Merge branch 'main' into refactor/relay-foreign-cache

# Conflicts:
#	shared/relay/client/manager.go

client/internal/peer/conn.go

@@ -6,6 +6,7 @@ import (
 	"net"
 	"net/netip"
 	"runtime"
+	"slices"
 	"sync"
 	"time"
 
@@ -29,6 +30,11 @@ import (
 	relayClient "github.com/netbirdio/netbird/shared/relay/client"
 )
 
+// wgTimeoutEscalationThreshold is the number of consecutive WireGuard
+// handshake timeouts after which the rosenpass state for the peer is
+// considered desynced and gets reset.
+const wgTimeoutEscalationThreshold = 3
+
 // MetricsRecorder is an interface for recording peer connection metrics
 type MetricsRecorder interface {
 	RecordConnectionStages(
@@ -117,6 +123,9 @@ type Conn struct {
 	wgWatcher       *WGWatcher
 	wgWatcherWg     sync.WaitGroup
 	wgWatcherCancel context.CancelFunc
+	// wgTimeouts counts consecutive WireGuard handshake timeouts without a
+	// successful handshake in between. Guarded by mu.
+	wgTimeouts int
 
 	// used to store the remote Rosenpass key for Relayed connection in case of connection update from ice
 	rosenpassRemoteKey []byte
@@ -136,6 +145,39 @@ type Conn struct {
 	// Connection stage timestamps for metrics
 	metricsRecorder MetricsRecorder
 	metricsStages   *MetricsStages
+
+	// pendingFirstPacket is the lazyconn-captured handshake init, replayed once the real
+	// transport is up.
+	pendingFirstPacket []byte
+}
+
+// injectPendingFirstPacket replays the captured handshake through the proxy if present, else
+// directly through the ICE conn. The packet is cleared only after a successful write, so a failed
+// or transport-less attempt leaves it available for a later reinjection. Caller must hold conn.mu.
+func (conn *Conn) injectPendingFirstPacket(proxy wgproxy.Proxy, directConn net.Conn) {
+	pkt := conn.pendingFirstPacket
+	if len(pkt) == 0 {
+		return
+	}
+
+	switch {
+	case proxy != nil:
+		if err := proxy.InjectPacket(pkt); err != nil {
+			conn.Log.Debugf("failed to reinject captured first packet via proxy: %v", err)
+			return
+		}
+	case directConn != nil:
+		if _, err := directConn.Write(pkt); err != nil {
+			conn.Log.Debugf("failed to reinject captured first packet via direct conn: %v", err)
+			return
+		}
+	default:
+		conn.Log.Debugf("no transport available to reinject captured first packet")
+		return
+	}
+
+	conn.pendingFirstPacket = nil
+	conn.Log.Debugf("reinjected captured first packet (%d bytes)", len(pkt))
 }
 
 // NewConn creates a new not opened Conn to the remote peer.
@@ -161,7 +203,6 @@ func NewConn(config ConnConfig, services ServiceDependencies) (*Conn, error) {
 		statusICE:          worker.NewAtomicStatus(),
 		dumpState:          dumpState,
 		endpointUpdater:    NewEndpointUpdater(connLog, config.WgConfig, isController(config)),
-		wgWatcher:          NewWGWatcher(connLog, config.WgConfig.WgInterface, config.Key, dumpState),
 		metricsRecorder:    services.MetricsRecorder,
 	}
 
@@ -172,6 +213,16 @@ func NewConn(config ConnConfig, services ServiceDependencies) (*Conn, error) {
 // It will try to establish a connection using ICE and in parallel with relay. The higher priority connection type will
 // be used.
 func (conn *Conn) Open(engineCtx context.Context) error {
+	return conn.open(engineCtx, nil)
+}
+
+// OpenWithFirstPacket opens the connection like Open and stashes firstPacket to be replayed once
+// the real transport is established. The packet is retained only on a successful open.
+func (conn *Conn) OpenWithFirstPacket(engineCtx context.Context, firstPacket []byte) error {
+	return conn.open(engineCtx, firstPacket)
+}
+
+func (conn *Conn) open(engineCtx context.Context, firstPacket []byte) error {
 	conn.mu.Lock()
 	defer conn.mu.Unlock()
 
@@ -227,6 +278,9 @@ func (conn *Conn) Open(engineCtx context.Context) error {
 		defer conn.wg.Done()
 		conn.guard.Start(conn.ctx, conn.onGuardEvent)
 	}()
+	if len(firstPacket) > 0 {
+		conn.pendingFirstPacket = slices.Clone(firstPacket)
+	}
 	conn.opened = true
 	return nil
 }
@@ -423,6 +477,8 @@ func (conn *Conn) onICEConnectionIsReady(priority conntype.ConnPriority, iceConn
 		conn.wgProxyRelay.RedirectAs(ep)
 	}
 
+	conn.injectPendingFirstPacket(wgProxy, iceConnInfo.RemoteConn)
+
 	conn.currentConnPriority = priority
 	conn.statusICE.SetConnected()
 	conn.updateIceState(iceConnInfo, updateTime)
@@ -546,6 +602,8 @@ func (conn *Conn) onRelayConnectionIsReady(rci RelayConnInfo) {
 
 	wgConfigWorkaround()
 
+	conn.injectPendingFirstPacket(wgProxy, nil)
+
 	conn.rosenpassRemoteKey = rci.rosenpassPubKey
 	conn.currentConnPriority = conntype.Relay
 	conn.statusRelay.SetConnected()
@@ -612,11 +670,12 @@ func (conn *Conn) onGuardEvent() {
 	}
 }
 
-func (conn *Conn) onWGDisconnected() {
+func (conn *Conn) onWGDisconnected(watcherCtx context.Context) {
 	conn.mu.Lock()
 	defer conn.mu.Unlock()
 
