diff options
Diffstat (limited to 'pkg')
229 files changed, 9363 insertions, 4279 deletions
diff --git a/pkg/abi/linux/fcntl.go b/pkg/abi/linux/fcntl.go index cc3571fad..d1ca56370 100644 --- a/pkg/abi/linux/fcntl.go +++ b/pkg/abi/linux/fcntl.go @@ -25,6 +25,8 @@ const ( F_SETLKW = 7 F_SETOWN = 8 F_GETOWN = 9 + F_SETSIG = 10 + F_GETSIG = 11 F_SETOWN_EX = 15 F_GETOWN_EX = 16 F_DUPFD_CLOEXEC = 1024 + 6 diff --git a/pkg/abi/linux/sem.go b/pkg/abi/linux/sem.go index 1b2f76c0b..0adff8dff 100644 --- a/pkg/abi/linux/sem.go +++ b/pkg/abi/linux/sem.go @@ -32,6 +32,23 @@ const ( SEM_STAT_ANY = 20 ) +// Information about system-wide sempahore limits and parameters. +// +// Source: include/uapi/linux/sem.h +const ( + SEMMNI = 32000 + SEMMSL = 32000 + SEMMNS = SEMMNI * SEMMSL + SEMOPM = 500 + SEMVMX = 32767 + SEMAEM = SEMVMX + + // followings are unused in kernel + SEMUME = SEMOPM + SEMMNU = SEMMNS + SEMMAP = SEMMNS +) + const SEM_UNDO = 0x1000 // Sembuf is equivalent to struct sembuf. @@ -42,3 +59,21 @@ type Sembuf struct { SemOp int16 SemFlg int16 } + +// SemInfo is equivalent to struct seminfo. +// +// Source: include/uapi/linux/sem.h +// +// +marshal +type SemInfo struct { + SemMap uint32 + SemMni uint32 + SemMns uint32 + SemMnu uint32 + SemMsl uint32 + SemOpm uint32 + SemUme uint32 + SemUsz uint32 + SemVmx uint32 + SemAem uint32 +} diff --git a/pkg/crypto/BUILD b/pkg/crypto/BUILD new file mode 100644 index 000000000..08fa772ca --- /dev/null +++ b/pkg/crypto/BUILD @@ -0,0 +1,12 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "crypto", + srcs = [ + "crypto.go", + "crypto_stdlib.go", + ], + visibility = ["//:sandbox"], +) diff --git a/pkg/sleep/empty.s b/pkg/crypto/crypto.go index fb37360ac..b26b55d37 100644 --- a/pkg/sleep/empty.s +++ b/pkg/crypto/crypto.go @@ -1,4 +1,4 @@ -// Copyright 2018 The gVisor Authors. +// 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. @@ -12,4 +12,5 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Empty assembly file so empty func definitions work. +// Package crypto wraps crypto primitives. +package crypto diff --git a/pkg/syncevent/waiter_asm_unsafe.go b/pkg/crypto/crypto_stdlib.go index 19d6b0b15..74a55a123 100644 --- a/pkg/syncevent/waiter_asm_unsafe.go +++ b/pkg/crypto/crypto_stdlib.go @@ -12,13 +12,21 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build amd64 arm64 - -package syncevent +package crypto import ( - "unsafe" + "crypto/ecdsa" + "crypto/sha512" + "math/big" ) -// See waiter_noasm_unsafe.go for a description of waiterUnlock. -func waiterUnlock(ptr unsafe.Pointer, wg *unsafe.Pointer) bool +// EcdsaVerify verifies the signature in r, s of hash using ECDSA and the +// public key, pub. Its return value records whether the signature is valid. +func EcdsaVerify(pub *ecdsa.PublicKey, hash []byte, r, s *big.Int) bool { + return ecdsa.Verify(pub, hash, r, s) +} + +// SumSha384 returns the SHA384 checksum of the data. +func SumSha384(data []byte) (sum384 [sha512.Size384]byte) { + return sha512.Sum384(data) +} diff --git a/pkg/flipcall/ctrl_futex.go b/pkg/flipcall/ctrl_futex.go index e7c3a3a0b..2e8452a02 100644 --- a/pkg/flipcall/ctrl_futex.go +++ b/pkg/flipcall/ctrl_futex.go @@ -40,17 +40,41 @@ func (ep *Endpoint) ctrlInit(opts ...EndpointOption) error { return nil } -type ctrlHandshakeRequest struct{} - -type ctrlHandshakeResponse struct{} - func (ep *Endpoint) ctrlConnect() error { if err := ep.enterFutexWait(); err != nil { return err } - _, err := ep.futexConnect(&ctrlHandshakeRequest{}) - ep.exitFutexWait() - return err + defer ep.exitFutexWait() + + // Write the connection request. + w := ep.NewWriter() + if err := json.NewEncoder(w).Encode(struct{}{}); err != nil { + return fmt.Errorf("error writing connection request: %v", err) + } + *ep.dataLen() = w.Len() + + // Exchange control with the server. + if err := ep.futexSetPeerActive(); err != nil { + return err + } + if err := ep.futexWakePeer(); err != nil { + return err + } + if err := ep.futexWaitUntilActive(); err != nil { + return err + } + + // Read the connection response. + var resp struct{} + respLen := atomic.LoadUint32(ep.dataLen()) + if respLen > ep.dataCap { + return fmt.Errorf("invalid connection response length %d (maximum %d)", respLen, ep.dataCap) + } + if err := json.NewDecoder(ep.NewReader(respLen)).Decode(&resp); err != nil { + return fmt.Errorf("error reading connection response: %v", err) + } + + return nil } func (ep *Endpoint) ctrlWaitFirst() error { @@ -59,52 +83,61 @@ func (ep *Endpoint) ctrlWaitFirst() error { } defer ep.exitFutexWait() - // Wait for the handshake request. - if err := ep.futexSwitchFromPeer(); err != nil { + // Wait for the connection request. + if err := ep.futexWaitUntilActive(); err != nil { return err } - // Read the handshake request. + // Read the connection request. reqLen := atomic.LoadUint32(ep.dataLen()) if reqLen > ep.dataCap { - return fmt.Errorf("invalid handshake request length %d (maximum %d)", reqLen, ep.dataCap) + return fmt.Errorf("invalid connection request length %d (maximum %d)", reqLen, ep.dataCap) } - var req ctrlHandshakeRequest + var req struct{} if err := json.NewDecoder(ep.NewReader(reqLen)).Decode(&req); err != nil { - return fmt.Errorf("error reading handshake request: %v", err) + return fmt.Errorf("error reading connection request: %v", err) } - // Write the handshake response. + // Write the connection response. w := ep.NewWriter() - if err := json.NewEncoder(w).Encode(ctrlHandshakeResponse{}); err != nil { - return fmt.Errorf("error writing handshake response: %v", err) + if err := json.NewEncoder(w).Encode(struct{}{}); err != nil { + return fmt.Errorf("error writing connection response: %v", err) } *ep.dataLen() = w.Len() // Return control to the client. raceBecomeInactive() - if err := ep.futexSwitchToPeer(); err != nil { + if err := ep.futexSetPeerActive(); err != nil { + return err + } + if err := ep.futexWakePeer(); err != nil { return err } - // Wait for the first non-handshake message. - return ep.futexSwitchFromPeer() + // Wait for the first non-connection message. + return ep.futexWaitUntilActive() } func (ep *Endpoint) ctrlRoundTrip() error { - if err := ep.futexSwitchToPeer(); err != nil { + if err := ep.enterFutexWait(); err != nil { return err } - if err := ep.enterFutexWait(); err != nil { + defer ep.exitFutexWait() + + if err := ep.futexSetPeerActive(); err != nil { return err } - err := ep.futexSwitchFromPeer() - ep.exitFutexWait() - return err + if err := ep.futexWakePeer(); err != nil { + return err + } + return ep.futexWaitUntilActive() } func (ep *Endpoint) ctrlWakeLast() error { - return ep.futexSwitchToPeer() + if err := ep.futexSetPeerActive(); err != nil { + return err + } + return ep.futexWakePeer() } func (ep *Endpoint) enterFutexWait() error { diff --git a/pkg/flipcall/flipcall_unsafe.go b/pkg/flipcall/flipcall_unsafe.go index ac974b232..580bf23a4 100644 --- a/pkg/flipcall/flipcall_unsafe.go +++ b/pkg/flipcall/flipcall_unsafe.go @@ -41,11 +41,11 @@ const ( ) func (ep *Endpoint) connState() *uint32 { - return (*uint32)((unsafe.Pointer)(ep.packet)) + return (*uint32)(unsafe.Pointer(ep.packet)) } func (ep *Endpoint) dataLen() *uint32 { - return (*uint32)((unsafe.Pointer)(ep.packet + 4)) + return (*uint32)(unsafe.Pointer(ep.packet + 4)) } // Data returns the datagram part of ep's packet window as a byte slice. @@ -63,7 +63,7 @@ func (ep *Endpoint) dataLen() *uint32 { // all. func (ep *Endpoint) Data() []byte { var bs []byte - bsReflect := (*reflect.SliceHeader)((unsafe.Pointer)(&bs)) + bsReflect := (*reflect.SliceHeader)(unsafe.Pointer(&bs)) bsReflect.Data = ep.packet + PacketHeaderBytes bsReflect.Len = int(ep.dataCap) bsReflect.Cap = int(ep.dataCap) @@ -76,12 +76,12 @@ var ioSync int64 func raceBecomeActive() { if sync.RaceEnabled { - sync.RaceAcquire((unsafe.Pointer)(&ioSync)) + sync.RaceAcquire(unsafe.Pointer(&ioSync)) } } func raceBecomeInactive() { if sync.RaceEnabled { - sync.RaceReleaseMerge((unsafe.Pointer)(&ioSync)) + sync.RaceReleaseMerge(unsafe.Pointer(&ioSync)) } } diff --git a/pkg/flipcall/futex_linux.go b/pkg/flipcall/futex_linux.go index 168c1ccff..0e559ee16 100644 --- a/pkg/flipcall/futex_linux.go +++ b/pkg/flipcall/futex_linux.go @@ -17,7 +17,6 @@ package flipcall import ( - "encoding/json" "fmt" "runtime" "sync/atomic" @@ -26,55 +25,26 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" ) -func (ep *Endpoint) futexConnect(req *ctrlHandshakeRequest) (ctrlHandshakeResponse, error) { - var resp ctrlHandshakeResponse - - // Write the handshake request. - w := ep.NewWriter() - if err := json.NewEncoder(w).Encode(req); err != nil { - return resp, fmt.Errorf("error writing handshake request: %v", err) - } - *ep.dataLen() = w.Len() - - // Exchange control with the server. - if err := ep.futexSwitchToPeer(); err != nil { - return resp, err +func (ep *Endpoint) futexSetPeerActive() error { + if atomic.CompareAndSwapUint32(ep.connState(), ep.activeState, ep.inactiveState) { + return nil } - if err := ep.futexSwitchFromPeer(); err != nil { - return resp, err + switch cs := atomic.LoadUint32(ep.connState()); cs { + case csShutdown: + return ShutdownError{} + default: + return fmt.Errorf("unexpected connection state before FUTEX_WAKE: %v", cs) } - - // Read the handshake response. - respLen := atomic.LoadUint32(ep.dataLen()) - if respLen > ep.dataCap { - return resp, fmt.Errorf("invalid handshake response length %d (maximum %d)", respLen, ep.dataCap) - } - if err := json.NewDecoder(ep.NewReader(respLen)).Decode(&resp); err != nil { - return resp, fmt.Errorf("error reading handshake response: %v", err) - } - - return resp, nil } -func (ep *Endpoint) futexSwitchToPeer() error { - // Update connection state to indicate that the peer should be active. - if !atomic.CompareAndSwapUint32(ep.connState(), ep.activeState, ep.inactiveState) { - switch cs := atomic.LoadUint32(ep.connState()); cs { - case csShutdown: - return ShutdownError{} - default: - return fmt.Errorf("unexpected connection state before FUTEX_WAKE: %v", cs) - } - } - - // Wake the peer's Endpoint.futexSwitchFromPeer(). +func (ep *Endpoint) futexWakePeer() error { if err := ep.futexWakeConnState(1); err != nil { return fmt.Errorf("failed to FUTEX_WAKE peer Endpoint: %v", err) } return nil } -func (ep *Endpoint) futexSwitchFromPeer() error { +func (ep *Endpoint) futexWaitUntilActive() error { for { switch cs := atomic.LoadUint32(ep.connState()); cs { case ep.activeState: diff --git a/pkg/goid/BUILD b/pkg/goid/BUILD index d855b702c..08832a8ae 100644 --- a/pkg/goid/BUILD +++ b/pkg/goid/BUILD @@ -9,6 +9,7 @@ go_library( "goid_amd64.s", "goid_arm64.s", ], + stateify = False, visibility = ["//visibility:public"], ) diff --git a/pkg/p9/client.go b/pkg/p9/client.go index 71e944c30..eadea390a 100644 --- a/pkg/p9/client.go +++ b/pkg/p9/client.go @@ -570,6 +570,8 @@ func (c *Client) Version() uint32 { func (c *Client) Close() { // unet.Socket.Shutdown() has no effect if unet.Socket.Close() has already // been called (by c.watch()). - c.socket.Shutdown() + if err := c.socket.Shutdown(); err != nil { + log.Warningf("Socket.Shutdown() failed (FD: %d): %v", c.socket.FD(), err) + } c.closedWg.Wait() } diff --git a/pkg/p9/client_file.go b/pkg/p9/client_file.go index 28fe081d6..8b46a2987 100644 --- a/pkg/p9/client_file.go +++ b/pkg/p9/client_file.go @@ -478,28 +478,23 @@ func (r *ReadWriterFile) ReadAt(p []byte, offset int64) (int, error) { } // Write implements part of the io.ReadWriter interface. +// +// Note that this may return a short write with a nil error. This violates the +// contract of io.Writer, but is more consistent with gVisor's pattern of +// returning errors that correspond to Linux errnos. Since short writes without +// error are common in Linux, returning a nil error is appropriate. func (r *ReadWriterFile) Write(p []byte) (int, error) { n, err := r.File.WriteAt(p, r.Offset) r.Offset += uint64(n) - if err != nil { - return n, err - } - if n < len(p) { - return n, io.ErrShortWrite - } - return n, nil + return n, err } // WriteAt implements the io.WriteAt interface. +// +// Note that this may return a short write with a nil error. This violates the +// contract of io.WriterAt. See comment on Write for justification. func (r *ReadWriterFile) WriteAt(p []byte, offset int64) (int, error) { - n, err := r.File.WriteAt(p, uint64(offset)) - if err != nil { - return n, err - } - if n < len(p) { - return n, io.ErrShortWrite - } - return n, nil + return r.File.WriteAt(p, uint64(offset)) } // Rename implements File.Rename. diff --git a/pkg/p9/handlers.go b/pkg/p9/handlers.go index abd237f46..81ceb37c5 100644 --- a/pkg/p9/handlers.go +++ b/pkg/p9/handlers.go @@ -296,25 +296,6 @@ func (t *Tlopen) handle(cs *connState) message { } defer ref.DecRef() - ref.openedMu.Lock() - defer ref.openedMu.Unlock() - - // Has it been opened already? - if ref.opened || !CanOpen(ref.mode) { - return newErr(syscall.EINVAL) - } - - if ref.mode.IsDir() { - // Directory must be opened ReadOnly. - if t.Flags&OpenFlagsModeMask != ReadOnly { - return newErr(syscall.EISDIR) - } - // Directory not truncatable. - if t.Flags&OpenTruncate != 0 { - return newErr(syscall.EISDIR) - } - } - var ( qid QID ioUnit uint32 @@ -326,6 +307,22 @@ func (t *Tlopen) handle(cs *connState) message { return syscall.EINVAL } + // Has it been opened already? + if ref.opened || !CanOpen(ref.mode) { + return syscall.EINVAL + } + + if ref.mode.IsDir() { + // Directory must be opened ReadOnly. + if t.Flags&OpenFlagsModeMask != ReadOnly { + return syscall.EISDIR + } + // Directory not truncatable. + if t.Flags&OpenTruncate != 0 { + return syscall.EISDIR + } + } + osFile, qid, ioUnit, err = ref.file.Open(t.Flags) return err }); err != nil { @@ -366,7 +363,7 @@ func (t *Tlcreate) do(cs *connState, uid UID) (*Rlcreate, error) { } // Not allowed on open directories. - if _, opened := ref.OpenFlags(); opened { + if ref.opened { return syscall.EINVAL } @@ -437,7 +434,7 @@ func (t *Tsymlink) do(cs *connState, uid UID) (*Rsymlink, error) { } // Not allowed on open directories. - if _, opened := ref.OpenFlags(); opened { + if ref.opened { return syscall.EINVAL } @@ -476,7 +473,7 @@ func (t *Tlink) handle(cs *connState) message { } // Not allowed on open directories. - if _, opened := ref.OpenFlags(); opened { + if ref.opened { return syscall.EINVAL } @@ -518,7 +515,7 @@ func (t *Trenameat) handle(cs *connState) message { } // Not allowed on open directories. - if _, opened := ref.OpenFlags(); opened { + if ref.opened { return syscall.EINVAL } @@ -561,7 +558,7 @@ func (t *Tunlinkat) handle(cs *connState) message { } // Not allowed on open directories. - if _, opened := ref.OpenFlags(); opened { + if ref.opened { return syscall.EINVAL } @@ -701,13 +698,12 @@ func (t *Tread) handle(cs *connState) message { ) if err := ref.safelyRead(func() (err error) { // Has it been opened already? - openFlags, opened := ref.OpenFlags() - if !opened { + if !ref.opened { return syscall.EINVAL } // Can it be read? Check permissions. - if openFlags&OpenFlagsModeMask == WriteOnly { + if ref.openFlags&OpenFlagsModeMask == WriteOnly { return syscall.EPERM } @@ -731,13 +727,12 @@ func (t *Twrite) handle(cs *connState) message { var n int if err := ref.safelyRead(func() (err error) { // Has it been opened already? - openFlags, opened := ref.OpenFlags() - if !opened { + if !ref.opened { return syscall.EINVAL } // Can it be written? Check permissions. - if openFlags&OpenFlagsModeMask == ReadOnly { + if ref.openFlags&OpenFlagsModeMask == ReadOnly { return syscall.EPERM } @@ -778,7 +773,7 @@ func (t *Tmknod) do(cs *connState, uid UID) (*Rmknod, error) { } // Not allowed on open directories. - if _, opened := ref.OpenFlags(); opened { + if ref.opened { return syscall.EINVAL } @@ -820,7 +815,7 @@ func (t *Tmkdir) do(cs *connState, uid UID) (*Rmkdir, error) { } // Not allowed on open directories. - if _, opened := ref.OpenFlags(); opened { + if ref.opened { return syscall.EINVAL } @@ -898,13 +893,12 @@ func (t *Tallocate) handle(cs *connState) message { if err := ref.safelyWrite(func() error { // Has it been opened already? - openFlags, opened := ref.OpenFlags() - if !opened { + if !ref.opened { return syscall.EINVAL } // Can it be written? Check permissions. - if openFlags&OpenFlagsModeMask == ReadOnly { + if ref.openFlags&OpenFlagsModeMask == ReadOnly { return syscall.EBADF } @@ -1049,8 +1043,8 @@ func (t *Treaddir) handle(cs *connState) message { return syscall.EINVAL } - // Has it been opened already? - if _, opened := ref.OpenFlags(); !opened { + // Has it been opened yet? + if !ref.opened { return syscall.EINVAL } @@ -1076,8 +1070,8 @@ func (t *Tfsync) handle(cs *connState) message { defer ref.DecRef() if err := ref.safelyRead(func() (err error) { - // Has it been opened already? - if _, opened := ref.OpenFlags(); !opened { + // Has it been opened yet? + if !ref.opened { return syscall.EINVAL } @@ -1185,8 +1179,13 @@ func doWalk(cs *connState, ref *fidRef, names []string, getattr bool) (qids []QI } // Has it been opened already? - if _, opened := ref.OpenFlags(); opened { - err = syscall.EBUSY + err = ref.safelyRead(func() (err error) { + if ref.opened { + return syscall.EBUSY + } + return nil + }) + if err != nil { return } diff --git a/pkg/p9/server.go b/pkg/p9/server.go index 3736f12a3..8c5c434fd 100644 --- a/pkg/p9/server.go +++ b/pkg/p9/server.go @@ -134,12 +134,11 @@ type fidRef struct { // The node above will be closed only when refs reaches zero. refs int64 - // openedMu protects opened and openFlags. - openedMu sync.Mutex - // opened indicates whether this has been opened already. // // This is updated in handlers.go. + // + // opened is protected by pathNode.opMu or renameMu (for write). opened bool // mode is the fidRef's mode from the walk. Only the type bits are @@ -151,6 +150,8 @@ type fidRef struct { // openFlags is the mode used in the open. // // This is updated in handlers.go. + // + // openFlags is protected by pathNode.opMu or renameMu (for write). openFlags OpenFlags // pathNode is the current pathNode for this FID. @@ -177,13 +178,6 @@ type fidRef struct { deleted uint32 } -// OpenFlags returns the flags the file was opened with and true iff the fid was opened previously. -func (f *fidRef) OpenFlags() (OpenFlags, bool) { - f.openedMu.Lock() - defer f.openedMu.Unlock() - return f.openFlags, f.opened -} - // IncRef increases the references on a fid. func (f *fidRef) IncRef() { atomic.AddInt64(&f.refs, 1) diff --git a/pkg/sentry/arch/signal.go b/pkg/sentry/arch/signal.go index c9fb55d00..5138f3bf5 100644 --- a/pkg/sentry/arch/signal.go +++ b/pkg/sentry/arch/signal.go @@ -251,3 +251,26 @@ func (s *SignalInfo) Arch() uint32 { func (s *SignalInfo) SetArch(val uint32) { usermem.ByteOrder.PutUint32(s.Fields[12:16], val) } + +// Band returns the si_band field. +func (s *SignalInfo) Band() int64 { + return int64(usermem.ByteOrder.Uint64(s.Fields[0:8])) +} + +// SetBand mutates the si_band field. +func (s *SignalInfo) SetBand(val int64) { + // Note: this assumes the platform uses `long` as `__ARCH_SI_BAND_T`. + // On some platforms, which gVisor doesn't support, `__ARCH_SI_BAND_T` is + // `int`. See siginfo.h. + usermem.ByteOrder.PutUint64(s.Fields[0:8], uint64(val)) +} + +// FD returns the si_fd field. +func (s *SignalInfo) FD() uint32 { + return usermem.ByteOrder.Uint32(s.Fields[8:12]) +} + +// SetFD mutates the si_fd field. +func (s *SignalInfo) SetFD(val uint32) { + usermem.ByteOrder.PutUint32(s.Fields[8:12], val) +} diff --git a/pkg/sentry/fs/BUILD b/pkg/sentry/fs/BUILD index ea85ab33c..5c3e852e9 100644 --- a/pkg/sentry/fs/BUILD +++ b/pkg/sentry/fs/BUILD @@ -49,13 +49,13 @@ go_library( "//pkg/amutex", "//pkg/context", "//pkg/log", - "//pkg/metric", "//pkg/p9", "//pkg/refs", "//pkg/secio", "//pkg/sentry/arch", "//pkg/sentry/device", "//pkg/sentry/fs/lock", + "//pkg/sentry/fsmetric", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", "//pkg/sentry/limits", diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index 72ea70fcf..57f904801 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -17,13 +17,12 @@ package fs import ( "math" "sync/atomic" - "time" "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/uniqueid" @@ -33,28 +32,6 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -var ( - // RecordWaitTime controls writing metrics for filesystem reads. - // Enabling this comes at a small CPU cost due to performing two - // monotonic clock reads per read call. - // - // Note that this is only performed in the direct read path, and may - // not be consistently applied for other forms of reads, such as - // splice. - RecordWaitTime = false - - reads = metric.MustCreateNewUint64Metric("/fs/reads", false /* sync */, "Number of file reads.") - readWait = metric.MustCreateNewUint64NanosecondsMetric("/fs/read_wait", false /* sync */, "Time waiting on file reads, in nanoseconds.") -) - -// IncrementWait increments the given wait time metric, if enabled. -func IncrementWait(m *metric.Uint64Metric, start time.Time) { - if !RecordWaitTime { - return - } - m.IncrementBy(uint64(time.Since(start))) -} - // FileMaxOffset is the maximum possible file offset. const FileMaxOffset = math.MaxInt64 @@ -257,22 +234,19 @@ func (f *File) Readdir(ctx context.Context, serializer DentrySerializer) error { // // Returns syserror.ErrInterrupted if reading was interrupted. func (f *File) Readv(ctx context.Context, dst usermem.IOSequence) (int64, error) { - var start time.Time - if RecordWaitTime { - start = time.Now() - } + start := fsmetric.StartReadWait() + defer fsmetric.FinishReadWait(fsmetric.ReadWait, start) + if !f.mu.Lock(ctx) { - IncrementWait(readWait, start) return 0, syserror.ErrInterrupted } - reads.Increment() + fsmetric.Reads.Increment() n, err := f.FileOperations.Read(ctx, f, dst, f.offset) if n > 0 && !f.flags.NonSeekable { atomic.AddInt64(&f.offset, n) } f.mu.Unlock() - IncrementWait(readWait, start) return n, err } @@ -282,19 +256,16 @@ func (f *File) Readv(ctx context.Context, dst usermem.IOSequence) (int64, error) // // Otherwise same as Readv. func (f *File) Preadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - var start time.Time - if RecordWaitTime { - start = time.Now() - } + start := fsmetric.StartReadWait() + defer fsmetric.FinishReadWait(fsmetric.ReadWait, start) + if !f.mu.Lock(ctx) { - IncrementWait(readWait, start) return 0, syserror.ErrInterrupted } - reads.Increment() + fsmetric.Reads.Increment() n, err := f.FileOperations.Read(ctx, f, dst, offset) f.mu.Unlock() - IncrementWait(readWait, start) return n, err } diff --git a/pkg/sentry/fs/fs.go b/pkg/sentry/fs/fs.go index d2dbff268..a020da53b 100644 --- a/pkg/sentry/fs/fs.go +++ b/pkg/sentry/fs/fs.go @@ -65,7 +65,7 @@ var ( // runs with the lock held for reading. AsyncBarrier will take the lock // for writing, thus ensuring that all Async work completes before // AsyncBarrier returns. - workMu sync.RWMutex + workMu sync.CrossGoroutineRWMutex // asyncError is used to store up to one asynchronous execution error. asyncError = make(chan error, 1) diff --git a/pkg/sentry/fs/gofer/BUILD b/pkg/sentry/fs/gofer/BUILD index fea135eea..4c30098cd 100644 --- a/pkg/sentry/fs/gofer/BUILD +++ b/pkg/sentry/fs/gofer/BUILD @@ -28,7 +28,6 @@ go_library( "//pkg/context", "//pkg/fd", "//pkg/log", - "//pkg/metric", "//pkg/p9", "//pkg/refs", "//pkg/safemem", @@ -38,6 +37,7 @@ go_library( "//pkg/sentry/fs/fdpipe", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/host", + "//pkg/sentry/fsmetric", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", diff --git a/pkg/sentry/fs/gofer/file.go b/pkg/sentry/fs/gofer/file.go index c0bc63a32..bb63448cb 100644 --- a/pkg/sentry/fs/gofer/file.go +++ b/pkg/sentry/fs/gofer/file.go @@ -21,27 +21,17 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) -var ( - opensWX = metric.MustCreateNewUint64Metric("/gofer/opened_write_execute_file", true /* sync */, "Number of times a writable+executable file was opened from a gofer.") - opens9P = metric.MustCreateNewUint64Metric("/gofer/opens_9p", false /* sync */, "Number of times a 9P file was opened from a gofer.") - opensHost = metric.MustCreateNewUint64Metric("/gofer/opens_host", false /* sync */, "Number of times a host file was opened from a gofer.") - reads9P = metric.MustCreateNewUint64Metric("/gofer/reads_9p", false /* sync */, "Number of 9P file reads from a gofer.") - readWait9P = metric.MustCreateNewUint64NanosecondsMetric("/gofer/read_wait_9p", false /* sync */, "Time waiting on 9P file reads from a gofer, in nanoseconds.") - readsHost = metric.MustCreateNewUint64Metric("/gofer/reads_host", false /* sync */, "Number of host file reads from a gofer.") - readWaitHost = metric.MustCreateNewUint64NanosecondsMetric("/gofer/read_wait_host", false /* sync */, "Time waiting on host file reads from a gofer, in nanoseconds.") -) - // fileOperations implements fs.FileOperations for a remote file system. // // +stateify savable @@ -101,14 +91,14 @@ func NewFile(ctx context.Context, dirent *fs.Dirent, name string, flags fs.FileF } if flags.Write { if err := dirent.Inode.CheckPermission(ctx, fs.PermMask{Execute: true}); err == nil { - opensWX.Increment() + fsmetric.GoferOpensWX.Increment() log.Warningf("Opened a writable executable: %q", name) } } if handles.Host != nil { - opensHost.Increment() + fsmetric.GoferOpensHost.Increment() } else { - opens9P.Increment() + fsmetric.GoferOpens9P.Increment() } return fs.NewFile(ctx, dirent, flags, f) } @@ -278,20 +268,17 @@ func (f *fileOperations) Write(ctx context.Context, file *fs.File, src usermem.I // use this function rather than using a defer in Read() to avoid the performance hit of defer. func (f *fileOperations) incrementReadCounters(start time.Time) { if f.handles.Host != nil { - readsHost.Increment() - fs.IncrementWait(readWaitHost, start) + fsmetric.GoferReadsHost.Increment() + fsmetric.FinishReadWait(fsmetric.GoferReadWaitHost, start) } else { - reads9P.Increment() - fs.IncrementWait(readWait9P, start) + fsmetric.GoferReads9P.Increment() + fsmetric.FinishReadWait(fsmetric.GoferReadWait9P, start) } } // Read implements fs.FileOperations.Read. func (f *fileOperations) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - var start time.Time - if fs.RecordWaitTime { - start = time.Now() - } + start := fsmetric.StartReadWait() if fs.IsDir(file.Dirent.Inode.StableAttr) { // Not all remote file systems enforce this so this client does. f.incrementReadCounters(start) diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 3a225fd39..9d6fdd08f 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -117,7 +117,7 @@ type inodeFileState struct { // loading is acquired when the inodeFileState begins an asynchronous // load. It releases when the load is complete. Callers that require all // state to be available should call waitForLoad() to ensure that. - loading sync.Mutex `state:".(struct{})"` + loading sync.CrossGoroutineMutex `state:".(struct{})"` // savedUAttr is only allocated during S/R. It points to the save-time // unstable attributes and is used to validate restore-time ones. diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 004910453..9b3d8166a 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -18,9 +18,9 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" @@ -28,8 +28,6 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -var opens = metric.MustCreateNewUint64Metric("/fs/opens", false /* sync */, "Number of file opens.") - // Inode is a file system object that can be simultaneously referenced by different // components of the VFS (Dirent, fs.File, etc). // @@ -247,7 +245,7 @@ func (i *Inode) GetFile(ctx context.Context, d *Dirent, flags FileFlags) (*File, if i.overlay != nil { return overlayGetFile(ctx, i.overlay, d, flags) } - opens.Increment() + fsmetric.Opens.Increment() return i.InodeOperations.GetFile(ctx, d, flags) } diff --git a/pkg/sentry/fs/proc/sys.go b/pkg/sentry/fs/proc/sys.go index f8aad2dbd..b998fb75d 100644 --- a/pkg/sentry/fs/proc/sys.go +++ b/pkg/sentry/fs/proc/sys.go @@ -84,6 +84,7 @@ func (p *proc) newKernelDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode children := map[string]*fs.Inode{ "hostname": newProcInode(ctx, &h, msrc, fs.SpecialFile, nil), + "sem": newStaticProcInode(ctx, msrc, []byte(fmt.Sprintf("%d\t%d\t%d\t%d\n", linux.SEMMSL, linux.SEMMNS, linux.SEMOPM, linux.SEMMNI))), "shmall": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMALL, 10))), "shmmax": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMAX, 10))), "shmmni": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMNI, 10))), diff --git a/pkg/sentry/fs/tmpfs/BUILD b/pkg/sentry/fs/tmpfs/BUILD index aa7199014..b521a86a2 100644 --- a/pkg/sentry/fs/tmpfs/BUILD +++ b/pkg/sentry/fs/tmpfs/BUILD @@ -15,12 +15,12 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/context", - "//pkg/metric", "//pkg/safemem", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/ramfs", + "//pkg/sentry/fsmetric", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", diff --git a/pkg/sentry/fs/tmpfs/inode_file.go b/pkg/sentry/fs/tmpfs/inode_file.go index d6c65301c..e04cd608d 100644 --- a/pkg/sentry/fs/tmpfs/inode_file.go +++ b/pkg/sentry/fs/tmpfs/inode_file.go @@ -18,14 +18,13 @@ import ( "fmt" "io" "math" - "time" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -35,13 +34,6 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -var ( - opensRO = metric.MustCreateNewUint64Metric("/in_memory_file/opens_ro", false /* sync */, "Number of times an in-memory file was opened in read-only mode.") - opensW = metric.MustCreateNewUint64Metric("/in_memory_file/opens_w", false /* sync */, "Number of times an in-memory file was opened in write mode.") - reads = metric.MustCreateNewUint64Metric("/in_memory_file/reads", false /* sync */, "Number of in-memory file reads.") - readWait = metric.MustCreateNewUint64NanosecondsMetric("/in_memory_file/read_wait", false /* sync */, "Time waiting on in-memory file reads, in nanoseconds.") -) - // fileInodeOperations implements fs.InodeOperations for a regular tmpfs file. // These files are backed by pages allocated from a platform.Memory, and may be // directly mapped. @@ -157,9 +149,9 @@ func (*fileInodeOperations) Rename(ctx context.Context, inode *fs.Inode, oldPare // GetFile implements fs.InodeOperations.GetFile. func (f *fileInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { if flags.Write { - opensW.Increment() + fsmetric.TmpfsOpensW.Increment() } else if flags.Read { - opensRO.Increment() + fsmetric.TmpfsOpensRO.Increment() } flags.Pread = true flags.Pwrite = true @@ -319,14 +311,12 @@ func (*fileInodeOperations) StatFS(context.Context) (fs.Info, error) { } func (f *fileInodeOperations) read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - var start time.Time - if fs.RecordWaitTime { - start = time.Now() - } - reads.Increment() + start := fsmetric.StartReadWait() + defer fsmetric.FinishReadWait(fsmetric.TmpfsReadWait, start) + fsmetric.TmpfsReads.Increment() + // Zero length reads for tmpfs are no-ops. if dst.NumBytes() == 0 { - fs.IncrementWait(readWait, start) return 0, nil } @@ -343,7 +333,6 @@ func (f *fileInodeOperations) read(ctx context.Context, file *fs.File, dst userm size := f.attr.Size f.dataMu.RUnlock() if offset >= size { - fs.IncrementWait(readWait, start) return 0, io.EOF } @@ -354,7 +343,6 @@ func (f *fileInodeOperations) read(ctx context.Context, file *fs.File, dst userm f.attr.AccessTime = ktime.NowFromContext(ctx) f.attrMu.Unlock() } - fs.IncrementWait(readWait, start) return n, err } diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index 89c3ef079..1bbe6fdb7 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -363,7 +363,7 @@ func (fd *DeviceFD) Readiness(mask waiter.EventMask) waiter.EventMask { func (fd *DeviceFD) readinessLocked(mask waiter.EventMask) waiter.EventMask { var ready waiter.EventMask - if fd.fs.umounted { + if fd.fs == nil || fd.fs.umounted { ready |= waiter.EventErr return ready & mask } diff --git a/pkg/sentry/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD index 4c3e9acf8..807b6ed1f 100644 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ b/pkg/sentry/fsimpl/gofer/BUILD @@ -59,6 +59,7 @@ go_library( "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/lock", "//pkg/sentry/fsimpl/host", + "//pkg/sentry/fsmetric", "//pkg/sentry/hostfd", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 2294c490e..df27554d3 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -24,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/fsimpl/host" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" @@ -985,14 +986,11 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open switch d.fileType() { case linux.S_IFREG: if !d.fs.opts.regularFilesUseSpecialFileFD { - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, trunc); err != nil { + if err := d.ensureSharedHandle(ctx, ats.MayRead(), ats.MayWrite(), trunc); err != nil { return nil, err } - fd := ®ularFileFD{} - fd.LockFD.Init(&d.locks) - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ - AllowDirectIO: true, - }); err != nil { + fd, err := newRegularFileFD(mnt, d, opts.Flags) + if err != nil { return nil, err } vfd = &fd.vfsfd @@ -1019,6 +1017,11 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { return nil, err } + if atomic.LoadInt32(&d.readFD) >= 0 { + fsmetric.GoferOpensHost.Increment() + } else { + fsmetric.GoferOpens9P.Increment() + } return &fd.vfsfd, nil case linux.S_IFLNK: // Can't open symlinks without O_PATH (which is unimplemented). @@ -1110,7 +1113,7 @@ retry: return nil, err } } - fd, err := newSpecialFileFD(h, mnt, d, &d.locks, opts.Flags) + fd, err := newSpecialFileFD(h, mnt, d, opts.Flags) if err != nil { h.close(ctx) return nil, err @@ -1205,11 +1208,8 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving // Finally, construct a file description representing the created file. var childVFSFD *vfs.FileDescription if useRegularFileFD { - fd := ®ularFileFD{} - fd.LockFD.Init(&child.locks) - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ - AllowDirectIO: true, - }); err != nil { + fd, err := newRegularFileFD(mnt, child, opts.Flags) + if err != nil { return nil, err } childVFSFD = &fd.vfsfd @@ -1221,7 +1221,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving if fdobj != nil { h.fd = int32(fdobj.Release()) } - fd, err := newSpecialFileFD(h, mnt, child, &d.locks, opts.Flags) + fd, err := newSpecialFileFD(h, mnt, child, opts.Flags) if err != nil { h.close(ctx) return nil, err diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index 75a836899..3cdb1e659 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -743,7 +743,9 @@ type dentry struct { // for memory mappings. If mmapFD is -1, no such FD is available, and the // internal page cache implementation is used for memory mappings instead. // - // These fields are protected by handleMu. + // These fields are protected by handleMu. readFD, writeFD, and mmapFD are + // additionally written using atomic memory operations, allowing them to be + // read (albeit racily) with atomic.LoadInt32() without locking handleMu. // // readFile and writeFile may or may not represent the same p9.File. Once // either p9.File transitions from closed (isNil() == true) to open @@ -1351,16 +1353,11 @@ func (d *dentry) checkCachingLocked(ctx context.Context) { return } if refs > 0 { - if d.cached { - // This isn't strictly necessary (fs.cachedDentries is permitted to - // contain dentries with non-zero refs, which are skipped by - // fs.evictCachedDentryLocked() upon reaching the end of the LRU), - // but since we are already holding fs.renameMu for writing we may - // as well. - d.fs.cachedDentries.Remove(d) - d.fs.cachedDentriesLen-- - d.cached = false - } + // This isn't strictly necessary (fs.cachedDentries is permitted to + // contain dentries with non-zero refs, which are skipped by + // fs.evictCachedDentryLocked() upon reaching the end of the LRU), but + // since we are already holding fs.renameMu for writing we may as well. + d.removeFromCacheLocked() return } // Deleted and invalidated dentries with zero references are no longer @@ -1369,20 +1366,18 @@ func (d *dentry) checkCachingLocked(ctx context.Context) { if d.isDeleted() { d.watches.HandleDeletion(ctx) } - if d.cached { - d.fs.cachedDentries.Remove(d) - d.fs.cachedDentriesLen-- - d.cached = false - } + d.removeFromCacheLocked() d.destroyLocked(ctx) return } - // If d still has inotify watches and it is not deleted or invalidated, we - // cannot cache it and allow it to be evicted. Otherwise, we will lose its - // watches, even if a new dentry is created for the same file in the future. - // Note that the size of d.watches cannot concurrently transition from zero - // to non-zero, because adding a watch requires holding a reference on d. + // If d still has inotify watches and it is not deleted or invalidated, it + // can't be evicted. Otherwise, we will lose its watches, even if a new + // dentry is created for the same file in the future. Note that the size of + // d.watches cannot concurrently transition from zero to non-zero, because + // adding a watch requires holding a reference on d. if d.watches.Size() > 0 { + // As in the refs > 0 case, this is not strictly necessary. + d.removeFromCacheLocked() return } @@ -1413,6 +1408,15 @@ func (d *dentry) checkCachingLocked(ctx context.Context) { } } +// Preconditions: d.fs.renameMu must be locked for writing. +func (d *dentry) removeFromCacheLocked() { + if d.cached { + d.fs.cachedDentries.Remove(d) + d.fs.cachedDentriesLen-- + d.cached = false + } +} + // Precondition: fs.renameMu must be locked for writing; it may be temporarily // unlocked. func (fs *filesystem) evictAllCachedDentriesLocked(ctx context.Context) { @@ -1426,12 +1430,10 @@ func (fs *filesystem) evictAllCachedDentriesLocked(ctx context.Context) { // * fs.cachedDentriesLen != 0. func (fs *filesystem) evictCachedDentryLocked(ctx context.Context) { victim := fs.cachedDentries.Back() - fs.cachedDentries.Remove(victim) - fs.cachedDentriesLen-- - victim.cached = false - // victim.refs may have become non-zero from an earlier path resolution - // since it was inserted into fs.cachedDentries. - if atomic.LoadInt64(&victim.refs) == 0 { + victim.removeFromCacheLocked() + // victim.refs or victim.watches.Size() may have become non-zero from an + // earlier path resolution since it was inserted into fs.cachedDentries. + if atomic.LoadInt64(&victim.refs) == 0 && victim.watches.Size() == 0 { if victim.parent != nil { victim.parent.dirMu.Lock() if !victim.vfsd.IsDead() { @@ -1668,7 +1670,7 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool } fdsToClose = append(fdsToClose, d.readFD) invalidateTranslations = true - d.readFD = h.fd + atomic.StoreInt32(&d.readFD, h.fd) } else { // Otherwise, we want to avoid invalidating existing // memmap.Translations (which is expensive); instead, use @@ -1689,15 +1691,15 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool h.fd = d.readFD } } else { - d.readFD = h.fd + atomic.StoreInt32(&d.readFD, h.fd) } if d.writeFD != h.fd && d.writeFD >= 0 { fdsToClose = append(fdsToClose, d.writeFD) } - d.writeFD = h.fd - d.mmapFD = h.fd + atomic.StoreInt32(&d.writeFD, h.fd) + atomic.StoreInt32(&d.mmapFD, h.fd) } else if openReadable && d.readFD < 0 { - d.readFD = h.fd + atomic.StoreInt32(&d.readFD, h.fd) // If the file has not been opened for writing, the new FD may // be used for read-only memory mappings. If the file was // previously opened for reading (without an FD), then existing @@ -1705,10 +1707,10 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool // invalidate those mappings. if d.writeFile.isNil() { invalidateTranslations = !d.readFile.isNil() - d.mmapFD = h.fd + atomic.StoreInt32(&d.mmapFD, h.fd) } } else if openWritable && d.writeFD < 0 { - d.writeFD = h.fd + atomic.StoreInt32(&d.writeFD, h.fd) if d.readFD >= 0 { // We have an existing read-only FD, but the file has just // been opened for writing, so we need to start supporting @@ -1717,7 +1719,7 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool // writable memory mappings. Switch to using the internal // page cache. invalidateTranslations = true - d.mmapFD = -1 + atomic.StoreInt32(&d.mmapFD, -1) } } else { // The new FD is not useful. @@ -1729,7 +1731,7 @@ func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool // memory mappings. However, we have no writable host FD. Switch to // using the internal page cache. invalidateTranslations = true - d.mmapFD = -1 + atomic.StoreInt32(&d.mmapFD, -1) } // Switch to new fids. diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 652142ecc..283b220bb 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/usage" @@ -48,6 +49,25 @@ type regularFileFD struct { off int64 } +func newRegularFileFD(mnt *vfs.Mount, d *dentry, flags uint32) (*regularFileFD, error) { + fd := ®ularFileFD{} + fd.LockFD.Init(&d.locks) + if err := fd.vfsfd.Init(fd, flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ + AllowDirectIO: true, + }); err != nil { + return nil, err + } + if fd.vfsfd.IsWritable() && (atomic.LoadUint32(&d.mode)&0111 != 0) { + fsmetric.GoferOpensWX.Increment() + } + if atomic.LoadInt32(&d.mmapFD) >= 0 { + fsmetric.GoferOpensHost.Increment() + } else { + fsmetric.GoferOpens9P.Increment() + } + return fd, nil +} + // Release implements vfs.FileDescriptionImpl.Release. func (fd *regularFileFD) Release(context.Context) { } @@ -89,6 +109,18 @@ func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + start := fsmetric.StartReadWait() + d := fd.dentry() + defer func() { + if atomic.LoadInt32(&d.readFD) >= 0 { + fsmetric.GoferReadsHost.Increment() + fsmetric.FinishReadWait(fsmetric.GoferReadWaitHost, start) + } else { + fsmetric.GoferReads9P.Increment() + fsmetric.FinishReadWait(fsmetric.GoferReadWait9P, start) + } + }() + if offset < 0 { return 0, syserror.EINVAL } @@ -102,7 +134,6 @@ func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs // Check for reading at EOF before calling into MM (but not under // InteropModeShared, which makes d.size unreliable). - d := fd.dentry() if d.cachedMetadataAuthoritative() && uint64(offset) >= atomic.LoadUint64(&d.size) { return 0, io.EOF } @@ -647,10 +678,7 @@ func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpt // Whether or not we have a host FD, we're not allowed to use it. return syserror.ENODEV } - d.handleMu.RLock() - haveFD := d.mmapFD >= 0 - d.handleMu.RUnlock() - if !haveFD { + if atomic.LoadInt32(&d.mmapFD) < 0 { return syserror.ENODEV } default: @@ -668,10 +696,7 @@ func (d *dentry) mayCachePages() bool { if d.fs.opts.forcePageCache { return true } - d.handleMu.RLock() - haveFD := d.mmapFD >= 0 - d.handleMu.RUnlock() - return haveFD + return atomic.LoadInt32(&d.mmapFD) >= 0 } // AddMapping implements memmap.Mappable.AddMapping. diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 625400c0b..089955a96 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -70,7 +71,7 @@ type specialFileFD struct { buf []byte } -func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, flags uint32) (*specialFileFD, error) { +func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, flags uint32) (*specialFileFD, error) { ftype := d.fileType() seekable := ftype == linux.S_IFREG || ftype == linux.S_IFCHR || ftype == linux.S_IFBLK haveQueue := (ftype == linux.S_IFIFO || ftype == linux.S_IFSOCK) && h.fd >= 0 @@ -80,7 +81,7 @@ func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, seekable: seekable, haveQueue: haveQueue, } - fd.LockFD.Init(locks) + fd.LockFD.Init(&d.locks) if haveQueue { if err := fdnotifier.AddFD(h.fd, &fd.queue); err != nil { return nil, err @@ -98,6 +99,14 @@ func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, d.fs.syncMu.Lock() d.fs.specialFileFDs[fd] = struct{}{} d.fs.syncMu.Unlock() + if fd.vfsfd.IsWritable() && (atomic.LoadUint32(&d.mode)&0111 != 0) { + fsmetric.GoferOpensWX.Increment() + } + if h.fd >= 0 { + fsmetric.GoferOpensHost.Increment() + } else { + fsmetric.GoferOpens9P.Increment() + } return fd, nil } @@ -161,6 +170,17 @@ func (fd *specialFileFD) Allocate(ctx context.Context, mode, offset, length uint // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + start := fsmetric.StartReadWait() + defer func() { + if fd.handle.fd >= 0 { + fsmetric.GoferReadsHost.Increment() + fsmetric.FinishReadWait(fsmetric.GoferReadWaitHost, start) + } else { + fsmetric.GoferReads9P.Increment() + fsmetric.FinishReadWait(fsmetric.GoferReadWait9P, start) + } + }() + if fd.seekable && offset < 0 { return 0, syserror.EINVAL } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index c14abcff4..565d723f0 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -286,7 +286,7 @@ func (d *Dentry) cacheLocked(ctx context.Context) { refs := atomic.LoadInt64(&d.refs) if refs == -1 { // Dentry has already been destroyed. - panic(fmt.Sprintf("cacheLocked called on a dentry which has already been destroyed: %v", d)) + return } if refs > 0 { if d.cached { diff --git a/pkg/sentry/fsimpl/overlay/copy_up.go b/pkg/sentry/fsimpl/overlay/copy_up.go index 469f3a33d..27b00cf6f 100644 --- a/pkg/sentry/fsimpl/overlay/copy_up.go +++ b/pkg/sentry/fsimpl/overlay/copy_up.go @@ -16,7 +16,6 @@ package overlay import ( "fmt" - "io" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" @@ -129,25 +128,9 @@ func (d *dentry) copyUpLocked(ctx context.Context) error { return err } defer newFD.DecRef(ctx) - bufIOSeq := usermem.BytesIOSequence(make([]byte, 32*1024)) // arbitrary buffer size - for { - readN, readErr := oldFD.Read(ctx, bufIOSeq, vfs.ReadOptions{}) - if readErr != nil && readErr != io.EOF { - cleanupUndoCopyUp() - return readErr - } - total := int64(0) - for total < readN { - writeN, writeErr := newFD.Write(ctx, bufIOSeq.DropFirst64(total), vfs.WriteOptions{}) - total += writeN - if writeErr != nil { - cleanupUndoCopyUp() - return writeErr - } - } - if readErr == io.EOF { - break - } + if _, err := vfs.CopyRegularFileData(ctx, newFD, oldFD); err != nil { + cleanupUndoCopyUp() + return err } d.mapsMu.Lock() defer d.mapsMu.Unlock() diff --git a/pkg/sentry/fsimpl/overlay/regular_file.go b/pkg/sentry/fsimpl/overlay/regular_file.go index 2b89a7a6d..25c785fd4 100644 --- a/pkg/sentry/fsimpl/overlay/regular_file.go +++ b/pkg/sentry/fsimpl/overlay/regular_file.go @@ -103,8 +103,8 @@ func (fd *regularFileFD) currentFDLocked(ctx context.Context) (*vfs.FileDescript for e, mask := range fd.lowerWaiters { fd.cachedFD.EventUnregister(e) upperFD.EventRegister(e, mask) - if ready&mask != 0 { - e.Callback.Callback(e) + if m := ready & mask; m != 0 { + e.Callback.Callback(e, m) } } } diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 5cf8a071a..d4f6a5a9b 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -208,7 +208,7 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { for _, se := range n.kernel.ListSockets() { s := se.SockVFS2 if !s.TryIncRef() { - log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) + // Racing with socket destruction, this is ok. continue } if family, _, _ := s.Impl().(socket.SocketVFS2).Type(); family != linux.AF_UNIX { @@ -351,7 +351,7 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, for _, se := range k.ListSockets() { s := se.SockVFS2 if !s.TryIncRef() { - log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) + // Racing with socket destruction, this is ok. continue } sops, ok := s.Impl().(socket.SocketVFS2) @@ -516,7 +516,7 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { for _, se := range d.kernel.ListSockets() { s := se.SockVFS2 if !s.TryIncRef() { - log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) + // Racing with socket destruction, this is ok. continue } sops, ok := s.Impl().(socket.SocketVFS2) diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index 7c7afdcfa..25c407d98 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -44,6 +44,7 @@ func (fs *filesystem) newSysDir(ctx context.Context, root *auth.Credentials, k * return fs.newStaticDir(ctx, root, map[string]kernfs.Inode{ "kernel": fs.newStaticDir(ctx, root, map[string]kernfs.Inode{ "hostname": fs.newInode(ctx, root, 0444, &hostnameData{}), + "sem": fs.newInode(ctx, root, 0444, newStaticFile(fmt.Sprintf("%d\t%d\t%d\t%d\n", linux.SEMMSL, linux.SEMMNS, linux.SEMOPM, linux.SEMMNI))), "shmall": fs.newInode(ctx, root, 0444, shmData(linux.SHMALL)), "shmmax": fs.newInode(ctx, root, 0444, shmData(linux.SHMMAX)), "shmmni": fs.newInode(ctx, root, 0444, shmData(linux.SHMMNI)), diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index fe520b6fd..09957c2b7 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -67,6 +67,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/lock", + "//pkg/sentry/fsmetric", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", "//pkg/sentry/kernel/time", diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index e39cd305b..9296db2fb 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -381,6 +382,8 @@ afterTrailingSymlink: creds := rp.Credentials() child := fs.newDentry(fs.newRegularFile(creds.EffectiveKUID, creds.EffectiveKGID, opts.Mode)) parentDir.insertChildLocked(child, name) + child.IncRef() + defer child.DecRef(ctx) unlock() fd, err := child.open(ctx, rp, &opts, true) if err != nil { @@ -437,6 +440,11 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open return nil, err } } + if fd.vfsfd.IsWritable() { + fsmetric.TmpfsOpensW.Increment() + } else if fd.vfsfd.IsReadable() { + fsmetric.TmpfsOpensRO.Increment() + } return &fd.vfsfd, nil case *directory: // Can't open directories writably. diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index f8e0cffb0..6255a7c84 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -359,6 +360,10 @@ func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + start := fsmetric.StartReadWait() + defer fsmetric.FinishReadWait(fsmetric.TmpfsReadWait, start) + fsmetric.TmpfsReads.Increment() + if offset < 0 { return 0, syserror.EINVAL } diff --git a/pkg/sentry/fsimpl/verity/filesystem.go b/pkg/sentry/fsimpl/verity/filesystem.go index add5dd48e..04e7110a3 100644 --- a/pkg/sentry/fsimpl/verity/filesystem.go +++ b/pkg/sentry/fsimpl/verity/filesystem.go @@ -107,8 +107,10 @@ func (fs *filesystem) renameMuUnlockAndCheckDrop(ctx context.Context, ds **[]*de // Dentries which may have a reference count of zero, and which therefore // should be dropped once traversal is complete, are appended to ds. // -// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. -// !rp.Done(). +// Preconditions: +// * fs.renameMu must be locked. +// * d.dirMu must be locked. +// * !rp.Done(). func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, mayFollowSymlinks bool, ds **[]*dentry) (*dentry, error) { if !d.isDir() { return nil, syserror.ENOTDIR @@ -158,15 +160,19 @@ afterSymlink: return child, nil } -// verifyChild verifies the hash of child against the already verified hash of -// the parent to ensure the child is expected. verifyChild triggers a sentry -// panic if unexpected modifications to the file system are detected. In +// verifyChildLocked verifies the hash of child against the already verified +// hash of the parent to ensure the child is expected. verifyChild triggers a +// sentry panic if unexpected modifications to the file system are detected. In // noCrashOnVerificationFailure mode it returns a syserror instead. -// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. +// +// Preconditions: +// * fs.renameMu must be locked. +// * d.dirMu must be locked. +// // TODO(b/166474175): Investigate all possible errors returned in this // function, and make sure we differentiate all errors that indicate unexpected // modifications to the file system from the ones that are not harmful. -func (fs *filesystem) verifyChild(ctx context.Context, parent *dentry, child *dentry) (*dentry, error) { +func (fs *filesystem) verifyChildLocked(ctx context.Context, parent *dentry, child *dentry) (*dentry, error) { vfsObj := fs.vfsfs.VirtualFilesystem() // Get the path to the child dentry. This is only used to provide path @@ -248,7 +254,7 @@ func (fs *filesystem) verifyChild(ctx context.Context, parent *dentry, child *de return nil, alertIntegrityViolation(fmt.Sprintf("Failed to convert xattr %s for %s to int: %v", merkleSizeXattr, childPath, err)) } - fdReader := vfs.FileReadWriteSeeker{ + fdReader := FileReadWriteSeeker{ FD: parentMerkleFD, Ctx: ctx, } @@ -268,7 +274,8 @@ func (fs *filesystem) verifyChild(ctx context.Context, parent *dentry, child *de // contain the hash of the children in the parent Merkle tree when // Verify returns with success. var buf bytes.Buffer - if _, err := merkletree.Verify(&merkletree.VerifyParams{ + parent.hashMu.RLock() + _, err = merkletree.Verify(&merkletree.VerifyParams{ Out: &buf, File: &fdReader, Tree: &fdReader, @@ -284,21 +291,27 @@ func (fs *filesystem) verifyChild(ctx context.Context, parent *dentry, child *de ReadSize: int64(merkletree.DigestSize(fs.alg.toLinuxHashAlg())), Expected: parent.hash, DataAndTreeInSameFile: true, - }); err != nil && err != io.EOF { + }) + parent.hashMu.RUnlock() + if err != nil && err != io.EOF { return nil, alertIntegrityViolation(fmt.Sprintf("Verification for %s failed: %v", childPath, err)) } // Cache child hash when it's verified the first time. + child.hashMu.Lock() if len(child.hash) == 0 { child.hash = buf.Bytes() } + child.hashMu.Unlock() return child, nil } -// verifyStatAndChildren verifies the stat and children names against the +// verifyStatAndChildrenLocked verifies the stat and children names against the // verified hash. The mode/uid/gid and childrenNames of the file is cached // after verified. -func (fs *filesystem) verifyStatAndChildren(ctx context.Context, d *dentry, stat linux.Statx) error { +// +// Preconditions: d.dirMu must be locked. +func (fs *filesystem) verifyStatAndChildrenLocked(ctx context.Context, d *dentry, stat linux.Statx) error { vfsObj := fs.vfsfs.VirtualFilesystem() // Get the path to the child dentry. This is only used to provide path @@ -384,12 +397,13 @@ func (fs *filesystem) verifyStatAndChildren(ctx context.Context, d *dentry, stat } } - fdReader := vfs.FileReadWriteSeeker{ + fdReader := FileReadWriteSeeker{ FD: fd, Ctx: ctx, } var buf bytes.Buffer + d.hashMu.RLock() params := &merkletree.VerifyParams{ Out: &buf, Tree: &fdReader, @@ -407,6 +421,7 @@ func (fs *filesystem) verifyStatAndChildren(ctx context.Context, d *dentry, stat Expected: d.hash, DataAndTreeInSameFile: false, } + d.hashMu.RUnlock() if atomic.LoadUint32(&d.mode)&linux.S_IFMT == linux.S_IFDIR { params.DataAndTreeInSameFile = true } @@ -421,7 +436,9 @@ func (fs *filesystem) verifyStatAndChildren(ctx context.Context, d *dentry, stat return nil } -// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. +// Preconditions: +// * fs.renameMu must be locked. +// * parent.dirMu must be locked. func (fs *filesystem) getChildLocked(ctx context.Context, parent *dentry, name string, ds **[]*dentry) (*dentry, error) { if child, ok := parent.children[name]; ok { // If verity is enabled on child, we should check again whether @@ -470,7 +487,7 @@ func (fs *filesystem) getChildLocked(ctx context.Context, parent *dentry, name s // be cached before enabled. if fs.allowRuntimeEnable { if parent.verityEnabled() { - if _, err := fs.verifyChild(ctx, parent, child); err != nil { + if _, err := fs.verifyChildLocked(ctx, parent, child); err != nil { return nil, err } } @@ -486,7 +503,7 @@ func (fs *filesystem) getChildLocked(ctx context.Context, parent *dentry, name s if err != nil { return nil, err } - if err := fs.verifyStatAndChildren(ctx, child, stat); err != nil { + if err := fs.verifyStatAndChildrenLocked(ctx, child, stat); err != nil { return nil, err } } @@ -506,7 +523,9 @@ func (fs *filesystem) getChildLocked(ctx context.Context, parent *dentry, name s return child, nil } -// Preconditions: fs.renameMu must be locked. parent.dirMu must be locked. +// Preconditions: +// * fs.renameMu must be locked. +// * parent.dirMu must be locked. func (fs *filesystem) lookupAndVerifyLocked(ctx context.Context, parent *dentry, name string) (*dentry, error) { vfsObj := fs.vfsfs.VirtualFilesystem() @@ -597,13 +616,13 @@ func (fs *filesystem) lookupAndVerifyLocked(ctx context.Context, parent *dentry, // allowRuntimeEnable mode and the parent directory hasn't been enabled // yet. if parent.verityEnabled() { - if _, err := fs.verifyChild(ctx, parent, child); err != nil { + if _, err := fs.verifyChildLocked(ctx, parent, child); err != nil { child.destroyLocked(ctx) return nil, err } } if child.verityEnabled() { - if err := fs.verifyStatAndChildren(ctx, child, stat); err != nil { + if err := fs.verifyStatAndChildrenLocked(ctx, child, stat); err != nil { child.destroyLocked(ctx) return nil, err } @@ -617,7 +636,9 @@ func (fs *filesystem) lookupAndVerifyLocked(ctx context.Context, parent *dentry, // rp.Start().Impl().(*dentry)). It does not check that the returned directory // is searchable by the provider of rp. // -// Preconditions: fs.renameMu must be locked. !rp.Done(). +// Preconditions: +// * fs.renameMu must be locked. +// * !rp.Done(). func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { for !rp.Final() { d.dirMu.Lock() @@ -958,11 +979,13 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err != nil { return linux.Statx{}, err } + d.dirMu.Lock() if d.verityEnabled() { - if err := fs.verifyStatAndChildren(ctx, d, stat); err != nil { + if err := fs.verifyStatAndChildrenLocked(ctx, d, stat); err != nil { return linux.Statx{}, err } } + d.dirMu.Unlock() return stat, nil } diff --git a/pkg/sentry/fsimpl/verity/verity.go b/pkg/sentry/fsimpl/verity/verity.go index 87dabe038..5788c661f 100644 --- a/pkg/sentry/fsimpl/verity/verity.go +++ b/pkg/sentry/fsimpl/verity/verity.go @@ -19,6 +19,18 @@ // The verity file system is read-only, except for one case: when // allowRuntimeEnable is true, additional Merkle files can be generated using // the FS_IOC_ENABLE_VERITY ioctl. +// +// Lock order: +// +// filesystem.renameMu +// dentry.dirMu +// fileDescription.mu +// filesystem.verityMu +// dentry.hashMu +// +// Locking dentry.dirMu in multiple dentries requires that parent dentries are +// locked before child dentries, and that filesystem.renameMu is locked to +// stabilize this relationship. package verity import ( @@ -79,9 +91,6 @@ var ( // noCrashOnVerificationFailure indicates whether the sandbox should panic // whenever verification fails. If true, an error is returned instead of // panicking. This should only be set for tests. - // - // TODO(b/165661693): Decide whether to panic or return error based on this - // flag. noCrashOnVerificationFailure bool // verityMu synchronizes concurrent operations that enable verity and perform @@ -372,12 +381,14 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return nil, nil, alertIntegrityViolation(fmt.Sprintf("Failed to deserialize childrenNames: %v", err)) } - if err := fs.verifyStatAndChildren(ctx, d, stat); err != nil { + if err := fs.verifyStatAndChildrenLocked(ctx, d, stat); err != nil { return nil, nil, err } } + d.hashMu.Lock() copy(d.hash, iopts.RootHash) + d.hashMu.Unlock() d.vfsd.Init(d) fs.rootDentry = d @@ -402,7 +413,8 @@ type dentry struct { fs *filesystem // mode, uid, gid and size are the file mode, owner, group, and size of - // the file in the underlying file system. + // the file in the underlying file system. They are set when a dentry + // is initialized, and never modified. mode uint32 uid uint32 gid uint32 @@ -425,18 +437,22 @@ type dentry struct { // childrenNames stores the name of all children of the dentry. This is // used by verity to check whether a child is expected. This is only - // populated by enableVerity. + // populated by enableVerity. childrenNames is also protected by dirMu. childrenNames map[string]struct{} - // lowerVD is the VirtualDentry in the underlying file system. + // lowerVD is the VirtualDentry in the underlying file system. It is + // never modified after initialized. lowerVD vfs.VirtualDentry // lowerMerkleVD is the VirtualDentry of the corresponding Merkle tree - // in the underlying file system. + // in the underlying file system. It is never modified after + // initialized. lowerMerkleVD vfs.VirtualDentry - // hash is the calculated hash for the current file or directory. - hash []byte + // hash is the calculated hash for the current file or directory. hash + // is protected by hashMu. + hashMu sync.RWMutex `state:"nosave"` + hash []byte } // newDentry creates a new dentry representing the given verity file. The @@ -519,7 +535,9 @@ func (d *dentry) checkDropLocked(ctx context.Context) { // destroyLocked destroys the dentry. // -// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. +// Preconditions: +// * d.fs.renameMu must be locked for writing. +// * d.refs == 0. func (d *dentry) destroyLocked(ctx context.Context) { switch atomic.LoadInt64(&d.refs) { case 0: @@ -599,6 +617,8 @@ func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) // mode, it returns true if the target has been enabled with // ioctl(FS_IOC_ENABLE_VERITY). func (d *dentry) verityEnabled() bool { + d.hashMu.RLock() + defer d.hashMu.RUnlock() return !d.fs.allowRuntimeEnable || len(d.hash) != 0 } @@ -678,11 +698,13 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu if err != nil { return linux.Statx{}, err } + fd.d.dirMu.Lock() if fd.d.verityEnabled() { - if err := fd.d.fs.verifyStatAndChildren(ctx, fd.d, stat); err != nil { + if err := fd.d.fs.verifyStatAndChildrenLocked(ctx, fd.d, stat); err != nil { return linux.Statx{}, err } } + fd.d.dirMu.Unlock() return stat, nil } @@ -718,22 +740,24 @@ func (fd *fileDescription) Seek(ctx context.Context, offset int64, whence int32) return offset, nil } -// generateMerkle generates a Merkle tree file for fd. If fd points to a file -// /foo/bar, a Merkle tree file /foo/.merkle.verity.bar is generated. The hash -// of the generated Merkle tree and the data size is returned. If fd points to -// a regular file, the data is the content of the file. If fd points to a -// directory, the data is all hahes of its children, written to the Merkle tree -// file. -func (fd *fileDescription) generateMerkle(ctx context.Context) ([]byte, uint64, error) { - fdReader := vfs.FileReadWriteSeeker{ +// generateMerkleLocked generates a Merkle tree file for fd. If fd points to a +// file /foo/bar, a Merkle tree file /foo/.merkle.verity.bar is generated. The +// hash of the generated Merkle tree and the data size is returned. If fd +// points to a regular file, the data is the content of the file. If fd points +// to a directory, the data is all hahes of its children, written to the Merkle +// tree file. +// +// Preconditions: fd.d.fs.verityMu must be locked. +func (fd *fileDescription) generateMerkleLocked(ctx context.Context) ([]byte, uint64, error) { + fdReader := FileReadWriteSeeker{ FD: fd.lowerFD, Ctx: ctx, } - merkleReader := vfs.FileReadWriteSeeker{ + merkleReader := FileReadWriteSeeker{ FD: fd.merkleReader, Ctx: ctx, } - merkleWriter := vfs.FileReadWriteSeeker{ + merkleWriter := FileReadWriteSeeker{ FD: fd.merkleWriter, Ctx: ctx, } @@ -793,11 +817,14 @@ func (fd *fileDescription) generateMerkle(ctx context.Context) ([]byte, uint64, return hash, uint64(params.Size), err } -// recordChildren writes the names of fd's children into the corresponding -// Merkle tree file, and saves the offset/size of the map into xattrs. +// recordChildrenLocked writes the names of fd's children into the +// corresponding Merkle tree file, and saves the offset/size of the map into +// xattrs. // -// Preconditions: fd.d.isDir() == true -func (fd *fileDescription) recordChildren(ctx context.Context) error { +// Preconditions: +// * fd.d.fs.verityMu must be locked. +// * fd.d.isDir() == true. +func (fd *fileDescription) recordChildrenLocked(ctx context.Context) error { // Record the children names in the Merkle tree file. childrenNames, err := json.Marshal(fd.d.childrenNames) if err != nil { @@ -847,7 +874,7 @@ func (fd *fileDescription) enableVerity(ctx context.Context) (uintptr, error) { return 0, alertIntegrityViolation("Unexpected verity fd: missing expected underlying fds") } - hash, dataSize, err := fd.generateMerkle(ctx) + hash, dataSize, err := fd.generateMerkleLocked(ctx) if err != nil { return 0, err } @@ -888,11 +915,13 @@ func (fd *fileDescription) enableVerity(ctx context.Context) (uintptr, error) { } if fd.d.isDir() { - if err := fd.recordChildren(ctx); err != nil { + if err := fd.recordChildrenLocked(ctx); err != nil { return 0, err } } - fd.d.hash = append(fd.d.hash, hash...) + fd.d.hashMu.Lock() + fd.d.hash = hash + fd.d.hashMu.Unlock() return 0, nil } @@ -904,6 +933,9 @@ func (fd *fileDescription) measureVerity(ctx context.Context, verityDigest userm } var metadata linux.DigestMetadata + fd.d.hashMu.RLock() + defer fd.d.hashMu.RUnlock() + // If allowRuntimeEnable is true, an empty fd.d.hash indicates that // verity is not enabled for the file. If allowRuntimeEnable is false, // this is an integrity violation because all files should have verity @@ -940,11 +972,13 @@ func (fd *fileDescription) measureVerity(ctx context.Context, verityDigest userm func (fd *fileDescription) verityFlags(ctx context.Context, flags usermem.Addr) (uintptr, error) { f := int32(0) + fd.d.hashMu.RLock() // All enabled files should store a hash. This flag is not settable via // FS_IOC_SETFLAGS. if len(fd.d.hash) != 0 { f |= linux.FS_VERITY_FL } + fd.d.hashMu.RUnlock() t := kernel.TaskFromContext(ctx) if t == nil { @@ -1013,16 +1047,17 @@ func (fd *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, of return 0, alertIntegrityViolation(fmt.Sprintf("Failed to convert xattr %s to int: %v", merkleSizeXattr, err)) } - dataReader := vfs.FileReadWriteSeeker{ + dataReader := FileReadWriteSeeker{ FD: fd.lowerFD, Ctx: ctx, } - merkleReader := vfs.FileReadWriteSeeker{ + merkleReader := FileReadWriteSeeker{ FD: fd.merkleReader, Ctx: ctx, } + fd.d.hashMu.RLock() n, err := merkletree.Verify(&merkletree.VerifyParams{ Out: dst.Writer(ctx), File: &dataReader, @@ -1040,6 +1075,7 @@ func (fd *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, of Expected: fd.d.hash, DataAndTreeInSameFile: false, }) + fd.d.hashMu.RUnlock() if err != nil { return 0, alertIntegrityViolation(fmt.Sprintf("Verification failed: %v", err)) } @@ -1065,3 +1101,45 @@ func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { return fd.lowerFD.UnlockPOSIX(ctx, uid, start, length, whence) } + +// FileReadWriteSeeker is a helper struct to pass a vfs.FileDescription as +// io.Reader/io.Writer/io.ReadSeeker/io.ReaderAt/io.WriterAt/etc. +type FileReadWriteSeeker struct { + FD *vfs.FileDescription + Ctx context.Context + ROpts vfs.ReadOptions + WOpts vfs.WriteOptions +} + +// ReadAt implements io.ReaderAt.ReadAt. +func (f *FileReadWriteSeeker) ReadAt(p []byte, off int64) (int, error) { + dst := usermem.BytesIOSequence(p) + n, err := f.FD.PRead(f.Ctx, dst, off, f.ROpts) + return int(n), err +} + +// Read implements io.ReadWriteSeeker.Read. +func (f *FileReadWriteSeeker) Read(p []byte) (int, error) { + dst := usermem.BytesIOSequence(p) + n, err := f.FD.Read(f.Ctx, dst, f.ROpts) + return int(n), err +} + +// Seek implements io.ReadWriteSeeker.Seek. +func (f *FileReadWriteSeeker) Seek(offset int64, whence int) (int64, error) { + return f.FD.Seek(f.Ctx, offset, int32(whence)) +} + +// WriteAt implements io.WriterAt.WriteAt. +func (f *FileReadWriteSeeker) WriteAt(p []byte, off int64) (int, error) { + dst := usermem.BytesIOSequence(p) + n, err := f.FD.PWrite(f.Ctx, dst, off, f.WOpts) + return int(n), err +} + +// Write implements io.ReadWriteSeeker.Write. +func (f *FileReadWriteSeeker) Write(p []byte) (int, error) { + buf := usermem.BytesIOSequence(p) + n, err := f.FD.Write(f.Ctx, buf, f.WOpts) + return int(n), err +} diff --git a/pkg/sentry/fsimpl/verity/verity_test.go b/pkg/sentry/fsimpl/verity/verity_test.go index 7196e74eb..6ced0afc9 100644 --- a/pkg/sentry/fsimpl/verity/verity_test.go +++ b/pkg/sentry/fsimpl/verity/verity_test.go @@ -35,16 +35,39 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) -// rootMerkleFilename is the name of the root Merkle tree file. -const rootMerkleFilename = "root.verity" +const ( + // rootMerkleFilename is the name of the root Merkle tree file. + rootMerkleFilename = "root.verity" + // maxDataSize is the maximum data size of a test file. + maxDataSize = 100000 +) + +var hashAlgs = []HashAlgorithm{SHA256, SHA512} -// maxDataSize is the maximum data size written to the file for test. -const maxDataSize = 100000 +func dentryFromVD(t *testing.T, vd vfs.VirtualDentry) *dentry { + t.Helper() + d, ok := vd.Dentry().Impl().(*dentry) + if !ok { + t.Fatalf("can't assert %T as a *dentry", vd) + } + return d +} + +// dentryFromFD returns the dentry corresponding to fd. +func dentryFromFD(t *testing.T, fd *vfs.FileDescription) *dentry { + t.Helper() + f, ok := fd.Impl().(*fileDescription) + if !ok { + t.Fatalf("can't assert %T as a *fileDescription", fd) + } + return f.d +} // newVerityRoot creates a new verity mount, and returns the root. The // underlying file system is tmpfs. If the error is not nil, then cleanup // should be called when the root is no longer needed. func newVerityRoot(t *testing.T, hashAlg HashAlgorithm) (*vfs.VirtualFilesystem, vfs.VirtualDentry, *kernel.Task, error) { + t.Helper() k, err := testutil.Boot() if err != nil { t.Fatalf("testutil.Boot: %v", err) @@ -92,7 +115,6 @@ func newVerityRoot(t *testing.T, hashAlg HashAlgorithm) (*vfs.VirtualFilesystem, t.Fatalf("testutil.CreateTask: %v", err) } - t.Helper() t.Cleanup(func() { root.DecRef(ctx) mntns.DecRef(ctx) @@ -100,21 +122,97 @@ func newVerityRoot(t *testing.T, hashAlg HashAlgorithm) (*vfs.VirtualFilesystem, return vfsObj, root, task, nil } -// newFileFD creates a new file in the verity mount, and returns the FD. The FD -// points to a file that has random data generated. -func newFileFD(ctx context.Context, vfsObj *vfs.VirtualFilesystem, root vfs.VirtualDentry, filePath string, mode linux.FileMode) (*vfs.FileDescription, int, error) { - creds := auth.CredentialsFromContext(ctx) - lowerRoot := root.Dentry().Impl().(*dentry).lowerVD +// openVerityAt opens a verity file. +// +// TODO(chongc): release reference from opening the file when done. +func openVerityAt(ctx context.Context, vfsObj *vfs.VirtualFilesystem, vd vfs.VirtualDentry, path string, flags uint32, mode linux.FileMode) (*vfs.FileDescription, error) { + return vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ + Root: vd, + Start: vd, + Path: fspath.Parse(path), + }, &vfs.OpenOptions{ + Flags: flags, + Mode: mode, + }) +} - // Create the file in the underlying file system. - lowerFD, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: lowerRoot, - Start: lowerRoot, - Path: fspath.Parse(filePath), +// openLowerAt opens the file in the underlying file system. +// +// TODO(chongc): release reference from opening the file when done. +func (d *dentry) openLowerAt(ctx context.Context, vfsObj *vfs.VirtualFilesystem, path string, flags uint32, mode linux.FileMode) (*vfs.FileDescription, error) { + return vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + Path: fspath.Parse(path), }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: linux.ModeRegular | mode, + Flags: flags, + Mode: mode, }) +} + +// openLowerMerkleAt opens the Merkle file in the underlying file system. +// +// TODO(chongc): release reference from opening the file when done. +func (d *dentry) openLowerMerkleAt(ctx context.Context, vfsObj *vfs.VirtualFilesystem, flags uint32, mode linux.FileMode) (*vfs.FileDescription, error) { + return vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ + Root: d.lowerMerkleVD, + Start: d.lowerMerkleVD, + }, &vfs.OpenOptions{ + Flags: flags, + Mode: mode, + }) +} + +// unlinkLowerAt deletes the file in the underlying file system. +func (d *dentry) unlinkLowerAt(ctx context.Context, vfsObj *vfs.VirtualFilesystem, path string) error { + return vfsObj.UnlinkAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + Path: fspath.Parse(path), + }) +} + +// unlinkLowerMerkleAt deletes the Merkle file in the underlying file system. +func (d *dentry) unlinkLowerMerkleAt(ctx context.Context, vfsObj *vfs.VirtualFilesystem, path string) error { + return vfsObj.UnlinkAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + Path: fspath.Parse(merklePrefix + path), + }) +} + +// renameLowerAt renames file name to newName in the underlying file system. +func (d *dentry) renameLowerAt(ctx context.Context, vfsObj *vfs.VirtualFilesystem, name string, newName string) error { + return vfsObj.RenameAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + Path: fspath.Parse(name), + }, &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + Path: fspath.Parse(newName), + }, &vfs.RenameOptions{}) +} + +// renameLowerMerkleAt renames Merkle file name to newName in the underlying +// file system. +func (d *dentry) renameLowerMerkleAt(ctx context.Context, vfsObj *vfs.VirtualFilesystem, name string, newName string) error { + return vfsObj.RenameAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + Path: fspath.Parse(merklePrefix + name), + }, &vfs.PathOperation{ + Root: d.lowerVD, + Start: d.lowerVD, + Path: fspath.Parse(merklePrefix + newName), + }, &vfs.RenameOptions{}) +} + +// newFileFD creates a new file in the verity mount, and returns the FD. The FD +// points to a file that has random data generated. +func newFileFD(ctx context.Context, t *testing.T, vfsObj *vfs.VirtualFilesystem, root vfs.VirtualDentry, filePath string, mode linux.FileMode) (*vfs.FileDescription, int, error) { + // Create the file in the underlying file system. + lowerFD, err := dentryFromVD(t, root).openLowerAt(ctx, vfsObj, filePath, linux.O_RDWR|linux.O_CREAT|linux.O_EXCL, linux.ModeRegular|mode) if err != nil { return nil, 0, err } @@ -137,20 +235,12 @@ func newFileFD(ctx context.Context, vfsObj *vfs.VirtualFilesystem, root vfs.Virt lowerFD.DecRef(ctx) // Now open the verity file descriptor. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filePath), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - Mode: linux.ModeRegular | mode, - }) + fd, err := openVerityAt(ctx, vfsObj, root, filePath, linux.O_RDONLY, mode) return fd, dataSize, err } -// corruptRandomBit randomly flips a bit in the file represented by fd. -func corruptRandomBit(ctx context.Context, fd *vfs.FileDescription, size int) error { - // Flip a random bit in the underlying file. +// flipRandomBit randomly flips a bit in the file represented by fd. +func flipRandomBit(ctx context.Context, fd *vfs.FileDescription, size int) error { randomPos := int64(rand.Intn(size)) byteToModify := make([]byte, 1) if _, err := fd.PRead(ctx, usermem.BytesIOSequence(byteToModify), randomPos, vfs.ReadOptions{}); err != nil { @@ -163,7 +253,14 @@ func corruptRandomBit(ctx context.Context, fd *vfs.FileDescription, size int) er return nil } -var hashAlgs = []HashAlgorithm{SHA256, SHA512} +func enableVerity(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { + t.Helper() + var args arch.SyscallArguments + args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} + if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { + t.Fatalf("enable verity: %v", err) + } +} // TestOpen ensures that when a file is created, the corresponding Merkle tree // file and the root Merkle tree file exist. @@ -175,30 +272,18 @@ func TestOpen(t *testing.T) { } filename := "verity-test-file" - if _, _, err := newFileFD(ctx, vfsObj, root, filename, 0644); err != nil { + fd, _, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) + if err != nil { t.Fatalf("newFileFD: %v", err) } // Ensure that the corresponding Merkle tree file is created. - lowerRoot := root.Dentry().Impl().(*dentry).lowerVD - if _, err = vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: lowerRoot, - Start: lowerRoot, - Path: fspath.Parse(merklePrefix + filename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }); err != nil { + if _, err = dentryFromFD(t, fd).openLowerMerkleAt(ctx, vfsObj, linux.O_RDONLY, linux.ModeRegular); err != nil { t.Errorf("OpenAt Merkle tree file %s: %v", merklePrefix+filename, err) } // Ensure the root merkle tree file is created. - if _, err = vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: lowerRoot, - Start: lowerRoot, - Path: fspath.Parse(merklePrefix + rootMerkleFilename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }); err != nil { + if _, err = dentryFromVD(t, root).openLowerMerkleAt(ctx, vfsObj, linux.O_RDONLY, linux.ModeRegular); err != nil { t.Errorf("OpenAt root Merkle tree file %s: %v", merklePrefix+rootMerkleFilename, err) } } @@ -214,17 +299,13 @@ func TestPReadUnmodifiedFileSucceeds(t *testing.T) { } filename := "verity-test-file" - fd, size, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, size, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file and confirm a normal read succeeds. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) buf := make([]byte, size) n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0 /* offset */, vfs.ReadOptions{}) @@ -248,17 +329,13 @@ func TestReadUnmodifiedFileSucceeds(t *testing.T) { } filename := "verity-test-file" - fd, size, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, size, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file and confirm a normal read succeeds. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) buf := make([]byte, size) n, err := fd.Read(ctx, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) @@ -282,27 +359,16 @@ func TestReopenUnmodifiedFileSucceeds(t *testing.T) { } filename := "verity-test-file" - fd, _, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, _, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file and confirms a normal read succeeds. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) // Ensure reopening the verity enabled file succeeds. - if _, err = vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - Mode: linux.ModeRegular, - }); err != nil { + if _, err = openVerityAt(ctx, vfsObj, root, filename, linux.O_RDONLY, linux.ModeRegular); err != nil { t.Errorf("reopen enabled file failed: %v", err) } } @@ -317,43 +383,24 @@ func TestOpenNonexistentFile(t *testing.T) { } filename := "verity-test-file" - fd, _, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, _, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file and confirms a normal read succeeds. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) // Enable verity on the parent directory. - parentFD, err := vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: root, - Start: root, - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }) + parentFD, err := openVerityAt(ctx, vfsObj, root, "", linux.O_RDONLY, linux.ModeRegular) if err != nil { t.Fatalf("OpenAt: %v", err) } - - if _, err := parentFD.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, parentFD) // Ensure open an unexpected file in the parent directory fails with // ENOENT rather than verification failure. - if _, err = vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename + "abc"), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - Mode: linux.ModeRegular, - }); err != syserror.ENOENT { + if _, err = openVerityAt(ctx, vfsObj, root, filename+"abc", linux.O_RDONLY, linux.ModeRegular); err != syserror.ENOENT { t.Errorf("OpenAt unexpected error: %v", err) } } @@ -368,33 +415,22 @@ func TestPReadModifiedFileFails(t *testing.T) { } filename := "verity-test-file" - fd, size, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, size, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) // Open a new lowerFD that's read/writable. - lowerVD := fd.Impl().(*fileDescription).d.lowerVD - - lowerFD, err := vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: lowerVD, - Start: lowerVD, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR, - }) + lowerFD, err := dentryFromFD(t, fd).openLowerAt(ctx, vfsObj, "", linux.O_RDWR, linux.ModeRegular) if err != nil { t.Fatalf("OpenAt: %v", err) } - if err := corruptRandomBit(ctx, lowerFD, size); err != nil { - t.Fatalf("corruptRandomBit: %v", err) + if err := flipRandomBit(ctx, lowerFD, size); err != nil { + t.Fatalf("flipRandomBit: %v", err) } // Confirm that read from the modified file fails. @@ -415,33 +451,22 @@ func TestReadModifiedFileFails(t *testing.T) { } filename := "verity-test-file" - fd, size, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, size, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) // Open a new lowerFD that's read/writable. - lowerVD := fd.Impl().(*fileDescription).d.lowerVD - - lowerFD, err := vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: lowerVD, - Start: lowerVD, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR, - }) + lowerFD, err := dentryFromFD(t, fd).openLowerAt(ctx, vfsObj, "", linux.O_RDWR, linux.ModeRegular) if err != nil { t.Fatalf("OpenAt: %v", err) } - if err := corruptRandomBit(ctx, lowerFD, size); err != nil { - t.Fatalf("corruptRandomBit: %v", err) + if err := flipRandomBit(ctx, lowerFD, size); err != nil { + t.Fatalf("flipRandomBit: %v", err) } // Confirm that read from the modified file fails. @@ -462,27 +487,16 @@ func TestModifiedMerkleFails(t *testing.T) { } filename := "verity-test-file" - fd, size, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, size, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) // Open a new lowerMerkleFD that's read/writable. - lowerMerkleVD := fd.Impl().(*fileDescription).d.lowerMerkleVD - - lowerMerkleFD, err := vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: lowerMerkleVD, - Start: lowerMerkleVD, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR, - }) + lowerMerkleFD, err := dentryFromFD(t, fd).openLowerMerkleAt(ctx, vfsObj, linux.O_RDWR, linux.ModeRegular) if err != nil { t.Fatalf("OpenAt: %v", err) } @@ -493,14 +507,13 @@ func TestModifiedMerkleFails(t *testing.T) { t.Errorf("lowerMerkleFD.Stat: %v", err) } - if err := corruptRandomBit(ctx, lowerMerkleFD, int(stat.Size)); err != nil { - t.Fatalf("corruptRandomBit: %v", err) + if err := flipRandomBit(ctx, lowerMerkleFD, int(stat.Size)); err != nil { + t.Fatalf("flipRandomBit: %v", err) } // Confirm that read from a file with modified Merkle tree fails. buf := make([]byte, size) if _, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0 /* offset */, vfs.ReadOptions{}); err == nil { - fmt.Println(buf) t.Fatalf("fd.PRead succeeded with modified Merkle file") } } @@ -517,42 +530,23 @@ func TestModifiedParentMerkleFails(t *testing.T) { } filename := "verity-test-file" - fd, _, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, _, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) // Enable verity on the parent directory. - parentFD, err := vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: root, - Start: root, - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }) + parentFD, err := openVerityAt(ctx, vfsObj, root, "", linux.O_RDONLY, linux.ModeRegular) if err != nil { t.Fatalf("OpenAt: %v", err) } - - if _, err := parentFD.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, parentFD) // Open a new lowerMerkleFD that's read/writable. - parentLowerMerkleVD := fd.Impl().(*fileDescription).d.parent.lowerMerkleVD - - parentLowerMerkleFD, err := vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: parentLowerMerkleVD, - Start: parentLowerMerkleVD, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR, - }) + parentLowerMerkleFD, err := dentryFromFD(t, fd).parent.openLowerMerkleAt(ctx, vfsObj, linux.O_RDWR, linux.ModeRegular) if err != nil { t.Fatalf("OpenAt: %v", err) } @@ -572,21 +566,14 @@ func TestModifiedParentMerkleFails(t *testing.T) { if err != nil { t.Fatalf("Failed convert size to int: %v", err) } - if err := corruptRandomBit(ctx, parentLowerMerkleFD, parentMerkleSize); err != nil { - t.Fatalf("corruptRandomBit: %v", err) + if err := flipRandomBit(ctx, parentLowerMerkleFD, parentMerkleSize); err != nil { + t.Fatalf("flipRandomBit: %v", err) } parentLowerMerkleFD.DecRef(ctx) // Ensure reopening the verity enabled file fails. - if _, err = vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - Mode: linux.ModeRegular, - }); err == nil { + if _, err = openVerityAt(ctx, vfsObj, root, filename, linux.O_RDONLY, linux.ModeRegular); err == nil { t.Errorf("OpenAt file with modified parent Merkle succeeded") } } @@ -602,18 +589,13 @@ func TestUnmodifiedStatSucceeds(t *testing.T) { } filename := "verity-test-file" - fd, _, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, _, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } - // Enable verity on the file and confirms stat succeeds. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("fd.Ioctl: %v", err) - } - + // Enable verity on the file and confirm that stat succeeds. + enableVerity(ctx, t, fd) if _, err := fd.Stat(ctx, vfs.StatOptions{}); err != nil { t.Errorf("fd.Stat: %v", err) } @@ -630,17 +612,13 @@ func TestModifiedStatFails(t *testing.T) { } filename := "verity-test-file" - fd, _, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, _, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("fd.Ioctl: %v", err) - } + enableVerity(ctx, t, fd) lowerFD := fd.Impl().(*fileDescription).lowerFD // Change the stat of the underlying file, and check that stat fails. @@ -663,73 +641,57 @@ func TestModifiedStatFails(t *testing.T) { // and/or the corresponding Merkle tree file fails with the verity error. func TestOpenDeletedFileFails(t *testing.T) { testCases := []struct { + name string // The original file is removed if changeFile is true. changeFile bool // The Merkle tree file is removed if changeMerkleFile is true. changeMerkleFile bool }{ { + name: "FileOnly", changeFile: true, changeMerkleFile: false, }, { + name: "MerkleOnly", changeFile: false, changeMerkleFile: true, }, { + name: "FileAndMerkle", changeFile: true, changeMerkleFile: true, }, } for _, tc := range testCases { - t.Run(fmt.Sprintf("changeFile:%t, changeMerkleFile:%t", tc.changeFile, tc.changeMerkleFile), func(t *testing.T) { + t.Run(tc.name, func(t *testing.T) { vfsObj, root, ctx, err := newVerityRoot(t, SHA256) if err != nil { t.Fatalf("newVerityRoot: %v", err) } filename := "verity-test-file" - fd, _, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, _, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) - rootLowerVD := root.Dentry().Impl().(*dentry).lowerVD if tc.changeFile { - if err := vfsObj.UnlinkAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: rootLowerVD, - Start: rootLowerVD, - Path: fspath.Parse(filename), - }); err != nil { + if err := dentryFromVD(t, root).unlinkLowerAt(ctx, vfsObj, filename); err != nil { t.Fatalf("UnlinkAt: %v", err) } } if tc.changeMerkleFile { - if err := vfsObj.UnlinkAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: rootLowerVD, - Start: rootLowerVD, - Path: fspath.Parse(merklePrefix + filename), - }); err != nil { + if err := dentryFromVD(t, root).unlinkLowerMerkleAt(ctx, vfsObj, filename); err != nil { t.Fatalf("UnlinkAt: %v", err) } } // Ensure reopening the verity enabled file fails. - if _, err = vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - Mode: linux.ModeRegular, - }); err != syserror.EIO { + if _, err = openVerityAt(ctx, vfsObj, root, filename, linux.O_RDONLY, linux.ModeRegular); err != syserror.EIO { t.Errorf("got OpenAt error: %v, expected EIO", err) } }) @@ -740,82 +702,58 @@ func TestOpenDeletedFileFails(t *testing.T) { // and/or the corresponding Merkle tree file fails with the verity error. func TestOpenRenamedFileFails(t *testing.T) { testCases := []struct { + name string // The original file is renamed if changeFile is true. changeFile bool // The Merkle tree file is renamed if changeMerkleFile is true. changeMerkleFile bool }{ { + name: "FileOnly", changeFile: true, changeMerkleFile: false, }, { + name: "MerkleOnly", changeFile: false, changeMerkleFile: true, }, { + name: "FileAndMerkle", changeFile: true, changeMerkleFile: true, }, } for _, tc := range testCases { - t.Run(fmt.Sprintf("changeFile:%t, changeMerkleFile:%t", tc.changeFile, tc.changeMerkleFile), func(t *testing.T) { + t.Run(tc.name, func(t *testing.T) { vfsObj, root, ctx, err := newVerityRoot(t, SHA256) if err != nil { t.Fatalf("newVerityRoot: %v", err) } filename := "verity-test-file" - fd, _, err := newFileFD(ctx, vfsObj, root, filename, 0644) + fd, _, err := newFileFD(ctx, t, vfsObj, root, filename, 0644) if err != nil { t.Fatalf("newFileFD: %v", err) } // Enable verity on the file. - var args arch.SyscallArguments - args[1] = arch.SyscallArgument{Value: linux.FS_IOC_ENABLE_VERITY} - if _, err := fd.Ioctl(ctx, nil /* uio */, args); err != nil { - t.Fatalf("Ioctl: %v", err) - } + enableVerity(ctx, t, fd) - rootLowerVD := root.Dentry().Impl().