summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/transport
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/transport')
-rw-r--r--pkg/tcpip/transport/icmp/endpoint.go23
-rw-r--r--pkg/tcpip/transport/tcp/connect.go11
-rw-r--r--pkg/tcpip/transport/tcp/endpoint.go44
-rw-r--r--pkg/tcpip/transport/tcp/rcv.go4
-rw-r--r--pkg/tcpip/transport/tcp/segment_heap.go1
-rw-r--r--pkg/tcpip/transport/udp/endpoint.go30
-rw-r--r--pkg/tcpip/transport/udp/endpoint_state.go3
7 files changed, 70 insertions, 46 deletions
diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go
index 426da1ee6..2a396e9bc 100644
--- a/pkg/tcpip/transport/icmp/endpoint.go
+++ b/pkg/tcpip/transport/icmp/endpoint.go
@@ -291,15 +291,13 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c
nicID = e.BindNICID
}
- toCopy := *to
- to = &toCopy
- netProto, err := e.checkV4Mapped(to)
+ dst, netProto, err := e.checkV4MappedLocked(*to)
if err != nil {
return 0, nil, err
}
- // Find the enpoint.
- r, err := e.stack.FindRoute(nicID, e.BindAddr, to.Addr, netProto, false /* multicastLoop */)
+ // Find the endpoint.
+ r, err := e.stack.FindRoute(nicID, e.BindAddr, dst.Addr, netProto, false /* multicastLoop */)
if err != nil {
return 0, nil, err
}
@@ -480,13 +478,14 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err
})
}
-func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) {
- unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, false /* v6only */)
+// checkV4MappedLocked determines the effective network protocol and converts
+// addr to its canonical form.
+func (e *endpoint) checkV4MappedLocked(addr tcpip.FullAddress) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) {
+ unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProtoLocked(addr, false /* v6only */)
if err != nil {
- return 0, err
+ return tcpip.FullAddress{}, 0, err
}
- *addr = unwrapped
- return netProto, nil
+ return unwrapped, netProto, nil
}
// Disconnect implements tcpip.Endpoint.Disconnect.
@@ -517,7 +516,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
- netProto, err := e.checkV4Mapped(&addr)
+ addr, netProto, err := e.checkV4MappedLocked(addr)
if err != nil {
return err
}
@@ -630,7 +629,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
- netProto, err := e.checkV4Mapped(&addr)
+ addr, netProto, err := e.checkV4MappedLocked(addr)
if err != nil {
return err
}
diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go
index cd247f3e1..ae4f3f3a9 100644
--- a/pkg/tcpip/transport/tcp/connect.go
+++ b/pkg/tcpip/transport/tcp/connect.go
@@ -295,6 +295,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error {
h.state = handshakeSynRcvd
h.ep.mu.Lock()
ttl := h.ep.ttl
+ amss := h.ep.amss
h.ep.setEndpointState(StateSynRecv)
h.ep.mu.Unlock()
synOpts := header.TCPSynOptions{
@@ -307,7 +308,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error {
// permits SACK. This is not explicitly defined in the RFC but
// this is the behaviour implemented by Linux.
SACKPermitted: rcvSynOpts.SACKPermitted,
- MSS: h.ep.amss,
+ MSS: amss,
}
if ttl == 0 {
ttl = s.route.DefaultTTL()
@@ -356,6 +357,10 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
+ h.ep.mu.RLock()
+ amss := h.ep.amss
+ h.ep.mu.RUnlock()
+
h.resetState()
synOpts := header.TCPSynOptions{
WS: h.rcvWndScale,
@@ -363,7 +368,7 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error {
TSVal: h.ep.timestamp(),
TSEcr: h.ep.recentTimestamp(),
SACKPermitted: h.ep.sackPermitted,
- MSS: h.ep.amss,
+ MSS: amss,
}
h.ep.sendSynTCP(&s.route, h.ep.ID, h.ep.ttl, h.ep.sendTOS, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
return nil
@@ -530,6 +535,7 @@ func (h *handshake) execute() *tcpip.Error {
// Send the initial SYN segment and loop until the handshake is
// completed.
