diff options
Diffstat (limited to 'pkg')
105 files changed, 3316 insertions, 1349 deletions
diff --git a/pkg/abi/linux/file.go b/pkg/abi/linux/file.go index 7d742871a..257f67222 100644 --- a/pkg/abi/linux/file.go +++ b/pkg/abi/linux/file.go @@ -271,7 +271,7 @@ type Statx struct { } // FileMode represents a mode_t. -type FileMode uint +type FileMode uint16 // Permissions returns just the permission bits. func (m FileMode) Permissions() FileMode { diff --git a/pkg/bits/BUILD b/pkg/bits/BUILD index 51967b811..1b5dac99a 100644 --- a/pkg/bits/BUILD +++ b/pkg/bits/BUILD @@ -11,8 +11,9 @@ go_library( "bits.go", "bits32.go", "bits64.go", - "uint64_arch_amd64.go", + "uint64_arch.go", "uint64_arch_amd64_asm.s", + "uint64_arch_arm64_asm.s", "uint64_arch_generic.go", ], importpath = "gvisor.dev/gvisor/pkg/bits", diff --git a/pkg/bits/uint64_arch_amd64.go b/pkg/bits/uint64_arch.go index faccaa61a..9f23eff77 100644 --- a/pkg/bits/uint64_arch_amd64.go +++ b/pkg/bits/uint64_arch.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build amd64 +// +build amd64 arm64 package bits diff --git a/pkg/bits/uint64_arch_arm64_asm.s b/pkg/bits/uint64_arch_arm64_asm.s new file mode 100644 index 000000000..814ba562d --- /dev/null +++ b/pkg/bits/uint64_arch_arm64_asm.s @@ -0,0 +1,33 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +TEXT ·TrailingZeros64(SB),$0-16 + MOVD x+0(FP), R0 + RBIT R0, R0 + CLZ R0, R0 // return 64 if x == 0 + MOVD R0, ret+8(FP) + RET + +TEXT ·MostSignificantOne64(SB),$0-16 + MOVD x+0(FP), R0 + CLZ R0, R0 // return 64 if x == 0 + MOVD $63, R1 + SUBS R0, R1, R0 // ret = 63 - CLZ + BPL end + MOVD $64, R0 // x == 0 +end: + MOVD R0, ret+8(FP) + RET diff --git a/pkg/bits/uint64_arch_generic.go b/pkg/bits/uint64_arch_generic.go index 7dd2d1480..9dd2098d1 100644 --- a/pkg/bits/uint64_arch_generic.go +++ b/pkg/bits/uint64_arch_generic.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build !amd64 +// +build !amd64,!arm64 package bits diff --git a/pkg/fd/BUILD b/pkg/fd/BUILD index afa8f7659..c7f549428 100644 --- a/pkg/fd/BUILD +++ b/pkg/fd/BUILD @@ -8,6 +8,9 @@ go_library( srcs = ["fd.go"], importpath = "gvisor.dev/gvisor/pkg/fd", visibility = ["//visibility:public"], + deps = [ + "//pkg/unet", + ], ) go_test( diff --git a/pkg/fd/fd.go b/pkg/fd/fd.go index 83bcfe220..7691b477b 100644 --- a/pkg/fd/fd.go +++ b/pkg/fd/fd.go @@ -22,6 +22,8 @@ import ( "runtime" "sync/atomic" "syscall" + + "gvisor.dev/gvisor/pkg/unet" ) // ReadWriter implements io.ReadWriter, io.ReaderAt, and io.WriterAt for fd. It @@ -185,6 +187,12 @@ func OpenAt(dir *FD, path string, flags int, mode uint32) (*FD, error) { return New(f), nil } +// DialUnix connects to a Unix Domain Socket and return the file descriptor. +func DialUnix(path string) (*FD, error) { + socket, err := unet.Connect(path, false) + return New(socket.FD()), err +} + // Close closes the file descriptor contained in the FD. // // Close is safe to call multiple times, but will return an error after the diff --git a/pkg/p9/client.go b/pkg/p9/client.go index 2412aa5e1..221516c6c 100644 --- a/pkg/p9/client.go +++ b/pkg/p9/client.go @@ -505,12 +505,27 @@ func (c *Client) sendRecvChannel(t message, r message) error { ch.active = false c.channelsMu.Unlock() c.channelsWg.Done() - return err + // Map all transport errors to EIO, but ensure that the real error + // is logged. + log.Warningf("p9.Client.sendRecvChannel: flipcall.Endpoint.Connect: %v", err) + return syscall.EIO } } - // Send the message. - err := ch.sendRecv(c, t, r) + // Send the request and receive the server's response. + rsz, err := ch.send(t) + if err != nil { + // See above. + c.channelsMu.Lock() + ch.active = false + c.channelsMu.Unlock() + c.channelsWg.Done() + log.Warningf("p9.Client.sendRecvChannel: p9.channel.send: %v", err) + return syscall.EIO + } + + // Parse the server's response. + _, retErr := ch.recv(r, rsz) // Release the channel. c.channelsMu.Lock() @@ -523,7 +538,7 @@ func (c *Client) sendRecvChannel(t message, r message) error { c.channelsMu.Unlock() c.channelsWg.Done() - return err + return retErr } // Version returns the negotiated 9P2000.L.Google version number. diff --git a/pkg/p9/server.go b/pkg/p9/server.go index 69c886a5d..e717e6161 100644 --- a/pkg/p9/server.go +++ b/pkg/p9/server.go @@ -452,7 +452,9 @@ func (cs *connState) initializeChannels() (err error) { cs.channelWg.Add(1) go func() { // S/R-SAFE: Server side. defer cs.channelWg.Done() - res.service(cs) + if err := res.service(cs); err != nil { + log.Warningf("p9.channel.service: %v", err) + } }() } diff --git a/pkg/p9/transport_flipcall.go b/pkg/p9/transport_flipcall.go index 7cdf4ecc3..233f825e3 100644 --- a/pkg/p9/transport_flipcall.go +++ b/pkg/p9/transport_flipcall.go @@ -132,7 +132,7 @@ func (ch *channel) send(m message) (uint32, error) { if filer, ok := m.(filer); ok { if f := filer.FilePayload(); f != nil { if err := ch.fds.SendFD(f.FD()); err != nil { - return 0, syscall.EIO // Map everything to EIO. + return 0, err } f.Close() // Per sendRecvLegacy. sentFD = true // To mark below. @@ -162,15 +162,7 @@ func (ch *channel) send(m message) (uint32, error) { } // Perform the one-shot communication. - n, err := ch.data.SendRecv(ssz) - if err != nil { - if n > 0 { - return n, nil - } - return 0, syscall.EIO // See above. - } - - return n, nil + return ch.data.SendRecv(ssz) } // recv decodes a message that exists on the channel. @@ -249,15 +241,3 @@ func (ch *channel) recv(r message, rsz uint32) (message, error) { return r, nil } - -// sendRecv sends the given message over the channel. -// -// This is used by the client. -func (ch *channel) sendRecv(c *Client, m, r message) error { - rsz, err := ch.send(m) - if err != nil { - return err - } - _, err = ch.recv(r, rsz) - return err -} diff --git a/pkg/seccomp/seccomp_unsafe.go b/pkg/seccomp/seccomp_unsafe.go index 0a3d92854..be328db12 100644 --- a/pkg/seccomp/seccomp_unsafe.go +++ b/pkg/seccomp/seccomp_unsafe.go @@ -35,7 +35,7 @@ type sockFprog struct { //go:nosplit func SetFilter(instrs []linux.BPFInstruction) syscall.Errno { // PR_SET_NO_NEW_PRIVS is required in order to enable seccomp. See seccomp(2) for details. - if _, _, errno := syscall.RawSyscall(syscall.SYS_PRCTL, linux.PR_SET_NO_NEW_PRIVS, 1, 0); errno != 0 { + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PRCTL, linux.PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0, 0); errno != 0 { return errno } diff --git a/pkg/sentry/fs/fsutil/host_mappable.go b/pkg/sentry/fs/fsutil/host_mappable.go index d2495cb83..693625ddc 100644 --- a/pkg/sentry/fs/fsutil/host_mappable.go +++ b/pkg/sentry/fs/fsutil/host_mappable.go @@ -144,7 +144,7 @@ func (h *HostMappable) Truncate(ctx context.Context, newSize int64) error { mask := fs.AttrMask{Size: true} attr := fs.UnstableAttr{Size: newSize} - if err := h.backingFile.SetMaskedAttributes(ctx, mask, attr); err != nil { + if err := h.backingFile.SetMaskedAttributes(ctx, mask, attr, false); err != nil { return err } diff --git a/pkg/sentry/fs/fsutil/inode_cached.go b/pkg/sentry/fs/fsutil/inode_cached.go index d404a79d4..dd80757dc 100644 --- a/pkg/sentry/fs/fsutil/inode_cached.go +++ b/pkg/sentry/fs/fsutil/inode_cached.go @@ -140,12 +140,16 @@ type CachedFileObject interface { // WriteFromBlocksAt may return a partial write without an error. WriteFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) - // SetMaskedAttributes sets the attributes in attr that are true in mask - // on the backing file. + // SetMaskedAttributes sets the attributes in attr that are true in + // mask on the backing file. If the mask contains only ATime or MTime + // and the CachedFileObject has an FD to the file, then this operation + // is a noop unless forceSetTimestamps is true. This avoids an extra + // RPC to the gofer in the open-read/write-close case, when the + // timestamps on the file will be updated by the host kernel for us. // // SetMaskedAttributes may be called at any point, regardless of whether // the file was opened. - SetMaskedAttributes(ctx context.Context, mask fs.AttrMask, attr fs.UnstableAttr) error + SetMaskedAttributes(ctx context.Context, mask fs.AttrMask, attr fs.UnstableAttr, forceSetTimestamps bool) error // Allocate allows the caller to reserve disk space for the inode. // It's equivalent to fallocate(2) with 'mode=0'. @@ -224,7 +228,7 @@ func (c *CachingInodeOperations) SetPermissions(ctx context.Context, inode *fs.I now := ktime.NowFromContext(ctx) masked := fs.AttrMask{Perms: true} - if err := c.backingFile.SetMaskedAttributes(ctx, masked, fs.UnstableAttr{Perms: perms}); err != nil { + if err := c.backingFile.SetMaskedAttributes(ctx, masked, fs.UnstableAttr{Perms: perms}, false); err != nil { return false } c.attr.Perms = perms @@ -246,7 +250,7 @@ func (c *CachingInodeOperations) SetOwner(ctx context.Context, inode *fs.Inode, UID: owner.UID.Ok(), GID: owner.GID.Ok(), } - if err := c.backingFile.SetMaskedAttributes(ctx, masked, fs.UnstableAttr{Owner: owner}); err != nil { + if err := c.backingFile.SetMaskedAttributes(ctx, masked, fs.UnstableAttr{Owner: owner}, false); err != nil { return err } if owner.UID.Ok() { @@ -282,7 +286,9 @@ func (c *CachingInodeOperations) SetTimestamps(ctx context.Context, inode *fs.In AccessTime: !ts.ATimeOmit, ModificationTime: !ts.MTimeOmit, } - if err := c.backingFile.SetMaskedAttributes(ctx, masked, fs.UnstableAttr{AccessTime: ts.ATime, ModificationTime: ts.MTime}); err != nil { + // Call SetMaskedAttributes with forceSetTimestamps = true to make sure + // the timestamp is updated. + if err := c.backingFile.SetMaskedAttributes(ctx, masked, fs.UnstableAttr{AccessTime: ts.ATime, ModificationTime: ts.MTime}, true); err != nil { return err } if !ts.ATimeOmit { @@ -305,7 +311,7 @@ func (c *CachingInodeOperations) Truncate(ctx context.Context, inode *fs.Inode, now := ktime.NowFromContext(ctx) masked := fs.AttrMask{Size: true} attr := fs.UnstableAttr{Size: size} - if err := c.backingFile.SetMaskedAttributes(ctx, masked, attr); err != nil { + if err := c.backingFile.SetMaskedAttributes(ctx, masked, attr, false); err != nil { c.dataMu.Unlock() return err } @@ -394,7 +400,7 @@ func (c *CachingInodeOperations) WriteOut(ctx context.Context, inode *fs.Inode) c.dirtyAttr.Size = false // Write out cached attributes. - if err := c.backingFile.SetMaskedAttributes(ctx, c.dirtyAttr, c.attr); err != nil { + if err := c.backingFile.SetMaskedAttributes(ctx, c.dirtyAttr, c.attr, false); err != nil { c.attrMu.Unlock() return err } diff --git a/pkg/sentry/fs/fsutil/inode_cached_test.go b/pkg/sentry/fs/fsutil/inode_cached_test.go index eb5730c35..129f314c8 100644 --- a/pkg/sentry/fs/fsutil/inode_cached_test.go +++ b/pkg/sentry/fs/fsutil/inode_cached_test.go @@ -39,7 +39,7 @@ func (noopBackingFile) WriteFromBlocksAt(ctx context.Context, srcs safemem.Block return srcs.NumBytes(), nil } -func (noopBackingFile) SetMaskedAttributes(context.Context, fs.AttrMask, fs.UnstableAttr) error { +func (noopBackingFile) SetMaskedAttributes(context.Context, fs.AttrMask, fs.UnstableAttr, bool) error { return nil } @@ -230,7 +230,7 @@ func (f *sliceBackingFile) WriteFromBlocksAt(ctx context.Context, srcs safemem.B return w.WriteFromBlocks(srcs) } -func (*sliceBackingFile) SetMaskedAttributes(context.Context, fs.AttrMask, fs.UnstableAttr) error { +func (*sliceBackingFile) SetMaskedAttributes(context.Context, fs.AttrMask, fs.UnstableAttr, bool) error { return nil } diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 95b064aea..d918d6620 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -215,8 +215,8 @@ func (i *inodeFileState) WriteFromBlocksAt(ctx context.Context, srcs safemem.Blo } // SetMaskedAttributes implements fsutil.CachedFileObject.SetMaskedAttributes. -func (i *inodeFileState) SetMaskedAttributes(ctx context.Context, mask fs.AttrMask, attr fs.UnstableAttr) error { - if i.skipSetAttr(mask) { +func (i *inodeFileState) SetMaskedAttributes(ctx context.Context, mask fs.AttrMask, attr fs.UnstableAttr, forceSetTimestamps bool) error { + if i.skipSetAttr(mask, forceSetTimestamps) { return nil } as, ans := attr.AccessTime.Unix() @@ -251,13 +251,14 @@ func (i *inodeFileState) SetMaskedAttributes(ctx context.Context, mask fs.AttrMa // when: // - Mask is empty // - Mask contains only attributes that cannot be set in the gofer -// - Mask contains only atime and/or mtime, and host FD exists +// - forceSetTimestamps is false and mask contains only atime and/or mtime +// and host FD exists // // Updates to atime and mtime can be skipped because cached value will be // "close enough" to host value, given that operation went directly to host FD. // Skipping atime updates is particularly important to reduce the number of // operations sent to the Gofer for readonly files. -func (i *inodeFileState) skipSetAttr(mask fs.AttrMask) bool { +func (i *inodeFileState) skipSetAttr(mask fs.AttrMask, forceSetTimestamps bool) bool { // First remove attributes that cannot be updated. cpy := mask cpy.Type = false @@ -277,6 +278,12 @@ func (i *inodeFileState) skipSetAttr(mask fs.AttrMask) bool { return false } + // If forceSetTimestamps was passed, then we cannot skip. + if forceSetTimestamps { + return false + } + + // Skip if we have a host FD. i.handlesMu.RLock() defer i.handlesMu.RUnlock() return (i.readHandles != nil && i.readHandles.Host != nil) || diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go index 894ab01f0..a6e4a09e3 100644 --- a/pkg/sentry/fs/host/inode.go +++ b/pkg/sentry/fs/host/inode.go @@ -114,7 +114,7 @@ func (i *inodeFileState) WriteFromBlocksAt(ctx context.Context, srcs safemem.Blo } // SetMaskedAttributes implements fsutil.CachedFileObject.SetMaskedAttributes. -func (i *inodeFileState) SetMaskedAttributes(ctx context.Context, mask fs.AttrMask, attr fs.UnstableAttr) error { +func (i *inodeFileState) SetMaskedAttributes(ctx context.Context, mask fs.AttrMask, attr fs.UnstableAttr, _ bool) error { if mask.Empty() { return nil } @@ -163,7 +163,7 @@ func (i *inodeFileState) unstableAttr(ctx context.Context) (fs.UnstableAttr, err return unstableAttr(i.mops, &s), nil } -// SetMaskedAttributes implements fsutil.CachedFileObject.SetMaskedAttributes. +// Allocate implements fsutil.CachedFileObject.Allocate. func (i *inodeFileState) Allocate(_ context.Context, offset, length int64) error { return syscall.Fallocate(i.FD(), 0, offset, length) } diff --git a/pkg/sentry/fs/host/tty.go b/pkg/sentry/fs/host/tty.go index 2526412a4..90331e3b2 100644 --- a/pkg/sentry/fs/host/tty.go +++ b/pkg/sentry/fs/host/tty.go @@ -43,12 +43,15 @@ type TTYFileOperations struct { // fgProcessGroup is the foreground process group that is currently // connected to this TTY. fgProcessGroup *kernel.ProcessGroup + + termios linux.KernelTermios } // newTTYFile returns a new fs.File that wraps a TTY FD. func newTTYFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags, iops *inodeOperations) *fs.File { return fs.NewFile(ctx, dirent, flags, &TTYFileOperations{ fileOperations: fileOperations{iops: iops}, + termios: linux.DefaultSlaveTermios, }) } @@ -97,9 +100,12 @@ func (t *TTYFileOperations) Write(ctx context.Context, file *fs.File, src userme t.mu.Lock() defer t.mu.Unlock() - // Are we allowed to do the write? - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - return 0, err + // Check whether TOSTOP is enabled. This corresponds to the check in + // drivers/tty/n_tty.c:n_tty_write(). + if t.termios.LEnabled(linux.TOSTOP) { + if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { + return 0, err + } } return t.fileOperations.Write(ctx, file, src, offset) } @@ -144,6 +150,9 @@ func (t *TTYFileOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO return 0, err } err := ioctlSetTermios(fd, ioctl, &termios) + if err == nil { + t.termios.FromTermios(termios) + } return 0, err case linux.TIOCGPGRP: diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 246b97161..5a388dad1 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -15,6 +15,7 @@ package fs import ( + "fmt" "strings" "gvisor.dev/gvisor/pkg/abi/linux" @@ -207,6 +208,11 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name } func overlayCreate(ctx context.Context, o *overlayEntry, parent *Dirent, name string, flags FileFlags, perm FilePermissions) (*File, error) { + // Sanity check. + if parent.Inode.overlay == nil { + panic(fmt.Sprintf("overlayCreate called with non-overlay parent inode (parent InodeOperations type is %T)", parent.Inode.InodeOperations)) + } + // Dirent.Create takes renameMu if the Inode is an overlay Inode. if err := copyUpLockedForRename(ctx, parent); err != nil { return nil, err diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 5e28982c5..f70239449 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -66,7 +66,7 @@ func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSo if s.SupportsIPv6() { contents["if_inet6"] = seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc) contents["ipv6_route"] = newStaticProcInode(ctx, msrc, []byte("")) - contents["tcp6"] = newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")) + contents["tcp6"] = seqfile.NewSeqFileInode(ctx, &netTCP6{k: k}, msrc) contents["udp6"] = newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")) } } @@ -310,44 +310,51 @@ func networkToHost16(n uint16) uint16 { return usermem.ByteOrder.Uint16(buf[:]) } -func writeInetAddr(w io.Writer, a linux.SockAddrInet) { - // linux.SockAddrInet.Port is stored in the network byte order and is - // printed like a number in host byte order. Note that all numbers in host - // byte order are printed with the most-significant byte first when - // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. - port := networkToHost16(a.Port) - - // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order - // (i.e. most-significant byte in index 0). Linux represents this as a - // __be32 which is a typedef for an unsigned int, and is printed with - // %X. This means that for a little-endian machine, Linux prints the - // least-significant byte of the address first. To emulate this, we first - // invert the byte order for the address using usermem.ByteOrder.Uint32, - // which makes it have the equivalent encoding to a __be32 on a little - // endian machine. Note that this operation is a no-op on a big endian - // machine. Then similar to Linux, we format it with %X, which will print - // the most-significant byte of the __be32 address first, which is now - // actually the least-significant byte of the original address in - // linux.SockAddrInet.Addr on little endian machines, due to the conversion. - addr := usermem.ByteOrder.Uint32(a.Addr[:]) - - fmt.Fprintf(w, "%08X:%04X ", addr, port) -} +func writeInetAddr(w io.Writer, family int, i linux.SockAddr) { + switch family { + case linux.AF_INET: + var a linux.SockAddrInet + if i != nil { + a = *i.(*linux.SockAddrInet) + } -// netTCP implements seqfile.SeqSource for /proc/net/tcp. -// -// +stateify savable -type netTCP struct { - k *kernel.Kernel -} + // linux.SockAddrInet.Port is stored in the network byte order and is + // printed like a number in host byte order. Note that all numbers in host + // byte order are printed with the most-significant byte first when + // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. + port := networkToHost16(a.Port) + + // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order + // (i.e. most-significant byte in index 0). Linux represents this as a + // __be32 which is a typedef for an unsigned int, and is printed with + // %X. This means that for a little-endian machine, Linux prints the + // least-significant byte of the address first. To emulate this, we first + // invert the byte order for the address using usermem.ByteOrder.Uint32, + // which makes it have the equivalent encoding to a __be32 on a little + // endian machine. Note that this operation is a no-op on a big endian + // machine. Then similar to Linux, we format it with %X, which will print + // the most-significant byte of the __be32 address first, which is now + // actually the least-significant byte of the original address in + // linux.SockAddrInet.Addr on little endian machines, due to the conversion. + addr := usermem.ByteOrder.Uint32(a.Addr[:]) + + fmt.Fprintf(w, "%08X:%04X ", addr, port) + case linux.AF_INET6: + var a linux.SockAddrInet6 + if i != nil { + a = *i.(*linux.SockAddrInet6) + } -// NeedsUpdate implements seqfile.SeqSource.NeedsUpdate. -func (*netTCP) NeedsUpdate(generation int64) bool { - return true + port := networkToHost16(a.Port) + addr0 := usermem.ByteOrder.Uint32(a.Addr[0:4]) + addr1 := usermem.ByteOrder.Uint32(a.Addr[4:8]) + addr2 := usermem.ByteOrder.Uint32(a.Addr[8:12]) + addr3 := usermem.ByteOrder.Uint32(a.Addr[12:16]) + fmt.Fprintf(w, "%08X%08X%08X%08X:%04X ", addr0, addr1, addr2, addr3, port) + } } -// ReadSeqFileData implements seqfile.SeqSource.ReadSeqFileData. -func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]seqfile.SeqData, int64) { +func commonReadSeqFileDataTCP(ctx context.Context, n seqfile.SeqHandle, k *kernel.Kernel, h seqfile.SeqHandle, fa int, header []byte) ([]seqfile.SeqData, int64) { // t may be nil here if our caller is not part of a task goroutine. This can // happen for example if we're here for "sentryctl cat". When t is nil, // degrade gracefully and retrieve what we can. @@ -358,7 +365,7 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se } var buf bytes.Buffer - for _, se := range n.k.ListSockets() { + for _, se := range k.ListSockets() { s := se.Sock.Get() if s == nil { log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) @@ -369,7 +376,7 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se if !ok { panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) } - if family, stype, _ := sops.Type(); !(family == linux.AF_INET && stype == linux.SOCK_STREAM) { + if family, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { s.DecRef() // Not tcp4 sockets. continue @@ -384,22 +391,22 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se fmt.Fprintf(&buf, "%4d: ", se.ID) // Field: local_adddress. - var localAddr linux.SockAddrInet + var localAddr linux.SockAddr if t != nil { if local, _, err := sops.GetSockName(t); err == nil { - localAddr = *local.(*linux.SockAddrInet) + localAddr = local } } - writeInetAddr(&buf, localAddr) + writeInetAddr(&buf, fa, localAddr) // Field: rem_address. - var remoteAddr linux.SockAddrInet + var remoteAddr linux.SockAddr if t != nil { if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = *remote.(*linux.SockAddrInet) + remoteAddr = remote } } - writeInetAddr(&buf, remoteAddr) + writeInetAddr(&buf, fa, remoteAddr) // Field: state; socket state. fmt.Fprintf(&buf, "%02X ", sops.State()) @@ -465,7 +472,7 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se data := []seqfile.SeqData{ { - Buf: []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n"), + Buf: header, Handle: n, }, { @@ -476,6 +483,42 @@ func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se return data, 0 } +// netTCP implements seqfile.SeqSource for /proc/net/tcp. +// +// +stateify savable +type netTCP struct { + k *kernel.Kernel +} + +// NeedsUpdate implements seqfile.SeqSource.NeedsUpdate. +func (*netTCP) NeedsUpdate(generation int64) bool { + return true +} + +// ReadSeqFileData implements seqfile.SeqSource.ReadSeqFileData. +func (n *netTCP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]seqfile.SeqData, int64) { + header := []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") + return commonReadSeqFileDataTCP(ctx, n, n.k, h, linux.AF_INET, header) +} + +// netTCP6 implements seqfile.SeqSource for /proc/net/tcp6. +// +// +stateify savable +type netTCP6 struct { + k *kernel.Kernel +} + +// NeedsUpdate implements seqfile.SeqSource.NeedsUpdate. +func (*netTCP6) NeedsUpdate(generation int64) bool { + return true +} + +// ReadSeqFileData implements seqfile.SeqSource.ReadSeqFileData. +func (n *netTCP6) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]seqfile.SeqData, int64) { + header := []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n") + return commonReadSeqFileDataTCP(ctx, n, n.k, h, linux.AF_INET6, header) +} + // netUDP implements seqfile.SeqSource for /proc/net/udp. // // +stateify savable @@ -529,7 +572,7 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se localAddr = *local.(*linux.SockAddrInet) } } - writeInetAddr(&buf, localAddr) + writeInetAddr(&buf, linux.AF_INET, &localAddr) // Field: rem_address. var remoteAddr linux.SockAddrInet @@ -538,7 +581,7 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se remoteAddr = *remote.(*linux.SockAddrInet) } } - writeInetAddr(&buf, remoteAddr) + writeInetAddr(&buf, linux.AF_INET, &remoteAddr) // Field: state; socket state. fmt.Fprintf(&buf, "%02X ", sops.State()) diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index 0ef13f2f5..56e92721e 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -230,7 +230,7 @@ func (rpf *rootProcFile) Readdir(ctx context.Context, file *fs.File, ser fs.Dent // But for whatever crazy reason, you can still walk to the given node. for _, tg := range rpf.iops.pidns.ThreadGroups() { if leader := tg.Leader(); leader != nil { - name := strconv.FormatUint(uint64(tg.ID()), 10) + name := strconv.FormatUint(uint64(rpf.iops.pidns.IDOfThreadGroup(tg)), 10) m[name] = fs.GenericDentAttr(fs.SpecialDirectory, device.ProcDevice) names = append(names, name) } diff --git a/pkg/sentry/fs/splice.go b/pkg/sentry/fs/splice.go index b03b7f836..311798811 100644 --- a/pkg/sentry/fs/splice.go +++ b/pkg/sentry/fs/splice.go @@ -139,7 +139,7 @@ func Splice(ctx context.Context, dst *File, src *File, opts SpliceOpts) (int64, // Attempt to do a WriteTo; this is likely the most efficient. n, err := src.FileOperations.WriteTo(ctx, src, w, opts.Length, opts.Dup) - if n == 0 && err != nil && err != syserror.ErrWouldBlock && !opts.Dup { + if n == 0 && err == syserror.ENOSYS && !opts.Dup { // Attempt as a ReadFrom. If a WriteTo, a ReadFrom may also be // more efficient than a copy if buffers are cached or readily // available. (It's unlikely that they can actually be donated). @@ -151,7 +151,7 @@ func Splice(ctx context.Context, dst *File, src *File, opts SpliceOpts) (int64, // if we block at some point, we could lose data. If the source is // not a pipe then reading is not destructive; if the destination // is a regular file, then it is guaranteed not to block writing. - if n == 0 && err != nil && err != syserror.ErrWouldBlock && !opts.Dup && (!dstPipe || !srcPipe) { + if n == 0 && err == syserror.ENOSYS && !opts.Dup && (!dstPipe || !srcPipe) { // Fallback to an in-kernel copy. n, err = io.Copy(w, &io.LimitedReader{ R: r, diff --git a/pkg/sentry/fs/timerfd/timerfd.go b/pkg/sentry/fs/timerfd/timerfd.go index 59403d9db..f8bf663bb 100644 --- a/pkg/sentry/fs/timerfd/timerfd.go +++ b/pkg/sentry/fs/timerfd/timerfd.go @@ -141,9 +141,10 @@ func (t *TimerOperations) Write(context.Context, *fs.File, usermem.IOSequence, i } // Notify implements ktime.TimerListener.Notify. -func (t *TimerOperations) Notify(exp uint64) { +func (t *TimerOperations) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) { atomic.AddUint64(&t.val, exp) t.events.Notify(waiter.EventIn) + return ktime.Setting{}, false } // Destroy implements ktime.TimerListener.Destroy. diff --git a/pkg/sentry/fs/tmpfs/tmpfs.go b/pkg/sentry/fs/tmpfs/tmpfs.go index 159fb7c08..69089c8a8 100644 --- a/pkg/sentry/fs/tmpfs/tmpfs.go +++ b/pkg/sentry/fs/tmpfs/tmpfs.go @@ -324,7 +324,7 @@ type Fifo struct { // NewFifo creates a new named pipe. func NewFifo(ctx context.Context, owner fs.FileOwner, perms fs.FilePermissions, msrc *fs.MountSource) *fs.Inode { // First create a pipe. - p := pipe.NewPipe(ctx, true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize) + p := pipe.NewPipe(true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize) // Build pipe InodeOperations. iops := pipe.NewInodeOperations(ctx, perms, p) diff --git a/pkg/sentry/fsimpl/memfs/directory.go b/pkg/sentry/fsimpl/memfs/directory.go index c620227c9..0bd82e480 100644 --- a/pkg/sentry/fsimpl/memfs/directory.go +++ b/pkg/sentry/fsimpl/memfs/directory.go @@ -32,7 +32,7 @@ type directory struct { childList dentryList } -func (fs *filesystem) newDirectory(creds *auth.Credentials, mode uint16) *inode { +func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *inode { dir := &directory{} dir.inode.init(dir, fs, creds, mode) dir.inode.nlink = 2 // from "." and parent directory or ".." for root diff --git a/pkg/sentry/fsimpl/memfs/memfs.go b/pkg/sentry/fsimpl/memfs/memfs.go index 45cd42b3e..b78471c0f 100644 --- a/pkg/sentry/fsimpl/memfs/memfs.go +++ b/pkg/sentry/fsimpl/memfs/memfs.go @@ -137,7 +137,7 @@ type inode struct { impl interface{} // immutable } -func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode uint16) { +func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { i.refs = 1 i.mode = uint32(mode) i.uid = uint32(creds.EffectiveKUID) diff --git a/pkg/sentry/fsimpl/memfs/regular_file.go b/pkg/sentry/fsimpl/memfs/regular_file.go index 55f869798..b7f4853b3 100644 --- a/pkg/sentry/fsimpl/memfs/regular_file.go +++ b/pkg/sentry/fsimpl/memfs/regular_file.go @@ -37,7 +37,7 @@ type regularFile struct { dataLen int64 } -func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode uint16) *inode { +func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { file := ®ularFile{} file.inode.init(file, fs, creds, mode) file.inode.nlink = 1 // from parent directory diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 8c1f79ab5..3cda03891 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -24,6 +24,7 @@ // TaskSet.mu // SignalHandlers.mu // Task.mu +// runningTasksMu // // Locking SignalHandlers.mu in multiple SignalHandlers requires locking // TaskSet.mu exclusively first. Locking Task.mu in multiple Tasks at the same @@ -135,6 +136,22 @@ type Kernel struct { // syslog is the kernel log. syslog syslog + // runningTasksMu synchronizes disable/enable of cpuClockTicker when + // the kernel is idle (runningTasks == 0). + // + // runningTasksMu is used to exclude critical sections when the timer + // disables itself and when the first active task enables the timer, + // ensuring that tasks always see a valid cpuClock value. + runningTasksMu sync.Mutex `state:"nosave"` + + // runningTasks is the total count of tasks currently in + // TaskGoroutineRunningSys or TaskGoroutineRunningApp. i.e., they are + // not blocked or stopped. + // + // runningTasks must be accessed atomically. Increments from 0 to 1 are + // further protected by runningTasksMu (see incRunningTasks). + runningTasks int64 + // cpuClock is incremented every linux.ClockTick. cpuClock is used to // measure task CPU usage, since sampling monotonicClock twice on every // syscall turns out to be unreasonably expensive. This is similar to how @@ -150,6 +167,22 @@ type Kernel struct { // cpuClockTicker increments cpuClock. cpuClockTicker *ktime.Timer `state:"nosave"` + // cpuClockTickerDisabled indicates that cpuClockTicker has been + // disabled because no tasks are running. + // + // cpuClockTickerDisabled is protected by runningTasksMu. + cpuClockTickerDisabled bool + + // cpuClockTickerSetting is the ktime.Setting of cpuClockTicker at the + // point it was disabled. It is cached here to avoid a lock ordering + // violation with cpuClockTicker.mu when runningTaskMu is held. + // + // cpuClockTickerSetting is only valid when cpuClockTickerDisabled is + // true. + // + // cpuClockTickerSetting is protected by runningTasksMu. + cpuClockTickerSetting ktime.Setting + // fdMapUids is an ever-increasing counter for generating FDTable uids. // // fdMapUids is mutable, and is accessed using atomic memory operations. @@ -912,6 +945,102 @@ func (k *Kernel) resumeTimeLocked() { } } +func (k *Kernel) incRunningTasks() { + for { + tasks := atomic.LoadInt64(&k.runningTasks) + if tasks != 0 { + // Standard case. Simply increment. + if !atomic.CompareAndSwapInt64(&k.runningTasks, tasks, tasks+1) { + continue + } + return + } + + // Transition from 0 -> 1. Synchronize with other transitions and timer. + k.runningTasksMu.Lock() + tasks = atomic.LoadInt64(&k.runningTasks) + if tasks != 0 { + // We're no longer the first task, no need to + // re-enable. + atomic.AddInt64(&k.runningTasks, 1) + k.runningTasksMu.Unlock() + return + } + + if !k.cpuClockTickerDisabled { + // Timer was never disabled. + atomic.StoreInt64(&k.runningTasks, 1) + k.runningTasksMu.Unlock() + return + } + + // We need to update cpuClock for all of the ticks missed while we + // slept, and then re-enable the timer. + // + // The Notify in Swap isn't sufficient. kernelCPUClockTicker.Notify + // always increments cpuClock by 1 regardless of the number of + // expirations as a heuristic to avoid over-accounting in cases of CPU + // throttling. + // + // We want to cover the normal case, when all time should be accounted, + // so we increment for all expirations. Throttling is less concerning + // here because the ticker is only disabled from Notify. This means + // that Notify must schedule and compensate for the throttled period + // before the timer is disabled. Throttling while the timer is disabled + // doesn't matter, as nothing is running or reading cpuClock anyways. + // + // S/R also adds complication, as there are two cases. Recall that + // monotonicClock will jump forward on restore. + // + // 1. If the ticker is enabled during save, then on Restore Notify is + // called with many expirations, covering the time jump, but cpuClock + // is only incremented by 1. + // + // 2. If the ticker is disabled during save, then after Restore the + // first wakeup will call this function and cpuClock will be + // incremented by the number of expirations across the S/R. + // + // These cause very different value of cpuClock. But again, since + // nothing was running while the ticker was disabled, those differences + // don't matter. + setting, exp := k.cpuClockTickerSetting.At(k.monotonicClock.Now()) + if exp > 0 { + atomic.AddUint64(&k.cpuClock, exp) + } + + // Now that cpuClock is updated it is safe to allow other tasks to + // transition to running. + atomic.StoreInt64(&k.runningTasks, 1) + + // N.B. we must unlock before calling Swap to maintain lock ordering. + // + // cpuClockTickerDisabled need not wait until after Swap to become + // true. It is sufficient that the timer *will* be enabled. + k.cpuClockTickerDisabled = false + k.runningTasksMu.Unlock() + + // This won't call Notify (unless it's been ClockTick since setting.At + // above). This means we skip the thread group work in Notify. However, + // since nothing was running while we were disabled, none of the timers + // could have expired. + k.cpuClockTicker.Swap(setting) + + return + } +} + +func (k *Kernel) decRunningTasks() { + tasks := atomic.AddInt64(&k.runningTasks, -1) + if tasks < 0 { + panic(fmt.Sprintf("Invalid running count %d", tasks)) + } + + // Nothing to do. The next CPU clock tick will disable the timer if + // there is still nothing running. This provides approximately one tick + // of slack in which we can switch back and forth between idle and + // active without an expensive transition. +} + // WaitExited blocks until all tasks in k have exited. func (k *Kernel) WaitExited() { k.tasks.liveGoroutines.Wait() diff --git a/pkg/sentry/kernel/pipe/node_test.go b/pkg/sentry/kernel/pipe/node_test.go index adbad7764..16fa80abe 100644 --- a/pkg/sentry/kernel/pipe/node_test.go +++ b/pkg/sentry/kernel/pipe/node_test.go @@ -85,11 +85,11 @@ func testOpen(ctx context.Context, t *testing.T, n fs.InodeOperations, flags fs. } func newNamedPipe(t *testing.T) *Pipe { - return NewPipe(contexttest.Context(t), true, DefaultPipeSize, usermem.PageSize) + return NewPipe(true, DefaultPipeSize, usermem.PageSize) } func newAnonPipe(t *testing.T) *Pipe { - return NewPipe(contexttest.Context(t), false, DefaultPipeSize, usermem.PageSize) + return NewPipe(false, DefaultPipeSize, usermem.PageSize) } // assertRecvBlocks ensures that a recv attempt on c blocks for at least diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 93b50669f..8e4e8e82e 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -98,7 +98,7 @@ type Pipe struct { // NewPipe initializes and returns a pipe. // // N.B. The size and atomicIOBytes will be bounded. -func NewPipe(ctx context.Context, isNamed bool, sizeBytes, atomicIOBytes int64) *Pipe { +func NewPipe(isNamed bool, sizeBytes, atomicIOBytes int64) *Pipe { if sizeBytes < MinimumPipeSize { sizeBytes = MinimumPipeSize } @@ -121,7 +121,7 @@ func NewPipe(ctx context.Context, isNamed bool, sizeBytes, atomicIOBytes int64) // NewConnectedPipe initializes a pipe and returns a pair of objects // representing the read and write ends of the pipe. func NewConnectedPipe(ctx context.Context, sizeBytes, atomicIOBytes int64) (*fs.File, *fs.File) { - p := NewPipe(ctx, false /* isNamed */, sizeBytes, atomicIOBytes) + p := NewPipe(false /* isNamed */, sizeBytes, atomicIOBytes) // Build an fs.Dirent for the pipe which will be shared by both // returned files. diff --git a/pkg/sentry/kernel/posixtimer.go b/pkg/sentry/kernel/posixtimer.go index c5d095af7..2e861a5a8 100644 --- a/pkg/sentry/kernel/posixtimer.go +++ b/pkg/sentry/kernel/posixtimer.go @@ -117,9 +117,9 @@ func (it *IntervalTimer) signalRejectedLocked() { } // Notify implements ktime.TimerListener.Notify. -func (it *IntervalTimer) Notify(exp uint64) { +func (it *IntervalTimer) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) { if it.target == nil { - return + return ktime.Setting{}, false } it.target.tg.pidns.owner.mu.RLock() @@ -129,7 +129,7 @@ func (it *IntervalTimer) Notify(exp uint64) { if it.sigpending { it.overrunCur += exp - return + return ktime.Setting{}, false } // sigpending must be set before sendSignalTimerLocked() so that it can be @@ -148,6 +148,8 @@ func (it *IntervalTimer) Notify(exp uint64) { if err := it.target.sendSignalTimerLocked(si, it.group, it); err != nil { it.signalRejectedLocked() } + + return ktime.Setting{}, false } // Destroy implements ktime.TimerListener.Destroy. Users of Timer should call diff --git a/pkg/sentry/kernel/task_identity.go b/pkg/sentry/kernel/task_identity.go index 78ff14b20..ce3e6ef28 100644 --- a/pkg/sentry/kernel/task_identity.go +++ b/pkg/sentry/kernel/task_identity.go @@ -465,8 +465,8 @@ func (t *Task) SetKeepCaps(k bool) { // disables the features we don't support anyway, is always set. This // drastically simplifies this function. // -// - We don't implement AT_SECURE, because no_new_privs always being set means -// that the conditions that require AT_SECURE never arise. (Compare Linux's +// - We don't set AT_SECURE = 1, because no_new_privs always being set means +// that the conditions that require AT_SECURE = 1 never arise. (Compare Linux's // security/commoncap.c:cap_bprm_set_creds() and cap_bprm_secureexec().) // // - We don't check for CAP_SYS_ADMIN in prctl(PR_SET_SECCOMP), since diff --git a/pkg/sentry/kernel/task_sched.go b/pkg/sentry/kernel/task_sched.go index e76c069b0..8b148db35 100644 --- a/pkg/sentry/kernel/task_sched.go +++ b/pkg/sentry/kernel/task_sched.go @@ -126,12 +126,22 @@ func (t *Task) accountTaskGoroutineEnter(state TaskGoroutineState) { t.gosched.Timestamp = now t.gosched.State = state t.goschedSeq.EndWrite() + + if state != TaskGoroutineRunningApp { + // Task is blocking/stopping. + t.k.decRunningTasks() + } } // Preconditions: The caller must be running on the task goroutine, and leaving // a state indicated by a previous call to // t.accountTaskGoroutineEnter(state). func (t *Task) accountTaskGoroutineLeave(state TaskGoroutineState) { + if state != TaskGoroutineRunningApp { + // Task is unblocking/continuing. + t.k.incRunningTasks() + } + now := t.k.CPUClockNow() if t.gosched.State != state { panic(fmt.Sprintf("Task goroutine switching from state %v (expected %v) to %v", t.gosched.State, state, TaskGoroutineRunningSys)) @@ -330,7 +340,7 @@ func newKernelCPUClockTicker(k *Kernel) *kernelCPUClockTicker { } // Notify implements ktime.TimerListener.Notify. -func (ticker *kernelCPUClockTicker) Notify(exp uint64) { +func (ticker *kernelCPUClockTicker) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) { // Only increment cpuClock by 1 regardless of the number of expirations. // This approximately compensates for cases where thread throttling or bad // Go runtime scheduling prevents the kernelCPUClockTicker goroutine, and @@ -426,6 +436,27 @@ func (ticker *kernelCPUClockTicker) Notify(exp uint64) { tgs[i] = nil } ticker.tgs = tgs[:0] + + // If nothing is running, we can disable the timer. + tasks := atomic.LoadInt64(&ticker.k.runningTasks) + if tasks == 0 { + ticker.k.runningTasksMu.Lock() + defer ticker.k.runningTasksMu.Unlock() + tasks := atomic.LoadInt64(&ticker.k.runningTasks) + if tasks != 0 { + // Raced with a 0 -> 1 transition. + return setting, false + } + + // Stop the timer. We must cache the current setting so the + // kernel can access it without violating the lock order. + ticker.k.cpuClockTickerSetting = setting + ticker.k.cpuClockTickerDisabled = true + setting.Enabled = false + return setting, true + } + + return setting, false } // Destroy implements ktime.TimerListener.Destroy. diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 0eef24bfb..72568d296 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -511,8 +511,9 @@ type itimerRealListener struct { } // Notify implements ktime.TimerListener.Notify. -func (l *itimerRealListener) Notify(exp uint64) { +func (l *itimerRealListener) Notify(exp uint64, setting ktime.Setting) (ktime.Setting, bool) { l.tg.SendSignal(SignalInfoPriv(linux.SIGALRM)) + return ktime.Setting{}, false } // Destroy implements ktime.TimerListener.Destroy. diff --git a/pkg/sentry/kernel/time/time.go b/pkg/sentry/kernel/time/time.go index aa6c75d25..107394183 100644 --- a/pkg/sentry/kernel/time/time.go +++ b/pkg/sentry/kernel/time/time.go @@ -280,13 +280,16 @@ func (ClockEventsQueue) Readiness(mask waiter.EventMask) waiter.EventMask { // A TimerListener receives expirations from a Timer. type TimerListener interface { // Notify is called when its associated Timer expires. exp is the number of - // expirations. + // expirations. setting is the next timer Setting. // // Notify is called with the associated Timer's mutex locked, so Notify // must not take any locks that precede Timer.mu in lock order. // + // If Notify returns true, the timer will use the returned setting + // rather than the passed one. + // // Preconditions: exp > 0. - Notify(exp uint64) + Notify(exp uint64, setting Setting) (newSetting Setting, update bool) // Destroy is called when the timer is destroyed. Destroy() @@ -533,7 +536,9 @@ func (t *Timer) Tick() { s, exp := t.setting.At(now) t.setting = s if exp > 0 { - t.listener.Notify(exp) + if newS, ok := t.listener.Notify(exp, t.setting); ok { + t.setting = newS + } } t.resetKickerLocked(now) } @@ -588,7 +593,9 @@ func (t *Timer) Get() (Time, Setting) { s, exp := t.setting.At(now) t.setting = s if exp > 0 { - t.listener.Notify(exp) + if newS, ok := t.listener.Notify(exp, t.setting); ok { + t.setting = newS + } } t.resetKickerLocked(now) return now, s @@ -620,7 +627,9 @@ func (t *Timer) SwapAnd(s Setting, f func()) (Time, Setting) { } oldS, oldExp := t.setting.At(now) if oldExp > 0 { - t.listener.Notify(oldExp) + t.listener.Notify(oldExp, oldS) + // N.B. The returned Setting doesn't matter because we're about + // to overwrite. } if f != nil { f() @@ -628,7 +637,9 @@ func (t *Timer) SwapAnd(s Setting, f func()) (Time, Setting) { newS, newExp := s.At(now) t.setting = newS if newExp > 0 { - t.listener.Notify(newExp) + if newS, ok := t.listener.Notify(newExp, t.setting); ok { + t.setting = newS + } } t.resetKickerLocked(now) return now, oldS @@ -683,11 +694,13 @@ func NewChannelNotifier() (TimerListener, <-chan struct{}) { } // Notify implements ktime.TimerListener.Notify. -func (c *ChannelNotifier) Notify(uint64) { +func (c *ChannelNotifier) Notify(uint64, Setting) (Setting, bool) { select { case c.tchan <- struct{}{}: default: } + + return Setting{}, false } // Destroy implements ktime.TimerListener.Destroy and will close the channel. diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index ba9c9ce12..2d9251e92 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -323,18 +323,22 @@ func mapSegment(ctx context.Context, m *mm.MemoryManager, f *fs.File, phdr *elf. return syserror.ENOEXEC } + // N.B. Linux uses vm_brk_flags to map these pages, which only + // honors the X bit, always mapping at least RW. ignoring These + // pages are not included in the final brk region. + prot := usermem.ReadWrite + if phdr.Flags&elf.PF_X == elf.PF_X { + prot.Execute = true + } + if _, err := m.MMap(ctx, memmap.MMapOpts{ Length: uint64(anonSize), Addr: anonAddr, // Fixed without Unmap will fail the mmap if something is // already at addr. - Fixed: true, - Private: true, - // N.B. Linux uses vm_brk to map these pages, ignoring - // the segment protections, instead always mapping RW. - // These pages are not included in the final brk - // region. - Perms: usermem.ReadWrite, + Fixed: true, + Private: true, + Perms: prot, MaxPerms: usermem.AnyAccess, }); err != nil { ctx.Infof("Error mapping PT_LOAD segment %v anonymous memory: %v", phdr, err) diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index f6f1ae762..089d1635b 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -308,6 +308,9 @@ func Load(ctx context.Context, m *mm.MemoryManager, mounts *fs.MountNamespace, r arch.AuxEntry{linux.AT_EUID, usermem.Addr(c.EffectiveKUID.In(c.UserNamespace).OrOverflow())}, arch.AuxEntry{linux.AT_GID, usermem.Addr(c.RealKGID.In(c.UserNamespace).OrOverflow())}, arch.AuxEntry{linux.AT_EGID, usermem.Addr(c.EffectiveKGID.In(c.UserNamespace).OrOverflow())}, + // The conditions that require AT_SECURE = 1 never arise. See + // kernel.Task.updateCredsForExecLocked. + arch.AuxEntry{linux.AT_SECURE, 0}, arch.AuxEntry{linux.AT_CLKTCK, linux.CLOCKS_PER_SEC}, arch.AuxEntry{linux.AT_EXECFN, execfn}, arch.AuxEntry{linux.AT_RANDOM, random}, diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 4f8f9c5d9..9f0ecfbe4 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -267,7 +267,7 @@ func (s *subprocess) newThread() *thread { // attach attaches to the thread. func (t *thread) attach() { - if _, _, errno := syscall.RawSyscall(syscall.SYS_PTRACE, syscall.PTRACE_ATTACH, uintptr(t.tid), 0); errno != 0 { + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_ATTACH, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("unable to attach: %v", errno)) } @@ -417,7 +417,7 @@ func (t *thread) syscall(regs *syscall.PtraceRegs) (uintptr, error) { for { // Execute the syscall instruction. - if _, _, errno := syscall.RawSyscall(syscall.SYS_PTRACE, syscall.PTRACE_SYSCALL, uintptr(t.tid), 0); errno != 0 { + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_SYSCALL, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) } @@ -435,7 +435,7 @@ func (t *thread) syscall(regs *syscall.PtraceRegs) (uintptr, error) { } // Complete the actual system call. - if _, _, errno := syscall.RawSyscall(syscall.SYS_PTRACE, syscall.PTRACE_SYSCALL, uintptr(t.tid), 0); errno != 0 { + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_SYSCALL, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) } @@ -526,17 +526,17 @@ func (s *subprocess) switchToApp(c *context, ac arch.Context) bool { for { // Start running until the next system call. if isSingleStepping(regs) { - if _, _, errno := syscall.RawSyscall( + if _, _, errno := syscall.RawSyscall6( syscall.SYS_PTRACE, syscall.PTRACE_SYSEMU_SINGLESTEP, - uintptr(t.tid), 0); errno != 0 { + uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace sysemu failed: %v", errno)) } } else { - if _, _, errno := syscall.RawSyscall( + if _, _, errno := syscall.RawSyscall6( syscall.SYS_PTRACE, syscall.PTRACE_SYSEMU, - uintptr(t.tid), 0); errno != 0 { + uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace sysemu failed: %v", errno)) } } diff --git a/pkg/sentry/platform/ptrace/subprocess_linux.go b/pkg/sentry/platform/ptrace/subprocess_linux.go index f09b0b3d0..c075b5f91 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux.go @@ -53,7 +53,7 @@ func probeSeccomp() bool { for { // Attempt an emulation. - if _, _, errno := syscall.RawSyscall(syscall.SYS_PTRACE, syscall.PTRACE_SYSEMU, uintptr(t.tid), 0); errno != 0 { + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_SYSEMU, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) } @@ -266,7 +266,7 @@ func attachedThread(flags uintptr, defaultAction linux.BPFAction) (*thread, erro // Enable cpuid-faulting; this may fail on older kernels or hardware, // so we just disregard the result. Host CPUID will be enabled. - syscall.RawSyscall(syscall.SYS_ARCH_PRCTL, linux.ARCH_SET_CPUID, 0, 0) + syscall.RawSyscall6(syscall.SYS_ARCH_PRCTL, linux.ARCH_SET_CPUID, 0, 0, 0, 0, 0) // Call the stub; should not return. stubCall(stubStart, ppid) diff --git a/pkg/sentry/sighandling/sighandling_unsafe.go b/pkg/sentry/sighandling/sighandling_unsafe.go index eace3766d..c303435d5 100644 --- a/pkg/sentry/sighandling/sighandling_unsafe.go +++ b/pkg/sentry/sighandling/sighandling_unsafe.go @@ -23,7 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" ) -// TODO(b/34161764): Move to pkg/abi/linux along with definitions in +// FIXME(gvisor.dev/issue/214): Move to pkg/abi/linux along with definitions in // pkg/sentry/arch. type sigaction struct { handler uintptr diff --git a/pkg/sentry/socket/epsocket/BUILD b/pkg/sentry/socket/netstack/BUILD index e927821e1..60523f79a 100644 --- a/pkg/sentry/socket/epsocket/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -3,15 +3,15 @@ package(licenses = ["notice"]) load("//tools/go_stateify:defs.bzl", "go_library") go_library( - name = "epsocket", + name = "netstack", srcs = [ "device.go", - "epsocket.go", + "netstack.go", "provider.go", "save_restore.go", "stack.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/epsocket", + importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netstack", visibility = [ "//pkg/sentry:internal", ], diff --git a/pkg/sentry/socket/epsocket/device.go b/pkg/sentry/socket/netstack/device.go index 85484d5b1..fbeb89fb8 100644 --- a/pkg/sentry/socket/epsocket/device.go +++ b/pkg/sentry/socket/netstack/device.go @@ -12,9 +12,9 @@ // See the License for the specific language governing permissions and // limitations under the License. -package epsocket +package netstack import "gvisor.dev/gvisor/pkg/sentry/device" -// epsocketDevice is the endpoint socket virtual device. -var epsocketDevice = device.NewAnonDevice() +// netstackDevice is the endpoint socket virtual device. +var netstackDevice = device.NewAnonDevice() diff --git a/pkg/sentry/socket/epsocket/epsocket.go b/pkg/sentry/socket/netstack/netstack.go index 3e66f9cbb..b3546471e 100644 --- a/pkg/sentry/socket/epsocket/epsocket.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package epsocket provides an implementation of the socket.Socket interface +// Package netstack provides an implementation of the socket.Socket interface // that is backed by a tcpip.Endpoint. // // It does not depend on any particular endpoint implementation, and thus can @@ -22,7 +22,7 @@ // Lock ordering: netstack => mm: ioSequencePayload copies user memory inside // tcpip.Endpoint.Write(). Netstack is allowed to (and does) hold locks during // this operation. -package epsocket +package netstack import ( "bytes" @@ -174,9 +174,13 @@ var Metrics = tcpip.Stats{ }, } +// DefaultTTL is linux's default TTL. All network protocols in all stacks used +// with this package must have this value set as their default TTL. +const DefaultTTL = 64 + const sizeOfInt32 int = 4 -var errStackType = syserr.New("expected but did not receive an epsocket.Stack", linux.EINVAL) +var errStackType = syserr.New("expected but did not receive a netstack.Stack", linux.EINVAL) // ntohs converts a 16-bit number from network byte order to host byte order. It // assumes that the host is little endian. @@ -262,8 +266,8 @@ type SocketOperations struct { // valid when timestampValid is true. It is protected by readMu. timestampNS int64 - // sockOptInq corresponds to TCP_INQ. It is implemented on the epsocket - // level, because it takes into account data from readView. + // sockOptInq corresponds to TCP_INQ. It is implemented at this level + // because it takes into account data from readView. sockOptInq bool } @@ -275,7 +279,7 @@ func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue } } - dirent := socket.NewDirent(t, epsocketDevice) + dirent := socket.NewDirent(t, netstackDevice) defer dirent.DecRef() return fs.NewFile(t, dirent, fs.FileFlags{Read: true, Write: true, NonSeekable: true}, &SocketOperations{ Queue: queue, @@ -760,7 +764,7 @@ func (s *SocketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error { // tcpip.Endpoint. func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { // TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is - // implemented specifically for epsocket.SocketOperations rather than + // implemented specifically for netstack.SocketOperations rather than // commonEndpoint. commonEndpoint should be extended to support socket // options where the implementation is not shared, as unix sockets need // their own support for SO_TIMESTAMP. @@ -833,7 +837,7 @@ func GetSockOpt(t *kernel.Task, s socket.Socket, ep commonEndpoint, family int, return getSockOptIPv6(t, ep, name, outLen) case linux.SOL_IP: - return getSockOptIP(t, ep, name, outLen) + return getSockOptIP(t, ep, name, outLen, family) case linux.SOL_UDP, linux.SOL_ICMPV6, @@ -942,6 +946,19 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family return int32(v), nil + case linux.SO_BINDTODEVICE: + var v tcpip.BindToDeviceOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + if len(v) == 0 { + return []byte{}, nil + } + if outLen < linux.IFNAMSIZ { + return nil, syserr.ErrInvalidArgument + } + return append([]byte(v), 0), nil + case linux.SO_BROADCAST: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument @@ -1163,8 +1180,25 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf } // getSockOptIP implements GetSockOpt when level is SOL_IP. -func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interface{}, *syserr.Error) { +func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family int) (interface{}, *syserr.Error) { switch name { + case linux.IP_TTL: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.TTLOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + // Fill in the default value, if needed. + if v == 0 { + v = DefaultTTL + } + + return int32(v), nil + case linux.IP_MULTICAST_TTL: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument @@ -1216,7 +1250,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfac // tcpip.Endpoint. func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVal []byte) *syserr.Error { // TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is - // implemented specifically for epsocket.SocketOperations rather than + // implemented specifically for netstack.SocketOperations rather than // commonEndpoint. commonEndpoint should be extended to support socket // options where the implementation is not shared, as unix sockets need // their own support for SO_TIMESTAMP. @@ -1305,6 +1339,13 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i v := usermem.ByteOrder.Uint32(optVal) return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.ReusePortOption(v))) + case linux.SO_BINDTODEVICE: + n := bytes.IndexByte(optVal, 0) + if n == -1 { + n = len(optVal) + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(optVal[:n]))) + case linux.SO_BROADCAST: if len(optVal) < sizeOfInt32 { return syserr.ErrInvalidArgument @@ -1628,6 +1669,20 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s t.Kernel().EmitUnimplementedEvent(t) return syserr.ErrInvalidArgument + case linux.IP_TTL: + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + + // -1 means default TTL. + if v == -1 { + v = 0 + } else if v < 1 || v > 255 { + return syserr.ErrInvalidArgument + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TTLOption(v))) + case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, @@ -1653,7 +1708,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s linux.IP_RETOPTS, linux.IP_TOS, linux.IP_TRANSPARENT, - linux.IP_TTL, linux.IP_UNBLOCK_SOURCE, linux.IP_UNICAST_IF, linux.IP_XFRM_POLICY, @@ -2215,7 +2269,7 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] // Ioctl implements fs.FileOperations.Ioctl. func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { - // SIOCGSTAMP is implemented by epsocket rather than all commonEndpoint + // SIOCGSTAMP is implemented by netstack rather than all commonEndpoint // sockets. // TODO(b/78348848): Add a commonEndpoint method to support SIOCGSTAMP. switch args[1].Int() { @@ -2518,7 +2572,7 @@ func ifconfIoctl(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error { // Flag values and meanings are described in greater detail in netdevice(7) in // the SIOCGIFFLAGS section. func interfaceStatusFlags(stack inet.Stack, name string) (uint32, *syserr.Error) { - // epsocket should only ever be passed an epsocket.Stack. + // We should only ever be passed a netstack.Stack. epstack, ok := stack.(*Stack) if !ok { return 0, errStackType diff --git a/pkg/sentry/socket/epsocket/provider.go b/pkg/sentry/socket/netstack/provider.go index 421f93dc4..357a664cc 100644 --- a/pkg/sentry/socket/epsocket/provider.go +++ b/pkg/sentry/socket/netstack/provider.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -package epsocket +package netstack import ( "syscall" @@ -65,7 +65,7 @@ func getTransportProtocol(ctx context.Context, stype linux.SockType, protocol in // Raw sockets require CAP_NET_RAW. creds := auth.CredentialsFromContext(ctx) if !creds.HasCapability(linux.CAP_NET_RAW) { - return 0, true, syserr.ErrPermissionDenied + return 0, true, syserr.ErrNotPermitted } switch protocol { diff --git a/pkg/sentry/socket/epsocket/save_restore.go b/pkg/sentry/socket/netstack/save_restore.go index f7b8c10cc..c7aaf722a 100644 --- a/pkg/sentry/socket/epsocket/save_restore.go +++ b/pkg/sentry/socket/netstack/save_restore.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -package epsocket +package netstack import ( "gvisor.dev/gvisor/pkg/tcpip/stack" diff --git a/pkg/sentry/socket/epsocket/stack.go b/pkg/sentry/socket/netstack/stack.go index 7cf7ff735..fda0156e5 100644 --- a/pkg/sentry/socket/epsocket/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -package epsocket +package netstack import ( "gvisor.dev/gvisor/pkg/abi/linux" diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index da9977fde..830f4da10 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -24,7 +24,7 @@ go_library( "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/control", - "//pkg/sentry/socket/epsocket", + "//pkg/sentry/socket/netstack", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usermem", "//pkg/syserr", diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 0d0cb68df..50c308134 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -31,7 +31,7 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/control" - "gvisor.dev/gvisor/pkg/sentry/socket/epsocket" + "gvisor.dev/gvisor/pkg/sentry/socket/netstack" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" @@ -40,8 +40,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -// SocketOperations is a Unix socket. It is similar to an epsocket, except it -// is backed by a transport.Endpoint instead of a tcpip.Endpoint. +// SocketOperations is a Unix socket. It is similar to a netstack socket, +// except it is backed by a transport.Endpoint instead of a tcpip.Endpoint. // // +stateify savable type SocketOperations struct { @@ -116,7 +116,7 @@ func (s *SocketOperations) Endpoint() transport.Endpoint { // extractPath extracts and validates the address. func extractPath(sockaddr []byte) (string, *syserr.Error) { - addr, _, err := epsocket.AddressAndFamily(linux.AF_UNIX, sockaddr, true /* strict */) + addr, _, err := netstack.AddressAndFamily(linux.AF_UNIX, sockaddr, true /* strict */) if err != nil { return "", err } @@ -143,7 +143,7 @@ func (s *SocketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, return nil, 0, syserr.TranslateNetstackError(err) } - a, l := epsocket.ConvertAddress(linux.AF_UNIX, addr) + a, l := netstack.ConvertAddress(linux.AF_UNIX, addr) return a, l, nil } @@ -155,19 +155,19 @@ func (s *SocketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, return nil, 0, syserr.TranslateNetstackError(err) } - a, l := epsocket.ConvertAddress(linux.AF_UNIX, addr) + a, l := netstack.ConvertAddress(linux.AF_UNIX, addr) return a, l, nil } // Ioctl implements fs.FileOperations.Ioctl. func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { - return epsocket.Ioctl(ctx, s.ep, io, args) + return netstack.Ioctl(ctx, s.ep, io, args) } // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // a transport.Endpoint. func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { - return epsocket.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen) + return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen) } // Listen implements the linux syscall listen(2) for sockets backed by @@ -474,13 +474,13 @@ func (s *SocketOperations) EventUnregister(e *waiter.Entry) { // SetSockOpt implements the linux syscall setsockopt(2) for sockets backed by // a transport.Endpoint. func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVal []byte) *syserr.Error { - return epsocket.SetSockOpt(t, s, s.ep, level, name, optVal) + return netstack.SetSockOpt(t, s, s.ep, level, name, optVal) } // Shutdown implements the linux syscall shutdown(2) for sockets backed by // a transport.Endpoint. func (s *SocketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error { - f, err := epsocket.ConvertShutdown(how) + f, err := netstack.ConvertShutdown(how) if err != nil { return err } @@ -546,7 +546,7 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags var from linux.SockAddr var fromLen uint32 if r.From != nil && len([]byte(r.From.Addr)) != 0 { - from, fromLen = epsocket.ConvertAddress(linux.AF_UNIX, *r.From) + from, fromLen = netstack.ConvertAddress(linux.AF_UNIX, *r.From) } if r.ControlTrunc { @@ -581,7 +581,7 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags var from linux.SockAddr var fromLen uint32 if r.From != nil { - from, fromLen = epsocket.ConvertAddress(linux.AF_UNIX, *r.From) + from, fromLen = netstack.ConvertAddress(linux.AF_UNIX, *r.From) } if r.ControlTrunc { diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index 7d7b42eba..72ebf766d 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -32,8 +32,8 @@ go_library( "//pkg/sentry/arch", "//pkg/sentry/kernel", "//pkg/sentry/socket/control", - "//pkg/sentry/socket/epsocket", "//pkg/sentry/socket/netlink", + "//pkg/sentry/socket/netstack", "//pkg/sentry/syscalls/linux", "//pkg/sentry/usermem", ], diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index f779186ad..94334f6d2 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -23,8 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket/control" - "gvisor.dev/gvisor/pkg/sentry/socket/epsocket" "gvisor.dev/gvisor/pkg/sentry/socket/netlink" + "gvisor.dev/gvisor/pkg/sentry/socket/netstack" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/usermem" ) @@ -332,7 +332,7 @@ func sockAddr(t *kernel.Task, addr usermem.Addr, length uint32) string { switch family { case linux.AF_INET, linux.AF_INET6, linux.AF_UNIX: - fa, _, err := epsocket.AddressAndFamily(int(family), b, true /* strict */) + fa, _, err := netstack.AddressAndFamily(int(family), b, true /* strict */) if err != nil { return fmt.Sprintf("%#x {Family: %s, error extracting address: %v}", addr, familyStr, err) } diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index e76ee27d2..cf2a56bed 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -8,6 +8,8 @@ go_library( "error.go", "flags.go", "linux64.go", + "linux64_amd64.go", + "linux64_arm64.go", "sigset.go", "sys_aio.go", "sys_capability.go", diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index 18d24ab61..b64c49ff5 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -1,4 +1,4 @@ -// Copyright 2018 The gVisor Authors. +// 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. @@ -15,377 +15,8 @@ // Package linux provides syscall tables for amd64 Linux. package linux -import ( - "gvisor.dev/gvisor/pkg/abi" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/syscalls" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserror" +const ( + _LINUX_SYSNAME = "Linux" + _LINUX_RELEASE = "4.4" + _LINUX_VERSION = "#1 SMP Sun Jan 10 15:06:54 PST 2016" ) - -// AUDIT_ARCH_X86_64 identifies the Linux syscall API on AMD64, and is taken -// from <linux/audit.h>. -const _AUDIT_ARCH_X86_64 = 0xc000003e - -// AMD64 is a table of Linux amd64 syscall API with the corresponding syscall -// numbers from Linux 4.4. -var AMD64 = &kernel.SyscallTable{ - OS: abi.Linux, - Arch: arch.AMD64, - Version: kernel.Version{ - // Version 4.4 is chosen as a stable, longterm version of Linux, which - // 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", - Release: "4.4", - Version: "#1 SMP Sun Jan 10 15:06:54 PST 2016", - }, - AuditNumber: _AUDIT_ARCH_X86_64, - Table: map[uintptr]kernel.Syscall{ - 0: syscalls.Supported("read", Read), - 1: syscalls.Supported("write", Write), - 2: syscalls.PartiallySupported("open", Open, "Options O_DIRECT, O_NOATIME, O_PATH, O_TMPFILE, O_SYNC are not supported.", nil), - 3: syscalls.Supported("close", Close), - 4: syscalls.Supported("stat", Stat), - 5: syscalls.Supported("fstat", Fstat), - 6: syscalls.Supported("lstat", Lstat), - 7: syscalls.Supported("poll", Poll), - 8: syscalls.Supported("lseek", Lseek), - 9: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), - 10: syscalls.Supported("mprotect", Mprotect), - 11: syscalls.Supported("munmap", Munmap), - 12: syscalls.Supported("brk", Brk), - 13: syscalls.Supported("rt_sigaction", RtSigaction), - 14: syscalls.Supported("rt_sigprocmask", RtSigprocmask), - 15: syscalls.Supported("rt_sigreturn", RtSigreturn), - 16: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), - 17: syscalls.Supported("pread64", Pread64), - 18: syscalls.Supported("pwrite64", Pwrite64), - 19: syscalls.Supported("readv", Readv), - 20: syscalls.Supported("writev", Writev), - 21: syscalls.Supported("access", Access), - 22: syscalls.Supported("pipe", Pipe), - 23: syscalls.Supported("select", Select), - 24: syscalls.Supported("sched_yield", SchedYield), - 25: syscalls.Supported("mremap", Mremap), - 26: syscalls.PartiallySupported("msync", Msync, "Full data flush is not guaranteed at this time.", nil), - 27: syscalls.PartiallySupported("mincore", Mincore, "Stub implementation. The sandbox does not have access to this information. Reports all mapped pages are resident.", nil), - 28: syscalls.PartiallySupported("madvise", Madvise, "Options MADV_DONTNEED, MADV_DONTFORK are supported. Other advice is ignored.", nil), - 29: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), - 30: syscalls.PartiallySupported("shmat", Shmat, "Option SHM_RND is not supported.", nil), - 31: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), - 32: syscalls.Supported("dup", Dup), - 33: syscalls.Supported("dup2", Dup2), - 34: syscalls.Supported("pause", Pause), - 35: syscalls.Supported("nanosleep", Nanosleep), - 36: syscalls.Supported("getitimer", Getitimer), - 37: syscalls.Supported("alarm", Alarm), - 38: syscalls.Supported("setitimer", Setitimer), - 39: syscalls.Supported("getpid", Getpid), - 40: syscalls.Supported("sendfile", Sendfile), - 41: syscalls.PartiallySupported("socket", Socket, "Limited support for AF_NETLINK, NETLINK_ROUTE sockets. Limited support for SOCK_RAW.", nil), - 42: syscalls.Supported("connect", Connect), - 43: syscalls.Supported("accept", Accept), - 44: syscalls.Supported("sendto", SendTo), - 45: syscalls.Supported("recvfrom", RecvFrom), - 46: syscalls.Supported("sendmsg", SendMsg), - 47: syscalls.PartiallySupported("recvmsg", RecvMsg, "Not all flags and control messages are supported.", nil), - 48: syscalls.PartiallySupported("shutdown", Shutdown, "Not all flags and control messages are supported.", nil), - 49: syscalls.PartiallySupported("bind", Bind, "Autobind for abstract Unix sockets is not supported.", nil), - 50: syscalls.Supported("listen", Listen), - 51: syscalls.Supported("getsockname", GetSockName), - 52: syscalls.Supported("getpeername", GetPeerName), - 53: syscalls.Supported("socketpair", SocketPair), - 54: syscalls.PartiallySupported("setsockopt", SetSockOpt, "Not all socket options are supported.", nil), - 55: syscalls.PartiallySupported("getsockopt", GetSockOpt, "Not all socket options are supported.", nil), - 56: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), - 57: syscalls.Supported("fork", Fork), - 58: syscalls.Supported("vfork", Vfork), - 59: syscalls.Supported("execve", Execve), - 60: syscalls.Supported("exit", Exit), - 61: syscalls.Supported("wait4", Wait4), - 62: syscalls.Supported("kill", Kill), - 63: syscalls.Supported("uname", Uname), - 64: syscalls.Supported("semget", Semget), - 65: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), - 66: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), - 67: syscalls.Supported("shmdt", Shmdt), - 68: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 69: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 70: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 71: syscalls.ErrorWithEvent("msgctl", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 72: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), - 73: syscalls.PartiallySupported("flock", Flock, "Locks are held within the sandbox only.", nil), - 74: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), - 75: syscalls.PartiallySupported("fdatasync", Fdatasync, "Full data flush is not guaranteed at this time.", nil), - 76: syscalls.Supported("truncate", Truncate), - 77: syscalls.Supported("ftruncate", Ftruncate), - 78: syscalls.Supported("getdents", Getdents), - 79: syscalls.Supported("getcwd", Getcwd), - 80: syscalls.Supported("chdir", Chdir), - 81: syscalls.Supported("fchdir", Fchdir), - 82: syscalls.Supported("rename", Rename), - 83: syscalls.Supported("mkdir", Mkdir), - 84: syscalls.Supported("rmdir", Rmdir), - 85: syscalls.Supported("creat", Creat), - 86: syscalls.Supported("link", Link), - 87: syscalls.Supported("unlink", Unlink), - 88: syscalls.Supported("symlink", Symlink), - 89: syscalls.Supported("readlink", Readlink), - 90: syscalls.Supported("chmod", Chmod), - 91: syscalls.PartiallySupported("fchmod", Fchmod, "Options S_ISUID and S_ISGID not supported.", nil), - 92: syscalls.Supported("chown", Chown), - 93: syscalls.Supported("fchown", Fchown), - 94: syscalls.Supported("lchown", Lchown), - 95: syscalls.Supported("umask", Umask), - 96: syscalls.Supported("gettimeofday", Gettimeofday), - 97: syscalls.Supported("getrlimit", Getrlimit), - 98: syscalls.PartiallySupported("getrusage", Getrusage, "Fields ru_maxrss, ru_minflt, ru_majflt, ru_inblock, ru_oublock are not supported. Fields ru_utime and ru_stime have low precision.", nil), - 99: syscalls.PartiallySupported("sysinfo", Sysinfo, "Fields loads, sharedram, bufferram, totalswap, freeswap, totalhigh, freehigh not supported.", nil), - 100: syscalls.Supported("times", Times), - 101: syscalls.PartiallySupported("ptrace", Ptrace, "Options PTRACE_PEEKSIGINFO, PTRACE_SECCOMP_GET_FILTER not supported.", nil), - 102: syscalls.Supported("getuid", Getuid), - 103: syscalls.PartiallySupported("syslog", Syslog, "Outputs a dummy message for security reasons.", nil), - 104: syscalls.Supported("getgid", Getgid), - 105: syscalls.Supported("setuid", Setuid), - 106: syscalls.Supported("setgid", Setgid), - 107: syscalls.Supported("geteuid", Geteuid), - 108: syscalls.Supported("getegid", Getegid), - 109: syscalls.Supported("setpgid", Setpgid), - 110: syscalls.Supported("getppid", Getppid), - 111: syscalls.Supported("getpgrp", Getpgrp), - 112: syscalls.Supported("setsid", Setsid), - 113: syscalls.Supported("setreuid", Setreuid), - 114: syscalls.Supported("setregid", Setregid), - 115: syscalls.Supported("getgroups", Getgroups), - 116: syscalls.Supported("setgroups", Setgroups), - 117: syscalls.Supported("setresuid", Setresuid), - 118: syscalls.Supported("getresuid", Getresuid), - 119: syscalls.Supported("setresgid", Setresgid), - 120: syscalls.Supported("getresgid", Getresgid), - 121: syscalls.Supported("getpgid", Getpgid), - 122: syscalls.ErrorWithEvent("setfsuid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) - 123: syscalls.ErrorWithEvent("setfsgid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) - 124: syscalls.Supported("getsid", Getsid), - 125: syscalls.Supported("capget", Capget), - 126: syscalls.Supported("capset", Capset), - 127: syscalls.Supported("rt_sigpending", RtSigpending), - 128: syscalls.Supported("rt_sigtimedwait", RtSigtimedwait), - 129: syscalls.Supported("rt_sigqueueinfo", RtSigqueueinfo), - 130: syscalls.Supported("rt_sigsuspend", RtSigsuspend), - 131: syscalls.Supported("sigaltstack", Sigaltstack), - 132: syscalls.Supported("utime", Utime), - 133: syscalls.PartiallySupported("mknod", Mknod, "Device creation is not generally supported. Only regular file and FIFO creation are supported.", nil), - 134: syscalls.Error("uselib", syserror.ENOSYS, "Obsolete", nil), - 135: syscalls.ErrorWithEvent("personality", syserror.EINVAL, "Unable to change personality.", nil), - 136: syscalls.ErrorWithEvent("ustat", syserror.ENOSYS, "Needs filesystem support.", nil), - 137: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), - 138: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), - 139: syscalls.ErrorWithEvent("sysfs", syserror.ENOSYS, "", []string{"gvisor.dev/issue/165"}), - 140: syscalls.PartiallySupported("getpriority", Getpriority, "Stub implementation.", nil), - 141: syscalls.PartiallySupported("setpriority", Setpriority, "Stub implementation.", nil), - 142: syscalls.CapError("sched_setparam", linux.CAP_SYS_NICE, "", nil), - 143: syscalls.PartiallySupported("sched_getparam", SchedGetparam, "Stub implementation.", nil), - 144: syscalls.PartiallySupported("sched_setscheduler", SchedSetscheduler, "Stub implementation.", nil), - 145: syscalls.PartiallySupported("sched_getscheduler", SchedGetscheduler, "Stub implementation.", nil), - 146: syscalls.PartiallySupported("sched_get_priority_max", SchedGetPriorityMax, "Stub implementation.", nil), - 147: syscalls.PartiallySupported("sched_get_priority_min", SchedGetPriorityMin, "Stub implementation.", nil), - 148: syscalls.ErrorWithEvent("sched_rr_get_interval", syserror.EPERM, "", nil), - 149: syscalls.PartiallySupported("mlock", Mlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 150: syscalls.PartiallySupported("munlock", Munlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 151: syscalls.PartiallySupported("mlockall", Mlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 152: syscalls.PartiallySupported("munlockall", Munlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), - 153: syscalls.CapError("vhangup", linux.CAP_SYS_TTY_CONFIG, "", nil), - 154: syscalls.Error("modify_ldt", syserror.EPERM, "", nil), - 155: syscalls.Error("pivot_root", syserror.EPERM, "", nil), - 156: syscalls.Error("sysctl", syserror.EPERM, "Deprecated. Use /proc/sys instead.", nil), - 157: syscalls.PartiallySupported("prctl", Prctl, "Not all options are supported.", nil), - 158: syscalls.PartiallySupported("arch_prctl", ArchPrctl, "Options ARCH_GET_GS, ARCH_SET_GS not supported.", nil), - 159: syscalls.CapError("adjtimex", linux.CAP_SYS_TIME, "", nil), - 160: syscalls.PartiallySupported("setrlimit", Setrlimit, "Not all rlimits are enforced.", nil), - 161: syscalls.Supported("chroot", Chroot), - 162: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), - 163: syscalls.CapError("acct", linux.CAP_SYS_PACCT, "", nil), - 164: syscalls.CapError("settimeofday", linux.CAP_SYS_TIME, "", nil), - 165: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), - 166: syscalls.PartiallySupported("umount2", Umount2, "Not all options or file systems are supported.", nil), - 167: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), - 168: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), - 169: syscalls.CapError("reboot", linux.CAP_SYS_BOOT, "", nil), - 170: syscalls.Supported("sethostname", Sethostname), - 171: syscalls.Supported("setdomainname", Setdomainname), - 172: syscalls.CapError("iopl", linux.CAP_SYS_RAWIO, "", nil), - 173: syscalls.CapError("ioperm", linux.CAP_SYS_RAWIO, "", nil), - 174: syscalls.CapError("create_module", linux.CAP_SYS_MODULE, "", nil), - 175: syscalls.CapError("init_module", linux.CAP_SYS_MODULE, "", nil), - 176: syscalls.CapError("delete_module", linux.CAP_SYS_MODULE, "", nil), - 177: syscalls.Error("get_kernel_syms", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), - 178: syscalls.Error("query_module", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), - 179: syscalls.CapError("quotactl", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_admin for most operations - 180: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), - 181: syscalls.Error("getpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), - 182: syscalls.Error("putpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), - 183: syscalls.Error("afs_syscall", syserror.ENOSYS, "Not implemented in Linux.", nil), - 184: syscalls.Error("tuxcall", syserror.ENOSYS, "Not implemented in Linux.", nil), - 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), - 186: syscalls.Supported("gettid", Gettid), - 187: syscalls.ErrorWithEvent("readahead", syserror.ENOSYS, "", []string{"gvisor.dev/issue/261"}), // TODO(b/29351341) - 188: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 191: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 197: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 198: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 199: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 200: syscalls.Supported("tkill", Tkill), - 201: syscalls.Supported("time", Time), - 202: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), - 203: syscalls.PartiallySupported("sched_setaffinity", SchedSetaffinity, "Stub implementation.", nil), - 204: syscalls.PartiallySupported("sched_getaffinity", SchedGetaffinity, "Stub implementation.", nil), - 205: syscalls.Error("set_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), - 206: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 207: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 208: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 209: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 210: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 211: syscalls.Error("get_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), - 212: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), - 213: syscalls.Supported("epoll_create", EpollCreate), - 214: syscalls.ErrorWithEvent("epoll_ctl_old", syserror.ENOSYS, "Deprecated.", nil), - 215: syscalls.ErrorWithEvent("epoll_wait_old", syserror.ENOSYS, "Deprecated.", nil), - 216: syscalls.ErrorWithEvent("remap_file_pages", syserror.ENOSYS, "Deprecated since Linux 3.16.", nil), - 217: syscalls.Supported("getdents64", Getdents64), - 218: syscalls.Supported("set_tid_address", SetTidAddress), - 219: syscalls.Supported("restart_syscall", RestartSyscall), - 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) - 221: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), - 222: syscalls.Supported("timer_create", TimerCreate), - 223: syscalls.Supported("timer_settime", TimerSettime), - 224: syscalls.Supported("timer_gettime", TimerGettime), - 225: syscalls.Supported("timer_getoverrun", TimerGetoverrun), - 226: syscalls.Supported("timer_delete", TimerDelete), - 227: syscalls.Supported("clock_settime", ClockSettime), - 228: syscalls.Supported("clock_gettime", ClockGettime), - 229: syscalls.Supported("clock_getres", ClockGetres), - 230: syscalls.Supported("clock_nanosleep", ClockNanosleep), - 231: syscalls.Supported("exit_group", ExitGroup), - 232: syscalls.Supported("epoll_wait", EpollWait), - 233: syscalls.Supported("epoll_ctl", EpollCtl), - 234: syscalls.Supported("tgkill", Tgkill), - 235: syscalls.Supported("utimes", Utimes), - 236: syscalls.Error("vserver", syserror.ENOSYS, "Not implemented by Linux", nil), - 237: syscalls.PartiallySupported("mbind", Mbind, "Stub implementation. Only a single NUMA node is advertised, and mempolicy is ignored accordingly, but mbind() will succeed and has effects reflected by get_mempolicy.", []string{"gvisor.dev/issue/262"}), - 238: syscalls.PartiallySupported("set_mempolicy", SetMempolicy, "Stub implementation.", nil), - 239: syscalls.PartiallySupported("get_mempolicy", GetMempolicy, "Stub implementation.", nil), - 240: syscalls.ErrorWithEvent("mq_open", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 241: syscalls.ErrorWithEvent("mq_unlink", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 242: syscalls.ErrorWithEvent("mq_timedsend", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 243: syscalls.ErrorWithEvent("mq_timedreceive", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 244: syscalls.ErrorWithEvent("mq_notify", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 245: syscalls.ErrorWithEvent("mq_getsetattr", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 246: syscalls.CapError("kexec_load", linux.CAP_SYS_BOOT, "", nil), - 247: syscalls.Supported("waitid", Waitid), - 248: syscalls.Error("add_key", syserror.EACCES, "Not available to user.", nil), - 249: syscalls.Error("request_key", syserror.EACCES, "Not available to user.", nil), - 250: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), - 251: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) - 252: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) - 253: syscalls.PartiallySupported("inotify_init", InotifyInit, "inotify events are only available inside the sandbox.", nil), - 254: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), - 255: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), - 256: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), - 257: syscalls.Supported("openat", Openat), - 258: syscalls.Supported("mkdirat", Mkdirat), - 259: syscalls.Supported("mknodat", Mknodat), - 260: syscalls.Supported("fchownat", Fchownat), - 261: syscalls.Supported("futimesat", Futimesat), - 262: syscalls.Supported("fstatat", Fstatat), - 263: syscalls.Supported("unlinkat", Unlinkat), - 264: syscalls.Supported("renameat", Renameat), - 265: syscalls.Supported("linkat", Linkat), - 266: syscalls.Supported("symlinkat", Symlinkat), - 267: syscalls.Supported("readlinkat", Readlinkat), - 268: syscalls.Supported("fchmodat", Fchmodat), - 269: syscalls.Supported("faccessat", Faccessat), - 270: syscalls.Supported("pselect", Pselect), - 271: syscalls.Supported("ppoll", Ppoll), - 272: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), - 273: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 274: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), - 275: syscalls.Supported("splice", Splice), - 276: syscalls.Supported("tee", Tee), - 277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), - 278: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) - 279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) - 280: syscalls.Supported("utimensat", Utimensat), - 281: syscalls.Supported("epoll_pwait", EpollPwait), - 282: syscalls.PartiallySupported("signalfd", Signalfd, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), - 283: syscalls.Supported("timerfd_create", TimerfdCreate), - 284: syscalls.Supported("eventfd", Eventfd), - 285: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), - 286: syscalls.Supported("timerfd_settime", TimerfdSettime), - 287: syscalls.Supported("timerfd_gettime", TimerfdGettime), - 288: syscalls.Supported("accept4", Accept4), - 289: syscalls.PartiallySupported("signalfd4", Signalfd4, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), - 290: syscalls.Supported("eventfd2", Eventfd2), - 291: syscalls.Supported("epoll_create1", EpollCreate1), - 292: syscalls.Supported("dup3", Dup3), - 293: syscalls.Supported("pipe2", Pipe2), - 294: syscalls.Supported("inotify_init1", InotifyInit1), - 295: syscalls.Supported("preadv", Preadv), - 296: syscalls.Supported("pwritev", Pwritev), - 297: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), - 298: syscalls.ErrorWithEvent("perf_event_open", syserror.ENODEV, "No support for perf counters", nil), - 299: syscalls.PartiallySupported("recvmmsg", RecvMMsg, "Not all flags and control messages are supported.", nil), - 300: syscalls.ErrorWithEvent("fanotify_init", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), - 301: syscalls.ErrorWithEvent("fanotify_mark", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), - 302: syscalls.Supported("prlimit64", Prlimit64), - 303: syscalls.Error("name_to_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), - 304: syscalls.Error("open_by_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), - 305: syscalls.CapError("clock_adjtime", linux.CAP_SYS_TIME, "", nil), - 306: syscalls.PartiallySupported("syncfs", Syncfs, "Depends on backing file system.", nil), - 307: syscalls.PartiallySupported("sendmmsg", SendMMsg, "Not all flags and control messages are supported.", nil), - 308: syscalls.ErrorWithEvent("setns", syserror.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) - 309: syscalls.Supported("getcpu", Getcpu), - 310: syscalls.ErrorWithEvent("process_vm_readv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), - 311: syscalls.ErrorWithEvent("process_vm_writev", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), - 312: syscalls.CapError("kcmp", linux.CAP_SYS_PTRACE, "", nil), - 313: syscalls.CapError("finit_module", linux.CAP_SYS_MODULE, "", nil), - 314: syscalls.ErrorWithEvent("sched_setattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) - 315: syscalls.ErrorWithEvent("sched_getattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) - 316: syscalls.ErrorWithEvent("renameat2", syserror.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) - 317: syscalls.Supported("seccomp", Seccomp), - 318: syscalls.Supported("getrandom", GetRandom), - 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) - 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), - - // Syscalls after 325 are "backports" from versions of Linux after 4.4. - 326: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), - 327: syscalls.Supported("preadv2", Preadv2), - 328: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), - 332: syscalls.Supported("statx", Statx), - }, - - Emulate: map[usermem.Addr]uintptr{ - 0xffffffffff600000: 96, // vsyscall gettimeofday(2) - 0xffffffffff600400: 201, // vsyscall time(2) - 0xffffffffff600800: 309, // vsyscall getcpu(2) - }, - Missing: func(t *kernel.Task, sysno uintptr, args arch.SyscallArguments) (uintptr, error) { - t.Kernel().EmitUnimplementedEvent(t) - return 0, syserror.ENOSYS - }, -} diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go new file mode 100644 index 000000000..e215ac049 --- /dev/null +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -0,0 +1,386 @@ +// 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 linux + +import ( + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" +) + +// AMD64 is a table of Linux amd64 syscall API with the corresponding syscall +// numbers from Linux 4.4. +var AMD64 = &kernel.SyscallTable{ + OS: abi.Linux, + Arch: arch.AMD64, + Version: kernel.Version{ + // Version 4.4 is chosen as a stable, longterm version of Linux, which + // 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, + }, + AuditNumber: linux.AUDIT_ARCH_X86_64, + Table: map[uintptr]kernel.Syscall{ + 0: syscalls.Supported("read", Read), + 1: syscalls.Supported("write", Write), + 2: syscalls.PartiallySupported("open", Open, "Options O_DIRECT, O_NOATIME, O_PATH, O_TMPFILE, O_SYNC are not supported.", nil), + 3: syscalls.Supported("close", Close), + 4: syscalls.Supported("stat", Stat), + 5: syscalls.Supported("fstat", Fstat), + 6: syscalls.Supported("lstat", Lstat), + 7: syscalls.Supported("poll", Poll), + 8: syscalls.Supported("lseek", Lseek), + 9: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), + 10: syscalls.Supported("mprotect", Mprotect), + 11: syscalls.Supported("munmap", Munmap), + 12: syscalls.Supported("brk", Brk), + 13: syscalls.Supported("rt_sigaction", RtSigaction), + 14: syscalls.Supported("rt_sigprocmask", RtSigprocmask), + 15: syscalls.Supported("rt_sigreturn", RtSigreturn), + 16: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), + 17: syscalls.Supported("pread64", Pread64), + 18: syscalls.Supported("pwrite64", Pwrite64), + 19: syscalls.Supported("readv", Readv), + 20: syscalls.Supported("writev", Writev), + 21: syscalls.Supported("access", Access), + 22: syscalls.Supported("pipe", Pipe), + 23: syscalls.Supported("select", Select), + 24: syscalls.Supported("sched_yield", SchedYield), + 25: syscalls.Supported("mremap", Mremap), + 26: syscalls.PartiallySupported("msync", Msync, "Full data flush is not guaranteed at this time.", nil), + 27: syscalls.PartiallySupported("mincore", Mincore, "Stub implementation. The sandbox does not have access to this information. Reports all mapped pages are resident.", nil), + 28: syscalls.PartiallySupported("madvise", Madvise, "Options MADV_DONTNEED, MADV_DONTFORK are supported. Other advice is ignored.", nil), + 29: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), + 30: syscalls.PartiallySupported("shmat", Shmat, "Option SHM_RND is not supported.", nil), + 31: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), + 32: syscalls.Supported("dup", Dup), + 33: syscalls.Supported("dup2", Dup2), + 34: syscalls.Supported("pause", Pause), + 35: syscalls.Supported("nanosleep", Nanosleep), + 36: syscalls.Supported("getitimer", Getitimer), + 37: syscalls.Supported("alarm", Alarm), + 38: syscalls.Supported("setitimer", Setitimer), + 39: syscalls.Supported("getpid", Getpid), + 40: syscalls.Supported("sendfile", Sendfile), + 41: syscalls.PartiallySupported("socket", Socket, "Limited support for AF_NETLINK, NETLINK_ROUTE sockets. Limited support for SOCK_RAW.", nil), + 42: syscalls.Supported("connect", Connect), + 43: syscalls.Supported("accept", Accept), + 44: syscalls.Supported("sendto", SendTo), + 45: syscalls.Supported("recvfrom", RecvFrom), + 46: syscalls.Supported("sendmsg", SendMsg), + 47: syscalls.PartiallySupported("recvmsg", RecvMsg, "Not all flags and control messages are supported.", nil), + 48: syscalls.PartiallySupported("shutdown", Shutdown, "Not all flags and control messages are supported.", nil), + 49: syscalls.PartiallySupported("bind", Bind, "Autobind for abstract Unix sockets is not supported.", nil), + 50: syscalls.Supported("listen", Listen), + 51: syscalls.Supported("getsockname", GetSockName), + 52: syscalls.Supported("getpeername", GetPeerName), + 53: syscalls.Supported("socketpair", SocketPair), + 54: syscalls.PartiallySupported("setsockopt", SetSockOpt, "Not all socket options are supported.", nil), + 55: syscalls.PartiallySupported("getsockopt", GetSockOpt, "Not all socket options are supported.", nil), + 56: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), + 57: syscalls.Supported("fork", Fork), + 58: syscalls.Supported("vfork", Vfork), + 59: syscalls.Supported("execve", Execve), + 60: syscalls.Supported("exit", Exit), + 61: syscalls.Supported("wait4", Wait4), + 62: syscalls.Supported("kill", Kill), + 63: syscalls.Supported("uname", Uname), + 64: syscalls.Supported("semget", Semget), + 65: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), + 66: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), + 67: syscalls.Supported("shmdt", Shmdt), + 68: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 69: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 70: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 71: syscalls.ErrorWithEvent("msgctl", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 72: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), + 73: syscalls.PartiallySupported("flock", Flock, "Locks are held within the sandbox only.", nil), + 74: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), + 75: syscalls.PartiallySupported("fdatasync", Fdatasync, "Full data flush is not guaranteed at this time.", nil), + 76: syscalls.Supported("truncate", Truncate), + 77: syscalls.Supported("ftruncate", Ftruncate), + 78: syscalls.Supported("getdents", Getdents), + 79: syscalls.Supported("getcwd", Getcwd), + 80: syscalls.Supported("chdir", Chdir), + 81: syscalls.Supported("fchdir", Fchdir), + 82: syscalls.Supported("rename", Rename), + 83: syscalls.Supported("mkdir", Mkdir), + 84: syscalls.Supported("rmdir", Rmdir), + 85: syscalls.Supported("creat", Creat), + 86: syscalls.Supported("link", Link), + 87: syscalls.Supported("unlink", Unlink), + 88: syscalls.Supported("symlink", Symlink), + 89: syscalls.Supported("readlink", Readlink), + 90: syscalls.Supported("chmod", Chmod), + 91: syscalls.PartiallySupported("fchmod", Fchmod, "Options S_ISUID and S_ISGID not supported.", nil), + 92: syscalls.Supported("chown", Chown), + 93: syscalls.Supported("fchown", Fchown), + 94: syscalls.Supported("lchown", Lchown), + 95: syscalls.Supported("umask", Umask), + 96: syscalls.Supported("gettimeofday", Gettimeofday), + 97: syscalls.Supported("getrlimit", Getrlimit), + 98: syscalls.PartiallySupported("getrusage", Getrusage, "Fields ru_maxrss, ru_minflt, ru_majflt, ru_inblock, ru_oublock are not supported. Fields ru_utime and ru_stime have low precision.", nil), + 99: syscalls.PartiallySupported("sysinfo", Sysinfo, "Fields loads, sharedram, bufferram, totalswap, freeswap, totalhigh, freehigh not supported.", nil), + 100: syscalls.Supported("times", Times), + 101: syscalls.PartiallySupported("ptrace", Ptrace, "Options PTRACE_PEEKSIGINFO, PTRACE_SECCOMP_GET_FILTER not supported.", nil), + 102: syscalls.Supported("getuid", Getuid), + 103: syscalls.PartiallySupported("syslog", Syslog, "Outputs a dummy message for security reasons.", nil), + 104: syscalls.Supported("getgid", Getgid), + 105: syscalls.Supported("setuid", Setuid), + 106: syscalls.Supported("setgid", Setgid), + 107: syscalls.Supported("geteuid", Geteuid), + 108: syscalls.Supported("getegid", Getegid), + 109: syscalls.Supported("setpgid", Setpgid), + 110: syscalls.Supported("getppid", Getppid), + 111: syscalls.Supported("getpgrp", Getpgrp), + 112: syscalls.Supported("setsid", Setsid), + 113: syscalls.Supported("setreuid", Setreuid), + 114: syscalls.Supported("setregid", Setregid), + 115: syscalls.Supported("getgroups", Getgroups), + 116: syscalls.Supported("setgroups", Setgroups), + 117: syscalls.Supported("setresuid", Setresuid), + 118: syscalls.Supported("getresuid", Getresuid), + 119: syscalls.Supported("setresgid", Setresgid), + 120: syscalls.Supported("getresgid", Getresgid), + 121: syscalls.Supported("getpgid", Getpgid), + 122: syscalls.ErrorWithEvent("setfsuid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 123: syscalls.ErrorWithEvent("setfsgid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 124: syscalls.Supported("getsid", Getsid), + 125: syscalls.Supported("capget", Capget), + 126: syscalls.Supported("capset", Capset), + 127: syscalls.Supported("rt_sigpending", RtSigpending), + 128: syscalls.Supported("rt_sigtimedwait", RtSigtimedwait), + 129: syscalls.Supported("rt_sigqueueinfo", RtSigqueueinfo), + 130: syscalls.Supported("rt_sigsuspend", RtSigsuspend), + 131: syscalls.Supported("sigaltstack", Sigaltstack), + 132: syscalls.Supported("utime", Utime), + 133: syscalls.PartiallySupported("mknod", Mknod, "Device creation is not generally supported. Only regular file and FIFO creation are supported.", nil), + 134: syscalls.Error("uselib", syserror.ENOSYS, "Obsolete", nil), + 135: syscalls.ErrorWithEvent("personality", syserror.EINVAL, "Unable to change personality.", nil), + 136: syscalls.ErrorWithEvent("ustat", syserror.ENOSYS, "Needs filesystem support.", nil), + 137: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), + 138: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), + 139: syscalls.ErrorWithEvent("sysfs", syserror.ENOSYS, "", []string{"gvisor.dev/issue/165"}), + 140: syscalls.PartiallySupported("getpriority", Getpriority, "Stub implementation.", nil), + 141: syscalls.PartiallySupported("setpriority", Setpriority, "Stub implementation.", nil), + 142: syscalls.CapError("sched_setparam", linux.CAP_SYS_NICE, "", nil), + 143: syscalls.PartiallySupported("sched_getparam", SchedGetparam, "Stub implementation.", nil), + 144: syscalls.PartiallySupported("sched_setscheduler", SchedSetscheduler, "Stub implementation.", nil), + 145: syscalls.PartiallySupported("sched_getscheduler", SchedGetscheduler, "Stub implementation.", nil), + 146: syscalls.PartiallySupported("sched_get_priority_max", SchedGetPriorityMax, "Stub implementation.", nil), + 147: syscalls.PartiallySupported("sched_get_priority_min", SchedGetPriorityMin, "Stub implementation.", nil), + 148: syscalls.ErrorWithEvent("sched_rr_get_interval", syserror.EPERM, "", nil), + 149: syscalls.PartiallySupported("mlock", Mlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 150: syscalls.PartiallySupported("munlock", Munlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 151: syscalls.PartiallySupported("mlockall", Mlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 152: syscalls.PartiallySupported("munlockall", Munlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 153: syscalls.CapError("vhangup", linux.CAP_SYS_TTY_CONFIG, "", nil), + 154: syscalls.Error("modify_ldt", syserror.EPERM, "", nil), + 155: syscalls.Error("pivot_root", syserror.EPERM, "", nil), + 156: syscalls.Error("sysctl", syserror.EPERM, "Deprecated. Use /proc/sys instead.", nil), + 157: syscalls.PartiallySupported("prctl", Prctl, "Not all options are supported.", nil), + 158: syscalls.PartiallySupported("arch_prctl", ArchPrctl, "Options ARCH_GET_GS, ARCH_SET_GS not supported.", nil), + 159: syscalls.CapError("adjtimex", linux.CAP_SYS_TIME, "", nil), + 160: syscalls.PartiallySupported("setrlimit", Setrlimit, "Not all rlimits are enforced.", nil), + 161: syscalls.Supported("chroot", Chroot), + 162: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), + 163: syscalls.CapError("acct", linux.CAP_SYS_PACCT, "", nil), + 164: syscalls.CapError("settimeofday", linux.CAP_SYS_TIME, "", nil), + 165: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), + 166: syscalls.PartiallySupported("umount2", Umount2, "Not all options or file systems are supported.", nil), + 167: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), + 168: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), + 169: syscalls.CapError("reboot", linux.CAP_SYS_BOOT, "", nil), + 170: syscalls.Supported("sethostname", Sethostname), + 171: syscalls.Supported("setdomainname", Setdomainname), + 172: syscalls.CapError("iopl", linux.CAP_SYS_RAWIO, "", nil), + 173: syscalls.CapError("ioperm", linux.CAP_SYS_RAWIO, "", nil), + 174: syscalls.CapError("create_module", linux.CAP_SYS_MODULE, "", nil), + 175: syscalls.CapError("init_module", linux.CAP_SYS_MODULE, "", nil), + 176: syscalls.CapError("delete_module", linux.CAP_SYS_MODULE, "", nil), + 177: syscalls.Error("get_kernel_syms", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), + 178: syscalls.Error("query_module", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), + 179: syscalls.CapError("quotactl", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_admin for most operations + 180: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), + 181: syscalls.Error("getpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), + 182: syscalls.Error("putpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), + 183: syscalls.Error("afs_syscall", syserror.ENOSYS, "Not implemented in Linux.", nil), + 184: syscalls.Error("tuxcall", syserror.ENOSYS, "Not implemented in Linux.", nil), + 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), + 186: syscalls.Supported("gettid", Gettid), + 187: syscalls.Supported("readahead", Readahead), + 188: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 191: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 197: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 198: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 199: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 200: syscalls.Supported("tkill", Tkill), + 201: syscalls.Supported("time", Time), + 202: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), + 203: syscalls.PartiallySupported("sched_setaffinity", SchedSetaffinity, "Stub implementation.", nil), + 204: syscalls.PartiallySupported("sched_getaffinity", SchedGetaffinity, "Stub implementation.", nil), + 205: syscalls.Error("set_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), + 206: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 207: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 208: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 209: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 210: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 211: syscalls.Error("get_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), + 212: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), + 213: syscalls.Supported("epoll_create", EpollCreate), + 214: syscalls.ErrorWithEvent("epoll_ctl_old", syserror.ENOSYS, "Deprecated.", nil), + 215: syscalls.ErrorWithEvent("epoll_wait_old", syserror.ENOSYS, "Deprecated.", nil), + 216: syscalls.ErrorWithEvent("remap_file_pages", syserror.ENOSYS, "Deprecated since Linux 3.16.", nil), + 217: syscalls.Supported("getdents64", Getdents64), + 218: syscalls.Supported("set_tid_address", SetTidAddress), + 219: syscalls.Supported("restart_syscall", RestartSyscall), + 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) + 221: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), + 222: syscalls.Supported("timer_create", TimerCreate), + 223: syscalls.Supported("timer_settime", TimerSettime), + 224: syscalls.Supported("timer_gettime", TimerGettime), + 225: syscalls.Supported("timer_getoverrun", TimerGetoverrun), + 226: syscalls.Supported("timer_delete", TimerDelete), + 227: syscalls.Supported("clock_settime", ClockSettime), + 228: syscalls.Supported("clock_gettime", ClockGettime), + 229: syscalls.Supported("clock_getres", ClockGetres), + 230: syscalls.Supported("clock_nanosleep", ClockNanosleep), + 231: syscalls.Supported("exit_group", ExitGroup), + 232: syscalls.Supported("epoll_wait", EpollWait), + 233: syscalls.Supported("epoll_ctl", EpollCtl), + 234: syscalls.Supported("tgkill", Tgkill), + 235: syscalls.Supported("utimes", Utimes), + 236: syscalls.Error("vserver", syserror.ENOSYS, "Not implemented by Linux", nil), + 237: syscalls.PartiallySupported("mbind", Mbind, "Stub implementation. Only a single NUMA node is advertised, and mempolicy is ignored accordingly, but mbind() will succeed and has effects reflected by get_mempolicy.", []string{"gvisor.dev/issue/262"}), + 238: syscalls.PartiallySupported("set_mempolicy", SetMempolicy, "Stub implementation.", nil), + 239: syscalls.PartiallySupported("get_mempolicy", GetMempolicy, "Stub implementation.", nil), + 240: syscalls.ErrorWithEvent("mq_open", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 241: syscalls.ErrorWithEvent("mq_unlink", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 242: syscalls.ErrorWithEvent("mq_timedsend", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 243: syscalls.ErrorWithEvent("mq_timedreceive", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 244: syscalls.ErrorWithEvent("mq_notify", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 245: syscalls.ErrorWithEvent("mq_getsetattr", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 246: syscalls.CapError("kexec_load", linux.CAP_SYS_BOOT, "", nil), + 247: syscalls.Supported("waitid", Waitid), + 248: syscalls.Error("add_key", syserror.EACCES, "Not available to user.", nil), + 249: syscalls.Error("request_key", syserror.EACCES, "Not available to user.", nil), + 250: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), + 251: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) + 252: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) + 253: syscalls.PartiallySupported("inotify_init", InotifyInit, "inotify events are only available inside the sandbox.", nil), + 254: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), + 255: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), + 256: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), + 257: syscalls.Supported("openat", Openat), + 258: syscalls.Supported("mkdirat", Mkdirat), + 259: syscalls.Supported("mknodat", Mknodat), + 260: syscalls.Supported("fchownat", Fchownat), + 261: syscalls.Supported("futimesat", Futimesat), + 262: syscalls.Supported("fstatat", Fstatat), + 263: syscalls.Supported("unlinkat", Unlinkat), + 264: syscalls.Supported("renameat", Renameat), + 265: syscalls.Supported("linkat", Linkat), + 266: syscalls.Supported("symlinkat", Symlinkat), + 267: syscalls.Supported("readlinkat", Readlinkat), + 268: syscalls.Supported("fchmodat", Fchmodat), + 269: syscalls.Supported("faccessat", Faccessat), + 270: syscalls.Supported("pselect", Pselect), + 271: syscalls.Supported("ppoll", Ppoll), + 272: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), + 273: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 274: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 275: syscalls.Supported("splice", Splice), + 276: syscalls.Supported("tee", Tee), + 277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), + 278: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) + 280: syscalls.Supported("utimensat", Utimensat), + 281: syscalls.Supported("epoll_pwait", EpollPwait), + 282: syscalls.PartiallySupported("signalfd", Signalfd, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), + 283: syscalls.Supported("timerfd_create", TimerfdCreate), + 284: syscalls.Supported("eventfd", Eventfd), + 285: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), + 286: syscalls.Supported("timerfd_settime", TimerfdSettime), + 287: syscalls.Supported("timerfd_gettime", TimerfdGettime), + 288: syscalls.Supported("accept4", Accept4), + 289: syscalls.PartiallySupported("signalfd4", Signalfd4, "Semantics are slightly different.", []string{"gvisor.dev/issue/139"}), + 290: syscalls.Supported("eventfd2", Eventfd2), + 291: syscalls.Supported("epoll_create1", EpollCreate1), + 292: syscalls.Supported("dup3", Dup3), + 293: syscalls.Supported("pipe2", Pipe2), + 294: syscalls.Supported("inotify_init1", InotifyInit1), + 295: syscalls.Supported("preadv", Preadv), + 296: syscalls.Supported("pwritev", Pwritev), + 297: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), + 298: syscalls.ErrorWithEvent("perf_event_open", syserror.ENODEV, "No support for perf counters", nil), + 299: syscalls.PartiallySupported("recvmmsg", RecvMMsg, "Not all flags and control messages are supported.", nil), + 300: syscalls.ErrorWithEvent("fanotify_init", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 301: syscalls.ErrorWithEvent("fanotify_mark", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 302: syscalls.Supported("prlimit64", Prlimit64), + 303: syscalls.Error("name_to_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 304: syscalls.Error("open_by_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 305: syscalls.CapError("clock_adjtime", linux.CAP_SYS_TIME, "", nil), + 306: syscalls.PartiallySupported("syncfs", Syncfs, "Depends on backing file system.", nil), + 307: syscalls.PartiallySupported("sendmmsg", SendMMsg, "Not all flags and control messages are supported.", nil), + 308: syscalls.ErrorWithEvent("setns", syserror.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) + 309: syscalls.Supported("getcpu", Getcpu), + 310: syscalls.ErrorWithEvent("process_vm_readv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 311: syscalls.ErrorWithEvent("process_vm_writev", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 312: syscalls.CapError("kcmp", linux.CAP_SYS_PTRACE, "", nil), + 313: syscalls.CapError("finit_module", linux.CAP_SYS_MODULE, "", nil), + 314: syscalls.ErrorWithEvent("sched_setattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 315: syscalls.ErrorWithEvent("sched_getattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 316: syscalls.ErrorWithEvent("renameat2", syserror.