Use cscli lapi status for CrowdSec readiness in installer

* [client] Suppress ICE signaling and periodic offers in force-relay mode

When NB_FORCE_RELAY is enabled, skip WorkerICE creation entirely,
suppress ICE credentials in offer/answer messages, disable the
periodic ICE candidate monitor, and fix isConnectedOnAllWay to
only check relay status so the guard stops sending unnecessary offers.

* [client] Dynamically suppress ICE based on remote peer's offer credentials

Track whether the remote peer includes ICE credentials in its
offers/answers. When remote stops sending ICE credentials, skip
ICE listener dispatch, suppress ICE credentials in responses, and
exclude ICE from the guard connectivity check. When remote resumes
sending ICE credentials, re-enable all ICE behavior.

* [client] Fix nil SessionID panic and force ICE teardown on relay-only transition

Fix nil pointer dereference in signalOfferAnswer when SessionID is nil
(relay-only offers). Close stale ICE agent immediately when remote peer
stops sending ICE credentials to avoid traffic black-hole during the
ICE disconnect timeout.

* [client] Add relay-only fallback check when ICE is unavailable

Ensure the relay connection is supported with the peer when ICE is disabled to prevent connectivity issues.

* [client] Add tri-state connection status to guard for smarter ICE retry (#5828)

* [client] Add tri-state connection status to guard for smarter ICE retry

Refactor isConnectedOnAllWay to return a ConnStatus enum (Connected,
Disconnected, PartiallyConnected) instead of a boolean. When relay is
up but ICE is not (PartiallyConnected), limit ICE offers to 3 retries
with exponential backoff then fall back to hourly attempts, reducing
unnecessary signaling traffic. Fully disconnected peers continue to
retry aggressively. External events (relay/ICE disconnect, signal/relay
reconnect) reset retry state to give ICE a fresh chance.

* [client] Clarify guard ICE retry state and trace log trigger

Split iceRetryState.attempt into shouldRetry (pure predicate) and
enterHourlyMode (explicit state transition) so the caller in
reconnectLoopWithRetry reads top-to-bottom. Restore the original
trace-log behavior in isConnectedOnAllWay so it only logs on full
disconnection, not on the new PartiallyConnected state.

* [client] Extract pure evalConnStatus and add unit tests

Split isConnectedOnAllWay into a thin method that snapshots state and
a pure evalConnStatus helper that takes a connStatusInputs struct, so
the tri-state decision logic can be exercised without constructing
full Worker or Handshaker objects. Add table-driven tests covering
force-relay, ICE-unavailable and fully-available code paths, plus
unit tests for iceRetryState budget/hourly transitions and reset.

* [client] Improve grammar in logs and refactor ICE credential checks

Makefile

@@ -5,7 +5,7 @@ GOLANGCI_LINT := $(shell pwd)/bin/golangci-lint
 $(GOLANGCI_LINT):
 	@echo "Installing golangci-lint..."
 	@mkdir -p ./bin
-	@GOBIN=$(shell pwd)/bin go install github.com/golangci/golangci-lint/cmd/golangci-lint@latest
+	@GOBIN=$(shell pwd)/bin go install github.com/golangci/golangci-lint/v2/cmd/golangci-lint@latest
 
 # Lint only changed files (fast, for pre-push)
 lint: $(GOLANGCI_LINT)

client/android/client.go

@@ -8,6 +8,7 @@ import (
 	"os"
 	"slices"
 	"sync"
+	"time"
 
 	"golang.org/x/exp/maps"
 
@@ -15,6 +16,7 @@ import (
 
 	"github.com/netbirdio/netbird/client/iface/device"
 	"github.com/netbirdio/netbird/client/internal"
+	"github.com/netbirdio/netbird/client/internal/debug"
 	"github.com/netbirdio/netbird/client/internal/dns"
 	"github.com/netbirdio/netbird/client/internal/listener"
 	"github.com/netbirdio/netbird/client/internal/peer"
@@ -26,6 +28,7 @@ import (
 	"github.com/netbirdio/netbird/formatter"
 	"github.com/netbirdio/netbird/route"
 	"github.com/netbirdio/netbird/shared/management/domain"
+	types "github.com/netbirdio/netbird/upload-server/types"
 )
 
