, clientFD, &sa); err != syscall.Errno(unix.EINPROGRESS) { + t.Errorf("expected connect to fail with EINPROGRESS, but got %v", err) + } + + // Get the SYN. + tcpLayers, err := conn.ExpectData(&testbench.TCP{Flags: testbench.Uint8(header.TCPFlagSyn)}, nil, time.Second) + if err != nil { + t.Fatalf("expected SYN: %s", err) + } + + // Send a host unreachable message. + rawConn := (*testbench.Connection)(&conn) + layers := rawConn.CreateFrame(nil) + layers = layers[:len(layers)-1] + const ipLayer = 1 + const tcpLayer = ipLayer + 1 + ip, ok := tcpLayers[ipLayer].(*testbench.IPv6) + if !ok { + t.Fatalf("expected %s to be IPv6", tcpLayers[ipLayer]) + } + tcp, ok := tcpLayers[tcpLayer].(*testbench.TCP) + if !ok { + t.Fatalf("expected %s to be TCP", tcpLayers[tcpLayer]) + } + var icmpv6 testbench.ICMPv6 = testbench.ICMPv6{ + Type: testbench.ICMPv6Type(header.ICMPv6DstUnreachable), + Code: testbench.Uint8(header.ICMPv6NetworkUnreachable), + // Per RFC 4443 3.1, the payload contains 4 zeroed bytes. + Payload: []byte{0, 0, 0, 0}, + } + layers = append(layers, &icmpv6, ip, tcp) + rawConn.SendFrameStateless(layers) + + if _, err = dut.ConnectWithErrno(ctx, clientFD, &sa); err != syscall.Errno(unix.ENETUNREACH) { + t.Errorf("expected connect to fail with ENETUNREACH, but got %v", err) + } +} diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index c06a75ada..c19b30b4a 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -400,6 +400,7 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:open_test", + vfs2 = "True", ) syscall_test( @@ -640,11 +641,13 @@ syscall_test( syscall_test( add_overlay = True, test = "//test/syscalls/linux:sendfile_socket_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:sendfile_test", + vfs2 = "True", ) syscall_test( diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 662d780d8..66a31cd28 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -943,6 +943,7 @@ cc_binary( "//test/util:eventfd_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/container:node_hash_set", "@com_google_absl//absl/strings", gtest, "//test/util:posix_error", diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 6fa16208e..1d0d584cd 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -161,6 +161,19 @@ TEST(DevTest, OpenDevFuse) { ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); } +TEST(DevTest, ReadDevFuseWithoutMount) { + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); + + std::vector<char> buf(1); + EXPECT_THAT(ReadFd(fd.get(), buf.data(), sizeof(buf)), + SyscallFailsWithErrno(EPERM)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/futex.cc b/test/syscalls/linux/futex.cc index 40c80a6e1..90b1f0508 100644 --- a/test/syscalls/linux/futex.cc +++ b/test/syscalls/linux/futex.cc @@ -18,6 +18,7 @@ #include <sys/syscall.h> #include <sys/time.h> #include <sys/types.h> +#include <syscall.h> #include <unistd.h> #include <algorithm> @@ -737,6 +738,97 @@ TEST_P(PrivateAndSharedFutexTest, PITryLockConcurrency_NoRandomSave) { } } +int get_robust_list(int pid, struct robust_list_head** head_ptr, + size_t* len_ptr) { + return syscall(__NR_get_robust_list, pid, head_ptr, len_ptr); +} + +int set_robust_list(struct robust_list_head* head, size_t len) { + return syscall(__NR_set_robust_list, head, len); +} + +TEST(RobustFutexTest, BasicSetGet) { + struct robust_list_head hd = {}; + struct robust_list_head* hd_ptr = &hd; + + // Set! + EXPECT_THAT(set_robust_list(hd_ptr, sizeof(hd)), SyscallSucceedsWithValue(0)); + + // Get! + struct robust_list_head* new_hd_ptr = hd_ptr; + size_t len; + EXPECT_THAT(get_robust_list(0, &new_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(new_hd_ptr, hd_ptr); + EXPECT_EQ(len, sizeof(hd)); +} + +TEST(RobustFutexTest, GetFromOtherTid) { + // Get the current tid and list head. + pid_t tid = gettid(); + struct robust_list_head* hd_ptr = {}; + size_t len; + EXPECT_THAT(get_robust_list(0, &hd_ptr, &len), SyscallSucceedsWithValue(0)); + + // Create a new thread. + ScopedThread t([&] { + // Current tid list head should be different from parent tid. + struct robust_list_head* got_hd_ptr = {}; + EXPECT_THAT(get_robust_list(0, &got_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_NE(hd_ptr, got_hd_ptr); + + // Get the parent list head by passing its tid. + EXPECT_THAT(get_robust_list(tid, &got_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(hd_ptr, got_hd_ptr); + }); + + // Wait for thread. + t.Join(); +} + +TEST(RobustFutexTest, InvalidSize) { + struct robust_list_head* hd = {}; + EXPECT_THAT(set_robust_list(hd, sizeof(*hd) + 1), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(RobustFutexTest, PthreadMutexAttr) { + constexpr int kNumMutexes = 3; + + // Create a bunch of robust mutexes. + pthread_mutexattr_t attrs[kNumMutexes]; + pthread_mutex_t mtxs[kNumMutexes]; + for (int i = 0; i < kNumMutexes; i++) { + TEST_PCHECK(pthread_mutexattr_init(&attrs[i]) == 0); + TEST_PCHECK(pthread_mutexattr_setrobust(&attrs[i], PTHREAD_MUTEX_ROBUST) == + 0); + TEST_PCHECK(pthread_mutex_init(&mtxs[i], &attrs[i]) == 0); + } + + // Start thread to lock the mutexes and then exit. + ScopedThread t([&] { + for (int i = 0; i < kNumMutexes; i++) { + TEST_PCHECK(pthread_mutex_lock(&mtxs[i]) == 0); + } + pthread_exit(NULL); + }); + + // Wait for thread. + t.Join(); + + // Now try to take the mutexes. + for (int i = 0; i < kNumMutexes; i++) { + // Should get EOWNERDEAD. + EXPECT_EQ(pthread_mutex_lock(&mtxs[i]), EOWNERDEAD); + // Make the mutex consistent. + EXPECT_EQ(pthread_mutex_consistent(&mtxs[i]), 0); + // Unlock. + EXPECT_EQ(pthread_mutex_unlock(&mtxs[i]), 0); + } +} + } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/getdents.cc b/test/syscalls/linux/getdents.cc index b147d6181..b040cdcf7 100644 --- a/test/syscalls/linux/getdents.cc +++ b/test/syscalls/linux/getdents.cc @@ -32,6 +32,7 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/container/node_hash_set.h" #include "absl/strings/numbers.h" #include "absl/strings/str_cat.h" #include "test/util/eventfd_util.h" @@ -393,7 +394,7 @@ TYPED_TEST(GetdentsTest, ProcSelfFd) { // Make the buffer very small since we want to iterate. typename TestFixture::DirentBufferType dirents( 2 * sizeof(typename TestFixture::LinuxDirentType)); - std::unordered_set<int> prev_fds; + absl::node_hash_set<int> prev_fds; while (true) { dirents.Reset(); int rv; diff --git a/test/syscalls/linux/mount.cc b/test/syscalls/linux/mount.cc index a3e9745cf..7664fa73d 100644 --- a/test/syscalls/linux/mount.cc +++ b/test/syscalls/linux/mount.cc @@ -321,6 +321,34 @@ TEST(MountTest, RenameRemoveMountPoint) { ASSERT_THAT(rmdir(dir.path().c_str()), SyscallFailsWithErrno(EBUSY)); } +TEST(MountTest, MountFuseFilesystemNoDevice) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(mount("", dir.path().c_str(), "fuse", 0, ""), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(MountTest, MountFuseFilesystem) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_WRONLY)); + std::string mopts = "fd=" + std::to_string(fd.get()); + + auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + auto const mount = + ASSERT_NO_ERRNO_AND_VALUE(Mount("", dir.path(), "fuse", 0, mopts, 0)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/raw_socket.cc b/test/syscalls/linux/raw_socket.cc index ce54dc064..8d6e5c913 100644 --- a/test/syscalls/linux/raw_socket.cc +++ b/test/syscalls/linux/raw_socket.cc @@ -262,6 +262,27 @@ TEST_P(RawSocketTest, SendWithoutConnectFails) { SyscallFailsWithErrno(EDESTADDRREQ)); } +// Wildcard Bind. +TEST_P(RawSocketTest, BindToWildcard) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + struct sockaddr_storage addr; + addr = {}; + + // We don't set ports because raw sockets don't have a notion of ports. + if (Family() == AF_INET) { + struct sockaddr_in* sin = reinterpret_cast<struct sockaddr_in*>(&addr); + sin->sin_family = AF_INET; + sin->sin_addr.s_addr = htonl(INADDR_ANY); + } else { + struct sockaddr_in6* sin6 = reinterpret_cast<struct sockaddr_in6*>(&addr); + sin6->sin6_family = AF_INET6; + sin6->sin6_addr = in6addr_any; + } + + ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); +} + // Bind to localhost. TEST_P(RawSocketTest, BindToLocalhost) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); diff --git a/tools/go_generics/defs.bzl b/tools/go_generics/defs.bzl index 7fe7d7e59..33329cf28 100644 --- a/tools/go_generics/defs.bzl +++ b/tools/go_generics/defs.bzl @@ -101,20 +101,20 @@ def _go_template_instance_impl(ctx): # Build the argument list. args = ["-i=%s" % template.file.path, "-o=%s" % output.path] if ctx.attr.package: - args += ["-p=%s" % ctx.attr.package] + args.append("-p=%s" % ctx.attr.package) if len(ctx.attr.prefix) > 0: - args += ["-prefix=%s" % ctx.attr.prefix] + args.append("-prefix=%s" % ctx.attr.prefix) if len(ctx.attr.suffix) > 0: - args += ["-suffix=%s" % ctx.attr.suffix] + args.append("-suffix=%s" % ctx.attr.suffix) args += [("-t=%s=%s" % (p[0], p[1])) for p in ctx.attr.types.items()] args += [("-c=%s=%s" % (p[0], p[1])) for p in ctx.attr.consts.items()] args += [("-import=%s=%s" % (p[0], p[1])) for p in ctx.attr.imports.items()] if ctx.attr.anon: - args += ["-anon"] + args.append("-anon") ctx.actions.run( inputs = [template.file], diff --git a/tools/go_generics/tests/all_stmts/BUILD b/tools/go_generics/tests/all_stmts/BUILD index fbc07dbff..a4a7c775a 100644 --- a/tools/go_generics/tests/all_stmts/BUILD +++ b/tools/go_generics/tests/all_stmts/BUILD @@ -8,3 +8,9 @@ go_generics_test( "T": "Q", }, ) + +# @unused +glaze_ignore = [ + "input.go", + "output.go", +] diff --git a/tools/go_generics/tests/all_types/BUILD b/tools/go_generics/tests/all_types/BUILD index 9144b0811..60b1fd314 100644 --- a/tools/go_generics/tests/all_types/BUILD +++ b/tools/go_generics/tests/all_types/BUILD @@ -8,3 +8,9 @@ go_generics_test( "T": "Q", }, ) + +# @unused +glaze_ignore = [ + "input.go", + "output.go", +] diff --git a/tools/go_generics/tests/anon/BUILD b/tools/go_generics/tests/anon/BUILD index 1890f2817..ef24f4b25 100644 --- a/tools/go_generics/tests/anon/BUILD +++ b/tools/go_generics/tests/anon/BUILD @@ -10,3 +10,9 @@ go_generics_test( "T": "Q", }, ) + +# @unused +glaze_ignore = [ + "input.go", + "output.go", +] diff --git a/tools/go_generics/tests/consts/BUILD b/tools/go_generics/tests/consts/BUILD index 1e3574593..fd7caccad 100644 --- a/tools/go_generics/tests/consts/BUILD +++ b/tools/go_generics/tests/consts/BUILD @@ -15,3 +15,9 @@ go_generics_test( inputs = ["input.go"], output = "output.go", ) + +# @unused +glaze_ignore = [ + "input.go", + "output.go", +] diff --git