diff options
Diffstat (limited to 'pkg')
68 files changed, 2805 insertions, 249 deletions
diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index d5b731390..2e2cc6be7 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -256,6 +256,17 @@ type SockAddrInet6 struct { Scope_id uint32 } +// SockAddrLink is a struct sockaddr_ll, from uapi/linux/if_packet.h. +type SockAddrLink struct { + Family uint16 + Protocol uint16 + InterfaceIndex int32 + ARPHardwareType uint16 + PacketType byte + HardwareAddrLen byte + HardwareAddr [8]byte +} + // UnixPathMax is the maximum length of the path in an AF_UNIX socket. // // From uapi/linux/un.h. @@ -278,6 +289,7 @@ type SockAddr interface { func (s *SockAddrInet) implementsSockAddr() {} func (s *SockAddrInet6) implementsSockAddr() {} +func (s *SockAddrLink) implementsSockAddr() {} func (s *SockAddrUnix) implementsSockAddr() {} func (s *SockAddrNetlink) implementsSockAddr() {} diff --git a/pkg/sentry/fsimpl/memfs/BUILD b/pkg/sentry/fsimpl/memfs/BUILD index 7e364c5fd..04d667273 100644 --- a/pkg/sentry/fsimpl/memfs/BUILD +++ b/pkg/sentry/fsimpl/memfs/BUILD @@ -24,14 +24,18 @@ go_library( "directory.go", "filesystem.go", "memfs.go", + "named_pipe.go", "regular_file.go", "symlink.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/memfs", deps = [ "//pkg/abi/linux", + "//pkg/amutex", + "//pkg/sentry/arch", "//pkg/sentry/context", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/pipe", "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", @@ -54,3 +58,19 @@ go_test( "//pkg/syserror", ], ) + +go_test( + name = "memfs_test", + size = "small", + srcs = ["pipe_test.go"], + embed = [":memfs"], + deps = [ + "//pkg/abi/linux", + "//pkg/sentry/context", + "//pkg/sentry/context/contexttest", + "//pkg/sentry/kernel/auth", + "//pkg/sentry/usermem", + "//pkg/sentry/vfs", + "//pkg/syserror", + ], +) diff --git a/pkg/sentry/fsimpl/memfs/filesystem.go b/pkg/sentry/fsimpl/memfs/filesystem.go index f79e2d9c8..f006c40cd 100644 --- a/pkg/sentry/fsimpl/memfs/filesystem.go +++ b/pkg/sentry/fsimpl/memfs/filesystem.go @@ -233,7 +233,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v if err != nil { return err } - _, err = checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(rp, parentVFSD, parentInode) if err != nil { return err } @@ -241,8 +241,40 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return err } defer rp.Mount().EndWrite() - // TODO: actually implement mknod - return syserror.EPERM + + switch opts.Mode.FileType() { + case 0: + // "Zero file type is equivalent to type S_IFREG." - mknod(2) + fallthrough + case linux.ModeRegular: + // TODO(b/138862511): Implement. + return syserror.EINVAL + + case linux.ModeNamedPipe: + child := fs.newDentry(fs.newNamedPipe(rp.Credentials(), opts.Mode)) + parentVFSD.InsertChild(&child.vfsd, pc) + parentInode.impl.(*directory).childList.PushBack(child) + return nil + + case linux.ModeSocket: + // TODO(b/138862511): Implement. + return syserror.EINVAL + + case linux.ModeCharacterDevice: + fallthrough + case linux.ModeBlockDevice: + // TODO(b/72101894): We don't support creating block or character + // devices at the moment. + // + // When we start supporting block and character devices, we'll + // need to check for CAP_MKNOD here. + return syserror.EPERM + + default: + // "EINVAL - mode requested creation of something other than a + // regular file, device special file, FIFO or socket." - mknod(2) + return syserror.EINVAL + } } // OpenAt implements vfs.FilesystemImpl.OpenAt. @@ -250,8 +282,9 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // Filter out flags that are not supported by memfs. O_DIRECTORY and // O_NOFOLLOW have no effect here (they're handled by VFS by setting // appropriate bits in rp), but are returned by - // FileDescriptionImpl.StatusFlags(). - opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW + // FileDescriptionImpl.StatusFlags(). O_NONBLOCK is supported only by + // pipes. + opts.Flags &= linux.O_ACCMODE | linux.O_CREAT | linux.O_EXCL | linux.O_TRUNC | linux.O_DIRECTORY | linux.O_NOFOLLOW | linux.O_NONBLOCK if opts.Flags&linux.O_CREAT == 0 { fs.mu.RLock() @@ -260,7 +293,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err != nil { return nil, err } - return inode.open(rp, vfsd, opts.Flags, false) + return inode.open(ctx, rp, vfsd, opts.Flags, false) } mustCreate := opts.Flags&linux.O_EXCL != 0 @@ -275,7 +308,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if mustCreate { return nil, syserror.EEXIST } - return inode.open(rp, vfsd, opts.Flags, false) + return inode.open(ctx, rp, vfsd, opts.Flags, false) } afterTrailingSymlink: // Walk to the parent directory of the last path component. @@ -320,7 +353,7 @@ afterTrailingSymlink: child := fs.newDentry(childInode) vfsd.InsertChild(&child.vfsd, pc) inode.impl.(*directory).childList.PushBack(child) - return childInode.open(rp, &child.vfsd, opts.Flags, true) + return childInode.open(ctx, rp, &child.vfsd, opts.Flags, true) } // Open existing file or follow symlink. if mustCreate { @@ -336,10 +369,10 @@ afterTrailingSymlink: // symlink target. goto afterTrailingSymlink } - return childInode.open(rp, childVFSD, opts.Flags, false) + return childInode.open(ctx, rp, childVFSD, opts.Flags, false) } -func (i *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32, afterCreate bool) (*vfs.FileDescription, error) { +func (i *inode) open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32, afterCreate bool) (*vfs.FileDescription, error) { ats := vfs.AccessTypesForOpenFlags(flags) if !afterCreate { if err := i.checkPermissions(rp.Credentials(), ats, i.isDir()); err != nil { @@ -378,6 +411,8 @@ func (i *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32, afte case *symlink: // Can't open symlinks without O_PATH (which is unimplemented). return nil, syserror.ELOOP + case *namedPipe: + return newNamedPipeFD(ctx, impl, rp, vfsd, flags) default: panic(fmt.Sprintf("unknown inode type: %T", i.impl)) } diff --git a/pkg/sentry/fsimpl/memfs/memfs.go b/pkg/sentry/fsimpl/memfs/memfs.go index b78471c0f..64c851c1a 100644 --- a/pkg/sentry/fsimpl/memfs/memfs.go +++ b/pkg/sentry/fsimpl/memfs/memfs.go @@ -227,6 +227,8 @@ func (i *inode) statTo(stat *linux.Statx) { stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS stat.Size = uint64(len(impl.target)) stat.Blocks = allocatedBlocksForSize(stat.Size) + case *namedPipe: + stat.Mode |= linux.S_IFIFO default: panic(fmt.Sprintf("unknown inode type: %T", i.impl)) } diff --git a/pkg/sentry/fsimpl/memfs/named_pipe.go b/pkg/sentry/fsimpl/memfs/named_pipe.go new file mode 100644 index 000000000..732ed7c58 --- /dev/null +++ b/pkg/sentry/fsimpl/memfs/named_pipe.go @@ -0,0 +1,59 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package memfs + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +type namedPipe struct { + inode inode + + pipe *pipe.VFSPipe +} + +// Preconditions: +// * fs.mu must be locked. +// * rp.Mount().CheckBeginWrite() has been called successfully. +func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { + file := &namedPipe{pipe: pipe.NewVFSPipe(pipe.DefaultPipeSize, usermem.PageSize)} + file.inode.init(file, fs, creds, mode) + file.inode.nlink = 1 // Only the parent has a link. + return &file.inode +} + +// namedPipeFD implements vfs.FileDescriptionImpl. Methods are implemented +// entirely via struct embedding. +type namedPipeFD struct { + fileDescription + + *pipe.VFSPipeFD +} + +func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + var err error + var fd namedPipeFD + fd.VFSPipeFD, err = np.pipe.NewVFSPipeFD(ctx, rp, vfsd, &fd.vfsfd, flags) + if err != nil { + return nil, err + } + fd.vfsfd.Init(&fd, rp.Mount(), vfsd) + return &fd.vfsfd, nil +} diff --git a/pkg/sentry/fsimpl/memfs/pipe_test.go b/pkg/sentry/fsimpl/memfs/pipe_test.go new file mode 100644 index 000000000..0674b81a3 --- /dev/null +++ b/pkg/sentry/fsimpl/memfs/pipe_test.go @@ -0,0 +1,233 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package memfs + +import ( + "bytes" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +const fileName = "mypipe" + +func TestSeparateFDs(t *testing.T) { + ctx, creds, vfsObj, root := setup(t) + defer root.DecRef() + + // Open the read side. This is done in a concurrently because opening + // One end the pipe blocks until the other end is opened. + pop := vfs.PathOperation{ + Root: root, + Start: root, + Pathname: fileName, + FollowFinalSymlink: true, + } + rfdchan := make(chan *vfs.FileDescription) + go func() { + openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY} + rfd, _ := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + rfdchan <- rfd + }() + + // Open the write side. + openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY} + wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != nil { + t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) + } + defer wfd.DecRef() + + rfd, ok := <-rfdchan + if !ok { + t.Fatalf("failed to open pipe for reading %q", fileName) + } + defer rfd.DecRef() + + const msg = "vamos azul" + checkEmpty(ctx, t, rfd) + checkWrite(ctx, t, wfd, msg) + checkRead(ctx, t, rfd, msg) +} + +func TestNonblockingRead(t *testing.T) { + ctx, creds, vfsObj, root := setup(t) + defer root.DecRef() + + // Open the read side as nonblocking. + pop := vfs.PathOperation{ + Root: root, + Start: root, + Pathname: fileName, + FollowFinalSymlink: true, + } + openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_NONBLOCK} + rfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != nil { + t.Fatalf("failed to open pipe for reading %q: %v", fileName, err) + } + defer rfd.DecRef() + + // Open the write side. + openOpts = vfs.OpenOptions{Flags: linux.O_WRONLY} + wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != nil { + t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) + } + defer wfd.DecRef() + + const msg = "geh blau" + checkEmpty(ctx, t, rfd) + checkWrite(ctx, t, wfd, msg) + checkRead(ctx, t, rfd, msg) +} + +func TestNonblockingWriteError(t *testing.T) { + ctx, creds, vfsObj, root := setup(t) + defer root.DecRef() + + // Open the write side as nonblocking, which should return ENXIO. + pop := vfs.PathOperation{ + Root: root, + Start: root, + Pathname: fileName, + FollowFinalSymlink: true, + } + openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY | linux.O_NONBLOCK} + _, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != syserror.ENXIO { + t.Fatalf("expected ENXIO, but got error: %v", err) + } +} + +func TestSingleFD(t *testing.T) { + ctx, creds, vfsObj, root := setup(t) + defer root.DecRef() + + // Open the pipe as readable and writable. + pop := vfs.PathOperation{ + Root: root, + Start: root, + Pathname: fileName, + FollowFinalSymlink: true, + } + openOpts := vfs.OpenOptions{Flags: linux.O_RDWR} + fd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) + if err != nil { + t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) + } + defer fd.DecRef() + + const msg = "forza blu" + checkEmpty(ctx, t, fd) + checkWrite(ctx, t, fd, msg) + checkRead(ctx, t, fd, msg) +} + +// setup creates a VFS with a pipe in the root directory at path fileName. The +// returned VirtualDentry must be DecRef()'d be the caller. It calls t.Fatal +// upon failure. +func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesystem, vfs.VirtualDentry) { + ctx := contexttest.Context(t) + creds := auth.CredentialsFromContext(ctx) + + // Create VFS. + vfsObj := vfs.New() + vfsObj.MustRegisterFilesystemType("memfs", FilesystemType{}) + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "memfs", &vfs.NewFilesystemOptions{}) + if err != nil { + t.Fatalf("failed to create tmpfs root mount: %v", err) + } + + // Create the pipe. + root := mntns.Root() + pop := vfs.PathOperation{ + Root: root, + Start: root, + Pathname: fileName, + FollowFinalSymlink: true, + } + mknodOpts := vfs.MknodOptions{Mode: linux.ModeNamedPipe | 0644} + if err := vfsObj.MknodAt(ctx, creds, &pop, &mknodOpts); err != nil { + t.Fatalf("failed to create file %q: %v", fileName, err) + } + + // Sanity check: the file pipe exists and has the correct mode. + stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Pathname: fileName, + FollowFinalSymlink: true, + }, &vfs.StatOptions{}) + if err != nil { + t.Fatalf("stat(%q) failed: %v", fileName, err) + } + if stat.Mode&^linux.S_IFMT != 0644 { + t.Errorf("got wrong permissions (%0o)", stat.Mode) + } + if stat.Mode&linux.S_IFMT != linux.ModeNamedPipe { + t.Errorf("got wrong file type (%0o)", stat.Mode) + } + + return ctx, creds, vfsObj, root +} + +// checkEmpty calls t.Fatal if the pipe in fd is not empty. +func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { + readData := make([]byte, 1) + dst := usermem.BytesIOSequence(readData) + bytesRead, err := fd.Impl().Read(ctx, dst, vfs.ReadOptions{}) + if err != syserror.ErrWouldBlock { + t.Fatalf("expected ErrWouldBlock reading from empty pipe %q, but got: %v", fileName, err) + } + if bytesRead != 0 { + t.Fatalf("expected to read 0 bytes, but got %d", bytesRead) + } +} + +// checkWrite calls t.Fatal if it fails to write all of msg to fd. +func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { + writeData := []byte(msg) + src := usermem.BytesIOSequence(writeData) + bytesWritten, err := fd.Impl().Write(ctx, src, vfs.WriteOptions{}) + if err != nil { + t.Fatalf("error writing to pipe %q: %v", fileName, err) + } + if bytesWritten != int64(len(writeData)) { + t.Fatalf("expected to write %d bytes, but wrote %d", len(writeData), bytesWritten) + } +} + +// checkRead calls t.Fatal if it fails to read msg from fd. +func checkRead(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { + readData := make([]byte, len(msg)) + dst := usermem.BytesIOSequence(readData) + bytesRead, err := fd.Impl().Read(ctx, dst, vfs.ReadOptions{}) + if err != nil { + t.Fatalf("error reading from pipe %q: %v", fileName, err) + } + if bytesRead != int64(len(msg)) { + t.Fatalf("expected to read %d bytes, but got %d", len(msg), bytesRead) + } + if !bytes.Equal(readData, []byte(msg)) { + t.Fatalf("expected to read %q from pipe, but got %q", msg, string(readData)) + } +} diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index 02320b830..6416e0dd8 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -182,7 +182,7 @@ func (fd *VFSPipeFD) Release() { } // OnClose implements vfs.FileDescriptionImpl.OnClose. -func (fd *VFSPipeFD) OnClose() error { +func (fd *VFSPipeFD) OnClose(_ context.Context) error { return nil } diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 9f0ecfbe4..b699b057d 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -327,6 +327,19 @@ func (t *thread) dumpAndPanic(message string) { panic(message) } +func (t *thread) unexpectedStubExit() { + msg, err := t.getEventMessage() + status := syscall.WaitStatus(msg) + if status.Signaled() && status.Signal() == syscall.SIGKILL { + // SIGKILL can be only sent by an user or OOM-killer. In both + // these cases, we don't need to panic. There is no reasons to + // think that something wrong in gVisor. + log.Warningf("The ptrace stub process %v has been killed by SIGKILL.", t.tgid) + syscall.Kill(os.Getpid(), syscall.SIGKILL) + } + t.dumpAndPanic(fmt.Sprintf("wait failed: the process %d:%d exited: %x (err %v)", t.tgid, t.tid, msg, err)) +} + // wait waits for a stop event. // // Precondition: outcome is a valid waitOutcome. @@ -355,8 +368,7 @@ func (t *thread) wait(outcome waitOutcome) syscall.Signal { } if stopSig == syscall.SIGTRAP { if status.TrapCause() == syscall.PTRACE_EVENT_EXIT { - msg, err := t.getEventMessage() - t.dumpAndPanic(fmt.Sprintf("wait failed: the process %d:%d exited: %x (err %v)", t.tgid, t.tid, msg, err)) + t.unexpectedStubExit() } // Re-encode the trap cause the way it's expected. return stopSig | syscall.Signal(status.TrapCause()<<8) diff --git a/pkg/sentry/platform/ptrace/subprocess_linux.go b/pkg/sentry/platform/ptrace/subprocess_linux.go index c075b5f91..3782d4332 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux.go @@ -129,6 +129,9 @@ func createStub() (*thread, error) { // transitively) will be killed as well. It's simply not possible to // safely handle a single stub getting killed: the exact state of // execution is unknown and not recoverable. + // + // In addition, we set the PTRACE_O_TRACEEXIT option to log more + // information about a stub process when it receives a fatal signal. return attachedThread(uintptr(syscall.SIGKILL)|syscall.CLONE_FILES, defaultAction) } diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 0ae573b45..debf9bce0 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -53,6 +53,7 @@ import ( "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/tcpip/transport/tcp" "gvisor.dev/gvisor/pkg/tcpip/transport/udp" @@ -296,6 +297,7 @@ func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue var sockAddrInetSize = int(binary.Size(linux.SockAddrInet{})) var sockAddrInet6Size = int(binary.Size(linux.SockAddrInet6{})) +var sockAddrLinkSize = int(binary.Size(linux.SockAddrLink{})) // bytesToIPAddress converts an IPv4 or IPv6 address from the user to the // netstack representation taking any addresses into account. @@ -307,12 +309,12 @@ func bytesToIPAddress(addr []byte) tcpip.Address { } // AddressAndFamily reads an sockaddr struct from the given address and -// converts it to the FullAddress format. It supports AF_UNIX, AF_INET and -// AF_INET6 