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) + 317: syscalls.Supported("seccomp", Seccomp), + 318: syscalls.Supported("getrandom", GetRandom), + 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) + 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), + + // Syscalls after 325 are "backports" from versions of Linux after 4.4. + 326: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), + 327: syscalls.Supported("preadv2", Preadv2), + 328: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), + 332: syscalls.Supported("statx", Statx), + }, + + Emulate: map[usermem.Addr]uintptr{ + 0xffffffffff600000: 96, // vsyscall gettimeofday(2) + 0xffffffffff600400: 201, // vsyscall time(2) + 0xffffffffff600800: 309, // vsyscall getcpu(2) + }, + Missing: func(t *kernel.Task, sysno uintptr, args arch.SyscallArguments) (uintptr, error) { + t.Kernel().EmitUnimplementedEvent(t) + return 0, syserror.ENOSYS + }, +} diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go new file mode 100644 index 000000000..1d3b63020 --- /dev/null +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -0,0 +1,313 @@ +// 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 linux + +import ( + "gvisor.dev/gvisor/pkg/abi" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/syscalls" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" +) + +// ARM64 is a table of Linux arm64 syscall API with the corresponding syscall +// numbers from Linux 4.4. +var ARM64 = &kernel.SyscallTable{ + OS: abi.Linux, + Arch: arch.ARM64, + Version: kernel.Version{ + Sysname: _LINUX_SYSNAME, + Release: _LINUX_RELEASE, + Version: _LINUX_VERSION, + }, + AuditNumber: linux.AUDIT_ARCH_AARCH64, + Table: map[uintptr]kernel.Syscall{ + 0: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 1: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 5: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 6: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 7: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 8: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 12: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 13: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 14: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 15: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 16: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 17: syscalls.Supported("getcwd", Getcwd), + 18: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), + 19: syscalls.Supported("eventfd2", Eventfd2), + 20: syscalls.Supported("epoll_create1", EpollCreate1), + 21: syscalls.Supported("epoll_ctl", EpollCtl), + 22: syscalls.Supported("epoll_pwait", EpollPwait), + 23: syscalls.Supported("dup", Dup), + 24: syscalls.Supported("dup3", Dup3), + 26: syscalls.Supported("inotify_init1", InotifyInit1), + 27: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), + 28: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), + 29: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), + 30: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) + 31: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) + 32: syscalls.PartiallySupported("flock", Flock, "Locks are held within the sandbox only.", nil), + 33: syscalls.Supported("mknodat", Mknodat), + 34: syscalls.Supported("mkdirat", Mkdirat), + 35: syscalls.Supported("unlinkat", Unlinkat), + 36: syscalls.Supported("symlinkat", Symlinkat), + 37: syscalls.Supported("linkat", Linkat), + 38: syscalls.Supported("renameat", Renameat), + 39: syscalls.PartiallySupported("umount2", Umount2, "Not all options or file systems are supported.", nil), + 40: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), + 41: syscalls.Error("pivot_root", syserror.EPERM, "", nil), + 42: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), + 44: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), + 46: syscalls.Supported("ftruncate", Ftruncate), + 47: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), + 48: syscalls.Supported("faccessat", Faccessat), + 49: syscalls.Supported("chdir", Chdir), + 50: syscalls.Supported("fchdir", Fchdir), + 51: syscalls.Supported("chroot", Chroot), + 52: syscalls.PartiallySupported("fchmod", Fchmod, "Options S_ISUID and S_ISGID not supported.", nil), + 53: syscalls.Supported("fchmodat", Fchmodat), + 54: syscalls.Supported("fchownat", Fchownat), + 55: syscalls.Supported("fchown", Fchown), + 56: syscalls.Supported("openat", Openat), + 57: syscalls.Supported("close", Close), + 58: syscalls.CapError("vhangup", linux.CAP_SYS_TTY_CONFIG, "", nil), + 59: syscalls.Supported("pipe2", Pipe2), + 60: syscalls.CapError("quotactl", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_admin for most operations + 61: syscalls.Supported("getdents64", Getdents64), + 62: syscalls.Supported("lseek", Lseek), + 63: syscalls.Supported("read", Read), + 64: syscalls.Supported("write", Write), + 65: syscalls.Supported("readv", Readv), + 66: syscalls.Supported("writev", Writev), + 67: syscalls.Supported("pread64", Pread64), + 68: syscalls.Supported("pwrite64", Pwrite64), + 69: syscalls.Supported("preadv", Preadv), + 70: syscalls.Supported("pwritev", Pwritev), + 71: syscalls.Supported("sendfile", Sendfile), + 72: syscalls.Supported("pselect", Pselect), + 73: syscalls.Supported("ppoll", Ppoll), + 74: syscalls.ErrorWithEvent("signalfd4", syserror.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426) + 75: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 76: syscalls.PartiallySupported("splice", Splice, "Stub implementation.", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 77: syscalls.ErrorWithEvent("tee", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 78: syscalls.Supported("readlinkat", Readlinkat), + 80: syscalls.Supported("fstat", Fstat), + 81: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), + 82: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), + 83: syscalls.PartiallySupported("fdatasync", Fdatasync, "Full data flush is not guaranteed at this time.", nil), + 84: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), + 85: syscalls.Supported("timerfd_create", TimerfdCreate), + 86: syscalls.Supported("timerfd_settime", TimerfdSettime), + 87: syscalls.Supported("timerfd_gettime", TimerfdGettime), + 88: syscalls.Supported("utimensat", Utimensat), + 89: syscalls.CapError("acct", linux.CAP_SYS_PACCT, "", nil), + 90: syscalls.Supported("capget", Capget), + 91: syscalls.Supported("capset", Capset), + 92: syscalls.ErrorWithEvent("personality", syserror.EINVAL, "Unable to change personality.", nil), + 93: syscalls.Supported("exit", Exit), + 94: syscalls.Supported("exit_group", ExitGroup), + 95: syscalls.Supported("waitid", Waitid), + 96: syscalls.Supported("set_tid_address", SetTidAddress), + 97: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), + 98: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), + 99: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 100: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 101: syscalls.Supported("nanosleep", Nanosleep), + 102: syscalls.Supported("getitimer", Getitimer), + 103: syscalls.Supported("setitimer", Setitimer), + 104: syscalls.CapError("kexec_load", linux.CAP_SYS_BOOT, "", nil), + 105: syscalls.CapError("init_module", linux.CAP_SYS_MODULE, "", nil), + 106: syscalls.CapError("delete_module", linux.CAP_SYS_MODULE, "", nil), + 107: syscalls.Supported("timer_create", TimerCreate), + 108: syscalls.Supported("timer_gettime", TimerGettime), + 109: syscalls.Supported("timer_getoverrun", TimerGetoverrun), + 110: syscalls.Supported("timer_settime", TimerSettime), + 111: syscalls.Supported("timer_delete", TimerDelete), + 112: syscalls.Supported("clock_settime", ClockSettime), + 113: syscalls.Supported("clock_gettime", ClockGettime), + 114: syscalls.Supported("clock_getres", ClockGetres), + 115: syscalls.Supported("clock_nanosleep", ClockNanosleep), + 116: syscalls.PartiallySupported("syslog", Syslog, "Outputs a dummy message for security reasons.", nil), + 117: syscalls.PartiallySupported("ptrace", Ptrace, "Options PTRACE_PEEKSIGINFO, PTRACE_SECCOMP_GET_FILTER not supported.", nil), + 118: syscalls.CapError("sched_setparam", linux.CAP_SYS_NICE, "", nil), + 119: syscalls.PartiallySupported("sched_setscheduler", SchedSetscheduler, "Stub implementation.", nil), + 120: syscalls.PartiallySupported("sched_getscheduler", SchedGetscheduler, "Stub implementation.", nil), + 121: syscalls.PartiallySupported("sched_getparam", SchedGetparam, "Stub implementation.", nil), + 122: syscalls.PartiallySupported("sched_setaffinity", SchedSetaffinity, "Stub implementation.", nil), + 123: syscalls.PartiallySupported("sched_getaffinity", SchedGetaffinity, "Stub implementation.", nil), + 124: syscalls.Supported("sched_yield", SchedYield), + 125: syscalls.PartiallySupported("sched_get_priority_max", SchedGetPriorityMax, "Stub implementation.", nil), + 126: syscalls.PartiallySupported("sched_get_priority_min", SchedGetPriorityMin, "Stub implementation.", nil), + 127: syscalls.ErrorWithEvent("sched_rr_get_interval", syserror.EPERM, "", nil), + 128: syscalls.Supported("restart_syscall", RestartSyscall), + 129: syscalls.Supported("kill", Kill), + 130: syscalls.Supported("tkill", Tkill), + 131: syscalls.Supported("tgkill", Tgkill), + 132: syscalls.Supported("sigaltstack", Sigaltstack), + 133: syscalls.Supported("rt_sigsuspend", RtSigsuspend), + 134: syscalls.Supported("rt_sigaction", RtSigaction), + 135: syscalls.Supported("rt_sigprocmask", RtSigprocmask), + 136: syscalls.Supported("rt_sigpending", RtSigpending), + 137: syscalls.Supported("rt_sigtimedwait", RtSigtimedwait), + 138: syscalls.Supported("rt_sigqueueinfo", RtSigqueueinfo), + 139: syscalls.Supported("rt_sigreturn", RtSigreturn), + 140: syscalls.PartiallySupported("setpriority", Setpriority, "Stub implementation.", nil), + 141: syscalls.PartiallySupported("getpriority", Getpriority, "Stub implementation.", nil), + 142: syscalls.CapError("reboot", linux.CAP_SYS_BOOT, "", nil), + 143: syscalls.Supported("setregid", Setregid), + 144: syscalls.Supported("setgid", Setgid), + 145: syscalls.Supported("setreuid", Setreuid), + 146: syscalls.Supported("setuid", Setuid), + 147: syscalls.Supported("setresuid", Setresuid), + 148: syscalls.Supported("getresuid", Getresuid), + 149: syscalls.Supported("setresgid", Setresgid), + 150: syscalls.Supported("getresgid", Getresgid), + 151: syscalls.ErrorWithEvent("setfsuid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 152: syscalls.ErrorWithEvent("setfsgid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 153: syscalls.Supported("times", Times), + 154: syscalls.Supported("setpgid", Setpgid), + 155: syscalls.Supported("getpgid", Getpgid), + 156: syscalls.Supported("getsid", Getsid), + 157: syscalls.Supported("setsid", Setsid), + 158: syscalls.Supported("getgroups", Getgroups), + 159: syscalls.Supported("setgroups", Setgroups), + 160: syscalls.Supported("uname", Uname), + 161: syscalls.Supported("sethostname", Sethostname), + 162: syscalls.Supported("setdomainname", Setdomainname), + 163: syscalls.Supported("getrlimit", Getrlimit), + 164: syscalls.PartiallySupported("setrlimit", Setrlimit, "Not all rlimits are enforced.", nil), + 165: syscalls.PartiallySupported("getrusage", Getrusage, "Fields ru_maxrss, ru_minflt, ru_majflt, ru_inblock, ru_oublock are not supported. Fields ru_utime and ru_stime have low precision.", nil), + 166: syscalls.Supported("umask", Umask), + 167: syscalls.PartiallySupported("prctl", Prctl, "Not all options are supported.", nil), + 168: syscalls.Supported("getcpu", Getcpu), + 169: syscalls.Supported("gettimeofday", Gettimeofday), + 170: syscalls.CapError("settimeofday", linux.CAP_SYS_TIME, "", nil), + 171: syscalls.CapError("adjtimex", linux.CAP_SYS_TIME, "", nil), + 172: syscalls.Supported("getpid", Getpid), + 173: syscalls.Supported("getppid", Getppid), + 174: syscalls.Supported("getuid", Getuid), + 175: syscalls.Supported("geteuid", Geteuid), + 176: syscalls.Supported("getgid", Getgid), + 177: syscalls.Supported("getegid", Getegid), + 178: syscalls.Supported("gettid", Gettid), + 179: syscalls.PartiallySupported("sysinfo", Sysinfo, "Fields loads, sharedram, bufferram, totalswap, freeswap, totalhigh, freehigh not supported.", nil), + 180: syscalls.ErrorWithEvent("mq_open", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 181: syscalls.ErrorWithEvent("mq_unlink", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 182: syscalls.ErrorWithEvent("mq_timedsend", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 183: syscalls.ErrorWithEvent("mq_timedreceive", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 184: syscalls.ErrorWithEvent("mq_notify", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 185: syscalls.ErrorWithEvent("mq_getsetattr", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 186: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 187: syscalls.ErrorWithEvent("msgctl", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 188: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 189: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 190: syscalls.Supported("semget", Semget), + 191: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), + 192: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) + 193: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), + 194: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), + 195: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), + 196: syscalls.PartiallySupported("shmat", Shmat, "Option SHM_RND is not supported.", nil), + 197: syscalls.Supported("shmdt", Shmdt), + 198: syscalls.PartiallySupported("socket", Socket, "Limited support for AF_NETLINK, NETLINK_ROUTE sockets. Limited support for SOCK_RAW.", nil), + 199: syscalls.Supported("socketpair", SocketPair), + 200: syscalls.PartiallySupported("bind", Bind, "Autobind for abstract Unix sockets is not supported.", nil), + 201: syscalls.Supported("listen", Listen), + 202: syscalls.Supported("accept", Accept), + 203: syscalls.Supported("connect", Connect), + 204: syscalls.Supported("getsockname", GetSockName), + 205: syscalls.Supported("getpeername", GetPeerName), + 206: syscalls.Supported("sendto", SendTo), + 207: syscalls.Supported("recvfrom", RecvFrom), + 208: syscalls.PartiallySupported("setsockopt", SetSockOpt, "Not all socket options are supported.", nil), + 209: syscalls.PartiallySupported("getsockopt", GetSockOpt, "Not all socket options are supported.", nil), + 210: syscalls.PartiallySupported("shutdown", Shutdown, "Not all flags and control messages are supported.", nil), + 211: syscalls.Supported("sendmsg", SendMsg), + 212: syscalls.PartiallySupported("recvmsg", RecvMsg, "Not all flags and control messages are supported.", nil), + 213: syscalls.ErrorWithEvent("readahead", syserror.ENOSYS, "", []string{"gvisor.dev/issue/261"}), // TODO(b/29351341) + 214: syscalls.Supported("brk", Brk), + 215: syscalls.Supported("munmap", Munmap), + 216: syscalls.Supported("mremap", Mremap), + 217: syscalls.Error("add_key", syserror.EACCES, "Not available to user.", nil), + 218: syscalls.Error("request_key", syserror.EACCES, "Not available to user.", nil), + 219: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), + 220: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), + 221: syscalls.Supported("execve", Execve), + 224: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), + 225: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), + 226: syscalls.Supported("mprotect", Mprotect), + 227: syscalls.PartiallySupported("msync", Msync, "Full data flush is not guaranteed at this time.", nil), + 228: syscalls.PartiallySupported("mlock", Mlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 229: syscalls.PartiallySupported("munlock", Munlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 230: syscalls.PartiallySupported("mlockall", Mlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 231: syscalls.PartiallySupported("munlockall", Munlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 232: syscalls.PartiallySupported("mincore", Mincore, "Stub implementation. The sandbox does not have access to this information. Reports all mapped pages are resident.", nil), + 233: syscalls.PartiallySupported("madvise", Madvise, "Options MADV_DONTNEED, MADV_DONTFORK are supported. Other advice is ignored.", nil), + 234: syscalls.ErrorWithEvent("remap_file_pages", syserror.ENOSYS, "Deprecated since Linux 3.16.", nil), + 235: syscalls.PartiallySupported("mbind", Mbind, "Stub implementation. Only a single NUMA node is advertised, and mempolicy is ignored accordingly, but mbind() will succeed and has effects reflected by get_mempolicy.", []string{"gvisor.dev/issue/262"}), + 236: syscalls.PartiallySupported("get_mempolicy", GetMempolicy, "Stub implementation.", nil), + 237: syscalls.PartiallySupported("set_mempolicy", SetMempolicy, "Stub implementation.", nil), + 238: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), + 239: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) + 240: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), + 241: syscalls.ErrorWithEvent("perf_event_open", syserror.ENODEV, "No support for perf counters", nil), + 242: syscalls.Supported("accept4", Accept4), + 243: syscalls.PartiallySupported("recvmmsg", RecvMMsg, "Not all flags and control messages are supported.", nil), + 260: syscalls.Supported("wait4", Wait4), + 261: syscalls.Supported("prlimit64", Prlimit64), + 262: syscalls.ErrorWithEvent("fanotify_init", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 263: syscalls.ErrorWithEvent("fanotify_mark", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 264: syscalls.Error("name_to_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 265: syscalls.Error("open_by_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 266: syscalls.CapError("clock_adjtime", linux.CAP_SYS_TIME, "", nil), + 267: syscalls.PartiallySupported("syncfs", Syncfs, "Depends on backing file system.", nil), + 268: syscalls.ErrorWithEvent("setns", syserror.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) + 269: syscalls.PartiallySupported("sendmmsg", SendMMsg, "Not all flags and control messages are supported.", nil), + 270: syscalls.ErrorWithEvent("process_vm_readv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 271: syscalls.ErrorWithEvent("process_vm_writev", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 272: syscalls.CapError("kcmp", linux.CAP_SYS_PTRACE, "", nil), + 273: syscalls.CapError("finit_module", linux.CAP_SYS_MODULE, "", nil), + 274: syscalls.ErrorWithEvent("sched_setattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 275: syscalls.ErrorWithEvent("sched_getattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 276: syscalls.ErrorWithEvent("renameat2", syserror.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) + 277: syscalls.Supported("seccomp", Seccomp), + 278: syscalls.Supported("getrandom", GetRandom), + 279: syscalls.Supported("memfd_create", MemfdCreate), + 280: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), + 281: syscalls.ErrorWithEvent("execveat", syserror.ENOSYS, "", []string{"gvisor.dev/issue/265"}), // TODO(b/118901836) + 282: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) + 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(b/118904897) + 284: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 285: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), + 286: syscalls.Supported("preadv2", Preadv2), + 287: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), + 291: syscalls.Supported("statx", Statx), + }, + Emulate: map[usermem.Addr]uintptr{}, + + Missing: func(t *kernel.Task, sysno uintptr, args arch.SyscallArguments) (uintptr, error) { + t.Kernel().EmitUnimplementedEvent(t) + return 0, syserror.ENOSYS + }, +} diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index 3ab54271c..cd31e0649 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -72,6 +72,39 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "read", file) } +// Readahead implements readahead(2). +func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + offset := args[1].Int64() + size := args[2].SizeT() + + file := t.GetFile(fd) + if file == nil { + return 0, nil, syserror.EBADF + } + defer file.DecRef() + + // Check that the file is readable. + if !file.Flags().Read { + return 0, nil, syserror.EBADF + } + + // Check that the size is valid. + if int(size) < 0 { + return 0, nil, syserror.EINVAL + } + + // Check that the offset is legitimate. + if offset < 0 { + return 0, nil, syserror.EINVAL + } + + // Return EINVAL; if the underlying file type does not support readahead, + // then Linux will return EINVAL to indicate as much. In the future, we + // may extend this function to actually support readahead hints. + return 0, nil, syserror.EINVAL +} + // Pread64 implements linux syscall pread64(2). func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 3bac4d90d..b5a72ce63 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -531,7 +531,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, syserror.ENOTSOCK } - if optLen <= 0 { + if optLen < 0 { return 0, nil, syserror.EINVAL } if optLen > maxOptLen { diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index f0a292f2f..dd3a5807f 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -159,9 +159,14 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc }, outFile.Flags().NonBlocking) } + // Sendfile can't lose any data because inFD is always a regual file. + if n != 0 { + err = nil + } + // We can only pass a single file to handleIOError, so pick inFile // arbitrarily. This is used only for debugging purposes. - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "sendfile", inFile) + return uintptr(n), nil, handleIOError(t, false, err, kernel.ERESTARTSYS, "sendfile", inFile) } // Splice implements splice(2). @@ -245,12 +250,12 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if inOffset != 0 || outOffset != 0 { return 0, nil, syserror.ESPIPE } - default: - return 0, nil, syserror.EINVAL - } - // We may not refer to the same pipe; otherwise it's a continuous loop. - if inFile.Dirent.Inode.StableAttr.InodeID == outFile.Dirent.Inode.StableAttr.InodeID { + // We may not refer to the same pipe; otherwise it's a continuous loop. + if inFile.Dirent.Inode.StableAttr.InodeID == outFile.Dirent.Inode.StableAttr.InodeID { + return 0, nil, syserror.EINVAL + } + default: return 0, nil, syserror.EINVAL } @@ -305,6 +310,11 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo Dup: true, }, nonBlock) + // Tee doesn't change a state of inFD, so it can't lose any data. + if n != 0 { + err = nil + } + // See above; inFile is chosen arbitrarily here. - return uintptr(n), nil, handleIOError(t, n != 0, err, kernel.ERESTARTSYS, "tee", inFile) + return uintptr(n), nil, handleIOError(t, false, err, kernel.ERESTARTSYS, "tee", inFile) } diff --git a/pkg/sentry/syscalls/linux/sys_time.go b/pkg/sentry/syscalls/linux/sys_time.go index 4b3f043a2..b887fa9d7 100644 --- a/pkg/sentry/syscalls/linux/sys_time.go +++ b/pkg/sentry/syscalls/linux/sys_time.go @@ -15,6 +15,7 @@ package linux import ( + "fmt" "time" "gvisor.dev/gvisor/pkg/abi/linux" @@ -228,41 +229,35 @@ func clockNanosleepFor(t *kernel.Task, c ktime.Clock, dur time.Duration, rem use timer.Destroy() - var remaining time.Duration - // Did we just block for the entire duration? - if err == syserror.ETIMEDOUT { - remaining = 0 - } else { - remaining = dur - after.Sub(start) + switch err { + case syserror.ETIMEDOUT: + // Slept for entire timeout. + return nil + case syserror.ErrInterrupted: + // Interrupted. + remaining := dur - after.Sub(start) if remaining < 0 { remaining = time.Duration(0) } - } - // Copy out remaining time. - if err != nil && rem != usermem.Addr(0) { - timeleft := linux.NsecToTimespec(remaining.Nanoseconds()) - if err := copyTimespecOut(t, rem, &timeleft); err != nil { - return err + // Copy out remaining time. + if rem != 0 { + timeleft := linux.NsecToTimespec(remaining.Nanoseconds()) + if err := copyTimespecOut(t, rem, &timeleft); err != nil { + return err + } } - } - - // Did we just block for the entire duration? - if err == syserror.ETIMEDOUT { - return nil - } - // If interrupted, arrange for a restart with the remaining duration. - if err == syserror.ErrInterrupted { + // Arrange for a restart with the remaining duration. t.SetSyscallRestartBlock(&clockNanosleepRestartBlock{ c: c, duration: remaining, rem: rem, }) return kernel.ERESTART_RESTARTBLOCK + default: + panic(fmt.Sprintf("Impossible BlockWithTimer error %v", err)) } - - return err } // Nanosleep implements linux syscall Nanosleep(2). diff --git a/pkg/sentry/syscalls/linux/sys_utsname.go b/pkg/sentry/syscalls/linux/sys_utsname.go index 271ace08e..748e8dd8d 100644 --- a/pkg/sentry/syscalls/linux/sys_utsname.go +++ b/pkg/sentry/syscalls/linux/sys_utsname.go @@ -79,11 +79,11 @@ func Sethostname(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, syserror.EINVAL } - name, err := t.CopyInString(nameAddr, int(size)) - if err != nil { + name := make([]byte, size) + if _, err := t.CopyInBytes(nameAddr, name); err != nil { return 0, nil, err } - utsns.SetHostName(name) + utsns.SetHostName(string(name)) return 0, nil, nil } diff --git a/pkg/sentry/usermem/usermem.go b/pkg/sentry/usermem/usermem.go index 6eced660a..7b1f312b1 100644 --- a/pkg/sentry/usermem/usermem.go +++ b/pkg/sentry/usermem/usermem.go @@ -16,6 +16,7 @@ package usermem import ( + "bytes" "errors" "io" "strconv" @@ -270,11 +271,10 @@ func CopyStringIn(ctx context.Context, uio IO, addr Addr, maxlen int, opts IOOpt n, err := uio.CopyIn(ctx, addr, buf[done:done+readlen], opts) // Look for the terminating zero byte, which may have occurred before // hitting err. - for i, c := range buf[done : done+n] { - if c == 0 { - return stringFromImmutableBytes(buf[:done+i]), nil - } + if i := bytes.IndexByte(buf[done:done+n], byte(0)); i >= 0 { + return stringFromImmutableBytes(buf[:done+i]), nil } + done += n if err != nil { return stringFromImmutableBytes(buf[:done]), err diff --git a/pkg/sentry/vfs/options.go b/pkg/sentry/vfs/options.go index 187e5410c..3aa73d911 100644 --- a/pkg/sentry/vfs/options.go +++ b/pkg/sentry/vfs/options.go @@ -31,14 +31,14 @@ type GetDentryOptions struct { // FilesystemImpl.MkdirAt(). type MkdirOptions struct { // Mode is the file mode bits for the created directory. - Mode uint16 + Mode linux.FileMode } // MknodOptions contains options to VirtualFilesystem.MknodAt() and // FilesystemImpl.MknodAt(). type MknodOptions struct { // Mode is the file type and mode bits for the created file. - Mode uint16 + Mode linux.FileMode // If Mode specifies a character or block device special file, DevMajor and // DevMinor are the major and minor device numbers for the created device. @@ -61,7 +61,7 @@ type OpenOptions struct { // If FilesystemImpl.OpenAt() creates a file, Mode is the file mode for the // created file. - Mode uint16 + Mode linux.FileMode } // ReadOptions contains options to FileDescription.PRead(), diff --git a/pkg/tcpip/adapters/gonet/gonet_test.go b/pkg/tcpip/adapters/gonet/gonet_test.go index 672f026b2..8ced960bb 100644 --- a/pkg/tcpip/adapters/gonet/gonet_test.go +++ b/pkg/tcpip/adapters/gonet/gonet_test.go @@ -60,7 +60,10 @@ func TestTimeouts(t *testing.T) { func newLoopbackStack() (*stack.Stack, *tcpip.Error) { // Create the stack and add a NIC. - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName}, []string{tcp.ProtocolName, udp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol(), udp.NewProtocol()}, + }) if err := s.CreateNIC(NICID, loopback.New()); err != nil { return nil, err diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go index 584db710e..7636418b1 100644 --- a/pkg/tcpip/link/fdbased/endpoint.go +++ b/pkg/tcpip/link/fdbased/endpoint.go @@ -82,7 +82,19 @@ const ( PacketMMap ) -// An endpoint implements the link-layer using a message-oriented file descriptor. +func (p PacketDispatchMode) String() string { + switch p { + case Readv: + return "Readv" + case RecvMMsg: + return "RecvMMsg" + case PacketMMap: + return "PacketMMap" + default: + return fmt.Sprintf("unknown packet dispatch mode %v", p) + } +} + type endpoint struct { // fds is the set of file descriptors each identifying one inbound/outbound // channel. The endpoint will dispatch from all inbound channels as well as diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go index fd6395fc1..26cf1c528 100644 --- a/pkg/tcpip/network/arp/arp.go +++ b/pkg/tcpip/network/arp/arp.go @@ -16,9 +16,9 @@ // IPv4 addresses into link-local MAC addresses, and advertises IPv4 // addresses of its stack with the local network. // -// To use it in the networking stack, pass arp.ProtocolName as one of the -// network protocols when calling stack.New. Then add an "arp" address to -// every NIC on the stack that should respond to ARP requests. That is: +// To use it in the networking stack, pass arp.NewProtocol() as one of the +// network protocols when calling stack.New. Then add an "arp" address to every +// NIC on the stack that should respond to ARP requests. That is: // // if err := s.AddAddress(1, arp.ProtocolNumber, "arp"); err != nil { // // handle err @@ -33,9 +33,6 @@ import ( ) const ( - // ProtocolName is the string representation of the ARP protocol name. - ProtocolName = "arp" - // ProtocolNumber is the ARP protocol number. ProtocolNumber = header.ARPProtocolNumber @@ -200,8 +197,7 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { var broadcastMAC = tcpip.LinkAddress([]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff}) -func init() { - stack.RegisterNetworkProtocolFactory(ProtocolName, func() stack.NetworkProtocol { - return &protocol{} - }) +// NewProtocol returns an ARP network protocol. +func NewProtocol() stack.NetworkProtocol { + return &protocol{} } diff --git a/pkg/tcpip/network/arp/arp_test.go b/pkg/tcpip/network/arp/arp_test.go index 387fca96e..88b57ec03 100644 --- a/pkg/tcpip/network/arp/arp_test.go +++ b/pkg/tcpip/network/arp/arp_test.go @@ -44,7 +44,10 @@ type testContext struct { } func newTestContext(t *testing.T) *testContext { - s := stack.New([]string{ipv4.ProtocolName, arp.ProtocolName}, []string{icmp.ProtocolName4}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), arp.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol4()}, + }) const defaultMTU = 65536 ep := channel.New(256, defaultMTU, stackLinkAddr) diff --git a/pkg/tcpip/network/ip_test.go b/pkg/tcpip/network/ip_test.go index 6a40e7ee3..a9741622e 100644 --- a/pkg/tcpip/network/ip_test.go +++ b/pkg/tcpip/network/ip_test.go @@ -172,7 +172,10 @@ func (t *testObject) WritePacket(_ *stack.Route, _ *stack.GSO, hdr buffer.Prepen } func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { - s := stack.New([]string{ipv4.ProtocolName}, []string{udp.ProtocolName, tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + }) s.CreateNIC(1, loopback.New()) s.AddAddress(1, ipv4.ProtocolNumber, local) s.SetRouteTable([]tcpip.Route{{ @@ -185,7 +188,10 @@ func buildIPv4Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { } func buildIPv6Route(local, remote tcpip.Address) (stack.Route, *tcpip.Error) { - s := stack.New([]string{ipv6.ProtocolName}, []string{udp.ProtocolName, tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol(), tcp.NewProtocol()}, + }) s.CreateNIC(1, loopback.New()) s.AddAddress(1, ipv6.ProtocolNumber, local) s.SetRouteTable([]tcpip.Route{{ diff --git a/pkg/tcpip/network/ipv4/icmp.go b/pkg/tcpip/network/ipv4/icmp.go index a25756443..c1cf6c222 100644 --- a/pkg/tcpip/network/ipv4/icmp.go +++ b/pkg/tcpip/network/ipv4/icmp.go @@ -95,7 +95,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V pkt.SetChecksum(0) pkt.SetChecksum(^header.Checksum(pkt, header.ChecksumVV(vv, 0))) sent := stats.ICMP.V4PacketsSent - if err := r.WritePacket(nil /* gso */, hdr, vv, header.ICMPv4ProtocolNumber, r.DefaultTTL()); err != nil { + if err := r.WritePacket(nil /* gso */, hdr, vv, header.ICMPv4ProtocolNumber, 0, true /* useDefaultTTL */); err != nil { sent.Dropped.Increment() return } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index b7a06f525..162aa1b4d 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -14,9 +14,9 @@ // Package ipv4 contains the implementation of the ipv4 network protocol. To use // it in the networking stack, this package must be added to the project, and -// activated on the stack by passing ipv4.ProtocolName (or "ipv4") as one of the -// network protocols when calling stack.New(). Then endpoints can be created -// by passing ipv4.ProtocolNumber as the network protocol number when calling +// activated on the stack by passing ipv4.NewProtocol() as one of the network +// protocols when calling stack.New(). Then endpoints can be created by passing +// ipv4.ProtocolNumber as the network protocol number when calling // Stack.NewEndpoint(). package ipv4 @@ -32,9 +32,6 @@ import ( ) const ( - // ProtocolName is the string representation of the ipv4 protocol name. - ProtocolName = "ipv4" - // ProtocolNumber is the ipv4 protocol number. ProtocolNumber = header.IPv4ProtocolNumber @@ -42,6 +39,9 @@ const ( // TotalLength field of the ipv4 header. MaxTotalSize = 0xffff + // DefaultTTL is the default time-to-live value for this endpoint. + DefaultTTL = 64 + // buckets is the number of identifier buckets. buckets = 2048 ) @@ -53,6 +53,7 @@ type endpoint struct { linkEP stack.LinkEndpoint dispatcher stack.TransportDispatcher fragmentation *fragmentation.Fragmentation + protocol *protocol } // NewEndpoint creates a new ipv4 endpoint. @@ -64,6 +65,7 @@ func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWi linkEP: linkEP, dispatcher: dispatcher, fragmentation: fragmentation.NewFragmentation(fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout), + protocol: p, } return e, nil @@ -71,7 +73,7 @@ func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWi // DefaultTTL is the default time-to-live value for this endpoint. func (e *endpoint) DefaultTTL() uint8 { - return 255 + return e.protocol.DefaultTTL() } // MTU implements stack.NetworkEndpoint.MTU. It returns the link-layer MTU minus @@ -204,7 +206,7 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prepen if length > header.IPv4MaximumHeaderSize+8 { // Packets of 68 bytes or less are required by RFC 791 to not be // fragmented, so we only assign ids to larger packets. - id = atomic.AddUint32(&ids[hashRoute(r, protocol)%buckets], 1) + id = atomic.AddUint32(&e.protocol.ids[hashRoute(r, protocol, e.protocol.hashIV)%buckets], 1) } ip.Encode(&header.IPv4Fields{ IHL: header.IPv4MinimumSize, @@ -267,7 +269,7 @@ func (e *endpoint) WriteHeaderIncludedPacket(r *stack.Route, payload buffer.Vect if payload.Size() > header.IPv4MaximumHeaderSize+8 { // Packets of 68 bytes or less are required by RFC 791 to not be // fragmented, so we only assign ids to larger packets. - id = atomic.AddUint32(&ids[hashRoute(r, 0 /* protocol */)%buckets], 1) + id = atomic.AddUint32(&e.protocol.ids[hashRoute(r, 0 /* protocol */, e.protocol.hashIV)%buckets], 1) } ip.SetID(uint16(id)) } @@ -325,14 +327,14 @@ func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { // Close cleans up resources associated with the endpoint. func (e *endpoint) Close() {} -type protocol struct{} +type protocol struct { + ids []uint32 + hashIV uint32 -// NewProtocol creates a new protocol ipv4 protocol descriptor. This is exported -// only for tests that short-circuit the stack. Regular use of the protocol is -// done via the stack, which gets a protocol descriptor from the init() function -// below. -func NewProtocol() stack.NetworkProtocol { - return &protocol{} + // defaultTTL is the current default TTL for the protocol. Only the + // uint8 portion of it is meaningful and it must be accessed + // atomically. + defaultTTL uint32 } // Number returns the ipv4 protocol number. @@ -358,12 +360,34 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) { // SetOption implements NetworkProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch v := option.(type) { + case tcpip.DefaultTTLOption: + p.SetDefaultTTL(uint8(v)) + return nil + default: + return tcpip.ErrUnknownProtocolOption + } } // Option implements NetworkProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch v := option.