Merge branch 'main' into cached-serial-check-on-sync

client/internal/connect.go

@@ -94,6 +94,7 @@ func (c *ConnectClient) RunOnAndroid(
 	dnsAddresses []netip.AddrPort,
 	dnsReadyListener dns.ReadyListener,
 	stateFilePath string,
+	cacheDir string,
 ) error {
 	// in case of non Android os these variables will be nil
 	mobileDependency := MobileDependency{
@@ -103,6 +104,7 @@ func (c *ConnectClient) RunOnAndroid(
 		HostDNSAddresses:      dnsAddresses,
 		DnsReadyListener:      dnsReadyListener,
 		StateFilePath:         stateFilePath,
+		TempDir:               cacheDir,
 	}
 	return c.run(mobileDependency, nil, "")
 }
@@ -338,6 +340,7 @@ func (c *ConnectClient) run(mobileDependency MobileDependency, runningChan chan
 			log.Error(err)
 			return wrapErr(err)
 		}
+		engineConfig.TempDir = mobileDependency.TempDir
 
 		relayManager := relayClient.NewManager(engineCtx, relayURLs, myPrivateKey.PublicKey().String(), engineConfig.MTU)
 		c.statusRecorder.SetRelayMgr(relayManager)

client/internal/debug/debug.go

@@ -16,7 +16,6 @@ import (
 	"path/filepath"
 	"runtime"
 	"runtime/pprof"
-	"slices"
 	"sort"
 	"strings"
 	"time"
@@ -31,7 +30,6 @@ import (
 	"github.com/netbirdio/netbird/client/internal/updater/installer"
 	nbstatus "github.com/netbirdio/netbird/client/status"
 	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
-	"github.com/netbirdio/netbird/util"
 )
 
 const readmeContent = `Netbird debug bundle
@@ -234,6 +232,7 @@ type BundleGenerator struct {
 	statusRecorder *peer.Status
 	syncResponse   *mgmProto.SyncResponse
 	logPath        string
+	tempDir        string
 	cpuProfile     []byte
 	refreshStatus  func() // Optional callback to refresh status before bundle generation
 	clientMetrics  MetricsExporter
@@ -256,6 +255,7 @@ type GeneratorDependencies struct {
 	StatusRecorder *peer.Status
 	SyncResponse   *mgmProto.SyncResponse
 	LogPath        string
+	TempDir        string // Directory for temporary bundle zip files. If empty, os.TempDir() is used.
 	CPUProfile     []byte
 	RefreshStatus  func() // Optional callback to refresh status before bundle generation
 	ClientMetrics  MetricsExporter
@@ -275,6 +275,7 @@ func NewBundleGenerator(deps GeneratorDependencies, cfg BundleConfig) *BundleGen
 		statusRecorder: deps.StatusRecorder,
 		syncResponse:   deps.SyncResponse,
 		logPath:        deps.LogPath,
+		tempDir:        deps.TempDir,
 		cpuProfile:     deps.CPUProfile,
 		refreshStatus:  deps.RefreshStatus,
 		clientMetrics:  deps.ClientMetrics,
@@ -287,7 +288,7 @@ func NewBundleGenerator(deps GeneratorDependencies, cfg BundleConfig) *BundleGen
 
 // Generate creates a debug bundle and returns the location.
 func (g *BundleGenerator) Generate() (resp string, err error) {
-	bundlePath, err := os.CreateTemp("", "netbird.debug.*.zip")
+	bundlePath, err := os.CreateTemp(g.tempDir, "netbird.debug.*.zip")
 	if err != nil {
 		return "", fmt.Errorf("create zip file: %w", err)
 	}
@@ -373,15 +374,8 @@ func (g *BundleGenerator) createArchive() error {
 		log.Errorf("failed to add wg show output: %v", err)
 	}
 
-	if g.logPath != "" && !slices.Contains(util.SpecialLogs, g.logPath) {
-		if err := g.addLogfile(); err != nil {
-			log.Errorf("failed to add log file to debug bundle: %v", err)
-			if err := g.trySystemdLogFallback(); err != nil {
-				log.Errorf("failed to add systemd logs as fallback: %v", err)
-			}
-		}
-	} else if err := g.trySystemdLogFallback(); err != nil {
-		log.Errorf("failed to add systemd logs: %v", err)
+	if err := g.addPlatformLog(); err != nil {
+		log.Errorf("failed to add logs to debug bundle: %v", err)
 	}
 
 	if err := g.addUpdateLogs(); err != nil {

client/internal/debug/debug_android.go

@@ -0,0 +1,41 @@
+//go:build android
+
+package debug
+
+import (
+	"fmt"
+	"io"
+	"os/exec"
+
+	log "github.com/sirupsen/logrus"
+)
+
+func (g *BundleGenerator) addPlatformLog() error {
+	cmd := exec.Command("/system/bin/logcat", "-d")
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		return fmt.Errorf("logcat stdout pipe: %w", err)
+	}
+
+	if err := cmd.Start(); err != nil {
+		return fmt.Errorf("start logcat: %w", err)
+	}
+
+	var logReader io.Reader = stdout
+	if g.anonymize {
+		var pw *io.PipeWriter
+		logReader, pw = io.Pipe()
+		go anonymizeLog(stdout, pw, g.anonymizer)
+	}
+
+	if err := g.addFileToZip(logReader, "logcat.txt"); err != nil {
+		return fmt.Errorf("add logcat to zip: %w", err)
+	}
+
+	if err := cmd.Wait(); err != nil {
+		return fmt.Errorf("wait logcat: %w", err)
+	}
+
+	log.Debug("added logcat output to debug bundle")
+	return nil
+}

client/internal/debug/debug_nonandroid.go

@@ -0,0 +1,25 @@
+//go:build !android
+
+package debug
+
+import (
+	"slices"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/util"
+)
+
+func (g *BundleGenerator) addPlatformLog() error {
+	if g.logPath != "" && !slices.Contains(util.SpecialLogs, g.logPath) {
+		if err := g.addLogfile(); err != nil {
+			log.Errorf("failed to add log file to debug bundle: %v", err)
+			if err := g.trySystemdLogFallback(); err != nil {
+				return err
+			}
+		}
+	} else if err := g.trySystemdLogFallback(); err != nil {
+		return err
+	}
+	return nil
+}

client/internal/engine.go

@@ -140,6 +140,7 @@ type EngineConfig struct {
 	ProfileConfig *profilemanager.Config
 
 	LogPath string
+	TempDir string
 }
 
 // EngineServices holds the external service dependencies required by the Engine.
@@ -569,7 +570,7 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, mgmtURL *url.URL)
 	e.connMgr.Start(e.ctx)
 
 	e.srWatcher = guard.NewSRWatcher(e.signal, e.relayManager, e.mobileDep.IFaceDiscover, iceCfg)
-	e.srWatcher.Start()
+	e.srWatcher.Start(peer.IsForceRelayed())
 
 	e.receiveSignalEvents()
 	e.receiveManagementEvents()
@@ -1095,6 +1096,7 @@ func (e *Engine) handleBundle(params *mgmProto.BundleParameters) (*mgmProto.JobR
 		StatusRecorder: e.statusRecorder,
 		SyncResponse:   syncResponse,
 		LogPath:        e.config.LogPath,
+		TempDir:        e.config.TempDir,
 		ClientMetrics:  e.clientMetrics,
 		RefreshStatus: func() {
 			e.RunHealthProbes(true)

client/internal/mobile_dependency.go

@@ -22,4 +22,8 @@ type MobileDependency struct {
 	DnsManager     dns.IosDnsManager
 	FileDescriptor int32
 	StateFilePath  string
+
+	// TempDir is a writable directory for temporary files (e.g., debug bundle zip).
+	// On Android, this should be set to the app's cache directory.
+	TempDir string
 }

client/internal/peer/conn.go

@@ -185,17 +185,20 @@ func (conn *Conn) Open(engineCtx context.Context) error {
 
 	conn.workerRelay = NewWorkerRelay(conn.ctx, conn.Log, isController(conn.config), conn.config, conn, conn.relayManager)
 