addresses. +// converts it to the FullAddress format. It supports AF_UNIX, AF_INET, +// AF_INET6, and AF_PACKET addresses. // // strict indicates whether addresses with the AF_UNSPEC family are accepted of not. // -// AddressAndFamily returns an address, its family. +// AddressAndFamily returns an address and its family. func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, uint16, *syserr.Error) { // Make sure we have at least 2 bytes for the address family. if len(addr) < 2 { @@ -371,6 +373,22 @@ func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, } return out, family, nil + case linux.AF_PACKET: + var a linux.SockAddrLink + if len(addr) < sockAddrLinkSize { + return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument + } + binary.Unmarshal(addr[:sockAddrLinkSize], usermem.ByteOrder, &a) + if a.Family != linux.AF_PACKET || a.HardwareAddrLen != header.EthernetAddressSize { + return tcpip.FullAddress{}, family, syserr.ErrInvalidArgument + } + + // TODO(b/129292371): Return protocol too. + return tcpip.FullAddress{ + NIC: tcpip.NICID(a.InterfaceIndex), + Addr: tcpip.Address(a.HardwareAddr[:header.EthernetAddressSize]), + }, family, nil + case linux.AF_UNSPEC: return tcpip.FullAddress{}, family, nil @@ -1951,12 +1969,14 @@ func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) return &out, uint32(2 + l) } return &out, uint32(3 + l) + case linux.AF_INET: var out linux.SockAddrInet copy(out.Addr[:], addr.Addr) out.Family = linux.AF_INET out.Port = htons(addr.Port) - return &out, uint32(binary.Size(out)) + return &out, uint32(sockAddrInetSize) + case linux.AF_INET6: var out linux.SockAddrInet6 if len(addr.Addr) == 4 { @@ -1972,7 +1992,17 @@ func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) if isLinkLocal(addr.Addr) { out.Scope_id = uint32(addr.NIC) } - return &out, uint32(binary.Size(out)) + return &out, uint32(sockAddrInet6Size) + + case linux.AF_PACKET: + // TODO(b/129292371): Return protocol too. + var out linux.SockAddrLink + out.Family = linux.AF_PACKET + out.InterfaceIndex = int32(addr.NIC) + out.HardwareAddrLen = header.EthernetAddressSize + copy(out.HardwareAddr[:], addr.Addr) + return &out, uint32(sockAddrLinkSize) + default: return nil, 0 } diff --git a/pkg/sentry/socket/netstack/provider.go b/pkg/sentry/socket/netstack/provider.go index 357a664cc..2d2c1ba2a 100644 --- a/pkg/sentry/socket/netstack/provider.go +++ b/pkg/sentry/socket/netstack/provider.go @@ -62,6 +62,10 @@ func getTransportProtocol(ctx context.Context, stype linux.SockType, protocol in } case linux.SOCK_RAW: + // TODO(b/142504697): "In order to create a raw socket, a + // process must have the CAP_NET_RAW capability in the user + // namespace that governs its network namespace." - raw(7) + // Raw sockets require CAP_NET_RAW. creds := auth.CredentialsFromContext(ctx) if !creds.HasCapability(linux.CAP_NET_RAW) { @@ -85,7 +89,8 @@ func getTransportProtocol(ctx context.Context, stype linux.SockType, protocol in return 0, true, syserr.ErrProtocolNotSupported } -// Socket creates a new socket object for the AF_INET or AF_INET6 family. +// Socket creates a new socket object for the AF_INET, AF_INET6, or AF_PACKET +// family. func (p *provider) Socket(t *kernel.Task, stype linux.SockType, protocol int) (*fs.File, *syserr.Error) { // Fail right away if we don't have a stack. stack := t.NetworkContext() @@ -99,6 +104,12 @@ func (p *provider) Socket(t *kernel.Task, stype linux.SockType, protocol int) (* return nil, nil } + // Packet sockets are handled separately, since they are neither INET + // nor INET6 specific. + if p.family == linux.AF_PACKET { + return packetSocket(t, eps, stype, protocol) + } + // Figure out the transport protocol. transProto, associated, err := getTransportProtocol(t, stype, protocol) if err != nil { @@ -121,12 +132,47 @@ func (p *provider) Socket(t *kernel.Task, stype linux.SockType, protocol int) (* return New(t, p.family, stype, int(transProto), wq, ep) } +func packetSocket(t *kernel.Task, epStack *Stack, stype linux.SockType, protocol int) (*fs.File, *syserr.Error) { + // TODO(b/142504697): "In order to create a packet socket, a process + // must have the CAP_NET_RAW capability in the user namespace that + // governs its network namespace." - packet(7) + + // Packet sockets require CAP_NET_RAW. + creds := auth.CredentialsFromContext(t) + if !creds.HasCapability(linux.CAP_NET_RAW) { + return nil, syserr.ErrNotPermitted + } + + // "cooked" packets don't contain link layer information. + var cooked bool + switch stype { + case linux.SOCK_DGRAM: + cooked = true + case linux.SOCK_RAW: + cooked = false + default: + return nil, syserr.ErrProtocolNotSupported + } + + // protocol is passed in network byte order, but netstack wants it in + // host order. + netProto := tcpip.NetworkProtocolNumber(ntohs(uint16(protocol))) + + wq := &waiter.Queue{} + ep, err := epStack.Stack.NewPacketEndpoint(cooked, netProto, wq) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return New(t, linux.AF_PACKET, stype, protocol, wq, ep) +} + // Pair just returns nil sockets (not supported). func (*provider) Pair(*kernel.Task, linux.SockType, int) (*fs.File, *fs.File, *syserr.Error) { return nil, nil, nil } -// init registers socket providers for AF_INET and AF_INET6. +// init registers socket providers for AF_INET, AF_INET6, and AF_PACKET. func init() { // Providers backed by netstack. p := []provider{ @@ -138,6 +184,9 @@ func init() { family: linux.AF_INET6, netProto: ipv6.ProtocolNumber, }, + { + family: linux.AF_PACKET, + }, } for i := range p { diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index b317cb99d..68589a377 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -16,7 +16,12 @@ package linux const ( - _LINUX_SYSNAME = "Linux" - _LINUX_RELEASE = "4.4.0" - _LINUX_VERSION = "#1 SMP Sun Jan 10 15:06:54 PST 2016" + // LinuxSysname is the OS name advertised by gVisor. + LinuxSysname = "Linux" + + // LinuxRelease is the Linux release version number advertised by gVisor. + LinuxRelease = "4.4.0" + + // LinuxVersion is the version info advertised by gVisor. + LinuxVersion = "#1 SMP Sun Jan 10 15:06:54 PST 2016" ) diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index e215ac049..aedb6d774 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -34,9 +34,9 @@ var AMD64 = &kernel.SyscallTable{ // guides the interface provided by this syscall table. The build // version is that for a clean build with default kernel config, at 5 // minutes after v4.4 was tagged. - Sysname: _LINUX_SYSNAME, - Release: _LINUX_RELEASE, - Version: _LINUX_VERSION, + Sysname: LinuxSysname, + Release: LinuxRelease, + Version: LinuxVersion, }, AuditNumber: linux.AUDIT_ARCH_X86_64, Table: map[uintptr]kernel.Syscall{ @@ -362,7 +362,7 @@ var AMD64 = &kernel.SyscallTable{ 319: syscalls.Supported("memfd_create", MemfdCreate), 320: syscalls.CapError("kexec_file_load", linux.CAP_SYS_BOOT, "", nil), 321: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), - 322: syscalls.ErrorWithEvent("execveat", syserror.ENOSYS, "", []string{"gvisor.dev/issue/265"}), // TODO(b/118901836) + 322: syscalls.PartiallySupported("execveat", Execveat, "No support for AT_EMPTY_PATH, AT_SYMLINK_FOLLOW.", nil), 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(b/118904897) 325: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index f82dfac31..4cf7f836a 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -30,9 +30,9 @@ var ARM64 = &kernel.SyscallTable{ OS: abi.Linux, Arch: arch.ARM64, Version: kernel.Version{ - Sysname: _LINUX_SYSNAME, - Release: _LINUX_RELEASE, - Version: _LINUX_VERSION, + Sysname: LinuxSysname, + Release: LinuxRelease, + Version: LinuxVersion, }, AuditNumber: linux.AUDIT_ARCH_AARCH64, Table: map[uintptr]kernel.Syscall{ diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 8ab7ffa25..6e425f1ec 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -15,10 +15,12 @@ package linux import ( + "path" "syscall" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -67,8 +69,22 @@ func Execve(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal argvAddr := args[1].Pointer() envvAddr := args[2].Pointer() - // Extract our arguments. - filename, err := t.CopyInString(filenameAddr, linux.PATH_MAX) + return execveat(t, linux.AT_FDCWD, filenameAddr, argvAddr, envvAddr, 0) +} + +// Execveat implements linux syscall execveat(2). +func Execveat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + dirFD := args[0].Int() + pathnameAddr := args[1].Pointer() + argvAddr := args[2].Pointer() + envvAddr := args[3].Pointer() + flags := args[4].Int() + + return execveat(t, dirFD, pathnameAddr, argvAddr, envvAddr, flags) +} + +func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr usermem.Addr, flags int32) (uintptr, *kernel.SyscallControl, error) { + pathname, err := t.CopyInString(pathnameAddr, linux.PATH_MAX) if err != nil { return 0, nil, err } @@ -89,14 +105,38 @@ func Execve(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } } + if flags != 0 { + // TODO(b/128449944): Handle AT_EMPTY_PATH and AT_SYMLINK_NOFOLLOW. + t.Kernel().EmitUnimplementedEvent(t) + return 0, nil, syserror.ENOSYS + } + root := t.FSContext().RootDirectory() defer root.DecRef() - wd := t.FSContext().WorkingDirectory() + + var wd *fs.Dirent + if dirFD == linux.AT_FDCWD || path.IsAbs(pathname) { + // If pathname is absolute, LoadTaskImage() will ignore the wd. + wd = t.FSContext().WorkingDirectory() + } else { + // Need to extract the given FD. + f := t.GetFile(dirFD) + if f == nil { + return 0, nil, syserror.EBADF + } + defer f.DecRef() + + wd = f.Dirent + wd.IncRef() + if !fs.IsDir(wd.Inode.StableAttr) { + return 0, nil, syserror.ENOTDIR + } + } defer wd.DecRef() // Load the new TaskContext. maxTraversals := uint(linux.MaxSymlinkTraversals) - tc, se := t.Kernel().LoadTaskImage(t, t.MountNamespace(), root, wd, &maxTraversals, filename, nil, argv, envv, t.Arch().FeatureSet()) + tc, se := t.Kernel().LoadTaskImage(t, t.MountNamespace(), root, wd, &maxTraversals, pathname, nil, argv, envv, t.Arch().FeatureSet()) if se != nil { return 0, nil, se.ToError() } diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go index 27cd2c336..ad4b67806 100644 --- a/pkg/sentry/syscalls/linux/sys_write.go +++ b/pkg/sentry/syscalls/linux/sys_write.go @@ -191,7 +191,6 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca } // Pwritev2 implements linux syscall pwritev2(2). -// TODO(b/120161091): Implement O_SYNC and D_SYNC functionality. func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { // While the syscall is // pwritev2(int fd, struct iovec* iov, int iov_cnt, off_t offset, int flags) diff --git a/pkg/sentry/vfs/syscalls.go b/pkg/sentry/vfs/syscalls.go index 23f2b9e08..abde0feaa 100644 --- a/pkg/sentry/vfs/syscalls.go +++ b/pkg/sentry/vfs/syscalls.go @@ -96,6 +96,26 @@ func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentia } } +// MknodAt creates a file of the given mode at the given path. It returns an +// error from the syserror package. +func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *MknodOptions) error { + rp, err := vfs.getResolvingPath(creds, pop) + if err != nil { + return nil + } + for { + if err = rp.mount.fs.impl.MknodAt(ctx, rp, *opts); err == nil { + vfs.putResolvingPath(rp) + return nil + } + // Handle mount traversals. + if !rp.handleError(err) { + vfs.putResolvingPath(rp) + return err + } + } +} + // OpenAt returns a FileDescription providing access to the file at the given // path. A reference is taken on the returned FileDescription. func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credentials, pop *PathOperation, opts *OpenOptions) (*FileDescription, error) { @@ -198,8 +218,6 @@ func (fd *FileDescription) SetStatusFlags(ctx context.Context, flags uint32) err // // - VFS.LinkAt() // -// - VFS.MknodAt() -// // - VFS.ReadlinkAt() // // - VFS.RenameAt() diff --git a/pkg/tcpip/buffer/prependable.go b/pkg/tcpip/buffer/prependable.go index 4287464f3..48a2a2713 100644 --- a/pkg/tcpip/buffer/prependable.go +++ b/pkg/tcpip/buffer/prependable.go @@ -41,6 +41,11 @@ func NewPrependableFromView(v View) Prependable { return Prependable{buf: v, usedIdx: 0} } +// NewEmptyPrependableFromView creates a new prependable buffer from a View. +func NewEmptyPrependableFromView(v View) Prependable { + return Prependable{buf: v, usedIdx: len(v)} +} + // View returns a View of the backing buffer that contains all prepended // data so far. func (p Prependable) View() View { diff --git a/pkg/tcpip/header/BUILD b/pkg/tcpip/header/BUILD index 07d09abed..a3485b35c 100644 --- a/pkg/tcpip/header/BUILD +++ b/pkg/tcpip/header/BUILD @@ -19,6 +19,7 @@ go_library( "ndp_neighbor_advert.go", "ndp_neighbor_solicit.go", "ndp_options.go", + "ndp_router_advert.go", "tcp.go", "udp.go", ], @@ -36,16 +37,23 @@ go_test( name = "header_x_test", size = "small", srcs = [ + "checksum_test.go", "ipversion_test.go", "tcp_test.go", ], - deps = [":header"], + deps = [ + ":header", + "//pkg/tcpip/buffer", + ], ) go_test( name = "header_test", size = "small", - srcs = ["ndp_test.go"], + srcs = [ + "eth_test.go", + "ndp_test.go", + ], embed = [":header"], deps = ["//pkg/tcpip"], ) diff --git a/pkg/tcpip/header/checksum.go b/pkg/tcpip/header/checksum.go index 39a4d69be..9749c7f4d 100644 --- a/pkg/tcpip/header/checksum.go +++ b/pkg/tcpip/header/checksum.go @@ -23,11 +23,17 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" ) -func calculateChecksum(buf []byte, initial uint32) uint16 { +func calculateChecksum(buf []byte, odd bool, initial uint32) (uint16, bool) { v := initial + if odd { + v += uint32(buf[0]) + buf = buf[1:] + } + l := len(buf) - if l&1 != 0 { + odd = l&1 != 0 + if odd { l-- v += uint32(buf[l]) << 8 } @@ -36,7 +42,7 @@ func calculateChecksum(buf []byte, initial uint32) uint16 { v += (uint32(buf[i]) << 8) + uint32(buf[i+1]) } - return ChecksumCombine(uint16(v), uint16(v>>16)) + return ChecksumCombine(uint16(v), uint16(v>>16)), odd } // Checksum calculates the checksum (as defined in RFC 1071) of the bytes in the @@ -44,7 +50,8 @@ func calculateChecksum(buf []byte, initial uint32) uint16 { // // The initial checksum must have been computed on an even number of bytes. func Checksum(buf []byte, initial uint16) uint16 { - return calculateChecksum(buf, uint32(initial)) + s, _ := calculateChecksum(buf, false, uint32(initial)) + return s } // ChecksumVV calculates the checksum (as defined in RFC 1071) of the bytes in @@ -52,19 +59,40 @@ func Checksum(buf []byte, initial uint16) uint16 { // // The initial checksum must have been computed on an even number of bytes. func ChecksumVV(vv buffer.VectorisedView, initial uint16) uint16 { - var odd bool + return ChecksumVVWithOffset(vv, initial, 0, vv.Size()) +} + +// ChecksumVVWithOffset calculates the checksum (as defined in RFC 1071) of the +// bytes in the given VectorizedView. +// +// The initial checksum must have been computed on an even number of bytes. +func ChecksumVVWithOffset(vv buffer.VectorisedView, initial uint16, off int, size int) uint16 { + odd := false sum := initial for _, v := range vv.Views() { if len(v) == 0 { continue } - s := uint32(sum) - if odd { - s += uint32(v[0]) - v = v[1:] + + if off >= len(v) { + off -= len(v) + continue + } + v = v[off:] + + l := len(v) + if l > size { + l = size + } + v = v[:l] + + sum, odd = calculateChecksum(v, odd, uint32(sum)) + + size -= len(v) + if size == 0 { + break } - odd = len(v)&1 != 0 - sum = calculateChecksum(v, s) + off = 0 } return sum } diff --git a/pkg/tcpip/header/checksum_test.go b/pkg/tcpip/header/checksum_test.go new file mode 100644 index 000000000..86b466c1c --- /dev/null +++ b/pkg/tcpip/header/checksum_test.go @@ -0,0 +1,109 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package header provides the implementation of the encoding and decoding of +// network protocol headers. +package header_test + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +func TestChecksumVVWithOffset(t *testing.T) { + testCases := []struct { + name string + vv buffer.VectorisedView + off, size int + initial uint16 + want uint16 + }{ + { + name: "empty", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}), + }), + off: 0, + size: 0, + want: 0, + }, + { + name: "OneView", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}), + }), + off: 0, + size: 5, + want: 1294, + }, + { + name: "TwoViews", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}), + }), + off: 0, + size: 11, + want: 33819, + }, + { + name: "TwoViewsWithOffset", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}), + }), + off: 1, + size: 11, + want: 33819, + }, + { + name: "ThreeViewsWithOffset", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123}), + }), + off: 7, + size: 11, + want: 33819, + }, + { + name: "ThreeViewsWithInitial", + vv: buffer.NewVectorisedView(0, []buffer.View{ + buffer.NewViewFromBytes([]byte{77, 11, 33, 0, 55, 44}), + buffer.NewViewFromBytes([]byte{98, 1, 9, 0, 5, 4}), + buffer.NewViewFromBytes([]byte{4, 3, 7, 1, 2, 123, 99}), + }), + initial: 77, + off: 7, + size: 11, + want: 33896, + }, + } + for _, tc := range testCases { + t.Run(tc.name, func(t *testing.T) { + if got, want := header.ChecksumVVWithOffset(tc.vv, tc.initial, tc.off, tc.size), tc.want; got != want { + t.Errorf("header.ChecksumVVWithOffset(%v) = %v, want: %v", tc, got, tc.want) + } + v := tc.vv.ToView() + v.TrimFront(tc.off) + v.CapLength(tc.size) + if