summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/transport/tcp/endpoint.go
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/transport/tcp/endpoint.go')
-rw-r--r--pkg/tcpip/transport/tcp/endpoint.go586
1 files changed, 411 insertions, 175 deletions
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index ac927569a..a1b784b49 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"
@@ -170,6 +172,101 @@ type rcvBufAutoTuneParams struct {
disabled bool
}
+// ReceiveErrors collect segment receive errors within transport layer.
+type ReceiveErrors struct {
+ tcpip.ReceiveErrors
+
+ // SegmentQueueDropped is the number of segments dropped due to
+ // a full segment queue.
+ SegmentQueueDropped tcpip.StatCounter
+
+ // ChecksumErrors is the number of segments dropped due to bad checksums.
+ ChecksumErrors tcpip.StatCounter
+
+ // ListenOverflowSynDrop is the number of times the listen queue overflowed
+ // and a SYN was dropped.
+ ListenOverflowSynDrop tcpip.StatCounter
+
+ // ListenOverflowAckDrop is the number of times the final ACK
+ // in the handshake was dropped due to overflow.
+ ListenOverflowAckDrop tcpip.StatCounter
+
+ // ZeroRcvWindowState is the number of times we advertised
+ // a zero receive window when rcvList is full.
+ ZeroRcvWindowState tcpip.StatCounter
+}
+
+// SendErrors collect segment send errors within the transport layer.
+type SendErrors struct {
+ tcpip.SendErrors
+
+ // SegmentSendToNetworkFailed is the number of TCP segments failed to be sent
+ // to the network endpoint.
+ SegmentSendToNetworkFailed tcpip.StatCounter
+
+ // SynSendToNetworkFailed is the number of TCP SYNs failed to be sent
+ // to the network endpoint.
+ SynSendToNetworkFailed tcpip.StatCounter
+
+ // Retransmits is the number of TCP segments retransmitted.
+ Retransmits tcpip.StatCounter
+
+ // FastRetransmit is the number of segments retransmitted in fast
+ // recovery.
+ FastRetransmit tcpip.StatCounter
+
+ // Timeouts is the number of times the RTO expired.
+ Timeouts tcpip.StatCounter
+}
+
+// Stats holds statistics about the endpoint.
+type Stats struct {
+ // SegmentsReceived is the number of TCP segments received that
+ // the transport layer successfully parsed.
+ SegmentsReceived tcpip.StatCounter
+
+ // SegmentsSent is the number of TCP segments sent.
+ SegmentsSent tcpip.StatCounter
+
+ // FailedConnectionAttempts is the number of times we saw Connect and
+ // Accept errors.
+ FailedConnectionAttempts tcpip.StatCounter
+
+ // ReceiveErrors collects segment receive errors within the
+ // transport layer.
+ ReceiveErrors ReceiveErrors
+
+ // ReadErrors collects segment read errors from an endpoint read call.
+ ReadErrors tcpip.ReadErrors
+
+ // SendErrors collects segment send errors within the transport layer.
+ SendErrors SendErrors
+
+ // WriteErrors collects segment write errors from an endpoint write call.
+ WriteErrors tcpip.WriteErrors
+}
+
+// IsEndpointStats is an empty method to implement the tcpip.EndpointStats
+// marker interface.
+func (*Stats) IsEndpointStats() {}
+
+// EndpointInfo holds useful information about a transport endpoint which
+// can be queried by monitoring tools.
+//
+// +stateify savable
+type EndpointInfo struct {
+ stack.TransportEndpointInfo
+
+ // HardError is meaningful only when state is stateError. It stores the
+ // error to be returned when read/write syscalls are called and the
+ // endpoint is in this state. HardError is protected by endpoint mu.
+ HardError *tcpip.Error `state:".(string)"`
+}
+
+// IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo
+// marker interface.
+func (*EndpointInfo) IsEndpointInfo() {}
+
// endpoint represents a TCP endpoint. This struct serves as the interface
// between users of the endpoint and the protocol implementation; it is legal to
// have concurrent goroutines make calls into the endpoint, they are properly
@@ -178,6 +275,8 @@ type rcvBufAutoTuneParams struct {
//
// +stateify savable
type endpoint struct {
+ EndpointInfo
+
// workMu is used to arbitrate which goroutine may perform protocol
// work. Only the main protocol goroutine is expected to call Lock() on
// it, but other goroutines (e.g., send) may call TryLock() to eagerly
@@ -186,8 +285,7 @@ type endpoint struct {
// The following fields are initialized at creation time and do not
// change throughout the lifetime of the endpoint.