-	relayIsSupportedLocally := conn.workerRelay.RelayIsSupportedLocally()
-	workerICE, err := NewWorkerICE(conn.ctx, conn.Log, conn.config, conn, conn.signaler, conn.iFaceDiscover, conn.statusRecorder, relayIsSupportedLocally)
-	if err != nil {
-		return err
+	forceRelay := IsForceRelayed()
+	if !forceRelay {
+		relayIsSupportedLocally := conn.workerRelay.RelayIsSupportedLocally()
+		workerICE, err := NewWorkerICE(conn.ctx, conn.Log, conn.config, conn, conn.signaler, conn.iFaceDiscover, conn.statusRecorder, relayIsSupportedLocally)
+		if err != nil {
+			return err
+		}
+		conn.workerICE = workerICE
 	}
-	conn.workerICE = workerICE
 
 	conn.handshaker = NewHandshaker(conn.Log, conn.config, conn.signaler, conn.workerICE, conn.workerRelay, conn.metricsStages)
 
 	conn.handshaker.AddRelayListener(conn.workerRelay.OnNewOffer)
-	if !isForceRelayed() {
+	if !forceRelay {
 		conn.handshaker.AddICEListener(conn.workerICE.OnNewOffer)
 	}
 
@@ -251,7 +254,9 @@ func (conn *Conn) Close(signalToRemote bool) {
 		conn.wgWatcherCancel()
 	}
 	conn.workerRelay.CloseConn()
-	conn.workerICE.Close()
+	if conn.workerICE != nil {
+		conn.workerICE.Close()
+	}
 
 	if conn.wgProxyRelay != nil {
 		err := conn.wgProxyRelay.CloseConn()
@@ -294,7 +299,9 @@ func (conn *Conn) OnRemoteAnswer(answer OfferAnswer) {
 // OnRemoteCandidate Handles ICE connection Candidate provided by the remote peer.
 func (conn *Conn) OnRemoteCandidate(candidate ice.Candidate, haRoutes route.HAMap) {
 	conn.dumpState.RemoteCandidate()
-	conn.workerICE.OnRemoteCandidate(candidate, haRoutes)
+	if conn.workerICE != nil {
+		conn.workerICE.OnRemoteCandidate(candidate, haRoutes)
+	}
 }
 
 // SetOnConnected sets a handler function to be triggered by Conn when a new connection to a remote peer established
@@ -712,33 +719,35 @@ func (conn *Conn) evalStatus() ConnStatus {
 	return StatusConnecting
 }
 
-func (conn *Conn) isConnectedOnAllWay() (connected bool) {
-	// would be better to protect this with a mutex, but it could cause deadlock with Close function
-
+// isConnectedOnAllWay evaluates the overall connection status based on ICE and Relay transports.
+//
+// The result is a tri-state:
+//   - ConnStatusConnected:          all available transports are up
+//   - ConnStatusPartiallyConnected: relay is up but ICE is still pending/reconnecting
+//   - ConnStatusDisconnected:       no working transport
+func (conn *Conn) isConnectedOnAllWay() (status guard.ConnStatus) {
 	defer func() {
-		if !connected {
+		if status == guard.ConnStatusDisconnected {
 			conn.logTraceConnState()
 		}
 	}()
 
-	// For JS platform: only relay connection is supported
-	if runtime.GOOS == "js" {
-		return conn.statusRelay.Get() == worker.StatusConnected
+	iceWorkerCreated := conn.workerICE != nil
+
+	var iceInProgress bool
+	if iceWorkerCreated {
+		iceInProgress = conn.workerICE.InProgress()
 	}
 
-	// For non-JS platforms: check ICE connection status
-	if conn.statusICE.Get() == worker.StatusDisconnected && !conn.workerICE.InProgress() {
-		return false
-	}
-
-	// If relay is supported with peer, it must also be connected
-	if conn.workerRelay.IsRelayConnectionSupportedWithPeer() {
-		if conn.statusRelay.Get() == worker.StatusDisconnected {
-			return false
-		}
-	}
-
-	return true
+	return evalConnStatus(connStatusInputs{
+		forceRelay:          IsForceRelayed(),
+		peerUsesRelay:       conn.workerRelay.IsRelayConnectionSupportedWithPeer(),
+		relayConnected:      conn.statusRelay.Get() == worker.StatusConnected,
+		remoteSupportsICE:   conn.handshaker.RemoteICESupported(),
+		iceWorkerCreated:    iceWorkerCreated,
+		iceStatusConnecting: conn.statusICE.Get() != worker.StatusDisconnected,
+		iceInProgress:       iceInProgress,
+	})
 }
 
 func (conn *Conn) enableWgWatcherIfNeeded(enabledTime time.Time) {
@@ -926,3 +935,43 @@ func isController(config ConnConfig) bool {
 func isRosenpassEnabled(remoteRosenpassPubKey []byte) bool {
 	return remoteRosenpassPubKey != nil
 }
+
+func evalConnStatus(in connStatusInputs) guard.ConnStatus {
+	// "Relay up and needed" — the peer uses relay and the transport is connected.
+	relayUsedAndUp := in.peerUsesRelay && in.relayConnected
+
+	// Force-relay mode: ICE never runs. Relay is the only transport and must be up.
+	if in.forceRelay {
+		return boolToConnStatus(relayUsedAndUp)
+	}
+
+	// Remote peer doesn't support ICE, or we haven't created the worker yet:
+	// relay is the only possible transport.
+	if !in.remoteSupportsICE || !in.iceWorkerCreated {
+		return boolToConnStatus(relayUsedAndUp)
+	}
+
+	// ICE counts as "up" when the status is anything other than Disconnected, OR
+	// when a negotiation is currently in progress (so we don't spam offers while one is in flight).
+	iceUp := in.iceStatusConnecting || in.iceInProgress
+
+	// Relay side is acceptable if the peer doesn't rely on relay, or relay is connected.
+	relayOK := !in.peerUsesRelay || in.relayConnected
+
+	switch {
+	case iceUp && relayOK:
+		return guard.ConnStatusConnected
+	case relayUsedAndUp:
+		// Relay is up but ICE is down — partially connected.
+		return guard.ConnStatusPartiallyConnected
+	default:
+		return guard.ConnStatusDisconnected
+	}
+}
+
+func boolToConnStatus(connected bool) guard.ConnStatus {
+	if connected {
+		return guard.ConnStatusConnected
+	}
+	return guard.ConnStatusDisconnected
+}

client/internal/peer/conn_status.go

@@ -13,6 +13,20 @@ const (
 	StatusConnected
 )
 
+// connStatusInputs is the primitive-valued snapshot of the state that drives the
+// tri-state connection classification. Extracted so the decision logic can be unit-tested
+// without constructing full Worker/Handshaker objects.
+type connStatusInputs struct {
+	forceRelay          bool // NB_FORCE_RELAY or JS/WASM
+	peerUsesRelay       bool // remote peer advertises relay support AND local has relay
+	relayConnected      bool // statusRelay reports Connected (independent of whether peer uses relay)
+	remoteSupportsICE   bool // remote peer sent ICE credentials
+	iceWorkerCreated    bool // local WorkerICE exists (false in force-relay mode)
+	iceStatusConnecting bool // statusICE is anything other than Disconnected
+	iceInProgress       bool // a negotiation is currently in flight
+}
+
+
 // ConnStatus describe the status of a peer's connection
 type ConnStatus int32
 