got, want := header.Checksum(v, tc.initial), tc.want; got != want { + t.Errorf("header.Checksum(%v) = %v, want: %v", tc, got, tc.want) + } + }) + } +} diff --git a/pkg/tcpip/header/eth.go b/pkg/tcpip/header/eth.go index 4c3d3311f..f5d2c127f 100644 --- a/pkg/tcpip/header/eth.go +++ b/pkg/tcpip/header/eth.go @@ -48,8 +48,48 @@ const ( // EthernetAddressSize is the size, in bytes, of an ethernet address. EthernetAddressSize = 6 + + // unspecifiedEthernetAddress is the unspecified ethernet address + // (all bits set to 0). + unspecifiedEthernetAddress = tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00") + + // unicastMulticastFlagMask is the mask of the least significant bit in + // the first octet (in network byte order) of an ethernet address that + // determines whether the ethernet address is a unicast or multicast. If + // the masked bit is a 1, then the address is a multicast, unicast + // otherwise. + // + // See the IEEE Std 802-2001 document for more details. Specifically, + // section 9.2.1 of http://ieee802.org/secmail/pdfocSP2xXA6d.pdf: + // "A 48-bit universal address consists of two parts. The first 24 bits + // correspond to the OUI as assigned by the IEEE, expect that the + // assignee may set the LSB of the first octet to 1 for group addresses + // or set it to 0 for individual addresses." + unicastMulticastFlagMask = 1 + + // unicastMulticastFlagByteIdx is the byte that holds the + // unicast/multicast flag. See unicastMulticastFlagMask. + unicastMulticastFlagByteIdx = 0 +) + +const ( + // EthernetProtocolAll is a catch-all for all protocols carried inside + // an ethernet frame. It is mainly used to create packet sockets that + // capture all traffic. + EthernetProtocolAll tcpip.NetworkProtocolNumber = 0x0003 + + // EthernetProtocolPUP is the PARC Universial Packet protocol ethertype. + EthernetProtocolPUP tcpip.NetworkProtocolNumber = 0x0200 ) +// Ethertypes holds the protocol numbers describing the payload of an ethernet +// frame. These types aren't necessarily supported by netstack, but can be used +// to catch all traffic of a type via packet endpoints. +var Ethertypes = []tcpip.NetworkProtocolNumber{ + EthernetProtocolAll, + EthernetProtocolPUP, +} + // SourceAddress returns the "MAC source" field of the ethernet frame header. func (b Ethernet) SourceAddress() tcpip.LinkAddress { return tcpip.LinkAddress(b[srcMAC:][:EthernetAddressSize]) @@ -72,3 +112,25 @@ func (b Ethernet) Encode(e *EthernetFields) { copy(b[srcMAC:][:EthernetAddressSize], e.SrcAddr) copy(b[dstMAC:][:EthernetAddressSize], e.DstAddr) } + +// IsValidUnicastEthernetAddress returns true if addr is a valid 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 new file mode 100644 index 000000000..6634c90f5 --- /dev/null +++ b/pkg/tcpip/header/eth_test.go @@ -0,0 +1,68 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package header + +import ( + "testing" + + "gvisor.dev/gvisor/pkg/tcpip" +) + +func TestIsValidUnicastEthernetAddress(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"), + false, + }, + { + "Valid", + tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06"), + true, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + if got := IsValidUnicastEthernetAddress(test.addr); got != test.expected { + t.Fatalf("got IsValidUnicastEthernetAddress = %t, want = %t", got, test.expected) + } + }) + } +} diff --git a/pkg/tcpip/header/ipv6.go b/pkg/tcpip/header/ipv6.go index b125bbea5..f1e60911b 100644 --- a/pkg/tcpip/header/ipv6.go +++ b/pkg/tcpip/header/ipv6.go @@ -101,6 +101,15 @@ var IPv6EmptySubnet = func() tcpip.Subnet { return subnet }() +// IPv6LinkLocalPrefix is the prefix for IPv6 link-local addresses, as defined +// by RFC 4291 section 2.5.6. +// +// The prefix is fe80::/64 +var IPv6LinkLocalPrefix = tcpip.AddressWithPrefix{ + Address: "\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00", + PrefixLen: 64, +} + // PayloadLength returns the value of the "payload length" field of the ipv6 // header. func (b IPv6) PayloadLength() uint16 { diff --git a/pkg/tcpip/header/ndp_neighbor_advert.go b/pkg/tcpip/header/ndp_neighbor_advert.go index 5c2b472c8..505c92668 100644 --- a/pkg/tcpip/header/ndp_neighbor_advert.go +++ b/pkg/tcpip/header/ndp_neighbor_advert.go @@ -18,6 +18,8 @@ import "gvisor.dev/gvisor/pkg/tcpip" // NDPNeighborAdvert is an NDP Neighbor Advertisement message. It will // only contain the body of an ICMPv6 packet. +// +// See RFC 4861 section 4.4 for more details. type NDPNeighborAdvert []byte const ( diff --git a/pkg/tcpip/header/ndp_neighbor_solicit.go b/pkg/tcpip/header/ndp_neighbor_solicit.go index 1dcb0fbc6..3a1b8e139 100644 --- a/pkg/tcpip/header/ndp_neighbor_solicit.go +++ b/pkg/tcpip/header/ndp_neighbor_solicit.go @@ -18,6 +18,8 @@ import "gvisor.dev/gvisor/pkg/tcpip" // NDPNeighborSolicit is an NDP Neighbor Solicitation message. It will only // contain the body of an ICMPv6 packet. +// +// See RFC 4861 section 4.3 for more details. type NDPNeighborSolicit []byte const ( diff --git a/pkg/tcpip/header/ndp_router_advert.go b/pkg/tcpip/header/ndp_router_advert.go new file mode 100644 index 000000000..bf7610863 --- /dev/null +++ b/pkg/tcpip/header/ndp_router_advert.go @@ -0,0 +1,112 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package header + +import ( + "encoding/binary" + "time" +) + +// NDPRouterAdvert is an NDP Router Advertisement message. It will only contain +// the body of an ICMPv6 packet. +// +// See RFC 4861 section 4.2 for more details. +type NDPRouterAdvert []byte + +const ( + // NDPRAMinimumSize is the minimum size of a valid NDP Router + // Advertisement message (body of an ICMPv6 packet). + NDPRAMinimumSize = 12 + + // ndpRACurrHopLimitOffset is the byte of the Curr Hop Limit field + // within an NDPRouterAdvert. + ndpRACurrHopLimitOffset = 0 + + // ndpRAFlagsOffset is the byte with the NDP RA bit-fields/flags + // within an NDPRouterAdvert. + ndpRAFlagsOffset = 1 + + // ndpRAManagedAddrConfFlagMask is the mask of the Managed Address + // Configuration flag within the bit-field/flags byte of an + // NDPRouterAdvert. + ndpRAManagedAddrConfFlagMask = (1 << 7) + + // ndpRAOtherConfFlagMask is the mask of the Other Configuration flag + // within the bit-field/flags byte of an NDPRouterAdvert. + ndpRAOtherConfFlagMask = (1 << 6) + + // ndpRARouterLifetimeOffset is the start of the 2-byte Router Lifetime + // field within an NDPRouterAdvert. + ndpRARouterLifetimeOffset = 2 + + // ndpRAReachableTimeOffset is the start of the 4-byte Reachable Time + // field within an NDPRouterAdvert. + ndpRAReachableTimeOffset = 4 + + // ndpRARetransTimerOffset is the start of the 4-byte Retrans Timer + // field within an NDPRouterAdvert. + ndpRARetransTimerOffset = 8 + + // ndpRAOptionsOffset is the start of the NDP options in an + // NDPRouterAdvert. + ndpRAOptionsOffset = 12 +) + +// CurrHopLimit returns the value of the Curr Hop Limit field. +func (b NDPRouterAdvert) CurrHopLimit() uint8 { + return b[ndpRACurrHopLimitOffset] +} + +// ManagedAddrConfFlag returns the value of the Managed Address Configuration +// flag. +func (b NDPRouterAdvert) ManagedAddrConfFlag() bool { + return b[ndpRAFlagsOffset]&ndpRAManagedAddrConfFlagMask != 0 +} + +// OtherConfFlag returns the value of the Other Configuration flag. +func (b NDPRouterAdvert) OtherConfFlag() bool { + return b[ndpRAFlagsOffset]&ndpRAOtherConfFlagMask != 0 +} + +// RouterLifetime returns the lifetime associated with the default router. A +// value of 0 means the source of the Router Advertisement is not a default +// router and SHOULD NOT appear on the default router list. Note, a value of 0 +// only means that the router should not be used as a default router, it does +// not apply to other information contained in the Router Advertisement. +func (b NDPRouterAdvert) RouterLifetime() time.Duration { + // The field is the time in seconds, as per RFC 4861 section 4.2. + return time.Second * time.Duration(binary.BigEndian.Uint16(b[ndpRARouterLifetimeOffset:])) +} + +// ReachableTime returns the time that a node assumes a neighbor is reachable +// after having received a reachability confirmation. A value of 0 means +// that it is unspecified by the source of the Router Advertisement message. +func (b NDPRouterAdvert) ReachableTime() time.Duration { + // The field is the time in milliseconds, as per RFC 4861 section 4.2. + return time.Millisecond * time.Duration(binary.BigEndian.Uint32(b[ndpRAReachableTimeOffset:])) +} + +// RetransTimer returns the time between retransmitted Neighbor Solicitation +// messages. A value of 0 means that it is unspecified by the source of the +// Router Advertisement message. +func (b NDPRouterAdvert) RetransTimer() time.Duration { + // The field is the time in milliseconds, as per RFC 4861 section 4.2. + return time.Millisecond * time.Duration(binary.BigEndian.Uint32(b[ndpRARetransTimerOffset:])) +} + +// Options returns an NDPOptions of the the options body. +func (b NDPRouterAdvert) Options() NDPOptions { + return NDPOptions(b[ndpRAOptionsOffset:]) +} diff --git a/pkg/tcpip/header/ndp_test.go b/pkg/tcpip/header/ndp_test.go index a431a6e61..0aac14f43 100644 --- a/pkg/tcpip/header/ndp_test.go +++ b/pkg/tcpip/header/ndp_test.go @@ -17,6 +17,7 @@ package header import ( "bytes" "testing" + "time" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -117,6 +118,40 @@ func TestNDPNeighborAdvert(t *testing.T) { } } +func TestNDPRouterAdvert(t *testing.T) { + b := []byte{ + 64, 128, 1, 2, + 3, 4, 5, 6, + 7, 8, 9, 10, + } + + ra := NDPRouterAdvert(b) + + if got := ra.CurrHopLimit(); got != 64 { + t.Fatalf("got ra.CurrHopLimit = %d, want = 64", got) + } + + if got := ra.ManagedAddrConfFlag(); !got { + t.Fatalf("got ManagedAddrConfFlag = false, want = true") + } + + if got := ra.OtherConfFlag(); got { + t.Fatalf("got OtherConfFlag = true, want = false") + } + + if got, want := ra.RouterLifetime(), time.Second*258; got != want { + t.Fatalf("got ra.RouterLifetime = %d, want = %d", got, want) + } + + if got, want := ra.ReachableTime(), time.Millisecond*50595078; got != want { + t.Fatalf("got ra.ReachableTime = %d, want = %d", got, want) + } + + if got, want := ra.RetransTimer(), time.Millisecond*117967114; got != want { + t.Fatalf("got ra.RetransTimer = %d, want = %d", got, want) + } +} + // TestNDPTargetLinkLayerAddressOptionSerialize tests serializing a // NDPTargetLinkLayerAddressOption. func TestNDPTargetLinkLayerAddressOptionSerialize(t *testing.T) { diff --git a/pkg/tcpip/link/channel/channel.go b/pkg/tcpip/link/channel/channel.go index 18adb2085..14f197a77 100644 --- a/pkg/tcpip/link/channel/channel.go +++ b/pkg/tcpip/link/channel/channel.go @@ -72,7 +72,7 @@ func (e *Endpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.Vector // InjectLinkAddr injects an inbound packet with a remote link address. func (e *Endpoint) InjectLinkAddr(protocol tcpip.NetworkProtocolNumber, remote tcpip.LinkAddress, vv buffer.VectorisedView) { - e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, vv.Clone(nil)) + e.dispatcher.DeliverNetworkPacket(e, remote, "" /* local */, protocol, vv.Clone(nil), nil /* linkHeader */) } // Attach saves the stack network-layer dispatcher for use later when packets @@ -96,7 +96,7 @@ func (e *Endpoint) MTU() uint32 { func (e *Endpoint) Capabilities() stack.LinkEndpointCapabilities { caps := stack.LinkEndpointCapabilities(0) if e.GSO { - caps |= stack.CapabilityGSO + caps |= stack.CapabilityHardwareGSO } return caps } @@ -134,5 +134,49 @@ func (e *Endpoint) WritePacket(_ *stack.Route, gso *stack.GSO, hdr buffer.Prepen return nil } +// WritePackets stores outbound packets into the channel. +func (e *Endpoint) WritePackets(_ *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + payloadView := payload.ToView() + n := 0 +packetLoop: + for i := range hdrs { + hdr := &hdrs[i].Hdr + off := hdrs[i].Off + size := hdrs[i].Size + p := PacketInfo{ + Header: hdr.View(), + Proto: protocol, + Payload: buffer.NewViewFromBytes(payloadView[off : off+size]), + GSO: gso, + } + + select { + case e.C <- p: + n++ + default: + break packetLoop + } + } + + return n, nil +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + p := PacketInfo{ + Header: packet.ToView(), + Proto: 0, + Payload: buffer.View{}, + GSO: nil, + } + + select { + case e.C <- p: + default: + } + + return nil +} + // Wait implements stack.LinkEndpoint.Wait. func (*Endpoint) Wait() {} diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index f80ac3435..ae4858529 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -165,6 +165,9 @@ type Options struct { // disabled. GSOMaxSize uint32 + // SoftwareGSOEnabled indicates whether software GSO is enabled or not. + SoftwareGSOEnabled bool + // PacketDispatchMode specifies the type of inbound dispatcher to be // used for this endpoint. PacketDispatchMode PacketDispatchMode @@ -242,7 +245,11 @@ func New(opts *Options) (stack.LinkEndpoint, error) { } if isSocket { if opts.GSOMaxSize != 0 { - e.caps |= stack.CapabilityGSO + if opts.SoftwareGSOEnabled { + e.caps |= stack.CapabilitySoftwareGSO + } else { + e.caps |= stack.CapabilityHardwareGSO + } e.gsoMaxSize = opts.GSOMaxSize } } @@ -397,7 +404,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen eth.Encode(ethHdr) } - if e.Capabilities()&stack.CapabilityGSO != 0 { + if e.Capabilities()&stack.CapabilityHardwareGSO != 0 { vnetHdr := virtioNetHdr{} vnetHdrBuf := vnetHdrToByteSlice(&vnetHdr) if gso != nil { @@ -430,8 +437,130 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return rawfile.NonBlockingWrite3(e.fds[0], hdr.View(), payload.ToView(), nil) } -// WriteRawPacket writes a raw packet directly to the file descriptor. -func (e *endpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error { +// WritePackets writes outbound packets to the file descriptor. If it is not +// currently writable, the packet is dropped. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + var ethHdrBuf []byte + // hdr + data + iovLen := 2 + if e.hdrSize > 0 { + // Add ethernet header if needed. + ethHdrBuf = make([]byte, header.EthernetMinimumSize) + eth := header.Ethernet(ethHdrBuf) + ethHdr := &header.EthernetFields{ + DstAddr: r.RemoteLinkAddress, + Type: protocol, + } + + // Preserve the src address if it's set in the route. + if r.LocalLinkAddress != "" { + ethHdr.SrcAddr = r.LocalLinkAddress + } else { + ethHdr.SrcAddr = e.addr + } + eth.Encode(ethHdr) + iovLen++ + } + + n := len(hdrs) + + views := payload.Views() + /* + * Each bondary in views can add one more iovec. + * + * payload | | | | + * ----------------------------- + * packets | | | | | | | + * ----------------------------- + * iovecs | | | | | | | | | + */ + iovec := make([]syscall.Iovec, n*iovLen+len(views)-1) + mmsgHdrs := make([]rawfile.MMsgHdr, n) + + iovecIdx := 0 + viewIdx := 0 + viewOff := 0 + off := 0 + nextOff := 0 + for i := range hdrs { + prevIovecIdx := iovecIdx + mmsgHdr := &mmsgHdrs[i] + mmsgHdr.Msg.Iov = &iovec[iovecIdx] + packetSize := hdrs[i].Size + hdr := &hdrs[i].Hdr + + off = hdrs[i].Off + if off != nextOff { + // We stop in a different point last time. + size := packetSize + viewIdx = 0 + viewOff = 0 + for size > 0 { + if size >= len(views[viewIdx]) { + viewIdx++ + viewOff = 0 + size -= len(views[viewIdx]) + } else { + viewOff = size + size = 0 + } + } + } + nextOff = off + packetSize + + if ethHdrBuf != nil { + v := &iovec[iovecIdx] + v.Base = ðHdrBuf[0] + v.Len = uint64(len(ethHdrBuf)) + iovecIdx++ + } + + v := &iovec[iovecIdx] + hdrView := hdr.View() + v.Base = &hdrView[0] + v.Len = uint64(len(hdrView)) + iovecIdx++ + + for packetSize > 0 { + vec := &iovec[iovecIdx] + iovecIdx++ + + v := views[viewIdx] + vec.Base = &v[viewOff] + s := len(v) - viewOff + if s <= packetSize { + viewIdx++ + viewOff = 0 + } else { + s = packetSize + viewOff += s + } + vec.Len = uint64(s) + packetSize -= s + } + + mmsgHdr.Msg.Iovlen = uint64(iovecIdx - prevIovecIdx) + } + + packets := 0 + for packets < n { + sent, err := rawfile.NonBlockingSendMMsg(e.fds[0], mmsgHdrs) + if err != nil { + return packets, err + } + packets += sent + mmsgHdrs = mmsgHdrs[sent:] + } + return packets, nil +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + return rawfile.NonBlockingWrite(e.fds[0], packet.ToView()) +} + +// InjectOutobund implements stack.InjectableEndpoint.InjectOutbound. +func (e *endpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { return rawfile.NonBlockingWrite(e.fds[0], packet) } @@ -468,9 +597,9 @@ func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) { e.dispatcher = dispatcher } -// Inject injects an inbound packet. -func (e *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv) +// InjectInbound injects an inbound packet. +func (e *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv, nil /* linkHeader */) } // NewInjectable creates a new fd-based InjectableEndpoint. diff --git a/pkg/tcpip/link/fdbased/endpoint_test.go b/pkg/tcpip/link/fdbased/endpoint_test.go index 04406bc9a..59378b96c 100644 --- a/pkg/tcpip/link/fdbased/endpoint_test.go +++ b/pkg/tcpip/link/fdbased/endpoint_test.go @@ -43,9 +43,10 @@ const ( ) type packetInfo struct { - raddr tcpip.LinkAddress - proto tcpip.NetworkProtocolNumber - contents buffer.View + raddr tcpip.LinkAddress + proto tcpip.NetworkProtocolNumber + contents buffer.View + linkHeader buffer.View } type context struct { @@ -92,8 +93,8 @@ func (c *context) cleanup() { syscall.Close(c.fds[1]) } -func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - c.ch <- packetInfo{remote, protocol, vv.ToView()} +func (c *context) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote tcpip.LinkAddress, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) { + c.ch <- packetInfo{remote, protocol, vv.ToView(), linkHeader} } func TestNoEthernetProperties(t *testing.T) { @@ -293,11 +294,12 @@ func TestDeliverPacket(t *testing.T) { b[i] = uint8(rand.Intn(256)) } + var hdr header.Ethernet if !eth { // So that it looks like an IPv4 packet. b[0] = 0x40 } else { - hdr := make(header.Ethernet, header.EthernetMinimumSize) + hdr = make(header.Ethernet, header.EthernetMinimumSize) hdr.Encode(&header.EthernetFields{ SrcAddr: raddr, DstAddr: laddr, @@ -315,9 +317,10 @@ func TestDeliverPacket(t *testing.T) { select { case pi := <-c.ch: want := packetInfo{ - raddr: raddr, - proto: proto, - contents: b, + raddr: raddr, + proto: proto, + contents: b, + linkHeader: buffer.View(hdr), } if !eth { want.proto = header.IPv4ProtocolNumber diff --git a/pkg/tcpip/link/fdbased/mmap.go b/pkg/tcpip/link/fdbased/mmap.go index 8bfeb97e4..554d45715 100644 --- a/pkg/tcpip/link/fdbased/mmap.go +++ b/pkg/tcpip/link/fdbased/mmap.go @@ -169,9 +169,10 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(pkt) + eth = header.Ethernet(pkt) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -189,6 +190,6 @@ func (d *packetMMapDispatcher) dispatch() (bool, *tcpip.Error) { } pkt = pkt[d.e.hdrSize:] - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, buffer.NewVectorisedView(len(pkt), []buffer.View{buffer.View(pkt)})) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, buffer.NewVectorisedView(len(pkt), []buffer.View{buffer.View(pkt)}), buffer.View(eth)) return true, nil } diff --git a/pkg/tcpip/link/fdbased/packet_dispatchers.go b/pkg/tcpip/link/fdbased/packet_dispatchers.go index 7ca217e5b..12168a1dc 100644 --- a/pkg/tcpip/link/fdbased/packet_dispatchers.go +++ b/pkg/tcpip/link/fdbased/packet_dispatchers.go @@ -53,7 +53,7 @@ func newReadVDispatcher(fd int, e *endpoint) (linkDispatcher, error) { d := &readVDispatcher{fd: fd, e: e} d.views = make([]buffer.View, len(BufConfig)) iovLen := len(BufConfig) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { iovLen++ } d.iovecs = make([]syscall.Iovec, iovLen) @@ -63,7 +63,7 @@ func newReadVDispatcher(fd int, e *endpoint) (linkDispatcher, error) { func (d *readVDispatcher) allocateViews(bufConfig []int) { var vnetHdr [virtioNetHdrSize]byte vnetHdrOff := 0 - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // The kernel adds virtioNetHdr before each packet, but // we don't use it, so so we allocate a buffer for it, // add it in iovecs but don't add it in a view. @@ -106,7 +106,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { if err != nil { return false, err } - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // Skip virtioNetHdr which is added before each packet, it // isn't used and it isn't in a view. n -= virtioNetHdrSize @@ -118,9 +118,10 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(d.views[0]) + eth = header.Ethernet(d.views[0][:header.EthernetMinimumSize]) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -141,7 +142,7 @@ func (d *readVDispatcher) dispatch() (bool, *tcpip.Error) { vv := buffer.NewVectorisedView(n, d.views[:used]) vv.TrimFront(d.e.hdrSize) - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv, buffer.View(eth)) // Prepare e.views for another packet: release used views. for i := 0; i < used; i++ { @@ -194,7 +195,7 @@ func newRecvMMsgDispatcher(fd int, e *endpoint) (linkDispatcher, error) { } d.iovecs = make([][]syscall.Iovec, MaxMsgsPerRecv) iovLen := len(BufConfig) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // virtioNetHdr is prepended before each packet. iovLen++ } @@ -225,7 +226,7 @@ func (d *recvMMsgDispatcher) allocateViews(bufConfig []int) { for k := 0; k < len(d.views); k++ { var vnetHdr [virtioNetHdrSize]byte vnetHdrOff := 0 - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { // The kernel adds virtioNetHdr before each packet, but // we don't use it, so so we allocate a buffer for it, // add it in iovecs but don't add it in a view. @@ -261,7 +262,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { // Process each of received packets. for k := 0; k < nMsgs; k++ { n := int(d.msgHdrs[k].Len) - if d.e.Capabilities()&stack.CapabilityGSO != 0 { + if d.e.Capabilities()&stack.CapabilityHardwareGSO != 0 { n -= virtioNetHdrSize } if n <= d.e.hdrSize { @@ -271,9 +272,10 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { var ( p tcpip.NetworkProtocolNumber remote, local tcpip.LinkAddress + eth header.Ethernet ) if d.e.hdrSize > 0 { - eth := header.Ethernet(d.views[k][0]) + eth = header.Ethernet(d.views[k][0]) p = eth.Type() remote = eth.SourceAddress() local = eth.DestinationAddress() @@ -293,7 +295,7 @@ func (d *recvMMsgDispatcher) dispatch() (bool, *tcpip.Error) { used := d.capViews(k, int(n), BufConfig) vv := buffer.NewVectorisedView(int(n), d.views[k][:used]) vv.TrimFront(d.e.hdrSize) - d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv) + d.e.dispatcher.DeliverNetworkPacket(d.e, remote, local, p, vv, buffer.View(eth)) // Prepare e.views for another packet: release used views. for i := 0; i < used; i++ { diff --git a/pkg/tcpip/link/loopback/BUILD b/pkg/tcpip/link/loopback/BUILD index 47a54845c..23e4d1418 100644 --- a/pkg/tcpip/link/loopback/BUILD +++ b/pkg/tcpip/link/loopback/BUILD @@ -10,6 +10,7 @@ go_library( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/header", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/link/loopback/loopback.go b/pkg/tcpip/link/loopback/loopback.go index b36629d2c..a3b48fa73 100644 --- a/pkg/tcpip/link/loopback/loopback.go +++ b/pkg/tcpip/link/loopback/loopback.go @@ -23,6 +23,7 @@ package loopback import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -70,6 +71,9 @@ func (*endpoint) LinkAddress() tcpip.LinkAddress { return "" } +// Wait implements stack.LinkEndpoint.Wait. +func (*endpoint) Wait() {} + // WritePacket implements stack.LinkEndpoint.WritePacket. It delivers outbound // packets to the network-layer dispatcher. func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { @@ -81,10 +85,27 @@ func (e *endpoint) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prependa // Because we're immediately turning around and writing the packet back to the // rx path, we intentionally don't preserve the remote and local link // addresses from the stack.Route we're passed. - e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv) + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv, nil /* linkHeader */) return nil } -// Wait implements stack.LinkEndpoint.Wait. -func (*endpoint) Wait() {} +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(_ *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + // Reject the packet if it's shorter than an ethernet header. + if packet.Size() < header.EthernetMinimumSize { + return tcpip.ErrBadAddress + } + + // There should be an ethernet header at the beginning of packet. + linkHeader := header.Ethernet(packet.First()[:header.EthernetMinimumSize]) + packet.TrimFront(len(linkHeader)) + e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, linkHeader.Type(), packet, buffer.View(linkHeader)) + + return nil +} diff --git a/pkg/tcpip/link/muxed/injectable.go b/pkg/tcpip/link/muxed/injectable.go index 7c946101d..682b60291 100644 --- a/pkg/tcpip/link/muxed/injectable.go +++ b/pkg/tcpip/link/muxed/injectable.go @@ -79,29 +79,47 @@ func (m *InjectableEndpoint) IsAttached() bool { return m.dispatcher != nil } -// Inject implements stack.InjectableLinkEndpoint. -func (m *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { - m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, vv) +// InjectInbound implements stack.InjectableLinkEndpoint. +func (m *InjectableEndpoint) InjectInbound(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { + m.dispatcher.DeliverNetworkPacket(m, "" /* remote */, "" /* local */, protocol, vv, nil /* linkHeader */) +} + +// WritePackets writes outbound packets to the appropriate +// LinkInjectableEndpoint based on the RemoteAddress. HandleLocal only works if +// r.RemoteAddress has a route registered in this endpoint. +func (m *InjectableEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + endpoint, ok := m.routes[r.RemoteAddress] + if !ok { + return 0, tcpip.ErrNoRoute + } + return endpoint.WritePackets(r, gso, hdrs, payload, protocol) } // WritePacket writes outbound packets to the appropriate LinkInjectableEndpoint // based on the RemoteAddress. HandleLocal only works if r.RemoteAddress has a // route registered in this endpoint. -func (m *InjectableEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (m *InjectableEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { if endpoint, ok := m.routes[r.RemoteAddress]; ok { - return endpoint.WritePacket(r, nil /* gso */, hdr, payload, protocol) + return endpoint.WritePacket(r, gso, hdr, payload, protocol) } return tcpip.ErrNoRoute } -// WriteRawPacket writes outbound packets to the appropriate +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (m *InjectableEndpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + // WriteRawPacket doesn't get a route or network address, so there's + // nowhere to write this. + return tcpip.ErrNoRoute +} + +// InjectOutbound writes outbound packets to the appropriate // LinkInjectableEndpoint based on the dest address. -func (m *InjectableEndpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error { +func (m *InjectableEndpoint) InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error { endpoint, ok := m.routes[dest] if !ok { return tcpip.ErrNoRoute } - return endpoint.WriteRawPacket(dest, packet) + return endpoint.InjectOutbound(dest, packet) } // Wait implements stack.LinkEndpoint.Wait. diff --git a/pkg/tcpip/link/muxed/injectable_test.go b/pkg/tcpip/link/muxed/injectable_test.go index 3086fec00..9cd300af8 100644 --- a/pkg/tcpip/link/muxed/injectable_test.go +++ b/pkg/tcpip/link/muxed/injectable_test.go @@ -31,7 +31,7 @@ import ( func TestInjectableEndpointRawDispatch(t *testing.T) { endpoint, sock, dstIP := makeTestInjectableEndpoint(t) - endpoint.WriteRawPacket(dstIP, []byte{0xFA}) + endpoint.InjectOutbound(dstIP, []byte{0xFA}) buf := make([]byte, ipv4.MaxTotalSize) bytesRead, err := sock.Read(buf) diff --git a/pkg/tcpip/link/rawfile/BUILD b/pkg/tcpip/link/rawfile/BUILD index 2e8bc772a..05c7b8024 100644 --- a/pkg/tcpip/link/rawfile/BUILD +++ b/pkg/tcpip/link/rawfile/BUILD @@ -16,5 +16,8 @@ go_library( visibility = [ "//visibility:public", ], - deps = ["//pkg/tcpip"], + deps = [ + "//pkg/tcpip", + "@org_golang_x_sys//unix:go_default_library", + ], ) diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go index 7e286a3a6..44e25d475 100644 --- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go +++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go @@ -22,6 +22,7 @@ import ( "syscall" "unsafe" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -101,6 +102,16 @@ func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error { return nil } +// NonBlockingSendMMsg sends multiple messages on a socket. +func NonBlockingSendMMsg(fd int, msgHdrs []MMsgHdr) (int, *tcpip.Error) { + n, _, e := syscall.RawSyscall6(unix.SYS_SENDMMSG, uintptr(fd), uintptr(unsafe.Pointer(&msgHdrs[0])), uintptr(len(msgHdrs)), syscall.MSG_DONTWAIT, 0, 0) + if e != 0 { + return 0, TranslateErrno(e) + } + + return int(n), nil +} + // PollEvent represents the pollfd structure passed to a poll() system call. type PollEvent struct { FD int32 diff --git a/pkg/tcpip/link/sharedmem/sharedmem.go b/pkg/tcpip/link/sharedmem/sharedmem.go index 9e71d4edf..279e2b457 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem.go +++ b/pkg/tcpip/link/sharedmem/sharedmem.go @@ -212,6 +212,26 @@ func (e *endpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.Prependa return nil } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + v := packet.ToView() + // Transmit the packet. + e.mu.Lock() + ok := e.tx.transmit(v, buffer.View{}) + e.mu.Unlock() + + if !ok { + return tcpip.ErrWouldBlock + } + + return nil +} + // dispatchLoop reads packets from the rx queue in a loop and dispatches them // to the network stack. func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { @@ -254,7 +274,7 @@ func (e *endpoint) dispatchLoop(d stack.NetworkDispatcher) { // Send packet up the stack. eth := header.Ethernet(b) - d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView()) + d.DeliverNetworkPacket(e, eth.SourceAddress(), eth.DestinationAddress(), eth.Type(), buffer.View(b[header.EthernetMinimumSize:]).ToVectorisedView(), buffer.View(eth)) } // Clean state. diff --git a/pkg/tcpip/link/sharedmem/sharedmem_test.go b/pkg/tcpip/link/sharedmem/sharedmem_test.go index 0e9ba0846..f3e9705c9 100644 --- a/pkg/tcpip/link/sharedmem/sharedmem_test.go +++ b/pkg/tcpip/link/sharedmem/sharedmem_test.go @@ -78,9 +78,10 @@ func (q *queueBuffers) cleanup() { } type packetInfo struct { - addr tcpip.LinkAddress - proto tcpip.NetworkProtocolNumber - vv buffer.VectorisedView + addr tcpip.LinkAddress + proto tcpip.NetworkProtocolNumber + vv buffer.VectorisedView + linkHeader buffer.View } type testContext struct { @@ -130,12 +131,13 @@ func newTestContext(t *testing.T, mtu, bufferSize uint32, addr tcpip.LinkAddress return c } -func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (c *testContext) DeliverNetworkPacket(_ stack.LinkEndpoint, remoteLinkAddr, localLinkAddr tcpip.LinkAddress, proto tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) { c.mu.Lock() c.packets = append(c.packets, packetInfo{ - addr: remoteLinkAddr, - proto: proto, - vv: vv.Clone(nil), + addr: remoteLinkAddr, + proto: proto, + vv: vv.Clone(nil), + linkHeader: linkHeader, }) c.mu.Unlock() diff --git a/pkg/tcpip/link/sniffer/sniffer.go b/pkg/tcpip/link/sniffer/sniffer.go index e401dce44..39757ea2a 100644 --- a/pkg/tcpip/link/sniffer/sniffer.go +++ b/pkg/tcpip/link/sniffer/sniffer.go @@ -116,7 +116,7 @@ func NewWithFile(lower stack.LinkEndpoint, file *os.File, snapLen uint32) (stack // DeliverNetworkPacket implements the stack.NetworkDispatcher interface. It is // called by the link-layer endpoint being wrapped when a packet arrives, and // logs the packet before forwarding to the actual dispatcher. -func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { logPacket("recv", protocol, vv.First(), nil) } @@ -147,7 +147,7 @@ func (e *endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local panic(err) } } - e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv) + e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv, linkHeader) } // Attach implements the stack.LinkEndpoint interface. It saves the dispatcher @@ -193,10 +193,7 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -// WritePacket implements the stack.LinkEndpoint interface. It is called by -// higher-level protocols to write packets; it just logs the packet and forwards -// the request to the lower endpoint. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { +func (e *endpoint) dumpPacket(gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) { if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { logPacket("send", protocol, hdr.View(), gso) } @@ -218,28 +215,74 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen panic(err) } length -= len(hdrBuf) - if length > 0 { - for _, v := range payload.Views() { - if len(v) > length { - v = v[:length] - } - n, err := buf.Write(v) - if err != nil { - panic(err) - } - length -= n - if length == 0 { - break - } - } - } + logVectorisedView(payload, length, buf) if _, err := e.file.Write(buf.Bytes()); err != nil { panic(err) } } +} + +// WritePacket implements the stack.LinkEndpoint interface. It is called by +// higher-level protocols to write packets; it just logs the packet and +// forwards the request to the lower endpoint. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error { + e.dumpPacket(gso, hdr, payload, protocol) return e.lower.WritePacket(r, gso, hdr, payload, protocol) } +// WritePackets implements the stack.LinkEndpoint interface. It is called by +// higher-level protocols to write packets; it just logs the packet and +// forwards the request to the lower endpoint. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + view := payload.ToView() + for _, d := range hdrs { + e.dumpPacket(gso, d.Hdr, buffer.NewVectorisedView(d.Size, []buffer.View{view[d.Off:][:d.Size]}), protocol) + } + return e.lower.WritePackets(r, gso, hdrs, payload, protocol) +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + if atomic.LoadUint32(&LogPackets) == 1 && e.file == nil { + logPacket("send", 0, buffer.View("[raw packet, no header available]"), nil /* gso */) + } + if e.file != nil && atomic.LoadUint32(&LogPacketsToFile) == 1 { + length := packet.Size() + if length > int(e.maxPCAPLen) { + length = int(e.maxPCAPLen) + } + + buf := bytes.NewBuffer(make([]byte, 0, pcapPacketHeaderLen+length)) + if err := binary.Write(buf, binary.BigEndian, newPCAPPacketHeader(uint32(length), uint32(packet.Size()))); err != nil { + panic(err) + } + logVectorisedView(packet, length, buf) + if _, err := e.file.Write(buf.Bytes()); err != nil { + panic(err) + } + } + return e.lower.WriteRawPacket(packet) +} + +func logVectorisedView(vv buffer.VectorisedView, length int, buf *bytes.Buffer) { + if length <= 0 { + return + } + for _, v := range vv.Views() { + if len(v) > length { + v = v[:length] + } + n, err := buf.Write(v) + if err != nil { + panic(err) + } + length -= n + if length == 0 { + return + } + } +} + // Wait implements stack.LinkEndpoint.Wait. func (*endpoint) Wait() {} diff --git a/pkg/tcpip/link/waitable/waitable.go b/pkg/tcpip/link/waitable/waitable.go index 5a1791cb5..a04fc1062 100644 --- a/pkg/tcpip/link/waitable/waitable.go +++ b/pkg/tcpip/link/waitable/waitable.go @@ -50,12 +50,12 @@ func New(lower stack.LinkEndpoint) *Endpoint { // It is called by the link-layer endpoint being wrapped when a packet arrives, // and only forwards to the actual dispatcher if Wait or WaitDispatch haven't // been called. -func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *Endpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) { if !e.dispatchGate.Enter() { return } - e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv) + e.dispatcher.DeliverNetworkPacket(e, remote, local, protocol, vv, linkHeader) e.dispatchGate.Leave() } @@ -109,6 +109,30 @@ func (e *Endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return err } +// WritePackets implements stack.LinkEndpoint.WritePackets. It is called by +// higher-level protocols to write packets. It only forwards packets to the +// lower endpoint if Wait or WaitWrite haven't been called. +func (e *Endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + if !e.writeGate.Enter() { + return len(hdrs), nil + } + + n, err := e.lower.WritePackets(r, gso, hdrs, payload, protocol) + e.writeGate.Leave() + return n, err +} + +// WriteRawPacket implements stack.LinkEndpoint.WriteRawPacket. +func (e *Endpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + if !e.writeGate.Enter() { + return nil + } + + err := e.lower.WriteRawPacket(packet) + e.writeGate.Leave() + return err +} + // WaitWrite prevents new calls to WritePacket from reaching the lower endpoint, // and waits for inflight ones to finish before returning. func (e *Endpoint) WaitWrite() { diff --git a/pkg/tcpip/link/waitable/waitable_test.go b/pkg/tcpip/link/waitable/waitable_test.go index ae23c96b7..5f0f8fa2d 100644 --- a/pkg/tcpip/link/waitable/waitable_test.go +++ b/pkg/tcpip/link/waitable/waitable_test.go @@ -35,7 +35,7 @@ type countedEndpoint struct { dispatcher stack.NetworkDispatcher } -func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (e *countedEndpoint) DeliverNetworkPacket(linkEP stack.LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) { e.dispatchCount++ } @@ -70,6 +70,17 @@ func (e *countedEndpoint) WritePacket(r *stack.Route, _ *stack.GSO, hdr buffer.P return nil } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *countedEndpoint) WritePackets(r *stack.Route, _ *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + e.writeCount += len(hdrs) + return len(hdrs), nil +} + +func (e *countedEndpoint) WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error { + e.writeCount++ + return nil +} + // Wait implements stack.LinkEndpoint.Wait. func (*countedEndpoint) Wait() {} @@ -109,21 +120,21 @@ func TestWaitDispatch(t *testing.T) { } // Dispatch and check that it goes through. - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}, buffer.View{}) if want := 1; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on writes, then try to dispatch. It must go through. wep.WaitWrite() - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}, buffer.View{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } // Wait on dispatches, then try to dispatch. It must not go through. wep.WaitDispatch() - ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}) + ep.dispatcher.DeliverNetworkPacket(ep, "", "", 0, buffer.VectorisedView{}, buffer.View{}) if want := 2; ep.dispatchCount != want { t.Fatalf("Unexpected dispatchCount: got=%v, want=%v", ep.dispatchCount, want) } diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index 6b1e854dc..46178459e 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -83,6 +83,11 @@ func (e *endpoint) WritePacket(*stack.Route, *stack.GSO, buffer.Prependable, buf return tcpip.ErrNotSupported } +// WritePackets implements stack.NetworkEndpoint.WritePackets. +func (e *endpoint) WritePackets(*stack.Route, *stack.GSO, []stack.PacketDescriptor, buffer.VectorisedView, stack.NetworkHeaderParams, stack.PacketLooping) (int, *tcpip.Error) { + return 0, tcpip.ErrNotSupported +} + func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { return tcpip.ErrNotSupported } diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index f644a8b08..8d74497ba 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -171,6 +171,15 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prepen return nil } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (t *testObject) WritePackets(_ *stack.Route, _ *stack.GSO, hdr []stack.PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) { + panic("not implemented") +} + +func (t *testObject) WriteRawPacket(_ buffer.VectorisedView) *tcpip.Error { + return tcpip.ErrNotSupported +} + func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { s := stack.New(stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 5cd895ff0..1339f8474 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -198,10 +198,9 @@ func (e *endpoint) writePacketFragments(r *stack.Route, gso *stack.GSO, hdr buff return nil } -// WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { +func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) { ip := header.IPv4(hdr.Prepend(header.IPv4MinimumSize)) - length := uint16(hdr.UsedLength() + payload.Size()) + length := uint16(hdr.UsedLength() + payloadSize) id := uint32(0) if length > header.IPv4MaximumHeaderSize+8 { // Packets of 68 bytes or less are required by RFC 791 to not be @@ -219,6 +218,11 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen DstAddr: r.RemoteAddress, }) ip.SetChecksum(^ip.CalculateChecksum()) +} + +// WritePacket writes a packet to the given destination address and protocol. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { + e.addIPHeader(r, &hdr, payload.Size(), params) if loop&stack.PacketLoop != 0 { views := make([]buffer.View, 1, 1+len(payload.Views())) @@ -242,6 +246,23 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return nil } +// WritePackets implements stack.NetworkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + if loop&stack.PacketLoop != 0 { + panic("multiple packets in local loop") + } + if loop&stack.PacketOut == 0 { + return len(hdrs), nil + } + + for i := range hdrs { + e.addIPHeader(r, &hdrs[i].Hdr, hdrs[i].Size, params) + } + n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err +} + // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index cd1e34085..5898f8f9e 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -97,9 +97,8 @@ func (e *endpoint) GSOMaxSize() uint32 { return 0 } -// WritePacket writes a packet to the given destination address and protocol. -func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { - length := uint16(hdr.UsedLength() + payload.Size()) +func (e *endpoint) addIPHeader(r *stack.Route, hdr *buffer.Prependable, payloadSize int, params stack.NetworkHeaderParams) { + length := uint16(hdr.UsedLength() + payloadSize) ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) ip.Encode(&header.IPv6Fields{ PayloadLength: length, @@ -109,6 +108,11 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) +} + +// WritePacket writes a packet to the given destination address and protocol. +func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) *tcpip.Error { + e.addIPHeader(r, &hdr, payload.Size(), params) if loop&stack.PacketLoop != 0 { views := make([]buffer.View, 1, 1+len(payload.Views())) @@ -127,6 +131,26 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen return e.linkEP.WritePacket(r, gso, hdr, payload, ProtocolNumber) } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + if loop&stack.PacketLoop != 0 { + panic("not implemented") + } + if loop&stack.PacketOut == 0 { + return len(hdrs), nil + } + + for i := range hdrs { + hdr := &hdrs[i].Hdr + size := hdrs[i].Size + e.addIPHeader(r, hdr, size, params) + } + + n, err := e.linkEP.WritePackets(r, gso, hdrs, payload, ProtocolNumber) + r.Stats().IP.PacketsSent.IncrementBy(uint64(n)) + return n, err +} + // WriteHeaderIncludedPacker implements stack.NetworkEndpoint. It is not yet // supported by IPv6. func (*endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index bfc03e90b..460db3cf8 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -73,6 +73,7 @@ go_test( "//pkg/tcpip/transport/icmp", "//pkg/tcpip/transport/udp", "//pkg/waiter", + "@com_github_google_go-cmp//cmp:go_default_library", ], ) diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index 849c2f35f..8995fbfc3 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -31,8 +31,7 @@ import ( const ( addr1 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" addr2 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" - linkAddr1 = "\x01\x02\x03\x04\x05\x06" - linkAddr2 = "\x01\x02\x03\x04\x05\x07" + linkAddr1 = "\x02\x02\x03\x04\x05\x06" ) // TestDADDisabled tests that an address successfully resolves immediately diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 38d066cd1..e456e05f4 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -40,6 +40,9 @@ type NIC struct { endpoints map[NetworkEndpointID]*referencedNetworkEndpoint addressRanges []tcpip.Subnet mcastJoins map[NetworkEndpointID]int32 + // packetEPs is protected by mu, but the contained PacketEndpoint + // values are not. + packetEPs map[tcpip.NetworkProtocolNumber][]PacketEndpoint stats NICStats @@ -78,7 +81,10 @@ const ( ) func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback bool) *NIC { - return &NIC{ + // TODO(b/141011931): Validate a LinkEndpoint (ep) is valid. For + // example, make sure that the link address it provides is a valid + // unicast ethernet address. + nic := &NIC{ stack: stack, id: id, name: name, @@ -87,6 +93,7 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback primary: make(map[tcpip.NetworkProtocolNumber][]*referencedNetworkEndpoint), endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint), mcastJoins: make(map[NetworkEndpointID]int32), + packetEPs: make(map[tcpip.NetworkProtocolNumber][]PacketEndpoint), stats: NICStats{ Tx: DirectionStats{ Packets: &tcpip.StatCounter{}, @@ -101,6 +108,16 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback dad: make(map[tcpip.Address]dadState), }, } + + // Register supported packet endpoint protocols. + for _, netProto := range header.Ethertypes { + nic.packetEPs[netProto] = []PacketEndpoint{} + } + for _, netProto := range stack.networkProtocols { + nic.packetEPs[netProto.Number()] = []PacketEndpoint{} + } + + return nic } // enable enables the NIC. enable will attach the link to its LinkEndpoint and @@ -125,11 +142,50 @@ func (n *NIC) enable() *tcpip.Error { // 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. - if _, ok := n.stack.networkProtocols[header.IPv6ProtocolNumber]; ok { - return n.joinGroup(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress) + // + // 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 { + return nil } - return nil + n.mu.Lock() + defer n.mu.Unlock() + + if err := n.joinGroupLocked(header.IPv6ProtocolNumber, header.IPv6AllNodesMulticastAddress); err != nil { + return err + } + + if !n.stack.autoGenIPv6LinkLocal { + return nil + } + + l2addr := n.linkEP.LinkAddress() + + // Only attempt to generate the link-local address if we have a + // valid MAC address. + // + // TODO(b/141011931): Validate a LinkEndpoint's link address + // (provided by LinkEndpoint.LinkAddress) before reaching this + // point. + if !header.IsValidUnicastEthernetAddress(l2addr) { + return nil + } + + addr := header.LinkLocalAddr(l2addr) + + _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: addr, + PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen, + }, + }, CanBePrimaryEndpoint) + + return err } // attachLinkEndpoint attaches the NIC to the endpoint, which will enable it @@ -279,9 +335,31 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p // The NIC already have a permanent endpoint with that address. return nil, tcpip.ErrDuplicateAddress case permanentExpired, temporary: - // Promote the endpoint to become permanent. + // Promote the endpoint to become permanent and respect + // the new peb. if ref.tryIncRef() { ref.setKind(permanent) + + refs := n.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.primary[r.protocol] = append(refs[:i], refs[i+1:]...) + case NeverPrimaryEndpoint: + n.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 @@ -350,12 +428,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar n.endpoints[id] = ref - switch peb { - case CanBePrimaryEndpoint: - n.primary[protocolAddress.Protocol] = append(n.primary[protocolAddress.Protocol], ref) - case FirstPrimaryEndpoint: - n.primary[protocolAddress.Protocol] = append([]*referencedNetworkEndpoint{ref}, n.primary[protocolAddress.Protocol]...) - } + n.insertPrimaryEndpointLocked(ref, peb) // If we are adding a tentative IPv6 address, start DAD. if isIPv6Unicast && kind == permanentTentative { @@ -477,6 +550,19 @@ func (n *NIC) AddressRanges() []tcpip.Subnet { return append(sns, n.addressRanges...) } +// 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.primary[r.protocol] = append(n.primary[r.protocol], r) + case FirstPrimaryEndpoint: + n.primary[r.protocol] = append([]*referencedNetworkEndpoint{r}, n.primary[r.protocol]...) + } +} + func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { id := *r.ep.ID() @@ -494,9 +580,10 @@ func (n *NIC) removeEndpointLocked(r *referencedNetworkEndpoint) { } delete(n.endpoints, id) - for i, ref := range n.primary[r.protocol] { + refs := n.primary[r.protocol] + for i, ref := range refs { if ref == r { - n.primary[r.protocol] = append(n.primary[r.protocol][:i], n.primary[r.protocol][i+1:]...) + n.primary[r.protocol] = append(refs[:i], refs[i+1:]...) break } } @@ -568,6 +655,11 @@ func (n *NIC) joinGroup(protocol tcpip.NetworkProtocolNumber, addr tcpip.Address // 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.mcastJoins[id] if joins == 0 { @@ -631,7 +723,7 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // Note that the ownership of the slice backing vv is retained by the caller. // This rule applies only to the slice itself, not to the items of the slice; // the ownership of the items is not retained by the caller. -func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) { +func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) { n.stats.Rx.Packets.Increment() n.stats.Rx.Bytes.IncrementBy(uint64(vv.Size())) @@ -641,6 +733,26 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr return } + // If no local link layer address is provided, assume it was sent + // directly to this NIC. + if local == "" { + local = n.linkEP.LinkAddress() + } + + // Are any packet sockets listening for this network protocol? + n.mu.RLock() + packetEPs := n.packetEPs[protocol] + // Check whether there are packet sockets listening for every protocol. + // If we received a packet with protocol EthernetProtocolAll, then the + // previous for loop will have handled it. + if protocol != header.EthernetProtocolAll { + packetEPs = append(packetEPs, n.packetEPs[header.EthernetProtocolAll]...) + } + n.mu.RUnlock() + for _, ep := range packetEPs { + ep.HandlePacket(n.id, local, protocol, vv, linkHeader) + } + if netProto.Number() == header.IPv4ProtocolNumber || netProto.Number() == header.IPv6ProtocolNumber { n.stack.stats.IP.PacketsReceived.Increment() } @@ -700,7 +812,10 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr return } - n.stack.stats.IP.InvalidAddressesReceived.Increment() + // If a packet socket handled the packet, don't treat it as invalid. + if len(packetEPs) == 0 { + n.stack.stats.IP.InvalidAddressesReceived.Increment() + } } // DeliverTransportPacket delivers the packets to the appropriate transport @@ -856,6 +971,36 @@ const ( temporary ) +func (n *NIC) registerPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error { + n.mu.Lock() + defer n.mu.Unlock() + + eps, ok := n.packetEPs[netProto] + if !ok { + return tcpip.ErrNotSupported + } + n.packetEPs[netProto] = append(eps, ep) + + return nil +} + +func (n *NIC) unregisterPacketEndpoint(netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) { + n.mu.Lock() + defer n.mu.Unlock() + + eps, ok := n.packetEPs[netProto] + if !ok { + return + } + + for i, epOther := range eps { + if epOther == ep { + n.packetEPs[netProto] = append(eps[:i], eps[i+1:]...) + return + } + } +} + type referencedNetworkEndpoint struct { ep NetworkEndpoint nic *NIC diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 9d6157f22..0869fb084 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -71,8 +71,8 @@ type TransportEndpoint interface { // RawTransportEndpoint is the interface that needs to be implemented by raw // transport protocol endpoints. RawTransportEndpoints receive the entire -// packet - including the link, network, and transport headers - as delivered -// to netstack. +// packet - including the network and transport headers - as delivered to +// netstack. type RawTransportEndpoint interface { // HandlePacket