(*dentry).lowerVD newFilename := "renamed-test-file" if tc.changeFile { - if err := vfsObj.RenameAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: rootLowerVD, - Start: rootLowerVD, - Path: fspath.Parse(filename), - }, &vfs.PathOperation{ - Root: rootLowerVD, - Start: rootLowerVD, - Path: fspath.Parse(newFilename), - }, &vfs.RenameOptions{}); err != nil { + if err := dentryFromVD(t, root).renameLowerAt(ctx, vfsObj, filename, newFilename); err != nil { t.Fatalf("RenameAt: %v", err) } } if tc.changeMerkleFile { - if err := vfsObj.RenameAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: rootLowerVD, - Start: rootLowerVD, - Path: fspath.Parse(merklePrefix + filename), - }, &vfs.PathOperation{ - Root: rootLowerVD, - Start: rootLowerVD, - Path: fspath.Parse(merklePrefix + newFilename), - }, &vfs.RenameOptions{}); err != nil { + if err := dentryFromVD(t, root).renameLowerMerkleAt(ctx, vfsObj, filename, newFilename); err != nil { t.Fatalf("UnlinkAt: %v", err) } } // Ensure reopening the verity enabled file fails. - if _, err = vfsObj.OpenAt(ctx, auth.CredentialsFromContext(ctx), &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - Mode: linux.ModeRegular, - }); err != syserror.EIO { + if _, err = openVerityAt(ctx, vfsObj, root, filename, linux.O_RDONLY, linux.ModeRegular); err != syserror.EIO { t.Errorf("got OpenAt error: %v, expected EIO", err) } }) diff --git a/pkg/sentry/fsmetric/BUILD b/pkg/sentry/fsmetric/BUILD new file mode 100644 index 000000000..4e86fbdd8 --- /dev/null +++ b/pkg/sentry/fsmetric/BUILD @@ -0,0 +1,10 @@ +load("//tools:defs.bzl", "go_library") + +licenses(["notice"]) + +go_library( + name = "fsmetric", + srcs = ["fsmetric.go"], + visibility = ["//pkg/sentry:internal"], + deps = ["//pkg/metric"], +) diff --git a/pkg/sentry/fsmetric/fsmetric.go b/pkg/sentry/fsmetric/fsmetric.go new file mode 100644 index 000000000..7e535b527 --- /dev/null +++ b/pkg/sentry/fsmetric/fsmetric.go @@ -0,0 +1,83 @@ +// 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 fsmetric defines filesystem metrics that are used by both VFS1 and +// VFS2. +// +// TODO(gvisor.dev/issue/1624): Once VFS1 is deleted, inline these metrics into +// VFS2. +package fsmetric + +import ( + "time" + + "gvisor.dev/gvisor/pkg/metric" +) + +// RecordWaitTime enables the ReadWait, GoferReadWait9P, GoferReadWaitHost, and +// TmpfsReadWait metrics. Enabling this comes at a CPU cost due to performing +// three clock reads per read call. +// +// Note that this is only performed in the direct read path, and may not be +// consistently applied for other forms of reads, such as splice. +var RecordWaitTime = false + +// Metrics that apply to all filesystems. +var ( + Opens = metric.MustCreateNewUint64Metric("/fs/opens", false /* sync */, "Number of file opens.") + Reads = metric.MustCreateNewUint64Metric("/fs/reads", false /* sync */, "Number of file reads.") + ReadWait = metric.MustCreateNewUint64NanosecondsMetric("/fs/read_wait", false /* sync */, "Time waiting on file reads, in nanoseconds.") +) + +// Metrics that only apply to fs/gofer and fsimpl/gofer. +var ( + GoferOpensWX = metric.MustCreateNewUint64Metric("/gofer/opened_write_execute_file", true /* sync */, "Number of times a executable file was opened writably from a gofer.") + GoferOpens9P = metric.MustCreateNewUint64Metric("/gofer/opens_9p", false /* sync */, "Number of times a file was opened from a gofer and did not have a host file descriptor.") + GoferOpensHost = metric.MustCreateNewUint64Metric("/gofer/opens_host", false /* sync */, "Number of times a file was opened from a gofer and did have a host file descriptor.") + GoferReads9P = metric.MustCreateNewUint64Metric("/gofer/reads_9p", false /* sync */, "Number of 9P file reads from a gofer.") + GoferReadWait9P = metric.MustCreateNewUint64NanosecondsMetric("/gofer/read_wait_9p", false /* sync */, "Time waiting on 9P file reads from a gofer, in nanoseconds.") + GoferReadsHost = metric.MustCreateNewUint64Metric("/gofer/reads_host", false /* sync */, "Number of host file reads from a gofer.") + GoferReadWaitHost = metric.MustCreateNewUint64NanosecondsMetric("/gofer/read_wait_host", false /* sync */, "Time waiting on host file reads from a gofer, in nanoseconds.") +) + +// Metrics that only apply to fs/tmpfs and fsimpl/tmpfs. +var ( + TmpfsOpensRO = metric.MustCreateNewUint64Metric("/in_memory_file/opens_ro", false /* sync */, "Number of times an in-memory file was opened in read-only mode.") + TmpfsOpensW = metric.MustCreateNewUint64Metric("/in_memory_file/opens_w", false /* sync */, "Number of times an in-memory file was opened in write mode.") + TmpfsReads = metric.MustCreateNewUint64Metric("/in_memory_file/reads", false /* sync */, "Number of in-memory file reads.") + TmpfsReadWait = metric.MustCreateNewUint64NanosecondsMetric("/in_memory_file/read_wait", false /* sync */, "Time waiting on in-memory file reads, in nanoseconds.") +) + +// StartReadWait indicates the beginning of a file read. +func StartReadWait() time.Time { + if !RecordWaitTime { + return time.Time{} + } + return time.Now() +} + +// FinishReadWait indicates the end of a file read whose time is accounted by +// m. start must be the value returned by the corresponding call to +// StartReadWait. +// +// FinishReadWait is marked nosplit for performance since it's often called +// from defer statements, which prevents it from being inlined +// (https://github.com/golang/go/issues/38471). +//go:nosplit +func FinishReadWait(m *metric.Uint64Metric, start time.Time) { + if !RecordWaitTime { + return + } + m.IncrementBy(uint64(time.Since(start).Nanoseconds())) +} diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 15519f0df..61aeca044 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -273,7 +273,7 @@ func (e *EventPoll) ReadEvents(max int) []linux.EpollEvent { // // Callback is called when one of the files we're polling becomes ready. It // moves said file to the readyList if it's currently in the waiting list. -func (p *pollEntry) Callback(*waiter.Entry) { +func (p *pollEntry) Callback(*waiter.Entry, waiter.EventMask) { e := p.epoll e.listsMu.Lock() @@ -306,9 +306,8 @@ func (e *EventPoll) initEntryReadiness(entry *pollEntry) { f.EventRegister(&entry.waiter, entry.mask) // Check if the file happens to already be in a ready state. - ready := f.Readiness(entry.mask) & entry.mask - if ready != 0 { - entry.Callback(&entry.waiter) + if ready := f.Readiness(entry.mask) & entry.mask; ready != 0 { + entry.Callback(&entry.waiter, ready) } } diff --git a/pkg/sentry/kernel/fasync/BUILD b/pkg/sentry/kernel/fasync/BUILD index 2b3955598..f855f038b 100644 --- a/pkg/sentry/kernel/fasync/BUILD +++ b/pkg/sentry/kernel/fasync/BUILD @@ -8,11 +8,13 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", + "//pkg/sentry/arch", "//pkg/sentry/fs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/vfs", "//pkg/sync", + "//pkg/syserror", "//pkg/waiter", ], ) diff --git a/pkg/sentry/kernel/fasync/fasync.go b/pkg/sentry/kernel/fasync/fasync.go index 153d2cd9b..b66d61c6f 100644 --- a/pkg/sentry/kernel/fasync/fasync.go +++ b/pkg/sentry/kernel/fasync/fasync.go @@ -17,22 +17,45 @@ package fasync import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) -// New creates a new fs.FileAsync. -func New() fs.FileAsync { - return &FileAsync{} +// Table to convert waiter event masks into si_band siginfo codes. +// Taken from fs/fcntl.c:band_table. +var bandTable = map[waiter.EventMask]int64{ + // POLL_IN + waiter.EventIn: linux.EPOLLIN | linux.EPOLLRDNORM, + // POLL_OUT + waiter.EventOut: linux.EPOLLOUT | linux.EPOLLWRNORM | linux.EPOLLWRBAND, + // POLL_ERR + waiter.EventErr: linux.EPOLLERR, + // POLL_PRI + waiter.EventPri: linux.EPOLLPRI | linux.EPOLLRDBAND, + // POLL_HUP + waiter.EventHUp: linux.EPOLLHUP | linux.EPOLLERR, } -// NewVFS2 creates a new vfs.FileAsync. -func NewVFS2() vfs.FileAsync { - return &FileAsync{} +// New returns a function that creates a new fs.FileAsync with the given file +// descriptor. +func New(fd int) func() fs.FileAsync { + return func() fs.FileAsync { + return &FileAsync{fd: fd} + } +} + +// NewVFS2 returns a function that creates a new vfs.FileAsync with the given +// file descriptor. +func NewVFS2(fd int) func() vfs.FileAsync { + return func() vfs.FileAsync { + return &FileAsync{fd: fd} + } } // FileAsync sends signals when the registered file is ready for IO. @@ -42,6 +65,12 @@ type FileAsync struct { // e is immutable after first use (which is protected by mu below). e waiter.Entry + // fd is the file descriptor to notify about. + // It is immutable, set at allocation time. This matches Linux semantics in + // fs/fcntl.c:fasync_helper. + // The fd value is passed to the signal recipient in siginfo.si_fd. + fd int + // regMu protects registeration and unregistration actions on e. // // regMu must be held while registration decisions are being made @@ -56,6 +85,10 @@ type FileAsync struct { mu sync.Mutex `state:"nosave"` requester *auth.Credentials registered bool + // signal is the signal to deliver upon I/O being available. + // The default value ("zero signal") means the default SIGIO signal will be + // delivered. + signal linux.Signal // Only one of the following is allowed to be non-nil. recipientPG *kernel.ProcessGroup @@ -64,10 +97,10 @@ type FileAsync struct { } // Callback sends a signal. -func (a *FileAsync) Callback(e *waiter.Entry) { +func (a *FileAsync) Callback(e *waiter.Entry, mask waiter.EventMask) { a.mu.Lock() + defer a.mu.Unlock() if !a.registered { - a.mu.Unlock() return } t := a.recipientT @@ -80,19 +113,34 @@ func (a *FileAsync) Callback(e *waiter.Entry) { } if t == nil { // No recipient has been registered. - a.mu.Unlock() return } c := t.Credentials() // Logic from sigio_perm in fs/fcntl.c. - if a.requester.EffectiveKUID == 0 || + permCheck := (a.requester.EffectiveKUID == 0 || a.requester.EffectiveKUID == c.SavedKUID || a.requester.EffectiveKUID == c.RealKUID || a.requester.RealKUID == c.SavedKUID || - a.requester.RealKUID == c.RealKUID { - t.SendSignal(kernel.SignalInfoPriv(linux.SIGIO)) + a.requester.RealKUID == c.RealKUID) + if !permCheck { + return } - a.mu.Unlock() + signalInfo := &arch.SignalInfo{ + Signo: int32(linux.SIGIO), + Code: arch.SignalInfoKernel, + } + if a.signal != 0 { + signalInfo.Signo = int32(a.signal) + signalInfo.SetFD(uint32(a.fd)) + var band int64 + for m, bandCode := range bandTable { + if m&mask != 0 { + band |= bandCode + } + } + signalInfo.SetBand(band) + } + t.SendSignal(signalInfo) } // Register sets the file which will be monitored for IO events. @@ -186,3 +234,25 @@ func (a *FileAsync) ClearOwner() { a.recipientTG = nil a.recipientPG = nil } + +// Signal returns which signal will be sent to the signal recipient. +// A value of zero means the signal to deliver wasn't customized, which means +// the default signal (SIGIO) will be delivered. +func (a *FileAsync) Signal() linux.Signal { + a.mu.Lock() + defer a.mu.Unlock() + return a.signal +} + +// SetSignal overrides which signal to send when I/O is available. +// The default behavior can be reset by specifying signal zero, which means +// to send SIGIO. +func (a *FileAsync) SetSignal(signal linux.Signal) error { + if signal != 0 && !signal.IsValid() { + return syserror.EINVAL + } + a.mu.Lock() + defer a.mu.Unlock() + a.signal = signal + return nil +} diff --git a/pkg/sentry/kernel/fd_table_unsafe.go b/pkg/sentry/kernel/fd_table_unsafe.go index 470d8bf83..f17f9c59c 100644 --- a/pkg/sentry/kernel/fd_table_unsafe.go +++ b/pkg/sentry/kernel/fd_table_unsafe.go @@ -121,18 +121,21 @@ func (f *FDTable) setAll(ctx context.Context, fd int32, file *fs.File, fileVFS2 panic("VFS1 and VFS2 files set") } - slice := *(*[]unsafe.Pointer)(atomic.LoadPointer(&f.slice)) + slicePtr := (*[]unsafe.Pointer)(atomic.LoadPointer(&f.slice)) // Grow the table as required. - if last := int32(len(slice)); fd >= last { + if last := int32(len(*slicePtr)); fd >= last { end := fd + 1 if end < 2*last { end = 2 * last } - slice = append(slice, make([]unsafe.Pointer, end-last)...) - atomic.StorePointer(&f.slice, unsafe.Pointer(&slice)) + newSlice := append(*slicePtr, make([]unsafe.Pointer, end-last)...) + slicePtr = &newSlice + atomic.StorePointer(&f.slice, unsafe.Pointer(slicePtr)) } + slice := *slicePtr + var desc *descriptor if file != nil || fileVFS2 != nil { desc = &descriptor{ diff --git a/pkg/sentry/kernel/semaphore/semaphore.go b/pkg/sentry/kernel/semaphore/semaphore.go index b99c0bffa..31198d772 100644 --- a/pkg/sentry/kernel/semaphore/semaphore.go +++ b/pkg/sentry/kernel/semaphore/semaphore.go @@ -29,17 +29,17 @@ import ( ) const ( - valueMax = 32767 // SEMVMX + // Maximum semaphore value. + valueMax = linux.SEMVMX - // semaphoresMax is "maximum number of semaphores per semaphore ID" (SEMMSL). - semaphoresMax = 32000 + // Maximum number of semaphore sets. + setsMax = linux.SEMMNI - // setMax is "system-wide limit on the number of semaphore sets" (SEMMNI). - setsMax = 32000 + // Maximum number of semaphroes in a semaphore set. + semsMax = linux.SEMMSL - // semaphoresTotalMax is "system-wide limit on the number of semaphores" - // (SEMMNS = SEMMNI*SEMMSL). - semaphoresTotalMax = 1024000000 + // Maximum number of semaphores in all semaphroe sets. + semsTotalMax = linux.SEMMNS ) // Registry maintains a set of semaphores that can be found by key or ID. @@ -122,7 +122,7 @@ func NewRegistry(userNS *auth.UserNamespace) *Registry { // be found. If exclusive is true, it fails if a set with the same key already // exists. func (r *Registry) FindOrCreate(ctx context.Context, key, nsems int32, mode linux.FileMode, private, create, exclusive bool) (*Set, error) { - if nsems < 0 || nsems > semaphoresMax { + if nsems < 0 || nsems > semsMax { return nil, syserror.EINVAL } @@ -166,7 +166,7 @@ func (r *Registry) FindOrCreate(ctx context.Context, key, nsems int32, mode linu if len(r.semaphores) >= setsMax { return nil, syserror.EINVAL } - if r.totalSems() > int(semaphoresTotalMax-nsems) { + if r.totalSems() > int(semsTotalMax-nsems) { return nil, syserror.EINVAL } @@ -176,6 +176,22 @@ func (r *Registry) FindOrCreate(ctx context.Context, key, nsems int32, mode linu return r.newSet(ctx, key, owner, owner, perms, nsems) } +// IPCInfo returns information about system-wide semaphore limits and parameters. +func (r *Registry) IPCInfo() *linux.SemInfo { + return &linux.SemInfo{ + SemMap: linux.SEMMAP, + SemMni: linux.SEMMNI, + SemMns: linux.SEMMNS, + SemMnu: linux.SEMMNU, + SemMsl: linux.SEMMSL, + SemOpm: linux.SEMOPM, + SemUme: linux.SEMUME, + SemUsz: 0, // SemUsz not supported. + SemVmx: linux.SEMVMX, + SemAem: linux.SEMAEM, + } +} + // RemoveID removes set with give 'id' from the registry and marks the set as // dead. All waiters will be awakened and fail. func (r *Registry) RemoveID(id int32, creds *auth.Credentials) error { diff --git a/pkg/sentry/kernel/syslog.go b/pkg/sentry/kernel/syslog.go index a83ce219c..3fee7aa68 100644 --- a/pkg/sentry/kernel/syslog.go +++ b/pkg/sentry/kernel/syslog.go @@ -75,6 +75,12 @@ func (s *syslog) Log() []byte { "Checking naughty and nice process list...", // Check it up to twice. "Granting licence to kill(2)...", // British spelling for British movie. "Letting the watchdogs out...", + "Conjuring /dev/null black hole...", + "Adversarially training Redcode AI...", + "Singleplexing /dev/ptmx...", + "Recruiting cron-ies...", + "Verifying that no non-zero bytes made their way into /dev/zero...", + "Accelerating teletypewriter to 9600 baud...", } selectMessage := func() string { diff --git a/pkg/sentry/kernel/task_block.go b/pkg/sentry/kernel/task_block.go index 147a3d286..9419f2e95 100644 --- a/pkg/sentry/kernel/task_block.go +++ b/pkg/sentry/kernel/task_block.go @@ -177,19 +177,23 @@ func (t *Task) SleepStart() <-chan struct{} { // SleepFinish implements context.ChannelSleeper.SleepFinish. func (t *Task) SleepFinish(success bool) { if !success { - // The interrupted notification is consumed only at the top-level - // (Run). Therefore we attempt to reset the pending notification. - // This will also elide our next entry back into the task, so we - // will process signals, state changes, etc. + // Our caller received from t.interruptChan; we need to re-send to it + // to ensure that t.interrupted() is still true. t.interruptSelf() } t.accountTaskGoroutineLeave(TaskGoroutineBlockedInterruptible) t.Activate() } -// Interrupted implements amutex.Sleeper.Interrupted +// Interrupted implements context.ChannelSleeper.Interrupted. func (t *Task) Interrupted() bool { - return len(t.interruptChan) != 0 + if t.interrupted() { + return true + } + // Indicate that t's task goroutine is still responsive (i.e. reset the + // watchdog timer). + t.accountTaskGoroutineRunning() + return false } // UninterruptibleSleepStart implements context.Context.UninterruptibleSleepStart. @@ -210,13 +214,17 @@ func (t *Task) UninterruptibleSleepFinish(activate bool) { } // interrupted returns true if interrupt or interruptSelf has been called at -// least once since the last call to interrupted. +// least once since the last call to unsetInterrupted. func (t *Task) interrupted() bool { + return len(t.interruptChan) != 0 +} + +// unsetInterrupted causes interrupted to return false until the next call to +// interrupt or interruptSelf. +func (t *Task) unsetInterrupted() { select { case <-t.interruptChan: - return true default: - return false } } @@ -232,9 +240,7 @@ func (t *Task) interrupt() { func (t *Task) interruptSelf() { select { case t.interruptChan <- struct{}{}: - t.Debugf("Interrupt queued") default: - t.Debugf("Dropping duplicate interrupt") } // platform.Context.Interrupt() is unnecessary since a task goroutine // calling interruptSelf() cannot also be blocked in diff --git a/pkg/sentry/kernel/task_sched.go b/pkg/sentry/kernel/task_sched.go index 52c55d13d..9ba5f8d78 100644 --- a/pkg/sentry/kernel/task_sched.go +++ b/pkg/sentry/kernel/task_sched.go @@ -157,6 +157,18 @@ func (t *Task) accountTaskGoroutineLeave(state TaskGoroutineState) { t.goschedSeq.EndWrite() } +// Preconditions: The caller must be running on the task goroutine. +func (t *Task) accountTaskGoroutineRunning() { + now := t.k.CPUClockNow() + if t.gosched.State != TaskGoroutineRunningSys { + panic(fmt.Sprintf("Task goroutine in state %v (expected %v)", t.gosched.State, TaskGoroutineRunningSys)) + } + t.goschedSeq.BeginWrite() + t.gosched.SysTicks += now - t.gosched.Timestamp + t.gosched.Timestamp = now + t.goschedSeq.EndWrite() +} + // TaskGoroutineSchedInfo returns a copy of t's task goroutine scheduling info. // Most clients should use t.CPUStats() instead. func (t *Task) TaskGoroutineSchedInfo() TaskGoroutineSchedInfo { diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index ebdb83061..42dd3e278 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -619,9 +619,6 @@ func (t *Task) setSignalMaskLocked(mask linux.SignalSet) { return } }) - // We have to re-issue the interrupt consumed by t.interrupted() since - // it might have been for a different reason. - t.interruptSelf() } // Conversely, if the new mask unblocks any signals that were blocked by @@ -931,10 +928,10 @@ func (t *Task) signalStop(target *Task, code int32, status int32) { type runInterrupt struct{} func (*runInterrupt) execute(t *Task) taskRunState { - // Interrupts are de-duplicated (if t is interrupted twice before - // t.interrupted() is called, t.interrupted() will only return true once), - // so early exits from this function must re-enter the runInterrupt state - // to check for more interrupt-signaled conditions. + // Interrupts are de-duplicated (t.unsetInterrupted() will undo the effect + // of all previous calls to t.interrupted() regardless of how many such + // calls there have been), so early exits from this function must re-enter + // the runInterrupt state to check for more interrupt-signaled conditions. t.tg.signalHandlers.mu.Lock() @@ -1080,6 +1077,7 @@ func (*runInterrupt) execute(t *Task) taskRunState { return t.deliverSignal(info, act) } + t.unsetInterrupted() t.tg.signalHandlers.mu.Unlock() return (*runApp)(nil) } diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index 7c297fb9e..d99be7f46 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -423,11 +423,7 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (memmap.File } if f.opts.ManualZeroing { - if err := f.forEachMappingSlice(fr, func(bs []byte) { - for i := range bs { - bs[i] = 0 - } - }); err != nil { + if err := f.manuallyZero(fr); err != nil { return memmap.FileRange{}, err } } @@ -560,19 +556,39 @@ func (f *MemoryFile) Decommit(fr memmap.FileRange) error { panic(fmt.Sprintf("invalid range: %v", fr)) } + if f.opts.ManualZeroing { + // FALLOC_FL_PUNCH_HOLE may not zero pages if ManualZeroing is in + // effect. + if err := f.manuallyZero(fr); err != nil { + return err + } + } else { + if err := f.decommitFile(fr); err != nil { + return err + } + } + + f.markDecommitted(fr) + return nil +} + +func (f *MemoryFile) manuallyZero(fr memmap.FileRange) error { + return f.forEachMappingSlice(fr, func(bs []byte) { + for i := range bs { + bs[i] = 0 + } + }) +} + +func (f *MemoryFile) decommitFile(fr memmap.FileRange) error { // "After a successful call, subsequent reads from this range will // return zeroes. The FALLOC_FL_PUNCH_HOLE flag must be ORed with // FALLOC_FL_KEEP_SIZE in mode ..." - fallocate(2) - err := syscall.Fallocate( + return syscall.Fallocate( int(f.file.Fd()), _FALLOC_FL_PUNCH_HOLE|_FALLOC_FL_KEEP_SIZE, int64(fr.Start), int64(fr.Length())) - if err != nil { - return err - } - f.markDecommitted(fr) - return nil } func (f *MemoryFile) markDecommitted(fr memmap.FileRange) { @@ -1044,20 +1060,20 @@ func (f *MemoryFile) runReclaim() { break } - if err := f.Decommit(fr); err != nil { - log.Warningf("Reclaim failed to decommit %v: %v", fr, err) - // Zero the pages manually. This won't reduce memory usage, but at - // least ensures that the pages will be zero when reallocated. - f.forEachMappingSlice(fr, func(bs []byte) { - for i := range bs { - bs[i] = 0 + // If ManualZeroing is in effect, pages will be zeroed on allocation + // and may not be freed by decommitFile, so calling decommitFile is + // unnecessary. + if !f.opts.ManualZeroing { + if err := f.decommitFile(fr); err != nil { + log.Warningf("Reclaim failed to decommit %v: %v", fr, err) + // Zero the pages manually. This won't reduce memory usage, but at + // least ensures that the pages will be zero when reallocated. + if err := f.manuallyZero(fr); err != nil { + panic(fmt.Sprintf("Reclaim failed to decommit or zero %v: %v", fr, err)) } - }) - // Pretend the pages were decommitted even though they weren't, - // since the memory accounting implementation has no idea how to - // deal with this. - f.markDecommitted(fr) + } } + f.markDecommitted(fr) f.markReclaimed(fr) } diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index fd92c3873..3f5be276b 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -263,13 +263,6 @@ func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) return usermem.NoAccess, platform.ErrContextInterrupt case ring0.El0SyncUndef: return c.fault(int32(syscall.SIGILL), info) - case ring0.El1SyncUndef: - *info = arch.SignalInfo{ - Signo: int32(syscall.SIGILL), - Code: 1, // ILL_ILLOPC (illegal opcode). - } - info.SetAddr(switchOpts.Registers.Pc) // Include address. - return usermem.AccessType{}, platform.ErrContextSignal default: panic(fmt.Sprintf("unexpected vector: 0x%x", vector)) } diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index f56aa3b79..571bfcc2e 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -18,8 +18,8 @@ // // In a nutshell, it works as follows: // -// The creation of a new address space creates a new child processes with a -// single thread which is traced by a single goroutine. +// The creation of a new address space creates a new child process with a single +// thread which is traced by a single goroutine. // // A context is just a collection of temporary variables. Calling Switch on a // context does the following: diff --git a/pkg/sentry/platform/ring0/aarch64.go b/pkg/sentry/platform/ring0/aarch64.go index 327d48465..c51df2811 100644 --- a/pkg/sentry/platform/ring0/aarch64.go +++ b/pkg/sentry/platform/ring0/aarch64.go @@ -90,6 +90,7 @@ const ( El0SyncIa El0SyncFpsimdAcc El0SyncSveAcc + El0SyncFpsimdExc El0SyncSys El0SyncSpPc El0SyncUndef diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 266817d82..cf0bf3528 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -290,6 +290,18 @@ MOVD CPU_TTBR0_KVM(from), RSV_REG; \ MSR RSV_REG, TTBR0_EL1; +TEXT ·EnableVFP(SB),NOSPLIT,$0 + MOVD $FPEN_ENABLE, R0 + WORD $0xd5181040 //MSR R0, CPACR_EL1 + ISB $15 + RET + +TEXT ·DisableVFP(SB),NOSPLIT,$0 + MOVD $0, R0 + WORD $0xd5181040 //MSR R0, CPACR_EL1 + ISB $15 + RET + #define VFP_ENABLE \ MOVD $FPEN_ENABLE, R0; \ WORD $0xd5181040; \ //MSR R0, CPACR_EL1 @@ -336,12 +348,12 @@ MOVD R4, CPU_REGISTERS+PTRACE_SP(RSV_REG); \ LOAD_KERNEL_STACK(RSV_REG); // Load the temporary stack. -// EXCEPTION_WITH_ERROR is a common exception handler function. -#define EXCEPTION_WITH_ERROR(user, vector) \ +// EXCEPTION_EL0 is a common el0 exception handler function. +#define EXCEPTION_EL0(vector) \ WORD $0xd538d092; \ //MRS TPIDR_EL1, R18 WORD $0xd538601a; \ //MRS FAR_EL1, R26 MOVD R26, CPU_FAULT_ADDR(RSV_REG); \ - MOVD $user, R3; \ + MOVD $1, R3; \ MOVD R3, CPU_ERROR_TYPE(RSV_REG); \ // Set error type to user. MOVD $vector, R3; \ MOVD R3, CPU_VECTOR_CODE(RSV_REG); \ @@ -349,6 +361,12 @@ MOVD R3, CPU_ERROR_CODE(RSV_REG); \ B ·kernelExitToEl1(SB); +// EXCEPTION_EL1 is a common el1 exception handler function. +#define EXCEPTION_EL1(vector) \ + MOVD $vector, R3; \ + MOVD R3, 8(RSP); \ + B ·HaltEl1ExceptionAndResume(SB); + // storeAppASID writes the application's asid value. TEXT ·storeAppASID(SB),NOSPLIT,$0-8 MOVD asid+0(FP), R1 @@ -396,6 +414,16 @@ TEXT ·HaltEl1SvcAndResume(SB),NOSPLIT,$0 CALL ·kernelSyscall(SB) // Call the trampoline. B ·kernelExitToEl1(SB) // Resume. +// HaltEl1ExceptionAndResume calls Hooks.KernelException and resume. +TEXT ·HaltEl1ExceptionAndResume(SB),NOSPLIT,$0-8 + WORD $0xd538d092 // MRS TPIDR_EL1, R18 + MOVD CPU_SELF(RSV_REG), R3 // Load vCPU. + MOVD R3, 8(RSP) // First argument (vCPU). + MOVD vector+0(FP), R3 + MOVD R3, 16(RSP) // Second argument (vector). + CALL ·kernelException(SB) // Call the trampoline. + B ·kernelExitToEl1(SB) // Resume. + // Shutdown stops the guest. TEXT ·Shutdown(SB),NOSPLIT,$0 // PSCI EVENT. @@ -558,39 +586,22 @@ TEXT ·El1_sync(SB),NOSPLIT,$0 B el1_invalid el1_da: + EXCEPTION_EL1(El1SyncDa) el1_ia: - WORD $0xd538d092 //MRS TPIDR_EL1, R18 - WORD $0xd538601a //MRS FAR_EL1, R26 - - MOVD R26, CPU_FAULT_ADDR(RSV_REG) - - MOVD $0, CPU_ERROR_TYPE(RSV_REG) - - MOVD $PageFault, R3 - MOVD R3, CPU_VECTOR_CODE(RSV_REG) - - B ·HaltAndResume(SB) - + EXCEPTION_EL1(El1SyncIa) el1_sp_pc: - B ·Shutdown(SB) - + EXCEPTION_EL1(El1SyncSpPc) el1_undef: - B ·Shutdown(SB) - + EXCEPTION_EL1(El1SyncUndef) el1_svc: - MOVD $0, CPU_ERROR_CODE(RSV_REG) - MOVD $0, CPU_ERROR_TYPE(RSV_REG) B ·HaltEl1SvcAndResume(SB) - el1_dbg: - B ·Shutdown(SB) - + EXCEPTION_EL1(El1SyncDbg) el1_fpsimd_acc: VFP_ENABLE B ·kernelExitToEl1(SB) // Resume. - el1_invalid: - B ·Shutdown(SB) + EXCEPTION_EL1(El1SyncInv) // El1_irq is the handler for El1_irq. TEXT ·El1_irq(SB),NOSPLIT,$0 @@ -646,28 +657,21 @@ el0_svc: el0_da: el0_ia: - EXCEPTION_WITH_ERROR(1, PageFault) - + EXCEPTION_EL0(PageFault) el0_fpsimd_acc: - B ·Shutdown(SB) - + EXCEPTION_EL0(El0SyncFpsimdAcc) el0_sve_acc: - B ·Shutdown(SB) - + EXCEPTION_EL0(El0SyncSveAcc) el0_fpsimd_exc: - B ·Shutdown(SB) - + EXCEPTION_EL0(El0SyncFpsimdExc) el0_sp_pc: - B ·Shutdown(SB) - + EXCEPTION_EL0(El0SyncSpPc) el0_undef: - EXCEPTION_WITH_ERROR(1, El0SyncUndef) - + EXCEPTION_EL0(El0SyncUndef) el0_dbg: - B ·Shutdown(SB) - + EXCEPTION_EL0(El0SyncDbg) el0_invalid: - B ·Shutdown(SB) + EXCEPTION_EL0(El0SyncInv) TEXT ·El0_irq(SB),NOSPLIT,$0 B ·Shutdown(SB) diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index 6cbbf001f..90a7b8392 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -24,6 +24,10 @@ func HaltAndResume() //go:nosplit func HaltEl1SvcAndResume() +// HaltEl1ExceptionAndResume calls Hooks.KernelException and resume. +//go:nosplit +func HaltEl1ExceptionAndResume() + // init initializes architecture-specific state. func (k *Kernel) init(maxCPUs int) { } @@ -61,11 +65,13 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { regs.Pstate &= ^uint64(PsrFlagsClear) regs.Pstate |= UserFlagsSet + EnableVFP() LoadFloatingPoint(switchOpts.FloatingPointState) kernelExitToEl0() SaveFloatingPoint(switchOpts.FloatingPointState) + DisableVFP() vector = c.vecCode diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index 842aa7e30..ef0d8974d 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -53,6 +53,12 @@ func LoadFloatingPoint(*byte) // SaveFloatingPoint saves floating point state. func SaveFloatingPoint(*byte) +// EnableVFP enables fpsimd. +func EnableVFP() + +// DisableVFP disables fpsimd. +func DisableVFP() + // Init sets function pointers based on architectural features. // // This must be called prior to using ring0. diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s index 19c1fca8b..6f4923539 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.s +++ b/pkg/sentry/platform/ring0/lib_arm64.s @@ -35,62 +35,47 @@ TEXT ·CPACREL1(SB),NOSPLIT,$0-8 RET TEXT ·GetFPCR(SB),NOSPLIT,$0-8 - WORD $0xd53b4201 // MRS NZCV, R1 + MOVD FPCR, R1 MOVD R1, ret+0(FP) RET TEXT ·GetFPSR(SB),NOSPLIT,$0-8 - WORD $0xd53b4421 // MRS FPSR, R1 + MOVD FPSR, R1 MOVD R1, ret+0(FP) RET TEXT ·SetFPCR(SB),NOSPLIT,$0-8 MOVD addr+0(FP), R1 - WORD $0xd51b4201 // MSR R1, NZCV + MOVD R1, FPCR RET TEXT ·SetFPSR(SB),NOSPLIT,$0-8 MOVD addr+0(FP), R1 - WORD $0xd51b4421 // MSR R1, FPSR + MOVD R1, FPSR RET TEXT ·SaveVRegs(SB),NOSPLIT,$0-8 MOVD addr+0(FP), R0 // Skip aarch64_ctx, fpsr, fpcr. - FMOVD F0, 16*1(R0) - FMOVD F1, 16*2(R0) - FMOVD F2, 16*3(R0) - FMOVD F3, 16*4(R0) - FMOVD F4, 16*5(R0) - FMOVD F5, 16*6(R0) - FMOVD F6, 16*7(R0) - FMOVD F7, 16*8(R0) - FMOVD F8, 16*9(R0) - FMOVD F9, 16*10(R0) - FMOVD F10, 16*11(R0) - FMOVD F11, 16*12(R0) - FMOVD F12, 16*13(R0) - FMOVD F13, 16*14(R0) - FMOVD F14, 16*15(R0) - FMOVD F15, 16*16(R0) - FMOVD F16, 16*17(R0) - FMOVD F17, 16*18(R0) - FMOVD F18, 16*19(R0) - FMOVD F19, 16*20(R0) - FMOVD F20, 16*21(R0) - FMOVD F21, 16*22(R0) - FMOVD F22, 16*23(R0) - FMOVD F23, 16*24(R0) - FMOVD F24, 16*25(R0) - FMOVD F25, 16*26(R0) - FMOVD F26, 16*27(R0) - FMOVD F27, 16*28(R0) - FMOVD F28, 16*29(R0) - FMOVD F29, 16*30(R0) - FMOVD F30, 16*31(R0) - FMOVD F31, 16*32(R0) - ISB $15 + ADD $16, R0, R0 + + WORD $0xad000400 // stp q0, q1, [x0] + WORD $0xad010c02 // stp q2, q3, [x0, #32] + WORD $0xad021404 // stp q4, q5, [x0, #64] + WORD $0xad031c06 // stp q6, q7, [x0, #96] + WORD $0xad042408 // stp q8, q9, [x0, #128] + WORD $0xad052c0a // stp q10, q11, [x0, #160] + WORD $0xad06340c // stp q12, q13, [x0, #192] + WORD $0xad073c0e // stp q14, q15, [x0, #224] + WORD $0xad084410 // stp q16, q17, [x0, #256] + WORD $0xad094c12 // stp q18, q19, [x0, #288] + WORD $0xad0a5414 // stp q20, q21, [x0, #320] + WORD $0xad0b5c16 // stp q22, q23, [x0, #352] + WORD $0xad0c6418 // stp q24, q25, [x0, #384] + WORD $0xad0d6c1a // stp q26, q27, [x0, #416] + WORD $0xad0e741c // stp q28, q29, [x0, #448] + WORD $0xad0f7c1e // stp q30, q31, [x0, #480] RET @@ -98,39 +83,24 @@ TEXT ·LoadVRegs(SB),NOSPLIT,$0-8 MOVD addr+0(FP), R0 // Skip aarch64_ctx, fpsr, fpcr. - FMOVD 16*1(R0), F0 - FMOVD 16*2(R0), F1 - FMOVD 16*3(R0), F2 - FMOVD 16*4(R0), F3 - FMOVD 16*5(R0), F4 - FMOVD 16*6(R0), F5 - FMOVD 16*7(R0), F6 - FMOVD 16*8(R0), F7 - FMOVD 16*9(R0), F8 - FMOVD 16*10(R0), F9 - FMOVD 16*11(R0), F10 - FMOVD 16*12(R0), F11 - FMOVD 16*13(R0), F12 - FMOVD 16*14(R0), F13 - FMOVD 16*15(R0), F14 - FMOVD 16*16(R0), F15 - FMOVD 16*17(R0), F16 - FMOVD 16*18(R0), F17 - FMOVD 16*19(R0), F18 - FMOVD 16*20(R0), F19 - FMOVD 16*21(R0), F20 - FMOVD 16*22(R0), F21 - FMOVD 16*23(R0), F22 - FMOVD 16*24(R0), F23 - FMOVD 16*25(R0), F24 - FMOVD 16*26(R0), F25 - FMOVD 16*27(R0), F26 - FMOVD 16*28(R0), F27 - FMOVD 16*29(R0), F28 - FMOVD 16*30(R0), F29 - FMOVD 16*31(R0), F30 - FMOVD 16*32(R0), F31 - ISB $15 + ADD $16, R0, R0 + + WORD $0xad400400 // ldp q0, q1, [x0] + WORD $0xad410c02 // ldp q2, q3, [x0, #32] + WORD $0xad421404 // ldp q4, q5, [x0, #64] + WORD $0xad431c06 // ldp q6, q7, [x0, #96] + WORD $0xad442408 // ldp q8, q9, [x0, #128] + WORD $0xad452c0a // ldp q10, q11, [x0, #160] + WORD $0xad46340c // ldp q12, q13, [x0, #192] + WORD $0xad473c0e // ldp q14, q15, [x0, #224] + WORD $0xad484410 // ldp q16, q17, [x0, #256] + WORD $0xad494c12 // ldp q18, q19, [x0, #288] + WORD $0xad4a5414 // ldp q20, q21, [x0, #320] + WORD $0xad4b5c16 // ldp q22, q23, [x0, #352] + WORD $0xad4c6418 // ldp q24, q25, [x0, #384] + WORD $0xad4d6c1a // ldp q26, q27, [x0, #416] + WORD $0xad4e741c // ldp q28, q29, [x0, #448] + WORD $0xad4f7c1e // ldp q30, q31, [x0, #480] RET diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/offsets_arm64.go index 53bc3353c..b5652deb9 100644 --- a/pkg/sentry/platform/ring0/offsets_arm64.go +++ b/pkg/sentry/platform/ring0/offsets_arm64.go @@ -70,6 +70,7 @@ func Emit(w io.Writer) { fmt.Fprintf(w, "#define El0SyncIa 0x%02x\n", El0SyncIa) fmt.Fprintf(w, "#define El0SyncFpsimdAcc 0x%02x\n", El0SyncFpsimdAcc) fmt.Fprintf(w, "#define El0SyncSveAcc 0x%02x\n", El0SyncSveAcc) + fmt.Fprintf(w, "#define El0SyncFpsimdExc 0x%02x\n", El0SyncFpsimdExc) fmt.Fprintf(w, "#define El0SyncSys 0x%02x\n", El0SyncSys) fmt.Fprintf(w, "#define El0SyncSpPc 0x%02x\n", El0SyncSpPc) fmt.Fprintf(w, "#define El0SyncUndef 0x%02x\n", El0SyncUndef) diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables.go b/pkg/sentry/platform/ring0/pagetables/pagetables.go index bc16a1622..7605d0cb2 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables.go @@ -58,6 +58,15 @@ type PageTables struct { readOnlyShared bool } +// Init initializes a set of PageTables. +// +//go:nosplit +func (p *PageTables) Init(allocator Allocator) { + p.Allocator = allocator + p.root = p.Allocator.NewPTEs() + p.rootPhysical = p.Allocator.PhysicalFor(p.root) +} + // NewWithUpper returns new PageTables. // // upperSharedPageTables are used for mapping the upper of addresses, @@ -73,14 +82,17 @@ type PageTables struct { func NewWithUpper(a Allocator, upperSharedPageTables *PageTables, upperStart uintptr) *PageTables { p := new(PageTables) p.Init(a) + if upperSharedPageTables != nil { if !upperSharedPageTables.readOnlyShared { panic("Only read-only shared pagetables can be used as upper") } p.upperSharedPageTables = upperSharedPageTables p.upperStart = upperStart - p.cloneUpperShared() } + + p.InitArch(a) + return p } diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go index a4e416af7..520161755 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go @@ -24,6 +24,14 @@ import ( // archPageTables is architecture-specific data. type archPageTables struct { + // root is the pagetable root for kernel space. + root *PTEs + + // rootPhysical is the cached physical address of the root. + // + // This is saved only to prevent constant translation. + rootPhysical uintptr + asid uint16 } @@ -38,7 +46,7 @@ func (p *PageTables) TTBR0_EL1(noFlush bool, asid uint16) uint64 { // //go:nosplit func (p *PageTables) TTBR1_EL1(noFlush bool, asid uint16) uint64 { - return uint64(p.upperSharedPageTables.rootPhysical) | (uint64(asid)&ttbrASIDMask)<<ttbrASIDOffset + return uint64(p.archPageTables.rootPhysical) | (uint64(asid)&ttbrASIDMask)<<ttbrASIDOffset } // Bits in page table entries. diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64.go index e7ab887e5..4bdde8448 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64.go @@ -41,13 +41,13 @@ const ( entriesPerPage = 512 ) -// Init initializes a set of PageTables. +// InitArch does some additional initialization related to the architecture. // //go:nosplit -func (p *PageTables) Init(allocator Allocator) { - p.Allocator = allocator - p.root = p.Allocator.NewPTEs() - p.rootPhysical = p.Allocator.PhysicalFor(p.root) +func (p *PageTables) InitArch(allocator Allocator) { + if p.upperSharedPageTables != nil { + p.cloneUpperShared() + } } func pgdIndex(upperStart uintptr) uintptr { diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go index 5392bf27a..ad0e30c88 100644 --- a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go +++ b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go @@ -42,13 +42,16 @@ const ( entriesPerPage = 512 ) -// Init initializes a set of PageTables. +// InitArch does some additional initialization related to the architecture. // //go:nosplit -func (p *PageTables) Init(allocator Allocator) { - p.Allocator = allocator - p.root = p.Allocator.NewPTEs() - p.rootPhysical = p.Allocator.PhysicalFor(p.root) +func (p *PageTables) InitArch(allocator Allocator) { + if p.upperSharedPageTables != nil { + p.cloneUpperShared() + } else { + p.archPageTables.root = p.Allocator.NewPTEs() + p.archPageTables.rootPhysical = p.Allocator.PhysicalFor(p.archPageTables.root) + } } // cloneUpperShared clone the upper from the upper shared page tables. @@ -59,7 +62,8 @@ func (p *PageTables) cloneUpperShared() { panic("upperStart should be the same as upperBottom") } - // nothing to do for arm. + p.archPageTables.root = p.upperSharedPageTables.archPageTables.root + p.archPageTables.rootPhysical = p.upperSharedPageTables.archPageTables.rootPhysical } // PTEs is a collection of entries. diff --git a/pkg/sentry/platform/ring0/pagetables/walker_arm64.go b/pkg/sentry/platform/ring0/pagetables/walker_arm64.go index 157c9a7cc..c261d393a 100644 --- a/pkg/sentry/platform/ring0/pagetables/walker_arm64.go +++ b/pkg/sentry/platform/ring0/pagetables/walker_arm64.go @@ -116,7 +116,7 @@ func next(start uintptr, size uintptr) uintptr { func (w *Walker) iterateRangeCanonical(start, end uintptr) { pgdEntryIndex := w.pageTables.root if start >= upperBottom { - pgdEntryIndex = w.pageTables.upperSharedPageTables.root + pgdEntryIndex = w.pageTables.archPageTables.root } for pgdIndex := (uint16((start & pgdMask) >> pgdShift)); start < end && pgdIndex < entriesPerPage; pgdIndex++ { diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD index a3f775d15..cc1f6bfcc 100644 --- a/pkg/sentry/socket/BUILD +++ b/pkg/sentry/socket/BUILD @@ -20,6 +20,7 @@ go_library( "//pkg/sentry/vfs", "//pkg/syserr", "//pkg/tcpip", + "//pkg/tcpip/header", "//pkg/usermem", ], ) diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 3baad098b..057f4d294 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -120,9 +120,6 @@ type socketOpsCommon struct { // fixed buffer but only consume this many bytes. sendBufferSize uint32 - // passcred indicates if this socket wants SCM credentials. - passcred bool - // filter indicates that this socket has a BPF filter "installed". // // TODO(gvisor.dev/issue/1119): We don't actually support filtering, @@ -201,10 +198,7 @@ func (s *socketOpsCommon) EventUnregister(e *waiter.Entry) { // Passcred implements transport.Credentialer.Passcred. func (s *socketOpsCommon) Passcred() bool { - s.mu.Lock() - passcred := s.passcred - s.mu.Unlock() - return passcred + return s.ep.SocketOptions().GetPassCred() } // ConnectedPasscred implements transport.Credentialer.ConnectedPasscred. @@ -419,9 +413,7 @@ func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt [] } passcred := usermem.ByteOrder.Uint32(opt) - s.mu.Lock() - s.passcred = passcred != 0 - s.mu.Unlock() + s.ep.SocketOptions().SetPassCred(passcred != 0) return nil case linux.SO_ATTACH_FILTER: diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 7d0ae15ca..5e9ab97ad 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -84,69 +84,73 @@ var Metrics = tcpip.Stats{ MalformedRcvdPackets: mustCreateMetric("/netstack/malformed_received_packets", "Number of packets received by netstack that were deemed malformed."), DroppedPackets: mustCreateMetric("/netstack/dropped_packets", "Number of packets dropped by netstack due to full queues."), ICMP: tcpip.ICMPStats{ - V4PacketsSent: tcpip.ICMPv4SentPacketStats{ - ICMPv4PacketStats: tcpip.ICMPv4PacketStats{ - Echo: mustCreateMetric("/netstack/icmp/v4/packets_sent/echo", "Total number of ICMPv4 echo packets sent by netstack."), - EchoReply: mustCreateMetric("/netstack/icmp/v4/packets_sent/echo_reply", "Total number of ICMPv4 echo reply packets sent by netstack."), - DstUnreachable: mustCreateMetric("/netstack/icmp/v4/packets_sent/dst_unreachable", "Total number of ICMPv4 destination unreachable packets sent by netstack."), - SrcQuench: mustCreateMetric("/netstack/icmp/v4/packets_sent/src_quench", "Total number of ICMPv4 source quench packets sent by netstack."), - Redirect: mustCreateMetric("/netstack/icmp/v4/packets_sent/redirect", "Total number of ICMPv4 redirect packets sent by netstack."), - TimeExceeded: mustCreateMetric("/netstack/icmp/v4/packets_sent/time_exceeded", "Total number of ICMPv4 time exceeded packets sent by netstack."), - ParamProblem: mustCreateMetric("/netstack/icmp/v4/packets_sent/param_problem", "Total number of ICMPv4 parameter problem packets sent by netstack."), - Timestamp: mustCreateMetric("/netstack/icmp/v4/packets_sent/timestamp", "Total number of ICMPv4 timestamp packets sent by netstack."), - TimestampReply: mustCreateMetric("/netstack/icmp/v4/packets_sent/timestamp_reply", "Total number of ICMPv4 timestamp reply packets sent by netstack."), - InfoRequest: mustCreateMetric("/netstack/icmp/v4/packets_sent/info_request", "Total number of ICMPv4 information request packets sent by netstack."), - InfoReply: mustCreateMetric("/netstack/icmp/v4/packets_sent/info_reply", "Total number of ICMPv4 information reply packets sent by netstack."), + V4: tcpip.ICMPv4Stats{ + PacketsSent: tcpip.ICMPv4SentPacketStats{ + ICMPv4PacketStats: tcpip.ICMPv4PacketStats{ + Echo: mustCreateMetric("/netstack/icmp/v4/packets_sent/echo", "Total number of ICMPv4 echo packets sent by netstack."), + EchoReply: mustCreateMetric("/netstack/icmp/v4/packets_sent/echo_reply", "Total number of ICMPv4 echo reply packets sent by netstack."), + DstUnreachable: mustCreateMetric("/netstack/icmp/v4/packets_sent/dst_unreachable", "Total number of ICMPv4 destination unreachable packets sent by netstack."), + SrcQuench: mustCreateMetric("/netstack/icmp/v4/packets_sent/src_quench", "Total number of ICMPv4 source quench packets sent by netstack."), + Redirect: mustCreateMetric("/netstack/icmp/v4/packets_sent/redirect", "Total number of ICMPv4 redirect packets sent by netstack."), + TimeExceeded: mustCreateMetric("/netstack/icmp/v4/packets_sent/time_exceeded", "Total number of ICMPv4 time exceeded packets sent by netstack."), + ParamProblem: mustCreateMetric("/netstack/icmp/v4/packets_sent/param_problem", "Total number of ICMPv4 parameter problem packets sent by netstack."), + Timestamp: mustCreateMetric("/netstack/icmp/v4/packets_sent/timestamp", "Total number of ICMPv4 timestamp packets sent by netstack."), + TimestampReply: mustCreateMetric("/netstack/icmp/v4/packets_sent/timestamp_reply", "Total number of ICMPv4 timestamp reply packets sent by netstack."), + InfoRequest: mustCreateMetric("/netstack/icmp/v4/packets_sent/info_request", "Total number of ICMPv4 information request packets sent by netstack."), + InfoReply: mustCreateMetric("/netstack/icmp/v4/packets_sent/info_reply", "Total number of ICMPv4 information reply packets sent by netstack."), + }, + Dropped: mustCreateMetric("/netstack/icmp/v4/packets_sent/dropped", "Total number of ICMPv4 packets dropped by netstack due to link layer errors."), }, - Dropped: mustCreateMetric("/netstack/icmp/v4/packets_sent/dropped", "Total number of ICMPv4 packets dropped by netstack due to link layer errors."), - }, - V4PacketsReceived: tcpip.ICMPv4ReceivedPacketStats{ - ICMPv4PacketStats: tcpip.ICMPv4PacketStats{ - Echo: mustCreateMetric("/netstack/icmp/v4/packets_received/echo", "Total number of ICMPv4 echo packets received by netstack."), - EchoReply: mustCreateMetric("/netstack/icmp/v4/packets_received/echo_reply", "Total number of ICMPv4 echo reply packets received by netstack."), - DstUnreachable: mustCreateMetric("/netstack/icmp/v4/packets_received/dst_unreachable", "Total number of ICMPv4 destination unreachable packets received by netstack."), - SrcQuench: mustCreateMetric("/netstack/icmp/v4/packets_received/src_quench", "Total number of ICMPv4 source quench packets received by netstack."), - Redirect: mustCreateMetric("/netstack/icmp/v4/packets_received/redirect", "Total number of ICMPv4 redirect packets received by netstack."), - TimeExceeded: mustCreateMetric("/netstack/icmp/v4/packets_received/time_exceeded", "Total number of ICMPv4 time exceeded packets received by netstack."), - ParamProblem: mustCreateMetric("/netstack/icmp/v4/packets_received/param_problem", "Total number of ICMPv4 parameter problem packets received by netstack."), - Timestamp: mustCreateMetric("/netstack/icmp/v4/packets_received/timestamp", "Total number of ICMPv4 timestamp packets received by netstack."), - TimestampReply: mustCreateMetric("/netstack/icmp/v4/packets_received/timestamp_reply", "Total number of ICMPv4 timestamp reply packets received by netstack."), - InfoRequest: mustCreateMetric("/netstack/icmp/v4/packets_received/info_request", "Total number of ICMPv4 information request packets received by netstack."), - InfoReply: mustCreateMetric("/netstack/icmp/v4/packets_received/info_reply", "Total number of ICMPv4 information reply packets received by netstack."), + PacketsReceived: tcpip.ICMPv4ReceivedPacketStats{ + ICMPv4PacketStats: tcpip.ICMPv4PacketStats{ + Echo: mustCreateMetric("/netstack/icmp/v4/packets_received/echo", "Total number of ICMPv4 echo packets received by netstack."), + EchoReply: mustCreateMetric("/netstack/icmp/v4/packets_received/echo_reply", "Total number of ICMPv4 echo reply packets received by netstack."), + DstUnreachable: mustCreateMetric("/netstack/icmp/v4/packets_received/dst_unreachable", "Total number of ICMPv4 destination unreachable packets received by netstack."), + SrcQuench: mustCreateMetric("/netstack/icmp/v4/packets_received/src_quench", "Total number of ICMPv4 source quench packets received by netstack."), + Redirect: mustCreateMetric("/netstack/icmp/v4/packets_received/redirect", "Total number of ICMPv4 redirect packets received by netstack."), + TimeExceeded: mustCreateMetric("/netstack/icmp/v4/packets_received/time_exceeded", "Total number of ICMPv4 time exceeded packets received by netstack."), + ParamProblem: mustCreateMetric("/netstack/icmp/v4/packets_received/param_problem", "Total number of ICMPv4 parameter problem packets received by netstack."), + Timestamp: mustCreateMetric("/netstack/icmp/v4/packets_received/timestamp", "Total number of ICMPv4 timestamp packets received by netstack."), + TimestampReply: mustCreateMetric("/netstack/icmp/v4/packets_received/timestamp_reply", "Total number of ICMPv4 timestamp reply packets received by netstack."), + InfoRequest: mustCreateMetric("/netstack/icmp/v4/packets_received/info_request", "Total number of ICMPv4 information request packets received by netstack."), + InfoReply: mustCreateMetric("/netstack/icmp/v4/packets_received/info_reply", "Total number of ICMPv4 information reply packets received by netstack."), + }, + Invalid: mustCreateMetric("/netstack/icmp/v4/packets_received/invalid", "Total number of ICMPv4 packets received that the transport layer could not parse."), }, - Invalid: mustCreateMetric("/netstack/icmp/v4/packets_received/invalid", "Total number of ICMPv4 packets received that the transport layer could not parse."), }, - V6PacketsSent: tcpip.ICMPv6SentPacketStats{ - ICMPv6PacketStats: tcpip.ICMPv6PacketStats{ - EchoRequest: mustCreateMetric("/netstack/icmp/v6/packets_sent/echo_request", "Total number of ICMPv6 echo request packets sent by netstack."), - EchoReply: mustCreateMetric("/netstack/icmp/v6/packets_sent/echo_reply", "Total number of ICMPv6 echo reply packets sent by netstack."), - DstUnreachable: mustCreateMetric("/netstack/icmp/v6/packets_sent/dst_unreachable", "Total number of ICMPv6 destination unreachable packets sent by netstack."), - PacketTooBig: mustCreateMetric("/netstack/icmp/v6/packets_sent/packet_too_big", "Total number of ICMPv6 packet too big packets sent by netstack."), - TimeExceeded: mustCreateMetric("/netstack/icmp/v6/packets_sent/time_exceeded", "Total number of ICMPv6 time exceeded packets sent by netstack."), - ParamProblem: mustCreateMetric("/netstack/icmp/v6/packets_sent/param_problem", "Total number of ICMPv6 parameter problem packets sent by netstack."), - RouterSolicit: mustCreateMetric("/netstack/icmp/v6/packets_sent/router_solicit", "Total number of ICMPv6 router solicit packets sent by netstack."), - RouterAdvert: mustCreateMetric("/netstack/icmp/v6/packets_sent/router_advert", "Total number of ICMPv6 router advert packets sent by netstack."), - NeighborSolicit: mustCreateMetric("/netstack/icmp/v6/packets_sent/neighbor_solicit", "Total number of ICMPv6 neighbor solicit packets sent by netstack."), - NeighborAdvert: mustCreateMetric("/netstack/icmp/v6/packets_sent/neighbor_advert", "Total number of ICMPv6 neighbor advert packets sent by netstack."), - RedirectMsg: mustCreateMetric("/netstack/icmp/v6/packets_sent/redirect_msg", "Total number of ICMPv6 redirect message packets sent by netstack."), + V6: tcpip.ICMPv6Stats{ + PacketsSent: tcpip.ICMPv6SentPacketStats{ + ICMPv6PacketStats: tcpip.ICMPv6PacketStats{ + EchoRequest: mustCreateMetric("/netstack/icmp/v6/packets_sent/echo_request", "Total number of ICMPv6 echo request packets sent by netstack."), + EchoReply: mustCreateMetric("/netstack/icmp/v6/packets_sent/echo_reply", "Total number of ICMPv6 echo reply packets sent by netstack."), + DstUnreachable: mustCreateMetric("/netstack/icmp/v6/packets_sent/dst_unreachable", "Total number of ICMPv6 destination unreachable packets sent by netstack."), + PacketTooBig: mustCreateMetric("/netstack/icmp/v6/packets_sent/packet_too_big", "Total number of ICMPv6 packet too big packets sent by netstack."), + TimeExceeded: mustCreateMetric("/netstack/icmp/v6/packets_sent/time_exceeded", "Total number of ICMPv6 time exceeded packets sent by netstack."), + ParamProblem: mustCreateMetric("/netstack/icmp/v6/packets_sent/param_problem", "Total number of ICMPv6 parameter problem packets sent by netstack."), + RouterSolicit: mustCreateMetric("/netstack/icmp/v6/packets_sent/router_solicit", "Total number of ICMPv6 router solicit packets sent by netstack."), + RouterAdvert: mustCreateMetric("/netstack/icmp/v6/packets_sent/router_advert", "Total number of ICMPv6 router advert packets sent by netstack."), + NeighborSolicit: mustCreateMetric("/netstack/icmp/v6/packets_sent/neighbor_solicit", "Total number of ICMPv6 neighbor solicit packets sent by netstack."), + NeighborAdvert: mustCreateMetric("/netstack/icmp/v6/packets_sent/neighbor_advert", "Total number of ICMPv6 neighbor advert packets sent by netstack."), + RedirectMsg: mustCreateMetric("/netstack/icmp/v6/packets_sent/redirect_msg", "Total number of ICMPv6 redirect message packets sent by netstack."), + }, + Dropped: mustCreateMetric("/netstack/icmp/v6/packets_sent/dropped", "Total number of ICMPv6 packets dropped by netstack due to link layer errors."), }, - Dropped: mustCreateMetric("/netstack/icmp/v6/packets_sent/dropped", "Total number of ICMPv6 packets dropped by netstack due to link layer errors."), - }, - V6PacketsReceived: tcpip.ICMPv6ReceivedPacketStats{ - ICMPv6PacketStats: tcpip.ICMPv6PacketStats{ - EchoRequest: mustCreateMetric("/netstack/icmp/v6/packets_received/echo_request", "Total number of ICMPv6 echo request packets received by netstack."), - EchoReply: mustCreateMetric("/netstack/icmp/v6/packets_received/echo_reply", "Total number of ICMPv6 echo reply packets received by netstack."), - DstUnreachable: mustCreateMetric("/netstack/icmp/v6/packets_received/dst_unreachable", "Total number of ICMPv6 destination unreachable packets received by netstack."), - PacketTooBig: mustCreateMetric("/netstack/icmp/v6/packets_received/packet_too_big", "Total number of ICMPv6 packet too big packets received by netstack."), - TimeExceeded: mustCreateMetric("/netstack/icmp/v6/packets_received/time_exceeded", "Total number of ICMPv6 time exceeded packets received by netstack."), - ParamProblem: mustCreateMetric("/netstack/icmp/v6/packets_received/param_problem", "Total number of ICMPv6 parameter problem packets received by netstack."), - RouterSolicit: mustCreateMetric("/netstack/icmp/v6/packets_received/router_solicit", "Total number of ICMPv6 router solicit packets received by netstack."), - RouterAdvert: mustCreateMetric("/netstack/icmp/v6/packets_received/router_advert", "Total number of ICMPv6 router advert packets received by netstack."), - NeighborSolicit: mustCreateMetric("/netstack/icmp/v6/packets_received/neighbor_solicit", "Total number of ICMPv6 neighbor solicit packets received by netstack."), - NeighborAdvert: mustCreateMetric("/netstack/icmp/v6/packets_received/neighbor_advert", "Total number of ICMPv6 neighbor advert packets received by netstack."), - RedirectMsg: mustCreateMetric("/netstack/icmp/v6/packets_received/redirect_msg", "Total number of ICMPv6 redirect message packets received by netstack."), + PacketsReceived: tcpip.ICMPv6ReceivedPacketStats{ + ICMPv6PacketStats: tcpip.ICMPv6PacketStats{ + EchoRequest: mustCreateMetric("/netstack/icmp/v6/packets_received/echo_request", "Total number of ICMPv6 echo request packets received by netstack."), + EchoReply: mustCreateMetric("/netstack/icmp/v6/packets_received/echo_reply", "Total number of ICMPv6 echo reply packets received by netstack."), + DstUnreachable: mustCreateMetric("/netstack/icmp/v6/packets_received/dst_unreachable", "Total number of ICMPv6 destination unreachable packets received by netstack."), + PacketTooBig: mustCreateMetric("/netstack/icmp/v6/packets_received/packet_too_big", "Total number of ICMPv6 packet too big packets received by netstack."), + TimeExceeded: mustCreateMetric("/netstack/icmp/v6/packets_received/time_exceeded", "Total number of ICMPv6 time exceeded packets received by netstack."), + ParamProblem: mustCreateMetric("/netstack/icmp/v6/packets_received/param_problem", "Total number of ICMPv6 parameter problem packets received by netstack."), + RouterSolicit: mustCreateMetric("/netstack/icmp/v6/packets_received/router_solicit", "Total number of ICMPv6 router solicit packets received by netstack."), + RouterAdvert: mustCreateMetric("/netstack/icmp/v6/packets_received/router_advert", "Total number of ICMPv6 router advert packets received by netstack."), + NeighborSolicit: mustCreateMetric("/netstack/icmp/v6/packets_received/neighbor_solicit", "Total number of ICMPv6 neighbor solicit packets received by netstack."), + NeighborAdvert: mustCreateMetric("/netstack/icmp/v6/packets_received/neighbor_advert", "Total number of ICMPv6 neighbor advert packets received by netstack."), + RedirectMsg: mustCreateMetric("/netstack/icmp/v6/packets_received/redirect_msg", "Total number of ICMPv6 redirect message packets received by netstack."), + }, + Invalid: mustCreateMetric("/netstack/icmp/v6/packets_received/invalid", "Total number of ICMPv6 packets received that the transport layer could not parse."), }, - Invalid: mustCreateMetric("/netstack/icmp/v6/packets_received/invalid", "Total number of ICMPv6 packets received that the transport layer could not parse."), }, }, IP: tcpip.IPStats{ @@ -209,18 +213,6 @@ const sizeOfInt32 int = 4 var errStackType = syserr.New("expected but did not receive a netstack.Stack", linux.EINVAL) -// ntohs converts a 16-bit number from network byte order to host byte order. It -// assumes that the host is little endian. -func ntohs(v uint16) uint16 { - return v<<8 | v>>8 -} - -// htons converts a 16-bit number from host byte order to network byte order. It -// assumes that the host is little endian. -func htons(v uint16) uint16 { - return ntohs(v) -} - // commonEndpoint represents the intersection of a tcpip.Endpoint and a // transport.Endpoint. type commonEndpoint interface { @@ -240,10 +232,6 @@ type commonEndpoint interface { // transport.Endpoint.SetSockOpt. SetSockOpt(tcpip.SettableSocketOption) *tcpip.Error - // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool and - // transport.Endpoint.SetSockOptBool. - SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error - // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt and // transport.Endpoint.SetSockOptInt. SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error @@ -252,18 +240,20 @@ type commonEndpoint interface { // transport.Endpoint.GetSockOpt. GetSockOpt(tcpip.GettableSocketOption) *tcpip.Error - // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool and - // transport.Endpoint.GetSockOpt. - GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) - // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt and // transport.Endpoint.GetSockOpt. GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) - // LastError implements tcpip.Endpoint.LastError. + // State returns a socket's lifecycle state. The returned value is + // protocol-specific and is primarily used for diagnostics. + State() uint32 + + // LastError implements tcpip.Endpoint.LastError and + // transport.Endpoint.LastError. LastError() *tcpip.Error - // SocketOptions implements tcpip.Endpoint.SocketOptions. + // SocketOptions implements tcpip.Endpoint.SocketOptions and + // transport.Endpoint.SocketOptions. SocketOptions() *tcpip.SocketOptions } @@ -332,9 +322,7 @@ type socketOpsCommon struct { // New creates a new endpoint socket. func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue *waiter.Queue, endpoint tcpip.Endpoint) (*fs.File, *syserr.Error) { if skType == linux.SOCK_STREAM { - if err := endpoint.SetSockOptBool(tcpip.DelayOption, true); err != nil { - return nil, syserr.TranslateNetstackError(err) - } + endpoint.SocketOptions().SetDelayOption(true) } dirent := socket.NewDirent(t, netstackDevice) @@ -363,88 +351,6 @@ func bytesToIPAddress(addr []byte) tcpip.Address { return tcpip.Address(addr) } -// AddressAndFamily reads an sockaddr struct from the given address and -// converts it to the FullAddress format. It supports AF_UNIX, AF_INET, -// AF_INET6, and AF_PACKET addresses. -// -// AddressAndFamily returns an address and its family. -func AddressAndFamily(addr []byte) (tcpip.FullAddress, uint16, *syserr.Error) { - // Make sure we have at least 2 bytes for the address family. - if len(addr) < 2 { - return tcpip.FullAddress{}, 0, syserr.ErrInvalidArgument - } - - // Get the rest of the fields based on the address family. - switch family := usermem.ByteOrder.Uint16(addr); family { - case linux.AF_UNIX: - path := addr[2:] - if len(path) > linux.UnixPathMax { - return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument - } - // Drop the terminating NUL (if one exists) and everything after - // it for filesystem (non-abstract) addresses. - if len(path) > 0 && path[0] != 0 { - if n := bytes.IndexByte(path[1:], 0); n >= 0 { - path = path[:n+1] - } - } - return tcpip.FullAddress{ - Addr: tcpip.Address(path), - }, family, nil - - case linux.AF_INET: - var a linux.SockAddrInet - if len(addr) < sockAddrInetSize { - return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument - } - binary.Unmarshal(addr[:sockAddrInetSize], usermem.ByteOrder, &a) - - out := tcpip.FullAddress{ - Addr: bytesToIPAddress(a.Addr[:]), - Port: ntohs(a.Port), - } - return out, family, nil - - case linux.AF_INET6: - var a linux.SockAddrInet6 - if len(addr) < sockAddrInet6Size { - return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument - } - binary.Unmarshal(addr[:sockAddrInet6Size], usermem.ByteOrder, &a) - - out := tcpip.FullAddress{ - Addr: bytesToIPAddress(a.Addr[:]), - Port: ntohs(a.Port), - } - if isLinkLocal(out.Addr) { - out.NIC = tcpip.NICID(a.Scope_id) - } - return out, family, nil - - case linux.AF_PACKET: - var a linux.SockAddrLink - if len(addr) < sockAddrLinkSize { - return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument - } - binary.Unmarshal(addr[:sockAddrLinkSize], usermem.ByteOrder, &a) - if a.Family != linux.AF_PACKET || a.HardwareAddrLen != header.EthernetAddressSize { - return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument - } - - // TODO(gvisor.dev/issue/173): Return protocol too. - return tcpip.FullAddress{ - NIC: tcpip.NICID(a.InterfaceIndex), - Addr: tcpip.Address(a.HardwareAddr[:header.EthernetAddressSize]), - }, family, nil - - case linux.AF_UNSPEC: - return tcpip.FullAddress{}, family, nil - - default: - return tcpip.FullAddress{}, 0, syserr.ErrAddressFamilyNotSupported - } -} - func (s *socketOpsCommon) isPacketBased() bool { return s.skType == linux.SOCK_DGRAM || s.skType == linux.SOCK_SEQPACKET || s.skType == linux.SOCK_RDM || s.skType == linux.SOCK_RAW } @@ -721,11 +627,7 @@ func (s *socketOpsCommon) checkFamily(family uint16, exact bool) *syserr.Error { return nil } if !exact && family == linux.AF_INET && s.family == linux.AF_INET6 { - v, err := s.Endpoint.GetSockOptBool(tcpip.V6OnlyOption) - if err != nil { - return syserr.TranslateNetstackError(err) - } - if !v { + if !s.Endpoint.SocketOptions().GetV6Only() { return nil } } @@ -749,7 +651,7 @@ func (s *socketOpsCommon) mapFamily(addr tcpip.FullAddress, family uint16) tcpip // Connect implements the linux syscall connect(2) for sockets backed by // tpcip.Endpoint. func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { - addr, family, err := AddressAndFamily(sockaddr) + addr, family, err := socket.AddressAndFamily(sockaddr) if err != nil { return err } @@ -830,7 +732,7 @@ func (s *socketOpsCommon) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { } } else { var err *syserr.Error - addr, family, err = AddressAndFamily(sockaddr) + addr, family, err = socket.AddressAndFamily(sockaddr) if err != nil { return err } @@ -921,7 +823,7 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, var addr linux.SockAddr var addrLen uint32 if peerAddr != nil { - addr, addrLen = ConvertAddress(s.family, *peerAddr) + addr, addrLen = socket.ConvertAddress(s.family, *peerAddr) } fd, e := t.NewFDFrom(0, ns, kernel.FDFlags{ @@ -1005,7 +907,7 @@ func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family in return getSockOptSocket(t, s, ep, family, skType, name, outLen) case linux.SOL_TCP: - return getSockOptTCP(t, ep, name, outLen) + return getSockOptTCP(t, s, ep, name, outLen) case linux.SOL_IPV6: return getSockOptIPv6(t, s, ep, name, outPtr, outLen) @@ -1068,13 +970,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.PasscredOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetPassCred())) + return &v, nil case linux.SO_SNDBUF: if outLen < sizeOfInt32 { @@ -1115,25 +1012,16 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.ReuseAddressOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetReuseAddress())) + return &v, nil case linux.SO_REUSEPORT: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.ReusePortOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetReusePort())) + return &v, nil case linux.SO_BINDTODEVICE: var v tcpip.BindToDeviceOption @@ -1174,13 +1062,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.KeepaliveEnabledOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetKeepAlive())) + return &v, nil case linux.SO_LINGER: if outLen < linux.SizeOfLinger { @@ -1235,21 +1118,18 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.NoChecksumOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetNoChecksum())) + return &v, nil case linux.SO_ACCEPTCONN: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.AcceptConnOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) + // This option is only viable for TCP endpoints. + var v bool + if _, skType, skProto := s.Type(); isTCPSocket(skType, skProto) { + v = tcp.EndpointState(ep.State()) == tcp.StateListen } vP := primitive.Int32(boolToInt32(v)) return &vP, nil @@ -1261,46 +1141,36 @@ 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) (marshal.Marshallable, *syserr.Error) { +func getSockOptTCP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name, outLen int) (marshal.Marshallable, *syserr.Error) { + if _, skType, skProto := s.Type(); !isTCPSocket(skType, skProto) { + log.Warningf("SOL_TCP options are only supported on TCP sockets: skType, skProto = %v, %d", skType, skProto) + return nil, syserr.ErrUnknownProtocolOption + } + switch name { case linux.TCP_NODELAY: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.DelayOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(!v)) - return &vP, nil + v := primitive.Int32(boolToInt32(!ep.SocketOptions().GetDelayOption())) + return &v, nil case linux.TCP_CORK: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.CorkOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetCorkOption())) + return &v, nil case linux.TCP_QUICKACK: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.QuickAckOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetQuickAck())) + return &v, nil case linux.TCP_MAXSEG: if outLen < sizeOfInt32 { @@ -1474,19 +1344,24 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (marshal // getSockOptIPv6 implements GetSockOpt when level is SOL_IPV6. func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { + if _, ok := ep.(tcpip.Endpoint); !ok { + log.Warningf("SOL_IPV6 options not supported on endpoints other than tcpip.Endpoint: option = %d", name) + return nil, syserr.ErrUnknownProtocolOption + } + + family, skType, _ := s.Type() + if family != linux.AF_INET6 { + return nil, syserr.ErrUnknownProtocolOption + } + switch name { case linux.IPV6_V6ONLY: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.V6OnlyOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetV6Only())) + return &v, nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1518,13 +1393,8 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.ReceiveTClassOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetReceiveTClass())) + return &v, nil case linux.IP6T_ORIGINAL_DST: if outLen < int(binary.Size(linux.SockAddrInet6{})) { @@ -1536,7 +1406,7 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name return nil, syserr.TranslateNetstackError(err) } - a, _ := ConvertAddress(linux.AF_INET6, tcpip.FullAddress(v)) + a, _ := socket.ConvertAddress(linux.AF_INET6, tcpip.FullAddress(v)) return a.(*linux.SockAddrInet6), nil case linux.IP6T_SO_GET_INFO: @@ -1545,7 +1415,7 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name } // Only valid for raw IPv6 sockets. - if family, skType, _ := s.Type(); family != linux.AF_INET6 || skType != linux.SOCK_RAW { + if skType != linux.SOCK_RAW { return nil, syserr.ErrProtocolNotAvailable } @@ -1565,7 +1435,7 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name return nil, syserr.ErrInvalidArgument } // Only valid for raw IPv6 sockets. - if family, skType, _ := s.Type(); family != linux.AF_INET6 || skType != linux.SOCK_RAW { + if skType != linux.SOCK_RAW { return nil, syserr.ErrProtocolNotAvailable } @@ -1585,7 +1455,7 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name } // Only valid for raw IPv6 sockets. - if family, skType, _ := s.Type(); family != linux.AF_INET6 || skType != linux.SOCK_RAW { + if skType != linux.SOCK_RAW { return nil, syserr.ErrProtocolNotAvailable } @@ -1607,6 +1477,11 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name // getSockOptIP implements GetSockOpt when level is SOL_IP. func getSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, outPtr usermem.Addr, outLen int, family int) (marshal.Marshallable, *syserr.Error) { + if _, ok := ep.(tcpip.Endpoint); !ok { + log.Warningf("SOL_IP options not supported on endpoints other than tcpip.Endpoint: option = %d", name) + return nil, syserr.ErrUnknownProtocolOption + } + switch name { case linux.IP_TTL: if outLen < sizeOfInt32 { @@ -1649,7 +1524,7 @@ func getSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in return nil, syserr.TranslateNetstackError(err) } - a, _ := ConvertAddress(linux.AF_INET, tcpip.FullAddress{Addr: v.InterfaceAddr}) + a, _ := socket.ConvertAddress(linux.AF_INET, tcpip.FullAddress{Addr: v.InterfaceAddr}) return &a.(*linux.SockAddrInet).Addr, nil @@ -1658,13 +1533,8 @@ func getSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.MulticastLoopOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetMulticastLoop())) + return &v, nil case linux.IP_TOS: // Length handling for parity with Linux. @@ -1688,26 +1558,24 @@ func getSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.ReceiveTOSOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) - } - - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetReceiveTOS())) + return &v, nil case linux.IP_PKTINFO: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptBool(tcpip.ReceiveIPPacketInfoOption) - if err != nil { - return nil, syserr.TranslateNetstackError(err) + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetReceivePacketInfo())) + return &v, nil + + case linux.IP_HDRINCL: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument } - vP := primitive.Int32(boolToInt32(v)) - return &vP, nil + v := primitive.Int32(boolToInt32(ep.SocketOptions().GetHeaderIncluded())) + return &v, nil case linux.SO_ORIGINAL_DST: if outLen < int(binary.Size(linux.SockAddrInet{})) { @@ -1719,7 +1587,7 @@ func getSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in return nil, syserr.TranslateNetstackError(err) } - a, _ := ConvertAddress(linux.AF_INET, tcpip.FullAddress(v)) + a, _ := socket.ConvertAddress(linux.AF_INET, tcpip.FullAddress(v)) return a.(*linux.SockAddrInet), nil case linux.IPT_SO_GET_INFO: @@ -1826,7 +1694,7 @@ func SetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, level int return setSockOptSocket(t, s, ep, name, optVal) case linux.SOL_TCP: - return setSockOptTCP(t, ep, name, optVal) + return setSockOptTCP(t, s, ep, name, optVal) case linux.SOL_IPV6: return setSockOptIPv6(t, s, ep, name, optVal) @@ -1876,7 +1744,8 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReuseAddressOption, v != 0)) + ep.SocketOptions().SetReuseAddress(v != 0) + return nil case linux.SO_REUSEPORT: if len(optVal) < sizeOfInt32 { @@ -1884,7 +1753,8 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReusePortOption, v != 0)) + ep.SocketOptions().SetReusePort(v != 0) + return nil case linux.SO_BINDTODEVICE: n := bytes.IndexByte(optVal, 0) @@ -1923,7 +1793,8 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.PasscredOption, v != 0)) + ep.SocketOptions().SetPassCred(v != 0) + return nil case linux.SO_KEEPALIVE: if len(optVal) < sizeOfInt32 { @@ -1931,7 +1802,8 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.KeepaliveEnabledOption, v != 0)) + ep.SocketOptions().SetKeepAlive(v != 0) + return nil case linux.SO_SNDTIMEO: if len(optVal) < linux.SizeOfTimeval { @@ -1979,7 +1851,8 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.NoChecksumOption, v != 0)) + ep.SocketOptions().SetNoChecksum(v != 0) + return nil case linux.SO_LINGER: if len(optVal) < linux.SizeOfLinger { @@ -2011,7 +1884,12 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } // setSockOptTCP implements SetSockOpt when level is SOL_TCP. -func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *syserr.Error { +func setSockOptTCP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, optVal []byte) *syserr.Error { + if _, skType, skProto := s.Type(); !isTCPSocket(skType, skProto) { + log.Warningf("SOL_TCP options are only supported on TCP sockets: skType, skProto = %v, %d", skType, skProto) + return syserr.ErrUnknownProtocolOption + } + switch name { case linux.TCP_NODELAY: if len(optVal) < sizeOfInt32 { @@ -2019,7 +1897,8 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.DelayOption, v == 0)) + ep.SocketOptions().SetDelayOption(v == 0) + return nil case linux.TCP_CORK: if len(optVal) < sizeOfInt32 { @@ -2027,7 +1906,8 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.CorkOption, v != 0)) + ep.SocketOptions().SetCorkOption(v != 0) + return nil case linux.TCP_QUICKACK: if len(optVal) < sizeOfInt32 { @@ -2035,7 +1915,8 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.QuickAckOption, v != 0)) + ep.SocketOptions().SetQuickAck(v != 0) + return nil case linux.TCP_MAXSEG: if len(optVal) < sizeOfInt32 { @@ -2147,14 +2028,31 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * // setSockOptIPv6 implements SetSockOpt when level is SOL_IPV6. func setSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, optVal []byte) *syserr.Error { + if _, ok := ep.(tcpip.Endpoint); !ok { + log.Warningf("SOL_IPV6 options not supported on endpoints other than tcpip.Endpoint: option = %d", name) + return syserr.ErrUnknownProtocolOption + } + + family, skType, skProto := s.Type() + if family != linux.AF_INET6 { + return syserr.ErrUnknownProtocolOption + } + switch name { case linux.IPV6_V6ONLY: if len(optVal) < sizeOfInt32 { return syserr.ErrInvalidArgument } + if isTCPSocket(skType, skProto) && tcp.EndpointState(ep.State()) != tcp.StateInitial { + return syserr.ErrInvalidEndpointState + } else if isUDPSocket(skType, skProto) && udp.EndpointState(ep.State()) != udp.StateInitial { + return syserr.ErrInvalidEndpointState + } + v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.V6OnlyOption, v != 0)) + ep.SocketOptions().SetV6Only(v != 0) + return nil case linux.IPV6_ADD_MEMBERSHIP, linux.IPV6_DROP_MEMBERSHIP, @@ -2193,7 +2091,8 @@ func setSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name return err } - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTClassOption, v != 0)) + ep.SocketOptions().SetReceiveTClass(v != 0) + return nil case linux.IP6T_SO_SET_REPLACE: if len(optVal) < linux.SizeOfIP6TReplace { @@ -2201,7 +2100,7 @@ func setSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name } // Only valid for raw IPv6 sockets. - if family, skType, _ := s.Type(); family != linux.AF_INET6 || skType != linux.SOCK_RAW { + if skType != linux.SOCK_RAW { return syserr.ErrProtocolNotAvailable } @@ -2276,6 +2175,11 @@ func parseIntOrChar(buf []byte) (int32, *syserr.Error) { // setSockOptIP implements SetSockOpt when level is SOL_IP. func setSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, optVal []byte) *syserr.Error { + if _, ok := ep.(tcpip.Endpoint); !ok { + log.Warningf("SOL_IP options not supported on endpoints other than tcpip.Endpoint: option = %d", name) + return syserr.ErrUnknownProtocolOption + } + switch name { case linux.IP_MULTICAST_TTL: v, err := parseIntOrChar(optVal) @@ -2328,7 +2232,7 @@ func setSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in return syserr.TranslateNetstackError(ep.SetSockOpt(&tcpip.MulticastInterfaceOption{ NIC: tcpip.NICID(req.InterfaceIndex), - InterfaceAddr: bytesToIPAddress(req.InterfaceAddr[:]), + InterfaceAddr: socket.BytesToIPAddress(req.InterfaceAddr[:]), })) case linux.IP_MULTICAST_LOOP: @@ -2337,7 +2241,8 @@ func setSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in return err } - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.MulticastLoopOption, v != 0)) + ep.SocketOptions().SetMulticastLoop(v != 0) + return nil case linux.MCAST_JOIN_GROUP: // FIXME(b/124219304): Implement MCAST_JOIN_GROUP. @@ -2373,7 +2278,8 @@ func setSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in if err != nil { return err } - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTOSOption, v != 0)) + ep.SocketOptions().SetReceiveTOS(v != 0) + return nil case linux.IP_PKTINFO: if len(optVal) == 0 { @@ -2383,7 +2289,8 @@ func setSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in if err != nil { return err } - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveIPPacketInfoOption, v != 0)) + ep.SocketOptions().SetReceivePacketInfo(v != 0) + return nil case linux.IP_HDRINCL: if len(optVal) == 0 { @@ -2393,7 +2300,8 @@ func setSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in if err != nil { return err } - return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.IPHdrIncludedOption, v != 0)) + ep.SocketOptions().SetHeaderIncluded(v != 0) + return nil case linux.IPT_SO_SET_REPLACE: if len(optVal) < linux.SizeOfIPTReplace { @@ -2535,7 +2443,6 @@ func emitUnimplementedEventIP(t *kernel.Task, name int) { switch name { case linux.IP_TOS, linux.IP_TTL, - linux.IP_HDRINCL, linux.IP_OPTIONS, linux.IP_ROUTER_ALERT, linux.IP_RECVOPTS, @@ -2582,72 +2489,6 @@ func emitUnimplementedEventIP(t *kernel.Task, name int) { } } -// isLinkLocal determines if the given IPv6 address is link-local. This is the -// case when it has the fe80::/10 prefix. This check is used to determine when -// the NICID is relevant for a given IPv6 address. -func isLinkLocal(addr tcpip.Address) bool { - return len(addr) >= 2 && addr[0] == 0xfe && addr[1]&0xc0 == 0x80 -} - -// ConvertAddress converts the given address to a native format. -func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) { - switch family { - case linux.AF_UNIX: - var out linux.SockAddrUnix - out.Family = linux.AF_UNIX - l := len([]byte(addr.Addr)) - for i := 0; i < l; i++ { - out.Path[i] = int8(addr.Addr[i]) - } - - // Linux returns the used length of the address struct (including the - // null terminator) for filesystem paths. The Family field is 2 bytes. - // It is sometimes allowed to exclude the null terminator if the - // address length is the max. Abstract and empty paths always return - // the full exact length. - if l == 0 || out.Path[0] == 0 || l == len(out.Path) { - return &out, uint32(2 + l) - } - return &out, uint32(3 + l) - - case linux.AF_INET: - var out linux.SockAddrInet - copy(out.Addr[:], addr.Addr) - out.Family = linux.AF_INET - out.Port = htons(addr.Port) - return &out, uint32(sockAddrInetSize) - - case linux.AF_INET6: - var out linux.SockAddrInet6 - if len(addr.Addr) == header.IPv4AddressSize { - // Copy address in v4-mapped format. - copy(out.Addr[12:], addr.Addr) - out.Addr[10] = 0xff - out.Addr[11] = 0xff - } else { - copy(out.Addr[:], addr.Addr) - } - out.Family = linux.AF_INET6 - out.Port = htons(addr.Port) - if isLinkLocal(addr.Addr) { - out.Scope_id = uint32(addr.NIC) - } - return &out, uint32(sockAddrInet6Size) - - case linux.AF_PACKET: - // TODO(gvisor.dev/issue/173): Return protocol too. - var out linux.SockAddrLink - out.Family = linux.AF_PACKET - out.InterfaceIndex = int32(addr.NIC) - out.HardwareAddrLen = header.EthernetAddressSize - copy(out.HardwareAddr[:], addr.Addr) - return &out, uint32(sockAddrLinkSize) - - default: - return nil, 0 - } -} - // GetSockName implements the linux syscall getsockname(2) for sockets backed by // tcpip.Endpoint. func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { @@ -2656,7 +2497,7 @@ func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, * return nil, 0, syserr.TranslateNetstackError(err) } - a, l := ConvertAddress(s.family, addr) + a, l := socket.ConvertAddress(s.family, addr) return a, l, nil } @@ -2668,7 +2509,7 @@ func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, * return nil, 0, syserr.TranslateNetstackError(err) } - a, l := ConvertAddress(s.family, addr) + a, l := socket.ConvertAddress(s.family, addr) return a, l, nil } @@ -2686,7 +2527,7 @@ func (s *socketOpsCommon) coalescingRead(ctx context.Context, dst usermem.IOSequ // Always do at least one fetchReadView, even if the number of bytes to // read is 0. err = s.fetchReadView() - if err != nil { + if err != nil || len(s.readView) == 0 { break } if dst.NumBytes() == 0 { @@ -2709,15 +2550,20 @@ func (s *socketOpsCommon) coalescingRead(ctx context.Context, dst usermem.IOSequ } copied += n s.readView.TrimFront(n) - if len(s.readView) == 0 { - atomic.StoreUint32(&s.readViewHasData, 0) - } dst = dst.DropFirst(n) if e != nil { err = syserr.FromError(e) break } + // If we are done reading requested data then stop. + if dst.NumBytes() == 0 { + break + } + } + + if len(s.readView) == 0 { + atomic.StoreUint32(&s.readViewHasData, 0) } // If we managed to copy something, we must deliver it. @@ -2812,10 +2658,10 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq var addr linux.SockAddr var addrLen uint32 if isPacket && senderRequested { - addr, addrLen = ConvertAddress(s.family, s.sender) + addr, addrLen = socket.ConvertAddress(s.family, s.sender) switch v := addr.(type) { case *linux.SockAddrLink: - v.Protocol = htons(uint16(s.linkPacketInfo.Protocol)) + v.Protocol = socket.Htons(uint16(s.linkPacketInfo.Protocol)) v.PacketType = toLinuxPacketType(s.linkPacketInfo.PktType) } } @@ -2980,7 +2826,7 @@ func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []b var addr *tcpip.FullAddress if len(to) > 0 { - addrBuf, family, err := AddressAndFamily(to) + addrBuf, family, err := socket.AddressAndFamily(to) if err != nil { return 0, err } @@ -3399,6 +3245,18 @@ func nicStateFlagsToLinux(f stack.NICStateFlags) uint32 { return rv } +func isTCPSocket(skType linux.SockType, skProto int) bool { + return skType == linux.SOCK_STREAM && (skProto == 0 || skProto == syscall.IPPROTO_TCP) +} + +func isUDPSocket(skType linux.SockType, skProto int) bool { + return skType == linux.SOCK_DGRAM && (skProto == 0 || skProto == syscall.IPPROTO_UDP) +} + +func isICMPSocket(skType linux.SockType, skProto int) bool { + return skType == linux.SOCK_DGRAM && (skProto == syscall.IPPROTO_ICMP || skProto == syscall.IPPROTO_ICMPV6) +} + // State implements socket.Socket.State. State translates the internal state // returned by netstack to values defined by Linux. func (s *socketOpsCommon) State() uint32 { @@ -3408,7 +3266,7 @@ func (s *socketOpsCommon) State() uint32 { } switch { - case s.skType == linux.SOCK_STREAM && s.protocol == 0 || s.protocol == syscall.IPPROTO_TCP: + case isTCPSocket(s.skType, s.protocol): // TCP socket. switch tcp.EndpointState(s.Endpoint.State()) { case tcp.StateEstablished: @@ -3437,7 +3295,7 @@ func (s *socketOpsCommon) State() uint32 { // Internal or unknown state. return 0 } - case s.skType == linux.SOCK_DGRAM && s.protocol == 0 || s.protocol == syscall.IPPROTO_UDP: + case isUDPSocket(s.skType, s.protocol): // UDP socket. switch udp.EndpointState(s.Endpoint.State()) { case udp.StateInitial, udp.StateBound, udp.StateClosed: @@ -3447,7 +3305,7 @@ func (s *socketOpsCommon) State() uint32 { default: return 0 } - case s.skType == linux.SOCK_DGRAM && s.protocol == syscall.IPPROTO_ICMP || s.protocol == syscall.IPPROTO_ICMPV6: + case isICMPSocket(s.skType, s.protocol): // TODO(b/112063468): Export states for ICMP sockets. case s.skType == linux.SOCK_RAW: // TODO(b/112063468): Export states for raw sockets. diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index b0d9e4d9e..b756bfca0 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -51,9 +51,7 @@ var _ = socket.SocketVFS2(&SocketVFS2{}) // NewVFS2 creates a new endpoint socket. func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, queue *waiter.Queue, endpoint tcpip.Endpoint) (*vfs.FileDescription, *syserr.Error) { if skType == linux.SOCK_STREAM { - if err := endpoint.SetSockOptBool(tcpip.DelayOption, true); err != nil { - return nil, syserr.TranslateNetstackError(err) - } + endpoint.SocketOptions().SetDelayOption(true) } mnt := t.Kernel().SocketMount() @@ -191,7 +189,7 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block var addrLen uint32 if peerAddr != nil { // Get address of the peer and write it to peer slice. - addr, addrLen = ConvertAddress(s.family, *peerAddr) + addr, addrLen = socket.ConvertAddress(s.family, *peerAddr) } fd, e := t.NewFDFromVFS2(0, ns, kernel.FDFlags{ diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go index ead3b2b79..c847ff1c7 100644 --- a/pkg/sentry/socket/netstack/provider.go +++ b/pkg/sentry/socket/netstack/provider.go @@ -158,7 +158,7 @@ func packetSocket(t *kernel.Task, epStack *Stack, stype linux.SockType, protocol // protocol is passed in network byte order, but netstack wants it in // host order. - netProto := tcpip.NetworkProtocolNumber(ntohs(uint16(protocol))) + netProto := tcpip.NetworkProtocolNumber(socket.Ntohs(uint16(protocol))) wq := &waiter.Queue{} ep, err := epStack.Stack.NewPacketEndpoint(cooked, netProto, wq) diff --git a/pkg/sentry/socket/netstack/provider_vfs2.go b/pkg/sentry/socket/netstack/provider_vfs2.go index 2a01143f6..0af805246 100644 --- a/pkg/sentry/socket/netstack/provider_vfs2.go +++ b/pkg/sentry/socket/netstack/provider_vfs2.go @@ -102,7 +102,7 @@ func packetSocketVFS2(t *kernel.Task, epStack *Stack, stype linux.SockType, prot // protocol is passed in network byte order, but netstack wants it in // host order. - netProto := tcpip.NetworkProtocolNumber(ntohs(uint16(protocol))) + netProto := tcpip.NetworkProtocolNumber(socket.Ntohs(uint16(protocol))) wq := &waiter.Queue{} ep, err := epStack.Stack.NewPacketEndpoint(cooked, netProto, wq) diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index fa9ac9059..cc0fadeb5 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -324,12 +324,12 @@ func (s *Stack) Statistics(stat interface{}, arg string) error { 0, // Support Ip/FragCreates. } case *inet.StatSNMPICMP: - in := Metrics.ICMP.V4PacketsReceived.ICMPv4PacketStats - out := Metrics.ICMP.V4PacketsSent.ICMPv4PacketStats + in := Metrics.ICMP.V4.PacketsReceived.ICMPv4PacketStats + out := Metrics.ICMP.V4.PacketsSent.ICMPv4PacketStats // TODO(gvisor.dev/issue/969) Support stubbed stats. *stats = inet.StatSNMPICMP{ 0, // Icmp/InMsgs. - Metrics.ICMP.V4PacketsSent.Dropped.Value(), // InErrors. + Metrics.ICMP.V4.PacketsSent.Dropped.Value(), // InErrors. 0, // Icmp/InCsumErrors. in.DstUnreachable.Value(), // InDestUnreachs. in.TimeExceeded.Value(), // InTimeExcds. @@ -343,18 +343,18 @@ func (s *Stack) Statistics(stat interface{}, arg string) error { in.InfoRequest.Value(), // InAddrMasks. in.InfoReply.Value(), // InAddrMaskReps. 