control/controlclient,ipn/ipnlocal,wgengine: avoid restarting wireguard when key is learned via tsmp (#19142)

When disco keys are learned on a node that is connected to control and
has a mapSession, wgengine will see the key as having changed, and
assume that any existing connections will need to be reset.

For keys learned via TSMP, the connection should not be reset as that
key is learned via an active wireguard connection. If wgengine resets
that connetion, a 15s timeout will occur.

This change adds a map to track new keys coming in via TSMP, and removes
them from the list of keys that needs to trigger wireguard resets. This
is done with an interface chain from controlclient down via localBackend
to userspaceEngine via the watchdog.

Once a key has been actively used for preventing a wireguard reset, the
key is removed from the map.

If mapSession becomes a long lived process instead of being dependent on
having a connection to control. This interface chain can be removed, and
the event sequence from wrap->controlClient->userspaceEngine, can be
changed to wrap->userspaceEngine->controlClient as we know the map will
not be gunked up with stale TSMP entries.

Updates #12639

Signed-off-by: Claus Lensbøl <claus@tailscale.com>

control/controlclient/auto.go

@@ -478,6 +478,27 @@ func (mrs mapRoutineState) UpdateNetmapDelta(muts []netmap.NodeMutation) bool {
 	return err == nil && ok
 }
 
+var _ patchDiscoKeyer = mapRoutineState{}
+
+func (mrs mapRoutineState) PatchDiscoKey(pub key.NodePublic, disco key.DiscoPublic) {
+	c := mrs.c
+	c.mu.Lock()
+	goodState := c.loggedIn && c.inMapPoll
+	dun, ok := c.observer.(patchDiscoKeyer)
+	c.mu.Unlock()
+
+	if !goodState || !ok {
+		return
+	}
+
+	ctx, cancel := context.WithTimeout(c.mapCtx, 2*time.Second)
+	defer cancel()
+
+	c.observerQueue.RunSync(ctx, func() {
+		dun.PatchDiscoKey(pub, disco)
+	})
+}
+
 // mapRoutine is responsible for keeping a read-only streaming connection to the
 // control server, and keeping the netmap up to date.
 func (c *Auto) mapRoutine() {

control/controlclient/direct.go

@@ -228,6 +228,15 @@ type NetmapDeltaUpdater interface {
 	UpdateNetmapDelta([]netmap.NodeMutation) (ok bool)
 }
 
+// patchDiscoKeyer is an optional interface that can be implemented by an [Observer] to be
+// notified about node disco keys received out-of-band from control, via
+// existing connection state.
+type patchDiscoKeyer interface {
+	// PatchDiscoKey reports to the receiver that the specified disco key
+	// for node was obtained out-of-band from control.
+	PatchDiscoKey(key.NodePublic, key.DiscoPublic)
+}
+
 var nextControlClientID atomic.Int64
 
 // NewDirect returns a new Direct client.
@@ -367,7 +376,7 @@ func NewDirect(opts Options) (*Direct, error) {
 			// mapSession has gone away, we want to fall back to pushing the key
 			// further down the chain.
 			if err := c.streamingMapSession.updateDiscoForNode(
-				peer.ID(), update.Key, time.Now(), false); err == nil ||
+				peer.ID(), peer.Key(), update.Key, time.Now(), false); err == nil ||
 				!errors.Is(err, ErrChangeQueueClosed) {
 				return
 			}
@@ -377,10 +386,7 @@ func NewDirect(opts Options) (*Direct, error) {
 		// not have a mapSession (we are not connected to control) or because the
 		// mapSession queue has closed.
 		c.logf("controlclient direct: updating discoKey for %v via magicsock", update.Src)
-		discoKeyPub.Publish(events.PeerDiscoKeyUpdate{
+		discoKeyPub.Publish(events.PeerDiscoKeyUpdate(update))
-			Src: update.Src,
-			Key: update.Key,
-		})
 	})
 
 	return c, nil
@@ -859,8 +865,10 @@ func (c *Direct) PollNetMap(ctx context.Context, nu NetmapUpdater) error {
 // update it observed. It is used by tests and [NetmapFromMapResponseForDebug].
 // It will report only the first netmap seen.
 type rememberLastNetmapUpdater struct {
-	last *netmap.NetworkMap
+	last          *netmap.NetworkMap
-	done chan any
+	lastTSMPKey   key.NodePublic
+	lastTSMPDisco key.DiscoPublic
+	done          chan any
 }
 
 func (nu *rememberLastNetmapUpdater) UpdateFullNetmap(nm *netmap.NetworkMap) {
@@ -871,6 +879,11 @@ func (nu *rememberLastNetmapUpdater) UpdateFullNetmap(nm *netmap.NetworkMap) {
 	}
 }
 