- stack *stack.Stack `state:"manual"`
- netProto tcpip.NetworkProtocolNumber
+ stack *stack.Stack `state:"manual"`
waiterQueue *waiter.Queue `state:"wait"`
// lastError represents the last error that the endpoint reported;
@@ -218,7 +316,6 @@ type endpoint struct {
// The following fields are protected by the mutex.
mu sync.RWMutex `state:"nosave"`
- id stack.TransportEndpointID
state EndpointState `state:".(EndpointState)"`
@@ -226,6 +323,7 @@ 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/
@@ -240,11 +338,6 @@ type endpoint struct {
// address).
effectiveNetProtos []tcpip.NetworkProtocolNumber `state:"manual"`
- // hardError is meaningful only when state is stateError, it stores the
- // error to be returned when read/write syscalls are called and the
- // endpoint is in this state. hardError is protected by mu.
- hardError *tcpip.Error `state:".(string)"`
-
// workerRunning specifies if a worker goroutine is running.
workerRunning bool
@@ -280,6 +373,9 @@ 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.
@@ -393,13 +489,19 @@ type endpoint struct {
probe stack.TCPProbeFunc `state:"nosave"`
// The following are only used to assist the restore run to re-connect.
- bindAddress tcpip.Address
connectingAddress tcpip.Address
// amss is the advertised MSS to the peer by this endpoint.
amss uint16
+ // sendTOS represents IPv4 TOS or IPv6 TrafficClass,
+ // applied while sending packets. Defaults to 0 as on Linux.
+ sendTOS uint8
+
gso *stack.GSO
+
+ // TODO(b/142022063): Add ability to save and restore per endpoint stats.
+ stats Stats `state:"nosave"`
}
// StopWork halts packet processing. Only to be used in tests.
@@ -427,10 +529,15 @@ type keepalive struct {
waker sleep.Waker `state:"nosave"`
}
-func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint {
+func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQueue *waiter.Queue) *endpoint {
e := &endpoint{
- stack: stack,
- netProto: netProto,
+ stack: s,
+ EndpointInfo: EndpointInfo{
+ TransportEndpointInfo: stack.TransportEndpointInfo{
+ NetProto: netProto,
+ TransProto: header.TCPProtocolNumber,
+ },
+ },
waiterQueue: waiterQueue,
state: StateInitial,
rcvBufSize: DefaultReceiveBufferSize,
@@ -446,26 +553,26 @@ func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, waite
}
var ss SendBufferSizeOption
- if err := stack.TransportProtocolOption(ProtocolNumber, &ss); err == nil {
+ if err := s.TransportProtocolOption(ProtocolNumber, &ss); err == nil {
e.sndBufSize = ss.Default
}
var rs ReceiveBufferSizeOption
- if err := stack.TransportProtocolOption(ProtocolNumber, &rs); err == nil {
+ if err := s.TransportProtocolOption(ProtocolNumber, &rs); err == nil {
e.rcvBufSize = rs.Default
}
var cs tcpip.CongestionControlOption
- if err := stack.TransportProtocolOption(ProtocolNumber, &cs); err == nil {
+ if err := s.TransportProtocolOption(ProtocolNumber, &cs); err == nil {
e.cc = cs
}
var mrb tcpip.ModerateReceiveBufferOption
- if err := stack.TransportProtocolOption(ProtocolNumber, &mrb); err == nil {
+ if err := s.TransportProtocolOption(ProtocolNumber, &mrb); err == nil {
e.rcvAutoParams.disabled = !bool(mrb)
}
- if p := stack.GetTCPProbe(); p != nil {
+ if p := s.GetTCPProbe(); p != nil {
e.probe = p
}
@@ -564,11 +671,11 @@ 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 +732,12 @@ 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
}
@@ -731,11 +838,12 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages,
bufUsed := e.rcvBufUsed
if s := e.state; !s.connected() && s != StateClose && bufUsed == 0 {
e.rcvListMu.Unlock()
- he := e.hardError
+ he := e.HardError
e.mu.RUnlock()
if s == StateError {
return buffer.View{}, tcpip.ControlMessages{}, he
}
+ e.stats.ReadErrors.InvalidEndpointState.Increment()
return buffer.View{}, tcpip.ControlMessages{}, tcpip.ErrInvalidEndpointState
}
@@ -744,6 +852,9 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages,
e.mu.RUnlock()
+ if err == tcpip.ErrClosedForReceive {
+ e.stats.ReadErrors.ReadClosed.Increment()
+ }
return v, tcpip.ControlMessages{}, err
}
@@ -787,7 +898,7 @@ func (e *endpoint) isEndpointWritableLocked() (int, *tcpip.Error) {
if !e.state.connected() {
switch e.state {
case StateError:
- return 0, e.hardError
+ return 0, e.HardError
default:
return 0, tcpip.ErrClosedForSend
}
@@ -806,7 +917,7 @@ func (e *endpoint) isEndpointWritableLocked() (int, *tcpip.Error) {
}
// Write writes data to the endpoint's peer.
