diff options
| -rw-r--r-- | pkg/abi/linux/linux_abi_autogen_unsafe.go | 12 | ||||
| -rw-r--r-- | pkg/tcpip/transport/tcp/snd.go | 31 |
2 files changed, 34 insertions, 9 deletions
diff --git a/pkg/abi/linux/linux_abi_autogen_unsafe.go b/pkg/abi/linux/linux_abi_autogen_unsafe.go index 1d601c5da..67513ba7f 100644 --- a/pkg/abi/linux/linux_abi_autogen_unsafe.go +++ b/pkg/abi/linux/linux_abi_autogen_unsafe.go @@ -124,12 +124,12 @@ func (s *Statx) UnmarshalBytes(src []byte) { // Packed implements marshal.Marshallable.Packed. //go:nosplit func (s *Statx) Packed() bool { - return s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() + return s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() } // MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. func (s *Statx) MarshalUnsafe(dst []byte) { - if s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() { + if s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { safecopy.CopyIn(dst, unsafe.Pointer(s)) } else { s.MarshalBytes(dst) @@ -138,7 +138,7 @@ func (s *Statx) MarshalUnsafe(dst []byte) { // UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. func (s *Statx) UnmarshalUnsafe(src []byte) { - if s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() { + if s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() { safecopy.CopyOut(unsafe.Pointer(s), src) } else { s.UnmarshalBytes(src) @@ -148,7 +148,7 @@ func (s *Statx) UnmarshalUnsafe(src []byte) { // CopyOutN implements marshal.Marshallable.CopyOutN. //go:nosplit func (s *Statx) CopyOutN(task marshal.Task, addr usermem.Addr, limit int) (int, error) { - if !s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { + if !s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() { // Type Statx doesn't have a packed layout in memory, fall back to MarshalBytes. buf := task.CopyScratchBuffer(s.SizeBytes()) // escapes: okay. s.MarshalBytes(buf) // escapes: fallback. @@ -178,7 +178,7 @@ func (s *Statx) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) { // CopyIn implements marshal.Marshallable.CopyIn. //go:nosplit func (s *Statx) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { - if !s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() { + if !s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { // Type Statx doesn't have a packed layout in memory, fall back to UnmarshalBytes. buf := task.CopyScratchBuffer(s.SizeBytes()) // escapes: okay. length, err := task.CopyInBytes(addr, buf) // escapes: okay. @@ -204,7 +204,7 @@ func (s *Statx) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) { // WriteTo implements io.WriterTo.WriteTo. func (s *Statx) WriteTo(w io.Writer) (int64, error) { - if !s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() { + if !s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() { // Type Statx doesn't have a packed layout in memory, fall back to MarshalBytes. buf := make([]byte, s.SizeBytes()) s.MarshalBytes(buf) diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index 3a19c4468..acacb42e4 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -618,6 +618,20 @@ func (s *sender) splitSeg(seg *segment, size int) { nSeg.data.TrimFront(size) nSeg.sequenceNumber.UpdateForward(seqnum.Size(size)) s.writeList.InsertAfter(seg, nSeg) + + // The segment being split does not carry PUSH flag because it is + // followed by the newly split segment. + // RFC1122 section 4.2.2.2: MUST set the PSH bit in the last buffered + // segment (i.e., when there is no more queued data to be sent). + // Linux removes PSH flag only when the segment is being split over MSS + // and retains it when we are splitting the segment over lack of sender + // window space. + // ref: net/ipv4/tcp_output.c::tcp_write_xmit(), tcp_mss_split_point() + // ref: net/ipv4/tcp_output.c::tcp_write_wakeup(), tcp_snd_wnd_test() + if seg.data.Size() > s.maxPayloadSize { + seg.flags ^= header.TCPFlagPsh + } + seg.data.CapLength(size) } @@ -739,7 +753,7 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se if !s.isAssignedSequenceNumber(seg) { // Merge segments if allowed. if seg.data.Size() != 0 { - available := int(seg.sequenceNumber.Size(end)) + available := int(s.sndNxt.Size(end)) if available > limit { available = limit } @@ -782,8 +796,11 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se // sent all at once. return false } - if atomic.LoadUint32(&s.ep.cork) != 0 { - // Hold back the segment until full. + // With TCP_CORK, hold back until minimum of the available + // send space and MSS. + // TODO(gvisor.dev/issue/2833): Drain the held segments after a + // timeout. + if seg.data.Size() < s.maxPayloadSize && atomic.LoadUint32(&s.ep.cork) != 0 { return false } } @@ -843,9 +860,17 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se if available == 0 { return false } + + // The segment size limit is computed as a function of sender congestion + // window and MSS. When sender congestion window is > 1, this limit can + // be larger than MSS. Ensure that the currently available send space + // is not greater than minimum of this limit and MSS. if available > limit { available = limit } + if available > s.maxPayloadSize { + available = s.maxPayloadSize + } if seg.data.Size() > available { s.splitSeg(seg, available) |