gvisor-lx/pkg/tcpip/transport/tcp/endpoint_state.go
Leadaxe 117243aa02 snapshot: sagernet/gvisor v0.0.0-20260727.0-sing-box-mod.1 + SPEC 048 guard
Обновление снапшота с v0.0.0-20250811.0 на пин, которого требует
sing-box после мержа 235 коммитов (upstream d620bbbf2 "Update gvisor to
20260727.0"). Прежний снапшот был взят 2026-08-04 ровно с той версии,
на которой тогда стоял апстрим; разрыв возник 2026-08-05 вместе с его
бампом.

За год апстрим-gvisor изменил ~14 000 строк в 292 файлах. Значимое для
нас — сетевой стек: tcp/connect.go (PMTU-discovery + исправление
начального RTT/RTO: раньше задержка ACK внутри стека завышала стартовый
таймаут на несколько RTT), tcp/snd.go, tcp/rcv.go, stack/conntrack.go,
stack/packet_buffer.go. Всего 30 файлов в TCP и 37 в stack.

Баг SPEC 048 апстрим НЕ исправил — проверено по коду новой версии:
handleConnecting по-прежнему проверяет состояние endpoint'а, но не ep.h,
а performHandshake так же зануляет h и отпускает мьютекс до Close().
Поэтому guard перенесён (12 строк) вместе со своим тестом (45 строк).

Red/green проверен на новой базе: без guard'а тест падает с той же
nil-паникой, что в полевом крашдампе; с ним зелёный.
2026-08-05 14:53:31 +03:00

