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.
This commit is contained in:
Leadaxe
parent
ffebe42860
commit
117243aa02
293 changed files with 16413 additions and 2842 deletions
|
|
@ -60,5 +60,5 @@ func queueDispatcherinitLockNames() {}
|
|||
|
||||
func init() {
|
||||
queueDispatcherinitLockNames()
|
||||
queueDispatcherprefixIndex = locking.NewMutexClass(reflect.TypeOf(queueDispatcherMutex{}), queueDispatcherlockNames)
|
||||
queueDispatcherprefixIndex = locking.NewMutexClass(reflect.TypeFor[queueDispatcherMutex](), queueDispatcherlockNames)
|
||||
}
|
||||
|
|
|
|||
|
|
@ -22,6 +22,7 @@ import (
|
|||
"github.com/sagernet/gvisor/pkg/sleep"
|
||||
"github.com/sagernet/gvisor/pkg/sync"
|
||||
"github.com/sagernet/gvisor/pkg/tcpip"
|
||||
"github.com/sagernet/gvisor/pkg/tcpip/link/qdisc"
|
||||
"github.com/sagernet/gvisor/pkg/tcpip/stack"
|
||||
)
|
||||
|
||||
|
|
@ -60,7 +61,7 @@ type queueDispatcher struct {
|
|||
|
||||
mu queueDispatcherMutex `state:"nosave"`
|
||||
// +checklocks:mu
|
||||
queue packetBufferCircularList
|
||||
queue qdisc.PacketBufferCircularList
|
||||
|
||||
newPacketWaker sleep.Waker `state:"nosave"`
|
||||
closeWaker sleep.Waker `state:"nosave"`
|
||||
|
|
@ -78,7 +79,7 @@ func New(lower stack.LinkWriter, n int, queueLen int) stack.QueueingDiscipline {
|
|||
for i := range d.dispatchers {
|
||||
qd := &d.dispatchers[i]
|
||||
qd.lower = lower
|
||||
qd.queue.init(queueLen)
|
||||
qd.queue.Init(queueLen)
|
||||
|
||||
d.wg.Add(1)
|
||||
go func() {
|
||||
|
|
@ -101,19 +102,19 @@ func (qd *queueDispatcher) dispatchLoop() {
|
|||
case &qd.newPacketWaker:
|
||||
case &qd.closeWaker:
|
||||
qd.mu.Lock()
|
||||
for p := qd.queue.removeFront(); p != nil; p = qd.queue.removeFront() {
|
||||
for p := qd.queue.RemoveFront(); p != nil; p = qd.queue.RemoveFront() {
|
||||
p.DecRef()
|
||||
}
|
||||
qd.queue.decRef()
|
||||
qd.queue.DecRef()
|
||||
qd.mu.Unlock()
|
||||
return
|
||||
default:
|
||||
panic("unknown waker")
|
||||
}
|
||||
qd.mu.Lock()
|
||||
for pkt := qd.queue.removeFront(); pkt != nil; pkt = qd.queue.removeFront() {
|
||||
for pkt := qd.queue.RemoveFront(); pkt != nil; pkt = qd.queue.RemoveFront() {
|
||||
batch.PushBack(pkt)
|
||||
if batch.Len() < BatchSize && !qd.queue.isEmpty() {
|
||||
if batch.Len() < BatchSize && !qd.queue.IsEmpty() {
|
||||
continue
|
||||
}
|
||||
qd.mu.Unlock()
|
||||
|
|
@ -137,9 +138,13 @@ func (d *discipline) WritePacket(pkt *stack.PacketBuffer) tcpip.Error {
|
|||
}
|
||||
qd := &d.dispatchers[int(pkt.Hash)%len(d.dispatchers)]
|
||||
qd.mu.Lock()
|
||||
haveSpace := qd.queue.hasSpace()
|
||||
if d.closed.Load() == qDiscClosed {
|
||||
qd.mu.Unlock()
|
||||
return &tcpip.ErrClosedForSend{}
|
||||
}
|
||||
haveSpace := qd.queue.HasSpace()
|
||||
if haveSpace {
|
||||
qd.queue.pushBack(pkt.IncRef())
|
||||
qd.queue.PushBack(pkt.IncRef())
|
||||
}
|
||||
qd.mu.Unlock()
|
||||
if !haveSpace {
|
||||
|
|
|
