diff options
Diffstat (limited to 'pkg')
327 files changed, 14503 insertions, 6409 deletions
diff --git a/pkg/abi/linux/BUILD b/pkg/abi/linux/BUILD index cdcaa8c73..4a26e28de 100644 --- a/pkg/abi/linux/BUILD +++ b/pkg/abi/linux/BUILD @@ -38,6 +38,7 @@ go_library( "ipc.go", "limits.go", "linux.go", + "membarrier.go", "mm.go", "netdevice.go", "netfilter.go", diff --git a/pkg/abi/linux/ioctl.go b/pkg/abi/linux/ioctl.go index 3356a2b4a..7df02dd6d 100644 --- a/pkg/abi/linux/ioctl.go +++ b/pkg/abi/linux/ioctl.go @@ -115,15 +115,33 @@ const ( // Constants from uapi/linux/fs.h. const ( - FS_IOC_GETFLAGS = 2147771905 + FS_IOC_GETFLAGS = 2148034049 FS_VERITY_FL = 1048576 ) // Constants from uapi/linux/fsverity.h. const ( - FS_IOC_ENABLE_VERITY = 1082156677 + FS_IOC_ENABLE_VERITY = 1082156677 + FS_IOC_MEASURE_VERITY = 3221513862 ) +// DigestMetadata is a helper struct for VerityDigest. +// +// +marshal +type DigestMetadata struct { + DigestAlgorithm uint16 + DigestSize uint16 +} + +// SizeOfDigestMetadata is the size of struct DigestMetadata. +const SizeOfDigestMetadata = 4 + +// VerityDigest is struct from uapi/linux/fsverity.h. +type VerityDigest struct { + Metadata DigestMetadata + Digest []byte +} + // IOC outputs the result of _IOC macro in asm-generic/ioctl.h. func IOC(dir, typ, nr, size uint32) uint32 { return uint32(dir)<<_IOC_DIRSHIFT | typ<<_IOC_TYPESHIFT | nr<<_IOC_NRSHIFT | size<<_IOC_SIZESHIFT diff --git a/pkg/abi/linux/membarrier.go b/pkg/abi/linux/membarrier.go new file mode 100644 index 000000000..4f6021a1d --- /dev/null +++ b/pkg/abi/linux/membarrier.go @@ -0,0 +1,34 @@ +// 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 linux + +// membarrier(2) commands, from include/uapi/linux/membarrier.h. +const ( + MEMBARRIER_CMD_QUERY = 0 + MEMBARRIER_CMD_GLOBAL = (1 << 0) + MEMBARRIER_CMD_GLOBAL_EXPEDITED = (1 << 1) + MEMBARRIER_CMD_REGISTER_GLOBAL_EXPEDITED = (1 << 2) + MEMBARRIER_CMD_PRIVATE_EXPEDITED = (1 << 3) + MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED = (1 << 4) + MEMBARRIER_CMD_PRIVATE_EXPEDITED_SYNC_CORE = (1 << 5) + MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED_SYNC_CORE = (1 << 6) + MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ = (1 << 7) + MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED_RSEQ = (1 << 8) +) + +// membarrier(2) flags, from include/uapi/linux/membarrier.h. +const ( + MEMBARRIER_CMD_FLAG_CPU = (1 << 0) +) diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 1c5b34711..b521144d9 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -265,6 +265,18 @@ type KernelXTEntryMatch struct { Data []byte } +// XTGetRevision corresponds to xt_get_revision in +// include/uapi/linux/netfilter/x_tables.h +// +// +marshal +type XTGetRevision struct { + Name ExtensionName + Revision uint8 +} + +// SizeOfXTGetRevision is the size of an XTGetRevision. +const SizeOfXTGetRevision = 30 + // XTEntryTarget holds a target for a rule. For example, it can specify that // packets matching the rule should DROP, ACCEPT, or use an extension target. // iptables-extension(8) has a list of possible targets. @@ -285,6 +297,13 @@ type XTEntryTarget struct { // SizeOfXTEntryTarget is the size of an XTEntryTarget. const SizeOfXTEntryTarget = 32 +// KernelXTEntryTarget is identical to XTEntryTarget, but contains a +// variable-length Data field. +type KernelXTEntryTarget struct { + XTEntryTarget + Data []byte +} + // XTStandardTarget is a built-in target, one of ACCEPT, DROP, JUMP, QUEUE, // RETURN, or jump. It corresponds to struct xt_standard_target in // include/uapi/linux/netfilter/x_tables.h. @@ -510,6 +529,8 @@ type IPTReplace struct { const SizeOfIPTReplace = 96 // ExtensionName holds the name of a netfilter extension. +// +// +marshal type ExtensionName [XT_EXTENSION_MAXNAMELEN]byte // String implements fmt.Stringer. diff --git a/pkg/abi/linux/netfilter_ipv6.go b/pkg/abi/linux/netfilter_ipv6.go index a137940b6..6d31eb5e3 100644 --- a/pkg/abi/linux/netfilter_ipv6.go +++ b/pkg/abi/linux/netfilter_ipv6.go @@ -321,3 +321,16 @@ const ( // Enable all flags. IP6T_INV_MASK = 0x7F ) + +// NFNATRange corresponds to struct nf_nat_range in +// include/uapi/linux/netfilter/nf_nat.h. +type NFNATRange struct { + Flags uint32 + MinAddr Inet6Addr + MaxAddr Inet6Addr + MinProto uint16 // Network byte order. + MaxProto uint16 // Network byte order. +} + +// SizeOfNFNATRange is the size of NFNATRange. +const SizeOfNFNATRange = 40 diff --git a/pkg/abi/linux/seccomp.go b/pkg/abi/linux/seccomp.go index b07cafe12..5be3f10f9 100644 --- a/pkg/abi/linux/seccomp.go +++ b/pkg/abi/linux/seccomp.go @@ -83,3 +83,22 @@ type SockFprog struct { pad [6]byte Filter *BPFInstruction } + +// SeccompData is equivalent to struct seccomp_data, which contains the data +// passed to seccomp-bpf filters. +// +// +marshal +type SeccompData struct { + // Nr is the system call number. + Nr int32 + + // Arch is an AUDIT_ARCH_* value indicating the system call convention. + Arch uint32 + + // InstructionPointer is the value of the instruction pointer at the time + // of the system call. + InstructionPointer uint64 + + // Args contains the first 6 system call arguments. + Args [6]uint64 +} diff --git a/pkg/abi/linux/signalfd.go b/pkg/abi/linux/signalfd.go index 85fad9956..468c6a387 100644 --- a/pkg/abi/linux/signalfd.go +++ b/pkg/abi/linux/signalfd.go @@ -23,6 +23,8 @@ const ( ) // SignalfdSiginfo is the siginfo encoding for signalfds. +// +// +marshal type SignalfdSiginfo struct { Signo uint32 Errno int32 @@ -41,5 +43,5 @@ type SignalfdSiginfo struct { STime uint64 Addr uint64 AddrLSB uint16 - _ [48]uint8 + _ [48]uint8 `marshal:"unaligned"` } diff --git a/pkg/buffer/BUILD b/pkg/buffer/BUILD index b03d46d18..1186f788e 100644 --- a/pkg/buffer/BUILD +++ b/pkg/buffer/BUILD @@ -20,6 +20,7 @@ go_library( srcs = [ "buffer.go", "buffer_list.go", + "pool.go", "safemem.go", "view.go", "view_unsafe.go", @@ -37,9 +38,13 @@ go_test( name = "buffer_test", size = "small", srcs = [ + "pool_test.go", "safemem_test.go", "view_test.go", ], library = ":buffer", - deps = ["//pkg/safemem"], + deps = [ + "//pkg/safemem", + "//pkg/state", + ], ) diff --git a/pkg/buffer/buffer.go b/pkg/buffer/buffer.go index c6d089fd9..311808ae9 100644 --- a/pkg/buffer/buffer.go +++ b/pkg/buffer/buffer.go @@ -14,36 +14,26 @@ // Package buffer provides the implementation of a buffer view. // -// A view is an flexible buffer, backed by a pool, supporting the safecopy -// operations natively as well as the ability to grow via either prepend or -// append, as well as shrink. +// A view is an flexible buffer, supporting the safecopy operations natively as +// well as the ability to grow via either prepend or append, as well as shrink. package buffer -import ( - "sync" -) - -const bufferSize = 8144 // See below. - // buffer encapsulates a queueable byte buffer. // -// Note that the total size is slightly less than two pages. This is done -// intentionally to ensure that the buffer object aligns with runtime -// internals. We have no hard size or alignment requirements. This two page -// size will effectively minimize internal fragmentation, but still have a -// large enough chunk to limit excessive segmentation. -// // +stateify savable type buffer struct { - data [bufferSize]byte + data []byte read int write int bufferEntry } -// reset resets internal data. -// -// This must be called before returning the buffer to the pool. +// init performs in-place initialization for zero value. +func (b *buffer) init(size int) { + b.data = make([]byte, size) +} + +// Reset resets read and write locations, effectively emptying the buffer. func (b *buffer) Reset() { b.read = 0 b.write = 0 @@ -85,10 +75,3 @@ func (b *buffer) WriteMove(n int) { func (b *buffer) WriteSlice() []byte { return b.data[b.write:] } - -// bufferPool is a pool for buffers. -var bufferPool = sync.Pool{ - New: func() interface{} { - return new(buffer) - }, -} diff --git a/pkg/buffer/pool.go b/pkg/buffer/pool.go new file mode 100644 index 000000000..7ad6132ab --- /dev/null +++ b/pkg/buffer/pool.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 buffer + +const ( + // embeddedCount is the number of buffer structures embedded in the pool. It + // is also the number for overflow allocations. + embeddedCount = 8 + + // defaultBufferSize is the default size for each underlying storage buffer. + // + // It is slightly less than two pages. This is done intentionally to ensure + // that the buffer object aligns with runtime internals. This two page size + // will effectively minimize internal fragmentation, but still have a large + // enough chunk to limit excessive segmentation. + defaultBufferSize = 8144 +) + +// pool allocates buffer. +// +// It contains an embedded buffer storage for fast path when the number of +// buffers needed is small. +// +// +stateify savable +type pool struct { + bufferSize int + avail []buffer `state:"nosave"` + embeddedStorage [embeddedCount]buffer `state:"wait"` +} + +// get gets a new buffer from p. +func (p *pool) get() *buffer { + if p.avail == nil { + p.avail = p.embeddedStorage[:] + } + if len(p.avail) == 0 { + p.avail = make([]buffer, embeddedCount) + } + if p.bufferSize <= 0 { + p.bufferSize = defaultBufferSize + } + buf := &p.avail[0] + buf.init(p.bufferSize) + p.avail = p.avail[1:] + return buf +} + +// put releases buf. +func (p *pool) put(buf *buffer) { + // Remove reference to the underlying storage, allowing it to be garbage + // collected. + buf.data = nil +} + +// setBufferSize sets the size of underlying storage buffer for future +// allocations. It can be called at any time. +func (p *pool) setBufferSize(size int) { + p.bufferSize = size +} + +// afterLoad is invoked by stateify. +func (p *pool) afterLoad() { + // S/R does not save subslice into embeddedStorage correctly. Restore + // available portion of embeddedStorage manually. Restore as nil if none used. + for i := len(p.embeddedStorage); i > 0; i-- { + if p.embeddedStorage[i-1].data != nil { + p.avail = p.embeddedStorage[i:] + break + } + } +} diff --git a/pkg/buffer/pool_test.go b/pkg/buffer/pool_test.go new file mode 100644 index 000000000..8584bac89 --- /dev/null +++ b/pkg/buffer/pool_test.go @@ -0,0 +1,51 @@ +// 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 buffer + +import ( + "testing" +) + +func TestGetDefaultBufferSize(t *testing.T) { + var p pool + for i := 0; i < embeddedCount*2; i++ { + buf := p.get() + if got, want := len(buf.data), defaultBufferSize; got != want { + t.Errorf("#%d len(buf.data) = %d, want %d", i, got, want) + } + } +} + +func TestGetCustomBufferSize(t *testing.T) { + const size = 100 + + var p pool + p.setBufferSize(size) + for i := 0; i < embeddedCount*2; i++ { + buf := p.get() + if got, want := len(buf.data), size; got != want { + t.Errorf("#%d len(buf.data) = %d, want %d", i, got, want) + } + } +} + +func TestPut(t *testing.T) { + var p pool + buf := p.get() + p.put(buf) + if buf.data != nil { + t.Errorf("buf.data = %x, want nil", buf.data) + } +} diff --git a/pkg/buffer/safemem.go b/pkg/buffer/safemem.go index b789e56e9..8b42575b4 100644 --- a/pkg/buffer/safemem.go +++ b/pkg/buffer/safemem.go @@ -44,7 +44,7 @@ func (v *View) WriteFromSafememReader(r safemem.Reader, count uint64) (uint64, e // Need at least one buffer. firstBuf := v.data.Back() if firstBuf == nil { - firstBuf = bufferPool.Get().(*buffer) + firstBuf = v.pool.get() v.data.PushBack(firstBuf) } @@ -56,7 +56,7 @@ func (v *View) WriteFromSafememReader(r safemem.Reader, count uint64) (uint64, e count -= l blocks = append(blocks, firstBuf.WriteBlock()) for count > 0 { - emptyBuf := bufferPool.Get().(*buffer) + emptyBuf := v.pool.get() v.data.PushBack(emptyBuf) block := emptyBuf.WriteBlock().TakeFirst64(count) count -= uint64(block.Len()) diff --git a/pkg/buffer/safemem_test.go b/pkg/buffer/safemem_test.go index 47f357e0c..721cc5934 100644 --- a/pkg/buffer/safemem_test.go +++ b/pkg/buffer/safemem_test.go @@ -23,6 +23,8 @@ import ( ) func TestSafemem(t *testing.T) { + const bufferSize = defaultBufferSize + testCases := []struct { name string input string diff --git a/pkg/buffer/view.go b/pkg/buffer/view.go index e6901eadb..00652d675 100644 --- a/pkg/buffer/view.go +++ b/pkg/buffer/view.go @@ -27,6 +27,7 @@ import ( type View struct { data bufferList size int64 + pool pool } // TrimFront removes the first count bytes from the buffer. @@ -81,7 +82,7 @@ func (v *View) advanceRead(count int64) { buf = buf.Next() // Iterate. v.data.Remove(oldBuf) oldBuf.Reset() - bufferPool.Put(oldBuf) + v.pool.put(oldBuf) // Update counts. count -= sz @@ -118,7 +119,7 @@ func (v *View) Truncate(length int64) { // Drop the buffer completely; see above. v.data.Remove(buf) buf.Reset() - bufferPool.Put(buf) + v.pool.put(buf) v.size -= sz } } @@ -137,7 +138,7 @@ func (v *View) Grow(length int64, zero bool) { // Is there some space in the last buffer? if buf == nil || buf.Full() { - buf = bufferPool.Get().(*buffer) + buf = v.pool.get() v.data.PushBack(buf) } @@ -181,7 +182,7 @@ func (v *View) Prepend(data []byte) { for len(data) > 0 { // Do we need an empty buffer? - buf := bufferPool.Get().(*buffer) + buf := v.pool.get() v.data.PushFront(buf) // The buffer is empty; copy last chunk. @@ -211,7 +212,7 @@ func (v *View) Append(data []byte) { // Ensure there's a buffer with space. if buf == nil || buf.Full() { - buf = bufferPool.Get().(*buffer) + buf = v.pool.get() v.data.PushBack(buf) } @@ -297,7 +298,7 @@ func (v *View) WriteFromReader(r io.Reader, count int64) (int64, error) { // Ensure we have an empty buffer. if buf == nil || buf.Full() { - buf = bufferPool.Get().(*buffer) + buf = v.pool.get() v.data.PushBack(buf) } diff --git a/pkg/buffer/view_test.go b/pkg/buffer/view_test.go index 3db1bc6ee..839af0223 100644 --- a/pkg/buffer/view_test.go +++ b/pkg/buffer/view_test.go @@ -16,11 +16,16 @@ package buffer import ( "bytes" + "context" "io" "strings" "testing" + + "gvisor.dev/gvisor/pkg/state" ) +const bufferSize = defaultBufferSize + func fillAppend(v *View, data []byte) { v.Append(data) } @@ -50,6 +55,30 @@ var fillFuncs = map[string]func(*View, []byte){ "writeFromReaderEnd": fillWriteFromReaderEnd, } +func BenchmarkReadAt(b *testing.B) { + b.ReportAllocs() + var v View + v.Append(make([]byte, 100)) + + buf := make([]byte, 10) + for i := 0; i < b.N; i++ { + v.ReadAt(buf, 0) + } +} + +func BenchmarkWriteRead(b *testing.B) { + b.ReportAllocs() + var v View + sz := 1000 + wbuf := make([]byte, sz) + rbuf := bytes.NewBuffer(make([]byte, sz)) + for i := 0; i < b.N; i++ { + v.Append(wbuf) + rbuf.Reset() + v.ReadToWriter(rbuf, int64(sz)) + } +} + func testReadAt(t *testing.T, v *View, offset int64, n int, wantStr string, wantErr error) { t.Helper() d := make([]byte, n) @@ -465,3 +494,51 @@ func TestView(t *testing.T) { } } } + +func doSaveAndLoad(t *testing.T, toSave, toLoad *View) { + t.Helper() + var buf bytes.Buffer + ctx := context.Background() + if _, err := state.Save(ctx, &buf, toSave); err != nil { + t.Fatal("state.Save:", err) + } + if _, err := state.Load(ctx, bytes.NewReader(buf.Bytes()), toLoad); err != nil { + t.Fatal("state.Load:", err) + } +} + +func TestSaveRestoreViewEmpty(t *testing.T) { + var toSave View + var v View + doSaveAndLoad(t, &toSave, &v) + + if got := v.pool.avail; got != nil { + t.Errorf("pool is not in zero state: v.pool.avail = %v, want nil", got) + } + if got := v.Flatten(); len(got) != 0 { + t.Errorf("v.Flatten() = %x, want []", got) + } +} + +func TestSaveRestoreView(t *testing.T) { + // Create data that fits 2.5 slots. + data := bytes.Join([][]byte{ + bytes.Repeat([]byte{1, 2}, defaultBufferSize), + bytes.Repeat([]byte{3}, defaultBufferSize/2), + }, nil) + + var toSave View + toSave.Append(data) + + var v View + doSaveAndLoad(t, &toSave, &v) + + // Next available slot at index 3; 0-2 slot are used. + i := 3 + if got, want := &v.pool.avail[0], &v.pool.embeddedStorage[i]; got != want { + t.Errorf("next available buffer points to %p, want %p (&v.pool.embeddedStorage[%d])", got, want, i) + } + if got := v.Flatten(); !bytes.Equal(got, data) { + t.Errorf("v.Flatten() = %x, want %x", got, data) + } +} diff --git a/pkg/coverage/coverage.go b/pkg/coverage/coverage.go index 6831adcce..a4f4b2c5e 100644 --- a/pkg/coverage/coverage.go +++ b/pkg/coverage/coverage.go @@ -100,12 +100,9 @@ var coveragePool = sync.Pool{ // instrumentation_filter. // // Note that we "consume", i.e. clear, coverdata when this function is run, to -// ensure that each event is only reported once. -// -// TODO(b/160639712): evaluate whether it is ok to reset the global coverage -// data every time this function is run. We could technically have each thread -// store a local snapshot against which we compare the most recent coverdata so -// that separate threads do not affect each other's view of the data. +// ensure that each event is only reported once. Due to the limitations of Go +// coverage tools, we reset the global coverage data every time this function is +// run. func ConsumeCoverageData(w io.Writer) int { once.Do(initCoverageData) @@ -117,23 +114,23 @@ func ConsumeCoverageData(w io.Writer) int { for fileIndex, file := range globalData.files { counters := coverdata.Cover.Counters[file] for index := 0; index < len(counters); index++ { - val := atomic.SwapUint32(&counters[index], 0) - if val != 0 { - // Calculate the synthetic PC. - pc := globalData.syntheticPCs[fileIndex][index] - - usermem.ByteOrder.PutUint64(pcBuffer[:], pc) - n, err := w.Write(pcBuffer[:]) - if err != nil { - if err == io.EOF { - // Simply stop writing if we encounter EOF; it's ok if we attempted to - // write more than we can hold. - return total + n - } - panic(fmt.Sprintf("Internal error writing PCs to kcov area: %v", err)) + if atomic.LoadUint32(&counters[index]) == 0 { + continue + } + // Non-zero coverage data found; consume it and report as a PC. + atomic.StoreUint32(&counters[index], 0) + pc := globalData.syntheticPCs[fileIndex][index] + usermem.ByteOrder.PutUint64(pcBuffer[:], pc) + n, err := w.Write(pcBuffer[:]) + if err != nil { + if err == io.EOF { + // Simply stop writing if we encounter EOF; it's ok if we attempted to + // write more than we can hold. + return total + n } - total += n + panic(fmt.Sprintf("Internal error writing PCs to kcov area: %v", err)) } + total += n } } diff --git a/pkg/marshal/BUILD b/pkg/marshal/BUILD index 4aec98218..aac0161fa 100644 --- a/pkg/marshal/BUILD +++ b/pkg/marshal/BUILD @@ -11,7 +11,5 @@ go_library( visibility = [ "//:sandbox", ], - deps = [ - "//pkg/usermem", - ], + deps = ["//pkg/usermem"], ) diff --git a/pkg/marshal/primitive/BUILD b/pkg/marshal/primitive/BUILD index 06741e6d1..d77a11c79 100644 --- a/pkg/marshal/primitive/BUILD +++ b/pkg/marshal/primitive/BUILD @@ -12,6 +12,7 @@ go_library( "//:sandbox", ], deps = [ + "//pkg/context", "//pkg/marshal", "//pkg/usermem", ], diff --git a/pkg/marshal/primitive/primitive.go b/pkg/marshal/primitive/primitive.go index dfdae5d60..4b342de6b 100644 --- a/pkg/marshal/primitive/primitive.go +++ b/pkg/marshal/primitive/primitive.go @@ -19,6 +19,7 @@ package primitive import ( "io" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/usermem" ) @@ -126,6 +127,46 @@ var _ marshal.Marshallable = (*ByteSlice)(nil) // Below, we define some convenience functions for marshalling primitive types // using the newtypes above, without requiring superfluous casts. +// 8-bit integers + +// CopyInt8In is a convenient wrapper for copying in an int8 from the task's +// memory. +func CopyInt8In(cc marshal.CopyContext, addr usermem.Addr, dst *int8) (int, error) { + var buf Int8 + n, err := buf.CopyIn(cc, addr) + if err != nil { + return n, err + } + *dst = int8(buf) + return n, nil +} + +// CopyInt8Out is a convenient wrapper for copying out an int8 to the task's +// memory. +func CopyInt8Out(cc marshal.CopyContext, addr usermem.Addr, src int8) (int, error) { + srcP := Int8(src) + return srcP.CopyOut(cc, addr) +} + +// CopyUint8In is a convenient wrapper for copying in a uint8 from the task's +// memory. +func CopyUint8In(cc marshal.CopyContext, addr usermem.Addr, dst *uint8) (int, error) { + var buf Uint8 + n, err := buf.CopyIn(cc, addr) + if err != nil { + return n, err + } + *dst = uint8(buf) + return n, nil +} + +// CopyUint8Out is a convenient wrapper for copying out a uint8 to the task's +// memory. +func CopyUint8Out(cc marshal.CopyContext, addr usermem.Addr, src uint8) (int, error) { + srcP := Uint8(src) + return srcP.CopyOut(cc, addr) +} + // 16-bit integers // CopyInt16In is a convenient wrapper for copying in an int16 from the task's @@ -245,3 +286,64 @@ func CopyUint64Out(cc marshal.CopyContext, addr usermem.Addr, src uint64) (int, srcP := Uint64(src) return srcP.CopyOut(cc, addr) } + +// CopyByteSliceIn is a convenient wrapper for copying in a []byte from the +// task's memory. +func CopyByteSliceIn(cc marshal.CopyContext, addr usermem.Addr, dst *[]byte) (int, error) { + var buf ByteSlice + n, err := buf.CopyIn(cc, addr) + if err != nil { + return n, err + } + *dst = []byte(buf) + return n, nil +} + +// CopyByteSliceOut is a convenient wrapper for copying out a []byte to the +// task's memory. +func CopyByteSliceOut(cc marshal.CopyContext, addr usermem.Addr, src []byte) (int, error) { + srcP := ByteSlice(src) + return srcP.CopyOut(cc, addr) +} + +// CopyStringIn is a convenient wrapper for copying in a string from the +// task's memory. +func CopyStringIn(cc marshal.CopyContext, addr usermem.Addr, dst *string) (int, error) { + var buf ByteSlice + n, err := buf.CopyIn(cc, addr) + if err != nil { + return n, err + } + *dst = string(buf) + return n, nil +} + +// CopyStringOut is a convenient wrapper for copying out a string to the task's +// memory. +func CopyStringOut(cc marshal.CopyContext, addr usermem.Addr, src string) (int, error) { + srcP := ByteSlice(src) + return srcP.CopyOut(cc, addr) +} + +// IOCopyContext wraps an object implementing usermem.IO to implement +// marshal.CopyContext. +type IOCopyContext struct { + Ctx context.Context + IO usermem.IO + Opts usermem.IOOpts +} + +// CopyScratchBuffer implements marshal.CopyContext.CopyScratchBuffer. +func (i *IOCopyContext) CopyScratchBuffer(size int) []byte { + return make([]byte, size) +} + +// CopyOutBytes implements marshal.CopyContext.CopyOutBytes. +func (i *IOCopyContext) CopyOutBytes(addr usermem.Addr, b []byte) (int, error) { + return i.IO.CopyOut(i.Ctx, addr, b, i.Opts) +} + +// CopyInBytes implements marshal.CopyContext.CopyInBytes. +func (i *IOCopyContext) CopyInBytes(addr usermem.Addr, b []byte) (int, error) { + return i.IO.CopyIn(i.Ctx, addr, b, i.Opts) +} diff --git a/pkg/merkletree/merkletree.go b/pkg/merkletree/merkletree.go index 4b4f9bd52..ccef83259 100644 --- a/pkg/merkletree/merkletree.go +++ b/pkg/merkletree/merkletree.go @@ -41,7 +41,7 @@ type Layout struct { blockSize int64 // digestSize is the size of a generated hash. digestSize int64 - // levelOffset contains the offset of the begnning of each level in + // levelOffset contains the offset of the beginning of each level in // bytes. The number of levels in the tree is the length of the slice. // The leaf nodes (level 0) contain hashes of blocks of the input data. // Each level N contains hashes of the blocks in level N-1. The highest diff --git a/pkg/refs/refcounter.go b/pkg/refs/refcounter.go index 57d8542b9..699ea8ac3 100644 --- a/pkg/refs/refcounter.go +++ b/pkg/refs/refcounter.go @@ -257,6 +257,8 @@ func (l *LeakMode) Get() interface{} { // String implements flag.Value. func (l *LeakMode) String() string { switch *l { + case UninitializedLeakChecking: + return "uninitialized" case NoLeakChecking: return "disabled" case LeaksLogWarning: @@ -264,7 +266,7 @@ func (l *LeakMode) String() string { case LeaksLogTraces: return "log-traces" } - panic(fmt.Sprintf("invalid ref leak mode %q", *l)) + panic(fmt.Sprintf("invalid ref leak mode %d", *l)) } // leakMode stores the current mode for the reference leak checker. diff --git a/pkg/seccomp/BUILD b/pkg/seccomp/BUILD index bdef7762c..e828894b0 100644 --- a/pkg/seccomp/BUILD +++ b/pkg/seccomp/BUILD @@ -49,7 +49,7 @@ go_test( library = ":seccomp", deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/bpf", + "//pkg/usermem", ], ) diff --git a/pkg/seccomp/seccomp_test.go b/pkg/seccomp/seccomp_test.go index 23f30678d..e1444d18b 100644 --- a/pkg/seccomp/seccomp_test.go +++ b/pkg/seccomp/seccomp_test.go @@ -28,17 +28,10 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/bpf" + "gvisor.dev/gvisor/pkg/usermem" ) -type seccompData struct { - nr uint32 - arch uint32 - instructionPointer uint64 - args [6]uint64 -} - // newVictim makes a victim binary. func newVictim() (string, error) { f, err := ioutil.TempFile("", "victim") @@ -58,9 +51,14 @@ func newVictim() (string, error) { return path, nil } -// asInput converts a seccompData to a bpf.Input. -func (d *seccompData) asInput() bpf.Input { - return bpf.InputBytes{binary.Marshal(nil, binary.LittleEndian, d), binary.LittleEndian} +// dataAsInput converts a linux.SeccompData to a bpf.Input. +func dataAsInput(d *linux.SeccompData) bpf.Input { + buf := make([]byte, d.SizeBytes()) + d.MarshalUnsafe(buf) + return bpf.InputBytes{ + Data: buf, + Order: usermem.ByteOrder, + } } func TestBasic(t *testing.T) { @@ -69,7 +67,7 @@ func TestBasic(t *testing.T) { desc string // data is the input data. - data seccompData + data linux.SeccompData // want is the expected return value of the BPF program. want linux.BPFAction @@ -95,12 +93,12 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "syscall allowed", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "syscall disallowed", - data: seccompData{nr: 2, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 2, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -131,22 +129,22 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "allowed (1a)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x1}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x1}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "allowed (1b)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, { desc: "syscall 1 matched 2nd rule", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, { desc: "no match", - data: seccompData{nr: 0, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 0, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_KILL_THREAD, }, }, @@ -168,42 +166,42 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "allowed (1)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "allowed (3)", - data: seccompData{nr: 3, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 3, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "allowed (5)", - data: seccompData{nr: 5, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 5, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "disallowed (0)", - data: seccompData{nr: 0, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 0, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, { desc: "disallowed (2)", - data: seccompData{nr: 2, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 2, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, { desc: "disallowed (4)", - data: seccompData{nr: 4, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 4, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, { desc: "disallowed (6)", - data: seccompData{nr: 6, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 6, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, { desc: "disallowed (100)", - data: seccompData{nr: 100, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 100, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -223,7 +221,7 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "arch (123)", - data: seccompData{nr: 1, arch: 123}, + data: linux.SeccompData{Nr: 1, Arch: 123}, want: linux.SECCOMP_RET_KILL_THREAD, }, }, @@ -243,7 +241,7 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "action trap", - data: seccompData{nr: 2, arch: LINUX_AUDIT_ARCH}, + data: linux.SeccompData{Nr: 2, Arch: LINUX_AUDIT_ARCH}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -268,12 +266,12 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "allowed", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0xf, 0xf}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0xf, 0xf}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "disallowed", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0xf, 0xe}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0xf, 0xe}}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -300,12 +298,12 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "match first rule", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0xf}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0xf}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "match 2nd rule", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0xe}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0xe}}, want: linux.SECCOMP_RET_ALLOW, }, }, @@ -331,28 +329,28 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "argument allowed (all match)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, - args: [6]uint64{0, math.MaxUint64 - 1, math.MaxUint32}, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, + Args: [6]uint64{0, math.MaxUint64 - 1, math.MaxUint32}, }, want: linux.SECCOMP_RET_ALLOW, }, { desc: "argument disallowed (one mismatch)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, - args: [6]uint64{0, math.MaxUint64, math.MaxUint32}, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, + Args: [6]uint64{0, math.MaxUint64, math.MaxUint32}, }, want: linux.SECCOMP_RET_TRAP, }, { desc: "argument disallowed (multiple mismatch)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, - args: [6]uint64{0, math.MaxUint64, math.MaxUint32 - 1}, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, + Args: [6]uint64{0, math.MaxUint64, math.MaxUint32 - 1}, }, want: linux.SECCOMP_RET_TRAP, }, @@ -379,28 +377,28 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "arg allowed", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, - args: [6]uint64{0, math.MaxUint64, math.MaxUint32 - 1}, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, + Args: [6]uint64{0, math.MaxUint64, math.MaxUint32 - 1}, }, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg disallowed (one equal)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, - args: [6]uint64{0x7aabbccdd, math.MaxUint64, math.MaxUint32 - 1}, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, + Args: [6]uint64{0x7aabbccdd, math.MaxUint64, math.MaxUint32 - 1}, }, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (all equal)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, - args: [6]uint64{0x7aabbccdd, math.MaxUint64 - 1, math.MaxUint32}, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, + Args: [6]uint64{0x7aabbccdd, math.MaxUint64 - 1, math.MaxUint32}, }, want: linux.SECCOMP_RET_TRAP, }, @@ -429,27 +427,27 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "high 32bits greater", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000003_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000003_00000002}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "high 32bits equal, low 32bits greater", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000003}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000003}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "high 32bits equal, low 32bits equal", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000002}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "high 32bits equal, low 32bits less", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000001}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000001}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "high 32bits less", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000001_00000003}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000001_00000003}}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -474,27 +472,27 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "arg allowed", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x10, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x10, 0xffffffff}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg disallowed (first arg equal)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0xf, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0xf, 0xffffffff}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (first arg smaller)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0xffffffff}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (second arg equal)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x10, 0xabcd000d}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x10, 0xabcd000d}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (second arg smaller)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x10, 0xa000ffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x10, 0xa000ffff}}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -522,27 +520,27 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "high 32bits greater", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000003_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000003_00000002}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "high 32bits equal, low 32bits greater", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000003}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000003}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "high 32bits equal, low 32bits equal", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000002}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "high 32bits equal, low 32bits less", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000001}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000001}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "high 32bits less", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000001_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000001_00000002}}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -567,32 +565,32 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "arg allowed (both greater)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x10, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x10, 0xffffffff}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg allowed (first arg equal)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0xf, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0xf, 0xffffffff}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg disallowed (first arg smaller)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0xffffffff}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg allowed (second arg equal)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x10, 0xabcd000d}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x10, 0xabcd000d}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg disallowed (second arg smaller)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x10, 0xa000ffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x10, 0xa000ffff}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (both arg smaller)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0xa000ffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0xa000ffff}}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -620,27 +618,27 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "high 32bits greater", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000003_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000003_00000002}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "high 32bits equal, low 32bits greater", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000003}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000003}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "high 32bits equal, low 32bits equal", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000002}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "high 32bits equal, low 32bits less", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000001}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000001}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "high 32bits less", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000001_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000001_00000002}}, want: linux.SECCOMP_RET_ALLOW, }, }, @@ -665,32 +663,32 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "arg allowed", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0x0}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0x0}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg disallowed (first arg equal)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x1, 0x0}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x1, 0x0}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (first arg greater)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x2, 0x0}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x2, 0x0}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (second arg equal)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0xabcd000d}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0xabcd000d}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (second arg greater)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0xffffffff}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (both arg greater)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x2, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x2, 0xffffffff}}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -718,27 +716,27 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "high 32bits greater", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000003_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000003_00000002}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "high 32bits equal, low 32bits greater", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000003}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000003}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "high 32bits equal, low 32bits equal", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000002}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "high 32bits equal, low 32bits less", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000002_00000001}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000002_00000001}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "high 32bits less", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x00000001_00000002}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x00000001_00000002}}, want: linux.SECCOMP_RET_ALLOW, }, }, @@ -764,32 +762,32 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "arg allowed", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0x0}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0x0}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg allowed (first arg equal)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x1, 0x0}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x1, 0x0}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg disallowed (first arg greater)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x2, 0x0}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x2, 0x0}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg allowed (second arg equal)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0xabcd000d}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0xabcd000d}}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg disallowed (second arg greater)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x0, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x0, 0xffffffff}}, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (both arg greater)", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{0x2, 0xffffffff}}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{0x2, 0xffffffff}}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -816,51 +814,51 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "arg allowed (low order mandatory bit)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, // 00000000 00000000 00000000 00000001 - args: [6]uint64{0x1}, + Args: [6]uint64{0x1}, }, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg allowed (low order optional bit)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, // 00000000 00000000 00000000 00000101 - args: [6]uint64{0x5}, + Args: [6]uint64{0x5}, }, want: linux.SECCOMP_RET_ALLOW, }, { desc: "arg disallowed (lowest order bit not set)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, // 00000000 00000000 00000000 00000010 - args: [6]uint64{0x2}, + Args: [6]uint64{0x2}, }, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (second lowest order bit set)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, // 00000000 00000000 00000000 00000011 - args: [6]uint64{0x3}, + Args: [6]uint64{0x3}, }, want: linux.SECCOMP_RET_TRAP, }, { desc: "arg disallowed (8th bit set)", - data: seccompData{ - nr: 1, - arch: LINUX_AUDIT_ARCH, + data: linux.SeccompData{ + Nr: 1, + Arch: LINUX_AUDIT_ARCH, // 00000000 00000000 00000001 00000000 - args: [6]uint64{0x100}, + Args: [6]uint64{0x100}, }, want: linux.SECCOMP_RET_TRAP, }, @@ -885,12 +883,12 @@ func TestBasic(t *testing.T) { specs: []spec{ { desc: "allowed", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{}, instructionPointer: 0x7aabbccdd}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{}, InstructionPointer: 0x7aabbccdd}, want: linux.SECCOMP_RET_ALLOW, }, { desc: "disallowed", - data: seccompData{nr: 1, arch: LINUX_AUDIT_ARCH, args: [6]uint64{}, instructionPointer: 0x711223344}, + data: linux.SeccompData{Nr: 1, Arch: LINUX_AUDIT_ARCH, Args: [6]uint64{}, InstructionPointer: 0x711223344}, want: linux.SECCOMP_RET_TRAP, }, }, @@ -906,7 +904,7 @@ func TestBasic(t *testing.T) { t.Fatalf("bpf.Compile() got error: %v", err) } for _, spec := range test.specs { - got, err := bpf.Exec(p, spec.data.asInput()) + got, err := bpf.Exec(p, dataAsInput(&spec.data)) if err != nil { t.Fatalf("%s: bpf.Exec() got error: %v", spec.desc, err) } @@ -947,8 +945,8 @@ func TestRandom(t *testing.T) { t.Fatalf("bpf.Compile() got error: %v", err) } for i := uint32(0); i < 200; i++ { - data := seccompData{nr: i, arch: LINUX_AUDIT_ARCH} - got, err := bpf.Exec(p, data.asInput()) + data := linux.SeccompData{Nr: int32(i), Arch: LINUX_AUDIT_ARCH} + got, err := bpf.Exec(p, dataAsInput(&data)) if err != nil { t.Errorf("bpf.Exec() got error: %v, for syscall %d", err, i) continue diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index 99e2b3389..4af4d6e84 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -22,6 +22,7 @@ go_library( "signal_info.go", "signal_stack.go", "stack.go", + "stack_unsafe.go", "syscalls_amd64.go", "syscalls_arm64.go", ], diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go index 0f433ee79..fd73751e7 100644 --- a/pkg/sentry/arch/arch_aarch64.go +++ b/pkg/sentry/arch/arch_aarch64.go @@ -154,6 +154,7 @@ func (s State) Proto() *rpb.Registers { Sp: s.Regs.Sp, Pc: s.Regs.Pc, Pstate: s.Regs.Pstate, + Tls: s.Regs.TPIDR_EL0, } return &rpb.Registers{Arch: &rpb.Registers_Arm64{Arm64: regs}} } @@ -232,6 +233,7 @@ func (s *State) RegisterMap() (map[string]uintptr, error) { "Sp": uintptr(s.Regs.Sp), "Pc": uintptr(s.Regs.Pc), "Pstate": uintptr(s.Regs.Pstate), + "Tls": uintptr(s.Regs.TPIDR_EL0), }, nil } diff --git a/pkg/sentry/arch/registers.proto b/pkg/sentry/arch/registers.proto index 60c027aab..2727ba08a 100644 --- a/pkg/sentry/arch/registers.proto +++ b/pkg/sentry/arch/registers.proto @@ -83,6 +83,7 @@ message ARM64Registers { uint64 sp = 32; uint64 pc = 33; uint64 pstate = 34; + uint64 tls = 35; } message Registers { oneof arch { diff --git a/pkg/sentry/arch/signal_amd64.go b/pkg/sentry/arch/signal_amd64.go index 6fb756f0e..72e07a988 100644 --- a/pkg/sentry/arch/signal_amd64.go +++ b/pkg/sentry/arch/signal_amd64.go @@ -17,17 +17,19 @@ package arch import ( - "encoding/binary" "math" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/usermem" ) // SignalContext64 is equivalent to struct sigcontext, the type passed as the // second argument to signal handlers set by signal(2). +// +// +marshal type SignalContext64 struct { R8 uint64 R9 uint64 @@ -68,6 +70,8 @@ const ( ) // UContext64 is equivalent to ucontext_t on 64-bit x86. +// +// +marshal type UContext64 struct { Flags uint64 Link uint64 @@ -172,12 +176,7 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt // "... the value (%rsp+8) is always a multiple of 16 (...) when // control is transferred to the function entry point." - AMD64 ABI - ucSize := binary.Size(uc) - if ucSize < 0 { - // This can only happen if we've screwed up the definition of - // UContext64. - panic("can't get size of UContext64") - } + ucSize := uc.SizeBytes() // st.Arch.Width() is for the restorer address. sizeof(siginfo) == 128. frameSize := int(st.Arch.Width()) + ucSize + 128 frameBottom := (sp-usermem.Addr(frameSize)) & ^usermem.Addr(15) - 8 @@ -195,18 +194,18 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt info.FixSignalCodeForUser() // Set up the stack frame. - infoAddr, err := st.Push(info) - if err != nil { + if _, err := info.CopyOut(st, StackBottomMagic); err != nil { return err } - ucAddr, err := st.Push(uc) - if err != nil { + infoAddr := st.Bottom + if _, err := uc.CopyOut(st, StackBottomMagic); err != nil { return err } + ucAddr := st.Bottom if act.HasRestorer() { // Push the restorer return address. // Note that this doesn't need to be popped. - if _, err := st.Push(usermem.Addr(act.Restorer)); err != nil { + if _, err := primitive.CopyUint64Out(st, StackBottomMagic, act.Restorer); err != nil { return err } } else { @@ -240,11 +239,11 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt func (c *context64) SignalRestore(st *Stack, rt bool) (linux.SignalSet, SignalStack, error) { // Copy out the stack frame. var uc UContext64 - if _, err := st.Pop(&uc); err != nil { + if _, err := uc.CopyIn(st, StackBottomMagic); err != nil { return 0, SignalStack{}, err } var info SignalInfo - if _, err := st.Pop(&info); err != nil { + if _, err := info.CopyIn(st, StackBottomMagic); err != nil { return 0, SignalStack{}, err } diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go index 642c79dda..7fde5d34e 100644 --- a/pkg/sentry/arch/signal_arm64.go +++ b/pkg/sentry/arch/signal_arm64.go @@ -12,10 +12,11 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build arm64 + package arch import ( - "encoding/binary" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" @@ -25,6 +26,8 @@ import ( // SignalContext64 is equivalent to struct sigcontext, the type passed as the // second argument to signal handlers set by signal(2). +// +// +marshal type SignalContext64 struct { FaultAddr uint64 Regs [31]uint64 @@ -36,6 +39,7 @@ type SignalContext64 struct { Reserved [3568]uint8 } +// +marshal type aarch64Ctx struct { Magic uint32 Size uint32 @@ -43,6 +47,8 @@ type aarch64Ctx struct { // FpsimdContext is equivalent to struct fpsimd_context on arm64 // (arch/arm64/include/uapi/asm/sigcontext.h). +// +// +marshal type FpsimdContext struct { Head aarch64Ctx Fpsr uint32 @@ -51,13 +57,15 @@ type FpsimdContext struct { } // UContext64 is equivalent to ucontext on arm64(arch/arm64/include/uapi/asm/ucontext.h). +// +// +marshal type UContext64 struct { Flags uint64 Link uint64 Stack SignalStack Sigset linux.SignalSet // glibc uses a 1024-bit sigset_t - _pad [(1024 - 64) / 8]byte + _pad [120]byte // (1024 - 64) / 8 = 120 // sigcontext must be aligned to 16-byte _pad2 [8]byte // last for future expansion @@ -94,11 +102,7 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt }, Sigset: sigset, } - - ucSize := binary.Size(uc) - if ucSize < 0 { - panic("can't get size of UContext64") - } + ucSize := uc.SizeBytes() // frameSize = ucSize + sizeof(siginfo). // sizeof(siginfo) == 128. @@ -119,14 +123,14 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt info.FixSignalCodeForUser() // Set up the stack frame. - infoAddr, err := st.Push(info) - if err != nil { + if _, err := info.CopyOut(st, StackBottomMagic); err != nil { return err } - ucAddr, err := st.Push(uc) - if err != nil { + infoAddr := st.Bottom + if _, err := uc.CopyOut(st, StackBottomMagic); err != nil { return err } + ucAddr := st.Bottom // Set up registers. c.Regs.Sp = uint64(st.Bottom) @@ -147,11 +151,11 @@ func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt func (c *context64) SignalRestore(st *Stack, rt bool) (linux.SignalSet, SignalStack, error) { // Copy out the stack frame. var uc UContext64 - if _, err := st.Pop(&uc); err != nil { + if _, err := uc.CopyIn(st, StackBottomMagic); err != nil { return 0, SignalStack{}, err } var info SignalInfo - if _, err := st.Pop(&info); err != nil { + if _, err := info.CopyIn(st, StackBottomMagic); err != nil { return 0, SignalStack{}, err } diff --git a/pkg/sentry/arch/stack.go b/pkg/sentry/arch/stack.go index 1108fa0bd..5f06c751d 100644 --- a/pkg/sentry/arch/stack.go +++ b/pkg/sentry/arch/stack.go @@ -15,14 +15,16 @@ package arch import ( - "encoding/binary" - "fmt" - "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/usermem" ) -// Stack is a simple wrapper around a usermem.IO and an address. +// Stack is a simple wrapper around a usermem.IO and an address. Stack +// implements marshal.CopyContext, and marshallable values can be pushed or +// popped from the stack through the marshal.Marshallable interface. +// +// Stack is not thread-safe. type Stack struct { // Our arch info. // We use this for automatic Native conversion of usermem.Addrs during @@ -34,105 +36,60 @@ type Stack struct { // Our current stack bottom. Bottom usermem.Addr -} -// Push pushes the given values on to the stack. -// -// (This method supports Addrs and treats them as native types.) -func (s *Stack) Push(vals ...interface{}) (usermem.Addr, error) { - for _, v := range vals { - - // We convert some types to well-known serializable quanities. - var norm interface{} - - // For array types, we will automatically add an appropriate - // terminal value. This is done simply to make the interface - // easier to use. - var term interface{} - - switch v.(type) { - case string: - norm = []byte(v.(string)) - term = byte(0) - case []int8, []uint8: - norm = v - term = byte(0) - case []int16, []uint16: - norm = v - term = uint16(0) - case []int32, []uint32: - norm = v - term = uint32(0) - case []int64, []uint64: - norm = v - term = uint64(0) - case []usermem.Addr: - // Special case: simply push recursively. - _, err := s.Push(s.Arch.Native(uintptr(0))) - if err != nil { - return 0, err - } - varr := v.([]usermem.Addr) - for i := len(varr) - 1; i >= 0; i-- { - _, err := s.Push(varr[i]) - if err != nil { - return 0, err - } - } - continue - case usermem.Addr: - norm = s.Arch.Native(uintptr(v.(usermem.Addr))) - default: - norm = v - } + // Scratch buffer used for marshalling to avoid having to repeatedly + // allocate scratch memory. + scratchBuf []byte +} - if term != nil { - _, err := s.Push(term) - if err != nil { - return 0, err - } - } +// scratchBufLen is the default length of Stack.scratchBuf. The +// largest structs the stack regularly serializes are arch.SignalInfo +// and arch.UContext64. We'll set the default size as the larger of +// the two, arch.UContext64. +var scratchBufLen = (*UContext64)(nil).SizeBytes() - c := binary.Size(norm) - if c < 0 { - return 0, fmt.Errorf("bad binary.Size for %T", v) - } - n, err := usermem.CopyObjectOut(context.Background(), s.IO, s.Bottom-usermem.Addr(c), norm, usermem.IOOpts{}) - if err != nil || c != n { - return 0, err - } +// CopyScratchBuffer implements marshal.CopyContext.CopyScratchBuffer. +func (s *Stack) CopyScratchBuffer(size int) []byte { + if len(s.scratchBuf) < size { + s.scratchBuf = make([]byte, size) + } + return s.scratchBuf[:size] +} +// StackBottomMagic is the special address callers must past to all stack +// marshalling operations to cause the src/dst address to be computed based on +// the current end of the stack. +const StackBottomMagic = ^usermem.Addr(0) // usermem.Addr(-1) + +// CopyOutBytes implements marshal.CopyContext.CopyOutBytes. CopyOutBytes +// computes an appropriate address based on the current end of the +// stack. Callers use the sentinel address StackBottomMagic to marshal methods +// to indicate this. +func (s *Stack) CopyOutBytes(sentinel usermem.Addr, b []byte) (int, error) { + if sentinel != StackBottomMagic { + panic("Attempted to copy out to stack with absolute address") + } + c := len(b) + n, err := s.IO.CopyOut(context.Background(), s.Bottom-usermem.Addr(c), b, usermem.IOOpts{}) + if err == nil && n == c { s.Bottom -= usermem.Addr(n) } - - return s.Bottom, nil + return n, err } -// Pop pops the given values off the stack. -// -// (This method supports Addrs and treats them as native types.) -func (s *Stack) Pop(vals ...interface{}) (usermem.Addr, error) { - for _, v := range vals { - - vaddr, isVaddr := v.(*usermem.Addr) - - var n int - var err error - if isVaddr { - value := s.Arch.Native(uintptr(0)) - n, err = usermem.CopyObjectIn(context.Background(), s.IO, s.Bottom, value, usermem.IOOpts{}) - *vaddr = usermem.Addr(s.Arch.Value(value)) - } else { - n, err = usermem.CopyObjectIn(context.Background(), s.IO, s.Bottom, v, usermem.IOOpts{}) - } - if err != nil { - return 0, err - } - +// CopyInBytes implements marshal.CopyContext.CopyInBytes. CopyInBytes computes +// an appropriate address based on the current end of the stack. Callers must +// use the sentinel address StackBottomMagic to marshal methods to indicate +// this. +func (s *Stack) CopyInBytes(sentinel usermem.Addr, b []byte) (int, error) { + if sentinel != StackBottomMagic { + panic("Attempted to copy in from stack with absolute address") + } + n, err := s.IO.CopyIn(context.Background(), s.Bottom, b, usermem.IOOpts{}) + if err == nil { s.Bottom += usermem.Addr(n) } - - return s.Bottom, nil + return n, err } // Align aligns the stack to the given offset. @@ -142,6 +99,22 @@ func (s *Stack) Align(offset int) { } } +// PushNullTerminatedByteSlice writes bs to the stack, followed by an extra null +// byte at the end. On error, the contents of the stack and the bottom cursor +// are undefined. +func (s *Stack) PushNullTerminatedByteSlice(bs []byte) (int, error) { + // Note: Stack grows up, so write the terminal null byte first. + nNull, err := primitive.CopyUint8Out(s, StackBottomMagic, 0) + if err != nil { + return 0, err + } + n, err := primitive.CopyByteSliceOut(s, StackBottomMagic, bs) + if err != nil { + return 0, err + } + return n + nNull, nil +} + // StackLayout describes the location of the arguments and environment on the // stack. type StackLayout struct { @@ -177,11 +150,10 @@ func (s *Stack) Load(args []string, env []string, aux Auxv) (StackLayout, error) l.EnvvEnd = s.Bottom envAddrs := make([]usermem.Addr, len(env)) for i := len(env) - 1; i >= 0; i-- { - addr, err := s.Push(env[i]) - if err != nil { + if _, err := s.PushNullTerminatedByteSlice([]byte(env[i])); err != nil { return StackLayout{}, err } - envAddrs[i] = addr + envAddrs[i] = s.Bottom } l.EnvvStart = s.Bottom @@ -189,11 +161,10 @@ func (s *Stack) Load(args []string, env []string, aux Auxv) (StackLayout, error) l.ArgvEnd = s.Bottom argAddrs := make([]usermem.Addr, len(args)) for i := len(args) - 1; i >= 0; i-- { - addr, err := s.Push(args[i]) - if err != nil { + if _, err := s.PushNullTerminatedByteSlice([]byte(args[i])); err != nil { return StackLayout{}, err } - argAddrs[i] = addr + argAddrs[i] = s.Bottom } l.ArgvStart = s.Bottom @@ -222,26 +193,26 @@ func (s *Stack) Load(args []string, env []string, aux Auxv) (StackLayout, error) auxv = append(auxv, usermem.Addr(a.Key), a.Value) } auxv = append(auxv, usermem.Addr(0)) - _, err := s.Push(auxv) + _, err := s.pushAddrSliceAndTerminator(auxv) if err != nil { return StackLayout{}, err } // Push environment. - _, err = s.Push(envAddrs) + _, err = s.pushAddrSliceAndTerminator(envAddrs) if err != nil { return StackLayout{}, err } // Push args. - _, err = s.Push(argAddrs) + _, err = s.pushAddrSliceAndTerminator(argAddrs) if err != nil { return StackLayout{}, err } // Push arg count. - _, err = s.Push(usermem.Addr(len(args))) - if err != nil { + lenP := s.Arch.Native(uintptr(len(args))) + if _, err = lenP.CopyOut(s, StackBottomMagic); err != nil { return StackLayout{}, err } diff --git a/pkg/sentry/arch/stack_unsafe.go b/pkg/sentry/arch/stack_unsafe.go new file mode 100644 index 000000000..a90d297ee --- /dev/null +++ b/pkg/sentry/arch/stack_unsafe.go @@ -0,0 +1,69 @@ +// 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 arch + +import ( + "reflect" + "runtime" + "unsafe" + + "gvisor.dev/gvisor/pkg/marshal/primitive" + "gvisor.dev/gvisor/pkg/usermem" +) + +// pushAddrSliceAndTerminator copies a slices of addresses to the stack, and +// also pushes an extra null address element at the end of the slice. +// +// Internally, we unsafely transmute the slice type from the arch-dependent +// []usermem.Addr type, to a slice of fixed-sized ints so that we can pass it to +// go-marshal. +// +// On error, the contents of the stack and the bottom cursor are undefined. +func (s *Stack) pushAddrSliceAndTerminator(src []usermem.Addr) (int, error) { + // Note: Stack grows upwards, so push the terminator first. + srcHdr := (*reflect.SliceHeader)(unsafe.Pointer(&src)) + switch s.Arch.Width() { + case 8: + nNull, err := primitive.CopyUint64Out(s, StackBottomMagic, 0) + if err != nil { + return 0, err + } + var dst []uint64 + dstHdr := (*reflect.SliceHeader)(unsafe.Pointer(&dst)) + dstHdr.Data = srcHdr.Data + dstHdr.Len = srcHdr.Len + dstHdr.Cap = srcHdr.Cap + n, err := primitive.CopyUint64SliceOut(s, StackBottomMagic, dst) + // Ensures src doesn't get GCed until we're done using it through dst. + runtime.KeepAlive(src) + return n + nNull, err + case 4: + nNull, err := primitive.CopyUint32Out(s, StackBottomMagic, 0) + if err != nil { + return 0, err + } + var dst []uint32 + dstHdr := (*reflect.SliceHeader)(unsafe.Pointer(&dst)) + dstHdr.Data = srcHdr.Data + dstHdr.Len = srcHdr.Len + dstHdr.Cap = srcHdr.Cap + n, err := primitive.CopyUint32SliceOut(s, StackBottomMagic, dst) + // Ensure src doesn't get GCed until we're done using it through dst. + runtime.KeepAlive(src) + return n + nNull, err + default: + panic("Unsupported arch width") + } +} diff --git a/pkg/sentry/devices/memdev/full.go b/pkg/sentry/devices/memdev/full.go index 511179e31..fece3e762 100644 --- a/pkg/sentry/devices/memdev/full.go +++ b/pkg/sentry/devices/memdev/full.go @@ -24,6 +24,8 @@ import ( const fullDevMinor = 7 // fullDevice implements vfs.Device for /dev/full. +// +// +stateify savable type fullDevice struct{} // Open implements vfs.Device.Open. @@ -38,6 +40,8 @@ func (fullDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, op } // fullFD implements vfs.FileDescriptionImpl for /dev/full. +// +// +stateify savable type fullFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/devices/memdev/null.go b/pkg/sentry/devices/memdev/null.go index 4918dbeeb..ff5837747 100644 --- a/pkg/sentry/devices/memdev/null.go +++ b/pkg/sentry/devices/memdev/null.go @@ -25,6 +25,8 @@ import ( const nullDevMinor = 3 // nullDevice implements vfs.Device for /dev/null. +// +// +stateify savable type nullDevice struct{} // Open implements vfs.Device.Open. @@ -39,6 +41,8 @@ func (nullDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, op } // nullFD implements vfs.FileDescriptionImpl for /dev/null. +// +// +stateify savable type nullFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/devices/memdev/random.go b/pkg/sentry/devices/memdev/random.go index 5e7fe0280..ac943e3ba 100644 --- a/pkg/sentry/devices/memdev/random.go +++ b/pkg/sentry/devices/memdev/random.go @@ -30,6 +30,8 @@ const ( ) // randomDevice implements vfs.Device for /dev/random and /dev/urandom. +// +// +stateify savable type randomDevice struct{} // Open implements vfs.Device.Open. @@ -44,6 +46,8 @@ func (randomDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, } // randomFD implements vfs.FileDescriptionImpl for /dev/random. +// +// +stateify savable type randomFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/devices/memdev/zero.go b/pkg/sentry/devices/memdev/zero.go index 60cfea888..1929e41cd 100644 --- a/pkg/sentry/devices/memdev/zero.go +++ b/pkg/sentry/devices/memdev/zero.go @@ -27,6 +27,8 @@ import ( const zeroDevMinor = 5 // zeroDevice implements vfs.Device for /dev/zero. +// +// +stateify savable type zeroDevice struct{} // Open implements vfs.Device.Open. @@ -41,6 +43,8 @@ func (zeroDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, op } // zeroFD implements vfs.FileDescriptionImpl for /dev/zero. +// +// +stateify savable type zeroFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/devices/ttydev/ttydev.go b/pkg/sentry/devices/ttydev/ttydev.go index 664e54498..a287c65ca 100644 --- a/pkg/sentry/devices/ttydev/ttydev.go +++ b/pkg/sentry/devices/ttydev/ttydev.go @@ -30,6 +30,8 @@ const ( ) // ttyDevice implements vfs.Device for /dev/tty. +// +// +stateify savable type ttyDevice struct{} // Open implements vfs.Device.Open. diff --git a/pkg/sentry/devices/tundev/BUILD b/pkg/sentry/devices/tundev/BUILD index 71c59287c..14a8bf9cd 100644 --- a/pkg/sentry/devices/tundev/BUILD +++ b/pkg/sentry/devices/tundev/BUILD @@ -17,6 +17,7 @@ go_library( "//pkg/sentry/vfs", "//pkg/syserror", "//pkg/tcpip/link/tun", + "//pkg/tcpip/network/arp", "//pkg/usermem", "//pkg/waiter", ], diff --git a/pkg/sentry/devices/tundev/tundev.go b/pkg/sentry/devices/tundev/tundev.go index 0b701a289..655ea549b 100644 --- a/pkg/sentry/devices/tundev/tundev.go +++ b/pkg/sentry/devices/tundev/tundev.go @@ -16,6 +16,8 @@ package tundev import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -26,6 +28,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip/link/tun" + "gvisor.dev/gvisor/pkg/tcpip/network/arp" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -84,7 +87,16 @@ func (fd *tunFD) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArg return 0, err } flags := usermem.ByteOrder.Uint16(req.Data[:]) - return 0, fd.device.SetIff(stack.Stack, req.Name(), flags) + created, err := fd.device.SetIff(stack.Stack, req.Name(), flags) + if err == nil && created { + // Always start with an ARP address for interfaces so they can handle ARP + // packets. + nicID := fd.device.NICID() + if err := stack.Stack.AddAddress(nicID, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + panic(fmt.Sprintf("failed to add ARP address after creating new TUN/TAP interface with ID = %d", nicID)) + } + } + return 0, err case linux.TUNGETIFF: var req linux.IFReq diff --git a/pkg/sentry/fs/dev/BUILD b/pkg/sentry/fs/dev/BUILD index 9379a4d7b..6b7b451b8 100644 --- a/pkg/sentry/fs/dev/BUILD +++ b/pkg/sentry/fs/dev/BUILD @@ -34,6 +34,7 @@ go_library( "//pkg/sentry/socket/netstack", "//pkg/syserror", "//pkg/tcpip/link/tun", + "//pkg/tcpip/network/arp", "//pkg/usermem", "//pkg/waiter", ], diff --git a/pkg/sentry/fs/dev/net_tun.go b/pkg/sentry/fs/dev/net_tun.go index 5f8c9b5a2..19ffdec47 100644 --- a/pkg/sentry/fs/dev/net_tun.go +++ b/pkg/sentry/fs/dev/net_tun.go @@ -15,6 +15,8 @@ package dev import ( + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -25,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip/link/tun" + "gvisor.dev/gvisor/pkg/tcpip/network/arp" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -60,7 +63,7 @@ func newNetTunDevice(ctx context.Context, owner fs.FileOwner, mode linux.FileMod } // GetFile implements fs.InodeOperations.GetFile. -func (iops *netTunInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { +func (*netTunInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { return fs.NewFile(ctx, d, flags, &netTunFileOperations{}), nil } @@ -80,12 +83,12 @@ type netTunFileOperations struct { var _ fs.FileOperations = (*netTunFileOperations)(nil) // Release implements fs.FileOperations.Release. -func (fops *netTunFileOperations) Release(ctx context.Context) { - fops.device.Release(ctx) +func (n *netTunFileOperations) Release(ctx context.Context) { + n.device.Release(ctx) } // Ioctl implements fs.FileOperations.Ioctl. -func (fops *netTunFileOperations) Ioctl(ctx context.Context, file *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (n *netTunFileOperations) Ioctl(ctx context.Context, file *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { request := args[1].Uint() data := args[2].Pointer() @@ -109,16 +112,25 @@ func (fops *netTunFileOperations) Ioctl(ctx context.Context, file *fs.File, io u return 0, err } flags := usermem.ByteOrder.Uint16(req.Data[:]) - return 0, fops.device.SetIff(stack.Stack, req.Name(), flags) + created, err := n.device.SetIff(stack.Stack, req.Name(), flags) + if err == nil && created { + // Always start with an ARP address for interfaces so they can handle ARP + // packets. + nicID := n.device.NICID() + if err := stack.Stack.AddAddress(nicID, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + panic(fmt.Sprintf("failed to add ARP address after creating new TUN/TAP interface with ID = %d", nicID)) + } + } + return 0, err case linux.TUNGETIFF: var req linux.IFReq - copy(req.IFName[:], fops.device.Name()) + copy(req.IFName[:], n.device.Name()) // Linux adds IFF_NOFILTER (the same value as IFF_NO_PI unfortunately) when // there is no sk_filter. See __tun_chr_ioctl() in net/drivers/tun.c. - flags := fops.device.Flags() | linux.IFF_NOFILTER + flags := n.device.Flags() | linux.IFF_NOFILTER usermem.ByteOrder.PutUint16(req.Data[:], flags) _, err := req.CopyOut(t, data) @@ -130,41 +142,41 @@ func (fops *netTunFileOperations) Ioctl(ctx context.Context, file *fs.File, io u } // Write implements fs.FileOperations.Write. -func (fops *netTunFileOperations) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { +func (n *netTunFileOperations) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { data := make([]byte, src.NumBytes()) if _, err := src.CopyIn(ctx, data); err != nil { return 0, err } - return fops.device.Write(data) + return n.device.Write(data) } // Read implements fs.FileOperations.Read. -func (fops *netTunFileOperations) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - data, err := fops.device.Read() +func (n *netTunFileOperations) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + data, err := n.device.Read() if err != nil { return 0, err } - n, err := dst.CopyOut(ctx, data) - if n > 0 && n < len(data) { + bytesCopied, err := dst.CopyOut(ctx, data) + if bytesCopied > 0 && bytesCopied < len(data) { // Not an error for partial copying. Packet truncated. err = nil } - return int64(n), err + return int64(bytesCopied), err } // Readiness implements watier.Waitable.Readiness. -func (fops *netTunFileOperations) Readiness(mask waiter.EventMask) waiter.EventMask { - return fops.device.Readiness(mask) +func (n *netTunFileOperations) Readiness(mask waiter.EventMask) waiter.EventMask { + return n.device.Readiness(mask) } // EventRegister implements watier.Waitable.EventRegister. -func (fops *netTunFileOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - fops.device.EventRegister(e, mask) +func (n *netTunFileOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + n.device.EventRegister(e, mask) } // EventUnregister implements watier.Waitable.EventUnregister. -func (fops *netTunFileOperations) EventUnregister(e *waiter.Entry) { - fops.device.EventUnregister(e) +func (n *netTunFileOperations) EventUnregister(e *waiter.Entry) { + n.device.EventUnregister(e) } // isNetTunSupported returns whether /dev/net/tun device is supported for s. diff --git a/pkg/sentry/fs/fsutil/inode_cached.go b/pkg/sentry/fs/fsutil/inode_cached.go index 9eb6f522e..85a23432b 100644 --- a/pkg/sentry/fs/fsutil/inode_cached.go +++ b/pkg/sentry/fs/fsutil/inode_cached.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -444,7 +443,7 @@ func (c *CachingInodeOperations) TouchAccessTime(ctx context.Context, inode *fs. // time. // // Preconditions: c.attrMu is locked for writing. -func (c *CachingInodeOperations) touchAccessTimeLocked(now time.Time) { +func (c *CachingInodeOperations) touchAccessTimeLocked(now ktime.Time) { c.attr.AccessTime = now c.dirtyAttr.AccessTime = true } @@ -461,7 +460,7 @@ func (c *CachingInodeOperations) TouchModificationAndStatusChangeTime(ctx contex // and status change times in-place to the current time. // // Preconditions: c.attrMu is locked for writing. -func (c *CachingInodeOperations) touchModificationAndStatusChangeTimeLocked(now time.Time) { +func (c *CachingInodeOperations) touchModificationAndStatusChangeTimeLocked(now ktime.Time) { c.attr.ModificationTime = now c.dirtyAttr.ModificationTime = true c.attr.StatusChangeTime = now @@ -480,7 +479,7 @@ func (c *CachingInodeOperations) TouchStatusChangeTime(ctx context.Context) { // in-place to the current time. // // Preconditions: c.attrMu is locked for writing. -func (c *CachingInodeOperations) touchStatusChangeTimeLocked(now time.Time) { +func (c *CachingInodeOperations) touchStatusChangeTimeLocked(now ktime.Time) { c.attr.StatusChangeTime = now c.dirtyAttr.StatusChangeTime = true } @@ -672,9 +671,6 @@ func (rw *inodeReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { // Continue. seg, gap = gap.NextSegment(), FileRangeGapIterator{} } - - default: - break } } unlock() @@ -768,9 +764,6 @@ func (rw *inodeReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error // Continue. seg, gap = gap.NextSegment(), FileRangeGapIterator{} - - default: - break } } rw.maybeGrowFile() diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 103bfc600..22d658acf 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -84,6 +84,7 @@ func (p *proc) newTaskDir(t *kernel.Task, msrc *fs.MountSource, isThreadGroup bo "auxv": newAuxvec(t, msrc), "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), "comm": newComm(t, msrc), + "cwd": newCwd(t, msrc), "environ": newExecArgInode(t, msrc, environExecArg), "exe": newExe(t, msrc), "fd": newFdDir(t, msrc), @@ -300,6 +301,49 @@ func (e *exe) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { return exec.PathnameWithDeleted(ctx), nil } +// cwd is an fs.InodeOperations symlink for the /proc/PID/cwd file. +// +// +stateify savable +type cwd struct { + ramfs.Symlink + + t *kernel.Task +} + +func newCwd(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { + cwdSymlink := &cwd{ + Symlink: *ramfs.NewSymlink(t, fs.RootOwner, ""), + t: t, + } + return newProcInode(t, cwdSymlink, msrc, fs.Symlink, t) +} + +// Readlink implements fs.InodeOperations. +func (e *cwd) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { + if !kernel.ContextCanTrace(ctx, e.t, false) { + return "", syserror.EACCES + } + if err := checkTaskState(e.t); err != nil { + return "", err + } + cwd := e.t.FSContext().WorkingDirectory() + if cwd == nil { + // It could have raced with process deletion. + return "", syserror.ESRCH + } + defer cwd.DecRef(ctx) + + root := fs.RootFromContext(ctx) + if root == nil { + // It could have raced with process deletion. + return "", syserror.ESRCH + } + defer root.DecRef(ctx) + + name, _ := cwd.FullName(root) + return name, nil +} + // namespaceSymlink represents a symlink in the namespacefs, such as the files // in /proc/<pid>/ns. // diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go index 323506d33..be0900030 100644 --- a/pkg/sentry/fsbridge/vfs.go +++ b/pkg/sentry/fsbridge/vfs.go @@ -122,7 +122,7 @@ func NewVFSLookup(mntns *vfs.MountNamespace, root, workingDir vfs.VirtualDentry) // remainingTraversals is not configurable in VFS2, all callers are using the // default anyways. func (l *vfsLookup) OpenPath(ctx context.Context, pathname string, opts vfs.OpenOptions, _ *uint, resolveFinal bool) (File, error) { - vfsObj := l.mntns.Root().Mount().Filesystem().VirtualFilesystem() + vfsObj := l.root.Mount().Filesystem().VirtualFilesystem() creds := auth.CredentialsFromContext(ctx) path := fspath.Parse(pathname) pop := &vfs.PathOperation{ diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index e73955126..903135fae 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -35,6 +35,8 @@ import ( const Name = "devpts" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // Name implements vfs.FilesystemType.Name. @@ -58,6 +60,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return fs.Filesystem.VFSFilesystem(), root.VFSDentry(), nil } +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -110,6 +113,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // rootInode is the root directory inode for the devpts mounts. +// +// +stateify savable type rootInode struct { implStatFS kernfs.AlwaysValid @@ -131,7 +136,7 @@ type rootInode struct { root *rootInode // mu protects the fields below. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // replicas maps pty ids to replica inodes. replicas map[uint32]*replicaInode @@ -187,8 +192,8 @@ func (i *rootInode) masterClose(t *Terminal) { } // Open implements kernfs.Inode.Open. -func (i *rootInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *rootInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndStaticEntries, }) if err != nil { @@ -242,6 +247,7 @@ func (i *rootInode) DecRef(context.Context) { i.rootInodeRefs.DecRef(i.Destroy) } +// +stateify savable type implStatFS struct{} // StatFS implements kernfs.Inode.StatFS. diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 83d790b38..69c2fe951 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -31,6 +31,8 @@ import ( ) // masterInode is the inode for the master end of the Terminal. +// +// +stateify savable type masterInode struct { implStatFS kernfs.InodeAttrs @@ -50,20 +52,18 @@ type masterInode struct { var _ kernfs.Inode = (*masterInode)(nil) // Open implements kernfs.Inode.Open. -func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { t, err := mi.root.allocateTerminal(rp.Credentials()) if err != nil { return nil, err } - mi.IncRef() fd := &masterFileDescription{ inode: mi, t: t, } fd.LockFD.Init(&mi.locks) - if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { - mi.DecRef(ctx) + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil @@ -89,6 +89,7 @@ func (mi *masterInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds return mi.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) } +// +stateify savable type masterFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -103,7 +104,6 @@ var _ vfs.FileDescriptionImpl = (*masterFileDescription)(nil) // Release implements vfs.FileDescriptionImpl.Release. func (mfd *masterFileDescription) Release(ctx context.Context) { mfd.inode.root.masterClose(mfd.t) - mfd.inode.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fsimpl/devpts/replica.go b/pkg/sentry/fsimpl/devpts/replica.go index 58f6c1d3a..6515c5536 100644 --- a/pkg/sentry/fsimpl/devpts/replica.go +++ b/pkg/sentry/fsimpl/devpts/replica.go @@ -30,6 +30,8 @@ import ( ) // replicaInode is the inode for the replica end of the Terminal. +// +// +stateify savable type replicaInode struct { implStatFS kernfs.InodeAttrs @@ -52,14 +54,12 @@ type replicaInode struct { var _ kernfs.Inode = (*replicaInode)(nil) // Open implements kernfs.Inode.Open. -func (si *replicaInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - si.IncRef() +func (ri *replicaInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &replicaFileDescription{ - inode: si, + inode: ri, } - fd.LockFD.Init(&si.locks) - if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { - si.DecRef(ctx) + fd.LockFD.Init(&ri.locks) + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil @@ -67,34 +67,35 @@ func (si *replicaInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *v } // Valid implements kernfs.Inode.Valid. -func (si *replicaInode) Valid(context.Context) bool { +func (ri *replicaInode) Valid(context.Context) bool { // Return valid if the replica still exists. - si.root.mu.Lock() - defer si.root.mu.Unlock() - _, ok := si.root.replicas[si.t.n] + ri.root.mu.Lock() + defer ri.root.mu.Unlock() + _, ok := ri.root.replicas[ri.t.n] return ok } // Stat implements kernfs.Inode.Stat. -func (si *replicaInode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - statx, err := si.InodeAttrs.Stat(ctx, vfsfs, opts) +func (ri *replicaInode) Stat(ctx context.Context, vfsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + statx, err := ri.InodeAttrs.Stat(ctx, vfsfs, opts) if err != nil { return linux.Statx{}, err } statx.Blksize = 1024 statx.RdevMajor = linux.UNIX98_PTY_REPLICA_MAJOR - statx.RdevMinor = si.t.n + statx.RdevMinor = ri.t.n return statx, nil } // SetStat implements kernfs.Inode.SetStat -func (si *replicaInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { +func (ri *replicaInode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { if opts.Stat.Mask&linux.STATX_SIZE != 0 { return syserror.EINVAL } - return si.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) + return ri.InodeAttrs.SetStat(ctx, vfsfs, creds, opts) } +// +stateify savable type replicaFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -106,37 +107,35 @@ type replicaFileDescription struct { var _ vfs.FileDescriptionImpl = (*replicaFileDescription)(nil) // Release implements fs.FileOperations.Release. -func (sfd *replicaFileDescription) Release(ctx context.Context) { - sfd.inode.DecRef(ctx) -} +func (rfd *replicaFileDescription) Release(ctx context.Context) {} // EventRegister implements waiter.Waitable.EventRegister. -func (sfd *replicaFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - sfd.inode.t.ld.replicaWaiter.EventRegister(e, mask) +func (rfd *replicaFileDescription) EventRegister(e *waiter.Entry, mask waiter.EventMask) { + rfd.inode.t.ld.replicaWaiter.EventRegister(e, mask) } // EventUnregister implements waiter.Waitable.EventUnregister. -func (sfd *replicaFileDescription) EventUnregister(e *waiter.Entry) { - sfd.inode.t.ld.replicaWaiter.EventUnregister(e) +func (rfd *replicaFileDescription) EventUnregister(e *waiter.Entry) { + rfd.inode.t.ld.replicaWaiter.EventUnregister(e) } // Readiness implements waiter.Waitable.Readiness. -func (sfd *replicaFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { - return sfd.inode.t.ld.replicaReadiness() +func (rfd *replicaFileDescription) Readiness(mask waiter.EventMask) waiter.EventMask { + return rfd.inode.t.ld.replicaReadiness() } // Read implements vfs.FileDescriptionImpl.Read. -func (sfd *replicaFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { - return sfd.inode.t.ld.inputQueueRead(ctx, dst) +func (rfd *replicaFileDescription) Read(ctx context.Context, dst usermem.IOSequence, _ vfs.ReadOptions) (int64, error) { + return rfd.inode.t.ld.inputQueueRead(ctx, dst) } // Write implements vfs.FileDescriptionImpl.Write. -func (sfd *replicaFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { - return sfd.inode.t.ld.outputQueueWrite(ctx, src) +func (rfd *replicaFileDescription) Write(ctx context.Context, src usermem.IOSequence, _ vfs.WriteOptions) (int64, error) { + return rfd.inode.t.ld.outputQueueWrite(ctx, src) } // Ioctl implements vfs.FileDescriptionImpl.Ioctl. -func (sfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (rfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { t := kernel.TaskFromContext(ctx) if t == nil { // ioctl(2) may only be called from a task goroutine. @@ -146,35 +145,35 @@ func (sfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, arg switch cmd := args[1].Uint(); cmd { case linux.FIONREAD: // linux.FIONREAD == linux.TIOCINQ // Get the number of bytes in the input queue read buffer. - return 0, sfd.inode.t.ld.inputQueueReadSize(t, io, args) + return 0, rfd.inode.t.ld.inputQueueReadSize(t, io, args) case linux.TCGETS: - return sfd.inode.t.ld.getTermios(t, args) + return rfd.inode.t.ld.getTermios(t, args) case linux.TCSETS: - return sfd.inode.t.ld.setTermios(t, args) + return rfd.inode.t.ld.setTermios(t, args) case linux.TCSETSW: // TODO(b/29356795): This should drain the output queue first. - return sfd.inode.t.ld.setTermios(t, args) + return rfd.inode.t.ld.setTermios(t, args) case linux.TIOCGPTN: - nP := primitive.Uint32(sfd.inode.t.n) + nP := primitive.Uint32(rfd.inode.t.n) _, err := nP.CopyOut(t, args[2].Pointer()) return 0, err case linux.TIOCGWINSZ: - return 0, sfd.inode.t.ld.windowSize(t, args) + return 0, rfd.inode.t.ld.windowSize(t, args) case linux.TIOCSWINSZ: - return 0, sfd.inode.t.ld.setWindowSize(t, args) + return 0, rfd.inode.t.ld.setWindowSize(t, args) case linux.TIOCSCTTY: // Make the given terminal the controlling terminal of the // calling process. - return 0, sfd.inode.t.setControllingTTY(ctx, args, false /* isMaster */) + return 0, rfd.inode.t.setControllingTTY(ctx, args, false /* isMaster */) case linux.TIOCNOTTY: // Release this process's controlling terminal. - return 0, sfd.inode.t.releaseControllingTTY(ctx, args, false /* isMaster */) + return 0, rfd.inode.t.releaseControllingTTY(ctx, args, false /* isMaster */) case linux.TIOCGPGRP: // Get the foreground process group. - return sfd.inode.t.foregroundProcessGroup(ctx, args, false /* isMaster */) + return rfd.inode.t.foregroundProcessGroup(ctx, args, false /* isMaster */) case linux.TIOCSPGRP: // Set the foreground process group. - return sfd.inode.t.setForegroundProcessGroup(ctx, args, false /* isMaster */) + return rfd.inode.t.setForegroundProcessGroup(ctx, args, false /* isMaster */) default: maybeEmitUnimplementedEvent(ctx, cmd) return 0, syserror.ENOTTY @@ -182,24 +181,24 @@ func (sfd *replicaFileDescription) Ioctl(ctx context.Context, io usermem.IO, arg } // SetStat implements vfs.FileDescriptionImpl.SetStat. -func (sfd *replicaFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { +func (rfd *replicaFileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { creds := auth.CredentialsFromContext(ctx) - fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() - return sfd.inode.SetStat(ctx, fs, creds, opts) + fs := rfd.vfsfd.VirtualDentry().Mount().Filesystem() + return rfd.inode.SetStat(ctx, fs, creds, opts) } // Stat implements vfs.FileDescriptionImpl.Stat. -func (sfd *replicaFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - fs := sfd.vfsfd.VirtualDentry().Mount().Filesystem() - return sfd.inode.Stat(ctx, fs, opts) +func (rfd *replicaFileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := rfd.vfsfd.VirtualDentry().Mount().Filesystem() + return rfd.inode.Stat(ctx, fs, opts) } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (sfd *replicaFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { - return sfd.Locks().LockPOSIX(ctx, &sfd.vfsfd, uid, t, start, length, whence, block) +func (rfd *replicaFileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { + return rfd.Locks().LockPOSIX(ctx, &rfd.vfsfd, uid, t, start, length, whence, block) } // UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (sfd *replicaFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { - return sfd.Locks().UnlockPOSIX(ctx, &sfd.vfsfd, uid, start, length, whence) +func (rfd *replicaFileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { + return rfd.Locks().UnlockPOSIX(ctx, &rfd.vfsfd, uid, start, length, whence) } diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index a23094e54..6d1753080 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -33,8 +33,10 @@ import ( const Name = "devtmpfs" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct { - initOnce sync.Once + initOnce sync.Once `state:"nosave"` // FIXME(gvisor.dev/issue/1664): not yet supported. initErr error // fs is the tmpfs filesystem that backs all mounts of this FilesystemType. diff --git a/pkg/sentry/fsimpl/eventfd/eventfd.go b/pkg/sentry/fsimpl/eventfd/eventfd.go index bb0bf3a07..1c27ad700 100644 --- a/pkg/sentry/fsimpl/eventfd/eventfd.go +++ b/pkg/sentry/fsimpl/eventfd/eventfd.go @@ -33,6 +33,8 @@ import ( // EventFileDescription implements vfs.FileDescriptionImpl for file-based event // notification (eventfd). Eventfds are usually internal to the Sentry but in // certain situations they may be converted into a host-backed eventfd. +// +// +stateify savable type EventFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD index abc610ef3..7b1eec3da 100644 --- a/pkg/sentry/fsimpl/ext/BUILD +++ b/pkg/sentry/fsimpl/ext/BUILD @@ -51,6 +51,8 @@ go_library( "//pkg/fd", "//pkg/fspath", "//pkg/log", + "//pkg/marshal", + "//pkg/marshal/primitive", "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/fs", @@ -86,9 +88,9 @@ go_test( library = ":ext", deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/context", "//pkg/fspath", + "//pkg/marshal/primitive", "//pkg/sentry/contexttest", "//pkg/sentry/fsimpl/ext/disklayout", "//pkg/sentry/kernel/auth", diff --git a/pkg/sentry/fsimpl/ext/block_map_file.go b/pkg/sentry/fsimpl/ext/block_map_file.go index 8bb104ff0..1165234f9 100644 --- a/pkg/sentry/fsimpl/ext/block_map_file.go +++ b/pkg/sentry/fsimpl/ext/block_map_file.go @@ -18,7 +18,7 @@ import ( "io" "math" - "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/syserror" ) @@ -34,19 +34,19 @@ type blockMapFile struct { // directBlks are the direct blocks numbers. The physical blocks pointed by // these holds file data. Contains file blocks 0 to 11. - directBlks [numDirectBlks]uint32 + directBlks [numDirectBlks]primitive.Uint32 // indirectBlk is the physical block which contains (blkSize/4) direct block // numbers (as uint32 integers). - indirectBlk uint32 + indirectBlk primitive.Uint32 // doubleIndirectBlk is the physical block which contains (blkSize/4) indirect // block numbers (as uint32 integers). - doubleIndirectBlk uint32 + doubleIndirectBlk primitive.Uint32 // tripleIndirectBlk is the physical block which contains (blkSize/4) doubly // indirect block numbers (as uint32 integers). - tripleIndirectBlk uint32 + tripleIndirectBlk primitive.Uint32 // coverage at (i)th index indicates the amount of file data a node at // height (i) covers. Height 0 is the direct block. @@ -68,10 +68,12 @@ func newBlockMapFile(args inodeArgs) (*blockMapFile, error) { } blkMap := file.regFile.inode.diskInode.Data() - binary.Unmarshal(blkMap[:numDirectBlks*4], binary.LittleEndian, &file.directBlks) - binary.Unmarshal(blkMap[numDirectBlks*4:(numDirectBlks+1)*4], binary.LittleEndian, &file.indirectBlk) - binary.Unmarshal(blkMap[(numDirectBlks+1)*4:(numDirectBlks+2)*4], binary.LittleEndian, &file.doubleIndirectBlk) - binary.Unmarshal(blkMap[(numDirectBlks+2)*4:(numDirectBlks+3)*4], binary.LittleEndian, &file.tripleIndirectBlk) + for i := 0; i < numDirectBlks; i++ { + file.directBlks[i].UnmarshalBytes(blkMap[i*4 : (i+1)*4]) + } + file.indirectBlk.UnmarshalBytes(blkMap[numDirectBlks*4 : (numDirectBlks+1)*4]) + file.doubleIndirectBlk.UnmarshalBytes(blkMap[(numDirectBlks+1)*4 : (numDirectBlks+2)*4]) + file.tripleIndirectBlk.UnmarshalBytes(blkMap[(numDirectBlks+2)*4 : (numDirectBlks+3)*4]) return file, nil } @@ -117,16 +119,16 @@ func (f *blockMapFile) ReadAt(dst []byte, off int64) (int, error) { switch { case offset < dirBlksEnd: // Direct block. - curR, err = f.read(f.directBlks[offset/f.regFile.inode.blkSize], offset%f.regFile.inode.blkSize, 0, dst[read:]) + curR, err = f.read(uint32(f.directBlks[offset/f.regFile.inode.blkSize]), offset%f.regFile.inode.blkSize, 0, dst[read:]) case offset < indirBlkEnd: // Indirect block. - curR, err = f.read(f.indirectBlk, offset-dirBlksEnd, 1, dst[read:]) + curR, err = f.read(uint32(f.indirectBlk), offset-dirBlksEnd, 1, dst[read:]) case offset < doubIndirBlkEnd: // Doubly indirect block. - curR, err = f.read(f.doubleIndirectBlk, offset-indirBlkEnd, 2, dst[read:]) + curR, err = f.read(uint32(f.doubleIndirectBlk), offset-indirBlkEnd, 2, dst[read:]) default: // Triply indirect block. - curR, err = f.read(f.tripleIndirectBlk, offset-doubIndirBlkEnd, 3, dst[read:]) + curR, err = f.read(uint32(f.tripleIndirectBlk), offset-doubIndirBlkEnd, 3, dst[read:]) } read += curR @@ -174,13 +176,13 @@ func (f *blockMapFile) read(curPhyBlk uint32, relFileOff uint64, height uint, ds read := 0 curChildOff := relFileOff % childCov for i := startIdx; i < endIdx; i++ { - var childPhyBlk uint32 + var childPhyBlk primitive.Uint32 err := readFromDisk(f.regFile.inode.fs.dev, curPhyBlkOff+int64(i*4), &childPhyBlk) if err != nil { return read, err } - n, err := f.read(childPhyBlk, curChildOff, height-1, dst[read:]) + n, err := f.read(uint32(childPhyBlk), curChildOff, height-1, dst[read:]) read += n if err != nil { return read, err diff --git a/pkg/sentry/fsimpl/ext/block_map_test.go b/pkg/sentry/fsimpl/ext/block_map_test.go index 6fa84e7aa..ed98b482e 100644 --- a/pkg/sentry/fsimpl/ext/block_map_test.go +++ b/pkg/sentry/fsimpl/ext/block_map_test.go @@ -20,7 +20,7 @@ import ( "testing" "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" ) @@ -87,29 +87,33 @@ func blockMapSetUp(t *testing.T) (*blockMapFile, []byte) { mockDisk := make([]byte, mockBMDiskSize) var fileData []byte blkNums := newBlkNumGen() - var data []byte + off := 0 + data := make([]byte, (numDirectBlks+3)*(*primitive.Uint32)(nil).SizeBytes()) // Write the direct blocks. for i := 0; i < numDirectBlks; i++ { - curBlkNum := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, curBlkNum) - fileData = append(fileData, writeFileDataToBlock(mockDisk, curBlkNum, 0, blkNums)...) + curBlkNum := primitive.Uint32(blkNums.next()) + curBlkNum.MarshalBytes(data[off:]) + off += curBlkNum.SizeBytes() + fileData = append(fileData, writeFileDataToBlock(mockDisk, uint32(curBlkNum), 0, blkNums)...) } // Write to indirect block. - indirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, indirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, indirectBlk, 1, blkNums)...) - - // Write to indirect block. - doublyIndirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, doublyIndirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, doublyIndirectBlk, 2, blkNums)...) - - // Write to indirect block. - triplyIndirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, triplyIndirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, triplyIndirectBlk, 3, blkNums)...) + indirectBlk := primitive.Uint32(blkNums.next()) + indirectBlk.MarshalBytes(data[off:]) + off += indirectBlk.SizeBytes() + fileData = append(fileData, writeFileDataToBlock(mockDisk, uint32(indirectBlk), 1, blkNums)...) + + // Write to double indirect block. + doublyIndirectBlk := primitive.Uint32(blkNums.next()) + doublyIndirectBlk.MarshalBytes(data[off:]) + off += doublyIndirectBlk.SizeBytes() + fileData = append(fileData, writeFileDataToBlock(mockDisk, uint32(doublyIndirectBlk), 2, blkNums)...) + + // Write to triple indirect block. + triplyIndirectBlk := primitive.Uint32(blkNums.next()) + triplyIndirectBlk.MarshalBytes(data[off:]) + fileData = append(fileData, writeFileDataToBlock(mockDisk, uint32(triplyIndirectBlk), 3, blkNums)...) args := inodeArgs{ fs: &filesystem{ @@ -142,9 +146,9 @@ func writeFileDataToBlock(disk []byte, blkNum uint32, height uint, blkNums *blkN var fileData []byte for off := blkNum * mockBMBlkSize; off < (blkNum+1)*mockBMBlkSize; off += 4 { - curBlkNum := blkNums.next() - copy(disk[off:off+4], binary.Marshal(nil, binary.LittleEndian, curBlkNum)) - fileData = append(fileData, writeFileDataToBlock(disk, curBlkNum, height-1, blkNums)...) + curBlkNum := primitive.Uint32(blkNums.next()) + curBlkNum.MarshalBytes(disk[off : off+4]) + fileData = append(fileData, writeFileDataToBlock(disk, uint32(curBlkNum), height-1, blkNums)...) } return fileData } diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index 7a1b4219f..9bfed883a 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -20,6 +20,8 @@ import ( ) // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 0fc01668d..0ad79b381 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -16,7 +16,6 @@ package ext import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -28,6 +27,8 @@ import ( ) // directory represents a directory inode. It holds the childList in memory. +// +// +stateify savable type directory struct { inode inode @@ -39,7 +40,7 @@ type directory struct { // Lock Order (outermost locks must be taken first): // directory.mu // filesystem.mu - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // childList is a list containing (1) child dirents and (2) fake dirents // (with diskDirent == nil) that represent the iteration position of @@ -98,7 +99,7 @@ func newDirectory(args inodeArgs, newDirent bool) (*directory, error) { } else { curDirent.diskDirent = &disklayout.DirentOld{} } - binary.Unmarshal(buf, binary.LittleEndian, curDirent.diskDirent) + curDirent.diskDirent.UnmarshalBytes(buf) if curDirent.diskDirent.Inode() != 0 && len(curDirent.diskDirent.FileName()) != 0 { // Inode number and name length fields being set to 0 is used to indicate @@ -120,6 +121,8 @@ func (i *inode) isDir() bool { } // dirent is the directory.childList node. +// +// +stateify savable type dirent struct { diskDirent disklayout.Dirent @@ -129,6 +132,8 @@ type dirent struct { // directoryFD represents a directory file description. It implements // vfs.FileDescriptionImpl. +// +// +stateify savable type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/ext/disklayout/BUILD b/pkg/sentry/fsimpl/ext/disklayout/BUILD index 9bd9c76c0..d98a05dd8 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/BUILD +++ b/pkg/sentry/fsimpl/ext/disklayout/BUILD @@ -22,10 +22,11 @@ go_library( "superblock_old.go", "test_utils.go", ], + marshal = True, visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/binary", + "//pkg/marshal", "//pkg/sentry/fs", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group.go b/pkg/sentry/fsimpl/ext/disklayout/block_group.go index ad6f4fef8..0d56ae9da 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/block_group.go +++ b/pkg/sentry/fsimpl/ext/disklayout/block_group.go @@ -14,6 +14,10 @@ package disklayout +import ( + "gvisor.dev/gvisor/pkg/marshal" +) + // BlockGroup represents a Linux ext block group descriptor. An ext file system // is split into a series of block groups. This provides an access layer to // information needed to access and use a block group. @@ -30,6 +34,8 @@ package disklayout // // See https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html#block-group-descriptors. type BlockGroup interface { + marshal.Marshallable + // InodeTable returns the absolute block number of the block containing the // inode table. This points to an array of Inode structs. Inode tables are // statically allocated at mkfs time. The superblock records the number of diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group_32.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_32.go index 3e16c76db..a35fa22a0 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/block_group_32.go +++ b/pkg/sentry/fsimpl/ext/disklayout/block_group_32.go @@ -17,6 +17,8 @@ package disklayout // BlockGroup32Bit emulates the first half of struct ext4_group_desc in // fs/ext4/ext4.h. It is the block group descriptor struct for ext2, ext3 and // 32-bit ext4 filesystems. It implements BlockGroup interface. +// +// +marshal type BlockGroup32Bit struct { BlockBitmapLo uint32 InodeBitmapLo uint32 diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group_64.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_64.go index 9a809197a..d54d1d345 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/block_group_64.go +++ b/pkg/sentry/fsimpl/ext/disklayout/block_group_64.go @@ -18,6 +18,8 @@ package disklayout // It is the block group descriptor struct for 64-bit ext4 filesystems. // It implements BlockGroup interface. It is an extension of the 32-bit // version of BlockGroup. +// +// +marshal type BlockGroup64Bit struct { // We embed the 32-bit struct here because 64-bit version is just an extension // of the 32-bit version. diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group_test.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_test.go index 0ef4294c0..e4ce484e4 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/block_group_test.go +++ b/pkg/sentry/fsimpl/ext/disklayout/block_group_test.go @@ -21,6 +21,8 @@ import ( // TestBlockGroupSize tests that the block group descriptor structs are of the // correct size. func TestBlockGroupSize(t *testing.T) { - assertSize(t, BlockGroup32Bit{}, 32) - assertSize(t, BlockGroup64Bit{}, 64) + var bgSmall BlockGroup32Bit + assertSize(t, &bgSmall, 32) + var bgBig BlockGroup64Bit + assertSize(t, &bgBig, 64) } diff --git a/pkg/sentry/fsimpl/ext/disklayout/dirent.go b/pkg/sentry/fsimpl/ext/disklayout/dirent.go index 417b6cf65..568c8cb4c 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent.go +++ b/pkg/sentry/fsimpl/ext/disklayout/dirent.go @@ -15,6 +15,7 @@ package disklayout import ( + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/fs" ) @@ -51,6 +52,8 @@ var ( // // See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#linear-classic-directories. type Dirent interface { + marshal.Marshallable + // Inode returns the absolute inode number of the underlying inode. // Inode number 0 signifies an unused dirent. Inode() uint32 diff --git a/pkg/sentry/fsimpl/ext/disklayout/dirent_new.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_new.go index 29ae4a5c2..51f9c2946 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent_new.go +++ b/pkg/sentry/fsimpl/ext/disklayout/dirent_new.go @@ -29,12 +29,14 @@ import ( // Note: This struct can be of variable size on disk. The one described below // is of maximum size and the FileName beyond NameLength bytes might contain // garbage. +// +// +marshal type DirentNew struct { InodeNumber uint32 RecordLength uint16 NameLength uint8 FileTypeRaw uint8 - FileNameRaw [MaxFileName]byte + FileNameRaw [MaxFileName]byte `marshal:"unaligned"` } // Compiles only if DirentNew implements Dirent. diff --git a/pkg/sentry/fsimpl/ext/disklayout/dirent_old.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_old.go index 6fff12a6e..d4b19e086 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent_old.go +++ b/pkg/sentry/fsimpl/ext/disklayout/dirent_old.go @@ -22,11 +22,13 @@ import "gvisor.dev/gvisor/pkg/sentry/fs" // Note: This struct can be of variable size on disk. The one described below // is of maximum size and the FileName beyond NameLength bytes might contain // garbage. +// +// +marshal type DirentOld struct { InodeNumber uint32 RecordLength uint16 NameLength uint16 - FileNameRaw [MaxFileName]byte + FileNameRaw [MaxFileName]byte `marshal:"unaligned"` } // Compiles only if DirentOld implements Dirent. diff --git a/pkg/sentry/fsimpl/ext/disklayout/dirent_test.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_test.go index 934919f8a..3486864dc 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent_test.go +++ b/pkg/sentry/fsimpl/ext/disklayout/dirent_test.go @@ -21,6 +21,8 @@ import ( // TestDirentSize tests that the dirent structs are of the correct // size. func TestDirentSize(t *testing.T) { - assertSize(t, DirentOld{}, uintptr(DirentSize)) - assertSize(t, DirentNew{}, uintptr(DirentSize)) + var dOld DirentOld + assertSize(t, &dOld, DirentSize) + var dNew DirentNew + assertSize(t, &dNew, DirentSize) } diff --git a/pkg/sentry/fsimpl/ext/disklayout/disklayout.go b/pkg/sentry/fsimpl/ext/disklayout/disklayout.go index bdf4e2132..0834e9ba8 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/disklayout.go +++ b/pkg/sentry/fsimpl/ext/disklayout/disklayout.go @@ -36,8 +36,6 @@ // escape analysis on an unknown implementation at compile time. // // Notes: -// - All fields in these structs are exported because binary.Read would -// panic otherwise. // - All structures on disk are in little-endian order. Only jbd2 (journal) // structures are in big-endian order. // - All OS dependent fields in these structures will be interpretted using diff --git a/pkg/sentry/fsimpl/ext/disklayout/extent.go b/pkg/sentry/fsimpl/ext/disklayout/extent.go index 4110649ab..b13999bfc 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/extent.go +++ b/pkg/sentry/fsimpl/ext/disklayout/extent.go @@ -14,6 +14,10 @@ package disklayout +import ( + "gvisor.dev/gvisor/pkg/marshal" +) + // Extents were introduced in ext4 and provide huge performance gains in terms // data locality and reduced metadata block usage. Extents are organized in // extent trees. The root node is contained in inode.BlocksRaw. @@ -64,6 +68,8 @@ type ExtentNode struct { // ExtentEntry represents an extent tree node entry. The entry can either be // an ExtentIdx or Extent itself. This exists to simplify navigation logic. type ExtentEntry interface { + marshal.Marshallable + // FileBlock returns the first file block number covered by this entry. FileBlock() uint32 @@ -75,6 +81,8 @@ type ExtentEntry interface { // tree node begins with this and is followed by `NumEntries` number of: // - Extent if `Depth` == 0 // - ExtentIdx otherwise +// +// +marshal type ExtentHeader struct { // Magic in the extent magic number, must be 0xf30a. Magic uint16 @@ -96,6 +104,8 @@ type ExtentHeader struct { // internal nodes. Sorted in ascending order based on FirstFileBlock since // Linux does a binary search on this. This points to a block containing the // child node. +// +// +marshal type ExtentIdx struct { FirstFileBlock uint32 ChildBlockLo uint32 @@ -121,6 +131,8 @@ func (ei *ExtentIdx) PhysicalBlock() uint64 { // nodes. Sorted in ascending order based on FirstFileBlock since Linux does a // binary search on this. This points to an array of data blocks containing the // file data. It covers `Length` data blocks starting from `StartBlock`. +// +// +marshal type Extent struct { FirstFileBlock uint32 Length uint16 diff --git a/pkg/sentry/fsimpl/ext/disklayout/extent_test.go b/pkg/sentry/fsimpl/ext/disklayout/extent_test.go index 8762b90db..c96002e19 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/extent_test.go +++ b/pkg/sentry/fsimpl/ext/disklayout/extent_test.go @@ -21,7 +21,10 @@ import ( // TestExtentSize tests that the extent structs are of the correct // size. func TestExtentSize(t *testing.T) { - assertSize(t, ExtentHeader{}, ExtentHeaderSize) - assertSize(t, ExtentIdx{}, ExtentEntrySize) - assertSize(t, Extent{}, ExtentEntrySize) + var h ExtentHeader + assertSize(t, &h, ExtentHeaderSize) + var i ExtentIdx + assertSize(t, &i, ExtentEntrySize) + var e Extent + assertSize(t, &e, ExtentEntrySize) } diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode.go b/pkg/sentry/fsimpl/ext/disklayout/inode.go index 88ae913f5..ef25040a9 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/inode.go +++ b/pkg/sentry/fsimpl/ext/disklayout/inode.go @@ -16,6 +16,7 @@ package disklayout import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/time" ) @@ -38,6 +39,8 @@ const ( // // See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#index-nodes. type Inode interface { + marshal.Marshallable + // Mode returns the linux file mode which is majorly used to extract // information like: // - File permissions (read/write/execute by user/group/others). diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode_new.go b/pkg/sentry/fsimpl/ext/disklayout/inode_new.go index 8f9f574ce..a4503f5cf 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/inode_new.go +++ b/pkg/sentry/fsimpl/ext/disklayout/inode_new.go @@ -27,6 +27,8 @@ import "gvisor.dev/gvisor/pkg/sentry/kernel/time" // are used to provide nanoscond precision. Hence, these timestamps will now // overflow in May 2446. // See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#inode-timestamps. +// +// +marshal type InodeNew struct { InodeOld diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode_old.go b/pkg/sentry/fsimpl/ext/disklayout/inode_old.go index db25b11b6..e6b28babf 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/inode_old.go +++ b/pkg/sentry/fsimpl/ext/disklayout/inode_old.go @@ -30,6 +30,8 @@ const ( // // All fields representing time are in seconds since the epoch. Which means that // they will overflow in January 2038. +// +// +marshal type InodeOld struct { ModeRaw uint16 UIDLo uint16 diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode_test.go b/pkg/sentry/fsimpl/ext/disklayout/inode_test.go index dd03ee50e..90744e956 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/inode_test.go +++ b/pkg/sentry/fsimpl/ext/disklayout/inode_test.go @@ -24,10 +24,12 @@ import ( // TestInodeSize tests that the inode structs are of the correct size. func TestInodeSize(t *testing.T) { - assertSize(t, InodeOld{}, OldInodeSize) + var iOld InodeOld + assertSize(t, &iOld, OldInodeSize) // This was updated from 156 bytes to 160 bytes in Oct 2015. - assertSize(t, InodeNew{}, 160) + var iNew InodeNew + assertSize(t, &iNew, 160) } // TestTimestampSeconds tests that the seconds part of [a/c/m] timestamps in diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock.go b/pkg/sentry/fsimpl/ext/disklayout/superblock.go index 8bb327006..70948ebe9 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock.go +++ b/pkg/sentry/fsimpl/ext/disklayout/superblock.go @@ -14,6 +14,10 @@ package disklayout +import ( + "gvisor.dev/gvisor/pkg/marshal" +) + const ( // SbOffset is the absolute offset at which the superblock is placed. SbOffset = 1024 @@ -38,6 +42,8 @@ const ( // // See https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html#super-block. type SuperBlock interface { + marshal.Marshallable + // InodesCount returns the total number of inodes in this filesystem. InodesCount() uint32 diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock_32.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_32.go index 53e515fd3..4dc6080fb 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock_32.go +++ b/pkg/sentry/fsimpl/ext/disklayout/superblock_32.go @@ -17,6 +17,8 @@ package disklayout // SuperBlock32Bit implements SuperBlock and represents the 32-bit version of // the ext4_super_block struct in fs/ext4/ext4.h. Should be used only if // RevLevel = DynamicRev and 64-bit feature is disabled. +// +// +marshal type SuperBlock32Bit struct { // We embed the old superblock struct here because the 32-bit version is just // an extension of the old version. diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock_64.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_64.go index 7c1053fb4..2c9039327 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock_64.go +++ b/pkg/sentry/fsimpl/ext/disklayout/superblock_64.go @@ -19,6 +19,8 @@ package disklayout // 1024 bytes (smallest possible block size) and hence the superblock always // fits in no more than one data block. Should only be used when the 64-bit // feature is set. +// +// +marshal type SuperBlock64Bit struct { // We embed the 32-bit struct here because 64-bit version is just an extension // of the 32-bit version. diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock_old.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_old.go index 9221e0251..e4709f23c 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock_old.go +++ b/pkg/sentry/fsimpl/ext/disklayout/superblock_old.go @@ -16,6 +16,8 @@ package disklayout // SuperBlockOld implements SuperBlock and represents the old version of the // superblock struct. Should be used only if RevLevel = OldRev. +// +// +marshal type SuperBlockOld struct { InodesCountRaw uint32 BlocksCountLo uint32 diff --git a/pkg/sentry/fsimpl/ext/disklayout/superblock_test.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_test.go index 463b5ba21..b734b6987 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/superblock_test.go +++ b/pkg/sentry/fsimpl/ext/disklayout/superblock_test.go @@ -21,7 +21,10 @@ import ( // TestSuperBlockSize tests that the superblock structs are of the correct // size. func TestSuperBlockSize(t *testing.T) { - assertSize(t, SuperBlockOld{}, 84) - assertSize(t, SuperBlock32Bit{}, 336) - assertSize(t, SuperBlock64Bit{}, 1024) + var sbOld SuperBlockOld + assertSize(t, &sbOld, 84) + var sb32 SuperBlock32Bit + assertSize(t, &sb32, 336) + var sb64 SuperBlock64Bit + assertSize(t, &sb64, 1024) } diff --git a/pkg/sentry/fsimpl/ext/disklayout/test_utils.go b/pkg/sentry/fsimpl/ext/disklayout/test_utils.go index 9c63f04c0..a4bc08411 100644 --- a/pkg/sentry/fsimpl/ext/disklayout/test_utils.go +++ b/pkg/sentry/fsimpl/ext/disklayout/test_utils.go @@ -18,13 +18,13 @@ import ( "reflect" "testing" - "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/marshal" ) -func assertSize(t *testing.T, v interface{}, want uintptr) { +func assertSize(t *testing.T, v marshal.Marshallable, want int) { t.Helper() - if got := binary.Size(v); got != want { + if got := v.SizeBytes(); got != want { t.Errorf("struct %s should be exactly %d bytes but is %d bytes", reflect.TypeOf(v).Name(), want, got) } } diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go index 08ffc2834..aca258d40 100644 --- a/pkg/sentry/fsimpl/ext/ext.go +++ b/pkg/sentry/fsimpl/ext/ext.go @@ -34,6 +34,8 @@ import ( const Name = "ext" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // Compiles only if FilesystemType implements vfs.FilesystemType. diff --git a/pkg/sentry/fsimpl/ext/extent_file.go b/pkg/sentry/fsimpl/ext/extent_file.go index c36225a7c..778460107 100644 --- a/pkg/sentry/fsimpl/ext/extent_file.go +++ b/pkg/sentry/fsimpl/ext/extent_file.go @@ -18,12 +18,13 @@ import ( "io" "sort" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/syserror" ) // extentFile is a type of regular file which uses extents to store file data. +// +// +stateify savable type extentFile struct { regFile regularFile @@ -58,7 +59,7 @@ func newExtentFile(args inodeArgs) (*extentFile, error) { func (f *extentFile) buildExtTree() error { rootNodeData := f.regFile.inode.diskInode.Data() - binary.Unmarshal(rootNodeData[:disklayout.ExtentHeaderSize], binary.LittleEndian, &f.root.Header) + f.root.Header.UnmarshalBytes(rootNodeData[:disklayout.ExtentHeaderSize]) // Root node can not have more than 4 entries: 60 bytes = 1 header + 4 entries. if f.root.Header.NumEntries > 4 { @@ -77,7 +78,7 @@ func (f *extentFile) buildExtTree() error { // Internal node. curEntry = &disklayout.ExtentIdx{} } - binary.Unmarshal(rootNodeData[off:off+disklayout.ExtentEntrySize], binary.LittleEndian, curEntry) + curEntry.UnmarshalBytes(rootNodeData[off : off+disklayout.ExtentEntrySize]) f.root.Entries[i].Entry = curEntry } diff --git a/pkg/sentry/fsimpl/ext/extent_test.go b/pkg/sentry/fsimpl/ext/extent_test.go index cd10d46ee..985f76ac0 100644 --- a/pkg/sentry/fsimpl/ext/extent_test.go +++ b/pkg/sentry/fsimpl/ext/extent_test.go @@ -21,7 +21,6 @@ import ( "github.com/google/go-cmp/cmp" "github.com/google/go-cmp/cmp/cmpopts" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" ) @@ -202,13 +201,14 @@ func extentTreeSetUp(t *testing.T, root *disklayout.ExtentNode) (*extentFile, [] // writeTree writes the tree represented by `root` to the inode and disk. It // also writes random file data on disk. func writeTree(in *inode, disk []byte, root *disklayout.ExtentNode, mockExtentBlkSize uint64) []byte { - rootData := binary.Marshal(nil, binary.LittleEndian, root.Header) + rootData := in.diskInode.Data() + root.Header.MarshalBytes(rootData) + off := root.Header.SizeBytes() for _, ep := range root.Entries { - rootData = binary.Marshal(rootData, binary.LittleEndian, ep.Entry) + ep.Entry.MarshalBytes(rootData[off:]) + off += ep.Entry.SizeBytes() } - copy(in.diskInode.Data(), rootData) - var fileData []byte for _, ep := range root.Entries { if root.Header.Height == 0 { @@ -223,13 +223,14 @@ func writeTree(in *inode, disk []byte, root *disklayout.ExtentNode, mockExtentBl // writeTreeToDisk is the recursive step for writeTree which writes the tree // on the disk only. Also writes random file data on disk. func writeTreeToDisk(disk []byte, curNode disklayout.ExtentEntryPair) []byte { - nodeData := binary.Marshal(nil, binary.LittleEndian, curNode.Node.Header) + nodeData := disk[curNode.Entry.PhysicalBlock()*mockExtentBlkSize:] + curNode.Node.Header.MarshalBytes(nodeData) + off := curNode.Node.Header.SizeBytes() for _, ep := range curNode.Node.Entries { - nodeData = binary.Marshal(nodeData, binary.LittleEndian, ep.Entry) + ep.Entry.MarshalBytes(nodeData[off:]) + off += ep.Entry.SizeBytes() } - copy(disk[curNode.Entry.PhysicalBlock()*mockExtentBlkSize:], nodeData) - var fileData []byte for _, ep := range curNode.Node.Entries { if curNode.Node.Header.Height == 0 { diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index a4a6d8c55..917f1873d 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -38,11 +38,13 @@ var ( ) // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem // mu serializes changes to the Dentry tree. - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` // dev represents the underlying fs device. It does not require protection // because io.ReaderAt permits concurrent read calls to it. It translates to diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go index 30636cf66..9009ba3c7 100644 --- a/pkg/sentry/fsimpl/ext/inode.go +++ b/pkg/sentry/fsimpl/ext/inode.go @@ -37,6 +37,8 @@ import ( // |-- regular-- // |-- extent file // |-- block map file +// +// +stateify savable type inode struct { // refs is a reference count. refs is accessed using atomic memory operations. refs int64 diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index e73e740d6..4a5539b37 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -31,6 +31,8 @@ import ( // regularFile represents a regular file's inode. This too follows the // inheritance pattern prevelant in the vfs layer described in // pkg/sentry/vfs/README.md. +// +// +stateify savable type regularFile struct { inode inode @@ -67,6 +69,8 @@ func (in *inode) isRegular() bool { // directoryFD represents a directory file description. It implements // vfs.FileDescriptionImpl. +// +// +stateify savable type regularFileFD struct { fileDescription vfs.LockFD @@ -75,7 +79,7 @@ type regularFileFD struct { off int64 // offMu serializes operations that may mutate off. - offMu sync.Mutex + offMu sync.Mutex `state:"nosave"` } // Release implements vfs.FileDescriptionImpl.Release. diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go index f33592d59..5e2bcc837 100644 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ b/pkg/sentry/fsimpl/ext/symlink.go @@ -23,6 +23,8 @@ import ( ) // symlink represents a symlink inode. +// +// +stateify savable type symlink struct { inode inode target string // immutable @@ -64,6 +66,8 @@ func (in *inode) isSymlink() bool { // symlinkFD represents a symlink file description and implements // vfs.FileDescriptionImpl. which may only be used if open options contains // O_PATH. For this reason most of the functions return EBADF. +// +// +stateify savable type symlinkFD struct { fileDescription vfs.NoLockFD diff --git a/pkg/sentry/fsimpl/ext/utils.go b/pkg/sentry/fsimpl/ext/utils.go index d8b728f8c..58ef7b9b8 100644 --- a/pkg/sentry/fsimpl/ext/utils.go +++ b/pkg/sentry/fsimpl/ext/utils.go @@ -17,21 +17,21 @@ package ext import ( "io" - "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" "gvisor.dev/gvisor/pkg/syserror" ) // readFromDisk performs a binary read from disk into the given struct from // the absolute offset provided. -func readFromDisk(dev io.ReaderAt, abOff int64, v interface{}) error { - n := binary.Size(v) +func readFromDisk(dev io.ReaderAt, abOff int64, v marshal.Marshallable) error { + n := v.SizeBytes() buf := make([]byte, n) if read, _ := dev.ReadAt(buf, abOff); read < int(n) { return syserror.EIO } - binary.Unmarshal(buf, binary.LittleEndian, v) + v.UnmarshalBytes(buf) return nil } diff --git a/pkg/sentry/fsimpl/fuse/connection.go b/pkg/sentry/fsimpl/fuse/connection.go index dbc5e1954..8ccda1264 100644 --- a/pkg/sentry/fsimpl/fuse/connection.go +++ b/pkg/sentry/fsimpl/fuse/connection.go @@ -39,15 +39,18 @@ const ( ) // connection is the struct by which the sentry communicates with the FUSE server daemon. +// // Lock order: // - conn.fd.mu // - conn.mu // - conn.asyncMu +// +// +stateify savable type connection struct { fd *DeviceFD // mu protects access to struct memebers. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // attributeVersion is the version of connection's attributes. attributeVersion uint64 @@ -75,7 +78,7 @@ type connection struct { initialized int32 // initializedChan is used to block requests before initialization. - initializedChan chan struct{} + initializedChan chan struct{} `state:".(bool)"` // connected (connection established) when a new FUSE file system is created. // Set to false when: @@ -113,7 +116,7 @@ type connection struct { // i.e. `!request.noReply` // asyncMu protects the async request fields. - asyncMu sync.Mutex + asyncMu sync.Mutex `state:"nosave"` // asyncNum is the number of async requests. // Protected by asyncMu. @@ -174,6 +177,22 @@ type connection struct { noOpen bool } +func (conn *connection) saveInitializedChan() bool { + select { + case <-conn.initializedChan: + return true // Closed. + default: + return false // Not closed. + } +} + +func (conn *connection) loadInitializedChan(closed bool) { + conn.initializedChan = make(chan struct{}, 1) + if closed { + close(conn.initializedChan) + } +} + // newFUSEConnection creates a FUSE connection to fd. func newFUSEConnection(_ context.Context, fd *vfs.FileDescription, opts *filesystemOptions) (*connection, error) { // Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index f690ef5ad..1b86a4b4c 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -31,6 +31,8 @@ import ( const fuseDevMinor = 229 // fuseDevice implements vfs.Device for /dev/fuse. +// +// +stateify savable type fuseDevice struct{} // Open implements vfs.Device.Open. @@ -49,6 +51,8 @@ func (fuseDevice) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, op } // DeviceFD implements vfs.FileDescriptionImpl for /dev/fuse. +// +// +stateify savable type DeviceFD struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -79,7 +83,7 @@ type DeviceFD struct { writeCursorFR *futureResponse // mu protects all the queues, maps, buffers and cursors and nextOpID. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // waitQueue is used to notify interested parties when the device becomes // readable or writable. @@ -88,12 +92,20 @@ type DeviceFD struct { // fullQueueCh is a channel used to synchronize the readers with the writers. // Writers (inbound requests to the filesystem) block if there are too many // unprocessed in-flight requests. - fullQueueCh chan struct{} + fullQueueCh chan struct{} `state:".(int)"` // fs is the FUSE filesystem that this FD is being used for. fs *filesystem } +func (fd *DeviceFD) saveFullQueueCh() int { + return cap(fd.fullQueueCh) +} + +func (fd *DeviceFD) loadFullQueueCh(capacity int) { + fd.fullQueueCh = make(chan struct{}, capacity) +} + // Release implements vfs.FileDescriptionImpl.Release. func (fd *DeviceFD) Release(ctx context.Context) { if fd.fs != nil { diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index 2144e72bd..65786e42a 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -41,8 +41,11 @@ const Name = "fuse" const maxActiveRequestsDefault = 10000 // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} +// +stateify savable type filesystemOptions struct { // userID specifies the numeric uid of the mount owner. // This option should not be specified by the filesystem owner. @@ -73,6 +76,8 @@ type filesystemOptions struct { } // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { kernfs.Filesystem devMinor uint32 @@ -240,6 +245,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // inode implements kernfs.Inode. +// +// +stateify savable type inode struct { inodeRefs kernfs.InodeAttrs @@ -280,6 +287,7 @@ func (fs *filesystem) newRootInode(creds *auth.Credentials, mode linux.FileMode) i := &inode{fs: fs} i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, 1, linux.ModeDirectory|0755) i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + i.EnableLeakCheck() i.dentry.Init(i) i.nodeID = 1 @@ -299,7 +307,7 @@ func (fs *filesystem) newInode(nodeID uint64, attr linux.FUSEAttr) *kernfs.Dentr } // Open implements kernfs.Inode.Open. -func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { isDir := i.InodeAttrs.Mode().IsDir() // return error if specified to open directory but inode is not a directory. if !isDir && opts.Mode.IsDir() { @@ -395,7 +403,7 @@ func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentr i.attributeTime = 0 } - if err := fd.vfsfd.Init(fdImpl, opts.Flags, rp.Mount(), vfsd, fdOptions); err != nil { + if err := fd.vfsfd.Init(fdImpl, opts.Flags, rp.Mount(), d.VFSDentry(), fdOptions); err != nil { return nil, err } return &fd.vfsfd, nil @@ -418,7 +426,7 @@ func (*inode) Valid(ctx context.Context) bool { } // NewFile implements kernfs.Inode.NewFile. -func (i *inode) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { +func (i *inode) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*kernfs.Dentry, error) { kernelTask := kernel.TaskFromContext(ctx) if kernelTask == nil { log.Warningf("fusefs.Inode.NewFile: couldn't get kernel task from context", i.nodeID) @@ -432,15 +440,11 @@ func (i *inode) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) }, Name: name, } - d, err := i.newEntry(ctx, name, linux.S_IFREG, linux.FUSE_CREATE, &in) - if err != nil { - return nil, err - } - return d.VFSDentry(), nil + return i.newEntry(ctx, name, linux.S_IFREG, linux.FUSE_CREATE, &in) } // NewNode implements kernfs.Inode.NewNode. -func (i *inode) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) { +func (i *inode) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*kernfs.Dentry, error) { in := linux.FUSEMknodIn{ MknodMeta: linux.FUSEMknodMeta{ Mode: uint32(opts.Mode), @@ -449,28 +453,20 @@ func (i *inode) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) }, Name: name, } - d, err := i.newEntry(ctx, name, opts.Mode.FileType(), linux.FUSE_MKNOD, &in) - if err != nil { - return nil, err - } - return d.VFSDentry(), nil + return i.newEntry(ctx, name, opts.Mode.FileType(), linux.FUSE_MKNOD, &in) } // NewSymlink implements kernfs.Inode.NewSymlink. -func (i *inode) NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) { +func (i *inode) NewSymlink(ctx context.Context, name, target string) (*kernfs.Dentry, error) { in := linux.FUSESymLinkIn{ Name: name, Target: target, } - d, err := i.newEntry(ctx, name, linux.S_IFLNK, linux.FUSE_SYMLINK, &in) - if err != nil { - return nil, err - } - return d.VFSDentry(), nil + return i.newEntry(ctx, name, linux.S_IFLNK, linux.FUSE_SYMLINK, &in) } // Unlink implements kernfs.Inode.Unlink. -func (i *inode) Unlink(ctx context.Context, name string, child *vfs.Dentry) error { +func (i *inode) Unlink(ctx context.Context, name string, child *kernfs.Dentry) error { kernelTask := kernel.TaskFromContext(ctx) if kernelTask == nil { log.Warningf("fusefs.Inode.newEntry: couldn't get kernel task from context", i.nodeID) @@ -493,7 +489,7 @@ func (i *inode) Unlink(ctx context.Context, name string, child *vfs.Dentry) erro } // NewDir implements kernfs.Inode.NewDir. -func (i *inode) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { +func (i *inode) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*kernfs.Dentry, error) { in := linux.FUSEMkdirIn{ MkdirMeta: linux.FUSEMkdirMeta{ Mode: uint32(opts.Mode), @@ -501,15 +497,11 @@ func (i *inode) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) }, Name: name, } - d, err := i.newEntry(ctx, name, linux.S_IFDIR, linux.FUSE_MKDIR, &in) - if err != nil { - return nil, err - } - return d.VFSDentry(), nil + return i.newEntry(ctx, name, linux.S_IFDIR, linux.FUSE_MKDIR, &in) } // RmDir implements kernfs.Inode.RmDir. -func (i *inode) RmDir(ctx context.Context, name string, child *vfs.Dentry) error { +func (i *inode) RmDir(ctx context.Context, name string, child *kernfs.Dentry) error { fusefs := i.fs task, creds := kernel.TaskFromContext(ctx), auth.CredentialsFromContext(ctx) @@ -527,12 +519,7 @@ func (i *inode) RmDir(ctx context.Context, name string, child *vfs.Dentry) error return err } - // TODO(Before merging): When creating new nodes, should we add nodes to the ordered children? - // If so we'll probably need to call this. We will also need to add them with the writable flag when - // appropriate. - // return i.OrderedChildren.RmDir(ctx, name, child) - - return nil + return i.dentry.RemoveChildLocked(name, child) } // newEntry calls FUSE server for entry creation and allocates corresponding entry according to response. @@ -562,11 +549,6 @@ func (i *inode) newEntry(ctx context.Context, name string, fileType linux.FileMo return nil, syserror.EIO } child := i.fs.newInode(out.NodeID, out.Attr) - if opcode == linux.FUSE_LOOKUP { - i.dentry.InsertChildLocked(name, child) - } else { - i.dentry.InsertChild(name, child) - } return child, nil } diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 91d2ae199..18c884b59 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -117,11 +117,12 @@ func (d *dentry) createSyntheticChildLocked(opts *createSyntheticOpts) { d.syntheticChildren++ } +// +stateify savable type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl - mu sync.Mutex + mu sync.Mutex `state:"nosave"` off int64 dirents []vfs.Dirent } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 97b9165cc..94d96261b 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -1416,11 +1416,11 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - if err := d.setStat(ctx, rp.Credentials(), &opts, rp.Mount()); err != nil { - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + err = d.setStat(ctx, rp.Credentials(), &opts, rp.Mount()) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + if err != nil { return err } - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) @@ -1556,11 +1556,11 @@ func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - if err := d.setXattr(ctx, rp.Credentials(), &opts); err != nil { - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + err = d.setXattr(ctx, rp.Credentials(), &opts) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + if err != nil { return err } - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil @@ -1575,11 +1575,11 @@ func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - if err := d.removeXattr(ctx, rp.Credentials(), name); err != nil { - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + err = d.removeXattr(ctx, rp.Credentials(), name) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) + if err != nil { return err } - fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index aaad9c0d9..8608471f8 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -62,9 +62,13 @@ import ( const Name = "9p" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem @@ -77,7 +81,7 @@ type filesystem struct { iopts InternalFilesystemOptions // client is the client used by this filesystem. client is immutable. - client *p9.Client + client *p9.Client `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. // clock is a realtime clock used to set timestamps in file operations. clock ktime.Clock @@ -95,7 +99,7 @@ type filesystem struct { // reference count (such that it is usable as vfs.ResolvingPath.Start() or // is reachable from its children), or if it is a child dentry (such that // it is reachable from its parent). - renameMu sync.RWMutex + renameMu sync.RWMutex `state:"nosave"` // cachedDentries contains all dentries with 0 references. (Due to race // conditions, it may also contain dentries with non-zero references.) @@ -107,7 +111,7 @@ type filesystem struct { // syncableDentries contains all dentries in this filesystem for which // !dentry.file.isNil(). specialFileFDs contains all open specialFileFDs. // These fields are protected by syncMu. - syncMu sync.Mutex + syncMu sync.Mutex `state:"nosave"` syncableDentries map[*dentry]struct{} specialFileFDs map[*specialFileFD]struct{} @@ -120,6 +124,8 @@ type filesystem struct { // dentries, it comes from QID.Path from the 9P server. Synthetic dentries // have have their inodeNumber generated sequentially, with the MSB reserved to // prevent conflicts with regular dentries. +// +// +stateify savable type inodeNumber uint64 // Reserve MSB for synthetic mounts. @@ -132,6 +138,7 @@ func inoFromPath(path uint64) inodeNumber { return inodeNumber(path &^ syntheticInoMask) } +// +stateify savable type filesystemOptions struct { // "Standard" 9P options. fd int @@ -177,6 +184,8 @@ type filesystemOptions struct { // InteropMode controls the client's interaction with other remote filesystem // users. +// +// +stateify savable type InteropMode uint32 const ( @@ -235,6 +244,8 @@ const ( // InternalFilesystemOptions may be passed as // vfs.GetFilesystemOptions.InternalData to FilesystemType.GetFilesystem. +// +// +stateify savable type InternalFilesystemOptions struct { // If LeakConnection is true, do not close the connection to the server // when the Filesystem is released. This is necessary for deployments in @@ -534,6 +545,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -563,7 +576,7 @@ type dentry struct { // If file.isNil(), this dentry represents a synthetic file, i.e. a file // that does not exist on the remote filesystem. As of this writing, the // only files that can be synthetic are sockets, pipes, and directories. - file p9file + file p9file `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. // If deleted is non-zero, the file represented by this dentry has been // deleted. deleted is accessed using atomic memory operations. @@ -575,7 +588,7 @@ type dentry struct { cached bool dentryEntry - dirMu sync.Mutex + dirMu sync.Mutex `state:"nosave"` // If this dentry represents a directory, children contains: // @@ -607,7 +620,7 @@ type dentry struct { // To mutate: // - Lock metadataMu and use atomic operations to update because we might // have atomic readers that don't hold the lock. - metadataMu sync.Mutex + metadataMu sync.Mutex `state:"nosave"` ino inodeNumber // immutable mode uint32 // type is immutable, perms are mutable uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic @@ -638,7 +651,7 @@ type dentry struct { // other metadata fields. nlink uint32 - mapsMu sync.Mutex + mapsMu sync.Mutex `state:"nosave"` // If this dentry represents a regular file, mappings tracks mappings of // the file into memmap.MappingSpaces. mappings is protected by mapsMu. @@ -662,12 +675,12 @@ type dentry struct { // either p9.File transitions from closed (isNil() == true) to open // (isNil() == false), it may be mutated with handleMu locked, but cannot // be closed until the dentry is destroyed. - handleMu sync.RWMutex - readFile p9file - writeFile p9file + handleMu sync.RWMutex `state:"nosave"` + readFile p9file `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. + writeFile p9file `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. hostFD int32 - dataMu sync.RWMutex + dataMu sync.RWMutex `state:"nosave"` // If this dentry represents a regular file that is client-cached, cache // maps offsets into the cached file to offsets into @@ -1627,12 +1640,14 @@ func (d *dentry) decLinks() { // fileDescription is embedded by gofer implementations of // vfs.FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.LockFD - lockLogging sync.Once + lockLogging sync.Once `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. } func (fd *fileDescription) filesystem() *filesystem { diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go index 104157512..a9ebe1206 100644 --- a/pkg/sentry/fsimpl/gofer/handle.go +++ b/pkg/sentry/fsimpl/gofer/handle.go @@ -25,6 +25,8 @@ import ( // handle represents a remote "open file descriptor", consisting of an opened // fid (p9.File) and optionally a host file descriptor. +// +// These are explicitly not savable. type handle struct { file p9file fd int32 // -1 if unavailable diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 24f03ee94..eeaf6e444 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -39,11 +39,12 @@ func (d *dentry) isRegularFile() bool { return d.fileType() == linux.S_IFREG } +// +stateify savable type regularFileFD struct { fileDescription // off is the file offset. off is protected by mu. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` off int64 } @@ -898,6 +899,8 @@ func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { // dentryPlatformFile is only used when a host FD representing the remote file // is available (i.e. dentry.hostFD >= 0), and that FD is used for application // memory mappings (i.e. !filesystem.opts.forcePageCache). +// +// +stateify savable type dentryPlatformFile struct { *dentry @@ -910,7 +913,7 @@ type dentryPlatformFile struct { hostFileMapper fsutil.HostFileMapper // hostFileMapperInitOnce is used to lazily initialize hostFileMapper. - hostFileMapperInitOnce sync.Once + hostFileMapperInitOnce sync.Once `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. } // IncRef implements memmap.File.IncRef. diff --git a/pkg/sentry/fsimpl/gofer/socket.go b/pkg/sentry/fsimpl/gofer/socket.go index 85d2bee72..326b940a7 100644 --- a/pkg/sentry/fsimpl/gofer/socket.go +++ b/pkg/sentry/fsimpl/gofer/socket.go @@ -36,12 +36,14 @@ func (d *dentry) isSocket() bool { // An endpoint's lifetime is the time between when filesystem.BoundEndpointAt() // is called and either BoundEndpoint.BidirectionalConnect or // BoundEndpoint.UnidirectionalConnect is called. +// +// +stateify savable type endpoint struct { // dentry is the filesystem dentry which produced this endpoint. dentry *dentry // file is the p9 file that contains a single unopened fid. - file p9.File + file p9.File `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. // path is the sentry path where this endpoint is bound. path string diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 576c57491..71581736c 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -34,11 +34,13 @@ import ( // special files, and (when filesystemOptions.regularFilesUseSpecialFileFD is // in effect) regular files. specialFileFD differs from regularFileFD by using // per-FD handles instead of shared per-dentry handles, and never buffering I/O. +// +// +stateify savable type specialFileFD struct { fileDescription // handle is used for file I/O. handle is immutable. - handle handle + handle handle `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. // isRegularFile is true if this FD represents a regular file which is only // possible when filesystemOptions.regularFilesUseSpecialFileFD is in @@ -56,7 +58,7 @@ type specialFileFD struct { queue waiter.Queue // If seekable is true, off is the file offset. off is protected by mu. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` off int64 } diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index db8536f26..ffe4ddb32 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -58,7 +58,7 @@ func newInode(fs *filesystem, hostFD int, fileType linux.FileMode, isTTY bool) ( canMap: fileType == linux.S_IFREG, } i.pf.inode = i - i.refs.EnableLeakCheck() + i.EnableLeakCheck() // Non-seekable files can't be memory mapped, assert this. if !i.seekable && i.canMap { @@ -126,7 +126,7 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) // For simplicity, fileDescription.offset is set to 0. Technically, we // should only set to 0 on files that are not seekable (sockets, pipes, // etc.), and use the offset from the host fd otherwise when importing. - return i.open(ctx, d.VFSDentry(), mnt, flags) + return i.open(ctx, d, mnt, flags) } // ImportFD sets up and returns a vfs.FileDescription from a donated fd. @@ -137,6 +137,8 @@ func ImportFD(ctx context.Context, mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs } // filesystemType implements vfs.FilesystemType. +// +// +stateify savable type filesystemType struct{} // GetFilesystem implements vfs.FilesystemType.GetFilesystem. @@ -166,6 +168,8 @@ func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { } // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -185,6 +189,8 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe } // inode implements kernfs.Inode. +// +// +stateify savable type inode struct { kernfs.InodeNoStatFS kernfs.InodeNotDirectory @@ -193,7 +199,7 @@ type inode struct { locks vfs.FileLocks // When the reference count reaches zero, the host fd is closed. - refs inodeRefs + inodeRefs // hostFD contains the host fd that this file was originally created from, // which must be available at time of restore. @@ -233,7 +239,7 @@ type inode struct { canMap bool // mapsMu protects mappings. - mapsMu sync.Mutex + mapsMu sync.Mutex `state:"nosave"` // If canMap is true, mappings tracks mappings of hostFD into // memmap.MappingSpaces. @@ -435,19 +441,9 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre return nil } -// IncRef implements kernfs.Inode.IncRef. -func (i *inode) IncRef() { - i.refs.IncRef() -} - -// TryIncRef implements kernfs.Inode.TryIncRef. -func (i *inode) TryIncRef() bool { - return i.refs.TryIncRef() -} - // DecRef implements kernfs.Inode.DecRef. func (i *inode) DecRef(ctx context.Context) { - i.refs.DecRef(func() { + i.inodeRefs.DecRef(func() { if i.wouldBlock { fdnotifier.RemoveFD(int32(i.hostFD)) } @@ -458,15 +454,15 @@ func (i *inode) DecRef(ctx context.Context) { } // Open implements kernfs.Inode.Open. -func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { // Once created, we cannot re-open a socket fd through /proc/[pid]/fd/. if i.Mode().FileType() == linux.S_IFSOCK { return nil, syserror.ENXIO } - return i.open(ctx, vfsd, rp.Mount(), opts.Flags) + return i.open(ctx, d, rp.Mount(), opts.Flags) } -func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, error) { +func (i *inode) open(ctx context.Context, d *kernfs.Dentry, mnt *vfs.Mount, flags uint32) (*vfs.FileDescription, error) { var s syscall.Stat_t if err := syscall.Fstat(i.hostFD, &s); err != nil { return nil, err @@ -490,7 +486,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u return nil, err } // Currently, we only allow Unix sockets to be imported. - return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d, &i.locks) + return unixsocket.NewFileDescription(ep, ep.Type(), flags, mnt, d.VFSDentry(), &i.locks) case syscall.S_IFREG, syscall.S_IFIFO, syscall.S_IFCHR: if i.isTTY { @@ -500,7 +496,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u } fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd - if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := vfsfd.Init(fd, flags, mnt, d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return vfsfd, nil @@ -509,7 +505,7 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u fd := &fileDescription{inode: i} fd.LockFD.Init(&i.locks) vfsfd := &fd.vfsfd - if err := vfsfd.Init(fd, flags, mnt, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := vfsfd.Init(fd, flags, mnt, d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return vfsfd, nil @@ -521,6 +517,8 @@ func (i *inode) open(ctx context.Context, d *vfs.Dentry, mnt *vfs.Mount, flags u } // fileDescription is embedded by host fd implementations of FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -535,7 +533,7 @@ type fileDescription struct { inode *inode // offsetMu protects offset. - offsetMu sync.Mutex + offsetMu sync.Mutex `state:"nosave"` // offset specifies the current file offset. It is only meaningful when // inode.seekable is true. diff --git a/pkg/sentry/fsimpl/host/mmap.go b/pkg/sentry/fsimpl/host/mmap.go index 65d3af38c..b51a17bed 100644 --- a/pkg/sentry/fsimpl/host/mmap.go +++ b/pkg/sentry/fsimpl/host/mmap.go @@ -27,11 +27,13 @@ import ( // cannot implement both kernfs.Inode.IncRef and memmap.File.IncRef. // // inodePlatformFile should only be used if inode.canMap is true. +// +// +stateify savable type inodePlatformFile struct { *inode // fdRefsMu protects fdRefs. - fdRefsMu sync.Mutex + fdRefsMu sync.Mutex `state:"nosave"` // fdRefs counts references on memmap.File offsets. It is used solely for // memory accounting. @@ -41,7 +43,7 @@ type inodePlatformFile struct { fileMapper fsutil.HostFileMapper // fileMapperInitOnce is used to lazily initialize fileMapper. - fileMapperInitOnce sync.Once + fileMapperInitOnce sync.Once `state:"nosave"` // FIXME(gvisor.dev/issue/1663): not yet supported. } // IncRef implements memmap.File.IncRef. diff --git a/pkg/sentry/fsimpl/host/socket.go b/pkg/sentry/fsimpl/host/socket.go index 131145b85..8a447e29f 100644 --- a/pkg/sentry/fsimpl/host/socket.go +++ b/pkg/sentry/fsimpl/host/socket.go @@ -348,10 +348,10 @@ func (e *SCMConnectedEndpoint) Init() error { func (e *SCMConnectedEndpoint) Release(ctx context.Context) { e.DecRef(func() { e.mu.Lock() + fdnotifier.RemoveFD(int32(e.fd)) if err := syscall.Close(e.fd); err != nil { log.Warningf("Failed to close host fd %d: %v", err) } - fdnotifier.RemoveFD(int32(e.fd)) e.destroyLocked() e.mu.Unlock() }) diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go index e02b9b8f6..f5c596fec 100644 --- a/pkg/sentry/fsimpl/host/tty.go +++ b/pkg/sentry/fsimpl/host/tty.go @@ -30,6 +30,8 @@ import ( // TTYFileDescription implements vfs.FileDescriptionImpl for a host file // descriptor that wraps a TTY FD. +// +// +stateify savable type TTYFileDescription struct { fileDescription diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index 1ee089620..b929118b1 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -56,9 +56,9 @@ func (f *DynamicBytesFile) Init(creds *auth.Credentials, devMajor, devMinor uint } // Open implements Inode.Open. -func (f *DynamicBytesFile) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (f *DynamicBytesFile) Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &DynamicBytesFD{} - if err := fd.Init(rp.Mount(), vfsd, f.data, &f.locks, opts.Flags); err != nil { + if err := fd.Init(rp.Mount(), d, f.data, &f.locks, opts.Flags); err != nil { return nil, err } return &fd.vfsfd, nil @@ -87,12 +87,12 @@ type DynamicBytesFD struct { } // Init initializes a DynamicBytesFD. -func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *vfs.Dentry, data vfs.DynamicBytesSource, locks *vfs.FileLocks, flags uint32) error { +func (fd *DynamicBytesFD) Init(m *vfs.Mount, d *Dentry, data vfs.DynamicBytesSource, locks *vfs.FileLocks, flags uint32) error { fd.LockFD.Init(locks) - if err := fd.vfsfd.Init(fd, flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, flags, m, d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return err } - fd.inode = d.Impl().(*Dentry).inode + fd.inode = d.inode fd.SetDataSource(data) return nil } diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 6518ff5cd..0a4cd4057 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -29,6 +29,8 @@ import ( ) // SeekEndConfig describes the SEEK_END behaviour for FDs. +// +// +stateify savable type SeekEndConfig int // Constants related to SEEK_END behaviour for FDs. @@ -41,6 +43,8 @@ const ( ) // GenericDirectoryFDOptions contains configuration for a GenericDirectoryFD. +// +// +stateify savable type GenericDirectoryFDOptions struct { SeekEnd SeekEndConfig } @@ -56,6 +60,8 @@ type GenericDirectoryFDOptions struct { // Must be initialize with Init before first use. // // Lock ordering: mu => children.mu. +// +// +stateify savable type GenericDirectoryFD struct { vfs.FileDescriptionDefaultImpl vfs.DirectoryFileDescriptionDefaultImpl @@ -68,7 +74,7 @@ type GenericDirectoryFD struct { children *OrderedChildren // mu protects the fields below. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // off is the current directory offset. Protected by "mu". off int64 @@ -76,12 +82,12 @@ type GenericDirectoryFD struct { // NewGenericDirectoryFD creates a new GenericDirectoryFD and returns its // dentry. -func NewGenericDirectoryFD(m *vfs.Mount, d *vfs.Dentry, children *OrderedChildren, locks *vfs.FileLocks, opts *vfs.OpenOptions, fdOpts GenericDirectoryFDOptions) (*GenericDirectoryFD, error) { +func NewGenericDirectoryFD(m *vfs.Mount, d *Dentry, children *OrderedChildren, locks *vfs.FileLocks, opts *vfs.OpenOptions, fdOpts GenericDirectoryFDOptions) (*GenericDirectoryFD, error) { fd := &GenericDirectoryFD{} if err := fd.Init(children, locks, opts, fdOpts); err != nil { return nil, err } - if err := fd.vfsfd.Init(fd, opts.Flags, m, d, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, m, d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return fd, nil @@ -195,8 +201,7 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent // these. childIdx := fd.off - 2 for it := fd.children.nthLocked(childIdx); it != nil; it = it.Next() { - inode := it.Dentry.Impl().(*Dentry).inode - stat, err := inode.Stat(ctx, fd.filesystem(), opts) + stat, err := it.Dentry.inode.Stat(ctx, fd.filesystem(), opts) if err != nil { return err } diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 03bcfb1e2..5cc1c4281 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -37,8 +37,7 @@ import ( // * !rp.Done(). // // Postcondition: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, mayFollowSymlinks bool) (*vfs.Dentry, error) { - d := vfsd.Impl().(*Dentry) +func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, mayFollowSymlinks bool) (*Dentry, error) { if !d.isDir() { return nil, syserror.ENOTDIR } @@ -55,20 +54,20 @@ afterSymlink: // calls d_revalidate(), but walk_component() => handle_dots() does not. if name == "." { rp.Advance() - return vfsd, nil + return d, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(ctx, vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, d.VFSDentry()); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() - return vfsd, nil + return d, nil } - if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, d.parent.VFSDentry()); err != nil { return nil, err } rp.Advance() - return &d.parent.vfsd, nil + return d.parent, nil } if len(name) > linux.NAME_MAX { return nil, syserror.ENAMETOOLONG @@ -79,7 +78,7 @@ afterSymlink: if err != nil { return nil, err } - if err := rp.CheckMount(ctx, &next.vfsd); err != nil { + if err := rp.CheckMount(ctx, next.VFSDentry()); err != nil { return nil, err } // Resolve any symlink at current path component. @@ -102,7 +101,7 @@ afterSymlink: goto afterSymlink } rp.Advance() - return &next.vfsd, nil + return next, nil } // revalidateChildLocked must be called after a call to parent.vfsd.Child(name) @@ -122,7 +121,7 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir if !child.inode.Valid(ctx) { delete(parent.children, name) vfsObj.InvalidateDentry(ctx, &child.vfsd) - fs.deferDecRef(&child.vfsd) // Reference from Lookup. + fs.deferDecRef(child) // Reference from Lookup. child = nil } } @@ -133,7 +132,8 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir if err != nil { return nil, err } - // Reference on childVFSD dropped by a corresponding Valid. + // Reference on c (provided by Lookup) will be dropped when the dentry + // fails validation. parent.InsertChildLocked(name, c) child = c } @@ -148,20 +148,19 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // Preconditions: Filesystem.mu must be locked for at least reading. // // Postconditions: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { - vfsd := rp.Start() +func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingPath) (*Dentry, error) { + d := rp.Start().Impl().