 // ConnectionListener export internal Listener for mobile
@@ -68,7 +71,30 @@ type Client struct {
 	uiVersion             string
 	networkChangeListener listener.NetworkChangeListener
 
+	stateMu       sync.RWMutex
 	connectClient *internal.ConnectClient
+	config        *profilemanager.Config
+	cacheDir      string
+}
+
+func (c *Client) setState(cfg *profilemanager.Config, cacheDir string, cc *internal.ConnectClient) {
+	c.stateMu.Lock()
+	defer c.stateMu.Unlock()
+	c.config = cfg
+	c.cacheDir = cacheDir
+	c.connectClient = cc
+}
+
+func (c *Client) stateSnapshot() (*profilemanager.Config, string, *internal.ConnectClient) {
+	c.stateMu.RLock()
+	defer c.stateMu.RUnlock()
+	return c.config, c.cacheDir, c.connectClient
+}
+
+func (c *Client) getConnectClient() *internal.ConnectClient {
+	c.stateMu.RLock()
+	defer c.stateMu.RUnlock()
+	return c.connectClient
 }
 
 // NewClient instantiate a new Client
@@ -93,6 +119,7 @@ func (c *Client) Run(platformFiles PlatformFiles, urlOpener URLOpener, isAndroid
 
 	cfgFile := platformFiles.ConfigurationFilePath()
 	stateFile := platformFiles.StateFilePath()
+	cacheDir := platformFiles.CacheDir()
 
 	log.Infof("Starting client with config: %s, state: %s", cfgFile, stateFile)
 
@@ -124,8 +151,9 @@ func (c *Client) Run(platformFiles PlatformFiles, urlOpener URLOpener, isAndroid
 
 	// todo do not throw error in case of cancelled context
 	ctx = internal.CtxInitState(ctx)
-	c.connectClient = internal.NewConnectClient(ctx, cfg, c.recorder)
+	connectClient := internal.NewConnectClient(ctx, cfg, c.recorder)
-	return c.connectClient.RunOnAndroid(c.tunAdapter, c.iFaceDiscover, c.networkChangeListener, slices.Clone(dns.items), dnsReadyListener, stateFile)
+	c.setState(cfg, cacheDir, connectClient)
+	return connectClient.RunOnAndroid(c.tunAdapter, c.iFaceDiscover, c.networkChangeListener, slices.Clone(dns.items), dnsReadyListener, stateFile, cacheDir)
 }
 
 // RunWithoutLogin we apply this type of run function when the backed has been started without UI (i.e. after reboot).
@@ -135,6 +163,7 @@ func (c *Client) RunWithoutLogin(platformFiles PlatformFiles, dns *DNSList, dnsR
 
 	cfgFile := platformFiles.ConfigurationFilePath()
 	stateFile := platformFiles.StateFilePath()
+	cacheDir := platformFiles.CacheDir()
 
 	log.Infof("Starting client without login with config: %s, state: %s", cfgFile, stateFile)
 
@@ -157,8 +186,9 @@ func (c *Client) RunWithoutLogin(platformFiles PlatformFiles, dns *DNSList, dnsR
 
 	// todo do not throw error in case of cancelled context
 	ctx = internal.CtxInitState(ctx)
-	c.connectClient = internal.NewConnectClient(ctx, cfg, c.recorder)
+	connectClient := internal.NewConnectClient(ctx, cfg, c.recorder)
-	return c.connectClient.RunOnAndroid(c.tunAdapter, c.iFaceDiscover, c.networkChangeListener, slices.Clone(dns.items), dnsReadyListener, stateFile)
+	c.setState(cfg, cacheDir, connectClient)
+	return connectClient.RunOnAndroid(c.tunAdapter, c.iFaceDiscover, c.networkChangeListener, slices.Clone(dns.items), dnsReadyListener, stateFile, cacheDir)
 }
 
 // Stop the internal client and free the resources
@@ -173,11 +203,12 @@ func (c *Client) Stop() {
 }
 
 func (c *Client) RenewTun(fd int) error {
-	if c.connectClient == nil {
+	cc := c.getConnectClient()
+	if cc == nil {
 		return fmt.Errorf("engine not running")
 	}
 
-	e := c.connectClient.Engine()
+	e := cc.Engine()
 	if e == nil {
 		return fmt.Errorf("engine not initialized")
 	}
@@ -185,6 +216,73 @@ func (c *Client) RenewTun(fd int) error {
 	return e.RenewTun(fd)
 }
 