+ h.ep.mu.Lock()
h.ep.amss = calculateAdvertisedMSS(h.ep.userMSS, h.ep.route)
synOpts := header.TCPSynOptions{
@@ -540,6 +546,7 @@ func (h *handshake) execute() *tcpip.Error {
SACKPermitted: bool(sackEnabled),
MSS: h.ep.amss,
}
+ h.ep.mu.Unlock()
// Execute is also called in a listen context so we want to make sure we
// only send the TS/SACK option when we received the TS/SACK in the
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index 9e72730bd..40cc664c0 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -959,15 +959,18 @@ func (e *endpoint) initialReceiveWindow() int {
// ModerateRecvBuf adjusts the receive buffer and the advertised window
// based on the number of bytes copied to user space.
func (e *endpoint) ModerateRecvBuf(copied int) {
+ e.mu.RLock()
e.rcvListMu.Lock()
if e.rcvAutoParams.disabled {
e.rcvListMu.Unlock()
+ e.mu.RUnlock()
return
}
now := time.Now()
if rtt := e.rcvAutoParams.rtt; rtt == 0 || now.Sub(e.rcvAutoParams.measureTime) < rtt {
e.rcvAutoParams.copied += copied
e.rcvListMu.Unlock()
+ e.mu.RUnlock()
return
}
prevRTTCopied := e.rcvAutoParams.copied + copied
@@ -1008,7 +1011,7 @@ func (e *endpoint) ModerateRecvBuf(copied int) {
e.rcvBufSize = rcvWnd
availAfter := e.receiveBufferAvailableLocked()
mask := uint32(notifyReceiveWindowChanged)
- if crossed, above := e.windowCrossedACKThreshold(availAfter - availBefore); crossed && above {
+ if crossed, above := e.windowCrossedACKThresholdLocked(availAfter - availBefore); crossed && above {
mask |= notifyNonZeroReceiveWindow
}
e.notifyProtocolGoroutine(mask)
@@ -1023,6 +1026,7 @@ func (e *endpoint) ModerateRecvBuf(copied int) {
e.rcvAutoParams.measureTime = now
e.rcvAutoParams.copied = 0
e.rcvListMu.Unlock()
+ e.mu.RUnlock()
}
// IPTables implements tcpip.Endpoint.IPTables.
@@ -1052,7 +1056,6 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages,
v, err := e.readLocked()
e.rcvListMu.Unlock()
-
e.mu.RUnlock()
if err == tcpip.ErrClosedForReceive {
@@ -1085,7 +1088,7 @@ func (e *endpoint) readLocked() (buffer.View, *tcpip.Error) {
// enough buffer space, to either fit an aMSS or half a receive buffer
// (whichever smaller), then notify the protocol goroutine to send a
// window update.
- if crossed, above := e.windowCrossedACKThreshold(len(v)); crossed && above {
+ if crossed, above := e.windowCrossedACKThresholdLocked(len(v)); crossed && above {
e.notifyProtocolGoroutine(notifyNonZeroReceiveWindow)
}
@@ -1303,9 +1306,9 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro
return num, tcpip.ControlMessages{}, nil
}
-// windowCrossedACKThreshold checks if the receive window to be announced now
-// would be under aMSS or under half receive buffer, whichever smaller. This is
-// useful as a receive side silly window syndrome prevention mechanism. If
+// windowCrossedACKThresholdLocked checks if the receive window to be announced
+// now would be under aMSS or under half receive buffer, whichever smaller. This
+// is useful as a receive side silly window syndrome prevention mechanism. If
// window grows to reasonable value, we should send ACK to the sender to inform
// the rx space is now large. We also want ensure a series of small read()'s
// won't trigger a flood of spurious tiny ACK's.
@@ -1316,7 +1319,9 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro
// crossed will be true if the window size crossed the ACK threshold.