a/tools/go_generics/tests/defs.bzl b/tools/go_generics/tests/defs.bzl index 038629852..6277c3947 100644 --- a/tools/go_generics/tests/defs.bzl +++ b/tools/go_generics/tests/defs.bzl @@ -31,17 +31,17 @@ _go_generics_test = rule( test = True, ) -""" -Instantiates a generics test. - -Args: - name: the name of the test. - inputs: all the input files. - output: the output files. - opts: the template options. -""" - def go_generics_test(name, inputs, output, types = None, consts = None, **kwargs): + """Instantiates a generics test. + + Args: + name: the name of the test. + inputs: all the input files. + output: the output files. + types: the template types (dictionary). + consts: the template consts (dictionary). + **kwargs: additional arguments for the template_instance. + """ if types == None: types = dict() if consts == None: diff --git a/tools/go_generics/tests/imports/BUILD b/tools/go_generics/tests/imports/BUILD index c4424ee9c..a86223d41 100644 --- a/tools/go_generics/tests/imports/BUILD +++ b/tools/go_generics/tests/imports/BUILD @@ -16,3 +16,9 @@ go_generics_test( "T": "sync.Mutex", }, ) + +# @unused +glaze_ignore = [ + "input.go", + "output.go", +] diff --git a/tools/go_generics/tests/remove_typedef/BUILD b/tools/go_generics/tests/remove_typedef/BUILD index 84f6ac1ab..46457cec6 100644 --- a/tools/go_generics/tests/remove_typedef/BUILD +++ b/tools/go_generics/tests/remove_typedef/BUILD @@ -8,3 +8,9 @@ go_generics_test( "T": "U", }, ) + +# @unused +glaze_ignore = [ + "input.go", + "output.go", +] diff --git a/tools/go_generics/tests/simple/BUILD b/tools/go_generics/tests/simple/BUILD index 204952b90..4b9265ea4 100644 --- a/tools/go_generics/tests/simple/BUILD +++ b/tools/go_generics/tests/simple/BUILD @@ -9,3 +9,9 @@ go_generics_test( "T": "Q", }, ) + +# @unused +glaze_ignore = [ + "input.go", + "output.go", +] diff --git a/tools/go_marshal/README.md b/tools/go_marshal/README.md index 4886efddf..68d759083 100644 --- a/tools/go_marshal/README.md +++ b/tools/go_marshal/README.md @@ -9,11 +9,9 @@ automatically generating code to marshal go data structures to memory. `binary.Marshal` by moving the go runtime reflection necessary to marshal a struct to compile-time. -`go_marshal` automatically generates implementations for `abi.Marshallable` and -`safemem.{Reader,Writer}`. Call-sites for serialization (typically syscall -implementations) can directly invoke `safemem.Reader.ReadToBlocks` and -`safemem.Writer.WriteFromBlocks`. Data structures that require custom -serialization will have manual implementations for these interfaces. +`go_marshal` automatically generates implementations for `marshal.Marshallable` +and `safemem.{Reader,Writer}`. Data structures that require custom serialization +will have manual implementations for these interfaces. Data structures can be flagged for code generation by adding a struct-level comment `// +marshal`. diff --git a/tools/go_marshal/gomarshal/generator.go b/tools/go_marshal/gomarshal/generator.go index 177013dbb..19bcd4e6a 100644 --- a/tools/go_marshal/gomarshal/generator.go +++ b/tools/go_marshal/gomarshal/generator.go @@ -413,13 +413,13 @@ func (g *Generator) Run() error { for _, t := range g.collectMarshallableTypes(a, fsets[i]) { impl := g.generateOne(t, fsets[i]) // Collect Marshallable types referenced by the generated code. - for ref, _ := range impl.ms { + for ref := range impl.ms { ms[ref] = struct{}{} } impls = append(impls, impl) // Collect imports referenced by the generated code and add them to // the list of