client/internal/peer/conn_status_eval_test.go

@@ -0,0 +1,201 @@
+package peer
+
+import (
+	"testing"
+
+	"github.com/netbirdio/netbird/client/internal/peer/guard"
+)
+
+func TestEvalConnStatus_ForceRelay(t *testing.T) {
+	tests := []struct {
+		name string
+		in   connStatusInputs
+		want guard.ConnStatus
+	}{
+		{
+			name: "force relay, peer uses relay, relay up",
+			in: connStatusInputs{
+				forceRelay:     true,
+				peerUsesRelay:  true,
+				relayConnected: true,
+			},
+			want: guard.ConnStatusConnected,
+		},
+		{
+			name: "force relay, peer uses relay, relay down",
+			in: connStatusInputs{
+				forceRelay:     true,
+				peerUsesRelay:  true,
+				relayConnected: false,
+			},
+			want: guard.ConnStatusDisconnected,
+		},
+		{
+			name: "force relay, peer does NOT use relay - disconnected forever",
+			in: connStatusInputs{
+				forceRelay:     true,
+				peerUsesRelay:  false,
+				relayConnected: true,
+			},
+			want: guard.ConnStatusDisconnected,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := evalConnStatus(tc.in); got != tc.want {
+				t.Fatalf("evalConnStatus = %v, want %v", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestEvalConnStatus_ICEUnavailable(t *testing.T) {
+	tests := []struct {
+		name string
+		in   connStatusInputs
+		want guard.ConnStatus
+	}{
+		{
+			name: "remote does not support ICE, peer uses relay, relay up",
+			in: connStatusInputs{
+				peerUsesRelay:     true,
+				relayConnected:    true,
+				remoteSupportsICE: false,
+				iceWorkerCreated:  true,
+			},
+			want: guard.ConnStatusConnected,
+		},
+		{
+			name: "remote does not support ICE, peer uses relay, relay down",
+			in: connStatusInputs{
+				peerUsesRelay:     true,
+				relayConnected:    false,
+				remoteSupportsICE: false,
+				iceWorkerCreated:  true,
+			},
+			want: guard.ConnStatusDisconnected,
+		},
+		{
+			name: "ICE worker not yet created, relay up",
+			in: connStatusInputs{
+				peerUsesRelay:     true,
+				relayConnected:    true,
+				remoteSupportsICE: true,
+				iceWorkerCreated:  false,
+			},
+			want: guard.ConnStatusConnected,
+		},
+		{
+			name: "remote does not support ICE, peer does not use relay",
+			in: connStatusInputs{
+				peerUsesRelay:     false,
+				relayConnected:    false,
+				remoteSupportsICE: false,
+				iceWorkerCreated:  true,
+			},
+			want: guard.ConnStatusDisconnected,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := evalConnStatus(tc.in); got != tc.want {
+				t.Fatalf("evalConnStatus = %v, want %v", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestEvalConnStatus_FullyAvailable(t *testing.T) {
+	base := connStatusInputs{
+		remoteSupportsICE: true,
+		iceWorkerCreated:  true,
+	}
+
+	tests := []struct {
+		name    string
+		mutator func(*connStatusInputs)
+		want    guard.ConnStatus
+	}{
+		{
+			name: "ICE connected, relay connected, peer uses relay",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = true
+				in.relayConnected = true
+				in.iceStatusConnecting = true
+			},
+			want: guard.ConnStatusConnected,
+		},
+		{
+			name: "ICE connected, peer does NOT use relay",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = false
+				in.relayConnected = false
+				in.iceStatusConnecting = true
+			},
+			want: guard.ConnStatusConnected,
+		},
+		{
+			name: "ICE InProgress only, peer does NOT use relay",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = false
+				in.iceStatusConnecting = false
+				in.iceInProgress = true
+			},
+			want: guard.ConnStatusConnected,
+		},
+		{
+			name: "ICE down, relay up, peer uses relay -> partial",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = true
+				in.relayConnected = true
+				in.iceStatusConnecting = false
+				in.iceInProgress = false
+			},
+			want: guard.ConnStatusPartiallyConnected,
+		},
+		{
+			name: "ICE down, peer does NOT use relay -> disconnected",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = false
+				in.relayConnected = false
+				in.iceStatusConnecting = false
+				in.iceInProgress = false
+			},
+			want: guard.ConnStatusDisconnected,
+		},
+		{
+			name: "ICE up, peer uses relay but relay down -> partial (relay required, ICE ignored)",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = true
+				in.relayConnected = false
+				in.iceStatusConnecting = true
+			},
+			// relayOK = false (peer uses relay but it's down), iceUp = true
+			// first switch arm fails (relayOK false), relayUsedAndUp = false (relay down),
+			// falls into default: Disconnected.
+			want: guard.ConnStatusDisconnected,
+		},
+		{
+			name: "ICE down, relay up but peer does not use relay -> disconnected",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = false
+				in.relayConnected = true // not actually used since peer doesn't rely on it
+				in.iceStatusConnecting = false
+				in.iceInProgress = false
+			},
+			want: guard.ConnStatusDisconnected,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			in := base
+			tc.mutator(&in)
+			if got := evalConnStatus(in); got != tc.want {
+				t.Fatalf("evalConnStatus = %v, want %v (inputs: %+v)", got, tc.want, in)
+			}
+		})
+	}
+}

client/internal/peer/env.go

@@ -10,7 +10,7 @@ const (
 	EnvKeyNBForceRelay = "NB_FORCE_RELAY"
 )
 
-func isForceRelayed() bool {
+func IsForceRelayed() bool {
 	if runtime.GOOS == "js" {
 		return true
 	}

client/internal/peer/guard/guard.go

@@ -8,7 +8,19 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
-type isConnectedFunc func() bool
+// ConnStatus represents the connection state as seen by the guard.
+type ConnStatus int
+
+const (
+	// ConnStatusDisconnected means neither ICE nor Relay is connected.
+	ConnStatusDisconnected ConnStatus = iota
+	// ConnStatusPartiallyConnected means Relay is connected but ICE is not.
+	ConnStatusPartiallyConnected
+	// ConnStatusConnected means all required connections are established.
+	ConnStatusConnected
+)
+
+type connStatusFunc func() ConnStatus
 
 // Guard is responsible for the reconnection logic.
 // It will trigger to send an offer to the peer then has connection issues.
@@ -20,14 +32,14 @@ type isConnectedFunc func() bool
 // - ICE candidate changes
 type Guard struct {
 	log                     *log.Entry
-	isConnectedOnAllWay     isConnectedFunc
+	isConnectedOnAllWay     connStatusFunc
 	timeout                 time.Duration
 	srWatcher               *SRWatcher
 	relayedConnDisconnected chan struct{}
 	iCEConnDisconnected     chan struct{}
 }
 
-func NewGuard(log *log.Entry, isConnectedFn isConnectedFunc, timeout time.Duration, srWatcher *SRWatcher) *Guard {
+func NewGuard(log *log.Entry, isConnectedFn connStatusFunc, timeout time.Duration, srWatcher *SRWatcher) *Guard {
 	return &Guard{
 		log:                     log,
 		isConnectedOnAllWay:     isConnectedFn,
@@ -57,8 +69,17 @@ func (g *Guard) SetICEConnDisconnected() {
 	}
 }
 
-// reconnectLoopWithRetry periodically check the connection status.
-// Try to send offer while the P2P is not established or while the Relay is not connected if is it supported
+// reconnectLoopWithRetry periodically checks the connection status and sends offers to re-establish connectivity.
+//
+// Behavior depends on the connection state reported by isConnectedOnAllWay:
+//   - Connected: no action, the peer is fully reachable.
+//   - Disconnected (neither ICE nor Relay): retries aggressively with exponential backoff (800ms doubling
+//     up to timeout), never gives up. This ensures rapid recovery when the peer has no connectivity at all.
+//   - PartiallyConnected (Relay up, ICE not): retries up to 3 times with exponential backoff, then switches
+//     to one attempt per hour. This limits signaling traffic when relay already provides connectivity.
+//
+// External events (relay/ICE disconnect, signal/relay reconnect, candidate changes) reset the retry
+// counter and backoff ticker, giving ICE a fresh chance after network conditions change.
 func (g *Guard) reconnectLoopWithRetry(ctx context.Context, callback func()) {
 	srReconnectedChan := g.srWatcher.NewListener()
 	defer g.srWatcher.RemoveListener(srReconnectedChan)
@@ -68,36 +89,47 @@ func (g *Guard) reconnectLoopWithRetry(ctx context.Context, callback func()) {
 
 	tickerChannel := ticker.C
 
+	iceState := &iceRetryState{log: g.log}
+	defer iceState.reset()
+
 	for {
 		select {
-		case t := <-tickerChannel:
-			if t.IsZero() {
-				g.log.Infof("retry timed out, stop periodic offer sending")
-				// after backoff timeout the ticker.C will be closed. We need to a dummy channel to avoid loop
-				tickerChannel = make(<-chan time.Time)
-				continue
+		case <-tickerChannel:
+			switch g.isConnectedOnAllWay() {
+			case ConnStatusConnected:
+				// all good, nothing to do
+			case ConnStatusDisconnected:
+				callback()
+			case ConnStatusPartiallyConnected:
+				if iceState.shouldRetry() {
+					callback()
+				} else {
+					iceState.enterHourlyMode()
+					ticker.Stop()
+					tickerChannel = iceState.hourlyC()
+				}
 			}
 