+func (nu *rememberLastNetmapUpdater) PatchDiscoKey(key key.NodePublic, disco key.DiscoPublic) {
+	nu.lastTSMPKey = key
+	nu.lastTSMPDisco = disco
+}
+
 // FetchNetMapForTest fetches the netmap once.
 func (c *Direct) FetchNetMapForTest(ctx context.Context) (*netmap.NetworkMap, error) {
 	var nu rememberLastNetmapUpdater

control/controlclient/map.go

@@ -37,6 +37,11 @@ import (
 	"tailscale.com/wgengine/filter"
 )
 
+type responseWithSource struct {
+	response *tailcfg.MapResponse
+	viaTSMP  bool
+}
+
 // mapSession holds the state over a long-polled "map" request to the
 // control plane.
 //
@@ -98,7 +103,7 @@ type mapSession struct {
 	lastTKAInfo            *tailcfg.TKAInfo
 	lastNetmapSummary      string // from NetworkMap.VeryConcise
 	cqmu                   sync.Mutex
-	changeQueue            chan (*tailcfg.MapResponse)
+	changeQueue            chan responseWithSource
 	changeQueueClosed      bool
 	processQueue           sync.WaitGroup
 }
@@ -123,7 +128,7 @@ func newMapSession(privateNodeKey key.NodePrivate, nu NetmapUpdater, controlKnob
 		cancel:            func() {},
 		onDebug:           func(context.Context, *tailcfg.Debug) error { return nil },
 		onSelfNodeChanged: func(*netmap.NetworkMap) {},
-		changeQueue:       make(chan *tailcfg.MapResponse),
+		changeQueue:       make(chan responseWithSource),
 		changeQueueClosed: false,
 	}
 	ms.sessionAliveCtx, ms.sessionAliveCtxClose = context.WithCancel(context.Background())
@@ -142,7 +147,7 @@ func (ms *mapSession) run() {
 	for {
 		select {
 		case change := <-ms.changeQueue:
-			ms.handleNonKeepAliveMapResponse(ms.sessionAliveCtx, change)
+			ms.handleNonKeepAliveMapResponse(ms.sessionAliveCtx, change.response, change.viaTSMP)
 		case <-ms.sessionAliveCtx.Done():
 			// Drain any remaining items in the queue before exiting.
 			// Lock the queue during this time to avoid updates through other channels
@@ -154,7 +159,7 @@ func (ms *mapSession) run() {
 			for {
 				select {
 				case change := <-ms.changeQueue:
-					ms.handleNonKeepAliveMapResponse(ms.sessionAliveCtx, change)
+					ms.handleNonKeepAliveMapResponse(ms.sessionAliveCtx, change.response, change.viaTSMP)
 				default:
 					// Queue is empty, close it and exit
 					close(ms.changeQueue)
@@ -190,7 +195,7 @@ func (ms *mapSession) Close() {
 
 var ErrChangeQueueClosed = errors.New("change queue closed")
 
-func (ms *mapSession) updateDiscoForNode(id tailcfg.NodeID, key key.DiscoPublic, lastSeen time.Time, online bool) error {
+func (ms *mapSession) updateDiscoForNode(id tailcfg.NodeID, key key.NodePublic, discoKey key.DiscoPublic, lastSeen time.Time, online bool) error {
 	ms.cqmu.Lock()
 
 	if ms.changeQueueClosed {
@@ -199,13 +204,17 @@ func (ms *mapSession) updateDiscoForNode(id tailcfg.NodeID, key key.DiscoPublic,
 		return ErrChangeQueueClosed
 	}
 
