Separate lifecircle of handshake, ice, relay connections

- fix Stun, Turn address update thread safety issue
- move conn worker login into peer package
This commit is contained in:
Zoltán Papp
2024-06-17 17:52:22 +02:00
parent a7760bf0a7
commit e407fe02c5
11 changed files with 1139 additions and 980 deletions
+30 -159
View File
@@ -12,6 +12,7 @@ import (
"runtime"
"strings"
"sync"
"sync/atomic"
"time"
"github.com/pion/ice/v3"
@@ -95,7 +96,8 @@ type EngineConfig struct {
// Engine is a mechanism responsible for reacting on Signal and Management stream events and managing connections to the remote peers.
type Engine struct {
// signal is a Signal Service client
signal signal.Client
signal signal.Client
signaler *Signaler
// mgmClient is a Management Service client
mgmClient mgm.Client
// peerConns is a map that holds all the peers that are known to this peer
@@ -116,7 +118,8 @@ type Engine struct {
// STUNs is a list of STUN servers used by ICE
STUNs []*stun.URI
// TURNs is a list of STUN servers used by ICE
TURNs []*stun.URI
TURNs []*stun.URI
StunTurn atomic.Value
// clientRoutes is the most recent list of clientRoutes received from the Management Service
clientRoutes route.HAMap
@@ -154,8 +157,6 @@ type Engine struct {
relayProbe *Probe
wgProbe *Probe
wgConnWorker sync.WaitGroup
relayManager *relayClient.Manager
}
@@ -207,11 +208,11 @@ func NewEngineWithProbes(
relayProbe *Probe,
wgProbe *Probe,
) *Engine {
return &Engine{
clientCtx: clientCtx,
clientCancel: clientCancel,
signal: signalClient,
signaler: NewSignaler(signalClient, config.WgPrivateKey),
mgmClient: mgmClient,
relayManager: relayManager,
peerConns: make(map[string]*peer.Conn),
@@ -258,7 +259,6 @@ func (e *Engine) Stop() error {
time.Sleep(500 * time.Millisecond)
e.close()
e.wgConnWorker.Wait()
log.Infof("stopped Netbird Engine")
return nil
}
@@ -457,72 +457,11 @@ func (e *Engine) removePeer(peerKey string) error {
conn, exists := e.peerConns[peerKey]
if exists {
delete(e.peerConns, peerKey)
err := conn.Close()
if err != nil {
switch err.(type) {
case *peer.ConnectionAlreadyClosedError:
return nil
default:
return err
}
}
conn.Close()
}
return nil
}
func signalCandidate(candidate ice.Candidate, myKey wgtypes.Key, remoteKey wgtypes.Key, s signal.Client) error {
err := s.Send(&sProto.Message{
Key: myKey.PublicKey().String(),
RemoteKey: remoteKey.String(),
Body: &sProto.Body{
Type: sProto.Body_CANDIDATE,
Payload: candidate.Marshal(),
},
})
if err != nil {
return err
}
return nil
}
func sendSignal(message *sProto.Message, s signal.Client) error {
return s.Send(message)
}
// SignalOfferAnswer signals either an offer or an answer to remote peer
func SignalOfferAnswer(offerAnswer peer.OfferAnswer, myKey wgtypes.Key, remoteKey wgtypes.Key, s signal.Client,
isAnswer bool) error {
var t sProto.Body_Type
if isAnswer {
t = sProto.Body_ANSWER
} else {
t = sProto.Body_OFFER
}
msg, err := signal.MarshalCredential(
myKey,
offerAnswer.WgListenPort,
remoteKey, &signal.Credential{
UFrag: offerAnswer.IceCredentials.UFrag,
Pwd: offerAnswer.IceCredentials.Pwd,
},
t,
offerAnswer.RosenpassPubKey,
offerAnswer.RosenpassAddr,
offerAnswer.RelaySrvAddress)
if err != nil {
return err
}
err = s.Send(msg)
if err != nil {
return err
}
return nil
}
func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
e.syncMsgMux.Lock()
defer e.syncMsgMux.Unlock()
@@ -538,6 +477,11 @@ func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
return err
}
var stunTurn []*stun.URI
stunTurn = append(stunTurn, e.STUNs...)
stunTurn = append(stunTurn, e.TURNs...)
e.StunTurn.Store(stunTurn)
// todo update relay address in the relay manager
// todo update signal
@@ -893,57 +837,10 @@ func (e *Engine) addNewPeer(peerConfig *mgmProto.RemotePeerConfig) error {
if err != nil {
log.Warnf("error adding peer %s to status recorder, got error: %v", peerKey, err)
}
e.wgConnWorker.Add(1)
go e.connWorker(conn, peerKey)
}
return nil
}
func (e *Engine) connWorker(conn *peer.Conn, peerKey string) {
defer e.wgConnWorker.Done()
for {
// randomize starting time a bit
min := 500
max := 2000
duration := time.Duration(rand.Intn(max-min)+min) * time.Millisecond
select {
case <-e.ctx.Done():
return
case <-time.After(duration):
}
// if peer has been removed -> give up
if !e.peerExists(peerKey) {
log.Debugf("peer %s doesn't exist anymore, won't retry connection", peerKey)
return
}
if !e.signal.Ready() {
log.Infof("signal client isn't ready, skipping connection attempt %s", peerKey)
continue
}
// we might have received new STUN and TURN servers meanwhile, so update them
e.syncMsgMux.Lock()
conn.UpdateStunTurn(append(e.STUNs, e.TURNs...))