-			if !g.isConnectedOnAllWay() {
-				callback()
-			}
 		case <-g.relayedConnDisconnected:
 			g.log.Debugf("Relay connection changed, reset reconnection ticker")
 			ticker.Stop()
-			ticker = g.prepareExponentTicker(ctx)
+			ticker = g.newReconnectTicker(ctx)
 			tickerChannel = ticker.C
+			iceState.reset()
 
 		case <-g.iCEConnDisconnected:
 			g.log.Debugf("ICE connection changed, reset reconnection ticker")
 			ticker.Stop()
-			ticker = g.prepareExponentTicker(ctx)
+			ticker = g.newReconnectTicker(ctx)
 			tickerChannel = ticker.C
+			iceState.reset()
 
 		case <-srReconnectedChan:
 			g.log.Debugf("has network changes, reset reconnection ticker")
 			ticker.Stop()
-			ticker = g.prepareExponentTicker(ctx)
+			ticker = g.newReconnectTicker(ctx)
 			tickerChannel = ticker.C
+			iceState.reset()
 
 		case <-ctx.Done():
 			g.log.Debugf("context is done, stop reconnect loop")
@@ -120,7 +152,7 @@ func (g *Guard) initialTicker(ctx context.Context) *backoff.Ticker {
 	return backoff.NewTicker(bo)
 }
 
-func (g *Guard) prepareExponentTicker(ctx context.Context) *backoff.Ticker {
+func (g *Guard) newReconnectTicker(ctx context.Context) *backoff.Ticker {
 	bo := backoff.WithContext(&backoff.ExponentialBackOff{
 		InitialInterval:     800 * time.Millisecond,
 		RandomizationFactor: 0.1,

client/internal/peer/guard/ice_retry_state.go

@@ -0,0 +1,61 @@
+package guard
+
+import (
+	"time"
+
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	// maxICERetries is the maximum number of ICE offer attempts when relay is connected
+	maxICERetries = 3
+	// iceRetryInterval is the periodic retry interval after ICE retries are exhausted
+	iceRetryInterval = 1 * time.Hour
+)
+
+// iceRetryState tracks the limited ICE retry attempts when relay is already connected.
+// After maxICERetries attempts it switches to a periodic hourly retry.
+type iceRetryState struct {
+	log     *log.Entry
+	retries int
+	hourly  *time.Ticker
+}
+
+func (s *iceRetryState) reset() {
+	s.retries = 0
+	if s.hourly != nil {
+		s.hourly.Stop()
+		s.hourly = nil
+	}
+}
+
+// shouldRetry reports whether the caller should send another ICE offer on this tick.
+// Returns false when the per-cycle retry budget is exhausted and the caller must switch
+// to the hourly ticker via enterHourlyMode + hourlyC.
+func (s *iceRetryState) shouldRetry() bool {
+	if s.hourly != nil {
+		s.log.Debugf("hourly ICE retry attempt")
+		return true
+	}
+
+	s.retries++
+	if s.retries <= maxICERetries {
+		s.log.Debugf("ICE retry attempt %d/%d", s.retries, maxICERetries)
+		return true
+	}
+
+	return false
+}
+
+// enterHourlyMode starts the hourly retry ticker. Must be called after shouldRetry returns false.
+func (s *iceRetryState) enterHourlyMode() {
+	s.log.Infof("ICE retries exhausted (%d/%d), switching to hourly retry", maxICERetries, maxICERetries)
+	s.hourly = time.NewTicker(iceRetryInterval)
+}
+
+func (s *iceRetryState) hourlyC() <-chan time.Time {
+	if s.hourly == nil {
+		return nil
+	}
+	return s.hourly.C
+}

client/internal/peer/guard/ice_retry_state_test.go

@@ -0,0 +1,103 @@
+package guard
+
+import (
+	"testing"
+
+	log "github.com/sirupsen/logrus"
+)
+
+func newTestRetryState() *iceRetryState {
+	return &iceRetryState{log: log.NewEntry(log.StandardLogger())}
+}
+
+func TestICERetryState_AllowsInitialBudget(t *testing.T) {
+	s := newTestRetryState()
+
+	for i := 1; i <= maxICERetries; i++ {
+		if !s.shouldRetry() {
+			t.Fatalf("shouldRetry returned false on attempt %d, want true (budget = %d)", i, maxICERetries)
+		}
+	}
+}
+
+func TestICERetryState_ExhaustsAfterBudget(t *testing.T) {
+	s := newTestRetryState()
+
+	for i := 0; i < maxICERetries; i++ {
+		_ = s.shouldRetry()
+	}
+
+	if s.shouldRetry() {
+		t.Fatalf("shouldRetry returned true after budget exhausted, want false")
+	}
+}
+
+func TestICERetryState_HourlyCNilBeforeEnterHourlyMode(t *testing.T) {
+	s := newTestRetryState()
+
+	if s.hourlyC() != nil {
+		t.Fatalf("hourlyC returned non-nil channel before enterHourlyMode")
+	}
+}
+
+func TestICERetryState_EnterHourlyModeArmsTicker(t *testing.T) {
+	s := newTestRetryState()
+	for i := 0; i < maxICERetries+1; i++ {
+		_ = s.shouldRetry()
+	}
+
+	s.enterHourlyMode()
+	defer s.reset()
+
+	if s.hourlyC() == nil {
+		t.Fatalf("hourlyC returned nil after enterHourlyMode")
+	}
+}
+
+func TestICERetryState_ShouldRetryTrueInHourlyMode(t *testing.T) {
+	s := newTestRetryState()
+	s.enterHourlyMode()
+	defer s.reset()
+
+	if !s.shouldRetry() {
+		t.Fatalf("shouldRetry returned false in hourly mode, want true")
+	}
+
+	// Subsequent calls also return true — we keep retrying on each hourly tick.
+	if !s.shouldRetry() {
+		t.Fatalf("second shouldRetry returned false in hourly mode, want true")
+	}
+}
+
+func TestICERetryState_ResetRestoresBudget(t *testing.T) {
+	s := newTestRetryState()
+	for i := 0; i < maxICERetries+1; i++ {
+		_ = s.shouldRetry()
+	}
+	s.enterHourlyMode()
+
+	s.reset()
+
+	if s.hourlyC() != nil {
+		t.Fatalf("hourlyC returned non-nil channel after reset")
+	}
+	if s.retries != 0 {
+		t.Fatalf("retries = %d after reset, want 0", s.retries)
+	}
+
+	for i := 1; i <= maxICERetries; i++ {
+		if !s.shouldRetry() {
+			t.Fatalf("shouldRetry returned false on attempt %d after reset, want true", i)
+		}
+	}
+}
+
+func TestICERetryState_ResetIsIdempotent(t *testing.T) {
+	s := newTestRetryState()
+	s.reset()
+	s.reset() // second call must not panic or re-stop a nil ticker
+
+	if s.hourlyC() != nil {
+		t.Fatalf("hourlyC non-nil after double reset")
+	}
+}

client/internal/peer/guard/sr_watcher.go

@@ -39,7 +39,7 @@ func NewSRWatcher(signalClient chNotifier, relayManager chNotifier, iFaceDiscove
 	return srw
 }
 