|||
|
|
@ -64,39 +64,7 @@ func (qd *queueDispatcher) StateLoad(ctx context.Context, stateSourceObject stat
|
|||
stateSourceObject.Load(1, &qd.queue)
|
||||
}
|
||||
|
||||
func (pl *packetBufferCircularList) StateTypeName() string {
|
||||
return "pkg/tcpip/link/qdisc/fifo.packetBufferCircularList"
|
||||
}
|
||||
|
||||
func (pl *packetBufferCircularList) StateFields() []string {
|
||||
return []string{
|
||||
"pbs",
|
||||
"head",
|
||||
"size",
|
||||
}
|
||||
}
|
||||
|
||||
func (pl *packetBufferCircularList) beforeSave() {}
|
||||
|
||||
// +checklocksignore
|
||||
func (pl *packetBufferCircularList) StateSave(stateSinkObject state.Sink) {
|
||||
pl.beforeSave()
|
||||
stateSinkObject.Save(0, &pl.pbs)
|
||||
stateSinkObject.Save(1, &pl.head)
|
||||
stateSinkObject.Save(2, &pl.size)
|
||||
}
|
||||
|
||||
func (pl *packetBufferCircularList) afterLoad(context.Context) {}
|
||||
|
||||
// +checklocksignore
|
||||
func (pl *packetBufferCircularList) StateLoad(ctx context.Context, stateSourceObject state.Source) {
|
||||
stateSourceObject.Load(0, &pl.pbs)
|
||||
stateSourceObject.Load(1, &pl.head)
|
||||
stateSourceObject.Load(2, &pl.size)
|
||||
}
|
||||
|
||||
func init() {
|
||||
state.Register((*discipline)(nil))
|
||||
state.Register((*queueDispatcher)(nil))
|
||||
state.Register((*packetBufferCircularList)(nil))
|
||||
}
|
||||
|
|
|
|||
|
|
@ -1,93 +0,0 @@
|
|||
// Copyright 2022 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 fifo
|
||||
|
||||
import "github.com/sagernet/gvisor/pkg/tcpip/stack"
|
||||
|
||||
// packetBufferCircularList is a slice-backed circular list. All operations are
|
||||
// O(1) unless otherwise noted. It only allocates once, during the call to
|
||||
// init().
|
||||
//
|
||||
// Users should call init() before using packetBufferCircularList.
|
||||
//
|
||||
// +stateify savable
|
||||
type packetBufferCircularList struct {
|
||||
pbs []*stack.PacketBuffer
|
||||
head int
|
||||
size int
|
||||
}
|
||||
|
||||
// init initializes the list with the given size.
|
||||
func (pl *packetBufferCircularList) init(size int) {
|
||||
pl.pbs = make([]*stack.PacketBuffer, size)
|
||||
}
|
||||
|
||||
// length returns the number of elements in the list.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *packetBufferCircularList) length() int {
|
||||
return pl.size
|
||||
}
|
||||
|
||||
// hasSpace returns whether there is space left in the list.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *packetBufferCircularList) hasSpace() bool {
|
||||
return pl.size < len(pl.pbs)
|
||||
}
|
||||
|
||||
// isEmpty returns whether the list is empty.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *packetBufferCircularList) isEmpty() bool {
|
||||
return pl.size == 0
|
||||
}
|
||||
|
||||
// pushBack inserts the PacketBuffer at the end of the list.
|
||||
//
|
||||
// Users must check beforehand that there is space via a call to hasSpace().