e.syncMsgMux.Unlock()
err := conn.Open(e.ctx)
if err != nil {
log.Debugf("connection to peer %s failed: %v", peerKey, err)
var connectionClosedError *peer.ConnectionClosedError
switch {
case errors.As(err, &connectionClosedError):
// conn has been forced to close, so we exit the loop
return
default:
}
}
}
}
func (e *Engine) peerExists(peerKey string) bool {
e.syncMsgMux.Lock()
defer e.syncMsgMux.Unlock()
@@ -953,9 +850,6 @@ func (e *Engine) peerExists(peerKey string) bool {
func (e *Engine) createPeerConn(pubKey string, allowedIPs string) (*peer.Conn, error) {
log.Debugf("creating peer connection %s", pubKey)
var stunTurn []*stun.URI
stunTurn = append(stunTurn, e.STUNs...)
stunTurn = append(stunTurn, e.TURNs...)
wgConfig := peer.WgConfig{
RemoteKey: pubKey,
@@ -988,52 +882,29 @@ func (e *Engine) createPeerConn(pubKey string, allowedIPs string) (*peer.Conn, e
// randomize connection timeout
timeout := time.Duration(rand.Intn(PeerConnectionTimeoutMax-PeerConnectionTimeoutMin)+PeerConnectionTimeoutMin) * time.Millisecond
config := peer.ConnConfig{
Key: pubKey,
LocalKey: e.config.WgPrivateKey.PublicKey().String(),
StunTurn: stunTurn,
InterfaceBlackList: e.config.IFaceBlackList,
DisableIPv6Discovery: e.config.DisableIPv6Discovery,
Timeout: timeout,
UDPMux: e.udpMux.UDPMuxDefault,
UDPMuxSrflx: e.udpMux,
WgConfig: wgConfig,
LocalWgPort: e.config.WgPort,
NATExternalIPs: e.parseNATExternalIPMappings(),
RosenpassPubKey: e.getRosenpassPubKey(),
RosenpassAddr: e.getRosenpassAddr(),
Key: pubKey,
LocalKey: e.config.WgPrivateKey.PublicKey().String(),
Timeout: timeout,
WgConfig: wgConfig,
LocalWgPort: e.config.WgPort,
RosenpassPubKey: e.getRosenpassPubKey(),
RosenpassAddr: e.getRosenpassAddr(),
ICEConfig: peer.ICEConfig{
StunTurn: e.StunTurn,
InterfaceBlackList: e.config.IFaceBlackList,
DisableIPv6Discovery: e.config.DisableIPv6Discovery,
UDPMux: e.udpMux.UDPMuxDefault,
UDPMuxSrflx: e.udpMux,
NATExternalIPs: e.parseNATExternalIPMappings(),
},
}
peerConn, err := peer.NewConn(config, e.statusRecorder, e.wgProxyFactory, e.mobileDep.TunAdapter, e.mobileDep.IFaceDiscover, e.relayManager)
peerConn, err := peer.NewConn(e.ctx, config, e.statusRecorder, e.wgProxyFactory, e.signaler, e.mobileDep.IFaceDiscover, e.relayManager)
if err != nil {
return nil, err
}
wgPubKey, err := wgtypes.ParseKey(pubKey)
if err != nil {
return nil, err
}
signalOffer := func(offerAnswer peer.OfferAnswer) error {
return SignalOfferAnswer(offerAnswer, e.config.WgPrivateKey, wgPubKey, e.signal, false)
}
signalCandidate := func(candidate ice.Candidate) error {
return signalCandidate(candidate, e.config.WgPrivateKey, wgPubKey, e.signal)
}
signalAnswer := func(offerAnswer peer.OfferAnswer) error {
return SignalOfferAnswer(offerAnswer, e.config.WgPrivateKey, wgPubKey, e.signal, true)
}
peerConn.SetSignalCandidate(signalCandidate)
peerConn.SetSignalOffer(signalOffer)
peerConn.SetSignalAnswer(signalAnswer)
peerConn.SetSendSignalMessage(func(message *sProto.Message) error {
return sendSignal(message, e.signal)
})
if e.rpManager != nil {
peerConn.SetOnConnected(e.rpManager.OnConnected)
peerConn.SetOnDisconnected(e.rpManager.OnDisconnected)
}
@@ -1107,7 +978,7 @@ func (e *Engine) receiveSignalEvents() {
return err
}
conn.OnRemoteCandidate(candidate, e.GetClientRoutes())
go conn.OnRemoteCandidate(candidate, e.GetClientRoutes())
case sProto.Body_MODE:
}
@@ -1402,7 +1273,7 @@ func (e *Engine) receiveProbeEvents() {
for _, peer := range e.peerConns {
key := peer.GetKey()
wgStats, err := peer.GetConf().WgConfig.WgInterface.GetStats(key)
wgStats, err := peer.WgConfig().WgInterface.GetStats(key)
if err != nil {
log.Debugf("failed to get wg stats for peer %s: %s", key, err)
}