(*Dentry) for !rp.Done() { var err error - vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd, true /* mayFollowSymlinks */) + d, err = fs.stepExistingLocked(ctx, rp, d, true /* mayFollowSymlinks */) if err != nil { - return nil, nil, err + return nil, err } } - d := vfsd.Impl().(*Dentry) if rp.MustBeDir() && !d.isDir() { - return nil, nil, syserror.ENOTDIR + return nil, syserror.ENOTDIR } - return vfsd, d.inode, nil + return d, nil } // walkParentDirLocked resolves all but the last path component of rp to an @@ -176,20 +175,19 @@ func (fs *Filesystem) walkExistingLocked(ctx context.Context, rp *vfs.ResolvingP // * !rp.Done(). // // Postconditions: Caller must call fs.processDeferredDecRefs*. -func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, Inode, error) { - vfsd := rp.Start() +func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath) (*Dentry, error) { + d := rp.Start().Impl().(*Dentry) for !rp.Final() { var err error - vfsd, err = fs.stepExistingLocked(ctx, rp, vfsd, true /* mayFollowSymlinks */) + d, err = fs.stepExistingLocked(ctx, rp, d, true /* mayFollowSymlinks */) if err != nil { - return nil, nil, err + return nil, err } } - d := vfsd.Impl().(*Dentry) if !d.isDir() { - return nil, nil, syserror.ENOTDIR + return nil, syserror.ENOTDIR } - return vfsd, d.inode, nil + return d, nil } // checkCreateLocked checks that a file named rp.Component() may be created in @@ -197,10 +195,9 @@ func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.Resolving // // Preconditions: // * Filesystem.mu must be locked for at least reading. -// * parentInode == parentVFSD.Impl().(*Dentry).Inode. // * isDir(parentInode) == true. -func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { +func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parent *Dentry) (string, error) { + if err := parent.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return "", err } pc := rp.Component() @@ -210,11 +207,10 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v if len(pc) > linux.NAME_MAX { return "", syserror.ENAMETOOLONG } - // FIXME(gvisor.dev/issue/1193): Data race due to not holding dirMu. - if _, ok := parentVFSD.Impl().(*Dentry).children[pc]; ok { + if _, ok := parent.children[pc]; ok { return "", syserror.EEXIST } - if parentVFSD.IsDead() { + if parent.VFSDentry().IsDead() { return "", syserror.ENOENT } return pc, nil @@ -223,8 +219,8 @@ func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *v // checkDeleteLocked checks that the file represented by vfsd may be deleted. // // Preconditions: Filesystem.mu must be locked for at least reading. -func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { - parent := vfsd.Impl().(*Dentry).parent +func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry) error { + parent := d.parent if parent == nil { return syserror.EBUSY } @@ -253,11 +249,11 @@ func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) if err != nil { return err } - return inode.CheckPermissions(ctx, creds, ats) + return d.inode.CheckPermissions(ctx, creds, ats) } // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. @@ -265,20 +261,20 @@ func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op fs.mu.RLock() defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) if err != nil { return nil, err } if opts.CheckSearchable { - d := vfsd.Impl().(*Dentry) if !d.isDir() { return nil, syserror.ENOTDIR } - if err := inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } } + vfsd := d.VFSDentry() vfsd.IncRef() // Ownership transferred to caller. return vfsd, nil } @@ -288,12 +284,12 @@ func (fs *Filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa fs.mu.RLock() defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() - vfsd, _, err := fs.walkParentDirLocked(ctx, rp) + d, err := fs.walkParentDirLocked(ctx, rp) if err != nil { return nil, err } - vfsd.IncRef() // Ownership transferred to caller. - return vfsd, nil + d.IncRef() // Ownership transferred to caller. + return d.VFSDentry(), nil } // LinkAt implements vfs.FilesystemImpl.LinkAt. @@ -303,12 +299,15 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. } fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) + + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -325,11 +324,11 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EPERM } - childVFSD, err := parentInode.NewLink(ctx, pc, d.inode) + child, err := parent.inode.NewLink(ctx, pc, d.inode) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) + parent.InsertChildLocked(pc, child) return nil } @@ -340,12 +339,15 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v } fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) + + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -353,14 +355,14 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return err } defer rp.Mount().EndWrite() - childVFSD, err := parentInode.NewDir(ctx, pc, opts) + child, err := parent.inode.NewDir(ctx, pc, opts) if err != nil { if !opts.ForSyntheticMountpoint || err == syserror.EEXIST { return err } - childVFSD = newSyntheticDirectory(rp.Credentials(), opts.Mode) + child = newSyntheticDirectory(rp.Credentials(), opts.Mode) } - parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) + parent.InsertChildLocked(pc, child) return nil } @@ -371,12 +373,15 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v } fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) + + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -384,11 +389,11 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return err } defer rp.Mount().EndWrite() - newVFSD, err := parentInode.NewNode(ctx, pc, opts) + newD, err := parent.inode.NewNode(ctx, pc, opts) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, newVFSD.Impl().(*Dentry)) + parent.InsertChildLocked(pc, newD) return nil } @@ -404,28 +409,27 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // Do not create new file. if opts.Flags&linux.O_CREAT == 0 { fs.mu.RLock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) if err != nil { fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) return nil, err } - if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) return nil, err } - inode.IncRef() - defer inode.DecRef(ctx) + d.inode.IncRef() + defer d.inode.DecRef(ctx) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) - return inode.Open(ctx, rp, vfsd, opts) + return d.inode.Open(ctx, rp, d, opts) } // May create new file. mustCreate := opts.Flags&linux.O_EXCL != 0 - vfsd := rp.Start() - inode := vfsd.Impl().(*Dentry).inode + d := rp.Start().Impl().(*Dentry) fs.mu.Lock() unlocked := false unlock := func() { @@ -442,22 +446,22 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if mustCreate { return nil, syserror.EEXIST } - if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } - inode.IncRef() - defer inode.DecRef(ctx) + d.inode.IncRef() + defer d.inode.DecRef(ctx) unlock() - return inode.Open(ctx, rp, vfsd, opts) + return d.inode.Open(ctx, rp, d, opts) } afterTrailingSymlink: - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return nil, err } // Check for search permission in the parent directory. - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { + if err := parent.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } // Reject attempts to open directories with O_CREAT. @@ -472,10 +476,10 @@ afterTrailingSymlink: return nil, syserror.ENAMETOOLONG } // Determine whether or not we need to create a file. - childVFSD, err := fs.stepExistingLocked(ctx, rp, parentVFSD, false /* mayFollowSymlinks */) + child, err := fs.stepExistingLocked(ctx, rp, parent, false /* mayFollowSymlinks */) if err == syserror.ENOENT { // Already checked for searchability above; now check for writability. - if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { + if err := parent.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } if err := rp.Mount().CheckBeginWrite(); err != nil { @@ -483,16 +487,18 @@ afterTrailingSymlink: } defer rp.Mount().EndWrite() // Create and open the child. - childVFSD, err = parentInode.NewFile(ctx, pc, opts) + child, err := parent.inode.NewFile(ctx, pc, opts) if err != nil { return nil, err } - child := childVFSD.Impl().(*Dentry) - parentVFSD.Impl().(*Dentry).InsertChild(pc, child) + // FIXME(gvisor.dev/issue/1193): Race between checking existence with + // fs.stepExistingLocked and parent.InsertChild. If possible, we should hold + // dirMu from one to the other. + parent.InsertChild(pc, child) child.inode.IncRef() defer child.inode.DecRef(ctx) unlock() - return child.inode.Open(ctx, rp, childVFSD, opts) + return child.inode.Open(ctx, rp, child, opts) } if err != nil { return nil, err @@ -501,7 +507,6 @@ afterTrailingSymlink: if mustCreate { return nil, syserror.EEXIST } - child := childVFSD.Impl().(*Dentry) if rp.ShouldFollowSymlink() && child.isSymlink() { targetVD, targetPathname, err := child.inode.Getlink(ctx, rp.Mount()) if err != nil { @@ -528,22 +533,22 @@ afterTrailingSymlink: child.inode.IncRef() defer child.inode.DecRef(ctx) unlock() - return child.inode.Open(ctx, rp, &child.vfsd, opts) + return child.inode.Open(ctx, rp, child, opts) } // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { fs.mu.RLock() - d, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { return "", err } - if !d.Impl().(*Dentry).isSymlink() { + if !d.isSymlink() { return "", syserror.EINVAL } - return inode.Readlink(ctx, rp.Mount()) + return d.inode.Readlink(ctx, rp.Mount()) } // RenameAt implements vfs.FilesystemImpl.RenameAt. @@ -560,11 +565,10 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // Resolve the destination directory first to verify that it's on this // Mount. - dstDirVFSD, dstDirInode, err := fs.walkParentDirLocked(ctx, rp) + dstDir, err := fs.walkParentDirLocked(ctx, rp) if err != nil { return err } - dstDir := dstDirVFSD.Impl().(*Dentry) mnt := rp.Mount() if mnt != oldParentVD.Mount() { return syserror.EXDEV @@ -582,16 +586,15 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if err != nil { return err } - srcVFSD := &src.vfsd // Can we remove the src dentry? - if err := checkDeleteLocked(ctx, rp, srcVFSD); err != nil { + if err := checkDeleteLocked(ctx, rp, src); err != nil { return err } // Can we create the dst dentry? var dst *Dentry - pc, err := checkCreateLocked(ctx, rp, dstDirVFSD, dstDirInode) + pc, err := checkCreateLocked(ctx, rp, dstDir) switch err { case nil: // Ok, continue with rename as replacement. @@ -602,14 +605,14 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } dst = dstDir.children[pc] if dst == nil { - panic(fmt.Sprintf("Child %q for parent Dentry %+v disappeared inside atomic section?", pc, dstDirVFSD)) + panic(fmt.Sprintf("Child %q for parent Dentry %+v disappeared inside atomic section?", pc, dstDir)) } default: return err } var dstVFSD *vfs.Dentry if dst != nil { - dstVFSD = &dst.vfsd + dstVFSD = dst.VFSDentry() } mntns := vfs.MountNamespaceFromContext(ctx) @@ -625,17 +628,18 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa defer dstDir.dirMu.Unlock() } + srcVFSD := src.VFSDentry() if err := virtfs.PrepareRenameDentry(mntns, srcVFSD, dstVFSD); err != nil { return err } - replaced, err := srcDir.inode.Rename(ctx, src.name, pc, srcVFSD, dstDirVFSD) + replaced, err := srcDir.inode.Rename(ctx, src.name, pc, src, dstDir) if err != nil { virtfs.AbortRenameDentry(srcVFSD, dstVFSD) return err } delete(srcDir.children, src.name) if srcDir != dstDir { - fs.deferDecRef(srcDirVFSD) + fs.deferDecRef(srcDir) dstDir.IncRef() } src.parent = dstDir @@ -644,7 +648,11 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa dstDir.children = make(map[string]*Dentry) } dstDir.children[pc] = src - virtfs.CommitRenameReplaceDentry(ctx, srcVFSD, replaced) + var replaceVFSD *vfs.Dentry + if replaced != nil { + replaceVFSD = replaced.VFSDentry() + } + virtfs.CommitRenameReplaceDentry(ctx, srcVFSD, replaceVFSD) return nil } @@ -653,7 +661,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error fs.mu.Lock() defer fs.mu.Unlock() - vfsd, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err @@ -662,14 +670,13 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } defer rp.Mount().EndWrite() - if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { + if err := checkDeleteLocked(ctx, rp, d); err != nil { return err } - d := vfsd.Impl().(*Dentry) if !d.isDir() { return syserror.ENOTDIR } - if inode.HasChildren() { + if d.inode.HasChildren() { return syserror.ENOTEMPTY } virtfs := rp.VirtualFilesystem() @@ -679,11 +686,12 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error mntns := vfs.MountNamespaceFromContext(ctx) defer mntns.DecRef(ctx) + vfsd := d.VFSDentry() if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } - if err := parentDentry.inode.RmDir(ctx, d.name, vfsd); err != nil { + if err := parentDentry.inode.RmDir(ctx, d.name, d); err != nil { virtfs.AbortDeleteDentry(vfsd) return err } @@ -694,7 +702,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error // SetStatAt implements vfs.FilesystemImpl.SetStatAt. func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { @@ -703,31 +711,31 @@ func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts if opts.Stat.Mask == 0 { return nil } - return inode.SetStat(ctx, fs.VFSFilesystem(), rp.Credentials(), opts) + return d.inode.SetStat(ctx, fs.VFSFilesystem(), rp.Credentials(), opts) } // StatAt implements vfs.FilesystemImpl.StatAt. func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { return linux.Statx{}, err } - return inode.Stat(ctx, fs.VFSFilesystem(), opts) + return d.inode.Stat(ctx, fs.VFSFilesystem(), opts) } // StatFSAt implements vfs.FilesystemImpl.StatFSAt. func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { return linux.Statfs{}, err } - return inode.StatFS(ctx, fs.VFSFilesystem()) + return d.inode.StatFS(ctx, fs.VFSFilesystem()) } // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. @@ -737,12 +745,15 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ } fs.mu.Lock() defer fs.mu.Unlock() - parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) + parent, err := fs.walkParentDirLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } - pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) + parent.dirMu.Lock() + defer parent.dirMu.Unlock() + + pc, err := checkCreateLocked(ctx, rp, parent) if err != nil { return err } @@ -750,11 +761,11 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ return err } defer rp.Mount().EndWrite() - childVFSD, err := parentInode.NewSymlink(ctx, pc, target) + child, err := parent.inode.NewSymlink(ctx, pc, target) if err != nil { return err } - parentVFSD.Impl().(*Dentry).InsertChild(pc, childVFSD.Impl().(*Dentry)) + parent.InsertChildLocked(pc, child) return nil } @@ -763,7 +774,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error fs.mu.Lock() defer fs.mu.Unlock() - vfsd, _, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { return err @@ -772,10 +783,9 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } defer rp.Mount().EndWrite() - if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { + if err := checkDeleteLocked(ctx, rp, d); err != nil { return err } - d := vfsd.Impl().(*Dentry) if d.isDir() { return syserror.EISDIR } @@ -785,10 +795,11 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error defer parentDentry.dirMu.Unlock() mntns := vfs.MountNamespaceFromContext(ctx) defer mntns.DecRef(ctx) + vfsd := d.VFSDentry() if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } - if err := parentDentry.inode.Unlink(ctx, d.name, vfsd); err != nil { + if err := parentDentry.inode.Unlink(ctx, d.name, d); err != nil { virtfs.AbortDeleteDentry(vfsd) return err } @@ -799,13 +810,13 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error // BoundEndpointAt implements vfs.FilesystemImpl.BoundEndpointAt. func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { fs.mu.RLock() - _, inode, err := fs.walkExistingLocked(ctx, rp) + d, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { return nil, err } - if err := inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } return nil, syserror.ECONNREFUSED @@ -814,7 +825,7 @@ func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath // ListXattrAt implements vfs.FilesystemImpl.ListXattrAt. func (fs *Filesystem) ListXattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { @@ -827,7 +838,7 @@ func (fs *Filesystem) ListXattrAt(ctx context.Context, rp *vfs.ResolvingPath, si // GetXattrAt implements vfs.FilesystemImpl.GetXattrAt. func (fs *Filesystem) GetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetXattrOptions) (string, error) { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { @@ -840,7 +851,7 @@ func (fs *Filesystem) GetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // SetXattrAt implements vfs.FilesystemImpl.SetXattrAt. func (fs *Filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetXattrOptions) error { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { @@ -853,7 +864,7 @@ func (fs *Filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // RemoveXattrAt implements vfs.FilesystemImpl.RemoveXattrAt. func (fs *Filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { fs.mu.RLock() - _, _, err := fs.walkExistingLocked(ctx, rp) + _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() fs.processDeferredDecRefs(ctx) if err != nil { diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 1ea4f1c7b..49210e748 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -31,6 +31,8 @@ import ( // count for inodes, performing no extra actions when references are obtained or // released. This is suitable for simple file inodes that don't reference any // resources. +// +// +stateify savable type InodeNoopRefCount struct { } @@ -50,30 +52,32 @@ func (InodeNoopRefCount) TryIncRef() bool { // InodeDirectoryNoNewChildren partially implements the Inode interface. // InodeDirectoryNoNewChildren represents a directory inode which does not // support creation of new children. +// +// +stateify savable type InodeDirectoryNoNewChildren struct{} // NewFile implements Inode.NewFile. -func (InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewFile(context.Context, string, vfs.OpenOptions) (*Dentry, error) { return nil, syserror.EPERM } // NewDir implements Inode.NewDir. -func (InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewDir(context.Context, string, vfs.MkdirOptions) (*Dentry, error) { return nil, syserror.EPERM } // NewLink implements Inode.NewLink. -func (InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewLink(context.Context, string, Inode) (*Dentry, error) { return nil, syserror.EPERM } // NewSymlink implements Inode.NewSymlink. -func (InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewSymlink(context.Context, string, string) (*Dentry, error) { return nil, syserror.EPERM } // NewNode implements Inode.NewNode. -func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { +func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOptions) (*Dentry, error) { return nil, syserror.EPERM } @@ -81,6 +85,8 @@ func (InodeDirectoryNoNewChildren) NewNode(context.Context, string, vfs.MknodOpt // inodeDirectory and inodeDynamicDirectory sub interfaces. Inodes that do not // represent directories can embed this to provide no-op implementations for // directory-related functions. +// +// +stateify savable type InodeNotDirectory struct { } @@ -90,42 +96,42 @@ func (InodeNotDirectory) HasChildren() bool { } // NewFile implements Inode.NewFile. -func (InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewFile(context.Context, string, vfs.OpenOptions) (*Dentry, error) { panic("NewFile called on non-directory inode") } // NewDir implements Inode.NewDir. -func (InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewDir(context.Context, string, vfs.MkdirOptions) (*Dentry, error) { panic("NewDir called on non-directory inode") } // NewLink implements Inode.NewLinkink. -func (InodeNotDirectory) NewLink(context.Context, string, Inode) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewLink(context.Context, string, Inode) (*Dentry, error) { panic("NewLink called on non-directory inode") } // NewSymlink implements Inode.NewSymlink. -func (InodeNotDirectory) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewSymlink(context.Context, string, string) (*Dentry, error) { panic("NewSymlink called on non-directory inode") } // NewNode implements Inode.NewNode. -func (InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { +func (InodeNotDirectory) NewNode(context.Context, string, vfs.MknodOptions) (*Dentry, error) { panic("NewNode called on non-directory inode") } // Unlink implements Inode.Unlink. -func (InodeNotDirectory) Unlink(context.Context, string, *vfs.Dentry) error { +func (InodeNotDirectory) Unlink(context.Context, string, *Dentry) error { panic("Unlink called on non-directory inode") } // RmDir implements Inode.RmDir. -func (InodeNotDirectory) RmDir(context.Context, string, *vfs.Dentry) error { +func (InodeNotDirectory) RmDir(context.Context, string, *Dentry) error { panic("RmDir called on non-directory inode") } // Rename implements Inode.Rename. -func (InodeNotDirectory) Rename(context.Context, string, string, *vfs.Dentry, *vfs.Dentry) (*vfs.Dentry, error) { +func (InodeNotDirectory) Rename(context.Context, string, string, *Dentry, *Dentry) (*Dentry, error) { panic("Rename called on non-directory inode") } @@ -149,6 +155,8 @@ func (InodeNotDirectory) Valid(context.Context) bool { // dymanic entries (i.e. entries that are not "hashed" into the // vfs.Dentry.children) can embed this to provide no-op implementations for // functions related to dynamic entries. +// +// +stateify savable type InodeNoDynamicLookup struct{} // Lookup implements Inode.Lookup. @@ -169,6 +177,8 @@ func (InodeNoDynamicLookup) Valid(ctx context.Context) bool { // InodeNotSymlink partially implements the Inode interface, specifically the // inodeSymlink sub interface. All inodes that are not symlinks may embed this // to return the appropriate errors from symlink-related functions. +// +// +stateify savable type InodeNotSymlink struct{} // Readlink implements Inode.Readlink. @@ -186,6 +196,8 @@ func (InodeNotSymlink) Getlink(context.Context, *vfs.Mount) (vfs.VirtualDentry, // inode attributes. // // Must be initialized by Init prior to first use. +// +// +stateify savable type InodeAttrs struct { devMajor uint32 devMinor uint32 @@ -330,13 +342,16 @@ func (a *InodeAttrs) DecLinks() { } } +// +stateify savable type slot struct { Name string - Dentry *vfs.Dentry + Dentry *Dentry slotEntry } // OrderedChildrenOptions contains initialization options for OrderedChildren. +// +// +stateify savable type OrderedChildrenOptions struct { // Writable indicates whether vfs.FilesystemImpl methods implemented by // OrderedChildren may modify the tracked children. This applies to @@ -352,12 +367,14 @@ type OrderedChildrenOptions struct { // directories. // // Must be initialize with Init before first use. +// +// +stateify savable type OrderedChildren struct { // Can children be modified by user syscalls? It set to false, interface // methods that would modify the children return EPERM. Immutable. writable bool - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` order slotList set map[string]*slot } @@ -390,7 +407,7 @@ func (o *OrderedChildren) Populate(d *Dentry, children map[string]*Dentry) uint3 if child.isDir() { links++ } - if err := o.Insert(name, child.VFSDentry()); err != nil { + if err := o.Insert(name, child); err != nil { panic(fmt.Sprintf("Collision when attempting to insert child %q (%+v) into %+v", name, child, d)) } d.InsertChild(name, child) @@ -407,7 +424,7 @@ func (o *OrderedChildren) HasChildren() bool { // Insert inserts child into o. This ignores the writability of o, as this is // not part of the vfs.FilesystemImpl interface, and is a lower-level operation. -func (o *OrderedChildren) Insert(name string, child *vfs.Dentry) error { +func (o *OrderedChildren) Insert(name string, child *Dentry) error { o.mu.Lock() defer o.mu.Unlock() if _, ok := o.set[name]; ok { @@ -431,10 +448,10 @@ func (o *OrderedChildren) removeLocked(name string) { } // Precondition: caller must hold o.mu for writing. -func (o *OrderedChildren) replaceChildLocked(name string, new *vfs.Dentry) *vfs.Dentry { +func (o *OrderedChildren) replaceChildLocked(name string, new *Dentry) *Dentry { if s, ok := o.set[name]; ok { // Existing slot with given name, simply replace the dentry. - var old *vfs.Dentry + var old *Dentry old, s.Dentry = s.Dentry, new return old } @@ -450,7 +467,7 @@ func (o *OrderedChildren) replaceChildLocked(name string, new *vfs.Dentry) *vfs. } // Precondition: caller must hold o.mu for reading or writing. -func (o *OrderedChildren) checkExistingLocked(name string, child *vfs.Dentry) error { +func (o *OrderedChildren) checkExistingLocked(name string, child *Dentry) error { s, ok := o.set[name] if !ok { return syserror.ENOENT @@ -462,7 +479,7 @@ func (o *OrderedChildren) checkExistingLocked(name string, child *vfs.Dentry) er } // Unlink implements Inode.Unlink. -func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *vfs.Dentry) error { +func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *Dentry) error { if !o.writable { return syserror.EPERM } @@ -478,12 +495,13 @@ func (o *OrderedChildren) Unlink(ctx context.Context, name string, child *vfs.De } // Rmdir implements Inode.Rmdir. -func (o *OrderedChildren) RmDir(ctx context.Context, name string, child *vfs.Dentry) error { +func (o *OrderedChildren) RmDir(ctx context.Context, name string, child *Dentry) error { // We're not responsible for checking that child is a directory, that it's // empty, or updating any link counts; so this is the same as unlink. return o.Unlink(ctx, name, child) } +// +stateify savable type renameAcrossDifferentImplementationsError struct{} func (renameAcrossDifferentImplementationsError) Error() string { @@ -499,8 +517,8 @@ func (renameAcrossDifferentImplementationsError) Error() string { // that will support Rename. // // Postcondition: reference on any replaced dentry transferred to caller. -func (o *OrderedChildren) Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (*vfs.Dentry, error) { - dst, ok := dstDir.Impl().(*Dentry).inode.(interface{}).(*OrderedChildren) +func (o *OrderedChildren) Rename(ctx context.Context, oldname, newname string, child, dstDir *Dentry) (*Dentry, error) { + dst, ok := dstDir.inode.(interface{}).(*OrderedChildren) if !ok { return nil, renameAcrossDifferentImplementationsError{} } @@ -542,12 +560,14 @@ func (o *OrderedChildren) nthLocked(i int64) *slot { } // InodeSymlink partially implements Inode interface for symlinks. +// +// +stateify savable type InodeSymlink struct { InodeNotDirectory } // Open implements Inode.Open. -func (InodeSymlink) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (InodeSymlink) Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return nil, syserror.ELOOP } @@ -574,6 +594,7 @@ var _ Inode = (*StaticDirectory)(nil) func NewStaticDir(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode, children map[string]*Dentry, fdOpts GenericDirectoryFDOptions) *Dentry { inode := &StaticDirectory{} inode.Init(creds, devMajor, devMinor, ino, perm, fdOpts) + inode.EnableLeakCheck() dentry := &Dentry{} dentry.Init(inode) @@ -595,8 +616,8 @@ func (s *StaticDirectory) Init(creds *auth.Credentials, devMajor, devMinor uint3 } // Open implements kernfs.Inode.Open. -func (s *StaticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := NewGenericDirectoryFD(rp.Mount(), vfsd, &s.OrderedChildren, &s.locks, &opts, s.fdOpts) +func (s *StaticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := NewGenericDirectoryFD(rp.Mount(), d, &s.OrderedChildren, &s.locks, &opts, s.fdOpts) if err != nil { return nil, err } @@ -614,6 +635,8 @@ func (s *StaticDirectory) DecRef(context.Context) { } // AlwaysValid partially implements kernfs.inodeDynamicLookup. +// +// +stateify savable type AlwaysValid struct{} // Valid implements kernfs.inodeDynamicLookup.Valid. @@ -623,6 +646,8 @@ func (*AlwaysValid) Valid(context.Context) bool { // InodeNoStatFS partially implements the Inode interface, where the client // filesystem doesn't support statfs(2). +// +// +stateify savable type InodeNoStatFS struct{} // StatFS implements Inode.StatFS. diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 7a63e1410..6d3d79333 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -29,7 +29,7 @@ // // Reference Model: // -// Kernfs dentries represents named pointers to inodes. Dentries and inode have +// Kernfs dentries represents named pointers to inodes. Dentries and inodes have // independent lifetimes and reference counts. A child dentry unconditionally // holds a reference on its parent directory's dentry. A dentry also holds a // reference on the inode it points to. Multiple dentries can point to the same @@ -66,15 +66,17 @@ import ( // Filesystem mostly implements vfs.FilesystemImpl for a generic in-memory // filesystem. Concrete implementations are expected to embed this in their own // Filesystem type. +// +// +stateify savable type Filesystem struct { vfsfs vfs.Filesystem - droppedDentriesMu sync.Mutex + droppedDentriesMu sync.Mutex `state:"nosave"` // droppedDentries is a list of dentries waiting to be DecRef()ed. This is // used to defer dentry destruction until mu can be acquired for // writing. Protected by droppedDentriesMu. - droppedDentries []*vfs.Dentry + droppedDentries []*Dentry // mu synchronizes the lifetime of Dentries on this filesystem. Holding it // for reading guarantees continued existence of any resolved dentries, but @@ -97,7 +99,7 @@ type Filesystem struct { // defer fs.mu.RUnlock() // ... // fs.deferDecRef(dentry) - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` // nextInoMinusOne is used to to allocate inode numbers on this // filesystem. Must be accessed by atomic operations. @@ -108,7 +110,7 @@ type Filesystem struct { // processDeferredDecRefs{,Locked}. See comment on Filesystem.mu. // // Precondition: d must not already be pending destruction. -func (fs *Filesystem) deferDecRef(d *vfs.Dentry) { +func (fs *Filesystem) deferDecRef(d *Dentry) { fs.droppedDentriesMu.Lock() fs.droppedDentries = append(fs.droppedDentries, d) fs.droppedDentriesMu.Unlock() @@ -160,6 +162,8 @@ const ( // to, and child dentries hold a reference on their parent. // // Must be initialized by Init prior to first use. +// +// +stateify savable type Dentry struct { DentryRefs @@ -173,7 +177,11 @@ type Dentry struct { name string // dirMu protects children and the names of child Dentries. - dirMu sync.Mutex + // + // Note that holding fs.mu for writing is not sufficient; + // revalidateChildLocked(), which is a very hot path, may modify children with + // fs.mu acquired for reading only. + dirMu sync.Mutex `state:"nosave"` children map[string]*Dentry inode Inode @@ -240,9 +248,8 @@ func (d *Dentry) Watches() *vfs.Watches { func (d *Dentry) OnZeroWatches(context.Context) {} // InsertChild inserts child into the vfs dentry cache with the given name under -// this dentry. This does not update the directory inode, so calling this on -// its own isn't sufficient to insert a child into a directory. InsertChild -// updates the link count on d if required. +// this dentry. This does not update the directory inode, so calling this on its +// own isn't sufficient to insert a child into a directory. // // Precondition: d must represent a directory inode. func (d *Dentry) InsertChild(name string, child *Dentry) { @@ -254,10 +261,12 @@ func (d *Dentry) InsertChild(name string, child *Dentry) { // InsertChildLocked is equivalent to InsertChild, with additional // preconditions. // -// Precondition: d.dirMu must be locked. +// Preconditions: +// * d must represent a directory inode. +// * d.dirMu must be locked. func (d *Dentry) InsertChildLocked(name string, child *Dentry) { if !d.isDir() { - panic(fmt.Sprintf("InsertChild called on non-directory Dentry: %+v.", d)) + panic(fmt.Sprintf("InsertChildLocked called on non-directory Dentry: %+v.", d)) } d.IncRef() // DecRef in child's Dentry.destroy. child.parent = d @@ -273,7 +282,7 @@ func (d *Dentry) InsertChildLocked(name string, child *Dentry) { // isn't sufficient to remove a child from a directory. // // Precondition: d must represent a directory inode. -func (d *Dentry) RemoveChild(name string, child *vfs.Dentry) error { +func (d *Dentry) RemoveChild(name string, child *Dentry) error { d.dirMu.Lock() defer d.dirMu.Unlock() return d.RemoveChildLocked(name, child) @@ -283,7 +292,7 @@ func (d *Dentry) RemoveChild(name string, child *vfs.Dentry) error { // preconditions. // // Precondition: d.dirMu must be locked. -func (d *Dentry) RemoveChildLocked(name string, child *vfs.Dentry) error { +func (d *Dentry) RemoveChildLocked(name string, child *Dentry) error { if !d.isDir() { panic(fmt.Sprintf("RemoveChild called on non-directory Dentry: %+v.", d)) } @@ -291,7 +300,7 @@ func (d *Dentry) RemoveChildLocked(name string, child *vfs.Dentry) error { if !ok { return syserror.ENOENT } - if &c.vfsd != child { + if c != child { panic(fmt.Sprintf("Dentry hashed into inode doesn't match what vfs thinks! Child: %+v, vfs: %+v", c, child)) } delete(d.children, name) @@ -318,7 +327,6 @@ func (d *Dentry) Inode() Inode { // // - Checking that dentries passed to methods are of the appropriate file type. // - Checking permissions. -// - Updating link and reference counts. // // Specific responsibilities of implementations are documented below. type Inode interface { @@ -328,7 +336,8 @@ type Inode interface { inodeRefs // Methods related to node metadata. A generic implementation is provided by - // InodeAttrs. + // InodeAttrs. Note that a concrete filesystem using kernfs is responsible for + // managing link counts. inodeMetadata // Method for inodes that represent symlink. InodeNotSymlink provides a @@ -346,11 +355,11 @@ type Inode interface { // Open creates a file description for the filesystem object represented by // this inode. The returned file description should hold a reference on the - // inode for its lifetime. + // dentry for its lifetime. // // Precondition: rp.Done(). vfsd.Impl() must be the kernfs Dentry containing // the inode on which Open() is being called. - Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) + Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) // StatFS returns filesystem statistics for the client filesystem. This // corresponds to vfs.FilesystemImpl.StatFSAt. If the client filesystem @@ -400,30 +409,30 @@ type inodeDirectory interface { HasChildren() bool // NewFile creates a new regular file inode. - NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) + NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*Dentry, error) // NewDir creates a new directory inode. - NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) + NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*Dentry, error) // NewLink creates a new hardlink to a specified inode in this // directory. Implementations should create a new kernfs Dentry pointing to // target, and update target's link count. - NewLink(ctx context.Context, name string, target Inode) (*vfs.Dentry, error) + NewLink(ctx context.Context, name string, target Inode) (*Dentry, error) // NewSymlink creates a new symbolic link inode. - NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) + NewSymlink(ctx context.Context, name, target string) (*Dentry, error) // NewNode creates a new filesystem node for a mknod syscall. - NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) + NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*Dentry, error) // Unlink removes a child dentry from this directory inode. - Unlink(ctx context.Context, name string, child *vfs.Dentry) error + Unlink(ctx context.Context, name string, child *Dentry) error // RmDir removes an empty child directory from this directory // inode. Implementations must update the parent directory's link count, // if required. Implementations are not responsible for checking that child // is a directory, checking for an empty directory. - RmDir(ctx context.Context, name string, child *vfs.Dentry) error + RmDir(ctx context.Context, name string, child *Dentry) error // Rename is called on the source directory containing an inode being // renamed. child should point to the resolved child in the source @@ -431,7 +440,7 @@ type inodeDirectory interface { // should return the replaced dentry or nil otherwise. // // Precondition: Caller must serialize concurrent calls to Rename. - Rename(ctx context.Context, oldname, newname string, child, dstDir *vfs.Dentry) (replaced *vfs.Dentry, err error) + Rename(ctx context.Context, oldname, newname string, child, dstDir *Dentry) (replaced *Dentry, err error) } type inodeDynamicLookup interface { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 09806a3f2..e413242dc 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -121,8 +121,8 @@ func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMod return &dir.dentry } -func (d *readonlyDir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (d *readonlyDir) Open(ctx context.Context, rp *vfs.ResolvingPath, kd *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), kd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndStaticEntries, }) if err != nil { @@ -162,8 +162,8 @@ func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, conte return &dir.dentry } -func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, kd *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), kd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndStaticEntries, }) if err != nil { @@ -176,38 +176,36 @@ func (d *dir) DecRef(context.Context) { d.dirRefs.DecRef(d.Destroy) } -func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { +func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*kernfs.Dentry, error) { creds := auth.CredentialsFromContext(ctx) dir := d.fs.newDir(creds, opts.Mode, nil) - dirVFSD := dir.VFSDentry() - if err := d.OrderedChildren.Insert(name, dirVFSD); err != nil { + if err := d.OrderedChildren.Insert(name, dir); err != nil { dir.DecRef(ctx) return nil, err } d.IncLinks(1) - return dirVFSD, nil + return dir, nil } -func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { +func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*kernfs.Dentry, error) { creds := auth.CredentialsFromContext(ctx) f := d.fs.newFile(creds, "") - fVFSD := f.VFSDentry() - if err := d.OrderedChildren.Insert(name, fVFSD); err != nil { + if err := d.OrderedChildren.Insert(name, f); err != nil { f.DecRef(ctx) return nil, err } - return fVFSD, nil + return f, nil } -func (*dir) NewLink(context.Context, string, kernfs.Inode) (*vfs.Dentry, error) { +func (*dir) NewLink(context.Context, string, kernfs.Inode) (*kernfs.Dentry, error) { return nil, syserror.EPERM } -func (*dir) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { +func (*dir) NewSymlink(context.Context, string, string) (*kernfs.Dentry, error) { return nil, syserror.EPERM } -func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { +func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*kernfs.Dentry, error) { return nil, syserror.EPERM } diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 443121c99..58a93eaac 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -24,6 +24,8 @@ import ( // StaticSymlink provides an Inode implementation for symlinks that point to // a immutable target. +// +// +stateify savable type StaticSymlink struct { InodeAttrs InodeNoopRefCount diff --git a/pkg/sentry/fsimpl/kernfs/synthetic_directory.go b/pkg/sentry/fsimpl/kernfs/synthetic_directory.go index 01ba72fa8..ea7f073eb 100644 --- a/pkg/sentry/fsimpl/kernfs/synthetic_directory.go +++ b/pkg/sentry/fsimpl/kernfs/synthetic_directory.go @@ -41,12 +41,12 @@ type syntheticDirectory struct { var _ Inode = (*syntheticDirectory)(nil) -func newSyntheticDirectory(creds *auth.Credentials, perm linux.FileMode) *vfs.Dentry { +func newSyntheticDirectory(creds *auth.Credentials, perm linux.FileMode) *Dentry { inode := &syntheticDirectory{} inode.Init(creds, 0 /* devMajor */, 0 /* devMinor */, 0 /* ino */, perm) d := &Dentry{} d.Init(inode) - return &d.vfsd + return d } func (dir *syntheticDirectory) Init(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64, perm linux.FileMode) { @@ -60,8 +60,8 @@ func (dir *syntheticDirectory) Init(creds *auth.Credentials, devMajor, devMinor } // Open implements Inode.Open. -func (dir *syntheticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := NewGenericDirectoryFD(rp.Mount(), vfsd, &dir.OrderedChildren, &dir.locks, &opts, GenericDirectoryFDOptions{}) +func (dir *syntheticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, d *Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := NewGenericDirectoryFD(rp.Mount(), d, &dir.OrderedChildren, &dir.locks, &opts, GenericDirectoryFDOptions{}) if err != nil { return nil, err } @@ -69,12 +69,12 @@ func (dir *syntheticDirectory) Open(ctx context.Context, rp *vfs.ResolvingPath, } // NewFile implements Inode.NewFile. -func (dir *syntheticDirectory) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*Dentry, error) { return nil, syserror.EPERM } // NewDir implements Inode.NewDir. -func (dir *syntheticDirectory) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*Dentry, error) { if !opts.ForSyntheticMountpoint { return nil, syserror.EPERM } @@ -87,16 +87,16 @@ func (dir *syntheticDirectory) NewDir(ctx context.Context, name string, opts vfs } // NewLink implements Inode.NewLink. -func (dir *syntheticDirectory) NewLink(ctx context.Context, name string, target Inode) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewLink(ctx context.Context, name string, target Inode) (*Dentry, error) { return nil, syserror.EPERM } // NewSymlink implements Inode.NewSymlink. -func (dir *syntheticDirectory) NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewSymlink(ctx context.Context, name, target string) (*Dentry, error) { return nil, syserror.EPERM } // NewNode implements Inode.NewNode. -func (dir *syntheticDirectory) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) { +func (dir *syntheticDirectory) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*Dentry, error) { return nil, syserror.EPERM } diff --git a/pkg/sentry/fsimpl/overlay/directory.go b/pkg/sentry/fsimpl/overlay/directory.go index 7ab42e71e..df4492346 100644 --- a/pkg/sentry/fsimpl/overlay/directory.go +++ b/pkg/sentry/fsimpl/overlay/directory.go @@ -100,12 +100,13 @@ func (d *dentry) collectWhiteoutsForRmdirLocked(ctx context.Context) (map[string return whiteouts, readdirErr } +// +stateify savable type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl - mu sync.Mutex + mu sync.Mutex `state:"nosave"` off int64 dirents []vfs.Dirent } @@ -116,10 +117,12 @@ func (fd *directoryFD) Release(ctx context.Context) { // IterDirents implements vfs.FileDescriptionImpl.IterDirents. func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { + d := fd.dentry() + defer d.InotifyWithParent(ctx, linux.IN_ACCESS, 0, vfs.PathEvent) + fd.mu.Lock() defer fd.mu.Unlock() - d := fd.dentry() if fd.dirents == nil { ds, err := d.getDirents(ctx) if err != nil { diff --git a/pkg/sentry/fsimpl/overlay/filesystem.go b/pkg/sentry/fsimpl/overlay/filesystem.go index e9ce4bde1..bd11372d5 100644 --- a/pkg/sentry/fsimpl/overlay/filesystem.go +++ b/pkg/sentry/fsimpl/overlay/filesystem.go @@ -499,7 +499,13 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if err := create(parent, name, childLayer == lookupLayerUpperWhiteout); err != nil { return err } + parent.dirents = nil + ev := linux.IN_CREATE + if dir { + ev |= linux.IN_ISDIR + } + parent.watches.Notify(ctx, name, uint32(ev), 0 /* cookie */, vfs.InodeEvent, false /* unlinked */) return nil } @@ -631,6 +637,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. } return err } + old.watches.Notify(ctx, "", linux.IN_ATTRIB, 0 /* cookie */, vfs.InodeEvent, false /* unlinked */) return nil }) } @@ -975,6 +982,7 @@ func (fs *filesystem) createAndOpenLocked(ctx context.Context, rp *vfs.Resolving // just can't open it anymore for some reason. return nil, err } + parent.watches.Notify(ctx, childName, linux.IN_CREATE, 0 /* cookie */, vfs.PathEvent, false /* unlinked */) return &fd.vfsfd, nil } @@ -1236,6 +1244,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } } + vfs.InotifyRename(ctx, &renamed.watches, &oldParent.watches, &newParent.watches, oldName, newName, renamed.isDir()) return nil } @@ -1352,6 +1361,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error delete(parent.children, name) ds = appendDentry(ds, child) parent.dirents = nil + parent.watches.Notify(ctx, name, linux.IN_DELETE|linux.IN_ISDIR, 0 /* cookie */, vfs.InodeEvent, true /* unlinked */) return nil } @@ -1359,12 +1369,25 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + return err + } + err = d.setStatLocked(ctx, rp, opts) + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + if err != nil { return err } + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ctx, ev, 0 /* cookie */, vfs.InodeEvent) + } + return nil +} + +// Precondition: d.fs.renameMu must be held for reading. +func (d *dentry) setStatLocked(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { mode := linux.FileMode(atomic.LoadUint32(&d.mode)) if err := vfs.CheckSetStat(ctx, rp.Credentials(), &opts, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { return err @@ -1555,11 +1578,14 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error panic(fmt.Sprintf("unrecoverable overlayfs inconsistency: failed to create whiteout during UnlinkAt: %v", err)) } + var cw *vfs.Watches if child != nil { vfsObj.CommitDeleteDentry(ctx, &child.vfsd) delete(parent.children, name) ds = appendDentry(ds, child) + cw = &child.watches } + vfs.InotifyRemoveChild(ctx, cw, &parent.watches, name) parent.dirents = nil return nil } @@ -1636,13 +1662,20 @@ func (fs *filesystem) getXattr(ctx context.Context, d *dentry, creds *auth.Crede func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetXattrOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) return err } - return fs.setXattrLocked(ctx, d, rp.Mount(), rp.Credentials(), &opts) + err = fs.setXattrLocked(ctx, d, rp.Mount(), rp.Credentials(), &opts) + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + if err != nil { + return err + } + + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0 /* cookie */, vfs.InodeEvent) + return nil } // Precondition: fs.renameMu must be locked. @@ -1673,13 +1706,20 @@ func (fs *filesystem) setXattrLocked(ctx context.Context, d *dentry, mnt *vfs.Mo func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) return err } - return fs.removeXattrLocked(ctx, d, rp.Mount(), rp.Credentials(), name) + err = fs.removeXattrLocked(ctx, d, rp.Mount(), rp.Credentials(), name) + fs.renameMuRUnlockAndCheckDrop(ctx, &ds) + if err != nil { + return err + } + + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0 /* cookie */, vfs.InodeEvent) + return nil } // Precondition: fs.renameMu must be locked. diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index 6e04705c7..853aee951 100644 --- a/pkg/sentry/fsimpl/overlay/non_directory.go +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -39,6 +39,7 @@ func (d *dentry) readlink(ctx context.Context) (string, error) { }) } +// +stateify savable type nonDirectoryFD struct { fileDescription @@ -47,7 +48,7 @@ type nonDirectoryFD struct { // fileDescription.dentry().upperVD. cachedFlags is the last known value of // cachedFD.StatusFlags(). copiedUp, cachedFD, and cachedFlags are // protected by mu. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` copiedUp bool cachedFD *vfs.FileDescription cachedFlags uint32 @@ -183,6 +184,9 @@ func (fd *nonDirectoryFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) return err } d.updateAfterSetStatLocked(&opts) + if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { + d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) + } return nil } diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go index d0d26185e..dfbccd05f 100644 --- a/pkg/sentry/fsimpl/overlay/overlay.go +++ b/pkg/sentry/fsimpl/overlay/overlay.go @@ -51,6 +51,8 @@ import ( const Name = "overlay" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // Name implements vfs.FilesystemType.Name. @@ -60,6 +62,8 @@ func (FilesystemType) Name() string { // FilesystemOptions may be passed as vfs.GetFilesystemOptions.InternalData to // FilesystemType.GetFilesystem. +// +// +stateify savable type FilesystemOptions struct { // Callers passing FilesystemOptions to // overlay.FilesystemType.GetFilesystem() are responsible for ensuring that @@ -76,6 +80,8 @@ type FilesystemOptions struct { } // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem @@ -98,7 +104,7 @@ type filesystem struct { // renameMu synchronizes renaming with non-renaming operations in order to // ensure consistent lock ordering between dentry.dirMu in different // dentries. - renameMu sync.RWMutex + renameMu sync.RWMutex `state:"nosave"` // lastDirIno is the last inode number assigned to a directory. lastDirIno // is accessed using atomic memory operations. @@ -367,6 +373,8 @@ func (fs *filesystem) newDirIno() uint64 { } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -399,7 +407,7 @@ type dentry struct { // and dirents (if not nil) is a cache of dirents as returned by // directoryFDs representing this directory. children is protected by // dirMu. - dirMu sync.Mutex + dirMu sync.Mutex `state:"nosave"` children map[string]*dentry dirents []vfs.Dirent @@ -409,7 +417,7 @@ type dentry struct { // If !upperVD.Ok(), it can transition to a valid vfs.VirtualDentry (i.e. // be copied up) with copyMu locked for writing; otherwise, it is // immutable. lowerVDs is always immutable. - copyMu sync.RWMutex + copyMu sync.RWMutex `state:"nosave"` upperVD vfs.VirtualDentry lowerVDs []vfs.VirtualDentry @@ -454,6 +462,13 @@ type dentry struct { isMappable uint32 locks vfs.FileLocks + + // watches is the set of inotify watches on the file repesented by this dentry. + // + // Note that hard links to the same file will not share the same set of + // watches, due to the fact that we do not have inode structures in this + // overlay implementation. + watches vfs.Watches } // newDentry creates a new dentry. The dentry initially has no references; it @@ -513,6 +528,14 @@ func (d *dentry) checkDropLocked(ctx context.Context) { if atomic.LoadInt64(&d.refs) != 0 { return } + + // Make sure that we do not lose watches on dentries that have not been + // deleted. Note that overlayfs never calls VFS.InvalidateDentry(), so + // d.vfsd.IsDead() indicates that d was deleted. + if !d.vfsd.IsDead() && d.watches.Size() > 0 { + return + } + // Refs is still zero; destroy it. d.destroyLocked(ctx) return @@ -541,6 +564,8 @@ func (d *dentry) destroyLocked(ctx context.Context) { lowerVD.DecRef(ctx) } + d.watches.HandleDeletion(ctx) + if d.parent != nil { d.parent.dirMu.Lock() if !d.vfsd.IsDead() { @@ -559,19 +584,36 @@ func (d *dentry) destroyLocked(ctx context.Context) { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. func (d *dentry) InotifyWithParent(ctx context.Context, events uint32, cookie uint32, et vfs.EventType) { - // TODO(gvisor.dev/issue/1479): Implement inotify. + if d.isDir() { + events |= linux.IN_ISDIR + } + + // overlayfs never calls VFS.InvalidateDentry(), so d.vfsd.IsDead() indicates + // that d was deleted. + deleted := d.vfsd.IsDead() + + d.fs.renameMu.RLock() + // The ordering below is important, Linux always notifies the parent first. + if d.parent != nil { + d.parent.watches.Notify(ctx, d.name, events, cookie, et, deleted) + } + d.watches.Notify(ctx, "", events, cookie, et, deleted) + d.fs.renameMu.RUnlock() } // Watches implements vfs.DentryImpl.Watches. func (d *dentry) Watches() *vfs.Watches { - // TODO(gvisor.dev/issue/1479): Implement inotify. - return nil + return &d.watches } // OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. -// -// TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *dentry) OnZeroWatches(context.Context) {} +func (d *dentry) OnZeroWatches(ctx context.Context) { + if atomic.LoadInt64(&d.refs) == 0 { + d.fs.renameMu.Lock() + d.checkDropLocked(ctx) + d.fs.renameMu.Unlock() + } +} // iterLayers invokes yield on each layer comprising d, from top to bottom. If // any call to yield returns false, iterLayer stops iteration. @@ -652,6 +694,8 @@ func (d *dentry) updateAfterSetStatLocked(opts *vfs.SetStatOptions) { // fileDescription is embedded by overlay implementations of // vfs.FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -679,17 +723,33 @@ func (fd *fileDescription) GetXattr(ctx context.Context, opts vfs.GetXattrOption // SetXattr implements vfs.FileDescriptionImpl.SetXattr. func (fd *fileDescription) SetXattr(ctx context.Context, opts vfs.SetXattrOptions) error { fs := fd.filesystem() + d := fd.dentry() + fs.renameMu.RLock() - defer fs.renameMu.RUnlock() - return fs.setXattrLocked(ctx, fd.dentry(), fd.vfsfd.Mount(), auth.CredentialsFromContext(ctx), &opts) + err := fs.setXattrLocked(ctx, d, fd.vfsfd.Mount(), auth.CredentialsFromContext(ctx), &opts) + fs.renameMu.RUnlock() + if err != nil { + return err + } + + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // RemoveXattr implements vfs.FileDescriptionImpl.RemoveXattr. func (fd *fileDescription) RemoveXattr(ctx context.Context, name string) error { fs := fd.filesystem() + d := fd.dentry() + fs.renameMu.RLock() - defer fs.renameMu.RUnlock() - return fs.removeXattrLocked(ctx, fd.dentry(), fd.vfsfd.Mount(), auth.CredentialsFromContext(ctx), name) + err := fs.removeXattrLocked(ctx, d, fd.vfsfd.Mount(), auth.CredentialsFromContext(ctx), name) + fs.renameMu.RUnlock() + if err != nil { + return err + } + + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) + return nil } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index 7053ad6db..4e2da4810 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type filesystemType struct{} // Name implements vfs.FilesystemType.Name. @@ -43,6 +44,7 @@ func (filesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile panic("pipefs.filesystemType.GetFilesystem should never be called") } +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -76,6 +78,8 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe } // inode implements kernfs.Inode. +// +// +stateify savable type inode struct { kernfs.InodeNotDirectory kernfs.InodeNotSymlink @@ -144,8 +148,8 @@ func (i *inode) SetStat(ctx context.Context, vfsfs *vfs.Filesystem, creds *auth. } // Open implements kernfs.Inode.Open. -func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - return i.pipe.Open(ctx, rp.Mount(), vfsd, opts.Flags, &i.locks) +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + return i.pipe.Open(ctx, rp.Mount(), d.VFSDentry(), opts.Flags, &i.locks) } // StatFS implements kernfs.Inode.StatFS. diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 03b5941b9..05d7948ea 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -41,6 +41,7 @@ func (FilesystemType) Name() string { return Name } +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -84,6 +85,8 @@ func (fs *filesystem) Release(ctx context.Context) { // dynamicInode is an overfitted interface for common Inodes with // dynamicByteSource types used in procfs. +// +// +stateify savable type dynamicInode interface { kernfs.Inode vfs.DynamicBytesSource @@ -99,6 +102,7 @@ func (fs *filesystem) newDentry(creds *auth.Credentials, ino uint64, perm linux. return d } +// +stateify savable type staticFile struct { kernfs.DynamicBytesFile vfs.StaticData @@ -118,10 +122,13 @@ func newStaticDir(creds *auth.Credentials, devMajor, devMinor uint32, ino uint64 // InternalData contains internal data passed in to the procfs mount via // vfs.GetFilesystemOptions.InternalData. +// +// +stateify savable type InternalData struct { Cgroups map[string]string } +// +stateify savable type implStatFS struct{} // StatFS implements kernfs.Inode.StatFS. diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 7277e431e..47ecd941c 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -116,6 +116,7 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb return offset, nil } +// +stateify savable type subtasksFD struct { kernfs.GenericDirectoryFD @@ -154,14 +155,14 @@ func (fd *subtasksFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) erro } // Open implements kernfs.Inode.Open. -func (i *subtasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *subtasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &subtasksFD{task: i.task} if err := fd.Init(&i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }); err != nil { return nil, err } - if err := fd.VFSFileDescription().Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.VFSFileDescription().Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return fd.VFSFileDescription(), nil diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index e24c8a031..a7cd6f57e 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -53,6 +53,7 @@ func (fs *filesystem) newTaskInode(task *kernel.Task, pidns *kernel.PIDNamespace "auxv": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &auxvData{task: task}), "cmdline": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), "comm": fs.newComm(task, fs.NextIno(), 0444), + "cwd": fs.newCwdSymlink(task, fs.NextIno()), "environ": fs.newTaskOwnedFile(task, fs.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), "exe": fs.newExeSymlink(task, fs.NextIno()), "fd": fs.newFDDirInode(task), @@ -107,8 +108,8 @@ func (i *taskInode) Valid(ctx context.Context) bool { } // Open implements kernfs.Inode.Open. -func (i *taskInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *taskInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) if err != nil { @@ -129,6 +130,8 @@ func (i *taskInode) DecRef(context.Context) { // taskOwnedInode implements kernfs.Inode and overrides inode owner with task // effective user and group. +// +// +stateify savable type taskOwnedInode struct { kernfs.Inode diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index f9dda7ad9..0866cea2b 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -51,6 +51,7 @@ func taskFDExists(ctx context.Context, t *kernel.Task, fd int32) bool { return true } +// +stateify savable type fdDir struct { locks vfs.FileLocks @@ -149,8 +150,8 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*kernfs.Dentry, e } // Open implements kernfs.Inode.Open. -func (i *fdDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *fdDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) if err != nil { @@ -286,8 +287,8 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*kernfs.Dentr } // Open implements kernfs.Inode.Open. -func (i *fdInfoDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *fdInfoDirInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) if err != nil { diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 8f7e9b801..3fbf081a6 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -669,18 +669,22 @@ func (fs *filesystem) newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentr // Readlink implements kernfs.Inode.Readlink. func (s *exeSymlink) Readlink(ctx context.Context, _ *vfs.Mount) (string, error) { - if !kernel.ContextCanTrace(ctx, s.task, false) { - return "", syserror.EACCES - } - - // Pull out the executable for /proc/[pid]/exe. - exec, err := s.executable() + exec, _, err := s.Getlink(ctx, nil) if err != nil { return "", err } defer exec.DecRef(ctx) - return exec.PathnameWithDeleted(ctx), nil + root := vfs.RootFromContext(ctx) + if !root.Ok() { + // It could have raced with process deletion. + return "", syserror.ESRCH + } + defer root.DecRef(ctx) + + vfsObj := exec.Mount().Filesystem().VirtualFilesystem() + name, _ := vfsObj.PathnameWithDeleted(ctx, root, exec) + return name, nil } // Getlink implements kernfs.Inode.Getlink. @@ -688,23 +692,12 @@ func (s *exeSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDent if !kernel.ContextCanTrace(ctx, s.task, false) { return vfs.VirtualDentry{}, "", syserror.EACCES } - - exec, err := s.executable() - if err != nil { - return vfs.VirtualDentry{}, "", err - } - defer exec.DecRef(ctx) - - vd := exec.(*fsbridge.VFSFile).FileDescription().VirtualDentry() - vd.IncRef() - return vd, "", nil -} - -func (s *exeSymlink) executable() (file fsbridge.File, err error) { if err := checkTaskState(s.task); err != nil { - return nil, err + return vfs.VirtualDentry{}, "", err } + var err error + var exec fsbridge.File s.task.WithMuLocked(func(t *kernel.Task) { mm := t.MemoryManager() if mm == nil { @@ -715,12 +708,78 @@ func (s *exeSymlink) executable() (file fsbridge.File, err error) { // The MemoryManager may be destroyed, in which case // MemoryManager.destroy will simply set the executable to nil // (with locks held). - file = mm.Executable() - if file == nil { + exec = mm.Executable() + if exec == nil { err = syserror.ESRCH } }) - return + if err != nil { + return vfs.VirtualDentry{}, "", err + } + defer exec.DecRef(ctx) + + vd := exec.(*fsbridge.VFSFile).FileDescription().VirtualDentry() + vd.IncRef() + return vd, "", nil +} + +// cwdSymlink is an symlink for the /proc/[pid]/cwd file. +// +// +stateify savable +type cwdSymlink struct { + implStatFS + kernfs.InodeAttrs + kernfs.InodeNoopRefCount + kernfs.InodeSymlink + + task *kernel.Task +} + +var _ kernfs.Inode = (*cwdSymlink)(nil) + +func (fs *filesystem) newCwdSymlink(task *kernel.Task, ino uint64) *kernfs.Dentry { + inode := &cwdSymlink{task: task} + inode.Init(task.Credentials(), linux.UNNAMED_MAJOR, fs.devMinor, ino, linux.ModeSymlink|0777) + + d := &kernfs.Dentry{} + d.Init(inode) + return d +} + +// Readlink implements kernfs.Inode.Readlink. +func (s *cwdSymlink) Readlink(ctx context.Context, _ *vfs.Mount) (string, error) { + cwd, _, err := s.Getlink(ctx, nil) + if err != nil { + return "", err + } + defer cwd.DecRef(ctx) + + root := vfs.RootFromContext(ctx) + if !root.Ok() { + // It could have raced with process deletion. + return "", syserror.ESRCH + } + defer root.DecRef(ctx) + + vfsObj := cwd.Mount().Filesystem().VirtualFilesystem() + name, _ := vfsObj.PathnameWithDeleted(ctx, root, cwd) + return name, nil +} + +// Getlink implements kernfs.Inode.Getlink. +func (s *cwdSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDentry, string, error) { + if !kernel.ContextCanTrace(ctx, s.task, false) { + return vfs.VirtualDentry{}, "", syserror.EACCES + } + if err := checkTaskState(s.task); err != nil { + return vfs.VirtualDentry{}, "", err + } + cwd := s.task.FSContext().WorkingDirectoryVFS2() + if !cwd.Ok() { + // It could have raced with process deletion. + return vfs.VirtualDentry{}, "", syserror.ESRCH + } + return cwd, "", nil } // mountInfoData is used to implement /proc/[pid]/mountinfo. @@ -785,6 +844,7 @@ func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { return nil } +// +stateify savable type namespaceSymlink struct { kernfs.StaticSymlink @@ -832,6 +892,8 @@ func (s *namespaceSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.Vir // namespaceInode is a synthetic inode created to represent a namespace in // /proc/[pid]/ns/*. +// +// +stateify savable type namespaceInode struct { implStatFS kernfs.InodeAttrs @@ -853,11 +915,11 @@ func (i *namespaceInode) Init(creds *auth.Credentials, devMajor, devMinor uint32 } // Open implements kernfs.Inode.Open. -func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { fd := &namespaceFD{inode: i} i.IncRef() fd.LockFD.Init(&i.locks) - if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, err } return &fd.vfsfd, nil @@ -865,6 +927,8 @@ func (i *namespaceInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd * // namespace FD is a synthetic file that represents a namespace in // /proc/[pid]/ns/*. +// +// +stateify savable type namespaceFD struct { vfs.FileDescriptionDefaultImpl vfs.LockFD diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 1607eac19..e7f748655 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -616,6 +616,7 @@ type netSnmpData struct { var _ dynamicInode = (*netSnmpData)(nil) +// +stateify savable type snmpLine struct { prefix string header string diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index 2ee9b9f6a..d8f5dd509 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -200,8 +200,8 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback } // Open implements kernfs.Inode.Open. -func (i *tasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (i *tasksInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), d, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndZero, }) if err != nil { @@ -236,6 +236,8 @@ func (i *tasksInode) DecRef(context.Context) { // staticFileSetStat implements a special static file that allows inode // attributes to be set. This is to support /proc files that are readonly, but // allow attributes to be set. +// +// +stateify savable type staticFileSetStat struct { dynamicBytesFileSetAttr vfs.StaticData diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 459a8e52e..f268c59b0 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type selfSymlink struct { implStatFS kernfs.InodeAttrs @@ -74,6 +75,7 @@ func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, return syserror.EPERM } +// +stateify savable type threadSelfSymlink struct { implStatFS kernfs.InodeAttrs @@ -121,6 +123,8 @@ func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Creden // dynamicBytesFileSetAttr implements a special file that allows inode // attributes to be set. This is to support /proc files that are readonly, but // allow attributes to be set. +// +// +stateify savable type dynamicBytesFileSetAttr struct { kernfs.DynamicBytesFile } @@ -131,6 +135,8 @@ func (d *dynamicBytesFileSetAttr) SetStat(ctx context.Context, fs *vfs.Filesyste } // cpuStats contains the breakdown of CPU time for /proc/stat. +// +// +stateify savable type cpuStats struct { // user is time spent in userspace tasks with non-positive niceness. user uint64 diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index a3ffbb15e..3312b0418 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type tcpMemDir int const ( diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index f693f9060..6975af5a7 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -67,6 +67,7 @@ var ( taskStaticFiles = map[string]testutil.DirentType{ "auxv": linux.DT_REG, "cgroup": linux.DT_REG, + "cwd": linux.DT_LNK, "cmdline": linux.DT_REG, "comm": linux.DT_REG, "environ": linux.DT_REG, diff --git a/pkg/sentry/fsimpl/signalfd/BUILD b/pkg/sentry/fsimpl/signalfd/BUILD index 067c1657f..adb610213 100644 --- a/pkg/sentry/fsimpl/signalfd/BUILD +++ b/pkg/sentry/fsimpl/signalfd/BUILD @@ -8,7 +8,6 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/context", "//pkg/sentry/kernel", "//pkg/sentry/vfs", diff --git a/pkg/sentry/fsimpl/signalfd/signalfd.go b/pkg/sentry/fsimpl/signalfd/signalfd.go index 3c02af8c9..10f1452ef 100644 --- a/pkg/sentry/fsimpl/signalfd/signalfd.go +++ b/pkg/sentry/fsimpl/signalfd/signalfd.go @@ -16,7 +16,6 @@ package signalfd import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -27,6 +26,8 @@ import ( ) // SignalFileDescription implements vfs.FileDescriptionImpl for signal fds. +// +// +stateify savable type SignalFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -43,7 +44,7 @@ type SignalFileDescription struct { target *kernel.Task // mu protects mask. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // mask is the signal mask. Protected by mu. mask linux.SignalSet @@ -93,8 +94,7 @@ func (sfd *SignalFileDescription) Read(ctx context.Context, dst usermem.IOSequen } // Copy out the signal info using the specified format. - var buf [128]byte - binary.Marshal(buf[:0], usermem.ByteOrder, &linux.SignalfdSiginfo{ + infoNative := linux.SignalfdSiginfo{ Signo: uint32(info.Signo), Errno: info.Errno, Code: info.Code, @@ -103,9 +103,13 @@ func (sfd *SignalFileDescription) Read(ctx context.Context, dst usermem.IOSequen Status: info.Status(), Overrun: uint32(info.Overrun()), Addr: info.Addr(), - }) - n, err := dst.CopyOut(ctx, buf[:]) - return int64(n), err + } + n, err := infoNative.WriteTo(dst.Writer(ctx)) + if err == usermem.ErrEndOfIOSequence { + // Partial copy-out ok. + err = nil + } + return n, err } // Readiness implements waiter.Waitable.Readiness. diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 80b41aa9e..29e5371d6 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -28,6 +28,8 @@ import ( ) // filesystemType implements vfs.FilesystemType. +// +// +stateify savable type filesystemType struct{} // GetFilesystem implements vfs.FilesystemType.GetFilesystem. @@ -44,6 +46,7 @@ func (filesystemType) Name() string { return "sockfs" } +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -80,6 +83,8 @@ func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDe } // inode implements kernfs.Inode. +// +// +stateify savable type inode struct { kernfs.InodeAttrs kernfs.InodeNoopRefCount @@ -88,7 +93,7 @@ type inode struct { } // Open implements kernfs.Inode.Open. -func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { return nil, syserror.ENXIO } diff --git a/pkg/sentry/fsimpl/sys/kcov.go b/pkg/sentry/fsimpl/sys/kcov.go index 73f3d3309..1a6749e53 100644 --- a/pkg/sentry/fsimpl/sys/kcov.go +++ b/pkg/sentry/fsimpl/sys/kcov.go @@ -36,6 +36,8 @@ func (fs *filesystem) newKcovFile(ctx context.Context, creds *auth.Credentials) } // kcovInode implements kernfs.Inode. +// +// +stateify savable type kcovInode struct { kernfs.InodeAttrs kernfs.InodeNoopRefCount @@ -44,7 +46,7 @@ type kcovInode struct { implStatFS } -func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { +func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, d *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { k := kernel.KernelFromContext(ctx) if k == nil { panic("KernelFromContext returned nil") @@ -54,7 +56,7 @@ func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.D kcov: k.NewKcov(), } - if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{ + if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), d.VFSDentry(), &vfs.FileDescriptionOptions{ DenyPRead: true, DenyPWrite: true, }); err != nil { @@ -63,6 +65,7 @@ func (i *kcovInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.D return &fd.vfsfd, nil } +// +stateify savable type kcovFD struct { vfs.FileDescriptionDefaultImpl vfs.NoLockFD @@ -101,7 +104,7 @@ func (fd *kcovFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) erro func (fd *kcovFD) Release(ctx context.Context) { // kcov instances have reference counts in Linux, but this seems sufficient // for our purposes. - fd.kcov.Reset() + fd.kcov.Clear() } // SetStat implements vfs.FileDescriptionImpl.SetStat. diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 39952d2d0..1568c581f 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -34,9 +34,13 @@ const Name = "sysfs" const defaultSysDirMode = linux.FileMode(0755) // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { kernfs.Filesystem @@ -117,6 +121,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // dir implements kernfs.Inode. +// +// +stateify savable type dir struct { dirRefs kernfs.InodeAttrs @@ -148,8 +154,8 @@ func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.Set } // Open implements kernfs.Inode.Open. -func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ +func (d *dir) Open(ctx context.Context, rp *vfs.ResolvingPath, kd *kernfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { + fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), kd, &d.OrderedChildren, &d.locks, &opts, kernfs.GenericDirectoryFDOptions{ SeekEnd: kernfs.SeekEndStaticEntries, }) if err != nil { @@ -169,6 +175,8 @@ func (d *dir) StatFS(ctx context.Context, fs *vfs.Filesystem) (linux.Statfs, err } // cpuFile implements kernfs.Inode. +// +// +stateify savable type cpuFile struct { implStatFS kernfs.DynamicBytesFile @@ -190,6 +198,7 @@ func (fs *filesystem) newCPUFile(creds *auth.Credentials, maxCores uint, mode li return d } +// +stateify savable type implStatFS struct{} // StatFS implements kernfs.Inode.StatFS. diff --git a/pkg/sentry/fsimpl/timerfd/timerfd.go b/pkg/sentry/fsimpl/timerfd/timerfd.go index ac8a4e3bb..8853c8ad2 100644 --- a/pkg/sentry/fsimpl/timerfd/timerfd.go +++ b/pkg/sentry/fsimpl/timerfd/timerfd.go @@ -28,6 +28,8 @@ import ( // TimerFileDescription implements vfs.FileDescriptionImpl for timer fds. It also // implements ktime.TimerListener. +// +// +stateify savable type TimerFileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go index ac54d420d..9129d35b7 100644 --- a/pkg/sentry/fsimpl/tmpfs/device_file.go +++ b/pkg/sentry/fsimpl/tmpfs/device_file.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" ) +// +stateify savable type deviceFile struct { inode inode kind vfs.DeviceKind diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 070c75e68..e90669cf0 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// +stateify savable type directory struct { // Since directories can't be hard-linked, each directory can only be // associated with a single dentry, which we can store in the directory @@ -44,7 +45,7 @@ type directory struct { // (with inode == nil) that represent the iteration position of // directoryFDs. childList is used to support directoryFD.IterDirents() // efficiently. childList is protected by iterMu. - iterMu sync.Mutex + iterMu sync.Mutex `state:"nosave"` childList dentryList } @@ -86,6 +87,7 @@ func (dir *directory) mayDelete(creds *auth.Credentials, child *dentry) error { return vfs.CheckDeleteSticky(creds, linux.FileMode(atomic.LoadUint32(&dir.inode.mode)), auth.KUID(atomic.LoadUint32(&child.inode.uid))) } +// +stateify savable type directoryFD struct { fileDescription vfs.DirectoryFileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 1362c1602..e39cd305b 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -673,11 +673,11 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.mu.RUnlock() return err } - if err := d.inode.setStat(ctx, rp.Credentials(), &opts); err != nil { - fs.mu.RUnlock() + err = d.inode.setStat(ctx, rp.Credentials(), &opts) + fs.mu.RUnlock() + if err != nil { return err } - fs.mu.RUnlock() if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) @@ -822,11 +822,11 @@ func (fs *filesystem) SetXattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.mu.RUnlock() return err } - if err := d.inode.setXattr(rp.Credentials(), &opts); err != nil { - fs.mu.RUnlock() + err = d.inode.setXattr(rp.Credentials(), &opts) + fs.mu.RUnlock() + if err != nil { return err } - fs.mu.RUnlock() d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil @@ -840,11 +840,11 @@ func (fs *filesystem) RemoveXattrAt(ctx context.Context, rp *vfs.ResolvingPath, fs.mu.RUnlock() return err } - if err := d.inode.removeXattr(rp.Credentials(), name); err != nil { - fs.mu.RUnlock() + err = d.inode.removeXattr(rp.Credentials(), name) + fs.mu.RUnlock() + if err != nil { return err } - fs.mu.RUnlock() d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go index 5b0471ff4..d772db9e9 100644 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ b/pkg/sentry/fsimpl/tmpfs/named_pipe.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// +stateify savable type namedPipe struct { inode inode diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index b8699d064..a199eb33d 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -36,6 +36,8 @@ import ( ) // regularFile is a regular (=S_IFREG) tmpfs file. +// +// +stateify savable type regularFile struct { inode inode @@ -66,7 +68,7 @@ type regularFile struct { writableMappingPages uint64 // dataMu protects the fields below. - dataMu sync.RWMutex + dataMu sync.RWMutex `state:"nosave"` // data maps offsets into the file to offsets into memFile that store // the file's data. @@ -325,13 +327,14 @@ func (*regularFile) InvalidateUnsavable(context.Context) error { return nil } +// +stateify savable type regularFileFD struct { fileDescription // off is the file offset. off is accessed using atomic memory operations. // offMu serializes operations that may mutate off. off int64 - offMu sync.Mutex + offMu sync.Mutex `state:"nosave"` } // Release implements vfs.FileDescriptionImpl.Release. diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go index 3ed650474..5699d5975 100644 --- a/pkg/sentry/fsimpl/tmpfs/socket_file.go +++ b/pkg/sentry/fsimpl/tmpfs/socket_file.go @@ -21,6 +21,8 @@ import ( ) // socketFile is a socket (=S_IFSOCK) tmpfs file. +// +// +stateify savable type socketFile struct { inode inode ep transport.BoundEndpoint diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go index b0de5fabe..a102a2ee2 100644 --- a/pkg/sentry/fsimpl/tmpfs/symlink.go +++ b/pkg/sentry/fsimpl/tmpfs/symlink.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) +// +stateify savable type symlink struct { inode inode target string // immutable diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 4658e1533..cefec8fde 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -51,9 +51,13 @@ import ( const Name = "tmpfs" // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem @@ -67,7 +71,7 @@ type filesystem struct { devMinor uint32 // mu serializes changes to the Dentry tree. - mu sync.RWMutex + mu sync.RWMutex `state:"nosave"` nextInoMinusOne uint64 // accessed using atomic memory operations } @@ -78,6 +82,8 @@ func (FilesystemType) Name() string { } // FilesystemOpts is used to pass configuration data to tmpfs. +// +// +stateify savable type FilesystemOpts struct { // RootFileType is the FileType of the filesystem root. Valid values // are: S_IFDIR, S_IFREG, and S_IFLNK. Defaults to S_IFDIR. @@ -221,6 +227,8 @@ var globalStatfs = linux.Statfs{ } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -300,6 +308,8 @@ func (d *dentry) Watches() *vfs.Watches { func (d *dentry) OnZeroWatches(context.Context) {} // inode represents a filesystem object. +// +// +stateify savable type inode struct { // fs is the owning filesystem. fs is immutable. fs *filesystem @@ -316,12 +326,12 @@ type inode struct { // Inode metadata. Writing multiple fields atomically requires holding // mu, othewise atomic operations can be used. - mu sync.Mutex - mode uint32 // file type and mode - nlink uint32 // protected by filesystem.mu instead of inode.mu - uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic - gid uint32 // auth.KGID, but ... - ino uint64 // immutable + mu sync.Mutex `state:"nosave"` + mode uint32 // file type and mode + nlink uint32 // protected by filesystem.mu instead of inode.mu + uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic + gid uint32 // auth.KGID, but ... + ino uint64 // immutable // Linux's tmpfs has no concept of btime. atime int64 // nanoseconds @@ -668,6 +678,8 @@ func (i *inode) checkXattrPermissions(creds *auth.Credentials, name string, ats // fileDescription is embedded by tmpfs implementations of // vfs.FileDescriptionImpl. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl diff --git a/pkg/sentry/fsimpl/verity/BUILD b/pkg/sentry/fsimpl/verity/BUILD index bc8e38431..0ca750281 100644 --- a/pkg/sentry/fsimpl/verity/BUILD +++ b/pkg/sentry/fsimpl/verity/BUILD @@ -1,4 +1,4 @@ -load("//tools:defs.bzl", "go_library") +load("//tools:defs.bzl", "go_library", "go_test") licenses(["notice"]) @@ -26,3 +26,22 @@ go_library( "//pkg/usermem", ], ) + +go_test( + name = "verity_test", + srcs = [ + "verity_test.go", + ], + library = ":verity", + deps = [ + "//pkg/abi/linux", + "//pkg/context", + "//pkg/fspath", + "//pkg/sentry/arch", + "//pkg/sentry/fsimpl/tmpfs", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/contexttest", + "//pkg/sentry/vfs", + "//pkg/usermem", + ], +) diff --git a/pkg/sentry/fsimpl/verity/filesystem.go b/pkg/sentry/fsimpl/verity/filesystem.go index 26b117ca4..a560b0797 100644 --- a/pkg/sentry/fsimpl/verity/filesystem.go +++ b/pkg/sentry/fsimpl/verity/filesystem.go @@ -693,22 +693,24 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf // be called if a verity FD is created successfully. defer merkleWriter.DecRef(ctx) - parentMerkleWriter, err = rp.VirtualFilesystem().OpenAt(ctx, d.fs.creds, &vfs.PathOperation{ - Root: d.parent.lowerMerkleVD, - Start: d.parent.lowerMerkleVD, - }, &vfs.OpenOptions{ - Flags: linux.O_WRONLY | linux.O_APPEND, - }) - if err != nil { - if err == syserror.ENOENT { - parentPath, _ := d.fs.vfsfs.VirtualFilesystem().PathnameWithDeleted(ctx, d.fs.rootDentry.lowerVD, d.parent.lowerVD) - return nil, alertIntegrityViolation(err, fmt.Sprintf("Merkle file for %s expected but not found", parentPath)) + if d.parent != nil { + parentMerkleWriter, err = rp.VirtualFilesystem().OpenAt(ctx, d.fs.creds, &vfs.PathOperation{ + Root: d.parent.lowerMerkleVD, + Start: d.parent.lowerMerkleVD, + }, &vfs.OpenOptions{ + Flags: linux.O_WRONLY | linux.O_APPEND, + }) + if err != nil { + if err == syserror.ENOENT { + parentPath, _ := d.fs.vfsfs.VirtualFilesystem().PathnameWithDeleted(ctx, d.fs.rootDentry.lowerVD, d.parent.lowerVD) + return nil, alertIntegrityViolation(err, fmt.Sprintf("Merkle file for %s expected but not found", parentPath)) + } + return nil, err } - return nil, err + // parentMerkleWriter is cleaned up if any error occurs. IncRef + // will be called if a verity FD is created successfully. + defer parentMerkleWriter.DecRef(ctx) } - // parentMerkleWriter is cleaned up if any error occurs. IncRef - // will be called if a verity FD is created successfully. - defer parentMerkleWriter.DecRef(ctx) } fd := &fileDescription{ diff --git a/pkg/sentry/fsimpl/verity/verity.go b/pkg/sentry/fsimpl/verity/verity.go index 9182df317..fc5eabbca 100644 --- a/pkg/sentry/fsimpl/verity/verity.go +++ b/pkg/sentry/fsimpl/verity/verity.go @@ -74,9 +74,13 @@ var noCrashOnVerificationFailure bool var verityMu sync.RWMutex // FilesystemType implements vfs.FilesystemType. +// +// +stateify savable type FilesystemType struct{} // filesystem implements vfs.FilesystemImpl. +// +// +stateify savable type filesystem struct { vfsfs vfs.Filesystem @@ -101,11 +105,13 @@ type filesystem struct { // renameMu synchronizes renaming with non-renaming operations in order // to ensure consistent lock ordering between dentry.dirMu in different // dentries. - renameMu sync.RWMutex + renameMu sync.RWMutex `state:"nosave"` } // InternalFilesystemOptions may be passed as // vfs.GetFilesystemOptions.InternalData to FilesystemType.GetFilesystem. +// +// +stateify savable type InternalFilesystemOptions struct { // RootMerkleFileName is the name of the verity root Merkle tree file. RootMerkleFileName string @@ -159,6 +165,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // verity, and should not be exposed or connected. mopts := &vfs.MountOptions{ GetFilesystemOptions: iopts.LowerGetFSOptions, + InternalMount: true, } mnt, err := vfsObj.MountDisconnected(ctx, creds, "", iopts.LowerName, mopts) if err != nil { @@ -258,6 +265,8 @@ func (fs *filesystem) Release(ctx context.Context) { } // dentry implements vfs.DentryImpl. +// +// +stateify savable type dentry struct { vfsd vfs.Dentry @@ -284,7 +293,7 @@ type dentry struct { // and dirents (if not nil) is a cache of dirents as returned by // directoryFDs representing this directory. children is protected by // dirMu. - dirMu sync.Mutex + dirMu sync.Mutex `state:"nosave"` children map[string]*dentry // lowerVD is the VirtualDentry in the underlying file system. @@ -428,6 +437,8 @@ func (d *dentry) readlink(ctx context.Context) (string, error) { // FileDescription is a wrapper of the underlying lowerFD, with support to build // Merkle trees through the Linux fs-verity API to verify contents read from // lowerFD. +// +// +stateify savable type fileDescription struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl @@ -547,7 +558,7 @@ func (fd *fileDescription) generateMerkle(ctx context.Context) ([]byte, uint64, // enableVerity enables verity features on fd by generating a Merkle tree file // and stores its root hash in its parent directory's Merkle tree. -func (fd *fileDescription) enableVerity(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (fd *fileDescription) enableVerity(ctx context.Context, uio usermem.IO) (uintptr, error) { if !fd.d.fs.allowRuntimeEnable { return 0, syserror.EPERM } @@ -557,7 +568,11 @@ func (fd *fileDescription) enableVerity(ctx context.Context, uio usermem.IO, arg verityMu.Lock() defer verityMu.Unlock() - if fd.lowerFD == nil || fd.merkleReader == nil || fd.merkleWriter == nil || fd.parentMerkleWriter == nil { + // In allowRuntimeEnable mode, the underlying fd and read/write fd for + // the Merkle tree file should have all been initialized. For any file + // or directory other than the root, the parent Merkle tree file should + // have also been initialized. + if fd.lowerFD == nil || fd.merkleReader == nil || fd.merkleWriter == nil || (fd.parentMerkleWriter == nil && fd.d != fd.d.fs.rootDentry) { return 0, alertIntegrityViolation(syserror.EIO, "Unexpected verity fd: missing expected underlying fds") } @@ -566,26 +581,28 @@ func (fd *fileDescription) enableVerity(ctx context.Context, uio usermem.IO, arg return 0, err } - stat, err := fd.parentMerkleWriter.Stat(ctx, vfs.StatOptions{}) - if err != nil { - return 0, err - } + if fd.parentMerkleWriter != nil { + stat, err := fd.parentMerkleWriter.Stat(ctx, vfs.StatOptions{}) + if err != nil { + return 0, err + } - // Write the root hash of fd to the parent directory's Merkle tree - // file, as it should be part of the parent Merkle tree data. - // parentMerkleWriter is open with O_APPEND, so it should write - // directly to the end of the file. - if _, err = fd.parentMerkleWriter.Write(ctx, usermem.BytesIOSequence(rootHash), vfs.WriteOptions{}); err != nil { - return 0, err - } + // Write the root hash of fd to the parent directory's Merkle + // tree file, as it should be part of the parent Merkle tree + // data. parentMerkleWriter is open with O_APPEND, so it + // should write directly to the end of the file. + if _, err = fd.parentMerkleWriter.Write(ctx, usermem.BytesIOSequence(rootHash), vfs.WriteOptions{}); err != nil { + return 0, err + } - // Record the offset of the root hash of fd in parent directory's - // Merkle tree file. - if err := fd.merkleWriter.SetXattr(ctx, &vfs.SetXattrOptions{ - Name: merkleOffsetInParentXattr, - Value: strconv.Itoa(int(stat.Size)), - }); err != nil { - return 0, err + // Record the offset of the root hash of fd in parent directory's + // Merkle tree file. + if err := fd.merkleWriter.SetXattr(ctx, &vfs.SetXattrOptions{ + Name: merkleOffsetInParentXattr, + Value: strconv.Itoa(int(stat.Size)), + }); err != nil { + return 0, err + } } // Record the size of the data being hashed for fd. @@ -599,7 +616,45 @@ func (fd *fileDescription) enableVerity(ctx context.Context, uio usermem.IO, arg return 0, nil } -func (fd *fileDescription) getFlags(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { +// measureVerity returns the root hash of fd, saved in args[2]. +func (fd *fileDescription) measureVerity(ctx context.Context, uio usermem.IO, verityDigest usermem.Addr) (uintptr, error) { + t := kernel.TaskFromContext(ctx) + var metadata linux.DigestMetadata + + // If allowRuntimeEnable is true, an empty fd.d.rootHash indicates that + // verity is not enabled for the file. If allowRuntimeEnable is false, + // this is an integrity violation because all files should have verity + // enabled, in which case fd.d.rootHash should be set. + if len(fd.d.rootHash) == 0 { + if fd.d.fs.allowRuntimeEnable { + return 0, syserror.ENODATA + } + return 0, alertIntegrityViolation(syserror.ENODATA, "Ioctl measureVerity: no root hash found") + } + + // The first part of VerityDigest is the metadata. + if _, err := metadata.CopyIn(t, verityDigest); err != nil { + return 0, err + } + if metadata.DigestSize < uint16(len(fd.d.rootHash)) { + return 0, syserror.EOVERFLOW + } + + // Populate the output digest size, since DigestSize is both input and + // output. + metadata.DigestSize = uint16(len(fd.d.rootHash)) + + // First copy the metadata. + if _, err := metadata.CopyOut(t, verityDigest); err != nil { + return 0, err + } + + // Now copy the root hash bytes to the memory after metadata. + _, err := t.CopyOutBytes(usermem.Addr(uintptr(verityDigest)+linux.SizeOfDigestMetadata), fd.d.rootHash) + return 0, err +} + +func (fd *fileDescription) verityFlags(ctx context.Context, uio usermem.IO, flags usermem.Addr) (uintptr, error) { f := int32(0) // All enabled files should store a root hash. This flag is not settable @@ -609,8 +664,7 @@ func (fd *fileDescription) getFlags(ctx context.Context, uio usermem.IO, args ar } t := kernel.TaskFromContext(ctx) - addr := args[2].Pointer() - _, err := primitive.CopyInt32Out(t, addr, f) + _, err := primitive.CopyInt32Out(t, flags, f) return 0, err } @@ -618,11 +672,15 @@ func (fd *fileDescription) getFlags(ctx context.Context, uio usermem.IO, args ar func (fd *fileDescription) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArguments) (uintptr, error) { switch cmd := args[1].Uint(); cmd { case linux.FS_IOC_ENABLE_VERITY: - return fd.enableVerity(ctx, uio, args) + return fd.enableVerity(ctx, uio) + case linux.FS_IOC_MEASURE_VERITY: + return fd.measureVerity(ctx, uio, args[2].Pointer()) case linux.FS_IOC_GETFLAGS: - return fd.getFlags(ctx, uio, args) + return fd.verityFlags(ctx, uio, args[2].Pointer()) default: - return fd.lowerFD.Ioctl(ctx, uio, args) + // TODO(b/169682228): Investigate which ioctl commands should + // be allowed. + return 0, syserror.ENOSYS } } @@ -676,10 +734,10 @@ func (fd *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, of // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. func (fd *fileDescription) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error { - return fd.Locks().LockPOSIX(ctx, &fd.vfsfd, uid, t, start, length, whence, block) + return fd.lowerFD.LockPOSIX(ctx, uid, t, start, length, whence, block) } // UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. func (fd *fileDescription) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error { - return fd.Locks().UnlockPOSIX(ctx, &fd.vfsfd, uid, start, length, whence) + return fd.lowerFD.UnlockPOSIX(ctx, uid, start, length, whence) } diff --git a/pkg/sentry/fsimpl/verity/verity_test.go b/pkg/sentry/fsimpl/verity/verity_test.go new file mode 100644 index 000000000..8bcc14131 --- /dev/null +++ b/pkg/sentry/fsimpl/verity/verity_test.go @@ -0,0 +1,429 @@ +// 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 verity + +import ( + "fmt" + "io" + "math/rand" + "testing" + "time" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +// rootMerkleFilename is the name of the root Merkle tree file. +const rootMerkleFilename = "root.verity" + +// maxDataSize is the maximum data size written to the file for test. +const maxDataSize = 100000 + +// 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(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { + rand.Seed(time.Now().UnixNano()) + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(ctx); err != nil { + return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) + } + + vfsObj.MustRegisterFilesystemType("verity", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + + vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + }) + + mntns, err := vfsObj.NewMountNamespace(ctx, auth.CredentialsFromContext(ctx), "", "verity", &vfs.MountOptions{ + GetFilesystemOptions: vfs.GetFilesystemOptions{ + InternalData: InternalFilesystemOptions{ + RootMerkleFileName: rootMerkleFilename, + LowerName: "tmpfs", + AllowRuntimeEnable: true, + NoCrashOnVerificationFailure: true, + }, + }, + }) + if err != nil { + return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("NewMountNamespace: %v", err) + } + root := mntns.Root() + return vfsObj, root, func() { + root.DecRef(ctx) + mntns.DecRef(ctx) + }, 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 + + // Create the file in the underlying file system. + lowerFD, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: lowerRoot, + Start: lowerRoot, + Path: fspath.Parse(filePath), + }, &vfs.OpenOptions{ + Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, + Mode: linux.ModeRegular | mode, + }) + if err != nil { + return nil, 0, err + } + + // Generate random data to be written to the file. + dataSize := rand.Intn(maxDataSize) + 1 + data := make([]byte, dataSize) + rand.Read(data) + + // Write directly to the underlying FD, since verity FD is read-only. + n, err := lowerFD.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) + if err != nil { + return nil, 0, err + } + + if n != int64(len(data)) { + return nil, 0, fmt.Errorf("lowerFD.Write got write length %d, want %d", n, len(data)) + } + + 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, + }) + 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. + randomPos := int64(rand.Intn(size)) + byteToModify := make([]byte, 1) + if _, err := fd.PRead(ctx, usermem.BytesIOSequence(byteToModify), randomPos, vfs.ReadOptions{}); err != nil { + return fmt.Errorf("lowerFD.PRead: %v", err) + } + byteToModify[0] ^= 1 + if _, err := fd.PWrite(ctx, usermem.BytesIOSequence(byteToModify), randomPos, vfs.WriteOptions{}); err != nil { + return fmt.Errorf("lowerFD.PWrite: %v", err) + } + return nil +} + +// TestOpen ensures that when a file is created, the corresponding Merkle tree +// file and the root Merkle tree file exist. +func TestOpen(t *testing.T) { + ctx := contexttest.Context(t) + vfsObj, root, cleanup, err := newVerityRoot(ctx) + if err != nil { + t.Fatalf("newVerityRoot: %v", err) + } + defer cleanup() + + filename := "verity-test-file" + if _, _, err := newFileFD(ctx, vfsObj, root, filename, 0644); 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 { + 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 { + t.Errorf("OpenAt root Merkle tree file %s: %v", merklePrefix+rootMerkleFilename, err) + } +} + +// TestUntouchedFileSucceeds ensures that read from an untouched verity file +// succeeds after enabling verity for it. +func TestReadUntouchedFileSucceeds(t *testing.T) { + ctx := contexttest.Context(t) + vfsObj, root, cleanup, err := newVerityRoot(ctx) + if err != nil { + t.Fatalf("newVerityRoot: %v", err) + } + defer cleanup() + + filename := "verity-test-file" + fd, size, err := newFileFD(ctx, 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) + } + + buf := make([]byte, size) + n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0 /* offset */, vfs.ReadOptions{}) + if err != nil && err != io.EOF { + t.Fatalf("fd.PRead: %v", err) + } + + if n != int64(size) { + t.Errorf("fd.PRead got read length %d, want %d", n, size) + } +} + +// TestReopenUntouchedFileSucceeds ensures that reopen an untouched verity file +// succeeds after enabling verity for it. +func TestReopenUntouchedFileSucceeds(t *testing.T) { + ctx := contexttest.Context(t) + vfsObj, root, cleanup, err := newVerityRoot(ctx) + if err != nil { + t.Fatalf("newVerityRoot: %v", err) + } + defer cleanup() + + filename := "verity-test-file" + fd, _, err := newFileFD(ctx, 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) + } + + // 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 { + t.Errorf("reopen enabled file failed: %v", err) + } +} + +// TestModifiedFileFails ensures that read from a modified verity file fails. +func TestModifiedFileFails(t *testing.T) { + ctx := contexttest.Context(t) + vfsObj, root, cleanup, err := newVerityRoot(ctx) + if err != nil { + t.Fatalf("newVerityRoot: %v", err) + } + defer cleanup() + + filename := "verity-test-file" + fd, size, err := newFileFD(ctx, 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) + } + + // 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, + }) + if err != nil { + t.Fatalf("OpenAt: %v", err) + } + + if err := corruptRandomBit(ctx, lowerFD, size); err != nil { + t.Fatalf("corruptRandomBit: %v", err) + } + + // Confirm that read from the modified file fails. + buf := make([]byte, size) + if _, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0 /* offset */, vfs.ReadOptions{}); err == nil { + t.Fatalf("fd.PRead succeeded with modified file") + } +} + +// TestModifiedMerkleFails ensures that read from a verity file fails if the +// corresponding Merkle tree file is modified. +func TestModifiedMerkleFails(t *testing.T) { + ctx := contexttest.Context(t) + vfsObj, root, cleanup, err := newVerityRoot(ctx) + if err != nil { + t.Fatalf("newVerityRoot: %v", err) + } + defer cleanup() + + filename := "verity-test-file" + fd, size, err := newFileFD(ctx, 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) + } + + // 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, + }) + if err != nil { + t.Fatalf("OpenAt: %v", err) + } + + // Flip a random bit in the Merkle tree file. + stat, err := lowerMerkleFD.Stat(ctx, vfs.StatOptions{}) + if err != nil { + t.Fatalf("stat: %v", err) + } + merkleSize := int(stat.Size) + if err := corruptRandomBit(ctx, lowerMerkleFD, merkleSize); err != nil { + t.Fatalf("corruptRandomBit: %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") + } +} + +// TestModifiedParentMerkleFails ensures that open a verity enabled file in a +// verity enabled directory fails if the hashes related to the target file in +// the parent Merkle tree file is modified. +func TestModifiedParentMerkleFails(t *testing.T) { + ctx := contexttest.Context(t) + vfsObj, root, cleanup, err := newVerityRoot(ctx) + if err != nil { + t.Fatalf("newVerityRoot: %v", err) + } + defer cleanup() + + filename := "verity-test-file" + fd, _, err := newFileFD(ctx, 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) + } + + // 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, + }) + if err != nil { + t.Fatalf("OpenAt: %v", err) + } + + if _, err := parentFD.Ioctl(ctx, nil /* uio */, args); err != nil { + t.Fatalf("Ioctl: %v", err) + } + + // 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, + }) + if err != nil { + t.Fatalf("OpenAt: %v", err) + } + + // Flip a random bit in the parent Merkle tree file. + // This parent directory contains only one child, so any random + // modification in the parent Merkle tree should cause verification + // failure when opening the child file. + stat, err := parentLowerMerkleFD.Stat(ctx, vfs.StatOptions{}) + if err != nil { + t.Fatalf("stat: %v", err) + } + parentMerkleSize := int(stat.Size) + if err := corruptRandomBit(ctx, parentLowerMerkleFD, parentMerkleSize); err != nil { + t.Fatalf("corruptRandomBit: %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 { + t.Errorf("OpenAt file with modified parent Merkle succeeded") + } +} diff --git a/pkg/sentry/hostmm/BUILD b/pkg/sentry/hostmm/BUILD index 61c78569d..300b7ccce 100644 --- a/pkg/sentry/hostmm/BUILD +++ b/pkg/sentry/hostmm/BUILD @@ -7,11 +7,14 @@ go_library( srcs = [ "cgroup.go", "hostmm.go", + "membarrier.go", ], visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/abi/linux", "//pkg/fd", "//pkg/log", "//pkg/usermem", + "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/hostmm/membarrier.go b/pkg/sentry/hostmm/membarrier.go new file mode 100644 index 000000000..4468d75f1 --- /dev/null +++ b/pkg/sentry/hostmm/membarrier.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 hostmm + +import ( + "syscall" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/log" +) + +var ( + haveMembarrierGlobal = false + haveMembarrierPrivateExpedited = false +) + +func init() { + supported, _, e := syscall.RawSyscall(unix.SYS_MEMBARRIER, linux.MEMBARRIER_CMD_QUERY, 0 /* flags */, 0 /* unused */) + if e != 0 { + if e != syscall.ENOSYS { + log.Warningf("membarrier(MEMBARRIER_CMD_QUERY) failed: %s", e.Error()) + } + return + } + // We don't use MEMBARRIER_CMD_GLOBAL_EXPEDITED because this sends IPIs to + // all CPUs running tasks that have previously invoked + // MEMBARRIER_CMD_REGISTER_GLOBAL_EXPEDITED, which presents a DOS risk. + // (MEMBARRIER_CMD_GLOBAL is synchronize_rcu(), i.e. it waits for an RCU + // grace period to elapse without bothering other CPUs. + // MEMBARRIER_CMD_PRIVATE_EXPEDITED sends IPIs only to CPUs running tasks + // sharing the caller's MM.) + if supported&linux.MEMBARRIER_CMD_GLOBAL != 0 { + haveMembarrierGlobal = true + } + if req := uintptr(linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED | linux.MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED); supported&req == req { + if _, _, e := syscall.RawSyscall(unix.SYS_MEMBARRIER, linux.MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED, 0 /* flags */, 0 /* unused */); e != 0 { + log.Warningf("membarrier(MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED) failed: %s", e.Error()) + } else { + haveMembarrierPrivateExpedited = true + } + } +} + +// HaveGlobalMemoryBarrier returns true if GlobalMemoryBarrier is supported. +func HaveGlobalMemoryBarrier() bool { + return haveMembarrierGlobal +} + +// GlobalMemoryBarrier blocks until "all running threads [in the host OS] have +// passed through a state where all memory accesses to user-space addresses +// match program order between entry to and return from [GlobalMemoryBarrier]", +// as for membarrier(2). +// +// Preconditions: HaveGlobalMemoryBarrier() == true. +func GlobalMemoryBarrier() error { + if _, _, e := syscall.Syscall(unix.SYS_MEMBARRIER, linux.MEMBARRIER_CMD_GLOBAL, 0 /* flags */, 0 /* unused */); e != 0 { + return e + } + return nil +} + +// HaveProcessMemoryBarrier returns true if ProcessMemoryBarrier is supported. +func HaveProcessMemoryBarrier() bool { + return haveMembarrierPrivateExpedited +} + +// ProcessMemoryBarrier is equivalent to GlobalMemoryBarrier, but only +// synchronizes with threads sharing a virtual address space (from the host OS' +// perspective) with the calling thread. +// +// Preconditions: HaveProcessMemoryBarrier() == true. +func ProcessMemoryBarrier() error { + if _, _, e := syscall.RawSyscall(unix.SYS_MEMBARRIER, linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED, 0 /* flags */, 0 /* unused */); e != 0 { + return e + } + return nil +} diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index a43c549f1..5de70aecb 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -69,8 +69,8 @@ go_template_instance( prefix = "socket", template = "//pkg/ilist:generic_list", types = { - "Element": "*SocketEntry", - "Linker": "*SocketEntry", + "Element": "*SocketRecordVFS1", + "Linker": "*SocketRecordVFS1", }, ) @@ -204,7 +204,6 @@ go_library( "//pkg/abi", "//pkg/abi/linux", "//pkg/amutex", - "//pkg/binary", "//pkg/bits", "//pkg/bpf", "//pkg/context", diff --git a/pkg/sentry/kernel/kcov.go b/pkg/sentry/kernel/kcov.go index aad63aa99..060c056df 100644 --- a/pkg/sentry/kernel/kcov.go +++ b/pkg/sentry/kernel/kcov.go @@ -89,6 +89,10 @@ func (kcov *Kcov) TaskWork(t *Task) { kcov.mu.Lock() defer kcov.mu.Unlock() + if kcov.mode != linux.KCOV_MODE_TRACE_PC { + return + } + rw := &kcovReadWriter{ mf: kcov.mfp.MemoryFile(), fr: kcov.mappable.FileRange(), @@ -142,7 +146,7 @@ func (kcov *Kcov) InitTrace(size uint64) error { } // EnableTrace performs the KCOV_ENABLE_TRACE ioctl. -func (kcov *Kcov) EnableTrace(ctx context.Context, traceMode uint8) error { +func (kcov *Kcov) EnableTrace(ctx context.Context, traceKind uint8) error { t := TaskFromContext(ctx) if t == nil { panic("kcovInode.EnableTrace() cannot be used outside of a task goroutine") @@ -156,9 +160,9 @@ func (kcov *Kcov) EnableTrace(ctx context.Context, traceMode uint8) error { return syserror.EINVAL } - switch traceMode { + switch traceKind { case linux.KCOV_TRACE_PC: - kcov.mode = traceMode + kcov.mode = linux.KCOV_MODE_TRACE_PC case linux.KCOV_TRACE_CMP: // We do not support KCOV_MODE_TRACE_CMP. return syserror.ENOTSUP @@ -171,6 +175,7 @@ func (kcov *Kcov) EnableTrace(ctx context.Context, traceMode uint8) error { } kcov.owningTask = t + t.SetKcov(kcov) t.RegisterWork(kcov) // Clear existing coverage data; the task expects to read only coverage data @@ -192,26 +197,35 @@ func (kcov *Kcov) DisableTrace(ctx context.Context) error { if t != kcov.owningTask { return syserror.EINVAL } - kcov.owningTask = nil kcov.mode = linux.KCOV_MODE_INIT - kcov.resetLocked() + kcov.owningTask = nil + kcov.mappable = nil return nil } -// Reset is called when the owning task exits. -func (kcov *Kcov) Reset() { +// Clear resets the mode and clears the owning task and memory mapping for kcov. +// It is called when the fd corresponding to kcov is closed. Note that the mode +// needs to be set so that the next call to kcov.TaskWork() will exit early. +func (kcov *Kcov) Clear() { kcov.mu.Lock() - kcov.resetLocked() + kcov.clearLocked() kcov.mu.Unlock() } -// The kcov instance is reset when the owning task exits or when tracing is -// disabled. -func (kcov *Kcov) resetLocked() { +func (kcov *Kcov) clearLocked() { + kcov.mode = linux.KCOV_MODE_INIT kcov.owningTask = nil - if kcov.mappable != nil { - kcov.mappable = nil - } + kcov.mappable = nil +} + +// OnTaskExit is called when the owning task exits. It is similar to +// kcov.Clear(), except the memory mapping is not cleared, so that the same +// mapping can be used in the future if kcov is enabled again by another task. +func (kcov *Kcov) OnTaskExit() { + kcov.mu.Lock() + kcov.mode = linux.KCOV_MODE_INIT + kcov.owningTask = nil + kcov.mu.Unlock() } // ConfigureMMap is called by the vfs.FileDescription for this kcov instance to diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 22f9bb006..d6c21adb7 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -220,13 +220,18 @@ type Kernel struct { // danglingEndpoints is used to save / restore tcpip.DanglingEndpoints. danglingEndpoints struct{} `state:".([]tcpip.Endpoint)"` - // sockets is the list of all network sockets the system. Protected by - // extMu. + // sockets is the list of all network sockets in the system. + // Protected by extMu. + // TODO(gvisor.dev/issue/1624): Only used by VFS1. sockets socketList - // nextSocketEntry is the next entry number to use in sockets. Protected + // socketsVFS2 records all network sockets in the system. Protected by + // extMu. + socketsVFS2 map[*vfs.FileDescription]*SocketRecord + + // nextSocketRecord is the next entry number to use in sockets. Protected // by extMu. - nextSocketEntry uint64 + nextSocketRecord uint64 // deviceRegistry is used to save/restore device.SimpleDevices. deviceRegistry struct{} `state:".(*device.Registry)"` @@ -414,6 +419,8 @@ func (k *Kernel) Init(args InitKernelArgs) error { return fmt.Errorf("failed to create sockfs mount: %v", err) } k.socketMount = socketMount + + k.socketsVFS2 = make(map[*vfs.FileDescription]*SocketRecord) } return nil @@ -507,6 +514,10 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // flushMountSourceRefs flushes the MountSources for all mounted filesystems // and open FDs. func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { + if VFS2Enabled { + return nil // Not relevant. + } + // Flush all mount sources for currently mounted filesystems in each task. flushed := make(map[*fs.MountNamespace]struct{}) k.tasks.mu.RLock() @@ -533,11 +544,6 @@ func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { // // Precondition: Must be called with the kernel paused. func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.FileDescription) error) (err error) { - // TODO(gvisor.dev/issue/1663): Add save support for VFS2. - if VFS2Enabled { - return nil - } - ts.mu.RLock() defer ts.mu.RUnlock() for t := range ts.Root.tids { @@ -556,6 +562,10 @@ func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.Fi func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. + if VFS2Enabled { + return nil + } + return ts.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { if flags := file.Flags(); !flags.Write { return nil @@ -1509,20 +1519,27 @@ func (k *Kernel) SupervisorContext() context.Context { } } -// SocketEntry represents a socket recorded in Kernel.sockets. It implements +// SocketRecord represents a socket recorded in Kernel.socketsVFS2. +// +// +stateify savable +type SocketRecord struct { + k *Kernel + Sock *refs.WeakRef // TODO(gvisor.dev/issue/1624): Only used by VFS1. + SockVFS2 *vfs.FileDescription // Only used by VFS2. + ID uint64 // Socket table entry number. +} + +// SocketRecordVFS1 represents a socket recorded in Kernel.sockets. It implements // refs.WeakRefUser for sockets stored in the socket table. // // +stateify savable -type SocketEntry struct { +type SocketRecordVFS1 struct { socketEntry - k *Kernel - Sock *refs.WeakRef - SockVFS2 *vfs.FileDescription - ID uint64 // Socket table entry number. + SocketRecord } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (s *SocketEntry) WeakRefGone(context.Context) { +func (s *SocketRecordVFS1) WeakRefGone(context.Context) { s.k.extMu.Lock() s.k.sockets.Remove(s) s.k.extMu.Unlock() @@ -1533,9 +1550,14 @@ func (s *SocketEntry) WeakRefGone(context.Context) { // Precondition: Caller must hold a reference to sock. func (k *Kernel) RecordSocket(sock *fs.File) { k.extMu.Lock() - id := k.nextSocketEntry - k.nextSocketEntry++ - s := &SocketEntry{k: k, ID: id} + id := k.nextSocketRecord + k.nextSocketRecord++ + s := &SocketRecordVFS1{ + SocketRecord: SocketRecord{ + k: k, + ID: id, + }, + } s.Sock = refs.NewWeakRef(sock, s) k.sockets.PushBack(s) k.extMu.Unlock() @@ -1547,29 +1569,45 @@ func (k *Kernel) RecordSocket(sock *fs.File) { // Precondition: Caller must hold a reference to sock. // // Note that the socket table will not hold a reference on the -// vfs.FileDescription, because we do not support weak refs on VFS2 files. +// vfs.FileDescription. func (k *Kernel) RecordSocketVFS2(sock *vfs.FileDescription) { k.extMu.Lock() - id := k.nextSocketEntry - k.nextSocketEntry++ - s := &SocketEntry{ + if _, ok := k.socketsVFS2[sock]; ok { + panic(fmt.Sprintf("Socket %p added twice", sock)) + } + id := k.nextSocketRecord + k.nextSocketRecord++ + s := &SocketRecord{ k: k, ID: id, SockVFS2: sock, } - k.sockets.PushBack(s) + k.socketsVFS2[sock] = s + k.extMu.Unlock() +} + +// DeleteSocketVFS2 removes a VFS2 socket from the system-wide socket table. +func (k *Kernel) DeleteSocketVFS2(sock *vfs.FileDescription) { + k.extMu.Lock() + delete(k.socketsVFS2, sock) k.extMu.Unlock() } // ListSockets returns a snapshot of all sockets. // -// Callers of ListSockets() in VFS2 should use SocketEntry.SockVFS2.TryIncRef() +// Callers of ListSockets() in VFS2 should use SocketRecord.SockVFS2.TryIncRef() // to get a reference on a socket in the table. -func (k *Kernel) ListSockets() []*SocketEntry { +func (k *Kernel) ListSockets() []*SocketRecord { k.extMu.Lock() - var socks []*SocketEntry - for s := k.sockets.Front(); s != nil; s = s.Next() { - socks = append(socks, s) + var socks []*SocketRecord + if VFS2Enabled { + for _, s := range k.socketsVFS2 { + socks = append(socks, s) + } + } else { + for s := k.sockets.Front(); s != nil; s = s.Next() { + socks = append(socks, &s.SocketRecord) + } } k.extMu.Unlock() return socks diff --git a/pkg/sentry/kernel/pipe/BUILD b/pkg/sentry/kernel/pipe/BUILD index 449643118..99134e634 100644 --- a/pkg/sentry/kernel/pipe/BUILD +++ b/pkg/sentry/kernel/pipe/BUILD @@ -21,6 +21,7 @@ go_library( "//pkg/amutex", "//pkg/buffer", "//pkg/context", + "//pkg/marshal/primitive", "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/device", diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index c410c96aa..67beb0ad6 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -17,6 +17,7 @@ package pipe import ( "fmt" + "io" "sync/atomic" "syscall" @@ -215,7 +216,7 @@ func (p *Pipe) readLocked(ctx context.Context, ops readOps) (int64, error) { if p.view.Size() == 0 { if !p.HasWriters() { // There are no writers, return EOF. - return 0, nil + return 0, io.EOF } return 0, syserror.ErrWouldBlock } diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index 6d58b682f..f665920cb 100644 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/buffer" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" @@ -145,9 +146,14 @@ func (p *Pipe) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArgume v = math.MaxInt32 // Silently truncate. } // Copy result to userspace. - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{ - AddressSpaceActive: true, - }) + iocc := primitive.IOCopyContext{ + IO: io, + Ctx: ctx, + Opts: usermem.IOOpts{ + AddressSpaceActive: true, + }, + } + _, err := primitive.CopyInt32Out(&iocc, args[2].Pointer(), int32(v)) return 0, err default: return 0, syscall.ENOTTY diff --git a/pkg/sentry/kernel/seccomp.go b/pkg/sentry/kernel/seccomp.go index c38c5a40c..387edfa91 100644 --- a/pkg/sentry/kernel/seccomp.go +++ b/pkg/sentry/kernel/seccomp.go @@ -18,7 +18,6 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/bpf" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/syserror" @@ -27,25 +26,18 @@ import ( const maxSyscallFilterInstructions = 1 << 15 -// seccompData is equivalent to struct seccomp_data, which contains the data -// passed to seccomp-bpf filters. -type seccompData struct { - // nr is the system call number. - nr int32 - - // arch is an AUDIT_ARCH_* value indicating the system call convention. - arch uint32 - - // instructionPointer is the value of the instruction pointer at the time - // of the system call. - instructionPointer uint64 - - // args contains the first 6 system call arguments. - args [6]uint64 -} - -func (d *seccompData) asBPFInput() bpf.Input { - return bpf.InputBytes{binary.Marshal(nil, usermem.ByteOrder, d), usermem.ByteOrder} +// dataAsBPFInput returns a serialized BPF program, only valid on the current task +// goroutine. +// +// Note: this is called for every syscall, which is a very hot path. +func dataAsBPFInput(t *Task, d *linux.SeccompData) bpf.Input { + buf := t.CopyScratchBuffer(d.SizeBytes()) + d.MarshalUnsafe(buf) + return bpf.InputBytes{ + Data: buf, + // Go-marshal always uses the native byte order. + Order: usermem.ByteOrder, + } } func seccompSiginfo(t *Task, errno, sysno int32, ip usermem.Addr) *arch.SignalInfo { @@ -112,20 +104,20 @@ func (t *Task) checkSeccompSyscall(sysno int32, args arch.SyscallArguments, ip u } func (t *Task) evaluateSyscallFilters(sysno int32, args arch.SyscallArguments, ip usermem.Addr) uint32 { - data := seccompData{ - nr: sysno, - arch: t.tc.st.AuditNumber, - instructionPointer: uint64(ip), + data := linux.SeccompData{ + Nr: sysno, + Arch: t.tc.st.AuditNumber, + InstructionPointer: uint64(ip), } // data.args is []uint64 and args is []arch.SyscallArgument (uintptr), so // we can't do any slicing tricks or even use copy/append here. for i, arg := range args { - if i >= len(data.args) { + if i >= len(data.Args) { break } - data.args[i] = arg.Uint64() + data.Args[i] = arg.Uint64() } - input := data.asBPFInput() + input := dataAsBPFInput(t, &data) ret := uint32(linux.SECCOMP_RET_ALLOW) f := t.syscallFilters.Load() diff --git a/pkg/sentry/kernel/signalfd/BUILD b/pkg/sentry/kernel/signalfd/BUILD index 3eb78e91b..76d472292 100644 --- a/pkg/sentry/kernel/signalfd/BUILD +++ b/pkg/sentry/kernel/signalfd/BUILD @@ -8,7 +8,6 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go index b07e1c1bd..78f718cfe 100644 --- a/pkg/sentry/kernel/signalfd/signalfd.go +++ b/pkg/sentry/kernel/signalfd/signalfd.go @@ -17,7 +17,6 @@ package signalfd import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" @@ -103,8 +102,7 @@ func (s *SignalOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS } // Copy out the signal info using the specified format. - var buf [128]byte - binary.Marshal(buf[:0], usermem.ByteOrder, &linux.SignalfdSiginfo{ + infoNative := linux.SignalfdSiginfo{ Signo: uint32(info.Signo), Errno: info.Errno, Code: info.Code, @@ -113,9 +111,13 @@ func (s *SignalOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOS Status: info.Status(), Overrun: uint32(info.Overrun()), Addr: info.Addr(), - }) - n, err := dst.CopyOut(ctx, buf[:]) - return int64(n), err + } + n, err := infoNative.WriteTo(dst.Writer(ctx)) + if err == usermem.ErrEndOfIOSequence { + // Partial copy-out ok. + err = nil + } + return n, err } // Readiness implements waiter.Waitable.Readiness. diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index a436610c9..f796e0fa3 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -917,7 +917,7 @@ func (t *Task) SetKcov(k *Kcov) { // ResetKcov clears the kcov instance associated with t. func (t *Task) ResetKcov() { if t.kcov != nil { - t.kcov.Reset() + t.kcov.OnTaskExit() t.kcov = nil } } diff --git a/pkg/sentry/kernel/task_context.go b/pkg/sentry/kernel/task_context.go index 9fa528384..d1136461a 100644 --- a/pkg/sentry/kernel/task_context.go +++ b/pkg/sentry/kernel/task_context.go @@ -126,7 +126,11 @@ func (t *Task) SyscallTable() *SyscallTable { // Preconditions: The caller must be running on the task goroutine, or t.mu // must be locked. func (t *Task) Stack() *arch.Stack { - return &arch.Stack{t.Arch(), t.MemoryManager(), usermem.Addr(t.Arch().Stack())} + return &arch.Stack{ + Arch: t.Arch(), + IO: t.MemoryManager(), + Bottom: usermem.Addr(t.Arch().Stack()), + } } // LoadTaskImage loads a specified file into a new TaskContext. diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index feaa38596..ebdb83061 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -259,7 +259,11 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) // Set up the signal handler. If we have a saved signal mask, the signal // handler should run with the current mask, but sigreturn should restore // the saved one. - st := &arch.Stack{t.Arch(), mm, sp} + st := &arch.Stack{ + Arch: t.Arch(), + IO: mm, + Bottom: sp, + } mask := t.signalMask if t.haveSavedSignalMask { mask = t.savedSignalMask diff --git a/pkg/sentry/kernel/threads.go b/pkg/sentry/kernel/threads.go index 5ae5906e8..fdadb52c0 100644 --- a/pkg/sentry/kernel/threads.go +++ b/pkg/sentry/kernel/threads.go @@ -265,6 +265,13 @@ func (ns *PIDNamespace) Tasks() []*Task { return tasks } +// NumTasks returns the number of tasks in ns. +func (ns *PIDNamespace) NumTasks() int { + ns.owner.mu.RLock() + defer ns.owner.mu.RUnlock() + return len(ns.tids) +} + // ThreadGroups returns a snapshot of the thread groups in ns. func (ns *PIDNamespace) ThreadGroups() []*ThreadGroup { return ns.ThreadGroupsAppend(nil) diff --git a/pkg/sentry/kernel/vdso.go b/pkg/sentry/kernel/vdso.go index e44a139b3..9bc452e67 100644 --- a/pkg/sentry/kernel/vdso.go +++ b/pkg/sentry/kernel/vdso.go @@ -17,7 +17,6 @@ package kernel import ( "fmt" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" @@ -28,6 +27,8 @@ import ( // // They are exposed to the VDSO via a parameter page managed by VDSOParamPage, // which also includes a sequence counter. +// +// +marshal type vdsoParams struct { monotonicReady uint64 monotonicBaseCycles int64 @@ -68,6 +69,13 @@ type VDSOParamPage struct { // checked in state_test_util tests, causing this field to change across // save / restore. seq uint64 + + // copyScratchBuffer is a temporary buffer used to marshal the params before + // copying it to the real parameter page. The parameter page is typically + // updated at a moderate frequency of ~O(seconds) throughout the lifetime of + // the sentry, so reusing this buffer is a good tradeoff between memory + // usage and the cost of allocation. + copyScratchBuffer []byte } // NewVDSOParamPage returns a VDSOParamPage. @@ -79,7 +87,11 @@ type VDSOParamPage struct { // * VDSOParamPage must be the only writer to fr. // * mfp.MemoryFile().MapInternal(fr) must return a single safemem.Block. func NewVDSOParamPage(mfp pgalloc.MemoryFileProvider, fr memmap.FileRange) *VDSOParamPage { - return &VDSOParamPage{mfp: mfp, fr: fr} + return &VDSOParamPage{ + mfp: mfp, + fr: fr, + copyScratchBuffer: make([]byte, (*vdsoParams)(nil).SizeBytes()), + } } // access returns a mapping of the param page. @@ -133,7 +145,8 @@ func (v *VDSOParamPage) Write(f func() vdsoParams) error { // Get the new params. p := f() - buf := binary.Marshal(nil, usermem.ByteOrder, p) + buf := v.copyScratchBuffer[:p.SizeBytes()] + p.MarshalUnsafe(buf) // Skip the sequence counter. if _, err := safemem.Copy(paramPage.DropFirst(8), safemem.BlockFromSafeSlice(buf)); err != nil { diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index 15c88aa7c..c69b62db9 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -122,7 +122,7 @@ func allocStack(ctx context.Context, m *mm.MemoryManager, a arch.Context) (*arch if err != nil { return nil, err } - return &arch.Stack{a, m, ar.End}, nil + return &arch.Stack{Arch: a, IO: m, Bottom: ar.End}, nil } const ( @@ -247,20 +247,20 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V } // Push the original filename to the stack, for AT_EXECFN. - execfn, err := stack.Push(args.Filename) - if err != nil { + if _, err := stack.PushNullTerminatedByteSlice([]byte(args.Filename)); err != nil { return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("Failed to push exec filename: %v", err), syserr.FromError(err).ToLinux()) } + execfn := stack.Bottom // Push 16 random bytes on the stack which AT_RANDOM will point to. var b [16]byte if _, err := rand.Read(b[:]); err != nil { return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("Failed to read random bytes: %v", err), syserr.FromError(err).ToLinux()) } - random, err := stack.Push(b) - if err != nil { + if _, err = stack.PushNullTerminatedByteSlice(b[:]); err != nil { return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("Failed to push random bytes: %v", err), syserr.FromError(err).ToLinux()) } + random := stack.Bottom c := auth.CredentialsFromContext(ctx) diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go index a44fa2b95..7fd77925f 100644 --- a/pkg/sentry/memmap/memmap.go +++ b/pkg/sentry/memmap/memmap.go @@ -127,7 +127,7 @@ func (t Translation) FileRange() FileRange { // Preconditions: Same as Mappable.Translate. func CheckTranslateResult(required, optional MappableRange, at usermem.AccessType, ts []Translation, terr error) error { // Verify that the inputs to Mappable.Translate were valid. - if !required.WellFormed() || required.Length() <= 0 { + if !required.WellFormed() || required.Length() == 0 { panic(fmt.Sprintf("invalid required range: %v", required)) } if !usermem.Addr(required.Start).IsPageAligned() || !usermem.Addr(required.End).IsPageAligned() { @@ -145,7 +145,7 @@ func CheckTranslateResult(required, optional MappableRange, at usermem.AccessTyp return fmt.Errorf("first Translation %+v does not cover start of required range %v", ts[0], required) } for i, t := range ts { - if !t.Source.WellFormed() || t.Source.Length() <= 0 { + if !t.Source.WellFormed() || t.Source.Length() == 0 { return fmt.Errorf("Translation %+v has invalid Source", t) } if !usermem.Addr(t.Source.Start).IsPageAligned() || !usermem.Addr(t.Source.End).IsPageAligned() { diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index 8c9f11cce..92cc87d84 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -235,6 +235,20 @@ type MemoryManager struct { // vdsoSigReturnAddr is the address of 'vdso_sigreturn'. vdsoSigReturnAddr uint64 + + // membarrierPrivateEnabled is non-zero if EnableMembarrierPrivate has + // previously been called. Since, as of this writing, + // MEMBARRIER_CMD_PRIVATE_EXPEDITED is implemented as a global memory + // barrier, membarrierPrivateEnabled has no other effect. + // + // membarrierPrivateEnabled is accessed using atomic memory operations. + membarrierPrivateEnabled uint32 + + // membarrierRSeqEnabled is non-zero if EnableMembarrierRSeq has previously + // been called. + // + // membarrierRSeqEnabled is accessed using atomic memory operations. + membarrierRSeqEnabled uint32 } // vma represents a virtual memory area. diff --git a/pkg/sentry/mm/pma.go b/pkg/sentry/mm/pma.go index 30facebf7..7e5f7de64 100644 --- a/pkg/sentry/mm/pma.go +++ b/pkg/sentry/mm/pma.go @@ -36,7 +36,7 @@ import ( // * ar.Length() != 0. func (mm *MemoryManager) existingPMAsLocked(ar usermem.AddrRange, at usermem.AccessType, ignorePermissions bool, needInternalMappings bool) pmaIterator { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 { + if !ar.WellFormed() || ar.Length() == 0 { panic(fmt.Sprintf("invalid ar: %v", ar)) } } @@ -100,7 +100,7 @@ func (mm *MemoryManager) existingVecPMAsLocked(ars usermem.AddrRangeSeq, at user // (i.e. permission checks must have been performed against vmas). func (mm *MemoryManager) getPMAsLocked(ctx context.Context, vseg vmaIterator, ar usermem.AddrRange, at usermem.AccessType) (pmaIterator, pmaGapIterator, error) { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 { + if !ar.WellFormed() || ar.Length() == 0 { panic(fmt.Sprintf("invalid ar: %v", ar)) } if !vseg.Ok() { @@ -193,7 +193,7 @@ func (mm *MemoryManager) getVecPMAsLocked(ctx context.Context, ars usermem.AddrR // getVecPMAsLocked; other clients should call one of those instead. func (mm *MemoryManager) getPMAsInternalLocked(ctx context.Context, vseg vmaIterator, ar usermem.AddrRange, at usermem.AccessType) (pmaIterator, pmaGapIterator, error) { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() { + if !ar.WellFormed() || ar.Length() == 0 || !ar.IsPageAligned() { panic(fmt.Sprintf("invalid ar: %v", ar)) } if !vseg.Ok() { @@ -223,7 +223,7 @@ func (mm *MemoryManager) getPMAsInternalLocked(ctx context.Context, vseg vmaIter // Need a pma here. optAR := vseg.Range().Intersect(pgap.Range()) if checkInvariants { - if optAR.Length() <= 0 { + if optAR.Length() == 0 { panic(fmt.Sprintf("vseg %v and pgap %v do not overlap", vseg, pgap)) } } @@ -560,7 +560,7 @@ func (mm *MemoryManager) isPMACopyOnWriteLocked(vseg vmaIterator, pseg pmaIterat // Invalidate implements memmap.MappingSpace.Invalidate. func (mm *MemoryManager) Invalidate(ar usermem.AddrRange, opts memmap.InvalidateOpts) { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() { + if !ar.WellFormed() || ar.Length() == 0 || !ar.IsPageAligned() { panic(fmt.Sprintf("invalid ar: %v", ar)) } } @@ -583,7 +583,7 @@ func (mm *MemoryManager) Invalidate(ar usermem.AddrRange, opts memmap.Invalidate // * ar must be page-aligned. func (mm *MemoryManager) invalidateLocked(ar usermem.AddrRange, invalidatePrivate, invalidateShared bool) { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() { + if !ar.WellFormed() || ar.Length() == 0 || !ar.IsPageAligned() { panic(fmt.Sprintf("invalid ar: %v", ar)) } } @@ -629,7 +629,7 @@ func (mm *MemoryManager) invalidateLocked(ar usermem.AddrRange, invalidatePrivat // * ar must be page-aligned. func (mm *MemoryManager) Pin(ctx context.Context, ar usermem.AddrRange, at usermem.AccessType, ignorePermissions bool) ([]PinnedRange, error) { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() { + if !ar.WellFormed() || ar.Length() == 0 || !ar.IsPageAligned() { panic(fmt.Sprintf("invalid ar: %v", ar)) } } @@ -715,10 +715,10 @@ func Unpin(prs []PinnedRange) { // * oldAR and newAR must be page-aligned. func (mm *MemoryManager) movePMAsLocked(oldAR, newAR usermem.AddrRange) { if checkInvariants { - if !oldAR.WellFormed() || oldAR.Length() <= 0 || !oldAR.IsPageAligned() { + if !oldAR.WellFormed() || oldAR.Length() == 0 || !oldAR.IsPageAligned() { panic(fmt.Sprintf("invalid oldAR: %v", oldAR)) } - if !newAR.WellFormed() || newAR.Length() <= 0 || !newAR.IsPageAligned() { + if !newAR.WellFormed() || newAR.Length() == 0 || !newAR.IsPageAligned() { panic(fmt.Sprintf("invalid newAR: %v", newAR)) } if oldAR.Length() > newAR.Length() { @@ -778,7 +778,7 @@ func (mm *MemoryManager) movePMAsLocked(oldAR, newAR usermem.AddrRange) { // into mm.pmas. func (mm *MemoryManager) getPMAInternalMappingsLocked(pseg pmaIterator, ar usermem.AddrRange) (pmaGapIterator, error) { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 { + if !ar.WellFormed() || ar.Length() == 0 { panic(fmt.Sprintf("invalid ar: %v", ar)) } if !pseg.Range().Contains(ar.Start) { @@ -831,7 +831,7 @@ func (mm *MemoryManager) getVecPMAInternalMappingsLocked(ars usermem.AddrRangeSe // * pseg.Range().Contains(ar.Start). func (mm *MemoryManager) internalMappingsLocked(pseg pmaIterator, ar usermem.AddrRange) safemem.BlockSeq { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 { + if !ar.WellFormed() || ar.Length() == 0 { panic(fmt.Sprintf("invalid ar: %v", ar)) } if !pseg.Range().Contains(ar.Start) { @@ -1050,7 +1050,7 @@ func (pseg pmaIterator) fileRangeOf(ar usermem.AddrRange) memmap.FileRange { if !pseg.Ok() { panic("terminal pma iterator") } - if !ar.WellFormed() || ar.Length() <= 0 { + if !ar.WellFormed() || ar.Length() == 0 { panic(fmt.Sprintf("invalid ar: %v", ar)) } if !pseg.Range().IsSupersetOf(ar) { diff --git a/pkg/sentry/mm/syscalls.go b/pkg/sentry/mm/syscalls.go index a2555ba1a..675efdc7c 100644 --- a/pkg/sentry/mm/syscalls.go +++ b/pkg/sentry/mm/syscalls.go @@ -17,6 +17,7 @@ package mm import ( "fmt" mrand "math/rand" + "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" @@ -1274,3 +1275,27 @@ func (mm *MemoryManager) VirtualDataSize() uint64 { defer mm.mappingMu.RUnlock() return mm.dataAS } + +// EnableMembarrierPrivate causes future calls to IsMembarrierPrivateEnabled to +// return true. +func (mm *MemoryManager) EnableMembarrierPrivate() { + atomic.StoreUint32(&mm.membarrierPrivateEnabled, 1) +} + +// IsMembarrierPrivateEnabled returns true if mm.EnableMembarrierPrivate() has +// previously been called. +func (mm *MemoryManager) IsMembarrierPrivateEnabled() bool { + return atomic.LoadUint32(&mm.membarrierPrivateEnabled) != 0 +} + +// EnableMembarrierRSeq causes future calls to IsMembarrierRSeqEnabled to +// return true. +func (mm *MemoryManager) EnableMembarrierRSeq() { + atomic.StoreUint32(&mm.membarrierRSeqEnabled, 1) +} + +// IsMembarrierRSeqEnabled returns true if mm.EnableMembarrierRSeq() has +// previously been called. +func (mm *MemoryManager) IsMembarrierRSeqEnabled() bool { + return atomic.LoadUint32(&mm.membarrierRSeqEnabled) != 0 +} diff --git a/pkg/sentry/mm/vma.go b/pkg/sentry/mm/vma.go index f769d8294..b8df72813 100644 --- a/pkg/sentry/mm/vma.go +++ b/pkg/sentry/mm/vma.go @@ -266,7 +266,7 @@ func (mm *MemoryManager) mlockedBytesRangeLocked(ar usermem.AddrRange) uint64 { // * ar.Length() != 0. func (mm *MemoryManager) getVMAsLocked(ctx context.Context, ar usermem.AddrRange, at usermem.AccessType, ignorePermissions bool) (vmaIterator, vmaGapIterator, error) { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 { + if !ar.WellFormed() || ar.Length() == 0 { panic(fmt.Sprintf("invalid ar: %v", ar)) } } @@ -350,7 +350,7 @@ const guardBytes = 256 * usermem.PageSize // * ar must be page-aligned. func (mm *MemoryManager) unmapLocked(ctx context.Context, ar usermem.AddrRange) vmaGapIterator { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() { + if !ar.WellFormed() || ar.Length() == 0 || !ar.IsPageAligned() { panic(fmt.Sprintf("invalid ar: %v", ar)) } } @@ -371,7 +371,7 @@ func (mm *MemoryManager) unmapLocked(ctx context.Context, ar usermem.AddrRange) // * ar must be page-aligned. func (mm *MemoryManager) removeVMAsLocked(ctx context.Context, ar usermem.AddrRange) vmaGapIterator { if checkInvariants { - if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() { + if !ar.WellFormed() || ar.Length() == 0 || !ar.IsPageAligned() { panic(fmt.Sprintf("invalid ar: %v", ar)) } } @@ -511,7 +511,7 @@ func (vseg vmaIterator) mappableRangeOf(ar usermem.AddrRange) memmap.MappableRan if vseg.ValuePtr().mappable == nil { panic("MappableRange is meaningless for anonymous vma") } - if !ar.WellFormed() || ar.Length() <= 0 { + if !ar.WellFormed() || ar.Length() == 0 { panic(fmt.Sprintf("invalid ar: %v", ar)) } if !vseg.Range().IsSupersetOf(ar) { @@ -536,7 +536,7 @@ func (vseg vmaIterator) addrRangeOf(mr memmap.MappableRange) usermem.AddrRange { if vseg.ValuePtr().mappable == nil { panic("MappableRange is meaningless for anonymous vma") } - if !mr.WellFormed() || mr.Length() <= 0 { + if !mr.WellFormed() || mr.Length() == 0 { panic(fmt.Sprintf("invalid mr: %v", mr)) } if !vseg.mappableRange().IsSupersetOf(mr) { diff --git a/pkg/sentry/platform/BUILD b/pkg/sentry/platform/BUILD index 209b28053..db7d55ef2 100644 --- a/pkg/sentry/platform/BUILD +++ b/pkg/sentry/platform/BUILD @@ -15,6 +15,7 @@ go_library( "//pkg/context", "//pkg/seccomp", "//pkg/sentry/arch", + "//pkg/sentry/hostmm", "//pkg/sentry/memmap", "//pkg/usermem", ], diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index 3970dd81d..dd2bbeb12 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -9,12 +9,12 @@ go_library( "bluepill.go", "bluepill_allocator.go", "bluepill_amd64.go", - "bluepill_amd64.s", "bluepill_amd64_unsafe.go", "bluepill_arm64.go", "bluepill_arm64.s", "bluepill_arm64_unsafe.go", "bluepill_fault.go", + "bluepill_impl_amd64.s", "bluepill_unsafe.go", "context.go", "filters_amd64.go", @@ -56,6 +56,7 @@ go_library( "//pkg/sentry/time", "//pkg/sync", "//pkg/usermem", + "@org_golang_x_sys//unix:go_default_library", ], ) @@ -78,6 +79,15 @@ go_test( "//pkg/sentry/platform/kvm/testutil", "//pkg/sentry/platform/ring0", "//pkg/sentry/platform/ring0/pagetables", + "//pkg/sentry/time", "//pkg/usermem", ], ) + +genrule( + name = "bluepill_impl_amd64", + srcs = ["bluepill_amd64.s"], + outs = ["bluepill_impl_amd64.s"], + cmd = "(echo -e '// build +amd64\\n' && $(location //pkg/sentry/platform/ring0/gen_offsets) && cat $(SRCS)) > $@", + tools = ["//pkg/sentry/platform/ring0/gen_offsets"], +) diff --git a/pkg/sentry/platform/kvm/bluepill_amd64.s b/pkg/sentry/platform/kvm/bluepill_amd64.s index 2bc34a435..025ea93b5 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64.s +++ b/pkg/sentry/platform/kvm/bluepill_amd64.s @@ -19,11 +19,6 @@ // This is guaranteed to be zero. #define VCPU_CPU 0x0 -// CPU_SELF is the self reference in ring0's percpu. -// -// This is guaranteed to be zero. -#define CPU_SELF 0x0 - // Context offsets. // // Only limited use of the context is done in the assembly stub below, most is @@ -44,7 +39,7 @@ begin: LEAQ VCPU_CPU(AX), BX BYTE CLI; check_vcpu: - MOVQ CPU_SELF(GS), CX + MOVQ ENTRY_CPU_SELF(GS), CX CMPQ BX, CX JE right_vCPU wrong_vcpu: diff --git a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go index 03a98512e..0a54dd30d 100644 --- a/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_amd64_unsafe.go @@ -83,5 +83,34 @@ func bluepillStopGuest(c *vCPU) { // //go:nosplit func bluepillReadyStopGuest(c *vCPU) bool { - return c.runData.readyForInterruptInjection != 0 + if c.runData.readyForInterruptInjection == 0 { + return false + } + + if c.runData.ifFlag == 0 { + // This is impossible if readyForInterruptInjection is 1. + throw("interrupts are disabled") + } + + // Disable interrupts if we are in the kernel space. + // + // When the Sentry switches into the kernel mode, it disables + // interrupts. But when goruntime switches on a goroutine which has + // been saved in the host mode, it restores flags and this enables + // interrupts. See the comment of UserFlagsSet for more details. + uregs := userRegs{} + err := c.getUserRegisters(&uregs) + if err != 0 { + throw("failed to get user registers") + } + + if ring0.IsKernelFlags(uregs.RFLAGS) { + uregs.RFLAGS &^= ring0.KernelFlagsClear + err = c.setUserRegisters(&uregs) + if err != 0 { + throw("failed to set user registers") + } + return false + } + return true } diff --git a/pkg/sentry/platform/kvm/filters_amd64.go b/pkg/sentry/platform/kvm/filters_amd64.go index 7d949f1dd..d3d216aa5 100644 --- a/pkg/sentry/platform/kvm/filters_amd64.go +++ b/pkg/sentry/platform/kvm/filters_amd64.go @@ -17,14 +17,23 @@ package kvm import ( "syscall" + "golang.org/x/sys/unix" + + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/seccomp" ) // SyscallFilters returns syscalls made exclusively by the KVM platform. func (*KVM) SyscallFilters() seccomp.SyscallRules { return seccomp.SyscallRules{ - syscall.SYS_ARCH_PRCTL: {}, - syscall.SYS_IOCTL: {}, + syscall.SYS_ARCH_PRCTL: {}, + syscall.SYS_IOCTL: {}, + unix.SYS_MEMBARRIER: []seccomp.Rule{ + { + seccomp.EqualTo(linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED), + seccomp.EqualTo(0), + }, + }, syscall.SYS_MMAP: {}, syscall.SYS_RT_SIGSUSPEND: {}, syscall.SYS_RT_SIGTIMEDWAIT: {}, diff --git a/pkg/sentry/platform/kvm/filters_arm64.go b/pkg/sentry/platform/kvm/filters_arm64.go index 9245d07c2..21abc2a3d 100644 --- a/pkg/sentry/platform/kvm/filters_arm64.go +++ b/pkg/sentry/platform/kvm/filters_arm64.go @@ -17,13 +17,22 @@ package kvm import ( "syscall" + "golang.org/x/sys/unix" + + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/seccomp" ) // SyscallFilters returns syscalls made exclusively by the KVM platform. func (*KVM) SyscallFilters() seccomp.SyscallRules { return seccomp.SyscallRules{ - syscall.SYS_IOCTL: {}, + syscall.SYS_IOCTL: {}, + unix.SYS_MEMBARRIER: []seccomp.Rule{ + { + seccomp.EqualTo(linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED), + seccomp.EqualTo(0), + }, + }, syscall.SYS_MMAP: {}, syscall.SYS_RT_SIGSUSPEND: {}, syscall.SYS_RT_SIGTIMEDWAIT: {}, diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index ae813e24e..dd45ad10b 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -63,6 +63,9 @@ type runData struct { type KVM struct { platform.NoCPUPreemptionDetection + // KVM never changes mm_structs. + platform.UseHostProcessMemoryBarrier + // machine is the backing VM. machine *machine } @@ -156,15 +159,7 @@ func (*KVM) MaxUserAddress() usermem.Addr { func (k *KVM) NewAddressSpace(_ interface{}) (platform.AddressSpace, <-chan struct{}, error) { // Allocate page tables and install system mappings. pageTables := pagetables.New(newAllocator()) - applyPhysicalRegions(func(pr physicalRegion) bool { - // Map the kernel in the upper half. - pageTables.Map( - usermem.Addr(ring0.KernelStartAddress|pr.virtual), - pr.length, - pagetables.MapOpts{AccessType: usermem.AnyAccess}, - pr.physical) - return true // Keep iterating. - }) + k.machine.mapUpperHalf(pageTables) // Return the new address space. return &addressSpace{ diff --git a/pkg/sentry/platform/kvm/kvm_const.go b/pkg/sentry/platform/kvm/kvm_const.go index 5c4b18899..6abaa21c4 100644 --- a/pkg/sentry/platform/kvm/kvm_const.go +++ b/pkg/sentry/platform/kvm/kvm_const.go @@ -26,12 +26,16 @@ const ( _KVM_RUN = 0xae80 _KVM_NMI = 0xae9a _KVM_CHECK_EXTENSION = 0xae03 + _KVM_GET_TSC_KHZ = 0xaea3 + _KVM_SET_TSC_KHZ = 0xaea2 _KVM_INTERRUPT = 0x4004ae86 _KVM_SET_MSRS = 0x4008ae89 _KVM_SET_USER_MEMORY_REGION = 0x4020ae46 _KVM_SET_REGS = 0x4090ae82 _KVM_SET_SREGS = 0x4138ae84 + _KVM_GET_MSRS = 0xc008ae88 _KVM_GET_REGS = 0x8090ae81 + _KVM_GET_SREGS = 0x8138ae83 _KVM_GET_SUPPORTED_CPUID = 0xc008ae05 _KVM_SET_CPUID2 = 0x4008ae90 _KVM_SET_SIGNAL_MASK = 0x4004ae8b @@ -79,11 +83,14 @@ const ( ) // KVM hypercall list. +// // Canonical list of hypercalls supported. const ( // On amd64, it uses 'HLT' to leave the guest. + // // Unlike amd64, arm64 can only uses mmio_exit/psci to leave the guest. - // _KVM_HYPERCALL_VMEXIT is only used on Arm64 for now. + // + // _KVM_HYPERCALL_VMEXIT is only used on arm64 for now. _KVM_HYPERCALL_VMEXIT int = iota _KVM_HYPERCALL_MAX ) diff --git a/pkg/sentry/platform/kvm/kvm_test.go b/pkg/sentry/platform/kvm/kvm_test.go index 45b3180f1..2e12470aa 100644 --- a/pkg/sentry/platform/kvm/kvm_test.go +++ b/pkg/sentry/platform/kvm/kvm_test.go @@ -27,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform/kvm/testutil" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + ktime "gvisor.dev/gvisor/pkg/sentry/time" "gvisor.dev/gvisor/pkg/usermem" ) @@ -442,6 +443,22 @@ func TestWrongVCPU(t *testing.T) { }) } +func TestRdtsc(t *testing.T) { + var i int // Iteration count. + kvmTest(t, nil, func(c *vCPU) bool { + start := ktime.Rdtsc() + bluepill(c) + guest := ktime.Rdtsc() + redpill() + end := ktime.Rdtsc() + if start > guest || guest > end { + t.Errorf("inconsistent time: start=%d, guest=%d, end=%d", start, guest, end) + } + i++ + return i < 100 + }) +} + func BenchmarkApplicationSyscall(b *testing.B) { var ( i int // Iteration includes machine.Get() / machine.Put(). diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index 372a4cbd7..75da253c5 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -155,7 +155,7 @@ func (m *machine) newVCPU() *vCPU { fd: int(fd), machine: m, } - c.CPU.Init(&m.kernel, c) + c.CPU.Init(&m.kernel, c.id, c) m.vCPUsByID[c.id] = c // Ensure the signal mask is correct. @@ -183,9 +183,6 @@ func newMachine(vm int) (*machine, error) { // Create the machine. m := &machine{fd: vm} m.available.L = &m.mu - m.kernel.Init(ring0.KernelOpts{ - PageTables: pagetables.New(newAllocator()), - }) // Pull the maximum vCPUs. maxVCPUs, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, uintptr(m.fd), _KVM_CHECK_EXTENSION, _KVM_CAP_MAX_VCPUS) @@ -197,6 +194,9 @@ func newMachine(vm int) (*machine, error) { log.Debugf("The maximum number of vCPUs is %d.", m.maxVCPUs) m.vCPUsByTID = make(map[uint64]*vCPU) m.vCPUsByID = make([]*vCPU, m.maxVCPUs) + m.kernel.Init(ring0.KernelOpts{ + PageTables: pagetables.New(newAllocator()), + }, m.maxVCPUs) // Pull the maximum slots. maxSlots, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, uintptr(m.fd), _KVM_CHECK_EXTENSION, _KVM_CAP_MAX_MEMSLOTS) @@ -219,15 +219,9 @@ func newMachine(vm int) (*machine, error) { pagetables.MapOpts{AccessType: usermem.AnyAccess}, pr.physical) - // And keep everything in the upper half. - m.kernel.PageTables.Map( - usermem.Addr(ring0.KernelStartAddress|pr.virtual), - pr.length, - pagetables.MapOpts{AccessType: usermem.AnyAccess}, - pr.physical) - return true // Keep iterating. }) + m.mapUpperHalf(m.kernel.PageTables) var physicalRegionsReadOnly []physicalRegion var physicalRegionsAvailable []physicalRegion @@ -365,6 +359,11 @@ func (m *machine) Destroy() { // Get gets an available vCPU. // // This will return with the OS thread locked. +// +// It is guaranteed that if any OS thread TID is in guest, m.vCPUs[TID] points +// to the vCPU in which the OS thread TID is running. So if Get() returns with +// the corrent context in guest, the vCPU of it must be the same as what +// Get() returns. func (m *machine) Get() *vCPU { m.mu.RLock() runtime.LockOSThread() diff --git a/pkg/sentry/platform/kvm/machine_amd64.go b/pkg/sentry/platform/kvm/machine_amd64.go index acc823ba6..451953008 100644 --- a/pkg/sentry/platform/kvm/machine_amd64.go +++ b/pkg/sentry/platform/kvm/machine_amd64.go @@ -18,14 +18,17 @@ package kvm import ( "fmt" + "math/big" "reflect" "runtime/debug" "syscall" + "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + ktime "gvisor.dev/gvisor/pkg/sentry/time" "gvisor.dev/gvisor/pkg/usermem" ) @@ -144,6 +147,7 @@ func (c *vCPU) initArchState() error { // Set the entrypoint for the kernel. kernelUserRegs.RIP = uint64(reflect.ValueOf(ring0.Start).Pointer()) kernelUserRegs.RAX = uint64(reflect.ValueOf(&c.CPU).Pointer()) + kernelUserRegs.RSP = c.StackTop() kernelUserRegs.RFLAGS = ring0.KernelFlagsSet // Set the system registers. @@ -152,8 +156,8 @@ func (c *vCPU) initArchState() error { } // Set the user registers. - if err := c.setUserRegisters(&kernelUserRegs); err != nil { - return err + if errno := c.setUserRegisters(&kernelUserRegs); errno != 0 { + return fmt.Errorf("error setting user registers: %v", errno) } // Allocate some floating point state save area for the local vCPU. @@ -166,6 +170,133 @@ func (c *vCPU) initArchState() error { return c.setSystemTime() } +// bitsForScaling returns the bits available for storing the fraction component +// of the TSC scaling ratio. This allows us to replicate the (bad) math done by +// the kernel below in scaledTSC, and ensure we can compute an exact zero +// offset in setSystemTime. +// +// These constants correspond to kvm_tsc_scaling_ratio_frac_bits. +var bitsForScaling = func() int64 { + fs := cpuid.HostFeatureSet() + if fs.Intel() { + return 48 // See vmx.c (kvm sources). + } else if fs.AMD() { + return 32 // See svm.c (svm sources). + } else { + return 63 // Unknown: theoretical maximum. + } +}() + +// scaledTSC returns the host TSC scaled by the given frequency. +// +// This assumes a current frequency of 1. We require only the unitless ratio of +// rawFreq to some current frequency. See setSystemTime for context. +// +// The kernel math guarantees that all bits of the multiplication and division +// will be correctly preserved and applied. However, it is not possible to +// actually store the ratio correctly. So we need to use the same schema in +// order to calculate the scaled frequency and get the same result. +// +// We can assume that the current frequency is (1), so we are calculating a +// strict inverse of this value. This simplifies this function considerably. +// +// Roughly, the returned value "scaledTSC" will have: +// scaledTSC/hostTSC == 1/rawFreq +// +//go:nosplit +func scaledTSC(rawFreq uintptr) int64 { + scale := int64(1 << bitsForScaling) + ratio := big.NewInt(scale / int64(rawFreq)) + ratio.Mul(ratio, big.NewInt(int64(ktime.Rdtsc()))) + ratio.Div(ratio, big.NewInt(scale)) + return ratio.Int64() +} + +// setSystemTime sets the vCPU to the system time. +func (c *vCPU) setSystemTime() error { + // First, scale down the clock frequency to the lowest value allowed by + // the API itself. How low we can go depends on the underlying + // hardware, but it is typically ~1/2^48 for Intel, ~1/2^32 for AMD. + // Even the lower bound here will take a 4GHz frequency down to 1Hz, + // meaning that everything should be able to handle a Khz setting of 1 + // with bits to spare. + // + // Note that reducing the clock does not typically require special + // capabilities as it is emulated in KVM. We don't actually use this + // capability, but it means that this method should be robust to + // different hardware configurations. + rawFreq, err := c.getTSCFreq() + if err != nil { + return c.setSystemTimeLegacy() + } + if err := c.setTSCFreq(1); err != nil { + return c.setSystemTimeLegacy() + } + + // Always restore the original frequency. + defer func() { + if err := c.setTSCFreq(rawFreq); err != nil { + panic(err.Error()) + } + }() + + // Attempt to set the system time in this compressed world. The + // calculation for offset normally looks like: + // + // offset = target_tsc - kvm_scale_tsc(vcpu, rdtsc()); + // + // So as long as the kvm_scale_tsc component is constant before and + // after the call to set the TSC value (and it is passes as the + // target_tsc), we will compute an offset value of zero. + // + // This is effectively cheating to make our "setSystemTime" call so + // unbelievably, incredibly fast that we do it "instantly" and all the + // calculations result in an offset of zero. + lastTSC := scaledTSC(rawFreq) + for { + if err := c.setTSC(uint64(lastTSC)); err != nil { + return