diff options
Diffstat (limited to 'pkg/tcpip/transport')
-rw-r--r-- | pkg/tcpip/transport/tcp/BUILD | 1 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/connect.go | 7 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/connect_unsafe.go | 30 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/endpoint.go | 18 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/segment.go | 6 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/segment_heap.go | 16 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/snd.go | 3 |
7 files changed, 58 insertions, 23 deletions
diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 272e8f570..a32f9eacf 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -32,6 +32,7 @@ go_library( srcs = [ "accept.go", "connect.go", + "connect_unsafe.go", "cubic.go", "cubic_state.go", "dispatcher.go", diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index ae4f3f3a9..be86af502 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -624,17 +624,17 @@ func parseSynSegmentOptions(s *segment) header.TCPSynOptions { var optionPool = sync.Pool{ New: func() interface{} { - return make([]byte, maxOptionSize) + return &[maxOptionSize]byte{} }, } func getOptions() []byte { - return optionPool.Get().([]byte) + return (*optionPool.Get().(*[maxOptionSize]byte))[:] } func putOptions(options []byte) { // Reslice to full capacity. - optionPool.Put(options[0:cap(options)]) + optionPool.Put(optionsToArray(options)) } func makeSynOptions(opts header.TCPSynOptions) []byte { @@ -1639,6 +1639,7 @@ func (e *endpoint) doTimeWait() (twReuse func()) { const timeWaitDone = 3 s := sleep.Sleeper{} + defer s.Done() s.AddWaker(&e.newSegmentWaker, newSegment) s.AddWaker(&e.notificationWaker, notification) diff --git a/pkg/tcpip/transport/tcp/connect_unsafe.go b/pkg/tcpip/transport/tcp/connect_unsafe.go new file mode 100644 index 000000000..cfc304616 --- /dev/null +++ b/pkg/tcpip/transport/tcp/connect_unsafe.go @@ -0,0 +1,30 @@ +// 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 tcp + +import ( + "reflect" + "unsafe" +) + +// optionsToArray converts a slice of capacity >-= maxOptionSize to an array. +// +// optionsToArray panics if the capacity of options is smaller than +// maxOptionSize. +func optionsToArray(options []byte) *[maxOptionSize]byte { + // Reslice to full capacity. + options = options[0:maxOptionSize] + return (*[maxOptionSize]byte)(unsafe.Pointer((*reflect.SliceHeader)(unsafe.Pointer(&options)).Data)) +} diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 40cc664c0..5187a5e25 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -825,6 +825,7 @@ func (e *endpoint) Abort() { func (e *endpoint) Close() { e.mu.Lock() closed := e.closed + e.closed = true e.mu.Unlock() if closed { return @@ -833,13 +834,7 @@ func (e *endpoint) Close() { // Issue a shutdown so that the peer knows we won't send any more data // if we're connected, or stop accepting if we're listening. e.Shutdown(tcpip.ShutdownWrite | tcpip.ShutdownRead) - e.closeNoShutdown() -} -// closeNoShutdown closes the endpoint without doing a full shutdown. This is -// used when a connection needs to be aborted with a RST and we want to skip -// a full 4 way TCP shutdown. -func (e *endpoint) closeNoShutdown() { e.mu.Lock() // For listening sockets, we always release ports inline so that they @@ -858,11 +853,8 @@ func (e *endpoint) closeNoShutdown() { e.boundPortFlags = ports.Flags{} } - // Mark endpoint as closed. - e.closed = true // Either perform the local cleanup or kick the worker to make sure it // knows it needs to cleanup. - tcpip.AddDanglingEndpoint(e) switch e.EndpointState() { // Sockets in StateSynRecv state(passive connections) are closed when // the handshake fails or if the listening socket is closed while @@ -876,6 +868,9 @@ func (e *endpoint) closeNoShutdown() { // do nothing. default: e.workerCleanup = true + tcpip.AddDanglingEndpoint(e) + // Worker will remove the dangling endpoint when the endpoint + // goroutine terminates. e.notifyProtocolGoroutine(notifyClose) } @@ -2117,10 +2112,13 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // Close for write. if (e.shutdownFlags & tcpip.ShutdownWrite) != 0 { e.sndBufMu.Lock() - if e.sndClosed { // Already closed. e.sndBufMu.Unlock() + if e.EndpointState() == StateTimeWait { + e.mu.Unlock() + return tcpip.ErrNotConnected + } break } diff --git a/pkg/tcpip/transport/tcp/segment.go b/pkg/tcpip/transport/tcp/segment.go index 1c10da5ca..5d0bc4f72 100644 --- a/pkg/tcpip/transport/tcp/segment.go +++ b/pkg/tcpip/transport/tcp/segment.go @@ -56,9 +56,9 @@ type segment struct { options []byte `state:".([]byte)"` hasNewSACKInfo bool rcvdTime time.Time `state:".(unixTime)"` - // xmitTime is the last transmit time of this segment. A zero value - // indicates that the segment has yet to be transmitted. - xmitTime time.Time `state:".(unixTime)"` + // xmitTime is the last transmit time of this segment. + xmitTime time.Time `state:".(unixTime)"` + xmitCount uint32 } func newSegment(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) *segment { diff --git a/pkg/tcpip/transport/tcp/segment_heap.go b/pkg/tcpip/transport/tcp/segment_heap.go index e28f213ba..8d3ddce4b 100644 --- a/pkg/tcpip/transport/tcp/segment_heap.go +++ b/pkg/tcpip/transport/tcp/segment_heap.go @@ -14,21 +14,25 @@ package tcp +import "container/heap" + type segmentHeap []*segment +var _ heap.Interface = (*segmentHeap)(nil) + // Len returns the length of h. -func (h segmentHeap) Len() int { - return len(h) +func (h *segmentHeap) Len() int { + return len(*h) } // Less determines whether the i-th element of h is less than the j-th element. -func (h segmentHeap) Less(i, j int) bool { - return h[i].sequenceNumber.LessThan(h[j].sequenceNumber) +func (h *segmentHeap) Less(i, j int) bool { + return (*h)[i].sequenceNumber.LessThan((*h)[j].sequenceNumber) } // Swap swaps the i-th and j-th elements of h. -func (h segmentHeap) Swap(i, j int) { - h[i], h[j] = h[j], h[i] +func (h *segmentHeap) Swap(i, j int) { + (*h)[i], (*h)[j] = (*h)[j], (*h)[i] } // Push adds x as the last element of h. diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index b74b61e7d..657c3146e 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -1229,7 +1229,7 @@ func (s *sender) handleRcvdSegment(seg *segment) { // sendSegment sends the specified segment. func (s *sender) sendSegment(seg *segment) *tcpip.Error { - if !seg.xmitTime.IsZero() { + if seg.xmitCount > 0 { s.ep.stack.Stats().TCP.Retransmits.Increment() s.ep.stats.SendErrors.Retransmits.Increment() if s.sndCwnd < s.sndSsthresh { @@ -1237,6 +1237,7 @@ func (s *sender) sendSegment(seg *segment) *tcpip.Error { } } seg.xmitTime = time.Now() + seg.xmitCount++ return s.sendSegmentFromView(seg.data, seg.flags, seg.sequenceNumber) } |