-	if conn.ctx.Err() != nil {
+	// watcherCtx guards against a stale watcher tearing down a connection that already superseded it.
+	if conn.ctx.Err() != nil || watcherCtx.Err() != nil {
 		return
 	}
 
@@ -632,6 +691,29 @@ func (conn *Conn) onWGDisconnected() {
 	default:
 		conn.Log.Debugf("No active connection to close on WG timeout")
 	}
+
+	conn.escalateWGTimeoutLocked()
+}
+
+// escalateWGTimeoutLocked resets the peer's rosenpass state after repeated
+// handshake timeouts. With rosenpass enabled, persistent timeouts mean the
+// preshared keys have desynced; the renewal exchange runs over the dead
+// tunnel and cannot resync them. Reporting the peer disconnected drops its
+// rosenpass state, so the next connection configuration programs the
+// rendezvous key and the tunnel can bootstrap again. Callers must hold mu.
+func (conn *Conn) escalateWGTimeoutLocked() {
+	if conn.config.RosenpassConfig.PubKey == nil {
+		return
+	}
+
+	conn.wgTimeouts++
+	if conn.wgTimeouts < wgTimeoutEscalationThreshold || conn.onDisconnected == nil {
+		return
+	}
+	conn.wgTimeouts = 0
+
+	conn.Log.Warnf("%d consecutive WireGuard handshake timeouts, resetting rosenpass state for peer", wgTimeoutEscalationThreshold)
+	conn.onDisconnected(conn.config.WgConfig.RemoteKey)
 }
 
 func (conn *Conn) updateRelayStatus(relayServerAddr string, rosenpassPubKey []byte, updateTime time.Time) {
@@ -751,23 +833,39 @@ func (conn *Conn) isConnectedOnAllWay() (status guard.ConnStatus) {
 	})
 }
 
+// enableWgWatcherIfNeeded starts a fresh watcher instance per connection attempt, so its
+// lifecycle stays bound to conn.mu and enable/disable can't race an old goroutine's shutdown.
+// Caller must hold conn.mu.
 func (conn *Conn) enableWgWatcherIfNeeded(enabledTime time.Time) {
-	if !conn.wgWatcher.IsEnabled() {
-		wgWatcherCtx, wgWatcherCancel := context.WithCancel(conn.ctx)
-		conn.wgWatcherCancel = wgWatcherCancel
-		conn.wgWatcherWg.Add(1)
-		go func() {
-			defer conn.wgWatcherWg.Done()
-			conn.wgWatcher.EnableWgWatcher(wgWatcherCtx, enabledTime, conn.onWGDisconnected, conn.onWGHandshakeSuccess)
-		}()
+	if conn.wgWatcher != nil {
+		return
 	}
+
+	watcher := NewWGWatcher(conn.Log, conn.config.WgConfig.WgInterface, conn.config.Key, conn.dumpState)
+	watcher.PrepareInitialHandshake()
+
+	wgWatcherCtx, wgWatcherCancel := context.WithCancel(conn.ctx)
+	conn.wgWatcher = watcher
+	conn.wgWatcherCancel = wgWatcherCancel
+
+	conn.wgWatcherWg.Add(1)
+	go func() {
+		defer conn.wgWatcherWg.Done()
+		onDisconnected := func() { conn.onWGDisconnected(wgWatcherCtx) }
+		watcher.EnableWgWatcher(wgWatcherCtx, enabledTime, onDisconnected, conn.onWGHandshakeSuccess, conn.onWGCheckSuccess)
+	}()
 }
 
+// disableWgWatcherIfNeeded cancels and drops the watcher once no transport is active. It never
+// waits for the goroutine: the timeout path reentrantly calls back here under conn.mu, so
+// blocking would deadlock. Caller must hold conn.mu.
 func (conn *Conn) disableWgWatcherIfNeeded() {
-	if conn.currentConnPriority == conntype.None && conn.wgWatcherCancel != nil {
-		conn.wgWatcherCancel()
-		conn.wgWatcherCancel = nil
+	if conn.currentConnPriority != conntype.None || conn.wgWatcher == nil {
+		return
 	}
+	conn.wgWatcherCancel()
+	conn.wgWatcher = nil
+	conn.wgWatcherCancel = nil
 }
 
 func (conn *Conn) newProxy(remoteConn net.Conn) (wgproxy.Proxy, error) {
@@ -790,7 +888,9 @@ func (conn *Conn) resetEndpoint() {
 		return
 	}
 	conn.Log.Infof("reset wg endpoint")
-	conn.wgWatcher.Reset()
+	if conn.wgWatcher != nil {
+		conn.wgWatcher.Reset()
+	}
 	if err := conn.endpointUpdater.RemoveEndpointAddress(); err != nil {
 		conn.Log.Warnf("failed to remove endpoint address before update: %v", err)
 	}
@@ -839,6 +939,15 @@ func (conn *Conn) onWGHandshakeSuccess(when time.Time) {
 	conn.recordConnectionMetrics()
 }
 
+// onWGCheckSuccess is called for every watcher check that observed a fresh
+// handshake, including handshakes of connections that were already up when
+// the watcher started.
+func (conn *Conn) onWGCheckSuccess() {
+	conn.mu.Lock()
+	conn.wgTimeouts = 0
+	conn.mu.Unlock()
+}
+
 // recordConnectionMetrics records connection stage timestamps as metrics
 func (conn *Conn) recordConnectionMetrics() {
 	if conn.metricsRecorder == nil {

client/internal/peer/conn_test.go

@@ -7,6 +7,7 @@ import (
 	"testing"
 	"time"
 
+	log "github.com/sirupsen/logrus"
 	"github.com/stretchr/testify/assert"
 