err + } + nextTSC := scaledTSC(rawFreq) + if lastTSC == nextTSC { + return nil + } + lastTSC = nextTSC // Try again. + } +} + +// setSystemTimeLegacy calibrates and sets an approximate system time. +func (c *vCPU) setSystemTimeLegacy() error { + const minIterations = 10 + minimum := uint64(0) + for iter := 0; ; iter++ { + // Try to set the TSC to an estimate of where it will be + // on the host during a "fast" system call iteration. + start := uint64(ktime.Rdtsc()) + if err := c.setTSC(start + (minimum / 2)); err != nil { + return err + } + // See if this is our new minimum call time. Note that this + // serves two functions: one, we make sure that we are + // accurately predicting the offset we need to set. Second, we + // don't want to do the final set on a slow call, which could + // produce a really bad result. + end := uint64(ktime.Rdtsc()) + if end < start { + continue // Totally bogus: unstable TSC? + } + current := end - start + if current < minimum || iter == 0 { + minimum = current // Set our new minimum. + } + // Is this past minIterations and within ~10% of minimum? + upperThreshold := (((minimum << 3) + minimum) >> 3) + if iter >= minIterations && current <= upperThreshold { + return nil + } + } +} + // nonCanonical generates a canonical address return. // //go:nosplit @@ -345,3 +476,41 @@ func rdonlyRegionsForSetMem() (phyRegions []physicalRegion) { func availableRegionsForSetMem() (phyRegions []physicalRegion) { return physicalRegions } + +var execRegions = func() (regions []region) { + applyVirtualRegions(func(vr virtualRegion) { + if excludeVirtualRegion(vr) || vr.filename == "[vsyscall]" { + return + } + if vr.accessType.Execute { + regions = append(regions, vr.region) + } + }) + return +}() + +func (m *machine) mapUpperHalf(pageTable *pagetables.PageTables) { + for _, r := range execRegions { + physical, length, ok := translateToPhysical(r.virtual) + if !ok || length < r.length { + panic("impossilbe translation") + } + pageTable.Map( + usermem.Addr(ring0.KernelStartAddress|r.virtual), + r.length, + pagetables.MapOpts{AccessType: usermem.Execute}, + physical) + } + for start, end := range m.kernel.EntryRegions() { + regionLen := end - start + physical, length, ok := translateToPhysical(start) + if !ok || length < regionLen { + panic("impossible translation") + } + pageTable.Map( + usermem.Addr(ring0.KernelStartAddress|start), + regionLen, + pagetables.MapOpts{AccessType: usermem.ReadWrite}, + physical) + } +} diff --git a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go index 290f035dd..b430f92c6 100644 --- a/pkg/sentry/platform/kvm/machine_amd64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_amd64_unsafe.go @@ -23,7 +23,6 @@ import ( "unsafe" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/time" ) // loadSegments copies the current segments. @@ -61,91 +60,63 @@ func (c *vCPU) setCPUID() error { return nil } -// setSystemTime sets the TSC for the vCPU. +// getTSCFreq gets the TSC frequency. // -// This has to make the call many times in order to minimize the intrinsic -// error in the offset. Unfortunately KVM does not expose a relative offset via -// the API, so this is an approximation. We do this via an iterative algorithm. -// This has the advantage that it can generally deal with highly variable -// system call times and should converge on the correct offset. -func (c *vCPU) setSystemTime() error { - const ( - _MSR_IA32_TSC = 0x00000010 - calibrateTries = 10 - ) - registers := modelControlRegisters{ - nmsrs: 1, - } - registers.entries[0] = modelControlRegister{ - index: _MSR_IA32_TSC, +// If mustSucceed is true, then this function panics on error. +func (c *vCPU) getTSCFreq() (uintptr, error) { + rawFreq, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_GET_TSC_KHZ, + 0 /* ignored */) + if errno != 0 { + return 0, errno } - target := uint64(^uint32(0)) - for done := 0; done < calibrateTries; { - start := uint64(time.Rdtsc()) - registers.entries[0].data = start + target - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_MSRS, - uintptr(unsafe.Pointer(®isters))); errno != 0 { - return fmt.Errorf("error setting system time: %v", errno) - } - // See if this is our new minimum call time. Note that this - // serves two functions: one, we make sure that we are - // accurately predicting the offset we need to set. Second, we - // don't want to do the final set on a slow call, which could - // produce a really bad result. So we only count attempts - // within +/- 6.25% of our minimum as an attempt. - end := uint64(time.Rdtsc()) - if end < start { - continue // Totally bogus. - } - half := (end - start) / 2 - if half < target { - target = half - } - if (half - target) < target/8 { - done++ - } + return rawFreq, nil +} + +// setTSCFreq sets the TSC frequency. +func (c *vCPU) setTSCFreq(freq uintptr) error { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_TSC_KHZ, + freq /* khz */); errno != 0 { + return fmt.Errorf("error setting TSC frequency: %v", errno) } return nil } -// setSignalMask sets the vCPU signal mask. -// -// This must be called prior to running the vCPU. -func (c *vCPU) setSignalMask() error { - // The layout of this structure implies that it will not necessarily be - // the same layout chosen by the Go compiler. It gets fudged here. - var data struct { - length uint32 - mask1 uint32 - mask2 uint32 - _ uint32 +// setTSC sets the TSC value. +func (c *vCPU) setTSC(value uint64) error { + const _MSR_IA32_TSC = 0x00000010 + registers := modelControlRegisters{ + nmsrs: 1, } - data.length = 8 // Fixed sigset size. - data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) - data.mask2 = ^uint32(bounceSignalMask >> 32) + registers.entries[0].index = _MSR_IA32_TSC + registers.entries[0].data = value if _, _, errno := syscall.RawSyscall( syscall.SYS_IOCTL, uintptr(c.fd), - _KVM_SET_SIGNAL_MASK, - uintptr(unsafe.Pointer(&data))); errno != 0 { - return fmt.Errorf("error setting signal mask: %v", errno) + _KVM_SET_MSRS, + uintptr(unsafe.Pointer(®isters))); errno != 0 { + return fmt.Errorf("error setting tsc: %v", errno) } return nil } // setUserRegisters sets user registers in the vCPU. -func (c *vCPU) setUserRegisters(uregs *userRegs) error { +// +//go:nosplit +func (c *vCPU) setUserRegisters(uregs *userRegs) syscall.Errno { if _, _, errno := syscall.RawSyscall( syscall.SYS_IOCTL, uintptr(c.fd), _KVM_SET_REGS, uintptr(unsafe.Pointer(uregs))); errno != 0 { - return fmt.Errorf("error setting user registers: %v", errno) + return errno } - return nil + return 0 } // getUserRegisters reloads user registers in the vCPU. @@ -175,3 +146,17 @@ func (c *vCPU) setSystemRegisters(sregs *systemRegs) error { } return nil } + +// getSystemRegisters sets system registers. +// +//go:nosplit +func (c *vCPU) getSystemRegisters(sregs *systemRegs) syscall.Errno { + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_GET_SREGS, + uintptr(unsafe.Pointer(sregs))); errno != 0 { + return errno + } + return 0 +} diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index 9db171af9..2df762991 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -19,6 +19,7 @@ package kvm import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/usermem" ) @@ -48,6 +49,18 @@ const ( poolPCIDs = 8 ) +func (m *machine) mapUpperHalf(pageTable *pagetables.PageTables) { + applyPhysicalRegions(func(pr physicalRegion) bool { + pageTable.Map( + usermem.Addr(ring0.KernelStartAddress|pr.virtual), + pr.length, + pagetables.MapOpts{AccessType: usermem.AnyAccess}, + pr.physical) + + return true // Keep iterating. + }) +} + // Get all read-only physicalRegions. func rdonlyRegionsForSetMem() (phyRegions []physicalRegion) { var rdonlyRegions []region diff --git a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go index 537419657..a163f956d 100644 --- a/pkg/sentry/platform/kvm/machine_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_arm64_unsafe.go @@ -191,42 +191,6 @@ func (c *vCPU) getOneRegister(reg *kvmOneReg) error { return nil } -// setCPUID sets the CPUID to be used by the guest. -func (c *vCPU) setCPUID() error { - return nil -} - -// setSystemTime sets the TSC for the vCPU. -func (c *vCPU) setSystemTime() error { - return nil -} - -// setSignalMask sets the vCPU signal mask. -// -// This must be called prior to running the vCPU. -func (c *vCPU) setSignalMask() error { - // The layout of this structure implies that it will not necessarily be - // the same layout chosen by the Go compiler. It gets fudged here. - var data struct { - length uint32 - mask1 uint32 - mask2 uint32 - _ uint32 - } - data.length = 8 // Fixed sigset size. - data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) - data.mask2 = ^uint32(bounceSignalMask >> 32) - if _, _, errno := syscall.RawSyscall( - syscall.SYS_IOCTL, - uintptr(c.fd), - _KVM_SET_SIGNAL_MASK, - uintptr(unsafe.Pointer(&data))); errno != 0 { - return fmt.Errorf("error setting signal mask: %v", errno) - } - - return nil -} - // SwitchToUser unpacks architectural-details. func (c *vCPU) SwitchToUser(switchOpts ring0.SwitchOpts, info *arch.SignalInfo) (usermem.AccessType, error) { // Check for canonical addresses. diff --git a/pkg/sentry/platform/kvm/machine_unsafe.go b/pkg/sentry/platform/kvm/machine_unsafe.go index 607c82156..1d6ca245a 100644 --- a/pkg/sentry/platform/kvm/machine_unsafe.go +++ b/pkg/sentry/platform/kvm/machine_unsafe.go @@ -143,3 +143,29 @@ func (c *vCPU) waitUntilNot(state uint32) { panic("futex wait error") } } + +// setSignalMask sets the vCPU signal mask. +// +// This must be called prior to running the vCPU. +func (c *vCPU) setSignalMask() error { + // The layout of this structure implies that it will not necessarily be + // the same layout chosen by the Go compiler. It gets fudged here. + var data struct { + length uint32 + mask1 uint32 + mask2 uint32 + _ uint32 + } + data.length = 8 // Fixed sigset size. + data.mask1 = ^uint32(bounceSignalMask & 0xffffffff) + data.mask2 = ^uint32(bounceSignalMask >> 32) + if _, _, errno := syscall.RawSyscall( + syscall.SYS_IOCTL, + uintptr(c.fd), + _KVM_SET_SIGNAL_MASK, + uintptr(unsafe.Pointer(&data))); errno != 0 { + return fmt.Errorf("error setting signal mask: %v", errno) + } + + return nil +} diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index 530e779b0..dcfe839a7 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/hostmm" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/usermem" ) @@ -52,6 +53,10 @@ type Platform interface { // can reliably return ErrContextCPUPreempted. DetectsCPUPreemption() bool + // HaveGlobalMemoryBarrier returns true if the GlobalMemoryBarrier method + // is supported. + HaveGlobalMemoryBarrier() bool + // MapUnit returns the alignment used for optional mappings into this // platform's AddressSpaces. Higher values indicate lower per-page costs // for AddressSpace.MapFile. As a special case, a MapUnit of 0 indicates @@ -97,6 +102,15 @@ type Platform interface { // called. PreemptAllCPUs() error + // GlobalMemoryBarrier blocks until all threads running application code + // (via Context.Switch) and all task goroutines "have passed through a + // state where all memory accesses to user-space addresses match program + // order between entry to and return from [GlobalMemoryBarrier]", as for + // membarrier(2). + // + // Preconditions: HaveGlobalMemoryBarrier() == true. + GlobalMemoryBarrier() error + // SyscallFilters returns syscalls made exclusively by this platform. SyscallFilters() seccomp.SyscallRules } @@ -115,6 +129,43 @@ func (NoCPUPreemptionDetection) PreemptAllCPUs() error { panic("This platform does not support CPU preemption detection") } +// UseHostGlobalMemoryBarrier implements Platform.HaveGlobalMemoryBarrier and +// Platform.GlobalMemoryBarrier by invoking equivalent functionality on the +// host. +type UseHostGlobalMemoryBarrier struct{} + +// HaveGlobalMemoryBarrier implements Platform.HaveGlobalMemoryBarrier. +func (UseHostGlobalMemoryBarrier) HaveGlobalMemoryBarrier() bool { + return hostmm.HaveGlobalMemoryBarrier() +} + +// GlobalMemoryBarrier implements Platform.GlobalMemoryBarrier. +func (UseHostGlobalMemoryBarrier) GlobalMemoryBarrier() error { + return hostmm.GlobalMemoryBarrier() +} + +// UseHostProcessMemoryBarrier implements Platform.HaveGlobalMemoryBarrier and +// Platform.GlobalMemoryBarrier by invoking a process-local memory barrier. +// This is faster than UseHostGlobalMemoryBarrier, but is only appropriate for +// platforms for which application code executes while using the sentry's +// mm_struct. +type UseHostProcessMemoryBarrier struct{} + +// HaveGlobalMemoryBarrier implements Platform.HaveGlobalMemoryBarrier. +func (UseHostProcessMemoryBarrier) HaveGlobalMemoryBarrier() bool { + // Fall back to a global memory barrier if a process-local one isn't + // available. + return hostmm.HaveProcessMemoryBarrier() || hostmm.HaveGlobalMemoryBarrier() +} + +// GlobalMemoryBarrier implements Platform.GlobalMemoryBarrier. +func (UseHostProcessMemoryBarrier) GlobalMemoryBarrier() error { + if hostmm.HaveProcessMemoryBarrier() { + return hostmm.ProcessMemoryBarrier() + } + return hostmm.GlobalMemoryBarrier() +} + // MemoryManager represents an abstraction above the platform address space // which manages memory mappings and their contents. type MemoryManager interface { diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index b52d0fbd8..f56aa3b79 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -192,6 +192,7 @@ func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} type PTrace struct { platform.MMapMinAddr platform.NoCPUPreemptionDetection + platform.UseHostGlobalMemoryBarrier } // New returns a new ptrace-based implementation of the platform interface. diff --git a/pkg/sentry/platform/ring0/defs_amd64.go b/pkg/sentry/platform/ring0/defs_amd64.go index 9c6c2cf5c..f617519fa 100644 --- a/pkg/sentry/platform/ring0/defs_amd64.go +++ b/pkg/sentry/platform/ring0/defs_amd64.go @@ -76,15 +76,42 @@ type KernelOpts struct { type KernelArchState struct { KernelOpts + // cpuEntries is array of kernelEntry for all cpus + cpuEntries []kernelEntry + // globalIDT is our set of interrupt gates. - globalIDT idt64 + globalIDT *idt64 } -// CPUArchState contains CPU-specific arch state. -type CPUArchState struct { +// kernelEntry contains minimal CPU-specific arch state +// that can be mapped at the upper of the address space. +// Malicious APP might steal info from it via CPU bugs. +type kernelEntry struct { // stack is the stack used for interrupts on this CPU. stack [256]byte + // scratch space for temporary usage. + scratch0 uint64 + scratch1 uint64 + + // stackTop is the top of the stack. + stackTop uint64 + + // cpuSelf is back reference to CPU. + cpuSelf *CPU + + // kernelCR3 is the cr3 used for sentry kernel. + kernelCR3 uintptr + + // gdt is the CPU's descriptor table. + gdt descriptorTable + + // tss is the CPU's task state. + tss TaskState64 +} + +// CPUArchState contains CPU-specific arch state. +type CPUArchState struct { // errorCode is the error code from the last exception. errorCode uintptr @@ -97,11 +124,7 @@ type CPUArchState struct { // exception. errorType uintptr - // gdt is the CPU's descriptor table. - gdt descriptorTable - - // tss is the CPU's task state. - tss TaskState64 + *kernelEntry } // ErrorCode returns the last error code. diff --git a/pkg/sentry/platform/ring0/entry_amd64.go b/pkg/sentry/platform/ring0/entry_amd64.go index 7fa43c2f5..d87b1fd00 100644 --- a/pkg/sentry/platform/ring0/entry_amd64.go +++ b/pkg/sentry/platform/ring0/entry_amd64.go @@ -36,12 +36,15 @@ func sysenter() // This must be called prior to sysret/iret. func swapgs() +// jumpToKernel jumps to the kernel version of the current RIP. +func jumpToKernel() + // sysret returns to userspace from a system call. // // The return code is the vector that interrupted execution. // // See stubs.go for a note regarding the frame size of this function. -func sysret(*CPU, *arch.Registers) Vector +func sysret(cpu *CPU, regs *arch.Registers, userCR3 uintptr) Vector // "iret is the cadillac of CPL switching." // @@ -50,7 +53,7 @@ func sysret(*CPU, *arch.Registers) Vector // iret is nearly identical to sysret, except an iret is used to fully restore // all user state. This must be called in cases where all registers need to be // restored. -func iret(*CPU, *arch.Registers) Vector +func iret(cpu *CPU, regs *arch.Registers, userCR3 uintptr) Vector // exception is the generic exception entry. // diff --git a/pkg/sentry/platform/ring0/entry_amd64.s b/pkg/sentry/platform/ring0/entry_amd64.s index 02df38331..f59747df3 100644 --- a/pkg/sentry/platform/ring0/entry_amd64.s +++ b/pkg/sentry/platform/ring0/entry_amd64.s @@ -63,6 +63,15 @@ MOVQ offset+PTRACE_RSI(reg), SI; \ MOVQ offset+PTRACE_RDI(reg), DI; +// WRITE_CR3() writes the given CR3 value. +// +// The code corresponds to: +// +// mov %rax, %cr3 +// +#define WRITE_CR3() \ + BYTE $0x0f; BYTE $0x22; BYTE $0xd8; + // SWAP_GS swaps the kernel GS (CPU). #define SWAP_GS() \ BYTE $0x0F; BYTE $0x01; BYTE $0xf8; @@ -75,15 +84,9 @@ #define SYSRET64() \ BYTE $0x48; BYTE $0x0f; BYTE $0x07; -// LOAD_KERNEL_ADDRESS loads a kernel address. -#define LOAD_KERNEL_ADDRESS(from, to) \ - MOVQ from, to; \ - ORQ ·KernelStartAddress(SB), to; - // LOAD_KERNEL_STACK loads the kernel stack. -#define LOAD_KERNEL_STACK(from) \ - LOAD_KERNEL_ADDRESS(CPU_SELF(from), SP); \ - LEAQ CPU_STACK_TOP(SP), SP; +#define LOAD_KERNEL_STACK(entry) \ + MOVQ ENTRY_STACK_TOP(entry), SP; // See kernel.go. TEXT ·Halt(SB),NOSPLIT,$0 @@ -95,58 +98,93 @@ TEXT ·swapgs(SB),NOSPLIT,$0 SWAP_GS() RET +// jumpToKernel changes execution to the kernel address space. +// +// This works by changing the return value to the kernel version. +TEXT ·jumpToKernel(SB),NOSPLIT,$0 + MOVQ 0(SP), AX + ORQ ·KernelStartAddress(SB), AX // Future return value. + MOVQ AX, 0(SP) + RET + // See entry_amd64.go. TEXT ·sysret(SB),NOSPLIT,$0-24 - // Save original state. - LOAD_KERNEL_ADDRESS(cpu+0(FP), BX) - LOAD_KERNEL_ADDRESS(regs+8(FP), AX) + CALL ·jumpToKernel(SB) + // Save original state and stack. sysenter() or exception() + // from APP(gr3) will switch to this stack, set the return + // value (vector: 32(SP)) and then do RET, which will also + // automatically return to the lower half. + MOVQ cpu+0(FP), BX + MOVQ regs+8(FP), AX + MOVQ userCR3+16(FP), CX MOVQ SP, CPU_REGISTERS+PTRACE_RSP(BX) MOVQ BP, CPU_REGISTERS+PTRACE_RBP(BX) MOVQ AX, CPU_REGISTERS+PTRACE_RAX(BX) + // save SP AX userCR3 on the kernel stack. + MOVQ CPU_ENTRY(BX), BX + LOAD_KERNEL_STACK(BX) + PUSHQ PTRACE_RSP(AX) + PUSHQ PTRACE_RAX(AX) + PUSHQ CX + // Restore user register state. REGISTERS_LOAD(AX, 0) MOVQ PTRACE_RIP(AX), CX // Needed for SYSRET. MOVQ PTRACE_FLAGS(AX), R11 // Needed for SYSRET. - MOVQ PTRACE_RSP(AX), SP // Restore the stack directly. - MOVQ PTRACE_RAX(AX), AX // Restore AX (scratch). + + // restore userCR3, AX, SP. + POPQ AX // Get userCR3. + WRITE_CR3() // Switch to userCR3. + POPQ AX // Restore AX. + POPQ SP // Restore SP. SYSRET64() // See entry_amd64.go. TEXT ·iret(SB),NOSPLIT,$0-24 - // Save original state. - LOAD_KERNEL_ADDRESS(cpu+0(FP), BX) - LOAD_KERNEL_ADDRESS(regs+8(FP), AX) + CALL ·jumpToKernel(SB) + // Save original state and stack. sysenter() or exception() + // from APP(gr3) will switch to this stack, set the return + // value (vector: 32(SP)) and then do RET, which will also + // automatically return to the lower half. + MOVQ cpu+0(FP), BX + MOVQ regs+8(FP), AX + MOVQ userCR3+16(FP), CX MOVQ SP, CPU_REGISTERS+PTRACE_RSP(BX) MOVQ BP, CPU_REGISTERS+PTRACE_RBP(BX) MOVQ AX, CPU_REGISTERS+PTRACE_RAX(BX) // Build an IRET frame & restore state. + MOVQ CPU_ENTRY(BX), BX LOAD_KERNEL_STACK(BX) - MOVQ PTRACE_SS(AX), BX; PUSHQ BX - MOVQ PTRACE_RSP(AX), CX; PUSHQ CX - MOVQ PTRACE_FLAGS(AX), DX; PUSHQ DX - MOVQ PTRACE_CS(AX), DI; PUSHQ DI - MOVQ PTRACE_RIP(AX), SI; PUSHQ SI - REGISTERS_LOAD(AX, 0) // Restore most registers. - MOVQ PTRACE_RAX(AX), AX // Restore AX (scratch). + PUSHQ PTRACE_SS(AX) + PUSHQ PTRACE_RSP(AX) + PUSHQ PTRACE_FLAGS(AX) + PUSHQ PTRACE_CS(AX) + PUSHQ PTRACE_RIP(AX) + PUSHQ PTRACE_RAX(AX) // Save AX on kernel stack. + PUSHQ CX // Save userCR3 on kernel stack. + REGISTERS_LOAD(AX, 0) // Restore most registers. + POPQ AX // Get userCR3. + WRITE_CR3() // Switch to userCR3. + POPQ AX // Restore AX. IRET() // See entry_amd64.go. TEXT ·resume(SB),NOSPLIT,$0 // See iret, above. - MOVQ CPU_REGISTERS+PTRACE_SS(GS), BX; PUSHQ BX - MOVQ CPU_REGISTERS+PTRACE_RSP(GS), CX; PUSHQ CX - MOVQ CPU_REGISTERS+PTRACE_FLAGS(GS), DX; PUSHQ DX - MOVQ CPU_REGISTERS+PTRACE_CS(GS), DI; PUSHQ DI - MOVQ CPU_REGISTERS+PTRACE_RIP(GS), SI; PUSHQ SI - REGISTERS_LOAD(GS, CPU_REGISTERS) - MOVQ CPU_REGISTERS+PTRACE_RAX(GS), AX + MOVQ ENTRY_CPU_SELF(GS), AX // Load vCPU. + PUSHQ CPU_REGISTERS+PTRACE_SS(AX) + PUSHQ CPU_REGISTERS+PTRACE_RSP(AX) + PUSHQ CPU_REGISTERS+PTRACE_FLAGS(AX) + PUSHQ CPU_REGISTERS+PTRACE_CS(AX) + PUSHQ CPU_REGISTERS+PTRACE_RIP(AX) + REGISTERS_LOAD(AX, CPU_REGISTERS) + MOVQ CPU_REGISTERS+PTRACE_RAX(AX), AX IRET() // See entry_amd64.go. TEXT ·Start(SB),NOSPLIT,$0 - LOAD_KERNEL_STACK(AX) // Set the stack. PUSHQ $0x0 // Previous frame pointer. MOVQ SP, BP // Set frame pointer. PUSHQ AX // First argument (CPU). @@ -155,53 +193,60 @@ TEXT ·Start(SB),NOSPLIT,$0 // See entry_amd64.go. TEXT ·sysenter(SB),NOSPLIT,$0 - // Interrupts are always disabled while we're executing in kernel mode - // and always enabled while executing in user mode. Therefore, we can - // reliably look at the flags in R11 to determine where this syscall - // was from. - TESTL $_RFLAGS_IF, R11 + // _RFLAGS_IOPL0 is always set in the user mode and it is never set in + // the kernel mode. See the comment of UserFlagsSet for more details. + TESTL $_RFLAGS_IOPL0, R11 JZ kernel - user: SWAP_GS() - XCHGQ CPU_REGISTERS+PTRACE_RSP(GS), SP // Swap stacks. - XCHGQ CPU_REGISTERS+PTRACE_RAX(GS), AX // Swap for AX (regs). + MOVQ AX, ENTRY_SCRATCH0(GS) // Save user AX on scratch. + MOVQ ENTRY_KERNEL_CR3(GS), AX // Get kernel cr3 on AX. + WRITE_CR3() // Switch to kernel cr3. + + MOVQ ENTRY_CPU_SELF(GS), AX // Load vCPU. + MOVQ CPU_REGISTERS+PTRACE_RAX(AX), AX // Get user regs. REGISTERS_SAVE(AX, 0) // Save all except IP, FLAGS, SP, AX. - MOVQ CPU_REGISTERS+PTRACE_RAX(GS), BX // Load saved AX value. - MOVQ BX, PTRACE_RAX(AX) // Save everything else. - MOVQ BX, PTRACE_ORIGRAX(AX) MOVQ CX, PTRACE_RIP(AX) MOVQ R11, PTRACE_FLAGS(AX) - MOVQ CPU_REGISTERS+PTRACE_RSP(GS), BX; MOVQ BX, PTRACE_RSP(AX) - MOVQ $0, CPU_ERROR_CODE(GS) // Clear error code. - MOVQ $1, CPU_ERROR_TYPE(GS) // Set error type to user. + MOVQ SP, PTRACE_RSP(AX) + MOVQ ENTRY_SCRATCH0(GS), CX // Load saved user AX value. + MOVQ CX, PTRACE_RAX(AX) // Save everything else. + MOVQ CX, PTRACE_ORIGRAX(AX) + + MOVQ ENTRY_CPU_SELF(GS), AX // Load vCPU. + MOVQ CPU_REGISTERS+PTRACE_RSP(AX), SP // Get stacks. + MOVQ $0, CPU_ERROR_CODE(AX) // Clear error code. + MOVQ $1, CPU_ERROR_TYPE(AX) // Set error type to user. // Return to the kernel, where the frame is: // - // vector (sp+24) + // vector (sp+32) + // userCR3 (sp+24) // regs (sp+16) // cpu (sp+8) // vcpu.Switch (sp+0) // - MOVQ CPU_REGISTERS+PTRACE_RBP(GS), BP // Original base pointer. - MOVQ $Syscall, 24(SP) // Output vector. + MOVQ CPU_REGISTERS+PTRACE_RBP(AX), BP // Original base pointer. + MOVQ $Syscall, 32(SP) // Output vector. RET kernel: // We can't restore the original stack, but we can access the registers // in the CPU state directly. No need for temporary juggling. - MOVQ AX, CPU_REGISTERS+PTRACE_ORIGRAX(GS) - MOVQ AX, CPU_REGISTERS+PTRACE_RAX(GS) - REGISTERS_SAVE(GS, CPU_REGISTERS) - MOVQ CX, CPU_REGISTERS+PTRACE_RIP(GS) - MOVQ R11, CPU_REGISTERS+PTRACE_FLAGS(GS) - MOVQ SP, CPU_REGISTERS+PTRACE_RSP(GS) - MOVQ $0, CPU_ERROR_CODE(GS) // Clear error code. - MOVQ $0, CPU_ERROR_TYPE(GS) // Set error type to kernel. + MOVQ AX, ENTRY_SCRATCH0(GS) + MOVQ ENTRY_CPU_SELF(GS), AX // Load vCPU. + REGISTERS_SAVE(AX, CPU_REGISTERS) + MOVQ CX, CPU_REGISTERS+PTRACE_RIP(AX) + MOVQ R11, CPU_REGISTERS+PTRACE_FLAGS(AX) + MOVQ SP, CPU_REGISTERS+PTRACE_RSP(AX) + MOVQ ENTRY_SCRATCH0(GS), BX + MOVQ BX, CPU_REGISTERS+PTRACE_ORIGRAX(AX) + MOVQ BX, CPU_REGISTERS+PTRACE_RAX(AX) + MOVQ $0, CPU_ERROR_CODE(AX) // Clear error code. + MOVQ $0, CPU_ERROR_TYPE(AX) // Set error type to kernel. // Call the syscall trampoline. LOAD_KERNEL_STACK(GS) - MOVQ CPU_SELF(GS), AX // Load vCPU. PUSHQ AX // First argument (vCPU). CALL ·kernelSyscall(SB) // Call the trampoline. POPQ AX // Pop vCPU. @@ -230,16 +275,21 @@ TEXT ·exception(SB),NOSPLIT,$0 // ERROR_CODE (sp+8) // VECTOR (sp+0) // - TESTL $_RFLAGS_IF, 32(SP) + TESTL $_RFLAGS_IOPL0, 32(SP) JZ kernel user: SWAP_GS() ADDQ $-8, SP // Adjust for flags. MOVQ $_KERNEL_FLAGS, 0(SP); BYTE $0x9d; // Reset flags (POPFQ). - XCHGQ CPU_REGISTERS+PTRACE_RAX(GS), AX // Swap for user regs. + PUSHQ AX // Save user AX on stack. + MOVQ ENTRY_KERNEL_CR3(GS), AX // Get kernel cr3 on AX. + WRITE_CR3() // Switch to kernel cr3. + + MOVQ ENTRY_CPU_SELF(GS), AX // Load vCPU. + MOVQ CPU_REGISTERS+PTRACE_RAX(AX), AX // Get user regs. REGISTERS_SAVE(AX, 0) // Save all except IP, FLAGS, SP, AX. - MOVQ CPU_REGISTERS+PTRACE_RAX(GS), BX // Restore original AX. + POPQ BX // Restore original AX. MOVQ BX, PTRACE_RAX(AX) // Save it. MOVQ BX, PTRACE_ORIGRAX(AX) MOVQ 16(SP), BX; MOVQ BX, PTRACE_RIP(AX) @@ -249,34 +299,36 @@ user: MOVQ 48(SP), SI; MOVQ SI, PTRACE_SS(AX) // Copy out and return. + MOVQ ENTRY_CPU_SELF(GS), AX // Load vCPU. MOVQ 0(SP), BX // Load vector. MOVQ 8(SP), CX // Load error code. - MOVQ CPU_REGISTERS+PTRACE_RSP(GS), SP // Original stack (kernel version). - MOVQ CPU_REGISTERS+PTRACE_RBP(GS), BP // Original base pointer. - MOVQ CX, CPU_ERROR_CODE(GS) // Set error code. - MOVQ $1, CPU_ERROR_TYPE(GS) // Set error type to user. - MOVQ BX, 24(SP) // Output vector. + MOVQ CPU_REGISTERS+PTRACE_RSP(AX), SP // Original stack (kernel version). + MOVQ CPU_REGISTERS+PTRACE_RBP(AX), BP // Original base pointer. + MOVQ CX, CPU_ERROR_CODE(AX) // Set error code. + MOVQ $1, CPU_ERROR_TYPE(AX) // Set error type to user. + MOVQ BX, 32(SP) // Output vector. RET kernel: // As per above, we can save directly. - MOVQ AX, CPU_REGISTERS+PTRACE_RAX(GS) - MOVQ AX, CPU_REGISTERS+PTRACE_ORIGRAX(GS) - REGISTERS_SAVE(GS, CPU_REGISTERS) - MOVQ 16(SP), AX; MOVQ AX, CPU_REGISTERS+PTRACE_RIP(GS) - MOVQ 32(SP), BX; MOVQ BX, CPU_REGISTERS+PTRACE_FLAGS(GS) - MOVQ 40(SP), CX; MOVQ CX, CPU_REGISTERS+PTRACE_RSP(GS) + PUSHQ AX + MOVQ ENTRY_CPU_SELF(GS), AX // Load vCPU. + REGISTERS_SAVE(AX, CPU_REGISTERS) + POPQ BX + MOVQ BX, CPU_REGISTERS+PTRACE_RAX(AX) + MOVQ BX, CPU_REGISTERS+PTRACE_ORIGRAX(AX) + MOVQ 16(SP), BX; MOVQ BX, CPU_REGISTERS+PTRACE_RIP(AX) + MOVQ 32(SP), BX; MOVQ BX, CPU_REGISTERS+PTRACE_FLAGS(AX) + MOVQ 40(SP), BX; MOVQ BX, CPU_REGISTERS+PTRACE_RSP(AX) // Set the error code and adjust the stack. - MOVQ 8(SP), AX // Load the error code. - MOVQ AX, CPU_ERROR_CODE(GS) // Copy out to the CPU. - MOVQ $0, CPU_ERROR_TYPE(GS) // Set error type to kernel. + MOVQ 8(SP), BX // Load the error code. + MOVQ BX, CPU_ERROR_CODE(AX) // Copy out to the CPU. + MOVQ $0, CPU_ERROR_TYPE(AX) // Set error type to kernel. MOVQ 0(SP), BX // BX contains the vector. - ADDQ $48, SP // Drop the exception frame. // Call the exception trampoline. LOAD_KERNEL_STACK(GS) - MOVQ CPU_SELF(GS), AX // Load vCPU. PUSHQ BX // Second argument (vector). PUSHQ AX // First argument (vCPU). CALL ·kernelException(SB) // Call the trampoline. diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 106762665..494baaa4d 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -343,6 +343,8 @@ ADD $16, RSP, RSP; \ MOVD RSV_REG, PTRACE_R18(R20); \ MOVD RSV_REG_APP, PTRACE_R9(R20); \ + MRS TPIDR_EL0, R3; \ + MOVD R3, PTRACE_TLS(R20); \ WORD $0xd5384003; \ // MRS SPSR_EL1, R3 MOVD R3, PTRACE_PSTATE(R20); \ MRS ELR_EL1, R3; \ @@ -355,6 +357,8 @@ WORD $0xd538d092; \ //MRS TPIDR_EL1, R18 REGISTERS_SAVE(RSV_REG, CPU_REGISTERS); \ // Save sentry context. MOVD RSV_REG_APP, CPU_REGISTERS+PTRACE_R9(RSV_REG); \ + MRS TPIDR_EL0, R4; \ + MOVD R4, CPU_REGISTERS+PTRACE_TLS(RSV_REG); \ WORD $0xd5384004; \ // MRS SPSR_EL1, R4 MOVD R4, CPU_REGISTERS+PTRACE_PSTATE(RSV_REG); \ MRS ELR_EL1, R4; \ @@ -436,6 +440,8 @@ TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 MRS TPIDR_EL1, RSV_REG REGISTERS_SAVE(RSV_REG, CPU_REGISTERS) MOVD RSV_REG_APP, CPU_REGISTERS+PTRACE_R9(RSV_REG) + MRS TPIDR_EL0, R3 + MOVD R3, CPU_REGISTERS+PTRACE_TLS(RSV_REG) WORD $0xd5384003 // MRS SPSR_EL1, R3 MOVD R3, CPU_REGISTERS+PTRACE_PSTATE(RSV_REG) @@ -462,8 +468,18 @@ TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 MOVD PTRACE_PSTATE(RSV_REG_APP), R1 WORD $0xd5184001 //MSR R1, SPSR_EL1 + // need use kernel space address to excute below code, since + // after SWITCH_TO_APP_PAGETABLE the ASID is changed to app's + // ASID. + WORD $0x10000061 // ADR R1, do_exit_to_el0 + ORR $0xffff000000000000, R1, R1 + JMP (R1) + +do_exit_to_el0: // RSV_REG & RSV_REG_APP will be loaded at the end. REGISTERS_LOAD(RSV_REG_APP, 0) + MOVD PTRACE_TLS(RSV_REG_APP), RSV_REG + MSR RSV_REG, TPIDR_EL0 // switch to user pagetable. MOVD PTRACE_R18(RSV_REG_APP), RSV_REG diff --git a/pkg/sentry/platform/ring0/gen_offsets/BUILD b/pkg/sentry/platform/ring0/gen_offsets/BUILD index 549f3d228..9742308d8 100644 --- a/pkg/sentry/platform/ring0/gen_offsets/BUILD +++ b/pkg/sentry/platform/ring0/gen_offsets/BUILD @@ -24,7 +24,10 @@ go_binary( "defs_impl_arm64.go", "main.go", ], - visibility = ["//pkg/sentry/platform/ring0:__pkg__"], + visibility = [ + "//pkg/sentry/platform/kvm:__pkg__", + "//pkg/sentry/platform/ring0:__pkg__", + ], deps = [ "//pkg/cpuid", "//pkg/sentry/arch", diff --git a/pkg/sentry/platform/ring0/kernel.go b/pkg/sentry/platform/ring0/kernel.go index 021693791..264be23d3 100644 --- a/pkg/sentry/platform/ring0/kernel.go +++ b/pkg/sentry/platform/ring0/kernel.go @@ -19,8 +19,8 @@ package ring0 // N.B. that constraints on KernelOpts must be satisfied. // //go:nosplit -func (k *Kernel) Init(opts KernelOpts) { - k.init(opts) +func (k *Kernel) Init(opts KernelOpts, maxCPUs int) { + k.init(opts, maxCPUs) } // Halt halts execution. @@ -49,6 +49,11 @@ func (defaultHooks) KernelException(Vector) { // kernelSyscall is a trampoline. // +// When in amd64, it is called with %rip on the upper half, so it can +// NOT access to any global data which is not mapped on upper and must +// call to function pointers or interfaces to switch to the lower half +// so that callee can access to global data. +// // +checkescape:hard,stack // //go:nosplit @@ -58,6 +63,11 @@ func kernelSyscall(c *CPU) { // kernelException is a trampoline. // +// When in amd64, it is called with %rip on the upper half, so it can +// NOT access to any global data which is not mapped on upper and must +// call to function pointers or interfaces to switch to the lower half +// so that callee can access to global data. +// // +checkescape:hard,stack // //go:nosplit @@ -68,10 +78,10 @@ func kernelException(c *CPU, vector Vector) { // Init initializes a new CPU. // // Init allows embedding in other objects. -func (c *CPU) Init(k *Kernel, hooks Hooks) { - c.self = c // Set self reference. - c.kernel = k // Set kernel reference. - c.init() // Perform architectural init. +func (c *CPU) Init(k *Kernel, cpuID int, hooks Hooks) { + c.self = c // Set self reference. + c.kernel = k // Set kernel reference. + c.init(cpuID) // Perform architectural init. // Require hooks. if hooks != nil { diff --git a/pkg/sentry/platform/ring0/kernel_amd64.go b/pkg/sentry/platform/ring0/kernel_amd64.go index d37981dbf..3a9dff4cc 100644 --- a/pkg/sentry/platform/ring0/kernel_amd64.go +++ b/pkg/sentry/platform/ring0/kernel_amd64.go @@ -18,13 +18,42 @@ package ring0 import ( "encoding/binary" + "reflect" + + "gvisor.dev/gvisor/pkg/usermem" ) // init initializes architecture-specific state. -func (k *Kernel) init(opts KernelOpts) { +func (k *Kernel) init(opts KernelOpts, maxCPUs int) { // Save the root page tables. k.PageTables = opts.PageTables + entrySize := reflect.TypeOf(kernelEntry{}).Size() + var ( + entries []kernelEntry + padding = 1 + ) + for { + entries = make([]kernelEntry, maxCPUs+padding-1) + totalSize := entrySize * uintptr(maxCPUs+padding-1) + addr := reflect.ValueOf(&entries[0]).Pointer() + if addr&(usermem.PageSize-1) == 0 && totalSize >= usermem.PageSize { + // The runtime forces power-of-2 alignment for allocations, and we are therefore + // safe once the first address is aligned and the chunk is at least a full page. + break + } + padding = padding << 1 + } + k.cpuEntries = entries + + k.globalIDT = &idt64{} + if reflect.TypeOf(idt64{}).Size() != usermem.PageSize { + panic("Size of globalIDT should be PageSize") + } + if reflect.ValueOf(k.globalIDT).Pointer()&(usermem.PageSize-1) != 0 { + panic("Allocated globalIDT should be page aligned") + } + // Setup the IDT, which is uniform. for v, handler := range handlers { // Allow Breakpoint and Overflow to be called from all @@ -39,8 +68,26 @@ func (k *Kernel) init(opts KernelOpts) { } } +func (k *Kernel) EntryRegions() map[uintptr]uintptr { + regions := make(map[uintptr]uintptr) + + addr := reflect.ValueOf(&k.cpuEntries[0]).Pointer() + size := reflect.TypeOf(kernelEntry{}).Size() * uintptr(len(k.cpuEntries)) + end, _ := usermem.Addr(addr + size).RoundUp() + regions[uintptr(usermem.Addr(addr).RoundDown())] = uintptr(end) + + addr = reflect.ValueOf(k.globalIDT).Pointer() + size = reflect.TypeOf(idt64{}).Size() + end, _ = usermem.Addr(addr + size).RoundUp() + regions[uintptr(usermem.Addr(addr).RoundDown())] = uintptr(end) + + return regions +} + // init initializes architecture-specific state. -func (c *CPU) init() { +func (c *CPU) init(cpuID int) { + c.kernelEntry = &c.kernel.cpuEntries[cpuID] + c.cpuSelf = c // Null segment. c.gdt[0].setNull() @@ -65,6 +112,7 @@ func (c *CPU) init() { // Set the kernel stack pointer in the TSS (virtual address). stackAddr := c.StackTop() + c.stackTop = stackAddr c.tss.rsp0Lo = uint32(stackAddr) c.tss.rsp0Hi = uint32(stackAddr >> 32) c.tss.ist1Lo = uint32(stackAddr) @@ -183,7 +231,7 @@ func IsCanonical(addr uint64) bool { //go:nosplit func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { userCR3 := switchOpts.PageTables.CR3(!switchOpts.Flush, switchOpts.UserPCID) - kernelCR3 := c.kernel.PageTables.CR3(true, switchOpts.KernelPCID) + c.kernelCR3 = uintptr(c.kernel.PageTables.CR3(true, switchOpts.KernelPCID)) // Sanitize registers. regs := switchOpts.Registers @@ -197,15 +245,11 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { WriteFS(uintptr(regs.Fs_base)) // escapes: no. Set application FS. WriteGS(uintptr(regs.Gs_base)) // escapes: no. Set application GS. LoadFloatingPoint(switchOpts.FloatingPointState) // escapes: no. Copy in floating point. - jumpToKernel() // Switch to upper half. - writeCR3(uintptr(userCR3)) // Change to user address space. if switchOpts.FullRestore { - vector = iret(c, regs) + vector = iret(c, regs, uintptr(userCR3)) } else { - vector = sysret(c, regs) + vector = sysret(c, regs, uintptr(userCR3)) } - writeCR3(uintptr(kernelCR3)) // Return to kernel address space. - jumpToUser() // Return to lower half. SaveFloatingPoint(switchOpts.FloatingPointState) // escapes: no. Copy out floating point. WriteFS(uintptr(c.registers.Fs_base)) // escapes: no. Restore kernel FS. return @@ -219,7 +263,7 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { //go:nosplit func start(c *CPU) { // Save per-cpu & FS segment. - WriteGS(kernelAddr(c)) + WriteGS(kernelAddr(c.kernelEntry)) WriteFS(uintptr(c.registers.Fs_base)) // Initialize floating point. diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index 14774c5db..b294ccc7c 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -25,13 +25,13 @@ func HaltAndResume() func HaltEl1SvcAndResume() // init initializes architecture-specific state. -func (k *Kernel) init(opts KernelOpts) { +func (k *Kernel) init(opts KernelOpts, maxCPUs int) { // Save the root page tables. k.PageTables = opts.PageTables } // init initializes architecture-specific state. -func (c *CPU) init() { +func (c *CPU) init(cpuID int) { // Set the kernel stack pointer(virtual address). c.registers.Sp = uint64(c.StackTop()) @@ -64,11 +64,9 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { regs.Pstate |= UserFlagsSet LoadFloatingPoint(switchOpts.FloatingPointState) - SetTLS(regs.TPIDR_EL0) kernelExitToEl0() - regs.TPIDR_EL0 = GetTLS() SaveFloatingPoint(switchOpts.FloatingPointState) vector = c.vecCode diff --git a/pkg/sentry/platform/ring0/lib_amd64.go b/pkg/sentry/platform/ring0/lib_amd64.go index ca968a036..0ec5c3bc5 100644 --- a/pkg/sentry/platform/ring0/lib_amd64.go +++ b/pkg/sentry/platform/ring0/lib_amd64.go @@ -61,21 +61,9 @@ func wrgsbase(addr uintptr) // wrgsmsr writes to the GS_BASE MSR. func wrgsmsr(addr uintptr) -// writeCR3 writes the CR3 value. -func writeCR3(phys uintptr) - -// readCR3 reads the current CR3 value. -func readCR3() uintptr - // readCR2 reads the current CR2 value. func readCR2() uintptr -// jumpToKernel jumps to the kernel version of the current RIP. -func jumpToKernel() - -// jumpToUser jumps to the user version of the current RIP. -func jumpToUser() - // fninit initializes the floating point unit. func fninit() diff --git a/pkg/sentry/platform/ring0/lib_amd64.s b/pkg/sentry/platform/ring0/lib_amd64.s index 75d742750..2fe83568a 100644 --- a/pkg/sentry/platform/ring0/lib_amd64.s +++ b/pkg/sentry/platform/ring0/lib_amd64.s @@ -127,53 +127,6 @@ TEXT ·wrgsmsr(SB),NOSPLIT,$0-8 BYTE $0x0f; BYTE $0x30; // WRMSR RET -// jumpToUser changes execution to the user address. -// -// This works by changing the return value to the user version. -TEXT ·jumpToUser(SB),NOSPLIT,$0 - MOVQ 0(SP), AX - MOVQ ·KernelStartAddress(SB), BX - NOTQ BX - ANDQ BX, SP // Switch the stack. - ANDQ BX, BP // Switch the frame pointer. - ANDQ BX, AX // Future return value. - MOVQ AX, 0(SP) - RET - -// jumpToKernel changes execution to the kernel address space. -// -// This works by changing the return value to the kernel version. -TEXT ·jumpToKernel(SB),NOSPLIT,$0 - MOVQ 0(SP), AX - MOVQ ·KernelStartAddress(SB), BX - ORQ BX, SP // Switch the stack. - ORQ BX, BP // Switch the frame pointer. - ORQ BX, AX // Future return value. - MOVQ AX, 0(SP) - RET - -// writeCR3 writes the given CR3 value. -// -// The code corresponds to: -// -// mov %rax, %cr3 -// -TEXT ·writeCR3(SB),NOSPLIT,$0-8 - MOVQ cr3+0(FP), AX - BYTE $0x0f; BYTE $0x22; BYTE $0xd8; - RET - -// readCR3 reads the current CR3 value. -// -// The code corresponds to: -// -// mov %cr3, %rax -// -TEXT ·readCR3(SB),NOSPLIT,$0-8 - BYTE $0x0f; BYTE $0x20; BYTE $0xd8; - MOVQ AX, ret+0(FP) - RET - // readCR2 reads the current CR2 value. // // The code corresponds to: diff --git a/pkg/sentry/platform/ring0/offsets_amd64.go b/pkg/sentry/platform/ring0/offsets_amd64.go index b8ab120a0..290d94bd6 100644 --- a/pkg/sentry/platform/ring0/offsets_amd64.go +++ b/pkg/sentry/platform/ring0/offsets_amd64.go @@ -30,14 +30,22 @@ func Emit(w io.Writer) { c := &CPU{} fmt.Fprintf(w, "\n// CPU offsets.\n") - fmt.Fprintf(w, "#define CPU_SELF 0x%02x\n", reflect.ValueOf(&c.self).Pointer()-reflect.ValueOf(c).Pointer()) fmt.Fprintf(w, "#define CPU_REGISTERS 0x%02x\n", reflect.ValueOf(&c.registers).Pointer()-reflect.ValueOf(c).Pointer()) - fmt.Fprintf(w, "#define CPU_STACK_TOP 0x%02x\n", reflect.ValueOf(&c.stack[0]).Pointer()-reflect.ValueOf(c).Pointer()+uintptr(len(c.stack))) fmt.Fprintf(w, "#define CPU_ERROR_CODE 0x%02x\n", reflect.ValueOf(&c.errorCode).Pointer()-reflect.ValueOf(c).Pointer()) fmt.Fprintf(w, "#define CPU_ERROR_TYPE 0x%02x\n", reflect.ValueOf(&c.errorType).Pointer()-reflect.ValueOf(c).Pointer()) + fmt.Fprintf(w, "#define CPU_ENTRY 0x%02x\n", reflect.ValueOf(&c.kernelEntry).Pointer()-reflect.ValueOf(c).Pointer()) + + e := &kernelEntry{} + fmt.Fprintf(w, "\n// CPU entry offsets.\n") + fmt.Fprintf(w, "#define ENTRY_SCRATCH0 0x%02x\n", reflect.ValueOf(&e.scratch0).Pointer()-reflect.ValueOf(e).Pointer()) + fmt.Fprintf(w, "#define ENTRY_SCRATCH1 0x%02x\n", reflect.ValueOf(&e.scratch1).Pointer()-reflect.ValueOf(e).Pointer()) + fmt.Fprintf(w, "#define ENTRY_STACK_TOP 0x%02x\n", reflect.ValueOf(&e.stackTop).Pointer()-reflect.ValueOf(e).Pointer()) + fmt.Fprintf(w, "#define ENTRY_CPU_SELF 0x%02x\n", reflect.ValueOf(&e.cpuSelf).Pointer()-reflect.ValueOf(e).Pointer()) + fmt.Fprintf(w, "#define ENTRY_KERNEL_CR3 0x%02x\n", reflect.ValueOf(&e.kernelCR3).Pointer()-reflect.ValueOf(e).Pointer()) fmt.Fprintf(w, "\n// Bits.\n") fmt.Fprintf(w, "#define _RFLAGS_IF 0x%02x\n", _RFLAGS_IF) + fmt.Fprintf(w, "#define _RFLAGS_IOPL0 0x%02x\n", _RFLAGS_IOPL0) fmt.Fprintf(w, "#define _KERNEL_FLAGS 0x%02x\n", KernelFlagsSet) fmt.Fprintf(w, "\n// Vectors.\n") diff --git a/pkg/sentry/platform/ring0/offsets_arm64.go b/pkg/sentry/platform/ring0/offsets_arm64.go index 1d86b4bcf..45eba960d 100644 --- a/pkg/sentry/platform/ring0/offsets_arm64.go +++ b/pkg/sentry/platform/ring0/offsets_arm64.go @@ -125,4 +125,5 @@ func Emit(w io.Writer) { fmt.Fprintf(w, "#define PTRACE_SP 0x%02x\n", reflect.ValueOf(&p.Sp).Pointer()-reflect.ValueOf(p).Pointer()) fmt.Fprintf(w, "#define PTRACE_PC 0x%02x\n", reflect.ValueOf(&p.Pc).Pointer()-reflect.ValueOf(p).Pointer()) fmt.Fprintf(w, "#define PTRACE_PSTATE 0x%02x\n", reflect.ValueOf(&p.Pstate).Pointer()-reflect.ValueOf(p).Pointer()) + fmt.Fprintf(w, "#define PTRACE_TLS 0x%02x\n", reflect.ValueOf(&p.TPIDR_EL0).Pointer()-reflect.ValueOf(p).Pointer()) } diff --git a/pkg/sentry/platform/ring0/x86.go b/pkg/sentry/platform/ring0/x86.go index 9da0ea685..34fbc1c35 100644 --- a/pkg/sentry/platform/ring0/x86.go +++ b/pkg/sentry/platform/ring0/x86.go @@ -39,7 +39,9 @@ const ( _RFLAGS_AC = 1 << 18 _RFLAGS_NT = 1 << 14 - _RFLAGS_IOPL = 3 << 12 + _RFLAGS_IOPL0 = 1 << 12 + _RFLAGS_IOPL1 = 1 << 13 + _RFLAGS_IOPL = _RFLAGS_IOPL0 | _RFLAGS_IOPL1 _RFLAGS_DF = 1 << 10 _RFLAGS_IF = 1 << 9 _RFLAGS_STEP = 1 << 8 @@ -67,15 +69,45 @@ const ( KernelFlagsSet = _RFLAGS_RESERVED // UserFlagsSet are always set in userspace. - UserFlagsSet = _RFLAGS_RESERVED | _RFLAGS_IF + // + // _RFLAGS_IOPL is a set of two bits and it shows the I/O privilege + // level. The Current Privilege Level (CPL) of the task must be less + // than or equal to the IOPL in order for the task or program to access + // I/O ports. + // + // Here, _RFLAGS_IOPL0 is used only to determine whether the task is + // running in the kernel or userspace mode. In the user mode, the CPL is + // always 3 and it doesn't matter what IOPL is set if it is bellow CPL. + // + // We need to have one bit which will be always different in user and + // kernel modes. And we have to remember that even though we have + // KernelFlagsClear, we still can see some of these flags in the kernel + // mode. This can happen when the goruntime switches on a goroutine + // which has been saved in the host mode. On restore, the popf + // instruction is used to restore flags and this means that all flags + // what the goroutine has in the host mode will be restored in the + // kernel mode. + // + // _RFLAGS_IOPL0 is never set in host and kernel modes and we always set + // it in the user mode. So if this flag is set, the task is running in + // the user mode and if it isn't set, the task is running in the kernel + // mode. + UserFlagsSet = _RFLAGS_RESERVED | _RFLAGS_IF | _RFLAGS_IOPL0 // KernelFlagsClear should always be clear in the kernel. KernelFlagsClear = _RFLAGS_STEP | _RFLAGS_IF | _RFLAGS_IOPL | _RFLAGS_AC | _RFLAGS_NT // UserFlagsClear are always cleared in userspace. - UserFlagsClear = _RFLAGS_NT | _RFLAGS_IOPL + UserFlagsClear = _RFLAGS_NT | _RFLAGS_IOPL1 ) +// IsKernelFlags returns true if rflags coresponds to the kernel mode. +// +// go:nosplit +func IsKernelFlags(rflags uint64) bool { + return rflags&_RFLAGS_IOPL0 == 0 +} + // Vector is an exception vector. type Vector uintptr @@ -104,7 +136,7 @@ const ( VirtualizationException SecurityException = 0x1e SyscallInt80 = 0x80 - _NR_INTERRUPTS = SyscallInt80 + 1 + _NR_INTERRUPTS = 0x100 ) // System call vectors. diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index 97bc6027f..163af329b 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -52,6 +52,7 @@ var _ = socket.SocketVFS2(&socketVFS2{}) func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol int, fd int, flags uint32) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) s := &socketVFS2{ socketOpsCommon: socketOpsCommon{ @@ -77,6 +78,13 @@ func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol in return vfsfd, nil } +// Release implements vfs.FileDescriptionImpl.Release. +func (s *socketVFS2) Release(ctx context.Context) { + t := kernel.TaskFromContext(ctx) + t.Kernel().DeleteSocketVFS2(&s.vfsfd) + s.socketOpsCommon.Release(ctx) +} + // Readiness implements waiter.Waitable.Readiness. func (s *socketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { return s.socketOpsCommon.Readiness(mask) diff --git a/pkg/sentry/socket/netfilter/extensions.go b/pkg/sentry/socket/netfilter/extensions.go index 0336a32d8..549787955 100644 --- a/pkg/sentry/socket/netfilter/extensions.go +++ b/pkg/sentry/socket/netfilter/extensions.go @@ -19,6 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -37,7 +39,7 @@ type matchMaker interface { // name is the matcher name as stored in the xt_entry_match struct. name() string - // marshal converts from an stack.Matcher to an ABI struct. + // marshal converts from a stack.Matcher to an ABI struct. marshal(matcher stack.Matcher) []byte // unmarshal converts from the ABI matcher struct to an @@ -93,3 +95,71 @@ func unmarshalMatcher(match linux.XTEntryMatch, filter stack.IPHeaderFilter, buf } return matchMaker.unmarshal(buf, filter) } + +// targetMaker knows how to (un)marshal a target. Once registered, +// marshalTarget and unmarshalTarget can be used. +type targetMaker interface { + // id uniquely identifies the target. + id() stack.TargetID + + // marshal converts from a stack.Target to an ABI struct. + marshal(target stack.Target) []byte + + // unmarshal converts from the ABI matcher struct to a stack.Target. + unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) +} + +// targetMakers maps the TargetID of supported targets to the targetMaker that +// marshals and unmarshals it. It is immutable after package initialization. +var targetMakers = map[stack.TargetID]targetMaker{} + +func targetRevision(name string, netProto tcpip.NetworkProtocolNumber, rev uint8) (uint8, bool) { + tid := stack.TargetID{ + Name: name, + NetworkProtocol: netProto, + Revision: rev, + } + if _, ok := targetMakers[tid]; !ok { + return 0, false + } + + // Return the highest supported revision unless rev is higher. + for _, other := range targetMakers { + otherID := other.id() + if name == otherID.Name && netProto == otherID.NetworkProtocol && otherID.Revision > rev { + rev = uint8(otherID.Revision) + } + } + return rev, true +} + +// registerTargetMaker should be called by target extensions to register them +// with the netfilter package. +func registerTargetMaker(tm targetMaker) { + if _, ok := targetMakers[tm.id()]; ok { + panic(fmt.Sprintf("multiple targets registered with name %q.", tm.id())) + } + targetMakers[tm.id()] = tm +} + +func marshalTarget(target stack.Target) []byte { + targetMaker, ok := targetMakers[target.ID()] + if !ok { + panic(fmt.Sprintf("unknown target of type %T with id %+v.", target, target.ID())) + } + return targetMaker.marshal(target) +} + +func unmarshalTarget(target linux.XTEntryTarget, filter stack.IPHeaderFilter, buf []byte) (stack.Target, *syserr.Error) { + tid := stack.TargetID{ + Name: target.Name.String(), + NetworkProtocol: filter.NetworkProtocol(), + Revision: target.Revision, + } + targetMaker, ok := targetMakers[tid] + if !ok { + nflog("unsupported target with name %q", target.Name.String()) + return nil, syserr.ErrInvalidArgument + } + return targetMaker.unmarshal(buf, filter) +} diff --git a/pkg/sentry/socket/netfilter/ipv4.go b/pkg/sentry/socket/netfilter/ipv4.go index e4c55a100..b560fae0d 100644 --- a/pkg/sentry/socket/netfilter/ipv4.go +++ b/pkg/sentry/socket/netfilter/ipv4.go @@ -181,18 +181,23 @@ func modifyEntries4(stk *stack.Stack, optVal []byte, replace *linux.IPTReplace, nflog("entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) return nil, syserr.ErrInvalidArgument } - target, err := parseTarget(filter, optVal[:targetSize]) - if err != nil { - nflog("failed to parse target: %v", err) - return nil, syserr.ErrInvalidArgument - } - optVal = optVal[targetSize:] - table.Rules = append(table.Rules, stack.Rule{ + rule := stack.Rule{ Filter: filter, - Target: target, Matchers: matchers, - }) + } + + { + target, err := parseTarget(filter, optVal[:targetSize], false /* ipv6 */) + if err != nil { + nflog("failed to parse target: %v", err) + return nil, err + } + rule.Target = target + } + optVal = optVal[targetSize:] + + table.Rules = append(table.Rules, rule) offsets[offset] = int(entryIdx) offset += uint32(entry.NextOffset) diff --git a/pkg/sentry/socket/netfilter/ipv6.go b/pkg/sentry/socket/netfilter/ipv6.go index 3b2c1becd..4253f7bf4 100644 --- a/pkg/sentry/socket/netfilter/ipv6.go +++ b/pkg/sentry/socket/netfilter/ipv6.go @@ -184,18 +184,23 @@ func modifyEntries6(stk *stack.Stack, optVal []byte, replace *linux.IPTReplace, nflog("entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) return nil, syserr.ErrInvalidArgument } - target, err := parseTarget(filter, optVal[:targetSize]) - if err != nil { - nflog("failed to parse target: %v", err) - return nil, syserr.ErrInvalidArgument - } - optVal = optVal[targetSize:] - table.Rules = append(table.Rules, stack.Rule{ + rule := stack.Rule{ Filter: filter, - Target: target, Matchers: matchers, - }) + } + + { + target, err := parseTarget(filter, optVal[:targetSize], true /* ipv6 */) + if err != nil { + nflog("failed to parse target: %v", err) + return nil, err + } + rule.Target = target + } + optVal = optVal[targetSize:] + + table.Rules = append(table.Rules, rule) offsets[offset] = int(entryIdx) offset += uint32(entry.NextOffset) diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 871ea80ee..904a12e38 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -146,10 +147,6 @@ func SetEntries(stk *stack.Stack, optVal []byte, ipv6 bool) *syserr.Error { case stack.FilterTable: table = stack.EmptyFilterTable() case stack.NATTable: - if ipv6 { - nflog("IPv6 redirection not yet supported (gvisor.dev/issue/3549)") - return syserr.ErrInvalidArgument - } table = stack.EmptyNATTable() default: nflog("we don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String()) @@ -199,7 +196,7 @@ func SetEntries(stk *stack.Stack, optVal []byte, ipv6 bool) *syserr.Error { // Check the user chains. for ruleIdx, rule := range table.Rules { - if _, ok := rule.Target.(stack.UserChainTarget); !ok { + if _, ok := rule.Target.(*stack.UserChainTarget); !ok { continue } @@ -220,7 +217,7 @@ func SetEntries(stk *stack.Stack, optVal []byte, ipv6 bool) *syserr.Error { // Set each jump to point to the appropriate rule. Right now they hold byte // offsets. for ruleIdx, rule := range table.Rules { - jump, ok := rule.Target.(JumpTarget) + jump, ok := rule.Target.(*JumpTarget) if !ok { continue } @@ -311,7 +308,7 @@ func validUnderflow(rule stack.Rule, ipv6 bool) bool { return false } switch rule.Target.(type) { - case stack.AcceptTarget, stack.DropTarget: + case *stack.AcceptTarget, *stack.DropTarget: return true default: return false @@ -322,7 +319,7 @@ func isUnconditionalAccept(rule stack.Rule, ipv6 bool) bool { if !validUnderflow(rule, ipv6) { return false } - _, ok := rule.Target.(stack.AcceptTarget) + _, ok := rule.Target.(*stack.AcceptTarget) return ok } @@ -341,3 +338,20 @@ func hookFromLinux(hook int) stack.Hook { } panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain", hook)) } + +// TargetRevision returns a linux.XTGetRevision for a given target. It sets +// Revision to the highest supported value, unless the provided revision number +// is larger. +func TargetRevision(t *kernel.Task, revPtr usermem.Addr, netProto tcpip.NetworkProtocolNumber) (linux.XTGetRevision, *syserr.Error) { + // Read in the target name and version. + var rev linux.XTGetRevision + if _, err := rev.CopyIn(t, revPtr); err != nil { + return linux.XTGetRevision{}, syserr.FromError(err) + } + maxSupported, ok := targetRevision(rev.Name.String(), netProto, rev.Revision) + if !ok { + return linux.XTGetRevision{}, syserr.ErrProtocolNotSupported + } + rev.Revision = maxSupported + return rev, nil +} diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go index 87e41abd8..0e14447fe 100644 --- a/pkg/sentry/socket/netfilter/targets.go +++ b/pkg/sentry/socket/netfilter/targets.go @@ -15,255 +15,357 @@ package netfilter import ( - "errors" "fmt" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) -// errorTargetName is used to mark targets as error targets. Error targets -// shouldn't be reached - an error has occurred if we fall through to one. -const errorTargetName = "ERROR" +func init() { + // Standard targets include ACCEPT, DROP, RETURN, and JUMP. + registerTargetMaker(&standardTargetMaker{ + NetworkProtocol: header.IPv4ProtocolNumber, + }) + registerTargetMaker(&standardTargetMaker{ + NetworkProtocol: header.IPv6ProtocolNumber, + }) + + // Both user chains and actual errors are represented in iptables by + // error targets. + registerTargetMaker(&errorTargetMaker{ + NetworkProtocol: header.IPv4ProtocolNumber, + }) + registerTargetMaker(&errorTargetMaker{ + NetworkProtocol: header.IPv6ProtocolNumber, + }) + + registerTargetMaker(&redirectTargetMaker{ + NetworkProtocol: header.IPv4ProtocolNumber, + }) + registerTargetMaker(&nfNATTargetMaker{ + NetworkProtocol: header.IPv6ProtocolNumber, + }) +} -// redirectTargetName is used to mark targets as redirect targets. Redirect -// targets should be reached for only NAT and Mangle tables. These targets will -// change the destination port/destination IP for packets. -const redirectTargetName = "REDIRECT" +type standardTargetMaker struct { + NetworkProtocol tcpip.NetworkProtocolNumber +} -func marshalTarget(target stack.Target) []byte { +func (sm *standardTargetMaker) id() stack.TargetID { + // Standard targets have the empty string as a name and no revisions. + return stack.TargetID{ + NetworkProtocol: sm.NetworkProtocol, + } +} +func (*standardTargetMaker) marshal(target stack.Target) []byte { + // Translate verdicts the same way as the iptables tool. + var verdict int32 switch tg := target.(type) { - case stack.AcceptTarget: - return marshalStandardTarget(stack.RuleAccept) - case stack.DropTarget: - return marshalStandardTarget(stack.RuleDrop) - case stack.ErrorTarget: - return marshalErrorTarget(errorTargetName) - case stack.UserChainTarget: - return marshalErrorTarget(tg.Name) - case stack.ReturnTarget: - return marshalStandardTarget(stack.RuleReturn) - case stack.RedirectTarget: - return marshalRedirectTarget(tg) - case JumpTarget: - return marshalJumpTarget(tg) + case *stack.AcceptTarget: + verdict = -linux.NF_ACCEPT - 1 + case *stack.DropTarget: + verdict = -linux.NF_DROP - 1 + case *stack.ReturnTarget: + verdict = linux.NF_RETURN + case *JumpTarget: + verdict = int32(tg.Offset) default: panic(fmt.Errorf("unknown target of type %T", target)) } -} - -func marshalStandardTarget(verdict stack.RuleVerdict) []byte { - nflog("convert to binary: marshalling standard target") // The target's name will be the empty string. - target := linux.XTStandardTarget{ + xt := linux.XTStandardTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTStandardTarget, }, - Verdict: translateFromStandardVerdict(verdict), + Verdict: verdict, } ret := make([]byte, 0, linux.SizeOfXTStandardTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) + return binary.Marshal(ret, usermem.ByteOrder, xt) +} + +func (*standardTargetMaker) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) { + if len(buf) != linux.SizeOfXTStandardTarget { + nflog("buf has wrong size for standard target %d", len(buf)) + return nil, syserr.ErrInvalidArgument + } + var standardTarget linux.XTStandardTarget + buf = buf[:linux.SizeOfXTStandardTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) + + if standardTarget.Verdict < 0 { + // A Verdict < 0 indicates a non-jump verdict. + return translateToStandardTarget(standardTarget.Verdict, filter.NetworkProtocol()) + } + // A verdict >= 0 indicates a jump. + return &JumpTarget{ + Offset: uint32(standardTarget.Verdict), + NetworkProtocol: filter.NetworkProtocol(), + }, nil +} + +type errorTargetMaker struct { + NetworkProtocol tcpip.NetworkProtocolNumber +} + +func (em *errorTargetMaker) id() stack.TargetID { + // Error targets have no revision. + return stack.TargetID{ + Name: stack.ErrorTargetName, + NetworkProtocol: em.NetworkProtocol, + } } -func marshalErrorTarget(errorName string) []byte { +func (*errorTargetMaker) marshal(target stack.Target) []byte { + var errorName string + switch tg := target.(type) { + case *stack.ErrorTarget: + errorName = stack.ErrorTargetName + case *stack.UserChainTarget: + errorName = tg.Name + default: + panic(fmt.Sprintf("errorMakerTarget cannot marshal unknown type %T", target)) + } + // This is an error target named error - target := linux.XTErrorTarget{ + xt := linux.XTErrorTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTErrorTarget, }, } - copy(target.Name[:], errorName) - copy(target.Target.Name[:], errorTargetName) + copy(xt.Name[:], errorName) + copy(xt.Target.Name[:], stack.ErrorTargetName) ret := make([]byte, 0, linux.SizeOfXTErrorTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) + return binary.Marshal(ret, usermem.ByteOrder, xt) } -func marshalRedirectTarget(rt stack.RedirectTarget) []byte { +func (*errorTargetMaker) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) { + if len(buf) != linux.SizeOfXTErrorTarget { + nflog("buf has insufficient size for error target %d", len(buf)) + return nil, syserr.ErrInvalidArgument + } + var errorTarget linux.XTErrorTarget + buf = buf[:linux.SizeOfXTErrorTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &errorTarget) + + // Error targets are used in 2 cases: + // * An actual error case. These rules have an error + // named stack.ErrorTargetName. The last entry of the table + // is usually an error case to catch any packets that + // somehow fall through every rule. + // * To mark the start of a user defined chain. These + // rules have an error with the name of the chain. + switch name := errorTarget.Name.String(); name { + case stack.ErrorTargetName: + return &stack.ErrorTarget{NetworkProtocol: filter.NetworkProtocol()}, nil + default: + // User defined chain. + return &stack.UserChainTarget{ + Name: name, + NetworkProtocol: filter.NetworkProtocol(), + }, nil + } +} + +type redirectTargetMaker struct { + NetworkProtocol tcpip.NetworkProtocolNumber +} + +func (rm *redirectTargetMaker) id() stack.TargetID { + return stack.TargetID{ + Name: stack.RedirectTargetName, + NetworkProtocol: rm.NetworkProtocol, + } +} + +func (*redirectTargetMaker) marshal(target stack.Target) []byte { + rt := target.(*stack.RedirectTarget) // This is a redirect target named redirect - target := linux.XTRedirectTarget{ + xt := linux.XTRedirectTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTRedirectTarget, }, } - copy(target.Target.Name[:], redirectTargetName) + copy(xt.Target.Name[:], stack.RedirectTargetName) ret := make([]byte, 0, linux.SizeOfXTRedirectTarget) - target.NfRange.RangeSize = 1 - if rt.RangeProtoSpecified { - target.NfRange.RangeIPV4.Flags |= linux.NF_NAT_RANGE_PROTO_SPECIFIED + xt.NfRange.RangeSize = 1 + xt.NfRange.RangeIPV4.Flags |= linux.NF_NAT_RANGE_PROTO_SPECIFIED + xt.NfRange.RangeIPV4.MinPort = htons(rt.Port) + xt.NfRange.RangeIPV4.MaxPort = xt.NfRange.RangeIPV4.MinPort + return binary.Marshal(ret, usermem.ByteOrder, xt) +} + +func (*redirectTargetMaker) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) { + if len(buf) < linux.SizeOfXTRedirectTarget { + nflog("redirectTargetMaker: buf has insufficient size for redirect target %d", len(buf)) + return nil, syserr.ErrInvalidArgument + } + + if p := filter.Protocol; p != header.TCPProtocolNumber && p != header.UDPProtocolNumber { + nflog("redirectTargetMaker: bad proto %d", p) + return nil, syserr.ErrInvalidArgument + } + + var redirectTarget linux.XTRedirectTarget + buf = buf[:linux.SizeOfXTRedirectTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) + + // Copy linux.XTRedirectTarget to stack.RedirectTarget. + target := stack.RedirectTarget{NetworkProtocol: filter.NetworkProtocol()} + + // RangeSize should be 1. + nfRange := redirectTarget.NfRange + if nfRange.RangeSize != 1 { + nflog("redirectTargetMaker: bad rangesize %d", nfRange.RangeSize) + return nil, syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/170): Check if the flags are valid. + // Also check if we need to map ports or IP. + // For now, redirect target only supports destination port change. + // Port range and IP range are not supported yet. + if nfRange.RangeIPV4.Flags != linux.NF_NAT_RANGE_PROTO_SPECIFIED { + nflog("redirectTargetMaker: invalid range flags %d", nfRange.RangeIPV4.Flags) + return nil, syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/170): Port range is not supported yet. + if nfRange.RangeIPV4.MinPort != nfRange.RangeIPV4.MaxPort { + nflog("redirectTargetMaker: MinPort != MaxPort (%d, %d)", nfRange.RangeIPV4.MinPort, nfRange.RangeIPV4.MaxPort) + return nil, syserr.ErrInvalidArgument } - // Convert port from little endian to big endian. - port := make([]byte, 2) - binary.LittleEndian.PutUint16(port, rt.MinPort) - target.NfRange.RangeIPV4.MinPort = binary.BigEndian.Uint16(port) - binary.LittleEndian.PutUint16(port, rt.MaxPort) - target.NfRange.RangeIPV4.MaxPort = binary.BigEndian.Uint16(port) - return binary.Marshal(ret, usermem.ByteOrder, target) + if nfRange.RangeIPV4.MinIP != nfRange.RangeIPV4.MaxIP { + nflog("redirectTargetMaker: MinIP != MaxIP (%d, %d)", nfRange.RangeIPV4.MinPort, nfRange.RangeIPV4.MaxPort) + return nil, syserr.ErrInvalidArgument + } + + target.Addr = tcpip.Address(nfRange.RangeIPV4.MinIP[:]) + target.Port = ntohs(nfRange.RangeIPV4.MinPort) + + return &target, nil } -func marshalJumpTarget(jt JumpTarget) []byte { - nflog("convert to binary: marshalling jump target") +type nfNATTarget struct { + Target linux.XTEntryTarget + Range linux.NFNATRange +} - // The target's name will be the empty string. - target := linux.XTStandardTarget{ +const nfNATMarhsalledSize = linux.SizeOfXTEntryTarget + linux.SizeOfNFNATRange + +type nfNATTargetMaker struct { + NetworkProtocol tcpip.NetworkProtocolNumber +} + +func (rm *nfNATTargetMaker) id() stack.TargetID { + return stack.TargetID{ + Name: stack.RedirectTargetName, + NetworkProtocol: rm.NetworkProtocol, + } +} + +func (*nfNATTargetMaker) marshal(target stack.Target) []byte { + rt := target.(*stack.RedirectTarget) + nt := nfNATTarget{ Target: linux.XTEntryTarget{ - TargetSize: linux.SizeOfXTStandardTarget, + TargetSize: nfNATMarhsalledSize, + }, + Range: linux.NFNATRange{ + Flags: linux.NF_NAT_RANGE_PROTO_SPECIFIED, }, - // Verdict is overloaded by the ABI. When positive, it holds - // the jump offset from the start of the table. - Verdict: int32(jt.Offset), } + copy(nt.Target.Name[:], stack.RedirectTargetName) + copy(nt.Range.MinAddr[:], rt.Addr) + copy(nt.Range.MaxAddr[:], rt.Addr) - ret := make([]byte, 0, linux.SizeOfXTStandardTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) + nt.Range.MinProto = htons(rt.Port) + nt.Range.MaxProto = nt.Range.MinProto + + ret := make([]byte, 0, nfNATMarhsalledSize) + return binary.Marshal(ret, usermem.ByteOrder, nt) } -// translateFromStandardVerdict translates verdicts the same way as the iptables -// tool. -func translateFromStandardVerdict(verdict stack.RuleVerdict) int32 { - switch verdict { - case stack.RuleAccept: - return -linux.NF_ACCEPT - 1 - case stack.RuleDrop: - return -linux.NF_DROP - 1 - case stack.RuleReturn: - return linux.NF_RETURN - default: - // TODO(gvisor.dev/issue/170): Support Jump. - panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) +func (*nfNATTargetMaker) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) { + if size := nfNATMarhsalledSize; len(buf) < size { + nflog("nfNATTargetMaker: buf has insufficient size (%d) for nfNAT target (%d)", len(buf), size) + return nil, syserr.ErrInvalidArgument } + + if p := filter.Protocol; p != header.TCPProtocolNumber && p != header.UDPProtocolNumber { + nflog("nfNATTargetMaker: bad proto %d", p) + return nil, syserr.ErrInvalidArgument + } + + var natRange linux.NFNATRange + buf = buf[linux.SizeOfXTEntryTarget:nfNATMarhsalledSize] + binary.Unmarshal(buf, usermem.ByteOrder, &natRange) + + // We don't support port or address ranges. + if natRange.MinAddr != natRange.MaxAddr { + nflog("nfNATTargetMaker: MinAddr and MaxAddr are different") + return nil, syserr.ErrInvalidArgument + } + if natRange.MinProto != natRange.MaxProto { + nflog("nfNATTargetMaker: MinProto and MaxProto are different") + return nil, syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/3549): Check for other flags. + // For now, redirect target only supports destination change. + if natRange.Flags != linux.NF_NAT_RANGE_PROTO_SPECIFIED { + nflog("nfNATTargetMaker: invalid range flags %d", natRange.Flags) + return nil, syserr.ErrInvalidArgument + } + + target := stack.RedirectTarget{ + NetworkProtocol: filter.NetworkProtocol(), + Addr: tcpip.Address(natRange.MinAddr[:]), + Port: ntohs(natRange.MinProto), + } + + return &target, nil } // translateToStandardTarget translates from the value in a // linux.XTStandardTarget to an stack.Verdict. -func translateToStandardTarget(val int32) (stack.Target, error) { +func translateToStandardTarget(val int32, netProto tcpip.NetworkProtocolNumber) (stack.Target, *syserr.Error) { // TODO(gvisor.dev/issue/170): Support other verdicts. switch val { case -linux.NF_ACCEPT - 1: - return stack.AcceptTarget{}, nil + return &stack.AcceptTarget{NetworkProtocol: netProto}, nil case -linux.NF_DROP - 1: - return stack.DropTarget{}, nil + return &stack.DropTarget{NetworkProtocol: netProto}, nil case -linux.NF_QUEUE - 1: - return nil, errors.New("unsupported iptables verdict QUEUE") + nflog("unsupported iptables verdict QUEUE") + return nil, syserr.ErrInvalidArgument case linux.NF_RETURN: - return stack.ReturnTarget{}, nil + return &stack.ReturnTarget{NetworkProtocol: netProto}, nil default: - return nil, fmt.Errorf("unknown iptables verdict %d", val) + nflog("unknown iptables verdict %d", val) + return nil, syserr.ErrInvalidArgument } } // parseTarget parses a target from optVal. optVal should contain only the // target. -func parseTarget(filter stack.IPHeaderFilter, optVal []byte) (stack.Target, error) { +func parseTarget(filter stack.IPHeaderFilter, optVal []byte, ipv6 bool) (stack.Target, *syserr.Error) { nflog("set entries: parsing target of size %d", len(optVal)) if len(optVal) < linux.SizeOfXTEntryTarget { - return nil, fmt.Errorf("optVal has insufficient size for entry target %d", len(optVal)) + nflog("optVal has insufficient size for entry target %d", len(optVal)) + return nil, syserr.ErrInvalidArgument } var target linux.XTEntryTarget buf := optVal[:linux.SizeOfXTEntryTarget] binary.Unmarshal(buf, usermem.ByteOrder, &target) - switch target.Name.String() { - case "": - // Standard target. - if len(optVal) != linux.SizeOfXTStandardTarget { - return nil, fmt.Errorf("optVal has wrong size for standard target %d", len(optVal)) - } - var standardTarget linux.XTStandardTarget - buf = optVal[:linux.SizeOfXTStandardTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) - - if standardTarget.Verdict < 0 { - // A Verdict < 0 indicates a non-jump verdict. - return translateToStandardTarget(standardTarget.Verdict) - } - // A verdict >= 0 indicates a jump. - return JumpTarget{Offset: uint32(standardTarget.Verdict)}, nil - - case errorTargetName: - // Error target. - if len(optVal) != linux.SizeOfXTErrorTarget { - return nil, fmt.Errorf("optVal has insufficient size for error target %d", len(optVal)) - } - var errorTarget linux.XTErrorTarget - buf = optVal[:linux.SizeOfXTErrorTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &errorTarget) - - // Error targets are used in 2 cases: - // * An actual error case. These rules have an error - // named errorTargetName. The last entry of the table - // is usually an error case to catch any packets that - // somehow fall through every rule. - // * To mark the start of a user defined chain. These - // rules have an error with the name of the chain. - switch name := errorTarget.Name.String(); name { - case errorTargetName: - nflog("set entries: error target") - return stack.ErrorTarget{}, nil - default: - // User defined chain. - nflog("set entries: user-defined target %q", name) - return stack.UserChainTarget{Name: name}, nil - } - - case redirectTargetName: - // Redirect target. - if len(optVal) < linux.SizeOfXTRedirectTarget { - return nil, fmt.Errorf("netfilter.SetEntries: optVal has insufficient size for redirect target %d", len(optVal)) - } - - if p := filter.Protocol; p != header.TCPProtocolNumber && p != header.UDPProtocolNumber { - return nil, fmt.Errorf("netfilter.SetEntries: bad proto %d", p) - } - - var redirectTarget linux.XTRedirectTarget - buf = optVal[:linux.SizeOfXTRedirectTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) - - // Copy linux.XTRedirectTarget to stack.RedirectTarget. - var target stack.RedirectTarget - nfRange := redirectTarget.NfRange - - // RangeSize should be 1. - if nfRange.RangeSize != 1 { - return nil, fmt.Errorf("netfilter.SetEntries: bad rangesize %d", nfRange.RangeSize) - } - - // TODO(gvisor.dev/issue/170): Check if the flags are valid. - // Also check if we need to map ports or IP. - // For now, redirect target only supports destination port change. - // Port range and IP range are not supported yet. - if nfRange.RangeIPV4.Flags&linux.NF_NAT_RANGE_PROTO_SPECIFIED == 0 { - return nil, fmt.Errorf("netfilter.SetEntries: invalid range flags %d", nfRange.RangeIPV4.Flags) - } - target.RangeProtoSpecified = true - - target.MinIP = tcpip.Address(nfRange.RangeIPV4.MinIP[:]) - target.MaxIP = tcpip.Address(nfRange.RangeIPV4.MaxIP[:]) - - // TODO(gvisor.dev/issue/170): Port range is not supported yet. - if nfRange.RangeIPV4.MinPort != nfRange.RangeIPV4.MaxPort { - return nil, fmt.Errorf("netfilter.SetEntries: minport != maxport (%d, %d)", nfRange.RangeIPV4.MinPort, nfRange.RangeIPV4.MaxPort) - } - - // Convert port from big endian to little endian. - port := make([]byte, 2) - binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MinPort) - target.MinPort = binary.LittleEndian.Uint16(port) - - binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MaxPort) - target.MaxPort = binary.LittleEndian.Uint16(port) - return target, nil - } - // Unknown target. - return nil, fmt.Errorf("unknown target %q doesn't exist or isn't supported yet", target.Name.String()) + return unmarshalTarget(target, filter, optVal) } // JumpTarget implements stack.Target. @@ -274,9 +376,31 @@ type JumpTarget struct { // RuleNum is the rule to jump to. RuleNum int + + // NetworkProtocol is the network protocol the target is used with. + NetworkProtocol tcpip.NetworkProtocolNumber +} + +// ID implements Target.ID. +func (jt *JumpTarget) ID() stack.TargetID { + return stack.TargetID{ + NetworkProtocol: jt.NetworkProtocol, + } } // Action implements stack.Target.Action. -func (jt JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrack, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) { +func (jt *JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrack, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) { return stack.RuleJump, jt.RuleNum } + +func ntohs(port uint16) uint16 { + buf := make([]byte, 2) + binary.BigEndian.PutUint16(buf, port) + return usermem.ByteOrder.Uint16(buf) +} + +func htons(port uint16) uint16 { + buf := make([]byte, 2) + usermem.ByteOrder.PutUint16(buf, port) + return binary.BigEndian.Uint16(buf) +} diff --git a/pkg/sentry/socket/netlink/provider_vfs2.go b/pkg/sentry/socket/netlink/provider_vfs2.go index bb205be0d..e8930f031 100644 --- a/pkg/sentry/socket/netlink/provider_vfs2.go +++ b/pkg/sentry/socket/netlink/provider_vfs2.go @@ -52,6 +52,7 @@ func (*socketProviderVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol vfsfd := &s.vfsfd mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, DenyPWrite: true, diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go index a38d25da9..c83b23242 100644 --- a/pkg/sentry/socket/netlink/socket_vfs2.go +++ b/pkg/sentry/socket/netlink/socket_vfs2.go @@ -82,6 +82,13 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV return fd, nil } +// Release implements vfs.FileDescriptionImpl.Release. +func (s *SocketVFS2) Release(ctx context.Context) { + t := kernel.TaskFromContext(ctx) + t.Kernel().DeleteSocketVFS2(&s.vfsfd) + s.socketOpsCommon.Release(ctx) +} + // Readiness implements waiter.Waitable.Readiness. func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { return s.socketOpsCommon.Readiness(mask) diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 6fede181a..87e30d742 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -198,7 +198,6 @@ var Metrics = tcpip.Stats{ PacketsSent: mustCreateMetric("/netstack/udp/packets_sent", "Number of UDP datagrams sent."), PacketSendErrors: mustCreateMetric("/netstack/udp/packet_send_errors", "Number of UDP datagrams failed to be sent."), ChecksumErrors: mustCreateMetric("/netstack/udp/checksum_errors", "Number of UDP datagrams dropped due to bad checksums."), - InvalidSourceAddress: mustCreateMetric("/netstack/udp/invalid_source", "Number of UDP datagrams dropped due to invalid source address."), }, } @@ -1513,8 +1512,17 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name return &vP, nil case linux.IP6T_ORIGINAL_DST: - // TODO(gvisor.dev/issue/170): ip6tables. - return nil, syserr.ErrInvalidArgument + if outLen < int(binary.Size(linux.SockAddrInet6{})) { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.OriginalDestinationOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + a, _ := ConvertAddress(linux.AF_INET6, tcpip.FullAddress(v)) + return a.(*linux.SockAddrInet6), nil case linux.IP6T_SO_GET_INFO: if outLen < linux.SizeOfIPTGetinfo { @@ -1556,6 +1564,26 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name } return &entries, nil + case linux.IP6T_SO_GET_REVISION_TARGET: + if outLen < linux.SizeOfXTGetRevision { + return nil, syserr.ErrInvalidArgument + } + + // Only valid for raw IPv6 sockets. + if family, skType, _ := s.Type(); family != linux.AF_INET6 || skType != linux.SOCK_RAW { + return nil, syserr.ErrProtocolNotAvailable + } + + stack := inet.StackFromContext(t) + if stack == nil { + return nil, syserr.ErrNoDevice + } + ret, err := netfilter.TargetRevision(t, outPtr, header.IPv6ProtocolNumber) + if err != nil { + return nil, err + } + return &ret, nil + default: emitUnimplementedEventIPv6(t, name) } @@ -1719,6 +1747,26 @@ func getSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in } return &entries, nil + case linux.IPT_SO_GET_REVISION_TARGET: + if outLen < linux.SizeOfXTGetRevision { + return nil, syserr.ErrInvalidArgument + } + + // Only valid for raw IPv4 sockets. + if family, skType, _ := s.Type(); family != linux.AF_INET || skType != linux.SOCK_RAW { + return nil, syserr.ErrProtocolNotAvailable + } + + stack := inet.StackFromContext(t) + if stack == nil { + return nil, syserr.ErrNoDevice + } + ret, err := netfilter.TargetRevision(t, outPtr, header.IPv4ProtocolNumber) + if err != nil { + return nil, err + } + return &ret, nil + default: emitUnimplementedEventIP(t, name) } diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index c0212ad76..4c6791fff 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -79,6 +79,13 @@ func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, qu return vfsfd, nil } +// Release implements vfs.FileDescriptionImpl.Release. +func (s *SocketVFS2) Release(ctx context.Context) { + t := kernel.TaskFromContext(ctx) + t.Kernel().DeleteSocketVFS2(&s.vfsfd) + s.socketOpsCommon.Release(ctx) +} + // Readiness implements waiter.Waitable.Readiness. func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { return s.socketOpsCommon.Readiness(mask) diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index a89583dad..cc7408698 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -7,10 +7,21 @@ go_template_instance( name = "socket_refs", out = "socket_refs.go", package = "unix", - prefix = "socketOpsCommon", + prefix = "socketOperations", template = "//pkg/refs_vfs2:refs_template", types = { - "T": "socketOpsCommon", + "T": "SocketOperations", + }, +) + +go_template_instance( + name = "socket_vfs2_refs", + out = "socket_vfs2_refs.go", + package = "unix", + prefix = "socketVFS2", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "SocketVFS2", }, ) @@ -20,6 +31,7 @@ go_library( "device.go", "io.go", "socket_refs.go", + "socket_vfs2_refs.go", "unix.go", "unix_vfs2.go", ], diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 917055cea..f80011ce4 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -55,6 +55,7 @@ type SocketOperations struct { fsutil.FileNoopFlush `state:"nosave"` fsutil.FileUseInodeUnstableAttr `state:"nosave"` + socketOperationsRefs socketOpsCommon } @@ -84,11 +85,27 @@ func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, sty return fs.NewFile(ctx, d, flags, &s) } +// DecRef implements RefCounter.DecRef. +func (s *SocketOperations) DecRef(ctx context.Context) { + s.socketOperationsRefs.DecRef(func() { + s.ep.Close(ctx) + if s.abstractNamespace != nil { + s.abstractNamespace.Remove(s.abstractName, s) + } + }) +} + +// Release implemements fs.FileOperations.Release. +func (s *SocketOperations) Release(ctx context.Context) { + // Release only decrements a reference on s because s may be referenced in + // the abstract socket namespace. + s.DecRef(ctx) +} + // socketOpsCommon contains the socket operations common to VFS1 and VFS2. // // +stateify savable type socketOpsCommon struct { - socketOpsCommonRefs socket.SendReceiveTimeout ep transport.Endpoint @@ -101,23 +118,6 @@ type socketOpsCommon struct { abstractNamespace *kernel.AbstractSocketNamespace } -// DecRef implements RefCounter.DecRef. -func (s *socketOpsCommon) DecRef(ctx context.Context) { - s.socketOpsCommonRefs.DecRef(func() { - s.ep.Close(ctx) - if s.abstractNamespace != nil { - s.abstractNamespace.Remove(s.abstractName, s) - } - }) -} - -// Release implemements fs.FileOperations.Release. -func (s *socketOpsCommon) Release(ctx context.Context) { - // Release only decrements a reference on s because s may be referenced in - // the abstract socket namespace. - s.DecRef(ctx) -} - func (s *socketOpsCommon) isPacket() bool { switch s.stype { case linux.SOCK_DGRAM, linux.SOCK_SEQPACKET: diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 3688f22d2..3345124cc 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -37,12 +37,15 @@ import ( // SocketVFS2 implements socket.SocketVFS2 (and by extension, // vfs.FileDescriptionImpl) for Unix sockets. +// +// +stateify savable type SocketVFS2 struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl vfs.LockFD + socketVFS2Refs socketOpsCommon } @@ -53,6 +56,7 @@ var _ = socket.SocketVFS2(&SocketVFS2{}) func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &vfs.FileLocks{}) if err != nil { @@ -88,6 +92,25 @@ func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint3 return vfsfd, nil } +// DecRef implements RefCounter.DecRef. +func (s *SocketVFS2) DecRef(ctx context.Context) { + s.socketVFS2Refs.DecRef(func() { + t := kernel.TaskFromContext(ctx) + t.Kernel().DeleteSocketVFS2(&s.vfsfd) + s.ep.Close(ctx) + if s.abstractNamespace != nil { + s.abstractNamespace.Remove(s.abstractName, s) + } + }) +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (s *SocketVFS2) Release(ctx context.Context) { + // Release only decrements a reference on s because s may be referenced in + // the abstract socket namespace. + s.DecRef(ctx) +} + // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // a transport.Endpoint. func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { diff --git a/pkg/sentry/state/state.go b/pkg/sentry/state/state.go index a06c9b8ab..245d2c5cf 100644 --- a/pkg/sentry/state/state.go +++ b/pkg/sentry/state/state.go @@ -61,8 +61,10 @@ func (opts SaveOpts) Save(k *kernel.Kernel, w *watchdog.Watchdog) error { log.Infof("Sandbox save started, pausing all tasks.") k.Pause() k.ReceiveTaskStates() - defer k.Unpause() - defer log.Infof("Tasks resumed after save.") + defer func() { + k.Unpause() + log.Infof("Tasks resumed after save.") + }() w.Stop() defer w.Start() diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 52281ccc2..396744597 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -17,7 +17,6 @@ package strace import ( - "encoding/binary" "fmt" "strconv" "strings" @@ -294,7 +293,7 @@ func itimerval(t *kernel.Task, addr usermem.Addr) string { } interval := timeval(t, addr) - value := timeval(t, addr+usermem.Addr(binary.Size(linux.Timeval{}))) + value := timeval(t, addr+usermem.Addr((*linux.Timeval)(nil).SizeBytes())) return fmt.Sprintf("%#x {interval=%s, value=%s}", addr, interval, value) } @@ -304,7 +303,7 @@ func itimerspec(t *kernel.Task, addr usermem.Addr) string { } interval := timespec(t, addr) - value := timespec(t, addr+usermem.Addr(binary.Size(linux.Timespec{}))) + value := timespec(t, addr+usermem.Addr((*linux.Timespec)(nil).SizeBytes())) return fmt.Sprintf("%#x {interval=%s, value=%s}", addr, interval, value) } diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 75752b2e6..a2e441448 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -21,6 +21,7 @@ go_library( "sys_identity.go", "sys_inotify.go", "sys_lseek.go", + "sys_membarrier.go", "sys_mempolicy.go", "sys_mmap.go", "sys_mount.go", diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index 5f26697d2..9c9def7cd 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -376,7 +376,7 @@ var AMD64 = &kernel.SyscallTable{ 321: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), 322: syscalls.Supported("execveat", Execveat), 323: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) - 324: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) + 324: syscalls.PartiallySupported("membarrier", Membarrier, "Not supported on all platforms.", nil), 325: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), // Syscalls implemented after 325 are "backports" from versions @@ -527,8 +527,8 @@ var ARM64 = &kernel.SyscallTable{ 96: syscalls.Supported("set_tid_address", SetTidAddress), 97: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), 98: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), - 99: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 100: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 99: syscalls.Supported("set_robust_list", SetRobustList), + 100: syscalls.Supported("get_robust_list", GetRobustList), 101: syscalls.Supported("nanosleep", Nanosleep), 102: syscalls.Supported("getitimer", Getitimer), 103: syscalls.Supported("setitimer", Setitimer), @@ -695,7 +695,7 @@ var ARM64 = &kernel.SyscallTable{ 280: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), 281: syscalls.Supported("execveat", Execveat), 282: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) - 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) + 283: syscalls.PartiallySupported("membarrier", Membarrier, "Not supported on all platforms.", nil), 284: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), // Syscalls after 284 are "backports" from versions of Linux after 4.4. diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 98331eb3c..519066a47 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -84,6 +84,7 @@ func fileOpOn(t *kernel.Task, dirFD int32, path string, resolve bool, fn func(ro } rel = f.Dirent if !fs.IsDir(rel.Inode.StableAttr) { + f.DecRef(t) return syserror.ENOTDIR } } diff --git a/pkg/sentry/syscalls/linux/sys_membarrier.go b/pkg/sentry/syscalls/linux/sys_membarrier.go new file mode 100644 index 000000000..63ee5d435 --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_membarrier.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 linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" +) + +// Membarrier implements syscall membarrier(2). +func Membarrier(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + cmd := args[0].Int() + flags := args[1].Uint() + + switch cmd { + case linux.MEMBARRIER_CMD_QUERY: + if flags != 0 { + return 0, nil, syserror.EINVAL + } + var supportedCommands uintptr + if t.Kernel().Platform.HaveGlobalMemoryBarrier() { + supportedCommands |= linux.MEMBARRIER_CMD_GLOBAL | + linux.MEMBARRIER_CMD_GLOBAL_EXPEDITED | + linux.MEMBARRIER_CMD_REGISTER_GLOBAL_EXPEDITED | + linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED | + linux.MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED + } + if t.RSeqAvailable() { + supportedCommands |= linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ | + linux.MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED_RSEQ + } + return supportedCommands, nil, nil + case linux.MEMBARRIER_CMD_GLOBAL, linux.MEMBARRIER_CMD_GLOBAL_EXPEDITED, linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED: + if flags != 0 { + return 0, nil, syserror.EINVAL + } + if !t.Kernel().Platform.HaveGlobalMemoryBarrier() { + return 0, nil, syserror.EINVAL + } + if cmd == linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED && !t.MemoryManager().IsMembarrierPrivateEnabled() { + return 0, nil, syserror.EPERM + } + return 0, nil, t.Kernel().Platform.GlobalMemoryBarrier() + case linux.MEMBARRIER_CMD_REGISTER_GLOBAL_EXPEDITED: + if flags != 0 { + return 0, nil, syserror.EINVAL + } + if !t.Kernel().Platform.HaveGlobalMemoryBarrier() { + return 0, nil, syserror.EINVAL + } + // no-op + return 0, nil, nil + case linux.MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED: + if flags != 0 { + return 0, nil, syserror.EINVAL + } + if !t.Kernel().Platform.HaveGlobalMemoryBarrier() { + return 0, nil, syserror.EINVAL + } + t.MemoryManager().EnableMembarrierPrivate() + return 0, nil, nil + case linux.MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ: + if flags&^linux.MEMBARRIER_CMD_FLAG_CPU != 0 { + return 0, nil, syserror.EINVAL + } + if !t.RSeqAvailable() { + return 0, nil, syserror.EINVAL + } + if !t.MemoryManager().IsMembarrierRSeqEnabled() { + return 0, nil, syserror.EPERM + } + // MEMBARRIER_CMD_FLAG_CPU and cpu_id are ignored since we don't have + // the ability to preempt specific CPUs. + return 0, nil, t.Kernel().Platform.PreemptAllCPUs() + case linux.MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED_RSEQ: + if flags != 0 { + return 0, nil, syserror.EINVAL + } + if !t.RSeqAvailable() { + return 0, nil, syserror.EINVAL + } + t.MemoryManager().EnableMembarrierRSeq() + return 0, nil, nil + default: + // Probably a command we don't implement. + t.Kernel().EmitUnimplementedEvent(t) + return 0, nil, syserror.EINVAL + } +} diff --git a/pkg/sentry/syscalls/linux/sys_sysinfo.go b/pkg/sentry/syscalls/linux/sys_sysinfo.go index 674d341b6..6320593f0 100644 --- a/pkg/sentry/syscalls/linux/sys_sysinfo.go +++ b/pkg/sentry/syscalls/linux/sys_sysinfo.go @@ -26,8 +26,12 @@ func Sysinfo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca addr := args[0].Pointer() mf := t.Kernel().MemoryFile() - mf.UpdateUsage() - _, totalUsage := usage.MemoryAccounting.Copy() + mfUsage, err := mf.TotalUsage() + if err != nil { + return 0, nil, err + } + memStats, _ := usage.MemoryAccounting.Copy() + totalUsage := mfUsage + memStats.Mapped totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) memFree := totalSize - totalUsage if memFree > totalSize { @@ -37,12 +41,12 @@ func Sysinfo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // Only a subset of the fields in sysinfo_t make sense to return. si := linux.Sysinfo{ - Procs: uint16(len(t.PIDNamespace().Tasks())), + Procs: uint16(t.Kernel().TaskSet().Root.NumTasks()), Uptime: t.Kernel().MonotonicClock().Now().Seconds(), TotalRAM: totalSize, FreeRAM: memFree, Unit: 1, } - _, err := si.CopyOut(t, addr) + _, err = si.CopyOut(t, addr) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index f55d74cd2..bf5c1171f 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" + slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -146,11 +147,6 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal panic("at least one end of splice must be a pipe") } - if n == 0 && err == io.EOF { - // We reached the end of the file. Eat the error and exit the loop. - err = nil - break - } if n != 0 || err != syserror.ErrWouldBlock || nonBlock { break } @@ -171,15 +167,16 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } } - if n == 0 { - return 0, nil, err + if n != 0 { + // On Linux, inotify behavior is not very consistent with splice(2). We try + // our best to emulate Linux for very basic calls to splice, where for some + // reason, events are generated for output files, but not input files. + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } - // On Linux, inotify behavior is not very consistent with splice(2). We try - // our best to emulate Linux for very basic calls to splice, where for some - // reason, events are generated for output files, but not input files. - outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + // 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, "splice", outFile) } // Tee implements Linux syscall tee(2). @@ -251,11 +248,20 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo break } } - if n == 0 { - return 0, nil, err + + if n != 0 { + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) + + // If a partial write is completed, the error is dropped. Log it here. + if err != nil && err != io.EOF && err != syserror.ErrWouldBlock { + log.Debugf("tee completed a partial write with error: %v", err) + err = nil + } } - outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + + // 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, "tee", inFile) } // Sendfile implements linux system call sendfile(2). @@ -348,11 +354,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc for n < count { var spliceN int64 spliceN, err = outPipeFD.SpliceFromNonPipe(t, inFile, offset, count) - if spliceN == 0 && err == io.EOF { - // We reached the end of the file. Eat the error and exit the loop. - err = nil - break - } if offset != -1 { offset += spliceN } @@ -375,13 +376,6 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } else { readN, err = inFile.Read(t, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) } - if readN == 0 && err != nil { - if err == io.EOF { - // We reached the end of the file. Eat the error before exiting the loop. - err = nil - } - break - } n += readN // Write all of the bytes that we read. This may need @@ -432,13 +426,20 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } } - if n == 0 { - return 0, nil, err + if n != 0 { + inFile.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) + + if err != nil && err != io.EOF && err != syserror.ErrWouldBlock { + // If a partial write is completed, the error is dropped. Log it here. + log.Debugf("sendfile completed a partial write with error: %v", err) + err = nil + } } - inFile.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) - outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) - return uintptr(n), nil, nil + // 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) } // dualWaiter is used to wait on one or both vfs.FileDescriptions. It is not diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go index 0df3bd449..c50fd97eb 100644 --- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go +++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go @@ -163,6 +163,7 @@ func Override() { // Override ARM64. s = linux.ARM64 + s.Table[2] = syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}) s.Table[5] = syscalls.Supported("setxattr", SetXattr) s.Table[6] = syscalls.Supported("lsetxattr", Lsetxattr) s.Table[7] = syscalls.Supported("fsetxattr", Fsetxattr) @@ -200,6 +201,7 @@ func Override() { s.Table[44] = syscalls.Supported("fstatfs", Fstatfs) s.Table[45] = syscalls.Supported("truncate", Truncate) s.Table[46] = syscalls.Supported("ftruncate", Ftruncate) + s.Table[47] = syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil) s.Table[48] = syscalls.Supported("faccessat", Faccessat) s.Table[49] = syscalls.Supported("chdir", Chdir) s.Table[50] = syscalls.Supported("fchdir", Fchdir) @@ -221,12 +223,14 @@ func Override() { s.Table[68] = syscalls.Supported("pwrite64", Pwrite64) s.Table[69] = syscalls.Supported("preadv", Preadv) s.Table[70] = syscalls.Supported("pwritev", Pwritev) + s.Table[71] = syscalls.Supported("sendfile", Sendfile) s.Table[72] = syscalls.Supported("pselect", Pselect) s.Table[73] = syscalls.Supported("ppoll", Ppoll) s.Table[74] = syscalls.Supported("signalfd4", Signalfd4) s.Table[76] = syscalls.Supported("splice", Splice) s.Table[77] = syscalls.Supported("tee", Tee) s.Table[78] = syscalls.Supported("readlinkat", Readlinkat) + s.Table[79] = syscalls.Supported("newfstatat", Newfstatat) s.Table[80] = syscalls.Supported("fstat", Fstat) s.Table[81] = syscalls.Supported("sync", Sync) s.Table[82] = syscalls.Supported("fsync", Fsync) @@ -251,8 +255,10 @@ func Override() { s.Table[210] = syscalls.Supported("shutdown", Shutdown) s.Table[211] = syscalls.Supported("sendmsg", SendMsg) s.Table[212] = syscalls.Supported("recvmsg", RecvMsg) + s.Table[213] = syscalls.Supported("readahead", Readahead) s.Table[221] = syscalls.Supported("execve", Execve) s.Table[222] = syscalls.Supported("mmap", Mmap) + s.Table[223] = syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil) s.Table[242] = syscalls.Supported("accept4", Accept4) s.Table[243] = syscalls.Supported("recvmmsg", RecvMMsg) s.Table[267] = syscalls.Supported("syncfs", Syncfs) diff --git a/pkg/sentry/vfs/BUILD b/pkg/sentry/vfs/BUILD index 8093ca55c..c855608db 100644 --- a/pkg/sentry/vfs/BUILD +++ b/pkg/sentry/vfs/BUILD @@ -92,7 +92,6 @@ go_library( visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/context", "//pkg/fd", "//pkg/fdnotifier", diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 9c4db3047..bdfd3ca8f 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -52,6 +52,8 @@ const ( ) // anonFilesystemType implements FilesystemType. +// +// +stateify savable type anonFilesystemType struct{} // GetFilesystem implements FilesystemType.GetFilesystem. @@ -69,12 +71,15 @@ func (anonFilesystemType) Name() string { // // Since all Dentries in anonFilesystem are non-directories, all FilesystemImpl // methods that would require an anonDentry to be a directory return ENOTDIR. +// +// +stateify savable type anonFilesystem struct { vfsfs Filesystem devMinor uint32 } +// +stateify savable type anonDentry struct { vfsd Dentry diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index a69a5b2f1..320ab7ce1 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -89,6 +89,8 @@ func (d *Dentry) Impl() DentryImpl { // DentryImpl contains implementation details for a Dentry. Implementations of // DentryImpl should contain their associated Dentry by value as their first // field. +// +// +stateify savable type DentryImpl interface { // IncRef increments the Dentry's reference count. A Dentry with a non-zero // reference count must remain coherent with the state of the filesystem. diff --git a/pkg/sentry/vfs/device.go b/pkg/sentry/vfs/device.go index 1e9dffc8f..dde2ad79b 100644 --- a/pkg/sentry/vfs/device.go +++ b/pkg/sentry/vfs/device.go @@ -22,6 +22,8 @@ import ( ) // DeviceKind indicates whether a device is a block or character device. +// +// +stateify savable type DeviceKind uint32 const ( @@ -44,6 +46,7 @@ func (kind DeviceKind) String() string { } } +// +stateify savable type devTuple struct { kind DeviceKind major uint32 diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 754e76aec..8f36c3e3b 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -27,6 +27,8 @@ import ( var epollCycleMu sync.Mutex // EpollInstance represents an epoll instance, as described by epoll(7). +// +// +stateify savable type EpollInstance struct { vfsfd FileDescription FileDescriptionDefaultImpl @@ -38,11 +40,11 @@ type EpollInstance struct { // interest is the set of file descriptors that are registered with the // EpollInstance for monitoring. interest is protected by interestMu. - interestMu sync.Mutex + interestMu sync.Mutex `state:"nosave"` interest map[epollInterestKey]*epollInterest // mu protects fields in registered epollInterests. - mu sync.Mutex + mu sync.Mutex `state:"nosave"` // ready is the set of file descriptors that may be "ready" for I/O. Note // that this must be an ordered list, not a map: "If more than maxevents @@ -55,6 +57,7 @@ type EpollInstance struct { ready epollInterestList } +// +stateify savable type epollInterestKey struct { // file is the registered FileDescription. No reference is held on file; // instead, when the last reference is dropped, FileDescription.DecRef() @@ -67,6 +70,8 @@ type epollInterestKey struct { } // epollInterest represents an EpollInstance's interest in a file descriptor. +// +// +stateify savable type epollInterest struct { // epoll is the owning EpollInstance. epoll is immutable. epoll *EpollInstance diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 73bb36d3e..1eba0270f 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -37,11 +37,13 @@ import ( // FileDescription methods require that a reference is held. // // FileDescription is analogous to Linux's struct file. +// +// +stateify savable type FileDescription struct { FileDescriptionRefs // flagsMu protects statusFlags and asyncHandler below. - flagsMu sync.Mutex + flagsMu sync.Mutex `state:"nosave"` // statusFlags contains status flags, "initialized by open(2) and possibly // modified by fcntl()" - fcntl(2). statusFlags can be read using atomic @@ -56,7 +58,7 @@ type FileDescription struct { // epolls is the set of epollInterests registered for this FileDescription. // epolls is protected by epollMu. - epollMu sync.Mutex + epollMu sync.Mutex `state:"nosave"` epolls map[*epollInterest]struct{} // vd is the filesystem location at which this FileDescription was opened. @@ -88,6 +90,8 @@ type FileDescription struct { } // FileDescriptionOptions contains options to FileDescription.Init(). +// +// +stateify savable type FileDescriptionOptions struct { // If AllowDirectIO is true, allow O_DIRECT to be set on the file. AllowDirectIO bool @@ -451,6 +455,8 @@ type FileDescriptionImpl interface { } // Dirent holds the information contained in struct linux_dirent64. +// +// +stateify savable type Dirent struct { // Name is the filename. Name string diff --git a/pkg/sentry/vfs/file_description_impl_util.go b/pkg/sentry/vfs/file_description_impl_util.go index 78da16bac..48ca9de44 100644 --- a/pkg/sentry/vfs/file_description_impl_util.go +++ b/pkg/sentry/vfs/file_description_impl_util.go @@ -42,6 +42,8 @@ import ( // FileDescriptionDefaultImpl may be embedded by implementations of // FileDescriptionImpl to obtain implementations of many FileDescriptionImpl // methods with default behavior analogous to Linux's. +// +// +stateify savable type FileDescriptionDefaultImpl struct{} // OnClose implements FileDescriptionImpl.OnClose analogously to @@ -166,6 +168,8 @@ func (FileDescriptionDefaultImpl) RemoveXattr(ctx context.Context, name string) // DirectoryFileDescriptionDefaultImpl may be embedded by implementations of // FileDescriptionImpl that always represent directories to obtain // implementations of non-directory I/O methods that return EISDIR. +// +// +stateify savable type DirectoryFileDescriptionDefaultImpl struct{} // Allocate implements DirectoryFileDescriptionDefaultImpl.Allocate. @@ -196,6 +200,8 @@ func (DirectoryFileDescriptionDefaultImpl) Write(ctx context.Context, src userme // DentryMetadataFileDescriptionImpl may be embedded by implementations of // FileDescriptionImpl for which FileDescriptionOptions.UseDentryMetadata is // true to obtain implementations of Stat and SetStat that panic. +// +// +stateify savable type DentryMetadataFileDescriptionImpl struct{} // Stat implements FileDescriptionImpl.Stat. @@ -210,12 +216,16 @@ func (DentryMetadataFileDescriptionImpl) SetStat(ctx context.Context, opts SetSt // DynamicBytesSource represents a data source for a // DynamicBytesFileDescriptionImpl. +// +// +stateify savable type DynamicBytesSource interface { // Generate writes the file's contents to buf. Generate(ctx context.Context, buf *bytes.Buffer) error } // StaticData implements DynamicBytesSource over a static string. +// +// +stateify savable type StaticData struct { Data string } @@ -242,14 +252,24 @@ type WritableDynamicBytesSource interface { // // DynamicBytesFileDescriptionImpl.SetDataSource() must be called before first // use. +// +// +stateify savable type DynamicBytesFileDescriptionImpl struct { data DynamicBytesSource // immutable - mu sync.Mutex // protects the following fields - buf bytes.Buffer + mu sync.Mutex `state:"nosave"` // protects the following fields + buf bytes.Buffer `state:".([]byte)"` off int64 lastRead int64 // offset at which the last Read, PRead, or Seek ended } +func (fd *DynamicBytesFileDescriptionImpl) saveBuf() []byte { + return fd.buf.Bytes() +} + +func (fd *DynamicBytesFileDescriptionImpl) loadBuf(p []byte) { + fd.buf.Write(p) +} + // SetDataSource must be called exactly once on fd before first use. func (fd *DynamicBytesFileDescriptionImpl) SetDataSource(data DynamicBytesSource) { fd.data = data @@ -382,6 +402,8 @@ func GenericConfigureMMap(fd *FileDescription, m memmap.Mappable, opts *memmap.M // LockFD may be used by most implementations of FileDescriptionImpl.Lock* // functions. Caller must call Init(). +// +// +stateify savable type LockFD struct { locks *FileLocks } @@ -409,6 +431,8 @@ func (fd *LockFD) UnlockBSD(ctx context.Context, uid fslock.UniqueID) error { // NoLockFD implements Lock*/Unlock* portion of FileDescriptionImpl interface // returning ENOLCK. +// +// +stateify savable type NoLockFD struct{} // LockBSD implements vfs.FileDescriptionImpl.LockBSD. diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 7dae4e7e8..c93d94634 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -506,6 +506,8 @@ type FilesystemImpl interface { // PrependPathAtVFSRootError is returned by implementations of // FilesystemImpl.PrependPath() when they encounter the contextual VFS root. +// +// +stateify savable type PrependPathAtVFSRootError struct{} // Error implements error.Error. @@ -516,6 +518,8 @@ func (PrependPathAtVFSRootError) Error() string { // PrependPathAtNonMountRootError is returned by implementations of // FilesystemImpl.PrependPath() when they encounter an independent ancestor // Dentry that is not the Mount root. +// +// +stateify savable type PrependPathAtNonMountRootError struct{} // Error implements error.Error. @@ -526,6 +530,8 @@ func (PrependPathAtNonMountRootError) Error() string { // PrependPathSyntheticError is returned by implementations of // FilesystemImpl.PrependPath() for which prepended names do not represent real // paths. +// +// +stateify savable type PrependPathSyntheticError struct{} // Error implements error.Error. diff --git a/pkg/sentry/vfs/filesystem_type.go b/pkg/sentry/vfs/filesystem_type.go index f2298f7f6..bc19db1d5 100644 --- a/pkg/sentry/vfs/filesystem_type.go +++ b/pkg/sentry/vfs/filesystem_type.go @@ -55,10 +55,13 @@ type registeredFilesystemType struct { // RegisterFilesystemTypeOptions contains options to // VirtualFilesystem.RegisterFilesystem(). +// +// +stateify savable type RegisterFilesystemTypeOptions struct { - // If AllowUserMount is true, allow calls to VirtualFilesystem.MountAt() - // for which MountOptions.InternalMount == false to use this filesystem - // type. + // AllowUserMount determines whether users are allowed to mount a file system + // of this type, i.e. through mount(2). If AllowUserMount is true, allow calls + // to VirtualFilesystem.MountAt() for which MountOptions.InternalMount == false + // to use this filesystem type. AllowUserMount bool // If AllowUserList is true, make this filesystem type visible in diff --git a/pkg/sentry/vfs/genericfstree/genericfstree.go b/pkg/sentry/vfs/genericfstree/genericfstree.go index 8882fa84a..2d27d9d35 100644 --- a/pkg/sentry/vfs/genericfstree/genericfstree.go +++ b/pkg/sentry/vfs/genericfstree/genericfstree.go @@ -27,6 +27,8 @@ import ( ) // Dentry is a required type parameter that is a struct with the given fields. +// +// +stateify savable type Dentry struct { // vfsd is the embedded vfs.Dentry corresponding to this vfs.DentryImpl. vfsd vfs.Dentry diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index aff220a61..3f0b8f45b 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -37,6 +37,8 @@ const inotifyEventBaseSize = 16 // // The way events are labelled appears somewhat arbitrary, but they must match // Linux so that IN_EXCL_UNLINK behaves as it does in Linux. +// +// +stateify savable type EventType uint8 // PathEvent and InodeEvent correspond to FSNOTIFY_EVENT_PATH and diff --git a/pkg/sentry/vfs/lock.go b/pkg/sentry/vfs/lock.go index 42666eebf..55783d4eb 100644 --- a/pkg/sentry/vfs/lock.go +++ b/pkg/sentry/vfs/lock.go @@ -33,6 +33,8 @@ import ( // Note that in Linux these two types of locks are _not_ cooperative, because // race and deadlock conditions make merging them prohibitive. We do the same // and keep them oblivious to each other. +// +// +stateify savable type FileLocks struct { // bsd is a set of BSD-style advisory file wide locks, see flock(2). bsd fslock.Locks diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 9da09d4c1..dfc3ae6c0 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -65,7 +65,7 @@ type Mount struct { // // Invariant: key.parent != nil iff key.point != nil. key.point belongs to // key.parent.fs. - key mountKey + key mountKey `state:".(VirtualDentry)"` // ns is the namespace in which this Mount was mounted. ns is protected by // VirtualFilesystem.mountMu. @@ -345,6 +345,7 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti return nil } +// +stateify savable type umountRecursiveOptions struct { // If eager is true, ensure that future calls to Mount.tryIncMountedRef() // on umounted mounts fail. @@ -414,7 +415,7 @@ func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns } } mnt.IncRef() // dropped by callers of umountRecursiveLocked - mnt.storeKey(vd) + mnt.setKey(vd) if vd.mount.children == nil { vd.mount.children = make(map[*Mount]struct{}) } @@ -439,13 +440,13 @@ func (vfs *VirtualFilesystem) connectLocked(mnt *Mount, vd VirtualDentry, mntns // * vfs.mounts.seq must be in a writer critical section. // * mnt.parent() != nil. func (vfs *VirtualFilesystem) disconnectLocked(mnt *Mount) VirtualDentry { - vd := mnt.loadKey() + vd := mnt.getKey() if checkInvariants { if vd.mount != nil { panic("VFS.disconnectLocked called on disconnected mount") } } - mnt.storeKey(VirtualDentry{}) + mnt.loadKey(VirtualDentry{}) delete(vd.mount.children, mnt) atomic.AddUint32(&vd.dentry.mounts, math.MaxUint32) // -1 mnt.ns.mountpoints[vd.dentry]-- diff --git a/pkg/sentry/vfs/mount_test.go b/pkg/sentry/vfs/mount_test.go index 3335e4057..cb8c56bd3 100644 --- a/pkg/sentry/vfs/mount_test.go +++ b/pkg/sentry/vfs/mount_test.go @@ -38,7 +38,7 @@ func TestMountTableInsertLookup(t *testing.T) { mt.Init() mount := &Mount{} - mount.storeKey(VirtualDentry{&Mount{}, &Dentry{}}) + mount.setKey(VirtualDentry{&Mount{}, &Dentry{}}) mt.Insert(mount) if m := mt.Lookup(mount.parent(), mount.point()); m != mount { @@ -79,7 +79,7 @@ const enableComparativeBenchmarks = false func newBenchMount() *Mount { mount := &Mount{} - mount.storeKey(VirtualDentry{&Mount{}, &Dentry{}}) + mount.loadKey(VirtualDentry{&Mount{}, &Dentry{}}) return mount } @@ -94,7 +94,7 @@ func BenchmarkMountTableParallelLookup(b *testing.B) { for i := 0; i < numMounts; i++ { mount := newBenchMount() mt.Insert(mount) - keys = append(keys, mount.loadKey()) + keys = append(keys, mount.saveKey()) } var ready sync.WaitGroup @@ -146,7 +146,7 @@ func BenchmarkMountMapParallelLookup(b *testing.B) { keys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { mount := newBenchMount() - key := mount.loadKey() + key := mount.saveKey() ms[key] = mount keys = append(keys, key) } @@ -201,7 +201,7 @@ func BenchmarkMountSyncMapParallelLookup(b *testing.B) { keys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { mount := newBenchMount() - key := mount.loadKey() + key := mount.getKey() ms.Store(key, mount) keys = append(keys, key) } @@ -283,7 +283,7 @@ func BenchmarkMountMapNegativeLookup(b *testing.B) { ms := make(map[VirtualDentry]*Mount) for i := 0; i < numMounts; i++ { mount := newBenchMount() - ms[mount.loadKey()] = mount + ms[mount.getKey()] = mount } negkeys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { @@ -318,7 +318,7 @@ func BenchmarkMountSyncMapNegativeLookup(b *testing.B) { var ms sync.Map for i := 0; i < numMounts; i++ { mount := newBenchMount() - ms.Store(mount.loadKey(), mount) + ms.Store(mount.saveKey(), mount) } negkeys := make([]VirtualDentry, 0, numMounts) for i := 0; i < numMounts; i++ { @@ -372,7 +372,7 @@ func BenchmarkMountMapInsert(b *testing.B) { b.ResetTimer() for i := range mounts { mount := mounts[i] - ms[mount.loadKey()] = mount + ms[mount.saveKey()] = mount } } @@ -392,7 +392,7 @@ func BenchmarkMountSyncMapInsert(b *testing.B) { b.ResetTimer() for i := range mounts { mount := mounts[i] - ms.Store(mount.loadKey(), mount) + ms.Store(mount.saveKey(), mount) } } @@ -425,13 +425,13 @@ func BenchmarkMountMapRemove(b *testing.B) { ms := make(map[VirtualDentry]*Mount) for i := range mounts { mount := mounts[i] - ms[mount.loadKey()] = mount + ms[mount.saveKey()] = mount } b.ResetTimer() for i := range mounts { mount := mounts[i] - delete(ms, mount.loadKey()) + delete(ms, mount.saveKey()) } } @@ -447,12 +447,12 @@ func BenchmarkMountSyncMapRemove(b *testing.B) { var ms sync.Map for i := range mounts { mount := mounts[i] - ms.Store(mount.loadKey(), mount) + ms.Store(mount.saveKey(), mount) } b.ResetTimer() for i := range mounts { mount := mounts[i] - ms.Delete(mount.loadKey()) + ms.Delete(mount.saveKey()) } } diff --git a/pkg/sentry/vfs/mount_unsafe.go b/pkg/sentry/vfs/mount_unsafe.go index da2a2e9c4..b7d122d22 100644 --- a/pkg/sentry/vfs/mount_unsafe.go +++ b/pkg/sentry/vfs/mount_unsafe.go @@ -34,6 +34,8 @@ import ( // structurally identical to VirtualDentry, but stores its fields as // unsafe.Pointer since mutators synchronize with VFS path traversal using // seqcounts. +// +// This is explicitly not savable. type mountKey struct { parent unsafe.Pointer // *Mount point unsafe.Pointer // *Dentry @@ -47,19 +49,23 @@ func (mnt *Mount) point() *Dentry { return (*Dentry)(atomic.LoadPointer(&mnt.key.point)) } -func (mnt *Mount) loadKey() VirtualDentry { +func (mnt *Mount) getKey() VirtualDentry { return VirtualDentry{ mount: mnt.parent(), dentry: mnt.point(), } } +func (mnt *Mount) saveKey() VirtualDentry { return mnt.getKey() } + // Invariant: mnt.key.parent == nil. vd.Ok(). -func (mnt *Mount) storeKey(vd VirtualDentry) { +func (mnt *Mount) setKey(vd VirtualDentry) { atomic.StorePointer(&mnt.key.parent, unsafe.Pointer(vd.mount)) atomic.StorePointer(&mnt.key.point, unsafe.Pointer(vd.dentry)) } +func (mnt *Mount) loadKey(vd VirtualDentry) { mnt.setKey(vd) } + // mountTable maps (mount parent, mount point) pairs to mounts. It supports // efficient concurrent lookup, even in the presence of concurrent mutators // (provided mutation is sufficiently uncommon). @@ -92,6 +98,7 @@ type mountTable struct { // length and cap in separate uint32s) for ~free. size uint64 + // FIXME(gvisor.dev/issue/1663): Slots need to be saved. slots unsafe.Pointer `state:"nosave"` // []mountSlot; never nil after Init } diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index b33d36cb1..bc79e5ecc 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -21,6 +21,8 @@ import ( // GetDentryOptions contains options to VirtualFilesystem.GetDentryAt() and // FilesystemImpl.GetDentryAt(). +// +// +stateify savable type GetDentryOptions struct { // If CheckSearchable is true, FilesystemImpl.GetDentryAt() must check that // the returned Dentry is a directory for which creds has search @@ -30,6 +32,8 @@ type GetDentryOptions struct { // MkdirOptions contains options to VirtualFilesystem.MkdirAt() and // FilesystemImpl.MkdirAt(). +// +// +stateify savable type MkdirOptions struct { // Mode is the file mode bits for the created directory. Mode linux.FileMode @@ -56,6 +60,8 @@ type MkdirOptions struct { // MknodOptions contains options to VirtualFilesystem.MknodAt() and // FilesystemImpl.MknodAt(). +// +// +stateify savable type MknodOptions struct { // Mode is the file type and mode bits for the created file. Mode linux.FileMode @@ -72,6 +78,8 @@ type MknodOptions struct { // MountFlags contains flags as specified for mount(2), e.g. MS_NOEXEC. // MS_RDONLY is not part of MountFlags because it's tracked in Mount.writers. +// +// +stateify savable type MountFlags struct { // NoExec is equivalent to MS_NOEXEC. NoExec bool @@ -93,6 +101,8 @@ type MountFlags struct { } // MountOptions contains options to VirtualFilesystem.MountAt(). +// +// +stateify savable type MountOptions struct { // Flags contains flags as specified for mount(2), e.g. MS_NOEXEC. Flags MountFlags @@ -103,13 +113,17 @@ type MountOptions struct { // GetFilesystemOptions contains options to FilesystemType.GetFilesystem(). GetFilesystemOptions GetFilesystemOptions - // If InternalMount is true, allow the use of filesystem types for which - // RegisterFilesystemTypeOptions.AllowUserMount == false. + // InternalMount indicates whether the mount operation is coming from the + // application, i.e. through mount(2). If InternalMount is true, allow the use + // of filesystem types for which RegisterFilesystemTypeOptions.AllowUserMount + // == false. InternalMount bool } // OpenOptions contains options to VirtualFilesystem.OpenAt() and // FilesystemImpl.OpenAt(). +// +// +stateify savable type OpenOptions struct { // Flags contains access mode and flags as specified for open(2). // @@ -135,6 +149,8 @@ type OpenOptions struct { // ReadOptions contains options to FileDescription.PRead(), // FileDescriptionImpl.PRead(), FileDescription.Read(), and // FileDescriptionImpl.Read(). +// +// +stateify savable type ReadOptions struct { // Flags contains flags as specified for preadv2(2). Flags uint32 @@ -142,6 +158,8 @@ type ReadOptions struct { // RenameOptions contains options to VirtualFilesystem.RenameAt() and // FilesystemImpl.RenameAt(). +// +// +stateify savable type RenameOptions struct { // Flags contains flags as specified for renameat2(2). Flags uint32 @@ -153,6 +171,8 @@ type RenameOptions struct { // SetStatOptions contains options to VirtualFilesystem.SetStatAt(), // FilesystemImpl.SetStatAt(), FileDescription.SetStat(), and // FileDescriptionImpl.SetStat(). +// +// +stateify savable type SetStatOptions struct { // Stat is the metadata that should be set. Only fields indicated by // Stat.Mask should be set. @@ -174,6 +194,8 @@ type SetStatOptions struct { // BoundEndpointOptions contains options to VirtualFilesystem.BoundEndpointAt() // and FilesystemImpl.BoundEndpointAt(). +// +// +stateify savable type BoundEndpointOptions struct { // Addr is the path of the file whose socket endpoint is being retrieved. // It is generally irrelevant: most endpoints are stored at a dentry that @@ -193,6 +215,8 @@ type BoundEndpointOptions struct { // GetXattrOptions contains options to VirtualFilesystem.GetXattrAt(), // FilesystemImpl.GetXattrAt(), FileDescription.GetXattr(), and // FileDescriptionImpl.GetXattr(). +// +// +stateify savable type GetXattrOptions struct { // Name is the name of the extended attribute to retrieve. Name string @@ -207,6 +231,8 @@ type GetXattrOptions struct { // SetXattrOptions contains options to VirtualFilesystem.SetXattrAt(), // FilesystemImpl.SetXattrAt(), FileDescription.SetXattr(), and // FileDescriptionImpl.SetXattr(). +// +// +stateify savable type SetXattrOptions struct { // Name is the name of the extended attribute being mutated. Name string @@ -221,6 +247,8 @@ type SetXattrOptions struct { // StatOptions contains options to VirtualFilesystem.StatAt(), // FilesystemImpl.StatAt(), FileDescription.Stat(), and // FileDescriptionImpl.Stat(). +// +// +stateify savable type StatOptions struct { // Mask is the set of fields in the returned Statx that the FilesystemImpl // or FileDescriptionImpl should provide. Bits are as in linux.Statx.Mask. @@ -238,6 +266,8 @@ type StatOptions struct { } // UmountOptions contains options to VirtualFilesystem.UmountAt(). +// +// +stateify savable type UmountOptions struct { // Flags contains flags as specified for umount2(2). Flags uint32 @@ -246,6 +276,8 @@ type UmountOptions struct { // WriteOptions contains options to FileDescription.PWrite(), // FileDescriptionImpl.PWrite(), FileDescription.Write(), and // FileDescriptionImpl.Write(). +// +// +stateify savable type WriteOptions struct { // Flags contains flags as specified for pwritev2(2). Flags uint32 diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index 00eeb8842..d48520d58 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -26,6 +26,8 @@ import ( ) // AccessTypes is a bitmask of Unix file permissions. +// +// +stateify savable type AccessTypes uint16 // Bits in AccessTypes. diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 3304372d9..e4fd55012 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -35,6 +35,8 @@ import ( // FilesystemImpl methods. // // ResolvingPath is loosely analogous to Linux's struct nameidata. +// +// +stateify savable type ResolvingPath struct { vfs *VirtualFilesystem root VirtualDentry // refs borrowed from PathOperation @@ -88,6 +90,7 @@ func init() { // so error "constants" are really mutable vars, necessitating somewhat // expensive interface object comparisons. +// +stateify savable type resolveMountRootOrJumpError struct{} // Error implements error.Error. @@ -95,6 +98,7 @@ func (resolveMountRootOrJumpError) Error() string { return "resolving mount root or jump" } +// +stateify savable type resolveMountPointError struct{} // Error implements error.Error. @@ -102,6 +106,7 @@ func (resolveMountPointError) Error() string { return "resolving mount point" } +// +stateify savable type resolveAbsSymlinkError struct{} // Error implements error.Error. diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 1ebf355ef..5bd756ea5 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -163,6 +163,8 @@ func (vfs *VirtualFilesystem) Init(ctx context.Context) error { // PathOperation is passed to VFS methods by pointer to reduce memory copying: // it's somewhat large and should never escape. (Options structs are passed by // pointer to VFS and FileDescription methods for the same reason.) +// +// +stateify savable type PathOperation struct { // Root is the VFS root. References on Root are borrowed from the provider // of the PathOperation. diff --git a/pkg/state/types.go b/pkg/state/types.go index 215ef80f8..84aed8732 100644 --- a/pkg/state/types.go +++ b/pkg/state/types.go @@ -107,6 +107,14 @@ func lookupNameFields(typ reflect.Type) (string, []string, bool) { } return name, nil, true } + // Sanity check the type. + if raceEnabled { + if _, ok := reverseTypeDatabase[typ]; !ok { + // The type was not registered? Must be an embedded + // structure or something else. + return "", nil, false + } + } // Extract the name from the object. name := t.StateTypeName() fields := t.StateFields() @@ -313,6 +321,9 @@ var primitiveTypeDatabase = func() map[string]reflect.Type { // globalTypeDatabase is used for dispatching interfaces on decode. var globalTypeDatabase = map[string]reflect.Type{} +// reverseTypeDatabase is a reverse mapping. +var reverseTypeDatabase = map[reflect.Type]string{} + // Register registers a type. // // This must be called on init and only done once. @@ -358,4 +369,7 @@ func Register(t Type) { Failf("conflicting name for %T: matches interfaceType", t) } globalTypeDatabase[name] = typ + if raceEnabled { + reverseTypeDatabase[typ] = name + } } diff --git a/pkg/tcpip/adapters/gonet/gonet_test.go b/pkg/tcpip/adapters/gonet/gonet_test.go index c975ad9cf..12b061def 100644 --- a/pkg/tcpip/adapters/gonet/gonet_test.go +++ b/pkg/tcpip/adapters/gonet/gonet_test.go @@ -61,8 +61,8 @@ func TestTimeouts(t *testing.T) { func newLoopbackStack() (*stack.Stack, *tcpip.Error) { // Create the stack and add a NIC. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol, udp.NewProtocol}, }) if err := s.CreateNIC(NICID, loopback.New()); err != nil { diff --git a/pkg/tcpip/buffer/view.go b/pkg/tcpip/buffer/view.go index ea0c5413d..8db70a700 100644 --- a/pkg/tcpip/buffer/view.go +++ b/pkg/tcpip/buffer/view.go @@ -84,8 +84,8 @@ type VectorisedView struct { size int } -// NewVectorisedView creates a new vectorised view from an already-allocated slice -// of View and sets its size. +// NewVectorisedView creates a new vectorised view from an already-allocated +// slice of View and sets its size. func NewVectorisedView(size int, views []View) VectorisedView { return VectorisedView{views: views, size: size} } @@ -170,8 +170,9 @@ func (vv *VectorisedView) CapLength(length int) { } // Clone returns a clone of this VectorisedView. -// If the buffer argument is large enough to contain all the Views of this VectorisedView, -// the method will avoid allocations and use the buffer to store the Views of the clone. +// If the buffer argument is large enough to contain all the Views of this +// VectorisedView, the method will avoid allocations and use the buffer to +// store the Views of the clone. func (vv *VectorisedView) Clone(buffer []View) VectorisedView { return VectorisedView{views: append(buffer[:0], vv.views...), size: vv.size} } @@ -209,7 +210,8 @@ func (vv *VectorisedView) PullUp(count int) (View, bool) { return newFirst, true } -// Size returns the size in bytes of the entire content stored in the vectorised view. +// Size returns the size in bytes of the entire content stored in the +// vectorised view. func (vv *VectorisedView) Size() int { return vv.size } @@ -222,6 +224,12 @@ func (vv *VectorisedView) ToView() View { if len(vv.views) == 1 { return vv.views[0] } + return vv.ToOwnedView() +} + +// ToOwnedView returns a single view containing the content of the vectorised +// view that vv does not own. +func (vv *VectorisedView) ToOwnedView() View { u := make([]byte, 0, vv.size) for _, v := range vv.views { u = append(u, v...) diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index b769094dc..d4d785cca 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -118,18 +118,82 @@ func TTL(ttl uint8) NetworkChecker { v = ip.HopLimit() } if v != ttl { - t.Fatalf("Bad TTL, got %v, want %v", v, ttl) + t.Fatalf("Bad TTL, got = %d, want = %d", v, ttl) + } + } +} + +// IPFullLength creates a checker for the full IP packet length. The +// expected size is checked against both the Total Length in the +// header and the number of bytes received. +func IPFullLength(packetLength uint16) NetworkChecker { + return func(t *testing.T, h []header.Network) { + t.Helper() + + var v uint16 + var l uint16 + switch ip := h[0].(type) { + case header.IPv4: + v = ip.TotalLength() + l = uint16(len(ip)) + case header.IPv6: + v = ip.PayloadLength() + header.IPv6FixedHeaderSize + l = uint16(len(ip)) + default: + t.Fatalf("unexpected network header passed to checker, got = %T, want = header.IPv4 or header.IPv6", ip) + } + if l != packetLength { + t.Errorf("bad packet length, got = %d, want = %d", l, packetLength) + } + if v != packetLength { + t.Errorf("unexpected packet length in header, got = %d, want = %d", v, packetLength) + } + } +} + +// IPv4HeaderLength creates a checker that checks the IPv4 Header length. +func IPv4HeaderLength(headerLength int) NetworkChecker { + return func(t *testing.T, h []header.Network) { + t.Helper() + + switch ip := h[0].(type) { + case header.IPv4: + if hl := ip.HeaderLength(); hl != uint8(headerLength) { + t.Errorf("Bad header length, got = %d, want = %d", hl, headerLength) + } + default: + t.Fatalf("unexpected network header passed to checker, got = %T, want = header.IPv4", ip) } } } // PayloadLen creates a checker that checks the payload length. -func PayloadLen(plen int) NetworkChecker { +func PayloadLen(payloadLength int) NetworkChecker { + return func(t *testing.T, h []header.Network) { + t.Helper() + + if l := len(h[0].Payload()); l != payloadLength { + t.Errorf("Bad payload length, got = %d, want = %d", l, payloadLength) + } + } +} + +// IPv4Options returns a checker that checks the options in an IPv4 packet. +func IPv4Options(want []byte) NetworkChecker { return func(t *testing.T, h []header.Network) { t.Helper() - if l := len(h[0].Payload()); l != plen { - t.Errorf("Bad payload length, got %v, want %v", l, plen) + ip, ok := h[0].(header.IPv4) + if !ok { + t.Fatalf("unexpected network header passed to checker, got = %T, want = header.IPv4", h[0]) + } + options := ip.Options() + // cmp.Diff does not consider nil slices equal to empty slices, but we do. + if len(want) == 0 && len(options) == 0 { + return + } + if diff := cmp.Diff(want, options); diff != "" { + t.Errorf("options mismatch (-want +got):\n%s", diff) } } } @@ -139,11 +203,11 @@ func FragmentOffset(offset uint16) NetworkChecker { return func(t *testing.T, h []header.Network) { t.Helper() - // We only do this of IPv4 for now. + // We only do this for IPv4 for now. switch ip := h[0].(type) { case header.IPv4: if v := ip.FragmentOffset(); v != offset { - t.Errorf("Bad fragment offset, got %v, want %v", v, offset) + t.Errorf("Bad fragment offset, got = %d, want = %d", v, offset) } } } @@ -154,11 +218,11 @@ func FragmentFlags(flags uint8) NetworkChecker { return func(t *testing.T, h []header.Network) { t.Helper() - // We only do this of IPv4 for now. + // We only do this for IPv4 for now. switch ip := h[0].(type) { case header.IPv4: if v := ip.Flags(); v != flags { - t.Errorf("Bad fragment offset, got %v, want %v", v, flags) + t.Errorf("Bad fragment offset, got = %d, want = %d", v, flags) } } } @@ -208,7 +272,7 @@ func TOS(tos uint8, label uint32) NetworkChecker { t.Helper() if v, l := h[0].TOS(); v != tos || l != label { - t.Errorf("Bad TOS, got (%v, %v), want (%v,%v)", v, l, tos, label) + t.Errorf("Bad TOS, got = (%d, %d), want = (%d,%d)", v, l, tos, label) } } } @@ -234,7 +298,7 @@ func IPv6Fragment(checkers ...NetworkChecker) NetworkChecker { t.Helper() if p := h[0].TransportProtocol(); p != header.IPv6FragmentHeader { - t.Errorf("Bad protocol, got %v, want %v", p, header.UDPProtocolNumber) + t.Errorf("Bad protocol, got = %d, want = %d", p, header.UDPProtocolNumber) } ipv6Frag := header.IPv6Fragment(h[0].Payload()) @@ -261,7 +325,7 @@ func TCP(checkers ...TransportChecker) NetworkChecker { last := h[len(h)-1] if p := last.TransportProtocol(); p != header.TCPProtocolNumber { - t.Errorf("Bad protocol, got %v, want %v", p, header.TCPProtocolNumber) + t.Errorf("Bad protocol, got = %d, want = %d", p, header.TCPProtocolNumber) } // Verify the checksum. @@ -297,7 +361,7 @@ func UDP(checkers ...TransportChecker) NetworkChecker { last := h[len(h)-1] if p := last.TransportProtocol(); p != header.UDPProtocolNumber { - t.Errorf("Bad protocol, got %v, want %v", p, header.UDPProtocolNumber) + t.Errorf("Bad protocol, got = %d, want = %d", p, header.UDPProtocolNumber) } udp := header.UDP(last.Payload()) @@ -316,7 +380,7 @@ func SrcPort(port uint16) TransportChecker { t.Helper() if p := h.SourcePort(); p != port { - t.Errorf("Bad source port, got %v, want %v", p, port) + t.Errorf("Bad source port, got = %d, want = %d", p, port) } } } @@ -327,7 +391,7 @@ func DstPort(port uint16) TransportChecker { t.Helper() if p := h.DestinationPort(); p != port { - t.Errorf("Bad destination port, got %v, want %v", p, port) + t.Errorf("Bad destination port, got = %d, want = %d", p, port) } } } @@ -339,7 +403,7 @@ func NoChecksum(noChecksum bool) TransportChecker { udp, ok := h.(header.UDP) if !ok { - return + t.Fatalf("UDP header not found in h: %T", h) } if b := udp.Checksum() == 0; b != noChecksum { @@ -348,50 +412,84 @@ func NoChecksum(noChecksum bool) TransportChecker { } } -// SeqNum creates a checker that checks the sequence number. -func SeqNum(seq uint32) TransportChecker { +// TCPSeqNum creates a checker that checks the sequence number. +func TCPSeqNum(seq uint32) TransportChecker { return func(t *testing.T, h header.Transport) { t.Helper() tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in h: %T", h) } if s := tcp.SequenceNumber(); s != seq { - t.Errorf("Bad sequence number, got %v, want %v", s, seq) + t.Errorf("Bad sequence number, got = %d, want = %d", s, seq) } } } -// AckNum creates a checker that checks the ack number. -func AckNum(seq uint32) TransportChecker { +// TCPAckNum creates a checker that checks the ack number. +func TCPAckNum(seq uint32) TransportChecker { return func(t *testing.T, h header.Transport) { t.Helper() tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in h: %T", h) } if s := tcp.AckNumber(); s != seq { - t.Errorf("Bad ack number, got %v, want %v", s, seq) + t.Errorf("Bad ack number, got = %d, want = %d", s, seq) } } } -// Window creates a checker that checks the tcp window. -func Window(window uint16) TransportChecker { +// TCPWindow creates a checker that checks the tcp window. +func TCPWindow(window uint16) TransportChecker { return func(t *testing.T, h header.Transport) { t.Helper() tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in hdr : %T", h) } if w := tcp.WindowSize(); w != window { - t.Errorf("Bad window, got 0x%x, want 0x%x", w, window) + t.Errorf("Bad window, got %d, want %d", w, window) + } + } +} + +// TCPWindowGreaterThanEq creates a checker that checks that the TCP window +// is greater than or equal to the provided value. +func TCPWindowGreaterThanEq(window uint16) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + tcp, ok := h.(header.TCP) + if !ok { + t.Fatalf("TCP header not found in h: %T", h) + } + + if w := tcp.WindowSize(); w < window { + t.Errorf("Bad window, got %d, want > %d", w, window) + } + } +} + +// TCPWindowLessThanEq creates a checker that checks that the tcp window +// is less than or equal to the provided value. +func TCPWindowLessThanEq(window uint16) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + tcp, ok := h.(header.TCP) + if !ok { + t.Fatalf("TCP header not found in h: %T", h) + } + + if w := tcp.WindowSize(); w > window { + t.Errorf("Bad window, got %d, want < %d", w, window) } } } @@ -403,7 +501,7 @@ func TCPFlags(flags uint8) TransportChecker { tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in h: %T", h) } if f := tcp.Flags(); f != flags { @@ -420,7 +518,7 @@ func TCPFlagsMatch(flags, mask uint8) TransportChecker { tcp, ok := h.(header.TCP) if !ok { - return + t.Fatalf("TCP header not found in h: %T", h) } if f := tcp.Flags(); (f & mask) != (flags & mask) { @@ -458,7 +556,7 @@ func TCPSynOptions(wantOpts header.TCPSynOptions) TransportChecker { case header.TCPOptionMSS: v := uint16(opts[i+2])<<8 | uint16(opts[i+3]) if wantOpts.MSS != v { - t.Errorf("Bad MSS: got %v, want %v", v, wantOpts.MSS) + t.Errorf("Bad MSS, got = %d, want = %d", v, wantOpts.MSS) } foundMSS = true i += 4 @@ -468,7 +566,7 @@ func TCPSynOptions(wantOpts header.TCPSynOptions) TransportChecker { } v := int(opts[i+2]) if v != wantOpts.WS { - t.Errorf("Bad WS: got %v, want %v", v, wantOpts.WS) + t.Errorf("Bad WS, got = %d, want = %d", v, wantOpts.WS) } foundWS = true i += 3 @@ -517,7 +615,7 @@ func TCPSynOptions(wantOpts header.TCPSynOptions) TransportChecker { t.Error("TS option specified but the timestamp value is zero") } if foundTS && tsEcr == 0 && wantOpts.TSEcr != 0 { - t.Errorf("TS option specified but TSEcr is incorrect: got %d, want: %d", tsEcr, wantOpts.TSEcr) + t.Errorf("TS option specified but TSEcr is incorrect, got = %d, want = %d", tsEcr, wantOpts.TSEcr) } if wantOpts.SACKPermitted && !foundSACKPermitted { t.Errorf("SACKPermitted option not found. Options: %x", opts) @@ -555,7 +653,7 @@ func TCPTimestampChecker(wantTS bool, wantTSVal uint32, wantTSEcr uint32) Transp t.Errorf("TS option found, but option is truncated, option length: %d, want 10 bytes", limit-i) } if opts[i+1] != 10 { - t.Errorf("TS option found, but bad length specified: %d, want: 10", opts[i+1]) + t.Errorf("TS option found, but bad length specified: got = %d, want = 10", opts[i+1]) } tsVal = binary.BigEndian.Uint32(opts[i+2:]) tsEcr = binary.BigEndian.Uint32(opts[i+6:]) @@ -575,19 +673,19 @@ func TCPTimestampChecker(wantTS bool, wantTSVal uint32, wantTSEcr uint32) Transp } if wantTS != foundTS { - t.Errorf("TS Option mismatch: got TS= %v, want TS= %v", foundTS, wantTS) + t.Errorf("TS Option mismatch, got TS= %t, want TS= %t", foundTS, wantTS) } if wantTS && wantTSVal != 0 && wantTSVal != tsVal { - t.Errorf("Timestamp value is incorrect: got: %d, want: %d", tsVal, wantTSVal) + t.Errorf("Timestamp value is incorrect, got = %d, want = %d", tsVal, wantTSVal) } if wantTS && wantTSEcr != 0 && tsEcr != wantTSEcr { - t.Errorf("Timestamp Echo Reply is incorrect: got: %d, want: %d", tsEcr, wantTSEcr) + t.Errorf("Timestamp Echo Reply is incorrect, got = %d, want = %d", tsEcr, wantTSEcr) } } } -// TCPNoSACKBlockChecker creates a checker that verifies that the segment does not -// contain any SACK blocks in the TCP options. +// TCPNoSACKBlockChecker creates a checker that verifies that the segment does +// not contain any SACK blocks in the TCP options. func TCPNoSACKBlockChecker() TransportChecker { return TCPSACKBlockChecker(nil) } @@ -645,7 +743,7 @@ func TCPSACKBlockChecker(sackBlocks []header.SACKBlock) TransportChecker { } if !reflect.DeepEqual(gotSACKBlocks, sackBlocks) { - t.Errorf("SACKBlocks are not equal, got: %v, want: %v", gotSACKBlocks, sackBlocks) + t.Errorf("SACKBlocks are not equal, got = %v, want = %v", gotSACKBlocks, sackBlocks) } } } @@ -661,8 +759,8 @@ func Payload(want []byte) TransportChecker { } } -// ICMPv4 creates a checker that checks that the transport protocol is ICMPv4 and -// potentially additional ICMPv4 header fields. +// ICMPv4 creates a checker that checks that the transport protocol is ICMPv4 +// and potentially additional ICMPv4 header fields. func ICMPv4(checkers ...TransportChecker) NetworkChecker { return func(t *testing.T, h []header.Network) { t.Helper() @@ -690,10 +788,10 @@ func ICMPv4Type(want header.ICMPv4Type) TransportChecker { icmpv4, ok := h.(header.ICMPv4) if !ok { - t.Fatalf("unexpected transport header passed to checker got: %+v, want: header.ICMPv4", h) + t.Fatalf("unexpected transport header passed to checker, got = %T, want = header.ICMPv4", h) } if got := icmpv4.Type(); got != want { - t.Fatalf("unexpected icmp type got: %d, want: %d", got, want) + t.Fatalf("unexpected icmp type, got = %d, want = %d", got, want) } } } @@ -705,10 +803,76 @@ func ICMPv4Code(want header.ICMPv4Code) TransportChecker { icmpv4, ok := h.(header.ICMPv4) if !ok { - t.Fatalf("unexpected transport header passed to checker got: %+v, want: header.ICMPv4", h) + t.Fatalf("unexpected transport header passed to checker, got = %T, want = header.ICMPv4", h) } if got := icmpv4.Code(); got != want { - t.Fatalf("unexpected ICMP code got: %d, want: %d", got, want) + t.Fatalf("unexpected ICMP code, got = %d, want = %d", got, want) + } + } +} + +// ICMPv4Ident creates a checker that checks the ICMPv4 echo Ident. +func ICMPv4Ident(want uint16) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmpv4, ok := h.(header.ICMPv4) + if !ok { + t.Fatalf("unexpected transport header passed to checker, got = %T, want = header.ICMPv4", h) + } + if got := icmpv4.Ident(); got != want { + t.Fatalf("unexpected ICMP ident, got = %d, want = %d", got, want) + } + } +} + +// ICMPv4Seq creates a checker that checks the ICMPv4 echo Sequence. +func ICMPv4Seq(want uint16) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmpv4, ok := h.(header.ICMPv4) + if !ok { + t.Fatalf("unexpected transport header passed to checker, got = %T, want = header.ICMPv4", h) + } + if got := icmpv4.Sequence(); got != want { + t.Fatalf("unexpected ICMP sequence, got = %d, want = %d", got, want) + } + } +} + +// ICMPv4Checksum creates a checker that checks the ICMPv4 Checksum. +// This assumes that the payload exactly makes up the rest of the slice. +func ICMPv4Checksum() TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmpv4, ok := h.(header.ICMPv4) + if !ok { + t.Fatalf("unexpected transport header passed to checker, got = %T, want = header.ICMPv4", h) + } + heldChecksum := icmpv4.Checksum() + icmpv4.SetChecksum(0) + newChecksum := ^header.Checksum(icmpv4, 0) + icmpv4.SetChecksum(heldChecksum) + if heldChecksum != newChecksum { + t.Errorf("unexpected ICMP checksum, got = %d, want = %d", heldChecksum, newChecksum) + } + } +} + +// ICMPv4Payload creates a checker that checks the payload in an ICMPv4 packet. +func ICMPv4Payload(want []byte) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmpv4, ok := h.(header.ICMPv4) + if !ok { + t.Fatalf("unexpected transport header passed to checker, got = %T, want = header.ICMPv4", h) + } + payload := icmpv4.Payload() + if diff := cmp.Diff(want, payload); diff != "" { + t.Errorf("ICMP payload mismatch (-want +got):\n%s", diff) } } } @@ -748,10 +912,10 @@ func ICMPv6Type(want header.ICMPv6Type) TransportChecker { icmpv6, ok := h.(header.ICMPv6) if !ok { - t.Fatalf("unexpected transport header passed to checker got: %+v, want: header.ICMPv6", h) + t.Fatalf("unexpected transport header passed to checker, got = %T, want = header.ICMPv6", h) } if got := icmpv6.Type(); got != want { - t.Fatalf("unexpected icmp type got: %d, want: %d", got, want) + t.Fatalf("unexpected icmp type, got = %d, want = %d", got, want) } } } @@ -763,10 +927,10 @@ func ICMPv6Code(want header.ICMPv6Code) TransportChecker { icmpv6, ok := h.(header.ICMPv6) if !ok { - t.Fatalf("unexpected transport header passed to checker got: %+v, want: header.ICMPv6", h) + t.Fatalf("unexpected transport header passed to checker, got = %T, want = header.ICMPv6", h) } if got := icmpv6.Code(); got != want { - t.Fatalf("unexpected ICMP code got: %d, want: %d", got, want) + t.Fatalf("unexpected ICMP code, got = %d, want = %d", got, want) } } } diff --git a/pkg/tcpip/faketime/BUILD b/pkg/tcpip/faketime/BUILD new file mode 100644 index 000000000..114d43df3 --- /dev/null +++ b/pkg/tcpip/faketime/BUILD @@ -0,0 +1,24 @@ +load("//tools:defs.bzl", "go_library", "go_test") + +package(licenses = ["notice"]) + +go_library( + name = "faketime", + srcs = ["faketime.go"], + visibility = ["//visibility:public"], + deps = [ + "//pkg/tcpip", + "@com_github_dpjacques_clockwork//:go_default_library", + ], +) + +go_test( + name = "faketime_test", + size = "small", + srcs = [ + "faketime_test.go", + ], + deps = [ + "//pkg/tcpip/faketime", + ], +) diff --git a/pkg/tcpip/stack/fake_time_test.go b/pkg/tcpip/faketime/faketime.go index 92c8cb534..f7a4fbde1 100644 --- a/pkg/tcpip/stack/fake_time_test.go +++ b/pkg/tcpip/faketime/faketime.go @@ -12,7 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. -package stack +// Package faketime provides a fake clock that implements tcpip.Clock interface. +package faketime import ( "container/heap" @@ -23,7 +24,29 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" ) -type fakeClock struct { +// NullClock implements a clock that never advances. +type NullClock struct{} + +var _ tcpip.Clock = (*NullClock)(nil) + +// NowNanoseconds implements tcpip.Clock.NowNanoseconds. +func (*NullClock) NowNanoseconds() int64 { + return 0 +} + +// NowMonotonic implements tcpip.Clock.NowMonotonic. +func (*NullClock) NowMonotonic() int64 { + return 0 +} + +// AfterFunc implements tcpip.Clock.AfterFunc. +func (*NullClock) AfterFunc(time.Duration, func()) tcpip.Timer { + return nil +} + +// ManualClock implements tcpip.Clock and only advances manually with Advance +// method. +type ManualClock struct { clock clockwork.FakeClock // mu protects the fields below. @@ -39,34 +62,35 @@ type fakeClock struct { waitGroups map[time.Time]*sync.WaitGroup } -func newFakeClock() *fakeClock { - return &fakeClock{ +// NewManualClock creates a new ManualClock instance. +func NewManualClock() *ManualClock { + return &ManualClock{ clock: clockwork.NewFakeClock(), times: &timeHeap{}, waitGroups: make(map[time.Time]*sync.WaitGroup), } } -var _ tcpip.Clock = (*fakeClock)(nil) +var _ tcpip.Clock = (*ManualClock)(nil) // NowNanoseconds implements tcpip.Clock.NowNanoseconds. -func (fc *fakeClock) NowNanoseconds() int64 { - return fc.clock.Now().UnixNano() +func (mc *ManualClock) NowNanoseconds() int64 { + return mc.clock.Now().UnixNano() } // NowMonotonic implements tcpip.Clock.NowMonotonic. -func (fc *fakeClock) NowMonotonic() int64 { - return fc.NowNanoseconds() +func (mc *ManualClock) NowMonotonic() int64 { + return mc.NowNanoseconds() } // AfterFunc implements tcpip.Clock.AfterFunc. -func (fc *fakeClock) AfterFunc(d time.Duration, f func()) tcpip.Timer { - until := fc.clock.Now().Add(d) - wg := fc.addWait(until) - return &fakeTimer{ - clock: fc, +func (mc *ManualClock) AfterFunc(d time.Duration, f func()) tcpip.Timer { + until := mc.clock.Now().Add(d) + wg := mc.addWait(until) + return &manualTimer{ + clock: mc, until: until, - timer: fc.clock.AfterFunc(d, func() { + timer: mc.clock.AfterFunc(d, func() { defer wg.Done() f() }), @@ -75,110 +99,113 @@ func (fc *fakeClock) AfterFunc(d time.Duration, f func()) tcpip.Timer { // addWait adds an additional wait to the WaitGroup for parallel execution of // all work scheduled for t. Returns a reference to the WaitGroup modified. -func (fc *fakeClock) addWait(t time.Time) *sync.WaitGroup { - fc.mu.RLock() - wg, ok := fc.waitGroups[t] - fc.mu.RUnlock() +func (mc *ManualClock) addWait(t time.Time) *sync.WaitGroup { + mc.mu.RLock() + wg, ok := mc.waitGroups[t] + mc.mu.RUnlock() if ok { wg.Add(1) return wg } - fc.mu.Lock() - heap.Push(fc.times, t) - fc.mu.Unlock() + mc.mu.Lock() + heap.Push(mc.times, t) + mc.mu.Unlock() wg = &sync.WaitGroup{} wg.Add(1) - fc.mu.Lock() - fc.waitGroups[t] = wg - fc.mu.Unlock() + mc.mu.Lock() + mc.waitGroups[t] = wg + mc.mu.Unlock() return wg } // removeWait removes a wait from the WaitGroup for parallel execution of all // work scheduled for t. -func (fc *fakeClock) removeWait(t time.Time) { - fc.mu.RLock() - defer fc.mu.RUnlock() +func (mc *ManualClock) removeWait(t time.Time) { + mc.mu.RLock() + defer mc.mu.RUnlock() - wg := fc.waitGroups[t] + wg := mc.waitGroups[t] wg.Done() } -// advance executes all work that have been scheduled to execute within d from -// the current fake time. Blocks until all work has completed execution. -func (fc *fakeClock) advance(d time.Duration) { +// Advance executes all work that have been scheduled to execute within d from +// the current time. Blocks until all work has completed execution. +func (mc *ManualClock) Advance(d time.Duration) { // Block until all the work is done - until := fc.clock.Now().Add(d) + until := mc.clock.Now().Add(d) for { - fc.mu.Lock() - if fc.times.Len() == 0 { - fc.mu.Unlock() - return + mc.mu.Lock() + if mc.times.Len() == 0 { + mc.mu.Unlock() + break } - t := heap.Pop(fc.times).(time.Time) + t := heap.Pop(mc.times).