-func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) {
+func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-chan struct{}, *tcpip.Error) {
// Linux completely ignores any address passed to sendto(2) for TCP sockets
// (without the MSG_FASTOPEN flag). Corking is unimplemented, so opts.More
// and opts.EndOfRecord are also ignored.
@@ -818,50 +929,57 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha
if err != nil {
e.sndBufMu.Unlock()
e.mu.RUnlock()
+ e.stats.WriteErrors.WriteClosed.Increment()
return 0, nil, err
}
- e.sndBufMu.Unlock()
- e.mu.RUnlock()
-
- // Nothing to do if the buffer is empty.
- if p.Size() == 0 {
- return 0, nil, nil
+ // We can release locks while copying data.
+ //
+ // This is not possible if atomic is set, because we can't allow the
+ // available buffer space to be consumed by some other caller while we
+ // are copying data in.
+ if !opts.Atomic {
+ e.sndBufMu.Unlock()
+ e.mu.RUnlock()
}
- // Copy in memory without holding sndBufMu so that worker goroutine can
- // make progress independent of this operation.
- v, perr := p.Get(avail)
- if perr != nil {
+ // Fetch data.
+ v, perr := p.Payload(avail)
+ if perr != nil || len(v) == 0 {
+ if opts.Atomic { // See above.
+ e.sndBufMu.Unlock()
+ e.mu.RUnlock()
+ }
+ // Note that perr may be nil if len(v) == 0.
return 0, nil, perr
}
- e.mu.RLock()
- e.sndBufMu.Lock()
+ if !opts.Atomic { // See above.
+ e.mu.RLock()
+ e.sndBufMu.Lock()
- // Because we released the lock before copying, check state again
- // to make sure the endpoint is still in a valid state for a
- // write.
- avail, err = e.isEndpointWritableLocked()
- if err != nil {
- e.sndBufMu.Unlock()
- e.mu.RUnlock()
- return 0, nil, err
- }
+ // Because we released the lock before copying, check state again
+ // to make sure the endpoint is still in a valid state for a write.
+ avail, err = e.isEndpointWritableLocked()
+ if err != nil {
+ e.sndBufMu.Unlock()
+ e.mu.RUnlock()
+ e.stats.WriteErrors.WriteClosed.Increment()
+ return 0, nil, err
+ }
- // Discard any excess data copied in due to avail being reduced due to a
- // simultaneous write call to the socket.
- if avail < len(v) {
- v = v[:avail]
+ // Discard any excess data copied in due to avail being reduced due
+ // to a simultaneous write call to the socket.
+ if avail < len(v) {
+ v = v[:avail]
+ }
}
// Add data to the send queue.
- l := len(v)
- s := newSegmentFromView(&e.route, e.id, v)
- e.sndBufUsed += l
- e.sndBufInQueue += seqnum.Size(l)
+ s := newSegmentFromView(&e.route, e.ID, v)
+ e.sndBufUsed += len(v)
+ e.sndBufInQueue += seqnum.Size(len(v))
e.sndQueue.PushBack(s)
-
e.sndBufMu.Unlock()
// Release the endpoint lock to prevent deadlocks due to lock
// order inversion when acquiring workMu.
@@ -875,7 +993,8 @@ func (e *endpoint) Write(p tcpip.Payload, opts tcpip.WriteOptions) (int64, <-cha
// Let the protocol goroutine do the work.
e.sndWaker.Assert()
}
- return int64(l), nil, nil
+
+ return int64(len(v)), nil, nil
}
// Peek reads data without consuming it from the endpoint.