 	"github.com/netbirdio/netbird/client/iface"
@@ -304,3 +305,84 @@ func TestConn_presharedKey_RosenpassManaged(t *testing.T) {
 		t.Fatalf("expected non-nil presharedKey before Rosenpass manages PSK")
 	}
 }
+
+func newWGTimeoutTestConn(rosenpassEnabled bool, disconnected *[]string) *Conn {
+	cfg := ConnConfig{
+		Key:      "LLHf3Ma6z6mdLbriAJbqhX7+nM/B71lgw2+91q3LfhU=",
+		LocalKey: "RRHf3Ma6z6mdLbriAJbqhX7+nM/B71lgw2+91q3LfhU=",
+		WgConfig: WgConfig{RemoteKey: "LLHf3Ma6z6mdLbriAJbqhX7+nM/B71lgw2+91q3LfhU="},
+	}
+	if rosenpassEnabled {
+		cfg.RosenpassConfig = RosenpassConfig{PubKey: []byte("dummykey")}
+	}
+
+	conn := &Conn{
+		ctx:           context.Background(),
+		config:        cfg,
+		Log:           log.WithField("peer", cfg.Key),
+		metricsStages: &MetricsStages{},
+	}
+	conn.SetOnDisconnected(func(remotePeer string) {
+		*disconnected = append(*disconnected, remotePeer)
+	})
+	return conn
+}
+
+// TestConn_onWGDisconnected_EscalatesToRosenpassReset: repeated handshake
+// timeouts with rosenpass enabled mean the preshared keys have desynced. The
+// renewal exchange runs over the dead tunnel and cannot resync them, so after
+// wgTimeoutEscalationThreshold consecutive timeouts the conn must report the
+// peer disconnected, dropping its rosenpass state so the next configuration
+// programs the rendezvous key.
+func TestConn_onWGDisconnected_EscalatesToRosenpassReset(t *testing.T) {
+	var disconnected []string
+	conn := newWGTimeoutTestConn(true, &disconnected)
+
+	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
+		conn.onWGDisconnected(conn.ctx)
+	}
+	assert.Empty(t, disconnected, "escalation must not fire below the threshold")
+
+	conn.onWGDisconnected(conn.ctx)
+	assert.Equal(t, []string{conn.config.WgConfig.RemoteKey}, disconnected,
+		"reaching the threshold must report the peer disconnected once")
+
+	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
+		conn.onWGDisconnected(conn.ctx)
+	}
+	assert.Len(t, disconnected, 1, "escalation must restart counting after firing")
+
+	conn.onWGDisconnected(conn.ctx)
+	assert.Len(t, disconnected, 2, "continued timeouts must escalate again")
+}
+
+// TestConn_onWGDisconnected_CheckSuccessResetsEscalation: a successful
+// handshake between timeouts means the tunnel recovered; the counter must
+// start over.
+func TestConn_onWGDisconnected_CheckSuccessResetsEscalation(t *testing.T) {
+	var disconnected []string
+	conn := newWGTimeoutTestConn(true, &disconnected)
+
+	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
+		conn.onWGDisconnected(conn.ctx)
+	}
+	conn.onWGCheckSuccess()
+
+	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
+		conn.onWGDisconnected(conn.ctx)
+	}
+	assert.Empty(t, disconnected, "handshake success must reset the timeout count")
+}
+
+// TestConn_onWGDisconnected_NoEscalationWithoutRosenpass: without rosenpass
+// there is no per-peer key state to reset; repeated timeouts must not report
+// disconnects.
+func TestConn_onWGDisconnected_NoEscalationWithoutRosenpass(t *testing.T) {
+	var disconnected []string
+	conn := newWGTimeoutTestConn(false, &disconnected)
+
+	for i := 0; i < wgTimeoutEscalationThreshold*3; i++ {
+		conn.onWGDisconnected(conn.ctx)
+	}
+	assert.Empty(t, disconnected, "escalation must be limited to rosenpass connections")
+}

client/internal/peer/guard/guard.go

@@ -85,7 +85,11 @@ func (g *Guard) reconnectLoopWithRetry(ctx context.Context, callback func()) {
 	defer g.srWatcher.RemoveListener(srReconnectedChan)
 
 	ticker := g.initialTicker(ctx)
-	defer ticker.Stop()
+	defer func() {
+		// If backoff.Ticker.send is blocked, context.Done will not close the Ticker goroutine.
+		// We have to explicitly call Stop, even if we use backoff.WithContext.
+		ticker.Stop()
+	}()
 