-func (w *SRWatcher) Start() {
+func (w *SRWatcher) Start(disableICEMonitor bool) {
 	w.mu.Lock()
 	defer w.mu.Unlock()
 
@@ -50,8 +50,10 @@ func (w *SRWatcher) Start() {
 	ctx, cancel := context.WithCancel(context.Background())
 	w.cancelIceMonitor = cancel
 
-	iceMonitor := NewICEMonitor(w.iFaceDiscover, w.iceConfig, GetICEMonitorPeriod())
-	go iceMonitor.Start(ctx, w.onICEChanged)
+	if !disableICEMonitor {
+		iceMonitor := NewICEMonitor(w.iFaceDiscover, w.iceConfig, GetICEMonitorPeriod())
+		go iceMonitor.Start(ctx, w.onICEChanged)
+	}
 	w.signalClient.SetOnReconnectedListener(w.onReconnected)
 	w.relayManager.SetOnReconnectedListener(w.onReconnected)
 

client/internal/peer/handshaker.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"sync"
+	"sync/atomic"
 
 	log "github.com/sirupsen/logrus"
 
@@ -43,6 +44,10 @@ type OfferAnswer struct {
 	SessionID *ICESessionID
 }
 
+func (o *OfferAnswer) hasICECredentials() bool {
+	return o.IceCredentials.UFrag != "" && o.IceCredentials.Pwd != ""
+}
+
 type Handshaker struct {
 	mu            sync.Mutex
 	log           *log.Entry
@@ -59,6 +64,10 @@ type Handshaker struct {
 	relayListener *AsyncOfferListener
 	iceListener   func(remoteOfferAnswer *OfferAnswer)
 
+	// remoteICESupported tracks whether the remote peer includes ICE credentials in its offers/answers.
+	// When false, the local side skips ICE listener dispatch and suppresses ICE credentials in responses.
+	remoteICESupported atomic.Bool
+
 	// remoteOffersCh is a channel used to wait for remote credentials to proceed with the connection
 	remoteOffersCh chan OfferAnswer
 	// remoteAnswerCh is a channel used to wait for remote credentials answer (confirmation of our offer) to proceed with the connection
@@ -66,7 +75,7 @@ type Handshaker struct {
 }
 
 func NewHandshaker(log *log.Entry, config ConnConfig, signaler *Signaler, ice *WorkerICE, relay *WorkerRelay, metricsStages *MetricsStages) *Handshaker {
-	return &Handshaker{
+	h := &Handshaker{
 		log:            log,
 		config:         config,
 		signaler:       signaler,
@@ -76,6 +85,13 @@ func NewHandshaker(log *log.Entry, config ConnConfig, signaler *Signaler, ice *W
 		remoteOffersCh: make(chan OfferAnswer),
 		remoteAnswerCh: make(chan OfferAnswer),
 	}
+	// assume remote supports ICE until we learn otherwise from received offers
+	h.remoteICESupported.Store(ice != nil)
+	return h
+}
+
+func (h *Handshaker) RemoteICESupported() bool {
+	return h.remoteICESupported.Load()
 }
 
 func (h *Handshaker) AddRelayListener(offer func(remoteOfferAnswer *OfferAnswer)) {
@@ -90,18 +106,20 @@ func (h *Handshaker) Listen(ctx context.Context) {
 	for {
 		select {
 		case remoteOfferAnswer := <-h.remoteOffersCh:
-			h.log.Infof("received offer, running version %s, remote WireGuard listen port %d, session id: %s", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString())
+			h.log.Infof("received offer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials())
 
 			// Record signaling received for reconnection attempts
 			if h.metricsStages != nil {
 				h.metricsStages.RecordSignalingReceived()
 			}
 
+			h.updateRemoteICEState(&remoteOfferAnswer)
+
 			if h.relayListener != nil {
 				h.relayListener.Notify(&remoteOfferAnswer)
 			}
 
-			if h.iceListener != nil {
+			if h.iceListener != nil && h.RemoteICESupported() {
 				h.iceListener(&remoteOfferAnswer)
 			}
 
@@ -110,18 +128,20 @@ func (h *Handshaker) Listen(ctx context.Context) {
 				continue
 			}
 		case remoteOfferAnswer := <-h.remoteAnswerCh:
-			h.log.Infof("received answer, running version %s, remote WireGuard listen port %d, session id: %s", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString())
+			h.log.Infof("received answer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials())
 
 			// Record signaling received for reconnection attempts
 			if h.metricsStages != nil {
 				h.metricsStages.RecordSignalingReceived()
 			}
 
+			h.updateRemoteICEState(&remoteOfferAnswer)
+
 			if h.relayListener != nil {
 				h.relayListener.Notify(&remoteOfferAnswer)
 			}
 
-			if h.iceListener != nil {
+			if h.iceListener != nil && h.RemoteICESupported() {
 				h.iceListener(&remoteOfferAnswer)
 			}
 		case <-ctx.Done():
@@ -183,15 +203,18 @@ func (h *Handshaker) sendAnswer() error {
 }
 
 func (h *Handshaker) buildOfferAnswer() OfferAnswer {
-	uFrag, pwd := h.ice.GetLocalUserCredentials()
-	sid := h.ice.SessionID()
 	answer := OfferAnswer{
-		IceCredentials:  IceCredentials{uFrag, pwd},
 		WgListenPort:    h.config.LocalWgPort,
 		Version:         version.NetbirdVersion(),
 		RosenpassPubKey: h.config.RosenpassConfig.PubKey,
 		RosenpassAddr:   h.config.RosenpassConfig.Addr,
-		SessionID:       &sid,
+	}
+
+	if h.ice != nil && h.RemoteICESupported() {
+		uFrag, pwd := h.ice.GetLocalUserCredentials()
+		sid := h.ice.SessionID()
+		answer.IceCredentials = IceCredentials{uFrag, pwd}
+		answer.SessionID = &sid
 	}
 
 	if addr, err := h.relay.RelayInstanceAddress(); err == nil {
@@ -200,3 +223,18 @@ func (h *Handshaker) buildOfferAnswer() OfferAnswer {
 
 	return answer
 }
+
+func (h *Handshaker) updateRemoteICEState(offer *OfferAnswer) {
+	hasICE := offer.hasICECredentials()
+	prev := h.remoteICESupported.Swap(hasICE)
+	if prev != hasICE {
+		if hasICE {
+			h.log.Infof("remote peer started sending ICE credentials")
+		} else {
+			h.log.Infof("remote peer stopped sending ICE credentials")
+			if h.ice != nil {
+				h.ice.Close()
+			}
+		}
+	}
+}

client/internal/peer/signaler.go

@@ -46,9 +46,13 @@ func (s *Signaler) Ready() bool {
 
 // SignalOfferAnswer signals either an offer or an answer to remote peer
 func (s *Signaler) signalOfferAnswer(offerAnswer OfferAnswer, remoteKey string, bodyType sProto.Body_Type) error {
-	sessionIDBytes, err := offerAnswer.SessionID.Bytes()
-	if err != nil {
-		log.Warnf("failed to get session ID bytes: %v", err)
+	var sessionIDBytes []byte
+	if offerAnswer.SessionID != nil {
+		var err error
+		sessionIDBytes, err = offerAnswer.SessionID.Bytes()
+		if err != nil {
+			log.Warnf("failed to get session ID bytes: %v", err)
+		}
 	}
 	msg, err := signal.MarshalCredential(
 		s.wgPrivateKey,

client/internal/routemanager/systemops/systemops_bsd_other.go

@@ -0,0 +1,10 @@
+//go:build (dragonfly || freebsd || netbsd || openbsd) && !darwin
+
+package systemops
+
+// Non-darwin BSDs don't support the IP_BOUND_IF + scoped default model. They
+// always fall through to the ref-counter exclusion-route path; these stubs
+// exist only so systemops_unix.go compiles.
+func (r *SysOps) setupAdvancedRouting() error   { return nil }
+func (r *SysOps) cleanupAdvancedRouting() error { return nil }
+func (r *SysOps) flushPlatformExtras() error    { return nil }