is called by the stack when new packets arrive to // this transport endpoint. The packet contains all data from the link @@ -80,6 +80,22 @@ type RawTransportEndpoint interface { HandlePacket(r *Route, netHeader buffer.View, packet buffer.VectorisedView) } +// PacketEndpoint is the interface that needs to be implemented by packet +// transport protocol endpoints. These endpoints receive link layer headers in +// addition to whatever they contain (usually network and transport layer +// headers and a payload). +type PacketEndpoint interface { + // HandlePacket is called by the stack when new packets arrive that + // match the endpoint. + // + // Implementers should treat packet as immutable and should copy it + // before before modification. + // + // linkHeader may have a length of 0, in which case the PacketEndpoint + // should construct its own ethernet header for applications. + HandlePacket(nicid tcpip.NICID, addr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, packet buffer.VectorisedView, linkHeader buffer.View) +} + // 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 { @@ -185,6 +201,10 @@ type NetworkEndpoint interface { // protocol. WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) *tcpip.Error + // WritePackets writes packets to the given destination address and + // protocol. + WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams, loop PacketLooping) (int, *tcpip.Error) + // WriteHeaderIncludedPacket writes a packet that includes a network // header to the given destination address. WriteHeaderIncludedPacket(r *Route, payload buffer.VectorisedView, loop PacketLooping) *tcpip.Error @@ -242,9 +262,10 @@ type NetworkProtocol interface { // packets to the appropriate network endpoint after it has been handled by // the data link layer. type NetworkDispatcher interface { - // DeliverNetworkPacket finds the appropriate network protocol - // endpoint and hands the packet over for further processing. - DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) + // DeliverNetworkPacket finds the appropriate network protocol endpoint + // and hands the packet over for further processing. linkHeader may have + // length 0 when the caller does not have ethernet data. + DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.LinkAddress, protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, linkHeader buffer.View) } // LinkEndpointCapabilities is the type associated with the capabilities @@ -266,7 +287,11 @@ const ( CapabilitySaveRestore CapabilityDisconnectOk CapabilityLoopback - CapabilityGSO + CapabilityHardwareGSO + + // CapabilitySoftwareGSO indicates the link endpoint supports of sending + // multiple packets using a single call (LinkEndpoint.WritePackets). + CapabilitySoftwareGSO ) // LinkEndpoint is the interface implemented by data link layer protocols (e.g., @@ -301,6 +326,18 @@ type LinkEndpoint interface { // r.LocalLinkAddress if it is provided. WritePacket(r *Route, gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error + // WritePackets writes packets with the given protocol through the + // given route. + // + // Right now, WritePackets is used only when the software segmentation + // offload is enabled. If it will be used for something else, it may + // require to change syscall filters. + WritePackets(r *Route, gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) (int, *tcpip.Error) + + // WriteRawPacket writes a packet directly to the link. The packet + // should already have an ethernet header. + WriteRawPacket(packet buffer.VectorisedView) *tcpip.Error + // Attach attaches the data link layer endpoint to the network-layer // dispatcher of the stack. Attach(dispatcher NetworkDispatcher) @@ -324,13 +361,14 @@ type LinkEndpoint interface { type InjectableLinkEndpoint interface { LinkEndpoint - // Inject injects an inbound packet. - Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) + // InjectInbound injects an inbound packet. + InjectInbound(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) - // WriteRawPacket writes a fully formed outbound packet directly to the link. + // InjectOutbound writes a fully formed outbound packet directly to the + // link. // // dest is used by endpoints with multiple raw destinations. - WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error + InjectOutbound(dest tcpip.Address, packet []byte) *tcpip.Error } // A LinkAddressResolver is an extension to a NetworkProtocol that @@ -379,11 +417,16 @@ type LinkAddressCache interface { RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) } -// UnassociatedEndpointFactory produces endpoints for writing packets not -// associated with a particular transport protocol. Such endpoints can be used -// to write arbitrary packets that include the IP header. -type UnassociatedEndpointFactory interface { - NewUnassociatedRawEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) +// RawFactory produces endpoints for writing various types of raw packets. +type RawFactory interface { + // NewUnassociatedEndpoint produces endpoints for writing packets not + // associated with a particular transport protocol. Such endpoints can + // be used to write arbitrary packets that include the network header. + NewUnassociatedEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) + + // NewPacketEndpoint produces endpoints for reading and writing packets + // that include network and (when cooked is false) link layer headers. + NewPacketEndpoint(stack *Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) } // GSOType is the type of GSO segments. @@ -394,8 +437,14 @@ type GSOType int // Types of gso segments. const ( GSONone GSOType = iota + + // Hardware GSO types: GSOTCPv4 GSOTCPv6 + + // GSOSW is used for software GSO segments which have to be sent by + // endpoint.WritePackets. + GSOSW ) // GSO contains generic segmentation offload properties. @@ -423,3 +472,7 @@ type GSOEndpoint interface { // GSOMaxSize returns the maximum GSO packet size. GSOMaxSize() uint32 } + +// SoftwareGSOMaxSize is a maximum allowed size of a software GSO segment. +// This isn't a hard limit, because it is never set into packet headers. +const SoftwareGSOMaxSize = (1 << 16) diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index e72373964..1a0a51b57 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -47,8 +47,8 @@ type Route struct { // starts. ref *referencedNetworkEndpoint - // loop controls where WritePacket should send packets. - loop PacketLooping + // Loop controls where WritePacket should send packets. + Loop PacketLooping } // makeRoute initializes a new route. It takes ownership of the provided @@ -69,7 +69,7 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip LocalLinkAddress: localLinkAddr, RemoteAddress: remoteAddr, ref: ref, - loop: loop, + Loop: loop, } } @@ -159,7 +159,7 @@ func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.Vec return tcpip.ErrInvalidEndpointState } - err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.loop) + err := r.ref.ep.WritePacket(r, gso, hdr, payload, params, r.Loop) if err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() } else { @@ -169,6 +169,44 @@ func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.Vec return err } +// PacketDescriptor is a packet descriptor which contains a packet header and +// offset and size of packet data in a payload view. +type PacketDescriptor struct { + Hdr buffer.Prependable + Off int + Size int +} + +// NewPacketDescriptors allocates a set of packet descriptors. +func NewPacketDescriptors(n int, hdrSize int) []PacketDescriptor { + buf := make([]byte, n*hdrSize) + hdrs := make([]PacketDescriptor, n) + for i := range hdrs { + hdrs[i].Hdr = buffer.NewEmptyPrependableFromView(buf[i*hdrSize:][:hdrSize]) + } + return hdrs +} + +// WritePackets writes the set of packets through the given route. +func (r *Route) WritePackets(gso *GSO, hdrs []PacketDescriptor, payload buffer.VectorisedView, params NetworkHeaderParams) (int, *tcpip.Error) { + if !r.ref.isValidForOutgoing() { + return 0, tcpip.ErrInvalidEndpointState + } + + n, err := r.ref.ep.WritePackets(r, gso, hdrs, payload, params, r.Loop) + if err != nil { + r.Stats().IP.OutgoingPacketErrors.IncrementBy(uint64(len(hdrs) - n)) + } + r.ref.nic.stats.Tx.Packets.IncrementBy(uint64(n)) + payloadSize := 0 + for i := 0; i < n; i++ { + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(hdrs[i].Hdr.UsedLength())) + payloadSize += hdrs[i].Size + } + r.ref.nic.stats.Tx.Bytes.IncrementBy(uint64(payloadSize)) + return n, err +} + // WriteHeaderIncludedPacket writes a packet already containing a network // header through the given route. func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip.Error { @@ -176,7 +214,7 @@ func (r *Route) WriteHeaderIncludedPacket(payload buffer.VectorisedView) *tcpip. return tcpip.ErrInvalidEndpointState } - if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.loop); err != nil { + if err := r.ref.ep.WriteHeaderIncludedPacket(r, payload, r.Loop); err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() return err } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index a199bc1cc..284280917 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -351,10 +351,9 @@ type Stack struct { networkProtocols map[tcpip.NetworkProtocolNumber]NetworkProtocol linkAddrResolvers map[tcpip.NetworkProtocolNumber]LinkAddressResolver - // unassociatedFactory creates unassociated endpoints. If nil, raw - // endpoints are disabled. It is set during Stack creation and is - // immutable. - unassociatedFactory UnassociatedEndpointFactory + // rawFactory creates raw endpoints. If nil, raw endpoints are + // disabled. It is set during Stack creation and is immutable. + rawFactory RawFactory demux *transportDemuxer @@ -402,6 +401,11 @@ type Stack struct { // ndpConfigs is the NDP configurations used by interfaces. ndpConfigs NDPConfigurations + + // autoGenIPv6LinkLocal determines whether or not the stack will attempt + // to auto-generate an IPv6 link-local address for newly enabled NICs. + // See the AutoGenIPv6LinkLocal field of Options for more details. + autoGenIPv6LinkLocal bool } // Options contains optional Stack configuration. @@ -425,16 +429,28 @@ type Options struct { // stack (false). HandleLocal bool - // UnassociatedFactory produces unassociated endpoints raw endpoints. - // Raw endpoints are enabled only if this is non-nil. - UnassociatedFactory UnassociatedEndpointFactory - // NDPConfigs is the 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 + + // AutoGenIPv6LinkLocal determins whether or not the stack will attempt + // to auto-generate an IPv6 link-local address for newly enabled 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 + + // RawFactory produces raw endpoints. Raw endpoints are enabled only if + // this is non-nil. + RawFactory RawFactory } // TransportEndpointInfo holds useful information about a transport endpoint @@ -485,18 +501,19 @@ func New(opts Options) *Stack { opts.NDPConfigs.validate() 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), - linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), - PortManager: ports.NewPortManager(), - clock: clock, - stats: opts.Stats.FillIn(), - handleLocal: opts.HandleLocal, - icmpRateLimiter: NewICMPRateLimiter(), - portSeed: generateRandUint32(), - ndpConfigs: opts.NDPConfigs, + transportProtocols: make(map[tcpip.TransportProtocolNumber]*transportProtocolState), + networkProtocols: make(map[tcpip.NetworkProtocolNumber]NetworkProtocol), + linkAddrResolvers: make(map[tcpip.NetworkProtocolNumber]LinkAddressResolver), + nics: make(map[tcpip.NICID]*NIC), + linkAddrCache: newLinkAddrCache(ageLimit, resolutionTimeout, resolutionAttempts), + PortManager: ports.NewPortManager(), + clock: clock, + stats: opts.Stats.FillIn(), + handleLocal: opts.HandleLocal, + icmpRateLimiter: NewICMPRateLimiter(), + portSeed: generateRandUint32(), + ndpConfigs: opts.NDPConfigs, + autoGenIPv6LinkLocal: opts.AutoGenIPv6LinkLocal, } // Add specified network protocols. @@ -514,8 +531,8 @@ func New(opts Options) *Stack { } } - // Add the factory for unassociated endpoints, if present. - s.unassociatedFactory = opts.UnassociatedFactory + // Add the factory for raw endpoints, if present. + s.rawFactory = opts.RawFactory // Create the global transport demuxer. s.demux = newTransportDemuxer(s) @@ -650,12 +667,12 @@ func (s *Stack) NewEndpoint(transport tcpip.TransportProtocolNumber, network tcp // protocol. Raw endpoints receive all traffic for a given protocol regardless // of address. func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue, associated bool) (tcpip.Endpoint, *tcpip.Error) { - if s.unassociatedFactory == nil { + if s.rawFactory == nil { return nil, tcpip.ErrNotPermitted } if !associated { - return s.unassociatedFactory.NewUnassociatedRawEndpoint(s, network, transport, waiterQueue) + return s.rawFactory.NewUnassociatedEndpoint(s, network, transport, waiterQueue) } t, ok := s.transportProtocols[transport] @@ -666,6 +683,16 @@ func (s *Stack) NewRawEndpoint(transport tcpip.TransportProtocolNumber, network return t.proto.NewRawEndpoint(s, network, waiterQueue) } +// NewPacketEndpoint creates a new packet endpoint listening for the given +// netProto. +func (s *Stack) NewPacketEndpoint(cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + if s.rawFactory == nil { + return nil, tcpip.ErrNotPermitted + } + + return s.rawFactory.NewPacketEndpoint(s, cooked, netProto, waiterQueue) +} + // createNIC creates a NIC with the provided id and link-layer endpoint, and // optionally enable it. func (s *Stack) createNIC(id tcpip.NICID, name string, ep LinkEndpoint, enabled, loopback bool) *tcpip.Error { @@ -1135,6 +1162,109 @@ func (s *Stack) Resume() { } } +// RegisterPacketEndpoint registers ep with the stack, causing it to receive +// all traffic of the specified netProto on the given NIC. If nicID is 0, it +// receives traffic from every NIC. +func (s *Stack) RegisterPacketEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) *tcpip.Error { + s.mu.Lock() + defer s.mu.Unlock() + + // If no NIC is specified, capture on all devices. + if nicID == 0 { + // Register with each NIC. + for _, nic := range s.nics { + if err := nic.registerPacketEndpoint(netProto, ep); err != nil { + s.unregisterPacketEndpointLocked(0, netProto, ep) + return err + } + } + return nil + } + + // Capture on a specific device. + nic, ok := s.nics[nicID] + if !ok { + return tcpip.ErrUnknownNICID + } + if err := nic.registerPacketEndpoint(netProto, ep); err != nil { + return err + } + + return nil +} + +// UnregisterPacketEndpoint unregisters ep for packets of the specified +// netProto from the specified NIC. If nicID is 0, ep is unregistered from all +// NICs. +func (s *Stack) UnregisterPacketEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) { + s.mu.Lock() + defer s.mu.Unlock() + s.unregisterPacketEndpointLocked(nicID, netProto, ep) +} + +func (s *Stack) unregisterPacketEndpointLocked(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, ep PacketEndpoint) { + // If no NIC is specified, unregister on all devices. + if nicID == 0 { + // Unregister with each NIC. + for _, nic := range s.nics { + nic.unregisterPacketEndpoint(netProto, ep) + } + return + } + + // Unregister in a single device. + nic, ok := s.nics[nicID] + if !ok { + return + } + nic.unregisterPacketEndpoint(netProto, ep) +} + +// WritePacket writes data directly to the specified NIC. It adds an ethernet +// header based on the arguments. +func (s *Stack) WritePacket(nicid tcpip.NICID, dst tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, payload buffer.VectorisedView) *tcpip.Error { + s.mu.Lock() + nic, ok := s.nics[nicid] + s.mu.Unlock() + if !ok { + return tcpip.ErrUnknownDevice + } + + // Add our own fake ethernet header. + ethFields := header.EthernetFields{ + SrcAddr: nic.linkEP.LinkAddress(), + DstAddr: dst, + Type: netProto, + } + fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) + fakeHeader.Encode(ðFields) + ethHeader := buffer.View(fakeHeader).ToVectorisedView() + ethHeader.Append(payload) + + if err := nic.linkEP.WriteRawPacket(ethHeader); err != nil { + return err + } + + return nil +} + +// WriteRawPacket writes data directly to the specified NIC without adding any +// headers. +func (s *Stack) WriteRawPacket(nicid tcpip.NICID, payload buffer.VectorisedView) *tcpip.Error { + s.mu.Lock() + nic, ok := s.nics[nicid] + s.mu.Unlock() + if !ok { + return tcpip.ErrUnknownDevice + } + + if err := nic.linkEP.WriteRawPacket(payload); err != nil { + return err + } + + return nil +} + // NetworkProtocolInstance returns the protocol instance in the stack for the // specified network protocol. This method is public for protocol implementers // and tests to use. diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 10fd1065f..9a8906a0d 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -24,11 +24,14 @@ import ( "sort" "strings" "testing" + "time" + "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/channel" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -144,6 +147,11 @@ func (f *fakeNetworkEndpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr bu return f.ep.WritePacket(r, gso, hdr, payload, fakeNetNumber) } +// WritePackets implements stack.LinkEndpoint.WritePackets. +func (f *fakeNetworkEndpoint) WritePackets(r *stack.Route, gso *stack.GSO, hdrs []stack.PacketDescriptor, payload buffer.VectorisedView, params stack.NetworkHeaderParams, loop stack.PacketLooping) (int, *tcpip.Error) { + panic("not implemented") +} + func (*fakeNetworkEndpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.VectorisedView, loop stack.PacketLooping) *tcpip.Error { return tcpip.ErrNotSupported } @@ -1864,3 +1872,281 @@ func TestNICForwarding(t *testing.T) { t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want) } } + +// TestNICAutoGenAddr tests the auto-generation of IPv6 link-local addresses +// (or lack there-of if disabled (default)). Note, DAD will be disabled in +// these tests. +func TestNICAutoGenAddr(t *testing.T) { + tests := []struct { + name string + autoGen bool + linkAddr tcpip.LinkAddress + shouldGen bool + }{ + { + "Disabled", + false, + linkAddr1, + false, + }, + { + "Enabled", + true, + linkAddr1, + true, + }, + { + "Nil MAC", + true, + tcpip.LinkAddress([]byte(nil)), + false, + }, + { + "Empty MAC", + true, + tcpip.LinkAddress(""), + false, + }, + { + "Invalid MAC", + true, + tcpip.LinkAddress("\x01\x02\x03"), + false, + }, + { + "Multicast MAC", + true, + tcpip.LinkAddress("\x01\x02\x03\x04\x05\x06"), + false, + }, + { + "Unspecified MAC", + true, + tcpip.LinkAddress("\x00\x00\x00\x00\x00\x00"), + false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + opts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + } + + if test.autoGen { + // Only set opts.AutoGenIPv6LinkLocal when + // test.autoGen is true because + // opts.AutoGenIPv6LinkLocal should be false by + // default. + opts.AutoGenIPv6LinkLocal = true + } + + e := channel.New(10, 1280, test.linkAddr) + s := stack.New(opts) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err) + } + + if test.shouldGen { + // Should have auto-generated an address and + // resolved immediately (DAD is disabled). + if want := (tcpip.AddressWithPrefix{Address: header.LinkLocalAddr(test.linkAddr), PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want) + } + } else { + // Should not have auto-generated an address. + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + } + }) + } +} + +// TestNICAutoGenAddrDoesDAD tests that the successful auto-generation of IPv6 +// link-local addresses will only be assigned after the DAD process resolves. +func TestNICAutoGenAddrDoesDAD(t *testing.T) { + opts := stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + RetransmitTimer: time.Second, + DupAddrDetectTransmits: 1, + }, + AutoGenIPv6LinkLocal: true, + } + + e := channel.New(10, 1280, linkAddr1) + s := stack.New(opts) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(_) = %s", err) + } + + // Address should not be considered bound to the + // NIC yet (DAD ongoing). + addr, err := s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (_, %v), want = (_, nil)", err) + } + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = (%s, nil), want = (%s, nil)", addr, want) + } + + // Wait for the address to resolve (an extra + // 250ms to make sure the address resolves). + // + // TODO(b/140896005): Use events from the + // netstack to know immediately when DAD + // completes. + time.Sleep(time.Second + 250*time.Millisecond) + + // Should have auto-generated an address and + // resolved (if DAD). + addr, err = s.GetMainNICAddress(1, header.IPv6ProtocolNumber) + if err != nil { + t.Fatalf("stack.GetMainNICAddress(_, _) err = %s", err) + } + if want := (tcpip.AddressWithPrefix{Address: header.LinkLocalAddr(linkAddr1), PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen}); addr != want { + t.Fatalf("got stack.GetMainNICAddress(_, _) = %s, want = %s", addr, want) + } +} + +// TestNewPEB tests that a new PrimaryEndpointBehavior value (peb) is respected +// when an address's kind gets "promoted" to permanent from permanentExpired. +func TestNewPEBOnPromotionToPermanent(t *testing.T) { + pebs := []stack.PrimaryEndpointBehavior{ + stack.NeverPrimaryEndpoint, + stack.CanBePrimaryEndpoint, + stack.FirstPrimaryEndpoint, + } + + for _, pi := range pebs { + 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()}, + }) + ep1 := channel.New(10, defaultMTU, "") + if err := s.CreateNIC(1, ep1); err != nil { + t.Fatal("CreateNIC failed:", err) + } + + // Add a permanent address with initial + // PrimaryEndpointBehavior (peb), pi. If pi is + // NeverPrimaryEndpoint, the address should not + // be returned by a call to GetMainNICAddress; + // else, it should. + if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", pi); err != nil { + t.Fatal("AddAddressWithOptions failed:", err) + } + addr, err := s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { + t.Fatal("s.GetMainNICAddress failed:", err) + } + if pi == stack.NeverPrimaryEndpoint { + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got GetMainNICAddress = %s, want = %s", addr, want) + + } + } else if addr.Address != "\x01" { + t.Fatalf("got GetMainNICAddress = %s, want = 1", addr.Address) + } + + { + subnet, err := tcpip.NewSubnet("\x00", "\x00") + if err != nil { + t.Fatalf("NewSubnet failed:", err) + } + s.SetRouteTable([]tcpip.Route{{Destination: subnet, Gateway: "\x00", NIC: 1}}) + } + + // Take a route through the address so its ref + // count gets incremented and does not actually + // get deleted when RemoveAddress is called + // below. This is because we want to test that a + // new peb is respected when an address gets + // "promoted" to permanent from a + // permanentExpired kind. + r, err := s.FindRoute(1, "\x01", "\x02", fakeNetNumber, false) + if err != nil { + t.Fatal("FindRoute failed:", err) + } + defer r.Release() + if err := s.RemoveAddress(1, "\x01"); err != nil { + t.Fatalf("RemoveAddress failed:", err) + } + + // + // At this point, the address should still be + // known by the NIC, but have its + // kind = permanentExpired. + // + + // Add some other address with peb set to + // FirstPrimaryEndpoint. + if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x03", stack.FirstPrimaryEndpoint); err != nil { + t.Fatal("AddAddressWithOptions failed:", err) + + } + + // Add back the address we removed earlier and + // make sure the new peb was respected. + // (The address should just be promoted now). + if err := s.AddAddressWithOptions(1, fakeNetNumber, "\x01", ps); err != nil { + t.Fatal("AddAddressWithOptions failed:", err) + } + var primaryAddrs []tcpip.Address + for _, pa := range s.NICInfo()[1].ProtocolAddresses { + primaryAddrs = append(primaryAddrs, pa.AddressWithPrefix.Address) + } + var expectedList []tcpip.Address + switch ps { + case stack.FirstPrimaryEndpoint: + expectedList = []tcpip.Address{ + "\x01", + "\x03", + } + case stack.CanBePrimaryEndpoint: + expectedList = []tcpip.Address{ + "\x03", + "\x01", + } + case stack.NeverPrimaryEndpoint: + expectedList = []tcpip.Address{ + "\x03", + } + } + if !cmp.Equal(primaryAddrs, expectedList) { + t.Fatalf("got NIC's primary addresses = %v, want = %v", primaryAddrs, expectedList) + } + + // Once we remove the other address, if the new + // peb, ps, was NeverPrimaryEndpoint, no address + // should be returned by a call to + // GetMainNICAddress; else, our original address + // should be returned. + if err := s.RemoveAddress(1, "\x03"); err != nil { + t.Fatalf("RemoveAddress failed:", err) + } + addr, err = s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { + t.Fatal("s.GetMainNICAddress failed:", err) + } + if ps == stack.NeverPrimaryEndpoint { + if want := (tcpip.AddressWithPrefix{}); addr != want { + t.Fatalf("got GetMainNICAddress = %s, want = %s", addr, want) + + } + } else { + if addr.Address != "\x01" { + t.Fatalf("got GetMainNICAddress = %s, want = 1", addr.Address) + } + } + }) + } + } +} diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index 92267ce4d..97a1aec4b 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -465,7 +465,7 @@ func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer func (d *transportDemuxer) registerRawEndpoint(netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error { eps, ok := d.protocol[protocolIDs{netProto, transProto}] if !ok { - return nil + return tcpip.ErrNotSupported } eps.mu.Lock() diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 678a94616..3d10001d8 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -255,7 +255,7 @@ type FullAddress struct { // This may not be used by all endpoint types. NIC NICID - // Addr is the network address. + // Addr is the network or link layer address. Addr Address // Port is the transport port. diff --git a/pkg/tcpip/transport/packet/BUILD b/pkg/tcpip/transport/packet/BUILD new file mode 100644 index 000000000..8ea2e6ee5 --- /dev/null +++ b/pkg/tcpip/transport/packet/BUILD @@ -0,0 +1,46 @@ +load("//tools/go_generics:defs.bzl", "go_template_instance") +load("//tools/go_stateify:defs.bzl", "go_library") + +package(licenses = ["notice"]) + +go_template_instance( + name = "packet_list", + out = "packet_list.go", + package = "packet", + prefix = "packet", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*packet", + "Linker": "*packet", + }, +) + +go_library( + name = "packet", + srcs = [ + "endpoint.go", + "endpoint_state.go", + "packet_list.go", + ], + importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/packet", + imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], + visibility = ["//visibility:public"], + deps = [ + "//pkg/log", + "//pkg/sleep", + "//pkg/tcpip", + "//pkg/tcpip/buffer", + "//pkg/tcpip/header", + "//pkg/tcpip/iptables", + "//pkg/tcpip/stack", + "//pkg/waiter", + ], +) + +filegroup( + name = "autogen", + srcs = [ + "packet_list.go", + ], + visibility = ["//:sandbox"], +) diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go new file mode 100644 index 000000000..73cdaa265 --- /dev/null +++ b/pkg/tcpip/transport/packet/endpoint.go @@ -0,0 +1,363 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package packet provides the implementation of packet sockets (see +// packet(7)). Packet sockets allow applications to: +// +// * manually write and inspect link, network, and transport headers +// * receive all traffic of a given network protocol, or all protocols +// +// Packet sockets are similar to raw sockets, but provide even more power to +// users, letting them effectively talk directly to the network device. +// +// Packet sockets skip the input and output iptables chains. +package packet + +import ( + "sync" + + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/iptables" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/waiter" +) + +// +stateify savable +type packet struct { + packetEntry + // data holds the actual packet data, including any headers and + // payload. + data buffer.VectorisedView `state:".(buffer.VectorisedView)"` + // views is pre-allocated space to back data. As long as the packet is + // made up of fewer than 8 buffer.Views, no extra allocation is + // necessary to store packet data. + views [8]buffer.View `state:"nosave"` + // timestampNS is the unix time at which the packet was received. + timestampNS int64 + // senderAddr is the network address of the sender. + senderAddr tcpip.FullAddress +} + +// endpoint is the packet socket implementation of tcpip.Endpoint. It is legal +// to have goroutines make concurrent calls into the endpoint. +// +// Lock order: +// endpoint.mu +// endpoint.rcvMu +// +// +stateify savable +type endpoint struct { + stack.TransportEndpointInfo + // The following fields are initialized at creation time and are + // immutable. + stack *stack.Stack `state:"manual"` + netProto tcpip.NetworkProtocolNumber + waiterQueue *waiter.Queue + cooked bool + + // The following fields are used to manage the receive queue and are + // protected by rcvMu. + rcvMu sync.Mutex `state:"nosave"` + rcvList packetList + rcvBufSizeMax int `state:".(int)"` + rcvBufSize int + rcvClosed bool + + // The following fields are protected by mu. + mu sync.RWMutex `state:"nosave"` + sndBufSize int + closed bool + stats tcpip.TransportEndpointStats `state:"nosave"` +} + +// NewEndpoint returns a new packet endpoint. +func NewEndpoint(s *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + ep := &endpoint{ + stack: s, + TransportEndpointInfo: stack.TransportEndpointInfo{ + NetProto: netProto, + }, + cooked: cooked, + netProto: netProto, + waiterQueue: waiterQueue, + rcvBufSizeMax: 32 * 1024, + sndBufSize: 32 * 1024, + } + + if err := s.RegisterPacketEndpoint(0, netProto, ep); err != nil { + return nil, err + } + return ep, nil +} + +// Close implements tcpip.Endpoint.Close. +func (ep *endpoint) Close() { + ep.mu.Lock() + defer ep.mu.Unlock() + + if ep.closed { + return + } + + ep.stack.UnregisterPacketEndpoint(0, ep.netProto, ep) + + ep.rcvMu.Lock() + defer ep.rcvMu.Unlock() + + // Clear the receive list. + ep.rcvClosed = true + ep.rcvBufSize = 0 + for !ep.rcvList.Empty() { + ep.rcvList.Remove(ep.rcvList.Front()) + } + + ep.closed = true + ep.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) +} + +// ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf. +func (ep *endpoint) ModerateRecvBuf(copied int) {} + +// IPTables implements tcpip.Endpoint.IPTables. +func (ep *endpoint) IPTables() (iptables.IPTables, error) { + return ep.stack.IPTables(), nil +} + +// Read implements tcpip.Endpoint.Read. +func (ep *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) { + ep.rcvMu.Lock() + + // If there's no data to read, return that read would block or that the + // endpoint is closed. + if ep.rcvList.Empty() { + err := tcpip.ErrWouldBlock + if ep.rcvClosed { + ep.stats.ReadErrors.ReadClosed.Increment() + err = tcpip.ErrClosedForReceive + } + ep.rcvMu.Unlock() + return buffer.View{}, tcpip.ControlMessages{}, err + } + + packet := ep.rcvList.Front() + ep.rcvList.Remove(packet) + ep.rcvBufSize -= packet.data.Size() + + ep.rcvMu.Unlock() + + if addr != nil { + *addr = packet.senderAddr + } + + return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil +} + +func (ep *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) { + // TODO(b/129292371): Implement. + return 0, nil, tcpip.ErrInvalidOptionValue +} + +// Peek implements tcpip.Endpoint.Peek. +func (ep *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { + return 0, tcpip.ControlMessages{}, nil +} + +// Disconnect implements tcpip.Endpoint.Disconnect. Packet sockets cannot be +// disconnected, and this function always returns tpcip.ErrNotSupported. +func (*endpoint) Disconnect() *tcpip.Error { + return tcpip.ErrNotSupported +} + +// Connect implements tcpip.Endpoint.Connect. Packet sockets cannot be +// connected, and this function always returnes tcpip.ErrNotSupported. +func (ep *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// Shutdown implements tcpip.Endpoint.Shutdown. Packet sockets cannot be used +// with Shutdown, and this function always returns tcpip.ErrNotSupported. +func (ep *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// Listen implements tcpip.Endpoint.Listen. Packet sockets cannot be used with +// Listen, and this function always returns tcpip.ErrNotSupported. +func (ep *endpoint) Listen(backlog int) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// Accept implements tcpip.Endpoint.Accept. Packet sockets cannot be used with +// Accept, and this function always returns tcpip.ErrNotSupported. +func (ep *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { + return nil, nil, tcpip.ErrNotSupported +} + +// Bind implements tcpip.Endpoint.Bind. +func (ep *endpoint) Bind(addr tcpip.FullAddress) *tcpip.Error { + // TODO(gvisor.dev/issue/173): Add Bind support. + + // "By default, all packets of the specified protocol type are passed + // to a packet socket. To get packets only from a specific interface + // use bind(2) specifying an address in a struct sockaddr_ll to bind + // the packet socket to an interface. Fields used for binding are + // sll_family (should be AF_PACKET), sll_protocol, and sll_ifindex." + // - packet(7). + + return tcpip.ErrNotSupported +} + +// GetLocalAddress implements tcpip.Endpoint.GetLocalAddress. +func (ep *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) { + return tcpip.FullAddress{}, tcpip.ErrNotSupported +} + +// GetRemoteAddress implements tcpip.Endpoint.GetRemoteAddress. +func (ep *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { + // Even a connected socket doesn't return a remote address. + return tcpip.FullAddress{}, tcpip.ErrNotConnected +} + +// Readiness implements tcpip.Endpoint.Readiness. +func (ep *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { + // The endpoint is always writable. + result := waiter.EventOut & mask + + // Determine whether the endpoint is readable. + if (mask & waiter.EventIn) != 0 { + ep.rcvMu.Lock() + if !ep.rcvList.Empty() || ep.rcvClosed { + result |= waiter.EventIn + } + ep.rcvMu.Unlock() + } + + return result +} + +// SetSockOpt implements tcpip.Endpoint.SetSockOpt. Packet sockets cannot be +// used with SetSockOpt, and this function always returns +// tcpip.ErrNotSupported. +func (ep *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. +func (ep *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { + return tcpip.ErrUnknownProtocolOption +} + +// GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. +func (ep *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { + return 0, tcpip.ErrNotSupported +} + +// GetSockOpt implements tcpip.Endpoint.GetSockOpt. +func (ep *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { + return tcpip.ErrNotSupported +} + +// HandlePacket implements stack.PacketEndpoint.HandlePacket. +func (ep *endpoint) HandlePacket(nicid tcpip.NICID, localAddr tcpip.LinkAddress, netProto tcpip.NetworkProtocolNumber, vv buffer.VectorisedView, ethHeader buffer.View) { + ep.rcvMu.Lock() + + // Drop the packet if our buffer is currently full. + if ep.rcvClosed { + ep.rcvMu.Unlock() + ep.stack.Stats().DroppedPackets.Increment() + ep.stats.ReceiveErrors.ClosedReceiver.Increment() + return + } + + if ep.rcvBufSize >= ep.rcvBufSizeMax { + ep.rcvMu.Unlock() + ep.stack.Stats().DroppedPackets.Increment() + ep.stats.ReceiveErrors.ReceiveBufferOverflow.Increment() + return + } + + wasEmpty := ep.rcvBufSize == 0 + + // Push new packet into receive list and increment the buffer size. + var packet packet + // TODO(b/129292371): Return network protocol. + if len(ethHeader) > 0 { + // Get info directly from the ethernet header. + hdr := header.Ethernet(ethHeader) + packet.senderAddr = tcpip.FullAddress{ + NIC: nicid, + Addr: tcpip.Address(hdr.SourceAddress()), + } + } else { + // Guess the would-be ethernet header. + packet.senderAddr = tcpip.FullAddress{ + NIC: nicid, + Addr: tcpip.Address(localAddr), + } + } + + if ep.cooked { + // Cooked packets can simply be queued. + packet.data = vv.Clone(packet.views[:]) + } else { + // Raw packets need their ethernet headers prepended before + // queueing. + if len(ethHeader) == 0 { + // We weren't provided with an actual ethernet header, + // so fake one. + ethFields := header.EthernetFields{ + SrcAddr: tcpip.LinkAddress([]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00}), + DstAddr: localAddr, + Type: netProto, + } + fakeHeader := make(header.Ethernet, header.EthernetMinimumSize) + fakeHeader.Encode(ðFields) + ethHeader = buffer.View(fakeHeader) + } + combinedVV := buffer.View(ethHeader).ToVectorisedView() + combinedVV.Append(vv) + packet.data = combinedVV.Clone(packet.views[:]) + } + packet.timestampNS = ep.stack.NowNanoseconds() + + ep.rcvList.PushBack(&packet) + ep.rcvBufSize += packet.data.Size() + + ep.rcvMu.Unlock() + ep.stats.PacketsReceived.Increment() + // Notify waiters that there's data to be read. + if wasEmpty { + ep.waiterQueue.Notify(waiter.EventIn) + } +} + +// State implements socket.Socket.State. +func (ep *endpoint) State() uint32 { + return 0 +} + +// Info returns a copy of the endpoint info. +func (ep *endpoint) Info() tcpip.EndpointInfo { + ep.mu.RLock() + // Make a copy of the endpoint info. + ret := ep.TransportEndpointInfo + ep.mu.RUnlock() + return &ret +} + +// Stats returns a pointer to the endpoint stats. +func (ep *endpoint) Stats() tcpip.EndpointStats { + return &ep.stats +} diff --git a/pkg/tcpip/transport/packet/endpoint_state.go b/pkg/tcpip/transport/packet/endpoint_state.go new file mode 100644 index 000000000..9b88f17e4 --- /dev/null +++ b/pkg/tcpip/transport/packet/endpoint_state.go @@ -0,0 +1,72 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package packet + +import ( + "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +// saveData saves packet.data field. +func (p *packet) saveData() buffer.VectorisedView { + // We cannot save p.data directly as p.data.views may alias to p.views, + // which is not allowed by state framework (in-struct pointer). + return p.data.Clone(nil) +} + +// loadData loads packet.data field. +func (p *packet) loadData(data buffer.VectorisedView) { + // NOTE: We cannot do the p.data = data.Clone(p.views[:]) optimization + // here because data.views is not guaranteed to be loaded by now. Plus, + // data.views will be allocated anyway so there really is little point + // of utilizing p.views for data.views. + p.data = data +} + +// beforeSave is invoked by stateify. +func (ep *endpoint) beforeSave() { + // Stop incoming packets from being handled (and mutate endpoint state). + // The lock will be released after saveRcvBufSizeMax(), which would have + // saved ep.rcvBufSizeMax and set it to 0 to continue blocking incoming + // packets. + ep.rcvMu.Lock() +} + +// saveRcvBufSizeMax is invoked by stateify. +func (ep *endpoint) saveRcvBufSizeMax() int { + max := ep.rcvBufSizeMax + // Make sure no new packets will be handled regardless of the lock. + ep.rcvBufSizeMax = 0 + // Release the lock acquired in beforeSave() so regular endpoint closing + // logic can proceed after save. + ep.rcvMu.Unlock() + return max +} + +// loadRcvBufSizeMax is invoked by stateify. +func (ep *endpoint) loadRcvBufSizeMax(max int) { + ep.rcvBufSizeMax = max +} + +// afterLoad is invoked by stateify. +func (ep *endpoint) afterLoad() { + // StackFromEnv is a stack used specifically for save/restore. + ep.stack = stack.StackFromEnv + + // TODO(gvisor.dev/173): Once bind is supported, choose the right NIC. + if err := ep.stack.RegisterPacketEndpoint(0, ep.netProto, ep); err != nil { + panic(*err) + } +} diff --git a/pkg/tcpip/transport/raw/BUILD b/pkg/tcpip/transport/raw/BUILD index fba598d51..4af49218c 100644 --- a/pkg/tcpip/transport/raw/BUILD +++ b/pkg/tcpip/transport/raw/BUILD @@ -4,14 +4,14 @@ load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) go_template_instance( - name = "packet_list", - out = "packet_list.go", + name = "raw_packet_list", + out = "raw_packet_list.go", package = "raw", - prefix = "packet", + prefix = "rawPacket", template = "//pkg/ilist:generic_list", types = { - "Element": "*packet", - "Linker": "*packet", + "Element": "*rawPacket", + "Linker": "*rawPacket", }, ) @@ -20,8 +20,8 @@ go_library( srcs = [ "endpoint.go", "endpoint_state.go", - "packet_list.go", "protocol.go", + "raw_packet_list.go", ], importpath = "gvisor.dev/gvisor/pkg/tcpip/transport/raw", imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], @@ -34,6 +34,7 @@ go_library( "//pkg/tcpip/header", "//pkg/tcpip/iptables", "//pkg/tcpip/stack", + "//pkg/tcpip/transport/packet", "//pkg/waiter", ], ) @@ -41,7 +42,7 @@ go_library( filegroup( name = "autogen", srcs = [ - "packet_list.go", + "raw_packet_list.go", ], visibility = ["//:sandbox"], ) diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index b4c660859..308f10d24 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -17,8 +17,7 @@ // // * manually write and inspect transport layer headers and payloads // * receive all traffic of a given transport protocol (e.g. ICMP or UDP) -// * optionally write and inspect network layer and link layer headers for -// packets +// * optionally write and inspect network layer headers of packets // // Raw sockets don't have any notion of ports, and incoming packets are // demultiplexed solely by protocol number. Thus, a raw UDP endpoint will @@ -38,8 +37,8 @@ import ( ) // +stateify savable -type packet struct { - packetEntry +type rawPacket struct { + rawPacketEntry // data holds the actual packet data, including any headers and // payload. data buffer.VectorisedView `state:".(buffer.VectorisedView)"` @@ -72,7 +71,7 @@ type endpoint struct { // The following fields are used to manage the receive queue and are // protected by rcvMu. rcvMu sync.Mutex `state:"nosave"` - rcvList packetList + rcvList rawPacketList rcvBufSizeMax int `state:".(int)"` rcvBufSize int rcvClosed bool @@ -90,7 +89,6 @@ type endpoint struct { } // NewEndpoint returns a raw endpoint for the given protocols. -// TODO(b/129292371): IP_HDRINCL and AF_PACKET. func NewEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return newEndpoint(stack, netProto, transProto, waiterQueue, true /* associated */) } @@ -187,17 +185,17 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess return buffer.View{}, tcpip.ControlMessages{}, err } - packet := e.rcvList.Front() - e.rcvList.Remove(packet) - e.rcvBufSize -= packet.data.Size() + pkt := e.rcvList.Front() + e.rcvList.Remove(pkt) + e.rcvBufSize -= pkt.data.Size() e.rcvMu.Unlock() if addr != nil { - *addr = packet.senderAddr + *addr = pkt.senderAddr } - return packet.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: packet.timestampNS}, nil + return pkt.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: pkt.timestampNS}, nil } // Write implements tcpip.Endpoint.Write. @@ -602,7 +600,7 @@ func (e *endpoint) HandlePacket(route *stack.Route, netHeader buffer.View, vv bu wasEmpty := e.rcvBufSize == 0 // Push new packet into receive list and increment the buffer size. - packet := &packet{ + pkt := &rawPacket{ senderAddr: tcpip.FullAddress{ NIC: route.NICID(), Addr: route.RemoteAddress, @@ -611,11 +609,11 @@ func (e *endpoint) HandlePacket(route *stack.Route, netHeader buffer.View, vv bu combinedVV := netHeader.ToVectorisedView() combinedVV.Append(vv) - packet.data = combinedVV.Clone(packet.views[:]) - packet.timestampNS = e.stack.NowNanoseconds() + pkt.data = combinedVV.Clone(pkt.views[:]) + pkt.timestampNS = e.stack.NowNanoseconds() - e.rcvList.PushBack(packet) - e.rcvBufSize += packet.data.Size() + e.rcvList.PushBack(pkt) + e.rcvBufSize += pkt.data.Size() e.rcvMu.Unlock() e.stats.PacketsReceived.Increment() diff --git a/pkg/tcpip/transport/raw/endpoint_state.go b/pkg/tcpip/transport/raw/endpoint_state.go index a6c7cc43a..33bfb56cd 100644 --- a/pkg/tcpip/transport/raw/endpoint_state.go +++ b/pkg/tcpip/transport/raw/endpoint_state.go @@ -20,15 +20,15 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/stack" ) -// saveData saves packet.data field. -func (p *packet) saveData() buffer.VectorisedView { +// saveData saves rawPacket.data field. +func (p *rawPacket) saveData() buffer.VectorisedView { // We cannot save p.data directly as p.data.views may alias to p.views, // which is not allowed by state framework (in-struct pointer). return p.data.Clone(nil) } -// loadData loads packet.data field. -func (p *packet) loadData(data buffer.VectorisedView) { +// loadData loads rawPacket.data field. +func (p *rawPacket) loadData(data buffer.VectorisedView) { // NOTE: We cannot do the p.data = data.Clone(p.views[:]) optimization // here because data.views is not guaranteed to be loaded by now. Plus, // data.views will be allocated anyway so there really is little point @@ -86,7 +86,9 @@ func (ep *endpoint) Resume(s *stack.Stack) { } } - if err := ep.stack.RegisterRawTransportEndpoint(ep.RegisterNICID, ep.NetProto, ep.TransProto, ep); err != nil { - panic(err) + if ep.associated { + if err := ep.stack.RegisterRawTransportEndpoint(ep.RegisterNICID, ep.NetProto, ep.TransProto, ep); err != nil { + panic(err) + } } } diff --git a/pkg/tcpip/transport/raw/protocol.go b/pkg/tcpip/transport/raw/protocol.go index a2512d666..f30aa2a4a 100644 --- a/pkg/tcpip/transport/raw/protocol.go +++ b/pkg/tcpip/transport/raw/protocol.go @@ -17,13 +17,19 @@ package raw import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/packet" "gvisor.dev/gvisor/pkg/waiter" ) -// EndpointFactory implements stack.UnassociatedEndpointFactory. +// EndpointFactory implements stack.RawFactory. type EndpointFactory struct{} -// NewUnassociatedRawEndpoint implements stack.UnassociatedEndpointFactory. -func (EndpointFactory) NewUnassociatedRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +// NewUnassociatedEndpoint implements stack.RawFactory.NewUnassociatedEndpoint. +func (EndpointFactory) NewUnassociatedEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return newEndpoint(stack, netProto, transProto, waiterQueue, false /* associated */) } + +// NewPacketEndpoint implements stack.RawFactory.NewPacketEndpoint. +func (EndpointFactory) NewPacketEndpoint(stack *stack.Stack, cooked bool, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { + return packet.NewEndpoint(stack, cooked, netProto, waiterQueue) +} diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index aed70e06f..f1dbc6f91 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -44,6 +44,7 @@ go_library( imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ + "//pkg/log", "//pkg/rand", "//pkg/sleep", "//pkg/tcpip", diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 5ea036bea..082135374 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -607,17 +607,11 @@ func (e *endpoint) sendTCP(r *stack.Route, id stack.TransportEndpointID, data bu return nil } -// sendTCP sends a TCP segment with the provided options via the provided -// network endpoint and under the provided identity. -func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { +func buildTCPHdr(r *stack.Route, id stack.TransportEndpointID, d *stack.PacketDescriptor, data buffer.VectorisedView, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) { optLen := len(opts) - // Allocate a buffer for the TCP header. - hdr := buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen) - - if rcvWnd > 0xffff { - rcvWnd = 0xffff - } - + hdr := &d.Hdr + packetSize := d.Size + off := d.Off // Initialize the header. tcp := header.TCP(hdr.Prepend(header.TCPMinimumSize + optLen)) tcp.Encode(&header.TCPFields{ @@ -631,7 +625,7 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise }) copy(tcp[header.TCPMinimumSize:], opts) - length := uint16(hdr.UsedLength() + data.Size()) + length := uint16(hdr.UsedLength() + packetSize) xsum := r.PseudoHeaderChecksum(ProtocolNumber, length) // Only calculate the checksum if offloading isn't supported. if gso != nil && gso.NeedsCsum { @@ -641,14 +635,71 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise // header and data and get the right sum of the TCP packet. tcp.SetChecksum(xsum) } else if r.Capabilities()&stack.CapabilityTXChecksumOffload == 0 { - xsum = header.ChecksumVV(data, xsum) + xsum = header.ChecksumVVWithOffset(data, xsum, off, packetSize) tcp.SetChecksum(^tcp.CalculateChecksum(xsum)) } +} + +func sendTCPBatch(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { + optLen := len(opts) + if rcvWnd > 0xffff { + rcvWnd = 0xffff + } + + mss := int(gso.MSS) + n := (data.Size() + mss - 1) / mss + + hdrs := stack.NewPacketDescriptors(n, header.TCPMinimumSize+int(r.MaxHeaderLength())+optLen) + + size := data.Size() + off := 0 + for i := 0; i < n; i++ { + packetSize := mss + if packetSize > size { + packetSize = size + } + size -= packetSize + hdrs[i].Off = off + hdrs[i].Size = packetSize + buildTCPHdr(r, id, &hdrs[i], data, flags, seq, ack, rcvWnd, opts, gso) + off += packetSize + seq = seq.Add(seqnum.Size(packetSize)) + } + if ttl == 0 { + ttl = r.DefaultTTL() + } + sent, err := r.WritePackets(gso, hdrs, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}) + if err != nil { + r.Stats().TCP.SegmentSendErrors.IncrementBy(uint64(n - sent)) + } + r.Stats().TCP.SegmentsSent.IncrementBy(uint64(sent)) + return err +} + +// sendTCP sends a TCP segment with the provided options via the provided +// network endpoint and under the provided identity. +func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.VectorisedView, ttl, tos uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts []byte, gso *stack.GSO) *tcpip.Error { + optLen := len(opts) + if rcvWnd > 0xffff { + rcvWnd = 0xffff + } + + if r.Loop&stack.PacketLoop == 0 && gso != nil && gso.Type == stack.GSOSW && int(gso.MSS) < data.Size() { + return sendTCPBatch(r, id, data, ttl, tos, flags, seq, ack, rcvWnd, opts, gso) + } + + d := &stack.PacketDescriptor{ + Hdr: buffer.NewPrependable(header.TCPMinimumSize + int(r.MaxHeaderLength()) + optLen), + Off: 0, + Size: data.Size(), + } + buildTCPHdr(r, id, d, data, flags, seq, ack, rcvWnd, opts, gso) + if ttl == 0 { ttl = r.DefaultTTL() } - if err := r.WritePacket(gso, hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { + if err := r.WritePacket(gso, d.Hdr, data, stack.NetworkHeaderParams{Protocol: ProtocolNumber, TTL: ttl, TOS: tos}); err != nil { r.Stats().TCP.SegmentSendErrors.Increment() return err } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index a1b784b49..8b9cb4c33 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -2327,11 +2327,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState { return s } -func (e *endpoint) initGSO() { - if e.route.Capabilities()&stack.CapabilityGSO == 0 { - return - } - +func (e *endpoint) initHardwareGSO() { gso := &stack.GSO{} switch e.route.NetProto { case header.IPv4ProtocolNumber: @@ -2349,6 +2345,18 @@ func (e *endpoint) initGSO() { e.gso = gso } +func (e *endpoint) initGSO() { + if e.route.Capabilities()&stack.CapabilityHardwareGSO != 0 { + e.initHardwareGSO() + } else if e.route.Capabilities()&stack.CapabilitySoftwareGSO != 0 { + e.gso = &stack.GSO{ + MaxSize: e.route.GSOMaxSize(), + Type: stack.GSOSW, + NeedsCsum: false, + } + } +} + // State implements tcpip.Endpoint.State. It exports the endpoint's protocol // state for diagnostics. func (e *endpoint) State() uint32 { diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index de026880f..5c3358a5e 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -121,8 +121,15 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans payloadLen = available } - payload := buffer.NewVectorisedView(len(netHeader), []buffer.View{netHeader}) - payload.Append(vv) + // The buffers used by vv and netHeader 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 vv and netHeader to prevent multiple ownership and SR + // errors. + newNetHeader := make(buffer.View, len(netHeader)) + copy(newNetHeader, netHeader) + payload := buffer.NewVectorisedView(len(newNetHeader), []buffer.View{newNetHeader}) + payload.Append(vv.ToView().ToVectorisedView()) payload.CapLength(payloadLen) hdr := buffer.NewPrependable(headerLen) |