|
||||
// Failing to do so may clobber existing entries.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *packetBufferCircularList) pushBack(pb *stack.PacketBuffer) {
|
||||
next := (pl.head + pl.size) % len(pl.pbs)
|
||||
pl.pbs[next] = pb
|
||||
pl.size++
|
||||
}
|
||||
|
||||
// removeFront returns the first element of the list or nil.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *packetBufferCircularList) removeFront() *stack.PacketBuffer {
|
||||
if pl.isEmpty() {
|
||||
return nil
|
||||
}
|
||||
ret := pl.pbs[pl.head]
|
||||
pl.pbs[pl.head] = nil
|
||||
pl.head = (pl.head + 1) % len(pl.pbs)
|
||||
pl.size--
|
||||
return ret
|
||||
}
|
||||
|
||||
// decRef decreases the reference count on each stack.PacketBuffer stored in
|
||||
// the list.
|
||||
//
|
||||
// NOTE: runs in O(n) time.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *packetBufferCircularList) decRef() {
|
||||
for i := 0; i < pl.size; i++ {
|
||||
pl.pbs[(pl.head+i)%len(pl.pbs)].DecRef()
|
||||
}
|
||||
}
|
||||
108
pkg/tcpip/link/qdisc/packet_buffer_circular_list.go
Normal file
108
pkg/tcpip/link/qdisc/packet_buffer_circular_list.go
Normal file
|
|
@ -0,0 +1,108 @@
|
|||
// Copyright 2022 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 qdisc provides shared building blocks used by queueing disciplines.
|
||||
package qdisc
|
||||
|
||||
import "github.com/sagernet/gvisor/pkg/tcpip/stack"
|
||||
|
||||
// PacketBufferCircularList is a slice-backed circular list. All operations are
|
||||
// O(1) unless otherwise noted. It only allocates once, during the call to
|
||||
// Init().
|
||||
//
|
||||
// Users should call Init() before using PacketBufferCircularList.
|
||||
//
|
||||
// +stateify savable
|
||||
type PacketBufferCircularList struct {
|
||||
pbs []*stack.PacketBuffer
|
||||
head int
|
||||
size int
|
||||
}
|
||||
|
||||
// Init initializes the list with the given size.
|
||||
func (pl *PacketBufferCircularList) Init(size int) {
|
||||
pl.pbs = make([]*stack.PacketBuffer, size)
|
||||
}
|
||||
|
||||
// Length returns the number of elements in the list.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *PacketBufferCircularList) Length() int {
|
||||
return pl.size
|
||||
}
|
||||
|
||||
// HasSpace returns whether there is space left in the list.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *PacketBufferCircularList) HasSpace() bool {
|
||||
return pl.size < len(pl.pbs)
|
||||
}
|
||||
|
||||
// IsEmpty returns whether the list is empty.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *PacketBufferCircularList) IsEmpty() bool {
|
||||
return pl.size == 0
|
||||
}
|
||||
|
||||
// PushBack inserts the PacketBuffer at the end of the list.
|
||||
//
|
||||
// Users must check beforehand that there is space via a call to HasSpace().
|
||||
// Failing to do so may clobber existing entries.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *PacketBufferCircularList) PushBack(pb *stack.PacketBuffer) {
|
||||
next := (pl.head + pl.size) % len(pl.pbs)
|
||||
pl.pbs[next] = pb
|
||||
pl.size++
|
||||
}
|
||||
|
||||
// PeekFront returns the first element of the list without removing it, or nil
|
||||
// if empty. The list retains its reference; the caller must not DecRef. To take
|
||||
// ownership, call RemoveFront, which returns the same pointer. The returned
|
||||
// pointer is only valid until the next mutation of the list.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *PacketBufferCircularList) PeekFront() *stack.PacketBuffer {
|
||||
if pl.IsEmpty() {
|
||||
return nil
|
||||
}
|
||||
return pl.pbs[pl.head]
|
||||
}
|
||||
|
||||
// RemoveFront returns the first element of the list or nil.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *PacketBufferCircularList) RemoveFront() *stack.PacketBuffer {
|
||||
if pl.IsEmpty() {
|
||||
return nil
|
||||
}
|
||||
ret := pl.pbs[pl.head]
|
||||
pl.pbs[pl.head] = nil
|
||||
pl.head = (pl.head + 1) % len(pl.pbs)
|
||||
pl.size--
|
||||
return ret
|
||||
}
|
||||
|
||||
// DecRef decreases the reference count on each stack.PacketBuffer stored in
|
||||
// the list.