@@ -889,8 +1008,9 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro
// but has some pending unread data.
if s := e.state; !s.connected() && s != StateClose {
if s == StateError {
- return 0, tcpip.ControlMessages{}, e.hardError
+ return 0, tcpip.ControlMessages{}, e.HardError
}
+ e.stats.ReadErrors.InvalidEndpointState.Increment()
return 0, tcpip.ControlMessages{}, tcpip.ErrInvalidEndpointState
}
@@ -899,6 +1019,7 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro
if e.rcvBufUsed == 0 {
if e.rcvClosed || !e.state.connected() {
+ e.stats.ReadErrors.ReadClosed.Increment()
return 0, tcpip.ControlMessages{}, tcpip.ErrClosedForReceive
}
return 0, tcpip.ControlMessages{}, tcpip.ErrWouldBlock
@@ -946,62 +1067,9 @@ func (e *endpoint) zeroReceiveWindow(scale uint8) bool {
return ((e.rcvBufSize - e.rcvBufUsed) >> scale) == 0
}
-// SetSockOpt sets a socket option.
-func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
- switch v := opt.(type) {
- case tcpip.DelayOption:
- if v == 0 {
- atomic.StoreUint32(&e.delay, 0)
-
- // Handle delayed data.
- e.sndWaker.Assert()
- } else {
- atomic.StoreUint32(&e.delay, 1)
- }
- return nil
-
- case tcpip.CorkOption:
- if v == 0 {
- atomic.StoreUint32(&e.cork, 0)
-
- // Handle the corked data.
- e.sndWaker.Assert()
- } else {
- atomic.StoreUint32(&e.cork, 1)
- }
- return nil
-
- case tcpip.ReuseAddressOption:
- e.mu.Lock()
- e.reuseAddr = v != 0
- e.mu.Unlock()
- return nil
-
- case tcpip.ReusePortOption:
- e.mu.Lock()
- e.reusePort = v != 0
- e.mu.Unlock()
- return nil
-
- case tcpip.QuickAckOption:
- if v == 0 {
- atomic.StoreUint32(&e.slowAck, 1)
- } else {
- atomic.StoreUint32(&e.slowAck, 0)
- }
- return nil
-
- case tcpip.MaxSegOption:
- userMSS := v
- if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS {
- return tcpip.ErrInvalidOptionValue
- }
- e.mu.Lock()
- e.userMSS = int(userMSS)
- e.mu.Unlock()
- e.notifyProtocolGoroutine(notifyMSSChanged)
- return nil
-
+// SetSockOptInt sets a socket option.
+func (e *endpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error {
+ switch opt {
case tcpip.ReceiveBufferSizeOption:
// Make sure the receive buffer size is within the min and max
// allowed.
@@ -1065,9 +1133,87 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
e.sndBufMu.Unlock()
return nil
+ default:
+ return nil
+ }
+}
+
+// SetSockOpt sets a socket option.
+func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
+ // Lower 2 bits represents ECN bits. RFC 3168, section 23.1
+ const inetECNMask = 3
+ switch v := opt.(type) {
+ case tcpip.DelayOption:
+ if v == 0 {
+ atomic.StoreUint32(&e.delay, 0)
+
+ // Handle delayed data.
+ e.sndWaker.Assert()
+ } else {
+ atomic.StoreUint32(&e.delay, 1)
+ }
+ return nil
+
+ case tcpip.CorkOption:
+ if v == 0 {
+ atomic.StoreUint32(&e.cork, 0)
+
+ // Handle the corked data.
+ e.sndWaker.Assert()
+ } else {
+ atomic.StoreUint32(&e.cork, 1)
+ }
+ return nil
+
+ case tcpip.ReuseAddressOption:
+ e.mu.Lock()
+ e.reuseAddr = v != 0
+ e.mu.Unlock()
+ return nil
+
+ case tcpip.ReusePortOption:
+ e.mu.Lock()
+ e.reusePort = v != 0
+ 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)
+ } else {
+ atomic.StoreUint32(&e.slowAck, 0)
+ }
+ return nil
+
+ case tcpip.MaxSegOption:
+ userMSS := v
+ if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS {
+ return tcpip.ErrInvalidOptionValue
+ }
+ e.mu.Lock()
+ e.userMSS = int(userMSS)
+ e.mu.Unlock()
+ e.notifyProtocolGoroutine(notifyMSSChanged)
+ return nil
+
case tcpip.V6OnlyOption:
// We only recognize this option on v6 endpoints.