(time.Time) if t.After(until) { // No work to do - heap.Push(fc.times, t) - fc.mu.Unlock() - return + heap.Push(mc.times, t) + mc.mu.Unlock() + break } - fc.mu.Unlock() + mc.mu.Unlock() - diff := t.Sub(fc.clock.Now()) - fc.clock.Advance(diff) + diff := t.Sub(mc.clock.Now()) + mc.clock.Advance(diff) - fc.mu.RLock() - wg := fc.waitGroups[t] - fc.mu.RUnlock() + mc.mu.RLock() + wg := mc.waitGroups[t] + mc.mu.RUnlock() wg.Wait() - fc.mu.Lock() - delete(fc.waitGroups, t) - fc.mu.Unlock() + mc.mu.Lock() + delete(mc.waitGroups, t) + mc.mu.Unlock() + } + if now := mc.clock.Now(); until.After(now) { + mc.clock.Advance(until.Sub(now)) } } -type fakeTimer struct { - clock *fakeClock +type manualTimer struct { + clock *ManualClock timer clockwork.Timer mu sync.RWMutex until time.Time } -var _ tcpip.Timer = (*fakeTimer)(nil) +var _ tcpip.Timer = (*manualTimer)(nil) // Reset implements tcpip.Timer.Reset. -func (ft *fakeTimer) Reset(d time.Duration) { - if !ft.timer.Reset(d) { +func (t *manualTimer) Reset(d time.Duration) { + if !t.timer.Reset(d) { return } - ft.mu.Lock() - defer ft.mu.Unlock() + t.mu.Lock() + defer t.mu.Unlock() - ft.clock.removeWait(ft.until) - ft.until = ft.clock.clock.Now().Add(d) - ft.clock.addWait(ft.until) + t.clock.removeWait(t.until) + t.until = t.clock.clock.Now().Add(d) + t.clock.addWait(t.until) } // Stop implements tcpip.Timer.Stop. -func (ft *fakeTimer) Stop() bool { - if !ft.timer.Stop() { +func (t *manualTimer) Stop() bool { + if !t.timer.Stop() { return false } - ft.mu.RLock() - defer ft.mu.RUnlock() + t.mu.RLock() + defer t.mu.RUnlock() - ft.clock.removeWait(ft.until) + t.clock.removeWait(t.until) return true } diff --git a/pkg/tcpip/faketime/faketime_test.go b/pkg/tcpip/faketime/faketime_test.go new file mode 100644 index 000000000..c2704df2c --- /dev/null +++ b/pkg/tcpip/faketime/faketime_test.go @@ -0,0 +1,95 @@ +// 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 faketime_test + +import ( + "testing" + "time" + + "gvisor.dev/gvisor/pkg/tcpip/faketime" +) + +func TestManualClockAdvance(t *testing.T) { + const timeout = time.Millisecond + clock := faketime.NewManualClock() + start := clock.NowMonotonic() + clock.Advance(timeout) + if got, want := time.Duration(clock.NowMonotonic()-start)*time.Nanosecond, timeout; got != want { + t.Errorf("got = %d, want = %d", got, want) + } +} + +func TestManualClockAfterFunc(t *testing.T) { + const ( + timeout1 = time.Millisecond // timeout for counter1 + timeout2 = 2 * time.Millisecond // timeout for counter2 + ) + tests := []struct { + name string + advance time.Duration + wantCounter1 int + wantCounter2 int + }{ + { + name: "before timeout1", + advance: timeout1 - 1, + wantCounter1: 0, + wantCounter2: 0, + }, + { + name: "timeout1", + advance: timeout1, + wantCounter1: 1, + wantCounter2: 0, + }, + { + name: "timeout2", + advance: timeout2, + wantCounter1: 1, + wantCounter2: 1, + }, + { + name: "after timeout2", + advance: timeout2 + 1, + wantCounter1: 1, + wantCounter2: 1, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + clock := faketime.NewManualClock() + counter1 := 0 + counter2 := 0 + clock.AfterFunc(timeout1, func() { + counter1++ + }) + clock.AfterFunc(timeout2, func() { + counter2++ + }) + start := clock.NowMonotonic() + clock.Advance(test.advance) + if got, want := counter1, test.wantCounter1; got != want { + t.Errorf("got counter1 = %d, want = %d", got, want) + } + if got, want := counter2, test.wantCounter2; got != want { + t.Errorf("got counter2 = %d, want = %d", got, want) + } + if got, want := time.Duration(clock.NowMonotonic()-start)*time.Nanosecond, test.advance; got != want { + t.Errorf("got elapsed = %d, want = %d", got, want) + } + }) + } +} diff --git a/pkg/tcpip/header/eth.go b/pkg/tcpip/header/eth.go index eaface8cb..95ade0e5c 100644 --- a/pkg/tcpip/header/eth.go +++ b/pkg/tcpip/header/eth.go @@ -117,25 +117,31 @@ func (b Ethernet) Encode(e *EthernetFields) { copy(b[dstMAC:][:EthernetAddressSize], e.DstAddr) } -// IsValidUnicastEthernetAddress returns true if addr is a valid unicast +// IsMulticastEthernetAddress returns true if the address is a multicast +// ethernet address. +func IsMulticastEthernetAddress(addr tcpip.LinkAddress) bool { + if len(addr) != EthernetAddressSize { + return false + } + + return addr[unicastMulticastFlagByteIdx]&unicastMulticastFlagMask != 0 +} + +// IsValidUnicastEthernetAddress returns true if the address is a unicast // ethernet address. func IsValidUnicastEthernetAddress(addr tcpip.LinkAddress) bool { - // Must be of the right length. if len(addr) != EthernetAddressSize { return false } - // Must not be unspecified. if addr == unspecifiedEthernetAddress { return false } - // Must not be a multicast. if addr[unicastMulticastFlagByteIdx]&unicastMulticastFlagMask != 0 { return false } - // addr is a valid unicast ethernet address. return true } diff --git a/pkg/tcpip/header/eth_test.go b/pkg/tcpip/header/eth_test.go index 14413f2ce..3bc8b2b21 100644 --- a/pkg/tcpip/header/eth_test.go +++ b/pkg/tcpip/header/eth_test.go @@ -67,6 +67,53 @@ func TestIsValidUnicastEthernetAddress(t *testing.T) { } } +func TestIsMulticastEthernetAddress(t *testing.T) { + tests := []struct { + name string + addr tcpip.LinkAddress + expected bool + }{ + { + "Nil", + tcpip.LinkAddress([]byte(nil)), + false, + }, + { + "Empty", + tcpip.LinkAddress(""), + false, + }, + { + "InvalidLength", + tcpip.LinkAddress("\x01\x02\x03"), + false, + }, + { + "Unspecified", + unspecifiedEthernetAddress, + false, + }, + { + "Multicast", + tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"), + true, + }, + { + "Unicast", + tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06"), + false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + if got := IsMulticastEthernetAddress(test.addr); got != test.expected { + t.Fatalf("got IsMulticastEthernetAddress = %t, want = %t", got, test.expected) + } + }) + } +} + func TestEthernetAddressFromMulticastIPv4Address(t *testing.T) { tests := []struct { name string diff --git a/pkg/tcpip/header/icmpv4.go b/pkg/tcpip/header/icmpv4.go index be03fb086..504408878 100644 --- a/pkg/tcpip/header/icmpv4.go +++ b/pkg/tcpip/header/icmpv4.go @@ -31,6 +31,27 @@ const ( // ICMPv4MinimumSize is the minimum size of a valid ICMP packet. ICMPv4MinimumSize = 8 + // ICMPv4MinimumErrorPayloadSize Is the smallest number of bytes of an + // errant packet's transport layer that an ICMP error type packet should + // attempt to send as per RFC 792 (see each type) and RFC 1122 + // section 3.2.2 which states: + // Every ICMP error message includes the Internet header and at + // least the first 8 data octets of the datagram that triggered + // the error; more than 8 octets MAY be sent; this header and data + // MUST be unchanged from the received datagram. + // + // RFC 792 shows: + // 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 | + // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ + // | unused | + // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ + // | Internet Header + 64 bits of Original Data Datagram | + // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ + ICMPv4MinimumErrorPayloadSize = 8 + // ICMPv4ProtocolNumber is the ICMP transport protocol number. ICMPv4ProtocolNumber tcpip.TransportProtocolNumber = 1 @@ -39,15 +60,19 @@ const ( icmpv4ChecksumOffset = 2 // icmpv4MTUOffset is the offset of the MTU field - // in a ICMPv4FragmentationNeeded message. + // in an ICMPv4FragmentationNeeded message. icmpv4MTUOffset = 6 // icmpv4IdentOffset is the offset of the ident field - // in a ICMPv4EchoRequest/Reply message. + // in an ICMPv4EchoRequest/Reply message. icmpv4IdentOffset = 4 + // icmpv4PointerOffset is the offset of the pointer field + // in an ICMPv4ParamProblem message. + icmpv4PointerOffset = 4 + // icmpv4SequenceOffset is the offset of the sequence field - // in a ICMPv4EchoRequest/Reply message. + // in an ICMPv4EchoRequest/Reply message. icmpv4SequenceOffset = 6 ) @@ -72,15 +97,23 @@ const ( ICMPv4InfoReply ICMPv4Type = 16 ) +// ICMP codes for ICMPv4 Time Exceeded messages as defined in RFC 792. +const ( + ICMPv4TTLExceeded ICMPv4Code = 0 +) + // ICMP codes for ICMPv4 Destination Unreachable messages as defined in RFC 792. const ( - ICMPv4TTLExceeded ICMPv4Code = 0 + ICMPv4NetUnreachable ICMPv4Code = 0 ICMPv4HostUnreachable ICMPv4Code = 1 ICMPv4ProtoUnreachable ICMPv4Code = 2 ICMPv4PortUnreachable ICMPv4Code = 3 ICMPv4FragmentationNeeded ICMPv4Code = 4 ) +// ICMPv4UnusedCode is a code to use in ICMP messages where no code is needed. +const ICMPv4UnusedCode ICMPv4Code = 0 + // Type is the ICMP type field. func (b ICMPv4) Type() ICMPv4Type { return ICMPv4Type(b[0]) } diff --git a/pkg/tcpip/header/icmpv6.go b/pkg/tcpip/header/icmpv6.go index 20b01d8f4..6be31beeb 100644 --- a/pkg/tcpip/header/icmpv6.go +++ b/pkg/tcpip/header/icmpv6.go @@ -54,9 +54,17 @@ const ( // address. ICMPv6NeighborAdvertSize = ICMPv6HeaderSize + NDPNAMinimumSize + NDPLinkLayerAddressSize - // ICMPv6EchoMinimumSize is the minimum size of a valid ICMP echo packet. + // ICMPv6EchoMinimumSize is the minimum size of a valid echo packet. ICMPv6EchoMinimumSize = 8 + // ICMPv6ErrorHeaderSize is the size of an ICMP error packet header, + // as per RFC 4443, Apendix A, item 4 and the errata. + // ... all ICMP error messages shall have exactly + // 32 bits of type-specific data, so that receivers can reliably find + // the embedded invoking packet even when they don't recognize the + // ICMP message Type. + ICMPv6ErrorHeaderSize = 8 + // ICMPv6DstUnreachableMinimumSize is the minimum size of a valid ICMP // destination unreachable packet. ICMPv6DstUnreachableMinimumSize = ICMPv6MinimumSize @@ -69,6 +77,10 @@ const ( // in an ICMPv6 message. icmpv6ChecksumOffset = 2 + // icmpv6PointerOffset is the offset of the pointer + // in an ICMPv6 Parameter problem message. + icmpv6PointerOffset = 4 + // icmpv6MTUOffset is the offset of the MTU field in an ICMPv6 // PacketTooBig message. icmpv6MTUOffset = 4 @@ -89,9 +101,10 @@ const ( NDPHopLimit = 255 ) -// ICMPv6Type is the ICMP type field described in RFC 4443 and friends. +// ICMPv6Type is the ICMP type field described in RFC 4443. type ICMPv6Type byte +// Values for use in the Type field of ICMPv6 packet from RFC 4433. const ( ICMPv6DstUnreachable ICMPv6Type = 1 ICMPv6PacketTooBig ICMPv6Type = 2 @@ -109,7 +122,18 @@ const ( ICMPv6RedirectMsg ICMPv6Type = 137 ) -// ICMPv6Code is the ICMP code field described in RFC 4443. +// IsErrorType returns true if the receiver is an ICMP error type. +func (typ ICMPv6Type) IsErrorType() bool { + // Per RFC 4443 section 2.1: + // ICMPv6 messages are grouped into two classes: error messages and + // informational messages. Error messages are identified as such by a + // zero in the high-order bit of their message Type field values. Thus, + // error messages have message types from 0 to 127; informational + // messages have message types from 128 to 255. + return typ&0x80 == 0 +} + +// ICMPv6Code is the ICMP Code field described in RFC 4443. type ICMPv6Code byte // ICMP codes used with Destination Unreachable (Type 1). As per RFC 4443 @@ -132,9 +156,14 @@ const ( // ICMP codes used with Parameter Problem (Type 4). As per RFC 4443 section 3.4. const ( + // ICMPv6ErroneousHeader indicates an erroneous header field was encountered. ICMPv6ErroneousHeader ICMPv6Code = 0 - ICMPv6UnknownHeader ICMPv6Code = 1 - ICMPv6UnknownOption ICMPv6Code = 2 + + // ICMPv6UnknownHeader indicates an unrecognized Next Header type encountered. + ICMPv6UnknownHeader ICMPv6Code = 1 + + // ICMPv6UnknownOption indicates an unrecognized IPv6 option was encountered. + ICMPv6UnknownOption ICMPv6Code = 2 ) // ICMPv6UnusedCode is the code value used with ICMPv6 messages which don't use @@ -153,6 +182,16 @@ func (b ICMPv6) Code() ICMPv6Code { return ICMPv6Code(b[1]) } // SetCode sets the ICMP code field. func (b ICMPv6) SetCode(c ICMPv6Code) { b[1] = byte(c) } +// TypeSpecific returns the type specific data field. +func (b ICMPv6) TypeSpecific() uint32 { + return binary.BigEndian.Uint32(b[icmpv6PointerOffset:]) +} + +// SetTypeSpecific sets the type specific data field. +func (b ICMPv6) SetTypeSpecific(val uint32) { + binary.BigEndian.PutUint32(b[icmpv6PointerOffset:], val) +} + // Checksum is the ICMP checksum field. func (b ICMPv6) Checksum() uint16 { return binary.BigEndian.Uint16(b[icmpv6ChecksumOffset:]) diff --git a/pkg/tcpip/header/ipv4.go b/pkg/tcpip/header/ipv4.go index e8816c3f4..4c6e4be64 100644 --- a/pkg/tcpip/header/ipv4.go +++ b/pkg/tcpip/header/ipv4.go @@ -16,10 +16,29 @@ package header import ( "encoding/binary" + "fmt" "gvisor.dev/gvisor/pkg/tcpip" ) +// RFC 971 defines the fields of the IPv4 header on page 11 using the following +// diagram: ("Figure 4") +// 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 +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// |Version| IHL |Type of Service| Total Length | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Identification |Flags| Fragment Offset | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Time to Live | Protocol | Header Checksum | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Source Address | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Destination Address | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// | Options | Padding | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// const ( versIHL = 0 tos = 1 @@ -33,6 +52,7 @@ const ( checksum = 10 srcAddr = 12 dstAddr = 16 + options = 20 ) // IPv4Fields contains the fields of an IPv4 packet. It is used to describe the @@ -76,11 +96,13 @@ type IPv4Fields struct { // IPv4 represents an ipv4 header stored in a byte array. // Most of the methods of IPv4 access to the underlying slice without // checking the boundaries and could panic because of 'index out of range'. -// Always call IsValid() to validate an instance of IPv4 before using other methods. +// Always call IsValid() to validate an instance of IPv4 before using other +// methods. type IPv4 []byte const ( - // IPv4MinimumSize is the minimum size of a valid IPv4 packet. + // IPv4MinimumSize is the minimum size of a valid IPv4 packet; + // i.e. a packet header with no options. IPv4MinimumSize = 20 // IPv4MaximumHeaderSize is the maximum size of an IPv4 header. Given @@ -150,13 +172,44 @@ func IPVersion(b []byte) int { if len(b) < versIHL+1 { return -1 } - return int(b[versIHL] >> 4) + return int(b[versIHL] >> ipVersionShift) } +// RFC 791 page 11 shows the header length (IHL) is in the lower 4 bits +// of the first byte, and is counted in multiples of 4 bytes. +// +// 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 +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// |Version| IHL |Type of Service| Total Length | +// +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +// (...) +// Version: 4 bits +// The Version field indicates the format of the internet header. This +// document describes version 4. +// +// IHL: 4 bits +// Internet Header Length is the length of the internet header in 32 +// bit words, and thus points to the beginning of the data. Note that +// the minimum value for a correct header is 5. +// +const ( + ipVersionShift = 4 + ipIHLMask = 0x0f + IPv4IHLStride = 4 +) + // HeaderLength returns the value of the "header length" field of the ipv4 // header. The length returned is in bytes. func (b IPv4) HeaderLength() uint8 { - return (b[versIHL] & 0xf) * 4 + return (b[versIHL] & ipIHLMask) * IPv4IHLStride +} + +// SetHeaderLength sets the value of the "Internet Header Length" field. +func (b IPv4) SetHeaderLength(hdrLen uint8) { + if hdrLen > IPv4MaximumHeaderSize { + panic(fmt.Sprintf("got IPv4 Header size = %d, want <= %d", hdrLen, IPv4MaximumHeaderSize)) + } + b[versIHL] = (IPv4Version << ipVersionShift) | ((hdrLen / IPv4IHLStride) & ipIHLMask) } // ID returns the value of the identifier field of the ipv4 header. @@ -210,6 +263,12 @@ func (b IPv4) DestinationAddress() tcpip.Address { return tcpip.Address(b[dstAddr : dstAddr+IPv4AddressSize]) } +// Options returns a a buffer holding the options. +func (b IPv4) Options() []byte { + hdrLen := b.HeaderLength() + return b[options:hdrLen:hdrLen] +} + // TransportProtocol implements Network.TransportProtocol. func (b IPv4) TransportProtocol() tcpip.TransportProtocolNumber { return tcpip.TransportProtocolNumber(b.Protocol()) @@ -235,6 +294,11 @@ func (b IPv4) SetTOS(v uint8, _ uint32) { b[tos] = v } +// SetTTL sets the "Time to Live" field of the IPv4 header. +func (b IPv4) SetTTL(v byte) { + b[ttl] = v +} + // SetTotalLength sets the "total length" field of the ipv4 header. func (b IPv4) SetTotalLength(totalLength uint16) { binary.BigEndian.PutUint16(b[IPv4TotalLenOffset:], totalLength) @@ -275,7 +339,7 @@ func (b IPv4) CalculateChecksum() uint16 { // Encode encodes all the fields of the ipv4 header. func (b IPv4) Encode(i *IPv4Fields) { - b[versIHL] = (4 << 4) | ((i.IHL / 4) & 0xf) + b.SetHeaderLength(i.IHL) b[tos] = i.TOS b.SetTotalLength(i.TotalLength) binary.BigEndian.PutUint16(b[id:], i.ID) @@ -327,7 +391,7 @@ func IsV4MulticastAddress(addr tcpip.Address) bool { } // IsV4LoopbackAddress determines if the provided address is an IPv4 loopback -// address (belongs to 127.0.0.1/8 subnet). +// address (belongs to 127.0.0.0/8 subnet). See RFC 1122 section 3.2.1.3. func IsV4LoopbackAddress(addr tcpip.Address) bool { if len(addr) != IPv4AddressSize { return false diff --git a/pkg/tcpip/header/ipv6.go b/pkg/tcpip/header/ipv6.go index 0761a1807..c5d8a3456 100644 --- a/pkg/tcpip/header/ipv6.go +++ b/pkg/tcpip/header/ipv6.go @@ -34,6 +34,9 @@ const ( hopLimit = 7 v6SrcAddr = 8 v6DstAddr = v6SrcAddr + IPv6AddressSize + + // IPv6FixedHeaderSize is the size of the fixed header. + IPv6FixedHeaderSize = v6DstAddr + IPv6AddressSize ) // IPv6Fields contains the fields of an IPv6 packet. It is used to describe the @@ -69,7 +72,7 @@ type IPv6 []byte const ( // IPv6MinimumSize is the minimum size of a valid IPv6 packet. - IPv6MinimumSize = 40 + IPv6MinimumSize = IPv6FixedHeaderSize // IPv6AddressSize is the size, in bytes, of an IPv6 address. IPv6AddressSize = 16 @@ -306,14 +309,21 @@ func IsV6UnicastAddress(addr tcpip.Address) bool { return addr[0] != 0xff } +const solicitedNodeMulticastPrefix = "\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\xff" + // SolicitedNodeAddr computes the solicited-node multicast address. This is // used for NDP. Described in RFC 4291. The argument must be a full-length IPv6 // address. func SolicitedNodeAddr(addr tcpip.Address) tcpip.Address { - const solicitedNodeMulticastPrefix = "\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\xff" return solicitedNodeMulticastPrefix + addr[len(addr)-3:] } +// IsSolicitedNodeAddr determines whether the address is a solicited-node +// multicast address. +func IsSolicitedNodeAddr(addr tcpip.Address) bool { + return solicitedNodeMulticastPrefix == addr[:len(addr)-3] +} + // EthernetAdddressToModifiedEUI64IntoBuf populates buf with a modified EUI-64 // from a 48-bit Ethernet/MAC address, as per RFC 4291 section 2.5.1. // diff --git a/pkg/tcpip/header/ipv6_extension_headers.go b/pkg/tcpip/header/ipv6_extension_headers.go index 3499d8399..583c2c5d3 100644 --- a/pkg/tcpip/header/ipv6_extension_headers.go +++ b/pkg/tcpip/header/ipv6_extension_headers.go @@ -149,6 +149,19 @@ func (b ipv6OptionsExtHdr) Iter() IPv6OptionsExtHdrOptionsIterator { // obtained before modification is no longer used. type IPv6OptionsExtHdrOptionsIterator struct { reader bytes.Reader + + // optionOffset is the number of bytes from the first byte of the + // options field to the beginning of the current option. + optionOffset uint32 + + // nextOptionOffset is the offset of the next option. + nextOptionOffset uint32 +} + +// OptionOffset returns the number of bytes parsed while processing the +// option field of the current Extension Header. +func (i *IPv6OptionsExtHdrOptionsIterator) OptionOffset() uint32 { + return i.optionOffset } // IPv6OptionUnknownAction is the action that must be taken if the processing @@ -226,6 +239,7 @@ func (*IPv6UnknownExtHdrOption) isIPv6ExtHdrOption() {} // the options data, or an error occured. func (i *IPv6OptionsExtHdrOptionsIterator) Next() (IPv6ExtHdrOption, bool, error) { for { + i.optionOffset = i.nextOptionOffset temp, err := i.reader.ReadByte() if err != nil { // If we can't read the first byte of a new option, then we know the @@ -238,6 +252,7 @@ func (i *IPv6OptionsExtHdrOptionsIterator) Next() (IPv6ExtHdrOption, bool, error // know the option does not have Length and Data fields. End processing of // the Pad1 option and continue processing the buffer as a new option. if id == ipv6Pad1ExtHdrOptionIdentifier { + i.nextOptionOffset = i.optionOffset + 1 continue } @@ -254,41 +269,40 @@ func (i *IPv6OptionsExtHdrOptionsIterator) Next() (IPv6ExtHdrOption, bool, error return nil, true, fmt.Errorf("error when reading the option's Length field for option with id = %d: %w", id, io.ErrUnexpectedEOF) } - // Special-case the variable length padding option to avoid a copy. - if id == ipv6PadNExtHdrOptionIdentifier { - // Do we have enough bytes in the reader for the PadN option? - if n := i.reader.Len(); n < int(length) { - // Reset the reader to effectively consume the remaining buffer. - i.reader.Reset(nil) - - // We return the same error as if we failed to read a non-padding option - // so consumers of this iterator don't need to differentiate between - // padding and non-padding options. - return nil, true, fmt.Errorf("read %d out of %d option data bytes for option with id = %d: %w", n, length, id, io.ErrUnexpectedEOF) - } + // Do we have enough bytes in the reader for the next option? + if n := i.reader.Len(); n < int(length) { + // Reset the reader to effectively consume the remaining buffer. + i.reader.Reset(nil) + + // We return the same error as if we failed to read a non-padding option + // so consumers of this iterator don't need to differentiate between + // padding and non-padding options. + return nil, true, fmt.Errorf("read %d out of %d option data bytes for option with id = %d: %w", n, length, id, io.ErrUnexpectedEOF) + } + + i.nextOptionOffset = i.optionOffset + uint32(length) + 1 /* option ID */ + 1 /* length byte */ + switch id { + case ipv6PadNExtHdrOptionIdentifier: + // Special-case the variable length padding option to avoid a copy. if _, err := i.reader.Seek(int64(length), io.SeekCurrent); err != nil { panic(fmt.Sprintf("error when skipping PadN (N = %d) option's data bytes: %s", length, err)) } - - // End processing of the PadN option and continue processing the buffer as - // a new option. continue - } - - bytes := make([]byte, length) - if n, err := io.ReadFull(&i.reader, bytes); err != nil { - // io.ReadFull may return io.EOF if i.reader has been exhausted. We use - // io.ErrUnexpectedEOF instead as the io.EOF is unexpected given the - // Length field found in the option. - if err == io.EOF { - err = io.ErrUnexpectedEOF + default: + bytes := make([]byte, length) + if n, err := io.ReadFull(&i.reader, bytes); err != nil { + // io.ReadFull may return io.EOF if i.reader has been exhausted. We use + // io.ErrUnexpectedEOF instead as the io.EOF is unexpected given the + // Length field found in the option. + if err == io.EOF { + err = io.ErrUnexpectedEOF + } + + return nil, true, fmt.Errorf("read %d out of %d option data bytes for option with id = %d: %w", n, length, id, err) } - - return nil, true, fmt.Errorf("read %d out of %d option data bytes for option with id = %d: %w", n, length, id, err) + return &IPv6UnknownExtHdrOption{Identifier: id, Data: bytes}, false, nil } - - return &IPv6UnknownExtHdrOption{Identifier: id, Data: bytes}, false, nil } } @@ -382,6 +396,29 @@ type IPv6PayloadIterator struct { // Indicates to the iterator that it should return the remaining payload as a // raw payload on the next call to Next. forceRaw bool + + // headerOffset is the offset of the beginning of the current extension + // header starting from the beginning of the fixed header. + headerOffset uint32 + + // parseOffset is the byte offset into the current extension header of the + // field we are currently examining. It can be added to the header offset + // if the absolute offset within the packet is required. + parseOffset uint32 + + // nextOffset is the offset of the next header. + nextOffset uint32 +} + +// HeaderOffset returns the offset to the start of the extension +// header most recently processed. +func (i IPv6PayloadIterator) HeaderOffset() uint32 { + return i.headerOffset +} + +// ParseOffset returns the number of bytes successfully parsed. +func (i IPv6PayloadIterator) ParseOffset() uint32 { + return i.headerOffset + i.parseOffset } // MakeIPv6PayloadIterator returns an iterator over the IPv6 payload containing @@ -397,7 +434,8 @@ func MakeIPv6PayloadIterator(nextHdrIdentifier IPv6ExtensionHeaderIdentifier, pa nextHdrIdentifier: nextHdrIdentifier, payload: payload.Clone(nil), // We need a buffer of size 1 for calls to bufio.Reader.ReadByte. - reader: *bufio.NewReaderSize(io.MultiReader(readerPs...), 1), + reader: *bufio.NewReaderSize(io.MultiReader(readerPs...), 1), + nextOffset: IPv6FixedHeaderSize, } } @@ -434,6 +472,8 @@ func (i *IPv6PayloadIterator) AsRawHeader(consume bool) IPv6RawPayloadHeader { // Next is unable to return anything because the iterator has reached the end of // the payload, or an error occured. func (i *IPv6PayloadIterator) Next() (IPv6PayloadHeader, bool, error) { + i.headerOffset = i.nextOffset + i.parseOffset = 0 // We could be forced to return i as a raw header when the previous header was // a fragment extension header as the data following the fragment extension // header may not be complete. @@ -461,7 +501,7 @@ func (i *IPv6PayloadIterator) Next() (IPv6PayloadHeader, bool, error) { return IPv6RoutingExtHdr(bytes), false, nil case IPv6FragmentExtHdrIdentifier: var data [6]byte - // We ignore the returned bytes becauase we know the fragment extension + // We ignore the returned bytes because we know the fragment extension // header specific data will fit in data. nextHdrIdentifier, _, err := i.nextHeaderData(true /* fragmentHdr */, data[:]) if err != nil { @@ -519,10 +559,12 @@ func (i *IPv6PayloadIterator) nextHeaderData(fragmentHdr bool, bytes []byte) (IP if err != nil { return 0, nil, fmt.Errorf("error when reading the Next Header field for extension header with id = %d: %w", i.nextHdrIdentifier, err) } + i.parseOffset++ var length uint8 length, err = i.reader.ReadByte() i.payload.TrimFront(1) + if err != nil { if fragmentHdr { return 0, nil, fmt.Errorf("error when reading the Length field for extension header with id = %d: %w", i.nextHdrIdentifier, err) @@ -534,6 +576,17 @@ func (i *IPv6PayloadIterator) nextHeaderData(fragmentHdr bool, bytes []byte) (IP length = 0 } + // Make parseOffset point to the first byte of the Extension Header + // specific data. + i.parseOffset++ + + // length is in 8 byte chunks but doesn't include the first one. + // See RFC 8200 for each header type, sections 4.3-4.6 and the requirement + // in section 4.8 for new extension headers at the top of page 24. + // [ Hdr Ext Len ] ... Length of the Destination Options header in 8-octet + // units, not including the first 8 octets. + i.nextOffset += uint32((length + 1) * ipv6ExtHdrLenBytesPerUnit) + bytesLen := int(length)*ipv6ExtHdrLenBytesPerUnit + ipv6ExtHdrLenBytesExcluded if bytes == nil { bytes = make([]byte, bytesLen) diff --git a/pkg/tcpip/header/ipversion_test.go b/pkg/tcpip/header/ipversion_test.go index b5540bf66..17a49d4fa 100644 --- a/pkg/tcpip/header/ipversion_test.go +++ b/pkg/tcpip/header/ipversion_test.go @@ -22,7 +22,7 @@ import ( func TestIPv4(t *testing.T) { b := header.IPv4(make([]byte, header.IPv4MinimumSize)) - b.Encode(&header.IPv4Fields{}) + b.Encode(&header.IPv4Fields{IHL: header.IPv4MinimumSize}) const want = header.IPv4Version if v := header.IPVersion(b); v != want { diff --git a/pkg/tcpip/header/parse/parse.go b/pkg/tcpip/header/parse/parse.go index 522135557..5ca75c834 100644 --- a/pkg/tcpip/header/parse/parse.go +++ b/pkg/tcpip/header/parse/parse.go @@ -139,6 +139,7 @@ traverseExtensions: // Returns true if the header was successfully parsed. func UDP(pkt *stack.PacketBuffer) bool { _, ok := pkt.TransportHeader().Consume(header.UDPMinimumSize) + pkt.TransportProtocolNumber = header.UDPProtocolNumber return ok } @@ -162,5 +163,6 @@ func TCP(pkt *stack.PacketBuffer) bool { } _, ok = pkt.TransportHeader().Consume(hdrLen) + pkt.TransportProtocolNumber = header.TCPProtocolNumber return ok } diff --git a/pkg/tcpip/link/pipe/BUILD b/pkg/tcpip/link/pipe/BUILD new file mode 100644 index 000000000..9f31c1ffc --- /dev/null +++ b/pkg/tcpip/link/pipe/BUILD @@ -0,0 +1,15 @@ +load("//tools:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_library( + name = "pipe", + srcs = ["pipe.go"], + visibility = ["//visibility:public"], + deps = [ + "//pkg/tcpip", + "//pkg/tcpip/buffer", + "//pkg/tcpip/header", + "//pkg/tcpip/stack", + ], +) diff --git a/pkg/tcpip/link/pipe/pipe.go b/pkg/tcpip/link/pipe/pipe.go new file mode 100644 index 000000000..76f563811 --- /dev/null +++ b/pkg/tcpip/link/pipe/pipe.go @@ -0,0 +1,124 @@ +// 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 pipe provides the implementation of pipe-like data-link layer +// endpoints. Such endpoints allow packets to be sent between two interfaces. +package pipe + +import ( + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +var _ stack.LinkEndpoint = (*Endpoint)(nil) + +// New returns both ends of a new pipe. +func New(linkAddr1, linkAddr2 tcpip.LinkAddress, capabilities stack.LinkEndpointCapabilities) (*Endpoint, *Endpoint) { + ep1 := &Endpoint{ + linkAddr: linkAddr1, + capabilities: capabilities, + } + ep2 := &Endpoint{ + linkAddr: linkAddr2, + linked: ep1, + capabilities: capabilities, + } + ep1.linked = ep2 + return ep1, ep2 +} + +// Endpoint is one end of a pipe. +type Endpoint struct { + capabilities stack.LinkEndpointCapabilities + linkAddr tcpip.LinkAddress + dispatcher stack.NetworkDispatcher + linked *Endpoint + onWritePacket func(*stack.PacketBuffer) +} + +// WritePacket implements stack.LinkEndpoint. +func (e *Endpoint) WritePacket(r *stack.Route, _ *stack.GSO, proto tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + if !e.linked.IsAttached() { + return nil + } + + // The pipe endpoint will accept all multicast/broadcast link traffic and only + // unicast traffic destined to itself. + if len(e.linked.linkAddr) != 0 && + r.RemoteLinkAddress != e.linked.linkAddr && + r.RemoteLinkAddress != header.EthernetBroadcastAddress && + !header.IsMulticastEthernetAddress(r.RemoteLinkAddress) { + return nil + } + + e.linked.dispatcher.DeliverNetworkPacket(e.linkAddr, r.RemoteLinkAddress, proto, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), + })) + + return nil +} + +// WritePackets implements stack.LinkEndpoint. +func (*Endpoint) WritePackets(*stack.Route, *stack.GSO, stack.PacketBufferList, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + 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 +} + +// IsAttached implements stack.LinkEndpoint. +func (e *Endpoint) IsAttached() bool { + return e.dispatcher != nil +} + +// Wait implements stack.LinkEndpoint. +func (*Endpoint) Wait() {} + +// MTU implements stack.LinkEndpoint. +func (*Endpoint) MTU() uint32 { + return header.IPv6MinimumMTU +} + +// Capabilities implements stack.LinkEndpoint. +func (e *Endpoint) Capabilities() stack.LinkEndpointCapabilities { + return e.capabilities +} + +// MaxHeaderLength implements stack.LinkEndpoint. +func (*Endpoint) MaxHeaderLength() uint16 { + return 0 +} + +// LinkAddress implements stack.LinkEndpoint. +func (e *Endpoint) LinkAddress() tcpip.LinkAddress { + return e.linkAddr +} + +// ARPHardwareType implements stack.LinkEndpoint. +func (*Endpoint) ARPHardwareType() header.ARPHardwareType { + return header.ARPHardwareEther +} + +// AddHeader implements stack.LinkEndpoint. +func (*Endpoint) AddHeader(_, _ tcpip.LinkAddress, _ tcpip.NetworkProtocolNumber, _ *stack.PacketBuffer) { +} diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index b6ddbe81e..f94491026 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -76,13 +76,29 @@ func (d *Device) Release(ctx context.Context) { } } +// NICID returns the NIC ID of the device. +// +// Must only be called after the device has been attached to an endpoint. +func (d *Device) NICID() tcpip.NICID { + d.mu.RLock() + defer d.mu.RUnlock() + + if d.endpoint == nil { + panic("called NICID on a device that has not been attached") + } + + return d.endpoint.nicID +} + // SetIff services TUNSETIFF ioctl(2) request. -func (d *Device) SetIff(s *stack.Stack, name string, flags uint16) error { +// +// Returns true if a new NIC was created; false if an existing one was attached. +func (d *Device) SetIff(s *stack.Stack, name string, flags uint16) (bool, error) { d.mu.Lock() defer d.mu.Unlock() if d.endpoint != nil { - return syserror.EINVAL + return false, syserror.EINVAL } // Input validations. @@ -90,7 +106,7 @@ func (d *Device) SetIff(s *stack.Stack, name string, flags uint16) error { isTap := flags&linux.IFF_TAP != 0 supportedFlags := uint16(linux.IFF_TUN | linux.IFF_TAP | linux.IFF_NO_PI) if isTap && isTun || !isTap && !isTun || flags&^supportedFlags != 0 { - return syserror.EINVAL + return false, syserror.EINVAL } prefix := "tun" @@ -103,32 +119,32 @@ func (d *Device) SetIff(s *stack.Stack, name string, flags uint16) error { linkCaps |= stack.CapabilityResolutionRequired } - endpoint, err := attachOrCreateNIC(s, name, prefix, linkCaps) + endpoint, created, err := attachOrCreateNIC(s, name, prefix, linkCaps) if err != nil { - return syserror.EINVAL + return false, syserror.EINVAL } d.endpoint = endpoint d.notifyHandle = d.endpoint.AddNotify(d) d.flags = flags - return nil + return created, nil } -func attachOrCreateNIC(s *stack.Stack, name, prefix string, linkCaps stack.LinkEndpointCapabilities) (*tunEndpoint, error) { +func attachOrCreateNIC(s *stack.Stack, name, prefix string, linkCaps stack.LinkEndpointCapabilities) (*tunEndpoint, bool, error) { for { // 1. Try to attach to an existing NIC. if name != "" { - if nic, found := s.GetNICByName(name); found { - endpoint, ok := nic.LinkEndpoint().(*tunEndpoint) + if linkEP := s.GetLinkEndpointByName(name); linkEP != nil { + endpoint, ok := linkEP.(*tunEndpoint) if !ok { // Not a NIC created by tun device. - return nil, syserror.EOPNOTSUPP + return nil, false, syserror.EOPNOTSUPP } if !endpoint.TryIncRef() { // Race detected: NIC got deleted in between. continue } - return endpoint, nil + return endpoint, false, nil } } @@ -151,12 +167,12 @@ func attachOrCreateNIC(s *stack.Stack, name, prefix string, linkCaps stack.LinkE }) switch err { case nil: - return endpoint, nil + return endpoint, true, nil case tcpip.ErrDuplicateNICID: // Race detected: A NIC has been created in between. continue default: - return nil, syserror.EINVAL + return nil, false, syserror.EINVAL } } } diff --git a/pkg/tcpip/network/BUILD b/pkg/tcpip/network/BUILD index 46083925c..59710352b 100644 --- a/pkg/tcpip/network/BUILD +++ b/pkg/tcpip/network/BUILD @@ -9,6 +9,7 @@ go_test( "ip_test.go", ], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", @@ -17,6 +18,7 @@ go_test( "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", "//pkg/tcpip/stack", + "//pkg/tcpip/transport/icmp", "//pkg/tcpip/transport/tcp", "//pkg/tcpip/transport/udp", ], diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index cb9225bd7..7df77c66e 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -15,17 +15,11 @@ // Package arp implements the ARP network protocol. It is used to resolve // IPv4 addresses into link-local MAC addresses, and advertises IPv4 // addresses of its stack with the local network. -// -// To use it in the networking stack, pass arp.NewProtocol() as one of the -// network protocols when calling stack.New. Then add an "arp" address to every -// NIC on the stack that should respond to ARP requests. That is: -// -// if err := s.AddAddress(1, arp.ProtocolNumber, "arp"); err != nil { -// // handle err -// } package arp import ( + "sync/atomic" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -41,38 +35,73 @@ const ( ProtocolAddress = tcpip.Address("arp") ) -// endpoint implements stack.NetworkEndpoint. +var _ stack.AddressableEndpoint = (*endpoint)(nil) +var _ stack.NetworkEndpoint = (*endpoint)(nil) + type endpoint struct { - protocol *protocol - nicID tcpip.NICID - linkEP stack.LinkEndpoint + stack.AddressableEndpointState + + protocol *protocol + + // enabled is set to 1 when the NIC is enabled and 0 when it is disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + + nic stack.NetworkInterface linkAddrCache stack.LinkAddressCache nud stack.NUDHandler } +func (e *endpoint) Enable() *tcpip.Error { + if !e.nic.Enabled() { + return tcpip.ErrNotPermitted + } + + e.setEnabled(true) + return nil +} + +func (e *endpoint) Enabled() bool { + return e.nic.Enabled() && e.isEnabled() +} + +// isEnabled returns true if the endpoint is enabled, regardless of the +// enabled status of the NIC. +func (e *endpoint) isEnabled() bool { + return atomic.LoadUint32(&e.enabled) == 1 +} + +// setEnabled sets the enabled status for the endpoint. +func (e *endpoint) setEnabled(v bool) { + if v { + atomic.StoreUint32(&e.enabled, 1) + } else { + atomic.StoreUint32(&e.enabled, 0) + } +} + +func (e *endpoint) Disable() { + e.setEnabled(false) +} + // DefaultTTL is unused for ARP. It implements stack.NetworkEndpoint. func (e *endpoint) DefaultTTL() uint8 { return 0 } func (e *endpoint) MTU() uint32 { - lmtu := e.linkEP.MTU() + lmtu := e.nic.MTU() return lmtu - uint32(e.MaxHeaderLength()) } -func (e *endpoint) NICID() tcpip.NICID { - return e.nicID -} - -func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { - return e.linkEP.Capabilities() -} - func (e *endpoint) MaxHeaderLength() uint16 { - return e.linkEP.MaxHeaderLength() + header.ARPSize + return e.nic.MaxHeaderLength() + header.ARPSize } -func (e *endpoint) Close() {} +func (e *endpoint) Close() { + e.AddressableEndpointState.Cleanup() +} func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, stack.NetworkHeaderParams, *stack.PacketBuffer) *tcpip.Error { return tcpip.ErrNotSupported @@ -93,6 +122,10 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu } func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { + if !e.isEnabled() { + return + } + h := header.ARP(pkt.NetworkHeader().View()) if !h.IsValid() { return @@ -103,15 +136,15 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { localAddr := tcpip.Address(h.ProtocolAddressTarget()) if e.nud == nil { - if e.linkAddrCache.CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { + if e.linkAddrCache.CheckLocalAddress(e.nic.ID(), header.IPv4ProtocolNumber, localAddr) == 0 { return // we have no useful answer, ignore the request } addr := tcpip.Address(h.ProtocolAddressSender()) linkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) - e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) + e.linkAddrCache.AddLinkAddress(e.nic.ID(), addr, linkAddr) } else { - if r.Stack().CheckLocalAddress(e.nicID, header.IPv4ProtocolNumber, localAddr) == 0 { + if r.Stack().CheckLocalAddress(e.nic.ID(), header.IPv4ProtocolNumber, localAddr) == 0 { return // we have no useful answer, ignore the request } @@ -120,24 +153,32 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { e.nud.HandleProbe(remoteAddr, localAddr, ProtocolNumber, remoteLinkAddr, e.protocol) } - pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - ReserveHeaderBytes: int(e.linkEP.MaxHeaderLength()) + header.ARPSize, + // As per RFC 826, under Packet Reception: + // Swap hardware and protocol fields, putting the local hardware and + // protocol addresses in the sender fields. + // + // Send the packet to the (new) target hardware address on the same + // hardware on which the request was received. + origSender := h.HardwareAddressSender() + r.RemoteLinkAddress = tcpip.LinkAddress(origSender) + respPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: int(e.nic.MaxHeaderLength()) + header.ARPSize, }) - packet := header.ARP(pkt.NetworkHeader().Push(header.ARPSize)) + packet := header.ARP(respPkt.NetworkHeader().Push(header.ARPSize)) packet.SetIPv4OverEthernet() packet.SetOp(header.ARPReply) copy(packet.HardwareAddressSender(), r.LocalLinkAddress[:]) copy(packet.ProtocolAddressSender(), h.ProtocolAddressTarget()) - copy(packet.HardwareAddressTarget(), h.HardwareAddressSender()) + copy(packet.HardwareAddressTarget(), origSender) copy(packet.ProtocolAddressTarget(), h.ProtocolAddressSender()) - _ = e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) + _ = e.nic.WritePacket(r, nil /* gso */, ProtocolNumber, respPkt) case header.ARPReply: addr := tcpip.Address(h.ProtocolAddressSender()) linkAddr := tcpip.LinkAddress(h.HardwareAddressSender()) if e.nud == nil { - e.linkAddrCache.AddLinkAddress(e.nicID, addr, linkAddr) + e.linkAddrCache.AddLinkAddress(e.nic.ID(), addr, linkAddr) return } @@ -169,14 +210,15 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { return tcpip.Address(h.ProtocolAddressSender()), ProtocolAddress } -func (p *protocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint { - return &endpoint{ +func (p *protocol) NewEndpoint(nic stack.NetworkInterface, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher) stack.NetworkEndpoint { + e := &endpoint{ protocol: p, - nicID: nicID, - linkEP: sender, + nic: nic, linkAddrCache: linkAddrCache, nud: nud, } + e.AddressableEndpointState.Init(e) + return e } // LinkAddressProtocol implements stack.LinkAddressResolver.LinkAddressProtocol. @@ -187,6 +229,7 @@ func (*protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { // LinkAddressRequest implements stack.LinkAddressResolver.LinkAddressRequest. func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP stack.LinkEndpoint) *tcpip.Error { r := &stack.Route{ + NetProto: ProtocolNumber, RemoteLinkAddress: remoteLinkAddr, } if len(r.RemoteLinkAddress) == 0 { @@ -239,6 +282,10 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu } // NewProtocol returns an ARP network protocol. -func NewProtocol() stack.NetworkProtocol { +// +// Note, to make sure that the ARP endpoint receives ARP packets, the "arp" +// address must be added to every NIC that should respond to ARP requests. See +// ProtocolAddress for more details. +func NewProtocol(*stack.Stack) stack.NetworkProtocol { return &protocol{} } diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 9c9a859e3..626af975a 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -176,8 +176,8 @@ func newTestContext(t *testing.T, useNeighborCache bool) *testContext { } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, arp.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol4}, NUDConfigs: c, NUDDisp: &d, UseNeighborCache: useNeighborCache, @@ -442,7 +442,7 @@ func TestLinkAddressRequest(t *testing.T) { } for _, test := range tests { - p := arp.NewProtocol() + p := arp.NewProtocol(nil) linkRes, ok := p.(stack.LinkAddressResolver) if !ok { t.Fatal("expected ARP protocol to implement stack.LinkAddressResolver") diff --git a/pkg/tcpip/network/fragmentation/BUILD b/pkg/tcpip/network/fragmentation/BUILD index 96c5f42f8..47fb63290 100644 --- a/pkg/tcpip/network/fragmentation/BUILD +++ b/pkg/tcpip/network/fragmentation/BUILD @@ -29,6 +29,8 @@ go_library( "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", + "//pkg/tcpip/stack", ], ) @@ -43,5 +45,8 @@ go_test( library = ":fragmentation", deps = [ "//pkg/tcpip/buffer", + "//pkg/tcpip/faketime", + "//pkg/tcpip/network/testutil", + "@com_github_google_go_cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index 6a4843f92..ed502a473 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -13,7 +13,7 @@ // limitations under the License. // Package fragmentation contains the implementation of IP fragmentation. -// It is based on RFC 791 and RFC 815. +// It is based on RFC 791, RFC 815 and RFC 8200. package fragmentation import ( @@ -25,12 +25,10 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) const ( - // DefaultReassembleTimeout is based on the linux stack: net.ipv4.ipfrag_time. - DefaultReassembleTimeout = 30 * time.Second - // HighFragThreshold is the threshold at which we start trimming old // fragmented packets. Linux uses a default value of 4 MB. See // net.ipv4.ipfrag_high_thresh for more information. @@ -81,6 +79,8 @@ type Fragmentation struct { size int timeout time.Duration blockSize uint16 + clock tcpip.Clock + releaseJob *tcpip.Job } // NewFragmentation creates a new Fragmentation. @@ -97,7 +97,7 @@ type Fragmentation struct { // reassemblingTimeout specifies the maximum time allowed to reassemble a packet. // Fragments are lazily evicted only when a new a packet with an // already existing fragmentation-id arrives after the timeout. -func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration) *Fragmentation { +func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, reassemblingTimeout time.Duration, clock tcpip.Clock) *Fragmentation { if lowMemoryLimit >= highMemoryLimit { lowMemoryLimit = highMemoryLimit } @@ -110,13 +110,17 @@ func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, rea blockSize = minBlockSize } - return &Fragmentation{ + f := &Fragmentation{ reassemblers: make(map[FragmentID]*reassembler), highLimit: highMemoryLimit, lowLimit: lowMemoryLimit, timeout: reassemblingTimeout, blockSize: blockSize, + clock: clock, } + f.releaseJob = tcpip.NewJob(f.clock, &f.mu, f.releaseReassemblersLocked) + + return f } // Process processes an incoming fragment belonging to an ID and returns a @@ -155,15 +159,17 @@ func (f *Fragmentation) Process( f.mu.Lock() r, ok := f.reassemblers[id] - if ok && r.tooOld(f.timeout) { - // This is very likely to be an id-collision or someone performing a slow-rate attack. - f.release(r) - ok = false - } if !ok { - r = newReassembler(id) + r = newReassembler(id, f.clock) f.reassemblers[id] = r + wasEmpty := f.rList.Empty() f.rList.PushFront(r) + if wasEmpty { + // If we have just pushed a first reassembler into an empty list, we + // should kickstart the release job. The release job will keep + // rescheduling itself until the list becomes empty. + f.releaseReassemblersLocked() + } } f.mu.Unlock() @@ -211,3 +217,102 @@ func (f *Fragmentation) release(r *reassembler) { f.size = 0 } } + +// releaseReassemblersLocked releases already-expired reassemblers, then +// schedules the job to call back itself for the remaining reassemblers if +// any. This function must be called with f.mu locked. +func (f *Fragmentation) releaseReassemblersLocked() { + now := f.clock.NowMonotonic() + for { + // The reassembler at the end of the list is the oldest. + r := f.rList.Back() + if r == nil { + // The list is empty. + break + } + elapsed := time.Duration(now-r.creationTime) * time.Nanosecond + if f.timeout > elapsed { + // If the oldest reassembler has not expired, schedule the release + // job so that this function is called back when it has expired. + f.releaseJob.Schedule(f.timeout - elapsed) + break + } + // If the oldest reassembler has already expired, release it. + f.release(r) + } +} + +// PacketFragmenter is the book-keeping struct for packet fragmentation. +type PacketFragmenter struct { + transportHeader buffer.View + data buffer.VectorisedView + reserve int + innerMTU int + fragmentCount int + currentFragment int + fragmentOffset int +} + +// MakePacketFragmenter prepares the struct needed for packet fragmentation. +// +// pkt is the packet to be fragmented. +// +// innerMTU is the maximum number of bytes of fragmentable data a fragment can +// have. +// +// reserve is the number of bytes that should be reserved for the headers in +// each generated fragment. +func MakePacketFragmenter(pkt *stack.PacketBuffer, innerMTU int, reserve int) PacketFragmenter { + // As per RFC 8200 Section 4.5, some IPv6 extension headers should not be + // repeated in each fragment. However we do not currently support any header + // of that kind yet, so the following computation is valid for both IPv4 and + // IPv6. + // TODO(gvisor.dev/issue/3912): Once Authentication or ESP Headers are + // supported for outbound packets, the fragmentable data should not include + // these headers. + var fragmentableData buffer.VectorisedView + fragmentableData.AppendView(pkt.TransportHeader().View()) + fragmentableData.Append(pkt.Data) + fragmentCount := (fragmentableData.Size() + innerMTU - 1) / innerMTU + + return PacketFragmenter{ + data: fragmentableData, + reserve: reserve, + innerMTU: innerMTU, + fragmentCount: fragmentCount, + } +} + +// BuildNextFragment returns a packet with the payload of the next fragment, +// along with the fragment's offset, the number of bytes copied and a boolean +// indicating if there are more fragments left or not. If this function is +// called again after it indicated that no more fragments were left, it will +// panic. +// +// Note that the returned packet will not have its network and link headers +// populated, but space for them will be reserved. The transport header will be +// stored in the packet's data. +func (pf *PacketFragmenter) BuildNextFragment() (*stack.PacketBuffer, int, int, bool) { + if pf.currentFragment >= pf.fragmentCount { + panic("BuildNextFragment should not be called again after the last fragment was returned") + } + + fragPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: pf.reserve, + }) + + // Copy data for the fragment. + copied := pf.data.ReadToVV(&fragPkt.Data, pf.innerMTU) + + offset := pf.fragmentOffset + pf.fragmentOffset += copied + pf.currentFragment++ + more := pf.currentFragment != pf.fragmentCount + + return fragPkt, offset, copied, more +} + +// RemainingFragmentCount returns the number of fragments left to be built. +func (pf *PacketFragmenter) RemainingFragmentCount() int { + return pf.fragmentCount - pf.currentFragment +} diff --git a/pkg/tcpip/network/fragmentation/fragmentation_test.go b/pkg/tcpip/network/fragmentation/fragmentation_test.go index 416604659..d3c7d7f92 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation_test.go +++ b/pkg/tcpip/network/fragmentation/fragmentation_test.go @@ -20,9 +20,16 @@ import ( "testing" "time" + "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/faketime" + "gvisor.dev/gvisor/pkg/tcpip/network/testutil" ) +// reassembleTimeout is dummy timeout used for testing, where the clock never +// advances. +const reassembleTimeout = 1 + // vv is a helper to build VectorisedView from different strings. func vv(size int, pieces ...string) buffer.VectorisedView { views := make([]buffer.View, len(pieces)) @@ -95,7 +102,7 @@ var processTestCases = []struct { func TestFragmentationProcess(t *testing.T) { for _, c := range processTestCases { t.Run(c.comment, func(t *testing.T) { - f := NewFragmentation(minBlockSize, 1024, 512, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 1024, 512, reassembleTimeout, &faketime.NullClock{}) firstFragmentProto := c.in[0].proto for i, in := range c.in { vv, proto, done, err := f.Process(in.id, in.first, in.last, in.more, in.proto, in.vv) @@ -131,25 +138,126 @@ func TestFragmentationProcess(t *testing.T) { } func TestReassemblingTimeout(t *testing.T) { - timeout := time.Millisecond - f := NewFragmentation(minBlockSize, 1024, 512, timeout) - // Send first fragment with id = 0, first = 0, last = 0, and more = true. - f.Process(FragmentID{}, 0, 0, true, 0xFF, vv(1, "0")) - // Sleep more than the timeout. - time.Sleep(2 * timeout) - // Send another fragment that completes a packet. - // However, no packet should be reassembled because the fragment arrived after the timeout. - _, _, done, err := f.Process(FragmentID{}, 1, 1, false, 0xFF, vv(1, "1")) - if err != nil { - t.Fatalf("f.Process(0, 1, 1, false, 0xFF, vv(1, \"1\")) failed: %v", err) + const ( + reassemblyTimeout = time.Millisecond + protocol = 0xff + ) + + type fragment struct { + first uint16 + last uint16 + more bool + data string } - if done { - t.Errorf("Fragmentation does not respect the reassembling timeout.") + + type event struct { + // name is a nickname of this event. + name string + + // clockAdvance is a duration to advance the clock. The clock advances + // before a fragment specified in the fragment field is processed. + clockAdvance time.Duration + + // fragment is a fragment to process. This can be nil if there is no + // fragment to process. + fragment *fragment + + // expectDone is true if the fragmentation instance should report the + // reassembly is done after the fragment is processd. + expectDone bool + + // sizeAfterEvent is the expected size of the fragmentation instance after + // the event. + sizeAfterEvent int + } + + half1 := &fragment{first: 0, last: 0, more: true, data: "0"} + half2 := &fragment{first: 1, last: 1, more: false, data: "1"} + + tests := []struct { + name string + events []event + }{ + { + name: "half1 and half2 are reassembled successfully", + events: []event{ + { + name: "half1", + fragment: half1, + expectDone: false, + sizeAfterEvent: 1, + }, + { + name: "half2", + fragment: half2, + expectDone: true, + sizeAfterEvent: 0, + }, + }, + }, + { + name: "half1 timeout, half2 timeout", + events: []event{ + { + name: "half1", + fragment: half1, + expectDone: false, + sizeAfterEvent: 1, + }, + { + name: "half1 just before reassembly timeout", + clockAdvance: reassemblyTimeout - 1, + sizeAfterEvent: 1, + }, + { + name: "half1 reassembly timeout", + clockAdvance: 1, + sizeAfterEvent: 0, + }, + { + name: "half2", + fragment: half2, + expectDone: false, + sizeAfterEvent: 1, + }, + { + name: "half2 just before reassembly timeout", + clockAdvance: reassemblyTimeout - 1, + sizeAfterEvent: 1, + }, + { + name: "half2 reassembly timeout", + clockAdvance: 1, + sizeAfterEvent: 0, + }, + }, + }, + } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + clock := faketime.NewManualClock() + f := NewFragmentation(minBlockSize, HighFragThreshold, LowFragThreshold, reassemblyTimeout, clock) + for _, event := range test.events { + clock.Advance(event.clockAdvance) + if frag := event.fragment; frag != nil { + _, _, done, err := f.Process(FragmentID{}, frag.first, frag.last, frag.more, protocol, vv(len(frag.data), frag.data)) + if err != nil { + t.Fatalf("%s: f.Process failed: %s", event.name, err) + } + if done != event.expectDone { + t.Fatalf("%s: got done = %t, want = %t", event.name, done, event.expectDone) + } + } + if got, want := f.size, event.sizeAfterEvent; got != want { + t.Errorf("%s: got f.size = %d, want = %d", event.name, got, want) + } + } + }) } } func TestMemoryLimits(t *testing.T) { - f := NewFragmentation(minBlockSize, 3, 1, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 3, 1, reassembleTimeout, &faketime.NullClock{}) // Send first fragment with id = 0. f.Process(FragmentID{ID: 0}, 0, 0, true, 0xFF, vv(1, "0")) // Send first fragment with id = 1. @@ -173,7 +281,7 @@ func TestMemoryLimits(t *testing.T) { } func TestMemoryLimitsIgnoresDuplicates(t *testing.T) { - f := NewFragmentation(minBlockSize, 1, 0, DefaultReassembleTimeout) + f := NewFragmentation(minBlockSize, 1, 0, reassembleTimeout, &faketime.NullClock{}) // Send first fragment with id = 0. f.Process(FragmentID{}, 0, 0, true, 0xFF, vv(1, "0")) // Send the same packet again. @@ -268,7 +376,7 @@ func TestErrors(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - f := NewFragmentation(test.blockSize, HighFragThreshold, LowFragThreshold, DefaultReassembleTimeout) + f := NewFragmentation(test.blockSize, HighFragThreshold, LowFragThreshold, reassembleTimeout, &faketime.NullClock{}) _, _, done, err := f.Process(FragmentID{}, test.first, test.last, test.more, 0, vv(len(test.data), test.data)) if !errors.Is(err, test.err) { t.Errorf("got Process(_, %d, %d, %t, _, %q) = (_, _, _, %v), want = (_, _, _, %v)", test.first, test.last, test.more, test.data, err, test.err) @@ -279,3 +387,113 @@ func TestErrors(t *testing.T) { }) } } + +type fragmentInfo struct { + remaining int + copied int + offset int + more bool +} + +func TestPacketFragmenter(t *testing.T) { + const ( + reserve = 60 + proto = 0 + ) + + tests := []struct { + name string + innerMTU int + transportHeaderLen int + payloadSize int + wantFragments []fragmentInfo + }{ + { + name: "Packet exactly fits in MTU", + innerMTU: 1280, + transportHeaderLen: 0, + payloadSize: 1280, + wantFragments: []fragmentInfo{ + {remaining: 0, copied: 1280, offset: 0, more: false}, + }, + }, + { + name: "Packet exactly does not fit in MTU", + innerMTU: 1000, + transportHeaderLen: 0, + payloadSize: 1001, + wantFragments: []fragmentInfo{ + {remaining: 1, copied: 1000, offset: 0, more: true}, + {remaining: 0, copied: 1, offset: 1000, more: false}, + }, + }, + { + name: "Packet has a transport header", + innerMTU: 560, + transportHeaderLen: 40, + payloadSize: 560, + wantFragments: []fragmentInfo{ + {remaining: 1, copied: 560, offset: 0, more: true}, + {remaining: 0, copied: 40, offset: 560, more: false}, + }, + }, + { + name: "Packet has a huge transport header", + innerMTU: 500, + transportHeaderLen: 1300, + payloadSize: 500, + wantFragments: []fragmentInfo{ + {remaining: 3, copied: 500, offset: 0, more: true}, + {remaining: 2, copied: 500, offset: 500, more: true}, + {remaining: 1, copied: 500, offset: 1000, more: true}, + {remaining: 0, copied: 300, offset: 1500, more: false}, + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + pkt := testutil.MakeRandPkt(test.transportHeaderLen, reserve, []int{test.payloadSize}, proto) + var originalPayload buffer.VectorisedView + originalPayload.AppendView(pkt.TransportHeader().View()) + originalPayload.Append(pkt.Data) + var reassembledPayload buffer.VectorisedView + pf := MakePacketFragmenter(pkt, test.innerMTU, reserve) + for i := 0; ; i++ { + fragPkt, offset, copied, more := pf.BuildNextFragment() + wantFragment := test.wantFragments[i] + if got := pf.RemainingFragmentCount(); got != wantFragment.remaining { + t.Errorf("(fragment #%d) got pf.RemainingFragmentCount() = %d, want = %d", i, got, wantFragment.remaining) + } + if copied != wantFragment.copied { + t.Errorf("(fragment #%d) got copied = %d, want = %d", i, copied, wantFragment.copied) + } + if offset != wantFragment.offset { + t.Errorf("(fragment #%d) got offset = %d, want = %d", i, offset, wantFragment.offset) + } + if more != wantFragment.more { + t.Errorf("(fragment #%d) got more = %t, want = %t", i, more, wantFragment.more) + } + if got := fragPkt.Size(); got > test.innerMTU { + t.Errorf("(fragment #%d) got fragPkt.Size() = %d, want <= %d", i, got, test.innerMTU) + } + if got := fragPkt.AvailableHeaderBytes(); got != reserve { + t.Errorf("(fragment #%d) got fragPkt.AvailableHeaderBytes() = %d, want = %d", i, got, reserve) + } + if got := fragPkt.TransportHeader().View().Size(); got != 0 { + t.Errorf("(fragment #%d) got fragPkt.TransportHeader().View().Size() = %d, want = 0", i, got) + } + reassembledPayload.Append(fragPkt.Data) + if !more { + if i != len(test.wantFragments)-1 { + t.Errorf("got fragment count = %d, want = %d", i, len(test.wantFragments)-1) + } + break + } + } + if diff := cmp.Diff(reassembledPayload.ToView(), originalPayload.ToView()); diff != "" { + t.Errorf("reassembledPayload mismatch (-want +got):\n%s", diff) + } + }) + } +} diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go index f044867dc..9bb051a30 100644 --- a/pkg/tcpip/network/fragmentation/reassembler.go +++ b/pkg/tcpip/network/fragmentation/reassembler.go @@ -18,9 +18,9 @@ import ( "container/heap" "fmt" "math" - "time" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) @@ -40,15 +40,15 @@ type reassembler struct { deleted int heap fragHeap done bool - creationTime time.Time + creationTime int64 } -func newReassembler(id FragmentID) *reassembler { +func newReassembler(id FragmentID, clock tcpip.Clock) *reassembler { r := &reassembler{ id: id, holes: make([]hole, 0, 16), heap: make(fragHeap, 0, 8), - creationTime: time.Now(), + creationTime: clock.NowMonotonic(), } r.holes = append(r.holes, hole{ first: 0, @@ -116,10 +116,6 @@ func (r *reassembler) process(first, last uint16, more bool, proto uint8, vv buf return res, r.proto, true, consumed, nil } -func (r *reassembler) tooOld(timeout time.Duration) bool { - return time.Now().Sub(r.creationTime) > timeout -} - func (r *reassembler) checkDoneOrMark() bool { r.mu.Lock() prev := r.done diff --git a/pkg/tcpip/network/fragmentation/reassembler_test.go b/pkg/tcpip/network/fragmentation/reassembler_test.go index dff7c9dcb..a0a04a027 100644 --- a/pkg/tcpip/network/fragmentation/reassembler_test.go +++ b/pkg/tcpip/network/fragmentation/reassembler_test.go @@ -18,6 +18,8 @@ import ( "math" "reflect" "testing" + + "gvisor.dev/gvisor/pkg/tcpip/faketime" ) type updateHolesInput struct { @@ -94,7 +96,7 @@ var holesTestCases = []struct { func TestUpdateHoles(t *testing.T) { for _, c := range holesTestCases { - r := newReassembler(FragmentID{}) + r := newReassembler(FragmentID{}, &faketime.NullClock{}) for _, i := range c.in { r.updateHoles(i.first, i.last, i.more) } diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index e45dd17f8..d436873b6 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -17,6 +17,7 @@ package ip_test import ( "testing" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -25,26 +26,35 @@ 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/tcp" "gvisor.dev/gvisor/pkg/tcpip/transport/udp" ) const ( - localIpv4Addr = "\x0a\x00\x00\x01" - localIpv4PrefixLen = 24 - remoteIpv4Addr = "\x0a\x00\x00\x02" - ipv4SubnetAddr = "\x0a\x00\x00\x00" - ipv4SubnetMask = "\xff\xff\xff\x00" - ipv4Gateway = "\x0a\x00\x00\x03" - localIpv6Addr = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" - localIpv6PrefixLen = 120 - remoteIpv6Addr = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" - ipv6SubnetAddr = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" - ipv6SubnetMask = "\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\x00" - ipv6Gateway = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03" - nicID = 1 + localIPv4Addr = "\x0a\x00\x00\x01" + remoteIPv4Addr = "\x0a\x00\x00\x02" + ipv4SubnetAddr = "\x0a\x00\x00\x00" + ipv4SubnetMask = "\xff\xff\xff\x00" + ipv4Gateway = "\x0a\x00\x00\x03" + localIPv6Addr = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" + remoteIPv6Addr = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" + ipv6SubnetAddr = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" + ipv6SubnetMask = "\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\x00" + ipv6Gateway = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03" + nicID = 1 ) +var localIPv4AddrWithPrefix = tcpip.AddressWithPrefix{ + Address: localIPv4Addr, + PrefixLen: 24, +} + +var localIPv6AddrWithPrefix = tcpip.AddressWithPrefix{ + Address: localIPv6Addr, + PrefixLen: 120, +} + // testObject implements two interfaces: LinkEndpoint and TransportDispatcher. // The former is used to pretend that it's a link endpoint so that we can // inspect packets written by the network endpoints. The latter is used to @@ -98,9 +108,10 @@ func (t *testObject) checkValues(protocol tcpip.TransportProtocolNumber, vv buff // DeliverTransportPacket is called by network endpoints after parsing incoming // packets. This is used by the test object to verify that the results of the // parsing are expected. -func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt *stack.PacketBuffer) { +func (t *testObject) DeliverTransportPacket(r *stack.Route, protocol tcpip.TransportProtocolNumber, pkt *stack.PacketBuffer) stack.TransportPacketDisposition { t.checkValues(protocol, pkt.Data, r.RemoteAddress, r.LocalAddress) t.dataCalls++ + return stack.TransportPacketHandled } // DeliverTransportControlPacket is called by network endpoints after parsing @@ -194,8 +205,8 @@ func (*testObject) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.Net func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, tcp.NewProtocol}, }) s.CreateNIC(nicID, loopback.New()) s.AddAddress(nicID, ipv4.ProtocolNumber, local) @@ -210,8 +221,8 @@ func buildIPv4Route(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.NetworkProtocol{ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, tcp.NewProtocol}, }) s.CreateNIC(nicID, loopback.New()) s.AddAddress(nicID, ipv6.ProtocolNumber, local) @@ -224,33 +235,290 @@ func buildIPv6Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { return s.FindRoute(nicID, local, remote, ipv6.ProtocolNumber, false /* multicastLoop */) } -func buildDummyStack(t *testing.T) *stack.Stack { +func buildDummyStackWithLinkEndpoint(t *testing.T) (*stack.Stack, *channel.Endpoint) { t.Helper() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol, tcp.NewProtocol}, }) e := channel.New(0, 1280, "") if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } - if err := s.AddAddress(nicID, header.IPv4ProtocolNumber, localIpv4Addr); err != nil { - t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv4ProtocolNumber, localIpv4Addr, err) + v4Addr := tcpip.ProtocolAddress{Protocol: header.IPv4ProtocolNumber, AddressWithPrefix: localIPv4AddrWithPrefix} + if err := s.AddProtocolAddress(nicID, v4Addr); err != nil { + t.Fatalf("AddProtocolAddress(%d, %#v) = %s", nicID, v4Addr, err) } - if err := s.AddAddress(nicID, header.IPv6ProtocolNumber, localIpv6Addr); err != nil { - t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, header.IPv6ProtocolNumber, localIpv6Addr, err) + v6Addr := tcpip.ProtocolAddress{Protocol: header.IPv6ProtocolNumber, AddressWithPrefix: localIPv6AddrWithPrefix} + if err := s.AddProtocolAddress(nicID, v6Addr); err != nil { + t.Fatalf("AddProtocolAddress(%d, %#v) = %s", nicID, v6Addr, err) } + return s, e +} + +func buildDummyStack(t *testing.T) *stack.Stack { + t.Helper() + + s, _ := buildDummyStackWithLinkEndpoint(t) return s } +var _ stack.NetworkInterface = (*testInterface)(nil) + +type testInterface struct { + testObject + + mu struct { + sync.RWMutex + disabled bool + } +} + +func (*testInterface) ID() tcpip.NICID { + return nicID +} + +func (*testInterface) IsLoopback() bool { + return false +} + +func (*testInterface) Name() string { + return "" +} + +func (t *testInterface) Enabled() bool { + t.mu.RLock() + defer t.mu.RUnlock() + return !t.mu.disabled +} + +func (t *testInterface) setEnabled(v bool) { + t.mu.Lock() + defer t.mu.Unlock() + t.mu.disabled = !v +} + +func TestSourceAddressValidation(t *testing.T) { + rxIPv4ICMP := func(e *channel.Endpoint, src tcpip.Address) { + totalLen := header.IPv4MinimumSize + header.ICMPv4MinimumSize + hdr := buffer.NewPrependable(totalLen) + pkt := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) + pkt.SetType(header.ICMPv4Echo) + pkt.SetCode(0) + pkt.SetChecksum(0) + pkt.SetChecksum(^header.Checksum(pkt, 0)) + ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize)) + ip.Encode(&header.IPv4Fields{ + IHL: header.IPv4MinimumSize, + TotalLength: uint16(totalLen), + Protocol: uint8(icmp.ProtocolNumber4), + TTL: ipv4.DefaultTTL, + SrcAddr: src, + DstAddr: localIPv4Addr, + }) + + e.InjectInbound(header.IPv4ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + })) + } + + rxIPv6ICMP := func(e *channel.Endpoint, src tcpip.Address) { + totalLen := header.IPv6MinimumSize + header.ICMPv6MinimumSize + hdr := buffer.NewPrependable(totalLen) + pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6MinimumSize)) + pkt.SetType(header.ICMPv6EchoRequest) + pkt.SetCode(0) + pkt.SetChecksum(0) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, src, localIPv6Addr, buffer.VectorisedView{})) + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: header.ICMPv6MinimumSize, + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: ipv6.DefaultTTL, + SrcAddr: src, + DstAddr: localIPv6Addr, + }) + + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + })) + } + + tests := []struct { + name string + srcAddress tcpip.Address + rxICMP func(*channel.Endpoint, tcpip.Address) + valid bool + }{ + { + name: "IPv4 valid", + srcAddress: "\x01\x02\x03\x04", + rxICMP: rxIPv4ICMP, + valid: true, + }, + { + name: "IPv6 valid", + srcAddress: "\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f\x10", + rxICMP: rxIPv6ICMP, + valid: true, + }, + { + name: "IPv4 unspecified", + srcAddress: header.IPv4Any, + rxICMP: rxIPv4ICMP, + valid: true, + }, + { + name: "IPv6 unspecified", + srcAddress: header.IPv4Any, + rxICMP: rxIPv6ICMP, + valid: true, + }, + { + name: "IPv4 multicast", + srcAddress: "\xe0\x00\x00\x01", + rxICMP: rxIPv4ICMP, + valid: false, + }, + { + name: "IPv6 multicast", + srcAddress: "\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01", + rxICMP: rxIPv6ICMP, + valid: false, + }, + { + name: "IPv4 broadcast", + srcAddress: header.IPv4Broadcast, + rxICMP: rxIPv4ICMP, + valid: false, + }, + { + name: "IPv4 subnet broadcast", + srcAddress: func() tcpip.Address { + subnet := localIPv4AddrWithPrefix.Subnet() + return subnet.Broadcast() + }(), + rxICMP: rxIPv4ICMP, + valid: false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + s, e := buildDummyStackWithLinkEndpoint(t) + test.rxICMP(e, test.srcAddress) + + var wantValid uint64 + if test.valid { + wantValid = 1 + } + + if got, want := s.Stats().IP.InvalidSourceAddressesReceived.Value(), 1-wantValid; got != want { + t.Errorf("got s.Stats().IP.InvalidSourceAddressesReceived.Value() = %d, want = %d", got, want) + } + if got := s.Stats().IP.PacketsDelivered.Value(); got != wantValid { + t.Errorf("got s.Stats().IP.PacketsDelivered.Value() = %d, want = %d", got, wantValid) + } + }) + } +} + +func TestEnableWhenNICDisabled(t *testing.T) { + tests := []struct { + name string + protocolFactory stack.NetworkProtocolFactory + protoNum tcpip.NetworkProtocolNumber + }{ + { + name: "IPv4", + protocolFactory: ipv4.NewProtocol, + protoNum: ipv4.ProtocolNumber, + }, + { + name: "IPv6", + protocolFactory: ipv6.NewProtocol, + protoNum: ipv6.ProtocolNumber, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + var nic testInterface + nic.setEnabled(false) + + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{test.protocolFactory}, + }) + p := s.NetworkProtocolInstance(test.protoNum) + + // We pass nil for all parameters except the NetworkInterface and Stack + // since Enable only depends on these. + ep := p.NewEndpoint(&nic, nil, nil, nil) + + // The endpoint should initially be disabled, regardless the NIC's enabled + // status. + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + nic.setEnabled(true) + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + + // Attempting to enable the endpoint while the NIC is disabled should + // fail. + nic.setEnabled(false) + if err := ep.Enable(); err != tcpip.ErrNotPermitted { + t.Fatalf("got ep.Enable() = %s, want = %s", err, tcpip.ErrNotPermitted) + } + // ep should consider the NIC's enabled status when determining its own + // enabled status so we "enable" the NIC to read just the endpoint's + // enabled status. + nic.setEnabled(true) + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + + // Enabling the interface after the NIC has been enabled should succeed. + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + if !ep.Enabled() { + t.Fatal("got ep.Enabled() = false, want = true") + } + + // ep should consider the NIC's enabled status when determining its own + // enabled status. + nic.setEnabled(false) + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + + // Disabling the endpoint when the NIC is enabled should make the endpoint + // disabled. + nic.setEnabled(true) + ep.Disable() + if ep.Enabled() { + t.Fatal("got ep.Enabled() = true, want = false") + } + }) + } +} + func TestIPv4Send(t *testing.T) { - o := testObject{t: t, v4: true} - proto := ipv4.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, nil, &o, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv4.ProtocolNumber) + nic := testInterface{ + testObject: testObject{ + t: t, + v4: true, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, nil) defer ep.Close() // Allocate and initialize the payload view. @@ -266,12 +534,12 @@ func TestIPv4Send(t *testing.T) { }) // Issue the write. - o.protocol = 123 - o.srcAddr = localIpv4Addr - o.dstAddr = remoteIpv4Addr - o.contents = payload + nic.testObject.protocol = 123 + nic.testObject.srcAddr = localIPv4Addr + nic.testObject.dstAddr = remoteIPv4Addr + nic.testObject.contents = payload - r, err := buildIPv4Route(localIpv4Addr, remoteIpv4Addr) + r, err := buildIPv4Route(localIPv4Addr, remoteIPv4Addr) if err != nil { t.Fatalf("could not find route: %v", err) } @@ -285,11 +553,21 @@ func TestIPv4Send(t *testing.T) { } func TestIPv4Receive(t *testing.T) { - o := testObject{t: t, v4: true} - proto := ipv4.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv4.ProtocolNumber) + nic := testInterface{ + testObject: testObject{ + t: t, + v4: true, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.testObject) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + totalLen := header.IPv4MinimumSize + 30 view := buffer.NewView(totalLen) ip := header.IPv4(view) @@ -298,8 +576,8 @@ func TestIPv4Receive(t *testing.T) { TotalLength: uint16(totalLen), TTL: 20, Protocol: 10, - SrcAddr: remoteIpv4Addr, - DstAddr: localIpv4Addr, + SrcAddr: remoteIPv4Addr, + DstAddr: localIPv4Addr, }) // Make payload be non-zero. @@ -308,12 +586,12 @@ func TestIPv4Receive(t *testing.T) { } // Give packet to ipv4 endpoint, dispatcher will validate that it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv4Addr - o.dstAddr = localIpv4Addr - o.contents = view[header.IPv4MinimumSize:totalLen] + nic.testObject.protocol = 10 + nic.testObject.srcAddr = remoteIPv4Addr + nic.testObject.dstAddr = localIPv4Addr + nic.testObject.contents = view[header.IPv4MinimumSize:totalLen] - r, err := buildIPv4Route(localIpv4Addr, remoteIpv4Addr) + r, err := buildIPv4Route(localIPv4Addr, remoteIPv4Addr) if err != nil { t.Fatalf("could not find route: %v", err) } @@ -324,8 +602,8 @@ func TestIPv4Receive(t *testing.T) { t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) } ep.HandlePacket(&r, pkt) - if o.dataCalls != 1 { - t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) + if nic.testObject.dataCalls != 1 { + t.Fatalf("Bad number of data calls: got %x, want 1", nic.testObject.dataCalls) } } @@ -349,17 +627,26 @@ func TestIPv4ReceiveControl(t *testing.T) { {"Non-zero fragment offset", 0, 100, header.ICMPv4PortUnreachable, stack.ControlPortUnreachable, 0, 0}, {"Zero-length packet", 0, 0, header.ICMPv4PortUnreachable, stack.ControlPortUnreachable, 0, 2*header.IPv4MinimumSize + header.ICMPv4MinimumSize + 8}, } - r, err := buildIPv4Route(localIpv4Addr, "\x0a\x00\x00\xbb") + r, err := buildIPv4Route(localIPv4Addr, "\x0a\x00\x00\xbb") if err != nil { t.Fatal(err) } for _, c := range cases { t.Run(c.name, func(t *testing.T) { - o := testObject{t: t} - proto := ipv4.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv4.ProtocolNumber) + nic := testInterface{ + testObject: testObject{ + t: t, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.testObject) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + const dataOffset = header.IPv4MinimumSize*2 + header.ICMPv4MinimumSize view := buffer.NewView(dataOffset + 8) @@ -371,7 +658,7 @@ func TestIPv4ReceiveControl(t *testing.T) { TTL: 20, Protocol: uint8(header.ICMPv4ProtocolNumber), SrcAddr: "\x0a\x00\x00\xbb", - DstAddr: localIpv4Addr, + DstAddr: localIPv4Addr, }) // Create the ICMP header. @@ -389,8 +676,8 @@ func TestIPv4ReceiveControl(t *testing.T) { TTL: 20, Protocol: 10, FragmentOffset: c.fragmentOffset, - SrcAddr: localIpv4Addr, - DstAddr: remoteIpv4Addr, + SrcAddr: localIPv4Addr, + DstAddr: remoteIPv4Addr, }) // Make payload be non-zero. @@ -400,27 +687,37 @@ func TestIPv4ReceiveControl(t *testing.T) { // Give packet to IPv4 endpoint, dispatcher will validate that // it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv4Addr - o.dstAddr = localIpv4Addr - o.contents = view[dataOffset:] - o.typ = c.expectedTyp - o.extra = c.expectedExtra + nic.testObject.protocol = 10 + nic.testObject.srcAddr = remoteIPv4Addr + nic.testObject.dstAddr = localIPv4Addr + nic.testObject.contents = view[dataOffset:] + nic.testObject.typ = c.expectedTyp + nic.testObject.extra = c.expectedExtra ep.HandlePacket(&r, truncatedPacket(view, c.trunc, header.IPv4MinimumSize)) - if want := c.expectedCount; o.controlCalls != want { - t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, o.controlCalls, want) + if want := c.expectedCount; nic.testObject.controlCalls != want { + t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, nic.testObject.controlCalls, want) } }) } } func TestIPv4FragmentationReceive(t *testing.T) { - o := testObject{t: t, v4: true} - proto := ipv4.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv4.ProtocolNumber) + nic := testInterface{ + testObject: testObject{ + t: t, + v4: true, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.testObject) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + totalLen := header.IPv4MinimumSize + 24 frag1 := buffer.NewView(totalLen) @@ -432,8 +729,8 @@ func TestIPv4FragmentationReceive(t *testing.T) { Protocol: 10, FragmentOffset: 0, Flags: header.IPv4FlagMoreFragments, - SrcAddr: remoteIpv4Addr, - DstAddr: localIpv4Addr, + SrcAddr: remoteIPv4Addr, + DstAddr: localIPv4Addr, }) // Make payload be non-zero. for i := header.IPv4MinimumSize; i < totalLen; i++ { @@ -448,8 +745,8 @@ func TestIPv4FragmentationReceive(t *testing.T) { TTL: 20, Protocol: 10, FragmentOffset: 24, - SrcAddr: remoteIpv4Addr, - DstAddr: localIpv4Addr, + SrcAddr: remoteIPv4Addr, + DstAddr: localIPv4Addr, }) // Make payload be non-zero. for i := header.IPv4MinimumSize; i < totalLen; i++ { @@ -457,12 +754,12 @@ func TestIPv4FragmentationReceive(t *testing.T) { } // Give packet to ipv4 endpoint, dispatcher will validate that it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv4Addr - o.dstAddr = localIpv4Addr - o.contents = append(frag1[header.IPv4MinimumSize:totalLen], frag2[header.IPv4MinimumSize:totalLen]...) + nic.testObject.protocol = 10 + nic.testObject.srcAddr = remoteIPv4Addr + nic.testObject.dstAddr = localIPv4Addr + nic.testObject.contents = append(frag1[header.IPv4MinimumSize:totalLen], frag2[header.IPv4MinimumSize:totalLen]...) - r, err := buildIPv4Route(localIpv4Addr, remoteIpv4Addr) + r, err := buildIPv4Route(localIPv4Addr, remoteIPv4Addr) if err != nil { t.Fatalf("could not find route: %v", err) } @@ -475,8 +772,8 @@ func TestIPv4FragmentationReceive(t *testing.T) { t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) } ep.HandlePacket(&r, pkt) - if o.dataCalls != 0 { - t.Fatalf("Bad number of data calls: got %x, want 0", o.dataCalls) + if nic.testObject.dataCalls != 0 { + t.Fatalf("Bad number of data calls: got %x, want 0", nic.testObject.dataCalls) } // Send second segment. @@ -487,17 +784,26 @@ func TestIPv4FragmentationReceive(t *testing.T) { t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) } ep.HandlePacket(&r, pkt) - if o.dataCalls != 1 { - t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) + if nic.testObject.dataCalls != 1 { + t.Fatalf("Bad number of data calls: got %x, want 1", nic.testObject.dataCalls) } } func TestIPv6Send(t *testing.T) { - o := testObject{t: t} - proto := ipv6.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, channel.New(0, 1280, ""), buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv6.ProtocolNumber) + nic := testInterface{ + testObject: testObject{ + t: t, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, nil) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + // Allocate and initialize the payload view. payload := buffer.NewView(100) for i := 0; i < len(payload); i++ { @@ -511,12 +817,12 @@ func TestIPv6Send(t *testing.T) { }) // Issue the write. - o.protocol = 123 - o.srcAddr = localIpv6Addr - o.dstAddr = remoteIpv6Addr - o.contents = payload + nic.testObject.protocol = 123 + nic.testObject.srcAddr = localIPv6Addr + nic.testObject.dstAddr = remoteIPv6Addr + nic.testObject.contents = payload - r, err := buildIPv6Route(localIpv6Addr, remoteIpv6Addr) + r, err := buildIPv6Route(localIPv6Addr, remoteIPv6Addr) if err != nil { t.Fatalf("could not find route: %v", err) } @@ -530,11 +836,20 @@ func TestIPv6Send(t *testing.T) { } func TestIPv6Receive(t *testing.T) { - o := testObject{t: t} - proto := ipv6.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv6.ProtocolNumber) + nic := testInterface{ + testObject: testObject{ + t: t, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.testObject) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + totalLen := header.IPv6MinimumSize + 30 view := buffer.NewView(totalLen) ip := header.IPv6(view) @@ -542,8 +857,8 @@ func TestIPv6Receive(t *testing.T) { PayloadLength: uint16(totalLen - header.IPv6MinimumSize), NextHeader: 10, HopLimit: 20, - SrcAddr: remoteIpv6Addr, - DstAddr: localIpv6Addr, + SrcAddr: remoteIPv6Addr, + DstAddr: localIPv6Addr, }) // Make payload be non-zero. @@ -552,12 +867,12 @@ func TestIPv6Receive(t *testing.T) { } // Give packet to ipv6 endpoint, dispatcher will validate that it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv6Addr - o.dstAddr = localIpv6Addr - o.contents = view[header.IPv6MinimumSize:totalLen] + nic.testObject.protocol = 10 + nic.testObject.srcAddr = remoteIPv6Addr + nic.testObject.dstAddr = localIPv6Addr + nic.testObject.contents = view[header.IPv6MinimumSize:totalLen] - r, err := buildIPv6Route(localIpv6Addr, remoteIpv6Addr) + r, err := buildIPv6Route(localIPv6Addr, remoteIPv6Addr) if err != nil { t.Fatalf("could not find route: %v", err) } @@ -569,8 +884,8 @@ func TestIPv6Receive(t *testing.T) { t.Fatalf("failed to parse packet: %x", pkt.Data.ToView()) } ep.HandlePacket(&r, pkt) - if o.dataCalls != 1 { - t.Fatalf("Bad number of data calls: got %x, want 1", o.dataCalls) + if nic.testObject.dataCalls != 1 { + t.Fatalf("Bad number of data calls: got %x, want 1", nic.testObject.dataCalls) } } @@ -601,7 +916,7 @@ func TestIPv6ReceiveControl(t *testing.T) { {"Zero-length packet", 0, nil, header.ICMPv6DstUnreachable, header.ICMPv6PortUnreachable, stack.ControlPortUnreachable, 0, 2*header.IPv6MinimumSize + header.ICMPv6DstUnreachableMinimumSize + 8}, } r, err := buildIPv6Route( - localIpv6Addr, + localIPv6Addr, "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xaa", ) if err != nil { @@ -609,11 +924,20 @@ func TestIPv6ReceiveControl(t *testing.T) { } for _, c := range cases { t.Run(c.name, func(t *testing.T) { - o := testObject{t: t} - proto := ipv6.NewProtocol() - ep := proto.NewEndpoint(nicID, nil, nil, &o, nil, buildDummyStack(t)) + s := buildDummyStack(t) + proto := s.NetworkProtocolInstance(ipv6.ProtocolNumber) + nic := testInterface{ + testObject: testObject{ + t: t, + }, + } + ep := proto.NewEndpoint(&nic, nil, nil, &nic.testObject) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + dataOffset := header.IPv6MinimumSize*2 + header.ICMPv6MinimumSize if c.fragmentOffset != nil { dataOffset += header.IPv6FragmentHeaderSize @@ -627,7 +951,7 @@ func TestIPv6ReceiveControl(t *testing.T) { NextHeader: uint8(header.ICMPv6ProtocolNumber), HopLimit: 20, SrcAddr: outerSrcAddr, - DstAddr: localIpv6Addr, + DstAddr: localIPv6Addr, }) // Create the ICMP header. @@ -643,8 +967,8 @@ func TestIPv6ReceiveControl(t *testing.T) { PayloadLength: 100, NextHeader: 10, HopLimit: 20, - SrcAddr: localIpv6Addr, - DstAddr: remoteIpv6Addr, + SrcAddr: localIPv6Addr, + DstAddr: remoteIPv6Addr, }) // Build the fragmentation header if needed. @@ -666,19 +990,19 @@ func TestIPv6ReceiveControl(t *testing.T) { // Give packet to IPv6 endpoint, dispatcher will validate that // it's ok. - o.protocol = 10 - o.srcAddr = remoteIpv6Addr - o.dstAddr = localIpv6Addr - o.contents = view[dataOffset:] - o.typ = c.expectedTyp - o.extra = c.expectedExtra + nic.testObject.protocol = 10 + nic.testObject.srcAddr = remoteIPv6Addr + nic.testObject.dstAddr = localIPv6Addr + nic.testObject.contents = view[dataOffset:] + nic.testObject.typ = c.expectedTyp + nic.testObject.extra = c.expectedExtra // Set ICMPv6 checksum. - icmp.SetChecksum(header.ICMPv6Checksum(icmp, outerSrcAddr, localIpv6Addr, buffer.VectorisedView{})) + icmp.SetChecksum(header.ICMPv6Checksum(icmp, outerSrcAddr, localIPv6Addr, buffer.VectorisedView{})) ep.HandlePacket(&r, truncatedPacket(view, c.trunc, header.IPv6MinimumSize)) - if want := c.expectedCount; o.controlCalls != want { - t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, o.controlCalls, want) + if want := c.expectedCount; nic.testObject.controlCalls != want { + t.Fatalf("Bad number of control calls for %q case: got %v, want %v", c.name, nic.testObject.controlCalls, want) } }) } diff --git a/pkg/tcpip/network/ipv4/BUILD b/pkg/tcpip/network/ipv4/BUILD index f9c2aa980..7fc12e229 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -10,6 +10,7 @@ go_library( ], visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", @@ -27,12 +28,15 @@ go_test( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/checker", "//pkg/tcpip/header", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/sniffer", + "//pkg/tcpip/network/arp", "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/testutil", "//pkg/tcpip/stack", + "//pkg/tcpip/transport/icmp", "//pkg/tcpip/transport/tcp", "//pkg/tcpip/transport/udp", "//pkg/waiter", diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index b5659a36b..3407755ed 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -15,6 +15,9 @@ package ipv4 import ( + "fmt" + + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -39,7 +42,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack // Drop packet if it doesn't have the basic IPv4 header or if the // original source address doesn't match an address we own. src := hdr.SourceAddress() - if e.stack.CheckLocalAddress(e.NICID(), ProtocolNumber, src) == 0 { + if e.protocol.stack.CheckLocalAddress(e.nic.ID(), ProtocolNumber, src) == 0 { return } @@ -76,69 +79,87 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { received.Echo.Increment() // Only send a reply if the checksum is valid. - wantChecksum := h.Checksum() - // Reset the checksum field to 0 to can calculate the proper - // checksum. We'll have to reset this before we hand the packet - // off. + headerChecksum := h.Checksum() h.SetChecksum(0) - gotChecksum := ^header.ChecksumVV(pkt.Data, 0 /* initial */) - if gotChecksum != wantChecksum { - // It's possible that a raw socket expects to receive this. - h.SetChecksum(wantChecksum) + calculatedChecksum := ^header.ChecksumVV(pkt.Data, 0 /* initial */) + h.SetChecksum(headerChecksum) + if calculatedChecksum != headerChecksum { + // It's possible that a raw socket still expects to receive this. e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, pkt) received.Invalid.Increment() return } - // Make a copy of data before pkt gets sent to raw socket. - // DeliverTransportPacket will take ownership of pkt. - replyData := pkt.Data.Clone(nil) - replyData.TrimFront(header.ICMPv4MinimumSize) + // DeliverTransportPacket will take ownership of pkt so don't use it beyond + // this point. Make a deep copy of the data before pkt gets sent as we will + // be modifying fields. + // + // TODO(gvisor.dev/issue/4399): The copy may not be needed if there are no + // waiting endpoints. Consider moving responsibility for doing the copy to + // DeliverTransportPacket so that is is only done when needed. + replyData := pkt.Data.ToOwnedView() + replyIPHdr := header.IPv4(append(buffer.View(nil), pkt.NetworkHeader().View()...)) - // It's possible that a raw socket expects to receive this. - h.SetChecksum(wantChecksum) e.dispatcher.DeliverTransportPacket(r, header.ICMPv4ProtocolNumber, pkt) - remoteLinkAddr := r.RemoteLinkAddress - // 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). localAddr := r.LocalAddress - if r.IsInboundBroadcast() || header.IsV4MulticastAddress(r.LocalAddress) { + if r.IsInboundBroadcast() || header.IsV4MulticastAddress(localAddr) { localAddr = "" } - r, err := r.Stack().FindRoute(e.NICID(), localAddr, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) + r, err := r.Stack().FindRoute(e.nic.ID(), localAddr, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) if err != nil { // If we cannot find a route to the destination, silently drop the packet. return } defer r.Release() - // Use the remote link address from the incoming packet. - r.ResolveWith(remoteLinkAddr) - - // Prepare a reply packet. - icmpHdr := make(header.ICMPv4, header.ICMPv4MinimumSize) - copy(icmpHdr, h) - icmpHdr.SetType(header.ICMPv4EchoReply) - icmpHdr.SetChecksum(0) - icmpHdr.SetChecksum(^header.Checksum(icmpHdr, header.ChecksumVV(replyData, 0))) - dataVV := buffer.View(icmpHdr).ToVectorisedView() - dataVV.Append(replyData) + // TODO(gvisor.dev/issue/3810:) When adding protocol numbers into the + // header information, we may have to change this code to handle the + // ICMP header no longer being in the data buffer. + + // Because IP and ICMP are so closely intertwined, we need to handcraft our + // IP header to be able to follow RFC 792. The wording on page 13 is as + // follows: + // IP Fields: + // Addresses + // The address of the source in an echo message will be the + // destination of the echo reply message. To form an echo reply + // message, the source and destination addresses are simply reversed, + // the type code changed to 0, and the checksum recomputed. + // + // This was interpreted by early implementors to mean that all options must + // be copied from the echo request IP header to the echo reply IP header + // and this behaviour is still relied upon by some applications. + // + // Create a copy of the IP header we received, options and all, and change + // The fields we need to alter. + // + // We need to produce the entire packet in the data segment in order to + // use WriteHeaderIncludedPacket(). + replyIPHdr.SetSourceAddress(r.LocalAddress) + replyIPHdr.SetDestinationAddress(r.RemoteAddress) + replyIPHdr.SetTTL(r.DefaultTTL()) + + replyICMPHdr := header.ICMPv4(replyData) + replyICMPHdr.SetType(header.ICMPv4EchoReply) + replyICMPHdr.SetChecksum(0) + replyICMPHdr.SetChecksum(^header.Checksum(replyData, 0)) + + replyVV := buffer.View(replyIPHdr).ToVectorisedView() + replyVV.AppendView(replyData) replyPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(r.MaxHeaderLength()), - Data: dataVV, + Data: replyVV, }) + replyPkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber - // Send out the reply packet. + // The checksum will be calculated so we don't need to do it here. sent := stats.ICMP.V4PacketsSent - if err := r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{ - Protocol: header.ICMPv4ProtocolNumber, - TTL: r.DefaultTTL(), - TOS: stack.DefaultTOS, - }, replyPkt); err != nil { + if err := r.WriteHeaderIncludedPacket(replyPkt); err != nil { sent.Dropped.Increment() return } @@ -193,3 +214,190 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer) { received.Invalid.Increment() } } + +// ======= ICMP Error packet generation ========= + +// icmpReason is a marker interface for IPv4 specific ICMP errors. +type icmpReason interface { + isICMPReason() +} + +// icmpReasonPortUnreachable is an error where the transport protocol has no +// listener and no alternative means to inform the sender. +type icmpReasonPortUnreachable struct{} + +func (*icmpReasonPortUnreachable) isICMPReason() {} + +// icmpReasonProtoUnreachable is an error where the transport protocol is +// not supported. +type icmpReasonProtoUnreachable struct{} + +func (*icmpReasonProtoUnreachable) isICMPReason() {} + +// returnError takes an error descriptor and generates the appropriate ICMP +// error packet for IPv4 and sends it back to the remote device that sent +// the problematic packet. It incorporates as much of that packet as +// possible as well as any error metadata as is available. returnError +// expects pkt to hold a valid IPv4 packet as per the wire format. +func (p *protocol) returnError(r *stack.Route, reason icmpReason, pkt *stack.PacketBuffer) *tcpip.Error { + // We check we are responding only when we are allowed to. + // See RFC 1812 section 4.3.2.7 (shown below). + // + // ========= + // 4.3.2.7 When Not to Send ICMP Errors + // + // An ICMP error message MUST NOT be sent as the result of receiving: + // + // o An ICMP error message, or + // + // o A packet which fails the IP header validation tests described in + // Section [5.2.2] (except where that section specifically permits + // the sending of an ICMP error message), or + // + // o A packet destined to an IP broadcast or IP multicast address, or + // + // o A packet sent as a Link Layer broadcast or multicast, or + // + // o Any fragment of a datagram other then the first fragment (i.e., a + // packet for which the fragment offset in the IP header is nonzero). + // + // TODO(gvisor.dev/issues/4058): Make sure we don't send ICMP errors in + // response to a non-initial fragment, but it currently can not happen. + + if r.IsInboundBroadcast() || header.IsV4MulticastAddress(r.LocalAddress) || r.RemoteAddress == header.IPv4Any { + return nil + } + + // Even if we were able to receive a packet from some remote, we may not have + // a route to it - the remote may be blocked via routing rules. We must always + // consult our routing table and find a route to the remote before sending any + // packet. + route, err := p.stack.FindRoute(r.NICID(), r.LocalAddress, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) + if err != nil { + return err + } + defer route.Release() + // From this point on, the incoming route should no longer be used; route + // must be used to send the ICMP error. + r = nil + + sent := p.stack.Stats().ICMP.V4PacketsSent + if !p.stack.AllowICMPMessage() { + sent.RateLimited.Increment() + return nil + } + + networkHeader := pkt.NetworkHeader().View() + transportHeader := pkt.TransportHeader().View() + + // Don't respond to icmp error packets. + if header.IPv4(networkHeader).Protocol() == uint8(header.ICMPv4ProtocolNumber) { + // TODO(gvisor.dev/issue/3810): + // Unfortunately the current stack pretty much always has ICMPv4 headers + // in the Data section of the packet but there is no guarantee that is the + // case. If this is the case grab the header to make it like all other + // packet types. When this is cleaned up the Consume should be removed. + if transportHeader.IsEmpty() { + var ok bool + transportHeader, ok = pkt.TransportHeader().Consume(header.ICMPv4MinimumSize) + if !ok { + return nil + } + } else if transportHeader.Size() < header.ICMPv4MinimumSize { + return nil + } + // We need to decide to explicitly name the packets we can respond to or + // the ones we can not respond to. The decision is somewhat arbitrary and + // if problems arise this could be reversed. It was judged less of a breach + // of protocol to not respond to unknown non-error packets than to respond + // to unknown error packets so we take the first approach. + switch header.ICMPv4(transportHeader).Type() { + case + header.ICMPv4EchoReply, + header.ICMPv4Echo, + header.ICMPv4Timestamp, + header.ICMPv4TimestampReply, + header.ICMPv4InfoRequest, + header.ICMPv4InfoReply: + default: + // Assume any type we don't know about may be an error type. + return nil + } + } + + // Now work out how much of the triggering packet we should return. + // As per RFC 1812 Section 4.3.2.3 + // + // ICMP datagram SHOULD contain as much of the original + // datagram as possible without the length of the ICMP + // datagram exceeding 576 bytes. + // + // NOTE: The above RFC referenced is different from the original + // recommendation in RFC 1122 and RFC 792 where it mentioned that at + // least 8 bytes of the payload must be included. Today linux and other + // systems implement the RFC 1812 definition and not the original + // requirement. We treat 8 bytes as the minimum but will try send more. + mtu := int(route.MTU()) + if mtu > header.IPv4MinimumProcessableDatagramSize { + mtu = header.IPv4MinimumProcessableDatagramSize + } + headerLen := int(route.MaxHeaderLength()) + header.ICMPv4MinimumSize + available := int(mtu) - headerLen + + if available < header.IPv4MinimumSize+header.ICMPv4MinimumErrorPayloadSize { + return nil + } + + payloadLen := networkHeader.Size() + transportHeader.Size() + pkt.Data.Size() + if payloadLen > available { + payloadLen = available + } + + // The buffers used by pkt may be used elsewhere in the system. + // For example, an AF_RAW or AF_PACKET socket may use what the transport + // protocol considers an unreachable destination. Thus we deep copy pkt to + // prevent multiple ownership and SR errors. The new copy is a vectorized + // view with the entire incoming IP packet reassembled and truncated as + // required. This is now the payload of the new ICMP packet and no longer + // considered a packet in its own right. + newHeader := append(buffer.View(nil), networkHeader...) + newHeader = append(newHeader, transportHeader...) + payload := newHeader.ToVectorisedView() + payload.AppendView(pkt.Data.ToView()) + payload.CapLength(payloadLen) + + icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: headerLen, + Data: payload, + }) + + icmpPkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber + + icmpHdr := header.ICMPv4(icmpPkt.TransportHeader().Push(header.ICMPv4MinimumSize)) + switch reason.(type) { + case *icmpReasonPortUnreachable: + icmpHdr.SetCode(header.ICMPv4PortUnreachable) + case *icmpReasonProtoUnreachable: + icmpHdr.SetCode(header.ICMPv4ProtoUnreachable) + default: + panic(fmt.Sprintf("unsupported ICMP type %T", reason)) + } + icmpHdr.SetType(header.ICMPv4DstUnreachable) + icmpHdr.SetChecksum(header.ICMPv4Checksum(icmpHdr, icmpPkt.Data)) + counter := sent.DstUnreachable + + if err := route.WritePacket( + nil, /* gso */ + stack.NetworkHeaderParams{ + Protocol: header.ICMPv4ProtocolNumber, + TTL: route.DefaultTTL(), + TOS: stack.DefaultTOS, + }, + icmpPkt, + ); err != nil { + sent.Dropped.Increment() + return err + } + counter.Increment() + return nil +} diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index b14b356d6..c5ac7b8b5 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -12,17 +12,15 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package ipv4 contains the implementation of the ipv4 network protocol. To use -// it in the networking stack, this package must be added to the project, and -// activated on the stack by passing ipv4.NewProtocol() as one of the network -// protocols when calling stack.New(). Then endpoints can be created by passing -// ipv4.ProtocolNumber as the network protocol number when calling -// Stack.NewEndpoint(). +// Package ipv4 contains the implementation of the ipv4 network protocol. package ipv4 import ( + "fmt" "sync/atomic" + "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -33,6 +31,15 @@ import ( ) const ( + // As per RFC 791 section 3.2: + // The current recommendation for the initial timer setting is 15 seconds. + // This may be changed as experience with this protocol accumulates. + // + // Considering that it is an old recommendation, we use the same reassembly + // timeout that linux defines, which is 30 seconds: + // https://github.com/torvalds/linux/blob/47ec5303d73ea344e84f46660fff693c57641386/include/net/ip.h#L138 + reassembleTimeout = 30 * time.Second + // ProtocolNumber is the ipv4 protocol number. ProtocolNumber = header.IPv4ProtocolNumber @@ -51,22 +58,113 @@ const ( fragmentblockSize = 8 ) +var ipv4BroadcastAddr = header.IPv4Broadcast.WithPrefix() + +var _ stack.GroupAddressableEndpoint = (*endpoint)(nil) +var _ stack.AddressableEndpoint = (*endpoint)(nil) +var _ stack.NetworkEndpoint = (*endpoint)(nil) + type endpoint struct { - nicID tcpip.NICID - linkEP stack.LinkEndpoint + nic stack.NetworkInterface dispatcher stack.TransportDispatcher protocol *protocol - stack *stack.Stack + + // enabled is set to 1 when the enpoint is enabled and 0 when it is + // disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + + mu struct { + sync.RWMutex + + addressableEndpointState stack.AddressableEndpointState + } } // NewEndpoint creates a new ipv4 endpoint. -func (p *protocol) NewEndpoint(nicID tcpip.NICID, _ stack.LinkAddressCache, _ stack.NUDHandler, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint { - return &endpoint{ - nicID: nicID, - linkEP: linkEP, +func (p *protocol) NewEndpoint(nic stack.NetworkInterface, _ stack.LinkAddressCache, _ stack.NUDHandler, dispatcher stack.TransportDispatcher) stack.NetworkEndpoint { + e := &endpoint{ + nic: nic, dispatcher: dispatcher, protocol: p, - stack: st, + } + e.mu.addressableEndpointState.Init(e) + return e +} + +// Enable implements stack.NetworkEndpoint. +func (e *endpoint) Enable() *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + + // If the NIC is not enabled, the endpoint can't do anything meaningful so + // don't enable the endpoint. + if !e.nic.Enabled() { + return tcpip.ErrNotPermitted + } + + // If the endpoint is already enabled, there is nothing for it to do. + if !e.setEnabled(true) { + return nil + } + + // Create an endpoint to receive broadcast packets on this interface. + ep, err := e.mu.addressableEndpointState.AddAndAcquirePermanentAddress(ipv4BroadcastAddr, stack.NeverPrimaryEndpoint, stack.AddressConfigStatic, false /* deprecated */) + if err != nil { + return err + } + // We have no need for the address endpoint. + ep.DecRef() + + // 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 +} + +// Enabled implements stack.NetworkEndpoint. +func (e *endpoint) Enabled() bool { + return e.nic.Enabled() && e.isEnabled() +} + +// isEnabled returns true if the endpoint is enabled, regardless of the +// enabled status of the NIC. +func (e *endpoint) isEnabled() bool { + return atomic.LoadUint32(&e.enabled) == 1 +} + +// setEnabled sets the enabled status for the endpoint. +// +// Returns true if the enabled status was updated. +func (e *endpoint) setEnabled(v bool) bool { + if v { + return atomic.SwapUint32(&e.enabled, 1) == 0 + } + return atomic.SwapUint32(&e.enabled, 0) == 1 +} + +// Disable implements stack.NetworkEndpoint. +func (e *endpoint) Disable() { + e.mu.Lock() + defer e.mu.Unlock() + e.disableLocked() +} + +func (e *endpoint) disableLocked() { + if !e.setEnabled(false) { + return + } + + // The endpoint may have already left the multicast group. + if _, err := e.mu.addressableEndpointState.LeaveGroup(header.IPv4AllSystems); err != nil && err != tcpip.ErrBadLocalAddress { + panic(fmt.Sprintf("unexpected error when leaving group = %s: %s", header.IPv4AllSystems, err)) + } + + // 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)) } } @@ -78,31 +176,13 @@ func (e *endpoint) DefaultTTL() uint8 { // MTU implements stack.NetworkEndpoint.MTU. It returns the link-layer MTU minus // the network layer max header length. func (e *endpoint) MTU() uint32 { - return calculateMTU(e.linkEP.MTU()) -} - -// Capabilities implements stack.NetworkEndpoint.Capabilities. -func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { - return e.linkEP.Capabilities() -} - -// NICID returns the ID of the NIC this endpoint belongs to. -func (e *endpoint) NICID() tcpip.NICID { - return e.nicID + return calculateMTU(e.nic.MTU()) } // MaxHeaderLength returns the maximum length needed by ipv4 headers (and // underlying protocols). func (e *endpoint) MaxHeaderLength() uint16 { - return e.linkEP.MaxHeaderLength() + header.IPv4MinimumSize -} - -// GSOMaxSize returns the maximum GSO packet size. -func (e *endpoint) GSOMaxSize() uint32 { - if gso, ok := e.linkEP.(stack.GSOEndpoint); ok { - return gso.GSOMaxSize() - } - return 0 + return e.nic.MaxHeaderLength() + header.IPv4MaximumHeaderSize } // NetworkProtocolNumber implements stack.NetworkEndpoint.NetworkProtocolNumber. @@ -110,98 +190,26 @@ func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return e.protocol.Number() } -// writePacketFragments calls e.linkEP.WritePacket with each packet fragment to -// write. It assumes that the IP header is already present in pkt.NetworkHeader. -// pkt.TransportHeader may be set. mtu includes the IP header and options. This -// does not support the DontFragment IP flag. -func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu int, pkt *stack.PacketBuffer) *tcpip.Error { - // This packet is too big, it needs to be fragmented. - ip := header.IPv4(pkt.NetworkHeader().View()) - flags := ip.Flags() - - // Update mtu to take into account the header, which will exist in all - // fragments anyway. - innerMTU := mtu - int(ip.HeaderLength()) - - // Round the MTU down to align to 8 bytes. Then calculate the number of - // fragments. Calculate fragment sizes as in RFC791. - innerMTU &^= 7 - n := (int(ip.PayloadLength()) + innerMTU - 1) / innerMTU - - outerMTU := innerMTU + int(ip.HeaderLength()) - offset := ip.FragmentOffset() - - // Keep the length reserved for link-layer, we need to create fragments with - // the same reserved length. - reservedForLink := pkt.AvailableHeaderBytes() - - // Destroy the packet, pull all payloads out for fragmentation. - transHeader, data := pkt.TransportHeader().View(), pkt.Data - - // Where possible, the first fragment that is sent has the same - // number of bytes reserved for header as the input packet. The link-layer - // endpoint may depend on this for looking at, eg, L4 headers. - transFitsFirst := len(transHeader) <= innerMTU - - for i := 0; i < n; i++ { - reserve := reservedForLink + int(ip.HeaderLength()) - if i == 0 && transFitsFirst { - // Reserve for transport header if it's going to be put in the first - // fragment. - reserve += len(transHeader) - } - fragPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - ReserveHeaderBytes: reserve, - }) - fragPkt.NetworkProtocolNumber = header.IPv4ProtocolNumber - - // Copy data for the fragment. - avail := innerMTU - - if n := len(transHeader); n > 0 { - if n > avail { - n = avail - } - if i == 0 && transFitsFirst { - copy(fragPkt.TransportHeader().Push(n), transHeader) - } else { - fragPkt.Data.AppendView(transHeader[:n:n]) - } - transHeader = transHeader[n:] - avail -= n - } - - if avail > 0 { - n := data.Size() - if n > avail { - n = avail - } - data.ReadToVV(&fragPkt.Data, n) - avail -= n - } - - copied := uint16(innerMTU - avail) - - // Set lengths in header and calculate checksum. - h := header.IPv4(fragPkt.NetworkHeader().Push(len(ip))) - copy(h, ip) - if i != n-1 { - h.SetTotalLength(uint16(outerMTU)) - h.SetFlagsFragmentOffset(flags|header.IPv4FlagMoreFragments, offset) - } else { - h.SetTotalLength(uint16(h.HeaderLength()) + copied) - h.SetFlagsFragmentOffset(flags, offset) - } - h.SetChecksum(0) - h.SetChecksum(^h.CalculateChecksum()) - offset += copied - - // Send out the fragment. - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, fragPkt); err != nil { +// writePacketFragments fragments pkt and writes the results on the link +// endpoint. The IP header must already present in the original packet. The mtu +// is the maximum size of the packets. +func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, mtu uint32, pkt *stack.PacketBuffer) *tcpip.Error { + networkHeader := header.IPv4(pkt.NetworkHeader().View()) + fragMTU := int(calculateFragmentInnerMTU(mtu, pkt)) + pf := fragmentation.MakePacketFragmenter(pkt, fragMTU, pkt.AvailableHeaderBytes()+len(networkHeader)) + + for { + fragPkt, more := buildNextFragment(&pf, networkHeader) + if err := e.nic.WritePacket(r, gso, ProtocolNumber, fragPkt); err != nil { + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pf.RemainingFragmentCount() + 1)) return err } r.Stats().IP.PacketsSent.Increment() + if !more { + break + } } + return nil } @@ -223,7 +231,7 @@ func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params s DstAddr: r.RemoteAddress, }) ip.SetChecksum(^ip.CalculateChecksum()) - pkt.NetworkProtocolNumber = header.IPv4ProtocolNumber + pkt.NetworkProtocolNumber = ProtocolNumber } // WritePacket writes a packet to the given destination address and protocol. @@ -232,8 +240,8 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // iptables filtering. All packets that reach here are locally // generated. - nicName := e.stack.FindNICNameFromID(e.NICID()) - ipt := e.stack.IPTables() + nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID()) + ipt := e.protocol.stack.IPTables() if ok := ipt.Check(stack.Output, pkt, gso, r, "", nicName); !ok { // iptables is telling us to drop the packet. r.Stats().IP.IPTablesOutputDropped.Increment() @@ -249,7 +257,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // short circuits broadcasts before they are sent out to other hosts. if pkt.NatDone { netHeader := header.IPv4(pkt.NetworkHeader().View()) - ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()) + ep, err := e.protocol.stack.FindNetworkEndpoint(ProtocolNumber, netHeader.DestinationAddress()) if err == nil { route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress()) ep.HandlePacket(&route, pkt) @@ -265,10 +273,11 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw if r.Loop&stack.PacketOut == 0 { return nil } - if pkt.Size() > int(e.linkEP.MTU()) && (gso == nil || gso.Type == stack.GSONone) { - return e.writePacketFragments(r, gso, int(e.linkEP.MTU()), pkt) + if pkt.Size() > int(e.nic.MTU()) && (gso == nil || gso.Type == stack.GSONone) { + return e.writePacketFragments(r, gso, e.nic.MTU(), pkt) } - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { + if err := e.nic.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { + r.Stats().IP.OutgoingPacketErrors.Increment() return err } r.Stats().IP.PacketsSent.Increment() @@ -289,16 +298,19 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe pkt = pkt.Next() } - nicName := e.stack.FindNICNameFromID(e.NICID()) + nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID()) // iptables filtering. All packets that reach here are locally // generated. - ipt := e.stack.IPTables() + ipt := e.protocol.stack.IPTables() dropped, natPkts := ipt.CheckPackets(stack.Output, pkts, gso, r, nicName) if len(dropped) == 0 && len(natPkts) == 0 { // Fast path: If no packets are to be dropped then we can just invoke the // faster WritePackets API directly. - n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) + n, err := e.nic.WritePackets(r, gso, pkts, ProtocolNumber) r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + if err != nil { + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pkts.Len() - n)) + } return n, err } r.Stats().IP.IPTablesOutputDropped.IncrementBy(uint64(len(dropped))) @@ -312,7 +324,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } if _, ok := natPkts[pkt]; ok { netHeader := header.IPv4(pkt.NetworkHeader().View()) - if ep, err := e.stack.FindNetworkEndpoint(header.IPv4ProtocolNumber, netHeader.DestinationAddress()); err == nil { + if ep, err := e.protocol.stack.FindNetworkEndpoint(ProtocolNumber, netHeader.DestinationAddress()); err == nil { src := netHeader.SourceAddress() dst := netHeader.DestinationAddress() route := r.ReverseRoute(src, dst) @@ -321,8 +333,9 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe continue } } - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { + if err := e.nic.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pkts.Len() - n - len(dropped))) // Dropped packets aren't errors, so include them in // the return value. return n + len(dropped), err @@ -381,23 +394,39 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBu return nil } + if err := e.nic.WritePacket(r, nil /* gso */, ProtocolNumber, pkt); err != nil { + r.Stats().IP.OutgoingPacketErrors.Increment() + return err + } r.Stats().IP.PacketsSent.Increment() - - return e.linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) + return nil } // HandlePacket is called by the link layer when new ipv4 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { + if !e.isEnabled() { + return + } + h := header.IPv4(pkt.NetworkHeader().View()) if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() return } + // 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 r.IsOutboundBroadcast() || header.IsV4MulticastAddress(r.RemoteAddress) { + r.Stats().IP.InvalidSourceAddressesReceived.Increment() + return + } + // iptables filtering. All packets that reach here are intended for // this machine and will not be forwarded. - ipt := e.stack.IPTables() + ipt := e.protocol.stack.IPTables() if ok := ipt.Check(stack.Input, pkt, nil, nil, "", ""); !ok { // iptables is telling us to drop the packet. r.Stats().IP.IPTablesInputDropped.Increment() @@ -453,27 +482,165 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { return } } + + r.Stats().IP.PacketsDelivered.Increment() p := h.TransportProtocol() if p == header.ICMPv4ProtocolNumber { + // TODO(gvisor.dev/issues/3810): when we sort out ICMP and transport + // headers, the setting of the transport number here should be + // unnecessary and removed. + pkt.TransportProtocolNumber = p e.handleICMP(r, pkt) return } - r.Stats().IP.PacketsDelivered.Increment() - e.dispatcher.DeliverTransportPacket(r, p, pkt) + + switch res := e.dispatcher.DeliverTransportPacket(r, p, pkt); res { + case stack.TransportPacketHandled: + case stack.TransportPacketDestinationPortUnreachable: + // As per RFC: 1122 Section 3.2.2.1 A host SHOULD generate Destination + // Unreachable messages with code: + // 3 (Port Unreachable), when the designated transport protocol + // (e.g., UDP) is unable to demultiplex the datagram but has no + // protocol mechanism to inform the sender. + _ = e.protocol.returnError(r, &icmpReasonPortUnreachable{}, pkt) + case stack.TransportPacketProtocolUnreachable: + // As per RFC: 1122 Section 3.2.2.1 + // A host SHOULD generate Destination Unreachable messages with code: + // 2 (Protocol Unreachable), when the designated transport protocol + // is not supported + _ = e.protocol.returnError(r, &icmpReasonProtoUnreachable{}, pkt) + default: + panic(fmt.Sprintf("unrecognized result from DeliverTransportPacket = %d", res)) + } } // Close cleans up resources associated with the endpoint. -func (e *endpoint) Close() {} +func (e *endpoint) Close() { + e.mu.Lock() + defer e.mu.Unlock() + + e.disableLocked() + e.mu.addressableEndpointState.Cleanup() +} + +// AddAndAcquirePermanentAddress implements stack.AddressableEndpoint. +func (e *endpoint) AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb stack.PrimaryEndpointBehavior, configType stack.AddressConfigType, deprecated bool) (stack.AddressEndpoint, *tcpip.Error) { + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.AddAndAcquirePermanentAddress(addr, peb, configType, deprecated) +} + +// RemovePermanentAddress implements stack.AddressableEndpoint. +func (e *endpoint) RemovePermanentAddress(addr tcpip.Address) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.RemovePermanentAddress(addr) +} + +// MainAddress implements stack.AddressableEndpoint. +func (e *endpoint) MainAddress() tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.MainAddress() +} + +// AcquireAssignedAddress implements stack.AddressableEndpoint. +func (e *endpoint) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB stack.PrimaryEndpointBehavior) stack.AddressEndpoint { + e.mu.Lock() + defer e.mu.Unlock() + + loopback := e.nic.IsLoopback() + addressEndpoint := e.mu.addressableEndpointState.ReadOnly().AddrOrMatching(localAddr, allowTemp, func(addressEndpoint stack.AddressEndpoint) bool { + subnet := addressEndpoint.AddressWithPrefix().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 +} + +// AcquireOutgoingPrimaryAddress implements stack.AddressableEndpoint. +func (e *endpoint) AcquireOutgoingPrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) stack.AddressEndpoint { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.AcquireOutgoingPrimaryAddress(remoteAddr, allowExpired) +} + +// PrimaryAddresses implements stack.AddressableEndpoint. +func (e *endpoint) PrimaryAddresses() []tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.PrimaryAddresses() +} + +// PermanentAddresses implements stack.AddressableEndpoint. +func (e *endpoint) PermanentAddresses() []tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.PermanentAddresses() +} + +// JoinGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) JoinGroup(addr tcpip.Address) (bool, *tcpip.Error) { + if !header.IsV4MulticastAddress(addr) { + return false, tcpip.ErrBadAddress + } + + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.JoinGroup(addr) +} + +// LeaveGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) LeaveGroup(addr tcpip.Address) (bool, *tcpip.Error) { + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.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) +} + +var _ stack.ForwardingNetworkProtocol = (*protocol)(nil) +var _ stack.NetworkProtocol = (*protocol)(nil) type protocol struct { - ids []uint32 - hashIV uint32 + stack *stack.Stack // defaultTTL is the current default TTL for the protocol. Only the - // uint8 portion of it is meaningful and it must be accessed - // atomically. + // uint8 portion of it is meaningful. + // + // Must be accessed using atomic operations. defaultTTL uint32 + // forwarding is set to 1 when the protocol has forwarding enabled and 0 + // when it is disabled. + // + // Must be accessed using atomic operations. + forwarding uint32 + + ids []uint32 + hashIV uint32 + fragmentation *fragmentation.Fragmentation } @@ -546,6 +713,20 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu return ipHdr.TransportProtocol(), !ipHdr.More() && ipHdr.FragmentOffset() == 0, true } +// Forwarding implements stack.ForwardingNetworkProtocol. +func (p *protocol) Forwarding() bool { + return uint8(atomic.LoadUint32(&p.forwarding)) == 1 +} + +// SetForwarding implements stack.ForwardingNetworkProtocol. +func (p *protocol) SetForwarding(v bool) { + if v { + atomic.StoreUint32(&p.forwarding, 1) + } else { + atomic.StoreUint32(&p.forwarding, 0) + } +} + // calculateMTU calculates the network-layer payload MTU based on the link-layer // payload mtu. func calculateMTU(mtu uint32) uint32 { @@ -555,19 +736,41 @@ func calculateMTU(mtu uint32) uint32 { return mtu - header.IPv4MinimumSize } +// calculateFragmentInnerMTU calculates the maximum number of bytes of +// fragmentable data a fragment can have, based on the link layer mtu and pkt's +// network header size. +func calculateFragmentInnerMTU(mtu uint32, pkt *stack.PacketBuffer) uint32 { + if mtu > MaxTotalSize { + mtu = MaxTotalSize + } + mtu -= uint32(pkt.NetworkHeader().View().Size()) + // Round the MTU down to align to 8 bytes. + mtu &^= 7 + return mtu +} + +// addressToUint32 translates an IPv4 address into its little endian uint32 +// representation. +// +// This function does the same thing as binary.LittleEndian.Uint32 but operates +// on a tcpip.Address (a string) without the need to convert it to a byte slice, +// which would cause an allocation. +func addressToUint32(addr tcpip.Address) uint32 { + _ = addr[3] // bounds check hint to compiler + 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 random initial -// value (generated once on initialization) to generate the hash. +// 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 { - t := r.LocalAddress - a := uint32(t[0]) | uint32(t[1])<<8 | uint32(t[2])<<16 | uint32(t[3])<<24 - t = r.RemoteAddress - b := uint32(t[0]) | uint32(t[1])<<8 | uint32(t[2])<<16 | uint32(t[3])<<24 + a := addressToUint32(r.LocalAddress) + b := addressToUint32(r.RemoteAddress) return hash.Hash3Words(a, b, uint32(protocol), hashIV) } // NewProtocol returns an IPv4 network protocol. -func NewProtocol() stack.NetworkProtocol { +func NewProtocol(s *stack.Stack) stack.NetworkProtocol { ids := make([]uint32, buckets) // Randomly initialize hashIV and the ids. @@ -578,9 +781,33 @@ func NewProtocol() stack.NetworkProtocol { hashIV := r[buckets] return &protocol{ + stack: s, ids: ids, hashIV: hashIV, defaultTTL: DefaultTTL, - fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), + fragmentation: fragmentation.NewFragmentation(fragmentblockSize, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, reassembleTimeout, s.Clock()), } } + +func buildNextFragment(pf *fragmentation.PacketFragmenter, originalIPHeader header.IPv4) (*stack.PacketBuffer, bool) { + fragPkt, offset, copied, more := pf.BuildNextFragment() + fragPkt.NetworkProtocolNumber = ProtocolNumber + + originalIPHeaderLength := len(originalIPHeader) + nextFragIPHeader := header.IPv4(fragPkt.NetworkHeader().Push(originalIPHeaderLength)) + + if copied := copy(nextFragIPHeader, originalIPHeader); copied != len(originalIPHeader) { + panic(fmt.Sprintf("wrong number of bytes copied into fragmentIPHeaders: got = %d, want = %d", copied, originalIPHeaderLength)) + } + + flags := originalIPHeader.Flags() + if more { + flags |= header.IPv4FlagMoreFragments + } + nextFragIPHeader.SetFlagsFragmentOffset(flags, uint16(offset)) + nextFragIPHeader.SetTotalLength(uint16(nextFragIPHeader.HeaderLength()) + uint16(copied)) + nextFragIPHeader.SetChecksum(0) + nextFragIPHeader.SetChecksum(^nextFragIPHeader.CalculateChecksum()) + + return fragPkt, more +} diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index b14bc98e8..9916d783f 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -16,18 +16,24 @@ package ipv4_test import ( "bytes" + "context" "encoding/hex" + "math" + "net" "testing" "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/checker" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" + "gvisor.dev/gvisor/pkg/tcpip/network/arp" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/testutil" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/icmp" "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/pkg/waiter" @@ -35,8 +41,8 @@ import ( func TestExcludeBroadcast(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) const defaultMTU = 65536 @@ -91,6 +97,276 @@ func TestExcludeBroadcast(t *testing.T) { }) } +// TestIPv4Sanity sends IP/ICMP packets with various problems to the stack and +// checks the response. +func TestIPv4Sanity(t *testing.T) { + const ( + defaultMTU = header.IPv6MinimumMTU + ttl = 255 + nicID = 1 + randomSequence = 123 + randomIdent = 42 + ) + var ( + ipv4Addr = tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("192.168.1.58").To4()), + PrefixLen: 24, + } + remoteIPv4Addr = tcpip.Address(net.ParseIP("10.0.0.1").To4()) + ) + + tests := []struct { + name string + headerLength uint8 // value of 0 means "use correct size" + maxTotalLength uint16 + transportProtocol uint8 + TTL uint8 + shouldFail bool + expectICMP bool + ICMPType header.ICMPv4Type + ICMPCode header.ICMPv4Code + options []byte + }{ + { + name: "valid", + maxTotalLength: defaultMTU, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + }, + // The TTL tests check that we are not rejecting an incoming packet + // with a zero or one TTL, which has been a point of confusion in the + // past as RFC 791 says: "If this field contains the value zero, then the + // datagram must be destroyed". However RFC 1122 section 3.2.1.7 clarifies + // for the case of the destination host, stating as follows. + // + // A host MUST NOT send a datagram with a Time-to-Live (TTL) + // value of zero. + // + // A host MUST NOT discard a datagram just because it was + // received with TTL less than 2. + { + name: "zero TTL", + maxTotalLength: defaultMTU, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: 0, + shouldFail: false, + }, + { + name: "one TTL", + maxTotalLength: defaultMTU, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: 1, + shouldFail: false, + }, + { + name: "End options", + maxTotalLength: defaultMTU, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: []byte{0, 0, 0, 0}, + }, + { + name: "NOP options", + maxTotalLength: defaultMTU, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: []byte{1, 1, 1, 1}, + }, + { + name: "NOP and End options", + maxTotalLength: defaultMTU, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + options: []byte{1, 1, 0, 0}, + }, + { + name: "bad header length", + headerLength: header.IPv4MinimumSize - 1, + maxTotalLength: defaultMTU, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + shouldFail: true, + expectICMP: false, + }, + { + name: "bad total length (0)", + maxTotalLength: 0, + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + shouldFail: true, + expectICMP: false, + }, + { + name: "bad total length (ip - 1)", + maxTotalLength: uint16(header.IPv4MinimumSize - 1), + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + shouldFail: true, + expectICMP: false, + }, + { + name: "bad total length (ip + icmp - 1)", + maxTotalLength: uint16(header.IPv4MinimumSize + header.ICMPv4MinimumSize - 1), + transportProtocol: uint8(header.ICMPv4ProtocolNumber), + TTL: ttl, + shouldFail: true, + expectICMP: false, + }, + { + name: "bad protocol", + maxTotalLength: defaultMTU, + transportProtocol: 99, + TTL: ttl, + shouldFail: true, + expectICMP: true, + ICMPType: header.ICMPv4DstUnreachable, + ICMPCode: header.ICMPv4ProtoUnreachable, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol4}, + }) + // We expect at most a single packet in response to our ICMP Echo Request. + e := channel.New(1, defaultMTU, "") + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID, err) + } + ipv4ProtoAddr := tcpip.ProtocolAddress{Protocol: header.IPv4ProtocolNumber, AddressWithPrefix: ipv4Addr} + if err := s.AddProtocolAddress(nicID, ipv4ProtoAddr); err != nil { + t.Fatalf("AddProtocolAddress(%d, %#v): %s", nicID, ipv4ProtoAddr, err) + } + + // Default routes for IPv4 so ICMP can find a route to the remote + // node when attempting to send the ICMP Echo Reply. + s.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: header.IPv4EmptySubnet, + NIC: nicID, + }, + }) + + // Round up the header size to the next multiple of 4 as RFC 791, page 11 + // says: "Internet Header Length is the length of the internet header + // in 32 bit words..." and on page 23: "The internet header padding is + // used to ensure that the internet header ends on a 32 bit boundary." + ipHeaderLength := ((header.IPv4MinimumSize + len(test.options)) + header.IPv4IHLStride - 1) & ^(header.IPv4IHLStride - 1) + + if ipHeaderLength > header.IPv4MaximumHeaderSize { + t.Fatalf("too many bytes in options: got = %d, want <= %d ", ipHeaderLength, header.IPv4MaximumHeaderSize) + } + totalLen := uint16(ipHeaderLength + header.ICMPv4MinimumSize) + hdr := buffer.NewPrependable(int(totalLen)) + icmp := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) + + // Specify ident/seq to make sure we get the same in the response. + icmp.SetIdent(randomIdent) + icmp.SetSequence(randomSequence) + icmp.SetType(header.ICMPv4Echo) + icmp.SetCode(header.ICMPv4UnusedCode) + icmp.SetChecksum(0) + icmp.SetChecksum(^header.Checksum(icmp, 0)) + ip := header.IPv4(hdr.Prepend(ipHeaderLength)) + if test.maxTotalLength < totalLen { + totalLen = test.maxTotalLength + } + ip.Encode(&header.IPv4Fields{ + IHL: uint8(ipHeaderLength), + TotalLength: totalLen, + Protocol: test.transportProtocol, + TTL: test.TTL, + SrcAddr: remoteIPv4Addr, + DstAddr: ipv4Addr.Address, + }) + if n := copy(ip.Options(), test.options); n != len(test.options) { + t.Fatalf("options larger than available space: copied %d/%d bytes", n, len(test.options)) + } + // Override the correct value if the test case specified one. + if test.headerLength != 0 { + ip.SetHeaderLength(test.headerLength) + } + requestPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + }) + e.InjectInbound(header.IPv4ProtocolNumber, requestPkt) + reply, ok := e.Read() + if !ok { + if test.shouldFail { + if test.expectICMP { + t.Fatal("expected ICMP error response missing") + } + return // Expected silent failure. + } + t.Fatal("expected ICMP echo reply missing") + } + + // Check the route that brought the packet to us. + if reply.Route.LocalAddress != ipv4Addr.Address { + t.Errorf("got pkt.Route.LocalAddress = %s, want = %s", reply.Route.LocalAddress, ipv4Addr.Address) + } + if reply.Route.RemoteAddress != remoteIPv4Addr { + t.Errorf("got pkt.Route.RemoteAddress = %s, want = %s", reply.Route.RemoteAddress, remoteIPv4Addr) + } + + // Make sure it's all in one buffer. + vv := buffer.NewVectorisedView(reply.Pkt.Size(), reply.Pkt.Views()) + replyIPHeader := header.IPv4(vv.ToView()) + + // At this stage we only know it's an IP header so verify that much. + checker.IPv4(t, replyIPHeader, + checker.SrcAddr(ipv4Addr.Address), + checker.DstAddr(remoteIPv4Addr), + ) + + // All expected responses are ICMP packets. + if got, want := replyIPHeader.Protocol(), uint8(header.ICMPv4ProtocolNumber); got != want { + t.Fatalf("not ICMP response, got protocol %d, want = %d", got, want) + } + replyICMPHeader := header.ICMPv4(replyIPHeader.Payload()) + + // Sanity check the response. + switch replyICMPHeader.Type() { + case header.ICMPv4DstUnreachable: + checker.IPv4(t, replyIPHeader, + checker.IPFullLength(uint16(header.IPv4MinimumSize+header.ICMPv4MinimumSize+requestPkt.Size())), + checker.IPv4HeaderLength(header.IPv4MinimumSize), + checker.ICMPv4( + checker.ICMPv4Code(test.ICMPCode), + checker.ICMPv4Checksum(), + checker.ICMPv4Payload([]byte(hdr.View())), + ), + ) + if !test.shouldFail || !test.expectICMP { + t.Fatalf("unexpected packet rejection, got ICMP error packet type %d, code %d", + header.ICMPv4DstUnreachable, replyICMPHeader.Code()) + } + return + case header.ICMPv4EchoReply: + checker.IPv4(t, replyIPHeader, + checker.IPv4HeaderLength(ipHeaderLength), + checker.IPv4Options(test.options), + checker.IPFullLength(uint16(requestPkt.Size())), + checker.ICMPv4( + checker.ICMPv4Code(header.ICMPv4UnusedCode), + checker.ICMPv4Seq(randomSequence), + checker.ICMPv4Ident(randomIdent), + checker.ICMPv4Checksum(), + ), + ) + if test.shouldFail { + t.Fatalf("unexpected Echo Reply packet\n") + } + default: + t.Fatalf("unexpected ICMP response, got type %d, want = %d or %d", + replyICMPHeader.Type(), header.ICMPv4EchoReply, header.ICMPv4DstUnreachable) + } + }) + } +} + // comparePayloads compared the contents of all the packets against the contents // of the source packet. func compareFragments(t *testing.T, packets []*stack.PacketBuffer, sourcePacketInfo *stack.PacketBuffer, mtu uint32) { @@ -122,16 +398,6 @@ func compareFragments(t *testing.T, packets []*stack.PacketBuffer, sourcePacketI if got, want := len(ip), int(mtu); got > want { t.Errorf("fragment is too large, got %d want %d", got, want) } - if i == 0 { - got := packet.NetworkHeader().View().Size() + packet.TransportHeader().View().Size() - // sourcePacketInfo does not have NetworkHeader added, simulate one. - want := header.IPv4MinimumSize + sourcePacketInfo.TransportHeader().View().Size() - // Check that it kept the transport header in packet.TransportHeader if - // it fits in the first fragment. - if want < int(mtu) && got != want { - t.Errorf("first fragment hdr parts should have unmodified length if possible: got %d, want %d", got, want) - } - } if got, want := packet.AvailableHeaderBytes(), sourcePacketInfo.AvailableHeaderBytes()-header.IPv4MinimumSize; got != want { t.Errorf("fragment #%d should have the same available space for prepending as source: got %d, want %d", i, got, want) } @@ -160,48 +426,9 @@ func compareFragments(t *testing.T, packets []*stack.PacketBuffer, sourcePacketI } } -type testRoute struct { - stack.Route - - linkEP *testutil.TestEndpoint -} - -func buildTestRoute(t *testing.T, ep *channel.Endpoint, packetCollectorErrors []*tcpip.Error) testRoute { - // Make the packet and write it. - s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - }) - testEP := testutil.NewTestEndpoint(ep, packetCollectorErrors) - s.CreateNIC(1, testEP) - const ( - src = "\x10\x00\x00\x01" - dst = "\x10\x00\x00\x02" - ) - s.AddAddress(1, ipv4.ProtocolNumber, src) - { - subnet, err := tcpip.NewSubnet(dst, tcpip.AddressMask(header.IPv4Broadcast)) - if err != nil { - t.Fatal(err) - } - s.SetRouteTable([]tcpip.Route{{ - Destination: subnet, - NIC: 1, - }}) - } - r, err := s.FindRoute(0, src, dst, ipv4.ProtocolNumber, false /* multicastLoop */) - if err != nil { - t.Fatalf("s.FindRoute got %v, want %v", err, nil) - } - t.Cleanup(func() { - testEP.Close() - }) - return testRoute{ - Route: r, - linkEP: testEP, - } -} - func TestFragmentation(t *testing.T) { + const ttl = 42 + var manyPayloadViewsSizes [1000]int for i := range manyPayloadViewsSizes { manyPayloadViewsSizes[i] = 7 @@ -215,38 +442,42 @@ func TestFragmentation(t *testing.T) { payloadViewsSizes []int expectedFrags int }{ - {"NoFragmentation", 2000, &stack.GSO{}, 0, header.IPv4MinimumSize, []int{1000}, 1}, - {"NoFragmentationWithBigHeader", 2000, &stack.GSO{}, 16, header.IPv4MinimumSize, []int{1000}, 1}, + {"No fragmentation", 2000, &stack.GSO{}, 0, header.IPv4MinimumSize, []int{1000}, 1}, + {"No fragmentation with big header", 2000, &stack.GSO{}, 16, header.IPv4MinimumSize, []int{1000}, 1}, {"Fragmented", 800, &stack.GSO{}, 0, header.IPv4MinimumSize, []int{1000}, 2}, - {"FragmentedWithGsoNil", 800, nil, 0, header.IPv4MinimumSize, []int{1000}, 2}, - {"FragmentedWithManyViews", 300, &stack.GSO{}, 0, header.IPv4MinimumSize, manyPayloadViewsSizes[:], 25}, - {"FragmentedWithManyViewsAndPrependableBytes", 300, &stack.GSO{}, 0, header.IPv4MinimumSize + 55, manyPayloadViewsSizes[:], 25}, - {"FragmentedWithBigHeader", 800, &stack.GSO{}, 20, header.IPv4MinimumSize, []int{1000}, 2}, - {"FragmentedWithBigHeaderAndPrependableBytes", 800, &stack.GSO{}, 20, header.IPv4MinimumSize + 66, []int{1000}, 2}, - {"FragmentedWithMTUSmallerThanHeaderAndPrependableBytes", 300, &stack.GSO{}, 1000, header.IPv4MinimumSize + 77, []int{500}, 6}, + {"Fragmented with gso nil", 800, nil, 0, header.IPv4MinimumSize, []int{1000}, 2}, + {"Fragmented with many views", 300, &stack.GSO{}, 0, header.IPv4MinimumSize, manyPayloadViewsSizes[:], 25}, + {"Fragmented with many views and prependable bytes", 300, &stack.GSO{}, 0, header.IPv4MinimumSize + 55, manyPayloadViewsSizes[:], 25}, + {"Fragmented with big header", 800, &stack.GSO{}, 20, header.IPv4MinimumSize, []int{1000}, 2}, + {"Fragmented with big header and prependable bytes", 800, &stack.GSO{}, 20, header.IPv4MinimumSize + 66, []int{1000}, 2}, + {"Fragmented with MTU smaller than header and prependable bytes", 300, &stack.GSO{}, 1000, header.IPv4MinimumSize + 77, []int{500}, 6}, } for _, ft := range fragTests { t.Run(ft.description, func(t *testing.T) { - r := buildTestRoute(t, channel.New(0, ft.mtu, ""), nil) + ep := testutil.NewMockLinkEndpoint(ft.mtu, nil, math.MaxInt32) + r := buildRoute(t, ep) pkt := testutil.MakeRandPkt(ft.transportHeaderLength, ft.extraHeaderReserveLength, ft.payloadViewsSizes, header.IPv4ProtocolNumber) source := pkt.Clone() err := r.WritePacket(ft.gso, stack.NetworkHeaderParams{ Protocol: tcp.ProtocolNumber, - TTL: 42, + TTL: ttl, TOS: stack.DefaultTOS, }, pkt) if err != nil { - t.Errorf("err got %v, want %v", err, nil) + t.Fatalf("r.WritePacket(_, _, _) = %s", err) } - if got, want := len(r.linkEP.WrittenPackets), ft.expectedFrags; got != want { - t.Errorf("len(r.linkEP.WrittenPackets) got %d, want %d", got, want) + if got := len(ep.WrittenPackets); got != ft.expectedFrags { + t.Errorf("got len(ep.WrittenPackets) = %d, want = %d", got, ft.expectedFrags) + } + if got, want := len(ep.WrittenPackets), int(r.Stats().IP.PacketsSent.Value()); got != want { + t.Errorf("no errors yet got len(ep.WrittenPackets) = %d, want = %d", got, want) } - if got, want := len(r.linkEP.WrittenPackets), int(r.Stats().IP.PacketsSent.Value()); got != want { - t.Errorf("no errors yet len(r.linkEP.WrittenPackets) got %d, want %d", got, want) + if got := r.Stats().IP.OutgoingPacketErrors.Value(); got != 0 { + t.Errorf("got r.Stats().IP.OutgoingPacketErrors.Value() = %d, want = 0", got) } - compareFragments(t, r.linkEP.WrittenPackets, source, ft.mtu) + compareFragments(t, ep.WrittenPackets, source, ft.mtu) }) } } @@ -254,40 +485,69 @@ func TestFragmentation(t *testing.T) { // TestFragmentationErrors checks that errors are returned from write packet // correctly. func TestFragmentationErrors(t *testing.T) { + const ttl = 42 + + expectedError := tcpip.ErrAborted fragTests := []struct { description string mtu uint32 transportHeaderLength int - payloadViewsSizes []int - packetCollectorErrors []*tcpip.Error + payloadSize int + allowPackets int + fragmentCount int }{ - {"NoFrag", 2000, 0, []int{1000}, []*tcpip.Error{tcpip.ErrAborted}}, - {"ErrorOnFirstFrag", 500, 0, []int{1000}, []*tcpip.Error{tcpip.ErrAborted}}, - {"ErrorOnSecondFrag", 500, 0, []int{1000}, []*tcpip.Error{nil, tcpip.ErrAborted}}, - {"ErrorOnFirstFragMTUSmallerThanHeader", 500, 1000, []int{500}, []*tcpip.Error{tcpip.ErrAborted}}, + { + description: "No frag", + mtu: 2000, + transportHeaderLength: 0, + payloadSize: 1000, + allowPackets: 0, + fragmentCount: 1, + }, + { + description: "Error on first frag", + mtu: 500, + transportHeaderLength: 0, + payloadSize: 1000, + allowPackets: 0, + fragmentCount: 3, + }, + { + description: "Error on second frag", + mtu: 500, + transportHeaderLength: 0, + payloadSize: 1000, + allowPackets: 1, + fragmentCount: 3, + }, + { + description: "Error on first frag MTU smaller than header", + mtu: 500, + transportHeaderLength: 1000, + payloadSize: 500, + allowPackets: 0, + fragmentCount: 4, + }, } for _, ft := range fragTests { t.Run(ft.description, func(t *testing.T) { - r := buildTestRoute(t, channel.New(0, ft.mtu, ""), ft.packetCollectorErrors) - pkt := testutil.MakeRandPkt(ft.transportHeaderLength, header.IPv4MinimumSize, ft.payloadViewsSizes, header.IPv4ProtocolNumber) + ep := testutil.NewMockLinkEndpoint(ft.mtu, expectedError, ft.allowPackets) + r := buildRoute(t, ep) + pkt := testutil.MakeRandPkt(ft.transportHeaderLength, header.IPv4MinimumSize, []int{ft.payloadSize}, header.IPv4ProtocolNumber) err := r.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{ Protocol: tcp.ProtocolNumber, - TTL: 42, + TTL: ttl, TOS: stack.DefaultTOS, }, pkt) - for i := 0; i < len(ft.packetCollectorErrors)-1; i++ { - if got, want := ft.packetCollectorErrors[i], (*tcpip.Error)(nil); got != want { - t.Errorf("ft.packetCollectorErrors[%d] got %v, want %v", i, got, want) - } + if err != expectedError { + t.Errorf("got WritePacket() = %s, want = %s", err, expectedError) } - // We only need to check that last error because all the ones before are - // nil. - if got, want := err, ft.packetCollectorErrors[len(ft.packetCollectorErrors)-1]; got != want { - t.Errorf("err got %v, want %v", got, want) + if got, want := len(ep.WrittenPackets), int(r.Stats().IP.PacketsSent.Value()); err != nil && got != want { + t.Errorf("got len(ep.WrittenPackets) = %d, want = %d", got, want) } - if got, want := len(r.linkEP.WrittenPackets), int(r.Stats().IP.PacketsSent.Value())+1; err != nil && got != want { - t.Errorf("after linkEP error len(result) got %d, want %d", got, want) + if got, want := int(r.Stats().IP.OutgoingPacketErrors.Value()), ft.fragmentCount-ft.allowPackets; got != want { + t.Errorf("got r.Stats().IP.OutgoingPacketErrors.Value() = %d, want = %d", got, want) } }) } @@ -561,8 +821,8 @@ func TestInvalidFragments(t *testing.T) { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ - ipv4.NewProtocol(), + NetworkProtocols: []stack.NetworkProtocolFactory{ + ipv4.NewProtocol, }, }) e := channel.New(0, 1500, linkAddr) @@ -973,8 +1233,8 @@ func TestReceiveFragments(t *testing.T) { t.Run(test.name, func(t *testing.T) { // Setup a stack and endpoint. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, 1280, tcpip.LinkAddress("\xf0\x00")) if err := s.CreateNIC(nicID, e); err != nil { @@ -1049,10 +1309,11 @@ func TestReceiveFragments(t *testing.T) { func TestWriteStats(t *testing.T) { const nPackets = 3 + tests := []struct { name string setup func(*testing.T, *stack.Stack) - linkEP func() stack.LinkEndpoint + allowPackets int expectSent int expectDropped int expectWritten int @@ -1061,7 +1322,7 @@ func TestWriteStats(t *testing.T) { name: "Accept all", // No setup needed, tables accept everything by default. setup: func(*testing.T, *stack.Stack) {}, - linkEP: func() stack.LinkEndpoint { return &limitedEP{nPackets} }, + allowPackets: math.MaxInt32, expectSent: nPackets, expectDropped: 0, expectWritten: nPackets, @@ -1069,7 +1330,7 @@ func TestWriteStats(t *testing.T) { name: "Accept all with error", // No setup needed, tables accept everything by default. setup: func(*testing.T, *stack.Stack) {}, - linkEP: func() stack.LinkEndpoint { return &limitedEP{nPackets - 1} }, + allowPackets: nPackets - 1, expectSent: nPackets - 1, expectDropped: 0, expectWritten: nPackets - 1, @@ -1084,12 +1345,12 @@ func TestWriteStats(t *testing.T) { t.Fatalf("failed to find filter table") } ruleIdx := filter.BuiltinChains[stack.Output] - filter.Rules[ruleIdx].Target = stack.DropTarget{} + filter.Rules[ruleIdx].Target = &stack.DropTarget{} if err := ipt.ReplaceTable(stack.FilterTable, filter, false /* ipv6 */); err != nil { - t.Fatalf("failed to replace table: %v", err) + t.Fatalf("failed to replace table: %s", err) } }, - linkEP: func() stack.LinkEndpoint { return &limitedEP{nPackets} }, + allowPackets: math.MaxInt32, expectSent: 0, expectDropped: nPackets, expectWritten: nPackets, @@ -1106,15 +1367,15 @@ func TestWriteStats(t *testing.T) { } // We'll match and DROP the last packet. ruleIdx := filter.BuiltinChains[stack.Output] - filter.Rules[ruleIdx].Target = stack.DropTarget{} + filter.Rules[ruleIdx].Target = &stack.DropTarget{} filter.Rules[ruleIdx].Matchers = []stack.Matcher{&limitedMatcher{nPackets - 1}} // Make sure the next rule is ACCEPT. - filter.Rules[ruleIdx+1].Target = stack.AcceptTarget{} + filter.Rules[ruleIdx+1].Target = &stack.AcceptTarget{} if err := ipt.ReplaceTable(stack.FilterTable, filter, false /* ipv6 */); err != nil { - t.Fatalf("failed to replace table: %v", err) + t.Fatalf("failed to replace table: %s", err) } }, - linkEP: func() stack.LinkEndpoint { return &limitedEP{nPackets} }, + allowPackets: math.MaxInt32, expectSent: nPackets - 1, expectDropped: 1, expectWritten: nPackets, @@ -1150,7 +1411,8 @@ func TestWriteStats(t *testing.T) { t.Run(writer.name, func(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - rt := buildRoute(t, nil, test.linkEP()) + ep := testutil.NewMockLinkEndpoint(header.IPv4MinimumSize+header.UDPMinimumSize, tcpip.ErrInvalidEndpointState, test.allowPackets) + rt := buildRoute(t, ep) var pkts stack.PacketBufferList for i := 0; i < nPackets; i++ { @@ -1181,117 +1443,255 @@ func TestWriteStats(t *testing.T) { } } -func buildRoute(t *testing.T, packetCollectorErrors []*tcpip.Error, linkEP stack.LinkEndpoint) stack.Route { +func buildRoute(t *testing.T, ep stack.LinkEndpoint) stack.Route { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, }) - s.CreateNIC(1, linkEP) + if err := s.CreateNIC(1, ep); err != nil { + t.Fatalf("CreateNIC(1, _) failed: %s", err) + } const ( src = "\x10\x00\x00\x01" dst = "\x10\x00\x00\x02" ) - s.AddAddress(1, ipv4.ProtocolNumber, src) + if err := s.AddAddress(1, ipv4.ProtocolNumber, src); err != nil { + t.Fatalf("AddAddress(1, %d, %s) failed: %s", ipv4.ProtocolNumber, src, err) + } { - subnet, err := tcpip.NewSubnet(dst, tcpip.AddressMask(header.IPv4Broadcast)) + mask := tcpip.AddressMask(header.IPv4Broadcast) + subnet, err := tcpip.NewSubnet(dst, mask) if err != nil { - t.Fatal(err) + t.Fatalf("NewSubnet(%s, %s) failed: %v", dst, mask, err) } s.SetRouteTable([]tcpip.Route{{ Destination: subnet, NIC: 1, }}) } - rt, err := s.FindRoute(0, src, dst, ipv4.ProtocolNumber, false /* multicastLoop */) + rt, err := s.FindRoute(1, src, dst, ipv4.ProtocolNumber, false /* multicastLoop */) if err != nil { - t.Fatalf("s.FindRoute got %v, want %v", err, nil) + t.Fatalf("FindRoute(1, %s, %s, %d, false) = %s", src, dst, ipv4.ProtocolNumber, err) } return rt } -// limitedEP is a link endpoint that writes up to a certain number of packets -// before returning errors. -type limitedEP struct { +// limitedMatcher is an iptables matcher that matches after a certain number of +// packets are checked against it. +type limitedMatcher struct { limit int } -// MTU implements LinkEndpoint.MTU. -func (*limitedEP) MTU() uint32 { - // Give an MTU that won't cause fragmentation for IPv4+UDP. - return header.IPv4MinimumSize + header.UDPMinimumSize +// Name implements Matcher.Name. +func (*limitedMatcher) Name() string { + return "limitedMatcher" } -// Capabilities implements LinkEndpoint.Capabilities. -func (*limitedEP) Capabilities() stack.LinkEndpointCapabilities { return 0 } +// Match implements Matcher.Match. +func (lm *limitedMatcher) Match(stack.Hook, *stack.PacketBuffer, string) (bool, bool) { + if lm.limit == 0 { + return true, false + } + lm.limit-- + return false, false +} -// MaxHeaderLength implements LinkEndpoint.MaxHeaderLength. -func (*limitedEP) MaxHeaderLength() uint16 { return 0 } +func TestPacketQueing(t *testing.T) { + const nicID = 1 -// LinkAddress implements LinkEndpoint.LinkAddress. -func (*limitedEP) LinkAddress() tcpip.LinkAddress { return "" } + var ( + host1NICLinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x06") + host2NICLinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x09") -// WritePacket implements LinkEndpoint.WritePacket. -func (ep *limitedEP) WritePacket(*stack.Route, *stack.GSO, tcpip.NetworkProtocolNumber, *stack.PacketBuffer) *tcpip.Error { - if ep.limit == 0 { - return tcpip.ErrInvalidEndpointState - } - ep.limit-- - return nil -} + host1IPv4Addr = tcpip.ProtocolAddress{ + Protocol: ipv4.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("192.168.0.1").To4()), + PrefixLen: 24, + }, + } + host2IPv4Addr = tcpip.ProtocolAddress{ + Protocol: ipv4.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("192.168.0.2").To4()), + PrefixLen: 8, + }, + } + ) -// WritePackets implements LinkEndpoint.WritePackets. -func (ep *limitedEP) WritePackets(_ *stack.Route, _ *stack.GSO, pkts stack.PacketBufferList, _ tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - if ep.limit == 0 { - return 0, tcpip.ErrInvalidEndpointState - } - nWritten := ep.limit - if nWritten > pkts.Len() { - nWritten = pkts.Len() - } - ep.limit -= nWritten - return nWritten, nil -} + tests := []struct { + name string + rxPkt func(*channel.Endpoint) + checkResp func(*testing.T, *channel.Endpoint) + }{ + { + name: "ICMP Error", + rxPkt: func(e *channel.Endpoint) { + hdr := buffer.NewPrependable(header.IPv4MinimumSize + header.UDPMinimumSize) + u := header.UDP(hdr.Prepend(header.UDPMinimumSize)) + u.Encode(&header.UDPFields{ + SrcPort: 5555, + DstPort: 80, + Length: header.UDPMinimumSize, + }) + sum := header.PseudoHeaderChecksum(udp.ProtocolNumber, host2IPv4Addr.AddressWithPrefix.Address, host1IPv4Addr.AddressWithPrefix.Address, header.UDPMinimumSize) + sum = header.Checksum(header.UDP([]byte{}), sum) + u.SetChecksum(^u.CalculateChecksum(sum)) + ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize)) + ip.Encode(&header.IPv4Fields{ + IHL: header.IPv4MinimumSize, + TotalLength: header.IPv4MinimumSize + header.UDPMinimumSize, + TTL: ipv4.DefaultTTL, + Protocol: uint8(udp.ProtocolNumber), + SrcAddr: host2IPv4Addr.AddressWithPrefix.Address, + DstAddr: host1IPv4Addr.AddressWithPrefix.Address, + }) + e.InjectInbound(ipv4.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + })) + }, + checkResp: func(t *testing.T, e *channel.Endpoint) { + p, ok := e.ReadContext(context.Background()) + if !ok { + t.Fatalf("timed out waiting for packet") + } + 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) + } + checker.IPv4(t, stack.PayloadSince(p.Pkt.NetworkHeader()), + checker.SrcAddr(host1IPv4Addr.AddressWithPrefix.Address), + checker.DstAddr(host2IPv4Addr.AddressWithPrefix.Address), + checker.ICMPv4( + checker.ICMPv4Type(header.ICMPv4DstUnreachable), + checker.ICMPv4Code(header.ICMPv4PortUnreachable))) + }, + }, -// WriteRawPacket implements LinkEndpoint.WriteRawPacket. -func (ep *limitedEP) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { - if ep.limit == 0 { - return tcpip.ErrInvalidEndpointState + { + name: "Ping", + rxPkt: func(e *channel.Endpoint) { + totalLen := header.IPv4MinimumSize + header.ICMPv4MinimumSize + hdr := buffer.NewPrependable(totalLen) + pkt := header.ICMPv4(hdr.Prepend(header.ICMPv4MinimumSize)) + pkt.SetType(header.ICMPv4Echo) + pkt.SetCode(0) + pkt.SetChecksum(0) + pkt.SetChecksum(^header.Checksum(pkt, 0)) + ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize)) + ip.Encode(&header.IPv4Fields{ + IHL: header.IPv4MinimumSize, + TotalLength: uint16(totalLen), + Protocol: uint8(icmp.ProtocolNumber4), + TTL: ipv4.DefaultTTL, + SrcAddr: host2IPv4Addr.AddressWithPrefix.Address, + DstAddr: host1IPv4Addr.AddressWithPrefix.Address, + }) + e.InjectInbound(header.IPv4ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + })) + }, + checkResp: func(t *testing.T, e *channel.Endpoint) { + p, ok := e.ReadContext(context.Background()) + if !ok { + t.Fatalf("timed out waiting for packet") + } + 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) + } + checker.IPv4(t, stack.PayloadSince(p.Pkt.NetworkHeader()), + checker.SrcAddr(host1IPv4Addr.AddressWithPrefix.Address), + checker.DstAddr(host2IPv4Addr.AddressWithPrefix.Address), + checker.ICMPv4( + checker.ICMPv4Type(header.ICMPv4EchoReply), + checker.ICMPv4Code(header.ICMPv4UnusedCode))) + }, + }, } - ep.limit-- - return nil -} -// Attach implements LinkEndpoint.Attach. -func (*limitedEP) Attach(_ stack.NetworkDispatcher) {} + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e := channel.New(1, header.IPv6MinimumMTU, host1NICLinkAddr) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{arp.NewProtocol, ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, + }) -// IsAttached implements LinkEndpoint.IsAttached. -func (*limitedEP) IsAttached() bool { return false } + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("s.CreateNIC(%d, _): %s", nicID, err) + } + if err := s.AddAddress(nicID, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + t.Fatalf("s.AddAddress(%d, %d, %s): %s", nicID, arp.ProtocolNumber, arp.ProtocolAddress, err) + } + if err := s.AddProtocolAddress(nicID, host1IPv4Addr); err != nil { + t.Fatalf("s.AddProtocolAddress(%d, %#v): %s", nicID, host1IPv4Addr, err) + } -// Wait implements LinkEndpoint.Wait. -func (*limitedEP) Wait() {} + s.