|
||||
//
|
||||
// NOTE: runs in O(n) time.
|
||||
//
|
||||
//go:nosplit
|
||||
func (pl *PacketBufferCircularList) DecRef() {
|
||||
for i := 0; i < pl.size; i++ {
|
||||
pl.pbs[(pl.head+i)%len(pl.pbs)].DecRef()
|
||||
}
|
||||
}
|
||||
44
pkg/tcpip/link/qdisc/qdisc_state_autogen.go
Normal file
44
pkg/tcpip/link/qdisc/qdisc_state_autogen.go
Normal file
|
|
@ -0,0 +1,44 @@
|
|||
// automatically generated by stateify.
|
||||
|
||||
package qdisc
|
||||
|
||||
import (
|
||||
"context"
|
||||
|
||||
"github.com/sagernet/gvisor/pkg/state"
|
||||
)
|
||||
|
||||
func (pl *PacketBufferCircularList) StateTypeName() string {
|
||||
return "pkg/tcpip/link/qdisc.PacketBufferCircularList"
|
||||
}
|
||||
|
||||
func (pl *PacketBufferCircularList) StateFields() []string {
|
||||
return []string{
|
||||
"pbs",
|
||||
"head",
|
||||
"size",
|
||||
}
|
||||
}
|
||||
|
||||
func (pl *PacketBufferCircularList) beforeSave() {}
|
||||
|
||||
// +checklocksignore
|
||||
func (pl *PacketBufferCircularList) StateSave(stateSinkObject state.Sink) {
|
||||
pl.beforeSave()
|
||||
stateSinkObject.Save(0, &pl.pbs)
|
||||
stateSinkObject.Save(1, &pl.head)
|
||||
stateSinkObject.Save(2, &pl.size)
|
||||
}
|
||||
|
||||
func (pl *PacketBufferCircularList) afterLoad(context.Context) {}
|
||||
|
||||
// +checklocksignore
|
||||
func (pl *PacketBufferCircularList) StateLoad(ctx context.Context, stateSourceObject state.Source) {
|
||||
stateSourceObject.Load(0, &pl.pbs)
|
||||
stateSourceObject.Load(1, &pl.head)
|
||||
stateSourceObject.Load(2, &pl.size)
|
||||
}
|
||||
|
||||
func init() {
|
||||
state.Register((*PacketBufferCircularList)(nil))
|
||||
}
|
||||
64
pkg/tcpip/link/qdisc/tbf/queue_mutex.go
Normal file
64
pkg/tcpip/link/qdisc/tbf/queue_mutex.go
Normal file
|
|
@ -0,0 +1,64 @@
|
|||
package tbf
|
||||
|
||||
import (
|
||||
"reflect"
|
||||
|
||||
"github.com/sagernet/gvisor/pkg/sync"
|
||||
"github.com/sagernet/gvisor/pkg/sync/locking"
|
||||
)
|
||||
|
||||
// Mutex is sync.Mutex with the correctness validator.
|
||||
type queueMutex struct {
|
||||
mu sync.Mutex
|
||||
}
|
||||
|
||||
var queueprefixIndex *locking.MutexClass
|
||||
|
||||
// lockNames is a list of user-friendly lock names.
|
||||
// Populated in init.
|
||||
var queuelockNames []string
|
||||
|
||||
// lockNameIndex is used as an index passed to NestedLock and NestedUnlock,
|
||||
// referring to an index within lockNames.