(type) { + case *tcpip.DefaultTTLOption: + *v = tcpip.DefaultTTLOption(p.DefaultTTL()) + return nil + default: + return tcpip.ErrUnknownProtocolOption + } +} + +// SetDefaultTTL sets the default TTL for endpoints created with this protocol. +func (p *protocol) SetDefaultTTL(ttl uint8) { + atomic.StoreUint32(&p.defaultTTL, uint32(ttl)) +} + +// DefaultTTL returns the default TTL for endpoints created with this protocol. +func (p *protocol) DefaultTTL() uint8 { + return uint8(atomic.LoadUint32(&p.defaultTTL)) } // calculateMTU calculates the network-layer payload MTU based on the link-layer @@ -378,7 +402,7 @@ func calculateMTU(mtu uint32) uint32 { // hashRoute calculates a hash value for the given route. It uses the source & // destination address, the transport protocol number, and a random initial // value (generated once on initialization) to generate the hash. -func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber) uint32 { +func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber, hashIV uint32) uint32 { t := r.LocalAddress a := uint32(t[0]) | uint32(t[1])<<8 | uint32(t[2])<<16 | uint32(t[3])<<24 t = r.RemoteAddress @@ -386,22 +410,16 @@ func hashRoute(r *stack.Route, protocol tcpip.TransportProtocolNumber) uint32 { return hash.Hash3Words(a, b, uint32(protocol), hashIV) } -var ( - ids []uint32 - hashIV uint32 -) - -func init() { - ids = make([]uint32, buckets) +// NewProtocol returns an IPv4 network protocol. +func NewProtocol() stack.NetworkProtocol { + ids := make([]uint32, buckets) // Randomly initialize hashIV and the ids. r := hash.RandN32(1 + buckets) for i := range ids { ids[i] = r[i] } - hashIV = r[buckets] + hashIV := r[buckets] - stack.RegisterNetworkProtocolFactory(ProtocolName, func() stack.NetworkProtocol { - return &protocol{} - }) + return &protocol{ids: ids, hashIV: hashIV, defaultTTL: DefaultTTL} } diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index ae827ca27..8b7500095 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -33,7 +33,10 @@ import ( ) func TestExcludeBroadcast(t *testing.T) { - s := stack.New([]string{ipv4.ProtocolName}, []string{udp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + }) const defaultMTU = 65536 ep := stack.LinkEndpoint(channel.New(256, defaultMTU, "")) @@ -44,9 +47,6 @@ func TestExcludeBroadcast(t *testing.T) { t.Fatalf("CreateNIC failed: %v", err) } - if err := s.AddAddress(1, ipv4.ProtocolNumber, header.IPv4Broadcast); err != nil { - t.Fatalf("AddAddress failed: %v", err) - } if err := s.AddAddress(1, ipv4.ProtocolNumber, header.IPv4Any); err != nil { t.Fatalf("AddAddress failed: %v", err) } @@ -238,7 +238,9 @@ type context struct { func buildContext(t *testing.T, packetCollectorErrors []*tcpip.Error, mtu uint32) context { // Make the packet and write it. - s := stack.New([]string{ipv4.ProtocolName}, []string{}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + }) ep := newErrorChannel(100 /* Enough for all tests. */, mtu, "", packetCollectorErrors) s.CreateNIC(1, ep) const ( @@ -300,7 +302,7 @@ func TestFragmentation(t *testing.T) { Payload: payload.Clone([]buffer.View{}), } c := buildContext(t, nil, ft.mtu) - err := c.Route.WritePacket(ft.gso, hdr, payload, tcp.ProtocolNumber, 42) + err := c.Route.WritePacket(ft.gso, hdr, payload, tcp.ProtocolNumber, 42 /* ttl */, false /* useDefaultTTL */) if err != nil { t.Errorf("err got %v, want %v", err, nil) } @@ -347,7 +349,7 @@ func TestFragmentationErrors(t *testing.T) { t.Run(ft.description, func(t *testing.T) { hdr, payload := makeHdrAndPayload(ft.hdrLength, header.IPv4MinimumSize, ft.payloadViewsSizes) c := buildContext(t, ft.packetCollectorErrors, ft.mtu) - err := c.Route.WritePacket(&stack.GSO{}, hdr, payload, tcp.ProtocolNumber, 42) + err := c.Route.WritePacket(&stack.GSO{}, hdr, payload, tcp.ProtocolNumber, 42 /* ttl */, false /* useDefaultTTL */) for i := 0; i < len(ft.packetCollectorErrors)-1; i++ { if got, want := ft.packetCollectorErrors[i], (*tcpip.Error)(nil); got != want { t.Errorf("ft.packetCollectorErrors[%d] got %v, want %v", i, got, want) diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index b4d0295bf..71c398027 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -154,7 +154,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V r.LocalAddress = targetAddr pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) - if err := r.WritePacket(nil /* gso */, hdr, buffer.VectorisedView{}, header.ICMPv6ProtocolNumber, r.DefaultTTL()); err != nil { + if err := r.WritePacket(nil /* gso */, hdr, buffer.VectorisedView{}, header.ICMPv6ProtocolNumber, 0, true /* useDefaultTTL */); err != nil { sent.Dropped.Increment() return } @@ -185,7 +185,7 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, vv buffer.V copy(pkt, h) pkt.SetType(header.ICMPv6EchoReply) pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, vv)) - if err := r.WritePacket(nil /* gso */, hdr, vv, header.ICMPv6ProtocolNumber, r.DefaultTTL()); err != nil { + if err := r.WritePacket(nil /* gso */, hdr, vv, header.ICMPv6ProtocolNumber, 0, true /* useDefaultTTL */); err != nil { sent.Dropped.Increment() return } @@ -262,7 +262,7 @@ func (*protocol) LinkAddressRequest(addr, localAddr tcpip.Address, linkEP stack. ip.Encode(&header.IPv6Fields{ PayloadLength: length, NextHeader: uint8(header.ICMPv6ProtocolNumber), - HopLimit: defaultIPv6HopLimit, + HopLimit: ndpHopLimit, SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index 653d984e9..501be208e 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -81,7 +81,10 @@ func (*stubLinkAddressCache) AddLinkAddress(tcpip.NICID, tcpip.Address, tcpip.Li } func TestICMPCounts(t *testing.T) { - s := stack.New([]string{ProtocolName}, []string{icmp.ProtocolName6}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + }) { if err := s.CreateNIC(1, &stubLinkEndpoint{}); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -141,7 +144,7 @@ func TestICMPCounts(t *testing.T) { ip.Encode(&header.IPv6Fields{ PayloadLength: uint16(payloadLength), NextHeader: uint8(header.ICMPv6ProtocolNumber), - HopLimit: r.DefaultTTL(), + HopLimit: ndpHopLimit, SrcAddr: r.LocalAddress, DstAddr: r.RemoteAddress, }) @@ -205,8 +208,14 @@ func (e endpointWithResolutionCapability) Capabilities() stack.LinkEndpointCapab func newTestContext(t *testing.T) *testContext { c := &testContext{ - s0: stack.New([]string{ProtocolName}, []string{icmp.ProtocolName6}, stack.Options{}), - s1: stack.New([]string{ProtocolName}, []string{icmp.ProtocolName6}, stack.Options{}), + s0: stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + }), + s1: stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + }), } const defaultMTU = 65536 diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 331a8bdaa..85c070c43 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -14,13 +14,15 @@ // Package ipv6 contains the implementation of the ipv6 network protocol. To use // it in the networking stack, this package must be added to the project, and -// activated on the stack by passing ipv6.ProtocolName (or "ipv6") as one of the -// network protocols when calling stack.New(). Then endpoints can be created -// by passing ipv6.ProtocolNumber as the network protocol number when calling +// activated on the stack by passing ipv6.NewProtocol() as one of the network +// protocols when calling stack.New(). Then endpoints can be created by passing +// ipv6.ProtocolNumber as the network protocol number when calling // Stack.NewEndpoint(). package ipv6 import ( + "sync/atomic" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" @@ -28,9 +30,6 @@ import ( ) const ( - // ProtocolName is the string representation of the ipv6 protocol name. - ProtocolName = "ipv6" - // ProtocolNumber is the ipv6 protocol number. ProtocolNumber = header.IPv6ProtocolNumber @@ -38,9 +37,9 @@ const ( // PayloadLength field of the ipv6 header. maxPayloadSize = 0xffff - // defaultIPv6HopLimit is the default hop limit for IPv6 Packets - // egressed by Netstack. - defaultIPv6HopLimit = 255 + // DefaultTTL is the default hop limit for IPv6 Packets egressed by + // Netstack. + DefaultTTL = 64 ) type endpoint struct { @@ -50,11 +49,12 @@ type endpoint struct { linkEP stack.LinkEndpoint linkAddrCache stack.LinkAddressCache dispatcher stack.TransportDispatcher + protocol *protocol } // DefaultTTL is the default hop limit for this endpoint. func (e *endpoint) DefaultTTL() uint8 { - return 255 + return e.protocol.DefaultTTL() } // MTU implements stack.NetworkEndpoint.MTU. It returns the link-layer MTU minus @@ -158,14 +158,11 @@ func (e *endpoint) HandlePacket(r *stack.Route, vv buffer.VectorisedView) { // Close cleans up resources associated with the endpoint. func (*endpoint) Close() {} -type protocol struct{} - -// NewProtocol creates a new protocol ipv6 protocol descriptor. This is exported -// only for tests that short-circuit the stack. Regular use of the protocol is -// done via the stack, which gets a protocol descriptor from the init() function -// below. -func NewProtocol() stack.NetworkProtocol { - return &protocol{} +type protocol struct { + // defaultTTL is the current default TTL for the protocol. Only the + // uint8 portion of it is meaningful and it must be accessed + // atomically. + defaultTTL uint32 } // Number returns the ipv6 protocol number. @@ -198,17 +195,40 @@ func (p *protocol) NewEndpoint(nicid tcpip.NICID, addrWithPrefix tcpip.AddressWi linkEP: linkEP, linkAddrCache: linkAddrCache, dispatcher: dispatcher, + protocol: p, }, nil } // SetOption implements NetworkProtocol.SetOption. func (p *protocol) SetOption(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch v := option.(type) { + case tcpip.DefaultTTLOption: + p.SetDefaultTTL(uint8(v)) + return nil + default: + return tcpip.ErrUnknownProtocolOption + } } // Option implements NetworkProtocol.Option. func (p *protocol) Option(option interface{}) *tcpip.Error { - return tcpip.ErrUnknownProtocolOption + switch v := option.(type) { + case *tcpip.DefaultTTLOption: + *v = tcpip.DefaultTTLOption(p.DefaultTTL()) + return nil + default: + return tcpip.ErrUnknownProtocolOption + } +} + +// SetDefaultTTL sets the default TTL for endpoints created with this protocol. +func (p *protocol) SetDefaultTTL(ttl uint8) { + atomic.StoreUint32(&p.defaultTTL, uint32(ttl)) +} + +// DefaultTTL returns the default TTL for endpoints created with this protocol. +func (p *protocol) DefaultTTL() uint8 { + return uint8(atomic.LoadUint32(&p.defaultTTL)) } // calculateMTU calculates the network-layer payload MTU based on the link-layer @@ -221,8 +241,7 @@ func calculateMTU(mtu uint32) uint32 { return maxPayloadSize } -func init() { - stack.RegisterNetworkProtocolFactory(ProtocolName, func() stack.NetworkProtocol { - return &protocol{} - }) +// NewProtocol returns an IPv6 network protocol. +func NewProtocol() stack.NetworkProtocol { + return &protocol{defaultTTL: DefaultTTL} } diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 57bcd5455..78c674c2c 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -124,17 +124,20 @@ func testReceiveUDP(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst // UDP packets destined to the IPv6 link-local all-nodes multicast address. func TestReceiveOnAllNodesMulticastAddr(t *testing.T) { tests := []struct { - name string - protocolName string - rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) + name string + protocolFactory stack.TransportProtocol + rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) }{ - {"ICMP", icmp.ProtocolName6, testReceiveICMP}, - {"UDP", udp.ProtocolName, testReceiveUDP}, + {"ICMP", icmp.NewProtocol6(), testReceiveICMP}, + {"UDP", udp.NewProtocol(), testReceiveUDP}, } for _, test := range tests { t.Run(test.name, func(t *testing.T) { - s := stack.New([]string{ProtocolName}, []string{test.protocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{test.protocolFactory}, + }) e := channel.New(10, 1280, linkAddr1) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -152,19 +155,22 @@ func TestReceiveOnAllNodesMulticastAddr(t *testing.T) { // address. func TestReceiveOnSolicitedNodeAddr(t *testing.T) { tests := []struct { - name string - protocolName string - rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) + name string + protocolFactory stack.TransportProtocol + rxf func(t *testing.T, s *stack.Stack, e *channel.Endpoint, src, dst tcpip.Address, want uint64) }{ - {"ICMP", icmp.ProtocolName6, testReceiveICMP}, - {"UDP", udp.ProtocolName, testReceiveUDP}, + {"ICMP", icmp.NewProtocol6(), testReceiveICMP}, + {"UDP", udp.NewProtocol(), testReceiveUDP}, } snmc := header.SolicitedNodeAddr(addr2) for _, test := range tests { t.Run(test.name, func(t *testing.T) { - s := stack.New([]string{ProtocolName}, []string{test.protocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{test.protocolFactory}, + }) e := channel.New(10, 1280, linkAddr1) if err := s.CreateNIC(1, e); err != nil { t.Fatalf("CreateNIC(_) = %s", err) @@ -237,7 +243,9 @@ func TestAddIpv6Address(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { - s := stack.New([]string{ProtocolName}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + }) if err := s.CreateNIC(1, &stubLinkEndpoint{}); err != nil { t.Fatalf("CreateNIC(_) = %s", err) } diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index 571915d3f..e30791fe3 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -31,7 +31,10 @@ import ( func setupStackAndEndpoint(t *testing.T, llladdr, rlladdr tcpip.Address) (*stack.Stack, stack.NetworkEndpoint) { t.Helper() - s := stack.New([]string{ProtocolName}, []string{icmp.ProtocolName6}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{icmp.NewProtocol6()}, + }) if err := s.CreateNIC(1, &stubLinkEndpoint{}); err != nil { t.Fatalf("CreateNIC(_) = %s", err) diff --git a/pkg/tcpip/ports/ports.go b/pkg/tcpip/ports/ports.go index 315780c0c..30cea8996 100644 --- a/pkg/tcpip/ports/ports.go +++ b/pkg/tcpip/ports/ports.go @@ -19,6 +19,7 @@ import ( "math" "math/rand" "sync" + "sync/atomic" "gvisor.dev/gvisor/pkg/tcpip" ) @@ -27,6 +28,10 @@ const ( // FirstEphemeral is the first ephemeral port. FirstEphemeral = 16000 + // numEphemeralPorts it the mnumber of available ephemeral ports to + // Netstack. + numEphemeralPorts = math.MaxUint16 - FirstEphemeral + 1 + anyIPAddress tcpip.Address = "" ) @@ -40,6 +45,13 @@ type portDescriptor struct { type PortManager struct { mu sync.RWMutex allocatedPorts map[portDescriptor]bindAddresses + + // hint is used to pick ports ephemeral ports in a stable order for + // a given port offset. + // + // hint must be accessed using the portHint/incPortHint helpers. + // TODO(gvisor.dev/issue/940): S/R this field. + hint uint32 } type portNode struct { @@ -47,43 +59,76 @@ type portNode struct { refs int } -// bindAddresses is a set of IP addresses. -type bindAddresses map[tcpip.Address]portNode +// deviceNode is never empty. When it has no elements, it is removed from the +// map that references it. +type deviceNode map[tcpip.NICID]portNode -// isAvailable checks whether an IP address is available to bind to. -func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool) bool { - if addr == anyIPAddress { - if len(b) == 0 { - return true - } +// isAvailable checks whether binding is possible by device. If not binding to a +// device, check against all portNodes. If binding to a specific device, check +// against the unspecified device and the provided device. +func (d deviceNode) isAvailable(reuse bool, bindToDevice tcpip.NICID) bool { + if bindToDevice == 0 { + // Trying to binding all devices. if !reuse { + // Can't bind because the (addr,port) is already bound. return false } - for _, n := range b { - if !n.reuse { + for _, p := range d { + if !p.reuse { + // Can't bind because the (addr,port) was previously bound without reuse. return false } } return true } - // If all addresses for this portDescriptor are already bound, no - // address is available. - if n, ok := b[anyIPAddress]; ok { - if !reuse { + if p, ok := d[0]; ok { + if !reuse || !p.reuse { return false } - if !n.reuse { + } + + if p, ok := d[bindToDevice]; ok { + if !reuse || !p.reuse { return false } } - if n, ok := b[addr]; ok { - if !reuse { + return true +} + +// bindAddresses is a set of IP addresses. +type bindAddresses map[tcpip.Address]deviceNode + +// isAvailable checks whether an IP address is available to bind to. If the +// address is the "any" address, check all other addresses. Otherwise, just +// check against the "any" address and the provided address. +func (b bindAddresses) isAvailable(addr tcpip.Address, reuse bool, bindToDevice tcpip.NICID) bool { + if addr == anyIPAddress { + // If binding to the "any" address then check that there are no conflicts + // with all addresses. + for _, d := range b { + if !d.isAvailable(reuse, bindToDevice) { + return false + } + } + return true + } + + // Check that there is no conflict with the "any" address. + if d, ok := b[anyIPAddress]; ok { + if !d.isAvailable(reuse, bindToDevice) { return false } - return n.reuse } + + // Check that this is no conflict with the provided address. + if d, ok := b[addr]; ok { + if !d.isAvailable(reuse, bindToDevice) { + return false + } + } + return true } @@ -97,11 +142,40 @@ func NewPortManager() *PortManager { // is suitable for its needs, and stopping when a port is found or an error // occurs. func (s *PortManager) PickEphemeralPort(testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { - count := uint16(math.MaxUint16 - FirstEphemeral + 1) - offset := uint16(rand.Int31n(int32(count))) + offset := uint32(rand.Int31n(numEphemeralPorts)) + return s.pickEphemeralPort(offset, numEphemeralPorts, testPort) +} + +// portHint atomically reads and returns the s.hint value. +func (s *PortManager) portHint() uint32 { + return atomic.LoadUint32(&s.hint) +} + +// incPortHint atomically increments s.hint by 1. +func (s *PortManager) incPortHint() { + atomic.AddUint32(&s.hint, 1) +} - for i := uint16(0); i < count; i++ { - port = FirstEphemeral + (offset+i)%count +// PickEphemeralPortStable starts at the specified offset + s.portHint and +// iterates over all ephemeral ports, allowing the caller to decide whether a +// given port is suitable for its needs and stopping when a port is found or an +// error occurs. +func (s *PortManager) PickEphemeralPortStable(offset uint32, testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { + p, err := s.pickEphemeralPort(s.portHint()+offset, numEphemeralPorts, testPort) + if err == nil { + s.incPortHint() + } + return p, err + +} + +// pickEphemeralPort starts at the offset specified from the FirstEphemeral port +// and iterates over the number of ports specified by count and allows the +// caller to decide whether a given port is suitable for its needs, and stopping +// when a port is found or an error occurs. +func (s *PortManager) pickEphemeralPort(offset, count uint32, testPort func(p uint16) (bool, *tcpip.Error)) (port uint16, err *tcpip.Error) { + for i := uint32(0); i < count; i++ { + port = uint16(FirstEphemeral + (offset+i)%count) ok, err := testPort(port) if err != nil { return 0, err @@ -116,17 +190,17 @@ func (s *PortManager) PickEphemeralPort(testPort func(p uint16) (bool, *tcpip.Er } // IsPortAvailable tests if the given port is available on all given protocols. -func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { +func (s *PortManager) IsPortAvailable(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { s.mu.Lock() defer s.mu.Unlock() - return s.isPortAvailableLocked(networks, transport, addr, port, reuse) + return s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) } -func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { +func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { for _, network := range networks { desc := portDescriptor{network, transport, port} if addrs, ok := s.allocatedPorts[desc]; ok { - if !addrs.isAvailable(addr, reuse) { + if !addrs.isAvailable(addr, reuse, bindToDevice) { return false } } @@ -138,14 +212,14 @@ func (s *PortManager) isPortAvailableLocked(networks []tcpip.NetworkProtocolNumb // reserved by another endpoint. If port is zero, ReservePort will search for // an unreserved ephemeral port and reserve it, returning its value in the // "port" return value. -func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) (reservedPort uint16, err *tcpip.Error) { +func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) (reservedPort uint16, err *tcpip.Error) { s.mu.Lock() defer s.mu.Unlock() // If a port is specified, just try to reserve it for all network // protocols. if port != 0 { - if !s.reserveSpecificPort(networks, transport, addr, port, reuse) { + if !s.reserveSpecificPort(networks, transport, addr, port, reuse, bindToDevice) { return 0, tcpip.ErrPortInUse } return port, nil @@ -153,13 +227,13 @@ func (s *PortManager) ReservePort(networks []tcpip.NetworkProtocolNumber, transp // A port wasn't specified, so try to find one. return s.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { - return s.reserveSpecificPort(networks, transport, addr, p, reuse), nil + return s.reserveSpecificPort(networks, transport, addr, p, reuse, bindToDevice), nil }) } // reserveSpecificPort tries to reserve the given port on all given protocols. -func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool) bool { - if !s.isPortAvailableLocked(networks, transport, addr, port, reuse) { +func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, reuse bool, bindToDevice tcpip.NICID) bool { + if !s.isPortAvailableLocked(networks, transport, addr, port, reuse, bindToDevice) { return false } @@ -171,11 +245,16 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber m = make(bindAddresses) s.allocatedPorts[desc] = m } - if n, ok := m[addr]; ok { + d, ok := m[addr] + if !ok { + d = make(deviceNode) + m[addr] = d + } + if n, ok := d[bindToDevice]; ok { n.refs++ - m[addr] = n + d[bindToDevice] = n } else { - m[addr] = portNode{reuse: reuse, refs: 1} + d[bindToDevice] = portNode{reuse: reuse, refs: 1} } } @@ -184,22 +263,28 @@ func (s *PortManager) reserveSpecificPort(networks []tcpip.NetworkProtocolNumber // ReleasePort releases the reservation on a port/IP combination so that it can // be reserved by other endpoints. -func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16) { +func (s *PortManager) ReleasePort(networks []tcpip.NetworkProtocolNumber, transport tcpip.TransportProtocolNumber, addr tcpip.Address, port uint16, bindToDevice tcpip.NICID) { s.mu.Lock() defer s.mu.Unlock() for _, network := range networks { desc := portDescriptor{network, transport, port} if m, ok := s.allocatedPorts[desc]; ok { - n, ok := m[addr] + d, ok := m[addr] + if !ok { + continue + } + n, ok := d[bindToDevice] if !ok { continue } n.refs-- + d[bindToDevice] = n if n.refs == 0 { + delete(d, bindToDevice) + } + if len(d) == 0 { delete(m, addr) - } else { - m[addr] = n } if len(m) == 0 { delete(s.allocatedPorts, desc) diff --git a/pkg/tcpip/ports/ports_test.go b/pkg/tcpip/ports/ports_test.go index 689401661..19f4833fc 100644 --- a/pkg/tcpip/ports/ports_test.go +++ b/pkg/tcpip/ports/ports_test.go @@ -15,6 +15,7 @@ package ports import ( + "math/rand" "testing" "gvisor.dev/gvisor/pkg/tcpip" @@ -34,6 +35,7 @@ type portReserveTestAction struct { want *tcpip.Error reuse bool release bool + device tcpip.NICID } func TestPortReservation(t *testing.T) { @@ -100,6 +102,112 @@ func TestPortReservation(t *testing.T) { {port: 24, ip: anyIPAddress, release: true}, {port: 24, ip: anyIPAddress, reuse: false, want: nil}, }, + }, { + tname: "bind twice with device fails", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 3, want: nil}, + {port: 24, ip: fakeIPAddress, device: 3, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind to device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 1, want: nil}, + {port: 24, ip: fakeIPAddress, device: 2, want: nil}, + }, + }, { + tname: "bind to device and then without device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind without device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, want: nil}, + {port: 24, ip: fakeIPAddress, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind with reuse", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + }, + }, { + tname: "binding with reuse and device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 123, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 999, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "mixing reuse and not reuse by binding to device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, want: nil}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 999, want: nil}, + }, + }, { + tname: "can't bind to 0 after mixing reuse and not reuse", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "bind and release", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil}, + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: tcpip.ErrPortInUse}, + {port: 24, ip: fakeIPAddress, device: 789, reuse: true, want: nil}, + + // Release the bind to device 0 and try again. + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil}, + }, + }, { + tname: "bind twice with reuse once", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 0, reuse: true, want: tcpip.ErrPortInUse}, + }, + }, { + tname: "release an unreserved device", + actions: []portReserveTestAction{ + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil}, + // The below don't exist. + {port: 24, ip: fakeIPAddress, device: 345, reuse: false, want: nil, release: true}, + {port: 9999, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + // Release all. + {port: 24, ip: fakeIPAddress, device: 123, reuse: false, want: nil, release: true}, + {port: 24, ip: fakeIPAddress, device: 456, reuse: false, want: nil, release: true}, + }, }, } { t.Run(test.tname, func(t *testing.T) { @@ -108,12 +216,12 @@ func TestPortReservation(t *testing.T) { for _, test := range test.actions { if test.release { - pm.ReleasePort(net, fakeTransNumber, test.ip, test.port) + pm.ReleasePort(net, fakeTransNumber, test.ip, test.port, test.device) continue } - gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse) + gotPort, err := pm.ReservePort(net, fakeTransNumber, test.ip, test.port, test.reuse, test.device) if err != test.want { - t.Fatalf("ReservePort(.., .., %s, %d, %t) = %v, want %v", test.ip, test.port, test.release, err, test.want) + t.Fatalf("ReservePort(.., .., %s, %d, %t, %d) = %v, want %v", test.ip, test.port, test.reuse, test.device, err, test.want) } if test.port == 0 && (gotPort == 0 || gotPort < FirstEphemeral) { t.Fatalf("ReservePort(.., .., .., 0) = %d, want port number >= %d to be picked", gotPort, FirstEphemeral) @@ -125,7 +233,6 @@ func TestPortReservation(t *testing.T) { } func TestPickEphemeralPort(t *testing.T) { - pm := NewPortManager() customErr := &tcpip.Error{} for _, test := range []struct { name string @@ -169,9 +276,63 @@ func TestPickEphemeralPort(t *testing.T) { }, } { t.Run(test.name, func(t *testing.T) { + pm := NewPortManager() if port, err := pm.PickEphemeralPort(test.f); port != test.wantPort || err != test.wantErr { t.Errorf("PickEphemeralPort(..) = (port %d, err %v); want (port %d, err %v)", port, err, test.wantPort, test.wantErr) } }) } } + +func TestPickEphemeralPortStable(t *testing.T) { + customErr := &tcpip.Error{} + for _, test := range []struct { + name string + f func(port uint16) (bool, *tcpip.Error) + wantErr *tcpip.Error + wantPort uint16 + }{ + { + name: "no-port-available", + f: func(port uint16) (bool, *tcpip.Error) { + return false, nil + }, + wantErr: tcpip.ErrNoPortAvailable, + }, + { + name: "port-tester-error", + f: func(port uint16) (bool, *tcpip.Error) { + return false, customErr + }, + wantErr: customErr, + }, + { + name: "only-port-16042-available", + f: func(port uint16) (bool, *tcpip.Error) { + if port == FirstEphemeral+42 { + return true, nil + } + return false, nil + }, + wantPort: FirstEphemeral + 42, + }, + { + name: "only-port-under-16000-available", + f: func(port uint16) (bool, *tcpip.Error) { + if port < FirstEphemeral { + return true, nil + } + return false, nil + }, + wantErr: tcpip.ErrNoPortAvailable, + }, + } { + t.Run(test.name, func(t *testing.T) { + pm := NewPortManager() + portOffset := uint32(rand.Int31n(int32(numEphemeralPorts))) + if port, err := pm.PickEphemeralPortStable(portOffset, test.f); port != test.wantPort || err != test.wantErr { + t.Errorf("PickEphemeralPort(..) = (port %d, err %v); want (port %d, err %v)", port, err, test.wantPort, test.wantErr) + } + }) + } +} diff --git a/pkg/tcpip/sample/tun_tcp_connect/main.go b/pkg/tcpip/sample/tun_tcp_connect/main.go index f12189580..2239c1e66 100644 --- a/pkg/tcpip/sample/tun_tcp_connect/main.go +++ b/pkg/tcpip/sample/tun_tcp_connect/main.go @@ -126,7 +126,10 @@ func main() { // Create the stack with ipv4 and tcp protocols, then add a tun-based // NIC and ipv4 address. - s := stack.New([]string{ipv4.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) mtu, err := rawfile.GetMTU(tunName) if err != nil { diff --git a/pkg/tcpip/sample/tun_tcp_echo/main.go b/pkg/tcpip/sample/tun_tcp_echo/main.go index 329941775..bca73cbb1 100644 --- a/pkg/tcpip/sample/tun_tcp_echo/main.go +++ b/pkg/tcpip/sample/tun_tcp_echo/main.go @@ -111,7 +111,10 @@ func main() { // Create the stack with ip and tcp protocols, then add a tun-based // NIC and address. - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName, arp.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol(), arp.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) mtu, err := rawfile.GetMTU(tunName) if err != nil { diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 28c49e8ff..baf88bfab 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -36,6 +36,7 @@ go_library( ], deps = [ "//pkg/ilist", + "//pkg/rand", "//pkg/sleep", "//pkg/tcpip", "//pkg/tcpip/buffer", @@ -54,6 +55,7 @@ go_test( size = "small", srcs = [ "stack_test.go", + "transport_demuxer_test.go", "transport_test.go", ], deps = [ @@ -64,6 +66,9 @@ go_test( "//pkg/tcpip/iptables", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/loopback", + "//pkg/tcpip/network/ipv4", + "//pkg/tcpip/network/ipv6", + "//pkg/tcpip/transport/udp", "//pkg/waiter", ], ) diff --git a/pkg/tcpip/stack/icmp_rate_limit.go b/pkg/tcpip/stack/icmp_rate_limit.go index f8156be47..3a20839da 100644 --- a/pkg/tcpip/stack/icmp_rate_limit.go +++ b/pkg/tcpip/stack/icmp_rate_limit.go @@ -15,8 +15,6 @@ package stack import ( - "sync" - "golang.org/x/time/rate" ) @@ -33,54 +31,11 @@ const ( // ICMPRateLimiter is a global rate limiter that controls the generation of // ICMP messages generated by the stack. type ICMPRateLimiter struct { - mu sync.RWMutex - l *rate.Limiter + *rate.Limiter } // NewICMPRateLimiter returns a global rate limiter for controlling the rate // at which ICMP messages are generated by the stack. func NewICMPRateLimiter() *ICMPRateLimiter { - return &ICMPRateLimiter{l: rate.NewLimiter(icmpLimit, icmpBurst)} -} - -// Allow returns true if we are allowed to send at least 1 message at the -// moment. -func (i *ICMPRateLimiter) Allow() bool { - i.mu.RLock() - allow := i.l.Allow() - i.mu.RUnlock() - return allow -} - -// Limit returns the maximum number of ICMP messages that can be sent in one -// second. -func (i *ICMPRateLimiter) Limit() rate.Limit { - i.mu.RLock() - defer i.mu.RUnlock() - return i.l.Limit() -} - -// SetLimit sets the maximum number of ICMP messages that can be sent in one -// second. -func (i *ICMPRateLimiter) SetLimit(newLimit rate.Limit) { - i.mu.RLock() - defer i.mu.RUnlock() - i.l.SetLimit(newLimit) -} - -// Burst returns how many ICMP messages can be sent at any single instant. -func (i *ICMPRateLimiter) Burst() int { - i.mu.RLock() - defer i.mu.RUnlock() - return i.l.Burst() -} - -// SetBurst sets the maximum number of ICMP messages allowed at any single -// instant. -// -// NOTE: Changing Burst causes the underlying rate limiter to be recreated. -func (i *ICMPRateLimiter) SetBurst(burst int) { - i.mu.Lock() - i.l = rate.NewLimiter(i.l.Limit(), burst) - i.mu.Unlock() + return &ICMPRateLimiter{Limiter: rate.NewLimiter(icmpLimit, icmpBurst)} } diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index a719058b4..5993fe582 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -34,8 +34,6 @@ type NIC struct { linkEP LinkEndpoint loopback bool - demux *transportDemuxer - mu sync.RWMutex spoofing bool promiscuous bool @@ -85,7 +83,6 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback name: name, linkEP: ep, loopback: loopback, - demux: newTransportDemuxer(stack), primary: make(map[tcpip.NetworkProtocolNumber]*ilist.List), endpoints: make(map[NetworkEndpointID]*referencedNetworkEndpoint), mcastJoins: make(map[NetworkEndpointID]int32), @@ -107,6 +104,16 @@ func newNIC(stack *Stack, id tcpip.NICID, name string, ep LinkEndpoint, loopback func (n *NIC) enable() *tcpip.Error { n.attachLinkEndpoint() + // Create an endpoint to receive broadcast packets on this interface. + if _, ok := n.stack.networkProtocols[header.IPv4ProtocolNumber]; ok { + if err := n.AddAddress(tcpip.ProtocolAddress{ + Protocol: header.IPv4ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{header.IPv4Broadcast, 8 * header.IPv4AddressSize}, + }, NeverPrimaryEndpoint); err != nil { + return err + } + } + // Join the IPv6 All-Nodes Multicast group if the stack is configured to // use IPv6. This is required to ensure that this node properly receives // and responds to the various NDP messages that are destined to the @@ -148,37 +155,6 @@ func (n *NIC) setSpoofing(enable bool) { n.mu.Unlock() } -func (n *NIC) getMainNICAddress(protocol tcpip.NetworkProtocolNumber) (tcpip.AddressWithPrefix, *tcpip.Error) { - n.mu.RLock() - defer n.mu.RUnlock() - - var r *referencedNetworkEndpoint - - // Check for a primary endpoint. - if list, ok := n.primary[protocol]; ok { - for e := list.Front(); e != nil; e = e.Next() { - ref := e.(*referencedNetworkEndpoint) - if ref.getKind() == permanent && ref.tryIncRef() { - r = ref - break - } - } - - } - - if r == nil { - return tcpip.AddressWithPrefix{}, tcpip.ErrNoLinkAddress - } - - addressWithPrefix := tcpip.AddressWithPrefix{ - Address: r.ep.ID().LocalAddress, - PrefixLen: r.ep.PrefixLen(), - } - r.decRef() - - return addressWithPrefix, nil -} - // primaryEndpoint returns the primary endpoint of n for the given network // protocol. func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedNetworkEndpoint { @@ -398,13 +374,15 @@ func (n *NIC) AddAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpo return err } -// Addresses returns the addresses associated with this NIC. -func (n *NIC) Addresses() []tcpip.ProtocolAddress { +// AllAddresses returns all addresses (primary and non-primary) associated with +// this NIC. +func (n *NIC) AllAddresses() []tcpip.ProtocolAddress { n.mu.RLock() defer n.mu.RUnlock() + addrs := make([]tcpip.ProtocolAddress, 0, len(n.endpoints)) for nid, ref := range n.endpoints { - // Don't include expired or tempory endpoints to avoid confusion and + // Don't include expired or temporary endpoints to avoid confusion and // prevent the caller from using those. switch ref.getKind() { case permanentExpired, temporary: @@ -421,6 +399,34 @@ func (n *NIC) Addresses() []tcpip.ProtocolAddress { return addrs } +// PrimaryAddresses returns the primary addresses associated with this NIC. +func (n *NIC) PrimaryAddresses() []tcpip.ProtocolAddress { + n.mu.RLock() + defer n.mu.RUnlock() + + var addrs []tcpip.ProtocolAddress + for proto, list := range n.primary { + for e := list.Front(); e != nil; e = e.Next() { + ref := e.