mirror of
https://github.com/lightningnetwork/lnd.git
synced 2024-11-19 09:53:54 +01:00
peer: ensure the peer has been started before Disconnect can be called
In this commit, we make sure that all the `wg.Add(1)` calls succeed before we attempt to wait on the shutdown of all the goroutines. Under rare scheduling scenarios, if both `Start` and `Disconnect` are called concurrently, then this internal race error can be hit, causing the panic to occur. Fixes https://github.com/lightningnetwork/lnd/issues/7853
This commit is contained in:
parent
798821db17
commit
325e844a96
@ -194,6 +194,10 @@ unlock or create.
|
||||
Tor](https://github.com/lightningnetwork/lnd/pull/7783), even when
|
||||
`tor.skip-proxy-for-clearnet-targets=true` was set.
|
||||
|
||||
* Fix a [concurrency issue related to rapid peer teardown and
|
||||
creation](https://github.com/lightningnetwork/lnd/pull/7856) that can arise
|
||||
under rare scenarios.
|
||||
|
||||
### Tooling and documentation
|
||||
|
||||
* Add support for [custom `RPCHOST` and
|
||||
@ -226,6 +230,7 @@ unlock or create.
|
||||
* Maxwell Sayles
|
||||
* Michael Street
|
||||
* MG-ng
|
||||
* Olaoluwa Osuntokun
|
||||
* Oliver Gugger
|
||||
* Pierre Beugnet
|
||||
* Satarupa Deb
|
||||
|
@ -469,8 +469,9 @@ type Brontide struct {
|
||||
// potentially holding lots of un-consumed events.
|
||||
channelEventClient *subscribe.Client
|
||||
|
||||
quit chan struct{}
|
||||
wg sync.WaitGroup
|
||||
startReady chan struct{}
|
||||
quit chan struct{}
|
||||
wg sync.WaitGroup
|
||||
|
||||
// log is a peer-specific logging instance.
|
||||
log btclog.Logger
|
||||
@ -500,6 +501,7 @@ func NewBrontide(cfg Config) *Brontide {
|
||||
linkFailures: make(chan linkFailureReport),
|
||||
chanCloseMsgs: make(chan *closeMsg),
|
||||
resentChanSyncMsg: make(map[lnwire.ChannelID]struct{}),
|
||||
startReady: make(chan struct{}),
|
||||
quit: make(chan struct{}),
|
||||
log: build.NewPrefixLog(logPrefix, peerLog),
|
||||
}
|
||||
@ -514,6 +516,11 @@ func (p *Brontide) Start() error {
|
||||
return nil
|
||||
}
|
||||
|
||||
// Once we've finished starting up the peer, we'll signal to other
|
||||
// goroutines that the they can move forward to tear down the peer, or
|
||||
// carry out other relevant changes.
|
||||
defer close(p.startReady)
|
||||
|
||||
p.log.Tracef("starting with conn[%v->%v]",
|
||||
p.cfg.Conn.LocalAddr(), p.cfg.Conn.RemoteAddr())
|
||||
|
||||
@ -1050,6 +1057,14 @@ func (p *Brontide) maybeSendNodeAnn(channels []*channeldb.OpenChannel) {
|
||||
// calling Disconnect will signal the quit channel and the method will not
|
||||
// block, since no goroutines were spawned.
|
||||
func (p *Brontide) WaitForDisconnect(ready chan struct{}) {
|
||||
// Before we try to call the `Wait` goroutine, we'll make sure the main
|
||||
// set of goroutines are already active.
|
||||
select {
|
||||
case <-p.startReady:
|
||||
case <-p.quit:
|
||||
return
|
||||
}
|
||||
|
||||
select {
|
||||
case <-ready:
|
||||
case <-p.quit:
|
||||
@ -1066,6 +1081,14 @@ func (p *Brontide) Disconnect(reason error) {
|
||||
return
|
||||
}
|
||||
|
||||
// Make sure initialization has completed before we try to tear things
|
||||
// down.
|
||||
select {
|
||||
case <-p.startReady:
|
||||
case <-p.quit:
|
||||
return
|
||||
}
|
||||
|
||||
err := fmt.Errorf("disconnecting %s, reason: %v", p, reason)
|
||||
p.storeError(err)
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user