|
||||
// Values are specified using the "consts" field of go_template_instance.
|
||||
type queuelockNameIndex int
|
||||
|
||||
// DO NOT REMOVE: The following function automatically replaced with lock index constants.
|
||||
// LOCK_NAME_INDEX_CONSTANTS
|
||||
const ()
|
||||
|
||||
// Lock locks m.
|
||||
// +checklocksignore
|
||||
func (m *queueMutex) Lock() {
|
||||
locking.AddGLock(queueprefixIndex, -1)
|
||||
m.mu.Lock()
|
||||
}
|
||||
|
||||
// NestedLock locks m knowing that another lock of the same type is held.
|
||||
// +checklocksignore
|
||||
func (m *queueMutex) NestedLock(i queuelockNameIndex) {
|
||||
locking.AddGLock(queueprefixIndex, int(i))
|
||||
m.mu.Lock()
|
||||
}
|
||||
|
||||
// Unlock unlocks m.
|
||||
// +checklocksignore
|
||||
func (m *queueMutex) Unlock() {
|
||||
locking.DelGLock(queueprefixIndex, -1)
|
||||
m.mu.Unlock()
|
||||
}
|
||||
|
||||
// NestedUnlock unlocks m knowing that another lock of the same type is held.
|
||||
// +checklocksignore
|
||||
func (m *queueMutex) NestedUnlock(i queuelockNameIndex) {
|
||||
locking.DelGLock(queueprefixIndex, int(i))
|
||||
m.mu.Unlock()
|
||||
}
|
||||
|
||||
// DO NOT REMOVE: The following function is automatically replaced.
|
||||
func queueinitLockNames() {}
|
||||
|
||||
func init() {
|
||||
queueinitLockNames()
|
||||
queueprefixIndex = locking.NewMutexClass(reflect.TypeFor[queueMutex](), queuelockNames)
|
||||
}
|
||||
239
pkg/tcpip/link/qdisc/tbf/tbf.go
Normal file
239
pkg/tcpip/link/qdisc/tbf/tbf.go
Normal file
|
|
@ -0,0 +1,239 @@
|
|||
// Copyright 2026 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 tbf provides a simplified Token Bucket Filter queueing discipline
|
||||
// modeled on Linux's net/sched/sch_tbf.c. Only the single-rate bucket is
|
||||
// implemented; peakrate/peakburst (Linux's second bucket) is not.
|
||||
package tbf
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"time"
|
||||
|
||||
"github.com/sagernet/gvisor/pkg/atomicbitops"
|
||||
"github.com/sagernet/gvisor/pkg/sleep"
|
||||
"github.com/sagernet/gvisor/pkg/sync"
|
||||
"github.com/sagernet/gvisor/pkg/tcpip"
|
||||
"github.com/sagernet/gvisor/pkg/tcpip/link/qdisc"
|
||||
"github.com/sagernet/gvisor/pkg/tcpip/stack"
|
||||
)
|
||||
|
||||
const (
|
||||
// BatchSize is the number of packets to write in each syscall. It is 47
|
||||
// because when GVisorGSO is in use then a single 65KB TCP segment can get
|
||||
// split into 46 segments of 1420 bytes and a single 216 byte segment.
|
||||
BatchSize = 47
|
||||
|
||||
qDiscClosed = 1
|
||||
)
|
||||
|
||||
var _ stack.QueueingDiscipline = (*discipline)(nil)
|
||||
|
||||
// +stateify savable
|
||||
type discipline struct {
|
||||
// Immutable configuration set by New.
|
||||
lower stack.LinkWriter
|
||||
clock tcpip.Clock `state:"nosave"`
|
||||
rate uint64 // max sustained throughput, bytes/sec
|
||||
burst uint32 // largest packet this TBF will pass, bytes
|
||||
buffer int64 // nanoseconds needed to transmit burst bytes at rate
|
||||
|
||||
// Shutdown state.