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: host1IPv4Addr.AddressWithPrefix.Subnet(), + NIC: nicID, + }, + }) -// ARPHardwareType implements LinkEndpoint.ARPHardwareType. -func (*limitedEP) ARPHardwareType() header.ARPHardwareType { return header.ARPHardwareEther } + // Receive a packet to trigger link resolution before a response is sent. + test.rxPkt(e) -// AddHeader implements LinkEndpoint.AddHeader. -func (*limitedEP) AddHeader(_, _ tcpip.LinkAddress, _ tcpip.NetworkProtocolNumber, _ *stack.PacketBuffer) { -} + // Wait for a ARP request since link address resolution should be + // performed. + { + p, ok := e.ReadContext(context.Background()) + if !ok { + t.Fatalf("timed out waiting for packet") + } + 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) + } + rep := header.ARP(p.Pkt.NetworkHeader().View()) + if got := rep.Op(); got != header.ARPRequest { + t.Errorf("got Op() = %d, want = %d", got, header.ARPRequest) + } + if got := tcpip.LinkAddress(rep.HardwareAddressSender()); got != host1NICLinkAddr { + t.Errorf("got HardwareAddressSender = %s, want = %s", got, host1NICLinkAddr) + } + if got := tcpip.Address(rep.ProtocolAddressSender()); got != host1IPv4Addr.AddressWithPrefix.Address { + t.Errorf("got ProtocolAddressSender = %s, want = %s", got, host1IPv4Addr.AddressWithPrefix.Address) + } + if got := tcpip.Address(rep.ProtocolAddressTarget()); got != host2IPv4Addr.AddressWithPrefix.Address { + t.Errorf("got ProtocolAddressTarget = %s, want = %s", got, host2IPv4Addr.AddressWithPrefix.Address) + } + } -// limitedMatcher is an iptables matcher that matches after a certain number of -// packets are checked against it. -type limitedMatcher struct { - limit int -} + // Send an ARP reply to complete link address resolution. + { + hdr := buffer.View(make([]byte, header.ARPSize)) + packet := header.ARP(hdr) + packet.SetIPv4OverEthernet() + packet.SetOp(header.ARPReply) + copy(packet.HardwareAddressSender(), host2NICLinkAddr) + copy(packet.ProtocolAddressSender(), host2IPv4Addr.AddressWithPrefix.Address) + copy(packet.HardwareAddressTarget(), host1NICLinkAddr) + copy(packet.ProtocolAddressTarget(), host1IPv4Addr.AddressWithPrefix.Address) + e.InjectInbound(arp.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.ToVectorisedView(), + })) + } -// Name implements Matcher.Name. -func (*limitedMatcher) Name() string { - return "limitedMatcher" -} + // Expect the response now that the link address has resolved. + test.checkResp(t, e) -// Match implements Matcher.Match. -func (lm *limitedMatcher) Match(stack.Hook, *stack.PacketBuffer, string) (bool, bool) { - if lm.limit == 0 { - return true, false + // Since link resolution was already performed, it shouldn't be performed + // again. + test.rxPkt(e) + test.checkResp(t, e) + }) } - lm.limit-- - return false, false } diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index cd5fe3ea8..a30437f02 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -5,16 +5,20 @@ package(licenses = ["notice"]) go_library( name = "ipv6", srcs = [ + "dhcpv6configurationfromndpra_string.go", "icmp.go", "ipv6.go", + "ndp.go", ], visibility = ["//visibility:public"], deps = [ + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/header/parse", "//pkg/tcpip/network/fragmentation", + "//pkg/tcpip/network/hash", "//pkg/tcpip/stack", ], ) @@ -35,8 +39,10 @@ go_test( "//pkg/tcpip/header", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/sniffer", + "//pkg/tcpip/network/testutil", "//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/stack/dhcpv6configurationfromndpra_string.go b/pkg/tcpip/network/ipv6/dhcpv6configurationfromndpra_string.go index d199ded6a..09ba133b1 100644 --- a/pkg/tcpip/stack/dhcpv6configurationfromndpra_string.go +++ b/pkg/tcpip/network/ipv6/dhcpv6configurationfromndpra_string.go @@ -14,7 +14,7 @@ // Code generated by "stringer -type DHCPv6ConfigurationFromNDPRA"; DO NOT EDIT. -package stack +package ipv6 import "strconv" diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index 7430b8fcd..7be35c78b 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -15,6 +15,8 @@ package ipv6 import ( + "fmt" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -39,7 +41,7 @@ func (e *endpoint) handleControl(typ stack.ControlType, extra uint32, pkt *stack // Drop packet if it doesn't have the basic IPv6 header or if the // original source address doesn't match an address we own. src := hdr.SourceAddress() - if e.stack.CheckLocalAddress(e.NICID(), ProtocolNumber, src) == 0 { + if e.protocol.stack.CheckLocalAddress(e.nic.ID(), ProtocolNumber, src) == 0 { return } @@ -207,14 +209,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme return } - s := r.Stack() - if isTentative, err := s.IsAddrTentative(e.nicID, targetAddr); err != nil { - // We will only get an error if the NIC is unrecognized, which should not - // happen. For now, drop this packet. - // - // TODO(b/141002840): Handle this better? - return - } else if isTentative { + if e.hasTentativeAddr(targetAddr) { // If the target address is tentative and the source of the packet is a // unicast (specified) address, then the source of the packet is // attempting to perform address resolution on the target. In this case, @@ -227,7 +222,20 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // stack know so it can handle such a scenario and do nothing further with // the NS. if r.RemoteAddress == header.IPv6Any { - s.DupTentativeAddrDetected(e.nicID, targetAddr) + // We would get an error if the address no longer exists or the address + // is no longer tentative (DAD resolved between the call to + // hasTentativeAddr and this point). Both of these are valid scenarios: + // 1) An address may be removed at any time. + // 2) As per RFC 4862 section 5.4, DAD is not a perfect: + // "Note that the method for detecting duplicates + // is not completely reliable, and it is possible that duplicate + // addresses will still exist" + // + // TODO(gvisor.dev/issue/4046): Handle the scenario when a duplicate + // address is detected for an assigned address. + if err := e.dupTentativeAddrDetected(targetAddr); err != nil && err != tcpip.ErrBadAddress && err != tcpip.ErrInvalidEndpointState { + panic(fmt.Sprintf("unexpected error handling duplicate tentative address: %s", err)) + } } // Do not handle neighbor solicitations targeted to an address that is @@ -240,7 +248,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // section 5.4.3. // Is the NS targeting us? - if s.CheckLocalAddress(e.nicID, ProtocolNumber, targetAddr) == 0 { + if r.Stack().CheckLocalAddress(e.nic.ID(), ProtocolNumber, targetAddr) == 0 { return } @@ -275,7 +283,18 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme } else if e.nud != nil { e.nud.HandleProbe(r.RemoteAddress, r.LocalAddress, header.IPv6ProtocolNumber, sourceLinkAddr, e.protocol) } else { - e.linkAddrCache.AddLinkAddress(e.nicID, r.RemoteAddress, sourceLinkAddr) + e.linkAddrCache.AddLinkAddress(e.nic.ID(), r.RemoteAddress, sourceLinkAddr) + } + + // As per RFC 4861 section 7.1.1: + // A node MUST silently discard any received Neighbor Solicitation + // messages that do not satisfy all of the following validity checks: + // ... + // - If the IP source address is the unspecified address, the IP + // destination address is a solicited-node multicast address. + if unspecifiedSource && !header.IsSolicitedNodeAddr(r.LocalAddress) { + received.Invalid.Increment() + return } // ICMPv6 Neighbor Solicit messages are always sent to @@ -318,6 +337,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme ReserveHeaderBytes: int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length()), }) packet := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6NeighborAdvertSize)) + pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber packet.SetType(header.ICMPv6NeighborAdvert) na := header.NDPNeighborAdvert(packet.NDPPayload()) na.SetSolicitedFlag(solicited) @@ -352,20 +372,26 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // NDP datagrams are very small and ToView() will not incur allocations. na := header.NDPNeighborAdvert(payload.ToView()) targetAddr := na.TargetAddress() - s := r.Stack() - - if isTentative, err := s.IsAddrTentative(e.nicID, targetAddr); err != nil { - // We will only get an error if the NIC is unrecognized, which should not - // happen. For now short-circuit this packet. - // - // TODO(b/141002840): Handle this better? - return - } else if isTentative { + if e.hasTentativeAddr(targetAddr) { // We just got an NA from a node that owns an address we are performing // DAD on, implying the address is not unique. In this case we let the // stack know so it can handle such a scenario and do nothing furthur with // the NDP NA. - s.DupTentativeAddrDetected(e.nicID, targetAddr) + // + // We would get an error if the address no longer exists or the address + // is no longer tentative (DAD resolved between the call to + // hasTentativeAddr and this point). Both of these are valid scenarios: + // 1) An address may be removed at any time. + // 2) As per RFC 4862 section 5.4, DAD is not a perfect: + // "Note that the method for detecting duplicates + // is not completely reliable, and it is possible that duplicate + // addresses will still exist" + // + // TODO(gvisor.dev/issue/4046): Handle the scenario when a duplicate + // address is detected for an assigned address. + if err := e.dupTentativeAddrDetected(targetAddr); err != nil && err != tcpip.ErrBadAddress && err != tcpip.ErrInvalidEndpointState { + panic(fmt.Sprintf("unexpected error handling duplicate tentative address: %s", err)) + } return } @@ -395,7 +421,7 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme // address cache with the link address for the target of the message. if len(targetLinkAddr) != 0 { if e.nud == nil { - e.linkAddrCache.AddLinkAddress(e.nicID, targetAddr, targetLinkAddr) + e.linkAddrCache.AddLinkAddress(e.nic.ID(), targetAddr, targetLinkAddr) return } @@ -414,8 +440,6 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme return } - remoteLinkAddr := r.RemoteLinkAddress - // As per RFC 4291 section 2.7, multicast addresses must not be used as // source addresses in IPv6 packets. localAddr := r.LocalAddress @@ -423,21 +447,19 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme localAddr = "" } - r, err := r.Stack().FindRoute(e.NICID(), localAddr, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) + r, err := r.Stack().FindRoute(e.nic.ID(), localAddr, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) if err != nil { // If we cannot find a route to the destination, silently drop the packet. return } defer r.Release() - // Use the link address from the source of the original packet. - r.ResolveWith(remoteLinkAddr) - replyPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(r.MaxHeaderLength()) + header.ICMPv6EchoMinimumSize, Data: pkt.Data, }) packet := header.ICMPv6(replyPkt.TransportHeader().Push(header.ICMPv6EchoMinimumSize)) + pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber copy(packet, icmpHdr) packet.SetType(header.ICMPv6EchoReply) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, pkt.Data)) @@ -566,9 +588,9 @@ func (e *endpoint) handleICMP(r *stack.Route, pkt *stack.PacketBuffer, hasFragme e.nud.HandleProbe(routerAddr, r.LocalAddress, header.IPv6ProtocolNumber, sourceLinkAddr, e.protocol) } - // Tell the NIC to handle the RA. - stack := r.Stack() - stack.HandleNDPRA(e.nicID, routerAddr, ra) + e.mu.Lock() + e.mu.ndp.handleRA(routerAddr, ra) + e.mu.Unlock() case header.ICMPv6RedirectMsg: // TODO(gvisor.dev/issue/2285): Call `e.nud.HandleProbe` after validating @@ -619,24 +641,28 @@ func (*protocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber { // LinkAddressRequest implements stack.LinkAddressResolver. func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAddr tcpip.LinkAddress, linkEP stack.LinkEndpoint) *tcpip.Error { - snaddr := header.SolicitedNodeAddr(addr) - // TODO(b/148672031): Use stack.FindRoute instead of manually creating the // route here. Note, we would need the nicID to do this properly so the right // NIC (associated to linkEP) is used to send the NDP NS message. - r := &stack.Route{ + r := stack.Route{ LocalAddress: localAddr, - RemoteAddress: snaddr, + RemoteAddress: addr, RemoteLinkAddress: remoteLinkAddr, } + + // If a remote address is not already known, then send a multicast + // solicitation since multicast addresses have a static mapping to link + // addresses. if len(r.RemoteLinkAddress) == 0 { - r.RemoteLinkAddress = header.EthernetAddressFromMulticastIPv6Address(snaddr) + r.RemoteAddress = header.SolicitedNodeAddr(addr) + r.RemoteLinkAddress = header.EthernetAddressFromMulticastIPv6Address(r.RemoteAddress) } pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(linkEP.MaxHeaderLength()) + header.IPv6MinimumSize + header.ICMPv6NeighborAdvertSize, }) icmpHdr := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6NeighborAdvertSize)) + pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber icmpHdr.SetType(header.ICMPv6NeighborSolicit) copy(icmpHdr[icmpV6OptOffset-len(addr):], addr) icmpHdr[icmpV6OptOffset] = ndpOptSrcLinkAddr @@ -655,7 +681,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, remoteLinkAdd }) // TODO(stijlist): count this in ICMP stats. - return linkEP.WritePacket(r, nil /* gso */, ProtocolNumber, pkt) + return linkEP.WritePacket(&r, nil /* gso */, ProtocolNumber, pkt) } // ResolveStaticAddress implements stack.LinkAddressResolver. @@ -665,3 +691,179 @@ func (*protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAddress, bo } return tcpip.LinkAddress([]byte(nil)), false } + +// ======= ICMP Error packet generation ========= + +// icmpReason is a marker interface for IPv6 specific ICMP errors. +type icmpReason interface { + isICMPReason() +} + +// icmpReasonParameterProblem is an error during processing of extension headers +// or the fixed header defined in RFC 4443 section 3.4. +type icmpReasonParameterProblem struct { + code header.ICMPv6Code + + // respondToMulticast indicates that we are sending a packet that falls under + // the exception outlined by RFC 4443 section 2.4 point e.3 exception 2: + // + // (e.3) A packet destined to an IPv6 multicast address. (There are + // two exceptions to this rule: (1) the Packet Too Big Message + // (Section 3.2) to allow Path MTU discovery to work for IPv6 + // multicast, and (2) the Parameter Problem Message, Code 2 + // (Section 3.4) reporting an unrecognized IPv6 option (see + // Section 4.2 of [IPv6]) that has the Option Type highest- + // order two bits set to 10). + respondToMulticast bool + + // pointer is defined in the RFC 4443 setion 3.4 which reads: + // + // Pointer Identifies the octet offset within the invoking packet + // where the error was detected. + // + // The pointer will point beyond the end of the ICMPv6 + // packet if the field in error is beyond what can fit + // in the maximum size of an ICMPv6 error message. + pointer uint32 +} + +func (*icmpReasonParameterProblem) isICMPReason() {} + +// icmpReasonPortUnreachable is an error where the transport protocol has no +// listener and no alternative means to inform the sender. +type icmpReasonPortUnreachable struct{} + +func (*icmpReasonPortUnreachable) isICMPReason() {} + +// returnError takes an error descriptor and generates the appropriate ICMP +// error packet for IPv6 and sends it. +func (p *protocol) returnError(r *stack.Route, reason icmpReason, pkt *stack.PacketBuffer) *tcpip.Error { + // Only send ICMP error if the address is not a multicast v6 + // address and the source is not the unspecified address. + // + // There are exceptions to this rule. + // See: point e.3) RFC 4443 section-2.4 + // + // (e) An ICMPv6 error message MUST NOT be originated as a result of + // receiving the following: + // + // (e.1) An ICMPv6 error message. + // + // (e.2) An ICMPv6 redirect message [IPv6-DISC]. + // + // (e.3) A packet destined to an IPv6 multicast address. (There are + // two exceptions to this rule: (1) the Packet Too Big Message + // (Section 3.2) to allow Path MTU discovery to work for IPv6 + // multicast, and (2) the Parameter Problem Message, Code 2 + // (Section 3.4) reporting an unrecognized IPv6 option (see + // Section 4.2 of [IPv6]) that has the Option Type highest- + // order two bits set to 10). + // + var allowResponseToMulticast bool + if reason, ok := reason.(*icmpReasonParameterProblem); ok { + allowResponseToMulticast = reason.respondToMulticast + } + + if (!allowResponseToMulticast && header.IsV6MulticastAddress(r.LocalAddress)) || r.RemoteAddress == header.IPv6Any { + return nil + } + + // Even if we were able to receive a packet from some remote, we may not have + // a route to it - the remote may be blocked via routing rules. We must always + // consult our routing table and find a route to the remote before sending any + // packet. + route, err := p.stack.FindRoute(r.NICID(), r.LocalAddress, r.RemoteAddress, ProtocolNumber, false /* multicastLoop */) + if err != nil { + return err + } + defer route.Release() + // From this point on, the incoming route should no longer be used; route + // must be used to send the ICMP error. + r = nil + + stats := p.stack.Stats().ICMP + sent := stats.V6PacketsSent + if !p.stack.AllowICMPMessage() { + sent.RateLimited.Increment() + return nil + } + + network, transport := pkt.NetworkHeader().View(), pkt.TransportHeader().View() + + if pkt.TransportProtocolNumber == header.ICMPv6ProtocolNumber { + // TODO(gvisor.dev/issues/3810): Sort this out when ICMP headers are stored. + // Unfortunately at this time ICMP Packets do not have a transport + // header separated out. It is in the Data part so we need to + // separate it out now. We will just pretend it is a minimal length + // ICMP packet as we don't really care if any later bits of a + // larger ICMP packet are in the header view or in the Data view. + transport, ok := pkt.TransportHeader().Consume(header.ICMPv6MinimumSize) + if !ok { + return nil + } + typ := header.ICMPv6(transport).Type() + if typ.IsErrorType() || typ == header.ICMPv6RedirectMsg { + return nil + } + } + + // As per RFC 4443 section 2.4 + // + // (c) Every ICMPv6 error message (type < 128) MUST include + // as much of the IPv6 offending (invoking) packet (the + // packet that caused the error) as possible without making + // the error message packet exceed the minimum IPv6 MTU + // [IPv6]. + mtu := int(route.MTU()) + if mtu > header.IPv6MinimumMTU { + mtu = header.IPv6MinimumMTU + } + headerLen := int(route.MaxHeaderLength()) + header.ICMPv6ErrorHeaderSize + available := int(mtu) - headerLen + if available < header.IPv6MinimumSize { + return nil + } + payloadLen := network.Size() + transport.Size() + pkt.Data.Size() + if payloadLen > available { + payloadLen = available + } + payload := buffer.NewVectorisedView(pkt.Size(), pkt.Views()) + payload.CapLength(payloadLen) + + newPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ + ReserveHeaderBytes: headerLen, + Data: payload, + }) + newPkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber + + icmpHdr := header.ICMPv6(newPkt.TransportHeader().Push(header.ICMPv6DstUnreachableMinimumSize)) + var counter *tcpip.StatCounter + switch reason := reason.(type) { + case *icmpReasonParameterProblem: + icmpHdr.SetType(header.ICMPv6ParamProblem) + icmpHdr.SetCode(reason.code) + icmpHdr.SetTypeSpecific(reason.pointer) + counter = sent.ParamProblem + case *icmpReasonPortUnreachable: + icmpHdr.SetType(header.ICMPv6DstUnreachable) + icmpHdr.SetCode(header.ICMPv6PortUnreachable) + counter = sent.DstUnreachable + default: + panic(fmt.Sprintf("unsupported ICMP type %T", reason)) + } + icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, route.LocalAddress, route.RemoteAddress, newPkt.Data)) + if err := route.WritePacket( + nil, /* gso */ + stack.NetworkHeaderParams{ + Protocol: header.ICMPv6ProtocolNumber, + TTL: route.DefaultTTL(), + TOS: stack.DefaultTOS, + }, + newPkt, + ); err != nil { + sent.Dropped.Increment() + return err + } + counter.Increment() + return nil +} diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 0f50bfb8e..3affcc4e4 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -16,17 +16,21 @@ package ipv6 import ( "context" + "net" "reflect" "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/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "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" ) @@ -39,6 +43,9 @@ const ( defaultChannelSize = 1 defaultMTU = 65536 + + // Extra time to use when waiting for an async event to occur. + defaultAsyncPositiveEventTimeout = 30 * time.Second ) var ( @@ -50,6 +57,10 @@ type stubLinkEndpoint struct { stack.LinkEndpoint } +func (*stubLinkEndpoint) MTU() uint32 { + return defaultMTU +} + func (*stubLinkEndpoint) Capabilities() stack.LinkEndpointCapabilities { // Indicate that resolution for link layer addresses is required to send // packets over this link. This is needed so the NIC knows to allocate a @@ -75,7 +86,8 @@ type stubDispatcher struct { stack.TransportDispatcher } -func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, *stack.PacketBuffer) { +func (*stubDispatcher) DeliverTransportPacket(*stack.Route, tcpip.TransportProtocolNumber, *stack.PacketBuffer) stack.TransportPacketDisposition { + return stack.TransportPacketHandled } type stubLinkAddressCache struct { @@ -102,6 +114,28 @@ func (*stubNUDHandler) HandleConfirmation(addr tcpip.Address, linkAddr tcpip.Lin func (*stubNUDHandler) HandleUpperLevelConfirmation(addr tcpip.Address) { } +var _ stack.NetworkInterface = (*testInterface)(nil) + +type testInterface struct { + stack.NetworkLinkEndpoint +} + +func (*testInterface) ID() tcpip.NICID { + return 0 +} + +func (*testInterface) IsLoopback() bool { + return false +} + +func (*testInterface) Name() string { + return "" +} + +func (*testInterface) Enabled() bool { + return true +} + func TestICMPCounts(t *testing.T) { tests := []struct { name string @@ -120,8 +154,8 @@ func TestICMPCounts(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, UseNeighborCache: test.useNeighborCache, }) { @@ -149,9 +183,13 @@ func TestICMPCounts(t *testing.T) { if netProto == nil { t.Fatalf("cannot find protocol instance for network protocol %d", ProtocolNumber) } - ep := netProto.NewEndpoint(0, &stubLinkAddressCache{}, &stubNUDHandler{}, &stubDispatcher{}, nil, s) + ep := netProto.NewEndpoint(&testInterface{}, &stubLinkAddressCache{}, &stubNUDHandler{}, &stubDispatcher{}) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + r, err := s.FindRoute(nicID, lladdr0, lladdr1, ProtocolNumber, false /* multicastLoop */) if err != nil { t.Fatalf("FindRoute(%d, %s, %s, _, false) = (_, %s), want = (_, nil)", nicID, lladdr0, lladdr1, err) @@ -258,8 +296,8 @@ func TestICMPCounts(t *testing.T) { func TestICMPCountsWithNeighborCache(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, UseNeighborCache: true, }) { @@ -287,9 +325,13 @@ func TestICMPCountsWithNeighborCache(t *testing.T) { if netProto == nil { t.Fatalf("cannot find protocol instance for network protocol %d", ProtocolNumber) } - ep := netProto.NewEndpoint(0, nil, &stubNUDHandler{}, &stubDispatcher{}, nil, s) + ep := netProto.NewEndpoint(&testInterface{}, nil, &stubNUDHandler{}, &stubDispatcher{}) defer ep.Close() + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + r, err := s.FindRoute(nicID, lladdr0, lladdr1, ProtocolNumber, false /* multicastLoop */) if err != nil { t.Fatalf("FindRoute(%d, %s, %s, _, false) = (_, %s), want = (_, nil)", nicID, lladdr0, lladdr1, err) @@ -423,12 +465,12 @@ func (e endpointWithResolutionCapability) Capabilities() stack.LinkEndpointCapab func newTestContext(t *testing.T) *testContext { c := &testContext{ s0: stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, }), s1: stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, }), } @@ -723,7 +765,7 @@ func TestICMPChecksumValidationSimple(t *testing.T) { e.LinkEPCapabilities |= stack.CapabilityResolutionRequired s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: test.useNeighborCache, }) if isRouter { @@ -919,7 +961,7 @@ func TestICMPChecksumValidationWithPayload(t *testing.T) { t.Run(typ.name, func(t *testing.T) { e := channel.New(10, 1280, linkAddr0) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(_, _) = %s", err) @@ -1097,7 +1139,7 @@ func TestICMPChecksumValidationWithPayloadMultipleViews(t *testing.T) { t.Run(typ.name, func(t *testing.T) { e := channel.New(10, 1280, linkAddr0) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -1186,24 +1228,30 @@ func TestLinkAddressRequest(t *testing.T) { mcaddr := header.EthernetAddressFromMulticastIPv6Address(snaddr) tests := []struct { - name string - remoteLinkAddr tcpip.LinkAddress - expectLinkAddr tcpip.LinkAddress + name string + remoteLinkAddr tcpip.LinkAddress + expectedLinkAddr tcpip.LinkAddress + expectedAddr tcpip.Address }{ { - name: "Unicast", - remoteLinkAddr: linkAddr1, - expectLinkAddr: linkAddr1, + name: "Unicast", + remoteLinkAddr: linkAddr1, + expectedLinkAddr: linkAddr1, + expectedAddr: lladdr0, }, { - name: "Multicast", - remoteLinkAddr: "", - expectLinkAddr: mcaddr, + name: "Multicast", + remoteLinkAddr: "", + expectedLinkAddr: mcaddr, + expectedAddr: snaddr, }, } for _, test := range tests { - p := NewProtocol() + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + }) + p := s.NetworkProtocolInstance(ProtocolNumber) linkRes, ok := p.(stack.LinkAddressResolver) if !ok { t.Fatalf("expected IPv6 protocol to implement stack.LinkAddressResolver") @@ -1218,9 +1266,229 @@ func TestLinkAddressRequest(t *testing.T) { if !ok { t.Fatal("expected to send a link address request") } + if pkt.Route.RemoteLinkAddress != test.expectedLinkAddr { + t.Errorf("got pkt.Route.RemoteLinkAddress = %s, want = %s", pkt.Route.RemoteLinkAddress, test.expectedLinkAddr) + } + if pkt.Route.RemoteAddress != test.expectedAddr { + t.Errorf("got pkt.Route.RemoteAddress = %s, want = %s", pkt.Route.RemoteAddress, test.expectedAddr) + } + if pkt.Route.LocalAddress != lladdr1 { + t.Errorf("got pkt.Route.LocalAddress = %s, want = %s", pkt.Route.LocalAddress, lladdr1) + } + checker.IPv6(t, stack.PayloadSince(pkt.Pkt.NetworkHeader()), + checker.SrcAddr(lladdr1), + checker.DstAddr(test.expectedAddr), + checker.TTL(header.NDPHopLimit), + checker.NDPNS( + checker.NDPNSTargetAddress(lladdr0), + checker.NDPNSOptions([]header.NDPOption{header.NDPSourceLinkLayerAddressOption(linkAddr0)}), + )) + } +} - if got, want := pkt.Route.RemoteLinkAddress, test.expectLinkAddr; got != want { - t.Errorf("got pkt.Route.RemoteLinkAddress = %s, want = %s", got, want) +func TestPacketQueing(t *testing.T) { + const nicID = 1 + + var ( + host1NICLinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x06") + host2NICLinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x09") + + host1IPv6Addr = tcpip.ProtocolAddress{ + Protocol: ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("a::1").To16()), + PrefixLen: 64, + }, } + host2IPv6Addr = tcpip.ProtocolAddress{ + Protocol: ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("a::2").To16()), + PrefixLen: 64, + }, + } + ) + + tests := []struct { + name string + rxPkt func(*channel.Endpoint) + checkResp func(*testing.T, *channel.Endpoint) + }{ + { + name: "ICMP Error", + rxPkt: func(e *channel.Endpoint) { + hdr := buffer.NewPrependable(header.IPv6MinimumSize + header.UDPMinimumSize) + u := header.UDP(hdr.Prepend(header.UDPMinimumSize)) + u.Encode(&header.UDPFields{ + SrcPort: 5555, + DstPort: 80, + Length: header.UDPMinimumSize, + }) + sum := header.PseudoHeaderChecksum(udp.ProtocolNumber, host2IPv6Addr.AddressWithPrefix.Address, host1IPv6Addr.AddressWithPrefix.Address, header.UDPMinimumSize) + sum = header.Checksum(header.UDP([]byte{}), sum) + u.SetChecksum(^u.CalculateChecksum(sum)) + payloadLength := hdr.UsedLength() + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(udp.ProtocolNumber), + HopLimit: DefaultTTL, + SrcAddr: host2IPv6Addr.AddressWithPrefix.Address, + DstAddr: host1IPv6Addr.AddressWithPrefix.Address, + }) + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + })) + }, + checkResp: func(t *testing.T, e *channel.Endpoint) { + p, ok := e.ReadContext(context.Background()) + if !ok { + t.Fatalf("timed out waiting for packet") + } + 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) + } + checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), + checker.SrcAddr(host1IPv6Addr.AddressWithPrefix.Address), + checker.DstAddr(host2IPv6Addr.AddressWithPrefix.Address), + checker.ICMPv6( + checker.ICMPv6Type(header.ICMPv6DstUnreachable), + checker.ICMPv6Code(header.ICMPv6PortUnreachable))) + }, + }, + + { + name: "Ping", + rxPkt: func(e *channel.Endpoint) { + totalLen := header.IPv6MinimumSize + header.ICMPv6MinimumSize + hdr := buffer.NewPrependable(totalLen) + pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6MinimumSize)) + pkt.SetType(header.ICMPv6EchoRequest) + pkt.SetCode(0) + pkt.SetChecksum(0) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, host2IPv6Addr.AddressWithPrefix.Address, host1IPv6Addr.AddressWithPrefix.Address, buffer.VectorisedView{})) + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: header.ICMPv6MinimumSize, + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: DefaultTTL, + SrcAddr: host2IPv6Addr.AddressWithPrefix.Address, + DstAddr: host1IPv6Addr.AddressWithPrefix.Address, + }) + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + })) + }, + checkResp: func(t *testing.T, e *channel.Endpoint) { + p, ok := e.ReadContext(context.Background()) + if !ok { + t.Fatalf("timed out waiting for packet") + } + 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) + } + checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), + checker.SrcAddr(host1IPv6Addr.AddressWithPrefix.Address), + checker.DstAddr(host2IPv6Addr.AddressWithPrefix.Address), + checker.ICMPv6( + checker.ICMPv6Type(header.ICMPv6EchoReply), + checker.ICMPv6Code(header.ICMPv6UnusedCode))) + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + + e := channel.New(1, header.IPv6MinimumMTU, host1NICLinkAddr) + e.LinkEPCapabilities |= stack.CapabilityResolutionRequired + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, + }) + + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("s.CreateNIC(%d, _): %s", nicID, err) + } + if err := s.AddProtocolAddress(nicID, host1IPv6Addr); err != nil { + t.Fatalf("s.AddProtocolAddress(%d, %#v): %s", nicID, host1IPv6Addr, err) + } + + s.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: host1IPv6Addr.AddressWithPrefix.Subnet(), + NIC: nicID, + }, + }) + + // Receive a packet to trigger link resolution before a response is sent. + test.rxPkt(e) + + // Wait for a neighbor solicitation since link address resolution should + // be performed. + { + p, ok := e.ReadContext(context.Background()) + if !ok { + t.Fatalf("timed out waiting for packet") + } + if p.Proto != ProtocolNumber { + 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) + } + checker.IPv6(t, stack.PayloadSince(p.Pkt.NetworkHeader()), + checker.SrcAddr(host1IPv6Addr.AddressWithPrefix.Address), + checker.DstAddr(snmc), + checker.TTL(header.NDPHopLimit), + checker.NDPNS( + checker.NDPNSTargetAddress(host2IPv6Addr.AddressWithPrefix.Address), + checker.NDPNSOptions([]header.NDPOption{header.NDPSourceLinkLayerAddressOption(host1NICLinkAddr)}), + )) + } + + // Send a neighbor advertisement to complete link address resolution. + { + naSize := header.ICMPv6NeighborAdvertMinimumSize + header.NDPLinkLayerAddressSize + hdr := buffer.NewPrependable(header.IPv6MinimumSize + naSize) + pkt := header.ICMPv6(hdr.Prepend(naSize)) + pkt.SetType(header.ICMPv6NeighborAdvert) + na := header.NDPNeighborAdvert(pkt.NDPPayload()) + na.SetSolicitedFlag(true) + na.SetOverrideFlag(true) + na.SetTargetAddress(host2IPv6Addr.AddressWithPrefix.Address) + na.Options().Serialize(header.NDPOptionsSerializer{ + header.NDPTargetLinkLayerAddressOption(host2NICLinkAddr), + }) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, host2IPv6Addr.AddressWithPrefix.Address, host1IPv6Addr.AddressWithPrefix.Address, buffer.VectorisedView{})) + payloadLength := hdr.UsedLength() + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: header.NDPHopLimit, + SrcAddr: host2IPv6Addr.AddressWithPrefix.Address, + DstAddr: host1IPv6Addr.AddressWithPrefix.Address, + }) + e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ + Data: hdr.View().ToVectorisedView(), + })) + } + + // Expect the response now that the link address has resolved. + test.checkResp(t, e) + + // Since link resolution was already performed, it shouldn't be performed + // again. + test.rxPkt(e) + test.checkResp(t, e) + }) } } diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index ee64d92d8..2bd8f4ece 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.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. @@ -12,27 +12,37 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package ipv6 contains the implementation of the ipv6 network protocol. To use -// it in the networking stack, this package must be added to the project, and -// activated on the stack by passing ipv6.NewProtocol() as one of the network -// protocols when calling stack.New(). Then endpoints can be created by passing -// ipv6.ProtocolNumber as the network protocol number when calling -// Stack.NewEndpoint(). +// Package ipv6 contains the implementation of the ipv6 network protocol. package ipv6 import ( + "encoding/binary" "fmt" + "hash/fnv" + "sort" "sync/atomic" + "time" + "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/header/parse" "gvisor.dev/gvisor/pkg/tcpip/network/fragmentation" + "gvisor.dev/gvisor/pkg/tcpip/network/hash" "gvisor.dev/gvisor/pkg/tcpip/stack" ) const ( + // 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. + // + // Linux also uses 60 seconds for reassembly timeout: + // https://github.com/torvalds/linux/blob/47ec5303d73ea344e84f46660fff693c57641386/include/net/ipv6.h#L456 + reassembleTimeout = 60 * time.Second + // ProtocolNumber is the ipv6 protocol number. ProtocolNumber = header.IPv6ProtocolNumber @@ -43,16 +53,306 @@ const ( // DefaultTTL is the default hop limit for IPv6 Packets egressed by // Netstack. DefaultTTL = 64 + + // buckets for fragment identifiers + buckets = 2048 ) +var _ stack.GroupAddressableEndpoint = (*endpoint)(nil) +var _ stack.AddressableEndpoint = (*endpoint)(nil) +var _ stack.NetworkEndpoint = (*endpoint)(nil) +var _ stack.NDPEndpoint = (*endpoint)(nil) +var _ NDPEndpoint = (*endpoint)(nil) + type endpoint struct { - nicID tcpip.NICID - linkEP stack.LinkEndpoint + nic stack.NetworkInterface linkAddrCache stack.LinkAddressCache nud stack.NUDHandler dispatcher stack.TransportDispatcher protocol *protocol stack *stack.Stack + + // enabled is set to 1 when the endpoint is enabled and 0 when it is + // disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + + mu struct { + sync.RWMutex + + addressableEndpointState stack.AddressableEndpointState + ndp ndpState + } +} + +// NICNameFromID is a function that returns a stable name for the specified NIC, +// even if different NIC IDs are used to refer to the same NIC in different +// program runs. It is used when generating opaque interface identifiers (IIDs). +// If the NIC was created with a name, it is passed to NICNameFromID. +// +// NICNameFromID SHOULD return unique NIC names so unique opaque IIDs are +// generated for the same prefix on differnt NICs. +type NICNameFromID func(tcpip.NICID, string) string + +// OpaqueInterfaceIdentifierOptions holds the options related to the generation +// of opaque interface indentifiers (IIDs) as defined by RFC 7217. +type OpaqueInterfaceIdentifierOptions struct { + // NICNameFromID is a function that returns a stable name for a specified NIC, + // even if the NIC ID changes over time. + // + // Must be specified to generate the opaque IID. + NICNameFromID NICNameFromID + + // SecretKey is a pseudo-random number used as the secret key when generating + // opaque IIDs as defined by RFC 7217. The key SHOULD be at least + // header.OpaqueIIDSecretKeyMinBytes bytes and MUST follow minimum randomness + // requirements for security as outlined by RFC 4086. SecretKey MUST NOT + // change between program runs, unless explicitly changed. + // + // OpaqueInterfaceIdentifierOptions takes ownership of SecretKey. SecretKey + // MUST NOT be modified after Stack is created. + // + // May be nil, but a nil value is highly discouraged to maintain + // some level of randomness between nodes. + SecretKey []byte +} + +// InvalidateDefaultRouter implements stack.NDPEndpoint. +func (e *endpoint) InvalidateDefaultRouter(rtr tcpip.Address) { + e.mu.Lock() + defer e.mu.Unlock() + e.mu.ndp.invalidateDefaultRouter(rtr) +} + +// SetNDPConfigurations implements NDPEndpoint. +func (e *endpoint) SetNDPConfigurations(c NDPConfigurations) { + c.validate() + e.mu.Lock() + defer e.mu.Unlock() + e.mu.ndp.configs = c +} + +// hasTentativeAddr returns true if addr is tentative on e. +func (e *endpoint) hasTentativeAddr(addr tcpip.Address) bool { + e.mu.RLock() + addressEndpoint := e.getAddressRLocked(addr) + e.mu.RUnlock() + return addressEndpoint != nil && addressEndpoint.GetKind() == stack.PermanentTentative +} + +// dupTentativeAddrDetected attempts to inform e that a tentative addr is a +// duplicate on a link. +// +// dupTentativeAddrDetected removes the tentative address if it exists. If the +// address was generated via SLAAC, an attempt is made to generate a new +// address. +func (e *endpoint) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + + addressEndpoint := e.getAddressRLocked(addr) + if addressEndpoint == nil { + return tcpip.ErrBadAddress + } + + if addressEndpoint.GetKind() != stack.PermanentTentative { + return tcpip.ErrInvalidEndpointState + } + + // If the address is a SLAAC address, do not invalidate its SLAAC prefix as an + // attempt will be made to generate a new address for it. + if err := e.removePermanentEndpointLocked(addressEndpoint, false /* allowSLAACInvalidation */); err != nil { + return err + } + + prefix := addressEndpoint.AddressWithPrefix().Subnet() + + switch t := addressEndpoint.ConfigType(); t { + case stack.AddressConfigStatic: + case stack.AddressConfigSlaac: + e.mu.ndp.regenerateSLAACAddr(prefix) + case stack.AddressConfigSlaacTemp: + // Do not reset the generation attempts counter for the prefix as the + // temporary address is being regenerated in response to a DAD conflict. + e.mu.ndp.regenerateTempSLAACAddr(prefix, false /* resetGenAttempts */) + default: + panic(fmt.Sprintf("unrecognized address config type = %d", t)) + } + + return nil +} + +// transitionForwarding transitions the endpoint's forwarding status to +// forwarding. +// +// Must only be called when the forwarding status changes. +func (e *endpoint) transitionForwarding(forwarding bool) { + e.mu.Lock() + defer e.mu.Unlock() + + if !e.Enabled() { + return + } + + if forwarding { + // When transitioning into an IPv6 router, host-only state (NDP discovered + // routers, discovered on-link prefixes, and auto-generated addresses) is + // cleaned up/invalidated and NDP router solicitations are stopped. + e.mu.ndp.stopSolicitingRouters() + e.mu.ndp.cleanupState(true /* hostOnly */) + } else { + // When transitioning into an IPv6 host, NDP router solicitations are + // started. + e.mu.ndp.startSolicitingRouters() + } +} + +// Enable implements stack.NetworkEndpoint. +func (e *endpoint) Enable() *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + + // If the NIC is not enabled, the endpoint can't do anything meaningful so + // don't enable the endpoint. + if !e.nic.Enabled() { + return tcpip.ErrNotPermitted + } + + // If the endpoint is already enabled, there is nothing for it to do. + if !e.setEnabled(true) { + return nil + } + + // 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 + // all-nodes multicast address. An example is the Neighbor Advertisement + // when we perform Duplicate Address Detection, or Router Advertisement + // when we do Router Discovery. See RFC 4862, section 5.4.2 and RFC 4861 + // section 4.2 for more information. + // + // Also auto-generate an IPv6 link-local address based on the endpoint's + // link address if it is configured to do so. Note, each interface is + // required to have IPv6 link-local unicast address, as per RFC 4291 + // section 2.1. + + // Join the All-Nodes multicast group before starting DAD as responses to DAD + // (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 + } + + // Perform DAD on the all the unicast IPv6 endpoints that are in the permanent + // state. + // + // 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 { + addr := addressEndpoint.AddressWithPrefix().Address + if !header.IsV6UnicastAddress(addr) { + return true + } + + switch addressEndpoint.GetKind() { + case stack.Permanent: + addressEndpoint.SetKind(stack.PermanentTentative) + fallthrough + case stack.PermanentTentative: + err = e.mu.ndp.startDuplicateAddressDetection(addr, addressEndpoint) + return err == nil + default: + return true + } + }) + if err != nil { + return err + } + + // Do not auto-generate an IPv6 link-local address for loopback devices. + if e.protocol.autoGenIPv6LinkLocal && !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) + } + + // If we are operating as a router, then do not solicit routers since we + // won't process the RAs anyway. + // + // Routers do not process Router Advertisements (RA) the same way a host + // does. That is, routers do not learn from RAs (e.g. on-link prefixes + // and default routers). Therefore, soliciting RAs from other routers on + // a link is unnecessary for routers. + if !e.protocol.Forwarding() { + e.mu.ndp.startSolicitingRouters() + } + + return nil +} + +// Enabled implements stack.NetworkEndpoint. +func (e *endpoint) Enabled() bool { + return e.nic.Enabled() && e.isEnabled() +} + +// isEnabled returns true if the endpoint is enabled, regardless of the +// enabled status of the NIC. +func (e *endpoint) isEnabled() bool { + return atomic.LoadUint32(&e.enabled) == 1 +} + +// setEnabled sets the enabled status for the endpoint. +// +// Returns true if the enabled status was updated. +func (e *endpoint) setEnabled(v bool) bool { + if v { + return atomic.SwapUint32(&e.enabled, 1) == 0 + } + return atomic.SwapUint32(&e.enabled, 0) == 1 +} + +// Disable implements stack.NetworkEndpoint. +func (e *endpoint) Disable() { + e.mu.Lock() + defer e.mu.Unlock() + e.disableLocked() +} + +func (e *endpoint) disableLocked() { + if !e.setEnabled(false) { + return + } + + e.mu.ndp.stopSolicitingRouters() + e.mu.ndp.cleanupState(false /* hostOnly */) + e.stopDADForPermanentAddressesLocked() + + // The endpoint may have already left the multicast group. + if _, err := e.mu.addressableEndpointState.LeaveGroup(header.IPv6AllNodesMulticastAddress); err != nil && err != tcpip.ErrBadLocalAddress { + panic(fmt.Sprintf("unexpected error when leaving group = %s: %s", header.IPv6AllNodesMulticastAddress, err)) + } +} + +// stopDADForPermanentAddressesLocked stops DAD for all permaneent addresses. +// +// 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 { + if addressEndpoint.GetKind() != stack.PermanentTentative { + return true + } + + addr := addressEndpoint.AddressWithPrefix().Address + if header.IsV6UnicastAddress(addr) { + e.mu.ndp.stopDuplicateAddressDetection(addr) + } + + return true + }) } // DefaultTTL is the default hop limit for this endpoint. @@ -63,31 +363,13 @@ func (e *endpoint) DefaultTTL() uint8 { // MTU implements stack.NetworkEndpoint.MTU. It returns the link-layer MTU minus // the network layer max header length. func (e *endpoint) MTU() uint32 { - return calculateMTU(e.linkEP.MTU()) -} - -// NICID returns the ID of the NIC this endpoint belongs to. -func (e *endpoint) NICID() tcpip.NICID { - return e.nicID -} - -// Capabilities implements stack.NetworkEndpoint.Capabilities. -func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities { - return e.linkEP.Capabilities() + return calculateMTU(e.nic.MTU()) } // MaxHeaderLength returns the maximum length needed by ipv6 headers (and // underlying protocols). func (e *endpoint) MaxHeaderLength() uint16 { - return e.linkEP.MaxHeaderLength() + header.IPv6MinimumSize -} - -// GSOMaxSize returns the maximum GSO packet size. -func (e *endpoint) GSOMaxSize() uint32 { - if gso, ok := e.linkEP.(stack.GSOEndpoint); ok { - return gso.GSOMaxSize() - } - return 0 + return e.nic.MaxHeaderLength() + header.IPv6MinimumSize } func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params stack.NetworkHeaderParams) { @@ -101,7 +383,44 @@ func (e *endpoint) addIPHeader(r *stack.Route, pkt *stack.PacketBuffer, params s SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) - pkt.NetworkProtocolNumber = header.IPv6ProtocolNumber + pkt.NetworkProtocolNumber = ProtocolNumber +} + +func (e *endpoint) packetMustBeFragmented(pkt *stack.PacketBuffer, gso *stack.GSO) bool { + return pkt.Size() > int(e.nic.MTU()) && (gso == nil || gso.Type == stack.GSONone) +} + +// handleFragments fragments pkt and calls the handler function on each +// fragment. It returns the number of fragments handled and the number of +// fragments left to be processed. The IP header must already be present in the +// original packet. The mtu is the maximum size of the packets. The transport +// header protocol number is required to avoid parsing the IPv6 extension +// headers. +func (e *endpoint) handleFragments(r *stack.Route, gso *stack.GSO, mtu uint32, pkt *stack.PacketBuffer, transProto tcpip.TransportProtocolNumber, handler func(*stack.PacketBuffer) *tcpip.Error) (int, int, *tcpip.Error) { + fragMTU := int(calculateFragmentInnerMTU(mtu, pkt)) + if fragMTU < pkt.TransportHeader().View().Size() { + // As per RFC 8200 Section 4.5, the Transport Header is expected to be small + // enough to fit in the first fragment. + return 0, 1, tcpip.ErrMessageTooLong + } + + pf := fragmentation.MakePacketFragmenter(pkt, fragMTU, calculateFragmentReserve(pkt)) + id := atomic.AddUint32(&e.protocol.ids[hashRoute(r, e.protocol.hashIV)%buckets], 1) + networkHeader := header.IPv6(pkt.NetworkHeader().View()) + + var n int + for { + fragPkt, more := buildNextFragment(&pf, networkHeader, transProto, id) + if err := handler(fragPkt); err != nil { + return n, pf.RemainingFragmentCount() + 1, err + } + n++ + if !more { + break + } + } + + return n, 0, nil } // WritePacket writes a packet to the given destination address and protocol. @@ -110,8 +429,8 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // iptables filtering. All packets that reach here are locally // generated. - nicName := e.stack.FindNICNameFromID(e.NICID()) - ipt := e.stack.IPTables() + nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID()) + ipt := e.protocol.stack.IPTables() if ok := ipt.Check(stack.Output, pkt, gso, r, "", nicName); !ok { // iptables is telling us to drop the packet. r.Stats().IP.IPTablesOutputDropped.Increment() @@ -127,7 +446,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw // short circuits broadcasts before they are sent out to other hosts. if pkt.NatDone { netHeader := header.IPv6(pkt.NetworkHeader().View()) - if ep, err := e.stack.FindNetworkEndpoint(header.IPv6ProtocolNumber, netHeader.DestinationAddress()); err == nil { + if ep, err := e.protocol.stack.FindNetworkEndpoint(ProtocolNumber, netHeader.DestinationAddress()); err == nil { route := r.ReverseRoute(netHeader.SourceAddress(), netHeader.DestinationAddress()) ep.HandlePacket(&route, pkt) return nil @@ -148,14 +467,29 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw return nil } - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { + if e.packetMustBeFragmented(pkt, gso) { + sent, remain, err := e.handleFragments(r, gso, e.nic.MTU(), pkt, params.Protocol, func(fragPkt *stack.PacketBuffer) *tcpip.Error { + // TODO(gvisor.dev/issue/3884): Evaluate whether we want to send each + // fragment one by one using WritePacket() (current strategy) or if we + // want to create a PacketBufferList from the fragments and feed it to + // WritePackets(). It'll be faster but cost more memory. + return e.nic.WritePacket(r, gso, ProtocolNumber, fragPkt) + }) + r.Stats().IP.PacketsSent.IncrementBy(uint64(sent)) + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(remain)) + return err + } + + if err := e.nic.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { + r.Stats().IP.OutgoingPacketErrors.Increment() return err } + r.Stats().IP.PacketsSent.Increment() return nil } -// WritePackets implements stack.LinkEndpoint.WritePackets. +// WritePackets implements stack.NetworkEndpoint.WritePackets. func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, params stack.NetworkHeaderParams) (int, *tcpip.Error) { if r.Loop&stack.PacketLoop != 0 { panic("not implemented") @@ -166,18 +500,38 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe for pb := pkts.Front(); pb != nil; pb = pb.Next() { e.addIPHeader(r, pb, params) + if e.packetMustBeFragmented(pb, gso) { + current := pb + _, _, err := e.handleFragments(r, gso, e.nic.MTU(), pb, params.Protocol, func(fragPkt *stack.PacketBuffer) *tcpip.Error { + // Modify the packet list in place with the new fragments. + pkts.InsertAfter(current, fragPkt) + current = current.Next() + return nil + }) + if err != nil { + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pkts.Len())) + return 0, err + } + // The fragmented packet can be released. The rest of the packets can be + // processed. + pkts.Remove(pb) + pb = current + } } // iptables filtering. All packets that reach here are locally // generated. - nicName := e.stack.FindNICNameFromID(e.NICID()) - ipt := e.stack.IPTables() + nicName := e.protocol.stack.FindNICNameFromID(e.nic.ID()) + ipt := e.protocol.stack.IPTables() dropped, natPkts := ipt.CheckPackets(stack.Output, pkts, gso, r, nicName) if len(dropped) == 0 && len(natPkts) == 0 { // Fast path: If no packets are to be dropped then we can just invoke the // faster WritePackets API directly. - n, err := e.linkEP.WritePackets(r, gso, pkts, ProtocolNumber) + n, err := e.nic.WritePackets(r, gso, pkts, ProtocolNumber) r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + if err != nil { + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pkts.Len() - n)) + } return n, err } r.Stats().IP.IPTablesOutputDropped.IncrementBy(uint64(len(dropped))) @@ -191,7 +545,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe } if _, ok := natPkts[pkt]; ok { netHeader := header.IPv6(pkt.NetworkHeader().View()) - if ep, err := e.stack.FindNetworkEndpoint(header.IPv6ProtocolNumber, netHeader.DestinationAddress()); err == nil { + if ep, err := e.protocol.stack.FindNetworkEndpoint(ProtocolNumber, netHeader.DestinationAddress()); err == nil { src := netHeader.SourceAddress() dst := netHeader.DestinationAddress() route := r.ReverseRoute(src, dst) @@ -200,8 +554,9 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe continue } } - if err := e.linkEP.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { + if err := e.nic.WritePacket(r, gso, ProtocolNumber, pkt); err != nil { r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(pkts.Len() - n + len(dropped))) // Dropped packets aren't errors, so include them in // the return value. return n + len(dropped), err @@ -224,12 +579,24 @@ func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack.PacketBuff // HandlePacket is called by the link layer when new ipv6 packets arrive for // this endpoint. func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { + if !e.isEnabled() { + return + } + h := header.IPv6(pkt.NetworkHeader().View()) if !h.IsValid(pkt.Data.Size() + pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size()) { r.Stats().IP.MalformedPacketsReceived.Increment() return } + // As per RFC 4291 section 2.7: + // Multicast addresses must not be used as source addresses in IPv6 + // packets or appear in any Routing header. + if header.IsV6MulticastAddress(r.RemoteAddress) { + r.Stats().IP.InvalidSourceAddressesReceived.Increment() + return + } + // vv consists of: // - Any IPv6 header bytes after the first 40 (i.e. extensions). // - The transport header, if present. @@ -241,15 +608,18 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { hasFragmentHeader := false // iptables filtering. All packets that reach here are intended for - // this machine and will not be forwarded. - ipt := e.stack.IPTables() + // this machine and need not be forwarded. + ipt := e.protocol.stack.IPTables() if ok := ipt.Check(stack.Input, pkt, nil, nil, "", ""); !ok { // iptables is telling us to drop the packet. r.Stats().IP.IPTablesInputDropped.Increment() return } - for firstHeader := true; ; firstHeader = false { + for { + // Keep track of the start of the previous header so we can report the + // special case of a Hop by Hop at a location other than at the start. + previousHeaderStart := it.HeaderOffset() extHdr, done, err := it.Next() if err != nil { r.Stats().IP.MalformedPacketsReceived.Increment() @@ -263,11 +633,11 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { case header.IPv6HopByHopOptionsExtHdr: // As per RFC 8200 section 4.1, the Hop By Hop extension header is // restricted to appear immediately after an IPv6 fixed header. - // - // TODO(b/152019344): Send an ICMPv6 Parameter Problem, Code 1 - // (unrecognized next header) error in response to an extension header's - // Next Header field with the Hop By Hop extension header identifier. - if !firstHeader { + if previousHeaderStart != 0 { + _ = e.protocol.returnError(r, &icmpReasonParameterProblem{ + code: header.ICMPv6UnknownHeader, + pointer: previousHeaderStart, + }, pkt) return } @@ -289,13 +659,25 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { case header.IPv6OptionUnknownActionSkip: case header.IPv6OptionUnknownActionDiscard: return - case header.IPv6OptionUnknownActionDiscardSendICMP: - // TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 2 for - // unrecognized IPv6 extension header options. - return case header.IPv6OptionUnknownActionDiscardSendICMPNoMulticastDest: - // TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 2 for - // unrecognized IPv6 extension header options. + if header.IsV6MulticastAddress(r.LocalAddress) { + return + } + fallthrough + case header.IPv6OptionUnknownActionDiscardSendICMP: + // This case satisfies a requirement of RFC 8200 section 4.2 + // which states that an unknown option starting with bits [10] should: + // + // discard the packet and, regardless of whether or not the + // packet's Destination Address was a multicast address, send an + // ICMP Parameter Problem, Code 2, message to the packet's + // Source Address, pointing to the unrecognized Option Type. + // + _ = e.protocol.returnError(r, &icmpReasonParameterProblem{ + code: header.ICMPv6UnknownOption, + pointer: it.ParseOffset() + optsIt.OptionOffset(), + respondToMulticast: true, + }, pkt) return default: panic(fmt.Sprintf("unrecognized action for an unrecognized Hop By Hop extension header option = %d", opt)) @@ -306,16 +688,20 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { // As per RFC 8200 section 4.4, if a node encounters a routing header with // an unrecognized routing type value, with a non-zero Segments Left // value, the node must discard the packet and send an ICMP Parameter - // Problem, Code 0. If the Segments Left is 0, the node must ignore the - // Routing extension header and process the next header in the packet. + // Problem, Code 0 to the packet's Source Address, pointing to the + // unrecognized Routing Type. + // + // If the Segments Left is 0, the node must ignore the Routing extension + // header and process the next header in the packet. // // Note, the stack does not yet handle any type of routing extension // header, so we just make sure Segments Left is zero before processing // the next extension header. - // - // TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 0 for - // unrecognized routing types with a non-zero Segments Left value. if extHdr.SegmentsLeft() != 0 { + _ = e.protocol.returnError(r, &icmpReasonParameterProblem{ + code: header.ICMPv6ErroneousHeader, + pointer: it.ParseOffset(), + }, pkt) return } @@ -348,7 +734,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { it, done, err := it.Next() if err != nil { r.Stats().IP.MalformedPacketsReceived.Increment() - r.Stats().IP.MalformedPacketsReceived.Increment() + r.Stats().IP.MalformedFragmentsReceived.Increment() return } if done { @@ -450,13 +836,25 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { case header.IPv6OptionUnknownActionSkip: case header.IPv6OptionUnknownActionDiscard: return - case header.IPv6OptionUnknownActionDiscardSendICMP: - // TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 2 for - // unrecognized IPv6 extension header options. - return case header.IPv6OptionUnknownActionDiscardSendICMPNoMulticastDest: - // TODO(b/152019344): Send an ICMPv6 Parameter Problem Code 2 for - // unrecognized IPv6 extension header options. + if header.IsV6MulticastAddress(r.LocalAddress) { + return + } + fallthrough + case header.IPv6OptionUnknownActionDiscardSendICMP: + // This case satisfies a requirement of RFC 8200 section 4.2 + // which states that an unknown option starting with bits [10] should: + // + // discard the packet and, regardless of whether or not the + // packet's Destination Address was a multicast address, send an + // ICMP Parameter Problem, Code 2, message to the packet's + // Source Address, pointing to the unrecognized Option Type. + // + _ = e.protocol.returnError(r, &icmpReasonParameterProblem{ + code: header.ICMPv6UnknownOption, + pointer: it.ParseOffset() + optsIt.OptionOffset(), + respondToMulticast: true, + }, pkt) return default: panic(fmt.Sprintf("unrecognized action for an unrecognized Destination extension header option = %d", opt)) @@ -475,21 +873,55 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { extHdr.Buf.TrimFront(pkt.TransportHeader().View().Size()) pkt.Data = extHdr.Buf + r.Stats().IP.PacketsDelivered.Increment() if p := tcpip.TransportProtocolNumber(extHdr.Identifier); p == header.ICMPv6ProtocolNumber { + pkt.TransportProtocolNumber = p e.handleICMP(r, pkt, hasFragmentHeader) } else { r.Stats().IP.PacketsDelivered.Increment() - // TODO(b/152019344): Send an ICMPv6 Parameter Problem, Code 1 error - // in response to unrecognized next header values. - e.dispatcher.DeliverTransportPacket(r, p, pkt) + switch res := e.dispatcher.DeliverTransportPacket(r, p, pkt); res { + case stack.TransportPacketHandled: + case stack.TransportPacketDestinationPortUnreachable: + // As per RFC 4443 section 3.1: + // A destination node SHOULD originate a Destination Unreachable + // message with Code 4 in response to a packet for which the + // transport protocol (e.g., UDP) has no listener, if that transport + // protocol has no alternative means to inform the sender. + _ = e.protocol.returnError(r, &icmpReasonPortUnreachable{}, pkt) + case stack.TransportPacketProtocolUnreachable: + // As per RFC 8200 section 4. (page 7): + // Extension headers are numbered from IANA IP Protocol Numbers + // [IANA-PN], the same values used for IPv4 and IPv6. When + // processing a sequence of Next Header values in a packet, the + // first one that is not an extension header [IANA-EH] indicates + // that the next item in the packet is the corresponding upper-layer + // header. + // With more related information on page 8: + // If, as a result of processing a header, the destination node is + // required to proceed to the next header but the Next Header value + // in the current header is unrecognized by the node, it should + // discard the packet and send an ICMP Parameter Problem message to + // the source of the packet, with an ICMP Code value of 1 + // ("unrecognized Next Header type encountered") and the ICMP + // Pointer field containing the offset of the unrecognized value + // within the original packet. + // + // Which when taken together indicate that an unknown protocol should + // be treated as an unrecognized next header value. + _ = e.protocol.returnError(r, &icmpReasonParameterProblem{ + code: header.ICMPv6UnknownHeader, + pointer: it.ParseOffset(), + }, pkt) + default: + panic(fmt.Sprintf("unrecognized result from DeliverTransportPacket = %d", res)) + } } default: - // If we receive a packet for an extension header we do not yet handle, - // drop the packet for now. - // - // TODO(b/152019344): Send an ICMPv6 Parameter Problem, Code 1 error - // in response to unrecognized next header values. + _ = e.protocol.returnError(r, &icmpReasonParameterProblem{ + code: header.ICMPv6UnknownHeader, + pointer: it.ParseOffset(), + }, pkt) r.Stats().UnknownProtocolRcvdPackets.Increment() return } @@ -497,19 +929,343 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } // Close cleans up resources associated with the endpoint. -func (*endpoint) Close() {} +func (e *endpoint) Close() { + e.mu.Lock() + e.disableLocked() + e.mu.ndp.removeSLAACAddresses(false /* keepLinkLocal */) + e.stopDADForPermanentAddressesLocked() + e.mu.addressableEndpointState.Cleanup() + e.mu.Unlock() + + e.protocol.forgetEndpoint(e) +} // NetworkProtocolNumber implements stack.NetworkEndpoint.NetworkProtocolNumber. func (e *endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return e.protocol.Number() } +// AddAndAcquirePermanentAddress implements stack.AddressableEndpoint. +func (e *endpoint) AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb stack.PrimaryEndpointBehavior, configType stack.AddressConfigType, deprecated bool) (stack.AddressEndpoint, *tcpip.Error) { + // TODO(b/169350103): add checks here after making sure we no longer receive + // an empty address. + e.mu.Lock() + defer e.mu.Unlock() + return e.addAndAcquirePermanentAddressLocked(addr, peb, configType, deprecated) +} + +// addAndAcquirePermanentAddressLocked is like AddAndAcquirePermanentAddress but +// with locking requirements. +// +// addAndAcquirePermanentAddressLocked also joins the passed address's +// solicited-node multicast group and start duplicate address detection. +// +// Precondition: e.mu must be write locked. +func (e *endpoint) addAndAcquirePermanentAddressLocked(addr tcpip.AddressWithPrefix, peb stack.PrimaryEndpointBehavior, configType stack.AddressConfigType, deprecated bool) (stack.AddressEndpoint, *tcpip.Error) { + addressEndpoint, err := e.mu.addressableEndpointState.AddAndAcquirePermanentAddress(addr, peb, configType, deprecated) + if err != nil { + return nil, err + } + + if !header.IsV6UnicastAddress(addr.Address) { + return addressEndpoint, nil + } + + snmc := header.SolicitedNodeAddr(addr.Address) + if _, err := e.mu.addressableEndpointState.JoinGroup(snmc); err != nil { + return nil, err + } + + addressEndpoint.SetKind(stack.PermanentTentative) + + if e.Enabled() { + if err := e.mu.ndp.startDuplicateAddressDetection(addr.Address, addressEndpoint); err != nil { + return nil, err + } + } + + return addressEndpoint, nil +} + +// RemovePermanentAddress implements stack.AddressableEndpoint. +func (e *endpoint) RemovePermanentAddress(addr tcpip.Address) *tcpip.Error { + e.mu.Lock() + defer e.mu.Unlock() + + addressEndpoint := e.getAddressRLocked(addr) + if addressEndpoint == nil || !addressEndpoint.GetKind().IsPermanent() { + return tcpip.ErrBadLocalAddress + } + + return e.removePermanentEndpointLocked(addressEndpoint, true) +} + +// removePermanentEndpointLocked is like removePermanentAddressLocked except +// it works with a stack.AddressEndpoint. +// +// Precondition: e.mu must be write locked. +func (e *endpoint) removePermanentEndpointLocked(addressEndpoint stack.AddressEndpoint, allowSLAACInvalidation bool) *tcpip.Error { + addr := addressEndpoint.AddressWithPrefix() + unicast := header.IsV6UnicastAddress(addr.Address) + if unicast { + e.mu.ndp.stopDuplicateAddressDetection(addr.Address) + + // If we are removing an address generated via SLAAC, cleanup + // its SLAAC resources and notify the integrator. + switch addressEndpoint.ConfigType() { + case stack.AddressConfigSlaac: + e.mu.ndp.cleanupSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) + case stack.AddressConfigSlaacTemp: + e.mu.ndp.cleanupTempSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) + } + } + + if err := e.mu.addressableEndpointState.RemovePermanentEndpoint(addressEndpoint); err != nil { + return err + } + + if !unicast { + return nil + } + + snmc := header.SolicitedNodeAddr(addr.Address) + if _, err := e.mu.addressableEndpointState.LeaveGroup(snmc); err != nil && err != tcpip.ErrBadLocalAddress { + return err + } + + return nil +} + +// hasPermanentAddressLocked returns true if the endpoint has a permanent +// address equal to the passed address. +// +// Precondition: e.mu must be read or write locked. +func (e *endpoint) hasPermanentAddressRLocked(addr tcpip.Address) bool { + addressEndpoint := e.getAddressRLocked(addr) + if addressEndpoint == nil { + return false + } + return addressEndpoint.GetKind().IsPermanent() +} + +// getAddressRLocked returns the endpoint for the passed address. +// +// 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) +} + +// MainAddress implements stack.AddressableEndpoint. +func (e *endpoint) MainAddress() tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.MainAddress() +} + +// AcquireAssignedAddress implements stack.AddressableEndpoint. +func (e *endpoint) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB stack.PrimaryEndpointBehavior) stack.AddressEndpoint { + e.mu.Lock() + defer e.mu.Unlock() + return e.acquireAddressOrCreateTempLocked(localAddr, allowTemp, tempPEB) +} + +// acquireAddressOrCreateTempLocked is like AcquireAssignedAddress but with +// locking requirements. +// +// Precondition: e.mu must be write locked. +func (e *endpoint) acquireAddressOrCreateTempLocked(localAddr tcpip.Address, allowTemp bool, tempPEB stack.PrimaryEndpointBehavior) stack.AddressEndpoint { + return e.mu.addressableEndpointState.AcquireAssignedAddress(localAddr, allowTemp, tempPEB) +} + +// AcquireOutgoingPrimaryAddress implements stack.AddressableEndpoint. +func (e *endpoint) AcquireOutgoingPrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) stack.AddressEndpoint { + e.mu.RLock() + defer e.mu.RUnlock() + return e.acquireOutgoingPrimaryAddressRLocked(remoteAddr, allowExpired) +} + +// acquireOutgoingPrimaryAddressRLocked is like AcquireOutgoingPrimaryAddress +// but with locking requirements. +// +// Precondition: e.mu must be read locked. +func (e *endpoint) acquireOutgoingPrimaryAddressRLocked(remoteAddr tcpip.Address, allowExpired bool) stack.AddressEndpoint { + // addrCandidate is a candidate for Source Address Selection, as per + // RFC 6724 section 5. + type addrCandidate struct { + addressEndpoint stack.AddressEndpoint + scope header.IPv6AddressScope + } + + if len(remoteAddr) == 0 { + return e.mu.addressableEndpointState.AcquireOutgoingPrimaryAddress(remoteAddr, allowExpired) + } + + // 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) { + // If r is not valid for outgoing connections, it is not a valid endpoint. + if !addressEndpoint.IsAssigned(allowExpired) { + return + } + + addr := addressEndpoint.AddressWithPrefix().Address + scope, err := header.ScopeForIPv6Address(addr) + if err != nil { + // Should never happen as we got r from the primary IPv6 endpoint list and + // ScopeForIPv6Address only returns an error if addr is not an IPv6 + // address. + panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", addr, err)) + } + + cs = append(cs, addrCandidate{ + addressEndpoint: addressEndpoint, + scope: scope, + }) + }) + + remoteScope, err := header.ScopeForIPv6Address(remoteAddr) + if err != nil { + // primaryIPv6Endpoint should never be called with an invalid IPv6 address. + panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", remoteAddr, err)) + } + + // Sort the addresses as per RFC 6724 section 5 rules 1-3. + // + // TODO(b/146021396): Implement rules 4-8 of RFC 6724 section 5. + sort.Slice(cs, func(i, j int) bool { + sa := cs[i] + sb := cs[j] + + // Prefer same address as per RFC 6724 section 5 rule 1. + if sa.addressEndpoint.AddressWithPrefix().Address == remoteAddr { + return true + } + if sb.addressEndpoint.AddressWithPrefix().Address == remoteAddr { + return false + } + + // Prefer appropriate scope as per RFC 6724 section 5 rule 2. + if sa.scope < sb.scope { + return sa.scope >= remoteScope + } else if sb.scope < sa.scope { + return sb.scope < remoteScope + } + + // Avoid deprecated addresses as per RFC 6724 section 5 rule 3. + if saDep, sbDep := sa.addressEndpoint.Deprecated(), sb.addressEndpoint.Deprecated(); saDep != sbDep { + // If sa is not deprecated, it is preferred over sb. + return sbDep + } + + // Prefer temporary addresses as per RFC 6724 section 5 rule 7. + if saTemp, sbTemp := sa.addressEndpoint.ConfigType() == stack.AddressConfigSlaacTemp, sb.addressEndpoint.ConfigType() == stack.AddressConfigSlaacTemp; saTemp != sbTemp { + return saTemp + } + + // sa and sb are equal, return the endpoint that is closest to the front of + // the primary endpoint list. + return i < j + }) + + // Return the most preferred address that can have its reference count + // incremented. + for _, c := range cs { + if c.addressEndpoint.IncRef() { + return c.addressEndpoint + } + } + + return nil +} + +// PrimaryAddresses implements stack.AddressableEndpoint. +func (e *endpoint) PrimaryAddresses() []tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.PrimaryAddresses() +} + +// PermanentAddresses implements stack.AddressableEndpoint. +func (e *endpoint) PermanentAddresses() []tcpip.AddressWithPrefix { + e.mu.RLock() + defer e.mu.RUnlock() + return e.mu.addressableEndpointState.PermanentAddresses() +} + +// JoinGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) JoinGroup(addr tcpip.Address) (bool, *tcpip.Error) { + if !header.IsV6MulticastAddress(addr) { + return false, tcpip.ErrBadAddress + } + + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.JoinGroup(addr) +} + +// LeaveGroup implements stack.GroupAddressableEndpoint. +func (e *endpoint) LeaveGroup(addr tcpip.Address) (bool, *tcpip.Error) { + e.mu.Lock() + defer e.mu.Unlock() + return e.mu.addressableEndpointState.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) +} + +var _ stack.ForwardingNetworkProtocol = (*protocol)(nil) +var _ stack.NetworkProtocol = (*protocol)(nil) + type protocol struct { + stack *stack.Stack + + mu struct { + sync.RWMutex + + eps map[*endpoint]struct{} + } + + ids []uint32 + hashIV uint32 + // defaultTTL is the current default TTL for the protocol. Only the - // uint8 portion of it is meaningful and it must be accessed - // atomically. - defaultTTL uint32 + // uint8 portion of it is meaningful. + // + // Must be accessed using atomic operations. + defaultTTL uint32 + + // forwarding is set to 1 when the protocol has forwarding enabled and 0 + // when it is disabled. + // + // Must be accessed using atomic operations. + 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. @@ -534,16 +1290,35 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { } // NewEndpoint creates a new ipv6 endpoint. -func (p *protocol) NewEndpoint(nicID tcpip.NICID, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) stack.NetworkEndpoint { - return &endpoint{ - nicID: nicID, - linkEP: linkEP, +func (p *protocol) NewEndpoint(nic stack.NetworkInterface, linkAddrCache stack.LinkAddressCache, nud stack.NUDHandler, dispatcher stack.TransportDispatcher) stack.NetworkEndpoint { + e := &endpoint{ + nic: nic, linkAddrCache: linkAddrCache, nud: nud, dispatcher: dispatcher, protocol: p, - stack: st, } + e.mu.addressableEndpointState.Init(e) + e.mu.ndp = ndpState{ + ep: e, + configs: p.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() + + p.mu.Lock() + defer p.mu.Unlock() + p.mu.eps[e] = struct{}{} + return e +} + +func (p *protocol) forgetEndpoint(e *endpoint) { + p.mu.Lock() + defer p.mu.Unlock() + delete(p.mu.eps, e) } // SetOption implements NetworkProtocol.SetOption. @@ -594,6 +1369,35 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) (proto tcpip.TransportProtocolNu return proto, !fragMore && fragOffset == 0, true } +// Forwarding implements stack.ForwardingNetworkProtocol. +func (p *protocol) Forwarding() bool { + return uint8(atomic.LoadUint32(&p.forwarding)) == 1 +} + +// setForwarding sets the forwarding status for the protocol. +// +// Returns true if the forwarding status was updated. +func (p *protocol) setForwarding(v bool) bool { + if v { + return atomic.SwapUint32(&p.forwarding, 1) == 0 + } + return atomic.SwapUint32(&p.forwarding, 0) == 1 +} + +// SetForwarding implements stack.ForwardingNetworkProtocol. +func (p *protocol) SetForwarding(v bool) { + p.mu.Lock() + defer p.mu.Unlock() + + if !p.setForwarding(v) { + return + } + + for ep := range p.mu.eps { + ep.transitionForwarding(v) + } +} + // calculateMTU calculates the network-layer payload MTU based on the link-layer // payload mtu. func calculateMTU(mtu uint32) uint32 { @@ -604,10 +1408,144 @@ func calculateMTU(mtu uint32) uint32 { return maxPayloadSize } -// NewProtocol returns an IPv6 network protocol. -func NewProtocol() stack.NetworkProtocol { - return &protocol{ - defaultTTL: DefaultTTL, - fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), +// Options holds options to configure a new protocol. +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 + // 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. + // + // The generated link-local address follows RFC 4291 Appendix A guidelines. + AutoGenIPv6LinkLocal bool + + // NDPDisp is the NDP event dispatcher that an integrator can provide to + // receive NDP related events. + NDPDisp NDPDispatcher + + // 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. + // + // Temporary SLAAC adresses are short-lived addresses which are unpredictable + // and random from the perspective of other nodes on the network. It is + // recommended that the seed be a random byte buffer of at least + // header.IIDSize bytes to make sure that temporary SLAAC addresses are + // sufficiently random. It should follow minimum randomness requirements for + // security as outlined by RFC 4086. + // + // Note: using a nil value, the same seed across netstack program runs, or a + // seed that is too small would reduce randomness and increase predictability, + // defeating the purpose of temporary SLAAC addresses. + TempIIDSeed []byte +} + +// NewProtocolWithOptions returns an IPv6 network protocol. +func NewProtocolWithOptions(opts Options) stack.NetworkProtocolFactory { + opts.NDPConfigs.validate() + + ids := hash.RandN32(buckets) + hashIV := hash.RandN32(1)[0] + + return func(s *stack.Stack) stack.NetworkProtocol { + p := &protocol{ + stack: s, + fragmentation: fragmentation.NewFragmentation(header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit, fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, reassembleTimeout, s.Clock()), + ids: ids, + hashIV: hashIV, + + ndpDisp: opts.NDPDisp, + ndpConfigs: opts.NDPConfigs, + opaqueIIDOpts: opts.OpaqueIIDOpts, + tempIIDSeed: opts.TempIIDSeed, + autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, + } + p.mu.eps = make(map[*endpoint]struct{}) + p.SetDefaultTTL(DefaultTTL) + return p } } + +// NewProtocol is equivalent to NewProtocolWithOptions with an empty Options. +func NewProtocol(s *stack.Stack) stack.NetworkProtocol { + return NewProtocolWithOptions(Options{})(s) +} + +// calculateFragmentInnerMTU calculates the maximum number of bytes of +// fragmentable data a fragment can have, based on the link layer mtu and pkt's +// network header size. +func calculateFragmentInnerMTU(mtu uint32, pkt *stack.PacketBuffer) uint32 { + // TODO(gvisor.dev/issue/3912): Once the Authentication or ESP Headers are + // supported for outbound packets, their length should not affect the fragment + // MTU because they should only be transmitted once. + mtu -= uint32(pkt.NetworkHeader().View().Size()) + mtu -= header.IPv6FragmentHeaderSize + // Round the MTU down to align to 8 bytes. + mtu &^= 7 + if mtu <= maxPayloadSize { + return mtu + } + return maxPayloadSize +} + +func calculateFragmentReserve(pkt *stack.PacketBuffer) int { + return pkt.AvailableHeaderBytes() + pkt.NetworkHeader().View().Size() + header.IPv6FragmentHeaderSize +} + +// hashRoute calculates a hash value for the given route. It uses the source & +// destination address and 32-bit number to generate the hash. +func hashRoute(r *stack.Route, hashIV uint32) uint32 { + // The FNV-1a was chosen because it is a fast hashing algorithm, and + // cryptographic properties are not needed here. + h := fnv.New32a() + if _, err := h.Write([]byte(r.LocalAddress)); err != nil { + panic(fmt.Sprintf("Hash.Write: %s, but Hash' implementation of Write is not expected to ever return an error", err)) + } + + if _, err := h.Write([]byte(r.RemoteAddress)); err != nil { + panic(fmt.Sprintf("Hash.Write: %s, but Hash' implementation of Write is not expected to ever return an error", err)) + } + + s := make([]byte, 4) + binary.LittleEndian.PutUint32(s, hashIV) + if _, err := h.Write(s); err != nil { + panic(fmt.Sprintf("Hash.Write: %s, but Hash' implementation of Write is not expected ever to return an error", err)) + } + + return h.Sum32() +} + +func buildNextFragment(pf *fragmentation.PacketFragmenter, originalIPHeaders header.IPv6, transportProto tcpip.TransportProtocolNumber, id uint32) (*stack.PacketBuffer, bool) { + fragPkt, offset, copied, more := pf.BuildNextFragment() + fragPkt.NetworkProtocolNumber = ProtocolNumber + + originalIPHeadersLength := len(originalIPHeaders) + fragmentIPHeadersLength := originalIPHeadersLength + header.IPv6FragmentHeaderSize + fragmentIPHeaders := header.IPv6(fragPkt.NetworkHeader().Push(fragmentIPHeadersLength)) + + // Copy the IPv6 header and any extension headers already populated. + if copied := copy(fragmentIPHeaders, originalIPHeaders); copied != originalIPHeadersLength { + panic(fmt.Sprintf("wrong number of bytes copied into fragmentIPHeaders: got %d, want %d", copied, originalIPHeadersLength)) + } + fragmentIPHeaders.SetNextHeader(header.IPv6FragmentHeader) + fragmentIPHeaders.SetPayloadLength(uint16(copied + fragmentIPHeadersLength - header.IPv6MinimumSize)) + + fragmentHeader := header.IPv6Fragment(fragmentIPHeaders[originalIPHeadersLength:]) + fragmentHeader.Encode(&header.IPv6FragmentFields{ + M: more, + FragmentOffset: uint16(offset / header.IPv6FragmentExtHdrFragmentOffsetBytesPerUnit), + Identification: id, + NextHeader: uint8(transportProto), + }) + + return fragPkt, more +} diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 9eea1de8d..e792ca9e2 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -15,16 +15,21 @@ package ipv6 import ( + "encoding/hex" + "fmt" "math" "testing" "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "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/testutil" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/tcpip/transport/icmp" + "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/pkg/waiter" ) @@ -135,23 +140,99 @@ func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst } } +func compareFragments(packets []*stack.PacketBuffer, sourcePacket *stack.PacketBuffer, mtu uint32, wantFragments []fragmentInfo, proto tcpip.TransportProtocolNumber) error { + // sourcePacket does not have its IP Header populated. Let's copy the one + // from the first fragment. + source := header.IPv6(packets[0].NetworkHeader().View()) + sourceIPHeadersLen := len(source) + vv := buffer.NewVectorisedView(sourcePacket.Size(), sourcePacket.Views()) + source = append(source, vv.ToView()...) + + var reassembledPayload buffer.VectorisedView + for i, fragment := range packets { + // Confirm that the packet is valid. + allBytes := buffer.NewVectorisedView(fragment.Size(), fragment.Views()) + fragmentIPHeaders := header.IPv6(allBytes.ToView()) + if !fragmentIPHeaders.IsValid(len(fragmentIPHeaders)) { + return fmt.Errorf("fragment #%d: IP packet is invalid:\n%s", i, hex.Dump(fragmentIPHeaders)) + } + + fragmentIPHeadersLength := fragment.NetworkHeader().View().Size() + if fragmentIPHeadersLength != sourceIPHeadersLen { + return fmt.Errorf("fragment #%d: got fragmentIPHeadersLength = %d, want = %d", i, fragmentIPHeadersLength, sourceIPHeadersLen) + } + + if got := len(fragmentIPHeaders); got > int(mtu) { + return fmt.Errorf("fragment #%d: got len(fragmentIPHeaders) = %d, want <= %d", i, got, mtu) + } + + sourceIPHeader := source[:header.IPv6MinimumSize] + fragmentIPHeader := fragmentIPHeaders[:header.IPv6MinimumSize] + + if got := fragmentIPHeaders.PayloadLength(); got != wantFragments[i].payloadSize { + return fmt.Errorf("fragment #%d: got fragmentIPHeaders.PayloadLength() = %d, want = %d", i, got, wantFragments[i].payloadSize) + } + + // We expect the IPv6 Header to be similar across each fragment, besides the + // payload length. + sourceIPHeader.SetPayloadLength(0) + fragmentIPHeader.SetPayloadLength(0) + if diff := cmp.Diff(fragmentIPHeader, sourceIPHeader); diff != "" { + return fmt.Errorf("fragment #%d: fragmentIPHeader mismatch (-want +got):\n%s", i, diff) + } + + if fragment.NetworkProtocolNumber != sourcePacket.NetworkProtocolNumber { + return fmt.Errorf("fragment #%d: got fragment.NetworkProtocolNumber = %d, want = %d", i, fragment.NetworkProtocolNumber, sourcePacket.NetworkProtocolNumber) + } + + if len(packets) > 1 { + // If the source packet was big enough that it needed fragmentation, let's + // inspect the fragment header. Because no other extension headers are + // supported, it will always be the last extension header. + fragmentHeader := header.IPv6Fragment(fragmentIPHeaders[fragmentIPHeadersLength-header.IPv6FragmentHeaderSize : fragmentIPHeadersLength]) + + if got := fragmentHeader.More(); got != wantFragments[i].more { + return fmt.Errorf("fragment #%d: got fragmentHeader.More() = %t, want = %t", i, got, wantFragments[i].more) + } + if got := fragmentHeader.FragmentOffset(); got != wantFragments[i].offset { + return fmt.Errorf("fragment #%d: got fragmentHeader.FragmentOffset() = %d, want = %d", i, got, wantFragments[i].offset) + } + if got := fragmentHeader.NextHeader(); got != uint8(proto) { + return fmt.Errorf("fragment #%d: got fragmentHeader.NextHeader() = %d, want = %d", i, got, uint8(proto)) + } + } + + // Store the reassembled payload as we parse each fragment. The payload + // includes the Transport header and everything after. + reassembledPayload.AppendView(fragment.TransportHeader().View()) + reassembledPayload.Append(fragment.Data) + } + + result := reassembledPayload.ToView() + if diff := cmp.Diff(result, buffer.View(source[sourceIPHeadersLen:])); diff != "" { + return fmt.Errorf("reassembledPayload mismatch (-want +got):\n%s", diff) + } + + return nil +} + // TestReceiveOnAllNodesMulticastAddr tests that IPv6 endpoints receive ICMP and // UDP packets destined to the IPv6 link-local all-nodes multicast address. func TestReceiveOnAllNodesMulticastAddr(t *testing.T) { tests := []struct { name string - protocolFactory stack.TransportProtocol + protocolFactory stack.TransportProtocolFactory rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) }{ - {"ICMP", icmp.NewProtocol6(), testReceiveICMP}, - {"UDP", udp.NewProtocol(), testReceiveUDP}, + {"ICMP", icmp.NewProtocol6, testReceiveICMP}, + {"UDP", udp.NewProtocol, testReceiveUDP}, } for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{test.protocolFactory}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{test.protocolFactory}, }) e := channel.New(10, 1280, linkAddr1) if err := s.CreateNIC(1, e); err != nil { @@ -169,15 +250,13 @@ func TestReceiveOnAllNodesMulticastAddr(t *testing.T) { // packets destined to the IPv6 solicited-node address of an assigned IPv6 // address. func TestReceiveOnSolicitedNodeAddr(t *testing.T) { - const nicID = 1 - tests := []struct { name string - protocolFactory stack.TransportProtocol + protocolFactory stack.TransportProtocolFactory rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) }{ - {"ICMP", icmp.NewProtocol6(), testReceiveICMP}, - {"UDP", udp.NewProtocol(), testReceiveUDP}, + {"ICMP", icmp.NewProtocol6, testReceiveICMP}, + {"UDP", udp.NewProtocol, testReceiveUDP}, } snmc := header.SolicitedNodeAddr(addr2) @@ -185,8 +264,8 @@ func TestReceiveOnSolicitedNodeAddr(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{test.protocolFactory}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{test.protocolFactory}, }) e := channel.New(1, 1280, linkAddr1) if err := s.CreateNIC(nicID, e); err != nil { @@ -194,7 +273,7 @@ func TestReceiveOnSolicitedNodeAddr(t *testing.T) { } s.SetRouteTable([]tcpip.Route{ - tcpip.Route{ + { Destination: header.IPv6EmptySubnet, NIC: nicID, }, @@ -272,7 +351,7 @@ func TestAddIpv6Address(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) if err := s.CreateNIC(1, &stubLinkEndpoint{}); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -294,17 +373,22 @@ func TestAddIpv6Address(t *testing.T) { } func TestReceiveIPv6ExtHdrs(t *testing.T) { - const nicID = 1 - tests := []struct { name string extHdr func(nextHdr uint8) ([]byte, uint8) shouldAccept bool + // Should we expect an ICMP response and if so, with what contents? + expectICMP bool + ICMPType header.ICMPv6Type + ICMPCode header.ICMPv6Code + pointer uint32 + multicast bool }{ { name: "None", extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{}, nextHdr }, shouldAccept: true, + expectICMP: false, }, { name: "hopbyhop with unknown option skippable action", @@ -335,9 +419,10 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { }, hopByHopExtHdrID }, shouldAccept: false, + expectICMP: false, }, { - name: "hopbyhop with unknown option discard and send icmp action", + name: "hopbyhop with unknown option discard and send icmp action (unicast)", extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{ nextHdr, 1, @@ -347,12 +432,58 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { // Discard & send ICMP if option is unknown. 191, 6, 1, 2, 3, 4, 5, 6, + //^ Unknown option. }, hopByHopExtHdrID }, shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownOption, + pointer: header.IPv6FixedHeaderSize + 8, }, { - name: "hopbyhop with unknown option discard and send icmp action unless multicast dest", + name: "hopbyhop with unknown option discard and send icmp action (multicast)", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 1, + + // Skippable unknown. + 63, 4, 1, 2, 3, 4, + + // Discard & send ICMP if option is unknown. + 191, 6, 1, 2, 3, 4, 5, 6, + //^ Unknown option. + }, hopByHopExtHdrID + }, + multicast: true, + shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownOption, + pointer: header.IPv6FixedHeaderSize + 8, + }, + { + name: "hopbyhop with unknown option discard and send icmp action unless multicast dest (unicast)", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 1, + + // Skippable unknown. + 63, 4, 1, 2, 3, 4, + + // Discard & send ICMP unless packet is for multicast destination if + // option is unknown. + 255, 6, 1, 2, 3, 4, 5, 6, + //^ Unknown option. + }, hopByHopExtHdrID + }, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownOption, + pointer: header.IPv6FixedHeaderSize + 8, + }, + { + name: "hopbyhop with unknown option discard and send icmp action unless multicast dest (multicast)", extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{ nextHdr, 1, @@ -363,39 +494,77 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { // Discard & send ICMP unless packet is for multicast destination if // option is unknown. 255, 6, 1, 2, 3, 4, 5, 6, + //^ Unknown option. }, hopByHopExtHdrID }, + multicast: true, shouldAccept: false, + expectICMP: false, }, { - name: "routing with zero segments left", - extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{nextHdr, 0, 1, 0, 2, 3, 4, 5}, routingExtHdrID }, + name: "routing with zero segments left", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 0, + 1, 0, 2, 3, 4, 5, + }, routingExtHdrID + }, shouldAccept: true, }, { - name: "routing with non-zero segments left", - extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{nextHdr, 0, 1, 1, 2, 3, 4, 5}, routingExtHdrID }, + name: "routing with non-zero segments left", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 0, + 1, 1, 2, 3, 4, 5, + }, routingExtHdrID + }, shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6ErroneousHeader, + pointer: header.IPv6FixedHeaderSize + 2, }, { - name: "atomic fragment with zero ID", - extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{nextHdr, 0, 0, 0, 0, 0, 0, 0}, fragmentExtHdrID }, + name: "atomic fragment with zero ID", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 0, + 0, 0, 0, 0, 0, 0, + }, fragmentExtHdrID + }, shouldAccept: true, }, { - name: "atomic fragment with non-zero ID", - extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{nextHdr, 0, 0, 0, 1, 2, 3, 4}, fragmentExtHdrID }, + name: "atomic fragment with non-zero ID", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 0, + 0, 0, 1, 2, 3, 4, + }, fragmentExtHdrID + }, shouldAccept: true, + expectICMP: false, }, { - name: "fragment", - extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{nextHdr, 0, 1, 0, 1, 2, 3, 4}, fragmentExtHdrID }, + name: "fragment", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 0, + 1, 0, 1, 2, 3, 4, + }, fragmentExtHdrID + }, shouldAccept: false, + expectICMP: false, }, { - name: "No next header", - extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{}, noNextHdrID }, + name: "No next header", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{}, + noNextHdrID + }, shouldAccept: false, + expectICMP: false, }, { name: "destination with unknown option skippable action", @@ -411,6 +580,7 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { }, destinationExtHdrID }, shouldAccept: true, + expectICMP: false, }, { name: "destination with unknown option discard action", @@ -426,9 +596,10 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { }, destinationExtHdrID }, shouldAccept: false, + expectICMP: false, }, { - name: "destination with unknown option discard and send icmp action", + name: "destination with unknown option discard and send icmp action (unicast)", extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{ nextHdr, 1, @@ -438,12 +609,38 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { // Discard & send ICMP if option is unknown. 191, 6, 1, 2, 3, 4, 5, 6, + //^ 191 is an unknown option. }, destinationExtHdrID }, shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownOption, + pointer: header.IPv6FixedHeaderSize + 8, }, { - name: "destination with unknown option discard and send icmp action unless multicast dest", + name: "destination with unknown option discard and send icmp action (muilticast)", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + nextHdr, 1, + + // Skippable unknown. + 63, 4, 1, 2, 3, 4, + + // Discard & send ICMP if option is unknown. + 191, 6, 1, 2, 3, 4, 5, 6, + //^ 191 is an unknown option. + }, destinationExtHdrID + }, + multicast: true, + shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownOption, + pointer: header.IPv6FixedHeaderSize + 8, + }, + { + name: "destination with unknown option discard and send icmp action unless multicast dest (unicast)", extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{ nextHdr, 1, @@ -454,22 +651,33 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { // Discard & send ICMP unless packet is for multicast destination if // option is unknown. 255, 6, 1, 2, 3, 4, 5, 6, + //^ 255 is unknown. }, destinationExtHdrID }, shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownOption, + pointer: header.IPv6FixedHeaderSize + 8, }, { - name: "routing - atomic fragment", + name: "destination with unknown option discard and send icmp action unless multicast dest (multicast)", extHdr: func(nextHdr uint8) ([]byte, uint8) { return []byte{ - // Routing extension header. - fragmentExtHdrID, 0, 1, 0, 2, 3, 4, 5, + nextHdr, 1, - // Fragment extension header. - nextHdr, 0, 0, 0, 1, 2, 3, 4, - }, routingExtHdrID + // Skippable unknown. + 63, 4, 1, 2, 3, 4, + + // Discard & send ICMP unless packet is for multicast destination if + // option is unknown. + 255, 6, 1, 2, 3, 4, 5, 6, + //^ 255 is unknown. + }, destinationExtHdrID }, - shouldAccept: true, + shouldAccept: false, + expectICMP: false, + multicast: true, }, { name: "atomic fragment - routing", @@ -503,12 +711,42 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { return []byte{ // Routing extension header. hopByHopExtHdrID, 0, 1, 0, 2, 3, 4, 5, + // ^^^ The HopByHop extension header may not appear after the first + // extension header. // Hop By Hop extension header with skippable unknown option. nextHdr, 0, 62, 4, 1, 2, 3, 4, }, routingExtHdrID }, shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownHeader, + pointer: header.IPv6FixedHeaderSize, + }, + { + name: "routing - hop by hop (with send icmp unknown)", + extHdr: func(nextHdr uint8) ([]byte, uint8) { + return []byte{ + // Routing extension header. + hopByHopExtHdrID, 0, 1, 0, 2, 3, 4, 5, + // ^^^ The HopByHop extension header may not appear after the first + // extension header. + + nextHdr, 1, + + // Skippable unknown. + 63, 4, 1, 2, 3, 4, + + // Skippable unknown. + 191, 6, 1, 2, 3, 4, 5, 6, + }, routingExtHdrID + }, + shouldAccept: false, + expectICMP: true, + ICMPType: header.ICMPv6ParamProblem, + ICMPCode: header.ICMPv6UnknownHeader, + pointer: header.IPv6FixedHeaderSize, }, { name: "No next header", @@ -552,6 +790,7 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { }, hopByHopExtHdrID }, shouldAccept: false, + expectICMP: false, }, { name: "hopbyhop (with skippable unknown) - routing - atomic fragment - destination (with discard unknown)", @@ -572,16 +811,17 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { }, hopByHopExtHdrID }, shouldAccept: false, + expectICMP: false, }, } for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) - e := channel.New(0, 1280, linkAddr1) + e := channel.New(1, 1280, linkAddr1) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } @@ -589,6 +829,14 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, ProtocolNumber, addr2, err) } + // Add a default route so that a return packet knows where to go. + s.SetRouteTable([]tcpip.Route{ + { + Destination: header.IPv6EmptySubnet, + NIC: nicID, + }, + }) + wq := waiter.Queue{} we, ch := waiter.NewChannelEntry(nil) wq.EventRegister(&we, waiter.EventIn) @@ -630,12 +878,16 @@ 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, HopLimit: 255, SrcAddr: addr1, - DstAddr: addr2, + DstAddr: dstAddr, }) e.InjectInbound(ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ @@ -649,6 +901,44 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { t.Errorf("got UDP Rx Packets = %d, want = 0", got) } + if !test.expectICMP { + if p, ok := e.Read(); ok { + t.Fatalf("unexpected packet received: %#v", p) + } + return + } + + // ICMP required. + p, ok := e.Read() + if !ok { + t.Fatalf("expected packet wasn't written out") + } + + // Pack the output packet into a single buffer.View as the checkers + // assume that. + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) + pkt := vv.ToView() + if got, want := len(pkt), header.IPv6FixedHeaderSize+header.ICMPv6MinimumSize+hdr.UsedLength(); got != want { + t.Fatalf("got an ICMP packet of size = %d, want = %d", got, want) + } + + ipHdr := header.IPv6(pkt) + checker.IPv6(t, ipHdr, checker.ICMPv6( + checker.ICMPv6Type(test.ICMPType), + checker.ICMPv6Code(test.ICMPCode))) + + // We know we are looking at no extension headers in the error ICMP + // packets. + icmpPkt := header.ICMPv6(ipHdr.Payload()) + // We know we sent small packets that won't be truncated when reflected + // back to us. + originalPacket := icmpPkt.Payload() + if got, want := icmpPkt.TypeSpecific(), test.pointer; got != want { + t.Errorf("unexpected ICMPv6 pointer, got = %d, want = %d\n", got, want) + } + if diff := cmp.Diff(hdr.View(), buffer.View(originalPacket)); diff != "" { + t.Errorf("ICMPv6 payload mismatch (-want +got):\n%s", diff) + } return } @@ -682,7 +972,6 @@ type fragmentData struct { func TestReceiveIPv6Fragments(t *testing.T) { const ( - nicID = 1 udpPayload1Length = 256 udpPayload2Length = 128 // Used to test cases where the fragment blocks are not a multiple of @@ -1548,8 +1837,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, 1280, linkAddr1) if err := s.CreateNIC(nicID, e); err != nil { @@ -1667,8 +1956,8 @@ func TestInvalidIPv6Fragments(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ - NewProtocol(), + NetworkProtocols: []stack.NetworkProtocolFactory{ + NewProtocol, }, }) e := channel.New(0, 1500, linkAddr1) @@ -1715,7 +2004,7 @@ func TestWriteStats(t *testing.T) { tests := []struct { name string setup func(*testing.T, *stack.Stack) - linkEP func() stack.LinkEndpoint + allowPackets int expectSent int expectDropped int expectWritten int @@ -1724,7 +2013,7 @@ func TestWriteStats(t *testing.T) { name: "Accept all", // No setup needed, tables accept everything by default. setup: func(*testing.T, *stack.Stack) {}, - linkEP: func() stack.LinkEndpoint { return &limitedEP{nPackets} }, + allowPackets: math.MaxInt32, expectSent: nPackets, expectDropped: 0, expectWritten: nPackets, @@ -1732,7 +2021,7 @@ func TestWriteStats(t *testing.T) { name: "Accept all with error", // No setup needed, tables accept everything by default. setup: func(*testing.T, *stack.Stack) {}, - linkEP: func() stack.LinkEndpoint { return &limitedEP{nPackets - 1} }, + allowPackets: nPackets - 1, expectSent: nPackets - 1, expectDropped: 0, expectWritten: nPackets - 1, @@ -1747,12 +2036,12 @@ func TestWriteStats(t *testing.T) { t.Fatalf("failed to find filter table") } ruleIdx := filter.BuiltinChains[stack.Output] - filter.Rules[ruleIdx].Target = stack.DropTarget{} + filter.Rules[ruleIdx].Target = &stack.DropTarget{} if err := ipt.ReplaceTable(stack.FilterTable, filter, true /* ipv6 */); err != nil { t.Fatalf("failed to replace table: %v", err) } }, - linkEP: func() stack.LinkEndpoint { return &limitedEP{nPackets} }, + allowPackets: math.MaxInt32, expectSent: 0, expectDropped: nPackets, expectWritten: nPackets, @@ -1769,15 +2058,15 @@ func TestWriteStats(t *testing.T) { } // We'll match and DROP the last packet. ruleIdx := filter.BuiltinChains[stack.Output] - filter.Rules[ruleIdx].Target = stack.DropTarget{} + filter.Rules[ruleIdx].Target = &stack.DropTarget{} filter.Rules[ruleIdx].Matchers = []stack.Matcher{&limitedMatcher{nPackets - 1}} // Make sure the next rule is ACCEPT. - filter.Rules[ruleIdx+1].Target = stack.AcceptTarget{} + filter.Rules[ruleIdx+1].Target = &stack.AcceptTarget{} if err := ipt.ReplaceTable(stack.FilterTable, filter, true /* ipv6 */); err != nil { t.Fatalf("failed to replace table: %v", err) } }, - linkEP: func() stack.LinkEndpoint { return &limitedEP{nPackets} }, + allowPackets: math.MaxInt32, expectSent: nPackets - 1, expectDropped: 1, expectWritten: nPackets, @@ -1812,8 +2101,8 @@ func TestWriteStats(t *testing.T) { t.Run(writer.name, func(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - rt := buildRoute(t, nil, test.linkEP()) - + ep := testutil.NewMockLinkEndpoint(header.IPv6MinimumMTU, tcpip.ErrInvalidEndpointState, test.allowPackets) + rt := buildRoute(t, ep) var pkts stack.PacketBufferList for i := 0; i < nPackets; i++ { pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ @@ -1843,116 +2132,371 @@ func TestWriteStats(t *testing.T) { } } -func buildRoute(t *testing.T, packetCollectorErrors []*tcpip.Error, linkEP stack.LinkEndpoint) stack.Route { +func buildRoute(t *testing.T, ep stack.LinkEndpoint) stack.Route { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) - s.CreateNIC(1, linkEP) + if err := s.CreateNIC(1, ep); err != nil { + t.Fatalf("CreateNIC(1, _) failed: %s", err) + } const ( src = "\xfc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" dst = "\xfc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" ) - s.AddAddress(1, ProtocolNumber, src) + if err := s.AddAddress(1, ProtocolNumber, src); err != nil { + t.Fatalf("AddAddress(1, %d, %s) failed: %s", ProtocolNumber, src, err) + } { - subnet, err := tcpip.NewSubnet(dst, tcpip.AddressMask("\xfc\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff")) + mask := tcpip.AddressMask("\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff") + subnet, err := tcpip.NewSubnet(dst, mask) if err != nil { - t.Fatal(err) + t.Fatalf("NewSubnet(%s, %s) failed: %v", dst, mask, err) } s.SetRouteTable([]tcpip.Route{{ Destination: subnet, NIC: 1, }}) } - rt, err := s.FindRoute(0, src, dst, ProtocolNumber, false /* multicastLoop */) + rt, err := s.FindRoute(1, src, dst, ProtocolNumber, false /* multicastLoop */) if err != nil { - t.Fatalf("s.FindRoute got %v, want %v", err, nil) + t.Fatalf("FindRoute(1, %s, %s, %d, false) = %s, want = nil", src, dst, ProtocolNumber, err) } return rt } -// limitedEP is a link endpoint that writes up to a certain number of packets -// before returning errors. -type limitedEP struct { +// limitedMatcher is an iptables matcher that matches after a certain number of +// packets are checked against it. +type limitedMatcher struct { limit int } -// MTU implements LinkEndpoint.MTU. -func (*limitedEP) MTU() uint32 { - return header.IPv6MinimumMTU +// Name implements Matcher.Name. +func (*limitedMatcher) Name() string { + return "limitedMatcher" } -// Capabilities implements LinkEndpoint.Capabilities. -func (*limitedEP) Capabilities() stack.LinkEndpointCapabilities { return 0 } +// Match implements Matcher.Match. +func (lm *limitedMatcher) Match(stack.Hook, *stack.PacketBuffer, string) (bool, bool) { + if lm.limit == 0 { + return true, false + } + lm.limit-- + return false, false +} -// MaxHeaderLength implements LinkEndpoint.MaxHeaderLength. -func (*limitedEP) MaxHeaderLength() uint16 { return 0 } +func TestClearEndpointFromProtocolOnClose(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + }) + proto := s.NetworkProtocolInstance(ProtocolNumber).(*protocol) + ep := proto.NewEndpoint(&testInterface{}, nil, nil, nil).(*endpoint) + { + proto.mu.Lock() + _, hasEP := proto.mu.eps[ep] + proto.mu.Unlock() + if !hasEP { + t.Fatalf("expected protocol to have ep = %p in set of endpoints", ep) + } + } -// LinkAddress implements LinkEndpoint.LinkAddress. -func (*limitedEP) LinkAddress() tcpip.LinkAddress { return "" } + ep.Close() -// WritePacket implements LinkEndpoint.WritePacket. -func (ep *limitedEP) WritePacket(*stack.Route, *stack.GSO, tcpip.NetworkProtocolNumber, *stack.PacketBuffer) *tcpip.Error { - if ep.limit == 0 { - return tcpip.ErrInvalidEndpointState + { + proto.mu.Lock() + _, hasEP := proto.mu.eps[ep] + proto.mu.Unlock() + if hasEP { + t.Fatalf("unexpectedly found ep = %p in set of protocol's endpoints", ep) + } } - ep.limit-- - return nil } -// WritePackets implements LinkEndpoint.WritePackets. -func (ep *limitedEP) WritePackets(_ *stack.Route, _ *stack.GSO, pkts stack.PacketBufferList, _ tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - if ep.limit == 0 { - return 0, tcpip.ErrInvalidEndpointState - } - nWritten := ep.limit - if nWritten > pkts.Len() { - nWritten = pkts.Len() - } - ep.limit -= nWritten - return nWritten, nil +type fragmentInfo struct { + offset uint16 + more bool + payloadSize uint16 +} + +type fragmentationTestCase struct { + description string + mtu uint32 + gso *stack.GSO + transHdrLen int + extraHdrLen int + payloadSize int + wantFragments []fragmentInfo + expectedFrags int +} + +var fragmentationTests = []fragmentationTestCase{ + { + description: "No Fragmentation", + mtu: 1280, + gso: &stack.GSO{}, + transHdrLen: 0, + extraHdrLen: header.IPv6MinimumSize, + payloadSize: 1000, + wantFragments: []fragmentInfo{ + {offset: 0, payloadSize: 1000, more: false}, + }, + }, + { + description: "Fragmented", + mtu: 1280, + gso: &stack.GSO{}, + transHdrLen: 0, + extraHdrLen: header.IPv6MinimumSize, + payloadSize: 2000, + wantFragments: []fragmentInfo{ + {offset: 0, payloadSize: 1240, more: true}, + {offset: 154, payloadSize: 776, more: false}, + }, + }, + { + description: "No fragmentation with big header", + mtu: 2000, + gso: &stack.GSO{}, + transHdrLen: 100, + extraHdrLen: header.IPv6MinimumSize, + payloadSize: 1000, + wantFragments: []fragmentInfo{ + {offset: 0, payloadSize: 1100, more: false}, + }, + }, + { + description: "Fragmented with gso nil", + mtu: 1280, + gso: nil, + transHdrLen: 0, + extraHdrLen: header.IPv6MinimumSize, + payloadSize: 1400, + wantFragments: []fragmentInfo{ + {offset: 0, payloadSize: 1240, more: true}, + {offset: 154, payloadSize: 176, more: false}, + }, + }, + { + description: "Fragmented with big header", + mtu: 1280, + gso: &stack.GSO{}, + transHdrLen: 100, + extraHdrLen: header.IPv6MinimumSize, + payloadSize: 1200, + wantFragments: []fragmentInfo{ + {offset: 0, payloadSize: 1240, more: true}, + {offset: 154, payloadSize: 76, more: false}, + }, + }, + { + description: "Fragmented with big header and prependable bytes", + mtu: 1280, + gso: &stack.GSO{}, + transHdrLen: 20, + extraHdrLen: header.IPv6MinimumSize + 66, + payloadSize: 1500, + wantFragments: []fragmentInfo{ + {offset: 0, payloadSize: 1240, more: true}, + {offset: 154, payloadSize: 296, more: false}, + }, + }, } -// WriteRawPacket implements LinkEndpoint.WriteRawPacket. -func (ep *limitedEP) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { - if ep.limit == 0 { - return tcpip.ErrInvalidEndpointState +func TestFragmentation(t *testing.T) { + const ( + ttl = 42 + tos = stack.DefaultTOS + transportProto = tcp.ProtocolNumber + ) + + for _, ft := range fragmentationTests { + t.Run(ft.description, func(t *testing.T) { + pkt := testutil.MakeRandPkt(ft.transHdrLen, ft.extraHdrLen, []int{ft.payloadSize}, header.IPv6ProtocolNumber) + source := pkt.Clone() + ep := testutil.NewMockLinkEndpoint(ft.mtu, nil, math.MaxInt32) + r := buildRoute(t, ep) + err := r.WritePacket(ft.gso, stack.NetworkHeaderParams{ + Protocol: tcp.ProtocolNumber, + TTL: ttl, + TOS: stack.DefaultTOS, + }, pkt) + if err != nil { + t.Fatalf("WritePacket(_, _, _): = %s", err) + } + if got := len(ep.WrittenPackets); got != len(ft.wantFragments) { + t.Errorf("got len(ep.WrittenPackets) = %d, want = %d", got, len(ft.wantFragments)) + } + if got := int(r.Stats().IP.PacketsSent.Value()); got != len(ft.wantFragments) { + t.Errorf("got c.Route.Stats().IP.PacketsSent.Value() = %d, want = %d", got, len(ft.wantFragments)) + } + if got := r.Stats().IP.OutgoingPacketErrors.Value(); got != 0 { + t.Errorf("got r.Stats().IP.OutgoingPacketErrors.Value() = %d, want = 0", got) + } + if len(ep.WrittenPackets) > 0 { + if err := compareFragments(ep.WrittenPackets, source, ft.mtu, ft.wantFragments, tcp.ProtocolNumber); err != nil { + t.Error(err) + } + } + }) } - ep.limit-- - return nil } -// Attach implements LinkEndpoint.Attach. -func (*limitedEP) Attach(_ stack.NetworkDispatcher) {} +func TestFragmentationWritePackets(t *testing.T) { + const ttl = 42 + tests := []struct { + description string + insertBefore int + insertAfter int + }{ + { + description: "Single packet", + insertBefore: 0, + insertAfter: 0, + }, + { + description: "With packet before", + insertBefore: 1, + insertAfter: 0, + }, + { + description: "With packet after", + insertBefore: 0, + insertAfter: 1, + }, + { + description: "With packet before and after", + insertBefore: 1, + insertAfter: 1, + }, + } + tinyPacket := testutil.MakeRandPkt(header.TCPMinimumSize, header.IPv6MinimumSize, []int{1}, header.IPv6ProtocolNumber) -// IsAttached implements LinkEndpoint.IsAttached. -func (*limitedEP) IsAttached() bool { return false } + for _, test := range tests { + t.Run(test.description, func(t *testing.T) { + for _, ft := range fragmentationTests { + t.Run(ft.description, func(t *testing.T) { + var pkts stack.PacketBufferList + for i := 0; i < test.insertBefore; i++ { + pkts.PushBack(tinyPacket.Clone()) + } + pkt := testutil.MakeRandPkt(ft.transHdrLen, ft.extraHdrLen, []int{ft.payloadSize}, header.IPv6ProtocolNumber) + source := pkt + pkts.PushBack(pkt.Clone()) + for i := 0; i < test.insertAfter; i++ { + pkts.PushBack(tinyPacket.Clone()) + } -// Wait implements LinkEndpoint.Wait. -func (*limitedEP) Wait() {} + ep := testutil.NewMockLinkEndpoint(ft.mtu, nil, math.MaxInt32) + r := buildRoute(t, ep) + + wantTotalPackets := len(ft.wantFragments) + test.insertBefore + test.insertAfter + n, err := r.WritePackets(ft.gso, pkts, stack.NetworkHeaderParams{ + Protocol: tcp.ProtocolNumber, + TTL: ttl, + TOS: stack.DefaultTOS, + }) + if n != wantTotalPackets || err != nil { + t.Errorf("got WritePackets(_, _, _) = (%d, %s), want = (%d, nil)", n, err, wantTotalPackets) + } + if got := len(ep.WrittenPackets); got != wantTotalPackets { + t.Errorf("got len(ep.WrittenPackets) = %d, want = %d", got, wantTotalPackets) + } + if got := int(r.Stats().IP.PacketsSent.Value()); got != wantTotalPackets { + t.Errorf("got c.Route.Stats().IP.PacketsSent.Value() = %d, want = %d", got, wantTotalPackets) + } + if got := r.Stats().IP.OutgoingPacketErrors.Value(); got != 0 { + t.Errorf("got r.Stats().IP.OutgoingPacketErrors.Value() = %d, want = 0", got) + } -// ARPHardwareType implements LinkEndpoint.ARPHardwareType. -func (*limitedEP) ARPHardwareType() header.ARPHardwareType { return header.ARPHardwareEther } + if wantTotalPackets == 0 { + return + } -// AddHeader implements LinkEndpoint.AddHeader. -func (*limitedEP) AddHeader(_, _ tcpip.LinkAddress, _ tcpip.NetworkProtocolNumber, _ *stack.PacketBuffer) { + fragments := ep.WrittenPackets[test.insertBefore : len(ft.wantFragments)+test.insertBefore] + if err := compareFragments(fragments, source, ft.mtu, ft.wantFragments, tcp.ProtocolNumber); err != nil { + t.Error(err) + } + }) + } + }) + } } -// limitedMatcher is an iptables matcher that matches after a certain number of -// packets are checked against it. -type limitedMatcher struct { - limit int -} +// TestFragmentationErrors checks that errors are returned from WritePacket +// correctly. +func TestFragmentationErrors(t *testing.T) { + const ttl = 42 -// Name implements Matcher.Name. -func (*limitedMatcher) Name() string { - return "limitedMatcher" -} + tests := []struct { + description string + mtu uint32 + transHdrLen int + payloadSize int + allowPackets int + outgoingErrors int + mockError *tcpip.Error + wantError *tcpip.Error + }{ + { + description: "No frag", + mtu: 2000, + payloadSize: 1000, + transHdrLen: 0, + allowPackets: 0, + outgoingErrors: 1, + mockError: tcpip.ErrAborted, + wantError: tcpip.ErrAborted, + }, + { + description: "Error on first frag", + mtu: 1300, + payloadSize: 3000, + transHdrLen: 0, + allowPackets: 0, + outgoingErrors: 3, + mockError: tcpip.ErrAborted, + wantError: tcpip.ErrAborted, + }, + { + description: "Error on second frag", + mtu: 1500, + payloadSize: 4000, + transHdrLen: 0, + allowPackets: 1, + outgoingErrors: 2, + mockError: tcpip.ErrAborted, + wantError: tcpip.ErrAborted, + }, + { + description: "Error on packet with MTU smaller than transport header", + mtu: 1280, + transHdrLen: 1500, + payloadSize: 500, + allowPackets: 0, + outgoingErrors: 1, + mockError: nil, + wantError: tcpip.ErrMessageTooLong, + }, + } -// Match implements Matcher.Match. -func (lm *limitedMatcher) Match(stack.Hook, *stack.PacketBuffer, string) (bool, bool) { - if lm.limit == 0 { - return true, false + for _, ft := range tests { + t.Run(ft.description, func(t *testing.T) { + pkt := testutil.MakeRandPkt(ft.transHdrLen, header.IPv6MinimumSize, []int{ft.payloadSize}, header.IPv6ProtocolNumber) + ep := testutil.NewMockLinkEndpoint(ft.mtu, ft.mockError, ft.allowPackets) + r := buildRoute(t, ep) + err := r.WritePacket(&stack.GSO{}, stack.NetworkHeaderParams{ + Protocol: tcp.ProtocolNumber, + TTL: ttl, + TOS: stack.DefaultTOS, + }, pkt) + if err != ft.wantError { + t.Errorf("got WritePacket(_, _, _) = %s, want = %s", err, ft.wantError) + } + if got := int(r.Stats().IP.PacketsSent.Value()); got != ft.allowPackets { + t.Errorf("got r.Stats().IP.PacketsSent.Value() = %d, want = %d", got, ft.allowPackets) + } + if got := int(r.Stats().IP.OutgoingPacketErrors.Value()); got != ft.outgoingErrors { + t.Errorf("got r.Stats().IP.OutgoingPacketErrors.Value() = %d, want = %d", got, ft.outgoingErrors) + } + }) } - lm.limit-- - return false, false } diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/network/ipv6/ndp.go index 97ca00d16..40da011f8 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/network/ipv6/ndp.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -package stack +package ipv6 import ( "fmt" @@ -23,9 +23,27 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/stack" ) const ( + // defaultRetransmitTimer is the default amount of time to wait between + // sending reachability probes. + // + // Default taken from RETRANS_TIMER of RFC 4861 section 10. + defaultRetransmitTimer = time.Second + + // minimumRetransmitTimer is the minimum amount of time to wait between + // sending reachability probes. + // + // Note, RFC 4861 does not impose a minimum Retransmit Timer, but we do here + // to make sure the messages are not sent all at once. We also come to this + // value because in the RetransmitTimer field of a Router Advertisement, a + // value of 0 means unspecified, so the smallest valid value is 1. Note, the + // unit of the RetransmitTimer field in the Router Advertisement is + // milliseconds. + minimumRetransmitTimer = time.Millisecond + // defaultDupAddrDetectTransmits is the default number of NDP Neighbor // Solicitation messages to send when doing Duplicate Address Detection // for a tentative address. @@ -34,7 +52,7 @@ const ( defaultDupAddrDetectTransmits = 1 // defaultMaxRtrSolicitations is the default number of Router - // Solicitation messages to send when a NIC becomes enabled. + // Solicitation messages to send when an IPv6 endpoint becomes enabled. // // Default = 3 (from RFC 4861 section 10). defaultMaxRtrSolicitations = 3 @@ -131,7 +149,7 @@ const ( minRegenAdvanceDuration = time.Duration(0) // maxSLAACAddrLocalRegenAttempts is the maximum number of times to attempt - // SLAAC address regenerations in response to a NIC-local conflict. + // SLAAC address regenerations in response to an IPv6 endpoint-local conflict. maxSLAACAddrLocalRegenAttempts = 10 ) @@ -163,7 +181,7 @@ var ( // This is exported as a variable (instead of a constant) so tests // can update it to a smaller value. // - // This value guarantees that a temporary address will be preferred for at + // This value guarantees that a temporary address is preferred for at // least 1hr if the SLAAC prefix is valid for at least that time. MinMaxTempAddrPreferredLifetime = defaultRegenAdvanceDuration + MaxDesyncFactor + time.Hour @@ -173,11 +191,17 @@ var ( // This is exported as a variable (instead of a constant) so tests // can update it to a smaller value. // - // This value guarantees that a temporary address will be valid for at least + // This value guarantees that a temporary address is valid for at least // 2hrs if the SLAAC prefix is valid for at least that time. MinMaxTempAddrValidLifetime = 2 * time.Hour ) +// NDPEndpoint is an endpoint that supports NDP. +type NDPEndpoint interface { + // SetNDPConfigurations sets the NDP configurations. + SetNDPConfigurations(NDPConfigurations) +} + // DHCPv6ConfigurationFromNDPRA is a configuration available via DHCPv6 that an // NDP Router Advertisement informed the Stack about. type DHCPv6ConfigurationFromNDPRA int @@ -192,7 +216,7 @@ const ( // DHCPv6ManagedAddress indicates that addresses are available via DHCPv6. // // DHCPv6ManagedAddress also implies DHCPv6OtherConfigurations because DHCPv6 - // will return all available configuration information. + // returns all available configuration information when serving addresses. DHCPv6ManagedAddress // DHCPv6OtherConfigurations indicates that other configuration information is @@ -207,19 +231,18 @@ const ( // NDPDispatcher is the interface integrators of netstack must implement to // receive and handle NDP related events. type NDPDispatcher interface { - // OnDuplicateAddressDetectionStatus will be called when the DAD process - // for an address (addr) on a NIC (with ID nicID) completes. resolved - // will be set to true if DAD completed successfully (no duplicate addr - // detected); false otherwise (addr was detected to be a duplicate on - // the link the NIC is a part of, or it was stopped for some other - // reason, such as the address being removed). If an error occured - // during DAD, err will be set and resolved must be ignored. + // OnDuplicateAddressDetectionStatus is called when the DAD process for an + // address (addr) on a NIC (with ID nicID) completes. resolved is set to true + // if DAD completed successfully (no duplicate addr detected); false otherwise + // (addr was detected to be a duplicate on the link the NIC is a part of, or + // it was stopped for some other reason, such as the address being removed). + // If an error occured during DAD, err is set and resolved must be ignored. // // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) - // OnDefaultRouterDiscovered will be called when a new default router is + // OnDefaultRouterDiscovered is called when a new default router is // discovered. Implementations must return true if the newly discovered // router should be remembered. // @@ -227,56 +250,55 @@ type NDPDispatcher interface { // is also not permitted to call into the stack. OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip.Address) bool - // OnDefaultRouterInvalidated will be called when a discovered default - // router that was remembered is invalidated. + // OnDefaultRouterInvalidated is called when a discovered default router that + // was remembered is invalidated. // // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address) - // OnOnLinkPrefixDiscovered will be called when a new on-link prefix is - // discovered. Implementations must return true if the newly discovered - // on-link prefix should be remembered. + // OnOnLinkPrefixDiscovered is called when a new on-link prefix is discovered. + // Implementations must return true if the newly discovered on-link prefix + // should be remembered. // // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip.Subnet) bool - // OnOnLinkPrefixInvalidated will be called when a discovered on-link - // prefix that was remembered is invalidated. + // OnOnLinkPrefixInvalidated is called when a discovered on-link prefix that + // was remembered is invalidated. // // This function is not permitted to block indefinitely. This function // is also not permitted to call into the stack. OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet) - // OnAutoGenAddress will be called when a new prefix with its - // autonomous address-configuration flag set has been received and SLAAC - // has been performed. Implementations may prevent the stack from - // assigning the address to the NIC by returning false. + // OnAutoGenAddress is called when a new prefix with its autonomous address- + // configuration flag set is received and SLAAC was performed. Implementations + // may prevent the stack from assigning the address to the NIC by returning + // false. // // This function is not permitted to block indefinitely. It must not // call functions on the stack itself. OnAutoGenAddress(tcpip.NICID, tcpip.AddressWithPrefix) bool - // OnAutoGenAddressDeprecated will be called when an auto-generated - // address (as part of SLAAC) has been deprecated, but is still - // considered valid. Note, if an address is invalidated at the same - // time it is deprecated, the deprecation event MAY be omitted. + // OnAutoGenAddressDeprecated is called when an auto-generated address (SLAAC) + // is deprecated, but is still considered valid. Note, if an address is + // invalidated at the same ime it is deprecated, the deprecation event may not + // be received. // // This function is not permitted to block indefinitely. It must not // call functions on the stack itself. OnAutoGenAddressDeprecated(tcpip.NICID, tcpip.AddressWithPrefix) - // OnAutoGenAddressInvalidated will be called when an auto-generated - // address (as part of SLAAC) has been invalidated. + // OnAutoGenAddressInvalidated is called when an auto-generated address + // (SLAAC) is invalidated. // // This function is not permitted to block indefinitely. It must not // call functions on the stack itself. OnAutoGenAddressInvalidated(tcpip.NICID, tcpip.AddressWithPrefix) - // OnRecursiveDNSServerOption will be called when an NDP option with - // recursive DNS servers has been received. Note, addrs may contain - // link-local addresses. + // OnRecursiveDNSServerOption is called when the stack learns of DNS servers + // through NDP. Note, the addresses may contain link-local addresses. // // It is up to the caller to use the DNS Servers only for their valid // lifetime. OnRecursiveDNSServerOption may be called for new or @@ -288,8 +310,8 @@ type NDPDispatcher interface { // call functions on the stack itself. OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) - // OnDNSSearchListOption will be called when an NDP option with a DNS - // search list has been received. + // OnDNSSearchListOption is called when the stack learns of DNS search lists + // through NDP. // // It is up to the caller to use the domain names in the search list // for only their valid lifetime. OnDNSSearchListOption may be called @@ -298,8 +320,8 @@ type NDPDispatcher interface { // be increased, decreased or completely invalidated when lifetime = 0. OnDNSSearchListOption(nicID tcpip.NICID, domainNames []string, lifetime time.Duration) - // OnDHCPv6Configuration will be called with an updated configuration that is - // available via DHCPv6 for a specified NIC. + // OnDHCPv6Configuration is called with an updated configuration that is + // available via DHCPv6 for the passed NIC. // // This function is not permitted to block indefinitely. It must not // call functions on the stack itself. @@ -320,7 +342,7 @@ type NDPConfigurations struct { // Must be greater than or equal to 1ms. RetransmitTimer time.Duration - // The number of Router Solicitation messages to send when the NIC + // The number of Router Solicitation messages to send when the IPv6 endpoint // becomes enabled. MaxRtrSolicitations uint8 @@ -335,24 +357,22 @@ type NDPConfigurations struct { // Must be greater than or equal to 0s. MaxRtrSolicitationDelay time.Duration - // HandleRAs determines whether or not Router Advertisements will be - // processed. + // HandleRAs determines whether or not Router Advertisements are processed. HandleRAs bool - // DiscoverDefaultRouters determines whether or not default routers will - // be discovered from Router Advertisements. This configuration is - // ignored if HandleRAs is false. + // DiscoverDefaultRouters determines whether or not default routers are + // discovered from Router Advertisements, as per RFC 4861 section 6. This + // configuration is ignored if HandleRAs is false. DiscoverDefaultRouters bool - // DiscoverOnLinkPrefixes determines whether or not on-link prefixes - // will be discovered from Router Advertisements' Prefix Information - // option. This configuration is ignored if HandleRAs is false. + // DiscoverOnLinkPrefixes determines whether or not on-link prefixes are + // discovered from Router Advertisements' Prefix Information option, as per + // RFC 4861 section 6. This configuration is ignored if HandleRAs is false. DiscoverOnLinkPrefixes bool - // AutoGenGlobalAddresses determines whether or not global IPv6 - // addresses will be generated for a NIC in response to receiving a new - // Prefix Information option with its Autonomous Address - // AutoConfiguration flag set, as a host, as per RFC 4862 (SLAAC). + // AutoGenGlobalAddresses determines whether or not an IPv6 endpoint performs + // SLAAC to auto-generate global SLAAC addresses in response to Prefix + // Information options, as per RFC 4862. // // Note, if an address was already generated for some unique prefix, as // part of SLAAC, this option does not affect whether or not the @@ -366,12 +386,12 @@ type NDPConfigurations struct { // // If the method used to generate the address does not support creating // alternative addresses (e.g. IIDs based on the modified EUI64 of a NIC's - // MAC address), then no attempt will be made to resolve the conflict. + // MAC address), then no attempt is made to resolve the conflict. AutoGenAddressConflictRetries uint8 // AutoGenTempGlobalAddresses determines whether or not temporary SLAAC - // addresses will be generated for a NIC as part of SLAAC privacy extensions, - // RFC 4941. + // addresses are generated for an IPv6 endpoint as part of SLAAC privacy + // extensions, as per RFC 4941. // // Ignored if AutoGenGlobalAddresses is false. AutoGenTempGlobalAddresses bool @@ -410,7 +430,7 @@ func DefaultNDPConfigurations() NDPConfigurations { } // validate modifies an NDPConfigurations with valid values. If invalid values -// are present in c, the corresponding default values will be used instead. +// are present in c, the corresponding default values are used instead. func (c *NDPConfigurations) validate() { if c.RetransmitTimer < minimumRetransmitTimer { c.RetransmitTimer = defaultRetransmitTimer @@ -439,8 +459,8 @@ func (c *NDPConfigurations) validate() { // ndpState is the per-interface NDP state. type ndpState struct { - // The NIC this ndpState is for. - nic *NIC + // The IPv6 endpoint this ndpState is for. + ep *endpoint // configs is the per-interface NDP configurations. configs NDPConfigurations @@ -458,8 +478,8 @@ type ndpState struct { // 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 NIC this ndpState is associated with. MUST be set when the timer is - // set. + // the IPv6 endpoint this ndpState is associated with. MUST be set when the + // timer is set. done *bool } @@ -492,7 +512,7 @@ type dadState struct { // Used to let the DAD timer know that it has been stopped. // // Must only be read from or written to while protected by the lock of - // the NIC this dadState is associated with. + // the IPv6 endpoint this dadState is associated with. done *bool } @@ -537,7 +557,7 @@ type tempSLAACAddrState struct { // The address's endpoint. // // Must not be nil. - ref *referencedNetworkEndpoint + addressEndpoint stack.AddressEndpoint // Has a new temporary SLAAC address already been regenerated? regenerated bool @@ -567,10 +587,10 @@ type slaacPrefixState struct { // // May only be nil when the address is being (re-)generated. Otherwise, // must not be nil as all SLAAC prefixes must have a stable address. - ref *referencedNetworkEndpoint + addressEndpoint stack.AddressEndpoint - // The number of times an address has been generated locally where the NIC - // already had the generated address. + // The number of times an address has been generated locally where the IPv6 + // endpoint already had the generated address. localGenerationFailures uint8 } @@ -578,11 +598,12 @@ type slaacPrefixState struct { tempAddrs map[tcpip.Address]tempSLAACAddrState // The next two fields are used by both stable and temporary addresses - // generated for a SLAAC prefix. This is safe as only 1 address will be - // in the generation and DAD process at any time. That is, no two addresses - // will be generated at the same time for a given SLAAC prefix. + // generated for a SLAAC prefix. This is safe as only 1 address is in the + // generation and DAD process at any time. That is, no two addresses are + // generated at the same time for a given SLAAC prefix. - // The number of times an address has been generated and added to the NIC. + // The number of times an address has been generated and added to the IPv6 + // endpoint. // // Addresses may be regenerated in reseponse to a DAD conflicts. generationAttempts uint8 @@ -597,16 +618,16 @@ type slaacPrefixState struct { // This function must only be called by IPv6 addresses that are currently // tentative. // -// The NIC that ndp belongs to MUST be locked. -func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *referencedNetworkEndpoint) *tcpip.Error { +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, addressEndpoint stack.AddressEndpoint) *tcpip.Error { // addr must be a valid unicast IPv6 address. if !header.IsV6UnicastAddress(addr) { return tcpip.ErrAddressFamilyNotSupported } - if ref.getKind() != permanentTentative { + if addressEndpoint.GetKind() != stack.PermanentTentative { // The endpoint should be marked as tentative since we are starting DAD. - panic(fmt.Sprintf("ndpdad: addr %s is not tentative on NIC(%d)", addr, ndp.nic.ID())) + panic(fmt.Sprintf("ndpdad: addr %s is not tentative on NIC(%d)", addr, ndp.ep.nic.ID())) } // Should not attempt to perform DAD on an address that is currently in the @@ -617,18 +638,18 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref // existed, we would get an error since we attempted to add a duplicate // address, or its reference count would have been increased without doing // the work that would have been done for an address that was brand new. - // See NIC.addAddressLocked. - panic(fmt.Sprintf("ndpdad: already performing DAD for addr %s on NIC(%d)", addr, ndp.nic.ID())) + // See endpoint.addAddressLocked. + panic(fmt.Sprintf("ndpdad: already performing DAD for addr %s on NIC(%d)", addr, ndp.ep.nic.ID())) } remaining := ndp.configs.DupAddrDetectTransmits if remaining == 0 { - ref.setKind(permanent) + addressEndpoint.SetKind(stack.Permanent) // Consider DAD to have resolved even if no DAD messages were actually // transmitted. - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, true, nil) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, true, nil) } return nil @@ -637,25 +658,25 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref 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 NIC's lock. This is effectively the same - // as starting a goroutine but we use a timer that fires immediately so we can - // reset it for the next DAD iteration. - timer = ndp.nic.stack.Clock().AfterFunc(0, func() { - ndp.nic.mu.Lock() - defer ndp.nic.mu.Unlock() + // 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 NIC lock and stopped DAD - // before this function obtained the NIC lock. Simply return here and do - // nothing further. + // 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 ref.getKind() != permanentTentative { + 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.nic.ID())) + panic(fmt.Sprintf("ndpdad: addr %s is no longer tentative on NIC(%d)", addr, ndp.ep.nic.ID())) } dadDone := remaining == 0 @@ -663,33 +684,34 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref var err *tcpip.Error if !dadDone { // Use the unspecified address as the source address when performing DAD. - ref := ndp.nic.getRefOrCreateTempLocked(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint) + 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 NIC. Note, DAD would be - // stopped if the NIC was disabled or removed, or if the address was - // removed. - ndp.nic.mu.Unlock() - err = ndp.sendDADPacket(addr, ref) - ndp.nic.mu.Lock() + // 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 done { // If we reach this point, it means that DAD was stopped after we released - // the NIC's read lock and before we obtained the write lock. + // the IPv6 endpoint's read lock and before we obtained the write lock. return } if dadDone { // DAD has resolved. - ref.setKind(permanent) + 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.nic.stack.ndpConfigs.RetransmitTimer) + timer.Reset(ndp.configs.RetransmitTimer) return } @@ -698,16 +720,16 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref // integrator know DAD has completed. delete(ndp.dad, addr) - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, dadDone, err) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, dadDone, err) } // If DAD resolved for a stable SLAAC address, attempt generation of a // temporary SLAAC address. - if dadDone && ref.configType == slaac { + 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(ref.addrWithPrefix().Subnet(), true /* resetGenAttempts */) + ndp.regenerateTempSLAACAddr(addressEndpoint.AddressWithPrefix().Subnet(), true /* resetGenAttempts */) } }) @@ -722,28 +744,31 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref // sendDADPacket sends a NS message to see if any nodes on ndp's NIC's link owns // addr. // -// addr must be a tentative IPv6 address on ndp's NIC. +// addr must be a tentative IPv6 address on ndp's IPv6 endpoint. // -// The NIC ndp belongs to MUST NOT be locked. -func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEndpoint) *tcpip.Error { +// 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 := makeRoute(header.IPv6ProtocolNumber, ref.address(), snmc, ndp.nic.linkEP.LinkAddress(), ref, false, false) + 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 NIC is not locked, - // the NIC could have been disabled or removed by another goroutine. + // 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.nic.ID(), 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.nic.ID())) + 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)) @@ -752,17 +777,16 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEnd ns.SetTargetAddress(addr) icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - pkt := NewPacketBuffer(PacketBufferOptions{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(r.MaxHeaderLength()), Data: buffer.View(icmpData).ToVectorisedView(), }) sent := r.Stats().ICMP.V6PacketsSent if err := r.WritePacket(nil, - NetworkHeaderParams{ + stack.NetworkHeaderParams{ Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, - TOS: DefaultTOS, }, pkt, ); err != nil { sent.Dropped.Increment() @@ -778,11 +802,9 @@ func (ndp *ndpState) sendDADPacket(addr tcpip.Address, ref *referencedNetworkEnd // such a state forever, unless some other external event resolves the DAD // process (receiving an NA from the true owner of addr, or an NS for addr // (implying another node is attempting to use addr)). It is up to the caller -// of this function to handle such a scenario. Normally, addr will be removed -// from n right after this function returns or the address successfully -// resolved. +// of this function to handle such a scenario. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) { dad, ok := ndp.dad[addr] if !ok { @@ -801,30 +823,30 @@ func (ndp *ndpState) stopDuplicateAddressDetection(addr tcpip.Address) { delete(ndp.dad, addr) // Let the integrator know DAD did not resolve. - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnDuplicateAddressDetectionStatus(ndp.nic.ID(), addr, false, nil) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, false, nil) } } // handleRA handles a Router Advertisement message that arrived on the NIC // this ndp is for. Does nothing if the NIC is configured to not handle RAs. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { - // Is the NIC configured to handle RAs at all? + // Is the IPv6 endpoint configured to handle RAs at all? // // Currently, the stack does not determine router interface status on a - // per-interface basis; it is a stack-wide configuration, so we check - // stack's forwarding flag to determine if the NIC is a routing - // interface. - if !ndp.configs.HandleRAs || ndp.nic.stack.Forwarding(header.IPv6ProtocolNumber) { + // per-interface basis; it is a protocol-wide configuration, so we check the + // protocol's forwarding flag to determine if the IPv6 endpoint is forwarding + // packets. + if !ndp.configs.HandleRAs || ndp.ep.protocol.Forwarding() { return } // 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.nic.stack.ndpDisp; ndpDisp != nil { + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { var configuration DHCPv6ConfigurationFromNDPRA switch { case ra.ManagedAddrConfFlag(): @@ -839,11 +861,11 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { if ndp.dhcpv6Configuration != configuration { ndp.dhcpv6Configuration = configuration - ndpDisp.OnDHCPv6Configuration(ndp.nic.ID(), configuration) + ndpDisp.OnDHCPv6Configuration(ndp.ep.nic.ID(), configuration) } } - // Is the NIC configured to discover default routers? + // Is the IPv6 endpoint configured to discover default routers? if ndp.configs.DiscoverDefaultRouters { rtr, ok := ndp.defaultRouters[ip] rl := ra.RouterLifetime() @@ -881,20 +903,20 @@ 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.nic.stack.ndpDisp == nil { + if ndp.ep.protocol.ndpDisp == nil { continue } addrs, _ := opt.Addresses() - ndp.nic.stack.ndpDisp.OnRecursiveDNSServerOption(ndp.nic.ID(), addrs, opt.Lifetime()) + ndp.ep.protocol.ndpDisp.OnRecursiveDNSServerOption(ndp.ep.nic.ID(), addrs, opt.Lifetime()) case header.NDPDNSSearchList: - if ndp.nic.stack.ndpDisp == nil { + if ndp.ep.protocol.ndpDisp == nil { continue } domainNames, _ := opt.DomainNames() - ndp.nic.stack.ndpDisp.OnDNSSearchListOption(ndp.nic.ID(), domainNames, opt.Lifetime()) + ndp.ep.protocol.ndpDisp.OnDNSSearchListOption(ndp.ep.nic.ID(), domainNames, opt.Lifetime()) case header.NDPPrefixInformation: prefix := opt.Subnet() @@ -928,7 +950,7 @@ func (ndp *ndpState) handleRA(ip tcpip.Address, ra header.NDPRouterAdvert) { // invalidateDefaultRouter invalidates a discovered default router. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { rtr, ok := ndp.defaultRouters[ip] @@ -942,32 +964,32 @@ func (ndp *ndpState) invalidateDefaultRouter(ip tcpip.Address) { delete(ndp.defaultRouters, ip) // Let the integrator know a discovered default router is invalidated. - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnDefaultRouterInvalidated(ndp.nic.ID(), ip) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnDefaultRouterInvalidated(ndp.ep.nic.ID(), ip) } } // rememberDefaultRouter remembers a newly discovered default router with IPv6 // link-local address ip with lifetime rl. // -// The router identified by ip MUST NOT already be known by the NIC. +// The router identified by ip MUST NOT already be known by the IPv6 endpoint. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { - ndpDisp := ndp.nic.stack.ndpDisp + ndpDisp := ndp.ep.protocol.ndpDisp if ndpDisp == nil { return } // Inform the integrator when we discovered a default router. - if !ndpDisp.OnDefaultRouterDiscovered(ndp.nic.ID(), ip) { + if !ndpDisp.OnDefaultRouterDiscovered(ndp.ep.nic.ID(), ip) { // Informed by the integrator to not remember the router, do // nothing further. return } state := defaultRouterState{ - invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + invalidationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { ndp.invalidateDefaultRouter(ip) }), } @@ -982,22 +1004,22 @@ func (ndp *ndpState) rememberDefaultRouter(ip tcpip.Address, rl time.Duration) { // // The prefix identified by prefix MUST NOT already be known. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) { - ndpDisp := ndp.nic.stack.ndpDisp + ndpDisp := ndp.ep.protocol.ndpDisp if ndpDisp == nil { return } // Inform the integrator when we discovered an on-link prefix. - if !ndpDisp.OnOnLinkPrefixDiscovered(ndp.nic.ID(), prefix) { + if !ndpDisp.OnOnLinkPrefixDiscovered(ndp.ep.nic.ID(), prefix) { // Informed by the integrator to not remember the prefix, do // nothing further. return } state := onLinkPrefixState{ - invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + invalidationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { ndp.invalidateOnLinkPrefix(prefix) }), } @@ -1011,7 +1033,7 @@ func (ndp *ndpState) rememberOnLinkPrefix(prefix tcpip.Subnet, l time.Duration) // invalidateOnLinkPrefix invalidates a discovered on-link prefix. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { s, ok := ndp.onLinkPrefixes[prefix] @@ -1025,8 +1047,8 @@ 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.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnOnLinkPrefixInvalidated(ndp.nic.ID(), prefix) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnOnLinkPrefixInvalidated(ndp.ep.nic.ID(), prefix) } } @@ -1036,7 +1058,7 @@ func (ndp *ndpState) invalidateOnLinkPrefix(prefix tcpip.Subnet) { // handleOnLinkPrefixInformation assumes that the prefix this pi is for is // not the link-local prefix and the on-link flag is set. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformation) { prefix := pi.Subnet() prefixState, ok := ndp.onLinkPrefixes[prefix] @@ -1089,7 +1111,7 @@ func (ndp *ndpState) handleOnLinkPrefixInformation(pi header.NDPPrefixInformatio // handleAutonomousPrefixInformation assumes that the prefix this pi is for is // not the link-local prefix and the autonomous flag is set. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) handleAutonomousPrefixInformation(pi header.NDPPrefixInformation) { vl := pi.ValidLifetime() pl := pi.PreferredLifetime() @@ -1125,7 +1147,7 @@ func (ndp *ndpState) handleAutonomousPrefixInformation(pi header.NDPPrefixInform // // pl is the new preferred lifetime. vl is the new valid lifetime. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { // If we do not already have an address for this prefix and the valid // lifetime is 0, no need to do anything further, as per RFC 4862 @@ -1142,15 +1164,15 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { } state := slaacPrefixState{ - deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + deprecationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the deprecated SLAAC prefix %s", prefix)) } - ndp.deprecateSLAACAddress(state.stableAddr.ref) + ndp.deprecateSLAACAddress(state.stableAddr.addressEndpoint) }), - invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + invalidationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { state, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for the invalidated SLAAC prefix %s", prefix)) @@ -1189,7 +1211,7 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { } // If the address is assigned (DAD resolved), generate a temporary address. - if state.stableAddr.ref.getKind() == permanent { + if state.stableAddr.addressEndpoint.GetKind() == stack.Permanent { // Reset the generation attempts counter as we are starting the generation // of a new address for the SLAAC prefix. ndp.generateTempSLAACAddr(prefix, &state, true /* resetGenAttempts */) @@ -1198,32 +1220,27 @@ func (ndp *ndpState) doSLAAC(prefix tcpip.Subnet, pl, vl time.Duration) { ndp.slaacPrefixes[prefix] = state } -// addSLAACAddr adds a SLAAC address to the NIC. +// addAndAcquireSLAACAddr adds a SLAAC address to the IPv6 endpoint. // -// The NIC that ndp belongs to MUST be locked. -func (ndp *ndpState) addSLAACAddr(addr tcpip.AddressWithPrefix, configType networkEndpointConfigType, deprecated bool) *referencedNetworkEndpoint { +// 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.nic.stack.ndpDisp + ndpDisp := ndp.ep.protocol.ndpDisp if ndpDisp == nil { return nil } - if !ndpDisp.OnAutoGenAddress(ndp.nic.ID(), addr) { + if !ndpDisp.OnAutoGenAddress(ndp.ep.nic.ID(), addr) { // Informed by the integrator not to add the address. return nil } - protocolAddr := tcpip.ProtocolAddress{ - Protocol: header.IPv6ProtocolNumber, - AddressWithPrefix: addr, - } - - ref, err := ndp.nic.addAddressLocked(protocolAddr, FirstPrimaryEndpoint, permanent, configType, deprecated) + addressEndpoint, err := ndp.ep.addAndAcquirePermanentAddressLocked(addr, stack.FirstPrimaryEndpoint, configType, deprecated) if err != nil { - panic(fmt.Sprintf("ndp: error when adding SLAAC address %+v: %s", protocolAddr, err)) + panic(fmt.Sprintf("ndp: error when adding SLAAC address %+v: %s", addr, err)) } - return ref + return addressEndpoint } // generateSLAACAddr generates a SLAAC address for prefix. @@ -1232,10 +1249,10 @@ func (ndp *ndpState) addSLAACAddr(addr tcpip.AddressWithPrefix, configType netwo // // Panics if the prefix is not a SLAAC prefix or it already has an address. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixState) bool { - if r := state.stableAddr.ref; r != nil { - panic(fmt.Sprintf("ndp: SLAAC prefix %s already has a permenant address %s", prefix, r.addrWithPrefix())) + if addressEndpoint := state.stableAddr.addressEndpoint; addressEndpoint != nil { + panic(fmt.Sprintf("ndp: SLAAC prefix %s already has a permenant address %s", prefix, addressEndpoint.AddressWithPrefix())) } // If we have already reached the maximum address generation attempts for the @@ -1255,11 +1272,11 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt } dadCounter := state.generationAttempts + state.stableAddr.localGenerationFailures - if oIID := ndp.nic.stack.opaqueIIDOpts; oIID.NICNameFromID != nil { + if oIID := ndp.ep.protocol.opaqueIIDOpts; oIID.NICNameFromID != nil { addrBytes = header.AppendOpaqueInterfaceIdentifier( addrBytes[:header.IIDOffsetInIPv6Address], prefix, - oIID.NICNameFromID(ndp.nic.ID(), ndp.nic.name), + oIID.NICNameFromID(ndp.ep.nic.ID(), ndp.ep.nic.Name()), dadCounter, oIID.SecretKey, ) @@ -1272,7 +1289,7 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt // // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by // LinkEndpoint.LinkAddress) before reaching this point. - linkAddr := ndp.nic.linkEP.LinkAddress() + linkAddr := ndp.ep.nic.LinkAddress() if !header.IsValidUnicastEthernetAddress(linkAddr) { return false } @@ -1291,15 +1308,15 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt PrefixLen: validPrefixLenForAutoGen, } - if !ndp.nic.hasPermanentAddrLocked(generatedAddr.Address) { + if !ndp.ep.hasPermanentAddressRLocked(generatedAddr.Address) { break } state.stableAddr.localGenerationFailures++ } - if ref := ndp.addSLAACAddr(generatedAddr, slaac, time.Since(state.preferredUntil) >= 0 /* deprecated */); ref != nil { - state.stableAddr.ref = ref + if addressEndpoint := ndp.addAndAcquireSLAACAddr(generatedAddr, stack.AddressConfigSlaac, time.Since(state.preferredUntil) >= 0 /* deprecated */); addressEndpoint != nil { + state.stableAddr.addressEndpoint = addressEndpoint state.generationAttempts++ return true } @@ -1309,10 +1326,9 @@ func (ndp *ndpState) generateSLAACAddr(prefix tcpip.Subnet, state *slaacPrefixSt // regenerateSLAACAddr regenerates an address for a SLAAC prefix. // -// If generating a new address for the prefix fails, the prefix will be -// invalidated. +// If generating a new address for the prefix fails, the prefix is invalidated. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) regenerateSLAACAddr(prefix tcpip.Subnet) { state, ok := ndp.slaacPrefixes[prefix] if !ok { @@ -1332,7 +1348,7 @@ func (ndp *ndpState) regenerateSLAACAddr(prefix tcpip.Subnet) { // generateTempSLAACAddr generates a new temporary SLAAC address. // -// If resetGenAttempts is true, the prefix's generation counter will be reset. +// If resetGenAttempts is true, the prefix's generation counter is reset. // // Returns true if a new address was generated. func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *slaacPrefixState, resetGenAttempts bool) bool { @@ -1353,7 +1369,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla return false } - stableAddr := prefixState.stableAddr.ref.address() + stableAddr := prefixState.stableAddr.addressEndpoint.AddressWithPrefix().Address now := time.Now() // As per RFC 4941 section 3.3 step 4, the valid lifetime of a temporary @@ -1392,7 +1408,8 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla return false } - // Attempt to generate a new address that is not already assigned to the NIC. + // Attempt to generate a new address that is not already assigned to the IPv6 + // endpoint. var generatedAddr tcpip.AddressWithPrefix for i := 0; ; i++ { // If we were unable to generate an address after the maximum SLAAC address @@ -1402,7 +1419,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla } generatedAddr = header.GenerateTempIPv6SLAACAddr(ndp.temporaryIIDHistory[:], stableAddr) - if !ndp.nic.hasPermanentAddrLocked(generatedAddr.Address) { + if !ndp.ep.hasPermanentAddressRLocked(generatedAddr.Address) { break } } @@ -1410,13 +1427,13 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla // As per RFC RFC 4941 section 3.3 step 5, we MUST NOT create a temporary // address with a zero preferred lifetime. The checks above ensure this // so we know the address is not deprecated. - ref := ndp.addSLAACAddr(generatedAddr, slaacTemp, false /* deprecated */) - if ref == nil { + addressEndpoint := ndp.addAndAcquireSLAACAddr(generatedAddr, stack.AddressConfigSlaacTemp, false /* deprecated */) + if addressEndpoint == nil { return false } state := tempSLAACAddrState{ - deprecationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + deprecationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to deprecate temporary address %s", prefix, generatedAddr)) @@ -1427,9 +1444,9 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla panic(fmt.Sprintf("ndp: must have a tempAddr entry to deprecate temporary address %s", generatedAddr)) } - ndp.deprecateSLAACAddress(tempAddrState.ref) + ndp.deprecateSLAACAddress(tempAddrState.addressEndpoint) }), - invalidationJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + invalidationJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to invalidate temporary address %s", prefix, generatedAddr)) @@ -1442,7 +1459,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla ndp.invalidateTempSLAACAddr(prefixState.tempAddrs, generatedAddr.Address, tempAddrState) }), - regenJob: ndp.nic.stack.newJob(&ndp.nic.mu, func() { + regenJob: ndp.ep.protocol.stack.NewJob(&ndp.ep.mu, func() { prefixState, ok := ndp.slaacPrefixes[prefix] if !ok { panic(fmt.Sprintf("ndp: must have a slaacPrefixes entry for %s to regenerate temporary address after %s", prefix, generatedAddr)) @@ -1465,8 +1482,8 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla prefixState.tempAddrs[generatedAddr.Address] = tempAddrState ndp.slaacPrefixes[prefix] = prefixState }), - createdAt: now, - ref: ref, + createdAt: now, + addressEndpoint: addressEndpoint, } state.deprecationJob.Schedule(pl) @@ -1481,7 +1498,7 @@ func (ndp *ndpState) generateTempSLAACAddr(prefix tcpip.Subnet, prefixState *sla // regenerateTempSLAACAddr regenerates a temporary address for a SLAAC prefix. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) regenerateTempSLAACAddr(prefix tcpip.Subnet, resetGenAttempts bool) { state, ok := ndp.slaacPrefixes[prefix] if !ok { @@ -1496,14 +1513,14 @@ func (ndp *ndpState) regenerateTempSLAACAddr(prefix tcpip.Subnet, resetGenAttemp // // pl is the new preferred lifetime. vl is the new valid lifetime. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixState *slaacPrefixState, pl, vl time.Duration) { // If the preferred lifetime is zero, then the prefix should be deprecated. deprecated := pl == 0 if deprecated { - ndp.deprecateSLAACAddress(prefixState.stableAddr.ref) + ndp.deprecateSLAACAddress(prefixState.stableAddr.addressEndpoint) } else { - prefixState.stableAddr.ref.deprecated = false + prefixState.stableAddr.addressEndpoint.SetDeprecated(false) } // If prefix was preferred for some finite lifetime before, cancel the @@ -1565,7 +1582,7 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // If DAD is not yet complete on the stable address, there is no need to do // work with temporary addresses. - if prefixState.stableAddr.ref.getKind() != permanent { + if prefixState.stableAddr.addressEndpoint.GetKind() != stack.Permanent { return } @@ -1608,9 +1625,9 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat newPreferredLifetime := preferredUntil.Sub(now) tempAddrState.deprecationJob.Cancel() if newPreferredLifetime <= 0 { - ndp.deprecateSLAACAddress(tempAddrState.ref) + ndp.deprecateSLAACAddress(tempAddrState.addressEndpoint) } else { - tempAddrState.ref.deprecated = false + tempAddrState.addressEndpoint.SetDeprecated(false) tempAddrState.deprecationJob.Schedule(newPreferredLifetime) } @@ -1635,8 +1652,8 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat // due to an update in preferred lifetime. // // If each temporay address has already been regenerated, no new temporary - // address will be generated. To ensure continuation of temporary SLAAC - // addresses, we manually try to regenerate an address here. + // address is generated. To ensure continuation of temporary SLAAC addresses, + // we manually try to regenerate an address here. if len(regenForAddr) != 0 || allAddressesRegenerated { // Reset the generation attempts counter as we are starting the generation // of a new address for the SLAAC prefix. @@ -1647,57 +1664,58 @@ func (ndp *ndpState) refreshSLAACPrefixLifetimes(prefix tcpip.Subnet, prefixStat } } -// deprecateSLAACAddress marks ref as deprecated and notifies the stack's NDP -// dispatcher that ref has been deprecated. +// deprecateSLAACAddress marks the address as deprecated and notifies the NDP +// dispatcher that address has been deprecated. // -// deprecateSLAACAddress does nothing if ref is already deprecated. +// deprecateSLAACAddress does nothing if the address is already deprecated. // -// The NIC that ndp belongs to MUST be locked. -func (ndp *ndpState) deprecateSLAACAddress(ref *referencedNetworkEndpoint) { - if ref.deprecated { +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) deprecateSLAACAddress(addressEndpoint stack.AddressEndpoint) { + if addressEndpoint.Deprecated() { return } - ref.deprecated = true - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnAutoGenAddressDeprecated(ndp.nic.ID(), ref.addrWithPrefix()) + addressEndpoint.SetDeprecated(true) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnAutoGenAddressDeprecated(ndp.ep.nic.ID(), addressEndpoint.AddressWithPrefix()) } } // invalidateSLAACPrefix invalidates a SLAAC prefix. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) invalidateSLAACPrefix(prefix tcpip.Subnet, state slaacPrefixState) { - if r := state.stableAddr.ref; r != nil { + ndp.cleanupSLAACPrefixResources(prefix, state) + + if addressEndpoint := state.stableAddr.addressEndpoint; addressEndpoint != nil { // Since we are already invalidating the prefix, do not invalidate the // prefix when removing the address. - if err := ndp.nic.removePermanentIPv6EndpointLocked(r, false /* allowSLAACInvalidation */); err != nil { - panic(fmt.Sprintf("ndp: error removing stable SLAAC address %s: %s", r.addrWithPrefix(), err)) + if err := ndp.ep.removePermanentEndpointLocked(addressEndpoint, false /* allowSLAACInvalidation */); err != nil { + panic(fmt.Sprintf("ndp: error removing stable SLAAC address %s: %s", addressEndpoint.AddressWithPrefix(), err)) } } - - ndp.cleanupSLAACPrefixResources(prefix, state) } // cleanupSLAACAddrResourcesAndNotify cleans up an invalidated SLAAC address's // resources. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPrefix, invalidatePrefix bool) { - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnAutoGenAddressInvalidated(ndp.nic.ID(), addr) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnAutoGenAddressInvalidated(ndp.ep.nic.ID(), addr) } prefix := addr.Subnet() state, ok := ndp.slaacPrefixes[prefix] - if !ok || state.stableAddr.ref == nil || addr.Address != state.stableAddr.ref.address() { + if !ok || state.stableAddr.addressEndpoint == nil || addr.Address != state.stableAddr.addressEndpoint.AddressWithPrefix().Address { return } if !invalidatePrefix { // If the prefix is not being invalidated, disassociate the address from the // prefix and do nothing further. - state.stableAddr.ref = nil + state.stableAddr.addressEndpoint.DecRef() + state.stableAddr.addressEndpoint = nil ndp.slaacPrefixes[prefix] = state return } @@ -1709,14 +1727,17 @@ func (ndp *ndpState) cleanupSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPr // // Panics if the SLAAC prefix is not known. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaacPrefixState) { // Invalidate all temporary addresses. for tempAddr, tempAddrState := range state.tempAddrs { ndp.invalidateTempSLAACAddr(state.tempAddrs, tempAddr, tempAddrState) } - state.stableAddr.ref = nil + if state.stableAddr.addressEndpoint != nil { + state.stableAddr.addressEndpoint.DecRef() + state.stableAddr.addressEndpoint = nil + } state.deprecationJob.Cancel() state.invalidationJob.Cancel() delete(ndp.slaacPrefixes, prefix) @@ -1724,12 +1745,12 @@ func (ndp *ndpState) cleanupSLAACPrefixResources(prefix tcpip.Subnet, state slaa // invalidateTempSLAACAddr invalidates a temporary SLAAC address. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) invalidateTempSLAACAddr(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) { // Since we are already invalidating the address, do not invalidate the // address when removing the address. - if err := ndp.nic.removePermanentIPv6EndpointLocked(tempAddrState.ref, false /* allowSLAACInvalidation */); err != nil { - panic(fmt.Sprintf("error removing temporary SLAAC address %s: %s", tempAddrState.ref.addrWithPrefix(), err)) + if err := ndp.ep.removePermanentEndpointLocked(tempAddrState.addressEndpoint, false /* allowSLAACInvalidation */); err != nil { + panic(fmt.Sprintf("error removing temporary SLAAC address %s: %s", tempAddrState.addressEndpoint.AddressWithPrefix(), err)) } ndp.cleanupTempSLAACAddrResources(tempAddrs, tempAddr, tempAddrState) @@ -1738,10 +1759,10 @@ func (ndp *ndpState) invalidateTempSLAACAddr(tempAddrs map[tcpip.Address]tempSLA // cleanupTempSLAACAddrResourcesAndNotify cleans up an invalidated temporary // SLAAC address's resources from ndp. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWithPrefix, invalidateAddr bool) { - if ndpDisp := ndp.nic.stack.ndpDisp; ndpDisp != nil { - ndpDisp.OnAutoGenAddressInvalidated(ndp.nic.ID(), addr) + if ndpDisp := ndp.ep.protocol.ndpDisp; ndpDisp != nil { + ndpDisp.OnAutoGenAddressInvalidated(ndp.ep.nic.ID(), addr) } if !invalidateAddr { @@ -1765,35 +1786,29 @@ func (ndp *ndpState) cleanupTempSLAACAddrResourcesAndNotify(addr tcpip.AddressWi // cleanupTempSLAACAddrResourcesAndNotify cleans up a temporary SLAAC address's // jobs and entry. // -// The NIC that ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupTempSLAACAddrResources(tempAddrs map[tcpip.Address]tempSLAACAddrState, tempAddr tcpip.Address, tempAddrState tempSLAACAddrState) { + tempAddrState.addressEndpoint.DecRef() + tempAddrState.addressEndpoint = nil tempAddrState.deprecationJob.Cancel() tempAddrState.invalidationJob.Cancel() tempAddrState.regenJob.Cancel() delete(tempAddrs, tempAddr) } -// cleanupState cleans up ndp's state. -// -// If hostOnly is true, then only host-specific state will be cleaned up. +// removeSLAACAddresses removes all SLAAC addresses. // -// cleanupState MUST be called with hostOnly set to true when ndp's NIC is -// transitioning from a host to a router. This function will invalidate all -// discovered on-link prefixes, discovered routers, and auto-generated -// addresses. -// -// If hostOnly is true, then the link-local auto-generated address will not be -// invalidated as routers are also expected to generate a link-local address. +// If keepLinkLocal is false, the SLAAC generated link-local address is removed. // -// The NIC that ndp belongs to MUST be locked. -func (ndp *ndpState) cleanupState(hostOnly bool) { +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) removeSLAACAddresses(keepLinkLocal bool) { linkLocalSubnet := header.IPv6LinkLocalPrefix.Subnet() - linkLocalPrefixes := 0 + var linkLocalPrefixes int for prefix, state := range ndp.slaacPrefixes { // RFC 4862 section 5 states that routers are also expected to generate a // link-local address so we do not invalidate them if we are cleaning up // host-only state. - if hostOnly && prefix == linkLocalSubnet { + if keepLinkLocal && prefix == linkLocalSubnet { linkLocalPrefixes++ continue } @@ -1804,6 +1819,21 @@ func (ndp *ndpState) cleanupState(hostOnly bool) { if got := len(ndp.slaacPrefixes); got != linkLocalPrefixes { panic(fmt.Sprintf("ndp: still have non-linklocal SLAAC prefixes after cleaning up; found = %d prefixes, of which %d are link-local", got, linkLocalPrefixes)) } +} + +// cleanupState cleans up ndp's state. +// +// If hostOnly is true, then only host-specific state is cleaned up. +// +// This function invalidates all discovered on-link prefixes, discovered +// routers, and auto-generated addresses. +// +// If hostOnly is true, then the link-local auto-generated address aren't +// invalidated as routers are also expected to generate a link-local address. +// +// The IPv6 endpoint that ndp belongs to MUST be locked. +func (ndp *ndpState) cleanupState(hostOnly bool) { + ndp.removeSLAACAddresses(hostOnly /* keepLinkLocal */) for prefix := range ndp.onLinkPrefixes { ndp.invalidateOnLinkPrefix(prefix) @@ -1827,7 +1857,7 @@ func (ndp *ndpState) cleanupState(hostOnly bool) { // startSolicitingRouters starts soliciting routers, as per RFC 4861 section // 6.3.7. If routers are already being solicited, this function does nothing. // -// The NIC ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) startSolicitingRouters() { if ndp.rtrSolicit.timer != nil { // We are already soliciting routers. @@ -1848,27 +1878,37 @@ func (ndp *ndpState) startSolicitingRouters() { var done bool ndp.rtrSolicit.done = &done - ndp.rtrSolicit.timer = ndp.nic.stack.Clock().AfterFunc(delay, func() { - ndp.nic.mu.Lock() + 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 NIC lock and stopped solicitations. - // Simply return here and do nothing further. - ndp.nic.mu.Unlock() + // goroutine already obtained the IPv6 endpoint lock and stopped + // solicitations. Simply return here and do nothing further. + ndp.ep.mu.Unlock() return } // 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. - ref := ndp.nic.primaryIPv6EndpointRLocked(header.IPv6AllRoutersMulticastAddress) - if ref == nil { - ref = ndp.nic.getRefOrCreateTempLocked(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint) + 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.nic.mu.Unlock() + ndp.ep.mu.Unlock() - localAddr := ref.address() - r := makeRoute(header.IPv6ProtocolNumber, localAddr, header.IPv6AllRoutersMulticastAddress, ndp.nic.linkEP.LinkAddress(), ref, false, false) + 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 @@ -1876,15 +1916,16 @@ func (ndp *ndpState) startSolicitingRouters() { // 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 NIC is not locked, - // the NIC could have been disabled or removed by another goroutine. + // 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.nic.ID(), err)) + 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.nic.ID())) + 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())) } // As per RFC 4861 section 4.1, an NDP RS SHOULD include the source @@ -1907,21 +1948,20 @@ func (ndp *ndpState) startSolicitingRouters() { rs.Options().Serialize(optsSerializer) icmpData.SetChecksum(header.ICMPv6Checksum(icmpData, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - pkt := NewPacketBuffer(PacketBufferOptions{ + pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ ReserveHeaderBytes: int(r.MaxHeaderLength()), Data: buffer.View(icmpData).ToVectorisedView(), }) sent := r.Stats().ICMP.V6PacketsSent if err := r.WritePacket(nil, - NetworkHeaderParams{ + stack.NetworkHeaderParams{ Protocol: header.ICMPv6ProtocolNumber, TTL: header.NDPHopLimit, - TOS: DefaultTOS, }, pkt, ); err != nil { sent.Dropped.Increment() - log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.nic.ID(), err) + 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. remaining = 0 } else { @@ -1929,19 +1969,19 @@ func (ndp *ndpState) startSolicitingRouters() { remaining-- } - ndp.nic.mu.Lock() + 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 NIC + // 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) } - ndp.nic.mu.Unlock() + ndp.ep.mu.Unlock() }) } @@ -1949,7 +1989,7 @@ func (ndp *ndpState) startSolicitingRouters() { // stopSolicitingRouters stops soliciting routers. If routers are not currently // being solicited, this function does nothing. // -// The NIC ndp belongs to MUST be locked. +// The IPv6 endpoint that ndp belongs to MUST be locked. func (ndp *ndpState) stopSolicitingRouters() { if ndp.rtrSolicit.timer == nil { // Nothing to do. @@ -1965,7 +2005,7 @@ func (ndp *ndpState) stopSolicitingRouters() { // initializeTempAddrState initializes state related to temporary SLAAC // addresses. func (ndp *ndpState) initializeTempAddrState() { - header.InitialTempIID(ndp.temporaryIIDHistory[:], ndp.nic.stack.tempIIDSeed, ndp.nic.ID()) + header.InitialTempIID(ndp.temporaryIIDHistory[:], ndp.ep.protocol.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 7434df4a1..9033a9ed5 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -17,6 +17,7 @@ package ipv6 import ( "strings" "testing" + "time" "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" @@ -35,8 +36,8 @@ func setupStackAndEndpoint(t *testing.T, llladdr, rlladdr tcpip.Address, useNeig t.Helper() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol6}, UseNeighborCache: useNeighborCache, }) @@ -65,10 +66,94 @@ func setupStackAndEndpoint(t *testing.T, llladdr, rlladdr tcpip.Address, useNeig t.Fatalf("cannot find protocol instance for network protocol %d", ProtocolNumber) } - ep := netProto.NewEndpoint(0, &stubLinkAddressCache{}, &stubNUDHandler{}, &stubDispatcher{}, nil, s) + ep := netProto.NewEndpoint(&testInterface{}, &stubLinkAddressCache{}, &stubNUDHandler{}, &stubDispatcher{}) + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + t.Cleanup(ep.Close) + return s, ep } +var _ NDPDispatcher = (*testNDPDispatcher)(nil) + +// testNDPDispatcher is an NDPDispatcher only allows default router discovery. +type testNDPDispatcher struct { + addr tcpip.Address +} + +func (*testNDPDispatcher) OnDuplicateAddressDetectionStatus(tcpip.NICID, tcpip.Address, bool, *tcpip.Error) { +} + +func (t *testNDPDispatcher) OnDefaultRouterDiscovered(_ tcpip.NICID, addr tcpip.Address) bool { + t.addr = addr + return true +} + +func (t *testNDPDispatcher) OnDefaultRouterInvalidated(_ tcpip.NICID, addr tcpip.Address) { + t.addr = addr +} + +func (*testNDPDispatcher) OnOnLinkPrefixDiscovered(tcpip.NICID, tcpip.Subnet) bool { + return false +} + +func (*testNDPDispatcher) OnOnLinkPrefixInvalidated(tcpip.NICID, tcpip.Subnet) { +} + +func (*testNDPDispatcher) OnAutoGenAddress(tcpip.NICID, tcpip.AddressWithPrefix) bool { + return false +} + +func (*testNDPDispatcher) OnAutoGenAddressDeprecated(tcpip.NICID, tcpip.AddressWithPrefix) { +} + +func (*testNDPDispatcher) OnAutoGenAddressInvalidated(tcpip.NICID, tcpip.AddressWithPrefix) { +} + +func (*testNDPDispatcher) OnRecursiveDNSServerOption(tcpip.NICID, []tcpip.Address, time.Duration) { +} + +func (*testNDPDispatcher) OnDNSSearchListOption(tcpip.NICID, []string, time.Duration) { +} + +func (*testNDPDispatcher) OnDHCPv6Configuration(tcpip.NICID, DHCPv6ConfigurationFromNDPRA) { +} + +func TestStackNDPEndpointInvalidateDefaultRouter(t *testing.T) { + var ndpDisp testNDPDispatcher + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocolWithOptions(Options{ + NDPDisp: &ndpDisp, + })}, + }) + + if err := s.CreateNIC(nicID, &stubLinkEndpoint{}); err != nil { + t.Fatalf("s.CreateNIC(%d, _): %s", nicID, err) + } + + ep, err := s.GetNetworkEndpoint(nicID, ProtocolNumber) + if err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, ProtocolNumber, err) + } + + ipv6EP := ep.(*endpoint) + ipv6EP.mu.Lock() + ipv6EP.mu.ndp.rememberDefaultRouter(lladdr1, time.Hour) + ipv6EP.mu.Unlock() + + if ndpDisp.addr != lladdr1 { + t.Fatalf("got ndpDisp.addr = %s, want = %s", ndpDisp.addr, lladdr1) + } + + ndpDisp.addr = "" + ndpEP := ep.(stack.NDPEndpoint) + ndpEP.InvalidateDefaultRouter(lladdr1) + if ndpDisp.addr != lladdr1 { + t.Fatalf("got ndpDisp.addr = %s, want = %s", ndpDisp.addr, lladdr1) + } +} + // TestNeighorSolicitationWithSourceLinkLayerOption tests that receiving a // valid NDP NS message with the Source Link Layer Address option results in a // new entry in the link address cache for the sender of the message. @@ -98,7 +183,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOption(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) e := channel.New(0, 1280, linkAddr0) if err := s.CreateNIC(nicID, e); err != nil { @@ -202,7 +287,7 @@ func TestNeighorSolicitationWithSourceLinkLayerOptionUsingNeighborCache(t *testi for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: true, }) e := channel.New(0, 1280, linkAddr0) @@ -325,7 +410,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { naDst tcpip.Address }{ { - name: "Unspecified source to multicast destination", + name: "Unspecified source to solicited-node multicast destination", nsOpts: nil, nsSrcLinkAddr: remoteLinkAddr0, nsSrc: header.IPv6Any, @@ -352,11 +437,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { nsSrcLinkAddr: remoteLinkAddr0, nsSrc: header.IPv6Any, nsDst: nicAddr, - nsInvalid: false, - naDstLinkAddr: remoteLinkAddr0, - naSolicited: false, - naSrc: nicAddr, - naDst: header.IPv6AllNodesMulticastAddress, + nsInvalid: true, }, { name: "Unspecified source with source ll option to unicast destination", @@ -475,7 +556,7 @@ func TestNeighorSolicitationResponse(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: stackTyp.useNeighborCache, }) e := channel.New(1, 1280, nicLinkAddr) @@ -596,7 +677,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOption(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, }) e := channel.New(0, 1280, linkAddr0) if err := s.CreateNIC(nicID, e); err != nil { @@ -707,7 +788,7 @@ func TestNeighorAdvertisementWithTargetLinkLayerOptionUsingNeighborCache(t *test for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: true, }) e := channel.New(0, 1280, linkAddr0) @@ -1172,7 +1253,7 @@ func TestRouterAdvertValidation(t *testing.T) { e := channel.New(10, 1280, linkAddr1) e.LinkEPCapabilities |= stack.CapabilityResolutionRequired s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{NewProtocol}, UseNeighborCache: stackTyp.useNeighborCache, }) diff --git a/pkg/tcpip/network/testutil/BUILD b/pkg/tcpip/network/testutil/BUILD index e218563d0..d0ffc299a 100644 --- a/pkg/tcpip/network/testutil/BUILD +++ b/pkg/tcpip/network/testutil/BUILD @@ -7,11 +7,15 @@ go_library( srcs = [ "testutil.go", ], - visibility = ["//pkg/tcpip/network/ipv4:__pkg__"], + visibility = [ + "//pkg/tcpip/network/fragmentation:__pkg__", + "//pkg/tcpip/network/ipv4:__pkg__", + "//pkg/tcpip/network/ipv6:__pkg__", + ], deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", - "//pkg/tcpip/link/channel", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/network/testutil/testutil.go b/pkg/tcpip/network/testutil/testutil.go index bf5ce74be..7cc52985e 100644 --- a/pkg/tcpip/network/testutil/testutil.go +++ b/pkg/tcpip/network/testutil/testutil.go @@ -22,48 +22,100 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) -// TestEndpoint is an endpoint used for testing, it stores packets written to it -// and can mock errors. -type TestEndpoint struct { - *channel.Endpoint - - // WrittenPackets is where we store packets written via WritePacket(). +// MockLinkEndpoint is an endpoint used for testing, it stores packets written +// to it and can mock errors. +type MockLinkEndpoint struct { + // WrittenPackets is where packets written to the endpoint are stored. WrittenPackets []*stack.PacketBuffer - packetCollectorErrors []*tcpip.Error + mtu uint32 + err *tcpip.Error + allowPackets int } -// NewTestEndpoint creates a new TestEndpoint endpoint. +// NewMockLinkEndpoint creates a new MockLinkEndpoint. // -// packetCollectorErrors can be used to set error values and each call to -// WritePacket will remove the first one from the slice and return it until -// the slice is empty - at that point it will return nil every time. -func NewTestEndpoint(ep *channel.Endpoint, packetCollectorErrors []*tcpip.Error) *TestEndpoint { - return &TestEndpoint{ - Endpoint: ep, - WrittenPackets: make([]*stack.PacketBuffer, 0), - packetCollectorErrors: packetCollectorErrors, +// err is the error that will be returned once allowPackets packets are written +// to the endpoint. +func NewMockLinkEndpoint(mtu uint32, err *tcpip.Error, allowPackets int) *MockLinkEndpoint { + return &MockLinkEndpoint{ + mtu: mtu, + err: err, + allowPackets: allowPackets, + } +} + +// MTU implements LinkEndpoint.MTU. +func (ep *MockLinkEndpoint) MTU() uint32 { return ep.mtu } + +// Capabilities implements LinkEndpoint.Capabilities. +func (*MockLinkEndpoint) Capabilities() stack.LinkEndpointCapabilities { return 0 } + +// MaxHeaderLength implements LinkEndpoint.MaxHeaderLength. +func (*MockLinkEndpoint) MaxHeaderLength() uint16 { return 0 } + +// LinkAddress implements LinkEndpoint.LinkAddress. +func (*MockLinkEndpoint) LinkAddress() tcpip.LinkAddress { return "" } + +// WritePacket implements LinkEndpoint.WritePacket. +func (ep *MockLinkEndpoint) WritePacket(_ *stack.Route, _ *stack.GSO, _ tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { + if ep.allowPackets == 0 { + return ep.err } + ep.allowPackets-- + ep.WrittenPackets = append(ep.WrittenPackets, pkt) + return nil } -// WritePacket stores outbound packets and may return an error if one was -// injected. -func (e *TestEndpoint) WritePacket(_ *stack.Route, _ *stack.GSO, _ tcpip.NetworkProtocolNumber, pkt *stack.PacketBuffer) *tcpip.Error { - e.WrittenPackets = append(e.WrittenPackets, pkt) +// WritePackets implements LinkEndpoint.WritePackets. +func (ep *MockLinkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + var n int - if len(e.packetCollectorErrors) > 0 { - nextError := e.packetCollectorErrors[0] - e.packetCollectorErrors = e.packetCollectorErrors[1:] - return nextError + for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() { + if err := ep.WritePacket(r, gso, protocol, pkt); err != nil { + return n, err + } + n++ } + 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) {} + +// IsAttached implements LinkEndpoint.IsAttached. +func (*MockLinkEndpoint) IsAttached() bool { return false } + +// Wait implements LinkEndpoint.Wait. +func (*MockLinkEndpoint) Wait() {} + +// ARPHardwareType implements LinkEndpoint.ARPHardwareType. +func (*MockLinkEndpoint) ARPHardwareType() header.ARPHardwareType { return header.ARPHardwareNone } + +// AddHeader implements LinkEndpoint.AddHeader. +func (*MockLinkEndpoint) AddHeader(_, _ tcpip.LinkAddress, _ tcpip.NetworkProtocolNumber, _ *stack.PacketBuffer) { +} + // MakeRandPkt generates a randomized packet. transportHeaderLength indicates // how many random bytes will be copied in the Transport Header. // extraHeaderReserveLength indicates how much extra space will be reserved for diff --git a/pkg/tcpip/sample/tun_tcp_connect/main.go b/pkg/tcpip/sample/tun_tcp_connect/main.go index 91fc26722..51d428049 100644 --- a/pkg/tcpip/sample/tun_tcp_connect/main.go +++ b/pkg/tcpip/sample/tun_tcp_connect/main.go @@ -127,8 +127,8 @@ func main() { // Create the stack with ipv4 and tcp protocols, then add a tun-based // NIC and ipv4 address. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) mtu, err := rawfile.GetMTU(tunName) diff --git a/pkg/tcpip/sample/tun_tcp_echo/main.go b/pkg/tcpip/sample/tun_tcp_echo/main.go index 3f58a15ea..8e0ee1cd7 100644 --- a/pkg/tcpip/sample/tun_tcp_echo/main.go +++ b/pkg/tcpip/sample/tun_tcp_echo/main.go @@ -112,8 +112,8 @@ func main() { // Create the stack with ip and tcp protocols, then add a tun-based // NIC and address. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol, arp.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) mtu, err := rawfile.GetMTU(tunName) diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 900938dd1..eba97334e 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -54,9 +54,8 @@ go_template_instance( go_library( name = "stack", srcs = [ + "addressable_endpoint_state.go", "conntrack.go", - "dhcpv6configurationfromndpra_string.go", - "forwarder.go", "headertype_string.go", "icmp_rate_limit.go", "iptables.go", @@ -65,7 +64,6 @@ go_library( "iptables_types.go", "linkaddrcache.go", "linkaddrentry_list.go", - "ndp.go", "neighbor_cache.go", "neighbor_entry.go", "neighbor_entry_list.go", @@ -74,6 +72,7 @@ go_library( "nud.go", "packet_buffer.go", "packet_buffer_list.go", + "pending_packets.go", "rand.go", "registration.go", "route.go", @@ -106,6 +105,7 @@ go_test( name = "stack_x_test", size = "medium", srcs = [ + "addressable_endpoint_state_test.go", "ndp_test.go", "nud_test.go", "stack_test.go", @@ -116,13 +116,13 @@ go_test( deps = [ ":stack", "//pkg/rand", + "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/checker", "//pkg/tcpip/header", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/loopback", - "//pkg/tcpip/network/arp", "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", "//pkg/tcpip/ports", @@ -138,8 +138,7 @@ go_test( name = "stack_test", size = "small", srcs = [ - "fake_time_test.go", - "forwarder_test.go", + "forwarding_test.go", "linkaddrcache_test.go", "neighbor_cache_test.go", "neighbor_entry_test.go", @@ -152,8 +151,8 @@ go_test( "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/faketime", "//pkg/tcpip/header", - "@com_github_dpjacques_clockwork//:go_default_library", "@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 new file mode 100644 index 000000000..4d3acab96 --- /dev/null +++ b/pkg/tcpip/stack/addressable_endpoint_state.go @@ -0,0 +1,753 @@ +// 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 stack + +import ( + "fmt" + + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" +) + +var _ GroupAddressableEndpoint = (*AddressableEndpointState)(nil) +var _ AddressableEndpoint = (*AddressableEndpointState)(nil) + +// AddressableEndpointState is an implementation of an AddressableEndpoint. +type AddressableEndpointState struct { + networkEndpoint NetworkEndpoint + + // Lock ordering (from outer to inner lock ordering): + // + // AddressableEndpointState.mu + // addressState.mu + mu struct { + sync.RWMutex + + 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 + } +} + +// Init initializes the AddressableEndpointState with networkEndpoint. +// +// Must be called before calling any other function on m. +func (a *AddressableEndpointState) Init(networkEndpoint NetworkEndpoint) { + a.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. +// +// 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. +// +// 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() + + ep, ok := m.inner.mu.endpoints[addr] + if !ok { + return nil + } + return ep +} + +// ForEach calls f for each address pair. +// +// 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() + + for _, ep := range m.inner.mu.endpoints { + if !f(ep) { + return + } + } +} + +// 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 { + 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() + a.releaseAddressStateLocked(addrState) +} + +// releaseAddressState removes addrState from s's address state (primary and endpoints list). +// +// Preconditions: a.mu must be write locked. +func (a *AddressableEndpointState) releaseAddressStateLocked(addrState *addressState) { + oldPrimary := a.mu.primary + for i, s := range a.mu.primary { + if s == addrState { + a.mu.primary = append(a.mu.primary[:i], a.mu.primary[i+1:]...) + oldPrimary[len(oldPrimary)-1] = nil + break + } + } + delete(a.mu.endpoints, addrState.addr.Address) +} + +// AddAndAcquirePermanentAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb PrimaryEndpointBehavior, configType AddressConfigType, deprecated bool) (AddressEndpoint, *tcpip.Error) { + a.mu.Lock() + defer a.mu.Unlock() + ep, err := a.addAndAcquireAddressLocked(addr, peb, configType, deprecated, true /* permanent */) + // From https://golang.org/doc/faq#nil_error: + // + // Under the covers, interfaces are implemented as two elements, a type T and + // a value V. + // + // An interface value is nil only if the V and T are both unset, (T=nil, V is + // not set), In particular, a nil interface will always hold a nil type. If we + // store a nil pointer of type *int inside an interface value, the inner type + // will be *int regardless of the value of the pointer: (T=*int, V=nil). Such + // an interface value will therefore be non-nil even when the pointer value V + // inside is nil. + // + // Since addAndAcquireAddressLocked returns a nil value with a non-nil type, + // we need to explicitly return nil below if ep is (a typed) nil. + if ep == nil { + return nil, err + } + return ep, err +} + +// AddAndAcquireTemporaryAddress adds a temporary address. +// +// Returns tcpip.ErrDuplicateAddress if the address exists. +// +// The temporary address's endpoint is acquired and returned. +func (a *AddressableEndpointState) AddAndAcquireTemporaryAddress(addr tcpip.AddressWithPrefix, peb PrimaryEndpointBehavior) (AddressEndpoint, *tcpip.Error) { + a.mu.Lock() + defer a.mu.Unlock() + ep, err := a.addAndAcquireAddressLocked(addr, peb, AddressConfigStatic, false /* deprecated */, false /* permanent */) + // From https://golang.org/doc/faq#nil_error: + // + // Under the covers, interfaces are implemented as two elements, a type T and + // a value V. + // + // An interface value is nil only if the V and T are both unset, (T=nil, V is + // not set), In particular, a nil interface will always hold a nil type. If we + // store a nil pointer of type *int inside an interface value, the inner type + // will be *int regardless of the value of the pointer: (T=*int, V=nil). Such + // an interface value will therefore be non-nil even when the pointer value V + // inside is nil. + // + // Since addAndAcquireAddressLocked returns a nil value with a non-nil type, + // we need to explicitly return nil below if ep is (a typed) nil. + if ep == nil { + return nil, err + } + return ep, err +} + +// addAndAcquireAddressLocked adds, acquires and returns a permanent or +// temporary address. +// +// If the addressable endpoint already has the address in a non-permanent state, +// and addAndAcquireAddressLocked is adding a permanent address, that address is +// promoted in place and its properties set to the properties provided. If the +// address already exists in any other state, then tcpip.ErrDuplicateAddress is +// returned, regardless the kind of address that is being added. +// +// Precondition: a.mu must be write locked. +func (a *AddressableEndpointState) addAndAcquireAddressLocked(addr tcpip.AddressWithPrefix, peb PrimaryEndpointBehavior, configType AddressConfigType, deprecated, permanent bool) (*addressState, *tcpip.Error) { + // attemptAddToPrimary is false when the address is already in the primary + // address list. + attemptAddToPrimary := true + addrState, ok := a.mu.endpoints[addr.Address] + if ok { + if !permanent { + // We are adding a non-permanent address but the address exists. No need + // to go any further since we can only promote existing temporary/expired + // addresses to permanent. + return nil, tcpip.ErrDuplicateAddress + } + + addrState.mu.Lock() + if addrState.mu.kind.IsPermanent() { + addrState.mu.Unlock() + // We are adding a permanent address but a permanent address already + // exists. + return nil, tcpip.ErrDuplicateAddress + } + + if addrState.mu.refs == 0 { + panic(fmt.Sprintf("found an address that should have been released (ref count == 0); address = %s", addrState.addr)) + } + + // We now promote the address. + for i, s := range a.mu.primary { + if s == addrState { + switch peb { + case CanBePrimaryEndpoint: + // The address is already in the primary address list. + attemptAddToPrimary = false + case FirstPrimaryEndpoint: + if i == 0 { + // The address is already first in the primary address list. + attemptAddToPrimary = false + } else { + a.mu.primary = append(a.mu.primary[:i], a.mu.primary[i+1:]...) + } + case NeverPrimaryEndpoint: + a.mu.primary = append(a.mu.primary[:i], a.mu.primary[i+1:]...) + default: + panic(fmt.Sprintf("unrecognized primary endpoint behaviour = %d", peb)) + } + break + } + } + } + + if addrState == nil { + addrState = &addressState{ + addressableEndpointState: a, + addr: addr, + } + a.mu.endpoints[addr.Address] = addrState + addrState.mu.Lock() + // We never promote an address to temporary - it can only be added as such. + // If we are actaully adding a permanent address, it is promoted below. + addrState.mu.kind = Temporary + } + + // At this point we have an address we are either promoting from an expired or + // temporary address to permanent, promoting an expired address to temporary, + // or we are adding a new temporary or permanent address. + // + // The address MUST be write locked at this point. + defer addrState.mu.Unlock() + + if permanent { + if addrState.mu.kind.IsPermanent() { + panic(fmt.Sprintf("only non-permanent addresses should be promoted to permanent; address = %s", addrState.addr)) + } + + // Primary addresses are biased by 1. + addrState.mu.refs++ + addrState.mu.kind = Permanent + } + // Acquire the address before returning it. + addrState.mu.refs++ + addrState.mu.deprecated = deprecated + addrState.mu.configType = configType + + if attemptAddToPrimary { + switch peb { + case NeverPrimaryEndpoint: + case CanBePrimaryEndpoint: + a.mu.primary = append(a.mu.primary, addrState) + case FirstPrimaryEndpoint: + if cap(a.mu.primary) == len(a.mu.primary) { + a.mu.primary = append([]*addressState{addrState}, a.mu.primary...) + } else { + // Shift all the endpoints by 1 to make room for the new address at the + // front. We could have just created a new slice but this saves + // allocations when the slice has capacity for the new address. + primaryCount := len(a.mu.primary) + a.mu.primary = append(a.mu.primary, nil) + if n := copy(a.mu.primary[1:], a.mu.primary); n != primaryCount { + panic(fmt.Sprintf("copied %d elements; expected = %d elements", n, primaryCount)) + } + a.mu.primary[0] = addrState + } + default: + panic(fmt.Sprintf("unrecognized primary endpoint behaviour = %d", peb)) + } + } + + return addrState, nil +} + +// RemovePermanentAddress implements AddressableEndpoint. +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) +} + +// removePermanentAddressLocked is like RemovePermanentAddress but with locking +// requirements. +// +// Precondition: a.mu must be write locked. +func (a *AddressableEndpointState) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { + addrState, ok := a.mu.endpoints[addr] + if !ok { + return tcpip.ErrBadLocalAddress + } + + return a.removePermanentEndpointLocked(addrState) +} + +// RemovePermanentEndpoint removes the passed endpoint if it is associated with +// a and permanent. +func (a *AddressableEndpointState) RemovePermanentEndpoint(ep AddressEndpoint) *tcpip.Error { + addrState, ok := ep.(*addressState) + if !ok || addrState.addressableEndpointState != a { + return tcpip.ErrInvalidEndpointState + } + + return a.removePermanentEndpointLocked(addrState) +} + +// removePermanentAddressLocked is like RemovePermanentAddress but with locking +// requirements. +// +// Precondition: a.mu must be write locked. +func (a *AddressableEndpointState) removePermanentEndpointLocked(addrState *addressState) *tcpip.Error { + if !addrState.GetKind().IsPermanent() { + return tcpip.ErrBadLocalAddress + } + + addrState.SetKind(PermanentExpired) + a.decAddressRefLocked(addrState) + return nil +} + +// decAddressRef decrements the address's reference count and releases it once +// the reference count hits 0. +func (a *AddressableEndpointState) decAddressRef(addrState *addressState) { + a.mu.Lock() + defer a.mu.Unlock() + a.decAddressRefLocked(addrState) +} + +// decAddressRefLocked is like decAddressRef but with locking requirements. +// +// Precondition: a.mu must be write locked. +func (a *AddressableEndpointState) decAddressRefLocked(addrState *addressState) { + addrState.mu.Lock() + defer addrState.mu.Unlock() + + if addrState.mu.refs == 0 { + panic(fmt.Sprintf("attempted to decrease ref count for AddressEndpoint w/ addr = %s when it is already released", addrState.addr)) + } + + addrState.mu.refs-- + + if addrState.mu.refs != 0 { + return + } + + // A non-expired permanent address must not have its reference count dropped + // to 0. + if addrState.mu.kind.IsPermanent() { + panic(fmt.Sprintf("permanent addresses should be removed through the AddressableEndpoint: addr = %s, kind = %d", addrState.addr, addrState.mu.kind)) + } + + a.releaseAddressStateLocked(addrState) +} + +// MainAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) MainAddress() tcpip.AddressWithPrefix { + a.mu.RLock() + defer a.mu.RUnlock() + + ep := a.acquirePrimaryAddressRLocked(func(ep *addressState) bool { + return ep.GetKind() == Permanent + }) + if ep == nil { + return tcpip.AddressWithPrefix{} + } + + addr := ep.AddressWithPrefix() + a.decAddressRefLocked(ep) + return addr +} + +// acquirePrimaryAddressRLocked returns an acquired primary address that is +// valid according to isValid. +// +// Precondition: e.mu must be read locked +func (a *AddressableEndpointState) acquirePrimaryAddressRLocked(isValid func(*addressState) bool) *addressState { + var deprecatedEndpoint *addressState + for _, ep := range a.mu.primary { + if !isValid(ep) { + continue + } + + if !ep.Deprecated() { + if ep.IncRef() { + // ep is not deprecated, so return it immediately. + // + // If we kept track of a deprecated endpoint, decrement its reference + // count since it was incremented when we decided to keep track of it. + if deprecatedEndpoint != nil { + a.decAddressRefLocked(deprecatedEndpoint) + deprecatedEndpoint = nil + } + + return ep + } + } else if deprecatedEndpoint == nil && ep.IncRef() { + // We prefer an endpoint that is not deprecated, but we keep track of + // ep in case a doesn't have any non-deprecated endpoints. + // + // If we end up finding a more preferred endpoint, ep's reference count + // will be decremented. + deprecatedEndpoint = ep + } + } + + return deprecatedEndpoint +} + +// AcquireAssignedAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB PrimaryEndpointBehavior) AddressEndpoint { + a.mu.Lock() + defer a.mu.Unlock() + + if addrState, ok := a.mu.endpoints[localAddr]; ok { + if !addrState.IsAssigned(allowTemp) { + return nil + } + + if !addrState.IncRef() { + panic(fmt.Sprintf("failed to increase the reference count for address = %s", addrState.addr)) + } + + return addrState + } + + if !allowTemp { + return nil + } + + addr := localAddr.WithPrefix() + ep, err := a.addAndAcquireAddressLocked(addr, tempPEB, AddressConfigStatic, false /* deprecated */, false /* permanent */) + if err != nil { + // addAndAcquireAddressLocked 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("a.addAndAcquireAddressLocked(%s, %d, %d, false, false): %s", addr, tempPEB, AddressConfigStatic, err)) + } + // From https://golang.org/doc/faq#nil_error: + // + // Under the covers, interfaces are implemented as two elements, a type T and + // a value V. + // + // An interface value is nil only if the V and T are both unset, (T=nil, V is + // not set), In particular, a nil interface will always hold a nil type. If we + // store a nil pointer of type *int inside an interface value, the inner type + // will be *int regardless of the value of the pointer: (T=*int, V=nil). Such + // an interface value will therefore be non-nil even when the pointer value V + // inside is nil. + // + // Since addAndAcquireAddressLocked returns a nil value with a non-nil type, + // we need to explicitly return nil below if ep is (a typed) nil. + if ep == nil { + return nil + } + return ep +} + +// AcquireOutgoingPrimaryAddress implements AddressableEndpoint. +func (a *AddressableEndpointState) AcquireOutgoingPrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) AddressEndpoint { + a.mu.RLock() + defer a.mu.RUnlock() + + ep := a.acquirePrimaryAddressRLocked(func(ep *addressState) bool { + return ep.IsAssigned(allowExpired) + }) + + // From https://golang.org/doc/faq#nil_error: + // + // Under the covers, interfaces are implemented as two elements, a type T and + // a value V. + // + // An interface value is nil only if the V and T are both unset, (T=nil, V is + // not set), In particular, a nil interface will always hold a nil type. If we + // store a nil pointer of type *int inside an interface value, the inner type + // will be *int regardless of the value of the pointer: (T=*int, V=nil). Such + // an interface value will therefore be non-nil even when the pointer value V + // inside is nil. + // + // Since acquirePrimaryAddressRLocked returns a nil value with a non-nil type, + // we need to explicitly return nil below if ep is (a typed) nil. + if ep == nil { + return nil + } + + return ep +} + +// PrimaryAddresses implements AddressableEndpoint. +func (a *AddressableEndpointState) PrimaryAddresses() []tcpip.AddressWithPrefix { + a.mu.RLock() + defer a.mu.RUnlock() + + var addrs []tcpip.AddressWithPrefix + for _, ep := range a.mu.primary { + // Don't include tentative, expired or temporary endpoints + // to avoid confusion and prevent the caller from using + // those. + switch ep.GetKind() { + case PermanentTentative, PermanentExpired, Temporary: + continue + } + + addrs = append(addrs, ep.AddressWithPrefix()) + } + + return addrs +} + +// PermanentAddresses implements AddressableEndpoint. +func (a *AddressableEndpointState) PermanentAddresses() []tcpip.AddressWithPrefix { + a.mu.RLock() + defer a.mu.RUnlock() + + var addrs []tcpip.AddressWithPrefix + for _, ep := range a.mu.endpoints { + if !ep.GetKind().IsPermanent() { + continue + } + + addrs = append(addrs, ep.AddressWithPrefix()) + } + + 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. + if err := a.removePermanentEndpointLocked(ep); err != nil && err != tcpip.ErrBadLocalAddress { + panic(fmt.Sprintf("unexpected error from removePermanentEndpointLocked(%s): %s", ep.addr, err)) + } + } +} + +var _ AddressEndpoint = (*addressState)(nil) + +// addressState holds state for an address. +type addressState struct { + addressableEndpointState *AddressableEndpointState + addr tcpip.AddressWithPrefix + + // Lock ordering (from outer to inner lock ordering): + // + // AddressableEndpointState.mu + // addressState.mu + mu struct { + sync.RWMutex + + refs uint32 + kind AddressKind + configType AddressConfigType + deprecated bool + } +} + +// AddressWithPrefix implements AddressEndpoint. +func (a *addressState) AddressWithPrefix() tcpip.AddressWithPrefix { + return a.addr +} + +// GetKind implements AddressEndpoint. +func (a *addressState) GetKind() AddressKind { + a.mu.RLock() + defer a.mu.RUnlock() + return a.mu.kind +} + +// SetKind implements AddressEndpoint. +func (a *addressState) SetKind(kind AddressKind) { + a.mu.Lock() + defer a.mu.Unlock() + a.mu.kind = kind +} + +// IsAssigned implements AddressEndpoint. +func (a *addressState) IsAssigned(allowExpired bool) bool { + if !a.addressableEndpointState.networkEndpoint.Enabled() { + return false + } + + switch a.GetKind() { + case PermanentTentative: + return false + case PermanentExpired: + return allowExpired + default: + return true + } +} + +// IncRef implements AddressEndpoint. +func (a *addressState) IncRef() bool { + a.mu.Lock() + defer a.mu.Unlock() + if a.mu.refs == 0 { + return false + } + + a.mu.refs++ + return true +} + +// DecRef implements AddressEndpoint. +func (a *addressState) DecRef() { + a.addressableEndpointState.decAddressRef(a) +} + +// ConfigType implements AddressEndpoint. +func (a *addressState) ConfigType() AddressConfigType { + a.mu.RLock() + defer a.mu.RUnlock() + return a.mu.configType +} + +// SetDeprecated implements AddressEndpoint. +func (a *addressState) SetDeprecated(d bool) { + a.mu.Lock() + defer a.mu.Unlock() + a.mu.deprecated = d +} + +// Deprecated implements AddressEndpoint. +func (a *addressState) Deprecated() bool { + a.mu.RLock() + defer a.mu.RUnlock() + return a.mu.deprecated +} diff --git a/pkg/tcpip/stack/addressable_endpoint_state_test.go b/pkg/tcpip/stack/addressable_endpoint_state_test.go new file mode 100644 index 000000000..26787d0a3 --- /dev/null +++ b/pkg/tcpip/stack/addressable_endpoint_state_test.go @@ -0,0 +1,77 @@ +// 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 stack_test + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +// TestAddressableEndpointStateCleanup tests that cleaning up an addressable +// endpoint state removes permanent addresses and leaves groups. +func TestAddressableEndpointStateCleanup(t *testing.T) { + var ep fakeNetworkEndpoint + if err := ep.Enable(); err != nil { + t.Fatalf("ep.Enable(): %s", err) + } + + var s stack.AddressableEndpointState + s.Init(&ep) + + addr := tcpip.AddressWithPrefix{ + Address: "\x01", + PrefixLen: 8, + } + + { + ep, err := s.AddAndAcquirePermanentAddress(addr, stack.NeverPrimaryEndpoint, stack.AddressConfigStatic, false /* deprecated */) + if err != nil { + t.Fatalf("s.AddAndAcquirePermanentAddress(%s, %d, %d, false): %s", addr, stack.NeverPrimaryEndpoint, stack.AddressConfigStatic, err) + } + // We don't need the address endpoint. + ep.DecRef() + } + { + ep := s.AcquireAssignedAddress(addr.Address, false /* allowTemp */, stack.NeverPrimaryEndpoint) + if ep == nil { + t.Fatalf("got s.AcquireAssignedAddress(%s, false, NeverPrimaryEndpoint) = nil, want = non-nil", addr.Address) + } + 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) + } +} diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index 836682ea0..0cd1da11f 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -196,13 +196,14 @@ type bucket struct { // packetToTupleID converts packet to a tuple ID. It fails when pkt lacks a valid // TCP header. +// +// Preconditions: pkt.NetworkHeader() is valid. func packetToTupleID(pkt *PacketBuffer) (tupleID, *tcpip.Error) { - // TODO(gvisor.dev/issue/170): Need to support for other - // protocols as well. - netHeader := header.IPv4(pkt.NetworkHeader().View()) - if len(netHeader) < header.IPv4MinimumSize || netHeader.TransportProtocol() != header.TCPProtocolNumber { + netHeader := pkt.Network() + if netHeader.TransportProtocol() != header.TCPProtocolNumber { return tupleID{}, tcpip.ErrUnknownProtocol } + tcpHeader := header.TCP(pkt.TransportHeader().View()) if len(tcpHeader) < header.TCPMinimumSize { return tupleID{}, tcpip.ErrUnknownProtocol @@ -214,7 +215,7 @@ func packetToTupleID(pkt *PacketBuffer) (tupleID, *tcpip.Error) { dstAddr: netHeader.DestinationAddress(), dstPort: tcpHeader.DestinationPort(), transProto: netHeader.TransportProtocol(), - netProto: header.IPv4ProtocolNumber, + netProto: pkt.NetworkProtocolNumber, }, nil } @@ -268,7 +269,7 @@ func (ct *ConnTrack) connForTID(tid tupleID) (*conn, direction) { return nil, dirOriginal } -func (ct *ConnTrack) insertRedirectConn(pkt *PacketBuffer, hook Hook, rt RedirectTarget) *conn { +func (ct *ConnTrack) insertRedirectConn(pkt *PacketBuffer, hook Hook, rt *RedirectTarget) *conn { tid, err := packetToTupleID(pkt) if err != nil { return nil @@ -281,8 +282,8 @@ func (ct *ConnTrack) insertRedirectConn(pkt *PacketBuffer, hook Hook, rt Redirec // rule. This tuple will be used to manipulate the packet in // handlePacket. replyTID := tid.reply() - replyTID.srcAddr = rt.MinIP - replyTID.srcPort = rt.MinPort + replyTID.srcAddr = rt.Addr + replyTID.srcPort = rt.Port var manip manipType switch hook { case Prerouting: @@ -344,7 +345,7 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { return } - netHeader := header.IPv4(pkt.NetworkHeader().View()) + netHeader := pkt.Network() tcpHeader := header.TCP(pkt.TransportHeader().View()) // For prerouting redirection, packets going in the original direction @@ -366,8 +367,12 @@ func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) { // support cases when they are validated, e.g. when we can't offload // receive checksumming. - netHeader.SetChecksum(0) - netHeader.SetChecksum(^netHeader.CalculateChecksum()) + // After modification, IPv4 packets need a valid checksum. + if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber { + netHeader := header.IPv4(pkt.NetworkHeader().View()) + netHeader.SetChecksum(0) + netHeader.SetChecksum(^netHeader.CalculateChecksum()) + } } // handlePacketOutput manipulates ports for packets in Output hook. @@ -377,7 +382,7 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d return } - netHeader := header.IPv4(pkt.NetworkHeader().View()) + netHeader := pkt.Network() tcpHeader := header.TCP(pkt.TransportHeader().View()) // For output redirection, packets going in the original direction @@ -396,7 +401,7 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d // Calculate the TCP checksum and set it. tcpHeader.SetChecksum(0) - length := uint16(pkt.Size()) - uint16(netHeader.HeaderLength()) + length := uint16(pkt.Size()) - uint16(len(pkt.NetworkHeader().View())) xsum := r.PseudoHeaderChecksum(header.TCPProtocolNumber, length) if gso != nil && gso.NeedsCsum { tcpHeader.SetChecksum(xsum) @@ -405,8 +410,11 @@ func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir d tcpHeader.SetChecksum(^tcpHeader.CalculateChecksum(xsum)) } - netHeader.SetChecksum(0) - netHeader.SetChecksum(^netHeader.CalculateChecksum()) + if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber { + netHeader := header.IPv4(pkt.NetworkHeader().View()) + netHeader.SetChecksum(0) + netHeader.SetChecksum(^netHeader.CalculateChecksum()) + } } // handlePacket will manipulate the port and address of the packet if the @@ -422,7 +430,7 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou } // TODO(gvisor.dev/issue/170): Support other transport protocols. - if nh := pkt.NetworkHeader().View(); nh.IsEmpty() || header.IPv4(nh).TransportProtocol() != header.TCPProtocolNumber { + if pkt.Network().TransportProtocol() != header.TCPProtocolNumber { return false } @@ -473,7 +481,7 @@ func (ct *ConnTrack) maybeInsertNoop(pkt *PacketBuffer, hook Hook) { } // We only track TCP connections. - if nh := pkt.NetworkHeader().View(); nh.IsEmpty() || header.IPv4(nh).TransportProtocol() != header.TCPProtocolNumber { + if pkt.Network().TransportProtocol() != header.TCPProtocolNumber { return } @@ -609,7 +617,7 @@ func (ct *ConnTrack) reapTupleLocked(tuple *tuple, bucket int, now time.Time) bo return true } -func (ct *ConnTrack) originalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) { +func (ct *ConnTrack) originalDst(epID TransportEndpointID, netProto tcpip.NetworkProtocolNumber) (tcpip.Address, uint16, *tcpip.Error) { // Lookup the connection. The reply's original destination // describes the original address. tid := tupleID{ @@ -618,7 +626,7 @@ func (ct *ConnTrack) originalDst(epID TransportEndpointID) (tcpip.Address, uint1 dstAddr: epID.RemoteAddress, dstPort: epID.RemotePort, transProto: header.TCPProtocolNumber, - netProto: header.IPv4ProtocolNumber, + netProto: netProto, } conn, _ := ct.connForTID(tid) if conn == nil { diff --git a/pkg/tcpip/stack/forwarder_test.go b/pkg/tcpip/stack/forwarding_test.go index 38c5bac71..cf042309e 100644 --- a/pkg/tcpip/stack/forwarder_test.go +++ b/pkg/tcpip/stack/forwarding_test.go @@ -20,6 +20,7 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -45,20 +46,27 @@ const ( // use the first three: destination address, source address, and transport // protocol. They're all one byte fields to simplify parsing. type fwdTestNetworkEndpoint struct { - nicID tcpip.NICID + AddressableEndpointState + + nic NetworkInterface proto *fwdTestNetworkProtocol dispatcher TransportDispatcher - ep LinkEndpoint } var _ NetworkEndpoint = (*fwdTestNetworkEndpoint)(nil) -func (f *fwdTestNetworkEndpoint) MTU() uint32 { - return f.ep.MTU() - uint32(f.MaxHeaderLength()) +func (*fwdTestNetworkEndpoint) Enable() *tcpip.Error { + return nil } -func (f *fwdTestNetworkEndpoint) NICID() tcpip.NICID { - return f.nicID +func (*fwdTestNetworkEndpoint) Enabled() bool { + return true +} + +func (*fwdTestNetworkEndpoint) Disable() {} + +func (f *fwdTestNetworkEndpoint) MTU() uint32 { + return f.nic.MTU() - uint32(f.MaxHeaderLength()) } func (*fwdTestNetworkEndpoint) DefaultTTL() uint8 { @@ -71,17 +79,13 @@ func (f *fwdTestNetworkEndpoint) HandlePacket(r *Route, pkt *PacketBuffer) { } func (f *fwdTestNetworkEndpoint) MaxHeaderLength() uint16 { - return f.ep.MaxHeaderLength() + fwdTestNetHeaderLen + return f.nic.MaxHeaderLength() + fwdTestNetHeaderLen } func (f *fwdTestNetworkEndpoint) PseudoHeaderChecksum(protocol tcpip.TransportProtocolNumber, dstAddr tcpip.Address) uint16 { return 0 } -func (f *fwdTestNetworkEndpoint) Capabilities() LinkEndpointCapabilities { - return f.ep.Capabilities() -} - func (f *fwdTestNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return f.proto.Number() } @@ -94,7 +98,7 @@ func (f *fwdTestNetworkEndpoint) WritePacket(r *Route, gso *GSO, params NetworkH b[srcAddrOffset] = r.LocalAddress[0] b[protocolNumberOffset] = byte(params.Protocol) - return f.ep.WritePacket(r, gso, fwdTestNetNumber, pkt) + return f.nic.WritePacket(r, gso, fwdTestNetNumber, pkt) } // WritePackets implements LinkEndpoint.WritePackets. @@ -106,7 +110,9 @@ func (*fwdTestNetworkEndpoint) WriteHeaderIncludedPacket(r *Route, pkt *PacketBu return tcpip.ErrNotSupported } -func (*fwdTestNetworkEndpoint) Close() {} +func (f *fwdTestNetworkEndpoint) Close() { + f.AddressableEndpointState.Cleanup() +} // fwdTestNetworkProtocol is a network-layer protocol that implements Address // resolution. @@ -116,6 +122,11 @@ type fwdTestNetworkProtocol struct { addrResolveDelay time.Duration onLinkAddressResolved func(cache *linkAddrCache, neigh *neighborCache, addr tcpip.Address, _ tcpip.LinkAddress) onResolveStaticAddress func(tcpip.Address) (tcpip.LinkAddress, bool) + + mu struct { + sync.RWMutex + forwarding bool + } } var _ NetworkProtocol = (*fwdTestNetworkProtocol)(nil) @@ -145,13 +156,14 @@ func (*fwdTestNetworkProtocol) Parse(pkt *PacketBuffer) (tcpip.TransportProtocol return tcpip.TransportProtocolNumber(netHeader[protocolNumberOffset]), true, true } -func (f *fwdTestNetworkProtocol) NewEndpoint(nicID tcpip.NICID, _ LinkAddressCache, _ NUDHandler, dispatcher TransportDispatcher, ep LinkEndpoint, _ *Stack) NetworkEndpoint { - return &fwdTestNetworkEndpoint{ - nicID: nicID, +func (f *fwdTestNetworkProtocol) NewEndpoint(nic NetworkInterface, _ LinkAddressCache, _ NUDHandler, dispatcher TransportDispatcher) NetworkEndpoint { + e := &fwdTestNetworkEndpoint{ + nic: nic, proto: f, dispatcher: dispatcher, - ep: ep, } + e.AddressableEndpointState.Init(e) + return e } func (*fwdTestNetworkProtocol) SetOption(tcpip.SettableNetworkProtocolOption) *tcpip.Error { @@ -186,6 +198,21 @@ func (*fwdTestNetworkProtocol) LinkAddressProtocol() tcpip.NetworkProtocolNumber return fwdTestNetNumber } +// Forwarding implements stack.ForwardingNetworkProtocol. +func (f *fwdTestNetworkProtocol) Forwarding() bool { + f.mu.RLock() + defer f.mu.RUnlock() + return f.mu.forwarding + +} + +// SetForwarding implements stack.ForwardingNetworkProtocol. +func (f *fwdTestNetworkProtocol) SetForwarding(v bool) { + f.mu.Lock() + defer f.mu.Unlock() + f.mu.forwarding = v +} + // fwdTestPacketInfo holds all the information about an outbound packet. type fwdTestPacketInfo struct { RemoteLinkAddress tcpip.LinkAddress @@ -307,7 +334,7 @@ func (e *fwdTestLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protoco func fwdTestNetFactory(t *testing.T, proto *fwdTestNetworkProtocol, useNeighborCache bool) (ep1, ep2 *fwdTestLinkEndpoint) { // Create a stack with the network protocol and two NICs. s := New(Options{ - NetworkProtocols: []NetworkProtocol{proto}, + NetworkProtocols: []NetworkProtocolFactory{func(*Stack) NetworkProtocol { return proto }}, UseNeighborCache: useNeighborCache, }) diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index 4a521eca9..8d6d9a7f1 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -60,11 +60,11 @@ func DefaultTables() *IPTables { v4Tables: [numTables]Table{ natID: Table{ Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &ErrorTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, }, BuiltinChains: [NumHooks]int{ Prerouting: 0, @@ -83,9 +83,9 @@ func DefaultTables() *IPTables { }, mangleID: Table{ Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &ErrorTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, }, BuiltinChains: [NumHooks]int{ Prerouting: 0, @@ -101,10 +101,10 @@ func DefaultTables() *IPTables { }, filterID: Table{ Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, + Rule{Target: &ErrorTarget{NetworkProtocol: header.IPv4ProtocolNumber}}, }, BuiltinChains: [NumHooks]int{ Prerouting: HookUnset, @@ -125,11 +125,11 @@ func DefaultTables() *IPTables { v6Tables: [numTables]Table{ natID: Table{ Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &ErrorTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, }, BuiltinChains: [NumHooks]int{ Prerouting: 0, @@ -148,9 +148,9 @@ func DefaultTables() *IPTables { }, mangleID: Table{ Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &ErrorTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, }, BuiltinChains: [NumHooks]int{ Prerouting: 0, @@ -166,10 +166,10 @@ func DefaultTables() *IPTables { }, filterID: Table{ Rules: []Rule{ - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: AcceptTarget{}}, - Rule{Target: ErrorTarget{}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &AcceptTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, + Rule{Target: &ErrorTarget{NetworkProtocol: header.IPv6ProtocolNumber}}, }, BuiltinChains: [NumHooks]int{ Prerouting: HookUnset, @@ -502,11 +502,11 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx // OriginalDst returns the original destination of redirected connections. It // returns an error if the connection doesn't exist or isn't redirected. -func (it *IPTables) OriginalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) { +func (it *IPTables) OriginalDst(epID TransportEndpointID, netProto tcpip.NetworkProtocolNumber) (tcpip.Address, uint16, *tcpip.Error) { it.mu.RLock() defer it.mu.RUnlock() if !it.modified { return "", 0, tcpip.ErrNotConnected } - return it.connections.originalDst(epID) + return it.connections.originalDst(epID, netProto) } diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go index 5f1b2af64..538c4625d 100644 --- a/pkg/tcpip/stack/iptables_targets.go +++ b/pkg/tcpip/stack/iptables_targets.go @@ -21,78 +21,139 @@ import ( ) // AcceptTarget accepts packets. -type AcceptTarget struct{} +type AcceptTarget struct { + // NetworkProtocol is the network protocol the target is used with. + NetworkProtocol tcpip.NetworkProtocolNumber +} + +// ID implements Target.ID. +func (at *AcceptTarget) ID() TargetID { + return TargetID{ + NetworkProtocol: at.NetworkProtocol, + } +} // Action implements Target.Action. -func (AcceptTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (*AcceptTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleAccept, 0 } // DropTarget drops packets. -type DropTarget struct{} +type DropTarget struct { + // NetworkProtocol is the network protocol the target is used with. + NetworkProtocol tcpip.NetworkProtocolNumber +} + +// ID implements Target.ID. +func (dt *DropTarget) ID() TargetID { + return TargetID{ + NetworkProtocol: dt.NetworkProtocol, + } +} // Action implements Target.Action. -func (DropTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (*DropTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleDrop, 0 } +// ErrorTargetName is used to mark targets as error targets. Error targets +// shouldn't be reached - an error has occurred if we fall through to one. +const ErrorTargetName = "ERROR" + // ErrorTarget logs an error and drops the packet. It represents a target that // should be unreachable. -type ErrorTarget struct{} +type ErrorTarget struct { + // NetworkProtocol is the network protocol the target is used with. + NetworkProtocol tcpip.NetworkProtocolNumber +} + +// ID implements Target.ID. +func (et *ErrorTarget) ID() TargetID { + return TargetID{ + Name: ErrorTargetName, + NetworkProtocol: et.NetworkProtocol, + } +} // Action implements Target.Action. -func (ErrorTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (*ErrorTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { log.Debugf("ErrorTarget triggered.") return RuleDrop, 0 } // UserChainTarget marks a rule as the beginning of a user chain. type UserChainTarget struct { + // Name is the chain name. Name string + + // NetworkProtocol is the network protocol the target is used with. + NetworkProtocol tcpip.NetworkProtocolNumber +} + +// ID implements Target.ID. +func (uc *UserChainTarget) ID() TargetID { + return TargetID{ + Name: ErrorTargetName, + NetworkProtocol: uc.NetworkProtocol, + } } // Action implements Target.Action. -func (UserChainTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (*UserChainTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { panic("UserChainTarget should never be called.") } // ReturnTarget returns from the current chain. If the chain is a built-in, the // hook's underflow should be called. -type ReturnTarget struct{} +type ReturnTarget struct { + // NetworkProtocol is the network protocol the target is used with. + NetworkProtocol tcpip.NetworkProtocolNumber +} + +// ID implements Target.ID. +func (rt *ReturnTarget) ID() TargetID { + return TargetID{ + NetworkProtocol: rt.NetworkProtocol, + } +} // Action implements Target.Action. -func (ReturnTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { +func (*ReturnTarget) Action(*PacketBuffer, *ConnTrack, Hook, *GSO, *Route, tcpip.Address) (RuleVerdict, int) { return RuleReturn, 0 } +// RedirectTargetName is used to mark targets as redirect targets. Redirect +// targets should be reached for only NAT and Mangle tables. These targets will +// change the destination port/destination IP for packets. +const RedirectTargetName = "REDIRECT" + // RedirectTarget redirects the packet by modifying the destination port/IP. -// Min and Max values for IP and Ports in the struct indicate the range of -// values which can be used to redirect. +// TODO(gvisor.dev/issue/170): Other flags need to be added after we support +// them. type RedirectTarget struct { - // TODO(gvisor.dev/issue/170): Other flags need to be added after - // we support them. - // RangeProtoSpecified flag indicates single port is specified to - // redirect. - RangeProtoSpecified bool + // Addr indicates address used to redirect. + Addr tcpip.Address - // MinIP indicates address used to redirect. - MinIP tcpip.Address + // Port indicates port used to redirect. + Port uint16 - // MaxIP indicates address used to redirect. - MaxIP tcpip.Address - - // MinPort indicates port used to redirect. - MinPort uint16 + // NetworkProtocol is the network protocol the target is used with. + NetworkProtocol tcpip.NetworkProtocolNumber +} - // MaxPort indicates port used to redirect. - MaxPort uint16 +// ID implements Target.ID. +func (rt *RedirectTarget) ID() TargetID { + return TargetID{ + Name: RedirectTargetName, + NetworkProtocol: rt.NetworkProtocol, + } } // Action implements Target.Action. // TODO(gvisor.dev/issue/170): Parse headers without copying. The current // implementation only works for PREROUTING and calls pkt.Clone(), neither // of which should be the case. -func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { +func (rt *RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) { // Packet is already manipulated. if pkt.NatDone { return RuleAccept, 0 @@ -103,34 +164,35 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso return RuleDrop, 0 } - // Change the address to localhost (127.0.0.1) in Output and - // to primary address of the incoming interface in Prerouting. + // Change the address to localhost (127.0.0.1 or ::1) in Output and to + // the primary address of the incoming interface in Prerouting. switch hook { case Output: - rt.MinIP = tcpip.Address([]byte{127, 0, 0, 1}) - rt.MaxIP = tcpip.Address([]byte{127, 0, 0, 1}) + if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber { + rt.Addr = tcpip.Address([]byte{127, 0, 0, 1}) + } else { + rt.Addr = header.IPv6Loopback + } case Prerouting: - rt.MinIP = address - rt.MaxIP = address + rt.Addr = address default: panic("redirect target is supported only on output and prerouting hooks") } // TODO(gvisor.dev/issue/170): Check Flags in RedirectTarget if // we need to change dest address (for OUTPUT chain) or ports. - netHeader := header.IPv4(pkt.NetworkHeader().View()) - switch protocol := netHeader.TransportProtocol(); protocol { + switch protocol := pkt.TransportProtocolNumber; protocol { case header.UDPProtocolNumber: udpHeader := header.UDP(pkt.TransportHeader().View()) - udpHeader.SetDestinationPort(rt.MinPort) + udpHeader.SetDestinationPort(rt.Port) // Calculate UDP checksum and set it. if hook == Output { udpHeader.SetChecksum(0) - length := uint16(pkt.Size()) - uint16(netHeader.HeaderLength()) // Only calculate the checksum if offloading isn't supported. if r.Capabilities()&CapabilityTXChecksumOffload == 0 { + length := uint16(pkt.Size()) - uint16(len(pkt.NetworkHeader().View())) xsum := r.PseudoHeaderChecksum(protocol, length) for _, v := range pkt.Data.Views() { xsum = header.Checksum(v, xsum) @@ -139,10 +201,15 @@ func (rt RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso udpHeader.SetChecksum(^udpHeader.CalculateChecksum(xsum)) } } - // Change destination address. - netHeader.SetDestinationAddress(rt.MinIP) - netHeader.SetChecksum(0) - netHeader.SetChecksum(^netHeader.CalculateChecksum()) + + pkt.Network().SetDestinationAddress(rt.Addr) + + // After modification, IPv4 packets need a valid checksum. + if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber { + netHeader := header.IPv4(pkt.NetworkHeader().View()) + netHeader.SetChecksum(0) + netHeader.SetChecksum(^netHeader.CalculateChecksum()) + } pkt.NatDone = true case header.TCPProtocolNumber: if ct == nil { diff --git a/pkg/tcpip/stack/iptables_types.go b/pkg/tcpip/stack/iptables_types.go index 093ee6881..7b3f3e88b 100644 --- a/pkg/tcpip/stack/iptables_types.go +++ b/pkg/tcpip/stack/iptables_types.go @@ -104,8 +104,20 @@ type IPTables struct { reaperDone chan struct{} } -// A Table defines a set of chains and hooks into the network stack. It is -// really just a list of rules. +// A Table defines a set of chains and hooks into the network stack. +// +// It is a list of Rules, entry points (BuiltinChains), and error handlers +// (Underflows). As packets traverse netstack, they hit hooks. When a packet +// hits a hook, iptables compares it to Rules starting from that hook's entry +// point. So if a packet hits the Input hook, we look up the corresponding +// entry point in BuiltinChains and jump to that point. +// +// If the Rule doesn't match the packet, iptables continues to the next Rule. +// If a Rule does match, it can issue a verdict on the packet (e.g. RuleAccept +// or RuleDrop) that causes the packet to stop traversing iptables. It can also +// jump to other rules or perform custom actions based on Rule.Target. +// +// Underflow Rules are invoked when a chain returns without reaching a verdict. // // +stateify savable type Table struct { @@ -260,6 +272,18 @@ func (fl IPHeaderFilter) match(pkt *PacketBuffer, hook Hook, nicName string) boo return true } +// NetworkProtocol returns the protocol (IPv4 or IPv6) on to which the header +// applies. +func (fl IPHeaderFilter) NetworkProtocol() tcpip.NetworkProtocolNumber { + switch len(fl.Src) { + case header.IPv4AddressSize: + return header.IPv4ProtocolNumber + case header.IPv6AddressSize: + return header.IPv6ProtocolNumber + } + panic(fmt.Sprintf("invalid address in IPHeaderFilter: %s", fl.Src)) +} + // filterAddress returns whether addr matches the filter. func filterAddress(addr, mask, filterAddr tcpip.Address, invert bool) bool { matches := true @@ -285,8 +309,23 @@ type Matcher interface { Match(hook Hook, packet *PacketBuffer, interfaceName string) (matches bool, hotdrop bool) } +// A TargetID uniquely identifies a target. +type TargetID struct { + // Name is the target name as stored in the xt_entry_target struct. + Name string + + // NetworkProtocol is the protocol to which the target applies. + NetworkProtocol tcpip.NetworkProtocolNumber + + // Revision is the version of the target. + Revision uint8 +} + // A Target is the interface for taking an action for a packet. type Target interface { + // ID uniquely identifies the Target. + ID() TargetID + // Action takes an action on the packet and returns a verdict on how // traversal should (or should not) continue. If the return value is // Jump, it also returns the index of the rule to jump to. diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 5e43a9b0b..73a01c2dd 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -150,10 +150,10 @@ type ndpDNSSLEvent struct { type ndpDHCPv6Event struct { nicID tcpip.NICID - configuration stack.DHCPv6ConfigurationFromNDPRA + configuration ipv6.DHCPv6ConfigurationFromNDPRA } -var _ stack.NDPDispatcher = (*ndpDispatcher)(nil) +var _ ipv6.NDPDispatcher = (*ndpDispatcher)(nil) // ndpDispatcher implements NDPDispatcher so tests can know when various NDP // related events happen for test purposes. @@ -170,7 +170,7 @@ type ndpDispatcher struct { dhcpv6ConfigurationC chan ndpDHCPv6Event } -// Implements stack.NDPDispatcher.OnDuplicateAddressDetectionStatus. +// Implements ipv6.NDPDispatcher.OnDuplicateAddressDetectionStatus. func (n *ndpDispatcher) OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, addr tcpip.Address, resolved bool, err *tcpip.Error) { if n.dadC != nil { n.dadC <- ndpDADEvent{ @@ -182,7 +182,7 @@ func (n *ndpDispatcher) OnDuplicateAddressDetectionStatus(nicID tcpip.NICID, add } } -// Implements stack.NDPDispatcher.OnDefaultRouterDiscovered. +// Implements ipv6.NDPDispatcher.OnDefaultRouterDiscovered. func (n *ndpDispatcher) OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip.Address) bool { if c := n.routerC; c != nil { c <- ndpRouterEvent{ @@ -195,7 +195,7 @@ func (n *ndpDispatcher) OnDefaultRouterDiscovered(nicID tcpip.NICID, addr tcpip. return n.rememberRouter } -// Implements stack.NDPDispatcher.OnDefaultRouterInvalidated. +// Implements ipv6.NDPDispatcher.OnDefaultRouterInvalidated. func (n *ndpDispatcher) OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip.Address) { if c := n.routerC; c != nil { c <- ndpRouterEvent{ @@ -206,7 +206,7 @@ func (n *ndpDispatcher) OnDefaultRouterInvalidated(nicID tcpip.NICID, addr tcpip } } -// Implements stack.NDPDispatcher.OnOnLinkPrefixDiscovered. +// Implements ipv6.NDPDispatcher.OnOnLinkPrefixDiscovered. func (n *ndpDispatcher) OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip.Subnet) bool { if c := n.prefixC; c != nil { c <- ndpPrefixEvent{ @@ -219,7 +219,7 @@ func (n *ndpDispatcher) OnOnLinkPrefixDiscovered(nicID tcpip.NICID, prefix tcpip return n.rememberPrefix } -// Implements stack.NDPDispatcher.OnOnLinkPrefixInvalidated. +// Implements ipv6.NDPDispatcher.OnOnLinkPrefixInvalidated. func (n *ndpDispatcher) OnOnLinkPrefixInvalidated(nicID tcpip.NICID, prefix tcpip.Subnet) { if c := n.prefixC; c != nil { c <- ndpPrefixEvent{ @@ -261,7 +261,7 @@ func (n *ndpDispatcher) OnAutoGenAddressInvalidated(nicID tcpip.NICID, addr tcpi } } -// Implements stack.NDPDispatcher.OnRecursiveDNSServerOption. +// Implements ipv6.NDPDispatcher.OnRecursiveDNSServerOption. func (n *ndpDispatcher) OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tcpip.Address, lifetime time.Duration) { if c := n.rdnssC; c != nil { c <- ndpRDNSSEvent{ @@ -274,7 +274,7 @@ func (n *ndpDispatcher) OnRecursiveDNSServerOption(nicID tcpip.NICID, addrs []tc } } -// Implements stack.NDPDispatcher.OnDNSSearchListOption. +// Implements ipv6.NDPDispatcher.OnDNSSearchListOption. func (n *ndpDispatcher) OnDNSSearchListOption(nicID tcpip.NICID, domainNames []string, lifetime time.Duration) { if n.dnsslC != nil { n.dnsslC <- ndpDNSSLEvent{ @@ -285,8 +285,8 @@ func (n *ndpDispatcher) OnDNSSearchListOption(nicID tcpip.NICID, domainNames []s } } -// Implements stack.NDPDispatcher.OnDHCPv6Configuration. -func (n *ndpDispatcher) OnDHCPv6Configuration(nicID tcpip.NICID, configuration stack.DHCPv6ConfigurationFromNDPRA) { +// Implements ipv6.NDPDispatcher.OnDHCPv6Configuration. +func (n *ndpDispatcher) OnDHCPv6Configuration(nicID tcpip.NICID, configuration ipv6.DHCPv6ConfigurationFromNDPRA) { if c := n.dhcpv6ConfigurationC; c != nil { c <- ndpDHCPv6Event{ nicID, @@ -319,13 +319,12 @@ func TestDADDisabled(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent, 1), } - opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPDisp: &ndpDisp, - } - e := channel.New(0, 1280, linkAddr1) - s := stack.New(opts) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + })}, + }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } @@ -413,19 +412,21 @@ func TestDADResolve(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent), } - opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPDisp: &ndpDisp, - } - opts.NDPConfigs.RetransmitTimer = test.retransTimer - opts.NDPConfigs.DupAddrDetectTransmits = test.dupAddrDetectTransmits e := channelLinkWithHeaderLength{ Endpoint: channel.New(int(test.dupAddrDetectTransmits), 1280, linkAddr1), headerLength: test.linkHeaderLen, } e.Endpoint.LinkEPCapabilities |= stack.CapabilityResolutionRequired - s := stack.New(opts) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + NDPConfigs: ipv6.NDPConfigurations{ + RetransmitTimer: test.retransTimer, + DupAddrDetectTransmits: test.dupAddrDetectTransmits, + }, + })}, + }) if err := s.CreateNIC(nicID, &e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } @@ -558,6 +559,26 @@ func TestDADResolve(t *testing.T) { } } +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.SetTargetAddress(tgt) + snmc := header.SolicitedNodeAddr(tgt) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, header.IPv6Any, snmc, buffer.VectorisedView{})) + payloadLength := hdr.UsedLength() + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(icmp.ProtocolNumber6), + HopLimit: 255, + SrcAddr: header.IPv6Any, + DstAddr: snmc, + }) + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{Data: hdr.View().ToVectorisedView()})) +} + // TestDADFail tests to make sure that the DAD process fails if another node is // detected to be performing DAD on the same address (receive an NS message from // a node doing DAD for the same address), or if another node is detected to own @@ -567,39 +588,19 @@ func TestDADFail(t *testing.T) { tests := []struct { name string - makeBuf func(tgt tcpip.Address) buffer.Prependable + rxPkt func(e *channel.Endpoint, tgt tcpip.Address) getStat func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter }{ { - "RxSolicit", - func(tgt tcpip.Address) buffer.Prependable { - hdr := buffer.NewPrependable(header.IPv6MinimumSize + header.ICMPv6NeighborSolicitMinimumSize) - pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) - pkt.SetType(header.ICMPv6NeighborSolicit) - ns := header.NDPNeighborSolicit(pkt.NDPPayload()) - ns.SetTargetAddress(tgt) - snmc := header.SolicitedNodeAddr(tgt) - pkt.SetChecksum(header.ICMPv6Checksum(pkt, header.IPv6Any, snmc, buffer.VectorisedView{})) - payloadLength := hdr.UsedLength() - ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) - ip.Encode(&header.IPv6Fields{ - PayloadLength: uint16(payloadLength), - NextHeader: uint8(icmp.ProtocolNumber6), - HopLimit: 255, - SrcAddr: header.IPv6Any, - DstAddr: snmc, - }) - - return hdr - - }, - func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + name: "RxSolicit", + rxPkt: rxNDPSolicit, + getStat: func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { return s.NeighborSolicit }, }, { - "RxAdvert", - func(tgt tcpip.Address) buffer.Prependable { + name: "RxAdvert", + rxPkt: func(e *channel.Endpoint, tgt tcpip.Address) { naSize := header.ICMPv6NeighborAdvertMinimumSize + header.NDPLinkLayerAddressSize hdr := buffer.NewPrependable(header.IPv6MinimumSize + naSize) pkt := header.ICMPv6(hdr.Prepend(naSize)) @@ -621,11 +622,9 @@ func TestDADFail(t *testing.T) { SrcAddr: tgt, DstAddr: header.IPv6AllNodesMulticastAddress, }) - - return hdr - + e.InjectInbound(header.IPv6ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{Data: hdr.View().ToVectorisedView()})) }, - func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { + getStat: func(s tcpip.ICMPv6ReceivedPacketStats) *tcpip.StatCounter { return s.NeighborAdvert }, }, @@ -636,16 +635,16 @@ func TestDADFail(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent, 1), } - ndpConfigs := stack.DefaultNDPConfigurations() - opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, - } - opts.NDPConfigs.RetransmitTimer = time.Second * 2 + ndpConfigs := ipv6.DefaultNDPConfigurations() + ndpConfigs.RetransmitTimer = time.Second * 2 e := channel.New(0, 1280, linkAddr1) - s := stack.New(opts) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + NDPConfigs: ndpConfigs, + })}, + }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } @@ -664,13 +663,8 @@ func TestDADFail(t *testing.T) { t.Fatalf("got stack.GetMainNICAddress(%d, %d) = (%s, nil), want = (%s, nil)", nicID, header.IPv6ProtocolNumber, addr, want) } - // Receive a packet to simulate multiple nodes owning or - // attempting to own the same address. - hdr := test.makeBuf(addr1) - pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - Data: hdr.View().ToVectorisedView(), - }) - e.InjectInbound(header.IPv6ProtocolNumber, pkt) + // Receive a packet to simulate an address conflict. + test.rxPkt(e, addr1) stat := test.getStat(s.Stats().ICMP.V6PacketsReceived) if got := stat.Value(); got != 1 { @@ -754,18 +748,19 @@ func TestDADStop(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent, 1), } - ndpConfigs := stack.NDPConfigurations{ + + ndpConfigs := ipv6.NDPConfigurations{ RetransmitTimer: time.Second, DupAddrDetectTransmits: 2, } - opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPDisp: &ndpDisp, - NDPConfigs: ndpConfigs, - } e := channel.New(0, 1280, linkAddr1) - s := stack.New(opts) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + NDPConfigs: ndpConfigs, + })}, + }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _): %s", nicID, err) } @@ -815,19 +810,6 @@ func TestDADStop(t *testing.T) { } } -// TestSetNDPConfigurationFailsForBadNICID tests to make sure we get an error if -// we attempt to update NDP configurations using an invalid NICID. -func TestSetNDPConfigurationFailsForBadNICID(t *testing.T) { - s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - }) - - // No NIC with ID 1 yet. - if got := s.SetNDPConfigurations(1, stack.NDPConfigurations{}); got != tcpip.ErrUnknownNICID { - t.Fatalf("got s.SetNDPConfigurations = %v, want = %s", got, tcpip.ErrUnknownNICID) - } -} - // TestSetNDPConfigurations tests that we can update and use per-interface NDP // configurations without affecting the default NDP configurations or other // interfaces' configurations. @@ -863,8 +845,9 @@ func TestSetNDPConfigurations(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDisp, + })}, }) expectDADEvent := func(nicID tcpip.NICID, addr tcpip.Address) { @@ -892,12 +875,15 @@ func TestSetNDPConfigurations(t *testing.T) { } // Update the NDP configurations on NIC(1) to use DAD. - configs := stack.NDPConfigurations{ + configs := ipv6.NDPConfigurations{ DupAddrDetectTransmits: test.dupAddrDetectTransmits, RetransmitTimer: test.retransmitTimer, } - if err := s.SetNDPConfigurations(nicID1, configs); err != nil { - t.Fatalf("got SetNDPConfigurations(%d, _) = %s", nicID1, err) + if ipv6Ep, err := s.GetNetworkEndpoint(nicID1, header.IPv6ProtocolNumber); err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID1, header.IPv6ProtocolNumber, err) + } else { + ndpEP := ipv6Ep.(ipv6.NDPEndpoint) + ndpEP.SetNDPConfigurations(configs) } // Created after updating NIC(1)'s NDP configurations @@ -1113,12 +1099,13 @@ func TestNoRouterDiscovery(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: handle, - DiscoverDefaultRouters: discover, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: handle, + DiscoverDefaultRouters: discover, + }, + NDPDisp: &ndpDisp, + })}, }) s.SetForwarding(ipv6.ProtocolNumber, forwarding) @@ -1151,12 +1138,13 @@ func TestRouterDiscoveryDispatcherNoRemember(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1192,12 +1180,13 @@ func TestRouterDiscovery(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + })}, }) expectRouterEvent := func(addr tcpip.Address, discovered bool) { @@ -1285,7 +1274,7 @@ func TestRouterDiscovery(t *testing.T) { } // TestRouterDiscoveryMaxRouters tests that only -// stack.MaxDiscoveredDefaultRouters discovered routers are remembered. +// ipv6.MaxDiscoveredDefaultRouters discovered routers are remembered. func TestRouterDiscoveryMaxRouters(t *testing.T) { ndpDisp := ndpDispatcher{ routerC: make(chan ndpRouterEvent, 1), @@ -1293,12 +1282,13 @@ func TestRouterDiscoveryMaxRouters(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1306,14 +1296,14 @@ func TestRouterDiscoveryMaxRouters(t *testing.T) { } // Receive an RA from 2 more than the max number of discovered routers. - for i := 1; i <= stack.MaxDiscoveredDefaultRouters+2; i++ { + for i := 1; i <= ipv6.MaxDiscoveredDefaultRouters+2; i++ { linkAddr := []byte{2, 2, 3, 4, 5, 0} linkAddr[5] = byte(i) llAddr := header.LinkLocalAddr(tcpip.LinkAddress(linkAddr)) e.InjectInbound(header.IPv6ProtocolNumber, raBuf(llAddr, 5)) - if i <= stack.MaxDiscoveredDefaultRouters { + if i <= ipv6.MaxDiscoveredDefaultRouters { select { case e := <-ndpDisp.routerC: if diff := checkRouterEvent(e, llAddr, true); diff != "" { @@ -1358,12 +1348,13 @@ func TestNoPrefixDiscovery(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: handle, - DiscoverOnLinkPrefixes: discover, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: handle, + DiscoverOnLinkPrefixes: discover, + }, + NDPDisp: &ndpDisp, + })}, }) s.SetForwarding(ipv6.ProtocolNumber, forwarding) @@ -1399,13 +1390,14 @@ func TestPrefixDiscoveryDispatcherNoRemember(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: false, - DiscoverOnLinkPrefixes: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: false, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1445,12 +1437,13 @@ func TestPrefixDiscovery(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverOnLinkPrefixes: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1545,12 +1538,13 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverOnLinkPrefixes: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1621,33 +1615,34 @@ func TestPrefixDiscoveryWithInfiniteLifetime(t *testing.T) { } // TestPrefixDiscoveryMaxRouters tests that only -// stack.MaxDiscoveredOnLinkPrefixes discovered on-link prefixes are remembered. +// ipv6.MaxDiscoveredOnLinkPrefixes discovered on-link prefixes are remembered. func TestPrefixDiscoveryMaxOnLinkPrefixes(t *testing.T) { ndpDisp := ndpDispatcher{ - prefixC: make(chan ndpPrefixEvent, stack.MaxDiscoveredOnLinkPrefixes+3), + prefixC: make(chan ndpPrefixEvent, ipv6.MaxDiscoveredOnLinkPrefixes+3), rememberPrefix: true, } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: false, - DiscoverOnLinkPrefixes: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: false, + DiscoverOnLinkPrefixes: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(1) = %s", err) } - optSer := make(header.NDPOptionsSerializer, stack.MaxDiscoveredOnLinkPrefixes+2) - prefixes := [stack.MaxDiscoveredOnLinkPrefixes + 2]tcpip.Subnet{} + optSer := make(header.NDPOptionsSerializer, ipv6.MaxDiscoveredOnLinkPrefixes+2) + prefixes := [ipv6.MaxDiscoveredOnLinkPrefixes + 2]tcpip.Subnet{} // Receive an RA with 2 more than the max number of discovered on-link // prefixes. - for i := 0; i < stack.MaxDiscoveredOnLinkPrefixes+2; i++ { + for i := 0; i < ipv6.MaxDiscoveredOnLinkPrefixes+2; i++ { prefixAddr := [16]byte{1, 2, 3, 4, 5, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 0} prefixAddr[7] = byte(i) prefix := tcpip.AddressWithPrefix{ @@ -1665,8 +1660,8 @@ func TestPrefixDiscoveryMaxOnLinkPrefixes(t *testing.T) { } e.InjectInbound(header.IPv6ProtocolNumber, raBufWithOpts(llAddr1, 0, optSer)) - for i := 0; i < stack.MaxDiscoveredOnLinkPrefixes+2; i++ { - if i < stack.MaxDiscoveredOnLinkPrefixes { + for i := 0; i < ipv6.MaxDiscoveredOnLinkPrefixes+2; i++ { + if i < ipv6.MaxDiscoveredOnLinkPrefixes { select { case e := <-ndpDisp.prefixC: if diff := checkPrefixEvent(e, prefixes[i], true); diff != "" { @@ -1716,12 +1711,13 @@ func TestNoAutoGenAddr(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: handle, - AutoGenGlobalAddresses: autogen, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: handle, + AutoGenGlobalAddresses: autogen, + }, + NDPDisp: &ndpDisp, + })}, }) s.SetForwarding(ipv6.ProtocolNumber, forwarding) @@ -1749,14 +1745,14 @@ func checkAutoGenAddrEvent(e ndpAutoGenAddrEvent, addr tcpip.AddressWithPrefix, // TestAutoGenAddr tests that an address is properly generated and invalidated // when configured to do so. -func TestAutoGenAddr(t *testing.T) { +func TestAutoGenAddr2(t *testing.T) { const newMinVL = 2 newMinVLDuration := newMinVL * time.Second - saved := stack.MinPrefixInformationValidLifetimeForUpdate + saved := ipv6.MinPrefixInformationValidLifetimeForUpdate defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = saved + ipv6.MinPrefixInformationValidLifetimeForUpdate = saved }() - stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration + ipv6.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) @@ -1766,12 +1762,13 @@ func TestAutoGenAddr(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -1876,14 +1873,14 @@ func TestAutoGenTempAddr(t *testing.T) { newMinVLDuration = newMinVL * time.Second ) - savedMinPrefixInformationValidLifetimeForUpdate := stack.MinPrefixInformationValidLifetimeForUpdate - savedMaxDesync := stack.MaxDesyncFactor + savedMinPrefixInformationValidLifetimeForUpdate := ipv6.MinPrefixInformationValidLifetimeForUpdate + savedMaxDesync := ipv6.MaxDesyncFactor defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = savedMinPrefixInformationValidLifetimeForUpdate - stack.MaxDesyncFactor = savedMaxDesync + ipv6.MinPrefixInformationValidLifetimeForUpdate = savedMinPrefixInformationValidLifetimeForUpdate + ipv6.MaxDesyncFactor = savedMaxDesync }() - stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration - stack.MaxDesyncFactor = time.Nanosecond + ipv6.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration + ipv6.MaxDesyncFactor = time.Nanosecond prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) @@ -1931,16 +1928,17 @@ func TestAutoGenTempAddr(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - DupAddrDetectTransmits: test.dupAddrTransmits, - RetransmitTimer: test.retransmitTimer, - HandleRAs: true, - AutoGenGlobalAddresses: true, - AutoGenTempGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, - TempIIDSeed: seed, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + DupAddrDetectTransmits: test.dupAddrTransmits, + RetransmitTimer: test.retransmitTimer, + HandleRAs: true, + AutoGenGlobalAddresses: true, + AutoGenTempGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + TempIIDSeed: seed, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2119,11 +2117,11 @@ func TestAutoGenTempAddr(t *testing.T) { func TestNoAutoGenTempAddrForLinkLocal(t *testing.T) { const nicID = 1 - savedMaxDesyncFactor := stack.MaxDesyncFactor + savedMaxDesyncFactor := ipv6.MaxDesyncFactor defer func() { - stack.MaxDesyncFactor = savedMaxDesyncFactor + ipv6.MaxDesyncFactor = savedMaxDesyncFactor }() - stack.MaxDesyncFactor = time.Nanosecond + ipv6.MaxDesyncFactor = time.Nanosecond tests := []struct { name string @@ -2160,12 +2158,13 @@ func TestNoAutoGenTempAddrForLinkLocal(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - AutoGenTempGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, - AutoGenIPv6LinkLocal: true, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + AutoGenTempGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + AutoGenIPv6LinkLocal: true, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2211,11 +2210,11 @@ func TestNoAutoGenTempAddrWithoutStableAddr(t *testing.T) { retransmitTimer = 2 * time.Second ) - savedMaxDesyncFactor := stack.MaxDesyncFactor + savedMaxDesyncFactor := ipv6.MaxDesyncFactor defer func() { - stack.MaxDesyncFactor = savedMaxDesyncFactor + ipv6.MaxDesyncFactor = savedMaxDesyncFactor }() - stack.MaxDesyncFactor = 0 + ipv6.MaxDesyncFactor = 0 prefix, _, addr := prefixSubnetAddr(0, linkAddr1) var tempIIDHistory [header.IIDSize]byte @@ -2228,15 +2227,16 @@ func TestNoAutoGenTempAddrWithoutStableAddr(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - DupAddrDetectTransmits: dadTransmits, - RetransmitTimer: retransmitTimer, - HandleRAs: true, - AutoGenGlobalAddresses: true, - AutoGenTempGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + DupAddrDetectTransmits: dadTransmits, + RetransmitTimer: retransmitTimer, + HandleRAs: true, + AutoGenGlobalAddresses: true, + AutoGenTempGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2294,17 +2294,17 @@ func TestAutoGenTempAddrRegen(t *testing.T) { newMinVLDuration = newMinVL * time.Second ) - savedMaxDesyncFactor := stack.MaxDesyncFactor - savedMinMaxTempAddrPreferredLifetime := stack.MinMaxTempAddrPreferredLifetime - savedMinMaxTempAddrValidLifetime := stack.MinMaxTempAddrValidLifetime + savedMaxDesyncFactor := ipv6.MaxDesyncFactor + savedMinMaxTempAddrPreferredLifetime := ipv6.MinMaxTempAddrPreferredLifetime + savedMinMaxTempAddrValidLifetime := ipv6.MinMaxTempAddrValidLifetime defer func() { - stack.MaxDesyncFactor = savedMaxDesyncFactor - stack.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime - stack.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime + ipv6.MaxDesyncFactor = savedMaxDesyncFactor + ipv6.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime + ipv6.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime }() - stack.MaxDesyncFactor = 0 - stack.MinMaxTempAddrPreferredLifetime = newMinVLDuration - stack.MinMaxTempAddrValidLifetime = newMinVLDuration + ipv6.MaxDesyncFactor = 0 + ipv6.MinMaxTempAddrPreferredLifetime = newMinVLDuration + ipv6.MinMaxTempAddrValidLifetime = newMinVLDuration prefix, _, addr := prefixSubnetAddr(0, linkAddr1) var tempIIDHistory [header.IIDSize]byte @@ -2317,16 +2317,17 @@ func TestAutoGenTempAddrRegen(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2), } e := channel.New(0, 1280, linkAddr1) - ndpConfigs := stack.NDPConfigurations{ + ndpConfigs := ipv6.NDPConfigurations{ HandleRAs: true, AutoGenGlobalAddresses: true, AutoGenTempGlobalAddresses: true, RegenAdvanceDuration: newMinVLDuration - regenAfter, } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2382,8 +2383,11 @@ func TestAutoGenTempAddrRegen(t *testing.T) { // Stop generating temporary addresses ndpConfigs.AutoGenTempGlobalAddresses = false - if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil { - t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err) + if ipv6Ep, err := s.GetNetworkEndpoint(nicID, header.IPv6ProtocolNumber); err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err) + } else { + ndpEP := ipv6Ep.(ipv6.NDPEndpoint) + ndpEP.SetNDPConfigurations(ndpConfigs) } // Wait for all the temporary addresses to get invalidated. @@ -2439,17 +2443,17 @@ func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { newMinVLDuration = newMinVL * time.Second ) - savedMaxDesyncFactor := stack.MaxDesyncFactor - savedMinMaxTempAddrPreferredLifetime := stack.MinMaxTempAddrPreferredLifetime - savedMinMaxTempAddrValidLifetime := stack.MinMaxTempAddrValidLifetime + savedMaxDesyncFactor := ipv6.MaxDesyncFactor + savedMinMaxTempAddrPreferredLifetime := ipv6.MinMaxTempAddrPreferredLifetime + savedMinMaxTempAddrValidLifetime := ipv6.MinMaxTempAddrValidLifetime defer func() { - stack.MaxDesyncFactor = savedMaxDesyncFactor - stack.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime - stack.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime + ipv6.MaxDesyncFactor = savedMaxDesyncFactor + ipv6.MinMaxTempAddrPreferredLifetime = savedMinMaxTempAddrPreferredLifetime + ipv6.MinMaxTempAddrValidLifetime = savedMinMaxTempAddrValidLifetime }() - stack.MaxDesyncFactor = 0 - stack.MinMaxTempAddrPreferredLifetime = newMinVLDuration - stack.MinMaxTempAddrValidLifetime = newMinVLDuration + ipv6.MaxDesyncFactor = 0 + ipv6.MinMaxTempAddrPreferredLifetime = newMinVLDuration + ipv6.MinMaxTempAddrValidLifetime = newMinVLDuration prefix, _, addr := prefixSubnetAddr(0, linkAddr1) var tempIIDHistory [header.IIDSize]byte @@ -2462,16 +2466,17 @@ func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2), } e := channel.New(0, 1280, linkAddr1) - ndpConfigs := stack.NDPConfigurations{ + ndpConfigs := ipv6.NDPConfigurations{ HandleRAs: true, AutoGenGlobalAddresses: true, AutoGenTempGlobalAddresses: true, RegenAdvanceDuration: newMinVLDuration - regenAfter, } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { @@ -2545,9 +2550,12 @@ func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { // as paased. ndpConfigs.MaxTempAddrValidLifetime = 100 * time.Second ndpConfigs.MaxTempAddrPreferredLifetime = 100 * time.Second - if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil { - t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err) + ipv6Ep, err := s.GetNetworkEndpoint(nicID, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err) } + ndpEP := ipv6Ep.(ipv6.NDPEndpoint) + ndpEP.SetNDPConfigurations(ndpConfigs) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100)) select { case e := <-ndpDisp.autoGenAddrC: @@ -2565,9 +2573,7 @@ func TestAutoGenTempAddrRegenJobUpdates(t *testing.T) { newLifetimes := newMinVLDuration + regenAfter + defaultAsyncNegativeEventTimeout ndpConfigs.MaxTempAddrValidLifetime = newLifetimes ndpConfigs.MaxTempAddrPreferredLifetime = newLifetimes - if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil { - t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err) - } + ndpEP.SetNDPConfigurations(ndpConfigs) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr2, 0, prefix, true, true, 100, 100)) expectAutoGenAddrEventAsync(tempAddr3, newAddr, regenAfter+defaultAsyncPositiveEventTimeout) } @@ -2655,20 +2661,21 @@ func TestMixedSLAACAddrConflictRegen(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, 2), } e := channel.New(0, 1280, linkAddr1) - ndpConfigs := stack.NDPConfigurations{ + ndpConfigs := ipv6.NDPConfigurations{ HandleRAs: true, AutoGenGlobalAddresses: true, AutoGenTempGlobalAddresses: test.tempAddrs, AutoGenAddressConflictRetries: 1, } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ - NICNameFromID: test.nicNameFromID, - }, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ + NICNameFromID: test.nicNameFromID, + }, + })}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) s.SetRouteTable([]tcpip.Route{{ @@ -2739,8 +2746,11 @@ func TestMixedSLAACAddrConflictRegen(t *testing.T) { ndpDisp.dadC = make(chan ndpDADEvent, 2) ndpConfigs.DupAddrDetectTransmits = dupAddrTransmits ndpConfigs.RetransmitTimer = retransmitTimer - if err := s.SetNDPConfigurations(nicID, ndpConfigs); err != nil { - t.Fatalf("s.SetNDPConfigurations(%d, _): %s", nicID, err) + if ipv6Ep, err := s.GetNetworkEndpoint(nicID, header.IPv6ProtocolNumber); err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err) + } else { + ndpEP := ipv6Ep.(ipv6.NDPEndpoint) + ndpEP.SetNDPConfigurations(ndpConfigs) } // Do SLAAC for prefix. @@ -2754,9 +2764,7 @@ func TestMixedSLAACAddrConflictRegen(t *testing.T) { // DAD failure to restart the local generation process. addr := test.addrs[maxSLAACAddrLocalRegenAttempts-1] expectAutoGenAddrAsyncEvent(addr, newAddr) - if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil { - t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err) - } + rxNDPSolicit(e, addr.Address) select { case e := <-ndpDisp.dadC: if diff := checkDADEvent(e, nicID, addr.Address, false, nil); diff != "" { @@ -2794,14 +2802,15 @@ func stackAndNdpDispatcherWithDefaultRoute(t *testing.T, nicID tcpip.NICID, useN } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: ndpDisp, - UseNeighborCache: useNeighborCache, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: ndpDisp, + })}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, + UseNeighborCache: useNeighborCache, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -3036,11 +3045,11 @@ func TestAutoGenAddrJobDeprecation(t *testing.T) { for _, stackTyp := range stacks { t.Run(stackTyp.name, func(t *testing.T) { - saved := stack.MinPrefixInformationValidLifetimeForUpdate + saved := ipv6.MinPrefixInformationValidLifetimeForUpdate defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = saved + ipv6.MinPrefixInformationValidLifetimeForUpdate = saved }() - stack.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration + ipv6.MinPrefixInformationValidLifetimeForUpdate = newMinVLDuration prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) @@ -3258,12 +3267,12 @@ func TestAutoGenAddrFiniteToInfiniteToFiniteVL(t *testing.T) { const infiniteVLSeconds = 2 const minVLSeconds = 1 savedIL := header.NDPInfiniteLifetime - savedMinVL := stack.MinPrefixInformationValidLifetimeForUpdate + savedMinVL := ipv6.MinPrefixInformationValidLifetimeForUpdate defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = savedMinVL + ipv6.MinPrefixInformationValidLifetimeForUpdate = savedMinVL header.NDPInfiniteLifetime = savedIL }() - stack.MinPrefixInformationValidLifetimeForUpdate = minVLSeconds * time.Second + ipv6.MinPrefixInformationValidLifetimeForUpdate = minVLSeconds * time.Second header.NDPInfiniteLifetime = infiniteVLSeconds * time.Second prefix, _, addr := prefixSubnetAddr(0, linkAddr1) @@ -3307,12 +3316,13 @@ func TestAutoGenAddrFiniteToInfiniteToFiniteVL(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -3357,11 +3367,11 @@ func TestAutoGenAddrFiniteToInfiniteToFiniteVL(t *testing.T) { func TestAutoGenAddrValidLifetimeUpdates(t *testing.T) { const infiniteVL = 4294967295 const newMinVL = 4 - saved := stack.MinPrefixInformationValidLifetimeForUpdate + saved := ipv6.MinPrefixInformationValidLifetimeForUpdate defer func() { - stack.MinPrefixInformationValidLifetimeForUpdate = saved + ipv6.MinPrefixInformationValidLifetimeForUpdate = saved }() - stack.MinPrefixInformationValidLifetimeForUpdate = newMinVL * time.Second + ipv6.MinPrefixInformationValidLifetimeForUpdate = newMinVL * time.Second prefix, _, addr := prefixSubnetAddr(0, linkAddr1) @@ -3449,12 +3459,13 @@ func TestAutoGenAddrValidLifetimeUpdates(t *testing.T) { } e := channel.New(10, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -3515,12 +3526,13 @@ func TestAutoGenAddrRemoval(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -3700,12 +3712,13 @@ func TestAutoGenAddrStaticConflict(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { @@ -3781,18 +3794,19 @@ func TestAutoGenAddrWithOpaqueIID(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ - NICNameFromID: func(_ tcpip.NICID, nicName string) string { - return nicName + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, }, - SecretKey: secretKey, - }, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ + NICNameFromID: func(_ tcpip.NICID, nicName string) string { + return nicName + }, + SecretKey: secretKey, + }, + })}, }) opts := stack.NICOptions{Name: nicName} if err := s.CreateNICWithOptions(nicID, e, opts); err != nil { @@ -3856,11 +3870,11 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { const lifetimeSeconds = 10 // Needed for the temporary address sub test. - savedMaxDesync := stack.MaxDesyncFactor + savedMaxDesync := ipv6.MaxDesyncFactor defer func() { - stack.MaxDesyncFactor = savedMaxDesync + ipv6.MaxDesyncFactor = savedMaxDesync }() - stack.MaxDesyncFactor = time.Nanosecond + ipv6.MaxDesyncFactor = time.Nanosecond var secretKeyBuf [header.OpaqueIIDSecretKeyMinBytes]byte secretKey := secretKeyBuf[:] @@ -3938,14 +3952,14 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { addrTypes := []struct { name string - ndpConfigs stack.NDPConfigurations + ndpConfigs ipv6.NDPConfigurations autoGenLinkLocal bool prepareFn func(t *testing.T, ndpDisp *ndpDispatcher, e *channel.Endpoint, tempIIDHistory []byte) []tcpip.AddressWithPrefix addrGenFn func(dadCounter uint8, tempIIDHistory []byte) tcpip.AddressWithPrefix }{ { name: "Global address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, HandleRAs: true, @@ -3963,7 +3977,7 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { }, { name: "LinkLocal address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, }, @@ -3977,7 +3991,7 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { }, { name: "Temporary address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, HandleRAs: true, @@ -4029,16 +4043,17 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { ndpConfigs := addrType.ndpConfigs ndpConfigs.AutoGenAddressConflictRetries = maxRetries s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, - NDPConfigs: ndpConfigs, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ - NICNameFromID: func(_ tcpip.NICID, nicName string) string { - return nicName + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, + NDPConfigs: ndpConfigs, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ + NICNameFromID: func(_ tcpip.NICID, nicName string) string { + return nicName + }, + SecretKey: secretKey, }, - SecretKey: secretKey, - }, + })}, }) opts := stack.NICOptions{Name: nicName} if err := s.CreateNICWithOptions(nicID, e, opts); err != nil { @@ -4059,9 +4074,7 @@ func TestAutoGenAddrInResponseToDADConflicts(t *testing.T) { } // Simulate a DAD conflict. - if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil { - t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err) - } + rxNDPSolicit(e, addr.Address) expectAutoGenAddrEvent(t, &ndpDisp, addr, invalidatedAddr) expectDADEvent(t, &ndpDisp, addr.Address, false) @@ -4119,14 +4132,14 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { addrTypes := []struct { name string - ndpConfigs stack.NDPConfigurations + ndpConfigs ipv6.NDPConfigurations autoGenLinkLocal bool subnet tcpip.Subnet triggerSLAACFn func(e *channel.Endpoint) }{ { name: "Global address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, HandleRAs: true, @@ -4142,7 +4155,7 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { }, { name: "LinkLocal address", - ndpConfigs: stack.NDPConfigurations{ + ndpConfigs: ipv6.NDPConfigurations{ DupAddrDetectTransmits: dadTransmits, RetransmitTimer: retransmitTimer, AutoGenAddressConflictRetries: maxRetries, @@ -4165,10 +4178,11 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, - NDPConfigs: addrType.ndpConfigs, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: addrType.autoGenLinkLocal, + NDPConfigs: addrType.ndpConfigs, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -4198,9 +4212,7 @@ func TestAutoGenAddrWithEUI64IIDNoDADRetries(t *testing.T) { expectAutoGenAddrEvent(addr, newAddr) // Simulate a DAD conflict. - if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil { - t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err) - } + rxNDPSolicit(e, addr.Address) expectAutoGenAddrEvent(addr, invalidatedAddr) select { case e := <-ndpDisp.dadC: @@ -4250,21 +4262,22 @@ func TestAutoGenAddrContinuesLifetimesAfterRetry(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - DupAddrDetectTransmits: dadTransmits, - RetransmitTimer: retransmitTimer, - HandleRAs: true, - AutoGenGlobalAddresses: true, - AutoGenAddressConflictRetries: maxRetries, - }, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ - NICNameFromID: func(_ tcpip.NICID, nicName string) string { - return nicName + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + DupAddrDetectTransmits: dadTransmits, + RetransmitTimer: retransmitTimer, + HandleRAs: true, + AutoGenGlobalAddresses: true, + AutoGenAddressConflictRetries: maxRetries, }, - SecretKey: secretKey, - }, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ + NICNameFromID: func(_ tcpip.NICID, nicName string) string { + return nicName + }, + SecretKey: secretKey, + }, + })}, }) opts := stack.NICOptions{Name: nicName} if err := s.CreateNICWithOptions(nicID, e, opts); err != nil { @@ -4296,9 +4309,7 @@ func TestAutoGenAddrContinuesLifetimesAfterRetry(t *testing.T) { // Simulate a DAD conflict after some time has passed. time.Sleep(failureTimer) - if err := s.DupTentativeAddrDetected(nicID, addr.Address); err != nil { - t.Fatalf("s.DupTentativeAddrDetected(%d, %s): %s", nicID, addr.Address, err) - } + rxNDPSolicit(e, addr.Address) expectAutoGenAddrEvent(addr, invalidatedAddr) select { case e := <-ndpDisp.dadC: @@ -4459,11 +4470,12 @@ func TestNDPRecursiveDNSServerDispatch(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(1) = %s", err) @@ -4509,11 +4521,12 @@ func TestNDPDNSSearchListDispatch(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -4694,15 +4707,16 @@ func TestCleanupNDPState(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, test.maxAutoGenAddrEvents), } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: true, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - DiscoverDefaultRouters: true, - DiscoverOnLinkPrefixes: true, - AutoGenGlobalAddresses: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: true, + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + DiscoverDefaultRouters: true, + DiscoverOnLinkPrefixes: true, + AutoGenGlobalAddresses: true, + }, + NDPDisp: &ndpDisp, + })}, }) expectRouterEvent := func() (bool, ndpRouterEvent) { @@ -4967,18 +4981,19 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { } e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + }, + NDPDisp: &ndpDisp, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) } - expectDHCPv6Event := func(configuration stack.DHCPv6ConfigurationFromNDPRA) { + expectDHCPv6Event := func(configuration ipv6.DHCPv6ConfigurationFromNDPRA) { t.Helper() select { case e := <-ndpDisp.dhcpv6ConfigurationC: @@ -5002,7 +5017,7 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // Even if the first RA reports no DHCPv6 configurations are available, the // dispatcher should get an event. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false)) - expectDHCPv6Event(stack.DHCPv6NoConfiguration) + expectDHCPv6Event(ipv6.DHCPv6NoConfiguration) // Receiving the same update again should not result in an event to the // dispatcher. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false)) @@ -5011,19 +5026,19 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // Receive an RA that updates the DHCPv6 configuration to Other // Configurations. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) - expectDHCPv6Event(stack.DHCPv6OtherConfigurations) + expectDHCPv6Event(ipv6.DHCPv6OtherConfigurations) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) expectNoDHCPv6Event() // Receive an RA that updates the DHCPv6 configuration to Managed Address. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false)) - expectDHCPv6Event(stack.DHCPv6ManagedAddress) + expectDHCPv6Event(ipv6.DHCPv6ManagedAddress) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, false)) expectNoDHCPv6Event() // Receive an RA that updates the DHCPv6 configuration to none. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false)) - expectDHCPv6Event(stack.DHCPv6NoConfiguration) + expectDHCPv6Event(ipv6.DHCPv6NoConfiguration) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, false)) expectNoDHCPv6Event() @@ -5031,7 +5046,7 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // // Note, when the M flag is set, the O flag is redundant. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true)) - expectDHCPv6Event(stack.DHCPv6ManagedAddress) + expectDHCPv6Event(ipv6.DHCPv6ManagedAddress) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, true, true)) expectNoDHCPv6Event() // Even though the DHCPv6 flags are different, the effective configuration is @@ -5044,7 +5059,7 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // Receive an RA that updates the DHCPv6 configuration to Other // Configurations. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) - expectDHCPv6Event(stack.DHCPv6OtherConfigurations) + expectDHCPv6Event(ipv6.DHCPv6OtherConfigurations) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) expectNoDHCPv6Event() @@ -5059,7 +5074,7 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { // Receive an RA that updates the DHCPv6 configuration to Other // Configurations. e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) - expectDHCPv6Event(stack.DHCPv6OtherConfigurations) + expectDHCPv6Event(ipv6.DHCPv6OtherConfigurations) e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) expectNoDHCPv6Event() } @@ -5217,12 +5232,13 @@ func TestRouterSolicitation(t *testing.T) { } } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - MaxRtrSolicitations: test.maxRtrSolicit, - RtrSolicitationInterval: test.rtrSolicitInt, - MaxRtrSolicitationDelay: test.maxRtrSolicitDelay, - }, + 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) @@ -5357,12 +5373,13 @@ func TestStopStartSolicitingRouters(t *testing.T) { checker.NDPRS()) } s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - MaxRtrSolicitations: maxRtrSolicitations, - RtrSolicitationInterval: interval, - MaxRtrSolicitationDelay: delay, - }, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + MaxRtrSolicitations: maxRtrSolicitations, + RtrSolicitationInterval: interval, + MaxRtrSolicitationDelay: delay, + }, + })}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) diff --git a/pkg/tcpip/stack/neighbor_cache.go b/pkg/tcpip/stack/neighbor_cache.go index 27e1feec0..4df288798 100644 --- a/pkg/tcpip/stack/neighbor_cache.go +++ b/pkg/tcpip/stack/neighbor_cache.go @@ -131,10 +131,17 @@ func (n *neighborCache) entry(remoteAddr, localAddr tcpip.Address, linkRes LinkA defer entry.mu.Unlock() switch s := entry.neigh.State; s { - case Reachable, Static: + case Stale: + entry.handlePacketQueuedLocked() + fallthrough + case Reachable, Static, Delay, Probe: + // As per RFC 4861 section 7.3.3: + // "Neighbor Unreachability Detection operates in parallel with the sending + // of packets to a neighbor. While reasserting a neighbor's reachability, + // a node continues sending packets to that neighbor using the cached + // link-layer address." return entry.neigh, nil, nil - - case Unknown, Incomplete, Stale, Delay, Probe: + case Unknown, Incomplete: entry.addWakerLocked(w) if entry.done == nil { @@ -147,10 +154,8 @@ func (n *neighborCache) entry(remoteAddr, localAddr tcpip.Address, linkRes LinkA entry.handlePacketQueuedLocked() return entry.neigh, entry.done, tcpip.ErrWouldBlock - case Failed: return entry.neigh, nil, tcpip.ErrNoLinkAddress - default: panic(fmt.Sprintf("Invalid cache entry state: %s", s)) } diff --git a/pkg/tcpip/stack/neighbor_cache_test.go b/pkg/tcpip/stack/neighbor_cache_test.go index b4fa69e3e..fcd54ed83 100644 --- a/pkg/tcpip/stack/neighbor_cache_test.go +++ b/pkg/tcpip/stack/neighbor_cache_test.go @@ -30,6 +30,7 @@ import ( "github.com/google/go-cmp/cmp/cmpopts" "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/faketime" ) const ( @@ -239,7 +240,7 @@ type entryEvent struct { func TestNeighborCacheGetConfig(t *testing.T) { nudDisp := testNUDDispatcher{} c := DefaultNUDConfigurations() - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, c, clock) if got, want := neigh.config(), c; got != want { @@ -257,7 +258,7 @@ func TestNeighborCacheGetConfig(t *testing.T) { func TestNeighborCacheSetConfig(t *testing.T) { nudDisp := testNUDDispatcher{} c := DefaultNUDConfigurations() - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, c, clock) c.MinRandomFactor = 1 @@ -279,7 +280,7 @@ func TestNeighborCacheSetConfig(t *testing.T) { func TestNeighborCacheEntry(t *testing.T) { c := DefaultNUDConfigurations() nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, c, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -298,7 +299,7 @@ func TestNeighborCacheEntry(t *testing.T) { t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - clock.advance(typicalLatency) + clock.Advance(typicalLatency) wantEvents := []testEntryEventInfo{ { @@ -339,7 +340,7 @@ func TestNeighborCacheRemoveEntry(t *testing.T) { config := DefaultNUDConfigurations() nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -358,7 +359,7 @@ func TestNeighborCacheRemoveEntry(t *testing.T) { t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - clock.advance(typicalLatency) + clock.Advance(typicalLatency) wantEvents := []testEntryEventInfo{ { @@ -409,7 +410,7 @@ func TestNeighborCacheRemoveEntry(t *testing.T) { } type testContext struct { - clock *fakeClock + clock *faketime.ManualClock neigh *neighborCache store *testEntryStore linkRes *testNeighborResolver @@ -418,7 +419,7 @@ type testContext struct { func newTestContext(c NUDConfigurations) testContext { nudDisp := &testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(nudDisp, c, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -454,7 +455,7 @@ func (c *testContext) overflowCache(opts overflowOptions) error { if _, _, err := c.neigh.entry(entry.Addr, entry.LocalAddr, c.linkRes, nil); err != tcpip.ErrWouldBlock { return fmt.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - c.clock.advance(c.neigh.config().RetransmitTimer) + c.clock.Advance(c.neigh.config().RetransmitTimer) var wantEvents []testEntryEventInfo @@ -567,7 +568,7 @@ func TestNeighborCacheRemoveEntryThenOverflow(t *testing.T) { if err != tcpip.ErrWouldBlock { t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - c.clock.advance(c.neigh.config().RetransmitTimer) + c.clock.Advance(c.neigh.config().RetransmitTimer) wantEvents := []testEntryEventInfo{ { EventType: entryTestAdded, @@ -803,7 +804,7 @@ func TestNeighborCacheOverwriteWithStaticEntryThenOverflow(t *testing.T) { if err != tcpip.ErrWouldBlock { t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - c.clock.advance(typicalLatency) + c.clock.Advance(typicalLatency) wantEvents := []testEntryEventInfo{ { EventType: entryTestAdded, @@ -876,7 +877,7 @@ func TestNeighborCacheNotifiesWaker(t *testing.T) { config := DefaultNUDConfigurations() nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -902,7 +903,7 @@ func TestNeighborCacheNotifiesWaker(t *testing.T) { if doneCh == nil { t.Fatalf("expected done channel from neigh.entry(%s, %s, _, _)", entry.Addr, entry.LocalAddr) } - clock.advance(typicalLatency) + clock.Advance(typicalLatency) select { case <-doneCh: @@ -944,7 +945,7 @@ func TestNeighborCacheRemoveWaker(t *testing.T) { config := DefaultNUDConfigurations() nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -974,7 +975,7 @@ func TestNeighborCacheRemoveWaker(t *testing.T) { // Remove the waker before the neighbor cache has the opportunity to send a // notification. neigh.removeWaker(entry.Addr, &w) - clock.advance(typicalLatency) + clock.Advance(typicalLatency) select { case <-doneCh: @@ -1073,7 +1074,7 @@ func TestNeighborCacheClear(t *testing.T) { config := DefaultNUDConfigurations() nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -1092,7 +1093,7 @@ func TestNeighborCacheClear(t *testing.T) { if err != tcpip.ErrWouldBlock { t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - clock.advance(typicalLatency) + clock.Advance(typicalLatency) wantEvents := []testEntryEventInfo{ { @@ -1188,7 +1189,7 @@ func TestNeighborCacheClearThenOverflow(t *testing.T) { if err != tcpip.ErrWouldBlock { t.Errorf("got c.neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - c.clock.advance(typicalLatency) + c.clock.Advance(typicalLatency) wantEvents := []testEntryEventInfo{ { EventType: entryTestAdded, @@ -1249,7 +1250,7 @@ func TestNeighborCacheKeepFrequentlyUsed(t *testing.T) { config.MaxRandomFactor = 1 nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -1277,7 +1278,7 @@ func TestNeighborCacheKeepFrequentlyUsed(t *testing.T) { if err != tcpip.ErrWouldBlock { t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - clock.advance(typicalLatency) + clock.Advance(typicalLatency) select { case <-doneCh: default: @@ -1325,7 +1326,7 @@ func TestNeighborCacheKeepFrequentlyUsed(t *testing.T) { if err != tcpip.ErrWouldBlock { t.Errorf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - clock.advance(typicalLatency) + clock.Advance(typicalLatency) select { case <-doneCh: default: @@ -1412,7 +1413,7 @@ func TestNeighborCacheConcurrent(t *testing.T) { config := DefaultNUDConfigurations() nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -1440,7 +1441,7 @@ func TestNeighborCacheConcurrent(t *testing.T) { wg.Wait() // Process all the requests for a single entry concurrently - clock.advance(typicalLatency) + clock.Advance(typicalLatency) } // All goroutines add in the same order and add more values than can fit in @@ -1472,7 +1473,7 @@ func TestNeighborCacheReplace(t *testing.T) { config := DefaultNUDConfigurations() nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -1491,7 +1492,7 @@ func TestNeighborCacheReplace(t *testing.T) { if err != tcpip.ErrWouldBlock { t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - clock.advance(typicalLatency) + clock.Advance(typicalLatency) select { case <-doneCh: default: @@ -1499,24 +1500,26 @@ func TestNeighborCacheReplace(t *testing.T) { } // Verify the entry exists - e, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) - if err != nil { - t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) - } - if doneCh != nil { - t.Errorf("unexpected done channel from neigh.entry(%s, %s, _, nil): %v", entry.Addr, entry.LocalAddr, doneCh) - } - if t.Failed() { - t.FailNow() - } - want := NeighborEntry{ - Addr: entry.Addr, - LocalAddr: entry.LocalAddr, - LinkAddr: entry.LinkAddr, - State: Reachable, - } - if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" { - t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LinkAddr, diff) + { + e, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != nil { + t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) + } + if doneCh != nil { + t.Errorf("unexpected done channel from neigh.entry(%s, %s, _, nil): %v", entry.Addr, entry.LocalAddr, doneCh) + } + if t.Failed() { + t.FailNow() + } + want := NeighborEntry{ + Addr: entry.Addr, + LocalAddr: entry.LocalAddr, + LinkAddr: entry.LinkAddr, + State: Reachable, + } + if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" { + t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LinkAddr, diff) + } } // Notify of a link address change @@ -1535,28 +1538,34 @@ func TestNeighborCacheReplace(t *testing.T) { IsRouter: false, }) - // Requesting the entry again should start address resolution + // Requesting the entry again should start neighbor reachability confirmation. + // + // Verify the entry's new link address and the new state. { - _, doneCh, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) - if err != tcpip.ErrWouldBlock { - t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) + e, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) + if err != nil { + t.Fatalf("neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) } - clock.advance(config.DelayFirstProbeTime + typicalLatency) - select { - case <-doneCh: - default: - t.Fatalf("expected notification from done channel returned by neigh.entry(%s, %s, _, nil)", entry.Addr, entry.LocalAddr) + want := NeighborEntry{ + Addr: entry.Addr, + LocalAddr: entry.LocalAddr, + LinkAddr: updatedLinkAddr, + State: Delay, + } + if diff := cmp.Diff(e, want, entryDiffOpts()...); diff != "" { + t.Errorf("neigh.entry(%s, %s, _, nil) mismatch (-got, +want):\n%s", entry.Addr, entry.LocalAddr, diff) } + clock.Advance(config.DelayFirstProbeTime + typicalLatency) } - // Verify the entry's new link address + // Verify that the neighbor is now reachable. { e, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) - clock.advance(typicalLatency) + clock.Advance(typicalLatency) if err != nil { t.Errorf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) } - want = NeighborEntry{ + want := NeighborEntry{ Addr: entry.Addr, LocalAddr: entry.LocalAddr, LinkAddr: updatedLinkAddr, @@ -1572,7 +1581,7 @@ func TestNeighborCacheResolutionFailed(t *testing.T) { config := DefaultNUDConfigurations() nudDisp := testNUDDispatcher{} - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(&nudDisp, config, clock) store := newTestEntryStore() @@ -1595,7 +1604,7 @@ func TestNeighborCacheResolutionFailed(t *testing.T) { if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrWouldBlock { t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } - clock.advance(typicalLatency) + clock.Advance(typicalLatency) got, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil) if err != nil { t.Fatalf("unexpected error from neigh.entry(%s, %s, _, nil): %s", entry.Addr, entry.LocalAddr, err) @@ -1618,7 +1627,7 @@ func TestNeighborCacheResolutionFailed(t *testing.T) { t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } waitFor := config.DelayFirstProbeTime + typicalLatency*time.Duration(config.MaxMulticastProbes) - clock.advance(waitFor) + clock.Advance(waitFor) if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrNoLinkAddress { t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrNoLinkAddress) } @@ -1636,7 +1645,7 @@ func TestNeighborCacheResolutionTimeout(t *testing.T) { config := DefaultNUDConfigurations() config.RetransmitTimer = time.Millisecond // small enough to cause timeout - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(nil, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ @@ -1654,7 +1663,7 @@ func TestNeighborCacheResolutionTimeout(t *testing.T) { t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrWouldBlock) } waitFor := config.RetransmitTimer * time.Duration(config.MaxMulticastProbes) - clock.advance(waitFor) + clock.Advance(waitFor) if _, _, err := neigh.entry(entry.Addr, entry.LocalAddr, linkRes, nil); err != tcpip.ErrNoLinkAddress { t.Fatalf("got neigh.entry(%s, %s, _, nil) = %v, want = %s", entry.Addr, entry.LocalAddr, err, tcpip.ErrNoLinkAddress) } @@ -1664,7 +1673,7 @@ func TestNeighborCacheResolutionTimeout(t *testing.T) { // resolved immediately and don't send resolution requests. func TestNeighborCacheStaticResolution(t *testing.T) { config := DefaultNUDConfigurations() - clock := newFakeClock() + clock := faketime.NewManualClock() neigh := newTestNeighborCache(nil, config, clock) store := newTestEntryStore() linkRes := &testNeighborResolver{ diff --git a/pkg/tcpip/stack/neighbor_entry.go b/pkg/tcpip/stack/neighbor_entry.go index 0068cacb8..4d69a4de1 100644 --- a/pkg/tcpip/stack/neighbor_entry.go +++ b/pkg/tcpip/stack/neighbor_entry.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/header" ) // NeighborEntry describes a neighboring device in the local network. @@ -73,8 +74,7 @@ const ( type neighborEntry struct { neighborEntryEntry - nic *NIC - protocol tcpip.NetworkProtocolNumber + nic *NIC // linkRes provides the functionality to send reachability probes, used in // Neighbor Unreachability Detection. @@ -236,7 +236,7 @@ func (e *neighborEntry) setStateLocked(next NeighborState) { return } - if err := e.linkRes.LinkAddressRequest(e.neigh.Addr, e.neigh.LocalAddr, "", e.nic.linkEP); err != nil { + if err := e.linkRes.LinkAddressRequest(e.neigh.Addr, e.neigh.LocalAddr, "", e.nic.LinkEndpoint); err != nil { // There is no need to log the error here; the NUD implementation may // assume a working link. A valid link should be the responsibility of // the NIC/stack.LinkEndpoint. @@ -277,7 +277,7 @@ func (e *neighborEntry) setStateLocked(next NeighborState) { return } - if err := e.linkRes.LinkAddressRequest(e.neigh.Addr, e.neigh.LocalAddr, e.neigh.LinkAddr, e.nic.linkEP); err != nil { + if err := e.linkRes.LinkAddressRequest(e.neigh.Addr, e.neigh.LocalAddr, e.neigh.LinkAddr, e.nic.LinkEndpoint); err != nil { e.dispatchRemoveEventLocked() e.setStateLocked(Failed) return @@ -440,7 +440,7 @@ func (e *neighborEntry) handleConfirmationLocked(linkAddr tcpip.LinkAddress, fla e.notifyWakersLocked() } - if e.isRouter && !flags.IsRouter { + if e.isRouter && !flags.IsRouter && header.IsV6UnicastAddress(e.neigh.Addr) { // "In those cases where the IsRouter flag changes from TRUE to FALSE as // a result of this update, the node MUST remove that router from the // Default Router List and update the Destination Cache entries for all @@ -448,9 +448,17 @@ func (e *neighborEntry) handleConfirmationLocked(linkAddr tcpip.LinkAddress, fla // 7.3.3. This is needed to detect when a node that is used as a router // stops forwarding packets due to being configured as a host." // - RFC 4861 section 7.2.5 - e.nic.mu.Lock() - e.nic.mu.ndp.invalidateDefaultRouter(e.neigh.Addr) - e.nic.mu.Unlock() + // + // TODO(gvisor.dev/issue/4085): Remove the special casing we do for IPv6 + // here. + ep, ok := e.nic.networkEndpoints[header.IPv6ProtocolNumber] + if !ok { + panic(fmt.Sprintf("have a neighbor entry for an IPv6 router but no IPv6 network endpoint")) + } + + if ndpEP, ok := ep.(NDPEndpoint); ok { + ndpEP.InvalidateDefaultRouter(e.neigh.Addr) + } } e.isRouter = flags.IsRouter diff --git a/pkg/tcpip/stack/neighbor_entry_test.go b/pkg/tcpip/stack/neighbor_entry_test.go index b769fb2fa..e79abebca 100644 --- a/pkg/tcpip/stack/neighbor_entry_test.go +++ b/pkg/tcpip/stack/neighbor_entry_test.go @@ -27,6 +27,8 @@ import ( "github.com/google/go-cmp/cmp/cmpopts" "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/faketime" + "gvisor.dev/gvisor/pkg/tcpip/header" ) const ( @@ -221,29 +223,27 @@ func (r *entryTestLinkResolver) LinkAddressProtocol() tcpip.NetworkProtocolNumbe return entryTestNetNumber } -func entryTestSetup(c NUDConfigurations) (*neighborEntry, *testNUDDispatcher, *entryTestLinkResolver, *fakeClock) { - clock := newFakeClock() +func entryTestSetup(c NUDConfigurations) (*neighborEntry, *testNUDDispatcher, *entryTestLinkResolver, *faketime.ManualClock) { + clock := faketime.NewManualClock() disp := testNUDDispatcher{} nic := NIC{ - id: entryTestNICID, - linkEP: nil, // entryTestLinkResolver doesn't use a LinkEndpoint + LinkEndpoint: nil, // entryTestLinkResolver doesn't use a LinkEndpoint + + id: entryTestNICID, stack: &Stack{ clock: clock, nudDisp: &disp, }, } + nic.networkEndpoints = map[tcpip.NetworkProtocolNumber]NetworkEndpoint{ + header.IPv6ProtocolNumber: (&testIPv6Protocol{}).NewEndpoint(&nic, nil, nil, nil), + } rng := rand.New(rand.NewSource(time.Now().UnixNano())) nudState := NewNUDState(c, rng) linkRes := entryTestLinkResolver{} entry := newNeighborEntry(&nic, entryTestAddr1 /* remoteAddr */, entryTestAddr2 /* localAddr */, nudState, &linkRes) - // Stub out ndpState to verify modification of default routers. - nic.mu.ndp = ndpState{ - nic: &nic, - defaultRouters: make(map[tcpip.Address]defaultRouterState), - } - // Stub out the neighbor cache to verify deletion from the cache. nic.neigh = &neighborCache{ nic: &nic, @@ -267,7 +267,7 @@ func TestEntryInitiallyUnknown(t *testing.T) { } e.mu.Unlock() - clock.advance(c.RetransmitTimer) + clock.Advance(c.RetransmitTimer) // No probes should have been sent. linkRes.mu.Lock() @@ -300,7 +300,7 @@ func TestEntryUnknownToUnknownWhenConfirmationWithUnknownAddress(t *testing.T) { } e.mu.Unlock() - clock.advance(time.Hour) + clock.Advance(time.Hour) // No probes should have been sent. linkRes.mu.Lock() @@ -410,7 +410,7 @@ func TestEntryIncompleteToIncompleteDoesNotChangeUpdatedAt(t *testing.T) { updatedAt := e.neigh.UpdatedAt e.mu.Unlock() - clock.advance(c.RetransmitTimer) + clock.Advance(c.RetransmitTimer) // UpdatedAt should remain the same during address resolution. wantProbes := []entryTestProbeInfo{ @@ -439,7 +439,7 @@ func TestEntryIncompleteToIncompleteDoesNotChangeUpdatedAt(t *testing.T) { } e.mu.Unlock() - clock.advance(c.RetransmitTimer) + clock.Advance(c.RetransmitTimer) // UpdatedAt should change after failing address resolution. Timing out after // sending the last probe transitions the entry to Failed. @@ -459,7 +459,7 @@ func TestEntryIncompleteToIncompleteDoesNotChangeUpdatedAt(t *testing.T) { } } - clock.advance(c.RetransmitTimer) + clock.Advance(c.RetransmitTimer) wantEvents := []testEntryEventInfo{ { @@ -748,7 +748,7 @@ func TestEntryIncompleteToFailed(t *testing.T) { e.mu.Unlock() waitFor := c.RetransmitTimer * time.Duration(c.MaxMulticastProbes) - clock.advance(waitFor) + clock.Advance(waitFor) wantProbes := []entryTestProbeInfo{ // The Incomplete-to-Incomplete state transition is tested here by @@ -816,6 +816,8 @@ func TestEntryStaysReachableWhenConfirmationWithRouterFlag(t *testing.T) { c := DefaultNUDConfigurations() e, nudDisp, linkRes, _ := entryTestSetup(c) + ipv6EP := e.nic.networkEndpoints[header.IPv6ProtocolNumber].(*testIPv6Endpoint) + e.mu.Lock() e.handlePacketQueuedLocked() e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ @@ -829,9 +831,7 @@ func TestEntryStaysReachableWhenConfirmationWithRouterFlag(t *testing.T) { if got, want := e.isRouter, true; got != want { t.Errorf("got e.isRouter = %t, want = %t", got, want) } - e.nic.mu.ndp.defaultRouters[entryTestAddr1] = defaultRouterState{ - invalidationJob: e.nic.stack.newJob(&testLocker{}, func() {}), - } + e.handleConfirmationLocked(entryTestLinkAddr1, ReachabilityConfirmationFlags{ Solicited: false, Override: false, @@ -840,8 +840,8 @@ func TestEntryStaysReachableWhenConfirmationWithRouterFlag(t *testing.T) { if got, want := e.isRouter, false; got != want { t.Errorf("got e.isRouter = %t, want = %t", got, want) } - if _, ok := e.nic.mu.ndp.defaultRouters[entryTestAddr1]; ok { - t.Errorf("unexpected defaultRouter for %s", entryTestAddr1) + if ipv6EP.invalidatedRtr != e.neigh.Addr { + t.Errorf("got ipv6EP.invalidatedRtr = %s, want = %s", ipv6EP.invalidatedRtr, e.neigh.Addr) } e.mu.Unlock() @@ -983,7 +983,7 @@ func TestEntryReachableToStaleWhenTimeout(t *testing.T) { t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) } - clock.advance(c.BaseReachableTime) + clock.Advance(c.BaseReachableTime) wantEvents := []testEntryEventInfo{ { @@ -1612,7 +1612,7 @@ func TestEntryDelayToReachableWhenUpperLevelConfirmation(t *testing.T) { t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) } - clock.advance(c.BaseReachableTime) + clock.Advance(c.BaseReachableTime) wantEvents := []testEntryEventInfo{ { @@ -1706,7 +1706,7 @@ func TestEntryDelayToReachableWhenSolicitedOverrideConfirmation(t *testing.T) { t.Fatalf("link address resolver probes mismatch (-got, +want):\n%s", diff) } - clock.advance(c.BaseReachableTime) + clock.Advance(c.BaseReachableTime) wantEvents := []testEntryEventInfo{ { @@ -1989,7 +1989,7 @@ func TestEntryDelayToProbe(t *testing.T) { } e.mu.Unlock() - clock.advance(c.DelayFirstProbeTime) + clock.Advance(c.DelayFirstProbeTime) wantProbes := []entryTestProbeInfo{ // The first probe is caused by the Unknown-to-Incomplete transition. @@ -2069,7 +2069,7 @@ func TestEntryProbeToStaleWhenProbeWithDifferentAddress(t *testing.T) { e.handlePacketQueuedLocked() e.mu.Unlock() - clock.advance(c.DelayFirstProbeTime) + clock.Advance(c.DelayFirstProbeTime) wantProbes := []entryTestProbeInfo{ // The first probe is caused by the Unknown-to-Incomplete transition. @@ -2166,7 +2166,7 @@ func TestEntryProbeToStaleWhenConfirmationWithDifferentAddress(t *testing.T) { e.handlePacketQueuedLocked() e.mu.Unlock() - clock.advance(c.DelayFirstProbeTime) + clock.Advance(c.DelayFirstProbeTime) wantProbes := []entryTestProbeInfo{ // The first probe is caused by the Unknown-to-Incomplete transition. @@ -2267,7 +2267,7 @@ func TestEntryStaysProbeWhenOverrideConfirmationWithSameAddress(t *testing.T) { e.handlePacketQueuedLocked() e.mu.Unlock() - clock.advance(c.DelayFirstProbeTime) + clock.Advance(c.DelayFirstProbeTime) wantProbes := []entryTestProbeInfo{ // The first probe is caused by the Unknown-to-Incomplete transition. @@ -2364,7 +2364,7 @@ func TestEntryUnknownToStaleToProbeToReachable(t *testing.T) { e.handlePacketQueuedLocked() e.mu.Unlock() - clock.advance(c.DelayFirstProbeTime) + clock.Advance(c.DelayFirstProbeTime) wantProbes := []entryTestProbeInfo{ // Probe caused by the Delay-to-Probe transition @@ -2398,7 +2398,7 @@ func TestEntryUnknownToStaleToProbeToReachable(t *testing.T) { } e.mu.Unlock() - clock.advance(c.BaseReachableTime) + clock.Advance(c.BaseReachableTime) wantEvents := []testEntryEventInfo{ { @@ -2463,7 +2463,7 @@ func TestEntryProbeToReachableWhenSolicitedOverrideConfirmation(t *testing.T) { e.handlePacketQueuedLocked() e.mu.Unlock() - clock.advance(c.DelayFirstProbeTime) + clock.Advance(c.DelayFirstProbeTime) wantProbes := []entryTestProbeInfo{ // The first probe is caused by the Unknown-to-Incomplete transition. @@ -2503,7 +2503,7 @@ func TestEntryProbeToReachableWhenSolicitedOverrideConfirmation(t *testing.T) { } e.mu.Unlock() - clock.advance(c.BaseReachableTime) + clock.Advance(c.BaseReachableTime) wantEvents := []testEntryEventInfo{ { @@ -2575,7 +2575,7 @@ func TestEntryProbeToReachableWhenSolicitedConfirmationWithSameAddress(t *testin e.handlePacketQueuedLocked() e.mu.Unlock() - clock.advance(c.DelayFirstProbeTime) + clock.Advance(c.DelayFirstProbeTime) wantProbes := []entryTestProbeInfo{ // The first probe is caused by the Unknown-to-Incomplete transition. @@ -2612,7 +2612,7 @@ func TestEntryProbeToReachableWhenSolicitedConfirmationWithSameAddress(t *testin } e.mu.Unlock() - clock.advance(c.BaseReachableTime) + clock.Advance(c.BaseReachableTime) wantEvents := []testEntryEventInfo{ { @@ -2682,7 +2682,7 @@ func TestEntryProbeToFailed(t *testing.T) { e.mu.Unlock() waitFor := c.DelayFirstProbeTime + c.RetransmitTimer*time.Duration(c.MaxUnicastProbes) - clock.advance(waitFor) + clock.Advance(waitFor) wantProbes := []entryTestProbeInfo{ // The first probe is caused by the Unknown-to-Incomplete transition. @@ -2787,7 +2787,7 @@ func TestEntryFailedGetsDeleted(t *testing.T) { e.mu.Unlock() waitFor := c.DelayFirstProbeTime + c.RetransmitTimer*time.Duration(c.MaxUnicastProbes) + c.UnreachableTime - clock.advance(waitFor) + clock.Advance(waitFor) wantProbes := []entryTestProbeInfo{ // The first probe is caused by the Unknown-to-Incomplete transition. diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index be274773c..8828cc5fe 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -18,7 +18,6 @@ import ( "fmt" "math/rand" "reflect" - "sort" "sync/atomic" "gvisor.dev/gvisor/pkg/sleep" @@ -28,39 +27,37 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" ) -var ipv4BroadcastAddr = tcpip.ProtocolAddress{ - Protocol: header.IPv4ProtocolNumber, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: header.IPv4Broadcast, - PrefixLen: 8 * header.IPv4AddressSize, - }, -} +var _ NetworkInterface = (*NIC)(nil) // NIC represents a "network interface card" to which the networking stack is // attached. type NIC struct { + LinkEndpoint + stack *Stack id tcpip.NICID name string - linkEP LinkEndpoint context NICContext - stats NICStats - neigh *neighborCache + stats NICStats + neigh *neighborCache + + // The network endpoints themselves may be modified by calling the interface's + // methods, but the map reference and entries must be constant. networkEndpoints map[tcpip.NetworkProtocolNumber]NetworkEndpoint + // enabled is set to 1 when the NIC is enabled and 0 when it is disabled. + // + // Must be accessed using atomic operations. + enabled uint32 + mu struct { sync.RWMutex - enabled bool spoofing bool promiscuous bool - primary map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint - endpoints map[NetworkEndpointID]*referencedNetworkEndpoint - mcastJoins map[NetworkEndpointID]uint32 // packetEPs is protected by mu, but the contained PacketEndpoint // values are not. packetEPs map[tcpip.NetworkProtocolNumber][]PacketEndpoint - ndp ndpState } } @@ -84,25 +81,6 @@ type DirectionStats struct { Bytes *tcpip.StatCounter } -// PrimaryEndpointBehavior is an enumeration of an endpoint's primacy behavior. -type PrimaryEndpointBehavior int - -const ( - // CanBePrimaryEndpoint indicates the endpoint can be used as a primary - // endpoint for new connections with no local address. This is the - // default when calling NIC.AddAddress. - CanBePrimaryEndpoint PrimaryEndpointBehavior = iota - - // FirstPrimaryEndpoint indicates the endpoint should be the first - // primary endpoint considered. If there are multiple endpoints with - // this behavior, the most recently-added one will be first. - FirstPrimaryEndpoint - - // NeverPrimaryEndpoint indicates the endpoint should never be a - // primary endpoint. - NeverPrimaryEndpoint -) - // newNIC returns a new NIC using the default NDP configurations from stack. func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICContext) *NIC { // TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For @@ -114,27 +92,16 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC // of IPv6 is supported on this endpoint's LinkEndpoint. nic := &NIC{ + LinkEndpoint: ep, + stack: stack, id: id, name: name, - linkEP: ep, context: ctx, stats: makeNICStats(), networkEndpoints: make(map[tcpip.NetworkProtocolNumber]NetworkEndpoint), } - nic.mu.primary = make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint) - nic.mu.endpoints = make(map[NetworkEndpointID]*referencedNetworkEndpoint) - nic.mu.mcastJoins = make(map[NetworkEndpointID]uint32) nic.mu.packetEPs = make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint) - nic.mu.ndp = ndpState{ - nic: nic, - configs: stack.