 	tickerChannel := ticker.C
 

client/internal/peer/guard/guard_leak_test.go

@@ -0,0 +1,92 @@
+package guard
+
+import (
+	"context"
+	"runtime"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/client/internal/peer/ice"
+)
+
+func newTestGuard(status connStatusFunc) *Guard {
+	srw := NewSRWatcher(nil, nil, nil, ice.Config{})
+	return NewGuard(log.WithField("test", "guard"), status, 50*time.Millisecond, srw)
+}
+
+// countBackoffTickerGoroutines returns how many goroutines are currently sitting
+// in backoff/v4.(*Ticker).run (a ticker goroutine that has not exited).
+func countBackoffTickerGoroutines() int {
+	buf := make([]byte, 1<<25) // 32MB
+	n := runtime.Stack(buf, true)
+	return strings.Count(string(buf[:n]), "backoff/v4.(*Ticker).run")
+}
+
+// TestGuard_ReconnectTicker_NoGoroutineLeakOnShutdown reproduces a observed
+// leak: after a shutdown burst, ticker run/send goroutines stay parked
+// forever even though every reconnect loop has exited.
+func TestGuard_ReconnectTicker_NoGoroutineLeakOnShutdown(t *testing.T) {
+	before := countBackoffTickerGoroutines()
+
+	const peers = 6000
+	cancels := make([]context.CancelFunc, 0, peers)
+	var wg sync.WaitGroup
+
+	// A status check slower than the tick cadence. This models the real
+	// isConnectedOnAllWay/callback doing work: while the loop is busy in the
+	// handler, the ticker fires the next tick and parks in send(), because
+	// send() never selects on ctx.
+	slowStatus := func() ConnStatus {
+		time.Sleep(70 * time.Millisecond)
+		return ConnStatusConnected
+	}
+
+	for range peers {
+		g := newTestGuard(slowStatus)
+		ctx, cancel := context.WithCancel(context.Background())
+		cancels = append(cancels, cancel)
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			g.Start(ctx, func() {})
+		}()
+		// Force the live ticker to be a newReconnectTicker.
+		g.SetRelayedConnDisconnected()
+	}
+
+	// Let the replacement tickers get past their 800ms initial interval, so
+	// many are parked in send() waiting on the (slow) consumer when we tear
+	// everything down.
+	time.Sleep(1500 * time.Millisecond)
+
+	// Shutdown burst: cancel every peer at once, like engine teardown.
+	for _, c := range cancels {
+		c()
+	}
+
+	// Every reconnect loop must return
+	waitCh := make(chan struct{})
+	go func() { wg.Wait(); close(waitCh) }()
+	select {
+	case <-waitCh:
+	case <-time.After(30 * time.Second):
+		t.Fatal("not all reconnect loops returned after ctx cancel")
+	}
+
+	// Give any correctly-stopped ticker goroutines time to unwind.
+	for range 50 {
+		runtime.Gosched()
+		time.Sleep(10 * time.Millisecond)
+	}
+
+	leaked := countBackoffTickerGoroutines() - before
+	t.Logf("backoff Ticker.run goroutines still parked after teardown of %d peers: %d", peers, leaked)
+	if leaked > 0 {
+		t.Errorf("LEAK: %d backoff ticker goroutines parked after all reconnect loops exited "+
+			"(defer ticker.Stop() stops the initial ticker, not the live replacement)", leaked)
+	}
+}

client/internal/peer/status.go

@@ -7,6 +7,7 @@ import (
 	"net/netip"
 	"slices"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/google/uuid"
@@ -191,23 +192,30 @@ func (s *StatusChangeSubscription) Events() chan map[string]RouterState {
 // every private-service request) don't contend against each other.
 // Pure read methods take RLock; anything that mutates state takes Lock.
 type Status struct {
-	mux                   sync.RWMutex
-	muxRelays             sync.RWMutex
-	peers                 map[string]State
-	ipToKey               map[string]string
-	changeNotify          map[string]map[string]*StatusChangeSubscription // map[peerID]map[subscriptionID]*StatusChangeSubscription
-	signalState           bool
-	signalError           error
-	managementState       bool
-	managementError       error
-	relayStates           []relay.ProbeResult
-	localPeer             LocalPeerState
-	offlinePeers          []State
-	mgmAddress            string
-	signalAddress         string
-	notifier              *notifier
-	rosenpassEnabled      bool
-	rosenpassPermissive   bool
+	mux                 sync.RWMutex
+	muxRelays           sync.RWMutex
+	peers               map[string]State
+	ipToKey             map[string]string
+	changeNotify        map[string]map[string]*StatusChangeSubscription // map[peerID]map[subscriptionID]*StatusChangeSubscription
+	signalState         bool
+	signalError         error
+	managementState     bool
+	managementError     error
+	relayStates         []relay.ProbeResult
+	localPeer           LocalPeerState
+	offlinePeers        []State
+	mgmAddress          string
+	signalAddress       string
+	notifier            *notifier
+	rosenpassEnabled    bool
+	rosenpassPermissive bool
+	// sessionExpiresAt is the absolute UTC instant at which the peer's SSO
+	// session expires. Zero when the peer is not SSO-tracked or login
+	// expiration is disabled. Populated from management LoginResponse /
+	// SyncResponse and exposed via the daemon's Status / SubscribeStatus RPC
+	// so the UI can show remaining time without itself talking to mgm.
+	sessionExpiresAt time.Time
+
 	nsGroupStates         []NSGroupState
 	resolvedDomainsStates map[domain.Domain]ResolvedDomainInfo
 	lazyConnectionEnabled bool
@@ -223,6 +231,21 @@ type Status struct {
 	eventStreams map[string]chan *proto.SystemEvent
 	eventQueue   *EventQueue
 
+	// stateChangeStreams fan-out connection-state changes (connected /
+	// disconnected / connecting / address change / peers list change) to
+	// every active SubscribeStatus gRPC stream. Each subscriber gets a
+	// buffered chan; the notifier non-blockingly pings them so a slow
+	// consumer can never stall the daemon.
+	stateChangeMux     sync.Mutex
+	stateChangeStreams map[string]chan struct{}
+
+	// networksRevision bumps whenever the routed-networks set or their
+	// selected state changes (driven by the route manager). Surfaced in the
+	// status snapshot so the UI can fingerprint on it and re-fetch
+	// ListNetworks only on a real change. Atomic so the snapshot builder can
+	// read it without taking mux.
+	networksRevision atomic.Uint64
+
 	ingressGwMgr *ingressgw.Manager
 
 	routeIDLookup routeIDLookup
@@ -237,6 +260,7 @@ func NewRecorder(mgmAddress string) *Status {
 		changeNotify:          make(map[string]map[string]*StatusChangeSubscription),
 		eventStreams:          make(map[string]chan *proto.SystemEvent),
 		eventQueue:            NewEventQueue(eventQueueSize),
+		stateChangeStreams:    make(map[string]chan struct{}),
 		offlinePeers:          make([]State, 0),
 		notifier:              newNotifier(),
 		mgmAddress:            mgmAddress,
@@ -401,6 +425,7 @@ func (d *Status) UpdatePeerState(receivedState State) error {
 	if notifyRouter {
 		d.dispatchRouterPeers(receivedState.PubKey, routerSnapshot)
 	}
+	d.notifyStateChange()
 	return nil
 }
 