0, // Icmp/OutMsgs. - Metrics.ICMP.V4PacketsReceived.Invalid.Value(), // OutErrors. - out.DstUnreachable.Value(), // OutDestUnreachs. - out.TimeExceeded.Value(), // OutTimeExcds. - out.ParamProblem.Value(), // OutParmProbs. - out.SrcQuench.Value(), // OutSrcQuenchs. - out.Redirect.Value(), // OutRedirects. - out.Echo.Value(), // OutEchos. - out.EchoReply.Value(), // OutEchoReps. - out.Timestamp.Value(), // OutTimestamps. - out.TimestampReply.Value(), // OutTimestampReps. - out.InfoRequest.Value(), // OutAddrMasks. - out.InfoReply.Value(), // OutAddrMaskReps. + Metrics.ICMP.V4.PacketsReceived.Invalid.Value(), // OutErrors. + out.DstUnreachable.Value(), // OutDestUnreachs. + out.TimeExceeded.Value(), // OutTimeExcds. + out.ParamProblem.Value(), // OutParmProbs. + out.SrcQuench.Value(), // OutSrcQuenchs. + out.Redirect.Value(), // OutRedirects. + out.Echo.Value(), // OutEchos. + out.EchoReply.Value(), // OutEchoReps. + out.Timestamp.Value(), // OutTimestamps. + out.TimestampReply.Value(), // OutTimestampReps. + out.InfoRequest.Value(), // OutAddrMasks. + out.InfoReply.Value(), // OutAddrMaskReps. } case *inet.StatSNMPTCP: tcp := Metrics.TCP diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index fd31479e5..9049e8a21 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -18,6 +18,7 @@ package socket import ( + "bytes" "fmt" "sync/atomic" "syscall" @@ -35,6 +36,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/usermem" ) @@ -460,3 +462,176 @@ func UnmarshalSockAddr(family int, data []byte) linux.SockAddr { panic(fmt.Sprintf("Unsupported socket family %v", family)) } } + +var sockAddrLinkSize = (&linux.SockAddrLink{}).SizeBytes() +var sockAddrInetSize = (&linux.SockAddrInet{}).SizeBytes() +var sockAddrInet6Size = (&linux.SockAddrInet6{}).SizeBytes() + +// Ntohs converts a 16-bit number from network byte order to host byte order. It +// assumes that the host is little endian. +func Ntohs(v uint16) uint16 { + return v<<8 | v>>8 +} + +// Htons converts a 16-bit number from host byte order to network byte order. It +// assumes that the host is little endian. +func Htons(v uint16) uint16 { + return Ntohs(v) +} + +// isLinkLocal determines if the given IPv6 address is link-local. This is the +// case when it has the fe80::/10 prefix. This check is used to determine when +// the NICID is relevant for a given IPv6 address. +func isLinkLocal(addr tcpip.Address) bool { + return len(addr) >= 2 && addr[0] == 0xfe && addr[1]&0xc0 == 0x80 +} + +// ConvertAddress converts the given address to a native format. +func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) { + switch family { + case linux.AF_UNIX: + var out linux.SockAddrUnix + out.Family = linux.AF_UNIX + l := len([]byte(addr.Addr)) + for i := 0; i < l; i++ { + out.Path[i] = int8(addr.Addr[i]) + } + + // Linux returns the used length of the address struct (including the + // null terminator) for filesystem paths. The Family field is 2 bytes. + // It is sometimes allowed to exclude the null terminator if the + // address length is the max. Abstract and empty paths always return + // the full exact length. + if l == 0 || out.Path[0] == 0 || l == len(out.Path) { + return &out, uint32(2 + l) + } + return &out, uint32(3 + l) + + case linux.AF_INET: + var out linux.SockAddrInet + copy(out.Addr[:], addr.Addr) + out.Family = linux.AF_INET + out.Port = Htons(addr.Port) + return &out, uint32(sockAddrInetSize) + + case linux.AF_INET6: + var out linux.SockAddrInet6 + if len(addr.Addr) == header.IPv4AddressSize { + // Copy address in v4-mapped format. + copy(out.Addr[12:], addr.Addr) + out.Addr[10] = 0xff + out.Addr[11] = 0xff + } else { + copy(out.Addr[:], addr.Addr) + } + out.Family = linux.AF_INET6 + out.Port = Htons(addr.Port) + if isLinkLocal(addr.Addr) { + out.Scope_id = uint32(addr.NIC) + } + return &out, uint32(sockAddrInet6Size) + + case linux.AF_PACKET: + // TODO(gvisor.dev/issue/173): Return protocol too. + var out linux.SockAddrLink + out.Family = linux.AF_PACKET + out.InterfaceIndex = int32(addr.NIC) + out.HardwareAddrLen = header.EthernetAddressSize + copy(out.HardwareAddr[:], addr.Addr) + return &out, uint32(sockAddrLinkSize) + + default: + return nil, 0 + } +} + +// BytesToIPAddress converts an IPv4 or IPv6 address from the user to the +// netstack representation taking any addresses into account. +func BytesToIPAddress(addr []byte) tcpip.Address { + if bytes.Equal(addr, make([]byte, 4)) || bytes.Equal(addr, make([]byte, 16)) { + return "" + } + return tcpip.Address(addr) +} + +// AddressAndFamily reads an sockaddr struct from the given address and +// converts it to the FullAddress format. It supports AF_UNIX, AF_INET, +// AF_INET6, and AF_PACKET addresses. +// +// AddressAndFamily returns an address and its family. +func AddressAndFamily(addr []byte) (tcpip.FullAddress, uint16, *syserr.Error) { + // Make sure we have at least 2 bytes for the address family. + if len(addr) < 2 { + return tcpip.FullAddress{}, 0, syserr.ErrInvalidArgument + } + + // Get the rest of the fields based on the address family. + switch family := usermem.ByteOrder.Uint16(addr); family { + case linux.AF_UNIX: + path := addr[2:] + if len(path) > linux.UnixPathMax { + return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument + } + // Drop the terminating NUL (if one exists) and everything after + // it for filesystem (non-abstract) addresses. + if len(path) > 0 && path[0] != 0 { + if n := bytes.IndexByte(path[1:], 0); n >= 0 { + path = path[:n+1] + } + } + return tcpip.FullAddress{ + Addr: tcpip.Address(path), + }, family, nil + + case linux.AF_INET: + var a linux.SockAddrInet + if len(addr) < sockAddrInetSize { + return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument + } + binary.Unmarshal(addr[:sockAddrInetSize], usermem.ByteOrder, &a) + + out := tcpip.FullAddress{ + Addr: BytesToIPAddress(a.Addr[:]), + Port: Ntohs(a.Port), + } + return out, family, nil + + case linux.AF_INET6: + var a linux.SockAddrInet6 + if len(addr) < sockAddrInet6Size { + return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument + } + binary.Unmarshal(addr[:sockAddrInet6Size], usermem.ByteOrder, &a) + + out := tcpip.FullAddress{ + Addr: BytesToIPAddress(a.Addr[:]), + Port: Ntohs(a.Port), + } + if isLinkLocal(out.Addr) { + out.NIC = tcpip.NICID(a.Scope_id) + } + return out, family, nil + + case linux.AF_PACKET: + var a linux.SockAddrLink + if len(addr) < sockAddrLinkSize { + return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument + } + binary.Unmarshal(addr[:sockAddrLinkSize], usermem.ByteOrder, &a) + if a.Family != linux.AF_PACKET || a.HardwareAddrLen != header.EthernetAddressSize { + return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/173): Return protocol too. + return tcpip.FullAddress{ + NIC: tcpip.NICID(a.InterfaceIndex), + Addr: tcpip.Address(a.HardwareAddr[:header.EthernetAddressSize]), + }, family, nil + + case linux.AF_UNSPEC: + return tcpip.FullAddress{}, family, nil + + default: + return tcpip.FullAddress{}, 0, syserr.ErrAddressFamilyNotSupported + } +} diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index 6d9e502bd..9f7aca305 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -118,28 +118,24 @@ var ( // NewConnectioned creates a new unbound connectionedEndpoint. func NewConnectioned(ctx context.Context, stype linux.SockType, uid UniqueIDProvider) Endpoint { - return &connectionedEndpoint{ + return newConnectioned(ctx, stype, uid) +} + +func newConnectioned(ctx context.Context, stype linux.SockType, uid UniqueIDProvider) *connectionedEndpoint { + ep := &connectionedEndpoint{ baseEndpoint: baseEndpoint{Queue: &waiter.Queue{}}, id: uid.UniqueID(), idGenerator: uid, stype: stype, } + ep.ops.InitHandler(ep) + return ep } // NewPair allocates a new pair of connected unix-domain connectionedEndpoints. func NewPair(ctx context.Context, stype linux.SockType, uid UniqueIDProvider) (Endpoint, Endpoint) { - a := &connectionedEndpoint{ - baseEndpoint: baseEndpoint{Queue: &waiter.Queue{}}, - id: uid.UniqueID(), - idGenerator: uid, - stype: stype, - } - b := &connectionedEndpoint{ - baseEndpoint: baseEndpoint{Queue: &waiter.Queue{}}, - id: uid.UniqueID(), - idGenerator: uid, - stype: stype, - } + a := newConnectioned(ctx, stype, uid) + b := newConnectioned(ctx, stype, uid) q1 := &queue{ReaderQueue: a.Queue, WriterQueue: b.Queue, limit: initialLimit} q1.InitRefs() @@ -171,12 +167,14 @@ func NewPair(ctx context.Context, stype linux.SockType, uid UniqueIDProvider) (E // NewExternal creates a new externally backed Endpoint. It behaves like a // socketpair. func NewExternal(ctx context.Context, stype linux.SockType, uid UniqueIDProvider, queue *waiter.Queue, receiver Receiver, connected ConnectedEndpoint) Endpoint { - return &connectionedEndpoint{ + ep := &connectionedEndpoint{ baseEndpoint: baseEndpoint{Queue: queue, receiver: receiver, connected: connected}, id: uid.UniqueID(), idGenerator: uid, stype: stype, } + ep.ops.InitHandler(ep) + return ep } // ID implements ConnectingEndpoint.ID. @@ -298,6 +296,7 @@ func (e *connectionedEndpoint) BidirectionalConnect(ctx context.Context, ce Conn idGenerator: e.idGenerator, stype: e.stype, } + ne.ops.InitHandler(ne) readQueue := &queue{ReaderQueue: ce.WaiterQueue(), WriterQueue: ne.Queue, limit: initialLimit} readQueue.InitRefs() diff --git a/pkg/sentry/socket/unix/transport/connectionless.go b/pkg/sentry/socket/unix/transport/connectionless.go index 1406971bc..0813ad87d 100644 --- a/pkg/sentry/socket/unix/transport/connectionless.go +++ b/pkg/sentry/socket/unix/transport/connectionless.go @@ -44,6 +44,7 @@ func NewConnectionless(ctx context.Context) Endpoint { q := queue{ReaderQueue: ep.Queue, WriterQueue: &waiter.Queue{}, limit: initialLimit} q.InitRefs() ep.receiver = &queueReceiver{readQueue: &q} + ep.ops.InitHandler(ep) return ep } diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 18a50e9f8..0247e93fa 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -16,8 +16,6 @@ package transport import ( - "sync/atomic" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" @@ -180,10 +178,6 @@ type Endpoint interface { // SetSockOpt sets a socket option. SetSockOpt(opt tcpip.SettableSocketOption) *tcpip.Error - // SetSockOptBool sets a socket option for simple cases when a value has - // the int type. - SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error - // SetSockOptInt sets a socket option for simple cases when a value has // the int type. SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error @@ -191,10 +185,6 @@ type Endpoint interface { // GetSockOpt gets a socket option. GetSockOpt(opt tcpip.GettableSocketOption) *tcpip.Error - // GetSockOptBool gets a socket option for simple cases when a return - // value has the int type. - GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) - // GetSockOptInt gets a socket option for simple cases when a return // value has the int type. GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) @@ -203,10 +193,11 @@ type Endpoint interface { // procfs. State() uint32 - // LastError implements tcpip.Endpoint.LastError. + // LastError clears and returns the last error reported by the endpoint. LastError() *tcpip.Error - // SocketOptions implements tcpip.Endpoint.SocketOptions. + // SocketOptions returns the structure which contains all the socket + // level options. SocketOptions() *tcpip.SocketOptions } @@ -739,10 +730,7 @@ func (e *connectedEndpoint) CloseUnread() { // +stateify savable type baseEndpoint struct { *waiter.Queue - - // passcred specifies whether SCM_CREDENTIALS socket control messages are - // enabled on this endpoint. Must be accessed atomically. - passcred int32 + tcpip.DefaultSocketOptionsHandler // Mutex protects the below fields. sync.Mutex `state:"nosave"` @@ -761,6 +749,7 @@ type baseEndpoint struct { // linger is used for SO_LINGER socket option. linger tcpip.LingerOption + // ops is used to get socket level options. ops tcpip.SocketOptions } @@ -786,7 +775,7 @@ func (e *baseEndpoint) EventUnregister(we *waiter.Entry) { // Passcred implements Credentialer.Passcred. func (e *baseEndpoint) Passcred() bool { - return atomic.LoadInt32(&e.passcred) != 0 + return e.SocketOptions().GetPassCred() } // ConnectedPasscred implements Credentialer.ConnectedPasscred. @@ -796,14 +785,6 @@ func (e *baseEndpoint) ConnectedPasscred() bool { return e.connected != nil && e.connected.Passcred() } -func (e *baseEndpoint) setPasscred(pc bool) { - if pc { - atomic.StoreInt32(&e.passcred, 1) - } else { - atomic.StoreInt32(&e.passcred, 0) - } -} - // Connected implements ConnectingEndpoint.Connected. func (e *baseEndpoint) Connected() bool { return e.receiver != nil && e.connected != nil @@ -868,17 +849,6 @@ func (e *baseEndpoint) SetSockOpt(opt tcpip.SettableSocketOption) *tcpip.Error { return nil } -func (e *baseEndpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - switch opt { - case tcpip.PasscredOption: - e.setPasscred(v) - case tcpip.ReuseAddressOption: - default: - log.Warningf("Unsupported socket option: %d", opt) - } - return nil -} - func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { switch opt { case tcpip.SendBufferSizeOption: @@ -889,20 +859,6 @@ func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return nil } -func (e *baseEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - switch opt { - case tcpip.KeepaliveEnabledOption, tcpip.AcceptConnOption: - return false, nil - - case tcpip.PasscredOption: - return e.Passcred(), nil - - default: - log.Warningf("Unsupported socket option: %d", opt) - return false, tcpip.ErrUnknownProtocolOption - } -} - func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 3e520d2ee..c59297c80 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -115,9 +115,6 @@ type socketOpsCommon struct { // bound, they cannot be modified. abstractName string abstractNamespace *kernel.AbstractSocketNamespace - - // ops is used to get socket level options. - ops tcpip.SocketOptions } func (s *socketOpsCommon) isPacket() bool { @@ -139,7 +136,7 @@ func (s *socketOpsCommon) Endpoint() transport.Endpoint { // extractPath extracts and validates the address. func extractPath(sockaddr []byte) (string, *syserr.Error) { - addr, family, err := netstack.AddressAndFamily(sockaddr) + addr, family, err := socket.AddressAndFamily(sockaddr) if err != nil { if err == syserr.ErrAddressFamilyNotSupported { err = syserr.ErrInvalidArgument @@ -172,7 +169,7 @@ func (s *socketOpsCommon) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, * return nil, 0, syserr.TranslateNetstackError(err) } - a, l := netstack.ConvertAddress(linux.AF_UNIX, addr) + a, l := socket.ConvertAddress(linux.AF_UNIX, addr) return a, l, nil } @@ -184,7 +181,7 @@ func (s *socketOpsCommon) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, * return nil, 0, syserr.TranslateNetstackError(err) } - a, l := netstack.ConvertAddress(linux.AF_UNIX, addr) + a, l := socket.ConvertAddress(linux.AF_UNIX, addr) return a, l, nil } @@ -258,7 +255,7 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, var addr linux.SockAddr var addrLen uint32 if peerAddr != nil { - addr, addrLen = netstack.ConvertAddress(linux.AF_UNIX, *peerAddr) + addr, addrLen = socket.ConvertAddress(linux.AF_UNIX, *peerAddr) } fd, e := t.NewFDFrom(0, ns, kernel.FDFlags{ @@ -650,7 +647,7 @@ func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags var from linux.SockAddr var fromLen uint32 if r.From != nil && len([]byte(r.From.Addr)) != 0 { - from, fromLen = netstack.ConvertAddress(linux.AF_UNIX, *r.From) + from, fromLen = socket.ConvertAddress(linux.AF_UNIX, *r.From) } if r.ControlTrunc { @@ -685,7 +682,7 @@ func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags var from linux.SockAddr var fromLen uint32 if r.From != nil { - from, fromLen = netstack.ConvertAddress(linux.AF_UNIX, *r.From) + from, fromLen = socket.ConvertAddress(linux.AF_UNIX, *r.From) } if r.ControlTrunc { diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index eaf0b0d26..27f705bb2 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -172,7 +172,7 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block var addr linux.SockAddr var addrLen uint32 if peerAddr != nil { - addr, addrLen = netstack.ConvertAddress(linux.AF_UNIX, *peerAddr) + addr, addrLen = socket.ConvertAddress(linux.AF_UNIX, *peerAddr) } fd, e := t.NewFDFromVFS2(0, ns, kernel.FDFlags{ diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index a920180d3..d36a64ffc 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -32,8 +32,8 @@ go_library( "//pkg/seccomp", "//pkg/sentry/arch", "//pkg/sentry/kernel", + "//pkg/sentry/socket", "//pkg/sentry/socket/netlink", - "//pkg/sentry/socket/netstack", "//pkg/sentry/syscalls/linux", "//pkg/usermem", ], diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index cc5f70cd4..d943a7cb1 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netlink" - "gvisor.dev/gvisor/pkg/sentry/socket/netstack" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/usermem" ) @@ -341,7 +341,7 @@ func sockAddr(t *kernel.Task, addr usermem.Addr, length uint32) string { switch family { case linux.AF_INET, linux.AF_INET6, linux.AF_UNIX: - fa, _, err := netstack.AddressAndFamily(b) + fa, _, err := socket.AddressAndFamily(b) if err != nil { return fmt.Sprintf("%#x {Family: %s, error extracting address: %v}", addr, familyStr, err) } diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index bb1f715e2..cff442846 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -118,7 +118,7 @@ var AMD64 = &kernel.SyscallTable{ 63: syscalls.Supported("uname", Uname), 64: syscalls.Supported("semget", Semget), 65: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), - 66: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, SEM_STAT, SEM_STAT_ANY not supported.", nil), + 66: syscalls.PartiallySupported("semctl", Semctl, "Options SEM_INFO, SEM_STAT, SEM_STAT_ANY not supported.", nil), 67: syscalls.Supported("shmdt", Shmdt), 68: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) 69: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) @@ -619,7 +619,7 @@ var ARM64 = &kernel.SyscallTable{ 188: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) 189: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) 190: syscalls.Supported("semget", Semget), - 191: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, SEM_STAT, SEM_STAT_ANY not supported.", nil), + 191: syscalls.PartiallySupported("semctl", Semctl, "Options SEM_INFO, SEM_STAT, SEM_STAT_ANY not supported.", nil), 192: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), 193: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), 194: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 519066a47..8db587401 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -646,7 +646,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } - fSetOwn(t, file, set) + fSetOwn(t, int(fd), file, set) return 0, nil, nil case linux.FIOGETOWN, linux.SIOCGPGRP: @@ -901,8 +901,8 @@ func fGetOwn(t *kernel.Task, file *fs.File) int32 { // // If who is positive, it represents a PID. If negative, it represents a PGID. // If the PID or PGID is invalid, the owner is silently unset. -func fSetOwn(t *kernel.Task, file *fs.File, who int32) error { - a := file.Async(fasync.New).(*fasync.FileAsync) +func fSetOwn(t *kernel.Task, fd int, file *fs.File, who int32) error { + a := file.Async(fasync.New(fd)).(*fasync.FileAsync) if who < 0 { // Check for overflow before flipping the sign. if who-1 > who { @@ -1049,7 +1049,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_GETOWN: return uintptr(fGetOwn(t, file)), nil, nil case linux.F_SETOWN: - return 0, nil, fSetOwn(t, file, args[2].Int()) + return 0, nil, fSetOwn(t, int(fd), file, args[2].Int()) case linux.F_GETOWN_EX: addr := args[2].Pointer() owner := fGetOwnEx(t, file) @@ -1062,7 +1062,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - a := file.Async(fasync.New).(*fasync.FileAsync) + a := file.Async(fasync.New(int(fd))).(*fasync.FileAsync) switch owner.Type { case linux.F_OWNER_TID: task := t.PIDNamespace().TaskWithID(kernel.ThreadID(owner.PID)) @@ -1111,6 +1111,12 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } n, err := sz.SetFifoSize(int64(args[2].Int())) return uintptr(n), nil, err + case linux.F_GETSIG: + a := file.Async(fasync.New(int(fd))).(*fasync.FileAsync) + return uintptr(a.Signal()), nil, nil + case linux.F_SETSIG: + a := file.Async(fasync.New(int(fd))).(*fasync.FileAsync) + return 0, nil, a.SetSignal(linux.Signal(args[2].Int())) default: // Everything else is not yet supported. return 0, nil, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/sys_sem.go b/pkg/sentry/syscalls/linux/sys_sem.go index e383a0a87..a1601676f 100644 --- a/pkg/sentry/syscalls/linux/sys_sem.go +++ b/pkg/sentry/syscalls/linux/sys_sem.go @@ -146,8 +146,15 @@ func Semctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal v, err := getNCnt(t, id, num) return uintptr(v), nil, err - case linux.IPC_INFO, - linux.SEM_INFO, + case linux.IPC_INFO: + buf := args[3].Pointer() + r := t.IPCNamespace().SemaphoreRegistry() + info := r.IPCInfo() + _, err := info.CopyOut(t, buf) + // TODO(gvisor.dev/issue/137): Return the index of the highest used entry. + return 0, nil, err + + case linux.SEM_INFO, linux.SEM_STAT, linux.SEM_STAT_ANY: diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 36e89700e..7dd9ef857 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -165,7 +165,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall ownerType = linux.F_OWNER_PGRP who = -who } - return 0, nil, setAsyncOwner(t, file, ownerType, who) + return 0, nil, setAsyncOwner(t, int(fd), file, ownerType, who) case linux.F_GETOWN_EX: owner, hasOwner := getAsyncOwner(t, file) if !hasOwner { @@ -179,7 +179,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - return 0, nil, setAsyncOwner(t, file, owner.Type, owner.PID) + return 0, nil, setAsyncOwner(t, int(fd), file, owner.Type, owner.PID) case linux.F_SETPIPE_SZ: pipefile, ok := file.Impl().(*pipe.VFSPipeFD) if !ok { @@ -207,6 +207,16 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err case linux.F_SETLK, linux.F_SETLKW: return 0, nil, posixLock(t, args, file, cmd) + case linux.F_GETSIG: + a := file.AsyncHandler() + if a == nil { + // Default behavior aka SIGIO. + return 0, nil, nil + } + return uintptr(a.(*fasync.FileAsync).Signal()), nil, nil + case linux.F_SETSIG: + a := file.SetAsyncHandler(fasync.NewVFS2(int(fd))).(*fasync.FileAsync) + return 0, nil, a.SetSignal(linux.Signal(args[2].Int())) default: // Everything else is not yet supported. return 0, nil, syserror.EINVAL @@ -241,7 +251,7 @@ func getAsyncOwner(t *kernel.Task, fd *vfs.FileDescription) (ownerEx linux.FOwne } } -func setAsyncOwner(t *kernel.Task, fd *vfs.FileDescription, ownerType, pid int32) error { +func setAsyncOwner(t *kernel.Task, fd int, file *vfs.FileDescription, ownerType, pid int32) error { switch ownerType { case linux.F_OWNER_TID, linux.F_OWNER_PID, linux.F_OWNER_PGRP: // Acceptable type. @@ -249,7 +259,7 @@ func setAsyncOwner(t *kernel.Task, fd *vfs.FileDescription, ownerType, pid int32 return syserror.EINVAL } - a := fd.SetAsyncHandler(fasync.NewVFS2).(*fasync.FileAsync) + a := file.SetAsyncHandler(fasync.NewVFS2(fd)).(*fasync.FileAsync) if pid == 0 { a.ClearOwner() return nil diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go index 2806c3f6f..20c264fef 100644 --- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go @@ -100,7 +100,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall ownerType = linux.F_OWNER_PGRP who = -who } - return 0, nil, setAsyncOwner(t, file, ownerType, who) + return 0, nil, setAsyncOwner(t, int(fd), file, ownerType, who) } ret, err := file.Ioctl(t, t.MemoryManager(), args) diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 9ce4f280a..8bb763a47 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -343,8 +343,8 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Copy data. var ( - n int64 - err error + total int64 + err error ) dw := dualWaiter{ inFile: inFile, @@ -357,13 +357,20 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // can block. nonBlock := outFile.StatusFlags()&linux.O_NONBLOCK != 0 if outIsPipe { - for n < count { - var spliceN int64 - spliceN, err = outPipeFD.SpliceFromNonPipe(t, inFile, offset, count) + for { + var n int64 + n, err = outPipeFD.SpliceFromNonPipe(t, inFile, offset, count-total) if offset != -1 { - offset += spliceN + offset += n + } + total += n + if total == count { + break + } + if err == nil && t.Interrupted() { + err = syserror.ErrInterrupted + break } - n += spliceN if err == syserror.ErrWouldBlock && !nonBlock { err = dw.waitForBoth(t) } @@ -374,7 +381,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } else { // Read inFile to buffer, then write the contents to outFile. buf := make([]byte, count) - for n < count { + for { var readN int64 if offset != -1 { readN, err = inFile.PRead(t, usermem.BytesIOSequence(buf), offset, vfs.ReadOptions{}) @@ -382,7 +389,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } else { readN, err = inFile.Read(t, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) } - n += readN // Write all of the bytes that we read. This may need // multiple write calls to complete. @@ -398,7 +404,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // We didn't complete the write. Only report the bytes that were actually // written, and rewind offsets as needed. notWritten := int64(len(wbuf)) - n -= notWritten + readN -= notWritten if offset == -1 { // We modified the offset of the input file itself during the read // operation. Rewind it. @@ -415,6 +421,16 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc break } } + + total += readN + buf = buf[readN:] + if total == count { + break + } + if err == nil && t.Interrupted() { + err = syserror.ErrInterrupted + break + } if err == syserror.ErrWouldBlock && !nonBlock { err = dw.waitForBoth(t) } @@ -432,7 +448,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } } - if n != 0 { + if total != 0 { inFile.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) @@ -445,7 +461,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // We can only pass a single file to handleIOError, so pick inFile arbitrarily. // This is used only for debugging purposes. - return uintptr(n), nil, slinux.HandleIOErrorVFS2(t, n != 0, err, syserror.ERESTARTSYS, "sendfile", inFile) + return uintptr(total), nil, slinux.HandleIOErrorVFS2(t, total != 0, err, syserror.ERESTARTSYS, "sendfile", inFile) } // dualWaiter is used to wait on one or both vfs.FileDescriptions. It is not diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 440c9307c..a3868bf16 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -105,6 +105,7 @@ go_library( "//pkg/sentry/arch", "//pkg/sentry/fs", "//pkg/sentry/fs/lock", + "//pkg/sentry/fsmetric", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", "//pkg/sentry/limits", diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index a98aac52b..072655fe8 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -204,8 +204,8 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, event lin file.EventRegister(&epi.waiter, wmask) // Check if the file is already ready. - if file.Readiness(wmask)&wmask != 0 { - epi.Callback(nil) + if m := file.Readiness(wmask) & wmask; m != 0 { + epi.Callback(nil, m) } // Add epi to file.epolls so that it is removed when the last @@ -274,8 +274,8 @@ func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, event file.EventRegister(&epi.waiter, wmask) // Check if the file is already ready with the new mask. - if file.Readiness(wmask)&wmask != 0 { - epi.Callback(nil) + if m := file.Readiness(wmask) & wmask; m != 0 { + epi.Callback(nil, m) } return nil @@ -311,7 +311,7 @@ func (ep *EpollInstance) DeleteInterest(file *FileDescription, num int32) error } // Callback implements waiter.EntryCallback.Callback. -func (epi *epollInterest) Callback(*waiter.Entry) { +func (epi *epollInterest) Callback(*waiter.Entry, waiter.EventMask) { newReady := false epi.epoll.mu.Lock() if !epi.ready { diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 936f9fc71..5321ac80a 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -15,12 +15,14 @@ package vfs import ( + "io" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sync" @@ -42,7 +44,7 @@ import ( type FileDescription struct { FileDescriptionRefs - // flagsMu protects statusFlags and asyncHandler below. + // flagsMu protects `statusFlags`, `saved`, and `asyncHandler` below. flagsMu sync.Mutex `state:"nosave"` // statusFlags contains status flags, "initialized by open(2) and possibly @@ -51,6 +53,11 @@ type FileDescription struct { // access to asyncHandler. statusFlags uint32 + // saved is true after beforeSave is called. This is used to prevent + // double-unregistration of asyncHandler. This does not work properly for + // save-resume, which is not currently supported in gVisor (see b/26588733). + saved bool `state:"nosave"` + // asyncHandler handles O_ASYNC signal generation. It is set with the // F_SETOWN or F_SETOWN_EX fcntls. For asyncHandler to be used, O_ASYNC must // also be set by fcntl(2). @@ -183,7 +190,7 @@ func (fd *FileDescription) DecRef(ctx context.Context) { } fd.vd.DecRef(ctx) fd.flagsMu.Lock() - if fd.statusFlags&linux.O_ASYNC != 0 && fd.asyncHandler != nil { + if !fd.saved && fd.statusFlags&linux.O_ASYNC != 0 && fd.asyncHandler != nil { fd.asyncHandler.Unregister(fd) } fd.asyncHandler = nil @@ -583,7 +590,11 @@ func (fd *FileDescription) PRead(ctx context.Context, dst usermem.IOSequence, of if !fd.readable { return 0, syserror.EBADF } - return fd.impl.PRead(ctx, dst, offset, opts) + start := fsmetric.StartReadWait() + n, err := fd.impl.PRead(ctx, dst, offset, opts) + fsmetric.Reads.Increment() + fsmetric.FinishReadWait(fsmetric.ReadWait, start) + return n, err } // Read is similar to PRead, but does not specify an offset. @@ -591,7 +602,11 @@ func (fd *FileDescription) Read(ctx context.Context, dst usermem.IOSequence, opt if !fd.readable { return 0, syserror.EBADF } - return fd.impl.Read(ctx, dst, opts) + start := fsmetric.StartReadWait() + n, err := fd.impl.Read(ctx, dst, opts) + fsmetric.Reads.Increment() + fsmetric.FinishReadWait(fsmetric.ReadWait, start) + return n, err } // PWrite writes src to the file represented by fd, starting at the given @@ -825,44 +840,27 @@ func (fd *FileDescription) SetAsyncHandler(newHandler func() FileAsync) FileAsyn return fd.asyncHandler } -// FileReadWriteSeeker is a helper struct to pass a FileDescription as -// io.Reader/io.Writer/io.ReadSeeker/io.ReaderAt/io.WriterAt/etc. -type FileReadWriteSeeker struct { - FD *FileDescription - Ctx context.Context - ROpts ReadOptions - WOpts WriteOptions -} - -// ReadAt implements io.ReaderAt.ReadAt. -func (f *FileReadWriteSeeker) ReadAt(p []byte, off int64) (int, error) { - dst := usermem.BytesIOSequence(p) - n, err := f.FD.PRead(f.Ctx, dst, off, f.ROpts) - return int(n), err -} - -// Read implements io.ReadWriteSeeker.Read. -func (f *FileReadWriteSeeker) Read(p []byte) (int, error) { - dst := usermem.BytesIOSequence(p) - n, err := f.FD.Read(f.Ctx, dst, f.ROpts) - return int(n), err -} - -// Seek implements io.ReadWriteSeeker.Seek. -func (f *FileReadWriteSeeker) Seek(offset int64, whence int) (int64, error) { - return f.FD.Seek(f.Ctx, offset, int32(whence)) -} - -// WriteAt implements io.WriterAt.WriteAt. -func (f *FileReadWriteSeeker) WriteAt(p []byte, off int64) (int, error) { - dst := usermem.BytesIOSequence(p) - n, err := f.FD.PWrite(f.Ctx, dst, off, f.WOpts) - return int(n), err -} - -// Write implements io.ReadWriteSeeker.Write. -func (f *FileReadWriteSeeker) Write(p []byte) (int, error) { - buf := usermem.BytesIOSequence(p) - n, err := f.FD.Write(f.Ctx, buf, f.WOpts) - return int(n), err +// CopyRegularFileData copies data from srcFD to dstFD until reading from srcFD +// returns EOF or an error. It returns the number of bytes copied. +func CopyRegularFileData(ctx context.Context, dstFD, srcFD *FileDescription) (int64, error) { + done := int64(0) + buf := usermem.BytesIOSequence(make([]byte, 32*1024)) // arbitrary buffer size + for { + readN, readErr := srcFD.Read(ctx, buf, ReadOptions{}) + if readErr != nil && readErr != io.EOF { + return done, readErr + } + src := buf.TakeFirst64(readN) + for src.NumBytes() != 0 { + writeN, writeErr := dstFD.Write(ctx, src, WriteOptions{}) + done += writeN + src = src.DropFirst64(writeN) + if writeErr != nil { + return done, writeErr + } + } + if readErr == io.EOF { + return done, nil + } + } } diff --git a/pkg/sentry/vfs/save_restore.go b/pkg/sentry/vfs/save_restore.go index 7723ed643..8f070ed53 100644 --- a/pkg/sentry/vfs/save_restore.go +++ b/pkg/sentry/vfs/save_restore.go @@ -18,8 +18,10 @@ import ( "fmt" "sync/atomic" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refsvfs2" + "gvisor.dev/gvisor/pkg/waiter" ) // FilesystemImplSaveRestoreExtension is an optional extension to @@ -120,5 +122,20 @@ func (mnt *Mount) afterLoad() { func (epi *epollInterest) afterLoad() { // Mark all epollInterests as ready after restore so that the next call to // EpollInstance.ReadEvents() rechecks their readiness. - epi.Callback(nil) + epi.Callback(nil, waiter.EventMaskFromLinux(epi.mask)) +} + +// beforeSave is called by stateify. +func (fd *FileDescription) beforeSave() { + fd.saved = true + if fd.statusFlags&linux.O_ASYNC != 0 && fd.asyncHandler != nil { + fd.asyncHandler.Unregister(fd) + } +} + +// afterLoad is called by stateify. +func (fd *FileDescription) afterLoad() { + if fd.statusFlags&linux.O_ASYNC != 0 && fd.asyncHandler != nil { + fd.asyncHandler.Register(fd) + } } diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 48d6252f7..6fd1bb0b2 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -41,6 +41,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/fsmetric" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sync" @@ -381,6 +382,8 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia // OpenAt returns a FileDescription providing access to the file at the given // path. A reference is taken on the returned FileDescription. func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { + fsmetric.Opens.Increment() + // Remove: // // - O_CLOEXEC, which affects file descriptors and therefore must be diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index 1d1062aeb..8e3146d8d 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -338,6 +338,7 @@ func (w *Watchdog) report(offenders map[*kernel.Task]*offender, newTaskFound boo tid := w.k.TaskSet().Root.IDOfTask(t) buf.WriteString(fmt.Sprintf("\tTask tid: %v (goroutine %d), entered RunSys state %v ago.\n", tid, t.GoroutineID(), now.Sub(o.lastUpdateTime))) } + buf.WriteString("Search for 'goroutine <id>' in the stack dump to find the offending goroutine(s)") // Force stack dump only if a new task is detected. w.doAction(w.TaskTimeoutAction, newTaskFound, &buf) diff --git a/pkg/shim/v2/service.go b/pkg/shim/v2/service.go index 2e39d2c4a..cba403cae 100644 --- a/pkg/shim/v2/service.go +++ b/pkg/shim/v2/service.go @@ -67,9 +67,15 @@ var ( var _ = (taskAPI.TaskService)(&service{}) -// configFile is the default config file name. For containerd 1.2, -// we assume that a config.toml should exist in the runtime root. -const configFile = "config.toml" +const ( + // configFile is the default config file name. For containerd 1.2, + // we assume that a config.toml should exist in the runtime root. + configFile = "config.toml" + + // shimAddressPath is the relative path to a file that contains the address + // to the shim UDS. See service.shimAddress. + shimAddressPath = "address" +) // New returns a new shim service that can be used via GRPC. func New(ctx context.Context, id string, publisher shim.Publisher, cancel func()) (shim.Shim, error) { @@ -101,6 +107,11 @@ func New(ctx context.Context, id string, publisher shim.Publisher, cancel func() return nil, fmt.Errorf("failed to initialized platform behavior: %w", err) } go s.forward(ctx, publisher) + + if address, err := shim.ReadAddress(shimAddressPath); err == nil { + s.shimAddress = address + } + return s, nil } @@ -152,6 +163,9 @@ type service struct { // cancel is a function that needs to be called before the shim stops. The // function is provided by the caller to New(). cancel func() + + // shimAddress is the location of the UDS used to communicate to containerd. + shimAddress string } func (s *service) newCommand(ctx context.Context, containerdBinary, containerdAddress string) (*exec.Cmd, error) { @@ -191,38 +205,58 @@ func (s *service) StartShim(ctx context.Context, id, containerdBinary, container if err != nil { return "", err } - address, err := shim.SocketAddress(ctx, id) + address, err := shim.SocketAddress(ctx, containerdAddress, id) if err != nil { return "", err } socket, err := shim.NewSocket(address) if err != nil { - return "", err + // The only time where this would happen is if there is a bug and the socket + // was not cleaned up in the cleanup method of the shim or we are using the + // grouping functionality where the new process should be run with the same + // shim as an existing container. + if !shim.SocketEaddrinuse(err) { + return "", fmt.Errorf("create new shim socket: %w", err) + } + if shim.CanConnect(address) { + if err := shim.WriteAddress(shimAddressPath, address); err != nil { + return "", fmt.Errorf("write existing socket for shim: %w", err) + } + return address, nil + } + if err := shim.RemoveSocket(address); err != nil { + return "", fmt.Errorf("remove pre-existing socket: %w", err) + } + if socket, err = shim.NewSocket(address); err != nil { + return "", fmt.Errorf("try create new shim socket 2x: %w", err) + } } - defer socket.Close() + cu := cleanup.Make(func() { + socket.Close() + _ = shim.RemoveSocket(address) + }) + defer cu.Clean() + f, err := socket.File() if err != nil { return "", err } - defer f.Close() cmd.ExtraFiles = append(cmd.ExtraFiles, f) log.L.Debugf("Executing: %q %s", cmd.Path, cmd.Args) if err := cmd.Start(); err != nil { + f.Close() return "", err } - cu := cleanup.Make(func() { - cmd.Process.Kill() - }) - defer cu.Clean() + cu.Add(func() { cmd.Process.Kill() }) // make sure to wait after start go cmd.Wait() if err := shim.WritePidFile("shim.pid", cmd.Process.Pid); err != nil { return "", err } - if err := shim.WriteAddress("address", address); err != nil { + if err := shim.WriteAddress(shimAddressPath, address); err != nil { return "", err } if err := shim.SetScore(cmd.Process.Pid); err != nil { @@ -675,8 +709,11 @@ func (s *service) Connect(ctx context.Context, r *taskAPI.ConnectRequest) (*task func (s *service) Shutdown(ctx context.Context, r *taskAPI.ShutdownRequest) (*types.Empty, error) { log.L.Debugf("Shutdown, id: %s", r.ID) s.cancel() + if s.shimAddress != "" { + _ = shim.RemoveSocket(s.shimAddress) + } os.Exit(0) - return empty, nil + panic("Should not get here") } func (s *service) Stats(ctx context.Context, r *taskAPI.StatsRequest) (*taskAPI.StatsResponse, error) { @@ -843,9 +880,7 @@ func (s *service) getContainerPids(ctx context.Context, id string) ([]uint32, er func (s *service) forward(ctx context.Context, publisher shim.Publisher) { for e := range s.events { - ctx, cancel := context.WithTimeout(ctx, 5*time.Second) err := publisher.Publish(ctx, getTopic(e), e) - cancel() if err != nil { // Should not happen. panic(fmt.Errorf("post event: %w", err)) diff --git a/pkg/sleep/BUILD b/pkg/sleep/BUILD index ae0fe1522..48bcdd62b 100644 --- a/pkg/sleep/BUILD +++ b/pkg/sleep/BUILD @@ -5,10 +5,6 @@ package(licenses = ["notice"]) go_library( name = "sleep", srcs = [ - "commit_amd64.s", - "commit_arm64.s", - "commit_asm.go", - "commit_noasm.go", "sleep_unsafe.go", ], visibility = ["//:sandbox"], diff --git a/pkg/sleep/commit_amd64.s b/pkg/sleep/commit_amd64.s deleted file mode 100644 index bc4ac2c3c..000000000 --- a/pkg/sleep/commit_amd64.s +++ /dev/null @@ -1,35 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -#include "textflag.h" - -#define preparingG 1 - -// See commit_noasm.go for a description of commitSleep. -// -// func commitSleep(g uintptr, waitingG *uintptr) bool -TEXT ·commitSleep(SB),NOSPLIT,$0-24 - MOVQ waitingG+8(FP), CX - MOVQ g+0(FP), DX - - // Store the G in waitingG if it's still preparingG. If it's anything - // else it means a waker has aborted the sleep. - MOVQ $preparingG, AX - LOCK - CMPXCHGQ DX, 0(CX) - - SETEQ AX - MOVB AX, ret+16(FP) - - RET diff --git a/pkg/sleep/commit_arm64.s b/pkg/sleep/commit_arm64.s deleted file mode 100644 index d0ef15b20..000000000 --- a/pkg/sleep/commit_arm64.s +++ /dev/null @@ -1,38 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -#include "textflag.h" - -#define preparingG 1 - -// See commit_noasm.go for a description of commitSleep. -// -// func commitSleep(g uintptr, waitingG *uintptr) bool -TEXT ·commitSleep(SB),NOSPLIT,$0-24 - MOVD waitingG+8(FP), R0 - MOVD $preparingG, R1 - MOVD G+0(FP), R2 - - // Store the G in waitingG if it's still preparingG. If it's anything - // else it means a waker has aborted the sleep. -again: - LDAXR (R0), R3 - CMP R1, R3 - BNE ok - STLXR R2, (R0), R3 - CBNZ R3, again -ok: - CSET EQ, R0 - MOVB R0, ret+16(FP) - RET diff --git a/pkg/sleep/commit_asm.go b/pkg/sleep/commit_asm.go deleted file mode 100644 index 75728a97d..000000000 --- a/pkg/sleep/commit_asm.go +++ /dev/null @@ -1,20 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build amd64 arm64 - -package sleep - -// See commit_noasm.go for a description of commitSleep. -func commitSleep(g uintptr, waitingG *uintptr) bool diff --git a/pkg/sleep/commit_noasm.go b/pkg/sleep/commit_noasm.go deleted file mode 100644 index f59061f37..000000000 --- a/pkg/sleep/commit_noasm.go +++ /dev/null @@ -1,33 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// +build !race -// +build !amd64,!arm64 - -package sleep - -import "sync/atomic" - -// commitSleep signals to wakers that the given g is now sleeping. Wakers can -// then fetch it and wake it. -// -// The commit may fail if wakers have been asserted after our last check, in -// which case they will have set s.waitingG to zero. -// -// It is written in assembly because it is called from g0, so it doesn't have -// a race context. -func commitSleep(g uintptr, waitingG *uintptr) bool { - // Try to store the G so that wakers know who to wake. - return atomic.CompareAndSwapUintptr(waitingG, preparingG, g) -} diff --git a/pkg/sleep/sleep_unsafe.go b/pkg/sleep/sleep_unsafe.go index 19bce2afb..c44206b1e 100644 --- a/pkg/sleep/sleep_unsafe.go +++ b/pkg/sleep/sleep_unsafe.go @@ -12,11 +12,6 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build go1.11 -// +build !go1.17 - -// Check go:linkname function signatures when updating Go version. - // Package sleep allows goroutines to efficiently sleep on multiple sources of // notifications (wakers). It offers O(1) complexity, which is different from // multi-channel selects which have O(n) complexity (where n is the number of @@ -91,12 +86,6 @@ var ( assertedSleeper Sleeper ) -//go:linkname gopark runtime.gopark -func gopark(unlockf func(uintptr, *uintptr) bool, wg *uintptr, reason uint8, traceEv byte, traceskip int) - -//go:linkname goready runtime.goready -func goready(g uintptr, traceskip int) - // Sleeper allows a goroutine to sleep and receive wake up notifications from // Wakers in an efficient way. // @@ -189,7 +178,7 @@ func (s *Sleeper) nextWaker(block bool) *Waker { // See:runtime2.go in the go runtime package for // the values to pass as the waitReason here. const waitReasonSelect = 9 - gopark(commitSleep, &s.waitingG, waitReasonSelect, traceEvGoBlockSelect, 0) + sync.Gopark(commitSleep, unsafe.Pointer(&s.waitingG), sync.WaitReasonSelect, sync.TraceEvGoBlockSelect, 0) } // Pull the shared list out and reverse it in the local @@ -212,6 +201,18 @@ func (s *Sleeper) nextWaker(block bool) *Waker { return w } +// commitSleep signals to wakers that the given g is now sleeping. Wakers can +// then fetch it and wake it. +// +// The commit may fail if wakers have been asserted after our last check, in +// which case they will have set s.waitingG to zero. +// +//go:norace +//go:nosplit +func commitSleep(g uintptr, waitingG unsafe.Pointer) bool { + return sync.RaceUncheckedAtomicCompareAndSwapUintptr((*uintptr)(waitingG), preparingG, g) +} + // Fetch fetches the next wake-up notification. If a notification is immediately // available, it is returned right away. Otherwise, the behavior depends on the // value of 'block': if true, the current goroutine blocks until a notification @@ -311,7 +312,7 @@ func (s *Sleeper) enqueueAssertedWaker(w *Waker) { case 0, preparingG: default: // We managed to get a G. Wake it up. - goready(g, 0) + sync.Goready(g, 0) } } diff --git a/pkg/sync/BUILD b/pkg/sync/BUILD index 68535c3b1..be5bc99fc 100644 --- a/pkg/sync/BUILD +++ b/pkg/sync/BUILD @@ -31,18 +31,26 @@ go_library( name = "sync", srcs = [ "aliases.go", - "memmove_unsafe.go", + "checklocks_off_unsafe.go", + "checklocks_on_unsafe.go", + "goyield_go113_unsafe.go", + "goyield_unsafe.go", "mutex_unsafe.go", "nocopy.go", "norace_unsafe.go", + "race_amd64.s", + "race_arm64.s", "race_unsafe.go", + "runtime_unsafe.go", "rwmutex_unsafe.go", "seqcount.go", - "spin_unsafe.go", "sync.go", ], marshal = False, stateify = False, + deps = [ + "//pkg/goid", + ], ) go_test( diff --git a/pkg/sync/checklocks_off_unsafe.go b/pkg/sync/checklocks_off_unsafe.go new file mode 100644 index 000000000..62c81b149 --- /dev/null +++ b/pkg/sync/checklocks_off_unsafe.go @@ -0,0 +1,18 @@ +// Copyright 2020 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build !checklocks + +package sync + +import ( + "unsafe" +) + +func noteLock(l unsafe.Pointer) { +} + +func noteUnlock(l unsafe.Pointer) { +} diff --git a/pkg/sync/checklocks_on_unsafe.go b/pkg/sync/checklocks_on_unsafe.go new file mode 100644 index 000000000..24f933ed1 --- /dev/null +++ b/pkg/sync/checklocks_on_unsafe.go @@ -0,0 +1,108 @@ +// Copyright 2020 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build checklocks + +package sync + +import ( + "fmt" + "strings" + "sync" + "unsafe" + + "gvisor.dev/gvisor/pkg/goid" +) + +// gLocks contains metadata about the locks held by a goroutine. +type gLocks struct { + locksHeld []unsafe.Pointer +} + +// map[goid int]*gLocks +// +// Each key may only be written by the G with the goid it refers to. +// +// Note that entries are not evicted when a G exit, causing unbounded growth +// with new G creation / destruction. If this proves problematic, entries could +// be evicted when no locks are held at the expense of more allocations when +// taking top-level locks. +var locksHeld sync.Map + +func getGLocks() *gLocks { + id := goid.Get() + + var locks *gLocks + if l, ok := locksHeld.Load(id); ok { + locks = l.(*gLocks) + } else { + locks = &gLocks{ + // Initialize space for a few locks. + locksHeld: make([]unsafe.Pointer, 0, 8), + } + locksHeld.Store(id, locks) + } + + return locks +} + +func noteLock(l unsafe.Pointer) { + locks := getGLocks() + + for _, lock := range locks.locksHeld { + if lock == l { + panic(fmt.Sprintf("Deadlock on goroutine %d! Double lock of %p: %+v", goid.Get(), l, locks)) + } + } + + // Commit only after checking for panic conditions so that this lock + // isn't on the list if the above panic is recovered. + locks.locksHeld = append(locks.locksHeld, l) +} + +func noteUnlock(l unsafe.Pointer) { + locks := getGLocks() + + if len(locks.locksHeld) == 0 { + panic(fmt.Sprintf("Unlock of %p on goroutine %d without any locks held! All locks:\n%s", l, goid.Get(), dumpLocks())) + } + + // Search backwards since callers are most likely to unlock in LIFO order. + length := len(locks.locksHeld) + for i := length - 1; i >= 0; i-- { + if l == locks.locksHeld[i] { + copy(locks.locksHeld[i:length-1], locks.locksHeld[i+1:length]) + // Clear last entry to ensure addr can be GC'd. + locks.locksHeld[length-1] = nil + locks.locksHeld = locks.locksHeld[:length-1] + return + } + } + + panic(fmt.Sprintf("Unlock of %p on goroutine %d without matching lock! All locks:\n%s", l, goid.Get(), dumpLocks())) +} + +func dumpLocks() string { + var s strings.Builder + locksHeld.Range(func(key, value interface{}) bool { + goid := key.(int64) + locks := value.(*gLocks) + + // N.B. accessing gLocks of another G is fundamentally racy. + + fmt.Fprintf(&s, "goroutine %d:\n", goid) + if len(locks.locksHeld) == 0 { + fmt.Fprintf(&s, "\t<none>\n") + } + for _, lock := range locks.locksHeld { + fmt.Fprintf(&s, "\t%p\n", lock) + } + fmt.Fprintf(&s, "\n") + + return true + }) + + return s.String() +} diff --git a/pkg/sync/goyield_go113_unsafe.go b/pkg/sync/goyield_go113_unsafe.go new file mode 100644 index 000000000..8aee0d455 --- /dev/null +++ b/pkg/sync/goyield_go113_unsafe.go @@ -0,0 +1,18 @@ +// Copyright 2020 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.13 +// +build !go1.14 + +package sync + +import ( + "runtime" +) + +func goyield() { + // goyield is not available until Go 1.14. + runtime.Gosched() +} diff --git a/pkg/sync/spin_unsafe.go b/pkg/sync/goyield_unsafe.go index cafb2d065..672ee274d 100644 --- a/pkg/sync/spin_unsafe.go +++ b/pkg/sync/goyield_unsafe.go @@ -3,7 +3,7 @@ // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. -// +build go1.13 +// +build go1.14 // +build !go1.17 // Check go:linkname function signatures when updating Go version. @@ -14,11 +14,5 @@ import ( _ "unsafe" // for go:linkname ) -//go:linkname canSpin sync.runtime_canSpin -func canSpin(i int) bool - -//go:linkname doSpin sync.runtime_doSpin -func doSpin() - //go:linkname goyield runtime.goyield func goyield() diff --git a/pkg/sync/memmove_unsafe.go b/pkg/sync/memmove_unsafe.go deleted file mode 100644 index f5e630009..000000000 --- a/pkg/sync/memmove_unsafe.go +++ /dev/null @@ -1,28 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Use of this source code is governed by a BSD-style -// license that can be found in the LICENSE file. - -// +build go1.12 -// +build !go1.17 - -// Check go:linkname function signatures when updating Go version. - -package sync - -import ( - "unsafe" -) - -//go:linkname memmove runtime.memmove -//go:noescape -func memmove(to, from unsafe.Pointer, n uintptr) - -// Memmove is exported for SeqAtomicLoad/SeqAtomicTryLoad<T>, which can't -// define it because go_generics can't update the go:linkname annotation. -// Furthermore, go:linkname silently doesn't work if the local name is exported -// (this is of course undocumented), which is why this indirection is -// necessary. -func Memmove(to, from unsafe.Pointer, n uintptr) { - memmove(to, from, n) -} diff --git a/pkg/sync/mutex_test.go b/pkg/sync/mutex_test.go index 0838248b4..4fb51a8ab 100644 --- a/pkg/sync/mutex_test.go +++ b/pkg/sync/mutex_test.go @@ -32,11 +32,11 @@ func TestStructSize(t *testing.T) { func TestFieldValues(t *testing.T) { var m Mutex m.Lock() - if got := *m.state(); got != mutexLocked { + if got := *m.m.state(); got != mutexLocked { t.Errorf("got locked sync.Mutex.state = %d, want = %d", got, mutexLocked) } m.Unlock() - if got := *m.state(); got != mutexUnlocked { + if got := *m.m.state(); got != mutexUnlocked { t.Errorf("got unlocked sync.Mutex.state = %d, want = %d", got, mutexUnlocked) } } diff --git a/pkg/sync/mutex_unsafe.go b/pkg/sync/mutex_unsafe.go index f4c2e9642..21084b857 100644 --- a/pkg/sync/mutex_unsafe.go +++ b/pkg/sync/mutex_unsafe.go @@ -17,8 +17,9 @@ import ( "unsafe" ) -// Mutex is a try lock. -type Mutex struct { +// CrossGoroutineMutex is equivalent to Mutex, but it need not be unlocked by a +// the same goroutine that locked the mutex. +type CrossGoroutineMutex struct { sync.Mutex } @@ -27,7 +28,7 @@ type syncMutex struct { sema uint32 } -func (m *Mutex) state() *int32 { +func (m *CrossGoroutineMutex) state() *int32 { return &(*syncMutex)(unsafe.Pointer(&m.Mutex)).state } @@ -36,9 +37,9 @@ const ( mutexLocked = 1 ) -// TryLock tries to aquire the mutex. It returns true if it succeeds and false +// TryLock tries to acquire the mutex. It returns true if it succeeds and false // otherwise. TryLock does not block. -func (m *Mutex) TryLock() bool { +func (m *CrossGoroutineMutex) TryLock() bool { if atomic.CompareAndSwapInt32(m.state(), mutexUnlocked, mutexLocked) { if RaceEnabled { RaceAcquire(unsafe.Pointer(&m.Mutex)) @@ -47,3 +48,43 @@ func (m *Mutex) TryLock() bool { } return false } + +// Mutex is a mutual exclusion lock. The zero value for a Mutex is an unlocked +// mutex. +// +// A Mutex must not be copied after first use. +// +// A Mutex must be unlocked by the same goroutine that locked it. This +// invariant is enforced with the 'checklocks' build tag. +type Mutex struct { + m CrossGoroutineMutex +} + +// Lock locks m. If the lock is already in use, the calling goroutine blocks +// until the mutex is available. +func (m *Mutex) Lock() { + noteLock(unsafe.Pointer(m)) + m.m.Lock() +} + +// Unlock unlocks m. +// +// Preconditions: +// * m is locked. +// * m was locked by this goroutine. +func (m *Mutex) Unlock() { + noteUnlock(unsafe.Pointer(m)) + m.m.Unlock() +} + +// TryLock tries to acquire the mutex. It returns true if it succeeds and false +// otherwise. TryLock does not block. +func (m *Mutex) TryLock() bool { + // Note lock first to enforce proper locking even if unsuccessful. + noteLock(unsafe.Pointer(m)) + locked := m.m.TryLock() + if !locked { + noteUnlock(unsafe.Pointer(m)) + } + return locked +} diff --git a/pkg/sync/norace_unsafe.go b/pkg/sync/norace_unsafe.go index 006055dd6..70b5f3a5e 100644 --- a/pkg/sync/norace_unsafe.go +++ b/pkg/sync/norace_unsafe.go @@ -8,6 +8,7 @@ package sync import ( + "sync/atomic" "unsafe" ) @@ -33,3 +34,13 @@ func RaceRelease(addr unsafe.Pointer) { // RaceReleaseMerge has the same semantics as runtime.RaceReleaseMerge. func RaceReleaseMerge(addr unsafe.Pointer) { } + +// RaceUncheckedAtomicCompareAndSwapUintptr is equivalent to +// sync/atomic.CompareAndSwapUintptr, but is not checked by the race detector. +// This is necessary when implementing gopark callbacks, since no race context +// is available during their execution. +func RaceUncheckedAtomicCompareAndSwapUintptr(ptr *uintptr, old, new uintptr) bool { + // Use atomic.CompareAndSwapUintptr outside of race builds for + // inlinability. + return atomic.CompareAndSwapUintptr(ptr, old, new) +} diff --git a/pkg/syncevent/waiter_amd64.s b/pkg/sync/race_amd64.s index 5e216b045..57bc0ec79 100644 --- a/pkg/syncevent/waiter_amd64.s +++ b/pkg/sync/race_amd64.s @@ -12,21 +12,22 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build race +// +build amd64 + #include "textflag.h" -// See waiter_noasm_unsafe.go for a description of waiterUnlock. -// -// func waiterUnlock(ptr unsafe.Pointer, wg *unsafe.Pointer) bool -TEXT ·waiterUnlock(SB),NOSPLIT,$0-24 +// func RaceUncheckedAtomicCompareAndSwapUintptr(ptr *uintptr, old, new uintptr) bool +TEXT ·RaceUncheckedAtomicCompareAndSwapUintptr(SB),NOSPLIT,$0-25 MOVQ ptr+0(FP), DI - MOVQ wg+8(FP), SI + MOVQ old+8(FP), AX + MOVQ new+16(FP), SI - MOVQ $·preparingG(SB), AX LOCK - CMPXCHGQ DI, 0(SI) + CMPXCHGQ SI, 0(DI) SETEQ AX - MOVB AX, ret+16(FP) + MOVB AX, ret+24(FP) RET diff --git a/pkg/syncevent/waiter_arm64.s b/pkg/sync/race_arm64.s index f4c06f194..88f091fda 100644 --- a/pkg/syncevent/waiter_arm64.s +++ b/pkg/sync/race_arm64.s @@ -12,15 +12,16 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build race +// +build arm64 + #include "textflag.h" -// See waiter_noasm_unsafe.go for a description of waiterUnlock. -// -// func waiterUnlock(ptr unsafe.Pointer, wg *unsafe.Pointer) bool -TEXT ·waiterUnlock(SB),NOSPLIT,$0-24 - MOVD wg+8(FP), R0 - MOVD $·preparingG(SB), R1 - MOVD ptr+0(FP), R2 +// func RaceUncheckedAtomicCompareAndSwapUintptr(ptr *uintptr, old, new uintptr) bool +TEXT ·RaceUncheckedAtomicCompareAndSwapUintptr(SB),NOSPLIT,$0-25 + MOVD ptr+0(FP), R0 + MOVD old+8(FP), R1 + MOVD new+16(FP), R1 again: LDAXR (R0), R3 CMP R1, R3 @@ -29,6 +30,6 @@ again: CBNZ R3, again ok: CSET EQ, R0 - MOVB R0, ret+16(FP) + MOVB R0, ret+24(FP) RET diff --git a/pkg/sync/race_unsafe.go b/pkg/sync/race_unsafe.go index 31d8fa9a6..59985c270 100644 --- a/pkg/sync/race_unsafe.go +++ b/pkg/sync/race_unsafe.go @@ -39,3 +39,9 @@ func RaceRelease(addr unsafe.Pointer) { func RaceReleaseMerge(addr unsafe.Pointer) { runtime.RaceReleaseMerge(addr) } + +// RaceUncheckedAtomicCompareAndSwapUintptr is equivalent to +// sync/atomic.CompareAndSwapUintptr, but is not checked by the race detector. +// This is necessary when implementing gopark callbacks, since no race context +// is available during their execution. +func RaceUncheckedAtomicCompareAndSwapUintptr(ptr *uintptr, old, new uintptr) bool diff --git a/pkg/sync/runtime_unsafe.go b/pkg/sync/runtime_unsafe.go new file mode 100644 index 000000000..7ad6a4434 --- /dev/null +++ b/pkg/sync/runtime_unsafe.go @@ -0,0 +1,76 @@ +// Copyright 2020 The gVisor Authors. +// +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build go1.13 +// +build !go1.17 + +// Check function signatures and constants when updating Go version. + +package sync + +import ( + "unsafe" +) + +// Note that go:linkname silently doesn't work if the local name is exported, +// necessitating an indirection for exported functions. + +// Memmove is runtime.memmove, exported for SeqAtomicLoad/SeqAtomicTryLoad<T>. +// +//go:nosplit +func Memmove(to, from unsafe.Pointer, n uintptr) { + memmove(to, from, n) +} + +//go:linkname memmove runtime.memmove +//go:noescape +func memmove(to, from unsafe.Pointer, n uintptr) + +// Gopark is runtime.gopark. Gopark calls unlockf(pointer to runtime.g, lock); +// if unlockf returns true, Gopark blocks until Goready(pointer to runtime.g) +// is called. unlockf and its callees must be nosplit and norace, since stack +// splitting and race context are not available where it is called. +// +//go:nosplit +func Gopark(unlockf func(uintptr, unsafe.Pointer) bool, lock unsafe.Pointer, reason uint8, traceEv byte, traceskip int) { + gopark(unlockf, lock, reason, traceEv, traceskip) +} + +//go:linkname gopark runtime.gopark +func gopark(unlockf func(uintptr, unsafe.Pointer) bool, lock unsafe.Pointer, reason uint8, traceEv byte, traceskip int) + +// Goready is runtime.goready. +// +//go:nosplit +func Goready(gp uintptr, traceskip int) { + goready(gp, traceskip) +} + +//go:linkname goready runtime.goready +func goready(gp uintptr, traceskip int) + +// Values for the reason argument to gopark, from Go's src/runtime/runtime2.go. +const ( + WaitReasonSelect uint8 = 9 +) + +// Values for the traceEv argument to gopark, from Go's src/runtime/trace.go. +const ( + TraceEvGoBlockSelect byte = 24 +) + +// These functions are only used within the sync package. + +//go:linkname semacquire sync.runtime_Semacquire +func semacquire(s *uint32) + +//go:linkname semrelease sync.runtime_Semrelease +func semrelease(s *uint32, handoff bool, skipframes int) + +//go:linkname canSpin sync.runtime_canSpin +func canSpin(i int) bool + +//go:linkname doSpin sync.runtime_doSpin +func doSpin() diff --git a/pkg/sync/rwmutex_unsafe.go b/pkg/sync/rwmutex_unsafe.go index b3b4dee78..4cf3fcd6e 100644 --- a/pkg/sync/rwmutex_unsafe.go +++ b/pkg/sync/rwmutex_unsafe.go @@ -3,11 +3,6 @@ // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. -// +build go1.13 -// +build !go1.17 - -// Check go:linkname function signatures when updating Go version. - // This is mostly copied from the standard library's sync/rwmutex.go. // // Happens-before relationships indicated to the race detector: @@ -23,16 +18,15 @@ import ( "unsafe" ) -//go:linkname runtimeSemacquire sync.runtime_Semacquire -func runtimeSemacquire(s *uint32) - -//go:linkname runtimeSemrelease sync.runtime_Semrelease -func runtimeSemrelease(s *uint32, handoff bool, skipframes int) - -// RWMutex is identical to sync.RWMutex, but adds the DowngradeLock, -// TryLock and TryRLock methods. -type RWMutex struct { - w Mutex // held if there are pending writers +// CrossGoroutineRWMutex is equivalent to RWMutex, but it need not be unlocked +// by a the same goroutine that locked the mutex. +type CrossGoroutineRWMutex struct { + // w is held if there are pending writers + // + // We use CrossGoroutineMutex rather than Mutex because the lock + // annotation instrumentation in Mutex will trigger false positives in + // the race detector when called inside of RaceDisable. + w CrossGoroutineMutex writerSem uint32 // semaphore for writers to wait for completing readers readerSem uint32 // semaphore for readers to wait for completing writers readerCount int32 // number of pending readers @@ -43,7 +37,7 @@ const rwmutexMaxReaders = 1 << 30 // TryRLock locks rw for reading. It returns true if it succeeds and false // otherwise. It does not block. -func (rw *RWMutex) TryRLock() bool { +func (rw *CrossGoroutineRWMutex) TryRLock() bool { if RaceEnabled { RaceDisable() } @@ -67,13 +61,17 @@ func (rw *RWMutex) TryRLock() bool { } // RLock locks rw for reading. -func (rw *RWMutex) RLock() { +// +// It should not be used for recursive read locking; a blocked Lock call +// excludes new readers from acquiring the lock. See the documentation on the +// RWMutex type. +func (rw *CrossGoroutineRWMutex) RLock() { if RaceEnabled { RaceDisable() } if atomic.AddInt32(&rw.readerCount, 1) < 0 { // A writer is pending, wait for it. - runtimeSemacquire(&rw.readerSem) + semacquire(&rw.readerSem) } if RaceEnabled { RaceEnable() @@ -82,7 +80,10 @@ func (rw *RWMutex) RLock() { } // RUnlock undoes a single RLock call. -func (rw *RWMutex) RUnlock() { +// +// Preconditions: +// * rw is locked for reading. +func (rw *CrossGoroutineRWMutex) RUnlock() { if RaceEnabled { RaceReleaseMerge(unsafe.Pointer(&rw.writerSem)) RaceDisable() @@ -94,7 +95,7 @@ func (rw *RWMutex) RUnlock() { // A writer is pending. if atomic.AddInt32(&rw.readerWait, -1) == 0 { // The last reader unblocks the writer. - runtimeSemrelease(&rw.writerSem, false, 0) + semrelease(&rw.writerSem, false, 0) } } if RaceEnabled { @@ -104,7 +105,7 @@ func (rw *RWMutex) RUnlock() { // TryLock locks rw for writing. It returns true if it succeeds and false // otherwise. It does not block. -func (rw *RWMutex) TryLock() bool { +func (rw *CrossGoroutineRWMutex) TryLock() bool { if RaceEnabled { RaceDisable() } @@ -130,8 +131,9 @@ func (rw *RWMutex) TryLock() bool { return true } -// Lock locks rw for writing. -func (rw *RWMutex) Lock() { +// Lock locks rw for writing. If the lock is already locked for reading or +// writing, Lock blocks until the lock is available. +func (rw *CrossGoroutineRWMutex) Lock() { if RaceEnabled { RaceDisable() } @@ -141,7 +143,7 @@ func (rw *RWMutex) Lock() { r := atomic.AddInt32(&rw.readerCount, -rwmutexMaxReaders) + rwmutexMaxReaders // Wait for active readers. if r != 0 && atomic.AddInt32(&rw.readerWait, r) != 0 { - runtimeSemacquire(&rw.writerSem) + semacquire(&rw.writerSem) } if RaceEnabled { RaceEnable() @@ -150,7 +152,10 @@ func (rw *RWMutex) Lock() { } // Unlock unlocks rw for writing. -func (rw *RWMutex) Unlock() { +// +// Preconditions: +// * rw is locked for writing. +func (rw *CrossGoroutineRWMutex) Unlock() { if RaceEnabled { RaceRelease(unsafe.Pointer(&rw.writerSem)) RaceRelease(unsafe.Pointer(&rw.readerSem)) @@ -163,7 +168,7 @@ func (rw *RWMutex) Unlock() { } // Unblock blocked readers, if any. for i := 0; i < int(r); i++ { - runtimeSemrelease(&rw.readerSem, false, 0) + semrelease(&rw.readerSem, false, 0) } // Allow other writers to proceed. rw.w.Unlock() @@ -173,7 +178,10 @@ func (rw *RWMutex) Unlock() { } // DowngradeLock atomically unlocks rw for writing and locks it for reading. -func (rw *RWMutex) DowngradeLock() { +// +// Preconditions: +// * rw is locked for writing. +func (rw *CrossGoroutineRWMutex) DowngradeLock() { if RaceEnabled { RaceRelease(unsafe.Pointer(&rw.readerSem)) RaceDisable() @@ -186,7 +194,7 @@ func (rw *RWMutex) DowngradeLock() { // Unblock blocked readers, if any. Note that this loop starts as 1 since r // includes this goroutine. for i := 1; i < int(r); i++ { - runtimeSemrelease(&rw.readerSem, false, 0) + semrelease(&rw.readerSem, false, 0) } // Allow other writers to proceed to rw.w.Lock(). Note that they will still // block on rw.writerSem since at least this reader exists, such that @@ -196,3 +204,91 @@ func (rw *RWMutex) DowngradeLock() { RaceEnable() } } + +// A RWMutex is a reader/writer mutual exclusion lock. The lock can be held by +// an arbitrary number of readers or a single writer. The zero value for a +// RWMutex is an unlocked mutex. +// +// A RWMutex must not be copied after first use. +// +// If a goroutine holds a RWMutex for reading and another goroutine might call +// Lock, no goroutine should expect to be able to acquire a read lock until the +// initial read lock is released. In particular, this prohibits recursive read +// locking. This is to ensure that the lock eventually becomes available; a +// blocked Lock call excludes new readers from acquiring the lock. +// +// A Mutex must be unlocked by the same goroutine that locked it. This +// invariant is enforced with the 'checklocks' build tag. +type RWMutex struct { + m CrossGoroutineRWMutex +} + +// TryRLock locks rw for reading. It returns true if it succeeds and false +// otherwise. It does not block. +func (rw *RWMutex) TryRLock() bool { + // Note lock first to enforce proper locking even if unsuccessful. + noteLock(unsafe.Pointer(rw)) + locked := rw.m.TryRLock() + if !locked { + noteUnlock(unsafe.Pointer(rw)) + } + return locked +} + +// RLock locks rw for reading. +// +// It should not be used for recursive read locking; a blocked Lock call +// excludes new readers from acquiring the lock. See the documentation on the +// RWMutex type. +func (rw *RWMutex) RLock() { + noteLock(unsafe.Pointer(rw)) + rw.m.RLock() +} + +// RUnlock undoes a single RLock call. +// +// Preconditions: +// * rw is locked for reading. +// * rw was locked by this goroutine. +func (rw *RWMutex) RUnlock() { + rw.m.RUnlock() + noteUnlock(unsafe.Pointer(rw)) +} + +// TryLock locks rw for writing. It returns true if it succeeds and false +// otherwise. It does not block. +func (rw *RWMutex) TryLock() bool { + // Note lock first to enforce proper locking even if unsuccessful. + noteLock(unsafe.Pointer(rw)) + locked := rw.m.TryLock() + if !locked { + noteUnlock(unsafe.Pointer(rw)) + } + return locked +} + +// Lock locks rw for writing. If the lock is already locked for reading or +// writing, Lock blocks until the lock is available. +func (rw *RWMutex) Lock() { + noteLock(unsafe.Pointer(rw)) + rw.m.Lock() +} + +// Unlock unlocks rw for writing. +// +// Preconditions: +// * rw is locked for writing. +// * rw was locked by this goroutine. +func (rw *RWMutex) Unlock() { + rw.m.Unlock() + noteUnlock(unsafe.Pointer(rw)) +} + +// DowngradeLock atomically unlocks rw for writing and locks it for reading. +// +// Preconditions: +// * rw is locked for writing. +func (rw *RWMutex) DowngradeLock() { + // No note change for DowngradeLock. + rw.m.DowngradeLock() +} diff --git a/pkg/sync/seqatomic_unsafe.go b/pkg/sync/seqatomic_unsafe.go index 2184cb5ab..780f3b8f8 100644 --- a/pkg/sync/seqatomic_unsafe.go +++ b/pkg/sync/seqatomic_unsafe.go @@ -8,20 +8,12 @@ package template import ( - "fmt" - "reflect" - "strings" "unsafe" "gvisor.dev/gvisor/pkg/sync" ) // Value is a required type parameter. -// -// Value must not contain any pointers, including interface objects, function -// objects, slices, maps, channels, unsafe.Pointer, and arrays or structs -// containing any of the above. An init() function will panic if this property -// does not hold. type Value struct{} // SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race @@ -55,12 +47,3 @@ func SeqAtomicTryLoad(seq *sync.SeqCount, epoch sync.SeqCountEpoch, ptr *Value) ok = seq.ReadOk(epoch) return } - -func init() { - var val Value - typ := reflect.TypeOf(val) - name := typ.Name() - if ptrs := sync.PointersInType(typ, name); len(ptrs) != 0 { - panic(fmt.Sprintf("SeqAtomicLoad<%s> is invalid since values %s of type %s contain pointers:\n%s", typ, name, typ, strings.Join(ptrs, "\n"))) - } -} diff --git a/pkg/sync/seqcount.go b/pkg/sync/seqcount.go index 2c5d3df99..1f025f33c 100644 --- a/pkg/sync/seqcount.go +++ b/pkg/sync/seqcount.go @@ -6,8 +6,6 @@ package sync import ( - "fmt" - "reflect" "sync/atomic" ) @@ -27,9 +25,6 @@ import ( // - SeqCount may be more flexible: correct use of SeqCount.ReadOk allows other // operations to be made atomic with reads of SeqCount-protected data. // -// - SeqCount may be less flexible: as of this writing, SeqCount-protected data -// cannot include pointers. -// // - SeqCount is more cumbersome to use; atomic reads of SeqCount-protected // data require instantiating function templates using go_generics (see // seqatomic.go). @@ -128,32 +123,3 @@ func (s *SeqCount) EndWrite() { panic("SeqCount.EndWrite outside writer critical section") } } - -// PointersInType returns a list of pointers reachable from values named -// valName of the given type. -// -// PointersInType is not exhaustive, but it is guaranteed that if typ contains -// at least one pointer, then PointersInTypeOf returns a non-empty list. -func PointersInType(typ reflect.Type, valName string) []string { - switch kind := typ.Kind(); kind { - case reflect.Bool, reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64, reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr, reflect.Float32, reflect.Float64, reflect.Complex64, reflect.Complex128: - return nil - - case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Ptr, reflect.Slice, reflect.String, reflect.UnsafePointer: - return []string{valName} - - case reflect.Array: - return PointersInType(typ.Elem(), valName+"[]") - - case reflect.Struct: - var ptrs []string - for i, n := 0, typ.NumField(); i < n; i++ { - field := typ.Field(i) - ptrs = append(ptrs, PointersInType(field.Type, fmt.Sprintf("%s.%s", valName, field.Name))...) - } - return ptrs - - default: - return []string{fmt.Sprintf("%s (of type %s with unknown kind %s)", valName, typ, kind)} - } -} diff --git a/pkg/sync/seqcount_test.go b/pkg/sync/seqcount_test.go index 6eb7b4b59..3f5592e3e 100644 --- a/pkg/sync/seqcount_test.go +++ b/pkg/sync/seqcount_test.go @@ -6,7 +6,6 @@ package sync import ( - "reflect" "testing" "time" ) @@ -99,55 +98,3 @@ func BenchmarkSeqCountReadUncontended(b *testing.B) { } }) } - -func TestPointersInType(t *testing.T) { - for _, test := range []struct { - name string // used for both test and value name - val interface{} - ptrs []string - }{ - { - name: "EmptyStruct", - val: struct{}{}, - }, - { - name: "Int", - val: int(0), - }, - { - name: "MixedStruct", - val: struct { - b bool - I int - ExportedPtr *struct{} - unexportedPtr *struct{} - arr [2]int - ptrArr [2]*int - nestedStruct struct { - nestedNonptr int - nestedPtr *int - } - structArr [1]struct { - nonptr int - ptr *int - } - }{}, - ptrs: []string{ - "MixedStruct.ExportedPtr", - "MixedStruct.unexportedPtr", - "MixedStruct.ptrArr[]", - "MixedStruct.nestedStruct.nestedPtr", - "MixedStruct.structArr[].ptr", - }, - }, - } { - t.Run(test.name, func(t *testing.T) { - typ := reflect.TypeOf(test.val) - ptrs := PointersInType(typ, test.name) - t.Logf("Found pointers: %v", ptrs) - if (len(ptrs) != 0 || len(test.ptrs) != 0) && !reflect.DeepEqual(ptrs, test.ptrs) { - t.Errorf("Got %v, wanted %v", ptrs, test.ptrs) - } - }) - } -} diff --git a/pkg/syncevent/BUILD b/pkg/syncevent/BUILD index 0500a22cf..42c553308 100644 --- a/pkg/syncevent/BUILD +++ b/pkg/syncevent/BUILD @@ -9,10 +9,6 @@ go_library( "receiver.go", "source.go", "syncevent.go", - "waiter_amd64.s", - "waiter_arm64.s", - "waiter_asm_unsafe.go", - "waiter_noasm_unsafe.go", "waiter_unsafe.go", ], visibility = ["//:sandbox"], diff --git a/pkg/syncevent/waiter_noasm_unsafe.go b/pkg/syncevent/waiter_noasm_unsafe.go deleted file mode 100644 index 0f74a689c..000000000 --- a/pkg/syncevent/waiter_noasm_unsafe.go +++ /dev/null @@ -1,39 +0,0 @@ -// Copyright 2020 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// waiterUnlock is called from g0, so when the race detector is enabled, -// waiterUnlock must be implemented in assembly since no race context is -// available. -// -// +build !race -// +build !amd64,!arm64 - -package syncevent - -import ( - "sync/atomic" - "unsafe" -) - -// waiterUnlock is the "unlock function" passed to runtime.gopark by -// Waiter.Wait*. wg is &Waiter.g, and g is a pointer to the calling runtime.g. -// waiterUnlock returns true if Waiter.Wait should sleep and false if sleeping -// should be aborted. -// -//go:nosplit -func waiterUnlock(ptr unsafe.Pointer, wg *unsafe.Pointer) bool { - // The only way this CAS can fail is if a call to Waiter.NotifyPending() - // has replaced *wg with nil, in which case we should not sleep. - return atomic.CompareAndSwapPointer(wg, (unsafe.Pointer)(&preparingG), ptr) -} diff --git a/pkg/syncevent/waiter_unsafe.go b/pkg/syncevent/waiter_unsafe.go index 518f18479..b6ed2852d 100644 --- a/pkg/syncevent/waiter_unsafe.go +++ b/pkg/syncevent/waiter_unsafe.go @@ -12,11 +12,6 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build go1.11 -// +build !go1.17 - -// Check go:linkname function signatures when updating Go version. - package syncevent import ( @@ -26,17 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sync" ) -//go:linkname gopark runtime.gopark -func gopark(unlockf func(unsafe.Pointer, *unsafe.Pointer) bool, wg *unsafe.Pointer, reason uint8, traceEv byte, traceskip int) - -//go:linkname goready runtime.goready -func goready(g unsafe.Pointer, traceskip int) - -const ( - waitReasonSelect = 9 // Go: src/runtime/runtime2.go - traceEvGoBlockSelect = 24 // Go: src/runtime/trace.go -) - // Waiter allows a goroutine to block on pending events received by a Receiver. // // Waiter.Init() must be called before first use. @@ -45,20 +29,19 @@ type Waiter struct { // g is one of: // - // - nil: No goroutine is blocking in Wait. + // - 0: No goroutine is blocking in Wait. // - // - &preparingG: A goroutine is in Wait preparing to sleep, but hasn't yet + // - preparingG: A goroutine is in Wait preparing to sleep, but hasn't yet // completed waiterUnlock(). Thus the wait can only be interrupted by - // replacing the value of g with nil (the G may not be in state Gwaiting - // yet, so we can't call goready.) + // replacing the value of g with 0 (the G may not be in state Gwaiting yet, + // so we can't call goready.) // // - Otherwise: g is a pointer to the runtime.g in state Gwaiting for the // goroutine blocked in Wait, which can only be woken by calling goready. - g unsafe.Pointer `state:"zerovalue"` + g uintptr `state:"zerovalue"` } -// Sentinel object for Waiter.g. -var preparingG struct{} +const preparingG = 1 // Init must be called before first use of w. func (w *Waiter) Init() { @@ -99,21 +82,29 @@ func (w *Waiter) WaitFor(es Set) Set { } // Indicate that we're preparing to go to sleep. - atomic.StorePointer(&w.g, (unsafe.Pointer)(&preparingG)) + atomic.StoreUintptr(&w.g, preparingG) // If an event is pending, abort the sleep. if p := w.r.Pending(); p&es != NoEvents { - atomic.StorePointer(&w.g, nil) + atomic.StoreUintptr(&w.g, 0) return p } // If w.g is still preparingG (i.e. w.NotifyPending() has not been - // called or has not reached atomic.SwapPointer()), go to sleep until + // called or has not reached atomic.SwapUintptr()), go to sleep until // w.NotifyPending() => goready(). - gopark(waiterUnlock, &w.g, waitReasonSelect, traceEvGoBlockSelect, 0) + sync.Gopark(waiterCommit, unsafe.Pointer(&w.g), sync.WaitReasonSelect, sync.TraceEvGoBlockSelect, 0) } } +//go:norace +//go:nosplit +func waiterCommit(g uintptr, wg unsafe.Pointer) bool { + // The only way this CAS can fail is if a call to Waiter.NotifyPending() + // has replaced *wg with nil, in which case we should not sleep. + return sync.RaceUncheckedAtomicCompareAndSwapUintptr((*uintptr)(wg), preparingG, g) +} + // Ack marks the given events as not pending. func (w *Waiter) Ack(es Set) { w.r.Ack(es) @@ -135,20 +126,20 @@ func (w *Waiter) WaitAndAckAll() Set { for { // Indicate that we're preparing to go to sleep. - atomic.StorePointer(&w.g, (unsafe.Pointer)(&preparingG)) + atomic.StoreUintptr(&w.g, preparingG) // If an event is pending, abort the sleep. if w.r.Pending() != NoEvents { if p := w.r.PendingAndAckAll(); p != NoEvents { - atomic.StorePointer(&w.g, nil) + atomic.StoreUintptr(&w.g, 0) return p } } // If w.g is still preparingG (i.e. w.NotifyPending() has not been - // called or has not reached atomic.SwapPointer()), go to sleep until + // called or has not reached atomic.SwapUintptr()), go to sleep until // w.NotifyPending() => goready(). - gopark(waiterUnlock, &w.g, waitReasonSelect, traceEvGoBlockSelect, 0) + sync.Gopark(waiterCommit, unsafe.Pointer(&w.g), sync.WaitReasonSelect, sync.TraceEvGoBlockSelect, 0) // Check for pending events. We call PendingAndAckAll() directly now since // we only expect to be woken after events become pending. @@ -171,14 +162,14 @@ func (w *Waiter) NotifyPending() { // goroutine. NotifyPending is called after w.r.Pending() is updated, so // concurrent and future calls to w.Wait() will observe pending events and // abort sleeping. - if atomic.LoadPointer(&w.g) == nil { + if atomic.LoadUintptr(&w.g) == 0 { return } // Wake a sleeping G, or prevent a G that is preparing to sleep from doing // so. Swap is needed here to ensure that only one call to NotifyPending // calls goready. - if g := atomic.SwapPointer(&w.g, nil); g != nil && g != (unsafe.Pointer)(&preparingG) { - goready(g, 0) + if g := atomic.SwapUintptr(&w.g, 0); g > preparingG { + sync.Goready(g, 0) } } diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 81f762e10..d3ae56ac6 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -20,6 +20,7 @@ import ( "encoding/binary" "reflect" "testing" + "time" "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" @@ -216,6 +217,42 @@ func IPv4Options(want header.IPv4Options) NetworkChecker { } } +// IPv4RouterAlert returns a checker that checks that the RouterAlert option is +// set in an IPv4 packet. +func IPv4RouterAlert() NetworkChecker { + return func(t *testing.T, h []header.Network) { + t.Helper() + ip, ok := h[0].(header.IPv4) + if !ok { + t.Fatalf("unexpected network header passed to checker, got = %T, want = header.IPv4", h[0]) + } + iterator := ip.Options().MakeIterator() + for { + opt, done, err := iterator.Next() + if err != nil { + t.Fatalf("error acquiring next IPv4 option %s", err) + } + if done { + break + } + if opt.Type() != header.IPv4OptionRouterAlertType { + continue + } + want := [header.IPv4OptionRouterAlertLength]byte{ + byte(header.IPv4OptionRouterAlertType), + header.IPv4OptionRouterAlertLength, + header.IPv4OptionRouterAlertValue, + header.IPv4OptionRouterAlertValue, + } + if diff := cmp.Diff(want[:], opt.Contents()); diff != "" { + t.Errorf("router alert option mismatch (-want +got):\n%s", diff) + } + return + } + t.Errorf("failed to find router alert option in %v", ip.Options()) + } +} + // FragmentOffset creates a checker that checks the FragmentOffset field. func FragmentOffset(offset uint16) NetworkChecker { return func(t *testing.T, h []header.Network) { @@ -1012,6 +1049,74 @@ func ICMPv6Payload(want []byte) TransportChecker { } } +// MLD creates a checker that checks that the packet contains a valid MLD +// message for type of mldType, with potentially additional checks specified by +// checkers. +// +// Checkers may assume that a valid ICMPv6 is passed to it containing a valid +// MLD message as far as the size of the message (minSize) is concerned. The +// values within the message are up to checkers to validate. +func MLD(msgType header.ICMPv6Type, minSize int, checkers ...TransportChecker) NetworkChecker { + return func(t *testing.T, h []header.Network) { + t.Helper() + + // Check normal ICMPv6 first. + ICMPv6( + ICMPv6Type(msgType), + ICMPv6Code(0))(t, h) + + last := h[len(h)-1] + + icmp := header.ICMPv6(last.Payload()) + if got := len(icmp.MessageBody()); got < minSize { + t.Fatalf("ICMPv6 MLD (type = %d) payload size of %d is less than the minimum size of %d", msgType, got, minSize) + } + + for _, f := range checkers { + f(t, icmp) + } + if t.Failed() { + t.FailNow() + } + } +} + +// MLDMaxRespDelay creates a checker that checks the Maximum Response Delay +// field of a MLD message. +// +// The returned TransportChecker assumes that a valid ICMPv6 is passed to it +// containing a valid MLD message as far as the size is concerned. +func MLDMaxRespDelay(want time.Duration) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmp := h.(header.ICMPv6) + ns := header.MLD(icmp.MessageBody()) + + if got := ns.MaximumResponseDelay(); got != want { + t.Errorf("got %T.MaximumResponseDelay() = %s, want = %s", ns, got, want) + } + } +} + +// MLDMulticastAddress creates a checker that checks the Multicast Address +// field of a MLD message. +// +// The returned TransportChecker assumes that a valid ICMPv6 is passed to it +// containing a valid MLD message as far as the size is concerned. +func MLDMulticastAddress(want tcpip.Address) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmp := h.(header.ICMPv6) + ns := header.MLD(icmp.MessageBody()) + + if got := ns.MulticastAddress(); got != want { + t.Errorf("got %T.MulticastAddress() = %s, want = %s", ns, got, want) + } + } +} + // NDP creates a checker that checks that the packet contains a valid NDP // message for type of ty, with potentially additional checks specified by // checkers. @@ -1031,7 +1136,7 @@ func NDP(msgType header.ICMPv6Type, minSize int, checkers ...TransportChecker) N last := h[len(h)-1] icmp := header.ICMPv6(last.Payload()) - if got := len(icmp.NDPPayload()); got < minSize { + if got := len(icmp.MessageBody()); got < minSize { t.Fatalf("ICMPv6 NDP (type = %d) payload size of %d is less than the minimum size of %d", msgType, got, minSize) } @@ -1065,7 +1170,7 @@ func NDPNSTargetAddress(want tcpip.Address) TransportChecker { t.Helper() icmp := h.(header.ICMPv6) - ns := header.NDPNeighborSolicit(icmp.NDPPayload()) + ns := header.NDPNeighborSolicit(icmp.MessageBody()) if got := ns.TargetAddress(); got != want { t.Errorf("got %T.TargetAddress() = %s, want = %s", ns, got, want) @@ -1094,7 +1199,7 @@ func NDPNATargetAddress(want tcpip.Address) TransportChecker { t.Helper() icmp := h.(header.ICMPv6) - na := header.NDPNeighborAdvert(icmp.NDPPayload()) + na := header.NDPNeighborAdvert(icmp.MessageBody()) if got := na.TargetAddress(); got != want { t.Errorf("got %T.TargetAddress() = %s, want = %s", na, got, want) @@ -1112,7 +1217,7 @@ func NDPNASolicitedFlag(want bool) TransportChecker { t.Helper() icmp := h.(header.ICMPv6) - na := header.NDPNeighborAdvert(icmp.NDPPayload()) + na := header.NDPNeighborAdvert(icmp.MessageBody()) if got := na.SolicitedFlag(); got != want { t.Errorf("got %T.SolicitedFlag = %t, want = %t", na, got, want) @@ -1183,7 +1288,7 @@ func NDPNAOptions(opts []header.NDPOption) TransportChecker { t.Helper() icmp := h.(header.ICMPv6) - na := header.NDPNeighborAdvert(icmp.NDPPayload()) + na := header.NDPNeighborAdvert(icmp.MessageBody()) ndpOptions(t, na.Options(), opts) } } @@ -1198,7 +1303,7 @@ func NDPNSOptions(opts []header.NDPOption) TransportChecker { t.Helper() icmp := h.(header.ICMPv6) - ns := header.NDPNeighborSolicit(icmp.NDPPayload()) + ns := header.NDPNeighborSolicit(icmp.MessageBody()) ndpOptions(t, ns.Options(), opts) } } @@ -1223,7 +1328,75 @@ func NDPRSOptions(opts []header.NDPOption) TransportChecker { t.Helper() icmp := h.(header.ICMPv6) - rs := header.NDPRouterSolicit(icmp.NDPPayload()) + rs := header.NDPRouterSolicit(icmp.MessageBody()) ndpOptions(t, rs.Options(), opts) } } + +// IGMP checks the validity and properties of the given IGMP packet. It is +// expected to be used in conjunction with other IGMP transport checkers for +// specific properties. +func IGMP(checkers ...TransportChecker) NetworkChecker { + return func(t *testing.T, h []header.Network) { + t.Helper() + + last := h[len(h)-1] + + if p := last.TransportProtocol(); p != header.IGMPProtocolNumber { + t.Fatalf("Bad protocol, got %d, want %d", p, header.IGMPProtocolNumber) + } + + igmp := header.IGMP(last.Payload()) + for _, f := range checkers { + f(t, igmp) + } + if t.Failed() { + t.FailNow() + } + } +} + +// IGMPType creates a checker that checks the IGMP Type field. +func IGMPType(want header.IGMPType) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + igmp, ok := h.(header.IGMP) + if !ok { + t.Fatalf("got transport header = %T, want = header.IGMP", h) + } + if got := igmp.Type(); got != want { + t.Errorf("got igmp.Type() = %d, want = %d", got, want) + } + } +} + +// IGMPMaxRespTime creates a checker that checks the IGMP Max Resp Time field. +func IGMPMaxRespTime(want time.Duration) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + igmp, ok := h.(header.IGMP) + if !ok { + t.Fatalf("got transport header = %T, want = header.IGMP", h) + } + if got := igmp.MaxRespTime(); got != want { + t.Errorf("got igmp.MaxRespTime() = %s, want = %s", got, want) + } + } +} + +// IGMPGroupAddress creates a checker that checks the IGMP Group Address field. +func IGMPGroupAddress(want tcpip.Address) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + igmp, ok := h.(header.IGMP) + if !ok { + t.Fatalf("got transport header = %T, want = header.IGMP", h) + } + if got := igmp.GroupAddress(); got != want { + t.Errorf("got igmp.GroupAddress() = %s, want = %s", got, want) + } + } +} diff --git a/pkg/tcpip/header/BUILD b/pkg/tcpip/header/BUILD index d87797617..0bdc12d53 100644 --- a/pkg/tcpip/header/BUILD +++ b/pkg/tcpip/header/BUILD @@ -11,11 +11,13 @@ go_library( "gue.go", "icmpv4.go", "icmpv6.go", + "igmp.go", "interfaces.go", "ipv4.go", "ipv6.go", "ipv6_extension_headers.go", "ipv6_fragment.go", + "mld.go", "ndp_neighbor_advert.go", "ndp_neighbor_solicit.go", "ndp_options.go", @@ -39,6 +41,8 @@ go_test( size = "small", srcs = [ "checksum_test.go", + "igmp_test.go", + "ipv4_test.go", "ipv6_test.go", "ipversion_test.go", "tcp_test.go", @@ -58,6 +62,7 @@ go_test( srcs = [ "eth_test.go", "ipv6_extension_headers_test.go", + "mld_test.go", "ndp_test.go", ], library = ":header", diff --git a/pkg/tcpip/header/icmpv6.go b/pkg/tcpip/header/icmpv6.go index 4303fc5d5..2eef64b4d 100644 --- a/pkg/tcpip/header/icmpv6.go +++ b/pkg/tcpip/header/icmpv6.go @@ -115,6 +115,12 @@ const ( ICMPv6NeighborSolicit ICMPv6Type = 135 ICMPv6NeighborAdvert ICMPv6Type = 136 ICMPv6RedirectMsg ICMPv6Type = 137 + + // Multicast Listener Discovery (MLD) messages, see RFC 2710. + + ICMPv6MulticastListenerQuery ICMPv6Type = 130 + ICMPv6MulticastListenerReport ICMPv6Type = 131 + ICMPv6MulticastListenerDone ICMPv6Type = 132 ) // IsErrorType returns true if the receiver is an ICMP error type. @@ -245,10 +251,9 @@ func (b ICMPv6) SetSequence(sequence uint16) { binary.BigEndian.PutUint16(b[icmpv6SequenceOffset:], sequence) } -// NDPPayload returns the NDP payload buffer. That is, it returns the ICMPv6 -// packet's message body as defined by RFC 4443 section 2.1; the portion of the -// ICMPv6 buffer after the first ICMPv6HeaderSize bytes. -func (b ICMPv6) NDPPayload() []byte { +// MessageBody returns the message body as defined by RFC 4443 section 2.1; the +// portion of the ICMPv6 buffer after the first ICMPv6HeaderSize bytes. +func (b ICMPv6) MessageBody() []byte { return b[ICMPv6HeaderSize:] } diff --git a/pkg/tcpip/header/igmp.go b/pkg/tcpip/header/igmp.go new file mode 100644 index 000000000..5c5be1b9d --- /dev/null +++ b/pkg/tcpip/header/igmp.go @@ -0,0 +1,181 @@ +// 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 header + +import ( + "encoding/binary" + "fmt" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +// IGMP represents an IGMP header stored in a byte array. +type IGMP []byte + +// IGMP implements `Transport`. +var _ Transport = (*IGMP)(nil) + +const ( + // IGMPMinimumSize is the minimum size of a valid IGMP packet in bytes, + // as per RFC 2236, Section 2, Page 2. + IGMPMinimumSize = 8 + + // IGMPQueryMinimumSize is the minimum size of a valid Membership Query + // Message in bytes, as per RFC 2236, Section 2, Page 2. + IGMPQueryMinimumSize = 8 + + // IGMPReportMinimumSize is the minimum size of a valid Report Message in + // bytes, as per RFC 2236, Section 2, Page 2. + IGMPReportMinimumSize = 8 + + // IGMPLeaveMessageMinimumSize is the minimum size of a valid Leave Message + // in bytes, as per RFC 2236, Section 2, Page 2. + IGMPLeaveMessageMinimumSize = 8 + + // IGMPTTL is the TTL for all IGMP messages, as per RFC 2236, Section 3, Page + // 3. + IGMPTTL = 1 + + // igmpTypeOffset defines the offset of the type field in an IGMP message. + igmpTypeOffset = 0 + + // igmpMaxRespTimeOffset defines the offset of the MaxRespTime field in an + // IGMP message. + igmpMaxRespTimeOffset = 1 + + // igmpChecksumOffset defines the offset of the checksum field in an IGMP + // message. + igmpChecksumOffset = 2 + + // igmpGroupAddressOffset defines the offset of the Group Address field in an + // IGMP message. + igmpGroupAddressOffset = 4 + + // IGMPProtocolNumber is IGMP's transport protocol number. + IGMPProtocolNumber tcpip.TransportProtocolNumber = 2 +) + +// IGMPType is the IGMP type field as per RFC 2236. +type IGMPType byte + +// Values for the IGMP Type described in RFC 2236 Section 2.1, Page 2. +// Descriptions below come from there. +const ( + // IGMPMembershipQuery indicates that the message type is Membership Query. + // "There are two sub-types of Membership Query messages: + // - General Query, used to learn which groups have members on an + // attached network. + // - Group-Specific Query, used to learn if a particular group + // has any members on an attached network. + // These two messages are differentiated by the Group Address, as + // described in section 1.4 ." + IGMPMembershipQuery IGMPType = 0x11 + // IGMPv1MembershipReport indicates that the message is a Membership Report + // generated by a host using the IGMPv1 protocol: "an additional type of + // message, for backwards-compatibility with IGMPv1" + IGMPv1MembershipReport IGMPType = 0x12 + // IGMPv2MembershipReport indicates that the Message type is a Membership + // Report generated by a host using the IGMPv2 protocol. + IGMPv2MembershipReport IGMPType = 0x16 + // IGMPLeaveGroup indicates that the message type is a Leave Group + // notification message. + IGMPLeaveGroup IGMPType = 0x17 +) + +// Type is the IGMP type field. +func (b IGMP) Type() IGMPType { return IGMPType(b[igmpTypeOffset]) } + +// SetType sets the IGMP type field. +func (b IGMP) SetType(t IGMPType) { b[igmpTypeOffset] = byte(t) } + +// MaxRespTime gets the MaxRespTimeField. This is meaningful only in Membership +// Query messages, in other cases it is set to 0 by the sender and ignored by +// the receiver. +func (b IGMP) MaxRespTime() time.Duration { + // As per RFC 2236 section 2.2, + // + // The Max Response Time field is meaningful only in Membership Query + // messages, and specifies the maximum allowed time before sending a + // responding report in units of 1/10 second. In all other messages, it + // is set to zero by the sender and ignored by receivers. + return DecisecondToDuration(b[igmpMaxRespTimeOffset]) +} + +// SetMaxRespTime sets the MaxRespTimeField. +func (b IGMP) SetMaxRespTime(m byte) { b[igmpMaxRespTimeOffset] = m } + +// Checksum is the IGMP checksum field. +func (b IGMP) Checksum() uint16 { + return binary.BigEndian.Uint16(b[igmpChecksumOffset:]) +} + +// SetChecksum sets the IGMP checksum field. +func (b IGMP) SetChecksum(checksum uint16) { + binary.BigEndian.PutUint16(b[igmpChecksumOffset:], checksum) +} + +// GroupAddress gets the Group Address field. +func (b IGMP) GroupAddress() tcpip.Address { + return tcpip.Address(b[igmpGroupAddressOffset:][:IPv4AddressSize]) +} + +// SetGroupAddress sets the Group Address field. +func (b IGMP) SetGroupAddress(address tcpip.Address) { + if n := copy(b[igmpGroupAddressOffset:], address); n != IPv4AddressSize { + panic(fmt.Sprintf("copied %d bytes, expected %d", n, IPv4AddressSize)) + } +} + +// SourcePort implements Transport.SourcePort. +func (IGMP) SourcePort() uint16 { + return 0 +} + +// DestinationPort implements Transport.DestinationPort. +func (IGMP) DestinationPort() uint16 { + return 0 +} + +// SetSourcePort implements Transport.SetSourcePort. +func (IGMP) SetSourcePort(uint16) { +} + +// SetDestinationPort implements Transport.SetDestinationPort. +func (IGMP) SetDestinationPort(uint16) { +} + +// Payload implements Transport.Payload. +func (IGMP) Payload() []byte { + return nil +} + +// IGMPCalculateChecksum calculates the IGMP checksum over the provided IGMP +// header. +func IGMPCalculateChecksum(h IGMP) uint16 { + // The header contains a checksum itself, set it aside to avoid checksumming + // the checksum and replace it afterwards. + existingXsum := h.Checksum() + h.SetChecksum(0) + xsum := ^Checksum(h, 0) + h.SetChecksum(existingXsum) + return xsum +} + +// DecisecondToDuration converts a value representing deci-seconds to a +// time.Duration. +func DecisecondToDuration(ds uint8) time.Duration { + return time.Duration(ds) * time.Second / 10 +} diff --git a/pkg/tcpip/header/igmp_test.go b/pkg/tcpip/header/igmp_test.go new file mode 100644 index 000000000..b6126d29a --- /dev/null +++ b/pkg/tcpip/header/igmp_test.go @@ -0,0 +1,110 @@ +// 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 header_test + +import ( + "testing" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +// TestIGMPHeader tests the functions within header.igmp +func TestIGMPHeader(t *testing.T) { + const maxRespTimeTenthSec = 0xF0 + b := []byte{ + 0x11, // IGMP Type, Membership Query + maxRespTimeTenthSec, // Maximum Response Time + 0xC0, 0xC0, // Checksum + 0x01, 0x02, 0x03, 0x04, // Group Address + } + + igmpHeader := header.IGMP(b) + + if got, want := igmpHeader.Type(), header.IGMPMembershipQuery; got != want { + t.Errorf("got igmpHeader.Type() = %x, want = %x", got, want) + } + + if got, want := igmpHeader.MaxRespTime(), header.DecisecondToDuration(maxRespTimeTenthSec); got != want { + t.Errorf("got igmpHeader.MaxRespTime() = %s, want = %s", got, want) + } + + if got, want := igmpHeader.Checksum(), uint16(0xC0C0); got != want { + t.Errorf("got igmpHeader.Checksum() = %x, want = %x", got, want) + } + + if got, want := igmpHeader.GroupAddress(), tcpip.Address("\x01\x02\x03\x04"); got != want { + t.Errorf("got igmpHeader.GroupAddress() = %s, want = %s", got, want) + } + + igmpType := header.IGMPv2MembershipReport + igmpHeader.SetType(igmpType) + if got := igmpHeader.Type(); got != igmpType { + t.Errorf("got igmpHeader.Type() = %x, want = %x", got, igmpType) + } + if got := header.IGMPType(b[0]); got != igmpType { + t.Errorf("got IGMPtype in backing buffer = %x, want %x", got, igmpType) + } + + respTime := byte(0x02) + igmpHeader.SetMaxRespTime(respTime) + if got, want := igmpHeader.MaxRespTime(), header.DecisecondToDuration(respTime); got != want { + t.Errorf("got igmpHeader.MaxRespTime() = %s, want = %s", got, want) + } + + checksum := uint16(0x0102) + igmpHeader.SetChecksum(checksum) + if got := igmpHeader.Checksum(); got != checksum { + t.Errorf("got igmpHeader.Checksum() = %x, want = %x", got, checksum) + } + + groupAddress := tcpip.Address("\x04\x03\x02\x01") + igmpHeader.SetGroupAddress(groupAddress) + if got := igmpHeader.GroupAddress(); got != groupAddress { + t.Errorf("got igmpHeader.GroupAddress() = %s, want = %s", got, groupAddress) + } +} + +// TestIGMPChecksum ensures that the checksum calculator produces the expected +// checksum. +func TestIGMPChecksum(t *testing.T) { + b := []byte{ + 0x11, // IGMP Type, Membership Query + 0xF0, // Maximum Response Time + 0xC0, 0xC0, // Checksum + 0x01, 0x02, 0x03, 0x04, // Group Address + } + + igmpHeader := header.IGMP(b) + + // Calculate the initial checksum after setting the checksum temporarily to 0 + // to avoid checksumming the checksum. + initialChecksum := igmpHeader.Checksum() + igmpHeader.SetChecksum(0) + checksum := ^header.Checksum(b, 0) + igmpHeader.SetChecksum(initialChecksum) + + if got := header.IGMPCalculateChecksum(igmpHeader); got != checksum { + t.Errorf("got IGMPCalculateChecksum = %x, want %x", got, checksum) + } +} + +func TestDecisecondToDuration(t *testing.T) { + const valueInDeciseconds = 5 + if got, want := header.DecisecondToDuration(valueInDeciseconds), valueInDeciseconds*time.Second/10; got != want { + t.Fatalf("got header.DecisecondToDuration(%d) = %s, want = %s", valueInDeciseconds, got, want) + } +} diff --git a/pkg/tcpip/header/ipv4.go b/pkg/tcpip/header/ipv4.go index 91fe7b6a5..e6103f4bc 100644 --- a/pkg/tcpip/header/ipv4.go +++ b/pkg/tcpip/header/ipv4.go @@ -100,7 +100,7 @@ type IPv4Fields struct { // // That leaves ten 32 bit (4 byte) fields for options. An attempt to encode // more will fail. - Options IPv4Options + Options IPv4OptionsSerializer } // IPv4 is an IPv4 header. @@ -157,6 +157,9 @@ const ( // IPv4Any is the non-routable IPv4 "any" meta address. IPv4Any tcpip.Address = "\x00\x00\x00\x00" + // IPv4AllRoutersGroup is a multicast address for all routers. + IPv4AllRoutersGroup tcpip.Address = "\xe0\x00\x00\x02" + // IPv4MinimumProcessableDatagramSize is the minimum size of an IP // packet that every IPv4 capable host must be able to // process/reassemble. @@ -282,18 +285,17 @@ func (b IPv4) DestinationAddress() tcpip.Address { return tcpip.Address(b[dstAddr : dstAddr+IPv4AddressSize]) } -// IPv4Options is a buffer that holds all the raw IP options. -type IPv4Options []byte - -// SizeWithPadding implements stack.NetOptions. -// It reports the size to allocate for the Options. RFC 791 page 23 (end of -// section 3.1) says of the padding at the end of the options: +// padIPv4OptionsLength returns the total length for IPv4 options of length l +// after applying padding according to RFC 791: // The internet header padding is used to ensure that the internet // header ends on a 32 bit boundary. -func (o IPv4Options) SizeWithPadding() int { - return (len(o) + IPv4IHLStride - 1) & ^(IPv4IHLStride - 1) +func padIPv4OptionsLength(length uint8) uint8 { + return (length + IPv4IHLStride - 1) & ^uint8(IPv4IHLStride-1) } +// IPv4Options is a buffer that holds all the raw IP options. +type IPv4Options []byte + // Options returns a buffer holding the options. func (b IPv4) Options() IPv4Options { hdrLen := b.HeaderLength() @@ -372,26 +374,16 @@ func (b IPv4) CalculateChecksum() uint16 { func (b IPv4) Encode(i *IPv4Fields) { // The size of the options defines the size of the whole header and thus the // IHL field. Options are rare and this is a heavily used function so it is - // worth a bit of optimisation here to keep the copy out of the fast path. - hdrLen := IPv4MinimumSize + // worth a bit of optimisation here to keep the serializer out of the fast + // path. + hdrLen := uint8(IPv4MinimumSize) if len(i.Options) != 0 { - // SizeWithPadding is always >= len(i.Options). - aLen := i.Options.SizeWithPadding() - hdrLen += aLen - if hdrLen > len(b) { - panic(fmt.Sprintf("encode received %d bytes, wanted >= %d", len(b), hdrLen)) - } - opts := b[options:] - // This avoids bounds checks on the next line(s) which would happen even - // if there's no work to do. - if n := copy(opts, i.Options); n != aLen { - padding := opts[n:][:aLen-n] - for i := range padding { - padding[i] = 0 - } - } + hdrLen += i.Options.Serialize(b[options:]) + } + if hdrLen > IPv4MaximumHeaderSize { + panic(fmt.Sprintf("%d is larger than maximum IPv4 header size of %d", hdrLen, IPv4MaximumHeaderSize)) } - b.SetHeaderLength(uint8(hdrLen)) + b.SetHeaderLength(hdrLen) b[tos] = i.TOS b.SetTotalLength(i.TotalLength) binary.BigEndian.PutUint16(b[id:], i.ID) @@ -471,6 +463,10 @@ const ( // options and may appear multiple times. IPv4OptionNOPType IPv4OptionType = 1 + // IPv4OptionRouterAlertType is the option type for the Router Alert option, + // defined in RFC 2113 Section 2.1. + IPv4OptionRouterAlertType IPv4OptionType = 20 | 0x80 + // IPv4OptionRecordRouteType is used by each router on the path of the packet // to record its path. It is carried over to an Echo Reply. IPv4OptionRecordRouteType IPv4OptionType = 7 @@ -871,3 +867,162 @@ func (rr *IPv4OptionRecordRoute) Size() uint8 { return uint8(len(*rr)) } // Contents implements IPv4Option. func (rr *IPv4OptionRecordRoute) Contents() []byte { return []byte(*rr) } + +// Router Alert option specific related constants. +// +// from RFC 2113 section 2.1: +// +// +--------+--------+--------+--------+ +// |10010100|00000100| 2 octet value | +// +--------+--------+--------+--------+ +// +// Type: +// Copied flag: 1 (all fragments must carry the option) +// Option class: 0 (control) +// Option number: 20 (decimal) +// +// Length: 4 +// +// Value: A two octet code with the following values: +// 0 - Router shall examine packet +// 1-65535 - Reserved +const ( + // IPv4OptionRouterAlertLength is the length of a Router Alert option. + IPv4OptionRouterAlertLength = 4 + + // IPv4OptionRouterAlertValue is the only permissible value of the 16 bit + // payload of the router alert option. + IPv4OptionRouterAlertValue = 0 + + // iPv4OptionRouterAlertValueOffset is the offset for the value of a + // RouterAlert option. + iPv4OptionRouterAlertValueOffset = 2 +) + +// IPv4SerializableOption is an interface to represent serializable IPv4 option +// types. +type IPv4SerializableOption interface { + // optionType returns the type identifier of the option. + optionType() IPv4OptionType +} + +// IPv4SerializableOptionPayload is an interface providing serialization of the +// payload of an IPv4 option. +type IPv4SerializableOptionPayload interface { + // length returns the size of the payload. + length() uint8 + + // serializeInto serializes the payload into the provided byte buffer. + // + // Note, the caller MUST provide a byte buffer with size of at least + // Length. Implementers of this function may assume that the byte buffer + // is of sufficient size. serializeInto MUST panic if the provided byte + // buffer is not of sufficient size. + // + // serializeInto will return the number of bytes that was used to + // serialize the receiver. Implementers must only use the number of + // bytes required to serialize the receiver. Callers MAY provide a + // larger buffer than required to serialize into. + serializeInto(buffer []byte) uint8 +} + +// IPv4OptionsSerializer is a serializer for IPv4 options. +type IPv4OptionsSerializer []IPv4SerializableOption + +// Length returns the total number of bytes required to serialize the options. +func (s IPv4OptionsSerializer) Length() uint8 { + var total uint8 + for _, opt := range s { + total++ + if withPayload, ok := opt.