fix: close stale relay connection on ErrConnAlreadyExists to recover …

client/internal/peer/conn.go

@@ -617,9 +617,31 @@ func (conn *Conn) onWGDisconnected() {
 	// Close the active connection based on current priority
 	switch conn.currentConnPriority {
 	case conntype.Relay:
+		// Mark the relay conn entry as stale so the next OnNewOffer closes
+		// and reopens it instead of reusing a dead pipe. MarkStale covers
+		// the case where CloseConn is a no-op (e.g. relayedConn already nil).
+		conn.workerRelay.MarkStale()
 		conn.workerRelay.CloseConn()
 		conn.handleRelayDisconnectedLocked()
+		// When running over relay, workerICE is not closed so its session ID is
+		// never rotated. The next offer would carry the same session ID, causing
+		// the remote peer to skip ICE agent recreation (it already has an agent
+		// for that session) and reuse stale candidates — preventing recovery
+		// after a NAT IP change (e.g. PPPoE reconnect). Force a new session ID
+		// so the remote peer creates a fresh ICE agent with current candidates.
+		if conn.workerICE != nil {
+			conn.workerICE.ResetSessionID()
+		}
 	case conntype.ICEP2P, conntype.ICETurn:
+		// WorkerICE.Close() sets agent=nil before pion's ICE library fires
+		// ConnectionStateClosed. By the time onConnectionStateChange runs
+		// closeAgent(), the w.agent==agent guard fails and the session ID
+		// is not rotated. Without rotation, the next offer carries the same
+		// session ID and the remote peer skips ICE agent recreation in
+		// OnNewOffer (sessionID match), reusing stale candidates from the
+		// previous network state. Rotate explicitly here so the remote peer
+		// always recreates its agent after a WG timeout on ICE.
+		conn.workerICE.ResetSessionID()
 		conn.workerICE.Close()
 	default:
 		conn.Log.Debugf("No active connection to close on WG timeout")

client/internal/peer/worker_ice.go

@@ -249,6 +249,24 @@ func (w *WorkerICE) SessionID() ICESessionID {
 	return w.sessionID
 }
 
+// ResetSessionID generates a new session ID and clears the remote session ID.
+// This must be called when the WireGuard handshake times out while using a relay
+// connection, so that the next ICE offer carries a fresh session ID and the remote
+// peer recreates its ICE agent with up-to-date candidates instead of skipping the
+// offer because the session ID matches the previous (failed) attempt.
+func (w *WorkerICE) ResetSessionID() {
+	w.muxAgent.Lock()
+	defer w.muxAgent.Unlock()
+
+	sessionID, err := NewICESessionID()
+	if err != nil {
+		w.log.Errorf("failed to create new session ID: %s", err)
+		return
+	}
+	w.sessionID = sessionID
+	w.remoteSessionID = ""
+}
+
 // will block until connection succeeded
 // but it won't release if ICE Agent went into Disconnected or Failed state,
 // so we have to cancel it with the provided context once agent detected a broken connection

client/internal/peer/worker_relay.go

@@ -30,6 +30,14 @@ type WorkerRelay struct {
 	relayLock   sync.Mutex
 
 	relaySupportedOnRemotePeer atomic.Bool
+
+	// relayConnStale is set to true when an event indicates that the current
+	// relay connection entry in the relay client's conns map is no longer
+	// backed by a live peer session (e.g. local WG handshake timeout, relay
+	// server close event, explicit CloseConn). When OnNewOffer observes
+	// ErrConnAlreadyExists, it only closes the stale entry if this flag is
+	// set; otherwise it bails out and reuses the existing healthy connection.
+	relayConnStale atomic.Bool
 }
 
 func NewWorkerRelay(ctx context.Context, log *log.Entry, ctrl bool, config ConnConfig, conn *Conn, relayManager *relayClient.Manager) *WorkerRelay {
@@ -64,11 +72,28 @@ func (w *WorkerRelay) OnNewOffer(remoteOfferAnswer *OfferAnswer) {
 	relayedConn, err := w.relayManager.OpenConn(w.peerCtx, srv, w.config.Key)
 	if err != nil {
 		if errors.Is(err, relayClient.ErrConnAlreadyExists) {
-			w.log.Debugf("handled offer by reusing existing relay connection")
+			// Only tear down the existing conn if something previously marked
+			// it as stale (local WG handshake timeout, relay server close, or
+			// explicit CloseConn). Without that signal, the existing conn is
+			// assumed healthy and is reused — unconditional close on every
+			// colliding offer causes an infinite tear-down/rebuild loop when
+			// the remote peer sends rapid successive offers.
+			if !w.relayConnStale.Load() {
+				w.log.Debugf("relay conn already exists and is not marked stale, reusing")
+				return
+			}
+			w.log.Infof("relay conn already exists and is marked stale, closing and retrying")
+			w.relayManager.CloseConnByPeerKey(srv, w.config.Key)
+			relayedConn, err = w.relayManager.OpenConn(w.peerCtx, srv, w.config.Key)
+			if err != nil {
+				w.log.Errorf("failed to reopen connection via Relay after closing stale: %s", err)
+				return
+			}
+			w.relayConnStale.Store(false)
+		} else {
+			w.log.Errorf("failed to open connection via Relay: %s", err)
 			return
 		}
-		w.log.Errorf("failed to open connection via Relay: %s", err)
-		return
 	}
 
 	w.relayLock.Lock()
@@ -109,11 +134,21 @@ func (w *WorkerRelay) CloseConn() {
 		return
 	}
 
+	w.relayConnStale.Store(true)
 	if err := w.relayedConn.Close(); err != nil {
 		w.log.Warnf("failed to close relay connection: %v", err)
 	}
 }
 
+// MarkStale marks the relay connection entry as stale so that the next
+// OnNewOffer call with ErrConnAlreadyExists will tear it down and open a
+// fresh one. Callers signal staleness from outside the relay client path,
+// e.g. when the local WG handshake watcher fires while the relay is the
+// active transport.
+func (w *WorkerRelay) MarkStale() {
+	w.relayConnStale.Store(true)
+}
+
 func (w *WorkerRelay) isRelaySupported(answer *OfferAnswer) bool {
 	if !w.relayManager.HasRelayAddress() {
 		return false
@@ -129,5 +164,6 @@ func (w *WorkerRelay) preferredRelayServer(myRelayAddress, remoteRelayAddress st
 }
 
 func (w *WorkerRelay) onRelayClientDisconnected() {
+	w.relayConnStale.Store(true)
 	go w.conn.onRelayDisconnected()
 }

shared/relay/client/client.go

@@ -627,6 +627,27 @@ func (c *Client) closeConn(containerRef *connContainer, id messages.PeerID) erro
 	return nil
 }
 
+// CloseConnByPeerKey closes an existing relay connection for the given peer key,
+// removing it from the internal connection map so a new one can be opened.
+func (c *Client) CloseConnByPeerKey(peerKey string) {
+	peerID := messages.HashID(peerKey)
+	c.mu.Lock()
+	container, ok := c.conns[peerID]
+	if !ok {
+		c.mu.Unlock()
+		return
+	}
+
+	c.log.Infof("force closing stale relay connection for peer: %s", peerID)
+	delete(c.conns, peerID)
+	c.mu.Unlock()
+
+	if err := c.stateSubscription.UnsubscribeStateChange([]messages.PeerID{peerID}); err != nil {
+		c.log.Errorf("failed to unsubscribe from peer state change: %s", err)
+	}
+	container.close()
+}
+
 func (c *Client) close(gracefullyExit bool) error {
 	c.readLoopMutex.Lock()
 	defer c.readLoopMutex.Unlock()

shared/relay/client/manager.go

@@ -147,6 +147,32 @@ func (m *Manager) OpenConn(ctx context.Context, serverAddress, peerKey string) (
 	return netConn, err
 }
 
+// CloseConnByPeerKey closes an existing relay connection for the given peer key
+// on the relay client associated with serverAddress, so that a subsequent
+// OpenConn can create a fresh one.
+func (m *Manager) CloseConnByPeerKey(serverAddress, peerKey string) {
+	m.relayClientMu.RLock()
+	homeClient := m.relayClient
+	m.relayClientMu.RUnlock()
+
+	if homeClient == nil {
+		return
+	}
+
+	homeAddr, err := homeClient.ServerInstanceURL()
+	if err == nil && homeAddr == serverAddress {
+		homeClient.CloseConnByPeerKey(peerKey)
+		return
+	}
+
+	m.relayClientsMutex.RLock()
+	rt, ok := m.relayClients[serverAddress]
+	m.relayClientsMutex.RUnlock()
+	if ok && rt.relayClient != nil {
+		rt.relayClient.CloseConnByPeerKey(peerKey)
+	}
+}
+
 // Ready returns true if the home Relay client is connected to the relay server.
 func (m *Manager) Ready() bool {
 	m.relayClientMu.RLock()