imports we need to copy to the generated code. - for name, _ := range impl.is { + for name := range impl.is { if !g.imports.markUsed(name) { panic(fmt.Sprintf("Generated code for '%s' referenced a non-existent import with local name '%s'. Either go-marshal needs to add an import to the generated file, or a package in an input source file has a package name differ from the final component of its path, which go-marshal doesn't know how to detect; use an import alias to work around this limitation.", impl.typeName(), name)) } diff --git a/tools/go_marshal/gomarshal/generator_interfaces_struct.go b/tools/go_marshal/gomarshal/generator_interfaces_struct.go index 9cd3c9579..4b9cea08a 100644 --- a/tools/go_marshal/gomarshal/generator_interfaces_struct.go +++ b/tools/go_marshal/gomarshal/generator_interfaces_struct.go @@ -268,6 +268,10 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { g.emit("// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe.\n") g.emit("func (%s *%s) MarshalUnsafe(dst []byte) {\n", g.r, g.typeName()) g.inIndent(func() { + fallback := func() { + g.emit("// Type %s doesn't have a packed layout in memory, fallback to MarshalBytes.\n", g.typeName()) + g.emit("%s.MarshalBytes(dst)\n", g.r) + } if thisPacked { g.recordUsedImport("safecopy") g.recordUsedImport("unsafe") @@ -277,16 +281,13 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { g.emit("safecopy.CopyIn(dst, unsafe.Pointer(%s))\n", g.r) }) g.emit("} else {\n") - g.inIndent(func() { - g.emit("%s.MarshalBytes(dst)\n", g.r) - }) + g.inIndent(fallback) g.emit("}\n") } else { g.emit("safecopy.CopyIn(dst, unsafe.Pointer(%s))\n", g.r) } } else { - g.emit("// Type %s doesn't have a packed layout in memory, fallback to MarshalBytes.\n", g.typeName()) - g.emit("%s.MarshalBytes(dst)\n", g.r) + fallback() } }) g.emit("}\n\n") @@ -294,6 +295,10 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { g.emit("// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe.\n") g.emit("func (%s *%s) UnmarshalUnsafe(src []byte) {\n", g.r, g.typeName()) g.inIndent(func() { + fallback := func() { + g.emit("// Type %s doesn't have a packed layout in memory, fallback to UnmarshalBytes.\n", g.typeName()) + g.emit("%s.UnmarshalBytes(src)\n", g.r) + } if thisPacked { g.recordUsedImport("safecopy") g.recordUsedImport("unsafe") @@ -303,16 +308,13 @@ func (g *interfaceGenerator) emitMarshallableForStruct(st *ast.StructType) { g.emit("safecopy.CopyOut(unsafe.Pointer(%s), src)\n", g.r) }) g.emit("} else {\n") - g.inIndent(func() { - g.emit("%s.UnmarshalBytes(src)\n", g.r) - }) + g.inIndent(fallback) g.emit("}\n") } else { g.emit("safecopy.CopyOut(unsafe.Pointer(%s), src)\n", g.r) } } else { - g.emit("// Type %s doesn't have a packed layout in memory, fall back to UnmarshalBytes.\n", g.typeName()) - g.emit("%s.UnmarshalBytes(src)\n", g.r) + fallback() } }) g.emit("}\n\n") @@ -463,8 +465,10 @@ func (g *interfaceGenerator) emitMarshallableSliceForStruct(st *ast.StructType, }) g.emit("}\n\n") - g.emit("// Handle any final partial object.\n") - g.emit("if length < size*count && length%size != 0 {\n") + g.emit("// Handle any final partial object. buf is guaranteed to be long enough for the\n") + g.emit("// final element, but may not contain valid data for the entire range. This may\n") + g.emit("// result in unmarshalling zero values for some parts of the object.