(IPv4SerializableOptionPayload); ok { + // Add 1 to reported length to account for the length byte. + total += 1 + withPayload.length() + } + } + return padIPv4OptionsLength(total) +} + +// Serialize serializes the provided list of IPV4 options into b. +// +// Note, b must be of sufficient size to hold all the options in s. See +// IPv4OptionsSerializer.Length for details on the getting the total size +// of a serialized IPv4OptionsSerializer. +// +// Serialize panics if b is not of sufficient size to hold all the options in s. +func (s IPv4OptionsSerializer) Serialize(b []byte) uint8 { + var total uint8 + for _, opt := range s { + ty := opt.optionType() + if withPayload, ok := opt.(IPv4SerializableOptionPayload); ok { + // Serialize first to reduce bounds checks. + l := 2 + withPayload.serializeInto(b[2:]) + b[0] = byte(ty) + b[1] = l + b = b[l:] + total += l + continue + } + // Options without payload consist only of the type field. + // + // NB: Repeating code from the branch above is intentional to minimize + // bounds checks. + b[0] = byte(ty) + b = b[1:] + total++ + } + + // According to RFC 791: + // + // The internet header padding is used to ensure that the internet + // header ends on a 32 bit boundary. The padding is zero. + padded := padIPv4OptionsLength(total) + b = b[:padded-total] + for i := range b { + b[i] = 0 + } + return padded +} + +var _ IPv4SerializableOptionPayload = (*IPv4SerializableRouterAlertOption)(nil) +var _ IPv4SerializableOption = (*IPv4SerializableRouterAlertOption)(nil) + +// IPv4SerializableRouterAlertOption provides serialization of the Router Alert +// IPv4 option according to RFC 2113. +type IPv4SerializableRouterAlertOption struct{} + +// Type implements IPv4SerializableOption. +func (*IPv4SerializableRouterAlertOption) optionType() IPv4OptionType { + return IPv4OptionRouterAlertType +} + +// Length implements IPv4SerializableOption. +func (*IPv4SerializableRouterAlertOption) length() uint8 { + return IPv4OptionRouterAlertLength - iPv4OptionRouterAlertValueOffset +} + +// SerializeInto implements IPv4SerializableOption. +func (o *IPv4SerializableRouterAlertOption) serializeInto(buffer []byte) uint8 { + binary.BigEndian.PutUint16(buffer, IPv4OptionRouterAlertValue) + return o.length() +} + +var _ IPv4SerializableOption = (*IPv4SerializableNOPOption)(nil) + +// IPv4SerializableNOPOption provides serialization for the IPv4 no-op option. +type IPv4SerializableNOPOption struct{} + +// Type implements IPv4SerializableOption. +func (*IPv4SerializableNOPOption) optionType() IPv4OptionType { + return IPv4OptionNOPType +} + +var _ IPv4SerializableOption = (*IPv4SerializableListEndOption)(nil) + +// IPv4SerializableListEndOption provides serialization for the IPv4 List End +// option. +type IPv4SerializableListEndOption struct{} + +// Type implements IPv4SerializableOption. +func (*IPv4SerializableListEndOption) optionType() IPv4OptionType { + return IPv4OptionListEndType +} diff --git a/pkg/tcpip/header/ipv4_test.go b/pkg/tcpip/header/ipv4_test.go new file mode 100644 index 000000000..6475cd694 --- /dev/null +++ b/pkg/tcpip/header/ipv4_test.go @@ -0,0 +1,179 @@ +// 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 header_test + +import ( + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +func TestIPv4OptionsSerializer(t *testing.T) { + optCases := []struct { + name string + option []header.IPv4SerializableOption + expect []byte + }{ + { + name: "NOP", + option: []header.IPv4SerializableOption{ + &header.IPv4SerializableNOPOption{}, + }, + expect: []byte{1, 0, 0, 0}, + }, + { + name: "ListEnd", + option: []header.IPv4SerializableOption{ + &header.IPv4SerializableListEndOption{}, + }, + expect: []byte{0, 0, 0, 0}, + }, + { + name: "RouterAlert", + option: []header.IPv4SerializableOption{ + &header.IPv4SerializableRouterAlertOption{}, + }, + expect: []byte{148, 4, 0, 0}, + }, { + name: "NOP and RouterAlert", + option: []header.IPv4SerializableOption{ + &header.IPv4SerializableNOPOption{}, + &header.IPv4SerializableRouterAlertOption{}, + }, + expect: []byte{1, 148, 4, 0, 0, 0, 0, 0}, + }, + } + + for _, opt := range optCases { + t.Run(opt.name, func(t *testing.T) { + s := header.IPv4OptionsSerializer(opt.option) + l := s.Length() + if got := len(opt.expect); got != int(l) { + t.Fatalf("s.Length() = %d, want = %d", got, l) + } + b := make([]byte, l) + for i := range b { + // Fill the buffer with full bytes to ensure padding is being set + // correctly. + b[i] = 0xFF + } + if serializedLength := s.Serialize(b); serializedLength != l { + t.Fatalf("s.Serialize(_) = %d, want %d", serializedLength, l) + } + if diff := cmp.Diff(opt.expect, b); diff != "" { + t.Errorf("mismatched serialized option (-want +got):\n%s", diff) + } + }) + } +} + +// TestIPv4Encode checks that ipv4.Encode correctly fills out the requested +// fields when options are supplied. +func TestIPv4EncodeOptions(t *testing.T) { + tests := []struct { + name string + numberOfNops int + encodedOptions header.IPv4Options // reply should look like this + wantIHL int + }{ + { + name: "valid no options", + wantIHL: header.IPv4MinimumSize, + }, + { + name: "one byte options", + numberOfNops: 1, + encodedOptions: header.IPv4Options{1, 0, 0, 0}, + wantIHL: header.IPv4MinimumSize + 4, + }, + { + name: "two byte options", + numberOfNops: 2, + encodedOptions: header.IPv4Options{1, 1, 0, 0}, + wantIHL: header.IPv4MinimumSize + 4, + }, + { + name: "three byte options", + numberOfNops: 3, + encodedOptions: header.IPv4Options{1, 1, 1, 0}, + wantIHL: header.IPv4MinimumSize + 4, + }, + { + name: "four byte options", + numberOfNops: 4, + encodedOptions: header.IPv4Options{1, 1, 1, 1}, + wantIHL: header.IPv4MinimumSize + 4, + }, + { + name: "five byte options", + numberOfNops: 5, + encodedOptions: header.IPv4Options{1, 1, 1, 1, 1, 0, 0, 0}, + wantIHL: header.IPv4MinimumSize + 8, + }, + { + name: "thirty nine byte options", + numberOfNops: 39, + encodedOptions: header.IPv4Options{ + 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 0, + }, + wantIHL: header.IPv4MinimumSize + 40, + }, + } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + serializeOpts := header.IPv4OptionsSerializer(make([]header.IPv4SerializableOption, test.numberOfNops)) + for i := range serializeOpts { + serializeOpts[i] = &header.IPv4SerializableNOPOption{} + } + paddedOptionLength := serializeOpts.Length() + ipHeaderLength := int(header.IPv4MinimumSize + paddedOptionLength) + if ipHeaderLength > header.IPv4MaximumHeaderSize { + t.Fatalf("IP header length too large: got = %d, want <= %d ", ipHeaderLength, header.IPv4MaximumHeaderSize) + } + totalLen := uint16(ipHeaderLength) + hdr := buffer.NewPrependable(int(totalLen)) + ip := header.IPv4(hdr.Prepend(ipHeaderLength)) + // To check the padding works, poison the last byte of the options space. + if paddedOptionLength != serializeOpts.Length() { + ip.SetHeaderLength(uint8(ipHeaderLength)) + ip.Options()[paddedOptionLength-1] = 0xff + ip.SetHeaderLength(0) + } + ip.Encode(&header.IPv4Fields{ + Options: serializeOpts, + }) + options := ip.Options() + wantOptions := test.encodedOptions + if got, want := int(ip.HeaderLength()), test.wantIHL; got != want { + t.Errorf("got IHL of %d, want %d", got, want) + } + + // cmp.Diff does not consider nil slices equal to empty slices, but we do. + if len(wantOptions) == 0 && len(options) == 0 { + return + } + + if diff := cmp.Diff(wantOptions, options); diff != "" { + t.Errorf("options mismatch (-want +got):\n%s", diff) + } + }) + } +} diff --git a/pkg/tcpip/header/ipv6_extension_headers.go b/pkg/tcpip/header/ipv6_extension_headers.go index 583c2c5d3..571eae233 100644 --- a/pkg/tcpip/header/ipv6_extension_headers.go +++ b/pkg/tcpip/header/ipv6_extension_headers.go @@ -47,6 +47,11 @@ const ( // IPv6NoNextHeaderIdentifier is the header identifier used to signify the end // of an IPv6 payload, as per RFC 8200 section 4.7. IPv6NoNextHeaderIdentifier IPv6ExtensionHeaderIdentifier = 59 + + // IPv6UnknownExtHdrIdentifier is reserved by IANA. + // https://www.iana.org/assignments/ipv6-parameters/ipv6-parameters.xhtml#extension-header + // "254 Use for experimentation and testing [RFC3692][RFC4727]" + IPv6UnknownExtHdrIdentifier IPv6ExtensionHeaderIdentifier = 254 ) const ( @@ -452,9 +457,11 @@ func (i *IPv6PayloadIterator) AsRawHeader(consume bool) IPv6RawPayloadHeader { // Since we consume the iterator, we return the payload as is. buf = i.payload - // Mark i as done. + // Mark i as done, but keep track of where we were for error reporting. *i = IPv6PayloadIterator{ nextHdrIdentifier: IPv6NoNextHeaderIdentifier, + headerOffset: i.headerOffset, + nextOffset: i.nextOffset, } } else { buf = i.payload.Clone(nil) diff --git a/pkg/tcpip/header/mld.go b/pkg/tcpip/header/mld.go new file mode 100644 index 000000000..ffe03c76a --- /dev/null +++ b/pkg/tcpip/header/mld.go @@ -0,0 +1,103 @@ +// 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 header + +import ( + "encoding/binary" + "fmt" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +const ( + // MLDMinimumSize is the minimum size for an MLD message. + MLDMinimumSize = 20 + + // MLDHopLimit is the Hop Limit for all IPv6 packets with an MLD message, as + // per RFC 2710 section 3. + MLDHopLimit = 1 + + // mldMaximumResponseDelayOffset is the offset to the Maximum Response Delay + // field within MLD. + mldMaximumResponseDelayOffset = 0 + + // mldMulticastAddressOffset is the offset to the Multicast Address field + // within MLD. + mldMulticastAddressOffset = 4 +) + +// MLD is a Multicast Listener Discovery message in an ICMPv6 packet. +// +// MLD will only contain the body of an ICMPv6 packet. +// +// As per RFC 2710 section 3, MLD messages have the following format (MLD only +// holds the bytes after the first four bytes in the diagram below): +// +// 0 1 2 3 +// 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Type | Code | Checksum | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Maximum Response Delay | Reserved | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | | +// + + +// | | +// + Multicast Address + +// | | +// + + +// | | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +type MLD []byte + +// MaximumResponseDelay returns the Maximum Response Delay. +func (m MLD) MaximumResponseDelay() time.Duration { + // As per RFC 2710 section 3.4: + // + // The Maximum Response Delay field is meaningful only in Query + // messages, and specifies the maximum allowed delay before sending a + // responding Report, in units of milliseconds. In all other messages, + // it is set to zero by the sender and ignored by receivers. + return time.Duration(binary.BigEndian.Uint16(m[mldMaximumResponseDelayOffset:])) * time.Millisecond +} + +// SetMaximumResponseDelay sets the Maximum Response Delay field. +// +// maxRespDelayMS is the value in milliseconds. +func (m MLD) SetMaximumResponseDelay(maxRespDelayMS uint16) { + binary.BigEndian.PutUint16(m[mldMaximumResponseDelayOffset:], maxRespDelayMS) +} + +// MulticastAddress returns the Multicast Address. +func (m MLD) MulticastAddress() tcpip.Address { + // As per RFC 2710 section 3.5: + // + // In a Query message, the Multicast Address field is set to zero when + // sending a General Query, and set to a specific IPv6 multicast address + // when sending a Multicast-Address-Specific Query. + // + // In a Report or Done message, the Multicast Address field holds a + // specific IPv6 multicast address to which the message sender is + // listening or is ceasing to listen, respectively. + return tcpip.Address(m[mldMulticastAddressOffset:][:IPv6AddressSize]) +} + +// SetMulticastAddress sets the Multicast Address field. +func (m MLD) SetMulticastAddress(multicastAddress tcpip.Address) { + if n := copy(m[mldMulticastAddressOffset:], multicastAddress); n != IPv6AddressSize { + panic(fmt.Sprintf("copied %d bytes, expected to copy %d bytes", n, IPv6AddressSize)) + } +} diff --git a/pkg/tcpip/header/mld_test.go b/pkg/tcpip/header/mld_test.go new file mode 100644 index 000000000..0cecf10d4 --- /dev/null +++ b/pkg/tcpip/header/mld_test.go @@ -0,0 +1,61 @@ +// 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 header + +import ( + "encoding/binary" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +func TestMLD(t *testing.T) { + b := []byte{ + // Maximum Response Delay + 0, 0, + + // Reserved + 0, 0, + + // MulticastAddress + 1, 2, 3, 4, 5, 6, 7, 8, 9, 0, 1, 2, 3, 4, 5, 6, + } + + const maxRespDelay = 513 + binary.BigEndian.PutUint16(b, maxRespDelay) + + mld := MLD(b) + + if got, want := mld.MaximumResponseDelay(), maxRespDelay*time.Millisecond; got != want { + t.Errorf("got mld.MaximumResponseDelay() = %s, want = %s", got, want) + } + + const newMaxRespDelay = 1234 + mld.SetMaximumResponseDelay(newMaxRespDelay) + if got, want := mld.MaximumResponseDelay(), newMaxRespDelay*time.Millisecond; got != want { + t.Errorf("got mld.MaximumResponseDelay() = %s, want = %s", got, want) + } + + if got, want := mld.MulticastAddress(), tcpip.Address([]byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 0, 1, 2, 3, 4, 5, 6}); got != want { + t.Errorf("got mld.MulticastAddress() = %s, want = %s", got, want) + } + + multicastAddress := tcpip.Address([]byte{15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0}) + mld.SetMulticastAddress(multicastAddress) + if got := mld.MulticastAddress(); got != multicastAddress { + t.Errorf("got mld.MulticastAddress() = %s, want = %s", got, multicastAddress) + } +} diff --git a/pkg/tcpip/header/ndp_options.go b/pkg/tcpip/header/ndp_options.go index 5d3975c56..554242f0c 100644 --- a/pkg/tcpip/header/ndp_options.go +++ b/pkg/tcpip/header/ndp_options.go @@ -298,7 +298,7 @@ func (b NDPOptions) Iter(check bool) (NDPOptionIterator, error) { return it, nil } -// Serialize serializes the provided list of NDP options into o. +// Serialize serializes the provided list of NDP options into b. // // Note, b must be of sufficient size to hold all the options in s. See // NDPOptionsSerializer.Length for details on the getting the total size diff --git a/pkg/tcpip/link/channel/BUILD b/pkg/tcpip/link/channel/BUILD index 39ca774ef..973f06cbc 100644 --- a/pkg/tcpip/link/channel/BUILD +++ b/pkg/tcpip/link/channel/BUILD @@ -9,7 +9,6 @@ go_library( deps = [ "//pkg/sync", "//pkg/tcpip", - "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/stack", ], diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index c95aef63c..0efbfb22b 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -32,7 +31,7 @@ type PacketInfo struct { Pkt *stack.PacketBuffer Proto tcpip.NetworkProtocolNumber GSO *stack.GSO - Route stack.Route + Route *stack.Route } // Notification is the interface for receiving notification from the packet @@ -271,21 +270,6 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe return n, nil } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - p := PacketInfo{ - Pkt: stack.NewPacketBuffer(stack.PacketBufferOptions{ - Data: vv, - }), - Proto: 0, - GSO: nil, - } - - e.q.Write(p) - - return nil -} - // Wait implements stack.LinkEndpoint.Wait. func (*Endpoint) Wait() {} diff --git a/pkg/tcpip/link/ethernet/ethernet.go b/pkg/tcpip/link/ethernet/ethernet.go index 3eef7cd56..beefcd008 100644 --- a/pkg/tcpip/link/ethernet/ethernet.go +++ b/pkg/tcpip/link/ethernet/ethernet.go @@ -62,7 +62,7 @@ func (e *Endpoint) Capabilities() stack.LinkEndpointCapabilities { // WritePacket implements stack.LinkEndpoint. func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, proto tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - e.AddHeader(e.Endpoint.LinkAddress(), r.RemoteLinkAddress, proto, pkt) + e.AddHeader(e.Endpoint.LinkAddress(), r.RemoteLinkAddress(), proto, pkt) return e.Endpoint.WritePacket(r, gso, proto, pkt) } @@ -71,7 +71,7 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe linkAddr := e.Endpoint.LinkAddress() for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { - e.AddHeader(linkAddr, r.RemoteLinkAddress, proto, pkt) + e.AddHeader(linkAddr, r.RemoteLinkAddress(), proto, pkt) } return e.Endpoint.WritePackets(r, gso, pkts, proto) diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 975309fc8..9f2084eae 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -410,7 +410,7 @@ func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.Net // currently writable, the packet is dropped. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { if e.hdrSize > 0 { - e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) + e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress(), protocol, pkt) } var builder iovec.Builder @@ -453,7 +453,7 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc mmsgHdrs := make([]rawfile.MMsgHdr, 0, len(batch)) for _, pkt := range batch { if e.hdrSize > 0 { - e.AddHeader(pkt.EgressRoute.LocalLinkAddress, pkt.EgressRoute.RemoteLinkAddress, pkt.NetworkProtocolNumber, pkt) + e.AddHeader(pkt.EgressRoute.LocalLinkAddress, pkt.EgressRoute.RemoteLinkAddress(), pkt.NetworkProtocolNumber, pkt) } var vnetHdrBuf []byte @@ -558,11 +558,6 @@ func viewsEqual(vs1, vs2 []buffer.View) bool { return len(vs1) == len(vs2) && (len(vs1) == 0 || &vs1[0] == &vs2[0]) } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - return rawfile.NonBlockingWrite(e.fds[0], vv.ToView()) -} - // InjectOutobund implements stack.InjectableEndpoint.InjectOutbound. func (e *endpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { return rawfile.NonBlockingWrite(e.fds[0], packet) diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 709f829c8..ce4da7230 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -183,9 +183,8 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32, hash u c := newContext(t, &Options{Address: laddr, MTU: mtu, EthernetHeader: eth, GSOMaxSize: gsoMaxSize}) defer c.cleanup() - r := &stack.Route{ - RemoteLinkAddress: raddr, - } + var r stack.Route + r.ResolveWith(raddr) // Build payload. payload := buffer.NewView(plen) @@ -220,7 +219,7 @@ func testWritePacket(t *testing.T, plen int, eth bool, gsoMaxSize uint32, hash u L3HdrLen: header.IPv4MaximumHeaderSize, } } - if err := c.ep.WritePacket(r, gso, proto, pkt); err != nil { + if err := c.ep.WritePacket(&r, gso, proto, pkt); err != nil { t.Fatalf("WritePacket failed: %v", err) } @@ -325,9 +324,9 @@ func TestPreserveSrcAddress(t *testing.T) { // Set LocalLinkAddress in route to the value of the bridged address. r := &stack.Route{ - RemoteLinkAddress: raddr, - LocalLinkAddress: baddr, + LocalLinkAddress: baddr, } + r.ResolveWith(raddr) pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ // WritePacket panics given a prependable with anything less than diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index 38aa694e4..edca57e4e 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -96,23 +96,6 @@ func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList panic("not implemented") } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - Data: vv, - }) - // There should be an ethernet header at the beginning of vv. - hdr, ok := pkt.LinkHeader().Consume(header.EthernetMinimumSize) - if !ok { - // Reject the packet if it's shorter than an ethernet header. - return tcpip.ErrBadAddress - } - linkHeader := header.Ethernet(hdr) - e.dispatcher.DeliverNetworkPacket("" /* remote */, "" /* local */, linkHeader.Type(), pkt) - - return nil -} - // ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. func (*endpoint) ARPHardwareType() header.ARPHardwareType { return header.ARPHardwareLoopback diff --git a/pkg/tcpip/link/muxed/BUILD b/pkg/tcpip/link/muxed/BUILD index e7493e5c5..cbda59775 100644 --- a/pkg/tcpip/link/muxed/BUILD +++ b/pkg/tcpip/link/muxed/BUILD @@ -8,7 +8,6 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/tcpip", - "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/stack", ], diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index 56a611825..22e79ce3a 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -17,7 +17,6 @@ package muxed import ( "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -106,13 +105,6 @@ func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, protoco return tcpip.ErrNoRoute } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (m *InjectableEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { - // WriteRawPacket doesn't get a route or network address, so there's - // nowhere to write this. - return tcpip.ErrNoRoute -} - // InjectOutbound writes outbound packets to the appropriate // LinkInjectableEndpoint based on the dest address. func (m *InjectableEndpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { diff --git a/pkg/tcpip/link/nested/BUILD b/pkg/tcpip/link/nested/BUILD index 2cdb23475..00b42b924 100644 --- a/pkg/tcpip/link/nested/BUILD +++ b/pkg/tcpip/link/nested/BUILD @@ -11,7 +11,6 @@ go_library( deps = [ "//pkg/sync", "//pkg/tcpip", - "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/stack", ], diff --git a/pkg/tcpip/link/nested/nested.go b/pkg/tcpip/link/nested/nested.go index d40de54df..0ee54c3d5 100644 --- a/pkg/tcpip/link/nested/nested.go +++ b/pkg/tcpip/link/nested/nested.go @@ -19,7 +19,6 @@ package nested import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -123,11 +122,6 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe return e.child.WritePackets(r, gso, pkts, protocol) } -// WriteRawPacket implements stack.LinkEndpoint. -func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - return e.child.WriteRawPacket(vv) -} - // Wait implements stack.LinkEndpoint. func (e *Endpoint) Wait() { e.child.Wait() diff --git a/pkg/tcpip/link/packetsocket/endpoint.go b/pkg/tcpip/link/packetsocket/endpoint.go index 3922c2a04..9a1b0c0c2 100644 --- a/pkg/tcpip/link/packetsocket/endpoint.go +++ b/pkg/tcpip/link/packetsocket/endpoint.go @@ -36,14 +36,14 @@ func New(lower stack.LinkEndpoint) stack.LinkEndpoint { // WritePacket implements stack.LinkEndpoint.WritePacket. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - e.Endpoint.DeliverOutboundPacket(r.RemoteLinkAddress, r.LocalLinkAddress, protocol, pkt) + e.Endpoint.DeliverOutboundPacket(r.RemoteLinkAddress(), r.LocalLinkAddress, protocol, pkt) return e.Endpoint.WritePacket(r, gso, protocol, pkt) } // WritePackets implements stack.LinkEndpoint.WritePackets. func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, proto tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { - e.Endpoint.DeliverOutboundPacket(pkt.EgressRoute.RemoteLinkAddress, pkt.EgressRoute.LocalLinkAddress, pkt.NetworkProtocolNumber, pkt) + e.Endpoint.DeliverOutboundPacket(pkt.EgressRoute.RemoteLinkAddress(), pkt.EgressRoute.LocalLinkAddress, pkt.NetworkProtocolNumber, pkt) } return e.Endpoint.WritePackets(r, gso, pkts, proto) diff --git a/pkg/tcpip/link/pipe/pipe.go b/pkg/tcpip/link/pipe/pipe.go index 523b0d24b..25c364391 100644 --- a/pkg/tcpip/link/pipe/pipe.go +++ b/pkg/tcpip/link/pipe/pipe.go @@ -55,7 +55,7 @@ func (e *Endpoint) WritePacket(r *stack.Route, _ *stack.GSO, proto tcpip.Network // remote address from the perspective of the other end of the pipe // (e.linked). Similarly, the remote address from the perspective of this // endpoint is the local address on the other end. - e.linked.dispatcher.DeliverNetworkPacket(r.LocalLinkAddress /* remote */, r.RemoteLinkAddress /* local */, proto, stack.NewPacketBuffer(stack.PacketBufferOptions{ + e.linked.dispatcher.DeliverNetworkPacket(r.LocalLinkAddress /* remote */, r.RemoteLinkAddress() /* local */, proto, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), })) @@ -67,11 +67,6 @@ func (*Endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList, panic("not implemented") } -// WriteRawPacket implements stack.LinkEndpoint. -func (*Endpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { - panic("not implemented") -} - // Attach implements stack.LinkEndpoint. func (e *Endpoint) Attach(dispatcher stack.NetworkDispatcher) { e.dispatcher = dispatcher diff --git a/pkg/tcpip/link/qdisc/fifo/BUILD b/pkg/tcpip/link/qdisc/fifo/BUILD index 1d0079bd6..5bea598eb 100644 --- a/pkg/tcpip/link/qdisc/fifo/BUILD +++ b/pkg/tcpip/link/qdisc/fifo/BUILD @@ -13,7 +13,6 @@ go_library( "//pkg/sleep", "//pkg/sync", "//pkg/tcpip", - "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/stack", ], diff --git a/pkg/tcpip/link/qdisc/fifo/endpoint.go b/pkg/tcpip/link/qdisc/fifo/endpoint.go index fc1e34fc7..27667f5f0 100644 --- a/pkg/tcpip/link/qdisc/fifo/endpoint.go +++ b/pkg/tcpip/link/qdisc/fifo/endpoint.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -156,7 +155,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // WritePacket caller's do not set the following fields in PacketBuffer // so we populate them here. newRoute := r.Clone() - pkt.EgressRoute = &newRoute + pkt.EgressRoute = newRoute pkt.GSOOptions = gso pkt.NetworkProtocolNumber = protocol d := e.dispatchers[int(pkt.Hash)%len(e.dispatchers)] @@ -183,7 +182,7 @@ func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, pkts stack.PacketB // the route here to ensure it doesn't get released while the // packet is still in our queue. newRoute := pkt.EgressRoute.Clone() - pkt.EgressRoute = &newRoute + pkt.EgressRoute = newRoute if !d.q.enqueue(pkt) { if enqueued > 0 { d.newPacketWaker.Assert() @@ -197,13 +196,6 @@ func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, pkts stack.PacketB return enqueued, nil } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - // TODO(gvisor.dev/issue/3267): Queue these packets as well once - // WriteRawPacket takes PacketBuffer instead of VectorisedView. - return e.lower.WriteRawPacket(vv) -} - // Wait implements stack.LinkEndpoint.Wait. func (e *endpoint) Wait() { e.lower.Wait() diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 7fb8a6c49..5660418fa 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -204,7 +204,7 @@ func (e *endpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.Net // WritePacket writes outbound packets to the file descriptor. If it is not // currently writable, the packet is dropped. func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress, protocol, pkt) + e.AddHeader(r.LocalLinkAddress, r.RemoteLinkAddress(), protocol, pkt) views := pkt.Views() // Transmit the packet. @@ -224,21 +224,6 @@ func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts stack.PacketB panic("not implemented") } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - views := vv.Views() - // Transmit the packet. - e.mu.Lock() - ok := e.tx.transmit(views...) - e.mu.Unlock() - - if !ok { - return tcpip.ErrWouldBlock - } - - return nil -} - // dispatchLoop reads packets from the rx queue in a loop and dispatches them // to the network stack. func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 22d5c97f1..7131392cc 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -260,9 +260,8 @@ func TestSimpleSend(t *testing.T) { defer c.cleanup() // Prepare route. - r := stack.Route{ - RemoteLinkAddress: remoteLinkAddr, - } + var r stack.Route + r.ResolveWith(remoteLinkAddr) for iters := 1000; iters > 0; iters-- { func() { @@ -342,9 +341,9 @@ func TestPreserveSrcAddressInSend(t *testing.T) { newLocalLinkAddress := tcpip.LinkAddress(strings.Repeat("0xFE", 6)) // Set both remote and local link address in route. r := stack.Route{ - RemoteLinkAddress: remoteLinkAddr, - LocalLinkAddress: newLocalLinkAddress, + LocalLinkAddress: newLocalLinkAddress, } + r.ResolveWith(remoteLinkAddr) pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ // WritePacket panics given a prependable with anything less than @@ -395,9 +394,8 @@ func TestFillTxQueue(t *testing.T) { defer c.cleanup() // Prepare to send a packet. - r := stack.Route{ - RemoteLinkAddress: remoteLinkAddr, - } + var r stack.Route + r.ResolveWith(remoteLinkAddr) buf := buffer.NewView(100) @@ -444,9 +442,8 @@ func TestFillTxQueueAfterBadCompletion(t *testing.T) { c.txq.rx.Flush() // Prepare to send a packet. - r := stack.Route{ - RemoteLinkAddress: remoteLinkAddr, - } + var r stack.Route + r.ResolveWith(remoteLinkAddr) buf := buffer.NewView(100) @@ -509,9 +506,8 @@ func TestFillTxMemory(t *testing.T) { defer c.cleanup() // Prepare to send a packet. - r := stack.Route{ - RemoteLinkAddress: remoteLinkAddr, - } + var r stack.Route + r.ResolveWith(remoteLinkAddr) buf := buffer.NewView(100) @@ -557,9 +553,8 @@ func TestFillTxMemoryWithMultiBuffer(t *testing.T) { defer c.cleanup() // Prepare to send a packet. - r := stack.Route{ - RemoteLinkAddress: remoteLinkAddr, - } + var r stack.Route + r.ResolveWith(remoteLinkAddr) buf := buffer.NewView(100) diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index b3e8c4b92..8d9a91020 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -53,16 +53,35 @@ type endpoint struct { nested.Endpoint writer io.Writer maxPCAPLen uint32 + logPrefix string } var _ stack.GSOEndpoint = (*endpoint)(nil) var _ stack.LinkEndpoint = (*endpoint)(nil) var _ stack.NetworkDispatcher = (*endpoint)(nil) +type direction int + +const ( + directionSend = iota + directionRecv +) + // New creates a new sniffer link-layer endpoint. It wraps around another // endpoint and logs packets and they traverse the endpoint. func New(lower stack.LinkEndpoint) stack.LinkEndpoint { - sniffer := &endpoint{} + return NewWithPrefix(lower, "") +} + +// NewWithPrefix creates a new sniffer link-layer endpoint. It wraps around +// another endpoint and logs packets prefixed with logPrefix as they traverse +// the endpoint. +// +// logPrefix is prepended to the log line without any separators. +// E.g. logPrefix = "NIC:en0/" will produce log lines like +// "NIC:en0/send udp [...]". +func NewWithPrefix(lower stack.LinkEndpoint, logPrefix string) stack.LinkEndpoint { + sniffer := &endpoint{logPrefix: logPrefix} sniffer.Endpoint.Init(lower, sniffer) return sniffer } @@ -120,7 +139,7 @@ func NewWithWriter(lower stack.LinkEndpoint, writer io.Writer, snapLen uint32) ( // called by the link-layer endpoint being wrapped when a packet arrives, and // logs the packet before forwarding to the actual dispatcher. func (e *endpoint) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { - e.dumpPacket("recv", nil, protocol, pkt) + e.dumpPacket(directionRecv, nil, protocol, pkt) e.Endpoint.DeliverNetworkPacket(remote, local, protocol, pkt) } @@ -129,10 +148,10 @@ func (e *endpoint) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protoc e.Endpoint.DeliverOutboundPacket(remote, local, protocol, pkt) } -func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { +func (e *endpoint) dumpPacket(dir direction, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) { writer := e.writer if writer == nil && atomic.LoadUint32(&LogPackets) == 1 { - logPacket(prefix, protocol, pkt, gso) + logPacket(e.logPrefix, dir, protocol, pkt, gso) } if writer != nil && atomic.LoadUint32(&LogPacketsToPCAP) == 1 { totalLength := pkt.Size() @@ -169,7 +188,7 @@ func (e *endpoint) dumpPacket(prefix string, gso *stack.GSO, protocol tcpip.Netw // higher-level protocols to write packets; it just logs the packet and // forwards the request to the lower endpoint. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - e.dumpPacket("send", gso, protocol, pkt) + e.dumpPacket(directionSend, gso, protocol, pkt) return e.Endpoint.WritePacket(r, gso, protocol, pkt) } @@ -178,20 +197,12 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne // forwards the request to the lower endpoint. func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { - e.dumpPacket("send", gso, protocol, pkt) + e.dumpPacket(directionSend, gso, protocol, pkt) } return e.Endpoint.WritePackets(r, gso, pkts, protocol) } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - e.dumpPacket("send", nil, 0, stack.NewPacketBuffer(stack.PacketBufferOptions{ - Data: vv, - })) - return e.Endpoint.WriteRawPacket(vv) -} - -func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { +func logPacket(prefix string, dir direction, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer, gso *stack.GSO) { // Figure out the network layer info. var transProto uint8 src := tcpip.Address("unknown") @@ -201,6 +212,16 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P var fragmentOffset uint16 var moreFragments bool + var directionPrefix string + switch dir { + case directionSend: + directionPrefix = "send" + case directionRecv: + directionPrefix = "recv" + default: + panic(fmt.Sprintf("unrecognized direction: %d", dir)) + } + // Clone the packet buffer to not modify the original. // // We don't clone the original packet buffer so that the new packet buffer @@ -248,15 +269,16 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P arp := header.ARP(pkt.NetworkHeader().View()) log.Infof( - "%s arp %s (%s) -> %s (%s) valid:%t", + "%s%s arp %s (%s) -> %s (%s) valid:%t", prefix, + directionPrefix, tcpip.Address(arp.ProtocolAddressSender()), tcpip.LinkAddress(arp.HardwareAddressSender()), tcpip.Address(arp.ProtocolAddressTarget()), tcpip.LinkAddress(arp.HardwareAddressTarget()), arp.IsValid(), ) return default: - log.Infof("%s unknown network protocol", prefix) + log.Infof("%s%s unknown network protocol", prefix, directionPrefix) return } @@ -300,7 +322,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P icmpType = "info reply" } } - log.Infof("%s %s %s -> %s %s len:%d id:%04x code:%d", prefix, transName, src, dst, icmpType, size, id, icmp.Code()) + log.Infof("%s%s %s %s -> %s %s len:%d id:%04x code:%d", prefix, directionPrefix, transName, src, dst, icmpType, size, id, icmp.Code()) return case header.ICMPv6ProtocolNumber: @@ -335,7 +357,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P case header.ICMPv6RedirectMsg: icmpType = "redirect message" } - log.Infof("%s %s %s -> %s %s len:%d id:%04x code:%d", prefix, transName, src, dst, icmpType, size, id, icmp.Code()) + log.Infof("%s%s %s %s -> %s %s len:%d id:%04x code:%d", prefix, directionPrefix, transName, src, dst, icmpType, size, id, icmp.Code()) return case header.UDPProtocolNumber: @@ -391,7 +413,7 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P } default: - log.Infof("%s %s -> %s unknown transport protocol: %d", prefix, src, dst, transProto) + log.Infof("%s%s %s -> %s unknown transport protocol: %d", prefix, directionPrefix, src, dst, transProto) return } @@ -399,5 +421,5 @@ func logPacket(prefix string, protocol tcpip.NetworkProtocolNumber, pkt *stack.P details += fmt.Sprintf(" gso: %+v", gso) } - log.Infof("%s %s %s:%d -> %s:%d len:%d id:%04x %s", prefix, transName, src, srcPort, dst, dstPort, size, id, details) + log.Infof("%s%s %s %s:%d -> %s:%d len:%d id:%04x %s", prefix, directionPrefix, transName, src, srcPort, dst, dstPort, size, id, details) } diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 9a76bdba7..a364c5801 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -264,7 +264,7 @@ func (d *Device) encodePkt(info *channel.PacketInfo) (buffer.View, bool) { // If the packet does not already have link layer header, and the route // does not exist, we can't compute it. This is possibly a raw packet, tun // device doesn't support this at the moment. - if info.Pkt.LinkHeader().View().IsEmpty() && info.Route.RemoteLinkAddress == "" { + if info.Pkt.LinkHeader().View().IsEmpty() && info.Route.RemoteLinkAddress() == "" { return nil, false } @@ -272,7 +272,7 @@ func (d *Device) encodePkt(info *channel.PacketInfo) (buffer.View, bool) { if d.hasFlags(linux.IFF_TAP) { // Add ethernet header if not provided. if info.Pkt.LinkHeader().View().IsEmpty() { - d.endpoint.AddHeader(info.Route.LocalLinkAddress, info.Route.RemoteLinkAddress, info.Proto, info.Pkt) + d.endpoint.AddHeader(info.Route.LocalLinkAddress, info.Route.RemoteLinkAddress(), info.Proto, info.Pkt) } vv.AppendView(info.Pkt.LinkHeader().View()) } diff --git a/pkg/tcpip/link/waitable/BUILD b/pkg/tcpip/link/waitable/BUILD index ee84c3d96..9b4602c1b 100644 --- a/pkg/tcpip/link/waitable/BUILD +++ b/pkg/tcpip/link/waitable/BUILD @@ -11,7 +11,6 @@ go_library( deps = [ "//pkg/gate", "//pkg/tcpip", - "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/stack", ], @@ -25,7 +24,6 @@ go_test( library = ":waitable", deps = [ "//pkg/tcpip", - "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/stack", ], diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index b152a0f26..cf0077f43 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -24,7 +24,6 @@ package waitable import ( "gvisor.dev/gvisor/pkg/gate" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -132,17 +131,6 @@ func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe return n, err } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *Endpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - if !e.writeGate.Enter() { - return nil - } - - err := e.lower.WriteRawPacket(vv) - e.writeGate.Leave() - return err -} - // WaitWrite prevents new calls to WritePacket from reaching the lower endpoint, // and waits for inflight ones to finish before returning. func (e *Endpoint) WaitWrite() { diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index 94827fc56..cf7fb5126 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -18,7 +18,6 @@ import ( "testing" "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -81,11 +80,6 @@ func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, pkts stack. return pkts.Len(), nil } -func (e *countedEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { - e.writeCount++ - return nil -} - // ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. func (*countedEndpoint) ARPHardwareType() header.ARPHardwareType { panic("unimplemented") diff --git a/pkg/tcpip/network/BUILD b/pkg/tcpip/network/BUILD index b38aff0b8..9ebf31b78 100644 --- a/pkg/tcpip/network/BUILD +++ b/pkg/tcpip/network/BUILD @@ -7,12 +7,14 @@ go_test( size = "small", srcs = [ "ip_test.go", + "multicast_group_test.go", ], deps = [ "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/checker", + "//pkg/tcpip/faketime", "//pkg/tcpip/header", "//pkg/tcpip/header/parse", "//pkg/tcpip/link/channel", diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index f462524c9..0fb373612 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -319,9 +319,9 @@ func TestDirectRequestWithNeighborCache(t *testing.T) { copy(h.HardwareAddressSender(), test.senderLinkAddr) copy(h.ProtocolAddressSender(), test.senderAddr) copy(h.ProtocolAddressTarget(), test.targetAddr) - c.linkEP.InjectInbound(arp.ProtocolNumber, &stack.PacketBuffer{ + c.linkEP.InjectInbound(arp.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: v.ToVectorisedView(), - }) + })) if !test.isValid { // No packets should be sent after receiving an invalid ARP request. @@ -442,9 +442,9 @@ func (*testInterface) Promiscuous() bool { func (t *testInterface) WritePacketToRemote(remoteLinkAddr tcpip.LinkAddress, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { r := stack.Route{ - NetProto: protocol, - RemoteLinkAddress: remoteLinkAddr, + NetProto: protocol, } + r.ResolveWith(remoteLinkAddr) return t.LinkEndpoint.WritePacket(&r, gso, protocol, pkt) } @@ -557,8 +557,8 @@ func TestLinkAddressRequest(t *testing.T) { t.Fatal("expected to send a link address request") } - if pkt.Route.RemoteLinkAddress != test.expectedRemoteLinkAddr { - t.Errorf("got pkt.Route.RemoteLinkAddress = %s, want = %s", pkt.Route.RemoteLinkAddress, test.expectedRemoteLinkAddr) + if got := pkt.Route.RemoteLinkAddress(); got != test.expectedRemoteLinkAddr { + t.Errorf("got pkt.Route.RemoteLinkAddress() = %s, want = %s", got, test.expectedRemoteLinkAddr) } rep := header.ARP(stack.PayloadSince(pkt.Pkt.NetworkHeader())) diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index c75ca7d71..d31296a41 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -46,9 +46,13 @@ const ( ) var ( - // ErrInvalidArgs indicates to the caller that that an invalid argument was + // ErrInvalidArgs indicates to the caller that an invalid argument was // provided. ErrInvalidArgs = errors.New("invalid args") + + // ErrFragmentOverlap indicates that, during reassembly, a fragment overlaps + // with another one. + ErrFragmentOverlap = errors.New("overlapping fragments") ) // FragmentID is the identifier for a fragment. diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go index 19f4920b3..04072d966 100644 --- a/pkg/tcpip/network/fragmentation/reassembler.go +++ b/pkg/tcpip/network/fragmentation/reassembler.go @@ -26,9 +26,9 @@ import ( ) type hole struct { - first uint16 - last uint16 - deleted bool + first uint16 + last uint16 + filled bool } type reassembler struct { @@ -38,7 +38,7 @@ type reassembler struct { proto uint8 mu sync.Mutex holes []hole - deleted int + filled int heap fragHeap done bool creationTime int64 @@ -53,44 +53,86 @@ func newReassembler(id FragmentID, clock tcpip.Clock) *reassembler { creationTime: clock.NowMonotonic(), } r.holes = append(r.holes, hole{ - first: 0, - last: math.MaxUint16, - deleted: false}) + first: 0, + last: math.MaxUint16, + filled: false, + }) return r } -// updateHoles updates the list of holes for an incoming fragment and -// returns true iff the fragment filled at least part of an existing hole. -func (r *reassembler) updateHoles(first, last uint16, more bool) bool { - used := false +// updateHoles updates the list of holes for an incoming fragment. It returns +// true if the fragment fits, it is not a duplicate and it does not overlap with +// another fragment. +// +// For IPv6, overlaps with an existing fragment are explicitly forbidden by +// RFC 8200 section 4.5: +// If any of the fragments being reassembled overlap with any other fragments +// being reassembled for the same packet, reassembly of that packet must be +// abandoned and all the fragments that have been received for that packet +// must be discarded, and no ICMP error messages should be sent. +// +// It is not explicitly forbidden for IPv4, but to keep parity with Linux we +// disallow it as well: +// https://github.com/torvalds/linux/blob/38525c6/net/ipv4/inet_fragment.c#L349 +func (r *reassembler) updateHoles(first, last uint16, more bool) (bool, error) { for i := range r.holes { - if r.holes[i].deleted || first > r.holes[i].last || last < r.holes[i].first { + currentHole := &r.holes[i] + + if currentHole.filled || last < currentHole.first || currentHole.last < first { continue } - used = true - r.deleted++ - r.holes[i].deleted = true - if first > r.holes[i].first { - r.holes = append(r.holes, hole{r.holes[i].first, first - 1, false}) + + if first < currentHole.first || currentHole.last < last { + // Incoming fragment only partially fits in the free hole. + return false, ErrFragmentOverlap + } + + r.filled++ + if first > currentHole.first { + r.holes = append(r.holes, hole{ + first: currentHole.first, + last: first - 1, + filled: false, + }) + } + if last < currentHole.last && more { + r.holes = append(r.holes, hole{ + first: last + 1, + last: currentHole.last, + filled: false, + }) } - if last < r.holes[i].last && more { - r.holes = append(r.holes, hole{last + 1, r.holes[i].last, false}) + // Update the current hole to precisely match the incoming fragment. + r.holes[i] = hole{ + first: first, + last: last, + filled: true, } + return true, nil } - return used + + // Incoming fragment is a duplicate/subset, or its offset comes after the end + // of the reassembled payload. + return false, nil } func (r *reassembler) process(first, last uint16, more bool, proto uint8, pkt *stack.PacketBuffer) (buffer.VectorisedView, uint8, bool, int, error) { r.mu.Lock() defer r.mu.Unlock() - consumed := 0 if r.done { // A concurrent goroutine might have already reassembled // the packet and emptied the heap while this goroutine // was waiting on the mutex. We don't have to do anything in this case. - return buffer.VectorisedView{}, 0, false, consumed, nil + return buffer.VectorisedView{}, 0, false, 0, nil } - if r.updateHoles(first, last, more) { + + used, err := r.updateHoles(first, last, more) + if err != nil { + return buffer.VectorisedView{}, 0, false, 0, fmt.Errorf("fragment reassembly failed: %w", err) + } + + var consumed int + if used { // For IPv6, it is possible to have different Protocol values between // fragments of a packet (because, unlike IPv4, the Protocol is not used to // identify a fragment). In this case, only the Protocol of the first @@ -109,13 +151,14 @@ func (r *reassembler) process(first, last uint16, more bool, proto uint8, pkt *s consumed = vv.Size() r.size += consumed } - // Check if all the holes have been deleted and we are ready to reassamble. - if r.deleted < len(r.holes) { + + // Check if all the holes have been filled and we are ready to reassemble. + if r.filled < len(r.holes) { return buffer.VectorisedView{}, 0, false, consumed, nil } res, err := r.heap.reassemble() if err != nil { - return buffer.VectorisedView{}, 0, false, consumed, fmt.Errorf("fragment reassembly failed: %w", err) + return buffer.VectorisedView{}, 0, false, 0, fmt.Errorf("fragment reassembly failed: %w", err) } return res, r.proto, true, consumed, nil } diff --git a/pkg/tcpip/network/fragmentation/reassembler_test.go b/pkg/tcpip/network/fragmentation/reassembler_test.go index a0a04a027..cee3063b1 100644 --- a/pkg/tcpip/network/fragmentation/reassembler_test.go +++ b/pkg/tcpip/network/fragmentation/reassembler_test.go @@ -16,92 +16,124 @@ package fragmentation import ( "math" - "reflect" "testing" + "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip/faketime" ) -type updateHolesInput struct { - first uint16 - last uint16 - more bool +type updateHolesParams struct { + first uint16 + last uint16 + more bool + wantUsed bool + wantError error } -var holesTestCases = []struct { - comment string - in []updateHolesInput - want []hole -}{ - { - comment: "No fragments. Expected holes: {[0 -> inf]}.", - in: []updateHolesInput{}, - want: []hole{{first: 0, last: math.MaxUint16, deleted: false}}, - }, - { - comment: "One fragment at beginning. Expected holes: {[2, inf]}.", - in: []updateHolesInput{{first: 0, last: 1, more: true}}, - want: []hole{ - {first: 0, last: math.MaxUint16, deleted: true}, - {first: 2, last: math.MaxUint16, deleted: false}, +func TestUpdateHoles(t *testing.T) { + var tests = []struct { + name string + params []updateHolesParams + want []hole + }{ + { + name: "No fragments", + params: nil, + want: []hole{{first: 0, last: math.MaxUint16, filled: false}}, }, - }, - { - comment: "One fragment in the middle. Expected holes: {[0, 0], [3, inf]}.", - in: []updateHolesInput{{first: 1, last: 2, more: true}}, - want: []hole{ - {first: 0, last: math.MaxUint16, deleted: true}, - {first: 0, last: 0, deleted: false}, - {first: 3, last: math.MaxUint16, deleted: false}, + { + name: "One fragment at beginning", + params: []updateHolesParams{{first: 0, last: 1, more: true, wantUsed: true, wantError: nil}}, + want: []hole{ + {first: 0, last: 1, filled: true}, + {first: 2, last: math.MaxUint16, filled: false}, + }, }, - }, - { - comment: "One fragment at the end. Expected holes: {[0, 0]}.", - in: []updateHolesInput{{first: 1, last: 2, more: false}}, - want: []hole{ - {first: 0, last: math.MaxUint16, deleted: true}, - {first: 0, last: 0, deleted: false}, + { + name: "One fragment in the middle", + params: []updateHolesParams{{first: 1, last: 2, more: true, wantUsed: true, wantError: nil}}, + want: []hole{ + {first: 1, last: 2, filled: true}, + {first: 0, last: 0, filled: false}, + {first: 3, last: math.MaxUint16, filled: false}, + }, }, - }, - { - comment: "One fragment completing a packet. Expected holes: {}.", - in: []updateHolesInput{{first: 0, last: 1, more: false}}, - want: []hole{ - {first: 0, last: math.MaxUint16, deleted: true}, + { + name: "One fragment at the end", + params: []updateHolesParams{{first: 1, last: 2, more: false, wantUsed: true, wantError: nil}}, + want: []hole{ + {first: 1, last: 2, filled: true}, + {first: 0, last: 0, filled: false}, + }, }, - }, - { - comment: "Two non-overlapping fragments completing a packet. Expected holes: {}.", - in: []updateHolesInput{ - {first: 0, last: 1, more: true}, - {first: 2, last: 3, more: false}, + { + name: "One fragment completing a packet", + params: []updateHolesParams{{first: 0, last: 1, more: false, wantUsed: true, wantError: nil}}, + want: []hole{ + {first: 0, last: 1, filled: true}, + }, }, - want: []hole{ - {first: 0, last: math.MaxUint16, deleted: true}, - {first: 2, last: math.MaxUint16, deleted: true}, + { + name: "Two fragments completing a packet", + params: []updateHolesParams{ + {first: 0, last: 1, more: true, wantUsed: true, wantError: nil}, + {first: 2, last: 3, more: false, wantUsed: true, wantError: nil}, + }, + want: []hole{ + {first: 0, last: 1, filled: true}, + {first: 2, last: 3, filled: true}, + }, }, - }, - { - comment: "Two overlapping fragments completing a packet. Expected holes: {}.", - in: []updateHolesInput{ - {first: 0, last: 2, more: true}, - {first: 2, last: 3, more: false}, + { + name: "Two fragments completing a packet with a duplicate", + params: []updateHolesParams{ + {first: 0, last: 1, more: true, wantUsed: true, wantError: nil}, + {first: 0, last: 1, more: true, wantUsed: false, wantError: nil}, + {first: 2, last: 3, more: false, wantUsed: true, wantError: nil}, + }, + want: []hole{ + {first: 0, last: 1, filled: true}, + {first: 2, last: 3, filled: true}, + }, }, - want: []hole{ - {first: 0, last: math.MaxUint16, deleted: true}, - {first: 3, last: math.MaxUint16, deleted: true}, + { + name: "Two overlapping fragments", + params: []updateHolesParams{ + {first: 0, last: 10, more: true, wantUsed: true, wantError: nil}, + {first: 5, last: 15, more: false, wantUsed: false, wantError: ErrFragmentOverlap}, + {first: 11, last: 15, more: false, wantUsed: true, wantError: nil}, + }, + want: []hole{ + {first: 0, last: 10, filled: true}, + {first: 11, last: 15, filled: true}, + }, }, - }, -} + { + name: "Out of bounds fragment", + params: []updateHolesParams{ + {first: 0, last: 10, more: true, wantUsed: true, wantError: nil}, + {first: 11, last: 15, more: false, wantUsed: true, wantError: nil}, + {first: 16, last: 20, more: false, wantUsed: false, wantError: nil}, + }, + want: []hole{ + {first: 0, last: 10, filled: true}, + {first: 11, last: 15, filled: true}, + }, + }, + } -func TestUpdateHoles(t *testing.T) { - for _, c := range holesTestCases { - r := newReassembler(FragmentID{}, &faketime.NullClock{}) - for _, i := range c.in { - r.updateHoles(i.first, i.last, i.more) - } - if !reflect.DeepEqual(r.holes, c.want) { - t.Errorf("Test \"%s\" produced unexepetced holes. Got %v. Want %v", c.comment, r.holes, c.want) - } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + r := newReassembler(FragmentID{}, &faketime.NullClock{}) + for _, param := range test.params { + used, err := r.updateHoles(param.first, param.last, param.more) + if used != param.wantUsed || err != param.wantError { + t.Errorf("got r.updateHoles(%d, %d, %t) = (%t, %v), want = (%t, %v)", param.first, param.last, param.more, used, err, param.wantUsed, param.wantError) + } + } + if diff := cmp.Diff(test.want, r.holes, cmp.AllowUnexported(hole{})); diff != "" { + t.Errorf("r.holes mismatch (-want +got):\n%s", diff) + } + }) } } diff --git a/pkg/tcpip/network/ip/BUILD b/pkg/tcpip/network/ip/BUILD new file mode 100644 index 000000000..6ca200b48 --- /dev/null +++ b/pkg/tcpip/network/ip/BUILD @@ -0,0 +1,25 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "ip", + srcs = ["generic_multicast_protocol.go"], + visibility = ["//visibility:public"], + deps = [ + "//pkg/sync", + "//pkg/tcpip", + ], +) + +go_test( + name = "ip_test", + size = "small", + srcs = ["generic_multicast_protocol_test.go"], + deps = [ + ":ip", + "//pkg/tcpip", + "//pkg/tcpip/faketime", + "@com_github_google_go_cmp//cmp:go_default_library", + ], +) diff --git a/pkg/tcpip/network/ip/generic_multicast_protocol.go b/pkg/tcpip/network/ip/generic_multicast_protocol.go new file mode 100644 index 000000000..f14e2a88a --- /dev/null +++ b/pkg/tcpip/network/ip/generic_multicast_protocol.go @@ -0,0 +1,546 @@ +// 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 ip holds IPv4/IPv6 common utilities. +package ip + +import ( + "fmt" + "math/rand" + "time" + + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" +) + +// hostState is the state a host may be in for a multicast group. +type hostState int + +// The states below are generic across IGMPv2 (RFC 2236 section 6) and MLDv1 +// (RFC 2710 section 5). Even though the states are generic across both IGMPv2 +// and MLDv1, IGMPv2 terminology will be used. +const ( + // nonMember is the "'Non-Member' state, when the host does not belong to the + // group on the interface. This is the initial state for all memberships on + // all network interfaces; it requires no storage in the host." + // + // 'Non-Listener' is the MLDv1 term used to describe this state. + // + // This state is used to keep track of groups that have been joined locally, + // but without advertising the membership to the network. + nonMember hostState = iota + + // delayingMember is the "'Delaying Member' state, when the host belongs to + // the group on the interface and has a report delay timer running for that + // membership." + // + // 'Delaying Listener' is the MLDv1 term used to describe this state. + delayingMember + + // idleMember is the "Idle Member" state, when the host belongs to the group + // on the interface and does not have a report delay timer running for that + // membership. + // + // 'Idle Listener' is the MLDv1 term used to describe this state. + idleMember +) + +// multicastGroupState holds the Generic Multicast Protocol state for a +// multicast group. +type multicastGroupState struct { + // joins is the number of times the group has been joined. + joins uint64 + + // state holds the host's state for the group. + state hostState + + // lastToSendReport is true if we sent the last report for the group. It is + // used to track whether there are other hosts on the subnet that are also + // members of the group. + // + // Defined in RFC 2236 section 6 page 9 for IGMPv2 and RFC 2710 section 5 page + // 8 for MLDv1. + lastToSendReport bool + + // delayedReportJob is used to delay sending responses to membership report + // messages in order to reduce duplicate reports from multiple hosts on the + // interface. + // + // Must not be nil. + delayedReportJob *tcpip.Job +} + +// GenericMulticastProtocolOptions holds options for the generic multicast +// protocol. +type GenericMulticastProtocolOptions struct { + // Enabled indicates whether the generic multicast protocol will be + // performed. + // + // When enabled, the protocol may transmit report and leave messages when + // joining and leaving multicast groups respectively, and handle incoming + // packets. + // + // When disabled, the protocol will still keep track of locally joined groups, + // it just won't transmit and handle packets, or update groups' state. + Enabled bool + + // Rand is the source of random numbers. + Rand *rand.Rand + + // Clock is the clock used to create timers. + Clock tcpip.Clock + + // Protocol is the implementation of the variant of multicast group protocol + // in use. + Protocol MulticastGroupProtocol + + // MaxUnsolicitedReportDelay is the maximum amount of time to wait between + // transmitting unsolicited reports. + // + // Unsolicited reports are transmitted when a group is newly joined. + MaxUnsolicitedReportDelay time.Duration + + // AllNodesAddress is a multicast address that all nodes on a network should + // be a member of. + // + // This address will not have the generic multicast protocol performed on it; + // it will be left in the non member/listener state, and packets will never + // be sent for it. + AllNodesAddress tcpip.Address +} + +// MulticastGroupProtocol is a multicast group protocol whose core state machine +// can be represented by GenericMulticastProtocolState. +type MulticastGroupProtocol interface { + // SendReport sends a multicast report for the specified group address. + SendReport(groupAddress tcpip.Address) *tcpip.Error + + // SendLeave sends a multicast leave for the specified group address. + SendLeave(groupAddress tcpip.Address) *tcpip.Error +} + +// GenericMulticastProtocolState is the per interface generic multicast protocol +// state. +// +// There is actually no protocol named "Generic Multicast Protocol". Instead, +// the term used to refer to a generic multicast protocol that applies to both +// IPv4 and IPv6. Specifically, Generic Multicast Protocol is the core state +// machine of IGMPv2 as defined by RFC 2236 and MLDv1 as defined by RFC 2710. +// +// GenericMulticastProtocolState.Init MUST be called before calling any of +// the methods on GenericMulticastProtocolState. +type GenericMulticastProtocolState struct { + opts GenericMulticastProtocolOptions + + mu struct { + sync.RWMutex + + // memberships holds group addresses and their associated state. + memberships map[tcpip.Address]multicastGroupState + } +} + +// Init initializes the Generic Multicast Protocol state. +// +// maxUnsolicitedReportDelay is the maximum time between sending unsolicited +// reports after joining a group. +func (g *GenericMulticastProtocolState) Init(opts GenericMulticastProtocolOptions) { + g.mu.Lock() + defer g.mu.Unlock() + g.opts = opts + g.mu.memberships = make(map[tcpip.Address]multicastGroupState) +} + +// MakeAllNonMember transitions all groups to the non-member state. +// +// The groups will still be considered joined locally. +func (g *GenericMulticastProtocolState) MakeAllNonMember() { + if !g.opts.Enabled { + return + } + + g.mu.Lock() + defer g.mu.Unlock() + + for groupAddress, info := range g.mu.memberships { + g.transitionToNonMemberLocked(groupAddress, &info) + g.mu.memberships[groupAddress] = info + } +} + +// InitializeGroups initializes each group, as if they were newly joined but +// without affecting the groups' join count. +// +// Must only be called after calling MakeAllNonMember as a group should not be +// initialized while it is not in the non-member state. +func (g *GenericMulticastProtocolState) InitializeGroups() { + if !g.opts.Enabled { + return + } + + g.mu.Lock() + defer g.mu.Unlock() + + for groupAddress, info := range g.mu.memberships { + g.initializeNewMemberLocked(groupAddress, &info) + g.mu.memberships[groupAddress] = info + } +} + +// JoinGroup handles joining a new group. +// +// If dontInitialize is true, the group will be not be initialized and will be +// left in the non-member state - no packets will be sent for it until it is +// initialized via InitializeGroups. +func (g *GenericMulticastProtocolState) JoinGroup(groupAddress tcpip.Address, dontInitialize bool) { + g.mu.Lock() + defer g.mu.Unlock() + + if info, ok := g.mu.memberships[groupAddress]; ok { + // The group has already been joined. + info.joins++ + g.mu.memberships[groupAddress] = info + return + } + + info := multicastGroupState{ + // Since we just joined the group, its count is 1. + joins: 1, + // The state will be updated below, if required. + state: nonMember, + lastToSendReport: false, + delayedReportJob: tcpip.NewJob(g.opts.Clock, &g.mu, func() { + info, ok := g.mu.memberships[groupAddress] + if !ok { + panic(fmt.Sprintf("expected to find group state for group = %s", groupAddress)) + } + + info.lastToSendReport = g.opts.Protocol.SendReport(groupAddress) == nil + info.state = idleMember + g.mu.memberships[groupAddress] = info + }), + } + + if !dontInitialize && g.opts.Enabled { + g.initializeNewMemberLocked(groupAddress, &info) + } + + g.mu.memberships[groupAddress] = info +} + +// IsLocallyJoined returns true if the group is locally joined. +func (g *GenericMulticastProtocolState) IsLocallyJoined(groupAddress tcpip.Address) bool { + g.mu.RLock() + defer g.mu.RUnlock() + _, ok := g.mu.memberships[groupAddress] + return ok +} + +// LeaveGroup handles leaving the group. +// +// Returns false if the group is not currently joined. +func (g *GenericMulticastProtocolState) LeaveGroup(groupAddress tcpip.Address) bool { + g.mu.Lock() + defer g.mu.Unlock() + + info, ok := g.mu.memberships[groupAddress] + if !ok { + return false + } + + if info.joins == 0 { + panic(fmt.Sprintf("tried to leave group %s with a join count of 0", groupAddress)) + } + info.joins-- + if info.joins != 0 { + // If we still have outstanding joins, then do nothing further. + g.mu.memberships[groupAddress] = info + return true + } + + g.transitionToNonMemberLocked(groupAddress, &info) + delete(g.mu.memberships, groupAddress) + return true +} + +// HandleQuery handles a query message with the specified maximum response time. +// +// If the group address is unspecified, then reports will be scheduled for all +// joined groups. +// +// Report(s) will be scheduled to be sent after a random duration between 0 and +// the maximum response time. +func (g *GenericMulticastProtocolState) HandleQuery(groupAddress tcpip.Address, maxResponseTime time.Duration) { + if !g.opts.Enabled { + return + } + + g.mu.Lock() + defer g.mu.Unlock() + + // As per RFC 2236 section 2.4 (for IGMPv2), + // + // In a Membership Query message, the group address field is set to zero + // when sending a General Query, and set to the group address being + // queried when sending a Group-Specific Query. + // + // As per RFC 2710 section 3.6 (for MLDv1), + // + // In a Query message, the Multicast Address field is set to zero when + // sending a General Query, and set to a specific IPv6 multicast address + // when sending a Multicast-Address-Specific Query. + if groupAddress.Unspecified() { + // This is a general query as the group address is unspecified. + for groupAddress, info := range g.mu.memberships { + g.setDelayTimerForAddressRLocked(groupAddress, &info, maxResponseTime) + g.mu.memberships[groupAddress] = info + } + } else if info, ok := g.mu.memberships[groupAddress]; ok { + g.setDelayTimerForAddressRLocked(groupAddress, &info, maxResponseTime) + g.mu.memberships[groupAddress] = info + } +} + +// HandleReport handles a report message. +// +// If the report is for a joined group, any active delayed report will be +// cancelled and the host state for the group transitions to idle. +func (g *GenericMulticastProtocolState) HandleReport(groupAddress tcpip.Address) { + if !g.opts.Enabled { + return + } + + g.mu.Lock() + defer g.mu.Unlock() + + // As per RFC 2236 section 3 pages 3-4 (for IGMPv2), + // + // If the host receives another host's Report (version 1 or 2) while it has + // a timer running, it stops its timer for the specified group and does not + // send a Report + // + // As per RFC 2710 section 4 page 6 (for MLDv1), + // + // If a node receives another node's Report from an interface for a + // multicast address while it has a timer running for that same address + // on that interface, it stops its timer and does not send a Report for + // that address, thus suppressing duplicate reports on the link. + if info, ok := g.mu.memberships[groupAddress]; ok && info.state == delayingMember { + info.delayedReportJob.Cancel() + info.lastToSendReport = false + info.state = idleMember + g.mu.memberships[groupAddress] = info + } +} + +// initializeNewMemberLocked initializes a new group membership. +// +// Precondition: g.mu must be locked. +func (g *GenericMulticastProtocolState) initializeNewMemberLocked(groupAddress tcpip.Address, info *multicastGroupState) { + if info.state != nonMember { + panic(fmt.Sprintf("state for group %s is not non-member; state = %d", groupAddress, info.state)) + } + + info.state = idleMember + + if groupAddress == g.opts.AllNodesAddress { + // As per RFC 2236 section 6 page 10 (for IGMPv2), + // + // The all-systems group (address 224.0.0.1) is handled as a special + // case. The host starts in Idle Member state for that group on every + // interface, never transitions to another state, and never sends a + // report for that group. + // + // As per RFC 2710 section 5 page 10 (for MLDv1), + // + // The link-scope all-nodes address (FF02::1) is handled as a special + // case. The node starts in Idle Listener state for that address on + // every interface, never transitions to another state, and never sends + // a Report or Done for that address. + return + } + + // As per RFC 2236 section 3 page 5 (for IGMPv2), + // + // When a host joins a multicast group, it should immediately transmit an + // unsolicited Version 2 Membership Report for that group" ... "it is + // recommended that it be repeated". + // + // As per RFC 2710 section 4 page 6 (for MLDv1), + // + // When a node starts listening to a multicast address on an interface, + // it should immediately transmit an unsolicited Report for that address + // on that interface, in case it is the first listener on the link. To + // cover the possibility of the initial Report being lost or damaged, it + // is recommended that it be repeated once or twice after short delays + // [Unsolicited Report Interval]. + // + // TODO(gvisor.dev/issue/4901): Support a configurable number of initial + // unsolicited reports. + info.lastToSendReport = g.opts.Protocol.SendReport(groupAddress) == nil + g.setDelayTimerForAddressRLocked(groupAddress, info, g.opts.MaxUnsolicitedReportDelay) +} + +// maybeSendLeave attempts to send a leave message. +func (g *GenericMulticastProtocolState) maybeSendLeave(groupAddress tcpip.Address, lastToSendReport bool) { + if !g.opts.Enabled || !lastToSendReport { + return + } + + if groupAddress == g.opts.AllNodesAddress { + // As per RFC 2236 section 6 page 10 (for IGMPv2), + // + // The all-systems group (address 224.0.0.1) is handled as a special + // case. The host starts in Idle Member state for that group on every + // interface, never transitions to another state, and never sends a + // report for that group. + // + // As per RFC 2710 section 5 page 10 (for MLDv1), + // + // The link-scope all-nodes address (FF02::1) is handled as a special + // case. The node starts in Idle Listener state for that address on + // every interface, never transitions to another state, and never sends + // a Report or Done for that address. + return + } + + // Okay to ignore the error here as if packet write failed, the multicast + // routers will eventually drop our membership anyways. If the interface is + // being disabled or removed, the generic multicast protocol's should be + // cleared eventually. + // + // As per RFC 2236 section 3 page 5 (for IGMPv2), + // + // When a router receives a Report, it adds the group being reported to + // the list of multicast group memberships on the network on which it + // received the Report and sets the timer for the membership to the + // [Group Membership Interval]. Repeated Reports refresh the timer. If + // no Reports are received for a particular group before this timer has + // expired, the router assumes that the group has no local members and + // that it need not forward remotely-originated multicasts for that + // group onto the attached network. + // + // As per RFC 2710 section 4 page 5 (for MLDv1), + // + // When a router receives a Report from a link, if the reported address + // is not already present in the router's list of multicast address + // having listeners on that link, the reported address is added to the + // list, its timer is set to [Multicast Listener Interval], and its + // appearance is made known to the router's multicast routing component. + // If a Report is received for a multicast address that is already + // present in the router's list, the timer for that address is reset to + // [Multicast Listener Interval]. If an address's timer expires, it is + // assumed that there are no longer any listeners for that address + // present on the link, so it is deleted from the list and its + // disappearance is made known to the multicast routing component. + // + // The requirement to send a leave message is also optional (it MAY be + // skipped): + // + // As per RFC 2236 section 6 page 8 (for IGMPv2), + // + // "send leave" for the group on the interface. If the interface + // state says the Querier is running IGMPv1, this action SHOULD be + // skipped. If the flag saying we were the last host to report is + // cleared, this action MAY be skipped. The Leave Message is sent to + // the ALL-ROUTERS group (224.0.0.2). + // + // As per RFC 2710 section 5 page 8 (for MLDv1), + // + // "send done" for the address on the interface. If the flag saying + // we were the last node to report is cleared, this action MAY be + // skipped. The Done message is sent to the link-scope all-routers + // address (FF02::2). + _ = g.opts.Protocol.SendLeave(groupAddress) +} + +// transitionToNonMemberLocked transitions the given multicast group the the +// non-member/listener state. +// +// Precondition: e.mu must be locked. +func (g *GenericMulticastProtocolState) transitionToNonMemberLocked(groupAddress tcpip.Address, info *multicastGroupState) { + if info.state == nonMember { + return + } + + info.delayedReportJob.Cancel() + g.maybeSendLeave(groupAddress, info.lastToSendReport) + info.lastToSendReport = false + info.state = nonMember +} + +// setDelayTimerForAddressRLocked sets timer to send a delay report. +// +// Precondition: g.mu MUST be read locked. +func (g *GenericMulticastProtocolState) setDelayTimerForAddressRLocked(groupAddress tcpip.Address, info *multicastGroupState, maxResponseTime time.Duration) { + if info.state == nonMember { + return + } + + if groupAddress == g.opts.AllNodesAddress { + // As per RFC 2236 section 6 page 10 (for IGMPv2), + // + // The all-systems group (address 224.0.0.1) is handled as a special + // case. The host starts in Idle Member state for that group on every + // interface, never transitions to another state, and never sends a + // report for that group. + // + // As per RFC 2710 section 5 page 10 (for MLDv1), + // + // The link-scope all-nodes address (FF02::1) is handled as a special + // case. The node starts in Idle Listener state for that address on + // every interface, never transitions to another state, and never sends + // a Report or Done for that address. + return + } + + // As per RFC 2236 section 3 page 3 (for IGMPv2), + // + // If a timer for the group is already unning, it is reset to the random + // value only if the requested Max Response Time is less than the remaining + // value of the running timer. + // + // As per RFC 2710 section 4 page 5 (for MLDv1), + // + // If a timer for any address is already running, it is reset to the new + // random value only if the requested Maximum Response Delay is less than + // the remaining value of the running timer. + if info.state == delayingMember { + // TODO: Reset the timer if time remaining is greater than maxResponseTime. + return + } + info.state = delayingMember + info.delayedReportJob.Cancel() + info.delayedReportJob.Schedule(g.calculateDelayTimerDuration(maxResponseTime)) +} + +// calculateDelayTimerDuration returns a random time between (0, maxRespTime]. +func (g *GenericMulticastProtocolState) calculateDelayTimerDuration(maxRespTime time.Duration) time.Duration { + // As per RFC 2236 section 3 page 3 (for IGMPv2), + // + // When a host receives a Group-Specific Query, it sets a delay timer to a + // random value selected from the range (0, Max Response Time]... + // + // As per RFC 2710 section 4 page 6 (for MLDv1), + // + // When a node receives a Multicast-Address-Specific Query, if it is + // listening to the queried Multicast Address on the interface from + // which the Query was received, it sets a delay timer for that address + // to a random value selected from the range [0, Maximum Response Delay], + // as above. + if maxRespTime == 0 { + return 0 + } + return time.Duration(g.opts.Rand.Int63n(int64(maxRespTime))) +} diff --git a/pkg/tcpip/network/ip/generic_multicast_protocol_test.go b/pkg/tcpip/network/ip/generic_multicast_protocol_test.go new file mode 100644 index 000000000..670be30d4 --- /dev/null +++ b/pkg/tcpip/network/ip/generic_multicast_protocol_test.go @@ -0,0 +1,576 @@ +// 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 ip_test + +import ( + "math/rand" + "testing" + "time" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/faketime" + "gvisor.dev/gvisor/pkg/tcpip/network/ip" +) + +const ( + addr1 = tcpip.Address("\x01") + addr2 = tcpip.Address("\x02") + addr3 = tcpip.Address("\x03") + addr4 = tcpip.Address("\x04") + + maxUnsolicitedReportDelay = time.Second +) + +var _ ip.MulticastGroupProtocol = (*mockMulticastGroupProtocol)(nil) + +type mockMulticastGroupProtocol struct { + sendReportGroupAddrCount map[tcpip.Address]int + sendLeaveGroupAddrCount map[tcpip.Address]int +} + +func (m *mockMulticastGroupProtocol) init() { + m.sendReportGroupAddrCount = make(map[tcpip.Address]int) + m.sendLeaveGroupAddrCount = make(map[tcpip.Address]int) +} + +func (m *mockMulticastGroupProtocol) SendReport(groupAddress tcpip.Address) *tcpip.Error { + m.sendReportGroupAddrCount[groupAddress]++ + return nil +} + +func (m *mockMulticastGroupProtocol) SendLeave(groupAddress tcpip.Address) *tcpip.Error { + m.sendLeaveGroupAddrCount[groupAddress]++ + return nil +} + +func checkProtocol(mgp *mockMulticastGroupProtocol, sendReportGroupAddresses []tcpip.Address, sendLeaveGroupAddresses []tcpip.Address) string { + sendReportGroupAddressesMap := make(map[tcpip.Address]int) + for _, a := range sendReportGroupAddresses { + sendReportGroupAddressesMap[a] = 1 + } + + sendLeaveGroupAddressesMap := make(map[tcpip.Address]int) + for _, a := range sendLeaveGroupAddresses { + sendLeaveGroupAddressesMap[a] = 1 + } + + diff := cmp.Diff(mockMulticastGroupProtocol{ + sendReportGroupAddrCount: sendReportGroupAddressesMap, + sendLeaveGroupAddrCount: sendLeaveGroupAddressesMap, + }, *mgp, cmp.AllowUnexported(mockMulticastGroupProtocol{})) + mgp.init() + return diff +} + +func TestJoinGroup(t *testing.T) { + tests := []struct { + name string + addr tcpip.Address + shouldSendReports bool + }{ + { + name: "Normal group", + addr: addr1, + shouldSendReports: true, + }, + { + name: "All-nodes group", + addr: addr2, + shouldSendReports: false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + var g ip.GenericMulticastProtocolState + var mgp mockMulticastGroupProtocol + mgp.init() + clock := faketime.NewManualClock() + g.Init(ip.GenericMulticastProtocolOptions{ + Enabled: true, + Rand: rand.New(rand.NewSource(0)), + Clock: clock, + Protocol: &mgp, + MaxUnsolicitedReportDelay: maxUnsolicitedReportDelay, + AllNodesAddress: addr2, + }) + + // Joining a group should send a report immediately and another after + // a random interval between 0 and the maximum unsolicited report delay. + g.JoinGroup(test.addr, false /* dontInitialize */) + if test.shouldSendReports { + if diff := checkProtocol(&mgp, []tcpip.Address{test.addr} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + clock.Advance(maxUnsolicitedReportDelay) + if diff := checkProtocol(&mgp, []tcpip.Address{test.addr} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + } + + // Should have no more messages to send. + clock.Advance(time.Hour) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + }) + } +} + +func TestLeaveGroup(t *testing.T) { + tests := []struct { + name string + addr tcpip.Address + shouldSendMessages bool + }{ + { + name: "Normal group", + addr: addr1, + shouldSendMessages: true, + }, + { + name: "All-nodes group", + addr: addr2, + shouldSendMessages: false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + var g ip.GenericMulticastProtocolState + var mgp mockMulticastGroupProtocol + mgp.init() + clock := faketime.NewManualClock() + g.Init(ip.GenericMulticastProtocolOptions{ + Enabled: true, + Rand: rand.New(rand.NewSource(1)), + Clock: clock, + Protocol: &mgp, + MaxUnsolicitedReportDelay: maxUnsolicitedReportDelay, + AllNodesAddress: addr2, + }) + + g.JoinGroup(test.addr, false /* dontInitialize */) + if test.shouldSendMessages { + if diff := checkProtocol(&mgp, []tcpip.Address{test.addr} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + } + + // Leaving a group should send a leave report immediately and cancel any + // delayed reports. + if !g.LeaveGroup(test.addr) { + t.Fatalf("got g.LeaveGroup(%s) = false, want = true", test.addr) + } + if test.shouldSendMessages { + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, []tcpip.Address{test.addr} /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + } + + // Should have no more messages to send. + clock.Advance(time.Hour) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + }) + } +} + +func TestHandleReport(t *testing.T) { + tests := []struct { + name string + reportAddr tcpip.Address + expectReportsFor []tcpip.Address + }{ + { + name: "Unpecified empty", + reportAddr: "", + expectReportsFor: []tcpip.Address{addr1, addr2}, + }, + { + name: "Unpecified any", + reportAddr: "\x00", + expectReportsFor: []tcpip.Address{addr1, addr2}, + }, + { + name: "Specified", + reportAddr: addr1, + expectReportsFor: []tcpip.Address{addr2}, + }, + { + name: "Specified all-nodes", + reportAddr: addr3, + expectReportsFor: []tcpip.Address{addr1, addr2}, + }, + { + name: "Specified other", + reportAddr: addr4, + expectReportsFor: []tcpip.Address{addr1, addr2}, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + var g ip.GenericMulticastProtocolState + var mgp mockMulticastGroupProtocol + mgp.init() + clock := faketime.NewManualClock() + g.Init(ip.GenericMulticastProtocolOptions{ + Enabled: true, + Rand: rand.New(rand.NewSource(2)), + Clock: clock, + Protocol: &mgp, + MaxUnsolicitedReportDelay: maxUnsolicitedReportDelay, + AllNodesAddress: addr3, + }) + + g.JoinGroup(addr1, false /* dontInitialize */) + if diff := checkProtocol(&mgp, []tcpip.Address{addr1} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + g.JoinGroup(addr2, false /* dontInitialize */) + if diff := checkProtocol(&mgp, []tcpip.Address{addr2} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + g.JoinGroup(addr3, false /* dontInitialize */) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + // Receiving a report for a group we have a timer scheduled for should + // cancel our delayed report timer for the group. + g.HandleReport(test.reportAddr) + if len(test.expectReportsFor) != 0 { + clock.Advance(maxUnsolicitedReportDelay) + if diff := checkProtocol(&mgp, test.expectReportsFor /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + } + + // Should have no more messages to send. + clock.Advance(time.Hour) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + }) + } +} + +func TestHandleQuery(t *testing.T) { + tests := []struct { + name string + queryAddr tcpip.Address + maxDelay time.Duration + expectReportsFor []tcpip.Address + }{ + { + name: "Unpecified empty", + queryAddr: "", + maxDelay: 0, + expectReportsFor: []tcpip.Address{addr1, addr2}, + }, + { + name: "Unpecified any", + queryAddr: "\x00", + maxDelay: 1, + expectReportsFor: []tcpip.Address{addr1, addr2}, + }, + { + name: "Specified", + queryAddr: addr1, + maxDelay: 2, + expectReportsFor: []tcpip.Address{addr1}, + }, + { + name: "Specified all-nodes", + queryAddr: addr3, + maxDelay: 3, + expectReportsFor: nil, + }, + { + name: "Specified other", + queryAddr: addr4, + maxDelay: 4, + expectReportsFor: nil, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + var g ip.GenericMulticastProtocolState + var mgp mockMulticastGroupProtocol + mgp.init() + clock := faketime.NewManualClock() + g.Init(ip.GenericMulticastProtocolOptions{ + Enabled: true, + Rand: rand.New(rand.NewSource(3)), + Clock: clock, + Protocol: &mgp, + MaxUnsolicitedReportDelay: maxUnsolicitedReportDelay, + AllNodesAddress: addr3, + }) + + g.JoinGroup(addr1, false /* dontInitialize */) + if diff := checkProtocol(&mgp, []tcpip.Address{addr1} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + g.JoinGroup(addr2, false /* dontInitialize */) + if diff := checkProtocol(&mgp, []tcpip.Address{addr2} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + g.JoinGroup(addr3, false /* dontInitialize */) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + clock.Advance(maxUnsolicitedReportDelay) + if diff := checkProtocol(&mgp, []tcpip.Address{addr1, addr2} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + // Receiving a query should make us schedule a new delayed report if it + // is a query directed at us or a general query. + g.HandleQuery(test.queryAddr, test.maxDelay) + if len(test.expectReportsFor) != 0 { + clock.Advance(test.maxDelay) + if diff := checkProtocol(&mgp, test.expectReportsFor /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + } + + // Should have no more messages to send. + clock.Advance(time.Hour) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + }) + } +} + +func TestJoinCount(t *testing.T) { + var g ip.GenericMulticastProtocolState + var mgp mockMulticastGroupProtocol + mgp.init() + clock := faketime.NewManualClock() + g.Init(ip.GenericMulticastProtocolOptions{ + Enabled: true, + Rand: rand.New(rand.NewSource(4)), + Clock: clock, + Protocol: &mgp, + MaxUnsolicitedReportDelay: time.Second, + }) + + // Set the join count to 2 for a group. + g.JoinGroup(addr1, false /* dontInitialize */) + if !g.IsLocallyJoined(addr1) { + t.Fatalf("got g.IsLocallyJoined(%s) = false, want = true", addr1) + } + // Only the first join should trigger a report to be sent. + if diff := checkProtocol(&mgp, []tcpip.Address{addr1} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + g.JoinGroup(addr1, false /* dontInitialize */) + if !g.IsLocallyJoined(addr1) { + t.Fatalf("got g.IsLocallyJoined(%s) = false, want = true", addr1) + } + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + // Group should still be considered joined after leaving once. + if !g.LeaveGroup(addr1) { + t.Fatalf("got g.LeaveGroup(%s) = false, want = true", addr1) + } + if !g.IsLocallyJoined(addr1) { + t.Fatalf("got g.IsLocallyJoined(%s) = false, want = true", addr1) + } + // A leave report should only be sent once the join count reaches 0. + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + // Leaving once more should actually remove us from the group. + if !g.LeaveGroup(addr1) { + t.Fatalf("got g.LeaveGroup(%s) = false, want = true", addr1) + } + if g.IsLocallyJoined(addr1) { + t.Fatalf("got g.IsLocallyJoined(%s) = true, want = false", addr1) + } + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, []tcpip.Address{addr1} /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + // Group should no longer be joined so we should not have anything to + // leave. + if g.LeaveGroup(addr1) { + t.Fatalf("got g.LeaveGroup(%s) = true, want = false", addr1) + } + if g.IsLocallyJoined(addr1) { + t.Fatalf("got g.IsLocallyJoined(%s) = true, want = false", addr1) + } + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + // Should have no more messages to send. + clock.Advance(time.Hour) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } +} + +func TestMakeAllNonMemberAndInitialize(t *testing.T) { + var g ip.GenericMulticastProtocolState + var mgp mockMulticastGroupProtocol + mgp.init() + clock := faketime.NewManualClock() + g.Init(ip.GenericMulticastProtocolOptions{ + Enabled: true, + Rand: rand.New(rand.NewSource(3)), + Clock: clock, + Protocol: &mgp, + MaxUnsolicitedReportDelay: maxUnsolicitedReportDelay, + AllNodesAddress: addr3, + }) + + g.JoinGroup(addr1, false /* dontInitialize */) + if diff := checkProtocol(&mgp, []tcpip.Address{addr1} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + g.JoinGroup(addr2, false /* dontInitialize */) + if diff := checkProtocol(&mgp, []tcpip.Address{addr2} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + g.JoinGroup(addr3, false /* dontInitialize */) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + // Should send the leave reports for each but still consider them locally + // joined. + g.MakeAllNonMember() + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, []tcpip.Address{addr1, addr2} /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + clock.Advance(time.Hour) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + for _, group := range []tcpip.Address{addr1, addr2, addr3} { + if !g.IsLocallyJoined(group) { + t.Fatalf("got g.IsLocallyJoined(%s) = false, want = true", group) + } + } + + // Should send the initial set of unsolcited reports. + g.InitializeGroups() + if diff := checkProtocol(&mgp, []tcpip.Address{addr1, addr2} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + clock.Advance(maxUnsolicitedReportDelay) + if diff := checkProtocol(&mgp, []tcpip.Address{addr1, addr2} /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + // Should have no more messages to send. + clock.Advance(time.Hour) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Errorf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } +} + +// TestGroupStateNonMember tests that groups do not send packets when in the +// non-member state, but are still considered locally joined. +func TestGroupStateNonMember(t *testing.T) { + tests := []struct { + name string + enabled bool + dontInitialize bool + }{ + { + name: "Disabled", + enabled: false, + dontInitialize: false, + }, + { + name: "Keep non-member", + enabled: true, + dontInitialize: true, + }, + { + name: "disabled and Keep non-member", + enabled: false, + dontInitialize: true, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + var g ip.GenericMulticastProtocolState + var mgp mockMulticastGroupProtocol + mgp.init() + clock := faketime.NewManualClock() + g.Init(ip.GenericMulticastProtocolOptions{ + Enabled: test.enabled, + Rand: rand.New(rand.NewSource(3)), + Clock: clock, + Protocol: &mgp, + MaxUnsolicitedReportDelay: maxUnsolicitedReportDelay, + }) + + g.JoinGroup(addr1, test.dontInitialize) + if !g.IsLocallyJoined(addr1) { + t.Fatalf("got g.IsLocallyJoined(%s) = false, want = true", addr1) + } + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + g.JoinGroup(addr2, test.dontInitialize) + if !g.IsLocallyJoined(addr2) { + t.Fatalf("got g.IsLocallyJoined(%s) = false, want = true", addr2) + } + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + g.HandleQuery(addr1, time.Nanosecond) + clock.Advance(time.Nanosecond) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + if !g.LeaveGroup(addr2) { + t.Errorf("got g.LeaveGroup(%s) = false, want = true", addr2) + } + if !g.IsLocallyJoined(addr1) { + t.Fatalf("got g.IsLocallyJoined(%s) = false, want = true", addr1) + } + if g.IsLocallyJoined(addr2) { + t.Fatalf("got g.IsLocallyJoined(%s) = true, want = false", addr2) + } + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + + clock.Advance(time.Hour) + if diff := checkProtocol(&mgp, nil /* sendReportGroupAddresses */, nil /* sendLeaveGroupAddresses */); diff != "" { + t.Fatalf("mockMulticastGroupProtocol mismatch (-want +got):\n%s", diff) + } + }) + } +} diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index d49c44846..a314dd386 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -193,10 +193,6 @@ func (*testObject) WritePackets(_ *stack.Route, _ *stack.GSO, pkt stack.PacketBu panic("not implemented") } -func (*testObject) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { - return tcpip.ErrNotSupported -} - // ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. func (*testObject) ARPHardwareType() header.ARPHardwareType { panic("not implemented") @@ -207,7 +203,7 @@ func (*testObject) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.Net panic("not implemented") } -func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { +func buildIPv4Route(local, remote tcpip.Address) (*stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, tcp.NewProtocol}, @@ -223,7 +219,7 @@ func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { return s.FindRoute(nicID, local, remote, ipv4.ProtocolNumber, false /* multicastLoop */) } -func buildIPv6Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { +func buildIPv6Route(local, remote tcpip.Address) (*stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, tcp.NewProtocol}, @@ -554,7 +550,7 @@ func TestIPv4Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{ + if err := ep.WritePacket(r, nil /* gso */, stack.NetworkHeaderParams{ Protocol: 123, TTL: 123, TOS: stack.DefaultTOS, @@ -937,7 +933,7 @@ func TestIPv6Send(t *testing.T) { if err != nil { t.Fatalf("could not find route: %v", err) } - if err := ep.WritePacket(&r, nil /* gso */, stack.NetworkHeaderParams{ + if err := ep.WritePacket(r, nil /* gso */, stack.NetworkHeaderParams{ Protocol: 123, TTL: 123, TOS: stack.DefaultTOS, @@ -1093,7 +1089,19 @@ func TestWriteHeaderIncludedPacket(t *testing.T) { dataBuf := [dataLen]byte{1, 2, 3, 4} data := dataBuf[:] - ipv4Options := header.IPv4Options{0, 1, 0, 1} + ipv4Options := header.IPv4OptionsSerializer{ + &header.IPv4SerializableListEndOption{}, + &header.IPv4SerializableNOPOption{}, + &header.IPv4SerializableListEndOption{}, + &header.IPv4SerializableNOPOption{}, + } + + expectOptions := header.IPv4Options{ + byte(header.IPv4OptionListEndType), + byte(header.IPv4OptionNOPType), + byte(header.IPv4OptionListEndType), + byte(header.IPv4OptionNOPType), + } ipv6FragmentExtHdrBuf := [header.IPv6FragmentExtHdrLength]byte{transportProto, 0, 62, 4, 1, 2, 3, 4} ipv6FragmentExtHdr := ipv6FragmentExtHdrBuf[:] @@ -1243,7 +1251,7 @@ func TestWriteHeaderIncludedPacket(t *testing.T) { nicAddr: localIPv4Addr, remoteAddr: remoteIPv4Addr, pktGen: func(t *testing.T, src tcpip.Address) buffer.VectorisedView { - ipHdrLen := header.IPv4MinimumSize + ipv4Options.SizeWithPadding() + ipHdrLen := int(header.IPv4MinimumSize + ipv4Options.Length()) totalLen := ipHdrLen + len(data) hdr := buffer.NewPrependable(totalLen) if n := copy(hdr.Prepend(len(data)), data); n != len(data) { @@ -1266,7 +1274,7 @@ func TestWriteHeaderIncludedPacket(t *testing.T) { netHdr := pkt.NetworkHeader() - hdrLen := header.IPv4MinimumSize + len(ipv4Options) + hdrLen := int(header.IPv4MinimumSize + ipv4Options.Length()) if len(netHdr.View()) != hdrLen { t.Errorf("got len(netHdr.View()) = %d, want = %d", len(netHdr.View()), hdrLen) } @@ -1276,7 +1284,7 @@ func TestWriteHeaderIncludedPacket(t *testing.T) { checker.DstAddr(remoteIPv4Addr), checker.IPv4HeaderLength(hdrLen), checker.IPFullLength(uint16(hdrLen+len(data))), - checker.IPv4Options(ipv4Options), + checker.IPv4Options(expectOptions), checker.IPPayload(data), ) }, @@ -1288,7 +1296,7 @@ func TestWriteHeaderIncludedPacket(t *testing.T) { nicAddr: localIPv4Addr, remoteAddr: remoteIPv4Addr, pktGen: func(t *testing.T, src tcpip.Address) buffer.VectorisedView { - ip := header.IPv4(make([]byte, header.IPv4MinimumSize+ipv4Options.SizeWithPadding())) + ip := header.IPv4(make([]byte, header.IPv4MinimumSize+ipv4Options.Length())) ip.Encode(&header.IPv4Fields{ Protocol: transportProto, TTL: ipv4.DefaultTTL, @@ -1307,7 +1315,7 @@ func TestWriteHeaderIncludedPacket(t *testing.T) { netHdr := pkt.NetworkHeader() - hdrLen := header.IPv4MinimumSize + len(ipv4Options) + hdrLen := int(header.IPv4MinimumSize + ipv4Options.Length()) if len(netHdr.View()) != hdrLen { t.Errorf("got len(netHdr.View()) = %d, want = %d", len(netHdr.View()), hdrLen) } @@ -1317,7 +1325,7 @@ func TestWriteHeaderIncludedPacket(t *testing.T) { checker.DstAddr(remoteIPv4Addr), checker.IPv4HeaderLength(hdrLen), checker.IPFullLength(uint16(hdrLen+len(data))), - checker.IPv4Options(ipv4Options), + checker.IPv4Options(expectOptions), checker.IPPayload(data), ) }, diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD index 6252614ec..32f53f217 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -6,6 +6,7 @@ go_library( name = "ipv4", srcs = [ "icmp.go", + "igmp.go", "ipv4.go", ], visibility = ["//visibility:public"], @@ -17,6 +18,7 @@ go_library( "//pkg/tcpip/header/parse", "//pkg/tcpip/network/fragmentation", "//pkg/tcpip/network/hash", + "//pkg/tcpip/network/ip", "//pkg/tcpip/stack", ], ) @@ -24,7 +26,10 @@ go_library( go_test( name = "ipv4_test", size = "small", - srcs = ["ipv4_test.go"], + srcs = [ + "igmp_test.go", + "ipv4_test.go", + ], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index 488945226..8e392f86c 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -63,7 +63,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack func (e *endpoint) handleICMP(pkt *stack.PacketBuffer) { stats := e.protocol.stack.Stats() - received := stats.ICMP.V4PacketsReceived + received := stats.ICMP.V4.PacketsReceived // TODO(gvisor.dev/issue/170): ICMP packets don't have their // TransportHeader fields set. See icmp/protocol.go:protocol.Parse for a // full explanation. @@ -130,7 +130,7 @@ func (e *endpoint) handleICMP(pkt *stack.PacketBuffer) { case header.ICMPv4Echo: received.Echo.Increment() - sent := stats.ICMP.V4PacketsSent + sent := stats.ICMP.V4.PacketsSent if !e.protocol.stack.AllowICMPMessage() { sent.RateLimited.Increment() return @@ -379,7 +379,7 @@ func (p *protocol) returnError(reason icmpReason, pkt *stack.PacketBuffer) *tcpi } defer route.Release() - sent := p.stack.Stats().ICMP.V4PacketsSent + sent := p.stack.Stats().ICMP.V4.PacketsSent if !p.stack.AllowICMPMessage() { sent.RateLimited.Increment() return nil diff --git a/pkg/tcpip/network/ipv4/igmp.go b/pkg/tcpip/network/ipv4/igmp.go new file mode 100644 index 000000000..0134fadc0 --- /dev/null +++ b/pkg/tcpip/network/ipv4/igmp.go @@ -0,0 +1,323 @@ +// 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 ipv4 + +import ( + "fmt" + "sync" + "sync/atomic" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/network/ip" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +const ( + // igmpV1PresentDefault is the initial state for igmpV1Present in the + // igmpState. As per RFC 2236 Page 9 says "No IGMPv1 Router Present ... is + // the initial state." + igmpV1PresentDefault = 0 + + // v1RouterPresentTimeout from RFC 2236 Section 8.11, Page 18 + // See note on igmpState.igmpV1Present for more detail. + v1RouterPresentTimeout = 400 * time.Second + + // v1MaxRespTime from RFC 2236 Section 4, Page 5. "The IGMPv1 router + // will send General Queries with the Max Response Time set to 0. This MUST + // be interpreted as a value of 100 (10 seconds)." + // + // Note that the Max Response Time field is a value in units of deciseconds. + v1MaxRespTime = 10 * time.Second + + // UnsolicitedReportIntervalMax is the maximum delay between sending + // unsolicited IGMP reports. + // + // Obtained from RFC 2236 Section 8.10, Page 19. + UnsolicitedReportIntervalMax = 10 * time.Second +) + +// IGMPOptions holds options for IGMP. +type IGMPOptions struct { + // Enabled indicates whether IGMP will be performed. + // + // When enabled, IGMP may transmit IGMP report and leave messages when + // joining and leaving multicast groups respectively, and handle incoming + // IGMP packets. + Enabled bool +} + +var _ ip.MulticastGroupProtocol = (*igmpState)(nil) + +// igmpState is the per-interface IGMP state. +// +// igmpState.init() MUST be called after creating an IGMP state. +type igmpState struct { + // The IPv4 endpoint this igmpState is for. + ep *endpoint + opts IGMPOptions + + // igmpV1Present is for maintaining compatibility with IGMPv1 Routers, from + // RFC 2236 Section 4 Page 6: "The IGMPv1 router expects Version 1 + // Membership Reports in response to its Queries, and will not pay + // attention to Version 2 Membership Reports. Therefore, a state variable + // MUST be kept for each interface, describing whether the multicast + // Querier on that interface is running IGMPv1 or IGMPv2. This variable + // MUST be based upon whether or not an IGMPv1 query was heard in the last + // [Version 1 Router Present Timeout] seconds". + // + // Must be accessed with atomic operations. Holds a value of 1 when true, 0 + // when false. + igmpV1Present uint32 + + mu struct { + sync.RWMutex + + genericMulticastProtocol ip.GenericMulticastProtocolState + + // igmpV1Job is scheduled when this interface receives an IGMPv1 style + // message, upon expiration the igmpV1Present flag is cleared. + // igmpV1Job may not be nil once igmpState is initialized. + igmpV1Job *tcpip.Job + } +} + +// SendReport implements ip.MulticastGroupProtocol. +func (igmp *igmpState) SendReport(groupAddress tcpip.Address) *tcpip.Error { + igmpType := header.IGMPv2MembershipReport + if igmp.v1Present() { + igmpType = header.IGMPv1MembershipReport + } + return igmp.writePacket(groupAddress, groupAddress, igmpType) +} + +// SendLeave implements ip.MulticastGroupProtocol. +func (igmp *igmpState) SendLeave(groupAddress tcpip.Address) *tcpip.Error { + // As per RFC 2236 Section 6, Page 8: "If the interface state says the + // Querier is running IGMPv1, this action SHOULD be skipped. If the flag + // saying we were the last host to report is cleared, this action MAY be + // skipped." + if igmp.v1Present() { + return nil + } + return igmp.writePacket(header.IPv4AllRoutersGroup, groupAddress, header.IGMPLeaveGroup) +} + +// init sets up an igmpState struct, and is required to be called before using +// a new igmpState. +func (igmp *igmpState) init(ep *endpoint, opts IGMPOptions) { + igmp.mu.Lock() + defer igmp.mu.Unlock() + igmp.ep = ep + igmp.opts = opts + igmp.mu.genericMulticastProtocol.Init(ip.GenericMulticastProtocolOptions{ + Enabled: opts.Enabled, + Rand: ep.protocol.stack.Rand(), + Clock: ep.protocol.stack.Clock(), + Protocol: igmp, + MaxUnsolicitedReportDelay: UnsolicitedReportIntervalMax, + AllNodesAddress: header.IPv4AllSystems, + }) + igmp.igmpV1Present = igmpV1PresentDefault + igmp.mu.igmpV1Job = igmp.ep.protocol.stack.NewJob(&igmp.mu, func() { + igmp.setV1Present(false) + }) +} + +func (igmp *igmpState) handleIGMP(pkt *stack.PacketBuffer) { + stats := igmp.ep.protocol.stack.Stats() + received := stats.IGMP.PacketsReceived + headerView, ok := pkt.Data.PullUp(header.IGMPMinimumSize) + if !ok { + received.Invalid.Increment() + return + } + h := header.IGMP(headerView) + + // Temporarily reset the checksum field to 0 in order to calculate the proper + // checksum. + wantChecksum := h.Checksum() + h.SetChecksum(0) + gotChecksum := ^header.ChecksumVV(pkt.Data, 0 /* initial */) + h.SetChecksum(wantChecksum) + + if gotChecksum != wantChecksum { + received.ChecksumErrors.Increment() + return + } + + switch h.Type() { + case header.IGMPMembershipQuery: + received.MembershipQuery.Increment() + if len(headerView) < header.IGMPQueryMinimumSize { + received.Invalid.Increment() + return + } + igmp.handleMembershipQuery(h.GroupAddress(), h.MaxRespTime()) + case header.IGMPv1MembershipReport: + received.V1MembershipReport.Increment() + if len(headerView) < header.IGMPReportMinimumSize { + received.Invalid.Increment() + return + } + igmp.handleMembershipReport(h.GroupAddress()) + case header.IGMPv2MembershipReport: + received.V2MembershipReport.Increment() + if len(headerView) < header.IGMPReportMinimumSize { + received.Invalid.Increment() + return + } + igmp.handleMembershipReport(h.GroupAddress()) + case header.IGMPLeaveGroup: + received.LeaveGroup.Increment() + // As per RFC 2236 Section 6, Page 7: "IGMP messages other than Query or + // Report, are ignored in all states" + + default: + // As per RFC 2236 Section 2.1 Page 3: "Unrecognized message types should + // be silently ignored. New message types may be used by newer versions of + // IGMP, by multicast routing protocols, or other uses." + received.Unrecognized.Increment() + } +} + +func (igmp *igmpState) v1Present() bool { + return atomic.LoadUint32(&igmp.igmpV1Present) == 1 +} + +func (igmp *igmpState) setV1Present(v bool) { + if v { + atomic.StoreUint32(&igmp.igmpV1Present, 1) + } else { + atomic.StoreUint32(&igmp.igmpV1Present, 0) + } +} + +func (igmp *igmpState) handleMembershipQuery(groupAddress tcpip.Address, maxRespTime time.Duration) { + igmp.mu.Lock() + defer igmp.mu.Unlock() + + // As per RFC 2236 Section 6, Page 10: If the maximum response time is zero + // then change the state to note that an IGMPv1 router is present and + // schedule the query received Job. + if maxRespTime == 0 && igmp.opts.Enabled { + igmp.mu.igmpV1Job.Cancel() + igmp.mu.igmpV1Job.Schedule(v1RouterPresentTimeout) + igmp.setV1Present(true) + maxRespTime = v1MaxRespTime + } + + igmp.mu.genericMulticastProtocol.HandleQuery(groupAddress, maxRespTime) +} + +func (igmp *igmpState) handleMembershipReport(groupAddress tcpip.Address) { + igmp.mu.Lock() + defer igmp.mu.Unlock() + igmp.mu.genericMulticastProtocol.HandleReport(groupAddress) +} + +// writePacket assembles and sends an IGMP packet with the provided fields, +// incrementing the provided stat counter on success. +func (igmp *igmpState) writePacket(destAddress tcpip.Address, groupAddress tcpip.Address, igmpType header.IGMPType) *tcpip.Error { + igmpData := header.IGMP(buffer.NewView(header.IGMPReportMinimumSize)) + igmpData.SetType(igmpType) + igmpData.SetGroupAddress(groupAddress) + igmpData.SetChecksum(header.IGMPCalculateChecksum(igmpData)) + + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(igmp.ep.MaxHeaderLength()), + Data: buffer.View(igmpData).ToVectorisedView(), + }) + + // TODO(gvisor.dev/issue/4888): We should not use the unspecified address, + // rather we should select an appropriate local address. + localAddr := header.IPv4Any + igmp.ep.addIPHeader(localAddr, destAddress, pkt, stack.NetworkHeaderParams{ + Protocol: header.IGMPProtocolNumber, + TTL: header.IGMPTTL, + TOS: stack.DefaultTOS, + }, header.IPv4OptionsSerializer{ + &header.IPv4SerializableRouterAlertOption{}, + }) + + sent := igmp.ep.protocol.stack.Stats().IGMP.PacketsSent + if err := igmp.