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), - } - nic.mu.ndp.initializeTempAddrState() // Check for Neighbor Unreachability Detection support. var nud NUDHandler @@ -162,37 +129,40 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, ctx NICC for _, netProto := range stack.networkProtocols { netNum := netProto.Number() nic.mu.packetEPs[netNum] = nil - nic.networkEndpoints[netNum] = netProto.NewEndpoint(id, stack, nud, nic, ep, stack) + nic.networkEndpoints[netNum] = netProto.NewEndpoint(nic, stack, nud, nic) } - nic.linkEP.Attach(nic) + nic.LinkEndpoint.Attach(nic) return nic } -// enabled returns true if n is enabled. -func (n *NIC) enabled() bool { - n.mu.RLock() - enabled := n.mu.enabled - n.mu.RUnlock() - return enabled +func (n *NIC) getNetworkEndpoint(proto tcpip.NetworkProtocolNumber) NetworkEndpoint { + return n.networkEndpoints[proto] } -// disable disables n. +// Enabled implements NetworkInterface. +func (n *NIC) Enabled() bool { + return atomic.LoadUint32(&n.enabled) == 1 +} + +// setEnabled sets the enabled status for the NIC. // -// It undoes the work done by enable. -func (n *NIC) disable() *tcpip.Error { - n.mu.RLock() - enabled := n.mu.enabled - n.mu.RUnlock() - if !enabled { - return nil +// Returns true if the enabled status was updated. +func (n *NIC) setEnabled(v bool) bool { + if v { + return atomic.SwapUint32(&n.enabled, 1) == 0 } + return atomic.SwapUint32(&n.enabled, 0) == 1 +} +// disable disables n. +// +// It undoes the work done by enable. +func (n *NIC) disable() { n.mu.Lock() - err := n.disableLocked() + n.disableLocked() n.mu.Unlock() - return err } // disableLocked disables n. @@ -200,9 +170,9 @@ func (n *NIC) disable() *tcpip.Error { // It undoes the work done by enable. // // n MUST be locked. -func (n *NIC) disableLocked() *tcpip.Error { - if !n.mu.enabled { - return nil +func (n *NIC) disableLocked() { + if !n.setEnabled(false) { + return } // TODO(gvisor.dev/issue/1491): Should Routes that are currently bound to n be @@ -210,38 +180,9 @@ func (n *NIC) disableLocked() *tcpip.Error { // again, and applications may not know that the underlying NIC was ever // disabled. - if _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]; ok { - n.mu.ndp.stopSolicitingRouters() - n.mu.ndp.cleanupState(false /* hostOnly */) - - // Stop DAD for all the unicast IPv6 endpoints that are in the - // permanentTentative state. - for _, r := range n.mu.endpoints { - if addr := r.address(); r.getKind() == permanentTentative && header.IsV6UnicastAddress(addr) { - n.mu.ndp.stopDuplicateAddressDetection(addr) - } - } - - // The NIC may have already left the multicast group. - if err := n.leaveGroupLocked(header.IPv6AllNodesMulticastAddress, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { - return err - } - } - - if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok { - // The NIC may have already left the multicast group. - if err := n.leaveGroupLocked(header.IPv4AllSystems, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { - return err - } - - // The address may have already been removed. - if err := n.removePermanentAddressLocked(ipv4BroadcastAddr.AddressWithPrefix.Address); err != nil && err != tcpip.ErrBadLocalAddress { - return err - } + for _, ep := range n.networkEndpoints { + ep.Disable() } - - n.mu.enabled = false - return nil } // enable enables n. @@ -251,162 +192,38 @@ func (n *NIC) disableLocked() *tcpip.Error { // routers if the stack is not operating as a router. If the stack is also // configured to auto-generate a link-local address, one will be generated. func (n *NIC) enable() *tcpip.Error { - n.mu.RLock() - enabled := n.mu.enabled - n.mu.RUnlock() - if enabled { - return nil - } - n.mu.Lock() defer n.mu.Unlock() - if n.mu.enabled { - return nil - } - - n.mu.enabled = true - - // Create an endpoint to receive broadcast packets on this interface. - if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok { - if _, err := n.addAddressLocked(ipv4BroadcastAddr, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { - return err - } - - // 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. - if err := n.joinGroupLocked(header.IPv4ProtocolNumber, header.IPv4AllSystems); err != nil { - return err - } - } - - // 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 - // all-nodes multicast address. An example is the Neighbor Advertisement - // when we perform Duplicate Address Detection, or Router Advertisement - // when we do Router Discovery. See RFC 4862, section 5.4.2 and RFC 4861 - // section 4.2 for more information. - // - // Also auto-generate an IPv6 link-local address based on the NIC's - // link address if it is configured to do so. Note, each interface is - // required to have IPv6 link-local unicast address, as per RFC 4291 - // section 2.1. - _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber] - if !ok { + if !n.setEnabled(true) { return nil } - // Join the All-Nodes multicast group before starting DAD as responses to DAD - // (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 := n.joinGroupLocked(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress); err != nil { - return err - } - - // Perform DAD on the all the unicast IPv6 endpoints that are in the permanent - // state. - // - // Addresses may have aleady completed DAD but in the time since the NIC was - // last enabled, other devices may have acquired the same addresses. - for _, r := range n.mu.endpoints { - addr := r.address() - if k := r.getKind(); (k != permanent && k != permanentTentative) || !header.IsV6UnicastAddress(addr) { - continue - } - - r.setKind(permanentTentative) - if err := n.mu.ndp.startDuplicateAddressDetection(addr, r); err != nil { + for _, ep := range n.networkEndpoints { + if err := ep.Enable(); err != nil { return err } } - // Do not auto-generate an IPv6 link-local address for loopback devices. - if n.stack.autoGenIPv6LinkLocal && !n.isLoopback() { - // The valid and preferred lifetime is infinite for the auto-generated - // link-local address. - n.mu.ndp.doSLAAC(header.IPv6LinkLocalPrefix.Subnet(), header.NDPInfiniteLifetime, header.NDPInfiniteLifetime) - } - - // If we are operating as a router, then do not solicit routers since we - // won't process the RAs anyways. - // - // Routers do not process Router Advertisements (RA) the same way a host - // does. That is, routers do not learn from RAs (e.g. on-link prefixes - // and default routers). Therefore, soliciting RAs from other routers on - // a link is unnecessary for routers. - if !n.stack.Forwarding(header.IPv6ProtocolNumber) { - n.mu.ndp.startSolicitingRouters() - } - return nil } -// remove detaches NIC from the link endpoint, and marks existing referenced -// network endpoints expired. This guarantees no packets between this NIC and -// the network stack. +// remove detaches NIC from the link endpoint and releases network endpoint +// resources. This guarantees no packets between this NIC and the network +// stack. func (n *NIC) remove() *tcpip.Error { n.mu.Lock() defer n.mu.Unlock() n.disableLocked() - // TODO(b/151378115): come up with a better way to pick an error than the - // first one. - var err *tcpip.Error - - // Forcefully leave multicast groups. - for nid := range n.mu.mcastJoins { - if tempErr := n.leaveGroupLocked(nid.LocalAddress, true /* force */); tempErr != nil && err == nil { - err = tempErr - } - } - - // Remove permanent and permanentTentative addresses, so no packet goes out. - for nid, ref := range n.mu.endpoints { - switch ref.getKind() { - case permanentTentative, permanent: - if tempErr := n.removePermanentAddressLocked(nid.LocalAddress); tempErr != nil && err == nil { - err = tempErr - } - } - } - - // Release any resources the network endpoint may hold. for _, ep := range n.networkEndpoints { ep.Close() } // Detach from link endpoint, so no packet comes in. - n.linkEP.Attach(nil) - - return err -} - -// becomeIPv6Router transitions n into an IPv6 router. -// -// When transitioning into an IPv6 router, host-only state (NDP discovered -// routers, discovered on-link prefixes, and auto-generated addresses) will -// be cleaned up/invalidated and NDP router solicitations will be stopped. -func (n *NIC) becomeIPv6Router() { - n.mu.Lock() - defer n.mu.Unlock() - - n.mu.ndp.cleanupState(true /* hostOnly */) - n.mu.ndp.stopSolicitingRouters() -} - -// becomeIPv6Host transitions n into an IPv6 host. -// -// When transitioning into an IPv6 host, NDP router solicitations will be -// started. -func (n *NIC) becomeIPv6Host() { - n.mu.Lock() - defer n.mu.Unlock() - - n.mu.ndp.startSolicitingRouters() + n.LinkEndpoint.Attach(nil) + return nil } // setPromiscuousMode enables or disables promiscuous mode. @@ -423,217 +240,113 @@ func (n *NIC) isPromiscuousMode() bool { return rv } -func (n *NIC) isLoopback() bool { - return n.linkEP.Capabilities()&CapabilityLoopback != 0 -} - -// setSpoofing enables or disables address spoofing. -func (n *NIC) setSpoofing(enable bool) { - n.mu.Lock() - n.mu.spoofing = enable - n.mu.Unlock() +// IsLoopback implements NetworkInterface. +func (n *NIC) IsLoopback() bool { + return n.LinkEndpoint.Capabilities()&CapabilityLoopback != 0 } -// primaryEndpoint will return the first non-deprecated endpoint if such an -// endpoint exists for the given protocol and remoteAddr. If no non-deprecated -// endpoint exists, the first deprecated endpoint will be returned. -// -// If an IPv6 primary endpoint is requested, Source Address Selection (as -// defined by RFC 6724 section 5) will be performed. -func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber, remoteAddr tcpip.Address) *referencedNetworkEndpoint { - if protocol == header.IPv6ProtocolNumber && len(remoteAddr) != 0 { - return n.primaryIPv6Endpoint(remoteAddr) - } - - n.mu.RLock() - defer n.mu.RUnlock() - - var deprecatedEndpoint *referencedNetworkEndpoint - for _, r := range n.mu.primary[protocol] { - if !r.isValidForOutgoingRLocked() { - continue - } - - if !r.deprecated { - if r.tryIncRef() { - // r is not deprecated, so return it immediately. - // - // If we kept track of a deprecated endpoint, decrement its reference - // count since it was incremented when we decided to keep track of it. - if deprecatedEndpoint != nil { - deprecatedEndpoint.decRefLocked() - deprecatedEndpoint = nil - } - - return r - } - } else if deprecatedEndpoint == nil && r.tryIncRef() { - // We prefer an endpoint that is not deprecated, but we keep track of r in - // case n doesn't have any non-deprecated endpoints. - // - // If we end up finding a more preferred endpoint, r's reference count - // will be decremented when such an endpoint is found. - deprecatedEndpoint = r +// WritePacket implements NetworkLinkEndpoint. +func (n *NIC) WritePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error { + // As per relevant RFCs, we should queue packets while we wait for link + // resolution to complete. + // + // RFC 1122 section 2.3.2.2 (for IPv4): + // The link layer SHOULD save (rather than discard) at least + // one (the latest) packet of each set of packets destined to + // the same unresolved IP address, and transmit the saved + // packet when the address has been resolved. + // + // RFC 4861 section 5.2 (for IPv6): + // Once the IP address of the next-hop node is known, the sender + // examines the Neighbor Cache for link-layer information about that + // neighbor. If no entry exists, the sender creates one, sets its state + // to INCOMPLETE, initiates Address Resolution, and then queues the data + // packet pending completion of address resolution. + if ch, err := r.Resolve(nil); err != nil { + if err == tcpip.ErrWouldBlock { + r := r.Clone() + n.stack.linkResQueue.enqueue(ch, &r, protocol, pkt) + return nil } + return err } - // n doesn't have any valid non-deprecated endpoints, so return - // deprecatedEndpoint (which may be nil if n doesn't have any valid deprecated - // endpoints either). - return deprecatedEndpoint + return n.writePacket(r, gso, protocol, pkt) } -// ipv6AddrCandidate is an IPv6 candidate for Source Address Selection (RFC -// 6724 section 5). -type ipv6AddrCandidate struct { - ref *referencedNetworkEndpoint - scope header.IPv6AddressScope -} +func (n *NIC) writePacket(r *Route, gso *GSO, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) *tcpip.Error { + // WritePacket takes ownership of pkt, calculate numBytes first. + numBytes := pkt.Size() -// primaryIPv6Endpoint returns an IPv6 endpoint following Source Address -// Selection (RFC 6724 section 5). -// -// Note, only rules 1-3 and 7 are followed. -// -// remoteAddr must be a valid IPv6 address. -func (n *NIC) primaryIPv6Endpoint(remoteAddr tcpip.Address) *referencedNetworkEndpoint { - n.mu.RLock() - ref := n.primaryIPv6EndpointRLocked(remoteAddr) - n.mu.RUnlock() - return ref -} - -// primaryIPv6EndpointLocked returns an IPv6 endpoint following Source Address -// Selection (RFC 6724 section 5). -// -// Note, only rules 1-3 and 7 are followed. -// -// remoteAddr must be a valid IPv6 address. -// -// n.mu MUST be read locked. -func (n *NIC) primaryIPv6EndpointRLocked(remoteAddr tcpip.Address) *referencedNetworkEndpoint { - primaryAddrs := n.mu.primary[header.IPv6ProtocolNumber] - - if len(primaryAddrs) == 0 { - return nil + if err := n.LinkEndpoint.WritePacket(r, gso, protocol, pkt); err != nil { + return err } - // Create a candidate set of available addresses we can potentially use as a - // source address. - cs := make([]ipv6AddrCandidate, 0, len(primaryAddrs)) - for _, r := range primaryAddrs { - // If r is not valid for outgoing connections, it is not a valid endpoint. - if !r.isValidForOutgoingRLocked() { - continue - } - - addr := r.address() - scope, err := header.ScopeForIPv6Address(addr) - if err != nil { - // Should never happen as we got r from the primary IPv6 endpoint list and - // ScopeForIPv6Address only returns an error if addr is not an IPv6 - // address. - panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", addr, err)) - } - - cs = append(cs, ipv6AddrCandidate{ - ref: r, - scope: scope, - }) - } + n.stats.Tx.Packets.Increment() + n.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) + return nil +} - remoteScope, err := header.ScopeForIPv6Address(remoteAddr) - if err != nil { - // primaryIPv6Endpoint should never be called with an invalid IPv6 address. - panic(fmt.Sprintf("header.ScopeForIPv6Address(%s): %s", remoteAddr, err)) +// WritePackets implements NetworkLinkEndpoint. +func (n *NIC) WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + // TODO(gvisor.dev/issue/4458): Queue packets whie link address resolution + // is being peformed like WritePacket. + writtenPackets, err := n.LinkEndpoint.WritePackets(r, gso, pkts, protocol) + n.stats.Tx.Packets.IncrementBy(uint64(writtenPackets)) + writtenBytes := 0 + for i, pb := 0, pkts.Front(); i < writtenPackets && pb != nil; i, pb = i+1, pb.Next() { + writtenBytes += pb.Size() } - // Sort the addresses as per RFC 6724 section 5 rules 1-3. - // - // TODO(b/146021396): Implement rules 4-8 of RFC 6724 section 5. - sort.Slice(cs, func(i, j int) bool { - sa := cs[i] - sb := cs[j] - - // Prefer same address as per RFC 6724 section 5 rule 1. - if sa.ref.address() == remoteAddr { - return true - } - if sb.ref.address() == remoteAddr { - return false - } - - // Prefer appropriate scope as per RFC 6724 section 5 rule 2. - if sa.scope < sb.scope { - return sa.scope >= remoteScope - } else if sb.scope < sa.scope { - return sb.scope < remoteScope - } - - // Avoid deprecated addresses as per RFC 6724 section 5 rule 3. - if saDep, sbDep := sa.ref.deprecated, sb.ref.deprecated; saDep != sbDep { - // If sa is not deprecated, it is preferred over sb. - return sbDep - } - - // Prefer temporary addresses as per RFC 6724 section 5 rule 7. - if saTemp, sbTemp := sa.ref.configType == slaacTemp, sb.ref.configType == slaacTemp; saTemp != sbTemp { - return saTemp - } - - // sa and sb are equal, return the endpoint that is closest to the front of - // the primary endpoint list. - return i < j - }) - - // Return the most preferred address that can have its reference count - // incremented. - for _, c := range cs { - if r := c.ref; r.tryIncRef() { - return r - } - } + n.stats.Tx.Bytes.IncrementBy(uint64(writtenBytes)) + return writtenPackets, err +} - return nil +// setSpoofing enables or disables address spoofing. +func (n *NIC) setSpoofing(enable bool) { + n.mu.Lock() + n.mu.spoofing = enable + n.mu.Unlock() } -// hasPermanentAddrLocked returns true if n has a permanent (including currently -// tentative) address, addr. -func (n *NIC) hasPermanentAddrLocked(addr tcpip.Address) bool { - ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] +// primaryAddress returns an address that can be used to communicate with +// remoteAddr. +func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber, remoteAddr tcpip.Address) AssignableAddressEndpoint { + n.mu.RLock() + spoofing := n.mu.spoofing + n.mu.RUnlock() + ep, ok := n.networkEndpoints[protocol] if !ok { - return false + return nil } - kind := ref.getKind() - - return kind == permanent || kind == permanentTentative + return ep.AcquireOutgoingPrimaryAddress(remoteAddr, spoofing) } -type getRefBehaviour int +type getAddressBehaviour int const ( // spoofing indicates that the NIC's spoofing flag should be observed when - // getting a NIC's referenced network endpoint. - spoofing getRefBehaviour = iota + // getting a NIC's address endpoint. + spoofing getAddressBehaviour = iota // promiscuous indicates that the NIC's promiscuous flag should be observed - // when getting a NIC's referenced network endpoint. + // when getting a NIC's address endpoint. promiscuous ) -func (n *NIC) getRef(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) *referencedNetworkEndpoint { - return n.getRefOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, promiscuous) +func (n *NIC) getAddress(protocol tcpip.NetworkProtocolNumber, dst tcpip.Address) AssignableAddressEndpoint { + return n.getAddressOrCreateTemp(protocol, dst, CanBePrimaryEndpoint, promiscuous) } // findEndpoint finds the endpoint, if any, with the given address. -func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) *referencedNetworkEndpoint { - return n.getRefOrCreateTemp(protocol, address, peb, spoofing) +func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) AssignableAddressEndpoint { + return n.getAddressOrCreateTemp(protocol, address, peb, spoofing) } -// getRefEpOrCreateTemp returns the referenced network endpoint for the given -// protocol and address. +// getAddressEpOrCreateTemp returns the address endpoint for the given protocol +// and address. // // If none exists a temporary one may be created if we are in promiscuous mode // or spoofing. Promiscuous mode will only be checked if promiscuous is true. @@ -641,9 +354,8 @@ func (n *NIC) findEndpoint(protocol tcpip.NetworkProtocolNumber, address tcpip.A // // If the address is the IPv4 broadcast address for an endpoint's network, that // endpoint will be returned. -func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, tempRef getRefBehaviour) *referencedNetworkEndpoint { +func (n *NIC) getAddressOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior, tempRef getAddressBehaviour) AssignableAddressEndpoint { n.mu.RLock() - var spoofingOrPromiscuous bool switch tempRef { case spoofing: @@ -651,274 +363,54 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t case promiscuous: spoofingOrPromiscuous = n.mu.promiscuous } - - if ref, ok := n.mu.endpoints[NetworkEndpointID{address}]; ok { - // An endpoint with this id exists, check if it can be used and return it. - if !ref.isAssignedRLocked(spoofingOrPromiscuous) { - n.mu.RUnlock() - return nil - } - - if ref.tryIncRef() { - n.mu.RUnlock() - return ref - } - } - - if protocol == header.IPv4ProtocolNumber { - if ref := n.getIPv4RefForBroadcastOrLoopbackRLocked(address); ref != nil { - n.mu.RUnlock() - return ref - } - } n.mu.RUnlock() - - if !spoofingOrPromiscuous { - return nil - } - - // Try again with the lock in exclusive mode. If we still can't get the - // endpoint, create a new "temporary" endpoint. It will only exist while - // there's a route through it. - n.mu.Lock() - ref := n.getRefOrCreateTempLocked(protocol, address, peb) - n.mu.Unlock() - return ref + return n.getAddressOrCreateTempInner(protocol, address, spoofingOrPromiscuous, peb) } -// getRefForBroadcastOrLoopbackRLocked returns an endpoint whose address is the -// broadcast address for the endpoint's network or an address in the endpoint's -// subnet if the NIC is a loopback interface. This matches linux behaviour. -// -// n.mu MUST be read or write locked. -func (n *NIC) getIPv4RefForBroadcastOrLoopbackRLocked(address tcpip.Address) *referencedNetworkEndpoint { - for _, ref := range n.mu.endpoints { - // Only IPv4 has a notion of broadcast addresses or considers the loopback - // interface bound to an address's whole subnet (on linux). - if ref.protocol != header.IPv4ProtocolNumber { - continue - } - - subnet := ref.addrWithPrefix().Subnet() - if (subnet.IsBroadcast(address) || (n.isLoopback() && subnet.Contains(address))) && ref.isValidForOutgoingRLocked() && ref.tryIncRef() { - return ref - } +// getAddressOrCreateTempInner is like getAddressEpOrCreateTemp except a boolean +// is passed to indicate whether or not we should generate temporary endpoints. +func (n *NIC) getAddressOrCreateTempInner(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, createTemp bool, peb PrimaryEndpointBehavior) AssignableAddressEndpoint { + if ep, ok := n.networkEndpoints[protocol]; ok { + return ep.AcquireAssignedAddress(address, createTemp, peb) } return nil } -/// getRefOrCreateTempLocked returns an existing endpoint for address or creates -/// and returns a temporary endpoint. -// -// If the address is the IPv4 broadcast address for an endpoint's network, that -// endpoint will be returned. -// -// n.mu must be write locked. -func (n *NIC) getRefOrCreateTempLocked(protocol tcpip.NetworkProtocolNumber, address tcpip.Address, peb PrimaryEndpointBehavior) *referencedNetworkEndpoint { - if ref, ok := n.mu.endpoints[NetworkEndpointID{address}]; ok { - // No need to check the type as we are ok with expired endpoints at this - // point. - if ref.tryIncRef() { - return ref - } - // tryIncRef failing means the endpoint is scheduled to be removed once the - // lock is released. Remove it here so we can create a new (temporary) one. - // The removal logic waiting for the lock handles this case. - n.removeEndpointLocked(ref) - } - - if protocol == header.IPv4ProtocolNumber { - if ref := n.getIPv4RefForBroadcastOrLoopbackRLocked(address); ref != nil { - return ref - } - } - - // Add a new temporary endpoint. - netProto, ok := n.stack.networkProtocols[protocol] - if !ok { - return nil - } - ref, _ := n.addAddressLocked(tcpip.ProtocolAddress{ - Protocol: protocol, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: address, - PrefixLen: netProto.DefaultPrefixLen(), - }, - }, peb, temporary, static, false) - return ref -} - -// addAddressLocked adds a new protocolAddress to n. -// -// If n already has the address in a non-permanent state, and the kind given is -// permanent, that address will be promoted in place and its properties set to -// the properties provided. Otherwise, it returns tcpip.ErrDuplicateAddress. -func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType, deprecated bool) (*referencedNetworkEndpoint, *tcpip.Error) { - // TODO(b/141022673): Validate IP addresses before adding them. - - // Sanity check. - id := NetworkEndpointID{LocalAddress: protocolAddress.AddressWithPrefix.Address} - if ref, ok := n.mu.endpoints[id]; ok { - // Endpoint already exists. - if kind != permanent { - return nil, tcpip.ErrDuplicateAddress - } - switch ref.getKind() { - case permanentTentative, permanent: - // The NIC already have a permanent endpoint with that address. - return nil, tcpip.ErrDuplicateAddress - case permanentExpired, temporary: - // Promote the endpoint to become permanent and respect the new peb, - // configType and deprecated status. - if ref.tryIncRef() { - // TODO(b/147748385): Perform Duplicate Address Detection when promoting - // an IPv6 endpoint to permanent. - ref.setKind(permanent) - ref.deprecated = deprecated - ref.configType = configType - - refs := n.mu.primary[ref.protocol] - for i, r := range refs { - if r == ref { - switch peb { - case CanBePrimaryEndpoint: - return ref, nil - case FirstPrimaryEndpoint: - if i == 0 { - return ref, nil - } - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - case NeverPrimaryEndpoint: - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - return ref, nil - } - } - } - - n.insertPrimaryEndpointLocked(ref, peb) - - return ref, nil - } - // tryIncRef failing means the endpoint is scheduled to be removed once - // the lock is released. Remove it here so we can create a new - // (permanent) one. The removal logic waiting for the lock handles this - // case. - n.removeEndpointLocked(ref) - } - } - +// addAddress adds a new address to n, so that it starts accepting packets +// targeted at the given address (and network protocol). +func (n *NIC) addAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) *tcpip.Error { ep, ok := n.networkEndpoints[protocolAddress.Protocol] if !ok { - return nil, tcpip.ErrUnknownProtocol - } - - isIPv6Unicast := protocolAddress.Protocol == header.IPv6ProtocolNumber && header.IsV6UnicastAddress(protocolAddress.AddressWithPrefix.Address) - - // If the address is an IPv6 address and it is a permanent address, - // mark it as tentative so it goes through the DAD process if the NIC is - // enabled. If the NIC is not enabled, DAD will be started when the NIC is - // enabled. - if isIPv6Unicast && kind == permanent { - kind = permanentTentative + return tcpip.ErrUnknownProtocol } - ref := &referencedNetworkEndpoint{ - refs: 1, - addr: protocolAddress.AddressWithPrefix, - ep: ep, - nic: n, - protocol: protocolAddress.Protocol, - kind: kind, - configType: configType, - deprecated: deprecated, + addressEndpoint, err := ep.AddAndAcquirePermanentAddress(protocolAddress.AddressWithPrefix, peb, AddressConfigStatic, false /* deprecated */) + if err == nil { + // We have no need for the address endpoint. + addressEndpoint.DecRef() } - - // Set up resolver if link address resolution exists for this protocol. - if n.linkEP.Capabilities()&CapabilityResolutionRequired != 0 { - if linkRes, ok := n.stack.linkAddrResolvers[protocolAddress.Protocol]; ok { - ref.linkCache = n.stack - ref.linkRes = linkRes - } - } - - // If we are adding an IPv6 unicast address, join the solicited-node - // multicast address. - if isIPv6Unicast { - snmc := header.SolicitedNodeAddr(protocolAddress.AddressWithPrefix.Address) - if err := n.joinGroupLocked(protocolAddress.Protocol, snmc); err != nil { - return nil, err - } - } - - n.mu.endpoints[id] = ref - - n.insertPrimaryEndpointLocked(ref, peb) - - // If we are adding a tentative IPv6 address, start DAD if the NIC is enabled. - if isIPv6Unicast && kind == permanentTentative && n.mu.enabled { - if err := n.mu.ndp.startDuplicateAddressDetection(protocolAddress.AddressWithPrefix.Address, ref); err != nil { - return nil, err - } - } - - return ref, nil -} - -// AddAddress adds a new address to n, so that it starts accepting packets -// targeted at the given address (and network protocol). -func (n *NIC) AddAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) *tcpip.Error { - // Add the endpoint. - n.mu.Lock() - _, err := n.addAddressLocked(protocolAddress, peb, permanent, static, false /* deprecated */) - n.mu.Unlock() - return err } -// AllAddresses returns all addresses (primary and non-primary) associated with +// allPermanentAddresses returns all permanent addresses associated with // this NIC. -func (n *NIC) AllAddresses() []tcpip.ProtocolAddress { - n.mu.RLock() - defer n.mu.RUnlock() - - addrs := make([]tcpip.ProtocolAddress, 0, len(n.mu.endpoints)) - for _, ref := range n.mu.endpoints { - // Don't include tentative, expired or temporary endpoints to - // avoid confusion and prevent the caller from using those. - switch ref.getKind() { - case permanentExpired, temporary: - continue +func (n *NIC) allPermanentAddresses() []tcpip.ProtocolAddress { + var addrs []tcpip.ProtocolAddress + for p, ep := range n.networkEndpoints { + for _, a := range ep.PermanentAddresses() { + addrs = append(addrs, tcpip.ProtocolAddress{Protocol: p, AddressWithPrefix: a}) } - - addrs = append(addrs, tcpip.ProtocolAddress{ - Protocol: ref.protocol, - AddressWithPrefix: ref.addrWithPrefix(), - }) } return addrs } -// PrimaryAddresses returns the primary addresses associated with this NIC. -func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { - n.mu.RLock() - defer n.mu.RUnlock() - +// primaryAddresses returns the primary addresses associated with this NIC. +func (n *NIC) primaryAddresses() []tcpip.ProtocolAddress { var addrs []tcpip.ProtocolAddress - for proto, list := range n.mu.primary { - for _, ref := range list { - // Don't include tentative, expired or tempory endpoints - // to avoid confusion and prevent the caller from using - // those. - switch ref.getKind() { - case permanentTentative, permanentExpired, temporary: - continue - } - - addrs = append(addrs, tcpip.ProtocolAddress{ - Protocol: proto, - AddressWithPrefix: ref.addrWithPrefix(), - }) + for p, ep := range n.networkEndpoints { + for _, a := range ep.PrimaryAddresses() { + addrs = append(addrs, tcpip.ProtocolAddress{Protocol: p, AddressWithPrefix: a}) } } return addrs @@ -930,147 +422,25 @@ func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { // address exists. If no non-deprecated address exists, the first deprecated // address will be returned. func (n *NIC) primaryAddress(proto tcpip.NetworkProtocolNumber) tcpip.AddressWithPrefix { - n.mu.RLock() - defer n.mu.RUnlock() - - list, ok := n.mu.primary[proto] + ep, ok := n.networkEndpoints[proto] if !ok { return tcpip.AddressWithPrefix{} } - var deprecatedEndpoint *referencedNetworkEndpoint - for _, ref := range list { - // Don't include tentative, expired or tempory endpoints to avoid confusion - // and prevent the caller from using those. - switch ref.getKind() { - case permanentTentative, permanentExpired, temporary: - continue - } - - if !ref.deprecated { - return ref.addrWithPrefix() - } - - if deprecatedEndpoint == nil { - deprecatedEndpoint = ref - } - } - - if deprecatedEndpoint != nil { - return deprecatedEndpoint.addrWithPrefix() - } - - return tcpip.AddressWithPrefix{} -} - -// insertPrimaryEndpointLocked adds r to n's primary endpoint list as required -// by peb. -// -// n MUST be locked. -func (n *NIC) insertPrimaryEndpointLocked(r *referencedNetworkEndpoint, peb PrimaryEndpointBehavior) { - switch peb { - case CanBePrimaryEndpoint: - n.mu.primary[r.protocol] = append(n.mu.primary[r.protocol], r) - case FirstPrimaryEndpoint: - n.mu.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.mu.primary[r.protocol]...) - } -} - -func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { - id := NetworkEndpointID{LocalAddress: r.address()} - - // Nothing to do if the reference has already been replaced with a different - // one. This happens in the case where 1) this endpoint's ref count hit zero - // and was waiting (on the lock) to be removed and 2) the same address was - // re-added in the meantime by removing this endpoint from the list and - // adding a new one. - if n.mu.endpoints[id] != r { - return - } - - if r.getKind() == permanent { - panic("Reference count dropped to zero before being removed") - } - - delete(n.mu.endpoints, id) - refs := n.mu.primary[r.protocol] - for i, ref := range refs { - if ref == r { - n.mu.primary[r.protocol] = append(refs[:i], refs[i+1:]...) - refs[len(refs)-1] = nil - break - } - } -} - -func (n *NIC) removeEndpoint(r *referencedNetworkEndpoint) { - n.mu.Lock() - n.removeEndpointLocked(r) - n.mu.Unlock() -} - -func (n *NIC) removePermanentAddressLocked(addr tcpip.Address) *tcpip.Error { - r, ok := n.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return tcpip.ErrBadLocalAddress - } - - kind := r.getKind() - if kind != permanent && kind != permanentTentative { - return tcpip.ErrBadLocalAddress - } - - switch r.protocol { - case header.IPv6ProtocolNumber: - return n.removePermanentIPv6EndpointLocked(r, true /* allowSLAACInvalidation */) - default: - r.expireLocked() - return nil - } + return ep.MainAddress() } -func (n *NIC) removePermanentIPv6EndpointLocked(r *referencedNetworkEndpoint, allowSLAACInvalidation bool) *tcpip.Error { - addr := r.addrWithPrefix() - - isIPv6Unicast := header.IsV6UnicastAddress(addr.Address) - - if isIPv6Unicast { - n.mu.ndp.stopDuplicateAddressDetection(addr.Address) - - // If we are removing an address generated via SLAAC, cleanup - // its SLAAC resources and notify the integrator. - switch r.configType { - case slaac: - n.mu.ndp.cleanupSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) - case slaacTemp: - n.mu.ndp.cleanupTempSLAACAddrResourcesAndNotify(addr, allowSLAACInvalidation) - } - } - - r.expireLocked() - - // At this point the endpoint is deleted. - - // If we are removing an IPv6 unicast address, leave the solicited-node - // multicast address. - // - // We ignore the tcpip.ErrBadLocalAddress error because the solicited-node - // multicast group may be left by user action. - if isIPv6Unicast { - snmc := header.SolicitedNodeAddr(addr.Address) - if err := n.leaveGroupLocked(snmc, false /* force */); err != nil && err != tcpip.ErrBadLocalAddress { +// removeAddress removes an address from n. +func (n *NIC) removeAddress(addr tcpip.Address) *tcpip.Error { + for _, ep := range n.networkEndpoints { + if err := ep.RemovePermanentAddress(addr); err == tcpip.ErrBadLocalAddress { + continue + } else { return err } } - return nil -} - -// RemoveAddress removes an address from n. -func (n *NIC) RemoveAddress(addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - return n.removePermanentAddressLocked(addr) + return tcpip.ErrBadLocalAddress } func (n *NIC) neighbors() ([]NeighborEntry, *tcpip.Error) { @@ -1121,91 +491,66 @@ func (n *NIC) clearNeighbors() *tcpip.Error { // joinGroup adds a new endpoint for the given multicast address, if none // exists yet. Otherwise it just increments its count. func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - - return n.joinGroupLocked(protocol, addr) -} - -// joinGroupLocked adds a new endpoint for the given multicast address, if none -// exists yet. Otherwise it just increments its count. n MUST be locked before -// joinGroupLocked is called. -func (n *NIC) joinGroupLocked(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { // TODO(b/143102137): When implementing MLD, make sure MLD packets are // not sent unless a valid link-local address is available for use on n // as an MLD packet's source address must be a link-local address as // outlined in RFC 3810 section 5. - id := NetworkEndpointID{addr} - joins := n.mu.mcastJoins[id] - if joins == 0 { - netProto, ok := n.stack.networkProtocols[protocol] - if !ok { - return tcpip.ErrUnknownProtocol - } - if _, err := n.addAddressLocked(tcpip.ProtocolAddress{ - Protocol: protocol, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: addr, - PrefixLen: netProto.DefaultPrefixLen(), - }, - }, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { - return err - } + ep, ok := n.networkEndpoints[protocol] + if !ok { + return tcpip.ErrNotSupported } - n.mu.mcastJoins[id] = joins + 1 - return nil + + gep, ok := ep.(GroupAddressableEndpoint) + if !ok { + return tcpip.ErrNotSupported + } + + _, err := gep.JoinGroup(addr) + return err } // leaveGroup decrements the count for the given multicast address, and when it // reaches zero removes the endpoint for this address. -func (n *NIC) leaveGroup(addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - - return n.leaveGroupLocked(addr, false /* force */) -} +func (n *NIC) leaveGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address) *tcpip.Error { + ep, ok := n.networkEndpoints[protocol] + if !ok { + return tcpip.ErrNotSupported + } -// leaveGroupLocked decrements the count for the given multicast address, and -// when it reaches zero removes the endpoint for this address. n MUST be locked -// before leaveGroupLocked is called. -// -// If force is true, then the count for the multicast addres is ignored and the -// endpoint will be removed immediately. -func (n *NIC) leaveGroupLocked(addr tcpip.Address, force bool) *tcpip.Error { - id := NetworkEndpointID{addr} - joins, ok := n.mu.mcastJoins[id] + gep, ok := ep.(GroupAddressableEndpoint) if !ok { - // There are no joins with this address on this NIC. - return tcpip.ErrBadLocalAddress + return tcpip.ErrNotSupported } - joins-- - if force || joins == 0 { - // There are no outstanding joins or we are forced to leave, clean up. - delete(n.mu.mcastJoins, id) - return n.removePermanentAddressLocked(addr) + if _, err := gep.LeaveGroup(addr); err != nil { + return err } - n.mu.mcastJoins[id] = joins return nil } // isInGroup returns true if n has joined the multicast group addr. func (n *NIC) isInGroup(addr tcpip.Address) bool { - n.mu.RLock() - joins := n.mu.mcastJoins[NetworkEndpointID{addr}] - n.mu.RUnlock() + for _, ep := range n.networkEndpoints { + gep, ok := ep.(GroupAddressableEndpoint) + if !ok { + continue + } - return joins != 0 + if gep.IsInGroup(addr) { + return true + } + } + + return false } -func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, localLinkAddr, remotelinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, pkt *PacketBuffer) { - r := makeRoute(protocol, dst, src, localLinkAddr, ref, false /* handleLocal */, false /* multicastLoop */) +func (n *NIC) handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, remotelinkAddr tcpip.LinkAddress, addressEndpoint AssignableAddressEndpoint, pkt *PacketBuffer) { + r := makeRoute(protocol, dst, src, n, addressEndpoint, false /* handleLocal */, false /* multicastLoop */) + defer r.Release() r.RemoteLinkAddress = remotelinkAddr - - ref.ep.HandlePacket(&r, pkt) - ref.decRef() + n.getNetworkEndpoint(protocol).HandlePacket(&r, pkt) } // DeliverNetworkPacket finds the appropriate network protocol endpoint and @@ -1216,7 +561,7 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // the ownership of the items is not retained by the caller. func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { n.mu.RLock() - enabled := n.mu.enabled + enabled := n.Enabled() // If the NIC is not yet enabled, don't receive any packets. if !enabled { n.mu.RUnlock() @@ -1239,12 +584,12 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp // If no local link layer address is provided, assume it was sent // directly to this NIC. if local == "" { - local = n.linkEP.LinkAddress() + local = n.LinkEndpoint.LinkAddress() } - // Are any packet sockets listening for this network protocol? + // Are any packet type sockets listening for this network protocol? packetEPs := n.mu.packetEPs[protocol] - // Add any other packet sockets that maybe listening for all protocols. + // Add any other packet type sockets that may be listening for all protocols. packetEPs = append(packetEPs, n.mu.packetEPs[header.EthernetProtocolAll]...) n.mu.RUnlock() for _, ep := range packetEPs { @@ -1265,6 +610,7 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp return } if hasTransportHdr { + pkt.TransportProtocolNumber = transProtoNum // Parse the transport header if present. if state, ok := n.stack.transportProtocols[transProtoNum]; ok { state.proto.Parse(pkt) @@ -1273,17 +619,21 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp src, dst := netProto.ParseAddresses(pkt.NetworkHeader().View()) - if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { - // The source address is one of our own, so we never should have gotten a - // packet like this unless handleLocal is false. Loopback also calls this - // function even though the packets didn't come from the physical interface - // so don't drop those. - n.stack.stats.IP.InvalidSourceAddressesReceived.Increment() - return + if n.stack.handleLocal && !n.IsLoopback() { + if r := n.getAddress(protocol, src); r != nil { + r.DecRef() + + // The source address is one of our own, so we never should have gotten a + // packet like this unless handleLocal is false. Loopback also calls this + // function even though the packets didn't come from the physical interface + // so don't drop those. + n.stack.stats.IP.InvalidSourceAddressesReceived.Increment() + return + } } // Loopback traffic skips the prerouting chain. - if !n.isLoopback() { + if !n.IsLoopback() { // iptables filtering. ipt := n.stack.IPTables() address := n.primaryAddress(protocol) @@ -1294,8 +644,8 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } } - if ref := n.getRef(protocol, dst); ref != nil { - handlePacket(protocol, dst, src, n.linkEP.LinkAddress(), remote, ref, pkt) + if addressEndpoint := n.getAddress(protocol, dst); addressEndpoint != nil { + n.handlePacket(protocol, dst, src, remote, addressEndpoint, pkt) return } @@ -1311,39 +661,39 @@ func (n *NIC) DeliverNetworkPacket(remote, local tcpip.LinkAddress, protocol tcp } // Found a NIC. - n := r.ref.nic - n.mu.RLock() - ref, ok := n.mu.endpoints[NetworkEndpointID{dst}] - ok = ok && ref.isValidForOutgoingRLocked() && ref.tryIncRef() - n.mu.RUnlock() - if ok { - r.LocalLinkAddress = n.linkEP.LinkAddress() - r.RemoteLinkAddress = remote - r.RemoteAddress = src - // TODO(b/123449044): Update the source NIC as well. - ref.ep.HandlePacket(&r, pkt) - ref.decRef() - r.Release() - return + n := r.nic + if addressEndpoint := n.getAddressOrCreateTempInner(protocol, dst, false, NeverPrimaryEndpoint); addressEndpoint != nil { + if n.isValidForOutgoing(addressEndpoint) { + r.LocalLinkAddress = n.LinkEndpoint.LinkAddress() + r.RemoteLinkAddress = remote + r.RemoteAddress = src + // TODO(b/123449044): Update the source NIC as well. + n.getNetworkEndpoint(protocol).HandlePacket(&r, pkt) + addressEndpoint.DecRef() + r.Release() + return + } + + addressEndpoint.DecRef() } // n doesn't have a destination endpoint. // Send the packet out of n. - // TODO(b/128629022): move this logic to route.WritePacket. // TODO(gvisor.dev/issue/1085): According to the RFC, we must decrease the TTL field for ipv4/ipv6. - if ch, err := r.Resolve(nil); err != nil { - if err == tcpip.ErrWouldBlock { - n.stack.forwarder.enqueue(ch, n, &r, protocol, pkt) - // forwarder will release route. - return - } + + // pkt may have set its header and may not have enough headroom for + // link-layer header for the other link to prepend. Here we create a new + // packet to forward. + fwdPkt := NewPacketBuffer(PacketBufferOptions{ + ReserveHeaderBytes: int(n.LinkEndpoint.MaxHeaderLength()), + Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), + }) + + // TODO(b/143425874) Decrease the TTL field in forwarded packets. + if err := n.WritePacket(&r, nil, protocol, fwdPkt); err != nil { n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() - r.Release() - return } - // The link-address resolution finished immediately. - n.forwardPacket(&r, protocol, pkt) r.Release() return } @@ -1367,41 +717,18 @@ func (n *NIC) DeliverOutboundPacket(remote, local tcpip.LinkAddress, protocol tc p.PktType = tcpip.PacketOutgoing // Add the link layer header as outgoing packets are intercepted // before the link layer header is created. - n.linkEP.AddHeader(local, remote, protocol, p) + n.LinkEndpoint.AddHeader(local, remote, protocol, p) ep.HandlePacket(n.id, local, protocol, p) } } -func (n *NIC) forwardPacket(r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { - // TODO(b/143425874) Decrease the TTL field in forwarded packets. - - // pkt may have set its header and may not have enough headroom for link-layer - // header for the other link to prepend. Here we create a new packet to - // forward. - fwdPkt := NewPacketBuffer(PacketBufferOptions{ - ReserveHeaderBytes: int(n.linkEP.MaxHeaderLength()), - Data: buffer.NewVectorisedView(pkt.Size(), pkt.Views()), - }) - - // WritePacket takes ownership of fwdPkt, calculate numBytes first. - numBytes := fwdPkt.Size() - - if err := n.linkEP.WritePacket(r, nil /* gso */, protocol, fwdPkt); err != nil { - r.Stats().IP.OutgoingPacketErrors.Increment() - return - } - - n.stats.Tx.Packets.Increment() - n.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) -} - // DeliverTransportPacket delivers the packets to the appropriate transport // protocol endpoint. -func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) { +func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) TransportPacketDisposition { state, ok := n.stack.transportProtocols[protocol] if !ok { n.stack.stats.UnknownProtocolRcvdPackets.Increment() - return + return TransportPacketProtocolUnreachable } transProto := state.proto @@ -1422,41 +749,47 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // we parse it using the minimum size. if _, ok := pkt.TransportHeader().Consume(transProto.MinimumPacketSize()); !ok { n.stack.stats.MalformedRcvdPackets.Increment() - return + // We consider a malformed transport packet handled because there is + // nothing the caller can do. + return TransportPacketHandled } - } else { - // This is either a bad packet or was re-assembled from fragments. - transProto.Parse(pkt) + } else if !transProto.Parse(pkt) { + n.stack.stats.MalformedRcvdPackets.Increment() + return TransportPacketHandled } } - if pkt.TransportHeader().View().Size() < transProto.MinimumPacketSize() { - n.stack.stats.MalformedRcvdPackets.Increment() - return - } - srcPort, dstPort, err := transProto.ParsePorts(pkt.TransportHeader().View()) if err != nil { n.stack.stats.MalformedRcvdPackets.Increment() - return + return TransportPacketHandled } id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress} if n.stack.demux.deliverPacket(r, protocol, pkt, id) { - return + return TransportPacketHandled } // Try to deliver to per-stack default handler. if state.defaultHandler != nil { if state.defaultHandler(r, id, pkt) { - return + return TransportPacketHandled } } - // We could not find an appropriate destination for this packet, so - // deliver it to the global handler. - if !transProto.HandleUnknownDestinationPacket(r, id, pkt) { + // We could not find an appropriate destination for this packet so + // give the protocol specific error handler a chance to handle it. + // If it doesn't handle it then we should do so. + switch res := transProto.HandleUnknownDestinationPacket(r, id, pkt); res { + case UnknownDestinationPacketMalformed: n.stack.stats.MalformedRcvdPackets.Increment() + return TransportPacketHandled + case UnknownDestinationPacketUnhandled: + return TransportPacketDestinationPortUnreachable + case UnknownDestinationPacketHandled: + return TransportPacketHandled + default: + panic(fmt.Sprintf("unrecognized result from HandleUnknownDestinationPacket = %d", res)) } } @@ -1489,96 +822,18 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp } } -// ID returns the identifier of n. +// ID implements NetworkInterface. func (n *NIC) ID() tcpip.NICID { return n.id } -// Name returns the name of n. +// Name implements NetworkInterface. func (n *NIC) Name() string { return n.name } -// Stack returns the instance of the Stack that owns this NIC. -func (n *NIC) Stack() *Stack { - return n.stack -} - -// LinkEndpoint returns the link endpoint of n. -func (n *NIC) LinkEndpoint() LinkEndpoint { - return n.linkEP -} - -// isAddrTentative returns true if addr is tentative on n. -// -// Note that if addr is not associated with n, then this function will return -// false. It will only return true if the address is associated with the NIC -// AND it is tentative. -func (n *NIC) isAddrTentative(addr tcpip.Address) bool { - n.mu.RLock() - defer n.mu.RUnlock() - - ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return false - } - - return ref.getKind() == permanentTentative -} - -// dupTentativeAddrDetected attempts to inform n that a tentative addr is a -// duplicate on a link. -// -// dupTentativeAddrDetected will remove the tentative address if it exists. If -// the address was generated via SLAAC, an attempt will be made to generate a -// new address. -func (n *NIC) dupTentativeAddrDetected(addr tcpip.Address) *tcpip.Error { - n.mu.Lock() - defer n.mu.Unlock() - - ref, ok := n.mu.endpoints[NetworkEndpointID{addr}] - if !ok { - return tcpip.ErrBadAddress - } - - if ref.getKind() != permanentTentative { - return tcpip.ErrInvalidEndpointState - } - - // If the address is a SLAAC address, do not invalidate its SLAAC prefix as a - // new address will be generated for it. - if err := n.removePermanentIPv6EndpointLocked(ref, false /* allowSLAACInvalidation */); err != nil { - return err - } - - prefix := ref.addrWithPrefix().Subnet() - - switch ref.configType { - case slaac: - n.mu.ndp.regenerateSLAACAddr(prefix) - case slaacTemp: - // Do not reset the generation attempts counter for the prefix as the - // temporary address is being regenerated in response to a DAD conflict. - n.mu.ndp.regenerateTempSLAACAddr(prefix, false /* resetGenAttempts */) - } - - return nil -} - -// setNDPConfigs sets the NDP configurations for n. -// -// Note, if c contains invalid NDP configuration values, it will be fixed to -// use default values for the erroneous values. -func (n *NIC) setNDPConfigs(c NDPConfigurations) { - c.validate() - - n.mu.Lock() - n.mu.ndp.configs = c - n.mu.Unlock() -} - -// NUDConfigs gets the NUD configurations for n. -func (n *NIC) NUDConfigs() (NUDConfigurations, *tcpip.Error) { +// nudConfigs gets the NUD configurations for n. +func (n *NIC) nudConfigs() (NUDConfigurations, *tcpip.Error) { if n.neigh == nil { return NUDConfigurations{}, tcpip.ErrNotSupported } @@ -1598,49 +853,6 @@ func (n *NIC) setNUDConfigs(c NUDConfigurations) *tcpip.Error { return nil } -// handleNDPRA handles an NDP Router Advertisement message that arrived on n. -func (n *NIC) handleNDPRA(ip tcpip.Address, ra header.NDPRouterAdvert) { - n.mu.Lock() - defer n.mu.Unlock() - - n.mu.ndp.handleRA(ip, ra) -} - -type networkEndpointKind int32 - -const ( - // A permanentTentative endpoint is a permanent address that is not yet - // considered to be fully bound to an interface in the traditional - // sense. That is, the address is associated with a NIC, but packets - // destined to the address MUST NOT be accepted and MUST be silently - // dropped, and the address MUST NOT be used as a source address for - // outgoing packets. For IPv6, addresses will be of this kind until - // NDP's Duplicate Address Detection has resolved, or be deleted if - // the process results in detecting a duplicate address. - permanentTentative networkEndpointKind = iota - - // A permanent endpoint is created by adding a permanent address (vs. a - // temporary one) to the NIC. Its reference count is biased by 1 to avoid - // removal when no route holds a reference to it. It is removed by explicitly - // removing the permanent address from the NIC. - permanent - - // An expired permanent endpoint is a permanent endpoint that had its address - // removed from the NIC, and it is waiting to be removed once no more routes - // hold a reference to it. This is achieved by decreasing its reference count - // by 1. If its address is re-added before the endpoint is removed, its type - // changes back to permanent and its reference count increases by 1 again. - permanentExpired - - // A temporary endpoint is created for spoofing outgoing packets, or when in - // promiscuous mode and accepting incoming packets that don't match any - // permanent endpoint. Its reference count is not biased by 1 and the - // endpoint is removed immediately when no more route holds a reference to - // it. A temporary endpoint can be promoted to permanent if its address - // is added permanently. - temporary -) - func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error { n.mu.Lock() defer n.mu.Unlock() @@ -1671,153 +883,12 @@ func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep } } -type networkEndpointConfigType int32 - -const ( - // A statically configured endpoint is an address that was added by - // some user-specified action (adding an explicit address, joining a - // multicast group). - static networkEndpointConfigType = iota - - // A SLAAC configured endpoint is an IPv6 endpoint that was added by - // SLAAC as per RFC 4862 section 5.5.3. - slaac - - // A temporary SLAAC configured endpoint is an IPv6 endpoint that was added by - // SLAAC as per RFC 4941. Temporary SLAAC addresses are short-lived and are - // not expected to be valid (or preferred) forever; hence the term temporary. - slaacTemp -) - -type referencedNetworkEndpoint struct { - ep NetworkEndpoint - addr tcpip.AddressWithPrefix - nic *NIC - protocol tcpip.NetworkProtocolNumber - - // linkCache is set if link address resolution is enabled for this - // protocol. Set to nil otherwise. - linkCache LinkAddressCache - - // linkRes is set if link address resolution is enabled for this protocol. - // Set to nil otherwise. - linkRes LinkAddressResolver - - // refs is counting references held for this endpoint. When refs hits zero it - // triggers the automatic removal of the endpoint from the NIC. - refs int32 - - // networkEndpointKind must only be accessed using {get,set}Kind(). - kind networkEndpointKind - - // configType is the method that was used to configure this endpoint. - // This must never change except during endpoint creation and promotion to - // permanent. - configType networkEndpointConfigType - - // deprecated indicates whether or not the endpoint should be considered - // deprecated. That is, when deprecated is true, other endpoints that are not - // deprecated should be preferred. - deprecated bool -} - -func (r *referencedNetworkEndpoint) address() tcpip.Address { - return r.addr.Address -} - -func (r *referencedNetworkEndpoint) addrWithPrefix() tcpip.AddressWithPrefix { - return r.addr -} - -func (r *referencedNetworkEndpoint) getKind() networkEndpointKind { - return networkEndpointKind(atomic.LoadInt32((*int32)(&r.kind))) -} - -func (r *referencedNetworkEndpoint) setKind(kind networkEndpointKind) { - atomic.StoreInt32((*int32)(&r.kind), int32(kind)) -} - // isValidForOutgoing returns true if the endpoint can be used to send out a // packet. It requires the endpoint to not be marked expired (i.e., its address) // has been removed) unless the NIC is in spoofing mode, or temporary. -func (r *referencedNetworkEndpoint) isValidForOutgoing() bool { - r.nic.mu.RLock() - defer r.nic.mu.RUnlock() - - return r.isValidForOutgoingRLocked() -} - -// isValidForOutgoingRLocked is the same as isValidForOutgoing but requires -// r.nic.mu to be read locked. -func (r *referencedNetworkEndpoint) isValidForOutgoingRLocked() bool { - if !r.nic.mu.enabled { - return false - } - - return r.isAssignedRLocked(r.nic.mu.spoofing) -} - -// isAssignedRLocked returns true if r is considered to be assigned to the NIC. -// -// r.nic.mu must be read locked. -func (r *referencedNetworkEndpoint) isAssignedRLocked(spoofingOrPromiscuous bool) bool { - switch r.getKind() { - case permanentTentative: - return false - case permanentExpired: - return spoofingOrPromiscuous - default: - return true - } -} - -// expireLocked decrements the reference count and marks the permanent endpoint -// as expired. -func (r *referencedNetworkEndpoint) expireLocked() { - r.setKind(permanentExpired) - r.decRefLocked() -} - -// decRef decrements the ref count and cleans up the endpoint once it reaches -// zero. -func (r *referencedNetworkEndpoint) decRef() { - if atomic.AddInt32(&r.refs, -1) == 0 { - r.nic.removeEndpoint(r) - } -} - -// decRefLocked is the same as decRef but assumes that the NIC.mu mutex is -// locked. -func (r *referencedNetworkEndpoint) decRefLocked() { - if atomic.AddInt32(&r.refs, -1) == 0 { - r.nic.removeEndpointLocked(r) - } -} - -// incRef increments the ref count. It must only be called when the caller is -// known to be holding a reference to the endpoint, otherwise tryIncRef should -// be used. -func (r *referencedNetworkEndpoint) incRef() { - atomic.AddInt32(&r.refs, 1) -} - -// tryIncRef attempts to increment the ref count from n to n+1, but only if n is -// not zero. That is, it will increment the count if the endpoint is still -// alive, and do nothing if it has already been clean up. -func (r *referencedNetworkEndpoint) tryIncRef() bool { - for { - v := atomic.LoadInt32(&r.refs) - if v == 0 { - return false - } - - if atomic.CompareAndSwapInt32(&r.refs, v, v+1) { - return true - } - } -} - -// stack returns the Stack instance that owns the underlying endpoint. -func (r *referencedNetworkEndpoint) stack() *Stack { - return r.nic.stack +func (n *NIC) isValidForOutgoing(ep AssignableAddressEndpoint) bool { + n.mu.RLock() + spoofing := n.mu.spoofing + n.mu.RUnlock() + return n.Enabled() && ep.IsAssigned(spoofing) } diff --git a/pkg/tcpip/stack/nic_test.go b/pkg/tcpip/stack/nic_test.go index dd6474297..97a96af62 100644 --- a/pkg/tcpip/stack/nic_test.go +++ b/pkg/tcpip/stack/nic_test.go @@ -15,96 +15,39 @@ package stack import ( - "math" "testing" - "time" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" ) -var _ LinkEndpoint = (*testLinkEndpoint)(nil) +var _ AddressableEndpoint = (*testIPv6Endpoint)(nil) +var _ NetworkEndpoint = (*testIPv6Endpoint)(nil) +var _ NDPEndpoint = (*testIPv6Endpoint)(nil) -// A LinkEndpoint that throws away outgoing packets. +// An IPv6 NetworkEndpoint that throws away outgoing packets. // -// We use this instead of the channel endpoint as the channel package depends on +// We use this instead of ipv6.endpoint because the ipv6 package depends on // the stack package which this test lives in, causing a cyclic dependency. -type testLinkEndpoint struct { - dispatcher NetworkDispatcher -} - -// Attach implements LinkEndpoint.Attach. -func (e *testLinkEndpoint) Attach(dispatcher NetworkDispatcher) { - e.dispatcher = dispatcher -} - -// IsAttached implements LinkEndpoint.IsAttached. -func (e *testLinkEndpoint) IsAttached() bool { - return e.dispatcher != nil -} - -// MTU implements LinkEndpoint.MTU. -func (*testLinkEndpoint) MTU() uint32 { - return math.MaxUint16 -} - -// Capabilities implements LinkEndpoint.Capabilities. -func (*testLinkEndpoint) Capabilities() LinkEndpointCapabilities { - return CapabilityResolutionRequired -} +type testIPv6Endpoint struct { + AddressableEndpointState -// MaxHeaderLength implements LinkEndpoint.MaxHeaderLength. -func (*testLinkEndpoint) MaxHeaderLength() uint16 { - return 0 -} + nic NetworkInterface + protocol *testIPv6Protocol -// LinkAddress returns the link address of this endpoint. -func (*testLinkEndpoint) LinkAddress() tcpip.LinkAddress { - return "" + invalidatedRtr tcpip.Address } -// Wait implements LinkEndpoint.Wait. -func (*testLinkEndpoint) Wait() {} - -// WritePacket implements LinkEndpoint.WritePacket. -func (e *testLinkEndpoint) WritePacket(*Route, *GSO, tcpip.NetworkProtocolNumber, *PacketBuffer) *tcpip.Error { +func (*testIPv6Endpoint) Enable() *tcpip.Error { return nil } -// WritePackets implements LinkEndpoint.WritePackets. -func (e *testLinkEndpoint) WritePackets(*Route, *GSO, PacketBufferList, tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { - // Our tests don't use this so we don't support it. - return 0, tcpip.ErrNotSupported -} - -// WriteRawPacket implements LinkEndpoint.WriteRawPacket. -func (e *testLinkEndpoint) WriteRawPacket(buffer.VectorisedView) *tcpip.Error { - // Our tests don't use this so we don't support it. - return tcpip.ErrNotSupported -} - -// ARPHardwareType implements stack.LinkEndpoint.ARPHardwareType. -func (*testLinkEndpoint) ARPHardwareType() header.ARPHardwareType { - panic("not implemented") -} - -// AddHeader implements stack.LinkEndpoint.AddHeader. -func (e *testLinkEndpoint) AddHeader(local, remote tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { - panic("not implemented") +func (*testIPv6Endpoint) Enabled() bool { + return true } -var _ NetworkEndpoint = (*testIPv6Endpoint)(nil) - -// An IPv6 NetworkEndpoint that throws away outgoing packets. -// -// We use this instead of ipv6.endpoint because the ipv6 package depends on -// the stack package which this test lives in, causing a cyclic dependency. -type testIPv6Endpoint struct { - nicID tcpip.NICID - linkEP LinkEndpoint - protocol *testIPv6Protocol -} +func (*testIPv6Endpoint) Disable() {} // DefaultTTL implements NetworkEndpoint.DefaultTTL. func (*testIPv6Endpoint) DefaultTTL() uint8 { @@ -113,17 +56,12 @@ func (*testIPv6Endpoint) DefaultTTL() uint8 { // MTU implements NetworkEndpoint.MTU. func (e *testIPv6Endpoint) MTU() uint32 { - return e.linkEP.MTU() - header.IPv6MinimumSize -} - -// Capabilities implements NetworkEndpoint.Capabilities. -func (e *testIPv6Endpoint) Capabilities() LinkEndpointCapabilities { - return e.linkEP.Capabilities() + return e.nic.MTU() - header.IPv6MinimumSize } // MaxHeaderLength implements NetworkEndpoint.MaxHeaderLength. func (e *testIPv6Endpoint) MaxHeaderLength() uint16 { - return e.linkEP.MaxHeaderLength() + header.IPv6MinimumSize + return e.nic.MaxHeaderLength() + header.IPv6MinimumSize } // WritePacket implements NetworkEndpoint.WritePacket. @@ -144,23 +82,24 @@ func (*testIPv6Endpoint) WriteHeaderIncludedPacket(*Route, *PacketBuffer) *tcpip return tcpip.ErrNotSupported } -// NICID implements NetworkEndpoint.NICID. -func (e *testIPv6Endpoint) NICID() tcpip.NICID { - return e.nicID -} - // HandlePacket implements NetworkEndpoint.HandlePacket. func (*testIPv6Endpoint) HandlePacket(*Route, *PacketBuffer) { } // Close implements NetworkEndpoint.Close. -func (*testIPv6Endpoint) Close() {} +func (e *testIPv6Endpoint) Close() { + e.AddressableEndpointState.Cleanup() +} // NetworkProtocolNumber implements NetworkEndpoint.NetworkProtocolNumber. func (*testIPv6Endpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return header.IPv6ProtocolNumber } +func (e *testIPv6Endpoint) InvalidateDefaultRouter(rtr tcpip.Address) { + e.invalidatedRtr = rtr +} + var _ NetworkProtocol = (*testIPv6Protocol)(nil) // An IPv6 NetworkProtocol that supports the bare minimum to make a stack @@ -192,12 +131,13 @@ func (*testIPv6Protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) } // NewEndpoint implements NetworkProtocol.NewEndpoint. -func (p *testIPv6Protocol) NewEndpoint(nicID tcpip.NICID, _ LinkAddressCache, _ NUDHandler, _ TransportDispatcher, linkEP LinkEndpoint, _ *Stack) NetworkEndpoint { - return &testIPv6Endpoint{ - nicID: nicID, - linkEP: linkEP, +func (p *testIPv6Protocol) NewEndpoint(nic NetworkInterface, _ LinkAddressCache, _ NUDHandler, _ TransportDispatcher) NetworkEndpoint { + e := &testIPv6Endpoint{ + nic: nic, protocol: p, } + e.AddressableEndpointState.Init(e) + return e } // SetOption implements NetworkProtocol.SetOption. @@ -241,38 +181,6 @@ func (*testIPv6Protocol) ResolveStaticAddress(addr tcpip.Address) (tcpip.LinkAdd return "", false } -// Test the race condition where a NIC is removed and an RS timer fires at the -// same time. -func TestRemoveNICWhileHandlingRSTimer(t *testing.T) { - const ( - nicID = 1 - - maxRtrSolicitations = 5 - ) - - e := testLinkEndpoint{} - s := New(Options{ - NetworkProtocols: []NetworkProtocol{&testIPv6Protocol{}}, - NDPConfigs: NDPConfigurations{ - MaxRtrSolicitations: maxRtrSolicitations, - RtrSolicitationInterval: minimumRtrSolicitationInterval, - }, - }) - - if err := s.CreateNIC(nicID, &e); err != nil { - t.Fatalf("s.CreateNIC(%d, _) = %s", nicID, err) - } - - s.mu.Lock() - // Wait for the router solicitation timer to fire and block trying to obtain - // the stack lock when doing link address resolution. - time.Sleep(minimumRtrSolicitationInterval * 2) - if err := s.removeNICLocked(nicID); err != nil { - t.Fatalf("s.removeNICLocked(%d) = %s", nicID, err) - } - s.mu.Unlock() -} - func TestDisabledRxStatsWhenNICDisabled(t *testing.T) { // When the NIC is disabled, the only field that matters is the stats field. // This test is limited to stats counter checks. diff --git a/pkg/tcpip/stack/nud_test.go b/pkg/tcpip/stack/nud_test.go index 2b97e5972..8cffb9fc6 100644 --- a/pkg/tcpip/stack/nud_test.go +++ b/pkg/tcpip/stack/nud_test.go @@ -60,7 +60,7 @@ func TestSetNUDConfigurationFailsForBadNICID(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The networking // stack will only allocate neighbor caches if a protocol providing link // address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, UseNeighborCache: true, }) @@ -137,7 +137,7 @@ func TestDefaultNUDConfigurations(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The networking // stack will only allocate neighbor caches if a protocol providing link // address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: stack.DefaultNUDConfigurations(), UseNeighborCache: true, }) @@ -192,7 +192,7 @@ func TestNUDConfigurationsBaseReachableTime(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -249,7 +249,7 @@ func TestNUDConfigurationsMinRandomFactor(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -329,7 +329,7 @@ func TestNUDConfigurationsMaxRandomFactor(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -391,7 +391,7 @@ func TestNUDConfigurationsRetransmitTimer(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -443,7 +443,7 @@ func TestNUDConfigurationsDelayFirstProbeTime(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -495,7 +495,7 @@ func TestNUDConfigurationsMaxMulticastProbes(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -547,7 +547,7 @@ func TestNUDConfigurationsMaxUnicastProbes(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) @@ -599,7 +599,7 @@ func TestNUDConfigurationsUnreachableTime(t *testing.T) { // A neighbor cache is required to store NUDConfigurations. The // networking stack will only allocate neighbor caches if a protocol // providing link address resolution is specified (e.g. ARP or IPv6). - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, NUDConfigs: c, UseNeighborCache: true, }) diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go index 1932aaeb7..105583c49 100644 --- a/pkg/tcpip/stack/packet_buffer.go +++ b/pkg/tcpip/stack/packet_buffer.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" ) type headerType int @@ -80,11 +81,17 @@ type PacketBuffer struct { // data are held in the same underlying buffer storage. header buffer.Prependable - // NetworkProtocolNumber is only valid when NetworkHeader is set. + // NetworkProtocolNumber is only valid when NetworkHeader().View().IsEmpty() + // returns false. // TODO(gvisor.dev/issue/3574): Remove the separately passed protocol // numbers in registration APIs that take a PacketBuffer. NetworkProtocolNumber tcpip.NetworkProtocolNumber + // TransportProtocol is only valid if it is non zero. + // TODO(gvisor.dev/issue/3810): This and the network protocol number should + // be moved into the headerinfo. This should resolve the validity issue. + TransportProtocolNumber tcpip.TransportProtocolNumber + // Hash is the transport layer hash of this packet. A value of zero // indicates no valid hash has been set. Hash uint32 @@ -234,20 +241,35 @@ func (pk *PacketBuffer) consume(typ headerType, size int) (v buffer.View, consum // underlying packet payload. func (pk *PacketBuffer) Clone() *PacketBuffer { newPk := &PacketBuffer{ - PacketBufferEntry: pk.PacketBufferEntry, - Data: pk.Data.Clone(nil), - headers: pk.headers, - header: pk.header, - Hash: pk.Hash, - Owner: pk.Owner, - EgressRoute: pk.EgressRoute, - GSOOptions: pk.GSOOptions, - NetworkProtocolNumber: pk.NetworkProtocolNumber, - NatDone: pk.NatDone, + PacketBufferEntry: pk.PacketBufferEntry, + Data: pk.Data.Clone(nil), + headers: pk.headers, + header: pk.header, + Hash: pk.Hash, + Owner: pk.Owner, + EgressRoute: pk.EgressRoute, + GSOOptions: pk.GSOOptions, + NetworkProtocolNumber: pk.NetworkProtocolNumber, + NatDone: pk.NatDone, + TransportProtocolNumber: pk.TransportProtocolNumber, } return newPk } +// Network returns the network header as a header.Network. +// +// Network should only be called when NetworkHeader has been set. +func (pk *PacketBuffer) Network() header.Network { + switch netProto := pk.NetworkProtocolNumber; netProto { + case header.IPv4ProtocolNumber: + return header.IPv4(pk.NetworkHeader().View()) + case header.IPv6ProtocolNumber: + return header.IPv6(pk.NetworkHeader().View()) + default: + panic(fmt.Sprintf("unknown network protocol number %d", netProto)) + } +} + // headerInfo stores metadata about a header in a packet. type headerInfo struct { // buf is the memorized slice for both prepended and consumed header. diff --git a/pkg/tcpip/stack/forwarder.go b/pkg/tcpip/stack/pending_packets.go index 3eff141e6..f838eda8d 100644 --- a/pkg/tcpip/stack/forwarder.go +++ b/pkg/tcpip/stack/pending_packets.go @@ -29,60 +29,60 @@ const ( ) type pendingPacket struct { - nic *NIC route *Route proto tcpip.NetworkProtocolNumber pkt *PacketBuffer } -type forwardQueue struct { +// packetsPendingLinkResolution is a queue of packets pending link resolution. +// +// Once link resolution completes successfully, the packets will be written. +type packetsPendingLinkResolution struct { sync.Mutex // The packets to send once the resolver completes. - packets map[<-chan struct{}][]*pendingPacket + packets map[<-chan struct{}][]pendingPacket // FIFO of channels used to cancel the oldest goroutine waiting for // link-address resolution. cancelChans []chan struct{} } -func newForwardQueue() *forwardQueue { - return &forwardQueue{packets: make(map[<-chan struct{}][]*pendingPacket)} +func (f *packetsPendingLinkResolution) init() { + f.Lock() + defer f.Unlock() + f.packets = make(map[<-chan struct{}][]pendingPacket) } -func (f *forwardQueue) enqueue(ch <-chan struct{}, n *NIC, r *Route, protocol tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { - shouldWait := false - +func (f *packetsPendingLinkResolution) enqueue(ch <-chan struct{}, r *Route, proto tcpip.NetworkProtocolNumber, pkt *PacketBuffer) { f.Lock() + defer f.Unlock() + packets, ok := f.packets[ch] - if !ok { - shouldWait = true - } - for len(packets) == maxPendingPacketsPerResolution { + if len(packets) == maxPendingPacketsPerResolution { p := packets[0] + packets[0] = pendingPacket{} packets = packets[1:] - p.nic.stack.stats.IP.OutgoingPacketErrors.Increment() + p.route.Stats().IP.OutgoingPacketErrors.Increment() p.route.Release() } + if l := len(packets); l >= maxPendingPacketsPerResolution { panic(fmt.Sprintf("max pending packets for resolution reached; got %d packets, max = %d", l, maxPendingPacketsPerResolution)) } - f.packets[ch] = append(packets, &pendingPacket{ - nic: n, + + f.packets[ch] = append(packets, pendingPacket{ route: r, - proto: protocol, + proto: proto, pkt: pkt, }) - f.Unlock() - if !shouldWait { + if ok { return } // Wait for the link-address resolution to complete. - // Start a goroutine with a forwarding-cancel channel so that we can - // limit the maximum number of goroutines running concurrently. - cancel := f.newCancelChannel() + cancel := f.newCancelChannelLocked() go func() { cancelled := false select { @@ -92,17 +92,21 @@ func (f *forwardQueue) enqueue(ch <-chan struct{}, n *NIC, r *Route, protocol tc } f.Lock() - packets := f.packets[ch] + packets, ok := f.packets[ch] delete(f.packets, ch) f.Unlock() + if !ok { + panic(fmt.Sprintf("link-resolution goroutine woke up but no entry exists in the queue of packets")) + } + for _, p := range packets { if cancelled { - p.nic.stack.stats.IP.OutgoingPacketErrors.Increment() + p.route.Stats().IP.OutgoingPacketErrors.Increment() } else if _, err := p.route.Resolve(nil); err != nil { - p.nic.stack.stats.IP.OutgoingPacketErrors.Increment() + p.route.Stats().IP.OutgoingPacketErrors.Increment() } else { - p.nic.forwardPacket(p.route, p.proto, p.pkt) + p.route.nic.writePacket(p.route, nil /* gso */, p.proto, p.pkt) } p.route.Release() } @@ -112,12 +116,10 @@ func (f *forwardQueue) enqueue(ch <-chan struct{}, n *NIC, r *Route, protocol tc // newCancelChannel creates a channel that can cancel a pending forwarding // activity. The oldest channel is closed if the number of open channels would // exceed maxPendingResolutions. -func (f *forwardQueue) newCancelChannel() chan struct{} { - f.Lock() - defer f.Unlock() - +func (f *packetsPendingLinkResolution) newCancelChannelLocked() chan struct{} { if len(f.cancelChans) == maxPendingResolutions { ch := f.cancelChans[0] + f.cancelChans[0] = nil f.cancelChans = f.cancelChans[1:] close(ch) } diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 4fa86a3ac..defb9129b 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -15,6 +15,8 @@ package stack import ( + "fmt" + "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -125,6 +127,26 @@ type PacketEndpoint interface { HandlePacket(nicID tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, pkt *PacketBuffer) } +// UnknownDestinationPacketDisposition enumerates the possible return vaues from +// HandleUnknownDestinationPacket(). +type UnknownDestinationPacketDisposition int + +const ( + // UnknownDestinationPacketMalformed denotes that the packet was malformed + // and no further processing should be attempted other than updating + // statistics. + UnknownDestinationPacketMalformed UnknownDestinationPacketDisposition = iota + + // UnknownDestinationPacketUnhandled tells the caller that the packet was + // well formed but that the issue was not handled and the stack should take + // the default action. + UnknownDestinationPacketUnhandled + + // UnknownDestinationPacketHandled tells the caller that it should do + // no further processing. + UnknownDestinationPacketHandled +) + // TransportProtocol is the interface that needs to be implemented by transport // protocols (e.g., tcp, udp) that want to be part of the networking stack. type TransportProtocol interface { @@ -132,10 +154,10 @@ type TransportProtocol interface { Number() tcpip.TransportProtocolNumber // NewEndpoint creates a new endpoint of the transport protocol. - NewEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, waitQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) + NewEndpoint(netProto tcpip.NetworkProtocolNumber, waitQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) // NewRawEndpoint creates a new raw endpoint of the transport protocol. - NewRawEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, waitQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) + NewRawEndpoint(netProto tcpip.NetworkProtocolNumber, waitQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) // MinimumPacketSize returns the minimum valid packet size of this // transport protocol. The stack automatically drops any packets smaller @@ -147,14 +169,12 @@ type TransportProtocol interface { ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) // HandleUnknownDestinationPacket handles packets targeted at this - // protocol but that don't match any existing endpoint. For example, - // it is targeted at a port that have no listeners. - // - // The return value indicates whether the packet was well-formed (for - // stats purposes only). + // protocol that don't match any existing endpoint. For example, + // it is targeted at a port that has no listeners. // - // HandleUnknownDestinationPacket takes ownership of pkt. - HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) bool + // HandleUnknownDestinationPacket takes ownership of pkt if it handles + // the issue. + HandleUnknownDestinationPacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) UnknownDestinationPacketDisposition // SetOption allows enabling/disabling protocol specific features. // SetOption returns an error if the option is not supported or the @@ -179,6 +199,25 @@ type TransportProtocol interface { Parse(pkt *PacketBuffer) (ok bool) } +// TransportPacketDisposition is the result from attempting to deliver a packet +// to the transport layer. +type TransportPacketDisposition int + +const ( + // TransportPacketHandled indicates that a transport packet was handled by the + // transport layer and callers need not take any further action. + TransportPacketHandled TransportPacketDisposition = iota + + // TransportPacketProtocolUnreachable indicates that the transport + // protocol requested in the packet is not supported. + TransportPacketProtocolUnreachable + + // TransportPacketDestinationPortUnreachable indicates that there weren't any + // listeners interested in the packet and the transport protocol has no means + // to notify the sender. + TransportPacketDestinationPortUnreachable +) + // TransportDispatcher contains the methods used by the network stack to deliver // packets to the appropriate transport endpoint after it has been handled by // the network layer. @@ -189,7 +228,7 @@ type TransportDispatcher interface { // pkt.NetworkHeader must be set before calling DeliverTransportPacket. // // DeliverTransportPacket takes ownership of pkt. - DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) + DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolNumber, pkt *PacketBuffer) TransportPacketDisposition // DeliverTransportControlPacket delivers control packets to the // appropriate transport protocol endpoint. @@ -226,9 +265,252 @@ type NetworkHeaderParams struct { TOS uint8 } +// GroupAddressableEndpoint is an endpoint that supports group addressing. +// +// An endpoint is considered to support group addressing when one or more +// endpoints may associate themselves with the same identifier (group address). +type GroupAddressableEndpoint interface { + // JoinGroup joins the spcified group. + // + // Returns true if the group was newly joined. + JoinGroup(group tcpip.Address) (bool, *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) + + // IsInGroup returns true if the endpoint is a member of the specified group. + IsInGroup(group tcpip.Address) bool +} + +// PrimaryEndpointBehavior is an enumeration of an AddressEndpoint's primary +// behavior. +type PrimaryEndpointBehavior int + +const ( + // CanBePrimaryEndpoint indicates the endpoint can be used as a primary + // endpoint for new connections with no local address. This is the + // default when calling NIC.AddAddress. + CanBePrimaryEndpoint PrimaryEndpointBehavior = iota + + // FirstPrimaryEndpoint indicates the endpoint should be the first + // primary endpoint considered. If there are multiple endpoints with + // this behavior, they are ordered by recency. + FirstPrimaryEndpoint + + // NeverPrimaryEndpoint indicates the endpoint should never be a + // primary endpoint. + NeverPrimaryEndpoint +) + +// AddressConfigType is the method used to add an address. +type AddressConfigType int + +const ( + // AddressConfigStatic is a statically configured address endpoint that was + // added by some user-specified action (adding an explicit address, joining a + // multicast group). + AddressConfigStatic AddressConfigType = iota + + // AddressConfigSlaac is an address endpoint added by SLAAC, as per RFC 4862 + // section 5.5.3. + AddressConfigSlaac + + // AddressConfigSlaacTemp is a temporary address endpoint added by SLAAC as + // per RFC 4941. Temporary SLAAC addresses are short-lived and are not + // to be valid (or preferred) forever; hence the term temporary. + AddressConfigSlaacTemp +) + +// AssignableAddressEndpoint is a reference counted address endpoint that may be +// assigned to a NetworkEndpoint. +type AssignableAddressEndpoint interface { + // AddressWithPrefix returns the endpoint's address. + AddressWithPrefix() tcpip.AddressWithPrefix + + // IsAssigned returns whether or not the endpoint is considered bound + // to its NetworkEndpoint. + IsAssigned(allowExpired bool) bool + + // IncRef increments this endpoint's reference count. + // + // Returns true if it was successfully incremented. If it returns false, then + // the endpoint is considered expired and should no longer be used. + IncRef() bool + + // DecRef decrements this endpoint's reference count. + DecRef() +} + +// AddressEndpoint is an endpoint representing an address assigned to an +// AddressableEndpoint. +type AddressEndpoint interface { + AssignableAddressEndpoint + + // GetKind returns the address kind for this endpoint. + GetKind() AddressKind + + // SetKind sets the address kind for this endpoint. + SetKind(AddressKind) + + // ConfigType returns the method used to add the address. + ConfigType() AddressConfigType + + // Deprecated returns whether or not this endpoint is deprecated. + Deprecated() bool + + // SetDeprecated sets this endpoint's deprecated status. + SetDeprecated(bool) +} + +// AddressKind is the kind of of an address. +// +// See the values of AddressKind for more details. +type AddressKind int + +const ( + // PermanentTentative is a permanent address endpoint that is not yet + // considered to be fully bound to an interface in the traditional + // sense. That is, the address is associated with a NIC, but packets + // destined to the address MUST NOT be accepted and MUST be silently + // dropped, and the address MUST NOT be used as a source address for + // outgoing packets. For IPv6, addresses are of this kind until NDP's + // Duplicate Address Detection (DAD) resolves. If DAD fails, the address + // is removed. + PermanentTentative AddressKind = iota + + // Permanent is a permanent endpoint (vs. a temporary one) assigned to the + // NIC. Its reference count is biased by 1 to avoid removal when no route + // holds a reference to it. It is removed by explicitly removing the address + // from the NIC. + Permanent + + // PermanentExpired is a permanent endpoint that had its address removed from + // the NIC, and it is waiting to be removed once no references to it are held. + // + // If the address is re-added before the endpoint is removed, its type + // changes back to Permanent. + PermanentExpired + + // Temporary is an endpoint, created on a one-off basis to temporarily + // consider the NIC bound an an address that it is not explictiy bound to + // (such as a permanent address). Its reference count must not be biased by 1 + // so that the address is removed immediately when references to it are no + // longer held. + // + // A temporary endpoint may be promoted to permanent if the address is added + // permanently. + Temporary +) + +// IsPermanent returns true if the AddressKind represents a permanent address. +func (k AddressKind) IsPermanent() bool { + switch k { + case Permanent, PermanentTentative: + return true + case Temporary, PermanentExpired: + return false + default: + panic(fmt.Sprintf("unrecognized address kind = %d", k)) + } +} + +// AddressableEndpoint is an endpoint that supports addressing. +// +// An endpoint is considered to support addressing when the endpoint may +// associate itself with an identifier (address). +type AddressableEndpoint interface { + // AddAndAcquirePermanentAddress adds the passed permanent address. + // + // Returns tcpip.ErrDuplicateAddress if the address exists. + // + // Acquires and returns the AddressEndpoint for the added address. + AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb PrimaryEndpointBehavior, configType AddressConfigType, deprecated bool) (AddressEndpoint, *tcpip.Error) + + // RemovePermanentAddress removes the passed address if it is a permanent + // address. + // + // Returns tcpip.ErrBadLocalAddress if the endpoint does not have the passed + // permanent address. + RemovePermanentAddress(addr tcpip.Address) *tcpip.Error + + // MainAddress returns the endpoint's primary permanent address. + MainAddress() tcpip.AddressWithPrefix + + // AcquireAssignedAddress returns an address endpoint for the passed address + // that is considered bound to the endpoint, optionally creating a temporary + // endpoint if requested and no existing address exists. + // + // The returned endpoint's reference count is incremented. + // + // Returns nil if the specified address is not local to this endpoint. + AcquireAssignedAddress(localAddr tcpip.Address, allowTemp bool, tempPEB PrimaryEndpointBehavior) AddressEndpoint + + // AcquireOutgoingPrimaryAddress returns a primary address that may be used as + // a source address when sending packets to the passed remote address. + // + // If allowExpired is true, expired addresses may be returned. + // + // The returned endpoint's reference count is incremented. + // + // Returns nil if a primary address is not available. + AcquireOutgoingPrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) AddressEndpoint + + // PrimaryAddresses returns the primary addresses. + PrimaryAddresses() []tcpip.AddressWithPrefix + + // PermanentAddresses returns all the permanent addresses. + PermanentAddresses() []tcpip.AddressWithPrefix +} + +// NDPEndpoint is a network endpoint that supports NDP. +type NDPEndpoint interface { + NetworkEndpoint + + // InvalidateDefaultRouter invalidates a default router discovered through + // NDP. + InvalidateDefaultRouter(tcpip.Address) +} + +// NetworkInterface is a network interface. +type NetworkInterface interface { + NetworkLinkEndpoint + + // ID returns the interface's ID. + ID() tcpip.NICID + + // IsLoopback returns true if the interface is a loopback interface. + IsLoopback() bool + + // Name returns the name of the interface. + // + // May return an empty string if the interface is not configured with a name. + Name() string + + // Enabled returns true if the interface is enabled. + Enabled() bool +} + // NetworkEndpoint is the interface that needs to be implemented by endpoints // of network layer protocols (e.g., ipv4, ipv6). type NetworkEndpoint interface { + AddressableEndpoint + + // Enable enables the endpoint. + // + // Must only be called when the stack is in a state that allows the endpoint + // to send and receive packets. + // + // Returns tcpip.ErrNotPermitted if the endpoint cannot be enabled. + Enable() *tcpip.Error + + // Enabled returns true if the endpoint is enabled. + Enabled() bool + + // Disable disables the endpoint. + Disable() + // DefaultTTL is the default time-to-live value (or hop limit, in ipv6) // for this endpoint. DefaultTTL() uint8 @@ -238,10 +520,6 @@ type NetworkEndpoint interface { // minus the network endpoint max header length. MTU() uint32 - // Capabilities returns the set of capabilities supported by the - // underlying link-layer endpoint. - Capabilities() LinkEndpointCapabilities - // MaxHeaderLength returns the maximum size the network (and lower // level layers combined) headers can have. Higher levels use this // information to reserve space in the front of the packets they're @@ -262,9 +540,6 @@ type NetworkEndpoint interface { // header to the given destination address. It takes ownership of pkt. WriteHeaderIncludedPacket(r *Route, pkt *PacketBuffer) *tcpip.Error - // NICID returns the id of the NIC this endpoint belongs to. - NICID() tcpip.NICID - // HandlePacket is called by the link layer when new packets arrive to // this network endpoint. It sets pkt.NetworkHeader. // @@ -279,6 +554,17 @@ type NetworkEndpoint interface { NetworkProtocolNumber() tcpip.NetworkProtocolNumber } +// ForwardingNetworkProtocol is a NetworkProtocol that may forward packets. +type ForwardingNetworkProtocol interface { + NetworkProtocol + + // Forwarding returns the forwarding configuration. + Forwarding() bool + + // SetForwarding sets the forwarding configuration. + SetForwarding(bool) +} + // NetworkProtocol is the interface that needs to be implemented by network // protocols (e.g., ipv4, ipv6) that want to be part of the networking stack. type NetworkProtocol interface { @@ -298,7 +584,7 @@ type NetworkProtocol interface { ParseAddresses(v buffer.View) (src, dst tcpip.Address) // NewEndpoint creates a new endpoint of this protocol. - NewEndpoint(nicID tcpip.NICID, linkAddrCache LinkAddressCache, nud NUDHandler, dispatcher TransportDispatcher, sender LinkEndpoint, st *Stack) NetworkEndpoint + NewEndpoint(nic NetworkInterface, linkAddrCache LinkAddressCache, nud NUDHandler, dispatcher TransportDispatcher) NetworkEndpoint // SetOption allows enabling/disabling protocol specific features. // SetOption returns an error if the option is not supported or the @@ -376,22 +662,15 @@ const ( CapabilitySoftwareGSO ) -// LinkEndpoint is the interface implemented by data link layer protocols (e.g., -// ethernet, loopback, raw) and used by network layer protocols to send packets -// out through the implementer's data link endpoint. When a link header exists, -// it sets each PacketBuffer's LinkHeader field before passing it up the -// stack. -type LinkEndpoint interface { +// NetworkLinkEndpoint is a data-link layer that supports sending network +// layer packets. +type NetworkLinkEndpoint interface { // MTU is the maximum transmission unit for this endpoint. This is // usually dictated by the backing physical network; when such a // physical network doesn't exist, the limit is generally 64k, which // includes the maximum size of an IP packet. MTU() uint32 - // Capabilities returns the set of capabilities supported by the - // endpoint. - Capabilities() LinkEndpointCapabilities - // MaxHeaderLength returns the maximum size the data link (and // lower level layers combined) headers can have. Higher levels use this // information to reserve space in the front of the packets they're @@ -399,7 +678,7 @@ type LinkEndpoint interface { MaxHeaderLength() uint16 // LinkAddress returns the link address (typically a MAC) of the - // link endpoint. + // endpoint. LinkAddress() tcpip.LinkAddress // WritePacket writes a packet with the given protocol through the @@ -419,6 +698,19 @@ type LinkEndpoint interface { // offload is enabled. If it will be used for something else, it may // require to change syscall filters. WritePackets(r *Route, gso *GSO, pkts PacketBufferList, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) +} + +// LinkEndpoint is the interface implemented by data link layer protocols (e.g., +// ethernet, loopback, raw) and used by network layer protocols to send packets +// out through the implementer's data link endpoint. When a link header exists, +// it sets each PacketBuffer's LinkHeader field before passing it up the +// stack. +type LinkEndpoint interface { + NetworkLinkEndpoint + + // Capabilities returns the set of capabilities supported by the + // endpoint. + Capabilities() LinkEndpointCapabilities // WriteRawPacket writes a packet directly to the link. The packet // should already have an ethernet header. It takes ownership of vv. @@ -427,8 +719,8 @@ type LinkEndpoint interface { // Attach attaches the data link layer endpoint to the network-layer // dispatcher of the stack. // - // Attach will be called with a nil dispatcher if the receiver's associated - // NIC is being removed. + // Attach is called with a nil dispatcher when the endpoint's NIC is being + // removed. Attach(dispatcher NetworkDispatcher) // IsAttached returns whether a NetworkDispatcher is attached to the diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 2cbbf0de8..25f80c1f8 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -42,17 +42,27 @@ type Route struct { // NetProto is the network-layer protocol. NetProto tcpip.NetworkProtocolNumber - // ref a reference to the network endpoint through which the route - // starts. - ref *referencedNetworkEndpoint - // Loop controls where WritePacket should send packets. Loop PacketLooping + + // nic is the NIC the route goes through. + nic *NIC + + // addressEndpoint is the local address this route is associated with. + addressEndpoint AssignableAddressEndpoint + + // linkCache is set if link address resolution is enabled for this protocol on + // the route's NIC. + linkCache LinkAddressCache + + // linkRes is set if link address resolution is enabled for this protocol on + // the route's NIC. + linkRes LinkAddressResolver } // makeRoute initializes a new route. It takes ownership of the provided -// reference to a network endpoint. -func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, localLinkAddr tcpip.LinkAddress, ref *referencedNetworkEndpoint, handleLocal, multicastLoop bool) Route { +// AssignableAddressEndpoint. +func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip.Address, nic *NIC, addressEndpoint AssignableAddressEndpoint, handleLocal, multicastLoop bool) Route { loop := PacketOut if handleLocal && localAddr != "" && remoteAddr == localAddr { loop = PacketLoop @@ -62,29 +72,39 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip loop |= PacketLoop } - return Route{ + r := Route{ NetProto: netProto, LocalAddress: localAddr, - LocalLinkAddress: localLinkAddr, + LocalLinkAddress: nic.LinkEndpoint.LinkAddress(), RemoteAddress: remoteAddr, - ref: ref, + addressEndpoint: addressEndpoint, + nic: nic, Loop: loop, } + + if r.nic.LinkEndpoint.Capabilities()&CapabilityResolutionRequired != 0 { + if linkRes, ok := r.nic.stack.linkAddrResolvers[r.NetProto]; ok { + r.linkRes = linkRes + r.linkCache = r.nic.stack + } + } + + return r } // NICID returns the id of the NIC from which this route originates. func (r *Route) NICID() tcpip.NICID { - return r.ref.ep.NICID() + return r.nic.ID() } // MaxHeaderLength forwards the call to the network endpoint's implementation. func (r *Route) MaxHeaderLength() uint16 { - return r.ref.ep.MaxHeaderLength() + return r.nic.getNetworkEndpoint(r.NetProto).MaxHeaderLength() } // Stats returns a mutable copy of current stats. func (r *Route) Stats() tcpip.Stats { - return r.ref.nic.stack.Stats() + return r.nic.stack.Stats() } // PseudoHeaderChecksum forwards the call to the network endpoint's @@ -95,23 +115,17 @@ func (r *Route) PseudoHeaderChecksum(protocol tcpip.TransportProtocolNumber, tot // Capabilities returns the link-layer capabilities of the route. func (r *Route) Capabilities() LinkEndpointCapabilities { - return r.ref.ep.Capabilities() + return r.nic.LinkEndpoint.Capabilities() } // GSOMaxSize returns the maximum GSO packet size. func (r *Route) GSOMaxSize() uint32 { - if gso, ok := r.ref.ep.(GSOEndpoint); ok { + if gso, ok := r.nic.LinkEndpoint.(GSOEndpoint); ok { return gso.GSOMaxSize() } return 0 } -// ResolveWith immediately resolves a route with the specified remote link -// address. -func (r *Route) ResolveWith(addr tcpip.LinkAddress) { - r.RemoteLinkAddress = addr -} - // Resolve attempts to resolve the link address if necessary. Returns ErrWouldBlock in // case address resolution requires blocking, e.g. wait for ARP reply. Waker is // notified when address resolution is complete (success or not). @@ -138,8 +152,8 @@ func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) { nextAddr = r.RemoteAddress } - if r.ref.nic.neigh != nil { - entry, ch, err := r.ref.nic.neigh.entry(nextAddr, r.LocalAddress, r.ref.linkRes, waker) + if neigh := r.nic.neigh; neigh != nil { + entry, ch, err := neigh.entry(nextAddr, r.LocalAddress, r.linkRes, waker) if err != nil { return ch, err } @@ -147,7 +161,7 @@ func (r *Route) Resolve(waker *sleep.Waker) (<-chan struct{}, *tcpip.Error) { return nil, nil } - linkAddr, ch, err := r.ref.linkCache.GetLinkAddress(r.ref.nic.ID(), nextAddr, r.LocalAddress, r.NetProto, waker) + linkAddr, ch, err := r.linkCache.GetLinkAddress(r.nic.ID(), nextAddr, r.LocalAddress, r.NetProto, waker) if err != nil { return ch, err } @@ -162,12 +176,12 @@ func (r *Route) RemoveWaker(waker *sleep.Waker) { nextAddr = r.RemoteAddress } - if r.ref.nic.neigh != nil { - r.ref.nic.neigh.removeWaker(nextAddr, waker) + if neigh := r.nic.neigh; neigh != nil { + neigh.removeWaker(nextAddr, waker) return } - r.ref.linkCache.RemoveWaker(r.ref.nic.ID(), nextAddr, waker) + r.linkCache.RemoveWaker(r.nic.ID(), nextAddr, waker) } // IsResolutionRequired returns true if Resolve() must be called to resolve @@ -175,104 +189,63 @@ func (r *Route) RemoveWaker(waker *sleep.Waker) { // // The NIC r uses must not be locked. func (r *Route) IsResolutionRequired() bool { - if r.ref.nic.neigh != nil { - return r.ref.isValidForOutgoing() && r.ref.linkRes != nil && r.RemoteLinkAddress == "" + if r.nic.neigh != nil { + return r.nic.isValidForOutgoing(r.addressEndpoint) && r.linkRes != nil && r.RemoteLinkAddress == "" } - return r.ref.isValidForOutgoing() && r.ref.linkCache != nil && r.RemoteLinkAddress == "" + return r.nic.isValidForOutgoing(r.addressEndpoint) && r.linkCache != nil && r.RemoteLinkAddress == "" } // WritePacket writes the packet through the given route. func (r *Route) WritePacket(gso *GSO, params NetworkHeaderParams, pkt *PacketBuffer) *tcpip.Error { - if !r.ref.isValidForOutgoing() { + if !r.nic.isValidForOutgoing(r.addressEndpoint) { return tcpip.ErrInvalidEndpointState } - // WritePacket takes ownership of pkt, calculate numBytes first. - numBytes := pkt.Size() - - err := r.ref.ep.WritePacket(r, gso, params, pkt) - if err != nil { - r.Stats().IP.OutgoingPacketErrors.Increment() - } else { - r.ref.nic.stats.Tx.Packets.Increment() - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) - } - return err + return r.nic.getNetworkEndpoint(r.NetProto).WritePacket(r, gso, params, pkt) } // WritePackets writes a list of n packets through the given route and returns // the number of packets written. func (r *Route) WritePackets(gso *GSO, pkts PacketBufferList, params NetworkHeaderParams) (int, *tcpip.Error) { - if !r.ref.isValidForOutgoing() { + if !r.nic.isValidForOutgoing(r.addressEndpoint) { return 0, tcpip.ErrInvalidEndpointState } - // WritePackets takes ownership of pkt, calculate length first. - numPkts := pkts.Len() - - n, err := r.ref.ep.WritePackets(r, gso, pkts, params) - if err != nil { - r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(numPkts - n)) - } - r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n)) - - writtenBytes := 0 - for i, pb := 0, pkts.Front(); i < n && pb != nil; i, pb = i+1, pb.Next() { - writtenBytes += pb.Size() - } - - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(writtenBytes)) - return n, err + return r.nic.getNetworkEndpoint(r.NetProto).WritePackets(r, gso, pkts, params) } // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. func (r *Route) WriteHeaderIncludedPacket(pkt *PacketBuffer) *tcpip.Error { - if !r.ref.isValidForOutgoing() { + if !r.nic.isValidForOutgoing(r.addressEndpoint) { return tcpip.ErrInvalidEndpointState } - // WriteHeaderIncludedPacket takes ownership of pkt, calculate numBytes first. - numBytes := pkt.Data.Size() - - if err := r.ref.ep.WriteHeaderIncludedPacket(r, pkt); err != nil { - r.Stats().IP.OutgoingPacketErrors.Increment() - return err - } - r.ref.nic.stats.Tx.Packets.Increment() - r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(numBytes)) - return nil + return r.nic.getNetworkEndpoint(r.NetProto).WriteHeaderIncludedPacket(r, pkt) } // DefaultTTL returns the default TTL of the underlying network endpoint. func (r *Route) DefaultTTL() uint8 { - return r.ref.ep.DefaultTTL() + return r.nic.getNetworkEndpoint(r.NetProto).DefaultTTL() } // MTU returns the MTU of the underlying network endpoint. func (r *Route) MTU() uint32 { - return r.ref.ep.MTU() -} - -// NetworkProtocolNumber returns the NetworkProtocolNumber of the underlying -// network endpoint. -func (r *Route) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { - return r.ref.ep.NetworkProtocolNumber() + return r.nic.getNetworkEndpoint(r.NetProto).MTU() } // Release frees all resources associated with the route. func (r *Route) Release() { - if r.ref != nil { - r.ref.decRef() - r.ref = nil + if r.addressEndpoint != nil { + r.addressEndpoint.DecRef() + r.addressEndpoint = nil } } -// Clone Clone a route such that the original one can be released and the new -// one will remain valid. +// Clone clones the route. func (r *Route) Clone() Route { - if r.ref != nil { - r.ref.incRef() + if r.addressEndpoint != nil { + _ = r.addressEndpoint.IncRef() } return *r } @@ -296,7 +269,7 @@ func (r *Route) MakeLoopedRoute() Route { // Stack returns the instance of the Stack that owns this route. func (r *Route) Stack() *Stack { - return r.ref.stack() + return r.nic.stack } func (r *Route) isV4Broadcast(addr tcpip.Address) bool { @@ -304,7 +277,7 @@ func (r *Route) isV4Broadcast(addr tcpip.Address) bool { return true } - subnet := r.ref.addrWithPrefix().Subnet() + subnet := r.addressEndpoint.AddressWithPrefix().Subnet() return subnet.IsBroadcast(addr) } @@ -330,7 +303,10 @@ func (r *Route) ReverseRoute(src tcpip.Address, dst tcpip.Address) Route { LocalLinkAddress: r.RemoteLinkAddress, RemoteAddress: src, RemoteLinkAddress: r.LocalLinkAddress, - ref: r.ref, Loop: r.Loop, + addressEndpoint: r.addressEndpoint, + nic: r.nic, + linkCache: r.linkCache, + linkRes: r.linkRes, } } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 68cf77de2..3a07577c8 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -144,10 +144,7 @@ type TCPReceiverState struct { // PendingBufUsed is the number of bytes pending in the receive // queue. - PendingBufUsed seqnum.Size - - // PendingBufSize is the size of the socket receive buffer. - PendingBufSize seqnum.Size + PendingBufUsed int } // TCPSenderState holds a copy of the internal state of the sender for @@ -366,38 +363,6 @@ func (u *uniqueIDGenerator) UniqueID() uint64 { return atomic.AddUint64((*uint64)(u), 1) } -// NICNameFromID is a function that returns a stable name for the specified NIC, -// even if different NIC IDs are used to refer to the same NIC in different -// program runs. It is used when generating opaque interface identifiers (IIDs). -// If the NIC was created with a name, it will be passed to NICNameFromID. -// -// NICNameFromID SHOULD return unique NIC names so unique opaque IIDs are -// generated for the same prefix on differnt NICs. -type NICNameFromID func(tcpip.NICID, string) string - -// OpaqueInterfaceIdentifierOptions holds the options related to the generation -// of opaque interface indentifiers (IIDs) as defined by RFC 7217. -type OpaqueInterfaceIdentifierOptions struct { - // NICNameFromID is a function that returns a stable name for a specified NIC, - // even if the NIC ID changes over time. - // - // Must be specified to generate the opaque IID. - NICNameFromID NICNameFromID - - // SecretKey is a pseudo-random number used as the secret key when generating - // opaque IIDs as defined by RFC 7217. The key SHOULD be at least - // header.OpaqueIIDSecretKeyMinBytes bytes and MUST follow minimum randomness - // requirements for security as outlined by RFC 4086. SecretKey MUST NOT - // change between program runs, unless explicitly changed. - // - // OpaqueInterfaceIdentifierOptions takes ownership of SecretKey. SecretKey - // MUST NOT be modified after Stack is created. - // - // May be nil, but a nil value is highly discouraged to maintain - // some level of randomness between nodes. - SecretKey []byte -} - // Stack is a networking stack, with all supported protocols, NICs, and route // table. type Stack struct { @@ -405,13 +370,6 @@ type Stack struct { networkProtocols map[tcpip.NetworkProtocolNumber]NetworkProtocol linkAddrResolvers map[tcpip.NetworkProtocolNumber]LinkAddressResolver - // forwarding contains the whether packet forwarding is enabled or not for - // different network protocols. - forwarding struct { - sync.RWMutex - protocols map[tcpip.NetworkProtocolNumber]bool - } - // rawFactory creates raw endpoints. If nil, raw endpoints are // disabled. It is set during Stack creation and is immutable. rawFactory RawFactory @@ -464,9 +422,6 @@ type Stack struct { // TODO(gvisor.dev/issue/940): S/R this field. seed uint32 - // ndpConfigs is the default NDP configurations used by interfaces. - ndpConfigs NDPConfigurations - // nudConfigs is the default NUD configurations used by interfaces. nudConfigs NUDConfigurations @@ -474,15 +429,6 @@ type Stack struct { // by the NIC's neighborCache instead of linkAddrCache. useNeighborCache bool - // autoGenIPv6LinkLocal determines whether or not the stack will attempt - // 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 - - // ndpDisp is the NDP event dispatcher that is used to send the netstack - // integrator NDP related events. - ndpDisp NDPDispatcher - // nudDisp is the NUD event dispatcher that is used to send the netstack // integrator NUD related events. nudDisp NUDDispatcher @@ -490,17 +436,9 @@ type Stack struct { // uniqueIDGenerator is a generator of unique identifiers. uniqueIDGenerator UniqueID - // 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 - - // forwarder holds the packets that wait for their link-address resolutions - // to complete, and forwards them when each resolution is done. - forwarder *forwardQueue + // linkResQueue holds packets that are waiting for link resolution to + // complete. + linkResQueue packetsPendingLinkResolution // randomGenerator is an injectable pseudo random generator that can be // used when a random number is required. @@ -520,13 +458,25 @@ type UniqueID interface { UniqueID() uint64 } +// NetworkProtocolFactory instantiates a network protocol. +// +// NetworkProtocolFactory must not attempt to modify the stack, it may only +// query the stack. +type NetworkProtocolFactory func(*Stack) NetworkProtocol + +// TransportProtocolFactory instantiates a transport protocol. +// +// TransportProtocolFactory must not attempt to modify the stack, it may only +// query the stack. +type TransportProtocolFactory func(*Stack) TransportProtocol + // Options contains optional Stack configuration. type Options struct { // NetworkProtocols lists the network protocols to enable. - NetworkProtocols []NetworkProtocol + NetworkProtocols []NetworkProtocolFactory // TransportProtocols lists the transport protocols to enable. - TransportProtocols []TransportProtocol + TransportProtocols []TransportProtocolFactory // Clock is an optional clock source used for timestampping packets. // @@ -544,13 +494,6 @@ type Options struct { // UniqueID is an optional generator of unique identifiers. UniqueID UniqueID - // NDPConfigs is the default NDP configurations used by interfaces. - // - // By default, NDPConfigs will have a zero value for its - // DupAddrDetectTransmits field, implying that DAD will not be performed - // before assigning an address to a NIC. - NDPConfigs NDPConfigurations - // NUDConfigs is the default NUD configurations used by interfaces. NUDConfigs NUDConfigurations @@ -561,24 +504,6 @@ type Options struct { // and ClearNeighbors. UseNeighborCache bool - // AutoGenIPv6LinkLocal determines whether or not the stack will attempt to - // auto-generate an IPv6 link-local address for newly enabled non-loopback - // NICs. - // - // Note, setting this to true does not mean that a link-local address - // will be assigned right away, or at all. If Duplicate Address Detection - // is enabled, an address will only be assigned if it successfully resolves. - // If it fails, no further attempt will be made to auto-generate an IPv6 - // link-local address. - // - // The generated link-local address will follow RFC 4291 Appendix A - // guidelines. - AutoGenIPv6LinkLocal bool - - // NDPDisp is the NDP event dispatcher that an integrator can provide to - // receive NDP related events. - NDPDisp NDPDispatcher - // NUDDisp is the NUD event dispatcher that an integrator can provide to // receive NUD related events. NUDDisp NUDDispatcher @@ -587,31 +512,12 @@ type Options struct { // this is non-nil. RawFactory RawFactory - // OpaqueIIDOpts hold the options for generating opaque interface - // identifiers (IIDs) as outlined by RFC 7217. - OpaqueIIDOpts OpaqueInterfaceIdentifierOptions - // RandSource is an optional source to use to generate random // numbers. If omitted it defaults to a Source seeded by the data // returned by rand.Read(). // // RandSource must be thread-safe. RandSource mathrand.Source - - // TempIIDSeed is used to seed the initial temporary interface identifier - // history value used to generate IIDs for temporary SLAAC addresses. - // - // Temporary SLAAC adresses are short-lived addresses which are unpredictable - // and random from the perspective of other nodes on the network. It is - // recommended that the seed be a random byte buffer of at least - // header.IIDSize bytes to make sure that temporary SLAAC addresses are - // sufficiently random. It should follow minimum randomness requirements for - // security as outlined by RFC 4086. - // - // Note: using a nil value, the same seed across netstack program runs, or a - // seed that is too small would reduce randomness and increase predictability, - // defeating the purpose of temporary SLAAC addresses. - TempIIDSeed []byte } // TransportEndpointInfo holds useful information about a transport endpoint @@ -644,8 +550,8 @@ type TransportEndpointInfo struct { // incompatible with the receiver. // // Preconditon: the parent endpoint mu must be held while calling this method. -func (e *TransportEndpointInfo) AddrNetProtoLocked(addr tcpip.FullAddress, v6only bool) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.NetProto +func (t *TransportEndpointInfo) AddrNetProtoLocked(addr tcpip.FullAddress, v6only bool) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) { + netProto := t.NetProto switch len(addr.Addr) { case header.IPv4AddressSize: netProto = header.IPv4ProtocolNumber @@ -659,7 +565,7 @@ func (e *TransportEndpointInfo) AddrNetProtoLocked(addr tcpip.FullAddress, v6onl } } - switch len(e.ID.LocalAddress) { + switch len(t.ID.LocalAddress) { case header.IPv4AddressSize: if len(addr.Addr) == header.IPv6AddressSize { return tcpip.FullAddress{}, 0, tcpip.ErrInvalidEndpointState @@ -671,8 +577,8 @@ func (e *TransportEndpointInfo) AddrNetProtoLocked(addr tcpip.FullAddress, v6onl } switch { - case netProto == e.NetProto: - case netProto == header.IPv4ProtocolNumber && e.NetProto == header.IPv6ProtocolNumber: + case netProto == t.NetProto: + case netProto == header.IPv4ProtocolNumber && t.NetProto == header.IPv6ProtocolNumber: if v6only { return tcpip.FullAddress{}, 0, tcpip.ErrNoRoute } @@ -714,36 +620,27 @@ func New(opts Options) *Stack { randSrc = &lockedRandomSource{src: mathrand.NewSource(generateRandInt64())} } - // Make sure opts.NDPConfigs contains valid values only. - opts.NDPConfigs.validate() - opts.NUDConfigs.resetInvalidFields() s := &Stack{ - transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState), - networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), - linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver), - nics: make(map[tcpip.NICID]*NIC), - cleanupEndpoints: make(map[TransportEndpoint]struct{}), - linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), - PortManager: ports.NewPortManager(), - clock: clock, - stats: opts.Stats.FillIn(), - handleLocal: opts.HandleLocal, - tables: DefaultTables(), - icmpRateLimiter: NewICMPRateLimiter(), - seed: generateRandUint32(), - ndpConfigs: opts.NDPConfigs, - nudConfigs: opts.NUDConfigs, - useNeighborCache: opts.UseNeighborCache, - autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, - uniqueIDGenerator: opts.UniqueID, - ndpDisp: opts.NDPDisp, - nudDisp: opts.NUDDisp, - opaqueIIDOpts: opts.OpaqueIIDOpts, - tempIIDSeed: opts.TempIIDSeed, - forwarder: newForwardQueue(), - randomGenerator: mathrand.New(randSrc), + transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState), + networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), + linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver), + nics: make(map[tcpip.NICID]*NIC), + cleanupEndpoints: make(map[TransportEndpoint]struct{}), + linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), + PortManager: ports.NewPortManager(), + clock: clock, + stats: opts.Stats.FillIn(), + handleLocal: opts.HandleLocal, + tables: DefaultTables(), + icmpRateLimiter: NewICMPRateLimiter(), + seed: generateRandUint32(), + nudConfigs: opts.NUDConfigs, + useNeighborCache: opts.UseNeighborCache, + uniqueIDGenerator: opts.UniqueID, + nudDisp: opts.NUDDisp, + randomGenerator: mathrand.New(randSrc), sendBufferSize: SendBufferSizeOption{ Min: MinBufferSize, Default: DefaultBufferSize, @@ -755,10 +652,11 @@ func New(opts Options) *Stack { Max: DefaultMaxBufferSize, }, } - s.forwarding.protocols = make(map[tcpip.NetworkProtocolNumber]bool) + s.linkResQueue.init() // Add specified network protocols. - for _, netProto := range opts.NetworkProtocols { + for _, netProtoFactory := range opts.NetworkProtocols { + netProto := netProtoFactory(s) s.networkProtocols[netProto.Number()] = netProto if r, ok := netProto.(LinkAddressResolver); ok { s.linkAddrResolvers[r.LinkAddressProtocol()] = r @@ -766,7 +664,8 @@ func New(opts Options) *Stack { } // Add specified transport protocols. - for _, transProto := range opts.TransportProtocols { + for _, transProtoFactory := range opts.TransportProtocols { + transProto := transProtoFactory(s) s.transportProtocols[transProto.Number()] = &transportProtocolState{ proto: transProto, } @@ -873,42 +772,37 @@ func (s *Stack) Stats() tcpip.Stats { return s.stats } -// SetForwarding enables or disables packet forwarding between NICs. -func (s *Stack) SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) { - s.forwarding.Lock() - defer s.forwarding.Unlock() - - // If this stack does not support the protocol, do nothing. - if _, ok := s.networkProtocols[protocol]; !ok { - return +// SetForwarding enables or disables packet forwarding between NICs for the +// passed protocol. +func (s *Stack) SetForwarding(protocolNum tcpip.NetworkProtocolNumber, enable bool) *tcpip.Error { + protocol, ok := s.networkProtocols[protocolNum] + if !ok { + return tcpip.ErrUnknownProtocol } - // If the forwarding value for this protocol hasn't changed then do - // nothing. - if forwarding := s.forwarding.protocols[protocol]; forwarding == enable { - return + forwardingProtocol, ok := protocol.(ForwardingNetworkProtocol) + if !ok { + return tcpip.ErrNotSupported } - s.forwarding.protocols[protocol] = enable + forwardingProtocol.SetForwarding(enable) + return nil +} - if protocol == header.IPv6ProtocolNumber { - if enable { - for _, nic := range s.nics { - nic.becomeIPv6Router() - } - } else { - for _, nic := range s.nics { - nic.becomeIPv6Host() - } - } +// Forwarding returns true if packet forwarding between NICs is enabled for the +// passed protocol. +func (s *Stack) Forwarding(protocolNum tcpip.NetworkProtocolNumber) bool { + protocol, ok := s.networkProtocols[protocolNum] + if !ok { + return false + } + + forwardingProtocol, ok := protocol.(ForwardingNetworkProtocol) + if !ok { + return false } -} -// Forwarding returns if packet forwarding between NICs is enabled. -func (s *Stack) Forwarding(protocol tcpip.NetworkProtocolNumber) bool { - s.forwarding.RLock() - defer s.forwarding.RUnlock() - return s.forwarding.protocols[protocol] + return forwardingProtocol.Forwarding() } // SetRouteTable assigns the route table to be used by this stack. It @@ -943,7 +837,7 @@ func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcp return nil, tcpip.ErrUnknownProtocol } - return t.proto.NewEndpoint(s, network, waiterQueue) + return t.proto.NewEndpoint(network, waiterQueue) } // NewRawEndpoint creates a new raw transport layer endpoint of the given @@ -963,7 +857,7 @@ func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network return nil, tcpip.ErrUnknownProtocol } - return t.proto.NewRawEndpoint(s, network, waiterQueue) + return t.proto.NewRawEndpoint(network, waiterQueue) } // NewPacketEndpoint creates a new packet endpoint listening for the given @@ -1034,16 +928,16 @@ func (s *Stack) CreateNIC(id tcpip.NICID, ep LinkEndpoint) *tcpip.Error { return s.CreateNICWithOptions(id, ep, NICOptions{}) } -// GetNICByName gets the NIC specified by name. -func (s *Stack) GetNICByName(name string) (*NIC, bool) { +// GetLinkEndpointByName gets the link endpoint specified by name. +func (s *Stack) GetLinkEndpointByName(name string) LinkEndpoint { s.mu.RLock() defer s.mu.RUnlock() for _, nic := range s.nics { if nic.Name() == name { - return nic, true + return nic.LinkEndpoint } } - return nil, false + return nil } // EnableNIC enables the given NIC so that the link-layer endpoint can start @@ -1070,7 +964,8 @@ func (s *Stack) DisableNIC(id tcpip.NICID) *tcpip.Error { return tcpip.ErrUnknownNICID } - return nic.disable() + nic.disable() + return nil } // CheckNIC checks if a NIC is usable. @@ -1083,7 +978,7 @@ func (s *Stack) CheckNIC(id tcpip.NICID) bool { return false } - return nic.enabled() + return nic.Enabled() } // RemoveNIC removes NIC and all related routes from the network stack. @@ -1161,19 +1056,19 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo { for id, nic := range s.nics { flags := NICStateFlags{ Up: true, // Netstack interfaces are always up. - Running: nic.enabled(), + Running: nic.Enabled(), Promiscuous: nic.isPromiscuousMode(), - Loopback: nic.isLoopback(), + Loopback: nic.IsLoopback(), } nics[id] = NICInfo{ Name: nic.name, - LinkAddress: nic.linkEP.LinkAddress(), - ProtocolAddresses: nic.PrimaryAddresses(), + LinkAddress: nic.LinkEndpoint.LinkAddress(), + ProtocolAddresses: nic.primaryAddresses(), Flags: flags, - MTU: nic.linkEP.MTU(), + MTU: nic.LinkEndpoint.MTU(), Stats: nic.stats, Context: nic.context, - ARPHardwareType: nic.linkEP.ARPHardwareType(), + ARPHardwareType: nic.LinkEndpoint.ARPHardwareType(), } } return nics @@ -1232,7 +1127,7 @@ func (s *Stack) AddProtocolAddressWithOptions(id tcpip.NICID, protocolAddress tc return tcpip.ErrUnknownNICID } - return nic.AddAddress(protocolAddress, peb) + return nic.addAddress(protocolAddress, peb) } // RemoveAddress removes an existing network-layer address from the specified @@ -1242,7 +1137,7 @@ func (s *Stack) RemoveAddress(id tcpip.NICID, addr tcpip.Address) *tcpip.Error { defer s.mu.RUnlock() if nic, ok := s.nics[id]; ok { - return nic.RemoveAddress(addr) + return nic.removeAddress(addr) } return tcpip.ErrUnknownNICID @@ -1256,7 +1151,7 @@ func (s *Stack) AllAddresses() map[tcpip.NICID][]tcpip.ProtocolAddress { nics := make(map[tcpip.NICID][]tcpip.ProtocolAddress) for id, nic := range s.nics { - nics[id] = nic.AllAddresses() + nics[id] = nic.allPermanentAddresses() } return nics } @@ -1278,7 +1173,7 @@ func (s *Stack) GetMainNICAddress(id tcpip.NICID, protocol tcpip.NetworkProtocol return nic.primaryAddress(protocol), nil } -func (s *Stack) getRefEP(nic *NIC, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (ref *referencedNetworkEndpoint) { +func (s *Stack) getAddressEP(nic *NIC, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) AssignableAddressEndpoint { if len(localAddr) == 0 { return nic.primaryEndpoint(netProto, remoteAddr) } @@ -1295,9 +1190,9 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n isMulticast := header.IsV4MulticastAddress(remoteAddr) || header.IsV6MulticastAddress(remoteAddr) needRoute := !(isLocalBroadcast || isMulticast || header.IsV6LinkLocalAddress(remoteAddr)) if id != 0 && !needRoute { - if nic, ok := s.nics[id]; ok && nic.enabled() { - if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil { - return makeRoute(netProto, ref.address(), remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.isLoopback(), multicastLoop && !nic.isLoopback()), nil + if nic, ok := s.nics[id]; ok && nic.Enabled() { + if addressEndpoint := s.getAddressEP(nic, localAddr, remoteAddr, netProto); addressEndpoint != nil { + return makeRoute(netProto, addressEndpoint.AddressWithPrefix().Address, remoteAddr, nic, addressEndpoint, s.handleLocal && !nic.IsLoopback(), multicastLoop && !nic.IsLoopback()), nil } } } else { @@ -1305,20 +1200,20 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n if (id != 0 && id != route.NIC) || (len(remoteAddr) != 0 && !route.Destination.Contains(remoteAddr)) { continue } - if nic, ok := s.nics[route.NIC]; ok && nic.enabled() { - if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil { + if nic, ok := s.nics[route.NIC]; ok && nic.Enabled() { + if addressEndpoint := s.getAddressEP(nic, localAddr, remoteAddr, netProto); addressEndpoint != nil { if len(remoteAddr) == 0 { // If no remote address was provided, then the route // provided will refer to the link local address. - remoteAddr = ref.address() + remoteAddr = addressEndpoint.AddressWithPrefix().Address } - r := makeRoute(netProto, ref.address(), remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.isLoopback(), multicastLoop && !nic.isLoopback()) + r := makeRoute(netProto, addressEndpoint.AddressWithPrefix().Address, remoteAddr, nic, addressEndpoint, s.handleLocal && !nic.IsLoopback(), multicastLoop && !nic.IsLoopback()) if len(route.Gateway) > 0 { if needRoute { r.NextHop = route.Gateway } - } else if subnet := ref.addrWithPrefix().Subnet(); subnet.IsBroadcast(remoteAddr) { + } else if subnet := addressEndpoint.AddressWithPrefix().Subnet(); subnet.IsBroadcast(remoteAddr) { r.RemoteLinkAddress = header.EthernetBroadcastAddress } @@ -1356,21 +1251,20 @@ func (s *Stack) CheckLocalAddress(nicID tcpip.NICID, protocol tcpip.NetworkProto return 0 } - ref := nic.findEndpoint(protocol, addr, CanBePrimaryEndpoint) - if ref == nil { + addressEndpoint := nic.findEndpoint(protocol, addr, CanBePrimaryEndpoint) + if addressEndpoint == nil { return 0 } - ref.decRef() + addressEndpoint.DecRef() return nic.id } // Go through all the NICs. for _, nic := range s.nics { - ref := nic.findEndpoint(protocol, addr, CanBePrimaryEndpoint) - if ref != nil { - ref.decRef() + if addressEndpoint := nic.findEndpoint(protocol, addr, CanBePrimaryEndpoint); addressEndpoint != nil { + addressEndpoint.DecRef() return nic.id } } @@ -1429,7 +1323,7 @@ func (s *Stack) GetLinkAddress(nicID tcpip.NICID, addr, localAddr tcpip.Address, fullAddr := tcpip.FullAddress{NIC: nicID, Addr: addr} linkRes := s.linkAddrResolvers[protocol] - return s.linkAddrCache.get(fullAddr, linkRes, localAddr, nic.linkEP, waker) + return s.linkAddrCache.get(fullAddr, linkRes, localAddr, nic.LinkEndpoint, waker) } // Neighbors returns all IP to MAC address associations. @@ -1645,7 +1539,7 @@ func (s *Stack) Wait() { s.mu.RLock() defer s.mu.RUnlock() for _, n := range s.nics { - n.linkEP.Wait() + n.LinkEndpoint.Wait() } } @@ -1733,7 +1627,7 @@ func (s *Stack) WritePacket(nicID tcpip.NICID, dst tcpip.LinkAddress, netProto t // Add our own fake ethernet header. ethFields := header.EthernetFields{ - SrcAddr: nic.linkEP.LinkAddress(), + SrcAddr: nic.LinkEndpoint.LinkAddress(), DstAddr: dst, Type: netProto, } @@ -1742,7 +1636,7 @@ func (s *Stack) WritePacket(nicID tcpip.NICID, dst tcpip.LinkAddress, netProto t vv := buffer.View(fakeHeader).ToVectorisedView() vv.Append(payload) - if err := nic.linkEP.WriteRawPacket(vv); err != nil { + if err := nic.LinkEndpoint.WriteRawPacket(vv); err != nil { return err } @@ -1759,7 +1653,7 @@ func (s *Stack) WriteRawPacket(nicID tcpip.NICID, payload buffer.VectorisedView) return tcpip.ErrUnknownDevice } - if err := nic.linkEP.WriteRawPacket(payload); err != nil { + if err := nic.LinkEndpoint.WriteRawPacket(payload); err != nil { return err } @@ -1839,7 +1733,7 @@ func (s *Stack) LeaveGroup(protocol tcpip.NetworkProtocolNumber, nicID tcpip.NIC defer s.mu.RUnlock() if nic, ok := s.nics[nicID]; ok { - return nic.leaveGroup(multicastAddr) + return nic.leaveGroup(protocol, multicastAddr) } return tcpip.ErrUnknownNICID } @@ -1891,53 +1785,18 @@ func (s *Stack) AllowICMPMessage() bool { return s.icmpRateLimiter.Allow() } -// IsAddrTentative returns true if addr is tentative on the NIC with ID id. -// -// Note that if addr is not associated with a NIC with id ID, then this -// function will return false. It will only return true if the address is -// associated with the NIC AND it is tentative. -func (s *Stack) IsAddrTentative(id tcpip.NICID, addr tcpip.Address) (bool, *tcpip.Error) { - s.mu.RLock() - defer s.mu.RUnlock() - - nic, ok := s.nics[id] - if !ok { - return false, tcpip.ErrUnknownNICID - } - - return nic.isAddrTentative(addr), nil -} - -// DupTentativeAddrDetected attempts to inform the NIC with ID id that a -// tentative addr on it is a duplicate on a link. -func (s *Stack) DupTentativeAddrDetected(id tcpip.NICID, addr tcpip.Address) *tcpip.Error { - s.mu.Lock() - defer s.mu.Unlock() - - nic, ok := s.nics[id] - if !ok { - return tcpip.ErrUnknownNICID - } - - return nic.dupTentativeAddrDetected(addr) -} - -// SetNDPConfigurations sets the per-interface NDP configurations on the NIC -// with ID id to c. -// -// Note, if c contains invalid NDP configuration values, it will be fixed to -// use default values for the erroneous values. -func (s *Stack) SetNDPConfigurations(id tcpip.NICID, c NDPConfigurations) *tcpip.Error { +// GetNetworkEndpoint returns the NetworkEndpoint with the specified protocol +// number installed on the specified NIC. +func (s *Stack) GetNetworkEndpoint(nicID tcpip.NICID, proto tcpip.NetworkProtocolNumber) (NetworkEndpoint, *tcpip.Error) { s.mu.Lock() defer s.mu.Unlock() - nic, ok := s.nics[id] + nic, ok := s.nics[nicID] if !ok { - return tcpip.ErrUnknownNICID + return nil, tcpip.ErrUnknownNICID } - nic.setNDPConfigs(c) - return nil + return nic.getNetworkEndpoint(proto), nil } // NUDConfigurations gets the per-interface NUD configurations. @@ -1950,7 +1809,7 @@ func (s *Stack) NUDConfigurations(id tcpip.NICID) (NUDConfigurations, *tcpip.Err return NUDConfigurations{}, tcpip.ErrUnknownNICID } - return nic.NUDConfigs() + return nic.nudConfigs() } // SetNUDConfigurations sets the per-interface NUD configurations. @@ -1969,22 +1828,6 @@ func (s *Stack) SetNUDConfigurations(id tcpip.NICID, c NUDConfigurations) *tcpip return nic.setNUDConfigs(c) } -// HandleNDPRA provides a NIC with ID id a validated NDP Router Advertisement -// message that it needs to handle. -func (s *Stack) HandleNDPRA(id tcpip.NICID, ip tcpip.Address, ra header.NDPRouterAdvert) *tcpip.Error { - s.mu.Lock() - defer s.mu.Unlock() - - nic, ok := s.nics[id] - if !ok { - return tcpip.ErrUnknownNICID - } - - nic.handleNDPRA(ip, ra) - - return nil -} - // Seed returns a 32 bit value that can be used as a seed value for port // picking, ISN generation etc. // @@ -2026,16 +1869,12 @@ func (s *Stack) FindNetworkEndpoint(netProto tcpip.NetworkProtocolNumber, addres defer s.mu.RUnlock() for _, nic := range s.nics { - id := NetworkEndpointID{address} - - if ref, ok := nic.mu.endpoints[id]; ok { - nic.mu.RLock() - defer nic.mu.RUnlock() - - // An endpoint with this id exists, check if it can be - // used and return it. - return ref.ep, nil + addressEndpoint := nic.getAddressOrCreateTempInner(netProto, address, false /* createTemp */, NeverPrimaryEndpoint) + if addressEndpoint == nil { + continue } + addressEndpoint.DecRef() + return nic.getNetworkEndpoint(netProto), nil } return nil, tcpip.ErrBadAddress } @@ -2052,3 +1891,8 @@ func (s *Stack) FindNICNameFromID(id tcpip.NICID) string { return nic.Name() } + +// NewJob returns a new tcpip.Job using the stack's clock. +func (s *Stack) NewJob(l sync.Locker, f func()) *tcpip.Job { + return tcpip.NewJob(s.clock, l, f) +} diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 7669ba672..38994cca1 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -21,7 +21,6 @@ import ( "bytes" "fmt" "math" - "net" "sort" "testing" "time" @@ -29,12 +28,12 @@ import ( "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" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" "gvisor.dev/gvisor/pkg/tcpip/link/loopback" - "gvisor.dev/gvisor/pkg/tcpip/network/arp" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -68,18 +67,40 @@ const ( // use the first three: destination address, source address, and transport // protocol. They're all one byte fields to simplify parsing. type fakeNetworkEndpoint struct { - nicID tcpip.NICID + stack.AddressableEndpointState + + mu struct { + sync.RWMutex + + enabled bool + } + + nic stack.NetworkInterface proto *fakeNetworkProtocol dispatcher stack.TransportDispatcher - ep stack.LinkEndpoint } -func (f *fakeNetworkEndpoint) MTU() uint32 { - return f.ep.MTU() - uint32(f.MaxHeaderLength()) +func (f *fakeNetworkEndpoint) Enable() *tcpip.Error { + f.mu.Lock() + defer f.mu.Unlock() + f.mu.enabled = true + return nil } -func (f *fakeNetworkEndpoint) NICID() tcpip.NICID { - return f.nicID +func (f *fakeNetworkEndpoint) Enabled() bool { + f.mu.RLock() + defer f.mu.RUnlock() + return f.mu.enabled +} + +func (f *fakeNetworkEndpoint) Disable() { + f.mu.Lock() + defer f.mu.Unlock() + f.mu.enabled = false +} + +func (f *fakeNetworkEndpoint) MTU() uint32 { + return f.nic.MTU() - uint32(f.MaxHeaderLength()) } func (*fakeNetworkEndpoint) DefaultTTL() uint8 { @@ -111,17 +132,13 @@ func (f *fakeNetworkEndpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuff } func (f *fakeNetworkEndpoint) MaxHeaderLength() uint16 { - return f.ep.MaxHeaderLength() + fakeNetHeaderLen + return f.nic.MaxHeaderLength() + fakeNetHeaderLen } func (f *fakeNetworkEndpoint) PseudoHeaderChecksum(protocol tcpip.TransportProtocolNumber, dstAddr tcpip.Address) uint16 { return 0 } -func (f *fakeNetworkEndpoint) Capabilities() stack.LinkEndpointCapabilities { - return f.ep.Capabilities() -} - func (f *fakeNetworkEndpoint) NetworkProtocolNumber() tcpip.NetworkProtocolNumber { return f.proto.Number() } @@ -144,7 +161,7 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, params return nil } - return f.ep.WritePacket(r, gso, fakeNetNumber, pkt) + return f.nic.WritePacket(r, gso, fakeNetNumber, pkt) } // WritePackets implements stack.LinkEndpoint.WritePackets. @@ -156,7 +173,9 @@ func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, pkt *stack return tcpip.ErrNotSupported } -func (*fakeNetworkEndpoint) Close() {} +func (f *fakeNetworkEndpoint) Close() { + f.AddressableEndpointState.Cleanup() +} // fakeNetworkProtocol is a network-layer protocol descriptor. It aggregates the // number of packets sent and received via endpoints of this protocol. The index @@ -165,6 +184,11 @@ type fakeNetworkProtocol struct { packetCount [10]int sendPacketCount [10]int defaultTTL uint8 + + mu struct { + sync.RWMutex + forwarding bool + } } func (f *fakeNetworkProtocol) Number() tcpip.NetworkProtocolNumber { @@ -187,13 +211,14 @@ func (*fakeNetworkProtocol) ParseAddresses(v buffer.View) (src, dst tcpip.Addres return tcpip.Address(v[srcAddrOffset : srcAddrOffset+1]), tcpip.Address(v[dstAddrOffset : dstAddrOffset+1]) } -func (f *fakeNetworkProtocol) NewEndpoint(nicID tcpip.NICID, _ stack.LinkAddressCache, _ stack.NUDHandler, dispatcher stack.TransportDispatcher, ep stack.LinkEndpoint, _ *stack.Stack) stack.NetworkEndpoint { - return &fakeNetworkEndpoint{ - nicID: nicID, +func (f *fakeNetworkProtocol) NewEndpoint(nic stack.NetworkInterface, _ stack.LinkAddressCache, _ stack.NUDHandler, dispatcher stack.TransportDispatcher) stack.NetworkEndpoint { + e := &fakeNetworkEndpoint{ + nic: nic, proto: f, dispatcher: dispatcher, - ep: ep, } + e.AddressableEndpointState.Init(e) + return e } func (f *fakeNetworkProtocol) SetOption(option tcpip.SettableNetworkProtocolOption) *tcpip.Error { @@ -216,13 +241,13 @@ func (f *fakeNetworkProtocol) Option(option tcpip.GettableNetworkProtocolOption) } } -// Close implements TransportProtocol.Close. +// Close implements NetworkProtocol.Close. func (*fakeNetworkProtocol) Close() {} -// Wait implements TransportProtocol.Wait. +// Wait implements NetworkProtocol.Wait. func (*fakeNetworkProtocol) Wait() {} -// Parse implements TransportProtocol.Parse. +// Parse implements NetworkProtocol.Parse. func (*fakeNetworkProtocol) Parse(pkt *stack.PacketBuffer) (tcpip.TransportProtocolNumber, bool, bool) { hdr, ok := pkt.NetworkHeader().Consume(fakeNetHeaderLen) if !ok { @@ -231,7 +256,21 @@ func (*fakeNetworkProtocol) Parse(pkt *stack.PacketBuffer) (tcpip.TransportProto return tcpip.TransportProtocolNumber(hdr[protocolNumberOffset]), true, true } -func fakeNetFactory() stack.NetworkProtocol { +// Forwarding implements stack.ForwardingNetworkProtocol. +func (f *fakeNetworkProtocol) Forwarding() bool { + f.mu.RLock() + defer f.mu.RUnlock() + return f.mu.forwarding +} + +// SetForwarding implements stack.ForwardingNetworkProtocol. +func (f *fakeNetworkProtocol) SetForwarding(v bool) { + f.mu.Lock() + defer f.mu.Unlock() + f.mu.forwarding = v +} + +func fakeNetFactory(*stack.Stack) stack.NetworkProtocol { return &fakeNetworkProtocol{} } @@ -268,7 +307,7 @@ func TestNetworkReceive(t *testing.T) { // addresses attached to it: 1 & 2. ep := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -428,7 +467,7 @@ func TestNetworkSend(t *testing.T) { // existing nic. ep := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) if err := s.CreateNIC(1, ep); err != nil { t.Fatal("NewNIC failed:", err) @@ -455,7 +494,7 @@ func TestNetworkSendMultiRoute(t *testing.T) { // addresses per nic, the first nic has odd address, the second one has // even addresses. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(10, defaultMTU, "") @@ -555,7 +594,7 @@ func TestAttachToLinkEndpointImmediately(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) e := linkEPWithMockedAttach{ @@ -574,7 +613,7 @@ func TestAttachToLinkEndpointImmediately(t *testing.T) { func TestDisableUnknownNIC(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) if err := s.DisableNIC(1); err != tcpip.ErrUnknownNICID { @@ -586,7 +625,7 @@ func TestDisabledNICsNICInfoAndCheckNIC(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) e := loopback.New() @@ -633,7 +672,7 @@ func TestDisabledNICsNICInfoAndCheckNIC(t *testing.T) { func TestRemoveUnknownNIC(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) if err := s.RemoveNIC(1); err != tcpip.ErrUnknownNICID { @@ -645,7 +684,7 @@ func TestRemoveNIC(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) e := linkEPWithMockedAttach{ @@ -706,7 +745,7 @@ func TestRouteWithDownNIC(t *testing.T) { setup := func(t *testing.T) (*stack.Stack, *channel.Endpoint, *channel.Endpoint) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(1, defaultMTU, "") @@ -872,7 +911,7 @@ func TestRoutes(t *testing.T) { // addresses per nic, the first nic has odd address, the second one has // even addresses. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(10, defaultMTU, "") @@ -952,7 +991,7 @@ func TestAddressRemoval(t *testing.T) { remoteAddr := tcpip.Address("\x02") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -999,7 +1038,7 @@ func TestAddressRemovalWithRouteHeld(t *testing.T) { remoteAddr := tcpip.Address("\x02") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1090,7 +1129,7 @@ func TestEndpointExpiration(t *testing.T) { for _, spoofing := range []bool{true, false} { t.Run(fmt.Sprintf("promiscuous=%t spoofing=%t", promiscuous, spoofing), func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1248,7 +1287,7 @@ func TestEndpointExpiration(t *testing.T) { func TestPromiscuousMode(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1300,7 +1339,7 @@ func TestSpoofingWithAddress(t *testing.T) { dstAddr := tcpip.Address("\x03") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1366,7 +1405,7 @@ func TestSpoofingNoAddress(t *testing.T) { dstAddr := tcpip.Address("\x02") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1429,7 +1468,7 @@ func verifyRoute(gotRoute, wantRoute stack.Route) error { func TestOutgoingBroadcastWithEmptyRouteTable(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1472,7 +1511,7 @@ func TestOutgoingBroadcastWithRouteTable(t *testing.T) { // Create a new stack with two NICs. s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1573,7 +1612,7 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) { } { t.Run(tc.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") @@ -1630,8 +1669,8 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) { func TestNetworkOption(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{}, }) opt := tcpip.DefaultTTLOption(5) @@ -1657,7 +1696,7 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { for never := 0; never < 3; never++ { t.Run(fmt.Sprintf("never=%d", never), func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1724,7 +1763,7 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { func TestGetMainNICAddressAddRemove(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1809,7 +1848,7 @@ func verifyAddresses(t *testing.T, expectedAddresses, gotAddresses []tcpip.Proto func TestAddAddress(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -1836,7 +1875,7 @@ func TestAddAddress(t *testing.T) { func TestAddProtocolAddress(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -1870,7 +1909,7 @@ func TestAddProtocolAddress(t *testing.T) { func TestAddAddressWithOptions(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -1901,7 +1940,7 @@ func TestAddAddressWithOptions(t *testing.T) { func TestAddProtocolAddressWithOptions(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -2022,7 +2061,7 @@ func TestCreateNICWithOptions(t *testing.T) { func TestNICStats(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { @@ -2063,7 +2102,7 @@ func TestNICStats(t *testing.T) { t.Errorf("got Tx.Packets.Value() = %d, ep1.Drain() = %d", got, want) } - if got, want := s.NICInfo()[1].Stats.Tx.Bytes.Value(), uint64(len(payload)); got != want { + if got, want := s.NICInfo()[1].Stats.Tx.Bytes.Value(), uint64(len(payload)+fakeNetHeaderLen); got != want { t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want) } } @@ -2089,7 +2128,7 @@ func TestNICForwarding(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) s.SetForwarding(fakeNetNumber, true) @@ -2213,7 +2252,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName string autoGen bool linkAddr tcpip.LinkAddress - iidOpts stack.OpaqueInterfaceIdentifierOptions + iidOpts ipv6.OpaqueInterfaceIdentifierOptions shouldGen bool expectedAddr tcpip.Address }{ @@ -2229,7 +2268,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "nic1", autoGen: false, linkAddr: linkAddr1, - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:], }, @@ -2274,7 +2313,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "nic1", autoGen: true, linkAddr: linkAddr1, - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:], }, @@ -2286,7 +2325,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { { name: "OIID Empty MAC and empty nicName", autoGen: true, - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:1], }, @@ -2298,7 +2337,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "test", autoGen: true, linkAddr: "\x01\x02\x03", - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:2], }, @@ -2310,7 +2349,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "test2", autoGen: true, linkAddr: "\x01\x02\x03\x04\x05\x06", - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, SecretKey: secretKey[:3], }, @@ -2322,7 +2361,7 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { nicName: "test3", autoGen: true, linkAddr: "\x00\x00\x00\x00\x00\x00", - iidOpts: stack.OpaqueInterfaceIdentifierOptions{ + iidOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: nicNameFunc, }, shouldGen: true, @@ -2336,10 +2375,11 @@ func TestNICAutoGenLinkLocalAddr(t *testing.T) { autoGenAddrC: make(chan ndpAutoGenAddrEvent, 1), } opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: test.autoGen, - NDPDisp: &ndpDisp, - OpaqueIIDOpts: test.iidOpts, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: test.autoGen, + NDPDisp: &ndpDisp, + OpaqueIIDOpts: test.iidOpts, + })}, } e := channel.New(0, 1280, test.linkAddr) @@ -2411,15 +2451,15 @@ func TestNoLinkLocalAutoGenForLoopbackNIC(t *testing.T) { tests := []struct { name string - opaqueIIDOpts stack.OpaqueInterfaceIdentifierOptions + opaqueIIDOpts ipv6.OpaqueInterfaceIdentifierOptions }{ { name: "IID From MAC", - opaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{}, + opaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{}, }, { name: "Opaque IID", - opaqueIIDOpts: stack.OpaqueInterfaceIdentifierOptions{ + opaqueIIDOpts: ipv6.OpaqueInterfaceIdentifierOptions{ NICNameFromID: func(_ tcpip.NICID, nicName string) string { return nicName }, @@ -2430,9 +2470,10 @@ func TestNoLinkLocalAutoGenForLoopbackNIC(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - AutoGenIPv6LinkLocal: true, - OpaqueIIDOpts: test.opaqueIIDOpts, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + AutoGenIPv6LinkLocal: true, + OpaqueIIDOpts: test.opaqueIIDOpts, + })}, } e := loopback.New() @@ -2461,12 +2502,13 @@ func TestNICAutoGenAddrDoesDAD(t *testing.T) { ndpDisp := ndpDispatcher{ dadC: make(chan ndpDADEvent), } - ndpConfigs := stack.DefaultNDPConfigurations() + ndpConfigs := ipv6.DefaultNDPConfigurations() opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: ndpConfigs, - AutoGenIPv6LinkLocal: true, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ndpConfigs, + AutoGenIPv6LinkLocal: true, + NDPDisp: &ndpDisp, + })}, } e := channel.New(int(ndpConfigs.DupAddrDetectTransmits), 1280, linkAddr1) @@ -2522,7 +2564,7 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) { for _, ps := range pebs { t.Run(fmt.Sprintf("%d-to-%d", pi, ps), func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { @@ -2813,14 +2855,15 @@ func TestIPv6SourceAddressSelectionScopeAndSameAddress(t *testing.T) { t.Run(test.name, func(t *testing.T) { e := channel.New(0, 1280, linkAddr1) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - HandleRAs: true, - AutoGenGlobalAddresses: true, - AutoGenTempGlobalAddresses: true, - }, - NDPDisp: &ndpDispatcher{}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + HandleRAs: true, + AutoGenGlobalAddresses: true, + AutoGenTempGlobalAddresses: true, + }, + NDPDisp: &ndpDispatcher{}, + })}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) if err := s.CreateNIC(nicID, e); err != nil { t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) @@ -2869,7 +2912,7 @@ func TestAddRemoveIPv4BroadcastAddressOnNICEnableDisable(t *testing.T) { e := loopback.New() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, }) nicOpts := stack.NICOptions{Disabled: true} if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil { @@ -2921,7 +2964,7 @@ func TestLeaveIPv6SolicitedNodeAddrBeforeAddrRemoval(t *testing.T) { const nicID = 1 s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocol}, }) e := channel.New(10, 1280, linkAddr1) if err := s.CreateNIC(1, e); err != nil { @@ -2982,7 +3025,7 @@ func TestJoinLeaveMulticastOnNICEnableDisable(t *testing.T) { t.Run(test.name, func(t *testing.T) { e := loopback.New() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, }) nicOpts := stack.NICOptions{Disabled: true} if err := s.CreateNICWithOptions(nicID, e, nicOpts); err != nil { @@ -3059,12 +3102,13 @@ func TestDoDADWhenNICEnabled(t *testing.T) { dadC: make(chan ndpDADEvent), } opts := stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, - NDPConfigs: stack.NDPConfigurations{ - DupAddrDetectTransmits: dadTransmits, - RetransmitTimer: retransmitTimer, - }, - NDPDisp: &ndpDisp, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPConfigs: ipv6.NDPConfigurations{ + DupAddrDetectTransmits: dadTransmits, + RetransmitTimer: retransmitTimer, + }, + NDPDisp: &ndpDisp, + })}, } e := channel.New(dadTransmits, 1280, linkAddr1) @@ -3423,7 +3467,7 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, }) ep := channel.New(0, defaultMTU, "") if err := s.CreateNIC(nicID1, ep); err != nil { @@ -3454,48 +3498,130 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { } } -func TestResolveWith(t *testing.T) { +// TestRouteReleaseAfterAddrRemoval tests that releasing a Route after its +// associated address is removed should not cause a panic. +func TestRouteReleaseAfterAddrRemoval(t *testing.T) { const ( - unspecifiedNICID = 0 - nicID = 1 + nicID = 1 + localAddr = tcpip.Address("\x01") + remoteAddr = tcpip.Address("\x02") ) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, }) + ep := channel.New(0, defaultMTU, "") - ep.LinkEPCapabilities |= stack.CapabilityResolutionRequired if err := s.CreateNIC(nicID, ep); err != nil { t.Fatalf("CreateNIC(%d, _): %s", nicID, err) } - addr := tcpip.ProtocolAddress{ - Protocol: header.IPv4ProtocolNumber, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: tcpip.Address(net.ParseIP("192.168.1.58").To4()), - PrefixLen: 24, - }, + if err := s.AddAddress(nicID, fakeNetNumber, localAddr); err != nil { + t.Fatalf("s.AddAddress(%d, %d, %s): %s", nicID, fakeNetNumber, localAddr, err) } - if err := s.AddProtocolAddress(nicID, addr); err != nil { - t.Fatalf("AddProtocolAddress(%d, %+v): %s", nicID, addr, err) + { + subnet, err := tcpip.NewSubnet("\x00", "\x00") + if err != nil { + t.Fatal(err) + } + s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}}) } - s.SetRouteTable([]tcpip.Route{{Destination: header.IPv4EmptySubnet, NIC: nicID}}) - - remoteAddr := tcpip.Address(net.ParseIP("192.168.1.59").To4()) - r, err := s.FindRoute(unspecifiedNICID, "" /* localAddr */, remoteAddr, header.IPv4ProtocolNumber, false /* multicastLoop */) + r, err := s.FindRoute(nicID, localAddr, remoteAddr, fakeNetNumber, false /* multicastLoop */) if err != nil { - t.Fatalf("FindRoute(%d, '', %s, %d): %s", unspecifiedNICID, remoteAddr, header.IPv4ProtocolNumber, err) + t.Fatalf("s.FindRoute(%d, %s, %s, %d, false): %s", nicID, localAddr, remoteAddr, fakeNetNumber, err) } + // Should not panic. defer r.Release() - // Should initially require resolution. - if !r.IsResolutionRequired() { - t.Fatal("got r.IsResolutionRequired() = false, want = true") + // Check that removing the same address fails. + if err := s.RemoveAddress(nicID, localAddr); err != nil { + t.Fatalf("s.RemoveAddress(%d, %s): %s", nicID, localAddr, err) + } +} + +func TestGetNetworkEndpoint(t *testing.T) { + const nicID = 1 + + tests := []struct { + name string + protoFactory stack.NetworkProtocolFactory + protoNum tcpip.NetworkProtocolNumber + }{ + { + name: "IPv4", + protoFactory: ipv4.NewProtocol, + protoNum: ipv4.ProtocolNumber, + }, + { + name: "IPv6", + protoFactory: ipv6.NewProtocol, + protoNum: ipv6.ProtocolNumber, + }, + } + + factories := make([]stack.NetworkProtocolFactory, 0, len(tests)) + for _, test := range tests { + factories = append(factories, test.protoFactory) + } + + s := stack.New(stack.Options{ + NetworkProtocols: factories, + }) + + if err := s.CreateNIC(nicID, channel.New(0, defaultMTU, "")); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID, err) + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + ep, err := s.GetNetworkEndpoint(nicID, test.protoNum) + if err != nil { + t.Fatalf("s.GetNetworkEndpoint(%d, %d): %s", nicID, test.protoNum, err) + } + + if got := ep.NetworkProtocolNumber(); got != test.protoNum { + t.Fatalf("got ep.NetworkProtocolNumber() = %d, want = %d", got, test.protoNum) + } + }) + } +} + +func TestGetMainNICAddressWhenNICDisabled(t *testing.T) { + const nicID = 1 + + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + }) + + if err := s.CreateNIC(nicID, channel.New(0, defaultMTU, "")); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID, err) } - // Manually resolving the route should no longer require resolution. - r.ResolveWith("\x01") - if r.IsResolutionRequired() { - t.Fatal("got r.IsResolutionRequired() = true, want = false") + protocolAddress := tcpip.ProtocolAddress{ + Protocol: fakeNetNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: "\x01", + PrefixLen: 8, + }, + } + if err := s.AddProtocolAddress(nicID, protocolAddress); err != nil { + t.Fatalf("AddProtocolAddress(%d, %#v): %s", nicID, protocolAddress, err) + } + + // Check that we get the right initial address and prefix length. + if gotAddr, err := s.GetMainNICAddress(nicID, fakeNetNumber); err != nil { + t.Fatalf("GetMainNICAddress(%d, %d): %s", nicID, fakeNetNumber, err) + } else if gotAddr != protocolAddress.AddressWithPrefix { + t.Fatalf("got GetMainNICAddress(%d, %d) = %s, want = %s", nicID, fakeNetNumber, gotAddr, protocolAddress.AddressWithPrefix) + } + + // Should still get the address when the NIC is diabled. + if err := s.DisableNIC(nicID); err != nil { + t.Fatalf("DisableNIC(%d): %s", nicID, err) + } + if gotAddr, err := s.GetMainNICAddress(nicID, fakeNetNumber); err != nil { + t.Fatalf("GetMainNICAddress(%d, %d): %s", nicID, fakeNetNumber, err) + } else if gotAddr != protocolAddress.AddressWithPrefix { + t.Fatalf("got GetMainNICAddress(%d, %d) = %s, want = %s", nicID, fakeNetNumber, gotAddr, protocolAddress.AddressWithPrefix) } } diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 0774b5382..35e5b1a2e 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -155,7 +155,7 @@ func (epsByNIC *endpointsByNIC) transportEndpoints() []TransportEndpoint { func (epsByNIC *endpointsByNIC) handlePacket(r *Route, id TransportEndpointID, pkt *PacketBuffer) { epsByNIC.mu.RLock() - mpep, ok := epsByNIC.endpoints[r.ref.nic.ID()] + mpep, ok := epsByNIC.endpoints[r.nic.ID()] if !ok { if mpep, ok = epsByNIC.endpoints[0]; !ok { epsByNIC.mu.RUnlock() // Don't use defer for performance reasons. @@ -544,9 +544,11 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto return true } - // If the packet is a TCP packet with a non-unicast source or destination - // address, then do nothing further and instruct the caller to do the same. - if protocol == header.TCPProtocolNumber && (!isInboundUnicast(r) || !isOutboundUnicast(r)) { + // If the packet is a TCP packet with a unspecified source or non-unicast + // destination address, then do nothing further and instruct the caller to do + // the same. The network layer handles address validation for specified source + // addresses. + if protocol == header.TCPProtocolNumber && (!isSpecified(r.LocalAddress) || !isSpecified(r.RemoteAddress) || isInboundMulticastOrBroadcast(r)) { // TCP can only be used to communicate between a single source and a // single destination; the addresses must be unicast. r.Stats().TCP.InvalidSegmentsReceived.Increment() @@ -626,7 +628,7 @@ func (d *transportDemuxer) findTransportEndpoint(netProto tcpip.NetworkProtocolN epsByNIC.mu.RLock() eps.mu.RUnlock() - mpep, ok := epsByNIC.endpoints[r.ref.nic.ID()] + mpep, ok := epsByNIC.endpoints[r.nic.ID()] if !ok { if mpep, ok = epsByNIC.endpoints[0]; !ok { epsByNIC.mu.RUnlock() // Don't use defer for performance reasons. @@ -681,10 +683,6 @@ func isInboundMulticastOrBroadcast(r *Route) bool { return r.IsInboundBroadcast() || header.IsV4MulticastAddress(r.LocalAddress) || header.IsV6MulticastAddress(r.LocalAddress) } -func isInboundUnicast(r *Route) bool { - return r.LocalAddress != header.IPv4Any && r.LocalAddress != header.IPv6Any && !isInboundMulticastOrBroadcast(r) -} - -func isOutboundUnicast(r *Route) bool { - return r.RemoteAddress != header.IPv4Any && r.RemoteAddress != header.IPv6Any && !r.IsOutboundBroadcast() && !header.IsV4MulticastAddress(r.RemoteAddress) && !header.IsV6MulticastAddress(r.RemoteAddress) +func isSpecified(addr tcpip.Address) bool { + return addr != header.IPv4Any && addr != header.IPv6Any } diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index 4d6d62eec..698c8609e 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -51,8 +51,8 @@ type testContext struct { // newDualTestContextMultiNIC creates the testing context and also linkEpIDs NICs. func newDualTestContextMultiNIC(t *testing.T, mtu uint32, linkEpIDs []tcpip.NICID) *testContext { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) linkEps := make(map[tcpip.NICID]*channel.Endpoint) for _, linkEpID := range linkEpIDs { @@ -182,8 +182,8 @@ func TestTransportDemuxerRegister(t *testing.T) { } { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) var wq waiter.Queue ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &wq) diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 64e44bc99..62ab6d92f 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -39,7 +39,7 @@ const ( // use it. type fakeTransportEndpoint struct { stack.TransportEndpointInfo - stack *stack.Stack + proto *fakeTransportProtocol peerAddr tcpip.Address route stack.Route @@ -59,8 +59,8 @@ func (*fakeTransportEndpoint) Stats() tcpip.EndpointStats { func (*fakeTransportEndpoint) SetOwner(owner tcpip.PacketOwner) {} -func newFakeTransportEndpoint(s *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint { - return &fakeTransportEndpoint{stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID} +func newFakeTransportEndpoint(proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint { + return &fakeTransportEndpoint{TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID} } func (f *fakeTransportEndpoint) Abort() { @@ -143,7 +143,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { f.peerAddr = addr.Addr // Find the route. - r, err := f.stack.FindRoute(addr.NIC, "", addr.Addr, fakeNetNumber, false /* multicastLoop */) + r, err := f.proto.stack.FindRoute(addr.NIC, "", addr.Addr, fakeNetNumber, false /* multicastLoop */) if err != nil { return tcpip.ErrNoRoute } @@ -151,7 +151,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // Try to register so that we can start receiving packets. f.ID.RemoteAddress = addr.Addr - err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.ID, f, ports.Flags{}, 0 /* bindToDevice */) + err = f.proto.