[client] Fix/ice handshake (#4281)

In this PR, speed up the GRPC message processing, force the recreation of the ICE agent when getting a new, remote offer (do not wait for local STUN timeout).
2026-04-16 07:16:38 +00:00 · 2025-08-18 20:09:50 +02:00
parent 6a3846a8b7
commit 12cad854b2
13 changed files with 563 additions and 357 deletions
--- a/shared/signal/client/client.go
+++ b/shared/signal/client/client.go
@@ -52,7 +52,7 @@ func UnMarshalCredential(msg *proto.Message) (*Credential, error) {
 }

 // MarshalCredential marshal a Credential instance and returns a Message object
-func MarshalCredential(myKey wgtypes.Key, myPort int, remoteKey string, credential *Credential, t proto.Body_Type, rosenpassPubKey []byte, rosenpassAddr string, relaySrvAddress string) (*proto.Message, error) {
+func MarshalCredential(myKey wgtypes.Key, myPort int, remoteKey string, credential *Credential, t proto.Body_Type, rosenpassPubKey []byte, rosenpassAddr string, relaySrvAddress string, sessionID []byte) (*proto.Message, error) {
 	return &proto.Message{
 		Key:       myKey.PublicKey().String(),
 		RemoteKey: remoteKey,
@@ -66,6 +66,7 @@ func MarshalCredential(myKey wgtypes.Key, myPort int, remoteKey string, credenti
 				RosenpassServerAddr: rosenpassAddr,
 			},
 			RelayServerAddress: relaySrvAddress,
+			SessionId:          sessionID,
 		},
 	}, nil
 }
--- a/shared/signal/client/grpc.go
+++ b/shared/signal/client/grpc.go
@@ -45,19 +45,10 @@ type GrpcClient struct {
 	connStateCallbackLock sync.RWMutex

 	onReconnectedListenerFn func()
-}

-func (c *GrpcClient) StreamConnected() bool {
-	return c.status == StreamConnected
-}
-
-func (c *GrpcClient) GetStatus() Status {
-	return c.status
-}
-
-// Close Closes underlying connections to the Signal Exchange
-func (c *GrpcClient) Close() error {
-	return c.signalConn.Close()
+	decryptionWorker       *Worker
+	decryptionWorkerCancel context.CancelFunc
+	decryptionWg           sync.WaitGroup
 }

 // NewClient creates a new Signal client
@@ -93,6 +84,25 @@ func NewClient(ctx context.Context, addr string, key wgtypes.Key, tlsEnabled boo
 	}, nil
 }

+func (c *GrpcClient) StreamConnected() bool {
+	return c.status == StreamConnected
+}
+
+func (c *GrpcClient) GetStatus() Status {
+	return c.status
+}
+
+// Close Closes underlying connections to the Signal Exchange
+func (c *GrpcClient) Close() error {
+	if c.decryptionWorkerCancel != nil {
+		c.decryptionWorkerCancel()
+	}
+	c.decryptionWg.Wait()
+	c.decryptionWorker = nil
+
+	return c.signalConn.Close()
+}
+
 // SetConnStateListener set the ConnStateNotifier
 func (c *GrpcClient) SetConnStateListener(notifier ConnStateNotifier) {
 	c.connStateCallbackLock.Lock()
@@ -148,8 +158,12 @@ func (c *GrpcClient) Receive(ctx context.Context, msgHandler func(msg *proto.Mes

 		log.Infof("connected to the Signal Service stream")
 		c.notifyConnected()
+
+		// Start worker pool if not already started
+		c.startEncryptionWorker(msgHandler)
+
 		// start receiving messages from the Signal stream (from other peers through signal)
-		err = c.receive(stream, msgHandler)
+		err = c.receive(stream)
 		if err != nil {
 			if s, ok := status.FromError(err); ok && s.Code() == codes.Canceled {
 				log.Debugf("signal connection context has been canceled, this usually indicates shutdown")
@@ -174,6 +188,7 @@ func (c *GrpcClient) Receive(ctx context.Context, msgHandler func(msg *proto.Mes

 	return nil
 }
+
 func (c *GrpcClient) notifyStreamDisconnected() {
 	c.mux.Lock()
 	defer c.mux.Unlock()
@@ -382,11 +397,11 @@ func (c *GrpcClient) Send(msg *proto.Message) error {
 }