379 lines
12 KiB
Go

// 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 (
"context"
"fmt"
"github.com/sagernet/gvisor/pkg/atomicbitops"
"github.com/sagernet/gvisor/pkg/log"
"github.com/sagernet/gvisor/pkg/sync"
"github.com/sagernet/gvisor/pkg/tcpip"
"github.com/sagernet/gvisor/pkg/tcpip/header"
"github.com/sagernet/gvisor/pkg/tcpip/seqnum"
"github.com/sagernet/gvisor/pkg/tcpip/stack"
)
// logDisconnectOnce ensures we don't spam logs when many connections are terminated.
var logDisconnectOnce sync.Once
func logDisconnect() {
logDisconnectOnce.Do(func() {
log.Infof("One or more TCP connections terminated during save restore")
})
}
// beforeSave is invoked by stateify.
func (e *Endpoint) beforeSave() {
// Stop incoming packets.
e.segmentQueue.freeze()
e.mu.Lock()
defer e.mu.Unlock()
epState := e.EndpointState()
switch {
case epState == StateInitial || epState == StateBound:
case epState.connected() || epState.handshake():
// Terminate valid connections only for restore.
if !e.stack.GetAllowConnectedOnSave() && !e.route.HasSaveRestoreCapability() {
if e.stack.GetRemoveConf() {
// Terminate the endpoint when resume=false.
e.terminateAtRestore = false
if !e.stack.AllowLiveTCPMigration() {
logDisconnect()
e.resetConnectionLocked(&tcpip.ErrConnectionAborted{})
e.mu.Unlock()
e.Close()
e.mu.Lock()
}
} else {
// This is set only when resume=true, the termination
// of this endpoint will happen during restore of the
// saved snapshot.
e.terminateAtRestore = true
}
}
fallthrough
case epState == StateListen:
// Nothing to do.
case epState.closed():
// Nothing to do.
default:
panic(fmt.Sprintf("endpoint in unknown state %v", e.EndpointState()))
}
e.stack.RegisterResumableEndpoint(e)
}
// saveEndpoints is invoked by stateify.
func (a *acceptQueue) saveEndpoints() []*Endpoint {
acceptedEndpoints := make([]*Endpoint, a.endpoints.Len())
for i, e := 0, a.endpoints.Front(); e != nil; i, e = i+1, e.Next() {
acceptedEndpoints[i] = e.Value.(*Endpoint)
}
return acceptedEndpoints
}
// loadEndpoints is invoked by stateify.
func (a *acceptQueue) loadEndpoints(_ context.Context, acceptedEndpoints []*Endpoint) {
for _, ep := range acceptedEndpoints {
a.endpoints.PushBack(ep)
}
}
// saveState is invoked by stateify.
func (e *Endpoint) saveState() EndpointState {
return e.EndpointState()
}
// Endpoint loading must be done in the following ordering by their state, to
// avoid dangling connecting w/o listening peer, and to avoid conflicts in port
// reservation.
var (
connectedLoading sync.WaitGroup
listenLoading sync.WaitGroup
connectingLoading sync.WaitGroup
)
// Bound endpoint loading happens last.
// loadState is invoked by stateify.
func (e *Endpoint) loadState(_ context.Context, epState EndpointState) {
// This is to ensure that the loading wait groups include all applicable
// endpoints before any asynchronous calls to the Wait() methods.
// For restore purposes we treat all endpoints with state after
// StateEstablished and before StateClosed like connected endpoint.
if epState.connected() {
connectedLoading.Add(1)
}
switch {
case epState == StateListen:
listenLoading.Add(1)
case epState.connecting():
connectingLoading.Add(1)
}
// Directly update the state here rather than using e.setEndpointState
// as the endpoint is still being loaded and the stack reference is not
// yet initialized.
e.state.Store(uint32(epState))
}
// afterLoad is invoked by stateify.
func (e *Endpoint) afterLoad(ctx context.Context) {
// RacyLoad() can be used because we are initializing e.
e.origEndpointState = e.state.RacyLoad()
// Restore the endpoint to InitialState as it will be moved to
// its origEndpointState during Restore.
e.state = atomicbitops.FromUint32(uint32(StateInitial))
e.stack.RegisterRestoredEndpoint(e)
}
// Close the endpoint during restore if terminateAtRestore was set for the endpoint.
func (e *Endpoint) closeEndpointAtRestore() {
e.mu.Lock()
defer e.mu.Unlock()
epState := EndpointState(e.origEndpointState)
if !epState.connected() && !epState.handshake() {
log.Debugf("endpoint was marked to terminate at restore in a wrong state, ID: %+v state: %v", e.ID, epState)
return
}
if epState.handshake() {
connectedLoading.Wait()
listenLoading.Wait()
}
// Put the endpoint in the error state and do cleanup. Do not
// attempt to send RST as route will be nil.
e.purgeReadQueue()
if epState.connected() {
e.purgeWriteQueue()
e.purgePendingRcvQueue()
e.cleanupLocked()
}
e.state.Store(uint32(StateError))
e.closeNoShutdownLocked()
tcpip.DeleteDanglingEndpoint(e)
if epState.connected() {
connectedLoading.Done()
} else {
connectingLoading.Done()
}
}
// Restore implements tcpip.RestoredEndpoint.Restore.
func (e *Endpoint) Restore(s *stack.Stack) {
if !e.EndpointState().closed() {
e.keepalive.timer.init(s.Clock(), timerHandler(e, e.keepaliveTimerExpired))
}
if snd := e.snd; snd != nil {
snd.resendTimer.init(s.Clock(), timerHandler(e, e.snd.retransmitTimerExpired))
snd.reorderTimer.init(s.Clock(), timerHandler(e, e.snd.rc.reorderTimerExpired))
snd.probeTimer.init(s.Clock(), timerHandler(e, e.snd.probeTimerExpired))
snd.corkTimer.init(s.Clock(), timerHandler(e, e.snd.corkTimerExpired))
}
e.ops.InitHandler(e, e.stack, GetTCPSendBufferLimits, GetTCPReceiveBufferLimits)
e.segmentQueue.thaw()
e.mu.Lock()
id := e.ID
terminateAtRestore := e.terminateAtRestore
e.mu.Unlock()
bind := func() {
e.mu.Lock()
defer e.mu.Unlock()
e.isPortReserved = true
// Mark endpoint as bound.
e.setEndpointState(StateBound)
}
if terminateAtRestore && !e.stack.AllowLiveTCPMigration() {
e.closeEndpointAtRestore()
return
}
epState := EndpointState(e.origEndpointState)
switch {
case epState.connected():
if e.stack.AllowLiveTCPMigration() {
// Handle dual stack addresses.
netProto := e.NetProto