(*referencedNetworkEndpoint) + // Don't include expired or tempory endpoints to avoid confusion and + // prevent the caller from using those. + switch ref.getKind() { + case permanentExpired, temporary: + continue + } + + addrs = append(addrs, tcpip.ProtocolAddress{ + Protocol: proto, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: ref.ep.ID().LocalAddress, + PrefixLen: ref.ep.PrefixLen(), + }, + }) + } + } + return addrs +} + // AddAddressRange adds a range of addresses to n, so that it starts accepting // packets targeted at the given addresses and network protocol. The range is // given by a subnet address, and all addresses contained in the subnet are @@ -628,21 +634,6 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, _ tcpip.LinkAddr n.stack.AddLinkAddress(n.id, src, remote) - // If the packet is destined to the IPv4 Broadcast address, then make a - // route to each IPv4 network endpoint and let each endpoint handle the - // packet. - if dst == header.IPv4Broadcast { - // n.endpoints is mutex protected so acquire lock. - n.mu.RLock() - for _, ref := range n.endpoints { - if ref.isValidForIncoming() && ref.protocol == header.IPv4ProtocolNumber && ref.tryIncRef() { - handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, vv) - } - } - n.mu.RUnlock() - return - } - if ref := n.getRef(protocol, dst); ref != nil { handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, vv) return @@ -708,9 +699,7 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN // Raw socket packets are delivered based solely on the transport // protocol number. We do not inspect the payload to ensure it's // validly formed. - if !n.demux.deliverRawPacket(r, protocol, netHeader, vv) { - n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv) - } + n.stack.demux.deliverRawPacket(r, protocol, netHeader, vv) if len(vv.First()) < transProto.MinimumPacketSize() { n.stack.stats.MalformedRcvdPackets.Increment() @@ -724,9 +713,6 @@ func (n *NIC) DeliverTransportPacket(r *Route, protocol tcpip.TransportProtocolN } id := TransportEndpointID{dstPort, r.LocalAddress, srcPort, r.RemoteAddress} - if n.demux.deliverPacket(r, protocol, netHeader, vv, id) { - return - } if n.stack.demux.deliverPacket(r, protocol, netHeader, vv, id) { return } @@ -768,10 +754,7 @@ func (n *NIC) DeliverTransportControlPacket(local, remote tcpip.Address, net tcp } id := TransportEndpointID{srcPort, local, dstPort, remote} - if n.demux.deliverControlPacket(net, trans, typ, extra, vv, id) { - return - } - if n.stack.demux.deliverControlPacket(net, trans, typ, extra, vv, id) { + if n.stack.demux.deliverControlPacket(n, net, trans, typ, extra, vv, id) { return } } diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go index 07e4c770d..80101d4bb 100644 --- a/pkg/tcpip/stack/registration.go +++ b/pkg/tcpip/stack/registration.go @@ -366,14 +366,6 @@ type LinkAddressCache interface { RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep.Waker) } -// TransportProtocolFactory functions are used by the stack to instantiate -// transport protocols. -type TransportProtocolFactory func() TransportProtocol - -// NetworkProtocolFactory provides methods to be used by the stack to -// instantiate network protocols. -type NetworkProtocolFactory func() NetworkProtocol - // 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. @@ -381,34 +373,6 @@ type UnassociatedEndpointFactory interface { NewUnassociatedRawEndpoint(stack *Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) } -var ( - transportProtocols = make(map[string]TransportProtocolFactory) - networkProtocols = make(map[string]NetworkProtocolFactory) - - unassociatedFactory UnassociatedEndpointFactory -) - -// RegisterTransportProtocolFactory registers a new transport protocol factory -// with the stack so that it becomes available to users of the stack. This -// function is intended to be called by init() functions of the protocols. -func RegisterTransportProtocolFactory(name string, p TransportProtocolFactory) { - transportProtocols[name] = p -} - -// RegisterNetworkProtocolFactory registers a new network protocol factory with -// the stack so that it becomes available to users of the stack. This function -// is intended to be called by init() functions of the protocols. -func RegisterNetworkProtocolFactory(name string, p NetworkProtocolFactory) { - networkProtocols[name] = p -} - -// RegisterUnassociatedFactory registers a factory to produce endpoints not -// associated with any particular transport protocol. This function is intended -// to be called by init() functions of the protocols. -func RegisterUnassociatedFactory(f UnassociatedEndpointFactory) { - unassociatedFactory = f -} - // GSOType is the type of GSO segments. // // +stateify savable diff --git a/pkg/tcpip/stack/route.go b/pkg/tcpip/stack/route.go index 5c8b7977a..22db619a1 100644 --- a/pkg/tcpip/stack/route.go +++ b/pkg/tcpip/stack/route.go @@ -59,6 +59,8 @@ func makeRoute(netProto tcpip.NetworkProtocolNumber, localAddr, remoteAddr tcpip loop = PacketLoop } else if multicastLoop && (header.IsV4MulticastAddress(remoteAddr) || header.IsV6MulticastAddress(remoteAddr)) { loop |= PacketLoop + } else if remoteAddr == header.IPv4Broadcast { + loop |= PacketLoop } return Route{ @@ -152,11 +154,15 @@ func (r *Route) IsResolutionRequired() bool { } // WritePacket writes the packet through the given route. -func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8) *tcpip.Error { +func (r *Route) WritePacket(gso *GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.TransportProtocolNumber, ttl uint8, useDefaultTTL bool) *tcpip.Error { if !r.ref.isValidForOutgoing() { return tcpip.ErrInvalidEndpointState } + if useDefaultTTL { + ttl = r.DefaultTTL() + } + err := r.ref.ep.WritePacket(r, gso, hdr, payload, protocol, ttl, r.loop) if err != nil { r.Stats().IP.OutgoingPacketErrors.Increment() @@ -208,10 +214,17 @@ func (r *Route) Clone() Route { return *r } -// MakeLoopedRoute duplicates the given route and tweaks it in case of multicast. +// MakeLoopedRoute duplicates the given route with special handling for routes +// used for sending multicast or broadcast packets. In those cases the +// multicast/broadcast address is the remote address when sending out, but for +// incoming (looped) packets it becomes the local address. Similarly, the local +// interface address that was the local address going out becomes the remote +// address coming in. This is different to unicast routes where local and +// remote addresses remain the same as they identify location (local vs remote) +// not direction (source vs destination). func (r *Route) MakeLoopedRoute() Route { l := r.Clone() - if header.IsV4MulticastAddress(r.RemoteAddress) || header.IsV6MulticastAddress(r.RemoteAddress) { + if r.RemoteAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(r.RemoteAddress) || header.IsV6MulticastAddress(r.RemoteAddress) { l.RemoteAddress, l.LocalAddress = l.LocalAddress, l.RemoteAddress l.RemoteLinkAddress = l.LocalLinkAddress } diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index 1fe21b68e..ff574a055 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -17,18 +17,15 @@ // // For consumers, the only function of interest is New(), everything else is // provided by the tcpip/public package. -// -// For protocol implementers, RegisterTransportProtocolFactory() and -// RegisterNetworkProtocolFactory() are used to register protocol factories with -// the stack, which will then be used to instantiate protocol objects when -// consumers interact with the stack. package stack import ( + "encoding/binary" "sync" "time" "golang.org/x/time/rate" + "gvisor.dev/gvisor/pkg/rand" "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -351,6 +348,9 @@ 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 demux *transportDemuxer @@ -359,10 +359,6 @@ type Stack struct { linkAddrCache *linkAddrCache - // raw indicates whether raw sockets may be created. It is set during - // Stack creation and is immutable. - raw bool - mu sync.RWMutex nics map[tcpip.NICID]*NIC forwarding bool @@ -394,10 +390,22 @@ type Stack struct { // icmpRateLimiter is a global rate limiter for all ICMP messages generated // by the stack. icmpRateLimiter *ICMPRateLimiter + + // portSeed is a one-time random value initialized at stack startup + // and is used to seed the TCP port picking on active connections + // + // TODO(gvisor.dev/issues/940): S/R this field. + portSeed uint32 } // Options contains optional Stack configuration. type Options struct { + // NetworkProtocols lists the network protocols to enable. + NetworkProtocols []NetworkProtocol + + // TransportProtocols lists the transport protocols to enable. + TransportProtocols []TransportProtocol + // Clock is an optional clock source used for timestampping packets. // // If no Clock is specified, the clock source will be time.Now. @@ -411,8 +419,9 @@ type Options struct { // stack (false). HandleLocal bool - // Raw indicates whether raw sockets may be created. - Raw bool + // UnassociatedFactory produces unassociated endpoints raw endpoints. + // Raw endpoints are enabled only if this is non-nil. + UnassociatedFactory UnassociatedEndpointFactory } // New allocates a new networking stack with only the requested networking and @@ -422,7 +431,7 @@ type Options struct { // SetNetworkProtocolOption/SetTransportProtocolOption methods provided by the // stack. Please refer to individual protocol implementations as to what options // are supported. -func New(network []string, transport []string, opts Options) *Stack { +func New(opts Options) *Stack { clock := opts.Clock if clock == nil { clock = &tcpip.StdClock{} @@ -438,17 +447,12 @@ func New(network []string, transport []string, opts Options) *Stack { clock: clock, stats: opts.Stats.FillIn(), handleLocal: opts.HandleLocal, - raw: opts.Raw, icmpRateLimiter: NewICMPRateLimiter(), + portSeed: generateRandUint32(), } // Add specified network protocols. - for _, name := range network { - netProtoFactory, ok := networkProtocols[name] - if !ok { - continue - } - netProto := netProtoFactory() + for _, netProto := range opts.NetworkProtocols { s.networkProtocols[netProto.Number()] = netProto if r, ok := netProto.(LinkAddressResolver); ok { s.linkAddrResolvers[r.LinkAddressProtocol()] = r @@ -456,18 +460,14 @@ func New(network []string, transport []string, opts Options) *Stack { } // Add specified transport protocols. - for _, name := range transport { - transProtoFactory, ok := transportProtocols[name] - if !ok { - continue - } - transProto := transProtoFactory() + for _, transProto := range opts.TransportProtocols { s.transportProtocols[transProto.Number()] = &transportProtocolState{ proto: transProto, } } - s.unassociatedFactory = unassociatedFactory + // Add the factory for unassociated endpoints, if present. + s.unassociatedFactory = opts.UnassociatedFactory // Create the global transport demuxer. s.demux = newTransportDemuxer(s) @@ -602,7 +602,7 @@ 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.raw { + if s.unassociatedFactory == nil { return nil, tcpip.ErrNotPermitted } @@ -738,7 +738,7 @@ func (s *Stack) NICInfo() map[tcpip.NICID]NICInfo { nics[id] = NICInfo{ Name: nic.name, LinkAddress: nic.linkEP.LinkAddress(), - ProtocolAddresses: nic.Addresses(), + ProtocolAddresses: nic.PrimaryAddresses(), Flags: flags, MTU: nic.linkEP.MTU(), Stats: nic.stats, @@ -845,19 +845,37 @@ func (s *Stack) RemoveAddress(id tcpip.NICID, addr tcpip.Address) *tcpip.Error { return tcpip.ErrUnknownNICID } -// GetMainNICAddress returns the first primary address (and the subnet that -// contains it) for the given NIC and protocol. Returns an arbitrary endpoint's -// address if no primary addresses exist. Returns an error if the NIC doesn't -// exist or has no endpoints. +// AllAddresses returns a map of NICIDs to their protocol addresses (primary +// and non-primary). +func (s *Stack) AllAddresses() map[tcpip.NICID][]tcpip.ProtocolAddress { + s.mu.RLock() + defer s.mu.RUnlock() + + nics := make(map[tcpip.NICID][]tcpip.ProtocolAddress) + for id, nic := range s.nics { + nics[id] = nic.AllAddresses() + } + return nics +} + +// GetMainNICAddress returns the first primary address and prefix for the given +// NIC and protocol. Returns an error if the NIC doesn't exist and an empty +// value if the NIC doesn't have a primary address for the given protocol. func (s *Stack) GetMainNICAddress(id tcpip.NICID, protocol tcpip.NetworkProtocolNumber) (tcpip.AddressWithPrefix, *tcpip.Error) { s.mu.RLock() defer s.mu.RUnlock() - if nic, ok := s.nics[id]; ok { - return nic.getMainNICAddress(protocol) + nic, ok := s.nics[id] + if !ok { + return tcpip.AddressWithPrefix{}, tcpip.ErrUnknownNICID } - return tcpip.AddressWithPrefix{}, tcpip.ErrUnknownNICID + for _, a := range nic.PrimaryAddresses() { + if a.Protocol == protocol { + return a.AddressWithPrefix, nil + } + } + return tcpip.AddressWithPrefix{}, nil } func (s *Stack) getRefEP(nic *NIC, localAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (ref *referencedNetworkEndpoint) { @@ -884,7 +902,7 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n } } else { for _, route := range s.routeTable { - if (id != 0 && id != route.NIC) || (len(remoteAddr) != 0 && !isBroadcast && !route.Destination.Contains(remoteAddr)) { + if (id != 0 && id != route.NIC) || (len(remoteAddr) != 0 && !route.Destination.Contains(remoteAddr)) { continue } if nic, ok := s.nics[route.NIC]; ok { @@ -1024,73 +1042,27 @@ func (s *Stack) RemoveWaker(nicid tcpip.NICID, addr tcpip.Address, waker *sleep. // transport dispatcher. Received packets that match the provided id will be // delivered to the given endpoint; specifying a nic is optional, but // nic-specific IDs have precedence over global ones. -func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { - if nicID == 0 { - return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort) - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic == nil { - return tcpip.ErrUnknownNICID - } - - return nic.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort) +func (s *Stack) RegisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { + return s.demux.registerEndpoint(netProtos, protocol, id, ep, reusePort, bindToDevice) } // UnregisterTransportEndpoint removes the endpoint with the given id from the // stack transport dispatcher. -func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint) { - if nicID == 0 { - s.demux.unregisterEndpoint(netProtos, protocol, id, ep) - return - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic != nil { - nic.demux.unregisterEndpoint(netProtos, protocol, id, ep) - } +func (s *Stack) UnregisterTransportEndpoint(nicID tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { + s.demux.unregisterEndpoint(netProtos, protocol, id, ep, bindToDevice) } // RegisterRawTransportEndpoint registers the given endpoint with the stack // transport dispatcher. Received packets that match the provided transport // protocol will be delivered to the given endpoint. func (s *Stack) RegisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) *tcpip.Error { - if nicID == 0 { - return s.demux.registerRawEndpoint(netProto, transProto, ep) - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic == nil { - return tcpip.ErrUnknownNICID - } - - return nic.demux.registerRawEndpoint(netProto, transProto, ep) + return s.demux.registerRawEndpoint(netProto, transProto, ep) } // UnregisterRawTransportEndpoint removes the endpoint for the transport // protocol from the stack transport dispatcher. func (s *Stack) UnregisterRawTransportEndpoint(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, ep RawTransportEndpoint) { - if nicID == 0 { - s.demux.unregisterRawEndpoint(netProto, transProto, ep) - return - } - - s.mu.RLock() - defer s.mu.RUnlock() - - nic := s.nics[nicID] - if nic != nil { - nic.demux.unregisterRawEndpoint(netProto, transProto, ep) - } + s.demux.unregisterRawEndpoint(netProto, transProto, ep) } // RegisterRestoredEndpoint records e as an endpoint that has been restored on @@ -1234,3 +1206,19 @@ func (s *Stack) SetICMPBurst(burst int) { func (s *Stack) AllowICMPMessage() bool { return s.icmpRateLimiter.Allow() } + +// PortSeed returns a 32 bit value that can be used as a seed value for port +// picking. +// +// NOTE: The seed is generated once during stack initialization only. +func (s *Stack) PortSeed() uint32 { + return s.portSeed +} + +func generateRandUint32() uint32 { + b := make([]byte, 4) + if _, err := rand.Read(b); err != nil { + panic(err) + } + return binary.LittleEndian.Uint32(b) +} diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index 0c26c9911..8f063038b 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -222,11 +222,17 @@ func (f *fakeNetworkProtocol) Option(option interface{}) *tcpip.Error { } } +func fakeNetFactory() stack.NetworkProtocol { + return &fakeNetworkProtocol{} +} + func TestNetworkReceive(t *testing.T) { // Create a stack with the fake network protocol, one nic, and two // addresses attached to it: 1 & 2. ep := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) } @@ -304,7 +310,7 @@ func sendTo(s *stack.Stack, addr tcpip.Address, payload buffer.View) *tcpip.Erro func send(r stack.Route, payload buffer.View) *tcpip.Error { hdr := buffer.NewPrependable(int(r.MaxHeaderLength())) - return r.WritePacket(nil /* gso */, hdr, payload.ToVectorisedView(), fakeTransNumber, 123) + return r.WritePacket(nil /* gso */, hdr, payload.ToVectorisedView(), fakeTransNumber, 123 /* ttl */, false /* useDefaultTTL */) } func testSendTo(t *testing.T, s *stack.Stack, addr tcpip.Address, ep *channel.Endpoint, payload buffer.View) { @@ -370,7 +376,9 @@ func TestNetworkSend(t *testing.T) { // address: 1. The route table sends all packets through the only // existing nic. ep := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) if err := s.CreateNIC(1, ep); err != nil { t.Fatal("NewNIC failed:", err) } @@ -395,7 +403,9 @@ func TestNetworkSendMultiRoute(t *testing.T) { // Create a stack with the fake network protocol, two nics, and two // addresses per nic, the first nic has odd address, the second one has // even addresses. - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { @@ -476,7 +486,9 @@ func TestRoutes(t *testing.T) { // Create a stack with the fake network protocol, two nics, and two // addresses per nic, the first nic has odd address, the second one has // even addresses. - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep1 := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep1); err != nil { @@ -554,7 +566,9 @@ func TestAddressRemoval(t *testing.T) { localAddr := tcpip.Address([]byte{localAddrByte}) remoteAddr := tcpip.Address("\x02") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -599,7 +613,9 @@ func TestAddressRemovalWithRouteHeld(t *testing.T) { localAddr := tcpip.Address([]byte{localAddrByte}) remoteAddr := tcpip.Address("\x02") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -688,7 +704,9 @@ func TestEndpointExpiration(t *testing.T) { for _, promiscuous := range []bool{true, false} { for _, spoofing := range []bool{true, false} { t.Run(fmt.Sprintf("promiscuous=%t spoofing=%t", promiscuous, spoofing), func(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { @@ -844,7 +862,9 @@ func TestEndpointExpiration(t *testing.T) { } func TestPromiscuousMode(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -894,7 +914,9 @@ func TestSpoofingWithAddress(t *testing.T) { nonExistentLocalAddr := tcpip.Address("\x02") dstAddr := tcpip.Address("\x03") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -930,10 +952,10 @@ func TestSpoofingWithAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != nonExistentLocalAddr { - t.Errorf("Route has wrong local address: got %v, wanted %v", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("got Route.LocalAddress = %s, want = %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %v, wanted %v", r.RemoteAddress, dstAddr) + t.Errorf("got Route.RemoteAddress = %s, want = %s", r.RemoteAddress, dstAddr) } // Sending a packet works. testSendTo(t, s, dstAddr, ep, nil) @@ -945,10 +967,10 @@ func TestSpoofingWithAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != localAddr { - t.Errorf("Route has wrong local address: got %v, wanted %v", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("got Route.LocalAddress = %s, want = %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %v, wanted %v", r.RemoteAddress, dstAddr) + t.Errorf("got Route.RemoteAddress = %s, want = %s", r.RemoteAddress, dstAddr) } // Sending a packet using the route works. testSend(t, r, ep, nil) @@ -958,7 +980,9 @@ func TestSpoofingNoAddress(t *testing.T) { nonExistentLocalAddr := tcpip.Address("\x01") dstAddr := tcpip.Address("\x02") - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -992,18 +1016,36 @@ func TestSpoofingNoAddress(t *testing.T) { t.Fatal("FindRoute failed:", err) } if r.LocalAddress != nonExistentLocalAddr { - t.Errorf("Route has wrong local address: got %v, wanted %v", r.LocalAddress, nonExistentLocalAddr) + t.Errorf("got Route.LocalAddress = %s, want = %s", r.LocalAddress, nonExistentLocalAddr) } if r.RemoteAddress != dstAddr { - t.Errorf("Route has wrong remote address: got %v, wanted %v", r.RemoteAddress, dstAddr) + t.Errorf("got Route.RemoteAddress = %s, want = %s", r.RemoteAddress, dstAddr) } // Sending a packet works. // FIXME(b/139841518):Spoofing doesn't work if there is no primary address. // testSendTo(t, s, remoteAddr, ep, nil) } -func TestBroadcastNeedsNoRoute(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) +func verifyRoute(gotRoute, wantRoute stack.Route) error { + if gotRoute.LocalAddress != wantRoute.LocalAddress { + return fmt.Errorf("bad local address: got %s, want = %s", gotRoute.LocalAddress, wantRoute.LocalAddress) + } + if gotRoute.RemoteAddress != wantRoute.RemoteAddress { + return fmt.Errorf("bad remote address: got %s, want = %s", gotRoute.RemoteAddress, wantRoute.RemoteAddress) + } + if gotRoute.RemoteLinkAddress != wantRoute.RemoteLinkAddress { + return fmt.Errorf("bad remote link address: got %s, want = %s", gotRoute.RemoteLinkAddress, wantRoute.RemoteLinkAddress) + } + if gotRoute.NextHop != wantRoute.NextHop { + return fmt.Errorf("bad next-hop address: got %s, want = %s", gotRoute.NextHop, wantRoute.NextHop) + } + return nil +} + +func TestOutgoingBroadcastWithEmptyRouteTable(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1013,28 +1055,99 @@ func TestBroadcastNeedsNoRoute(t *testing.T) { // If there is no endpoint, it won't work. if _, err := s.FindRoute(1, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */); err != tcpip.ErrNetworkUnreachable { - t.Fatalf("got FindRoute(1, %v, %v, %v) = %v, want = %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) + t.Fatalf("got FindRoute(1, %s, %s, %d) = %s, want = %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) } - if err := s.AddAddress(1, fakeNetNumber, header.IPv4Any); err != nil { - t.Fatalf("AddAddress(%v, %v) failed: %s", fakeNetNumber, header.IPv4Any, err) + protoAddr := tcpip.ProtocolAddress{Protocol: fakeNetNumber, AddressWithPrefix: tcpip.AddressWithPrefix{header.IPv4Any, 0}} + if err := s.AddProtocolAddress(1, protoAddr); err != nil { + t.Fatalf("AddProtocolAddress(1, %s) failed: %s", protoAddr, err) } r, err := s.FindRoute(1, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) if err != nil { - t.Fatalf("FindRoute(1, %v, %v, %v) failed: %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) + t.Fatalf("FindRoute(1, %s, %s, %d) failed: %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) + } + if err := verifyRoute(r, stack.Route{LocalAddress: header.IPv4Any, RemoteAddress: header.IPv4Broadcast}); err != nil { + t.Errorf("FindRoute(1, %s, %s, %d) returned unexpected Route: %s)", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err) } - if r.LocalAddress != header.IPv4Any { - t.Errorf("Bad local address: got %v, want = %v", r.LocalAddress, header.IPv4Any) + // If the NIC doesn't exist, it won't work. + if _, err := s.FindRoute(2, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */); err != tcpip.ErrNetworkUnreachable { + t.Fatalf("got FindRoute(2, %s, %s, %d) = %s want = %s", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) } +} - if r.RemoteAddress != header.IPv4Broadcast { - t.Errorf("Bad remote address: got %v, want = %v", r.RemoteAddress, header.IPv4Broadcast) +func TestOutgoingBroadcastWithRouteTable(t *testing.T) { + defaultAddr := tcpip.AddressWithPrefix{header.IPv4Any, 0} + // Local subnet on NIC1: 192.168.1.58/24, gateway 192.168.1.1. + nic1Addr := tcpip.AddressWithPrefix{"\xc0\xa8\x01\x3a", 24} + nic1Gateway := tcpip.Address("\xc0\xa8\x01\x01") + // Local subnet on NIC2: 10.10.10.5/24, gateway 10.10.10.1. + nic2Addr := tcpip.AddressWithPrefix{"\x0a\x0a\x0a\x05", 24} + nic2Gateway := tcpip.Address("\x0a\x0a\x0a\x01") + + // Create a new stack with two NICs. + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) + ep := channel.New(10, defaultMTU, "") + if err := s.CreateNIC(1, ep); err != nil { + t.Fatalf("CreateNIC failed: %s", err) + } + if err := s.CreateNIC(2, ep); err != nil { + t.Fatalf("CreateNIC failed: %s", err) + } + nic1ProtoAddr := tcpip.ProtocolAddress{fakeNetNumber, nic1Addr} + if err := s.AddProtocolAddress(1, nic1ProtoAddr); err != nil { + t.Fatalf("AddProtocolAddress(1, %s) failed: %s", nic1ProtoAddr, err) } - // If the NIC doesn't exist, it won't work. - if _, err := s.FindRoute(2, header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */); err != tcpip.ErrNetworkUnreachable { - t.Fatalf("got FindRoute(2, %v, %v, %v) = %v want = %v", header.IPv4Any, header.IPv4Broadcast, fakeNetNumber, err, tcpip.ErrNetworkUnreachable) + nic2ProtoAddr := tcpip.ProtocolAddress{fakeNetNumber, nic2Addr} + if err := s.AddProtocolAddress(2, nic2ProtoAddr); err != nil { + t.Fatalf("AddAddress(2, %s) failed: %s", nic2ProtoAddr, err) + } + + // Set the initial route table. + rt := []tcpip.Route{ + {Destination: nic1Addr.Subnet(), NIC: 1}, + {Destination: nic2Addr.Subnet(), NIC: 2}, + {Destination: defaultAddr.Subnet(), Gateway: nic2Gateway, NIC: 2}, + {Destination: defaultAddr.Subnet(), Gateway: nic1Gateway, NIC: 1}, + } + s.SetRouteTable(rt) + + // When an interface is given, the route for a broadcast goes through it. + r, err := s.FindRoute(1, nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) + if err != nil { + t.Fatalf("FindRoute(1, %s, %s, %d) failed: %s", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) + } + if err := verifyRoute(r, stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + t.Errorf("FindRoute(1, %s, %s, %d) returned unexpected Route: %s)", nic1Addr.Address, header.IPv4Broadcast, fakeNetNumber, err) + } + + // When an interface is not given, it consults the route table. + // 1. Case: Using the default route. + r, err = s.FindRoute(0, "", header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) + if err != nil { + t.Fatalf("FindRoute(0, \"\", %s, %d) failed: %s", header.IPv4Broadcast, fakeNetNumber, err) + } + if err := verifyRoute(r, stack.Route{LocalAddress: nic2Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + t.Errorf("FindRoute(0, \"\", %s, %d) returned unexpected Route: %s)", header.IPv4Broadcast, fakeNetNumber, err) + } + + // 2. Case: Having an explicit route for broadcast will select that one. + rt = append( + []tcpip.Route{ + {Destination: tcpip.AddressWithPrefix{header.IPv4Broadcast, 8 * header.IPv4AddressSize}.Subnet(), NIC: 1}, + }, + rt..., + ) + s.SetRouteTable(rt) + r, err = s.FindRoute(0, "", header.IPv4Broadcast, fakeNetNumber, false /* multicastLoop */) + if err != nil { + t.Fatalf("FindRoute(0, \"\", %s, %d) failed: %s", header.IPv4Broadcast, fakeNetNumber, err) + } + if err := verifyRoute(r, stack.Route{LocalAddress: nic1Addr.Address, RemoteAddress: header.IPv4Broadcast}); err != nil { + t.Errorf("FindRoute(0, \"\", %s, %d) returned unexpected Route: %s)", header.IPv4Broadcast, fakeNetNumber, err) } } @@ -1074,7 +1187,9 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) { {"IPv6 Unicast Not Link-Local 7", true, "\xf0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"}, } { t.Run(tc.name, func(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1130,7 +1245,9 @@ func TestMulticastOrIPv6LinkLocalNeedsNoRoute(t *testing.T) { // Add a range of addresses, then check that a packet is delivered. func TestAddressRangeAcceptsMatchingPacket(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1196,7 +1313,9 @@ func testNicForAddressRange(t *testing.T, nicID tcpip.NICID, s *stack.Stack, sub // existent. func TestCheckLocalAddressForSubnet(t *testing.T) { const nicID tcpip.NICID = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicID, ep); err != nil { @@ -1234,7 +1353,9 @@ func TestCheckLocalAddressForSubnet(t *testing.T) { // Set a range of addresses, then send a packet to a destination outside the // range and then check it doesn't get delivered. func TestAddressRangeRejectsNonmatchingPacket(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { @@ -1266,7 +1387,10 @@ func TestAddressRangeRejectsNonmatchingPacket(t *testing.T) { } func TestNetworkOptions(t *testing.T) { - s := stack.New([]string{"fakeNet"}, []string{}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{}, + }) // Try an unsupported network protocol. if err := s.SetNetworkProtocolOption(tcpip.NetworkProtocolNumber(99999), fakeNetGoodOption(false)); err != tcpip.ErrUnknownProtocol { @@ -1319,7 +1443,9 @@ func stackContainsAddressRange(s *stack.Stack, id tcpip.NICID, addrRange tcpip.S } func TestAddresRangeAddRemove(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1360,7 +1486,9 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { t.Run(fmt.Sprintf("canBe=%d", canBe), func(t *testing.T) { for never := 0; never < 3; never++ { t.Run(fmt.Sprintf("never=%d", never), func(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1400,20 +1528,20 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { // Check that GetMainNICAddress returns an address if at least // one primary address was added. In that case make sure the // address/prefixLen matches what we added. + gotAddr, err := s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { + t.Fatal("GetMainNICAddress failed:", err) + } if len(primaryAddrAdded) == 0 { - // No primary addresses present, expect an error. - if _, err := s.GetMainNICAddress(1, fakeNetNumber); err != tcpip.ErrNoLinkAddress { - t.Fatalf("got s.GetMainNICAddress(...) = %v, wanted = %s", err, tcpip.ErrNoLinkAddress) + // No primary addresses present. + if wantAddr := (tcpip.AddressWithPrefix{}); gotAddr != wantAddr { + t.Fatalf("GetMainNICAddress: got addr = %s, want = %s", gotAddr, wantAddr) } } else { - // At least one primary address was added, expect a valid - // address and prefixLen. - gotAddressWithPefix, err := s.GetMainNICAddress(1, fakeNetNumber) - if err != nil { - t.Fatal("GetMainNICAddress failed:", err) - } - if _, ok := primaryAddrAdded[gotAddressWithPefix]; !ok { - t.Fatalf("GetMainNICAddress: got addressWithPrefix = %v, wanted any in {%v}", gotAddressWithPefix, primaryAddrAdded) + // At least one primary address was added, verify the returned + // address is in the list of primary addresses we added. + if _, ok := primaryAddrAdded[gotAddr]; !ok { + t.Fatalf("GetMainNICAddress: got = %s, want any in {%v}", gotAddr, primaryAddrAdded) } } }) @@ -1425,7 +1553,9 @@ func TestGetMainNICAddressAddPrimaryNonPrimary(t *testing.T) { } func TestGetMainNICAddressAddRemove(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(1, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1452,19 +1582,25 @@ func TestGetMainNICAddressAddRemove(t *testing.T) { } // Check that we get the right initial address and prefix length. - if gotAddressWithPrefix, err := s.GetMainNICAddress(1, fakeNetNumber); err != nil { + gotAddr, err := s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { t.Fatal("GetMainNICAddress failed:", err) - } else if gotAddressWithPrefix != protocolAddress.AddressWithPrefix { - t.Fatalf("got GetMainNICAddress = %+v, want = %+v", gotAddressWithPrefix, protocolAddress.AddressWithPrefix) + } + if wantAddr := protocolAddress.AddressWithPrefix; gotAddr != wantAddr { + t.Fatalf("got s.GetMainNICAddress(...) = %s, want = %s", gotAddr, wantAddr) } if err := s.RemoveAddress(1, protocolAddress.AddressWithPrefix.Address); err != nil { t.Fatal("RemoveAddress failed:", err) } - // Check that we get an error after removal. - if _, err := s.GetMainNICAddress(1, fakeNetNumber); err != tcpip.ErrNoLinkAddress { - t.Fatalf("got s.GetMainNICAddress(...) = %v, want = %s", err, tcpip.ErrNoLinkAddress) + // Check that we get no address after removal. + gotAddr, err = s.GetMainNICAddress(1, fakeNetNumber) + if err != nil { + t.Fatal("GetMainNICAddress failed:", err) + } + if wantAddr := (tcpip.AddressWithPrefix{}); gotAddr != wantAddr { + t.Fatalf("got GetMainNICAddress(...) = %s, want = %s", gotAddr, wantAddr) } }) } @@ -1479,8 +1615,10 @@ func (g *addressGenerator) next(addrLen int) tcpip.Address { } func verifyAddresses(t *testing.T, expectedAddresses, gotAddresses []tcpip.ProtocolAddress) { + t.Helper() + if len(gotAddresses) != len(expectedAddresses) { - t.Fatalf("got len(addresses) = %d, wanted = %d", len(gotAddresses), len(expectedAddresses)) + t.Fatalf("got len(addresses) = %d, want = %d", len(gotAddresses), len(expectedAddresses)) } sort.Slice(gotAddresses, func(i, j int) bool { @@ -1500,7 +1638,9 @@ func verifyAddresses(t *testing.T, expectedAddresses, gotAddresses []tcpip.Proto func TestAddAddress(t *testing.T) { const nicid = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1519,13 +1659,15 @@ func TestAddAddress(t *testing.T) { }) } - gotAddresses := s.NICInfo()[nicid].ProtocolAddresses + gotAddresses := s.AllAddresses()[nicid] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddProtocolAddress(t *testing.T) { const nicid = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1551,13 +1693,15 @@ func TestAddProtocolAddress(t *testing.T) { } } - gotAddresses := s.NICInfo()[nicid].ProtocolAddresses + gotAddresses := s.AllAddresses()[nicid] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddAddressWithOptions(t *testing.T) { const nicid = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1580,13 +1724,15 @@ func TestAddAddressWithOptions(t *testing.T) { } } - gotAddresses := s.NICInfo()[nicid].ProtocolAddresses + gotAddresses := s.AllAddresses()[nicid] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestAddProtocolAddressWithOptions(t *testing.T) { const nicid = 1 - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) ep := channel.New(10, defaultMTU, "") if err := s.CreateNIC(nicid, ep); err != nil { t.Fatal("CreateNIC failed:", err) @@ -1615,12 +1761,14 @@ func TestAddProtocolAddressWithOptions(t *testing.T) { } } - gotAddresses := s.NICInfo()[nicid].ProtocolAddresses + gotAddresses := s.AllAddresses()[nicid] verifyAddresses(t, expectedAddresses, gotAddresses) } func TestNICStats(t *testing.T) { - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + 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) @@ -1666,7 +1814,9 @@ func TestNICStats(t *testing.T) { func TestNICForwarding(t *testing.T) { // Create a stack with the fake network protocol, two NICs, each with // an address. - s := stack.New([]string{"fakeNet"}, nil, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + }) s.SetForwarding(true) ep1 := channel.New(10, defaultMTU, "") @@ -1714,9 +1864,3 @@ func TestNICForwarding(t *testing.T) { t.Errorf("got Tx.Bytes.Value() = %d, want = %d", got, want) } } - -func init() { - stack.RegisterNetworkProtocolFactory("fakeNet", func() stack.NetworkProtocol { - return &fakeNetworkProtocol{} - }) -} diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index cf8a6d129..92267ce4d 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -35,25 +35,109 @@ type protocolIDs struct { type transportEndpoints struct { // mu protects all fields of the transportEndpoints. mu sync.RWMutex - endpoints map[TransportEndpointID]TransportEndpoint + endpoints map[TransportEndpointID]*endpointsByNic // rawEndpoints contains endpoints for raw sockets, which receive all // traffic of a given protocol regardless of port. rawEndpoints []RawTransportEndpoint } +type endpointsByNic struct { + mu sync.RWMutex + endpoints map[tcpip.NICID]*multiPortEndpoint + // seed is a random secret for a jenkins hash. + seed uint32 +} + +// HandlePacket is called by the stack when new packets arrive to this transport +// endpoint. +func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { + epsByNic.mu.RLock() + + mpep, ok := epsByNic.endpoints[r.ref.nic.ID()] + if !ok { + if mpep, ok = epsByNic.endpoints[0]; !ok { + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return + } + } + + // If this is a broadcast or multicast datagram, deliver the datagram to all + // endpoints bound to the right device. + if isMulticastOrBroadcast(id.LocalAddress) { + mpep.handlePacketAll(r, id, vv) + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. + return + } + + // multiPortEndpoints are guaranteed to have at least one element. + selectEndpoint(id, mpep, epsByNic.seed).HandlePacket(r, id, vv) + epsByNic.mu.RUnlock() // Don't use defer for performance reasons. +} + +// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. +func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) { + epsByNic.mu.RLock() + defer epsByNic.mu.RUnlock() + + mpep, ok := epsByNic.endpoints[n.ID()] + if !ok { + mpep, ok = epsByNic.endpoints[0] + } + if !ok { + return + } + + // TODO(eyalsoha): Why don't we look at id to see if this packet needs to + // broadcast like we are doing with handlePacket above? + + // multiPortEndpoints are guaranteed to have at least one element. + selectEndpoint(id, mpep, epsByNic.seed).HandleControlPacket(id, typ, extra, vv) +} + +// registerEndpoint returns true if it succeeds. It fails and returns +// false if ep already has an element with the same key. +func (epsByNic *endpointsByNic) registerEndpoint(t TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { + epsByNic.mu.Lock() + defer epsByNic.mu.Unlock() + + if multiPortEp, ok := epsByNic.endpoints[bindToDevice]; ok { + // There was already a bind. + return multiPortEp.singleRegisterEndpoint(t, reusePort) + } + + // This is a new binding. + multiPortEp := &multiPortEndpoint{} + multiPortEp.endpointsMap = make(map[TransportEndpoint]int) + multiPortEp.reuse = reusePort + epsByNic.endpoints[bindToDevice] = multiPortEp + return multiPortEp.singleRegisterEndpoint(t, reusePort) +} + +// unregisterEndpoint returns true if endpointsByNic has to be unregistered. +func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t TransportEndpoint) bool { + epsByNic.mu.Lock() + defer epsByNic.mu.Unlock() + multiPortEp, ok := epsByNic.endpoints[bindToDevice] + if !ok { + return false + } + if multiPortEp.unregisterEndpoint(t) { + delete(epsByNic.endpoints, bindToDevice) + } + return len(epsByNic.endpoints) == 0 +} + // unregisterEndpoint unregisters the endpoint with the given id such that it // won't receive any more packets. -func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint) { +func (eps *transportEndpoints) unregisterEndpoint(id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { eps.mu.Lock() defer eps.mu.Unlock() - e, ok := eps.endpoints[id] + epsByNic, ok := eps.endpoints[id] if !ok { return } - if multiPortEp, ok := e.