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv4Address(destAddress), nil /* gso */, ProtocolNumber, pkt); err != nil { + sent.Dropped.Increment() + return err + } + switch igmpType { + case header.IGMPv1MembershipReport: + sent.V1MembershipReport.Increment() + case header.IGMPv2MembershipReport: + sent.V2MembershipReport.Increment() + case header.IGMPLeaveGroup: + sent.LeaveGroup.Increment() + default: + panic(fmt.Sprintf("unrecognized igmp type = %d", igmpType)) + } + return nil +} + +// joinGroup handles adding a new group to the membership map, setting up the +// IGMP state for the group, and sending and scheduling the required +// messages. +// +// If the group already exists in the membership map, returns +// tcpip.ErrDuplicateAddress. +func (igmp *igmpState) joinGroup(groupAddress tcpip.Address) { + igmp.mu.Lock() + defer igmp.mu.Unlock() + igmp.mu.genericMulticastProtocol.JoinGroup(groupAddress, !igmp.ep.Enabled() /* dontInitialize */) +} + +// isInGroup returns true if the specified group has been joined locally. +func (igmp *igmpState) isInGroup(groupAddress tcpip.Address) bool { + igmp.mu.Lock() + defer igmp.mu.Unlock() + return igmp.mu.genericMulticastProtocol.IsLocallyJoined(groupAddress) +} + +// leaveGroup handles removing the group from the membership map, cancels any +// delay timers associated with that group, and sends the Leave Group message +// if required. +func (igmp *igmpState) leaveGroup(groupAddress tcpip.Address) *tcpip.Error { + igmp.mu.Lock() + defer igmp.mu.Unlock() + + // LeaveGroup returns false only if the group was not joined. + if igmp.mu.genericMulticastProtocol.LeaveGroup(groupAddress) { + return nil + } + + return tcpip.ErrBadLocalAddress +} + +// softLeaveAll leaves all groups from the perspective of IGMP, but remains +// joined locally. +func (igmp *igmpState) softLeaveAll() { + igmp.mu.Lock() + defer igmp.mu.Unlock() + igmp.mu.genericMulticastProtocol.MakeAllNonMember() +} + +// initializeAll attemps to initialize the IGMP state for each group that has +// been joined locally. +func (igmp *igmpState) initializeAll() { + igmp.mu.Lock() + defer igmp.mu.Unlock() + igmp.mu.genericMulticastProtocol.InitializeGroups() +} diff --git a/pkg/tcpip/network/ipv4/igmp_test.go b/pkg/tcpip/network/ipv4/igmp_test.go new file mode 100644 index 000000000..5e139377b --- /dev/null +++ b/pkg/tcpip/network/ipv4/igmp_test.go @@ -0,0 +1,156 @@ +// 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 ipv4_test + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/checker" + "gvisor.dev/gvisor/pkg/tcpip/faketime" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +const ( + linkAddr = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") + multicastAddr = tcpip.Address("\xe0\x00\x00\x03") + nicID = 1 +) + +// validateIgmpPacket checks that a passed PacketInfo is an IPv4 IGMP packet +// sent to the provided address with the passed fields set. Raises a t.Error if +// any field does not match. +func validateIgmpPacket(t *testing.T, p channel.PacketInfo, remoteAddress tcpip.Address, igmpType header.IGMPType, maxRespTime byte, groupAddress tcpip.Address) { + t.Helper() + + payload := header.IPv4(stack.PayloadSince(p.Pkt.NetworkHeader())) + checker.IPv4(t, payload, + checker.DstAddr(remoteAddress), + // TTL for an IGMP message must be 1 as per RFC 2236 section 2. + checker.TTL(1), + checker.IPv4RouterAlert(), + checker.IGMP( + checker.IGMPType(igmpType), + checker.IGMPMaxRespTime(header.DecisecondToDuration(maxRespTime)), + checker.IGMPGroupAddress(groupAddress), + ), + ) +} + +func createStack(t *testing.T, igmpEnabled bool) (*channel.Endpoint, *stack.Stack, *faketime.ManualClock) { + t.Helper() + + // Create an endpoint of queue size 1, since no more than 1 packets are ever + // queued in the tests in this file. + e := channel.New(1, 1280, linkAddr) + clock := faketime.NewManualClock() + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocolWithOptions(ipv4.Options{ + IGMP: ipv4.IGMPOptions{ + Enabled: igmpEnabled, + }, + })}, + Clock: clock, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + + return e, s, clock +} + +func createAndInjectIGMPPacket(e *channel.Endpoint, igmpType header.IGMPType, maxRespTime byte, groupAddress tcpip.Address) { + buf := buffer.NewView(header.IPv4MinimumSize + header.IGMPQueryMinimumSize) + + ip := header.IPv4(buf) + ip.Encode(&header.IPv4Fields{ + TotalLength: uint16(len(buf)), + TTL: 1, + Protocol: uint8(header.IGMPProtocolNumber), + SrcAddr: header.IPv4Any, + DstAddr: header.IPv4AllSystems, + }) + ip.SetChecksum(^ip.CalculateChecksum()) + + igmp := header.IGMP(buf[header.IPv4MinimumSize:]) + igmp.SetType(igmpType) + igmp.SetMaxRespTime(maxRespTime) + igmp.SetGroupAddress(groupAddress) + igmp.SetChecksum(header.IGMPCalculateChecksum(igmp)) + + e.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) +} + +// TestIgmpV1Present tests the handling of the case where an IGMPv1 router is +// present on the network. The IGMP stack will then send IGMPv1 Membership +// reports for backwards compatibility. +func TestIgmpV1Present(t *testing.T) { + e, s, clock := createStack(t, true) + + if err := s.JoinGroup(ipv4.ProtocolNumber, nicID, multicastAddr); err != nil { + t.Fatalf("JoinGroup(ipv4, nic, %s) = %s", multicastAddr, err) + } + + // This NIC will send an IGMPv2 report immediately, before this test can get + // the IGMPv1 General Membership Query in. + p, ok := e.Read() + if !ok { + t.Fatal("unable to Read IGMP packet, expected V2MembershipReport") + } + if got := s.Stats().IGMP.PacketsSent.V2MembershipReport.Value(); got != 1 { + t.Fatalf("got V2MembershipReport messages sent = %d, want = 1", got) + } + validateIgmpPacket(t, p, multicastAddr, header.IGMPv2MembershipReport, 0, multicastAddr) + if t.Failed() { + t.FailNow() + } + + // Inject an IGMPv1 General Membership Query which is identical to a standard + // membership query except the Max Response Time is set to 0, which will tell + // the stack that this is a router using IGMPv1. Send it to the all systems + // group which is the only group this host belongs to. + createAndInjectIGMPPacket(e, header.IGMPMembershipQuery, 0, header.IPv4AllSystems) + if got := s.Stats().IGMP.PacketsReceived.MembershipQuery.Value(); got != 1 { + t.Fatalf("got Membership Queries received = %d, want = 1", got) + } + + // Before advancing the clock, verify that this host has not sent a + // V1MembershipReport yet. + if got := s.Stats().IGMP.PacketsSent.V1MembershipReport.Value(); got != 0 { + t.Fatalf("got V1MembershipReport messages sent = %d, want = 0", got) + } + + // Verify the solicited Membership Report is sent. Now that this NIC has seen + // an IGMPv1 query, it should send an IGMPv1 Membership Report. + p, ok = e.Read() + if ok { + t.Fatalf("sent unexpected packet, expected V1MembershipReport only after advancing the clock = %+v", p.Pkt) + } + clock.Advance(ipv4.UnsolicitedReportIntervalMax) + p, ok = e.Read() + if !ok { + t.Fatal("unable to Read IGMP packet, expected V1MembershipReport") + } + if got := s.Stats().IGMP.PacketsSent.V1MembershipReport.Value(); got != 1 { + t.Fatalf("got V1MembershipReport messages sent = %d, want = 1", got) + } + validateIgmpPacket(t, p, multicastAddr, header.IGMPv1MembershipReport, 0, multicastAddr) +} diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 1efe6297a..3076185cd 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -72,6 +72,7 @@ type endpoint struct { nic stack.NetworkInterface dispatcher stack.TransportDispatcher protocol *protocol + igmp igmpState // enabled is set to 1 when the enpoint is enabled and 0 when it is // disabled. @@ -94,6 +95,7 @@ func (p *protocol) NewEndpoint(nic stack.NetworkInterface, _ stack.LinkAddressCa protocol: p, } e.mu.addressableEndpointState.Init(e) + e.igmp.init(e, p.options.IGMP) return e } @@ -121,11 +123,22 @@ func (e *endpoint) Enable() *tcpip.Error { // We have no need for the address endpoint. ep.DecRef() + // Groups may have been joined while the endpoint was disabled, or the + // endpoint may have left groups from the perspective of IGMP when the + // endpoint was disabled. Either way, we need to let routers know to + // send us multicast traffic. + e.igmp.initializeAll() + // As per RFC 1122 section 3.3.7, all hosts should join the all-hosts // multicast group. Note, the IANA calls the all-hosts multicast group the // all-systems multicast group. - _, err = e.mu.addressableEndpointState.JoinGroup(header.IPv4AllSystems) - return err + if err := e.joinGroupLocked(header.IPv4AllSystems); err != nil { + // joinGroupLocked only returns an error if the group address is not a valid + // IPv4 multicast address. + panic(fmt.Sprintf("e.joinGroupLocked(%s): %s", header.IPv4AllSystems, err)) + } + + return nil } // Enabled implements stack.NetworkEndpoint. @@ -162,10 +175,14 @@ func (e *endpoint) disableLocked() { } // The endpoint may have already left the multicast group. - if _, err := e.mu.addressableEndpointState.LeaveGroup(header.IPv4AllSystems); err != nil && err != tcpip.ErrBadLocalAddress { + if err := e.leaveGroupLocked(header.IPv4AllSystems); err != nil && err != tcpip.ErrBadLocalAddress { panic(fmt.Sprintf("unexpected error when leaving group = %s: %s", header.IPv4AllSystems, err)) } + // Leave groups from the perspective of IGMP so that routers know that + // we are no longer interested in the group. + e.igmp.softLeaveAll() + // The address may have already been removed. if err := e.mu.addressableEndpointState.RemovePermanentAddress(ipv4BroadcastAddr.Address); err != nil && err != tcpip.ErrBadLocalAddress { panic(fmt.Sprintf("unexpected error when removing address = %s: %s", ipv4BroadcastAddr.Address, err)) @@ -198,37 +215,34 @@ func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return e.protocol.Number() } -func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) { +func (e *endpoint) addIPHeader(srcAddr, dstAddr tcpip.Address, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams, options header.IPv4OptionsSerializer) { hdrLen := header.IPv4MinimumSize - var opts header.IPv4Options - if params.Options != nil { - var ok bool - if opts, ok = params.Options.(header.IPv4Options); !ok { - panic(fmt.Sprintf("want IPv4Options, got %T", params.Options)) - } - hdrLen += opts.SizeWithPadding() - if hdrLen > header.IPv4MaximumHeaderSize { - // Since we have no way to report an error we must either panic or create - // a packet which is different to what was requested. Choose panic as this - // would be a programming error that should be caught in testing. - panic(fmt.Sprintf("IPv4 Options %d bytes, Max %d", params.Options.SizeWithPadding(), header.IPv4MaximumOptionsSize)) - } + var optLen int + if options != nil { + optLen = int(options.Length()) + } + hdrLen += optLen + if hdrLen > header.IPv4MaximumHeaderSize { + // Since we have no way to report an error we must either panic or create + // a packet which is different to what was requested. Choose panic as this + // would be a programming error that should be caught in testing. + panic(fmt.Sprintf("IPv4 Options %d bytes, Max %d", optLen, header.IPv4MaximumOptionsSize)) } ip := header.IPv4(pkt.NetworkHeader().Push(hdrLen)) length := uint16(pkt.Size()) // RFC 6864 section 4.3 mandates uniqueness of ID values for non-atomic // datagrams. Since the DF bit is never being set here, all datagrams // are non-atomic and need an ID. - id := atomic.AddUint32(&e.protocol.ids[hashRoute(r, params.Protocol, e.protocol.hashIV)%buckets], 1) + id := atomic.AddUint32(&e.protocol.ids[hashRoute(srcAddr, dstAddr, params.Protocol, e.protocol.hashIV)%buckets], 1) ip.Encode(&header.IPv4Fields{ TotalLength: length, ID: uint16(id), TTL: params.TTL, TOS: params.TOS, Protocol: uint8(params.Protocol), - SrcAddr: r.LocalAddress, - DstAddr: r.RemoteAddress, - Options: opts, + SrcAddr: srcAddr, + DstAddr: dstAddr, + Options: options, }) ip.SetChecksum(^ip.CalculateChecksum()) pkt.NetworkProtocolNumber = ProtocolNumber @@ -259,7 +273,7 @@ func (e *endpoint) handleFragments(r *stack.Route, gso *stack.GSO, networkMTU ui // WritePacket writes a packet to the given destination address and protocol. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { - e.addIPHeader(r, pkt, params) + e.addIPHeader(r.LocalAddress, r.RemoteAddress, pkt, params, nil /* options */) // iptables filtering. All packets that reach here are locally // generated. @@ -347,7 +361,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { - e.addIPHeader(r, pkt, params) + e.addIPHeader(r.LocalAddress, r.RemoteAddress, pkt, params, nil /* options */) networkMTU, err := calculateNetworkMTU(e.nic.MTU(), uint32(pkt.NetworkHeader().View().Size())) if err != nil { r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pkts.Len())) @@ -461,7 +475,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu // non-atomic datagrams, so assign an ID to all such datagrams // according to the definition given in RFC 6864 section 4. if ip.Flags()&header.IPv4FlagDontFragment == 0 || ip.Flags()&header.IPv4FlagMoreFragments != 0 || ip.FragmentOffset() > 0 { - ip.SetID(uint16(atomic.AddUint32(&e.protocol.ids[hashRoute(r, 0 /* protocol */, e.protocol.hashIV)%buckets], 1))) + ip.SetID(uint16(atomic.AddUint32(&e.protocol.ids[hashRoute(r.LocalAddress, r.RemoteAddress, 0 /* protocol */, e.protocol.hashIV)%buckets], 1))) } } @@ -566,21 +580,6 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { stats.IP.MalformedPacketsReceived.Increment() return } - srcAddr := h.SourceAddress() - dstAddr := h.DestinationAddress() - - addressEndpoint := e.AcquireAssignedAddress(dstAddr, e.nic.Promiscuous(), stack.CanBePrimaryEndpoint) - if addressEndpoint == nil { - if !e.protocol.Forwarding() { - stats.IP.InvalidDestinationAddressesReceived.Increment() - return - } - - _ = e.forwardPacket(pkt) - return - } - subnet := addressEndpoint.AddressWithPrefix().Subnet() - addressEndpoint.DecRef() // There has been some confusion regarding verifying checksums. We need // just look for negative 0 (0xffff) as the checksum, as it's not possible to @@ -608,16 +607,42 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { return } + srcAddr := h.SourceAddress() + dstAddr := h.DestinationAddress() + // As per RFC 1122 section 3.2.1.3: // When a host sends any datagram, the IP source address MUST // be one of its own IP addresses (but not a broadcast or // multicast address). - if directedBroadcast := subnet.IsBroadcast(srcAddr); directedBroadcast || srcAddr == header.IPv4Broadcast || header.IsV4MulticastAddress(srcAddr) { + if srcAddr == header.IPv4Broadcast || header.IsV4MulticastAddress(srcAddr) { stats.IP.InvalidSourceAddressesReceived.Increment() return } + // Make sure the source address is not a subnet-local broadcast address. + if addressEndpoint := e.AcquireAssignedAddress(srcAddr, false /* createTemp */, stack.NeverPrimaryEndpoint); addressEndpoint != nil { + subnet := addressEndpoint.Subnet() + addressEndpoint.DecRef() + if subnet.IsBroadcast(srcAddr) { + stats.IP.InvalidSourceAddressesReceived.Increment() + return + } + } - pkt.NetworkPacketInfo.LocalAddressBroadcast = subnet.IsBroadcast(dstAddr) || dstAddr == header.IPv4Broadcast + // The destination address should be an address we own or a group we joined + // for us to receive the packet. Otherwise, attempt to forward the packet. + if addressEndpoint := e.AcquireAssignedAddress(dstAddr, e.nic.Promiscuous(), stack.CanBePrimaryEndpoint); addressEndpoint != nil { + subnet := addressEndpoint.AddressWithPrefix().Subnet() + addressEndpoint.DecRef() + pkt.NetworkPacketInfo.LocalAddressBroadcast = subnet.IsBroadcast(dstAddr) || dstAddr == header.IPv4Broadcast + } else if !e.IsInGroup(dstAddr) { + if !e.protocol.Forwarding() { + stats.IP.InvalidDestinationAddressesReceived.Increment() + return + } + + _ = e.forwardPacket(pkt) + return + } // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. @@ -692,6 +717,10 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { e.handleICMP(pkt) return } + if p == header.IGMPProtocolNumber { + e.igmp.handleIGMP(pkt) + return + } if opts := h.Options(); len(opts) != 0 { // TODO(gvisor.dev/issue/4586): // When we add forwarding support we should use the verified options @@ -770,28 +799,12 @@ func (e *endpoint) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp boo defer e.mu.Unlock() loopback := e.nic.IsLoopback() - addressEndpoint := e.mu.addressableEndpointState.ReadOnly().AddrOrMatching(localAddr, allowTemp, func(addressEndpoint stack.AddressEndpoint) bool { + return e.mu.addressableEndpointState.AcquireAssignedAddressOrMatching(localAddr, func(addressEndpoint stack.AddressEndpoint) bool { subnet := addressEndpoint.Subnet() // IPv4 has a notion of a subnet broadcast address and considers the // loopback interface bound to an address's whole subnet (on linux). return subnet.IsBroadcast(localAddr) || (loopback && subnet.Contains(localAddr)) - }) - if addressEndpoint != nil { - return addressEndpoint - } - - if !allowTemp { - return nil - } - - addr := localAddr.WithPrefix() - addressEndpoint, err := e.mu.addressableEndpointState.AddAndAcquireTemporaryAddress(addr, tempPEB) - if err != nil { - // AddAddress only returns an error if the address is already assigned, - // but we just checked above if the address exists so we expect no error. - panic(fmt.Sprintf("e.mu.addressableEndpointState.AddAndAcquireTemporaryAddress(%s, %d): %s", addr, tempPEB, err)) - } - return addressEndpoint + }, allowTemp, tempPEB) } // AcquireOutgoingPrimaryAddress implements stack.AddressableEndpoint. @@ -816,28 +829,43 @@ func (e *endpoint) PermanentAddresses() []tcpip.AddressWithPrefix { } // JoinGroup implements stack.GroupAddressableEndpoint. -func (e *endpoint) JoinGroup(addr tcpip.Address) (bool, *tcpip.Error) { +func (e *endpoint) JoinGroup(addr tcpip.Address) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + return e.joinGroupLocked(addr) +} + +// joinGroupLocked is like JoinGroup but with locking requirements. +// +// Precondition: e.mu must be locked. +func (e *endpoint) joinGroupLocked(addr tcpip.Address) *tcpip.Error { if !header.IsV4MulticastAddress(addr) { - return false, tcpip.ErrBadAddress + return tcpip.ErrBadAddress } - e.mu.Lock() - defer e.mu.Unlock() - return e.mu.addressableEndpointState.JoinGroup(addr) + e.igmp.joinGroup(addr) + return nil } // LeaveGroup implements stack.GroupAddressableEndpoint. -func (e *endpoint) LeaveGroup(addr tcpip.Address) (bool, *tcpip.Error) { +func (e *endpoint) LeaveGroup(addr tcpip.Address) *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - return e.mu.addressableEndpointState.LeaveGroup(addr) + return e.leaveGroupLocked(addr) +} + +// leaveGroupLocked is like LeaveGroup but with locking requirements. +// +// Precondition: e.mu must be locked. +func (e *endpoint) leaveGroupLocked(addr tcpip.Address) *tcpip.Error { + return e.igmp.leaveGroup(addr) } // IsInGroup implements stack.GroupAddressableEndpoint. func (e *endpoint) IsInGroup(addr tcpip.Address) bool { e.mu.RLock() defer e.mu.RUnlock() - return e.mu.addressableEndpointState.IsInGroup(addr) + return e.igmp.isInGroup(addr) } var _ stack.ForwardingNetworkProtocol = (*protocol)(nil) @@ -863,6 +891,8 @@ type protocol struct { hashIV uint32 fragmentation *fragmentation.Fragmentation + + options Options } // Number returns the ipv4 protocol number. @@ -987,17 +1017,23 @@ func addressToUint32(addr tcpip.Address) uint32 { return uint32(addr[0]) | uint32(addr[1])<<8 | uint32(addr[2])<<16 | uint32(addr[3])<<24 } -// hashRoute calculates a hash value for the given route. It uses the source & -// destination address, the transport protocol number and a 32-bit number to -// generate the hash. -func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber, hashIV uint32) uint32 { - a := addressToUint32(r.LocalAddress) - b := addressToUint32(r.RemoteAddress) +// hashRoute calculates a hash value for the given source/destination pair using +// the addresses, transport protocol number and a 32-bit number to generate the +// hash. +func hashRoute(srcAddr, dstAddr tcpip.Address, protocol tcpip.TransportProtocolNumber, hashIV uint32) uint32 { + a := addressToUint32(srcAddr) + b := addressToUint32(dstAddr) return hash.Hash3Words(a, b, uint32(protocol), hashIV) } -// NewProtocol returns an IPv4 network protocol. -func NewProtocol(s *stack.Stack) stack.NetworkProtocol { +// Options holds options to configure a new protocol. +type Options struct { + // IGMP holds options for IGMP. + IGMP IGMPOptions +} + +// NewProtocolWithOptions returns an IPv4 network protocol. +func NewProtocolWithOptions(opts Options) stack.NetworkProtocolFactory { ids := make([]uint32, buckets) // Randomly initialize hashIV and the ids. @@ -1007,14 +1043,22 @@ func NewProtocol(s *stack.Stack) stack.NetworkProtocol { } hashIV := r[buckets] - p := &protocol{ - stack: s, - ids: ids, - hashIV: hashIV, - defaultTTL: DefaultTTL, + return func(s *stack.Stack) stack.NetworkProtocol { + p := &protocol{ + stack: s, + ids: ids, + hashIV: hashIV, + defaultTTL: DefaultTTL, + options: opts, + } + p.fragmentation = fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, ReassembleTimeout, s.Clock(), p) + return p } - p.fragmentation = fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, ReassembleTimeout, s.Clock(), p) - return p +} + +// NewProtocol is equivalent to NewProtocolWithOptions with an empty Options. +func NewProtocol(s *stack.Stack) stack.NetworkProtocol { + return NewProtocolWithOptions(Options{})(s) } func buildNextFragment(pf *fragmentation.PacketFragmenter, originalIPHeader header.IPv4) (*stack.PacketBuffer, bool) { @@ -1129,6 +1173,12 @@ func handleTimestamp(tsOpt header.IPv4OptionTimestamp, localAddress tcpip.Addres } pointer := tsOpt.Pointer() + // RFC 791 page 22 states: "The smallest legal value is 5." + // Since the pointer is 1 based, and the header is 4 bytes long the + // pointer must point beyond the header therefore 4 or less is bad. + if pointer <= header.IPv4OptionTimestampHdrLength { + return header.IPv4OptTSPointerOffset, errIPv4TimestampOptInvalidPointer + } // To simplify processing below, base further work on the array of timestamps // beyond the header, rather than on the whole option. Also to aid // calculations set 'nextSlot' to be 0 based as in the packet it is 1 based. @@ -1215,7 +1265,15 @@ func handleRecordRoute(rrOpt header.IPv4OptionRecordRoute, localAddress tcpip.Ad return header.IPv4OptionLengthOffset, errIPv4RecordRouteOptInvalidLength } - nextSlot := rrOpt.Pointer() - 1 // Pointer is 1 based. + pointer := rrOpt.Pointer() + // RFC 791 page 20 states: + // The pointer is relative to this option, and the + // smallest legal value for the pointer is 4. + // Since the pointer is 1 based, and the header is 3 bytes long the + // pointer must point beyond the header therefore 3 or less is bad. + if pointer <= header.IPv4OptionRecordRouteHdrLength { + return header.IPv4OptRRPointerOffset, errIPv4RecordRouteOptInvalidPointer + } // RFC 791 page 21 says // If the route data area is already full (the pointer exceeds the @@ -1230,14 +1288,14 @@ func handleRecordRoute(rrOpt header.IPv4OptionRecordRoute, localAddress tcpip.Ad // do this (as do most implementations). It is probable that the inclusion // of these words is a copy/paste error from the timestamp option where // there are two failure reasons given. - if nextSlot >= optlen { + if pointer > optlen { return 0, nil } // The data area isn't full but there isn't room for a new entry. // Either Length or Pointer could be bad. We must select Pointer for Linux - // compatibility, even if only the length is bad. - if nextSlot+header.IPv4AddressSize > optlen { + // compatibility, even if only the length is bad. NB. pointer is 1 based. + if pointer+header.IPv4AddressSize > optlen+1 { if false { // This is what we would do if we were not being Linux compatible. // Check for bad pointer or length value. Must be a multiple of 4 after diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index 4e4e1f3b4..9e2d2cfd6 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -103,105 +103,6 @@ func TestExcludeBroadcast(t *testing.T) { }) } -// TestIPv4Encode checks that ipv4.Encode correctly fills out the requested -// fields when options are supplied. -func TestIPv4EncodeOptions(t *testing.T) { - tests := []struct { - name string - options header.IPv4Options - encodedOptions header.IPv4Options // reply should look like this - wantIHL int - }{ - { - name: "valid no options", - wantIHL: header.IPv4MinimumSize, - }, - { - name: "one byte options", - options: header.IPv4Options{1}, - encodedOptions: header.IPv4Options{1, 0, 0, 0}, - wantIHL: header.IPv4MinimumSize + 4, - }, - { - name: "two byte options", - options: header.IPv4Options{1, 1}, - encodedOptions: header.IPv4Options{1, 1, 0, 0}, - wantIHL: header.IPv4MinimumSize + 4, - }, - { - name: "three byte options", - options: header.IPv4Options{1, 1, 1}, - encodedOptions: header.IPv4Options{1, 1, 1, 0}, - wantIHL: header.IPv4MinimumSize + 4, - }, - { - name: "four byte options", - options: header.IPv4Options{1, 1, 1, 1}, - encodedOptions: header.IPv4Options{1, 1, 1, 1}, - wantIHL: header.IPv4MinimumSize + 4, - }, - { - name: "five byte options", - options: header.IPv4Options{1, 1, 1, 1, 1}, - encodedOptions: header.IPv4Options{1, 1, 1, 1, 1, 0, 0, 0}, - wantIHL: header.IPv4MinimumSize + 8, - }, - { - name: "thirty nine byte options", - options: header.IPv4Options{ - 1, 2, 3, 4, 5, 6, 7, 8, - 9, 10, 11, 12, 13, 14, 15, 16, - 17, 18, 19, 20, 21, 22, 23, 24, - 25, 26, 27, 28, 29, 30, 31, 32, - 33, 34, 35, 36, 37, 38, 39, - }, - encodedOptions: header.IPv4Options{ - 1, 2, 3, 4, 5, 6, 7, 8, - 9, 10, 11, 12, 13, 14, 15, 16, - 17, 18, 19, 20, 21, 22, 23, 24, - 25, 26, 27, 28, 29, 30, 31, 32, - 33, 34, 35, 36, 37, 38, 39, 0, - }, - wantIHL: header.IPv4MinimumSize + 40, - }, - } - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - paddedOptionLength := test.options.SizeWithPadding() - ipHeaderLength := header.IPv4MinimumSize + paddedOptionLength - if ipHeaderLength > header.IPv4MaximumHeaderSize { - t.Fatalf("IP header length too large: got = %d, want <= %d ", ipHeaderLength, header.IPv4MaximumHeaderSize) - } - totalLen := uint16(ipHeaderLength) - hdr := buffer.NewPrependable(int(totalLen)) - ip := header.IPv4(hdr.Prepend(ipHeaderLength)) - // To check the padding works, poison the last byte of the options space. - if paddedOptionLength != len(test.options) { - ip.SetHeaderLength(uint8(ipHeaderLength)) - ip.Options()[paddedOptionLength-1] = 0xff - ip.SetHeaderLength(0) - } - ip.Encode(&header.IPv4Fields{ - Options: test.options, - }) - options := ip.Options() - wantOptions := test.encodedOptions - if got, want := int(ip.HeaderLength()), test.wantIHL; got != want { - t.Errorf("got IHL of %d, want %d", got, want) - } - - // cmp.Diff does not consider nil slices equal to empty slices, but we do. - if len(wantOptions) == 0 && len(options) == 0 { - return - } - - if diff := cmp.Diff(wantOptions, options); diff != "" { - t.Errorf("options mismatch (-want +got):\n%s", diff) - } - }) - } -} - func TestForwarding(t *testing.T) { const ( nicID1 = 1 @@ -453,14 +354,6 @@ func TestIPv4Sanity(t *testing.T) { replyOptions: header.IPv4Options{1, 1, 0, 0}, }, { - name: "Check option padding", - maxTotalLength: ipv4.MaxTotalSize, - transportProtocol: uint8(header.ICMPv4ProtocolNumber), - TTL: ttl, - options: header.IPv4Options{1, 1, 1}, - replyOptions: header.IPv4Options{1, 1, 1, 0}, - }, - { name: "bad header length", headerLength: header.IPv4MinimumSize - 1, maxTotalLength: ipv4.MaxTotalSize, @@ -583,7 +476,7 @@ func TestIPv4Sanity(t *testing.T) { 68, 7, 5, 0, // ^ ^ Linux points here which is wrong. // | Not a multiple of 4 - 1, 2, 3, + 1, 2, 3, 0, }, shouldFail: true, expectErrorICMP: true, @@ -662,6 +555,56 @@ func TestIPv4Sanity(t *testing.T) { }, }, { + // Timestamp pointer uses one based counting so 0 is invalid. + name: "timestamp pointer invalid", + maxTotalLength: ipv4.MaxTotalSize, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: header.IPv4Options{ + 68, 8, 0, 0x00, + // ^ 0 instead of 5 or more. + 0, 0, 0, 0, + }, + shouldFail: true, + expectErrorICMP: true, + ICMPType: header.ICMPv4ParamProblem, + ICMPCode: header.ICMPv4UnusedCode, + paramProblemPointer: header.IPv4MinimumSize + 2, + }, + { + // Timestamp pointer cannot be less than 5. It must point past the header + // which is 4 bytes. (1 based counting) + name: "timestamp pointer too small by 1", + maxTotalLength: ipv4.MaxTotalSize, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: header.IPv4Options{ + 68, 8, header.IPv4OptionTimestampHdrLength, 0x00, + // ^ header is 4 bytes, so 4 should fail. + 0, 0, 0, 0, + }, + shouldFail: true, + expectErrorICMP: true, + ICMPType: header.ICMPv4ParamProblem, + ICMPCode: header.ICMPv4UnusedCode, + paramProblemPointer: header.IPv4MinimumSize + 2, + }, + { + name: "valid timestamp pointer", + maxTotalLength: ipv4.MaxTotalSize, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: header.IPv4Options{ + 68, 8, header.IPv4OptionTimestampHdrLength + 1, 0x00, + // ^ header is 4 bytes, so 5 should succeed. + 0, 0, 0, 0, + }, + replyOptions: header.IPv4Options{ + 68, 8, 9, 0x00, + 0x00, 0xad, 0x1c, 0x40, // time we expect from fakeclock + }, + }, + { // Needs 8 bytes for a type 1 timestamp but there are only 4 free. name: "bad timer element alignment", maxTotalLength: ipv4.MaxTotalSize, @@ -792,7 +735,61 @@ func TestIPv4Sanity(t *testing.T) { }, }, { - // Confirm linux bug for bug compatibility. + // Pointer uses one based counting so 0 is invalid. + name: "record route pointer zero", + maxTotalLength: ipv4.MaxTotalSize, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: header.IPv4Options{ + 7, 8, 0, // 3 byte header + 0, 0, 0, 0, + 0, + }, + shouldFail: true, + expectErrorICMP: true, + ICMPType: header.ICMPv4ParamProblem, + ICMPCode: header.ICMPv4UnusedCode, + paramProblemPointer: header.IPv4MinimumSize + 2, + }, + { + // Pointer must be 4 or more as it must point past the 3 byte header + // using 1 based counting. 3 should fail. + name: "record route pointer too small by 1", + maxTotalLength: ipv4.MaxTotalSize, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: header.IPv4Options{ + 7, 8, header.IPv4OptionRecordRouteHdrLength, // 3 byte header + 0, 0, 0, 0, + 0, + }, + shouldFail: true, + expectErrorICMP: true, + ICMPType: header.ICMPv4ParamProblem, + ICMPCode: header.ICMPv4UnusedCode, + paramProblemPointer: header.IPv4MinimumSize + 2, + }, + { + // Pointer must be 4 or more as it must point past the 3 byte header + // using 1 based counting. Check 4 passes. (Duplicates "single + // record route with room") + name: "valid record route pointer", + maxTotalLength: ipv4.MaxTotalSize, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: header.IPv4Options{ + 7, 7, header.IPv4OptionRecordRouteHdrLength + 1, // 3 byte header + 0, 0, 0, 0, + 0, + }, + replyOptions: header.IPv4Options{ + 7, 7, 8, // 3 byte header + 192, 168, 1, 58, // New IP Address. + 0, // padding to multiple of 4 bytes. + }, + }, + { + // Confirm Linux bug for bug compatibility. // Linux returns slot 22 but the error is in slot 21. name: "multiple record route with not enough room", maxTotalLength: ipv4.MaxTotalSize, @@ -863,8 +860,10 @@ func TestIPv4Sanity(t *testing.T) { }, }) - paddedOptionLength := test.options.SizeWithPadding() - ipHeaderLength := header.IPv4MinimumSize + paddedOptionLength + if len(test.options)%4 != 0 { + t.Fatalf("options must be aligned to 32 bits, invalid test options: %x (len=%d)", test.options, len(test.options)) + } + ipHeaderLength := header.IPv4MinimumSize + len(test.options) if ipHeaderLength > header.IPv4MaximumHeaderSize { t.Fatalf("IP header length too large: got = %d, want <= %d ", ipHeaderLength, header.IPv4MaximumHeaderSize) } @@ -883,11 +882,6 @@ func TestIPv4Sanity(t *testing.T) { if test.maxTotalLength < totalLen { totalLen = test.maxTotalLength } - // To check the padding works, poison the options space. - if paddedOptionLength != len(test.options) { - ip.SetHeaderLength(uint8(ipHeaderLength)) - ip.Options()[paddedOptionLength-1] = 0x01 - } ip.Encode(&header.IPv4Fields{ TotalLength: totalLen, @@ -895,10 +889,19 @@ func TestIPv4Sanity(t *testing.T) { TTL: test.TTL, SrcAddr: remoteIPv4Addr, DstAddr: ipv4Addr.Address, - Options: test.options, }) if test.headerLength != 0 { ip.SetHeaderLength(test.headerLength) + } else { + // Set the calculated header length, since we may manually add options. + ip.SetHeaderLength(uint8(ipHeaderLength)) + } + if len(test.options) != 0 { + // Copy options manually. We do not use Encode for options so we can + // verify malformed options with handcrafted payloads. + if want, got := copy(ip.Options(), test.options), len(test.options); want != got { + t.Fatalf("got copy(ip.Options(), test.options) = %d, want = %d", got, want) + } } ip.SetChecksum(0) ipHeaderChecksum := ip.CalculateChecksum() @@ -1003,7 +1006,7 @@ func TestIPv4Sanity(t *testing.T) { } // If the IP options change size then the packet will change size, so // some IP header fields will need to be adjusted for the checks. - sizeChange := len(test.replyOptions) - paddedOptionLength + sizeChange := len(test.replyOptions) - len(test.options) checker.IPv4(t, replyIPHeader, checker.IPv4HeaderLength(ipHeaderLength+sizeChange), @@ -2320,6 +2323,28 @@ func TestReceiveFragments(t *testing.T) { }, expectedPayloads: [][]byte{udpPayload4Addr1ToAddr2}, }, + { + name: "Two fragments with MF flag reassembled into a maximum UDP packet", + fragments: []fragmentData{ + { + srcAddr: addr1, + dstAddr: addr2, + id: 1, + flags: header.IPv4FlagMoreFragments, + fragmentOffset: 0, + payload: ipv4Payload4Addr1ToAddr2[:65512], + }, + { + srcAddr: addr1, + dstAddr: addr2, + id: 1, + flags: header.IPv4FlagMoreFragments, + fragmentOffset: 65512, + payload: ipv4Payload4Addr1ToAddr2[65512:], + }, + }, + expectedPayloads: nil, + }, } for _, test := range tests { @@ -2513,7 +2538,7 @@ func TestWriteStats(t *testing.T) { test.setup(t, rt.Stack()) - nWritten, _ := writer.writePackets(&rt, pkts) + nWritten, _ := writer.writePackets(rt, pkts) if got := int(rt.Stats().IP.PacketsSent.Value()); got != test.expectSent { t.Errorf("sent %d packets, but expected to send %d", got, test.expectSent) @@ -2530,7 +2555,7 @@ func TestWriteStats(t *testing.T) { } } -func buildRoute(t *testing.T, ep stack.LinkEndpoint) stack.Route { +func buildRoute(t *testing.T, ep stack.LinkEndpoint) *stack.Route { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, }) @@ -2644,8 +2669,8 @@ func TestPacketQueing(t *testing.T) { if p.Proto != header.IPv4ProtocolNumber { t.Errorf("got p.Proto = %d, want = %d", p.Proto, header.IPv4ProtocolNumber) } - if p.Route.RemoteLinkAddress != host2NICLinkAddr { - t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, host2NICLinkAddr) + if got := p.Route.RemoteLinkAddress(); got != host2NICLinkAddr { + t.Errorf("got p.Route.RemoteLinkAddress() = %s, want = %s", got, host2NICLinkAddr) } checker.IPv4(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(host1IPv4Addr.AddressWithPrefix.Address), @@ -2687,8 +2712,8 @@ func TestPacketQueing(t *testing.T) { if p.Proto != header.IPv4ProtocolNumber { t.Errorf("got p.Proto = %d, want = %d", p.Proto, header.IPv4ProtocolNumber) } - if p.Route.RemoteLinkAddress != host2NICLinkAddr { - t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, host2NICLinkAddr) + if got := p.Route.RemoteLinkAddress(); got != host2NICLinkAddr { + t.Errorf("got p.Route.RemoteLinkAddress() = %s, want = %s", got, host2NICLinkAddr) } checker.IPv4(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(host1IPv4Addr.AddressWithPrefix.Address), @@ -2736,8 +2761,8 @@ func TestPacketQueing(t *testing.T) { if p.Proto != arp.ProtocolNumber { t.Errorf("got p.Proto = %d, want = %d", p.Proto, arp.ProtocolNumber) } - if p.Route.RemoteLinkAddress != header.EthernetBroadcastAddress { - t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, header.EthernetBroadcastAddress) + if got := p.Route.RemoteLinkAddress(); got != header.EthernetBroadcastAddress { + t.Errorf("got p.Route.RemoteLinkAddress() = %s, want = %s", got, header.EthernetBroadcastAddress) } rep := header.ARP(p.Pkt.NetworkHeader().View()) if got := rep.Op(); got != header.ARPRequest { diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index 0ac24a6fb..5e75c8740 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -8,6 +8,7 @@ go_library( "dhcpv6configurationfromndpra_string.go", "icmp.go", "ipv6.go", + "mld.go", "ndp.go", ], visibility = ["//visibility:public"], @@ -19,6 +20,7 @@ go_library( "//pkg/tcpip/header/parse", "//pkg/tcpip/network/fragmentation", "//pkg/tcpip/network/hash", + "//pkg/tcpip/network/ip", "//pkg/tcpip/stack", ], ) @@ -49,3 +51,16 @@ go_test( "@com_github_google_go_cmp//cmp:go_default_library", ], ) + +go_test( + name = "ipv6_x_test", + size = "small", + srcs = ["mld_test.go"], + deps = [ + ":ipv6", + "//pkg/tcpip/checker", + "//pkg/tcpip/header", + "//pkg/tcpip/link/channel", + "//pkg/tcpip/stack", + ], +) diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index beb8f562e..510276b8e 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -126,8 +126,8 @@ func getTargetLinkAddr(it header.NDPOptionIterator) (tcpip.LinkAddress, bool) { func (e *endpoint) handleICMP(pkt *stack.PacketBuffer, hasFragmentHeader bool) { stats := e.protocol.stack.Stats().ICMP - sent := stats.V6PacketsSent - received := stats.V6PacketsReceived + sent := stats.V6.PacketsSent + received := stats.V6.PacketsReceived // TODO(gvisor.dev/issue/170): ICMP packets don't have their // TransportHeader fields set. See icmp/protocol.go:protocol.Parse for a // full explanation. @@ -163,7 +163,7 @@ func (e *endpoint) handleICMP(pkt *stack.PacketBuffer, hasFragmentHeader bool) { } // TODO(b/112892170): Meaningfully handle all ICMP types. - switch h.Type() { + switch icmpType := h.Type(); icmpType { case header.ICMPv6PacketTooBig: received.PacketTooBig.Increment() hdr, ok := pkt.Data.PullUp(header.ICMPv6PacketTooBigMinimumSize) @@ -358,7 +358,7 @@ func (e *endpoint) handleICMP(pkt *stack.PacketBuffer, hasFragmentHeader bool) { pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber packet := header.ICMPv6(pkt.TransportHeader().Push(neighborAdvertSize)) packet.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(packet.NDPPayload()) + na := header.NDPNeighborAdvert(packet.MessageBody()) // As per RFC 4861 section 7.2.4: // @@ -644,8 +644,31 @@ func (e *endpoint) handleICMP(pkt *stack.PacketBuffer, hasFragmentHeader bool) { return } + case header.ICMPv6MulticastListenerQuery, header.ICMPv6MulticastListenerReport, header.ICMPv6MulticastListenerDone: + var handler func(header.MLD) + switch icmpType { + case header.ICMPv6MulticastListenerQuery: + received.MulticastListenerQuery.Increment() + handler = e.mld.handleMulticastListenerQuery + case header.ICMPv6MulticastListenerReport: + received.MulticastListenerReport.Increment() + handler = e.mld.handleMulticastListenerReport + case header.ICMPv6MulticastListenerDone: + received.MulticastListenerDone.Increment() + default: + panic(fmt.Sprintf("unrecognized MLD message = %d", icmpType)) + } + if pkt.Data.Size()-header.ICMPv6HeaderSize < header.MLDMinimumSize { + received.Invalid.Increment() + return + } + + if handler != nil { + handler(header.MLD(payload.ToView())) + } + default: - received.Invalid.Increment() + received.Unrecognized.Increment() } } @@ -681,12 +704,12 @@ func (p *protocol) LinkAddressRequest(targetAddr, localAddr tcpip.Address, remot pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber packet := header.ICMPv6(pkt.TransportHeader().Push(neighborSolicitSize)) packet.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(packet.NDPPayload()) + ns := header.NDPNeighborSolicit(packet.MessageBody()) ns.SetTargetAddress(targetAddr) ns.Options().Serialize(optsSerializer) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - stat := p.stack.Stats().ICMP.V6PacketsSent + stat := p.stack.Stats().ICMP.V6.PacketsSent if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, @@ -796,7 +819,8 @@ func (p *protocol) returnError(reason icmpReason, pkt *stack.PacketBuffer) *tcpi allowResponseToMulticast = reason.respondToMulticast } - if (!allowResponseToMulticast && header.IsV6MulticastAddress(origIPHdrDst)) || origIPHdrSrc == header.IPv6Any { + isOrigDstMulticast := header.IsV6MulticastAddress(origIPHdrDst) + if (!allowResponseToMulticast && isOrigDstMulticast) || origIPHdrSrc == header.IPv6Any { return nil } @@ -812,8 +836,13 @@ func (p *protocol) returnError(reason icmpReason, pkt *stack.PacketBuffer) *tcpi // If we are operating as a router, do not use the packet's destination // address as the response's source address as we should not own the // destination address of a packet we are forwarding. + // + // If the packet was originally destined to a multicast address, then do not + // use the packet's destination address as the source for the response ICMP + // packet as "multicast addresses must not be used as source addresses in IPv6 + // packets", as per RFC 4291 section 2.7. localAddr := origIPHdrDst - if _, ok := reason.(*icmpReasonHopLimitExceeded); ok { + if _, ok := reason.(*icmpReasonHopLimitExceeded); ok || isOrigDstMulticast { localAddr = "" } // Even if we were able to receive a packet from some remote, we may not have @@ -827,7 +856,7 @@ func (p *protocol) returnError(reason icmpReason, pkt *stack.PacketBuffer) *tcpi defer route.Release() stats := p.stack.Stats().ICMP - sent := stats.V6PacketsSent + sent := stats.V6.PacketsSent if !p.stack.AllowICMPMessage() { sent.RateLimited.Increment() return nil diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 9bc02d851..32adb5c83 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -150,9 +150,9 @@ func (*testInterface) Promiscuous() bool { func (t *testInterface) WritePacketToRemote(remoteLinkAddr tcpip.LinkAddress, gso *stack.GSO, protocol tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { r := stack.Route{ - NetProto: protocol, - RemoteLinkAddress: remoteLinkAddr, + NetProto: protocol, } + r.ResolveWith(remoteLinkAddr) return t.LinkEndpoint.WritePacket(&r, gso, protocol, pkt) } @@ -271,6 +271,22 @@ func TestICMPCounts(t *testing.T) { typ: header.ICMPv6RedirectMsg, size: header.ICMPv6MinimumSize, }, + { + typ: header.ICMPv6MulticastListenerQuery, + size: header.MLDMinimumSize + header.ICMPv6HeaderSize, + }, + { + typ: header.ICMPv6MulticastListenerReport, + size: header.MLDMinimumSize + header.ICMPv6HeaderSize, + }, + { + typ: header.ICMPv6MulticastListenerDone, + size: header.MLDMinimumSize + header.ICMPv6HeaderSize, + }, + { + typ: 255, /* Unrecognized */ + size: 50, + }, } handleIPv6Payload := func(icmp header.ICMPv6) { @@ -301,7 +317,7 @@ func TestICMPCounts(t *testing.T) { // Stats().ICMP.ICMPv6ReceivedPacketStats.Invalid is incremented. handleIPv6Payload(header.ICMPv6(buffer.NewView(header.IPv6MinimumSize))) - icmpv6Stats := s.Stats().ICMP.V6PacketsReceived + icmpv6Stats := s.Stats().ICMP.V6.PacketsReceived visitStats(reflect.ValueOf(&icmpv6Stats).Elem(), func(name string, s *tcpip.StatCounter) { if got, want := s.Value(), uint64(1); got != want { t.Errorf("got %s = %d, want = %d", name, got, want) @@ -413,6 +429,22 @@ func TestICMPCountsWithNeighborCache(t *testing.T) { typ: header.ICMPv6RedirectMsg, size: header.ICMPv6MinimumSize, }, + { + typ: header.ICMPv6MulticastListenerQuery, + size: header.MLDMinimumSize + header.ICMPv6HeaderSize, + }, + { + typ: header.ICMPv6MulticastListenerReport, + size: header.MLDMinimumSize + header.ICMPv6HeaderSize, + }, + { + typ: header.ICMPv6MulticastListenerDone, + size: header.MLDMinimumSize + header.ICMPv6HeaderSize, + }, + { + typ: 255, /* Unrecognized */ + size: 50, + }, } handleIPv6Payload := func(icmp header.ICMPv6) { @@ -443,7 +475,7 @@ func TestICMPCountsWithNeighborCache(t *testing.T) { // Stats().ICMP.ICMPv6ReceivedPacketStats.Invalid is incremented. handleIPv6Payload(header.ICMPv6(buffer.NewView(header.IPv6MinimumSize))) - icmpv6Stats := s.Stats().ICMP.V6PacketsReceived + icmpv6Stats := s.Stats().ICMP.V6.PacketsReceived visitStats(reflect.ValueOf(&icmpv6Stats).Elem(), func(name string, s *tcpip.StatCounter) { if got, want := s.Value(), uint64(1); got != want { t.Errorf("got %s = %d, want = %d", name, got, want) @@ -568,8 +600,8 @@ func routeICMPv6Packet(t *testing.T, args routeArgs, fn func(*testing.T, header. return } - if len(args.remoteLinkAddr) != 0 && args.remoteLinkAddr != pi.Route.RemoteLinkAddress { - t.Errorf("got remote link address = %s, want = %s", pi.Route.RemoteLinkAddress, args.remoteLinkAddr) + if got := pi.Route.RemoteLinkAddress(); len(args.remoteLinkAddr) != 0 && got != args.remoteLinkAddr { + t.Errorf("got remote link address = %s, want = %s", got, args.remoteLinkAddr) } // Pull the full payload since network header. Needed for header.IPv6 to @@ -833,7 +865,7 @@ func TestICMPChecksumValidationSimple(t *testing.T) { e.InjectInbound(ProtocolNumber, pkt) } - stats := s.Stats().ICMP.V6PacketsReceived + stats := s.Stats().ICMP.V6.PacketsReceived invalid := stats.Invalid routerOnly := stats.RouterOnlyPacketsDroppedByHost typStat := typ.statCounter(stats) @@ -1028,7 +1060,7 @@ func TestICMPChecksumValidationWithPayload(t *testing.T) { e.InjectInbound(ProtocolNumber, pkt) } - stats := s.Stats().ICMP.V6PacketsReceived + stats := s.Stats().ICMP.V6.PacketsReceived invalid := stats.Invalid typStat := typ.statCounter(stats) @@ -1207,7 +1239,7 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { e.InjectInbound(ProtocolNumber, pkt) } - stats := s.Stats().ICMP.V6PacketsReceived + stats := s.Stats().ICMP.V6.PacketsReceived invalid := stats.Invalid typStat := typ.statCounter(stats) @@ -1349,8 +1381,8 @@ func TestLinkAddressRequest(t *testing.T) { if !ok { t.Fatal("expected to send a link address request") } - if pkt.Route.RemoteLinkAddress != test.expectedRemoteLinkAddr { - t.Errorf("got pkt.Route.RemoteLinkAddress = %s, want = %s", pkt.Route.RemoteLinkAddress, test.expectedRemoteLinkAddr) + if got := pkt.Route.RemoteLinkAddress(); got != test.expectedRemoteLinkAddr { + t.Errorf("got pkt.Route.RemoteLinkAddress() = %s, want = %s", got, test.expectedRemoteLinkAddr) } if pkt.Route.RemoteAddress != test.expectedRemoteAddr { t.Errorf("got pkt.Route.RemoteAddress = %s, want = %s", pkt.Route.RemoteAddress, test.expectedRemoteAddr) @@ -1431,8 +1463,8 @@ func TestPacketQueing(t *testing.T) { if p.Proto != ProtocolNumber { t.Errorf("got p.Proto = %d, want = %d", p.Proto, ProtocolNumber) } - if p.Route.RemoteLinkAddress != host2NICLinkAddr { - t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, host2NICLinkAddr) + if got := p.Route.RemoteLinkAddress(); got != host2NICLinkAddr { + t.Errorf("got p.Route.RemoteLinkAddress() = %s, want = %s", got, host2NICLinkAddr) } checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(host1IPv6Addr.AddressWithPrefix.Address), @@ -1473,8 +1505,8 @@ func TestPacketQueing(t *testing.T) { if p.Proto != ProtocolNumber { t.Errorf("got p.Proto = %d, want = %d", p.Proto, ProtocolNumber) } - if p.Route.RemoteLinkAddress != host2NICLinkAddr { - t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, host2NICLinkAddr) + if got := p.Route.RemoteLinkAddress(); got != host2NICLinkAddr { + t.Errorf("got p.Route.RemoteLinkAddress() = %s, want = %s", got, host2NICLinkAddr) } checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(host1IPv6Addr.AddressWithPrefix.Address), @@ -1524,8 +1556,8 @@ func TestPacketQueing(t *testing.T) { t.Errorf("got Proto = %d, want = %d", p.Proto, ProtocolNumber) } snmc := header.SolicitedNodeAddr(host2IPv6Addr.AddressWithPrefix.Address) - if want := header.EthernetAddressFromMulticastIPv6Address(snmc); p.Route.RemoteLinkAddress != want { - t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, want) + if got, want := p.Route.RemoteLinkAddress(), header.EthernetAddressFromMulticastIPv6Address(snmc); got != want { + t.Errorf("got p.Route.RemoteLinkAddress() = %s, want = %s", got, want) } checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), checker.SrcAddr(host1IPv6Addr.AddressWithPrefix.Address), @@ -1543,7 +1575,7 @@ func TestPacketQueing(t *testing.T) { hdr := buffer.NewPrependable(header.IPv6MinimumSize + naSize) pkt := header.ICMPv6(hdr.Prepend(naSize)) pkt.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(pkt.NDPPayload()) + na := header.NDPNeighborAdvert(pkt.MessageBody()) na.SetSolicitedFlag(true) na.SetOverrideFlag(true) na.SetTargetAddress(host2IPv6Addr.AddressWithPrefix.Address) @@ -1592,7 +1624,7 @@ func TestCallsToNeighborCache(t *testing.T) { nsSize := header.ICMPv6NeighborSolicitMinimumSize + header.NDPLinkLayerAddressSize icmp := header.ICMPv6(buffer.NewView(nsSize)) icmp.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(icmp.NDPPayload()) + ns := header.NDPNeighborSolicit(icmp.MessageBody()) ns.SetTargetAddress(lladdr0) return icmp }, @@ -1612,7 +1644,7 @@ func TestCallsToNeighborCache(t *testing.T) { nsSize := header.ICMPv6NeighborSolicitMinimumSize + header.NDPLinkLayerAddressSize icmp := header.ICMPv6(buffer.NewView(nsSize)) icmp.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(icmp.NDPPayload()) + ns := header.NDPNeighborSolicit(icmp.MessageBody()) ns.SetTargetAddress(lladdr0) ns.Options().Serialize(header.NDPOptionsSerializer{ header.NDPSourceLinkLayerAddressOption(linkAddr1), @@ -1629,7 +1661,7 @@ func TestCallsToNeighborCache(t *testing.T) { nsSize := header.ICMPv6NeighborSolicitMinimumSize + header.NDPLinkLayerAddressSize icmp := header.ICMPv6(buffer.NewView(nsSize)) icmp.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(icmp.NDPPayload()) + ns := header.NDPNeighborSolicit(icmp.MessageBody()) ns.SetTargetAddress(lladdr0) return icmp }, @@ -1645,7 +1677,7 @@ func TestCallsToNeighborCache(t *testing.T) { nsSize := header.ICMPv6NeighborSolicitMinimumSize + header.NDPLinkLayerAddressSize icmp := header.ICMPv6(buffer.NewView(nsSize)) icmp.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(icmp.NDPPayload()) + ns := header.NDPNeighborSolicit(icmp.MessageBody()) ns.SetTargetAddress(lladdr0) ns.Options().Serialize(header.NDPOptionsSerializer{ header.NDPSourceLinkLayerAddressOption(linkAddr1), @@ -1662,7 +1694,7 @@ func TestCallsToNeighborCache(t *testing.T) { naSize := header.ICMPv6NeighborAdvertMinimumSize icmp := header.ICMPv6(buffer.NewView(naSize)) icmp.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(icmp.NDPPayload()) + na := header.NDPNeighborAdvert(icmp.MessageBody()) na.SetSolicitedFlag(true) na.SetOverrideFlag(false) na.SetTargetAddress(lladdr1) @@ -1683,7 +1715,7 @@ func TestCallsToNeighborCache(t *testing.T) { naSize := header.ICMPv6NeighborAdvertMinimumSize + header.NDPLinkLayerAddressSize icmp := header.ICMPv6(buffer.NewView(naSize)) icmp.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(icmp.NDPPayload()) + na := header.NDPNeighborAdvert(icmp.MessageBody()) na.SetSolicitedFlag(true) na.SetOverrideFlag(false) na.SetTargetAddress(lladdr1) @@ -1702,7 +1734,7 @@ func TestCallsToNeighborCache(t *testing.T) { naSize := header.ICMPv6NeighborAdvertMinimumSize + header.NDPLinkLayerAddressSize icmp := header.ICMPv6(buffer.NewView(naSize)) icmp.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(icmp.NDPPayload()) + na := header.NDPNeighborAdvert(icmp.MessageBody()) na.SetSolicitedFlag(false) na.SetOverrideFlag(false) na.SetTargetAddress(lladdr1) @@ -1722,7 +1754,7 @@ func TestCallsToNeighborCache(t *testing.T) { naSize := header.ICMPv6NeighborAdvertMinimumSize + header.NDPLinkLayerAddressSize icmp := header.ICMPv6(buffer.NewView(naSize)) icmp.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(icmp.NDPPayload()) + na := header.NDPNeighborAdvert(icmp.MessageBody()) na.SetSolicitedFlag(false) na.SetOverrideFlag(false) na.SetTargetAddress(lladdr1) diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 7a00f6314..8bf84601f 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -34,7 +34,9 @@ import ( ) const ( + // ReassembleTimeout controls how long a fragment will be held. // As per RFC 8200 section 4.5: + // // If insufficient fragments are received to complete reassembly of a packet // within 60 seconds of the reception of the first-arriving fragment of that // packet, reassembly of that packet must be abandoned. @@ -84,6 +86,8 @@ type endpoint struct { addressableEndpointState stack.AddressableEndpointState ndp ndpState } + + mld mldState } // NICNameFromID is a function that returns a stable name for the specified NIC, @@ -224,6 +228,12 @@ func (e *endpoint) Enable() *tcpip.Error { return nil } + // Groups may have been joined when the endpoint was disabled, or the + // endpoint may have left groups from the perspective of MLD when the + // endpoint was disabled. Either way, we need to let routers know to + // send us multicast traffic. + e.mld.initializeAll() + // Join the IPv6 All-Nodes Multicast group if the stack is configured to // use IPv6. This is required to ensure that this node properly receives // and responds to the various NDP messages that are destined to the @@ -241,8 +251,10 @@ func (e *endpoint) Enable() *tcpip.Error { // (NDP NS) messages may be sent to the All-Nodes multicast group if the // source address of the NDP NS is the unspecified address, as per RFC 4861 // section 7.2.4. - if _, err := e.mu.addressableEndpointState.JoinGroup(header.IPv6AllNodesMulticastAddress); err != nil { - return err + if err := e.joinGroupLocked(header.IPv6AllNodesMulticastAddress); err != nil { + // joinGroupLocked only returns an error if the group address is not a valid + // IPv6 multicast address. + panic(fmt.Sprintf("e.joinGroupLocked(%s): %s", header.IPv6AllNodesMulticastAddress, err)) } // Perform DAD on the all the unicast IPv6 endpoints that are in the permanent @@ -251,7 +263,7 @@ func (e *endpoint) Enable() *tcpip.Error { // Addresses may have aleady completed DAD but in the time since the endpoint // was last enabled, other devices may have acquired the same addresses. var err *tcpip.Error - e.mu.addressableEndpointState.ReadOnly().ForEach(func(addressEndpoint stack.AddressEndpoint) bool { + e.mu.addressableEndpointState.ForEachEndpoint(func(addressEndpoint stack.AddressEndpoint) bool { addr := addressEndpoint.AddressWithPrefix().Address if !header.IsV6UnicastAddress(addr) { return true @@ -273,7 +285,7 @@ func (e *endpoint) Enable() *tcpip.Error { } // Do not auto-generate an IPv6 link-local address for loopback devices. - if e.protocol.autoGenIPv6LinkLocal && !e.nic.IsLoopback() { + if e.protocol.options.AutoGenLinkLocal && !e.nic.IsLoopback() { // The valid and preferred lifetime is infinite for the auto-generated // link-local address. e.mu.ndp.doSLAAC(header.IPv6LinkLocalPrefix.Subnet(), header.NDPInfiniteLifetime, header.NDPInfiniteLifetime) @@ -331,9 +343,13 @@ func (e *endpoint) disableLocked() { e.stopDADForPermanentAddressesLocked() // The endpoint may have already left the multicast group. - if _, err := e.mu.addressableEndpointState.LeaveGroup(header.IPv6AllNodesMulticastAddress); err != nil && err != tcpip.ErrBadLocalAddress { + if err := e.leaveGroupLocked(header.IPv6AllNodesMulticastAddress); err != nil && err != tcpip.ErrBadLocalAddress { panic(fmt.Sprintf("unexpected error when leaving group = %s: %s", header.IPv6AllNodesMulticastAddress, err)) } + + // Leave groups from the perspective of MLD so that routers know that + // we are no longer interested in the group. + e.mld.softLeaveAll() } // stopDADForPermanentAddressesLocked stops DAD for all permaneent addresses. @@ -341,7 +357,7 @@ func (e *endpoint) disableLocked() { // Precondition: e.mu must be write locked. func (e *endpoint) stopDADForPermanentAddressesLocked() { // Stop DAD for all the tentative unicast addresses. - e.mu.addressableEndpointState.ReadOnly().ForEach(func(addressEndpoint stack.AddressEndpoint) bool { + e.mu.addressableEndpointState.ForEachEndpoint(func(addressEndpoint stack.AddressEndpoint) bool { if addressEndpoint.GetKind() != stack.PermanentTentative { return true } @@ -376,7 +392,7 @@ func (e *endpoint) MaxHeaderLength() uint16 { return e.nic.MaxHeaderLength() + header.IPv6MinimumSize } -func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) { +func (e *endpoint) addIPHeader(srcAddr, dstAddr tcpip.Address, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) { length := uint16(pkt.Size()) ip := header.IPv6(pkt.NetworkHeader().Push(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ @@ -384,8 +400,8 @@ func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params s NextHeader: uint8(params.Protocol), HopLimit: params.TTL, TrafficClass: params.TOS, - SrcAddr: r.LocalAddress, - DstAddr: r.RemoteAddress, + SrcAddr: srcAddr, + DstAddr: dstAddr, }) pkt.NetworkProtocolNumber = ProtocolNumber } @@ -440,7 +456,7 @@ func (e *endpoint) handleFragments(r *stack.Route, gso *stack.GSO, networkMTU ui // WritePacket writes a packet to the given destination address and protocol. func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.NetworkHeaderParams, pkt *stack.PacketBuffer) *tcpip.Error { - e.addIPHeader(r, pkt, params) + e.addIPHeader(r.LocalAddress, r.RemoteAddress, pkt, params) // iptables filtering. All packets that reach here are locally // generated. @@ -529,7 +545,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe linkMTU := e.nic.MTU() for pb := pkts.Front(); pb != nil; pb = pb.Next() { - e.addIPHeader(r, pb, params) + e.addIPHeader(r.LocalAddress, r.RemoteAddress, pb, params) networkMTU, err := calculateNetworkMTU(linkMTU, uint32(pb.NetworkHeader().View().Size())) if err != nil { @@ -737,8 +753,11 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { return } - addressEndpoint := e.AcquireAssignedAddress(dstAddr, e.nic.Promiscuous(), stack.CanBePrimaryEndpoint) - if addressEndpoint == nil { + // The destination address should be an address we own or a group we joined + // for us to receive the packet. Otherwise, attempt to forward the packet. + if addressEndpoint := e.AcquireAssignedAddress(dstAddr, e.nic.Promiscuous(), stack.CanBePrimaryEndpoint); addressEndpoint != nil { + addressEndpoint.DecRef() + } else if !e.IsInGroup(dstAddr) { if !e.protocol.Forwarding() { stats.IP.InvalidDestinationAddressesReceived.Increment() return @@ -747,7 +766,6 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { _ = e.forwardPacket(pkt) return } - addressEndpoint.DecRef() // vv consists of: // - Any IPv6 header bytes after the first 40 (i.e. extensions). @@ -1090,9 +1108,16 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { // // Which when taken together indicate that an unknown protocol should // be treated as an unrecognized next header value. + // The location of the Next Header field is in a different place in + // the initial IPv6 header than it is in the extension headers so + // treat it specially. + prevHdrIDOffset := uint32(header.IPv6NextHeaderOffset) + if previousHeaderStart != 0 { + prevHdrIDOffset = previousHeaderStart + } _ = e.protocol.returnError(&icmpReasonParameterProblem{ code: header.ICMPv6UnknownHeader, - pointer: it.ParseOffset(), + pointer: prevHdrIDOffset, }, pkt) default: panic(fmt.Sprintf("unrecognized result from DeliverTransportPacket = %d", res)) @@ -1100,12 +1125,11 @@ func (e *endpoint) handlePacket(pkt *stack.PacketBuffer) { } default: - _ = e.protocol.returnError(&icmpReasonParameterProblem{ - code: header.ICMPv6UnknownHeader, - pointer: it.ParseOffset(), - }, pkt) - stats.UnknownProtocolRcvdPackets.Increment() - return + // Since the iterator returns IPv6RawPayloadHeader for unknown Extension + // Header IDs this should never happen unless we missed a supported type + // here. + panic(fmt.Sprintf("unrecognized type from it.Next() = %T", extHdr)) + } } } @@ -1154,8 +1178,10 @@ func (e *endpoint) addAndAcquirePermanentAddressLocked(addr tcpip.AddressWithPre } snmc := header.SolicitedNodeAddr(addr.Address) - if _, err := e.mu.addressableEndpointState.JoinGroup(snmc); err != nil { - return nil, err + if err := e.joinGroupLocked(snmc); err != nil { + // joinGroupLocked only returns an error if the group address is not a valid + // IPv6 multicast address. + panic(fmt.Sprintf("e.joinGroupLocked(%s): %s", snmc, err)) } addressEndpoint.SetKind(stack.PermanentTentative) @@ -1211,7 +1237,8 @@ func (e *endpoint) removePermanentEndpointLocked(addressEndpoint stack.AddressEn } snmc := header.SolicitedNodeAddr(addr.Address) - if _, err := e.mu.addressableEndpointState.LeaveGroup(snmc); err != nil && err != tcpip.ErrBadLocalAddress { + // The endpoint may have already left the multicast group. + if err := e.leaveGroupLocked(snmc); err != nil && err != tcpip.ErrBadLocalAddress { return err } @@ -1234,7 +1261,7 @@ func (e *endpoint) hasPermanentAddressRLocked(addr tcpip.Address) bool { // // Precondition: e.mu must be read or write locked. func (e *endpoint) getAddressRLocked(localAddr tcpip.Address) stack.AddressEndpoint { - return e.mu.addressableEndpointState.ReadOnly().Lookup(localAddr) + return e.mu.addressableEndpointState.GetAddress(localAddr) } // MainAddress implements stack.AddressableEndpoint. @@ -1285,7 +1312,7 @@ func (e *endpoint) acquireOutgoingPrimaryAddressRLocked(remoteAddr tcpip.Address // Create a candidate set of available addresses we can potentially use as a // source address. var cs []addrCandidate - e.mu.addressableEndpointState.ReadOnly().ForEachPrimaryEndpoint(func(addressEndpoint stack.AddressEndpoint) { + e.mu.addressableEndpointState.ForEachPrimaryEndpoint(func(addressEndpoint stack.AddressEndpoint) { // If r is not valid for outgoing connections, it is not a valid endpoint. if !addressEndpoint.IsAssigned(allowExpired) { return @@ -1376,28 +1403,43 @@ func (e *endpoint) PermanentAddresses() []tcpip.AddressWithPrefix { } // JoinGroup implements stack.GroupAddressableEndpoint. -func (e *endpoint) JoinGroup(addr tcpip.Address) (bool, *tcpip.Error) { +func (e *endpoint) JoinGroup(addr tcpip.Address) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + return e.joinGroupLocked(addr) +} + +// joinGroupLocked is like JoinGroup but with locking requirements. +// +// Precondition: e.mu must be locked. +func (e *endpoint) joinGroupLocked(addr tcpip.Address) *tcpip.Error { if !header.IsV6MulticastAddress(addr) { - return false, tcpip.ErrBadAddress + return tcpip.ErrBadAddress } - e.mu.Lock() - defer e.mu.Unlock() - return e.mu.addressableEndpointState.JoinGroup(addr) + e.mld.joinGroup(addr) + return nil } // LeaveGroup implements stack.GroupAddressableEndpoint. -func (e *endpoint) LeaveGroup(addr tcpip.Address) (bool, *tcpip.Error) { +func (e *endpoint) LeaveGroup(addr tcpip.Address) *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - return e.mu.addressableEndpointState.LeaveGroup(addr) + return e.leaveGroupLocked(addr) +} + +// leaveGroupLocked is like LeaveGroup but with locking requirements. +// +// Precondition: e.mu must be locked. +func (e *endpoint) leaveGroupLocked(addr tcpip.Address) *tcpip.Error { + return e.mld.leaveGroup(addr) } // IsInGroup implements stack.GroupAddressableEndpoint. func (e *endpoint) IsInGroup(addr tcpip.Address) bool { e.mu.RLock() defer e.mu.RUnlock() - return e.mu.addressableEndpointState.IsInGroup(addr) + return e.mld.isInGroup(addr) } var _ stack.ForwardingNetworkProtocol = (*protocol)(nil) @@ -1405,7 +1447,8 @@ var _ stack.NetworkProtocol = (*protocol)(nil) var _ fragmentation.TimeoutHandler = (*protocol)(nil) type protocol struct { - stack *stack.Stack + stack *stack.Stack + options Options mu struct { sync.RWMutex @@ -1429,26 +1472,6 @@ type protocol struct { forwarding uint32 fragmentation *fragmentation.Fragmentation - - // ndpDisp is the NDP event dispatcher that is used to send the netstack - // integrator NDP related events. - ndpDisp NDPDispatcher - - // ndpConfigs is the default NDP configurations used by an IPv6 endpoint. - ndpConfigs NDPConfigurations - - // opaqueIIDOpts hold the options for generating opaque interface identifiers - // (IIDs) as outlined by RFC 7217. - opaqueIIDOpts OpaqueInterfaceIdentifierOptions - - // tempIIDSeed is used to seed the initial temporary interface identifier - // history value used to generate IIDs for temporary SLAAC addresses. - tempIIDSeed []byte - - // autoGenIPv6LinkLocal determines whether or not the stack attempts to - // auto-generate an IPv6 link-local address for newly enabled non-loopback - // NICs. See the AutoGenIPv6LinkLocal field of Options for more details. - autoGenIPv6LinkLocal bool } // Number returns the ipv6 protocol number. @@ -1484,13 +1507,14 @@ func (p *protocol) NewEndpoint(nic stack.NetworkInterface, linkAddrCache stack.L e.mu.addressableEndpointState.Init(e) e.mu.ndp = ndpState{ ep: e, - configs: p.ndpConfigs, + configs: p.options.NDPConfigs, dad: make(map[tcpip.Address]dadState), defaultRouters: make(map[tcpip.Address]defaultRouterState), onLinkPrefixes: make(map[tcpip.Subnet]onLinkPrefixState), slaacPrefixes: make(map[tcpip.Subnet]slaacPrefixState), } e.mu.ndp.initializeTempAddrState() + e.mld.init(e, p.options.MLD) p.mu.Lock() defer p.mu.Unlock() @@ -1613,17 +1637,17 @@ type Options struct { // NDPConfigs is the default NDP configurations used by interfaces. NDPConfigs NDPConfigurations - // AutoGenIPv6LinkLocal determines whether or not the stack attempts to - // auto-generate an IPv6 link-local address for newly enabled non-loopback + // AutoGenLinkLocal determines whether or not the stack attempts to + // auto-generate a link-local address for newly enabled non-loopback // NICs. // // Note, setting this to true does not mean that a link-local address is // assigned right away, or at all. If Duplicate Address Detection is enabled, // an address is only assigned if it successfully resolves. If it fails, no - // further attempts are made to auto-generate an IPv6 link-local adddress. + // further attempts are made to auto-generate a link-local adddress. // // The generated link-local address follows RFC 4291 Appendix A guidelines. - AutoGenIPv6LinkLocal bool + AutoGenLinkLocal bool // NDPDisp is the NDP event dispatcher that an integrator can provide to // receive NDP related events. @@ -1647,6 +1671,9 @@ type Options struct { // seed that is too small would reduce randomness and increase predictability, // defeating the purpose of temporary SLAAC addresses. TempIIDSeed []byte + + // MLD holds options for MLD. + MLD MLDOptions } // NewProtocolWithOptions returns an IPv6 network protocol. @@ -1658,15 +1685,11 @@ func NewProtocolWithOptions(opts Options) stack.NetworkProtocolFactory { return func(s *stack.Stack) stack.NetworkProtocol { p := &protocol{ - stack: s, + stack: s, + options: opts, + ids: ids, hashIV: hashIV, - - ndpDisp: opts.NDPDisp, - ndpConfigs: opts.NDPConfigs, - opaqueIIDOpts: opts.OpaqueIIDOpts, - tempIIDSeed: opts.TempIIDSeed, - autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, } p.fragmentation = fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, ReassembleTimeout, s.Clock(), p) p.mu.eps = make(map[*endpoint]struct{}) diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index a671d4bac..1c01f17ab 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -51,6 +51,7 @@ const ( fragmentExtHdrID = uint8(header.IPv6FragmentExtHdrIdentifier) destinationExtHdrID = uint8(header.IPv6DestinationOptionsExtHdrIdentifier) noNextHdrID = uint8(header.IPv6NoNextHeaderIdentifier) + unknownHdrID = uint8(header.IPv6UnknownExtHdrIdentifier) extraHeaderReserve = 50 ) @@ -79,7 +80,7 @@ func testReceiveICMP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst Data: hdr.View().ToVectorisedView(), })) - stats := s.Stats().ICMP.V6PacketsReceived + stats := s.Stats().ICMP.V6.PacketsReceived if got := stats.NeighborAdvert.Value(); got != want { t.Fatalf("got NeighborAdvert = %d, want = %d", got, want) @@ -573,6 +574,33 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { expectICMP: false, }, { + name: "unknown next header (first)", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 0, 63, 4, 1, 2, 3, 4, + }, unknownHdrID + }, + shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownHeader, + pointer: header.IPv6NextHeaderOffset, + }, + { + name: "unknown next header (not first)", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + unknownHdrID, 0, + 63, 4, 1, 2, 3, 4, + }, hopByHopExtHdrID + }, + shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownHeader, + pointer: header.IPv6FixedHeaderSize, + }, + { name: "destination with unknown option skippable action", extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{ @@ -755,11 +783,6 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { pointer: header.IPv6FixedHeaderSize, }, { - name: "No next header", - extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{}, noNextHdrID }, - shouldAccept: false, - }, - { name: "hopbyhop (with skippable unknown) - routing - atomic fragment - destination (with skippable unknown)", extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{ @@ -873,7 +896,13 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { Length: uint16(udpLength), }) copy(u.Payload(), udpPayload) - sum := header.PseudoHeaderChecksum(udp.ProtocolNumber, addr1, addr2, uint16(udpLength)) + + dstAddr := tcpip.Address(addr2) + if test.multicast { + dstAddr = header.IPv6AllNodesMulticastAddress + } + + sum := header.PseudoHeaderChecksum(udp.ProtocolNumber, addr1, dstAddr, uint16(udpLength)) sum = header.Checksum(udpPayload, sum) u.SetChecksum(^u.CalculateChecksum(sum)) @@ -884,10 +913,6 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { // Serialize IPv6 fixed header. payloadLength := hdr.UsedLength() ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) - dstAddr := tcpip.Address(addr2) - if test.multicast { - dstAddr = header.IPv6AllNodesMulticastAddress - } ip.Encode(&header.IPv6Fields{ PayloadLength: uint16(payloadLength), NextHeader: ipv6NextHdr, @@ -982,9 +1007,10 @@ func TestReceiveIPv6Fragments(t *testing.T) { udpPayload2Length = 128 // Used to test cases where the fragment blocks are not a multiple of // the fragment block size of 8 (RFC 8200 section 4.5). - udpPayload3Length = 127 - udpPayload4Length = header.IPv6MaximumPayloadSize - header.UDPMinimumSize - fragmentExtHdrLen = 8 + udpPayload3Length = 127 + udpPayload4Length = header.IPv6MaximumPayloadSize - header.UDPMinimumSize + udpMaximumSizeMinus15 = header.UDPMaximumSize - 15 + fragmentExtHdrLen = 8 // Note, not all routing extension headers will be 8 bytes but this test // uses 8 byte routing extension headers for most sub tests. routingExtHdrLen = 8 @@ -1328,14 +1354,14 @@ func TestReceiveIPv6Fragments(t *testing.T) { dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+65520, + fragmentExtHdrLen+udpMaximumSizeMinus15, []buffer.View{ // Fragment extension header. // // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload4Addr1ToAddr2[:65520], + ipv6Payload4Addr1ToAddr2[:udpMaximumSizeMinus15], }, ), }, @@ -1344,14 +1370,17 @@ func TestReceiveIPv6Fragments(t *testing.T) { dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload4Addr1ToAddr2)-65520, + fragmentExtHdrLen+len(ipv6Payload4Addr1ToAddr2)-udpMaximumSizeMinus15, []buffer.View{ // Fragment extension header. // - // Fragment offset = 8190, More = false, ID = 1 - buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 255, 240, 0, 0, 0, 1}), + // Fragment offset = udpMaximumSizeMinus15/8, More = false, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, + udpMaximumSizeMinus15 >> 8, + udpMaximumSizeMinus15 & 0xff, + 0, 0, 0, 1}), - ipv6Payload4Addr1ToAddr2[65520:], + ipv6Payload4Addr1ToAddr2[udpMaximumSizeMinus15:], }, ), }, @@ -1359,6 +1388,47 @@ func TestReceiveIPv6Fragments(t *testing.T) { expectedPayloads: [][]byte{udpPayload4Addr1ToAddr2}, }, { + name: "Two fragments with MF flag reassembled into a maximum UDP packet", + fragments: []fragmentData{ + { + srcAddr: addr1, + dstAddr: addr2, + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+udpMaximumSizeMinus15, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 0, More = true, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), + + ipv6Payload4Addr1ToAddr2[:udpMaximumSizeMinus15], + }, + ), + }, + { + srcAddr: addr1, + dstAddr: addr2, + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+len(ipv6Payload4Addr1ToAddr2)-udpMaximumSizeMinus15, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = udpMaximumSizeMinus15/8, More = true, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, + udpMaximumSizeMinus15 >> 8, + (udpMaximumSizeMinus15 & 0xff) + 1, + 0, 0, 0, 1}), + + ipv6Payload4Addr1ToAddr2[udpMaximumSizeMinus15:], + }, + ), + }, + }, + expectedPayloads: nil, + }, + { name: "Two fragments with per-fragment routing header with zero segments left", fragments: []fragmentData{ { @@ -2439,7 +2509,7 @@ func TestWriteStats(t *testing.T) { test.setup(t, rt.Stack()) - nWritten, _ := writer.writePackets(&rt, pkts) + nWritten, _ := writer.writePackets(rt, pkts) if got := int(rt.Stats().IP.PacketsSent.Value()); got != test.expectSent { t.Errorf("sent %d packets, but expected to send %d", got, test.expectSent) @@ -2456,7 +2526,7 @@ func TestWriteStats(t *testing.T) { } } -func buildRoute(t *testing.T, ep stack.LinkEndpoint) stack.Route { +func buildRoute(t *testing.T, ep stack.LinkEndpoint) *stack.Route { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) diff --git a/pkg/tcpip/network/ipv6/mld.go b/pkg/tcpip/network/ipv6/mld.go new file mode 100644 index 000000000..4c06b3f0c --- /dev/null +++ b/pkg/tcpip/network/ipv6/mld.go @@ -0,0 +1,164 @@ +// 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 ipv6 + +import ( + "fmt" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/network/ip" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +const ( + // UnsolicitedReportIntervalMax is the maximum delay between sending + // unsolicited MLD reports. + // + // Obtained from RFC 2710 Section 7.10. + UnsolicitedReportIntervalMax = 10 * time.Second +) + +// MLDOptions holds options for MLD. +type MLDOptions struct { + // Enabled indicates whether MLD will be performed. + // + // When enabled, MLD may transmit MLD report and done messages when + // joining and leaving multicast groups respectively, and handle incoming + // MLD packets. + Enabled bool +} + +var _ ip.MulticastGroupProtocol = (*mldState)(nil) + +// mldState is the per-interface MLD state. +// +// mldState.init MUST be called to initialize the MLD state. +type mldState struct { + // The IPv6 endpoint this mldState is for. + ep *endpoint + + genericMulticastProtocol ip.GenericMulticastProtocolState +} + +// SendReport implements ip.MulticastGroupProtocol. +func (mld *mldState) SendReport(groupAddress tcpip.Address) *tcpip.Error { + return mld.writePacket(groupAddress, groupAddress, header.ICMPv6MulticastListenerReport) +} + +// SendLeave implements ip.MulticastGroupProtocol. +func (mld *mldState) SendLeave(groupAddress tcpip.Address) *tcpip.Error { + return mld.writePacket(header.IPv6AllRoutersMulticastAddress, groupAddress, header.ICMPv6MulticastListenerDone) +} + +// init sets up an mldState struct, and is required to be called before using +// a new mldState. +func (mld *mldState) init(ep *endpoint, opts MLDOptions) { + mld.ep = ep + mld.genericMulticastProtocol.Init(ip.GenericMulticastProtocolOptions{ + Enabled: opts.Enabled, + Rand: ep.protocol.stack.Rand(), + Clock: ep.protocol.stack.Clock(), + Protocol: mld, + MaxUnsolicitedReportDelay: UnsolicitedReportIntervalMax, + AllNodesAddress: header.IPv6AllNodesMulticastAddress, + }) +} + +func (mld *mldState) handleMulticastListenerQuery(mldHdr header.MLD) { + mld.genericMulticastProtocol.HandleQuery(mldHdr.MulticastAddress(), mldHdr.MaximumResponseDelay()) +} + +func (mld *mldState) handleMulticastListenerReport(mldHdr header.MLD) { + mld.genericMulticastProtocol.HandleReport(mldHdr.MulticastAddress()) +} + +// joinGroup handles joining a new group and sending and scheduling the required +// messages. +// +// If the group is already joined, returns tcpip.ErrDuplicateAddress. +func (mld *mldState) joinGroup(groupAddress tcpip.Address) { + mld.genericMulticastProtocol.JoinGroup(groupAddress, !mld.ep.Enabled() /* dontInitialize */) +} + +// isInGroup returns true if the specified group has been joined locally. +func (mld *mldState) isInGroup(groupAddress tcpip.Address) bool { + return mld.genericMulticastProtocol.IsLocallyJoined(groupAddress) +} + +// leaveGroup handles removing the group from the membership map, cancels any +// delay timers associated with that group, and sends the Done message, if +// required. +func (mld *mldState) leaveGroup(groupAddress tcpip.Address) *tcpip.Error { + // LeaveGroup returns false only if the group was not joined. + if mld.genericMulticastProtocol.LeaveGroup(groupAddress) { + return nil + } + + return tcpip.ErrBadLocalAddress +} + +// softLeaveAll leaves all groups from the perspective of MLD, but remains +// joined locally. +func (mld *mldState) softLeaveAll() { + mld.genericMulticastProtocol.MakeAllNonMember() +} + +// initializeAll attemps to initialize the MLD state for each group that has +// been joined locally. +func (mld *mldState) initializeAll() { + mld.genericMulticastProtocol.InitializeGroups() +} + +func (mld *mldState) writePacket(destAddress, groupAddress tcpip.Address, mldType header.ICMPv6Type) *tcpip.Error { + sentStats := mld.ep.protocol.stack.Stats().ICMP.V6.PacketsSent + var mldStat *tcpip.StatCounter + switch mldType { + case header.ICMPv6MulticastListenerReport: + mldStat = sentStats.MulticastListenerReport + case header.ICMPv6MulticastListenerDone: + mldStat = sentStats.MulticastListenerDone + default: + panic(fmt.Sprintf("unrecognized mld type = %d", mldType)) + } + + icmp := header.ICMPv6(buffer.NewView(header.ICMPv6HeaderSize + header.MLDMinimumSize)) + icmp.SetType(mldType) + header.MLD(icmp.MessageBody()).SetMulticastAddress(groupAddress) + // TODO(gvisor.dev/issue/4888): We should not use the unspecified address, + // rather we should select an appropriate local address. + localAddress := header.IPv6Any + icmp.SetChecksum(header.ICMPv6Checksum(icmp, localAddress, destAddress, buffer.VectorisedView{})) + + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(mld.ep.MaxHeaderLength()), + Data: buffer.View(icmp).ToVectorisedView(), + }) + + mld.ep.addIPHeader(localAddress, destAddress, pkt, stack.NetworkHeaderParams{ + Protocol: header.ICMPv6ProtocolNumber, + TTL: header.MLDHopLimit, + }) + // TODO(b/162198658): set the ROUTER_ALERT option when sending Host + // Membership Reports. + if err := mld.