\n") + g.emit("if length%size != 0 {\n") g.inIndent(func() { g.emit("idx := limit\n") g.emit("dst[idx].UnmarshalBytes(buf[size*idx:size*(idx+1)])\n") diff --git a/tools/go_marshal/marshal/marshal.go b/tools/go_marshal/marshal/marshal.go index cb2166252..85b196f08 100644 --- a/tools/go_marshal/marshal/marshal.go +++ b/tools/go_marshal/marshal/marshal.go @@ -58,18 +58,12 @@ type Marshallable interface { // likely make use of the type of these fields). SizeBytes() int - // MarshalBytes serializes a copy of a type to dst. dst may be smaller than - // SizeBytes(), which results in a part of the struct being marshalled. Note - // that this may have unexpected results for non-packed types, as implicit - // padding needs to be taken into account when reasoning about how much of - // the type is serialized. + // MarshalBytes serializes a copy of a type to dst. + // Precondition: dst must be at least SizeBytes() in length. MarshalBytes(dst []byte) - // UnmarshalBytes deserializes a type from src. src may be smaller than - // SizeBytes(), which results in a partially deserialized struct. Note that - // this may have unexpected results for non-packed types, as implicit - // padding needs to be taken into account when reasoning about how much of - // the type is deserialized. + // UnmarshalBytes deserializes a type from src. + // Precondition: src must be at least SizeBytes() in length. UnmarshalBytes(src []byte) // Packed returns true if the marshalled size of the type is the same as the @@ -89,8 +83,8 @@ type Marshallable interface { // representation to the dst buffer. This is only safe to do when the type // has no implicit padding, see Marshallable.Packed. When Packed would // return false, MarshalUnsafe should fall back to the safer but slower - // MarshalBytes. dst may be smaller than SizeBytes(), see comment for - // MarshalBytes for implications. + // MarshalBytes. + // Precondition: dst must be at least SizeBytes() in length. MarshalUnsafe(dst []byte) // UnmarshalUnsafe deserializes a type by directly copying to the underlying @@ -99,8 +93,8 @@ type Marshallable interface { // This allows much faster unmarshalling of types which have no implicit // padding, see Marshallable.Packed. When Packed would return false, // UnmarshalUnsafe should fall back to the safer but slower unmarshal - // mechanism implemented in UnmarshalBytes. src may be smaller than - // SizeBytes(), see comment for UnmarshalBytes for implications. + // mechanism implemented in UnmarshalBytes. + // Precondition: src must be at least SizeBytes() in length. UnmarshalUnsafe(src []byte) // CopyIn deserializes a Marshallable type from a task's memory. This may @@ -149,14 +143,16 @@ type Marshallable interface { // // Generates four additional functions for marshalling slices of Foos like this: // -// // MarshalUnsafeFooSlice is like Foo.MarshalUnsafe, buf for a []Foo. It's -// // more efficient that repeatedly calling calling Foo.MarshalUnsafe over a -// // []Foo in a loop. +// // MarshalUnsafeFooSlice is like Foo.MarshalUnsafe, buf for a []Foo. It +// // might be more efficient that repeatedly calling Foo.MarshalUnsafe +// // over a []Foo in a loop if the type is Packed. +// // Preconditions: dst must be at least len(src)*Foo.SizeBytes() in length. // func MarshalUnsafeFooSlice(src []Foo, dst []byte) (int, error) { ... } // -// // UnmarshalUnsafeFooSlice is like Foo.UnmarshalUnsafe, buf for a []Foo. It's -// // more efficient that repeatedly calling calling Foo.UnmarshalUnsafe over a -// // []Foo in a loop. +// // UnmarshalUnsafeFooSlice is like Foo.UnmarshalUnsafe, buf for a []Foo. It +// // might be more efficient that repeatedly calling