(*multiPortEndpoint); ok { - if !multiPortEp.unregisterEndpoint(ep) { - return - } + if !epsByNic.unregisterEndpoint(bindToDevice, ep) { + return } delete(eps.endpoints, id) } @@ -75,7 +159,7 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer { for netProto := range stack.networkProtocols { for proto := range stack.transportProtocols { d.protocol[protocolIDs{netProto, proto}] = &transportEndpoints{ - endpoints: make(map[TransportEndpointID]TransportEndpoint), + endpoints: make(map[TransportEndpointID]*endpointsByNic), } } } @@ -85,10 +169,10 @@ func newTransportDemuxer(stack *Stack) *transportDemuxer { // registerEndpoint registers the given endpoint with the dispatcher such that // packets that match the endpoint ID are delivered to it. -func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { +func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { for i, n := range netProtos { - if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort); err != nil { - d.unregisterEndpoint(netProtos[:i], protocol, id, ep) + if err := d.singleRegisterEndpoint(n, protocol, id, ep, reusePort, bindToDevice); err != nil { + d.unregisterEndpoint(netProtos[:i], protocol, id, ep, bindToDevice) return err } } @@ -97,13 +181,14 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum } // multiPortEndpoint is a container for TransportEndpoints which are bound to -// the same pair of address and port. +// the same pair of address and port. endpointsArr always has at least one +// element. type multiPortEndpoint struct { mu sync.RWMutex endpointsArr []TransportEndpoint endpointsMap map[TransportEndpoint]int - // seed is a random secret for a jenkins hash. - seed uint32 + // reuse indicates if more than one endpoint is allowed. + reuse bool } // reciprocalScale scales a value into range [0, n). @@ -117,9 +202,10 @@ func reciprocalScale(val, n uint32) uint32 { // selectEndpoint calculates a hash of destination and source addresses and // ports then uses it to select a socket. In this case, all packets from one // address will be sent to same endpoint. -func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEndpoint { - ep.mu.RLock() - defer ep.mu.RUnlock() +func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32) TransportEndpoint { + if len(mpep.endpointsArr) == 1 { + return mpep.endpointsArr[0] + } payload := []byte{ byte(id.LocalPort), @@ -128,51 +214,50 @@ func (ep *multiPortEndpoint) selectEndpoint(id TransportEndpointID) TransportEnd byte(id.RemotePort >> 8), } - h := jenkins.Sum32(ep.seed) + h := jenkins.Sum32(seed) h.Write(payload) h.Write([]byte(id.LocalAddress)) h.Write([]byte(id.RemoteAddress)) hash := h.Sum32() - idx := reciprocalScale(hash, uint32(len(ep.endpointsArr))) - return ep.endpointsArr[idx] + idx := reciprocalScale(hash, uint32(len(mpep.endpointsArr))) + return mpep.endpointsArr[idx] } -// HandlePacket is called by the stack when new packets arrive to this transport -// endpoint. -func (ep *multiPortEndpoint) HandlePacket(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { - // If this is a broadcast or multicast datagram, deliver the datagram to all - // endpoints managed by ep. - if id.LocalAddress == header.IPv4Broadcast || header.IsV4MulticastAddress(id.LocalAddress) || header.IsV6MulticastAddress(id.LocalAddress) { - for i, endpoint := range ep.endpointsArr { - // HandlePacket modifies vv, so each endpoint needs its own copy. - if i == len(ep.endpointsArr)-1 { - endpoint.HandlePacket(r, id, vv) - break - } - vvCopy := buffer.NewView(vv.Size()) - copy(vvCopy, vv.ToView()) - endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView()) +func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, vv buffer.VectorisedView) { + ep.mu.RLock() + for i, endpoint := range ep.endpointsArr { + // HandlePacket modifies vv, so each endpoint needs its own copy except for + // the final one. + if i == len(ep.endpointsArr)-1 { + endpoint.HandlePacket(r, id, vv) + break } - } else { - ep.selectEndpoint(id).HandlePacket(r, id, vv) + vvCopy := buffer.NewView(vv.Size()) + copy(vvCopy, vv.ToView()) + endpoint.HandlePacket(r, id, vvCopy.ToVectorisedView()) } + ep.mu.RUnlock() // Don't use defer for performance reasons. } -// HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. -func (ep *multiPortEndpoint) HandleControlPacket(id TransportEndpointID, typ ControlType, extra uint32, vv buffer.VectorisedView) { - ep.selectEndpoint(id).HandleControlPacket(id, typ, extra, vv) -} - -func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint) { +// singleRegisterEndpoint tries to add an endpoint to the multiPortEndpoint +// list. The list might be empty already. +func (ep *multiPortEndpoint) singleRegisterEndpoint(t TransportEndpoint, reusePort bool) *tcpip.Error { ep.mu.Lock() defer ep.mu.Unlock() - // A new endpoint is added into endpointsArr and its index there is - // saved in endpointsMap. This will allows to remove endpoint from - // the array fast. + if len(ep.endpointsArr) > 0 { + // If it was previously bound, we need to check if we can bind again. + if !ep.reuse || !reusePort { + return tcpip.ErrPortInUse + } + } + + // A new endpoint is added into endpointsArr and its index there is saved in + // endpointsMap. This will allow us to remove endpoint from the array fast. ep.endpointsMap[t] = len(ep.endpointsArr) ep.endpointsArr = append(ep.endpointsArr, t) + return nil } // unregisterEndpoint returns true if multiPortEndpoint has to be unregistered. @@ -197,53 +282,41 @@ func (ep *multiPortEndpoint) unregisterEndpoint(t TransportEndpoint) bool { return true } -func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool) *tcpip.Error { +func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { if id.RemotePort != 0 { + // TODO(eyalsoha): Why? reusePort = false } eps, ok := d.protocol[protocolIDs{netProto, protocol}] if !ok { - return nil + return tcpip.ErrUnknownProtocol } eps.mu.Lock() defer eps.mu.Unlock() - var multiPortEp *multiPortEndpoint - if _, ok := eps.endpoints[id]; ok { - if !reusePort { - return tcpip.ErrPortInUse - } - multiPortEp, ok = eps.endpoints[id].(*multiPortEndpoint) - if !ok { - return tcpip.ErrPortInUse - } + if epsByNic, ok := eps.endpoints[id]; ok { + // There was already a binding. + return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) } - if reusePort { - if multiPortEp == nil { - multiPortEp = &multiPortEndpoint{} - multiPortEp.endpointsMap = make(map[TransportEndpoint]int) - multiPortEp.seed = rand.Uint32() - eps.endpoints[id] = multiPortEp - } - - multiPortEp.singleRegisterEndpoint(ep) - - return nil + // This is a new binding. + epsByNic := &endpointsByNic{ + endpoints: make(map[tcpip.NICID]*multiPortEndpoint), + seed: rand.Uint32(), } - eps.endpoints[id] = ep + eps.endpoints[id] = epsByNic - return nil + return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) } // unregisterEndpoint unregisters the endpoint with the given id such that it // won't receive any more packets. -func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint) { +func (d *transportDemuxer) unregisterEndpoint(netProtos []tcpip.NetworkProtocolNumber, protocol tcpip.TransportProtocolNumber, id TransportEndpointID, ep TransportEndpoint, bindToDevice tcpip.NICID) { for _, n := range netProtos { if eps, ok := d.protocol[protocolIDs{n, protocol}]; ok { - eps.unregisterEndpoint(id, ep) + eps.unregisterEndpoint(id, ep, bindToDevice) } } } @@ -265,23 +338,14 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto return false } - // If a sender bound to the Loopback interface sends a broadcast, - // that broadcast must not be delivered to the sender. - if loopbackSubnet.Contains(r.RemoteAddress) && r.LocalAddress == header.IPv4Broadcast && id.LocalPort == id.RemotePort { - return false - } - - // If the packet is a broadcast, then find all matching transport endpoints. - // Otherwise, try to find a single matching transport endpoint. - destEps := make([]TransportEndpoint, 0, 1) eps.mu.RLock() - if protocol == header.UDPProtocolNumber && id.LocalAddress == header.IPv4Broadcast { - for epID, endpoint := range eps.endpoints { - if epID.LocalPort == id.LocalPort { - destEps = append(destEps, endpoint) - } - } + // Determine which transport endpoint or endpoints to deliver this packet to. + // If the packet is a broadcast or multicast, then find all matching + // transport endpoints. + var destEps []*endpointsByNic + if protocol == header.UDPProtocolNumber && isMulticastOrBroadcast(id.LocalAddress) { + destEps = d.findAllEndpointsLocked(eps, vv, id) } else if ep := d.findEndpointLocked(eps, vv, id); ep != nil { destEps = append(destEps, ep) } @@ -299,7 +363,7 @@ func (d *transportDemuxer) deliverPacket(r *Route, protocol tcpip.TransportProto // Deliver the packet. for _, ep := range destEps { - ep.HandlePacket(r, id, vv) + ep.handlePacket(r, id, vv) } return true @@ -331,7 +395,7 @@ func (d *transportDemuxer) deliverRawPacket(r *Route, protocol tcpip.TransportPr // deliverControlPacket attempts to deliver the given control packet. Returns // true if it found an endpoint, false otherwise. -func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool { +func (d *transportDemuxer) deliverControlPacket(n *NIC, net tcpip.NetworkProtocolNumber, trans tcpip.TransportProtocolNumber, typ ControlType, extra uint32, vv buffer.VectorisedView, id TransportEndpointID) bool { eps, ok := d.protocol[protocolIDs{net, trans}] if !ok { return false @@ -348,15 +412,16 @@ func (d *transportDemuxer) deliverControlPacket(net tcpip.NetworkProtocolNumber, } // Deliver the packet. - ep.HandleControlPacket(id, typ, extra, vv) + ep.handleControlPacket(n, id, typ, extra, vv) return true } -func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) TransportEndpoint { +func (d *transportDemuxer) findAllEndpointsLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) []*endpointsByNic { + var matchedEPs []*endpointsByNic // Try to find a match with the id as provided. if ep, ok := eps.endpoints[id]; ok { - return ep + matchedEPs = append(matchedEPs, ep) } // Try to find a match with the id minus the local address. @@ -364,7 +429,7 @@ func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer nid.LocalAddress = "" if ep, ok := eps.endpoints[nid]; ok { - return ep + matchedEPs = append(matchedEPs, ep) } // Try to find a match with the id minus the remote part. @@ -372,15 +437,24 @@ func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer nid.RemoteAddress = "" nid.RemotePort = 0 if ep, ok := eps.endpoints[nid]; ok { - return ep + matchedEPs = append(matchedEPs, ep) } // Try to find a match with only the local port. nid.LocalAddress = "" if ep, ok := eps.endpoints[nid]; ok { - return ep + matchedEPs = append(matchedEPs, ep) } + return matchedEPs +} + +// findEndpointLocked returns the endpoint that most closely matches the given +// id. +func (d *transportDemuxer) findEndpointLocked(eps *transportEndpoints, vv buffer.VectorisedView, id TransportEndpointID) *endpointsByNic { + if matchedEPs := d.findAllEndpointsLocked(eps, vv, id); len(matchedEPs) > 0 { + return matchedEPs[0] + } return nil } @@ -418,3 +492,7 @@ func (d *transportDemuxer) unregisterRawEndpoint(netProto tcpip.NetworkProtocolN } } } + +func isMulticastOrBroadcast(addr tcpip.Address) bool { + return addr == header.IPv4Broadcast || header.IsV4MulticastAddress(addr) || header.IsV6MulticastAddress(addr) +} diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go new file mode 100644 index 000000000..210233dc0 --- /dev/null +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -0,0 +1,352 @@ +// 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 stack_test + +import ( + "math" + "math/rand" + "testing" + + "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/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" + "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/udp" + "gvisor.dev/gvisor/pkg/waiter" +) + +const ( + stackV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" + testV6Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" + + stackAddr = "\x0a\x00\x00\x01" + stackPort = 1234 + testPort = 4096 +) + +type testContext struct { + t *testing.T + linkEPs map[string]*channel.Endpoint + s *stack.Stack + + ep tcpip.Endpoint + wq waiter.Queue +} + +func (c *testContext) cleanup() { + if c.ep != nil { + c.ep.Close() + } +} + +func (c *testContext) createV6Endpoint(v6only bool) { + var err *tcpip.Error + c.ep, err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &c.wq) + if err != nil { + c.t.Fatalf("NewEndpoint failed: %v", err) + } + + var v tcpip.V6OnlyOption + if v6only { + v = 1 + } + if err := c.ep.SetSockOpt(v); err != nil { + c.t.Fatalf("SetSockOpt failed: %v", err) + } +} + +// newDualTestContextMultiNic creates the testing context and also linkEpNames +// named NICs. +func newDualTestContextMultiNic(t *testing.T, mtu uint32, linkEpNames []string) *testContext { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + linkEPs := make(map[string]*channel.Endpoint) + for i, linkEpName := range linkEpNames { + channelEP := channel.New(256, mtu, "") + nicid := tcpip.NICID(i + 1) + if err := s.CreateNamedNIC(nicid, linkEpName, channelEP); err != nil { + t.Fatalf("CreateNIC failed: %v", err) + } + linkEPs[linkEpName] = channelEP + + if err := s.AddAddress(nicid, ipv4.ProtocolNumber, stackAddr); err != nil { + t.Fatalf("AddAddress IPv4 failed: %v", err) + } + + if err := s.AddAddress(nicid, ipv6.ProtocolNumber, stackV6Addr); err != nil { + t.Fatalf("AddAddress IPv6 failed: %v", err) + } + } + + s.SetRouteTable([]tcpip.Route{ + { + Destination: header.IPv4EmptySubnet, + NIC: 1, + }, + { + Destination: header.IPv6EmptySubnet, + NIC: 1, + }, + }) + + return &testContext{ + t: t, + s: s, + linkEPs: linkEPs, + } +} + +type headers struct { + srcPort uint16 + dstPort uint16 +} + +func newPayload() []byte { + b := make([]byte, 30+rand.Intn(100)) + for i := range b { + b[i] = byte(rand.Intn(256)) + } + return b +} + +func (c *testContext) sendV6Packet(payload []byte, h *headers, linkEpName string) { + // Allocate a buffer for data and headers. + buf := buffer.NewView(header.UDPMinimumSize + header.IPv6MinimumSize + len(payload)) + copy(buf[len(buf)-len(payload):], payload) + + // Initialize the IP header. + ip := header.IPv6(buf) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(header.UDPMinimumSize + len(payload)), + NextHeader: uint8(udp.ProtocolNumber), + HopLimit: 65, + SrcAddr: testV6Addr, + DstAddr: stackV6Addr, + }) + + // Initialize the UDP header. + u := header.UDP(buf[header.IPv6MinimumSize:]) + u.Encode(&header.UDPFields{ + SrcPort: h.srcPort, + DstPort: h.dstPort, + Length: uint16(header.UDPMinimumSize + len(payload)), + }) + + // Calculate the UDP pseudo-header checksum. + xsum := header.PseudoHeaderChecksum(udp.ProtocolNumber, testV6Addr, stackV6Addr, uint16(len(u))) + + // Calculate the UDP checksum and set it. + xsum = header.Checksum(payload, xsum) + u.SetChecksum(^u.CalculateChecksum(xsum)) + + // Inject packet. + c.linkEPs[linkEpName].Inject(ipv6.ProtocolNumber, buf.ToVectorisedView()) +} + +func TestTransportDemuxerRegister(t *testing.T) { + for _, test := range []struct { + name string + proto tcpip.NetworkProtocolNumber + want *tcpip.Error + }{ + {"failure", ipv6.ProtocolNumber, tcpip.ErrUnknownProtocol}, + {"success", ipv4.ProtocolNumber, nil}, + } { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) + if got, want := s.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{test.proto}, udp.ProtocolNumber, stack.TransportEndpointID{}, nil, false, 0), test.want; got != want { + t.Fatalf("s.RegisterTransportEndpoint(...) = %v, want %v", got, want) + } + }) + } +} + +// TestReuseBindToDevice injects varied packets on input devices and checks that +// the distribution of packets received matches expectations. +func TestDistribution(t *testing.T) { + type endpointSockopts struct { + reuse int + bindToDevice string + } + for _, test := range []struct { + name string + // endpoints will received the inject packets. + endpoints []endpointSockopts + // wantedDistribution is the wanted ratio of packets received on each + // endpoint for each NIC on which packets are injected. + wantedDistributions map[string][]float64 + }{ + { + "BindPortReuse", + // 5 endpoints that all have reuse set. + []endpointSockopts{ + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + endpointSockopts{1, ""}, + }, + map[string][]float64{ + // Injected packets on dev0 get distributed evenly. + "dev0": []float64{0.2, 0.2, 0.2, 0.2, 0.2}, + }, + }, + { + "BindToDevice", + // 3 endpoints with various bindings. + []endpointSockopts{ + endpointSockopts{0, "dev0"}, + endpointSockopts{0, "dev1"}, + endpointSockopts{0, "dev2"}, + }, + map[string][]float64{ + // Injected packets on dev0 go only to the endpoint bound to dev0. + "dev0": []float64{1, 0, 0}, + // Injected packets on dev1 go only to the endpoint bound to dev1. + "dev1": []float64{0, 1, 0}, + // Injected packets on dev2 go only to the endpoint bound to dev2. + "dev2": []float64{0, 0, 1}, + }, + }, + { + "ReuseAndBindToDevice", + // 6 endpoints with various bindings. + []endpointSockopts{ + endpointSockopts{1, "dev0"}, + endpointSockopts{1, "dev0"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, "dev1"}, + endpointSockopts{1, ""}, + }, + map[string][]float64{ + // Injected packets on dev0 get distributed among endpoints bound to + // dev0. + "dev0": []float64{0.5, 0.5, 0, 0, 0, 0}, + // Injected packets on dev1 get distributed among endpoints bound to + // dev1 or unbound. + "dev1": []float64{0, 0, 1. / 3, 1. / 3, 1. / 3, 0}, + // Injected packets on dev999 go only to the unbound. + "dev999": []float64{0, 0, 0, 0, 0, 1}, + }, + }, + } { + t.Run(test.name, func(t *testing.T) { + for device, wantedDistribution := range test.wantedDistributions { + t.Run(device, func(t *testing.T) { + var devices []string + for d := range test.wantedDistributions { + devices = append(devices, d) + } + c := newDualTestContextMultiNic(t, defaultMTU, devices) + defer c.cleanup() + + c.createV6Endpoint(false) + + eps := make(map[tcpip.Endpoint]int) + + pollChannel := make(chan tcpip.Endpoint) + for i, endpoint := range test.endpoints { + // Try to receive the data. + wq := waiter.Queue{} + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + defer close(ch) + + var err *tcpip.Error + ep, err := c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq) + if err != nil { + c.t.Fatalf("NewEndpoint failed: %v", err) + } + eps[ep] = i + + go func(ep tcpip.Endpoint) { + for range ch { + pollChannel <- ep + } + }(ep) + + defer ep.Close() + reusePortOption := tcpip.ReusePortOption(endpoint.reuse) + if err := ep.SetSockOpt(reusePortOption); err != nil { + c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", reusePortOption, i, err) + } + bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice) + if err := ep.SetSockOpt(bindToDeviceOption); err != nil { + c.t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %v", bindToDeviceOption, i, err) + } + if err := ep.Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil { + t.Fatalf("ep.Bind(...) on endpoint %d failed: %v", i, err) + } + } + + npackets := 100000 + nports := 10000 + if got, want := len(test.endpoints), len(wantedDistribution); got != want { + t.Fatalf("got len(test.endpoints) = %d, want %d", got, want) + } + ports := make(map[uint16]tcpip.Endpoint) + stats := make(map[tcpip.Endpoint]int) + for i := 0; i < npackets; i++ { + // Send a packet. + port := uint16(i % nports) + payload := newPayload() + c.sendV6Packet(payload, + &headers{ + srcPort: testPort + port, + dstPort: stackPort}, + device) + + var addr tcpip.FullAddress + ep := <-pollChannel + _, _, err := ep.Read(&addr) + if err != nil { + c.t.Fatalf("Read on endpoint %d failed: %v", eps[ep], err) + } + stats[ep]++ + if i < nports { + ports[uint16(i)] = ep + } else { + // Check that all packets from one client are handled by the same + // socket. + if want, got := ports[port], ep; want != got { + t.Fatalf("Packet sent on port %d expected on endpoint %d but received on endpoint %d", port, eps[want], eps[got]) + } + } + } + + // Check that a packet distribution is as expected. + for ep, i := range eps { + wantedRatio := wantedDistribution[i] + wantedRecv := wantedRatio * float64(npackets) + actualRecv := stats[ep] + actualRatio := float64(stats[ep]) / float64(npackets) + // The deviation is less than 10%. + if math.Abs(actualRatio-wantedRatio) > 0.05 { + t.Errorf("wanted about %.0f%% (%.0f of %d) packets to arrive on endpoint %d, got %.0f%% (%d of %d)", wantedRatio*100, wantedRecv, npackets, i, actualRatio*100, actualRecv, npackets) + } + } + }) + } + }) + } +} diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go index 0e69ac7c8..db290c404 100644 --- a/pkg/tcpip/stack/transport_test.go +++ b/pkg/tcpip/stack/transport_test.go @@ -75,7 +75,7 @@ func (f *fakeTransportEndpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions if err != nil { return 0, nil, err } - if err := f.route.WritePacket(nil /* gso */, hdr, buffer.View(v).ToVectorisedView(), fakeTransNumber, 123); err != nil { + if err := f.route.WritePacket(nil /* gso */, hdr, buffer.View(v).ToVectorisedView(), fakeTransNumber, 123 /* ttl */, false /* useDefaultTTL */); err != nil { return 0, nil, err } @@ -127,7 +127,7 @@ func (f *fakeTransportEndpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // Try to register so that we can start receiving packets. f.id.RemoteAddress = addr.Addr - err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false) + err = f.stack.RegisterTransportEndpoint(0, []tcpip.NetworkProtocolNumber{fakeNetNumber}, fakeTransNumber, f.id, f, false /* reuse */, 0 /* bindToDevice */) if err != nil { return err } @@ -168,7 +168,8 @@ func (f *fakeTransportEndpoint) Bind(a tcpip.FullAddress) *tcpip.Error { fakeTransNumber, stack.TransportEndpointID{LocalAddress: a.Addr}, f, - false, + false, /* reuse */ + 0, /* bindtoDevice */ ); err != nil { return err } @@ -282,9 +283,16 @@ func (f *fakeTransportProtocol) Option(option interface{}) *tcpip.Error { } } +func fakeTransFactory() stack.TransportProtocol { + return &fakeTransportProtocol{} +} + func TestTransportReceive(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -346,7 +354,10 @@ func TestTransportReceive(t *testing.T) { func TestTransportControlReceive(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -414,7 +425,10 @@ func TestTransportControlReceive(t *testing.T) { func TestTransportSend(t *testing.T) { linkEP := channel.New(10, defaultMTU, "") - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) if err := s.CreateNIC(1, linkEP); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -457,7 +471,10 @@ func TestTransportSend(t *testing.T) { } func TestTransportOptions(t *testing.T) { - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) // Try an unsupported transport protocol. if err := s.SetTransportProtocolOption(tcpip.TransportProtocolNumber(99999), fakeTransportGoodOption(false)); err != tcpip.ErrUnknownProtocol { @@ -498,7 +515,10 @@ func TestTransportOptions(t *testing.T) { } func TestTransportForwarding(t *testing.T) { - s := stack.New([]string{"fakeNet"}, []string{"fakeTrans"}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{fakeNetFactory()}, + TransportProtocols: []stack.TransportProtocol{fakeTransFactory()}, + }) s.SetForwarding(true) // TODO(b/123449044): Change this to a channel NIC. @@ -576,9 +596,3 @@ func TestTransportForwarding(t *testing.T) { t.Errorf("Response packet has incorrect source addresss: got = %d, want = 3", src) } } - -func init() { - stack.RegisterTransportProtocolFactory("fakeTrans", func() stack.TransportProtocol { - return &fakeTransportProtocol{} - }) -} diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index c021c67ac..a35d6562e 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -57,6 +57,9 @@ type Error struct { // String implements fmt.Stringer.String. func (e *Error) String() string { + if e == nil { + return "<nil>" + } return e.msg } @@ -495,6 +498,10 @@ type ReuseAddressOption int // to be bound to an identical socket address. type ReusePortOption int +// BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets +// should bind only on a specific NIC. +type BindToDeviceOption string + // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. type QuickAckOption int @@ -546,6 +553,12 @@ type ModerateReceiveBufferOption bool // Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. type MaxSegOption int +// TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop +// limit value for unicast messages. The default is protocol specific. +// +// A zero value indicates the default. +type TTLOption uint8 + // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default // TTL value for multicast messages. The default is 1. type MulticastTTLOption uint8 @@ -587,6 +600,10 @@ type OutOfBandInlineOption int // datagram sockets are allowed to send packets to a broadcast address. type BroadcastOption int +// DefaultTTLOption is used by stack.(*Stack).NetworkProtocolOption to specify +// a default TTL. +type DefaultTTLOption uint8 + // Route is a row in the routing table. It specifies through which NIC (and // gateway) sets of packets should be routed. A row is considered viable if the // masked target address matches the destination address in the row. @@ -1091,6 +1108,47 @@ func (a AddressWithPrefix) String() string { return fmt.Sprintf("%s/%d", a.Address, a.PrefixLen) } +// Subnet converts the address and prefix into a Subnet value and returns it. +func (a AddressWithPrefix) Subnet() Subnet { + addrLen := len(a.Address) + if a.PrefixLen <= 0 { + return Subnet{ + address: Address(strings.Repeat("\x00", addrLen)), + mask: AddressMask(strings.Repeat("\x00", addrLen)), + } + } + if a.PrefixLen >= addrLen*8 { + return Subnet{ + address: a.Address, + mask: AddressMask(strings.Repeat("\xff", addrLen)), + } + } + + sa := make([]byte, addrLen) + sm := make([]byte, addrLen) + n := uint(a.PrefixLen) + for i := 0; i < addrLen; i++ { + if n >= 8 { + sa[i] = a.Address[i] + sm[i] = 0xff + n -= 8 + continue + } + sm[i] = ^byte(0xff >> n) + sa[i] = a.Address[i] & sm[i] + n = 0 + } + + // For extra caution, call NewSubnet rather than directly creating the Subnet + // value. If that fails it indicates a serious bug in this code, so panic is + // in order. + s, err := NewSubnet(Address(sa), AddressMask(sm)) + if err != nil { + panic("invalid subnet: " + err.Error()) + } + return s +} + // ProtocolAddress is an address and the network protocol it is associated // with. type ProtocolAddress struct { diff --git a/pkg/tcpip/tcpip_test.go b/pkg/tcpip/tcpip_test.go index fb3a0a5ee..8c0aacffa 100644 --- a/pkg/tcpip/tcpip_test.go +++ b/pkg/tcpip/tcpip_test.go @@ -195,3 +195,34 @@ func TestStatsString(t *testing.T) { t.Logf(`got = fmt.Sprintf("%%+v", Stats{}.FillIn()) = %q`, got) } } + +func TestAddressWithPrefixSubnet(t *testing.T) { + tests := []struct { + addr Address + prefixLen int + subnetAddr Address + subnetMask AddressMask + }{ + {"\xaa\x55\x33\x42", -1, "\x00\x00\x00\x00", "\x00\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 0, "\x00\x00\x00\x00", "\x00\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 1, "\x80\x00\x00\x00", "\x80\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 7, "\xaa\x00\x00\x00", "\xfe\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 8, "\xaa\x00\x00\x00", "\xff\x00\x00\x00"}, + {"\xaa\x55\x33\x42", 24, "\xaa\x55\x33\x00", "\xff\xff\xff\x00"}, + {"\xaa\x55\x33\x42", 31, "\xaa\x55\x33\x42", "\xff\xff\xff\xfe"}, + {"\xaa\x55\x33\x42", 32, "\xaa\x55\x33\x42", "\xff\xff\xff\xff"}, + {"\xaa\x55\x33\x42", 33, "\xaa\x55\x33\x42", "\xff\xff\xff\xff"}, + } + for _, tt := range tests { + ap := AddressWithPrefix{Address: tt.addr, PrefixLen: tt.prefixLen} + gotSubnet := ap.Subnet() + wantSubnet, err := NewSubnet(tt.subnetAddr, tt.subnetMask) + if err != nil { + t.Error("NewSubnet(%q, %q) failed: %s", tt.subnetAddr, tt.subnetMask, err) + continue + } + if gotSubnet != wantSubnet { + t.Errorf("got subnet = %q, want = %q", gotSubnet, wantSubnet) + } + } +} diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index a111fdb2a..e17e737e4 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -84,6 +84,7 @@ type endpoint struct { // NIC. regNICID tcpip.NICID route stack.Route `state:"manual"` + ttl uint8 } func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { @@ -104,7 +105,7 @@ func (e *endpoint) Close() { e.shutdownFlags = tcpip.ShutdownRead | tcpip.ShutdownWrite switch e.state { case stateBound, stateConnected: - e.stack.UnregisterTransportEndpoint(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.transProto, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, []tcpip.NetworkProtocolNumber{e.netProto}, e.transProto, e.id, e, 0 /* bindToDevice */) } // Close the receive list and drain it. @@ -296,10 +297,10 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c switch e.netProto { case header.IPv4ProtocolNumber: - err = send4(route, e.id.LocalPort, v) + err = send4(route, e.id.LocalPort, v, e.ttl) case header.IPv6ProtocolNumber: - err = send6(route, e.id.LocalPort, v) + err = send6(route, e.id.LocalPort, v, e.ttl) } if err != nil { @@ -314,8 +315,15 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { return 0, tcpip.ControlMessages{}, nil } -// SetSockOpt sets a socket option. Currently not supported. +// SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch o := opt.(type) { + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(o) + e.mu.Unlock() + } + return nil } @@ -362,12 +370,18 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { *o = 0 return nil + case *tcpip.TTLOption: + e.rcvMu.Lock() + *o = tcpip.TTLOption(e.ttl) + e.rcvMu.Unlock() + return nil + default: return tcpip.ErrUnknownProtocolOption } } -func send4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { +func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { if len(data) < header.ICMPv4MinimumSize { return tcpip.ErrInvalidEndpointState } @@ -389,10 +403,10 @@ func send4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { icmpv4.SetChecksum(0) icmpv4.SetChecksum(^header.Checksum(icmpv4, header.Checksum(data, 0))) - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv4ProtocolNumber, r.DefaultTTL()) + return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv4ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */) } -func send6(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { +func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error { if len(data) < header.ICMPv6EchoMinimumSize { return tcpip.ErrInvalidEndpointState } @@ -412,7 +426,7 @@ func send6(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error { icmpv6.SetChecksum(0) icmpv6.SetChecksum(^header.Checksum(icmpv6, header.Checksum(data, 0))) - return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv6ProtocolNumber, r.DefaultTTL()) + return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv6ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */) } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { @@ -543,14 +557,14 @@ func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.Networ if id.LocalPort != 0 { // The endpoint already has a local port, just attempt to // register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false /* reuse */, 0 /* bindToDevice */) return id, err } // We need to find a port for the endpoint. _, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { id.LocalPort = p - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, e.transProto, id, e, false /* reuse */, 0 /* bindtodevice */) switch err { case nil: return true, nil diff --git a/pkg/tcpip/transport/icmp/protocol.go b/pkg/tcpip/transport/icmp/protocol.go index 1eb790932..bfb16f7c3 100644 --- a/pkg/tcpip/transport/icmp/protocol.go +++ b/pkg/tcpip/transport/icmp/protocol.go @@ -14,10 +14,9 @@ // Package icmp contains the implementation of the ICMP and IPv6-ICMP transport // protocols for use in ping. To use it in the networking stack, this package -// must be added to the project, and -// activated on the stack by passing icmp.ProtocolName (or "icmp") and/or -// icmp.ProtocolName6 (or "icmp6") as one of the transport protocols when -// calling stack.New(). Then endpoints can be created by passing +// must be added to the project, and activated on the stack by passing +// icmp.NewProtocol4() and/or icmp.NewProtocol6() as one of the transport +// protocols when calling stack.New(). Then endpoints can be created by passing // icmp.ProtocolNumber or icmp.ProtocolNumber6 as the transport protocol number // when calling Stack.NewEndpoint(). package icmp @@ -34,15 +33,9 @@ import ( ) const ( - // ProtocolName4 is the string representation of the icmp protocol name. - ProtocolName4 = "icmp4" - // ProtocolNumber4 is the ICMP protocol number. ProtocolNumber4 = header.ICMPv4ProtocolNumber - // ProtocolName6 is the string representation of the icmp protocol name. - ProtocolName6 = "icmp6" - // ProtocolNumber6 is the IPv6-ICMP protocol number. ProtocolNumber6 = header.ICMPv6ProtocolNumber ) @@ -125,12 +118,12 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } -func init() { - stack.RegisterTransportProtocolFactory(ProtocolName4, func() stack.TransportProtocol { - return &protocol{ProtocolNumber4} - }) +// NewProtocol4 returns an ICMPv4 transport protocol. +func NewProtocol4() stack.TransportProtocol { + return &protocol{ProtocolNumber4} +} - stack.RegisterTransportProtocolFactory(ProtocolName6, func() stack.TransportProtocol { - return &protocol{ProtocolNumber6} - }) +// NewProtocol6 returns an ICMPv6 transport protocol. +func NewProtocol6() stack.TransportProtocol { + return &protocol{ProtocolNumber6} } diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index a02731a5d..8ae132efa 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -332,7 +332,7 @@ func (ep *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64, break } hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength())) - if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, route.DefaultTTL()); err != nil { + if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, 0, true /* useDefaultTTL */); err != nil { return 0, nil, err } diff --git a/pkg/tcpip/transport/raw/protocol.go b/pkg/tcpip/transport/raw/protocol.go index 783c21e6b..a2512d666 100644 --- a/pkg/tcpip/transport/raw/protocol.go +++ b/pkg/tcpip/transport/raw/protocol.go @@ -20,13 +20,10 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) -type factory struct{} +// EndpointFactory implements stack.UnassociatedEndpointFactory. +type EndpointFactory struct{} // NewUnassociatedRawEndpoint implements stack.UnassociatedEndpointFactory. -func (factory) NewUnassociatedRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { +func (EndpointFactory) NewUnassociatedRawEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { return newEndpoint(stack, netProto, transProto, waiterQueue, false /* associated */) } - -func init() { - stack.RegisterUnassociatedFactory(factory{}) -} diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 39a839ab7..a42e1f4a2 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -49,6 +49,7 @@ go_library( "//pkg/sleep", "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/hash/jenkins", "//pkg/tcpip/header", "//pkg/tcpip/iptables", "//pkg/tcpip/seqnum", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 0802e984e..b8b4bcee8 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -242,7 +242,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i n.initGSO() // Register new endpoint so that packets are routed to it. - if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.id, n, n.reusePort); err != nil { + if err := n.stack.RegisterTransportEndpoint(n.boundNICID, n.effectiveNetProtos, ProtocolNumber, n.id, n, n.reusePort, n.bindToDevice); err != nil { n.Close() return nil, err } @@ -439,7 +439,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { TSEcr: opts.TSVal, MSS: uint16(mss), } - sendSynTCP(&s.route, s.id, header.TCPFlagSyn|header.TCPFlagAck, cookie, s.sequenceNumber+1, ctx.rcvWnd, synOpts) + sendSynTCP(&s.route, s.id, e.ttl, header.TCPFlagSyn|header.TCPFlagAck, cookie, s.sequenceNumber+1, ctx.rcvWnd, synOpts) e.stack.Stats().TCP.ListenOverflowSynCookieSent.Increment() } diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 21038a65a..1d6e7f5f3 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -238,6 +238,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { h.state = handshakeSynRcvd h.ep.mu.Lock() h.ep.state = StateSynRecv + ttl := h.ep.ttl h.ep.mu.Unlock() synOpts := header.TCPSynOptions{ WS: int(h.effectiveRcvWndScale()), @@ -251,7 +252,10 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { SACKPermitted: rcvSynOpts.SACKPermitted, MSS: h.ep.amss, } - sendSynTCP(&s.