- if e.netProto != header.IPv6ProtocolNumber {
+ if e.NetProto != header.IPv6ProtocolNumber {
return tcpip.ErrInvalidEndpointState
}
@@ -1082,6 +1228,12 @@ 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
@@ -1150,6 +1302,23 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
// Linux returns ENOENT when an invalid congestion
// control algorithm is specified.
return tcpip.ErrNoSuchFile
+
+ case tcpip.IPv4TOSOption:
+ e.mu.Lock()
+ // TODO(gvisor.dev/issue/995): ECN is not currently supported,
+ // ignore the bits for now.
+ e.sendTOS = uint8(v) & ^uint8(inetECNMask)
+ e.mu.Unlock()
+ return nil
+
+ case tcpip.IPv6TrafficClassOption:
+ e.mu.Lock()
+ // TODO(gvisor.dev/issue/995): ECN is not currently supported,
+ // ignore the bits for now.
+ e.sendTOS = uint8(v) & ^uint8(inetECNMask)
+ e.mu.Unlock()
+ return nil
+
default:
return nil
}
@@ -1176,6 +1345,18 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) {
switch opt {
case tcpip.ReceiveQueueSizeOption:
return e.readyReceiveSize()
+ case tcpip.SendBufferSizeOption:
+ e.sndBufMu.Lock()
+ v := e.sndBufSize
+ e.sndBufMu.Unlock()
+ return v, nil
+
+ case tcpip.ReceiveBufferSizeOption:
+ e.rcvListMu.Lock()
+ v := e.rcvBufSize
+ e.rcvListMu.Unlock()
+ return v, nil
+
}
return -1, tcpip.ErrUnknownProtocolOption
}
@@ -1198,18 +1379,6 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
*o = header.TCPDefaultMSS
return nil
- case *tcpip.SendBufferSizeOption:
- e.sndBufMu.Lock()
- *o = tcpip.SendBufferSizeOption(e.sndBufSize)
- e.sndBufMu.Unlock()
- return nil
-
- case *tcpip.ReceiveBufferSizeOption:
- e.rcvListMu.Lock()
- *o = tcpip.ReceiveBufferSizeOption(e.rcvBufSize)
- e.rcvListMu.Unlock()
- return nil
-
case *tcpip.DelayOption:
*o = 0
if v := atomic.LoadUint32(&e.delay); v != 0 {
@@ -1246,6 +1415,16 @@ 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 {
@@ -1255,7 +1434,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
case *tcpip.V6OnlyOption:
// We only recognize this option on v6 endpoints.
- if e.netProto != header.IPv6ProtocolNumber {
+ if e.NetProto != header.IPv6ProtocolNumber {
return tcpip.ErrUnknownProtocolOption
}
@@ -1269,6 +1448,12 @@ 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()
@@ -1333,13 +1518,25 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
e.mu.Unlock()
return nil
+ case *tcpip.IPv4TOSOption:
+ e.mu.RLock()
+ *o = tcpip.IPv4TOSOption(e.sendTOS)
+ e.mu.RUnlock()
+ return nil
+
+ case *tcpip.IPv6TrafficClassOption:
+ e.mu.RLock()
+ *o = tcpip.IPv6TrafficClassOption(e.sendTOS)
+ e.mu.RUnlock()
+ return nil
+
default:
return tcpip.ErrUnknownProtocolOption
}
}
func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) {
- netProto := e.netProto
+ netProto := e.NetProto
if header.IsV4MappedAddress(addr.Addr) {
// Fail if using a v4 mapped address on a v6only endpoint.
if e.v6only {
@@ -1355,7 +1552,7 @@ func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocol
// Fail if we're bound to an address length different from the one we're
// checking.
- if l := len(e.id.LocalAddress); l != 0 && len(addr.Addr) != 0 && l != len(addr.Addr) {
+ if l := len(e.ID.LocalAddress); l != 0 && len(addr.Addr) != 0 && l != len(addr.Addr) {
return 0, tcpip.ErrInvalidEndpointState
}
@@ -1369,7 +1566,12 @@ func (*endpoint) Disconnect() *tcpip.Error {
// Connect connects the endpoint to its peer.
func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
- return e.connect(addr, true, true)
+ err := e.connect(addr, true, true)
+ if err != nil && !err.IgnoreStats() {
+ e.stack.Stats().TCP.FailedConnectionAttempts.Increment()
+ e.stats.FailedConnectionAttempts.Increment()
+ }
+ return err
}
// connect connects the endpoint to its peer. In the normal non-S/R case, the
@@ -1378,14 +1580,9 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error {
// created (so no new handshaking is done); for stack-accepted connections not
// yet accepted by the app, they are restored without running the main goroutine
// here.