Foo.UnmarshalUnsafe +// // over a []Foo in a loop if the type is Packed. +// // Preconditions: src must be at least len(dst)*Foo.SizeBytes() in length. // func UnmarshalUnsafeFooSlice(dst []Foo, src []byte) (int, error) { ... } // // // CopyFooSliceIn copies in a slice of Foo objects from the task's memory. diff --git a/tools/go_marshal/primitive/primitive.go b/tools/go_marshal/primitive/primitive.go index ebcf130ae..d93edda8b 100644 --- a/tools/go_marshal/primitive/primitive.go +++ b/tools/go_marshal/primitive/primitive.go @@ -17,10 +17,22 @@ package primitive import ( + "io" + "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/tools/go_marshal/marshal" ) +// Int8 is a marshal.Marshallable implementation for int8. +// +// +marshal slice:Int8Slice:inner +type Int8 int8 + +// Uint8 is a marshal.Marshallable implementation for uint8. +// +// +marshal slice:Uint8Slice:inner +type Uint8 uint8 + // Int16 is a marshal.Marshallable implementation for int16. // // +marshal slice:Int16Slice:inner @@ -51,6 +63,66 @@ type Int64 int64 // +marshal slice:Uint64Slice:inner type Uint64 uint64 +// ByteSlice is a marshal.Marshallable implementation for []byte. +// This is a convenience wrapper around a dynamically sized type, and can't be +// embedded in other marshallable types because it breaks assumptions made by +// go-marshal internals. It violates the "no dynamically-sized types" +// constraint of the go-marshal library. +type ByteSlice []byte + +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (b *ByteSlice) SizeBytes() int { + return len(*b) +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (b *ByteSlice) MarshalBytes(dst []byte) { + copy(dst, *b) +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (b *ByteSlice) UnmarshalBytes(src []byte) { + copy(*b, src) +} + +// Packed implements marshal.Marshallable.Packed. +func (b *ByteSlice) Packed() bool { + return false +} + +// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. +func (b *ByteSlice) MarshalUnsafe(dst []byte) { + b.MarshalBytes(dst) +} + +// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. +func (b *ByteSlice) UnmarshalUnsafe(src []byte) { + b.UnmarshalBytes(src) +} + +// CopyIn implements marshal.Marshallable.CopyIn. +func (b *ByteSlice) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { + return task.CopyInBytes(addr, *b) +} + +// CopyOut implements marshal.Marshallable.CopyOut. +func (b *ByteSlice) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) { + return task.CopyOutBytes(addr, *b) +} + +// CopyOutN implements marshal.Marshallable.CopyOutN. +func (b *ByteSlice) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) { + return task.CopyOutBytes(addr, (*b)[:limit]) +} + +// WriteTo implements io.WriterTo.WriteTo. +func (b *ByteSlice) WriteTo(w io.Writer) (int64, error) { + n, err := w.Write(*b) + return int64(n), err +} + +var _ marshal.Marshallable = (*ByteSlice)(nil) + // Below, we define some convenience functions for marshalling primitive types // using the newtypes above, without requiring superfluous casts. diff --git a/tools/vm/ubuntu1604/30_docker.sh b/tools/vm/ubuntu1604/30_docker.sh index 332a03dfb..d393133e4 100755 --- a/tools/vm/ubuntu1604/30_docker.sh +++ b/tools/vm/ubuntu1604/30_docker.sh @@ -54,8 +54,11 @@ while true; do done # Enable experimental features, for cross-building aarch64 images. +# Enable Docker IPv6. cat > /etc/docker/daemon.json <<EOF { - "experimental": true + "experimental": true, + "fixed-cidr-v6": "2001:db8:1::/64", + "ipv6": true } EOF |