|
||||
wg sync.WaitGroup `state:"nosave"`
|
||||
closed atomicbitops.Int32
|
||||
|
||||
// Wakers driving dispatchLoop.
|
||||
newPacketWaker sleep.Waker `state:"nosave"`
|
||||
tokenWaker sleep.Waker `state:"nosave"`
|
||||
closeWaker sleep.Waker `state:"nosave"`
|
||||
|
||||
mu queueMutex `state:"nosave"`
|
||||
// +checklocks:mu
|
||||
queue qdisc.PacketBufferCircularList
|
||||
|
||||
// Dispatcher state: mutated only inside dispatchLoop and
|
||||
// thus not protected by mu.
|
||||
tokens int64 // current bucket level, ns
|
||||
timeCheckpoint tcpip.MonotonicTime
|
||||
watchdog tcpip.Timer `state:"nosave"`
|
||||
}
|
||||
|
||||
// len2TimeNS returns the number of ns to transmit len bytes at rate bytes/sec.
|
||||
// Linux's psched_l2t_ns avoids the divide via a precomputed mult/shift; see
|
||||
// psched_ratecfg_precompute__ in net/sched/sch_generic.c.
|
||||
func len2TimeNS(rate uint64, len uint32) uint64 {
|
||||
const nsecPerSec = 1000000000
|
||||
return uint64(len) * nsecPerSec / rate
|
||||
}
|
||||
|
||||
func (d *discipline) dispatchLoop() {
|
||||
s := sleep.Sleeper{}
|
||||
s.AddWaker(&d.newPacketWaker)
|
||||
s.AddWaker(&d.tokenWaker)
|
||||
s.AddWaker(&d.closeWaker)
|
||||
defer s.Done()
|
||||
|
||||
var batch stack.PacketBufferList
|
||||
for {
|
||||
switch w := s.Fetch(true); w {
|
||||
case &d.newPacketWaker, &d.tokenWaker:
|
||||
case &d.closeWaker:
|
||||
if d.watchdog != nil {
|
||||
d.watchdog.Stop()
|
||||
}
|
||||
d.mu.Lock()
|
||||
for p := d.queue.RemoveFront(); p != nil; p = d.queue.RemoveFront() {
|
||||
p.DecRef()
|
||||
}
|
||||
d.queue.DecRef()
|
||||
d.mu.Unlock()
|
||||
return
|
||||
default:
|
||||
panic("unknown waker")
|
||||
}
|
||||
|
||||
d.mu.Lock()
|
||||
for pkt := d.queue.PeekFront(); pkt != nil; pkt = d.queue.PeekFront() {
|
||||
pktLen := pkt.Size()
|
||||
now := d.clock.NowMonotonic()
|
||||
toks := min(now.Sub(d.timeCheckpoint).Nanoseconds(), d.buffer)
|
||||
toks += d.tokens
|
||||
if toks > d.buffer {
|
||||
toks = d.buffer
|
||||
}
|
||||
toks -= int64(len2TimeNS(d.rate, uint32(pktLen)))
|
||||
sufficientTokens := toks >= 0
|
||||
if !sufficientTokens {
|
||||
// -toks is the deficit in ns: how long until enough tokens accumulate.