@@ -426,6 +451,7 @@ func (d *Status) AddPeerStateRoute(peer string, route string, resourceId route.R
 
 	// todo: consider to make sense of this notification or not
 	d.notifier.peerListChanged(numPeers)
+	d.notifyStateChange()
 	return nil
 }
 
@@ -451,6 +477,7 @@ func (d *Status) RemovePeerStateRoute(peer string, route string) error {
 
 	// todo: consider to make sense of this notification or not
 	d.notifier.peerListChanged(numPeers)
+	d.notifyStateChange()
 	return nil
 }
 
@@ -500,6 +527,7 @@ func (d *Status) UpdatePeerICEState(receivedState State) error {
 	if notifyRouter {
 		d.dispatchRouterPeers(receivedState.PubKey, routerSnapshot)
 	}
+	d.notifyStateChange()
 	return nil
 }
 
@@ -536,6 +564,7 @@ func (d *Status) UpdatePeerRelayedState(receivedState State) error {
 	if notifyRouter {
 		d.dispatchRouterPeers(receivedState.PubKey, routerSnapshot)
 	}
+	d.notifyStateChange()
 	return nil
 }
 
@@ -571,6 +600,7 @@ func (d *Status) UpdatePeerRelayedStateToDisconnected(receivedState State) error
 	if notifyRouter {
 		d.dispatchRouterPeers(receivedState.PubKey, routerSnapshot)
 	}
+	d.notifyStateChange()
 	return nil
 }
 
@@ -609,6 +639,7 @@ func (d *Status) UpdatePeerICEStateToDisconnected(receivedState State) error {
 	if notifyRouter {
 		d.dispatchRouterPeers(receivedState.PubKey, routerSnapshot)
 	}
+	d.notifyStateChange()
 	return nil
 }
 
@@ -702,6 +733,7 @@ func (d *Status) FinishPeerListModifications() {
 	for _, rd := range dispatches {
 		d.dispatchRouterPeers(rd.peerID, rd.snapshot)
 	}
+	d.notifyStateChange()
 }
 
 func (d *Status) SubscribeToPeerStateChanges(ctx context.Context, peerID string) *StatusChangeSubscription {
@@ -760,6 +792,41 @@ func (d *Status) UpdateLocalPeerState(localPeerState LocalPeerState) {
 	d.mux.Unlock()
 
 	d.notifier.localAddressChanged(fqdn, ip)
+	d.notifyStateChange()
+}
+
+// SetSessionExpiresAt records the absolute UTC instant at which the peer's
+// SSO session is set to expire. Pass the zero value to clear (e.g. when the
+// management server stops publishing a deadline because login expiration was
+// disabled or the peer is not SSO-tracked). Same-value updates are no-ops;
+// real changes fan out via notifyStateChange so SubscribeStatus consumers
+// pick up the new deadline on their next read.
+func (d *Status) SetSessionExpiresAt(deadline time.Time) {
+	d.mux.Lock()
+	if d.sessionExpiresAt.Equal(deadline) {
+		d.mux.Unlock()
+		return
+	}
+	d.sessionExpiresAt = deadline
+	d.mux.Unlock()
+	d.notifyStateChange()
+}
+
+// GetSessionExpiresAt returns the most recently recorded SSO session deadline,
+// or the zero value when no deadline is tracked. A deadline that has already
+// slipped into the past reports as "none": once the session has expired it is
+// no longer a meaningful countdown, and the sessionwatch.Watcher does not
+// arm a timer at the deadline itself to clear it (only the two pre-expiry
+// warnings). Without this guard the UI would keep painting a stale
+// "expires in …" against a moment that has passed until the next login,
+// extend, or teardown rewrote the value.
+func (d *Status) GetSessionExpiresAt() time.Time {
+	d.mux.Lock()
+	defer d.mux.Unlock()
+	if !d.sessionExpiresAt.IsZero() && d.sessionExpiresAt.Before(time.Now()) {
+		return time.Time{}
+	}
+	return d.sessionExpiresAt
 }
 
 // AddLocalPeerStateRoute adds a route to the local peer state
@@ -828,11 +895,19 @@ func (d *Status) CleanLocalPeerState() {
 	d.mux.Unlock()
 
 	d.notifier.localAddressChanged(fqdn, ip)
+	d.notifyStateChange()
 }
 
 // MarkManagementDisconnected sets ManagementState to disconnected
 func (d *Status) MarkManagementDisconnected(err error) {
 	d.mux.Lock()
+	// Health checks re-mark the same state on every probe; skip the fan-out
+	// when nothing actually changed so we don't flood SubscribeStatus
+	// consumers with identical snapshots.
+	if !d.managementState && errors.Is(d.managementError, err) {
+		d.mux.Unlock()
+		return
+	}
 	d.managementState = false
 	d.managementError = err
 	mgm := d.managementState
@@ -840,11 +915,16 @@ func (d *Status) MarkManagementDisconnected(err error) {
 	d.mux.Unlock()
 
 	d.notifier.updateServerStates(mgm, sig)
+	d.notifyStateChange()
 }
 
 // MarkManagementConnected sets ManagementState to connected
 func (d *Status) MarkManagementConnected() {
 	d.mux.Lock()
+	if d.managementState && d.managementError == nil {
+		d.mux.Unlock()
+		return
+	}
 	d.managementState = true
 	d.managementError = nil
 	mgm := d.managementState
@@ -852,6 +932,7 @@ func (d *Status) MarkManagementConnected() {
 	d.mux.Unlock()
 