client/internal/routemanager/systemops/systemops_darwin.go

@@ -0,0 +1,241 @@
+//go:build darwin && !ios
+
+package systemops
+
+import (
+	"errors"
+	"fmt"
+	"net/netip"
+	"os"
+	"time"
+
+	"github.com/hashicorp/go-multierror"
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/net/route"
+	"golang.org/x/sys/unix"
+
+	nberrors "github.com/netbirdio/netbird/client/errors"
+	"github.com/netbirdio/netbird/client/internal/routemanager/vars"
+	nbnet "github.com/netbirdio/netbird/client/net"
+)
+
+// scopedRouteBudget bounds retries for the scoped default route. Installing or
+// deleting it matters enough that we're willing to spend longer waiting for the
+// kernel reply than for per-prefix exclusion routes.
+const scopedRouteBudget = 5 * time.Second
+
+// setupAdvancedRouting installs an RTF_IFSCOPE default route per address family
+// pinned to the current physical egress, so IP_BOUND_IF scoped lookups can
+// resolve gateway'd destinations while the VPN's split default owns the
+// unscoped table.
+//
+// Timing note: this runs during routeManager.Init, which happens before the
+// VPN interface is created and before any peer routes propagate. The initial
+// mgmt / signal / relay TCP dials always fire before this runs, so those
+// sockets miss the IP_BOUND_IF binding and rely on the kernel's normal route
+// lookup, which at that point correctly picks the physical default. Those
+// already-established TCP flows keep their originally-selected interface for
+// their lifetime on Darwin because the kernel caches the egress route
+// per-socket at connect time; adding the VPN's 0/1 + 128/1 split default
+// afterwards does not migrate them since the original en0 default stays in
+// the table. Any subsequent reconnect via nbnet.NewDialer picks up the
+// populated bound-iface cache and gets IP_BOUND_IF set cleanly.
+func (r *SysOps) setupAdvancedRouting() error {
+	// Drop any previously-cached egress interface before reinstalling. On a
+	// refresh, a family that no longer resolves would otherwise keep the stale
+	// binding, causing new sockets to scope to an interface without a matching
+	// scoped default.
+	nbnet.ClearBoundInterfaces()
+
+	if err := r.flushScopedDefaults(); err != nil {
+		log.Warnf("flush residual scoped defaults: %v", err)
+	}
+
+	var merr *multierror.Error
+	installed := 0
+
+	for _, unspec := range []netip.Addr{netip.IPv4Unspecified(), netip.IPv6Unspecified()} {
+		ok, err := r.installScopedDefaultFor(unspec)
+		if err != nil {
+			merr = multierror.Append(merr, err)
+			continue
+		}
+		if ok {
+			installed++
+		}
+	}
+
+	if installed == 0 && merr != nil {
+		return nberrors.FormatErrorOrNil(merr)
+	}
+	if merr != nil {
+		log.Warnf("advanced routing setup partially succeeded: %v", nberrors.FormatErrorOrNil(merr))
+	}
+	return nil
+}
+
+// installScopedDefaultFor resolves the physical default nexthop for the given
+// address family, installs a scoped default via it, and caches the iface for
+// subsequent IP_BOUND_IF / IPV6_BOUND_IF socket binds.
+func (r *SysOps) installScopedDefaultFor(unspec netip.Addr) (bool, error) {
+	nexthop, err := GetNextHop(unspec)
+	if err != nil {
+		if errors.Is(err, vars.ErrRouteNotFound) {
+			return false, nil
+		}
+		return false, fmt.Errorf("get default nexthop for %s: %w", unspec, err)
+	}
+	if nexthop.Intf == nil {
+		return false, fmt.Errorf("unusable default nexthop for %s (no interface)", unspec)
+	}
+
+	if err := r.addScopedDefault(unspec, nexthop); err != nil {
+		return false, fmt.Errorf("add scoped default on %s: %w", nexthop.Intf.Name, err)
+	}
+
+	af := unix.AF_INET
+	if unspec.Is6() {
+		af = unix.AF_INET6
+	}
+	nbnet.SetBoundInterface(af, nexthop.Intf)
+	via := "point-to-point"
+	if nexthop.IP.IsValid() {
+		via = nexthop.IP.String()
+	}
+	log.Infof("installed scoped default route via %s on %s for %s", via, nexthop.Intf.Name, afOf(unspec))
+	return true, nil
+}
+
+func (r *SysOps) cleanupAdvancedRouting() error {
+	nbnet.ClearBoundInterfaces()
+	return r.flushScopedDefaults()
+}
+
+// flushPlatformExtras runs darwin-specific residual cleanup hooked into the
+// generic FlushMarkedRoutes path, so a crashed daemon's scoped defaults get
+// removed on the next boot regardless of whether a profile is brought up.
+func (r *SysOps) flushPlatformExtras() error {
+	return r.flushScopedDefaults()
+}
+
+// flushScopedDefaults removes any scoped default routes tagged with routeProtoFlag.
+// Safe to call at startup to clear residual entries from a prior session.
+func (r *SysOps) flushScopedDefaults() error {
+	rib, err := retryFetchRIB()
+	if err != nil {
+		return fmt.Errorf("fetch routing table: %w", err)
+	}
+
+	msgs, err := route.ParseRIB(route.RIBTypeRoute, rib)
+	if err != nil {
+		return fmt.Errorf("parse routing table: %w", err)
+	}
+
+	var merr *multierror.Error
+	removed := 0
+
+	for _, msg := range msgs {
+		rtMsg, ok := msg.(*route.RouteMessage)
+		if !ok {
+			continue
+		}
+		if rtMsg.Flags&routeProtoFlag == 0 {
+			continue
+		}
+		if rtMsg.Flags&unix.RTF_IFSCOPE == 0 {
+			continue
+		}
+
+		info, err := MsgToRoute(rtMsg)
+		if err != nil {
+			log.Debugf("skip scoped flush: %v", err)
+			continue
+		}
+		if !info.Dst.IsValid() || info.Dst.Bits() != 0 {
+			continue
+		}
+
+		if err := r.deleteScopedRoute(rtMsg); err != nil {
+			merr = multierror.Append(merr, fmt.Errorf("delete scoped default %s on index %d: %w",
+				info.Dst, rtMsg.Index, err))
+			continue
+		}
+		removed++
+		log.Debugf("flushed residual scoped default %s on index %d", info.Dst, rtMsg.Index)
+	}
+
+	if removed > 0 {
+		log.Infof("flushed %d residual scoped default route(s)", removed)
+	}
+	return nberrors.FormatErrorOrNil(merr)
+}
+
+func (r *SysOps) addScopedDefault(unspec netip.Addr, nexthop Nexthop) error {
+	return r.scopedRouteSocket(unix.RTM_ADD, unspec, nexthop)
+}
+
+func (r *SysOps) deleteScopedRoute(rtMsg *route.RouteMessage) error {
+	// Preserve identifying flags from the stored route (including RTF_GATEWAY
+	// only if present); kernel-set bits like RTF_DONE don't belong on RTM_DELETE.
+	keep := unix.RTF_UP | unix.RTF_STATIC | unix.RTF_GATEWAY | unix.RTF_IFSCOPE | routeProtoFlag
+	del := &route.RouteMessage{
+		Type:    unix.RTM_DELETE,
+		Flags:   rtMsg.Flags & keep,
+		Version: unix.RTM_VERSION,
+		Seq:     r.getSeq(),
+		Index:   rtMsg.Index,
+		Addrs:   rtMsg.Addrs,
+	}
+	return r.writeRouteMessage(del, scopedRouteBudget)
+}
+
+func (r *SysOps) scopedRouteSocket(action int, unspec netip.Addr, nexthop Nexthop) error {
+	flags := unix.RTF_UP | unix.RTF_STATIC | unix.RTF_IFSCOPE | routeProtoFlag
+
+	msg := &route.RouteMessage{
+		Type:    action,
+		Flags:   flags,
+		Version: unix.RTM_VERSION,
+		ID:      uintptr(os.Getpid()),
+		Seq:     r.getSeq(),
+		Index:   nexthop.Intf.Index,
+	}
+
+	const numAddrs = unix.RTAX_NETMASK + 1
+	addrs := make([]route.Addr, numAddrs)
+
+	dst, err := addrToRouteAddr(unspec)
+	if err != nil {
+		return fmt.Errorf("build destination: %w", err)
+	}
+	mask, err := prefixToRouteNetmask(netip.PrefixFrom(unspec, 0))
+	if err != nil {
+		return fmt.Errorf("build netmask: %w", err)
+	}
+	addrs[unix.RTAX_DST] = dst
+	addrs[unix.RTAX_NETMASK] = mask
+
+	if nexthop.IP.IsValid() {
+		msg.Flags |= unix.RTF_GATEWAY
+		gw, err := addrToRouteAddr(nexthop.IP.Unmap())
+		if err != nil {
+			return fmt.Errorf("build gateway: %w", err)
+		}
+		addrs[unix.RTAX_GATEWAY] = gw
+	} else {
+		addrs[unix.RTAX_GATEWAY] = &route.LinkAddr{
+			Index: nexthop.Intf.Index,
+			Name:  nexthop.Intf.Name,
+		}
+	}
+	msg.Addrs = addrs
+
+	return r.writeRouteMessage(msg, scopedRouteBudget)
+}
+
+func afOf(a netip.Addr) string {
+	if a.Is4() {
+		return "IPv4"
+	}
+	return "IPv6"
+}