-func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (err *tcpip.Error) {
+func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tcpip.Error {
e.mu.Lock()
defer e.mu.Unlock()
- defer func() {
- if err != nil && !err.IgnoreStats() {
- e.stack.Stats().TCP.FailedConnectionAttempts.Increment()
- }
- }()
connectingAddr := addr.Addr
@@ -1430,29 +1627,29 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er
return tcpip.ErrAlreadyConnecting
case StateError:
- return e.hardError
+ return e.HardError
default:
return tcpip.ErrInvalidEndpointState
}
// Find a route to the desired destination.
- r, err := e.stack.FindRoute(nicid, e.id.LocalAddress, addr.Addr, netProto, false /* multicastLoop */)
+ r, err := e.stack.FindRoute(nicid, e.ID.LocalAddress, addr.Addr, netProto, false /* multicastLoop */)
if err != nil {
return err
}
defer r.Release()
- origID := e.id
+ origID := e.ID
netProtos := []tcpip.NetworkProtocolNumber{netProto}
- e.id.LocalAddress = r.LocalAddress
- e.id.RemoteAddress = r.RemoteAddress
- e.id.RemotePort = addr.Port
+ e.ID.LocalAddress = r.LocalAddress
+ e.ID.RemoteAddress = r.RemoteAddress
+ e.ID.RemotePort = addr.Port
- if e.id.LocalPort != 0 {
+ 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
}
@@ -1461,20 +1658,35 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er
// one. Make sure that it isn't one that will result in the same
// 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) {
- if sameAddr && p == e.id.RemotePort {
+ sameAddr := e.ID.LocalAddress == e.ID.RemoteAddress
+
+ // 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 := 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
+ e.ID = id
return true, nil
case tcpip.ErrPortInUse:
return false, nil
@@ -1490,7 +1702,7 @@ 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
}
@@ -1509,7 +1721,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) (er
e.segmentQueue.mu.Lock()
for _, l := range []segmentList{e.segmentQueue.list, e.sndQueue, e.snd.writeList} {
for s := l.Front(); s != nil; s = s.Next() {
- s.id = e.id
+ s.id = e.ID
s.route = r.Clone()
e.sndWaker.Assert()
}
@@ -1569,7 +1781,7 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error {
}
// Queue fin segment.
- s := newSegmentFromView(&e.route, e.id, nil)
+ s := newSegmentFromView(&e.route, e.ID, nil)
e.sndQueue.PushBack(s)
e.sndBufInQueue++
@@ -1597,14 +1809,18 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error {
// Listen puts the endpoint in "listen" mode, which allows it to accept
// new connections.
-func (e *endpoint) Listen(backlog int) (err *tcpip.Error) {
+func (e *endpoint) Listen(backlog int) *tcpip.Error {
+ err := e.listen(backlog)
+ if err != nil && !err.IgnoreStats() {
+ e.stack.Stats().TCP.FailedConnectionAttempts.Increment()
+ e.stats.FailedConnectionAttempts.Increment()
+ }
+ return err
+}
+
+func (e *endpoint) listen(backlog int) *tcpip.Error {
e.mu.Lock()
defer e.mu.Unlock()
- defer func() {
- if err != nil && !err.IgnoreStats() {
- e.stack.Stats().TCP.FailedConnectionAttempts.Increment()
- }
- }()
// Allow the backlog to be adjusted if the endpoint is not shutting down.
// When the endpoint shuts down, it sets workerCleanup to true, and from
@@ -1630,11 +1846,12 @@ func (e *endpoint) Listen(backlog int) (err *tcpip.Error) {
// Endpoint must be bound before it can transition to listen mode.
if e.state != StateBound {
+ e.stats.ReadErrors.InvalidEndpointState.Increment()
return tcpip.ErrInvalidEndpointState
}
// 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
}
@@ -1698,7 +1915,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) {
return tcpip.ErrAlreadyBound
}
- e.bindAddress = addr.Addr
+ e.BindAddr = addr.Addr
netProto, err := e.checkV4Mapped(&addr)
if err != nil {
return err
@@ -1715,26 +1932,26 @@ 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
}
e.isPortReserved = true
e.effectiveNetProtos = netProtos
- e.id.LocalPort = port
+ 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.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.