-	resp := &tailcfg.MapResponse{
+	resp := responseWithSource{
-		PeersChangedPatch: []*tailcfg.PeerChange{{
+		response: &tailcfg.MapResponse{
-			NodeID:   id,
+			PeersChangedPatch: []*tailcfg.PeerChange{{
-			LastSeen: &lastSeen,
+				NodeID:   id,
-			Online:   &online,
+				Key:      &key,
-			DiscoKey: &key,
+				LastSeen: &lastSeen,
-		}},
+				Online:   &online,
+				DiscoKey: &discoKey,
+			}},
+		},
+		viaTSMP: true,
 	}
 	ms.changeQueue <- resp
 	ms.cqmu.Unlock()
@@ -221,7 +230,12 @@ func (ms *mapSession) HandleNonKeepAliveMapResponse(ctx context.Context, resp *t
 		return ErrChangeQueueClosed
 	}
 
-	ms.changeQueue <- resp
+	change := responseWithSource{
+		response: resp,
+		viaTSMP:  false,
+	}
+
+	ms.changeQueue <- change
 	ms.cqmu.Unlock()
 	return nil
 }
@@ -234,7 +248,7 @@ func (ms *mapSession) HandleNonKeepAliveMapResponse(ctx context.Context, resp *t
 //
 // TODO(bradfitz): make this handle all fields later. For now (2023-08-20) this
 // is [re]factoring progress enough.
-func (ms *mapSession) handleNonKeepAliveMapResponse(ctx context.Context, resp *tailcfg.MapResponse) error {
+func (ms *mapSession) handleNonKeepAliveMapResponse(ctx context.Context, resp *tailcfg.MapResponse, viaTSMP bool) error {
 	if debug := resp.Debug; debug != nil {
 		if err := ms.onDebug(ctx, debug); err != nil {
 			return err
@@ -284,6 +298,13 @@ func (ms *mapSession) handleNonKeepAliveMapResponse(ctx context.Context, resp *t
 
 	ms.updateStateFromResponse(resp)
 
+	// If source was learned via TSMP, the updated disco key need to be marked in
+	// userspaceEngine as an update that should not reconfigure the wireguard
+	// connection.
+	if viaTSMP {
+		ms.tryMarkDiscoAsLearnedFromTSMP(resp)
+	}
+
 	if ms.tryHandleIncrementally(resp) {
 		ms.occasionallyPrintSummary(ms.lastNetmapSummary)
 		return nil
@@ -312,6 +333,21 @@ func (ms *mapSession) handleNonKeepAliveMapResponse(ctx context.Context, resp *t
 	return nil
 }
 
+func (ms *mapSession) tryMarkDiscoAsLearnedFromTSMP(res *tailcfg.MapResponse) {
+	dun, ok := ms.netmapUpdater.(patchDiscoKeyer)
+	if !ok {
+		return
+	}
+
+	// In reality we should never really have more than one change here over TSMP.
+	for _, change := range res.PeersChangedPatch {
+		if change == nil || change.DiscoKey == nil || change.Key == nil {
+			continue
+		}
+		dun.PatchDiscoKey(*change.Key, *change.DiscoKey)
+	}
+}
+
 // upgradeNode upgrades Node fields from the server into the modern forms
 // not using deprecated fields.
 func upgradeNode(n *tailcfg.Node) {

control/controlclient/map_test.go

@@ -33,7 +33,9 @@ import (
 	"tailscale.com/util/eventbus/eventbustest"
 	"tailscale.com/util/mak"
 	"tailscale.com/util/must"
+	"tailscale.com/util/usermetric"
 	"tailscale.com/util/zstdframe"
+	"tailscale.com/wgengine"
 )
 
 func eps(s ...string) []netip.AddrPort {
@@ -678,6 +680,7 @@ func TestUpdateDiscoForNode(t *testing.T) {
 			// Insert existing node
 			node := tailcfg.Node{
 				ID:       1,
+				Key:      key.NewNode().Public(),
 				DiscoKey: oldKey.Public(),
 				Online:   &tt.initialOnline,
 				LastSeen: &tt.initialLastSeen,
@@ -690,7 +693,7 @@ func TestUpdateDiscoForNode(t *testing.T) {
 			}
 
 			newKey := key.NewDisco()
-			ms.updateDiscoForNode(node.ID, newKey.Public(), tt.updateLastSeen, tt.updateOnline)
+			ms.updateDiscoForNode(node.ID, node.Key, newKey.Public(), tt.updateLastSeen, tt.updateOnline)
 			<-nu.done
 
 			nm := ms.netmap()
@@ -707,6 +710,82 @@ func TestUpdateDiscoForNode(t *testing.T) {
 	}
 }
 