client/internal/routemanager/systemops/systemops_generic.go

@@ -21,6 +21,7 @@ import (
 	"github.com/netbirdio/netbird/client/internal/routemanager/util"
 	"github.com/netbirdio/netbird/client/internal/routemanager/vars"
 	"github.com/netbirdio/netbird/client/internal/statemanager"
+	nbnet "github.com/netbirdio/netbird/client/net"
 	"github.com/netbirdio/netbird/client/net/hooks"
 )
 
@@ -31,8 +32,6 @@ var splitDefaultv4_2 = netip.PrefixFrom(netip.AddrFrom4([4]byte{128}), 1)
 var splitDefaultv6_1 = netip.PrefixFrom(netip.IPv6Unspecified(), 1)
 var splitDefaultv6_2 = netip.PrefixFrom(netip.AddrFrom16([16]byte{0x80}), 1)
 
-var ErrRoutingIsSeparate = errors.New("routing is separate")
-
 func (r *SysOps) setupRefCounter(initAddresses []net.IP, stateManager *statemanager.Manager) error {
 	stateManager.RegisterState(&ShutdownState{})
 
@@ -397,12 +396,16 @@ func ipToAddr(ip net.IP, intf *net.Interface) (netip.Addr, error) {
 }
 
 // IsAddrRouted checks if the candidate address would route to the vpn, in which case it returns true and the matched prefix.
+// When advanced routing is active the WG socket is bound to the physical interface (fwmark on linux,
+// IP_UNICAST_IF on windows, IP_BOUND_IF on darwin) and bypasses the main routing table, so the check is skipped.
 func IsAddrRouted(addr netip.Addr, vpnRoutes []netip.Prefix) (bool, netip.Prefix) {
-	localRoutes, err := hasSeparateRouting()
+	if nbnet.AdvancedRouting() {
+		return false, netip.Prefix{}
+	}
+
+	localRoutes, err := GetRoutesFromTable()
 	if err != nil {
-		if !errors.Is(err, ErrRoutingIsSeparate) {
-			log.Errorf("Failed to get routes: %v", err)
-		}
+		log.Errorf("Failed to get routes: %v", err)
 		return false, netip.Prefix{}
 	}
 

client/internal/routemanager/systemops/systemops_js.go

@@ -22,10 +22,6 @@ func GetRoutesFromTable() ([]netip.Prefix, error) {
 	return []netip.Prefix{}, nil
 }
 
-func hasSeparateRouting() ([]netip.Prefix, error) {
-	return []netip.Prefix{}, nil
-}
-
 // GetDetailedRoutesFromTable returns empty routes for WASM.
 func GetDetailedRoutesFromTable() ([]DetailedRoute, error) {
 	return []DetailedRoute{}, nil

client/internal/routemanager/systemops/systemops_linux.go

@@ -894,13 +894,6 @@ func getAddressFamily(prefix netip.Prefix) int {
 	return netlink.FAMILY_V6
 }
 
-func hasSeparateRouting() ([]netip.Prefix, error) {
-	if !nbnet.AdvancedRouting() {
-		return GetRoutesFromTable()
-	}
-	return nil, ErrRoutingIsSeparate
-}
-
 func isOpErr(err error) bool {
 	// EAFTNOSUPPORT when ipv6 is disabled via sysctl, EOPNOTSUPP when disabled in boot options or otherwise not supported
 	if errors.Is(err, syscall.EAFNOSUPPORT) || errors.Is(err, syscall.EOPNOTSUPP) {

client/internal/routemanager/systemops/systemops_nonlinux.go

@@ -48,10 +48,6 @@ func EnableIPForwarding() error {
 	return nil
 }
 
-func hasSeparateRouting() ([]netip.Prefix, error) {
-	return GetRoutesFromTable()
-}
-
 // GetIPRules returns IP rules for debugging (not supported on non-Linux platforms)
 func GetIPRules() ([]IPRule, error) {
 	log.Infof("IP rules collection is not supported on %s", runtime.GOOS)

client/internal/routemanager/systemops/systemops_unix.go

@@ -25,6 +25,9 @@ import (
 
 const (
 	envRouteProtoFlag = "NB_ROUTE_PROTO_FLAG"
+
+	// routeBudget bounds retries for per-prefix exclusion route programming.
+	routeBudget = 1 * time.Second
 )
 
 var routeProtoFlag int
@@ -41,26 +44,42 @@ func init() {
 }
 
 func (r *SysOps) SetupRouting(initAddresses []net.IP, stateManager *statemanager.Manager, advancedRouting bool) error {
+	if advancedRouting {
+		return r.setupAdvancedRouting()
+	}
+
+	log.Infof("Using legacy routing setup with ref counters")
 	return r.setupRefCounter(initAddresses, stateManager)
 }
 
 func (r *SysOps) CleanupRouting(stateManager *statemanager.Manager, advancedRouting bool) error {
+	if advancedRouting {
+		return r.cleanupAdvancedRouting()
+	}
+
 	return r.cleanupRefCounter(stateManager)
 }
 
 // FlushMarkedRoutes removes single IP exclusion routes marked with the configured RTF_PROTO flag.
+// On darwin it also flushes residual RTF_IFSCOPE scoped default routes so a
+// crashed prior session can't leave crud in the table.
 func (r *SysOps) FlushMarkedRoutes() error {
+	var merr *multierror.Error
+
+	if err := r.flushPlatformExtras(); err != nil {
+		merr = multierror.Append(merr, fmt.Errorf("flush platform extras: %w", err))
+	}
+
 	rib, err := retryFetchRIB()
 	if err != nil {
-		return fmt.Errorf("fetch routing table: %w", err)
+		return nberrors.FormatErrorOrNil(multierror.Append(merr, fmt.Errorf("fetch routing table: %w", err)))
 	}
 
 	msgs, err := route.ParseRIB(route.RIBTypeRoute, rib)
 	if err != nil {
-		return fmt.Errorf("parse routing table: %w", err)
+		return nberrors.FormatErrorOrNil(multierror.Append(merr, fmt.Errorf("parse routing table: %w", err)))
 	}
 
-	var merr *multierror.Error
 	flushedCount := 0
 
 	for _, msg := range msgs {
@@ -117,12 +136,12 @@ func (r *SysOps) routeSocket(action int, prefix netip.Prefix, nexthop Nexthop) e
 		return fmt.Errorf("invalid prefix: %s", prefix)
 	}
 