@@ -1745,7 +1962,7 @@ func (e *endpoint) Bind(addr tcpip.FullAddress) (err *tcpip.Error) {
}
e.boundNICID = nic
- e.id.LocalAddress = addr.Addr
+ e.ID.LocalAddress = addr.Addr
}
// Mark endpoint as bound.
@@ -1760,8 +1977,8 @@ func (e *endpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) {
defer e.mu.RUnlock()
return tcpip.FullAddress{
- Addr: e.id.LocalAddress,
- Port: e.id.LocalPort,
+ Addr: e.ID.LocalAddress,
+ Port: e.ID.LocalPort,
NIC: e.boundNICID,
}, nil
}
@@ -1776,8 +1993,8 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) {
}
return tcpip.FullAddress{
- Addr: e.id.RemoteAddress,
- Port: e.id.RemotePort,
+ Addr: e.ID.RemoteAddress,
+ Port: e.ID.RemotePort,
NIC: e.boundNICID,
}, nil
}
@@ -1789,6 +2006,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv
if !s.parse() {
e.stack.Stats().MalformedRcvdPackets.Increment()
e.stack.Stats().TCP.InvalidSegmentsReceived.Increment()
+ e.stats.ReceiveErrors.MalformedPacketsReceived.Increment()
s.decRef()
return
}
@@ -1796,11 +2014,13 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv
if !s.csumValid {
e.stack.Stats().MalformedRcvdPackets.Increment()
e.stack.Stats().TCP.ChecksumErrors.Increment()
+ e.stats.ReceiveErrors.ChecksumErrors.Increment()
s.decRef()
return
}
e.stack.Stats().TCP.ValidSegmentsReceived.Increment()
+ e.stats.SegmentsReceived.Increment()
if (s.flags & header.TCPFlagRst) != 0 {
e.stack.Stats().TCP.ResetsReceived.Increment()
}
@@ -1811,6 +2031,7 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, vv
} else {
// The queue is full, so we drop the segment.
e.stack.Stats().DroppedPackets.Increment()
+ e.stats.ReceiveErrors.SegmentQueueDropped.Increment()
s.decRef()
}
}
@@ -1860,6 +2081,7 @@ func (e *endpoint) readyToRead(s *segment) {
// that a subsequent read of the segment will correctly trigger
// a non-zero notification.
if avail := e.receiveBufferAvailableLocked(); avail>>e.rcv.rcvWndScale == 0 {
+ e.stats.ReceiveErrors.ZeroRcvWindowState.Increment()
e.zeroWindow = true
}
e.rcvList.PushBack(s)
@@ -2012,7 +2234,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState {
// Copy EndpointID.
e.mu.Lock()
- s.ID = stack.TCPEndpointID(e.id)
+ s.ID = stack.TCPEndpointID(e.ID)
e.mu.Unlock()
// Copy endpoint rcv state.
@@ -2119,7 +2341,7 @@ func (e *endpoint) initGSO() {
gso.Type = stack.GSOTCPv6
gso.L3HdrLen = header.IPv6MinimumSize
default:
- panic(fmt.Sprintf("Unknown netProto: %v", e.netProto))
+ panic(fmt.Sprintf("Unknown netProto: %v", e.NetProto))
}
gso.NeedsCsum = true
gso.CsumOffset = header.TCPChecksumOffset
@@ -2135,6 +2357,20 @@ func (e *endpoint) State() uint32 {
return uint32(e.state)
}
+// Info returns a copy of the endpoint info.
+func (e *endpoint) Info() tcpip.EndpointInfo {
+ e.mu.RLock()
+ // Make a copy of the endpoint info.
+ ret := e.EndpointInfo
+ e.mu.RUnlock()
+ return &ret
+}
+
+// Stats returns a pointer to the endpoint stats.
+func (e *endpoint) Stats() tcpip.EndpointStats {
+ return &e.stats
+}
+
func mssForRoute(r *stack.Route) uint16 {
return uint16(r.MTU() - header.TCPMinimumSize)
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-05 03:33:07 +0000