 // receive receives messages from other peers coming through the Signal Exchange
-func (c *GrpcClient) receive(stream proto.SignalExchange_ConnectStreamClient,
-	msgHandler func(msg *proto.Message) error) error {
-
+// and distributes them to worker threads for processing
+func (c *GrpcClient) receive(stream proto.SignalExchange_ConnectStreamClient) error {
 	for {
 		msg, err := stream.Recv()
+		// Handle errors immediately
 		switch s, ok := status.FromError(err); {
 		case ok && s.Code() == codes.Canceled:
 			log.Debugf("stream canceled (usually indicates shutdown)")
@@ -398,24 +413,37 @@ func (c *GrpcClient) receive(stream proto.SignalExchange_ConnectStreamClient,
 			log.Debugf("Signal Service stream closed by server")
 			return err
 		case err != nil:
+			log.Errorf("Stream receive error: %v", err)
 			return err
 		}
-		log.Tracef("received a new message from Peer [fingerprint: %s]", msg.Key)

-		decryptedMessage, err := c.decryptMessage(msg)
-		if err != nil {
-			log.Errorf("failed decrypting message of Peer [key: %s] error: [%s]", msg.Key, err.Error())
+		if msg == nil {
+			continue
 		}

-		err = msgHandler(decryptedMessage)
-
-		if err != nil {
-			log.Errorf("error while handling message of Peer [key: %s] error: [%s]", msg.Key, err.Error())
-			// todo send something??
+		if err := c.decryptionWorker.AddMsg(c.ctx, msg); err != nil {
+			log.Errorf("failed to add message to decryption worker: %v", err)
 		}
 	}
 }

+func (c *GrpcClient) startEncryptionWorker(handler func(msg *proto.Message) error) {
+	if c.decryptionWorker != nil {
+		return
+	}
+
+	c.decryptionWorker = NewWorker(c.decryptMessage, handler)
+	workerCtx, workerCancel := context.WithCancel(context.Background())
+	c.decryptionWorkerCancel = workerCancel
+
+	c.decryptionWg.Add(1)
+	go func() {
+		defer workerCancel()
+		c.decryptionWorker.Work(workerCtx)
+		c.decryptionWg.Done()
+	}()
+}
+
 func (c *GrpcClient) notifyDisconnected(err error) {
 	c.connStateCallbackLock.RLock()
 	defer c.connStateCallbackLock.RUnlock()
--- a/shared/signal/client/worker.go
+++ b/shared/signal/client/worker.go
@@ -0,0 +1,55 @@
+package client
+
+import (
+	"context"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/shared/signal/proto"
+)
+
+type Worker struct {
+	decryptMessage func(msg *proto.EncryptedMessage) (*proto.Message, error)
+	handler        func(msg *proto.Message) error
+
+	encryptedMsgPool chan *proto.EncryptedMessage
+}
+
+func NewWorker(decryptFn func(msg *proto.EncryptedMessage) (*proto.Message, error), handlerFn func(msg *proto.Message) error) *Worker {
+	return &Worker{
+		decryptMessage:   decryptFn,
+		handler:          handlerFn,
+		encryptedMsgPool: make(chan *proto.EncryptedMessage, 1),
+	}
+}
+
+func (w *Worker) AddMsg(ctx context.Context, msg *proto.EncryptedMessage) error {
+	// this is blocker because do not want to drop messages here
+	select {
+	case w.encryptedMsgPool <- msg:
+	case <-ctx.Done():
+	}
+	return nil
+}
+
+func (w *Worker) Work(ctx context.Context) {
+	for {
+		select {
+		case msg := <-w.encryptedMsgPool:
+			decryptedMessage, err := w.decryptMessage(msg)
+			if err != nil {
+				log.Errorf("failed to decrypt message: %v", err)
+				continue
+			}
+
+			if err := w.handler(decryptedMessage); err != nil {
+				log.Errorf("failed to handle message: %v", err)
+				continue
+			}
+
+		case <-ctx.Done():
+			log.Infof("Message worker stopping due to context cancellation")
+			return
+		}
+	}
+}
--- a/shared/signal/proto/signalexchange.pb.go
+++ b/shared/signal/proto/signalexchange.pb.go
@@ -230,6 +230,7 @@ type Body struct {
 	RosenpassConfig *RosenpassConfig `protobuf:"bytes,7,opt,name=rosenpassConfig,proto3" json:"rosenpassConfig,omitempty"`
 	// relayServerAddress is url of the relay server
 	RelayServerAddress string `protobuf:"bytes,8,opt,name=relayServerAddress,proto3" json:"relayServerAddress,omitempty"`
+	SessionId          []byte `protobuf:"bytes,10,opt,name=sessionId,proto3,oneof" json:"sessionId,omitempty"`
 }

 func (x *Body) Reset() {
@@ -320,6 +321,13 @@ func (x *Body) GetRelayServerAddress() string {
 	return ""
 }

+func (x *Body) GetSessionId() []byte {
+	if x != nil {
+		return x.SessionId
+	}
+	return nil
+}
+
 // Mode indicates a connection mode
 type Mode struct {
 	state         protoimpl.MessageState
@@ -443,7 +451,7 @@ var file_signalexchange_proto_rawDesc = []byte{
 	0x52, 0x09, 0x72, 0x65, 0x6d, 0x6f, 0x74, 0x65, 0x4b, 0x65, 0x79, 0x12, 0x28, 0x0a, 0x04, 0x62,
 	0x6f, 0x64, 0x79, 0x18, 0x04, 0x20, 0x01, 0x28, 0x0b, 0x32, 0x14, 0x2e, 0x73, 0x69, 0x67, 0x6e,
 	0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x42, 0x6f, 0x64, 0x79, 0x52,
-	0x04, 0x62, 0x6f, 0x64, 0x79, 0x22, 0xb3, 0x03, 0x0a, 0x04, 0x42, 0x6f, 0x64, 0x79, 0x12, 0x2d,
+	0x04, 0x62, 0x6f, 0x64, 0x79, 0x22, 0xe4, 0x03, 0x0a, 0x04, 0x42, 0x6f, 0x64, 0x79, 0x12, 0x2d,
 	0x0a, 0x04, 0x74, 0x79, 0x70, 0x65, 0x18, 0x01, 0x20, 0x01, 0x28, 0x0e, 0x32, 0x19, 0x2e, 0x73,
 	0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x42, 0x6f,
 	0x64, 0x79, 0x2e, 0x54, 0x79, 0x70, 0x65, 0x52, 0x04, 0x74, 0x79, 0x70, 0x65, 0x12, 0x18, 0x0a,
@@ -466,34 +474,37 @@ var file_signalexchange_proto_rawDesc = []byte{
 	0x43, 0x6f, 0x6e, 0x66, 0x69, 0x67, 0x12, 0x2e, 0x0a, 0x12, 0x72, 0x65, 0x6c, 0x61, 0x79, 0x53,
 	0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72, 0x65, 0x73, 0x73, 0x18, 0x08, 0x20, 0x01,
 	0x28, 0x09, 0x52, 0x12, 0x72, 0x65, 0x6c, 0x61, 0x79, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x41,
-	0x64, 0x64, 0x72, 0x65, 0x73, 0x73, 0x22, 0x43, 0x0a, 0x04, 0x54, 0x79, 0x70, 0x65, 0x12, 0x09,
-	0x0a, 0x05, 0x4f, 0x46, 0x46, 0x45, 0x52, 0x10, 0x00, 0x12, 0x0a, 0x0a, 0x06, 0x41, 0x4e, 0x53,
-	0x57, 0x45, 0x52, 0x10, 0x01, 0x12, 0x0d, 0x0a, 0x09, 0x43, 0x41, 0x4e, 0x44, 0x49, 0x44, 0x41,
-	0x54, 0x45, 0x10, 0x02, 0x12, 0x08, 0x0a, 0x04, 0x4d, 0x4f, 0x44, 0x45, 0x10, 0x04, 0x12, 0x0b,
-	0x0a, 0x07, 0x47, 0x4f, 0x5f, 0x49, 0x44, 0x4c, 0x45, 0x10, 0x05, 0x22, 0x2e, 0x0a, 0x04, 0x4d,
-	0x6f, 0x64, 0x65, 0x12, 0x1b, 0x0a, 0x06, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x18, 0x01, 0x20,
-	0x01, 0x28, 0x08, 0x48, 0x00, 0x52, 0x06, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x88, 0x01, 0x01,
-	0x42, 0x09, 0x0a, 0x07, 0x5f, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x22, 0x6d, 0x0a, 0x0f, 0x52,
-	0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x43, 0x6f, 0x6e, 0x66, 0x69, 0x67, 0x12, 0x28,
-	0x0a, 0x0f, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x50, 0x75, 0x62, 0x4b, 0x65,
-	0x79, 0x18, 0x01, 0x20, 0x01, 0x28, 0x0c, 0x52, 0x0f, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61,
-	0x73, 0x73, 0x50, 0x75, 0x62, 0x4b, 0x65, 0x79, 0x12, 0x30, 0x0a, 0x13, 0x72, 0x6f, 0x73, 0x65,
-	0x6e, 0x70, 0x61, 0x73, 0x73, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72, 0x18,
-	0x02, 0x20, 0x01, 0x28, 0x09, 0x52, 0x13, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73,
-	0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72, 0x32, 0xb9, 0x01, 0x0a, 0x0e, 0x53,
-	0x69, 0x67, 0x6e, 0x61, 0x6c, 0x45, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x12, 0x4c, 0x0a,
-	0x04, 0x53, 0x65, 0x6e, 0x64, 0x12, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78,
-	0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64,
-	0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x1a, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c,
-	0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74,
-	0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x22, 0x00, 0x12, 0x59, 0x0a, 0x0d, 0x43,
-	0x6f, 0x6e, 0x6e, 0x65, 0x63, 0x74, 0x53, 0x74, 0x72, 0x65, 0x61, 0x6d, 0x12, 0x20, 0x2e, 0x73,
-	0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e,
-	0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x1a, 0x20,
-	0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e,
-	0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65,
-	0x22, 0x00, 0x28, 0x01, 0x30, 0x01, 0x42, 0x08, 0x5a, 0x06, 0x2f, 0x70, 0x72, 0x6f, 0x74, 0x6f,
-	0x62, 0x06, 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x33,
+	0x64, 0x64, 0x72, 0x65, 0x73, 0x73, 0x12, 0x21, 0x0a, 0x09, 0x73, 0x65, 0x73, 0x73, 0x69, 0x6f,
+	0x6e, 0x49, 0x64, 0x18, 0x0a, 0x20, 0x01, 0x28, 0x0c, 0x48, 0x00, 0x52, 0x09, 0x73, 0x65, 0x73,
+	0x73, 0x69, 0x6f, 0x6e, 0x49, 0x64, 0x88, 0x01, 0x01, 0x22, 0x43, 0x0a, 0x04, 0x54, 0x79, 0x70,
+	0x65, 0x12, 0x09, 0x0a, 0x05, 0x4f, 0x46, 0x46, 0x45, 0x52, 0x10, 0x00, 0x12, 0x0a, 0x0a, 0x06,
+	0x41, 0x4e, 0x53, 0x57, 0x45, 0x52, 0x10, 0x01, 0x12, 0x0d, 0x0a, 0x09, 0x43, 0x41, 0x4e, 0x44,
+	0x49, 0x44, 0x41, 0x54, 0x45, 0x10, 0x02, 0x12, 0x08, 0x0a, 0x04, 0x4d, 0x4f, 0x44, 0x45, 0x10,
+	0x04, 0x12, 0x0b, 0x0a, 0x07, 0x47, 0x4f, 0x5f, 0x49, 0x44, 0x4c, 0x45, 0x10, 0x05, 0x42, 0x0c,
+	0x0a, 0x0a, 0x5f, 0x73, 0x65, 0x73, 0x73, 0x69, 0x6f, 0x6e, 0x49, 0x64, 0x22, 0x2e, 0x0a, 0x04,
+	0x4d, 0x6f, 0x64, 0x65, 0x12, 0x1b, 0x0a, 0x06, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x18, 0x01,
+	0x20, 0x01, 0x28, 0x08, 0x48, 0x00, 0x52, 0x06, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x88, 0x01,
+	0x01, 0x42, 0x09, 0x0a, 0x07, 0x5f, 0x64, 0x69, 0x72, 0x65, 0x63, 0x74, 0x22, 0x6d, 0x0a, 0x0f,
+	0x52, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x43, 0x6f, 0x6e, 0x66, 0x69, 0x67, 0x12,
+	0x28, 0x0a, 0x0f, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x50, 0x75, 0x62, 0x4b,
+	0x65, 0x79, 0x18, 0x01, 0x20, 0x01, 0x28, 0x0c, 0x52, 0x0f, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70,
+	0x61, 0x73, 0x73, 0x50, 0x75, 0x62, 0x4b, 0x65, 0x79, 0x12, 0x30, 0x0a, 0x13, 0x72, 0x6f, 0x73,
+	0x65, 0x6e, 0x70, 0x61, 0x73, 0x73, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72,
+	0x18, 0x02, 0x20, 0x01, 0x28, 0x09, 0x52, 0x13, 0x72, 0x6f, 0x73, 0x65, 0x6e, 0x70, 0x61, 0x73,
+	0x73, 0x53, 0x65, 0x72, 0x76, 0x65, 0x72, 0x41, 0x64, 0x64, 0x72, 0x32, 0xb9, 0x01, 0x0a, 0x0e,
+	0x53, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x45, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x12, 0x4c,
+	0x0a, 0x04, 0x53, 0x65, 0x6e, 0x64, 0x12, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65,
+	0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65,
+	0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x1a, 0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61,
+	0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70,
+	0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x22, 0x00, 0x12, 0x59, 0x0a, 0x0d,
+	0x43, 0x6f, 0x6e, 0x6e, 0x65, 0x63, 0x74, 0x53, 0x74, 0x72, 0x65, 0x61, 0x6d, 0x12, 0x20, 0x2e,
+	0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65, 0x2e, 0x45,
+	0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x1a,
+	0x20, 0x2e, 0x73, 0x69, 0x67, 0x6e, 0x61, 0x6c, 0x65, 0x78, 0x63, 0x68, 0x61, 0x6e, 0x67, 0x65,
+	0x2e, 0x45, 0x6e, 0x63, 0x72, 0x79, 0x70, 0x74, 0x65, 0x64, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67,
+	0x65, 0x22, 0x00, 0x28, 0x01, 0x30, 0x01, 0x42, 0x08, 0x5a, 0x06, 0x2f, 0x70, 0x72, 0x6f, 0x74,
+	0x6f, 0x62, 0x06, 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x33,
 }

 var (
@@ -601,6 +612,7 @@ func file_signalexchange_proto_init() {
 			}
 		}
 	}
+	file_signalexchange_proto_msgTypes[2].OneofWrappers = []interface{}{}
 	file_signalexchange_proto_msgTypes[3].OneofWrappers = []interface{}{}
 	type x struct{}
 	out := protoimpl.TypeBuilder{
--- a/shared/signal/proto/signalexchange.proto
+++ b/shared/signal/proto/signalexchange.proto
@@ -64,6 +64,8 @@ message Body {

  // relayServerAddress is url of the relay server
  string relayServerAddress = 8;
+
+  optional bytes sessionId = 10;
 }

 // Mode indicates a connection mode