+// DebugBundle generates a debug bundle, uploads it, and returns the upload key.
+// It works both with and without a running engine.
+func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool) (string, error) {
+	cfg, cacheDir, cc := c.stateSnapshot()
+
+	// If the engine hasn't been started, load config from disk
+	if cfg == nil {
+		var err error
+		cfg, err = profilemanager.UpdateOrCreateConfig(profilemanager.ConfigInput{
+			ConfigPath: platformFiles.ConfigurationFilePath(),
+		})
+		if err != nil {
+			return "", fmt.Errorf("load config: %w", err)
+		}
+		cacheDir = platformFiles.CacheDir()
+	}
+
+	deps := debug.GeneratorDependencies{
+		InternalConfig: cfg,
+		StatusRecorder: c.recorder,
+		TempDir:        cacheDir,
+	}
+
+	if cc != nil {
+		resp, err := cc.GetLatestSyncResponse()
+		if err != nil {
+			log.Warnf("get latest sync response: %v", err)
+		}
+		deps.SyncResponse = resp
+
+		if e := cc.Engine(); e != nil {
+			if cm := e.GetClientMetrics(); cm != nil {
+				deps.ClientMetrics = cm
+			}
+		}
+	}
+
+	bundleGenerator := debug.NewBundleGenerator(
+		deps,
+		debug.BundleConfig{
+			Anonymize:         anonymize,
+			IncludeSystemInfo: true,
+		},
+	)
+
+	path, err := bundleGenerator.Generate()
+	if err != nil {
+		return "", fmt.Errorf("generate debug bundle: %w", err)
+	}
+	defer func() {
+		if err := os.Remove(path); err != nil {
+			log.Errorf("failed to remove debug bundle file: %v", err)
+		}
+	}()
+
+	uploadCtx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
+	defer cancel()
+
+	key, err := debug.UploadDebugBundle(uploadCtx, types.DefaultBundleURL, cfg.ManagementURL.String(), path)
+	if err != nil {
+		return "", fmt.Errorf("upload debug bundle: %w", err)
+	}
+
+	log.Infof("debug bundle uploaded with key %s", key)
+	return key, nil
+}
+
 // SetTraceLogLevel configure the logger to trace level
 func (c *Client) SetTraceLogLevel() {
 	log.SetLevel(log.TraceLevel)
@@ -214,12 +312,13 @@ func (c *Client) PeersList() *PeerInfoArray {
 }
 
 func (c *Client) Networks() *NetworkArray {
-	if c.connectClient == nil {
+	cc := c.getConnectClient()
+	if cc == nil {
 		log.Error("not connected")
 		return nil
 	}
 
-	engine := c.connectClient.Engine()
+	engine := cc.Engine()
 	if engine == nil {
 		log.Error("could not get engine")
 		return nil
@@ -300,7 +399,7 @@ func (c *Client) toggleRoute(command routeCommand) error {
 }
 
 func (c *Client) getRouteManager() (routemanager.Manager, error) {
-	client := c.connectClient
+	client := c.getConnectClient()
 	if client == nil {
 		return nil, fmt.Errorf("not connected")
 	}

client/android/platform_files.go

@@ -7,4 +7,5 @@ package android
 type PlatformFiles interface {
 	ConfigurationFilePath() string
 	StateFilePath() string
+	CacheDir() string
 }

client/firewall/firewalld/firewalld.go

@@ -0,0 +1,11 @@
+// Package firewalld integrates with the firewalld daemon so NetBird can place
+// its wg interface into firewalld's "trusted" zone. This is required because
+// firewalld's nftables chains are created with NFT_CHAIN_OWNER on recent
+// versions, which returns EPERM to any other process that tries to insert
+// rules into them. The workaround mirrors what Tailscale does: let firewalld
+// itself add the accept rules to its own chains by trusting the interface.
+package firewalld
+
+// TrustedZone is the firewalld zone name used for interfaces whose traffic
+// should bypass firewalld filtering.
+const TrustedZone = "trusted"