 	d.notifier.updateServerStates(mgm, sig)
+	d.notifyStateChange()
 }
 
 // UpdateSignalAddress update the address of the signal server
@@ -885,6 +966,10 @@ func (d *Status) UpdateLazyConnection(enabled bool) {
 // MarkSignalDisconnected sets SignalState to disconnected
 func (d *Status) MarkSignalDisconnected(err error) {
 	d.mux.Lock()
+	if !d.signalState && errors.Is(d.signalError, err) {
+		d.mux.Unlock()
+		return
+	}
 	d.signalState = false
 	d.signalError = err
 	mgm := d.managementState
@@ -892,11 +977,16 @@ func (d *Status) MarkSignalDisconnected(err error) {
 	d.mux.Unlock()
 
 	d.notifier.updateServerStates(mgm, sig)
+	d.notifyStateChange()
 }
 
 // MarkSignalConnected sets SignalState to connected
 func (d *Status) MarkSignalConnected() {
 	d.mux.Lock()
+	if d.signalState && d.signalError == nil {
+		d.mux.Unlock()
+		return
+	}
 	d.signalState = true
 	d.signalError = nil
 	mgm := d.managementState
@@ -904,6 +994,7 @@ func (d *Status) MarkSignalConnected() {
 	d.mux.Unlock()
 
 	d.notifier.updateServerStates(mgm, sig)
+	d.notifyStateChange()
 }
 
 func (d *Status) UpdateRelayStates(relayResults []relay.ProbeResult) {
@@ -1110,16 +1201,19 @@ func (d *Status) GetFullStatus() FullStatus {
 // ClientStart will notify all listeners about the new service state
 func (d *Status) ClientStart() {
 	d.notifier.clientStart()
+	d.notifyStateChange()
 }
 
 // ClientStop will notify all listeners about the new service state
 func (d *Status) ClientStop() {
 	d.notifier.clientStop()
+	d.notifyStateChange()
 }
 
 // ClientTeardown will notify all listeners about the service is under teardown
 func (d *Status) ClientTeardown() {
 	d.notifier.clientTearDown()
+	d.notifyStateChange()
 }
 
 // SetConnectionListener set a listener to the notifier
@@ -1261,6 +1355,79 @@ func (d *Status) GetEventHistory() []*proto.SystemEvent {
 	return d.eventQueue.GetAll()
 }
 
+// SubscribeToStateChanges hands back a channel that receives a tick on
+// every connection-state change (connected / disconnected / connecting /
+// address change / peers-list change). The channel is buffered to one
+// pending tick so a coalesced burst still wakes the consumer exactly
+// once. Pass the returned id to UnsubscribeFromStateChanges to detach.
+func (d *Status) SubscribeToStateChanges() (string, <-chan struct{}) {
+	d.stateChangeMux.Lock()
+	defer d.stateChangeMux.Unlock()
+
+	id := uuid.New().String()
+	ch := make(chan struct{}, 1)
+	d.stateChangeStreams[id] = ch
+	return id, ch
+}
+
+// UnsubscribeFromStateChanges releases a SubscribeToStateChanges channel
+// and closes it so any consumer goroutine selecting on the channel
+// unblocks cleanly.
+func (d *Status) UnsubscribeFromStateChanges(id string) {
+	d.stateChangeMux.Lock()
+	defer d.stateChangeMux.Unlock()
+
+	if ch, ok := d.stateChangeStreams[id]; ok {
+		close(ch)
+		delete(d.stateChangeStreams, id)
+	}
+}
+
+// notifyStateChange wakes every SubscribeToStateChanges subscriber. Drops
+// the tick if a subscriber's buffer is full — by definition the consumer
+// is already going to fetch the latest snapshot, so multiple pending ticks
+// would be redundant.
+func (d *Status) notifyStateChange() {
+	d.stateChangeMux.Lock()
+	defer d.stateChangeMux.Unlock()
+
+	for _, ch := range d.stateChangeStreams {
+		select {
+		case ch <- struct{}{}:
+		default:
+		}
+	}
+}
+
+// NotifyStateChange is the public wake-the-subscribers entry point used by
+// callers that mutate state outside the peer recorder — most importantly
+// the connect-state machine, which writes StatusNeedsLogin into the
+// shared contextState (client/internal/state.go) without touching any
+// recorder field. Without this push the SubscribeStatus stream stays on
+// the previous snapshot until an unrelated peer/management/signal
+// change happens to fire notifyStateChange, leaving the UI's status
+// out of sync with the daemon.
+func (d *Status) NotifyStateChange() {
+	d.notifyStateChange()
+}
+
+// BumpNetworksRevision increments the routed-networks revision and wakes every
+// SubscribeStatus subscriber. The route manager calls it when a network map
+// changes the available routes or when a selection is applied — the peer
+// status itself only records actively-routed (chosen) networks, so without
+// this bump a candidate route appearing/disappearing would never reach the UI.
+func (d *Status) BumpNetworksRevision() {
+	d.networksRevision.Add(1)
+	d.notifyStateChange()
+}
+
+// GetNetworksRevision returns the current routed-networks revision, surfaced in
+// the status snapshot so the UI can detect route/selection changes (see
+// BumpNetworksRevision).
+func (d *Status) GetNetworksRevision() uint64 {
+	return d.networksRevision.Load()
+}
+
 func (d *Status) SetWgIface(wgInterface WGIfaceStatus) {
 	d.mux.Lock()
 	defer d.mux.Unlock()