switch e.TransportEndpointInfo.ID.LocalAddress.BitLen() {
case header.IPv4AddressSizeBits:
netProto = header.IPv4ProtocolNumber
case header.IPv6AddressSizeBits:
netProto = header.IPv6ProtocolNumber
}
// Get the new local NIC for source IP and do a FindRoute here to
// identify if the network config is same. Then only attempt restore,
// else close the connection on our end.
r, err := e.stack.FindRoute(0, e.TransportEndpointInfo.ID.LocalAddress, e.TransportEndpointInfo.ID.RemoteAddress, netProto, false /* multicastLoop */)
if err != nil {
e.closeEndpointAtRestore()
log.Infof("Cannot find the route %+v", e.TransportEndpointInfo.ID)
return
}
e.boundNICID = r.NICID()
r.Release()
}
bind()
if e.connectingAddress.BitLen() == 0 {
e.connectingAddress = e.TransportEndpointInfo.ID.RemoteAddress
// This endpoint is accepted by netstack but not yet by
// the app. If the endpoint is IPv6 but the remote
// address is IPv4, we need to connect as IPv6 so that
// dual-stack mode can be properly activated.
if e.NetProto == header.IPv6ProtocolNumber && e.TransportEndpointInfo.ID.RemoteAddress.BitLen() != header.IPv6AddressSizeBits {
e.connectingAddress = tcpip.AddrFrom16Slice(append(
[]byte{0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xff, 0xff},
e.TransportEndpointInfo.ID.RemoteAddress.AsSlice()...,
))
}
}
// Reset the scoreboard to reinitialize the sack information as
// we do not restore SACK information.
e.scoreboard.Reset()
// Unregister the endpoint before registering again during Connect.
e.stack.UnregisterTransportEndpoint(e.effectiveNetProtos, header.TCPProtocolNumber, e.TransportEndpointInfo.ID, e, e.boundPortFlags, e.boundBindToDevice)
e.mu.Lock()
err := e.connect(tcpip.FullAddress{NIC: e.boundNICID, Addr: e.connectingAddress, Port: e.TransportEndpointInfo.ID.RemotePort}, false /* handshake */)
if _, ok := err.(*tcpip.ErrConnectStarted); !ok {
log.Warningf("TCP endpoint connect failed for connected endpoint with ID: %+v err: %v", id, err)
e.mu.Unlock()
e.Close()
connectedLoading.Done()
return
}
e.state.Store(e.origEndpointState)
log.Infof("connect success: %+v", e.TransportEndpointInfo.ID)
// For FIN-WAIT-2 and TIME-WAIT we need to start the appropriate timers so
// that the socket is closed correctly.
switch epState {
case StateFinWait2:
e.finWait2Timer = e.stack.Clock().AfterFunc(e.tcpLingerTimeout, e.finWait2TimerExpired)
case StateTimeWait:
e.timeWaitTimer = e.stack.Clock().AfterFunc(e.getTimeWaitDuration(), e.timeWaitTimerExpired)
}
if e.ops.GetCorkOption() {
// Rearm the timer if TCP_CORK is enabled which will
// drain all the segments in the queue after restore.
e.snd.corkTimer.enable(MinRTO)
}
e.mu.Unlock()
e.requeueOnRestore()
connectedLoading.Done()
case epState == StateListen:
tcpip.AsyncLoading.Add(1)
go func() {
connectedLoading.Wait()
e.LockUser()
// All endpoints will be moved to initial state after
// restore. Set endpoint to its originial listen state.
e.setEndpointState(StateListen)
// Initialize the listening context.
rcvWnd := seqnum.Size(e.receiveBufferAvailable())
e.listenCtx = newListenContext(e.stack, e.protocol, e, rcvWnd, e.ops.GetV6Only(), e.NetProto)
e.UnlockUser()
e.requeueOnRestore()
listenLoading.Done()
tcpip.AsyncLoading.Done()
}()
case epState == StateConnecting:
// Initial SYN hasn't been sent yet so initiate a connect.
tcpip.AsyncLoading.Add(1)
go func() {
connectedLoading.Wait()
listenLoading.Wait()
bind()
err := e.Connect(tcpip.FullAddress{NIC: e.boundNICID, Addr: e.connectingAddress, Port: e.TransportEndpointInfo.ID.RemotePort})
if _, ok := err.(*tcpip.ErrConnectStarted); !ok {
log.Warningf("TCP endpoint connect failed for connecting endpoint with ID: %+v err: %v", id, err)
e.Close()
}
connectingLoading.Done()
tcpip.AsyncLoading.Done()
}()
case epState == StateSynSent || epState == StateSynRecv:
tcpip.AsyncLoading.Add(1)
go func() {
connectedLoading.Wait()
listenLoading.Wait()
// Initial SYN has been sent/received so we should bind the
// ports start the retransmit timer for the SYNs and let it
// naturally complete the connection.
bind()
e.mu.Lock()
e.setEndpointState(epState)
r, err := e.stack.FindRoute(e.boundNICID, e.TransportEndpointInfo.ID.LocalAddress, e.TransportEndpointInfo.ID.RemoteAddress, e.effectiveNetProtos[0], false /* multicastLoop */)
if err != nil {
e.mu.Unlock()
log.Warningf("FindRoute failed when restoring endpoint w/ ID: %+v err: %v", id, err)
e.Close()
connectingLoading.Done()
tcpip.AsyncLoading.Done()
return
}
e.route = r
timer, err := newBackoffTimer(e.stack.Clock(), InitialRTO, MaxRTO, timerHandler(e, e.h.retransmitHandlerLocked))
if err != nil {
panic(fmt.Sprintf("newBackOffTimer(_, %s, %s, _) failed: %s", InitialRTO, MaxRTO, err))
}
e.h.retransmitTimer = timer
connectingLoading.Done()
tcpip.AsyncLoading.Done()
e.mu.Unlock()
e.requeueOnRestore()
}()
case epState == StateBound:
tcpip.AsyncLoading.Add(1)
go func() {
connectedLoading.Wait()
listenLoading.Wait()
connectingLoading.Wait()
bind()
tcpip.AsyncLoading.Done()
}()
case epState == StateClose:
e.isPortReserved = false
e.state.Store(uint32(StateClose))
e.stack.CompleteTransportEndpointCleanup(e)
tcpip.DeleteDanglingEndpoint(e)
case epState == StateError:
e.state.Store(uint32(StateError))
e.stack.CompleteTransportEndpointCleanup(e)
tcpip.DeleteDanglingEndpoint(e)
}
}
// Resume implements tcpip.ResumableEndpoint.Resume.
func (e *Endpoint) Resume() {
e.segmentQueue.thaw()
}
// requeueOnRestore re-adds the endpoint to its processor's run-queue if it has
// queued segments. The run-queue is not saved across checkpoint/restore.
func (e *Endpoint) requeueOnRestore() {
if e.segmentQueue.empty() || e.isOwnedByUser() {
return
}
e.protocol.dispatcher.selectProcessor(e.TransportEndpointInfo.ID).queueEndpoint(e)
}