client/firewall/firewalld/firewalld_linux.go

@@ -0,0 +1,260 @@
+//go:build linux
+
+package firewalld
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os/exec"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/godbus/dbus/v5"
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	dbusDest      = "org.fedoraproject.FirewallD1"
+	dbusPath      = "/org/fedoraproject/FirewallD1"
+	dbusRootIface = "org.fedoraproject.FirewallD1"
+	dbusZoneIface = "org.fedoraproject.FirewallD1.zone"
+
+	errZoneAlreadySet = "ZONE_ALREADY_SET"
+	errAlreadyEnabled = "ALREADY_ENABLED"
+	errUnknownIface   = "UNKNOWN_INTERFACE"
+	errNotEnabled     = "NOT_ENABLED"
+
+	// callTimeout bounds each individual DBus or firewall-cmd invocation.
+	// A fresh context is created for each call so a slow DBus probe can't
+	// exhaust the deadline before the firewall-cmd fallback gets to run.
+	callTimeout = 3 * time.Second
+)
+
+var (
+	errDBusUnavailable = errors.New("firewalld dbus unavailable")
+
+	// trustLogOnce ensures the "added to trusted zone" message is logged at
+	// Info level only for the first successful add per process; repeat adds
+	// from other init paths are quieter.
+	trustLogOnce sync.Once
+
+	parentCtxMu sync.RWMutex
+	parentCtx   context.Context = context.Background()
+)
+
+// SetParentContext installs a parent context whose cancellation aborts any
+// in-flight TrustInterface call. It does not affect UntrustInterface, which
+// always uses a fresh Background-rooted timeout so cleanup can still run
+// during engine shutdown when the engine context is already cancelled.
+func SetParentContext(ctx context.Context) {
+	parentCtxMu.Lock()
+	parentCtx = ctx
+	parentCtxMu.Unlock()
+}
+
+func getParentContext() context.Context {
+	parentCtxMu.RLock()
+	defer parentCtxMu.RUnlock()
+	return parentCtx
+}
+
+// TrustInterface places iface into firewalld's trusted zone if firewalld is
+// running. It is idempotent and best-effort: errors are returned so callers
+// can log, but a non-running firewalld is not an error. Only the first
+// successful call per process logs at Info. Respects the parent context set
+// via SetParentContext so startup-time cancellation unblocks it.
+func TrustInterface(iface string) error {
+	parent := getParentContext()
+	if !isRunning(parent) {
+		return nil
+	}
+	if err := addTrusted(parent, iface); err != nil {
+		return fmt.Errorf("add %s to firewalld trusted zone: %w", iface, err)
+	}
+	trustLogOnce.Do(func() {
+		log.Infof("added %s to firewalld trusted zone", iface)
+	})
+	log.Debugf("firewalld: ensured %s is in trusted zone", iface)
+	return nil
+}
+
+// UntrustInterface removes iface from firewalld's trusted zone if firewalld
+// is running. Idempotent. Uses a Background-rooted timeout so it still runs
+// during shutdown after the engine context has been cancelled.
+func UntrustInterface(iface string) error {
+	if !isRunning(context.Background()) {
+		return nil
+	}
+	if err := removeTrusted(context.Background(), iface); err != nil {
+		return fmt.Errorf("remove %s from firewalld trusted zone: %w", iface, err)
+	}
+	return nil
+}
+
+func newCallContext(parent context.Context) (context.Context, context.CancelFunc) {
+	return context.WithTimeout(parent, callTimeout)
+}
+
+func isRunning(parent context.Context) bool {
+	ctx, cancel := newCallContext(parent)
+	ok, err := isRunningDBus(ctx)
+	cancel()
+	if err == nil {
+		return ok
+	}
+	if errors.Is(err, errDBusUnavailable) || errors.Is(err, context.DeadlineExceeded) {
+		ctx, cancel = newCallContext(parent)
+		defer cancel()
+		return isRunningCLI(ctx)
+	}
+	return false
+}
+
+func addTrusted(parent context.Context, iface string) error {
+	ctx, cancel := newCallContext(parent)