-	expBackOff := backoff.NewExponentialBackOff()
-	expBackOff.InitialInterval = 50 * time.Millisecond
-	expBackOff.MaxInterval = 500 * time.Millisecond
-	expBackOff.MaxElapsedTime = 1 * time.Second
+	msg, err := r.buildRouteMessage(action, prefix, nexthop)
+	if err != nil {
+		return fmt.Errorf("build route message: %w", err)
+	}
 
-	if err := backoff.Retry(r.routeOp(action, prefix, nexthop), expBackOff); err != nil {
+	if err := r.writeRouteMessage(msg, routeBudget); err != nil {
 		a := "add"
 		if action == unix.RTM_DELETE {
 			a = "remove"
@@ -132,50 +151,91 @@ func (r *SysOps) routeSocket(action int, prefix netip.Prefix, nexthop Nexthop) e
 	return nil
 }
 
-func (r *SysOps) routeOp(action int, prefix netip.Prefix, nexthop Nexthop) func() error {
-	operation := func() error {
-		fd, err := unix.Socket(syscall.AF_ROUTE, syscall.SOCK_RAW, syscall.AF_UNSPEC)
-		if err != nil {
-			return fmt.Errorf("open routing socket: %w", err)
+// writeRouteMessage sends a route message over AF_ROUTE and waits for the
+// kernel's matching reply, retrying transient failures until budget elapses.
+// Callers do not need to manage sockets or seq numbers themselves.
+func (r *SysOps) writeRouteMessage(msg *route.RouteMessage, budget time.Duration) error {
+	expBackOff := backoff.NewExponentialBackOff()
+	expBackOff.InitialInterval = 50 * time.Millisecond
+	expBackOff.MaxInterval = 500 * time.Millisecond
+	expBackOff.MaxElapsedTime = budget
+
+	return backoff.Retry(func() error { return routeMessageRoundtrip(msg) }, expBackOff)
+}
+
+func routeMessageRoundtrip(msg *route.RouteMessage) error {
+	fd, err := unix.Socket(syscall.AF_ROUTE, syscall.SOCK_RAW, syscall.AF_UNSPEC)
+	if err != nil {
+		return fmt.Errorf("open routing socket: %w", err)
+	}
+	defer func() {
+		if err := unix.Close(fd); err != nil && !errors.Is(err, unix.EBADF) {
+			log.Warnf("close routing socket: %v", err)
 		}
-		defer func() {
-			if err := unix.Close(fd); err != nil && !errors.Is(err, unix.EBADF) {
-				log.Warnf("failed to close routing socket: %v", err)
+	}()
+
+	tv := unix.Timeval{Sec: 1}
+	if err := unix.SetsockoptTimeval(fd, unix.SOL_SOCKET, unix.SO_RCVTIMEO, &tv); err != nil {
+		return backoff.Permanent(fmt.Errorf("set recv timeout: %w", err))
+	}
+
+	// AF_ROUTE is a broadcast channel: every route socket on the host sees
+	// every RTM_* event. With concurrent route programming the default
+	// per-socket queue overflows and our own reply gets dropped.
+	if err := unix.SetsockoptInt(fd, unix.SOL_SOCKET, unix.SO_RCVBUF, 1<<20); err != nil {
+		log.Debugf("set SO_RCVBUF on route socket: %v", err)
+	}
+
+	bytes, err := msg.Marshal()
+	if err != nil {
+		return backoff.Permanent(fmt.Errorf("marshal: %w", err))
+	}
+
+	if _, err = unix.Write(fd, bytes); err != nil {
+		if errors.Is(err, unix.ENOBUFS) || errors.Is(err, unix.EAGAIN) {
+			return fmt.Errorf("write: %w", err)
+		}
+		return backoff.Permanent(fmt.Errorf("write: %w", err))
+	}
+	return readRouteResponse(fd, msg.Type, msg.Seq)
+}
+
+// readRouteResponse reads from the AF_ROUTE socket until it sees a reply
+// matching our write (same type, seq, and pid). AF_ROUTE SOCK_RAW is a
+// broadcast channel: interface up/down, third-party route changes and neighbor
+// discovery events can all land between our write and read, so we must filter.
+func readRouteResponse(fd, wantType, wantSeq int) error {
+	pid := int32(os.Getpid())
+	resp := make([]byte, 2048)
+	deadline := time.Now().Add(time.Second)
+	for {
+		if time.Now().After(deadline) {
+			// Transient: under concurrent pressure the kernel can drop our reply
+			// from the socket buffer. Let backoff.Retry re-send with a fresh seq.
+			return fmt.Errorf("read: timeout waiting for route reply type=%d seq=%d", wantType, wantSeq)
+		}
+		n, err := unix.Read(fd, resp)
+		if err != nil {
+			if errors.Is(err, unix.EAGAIN) || errors.Is(err, unix.EWOULDBLOCK) {
+				// SO_RCVTIMEO fired while waiting; loop to re-check the absolute deadline.
+				continue
 			}
-		}()
-
-		msg, err := r.buildRouteMessage(action, prefix, nexthop)
-		if err != nil {
-			return backoff.Permanent(fmt.Errorf("build route message: %w", err))
+			return backoff.Permanent(fmt.Errorf("read: %w", err))
 		}
-
-		msgBytes, err := msg.Marshal()
-		if err != nil {
-			return backoff.Permanent(fmt.Errorf("marshal route message: %w", err))
+		if n < int(unsafe.Sizeof(unix.RtMsghdr{})) {
+			continue
 		}
-
-		if _, err = unix.Write(fd, msgBytes); err != nil {
-			if errors.Is(err, unix.ENOBUFS) || errors.Is(err, unix.EAGAIN) {
-				return fmt.Errorf("write: %w", err)
-			}
-			return backoff.Permanent(fmt.Errorf("write: %w", err))
+		hdr := (*unix.RtMsghdr)(unsafe.Pointer(&resp[0]))
+		// Darwin reflects the sender's pid on replies; matching (Type, Seq, Pid)
+		// uniquely identifies our own reply among broadcast traffic.
+		if int(hdr.Type) != wantType || int(hdr.Seq) != wantSeq || hdr.Pid != pid {
+			continue
 		}
-
-		respBuf := make([]byte, 2048)
-		n, err := unix.Read(fd, respBuf)
-		if err != nil {
-			return backoff.Permanent(fmt.Errorf("read route response: %w", err))
+		if hdr.Errno != 0 {
+			return backoff.Permanent(fmt.Errorf("kernel: %w", syscall.Errno(hdr.Errno)))
 		}
-
-		if n > 0 {
-			if err := r.parseRouteResponse(respBuf[:n]); err != nil {
-				return backoff.Permanent(err)
-			}
-		}
-
 		return nil
 	}
-	return operation
 }
 
 func (r *SysOps) buildRouteMessage(action int, prefix netip.Prefix, nexthop Nexthop) (msg *route.RouteMessage, err error) {
@@ -183,6 +243,7 @@ func (r *SysOps) buildRouteMessage(action int, prefix netip.Prefix, nexthop Next
 		Type:    action,
 		Flags:   unix.RTF_UP | routeProtoFlag,
 		Version: unix.RTM_VERSION,
+		ID:      uintptr(os.Getpid()),
 		Seq:     r.getSeq(),
 	}
 
@@ -221,19 +282,6 @@ func (r *SysOps) buildRouteMessage(action int, prefix netip.Prefix, nexthop Next
 	return msg, nil
 }
 
-func (r *SysOps) parseRouteResponse(buf []byte) error {
-	if len(buf) < int(unsafe.Sizeof(unix.RtMsghdr{})) {
-		return nil
-	}
-
-	rtMsg := (*unix.RtMsghdr)(unsafe.Pointer(&buf[0]))
-	if rtMsg.Errno != 0 {
-		return fmt.Errorf("parse: %d", rtMsg.Errno)
-	}
-
-	return nil
-}
-
 // addrToRouteAddr converts a netip.Addr to the appropriate route.Addr (*route.Inet4Addr or *route.Inet6Addr).
 func addrToRouteAddr(addr netip.Addr) (route.Addr, error) {
 	if addr.Is4() {