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.ID, f, ports.Flags{}, 0 /* bindToDevice */) if err != nil { return err } @@ -190,7 +190,7 @@ func (f *fakeTransportEndpoint) Accept(*tcpip.FullAddress) (tcpip.Endpoint, *wai } func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error { - if err := f.stack.RegisterTransportEndpoint( + if err := f.proto.stack.RegisterTransportEndpoint( a.NIC, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, @@ -218,7 +218,6 @@ func (f *fakeTransportEndpoint) HandlePacket(r *stack.Route, id stack.TransportE f.proto.packetCount++ if f.acceptQueue != nil { f.acceptQueue = append(f.acceptQueue, fakeTransportEndpoint{ - stack: f.stack, TransportEndpointInfo: stack.TransportEndpointInfo{ ID: f.ID, NetProto: f.NetProto, @@ -262,6 +261,8 @@ type fakeTransportProtocolOptions struct { // fakeTransportProtocol is a transport-layer protocol descriptor. It // aggregates the number of packets received via endpoints of this protocol. type fakeTransportProtocol struct { + stack *stack.Stack + packetCount int controlCount int opts fakeTransportProtocolOptions @@ -271,11 +272,11 @@ func (*fakeTransportProtocol) Number() tcpip.TransportProtocolNumber { return fakeTransNumber } -func (f *fakeTransportProtocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return newFakeTransportEndpoint(stack, f, netProto, stack.UniqueID()), nil +func (f *fakeTransportProtocol) NewEndpoint(netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return newFakeTransportEndpoint(f, netProto, f.stack.UniqueID()), nil } -func (*fakeTransportProtocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, _ *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (*fakeTransportProtocol) NewRawEndpoint(tcpip.NetworkProtocolNumber, *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return nil, tcpip.ErrUnknownProtocol } @@ -287,8 +288,8 @@ func (*fakeTransportProtocol) ParsePorts(buffer.View) (src, dst uint16, err *tcp return 0, 0, nil } -func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) bool { - return true +func (*fakeTransportProtocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) stack.UnknownDestinationPacketDisposition { + return stack.UnknownDestinationPacketHandled } func (f *fakeTransportProtocol) SetOption(option tcpip.SettableTransportProtocolOption) *tcpip.Error { @@ -326,15 +327,15 @@ func (*fakeTransportProtocol) Parse(pkt *stack.PacketBuffer) bool { return ok } -func fakeTransFactory() stack.TransportProtocol { - return &fakeTransportProtocol{} +func fakeTransFactory(s *stack.Stack) stack.TransportProtocol { + return &fakeTransportProtocol{stack: s} } func TestTransportReceive(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) @@ -404,8 +405,8 @@ func TestTransportReceive(t *testing.T) { func TestTransportControlReceive(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) @@ -481,8 +482,8 @@ func TestTransportControlReceive(t *testing.T) { func TestTransportSend(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) @@ -527,8 +528,8 @@ func TestTransportSend(t *testing.T) { func TestTransportOptions(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) v := tcpip.TCPModerateReceiveBufferOption(true) @@ -546,8 +547,8 @@ func TestTransportOptions(t *testing.T) { func TestTransportForwarding(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, - TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + NetworkProtocols: []stack.NetworkProtocolFactory{fakeNetFactory}, + TransportProtocols: []stack.TransportProtocolFactory{fakeTransFactory}, }) s.SetForwarding(fakeNetNumber, true) diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 464608dee..c42bb0991 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -237,6 +237,14 @@ type Timer interface { // network node. Or, in the case of unix endpoints, it may represent a path. type Address string +// WithPrefix returns the address with a prefix that represents a point subnet. +func (a Address) WithPrefix() AddressWithPrefix { + return AddressWithPrefix{ + Address: a, + PrefixLen: len(a) * 8, + } +} + // AddressMask is a bitmask for an address. type AddressMask string @@ -1614,9 +1622,6 @@ type UDPStats struct { // ChecksumErrors is the number of datagrams dropped due to bad checksums. ChecksumErrors *StatCounter - - // InvalidSourceAddress is the number of invalid sourced datagrams dropped. - InvalidSourceAddress *StatCounter } // Stats holds statistics about the networking stack. diff --git a/pkg/tcpip/tests/integration/BUILD b/pkg/tcpip/tests/integration/BUILD index 06c7a3cd3..a4f141253 100644 --- a/pkg/tcpip/tests/integration/BUILD +++ b/pkg/tcpip/tests/integration/BUILD @@ -6,6 +6,8 @@ go_test( name = "integration_test", size = "small", srcs = [ + "forward_test.go", + "link_resolution_test.go", "loopback_test.go", "multicast_broadcast_test.go", ], @@ -15,6 +17,8 @@ go_test( "//pkg/tcpip/header", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/loopback", + "//pkg/tcpip/link/pipe", + "//pkg/tcpip/network/arp", "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", "//pkg/tcpip/stack", diff --git a/pkg/tcpip/tests/integration/forward_test.go b/pkg/tcpip/tests/integration/forward_test.go new file mode 100644 index 000000000..ffd38ee1a --- /dev/null +++ b/pkg/tcpip/tests/integration/forward_test.go @@ -0,0 +1,378 @@ +// 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 integration_test + +import ( + "net" + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/link/pipe" + "gvisor.dev/gvisor/pkg/tcpip/network/arp" + "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/udp" + "gvisor.dev/gvisor/pkg/waiter" +) + +func TestForwarding(t *testing.T) { + const ( + host1NICLinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x06") + routerNIC1LinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x07") + routerNIC2LinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x08") + host2NICLinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x09") + + host1NICID = 1 + routerNICID1 = 2 + routerNICID2 = 3 + host2NICID = 4 + + listenPort = 8080 + ) + + host1IPv4Addr := tcpip.ProtocolAddress{ + Protocol: ipv4.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("192.168.0.2").To4()), + PrefixLen: 24, + }, + } + routerNIC1IPv4Addr := tcpip.ProtocolAddress{ + Protocol: ipv4.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("192.168.0.1").To4()), + PrefixLen: 24, + }, + } + routerNIC2IPv4Addr := tcpip.ProtocolAddress{ + Protocol: ipv4.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("10.0.0.1").To4()), + PrefixLen: 8, + }, + } + host2IPv4Addr := tcpip.ProtocolAddress{ + Protocol: ipv4.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("10.0.0.2").To4()), + PrefixLen: 8, + }, + } + host1IPv6Addr := tcpip.ProtocolAddress{ + Protocol: ipv6.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("a::2").To16()), + PrefixLen: 64, + }, + } + routerNIC1IPv6Addr := tcpip.ProtocolAddress{ + Protocol: ipv6.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("a::1").To16()), + PrefixLen: 64, + }, + } + routerNIC2IPv6Addr := tcpip.ProtocolAddress{ + Protocol: ipv6.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("b::1").To16()), + PrefixLen: 64, + }, + } + host2IPv6Addr := tcpip.ProtocolAddress{ + Protocol: ipv6.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("b::2").To16()), + PrefixLen: 64, + }, + } + + type endpointAndAddresses struct { + serverEP tcpip.Endpoint + serverAddr tcpip.Address + serverReadableCH chan struct{} + + clientEP tcpip.Endpoint + clientAddr tcpip.Address + clientReadableCH chan struct{} + } + + newEP := func(t *testing.T, s *stack.Stack, transProto tcpip.TransportProtocolNumber, netProto tcpip.NetworkProtocolNumber) (tcpip.Endpoint, chan struct{}) { + t.Helper() + var wq waiter.Queue + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + ep, err := s.NewEndpoint(transProto, netProto, &wq) + if err != nil { + t.Fatalf("s.NewEndpoint(%d, %d, _): %s", transProto, netProto, err) + } + + t.Cleanup(func() { + wq.EventUnregister(&we) + }) + + return ep, ch + } + + tests := []struct { + name string + epAndAddrs func(t *testing.T, host1Stack, routerStack, host2Stack *stack.Stack) endpointAndAddresses + }{ + { + name: "IPv4 host1 server with host2 client", + epAndAddrs: func(t *testing.T, host1Stack, routerStack, host2Stack *stack.Stack) endpointAndAddresses { + ep1, ep1WECH := newEP(t, host1Stack, udp.ProtocolNumber, ipv4.ProtocolNumber) + ep2, ep2WECH := newEP(t, host2Stack, udp.ProtocolNumber, ipv4.ProtocolNumber) + return endpointAndAddresses{ + serverEP: ep1, + serverAddr: host1IPv4Addr.AddressWithPrefix.Address, + serverReadableCH: ep1WECH, + + clientEP: ep2, + clientAddr: host2IPv4Addr.AddressWithPrefix.Address, + clientReadableCH: ep2WECH, + } + }, + }, + { + name: "IPv6 host2 server with host1 client", + epAndAddrs: func(t *testing.T, host1Stack, routerStack, host2Stack *stack.Stack) endpointAndAddresses { + ep1, ep1WECH := newEP(t, host2Stack, udp.ProtocolNumber, ipv6.ProtocolNumber) + ep2, ep2WECH := newEP(t, host1Stack, udp.ProtocolNumber, ipv6.ProtocolNumber) + return endpointAndAddresses{ + serverEP: ep1, + serverAddr: host2IPv6Addr.AddressWithPrefix.Address, + serverReadableCH: ep1WECH, + + clientEP: ep2, + clientAddr: host1IPv6Addr.AddressWithPrefix.Address, + clientReadableCH: ep2WECH, + } + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + stackOpts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{arp.NewProtocol, ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, + } + + host1Stack := stack.New(stackOpts) + routerStack := stack.New(stackOpts) + host2Stack := stack.New(stackOpts) + + host1NIC, routerNIC1 := pipe.New(host1NICLinkAddr, routerNIC1LinkAddr, stack.CapabilityResolutionRequired) + routerNIC2, host2NIC := pipe.New(routerNIC2LinkAddr, host2NICLinkAddr, stack.CapabilityResolutionRequired) + + if err := host1Stack.CreateNIC(host1NICID, host1NIC); err != nil { + t.Fatalf("host1Stack.CreateNIC(%d, _): %s", host1NICID, err) + } + if err := routerStack.CreateNIC(routerNICID1, routerNIC1); err != nil { + t.Fatalf("routerStack.CreateNIC(%d, _): %s", routerNICID1, err) + } + if err := routerStack.CreateNIC(routerNICID2, routerNIC2); err != nil { + t.Fatalf("routerStack.CreateNIC(%d, _): %s", routerNICID2, err) + } + if err := host2Stack.CreateNIC(host2NICID, host2NIC); err != nil { + t.Fatalf("host2Stack.CreateNIC(%d, _): %s", host2NICID, err) + } + + if err := routerStack.SetForwarding(ipv4.ProtocolNumber, true); err != nil { + t.Fatalf("routerStack.SetForwarding(%d): %s", ipv4.ProtocolNumber, err) + } + if err := routerStack.SetForwarding(ipv6.ProtocolNumber, true); err != nil { + t.Fatalf("routerStack.SetForwarding(%d): %s", ipv6.ProtocolNumber, err) + } + + if err := host1Stack.AddAddress(host1NICID, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + t.Fatalf("host1Stack.AddAddress(%d, %d, %s): %s", host1NICID, arp.ProtocolNumber, arp.ProtocolAddress, err) + } + if err := routerStack.AddAddress(routerNICID1, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + t.Fatalf("routerStack.AddAddress(%d, %d, %s): %s", routerNICID1, arp.ProtocolNumber, arp.ProtocolAddress, err) + } + if err := routerStack.AddAddress(routerNICID2, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + t.Fatalf("routerStack.AddAddress(%d, %d, %s): %s", routerNICID2, arp.ProtocolNumber, arp.ProtocolAddress, err) + } + if err := host2Stack.AddAddress(host2NICID, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + t.Fatalf("host2Stack.AddAddress(%d, %d, %s): %s", host2NICID, arp.ProtocolNumber, arp.ProtocolAddress, err) + } + + if err := host1Stack.AddProtocolAddress(host1NICID, host1IPv4Addr); err != nil { + t.Fatalf("host1Stack.AddProtocolAddress(%d, %#v): %s", host1NICID, host1IPv4Addr, err) + } + if err := routerStack.AddProtocolAddress(routerNICID1, routerNIC1IPv4Addr); err != nil { + t.Fatalf("routerStack.AddProtocolAddress(%d, %#v): %s", routerNICID1, routerNIC1IPv4Addr, err) + } + if err := routerStack.AddProtocolAddress(routerNICID2, routerNIC2IPv4Addr); err != nil { + t.Fatalf("routerStack.AddProtocolAddress(%d, %#v): %s", routerNICID2, routerNIC2IPv4Addr, err) + } + if err := host2Stack.AddProtocolAddress(host2NICID, host2IPv4Addr); err != nil { + t.Fatalf("host2Stack.AddProtocolAddress(%d, %#v): %s", host2NICID, host2IPv4Addr, err) + } + if err := host1Stack.AddProtocolAddress(host1NICID, host1IPv6Addr); err != nil { + t.Fatalf("host1Stack.AddProtocolAddress(%d, %#v): %s", host1NICID, host1IPv6Addr, err) + } + if err := routerStack.AddProtocolAddress(routerNICID1, routerNIC1IPv6Addr); err != nil { + t.Fatalf("routerStack.AddProtocolAddress(%d, %#v): %s", routerNICID1, routerNIC1IPv6Addr, err) + } + if err := routerStack.AddProtocolAddress(routerNICID2, routerNIC2IPv6Addr); err != nil { + t.Fatalf("routerStack.AddProtocolAddress(%d, %#v): %s", routerNICID2, routerNIC2IPv6Addr, err) + } + if err := host2Stack.AddProtocolAddress(host2NICID, host2IPv6Addr); err != nil { + t.Fatalf("host2Stack.AddProtocolAddress(%d, %#v): %s", host2NICID, host2IPv6Addr, err) + } + + host1Stack.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: host1IPv4Addr.AddressWithPrefix.Subnet(), + NIC: host1NICID, + }, + tcpip.Route{ + Destination: host1IPv6Addr.AddressWithPrefix.Subnet(), + NIC: host1NICID, + }, + tcpip.Route{ + Destination: host2IPv4Addr.AddressWithPrefix.Subnet(), + Gateway: routerNIC1IPv4Addr.AddressWithPrefix.Address, + NIC: host1NICID, + }, + tcpip.Route{ + Destination: host2IPv6Addr.AddressWithPrefix.Subnet(), + Gateway: routerNIC1IPv6Addr.AddressWithPrefix.Address, + NIC: host1NICID, + }, + }) + routerStack.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: routerNIC1IPv4Addr.AddressWithPrefix.Subnet(), + NIC: routerNICID1, + }, + tcpip.Route{ + Destination: routerNIC1IPv6Addr.AddressWithPrefix.Subnet(), + NIC: routerNICID1, + }, + tcpip.Route{ + Destination: routerNIC2IPv4Addr.AddressWithPrefix.Subnet(), + NIC: routerNICID2, + }, + tcpip.Route{ + Destination: routerNIC2IPv6Addr.AddressWithPrefix.Subnet(), + NIC: routerNICID2, + }, + }) + host2Stack.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: host2IPv4Addr.AddressWithPrefix.Subnet(), + NIC: host2NICID, + }, + tcpip.Route{ + Destination: host2IPv6Addr.AddressWithPrefix.Subnet(), + NIC: host2NICID, + }, + tcpip.Route{ + Destination: host1IPv4Addr.AddressWithPrefix.Subnet(), + Gateway: routerNIC2IPv4Addr.AddressWithPrefix.Address, + NIC: host2NICID, + }, + tcpip.Route{ + Destination: host1IPv6Addr.AddressWithPrefix.Subnet(), + Gateway: routerNIC2IPv6Addr.AddressWithPrefix.Address, + NIC: host2NICID, + }, + }) + + epsAndAddrs := test.epAndAddrs(t, host1Stack, routerStack, host2Stack) + defer epsAndAddrs.serverEP.Close() + defer epsAndAddrs.clientEP.Close() + + serverAddr := tcpip.FullAddress{Addr: epsAndAddrs.serverAddr, Port: listenPort} + if err := epsAndAddrs.serverEP.Bind(serverAddr); err != nil { + t.Fatalf("epsAndAddrs.serverEP.Bind(%#v): %s", serverAddr, err) + } + clientAddr := tcpip.FullAddress{Addr: epsAndAddrs.clientAddr} + if err := epsAndAddrs.clientEP.Bind(clientAddr); err != nil { + t.Fatalf("epsAndAddrs.clientEP.Bind(%#v): %s", clientAddr, err) + } + + write := func(ep tcpip.Endpoint, data []byte, to *tcpip.FullAddress) { + t.Helper() + + dataPayload := tcpip.SlicePayload(data) + wOpts := tcpip.WriteOptions{To: to} + n, ch, err := ep.Write(dataPayload, wOpts) + if err == tcpip.ErrNoLinkAddress { + // Wait for link resolution to complete. + <-ch + + n, _, err = ep.Write(dataPayload, wOpts) + } else if err != nil { + t.Fatalf("ep.Write(_, _): %s", err) + } + + if err != nil { + t.Fatalf("ep.Write(_, _): %s", err) + } + if want := int64(len(data)); n != want { + t.Fatalf("got ep.Write(_, _) = (%d, _, _), want = (%d, _, _)", n, want) + } + } + + data := []byte{1, 2, 3, 4} + write(epsAndAddrs.clientEP, data, &serverAddr) + + read := func(ch chan struct{}, ep tcpip.Endpoint, data []byte, expectedFrom tcpip.Address) tcpip.FullAddress { + t.Helper() + + // Wait for the endpoint to be readable. + <-ch + + var addr tcpip.FullAddress + v, _, err := ep.Read(&addr) + if err != nil { + t.Fatalf("ep.Read(_): %s", err) + } + + if diff := cmp.Diff(v, buffer.View(data)); diff != "" { + t.Errorf("received data mismatch (-want +got):\n%s", diff) + } + if addr.Addr != expectedFrom { + t.Errorf("got addr.Addr = %s, want = %s", addr.Addr, expectedFrom) + } + + if t.Failed() { + t.FailNow() + } + + return addr + } + + addr := read(epsAndAddrs.serverReadableCH, epsAndAddrs.serverEP, data, epsAndAddrs.clientAddr) + // Unspecify the NIC since NIC IDs are meaningless across stacks. + addr.NIC = 0 + + data = tcpip.SlicePayload([]byte{5, 6, 7, 8, 9, 10, 11, 12}) + write(epsAndAddrs.serverEP, data, &addr) + addr = read(epsAndAddrs.clientReadableCH, epsAndAddrs.clientEP, data, epsAndAddrs.serverAddr) + if addr.Port != listenPort { + t.Errorf("got addr.Port = %d, want = %d", addr.Port, listenPort) + } + }) + } +} diff --git a/pkg/tcpip/tests/integration/link_resolution_test.go b/pkg/tcpip/tests/integration/link_resolution_test.go new file mode 100644 index 000000000..bf3a6f6ee --- /dev/null +++ b/pkg/tcpip/tests/integration/link_resolution_test.go @@ -0,0 +1,219 @@ +// 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 integration_test + +import ( + "net" + "testing" + + "github.com/google/go-cmp/cmp" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/link/pipe" + "gvisor.dev/gvisor/pkg/tcpip/network/arp" + "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/waiter" +) + +var ( + host1NICLinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x06") + host2NICLinkAddr = tcpip.LinkAddress("\x02\x03\x03\x04\x05\x09") + + host1IPv4Addr = tcpip.ProtocolAddress{ + Protocol: ipv4.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("192.168.0.1").To4()), + PrefixLen: 24, + }, + } + host2IPv4Addr = tcpip.ProtocolAddress{ + Protocol: ipv4.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("192.168.0.2").To4()), + PrefixLen: 8, + }, + } + host1IPv6Addr = tcpip.ProtocolAddress{ + Protocol: ipv6.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("a::1").To16()), + PrefixLen: 64, + }, + } + host2IPv6Addr = tcpip.ProtocolAddress{ + Protocol: ipv6.ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: tcpip.Address(net.ParseIP("a::2").To16()), + PrefixLen: 64, + }, + } +) + +// TestPing tests that two hosts can ping eachother when link resolution is +// enabled. +func TestPing(t *testing.T) { + const ( + host1NICID = 1 + host2NICID = 4 + + // icmpDataOffset is the offset to the data in both ICMPv4 and ICMPv6 echo + // request/reply packets. + icmpDataOffset = 8 + ) + + tests := []struct { + name string + transProto tcpip.TransportProtocolNumber + netProto tcpip.NetworkProtocolNumber + remoteAddr tcpip.Address + icmpBuf func(*testing.T) buffer.View + }{ + { + name: "IPv4 Ping", + transProto: icmp.ProtocolNumber4, + netProto: ipv4.ProtocolNumber, + remoteAddr: host2IPv4Addr.AddressWithPrefix.Address, + icmpBuf: func(t *testing.T) buffer.View { + data := [8]byte{1, 2, 3, 4, 5, 6, 7, 8} + hdr := header.ICMPv4(make([]byte, header.ICMPv4MinimumSize+len(data))) + hdr.SetType(header.ICMPv4Echo) + if n := copy(hdr.Payload(), data[:]); n != len(data) { + t.Fatalf("copied %d bytes but expected to copy %d bytes", n, len(data)) + } + return buffer.View(hdr) + }, + }, + { + name: "IPv6 Ping", + transProto: icmp.ProtocolNumber6, + netProto: ipv6.ProtocolNumber, + remoteAddr: host2IPv6Addr.AddressWithPrefix.Address, + icmpBuf: func(t *testing.T) buffer.View { + data := [8]byte{1, 2, 3, 4, 5, 6, 7, 8} + hdr := header.ICMPv6(make([]byte, header.ICMPv6MinimumSize+len(data))) + hdr.SetType(header.ICMPv6EchoRequest) + if n := copy(hdr.Payload(), data[:]); n != len(data) { + t.Fatalf("copied %d bytes but expected to copy %d bytes", n, len(data)) + } + return buffer.View(hdr) + }, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + stackOpts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{arp.NewProtocol, ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol4, icmp.NewProtocol6}, + } + + host1Stack := stack.New(stackOpts) + host2Stack := stack.New(stackOpts) + + host1NIC, host2NIC := pipe.New(host1NICLinkAddr, host2NICLinkAddr, stack.CapabilityResolutionRequired) + + if err := host1Stack.CreateNIC(host1NICID, host1NIC); err != nil { + t.Fatalf("host1Stack.CreateNIC(%d, _): %s", host1NICID, err) + } + if err := host2Stack.CreateNIC(host2NICID, host2NIC); err != nil { + t.Fatalf("host2Stack.CreateNIC(%d, _): %s", host2NICID, err) + } + + if err := host1Stack.AddAddress(host1NICID, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + t.Fatalf("host1Stack.AddAddress(%d, %d, %s): %s", host1NICID, arp.ProtocolNumber, arp.ProtocolAddress, err) + } + if err := host2Stack.AddAddress(host2NICID, arp.ProtocolNumber, arp.ProtocolAddress); err != nil { + t.Fatalf("host2Stack.AddAddress(%d, %d, %s): %s", host2NICID, arp.ProtocolNumber, arp.ProtocolAddress, err) + } + + if err := host1Stack.AddProtocolAddress(host1NICID, host1IPv4Addr); err != nil { + t.Fatalf("host1Stack.AddProtocolAddress(%d, %#v): %s", host1NICID, host1IPv4Addr, err) + } + if err := host2Stack.AddProtocolAddress(host2NICID, host2IPv4Addr); err != nil { + t.Fatalf("host2Stack.AddProtocolAddress(%d, %#v): %s", host2NICID, host2IPv4Addr, err) + } + if err := host1Stack.AddProtocolAddress(host1NICID, host1IPv6Addr); err != nil { + t.Fatalf("host1Stack.AddProtocolAddress(%d, %#v): %s", host1NICID, host1IPv6Addr, err) + } + if err := host2Stack.AddProtocolAddress(host2NICID, host2IPv6Addr); err != nil { + t.Fatalf("host2Stack.AddProtocolAddress(%d, %#v): %s", host2NICID, host2IPv6Addr, err) + } + + host1Stack.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: host1IPv4Addr.AddressWithPrefix.Subnet(), + NIC: host1NICID, + }, + tcpip.Route{ + Destination: host1IPv6Addr.AddressWithPrefix.Subnet(), + NIC: host1NICID, + }, + }) + host2Stack.SetRouteTable([]tcpip.Route{ + tcpip.Route{ + Destination: host2IPv4Addr.AddressWithPrefix.Subnet(), + NIC: host2NICID, + }, + tcpip.Route{ + Destination: host2IPv6Addr.AddressWithPrefix.Subnet(), + NIC: host2NICID, + }, + }) + + var wq waiter.Queue + we, waiterCH := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + ep, err := host1Stack.NewEndpoint(test.transProto, test.netProto, &wq) + if err != nil { + t.Fatalf("host1Stack.NewEndpoint(%d, %d, _): %s", test.transProto, test.netProto, err) + } + defer ep.Close() + + // The first write should trigger link resolution. + icmpBuf := test.icmpBuf(t) + wOpts := tcpip.WriteOptions{To: &tcpip.FullAddress{Addr: test.remoteAddr}} + if _, ch, err := ep.Write(tcpip.SlicePayload(icmpBuf), wOpts); err != tcpip.ErrNoLinkAddress { + t.Fatalf("got ep.Write(_, _) = %s, want = %s", err, tcpip.ErrNoLinkAddress) + } else { + // Wait for link resolution to complete. + <-ch + } + if n, _, err := ep.Write(tcpip.SlicePayload(icmpBuf), wOpts); err != nil { + t.Fatalf("ep.Write(_, _): %s", err) + } else if want := int64(len(icmpBuf)); n != want { + t.Fatalf("got ep.Write(_, _) = (%d, _, _), want = (%d, _, _)", n, want) + } + + // Wait for the endpoint to be readable. + <-waiterCH + + var addr tcpip.FullAddress + v, _, err := ep.Read(&addr) + if err != nil { + t.Fatalf("ep.Read(_): %s", err) + } + if diff := cmp.Diff(v[icmpDataOffset:], icmpBuf[icmpDataOffset:]); diff != "" { + t.Errorf("received data mismatch (-want +got):\n%s", diff) + } + if addr.Addr != test.remoteAddr { + t.Errorf("got addr.Addr = %s, want = %s", addr.Addr, test.remoteAddr) + } + }) + } +} diff --git a/pkg/tcpip/tests/integration/loopback_test.go b/pkg/tcpip/tests/integration/loopback_test.go index fecbe7ba7..e8caf09ba 100644 --- a/pkg/tcpip/tests/integration/loopback_test.go +++ b/pkg/tcpip/tests/integration/loopback_test.go @@ -16,6 +16,7 @@ package integration_test import ( "testing" + "time" "github.com/google/go-cmp/cmp" "gvisor.dev/gvisor/pkg/tcpip" @@ -29,6 +30,69 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +var _ ipv6.NDPDispatcher = (*ndpDispatcher)(nil) + +type ndpDispatcher struct{} + +func (*ndpDispatcher) OnDuplicateAddressDetectionStatus(tcpip.NICID, tcpip.Address, bool, *tcpip.Error) { +} + +func (*ndpDispatcher) OnDefaultRouterDiscovered(tcpip.NICID, tcpip.Address) bool { + return false +} + +func (*ndpDispatcher) OnDefaultRouterInvalidated(tcpip.NICID, tcpip.Address) {} + +func (*ndpDispatcher) OnOnLinkPrefixDiscovered(tcpip.NICID, tcpip.Subnet) bool { + return false +} + +func (*ndpDispatcher) OnOnLinkPrefixInvalidated(tcpip.NICID, tcpip.Subnet) {} + +func (*ndpDispatcher) OnAutoGenAddress(tcpip.NICID, tcpip.AddressWithPrefix) bool { + return true +} + +func (*ndpDispatcher) OnAutoGenAddressDeprecated(tcpip.NICID, tcpip.AddressWithPrefix) {} + +func (*ndpDispatcher) OnAutoGenAddressInvalidated(tcpip.NICID, tcpip.AddressWithPrefix) {} + +func (*ndpDispatcher) OnRecursiveDNSServerOption(tcpip.NICID, []tcpip.Address, time.Duration) {} + +func (*ndpDispatcher) OnDNSSearchListOption(tcpip.NICID, []string, time.Duration) {} + +func (*ndpDispatcher) OnDHCPv6Configuration(tcpip.NICID, ipv6.DHCPv6ConfigurationFromNDPRA) {} + +// TestInitialLoopbackAddresses tests that the loopback interface does not +// auto-generate a link-local address when it is brought up. +func TestInitialLoopbackAddresses(t *testing.T) { + const nicID = 1 + + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocolWithOptions(ipv6.Options{ + NDPDisp: &ndpDispatcher{}, + AutoGenIPv6LinkLocal: 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) + return "" + }, + }, + })}, + }) + + if err := s.CreateNIC(nicID, loopback.New()); err != nil { + t.Fatalf("CreateNIC(%d, _): %s", nicID, err) + } + + nicsInfo := s.NICInfo() + if nicInfo, ok := nicsInfo[nicID]; !ok { + t.Fatalf("did not find NIC with ID = %d in s.NICInfo() = %#v", nicID, nicsInfo) + } else if got := len(nicInfo.ProtocolAddresses); got != 0 { + t.Fatalf("got len(nicInfo.ProtocolAddresses) = %d, want = 0; nicInfo.ProtocolAddresses = %#v", got, nicInfo.ProtocolAddresses) + } +} + // TestLoopbackAcceptAllInSubnet tests that a loopback interface considers // itself bound to all addresses in the subnet of an assigned address. func TestLoopbackAcceptAllInSubnet(t *testing.T) { @@ -120,8 +184,8 @@ func TestLoopbackAcceptAllInSubnet(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) if err := s.CreateNIC(nicID, loopback.New()); err != nil { t.Fatalf("CreateNIC(%d, _): %s", nicID, err) @@ -203,7 +267,7 @@ func TestLoopbackSubnetLifetimeBoundToAddr(t *testing.T) { otherAddr := tcpip.Address(addrBytes) s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, }) if err := s.CreateNIC(nicID, loopback.New()); err != nil { t.Fatalf("s.CreateNIC(%d, _): %s", nicID, err) diff --git a/pkg/tcpip/tests/integration/multicast_broadcast_test.go b/pkg/tcpip/tests/integration/multicast_broadcast_test.go index 659acbc7a..4f2ca7f54 100644 --- a/pkg/tcpip/tests/integration/multicast_broadcast_test.go +++ b/pkg/tcpip/tests/integration/multicast_broadcast_test.go @@ -140,11 +140,9 @@ func TestPingMulticastBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - ipv4Proto := ipv4.NewProtocol() - ipv6Proto := ipv6.NewProtocol() s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4Proto, ipv6Proto}, - TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4(), icmp.NewProtocol6()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{icmp.NewProtocol4, icmp.NewProtocol6}, }) // We only expect a single packet in response to our ICMP Echo Request. e := channel.New(1, defaultMTU, "") @@ -176,18 +174,18 @@ func TestPingMulticastBroadcast(t *testing.T) { var rxICMP func(*channel.Endpoint, tcpip.Address) var expectedSrc tcpip.Address var expectedDst tcpip.Address - var proto stack.NetworkProtocol + var protoNum tcpip.NetworkProtocolNumber switch l := len(test.dstAddr); l { case header.IPv4AddressSize: rxICMP = rxIPv4ICMP expectedSrc = ipv4Addr.Address expectedDst = remoteIPv4Addr - proto = ipv4Proto + protoNum = header.IPv4ProtocolNumber case header.IPv6AddressSize: rxICMP = rxIPv6ICMP expectedSrc = ipv6Addr.Address expectedDst = remoteIPv6Addr - proto = ipv6Proto + protoNum = header.IPv6ProtocolNumber default: t.Fatalf("got unexpected address length = %d bytes", l) } @@ -205,7 +203,7 @@ func TestPingMulticastBroadcast(t *testing.T) { t.Errorf("got pkt.Route.RemoteAddress = %s, want = %s", pkt.Route.RemoteAddress, expectedDst) } - src, dst := proto.ParseAddresses(pkt.Pkt.NetworkHeader().View()) + src, dst := s.NetworkProtocolInstance(protoNum).ParseAddresses(stack.PayloadSince(pkt.Pkt.NetworkHeader())) if src != expectedSrc { t.Errorf("got pkt source = %s, want = %s", src, expectedSrc) } @@ -380,8 +378,8 @@ func TestIncomingMulticastAndBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, defaultMTU, "") if err := s.CreateNIC(nicID, e); err != nil { @@ -466,8 +464,8 @@ func TestReuseAddrAndBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) if err := s.CreateNIC(nicID, loopback.New()); err != nil { t.Fatalf("CreateNIC(%d, _): %s", nicID, err) diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 31116309e..41eb0ca44 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -446,6 +446,7 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpi pkt.Owner = owner icmpv4 := header.ICMPv4(pkt.TransportHeader().Push(header.ICMPv4MinimumSize)) + pkt.TransportProtocolNumber = header.ICMPv4ProtocolNumber copy(icmpv4, data) // Set the ident to the user-specified port. Sequence number should // already be set by the user. @@ -478,6 +479,7 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err }) icmpv6 := header.ICMPv6(pkt.TransportHeader().Push(header.ICMPv6MinimumSize)) + pkt.TransportProtocolNumber = header.ICMPv6ProtocolNumber copy(icmpv6, data) // Set the ident. Sequence number is provided by the user. icmpv6.SetIdent(ident) diff --git a/pkg/tcpip/transport/icmp/protocol.go b/pkg/tcpip/transport/icmp/protocol.go index bb11e4e83..87d510f96 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -13,12 +13,7 @@ // limitations under the License. // Package icmp contains the implementation of the ICMP and IPv6-ICMP transport -// protocols for use in ping. To use it in the networking stack, this package -// must be added to the project, and activated on the stack by passing -// icmp.NewProtocol4() and/or icmp.NewProtocol6() as one of the transport -// protocols when calling stack.New(). Then endpoints can be created by passing -// icmp.ProtocolNumber or icmp.ProtocolNumber6 as the transport protocol number -// when calling Stack.NewEndpoint(). +// protocols for use in ping. package icmp import ( @@ -42,6 +37,8 @@ const ( // protocol implements stack.TransportProtocol. type protocol struct { + stack *stack.Stack + number tcpip.TransportProtocolNumber } @@ -62,20 +59,20 @@ func (p *protocol) netProto() tcpip.NetworkProtocolNumber { // NewEndpoint creates a new icmp endpoint. It implements // stack.TransportProtocol.NewEndpoint. -func (p *protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (p *protocol) NewEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { if netProto != p.netProto() { return nil, tcpip.ErrUnknownProtocol } - return newEndpoint(stack, netProto, p.number, waiterQueue) + return newEndpoint(p.stack, netProto, p.number, waiterQueue) } // NewRawEndpoint creates a new raw icmp endpoint. It implements // stack.TransportProtocol.NewRawEndpoint. -func (p *protocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (p *protocol) NewRawEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { if netProto != p.netProto() { return nil, tcpip.ErrUnknownProtocol } - return raw.NewEndpoint(stack, netProto, p.number, waiterQueue) + return raw.NewEndpoint(p.stack, netProto, p.number, waiterQueue) } // MinimumPacketSize returns the minimum valid icmp packet size. @@ -104,8 +101,8 @@ func (p *protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. -func (*protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) bool { - return true +func (*protocol) HandleUnknownDestinationPacket(*stack.Route, stack.TransportEndpointID, *stack.PacketBuffer) stack.UnknownDestinationPacketDisposition { + return stack.UnknownDestinationPacketHandled } // SetOption implements stack.TransportProtocol.SetOption. @@ -135,11 +132,11 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { } // NewProtocol4 returns an ICMPv4 transport protocol. -func NewProtocol4() stack.TransportProtocol { - return &protocol{ProtocolNumber4} +func NewProtocol4(s *stack.Stack) stack.TransportProtocol { + return &protocol{stack: s, number: ProtocolNumber4} } // NewProtocol6 returns an ICMPv6 transport protocol. -func NewProtocol6() stack.TransportProtocol { - return &protocol{ProtocolNumber6} +func NewProtocol6(s *stack.Stack) stack.TransportProtocol { + return &protocol{stack: s, number: ProtocolNumber6} } diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 4778e7b1c..518449602 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -94,6 +94,7 @@ go_test( shard_count = 10, deps = [ ":tcp", + "//pkg/rand", "//pkg/sync", "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 09d53d158..189c01c8f 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -747,6 +747,7 @@ func (e *endpoint) sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedV func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *stack.GSO) { optLen := len(tf.opts) tcp := header.TCP(pkt.TransportHeader().Push(header.TCPMinimumSize + optLen)) + pkt.TransportProtocolNumber = header.TCPProtocolNumber tcp.Encode(&header.TCPFields{ SrcPort: tf.id.LocalPort, DstPort: tf.id.RemotePort, @@ -803,7 +804,7 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso pkt.Owner = owner pkt.EgressRoute = r pkt.GSOOptions = gso - pkt.NetworkProtocolNumber = r.NetworkProtocolNumber() + pkt.NetworkProtocolNumber = r.NetProto data.ReadToVV(&pkt.Data, packetSize) buildTCPHdr(r, tf, pkt, gso) tf.seq = tf.seq.Add(seqnum.Size(packetSize)) @@ -897,7 +898,7 @@ func (e *endpoint) makeOptions(sackBlocks []header.SACKBlock) []byte { // sendRaw sends a TCP segment to the endpoint's peer. func (e *endpoint) sendRaw(data buffer.VectorisedView, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size) *tcpip.Error { var sackBlocks []header.SACKBlock - if e.EndpointState() == StateEstablished && e.rcv.pendingBufSize > 0 && (flags&header.TCPFlagAck != 0) { + if e.EndpointState() == StateEstablished && e.rcv.pendingRcvdSegments.Len() > 0 && (flags&header.TCPFlagAck != 0) { sackBlocks = e.sack.Blocks[:e.sack.NumBlocks] } options := e.makeOptions(sackBlocks) @@ -1002,9 +1003,8 @@ func (e *endpoint) transitionToStateEstablishedLocked(h *handshake) { // (indicated by a negative send window scale). e.snd = newSender(e, h.iss, h.ackNum-1, h.sndWnd, h.mss, h.sndWndScale) - rcvBufSize := seqnum.Size(e.receiveBufferSize()) e.rcvListMu.Lock() - e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale(), rcvBufSize) + e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale()) // Bootstrap the auto tuning algorithm. Starting at zero will // result in a really large receive window after the first auto // tuning adjustment. @@ -1135,12 +1135,11 @@ func (e *endpoint) handleSegments(fastPath bool) *tcpip.Error { } cont, err := e.handleSegment(s) + s.decRef() if err != nil { - s.decRef() return err } if !cont { - s.decRef() return nil } } @@ -1220,6 +1219,12 @@ func (e *endpoint) handleSegment(s *segment) (cont bool, err *tcpip.Error) { return true, nil } + // Increase counter if after processing the segment we would potentially + // advertise a zero window. + if crossed, above := e.windowCrossedACKThresholdLocked(-s.segMemSize()); crossed && !above { + e.stats.ReceiveErrors.ZeroRcvWindowState.Increment() + } + // Now check if the received segment has caused us to transition // to a CLOSED state, if yes then terminate processing and do // not invoke the sender. @@ -1232,7 +1237,6 @@ func (e *endpoint) handleSegment(s *segment) (cont bool, err *tcpip.Error) { // or a notification from the protocolMainLoop (caller goroutine). // This means that with this return, the segment dequeue below can // never occur on a closed endpoint. - s.decRef() return false, nil } @@ -1424,10 +1428,6 @@ func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{ e.rcv.nonZeroWindow() } - if n¬ifyReceiveWindowChanged != 0 { - e.rcv.pendingBufSize = seqnum.Size(e.receiveBufferSize()) - } - if n¬ifyMTUChanged != 0 { e.sndBufMu.Lock() count := e.packetTooBigCount diff --git a/pkg/tcpip/transport/tcp/dual_stack_test.go b/pkg/tcpip/transport/tcp/dual_stack_test.go index 94207c141..560b4904c 100644 --- a/pkg/tcpip/transport/tcp/dual_stack_test.go +++ b/pkg/tcpip/transport/tcp/dual_stack_test.go @@ -78,8 +78,8 @@ func testV4Connect(t *testing.T, c *context.Context, checkers ...checker.Network ackCheckers := append(checkers, checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+1), )) checker.IPv4(t, c.GetPacket(), ackCheckers...) @@ -185,8 +185,8 @@ func testV6Connect(t *testing.T, c *context.Context, checkers ...checker.Network ackCheckers := append(checkers, checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+1), )) checker.IPv6(t, c.GetV6Packet(), ackCheckers...) @@ -283,7 +283,7 @@ func TestV4RefuseOnV6Only(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.AckNum(uint32(irs)+1), + checker.TCPAckNum(uint32(irs)+1), ), ) } @@ -319,7 +319,7 @@ func TestV6RefuseOnBoundToV4Mapped(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.AckNum(uint32(irs)+1), + checker.TCPAckNum(uint32(irs)+1), ), ) } @@ -352,7 +352,7 @@ func testV4Accept(t *testing.T, c *context.Context) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1), + checker.TCPAckNum(uint32(irs)+1), ), ) @@ -492,7 +492,7 @@ func TestV6AcceptOnV6(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1), + checker.TCPAckNum(uint32(irs)+1), ), ) diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 120483838..ae817091a 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -63,6 +63,17 @@ const ( StateClosing ) +const ( + // rcvAdvWndScale is used to split the available socket buffer into + // application buffer and the window to be advertised to the peer. This is + // currently hard coded to split the available space equally. + rcvAdvWndScale = 1 + + // SegOverheadFactor is used to multiply the value provided by the + // user on a SetSockOpt for setting the socket send/receive buffer sizes. + SegOverheadFactor = 2 +) + // connected returns true when s is one of the states representing an // endpoint connected to a peer. func (s EndpointState) connected() bool { @@ -149,7 +160,6 @@ func (s EndpointState) String() string { // Reasons for notifying the protocol goroutine. const ( notifyNonZeroReceiveWindow = 1 << iota - notifyReceiveWindowChanged notifyClose notifyMTUChanged notifyDrain @@ -384,13 +394,26 @@ type endpoint struct { // to indicate to users that no more data is coming. // // rcvListMu can be taken after the endpoint mu below. - rcvListMu sync.Mutex `state:"nosave"` - rcvList segmentList `state:"wait"` - rcvClosed bool - rcvBufSize int + rcvListMu sync.Mutex `state:"nosave"` + rcvList segmentList `state:"wait"` + rcvClosed bool + // rcvBufSize is the total size of the receive buffer. + rcvBufSize int + // rcvBufUsed is the actual number of payload bytes held in the receive buffer + // not counting any overheads of the segments itself. NOTE: This will always + // be strictly <= rcvMemUsed below. rcvBufUsed int rcvAutoParams rcvBufAutoTuneParams + // rcvMemUsed tracks the total amount of memory in use by received segments + // held in rcvList, pendingRcvdSegments and the segment queue. This is used to + // compute the window and the actual available buffer space. This is distinct + // from rcvBufUsed above which is the actual number of payload bytes held in + // the buffer not including any segment overheads. + // + // rcvMemUsed must be accessed atomically. + rcvMemUsed int32 + // mu protects all endpoint fields unless documented otherwise. mu must // be acquired before interacting with the endpoint fields. mu sync.Mutex `state:"nosave"` @@ -891,7 +914,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue e.probe = p } - e.segmentQueue.setLimit(MaxUnprocessedSegments) + e.segmentQueue.ep = e e.tsOffset = timeStampOffset() e.acceptCond = sync.NewCond(&e.acceptMu) @@ -904,7 +927,12 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { result := waiter.EventMask(0) switch e.EndpointState() { - case StateInitial, StateBound, StateConnecting, StateSynSent, StateSynRecv: + case StateInitial, StateBound: + // This prevents blocking of new sockets which are not + // connected when SO_LINGER is set. + result |= waiter.EventHUp + + case StateConnecting, StateSynSent, StateSynRecv: // Ready for nothing. case StateClose, StateError, StateTimeWait: @@ -1075,6 +1103,8 @@ func (e *endpoint) closeNoShutdownLocked() { e.notifyProtocolGoroutine(notifyClose) } else { e.transitionToStateCloseLocked() + // Notify that the endpoint is closed. + e.waiterQueue.Notify(waiter.EventHUp) } } @@ -1129,10 +1159,16 @@ func (e *endpoint) cleanupLocked() { tcpip.DeleteDanglingEndpoint(e) } +// wndFromSpace returns the window that we can advertise based on the available +// receive buffer space. +func wndFromSpace(space int) int { + return space / (1 << rcvAdvWndScale) +} + // initialReceiveWindow returns the initial receive window to advertise in the // SYN/SYN-ACK. func (e *endpoint) initialReceiveWindow() int { - rcvWnd := e.receiveBufferAvailable() + rcvWnd := wndFromSpace(e.receiveBufferAvailable()) if rcvWnd > math.MaxUint16 { rcvWnd = math.MaxUint16 } @@ -1209,14 +1245,12 @@ func (e *endpoint) ModerateRecvBuf(copied int) { // reject valid data that might already be in flight as the // acceptable window will shrink. if rcvWnd > e.rcvBufSize { - availBefore := e.receiveBufferAvailableLocked() + availBefore := wndFromSpace(e.receiveBufferAvailableLocked()) e.rcvBufSize = rcvWnd - availAfter := e.receiveBufferAvailableLocked() - mask := uint32(notifyReceiveWindowChanged) + availAfter := wndFromSpace(e.receiveBufferAvailableLocked()) if crossed, above := e.windowCrossedACKThresholdLocked(availAfter - availBefore); crossed && above { - mask |= notifyNonZeroReceiveWindow + e.notifyProtocolGoroutine(notifyNonZeroReceiveWindow) } - e.notifyProtocolGoroutine(mask) } // We only update prevCopied when we grow the buffer because in cases @@ -1293,18 +1327,22 @@ func (e *endpoint) readLocked() (buffer.View, *tcpip.Error) { v := views[s.viewToDeliver] s.viewToDeliver++ + var delta int if s.viewToDeliver >= len(views) { e.rcvList.Remove(s) + // We only free up receive buffer space when the segment is released as the + // segment is still holding on to the views even though some views have been + // read out to the user. + delta = s.segMemSize() s.decRef() } e.rcvBufUsed -= len(v) - // If the window was small before this read and if the read freed up // enough buffer space, to either fit an aMSS or half a receive buffer // (whichever smaller), then notify the protocol goroutine to send a // window update. - if crossed, above := e.windowCrossedACKThresholdLocked(len(v)); crossed && above { + if crossed, above := e.windowCrossedACKThresholdLocked(delta); crossed && above { e.notifyProtocolGoroutine(notifyNonZeroReceiveWindow) } @@ -1481,11 +1519,11 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro } // windowCrossedACKThresholdLocked checks if the receive window to be announced -// now would be under aMSS or under half receive buffer, whichever smaller. This -// is useful as a receive side silly window syndrome prevention mechanism. If -// window grows to reasonable value, we should send ACK to the sender to inform -// the rx space is now large. We also want ensure a series of small read()'s -// won't trigger a flood of spurious tiny ACK's. +// would be under aMSS or under the window derived from half receive buffer, +// whichever smaller. This is useful as a receive side silly window syndrome +// prevention mechanism. If window grows to reasonable value, we should send ACK +// to the sender to inform the rx space is now large. We also want ensure a +// series of small read()'s won't trigger a flood of spurious tiny ACK's. // // For large receive buffers, the threshold is aMSS - once reader reads more // than aMSS we'll send ACK. For tiny receive buffers, the threshold is half of @@ -1496,17 +1534,18 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro // // Precondition: e.mu and e.rcvListMu must be held. func (e *endpoint) windowCrossedACKThresholdLocked(deltaBefore int) (crossed bool, above bool) { - newAvail := e.receiveBufferAvailableLocked() + newAvail := wndFromSpace(e.receiveBufferAvailableLocked()) oldAvail := newAvail - deltaBefore if oldAvail < 0 { oldAvail = 0 } - threshold := int(e.amss) - if threshold > e.rcvBufSize/2 { - threshold = e.rcvBufSize / 2 + // rcvBufFraction is the inverse of the fraction of receive buffer size that + // is used to decide if the available buffer space is now above it. + const rcvBufFraction = 2 + if wndThreshold := wndFromSpace(e.rcvBufSize / rcvBufFraction); threshold > wndThreshold { + threshold = wndThreshold } - switch { case oldAvail < threshold && newAvail >= threshold: return true, true @@ -1636,17 +1675,23 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { // Make sure the receive buffer size is within the min and max // allowed. var rs tcpip.TCPReceiveBufferSizeRangeOption - if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil { + if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { + panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %#v) = %s", ProtocolNumber, &rs, err)) + } + + if v > rs.Max { + v = rs.Max + } + + if v < math.MaxInt32/SegOverheadFactor { + v *= SegOverheadFactor if v < rs.Min { v = rs.Min } - if v > rs.Max { - v = rs.Max - } + } else { + v = math.MaxInt32 } - mask := uint32(notifyReceiveWindowChanged) - e.LockUser() e.rcvListMu.Lock() @@ -1660,14 +1705,9 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { v = 1 << scale } - // Make sure 2*size doesn't overflow. - if v > math.MaxInt32/2 { - v = math.MaxInt32 / 2 - } - - availBefore := e.receiveBufferAvailableLocked() + availBefore := wndFromSpace(e.receiveBufferAvailableLocked()) e.rcvBufSize = v - availAfter := e.receiveBufferAvailableLocked() + availAfter := wndFromSpace(e.receiveBufferAvailableLocked()) e.rcvAutoParams.disabled = true @@ -1675,24 +1715,31 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { // syndrome prevetion, when our available space grows above aMSS // or half receive buffer, whichever smaller. if crossed, above := e.windowCrossedACKThresholdLocked(availAfter - availBefore); crossed && above { - mask |= notifyNonZeroReceiveWindow + e.notifyProtocolGoroutine(notifyNonZeroReceiveWindow) } e.rcvListMu.Unlock() e.UnlockUser() - e.notifyProtocolGoroutine(mask) case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max // allowed. var ss tcpip.TCPSendBufferSizeRangeOption - if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil { + if err := e.stack.TransportProtocolOption(ProtocolNumber, &ss); err != nil { + panic(fmt.Sprintf("e.stack.TransportProtocolOption(%d, %#v) = %s", ProtocolNumber, &ss, err)) + } + + if v > ss.Max { + v = ss.Max + } + + if v < math.MaxInt32/SegOverheadFactor { + v *= SegOverheadFactor if v < ss.Min { v = ss.Min } - if v > ss.Max { - v = ss.Max - } + } else { + v = math.MaxInt32 } e.sndBufMu.Lock() @@ -2052,7 +2099,7 @@ func (e *endpoint) GetSockOpt(opt tcpip.GettableSocketOption) *tcpip.Error { case *tcpip.OriginalDestinationOption: e.LockUser() ipt := e.stack.IPTables() - addr, port, err := ipt.OriginalDst(e.ID) + addr, port, err := ipt.OriginalDst(e.ID, e.NetProto) e.UnlockUser() if err != nil { return err @@ -2699,13 +2746,8 @@ func (e *endpoint) updateSndBufferUsage(v int) { func (e *endpoint) readyToRead(s *segment) { e.rcvListMu.Lock() if s != nil { + e.rcvBufUsed += s.payloadSize() s.incRef() - e.rcvBufUsed += s.data.Size() - // Increase counter if the receive window falls down below MSS - // or half receive buffer size, whichever smaller. - if crossed, above := e.windowCrossedACKThresholdLocked(-s.data.Size()); crossed && !above { - e.stats.ReceiveErrors.ZeroRcvWindowState.Increment() - } e.rcvList.PushBack(s) } else { e.rcvClosed = true @@ -2720,15 +2762,17 @@ func (e *endpoint) readyToRead(s *segment) { func (e *endpoint) receiveBufferAvailableLocked() int { // We may use more bytes than the buffer size when the receive buffer // shrinks. - if e.rcvBufUsed >= e.rcvBufSize { + memUsed := e.receiveMemUsed() + if memUsed >= e.rcvBufSize { return 0 } - return e.rcvBufSize - e.rcvBufUsed + return e.rcvBufSize - memUsed } // receiveBufferAvailable calculates how many bytes are still available in the -// receive buffer. +// receive buffer based on the actual memory used by all segments held in +// receive buffer/pending and segment queue. func (e *endpoint) receiveBufferAvailable() int { e.rcvListMu.Lock() available := e.receiveBufferAvailableLocked() @@ -2736,14 +2780,35 @@ func (e *endpoint) receiveBufferAvailable() int { return available } +// receiveBufferUsed returns the amount of in-use receive buffer. +func (e *endpoint) receiveBufferUsed() int { + e.rcvListMu.Lock() + used := e.rcvBufUsed + e.rcvListMu.Unlock() + return used +} + +// receiveBufferSize returns the current size of the receive buffer. func (e *endpoint) receiveBufferSize() int { e.rcvListMu.Lock() size := e.rcvBufSize e.rcvListMu.Unlock() - return size } +// receiveMemUsed returns the total memory in use by segments held by this +// endpoint. +func (e *endpoint) receiveMemUsed() int { + return int(atomic.LoadInt32(&e.rcvMemUsed)) +} + +// updateReceiveMemUsed adds the provided delta to e.rcvMemUsed. +func (e *endpoint) updateReceiveMemUsed(delta int) { + atomic.AddInt32(&e.rcvMemUsed, int32(delta)) +} + +// maxReceiveBufferSize returns the stack wide maximum receive buffer size for +// an endpoint. func (e *endpoint) maxReceiveBufferSize() int { var rs tcpip.TCPReceiveBufferSizeRangeOption if err := e.stack.TransportProtocolOption(ProtocolNumber, &rs); err != nil { @@ -2894,7 +2959,6 @@ func (e *endpoint) completeState() stack.TCPEndpointState { RcvAcc: e.rcv.rcvAcc, RcvWndScale: e.rcv.rcvWndScale, PendingBufUsed: e.rcv.pendingBufUsed, - PendingBufSize: e.rcv.pendingBufSize, } // Copy sender state. diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 41d0050f3..b25431467 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -44,7 +44,7 @@ func (e *endpoint) drainSegmentLocked() { // beforeSave is invoked by stateify. func (e *endpoint) beforeSave() { // Stop incoming packets. - e.segmentQueue.setLimit(0) + e.segmentQueue.freeze() e.mu.Lock() defer e.mu.Unlock() @@ -178,7 +178,7 @@ func (e *endpoint) afterLoad() { // Resume implements tcpip.ResumableEndpoint.Resume. func (e *endpoint) Resume(s *stack.Stack) { e.stack = s - e.segmentQueue.setLimit(MaxUnprocessedSegments) + e.segmentQueue.thaw() epState := e.origEndpointState switch epState { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 74a17af79..5bce73605 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -12,12 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package tcp contains the implementation of the TCP transport protocol. To use -// it in the networking stack, this package must be added to the project, and -// activated on the stack by passing tcp.NewProtocol() as one of the -// transport protocols when calling stack.New(). Then endpoints can be created -// by passing tcp.ProtocolNumber as the transport protocol number when calling -// Stack.NewEndpoint(). +// Package tcp contains the implementation of the TCP transport protocol. package tcp import ( @@ -138,6 +133,8 @@ func (s *synRcvdCounter) Threshold() uint64 { } type protocol struct { + stack *stack.Stack + mu sync.RWMutex sackEnabled bool recovery tcpip.TCPRecovery @@ -164,14 +161,14 @@ func (*protocol) Number() tcpip.TransportProtocolNumber { } // NewEndpoint creates a new tcp endpoint. -func (p *protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return newEndpoint(stack, netProto, waiterQueue), nil +func (p *protocol) NewEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return newEndpoint(p.stack, netProto, waiterQueue), nil } // NewRawEndpoint creates a new raw TCP endpoint. Raw TCP sockets are currently // unsupported. It implements stack.TransportProtocol.NewRawEndpoint. -func (p *protocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return raw.NewEndpoint(stack, netProto, header.TCPProtocolNumber, waiterQueue) +func (p *protocol) NewRawEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return raw.NewEndpoint(p.stack, netProto, header.TCPProtocolNumber, waiterQueue) } // MinimumPacketSize returns the minimum valid tcp packet size. @@ -201,21 +198,20 @@ func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id st // a reset is sent in response to any incoming segment except another reset. In // particular, SYNs addressed to a non-existent connection are rejected by this // means." -func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { + +func (*protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) stack.UnknownDestinationPacketDisposition { s := newSegment(r, id, pkt) defer s.decRef() if !s.parse() || !s.csumValid { - return false + return stack.UnknownDestinationPacketMalformed } - // There's nothing to do if this is already a reset packet. - if s.flagIsSet(header.TCPFlagRst) { - return true + if !s.flagIsSet(header.TCPFlagRst) { + replyWithReset(s, stack.DefaultTOS, s.route.DefaultTTL()) } - replyWithReset(s, stack.DefaultTOS, s.route.DefaultTTL()) - return true + return stack.UnknownDestinationPacketHandled } // replyWithReset replies to the given segment with a reset segment. @@ -511,8 +507,9 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { } // NewProtocol returns a TCP transport protocol. -func NewProtocol() stack.TransportProtocol { +func NewProtocol(s *stack.Stack) stack.TransportProtocol { p := protocol{ + stack: s, sendBufferSize: tcpip.TCPSendBufferSizeRangeOption{ Min: MinBufferSize, Default: DefaultSendBufferSize, diff --git a/pkg/tcpip/transport/tcp/rack.go b/pkg/tcpip/transport/tcp/rack.go index d969ca23a..439932595 100644 --- a/pkg/tcpip/transport/tcp/rack.go +++ b/pkg/tcpip/transport/tcp/rack.go @@ -39,6 +39,9 @@ type rackControl struct { // sequence. fack seqnum.Value + // minRTT is the estimated minimum RTT of the connection. + minRTT time.Duration + // rtt is the RTT of the most recently delivered packet on the // connection (either cumulatively acknowledged or selectively // acknowledged) that was not marked invalid as a possible spurious @@ -48,7 +51,7 @@ type rackControl struct { // Update will update the RACK related fields when an ACK has been received. // See: https://tools.ietf.org/html/draft-ietf-tcpm-rack-08#section-7.2 -func (rc *rackControl) Update(seg *segment, ackSeg *segment, srtt time.Duration, offset uint32) { +func (rc *rackControl) Update(seg *segment, ackSeg *segment, offset uint32) { rtt := time.Now().Sub(seg.xmitTime) // If the ACK is for a retransmitted packet, do not update if it is a @@ -65,12 +68,21 @@ func (rc *rackControl) Update(seg *segment, ackSeg *segment, srtt time.Duration, return } } - if rtt < srtt { + if rtt < rc.minRTT { return } } rc.rtt = rtt + + // The sender can either track a simple global minimum of all RTT + // measurements from the connection, or a windowed min-filtered value + // of recent RTT measurements. This implementation keeps track of the + // simple global minimum of all RTTs for the connection. + if rtt < rc.minRTT || rc.minRTT == 0 { + rc.minRTT = rtt + } + // Update rc.xmitTime and rc.endSequence to the transmit time and // ending sequence number of the packet which has been acknowledged // most recently. diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index cfd43b5e3..48bf196d8 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -47,22 +47,24 @@ type receiver struct { closed bool + // pendingRcvdSegments is bounded by the receive buffer size of the + // endpoint. pendingRcvdSegments segmentHeap - pendingBufUsed seqnum.Size - pendingBufSize seqnum.Size + // pendingBufUsed tracks the total number of bytes (including segment + // overhead) currently queued in pendingRcvdSegments. + pendingBufUsed int // Time when the last ack was received. lastRcvdAckTime time.Time `state:".(unixTime)"` } -func newReceiver(ep *endpoint, irs seqnum.Value, rcvWnd seqnum.Size, rcvWndScale uint8, pendingBufSize seqnum.Size) *receiver { +func newReceiver(ep *endpoint, irs seqnum.Value, rcvWnd seqnum.Size, rcvWndScale uint8) *receiver { return &receiver{ ep: ep, rcvNxt: irs + 1, rcvAcc: irs.Add(rcvWnd + 1), rcvWnd: rcvWnd, rcvWndScale: rcvWndScale, - pendingBufSize: pendingBufSize, lastRcvdAckTime: time.Now(), } } @@ -85,15 +87,30 @@ func (r *receiver) acceptable(segSeq seqnum.Value, segLen seqnum.Size) bool { // getSendParams returns the parameters needed by the sender when building // segments to send. func (r *receiver) getSendParams() (rcvNxt seqnum.Value, rcvWnd seqnum.Size) { - // Calculate the window size based on the available buffer space. - receiveBufferAvailable := r.ep.receiveBufferAvailable() - acc := r.rcvNxt.Add(seqnum.Size(receiveBufferAvailable)) - if r.rcvAcc.LessThan(acc) { - r.rcvAcc = acc + avail := wndFromSpace(r.ep.receiveBufferAvailable()) + if avail == 0 { + // We have no space available to accept any data, move to zero window + // state. + r.rcvWnd = 0 + return r.rcvNxt, 0 + } + + acc := r.rcvNxt.Add(seqnum.Size(avail)) + newWnd := r.rcvNxt.Size(acc) + curWnd := r.rcvNxt.Size(r.rcvAcc) + + // 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 + // would end up dropping bytes that might already be in flight. + if newWnd > curWnd { + r.rcvAcc = r.rcvNxt.Add(newWnd) + } else { + newWnd = curWnd } // Stash away the non-scaled receive window as we use it for measuring // receiver's estimated RTT. - r.rcvWnd = r.rcvNxt.Size(r.rcvAcc) + r.rcvWnd = newWnd return r.rcvNxt, r.rcvWnd >> r.rcvWndScale } @@ -195,7 +212,9 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum } for i := first; i < len(r.pendingRcvdSegments); i++ { + r.pendingBufUsed -= r.pendingRcvdSegments[i].segMemSize() r.pendingRcvdSegments[i].decRef() + // Note that slice truncation does not allow garbage collection of // truncated items, thus truncated items must be set to nil to avoid // memory leaks. @@ -384,10 +403,16 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { // Defer segment processing if it can't be consumed now. if !r.consumeSegment(s, segSeq, segLen) { if segLen > 0 || s.flagIsSet(header.TCPFlagFin) { - // We only store the segment if it's within our buffer - // size limit. - if r.pendingBufUsed < r.pendingBufSize { - r.pendingBufUsed += seqnum.Size(s.segMemSize()) + // We only store the segment if it's within our buffer size limit. + // + // Only use 75% of the receive buffer queue for out-of-order + // segments. This ensures that we always leave some space for the inorder + // segments to arrive allowing pending segments to be processed and + // delivered to the user. + if r.ep.receiveBufferAvailable() > 0 && r.pendingBufUsed < r.ep.receiveBufferSize()>>2 { + r.ep.rcvListMu.Lock() + r.pendingBufUsed += s.segMemSize() + r.ep.rcvListMu.Unlock() s.incRef() heap.Push(&r.pendingRcvdSegments, s) UpdateSACKBlocks(&r.ep.sack, segSeq, segSeq.Add(segLen), r.rcvNxt) @@ -421,7 +446,9 @@ func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { } heap.Pop(&r.pendingRcvdSegments) - r.pendingBufUsed -= seqnum.Size(s.segMemSize()) + r.ep.rcvListMu.Lock() + r.pendingBufUsed -= s.segMemSize() + r.ep.rcvListMu.Unlock() s.decRef() } return false, nil diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 94307d31a..13acaf753 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -15,6 +15,7 @@ package tcp import ( + "fmt" "sync/atomic" "time" @@ -24,6 +25,15 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/stack" ) +// queueFlags are used to indicate which queue of an endpoint a particular segment +// belongs to. This is used to track memory accounting correctly. +type queueFlags uint8 + +const ( + recvQ queueFlags = 1 << iota + sendQ +) + // segment represents a TCP segment. It holds the payload and parsed TCP segment // information, and can be added to intrusive lists. // segment is mostly immutable, the only field allowed to change is viewToDeliver. @@ -32,6 +42,8 @@ import ( type segment struct { segmentEntry refCnt int32 + ep *endpoint + qFlags queueFlags id stack.TransportEndpointID `state:"manual"` route stack.Route `state:"manual"` data buffer.VectorisedView `state:".(buffer.VectorisedView)"` @@ -100,6 +112,8 @@ func (s *segment) clone() *segment { rcvdTime: s.rcvdTime, xmitTime: s.xmitTime, xmitCount: s.xmitCount, + ep: s.ep, + qFlags: s.qFlags, } t.data = s.data.Clone(t.views[:]) return t @@ -115,8 +129,34 @@ func (s *segment) flagsAreSet(flags uint8) bool { return s.flags&flags == flags } +// setOwner sets the owning endpoint for this segment. Its required +// to be called to ensure memory accounting for receive/send buffer +// queues is done properly. +func (s *segment) setOwner(ep *endpoint, qFlags queueFlags) { + switch qFlags { + case recvQ: + ep.updateReceiveMemUsed(s.segMemSize()) + case sendQ: + // no memory account for sendQ yet. + default: + panic(fmt.Sprintf("unexpected queue flag %b", qFlags)) + } + s.ep = ep + s.qFlags = qFlags +} + func (s *segment) decRef() { if atomic.AddInt32(&s.refCnt, -1) == 0 { + if s.ep != nil { + switch s.qFlags { + case recvQ: + s.ep.updateReceiveMemUsed(-s.segMemSize()) + case sendQ: + // no memory accounting for sendQ yet. + default: + panic(fmt.Sprintf("unexpected queue flag %b set for segment", s.qFlags)) + } + } s.route.Release() } } @@ -138,6 +178,11 @@ func (s *segment) logicalLen() seqnum.Size { return l } +// payloadSize is the size of s.data. +func (s *segment) payloadSize() int { + return s.data.Size() +} + // segMemSize is the amount of memory used to hold the segment data and // the associated metadata. func (s *segment) segMemSize() int { diff --git a/pkg/tcpip/transport/tcp/segment_queue.go b/pkg/tcpip/transport/tcp/segment_queue.go index 48a257137..54545a1b1 100644 --- a/pkg/tcpip/transport/tcp/segment_queue.go +++ b/pkg/tcpip/transport/tcp/segment_queue.go @@ -22,16 +22,16 @@ import ( // // +stateify savable type segmentQueue struct { - mu sync.Mutex `state:"nosave"` - list segmentList `state:"wait"` - limit int - used int + mu sync.Mutex `state:"nosave"` + list segmentList `state:"wait"` + ep *endpoint + frozen bool } // emptyLocked determines if the queue is empty. // Preconditions: q.mu must be held. func (q *segmentQueue) emptyLocked() bool { - return q.used == 0 + return q.list.Empty() } // empty determines if the queue is empty. @@ -43,14 +43,6 @@ func (q *segmentQueue) empty() bool { return r } -// setLimit updates the limit. No segments are immediately dropped in case the -// queue becomes full due to the new limit. -func (q *segmentQueue) setLimit(limit int) { - q.mu.Lock() - q.limit = limit - q.mu.Unlock() -} - // enqueue adds the given segment to the queue. // // Returns true when the segment is successfully added to the queue, in which @@ -58,15 +50,23 @@ func (q *segmentQueue) setLimit(limit int) { // false if the queue is full, in which case ownership is retained by the // caller. func (q *segmentQueue) enqueue(s *segment) bool { + // q.ep.receiveBufferParams() must be called without holding q.mu to + // avoid lock order inversion. + bufSz := q.ep.receiveBufferSize() + used := q.ep.receiveMemUsed() q.mu.Lock() - r := q.used < q.limit - if r { + // Allow zero sized segments (ACK/FIN/RSTs etc even if the segment queue + // is currently full). + allow := (used <= bufSz || s.payloadSize() == 0) && !q.frozen + + if allow { q.list.PushBack(s) - q.used++ + // Set the owner now that the endpoint owns the segment. + s.setOwner(q.ep, recvQ) } q.mu.Unlock() - return r + return allow } // dequeue removes and returns the next segment from queue, if one exists. @@ -77,9 +77,25 @@ func (q *segmentQueue) dequeue() *segment { s := q.list.Front() if s != nil { q.list.Remove(s) - q.used-- } q.mu.Unlock() return s } + +// freeze prevents any more segments from being added to the queue. i.e all +// future segmentQueue.enqueue will return false and not add the segment to the +// queue till the queue is unfroze with a corresponding segmentQueue.thaw call. +func (q *segmentQueue) freeze() { + q.mu.Lock() + q.frozen = true + q.mu.Unlock() +} + +// thaw unfreezes a previously frozen queue using segmentQueue.freeze() and +// allows new segments to be queued again. +func (q *segmentQueue) thaw() { + q.mu.Lock() + q.frozen = false + q.mu.Unlock() +} diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index c55589c45..4c9a86cda 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -1365,9 +1365,6 @@ func (s *sender) handleRcvdSegment(rcvdSeg *segment) { ackLeft := acked originalOutstanding := s.outstanding - s.rtt.Lock() - srtt := s.rtt.srtt - s.rtt.Unlock() for ackLeft > 0 { // We use logicalLen here because we can have FIN // segments (which are always at the end of list) that @@ -1389,7 +1386,7 @@ func (s *sender) handleRcvdSegment(rcvdSeg *segment) { // Update the RACK fields if SACK is enabled. if s.ep.sackPermitted { - s.rc.Update(seg, rcvdSeg, srtt, s.ep.tsOffset) + s.rc.Update(seg, rcvdSeg, s.ep.tsOffset) } s.writeList.Remove(seg) diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index b1e5f1b24..5b504d0d1 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -21,6 +21,7 @@ import ( "testing" "time" + "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -240,6 +241,38 @@ func TestTCPResetsSentIncrement(t *testing.T) { } } +// TestTCPResetsSentNoICMP confirms that we don't get an ICMP +// DstUnreachable packet when we try send a packet which is not part +// of an active session. +func TestTCPResetsSentNoICMP(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + stats := c.Stack().Stats() + + // Send a SYN request for a closed port. This should elicit an RST + // but NOT an ICMPv4 DstUnreachable packet. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + }) + + // Receive whatever comes back. + b := c.GetPacket() + ipHdr := header.IPv4(b) + if got, want := ipHdr.Protocol(), uint8(header.TCPProtocolNumber); got != want { + t.Errorf("unexpected protocol, got = %d, want = %d", got, want) + } + + // Read outgoing ICMP stats and check no ICMP DstUnreachable was recorded. + sent := stats.ICMP.V4PacketsSent + if got, want := sent.DstUnreachable.Value(), uint64(0); got != want { + t.Errorf("got ICMP DstUnreachable.Value() = %d, want = %d", got, want) + } +} + // TestTCPResetSentForACKWhenNotUsingSynCookies checks that the stack generates // a RST if an ACK is received on the listening socket for which there is no // active handshake in progress and we are not using SYN cookies. @@ -317,8 +350,8 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ SrcPort: context.TestPort, @@ -348,8 +381,8 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(0), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst))) } @@ -447,8 +480,8 @@ func TestConnectResetAfterClose(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -489,8 +522,8 @@ func TestConnectResetAfterClose(t *testing.T) { // RST is always generated with sndNxt which if the FIN // has been sent will be 1 higher than the sequence number // of the FIN itself. - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(0), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -529,8 +562,8 @@ func TestCurrentConnectedIncrement(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -565,8 +598,8 @@ func TestCurrentConnectedIncrement(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -612,8 +645,8 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -633,8 +666,8 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -693,8 +726,8 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(0), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -745,8 +778,8 @@ func TestSimpleReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -998,7 +1031,7 @@ func TestSendRstOnListenerRxSynAckV4(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst), - checker.SeqNum(200))) + checker.TCPSeqNum(200))) } func TestSendRstOnListenerRxSynAckV6(t *testing.T) { @@ -1026,7 +1059,7 @@ func TestSendRstOnListenerRxSynAckV6(t *testing.T) { checker.IPv6(t, c.GetV6Packet(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst), - checker.SeqNum(200))) + checker.TCPSeqNum(200))) } // TestTCPAckBeforeAcceptV4 tests that once the 3-way handshake is complete, @@ -1063,8 +1096,8 @@ func TestTCPAckBeforeAcceptV4(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) } // TestTCPAckBeforeAcceptV6 tests that once the 3-way handshake is complete, @@ -1101,8 +1134,8 @@ func TestTCPAckBeforeAcceptV6(t *testing.T) { checker.IPv6(t, c.GetV6Packet(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) } func TestSendRstOnListenerRxAckV4(t *testing.T) { @@ -1130,7 +1163,7 @@ func TestSendRstOnListenerRxAckV4(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst), - checker.SeqNum(200))) + checker.TCPSeqNum(200))) } func TestSendRstOnListenerRxAckV6(t *testing.T) { @@ -1158,7 +1191,7 @@ func TestSendRstOnListenerRxAckV6(t *testing.T) { checker.IPv6(t, c.GetV6Packet(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst), - checker.SeqNum(200))) + checker.TCPSeqNum(200))) } // TestListenShutdown tests for the listening endpoint replying with RST @@ -1274,8 +1307,8 @@ func TestTOSV4(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), // Acknum is initial sequence number + 1 + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), // Acknum is initial sequence number + 1 checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), checker.TOS(tos, 0), @@ -1323,8 +1356,8 @@ func TestTrafficClassV6(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), checker.TOS(tos, 0), @@ -1514,8 +1547,8 @@ func TestOutOfOrderReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1565,8 +1598,8 @@ func TestOutOfOrderReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1576,8 +1609,8 @@ func TestOutOfOrderFlood(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - // Create a new connection with initial window size of 10. - c.CreateConnected(789, 30000, 10) + rcvBufSz := math.MaxUint16 + c.CreateConnected(789, 30000, rcvBufSz) if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrWouldBlock) @@ -1598,8 +1631,8 @@ func TestOutOfOrderFlood(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1619,8 +1652,8 @@ func TestOutOfOrderFlood(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1639,8 +1672,8 @@ func TestOutOfOrderFlood(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(793), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(793), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1681,8 +1714,8 @@ func TestRstOnCloseWithUnreadData(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1696,7 +1729,7 @@ func TestRstOnCloseWithUnreadData(t *testing.T) { checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), // We shouldn't consume a sequence number on RST. - checker.SeqNum(uint32(c.IRS)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), )) // The RST puts the endpoint into an error state. if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { @@ -1750,8 +1783,8 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -1764,7 +1797,7 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), - checker.SeqNum(uint32(c.IRS)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), )) if got, want := tcp.EndpointState(c.EP.State()), tcp.StateFinWait1; got != want { @@ -1783,7 +1816,7 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { // RST is always generated with sndNxt which if the FIN // has been sent will be 1 higher than the sequence // number of the FIN itself. - checker.SeqNum(uint32(c.IRS)+2), + checker.TCPSeqNum(uint32(c.IRS)+2), )) // The RST puts the endpoint into an error state. if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { @@ -1829,7 +1862,8 @@ func TestFullWindowReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - c.CreateConnected(789, 30000, 10) + const rcvBufSz = 10 + c.CreateConnected(789, 30000, rcvBufSz) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -1840,8 +1874,13 @@ func TestFullWindowReceive(t *testing.T) { t.Fatalf("Read failed: %s", err) } - // Fill up the window. - data := []byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10} + // Fill up the window w/ tcp.SegOverheadFactor*rcvBufSz as netstack multiplies + // the provided buffer value by tcp.SegOverheadFactor to calculate the actual + // receive buffer size. + data := make([]byte, tcp.SegOverheadFactor*rcvBufSz) + for i := range data { + data[i] = byte(i % 255) + } c.SendPacket(data, &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, @@ -1862,10 +1901,10 @@ func TestFullWindowReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), - checker.Window(0), + checker.TCPWindow(0), ), ) @@ -1888,10 +1927,10 @@ func TestFullWindowReceive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+len(data))), checker.TCPFlags(header.TCPFlagAck), - checker.Window(10), + checker.TCPWindow(10), ), ) } @@ -1900,12 +1939,15 @@ func TestNoWindowShrinking(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - // Start off with a window size of 10, then shrink it to 5. - c.CreateConnected(789, 30000, 10) - - if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 5); err != nil { - t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 5) failed: %s", err) - } + // Start off with a certain receive buffer then cut it in half and verify that + // the right edge of the window does not shrink. + // NOTE: Netstack doubles the value specified here. + rcvBufSize := 65536 + iss := seqnum.Value(789) + // Enable window scaling with a scale of zero from our end. + c.CreateConnectedWithRawOptions(iss, 30000, rcvBufSize, []byte{ + header.TCPOptionWS, 3, 0, header.TCPOptionNOP, + }) we, ch := waiter.NewChannelEntry(nil) c.WQ.EventRegister(&we, waiter.EventIn) @@ -1914,14 +1956,15 @@ func TestNoWindowShrinking(t *testing.T) { if _, _, err := c.EP.Read(nil); err != tcpip.ErrWouldBlock { t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrWouldBlock) } - - // Send 3 bytes, check that the peer acknowledges them. - data := []byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10} - c.SendPacket(data[:3], &context.Headers{ + // Send a 1 byte payload so that we can record the current receive window. + // Send a payload of half the size of rcvBufSize. + seqNum := iss.Add(1) + payload := []byte{1} + c.SendPacket(payload, &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, Flags: header.TCPFlagAck, - SeqNum: 790, + SeqNum: seqNum, AckNum: c.IRS.Add(1), RcvWnd: 30000, }) @@ -1933,46 +1976,93 @@ func TestNoWindowShrinking(t *testing.T) { t.Fatalf("Timed out waiting for data to arrive") } - // Check that data is acknowledged, and that window doesn't go to zero - // just yet because it was previously set to 10. It must go to 7 now. - checker.IPv4(t, c.GetPacket(), + // Read the 1 byte payload we just sent. + v, _, err := c.EP.Read(nil) + if err != nil { + t.Fatalf("Read failed: %s", err) + } + if got, want := payload, v; !bytes.Equal(got, want) { + t.Fatalf("got data: %v, want: %v", got, want) + } + + seqNum = seqNum.Add(1) + // Verify that the ACK does not shrink the window. + pkt := c.GetPacket() + checker.IPv4(t, pkt, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(793), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(seqNum)), checker.TCPFlags(header.TCPFlagAck), - checker.Window(7), ), ) + // Stash the initial window. + initialWnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize() << c.RcvdWindowScale + initialLastAcceptableSeq := seqNum.Add(seqnum.Size(initialWnd)) + // Now shrink the receive buffer to half its original size. + if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufSize/2); err != nil { + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 5) failed: %s", err) + } - // Send 7 more bytes, check that the window fills up. - c.SendPacket(data[3:], &context.Headers{ + data := generateRandomPayload(t, rcvBufSize) + // Send a payload of half the size of rcvBufSize. + c.SendPacket(data[:rcvBufSize/2], &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, Flags: header.TCPFlagAck, - SeqNum: 793, + SeqNum: seqNum, AckNum: c.IRS.Add(1), RcvWnd: 30000, }) + seqNum = seqNum.Add(seqnum.Size(rcvBufSize / 2)) - select { - case <-ch: - case <-time.After(5 * time.Second): - t.Fatalf("Timed out waiting for data to arrive") + // Verify that the ACK does not shrink the window. + pkt = c.GetPacket() + checker.IPv4(t, pkt, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(seqNum)), + checker.TCPFlags(header.TCPFlagAck), + ), + ) + newWnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize() << c.RcvdWindowScale + newLastAcceptableSeq := seqNum.Add(seqnum.Size(newWnd)) + if newLastAcceptableSeq.LessThan(initialLastAcceptableSeq) { + t.Fatalf("receive window shrunk unexpectedly got: %d, want >= %d", newLastAcceptableSeq, initialLastAcceptableSeq) } + // Send another payload of half the size of rcvBufSize. This should fill up the + // socket receive buffer and we should see a zero window. + c.SendPacket(data[rcvBufSize/2:], &context.Headers{ + SrcPort: context.TestPort, + DstPort: c.Port, + Flags: header.TCPFlagAck, + SeqNum: seqNum, + AckNum: c.IRS.Add(1), + RcvWnd: 30000, + }) + seqNum = seqNum.Add(seqnum.Size(rcvBufSize / 2)) + checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(seqNum)), checker.TCPFlags(header.TCPFlagAck), - checker.Window(0), + checker.TCPWindow(0), ), ) + // Wait for receive to be notified. + select { + case <-ch: + case <-time.After(5 * time.Second): + t.Fatalf("Timed out waiting for data to arrive") + } + // Receive data and check it. - read := make([]byte, 0, 10) + read := make([]byte, 0, rcvBufSize) for len(read) < len(data) { v, _, err := c.EP.Read(nil) if err != nil { @@ -1986,15 +2076,15 @@ func TestNoWindowShrinking(t *testing.T) { t.Fatalf("got data = %v, want = %v", read, data) } - // Check that we get an ACK for the newly non-zero window, which is the - // new size. + // Check that we get an ACK for the newly non-zero window, which is the new + // receive buffer size we set after the connection was established. checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(seqNum)), checker.TCPFlags(header.TCPFlagAck), - checker.Window(5), + checker.TCPWindow(uint16(rcvBufSize/2)>>c.RcvdWindowScale), ), ) } @@ -2019,8 +2109,8 @@ func TestSimpleSend(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2061,8 +2151,8 @@ func TestZeroWindowSend(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2083,8 +2173,8 @@ func TestZeroWindowSend(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2123,16 +2213,16 @@ func TestScaledWindowConnect(t *testing.T) { t.Fatalf("Write failed: %s", err) } - // Check that data is received, and that advertised window is 0xbfff, + // Check that data is received, and that advertised window is 0x5fff, // that is, that it is scaled. b := c.GetPacket() checker.IPv4(t, b, checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xbfff), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(0x5fff), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2162,9 +2252,9 @@ func TestNonScaledWindowConnect(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xffff), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(0xffff), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2198,7 +2288,8 @@ func TestScaledWindowAccept(t *testing.T) { } // Do 3-way handshake. - c.PassiveConnectWithOptions(100, 2, header.TCPSynOptions{MSS: defaultIPv4MSS}) + // wndScale expected is 3 as 65535 * 3 * 2 < 65535 * 2^3 but > 65535 *2 *2 + c.PassiveConnectWithOptions(100, 3 /* wndScale */, header.TCPSynOptions{MSS: defaultIPv4MSS}) // Try to accept the connection. we, ch := waiter.NewChannelEntry(nil) @@ -2228,16 +2319,16 @@ func TestScaledWindowAccept(t *testing.T) { t.Fatalf("Write failed: %s", err) } - // Check that data is received, and that advertised window is 0xbfff, + // Check that data is received, and that advertised window is 0x5fff, // that is, that it is scaled. b := c.GetPacket() checker.IPv4(t, b, checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xbfff), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(0x5fff), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2309,9 +2400,9 @@ func TestNonScaledWindowAccept(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(0xffff), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(0xffff), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2324,18 +2415,19 @@ func TestZeroScaledWindowReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - // Set the window size such that a window scale of 4 will be used. - const wnd = 65535 * 10 - const ws = uint32(4) - c.CreateConnectedWithRawOptions(789, 30000, wnd, []byte{ + // Set the buffer size such that a window scale of 5 will be used. + const bufSz = 65535 * 10 + const ws = uint32(5) + c.CreateConnectedWithRawOptions(789, 30000, bufSz, []byte{ header.TCPOptionWS, 3, 0, header.TCPOptionNOP, }) // Write chunks of 50000 bytes. - remain := wnd + remain := 0 sent := 0 data := make([]byte, 50000) - for remain > len(data) { + // Keep writing till the window drops below len(data). + for { c.SendPacket(data, &context.Headers{ SrcPort: context.TestPort, DstPort: c.Port, @@ -2345,21 +2437,25 @@ func TestZeroScaledWindowReceive(t *testing.T) { RcvWnd: 30000, }) sent += len(data) - remain -= len(data) - checker.IPv4(t, c.GetPacket(), + pkt := c.GetPacket() + checker.IPv4(t, pkt, checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(remain>>ws)), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), checker.TCPFlags(header.TCPFlagAck), ), ) + // Don't reduce window to zero here. + if wnd := int(header.TCP(header.IPv4(pkt).Payload()).WindowSize()); wnd<<ws < len(data) { + remain = wnd << ws + break + } } // Make the window non-zero, but the scaled window zero. - if remain >= 16 { + for remain >= 16 { data = data[:remain-15] c.SendPacket(data, &context.Headers{ SrcPort: context.TestPort, @@ -2370,22 +2466,35 @@ func TestZeroScaledWindowReceive(t *testing.T) { RcvWnd: 30000, }) sent += len(data) - remain -= len(data) - checker.IPv4(t, c.GetPacket(), + pkt := c.GetPacket() + checker.IPv4(t, pkt, checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(0), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), checker.TCPFlags(header.TCPFlagAck), ), ) + // Since the receive buffer is split between window advertisement and + // application data buffer the window does not always reflect the space + // available and actual space available can be a bit more than what is + // advertised in the window. + wnd := int(header.TCP(header.IPv4(pkt).Payload()).WindowSize()) + if wnd == 0 { + break + } + remain = wnd << ws } - // Read at least 1MSS of data. An ack should be sent in response to that. + // Read at least 2MSS of data. An ack should be sent in response to that. + // Since buffer space is now split in half between window and application + // data we need to read more than 1 MSS(65536) of data for a non-zero window + // update to be sent. For 1MSS worth of window to be available we need to + // read at least 128KB. Since our segments above were 50KB each it means + // we need to read at 3 packets. sz := 0 - for sz < defaultMTU { + for sz < defaultMTU*2 { v, _, err := c.EP.Read(nil) if err != nil { t.Fatalf("Read failed: %s", err) @@ -2397,9 +2506,9 @@ func TestZeroScaledWindowReceive(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(sz>>ws)), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), + checker.TCPWindowGreaterThanEq(uint16(defaultMTU>>ws)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -2466,8 +2575,8 @@ func TestSegmentMerging(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize+1), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+uint32(i)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+uint32(i)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2489,8 +2598,8 @@ func TestSegmentMerging(t *testing.T) { checker.PayloadLen(len(allData)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+11), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+11), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2537,8 +2646,8 @@ func TestDelay(t *testing.T) { checker.PayloadLen(len(want)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(seq)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2584,8 +2693,8 @@ func TestUndelay(t *testing.T) { checker.PayloadLen(len(allData[0])+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(seq)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2607,8 +2716,8 @@ func TestUndelay(t *testing.T) { checker.PayloadLen(len(allData[1])+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(seq)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2669,8 +2778,8 @@ func TestMSSNotDelayed(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(seq)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(seq)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2721,8 +2830,8 @@ func testBrokenUpWrite(t *testing.T, c *context.Context, maxPayload int) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1+uint32(bytesReceived)), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1+uint32(bytesReceived)), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -2964,7 +3073,7 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { // Set the buffer size to a deterministic size so that we can check the // window scaling option. const rcvBufferSize = 0x20000 - const wndScale = 2 + const wndScale = 3 if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufferSize); err != nil { t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, %d) failed failed: %s", rcvBufferSize, err) } @@ -2999,7 +3108,7 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagSyn), checker.SrcPort(tcpHdr.SourcePort()), - checker.SeqNum(tcpHdr.SequenceNumber()), + checker.TCPSeqNum(tcpHdr.SequenceNumber()), checker.TCPSynOptions(header.TCPSynOptions{MSS: mss, WS: wndScale}), ), ) @@ -3020,8 +3129,8 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+1), ), ) @@ -3314,8 +3423,8 @@ func TestFinImmediately(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3335,8 +3444,8 @@ func TestFinImmediately(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3357,8 +3466,8 @@ func TestFinRetransmit(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3368,8 +3477,8 @@ func TestFinRetransmit(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3389,8 +3498,8 @@ func TestFinRetransmit(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(791), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3413,8 +3522,8 @@ func TestFinWithNoPendingData(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3438,8 +3547,8 @@ func TestFinWithNoPendingData(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3460,8 +3569,8 @@ func TestFinWithNoPendingData(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3488,8 +3597,8 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3507,8 +3616,8 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3527,8 +3636,8 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3548,8 +3657,8 @@ func TestFinWithPendingDataCwndFull(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3572,8 +3681,8 @@ func TestFinWithPendingData(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3597,8 +3706,8 @@ func TestFinWithPendingData(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3613,8 +3722,8 @@ func TestFinWithPendingData(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3634,8 +3743,8 @@ func TestFinWithPendingData(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3659,8 +3768,8 @@ func TestFinWithPartialAck(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3680,8 +3789,8 @@ func TestFinWithPartialAck(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3695,8 +3804,8 @@ func TestFinWithPartialAck(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3711,8 +3820,8 @@ func TestFinWithPartialAck(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(791), + checker.TCPSeqNum(next), + checker.TCPAckNum(791), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -3803,8 +3912,8 @@ func scaledSendWindow(t *testing.T, scale uint8) { checker.PayloadLen((1<<scale)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -3942,7 +4051,7 @@ func TestReceivedSegmentQueuing(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -3993,8 +4102,8 @@ func TestReadAfterClosedState(t *testing.T) { checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagFin), ), ) @@ -4018,8 +4127,8 @@ func TestReadAfterClosedState(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+2), - checker.AckNum(uint32(791+len(data))), + checker.TCPSeqNum(uint32(c.IRS)+2), + checker.TCPAckNum(uint32(791+len(data))), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -4191,8 +4300,8 @@ func checkSendBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { func TestDefaultBufferSizes(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) // Check the default values. @@ -4254,8 +4363,8 @@ func TestDefaultBufferSizes(t *testing.T) { func TestMinMaxBufferSizes(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) // Check the default values. @@ -4280,14 +4389,14 @@ func TestMinMaxBufferSizes(t *testing.T) { } } - // Set values below the min. - if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 199); err != nil { + // Set values below the min/2. + if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 99); err != nil { t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 199) failed: %s", err) } checkRecvBufferSize(t, ep, 200) - if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 299); err != nil { + if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 149); err != nil { t.Fatalf("SetSockOptInt(SendBufferSizeOption, 299) failed: %s", err) } @@ -4298,19 +4407,21 @@ func TestMinMaxBufferSizes(t *testing.T) { t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption) failed: %s", err) } - checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20) + // Values above max are capped at max and then doubled. + checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20*2) if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 1+tcp.DefaultSendBufferSize*30); err != nil { t.Fatalf("SetSockOptInt(SendBufferSizeOption) failed: %s", err) } - checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) + // Values above max are capped at max and then doubled. + checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30*2) } func TestBindToDeviceOption(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}}) + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}}) ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) if err != nil { @@ -4359,11 +4470,11 @@ func TestBindToDeviceOption(t *testing.T) { func makeStack() (*stack.Stack, *tcpip.Error) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ - ipv4.NewProtocol(), - ipv6.NewProtocol(), + NetworkProtocols: []stack.NetworkProtocolFactory{ + ipv4.NewProtocol, + ipv6.NewProtocol, }, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) id := loopback.New() @@ -4646,8 +4757,8 @@ func TestPathMTUDiscovery(t *testing.T) { checker.PayloadLen(size+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(seqNum), - checker.AckNum(790), + checker.TCPSeqNum(seqNum), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -4898,8 +5009,8 @@ func TestKeepalive(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)), - checker.AckNum(uint32(790)), + checker.TCPSeqNum(uint32(c.IRS)), + checker.TCPAckNum(uint32(790)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -4932,8 +5043,8 @@ func TestKeepalive(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -4944,8 +5055,8 @@ func TestKeepalive(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagPsh), ), ) @@ -4970,8 +5081,8 @@ func TestKeepalive(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(next-1)), - checker.AckNum(uint32(790)), + checker.TCPSeqNum(uint32(next-1)), + checker.TCPAckNum(uint32(790)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -4997,8 +5108,8 @@ func TestKeepalive(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(next)), - checker.AckNum(uint32(0)), + checker.TCPSeqNum(uint32(next)), + checker.TCPAckNum(uint32(0)), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -5038,7 +5149,7 @@ func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCooki checker.SrcPort(context.StackPort), checker.DstPort(srcPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), + checker.TCPAckNum(uint32(irs) + 1), } if synCookieInUse { @@ -5082,7 +5193,7 @@ func executeV6Handshake(t *testing.T, c *context.Context, srcPort uint16, synCoo checker.SrcPort(context.StackPort), checker.DstPort(srcPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), + checker.TCPAckNum(uint32(irs) + 1), } if synCookieInUse { @@ -5316,7 +5427,7 @@ func TestListenNoAcceptNonUnicastV4(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1))) + checker.TCPAckNum(uint32(irs)+1))) }) } } @@ -5324,8 +5435,8 @@ func TestListenNoAcceptNonUnicastV4(t *testing.T) { // TestListenNoAcceptMulticastBroadcastV6 makes sure that TCP segments with a // non unicast IPv6 address are not accepted. func TestListenNoAcceptNonUnicastV6(t *testing.T) { - multicastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x01") - otherMulticastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x02") + multicastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x01") + otherMulticastAddr := tcpip.Address("\xff\x0e\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x02") tests := []struct { name string @@ -5416,7 +5527,7 @@ func TestListenNoAcceptNonUnicastV6(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1))) + checker.TCPAckNum(uint32(irs)+1))) }) } } @@ -5464,7 +5575,7 @@ func TestListenSynRcvdQueueFull(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), + checker.TCPAckNum(uint32(irs) + 1), } checker.IPv4(t, b, checker.TCP(tcpCheckers...)) @@ -5642,7 +5753,7 @@ func TestSynRcvdBadSeqNumber(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(irs) + 1), + checker.TCPAckNum(uint32(irs) + 1), } checker.IPv4(t, b, checker.TCP(tcpCheckers...)) @@ -5663,8 +5774,8 @@ func TestSynRcvdBadSeqNumber(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.AckNum(uint32(irs) + 1), - checker.SeqNum(uint32(iss + 1)), + checker.TCPAckNum(uint32(irs) + 1), + checker.TCPSeqNum(uint32(iss + 1)), } checker.IPv4(t, b, checker.TCP(tcpCheckers...)) @@ -5962,16 +6073,14 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { time.Sleep(latency) rawEP.SendPacketWithTS([]byte{1}, tsVal) - // Verify that the ACK has the expected window. - wantRcvWnd := receiveBufferSize - wantRcvWnd = (wantRcvWnd >> uint32(c.WindowScale)) - rawEP.VerifyACKRcvWnd(uint16(wantRcvWnd - 1)) + 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. - b := make([]byte, int(receiveBufferSize)*2) - offset := 0 - payloadSize := receiveBufferSize - 1 + payloadSize := receiveBufferSize * 2 + b := make([]byte, int(payloadSize)) + worker := (c.EP).(interface { StopWork() ResumeWork() @@ -5980,11 +6089,15 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // Stop the worker goroutine. worker.StopWork() - start := offset - end := offset + payloadSize + start := 0 + end := payloadSize / 2 packetsSent := 0 for ; start < end; start += mss { - rawEP.SendPacketWithTS(b[start:start+mss], tsVal) + packetEnd := start + mss + if start+mss > end { + packetEnd = end + } + rawEP.SendPacketWithTS(b[start:packetEnd], tsVal) packetsSent++ } @@ -5992,29 +6105,20 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // are waiting to be read. worker.ResumeWork() - // Since we read no bytes the window should goto zero till the - // application reads some of the data. - // Discard all intermediate acks except the last one. - if packetsSent > 100 { - for i := 0; i < (packetsSent / 100); i++ { - _ = c.GetPacket() - } + // Since we sent almost the full receive buffer worth of data (some may have + // been dropped due to segment overheads), we should get a zero window back. + pkt = c.GetPacket() + tcpHdr := header.TCP(header.IPv4(pkt).Payload()) + gotRcvWnd := tcpHdr.WindowSize() + wantAckNum := tcpHdr.AckNumber() + if got, want := int(gotRcvWnd), 0; got != want { + t.Fatalf("got rcvWnd: %d, want: %d", got, want) } - rawEP.VerifyACKRcvWnd(0) time.Sleep(25 * time.Millisecond) - // Verify that sending more data when window is closed is dropped and - // not acked. + // Verify that sending more data when receiveBuffer is exhausted. rawEP.SendPacketWithTS(b[start:start+mss], tsVal) - // Verify that the stack sends us back an ACK with the sequence number - // of the last packet sent indicating it was dropped. - p := c.GetPacket() - checker.IPv4(t, p, checker.TCP( - checker.AckNum(uint32(rawEP.NextSeqNum)-uint32(mss)), - checker.Window(0), - )) - // Now read all the data from the endpoint and verify that advertised // window increases to the full available buffer size. for { @@ -6027,23 +6131,26 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { // Verify that we receive a non-zero window update ACK. When running // under thread santizer this test can end up sending more than 1 // ack, 1 for the non-zero window - p = c.GetPacket() + p := c.GetPacket() checker.IPv4(t, p, checker.TCP( - checker.AckNum(uint32(rawEP.NextSeqNum)-uint32(mss)), + checker.TCPAckNum(uint32(wantAckNum)), func(t *testing.T, h header.Transport) { tcp, ok := h.(header.TCP) if !ok { return } - if w := tcp.WindowSize(); w == 0 || w > uint16(wantRcvWnd) { - t.Errorf("expected a non-zero window: got %d, want <= wantRcvWnd", w) + // We use 10% here as the error margin upwards as the initial window we + // got was afer 1 segment was already in the receive buffer queue. + tolerance := 1.1 + if w := tcp.WindowSize(); w == 0 || w > uint16(float64(rcvWnd)*tolerance) { + t.Errorf("expected a non-zero window: got %d, want <= %d", w, uint16(float64(rcvWnd)*tolerance)) } }, )) } -// This test verifies that the auto tuning does not grow the receive buffer if -// the application is not reading the data actively. +// This test verifies that the advertised window is auto-tuned up as the +// application is reading the data that is being received. func TestReceiveBufferAutoTuning(t *testing.T) { const mtu = 1500 const mss = mtu - header.IPv4MinimumSize - header.TCPMinimumSize @@ -6053,9 +6160,6 @@ func TestReceiveBufferAutoTuning(t *testing.T) { // Enable Auto-tuning. stk := c.Stack() - // Set lower limits for auto-tuning tests. This is required because the - // test stops the worker which can cause packets to be dropped because - // the segment queue holding unprocessed packets is limited to 300. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 { @@ -6077,8 +6181,12 @@ func TestReceiveBufferAutoTuning(t *testing.T) { c.WindowScale = uint8(tcp.FindWndScale(maxReceiveBufferSize)) rawEP := c.CreateConnectedWithOptions(header.TCPSynOptions{TS: true, WS: 4}) - - wantRcvWnd := receiveBufferSize + tsVal := uint32(rawEP.TSVal) + rawEP.NextSeqNum-- + rawEP.SendPacketWithTS(nil, tsVal) + rawEP.NextSeqNum++ + pkt := rawEP.VerifyAndReturnACKWithTS(tsVal) + curRcvWnd := int(header.TCP(header.IPv4(pkt).Payload()).WindowSize()) << c.WindowScale scaleRcvWnd := func(rcvWnd int) uint16 { return uint16(rcvWnd >> uint16(c.WindowScale)) } @@ -6095,14 +6203,8 @@ func TestReceiveBufferAutoTuning(t *testing.T) { StopWork() ResumeWork() }) - tsVal := rawEP.TSVal - // We are going to do our own computation of what the moderated receive - // buffer should be based on sent/copied data per RTT and verify that - // the advertised window by the stack matches our calculations. - prevCopied := 0 - done := false latency := 1 * time.Millisecond - for i := 0; !done; i++ { + for i := 0; i < 5; i++ { tsVal++ // Stop the worker goroutine. @@ -6124,15 +6226,20 @@ func TestReceiveBufferAutoTuning(t *testing.T) { // Give 1ms for the worker to process the packets. time.Sleep(1 * time.Millisecond) - // Verify that the advertised window on the ACK is reduced by - // the total bytes sent. - expectedWnd := wantRcvWnd - totalSent - if packetsSent > 100 { - for i := 0; i < (packetsSent / 100); i++ { - _ = c.GetPacket() + lastACK := c.GetPacket() + // Discard any intermediate ACKs and only check the last ACK we get in a + // short time period of few ms. + for { + time.Sleep(1 * time.Millisecond) + pkt := c.GetPacketNonBlocking() + if pkt == nil { + break } + lastACK = pkt + } + if got, want := int(header.TCP(header.IPv4(lastACK).Payload()).WindowSize()), int(scaleRcvWnd(curRcvWnd)); got > want { + t.Fatalf("advertised window got: %d, want <= %d", got, want) } - rawEP.VerifyACKRcvWnd(scaleRcvWnd(expectedWnd)) // Now read all the data from the endpoint and invoke the // moderation API to allow for receive buffer auto-tuning @@ -6157,35 +6264,20 @@ func TestReceiveBufferAutoTuning(t *testing.T) { rawEP.NextSeqNum-- rawEP.SendPacketWithTS(nil, tsVal) rawEP.NextSeqNum++ - if i == 0 { // In the first iteration the receiver based RTT is not // yet known as a result the moderation code should not // increase the advertised window. - rawEP.VerifyACKRcvWnd(scaleRcvWnd(wantRcvWnd)) - prevCopied = totalCopied + rawEP.VerifyACKRcvWnd(scaleRcvWnd(curRcvWnd)) } else { - rttCopied := totalCopied - if i == 1 { - // The moderation code accumulates copied bytes till - // RTT is established. So add in the bytes sent in - // the first iteration to the total bytes for this - // RTT. - rttCopied += prevCopied - // Now reset it to the initial value used by the - // auto tuning logic. - prevCopied = tcp.InitialCwnd * mss * 2 - } - newWnd := rttCopied<<1 + 16*mss - grow := (newWnd * (rttCopied - prevCopied)) / prevCopied - newWnd += (grow << 1) - if newWnd > maxReceiveBufferSize { - newWnd = maxReceiveBufferSize - done = true + pkt := c.GetPacket() + curRcvWnd = int(header.TCP(header.IPv4(pkt).Payload()).WindowSize()) << c.WindowScale + // If thew new current window is close maxReceiveBufferSize then terminate + // the loop. This can happen before all iterations are done due to timing + // differences when running the test. + if int(float64(curRcvWnd)*1.1) > maxReceiveBufferSize/2 { + break } - rawEP.VerifyACKRcvWnd(scaleRcvWnd(newWnd)) - wantRcvWnd = newWnd - prevCopied = rttCopied // Increase the latency after first two iterations to // establish a low RTT value in the receiver since it // only tracks the lowest value. This ensures that when @@ -6198,6 +6290,12 @@ func TestReceiveBufferAutoTuning(t *testing.T) { offset += payloadSize payloadSize *= 2 } + // Check that at the end of our iterations the receive window grew close to the maximum + // permissible size of maxReceiveBufferSize/2 + if got, want := int(float64(curRcvWnd)*1.1), maxReceiveBufferSize/2; got < want { + t.Fatalf("unexpected rcvWnd got: %d, want > %d", got, want) + } + } func TestDelayEnabled(t *testing.T) { @@ -6349,8 +6447,8 @@ func TestTCPTimeWaitRSTIgnored(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ @@ -6367,8 +6465,8 @@ func TestTCPTimeWaitRSTIgnored(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Now send a RST and this should be ignored and not @@ -6396,8 +6494,8 @@ func TestTCPTimeWaitRSTIgnored(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) } @@ -6468,8 +6566,8 @@ func TestTCPTimeWaitOutOfOrder(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ @@ -6486,8 +6584,8 @@ func TestTCPTimeWaitOutOfOrder(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Out of order ACK should generate an immediate ACK in @@ -6503,8 +6601,8 @@ func TestTCPTimeWaitOutOfOrder(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) } @@ -6575,8 +6673,8 @@ func TestTCPTimeWaitNewSyn(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ @@ -6593,8 +6691,8 @@ func TestTCPTimeWaitNewSyn(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Send a SYN request w/ sequence number lower than @@ -6732,8 +6830,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) finHeaders := &context.Headers{ @@ -6750,8 +6848,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) time.Sleep(2 * time.Second) @@ -6765,8 +6863,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+2)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+2)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Sleep for 4 seconds so at this point we are 1 second past the @@ -6794,8 +6892,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(ackHeaders.AckNum)), - checker.AckNum(0), + checker.TCPSeqNum(uint32(ackHeaders.AckNum)), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst))) if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != want { @@ -6894,8 +6992,8 @@ func TestTCPCloseWithData(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(iss)+2), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(iss)+2), checker.TCPFlags(header.TCPFlagAck))) // Now write a few bytes and then close the endpoint. @@ -6913,8 +7011,8 @@ func TestTCPCloseWithData(t *testing.T) { checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+2), // Acknum is initial sequence number + 1 + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+2), // Acknum is initial sequence number + 1 checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -6928,8 +7026,8 @@ func TestTCPCloseWithData(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)+uint32(len(data))), - checker.AckNum(uint32(iss+2)), + checker.TCPSeqNum(uint32(c.IRS+1)+uint32(len(data))), + checker.TCPAckNum(uint32(iss+2)), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) // First send a partial ACK. @@ -6974,8 +7072,8 @@ func TestTCPCloseWithData(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), - checker.SeqNum(uint32(ackHeaders.AckNum)), - checker.AckNum(0), + checker.TCPSeqNum(uint32(ackHeaders.AckNum)), + checker.TCPAckNum(0), checker.TCPFlags(header.TCPFlagRst))) } @@ -7011,8 +7109,8 @@ func TestTCPUserTimeout(t *testing.T) { checker.PayloadLen(len(view)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(next), - checker.AckNum(790), + checker.TCPSeqNum(next), + checker.TCPAckNum(790), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -7046,8 +7144,8 @@ func TestTCPUserTimeout(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(next)), - checker.AckNum(uint32(0)), + checker.TCPSeqNum(uint32(next)), + checker.TCPAckNum(uint32(0)), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -7108,8 +7206,8 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)), - checker.AckNum(uint32(790)), + checker.TCPSeqNum(uint32(c.IRS)), + checker.TCPAckNum(uint32(790)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -7134,8 +7232,8 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS+1)), - checker.AckNum(uint32(0)), + checker.TCPSeqNum(uint32(c.IRS+1)), + checker.TCPAckNum(uint32(0)), checker.TCPFlags(header.TCPFlagRst), ), ) @@ -7151,9 +7249,9 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { } } -func TestIncreaseWindowOnReceive(t *testing.T) { +func TestIncreaseWindowOnRead(t *testing.T) { // This test ensures that the endpoint sends an ack, - // after recv() when the window grows to more than 1 MSS. + // after read() when the window grows by more than 1 MSS. c := context.New(t, defaultMTU) defer c.Cleanup() @@ -7162,10 +7260,9 @@ func TestIncreaseWindowOnReceive(t *testing.T) { // Write chunks of ~30000 bytes. It's important that two // payloads make it equal or longer than MSS. - remain := rcvBuf + remain := rcvBuf * 2 sent := 0 data := make([]byte, defaultMTU/2) - lastWnd := uint16(0) for remain > len(data) { c.SendPacket(data, &context.Headers{ @@ -7178,46 +7275,43 @@ func TestIncreaseWindowOnReceive(t *testing.T) { }) sent += len(data) remain -= len(data) - - lastWnd = uint16(remain) - if remain > 0xffff { - lastWnd = 0xffff - } - checker.IPv4(t, c.GetPacket(), + pkt := c.GetPacket() + checker.IPv4(t, pkt, checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(lastWnd), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), checker.TCPFlags(header.TCPFlagAck), ), ) + // Break once the window drops below defaultMTU/2 + if wnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize(); wnd < defaultMTU/2 { + break + } } - if lastWnd == 0xffff || lastWnd == 0 { - t.Fatalf("expected small, non-zero window: %d", lastWnd) - } - - // We now have < 1 MSS in the buffer space. Read the data! An - // ack should be sent in response to that. The window was not - // zero, but it grew to larger than MSS. - if _, _, err := c.EP.Read(nil); err != nil { - t.Fatalf("Read failed: %s", err) - } - - if _, _, err := c.EP.Read(nil); err != nil { - t.Fatalf("Read failed: %s", err) + // We now have < 1 MSS in the buffer space. Read at least > 2 MSS + // worth of data as receive buffer space + read := 0 + // defaultMTU is a good enough estimate for the MSS used for this + // connection. + for read < defaultMTU*2 { + v, _, err := c.EP.Read(nil) + if err != nil { + t.Fatalf("Read failed: %s", err) + } + read += len(v) } - // After reading two packets, we surely crossed MSS. See the ack: + // After reading > MSS worth of data, we surely crossed MSS. See the ack: checker.IPv4(t, c.GetPacket(), checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(0xffff)), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), + checker.TCPWindow(uint16(0xffff)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -7234,10 +7328,9 @@ func TestIncreaseWindowOnBufferResize(t *testing.T) { // Write chunks of ~30000 bytes. It's important that two // payloads make it equal or longer than MSS. - remain := rcvBuf + remain := rcvBuf * 2 sent := 0 data := make([]byte, defaultMTU/2) - lastWnd := uint16(0) for remain > len(data) { c.SendPacket(data, &context.Headers{ @@ -7251,38 +7344,29 @@ func TestIncreaseWindowOnBufferResize(t *testing.T) { sent += len(data) remain -= len(data) - lastWnd = uint16(remain) - if remain > 0xffff { - lastWnd = 0xffff - } checker.IPv4(t, c.GetPacket(), checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(lastWnd), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), + checker.TCPWindowLessThanEq(0xffff), checker.TCPFlags(header.TCPFlagAck), ), ) } - if lastWnd == 0xffff || lastWnd == 0 { - t.Fatalf("expected small, non-zero window: %d", lastWnd) - } - // Increasing the buffer from should generate an ACK, // since window grew from small value to larger equal MSS c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBuf*2) - // After reading two packets, we surely crossed MSS. See the ack: checker.IPv4(t, c.GetPacket(), checker.PayloadLen(header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(790+sent)), - checker.Window(uint16(0xffff)), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(790+sent)), + checker.TCPWindow(uint16(0xffff)), checker.TCPFlags(header.TCPFlagAck), ), ) @@ -7327,8 +7411,8 @@ func TestTCPDeferAccept(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) // Give a bit of time for the socket to be delivered to the accept queue. time.Sleep(50 * time.Millisecond) @@ -7342,8 +7426,8 @@ func TestTCPDeferAccept(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) } func TestTCPDeferAcceptTimeout(t *testing.T) { @@ -7380,7 +7464,7 @@ func TestTCPDeferAcceptTimeout(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagSyn), - checker.AckNum(uint32(irs)+1))) + checker.TCPAckNum(uint32(irs)+1))) // Send data. This should result in an acceptable endpoint. c.SendPacket([]byte{1, 2, 3, 4}, &context.Headers{ @@ -7396,8 +7480,8 @@ func TestTCPDeferAcceptTimeout(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) // Give sometime for the endpoint to be delivered to the accept queue. time.Sleep(50 * time.Millisecond) @@ -7412,8 +7496,8 @@ func TestTCPDeferAcceptTimeout(t *testing.T) { checker.SrcPort(context.StackPort), checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck), - checker.SeqNum(uint32(iss+1)), - checker.AckNum(uint32(irs+5)))) + checker.TCPSeqNum(uint32(iss+1)), + checker.TCPAckNum(uint32(irs+5)))) } func TestResetDuringClose(t *testing.T) { @@ -7438,8 +7522,8 @@ func TestResetDuringClose(t *testing.T) { checker.IPv4(t, c.GetPacket(), checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(irs.Add(1))), - checker.AckNum(uint32(iss.Add(5))))) + checker.TCPSeqNum(uint32(irs.Add(1))), + checker.TCPAckNum(uint32(iss.Add(5))))) // Close in a separate goroutine so that we can trigger // a race with the RST we send below. This should not @@ -7520,3 +7604,14 @@ func TestSetStackTimeWaitReuse(t *testing.T) { } } } + +// generateRandomPayload generates a random byte slice of the specified length +// causing a fatal test failure if it is unable to do so. +func generateRandomPayload(t *testing.T, n int) []byte { + t.Helper() + buf := make([]byte, n) + if _, err := rand.Read(buf); err != nil { + t.Fatalf("rand.Read(buf) failed: %s", err) + } + return buf +} diff --git a/pkg/tcpip/transport/tcp/tcp_timestamp_test.go b/pkg/tcpip/transport/tcp/tcp_timestamp_test.go index 44593ed98..0f9ed06cd 100644 --- a/pkg/tcpip/transport/tcp/tcp_timestamp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_timestamp_test.go @@ -159,9 +159,9 @@ func timeStampEnabledAccept(t *testing.T, cookieEnabled bool, wndScale int, wndS checker.PayloadLen(len(data)+header.TCPMinimumSize+12), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(wndSize), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(wndSize), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), checker.TCPTimestampChecker(true, 0, tsVal+1), ), @@ -181,7 +181,8 @@ func TestTimeStampEnabledAccept(t *testing.T) { wndSize uint16 }{ {true, -1, 0xffff}, // When cookie is used window scaling is disabled. - {false, 5, 0x8000}, // DefaultReceiveBufferSize is 1MB >> 5. + // DefaultReceiveBufferSize is 1MB >> 5. Advertised window will be 1/2 of that. + {false, 5, 0x4000}, } for _, tc := range testCases { timeStampEnabledAccept(t, tc.cookieEnabled, tc.wndScale, tc.wndSize) @@ -219,9 +220,9 @@ func timeStampDisabledAccept(t *testing.T, cookieEnabled bool, wndScale int, wnd checker.PayloadLen(len(data)+header.TCPMinimumSize), checker.TCP( checker.DstPort(context.TestPort), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(790), - checker.Window(wndSize), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(790), + checker.TCPWindow(wndSize), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), checker.TCPTimestampChecker(false, 0, 0), ), @@ -237,7 +238,9 @@ func TestTimeStampDisabledAccept(t *testing.T) { wndSize uint16 }{ {true, -1, 0xffff}, // When cookie is used window scaling is disabled. - {false, 5, 0x8000}, // DefaultReceiveBufferSize is 1MB >> 5. + // DefaultReceiveBufferSize is 1MB >> 5. Advertised window will be half of + // that. + {false, 5, 0x4000}, } for _, tc := range testCases { timeStampDisabledAccept(t, tc.cookieEnabled, tc.wndScale, tc.wndSize) diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 85e8c1c75..faf51ef95 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -145,14 +145,18 @@ type Context struct { // WindowScale is the expected window scale in SYN packets sent by // the stack. WindowScale uint8 + + // RcvdWindowScale is the actual window scale sent by the stack in + // SYN/SYN-ACK. + RcvdWindowScale uint8 } // New allocates and initializes a test context containing a new // stack and a link-layer endpoint. func New(t *testing.T, mtu uint32) *Context { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol}, }) const sendBufferSize = 1 << 20 // 1 MiB @@ -261,18 +265,17 @@ func (c *Context) CheckNoPacket(errMsg string) { c.CheckNoPacketTimeout(errMsg, 1*time.Second) } -// GetPacket reads a packet from the link layer endpoint and verifies +// GetPacketWithTimeout reads a packet from the link layer endpoint and verifies // that it is an IPv4 packet with the expected source and destination -// addresses. It will fail with an error if no packet is received for -// 2 seconds. -func (c *Context) GetPacket() []byte { +// addresses. If no packet is received in the specified timeout it will return +// nil. +func (c *Context) GetPacketWithTimeout(timeout time.Duration) []byte { c.t.Helper() - ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second) + ctx, cancel := context.WithTimeout(context.Background(), timeout) defer cancel() p, ok := c.linkEP.ReadContext(ctx) if !ok { - c.t.Fatalf("Packet wasn't written out") return nil } @@ -280,6 +283,14 @@ func (c *Context) GetPacket() []byte { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) } + // Just check that the stack set the transport protocol number for outbound + // TCP messages. + // TODO(gvisor.dev/issues/3810): Remove when protocol numbers are part + // of the headerinfo. + if p.Pkt.TransportProtocolNumber != tcp.ProtocolNumber { + c.t.Fatalf("got p.Pkt.TransportProtocolNumber = %d, want = %d", p.Pkt.TransportProtocolNumber, tcp.ProtocolNumber) + } + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) b := vv.ToView() @@ -291,6 +302,21 @@ func (c *Context) GetPacket() []byte { return b } +// GetPacket reads a packet from the link layer endpoint and verifies +// that it is an IPv4 packet with the expected source and destination +// addresses. +func (c *Context) GetPacket() []byte { + c.t.Helper() + + p := c.GetPacketWithTimeout(5 * time.Second) + if p == nil { + c.t.Fatalf("Packet wasn't written out") + return nil + } + + return p +} + // GetPacketNonBlocking reads a packet from the link layer endpoint // and verifies that it is an IPv4 packet with the expected source // and destination address. If no packet is available it will return @@ -307,6 +333,14 @@ func (c *Context) GetPacketNonBlocking() []byte { c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, ipv4.ProtocolNumber) } + // Just check that the stack set the transport protocol number for outbound + // TCP messages. + // TODO(gvisor.dev/issues/3810): Remove when protocol numbers are part + // of the headerinfo. + if p.Pkt.TransportProtocolNumber != tcp.ProtocolNumber { + c.t.Fatalf("got p.Pkt.TransportProtocolNumber = %d, want = %d", p.Pkt.TransportProtocolNumber, tcp.ProtocolNumber) + } + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) b := vv.ToView() @@ -470,8 +504,8 @@ func (c *Context) ReceiveAndCheckPacketWithOptions(data []byte, offset, size, op checker.PayloadLen(size+header.TCPMinimumSize+optlen), checker.TCP( checker.DstPort(TestPort), - checker.SeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), - checker.AckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), + checker.TCPSeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), + checker.TCPAckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -497,8 +531,8 @@ func (c *Context) ReceiveNonBlockingAndCheckPacket(data []byte, offset, size int checker.PayloadLen(size+header.TCPMinimumSize), checker.TCP( checker.DstPort(TestPort), - checker.SeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), - checker.AckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), + checker.TCPSeqNum(uint32(c.IRS.Add(seqnum.Size(1+offset)))), + checker.TCPAckNum(uint32(seqnum.Value(testInitialSequenceNumber).Add(1))), checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), ), ) @@ -636,6 +670,7 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) } tcpHdr := header.TCP(header.IPv4(b).Payload()) + synOpts := header.ParseSynOptions(tcpHdr.Options(), false /* isAck */) c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) c.SendPacket(nil, &Headers{ @@ -653,8 +688,8 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) checker.TCP( checker.DstPort(TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS)+1), - checker.AckNum(uint32(iss)+1), + checker.TCPSeqNum(uint32(c.IRS)+1), + checker.TCPAckNum(uint32(iss)+1), ), ) @@ -671,6 +706,7 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) c.t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) } + c.RcvdWindowScale = uint8(synOpts.WS) c.Port = tcpHdr.SourcePort() } @@ -742,17 +778,18 @@ func (r *RawEndpoint) SendPacket(payload []byte, opts []byte) { r.NextSeqNum = r.NextSeqNum.Add(seqnum.Size(len(payload))) } -// VerifyACKWithTS verifies that the tsEcr field in the ack matches the provided -// tsVal. -func (r *RawEndpoint) VerifyACKWithTS(tsVal uint32) { +// VerifyAndReturnACKWithTS verifies that the tsEcr field int he ACK matches +// the provided tsVal as well as returns the original packet. +func (r *RawEndpoint) VerifyAndReturnACKWithTS(tsVal uint32) []byte { + r.C.t.Helper() // Read ACK and verify that tsEcr of ACK packet is [1,2,3,4] ackPacket := r.C.GetPacket() checker.IPv4(r.C.t, ackPacket, checker.TCP( checker.DstPort(r.SrcPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), + checker.TCPSeqNum(uint32(r.AckNum)), + checker.TCPAckNum(uint32(r.NextSeqNum)), checker.TCPTimestampChecker(true, 0, tsVal), ), ) @@ -760,19 +797,28 @@ func (r *RawEndpoint) VerifyACKWithTS(tsVal uint32) { tcpSeg := header.TCP(header.IPv4(ackPacket).Payload()) opts := tcpSeg.ParsedOptions() r.RecentTS = opts.TSVal + return ackPacket +} + +// VerifyACKWithTS verifies that the tsEcr field in the ack matches the provided +// tsVal. +func (r *RawEndpoint) VerifyACKWithTS(tsVal uint32) { + r.C.t.Helper() + _ = r.VerifyAndReturnACKWithTS(tsVal) } // VerifyACKRcvWnd verifies that the window advertised by the incoming ACK // matches the provided rcvWnd. func (r *RawEndpoint) VerifyACKRcvWnd(rcvWnd uint16) { + r.C.t.Helper() ackPacket := r.C.GetPacket() checker.IPv4(r.C.t, ackPacket, checker.TCP( checker.DstPort(r.SrcPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), - checker.Window(rcvWnd), + checker.TCPSeqNum(uint32(r.AckNum)), + checker.TCPAckNum(uint32(r.NextSeqNum)), + checker.TCPWindow(rcvWnd), ), ) } @@ -791,8 +837,8 @@ func (r *RawEndpoint) VerifyACKHasSACK(sackBlocks []header.SACKBlock) { checker.TCP( checker.DstPort(r.SrcPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(r.AckNum)), - checker.AckNum(uint32(r.NextSeqNum)), + checker.TCPSeqNum(uint32(r.AckNum)), + checker.TCPAckNum(uint32(r.NextSeqNum)), checker.TCPSACKBlockChecker(sackBlocks), ), ) @@ -884,8 +930,8 @@ func (c *Context) CreateConnectedWithOptions(wantOptions header.TCPSynOptions) * tcpCheckers := []checker.TransportChecker{ checker.DstPort(TestPort), checker.TCPFlags(header.TCPFlagAck), - checker.SeqNum(uint32(c.IRS) + 1), - checker.AckNum(uint32(iss) + 1), + checker.TCPSeqNum(uint32(c.IRS) + 1), + checker.TCPAckNum(uint32(iss) + 1), } // Verify that tsEcr of ACK packet is wantOptions.TSVal if the @@ -920,7 +966,7 @@ func (c *Context) CreateConnectedWithOptions(wantOptions header.TCPSynOptions) * // Mark in context that timestamp option is enabled for this endpoint. c.TimeStampEnabled = true - + c.RcvdWindowScale = uint8(synOptions.WS) return &RawEndpoint{ C: c, SrcPort: tcpSeg.DestinationPort(), @@ -1013,6 +1059,7 @@ func (c *Context) PassiveConnect(maxPayload, wndScale int, synOptions header.TCP // value of the window scaling option to be sent in the SYN. If synOptions.WS > // 0 then we send the WindowScale option. func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions header.TCPSynOptions) *RawEndpoint { + c.t.Helper() opts := make([]byte, header.TCPOptionsMaximumSize) offset := 0 offset += header.EncodeMSSOption(uint32(maxPayload), opts) @@ -1051,13 +1098,14 @@ func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions // are present. b := c.GetPacket() tcp := header.TCP(header.IPv4(b).Payload()) + rcvdSynOptions := header.ParseSynOptions(tcp.Options(), true /* isAck */) c.IRS = seqnum.Value(tcp.SequenceNumber()) tcpCheckers := []checker.TransportChecker{ checker.SrcPort(StackPort), checker.DstPort(TestPort), checker.TCPFlags(header.TCPFlagAck | header.TCPFlagSyn), - checker.AckNum(uint32(iss) + 1), + checker.TCPAckNum(uint32(iss) + 1), checker.TCPSynOptions(header.TCPSynOptions{MSS: synOptions.MSS, WS: wndScale, SACKPermitted: synOptions.SACKPermitted && c.SACKEnabled()}), } @@ -1100,6 +1148,7 @@ func (c *Context) PassiveConnectWithOptions(maxPayload, wndScale int, synOptions // Send ACK. c.SendPacket(nil, ackHeaders) + c.RcvdWindowScale = uint8(rcvdSynOptions.WS) c.Port = StackPort return &RawEndpoint{ diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 518f636f0..d57ed5d79 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -996,6 +996,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u // Initialize the UDP header. udp := header.UDP(pkt.TransportHeader().Push(header.UDPMinimumSize)) + pkt.TransportProtocolNumber = ProtocolNumber length := uint16(pkt.Size()) udp.Encode(&header.UDPFields{ @@ -1396,15 +1397,6 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk return } - // Never receive from a multicast address. - if header.IsV4MulticastAddress(id.RemoteAddress) || - header.IsV6MulticastAddress(id.RemoteAddress) { - e.stack.Stats().UDP.InvalidSourceAddress.Increment() - e.stack.Stats().IP.InvalidSourceAddressesReceived.Increment() - e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() - return - } - if !verifyChecksum(r, hdr, pkt) { // Checksum Error. e.stack.Stats().UDP.ChecksumErrors.Increment() diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 7d6b91a75..da5b1deb2 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -12,12 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package udp contains the implementation of the UDP transport protocol. To use -// it in the networking stack, this package must be added to the project, and -// activated on the stack by passing udp.NewProtocol() as one of the -// transport protocols when calling stack.New(). Then endpoints can be created -// by passing udp.ProtocolNumber as the transport protocol number when calling -// Stack.NewEndpoint(). +// Package udp contains the implementation of the UDP transport protocol. package udp import ( @@ -50,6 +45,7 @@ const ( ) type protocol struct { + stack *stack.Stack } // Number returns the udp protocol number. @@ -58,14 +54,14 @@ func (*protocol) Number() tcpip.TransportProtocolNumber { } // NewEndpoint creates a new udp endpoint. -func (*protocol) NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return newEndpoint(stack, netProto, waiterQueue), nil +func (p *protocol) NewEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return newEndpoint(p.stack, netProto, waiterQueue), nil } // NewRawEndpoint creates a new raw UDP endpoint. It implements // stack.TransportProtocol.NewRawEndpoint. -func (p *protocol) NewRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { - return raw.NewEndpoint(stack, netProto, header.UDPProtocolNumber, waiterQueue) +func (p *protocol) NewRawEndpoint(netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return raw.NewEndpoint(p.stack, netProto, header.UDPProtocolNumber, waiterQueue) } // MinimumPacketSize returns the minimum valid udp packet size. @@ -80,126 +76,21 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { return h.SourcePort(), h.DestinationPort(), nil } -// HandleUnknownDestinationPacket handles packets targeted at this protocol but -// that don't match any existing endpoint. -func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) bool { +// HandleUnknownDestinationPacket handles packets that are targeted at this +// protocol but don't match any existing endpoint. +func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.TransportEndpointID, pkt *stack.PacketBuffer) stack.UnknownDestinationPacketDisposition { hdr := header.UDP(pkt.TransportHeader().View()) if int(hdr.Length()) > pkt.Data.Size()+header.UDPMinimumSize { - // Malformed packet. r.Stack().Stats().UDP.MalformedPacketsReceived.Increment() - return true + return stack.UnknownDestinationPacketMalformed } if !verifyChecksum(r, hdr, pkt) { - // Checksum Error. r.Stack().Stats().UDP.ChecksumErrors.Increment() - return true + return stack.UnknownDestinationPacketMalformed } - // Only send ICMP error if the address is not a multicast/broadcast - // v4/v6 address or the source is not the unspecified address. - // - // See: point e) in https://tools.ietf.org/html/rfc4443#section-2.4 - if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) || id.RemoteAddress == header.IPv6Any || id.RemoteAddress == header.IPv4Any { - return true - } - - // As per RFC: 1122 Section 3.2.2.1 A host SHOULD generate Destination - // Unreachable messages with code: - // - // 2 (Protocol Unreachable), when the designated transport protocol - // is not supported; or - // - // 3 (Port Unreachable), when the designated transport protocol - // (e.g., UDP) is unable to demultiplex the datagram but has no - // protocol mechanism to inform the sender. - switch len(id.LocalAddress) { - case header.IPv4AddressSize: - if !r.Stack().AllowICMPMessage() { - r.Stack().Stats().ICMP.V4PacketsSent.RateLimited.Increment() - return true - } - // As per RFC 1812 Section 4.3.2.3 - // - // ICMP datagram SHOULD contain as much of the original - // datagram as possible without the length of the ICMP - // datagram exceeding 576 bytes - // - // NOTE: The above RFC referenced is different from the original - // recommendation in RFC 1122 where it mentioned that at least 8 - // bytes of the payload must be included. Today linux and other - // systems implement the] RFC1812 definition and not the original - // RFC 1122 requirement. - mtu := int(r.MTU()) - if mtu > header.IPv4MinimumProcessableDatagramSize { - mtu = header.IPv4MinimumProcessableDatagramSize - } - headerLen := int(r.MaxHeaderLength()) + header.ICMPv4MinimumSize - available := int(mtu) - headerLen - payloadLen := pkt.NetworkHeader().View().Size() + pkt.TransportHeader().View().Size() + pkt.Data.Size() - if payloadLen > available { - payloadLen = available - } - - // The buffers used by pkt may be used elsewhere in the system. - // For example, a raw or packet socket may use what UDP - // considers an unreachable destination. Thus we deep copy pkt - // to prevent multiple ownership and SR errors. - newHeader := append(buffer.View(nil), pkt.NetworkHeader().View()...) - newHeader = append(newHeader, pkt.TransportHeader().View()...) - payload := newHeader.ToVectorisedView() - payload.AppendView(pkt.Data.ToView()) - payload.CapLength(payloadLen) - - icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - ReserveHeaderBytes: headerLen, - Data: payload, - }) - icmpHdr := header.ICMPv4(icmpPkt.TransportHeader().Push(header.ICMPv4MinimumSize)) - icmpHdr.SetType(header.ICMPv4DstUnreachable) - icmpHdr.SetCode(header.ICMPv4PortUnreachable) - icmpHdr.SetChecksum(header.ICMPv4Checksum(icmpHdr, icmpPkt.Data)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv4ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, icmpPkt) - - case header.IPv6AddressSize: - if !r.Stack().AllowICMPMessage() { - r.Stack().Stats().ICMP.V6PacketsSent.RateLimited.Increment() - return true - } - - // As per RFC 4443 section 2.4 - // - // (c) Every ICMPv6 error message (type < 128) MUST include - // as much of the IPv6 offending (invoking) packet (the - // packet that caused the error) as possible without making - // the error message packet exceed the minimum IPv6 MTU - // [IPv6]. - mtu := int(r.MTU()) - if mtu > header.IPv6MinimumMTU { - mtu = header.IPv6MinimumMTU - } - headerLen := int(r.MaxHeaderLength()) + header.ICMPv6DstUnreachableMinimumSize - available := int(mtu) - headerLen - network, transport := pkt.NetworkHeader().View(), pkt.TransportHeader().View() - payloadLen := len(network) + len(transport) + pkt.Data.Size() - if payloadLen > available { - payloadLen = available - } - payload := buffer.NewVectorisedView(len(network)+len(transport), []buffer.View{network, transport}) - payload.Append(pkt.Data) - payload.CapLength(payloadLen) - - icmpPkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ - ReserveHeaderBytes: headerLen, - Data: payload, - }) - icmpHdr := header.ICMPv6(icmpPkt.TransportHeader().Push(header.ICMPv6DstUnreachableMinimumSize)) - icmpHdr.SetType(header.ICMPv6DstUnreachable) - icmpHdr.SetCode(header.ICMPv6PortUnreachable) - icmpHdr.SetChecksum(header.ICMPv6Checksum(icmpHdr, r.LocalAddress, r.RemoteAddress, icmpPkt.Data)) - r.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: header.ICMPv6ProtocolNumber, TTL: r.DefaultTTL(), TOS: stack.DefaultTOS}, icmpPkt) - } - return true + return stack.UnknownDestinationPacketUnhandled } // SetOption implements stack.TransportProtocol.SetOption. @@ -224,6 +115,6 @@ func (*protocol) Parse(pkt *stack.PacketBuffer) bool { } // NewProtocol returns a UDP transport protocol. -func NewProtocol() stack.TransportProtocol { - return &protocol{} +func NewProtocol(s *stack.Stack) stack.TransportProtocol { + return &protocol{stack: s} } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index d5881d183..b4604ba35 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -294,8 +294,8 @@ type testContext struct { func newDualTestContext(t *testing.T, mtu uint32) *testContext { t.Helper() return newDualTestContextWithOptions(t, mtu, stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) } @@ -388,6 +388,10 @@ func (c *testContext) getPacketAndVerify(flow testFlow, checkers ...checker.Netw c.t.Fatalf("Bad network protocol: got %v, wanted %v", p.Proto, flow.netProto()) } + if got, want := p.Pkt.TransportProtocolNumber, header.UDPProtocolNumber; got != want { + c.t.Errorf("got p.Pkt.TransportProtocolNumber = %d, want = %d", got, want) + } + vv := buffer.NewVectorisedView(p.Pkt.Size(), p.Pkt.Views()) b := vv.ToView() @@ -528,8 +532,8 @@ func newMinPayload(minSize int) []byte { func TestBindToDeviceOption(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}}) ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) if err != nil { @@ -803,8 +807,8 @@ func TestV4ReadSelfSource(t *testing.T) { } { t.Run(tt.name, func(t *testing.T) { c := newDualTestContextWithOptions(t, defaultMTU, stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, HandleLocal: tt.handleLocal, }) defer c.cleanup() @@ -924,42 +928,6 @@ func TestReadFromMulticast(t *testing.T) { } } -// TestReadFromMulticaststats checks that a discarded packet -// that that was sent with multicast SOURCE address increments -// the correct counters and that a regular packet does not. -func TestReadFromMulticastStats(t *testing.T) { - t.Helper() - for _, flow := range []testFlow{reverseMulticast4, reverseMulticast6, unicastV4} { - t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpointForFlow(flow) - - if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { - t.Fatalf("Bind failed: %s", err) - } - - payload := newPayload() - c.injectPacket(flow, payload, false) - - var want uint64 = 0 - if flow.isReverseMulticast() { - want = 1 - } - if got := c.s.Stats().IP.InvalidSourceAddressesReceived.Value(); got != want { - t.Errorf("got stats.IP.InvalidSourceAddressesReceived.Value() = %d, want = %d", got, want) - } - if got := c.s.Stats().UDP.InvalidSourceAddress.Value(); got != want { - t.Errorf("got stats.UDP.InvalidSourceAddress.Value() = %d, want = %d", got, want) - } - if got := c.ep.Stats().(*tcpip.TransportEndpointStats).ReceiveErrors.MalformedPacketsReceived.Value(); got != want { - t.Errorf("got EP Stats.ReceiveErrors.MalformedPacketsReceived stats = %d, want = %d", got, want) - } - }) - } -} - // TestV4ReadBroadcastOnBoundToWildcard checks that an endpoint can bind to ANY // and receive broadcast and unicast data. func TestV4ReadBroadcastOnBoundToWildcard(t *testing.T) { @@ -1462,6 +1430,28 @@ func TestNoChecksum(t *testing.T) { } } +var _ stack.NetworkInterface = (*testInterface)(nil) + +type testInterface struct { + stack.NetworkLinkEndpoint +} + +func (*testInterface) ID() tcpip.NICID { + return 0 +} + +func (*testInterface) IsLoopback() bool { + return false +} + +func (*testInterface) Name() string { + return "" +} + +func (*testInterface) Enabled() bool { + return true +} + func TestTTL(t *testing.T) { for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, multicastV4, multicastV4in6, multicastV6, broadcast, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { @@ -1479,16 +1469,19 @@ func TestTTL(t *testing.T) { if flow.isMulticast() { wantTTL = multicastTTL } else { - var p stack.NetworkProtocol + var p stack.NetworkProtocolFactory + var n tcpip.NetworkProtocolNumber if flow.isV4() { - p = ipv4.NewProtocol() + p = ipv4.NewProtocol + n = ipv4.ProtocolNumber } else { - p = ipv6.NewProtocol() + p = ipv6.NewProtocol + n = ipv6.ProtocolNumber } - ep := p.NewEndpoint(0, nil, nil, nil, nil, stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - })) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocolFactory{p}, + }) + ep := s.NetworkProtocolInstance(n).NewEndpoint(&testInterface{}, nil, nil, nil) wantTTL = ep.DefaultTTL() ep.Close() } @@ -1512,18 +1505,6 @@ func TestSetTTL(t *testing.T) { c.t.Fatalf("SetSockOptInt(TTLOption, %d) failed: %s", wantTTL, err) } - var p stack.NetworkProtocol - if flow.isV4() { - p = ipv4.NewProtocol() - } else { - p = ipv6.NewProtocol() - } - ep := p.NewEndpoint(0, nil, nil, nil, nil, stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, - })) - ep.Close() - testWrite(c, flow, checker.TTL(wantTTL)) }) } @@ -1797,16 +1778,26 @@ func TestV4UnknownDestination(t *testing.T) { checker.ICMPv4Type(header.ICMPv4DstUnreachable), checker.ICMPv4Code(header.ICMPv4PortUnreachable))) + // We need to compare the included data part of the UDP packet that is in + // the ICMP packet with the matching original data. icmpPkt := header.ICMPv4(hdr.Payload()) payloadIPHeader := header.IPv4(icmpPkt.Payload()) + incomingHeaderLength := header.IPv4MinimumSize + header.UDPMinimumSize wantLen := len(payload) if tc.largePayload { - wantLen = header.IPv4MinimumProcessableDatagramSize - header.IPv4MinimumSize*2 - header.ICMPv4MinimumSize - header.UDPMinimumSize + // To work out the data size we need to simulate what the sender would + // have done. The wanted size is the total available minus the sum of + // the headers in the UDP AND ICMP packets, given that we know the test + // had only a minimal IP header but the ICMP sender will have allowed + // for a maximally sized packet header. + wantLen = header.IPv4MinimumProcessableDatagramSize - header.IPv4MaximumHeaderSize - header.ICMPv4MinimumSize - incomingHeaderLength + } - // In case of large payloads the IP packet may be truncated. Update + // In the case of large payloads the IP packet may be truncated. Update // the length field before retrieving the udp datagram payload. - payloadIPHeader.SetTotalLength(uint16(wantLen + header.UDPMinimumSize + header.IPv4MinimumSize)) + // Add back the two headers within the payload. + payloadIPHeader.SetTotalLength(uint16(wantLen + incomingHeaderLength)) origDgram := header.UDP(payloadIPHeader.Payload()) if got, want := len(origDgram.Payload()), wantLen; got != want { @@ -2032,7 +2023,8 @@ func TestPayloadModifiedV4(t *testing.T) { payload := newPayload() h := unicastV4.header4Tuple(incoming) buf := c.buildV4Packet(payload, &h) - // Modify the payload so that the checksum value in the UDP header will be incorrect. + // Modify the payload so that the checksum value in the UDP header will be + // incorrect. buf[len(buf)-1]++ c.linkEP.InjectInbound(ipv4.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), @@ -2062,7 +2054,8 @@ func TestPayloadModifiedV6(t *testing.T) { payload := newPayload() h := unicastV6.header4Tuple(incoming) buf := c.buildV6Packet(payload, &h) - // Modify the payload so that the checksum value in the UDP header will be incorrect. + // Modify the payload so that the checksum value in the UDP header will be + // incorrect. buf[len(buf)-1]++ c.linkEP.InjectInbound(ipv6.ProtocolNumber, stack.NewPacketBuffer(stack.PacketBufferOptions{ Data: buf.ToVectorisedView(), @@ -2353,9 +2346,8 @@ func TestOutgoingSubnetBroadcast(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { s := stack.New(stack.Options{ - NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, - - TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + NetworkProtocols: []stack.NetworkProtocolFactory{ipv4.NewProtocol, ipv6.NewProtocol}, + TransportProtocols: []stack.TransportProtocolFactory{udp.NewProtocol}, }) e := channel.New(0, defaultMTU, "") if err := s.CreateNIC(nicID1, e); err != nil { diff --git a/pkg/test/testutil/testutil.go b/pkg/test/testutil/testutil.go index 06fb823f6..49ab87c58 100644 --- a/pkg/test/testutil/testutil.go +++ b/pkg/test/testutil/testutil.go @@ -270,7 +270,7 @@ func RandomID(prefix string) string { // same name, sometimes between test runs the socket does not get cleaned up // quickly enough, causing container creation to fail. func RandomContainerID() string { - return RandomID("test-container-") + return RandomID("test-container") } // Copy copies file from src to dst. diff --git a/pkg/usermem/usermem.go b/pkg/usermem/usermem.go index 27279b409..9b1e7a085 100644 --- a/pkg/usermem/usermem.go +++ b/pkg/usermem/usermem.go @@ -21,7 +21,6 @@ import ( "io" "strconv" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/gohacks" "gvisor.dev/gvisor/pkg/safemem" @@ -184,51 +183,6 @@ func (rw *IOReadWriter) Write(src []byte) (int, error) { return n, err } -// CopyObjectOut copies a fixed-size value or slice of fixed-size values from -// src to the memory mapped at addr in uio. It returns the number of bytes -// copied. -// -// CopyObjectOut must use reflection to encode src; performance-sensitive -// clients should do encoding manually and use uio.CopyOut directly. -// -// Preconditions: Same as IO.CopyOut. -func CopyObjectOut(ctx context.Context, uio IO, addr Addr, src interface{}, opts IOOpts) (int, error) { - w := &IOReadWriter{ - Ctx: ctx, - IO: uio, - Addr: addr, - Opts: opts, - } - // Allocate a byte slice the size of the object being marshaled. This - // adds an extra reflection call, but avoids needing to grow the slice - // during encoding, which can result in many heap-allocated slices. - b := make([]byte, 0, binary.Size(src)) - return w.Write(binary.Marshal(b, ByteOrder, src)) -} - -// CopyObjectIn copies a fixed-size value or slice of fixed-size values from -// the memory mapped at addr in uio to dst. It returns the number of bytes -// copied. -// -// CopyObjectIn must use reflection to decode dst; performance-sensitive -// clients should use uio.CopyIn directly and do decoding manually. -// -// Preconditions: Same as IO.CopyIn. -func CopyObjectIn(ctx context.Context, uio IO, addr Addr, dst interface{}, opts IOOpts) (int, error) { - r := &IOReadWriter{ - Ctx: ctx, - IO: uio, - Addr: addr, - Opts: opts, - } - buf := make([]byte, binary.Size(dst)) - if _, err := io.ReadFull(r, buf); err != nil { - return 0, err - } - binary.Unmarshal(buf, ByteOrder, dst) - return int(r.Addr - addr), nil -} - // CopyStringIn tuning parameters, defined outside that function for tests. const ( copyStringIncrement = 64 diff --git a/pkg/usermem/usermem_test.go b/pkg/usermem/usermem_test.go index bf3c5df2b..da60b0cc7 100644 --- a/pkg/usermem/usermem_test.go +++ b/pkg/usermem/usermem_test.go @@ -16,7 +16,6 @@ package usermem import ( "bytes" - "encoding/binary" "fmt" "reflect" "strings" @@ -174,23 +173,6 @@ type testStruct struct { Uint64 uint64 } -func TestCopyObject(t *testing.T) { - wantObj := testStruct{1, 2, 3, 4, 5, 6, 7, 8} - wantN := binary.Size(wantObj) - b := &BytesIO{make([]byte, wantN)} - ctx := newContext() - if n, err := CopyObjectOut(ctx, b, 0, &wantObj, IOOpts{}); n != wantN || err != nil { - t.Fatalf("CopyObjectOut: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - var gotObj testStruct - if n, err := CopyObjectIn(ctx, b, 0, &gotObj, IOOpts{}); n != wantN || err != nil { - t.Errorf("CopyObjectIn: got (%v, %v), wanted (%v, nil)", n, err, wantN) - } - if gotObj != wantObj { - t.Errorf("CopyObject round trip: got %+v, wanted %+v", gotObj, wantObj) - } -} - func TestCopyStringInShort(t *testing.T) { // Tests for string length <= copyStringIncrement. want := strings.Repeat("A", copyStringIncrement-2) |