client/internal/peer/status_test.go

@@ -314,3 +314,39 @@ func TestGetFullStatus(t *testing.T) {
 	assert.Equal(t, signalState, fullStatus.SignalState, "signal status should be equal")
 	assert.ElementsMatch(t, []State{peerState1, peerState2}, fullStatus.Peers, "peers states should match")
 }
+
+// notified reports whether a state-change tick is pending on ch, draining it.
+func notified(ch <-chan struct{}) bool {
+	select {
+	case <-ch:
+		return true
+	default:
+		return false
+	}
+}
+
+func TestMarkServerStateDoesNotNotifyWhenUnchanged(t *testing.T) {
+	status := NewRecorder("https://mgm")
+	_, ch := status.SubscribeToStateChanges()
+
+	// First transition is a real change and must notify.
+	status.MarkManagementConnected()
+	require.True(t, notified(ch), "first connect should notify")
+
+	// Re-marking the same state must not notify again.
+	status.MarkManagementConnected()
+	assert.False(t, notified(ch), "redundant connect should not notify")
+
+	// Same for signal.
+	status.MarkSignalConnected()
+	require.True(t, notified(ch), "first signal connect should notify")
+	status.MarkSignalConnected()
+	assert.False(t, notified(ch), "redundant signal connect should not notify")
+
+	// A genuine change (disconnect with an error) notifies again.
+	err := errors.New("boom")
+	status.MarkManagementDisconnected(err)
+	require.True(t, notified(ch), "disconnect should notify")
+	status.MarkManagementDisconnected(err)
+	assert.False(t, notified(ch), "redundant disconnect should not notify")
+}

client/internal/peer/wg_watcher.go

@@ -3,7 +3,6 @@ package peer
 import (
 	"context"
 	"fmt"
-	"sync"
 	"time"
 
 	log "github.com/sirupsen/logrus"
@@ -24,14 +23,16 @@ type WGInterfaceStater interface {
 	GetStats() (map[string]configurer.WGStats, error)
 }
 
+// WGWatcher is single-shot: one instance per connection attempt, run once, then discarded.
+// Lifecycle is owned by Conn under conn.mu, so it keeps no "enabled" state to go stale.
 type WGWatcher struct {
 	log           *log.Entry
 	wgIfaceStater WGInterfaceStater
 	peerKey       string
 	stateDump     *stateDump
 
-	enabled   bool
-	muEnabled sync.RWMutex
+	// initialHandshake is not thread-safe; never call PrepareInitialHandshake and EnableWgWatcher concurrently.
+	initialHandshake time.Time
 
 	resetCh chan struct{}
 }
@@ -46,36 +47,23 @@ func NewWGWatcher(log *log.Entry, wgIfaceStater WGInterfaceStater, peerKey strin
 	}
 }
 
-// EnableWgWatcher starts the WireGuard watcher. If it is already enabled, it will return immediately and do nothing.
-// The watcher runs until ctx is cancelled. Caller is responsible for context lifecycle management.
-func (w *WGWatcher) EnableWgWatcher(ctx context.Context, enabledTime time.Time, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time)) {
-	w.muEnabled.Lock()
-	if w.enabled {
-		w.muEnabled.Unlock()
-		return
-	}
-
+// PrepareInitialHandshake reads the peer's current WireGuard handshake time. It must be
+// called before the peer is (re)configured on the WireGuard interface, so the captured
+// baseline reflects the state prior to this connection attempt instead of racing with
+// that configuration.
+func (w *WGWatcher) PrepareInitialHandshake() {
 	w.log.Debugf("enable WireGuard watcher")
-	w.enabled = true
-	w.muEnabled.Unlock()
-
-	initialHandshake, err := w.wgState()
-	if err != nil {
-		w.log.Warnf("failed to read initial wg stats: %v", err)
-	}
-
-	w.periodicHandshakeCheck(ctx, onDisconnectedFn, onHandshakeSuccessFn, enabledTime, initialHandshake)
-
-	w.muEnabled.Lock()
-	w.enabled = false
-	w.muEnabled.Unlock()
+	handshake, _ := w.wgState()
+	w.initialHandshake = handshake
 }
 
-// IsEnabled returns true if the WireGuard watcher is currently enabled
-func (w *WGWatcher) IsEnabled() bool {
-	w.muEnabled.RLock()
-	defer w.muEnabled.RUnlock()
-	return w.enabled
+// EnableWgWatcher runs the WireGuard watcher loop using the handshake baseline captured by
+// PrepareInitialHandshake. The watcher runs until ctx is cancelled. Caller is responsible
+// for context lifecycle management. onHandshakeSuccessFn is called only for the first
+// handshake observed by this run, onCheckSuccessFn for every check that observed a fresh
+// handshake, including the first.
+func (w *WGWatcher) EnableWgWatcher(ctx context.Context, enabledTime time.Time, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time), onCheckSuccessFn func()) {
+	w.periodicHandshakeCheck(ctx, onDisconnectedFn, onHandshakeSuccessFn, onCheckSuccessFn, enabledTime, w.initialHandshake)
 }
 
 // Reset signals the watcher that the WireGuard peer has been reset and a new
@@ -88,7 +76,7 @@ func (w *WGWatcher) Reset() {
 }
 
 // wgStateCheck help to check the state of the WireGuard handshake and relay connection
-func (w *WGWatcher) periodicHandshakeCheck(ctx context.Context, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time), enabledTime time.Time, initialHandshake time.Time) {
+func (w *WGWatcher) periodicHandshakeCheck(ctx context.Context, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time), onCheckSuccessFn func(), enabledTime time.Time, initialHandshake time.Time) {
 	w.log.Infof("WireGuard watcher started")
 
 	timer := time.NewTimer(wgHandshakeOvertime)
@@ -101,17 +89,25 @@ func (w *WGWatcher) periodicHandshakeCheck(ctx context.Context, onDisconnectedFn
 		case <-timer.C:
 			handshake, ok := w.handshakeCheck(lastHandshake)
 			if !ok {
+				// early ctx cancel check return
+				if ctx.Err() != nil {
+					return
+				}
 				onDisconnectedFn()
 				return
 			}
 			if lastHandshake.IsZero() {
 				elapsed := calcElapsed(enabledTime, *handshake)
 				w.log.Infof("first wg handshake detected within: %.2fsec, (%s)", elapsed, handshake)
-				if onHandshakeSuccessFn != nil {
+				if onHandshakeSuccessFn != nil && ctx.Err() == nil {
 					onHandshakeSuccessFn(*handshake)
 				}
 			}
 