|
||||
if d.watchdog != nil {
|
||||
d.watchdog.Stop()
|
||||
}
|
||||
d.watchdog = d.clock.AfterFunc(time.Duration(-toks), d.tokenWaker.Assert)
|
||||
break
|
||||
}
|
||||
d.queue.RemoveFront()
|
||||
d.timeCheckpoint = now
|
||||
d.tokens = toks
|
||||
batch.PushBack(pkt)
|
||||
|
||||
possiblyAnotherPacket := batch.Len() < BatchSize && !d.queue.IsEmpty()
|
||||
if possiblyAnotherPacket {
|
||||
continue
|
||||
}
|
||||
d.mu.Unlock()
|
||||
_, _ = d.lower.WritePackets(batch)
|
||||
batch.Reset()
|
||||
d.mu.Lock()
|
||||
}
|
||||
if batch.Len() > 0 {
|
||||
d.mu.Unlock()
|
||||
_, _ = d.lower.WritePackets(batch)
|
||||
batch.Reset()
|
||||
d.mu.Lock()
|
||||
}
|
||||
d.mu.Unlock()
|
||||
}
|
||||
}
|
||||
|
||||
// New creates a new TBF queueing discipline that will rate-limit lower to
|
||||
// rate bytes/sec with bursts of up to burst bytes, queueing up to queueLen
|
||||
// packets of backlog before dropping. Note that queueLen counts packets,
|
||||
// not bytes as in Linux's sch_tbf.c, for consistency with the fifo qdisc.
|
||||
//
|
||||
// +checklocksignore: we don't have to hold locks during initialization.
|
||||
func New(lower stack.LinkEndpoint, clock tcpip.Clock, rate uint64, burst, queueLen uint32) (stack.QueueingDiscipline, error) {
|
||||
if rate == 0 {
|
||||
return nil, fmt.Errorf("qdisc=tbf requires setting qdisc-tbf-rate")
|
||||
}
|
||||
|
||||
if burst == 0 {
|
||||
return nil, fmt.Errorf("qdisc=tbf requires setting qdisc-tbf-burst")
|
||||
}
|
||||
|
||||
if gsoEP, ok := lower.(stack.GSOEndpoint); ok {
|
||||
// HostGSOSupported endpoints can hand WritePacket a single GSO
|
||||
// super-packet up to GSOMaxSize+MaxHeaderLength bytes, so the bucket
|
||||
// must be able to hold one. GVisorGSOSupported segments above the
|
||||
// qdisc and GSONotSupported never produces packets above the link
|
||||
// MTU, both covered by the next check.
|
||||
maxGSOPktLen := gsoEP.GSOMaxSize() + uint32(lower.MaxHeaderLength())
|
||||
if gsoEP.SupportedGSO() == stack.HostGSOSupported && burst < uint32(maxGSOPktLen) {
|
||||
return nil, fmt.Errorf("burst (%d bytes) is smaller than link's max GSO packet size (%d bytes); either increase burst or disable host GSO via --gso=false", burst, maxGSOPktLen)
|
||||
}
|
||||
}
|
||||
|
||||
maxPktLen := lower.MTU() + uint32(lower.MaxHeaderLength())
|
||||
if burst < maxPktLen {
|
||||
return nil, fmt.Errorf("burst (%d bytes) is smaller than max packet length (%d bytes)", burst, maxPktLen)
|
||||
}
|
||||
|
||||
buffer := int64(len2TimeNS(rate, burst))
|
||||
if buffer == 0 {
|
||||
return nil, fmt.Errorf("rate (%d bytes/sec) is too high relative to burst (%d bytes); reduce qdisc-tbf-rate or increase qdisc-tbf-burst", rate, burst)
|
||||
}
|
||||
|
||||
d := &discipline{
|
||||
lower: lower,
|
||||
clock: clock,
|
||||
rate: rate,
|
||||
burst: burst,
|
||||
buffer: buffer,
|
||||
tokens: buffer,
|
||||
timeCheckpoint: clock.NowMonotonic(),
|
||||
}
|
||||
d.queue.Init(int(queueLen))
|
||||
d.wg.Add(1)
|
||||
go func() {
|
||||
defer d.wg.Done()
|
||||
d.dispatchLoop()
|
||||
}()
|
||||
return d, nil
|
||||
}
|
||||
|
||||
// WritePacket implements stack.QueueingDiscipline.WritePacket.