+func TestUpdateDiscoForNodeCallback(t *testing.T) {
+	t.Run("key_wired_through_to_updater", func(t *testing.T) {
+		nu := &rememberLastNetmapUpdater{
+			done: make(chan any, 1),
+		}
+		ms := newTestMapSession(t, nu)
+
+		oldKey := key.NewDisco()
+
+		// Insert existing node
+		node := tailcfg.Node{
+			ID:       1,
+			Key:      key.NewNode().Public(),
+			DiscoKey: oldKey.Public(),
+			Online:   new(false),
+			LastSeen: new(time.Unix(1, 0)),
+		}
+
+		if nm := ms.netmapForResponse(&tailcfg.MapResponse{
+			Peers: []*tailcfg.Node{&node},
+		}); len(nm.Peers) != 1 {
+			t.Fatalf("node not inserted")
+		}
+
+		newKey := key.NewDisco()
+		ms.updateDiscoForNode(node.ID, node.Key, newKey.Public(), time.Now(), false)
+		<-nu.done
+
+		if nu.lastTSMPKey != node.Key || nu.lastTSMPDisco != newKey.Public() {
+			t.Fatalf("expected [%s]=%s, got [%s]=%s", node.Key, newKey.Public(),
+				nu.lastTSMPKey, nu.lastTSMPDisco)
+		}
+	})
+	t.Run("key_not_wired_through_to_updater", func(t *testing.T) {
+		nu := &rememberLastNetmapUpdater{
+			done: make(chan any, 1),
+		}
+		ms := newTestMapSession(t, nu)
+
+		oldKey := key.NewDisco()
+
+		// Insert existing node
+		node := tailcfg.Node{
+			ID:       1,
+			Key:      key.NewNode().Public(),
+			DiscoKey: oldKey.Public(),
+			Online:   new(false),
+			LastSeen: new(time.Unix(1, 0)),
+		}
+
+		if nm := ms.netmapForResponse(&tailcfg.MapResponse{
+			Peers: []*tailcfg.Node{&node},
+		}); len(nm.Peers) != 1 {
+			t.Fatalf("node not inserted")
+		}
+
+		newKey := key.NewDisco().Public()
+		resp := &tailcfg.MapResponse{
+			PeersChangedPatch: []*tailcfg.PeerChange{{
+				NodeID:   node.ID,
+				Key:      &node.Key,
+				LastSeen: new(time.Now()),
+				Online:   new(true),
+				DiscoKey: &newKey,
+			}},
+		}
+		ms.HandleNonKeepAliveMapResponse(t.Context(), resp)
+		<-nu.done
+
+		if !nu.lastTSMPKey.IsZero() || !nu.lastTSMPDisco.IsZero() {
+			t.Fatalf("expected zero keys, got [%s]=%s",
+				nu.lastTSMPKey, nu.lastTSMPDisco)
+		}
+	})
+}
+
 func first[T any](s []T) T {
 	if len(s) == 0 {
 		var zero T
@@ -1568,3 +1647,22 @@ func TestLearnZstdOfKeepAlive(t *testing.T) {
 		t.Fatalf("got %d zstd decodes; want %d", got, want)
 	}
 }
+
+func TestPathDiscokeyerImplementations(t *testing.T) {
+	bus := eventbustest.NewBus(t)
+	ht := health.NewTracker(bus)
+	reg := new(usermetric.Registry)
+	e, err := wgengine.NewFakeUserspaceEngine(t.Logf, 0, ht, reg, bus)
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(e.Close)
+	if _, ok := e.(patchDiscoKeyer); !ok {
+		t.Error("wgengine.userspaceEngine must implement patchDiscoKeyer")
+	}
+
+	wd := wgengine.NewWatchdog(e)
+	if _, ok := wd.(patchDiscoKeyer); !ok {
+		t.Error("wgengine.watchdogEngine must implement patchDiscoKeyer")
+	}
+}

ipn/ipnlocal/local.go

@@ -1857,6 +1857,18 @@ func (b *LocalBackend) setControlClientStatusLocked(c controlclient.Client, st c
 	b.authReconfigLocked()
 }
 