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv6Address(destAddress), nil /* gso */, ProtocolNumber, pkt); err != nil { + sentStats.Dropped.Increment() + return err + } + mldStat.Increment() + return nil +} diff --git a/pkg/tcpip/network/ipv6/mld_test.go b/pkg/tcpip/network/ipv6/mld_test.go new file mode 100644 index 000000000..5677bdd54 --- /dev/null +++ b/pkg/tcpip/network/ipv6/mld_test.go @@ -0,0 +1,90 @@ +// 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 ipv6_test + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip/checker" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +const ( + addr1 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" +) + +func TestIPv6JoinLeaveSolicitedNodeAddressPerformsMLD(t *testing.T) { + const nicID = 1 + + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + MLD: ipv6.MLDOptions{ + Enabled: true, + }, + })}, + }) + e := channel.New(1, header.IPv6MinimumMTU, "") + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID, err) + } + + // The stack will join an address's solicited node multicast address when + // an address is added. An MLD report message should be sent for the + // solicited-node group. + if err := s.AddAddress(nicID, ipv6.ProtocolNumber, addr1); err != nil { + t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, ipv6.ProtocolNumber, addr1, err) + } + { + p, ok := e.Read() + if !ok { + t.Fatal("expected a report message to be sent") + } + snmc := header.SolicitedNodeAddr(addr1) + checker.IPv6(t, header.IPv6(stack.PayloadSince(p.Pkt.NetworkHeader())), + checker.DstAddr(snmc), + // Hop Limit for an MLD message must be 1 as per RFC 2710 section 3. + checker.TTL(1), + checker.MLD(header.ICMPv6MulticastListenerReport, header.MLDMinimumSize, + checker.MLDMaxRespDelay(0), + checker.MLDMulticastAddress(snmc), + ), + ) + } + + // The stack will leave an address's solicited node multicast address when + // an address is removed. An MLD done message should be sent for the + // solicited-node group. + if err := s.RemoveAddress(nicID, addr1); err != nil { + t.Fatalf("RemoveAddress(%d, %s) = %s", nicID, addr1, err) + } + { + p, ok := e.Read() + if !ok { + t.Fatal("expected a done message to be sent") + } + snmc := header.SolicitedNodeAddr(addr1) + checker.IPv6(t, header.IPv6(stack.PayloadSince(p.Pkt.NetworkHeader())), + checker.DstAddr(header.IPv6AllRoutersMulticastAddress), + checker.TTL(1), + checker.MLD(header.ICMPv6MulticastListenerDone, header.MLDMinimumSize, + checker.MLDMaxRespDelay(0), + checker.MLDMulticastAddress(snmc), + ), + ) + } +} diff --git a/pkg/tcpip/network/ipv6/ndp.go b/pkg/tcpip/network/ipv6/ndp.go index 40da011f8..8cb7d4dab 100644 --- a/pkg/tcpip/network/ipv6/ndp.go +++ b/pkg/tcpip/network/ipv6/ndp.go @@ -471,17 +471,8 @@ type ndpState struct { // The default routers discovered through Router Advertisements. defaultRouters map[tcpip.Address]defaultRouterState - rtrSolicit struct { - // The timer used to send the next router solicitation message. - timer tcpip.Timer - - // Used to let the Router Solicitation timer know that it has been stopped. - // - // Must only be read from or written to while protected by the lock of - // the IPv6 endpoint this ndpState is associated with. MUST be set when the - // timer is set. - done *bool - } + // The job used to send the next router solicitation message. + rtrSolicitJob *tcpip.Job // The on-link prefixes discovered through Router Advertisements' Prefix // Information option. @@ -507,7 +498,7 @@ 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 tcpip.Timer + job *tcpip.Job // Used to let the DAD timer know that it has been stopped. // @@ -648,96 +639,70 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, addressE // Consider DAD to have resolved even if no DAD messages were actually // transmitted. - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, true, nil) } return nil } - var done bool - var timer tcpip.Timer - // We initially start a timer to fire immediately because some of the DAD work - // cannot be done while holding the IPv6 endpoint'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 = ndp.ep.protocol.stack.Clock().AfterFunc(0, func() { - ndp.ep.mu.Lock() - defer ndp.ep.mu.Unlock() - - if done { - // If we reach this point, it means that the DAD timer fired after - // another goroutine already obtained the IPv6 endpoint lock and stopped - // DAD before this function obtained the NIC lock. Simply return here and - // do nothing further. - return - } - - if addressEndpoint.GetKind() != stack.PermanentTentative { - // The endpoint should still be marked as tentative since we are still - // performing DAD on it. - panic(fmt.Sprintf("ndpdad: addr %s is no longer tentative on NIC(%d)", addr, ndp.ep.nic.ID())) - } + state := dadState{ + job: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { + state, ok := ndp.dad[addr] + if !ok { + panic(fmt.Sprintf("ndpdad: DAD timer fired but missing state for %s on NIC(%d)", addr, ndp.ep.nic.ID())) + } - dadDone := remaining == 0 - - var err *tcpip.Error - if !dadDone { - // Use the unspecified address as the source address when performing DAD. - addressEndpoint := ndp.ep.acquireAddressOrCreateTempLocked(header.IPv6Any, true /* createTemp */, stack.NeverPrimaryEndpoint) - - // Do not hold the lock when sending packets which may be a long running - // task or may block link address resolution. We know this is safe - // because immediately after obtaining the lock again, we check if DAD - // has been stopped before doing any work with the IPv6 endpoint. Note, - // DAD would be stopped if the IPv6 endpoint was disabled or closed, or if - // the address was removed. - ndp.ep.mu.Unlock() - err = ndp.sendDADPacket(addr, addressEndpoint) - ndp.ep.mu.Lock() - addressEndpoint.DecRef() - } + if addressEndpoint.GetKind() != stack.PermanentTentative { + // The endpoint should still be marked as tentative since we are still + // performing DAD on it. + panic(fmt.Sprintf("ndpdad: addr %s is no longer tentative on NIC(%d)", addr, ndp.ep.nic.ID())) + } - if done { - // If we reach this point, it means that DAD was stopped after we released - // the IPv6 endpoint's read lock and before we obtained the write lock. - return - } + dadDone := remaining == 0 - if dadDone { - // DAD has resolved. - addressEndpoint.SetKind(stack.Permanent) - } else if err == nil { - // DAD is not done and we had no errors when sending the last NDP NS, - // schedule the next DAD timer. - remaining-- - timer.Reset(ndp.configs.RetransmitTimer) - return - } + var err *tcpip.Error + if !dadDone { + err = ndp.sendDADPacket(addr, addressEndpoint) + } - // At this point we know that either DAD is done or we hit an error sending - // the last NDP NS. Either way, clean up addr's DAD state and let the - // integrator know DAD has completed. - delete(ndp.dad, addr) + if dadDone { + // DAD has resolved. + addressEndpoint.SetKind(stack.Permanent) + } else if err == nil { + // DAD is not done and we had no errors when sending the last NDP NS, + // schedule the next DAD timer. + remaining-- + state.job.Schedule(ndp.configs.RetransmitTimer) + return + } - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { - ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, dadDone, err) - } + // At this point we know that either DAD is done or we hit an error + // sending the last NDP NS. Either way, clean up addr's DAD state and let + // the integrator know DAD has completed. + delete(ndp.dad, addr) - // If DAD resolved for a stable SLAAC address, attempt generation of a - // temporary SLAAC address. - if dadDone && addressEndpoint.ConfigType() == stack.AddressConfigSlaac { - // Reset the generation attempts counter as we are starting the generation - // of a new address for the SLAAC prefix. - ndp.regenerateTempSLAACAddr(addressEndpoint.AddressWithPrefix().Subnet(), true /* resetGenAttempts */) - } - }) + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { + ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, dadDone, err) + } - ndp.dad[addr] = dadState{ - timer: timer, - done: &done, + // If DAD resolved for a stable SLAAC address, attempt generation of a + // temporary SLAAC address. + if dadDone && addressEndpoint.ConfigType() == stack.AddressConfigSlaac { + // Reset the generation attempts counter as we are starting the generation + // of a new address for the SLAAC prefix. + ndp.regenerateTempSLAACAddr(addressEndpoint.AddressWithPrefix().Subnet(), true /* resetGenAttempts */) + } + }), } + // We initially start a timer to fire immediately because some of the DAD work + // cannot be done while holding the IPv6 endpoint'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. + state.job.Schedule(0) + ndp.dad[addr] = state + return nil } @@ -745,55 +710,31 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, addressE // addr. // // addr must be a tentative IPv6 address on ndp's IPv6 endpoint. -// -// The IPv6 endpoint that ndp belongs to MUST NOT be locked. func (ndp *ndpState) sendDADPacket(addr tcpip.Address, addressEndpoint stack.AddressEndpoint) *tcpip.Error { snmc := header.SolicitedNodeAddr(addr) - r, err := ndp.ep.protocol.stack.FindRoute(ndp.ep.nic.ID(), header.IPv6Any, snmc, ProtocolNumber, false /* multicastLoop */) - if err != nil { - return err - } - defer r.Release() - - // Route should resolve immediately since snmc is a multicast address so a - // remote link address can be calculated without a resolution process. - if c, err := r.Resolve(nil); err != nil { - // Do not consider the NIC being unknown or disabled as a fatal error. - // Since this method is required to be called when the IPv6 endpoint is not - // locked, the NIC could have been disabled or removed by another goroutine. - if err == tcpip.ErrUnknownNICID || err != tcpip.ErrInvalidEndpointState { - return err - } - - panic(fmt.Sprintf("ndp: error when resolving route to send NDP NS for DAD (%s -> %s on NIC(%d)): %s", header.IPv6Any, snmc, ndp.ep.nic.ID(), err)) - } else if c != nil { - panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP NS for DAD (%s -> %s on NIC(%d))", header.IPv6Any, snmc, ndp.ep.nic.ID())) - } - - icmpData := header.ICMPv6(buffer.NewView(header.ICMPv6NeighborSolicitMinimumSize)) - icmpData.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(icmpData.NDPPayload()) + icmp := header.ICMPv6(buffer.NewView(header.ICMPv6NeighborSolicitMinimumSize)) + icmp.SetType(header.ICMPv6NeighborSolicit) + ns := header.NDPNeighborSolicit(icmp.MessageBody()) ns.SetTargetAddress(addr) - icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + icmp.SetChecksum(header.ICMPv6Checksum(icmp, header.IPv6Any, snmc, buffer.VectorisedView{})) pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - ReserveHeaderBytes: int(r.MaxHeaderLength()), - Data: buffer.View(icmpData).ToVectorisedView(), + ReserveHeaderBytes: int(ndp.ep.MaxHeaderLength()), + Data: buffer.View(icmp).ToVectorisedView(), }) - sent := r.Stats().ICMP.V6PacketsSent - if err := r.WritePacket(nil, - stack.NetworkHeaderParams{ - Protocol: header.ICMPv6ProtocolNumber, - TTL: header.NDPHopLimit, - }, pkt, - ); err != nil { + sent := ndp.ep.protocol.stack.Stats().ICMP.V6.PacketsSent + ndp.ep.addIPHeader(header.IPv6Any, snmc, pkt, stack.NetworkHeaderParams{ + Protocol: header.ICMPv6ProtocolNumber, + TTL: header.NDPHopLimit, + }) + + if err := ndp.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv6Address(snmc), nil /* gso */, ProtocolNumber, pkt); err != nil { sent.Dropped.Increment() return err } sent.NeighborSolicit.Increment() - return nil } @@ -812,18 +753,11 @@ func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) { return } - if dad.timer != nil { - dad.timer.Stop() - dad.timer = nil - - *dad.done = true - dad.done = nil - } - + dad.job.Cancel() delete(ndp.dad, addr) // Let the integrator know DAD did not resolve. - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, false, nil) } } @@ -846,7 +780,7 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { // Only worry about the DHCPv6 configuration if we have an NDPDispatcher as we // only inform the dispatcher on configuration changes. We do nothing else // with the information. - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { var configuration DHCPv6ConfigurationFromNDPRA switch { case ra.ManagedAddrConfFlag(): @@ -903,20 +837,20 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { for opt, done, _ := it.Next(); !done; opt, done, _ = it.Next() { switch opt := opt.(type) { case header.NDPRecursiveDNSServer: - if ndp.ep.protocol.ndpDisp == nil { + if ndp.ep.protocol.options.NDPDisp == nil { continue } addrs, _ := opt.Addresses() - ndp.ep.protocol.ndpDisp.OnRecursiveDNSServerOption(ndp.ep.nic.ID(), addrs, opt.Lifetime()) + ndp.ep.protocol.options.NDPDisp.OnRecursiveDNSServerOption(ndp.ep.nic.ID(), addrs, opt.Lifetime()) case header.NDPDNSSearchList: - if ndp.ep.protocol.ndpDisp == nil { + if ndp.ep.protocol.options.NDPDisp == nil { continue } domainNames, _ := opt.DomainNames() - ndp.ep.protocol.ndpDisp.OnDNSSearchListOption(ndp.ep.nic.ID(), domainNames, opt.Lifetime()) + ndp.ep.protocol.options.NDPDisp.OnDNSSearchListOption(ndp.ep.nic.ID(), domainNames, opt.Lifetime()) case header.NDPPrefixInformation: prefix := opt.Subnet() @@ -964,7 +898,7 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { delete(ndp.defaultRouters, ip) // Let the integrator know a discovered default router is invalidated. - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { ndpDisp.OnDefaultRouterInvalidated(ndp.ep.nic.ID(), ip) } } @@ -976,7 +910,7 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { // // The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { - ndpDisp := ndp.ep.protocol.ndpDisp + ndpDisp := ndp.ep.protocol.options.NDPDisp if ndpDisp == nil { return } @@ -1006,7 +940,7 @@ func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { // // The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) { - ndpDisp := ndp.ep.protocol.ndpDisp + ndpDisp := ndp.ep.protocol.options.NDPDisp if ndpDisp == nil { return } @@ -1047,7 +981,7 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { delete(ndp.onLinkPrefixes, prefix) // Let the integrator know a discovered on-link prefix is invalidated. - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { ndpDisp.OnOnLinkPrefixInvalidated(ndp.ep.nic.ID(), prefix) } } @@ -1225,7 +1159,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { // The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) addAndAcquireSLAACAddr(addr tcpip.AddressWithPrefix, configType stack.AddressConfigType, deprecated bool) stack.AddressEndpoint { // Inform the integrator that we have a new SLAAC address. - ndpDisp := ndp.ep.protocol.ndpDisp + ndpDisp := ndp.ep.protocol.options.NDPDisp if ndpDisp == nil { return nil } @@ -1272,7 +1206,7 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt } dadCounter := state.generationAttempts + state.stableAddr.localGenerationFailures - if oIID := ndp.ep.protocol.opaqueIIDOpts; oIID.NICNameFromID != nil { + if oIID := ndp.ep.protocol.options.OpaqueIIDOpts; oIID.NICNameFromID != nil { addrBytes = header.AppendOpaqueInterfaceIdentifier( addrBytes[:header.IIDOffsetInIPv6Address], prefix, @@ -1676,7 +1610,7 @@ func (ndp *ndpState) deprecateSLAACAddress(addressEndpoint stack.AddressEndpoint } addressEndpoint.SetDeprecated(true) - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { ndpDisp.OnAutoGenAddressDeprecated(ndp.ep.nic.ID(), addressEndpoint.AddressWithPrefix()) } } @@ -1701,7 +1635,7 @@ func (ndp *ndpState) invalidateSLAACPrefix(prefix tcpip.Subnet, state slaacPrefi // // The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPrefix, invalidatePrefix bool) { - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { ndpDisp.OnAutoGenAddressInvalidated(ndp.ep.nic.ID(), addr) } @@ -1761,7 +1695,7 @@ func (ndp *ndpState) invalidateTempSLAACAddr(tempAddrs map[tcpip.Address]tempSLA // // The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPrefix, invalidateAddr bool) { - if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.options.NDPDisp; ndpDisp != nil { ndpDisp.OnAutoGenAddressInvalidated(ndp.ep.nic.ID(), addr) } @@ -1859,7 +1793,7 @@ func (ndp *ndpState) cleanupState(hostOnly bool) { // // The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) startSolicitingRouters() { - if ndp.rtrSolicit.timer != nil { + if ndp.rtrSolicitJob != nil { // We are already soliciting routers. return } @@ -1876,56 +1810,14 @@ func (ndp *ndpState) startSolicitingRouters() { delay = time.Duration(rand.Int63n(int64(ndp.configs.MaxRtrSolicitationDelay))) } - var done bool - ndp.rtrSolicit.done = &done - ndp.rtrSolicit.timer = ndp.ep.protocol.stack.Clock().AfterFunc(delay, func() { - ndp.ep.mu.Lock() - if done { - // If we reach this point, it means that the RS timer fired after another - // goroutine already obtained the IPv6 endpoint lock and stopped - // solicitations. Simply return here and do nothing further. - ndp.ep.mu.Unlock() - return - } - + ndp.rtrSolicitJob = ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { // As per RFC 4861 section 4.1, the source of the RS is an address assigned // to the sending interface, or the unspecified address if no address is // assigned to the sending interface. - addressEndpoint := ndp.ep.acquireOutgoingPrimaryAddressRLocked(header.IPv6AllRoutersMulticastAddress, false) - if addressEndpoint == nil { - // Incase this ends up creating a new temporary address, we need to hold - // onto the endpoint until a route is obtained. If we decrement the - // reference count before obtaing a route, the address's resources would - // be released and attempting to obtain a route after would fail. Once a - // route is obtainted, it is safe to decrement the reference count since - // obtaining a route increments the address's reference count. - addressEndpoint = ndp.ep.acquireAddressOrCreateTempLocked(header.IPv6Any, true /* createTemp */, stack.NeverPrimaryEndpoint) - } - ndp.ep.mu.Unlock() - - localAddr := addressEndpoint.AddressWithPrefix().Address - r, err := ndp.ep.protocol.stack.FindRoute(ndp.ep.nic.ID(), localAddr, header.IPv6AllRoutersMulticastAddress, ProtocolNumber, false /* multicastLoop */) - addressEndpoint.DecRef() - if err != nil { - return - } - defer r.Release() - - // Route should resolve immediately since - // header.IPv6AllRoutersMulticastAddress is a multicast address so a - // remote link address can be calculated without a resolution process. - if c, err := r.Resolve(nil); err != nil { - // Do not consider the NIC being unknown or disabled as a fatal error. - // Since this method is required to be called when the IPv6 endpoint is - // not locked, the IPv6 endpoint could have been disabled or removed by - // another goroutine. - if err == tcpip.ErrUnknownNICID || err == tcpip.ErrInvalidEndpointState { - return - } - - panic(fmt.Sprintf("ndp: error when resolving route to send NDP RS (%s -> %s on NIC(%d)): %s", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.ep.nic.ID(), err)) - } else if c != nil { - panic(fmt.Sprintf("ndp: route resolution not immediate for route to send NDP RS (%s -> %s on NIC(%d))", header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.ep.nic.ID())) + localAddr := header.IPv6Any + if addressEndpoint := ndp.ep.acquireOutgoingPrimaryAddressRLocked(header.IPv6AllRoutersMulticastAddress, false); addressEndpoint != nil { + localAddr = addressEndpoint.AddressWithPrefix().Address + addressEndpoint.DecRef() } // As per RFC 4861 section 4.1, an NDP RS SHOULD include the source @@ -1936,30 +1828,31 @@ func (ndp *ndpState) startSolicitingRouters() { // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by // LinkEndpoint.LinkAddress) before reaching this point. var optsSerializer header.NDPOptionsSerializer - if localAddr != header.IPv6Any && header.IsValidUnicastEthernetAddress(r.LocalLinkAddress) { + linkAddress := ndp.ep.nic.LinkAddress() + if localAddr != header.IPv6Any && header.IsValidUnicastEthernetAddress(linkAddress) { optsSerializer = header.NDPOptionsSerializer{ - header.NDPSourceLinkLayerAddressOption(r.LocalLinkAddress), + header.NDPSourceLinkLayerAddressOption(linkAddress), } } payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize + int(optsSerializer.Length()) icmpData := header.ICMPv6(buffer.NewView(payloadSize)) icmpData.SetType(header.ICMPv6RouterSolicit) - rs := header.NDPRouterSolicit(icmpData.NDPPayload()) + rs := header.NDPRouterSolicit(icmpData.MessageBody()) rs.Options().Serialize(optsSerializer) - icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, localAddr, header.IPv6AllRoutersMulticastAddress, buffer.VectorisedView{})) pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - ReserveHeaderBytes: int(r.MaxHeaderLength()), + ReserveHeaderBytes: int(ndp.ep.MaxHeaderLength()), Data: buffer.View(icmpData).ToVectorisedView(), }) - sent := r.Stats().ICMP.V6PacketsSent - if err := r.WritePacket(nil, - stack.NetworkHeaderParams{ - Protocol: header.ICMPv6ProtocolNumber, - TTL: header.NDPHopLimit, - }, pkt, - ); err != nil { + sent := ndp.ep.protocol.stack.Stats().ICMP.V6.PacketsSent + ndp.ep.addIPHeader(localAddr, header.IPv6AllRoutersMulticastAddress, pkt, stack.NetworkHeaderParams{ + Protocol: header.ICMPv6ProtocolNumber, + TTL: header.NDPHopLimit, + }) + + if err := ndp.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv6Address(header.IPv6AllRoutersMulticastAddress), nil /* gso */, ProtocolNumber, pkt); err != nil { sent.Dropped.Increment() log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.ep.nic.ID(), err) // Don't send any more messages if we had an error. @@ -1969,21 +1862,12 @@ func (ndp *ndpState) startSolicitingRouters() { remaining-- } - ndp.ep.mu.Lock() - if done || remaining == 0 { - ndp.rtrSolicit.timer = nil - ndp.rtrSolicit.done = nil - } else if ndp.rtrSolicit.timer != nil { - // Note, we need to explicitly check to make sure that - // the timer field is not nil because if it was nil but - // we still reached this point, then we know the IPv6 endpoint - // was requested to stop soliciting routers so we don't - // need to send the next Router Solicitation message. - ndp.rtrSolicit.timer.Reset(ndp.configs.RtrSolicitationInterval) + if remaining != 0 { + ndp.rtrSolicitJob.Schedule(ndp.configs.RtrSolicitationInterval) } - ndp.ep.mu.Unlock() }) + ndp.rtrSolicitJob.Schedule(delay) } // stopSolicitingRouters stops soliciting routers. If routers are not currently @@ -1991,21 +1875,19 @@ func (ndp *ndpState) startSolicitingRouters() { // // The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) stopSolicitingRouters() { - if ndp.rtrSolicit.timer == nil { + if ndp.rtrSolicitJob == nil { // Nothing to do. return } - *ndp.rtrSolicit.done = true - ndp.rtrSolicit.timer.Stop() - ndp.rtrSolicit.timer = nil - ndp.rtrSolicit.done = nil + ndp.rtrSolicitJob.Cancel() + ndp.rtrSolicitJob = nil } // initializeTempAddrState initializes state related to temporary SLAAC // addresses. func (ndp *ndpState) initializeTempAddrState() { - header.InitialTempIID(ndp.temporaryIIDHistory[:], ndp.ep.protocol.tempIIDSeed, ndp.ep.nic.ID()) + header.InitialTempIID(ndp.temporaryIIDHistory[:], ndp.ep.protocol.options.TempIIDSeed, ndp.ep.nic.ID()) if MaxDesyncFactor != 0 { ndp.temporaryAddressDesyncFactor = time.Duration(rand.Int63n(int64(MaxDesyncFactor))) diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index 37e8b1083..95c626bb8 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -205,7 +205,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOption(t *testing.T) { hdr := buffer.NewPrependable(header.IPv6MinimumSize + ndpNSSize) pkt := header.ICMPv6(hdr.Prepend(ndpNSSize)) pkt.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + ns := header.NDPNeighborSolicit(pkt.MessageBody()) ns.SetTargetAddress(lladdr0) opts := ns.Options() copy(opts, test.optsBuf) @@ -220,7 +220,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOption(t *testing.T) { DstAddr: lladdr0, }) - invalid := s.Stats().ICMP.V6PacketsReceived.Invalid + invalid := s.Stats().ICMP.V6.PacketsReceived.Invalid // Invalid count should initially be 0. if got := invalid.Value(); got != 0 { @@ -311,7 +311,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOptionUsingNeighborCache(t *testi hdr := buffer.NewPrependable(header.IPv6MinimumSize + ndpNSSize) pkt := header.ICMPv6(hdr.Prepend(ndpNSSize)) pkt.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + ns := header.NDPNeighborSolicit(pkt.MessageBody()) ns.SetTargetAddress(lladdr0) opts := ns.Options() copy(opts, test.optsBuf) @@ -326,16 +326,16 @@ func TestNeighorSolicitationWithSourceLinkLayerOptionUsingNeighborCache(t *testi DstAddr: lladdr0, }) - invalid := s.Stats().ICMP.V6PacketsReceived.Invalid + invalid := s.Stats().ICMP.V6.PacketsReceived.Invalid // Invalid count should initially be 0. if got := invalid.Value(); got != 0 { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) neighbors, err := s.Neighbors(nicID) if err != nil { @@ -591,7 +591,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { hdr := buffer.NewPrependable(header.IPv6MinimumSize + ndpNSSize) pkt := header.ICMPv6(hdr.Prepend(ndpNSSize)) pkt.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + ns := header.NDPNeighborSolicit(pkt.MessageBody()) ns.SetTargetAddress(nicAddr) opts := ns.Options() opts.Serialize(test.nsOpts) @@ -606,7 +606,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { DstAddr: test.nsDst, }) - invalid := s.Stats().ICMP.V6PacketsReceived.Invalid + invalid := s.Stats().ICMP.V6.PacketsReceived.Invalid // Invalid count should initially be 0. if got := invalid.Value(); got != 0 { @@ -650,8 +650,8 @@ func TestNeighorSolicitationResponse(t *testing.T) { if p.Route.RemoteAddress != respNSDst { t.Errorf("got p.Route.RemoteAddress = %s, want = %s", p.Route.RemoteAddress, respNSDst) } - if want := header.EthernetAddressFromMulticastIPv6Address(respNSDst); p.Route.RemoteLinkAddress != want { - t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, want) + if got, want := p.Route.RemoteLinkAddress(), header.EthernetAddressFromMulticastIPv6Address(respNSDst); got != want { + t.Errorf("got p.Route.RemoteLinkAddress() = %s, want = %s", got, want) } checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), @@ -672,7 +672,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { hdr := buffer.NewPrependable(header.IPv6MinimumSize + ndpNASize) pkt := header.ICMPv6(hdr.Prepend(ndpNASize)) pkt.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(pkt.NDPPayload()) + na := header.NDPNeighborAdvert(pkt.MessageBody()) na.SetSolicitedFlag(true) na.SetOverrideFlag(true) na.SetTargetAddress(test.nsSrc) @@ -706,8 +706,8 @@ func TestNeighorSolicitationResponse(t *testing.T) { if p.Route.RemoteAddress != test.naDst { t.Errorf("got p.Route.RemoteAddress = %s, want = %s", p.Route.RemoteAddress, test.naDst) } - if p.Route.RemoteLinkAddress != test.naDstLinkAddr { - t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, test.naDstLinkAddr) + if got := p.Route.RemoteLinkAddress(); got != test.naDstLinkAddr { + t.Errorf("got p.Route.RemoteLinkAddress() = %s, want = %s", got, test.naDstLinkAddr) } checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), @@ -777,7 +777,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOption(t *testing.T) { hdr := buffer.NewPrependable(header.IPv6MinimumSize + ndpNASize) pkt := header.ICMPv6(hdr.Prepend(ndpNASize)) pkt.SetType(header.ICMPv6NeighborAdvert) - ns := header.NDPNeighborAdvert(pkt.NDPPayload()) + ns := header.NDPNeighborAdvert(pkt.MessageBody()) ns.SetTargetAddress(lladdr1) opts := ns.Options() copy(opts, test.optsBuf) @@ -792,7 +792,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOption(t *testing.T) { DstAddr: lladdr0, }) - invalid := s.Stats().ICMP.V6PacketsReceived.Invalid + invalid := s.Stats().ICMP.V6.PacketsReceived.Invalid // Invalid count should initially be 0. if got := invalid.Value(); got != 0 { @@ -890,7 +890,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOptionUsingNeighborCache(t *test hdr := buffer.NewPrependable(header.IPv6MinimumSize + ndpNASize) pkt := header.ICMPv6(hdr.Prepend(ndpNASize)) pkt.SetType(header.ICMPv6NeighborAdvert) - ns := header.NDPNeighborAdvert(pkt.NDPPayload()) + ns := header.NDPNeighborAdvert(pkt.MessageBody()) ns.SetTargetAddress(lladdr1) opts := ns.Options() copy(opts, test.optsBuf) @@ -905,16 +905,16 @@ func TestNeighorAdvertisementWithTargetLinkLayerOptionUsingNeighborCache(t *test DstAddr: lladdr0, }) - invalid := s.Stats().ICMP.V6PacketsReceived.Invalid + invalid := s.Stats().ICMP.V6.PacketsReceived.Invalid // Invalid count should initially be 0. if got := invalid.Value(); got != 0 { t.Fatalf("got invalid = %d, want = 0", got) } - e.InjectInbound(ProtocolNumber, &stack.PacketBuffer{ + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: hdr.View().ToVectorisedView(), - }) + })) neighbors, err := s.Neighbors(nicID) if err != nil { @@ -1122,7 +1122,7 @@ func TestNDPValidation(t *testing.T) { s.SetForwarding(ProtocolNumber, true) } - stats := s.Stats().ICMP.V6PacketsReceived + stats := s.Stats().ICMP.V6.PacketsReceived invalid := stats.Invalid routerOnly := stats.RouterOnlyPacketsDroppedByHost typStat := typ.statCounter(stats) @@ -1346,7 +1346,7 @@ func TestRouterAdvertValidation(t *testing.T) { pkt := header.ICMPv6(hdr.Prepend(icmpSize)) pkt.SetType(header.ICMPv6RouterAdvert) pkt.SetCode(test.code) - copy(pkt.NDPPayload(), test.ndpPayload) + copy(pkt.MessageBody(), test.ndpPayload) payloadLength := hdr.UsedLength() pkt.SetChecksum(header.ICMPv6Checksum(pkt, test.src, header.IPv6AllNodesMulticastAddress, buffer.VectorisedView{})) ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) @@ -1358,7 +1358,7 @@ func TestRouterAdvertValidation(t *testing.T) { DstAddr: header.IPv6AllNodesMulticastAddress, }) - stats := s.Stats().ICMP.V6PacketsReceived + stats := s.Stats().ICMP.V6.PacketsReceived invalid := stats.Invalid rxRA := stats.RouterAdvert diff --git a/pkg/tcpip/network/multicast_group_test.go b/pkg/tcpip/network/multicast_group_test.go new file mode 100644 index 000000000..95fb67986 --- /dev/null +++ b/pkg/tcpip/network/multicast_group_test.go @@ -0,0 +1,1069 @@ +// 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 ip_test + +import ( + "fmt" + "strings" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/checker" + "gvisor.dev/gvisor/pkg/tcpip/faketime" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +const ( + linkAddr = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") + + ipv4MulticastAddr1 = tcpip.Address("\xe0\x00\x00\x03") + ipv4MulticastAddr2 = tcpip.Address("\xe0\x00\x00\x04") + ipv4MulticastAddr3 = tcpip.Address("\xe0\x00\x00\x05") + ipv6MulticastAddr1 = tcpip.Address("\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03") + ipv6MulticastAddr2 = tcpip.Address("\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x04") + ipv6MulticastAddr3 = tcpip.Address("\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x05") + + igmpMembershipQuery = uint8(header.IGMPMembershipQuery) + igmpv1MembershipReport = uint8(header.IGMPv1MembershipReport) + igmpv2MembershipReport = uint8(header.IGMPv2MembershipReport) + igmpLeaveGroup = uint8(header.IGMPLeaveGroup) + mldQuery = uint8(header.ICMPv6MulticastListenerQuery) + mldReport = uint8(header.ICMPv6MulticastListenerReport) + mldDone = uint8(header.ICMPv6MulticastListenerDone) +) + +var ( + // unsolicitedIGMPReportIntervalMaxTenthSec is the maximum amount of time the + // NIC will wait before sending an unsolicited report after joining a + // multicast group, in deciseconds. + unsolicitedIGMPReportIntervalMaxTenthSec = func() uint8 { + const decisecond = time.Second / 10 + if ipv4.UnsolicitedReportIntervalMax%decisecond != 0 { + panic(fmt.Sprintf("UnsolicitedReportIntervalMax of %d is a lossy conversion to deciseconds", ipv4.UnsolicitedReportIntervalMax)) + } + return uint8(ipv4.UnsolicitedReportIntervalMax / decisecond) + }() +) + +// validateMLDPacket checks that a passed PacketInfo is an IPv6 MLD packet +// sent to the provided address with the passed fields set. +func validateMLDPacket(t *testing.T, p channel.PacketInfo, remoteAddress tcpip.Address, mldType uint8, maxRespTime byte, groupAddress tcpip.Address) { + t.Helper() + + payload := header.IPv6(stack.PayloadSince(p.Pkt.NetworkHeader())) + checker.IPv6(t, payload, + checker.DstAddr(remoteAddress), + // Hop Limit for an MLD message must be 1 as per RFC 2710 section 3. + checker.TTL(1), + checker.MLD(header.ICMPv6Type(mldType), header.MLDMinimumSize, + checker.MLDMaxRespDelay(time.Duration(maxRespTime)*time.Millisecond), + checker.MLDMulticastAddress(groupAddress), + ), + ) +} + +// validateIGMPPacket checks that a passed PacketInfo is an IPv4 IGMP packet +// sent to the provided address with the passed fields set. +func validateIGMPPacket(t *testing.T, p channel.PacketInfo, remoteAddress tcpip.Address, igmpType uint8, maxRespTime byte, groupAddress tcpip.Address) { + t.Helper() + + payload := header.IPv4(stack.PayloadSince(p.Pkt.NetworkHeader())) + checker.IPv4(t, payload, + checker.DstAddr(remoteAddress), + // TTL for an IGMP message must be 1 as per RFC 2236 section 2. + checker.TTL(1), + checker.IPv4RouterAlert(), + checker.IGMP( + checker.IGMPType(header.IGMPType(igmpType)), + checker.IGMPMaxRespTime(header.DecisecondToDuration(maxRespTime)), + checker.IGMPGroupAddress(groupAddress), + ), + ) +} + +func createStack(t *testing.T, mgpEnabled bool) (*channel.Endpoint, *stack.Stack, *faketime.ManualClock) { + t.Helper() + + // Create an endpoint of queue size 2, since no more than 2 packets are ever + // queued in the tests in this file. + e := channel.New(2, 1280, linkAddr) + clock := faketime.NewManualClock() + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ + ipv4.NewProtocolWithOptions(ipv4.Options{ + IGMP: ipv4.IGMPOptions{ + Enabled: mgpEnabled, + }, + }), + ipv6.NewProtocolWithOptions(ipv6.Options{ + MLD: ipv6.MLDOptions{ + Enabled: mgpEnabled, + }, + }), + }, + Clock: clock, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + + return e, s, clock +} + +// createAndInjectIGMPPacket creates and injects an IGMP packet with the +// specified fields. +// +// Note, the router alert option is not included in this packet. +// +// TODO(b/162198658): set the router alert option. +func createAndInjectIGMPPacket(e *channel.Endpoint, igmpType byte, maxRespTime byte, groupAddress tcpip.Address) { + buf := buffer.NewView(header.IPv4MinimumSize + header.IGMPQueryMinimumSize) + + ip := header.IPv4(buf) + ip.Encode(&header.IPv4Fields{ + TotalLength: uint16(len(buf)), + TTL: header.IGMPTTL, + Protocol: uint8(header.IGMPProtocolNumber), + SrcAddr: header.IPv4Any, + DstAddr: header.IPv4AllSystems, + }) + ip.SetChecksum(^ip.CalculateChecksum()) + + igmp := header.IGMP(buf[header.IPv4MinimumSize:]) + igmp.SetType(header.IGMPType(igmpType)) + igmp.SetMaxRespTime(maxRespTime) + igmp.SetGroupAddress(groupAddress) + igmp.SetChecksum(header.IGMPCalculateChecksum(igmp)) + + e.InjectInbound(ipv4.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) +} + +// createAndInjectMLDPacket creates and injects an MLD packet with the +// specified fields. +// +// Note, the router alert option is not included in this packet. +// +// TODO(b/162198658): set the router alert option. +func createAndInjectMLDPacket(e *channel.Endpoint, mldType uint8, maxRespDelay byte, groupAddress tcpip.Address) { + icmpSize := header.ICMPv6HeaderSize + header.MLDMinimumSize + buf := buffer.NewView(header.IPv6MinimumSize + icmpSize) + + ip := header.IPv6(buf) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(icmpSize), + HopLimit: header.MLDHopLimit, + NextHeader: uint8(header.ICMPv6ProtocolNumber), + SrcAddr: header.IPv4Any, + DstAddr: header.IPv6AllNodesMulticastAddress, + }) + + icmp := header.ICMPv6(buf[header.IPv6MinimumSize:]) + icmp.SetType(header.ICMPv6Type(mldType)) + mld := header.MLD(icmp.MessageBody()) + mld.SetMaximumResponseDelay(uint16(maxRespDelay)) + mld.SetMulticastAddress(groupAddress) + icmp.SetChecksum(header.ICMPv6Checksum(icmp, header.IPv6Any, header.IPv6AllNodesMulticastAddress, buffer.VectorisedView{})) + + e.InjectInbound(ipv6.ProtocolNumber, &stack.PacketBuffer{ + Data: buf.ToVectorisedView(), + }) +} + +// TestMGPDisabled tests that the multicast group protocol is not enabled by +// default. +func TestMGPDisabled(t *testing.T) { + tests := []struct { + name string + protoNum tcpip.NetworkProtocolNumber + multicastAddr tcpip.Address + sentReportStat func(*stack.Stack) *tcpip.StatCounter + receivedQueryStat func(*stack.Stack) *tcpip.StatCounter + rxQuery func(*channel.Endpoint) + }{ + { + name: "IGMP", + protoNum: ipv4.ProtocolNumber, + multicastAddr: ipv4MulticastAddr1, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.V2MembershipReport + }, + receivedQueryStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsReceived.MembershipQuery + }, + rxQuery: func(e *channel.Endpoint) { + createAndInjectIGMPPacket(e, igmpMembershipQuery, unsolicitedIGMPReportIntervalMaxTenthSec, header.IPv4Any) + }, + }, + { + name: "MLD", + protoNum: ipv6.ProtocolNumber, + multicastAddr: ipv6MulticastAddr1, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerReport + }, + receivedQueryStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsReceived.MulticastListenerQuery + }, + rxQuery: func(e *channel.Endpoint) { + createAndInjectMLDPacket(e, mldQuery, 0, header.IPv6Any) + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e, s, clock := createStack(t, false) + + // This NIC may join multicast groups when it is enabled but since MGP is + // disabled, no reports should be sent. + sentReportStat := test.sentReportStat(s) + if got := sentReportStat.Value(); got != 0 { + t.Fatalf("got sentReportState.Value() = %d, want = 0", got) + } + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet, stack with disabled MGP sent packet = %#v", p.Pkt) + } + + // Test joining a specific group explicitly and verify that no reports are + // sent. + if err := s.JoinGroup(test.protoNum, nicID, test.multicastAddr); err != nil { + t.Fatalf("JoinGroup(%d, %d, %s): %s", test.protoNum, nicID, test.multicastAddr, err) + } + if got := sentReportStat.Value(); got != 0 { + t.Fatalf("got sentReportState.Value() = %d, want = 0", got) + } + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet, stack with disabled IGMP sent packet = %#v", p.Pkt) + } + + // Inject a general query message. This should only trigger a report to be + // sent if the MGP was enabled. + test.rxQuery(e) + if got := test.receivedQueryStat(s).Value(); got != 1 { + t.Fatalf("got receivedQueryStat(_).Value() = %d, want = 1", got) + } + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet, stack with disabled IGMP sent packet = %+v", p.Pkt) + } + }) + } +} + +func TestMGPReceiveCounters(t *testing.T) { + tests := []struct { + name string + headerType uint8 + maxRespTime byte + groupAddress tcpip.Address + statCounter func(*stack.Stack) *tcpip.StatCounter + rxMGPkt func(*channel.Endpoint, byte, byte, tcpip.Address) + }{ + { + name: "IGMP Membership Query", + headerType: igmpMembershipQuery, + maxRespTime: unsolicitedIGMPReportIntervalMaxTenthSec, + groupAddress: header.IPv4Any, + statCounter: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsReceived.MembershipQuery + }, + rxMGPkt: createAndInjectIGMPPacket, + }, + { + name: "IGMPv1 Membership Report", + headerType: igmpv1MembershipReport, + maxRespTime: 0, + groupAddress: header.IPv4AllSystems, + statCounter: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsReceived.V1MembershipReport + }, + rxMGPkt: createAndInjectIGMPPacket, + }, + { + name: "IGMPv2 Membership Report", + headerType: igmpv2MembershipReport, + maxRespTime: 0, + groupAddress: header.IPv4AllSystems, + statCounter: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsReceived.V2MembershipReport + }, + rxMGPkt: createAndInjectIGMPPacket, + }, + { + name: "IGMP Leave Group", + headerType: igmpLeaveGroup, + maxRespTime: 0, + groupAddress: header.IPv4AllRoutersGroup, + statCounter: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsReceived.LeaveGroup + }, + rxMGPkt: createAndInjectIGMPPacket, + }, + { + name: "MLD Query", + headerType: mldQuery, + maxRespTime: 0, + groupAddress: header.IPv6Any, + statCounter: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsReceived.MulticastListenerQuery + }, + rxMGPkt: createAndInjectMLDPacket, + }, + { + name: "MLD Report", + headerType: mldReport, + maxRespTime: 0, + groupAddress: header.IPv6Any, + statCounter: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsReceived.MulticastListenerReport + }, + rxMGPkt: createAndInjectMLDPacket, + }, + { + name: "MLD Done", + headerType: mldDone, + maxRespTime: 0, + groupAddress: header.IPv6Any, + statCounter: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsReceived.MulticastListenerDone + }, + rxMGPkt: createAndInjectMLDPacket, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e, s, _ := createStack(t, true) + + test.rxMGPkt(e, test.headerType, test.maxRespTime, test.groupAddress) + if got := test.statCounter(s).Value(); got != 1 { + t.Fatalf("got %s received = %d, want = 1", test.name, got) + } + }) + } +} + +// TestMGPJoinGroup tests that when explicitly joining a multicast group, the +// stack schedules and sends correct Membership Reports. +func TestMGPJoinGroup(t *testing.T) { + tests := []struct { + name string + protoNum tcpip.NetworkProtocolNumber + multicastAddr tcpip.Address + maxUnsolicitedResponseDelay time.Duration + sentReportStat func(*stack.Stack) *tcpip.StatCounter + receivedQueryStat func(*stack.Stack) *tcpip.StatCounter + validateReport func(*testing.T, channel.PacketInfo) + }{ + { + name: "IGMP", + protoNum: ipv4.ProtocolNumber, + multicastAddr: ipv4MulticastAddr1, + maxUnsolicitedResponseDelay: ipv4.UnsolicitedReportIntervalMax, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.V2MembershipReport + }, + receivedQueryStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsReceived.MembershipQuery + }, + validateReport: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateIGMPPacket(t, p, ipv4MulticastAddr1, igmpv2MembershipReport, 0, ipv4MulticastAddr1) + }, + }, + { + name: "MLD", + protoNum: ipv6.ProtocolNumber, + multicastAddr: ipv6MulticastAddr1, + maxUnsolicitedResponseDelay: ipv6.UnsolicitedReportIntervalMax, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerReport + }, + receivedQueryStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsReceived.MulticastListenerQuery + }, + validateReport: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateMLDPacket(t, p, ipv6MulticastAddr1, mldReport, 0, ipv6MulticastAddr1) + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e, s, clock := createStack(t, true) + + // Test joining a specific address explicitly and verify a Report is sent + // immediately. + if err := s.JoinGroup(test.protoNum, nicID, test.multicastAddr); err != nil { + t.Fatalf("JoinGroup(%d, %d, %s): %s", test.protoNum, nicID, test.multicastAddr, err) + } + sentReportStat := test.sentReportStat(s) + if got := sentReportStat.Value(); got != 1 { + t.Errorf("got sentReportState.Value() = %d, want = 1", got) + } + if p, ok := e.Read(); !ok { + t.Fatal("expected a report message to be sent") + } else { + test.validateReport(t, p) + } + if t.Failed() { + t.FailNow() + } + + // Verify the second report is sent by the maximum unsolicited response + // interval. + p, ok := e.Read() + if ok { + t.Fatalf("sent unexpected packet, expected report only after advancing the clock = %#v", p.Pkt) + } + clock.Advance(test.maxUnsolicitedResponseDelay) + if got := sentReportStat.Value(); got != 2 { + t.Errorf("got sentReportState.Value() = %d, want = 2", got) + } + if p, ok := e.Read(); !ok { + t.Fatal("expected a report message to be sent") + } else { + test.validateReport(t, p) + } + + // Should not send any more packets. + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet = %#v", p) + } + }) + } +} + +// TestMGPLeaveGroup tests that when leaving a previously joined multicast +// group the stack sends a leave/done message. +func TestMGPLeaveGroup(t *testing.T) { + tests := []struct { + name string + protoNum tcpip.NetworkProtocolNumber + multicastAddr tcpip.Address + sentReportStat func(*stack.Stack) *tcpip.StatCounter + sentLeaveStat func(*stack.Stack) *tcpip.StatCounter + validateReport func(*testing.T, channel.PacketInfo) + validateLeave func(*testing.T, channel.PacketInfo) + }{ + { + name: "IGMP", + protoNum: ipv4.ProtocolNumber, + multicastAddr: ipv4MulticastAddr1, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.V2MembershipReport + }, + sentLeaveStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.LeaveGroup + }, + validateReport: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateIGMPPacket(t, p, ipv4MulticastAddr1, igmpv2MembershipReport, 0, ipv4MulticastAddr1) + }, + validateLeave: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateIGMPPacket(t, p, header.IPv4AllRoutersGroup, igmpLeaveGroup, 0, ipv4MulticastAddr1) + }, + }, + { + name: "MLD", + protoNum: ipv6.ProtocolNumber, + multicastAddr: ipv6MulticastAddr1, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerReport + }, + sentLeaveStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerDone + }, + validateReport: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateMLDPacket(t, p, ipv6MulticastAddr1, mldReport, 0, ipv6MulticastAddr1) + }, + validateLeave: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateMLDPacket(t, p, header.IPv6AllRoutersMulticastAddress, mldDone, 0, ipv6MulticastAddr1) + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e, s, clock := createStack(t, true) + + if err := s.JoinGroup(test.protoNum, nicID, test.multicastAddr); err != nil { + t.Fatalf("JoinGroup(%d, %d, %s): %s", test.protoNum, nicID, test.multicastAddr, err) + } + if got := test.sentReportStat(s).Value(); got != 1 { + t.Errorf("got sentReportStat(_).Value() = %d, want = 1", got) + } + if p, ok := e.Read(); !ok { + t.Fatal("expected a report message to be sent") + } else { + test.validateReport(t, p) + } + if t.Failed() { + t.FailNow() + } + + // Leaving the group should trigger an leave/done message to be sent. + if err := s.LeaveGroup(test.protoNum, nicID, test.multicastAddr); err != nil { + t.Fatalf("LeaveGroup(%d, nic, %s): %s", test.protoNum, test.multicastAddr, err) + } + if got := test.sentLeaveStat(s).Value(); got != 1 { + t.Fatalf("got sentLeaveStat(_).Value() = %d, want = 1", got) + } + if p, ok := e.Read(); !ok { + t.Fatal("expected a leave message to be sent") + } else { + test.validateLeave(t, p) + } + + // Should not send any more packets. + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet = %#v", p) + } + }) + } +} + +// TestMGPQueryMessages tests that a report is sent in response to query +// messages. +func TestMGPQueryMessages(t *testing.T) { + tests := []struct { + name string + protoNum tcpip.NetworkProtocolNumber + multicastAddr tcpip.Address + maxUnsolicitedResponseDelay time.Duration + sentReportStat func(*stack.Stack) *tcpip.StatCounter + receivedQueryStat func(*stack.Stack) *tcpip.StatCounter + rxQuery func(*channel.Endpoint, uint8, tcpip.Address) + validateReport func(*testing.T, channel.PacketInfo) + maxRespTimeToDuration func(uint8) time.Duration + }{ + { + name: "IGMP", + protoNum: ipv4.ProtocolNumber, + multicastAddr: ipv4MulticastAddr1, + maxUnsolicitedResponseDelay: ipv4.UnsolicitedReportIntervalMax, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.V2MembershipReport + }, + receivedQueryStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsReceived.MembershipQuery + }, + rxQuery: func(e *channel.Endpoint, maxRespTime uint8, groupAddress tcpip.Address) { + createAndInjectIGMPPacket(e, igmpMembershipQuery, maxRespTime, groupAddress) + }, + validateReport: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateIGMPPacket(t, p, ipv4MulticastAddr1, igmpv2MembershipReport, 0, ipv4MulticastAddr1) + }, + maxRespTimeToDuration: header.DecisecondToDuration, + }, + { + name: "MLD", + protoNum: ipv6.ProtocolNumber, + multicastAddr: ipv6MulticastAddr1, + maxUnsolicitedResponseDelay: ipv6.UnsolicitedReportIntervalMax, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerReport + }, + receivedQueryStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsReceived.MulticastListenerQuery + }, + rxQuery: func(e *channel.Endpoint, maxRespTime uint8, groupAddress tcpip.Address) { + createAndInjectMLDPacket(e, mldQuery, maxRespTime, groupAddress) + }, + validateReport: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateMLDPacket(t, p, ipv6MulticastAddr1, mldReport, 0, ipv6MulticastAddr1) + }, + maxRespTimeToDuration: func(d uint8) time.Duration { + return time.Duration(d) * time.Millisecond + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + subTests := []struct { + name string + multicastAddr tcpip.Address + expectReport bool + }{ + { + name: "Unspecified", + multicastAddr: tcpip.Address(strings.Repeat("\x00", len(test.multicastAddr))), + expectReport: true, + }, + { + name: "Specified", + multicastAddr: test.multicastAddr, + expectReport: true, + }, + { + name: "Specified other address", + multicastAddr: func() tcpip.Address { + addrBytes := []byte(test.multicastAddr) + addrBytes[len(addrBytes)-1]++ + return tcpip.Address(addrBytes) + }(), + expectReport: false, + }, + } + + for _, subTest := range subTests { + t.Run(subTest.name, func(t *testing.T) { + e, s, clock := createStack(t, true) + + if err := s.JoinGroup(test.protoNum, nicID, test.multicastAddr); err != nil { + t.Fatalf("JoinGroup(%d, %d, %s): %s", test.protoNum, nicID, test.multicastAddr, err) + } + sentReportStat := test.sentReportStat(s) + for i := uint64(1); i <= 2; i++ { + sentReportStat := test.sentReportStat(s) + if got := sentReportStat.Value(); got != i { + t.Errorf("(i=%d) got sentReportState.Value() = %d, want = %d", i, got, i) + } + if p, ok := e.Read(); !ok { + t.Fatalf("expected %d-th report message to be sent", i) + } else { + test.validateReport(t, p) + } + clock.Advance(test.maxUnsolicitedResponseDelay) + } + if t.Failed() { + t.FailNow() + } + + // Should not send any more packets until a query. + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet = %#v", p) + } + + // Receive a query message which should trigger a report to be sent at + // some time before the maximum response time if the report is + // targeted at the host. + const maxRespTime = 100 + test.rxQuery(e, maxRespTime, subTest.multicastAddr) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet = %#v", p.Pkt) + } + + if subTest.expectReport { + clock.Advance(test.maxRespTimeToDuration(maxRespTime)) + if got := sentReportStat.Value(); got != 3 { + t.Errorf("got sentReportState.Value() = %d, want = 3", got) + } + if p, ok := e.Read(); !ok { + t.Fatal("expected a report message to be sent") + } else { + test.validateReport(t, p) + } + } + + // Should not send any more packets. + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet = %#v", p) + } + }) + } + }) + } +} + +// TestMGPQueryMessages tests that no further reports or leave/done messages +// are sent after receiving a report. +func TestMGPReportMessages(t *testing.T) { + tests := []struct { + name string + protoNum tcpip.NetworkProtocolNumber + multicastAddr tcpip.Address + sentReportStat func(*stack.Stack) *tcpip.StatCounter + sentLeaveStat func(*stack.Stack) *tcpip.StatCounter + rxReport func(*channel.Endpoint) + validateReport func(*testing.T, channel.PacketInfo) + maxRespTimeToDuration func(uint8) time.Duration + }{ + { + name: "IGMP", + protoNum: ipv4.ProtocolNumber, + multicastAddr: ipv4MulticastAddr1, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.V2MembershipReport + }, + sentLeaveStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.LeaveGroup + }, + rxReport: func(e *channel.Endpoint) { + createAndInjectIGMPPacket(e, igmpv2MembershipReport, 0, ipv4MulticastAddr1) + }, + validateReport: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateIGMPPacket(t, p, ipv4MulticastAddr1, igmpv2MembershipReport, 0, ipv4MulticastAddr1) + }, + maxRespTimeToDuration: header.DecisecondToDuration, + }, + { + name: "MLD", + protoNum: ipv6.ProtocolNumber, + multicastAddr: ipv6MulticastAddr1, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerReport + }, + sentLeaveStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerDone + }, + rxReport: func(e *channel.Endpoint) { + createAndInjectMLDPacket(e, mldReport, 0, ipv6MulticastAddr1) + }, + validateReport: func(t *testing.T, p channel.PacketInfo) { + t.Helper() + + validateMLDPacket(t, p, ipv6MulticastAddr1, mldReport, 0, ipv6MulticastAddr1) + }, + maxRespTimeToDuration: func(d uint8) time.Duration { + return time.Duration(d) * time.Millisecond + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e, s, clock := createStack(t, true) + + if err := s.JoinGroup(test.protoNum, nicID, test.multicastAddr); err != nil { + t.Fatalf("JoinGroup(%d, %d, %s): %s", test.protoNum, nicID, test.multicastAddr, err) + } + sentReportStat := test.sentReportStat(s) + if got := sentReportStat.Value(); got != 1 { + t.Errorf("got sentReportStat.Value() = %d, want = 1", got) + } + if p, ok := e.Read(); !ok { + t.Fatal("expected a report message to be sent") + } else { + test.validateReport(t, p) + } + if t.Failed() { + t.FailNow() + } + + // Receiving a report for a group we joined should cancel any further + // reports. + test.rxReport(e) + clock.Advance(time.Hour) + if got := sentReportStat.Value(); got != 1 { + t.Errorf("got sentReportStat.Value() = %d, want = 1", got) + } + if p, ok := e.Read(); ok { + t.Errorf("sent unexpected packet = %#v", p) + } + if t.Failed() { + t.FailNow() + } + + // Leaving a group after getting a report should not send a leave/done + // message. + if err := s.LeaveGroup(test.protoNum, nicID, test.multicastAddr); err != nil { + t.Fatalf("LeaveGroup(%d, nic, %s): %s", test.protoNum, test.multicastAddr, err) + } + clock.Advance(time.Hour) + if got := test.sentLeaveStat(s).Value(); got != 0 { + t.Fatalf("got sentLeaveStat(_).Value() = %d, want = 0", got) + } + + // Should not send any more packets. + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet = %#v", p) + } + }) + } +} + +func TestMGPWithNICLifecycle(t *testing.T) { + tests := []struct { + name string + protoNum tcpip.NetworkProtocolNumber + multicastAddrs []tcpip.Address + finalMulticastAddr tcpip.Address + maxUnsolicitedResponseDelay time.Duration + sentReportStat func(*stack.Stack) *tcpip.StatCounter + sentLeaveStat func(*stack.Stack) *tcpip.StatCounter + validateReport func(*testing.T, channel.PacketInfo, tcpip.Address) + validateLeave func(*testing.T, channel.PacketInfo, tcpip.Address) + getAndCheckGroupAddress func(*testing.T, map[tcpip.Address]bool, channel.PacketInfo) tcpip.Address + }{ + { + name: "IGMP", + protoNum: ipv4.ProtocolNumber, + multicastAddrs: []tcpip.Address{ipv4MulticastAddr1, ipv4MulticastAddr2}, + finalMulticastAddr: ipv4MulticastAddr3, + maxUnsolicitedResponseDelay: ipv4.UnsolicitedReportIntervalMax, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.V2MembershipReport + }, + sentLeaveStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().IGMP.PacketsSent.LeaveGroup + }, + validateReport: func(t *testing.T, p channel.PacketInfo, addr tcpip.Address) { + t.Helper() + + validateIGMPPacket(t, p, addr, igmpv2MembershipReport, 0, addr) + }, + validateLeave: func(t *testing.T, p channel.PacketInfo, addr tcpip.Address) { + t.Helper() + + validateIGMPPacket(t, p, header.IPv4AllRoutersGroup, igmpLeaveGroup, 0, addr) + }, + getAndCheckGroupAddress: func(t *testing.T, seen map[tcpip.Address]bool, p channel.PacketInfo) tcpip.Address { + t.Helper() + + ipv4 := header.IPv4(stack.PayloadSince(p.Pkt.NetworkHeader())) + if got := tcpip.TransportProtocolNumber(ipv4.Protocol()); got != header.IGMPProtocolNumber { + t.Fatalf("got ipv4.Protocol() = %d, want = %d", got, header.IGMPProtocolNumber) + } + addr := header.IGMP(ipv4.Payload()).GroupAddress() + s, ok := seen[addr] + if !ok { + t.Fatalf("unexpectedly got a packet for group %s", addr) + } + if s { + t.Fatalf("already saw packet for group %s", addr) + } + seen[addr] = true + return addr + }, + }, + { + name: "MLD", + protoNum: ipv6.ProtocolNumber, + multicastAddrs: []tcpip.Address{ipv6MulticastAddr1, ipv6MulticastAddr2}, + finalMulticastAddr: ipv6MulticastAddr3, + maxUnsolicitedResponseDelay: ipv6.UnsolicitedReportIntervalMax, + sentReportStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerReport + }, + sentLeaveStat: func(s *stack.Stack) *tcpip.StatCounter { + return s.Stats().ICMP.V6.PacketsSent.MulticastListenerDone + }, + validateReport: func(t *testing.T, p channel.PacketInfo, addr tcpip.Address) { + t.Helper() + + validateMLDPacket(t, p, addr, mldReport, 0, addr) + }, + validateLeave: func(t *testing.T, p channel.PacketInfo, addr tcpip.Address) { + t.Helper() + + validateMLDPacket(t, p, header.IPv6AllRoutersMulticastAddress, mldDone, 0, addr) + }, + getAndCheckGroupAddress: func(t *testing.T, seen map[tcpip.Address]bool, p channel.PacketInfo) tcpip.Address { + t.Helper() + + ipv6 := header.IPv6(stack.PayloadSince(p.Pkt.NetworkHeader())) + if got := tcpip.TransportProtocolNumber(ipv6.NextHeader()); got != header.ICMPv6ProtocolNumber { + t.Fatalf("got ipv6.NextHeader() = %d, want = %d", got, header.ICMPv6ProtocolNumber) + } + icmpv6 := header.ICMPv6(ipv6.Payload()) + if got := icmpv6.Type(); got != header.ICMPv6MulticastListenerReport && got != header.ICMPv6MulticastListenerDone { + t.Fatalf("got icmpv6.Type() = %d, want = %d or %d", got, header.ICMPv6MulticastListenerReport, header.ICMPv6MulticastListenerDone) + } + addr := header.MLD(icmpv6.MessageBody()).MulticastAddress() + s, ok := seen[addr] + if !ok { + t.Fatalf("unexpectedly got a packet for group %s", addr) + } + if s { + t.Fatalf("already saw packet for group %s", addr) + } + seen[addr] = true + return addr + + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e, s, clock := createStack(t, true) + + sentReportStat := test.sentReportStat(s) + var reportCounter uint64 + for _, a := range test.multicastAddrs { + if err := s.JoinGroup(test.protoNum, nicID, a); err != nil { + t.Fatalf("JoinGroup(%d, %d, %s): %s", test.protoNum, nicID, a, err) + } + reportCounter++ + if got := sentReportStat.Value(); got != reportCounter { + t.Errorf("got sentReportStat.Value() = %d, want = %d", got, reportCounter) + } + if p, ok := e.Read(); !ok { + t.Fatalf("expected a report message to be sent for %s", a) + } else { + test.validateReport(t, p, a) + } + } + if t.Failed() { + t.FailNow() + } + + // Leave messages should be sent for the joined groups when the NIC is + // disabled. + if err := s.DisableNIC(nicID); err != nil { + t.Fatalf("DisableNIC(%d): %s", nicID, err) + } + sentLeaveStat := test.sentLeaveStat(s) + leaveCounter := uint64(len(test.multicastAddrs)) + if got := sentLeaveStat.Value(); got != leaveCounter { + t.Errorf("got sentLeaveStat.Value() = %d, want = %d", got, leaveCounter) + } + { + seen := make(map[tcpip.Address]bool) + for _, a := range test.multicastAddrs { + seen[a] = false + } + + for i, _ := range test.multicastAddrs { + p, ok := e.Read() + if !ok { + t.Fatalf("expected (%d-th) leave message to be sent", i) + } + + test.validateLeave(t, p, test.getAndCheckGroupAddress(t, seen, p)) + } + } + if t.Failed() { + t.FailNow() + } + + // Reports should be sent for the joined groups when the NIC is enabled. + if err := s.EnableNIC(nicID); err != nil { + t.Fatalf("EnableNIC(%d): %s", nicID, err) + } + reportCounter += uint64(len(test.multicastAddrs)) + if got := sentReportStat.Value(); got != reportCounter { + t.Errorf("got sentReportStat.Value() = %d, want = %d", got, reportCounter) + } + { + seen := make(map[tcpip.Address]bool) + for _, a := range test.multicastAddrs { + seen[a] = false + } + + for i, _ := range test.multicastAddrs { + p, ok := e.Read() + if !ok { + t.Fatalf("expected (%d-th) report message to be sent", i) + } + + test.validateReport(t, p, test.getAndCheckGroupAddress(t, seen, p)) + } + } + if t.Failed() { + t.FailNow() + } + + // Joining/leaving a group while disabled should not send any messages. + if err := s.DisableNIC(nicID); err != nil { + t.Fatalf("DisableNIC(%d): %s", nicID, err) + } + leaveCounter += uint64(len(test.multicastAddrs)) + if got := sentLeaveStat.Value(); got != leaveCounter { + t.Errorf("got sentLeaveStat.Value() = %d, want = %d", got, leaveCounter) + } + for i, _ := range test.multicastAddrs { + if _, ok := e.Read(); !ok { + t.Fatalf("expected (%d-th) leave message to be sent", i) + } + } + for _, a := range test.multicastAddrs { + if err := s.LeaveGroup(test.protoNum, nicID, a); err != nil { + t.Fatalf("LeaveGroup(%d, nic, %s): %s", test.protoNum, a, err) + } + if got := sentLeaveStat.Value(); got != leaveCounter { + t.Errorf("got sentLeaveStat.Value() = %d, want = %d", got, leaveCounter) + } + if p, ok := e.Read(); ok { + t.Fatalf("leaving group %s on disabled NIC sent unexpected packet = %#v", a, p.Pkt) + } + } + if err := s.JoinGroup(test.protoNum, nicID, test.finalMulticastAddr); err != nil { + t.Fatalf("JoinGroup(%d, %d, %s): %s", test.protoNum, nicID, test.finalMulticastAddr, err) + } + if got := sentReportStat.Value(); got != reportCounter { + t.Errorf("got sentReportStat.Value() = %d, want = %d", got, reportCounter) + } + if p, ok := e.Read(); ok { + t.Fatalf("joining group %s on disabled NIC sent unexpected packet = %#v", test.finalMulticastAddr, p.Pkt) + } + + // A report should only be sent for the group we last joined after + // enabling the NIC since the original groups were all left. + if err := s.EnableNIC(nicID); err != nil { + t.Fatalf("EnableNIC(%d): %s", nicID, err) + } + reportCounter++ + if got := sentReportStat.Value(); got != reportCounter { + t.Errorf("got sentReportStat.Value() = %d, want = %d", got, reportCounter) + } + if p, ok := e.Read(); !ok { + t.Fatal("expected a report message to be sent") + } else { + test.validateReport(t, p, test.finalMulticastAddr) + } + + clock.Advance(test.maxUnsolicitedResponseDelay) + reportCounter++ + if got := sentReportStat.Value(); got != reportCounter { + t.Errorf("got sentReportState.Value() = %d, want = %d", got, reportCounter) + } + if p, ok := e.Read(); !ok { + t.Fatal("expected a report message to be sent") + } else { + test.validateReport(t, p, test.finalMulticastAddr) + } + + // Should not send any more packets. + clock.Advance(time.Hour) + if p, ok := e.Read(); ok { + t.Fatalf("sent unexpected packet = %#v", p) + } + }) + } +} diff --git a/pkg/tcpip/network/testutil/testutil.go b/pkg/tcpip/network/testutil/testutil.go index 7cc52985e..5c3363759 100644 --- a/pkg/tcpip/network/testutil/testutil.go +++ b/pkg/tcpip/network/testutil/testutil.go @@ -85,21 +85,6 @@ func (ep *MockLinkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts st return n, nil } -// WriteRawPacket implements LinkEndpoint.WriteRawPacket. -func (ep *MockLinkEndpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - if ep.allowPackets == 0 { - return ep.err - } - ep.allowPackets-- - - pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - Data: vv, - }) - ep.WrittenPackets = append(ep.WrittenPackets, pkt) - - return nil -} - // Attach implements LinkEndpoint.Attach. func (*MockLinkEndpoint) Attach(stack.NetworkDispatcher) {} diff --git a/pkg/tcpip/socketops.go b/pkg/tcpip/socketops.go index 2a6c7c7c0..c53698a6a 100644 --- a/pkg/tcpip/socketops.go +++ b/pkg/tcpip/socketops.go @@ -15,31 +15,290 @@ package tcpip import ( - "gvisor.dev/gvisor/pkg/sync" + "sync/atomic" ) -// SocketOptions contains all the variables which store values for socket -// level options. +// SocketOptionsHandler holds methods that help define endpoint specific +// behavior for socket level socket options. These must be implemented by +// endpoints to get notified when socket level options are set. +type SocketOptionsHandler interface { + // OnReuseAddressSet is invoked when SO_REUSEADDR is set for an endpoint. + OnReuseAddressSet(v bool) + + // OnReusePortSet is invoked when SO_REUSEPORT is set for an endpoint. + OnReusePortSet(v bool) + + // OnKeepAliveSet is invoked when SO_KEEPALIVE is set for an endpoint. + OnKeepAliveSet(v bool) + + // OnDelayOptionSet is invoked when TCP_NODELAY is set for an endpoint. + // Note that v will be the inverse of TCP_NODELAY option. + OnDelayOptionSet(v bool) + + // OnCorkOptionSet is invoked when TCP_CORK is set for an endpoint. + OnCorkOptionSet(v bool) +} + +// DefaultSocketOptionsHandler is an embeddable type that implements no-op +// implementations for SocketOptionsHandler methods. +type DefaultSocketOptionsHandler struct{} + +var _ SocketOptionsHandler = (*DefaultSocketOptionsHandler)(nil) + +// OnReuseAddressSet implements SocketOptionsHandler.OnReuseAddressSet. +func (*DefaultSocketOptionsHandler) OnReuseAddressSet(bool) {} + +// OnReusePortSet implements SocketOptionsHandler.OnReusePortSet. +func (*DefaultSocketOptionsHandler) OnReusePortSet(bool) {} + +// OnKeepAliveSet implements SocketOptionsHandler.OnKeepAliveSet. +func (*DefaultSocketOptionsHandler) OnKeepAliveSet(bool) {} + +// OnDelayOptionSet implements SocketOptionsHandler.OnDelayOptionSet. +func (*DefaultSocketOptionsHandler) OnDelayOptionSet(bool) {} + +// OnCorkOptionSet implements SocketOptionsHandler.OnCorkOptionSet. +func (*DefaultSocketOptionsHandler) OnCorkOptionSet(bool) {} + +// SocketOptions contains all the variables which store values for SOL_SOCKET, +// SOL_IP, SOL_IPV6 and SOL_TCP level options. // // +stateify savable type SocketOptions struct { - // mu protects fields below. - mu sync.Mutex `state:"nosave"` - broadcastEnabled bool + handler SocketOptionsHandler + + // These fields are accessed and modified using atomic operations. + + // broadcastEnabled determines whether datagram sockets are allowed to send + // packets to a broadcast address. + broadcastEnabled uint32 + + // passCredEnabled determines whether SCM_CREDENTIALS socket control messages + // are enabled. + passCredEnabled uint32 + + // noChecksumEnabled determines whether UDP checksum is disabled while + // transmitting for this socket. + noChecksumEnabled uint32 + + // reuseAddressEnabled determines whether Bind() should allow reuse of local + // address. + reuseAddressEnabled uint32 + + // reusePortEnabled determines whether to permit multiple sockets to be bound + // to an identical socket address. + reusePortEnabled uint32 + + // keepAliveEnabled determines whether TCP keepalive is enabled for this + // socket. + keepAliveEnabled uint32 + + // multicastLoopEnabled determines whether multicast packets sent over a + // non-loopback interface will be looped back. Analogous to inet->mc_loop. + multicastLoopEnabled uint32 + + // receiveTOSEnabled is used to specify if the TOS ancillary message is + // passed with incoming packets. + receiveTOSEnabled uint32 + + // receiveTClassEnabled is used to specify if the IPV6_TCLASS ancillary + // message is passed with incoming packets. + receiveTClassEnabled uint32 + + // receivePacketInfoEnabled is used to specify if more inforamtion is + // provided with incoming packets such as interface index and address. + receivePacketInfoEnabled uint32 + + // hdrIncludeEnabled is used to indicate for a raw endpoint that all packets + // being written have an IP header and the endpoint should not attach an IP + // header. + hdrIncludedEnabled uint32 + + // v6OnlyEnabled is used to determine whether an IPv6 socket is to be + // restricted to sending and receiving IPv6 packets only. + v6OnlyEnabled uint32 + + // quickAckEnabled is used to represent the value of TCP_QUICKACK option. + // It currently does not have any effect on the TCP endpoint. + quickAckEnabled uint32 + + // delayOptionEnabled is used to specify if data should be sent out immediately + // by the transport protocol. For TCP, it determines if the Nagle algorithm + // is on or off. + delayOptionEnabled uint32 + + // corkOptionEnabled is used to specify if data should be held until segments + // are full by the TCP transport protocol. + corkOptionEnabled uint32 +} + +// InitHandler initializes the handler. This must be called before using the +// socket options utility. +func (so *SocketOptions) InitHandler(handler SocketOptionsHandler) { + so.handler = handler +} + +func storeAtomicBool(addr *uint32, v bool) { + var val uint32 + if v { + val = 1 + } + atomic.StoreUint32(addr, val) } // GetBroadcast gets value for SO_BROADCAST option. func (so *SocketOptions) GetBroadcast() bool { - so.mu.Lock() - defer so.mu.Unlock() - - return so.broadcastEnabled + return atomic.LoadUint32(&so.broadcastEnabled) != 0 } // SetBroadcast sets value for SO_BROADCAST option. func (so *SocketOptions) SetBroadcast(v bool) { - so.mu.Lock() - defer so.mu.Unlock() + storeAtomicBool(&so.broadcastEnabled, v) +} + +// GetPassCred gets value for SO_PASSCRED option. +func (so *SocketOptions) GetPassCred() bool { + return atomic.LoadUint32(&so.passCredEnabled) != 0 +} + +// SetPassCred sets value for SO_PASSCRED option. +func (so *SocketOptions) SetPassCred(v bool) { + storeAtomicBool(&so.passCredEnabled, v) +} + +// GetNoChecksum gets value for SO_NO_CHECK option. +func (so *SocketOptions) GetNoChecksum() bool { + return atomic.LoadUint32(&so.noChecksumEnabled) != 0 +} + +// SetNoChecksum sets value for SO_NO_CHECK option. +func (so *SocketOptions) SetNoChecksum(v bool) { + storeAtomicBool(&so.noChecksumEnabled, v) +} + +// GetReuseAddress gets value for SO_REUSEADDR option. +func (so *SocketOptions) GetReuseAddress() bool { + return atomic.LoadUint32(&so.reuseAddressEnabled) != 0 +} + +// SetReuseAddress sets value for SO_REUSEADDR option. +func (so *SocketOptions) SetReuseAddress(v bool) { + storeAtomicBool(&so.reuseAddressEnabled, v) + so.handler.OnReuseAddressSet(v) +} + +// GetReusePort gets value for SO_REUSEPORT option. +func (so *SocketOptions) GetReusePort() bool { + return atomic.LoadUint32(&so.reusePortEnabled) != 0 +} + +// SetReusePort sets value for SO_REUSEPORT option. +func (so *SocketOptions) SetReusePort(v bool) { + storeAtomicBool(&so.reusePortEnabled, v) + so.handler.OnReusePortSet(v) +} + +// GetKeepAlive gets value for SO_KEEPALIVE option. +func (so *SocketOptions) GetKeepAlive() bool { + return atomic.LoadUint32(&so.keepAliveEnabled) != 0 +} + +// SetKeepAlive sets value for SO_KEEPALIVE option. +func (so *SocketOptions) SetKeepAlive(v bool) { + storeAtomicBool(&so.keepAliveEnabled, v) + so.handler.OnKeepAliveSet(v) +} + +// GetMulticastLoop gets value for IP_MULTICAST_LOOP option. +func (so *SocketOptions) GetMulticastLoop() bool { + return atomic.LoadUint32(&so.multicastLoopEnabled) != 0 +} + +// SetMulticastLoop sets value for IP_MULTICAST_LOOP option. +func (so *SocketOptions) SetMulticastLoop(v bool) { + storeAtomicBool(&so.multicastLoopEnabled, v) +} + +// GetReceiveTOS gets value for IP_RECVTOS option. +func (so *SocketOptions) GetReceiveTOS() bool { + return atomic.LoadUint32(&so.receiveTOSEnabled) != 0 +} + +// SetReceiveTOS sets value for IP_RECVTOS option. +func (so *SocketOptions) SetReceiveTOS(v bool) { + storeAtomicBool(&so.receiveTOSEnabled, v) +} + +// GetReceiveTClass gets value for IPV6_RECVTCLASS option. +func (so *SocketOptions) GetReceiveTClass() bool { + return atomic.LoadUint32(&so.receiveTClassEnabled) != 0 +} + +// SetReceiveTClass sets value for IPV6_RECVTCLASS option. +func (so *SocketOptions) SetReceiveTClass(v bool) { + storeAtomicBool(&so.receiveTClassEnabled, v) +} + +// GetReceivePacketInfo gets value for IP_PKTINFO option. +func (so *SocketOptions) GetReceivePacketInfo() bool { + return atomic.LoadUint32(&so.receivePacketInfoEnabled) != 0 +} + +// SetReceivePacketInfo sets value for IP_PKTINFO option. +func (so *SocketOptions) SetReceivePacketInfo(v bool) { + storeAtomicBool(&so.receivePacketInfoEnabled, v) +} + +// GetHeaderIncluded gets value for IP_HDRINCL option. +func (so *SocketOptions) GetHeaderIncluded() bool { + return atomic.LoadUint32(&so.hdrIncludedEnabled) != 0 +} + +// SetHeaderIncluded sets value for IP_HDRINCL option. +func (so *SocketOptions) SetHeaderIncluded(v bool) { + storeAtomicBool(&so.hdrIncludedEnabled, v) +} + +// GetV6Only gets value for IPV6_V6ONLY option. +func (so *SocketOptions) GetV6Only() bool { + return atomic.LoadUint32(&so.v6OnlyEnabled) != 0 +} + +// SetV6Only sets value for IPV6_V6ONLY option. +// +// Preconditions: the backing TCP or UDP endpoint must be in initial state. +func (so *SocketOptions) SetV6Only(v bool) { + storeAtomicBool(&so.v6OnlyEnabled, v) +} + +// GetQuickAck gets value for TCP_QUICKACK option. +func (so *SocketOptions) GetQuickAck() bool { + return atomic.LoadUint32(&so.quickAckEnabled) != 0 +} + +// SetQuickAck sets value for TCP_QUICKACK option. +func (so *SocketOptions) SetQuickAck(v bool) { + storeAtomicBool(&so.quickAckEnabled, v) +} + +// GetDelayOption gets inverted value for TCP_NODELAY option. +func (so *SocketOptions) GetDelayOption() bool { + return atomic.LoadUint32(&so.delayOptionEnabled) != 0 +} + +// SetDelayOption sets inverted value for TCP_NODELAY option. +func (so *SocketOptions) SetDelayOption(v bool) { + storeAtomicBool(&so.delayOptionEnabled, v) + so.handler.OnDelayOptionSet(v) +} + +// GetCorkOption gets value for TCP_CORK option. +func (so *SocketOptions) GetCorkOption() bool { + return atomic.LoadUint32(&so.corkOptionEnabled) != 0 +} - so.broadcastEnabled = v +// SetCorkOption sets value for TCP_CORK option. +func (so *SocketOptions) SetCorkOption(v bool) { + storeAtomicBool(&so.corkOptionEnabled, v) + so.handler.OnCorkOptionSet(v) } diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index d09ebe7fa..9cc6074da 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test", "most_shards") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) @@ -112,7 +112,7 @@ go_test( "transport_demuxer_test.go", "transport_test.go", ], - shard_count = 20, + shard_count = most_shards, deps = [ ":stack", "//pkg/rand", @@ -120,6 +120,7 @@ go_test( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/checker", + "//pkg/tcpip/faketime", "//pkg/tcpip/header", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/loopback", @@ -131,7 +132,6 @@ go_test( "//pkg/tcpip/transport/udp", "//pkg/waiter", "@com_github_google_go_cmp//cmp:go_default_library", - "@com_github_google_go_cmp//cmp/cmpopts:go_default_library", ], ) diff --git a/pkg/tcpip/stack/addressable_endpoint_state.go b/pkg/tcpip/stack/addressable_endpoint_state.go index 9478f3fb7..6e4f5fa46 100644 --- a/pkg/tcpip/stack/addressable_endpoint_state.go +++ b/pkg/tcpip/stack/addressable_endpoint_state.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" ) -var _ GroupAddressableEndpoint = (*AddressableEndpointState)(nil) var _ AddressableEndpoint = (*AddressableEndpointState)(nil) // AddressableEndpointState is an implementation of an AddressableEndpoint. @@ -37,10 +36,6 @@ type AddressableEndpointState struct { endpoints map[tcpip.Address]*addressState primary []*addressState - - // groups holds the mapping between group addresses and the number of times - // they have been joined. - groups map[tcpip.Address]uint32 } } @@ -53,65 +48,33 @@ func (a *AddressableEndpointState) Init(networkEndpoint NetworkEndpoint) { a.mu.Lock() defer a.mu.Unlock() a.mu.endpoints = make(map[tcpip.Address]*addressState) - a.mu.groups = make(map[tcpip.Address]uint32) -} - -// ReadOnlyAddressableEndpointState provides read-only access to an -// AddressableEndpointState. -type ReadOnlyAddressableEndpointState struct { - inner *AddressableEndpointState } -// AddrOrMatching returns an endpoint for the passed address that is consisdered -// bound to the wrapped AddressableEndpointState. +// GetAddress returns the AddressEndpoint for the passed address. // -// If addr is an exact match with an existing address, that address is returned. -// Otherwise, f is called with each address and the address that f returns true -// for is returned. -// -// Returns nil of no address matches. -func (m ReadOnlyAddressableEndpointState) AddrOrMatching(addr tcpip.Address, spoofingOrPrimiscuous bool, f func(AddressEndpoint) bool) AddressEndpoint { - m.inner.mu.RLock() - defer m.inner.mu.RUnlock() - - if ep, ok := m.inner.mu.endpoints[addr]; ok { - if ep.IsAssigned(spoofingOrPrimiscuous) && ep.IncRef() { - return ep - } - } - - for _, ep := range m.inner.mu.endpoints { - if ep.IsAssigned(spoofingOrPrimiscuous) && f(ep) && ep.IncRef() { - return ep - } - } - - return nil -} - -// Lookup returns the AddressEndpoint for the passed address. +// GetAddress does not increment the address's reference count or check if the +// address is considered bound to the endpoint. // -// Returns nil if the passed address is not associated with the -// AddressableEndpointState. -func (m ReadOnlyAddressableEndpointState) Lookup(addr tcpip.Address) AddressEndpoint { - m.inner.mu.RLock() - defer m.inner.mu.RUnlock() +// Returns nil if the passed address is not associated with the endpoint. +func (a *AddressableEndpointState) GetAddress(addr tcpip.Address) AddressEndpoint { + a.mu.RLock() + defer a.mu.RUnlock() - ep, ok := m.inner.mu.endpoints[addr] + ep, ok := a.mu.endpoints[addr] if !ok { return nil } return ep } -// ForEach calls f for each address pair. +// ForEachEndpoint calls f for each address. // -// If f returns false, f is no longer be called. -func (m ReadOnlyAddressableEndpointState) ForEach(f func(AddressEndpoint) bool) { - m.inner.mu.RLock() - defer m.inner.mu.RUnlock() +// Once f returns false, f will no longer be called. +func (a *AddressableEndpointState) ForEachEndpoint(f func(AddressEndpoint) bool) { + a.mu.RLock() + defer a.mu.RUnlock() - for _, ep := range m.inner.mu.endpoints { + for _, ep := range a.mu.endpoints { if !f(ep) { return } @@ -119,21 +82,15 @@ func (m ReadOnlyAddressableEndpointState) ForEach(f func(AddressEndpoint) bool) } // ForEachPrimaryEndpoint calls f for each primary address. -// -// If f returns false, f is no longer be called. -func (m ReadOnlyAddressableEndpointState) ForEachPrimaryEndpoint(f func(AddressEndpoint)) { - m.inner.mu.RLock() - defer m.inner.mu.RUnlock() - for _, ep := range m.inner.mu.primary { +func (a *AddressableEndpointState) ForEachPrimaryEndpoint(f func(AddressEndpoint)) { + a.mu.RLock() + defer a.mu.RUnlock() + + for _, ep := range a.mu.primary { f(ep) } } -// ReadOnly returns a readonly reference to a. -func (a *AddressableEndpointState) ReadOnly() ReadOnlyAddressableEndpointState { - return ReadOnlyAddressableEndpointState{inner: a} -} - func (a *AddressableEndpointState) releaseAddressState(addrState *addressState) { a.mu.Lock() defer a.mu.Unlock() @@ -335,11 +292,6 @@ func (a *AddressableEndpointState) addAndAcquireAddressLocked(addr tcpip.Address func (a *AddressableEndpointState) RemovePermanentAddress(addr tcpip.Address) *tcpip.Error { a.mu.Lock() defer a.mu.Unlock() - - if _, ok := a.mu.groups[addr]; ok { - panic(fmt.Sprintf("group address = %s must be removed with LeaveGroup", addr)) - } - return a.removePermanentAddressLocked(addr) } @@ -471,8 +423,19 @@ func (a *AddressableEndpointState) acquirePrimaryAddressRLocked(isValid func(*ad return deprecatedEndpoint } -// AcquireAssignedAddress implements AddressableEndpoint. -func (a *AddressableEndpointState) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB PrimaryEndpointBehavior) AddressEndpoint { +// AcquireAssignedAddressOrMatching returns an address endpoint that is +// considered assigned to the addressable endpoint. +// +// If the address is an exact match with an existing address, that address is +// returned. Otherwise, if f is provided, f is called with each address and +// the address that f returns true for is returned. +// +// If there is no matching address, a temporary address will be returned if +// allowTemp is true. +// +// Regardless how the address was obtained, it will be acquired before it is +// returned. +func (a *AddressableEndpointState) AcquireAssignedAddressOrMatching(localAddr tcpip.Address, f func(AddressEndpoint) bool, allowTemp bool, tempPEB PrimaryEndpointBehavior) AddressEndpoint { a.mu.Lock() defer a.mu.Unlock() @@ -488,6 +451,14 @@ func (a *AddressableEndpointState) AcquireAssignedAddress(localAddr tcpip.Addres return addrState } + if f != nil { + for _, addrState := range a.mu.endpoints { + if addrState.IsAssigned(allowTemp) && f(addrState) && addrState.IncRef() { + return addrState + } + } + } + if !allowTemp { return nil } @@ -520,6 +491,11 @@ func (a *AddressableEndpointState) AcquireAssignedAddress(localAddr tcpip.Addres return ep } +// AcquireAssignedAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB PrimaryEndpointBehavior) AddressEndpoint { + return a.AcquireAssignedAddressOrMatching(localAddr, nil, allowTemp, tempPEB) +} + // AcquireOutgoingPrimaryAddress implements AddressableEndpoint. func (a *AddressableEndpointState) AcquireOutgoingPrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) AddressEndpoint { a.mu.RLock() @@ -588,72 +564,11 @@ func (a *AddressableEndpointState) PermanentAddresses() []tcpip.AddressWithPrefi return addrs } -// JoinGroup implements GroupAddressableEndpoint. -func (a *AddressableEndpointState) JoinGroup(group tcpip.Address) (bool, *tcpip.Error) { - a.mu.Lock() - defer a.mu.Unlock() - - joins, ok := a.mu.groups[group] - if !ok { - ep, err := a.addAndAcquireAddressLocked(group.WithPrefix(), NeverPrimaryEndpoint, AddressConfigStatic, false /* deprecated */, true /* permanent */) - if err != nil { - return false, err - } - // We have no need for the address endpoint. - a.decAddressRefLocked(ep) - } - - a.mu.groups[group] = joins + 1 - return !ok, nil -} - -// LeaveGroup implements GroupAddressableEndpoint. -func (a *AddressableEndpointState) LeaveGroup(group tcpip.Address) (bool, *tcpip.Error) { - a.mu.Lock() - defer a.mu.Unlock() - - joins, ok := a.mu.groups[group] - if !ok { - return false, tcpip.ErrBadLocalAddress - } - - if joins == 1 { - a.removeGroupAddressLocked(group) - delete(a.mu.groups, group) - return true, nil - } - - a.mu.groups[group] = joins - 1 - return false, nil -} - -// IsInGroup implements GroupAddressableEndpoint. -func (a *AddressableEndpointState) IsInGroup(group tcpip.Address) bool { - a.mu.RLock() - defer a.mu.RUnlock() - _, ok := a.mu.groups[group] - return ok -} - -func (a *AddressableEndpointState) removeGroupAddressLocked(group tcpip.Address) { - if err := a.removePermanentAddressLocked(group); err != nil { - // removePermanentEndpointLocked would only return an error if group is - // not bound to the addressable endpoint, but we know it MUST be assigned - // since we have group in our map of groups. - panic(fmt.Sprintf("error removing group address = %s: %s", group, err)) - } -} - // Cleanup forcefully leaves all groups and removes all permanent addresses. func (a *AddressableEndpointState) Cleanup() { a.mu.Lock() defer a.mu.Unlock() - for group := range a.mu.groups { - a.removeGroupAddressLocked(group) - } - a.mu.groups = make(map[tcpip.Address]uint32) - for _, ep := range a.mu.endpoints { // removePermanentEndpointLocked returns tcpip.ErrBadLocalAddress if ep is // not a permanent address. diff --git a/pkg/tcpip/stack/addressable_endpoint_state_test.go b/pkg/tcpip/stack/addressable_endpoint_state_test.go index 26787d0a3..140f146f6 100644 --- a/pkg/tcpip/stack/addressable_endpoint_state_test.go +++ b/pkg/tcpip/stack/addressable_endpoint_state_test.go @@ -53,25 +53,9 @@ func TestAddressableEndpointStateCleanup(t *testing.T) { ep.DecRef() } - group := tcpip.Address("\x02") - if added, err := s.JoinGroup(group); err != nil { - t.Fatalf("s.JoinGroup(%s): %s", group, err) - } else if !added { - t.Fatalf("got s.JoinGroup(%s) = false, want = true", group) - } - if !s.IsInGroup(group) { - t.Fatalf("got s.IsInGroup(%s) = false, want = true", group) - } - s.Cleanup() - { - ep := s.AcquireAssignedAddress(addr.Address, false /* allowTemp */, stack.NeverPrimaryEndpoint) - if ep != nil { - ep.DecRef() - t.Fatalf("got s.AcquireAssignedAddress(%s, false, NeverPrimaryEndpoint) = %s, want = nil", addr.Address, ep.AddressWithPrefix()) - } - } - if s.IsInGroup(group) { - t.Fatalf("got s.IsInGroup(%s) = true, want = false", group) + if ep := s.AcquireAssignedAddress(addr.Address, false /* allowTemp */, stack.NeverPrimaryEndpoint); ep != nil { + ep.DecRef() + t.Fatalf("got s.AcquireAssignedAddress(%s, false, NeverPrimaryEndpoint) = %s, want = nil", addr.Address, ep.AddressWithPrefix()) } } diff --git a/pkg/tcpip/stack/forwarding_test.go b/pkg/tcpip/stack/forwarding_test.go index 6dc9e7859..5ec9b3411 100644 --- a/pkg/tcpip/stack/forwarding_test.go +++ b/pkg/tcpip/stack/forwarding_test.go @@ -309,7 +309,7 @@ func (e *fwdTestLinkEndpoint) LinkAddress() tcpip.LinkAddress { func (e fwdTestLinkEndpoint) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error { p := fwdTestPacketInfo{ - RemoteLinkAddress: r.RemoteLinkAddress, + RemoteLinkAddress: r.RemoteLinkAddress(), LocalLinkAddress: r.LocalLinkAddress, Pkt: pkt, } @@ -333,20 +333,6 @@ func (e *fwdTestLinkEndpoint) WritePackets(r *Route, gso *GSO, pkts PacketBuffer return n, nil } -// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. -func (e *fwdTestLinkEndpoint) WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error { - p := fwdTestPacketInfo{ - Pkt: NewPacketBuffer(PacketBufferOptions{Data: vv}), - } - - select { - case e.C <- p: - default: - } - - return nil -} - // Wait implements stack.LinkEndpoint.Wait. func (*fwdTestLinkEndpoint) Wait() {} diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 73a01c2dd..31b67b987 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/checker" + "gvisor.dev/gvisor/pkg/tcpip/faketime" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" @@ -352,7 +353,7 @@ func TestDADDisabled(t *testing.T) { } // We should not have sent any NDP NS messages. - if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got != 0 { + if got := s.Stats().ICMP.V6.PacketsSent.NeighborSolicit.Value(); got != 0 { t.Fatalf("got NeighborSolicit = %d, want = 0", got) } } @@ -465,14 +466,18 @@ func TestDADResolve(t *testing.T) { if err != tcpip.ErrNoRoute { t.Errorf("got FindRoute(%d, '', %s, %d, false) = (%+v, %v), want = (_, %s)", nicID, addr2, header.IPv6ProtocolNumber, r, err, tcpip.ErrNoRoute) } - r.Release() + if r != nil { + r.Release() + } } { r, err := s.FindRoute(nicID, addr1, addr2, header.IPv6ProtocolNumber, false) if err != tcpip.ErrNoRoute { t.Errorf("got FindRoute(%d, %s, %s, %d, false) = (%+v, %v), want = (_, %s)", nicID, addr1, addr2, header.IPv6ProtocolNumber, r, err, tcpip.ErrNoRoute) } - r.Release() + if r != nil { + r.Release() + } } if t.Failed() { @@ -510,7 +515,9 @@ func TestDADResolve(t *testing.T) { } else if r.LocalAddress != addr1 { t.Errorf("got r.LocalAddress = %s, want = %s", r.LocalAddress, addr1) } - r.Release() + if r != nil { + r.Release() + } } if t.Failed() { @@ -518,7 +525,7 @@ func TestDADResolve(t *testing.T) { } // Should not have sent any more NS messages. - if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got != uint64(test.dupAddrDetectTransmits) { + if got := s.Stats().ICMP.V6.PacketsSent.NeighborSolicit.Value(); got != uint64(test.dupAddrDetectTransmits) { t.Fatalf("got NeighborSolicit = %d, want = %d", got, test.dupAddrDetectTransmits) } @@ -533,8 +540,8 @@ func TestDADResolve(t *testing.T) { // Make sure the right remote link address is used. snmc := header.SolicitedNodeAddr(addr1) - if want := header.EthernetAddressFromMulticastIPv6Address(snmc); p.Route.RemoteLinkAddress != want { - t.Errorf("got remote link address = %s, want = %s", p.Route.RemoteLinkAddress, want) + if got, want := p.Route.RemoteLinkAddress(), header.EthernetAddressFromMulticastIPv6Address(snmc); got != want { + t.Errorf("got remote link address = %s, want = %s", got, want) } // Check NDP NS packet. @@ -563,7 +570,7 @@ func rxNDPSolicit(e *channel.Endpoint, tgt tcpip.Address) { hdr := buffer.NewPrependable(header.IPv6MinimumSize + header.ICMPv6NeighborSolicitMinimumSize) pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) pkt.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + ns := header.NDPNeighborSolicit(pkt.MessageBody()) ns.SetTargetAddress(tgt) snmc := header.SolicitedNodeAddr(tgt) pkt.SetChecksum(header.ICMPv6Checksum(pkt, header.IPv6Any, snmc, buffer.VectorisedView{})) @@ -605,7 +612,7 @@ func TestDADFail(t *testing.T) { hdr := buffer.NewPrependable(header.IPv6MinimumSize + naSize) pkt := header.ICMPv6(hdr.Prepend(naSize)) pkt.