// above will be true if the new window is >= ACK threshold and false
// otherwise.
-func (e *endpoint) windowCrossedACKThreshold(deltaBefore int) (crossed bool, above bool) {
+//
+// Precondition: e.mu and e.rcvListMu must be held.
+func (e *endpoint) windowCrossedACKThresholdLocked(deltaBefore int) (crossed bool, above bool) {
newAvail := e.receiveBufferAvailableLocked()
oldAvail := newAvail - deltaBefore
if oldAvail < 0 {
@@ -1379,6 +1384,7 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error {
mask := uint32(notifyReceiveWindowChanged)
+ e.mu.RLock()
e.rcvListMu.Lock()
// Make sure the receive buffer size allows us to send a
@@ -1405,11 +1411,11 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error {
// Immediately send an ACK to uncork the sender silly window
// syndrome prevetion, when our available space grows above aMSS
// or half receive buffer, whichever smaller.
- if crossed, above := e.windowCrossedACKThreshold(availAfter - availBefore); crossed && above {
+ if crossed, above := e.windowCrossedACKThresholdLocked(availAfter - availBefore); crossed && above {
mask |= notifyNonZeroReceiveWindow
}
e.rcvListMu.Unlock()
-
+ e.mu.RUnlock()
e.notifyProtocolGoroutine(mask)
return nil
@@ -1868,13 +1874,14 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
}
}
-func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) {
- unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, e.v6only)
+// checkV4MappedLocked determines the effective network protocol and converts
+// addr to its canonical form.
+func (e *endpoint) checkV4MappedLocked(addr tcpip.FullAddress) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) {
+ unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProtoLocked(addr, e.v6only)
if err != nil {
- return 0, err
+ return tcpip.FullAddress{}, 0, err
}
- *addr = unwrapped
- return netProto, nil
+ return unwrapped, netProto, nil
}
// Disconnect implements tcpip.Endpoint.Disconnect.
@@ -1904,7 +1911,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc
connectingAddr := addr.Addr
- netProto, err := e.checkV4Mapped(&addr)
+ addr, netProto, err := e.checkV4MappedLocked(addr)
if err != nil {
return err
}
@@ -2270,7 +2277,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) {
}
e.BindAddr = addr.Addr
- netProto, err := e.checkV4Mapped(&addr)
+ addr, netProto, err := e.checkV4MappedLocked(addr)
if err != nil {
return err
}
@@ -2414,13 +2421,14 @@ func (e *endpoint) updateSndBufferUsage(v int) {
// to be read, or when the connection is closed for receiving (in which case
// s will be nil).
func (e *endpoint) readyToRead(s *segment) {
+ e.mu.RLock()
e.rcvListMu.Lock()
if s != nil {
s.incRef()
e.rcvBufUsed += s.data.Size()
// Increase counter if the receive window falls down below MSS
// or half receive buffer size, whichever smaller.
- if crossed, above := e.windowCrossedACKThreshold(-s.data.Size()); crossed && !above {
+ if crossed, above := e.windowCrossedACKThresholdLocked(-s.data.Size()); crossed && !above {
e.stats.ReceiveErrors.ZeroRcvWindowState.Increment()
}
e.rcvList.PushBack(s)
@@ -2428,7 +2436,7 @@ func (e *endpoint) readyToRead(s *segment) {
e.rcvClosed = true
}
e.rcvListMu.Unlock()
-
+ e.mu.RUnlock()
e.waiterQueue.Notify(waiter.EventIn)
}
diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go
index 958f03ac1..d80aff1b6 100644
--- a/pkg/tcpip/transport/tcp/rcv.go
+++ b/pkg/tcpip/transport/tcp/rcv.go
@@ -195,6 +195,10 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum
for i := first; i < len(r.pendingRcvdSegments); i++ {
r.pendingRcvdSegments[i].decRef()
+ // Note that slice truncation does not allow garbage collection of
+ // truncated items, thus truncated items must be set to nil to avoid
+ // memory leaks.
+ r.pendingRcvdSegments[i] = nil
}
r.pendingRcvdSegments = r.pendingRcvdSegments[:first]
diff --git a/pkg/tcpip/transport/tcp/segment_heap.go b/pkg/tcpip/transport/tcp/segment_heap.go
index 9fd061d7d..e28f213ba 100644
--- a/pkg/tcpip/transport/tcp/segment_heap.go
+++ b/pkg/tcpip/transport/tcp/segment_heap.go
@@ -41,6 +41,7 @@ func (h *segmentHeap) Pop() interface{} {
old := *h
n := len(old)
x := old[n-1]
+ old[n-1] = nil
*h = old[:n-1]
return x
}
diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go
index 1c6a600b8..0af4514e1 100644
--- a/pkg/tcpip/transport/udp/endpoint.go
+++ b/pkg/tcpip/transport/udp/endpoint.go
@@ -443,19 +443,19 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c
return 0, nil, tcpip.ErrBroadcastDisabled
}
- netProto, err := e.checkV4Mapped(to)
+ dst, netProto, err := e.checkV4MappedLocked(*to)
if err != nil {
return 0, nil, err
}
- r, _, err := e.connectRoute(nicID, *to, netProto)
+ r, _, err := e.connectRoute(nicID, dst, netProto)
if err != nil {
return 0, nil, err
}
defer r.Release()
route = &r
- dstPort = to.Port
+ dstPort = dst.Port
}
if route.IsResolutionRequired() {
@@ -566,7 +566,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
defer e.mu.Unlock()
fa := tcpip.FullAddress{Addr: v.InterfaceAddr}
- netProto, err := e.checkV4Mapped(&fa)
+ fa, netProto, err := e.checkV4MappedLocked(fa)
if err != nil {
return err
}
@@ -927,13 +927,14 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u
return nil
}
-func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) {
- unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, e.v6only)
+// checkV4MappedLocked determines the effective network protocol and converts
+// addr to its canonical form.
+func (e *endpoint) checkV4MappedLocked(addr tcpip.FullAddress) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) {
+ unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProtoLocked(addr, e.v6only)
if err != nil {
- return 0, err
+ return tcpip.FullAddress{}, 0, err
}
- *addr = unwrapped
- return netProto, nil
+ return unwrapped, netProto, nil
}
// Disconnect implements tcpip.Endpoint.Disconnect.
@@ -981,10 +982,6 @@ func (e *endpoint) Disconnect() *tcpip.Error {
// Connect connects the endpoint to its peer. Specifying a NIC is optional.
func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
- netProto, err := e.checkV4Mapped(&addr)
- if err != nil {
- return err
- }
if addr.Port == 0 {
// We don't support connecting to port zero.
return tcpip.ErrInvalidEndpointState
@@ -1012,6 +1009,11 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
+ addr, netProto, err := e.checkV4MappedLocked(addr)
+ if err != nil {
+ return err
+ }
+
r, nicID, err := e.connectRoute(nicID, addr, netProto)
if err != nil {
return err
@@ -1139,7 +1141,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error {
return tcpip.ErrInvalidEndpointState
}
- netProto, err := e.checkV4Mapped(&addr)
+ addr, netProto, err := e.checkV4MappedLocked(addr)
if err != nil {
return err
}
diff --git a/pkg/tcpip/transport/udp/endpoint_state.go b/pkg/tcpip/transport/udp/endpoint_state.go
index 43fb047ed..466bd9381 100644
--- a/pkg/tcpip/transport/udp/endpoint_state.go
+++ b/pkg/tcpip/transport/udp/endpoint_state.go
@@ -69,6 +69,9 @@ func (e *endpoint) afterLoad() {
// Resume implements tcpip.ResumableEndpoint.Resume.
func (e *endpoint) Resume(s *stack.Stack) {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+
e.stack = s
for _, m := range e.multicastMemberships {
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-28 16:49:56 +0000