+func (b *LocalBackend) PatchDiscoKey(pub key.NodePublic, disco key.DiscoPublic) {
+	// PatchDiscoKey mirrors the implementation of [controlclient.patchDiscoKeyer].
+	// It is implemented here to avoid the dependency edge to controlclient, but must be kept
+	// in sync with the original implementation.
+	type patchDiscoKeyer interface {
+		PatchDiscoKey(key.NodePublic, key.DiscoPublic)
+	}
+	if e, ok := b.e.(patchDiscoKeyer); ok {
+		e.PatchDiscoKey(pub, disco)
+	}
+}
+
 type preferencePolicyInfo struct {
 	key pkey.Key
 	get func(ipn.PrefsView) bool

wgengine/userspace.go

@@ -121,7 +121,8 @@ type userspaceEngine struct {
 	birdClient     BIRDClient          // or nil
 	controlKnobs   *controlknobs.Knobs // or nil
 
-	testMaybeReconfigHook func() // for tests; if non-nil, fires if maybeReconfigWireguardLocked called
+	testMaybeReconfigHook func()                        // for tests; if non-nil, fires if maybeReconfigWireguardLocked called
+	testDiscoChangedHook  func(map[key.NodePublic]bool) // for tests; if non-nil, fires after assembling discoChanged map
 
 	// isLocalAddr reports the whether an IP is assigned to the local
 	// tunnel interface. It's used to reflect local packets
@@ -167,6 +168,10 @@ type userspaceEngine struct {
 	// networkLogger logs statistics about network connections.
 	networkLogger netlog.Logger
 
+	// tsmpLearnedDisco tracks per node key if a peer disco key was learned via TSMP.
+	// wgLock must be held when using this map.
+	tsmpLearnedDisco map[key.NodePublic]key.DiscoPublic
+
 	// Lock ordering: magicsock.Conn.mu, wgLock, then mu.
 }
 
@@ -1028,6 +1033,12 @@ func (e *userspaceEngine) ResetAndStop() (*Status, error) {
 	}
 }
 
+func (e *userspaceEngine) PatchDiscoKey(pub key.NodePublic, disco key.DiscoPublic) {
+	e.wgLock.Lock()
+	defer e.wgLock.Unlock()
+	mak.Set(&e.tsmpLearnedDisco, pub, disco)
+}
+
 func (e *userspaceEngine) Reconfig(cfg *wgcfg.Config, routerCfg *router.Config, dnsCfg *dns.Config) error {
 	if routerCfg == nil {
 		panic("routerCfg must not be nil")
@@ -1119,14 +1130,31 @@ func (e *userspaceEngine) Reconfig(cfg *wgcfg.Config, routerCfg *router.Config,
 			if p.DiscoKey.IsZero() {
 				continue
 			}
+
+			// If the key changed, mark the connection for reconfiguration.
 			pub := p.PublicKey
 			if old, ok := prevEP[pub]; ok && old != p.DiscoKey {
+				// If the disco key was learned via TSMP, we do not need to reset the
+				// wireguard config as the new key was received over an existing wireguard
+				// connection.
+				if discoTSMP, okTSMP := e.tsmpLearnedDisco[p.PublicKey]; okTSMP &&
+					discoTSMP == p.DiscoKey {
+					delete(e.tsmpLearnedDisco, p.PublicKey)
+					e.logf("wgengine: Skipping reconfig (TSMP key): %s changed from %q to %q", pub.ShortString(), old, p.DiscoKey)
+					continue
+				}
+
 				discoChanged[pub] = true
 				e.logf("wgengine: Reconfig: %s changed from %q to %q", pub.ShortString(), old, p.DiscoKey)
 			}
 		}
 	}
 
+	// For tests, what disco connections needs to be changed.
+	if e.testDiscoChangedHook != nil {
+		e.testDiscoChangedHook(discoChanged)
+	}
+
 	e.lastCfgFull = *cfg.Clone()
 