+			if onCheckSuccessFn != nil && ctx.Err() == nil {
+				onCheckSuccessFn()
+			}
+
 			lastHandshake = *handshake
 
 			resetTime := time.Until(handshake.Add(checkPeriod))
@@ -142,9 +138,9 @@ func (w *WGWatcher) handshakeCheck(lastHandshake time.Time) (*time.Time, bool) {
 
 	w.log.Tracef("previous handshake, handshake: %v, %v", lastHandshake, handshake)
 
-	// the current know handshake did not change
+	// the current known handshake did not change
 	if handshake.Equal(lastHandshake) {
-		w.log.Warnf("WireGuard handshake timed out: %v", handshake)
+		w.log.Warnf("WireGuard handshake not updated: %v", handshake)
 		return nil, false
 	}
 

client/internal/peer/wg_watcher_test.go

@@ -23,6 +23,72 @@ func (m *MocWgIface) disconnect() {
 	m.stop = true
 }
 
+type mockHandshakeStats struct {
+	mu        sync.Mutex
+	handshake time.Time
+}
+
+func (m *mockHandshakeStats) GetStats() (map[string]configurer.WGStats, error) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	return map[string]configurer.WGStats{"": {LastHandshake: m.handshake}}, nil
+}
+
+func (m *mockHandshakeStats) advance() {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.handshake = time.Now()
+}
+
+// TestWGWatcher_CheckSuccessCallback: onCheckSuccessFn must fire for a fresh
+// handshake even when the watcher started with an existing handshake baseline,
+// the case where onHandshakeSuccessFn stays silent.
+func TestWGWatcher_CheckSuccessCallback(t *testing.T) {
+	// checkPeriod bounds how stale a handshake may be before the watcher treats it
+	// as a suspended-machine timeout. The first check fires after wgHandshakeOvertime,
+	// so keep checkPeriod well above any scheduling jitter to avoid a false timeout
+	// converting the expected success into a disconnect on a loaded runner.
+	checkPeriod = 1 * time.Minute
+	wgHandshakeOvertime = 1 * time.Second
+
+	mlog := log.WithField("peer", "tet")
+	// Use an old baseline so advance() yields a strictly newer handshake even on
+	// platforms with coarse clock resolution (Windows), where two time.Now() calls
+	// microseconds apart can return the same instant and read as a timed-out handshake.
+	stats := &mockHandshakeStats{handshake: time.Now().Add(-time.Hour)}
+	watcher := NewWGWatcher(mlog, stats, "", newStateDump("peer", mlog, &Status{}))
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	watcher.PrepareInitialHandshake()
+
+	firstHandshake := make(chan struct{}, 1)
+	checkSuccess := make(chan struct{}, 1)
+	go watcher.EnableWgWatcher(ctx, time.Now(), func() {}, func(when time.Time) {
+		firstHandshake <- struct{}{}
+	}, func() {
+		select {
+		case checkSuccess <- struct{}{}:
+		default:
+		}
+	})
+
+	stats.advance()
+
+	select {
+	case <-checkSuccess:
+	case <-time.After(10 * time.Second):
+		t.Errorf("timeout waiting for check success callback")
+	}
+
+	select {
+	case <-firstHandshake:
+		t.Errorf("first-handshake callback must not fire for a non-zero baseline")
+	default:
+	}
+}
+
 func TestWGWatcher_EnableWgWatcher(t *testing.T) {
 	checkPeriod = 5 * time.Second
 	wgHandshakeOvertime = 1 * time.Second
@@ -34,13 +100,15 @@ func TestWGWatcher_EnableWgWatcher(t *testing.T) {
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 
+	watcher.PrepareInitialHandshake()
+
 	onDisconnected := make(chan struct{}, 1)
 	go watcher.EnableWgWatcher(ctx, time.Now(), func() {
 		mlog.Infof("onDisconnectedFn")
 		onDisconnected <- struct{}{}
 	}, func(when time.Time) {
 		mlog.Infof("onHandshakeSuccess: %v", when)
-	})
+	}, nil)
 
 	// wait for initial reading
 	time.Sleep(2 * time.Second)
@@ -62,11 +130,13 @@ func TestWGWatcher_ReEnable(t *testing.T) {
 	watcher := NewWGWatcher(mlog, mocWgIface, "", newStateDump("peer", mlog, &Status{}))
 
 	ctx, cancel := context.WithCancel(context.Background())
+	watcher.PrepareInitialHandshake()
+
 	wg := &sync.WaitGroup{}
 	wg.Add(1)
 	go func() {
 		defer wg.Done()
-		watcher.EnableWgWatcher(ctx, time.Now(), func() {}, func(when time.Time) {})
+		watcher.EnableWgWatcher(ctx, time.Now(), func() {}, func(when time.Time) {}, nil)
 	}()
 	cancel()
 
@@ -76,10 +146,12 @@ func TestWGWatcher_ReEnable(t *testing.T) {
 	ctx, cancel = context.WithCancel(context.Background())
 	defer cancel()
 
+	watcher.PrepareInitialHandshake()
+
 	onDisconnected := make(chan struct{}, 1)
 	go watcher.EnableWgWatcher(ctx, time.Now(), func() {
 		onDisconnected <- struct{}{}
-	}, func(when time.Time) {})
+	}, func(when time.Time) {}, nil)
 
 	time.Sleep(2 * time.Second)
 	mocWgIface.disconnect()