wireguard-go/device/channels.go

138 lines
3.6 KiB
Go
Raw Permalink Normal View History

/* SPDX-License-Identifier: MIT
*
* Copyright (C) 2017-2023 WireGuard LLC. All Rights Reserved.
*/
package device
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
import (
"runtime"
"sync"
)
// An outboundQueue is a channel of QueueOutboundElements awaiting encryption.
// An outboundQueue is ref-counted using its wg field.
// An outboundQueue created with newOutboundQueue has one reference.
// Every additional writer must call wg.Add(1).
// Every completed writer must call wg.Done().
// When no further writers will be added,
// call wg.Done to remove the initial reference.
// When the refcount hits 0, the queue's channel is closed.
type outboundQueue struct {
c chan *QueueOutboundElementsContainer
wg sync.WaitGroup
}
func newOutboundQueue() *outboundQueue {
q := &outboundQueue{
c: make(chan *QueueOutboundElementsContainer, QueueOutboundSize),
}
q.wg.Add(1)
go func() {
q.wg.Wait()
close(q.c)
}()
return q
}
// A inboundQueue is similar to an outboundQueue; see those docs.
type inboundQueue struct {
c chan *QueueInboundElementsContainer
wg sync.WaitGroup
}
func newInboundQueue() *inboundQueue {
q := &inboundQueue{
c: make(chan *QueueInboundElementsContainer, QueueInboundSize),
}
q.wg.Add(1)
go func() {
q.wg.Wait()
close(q.c)
}()
return q
}
// A handshakeQueue is similar to an outboundQueue; see those docs.
type handshakeQueue struct {
c chan QueueHandshakeElement
wg sync.WaitGroup
}
func newHandshakeQueue() *handshakeQueue {
q := &handshakeQueue{
c: make(chan QueueHandshakeElement, QueueHandshakeSize),
}
q.wg.Add(1)
go func() {
q.wg.Wait()
close(q.c)
}()
return q
}
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
type autodrainingInboundQueue struct {
c chan *QueueInboundElementsContainer
}
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
// newAutodrainingInboundQueue returns a channel that will be drained when it gets GC'd.
// It is useful in cases in which is it hard to manage the lifetime of the channel.
// The returned channel must not be closed. Senders should signal shutdown using
// some other means, such as sending a sentinel nil values.
func newAutodrainingInboundQueue(device *Device) *autodrainingInboundQueue {
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
q := &autodrainingInboundQueue{
c: make(chan *QueueInboundElementsContainer, QueueInboundSize),
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
}
runtime.SetFinalizer(q, device.flushInboundQueue)
return q
}
func (device *Device) flushInboundQueue(q *autodrainingInboundQueue) {
for {
select {
case elemsContainer := <-q.c:
elemsContainer.Lock()
for _, elem := range elemsContainer.elems {
device.PutMessageBuffer(elem.buffer)
device.PutInboundElement(elem)
}
device.PutInboundElementsContainer(elemsContainer)
default:
return
}
}
}
type autodrainingOutboundQueue struct {
c chan *QueueOutboundElementsContainer
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
}
// newAutodrainingOutboundQueue returns a channel that will be drained when it gets GC'd.
// It is useful in cases in which is it hard to manage the lifetime of the channel.
// The returned channel must not be closed. Senders should signal shutdown using
// some other means, such as sending a sentinel nil values.
// All sends to the channel must be best-effort, because there may be no receivers.
func newAutodrainingOutboundQueue(device *Device) *autodrainingOutboundQueue {
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
q := &autodrainingOutboundQueue{
c: make(chan *QueueOutboundElementsContainer, QueueOutboundSize),
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
}
runtime.SetFinalizer(q, device.flushOutboundQueue)
return q
device: remove mutex from Peer send/receive The immediate motivation for this change is an observed deadlock. 1. A goroutine calls peer.Stop. That calls peer.queue.Lock(). 2. Another goroutine is in RoutineSequentialReceiver. It receives an elem from peer.queue.inbound. 3. The peer.Stop goroutine calls close(peer.queue.inbound), close(peer.queue.outbound), and peer.stopping.Wait(). It blocks waiting for RoutineSequentialReceiver and RoutineSequentialSender to exit. 4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets(). SendStagedPackets attempts peer.queue.RLock(). That blocks forever because the peer.Stop goroutine holds a write lock on that mutex. A background motivation for this change is that it can be expensive to have a mutex in the hot code path of RoutineSequential*. The mutex was necessary to avoid attempting to send elems on a closed channel. This commit removes that danger by never closing the channel. Instead, we send a sentinel nil value on the channel to indicate to the receiver that it should exit. The only problem with this is that if the receiver exits, we could write an elem into the channel which would never get received. If it never gets received, it cannot get returned to the device pools. To work around this, we use a finalizer. When the channel can be GC'd, the finalizer drains any remaining elements from the channel and restores them to the device pool. After that change, peer.queue.RWMutex no longer makes sense where it is. It is only used to prevent concurrent calls to Start and Stop. Move it to a more sensible location and make it a plain sync.Mutex. Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
2021-02-08 22:02:52 +01:00
}
func (device *Device) flushOutboundQueue(q *autodrainingOutboundQueue) {
for {
select {
case elemsContainer := <-q.c:
elemsContainer.Lock()
for _, elem := range elemsContainer.elems {
device.PutMessageBuffer(elem.buffer)
device.PutOutboundElement(elem)
}
device.PutOutboundElementsContainer(elemsContainer)
default:
return
}
}
}