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + if ttl == 0 { + ttl = s.route.DefaultTTL() + } + sendSynTCP(&s.route, h.ep.id, ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) return nil } @@ -296,7 +300,7 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { SACKPermitted: h.ep.sackPermitted, MSS: h.ep.amss, } - sendSynTCP(&s.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + sendSynTCP(&s.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) return nil } @@ -460,7 +464,7 @@ func (h *handshake) execute() *tcpip.Error { synOpts.WS = -1 } } - sendSynTCP(&h.ep.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + sendSynTCP(&h.ep.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) for h.state != handshakeCompleted { switch index, _ := s.Fetch(true); index { case wakerForResend: @@ -469,7 +473,7 @@ func (h *handshake) execute() *tcpip.Error { return tcpip.ErrTimeout } rt.Reset(timeOut) - sendSynTCP(&h.ep.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) + sendSynTCP(&h.ep.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts) case wakerForNotification: n := h.ep.fetchNotifications() @@ -579,9 +583,9 @@ func makeSynOptions(opts header.TCPSynOptions) []byte { return options[:offset] } -func sendSynTCP(r *stack.Route, id stack.TransportEndpointID, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts header.TCPSynOptions) *tcpip.Error { +func sendSynTCP(r *stack.Route, id stack.TransportEndpointID, ttl uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts header.TCPSynOptions) *tcpip.Error { options := makeSynOptions(opts) - err := sendTCP(r, id, buffer.VectorisedView{}, r.DefaultTTL(), flags, seq, ack, rcvWnd, options, nil) + err := sendTCP(r, id, buffer.VectorisedView{}, ttl, flags, seq, ack, rcvWnd, options, nil) putOptions(options) return err } @@ -629,7 +633,7 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise r.Stats().TCP.ResetsSent.Increment() } - return r.WritePacket(gso, hdr, data, ProtocolNumber, ttl) + return r.WritePacket(gso, hdr, data, ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */) } // makeOptions makes an options slice. @@ -678,7 +682,7 @@ func (e *endpoint) sendRaw(data buffer.VectorisedView, flags byte, seq, ack seqn sackBlocks = e.sack.Blocks[:e.sack.NumBlocks] } options := e.makeOptions(sackBlocks) - err := sendTCP(&e.route, e.id, data, e.route.DefaultTTL(), flags, seq, ack, rcvWnd, options, e.gso) + err := sendTCP(&e.route, e.id, data, e.ttl, flags, seq, ack, rcvWnd, options, e.gso) putOptions(options) return err } diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 35b489c68..83d92b3e1 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -15,6 +15,7 @@ package tcp import ( + "encoding/binary" "fmt" "math" "strings" @@ -26,6 +27,7 @@ import ( "gvisor.dev/gvisor/pkg/sleep" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/iptables" "gvisor.dev/gvisor/pkg/tcpip/seqnum" @@ -226,6 +228,7 @@ type endpoint struct { isRegistered bool boundNICID tcpip.NICID `state:"manual"` route stack.Route `state:"manual"` + ttl uint8 v6only bool isConnectNotified bool // TCP should never broadcast but Linux nevertheless supports enabling/ @@ -280,6 +283,9 @@ type endpoint struct { // reusePort is set to true if SO_REUSEPORT is enabled. reusePort bool + // bindToDevice is set to the NIC on which to bind or disabled if 0. + bindToDevice tcpip.NICID + // delay enables Nagle's algorithm. // // delay is a boolean (0 is false) and must be accessed atomically. @@ -564,11 +570,11 @@ func (e *endpoint) Close() { // in Listen() when trying to register. if e.state == StateListen && e.isPortReserved { if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.isRegistered = false } - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -625,12 +631,12 @@ func (e *endpoint) cleanupLocked() { e.workerCleanup = false if e.isRegistered { - e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.isRegistered = false } if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -1060,6 +1066,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Unlock() return nil + case tcpip.BindToDeviceOption: + e.mu.Lock() + defer e.mu.Unlock() + if v == "" { + e.bindToDevice = 0 + return nil + } + for nicid, nic := range e.stack.NICInfo() { + if nic.Name == string(v) { + e.bindToDevice = nicid + return nil + } + } + return tcpip.ErrUnknownDevice + case tcpip.QuickAckOption: if v == 0 { atomic.StoreUint32(&e.slowAck, 1) @@ -1096,6 +1117,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.v6only = v != 0 return nil + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(v) + e.mu.Unlock() + return nil + case tcpip.KeepaliveEnabledOption: e.keepalive.Lock() e.keepalive.enabled = v != 0 @@ -1260,6 +1287,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.BindToDeviceOption: + e.mu.RLock() + defer e.mu.RUnlock() + if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { + *o = tcpip.BindToDeviceOption(nic.Name) + return nil + } + *o = "" + return nil + case *tcpip.QuickAckOption: *o = 1 if v := atomic.LoadUint32(&e.slowAck); v != 0 { @@ -1283,6 +1320,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.TTLOption: + e.mu.Lock() + *o = tcpip.TTLOption(e.ttl) + e.mu.Unlock() + return nil + case *tcpip.TCPInfoOption: *o = tcpip.TCPInfoOption{} e.mu.RLock() @@ -1466,7 +1509,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er if e.id.LocalPort != 0 { // The endpoint is bound to a port, attempt to register it. - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, e.id, e, e.reusePort, e.bindToDevice) if err != nil { return err } @@ -1476,17 +1519,32 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er // address/port for both local and remote (otherwise this // endpoint would be trying to connect to itself). sameAddr := e.id.LocalAddress == e.id.RemoteAddress - if _, err := e.stack.PickEphemeralPort(func(p uint16) (bool, *tcpip.Error) { + + // Calculate a port offset based on the destination IP/port and + // src IP to ensure that for a given tuple (srcIP, destIP, + // destPort) the offset used as a starting point is the same to + // ensure that we can cycle through the port space effectively. + h := jenkins.Sum32(e.stack.PortSeed()) + h.Write([]byte(e.id.LocalAddress)) + h.Write([]byte(e.id.RemoteAddress)) + portBuf := make([]byte, 2) + binary.LittleEndian.PutUint16(portBuf, e.id.RemotePort) + h.Write(portBuf) + portOffset := h.Sum32() + + if _, err := e.stack.PickEphemeralPortStable(portOffset, func(p uint16) (bool, *tcpip.Error) { if sameAddr && p == e.id.RemotePort { return false, nil } - if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false) { + // reusePort is false below because connect cannot reuse a port even if + // reusePort was set. + if !e.stack.IsPortAvailable(netProtos, ProtocolNumber, e.id.LocalAddress, p, false /* reusePort */, e.bindToDevice) { return false, nil } id := e.id id.LocalPort = p - switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) { + switch e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) { case nil: e.id = id return true, nil @@ -1504,7 +1562,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er // before Connect: in such a case we don't want to hold on to // reservations anymore. if e.isPortReserved { - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, origID.LocalAddress, origID.LocalPort, e.bindToDevice) e.isPortReserved = false } @@ -1648,7 +1706,7 @@ func (e *endpoint) Listen(backlog int) (err *tcpip.Error) { } // Register the endpoint. - if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort); err != nil { + if err := e.stack.RegisterTransportEndpoint(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.reusePort, e.bindToDevice); err != nil { return err } @@ -1729,7 +1787,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { } } - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, addr.Addr, addr.Port, e.reusePort, e.bindToDevice) if err != nil { return err } @@ -1739,16 +1797,16 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) { e.id.LocalPort = port // Any failures beyond this point must remove the port registration. - defer func() { + defer func(bindToDevice tcpip.NICID) { if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port) + e.stack.ReleasePort(netProtos, ProtocolNumber, addr.Addr, port, bindToDevice) e.isPortReserved = false e.effectiveNetProtos = nil e.id.LocalPort = 0 e.id.LocalAddress = "" e.boundNICID = 0 } - }() + }(e.bindToDevice) // If an address is specified, we must ensure that it's one of our // local addresses. diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 2a13b2022..d5d8ab96a 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -14,7 +14,7 @@ // Package tcp contains the implementation of the TCP transport protocol. To use // it in the networking stack, this package must be added to the project, and -// activated on the stack by passing tcp.ProtocolName (or "tcp") as one of the +// activated on the stack by passing tcp.NewProtocol() as one of the // transport protocols when calling stack.New(). Then endpoints can be created // by passing tcp.ProtocolNumber as the transport protocol number when calling // Stack.NewEndpoint(). @@ -34,9 +34,6 @@ import ( ) const ( - // ProtocolName is the string representation of the tcp protocol name. - ProtocolName = "tcp" - // ProtocolNumber is the tcp protocol number. ProtocolNumber = header.TCPProtocolNumber @@ -254,13 +251,12 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { } } -func init() { - stack.RegisterTransportProtocolFactory(ProtocolName, func() stack.TransportProtocol { - return &protocol{ - sendBufferSize: SendBufferSizeOption{MinBufferSize, DefaultSendBufferSize, MaxBufferSize}, - recvBufferSize: ReceiveBufferSizeOption{MinBufferSize, DefaultReceiveBufferSize, MaxBufferSize}, - congestionControl: ccReno, - availableCongestionControl: []string{ccReno, ccCubic}, - } - }) +// NewProtocol returns a TCP transport protocol. +func NewProtocol() stack.TransportProtocol { + return &protocol{ + sendBufferSize: SendBufferSizeOption{MinBufferSize, DefaultSendBufferSize, MaxBufferSize}, + recvBufferSize: ReceiveBufferSizeOption{MinBufferSize, DefaultReceiveBufferSize, MaxBufferSize}, + congestionControl: ccReno, + availableCongestionControl: []string{ccReno, ccCubic}, + } } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 7fa5cfb6e..a86123829 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -465,6 +465,66 @@ func TestSimpleReceive(t *testing.T) { ) } +func TestConnectBindToDevice(t *testing.T) { + for _, test := range []struct { + name string + device string + want tcp.EndpointState + }{ + {"RightDevice", "nic1", tcp.StateEstablished}, + {"WrongDevice", "nic2", tcp.StateSynSent}, + {"AnyDevice", "", tcp.StateEstablished}, + } { + t.Run(test.name, func(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + c.Create(-1) + bindToDevice := tcpip.BindToDeviceOption(test.device) + c.EP.SetSockOpt(bindToDevice) + // Start connection attempt. + waitEntry, _ := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&waitEntry, waiter.EventOut) + defer c.WQ.EventUnregister(&waitEntry) + + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet. + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + ), + ) + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { + t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) + } + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + iss := seqnum.Value(789) + rcvWnd := seqnum.Size(30000) + c.SendPacket(nil, &context.Headers{ + SrcPort: tcpHdr.DestinationPort(), + DstPort: tcpHdr.SourcePort(), + Flags: header.TCPFlagSyn | header.TCPFlagAck, + SeqNum: iss, + AckNum: c.IRS.Add(1), + RcvWnd: rcvWnd, + TCPOpts: nil, + }) + + c.GetPacket() + if got, want := tcp.EndpointState(c.EP.State()), test.want; got != want { + t.Fatalf("Unexpected endpoint state: want %v, got %v", want, got) + } + }) + } +} + func TestOutOfOrderReceive(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -1692,6 +1752,34 @@ func TestSendGreaterThanMTU(t *testing.T) { testBrokenUpWrite(t, c, maxPayload) } +func TestSetTTL(t *testing.T) { + for _, wantTTL := range []uint8{1, 2, 50, 64, 128, 254, 255} { + t.Run(fmt.Sprintf("TTL:%d", wantTTL), func(t *testing.T) { + c := context.New(t, 65535) + defer c.Cleanup() + + var err *tcpip.Error + c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed: %v", err) + } + + if err := c.EP.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { + t.Fatalf("SetSockOpt failed: %v", err) + } + + if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { + t.Fatalf("Unexpected return value from Connect: %v", err) + } + + // Receive SYN packet. + b := c.GetPacket() + + checker.IPv4(t, b, checker.TTL(wantTTL)) + }) + } +} + func TestActiveSendMSSLessThanMTU(t *testing.T) { const maxPayload = 100 c := context.New(t, 65535) @@ -2873,7 +2961,10 @@ func checkSendBufferSize(t *testing.T, ep tcpip.Endpoint, v int) { } func TestDefaultBufferSizes(t *testing.T) { - s := stack.New([]string{ipv4.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) // Check the default values. ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) @@ -2919,7 +3010,10 @@ func TestDefaultBufferSizes(t *testing.T) { } func TestMinMaxBufferSizes(t *testing.T) { - s := stack.New([]string{ipv4.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) // Check the default values. ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) @@ -2964,11 +3058,70 @@ func TestMinMaxBufferSizes(t *testing.T) { checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) } +func TestBindToDeviceOption(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}}) + + ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed; %v", err) + } + defer ep.Close() + + if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil { + t.Errorf("CreateNamedNIC failed: %v", err) + } + + // Make an nameless NIC. + if err := s.CreateNIC(54321, loopback.New()); err != nil { + t.Errorf("CreateNIC failed: %v", err) + } + + // strPtr is used instead of taking the address of string literals, which is + // a compiler error. + strPtr := func(s string) *string { + return &s + } + + testActions := []struct { + name string + setBindToDevice *string + setBindToDeviceError *tcpip.Error + getBindToDevice tcpip.BindToDeviceOption + }{ + {"GetDefaultValue", nil, nil, ""}, + {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, + {"BindToExistent", strPtr("my_device"), nil, "my_device"}, + {"UnbindToDevice", strPtr(""), nil, ""}, + } + for _, testAction := range testActions { + t.Run(testAction.name, func(t *testing.T) { + if testAction.setBindToDevice != nil { + bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) + if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + } + } + bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") + if ep.GetSockOpt(&bindToDevice) != nil { + t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + } + if got, want := bindToDevice, testAction.getBindToDevice; got != want { + t.Errorf("bindToDevice got %q, want %q", got, want) + } + }) + } +} + func makeStack() (*stack.Stack, *tcpip.Error) { - s := stack.New([]string{ - ipv4.ProtocolName, - ipv6.ProtocolName, - }, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ + ipv4.NewProtocol(), + ipv6.NewProtocol(), + }, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) id := loopback.New() if testing.Verbose() { diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 78eff5c3a..ef823e4ae 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -137,7 +137,10 @@ type Context struct { // New allocates and initializes a test context containing a new // stack and a link-layer endpoint. func New(t *testing.T, mtu uint32) *Context { - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName}, []string{tcp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{tcp.NewProtocol()}, + }) // Allow minimum send/receive buffer sizes to be 1 during tests. if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SendBufferSizeOption{1, tcp.DefaultSendBufferSize, 10 * tcp.DefaultSendBufferSize}); err != nil { @@ -155,7 +158,14 @@ func New(t *testing.T, mtu uint32) *Context { if testing.Verbose() { wep = sniffer.New(ep) } - if err := s.CreateNIC(1, wep); err != nil { + if err := s.CreateNamedNIC(1, "nic1", wep); err != nil { + t.Fatalf("CreateNIC failed: %v", err) + } + wep2 := stack.LinkEndpoint(channel.New(1000, mtu, "")) + if testing.Verbose() { + wep2 = sniffer.New(channel.New(1000, mtu, "")) + } + if err := s.CreateNamedNIC(2, "nic2", wep2); err != nil { t.Fatalf("CreateNIC failed: %v", err) } @@ -585,12 +595,8 @@ func (c *Context) Connect(iss seqnum.Value, rcvWnd seqnum.Size, options []byte) c.Port = tcpHdr.SourcePort() } -// CreateConnectedWithRawOptions creates a connected TCP endpoint and sends -// the specified option bytes as the Option field in the initial SYN packet. -// -// It also sets the receive buffer for the endpoint to the specified -// value in epRcvBuf. -func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf int, options []byte) { +// Create creates a TCP endpoint. +func (c *Context) Create(epRcvBuf int) { // Create TCP endpoint. var err *tcpip.Error c.EP, err = c.s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) @@ -603,6 +609,15 @@ func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum. c.t.Fatalf("SetSockOpt failed failed: %v", err) } } +} + +// CreateConnectedWithRawOptions creates a connected TCP endpoint and sends +// the specified option bytes as the Option field in the initial SYN packet. +// +// It also sets the receive buffer for the endpoint to the specified +// value in epRcvBuf. +func (c *Context) CreateConnectedWithRawOptions(iss seqnum.Value, rcvWnd seqnum.Size, epRcvBuf int, options []byte) { + c.Create(epRcvBuf) c.Connect(iss, rcvWnd, options) } diff --git a/pkg/tcpip/transport/udp/BUILD b/pkg/tcpip/transport/udp/BUILD index c1ca22b35..7a635ab8d 100644 --- a/pkg/tcpip/transport/udp/BUILD +++ b/pkg/tcpip/transport/udp/BUILD @@ -52,6 +52,7 @@ go_test( "//pkg/tcpip/checker", "//pkg/tcpip/header", "//pkg/tcpip/link/channel", + "//pkg/tcpip/link/loopback", "//pkg/tcpip/link/sniffer", "//pkg/tcpip/network/ipv4", "//pkg/tcpip/network/ipv6", diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 0bec7e62d..8ae83437e 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -83,11 +83,13 @@ type endpoint struct { route stack.Route `state:"manual"` dstPort uint16 v6only bool + ttl uint8 multicastTTL uint8 multicastAddr tcpip.Address multicastNICID tcpip.NICID multicastLoop bool reusePort bool + bindToDevice tcpip.NICID broadcast bool // shutdownFlags represent the current shutdown state of the endpoint. @@ -144,8 +146,8 @@ func (e *endpoint) Close() { switch e.state { case StateBound, StateConnected: - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) } for _, mem := range e.multicastMemberships { @@ -373,12 +375,16 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, tcpip.ErrMessageTooLong } - ttl := route.DefaultTTL() + ttl := e.ttl + useDefaultTTL := ttl == 0 + if header.IsV4MulticastAddress(route.RemoteAddress) || header.IsV6MulticastAddress(route.RemoteAddress) { ttl = e.multicastTTL + // Multicast allows a 0 TTL. + useDefaultTTL = false } - if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.id.LocalPort, dstPort, ttl); err != nil { + if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.id.LocalPort, dstPort, ttl, useDefaultTTL); err != nil { return 0, nil, err } return int64(len(v)), nil, nil @@ -413,6 +419,11 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.v6only = v != 0 + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(v) + e.mu.Unlock() + case tcpip.MulticastTTLOption: e.mu.Lock() e.multicastTTL = uint8(v) @@ -551,6 +562,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.reusePort = v != 0 e.mu.Unlock() + case tcpip.BindToDeviceOption: + e.mu.Lock() + defer e.mu.Unlock() + if v == "" { + e.bindToDevice = 0 + return nil + } + for nicid, nic := range e.stack.NICInfo() { + if nic.Name == string(v) { + e.bindToDevice = nicid + return nil + } + } + return tcpip.ErrUnknownDevice + case tcpip.BroadcastOption: e.mu.Lock() e.broadcast = v != 0 @@ -612,6 +638,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.TTLOption: + e.mu.Lock() + *o = tcpip.TTLOption(e.ttl) + e.mu.Unlock() + return nil + case *tcpip.MulticastTTLOption: e.mu.Lock() *o = tcpip.MulticastTTLOption(e.multicastTTL) @@ -646,6 +678,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } return nil + case *tcpip.BindToDeviceOption: + e.mu.RLock() + defer e.mu.RUnlock() + if nic, ok := e.stack.NICInfo()[e.bindToDevice]; ok { + *o = tcpip.BindToDeviceOption(nic.Name) + return nil + } + *o = tcpip.BindToDeviceOption("") + return nil + case *tcpip.KeepaliveEnabledOption: *o = 0 return nil @@ -668,7 +710,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // sendUDP sends a UDP segment via the provided network endpoint and under the // provided identity. -func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8) *tcpip.Error { +func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool) *tcpip.Error { // Allocate a buffer for the UDP header. hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength())) @@ -694,7 +736,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u // Track count of packets sent. r.Stats().UDP.PacketsSent.Increment() - return r.WritePacket(nil /* gso */, hdr, data, ProtocolNumber, ttl) + return r.WritePacket(nil /* gso */, hdr, data, ProtocolNumber, ttl, useDefaultTTL) } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { @@ -753,12 +795,12 @@ func (e *endpoint) Disconnect() *tcpip.Error { } else { if e.id.LocalPort != 0 { // Release the ephemeral port. - e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort) + e.stack.ReleasePort(e.effectiveNetProtos, ProtocolNumber, e.id.LocalAddress, e.id.LocalPort, e.bindToDevice) } e.state = StateInitial } - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) e.id = id e.route.Release() e.route = stack.Route{} @@ -835,7 +877,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { // Remove the old registration. if e.id.LocalPort != 0 { - e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e) + e.stack.UnregisterTransportEndpoint(e.regNICID, e.effectiveNetProtos, ProtocolNumber, e.id, e, e.bindToDevice) } e.id = id @@ -898,16 +940,16 @@ func (*endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { func (e *endpoint) registerWithStack(nicid tcpip.NICID, netProtos []tcpip.NetworkProtocolNumber, id stack.TransportEndpointID) (stack.TransportEndpointID, *tcpip.Error) { if e.id.LocalPort == 0 { - port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort) + port, err := e.stack.ReservePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.reusePort, e.bindToDevice) if err != nil { return id, err } id.LocalPort = port } - err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort) + err := e.stack.RegisterTransportEndpoint(nicid, netProtos, ProtocolNumber, id, e, e.reusePort, e.bindToDevice) if err != nil { - e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort) + e.stack.ReleasePort(netProtos, ProtocolNumber, id.LocalAddress, id.LocalPort, e.bindToDevice) } return id, err } diff --git a/pkg/tcpip/transport/udp/forwarder.go b/pkg/tcpip/transport/udp/forwarder.go index a9edc2c8d..2d0bc5221 100644 --- a/pkg/tcpip/transport/udp/forwarder.go +++ b/pkg/tcpip/transport/udp/forwarder.go @@ -74,7 +74,7 @@ func (r *ForwarderRequest) ID() stack.TransportEndpointID { // CreateEndpoint creates a connected UDP endpoint for the session request. func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) { ep := newEndpoint(r.stack, r.route.NetProto, queue) - if err := r.stack.RegisterTransportEndpoint(r.route.NICID(), []tcpip.NetworkProtocolNumber{r.route.NetProto}, ProtocolNumber, r.id, ep, ep.reusePort); err != nil { + if err := r.stack.RegisterTransportEndpoint(r.route.NICID(), []tcpip.NetworkProtocolNumber{r.route.NetProto}, ProtocolNumber, r.id, ep, ep.reusePort, ep.bindToDevice); err != nil { ep.Close() return nil, err } diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go index 068d9a272..0b24bbc85 100644 --- a/pkg/tcpip/transport/udp/protocol.go +++ b/pkg/tcpip/transport/udp/protocol.go @@ -14,7 +14,7 @@ // Package udp contains the implementation of the UDP transport protocol. To use // it in the networking stack, this package must be added to the project, and -// activated on the stack by passing udp.ProtocolName (or "udp") as one of the +// activated on the stack by passing udp.NewProtocol() as one of the // transport protocols when calling stack.New(). Then endpoints can be created // by passing udp.ProtocolNumber as the transport protocol number when calling // Stack.NewEndpoint(). @@ -30,9 +30,6 @@ import ( ) const ( - // ProtocolName is the string representation of the udp protocol name. - ProtocolName = "udp" - // ProtocolNumber is the udp protocol number. ProtocolNumber = header.UDPProtocolNumber ) @@ -133,7 +130,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv4DstUnreachable) pkt.SetCode(header.ICMPv4PortUnreachable) pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload)) - r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv4ProtocolNumber, r.DefaultTTL()) + r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv4ProtocolNumber, 0, true /* useDefaultTTL */) case header.IPv6AddressSize: if !r.Stack().AllowICMPMessage() { @@ -167,7 +164,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans pkt.SetType(header.ICMPv6DstUnreachable) pkt.SetCode(header.ICMPv6PortUnreachable) pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload)) - r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv6ProtocolNumber, r.DefaultTTL()) + r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv6ProtocolNumber, 0, true /* useDefaultTTL */) } return true } @@ -182,8 +179,7 @@ func (p *protocol) Option(option interface{}) *tcpip.Error { return tcpip.ErrUnknownProtocolOption } -func init() { - stack.RegisterTransportProtocolFactory(ProtocolName, func() stack.TransportProtocol { - return &protocol{} - }) +// NewProtocol returns a UDP transport protocol. +func NewProtocol() stack.TransportProtocol { + return &protocol{} } diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index c6deab892..faa728b68 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -17,7 +17,6 @@ package udp_test import ( "bytes" "fmt" - "math" "math/rand" "testing" "time" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/checker" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" + "gvisor.dev/gvisor/pkg/tcpip/link/loopback" "gvisor.dev/gvisor/pkg/tcpip/link/sniffer" "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" @@ -274,7 +274,10 @@ type testContext struct { func newDualTestContext(t *testing.T, mtu uint32) *testContext { t.Helper() - s := stack.New([]string{ipv4.ProtocolName, ipv6.ProtocolName}, []string{udp.ProtocolName}, stack.Options{}) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + }) ep := channel.New(256, mtu, "") wep := stack.LinkEndpoint(ep) @@ -473,94 +476,67 @@ func newMinPayload(minSize int) []byte { return b } -func TestBindPortReuse(t *testing.T) { - c := newDualTestContext(t, defaultMTU) - defer c.cleanup() - - c.createEndpoint(ipv6.ProtocolNumber) - - var eps [5]tcpip.Endpoint - reusePortOpt := tcpip.ReusePortOption(1) - - pollChannel := make(chan tcpip.Endpoint) - for i := 0; i < len(eps); i++ { - // Try to receive the data. - wq := waiter.Queue{} - we, ch := waiter.NewChannelEntry(nil) - wq.EventRegister(&we, waiter.EventIn) - defer wq.EventUnregister(&we) - defer close(ch) - - var err *tcpip.Error - eps[i], err = c.s.NewEndpoint(udp.ProtocolNumber, ipv6.ProtocolNumber, &wq) - if err != nil { - c.t.Fatalf("NewEndpoint failed: %v", err) - } - - go func(ep tcpip.Endpoint) { - for range ch { - pollChannel <- ep - } - }(eps[i]) +func TestBindToDeviceOption(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}}) - defer eps[i].Close() - if err := eps[i].SetSockOpt(reusePortOpt); err != nil { - c.t.Fatalf("SetSockOpt failed failed: %v", err) - } - if err := eps[i].Bind(tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}); err != nil { - t.Fatalf("ep.Bind(...) failed: %v", err) - } + ep, err := s.NewEndpoint(udp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{}) + if err != nil { + t.Fatalf("NewEndpoint failed; %v", err) } + defer ep.Close() - npackets := 100000 - nports := 10000 - ports := make(map[uint16]tcpip.Endpoint) - stats := make(map[tcpip.Endpoint]int) - for i := 0; i < npackets; i++ { - // Send a packet. - port := uint16(i % nports) - payload := newPayload() - c.injectV6Packet(payload, &header4Tuple{ - srcAddr: tcpip.FullAddress{Addr: testV6Addr, Port: testPort + port}, - dstAddr: tcpip.FullAddress{Addr: stackV6Addr, Port: stackPort}, - }) + if err := s.CreateNamedNIC(321, "my_device", loopback.New()); err != nil { + t.Errorf("CreateNamedNIC failed: %v", err) + } - var addr tcpip.FullAddress - ep := <-pollChannel - _, _, err := ep.Read(&addr) - if err != nil { - c.t.Fatalf("Read failed: %v", err) - } - stats[ep]++ - if i < nports { - ports[uint16(i)] = ep - } else { - // Check that all packets from one client are handled - // by the same socket. - if ports[port] != ep { - t.Fatalf("Port mismatch") - } - } + // Make an nameless NIC. + if err := s.CreateNIC(54321, loopback.New()); err != nil { + t.Errorf("CreateNIC failed: %v", err) } - if len(stats) != len(eps) { - t.Fatalf("Only %d(expected %d) sockets received packets", len(stats), len(eps)) + // strPtr is used instead of taking the address of string literals, which is + // a compiler error. + strPtr := func(s string) *string { + return &s } - // Check that a packet distribution is fair between sockets. - for _, c := range stats { - n := float64(npackets) / float64(len(eps)) - // The deviation is less than 10%. - if math.Abs(float64(c)-n) > n/10 { - t.Fatal(c, n) - } + testActions := []struct { + name string + setBindToDevice *string + setBindToDeviceError *tcpip.Error + getBindToDevice tcpip.BindToDeviceOption + }{ + {"GetDefaultValue", nil, nil, ""}, + {"BindToNonExistent", strPtr("non_existent_device"), tcpip.ErrUnknownDevice, ""}, + {"BindToExistent", strPtr("my_device"), nil, "my_device"}, + {"UnbindToDevice", strPtr(""), nil, ""}, + } + for _, testAction := range testActions { + t.Run(testAction.name, func(t *testing.T) { + if testAction.setBindToDevice != nil { + bindToDevice := tcpip.BindToDeviceOption(*testAction.setBindToDevice) + if got, want := ep.SetSockOpt(bindToDevice), testAction.setBindToDeviceError; got != want { + t.Errorf("SetSockOpt(%v) got %v, want %v", bindToDevice, got, want) + } + } + bindToDevice := tcpip.BindToDeviceOption("to be modified by GetSockOpt") + if ep.GetSockOpt(&bindToDevice) != nil { + t.Errorf("GetSockOpt got %v, want %v", ep.GetSockOpt(&bindToDevice), nil) + } + if got, want := bindToDevice, testAction.getBindToDevice; got != want { + t.Errorf("bindToDevice got %q, want %q", got, want) + } + }) } } -// testRead sends a packet of the given test flow into the stack by injecting it -// into the link endpoint. It then reads it from the UDP endpoint and verifies -// its correctness. -func testRead(c *testContext, flow testFlow) { +// testReadInternal sends a packet of the given test flow into the stack by +// injecting it into the link endpoint. It then attempts to read it from the +// UDP endpoint and depending on if this was expected to succeed verifies its +// correctness. +func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped bool) { c.t.Helper() payload := newPayload() @@ -578,27 +554,51 @@ func testRead(c *testContext, flow testFlow) { select { case <-ch: v, _, err = c.ep.Read(&addr) - if err != nil { - c.t.Fatalf("Read failed: %v", err) - } - case <-time.After(1 * time.Second): - c.t.Fatalf("Timed out waiting for data") + case <-time.After(300 * time.Millisecond): + if packetShouldBeDropped { + return // expected to time out + } + c.t.Fatal("timed out waiting for data") } } + if err != nil { + c.t.Fatal("Read failed:", err) + } + + if packetShouldBeDropped { + c.t.Fatalf("Read unexpectedly received data from %s", addr.Addr) + } + // Check the peer address. h := flow.header4Tuple(incoming) if addr.Addr != h.srcAddr.Addr { - c.t.Fatalf("Unexpected remote address: got %v, want %v", addr.Addr, h.srcAddr) + c.t.Fatalf("unexpected remote address: got %s, want %s", addr.Addr, h.srcAddr) } // Check the payload. if !bytes.Equal(payload, v) { - c.t.Fatalf("Bad payload: got %x, want %x", v, payload) + c.t.Fatalf("bad payload: got %x, want %x", v, payload) } } +// testRead sends a packet of the given test flow into the stack by injecting it +// into the link endpoint. It then reads it from the UDP endpoint and verifies +// its correctness. +func testRead(c *testContext, flow testFlow) { + c.t.Helper() + testReadInternal(c, flow, false /* packetShouldBeDropped */) +} + +// testFailingRead sends a packet of the given test flow into the stack by +// injecting it into the link endpoint. It then tries to read it from the UDP +// endpoint and expects this to fail. +func testFailingRead(c *testContext, flow testFlow) { + c.t.Helper() + testReadInternal(c, flow, true /* packetShouldBeDropped */) +} + func TestBindEphemeralPort(t *testing.T) { c := newDualTestContext(t, defaultMTU) defer c.cleanup() @@ -768,13 +768,17 @@ func TestReadOnBoundToMulticast(t *testing.T) { c.t.Fatal("SetSockOpt failed:", err) } + // Check that we receive multicast packets but not unicast or broadcast + // ones. testRead(c, flow) + testFailingRead(c, broadcast) + testFailingRead(c, unicastV4) }) } } // TestV4ReadOnBoundToBroadcast checks that an endpoint can bind to a broadcast -// address and receive broadcast data on it. +// address and can receive only broadcast data. func TestV4ReadOnBoundToBroadcast(t *testing.T) { for _, flow := range []testFlow{broadcast, broadcastIn6} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { @@ -789,8 +793,31 @@ func TestV4ReadOnBoundToBroadcast(t *testing.T) { c.t.Fatalf("Bind failed: %s", err) } - // Test acceptance. + // Check that we receive broadcast packets but not unicast ones. testRead(c, flow) + testFailingRead(c, unicastV4) + }) + } +} + +// TestV4ReadBroadcastOnBoundToWildcard checks that an endpoint can bind to ANY +// and receive broadcast and unicast data. +func TestV4ReadBroadcastOnBoundToWildcard(t *testing.T) { + for _, flow := range []testFlow{broadcast, broadcastIn6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + // Bind to wildcard. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatalf("Bind failed: %s (", err) + } + + // Check that we receive both broadcast and unicast packets. + testRead(c, flow) + testRead(c, unicastV4) }) } } @@ -1180,6 +1207,39 @@ func TestTTL(t *testing.T) { } } +func TestSetTTL(t *testing.T) { + for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + for _, wantTTL := range []uint8{1, 2, 50, 64, 128, 254, 255} { + t.Run(fmt.Sprintf("TTL:%d", wantTTL), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + if err := c.ep.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { + c.t.Fatalf("SetSockOpt failed: %v", err) + } + + var p stack.NetworkProtocol + if flow.isV4() { + p = ipv4.NewProtocol() + } else { + p = ipv6.NewProtocol() + } + ep, err := p.NewEndpoint(0, tcpip.AddressWithPrefix{}, nil, nil, nil) + if err != nil { + t.Fatal(err) + } + ep.Close() + + testWrite(c, flow, checker.TTL(wantTTL)) + }) + } + }) + } +} + func TestMulticastInterfaceOption(t *testing.T) { for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { |