diff options
Diffstat (limited to 'pkg/tcpip/transport/tcp')
-rw-r--r-- | pkg/tcpip/transport/tcp/accept.go | 4 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/connect.go | 89 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/rcv.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/snd.go | 1 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/tcp_test.go | 227 | ||||
-rw-r--r-- | pkg/tcpip/transport/tcp/testing/context/context.go | 3 |
6 files changed, 285 insertions, 41 deletions
diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index f543a6105..74df3edfb 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -298,8 +298,6 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head return nil, err } ep.mu.Lock() - ep.stack.Stats().TCP.CurrentEstablished.Increment() - ep.state = StateEstablished ep.isConnectNotified = true ep.mu.Unlock() @@ -546,6 +544,8 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { n.tsOffset = 0 // Switch state to connected. + // We do not use transitionToStateEstablishedLocked here as there is + // no handshake state available when doing a SYN cookie based accept. n.stack.Stats().TCP.CurrentEstablished.Increment() n.state = StateEstablished n.isConnectNotified = true diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 4206db8b6..3d059c302 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -252,6 +252,11 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { // and the handshake is completed. if s.flagIsSet(header.TCPFlagAck) { h.state = handshakeCompleted + + h.ep.mu.Lock() + h.ep.transitionToStateEstablishedLocked(h) + h.ep.mu.Unlock() + h.ep.sendRaw(buffer.VectorisedView{}, header.TCPFlagAck, h.iss+1, h.ackNum, h.rcvWnd>>h.effectiveRcvWndScale()) return nil } @@ -352,6 +357,10 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { h.ep.updateRecentTimestamp(s.parsedOptions.TSVal, h.ackNum, s.sequenceNumber) } h.state = handshakeCompleted + h.ep.mu.Lock() + h.ep.transitionToStateEstablishedLocked(h) + h.ep.mu.Unlock() + return nil } @@ -880,6 +889,30 @@ func (e *endpoint) completeWorkerLocked() { } } +// transitionToStateEstablisedLocked transitions a given endpoint +// to an established state using the handshake parameters provided. +// It also initializes sender/receiver if required. +func (e *endpoint) transitionToStateEstablishedLocked(h *handshake) { + if e.snd == nil { + // Transfer handshake state to TCP connection. We disable + // receive window scaling if the peer doesn't support it + // (indicated by a negative send window scale). + e.snd = newSender(e, h.iss, h.ackNum-1, h.sndWnd, h.mss, h.sndWndScale) + } + if e.rcv == nil { + rcvBufSize := seqnum.Size(e.receiveBufferSize()) + e.rcvListMu.Lock() + e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale(), rcvBufSize) + // Bootstrap the auto tuning algorithm. Starting at zero will + // result in a really large receive window after the first auto + // tuning adjustment. + e.rcvAutoParams.prevCopied = int(h.rcvWnd) + e.rcvListMu.Unlock() + } + h.ep.stack.Stats().TCP.CurrentEstablished.Increment() + e.state = StateEstablished +} + // transitionToStateCloseLocked ensures that the endpoint is // cleaned up from the transport demuxer, "before" moving to // StateClose. This will ensure that no packet will be @@ -891,6 +924,7 @@ func (e *endpoint) transitionToStateCloseLocked() { } e.cleanupLocked() e.state = StateClose + e.stack.Stats().TCP.EstablishedClosed.Increment() } // tryDeliverSegmentFromClosedEndpoint attempts to deliver the parsed @@ -953,20 +987,6 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { func (e *endpoint) handleSegments() *tcpip.Error { checkRequeue := true for i := 0; i < maxSegmentsPerWake; i++ { - e.mu.RLock() - state := e.state - e.mu.RUnlock() - if state == StateClose { - // When we get into StateClose while processing from the queue, - // return immediately and let the protocolMainloop handle it. - // - // We can reach StateClose only while processing a previous segment - // or a notification from the protocolMainLoop (caller goroutine). - // This means that with this return, the segment dequeue below can - // never occur on a closed endpoint. - return nil - } - s := e.segmentQueue.dequeue() if s == nil { checkRequeue = false @@ -1024,6 +1044,24 @@ func (e *endpoint) handleSegments() *tcpip.Error { s.decRef() continue } + + // Now check if the received segment has caused us to transition + // to a CLOSED state, if yes then terminate processing and do + // not invoke the sender. + e.mu.RLock() + state := e.state + e.mu.RUnlock() + if state == StateClose { + // When we get into StateClose while processing from the queue, + // return immediately and let the protocolMainloop handle it. + // + // We can reach StateClose only while processing a previous segment + // or a notification from the protocolMainLoop (caller goroutine). + // This means that with this return, the segment dequeue below can + // never occur on a closed endpoint. + s.decRef() + return nil + } e.snd.handleRcvdSegment(s) } s.decRef() @@ -1057,6 +1095,7 @@ func (e *endpoint) keepaliveTimerExpired() *tcpip.Error { if e.keepalive.unacked >= e.keepalive.count { e.keepalive.Unlock() + e.stack.Stats().TCP.EstablishedTimedout.Increment() return tcpip.ErrTimeout } @@ -1142,8 +1181,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.lastErrorMu.Unlock() e.mu.Lock() - e.stack.Stats().TCP.EstablishedResets.Increment() - e.stack.Stats().TCP.CurrentEstablished.Decrement() e.state = StateError e.HardError = err @@ -1152,25 +1189,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { return err } - - // Transfer handshake state to TCP connection. We disable - // receive window scaling if the peer doesn't support it - // (indicated by a negative send window scale). - e.snd = newSender(e, h.iss, h.ackNum-1, h.sndWnd, h.mss, h.sndWndScale) - - rcvBufSize := seqnum.Size(e.receiveBufferSize()) - e.rcvListMu.Lock() - e.rcv = newReceiver(e, h.ackNum-1, h.rcvWnd, h.effectiveRcvWndScale(), rcvBufSize) - // boot strap the auto tuning algorithm. Starting at zero will - // result in a large step function on the first proper causing - // the window to just go to a really large value after the first - // RTT itself. - e.rcvAutoParams.prevCopied = initialRcvWnd - e.rcvListMu.Unlock() - e.stack.Stats().TCP.CurrentEstablished.Increment() - e.mu.Lock() - e.state = StateEstablished - e.mu.Unlock() } e.keepalive.timer.init(&e.keepalive.waker) @@ -1371,7 +1389,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // Mark endpoint as closed. e.mu.Lock() if e.state != StateError { - e.stack.Stats().TCP.EstablishedResets.Increment() e.stack.Stats().TCP.CurrentEstablished.Decrement() e.transitionToStateCloseLocked() } diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index 857dc445f..5ee499c36 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -205,7 +205,7 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum // Handle ACK (not FIN-ACK, which we handled above) during one of the // shutdown states. - if s.flagIsSet(header.TCPFlagAck) { + if s.flagIsSet(header.TCPFlagAck) && s.ackNumber == r.ep.snd.sndNxt { r.ep.mu.Lock() switch r.ep.state { case StateFinWait1: diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index d3f7c9125..8332a0179 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -674,7 +674,6 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se default: s.ep.state = StateFinWait1 } - s.ep.stack.Stats().TCP.CurrentEstablished.Decrement() s.ep.mu.Unlock() } else { // We're sending a non-FIN segment. diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 50829ae27..bc5cfcf0e 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -75,6 +75,20 @@ func TestGiveUpConnect(t *testing.T) { if err := ep.GetSockOpt(tcpip.ErrorOption{}); err != tcpip.ErrAborted { t.Fatalf("got ep.GetSockOpt(tcpip.ErrorOption{}) = %v, want = %v", err, tcpip.ErrAborted) } + + // Call Connect again to retreive the handshake failure status + // and stats updates. + if err := ep.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrAborted { + t.Fatalf("got ep.Connect(...) = %v, want = %v", err, tcpip.ErrAborted) + } + + if got := c.Stack().Stats().TCP.FailedConnectionAttempts.Value(); got != 1 { + t.Errorf("got stats.TCP.FailedConnectionAttempts.Value() = %v, want = 1", got) + } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestConnectIncrementActiveConnection(t *testing.T) { @@ -541,13 +555,21 @@ func TestClosingWithEnqueuedSegments(t *testing.T) { ep.(interface{ ResumeWork() }).ResumeWork() // Wait for the protocolMainLoop to resume and update state. - time.Sleep(1 * time.Millisecond) + time.Sleep(10 * time.Millisecond) // Expect the endpoint to be closed. if got, want := tcp.EndpointState(ep.State()), tcp.StateClose; got != want { t.Errorf("Unexpected endpoint state: want %v, got %v", want, got) } + if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != 1 { + t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = 1", got) + } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } + // Check if the endpoint was moved to CLOSED and netstack a reset in // response to the ACK packet that we sent after last-ACK. checker.IPv4(t, c.GetPacket(), @@ -2694,6 +2716,13 @@ loop: if tcp.EndpointState(c.EP.State()) != tcp.StateError { t.Fatalf("got EP state is not StateError") } + + if got := c.Stack().Stats().TCP.EstablishedResets.Value(); got != 1 { + t.Errorf("got stats.TCP.EstablishedResets.Value() = %v, want = 1", got) + } + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func TestSendOnResetConnection(t *testing.T) { @@ -4363,9 +4392,17 @@ func TestKeepalive(t *testing.T) { ), ) + if got := c.Stack().Stats().TCP.EstablishedTimedout.Value(); got != 1 { + t.Errorf("got c.Stack().Stats().TCP.EstablishedTimedout.Value() = %v, want = 1", got) + } + if _, _, err := c.EP.Read(nil); err != tcpip.ErrTimeout { t.Fatalf("got c.EP.Read(nil) = %v, want = %v", err, tcpip.ErrTimeout) } + + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } } func executeHandshake(t *testing.T, c *context.Context, srcPort uint16, synCookieInUse bool) (irs, iss seqnum.Value) { @@ -5632,6 +5669,7 @@ func TestTCPTimeWaitRSTIgnored(t *testing.T) { DstPort: context.StackPort, Flags: header.TCPFlagSyn, SeqNum: iss, + RcvWnd: 30000, }) // Receive the SYN-ACK reply. @@ -5750,6 +5788,7 @@ func TestTCPTimeWaitOutOfOrder(t *testing.T) { DstPort: context.StackPort, Flags: header.TCPFlagSyn, SeqNum: iss, + RcvWnd: 30000, }) // Receive the SYN-ACK reply. @@ -5856,6 +5895,7 @@ func TestTCPTimeWaitNewSyn(t *testing.T) { DstPort: context.StackPort, Flags: header.TCPFlagSyn, SeqNum: iss, + RcvWnd: 30000, }) // Receive the SYN-ACK reply. @@ -5929,6 +5969,7 @@ func TestTCPTimeWaitNewSyn(t *testing.T) { DstPort: context.StackPort, Flags: header.TCPFlagSyn, SeqNum: iss, + RcvWnd: 30000, }) c.CheckNoPacketTimeout("unexpected packet received in response to SYN", 1*time.Second) @@ -5941,6 +5982,7 @@ func TestTCPTimeWaitNewSyn(t *testing.T) { DstPort: context.StackPort, Flags: header.TCPFlagSyn, SeqNum: iss, + RcvWnd: 30000, }) // Receive the SYN-ACK reply. @@ -5987,6 +6029,8 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPLingerTimeoutOption(%d) failed: %s", tcpTimeWaitTimeout, err) } + want := c.Stack().Stats().TCP.EstablishedClosed.Value() + 1 + wq := &waiter.Queue{} ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) if err != nil { @@ -6007,6 +6051,7 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { DstPort: context.StackPort, Flags: header.TCPFlagSyn, SeqNum: iss, + RcvWnd: 30000, }) // Receive the SYN-ACK reply. @@ -6114,4 +6159,184 @@ func TestTCPTimeWaitDuplicateFINExtendsTimeWait(t *testing.T) { checker.SeqNum(uint32(ackHeaders.AckNum)), checker.AckNum(uint32(ackHeaders.SeqNum)), checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + + if got := c.Stack().Stats().TCP.EstablishedClosed.Value(); got != want { + t.Errorf("got c.Stack().Stats().TCP.EstablishedClosed = %v, want = %v", got, want) + } + if got := c.Stack().Stats().TCP.CurrentEstablished.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentEstablished.Value() = %v, want = 0", got) + } +} + +func TestTCPCloseWithData(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Set TCPTimeWaitTimeout to 5 seconds so that sockets are marked closed + // after 5 seconds in TIME_WAIT state. + tcpTimeWaitTimeout := 5 * time.Second + if err := c.Stack().SetTransportProtocolOption(tcp.ProtocolNumber, tcpip.TCPTimeWaitTimeoutOption(tcpTimeWaitTimeout)); err != nil { + t.Fatalf("c.stack.SetTransportProtocolOption(tcp, tcpip.TCPLingerTimeoutOption(%d) failed: %s", tcpTimeWaitTimeout, err) + } + + wq := &waiter.Queue{} + ep, err := c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, wq) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + if err := ep.Listen(10); err != nil { + t.Fatalf("Listen failed: %s", err) + } + + // Send a SYN request. + iss := seqnum.Value(789) + c.SendPacket(nil, &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagSyn, + SeqNum: iss, + RcvWnd: 30000, + }) + + // Receive the SYN-ACK reply. + b := c.GetPacket() + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + + ackHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 1, + AckNum: c.IRS + 1, + RcvWnd: 30000, + } + + // Send ACK. + c.SendPacket(nil, ackHeaders) + + // Try to accept the connection. + we, ch := waiter.NewChannelEntry(nil) + wq.EventRegister(&we, waiter.EventIn) + defer wq.EventUnregister(&we) + + c.EP, _, err = ep.Accept() + if err == tcpip.ErrWouldBlock { + // Wait for connection to be established. + select { + case <-ch: + c.EP, _, err = ep.Accept() + if err != nil { + t.Fatalf("Accept failed: %s", err) + } + + case <-time.After(1 * time.Second): + t.Fatalf("Timed out waiting for accept") + } + } + + // Now trigger a passive close by sending a FIN. + finHeaders := &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck | header.TCPFlagFin, + SeqNum: iss + 1, + AckNum: c.IRS + 2, + RcvWnd: 30000, + } + + c.SendPacket(nil, finHeaders) + + // Get the ACK to the FIN we just sent. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)), + checker.AckNum(uint32(iss)+2), + checker.TCPFlags(header.TCPFlagAck))) + + // Now write a few bytes and then close the endpoint. + data := []byte{1, 2, 3} + view := buffer.NewView(len(data)) + copy(view, data) + + if _, _, err := c.EP.Write(tcpip.SlicePayload(view), tcpip.WriteOptions{}); err != nil { + t.Fatalf("Write failed: %s", err) + } + + // Check that data is received. + b = c.GetPacket() + checker.IPv4(t, b, + checker.PayloadLen(len(data)+header.TCPMinimumSize), + checker.TCP( + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS)+1), + checker.AckNum(uint32(iss)+2), // Acknum is initial sequence number + 1 + checker.TCPFlagsMatch(header.TCPFlagAck, ^uint8(header.TCPFlagPsh)), + ), + ) + + if p := b[header.IPv4MinimumSize+header.TCPMinimumSize:]; !bytes.Equal(data, p) { + t.Errorf("got data = %x, want = %x", p, data) + } + + c.EP.Close() + // Check the FIN. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(c.IRS+1)+uint32(len(data))), + checker.AckNum(uint32(iss+2)), + checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) + + // First send a partial ACK. + ackHeaders = &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 2, + AckNum: c.IRS + 1 + seqnum.Value(len(data)-1), + RcvWnd: 30000, + } + c.SendPacket(nil, ackHeaders) + + // Now send a full ACK. + ackHeaders = &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 2, + AckNum: c.IRS + 1 + seqnum.Value(len(data)), + RcvWnd: 30000, + } + c.SendPacket(nil, ackHeaders) + + // Now ACK the FIN. + ackHeaders.AckNum++ + c.SendPacket(nil, ackHeaders) + + // Now send an ACK and we should get a RST back as the endpoint should + // be in CLOSED state. + ackHeaders = &context.Headers{ + SrcPort: context.TestPort, + DstPort: context.StackPort, + Flags: header.TCPFlagAck, + SeqNum: iss + 2, + AckNum: c.IRS + 1 + seqnum.Value(len(data)), + RcvWnd: 30000, + } + c.SendPacket(nil, ackHeaders) + + // Check the RST. + checker.IPv4(t, c.GetPacket(), checker.TCP( + checker.SrcPort(context.StackPort), + checker.DstPort(context.TestPort), + checker.SeqNum(uint32(ackHeaders.AckNum)), + checker.AckNum(uint32(ackHeaders.SeqNum)), + checker.TCPFlags(header.TCPFlagRst|header.TCPFlagAck))) + } diff --git a/pkg/tcpip/transport/tcp/testing/context/context.go b/pkg/tcpip/transport/tcp/testing/context/context.go index 6cb66c1af..b0a376eba 100644 --- a/pkg/tcpip/transport/tcp/testing/context/context.go +++ b/pkg/tcpip/transport/tcp/testing/context/context.go @@ -231,6 +231,7 @@ func (c *Context) CheckNoPacket(errMsg string) { // addresses. It will fail with an error if no packet is received for // 2 seconds. func (c *Context) GetPacket() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv4.ProtocolNumber { @@ -259,6 +260,7 @@ func (c *Context) GetPacket() []byte { // and destination address. If no packet is available it will return // nil immediately. func (c *Context) GetPacketNonBlocking() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv4.ProtocolNumber { @@ -483,6 +485,7 @@ func (c *Context) CreateV6Endpoint(v6only bool) { // GetV6Packet reads a single packet from the link layer endpoint of the context // and asserts that it is an IPv6 Packet with the expected src/dest addresses. func (c *Context) GetV6Packet() []byte { + c.t.Helper() select { case p := <-c.linkEP.C: if p.Proto != ipv6.ProtocolNumber { |