|
||||
func (d *discipline) WritePacket(pkt *stack.PacketBuffer) tcpip.Error {
|
||||
if d.closed.Load() == qDiscClosed {
|
||||
return &tcpip.ErrClosedForSend{}
|
||||
}
|
||||
|
||||
if uint32(pkt.Size()) > d.burst {
|
||||
// if the burst parameter is not smaller than the expected packet size,
|
||||
// oversize packets should be impossible with New's GSO check
|
||||
return &tcpip.ErrMessageTooLong{}
|
||||
}
|
||||
|
||||
d.mu.Lock()
|
||||
if d.closed.Load() == qDiscClosed {
|
||||
d.mu.Unlock()
|
||||
return &tcpip.ErrClosedForSend{}
|
||||
}
|
||||
haveSpace := d.queue.HasSpace()
|
||||
if haveSpace {
|
||||
d.queue.PushBack(pkt.IncRef())
|
||||
}
|
||||
d.mu.Unlock()
|
||||
if !haveSpace {
|
||||
return &tcpip.ErrNoBufferSpace{}
|
||||
}
|
||||
|
||||
d.newPacketWaker.Assert()
|
||||
return nil
|
||||
}
|
||||
|
||||
// Close implements stack.QueueingDiscipline.Close.
|
||||
func (d *discipline) Close() {
|
||||
d.closed.Store(qDiscClosed)
|
||||
d.closeWaker.Assert()
|
||||
d.wg.Wait()
|
||||
}
|
||||
59
pkg/tcpip/link/qdisc/tbf/tbf_state_autogen.go
Normal file
59
pkg/tcpip/link/qdisc/tbf/tbf_state_autogen.go
Normal file
|
|
@ -0,0 +1,59 @@
|
|||
// automatically generated by stateify.
|
||||
|
||||
package tbf
|
||||
|
||||
import (
|
||||
"context"
|
||||
|
||||
"github.com/sagernet/gvisor/pkg/state"
|
||||
)
|
||||
|
||||
func (d *discipline) StateTypeName() string {
|
||||
return "pkg/tcpip/link/qdisc/tbf.discipline"
|
||||
}
|
||||
|
||||
func (d *discipline) StateFields() []string {
|
||||
return []string{
|
||||
"lower",
|
||||
"rate",
|
||||
"burst",
|
||||
"buffer",
|
||||
"closed",
|
||||
"queue",
|
||||
"tokens",
|
||||
"timeCheckpoint",
|
||||
}
|
||||
}
|
||||
|
||||
func (d *discipline) beforeSave() {}
|
||||
|
||||
// +checklocksignore
|
||||
func (d *discipline) StateSave(stateSinkObject state.Sink) {
|
||||
d.beforeSave()
|
||||
stateSinkObject.Save(0, &d.lower)
|
||||
stateSinkObject.Save(1, &d.rate)
|
||||
stateSinkObject.Save(2, &d.burst)
|
||||
stateSinkObject.Save(3, &d.buffer)
|
||||
stateSinkObject.Save(4, &d.closed)
|
||||
stateSinkObject.Save(5, &d.queue)
|
||||
stateSinkObject.Save(6, &d.tokens)
|
||||
stateSinkObject.Save(7, &d.timeCheckpoint)
|
||||
}
|
||||
|
||||
func (d *discipline) afterLoad(context.Context) {}
|
||||
|
||||
// +checklocksignore
|
||||
func (d *discipline) StateLoad(ctx context.Context, stateSourceObject state.Source) {
|
||||
stateSourceObject.Load(0, &d.lower)
|
||||
stateSourceObject.Load(1, &d.rate)
|
||||
stateSourceObject.Load(2, &d.burst)
|
||||
stateSourceObject.Load(3, &d.buffer)
|
||||
stateSourceObject.Load(4, &d.closed)
|
||||
stateSourceObject.Load(5, &d.queue)
|
||||
stateSourceObject.Load(6, &d.tokens)
|
||||
stateSourceObject.Load(7, &d.timeCheckpoint)
|
||||
}
|
||||
|
||||
func init() {
|
||||
state.Register((*discipline)(nil))
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue