From 1bfda2793bf30bf899f0cf3193a49d7beb9c00c1 Mon Sep 17 00:00:00 2001
From: riccardom <riccardomanfrin@gmail.com>
Date: Fri, 31 Jul 2026 10:47:07 +0200
Subject: [PATCH] pqkem: recover from persistent rekey failure by
 re-bootstrapping over signal

OnRekeyFailed now re-runs the KEM bootstrap over Signal (conn.RequestReoffer ->
handshaker.SendOffer) instead of only logging: a fresh signalling offer starts a new
exchange that overwrites the stalled PSK on both sides, resyncing after a persistent
data-path desync. Chosen over a responder-side awaitingAck revert (which fights the
confirm-less ack timing) and a full tunnel teardown (heavier). The tunnel stays up on
the previous PSK meanwhile since Signal is independent of the broken data path.
---
 client/internal/engine.go        | 13 ++++++++++++-
 client/internal/peer/conn.go     | 16 ++++++++++++++++
 client/internal/pqkem_adapter.go | 15 ++++++++++++---
 3 files changed, 40 insertions(+), 4 deletions(-)

diff --git a/client/internal/engine.go b/client/internal/engine.go
index f60d43743..3fbd8b64c 100644
--- a/client/internal/engine.go
+++ b/client/internal/engine.go
@@ -663,7 +663,18 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, mgmtURL *url.URL)
 		if pqErr != nil {
 			log.Errorf("pqkem: transport bind failed, exchange disabled: %v", pqErr)
 		} else {
-			e.pqkemManager = pqkem.NewManager(pqkem.LocalID(publicKey.String()), pqCallbackHandler{wg: e.wgInterface}, pqkem.NewLogger())
+			cbHandler := pqCallbackHandler{
+				wg: e.wgInterface,
+				// On a persistent rekey failure, re-bootstrap the KEM over Signal: a
+				// fresh signalling offer starts a new exchange that overwrites the
+				// stalled PSK on both sides, recovering from a data-path desync.
+				reoffer: func(remoteKey string) {
+					if conn, ok := e.peerStore.PeerConn(remoteKey); ok {
+						conn.RequestReoffer()
+					}
+				},
+			}
+			e.pqkemManager = pqkem.NewManager(pqkem.LocalID(publicKey.String()), cbHandler, pqkem.NewLogger())
 			e.pqkemManager.Start(tr)
 			log.Infof("pqkem: enabled (udp port %d on overlay %s)", e.pqkemManager.LocalPort(), e.config.WgAddr.IP)
 		}
diff --git a/client/internal/peer/conn.go b/client/internal/peer/conn.go
index ddc57c0d5..1e0904bf3 100644
--- a/client/internal/peer/conn.go
+++ b/client/internal/peer/conn.go
@@ -719,6 +719,22 @@ func (conn *Conn) onGuardEvent() {
 	}
 }
 
+// RequestReoffer sends a fresh signalling offer for the peer, re-running the
+// post-quantum bootstrap over Signal. Used to recover from a persistent data-path
+// rekey failure: a new exchange overwrites the stalled PSK on both sides. No-op if the
+// connection is not open yet.
+func (conn *Conn) RequestReoffer() {
+	conn.mu.Lock()
+	h := conn.handshaker
+	conn.mu.Unlock()
+	if h == nil {
+		return
+	}
+	if err := h.SendOffer(); err != nil {
+		conn.Log.Debugf("pqkem: recovery re-offer failed: %v", err)
+	}
+}
+
 func (conn *Conn) onWGDisconnected(watcherCtx context.Context) {
 	conn.mu.Lock()
 	defer conn.mu.Unlock()
diff --git a/client/internal/pqkem_adapter.go b/client/internal/pqkem_adapter.go
index fe41082ad..7a66c8a69 100644
--- a/client/internal/pqkem_adapter.go
+++ b/client/internal/pqkem_adapter.go
@@ -20,6 +20,9 @@ type pqPresharedKeySetter interface {
 // engine-side implementation of pqkem.CallbackHandler.
 type pqCallbackHandler struct {
 	wg pqPresharedKeySetter
+	// reoffer re-bootstraps the KEM over Signal for a peer (a fresh signalling offer)
+	// to recover from a persistent data-path rekey failure. Nil disables recovery.
+	reoffer func(remoteKey string)
 }
 
 // OnNewPSKReady programs the freshly derived PSK for the peer (updateOnly: a no-op
@@ -32,10 +35,16 @@ func (h pqCallbackHandler) OnNewPSKReady(remoteID pqkem.RemoteID, psk pqkem.PSK)
 	return h.wg.SetPresharedKey(string(remoteID), wgtypes.Key(psk), true)
 }
 
-// OnRekeyFailed reports a failed PQ (re)key convergence.
-// TODO(NET-1406): tear the peer connection down / trigger ICE reconnect.
+// OnRekeyFailed reports a failed PQ (re)key convergence and re-bootstraps the KEM over
+// Signal to recover: a fresh signalling offer starts a new exchange that overwrites the
+// stalled PSK on both sides, resyncing after a persistent data-path desync. The tunnel
+// stays up on the previous PSK meanwhile (the Signal channel is independent of the
+// broken data path).
 func (h pqCallbackHandler) OnRekeyFailed(remoteID pqkem.RemoteID) error {
-	log.Warnf("pqkem: post-quantum rekey failed for peer %s", remoteID)
+	log.Warnf("pqkem: post-quantum rekey failed for peer %s, re-bootstrapping over signal", remoteID)
+	if h.reoffer != nil {
+		h.reoffer(string(remoteID))
+	}
 	return nil
 }