 	// Tell magicsock about the new (or initial) private key
@@ -1144,6 +1172,13 @@ func (e *userspaceEngine) Reconfig(cfg *wgcfg.Config, routerCfg *router.Config,
 		return err
 	}
 
+	// Cleanup map of tsmp marks for peers that no longer exists in config.
+	for nodeKey := range e.tsmpLearnedDisco {
+		if !peerSet.Contains(nodeKey) {
+			delete(e.tsmpLearnedDisco, nodeKey)
+		}
+	}
+
 	// Shutdown the network logger because the IDs changed.
 	// Let it be started back up by subsequent logic.
 	if buildfeatures.HasNetLog && netLogIDsChanged && e.networkLogger.Running() {

wgengine/userspace_test.go

@@ -164,6 +164,79 @@ func TestUserspaceEngineReconfig(t *testing.T) {
 	}
 }
 
+func TestUserspaceEngineTSMPLearned(t *testing.T) {
+	bus := eventbustest.NewBus(t)
+
+	ht := health.NewTracker(bus)
+	reg := new(usermetric.Registry)
+	e, err := NewFakeUserspaceEngine(t.Logf, 0, ht, reg, bus)
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(e.Close)
+	ue := e.(*userspaceEngine)
+
+	discoChangedChan := make(chan map[key.NodePublic]bool, 1)
+	ue.testDiscoChangedHook = func(m map[key.NodePublic]bool) {
+		discoChangedChan <- m
+	}
+
+	routerCfg := &router.Config{}
+
+	keyChanges := []struct {
+		tsmp  bool
+		inMap bool
+	}{
+		{tsmp: false, inMap: false},
+		{tsmp: true, inMap: false},
+		{tsmp: false, inMap: true},
+	}
+
+	nkHex := "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
+	for _, change := range keyChanges {
+		oldDisco := key.NewDisco()
+		nm := &netmap.NetworkMap{
+			Peers: nodeViews([]*tailcfg.Node{
+				{
+					ID:       1,
+					Key:      nkFromHex(nkHex),
+					DiscoKey: oldDisco.Public(),
+				},
+			}),
+		}
+		nk, err := key.ParseNodePublicUntyped(mem.S(nkHex))
+		if err != nil {
+			t.Fatal(err)
+		}
+		e.SetNetworkMap(nm)
+
+		newDisco := key.NewDisco()
+		cfg := &wgcfg.Config{
+			Peers: []wgcfg.Peer{
+				{
+					PublicKey: nk,
+					DiscoKey:  newDisco.Public(),
+				},
+			},
+		}
+
+		if change.tsmp {
+			ue.PatchDiscoKey(nk, newDisco.Public())
+		}
+		err = e.Reconfig(cfg, routerCfg, &dns.Config{})
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		changeMap := <-discoChangedChan
+
+		if _, ok := changeMap[nk]; ok != change.inMap {
+			t.Fatalf("expect key %v in map %v to be %t, got %t", nk, changeMap,
+				change.inMap, ok)
+		}
+	}
+}
+
 func TestUserspaceEnginePortReconfig(t *testing.T) {
 	flakytest.Mark(t, "https://github.com/tailscale/tailscale/issues/2855")
 	const defaultPort = 49983

wgengine/watchdog.go

@@ -242,3 +242,15 @@ func (e *watchdogEngine) InstallCaptureHook(cb packet.CaptureCallback) {
 func (e *watchdogEngine) PeerByKey(pubKey key.NodePublic) (_ wgint.Peer, ok bool) {
 	return e.wrap.PeerByKey(pubKey)
 }
+
+func (e *watchdogEngine) PatchDiscoKey(pub key.NodePublic, disco key.DiscoPublic) {
+	// PatchDiscoKey mirrors the implementation of [controlclient.patchDiscoKeyer ].
+	// It is implemented here to avoid the dependency edge to controlclient, but must be kept
+	// in sync with the original implementation.
+	type patchDiscoKeyer interface {
+		PatchDiscoKey(key.NodePublic, key.DiscoPublic)
+	}
+	if n, ok := e.wrap.(patchDiscoKeyer); ok {
+		n.PatchDiscoKey(pub, disco)
+	}
+}