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(pkt.NDPPayload()) + na := header.NDPNeighborAdvert(pkt.MessageBody()) na.SetSolicitedFlag(true) na.SetOverrideFlag(true) na.SetTargetAddress(tgt) @@ -666,7 +673,7 @@ func TestDADFail(t *testing.T) { // Receive a packet to simulate an address conflict. test.rxPkt(e, addr1) - stat := test.getStat(s.Stats().ICMP.V6PacketsReceived) + stat := test.getStat(s.Stats().ICMP.V6.PacketsReceived) if got := stat.Value(); got != 1 { t.Fatalf("got stat = %d, want = 1", got) } @@ -803,7 +810,7 @@ func TestDADStop(t *testing.T) { } // Should not have sent more than 1 NS message. - if got := s.Stats().ICMP.V6PacketsSent.NeighborSolicit.Value(); got > 1 { + if got := s.Stats().ICMP.V6.PacketsSent.NeighborSolicit.Value(); got > 1 { t.Errorf("got NeighborSolicit = %d, want <= 1", got) } }) @@ -982,7 +989,7 @@ func raBufWithOptsAndDHCPv6(ip tcpip.Address, rl uint16, managedAddress, otherCo pkt := header.ICMPv6(hdr.Prepend(icmpSize)) pkt.SetType(header.ICMPv6RouterAdvert) pkt.SetCode(0) - raPayload := pkt.NDPPayload() + raPayload := pkt.MessageBody() ra := header.NDPRouterAdvert(raPayload) // Populate the Router Lifetime. binary.BigEndian.PutUint16(raPayload[2:], rl) @@ -2162,8 +2169,8 @@ func TestNoAutoGenTempAddrForLinkLocal(t *testing.T) { NDPConfigs: ipv6.NDPConfigurations{ AutoGenTempGlobalAddresses: true, }, - NDPDisp: &ndpDisp, - AutoGenIPv6LinkLocal: true, + NDPDisp: &ndpDisp, + AutoGenLinkLocal: true, })}, }) @@ -2843,9 +2850,7 @@ func addrForNewConnectionTo(t *testing.T, s *stack.Stack, addr tcpip.FullAddress t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err) } defer ep.Close() - if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err) - } + ep.SocketOptions().SetV6Only(true) if err := ep.Connect(addr); err != nil { t.Fatalf("ep.Connect(%+v): %s", addr, err) } @@ -2879,9 +2884,7 @@ func addrForNewConnectionWithAddr(t *testing.T, s *stack.Stack, addr tcpip.FullA t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err) } defer ep.Close() - if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err) - } + ep.SocketOptions().SetV6Only(true) if err := ep.Bind(addr); err != nil { t.Fatalf("ep.Bind(%+v): %s", addr, err) } @@ -3250,9 +3253,7 @@ func TestAutoGenAddrJobDeprecation(t *testing.T) { t.Fatalf("s.NewEndpoint(%d, %d, _): %s", header.UDPProtocolNumber, header.IPv6ProtocolNumber, err) } defer ep.Close() - if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err) - } + ep.SocketOptions().SetV6Only(true) if err := ep.Connect(dstAddr); err != tcpip.ErrNoRoute { t.Errorf("got ep.Connect(%+v) = %s, want = %s", dstAddr, err, tcpip.ErrNoRoute) @@ -4044,9 +4045,9 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { ndpConfigs.AutoGenAddressConflictRetries = maxRetries s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ - AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, + AutoGenLinkLocal: addrType.autoGenLinkLocal, + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: func(_ tcpip.NICID, nicName string) string { return nicName @@ -4179,9 +4180,9 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ - AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, - NDPConfigs: addrType.ndpConfigs, - NDPDisp: &ndpDisp, + AutoGenLinkLocal: addrType.autoGenLinkLocal, + NDPConfigs: addrType.ndpConfigs, + NDPDisp: &ndpDisp, })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -4708,7 +4709,7 @@ func TestCleanupNDPState(t *testing.T) { } s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ - AutoGenIPv6LinkLocal: true, + AutoGenLinkLocal: true, NDPConfigs: ipv6.NDPConfigurations{ HandleRAs: true, DiscoverDefaultRouters: true, @@ -5174,113 +5175,99 @@ func TestRouterSolicitation(t *testing.T) { }, } - // This Run will not return until the parallel tests finish. - // - // We need this because we need to do some teardown work after the - // parallel tests complete. - // - // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for - // more details. - t.Run("group", func(t *testing.T) { - for _, test := range tests { - test := test - - t.Run(test.name, func(t *testing.T) { - t.Parallel() + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + clock := faketime.NewManualClock() + e := channelLinkWithHeaderLength{ + Endpoint: channel.New(int(test.maxRtrSolicit), 1280, test.linkAddr), + headerLength: test.linkHeaderLen, + } + e.Endpoint.LinkEPCapabilities |= stack.CapabilityResolutionRequired + waitForPkt := func(timeout time.Duration) { + t.Helper() - e := channelLinkWithHeaderLength{ - Endpoint: channel.New(int(test.maxRtrSolicit), 1280, test.linkAddr), - headerLength: test.linkHeaderLen, + clock.Advance(timeout) + p, ok := e.Read() + if !ok { + t.Fatal("expected router solicitation packet") } - e.Endpoint.LinkEPCapabilities |= stack.CapabilityResolutionRequired - waitForPkt := func(timeout time.Duration) { - t.Helper() - ctx, cancel := context.WithTimeout(context.Background(), timeout) - defer cancel() - p, ok := e.ReadContext(ctx) - if !ok { - t.Fatal("timed out waiting for packet") - return - } - if p.Proto != header.IPv6ProtocolNumber { - t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) - } + if p.Proto != header.IPv6ProtocolNumber { + t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) + } - // Make sure the right remote link address is used. - if want := header.EthernetAddressFromMulticastIPv6Address(header.IPv6AllRoutersMulticastAddress); p.Route.RemoteLinkAddress != want { - t.Errorf("got remote link address = %s, want = %s", p.Route.RemoteLinkAddress, want) - } + // Make sure the right remote link address is used. + if got, want := p.Route.RemoteLinkAddress(), header.EthernetAddressFromMulticastIPv6Address(header.IPv6AllRoutersMulticastAddress); got != want { + t.Errorf("got remote link address = %s, want = %s", got, want) + } - checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), - checker.SrcAddr(test.expectedSrcAddr), - checker.DstAddr(header.IPv6AllRoutersMulticastAddress), - checker.TTL(header.NDPHopLimit), - checker.NDPRS(checker.NDPRSOptions(test.expectedNDPOpts)), - ) + checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), + checker.SrcAddr(test.expectedSrcAddr), + checker.DstAddr(header.IPv6AllRoutersMulticastAddress), + checker.TTL(header.NDPHopLimit), + checker.NDPRS(checker.NDPRSOptions(test.expectedNDPOpts)), + ) - if l, want := p.Pkt.AvailableHeaderBytes(), int(test.linkHeaderLen); l != want { - t.Errorf("got p.Pkt.AvailableHeaderBytes() = %d; want = %d", l, want) - } - } - waitForNothing := func(timeout time.Duration) { - t.Helper() - ctx, cancel := context.WithTimeout(context.Background(), timeout) - defer cancel() - if _, ok := e.ReadContext(ctx); ok { - t.Fatal("unexpectedly got a packet") - } - } - s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ - NDPConfigs: ipv6.NDPConfigurations{ - MaxRtrSolicitations: test.maxRtrSolicit, - RtrSolicitationInterval: test.rtrSolicitInt, - MaxRtrSolicitationDelay: test.maxRtrSolicitDelay, - }, - })}, - }) - if err := s.CreateNIC(nicID, &e); err != nil { - t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + if l, want := p.Pkt.AvailableHeaderBytes(), int(test.linkHeaderLen); l != want { + t.Errorf("got p.Pkt.AvailableHeaderBytes() = %d; want = %d", l, want) } + } + waitForNothing := func(timeout time.Duration) { + t.Helper() - if addr := test.nicAddr; addr != "" { - if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr); err != nil { - t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr, err) - } + clock.Advance(timeout) + if p, ok := e.Read(); ok { + t.Fatalf("unexpectedly got a packet = %#v", p) } + } + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + MaxRtrSolicitations: test.maxRtrSolicit, + RtrSolicitationInterval: test.rtrSolicitInt, + MaxRtrSolicitationDelay: test.maxRtrSolicitDelay, + }, + })}, + Clock: clock, + }) + if err := s.CreateNIC(nicID, &e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } - // Make sure each RS is sent at the right time. - remaining := test.maxRtrSolicit - if remaining > 0 { - waitForPkt(test.effectiveMaxRtrSolicitDelay + defaultAsyncPositiveEventTimeout) - remaining-- + if addr := test.nicAddr; addr != "" { + if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, addr); err != nil { + t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, addr, err) } + } - for ; remaining > 0; remaining-- { - if test.effectiveRtrSolicitInt > defaultAsyncPositiveEventTimeout { - waitForNothing(test.effectiveRtrSolicitInt - defaultAsyncNegativeEventTimeout) - waitForPkt(defaultAsyncPositiveEventTimeout) - } else { - waitForPkt(test.effectiveRtrSolicitInt + defaultAsyncPositiveEventTimeout) - } - } + // Make sure each RS is sent at the right time. + remaining := test.maxRtrSolicit + if remaining > 0 { + waitForPkt(test.effectiveMaxRtrSolicitDelay) + remaining-- + } - // Make sure no more RS. - if test.effectiveRtrSolicitInt > test.effectiveMaxRtrSolicitDelay { - waitForNothing(test.effectiveRtrSolicitInt + defaultAsyncNegativeEventTimeout) + for ; remaining > 0; remaining-- { + if test.effectiveRtrSolicitInt > defaultAsyncPositiveEventTimeout { + waitForNothing(test.effectiveRtrSolicitInt - time.Nanosecond) + waitForPkt(time.Nanosecond) } else { - waitForNothing(test.effectiveMaxRtrSolicitDelay + defaultAsyncNegativeEventTimeout) + waitForPkt(test.effectiveRtrSolicitInt) } + } - // Make sure the counter got properly - // incremented. - if got, want := s.Stats().ICMP.V6PacketsSent.RouterSolicit.Value(), uint64(test.maxRtrSolicit); got != want { - t.Fatalf("got sent RouterSolicit = %d, want = %d", got, want) - } - }) - } - }) + // Make sure no more RS. + if test.effectiveRtrSolicitInt > test.effectiveMaxRtrSolicitDelay { + waitForNothing(test.effectiveRtrSolicitInt) + } else { + waitForNothing(test.effectiveMaxRtrSolicitDelay) + } + + if got, want := s.Stats().ICMP.V6.PacketsSent.RouterSolicit.Value(), uint64(test.maxRtrSolicit); got != want { + t.Fatalf("got sent RouterSolicit = %d, want = %d", got, want) + } + }) + } } func TestStopStartSolicitingRouters(t *testing.T) { diff --git a/pkg/tcpip/stack/neighbor_cache.go b/pkg/tcpip/stack/neighbor_cache.go index 177bf5516..317f6871d 100644 --- a/pkg/tcpip/stack/neighbor_cache.go +++ b/pkg/tcpip/stack/neighbor_cache.go @@ -24,9 +24,16 @@ import ( const neighborCacheSize = 512 // max entries per interface +// NeighborStats holds metrics for the neighbor table. +type NeighborStats struct { + // FailedEntryLookups counts the number of lookups performed on an entry in + // Failed state. + FailedEntryLookups *tcpip.StatCounter +} + // neighborCache maps IP addresses to link addresses. It uses the Least // Recently Used (LRU) eviction strategy to implement a bounded cache for -// dynmically acquired entries. It contains the state machine and configuration +// dynamically acquired entries. It contains the state machine and configuration // for running Neighbor Unreachability Detection (NUD). // // There are two types of entries in the neighbor cache: @@ -175,14 +182,15 @@ func (n *neighborCache) removeWaker(addr tcpip.Address, waker *sleep.Waker) { // entries returns all entries in the neighbor cache. func (n *neighborCache) entries() []NeighborEntry { - entries := make([]NeighborEntry, 0, len(n.cache)) n.mu.RLock() + defer n.mu.RUnlock() + + entries := make([]NeighborEntry, 0, len(n.cache)) for _, entry := range n.cache { entry.mu.RLock() entries = append(entries, entry.neigh) entry.mu.RUnlock() } - n.mu.RUnlock() return entries } @@ -226,6 +234,8 @@ func (n *neighborCache) addStaticEntry(addr tcpip.Address, linkAddr tcpip.LinkAd } // removeEntryLocked removes the specified entry from the neighbor cache. +// +// Prerequisite: n.mu and entry.mu MUST be locked. func (n *neighborCache) removeEntryLocked(entry *neighborEntry) { if entry.neigh.State != Static { n.dynamic.lru.Remove(entry) diff --git a/pkg/tcpip/stack/neighbor_cache_test.go b/pkg/tcpip/stack/neighbor_cache_test.go index ed33418f3..732a299f7 100644 --- a/pkg/tcpip/stack/neighbor_cache_test.go +++ b/pkg/tcpip/stack/neighbor_cache_test.go @@ -80,17 +80,20 @@ func entryDiffOptsWithSort() []cmp.Option { func newTestNeighborCache(nudDisp NUDDispatcher, config NUDConfigurations, clock tcpip.Clock) *neighborCache { config.resetInvalidFields() rng := rand.New(rand.NewSource(time.Now().UnixNano())) - return &neighborCache{ + neigh := &neighborCache{ nic: &NIC{ stack: &Stack{ clock: clock, nudDisp: nudDisp, }, - id: 1, + id: 1, + stats: makeNICStats(), }, state: NewNUDState(config, rng), cache: make(map[tcpip.Address]*neighborEntry, neighborCacheSize), } + neigh.nic.neigh = neigh + return neigh } // testEntryStore contains a set of IP to NeighborEntry mappings. diff --git a/pkg/tcpip/stack/neighbor_entry.go b/pkg/tcpip/stack/neighbor_entry.go index 493e48031..32399b4f5 100644 --- a/pkg/tcpip/stack/neighbor_entry.go +++ b/pkg/tcpip/stack/neighbor_entry.go @@ -258,7 +258,7 @@ func (e *neighborEntry) setStateLocked(next NeighborState) { case Failed: e.notifyWakersLocked() - e.job = e.nic.stack.newJob(&e.mu, func() { + e.job = e.nic.stack.newJob(&doubleLock{first: &e.nic.neigh.mu, second: &e.mu}, func() { e.nic.neigh.removeEntryLocked(e) }) e.job.Schedule(config.UnreachableTime) @@ -347,9 +347,10 @@ func (e *neighborEntry) handlePacketQueuedLocked(localAddr tcpip.Address) { e.setStateLocked(Delay) e.dispatchChangeEventLocked() - case Incomplete, Reachable, Delay, Probe, Static, Failed: + case Incomplete, Reachable, Delay, Probe, Static: // Do nothing - + case Failed: + e.nic.stats.Neighbor.FailedEntryLookups.Increment() default: panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State)) } @@ -511,3 +512,23 @@ func (e *neighborEntry) handleUpperLevelConfirmationLocked() { panic(fmt.Sprintf("Invalid cache entry state: %s", e.neigh.State)) } } + +// doubleLock combines two locks into one while maintaining lock ordering. +// +// TODO(gvisor.dev/issue/4796): Remove this once subsequent traffic to a Failed +// neighbor is allowed. +type doubleLock struct { + first, second sync.Locker +} + +// Lock locks both locks in order: first then second. +func (l *doubleLock) Lock() { + l.first.Lock() + l.second.Lock() +} + +// Unlock unlocks both locks in reverse order: second then first. +func (l *doubleLock) Unlock() { + l.second.Unlock() + l.first.Unlock() +} diff --git a/pkg/tcpip/stack/neighbor_entry_test.go b/pkg/tcpip/stack/neighbor_entry_test.go index c2b763325..c497d3932 100644 --- a/pkg/tcpip/stack/neighbor_entry_test.go +++ b/pkg/tcpip/stack/neighbor_entry_test.go @@ -89,7 +89,7 @@ func eventDiffOptsWithSort() []cmp.Option { // | Stale | Reachable | Solicited confirmation w/o address | Notify wakers | Changed | // | Stale | Stale | Override confirmation | Update LinkAddr | Changed | // | Stale | Stale | Probe w/ different address | Update LinkAddr | Changed | -// | Stale | Delay | Packet sent | | Changed | +// | Stale | Delay | Packet queued | | Changed | // | Delay | Reachable | Upper-layer confirmation | | Changed | // | Delay | Reachable | Solicited override confirmation | Update LinkAddr | Changed | // | Delay | Reachable | Solicited confirmation w/o address | Notify wakers | Changed | @@ -101,6 +101,7 @@ func eventDiffOptsWithSort() []cmp.Option { // | Probe | Stale | Probe or confirmation w/ different address | | Changed | // | Probe | Probe | Retransmit timer expired | Send probe | Changed | // | Probe | Failed | Max probes sent without reply | Notify wakers | Removed | +// | Failed | Failed | Packet queued | | | // | Failed | | Unreachability timer expired | Delete entry | | type testEntryEventType uint8 @@ -228,6 +229,7 @@ func entryTestSetup(c NUDConfigurations) (*neighborEntry, *testNUDDispatcher, *e clock: clock, nudDisp: &disp, }, + stats: makeNICStats(), } nic.networkEndpoints = map[tcpip.NetworkProtocolNumber]NetworkEndpoint{ header.IPv6ProtocolNumber: (&testIPv6Protocol{}).NewEndpoint(&nic, nil, nil, nil), @@ -3433,6 +3435,146 @@ func TestEntryProbeToFailed(t *testing.T) { nudDisp.mu.Unlock() } +func TestEntryFailedToFailed(t *testing.T) { + c := DefaultNUDConfigurations() + c.MaxMulticastProbes = 3 + c.MaxUnicastProbes = 3 + e, nudDisp, linkRes, clock := entryTestSetup(c) + + // Verify the cache contains the entry. + if _, ok := e.nic.neigh.cache[entryTestAddr1]; !ok { + t.Errorf("expected entry %q to exist in the neighbor cache", entryTestAddr1) + } + + // TODO(gvisor.dev/issue/4872): Use helper functions to start entry tests in + // their expected state. + e.mu.Lock() + e.handlePacketQueuedLocked(entryTestAddr2) + e.mu.Unlock() + + runImmediatelyScheduledJobs(clock) + { + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + LocalAddress: entryTestAddr2, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.probes = nil + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + } + + e.mu.Lock() + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ + Solicited: false, + Override: false, + IsRouter: false, + }) + e.handlePacketQueuedLocked(entryTestAddr2) + e.mu.Unlock() + + waitFor := c.DelayFirstProbeTime + c.RetransmitTimer*time.Duration(c.MaxUnicastProbes) + clock.Advance(waitFor) + { + wantProbes := []entryTestProbeInfo{ + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + }, + { + RemoteAddress: entryTestAddr1, + RemoteLinkAddress: entryTestLinkAddr1, + }, + } + linkRes.mu.Lock() + diff := cmp.Diff(linkRes.probes, wantProbes) + linkRes.mu.Unlock() + if diff != "" { + t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) + } + } + + wantEvents := []testEntryEventInfo{ + { + EventType: entryTestAdded, + NICID: entryTestNICID, + Entry: NeighborEntry{ + Addr: entryTestAddr1, + LinkAddr: tcpip.LinkAddress(""), + State: Incomplete, + }, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Entry: NeighborEntry{ + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Stale, + }, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Entry: NeighborEntry{ + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Delay, + }, + }, + { + EventType: entryTestChanged, + NICID: entryTestNICID, + Entry: NeighborEntry{ + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + }, + { + EventType: entryTestRemoved, + NICID: entryTestNICID, + Entry: NeighborEntry{ + Addr: entryTestAddr1, + LinkAddr: entryTestLinkAddr1, + State: Probe, + }, + }, + } + nudDisp.mu.Lock() + if diff := cmp.Diff(nudDisp.events, wantEvents, eventDiffOpts()...); diff != "" { + t.Errorf("nud dispatcher events mismatch (-got, +want):\n%s", diff) + } + nudDisp.mu.Unlock() + + failedLookups := e.nic.stats.Neighbor.FailedEntryLookups + if got := failedLookups.Value(); got != 0 { + t.Errorf("got Neighbor.FailedEntryLookups = %d, want = 0", got) + } + + e.mu.Lock() + // Verify queuing a packet to the entry immediately fails. + e.handlePacketQueuedLocked(entryTestAddr2) + state := e.neigh.State + e.mu.Unlock() + if state != Failed { + t.Errorf("got e.neigh.State = %q, want = %q", state, Failed) + } + + if got := failedLookups.Value(); got != 1 { + t.Errorf("got Neighbor.FailedEntryLookups = %d, want = 1", got) + } +} + func TestEntryFailedGetsDeleted(t *testing.T) { c := DefaultNUDConfigurations() c.MaxMulticastProbes = 3 diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 3e6ceff28..5887aa1ed 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -60,12 +60,14 @@ type NIC struct { } } -// NICStats includes transmitted and received stats. +// NICStats hold statistics for a NIC. type NICStats struct { Tx DirectionStats Rx DirectionStats DisabledRx DirectionStats + + Neighbor NeighborStats } func makeNICStats() NICStats { @@ -265,7 +267,7 @@ func (n *NIC) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumb if ch, err := r.Resolve(nil); err != nil { if err == tcpip.ErrWouldBlock { r := r.Clone() - n.stack.linkResQueue.enqueue(ch, &r, protocol, pkt) + n.stack.linkResQueue.enqueue(ch, r, protocol, pkt) return nil } return err @@ -277,9 +279,9 @@ func (n *NIC) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumb // WritePacketToRemote implements NetworkInterface. func (n *NIC) WritePacketToRemote(remoteLinkAddr tcpip.LinkAddress, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error { r := Route{ - NetProto: protocol, - RemoteLinkAddress: remoteLinkAddr, + NetProto: protocol, } + r.ResolveWith(remoteLinkAddr) return n.writePacket(&r, gso, protocol, pkt) } @@ -561,8 +563,7 @@ func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address return tcpip.ErrNotSupported } - _, err := gep.JoinGroup(addr) - return err + return gep.JoinGroup(addr) } // leaveGroup decrements the count for the given multicast address, and when it @@ -578,11 +579,7 @@ func (n *NIC) leaveGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Addres return tcpip.ErrNotSupported } - if _, err := gep.LeaveGroup(addr); err != nil { - return err - } - - return nil + return gep.LeaveGroup(addr) } // isInGroup returns true if n has joined the multicast group addr. diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 2cb13c6fa..b334e27c4 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -259,15 +259,6 @@ const ( PacketLoop ) -// NetOptions is an interface that allows us to pass network protocol specific -// options through the Stack layer code. -type NetOptions interface { - // SizeWithPadding returns the amount of memory that must be allocated to - // hold the options given that the value must be rounded up to the next - // multiple of 4 bytes. - SizeWithPadding() int -} - // NetworkHeaderParams are the header parameters given as input by the // transport endpoint to the network. type NetworkHeaderParams struct { @@ -279,10 +270,6 @@ type NetworkHeaderParams struct { // TOS refers to TypeOfService or TrafficClass field of the IP-header. TOS uint8 - - // Options is a set of options to add to a network header (or nil). - // It will be protocol specific opaque information from higher layers. - Options NetOptions } // GroupAddressableEndpoint is an endpoint that supports group addressing. @@ -291,14 +278,10 @@ type NetworkHeaderParams struct { // endpoints may associate themselves with the same identifier (group address). type GroupAddressableEndpoint interface { // JoinGroup joins the specified group. - // - // Returns true if the group was newly joined. - JoinGroup(group tcpip.Address) (bool, *tcpip.Error) + JoinGroup(group tcpip.Address) *tcpip.Error // LeaveGroup attempts to leave the specified group. - // - // Returns tcpip.ErrBadLocalAddress if the endpoint has not joined the group. - LeaveGroup(group tcpip.Address) (bool, *tcpip.Error) + LeaveGroup(group tcpip.Address) *tcpip.Error // IsInGroup returns true if the endpoint is a member of the specified group. IsInGroup(group tcpip.Address) bool @@ -739,10 +722,6 @@ type LinkEndpoint interface { // endpoint. Capabilities() LinkEndpointCapabilities - // WriteRawPacket writes a packet directly to the link. The packet - // should already have an ethernet header. It takes ownership of vv. - WriteRawPacket(vv buffer.VectorisedView) *tcpip.Error - // Attach attaches the data link layer endpoint to the network-layer // dispatcher of the stack. // diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 53cb6694f..de5fe6ffe 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -18,19 +18,22 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" ) // Route represents a route through the networking stack to a given destination. +// +// It is safe to call Route's methods from multiple goroutines. +// +// The exported fields are immutable. +// +// TODO(gvisor.dev/issue/4902): Unexpose immutable fields. type Route struct { // RemoteAddress is the final destination of the route. RemoteAddress tcpip.Address - // RemoteLinkAddress is the link-layer (MAC) address of the - // final destination of the route. - RemoteLinkAddress tcpip.LinkAddress - // LocalAddress is the local address where the route starts. LocalAddress tcpip.Address @@ -52,8 +55,16 @@ type Route struct { // address's assigned status without the NIC. localAddressNIC *NIC - // localAddressEndpoint is the local address this route is associated with. - localAddressEndpoint AssignableAddressEndpoint + mu struct { + sync.RWMutex + + // localAddressEndpoint is the local address this route is associated with. + localAddressEndpoint AssignableAddressEndpoint + + // remoteLinkAddress is the link-layer (MAC) address of the next hop in the + // route. + remoteLinkAddress tcpip.LinkAddress + } // outgoingNIC is the interface this route uses to write packets. outgoingNIC *NIC @@ -71,22 +82,24 @@ type Route struct { // ownership of the provided local address. // // Returns an empty route if validation fails. -func constructAndValidateRoute(netProto tcpip.NetworkProtocolNumber, addressEndpoint AssignableAddressEndpoint, localAddressNIC, outgoingNIC *NIC, gateway, remoteAddr tcpip.Address, handleLocal, multicastLoop bool) Route { - addrWithPrefix := addressEndpoint.AddressWithPrefix() +func constructAndValidateRoute(netProto tcpip.NetworkProtocolNumber, addressEndpoint AssignableAddressEndpoint, localAddressNIC, outgoingNIC *NIC, gateway, localAddr, remoteAddr tcpip.Address, handleLocal, multicastLoop bool) *Route { + if len(localAddr) == 0 { + localAddr = addressEndpoint.AddressWithPrefix().Address + } - if localAddressNIC != outgoingNIC && header.IsV6LinkLocalAddress(addrWithPrefix.Address) { + if localAddressNIC != outgoingNIC && header.IsV6LinkLocalAddress(localAddr) { addressEndpoint.DecRef() - return Route{} + return nil } // If no remote address is provided, use the local address. if len(remoteAddr) == 0 { - remoteAddr = addrWithPrefix.Address + remoteAddr = localAddr } r := makeRoute( netProto, - addrWithPrefix.Address, + localAddr, remoteAddr, outgoingNIC, localAddressNIC, @@ -99,8 +112,8 @@ func constructAndValidateRoute(netProto tcpip.NetworkProtocolNumber, addressEndp // broadcast it. if len(gateway) > 0 { r.NextHop = gateway - } else if subnet := addrWithPrefix.Subnet(); subnet.IsBroadcast(remoteAddr) { - r.RemoteLinkAddress = header.EthernetBroadcastAddress + } else if subnet := addressEndpoint.Subnet(); subnet.IsBroadcast(remoteAddr) { + r.ResolveWith(header.EthernetBroadcastAddress) } return r @@ -108,11 +121,15 @@ func constructAndValidateRoute(netProto tcpip.NetworkProtocolNumber, addressEndp // makeRoute initializes a new route. It takes ownership of the provided // AssignableAddressEndpoint. -func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, outgoingNIC, localAddressNIC *NIC, localAddressEndpoint AssignableAddressEndpoint, handleLocal, multicastLoop bool) Route { +func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, outgoingNIC, localAddressNIC *NIC, localAddressEndpoint AssignableAddressEndpoint, handleLocal, multicastLoop bool) *Route { if localAddressNIC.stack != outgoingNIC.stack { panic(fmt.Sprintf("cannot create a route with NICs from different stacks")) } + if len(localAddr) == 0 { + localAddr = localAddressEndpoint.AddressWithPrefix().Address + } + loop := PacketOut // TODO(gvisor.dev/issue/4689): Loopback interface loops back packets at the @@ -133,18 +150,21 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip return makeRouteInner(netProto, localAddr, remoteAddr, outgoingNIC, localAddressNIC, localAddressEndpoint, loop) } -func makeRouteInner(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, outgoingNIC, localAddressNIC *NIC, localAddressEndpoint AssignableAddressEndpoint, loop PacketLooping) Route { - r := Route{ - NetProto: netProto, - LocalAddress: localAddr, - LocalLinkAddress: outgoingNIC.LinkEndpoint.LinkAddress(), - RemoteAddress: remoteAddr, - localAddressNIC: localAddressNIC, - localAddressEndpoint: localAddressEndpoint, - outgoingNIC: outgoingNIC, - Loop: loop, +func makeRouteInner(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, outgoingNIC, localAddressNIC *NIC, localAddressEndpoint AssignableAddressEndpoint, loop PacketLooping) *Route { + r := &Route{ + NetProto: netProto, + LocalAddress: localAddr, + LocalLinkAddress: outgoingNIC.LinkEndpoint.LinkAddress(), + RemoteAddress: remoteAddr, + localAddressNIC: localAddressNIC, + outgoingNIC: outgoingNIC, + Loop: loop, } + r.mu.Lock() + r.mu.localAddressEndpoint = localAddressEndpoint + r.mu.Unlock() + if r.outgoingNIC.LinkEndpoint.Capabilities()&CapabilityResolutionRequired != 0 { if linkRes, ok := r.outgoingNIC.stack.linkAddrResolvers[r.NetProto]; ok { r.linkRes = linkRes @@ -159,7 +179,7 @@ func makeRouteInner(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr // provided AssignableAddressEndpoint. // // A local route is a route to a destination that is local to the stack. -func makeLocalRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, outgoingNIC, localAddressNIC *NIC, localAddressEndpoint AssignableAddressEndpoint) Route { +func makeLocalRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, outgoingNIC, localAddressNIC *NIC, localAddressEndpoint AssignableAddressEndpoint) *Route { loop := PacketLoop // TODO(gvisor.dev/issue/4689): Loopback interface loops back packets at the // link endpoint level. We can remove this check once loopback interfaces @@ -170,6 +190,14 @@ func makeLocalRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr return makeRouteInner(netProto, localAddr, remoteAddr, outgoingNIC, localAddressNIC, localAddressEndpoint, loop) } +// RemoteLinkAddress returns the link-layer (MAC) address of the next hop in +// the route. +func (r *Route) RemoteLinkAddress() tcpip.LinkAddress { + r.mu.RLock() + defer r.mu.RUnlock() + return r.mu.remoteLinkAddress +} + // NICID returns the id of the NIC from which this route originates. func (r *Route) NICID() tcpip.NICID { return r.outgoingNIC.ID() @@ -231,7 +259,9 @@ func (r *Route) GSOMaxSize() uint32 { // ResolveWith immediately resolves a route with the specified remote link // address. func (r *Route) ResolveWith(addr tcpip.LinkAddress) { - r.RemoteLinkAddress = addr + r.mu.Lock() + defer r.mu.Unlock() + r.mu.remoteLinkAddress = addr } // Resolve attempts to resolve the link address if necessary. Returns ErrWouldBlock in @@ -244,7 +274,10 @@ func (r *Route) ResolveWith(addr tcpip.LinkAddress) { // // The NIC r uses must not be locked. func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) { - if !r.IsResolutionRequired() { + r.mu.Lock() + defer r.mu.Unlock() + + if !r.isResolutionRequiredRLocked() { // Nothing to do if there is no cache (which does the resolution on cache miss) or // link address is already known. return nil, nil @@ -254,7 +287,7 @@ func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) { if nextAddr == "" { // Local link address is already known. if r.RemoteAddress == r.LocalAddress { - r.RemoteLinkAddress = r.LocalLinkAddress + r.mu.remoteLinkAddress = r.LocalLinkAddress return nil, nil } nextAddr = r.RemoteAddress @@ -272,7 +305,7 @@ func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) { if err != nil { return ch, err } - r.RemoteLinkAddress = entry.LinkAddr + r.mu.remoteLinkAddress = entry.LinkAddr return nil, nil } @@ -280,7 +313,7 @@ func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) { if err != nil { return ch, err } - r.RemoteLinkAddress = linkAddr + r.mu.remoteLinkAddress = linkAddr return nil, nil } @@ -309,7 +342,13 @@ func (r *Route) local() bool { // // The NICs the route is associated with must not be locked. func (r *Route) IsResolutionRequired() bool { - if !r.isValidForOutgoing() || r.RemoteLinkAddress != "" || r.local() { + r.mu.RLock() + defer r.mu.RUnlock() + return r.isResolutionRequiredRLocked() +} + +func (r *Route) isResolutionRequiredRLocked() bool { + if !r.isValidForOutgoingRLocked() || r.mu.remoteLinkAddress != "" || r.local() { return false } @@ -317,11 +356,18 @@ func (r *Route) IsResolutionRequired() bool { } func (r *Route) isValidForOutgoing() bool { + r.mu.RLock() + defer r.mu.RUnlock() + return r.isValidForOutgoingRLocked() +} + +func (r *Route) isValidForOutgoingRLocked() bool { if !r.outgoingNIC.Enabled() { return false } - if !r.localAddressNIC.isValidForOutgoing(r.localAddressEndpoint) { + localAddressEndpoint := r.mu.localAddressEndpoint + if localAddressEndpoint == nil || !r.localAddressNIC.isValidForOutgoing(localAddressEndpoint) { return false } @@ -375,37 +421,44 @@ func (r *Route) MTU() uint32 { // Release frees all resources associated with the route. func (r *Route) Release() { - if r.localAddressEndpoint != nil { - r.localAddressEndpoint.DecRef() - r.localAddressEndpoint = nil + r.mu.Lock() + defer r.mu.Unlock() + + if r.mu.localAddressEndpoint != nil { + r.mu.localAddressEndpoint.DecRef() + r.mu.localAddressEndpoint = nil } } // Clone clones the route. -func (r *Route) Clone() Route { - if r.localAddressEndpoint != nil { - if !r.localAddressEndpoint.IncRef() { - panic(fmt.Sprintf("failed to increment reference count for local address endpoint = %s", r.LocalAddress)) - } +func (r *Route) Clone() *Route { + r.mu.RLock() + defer r.mu.RUnlock() + + newRoute := &Route{ + RemoteAddress: r.RemoteAddress, + LocalAddress: r.LocalAddress, + LocalLinkAddress: r.LocalLinkAddress, + NextHop: r.NextHop, + NetProto: r.NetProto, + Loop: r.Loop, + localAddressNIC: r.localAddressNIC, + outgoingNIC: r.outgoingNIC, + linkCache: r.linkCache, + linkRes: r.linkRes, } - return *r -} -// MakeLoopedRoute duplicates the given route with special handling for routes -// used for sending multicast or broadcast packets. In those cases the -// multicast/broadcast address is the remote address when sending out, but for -// incoming (looped) packets it becomes the local address. Similarly, the local -// interface address that was the local address going out becomes the remote -// address coming in. This is different to unicast routes where local and -// remote addresses remain the same as they identify location (local vs remote) -// not direction (source vs destination). -func (r *Route) MakeLoopedRoute() Route { - l := r.Clone() - if r.RemoteAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(r.RemoteAddress) || header.IsV6MulticastAddress(r.RemoteAddress) { - l.RemoteAddress, l.LocalAddress = l.LocalAddress, l.RemoteAddress - l.RemoteLinkAddress = l.LocalLinkAddress + newRoute.mu.Lock() + defer newRoute.mu.Unlock() + newRoute.mu.localAddressEndpoint = r.mu.localAddressEndpoint + if newRoute.mu.localAddressEndpoint != nil { + if !newRoute.mu.localAddressEndpoint.IncRef() { + panic(fmt.Sprintf("failed to increment reference count for local address endpoint = %s", newRoute.LocalAddress)) + } } - return l + newRoute.mu.remoteLinkAddress = r.mu.remoteLinkAddress + + return newRoute } // Stack returns the instance of the Stack that owns this route. @@ -418,7 +471,14 @@ func (r *Route) isV4Broadcast(addr tcpip.Address) bool { return true } - subnet := r.localAddressEndpoint.Subnet() + r.mu.RLock() + localAddressEndpoint := r.mu.localAddressEndpoint + r.mu.RUnlock() + if localAddressEndpoint == nil { + return false + } + + subnet := localAddressEndpoint.Subnet() return subnet.IsBroadcast(addr) } @@ -428,27 +488,3 @@ func (r *Route) IsOutboundBroadcast() bool { // Only IPv4 has a notion of broadcast. return r.isV4Broadcast(r.RemoteAddress) } - -// isInboundBroadcast returns true if the route is for an inbound broadcast -// packet. -func (r *Route) isInboundBroadcast() bool { - // Only IPv4 has a notion of broadcast. - return r.isV4Broadcast(r.LocalAddress) -} - -// ReverseRoute returns new route with given source and destination address. -func (r *Route) ReverseRoute(src tcpip.Address, dst tcpip.Address) Route { - return Route{ - NetProto: r.NetProto, - LocalAddress: dst, - LocalLinkAddress: r.RemoteLinkAddress, - RemoteAddress: src, - RemoteLinkAddress: r.LocalLinkAddress, - Loop: r.Loop, - localAddressNIC: r.localAddressNIC, - localAddressEndpoint: r.localAddressEndpoint, - outgoingNIC: r.outgoingNIC, - linkCache: r.linkCache, - linkRes: r.linkRes, - } -} diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index e0025e0a9..dc4f5b3e7 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -1118,6 +1118,16 @@ func (s *Stack) AddAddress(id tcpip.NICID, protocol tcpip.NetworkProtocolNumber, return s.AddAddressWithOptions(id, protocol, addr, CanBePrimaryEndpoint) } +// AddAddressWithPrefix is the same as AddAddress, but allows you to specify +// the address prefix. +func (s *Stack) AddAddressWithPrefix(id tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.AddressWithPrefix) *tcpip.Error { + ap := tcpip.ProtocolAddress{ + Protocol: protocol, + AddressWithPrefix: addr, + } + return s.AddProtocolAddressWithOptions(id, ap, CanBePrimaryEndpoint) +} + // AddProtocolAddress adds a new network-layer protocol address to the // specified NIC. func (s *Stack) AddProtocolAddress(id tcpip.NICID, protocolAddress tcpip.ProtocolAddress) *tcpip.Error { @@ -1208,10 +1218,10 @@ func (s *Stack) getAddressEP(nic *NIC, localAddr, remoteAddr tcpip.Address, netP // from the specified NIC. // // Precondition: s.mu must be read locked. -func (s *Stack) findLocalRouteFromNICRLocked(localAddressNIC *NIC, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (route Route, ok bool) { +func (s *Stack) findLocalRouteFromNICRLocked(localAddressNIC *NIC, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) *Route { localAddressEndpoint := localAddressNIC.getAddressOrCreateTempInner(netProto, localAddr, false /* createTemp */, NeverPrimaryEndpoint) if localAddressEndpoint == nil { - return Route{}, false + return nil } var outgoingNIC *NIC @@ -1235,12 +1245,12 @@ func (s *Stack) findLocalRouteFromNICRLocked(localAddressNIC *NIC, localAddr, re // route. if outgoingNIC == nil { localAddressEndpoint.DecRef() - return Route{}, false + return nil } r := makeLocalRoute( netProto, - localAddressEndpoint.AddressWithPrefix().Address, + localAddr, remoteAddr, outgoingNIC, localAddressNIC, @@ -1249,10 +1259,10 @@ func (s *Stack) findLocalRouteFromNICRLocked(localAddressNIC *NIC, localAddr, re if r.IsOutboundBroadcast() { r.Release() - return Route{}, false + return nil } - return r, true + return r } // findLocalRouteRLocked returns a local route. @@ -1261,26 +1271,26 @@ func (s *Stack) findLocalRouteFromNICRLocked(localAddressNIC *NIC, localAddr, re // is, a local route is a route where packets never have to leave the stack. // // Precondition: s.mu must be read locked. -func (s *Stack) findLocalRouteRLocked(localAddressNICID tcpip.NICID, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (route Route, ok bool) { +func (s *Stack) findLocalRouteRLocked(localAddressNICID tcpip.NICID, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) *Route { if len(localAddr) == 0 { localAddr = remoteAddr } if localAddressNICID == 0 { for _, localAddressNIC := range s.nics { - if r, ok := s.findLocalRouteFromNICRLocked(localAddressNIC, localAddr, remoteAddr, netProto); ok { - return r, true + if r := s.findLocalRouteFromNICRLocked(localAddressNIC, localAddr, remoteAddr, netProto); r != nil { + return r } } - return Route{}, false + return nil } if localAddressNIC, ok := s.nics[localAddressNICID]; ok { return s.findLocalRouteFromNICRLocked(localAddressNIC, localAddr, remoteAddr, netProto) } - return Route{}, false + return nil } // FindRoute creates a route to the given destination address, leaving through @@ -1294,7 +1304,7 @@ func (s *Stack) findLocalRouteRLocked(localAddressNICID tcpip.NICID, localAddr, // If no local address is provided, the stack will select a local address. If no // remote address is provided, the stack wil use a remote address equal to the // local address. -func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber, multicastLoop bool) (Route, *tcpip.Error) { +func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber, multicastLoop bool) (*Route, *tcpip.Error) { s.mu.RLock() defer s.mu.RUnlock() @@ -1305,7 +1315,7 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n needRoute := !(isLocalBroadcast || isMulticast || isLinkLocal || isLoopback) if s.handleLocal && !isMulticast && !isLocalBroadcast { - if r, ok := s.findLocalRouteRLocked(id, localAddr, remoteAddr, netProto); ok { + if r := s.findLocalRouteRLocked(id, localAddr, remoteAddr, netProto); r != nil { return r, nil } } @@ -1317,7 +1327,7 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n if addressEndpoint := s.getAddressEP(nic, localAddr, remoteAddr, netProto); addressEndpoint != nil { return makeRoute( netProto, - addressEndpoint.AddressWithPrefix().Address, + localAddr, remoteAddr, nic, /* outboundNIC */ nic, /* localAddressNIC*/ @@ -1329,9 +1339,9 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n } if isLoopback { - return Route{}, tcpip.ErrBadLocalAddress + return nil, tcpip.ErrBadLocalAddress } - return Route{}, tcpip.ErrNetworkUnreachable + return nil, tcpip.ErrNetworkUnreachable } canForward := s.Forwarding(netProto) && !header.IsV6LinkLocalAddress(localAddr) && !isLinkLocal @@ -1354,8 +1364,8 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n if needRoute { gateway = route.Gateway } - r := constructAndValidateRoute(netProto, addressEndpoint, nic /* outgoingNIC */, nic /* outgoingNIC */, gateway, remoteAddr, s.handleLocal, multicastLoop) - if r == (Route{}) { + r := constructAndValidateRoute(netProto, addressEndpoint, nic /* outgoingNIC */, nic /* outgoingNIC */, gateway, localAddr, remoteAddr, s.handleLocal, multicastLoop) + if r == nil { panic(fmt.Sprintf("non-forwarding route validation failed with route table entry = %#v, id = %d, localAddr = %s, remoteAddr = %s", route, id, localAddr, remoteAddr)) } return r, nil @@ -1391,13 +1401,13 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n if id != 0 { if aNIC, ok := s.nics[id]; ok { if addressEndpoint := s.getAddressEP(aNIC, localAddr, remoteAddr, netProto); addressEndpoint != nil { - if r := constructAndValidateRoute(netProto, addressEndpoint, aNIC /* localAddressNIC */, nic /* outgoingNIC */, gateway, remoteAddr, s.handleLocal, multicastLoop); r != (Route{}) { + if r := constructAndValidateRoute(netProto, addressEndpoint, aNIC /* localAddressNIC */, nic /* outgoingNIC */, gateway, localAddr, remoteAddr, s.handleLocal, multicastLoop); r != nil { return r, nil } } } - return Route{}, tcpip.ErrNoRoute + return nil, tcpip.ErrNoRoute } if id == 0 { @@ -1409,7 +1419,7 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n continue } - if r := constructAndValidateRoute(netProto, addressEndpoint, aNIC /* localAddressNIC */, nic /* outgoingNIC */, gateway, remoteAddr, s.handleLocal, multicastLoop); r != (Route{}) { + if r := constructAndValidateRoute(netProto, addressEndpoint, aNIC /* localAddressNIC */, nic /* outgoingNIC */, gateway, localAddr, remoteAddr, s.handleLocal, multicastLoop); r != nil { return r, nil } } @@ -1417,12 +1427,12 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n } if needRoute { - return Route{}, tcpip.ErrNoRoute + return nil, tcpip.ErrNoRoute } if header.IsV6LoopbackAddress(remoteAddr) { - return Route{}, tcpip.ErrBadLocalAddress + return nil, tcpip.ErrBadLocalAddress } - return Route{}, tcpip.ErrNetworkUnreachable + return nil, tcpip.ErrNetworkUnreachable } // CheckNetworkProtocol checks if a given network protocol is enabled in the @@ -1810,49 +1820,20 @@ func (s *Stack) unregisterPacketEndpointLocked(nicID tcpip.NICID, netProto tcpip nic.unregisterPacketEndpoint(netProto, ep) } -// WritePacket writes data directly to the specified NIC. It adds an ethernet -// header based on the arguments. -func (s *Stack) WritePacket(nicID tcpip.NICID, dst tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, payload buffer.VectorisedView) *tcpip.Error { +// WritePacketToRemote writes a payload on the specified NIC using the provided +// network protocol and remote link address. +func (s *Stack) WritePacketToRemote(nicID tcpip.NICID, remote tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, payload buffer.VectorisedView) *tcpip.Error { s.mu.Lock() nic, ok := s.nics[nicID] s.mu.Unlock() if !ok { return tcpip.ErrUnknownDevice } - - // Add our own fake ethernet header. - ethFields := header.EthernetFields{ - SrcAddr: nic.LinkEndpoint.LinkAddress(), - DstAddr: dst, - Type: netProto, - } - fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) - fakeHeader.Encode(ðFields) - vv := buffer.View(fakeHeader).ToVectorisedView() - vv.Append(payload) - - if err := nic.LinkEndpoint.WriteRawPacket(vv); err != nil { - return err - } - - return nil -} - -// WriteRawPacket writes data directly to the specified NIC without adding any -// headers. -func (s *Stack) WriteRawPacket(nicID tcpip.NICID, payload buffer.VectorisedView) *tcpip.Error { - s.mu.Lock() - nic, ok := s.nics[nicID] - s.mu.Unlock() - if !ok { - return tcpip.ErrUnknownDevice - } - - if err := nic.LinkEndpoint.WriteRawPacket(payload); err != nil { - return err - } - - return nil + pkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(nic.MaxHeaderLength()), + Data: payload, + }) + return nic.WritePacketToRemote(remote, nil, netProto, pkt) } // NetworkProtocolInstance returns the protocol instance in the stack for the @@ -1912,7 +1893,6 @@ func (s *Stack) RemoveTCPProbe() { // JoinGroup joins the given multicast group on the given NIC. func (s *Stack) JoinGroup(protocol tcpip.NetworkProtocolNumber, nicID tcpip.NICID, multicastAddr tcpip.Address) *tcpip.Error { - // TODO: notify network of subscription via igmp protocol. s.mu.RLock() defer s.mu.RUnlock() @@ -2159,3 +2139,43 @@ func (s *Stack) networkProtocolNumbers() []tcpip.NetworkProtocolNumber { } return protos } + +func isSubnetBroadcastOnNIC(nic *NIC, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) bool { + addressEndpoint := nic.getAddressOrCreateTempInner(protocol, addr, false /* createTemp */, NeverPrimaryEndpoint) + if addressEndpoint == nil { + return false + } + + subnet := addressEndpoint.Subnet() + addressEndpoint.DecRef() + return subnet.IsBroadcast(addr) +} + +// IsSubnetBroadcast returns true if the provided address is a subnet-local +// broadcast address on the specified NIC and protocol. +// +// Returns false if the NIC is unknown or if the protocol is unknown or does +// not support addressing. +// +// If the NIC is not specified, the stack will check all NICs. +func (s *Stack) IsSubnetBroadcast(nicID tcpip.NICID, protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) bool { + s.mu.RLock() + defer s.mu.RUnlock() + + if nicID != 0 { + nic, ok := s.nics[nicID] + if !ok { + return false + } + + return isSubnetBroadcastOnNIC(nic, protocol, addr) + } + + for _, nic := range s.nics { + if isSubnetBroadcastOnNIC(nic, protocol, addr) { + return true + } + } + + return false +} diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 61db3164b..457990945 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -27,7 +27,6 @@ import ( "time" "github.com/google/go-cmp/cmp" - "github.com/google/go-cmp/cmp/cmpopts" "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" @@ -407,7 +406,7 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro return send(r, payload) } -func send(r stack.Route, payload buffer.View) *tcpip.Error { +func send(r *stack.Route, payload buffer.View) *tcpip.Error { return r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: fakeTransNumber, TTL: 123, TOS: stack.DefaultTOS}, stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(r.MaxHeaderLength()), Data: payload.ToVectorisedView(), @@ -425,7 +424,7 @@ func testSendTo(t *testing.T, s *stack.Stack, addr tcpip.Address, ep *channel.En } } -func testSend(t *testing.T, r stack.Route, ep *channel.Endpoint, payload buffer.View) { +func testSend(t *testing.T, r *stack.Route, ep *channel.Endpoint, payload buffer.View) { t.Helper() ep.Drain() if err := send(r, payload); err != nil { @@ -436,7 +435,7 @@ func testSend(t *testing.T, r stack.Route, ep *channel.Endpoint, payload buffer. } } -func testFailingSend(t *testing.T, r stack.Route, ep *channel.Endpoint, payload buffer.View, wantErr *tcpip.Error) { +func testFailingSend(t *testing.T, r *stack.Route, ep *channel.Endpoint, payload buffer.View, wantErr *tcpip.Error) { t.Helper() if gotErr := send(r, payload); gotErr != wantErr { t.Errorf("send failed: got = %s, want = %s ", gotErr, wantErr) @@ -1563,15 +1562,15 @@ func TestSpoofingNoAddress(t *testing.T) { // testSendTo(t, s, remoteAddr, ep, nil) } -func verifyRoute(gotRoute, wantRoute stack.Route) error { +func verifyRoute(gotRoute, wantRoute *stack.Route) error { if gotRoute.LocalAddress != wantRoute.LocalAddress { return fmt.Errorf("bad local address: got %s, want = %s", gotRoute.LocalAddress, wantRoute.LocalAddress) } if gotRoute.RemoteAddress != wantRoute.RemoteAddress { return fmt.Errorf("bad remote address: got %s, want = %s", gotRoute.RemoteAddress, wantRoute.RemoteAddress) } - if gotRoute.RemoteLinkAddress != wantRoute.RemoteLinkAddress { - return fmt.Errorf("bad remote link address: got %s, want = %s", gotRoute.RemoteLinkAddress, wantRoute.RemoteLinkAddress) + if got, want := gotRoute.RemoteLinkAddress(), wantRoute.RemoteLinkAddress(); got != want { + return fmt.Errorf("bad remote link address: got %s, want = %s", got, want) } if gotRoute.NextHop != wantRoute.NextHop { return fmt.Errorf("bad next-hop address: got %s, want = %s", gotRoute.NextHop, wantRoute.NextHop) @@ -1603,7 +1602,7 @@ func TestOutgoingBroadcastWithEmptyRouteTable(t *testing.T) { if err != nil { t.Fatalf("FindRoute(1, %v, %v, %d) failed: %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) } - if err := verifyRoute(r, stack.Route{LocalAddress: header.IPv4Any, RemoteAddress: header.IPv4Broadcast}); err != nil { + if err := verifyRoute(r, &stack.Route{LocalAddress: header.IPv4Any, RemoteAddress: header.IPv4Broadcast}); err != nil { t.Errorf("FindRoute(1, %v, %v, %d) returned unexpected Route: %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) } @@ -1657,7 +1656,7 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) { if err != nil { t.Fatalf("FindRoute(1, %v, %v, %d) failed: %v", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) } - if err := verifyRoute(r, stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + if err := verifyRoute(r, &stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { t.Errorf("FindRoute(1, %v, %v, %d) returned unexpected Route: %v", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) } @@ -1667,7 +1666,7 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) { if err != nil { t.Fatalf("FindRoute(0, \"\", %s, %d) failed: %s", header.IPv4Broadcast, fakeNetNumber, err) } - if err := verifyRoute(r, stack.Route{LocalAddress: nic2Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + if err := verifyRoute(r, &stack.Route{LocalAddress: nic2Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { t.Errorf("FindRoute(0, \"\", %s, %d) returned unexpected Route: %s)", header.IPv4Broadcast, fakeNetNumber, err) } @@ -1683,7 +1682,7 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) { if err != nil { t.Fatalf("FindRoute(0, \"\", %s, %d) failed: %s", header.IPv4Broadcast, fakeNetNumber, err) } - if err := verifyRoute(r, stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + if err := verifyRoute(r, &stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { t.Errorf("FindRoute(0, \"\", %s, %d) returned unexpected Route: %s)", header.IPv4Broadcast, fakeNetNumber, err) } } @@ -2407,9 +2406,9 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { } opts := stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ - AutoGenIPv6LinkLocal: test.autoGen, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: test.iidOpts, + AutoGenLinkLocal: test.autoGen, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: test.iidOpts, })}, } @@ -2502,8 +2501,8 @@ func TestNoLinkLocalAutoGenForLoopbackNIC(t *testing.T) { t.Run(test.name, func(t *testing.T) { opts := stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ - AutoGenIPv6LinkLocal: true, - OpaqueIIDOpts: test.opaqueIIDOpts, + AutoGenLinkLocal: true, + OpaqueIIDOpts: test.opaqueIIDOpts, })}, } @@ -2536,9 +2535,9 @@ func TestNICAutoGenAddrDoesDAD(t *testing.T) { ndpConfigs := ipv6.DefaultNDPConfigurations() opts := stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ - NDPConfigs: ndpConfigs, - AutoGenIPv6LinkLocal: true, - NDPDisp: &ndpDisp, + NDPConfigs: ndpConfigs, + AutoGenLinkLocal: true, + NDPDisp: &ndpDisp, })}, } @@ -3351,11 +3350,16 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { remNetSubnetBcast := remNetSubnet.Broadcast() tests := []struct { - name string - nicAddr tcpip.ProtocolAddress - routes []tcpip.Route - remoteAddr tcpip.Address - expectedRoute stack.Route + name string + nicAddr tcpip.ProtocolAddress + routes []tcpip.Route + remoteAddr tcpip.Address + expectedLocalAddress tcpip.Address + expectedRemoteAddress tcpip.Address + expectedRemoteLinkAddress tcpip.LinkAddress + expectedNextHop tcpip.Address + expectedNetProto tcpip.NetworkProtocolNumber + expectedLoop stack.PacketLooping }{ // Broadcast to a locally attached subnet populates the broadcast MAC. { @@ -3370,14 +3374,12 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { NIC: nicID1, }, }, - remoteAddr: ipv4SubnetBcast, - expectedRoute: stack.Route{ - LocalAddress: ipv4Addr.Address, - RemoteAddress: ipv4SubnetBcast, - RemoteLinkAddress: header.EthernetBroadcastAddress, - NetProto: header.IPv4ProtocolNumber, - Loop: stack.PacketOut | stack.PacketLoop, - }, + remoteAddr: ipv4SubnetBcast, + expectedLocalAddress: ipv4Addr.Address, + expectedRemoteAddress: ipv4SubnetBcast, + expectedRemoteLinkAddress: header.EthernetBroadcastAddress, + expectedNetProto: header.IPv4ProtocolNumber, + expectedLoop: stack.PacketOut | stack.PacketLoop, }, // Broadcast to a locally attached /31 subnet does not populate the // broadcast MAC. @@ -3393,13 +3395,11 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { NIC: nicID1, }, }, - remoteAddr: ipv4Subnet31Bcast, - expectedRoute: stack.Route{ - LocalAddress: ipv4AddrPrefix31.Address, - RemoteAddress: ipv4Subnet31Bcast, - NetProto: header.IPv4ProtocolNumber, - Loop: stack.PacketOut, - }, + remoteAddr: ipv4Subnet31Bcast, + expectedLocalAddress: ipv4AddrPrefix31.Address, + expectedRemoteAddress: ipv4Subnet31Bcast, + expectedNetProto: header.IPv4ProtocolNumber, + expectedLoop: stack.PacketOut, }, // Broadcast to a locally attached /32 subnet does not populate the // broadcast MAC. @@ -3415,13 +3415,11 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { NIC: nicID1, }, }, - remoteAddr: ipv4Subnet32Bcast, - expectedRoute: stack.Route{ - LocalAddress: ipv4AddrPrefix32.Address, - RemoteAddress: ipv4Subnet32Bcast, - NetProto: header.IPv4ProtocolNumber, - Loop: stack.PacketOut, - }, + remoteAddr: ipv4Subnet32Bcast, + expectedLocalAddress: ipv4AddrPrefix32.Address, + expectedRemoteAddress: ipv4Subnet32Bcast, + expectedNetProto: header.IPv4ProtocolNumber, + expectedLoop: stack.PacketOut, }, // IPv6 has no notion of a broadcast. { @@ -3436,13 +3434,11 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { NIC: nicID1, }, }, - remoteAddr: ipv6SubnetBcast, - expectedRoute: stack.Route{ - LocalAddress: ipv6Addr.Address, - RemoteAddress: ipv6SubnetBcast, - NetProto: header.IPv6ProtocolNumber, - Loop: stack.PacketOut, - }, + remoteAddr: ipv6SubnetBcast, + expectedLocalAddress: ipv6Addr.Address, + expectedRemoteAddress: ipv6SubnetBcast, + expectedNetProto: header.IPv6ProtocolNumber, + expectedLoop: stack.PacketOut, }, // Broadcast to a remote subnet in the route table is send to the next-hop // gateway. @@ -3459,14 +3455,12 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { NIC: nicID1, }, }, - remoteAddr: remNetSubnetBcast, - expectedRoute: stack.Route{ - LocalAddress: ipv4Addr.Address, - RemoteAddress: remNetSubnetBcast, - NextHop: ipv4Gateway, - NetProto: header.IPv4ProtocolNumber, - Loop: stack.PacketOut, - }, + remoteAddr: remNetSubnetBcast, + expectedLocalAddress: ipv4Addr.Address, + expectedRemoteAddress: remNetSubnetBcast, + expectedNextHop: ipv4Gateway, + expectedNetProto: header.IPv4ProtocolNumber, + expectedLoop: stack.PacketOut, }, // Broadcast to an unknown subnet follows the default route. Note that this // is essentially just routing an unknown destination IP, because w/o any @@ -3484,14 +3478,12 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { NIC: nicID1, }, }, - remoteAddr: remNetSubnetBcast, - expectedRoute: stack.Route{ - LocalAddress: ipv4Addr.Address, - RemoteAddress: remNetSubnetBcast, - NextHop: ipv4Gateway, - NetProto: header.IPv4ProtocolNumber, - Loop: stack.PacketOut, - }, + remoteAddr: remNetSubnetBcast, + expectedLocalAddress: ipv4Addr.Address, + expectedRemoteAddress: remNetSubnetBcast, + expectedNextHop: ipv4Gateway, + expectedNetProto: header.IPv4ProtocolNumber, + expectedLoop: stack.PacketOut, }, } @@ -3520,10 +3512,27 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { t.Fatalf("got unexpected address length = %d bytes", l) } - if r, err := s.FindRoute(unspecifiedNICID, "" /* localAddr */, test.remoteAddr, netProto, false /* multicastLoop */); err != nil { + r, err := s.FindRoute(unspecifiedNICID, "" /* localAddr */, test.remoteAddr, netProto, false /* multicastLoop */) + if err != nil { t.Fatalf("FindRoute(%d, '', %s, %d): %s", unspecifiedNICID, test.remoteAddr, netProto, err) - } else if diff := cmp.Diff(r, test.expectedRoute, cmpopts.IgnoreUnexported(r)); diff != "" { - t.Errorf("route mismatch (-want +got):\n%s", diff) + } + if r.LocalAddress != test.expectedLocalAddress { + t.Errorf("got r.LocalAddress = %s, want = %s", r.LocalAddress, test.expectedLocalAddress) + } + if r.RemoteAddress != test.expectedRemoteAddress { + t.Errorf("got r.RemoteAddress = %s, want = %s", r.RemoteAddress, test.expectedRemoteAddress) + } + if got := r.RemoteLinkAddress(); got != test.expectedRemoteLinkAddress { + t.Errorf("got r.RemoteLinkAddress() = %s, want = %s", got, test.expectedRemoteLinkAddress) + } + if r.NextHop != test.expectedNextHop { + t.Errorf("got r.NextHop = %s, want = %s", r.NextHop, test.expectedNextHop) + } + if r.NetProto != test.expectedNetProto { + t.Errorf("got r.NetProto = %d, want = %d", r.NetProto, test.expectedNetProto) + } + if r.Loop != test.expectedLoop { + t.Errorf("got r.Loop = %x, want = %x", r.Loop, test.expectedLoop) } }) } @@ -4091,10 +4100,12 @@ func TestFindRouteWithForwarding(t *testing.T) { s.SetRouteTable([]tcpip.Route{{Destination: test.netCfg.remoteAddr.WithPrefix().Subnet(), NIC: nicID2}}) r, err := s.FindRoute(test.addrNIC, test.localAddr, test.netCfg.remoteAddr, test.netCfg.proto, false /* multicastLoop */) + if r != nil { + defer r.Release() + } if err != test.findRouteErr { t.Fatalf("FindRoute(%d, %s, %s, %d, false) = %s, want = %s", test.addrNIC, test.localAddr, test.netCfg.remoteAddr, test.netCfg.proto, err, test.findRouteErr) } - defer r.Release() if test.findRouteErr != nil { return @@ -4152,3 +4163,63 @@ func TestFindRouteWithForwarding(t *testing.T) { }) } } + +func TestWritePacketToRemote(t *testing.T) { + const nicID = 1 + const MTU = 1280 + e := channel.New(1, MTU, linkAddr1) + s := stack.New(stack.Options{}) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + if err := s.EnableNIC(nicID); err != nil { + t.Fatalf("CreateNIC(%d) = %s", nicID, err) + } + tests := []struct { + name string + protocol tcpip.NetworkProtocolNumber + payload []byte + }{ + { + name: "SuccessIPv4", + protocol: header.IPv4ProtocolNumber, + payload: []byte{1, 2, 3, 4}, + }, + { + name: "SuccessIPv6", + protocol: header.IPv6ProtocolNumber, + payload: []byte{5, 6, 7, 8}, + }, + } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + if err := s.WritePacketToRemote(nicID, linkAddr2, test.protocol, buffer.View(test.payload).ToVectorisedView()); err != nil { + t.Fatalf("s.WritePacketToRemote(_, _, _, _) = %s", err) + } + + pkt, ok := e.Read() + if got, want := ok, true; got != want { + t.Fatalf("e.Read() = %t, want %t", got, want) + } + if got, want := pkt.Proto, test.protocol; got != want { + t.Fatalf("pkt.Proto = %d, want %d", got, want) + } + if got, want := pkt.Route.RemoteLinkAddress(), linkAddr2; got != want { + t.Fatalf("pkt.Route.RemoteAddress = %s, want %s", got, want) + } + if diff := cmp.Diff(pkt.Pkt.Data.ToView(), buffer.View(test.payload)); diff != "" { + t.Errorf("pkt.Pkt.Data mismatch (-want +got):\n%s", diff) + } + }) + } + + t.Run("InvalidNICID", func(t *testing.T) { + if got, want := s.WritePacketToRemote(234, linkAddr2, header.IPv4ProtocolNumber, buffer.View([]byte{1}).ToVectorisedView()), tcpip.ErrUnknownDevice; got != want { + t.Fatalf("s.WritePacketToRemote(_, _, _, _) = %s, want = %s", got, want) + } + pkt, ok := e.Read() + if got, want := ok, false; got != want { + t.Fatalf("e.Read() = %t, %v; want %t", got, pkt, want) + } + }) +} diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index 41a8e5ad0..2cdb5ca79 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -307,9 +307,7 @@ func TestBindToDeviceDistribution(t *testing.T) { }(ep) defer ep.Close() - if err := ep.SetSockOptBool(tcpip.ReusePortOption, endpoint.reuse); err != nil { - t.Fatalf("SetSockOptBool(ReusePortOption, %t) on endpoint %d failed: %s", endpoint.reuse, i, err) - } + ep.SocketOptions().SetReusePort(endpoint.reuse) bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice) if err := ep.SetSockOpt(&bindToDeviceOption); err != nil { t.Fatalf("SetSockOpt(&%T(%d)) on endpoint %d failed: %s", bindToDeviceOption, bindToDeviceOption, i, err) diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 5b9043d85..d9769e47d 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -38,14 +38,15 @@ const ( // use it. type fakeTransportEndpoint struct { stack.TransportEndpointInfo + tcpip.DefaultSocketOptionsHandler proto *fakeTransportProtocol peerAddr tcpip.Address - route stack.Route + route *stack.Route uniqueID uint64 // acceptQueue is non-nil iff bound. - acceptQueue []fakeTransportEndpoint + acceptQueue []*fakeTransportEndpoint // ops is used to set and get socket options. ops tcpip.SocketOptions @@ -64,8 +65,11 @@ func (*fakeTransportEndpoint) SetOwner(owner tcpip.PacketOwner) {} func (f *fakeTransportEndpoint) SocketOptions() *tcpip.SocketOptions { return &f.ops } + func newFakeTransportEndpoint(proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint { - return &fakeTransportEndpoint{TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID} + ep := &fakeTransportEndpoint{TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID} + ep.ops.InitHandler(ep) + return ep } func (f *fakeTransportEndpoint) Abort() { @@ -114,21 +118,11 @@ func (*fakeTransportEndpoint) SetSockOpt(tcpip.SettableSocketOption) *tcpip.Erro return tcpip.ErrInvalidEndpointState } -// SetSockOptBool sets a socket option. Currently not supported. -func (*fakeTransportEndpoint) SetSockOptBool(tcpip.SockOptBool, bool) *tcpip.Error { - return tcpip.ErrInvalidEndpointState -} - // SetSockOptInt sets a socket option. Currently not supported. func (*fakeTransportEndpoint) SetSockOptInt(tcpip.SockOptInt, int) *tcpip.Error { return tcpip.ErrInvalidEndpointState } -// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. -func (*fakeTransportEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - return false, tcpip.ErrUnknownProtocolOption -} - // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (*fakeTransportEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { return -1, tcpip.ErrUnknownProtocolOption @@ -189,7 +183,7 @@ func (f *fakeTransportEndpoint) Accept(*tcpip.FullAddress) (tcpip.Endpoint, *wai if len(f.acceptQueue) == 0 { return nil, nil, nil } - a := &f.acceptQueue[0] + a := f.acceptQueue[0] f.acceptQueue = f.acceptQueue[1:] return a, nil, nil } @@ -206,7 +200,7 @@ func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error { ); err != nil { return err } - f.acceptQueue = []fakeTransportEndpoint{} + f.acceptQueue = []*fakeTransportEndpoint{} return nil } @@ -232,7 +226,7 @@ func (f *fakeTransportEndpoint) HandlePacket(id stack.TransportEndpointID, pkt * } route.ResolveWith(pkt.SourceLinkAddress()) - f.acceptQueue = append(f.acceptQueue, fakeTransportEndpoint{ + ep := &fakeTransportEndpoint{ TransportEndpointInfo: stack.TransportEndpointInfo{ ID: f.ID, NetProto: f.NetProto, @@ -240,7 +234,9 @@ func (f *fakeTransportEndpoint) HandlePacket(id stack.TransportEndpointID, pkt * proto: f.proto, peerAddr: route.RemoteAddress, route: route, - }) + } + ep.ops.InitHandler(ep) + f.acceptQueue = append(f.acceptQueue, ep) } func (f *fakeTransportEndpoint) HandleControlPacket(stack.TransportEndpointID, stack.ControlType, uint32, *stack.PacketBuffer) { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index f9e83dd1c..5c9225b5d 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -247,6 +247,16 @@ func (a Address) WithPrefix() AddressWithPrefix { } } +// Unspecified returns true if the address is unspecified. +func (a Address) Unspecified() bool { + for _, b := range a { + if b != 0 { + return false + } + } + return true +} + // AddressMask is a bitmask for an address. type AddressMask string @@ -593,10 +603,6 @@ type Endpoint interface { // SetSockOpt sets a socket option. SetSockOpt(opt SettableSocketOption) *Error - // SetSockOptBool sets a socket option, for simple cases where a value - // has the bool type. - SetSockOptBool(opt SockOptBool, v bool) *Error - // SetSockOptInt sets a socket option, for simple cases where a value // has the int type. SetSockOptInt(opt SockOptInt, v int) *Error @@ -604,10 +610,6 @@ type Endpoint interface { // GetSockOpt gets a socket option. GetSockOpt(opt GettableSocketOption) *Error - // GetSockOptBool gets a socket option for simple cases where a return - // value has the bool type. - GetSockOptBool(SockOptBool) (bool, *Error) - // GetSockOptInt gets a socket option for simple cases where a return // value has the int type. GetSockOptInt(SockOptInt) (int, *Error) @@ -694,79 +696,6 @@ type WriteOptions struct { Atomic bool } -// SockOptBool represents socket options which values have the bool type. -type SockOptBool int - -const ( - // CorkOption is used by SetSockOptBool/GetSockOptBool to specify if - // data should be held until segments are full by the TCP transport - // protocol. - CorkOption SockOptBool = iota - - // DelayOption is used by SetSockOptBool/GetSockOptBool to specify if - // data should be sent out immediately by the transport protocol. For - // TCP, it determines if the Nagle algorithm is on or off. - DelayOption - - // KeepaliveEnabledOption is used by SetSockOptBool/GetSockOptBool to - // specify whether TCP keepalive is enabled for this socket. - KeepaliveEnabledOption - - // MulticastLoopOption is used by SetSockOptBool/GetSockOptBool to - // specify whether multicast packets sent over a non-loopback interface - // will be looped back. - MulticastLoopOption - - // NoChecksumOption is used by SetSockOptBool/GetSockOptBool to specify - // whether UDP checksum is disabled for this socket. - NoChecksumOption - - // PasscredOption is used by SetSockOptBool/GetSockOptBool to specify - // whether SCM_CREDENTIALS socket control messages are enabled. - // - // Only supported on Unix sockets. - PasscredOption - - // QuickAckOption is stubbed out in SetSockOptBool/GetSockOptBool. - QuickAckOption - - // ReceiveTClassOption is used by SetSockOptBool/GetSockOptBool to - // specify if the IPV6_TCLASS ancillary message is passed with incoming - // packets. - ReceiveTClassOption - - // ReceiveTOSOption is used by SetSockOptBool/GetSockOptBool to specify - // if the TOS ancillary message is passed with incoming packets. - ReceiveTOSOption - - // ReceiveIPPacketInfoOption is used by SetSockOptBool/GetSockOptBool to - // specify if more inforamtion is provided with incoming packets such as - // interface index and address. - ReceiveIPPacketInfoOption - - // ReuseAddressOption is used by SetSockOptBool/GetSockOptBool to - // specify whether Bind() should allow reuse of local address. - ReuseAddressOption - - // ReusePortOption is used by SetSockOptBool/GetSockOptBool to permit - // multiple sockets to be bound to an identical socket address. - ReusePortOption - - // V6OnlyOption is used by SetSockOptBool/GetSockOptBool to specify - // whether an IPv6 socket is to be restricted to sending and receiving - // IPv6 packets only. - V6OnlyOption - - // IPHdrIncludedOption is used by SetSockOpt to indicate for a raw - // endpoint that all packets being written have an IP header and the - // endpoint should not attach an IP header. - IPHdrIncludedOption - - // AcceptConnOption is used by GetSockOptBool to indicate if the - // socket is a listening socket. - AcceptConnOption -) - // SockOptInt represents socket options which values have the int type. type SockOptInt int @@ -1389,6 +1318,18 @@ type ICMPv6PacketStats struct { // RedirectMsg is the total number of ICMPv6 redirect message packets // counted. RedirectMsg *StatCounter + + // MulticastListenerQuery is the total number of Multicast Listener Query + // messages counted. + MulticastListenerQuery *StatCounter + + // MulticastListenerReport is the total number of Multicast Listener Report + // messages counted. + MulticastListenerReport *StatCounter + + // MulticastListenerDone is the total number of Multicast Listener Done + // messages counted. + MulticastListenerDone *StatCounter } // ICMPv4SentPacketStats collects outbound ICMPv4-specific stats. @@ -1430,6 +1371,10 @@ type ICMPv6SentPacketStats struct { type ICMPv6ReceivedPacketStats struct { ICMPv6PacketStats + // Unrecognized is the total number of ICMPv6 packets received that the + // transport layer does not know how to parse. + Unrecognized *StatCounter + // Invalid is the total number of ICMPv6 packets received that the // transport layer could not parse. Invalid *StatCounter @@ -1439,25 +1384,91 @@ type ICMPv6ReceivedPacketStats struct { RouterOnlyPacketsDroppedByHost *StatCounter } -// ICMPStats collects ICMP-specific stats (both v4 and v6). -type ICMPStats struct { +// ICMPv4Stats collects ICMPv4-specific stats. +type ICMPv4Stats struct { // ICMPv4SentPacketStats contains counts of sent packets by ICMPv4 packet type // and a single count of packets which failed to write to the link // layer. - V4PacketsSent ICMPv4SentPacketStats + PacketsSent ICMPv4SentPacketStats // ICMPv4ReceivedPacketStats contains counts of received packets by ICMPv4 // packet type and a single count of invalid packets received. - V4PacketsReceived ICMPv4ReceivedPacketStats + PacketsReceived ICMPv4ReceivedPacketStats +} +// ICMPv6Stats collects ICMPv6-specific stats. +type ICMPv6Stats struct { // ICMPv6SentPacketStats contains counts of sent packets by ICMPv6 packet type // and a single count of packets which failed to write to the link // layer. - V6PacketsSent ICMPv6SentPacketStats + PacketsSent ICMPv6SentPacketStats // ICMPv6ReceivedPacketStats contains counts of received packets by ICMPv6 // packet type and a single count of invalid packets received. - V6PacketsReceived ICMPv6ReceivedPacketStats + PacketsReceived ICMPv6ReceivedPacketStats +} + +// ICMPStats collects ICMP-specific stats (both v4 and v6). +type ICMPStats struct { + // V4 contains the ICMPv4-specifics stats. + V4 ICMPv4Stats + + // V6 contains the ICMPv4-specifics stats. + V6 ICMPv6Stats +} + +// IGMPPacketStats enumerates counts for all IGMP packet types. +type IGMPPacketStats struct { + // MembershipQuery is the total number of Membership Query messages counted. + MembershipQuery *StatCounter + + // V1MembershipReport is the total number of Version 1 Membership Report + // messages counted. + V1MembershipReport *StatCounter + + // V2MembershipReport is the total number of Version 2 Membership Report + // messages counted. + V2MembershipReport *StatCounter + + // LeaveGroup is the total number of Leave Group messages counted. + LeaveGroup *StatCounter +} + +// IGMPSentPacketStats collects outbound IGMP-specific stats. +type IGMPSentPacketStats struct { + IGMPPacketStats + + // Dropped is the total number of IGMP packets dropped due to link layer + // errors. + Dropped *StatCounter +} + +// IGMPReceivedPacketStats collects inbound IGMP-specific stats. +type IGMPReceivedPacketStats struct { + IGMPPacketStats + + // Invalid is the total number of IGMP packets received that IGMP could not + // parse. + Invalid *StatCounter + + // ChecksumErrors is the total number of IGMP packets dropped due to bad + // checksums. + ChecksumErrors *StatCounter + + // Unrecognized is the total number of unrecognized messages counted, these + // are silently ignored for forward-compatibilty. + Unrecognized *StatCounter +} + +// IGMPStats colelcts IGMP-specific stats. +type IGMPStats struct { + // IGMPSentPacketStats contains counts of sent packets by IGMP packet type + // and a single count of invalid packets received. + PacketsSent IGMPSentPacketStats + + // IGMPReceivedPacketStats contains counts of received packets by IGMP packet + // type and a single count of invalid packets received. + PacketsReceived IGMPReceivedPacketStats } // IPStats collects IP-specific stats (both v4 and v6). @@ -1665,6 +1676,9 @@ type Stats struct { // ICMP breaks out ICMP-specific stats (both v4 and v6). ICMP ICMPStats + // IGMP breaks out IGMP-specific stats. + IGMP IGMPStats + // IP breaks out IP-specific stats (both v4 and v6). IP IPStats diff --git a/pkg/tcpip/tcpip_test.go b/pkg/tcpip/tcpip_test.go index 1c8e2bc34..c461da137 100644 --- a/pkg/tcpip/tcpip_test.go +++ b/pkg/tcpip/tcpip_test.go @@ -226,3 +226,47 @@ func TestAddressWithPrefixSubnet(t *testing.T) { } } } + +func TestAddressUnspecified(t *testing.T) { + tests := []struct { + addr Address + unspecified bool + }{ + { + addr: "", + unspecified: true, + }, + { + addr: "\x00", + unspecified: true, + }, + { + addr: "\x01", + unspecified: false, + }, + { + addr: "\x00\x00", + unspecified: true, + }, + { + addr: "\x01\x00", + unspecified: false, + }, + { + addr: "\x00\x01", + unspecified: false, + }, + { + addr: "\x01\x01", + unspecified: false, + }, + } + + for _, test := range tests { + t.Run(fmt.Sprintf("addr=%s", test.addr), func(t *testing.T) { + if got := test.addr.Unspecified(); got != test.unspecified { + t.Fatalf("got addr.Unspecified() = %t, want = %t", got, test.unspecified) + } + }) + } +} diff --git a/pkg/tcpip/tests/integration/BUILD b/pkg/tcpip/tests/integration/BUILD index 9b0f3b675..800025fb9 100644 --- a/pkg/tcpip/tests/integration/BUILD +++ b/pkg/tcpip/tests/integration/BUILD @@ -25,6 +25,7 @@ go_test( "//pkg/tcpip/network/ipv6", "//pkg/tcpip/stack", "//pkg/tcpip/transport/icmp", + "//pkg/tcpip/transport/tcp", "//pkg/tcpip/transport/udp", "//pkg/waiter", "@com_github_google_go_cmp//cmp:go_default_library", diff --git a/pkg/tcpip/tests/integration/loopback_test.go b/pkg/tcpip/tests/integration/loopback_test.go index 421da1add..baaa741cd 100644 --- a/pkg/tcpip/tests/integration/loopback_test.go +++ b/pkg/tcpip/tests/integration/loopback_test.go @@ -26,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/pkg/waiter" ) @@ -70,8 +71,8 @@ func TestInitialLoopbackAddresses(t *testing.T) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocolWithOptions(ipv6.Options{ - NDPDisp: &ndpDispatcher{}, - AutoGenIPv6LinkLocal: true, + NDPDisp: &ndpDispatcher{}, + AutoGenLinkLocal: true, OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: func(nicID tcpip.NICID, nicName string) string { t.Fatalf("should not attempt to get name for NIC with ID = %d; nicName = %s", nicID, nicName) @@ -93,9 +94,10 @@ func TestInitialLoopbackAddresses(t *testing.T) { } } -// TestLoopbackAcceptAllInSubnet tests that a loopback interface considers -// itself bound to all addresses in the subnet of an assigned address. -func TestLoopbackAcceptAllInSubnet(t *testing.T) { +// TestLoopbackAcceptAllInSubnetUDP tests that a loopback interface considers +// itself bound to all addresses in the subnet of an assigned address and UDP +// traffic is sent/received correctly. +func TestLoopbackAcceptAllInSubnetUDP(t *testing.T) { const ( nicID = 1 localPort = 80 @@ -107,7 +109,7 @@ func TestLoopbackAcceptAllInSubnet(t *testing.T) { Protocol: header.IPv4ProtocolNumber, AddressWithPrefix: ipv4Addr, } - ipv4Bytes := []byte(ipv4Addr.Address) + ipv4Bytes := []byte(ipv4ProtocolAddress.AddressWithPrefix.Address) ipv4Bytes[len(ipv4Bytes)-1]++ otherIPv4Address := tcpip.Address(ipv4Bytes) @@ -129,7 +131,7 @@ func TestLoopbackAcceptAllInSubnet(t *testing.T) { { name: "IPv4 bind to wildcard and send to assigned address", addAddress: ipv4ProtocolAddress, - dstAddr: ipv4Addr.Address, + dstAddr: ipv4ProtocolAddress.AddressWithPrefix.Address, expectRx: true, }, { @@ -148,7 +150,7 @@ func TestLoopbackAcceptAllInSubnet(t *testing.T) { name: "IPv4 bind to other subnet-local address and send to assigned address", addAddress: ipv4ProtocolAddress, bindAddr: otherIPv4Address, - dstAddr: ipv4Addr.Address, + dstAddr: ipv4ProtocolAddress.AddressWithPrefix.Address, expectRx: false, }, { @@ -161,7 +163,7 @@ func TestLoopbackAcceptAllInSubnet(t *testing.T) { { name: "IPv4 bind to assigned address and send to other subnet-local address", addAddress: ipv4ProtocolAddress, - bindAddr: ipv4Addr.Address, + bindAddr: ipv4ProtocolAddress.AddressWithPrefix.Address, dstAddr: otherIPv4Address, expectRx: false, }, @@ -236,13 +238,17 @@ func TestLoopbackAcceptAllInSubnet(t *testing.T) { t.Fatalf("got sep.Write(_, _) = (%d, _, nil), want = (%d, _, nil)", n, want) } - if gotPayload, _, err := rep.Read(nil); test.expectRx { + var addr tcpip.FullAddress + if gotPayload, _, err := rep.Read(&addr); test.expectRx { if err != nil { - t.Fatalf("reep.Read(nil): %s", err) + t.Fatalf("reep.Read(_): %s", err) } if diff := cmp.Diff(buffer.View(data), gotPayload); diff != "" { t.Errorf("got UDP payload mismatch (-want +got):\n%s", diff) } + if addr.Addr != test.addAddress.AddressWithPrefix.Address { + t.Errorf("got addr.Addr = %s, want = %s", addr.Addr, test.addAddress.AddressWithPrefix.Address) + } } else { if err != tcpip.ErrWouldBlock { t.Fatalf("got rep.Read(nil) = (%x, _, %s), want = (_, _, %s)", gotPayload, err, tcpip.ErrWouldBlock) @@ -312,3 +318,168 @@ func TestLoopbackSubnetLifetimeBoundToAddr(t *testing.T) { t.Fatalf("got r.WritePacket(nil, %#v, _) = %s, want = %s", params, err, tcpip.ErrInvalidEndpointState) } } + +// TestLoopbackAcceptAllInSubnetTCP tests that a loopback interface considers +// itself bound to all addresses in the subnet of an assigned address and TCP +// traffic is sent/received correctly. +func TestLoopbackAcceptAllInSubnetTCP(t *testing.T) { + const ( + nicID = 1 + localPort = 80 + ) + + ipv4ProtocolAddress := tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: ipv4Addr, + } + ipv4ProtocolAddress.AddressWithPrefix.PrefixLen = 8 + ipv4Bytes := []byte(ipv4ProtocolAddress.AddressWithPrefix.Address) + ipv4Bytes[len(ipv4Bytes)-1]++ + otherIPv4Address := tcpip.Address(ipv4Bytes) + + ipv6ProtocolAddress := tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: ipv6Addr, + } + ipv6Bytes := []byte(ipv6Addr.Address) + ipv6Bytes[len(ipv6Bytes)-1]++ + otherIPv6Address := tcpip.Address(ipv6Bytes) + + tests := []struct { + name string + addAddress tcpip.ProtocolAddress + bindAddr tcpip.Address + dstAddr tcpip.Address + expectAccept bool + }{ + { + name: "IPv4 bind to wildcard and send to assigned address", + addAddress: ipv4ProtocolAddress, + dstAddr: ipv4ProtocolAddress.AddressWithPrefix.Address, + expectAccept: true, + }, + { + name: "IPv4 bind to wildcard and send to other subnet-local address", + addAddress: ipv4ProtocolAddress, + dstAddr: otherIPv4Address, + expectAccept: true, + }, + { + name: "IPv4 bind to wildcard send to other address", + addAddress: ipv4ProtocolAddress, + dstAddr: remoteIPv4Addr, + expectAccept: false, + }, + { + name: "IPv4 bind to other subnet-local address and send to assigned address", + addAddress: ipv4ProtocolAddress, + bindAddr: otherIPv4Address, + dstAddr: ipv4ProtocolAddress.AddressWithPrefix.Address, + expectAccept: false, + }, + { + name: "IPv4 bind and send to other subnet-local address", + addAddress: ipv4ProtocolAddress, + bindAddr: otherIPv4Address, + dstAddr: otherIPv4Address, + expectAccept: true, + }, + { + name: "IPv4 bind to assigned address and send to other subnet-local address", + addAddress: ipv4ProtocolAddress, + bindAddr: ipv4ProtocolAddress.AddressWithPrefix.Address, + dstAddr: otherIPv4Address, + expectAccept: false, + }, + + { + name: "IPv6 bind and send to assigned address", + addAddress: ipv6ProtocolAddress, + bindAddr: ipv6Addr.Address, + dstAddr: ipv6Addr.Address, + expectAccept: true, + }, + { + name: "IPv6 bind to wildcard and send to other subnet-local address", + addAddress: ipv6ProtocolAddress, + dstAddr: otherIPv6Address, + expectAccept: false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, + }) + if err := s.CreateNIC(nicID, loopback.New()); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID, err) + } + if err := s.AddProtocolAddress(nicID, test.addAddress); err != nil { + t.Fatalf("AddProtocolAddress(%d, %#v): %s", nicID, test.addAddress, err) + } + s.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: header.IPv4EmptySubnet, + NIC: nicID, + }, + tcpip.Route{ + Destination: header.IPv6EmptySubnet, + NIC: nicID, + }, + }) + + var wq waiter.Queue + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + listeningEndpoint, err := s.NewEndpoint(tcp.ProtocolNumber, test.addAddress.Protocol, &wq) + if err != nil { + t.Fatalf("NewEndpoint(%d, %d, _): %s", udp.ProtocolNumber, test.addAddress.Protocol, err) + } + defer listeningEndpoint.Close() + + bindAddr := tcpip.FullAddress{Addr: test.bindAddr, Port: localPort} + if err := listeningEndpoint.Bind(bindAddr); err != nil { + t.Fatalf("listeningEndpoint.Bind(%#v): %s", bindAddr, err) + } + + if err := listeningEndpoint.Listen(1); err != nil { + t.Fatalf("listeningEndpoint.Listen(1): %s", err) + } + + connectingEndpoint, err := s.NewEndpoint(tcp.ProtocolNumber, test.addAddress.Protocol, &wq) + if err != nil { + t.Fatalf("s.NewEndpoint(%d, %d, _): %s", udp.ProtocolNumber, test.addAddress.Protocol, err) + } + defer connectingEndpoint.Close() + + connectAddr := tcpip.FullAddress{ + Addr: test.dstAddr, + Port: localPort, + } + if err := connectingEndpoint.Connect(connectAddr); err != tcpip.ErrConnectStarted { + t.Fatalf("connectingEndpoint.Connect(%#v): %s", connectAddr, err) + } + + if !test.expectAccept { + if _, _, err := listeningEndpoint.Accept(nil); err != tcpip.ErrWouldBlock { + t.Fatalf("got listeningEndpoint.Accept(nil) = %s, want = %s", err, tcpip.ErrWouldBlock) + } + return + } + + // Wait for the listening endpoint to be "readable". That is, wait for a + // new connection. + <-ch + var addr tcpip.FullAddress + if _, _, err := listeningEndpoint.Accept(&addr); err != nil { + t.Fatalf("listeningEndpoint.Accept(nil): %s", err) + } + if addr.Addr != test.addAddress.AddressWithPrefix.Address { + t.Errorf("got addr.Addr = %s, want = %s", addr.Addr, test.addAddress.AddressWithPrefix.Address) + } + }) + } +} diff --git a/pkg/tcpip/tests/integration/multicast_broadcast_test.go b/pkg/tcpip/tests/integration/multicast_broadcast_test.go index 9d30329f5..8be791a00 100644 --- a/pkg/tcpip/tests/integration/multicast_broadcast_test.go +++ b/pkg/tcpip/tests/integration/multicast_broadcast_test.go @@ -510,10 +510,7 @@ func TestReuseAddrAndBroadcast(t *testing.T) { } defer ep.Close() - if err := ep.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { - t.Fatalf("eps[%d].SetSockOptBool(tcpip.ReuseAddressOption, true): %s", len(eps), err) - } - + ep.SocketOptions().SetReuseAddress(true) ep.SocketOptions().SetBroadcast(true) bindAddr := tcpip.FullAddress{Port: localPort} diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 440cb0352..94fcd72d9 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -49,6 +49,7 @@ const ( // +stateify savable type endpoint struct { stack.TransportEndpointInfo + tcpip.DefaultSocketOptionsHandler // The following fields are initialized at creation time and are // immutable. @@ -71,7 +72,7 @@ type endpoint struct { // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags state endpointState - route stack.Route `state:"manual"` + route *stack.Route `state:"manual"` ttl uint8 stats tcpip.TransportEndpointStats `state:"nosave"` // linger is used for SO_LINGER socket option. @@ -85,7 +86,7 @@ type endpoint struct { } func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return &endpoint{ + ep := &endpoint{ stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{ NetProto: netProto, @@ -96,7 +97,9 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt sndBufSize: 32 * 1024, state: stateInitial, uniqueID: s.UniqueID(), - }, nil + } + ep.ops.InitHandler(ep) + return ep, nil } // UniqueID implements stack.TransportEndpoint.UniqueID. @@ -129,7 +132,10 @@ func (e *endpoint) Close() { } e.rcvMu.Unlock() - e.route.Release() + if e.route != nil { + e.route.Release() + e.route = nil + } // Update the state. e.state = stateClosed @@ -142,6 +148,7 @@ func (e *endpoint) Close() { // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. func (e *endpoint) ModerateRecvBuf(copied int) {} +// SetOwner implements tcpip.Endpoint.SetOwner. func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } @@ -267,26 +274,8 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } } - var route *stack.Route - if to == nil { - route = &e.route - - if route.IsResolutionRequired() { - // Promote lock to exclusive if using a shared route, - // given that it may need to change in Route.Resolve() - // call below. - e.mu.RUnlock() - defer e.mu.RLock() - - e.mu.Lock() - defer e.mu.Unlock() - - // Recheck state after lock was re-acquired. - if e.state != stateConnected { - return 0, nil, tcpip.ErrInvalidEndpointState - } - } - } else { + route := e.route + if to != nil { // Reject destination address if it goes through a different // NIC than the endpoint was bound to. nicID := to.NIC @@ -310,7 +299,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } defer r.Release() - route = &r + route = r } if route.IsResolutionRequired() { @@ -361,11 +350,6 @@ func (e *endpoint) SetSockOpt(opt tcpip.SettableSocketOption) *tcpip.Error { return nil } -// SetSockOptBool sets a socket option. Currently not supported. -func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - return nil -} - // SetSockOptInt sets a socket option. Currently not supported. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { switch opt { @@ -378,17 +362,6 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { return nil } -// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. -func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - switch opt { - case tcpip.KeepaliveEnabledOption, tcpip.AcceptConnOption: - return false, nil - - default: - return false, tcpip.ErrUnknownProtocolOption - } -} - // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { @@ -857,6 +830,7 @@ func (*endpoint) LastError() *tcpip.Error { return nil } +// SocketOptions implements tcpip.Endpoint.SocketOptions. func (e *endpoint) SocketOptions() *tcpip.SocketOptions { return &e.ops } diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go index 3bff3755a..3666bac0f 100644 --- a/pkg/tcpip/transport/packet/endpoint.go +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -60,6 +60,8 @@ type packet struct { // +stateify savable type endpoint struct { stack.TransportEndpointInfo + tcpip.DefaultSocketOptionsHandler + // The following fields are initialized at creation time and are // immutable. stack *stack.Stack `state:"manual"` @@ -107,6 +109,7 @@ func NewEndpoint(s *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumb rcvBufSizeMax: 32 * 1024, sndBufSize: 32 * 1024, } + ep.ops.InitHandler(ep) // Override with stack defaults. var ss stack.SendBufferSizeOption @@ -318,11 +321,6 @@ func (ep *endpoint) SetSockOpt(opt tcpip.SettableSocketOption) *tcpip.Error { } } -// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. -func (ep *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption -} - // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (ep *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { switch opt { @@ -390,16 +388,6 @@ func (ep *endpoint) GetSockOpt(opt tcpip.GettableSocketOption) *tcpip.Error { } } -// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. -func (*endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - switch opt { - case tcpip.AcceptConnOption: - return false, nil - default: - return false, tcpip.ErrNotSupported - } -} - // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (ep *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { @@ -551,8 +539,10 @@ func (ep *endpoint) Stats() tcpip.EndpointStats { return &ep.stats } +// SetOwner implements tcpip.Endpoint.SetOwner. func (ep *endpoint) SetOwner(owner tcpip.PacketOwner) {} +// SocketOptions implements tcpip.Endpoint.SocketOptions. func (ep *endpoint) SocketOptions() *tcpip.SocketOptions { return &ep.ops } diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 4ae1f92ab..0840a4b3d 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -58,12 +58,13 @@ type rawPacket struct { // +stateify savable type endpoint struct { stack.TransportEndpointInfo + tcpip.DefaultSocketOptionsHandler + // The following fields are initialized at creation time and are // immutable. stack *stack.Stack `state:"manual"` waiterQueue *waiter.Queue associated bool - hdrIncluded bool // The following fields are used to manage the receive queue and are // protected by rcvMu. @@ -82,7 +83,7 @@ type endpoint struct { bound bool // route is the route to a remote network endpoint. It is set via // Connect(), and is valid only when conneted is true. - route stack.Route `state:"manual"` + route *stack.Route `state:"manual"` stats tcpip.TransportEndpointStats `state:"nosave"` // linger is used for SO_LINGER socket option. linger tcpip.LingerOption @@ -114,8 +115,9 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt rcvBufSizeMax: 32 * 1024, sndBufSizeMax: 32 * 1024, associated: associated, - hdrIncluded: !associated, } + e.ops.InitHandler(e) + e.ops.SetHeaderIncluded(!associated) // Override with stack defaults. var ss stack.SendBufferSizeOption @@ -170,9 +172,11 @@ func (e *endpoint) Close() { e.rcvList.Remove(e.rcvList.Front()) } - if e.connected { + e.connected = false + + if e.route != nil { e.route.Release() - e.connected = false + e.route = nil } e.closed = true @@ -266,7 +270,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c // If this is an unassociated socket and callee provided a nonzero // destination address, route using that address. - if e.hdrIncluded { + if e.ops.GetHeaderIncluded() { ip := header.IPv4(payloadBytes) if !ip.IsValid(len(payloadBytes)) { e.mu.RUnlock() @@ -296,7 +300,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } if e.route.IsResolutionRequired() { - savedRoute := &e.route + savedRoute := e.route // Promote lock to exclusive if using a shared route, // given that it may need to change in finishWrite. e.mu.RUnlock() @@ -304,7 +308,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c // Make sure that the route didn't change during the // time we didn't hold the lock. - if !e.connected || savedRoute != &e.route { + if !e.connected || savedRoute != e.route { e.mu.Unlock() return 0, nil, tcpip.ErrInvalidEndpointState } @@ -314,7 +318,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return n, ch, err } - n, ch, err := e.finishWrite(payloadBytes, &e.route) + n, ch, err := e.finishWrite(payloadBytes, e.route) e.mu.RUnlock() return n, ch, err } @@ -335,7 +339,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, err } - n, ch, err := e.finishWrite(payloadBytes, &route) + n, ch, err := e.finishWrite(payloadBytes, route) route.Release() e.mu.RUnlock() return n, ch, err @@ -356,7 +360,7 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, } } - if e.hdrIncluded { + if e.ops.GetHeaderIncluded() { pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buffer.View(payloadBytes).ToVectorisedView(), }) @@ -531,18 +535,6 @@ func (e *endpoint) SetSockOpt(opt tcpip.SettableSocketOption) *tcpip.Error { } } -// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. -func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - switch opt { - case tcpip.IPHdrIncludedOption: - e.mu.Lock() - e.hdrIncluded = v - e.mu.Unlock() - return nil - } - return tcpip.ErrUnknownProtocolOption -} - // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { switch opt { @@ -601,23 +593,6 @@ func (e *endpoint) GetSockOpt(opt tcpip.GettableSocketOption) *tcpip.Error { } } -// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. -func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - switch opt { - case tcpip.KeepaliveEnabledOption, tcpip.AcceptConnOption: - return false, nil - - case tcpip.IPHdrIncludedOption: - e.mu.Lock() - v := e.hdrIncluded - e.mu.Unlock() - return v, nil - - default: - return false, tcpip.ErrUnknownProtocolOption - } -} - // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { @@ -756,10 +731,12 @@ func (e *endpoint) Stats() tcpip.EndpointStats { // Wait implements stack.TransportEndpoint.Wait. func (*endpoint) Wait() {} +// LastError implements tcpip.Endpoint.LastError. func (*endpoint) LastError() *tcpip.Error { return nil } +// SocketOptions implements tcpip.Endpoint.SocketOptions. func (e *endpoint) SocketOptions() *tcpip.SocketOptions { return &e.ops } diff --git a/pkg/tcpip/transport/raw/endpoint_state.go b/pkg/tcpip/transport/raw/endpoint_state.go index 7d97cbdc7..4a7e1c039 100644 --- a/pkg/tcpip/transport/raw/endpoint_state.go +++ b/pkg/tcpip/transport/raw/endpoint_state.go @@ -73,7 +73,13 @@ func (e *endpoint) Resume(s *stack.Stack) { // If the endpoint is connected, re-connect. if e.connected { var err *tcpip.Error - e.route, err = e.stack.FindRoute(e.RegisterNICID, e.BindAddr, e.route.RemoteAddress, e.NetProto, false) + // TODO(gvisor.dev/issue/4906): Properly restore the route with the right + // remote address. We used to pass e.remote.RemoteAddress which was + // effectively the empty address but since moving e.route to hold a pointer + // to a route instead of the route by value, we pass the empty address + // directly. Obviously this was always wrong since we should provide the + // remote address we were connected to, to properly restore the route. + e.route, err = e.stack.FindRoute(e.RegisterNICID, e.BindAddr, "", e.NetProto, false) if err != nil { panic(err) } diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 518449602..cf232b508 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library", "go_test") +load("//tools:defs.bzl", "go_library", "go_test", "more_shards") load("//tools/go_generics:defs.bzl", "go_template_instance") package(licenses = ["notice"]) @@ -45,7 +45,9 @@ go_library( "rcv.go", "rcv_state.go", "reno.go", + "reno_recovery.go", "sack.go", + "sack_recovery.go", "sack_scoreboard.go", "segment.go", "segment_heap.go", @@ -91,7 +93,7 @@ go_test( "tcp_test.go", "tcp_timestamp_test.go", ], - shard_count = 10, + shard_count = more_shards, deps = [ ":tcp", "//pkg/rand", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 6e5adc383..3e1041cbe 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -213,7 +213,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i route.ResolveWith(s.remoteLinkAddr) n := newEndpoint(l.stack, netProto, queue) - n.v6only = l.v6Only + n.ops.SetV6Only(l.v6Only) n.ID = s.id n.boundNICID = s.nicID n.route = route @@ -599,7 +599,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) *tcpip.Er ack: s.sequenceNumber + 1, rcvWnd: ctx.rcvWnd, } - if err := e.sendSynTCP(&route, fields, synOpts); err != nil { + if err := e.sendSynTCP(route, fields, synOpts); err != nil { return err } e.stack.Stats().TCP.ListenOverflowSynCookieSent.Increment() @@ -752,7 +752,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) *tcpip.Er // its own goroutine and is responsible for handling connection requests. func (e *endpoint) protocolListenLoop(rcvWnd seqnum.Size) { e.mu.Lock() - v6Only := e.v6only + v6Only := e.ops.GetV6Only() ctx := newListenContext(e.stack, e, rcvWnd, v6Only, e.NetProto) defer func() { diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index ac6d879a7..c944dccc0 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -16,6 +16,7 @@ package tcp import ( "encoding/binary" + "math" "time" "gvisor.dev/gvisor/pkg/rand" @@ -133,7 +134,7 @@ func FindWndScale(wnd seqnum.Size) int { return 0 } - max := seqnum.Size(0xffff) + max := seqnum.Size(math.MaxUint16) s := 0 for wnd > max && s < header.MaxWndScale { s++ @@ -300,7 +301,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { if ttl == 0 { ttl = h.ep.route.DefaultTTL() } - h.ep.sendSynTCP(&h.ep.route, tcpFields{ + h.ep.sendSynTCP(h.ep.route, tcpFields{ id: h.ep.ID, ttl: ttl, tos: h.ep.sendTOS, @@ -361,7 +362,7 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { SACKPermitted: h.ep.sackPermitted, MSS: h.ep.amss, } - h.ep.sendSynTCP(&h.ep.route, tcpFields{ + h.ep.sendSynTCP(h.ep.route, tcpFields{ id: h.ep.ID, ttl: h.ep.ttl, tos: h.ep.sendTOS, @@ -496,7 +497,7 @@ func (h *handshake) resolveRoute() *tcpip.Error { h.ep.mu.Lock() } if n¬ifyError != 0 { - return h.ep.LastError() + return h.ep.lastErrorLocked() } } @@ -547,7 +548,7 @@ func (h *handshake) start() *tcpip.Error { } h.sendSYNOpts = synOpts - h.ep.sendSynTCP(&h.ep.route, tcpFields{ + h.ep.sendSynTCP(h.ep.route, tcpFields{ id: h.ep.ID, ttl: h.ep.ttl, tos: h.ep.sendTOS, @@ -575,7 +576,6 @@ func (h *handshake) complete() *tcpip.Error { return err } defer timer.stop() - for h.state != handshakeCompleted { // Unlock before blocking, and reacquire again afterwards (h.ep.mu is held // throughout handshake processing). @@ -597,7 +597,7 @@ func (h *handshake) complete() *tcpip.Error { // the connection with another ACK or data (as ACKs are never // retransmitted on their own). if h.active || !h.acked || h.deferAccept != 0 && time.Since(h.startTime) > h.deferAccept { - h.ep.sendSynTCP(&h.ep.route, tcpFields{ + h.ep.sendSynTCP(h.ep.route, tcpFields{ id: h.ep.ID, ttl: h.ep.ttl, tos: h.ep.sendTOS, @@ -631,9 +631,8 @@ func (h *handshake) complete() *tcpip.Error { h.ep.mu.Lock() } if n¬ifyError != 0 { - return h.ep.LastError() + return h.ep.lastErrorLocked() } - case wakerForNewSegment: if err := h.processSegments(); err != nil { return err @@ -820,8 +819,8 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso data = data.Clone(nil) optLen := len(tf.opts) - if tf.rcvWnd > 0xffff { - tf.rcvWnd = 0xffff + if tf.rcvWnd > math.MaxUint16 { + tf.rcvWnd = math.MaxUint16 } mss := int(gso.MSS) @@ -865,8 +864,8 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso // network endpoint and under the provided identity. func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO, owner tcpip.PacketOwner) *tcpip.Error { optLen := len(tf.opts) - if tf.rcvWnd > 0xffff { - tf.rcvWnd = 0xffff + if tf.rcvWnd > math.MaxUint16 { + tf.rcvWnd = math.MaxUint16 } if r.Loop&stack.PacketLoop == 0 && gso != nil && gso.Type == stack.GSOSW && int(gso.MSS) < data.Size() { @@ -941,7 +940,7 @@ func (e *endpoint) sendRaw(data buffer.VectorisedView, flags byte, seq, ack seqn sackBlocks = e.sack.Blocks[:e.sack.NumBlocks] } options := e.makeOptions(sackBlocks) - err := e.sendTCP(&e.route, tcpFields{ + err := e.sendTCP(e.route, tcpFields{ id: e.ID, ttl: e.ttl, tos: e.sendTOS, @@ -1002,7 +1001,7 @@ func (e *endpoint) resetConnectionLocked(err *tcpip.Error) { // Only send a reset if the connection is being aborted for a reason // other than receiving a reset. e.setEndpointState(StateError) - e.HardError = err + e.hardError = err if err != tcpip.ErrConnectionReset && err != tcpip.ErrTimeout { // The exact sequence number to be used for the RST is the same as the // one used by Linux. We need to handle the case of window being shrunk @@ -1080,7 +1079,7 @@ func (e *endpoint) transitionToStateCloseLocked() { // to any other listening endpoint. We reply with RST if we cannot find one. func (e *endpoint) tryDeliverSegmentFromClosedEndpoint(s *segment) { ep := e.stack.FindTransportEndpoint(e.NetProto, e.TransProto, e.ID, s.nicID) - if ep == nil && e.NetProto == header.IPv6ProtocolNumber && e.EndpointInfo.TransportEndpointInfo.ID.LocalAddress.To4() != "" { + if ep == nil && e.NetProto == header.IPv6ProtocolNumber && e.TransportEndpointInfo.ID.LocalAddress.To4() != "" { // Dual-stack socket, try IPv4. ep = e.stack.FindTransportEndpoint(header.IPv4ProtocolNumber, e.TransProto, e.ID, s.nicID) } @@ -1141,7 +1140,7 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { // delete the TCB, and return. case StateCloseWait: e.transitionToStateCloseLocked() - e.HardError = tcpip.ErrAborted + e.hardError = tcpip.ErrAborted e.notifyProtocolGoroutine(notifyTickleWorker) return false, nil default: @@ -1286,7 +1285,7 @@ func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { userTimeout := e.userTimeout e.keepalive.Lock() - if !e.keepalive.enabled || !e.keepalive.timer.checkExpiration() { + if !e.SocketOptions().GetKeepAlive() || !e.keepalive.timer.checkExpiration() { e.keepalive.Unlock() return nil } @@ -1323,7 +1322,7 @@ func (e *endpoint) resetKeepaliveTimer(receivedData bool) { } // Start the keepalive timer IFF it's enabled and there is no pending // data to send. - if !e.keepalive.enabled || e.snd == nil || e.snd.sndUna != e.snd.sndNxt { + if !e.SocketOptions().GetKeepAlive() || e.snd == nil || e.snd.sndUna != e.snd.sndNxt { e.keepalive.timer.disable() e.keepalive.Unlock() return @@ -1353,7 +1352,6 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ epilogue := func() { // e.mu is expected to be hold upon entering this section. - if e.snd != nil { e.snd.resendTimer.cleanup() } @@ -1383,7 +1381,7 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ e.lastErrorMu.Unlock() e.setEndpointState(StateError) - e.HardError = err + e.hardError = err e.workerCleanup = true // Lock released below. @@ -1638,7 +1636,7 @@ func (e *endpoint) handleTimeWaitSegments() (extendTimeWait bool, reuseTW func() } extTW, newSyn := e.rcv.handleTimeWaitSegment(s) if newSyn { - info := e.EndpointInfo.TransportEndpointInfo + info := e.TransportEndpointInfo newID := info.ID newID.RemoteAddress = "" newID.RemotePort = 0 diff --git a/pkg/tcpip/transport/tcp/dual_stack_test.go b/pkg/tcpip/transport/tcp/dual_stack_test.go index a6f25896b..1d1b01a6c 100644 --- a/pkg/tcpip/transport/tcp/dual_stack_test.go +++ b/pkg/tcpip/transport/tcp/dual_stack_test.go @@ -405,14 +405,6 @@ func testV4Accept(t *testing.T, c *context.Context) { } } - // Make sure we get the same error when calling the original ep and the - // new one. This validates that v4-mapped endpoints are still able to - // query the V6Only flag, whereas pure v4 endpoints are not. - _, expected := c.EP.GetSockOptBool(tcpip.V6OnlyOption) - if _, err := nep.GetSockOptBool(tcpip.V6OnlyOption); err != expected { - t.Fatalf("GetSockOpt returned unexpected value: got %v, want %v", err, expected) - } - // Check the peer address. addr, err := nep.GetRemoteAddress() if err != nil { @@ -530,12 +522,12 @@ func TestV6AcceptOnV6(t *testing.T) { c.WQ.EventRegister(&we, waiter.EventIn) defer c.WQ.EventUnregister(&we) var addr tcpip.FullAddress - nep, _, err := c.EP.Accept(&addr) + _, _, err := c.EP.Accept(&addr) if err == tcpip.ErrWouldBlock { // Wait for connection to be established. select { case <-ch: - nep, _, err = c.EP.Accept(&addr) + _, _, err = c.EP.Accept(&addr) if err != nil { t.Fatalf("Accept failed: %v", err) } @@ -548,12 +540,6 @@ func TestV6AcceptOnV6(t *testing.T) { if addr.Addr != context.TestV6Addr { t.Errorf("Unexpected remote address: got %s, want %s", addr.Addr, context.TestV6Addr) } - - // Make sure we can still query the v6 only status of the new endpoint, - // that is, that it is in fact a v6 socket. - if _, err := nep.GetSockOptBool(tcpip.V6OnlyOption); err != nil { - t.Errorf("GetSockOptBool(tcpip.V6OnlyOption) failed: %s", err) - } } func TestV4AcceptOnV4(t *testing.T) { diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 4f4f4c65e..87eda2efb 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -310,16 +310,12 @@ type Stats struct { func (*Stats) IsEndpointStats() {} // EndpointInfo holds useful information about a transport endpoint which -// can be queried by monitoring tools. +// can be queried by monitoring tools. This exists to allow tcp-only state to +// be exposed. // // +stateify savable type EndpointInfo struct { stack.TransportEndpointInfo - - // HardError is meaningful only when state is stateError. It stores the - // error to be returned when read/write syscalls are called and the - // endpoint is in this state. HardError is protected by endpoint mu. - HardError *tcpip.Error `state:".(string)"` } // IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo @@ -367,6 +363,7 @@ func (*EndpointInfo) IsEndpointInfo() {} // +stateify savable type endpoint struct { EndpointInfo + tcpip.DefaultSocketOptionsHandler // endpointEntry is used to queue endpoints for processing to the // a given tcp processor goroutine. @@ -386,6 +383,11 @@ type endpoint struct { waiterQueue *waiter.Queue `state:"wait"` uniqueID uint64 + // hardError is meaningful only when state is stateError. It stores the + // error to be returned when read/write syscalls are called and the + // endpoint is in this state. hardError is protected by endpoint mu. + hardError *tcpip.Error `state:".(string)"` + // lastError represents the last error that the endpoint reported; // access to it is protected by the following mutex. lastErrorMu sync.Mutex `state:"nosave"` @@ -421,7 +423,10 @@ type endpoint struct { // mu protects all endpoint fields unless documented otherwise. mu must // be acquired before interacting with the endpoint fields. - mu sync.Mutex `state:"nosave"` + // + // During handshake, mu is locked by the protocol listen goroutine and + // released by the handshake completion goroutine. + mu sync.CrossGoroutineMutex `state:"nosave"` ownedByUser uint32 // state must be read/set using the EndpointState()/setEndpointState() @@ -436,9 +441,8 @@ type endpoint struct { isPortReserved bool `state:"manual"` isRegistered bool `state:"manual"` boundNICID tcpip.NICID - route stack.Route `state:"manual"` + route *stack.Route `state:"manual"` ttl uint8 - v6only bool isConnectNotified bool // h stores a reference to the current handshake state if the endpoint is in @@ -506,24 +510,9 @@ type endpoint struct { // delay is a boolean (0 is false) and must be accessed atomically. delay uint32 - // cork holds back segments until full. - // - // cork is a boolean (0 is false) and must be accessed atomically. - cork uint32 - // scoreboard holds TCP SACK Scoreboard information for this endpoint. scoreboard *SACKScoreboard - // The options below aren't implemented, but we remember the user - // settings because applications expect to be able to set/query these - // options. - - // slowAck holds the negated state of quick ack. It is stubbed out and - // does nothing. - // - // slowAck is a boolean (0 is false) and must be accessed atomically. - slowAck uint32 - // segmentQueue is used to hand received segments to the protocol // goroutine. Segments are queued as long as the queue is not full, // and dropped when it is. @@ -701,7 +690,7 @@ func (e *endpoint) UniqueID() uint64 { // // If userMSS is non-zero and is not greater than the maximum possible MSS for // r, it will be used; otherwise, the maximum possible MSS will be used. -func calculateAdvertisedMSS(userMSS uint16, r stack.Route) uint16 { +func calculateAdvertisedMSS(userMSS uint16, r *stack.Route) uint16 { // The maximum possible MSS is dependent on the route. // TODO(b/143359391): Respect TCP Min and Max size. maxMSS := uint16(r.MTU() - header.TCPMinimumSize) @@ -850,7 +839,6 @@ func (e *endpoint) recentTimestamp() uint32 { // +stateify savable type keepalive struct { sync.Mutex `state:"nosave"` - enabled bool idle time.Duration interval time.Duration count int @@ -884,6 +872,9 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue windowClamp: DefaultReceiveBufferSize, maxSynRetries: DefaultSynRetries, } + e.ops.InitHandler(e) + e.ops.SetMulticastLoop(true) + e.ops.SetQuickAck(true) var ss tcpip.TCPSendBufferSizeRangeOption if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil { @@ -907,7 +898,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue var de tcpip.TCPDelayEnabled if err := s.TransportProtocolOption(ProtocolNumber, &de); err == nil && de { - e.SetSockOptBool(tcpip.DelayOption, true) + e.ops.SetDelayOption(true) } var tcpLT tcpip.TCPLingerTimeoutOption @@ -1169,7 +1160,11 @@ func (e *endpoint) cleanupLocked() { e.boundPortFlags = ports.Flags{} e.boundDest = tcpip.FullAddress{} - e.route.Release() + if e.route != nil { + e.route.Release() + e.route = nil + } + e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } @@ -1279,11 +1274,20 @@ func (e *endpoint) ModerateRecvBuf(copied int) { e.rcvListMu.Unlock() } +// SetOwner implements tcpip.Endpoint.SetOwner. func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } -func (e *endpoint) LastError() *tcpip.Error { +// Preconditions: e.mu must be held to call this function. +func (e *endpoint) hardErrorLocked() *tcpip.Error { + err := e.hardError + e.hardError = nil + return err +} + +// Preconditions: e.mu must be held to call this function. +func (e *endpoint) lastErrorLocked() *tcpip.Error { e.lastErrorMu.Lock() defer e.lastErrorMu.Unlock() err := e.lastError @@ -1291,6 +1295,16 @@ func (e *endpoint) LastError() *tcpip.Error { return err } +// LastError implements tcpip.Endpoint.LastError. +func (e *endpoint) LastError() *tcpip.Error { + e.LockUser() + defer e.UnlockUser() + if err := e.hardErrorLocked(); err != nil { + return err + } + return e.lastErrorLocked() +} + // Read reads data from the endpoint. func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { e.LockUser() @@ -1312,9 +1326,11 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, bufUsed := e.rcvBufUsed if s := e.EndpointState(); !s.connected() && s != StateClose && bufUsed == 0 { e.rcvListMu.Unlock() - he := e.HardError if s == StateError { - return buffer.View{}, tcpip.ControlMessages{}, he + if err := e.hardErrorLocked(); err != nil { + return buffer.View{}, tcpip.ControlMessages{}, err + } + return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrClosedForReceive } e.stats.ReadErrors.NotConnected.Increment() return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrNotConnected @@ -1370,9 +1386,13 @@ func (e *endpoint) readLocked() (buffer.View, *tcpip.Error) { // indicating the reason why it's not writable. // Caller must hold e.mu and e.sndBufMu func (e *endpoint) isEndpointWritableLocked() (int, *tcpip.Error) { + // The endpoint cannot be written to if it's not connected. switch s := e.EndpointState(); { case s == StateError: - return 0, e.HardError + if err := e.hardErrorLocked(); err != nil { + return 0, err + } + return 0, tcpip.ErrClosedForSend case !s.connecting() && !s.connected(): return 0, tcpip.ErrClosedForSend case s.connecting(): @@ -1486,7 +1506,7 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro // but has some pending unread data. if s := e.EndpointState(); !s.connected() && s != StateClose { if s == StateError { - return 0, tcpip.ControlMessages{}, e.HardError + return 0, tcpip.ControlMessages{}, e.hardErrorLocked() } e.stats.ReadErrors.InvalidEndpointState.Increment() return 0, tcpip.ControlMessages{}, tcpip.ErrInvalidEndpointState @@ -1602,72 +1622,39 @@ func (e *endpoint) windowCrossedACKThresholdLocked(deltaBefore int) (crossed boo return false, false } -// SetSockOptBool sets a socket option. -func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - switch opt { - - case tcpip.CorkOption: - e.LockUser() - if !v { - atomic.StoreUint32(&e.cork, 0) - - // Handle the corked data. - e.sndWaker.Assert() - } else { - atomic.StoreUint32(&e.cork, 1) - } - e.UnlockUser() - - case tcpip.DelayOption: - if v { - atomic.StoreUint32(&e.delay, 1) - } else { - atomic.StoreUint32(&e.delay, 0) - - // Handle delayed data. - e.sndWaker.Assert() - } - - case tcpip.KeepaliveEnabledOption: - e.keepalive.Lock() - e.keepalive.enabled = v - e.keepalive.Unlock() - e.notifyProtocolGoroutine(notifyKeepaliveChanged) - - case tcpip.QuickAckOption: - o := uint32(1) - if v { - o = 0 - } - atomic.StoreUint32(&e.slowAck, o) - - case tcpip.ReuseAddressOption: - e.LockUser() - e.portFlags.TupleOnly = v - e.UnlockUser() - - case tcpip.ReusePortOption: - e.LockUser() - e.portFlags.LoadBalanced = v - e.UnlockUser() +// OnReuseAddressSet implements tcpip.SocketOptionsHandler.OnReuseAddressSet. +func (e *endpoint) OnReuseAddressSet(v bool) { + e.LockUser() + e.portFlags.TupleOnly = v + e.UnlockUser() +} - case tcpip.V6OnlyOption: - // We only recognize this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return tcpip.ErrInvalidEndpointState - } +// OnReusePortSet implements tcpip.SocketOptionsHandler.OnReusePortSet. +func (e *endpoint) OnReusePortSet(v bool) { + e.LockUser() + e.portFlags.LoadBalanced = v + e.UnlockUser() +} - // We only allow this to be set when we're in the initial state. - if e.EndpointState() != StateInitial { - return tcpip.ErrInvalidEndpointState - } +// OnKeepAliveSet implements tcpip.SocketOptionsHandler.OnKeepAliveSet. +func (e *endpoint) OnKeepAliveSet(v bool) { + e.notifyProtocolGoroutine(notifyKeepaliveChanged) +} - e.LockUser() - e.v6only = v - e.UnlockUser() +// OnDelayOptionSet implements tcpip.SocketOptionsHandler.OnDelayOptionSet. +func (e *endpoint) OnDelayOptionSet(v bool) { + if !v { + // Handle delayed data. + e.sndWaker.Assert() } +} - return nil +// OnCorkOptionSet implements tcpip.SocketOptionsHandler.OnCorkOptionSet. +func (e *endpoint) OnCorkOptionSet(v bool) { + if !v { + // Handle the corked data. + e.sndWaker.Assert() + } } // SetSockOptInt sets a socket option. @@ -1949,67 +1936,6 @@ func (e *endpoint) readyReceiveSize() (int, *tcpip.Error) { return e.rcvBufUsed, nil } -// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. -func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - switch opt { - - case tcpip.CorkOption: - return atomic.LoadUint32(&e.cork) != 0, nil - - case tcpip.DelayOption: - return atomic.LoadUint32(&e.delay) != 0, nil - - case tcpip.KeepaliveEnabledOption: - e.keepalive.Lock() - v := e.keepalive.enabled - e.keepalive.Unlock() - - return v, nil - - case tcpip.QuickAckOption: - v := atomic.LoadUint32(&e.slowAck) == 0 - return v, nil - - case tcpip.ReuseAddressOption: - e.LockUser() - v := e.portFlags.TupleOnly - e.UnlockUser() - - return v, nil - - case tcpip.ReusePortOption: - e.LockUser() - v := e.portFlags.LoadBalanced - e.UnlockUser() - - return v, nil - - case tcpip.V6OnlyOption: - // We only recognize this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return false, tcpip.ErrUnknownProtocolOption - } - - e.LockUser() - v := e.v6only - e.UnlockUser() - - return v, nil - - case tcpip.MulticastLoopOption: - return true, nil - - case tcpip.AcceptConnOption: - e.LockUser() - defer e.UnlockUser() - - return e.EndpointState() == StateListen, nil - - default: - return false, tcpip.ErrUnknownProtocolOption - } -} - // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { @@ -2166,7 +2092,7 @@ func (e *endpoint) GetSockOpt(opt tcpip.GettableSocketOption) *tcpip.Error { // checkV4MappedLocked determines the effective network protocol and converts // addr to its canonical form. func (e *endpoint) checkV4MappedLocked(addr tcpip.FullAddress) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) { - unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProtoLocked(addr, e.v6only) + unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProtoLocked(addr, e.ops.GetV6Only()) if err != nil { return tcpip.FullAddress{}, 0, err } @@ -2243,7 +2169,10 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc return tcpip.ErrAlreadyConnecting case StateError: - return e.HardError + if err := e.hardErrorLocked(); err != nil { + return err + } + return tcpip.ErrConnectionAborted default: return tcpip.ErrInvalidEndpointState @@ -2417,7 +2346,7 @@ func (e *endpoint) startMainLoop(handshake bool) *tcpip.Error { e.lastErrorMu.Unlock() e.setEndpointState(StateError) - e.HardError = err + e.hardError = err // Call cleanupLocked to free up any reservations. e.cleanupLocked() @@ -2697,7 +2626,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) { // v6only set to false. if netProto == header.IPv6ProtocolNumber { stackHasV4 := e.stack.CheckNetworkProtocol(header.IPv4ProtocolNumber) - alsoBindToV4 := !e.v6only && addr.Addr == "" && stackHasV4 + alsoBindToV4 := !e.ops.GetV6Only() && addr.Addr == "" && stackHasV4 if alsoBindToV4 { netProtos = append(netProtos, header.IPv4ProtocolNumber) } @@ -2782,7 +2711,7 @@ func (e *endpoint) getRemoteAddress() tcpip.FullAddress { func (*endpoint) HandlePacket(stack.TransportEndpointID, *stack.PacketBuffer) { // TCP HandlePacket is not required anymore as inbound packets first - // land at the Dispatcher which then can either delivery using the + // land at the Dispatcher which then can either deliver using the // worker go routine or directly do the invoke the tcp processing inline // based on the state of the endpoint. } @@ -3161,7 +3090,7 @@ func (e *endpoint) State() uint32 { func (e *endpoint) Info() tcpip.EndpointInfo { e.LockUser() // Make a copy of the endpoint info. - ret := e.EndpointInfo + ret := e.TransportEndpointInfo e.UnlockUser() return &ret } @@ -3187,6 +3116,7 @@ func (e *endpoint) Wait() { } } +// SocketOptions implements tcpip.Endpoint.SocketOptions. func (e *endpoint) SocketOptions() *tcpip.SocketOptions { return &e.ops } diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index bb901c0f8..ba67176b5 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -321,21 +321,21 @@ func (e *endpoint) loadRecentTSTime(unix unixTime) { } // saveHardError is invoked by stateify. -func (e *EndpointInfo) saveHardError() string { - if e.HardError == nil { +func (e *endpoint) saveHardError() string { + if e.hardError == nil { return "" } - return e.HardError.String() + return e.hardError.String() } // loadHardError is invoked by stateify. -func (e *EndpointInfo) loadHardError(s string) { +func (e *endpoint) loadHardError(s string) { if s == "" { return } - e.HardError = tcpip.StringToError(s) + e.hardError = tcpip.StringToError(s) } // saveMeasureTime is invoked by stateify. diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 2329aca4b..672159eed 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -250,7 +250,7 @@ func replyWithReset(stack *stack.Stack, s *segment, tos, ttl uint8) *tcpip.Error ttl = route.DefaultTTL() } - return sendTCP(&route, tcpFields{ + return sendTCP(route, tcpFields{ id: s.id, ttl: ttl, tos: tos, diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index 8e0b7c843..f2b1b68da 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -16,6 +16,7 @@ package tcp import ( "container/heap" + "math" "time" "gvisor.dev/gvisor/pkg/tcpip" @@ -48,6 +49,10 @@ type receiver struct { rcvWndScale uint8 + // prevBufused is the snapshot of endpoint rcvBufUsed taken when we + // advertise a receive window. + prevBufUsed int + closed bool // pendingRcvdSegments is bounded by the receive buffer size of the @@ -80,9 +85,9 @@ func (r *receiver) acceptable(segSeq seqnum.Value, segLen seqnum.Size) bool { // outgoing packets, we should use what we have advertised for acceptability // test. scaledWindowSize := r.rcvWnd >> r.rcvWndScale - if scaledWindowSize > 0xffff { + if scaledWindowSize > math.MaxUint16 { // This is what we actually put in the Window field. - scaledWindowSize = 0xffff + scaledWindowSize = math.MaxUint16 } advertisedWindowSize := scaledWindowSize << r.rcvWndScale return header.Acceptable(segSeq, segLen, r.rcvNxt, r.rcvNxt.Add(advertisedWindowSize)) @@ -106,6 +111,34 @@ func (r *receiver) currentWindow() (curWnd seqnum.Size) { func (r *receiver) getSendParams() (rcvNxt seqnum.Value, rcvWnd seqnum.Size) { newWnd := r.ep.selectWindow() curWnd := r.currentWindow() + unackLen := int(r.ep.snd.maxSentAck.Size(r.rcvNxt)) + bufUsed := r.ep.receiveBufferUsed() + + // Grow the right edge of the window only for payloads larger than the + // the segment overhead OR if the application is actively consuming data. + // + // Avoiding growing the right edge otherwise, addresses a situation below: + // An application has been slow in reading data and we have burst of + // incoming segments lengths < segment overhead. Here, our available free + // memory would reduce drastically when compared to the advertised receive + // window. + // + // For example: With incoming 512 bytes segments, segment overhead of + // 552 bytes (at the time of writing this comment), with receive window + // starting from 1MB and with rcvAdvWndScale being 1, buffer would reach 0 + // when the curWnd is still 19436 bytes, because for every incoming segment + // newWnd would reduce by (552+512) >> rcvAdvWndScale (current value 1), + // while curWnd would reduce by 512 bytes. + // Such a situation causes us to keep tail dropping the incoming segments + // and never advertise zero receive window to the peer. + // + // Linux does a similar check for minimal sk_buff size (128): + // https://github.com/torvalds/linux/blob/d5beb3140f91b1c8a3d41b14d729aefa4dcc58bc/net/ipv4/tcp_input.c#L783 + // + // Also, if the application is reading the data, we keep growing the right + // edge, as we are still advertising a window that we think can be serviced. + toGrow := unackLen >= SegSize || bufUsed <= r.prevBufUsed + // Update rcvAcc only if new window is > previously advertised window. We // should never shrink the acceptable sequence space once it has been // advertised the peer. If we shrink the acceptable sequence space then we @@ -115,7 +148,7 @@ func (r *receiver) getSendParams() (rcvNxt seqnum.Value, rcvWnd seqnum.Size) { // rcvWUP rcvNxt rcvAcc new rcvAcc // <=====curWnd ===> // <========= newWnd > curWnd ========= > - if r.rcvNxt.Add(seqnum.Size(curWnd)).LessThan(r.rcvNxt.Add(seqnum.Size(newWnd))) { + if r.rcvNxt.Add(seqnum.Size(curWnd)).LessThan(r.rcvNxt.Add(seqnum.Size(newWnd))) && toGrow { // If the new window moves the right edge, then update rcvAcc. r.rcvAcc = r.rcvNxt.Add(seqnum.Size(newWnd)) } else { @@ -130,11 +163,24 @@ func (r *receiver) getSendParams() (rcvNxt seqnum.Value, rcvWnd seqnum.Size) { // receiver's estimated RTT. r.rcvWnd = newWnd r.rcvWUP = r.rcvNxt + r.prevBufUsed = bufUsed scaledWnd := r.rcvWnd >> r.rcvWndScale if scaledWnd == 0 { // Increment a metric if we are advertising an actual zero window. r.ep.stats.ReceiveErrors.ZeroRcvWindowState.Increment() } + + // If we started off with a window larger than what can he held in + // the 16bit window field, we ceil the value to the max value. + // While ceiling, we still do not want to grow the right edge when + // not applicable. + if scaledWnd > math.MaxUint16 { + if toGrow { + scaledWnd = seqnum.Size(math.MaxUint16) + } else { + scaledWnd = seqnum.Size(uint16(scaledWnd)) + } + } return r.rcvNxt, scaledWnd } diff --git a/pkg/tcpip/transport/tcp/reno_recovery.go b/pkg/tcpip/transport/tcp/reno_recovery.go new file mode 100644 index 000000000..2aa708e97 --- /dev/null +++ b/pkg/tcpip/transport/tcp/reno_recovery.go @@ -0,0 +1,67 @@ +// 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 + +// renoRecovery stores the variables related to TCP Reno loss recovery +// algorithm. +// +// +stateify savable +type renoRecovery struct { + s *sender +} + +func newRenoRecovery(s *sender) *renoRecovery { + return &renoRecovery{s: s} +} + +func (rr *renoRecovery) DoRecovery(rcvdSeg *segment, fastRetransmit bool) { + ack := rcvdSeg.ackNumber + snd := rr.s + + // We are in fast recovery mode. Ignore the ack if it's out of range. + if !ack.InRange(snd.sndUna, snd.sndNxt+1) { + return + } + + // Don't count this as a duplicate if it is carrying data or + // updating the window. + if rcvdSeg.logicalLen() != 0 || snd.sndWnd != rcvdSeg.window { + return + } + + // Inflate the congestion window if we're getting duplicate acks + // for the packet we retransmitted. + if !fastRetransmit && ack == snd.fr.first { + // We received a dup, inflate the congestion window by 1 packet + // if we're not at the max yet. Only inflate the window if + // regular FastRecovery is in use, RFC6675 does not require + // inflating cwnd on duplicate ACKs. + if snd.sndCwnd < snd.fr.maxCwnd { + snd.sndCwnd++ + } + return + } + + // A partial ack was received. Retransmit this packet and remember it + // so that we don't retransmit it again. + // + // We don't inflate the window because we're putting the same packet + // back onto the wire. + // + // N.B. The retransmit timer will be reset by the caller. + snd.fr.first = ack + snd.dupAckCount = 0 + snd.resendSegment() +} diff --git a/pkg/tcpip/transport/tcp/sack_recovery.go b/pkg/tcpip/transport/tcp/sack_recovery.go new file mode 100644 index 000000000..7e813fa96 --- /dev/null +++ b/pkg/tcpip/transport/tcp/sack_recovery.go @@ -0,0 +1,120 @@ +// 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 + +import "gvisor.dev/gvisor/pkg/tcpip/seqnum" + +// sackRecovery stores the variables related to TCP SACK loss recovery +// algorithm. +// +// +stateify savable +type sackRecovery struct { + s *sender +} + +func newSACKRecovery(s *sender) *sackRecovery { + return &sackRecovery{s: s} +} + +// handleSACKRecovery implements the loss recovery phase as described in RFC6675 +// section 5, step C. +func (sr *sackRecovery) handleSACKRecovery(limit int, end seqnum.Value) (dataSent bool) { + snd := sr.s + snd.SetPipe() + + if smss := int(snd.ep.scoreboard.SMSS()); limit > smss { + // Cap segment size limit to s.smss as SACK recovery requires + // that all retransmissions or new segments send during recovery + // be of <= SMSS. + limit = smss + } + + nextSegHint := snd.writeList.Front() + for snd.outstanding < snd.sndCwnd { + var nextSeg *segment + var rescueRtx bool + nextSeg, nextSegHint, rescueRtx = snd.NextSeg(nextSegHint) + if nextSeg == nil { + return dataSent + } + if !snd.isAssignedSequenceNumber(nextSeg) || snd.sndNxt.LessThanEq(nextSeg.sequenceNumber) { + // New data being sent. + + // Step C.3 described below is handled by + // maybeSendSegment which increments sndNxt when + // a segment is transmitted. + // + // Step C.3 "If any of the data octets sent in + // (C.1) are above HighData, HighData must be + // updated to reflect the transmission of + // previously unsent data." + // + // We pass s.smss as the limit as the Step 2) requires that + // new data sent should be of size s.smss or less. + if sent := snd.maybeSendSegment(nextSeg, limit, end); !sent { + return dataSent + } + dataSent = true + snd.outstanding++ + snd.writeNext = nextSeg.Next() + continue + } + + // Now handle the retransmission case where we matched either step 1,3 or 4 + // of the NextSeg algorithm. + // RFC 6675, Step C.4. + // + // "The estimate of the amount of data outstanding in the network + // must be updated by incrementing pipe by the number of octets + // transmitted in (C.1)." + snd.outstanding++ + dataSent = true + snd.sendSegment(nextSeg) + + segEnd := nextSeg.sequenceNumber.Add(nextSeg.logicalLen()) + if rescueRtx { + // We do the last part of rule (4) of NextSeg here to update + // RescueRxt as until this point we don't know if we are going + // to use the rescue transmission. + snd.fr.rescueRxt = snd.fr.last + } else { + // RFC 6675, Step C.2 + // + // "If any of the data octets sent in (C.1) are below + // HighData, HighRxt MUST be set to the highest sequence + // number of the retransmitted segment unless NextSeg () + // rule (4) was invoked for this retransmission." + snd.fr.highRxt = segEnd - 1 + } + } + return dataSent +} + +func (sr *sackRecovery) DoRecovery(rcvdSeg *segment, fastRetransmit bool) { + snd := sr.s + if fastRetransmit { + snd.resendSegment() + } + + // We are in fast recovery mode. Ignore the ack if it's out of range. + if ack := rcvdSeg.ackNumber; !ack.InRange(snd.sndUna, snd.sndNxt+1) { + return + } + + // RFC 6675 recovery algorithm step C 1-5. + end := snd.sndUna.Add(snd.sndWnd) + dataSent := sr.handleSACKRecovery(snd.maxPayloadSize, end) + snd.postXmit(dataSent) +} diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 2091989cc..5ef73ec74 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -204,7 +204,7 @@ func (s *segment) payloadSize() int { // segMemSize is the amount of memory used to hold the segment data and // the associated metadata. func (s *segment) segMemSize() int { - return segSize + s.data.Size() + return SegSize + s.data.Size() } // parse populates the sequence & ack numbers, flags, and window fields of the diff --git a/pkg/tcpip/transport/tcp/segment_unsafe.go b/pkg/tcpip/transport/tcp/segment_unsafe.go index 0ab7b8f56..392ff0859 100644 --- a/pkg/tcpip/transport/tcp/segment_unsafe.go +++ b/pkg/tcpip/transport/tcp/segment_unsafe.go @@ -19,5 +19,6 @@ import ( ) const ( - segSize = int(unsafe.Sizeof(segment{})) + // SegSize is the minimal size of the segment overhead. + SegSize = int(unsafe.Sizeof(segment{})) ) diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index 0e0fdf14c..baec762e1 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -18,7 +18,6 @@ import ( "fmt" "math" "sort" - "sync/atomic" "time" "gvisor.dev/gvisor/pkg/sleep" @@ -92,6 +91,17 @@ type congestionControl interface { PostRecovery() } +// lossRecovery is an interface that must be implemented by any supported +// loss recovery algorithm. +type lossRecovery interface { + // DoRecovery is invoked when loss is detected and segments need + // to be retransmitted. The cumulative or selective ACK is passed along + // with the flag which identifies whether the connection entered fast + // retransmit with this ACK and to retransmit the first unacknowledged + // segment. + DoRecovery(rcvdSeg *segment, fastRetransmit bool) +} + // sender holds the state necessary to send TCP segments. // // +stateify savable @@ -108,6 +118,9 @@ type sender struct { // fr holds state related to fast recovery. fr fastRecovery + // lr is the loss recovery algorithm used by the sender. + lr lossRecovery + // sndCwnd is the congestion window, in packets. sndCwnd int @@ -276,6 +289,8 @@ func newSender(ep *endpoint, iss, irs seqnum.Value, sndWnd seqnum.Size, mss uint s.cc = s.initCongestionControl(ep.cc) + s.lr = s.initLossRecovery() + // A negative sndWndScale means that no scaling is in use, otherwise we // store the scaling value. if sndWndScale > 0 { @@ -330,6 +345,14 @@ func (s *sender) initCongestionControl(congestionControlName tcpip.CongestionCon } } +// initLossRecovery initiates the loss recovery algorithm for the sender. +func (s *sender) initLossRecovery() lossRecovery { + if s.ep.sackPermitted { + return newSACKRecovery(s) + } + return newRenoRecovery(s) +} + // updateMaxPayloadSize updates the maximum payload size based on the given // MTU. If this is in response to "packet too big" control packets (indicated // by the count argument), it also reduces the number of outstanding packets and @@ -550,7 +573,7 @@ func (s *sender) retransmitTimerExpired() bool { // We were attempting fast recovery but were not successful. // Leave the state. We don't need to update ssthresh because it // has already been updated when entered fast-recovery. - s.leaveFastRecovery() + s.leaveRecovery() } s.state = RTORecovery @@ -789,7 +812,7 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se } if !nextTooBig && seg.data.Size() < available { // Segment is not full. - if s.outstanding > 0 && atomic.LoadUint32(&s.ep.delay) != 0 { + if s.outstanding > 0 && s.ep.ops.GetDelayOption() { // Nagle's algorithm. From Wikipedia: // Nagle's algorithm works by // combining a number of small @@ -808,7 +831,7 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se // send space and MSS. // TODO(gvisor.dev/issue/2833): Drain the held segments after a // timeout. - if seg.data.Size() < s.maxPayloadSize && atomic.LoadUint32(&s.ep.cork) != 0 { + if seg.data.Size() < s.maxPayloadSize && s.ep.ops.GetCorkOption() { return false } } @@ -913,79 +936,6 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se return true } -// handleSACKRecovery implements the loss recovery phase as described in RFC6675 -// section 5, step C. -func (s *sender) handleSACKRecovery(limit int, end seqnum.Value) (dataSent bool) { - s.SetPipe() - - if smss := int(s.ep.scoreboard.SMSS()); limit > smss { - // Cap segment size limit to s.smss as SACK recovery requires - // that all retransmissions or new segments send during recovery - // be of <= SMSS. - limit = smss - } - - nextSegHint := s.writeList.Front() - for s.outstanding < s.sndCwnd { - var nextSeg *segment - var rescueRtx bool - nextSeg, nextSegHint, rescueRtx = s.NextSeg(nextSegHint) - if nextSeg == nil { - return dataSent - } - if !s.isAssignedSequenceNumber(nextSeg) || s.sndNxt.LessThanEq(nextSeg.sequenceNumber) { - // New data being sent. - - // Step C.3 described below is handled by - // maybeSendSegment which increments sndNxt when - // a segment is transmitted. - // - // Step C.3 "If any of the data octets sent in - // (C.1) are above HighData, HighData must be - // updated to reflect the transmission of - // previously unsent data." - // - // We pass s.smss as the limit as the Step 2) requires that - // new data sent should be of size s.smss or less. - if sent := s.maybeSendSegment(nextSeg, limit, end); !sent { - return dataSent - } - dataSent = true - s.outstanding++ - s.writeNext = nextSeg.Next() - continue - } - - // Now handle the retransmission case where we matched either step 1,3 or 4 - // of the NextSeg algorithm. - // RFC 6675, Step C.4. - // - // "The estimate of the amount of data outstanding in the network - // must be updated by incrementing pipe by the number of octets - // transmitted in (C.1)." - s.outstanding++ - dataSent = true - s.sendSegment(nextSeg) - - segEnd := nextSeg.sequenceNumber.Add(nextSeg.logicalLen()) - if rescueRtx { - // We do the last part of rule (4) of NextSeg here to update - // RescueRxt as until this point we don't know if we are going - // to use the rescue transmission. - s.fr.rescueRxt = s.fr.last - } else { - // RFC 6675, Step C.2 - // - // "If any of the data octets sent in (C.1) are below - // HighData, HighRxt MUST be set to the highest sequence - // number of the retransmitted segment unless NextSeg () - // rule (4) was invoked for this retransmission." - s.fr.highRxt = segEnd - 1 - } - } - return dataSent -} - func (s *sender) sendZeroWindowProbe() { ack, win := s.ep.rcv.getSendParams() s.unackZeroWindowProbes++ @@ -1014,6 +964,30 @@ func (s *sender) disableZeroWindowProbing() { s.resendTimer.disable() } +func (s *sender) postXmit(dataSent bool) { + if dataSent { + // We sent data, so we should stop the keepalive timer to ensure + // that no keepalives are sent while there is pending data. + s.ep.disableKeepaliveTimer() + } + + // If the sender has advertized zero receive window and we have + // data to be sent out, start zero window probing to query the + // the remote for it's receive window size. + if s.writeNext != nil && s.sndWnd == 0 { + s.enableZeroWindowProbing() + } + + // Enable the timer if we have pending data and it's not enabled yet. + if !s.resendTimer.enabled() && s.sndUna != s.sndNxt { + s.resendTimer.enable(s.rto) + } + // If we have no more pending data, start the keepalive timer. + if s.sndUna == s.sndNxt { + s.ep.resetKeepaliveTimer(false) + } +} + // sendData sends new data segments. It is called when data becomes available or // when the send window opens up. func (s *sender) sendData() { @@ -1034,55 +1008,29 @@ func (s *sender) sendData() { } var dataSent bool - - // RFC 6675 recovery algorithm step C 1-5. - if s.fr.active && s.ep.sackPermitted { - dataSent = s.handleSACKRecovery(s.maxPayloadSize, end) - } else { - for seg := s.writeNext; seg != nil && s.outstanding < s.sndCwnd; seg = seg.Next() { - cwndLimit := (s.sndCwnd - s.outstanding) * s.maxPayloadSize - if cwndLimit < limit { - limit = cwndLimit - } - if s.isAssignedSequenceNumber(seg) && s.ep.sackPermitted && s.ep.scoreboard.IsSACKED(seg.sackBlock()) { - // Move writeNext along so that we don't try and scan data that - // has already been SACKED. - s.writeNext = seg.Next() - continue - } - if sent := s.maybeSendSegment(seg, limit, end); !sent { - break - } - dataSent = true - s.outstanding += s.pCount(seg) + for seg := s.writeNext; seg != nil && s.outstanding < s.sndCwnd; seg = seg.Next() { + cwndLimit := (s.sndCwnd - s.outstanding) * s.maxPayloadSize + if cwndLimit < limit { + limit = cwndLimit + } + if s.isAssignedSequenceNumber(seg) && s.ep.sackPermitted && s.ep.scoreboard.IsSACKED(seg.sackBlock()) { + // Move writeNext along so that we don't try and scan data that + // has already been SACKED. s.writeNext = seg.Next() + continue } + if sent := s.maybeSendSegment(seg, limit, end); !sent { + break + } + dataSent = true + s.outstanding += s.pCount(seg) + s.writeNext = seg.Next() } - if dataSent { - // We sent data, so we should stop the keepalive timer to ensure - // that no keepalives are sent while there is pending data. - s.ep.disableKeepaliveTimer() - } - - // If the sender has advertized zero receive window and we have - // data to be sent out, start zero window probing to query the - // the remote for it's receive window size. - if s.writeNext != nil && s.sndWnd == 0 { - s.enableZeroWindowProbing() - } - - // Enable the timer if we have pending data and it's not enabled yet. - if !s.resendTimer.enabled() && s.sndUna != s.sndNxt { - s.resendTimer.enable(s.rto) - } - // If we have no more pending data, start the keepalive timer. - if s.sndUna == s.sndNxt { - s.ep.resetKeepaliveTimer(false) - } + s.postXmit(dataSent) } -func (s *sender) enterFastRecovery() { +func (s *sender) enterRecovery() { s.fr.active = true // Save state to reflect we're now in fast recovery. // @@ -1104,7 +1052,7 @@ func (s *sender) enterFastRecovery() { s.ep.stack.Stats().TCP.FastRecovery.Increment() } -func (s *sender) leaveFastRecovery() { +func (s *sender) leaveRecovery() { s.fr.active = false s.fr.maxCwnd = 0 s.dupAckCount = 0 @@ -1115,57 +1063,6 @@ func (s *sender) leaveFastRecovery() { s.cc.PostRecovery() } -func (s *sender) handleFastRecovery(seg *segment) (rtx bool) { - ack := seg.ackNumber - // We are in fast recovery mode. Ignore the ack if it's out of - // range. - if !ack.InRange(s.sndUna, s.sndNxt+1) { - return false - } - - // Leave fast recovery if it acknowledges all the data covered by - // this fast recovery session. - if s.fr.last.LessThan(ack) { - s.leaveFastRecovery() - return false - } - - if s.ep.sackPermitted { - // When SACK is enabled we let retransmission be governed by - // the SACK logic. - return false - } - - // Don't count this as a duplicate if it is carrying data or - // updating the window. - if seg.logicalLen() != 0 || s.sndWnd != seg.window { - return false - } - - // Inflate the congestion window if we're getting duplicate acks - // for the packet we retransmitted. - if ack == s.fr.first { - // We received a dup, inflate the congestion window by 1 packet - // if we're not at the max yet. Only inflate the window if - // regular FastRecovery is in use, RFC6675 does not require - // inflating cwnd on duplicate ACKs. - if s.sndCwnd < s.fr.maxCwnd { - s.sndCwnd++ - } - return false - } - - // A partial ack was received. Retransmit this packet and - // remember it so that we don't retransmit it again. We don't - // inflate the window because we're putting the same packet back - // onto the wire. - // - // N.B. The retransmit timer will be reset by the caller. - s.fr.first = ack - s.dupAckCount = 0 - return true -} - // isAssignedSequenceNumber relies on the fact that we only set flags once a // sequencenumber is assigned and that is only done right before we send the // segment. As a result any segment that has a non-zero flag has a valid @@ -1228,14 +1125,11 @@ func (s *sender) SetPipe() { s.outstanding = pipe } -// checkDuplicateAck is called when an ack is received. It manages the state -// related to duplicate acks and determines if a retransmit is needed according -// to the rules in RFC 6582 (NewReno). -func (s *sender) checkDuplicateAck(seg *segment) (rtx bool) { +// detectLoss is called when an ack is received and returns whether a loss is +// detected. It manages the state related to duplicate acks and determines if +// a retransmit is needed according to the rules in RFC 6582 (NewReno). +func (s *sender) detectLoss(seg *segment) (fastRetransmit bool) { ack := seg.ackNumber - if s.fr.active { - return s.handleFastRecovery(seg) - } // We're not in fast recovery yet. A segment is considered a duplicate // only if it doesn't carry any data and doesn't update the send window, @@ -1266,14 +1160,14 @@ func (s *sender) checkDuplicateAck(seg *segment) (rtx bool) { // See: https://tools.ietf.org/html/rfc6582#section-3.2 Step 2 // // We only do the check here, the incrementing of last to the highest - // sequence number transmitted till now is done when enterFastRecovery + // sequence number transmitted till now is done when enterRecovery // is invoked. if !s.fr.last.LessThan(seg.ackNumber) { s.dupAckCount = 0 return false } s.cc.HandleNDupAcks() - s.enterFastRecovery() + s.enterRecovery() s.dupAckCount = 0 return true } @@ -1415,14 +1309,23 @@ func (s *sender) handleRcvdSegment(rcvdSeg *segment) { s.SetPipe() } - // Count the duplicates and do the fast retransmit if needed. - rtx := s.checkDuplicateAck(rcvdSeg) + ack := rcvdSeg.ackNumber + fastRetransmit := false + // Do not leave fast recovery, if the ACK is out of range. + if s.fr.active { + // Leave fast recovery if it acknowledges all the data covered by + // this fast recovery session. + if ack.InRange(s.sndUna, s.sndNxt+1) && s.fr.last.LessThan(ack) { + s.leaveRecovery() + } + } else { + // Detect loss by counting the duplicates and enter recovery. + fastRetransmit = s.detectLoss(rcvdSeg) + } // Stash away the current window size. s.sndWnd = rcvdSeg.window - ack := rcvdSeg.ackNumber - // Disable zero window probing if remote advertizes a non-zero receive // window. This can be with an ACK to the zero window probe (where the // acknumber refers to the already acknowledged byte) OR to any previously @@ -1539,19 +1442,24 @@ func (s *sender) handleRcvdSegment(rcvdSeg *segment) { s.resendTimer.disable() } } + // Now that we've popped all acknowledged data from the retransmit // queue, retransmit if needed. - if rtx { - s.resendSegment() + if s.fr.active { + s.lr.DoRecovery(rcvdSeg, fastRetransmit) + // When SACK is enabled data sending is governed by steps in + // RFC 6675 Section 5 recovery steps A-C. + // See: https://tools.ietf.org/html/rfc6675#section-5. + if s.ep.sackPermitted { + return + } } // Send more data now that some of the pending data has been ack'd, or // that the window opened up, or the congestion window was inflated due // to a duplicate ack during fast recovery. This will also re-enable // the retransmit timer if needed. - if !s.ep.sackPermitted || s.fr.active || s.dupAckCount == 0 || rcvdSeg.hasNewSACKInfo { - s.sendData() - } + s.sendData() } // sendSegment sends the specified segment. diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 9f0fb41e3..1759ebea9 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -75,9 +75,6 @@ func TestGiveUpConnect(t *testing.T) { // Wait for ep to become writable. <-notifyCh - if err := ep.LastError(); err != tcpip.ErrAborted { - t.Fatalf("got ep.LastError() = %s, want = %s", err, tcpip.ErrAborted) - } // Call Connect again to retreive the handshake failure status // and stats updates. @@ -267,7 +264,7 @@ func TestTCPResetsSentNoICMP(t *testing.T) { } // Read outgoing ICMP stats and check no ICMP DstUnreachable was recorded. - sent := stats.ICMP.V4PacketsSent + sent := stats.ICMP.V4.PacketsSent if got, want := sent.DstUnreachable.Value(), uint64(0); got != want { t.Errorf("got ICMP DstUnreachable.Value() = %d, want = %d", got, want) } @@ -2532,10 +2529,10 @@ func TestSegmentMerging(t *testing.T) { { "cork", func(ep tcpip.Endpoint) { - ep.SetSockOptBool(tcpip.CorkOption, true) + ep.SocketOptions().SetCorkOption(true) }, func(ep tcpip.Endpoint) { - ep.SetSockOptBool(tcpip.CorkOption, false) + ep.SocketOptions().SetCorkOption(false) }, }, } @@ -2627,7 +2624,7 @@ func TestDelay(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - c.EP.SetSockOptBool(tcpip.DelayOption, true) + c.EP.SocketOptions().SetDelayOption(true) var allData []byte for i, data := range [][]byte{{0}, {1, 2, 3, 4}, {5, 6, 7}, {8, 9}, {10}, {11}} { @@ -2675,7 +2672,7 @@ func TestUndelay(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - c.EP.SetSockOptBool(tcpip.DelayOption, true) + c.EP.SocketOptions().SetDelayOption(true) allData := [][]byte{{0}, {1, 2, 3}} for i, data := range allData { @@ -2708,7 +2705,7 @@ func TestUndelay(t *testing.T) { // Check that we don't get the second packet yet. c.CheckNoPacketTimeout("delayed second packet transmitted", 100*time.Millisecond) - c.EP.SetSockOptBool(tcpip.DelayOption, false) + c.EP.SocketOptions().SetDelayOption(false) // Check that data is received. second := c.GetPacket() @@ -2745,8 +2742,8 @@ func TestMSSNotDelayed(t *testing.T) { fn func(tcpip.Endpoint) }{ {"no-op", func(tcpip.Endpoint) {}}, - {"delay", func(ep tcpip.Endpoint) { ep.SetSockOptBool(tcpip.DelayOption, true) }}, - {"cork", func(ep tcpip.Endpoint) { ep.SetSockOptBool(tcpip.CorkOption, true) }}, + {"delay", func(ep tcpip.Endpoint) { ep.SocketOptions().SetDelayOption(true) }}, + {"cork", func(ep tcpip.Endpoint) { ep.SocketOptions().SetCorkOption(true) }}, } for _, test := range tests { @@ -3198,6 +3195,11 @@ loop: case tcpip.ErrWouldBlock: select { case <-ch: + // Expect the state to be StateError and subsequent Reads to fail with HardError. + if _, _, err := c.EP.Read(nil); err != tcpip.ErrConnectionReset { + t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrConnectionReset) + } + break loop case <-time.After(1 * time.Second): t.Fatalf("Timed out waiting for reset to arrive") } @@ -3207,14 +3209,10 @@ loop: t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrConnectionReset) } } - // Expect the state to be StateError and subsequent Reads to fail with HardError. - if _, _, err := c.EP.Read(nil); err != tcpip.ErrConnectionReset { - t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrConnectionReset) - } + if tcp.EndpointState(c.EP.State()) != tcp.StateError { t.Fatalf("got EP state is not StateError") } - if got := c.Stack().Stats().TCP.EstablishedResets.Value(); got != 1 { t.Errorf("got stats.TCP.EstablishedResets.Value() = %d, want = 1", got) } @@ -4193,9 +4191,7 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } - if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { - t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) - } + c.EP.SocketOptions().SetReuseAddress(true) if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -4205,9 +4201,7 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } - if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { - t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) - } + c.EP.SocketOptions().SetReuseAddress(true) if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -4218,9 +4212,7 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } - if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { - t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) - } + c.EP.SocketOptions().SetReuseAddress(true) if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -4233,9 +4225,7 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } - if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { - t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) - } + c.EP.SocketOptions().SetReuseAddress(true) if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -4246,9 +4236,7 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } - if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { - t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) - } + c.EP.SocketOptions().SetReuseAddress(true) if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -4261,9 +4249,7 @@ func TestReusePort(t *testing.T) { if err != nil { t.Fatalf("NewEndpoint failed; %s", err) } - if err := c.EP.SetSockOptBool(tcpip.ReuseAddressOption, true); err != nil { - t.Fatalf("SetSockOptBool ReuseAddressOption failed: %s", err) - } + c.EP.SocketOptions().SetReuseAddress(true) if err := c.EP.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { t.Fatalf("Bind failed: %s", err) } @@ -4656,13 +4642,9 @@ func TestConnectAvoidsBoundPorts(t *testing.T) { switch network { case "ipv4": case "ipv6": - if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - t.Fatalf("SetSockOptBool(V6OnlyOption(true)) failed: %s", err) - } + ep.SocketOptions().SetV6Only(true) case "dual": - if err := ep.SetSockOptBool(tcpip.V6OnlyOption, false); err != nil { - t.Fatalf("SetSockOptBool(V6OnlyOption(false)) failed: %s", err) - } + ep.SocketOptions().SetV6Only(false) default: t.Fatalf("unknown network: '%s'", network) } @@ -4998,9 +4980,7 @@ func TestKeepalive(t *testing.T) { if err := c.EP.SetSockOptInt(tcpip.KeepaliveCountOption, 5); err != nil { t.Fatalf("c.EP.SetSockOptInt(tcpip.KeepaliveCountOption, 5): %s", err) } - if err := c.EP.SetSockOptBool(tcpip.KeepaliveEnabledOption, true); err != nil { - t.Fatalf("c.EP.SetSockOptBool(tcpip.KeepaliveEnabledOption, true): %s", err) - } + c.EP.SocketOptions().SetKeepAlive(true) // 5 unacked keepalives are sent. ACK each one, and check that the // connection stays alive after 5. @@ -6118,10 +6098,13 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // Introduce a 25ms latency by delaying the first byte. latency := 25 * time.Millisecond time.Sleep(latency) - rawEP.SendPacketWithTS([]byte{1}, tsVal) + // Send an initial payload with atleast segment overhead size. The receive + // window would not grow for smaller segments. + rawEP.SendPacketWithTS(make([]byte, tcp.SegSize), tsVal) pkt := rawEP.VerifyAndReturnACKWithTS(tsVal) rcvWnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize() + time.Sleep(25 * time.Millisecond) // Allocate a large enough payload for the test. @@ -6394,10 +6377,7 @@ func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcpip.T if err != nil { t.Fatalf("NewEndPoint(tcp, ipv4, new(waiter.Queue)) failed: %s", err) } - gotDelayOption, err := ep.GetSockOptBool(tcpip.DelayOption) - if err != nil { - t.Fatalf("ep.GetSockOptBool(tcpip.DelayOption) failed: %s", err) - } + gotDelayOption := ep.SocketOptions().GetDelayOption() if gotDelayOption != wantDelayOption { t.Errorf("ep.GetSockOptBool(tcpip.DelayOption) got: %t, want: %t", gotDelayOption, wantDelayOption) } @@ -7250,9 +7230,7 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { if err := c.EP.SetSockOptInt(tcpip.KeepaliveCountOption, 10); err != nil { t.Fatalf("c.EP.SetSockOptInt(tcpip.KeepaliveCountOption, 10): %s", err) } - if err := c.EP.SetSockOptBool(tcpip.KeepaliveEnabledOption, true); err != nil { - t.Fatalf("c.EP.SetSockOptBool(tcpip.KeepaliveEnabledOption, true): %s", err) - } + c.EP.SocketOptions().SetKeepAlive(true) // Set userTimeout to be the duration to be 1 keepalive // probes. Which means that after the first probe is sent diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index e6aa4fc4b..010a23e45 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -592,9 +592,7 @@ func (c *Context) CreateV6Endpoint(v6only bool) { c.t.Fatalf("NewEndpoint failed: %v", err) } - if err := c.EP.SetSockOptBool(tcpip.V6OnlyOption, v6only); err != nil { - c.t.Fatalf("SetSockOpt failed failed: %v", err) - } + c.EP.SocketOptions().SetV6Only(v6only) } // GetV6Packet reads a single packet from the link layer endpoint of the context diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index c78549424..7ebae63d8 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -56,6 +56,7 @@ go_test( "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", "//pkg/tcpip/stack", + "//pkg/tcpip/transport/icmp", "//pkg/waiter", ], ) diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 57976d4e3..5043e7aa5 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -16,8 +16,8 @@ package udp import ( "fmt" + "sync/atomic" - "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -77,6 +77,7 @@ func (s EndpointState) String() string { // +stateify savable type endpoint struct { stack.TransportEndpointInfo + tcpip.DefaultSocketOptionsHandler // The following fields are initialized at creation time and do not // change throughout the lifetime of the endpoint. @@ -94,21 +95,20 @@ type endpoint struct { rcvClosed bool // The following fields are protected by the mu mutex. - mu sync.RWMutex `state:"nosave"` - sndBufSize int - sndBufSizeMax int + mu sync.RWMutex `state:"nosave"` + sndBufSize int + sndBufSizeMax int + // state must be read/set using the EndpointState()/setEndpointState() + // methods. state EndpointState - route stack.Route `state:"manual"` + route *stack.Route `state:"manual"` dstPort uint16 - v6only bool ttl uint8 multicastTTL uint8 multicastAddr tcpip.Address multicastNICID tcpip.NICID - multicastLoop bool portFlags ports.Flags bindToDevice tcpip.NICID - noChecksum bool lastErrorMu sync.Mutex `state:"nosave"` lastError *tcpip.Error `state:".(string)"` @@ -122,17 +122,6 @@ type endpoint struct { // applied while sending packets. Defaults to 0 as on Linux. sendTOS uint8 - // receiveTOS determines if the incoming IPv4 TOS header field is passed - // as ancillary data to ControlMessages on Read. - receiveTOS bool - - // receiveTClass determines if the incoming IPv6 TClass header field is - // passed as ancillary data to ControlMessages on Read. - receiveTClass bool - - // receiveIPPacketInfo determines if the packet info is returned by Read. - receiveIPPacketInfo bool - // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags @@ -188,13 +177,14 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue // // Linux defaults to TTL=1. multicastTTL: 1, - multicastLoop: true, rcvBufSizeMax: 32 * 1024, sndBufSizeMax: 32 * 1024, multicastMemberships: make(map[multicastMembership]struct{}), state: StateInitial, uniqueID: s.UniqueID(), } + e.ops.InitHandler(e) + e.ops.SetMulticastLoop(true) // Override with stack defaults. var ss stack.SendBufferSizeOption @@ -210,6 +200,20 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue return e } +// setEndpointState updates the state of the endpoint to state atomically. This +// method is unexported as the only place we should update the state is in this +// package but we allow the state to be read freely without holding e.mu. +// +// Precondition: e.mu must be held to call this method. +func (e *endpoint) setEndpointState(state EndpointState) { + atomic.StoreUint32((*uint32)(&e.state), uint32(state)) +} + +// EndpointState() returns the current state of the endpoint. +func (e *endpoint) EndpointState() EndpointState { + return EndpointState(atomic.LoadUint32((*uint32)(&e.state))) +} + // UniqueID implements stack.TransportEndpoint.UniqueID. func (e *endpoint) UniqueID() uint64 { return e.uniqueID @@ -235,7 +239,7 @@ func (e *endpoint) Close() { e.mu.Lock() e.shutdownFlags = tcpip.ShutdownRead | tcpip.ShutdownWrite - switch e.state { + switch e.EndpointState() { case StateBound, StateConnected: e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundPortFlags, e.boundBindToDevice) e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, e.boundPortFlags, e.boundBindToDevice, tcpip.FullAddress{}) @@ -258,10 +262,13 @@ func (e *endpoint) Close() { } e.rcvMu.Unlock() - e.route.Release() + if e.route != nil { + e.route.Release() + e.route = nil + } // Update the state. - e.state = StateClosed + e.setEndpointState(StateClosed) e.mu.Unlock() @@ -303,21 +310,16 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess HasTimestamp: true, Timestamp: p.timestamp, } - e.mu.RLock() - receiveTOS := e.receiveTOS - receiveTClass := e.receiveTClass - receiveIPPacketInfo := e.receiveIPPacketInfo - e.mu.RUnlock() - if receiveTOS { + if e.ops.GetReceiveTOS() { cm.HasTOS = true cm.TOS = p.tos } - if receiveTClass { + if e.ops.GetReceiveTClass() { cm.HasTClass = true // Although TClass is an 8-bit value it's read in the CMsg as a uint32. cm.TClass = uint32(p.tos) } - if receiveIPPacketInfo { + if e.ops.GetReceivePacketInfo() { cm.HasIPPacketInfo = true cm.PacketInfo = p.packetInfo } @@ -330,7 +332,7 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess // // Returns true for retry if preparation should be retried. func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpip.Error) { - switch e.state { + switch e.EndpointState() { case StateInitial: case StateConnected: return false, nil @@ -352,7 +354,7 @@ func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpi // The state changed when we released the shared locked and re-acquired // it in exclusive mode. Try again. - if e.state != StateInitial { + if e.EndpointState() != StateInitial { return true, nil } @@ -367,9 +369,9 @@ func (e *endpoint) prepareForWrite(to *tcpip.FullAddress) (retry bool, err *tcpi // connectRoute establishes a route to the specified interface or the // configured multicast interface if no interface is specified and the // specified address is a multicast address. -func (e *endpoint) connectRoute(nicID tcpip.NICID, addr tcpip.FullAddress, netProto tcpip.NetworkProtocolNumber) (stack.Route, tcpip.NICID, *tcpip.Error) { +func (e *endpoint) connectRoute(nicID tcpip.NICID, addr tcpip.FullAddress, netProto tcpip.NetworkProtocolNumber) (*stack.Route, tcpip.NICID, *tcpip.Error) { localAddr := e.ID.LocalAddress - if isBroadcastOrMulticast(localAddr) { + if e.isBroadcastOrMulticast(nicID, netProto, localAddr) { // A packet can only originate from a unicast address (i.e., an interface). localAddr = "" } @@ -384,9 +386,9 @@ func (e *endpoint) connectRoute(nicID tcpip.NICID, addr tcpip.FullAddress, netPr } // Find a route to the desired destination. - r, err := e.stack.FindRoute(nicID, localAddr, addr.Addr, netProto, e.multicastLoop) + r, err := e.stack.FindRoute(nicID, localAddr, addr.Addr, netProto, e.ops.GetMulticastLoop()) if err != nil { - return stack.Route{}, 0, err + return nil, 0, err } return r, nicID, nil } @@ -429,7 +431,13 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c to := opts.To e.mu.RLock() - defer e.mu.RUnlock() + lockReleased := false + defer func() { + if lockReleased { + return + } + e.mu.RUnlock() + }() // If we've shutdown with SHUT_WR we are in an invalid state for sending. if e.shutdownFlags&tcpip.ShutdownWrite != 0 { @@ -448,36 +456,9 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } } - var route *stack.Route - var resolve func(waker *sleep.Waker) (ch <-chan struct{}, err *tcpip.Error) - var dstPort uint16 - if to == nil { - route = &e.route - dstPort = e.dstPort - resolve = func(waker *sleep.Waker) (ch <-chan struct{}, err *tcpip.Error) { - // Promote lock to exclusive if using a shared route, given that it may - // need to change in Route.Resolve() call below. - e.mu.RUnlock() - e.mu.Lock() - - // Recheck state after lock was re-acquired. - if e.state != StateConnected { - err = tcpip.ErrInvalidEndpointState - } - if err == nil && route.IsResolutionRequired() { - ch, err = route.Resolve(waker) - } - - e.mu.Unlock() - e.mu.RLock() - - // Recheck state after lock was re-acquired. - if e.state != StateConnected { - err = tcpip.ErrInvalidEndpointState - } - return - } - } else { + route := e.route + dstPort := e.dstPort + if to != nil { // Reject destination address if it goes through a different // NIC than the endpoint was bound to. nicID := to.NIC @@ -505,9 +486,8 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } defer r.Release() - route = &r + route = r dstPort = dst.Port - resolve = route.Resolve } if !e.ops.GetBroadcast() && route.IsOutboundBroadcast() { @@ -515,7 +495,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c } if route.IsResolutionRequired() { - if ch, err := resolve(nil); err != nil { + if ch, err := route.Resolve(nil); err != nil { if err == tcpip.ErrWouldBlock { return 0, ch, tcpip.ErrNoLinkAddress } @@ -541,7 +521,24 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c useDefaultTTL = false } - if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS, e.owner, e.noChecksum); err != nil { + localPort := e.ID.LocalPort + sendTOS := e.sendTOS + owner := e.owner + noChecksum := e.SocketOptions().GetNoChecksum() + lockReleased = true + e.mu.RUnlock() + + // Do not hold lock when sending as loopback is synchronous and if the UDP + // datagram ends up generating an ICMP response then it can result in a + // deadlock where the ICMP response handling ends up acquiring this endpoint's + // mutex using e.mu.RLock() in endpoint.HandleControlPacket which can cause a + // deadlock if another caller is trying to acquire e.mu in exclusive mode w/ + // e.mu.Lock(). Since e.mu.Lock() prevents any new read locks to ensure the + // lock can be eventually acquired. + // + // See: https://golang.org/pkg/sync/#RWMutex for details on why recursive read + // locking is prohibited. + if err := sendUDP(route, buffer.View(v).ToVectorisedView(), localPort, dstPort, ttl, useDefaultTTL, sendTOS, owner, noChecksum); err != nil { return 0, nil, err } return int64(len(v)), nil, nil @@ -552,66 +549,18 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { return 0, tcpip.ControlMessages{}, nil } -// SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. -func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - switch opt { - case tcpip.MulticastLoopOption: - e.mu.Lock() - e.multicastLoop = v - e.mu.Unlock() - - case tcpip.NoChecksumOption: - e.mu.Lock() - e.noChecksum = v - e.mu.Unlock() - - case tcpip.ReceiveTOSOption: - e.mu.Lock() - e.receiveTOS = v - e.mu.Unlock() - - case tcpip.ReceiveTClassOption: - // We only support this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return tcpip.ErrNotSupported - } - - e.mu.Lock() - e.receiveTClass = v - e.mu.Unlock() - - case tcpip.ReceiveIPPacketInfoOption: - e.mu.Lock() - e.receiveIPPacketInfo = v - e.mu.Unlock() - - case tcpip.ReuseAddressOption: - e.mu.Lock() - e.portFlags.MostRecent = v - e.mu.Unlock() - - case tcpip.ReusePortOption: - e.mu.Lock() - e.portFlags.LoadBalanced = v - e.mu.Unlock() - - case tcpip.V6OnlyOption: - // We only recognize this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return tcpip.ErrInvalidEndpointState - } - - e.mu.Lock() - defer e.mu.Unlock() - - // We only allow this to be set when we're in the initial state. - if e.state != StateInitial { - return tcpip.ErrInvalidEndpointState - } +// OnReuseAddressSet implements tcpip.SocketOptionsHandler.OnReuseAddressSet. +func (e *endpoint) OnReuseAddressSet(v bool) { + e.mu.Lock() + e.portFlags.MostRecent = v + e.mu.Unlock() +} - e.v6only = v - } - return nil +// OnReusePortSet implements tcpip.SocketOptionsHandler.OnReusePortSet. +func (e *endpoint) OnReusePortSet(v bool) { + e.mu.Lock() + e.portFlags.LoadBalanced = v + e.mu.Unlock() } // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. @@ -823,81 +772,6 @@ func (e *endpoint) SetSockOpt(opt tcpip.SettableSocketOption) *tcpip.Error { return nil } -// GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. -func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - switch opt { - case tcpip.KeepaliveEnabledOption: - return false, nil - - case tcpip.MulticastLoopOption: - e.mu.RLock() - v := e.multicastLoop - e.mu.RUnlock() - return v, nil - - case tcpip.NoChecksumOption: - e.mu.RLock() - v := e.noChecksum - e.mu.RUnlock() - return v, nil - - case tcpip.ReceiveTOSOption: - e.mu.RLock() - v := e.receiveTOS - e.mu.RUnlock() - return v, nil - - case tcpip.ReceiveTClassOption: - // We only support this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return false, tcpip.ErrNotSupported - } - - e.mu.RLock() - v := e.receiveTClass - e.mu.RUnlock() - return v, nil - - case tcpip.ReceiveIPPacketInfoOption: - e.mu.RLock() - v := e.receiveIPPacketInfo - e.mu.RUnlock() - return v, nil - - case tcpip.ReuseAddressOption: - e.mu.RLock() - v := e.portFlags.MostRecent - e.mu.RUnlock() - - return v, nil - - case tcpip.ReusePortOption: - e.mu.RLock() - v := e.portFlags.LoadBalanced - e.mu.RUnlock() - - return v, nil - - case tcpip.V6OnlyOption: - // We only recognize this option on v6 endpoints. - if e.NetProto != header.IPv6ProtocolNumber { - return false, tcpip.ErrUnknownProtocolOption - } - - e.mu.RLock() - v := e.v6only - e.mu.RUnlock() - - return v, nil - - case tcpip.AcceptConnOption: - return false, nil - - default: - return false, tcpip.ErrUnknownProtocolOption - } -} - // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { @@ -1036,7 +910,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u // checkV4MappedLocked determines the effective network protocol and converts // addr to its canonical form. func (e *endpoint) checkV4MappedLocked(addr tcpip.FullAddress) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) { - unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProtoLocked(addr, e.v6only) + unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProtoLocked(addr, e.ops.GetV6Only()) if err != nil { return tcpip.FullAddress{}, 0, err } @@ -1048,7 +922,7 @@ func (e *endpoint) Disconnect() *tcpip.Error { e.mu.Lock() defer e.mu.Unlock() - if e.state != StateConnected { + if e.EndpointState() != StateConnected { return nil } var ( @@ -1071,7 +945,7 @@ func (e *endpoint) Disconnect() *tcpip.Error { if err != nil { return err } - e.state = StateBound + e.setEndpointState(StateBound) boundPortFlags = e.boundPortFlags } else { if e.ID.LocalPort != 0 { @@ -1079,14 +953,14 @@ func (e *endpoint) Disconnect() *tcpip.Error { e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.ID.LocalAddress, e.ID.LocalPort, boundPortFlags, e.boundBindToDevice, tcpip.FullAddress{}) e.boundPortFlags = ports.Flags{} } - e.state = StateInitial + e.setEndpointState(StateInitial) } e.stack.UnregisterTransportEndpoint(e.RegisterNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, boundPortFlags, e.boundBindToDevice) e.ID = id e.boundBindToDevice = btd e.route.Release() - e.route = stack.Route{} + e.route = nil e.dstPort = 0 return nil @@ -1104,7 +978,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { nicID := addr.NIC var localPort uint16 - switch e.state { + switch e.EndpointState() { case StateInitial: case StateBound, StateConnected: localPort = e.ID.LocalPort @@ -1139,7 +1013,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { RemoteAddress: r.RemoteAddress, } - if e.state == StateInitial { + if e.EndpointState() == StateInitial { id.LocalAddress = r.LocalAddress } @@ -1147,7 +1021,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // packets on a different network protocol, so we register both even if // v6only is set to false and this is an ipv6 endpoint. netProtos := []tcpip.NetworkProtocolNumber{netProto} - if netProto == header.IPv6ProtocolNumber && !e.v6only { + if netProto == header.IPv6ProtocolNumber && !e.ops.GetV6Only() { netProtos = []tcpip.NetworkProtocolNumber{ header.IPv4ProtocolNumber, header.IPv6ProtocolNumber, @@ -1173,7 +1047,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { e.RegisterNICID = nicID e.effectiveNetProtos = netProtos - e.state = StateConnected + e.setEndpointState(StateConnected) e.rcvMu.Lock() e.rcvReady = true @@ -1195,7 +1069,7 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // A socket in the bound state can still receive multicast messages, // so we need to notify waiters on shutdown. - if e.state != StateBound && e.state != StateConnected { + if state := e.EndpointState(); state != StateBound && state != StateConnected { return tcpip.ErrNotConnected } @@ -1246,7 +1120,7 @@ func (e *endpoint) registerWithStack(nicID tcpip.NICID, netProtos []tcpip.Networ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { // Don't allow binding once endpoint is not in the initial state // anymore. - if e.state != StateInitial { + if e.EndpointState() != StateInitial { return tcpip.ErrInvalidEndpointState } @@ -1259,7 +1133,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { // wildcard (empty) address, and this is an IPv6 endpoint with v6only // set to false. netProtos := []tcpip.NetworkProtocolNumber{netProto} - if netProto == header.IPv6ProtocolNumber && !e.v6only && addr.Addr == "" { + if netProto == header.IPv6ProtocolNumber && !e.ops.GetV6Only() && addr.Addr == "" { netProtos = []tcpip.NetworkProtocolNumber{ header.IPv6ProtocolNumber, header.IPv4ProtocolNumber, @@ -1267,7 +1141,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { } nicID := addr.NIC - if len(addr.Addr) != 0 && !isBroadcastOrMulticast(addr.Addr) { + if len(addr.Addr) != 0 && !e.isBroadcastOrMulticast(addr.NIC, netProto, addr.Addr) { // A local unicast address was specified, verify that it's valid. nicID = e.stack.CheckLocalAddress(addr.NIC, netProto, addr.Addr) if nicID == 0 { @@ -1290,7 +1164,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { e.effectiveNetProtos = netProtos // Mark endpoint as bound. - e.state = StateBound + e.setEndpointState(StateBound) e.rcvMu.Lock() e.rcvReady = true @@ -1322,7 +1196,7 @@ func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { defer e.mu.RUnlock() addr := e.ID.LocalAddress - if e.state == StateConnected { + if e.EndpointState() == StateConnected { addr = e.route.LocalAddress } @@ -1338,7 +1212,7 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { e.mu.RLock() defer e.mu.RUnlock() - if e.state != StateConnected { + if e.EndpointState() != StateConnected { return tcpip.FullAddress{}, tcpip.ErrNotConnected } @@ -1470,25 +1344,20 @@ func (e *endpoint) HandlePacket(id stack.TransportEndpointID, pkt *stack.PacketB // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) { if typ == stack.ControlPortUnreachable { - e.mu.RLock() - if e.state == StateConnected { + if e.EndpointState() == StateConnected { e.lastErrorMu.Lock() e.lastError = tcpip.ErrConnectionRefused e.lastErrorMu.Unlock() - e.mu.RUnlock() e.waiterQueue.Notify(waiter.EventErr) return } - e.mu.RUnlock() } } // State implements tcpip.Endpoint.State. func (e *endpoint) State() uint32 { - e.mu.Lock() - defer e.mu.Unlock() - return uint32(e.state) + return uint32(e.EndpointState()) } // Info returns a copy of the endpoint info. @@ -1508,14 +1377,16 @@ func (e *endpoint) Stats() tcpip.EndpointStats { // Wait implements tcpip.Endpoint.Wait. func (*endpoint) Wait() {} -func isBroadcastOrMulticast(a tcpip.Address) bool { - return a == header.IPv4Broadcast || header.IsV4MulticastAddress(a) || header.IsV6MulticastAddress(a) +func (e *endpoint) isBroadcastOrMulticast(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, addr tcpip.Address) bool { + return addr == header.IPv4Broadcast || header.IsV4MulticastAddress(addr) || header.IsV6MulticastAddress(addr) || e.stack.IsSubnetBroadcast(nicID, netProto, addr) } +// SetOwner implements tcpip.Endpoint.SetOwner. func (e *endpoint) SetOwner(owner tcpip.PacketOwner) { e.owner = owner } +// SocketOptions implements tcpip.Endpoint.SocketOptions. func (e *endpoint) SocketOptions() *tcpip.SocketOptions { return &e.ops } diff --git a/pkg/tcpip/transport/udp/endpoint_state.go b/pkg/tcpip/transport/udp/endpoint_state.go index 858c99a45..13b72dc88 100644 --- a/pkg/tcpip/transport/udp/endpoint_state.go +++ b/pkg/tcpip/transport/udp/endpoint_state.go @@ -98,7 +98,8 @@ func (e *endpoint) Resume(s *stack.Stack) { } } - if e.state != StateBound && e.state != StateConnected { + state := e.EndpointState() + if state != StateBound && state != StateConnected { return } @@ -113,12 +114,12 @@ func (e *endpoint) Resume(s *stack.Stack) { } var err *tcpip.Error - if e.state == StateConnected { - e.route, err = e.stack.FindRoute(e.RegisterNICID, e.ID.LocalAddress, e.ID.RemoteAddress, netProto, e.multicastLoop) + if state == StateConnected { + e.route, err = e.stack.FindRoute(e.RegisterNICID, e.ID.LocalAddress, e.ID.RemoteAddress, netProto, e.ops.GetMulticastLoop()) if err != nil { panic(err) } - } else if len(e.ID.LocalAddress) != 0 && !isBroadcastOrMulticast(e.ID.LocalAddress) { // stateBound + } else if len(e.ID.LocalAddress) != 0 && !e.isBroadcastOrMulticast(e.RegisterNICID, netProto, e.ID.LocalAddress) { // stateBound // A local unicast address is specified, verify that it's valid. if e.stack.CheckLocalAddress(e.RegisterNICID, netProto, e.ID.LocalAddress) == 0 { panic(tcpip.ErrBadLocalAddress) diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index 764ad0857..e384f52dd 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -32,6 +32,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/icmp" "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/pkg/waiter" ) @@ -54,6 +55,7 @@ const ( stackPort = 1234 testAddr = "\x0a\x00\x00\x02" testPort = 4096 + invalidPort = 8192 multicastAddr = "\xe8\x2b\xd3\xea" multicastV6Addr = "\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" broadcastAddr = header.IPv4Broadcast @@ -295,7 +297,8 @@ func newDualTestContext(t *testing.T, mtu uint32) *testContext { t.Helper() return newDualTestContextWithOptions(t, mtu, stack.Options{ NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, - TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, icmp.NewProtocol6, icmp.NewProtocol4}, + HandleLocal: true, }) } @@ -360,9 +363,7 @@ func (c *testContext) createEndpointForFlow(flow testFlow) { c.createEndpoint(flow.sockProto()) if flow.isV6Only() { - if err := c.ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - c.t.Fatalf("SetSockOptBool failed: %s", err) - } + c.ep.SocketOptions().SetV6Only(true) } else if flow.isBroadcast() { c.ep.SocketOptions().SetBroadcast(true) } @@ -972,7 +973,7 @@ func testFailingWrite(c *testContext, flow testFlow, wantErr *tcpip.Error) { // provided. func testWrite(c *testContext, flow testFlow, checkers ...checker.NetworkChecker) uint16 { c.t.Helper() - return testWriteInternal(c, flow, true, checkers...) + return testWriteAndVerifyInternal(c, flow, true, checkers...) } // testWriteWithoutDestination sends a packet of the given test flow from the @@ -981,10 +982,10 @@ func testWrite(c *testContext, flow testFlow, checkers ...checker.NetworkChecker // checker functions provided. func testWriteWithoutDestination(c *testContext, flow testFlow, checkers ...checker.NetworkChecker) uint16 { c.t.Helper() - return testWriteInternal(c, flow, false, checkers...) + return testWriteAndVerifyInternal(c, flow, false, checkers...) } -func testWriteInternal(c *testContext, flow testFlow, setDest bool, checkers ...checker.NetworkChecker) uint16 { +func testWriteNoVerify(c *testContext, flow testFlow, setDest bool) buffer.View { c.t.Helper() // Take a snapshot of the stats to validate them at the end of the test. epstats := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() @@ -1006,6 +1007,12 @@ func testWriteInternal(c *testContext, flow testFlow, setDest bool, checkers ... c.t.Fatalf("Bad number of bytes written: got %v, want %v", n, len(payload)) } c.checkEndpointWriteStats(1, epstats, err) + return payload +} + +func testWriteAndVerifyInternal(c *testContext, flow testFlow, setDest bool, checkers ...checker.NetworkChecker) uint16 { + c.t.Helper() + payload := testWriteNoVerify(c, flow, setDest) // Received the packet and check the payload. b := c.getPacketAndVerify(flow, checkers...) var udp header.UDP @@ -1150,6 +1157,39 @@ func TestV4WriteOnConnected(t *testing.T) { testWriteWithoutDestination(c, unicastV4) } +func TestWriteOnConnectedInvalidPort(t *testing.T) { + protocols := map[string]tcpip.NetworkProtocolNumber{ + "ipv4": ipv4.ProtocolNumber, + "ipv6": ipv6.ProtocolNumber, + } + for name, pn := range protocols { + t.Run(name, func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpoint(pn) + if err := c.ep.Connect(tcpip.FullAddress{Addr: stackAddr, Port: invalidPort}); err != nil { + c.t.Fatalf("Connect failed: %s", err) + } + writeOpts := tcpip.WriteOptions{ + To: &tcpip.FullAddress{Addr: stackAddr, Port: invalidPort}, + } + payload := buffer.View(newPayload()) + n, _, err := c.ep.Write(tcpip.SlicePayload(payload), writeOpts) + if err != nil { + c.t.Fatalf("c.ep.Write(...) = %+s, want nil", err) + } + if got, want := n, int64(len(payload)); got != want { + c.t.Fatalf("c.ep.Write(...) wrote %d bytes, want %d bytes", got, want) + } + + if err := c.ep.LastError(); err != tcpip.ErrConnectionRefused { + c.t.Fatalf("expected c.ep.LastError() == ErrConnectionRefused, got: %+v", err) + } + }) + } +} + // TestWriteOnBoundToV4Multicast checks that we can send packets out of a socket // that is bound to a V4 multicast address. func TestWriteOnBoundToV4Multicast(t *testing.T) { @@ -1372,9 +1412,7 @@ func TestReadIPPacketInfo(t *testing.T) { } } - if err := c.ep.SetSockOptBool(tcpip.ReceiveIPPacketInfoOption, true); err != nil { - t.Fatalf("c.ep.SetSockOptBool(tcpip.ReceiveIPPacketInfoOption, true): %s", err) - } + c.ep.SocketOptions().SetReceivePacketInfo(true) testRead(c, test.flow, checker.ReceiveIPPacketInfo(tcpip.IPPacketInfo{ NIC: 1, @@ -1412,16 +1450,12 @@ func TestNoChecksum(t *testing.T) { c.createEndpointForFlow(flow) // Disable the checksum generation. - if err := c.ep.SetSockOptBool(tcpip.NoChecksumOption, true); err != nil { - t.Fatalf("SetSockOptBool failed: %s", err) - } + c.ep.SocketOptions().SetNoChecksum(true) // This option is effective on IPv4 only. testWrite(c, flow, checker.UDP(checker.NoChecksum(flow.isV4()))) // Enable the checksum generation. - if err := c.ep.SetSockOptBool(tcpip.NoChecksumOption, false); err != nil { - t.Fatalf("SetSockOptBool failed: %s", err) - } + c.ep.SocketOptions().SetNoChecksum(false) testWrite(c, flow, checker.UDP(checker.NoChecksum(false))) }) } @@ -1591,13 +1625,15 @@ func TestSetTClass(t *testing.T) { } func TestReceiveTosTClass(t *testing.T) { + const RcvTOSOpt = "ReceiveTosOption" + const RcvTClassOpt = "ReceiveTClassOption" + testCases := []struct { - name string - getReceiveOption tcpip.SockOptBool - tests []testFlow + name string + tests []testFlow }{ - {"ReceiveTosOption", tcpip.ReceiveTOSOption, []testFlow{unicastV4, broadcast}}, - {"ReceiveTClassOption", tcpip.ReceiveTClassOption, []testFlow{unicastV4in6, unicastV6, unicastV6Only, broadcastIn6}}, + {RcvTOSOpt, []testFlow{unicastV4, broadcast}}, + {RcvTClassOpt, []testFlow{unicastV4in6, unicastV6, unicastV6Only, broadcastIn6}}, } for _, testCase := range testCases { for _, flow := range testCase.tests { @@ -1606,29 +1642,32 @@ func TestReceiveTosTClass(t *testing.T) { defer c.cleanup() c.createEndpointForFlow(flow) - option := testCase.getReceiveOption name := testCase.name - // Verify that setting and reading the option works. - v, err := c.ep.GetSockOptBool(option) - if err != nil { - c.t.Errorf("GetSockOptBool(%s) failed: %s", name, err) + var optionGetter func() bool + var optionSetter func(bool) + switch name { + case RcvTOSOpt: + optionGetter = c.ep.SocketOptions().GetReceiveTOS + optionSetter = c.ep.SocketOptions().SetReceiveTOS + case RcvTClassOpt: + optionGetter = c.ep.SocketOptions().GetReceiveTClass + optionSetter = c.ep.SocketOptions().SetReceiveTClass + default: + t.Fatalf("unkown test variant: %s", name) } + + // Verify that setting and reading the option works. + v := optionGetter() // Test for expected default value. if v != false { c.t.Errorf("got GetSockOptBool(%s) = %t, want = %t", name, v, false) } want := true - if err := c.ep.SetSockOptBool(option, want); err != nil { - c.t.Fatalf("SetSockOptBool(%s, %t) failed: %s", name, want, err) - } - - got, err := c.ep.GetSockOptBool(option) - if err != nil { - c.t.Errorf("GetSockOptBool(%s) failed: %s", name, err) - } + optionSetter(want) + got := optionGetter() if got != want { c.t.Errorf("got GetSockOptBool(%s) = %t, want = %t", name, got, want) } @@ -1638,10 +1677,10 @@ func TestReceiveTosTClass(t *testing.T) { if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { c.t.Fatalf("Bind failed: %s", err) } - switch option { - case tcpip.ReceiveTClassOption: + switch name { + case RcvTClassOpt: testRead(c, flow, checker.ReceiveTClass(testTOS)) - case tcpip.ReceiveTOSOption: + case RcvTOSOpt: testRead(c, flow, checker.ReceiveTOS(testTOS)) default: t.Fatalf("unknown test variant: %s", name) diff --git a/pkg/test/criutil/criutil.go b/pkg/test/criutil/criutil.go index 70945f234..e41769017 100644 --- a/pkg/test/criutil/criutil.go +++ b/pkg/test/criutil/criutil.go @@ -54,14 +54,20 @@ func ResolvePath(executable string) string { } } + // Favor /usr/local/bin, if it exists. + localBin := fmt.Sprintf("/usr/local/bin/%s", executable) + if _, err := os.Stat(localBin); err == nil { + return localBin + } + // Try to find via the path. - guess, err := exec.LookPath(executable) + guess, _ := exec.LookPath(executable) if err == nil { return guess } - // Return a default path. - return fmt.Sprintf("/usr/local/bin/%s", executable) + // Return a bare path; this generates a suitable error. + return executable } // NewCrictl returns a Crictl configured with a timeout and an endpoint over diff --git a/pkg/test/dockerutil/container.go b/pkg/test/dockerutil/container.go index 64d17f661..2bf0a22ff 100644 --- a/pkg/test/dockerutil/container.go +++ b/pkg/test/dockerutil/container.go @@ -17,6 +17,7 @@ package dockerutil import ( "bytes" "context" + "errors" "fmt" "io/ioutil" "net" @@ -351,6 +352,9 @@ func (c *Container) SandboxPid(ctx context.Context) (int, error) { return resp.ContainerJSONBase.State.Pid, nil } +// ErrNoIP indicates that no IP address is available. +var ErrNoIP = errors.New("no IP available") + // FindIP returns the IP address of the container. func (c *Container) FindIP(ctx context.Context, ipv6 bool) (net.IP, error) { resp, err := c.client.ContainerInspect(ctx, c.id) @@ -365,7 +369,7 @@ func (c *Container) FindIP(ctx context.Context, ipv6 bool) (net.IP, error) { ip = net.ParseIP(resp.NetworkSettings.DefaultNetworkSettings.IPAddress) } if ip == nil { - return net.IP{}, fmt.Errorf("invalid IP: %q", ip) + return net.IP{}, ErrNoIP } return ip, nil } diff --git a/pkg/test/dockerutil/exec.go b/pkg/test/dockerutil/exec.go index 4c739c9e9..bf968acec 100644 --- a/pkg/test/dockerutil/exec.go +++ b/pkg/test/dockerutil/exec.go @@ -77,11 +77,6 @@ func (c *Container) doExec(ctx context.Context, r ExecOpts, args []string) (Proc return Process{}, fmt.Errorf("exec attach failed with err: %v", err) } - if err := c.client.ContainerExecStart(ctx, resp.ID, types.ExecStartCheck{}); err != nil { - hijack.Close() - return Process{}, fmt.Errorf("exec start failed with err: %v", err) - } - return Process{ container: c, execid: resp.ID, diff --git a/pkg/test/testutil/testutil.go b/pkg/test/testutil/testutil.go index 49ab87c58..fdd416b5e 100644 --- a/pkg/test/testutil/testutil.go +++ b/pkg/test/testutil/testutil.go @@ -36,7 +36,6 @@ import ( "path/filepath" "strconv" "strings" - "sync/atomic" "syscall" "testing" "time" @@ -49,7 +48,10 @@ import ( ) var ( - checkpoint = flag.Bool("checkpoint", true, "control checkpoint/restore support") + checkpoint = flag.Bool("checkpoint", true, "control checkpoint/restore support") + partition = flag.Int("partition", 1, "partition number, this is 1-indexed") + totalPartitions = flag.Int("total_partitions", 1, "total number of partitions") + isRunningWithHostNet = flag.Bool("hostnet", false, "whether test is running with hostnet") ) // IsCheckpointSupported returns the relevant command line flag. @@ -57,6 +59,11 @@ func IsCheckpointSupported() bool { return *checkpoint } +// IsRunningWithHostNet returns the relevant command line flag. +func IsRunningWithHostNet() bool { + return *isRunningWithHostNet +} + // ImageByName mangles the image name used locally. This depends on the image // build infrastructure in images/ and tools/vm. func ImageByName(name string) string { @@ -249,14 +256,25 @@ func writeSpec(dir string, spec *specs.Spec) error { // idRandomSrc is a pseudo random generator used to in RandomID. var idRandomSrc = rand.New(rand.NewSource(time.Now().UnixNano())) +// idRandomSrcMtx is the mutex protecting idRandomSrc.Read from being used +// concurrently in differnt goroutines. +var idRandomSrcMtx sync.Mutex + // RandomID returns 20 random bytes following the given prefix. func RandomID(prefix string) string { // Read 20 random bytes. b := make([]byte, 20) + // Rand.Read is not safe for concurrent use. Packetimpact tests can be run in + // parallel now, so we have to protect the Read with a mutex. Otherwise we'll + // run into name conflicts. + // https://golang.org/pkg/math/rand/#Rand.Read + idRandomSrcMtx.Lock() // "[Read] always returns len(p) and a nil error." --godoc if _, err := idRandomSrc.Read(b); err != nil { + idRandomSrcMtx.Unlock() panic("rand.Read failed: " + err.Error()) } + idRandomSrcMtx.Unlock() if prefix != "" { prefix = prefix + "-" } @@ -417,33 +435,35 @@ func StartReaper() func() { // WaitUntilRead reads from the given reader until the wanted string is found // or until timeout. -func WaitUntilRead(r io.Reader, want string, split bufio.SplitFunc, timeout time.Duration) error { +func WaitUntilRead(r io.Reader, want string, timeout time.Duration) error { sc := bufio.NewScanner(r) - if split != nil { - sc.Split(split) - } // done must be accessed atomically. A value greater than 0 indicates // that the read loop can exit. - var done uint32 - doneCh := make(chan struct{}) + doneCh := make(chan bool) + defer close(doneCh) go func() { for sc.Scan() { t := sc.Text() if strings.Contains(t, want) { - atomic.StoreUint32(&done, 1) - close(doneCh) - break + doneCh <- true + return } - if atomic.LoadUint32(&done) > 0 { - break + select { + case <-doneCh: + return + default: } } + doneCh <- false }() + select { case <-time.After(timeout): - atomic.StoreUint32(&done, 1) return fmt.Errorf("timeout waiting to read %q", want) - case <-doneCh: + case res := <-doneCh: + if !res { + return fmt.Errorf("reader closed while waiting to read %q", want) + } return nil } } @@ -509,7 +529,8 @@ func TouchShardStatusFile() error { } // TestIndicesForShard returns indices for this test shard based on the -// TEST_SHARD_INDEX and TEST_TOTAL_SHARDS environment vars. +// TEST_SHARD_INDEX and TEST_TOTAL_SHARDS environment vars, as well as +// the passed partition flags. // // If either of the env vars are not present, then the function will return all // tests. If there are more shards than there are tests, then the returned list @@ -534,6 +555,11 @@ func TestIndicesForShard(numTests int) ([]int, error) { } } + // Combine with the partitions. + partitionSize := shardTotal + shardTotal = (*totalPartitions) * shardTotal + shardIndex = partitionSize*(*partition-1) + shardIndex + // Calculate! var indices []int numBlocks := int(math.Ceil(float64(numTests) / float64(shardTotal))) diff --git a/pkg/waiter/waiter.go b/pkg/waiter/waiter.go index 08519d986..83d4f893a 100644 --- a/pkg/waiter/waiter.go +++ b/pkg/waiter/waiter.go @@ -119,7 +119,10 @@ type EntryCallback interface { // The callback is supposed to perform minimal work, and cannot call // any method on the queue itself because it will be locked while the // callback is running. - Callback(e *Entry) + // + // The mask indicates the events that occurred and that the entry is + // interested in. + Callback(e *Entry, mask EventMask) } // Entry represents a waiter that can be add to the a wait queue. It can @@ -140,7 +143,7 @@ type channelCallback struct { } // Callback implements EntryCallback.Callback. -func (c *channelCallback) Callback(*Entry) { +func (c *channelCallback) Callback(*Entry, EventMask) { select { case c.ch <- struct{}{}: default: @@ -193,8 +196,8 @@ func (q *Queue) EventUnregister(e *Entry) { func (q *Queue) Notify(mask EventMask) { q.mu.RLock() for e := q.list.Front(); e != nil; e = e.Next() { - if mask&e.mask != 0 { - e.Callback.Callback(e) + if m := mask & e.mask; m != 0 { + e.Callback.Callback(e, m) } } q.mu.RUnlock() diff --git a/pkg/waiter/waiter_test.go b/pkg/waiter/waiter_test.go index c1b94a4f3..6928f28b4 100644 --- a/pkg/waiter/waiter_test.go +++ b/pkg/waiter/waiter_test.go @@ -20,12 +20,12 @@ import ( ) type callbackStub struct { - f func(e *Entry) + f func(e *Entry, m EventMask) } // Callback implements EntryCallback.Callback. -func (c *callbackStub) Callback(e *Entry) { - c.f(e) +func (c *callbackStub) Callback(e *Entry, m EventMask) { + c.f(e, m) } func TestEmptyQueue(t *testing.T) { @@ -36,7 +36,7 @@ func TestEmptyQueue(t *testing.T) { // Register then unregister a waiter, then notify the queue. cnt := 0 - e := Entry{Callback: &callbackStub{func(*Entry) { cnt++ }}} + e := Entry{Callback: &callbackStub{func(*Entry, EventMask) { cnt++ }}} q.EventRegister(&e, EventIn) q.EventUnregister(&e) q.Notify(EventIn) @@ -49,7 +49,7 @@ func TestMask(t *testing.T) { // Register a waiter. var q Queue var cnt int - e := Entry{Callback: &callbackStub{func(*Entry) { cnt++ }}} + e := Entry{Callback: &callbackStub{func(*Entry, EventMask) { cnt++ }}} q.EventRegister(&e, EventIn|EventErr) // Notify with an overlapping mask. @@ -101,11 +101,14 @@ func TestConcurrentRegistration(t *testing.T) { for i := 0; i < concurrency; i++ { go func() { var e Entry - e.Callback = &callbackStub{func(entry *Entry) { + e.Callback = &callbackStub{func(entry *Entry, mask EventMask) { cnt++ if entry != &e { t.Errorf("entry = %p, want %p", entry, &e) } + if mask != EventIn { + t.Errorf("mask = %#x want %#x", mask, EventIn) + } }} // Wait for notification, then register. @@ -158,11 +161,14 @@ func TestConcurrentNotification(t *testing.T) { // Register waiters. for i := 0; i < waiterCount; i++ { var e Entry - e.Callback = &callbackStub{func(entry *Entry) { + e.Callback = &callbackStub{func(entry *Entry, mask EventMask) { atomic.AddInt32(&cnt, 1) if entry != &e { t.Errorf("entry = %p, want %p", entry, &e) } + if mask != EventIn { + t.Errorf("mask = %#x want %#x", mask, EventIn) + } }} q.EventRegister(&e, EventIn|EventErr) |