+	err := addDBus(ctx, iface)
+	cancel()
+	if err == nil {
+		return nil
+	}
+	if !errors.Is(err, errDBusUnavailable) {
+		log.Debugf("firewalld: dbus add failed, falling back to firewall-cmd: %v", err)
+	}
+	ctx, cancel = newCallContext(parent)
+	defer cancel()
+	return addCLI(ctx, iface)
+}
+
+func removeTrusted(parent context.Context, iface string) error {
+	ctx, cancel := newCallContext(parent)
+	err := removeDBus(ctx, iface)
+	cancel()
+	if err == nil {
+		return nil
+	}
+	if !errors.Is(err, errDBusUnavailable) {
+		log.Debugf("firewalld: dbus remove failed, falling back to firewall-cmd: %v", err)
+	}
+	ctx, cancel = newCallContext(parent)
+	defer cancel()
+	return removeCLI(ctx, iface)
+}
+
+func isRunningDBus(ctx context.Context) (bool, error) {
+	conn, err := dbus.SystemBus()
+	if err != nil {
+		return false, fmt.Errorf("%w: %v", errDBusUnavailable, err)
+	}
+	obj := conn.Object(dbusDest, dbusPath)
+
+	var zone string
+	if err := obj.CallWithContext(ctx, dbusRootIface+".getDefaultZone", 0).Store(&zone); err != nil {
+		return false, fmt.Errorf("firewalld getDefaultZone: %w", err)
+	}
+	return true, nil
+}
+
+func isRunningCLI(ctx context.Context) bool {
+	if _, err := exec.LookPath("firewall-cmd"); err != nil {
+		return false
+	}
+	return exec.CommandContext(ctx, "firewall-cmd", "--state").Run() == nil
+}
+
+func addDBus(ctx context.Context, iface string) error {
+	conn, err := dbus.SystemBus()
+	if err != nil {
+		return fmt.Errorf("%w: %v", errDBusUnavailable, err)
+	}
+	obj := conn.Object(dbusDest, dbusPath)
+
+	call := obj.CallWithContext(ctx, dbusZoneIface+".addInterface", 0, TrustedZone, iface)
+	if call.Err == nil {
+		return nil
+	}
+
+	if dbusErrContains(call.Err, errAlreadyEnabled) {
+		return nil
+	}
+
+	if dbusErrContains(call.Err, errZoneAlreadySet) {
+		move := obj.CallWithContext(ctx, dbusZoneIface+".changeZoneOfInterface", 0, TrustedZone, iface)
+		if move.Err != nil {
+			return fmt.Errorf("firewalld changeZoneOfInterface: %w", move.Err)
+		}
+		return nil
+	}
+
+	return fmt.Errorf("firewalld addInterface: %w", call.Err)
+}
+
+func removeDBus(ctx context.Context, iface string) error {
+	conn, err := dbus.SystemBus()
+	if err != nil {
+		return fmt.Errorf("%w: %v", errDBusUnavailable, err)
+	}
+	obj := conn.Object(dbusDest, dbusPath)
+
+	call := obj.CallWithContext(ctx, dbusZoneIface+".removeInterface", 0, TrustedZone, iface)
+	if call.Err == nil {
+		return nil
+	}
+
+	if dbusErrContains(call.Err, errUnknownIface) || dbusErrContains(call.Err, errNotEnabled) {
+		return nil
+	}
+
+	return fmt.Errorf("firewalld removeInterface: %w", call.Err)
+}
+
+func addCLI(ctx context.Context, iface string) error {
+	if _, err := exec.LookPath("firewall-cmd"); err != nil {
+		return fmt.Errorf("firewall-cmd not available: %w", err)
+	}
+
+	// --change-interface (no --permanent) binds the interface for the
+	// current runtime only; we do not want membership to persist across
+	// reboots because netbird re-asserts it on every startup.
+	out, err := exec.CommandContext(ctx,
+		"firewall-cmd", "--zone="+TrustedZone, "--change-interface="+iface,
+	).CombinedOutput()
+	if err != nil {
+		return fmt.Errorf("firewall-cmd change-interface: %w: %s", err, strings.TrimSpace(string(out)))
+	}
+	return nil
+}
+
+func removeCLI(ctx context.Context, iface string) error {
+	if _, err := exec.LookPath("firewall-cmd"); err != nil {
+		return fmt.Errorf("firewall-cmd not available: %w", err)
+	}
+
+	out, err := exec.CommandContext(ctx,
+		"firewall-cmd", "--zone="+TrustedZone, "--remove-interface="+iface,
+	).CombinedOutput()
+	if err != nil {
+		msg := strings.TrimSpace(string(out))
+		if strings.Contains(msg, errUnknownIface) || strings.Contains(msg, errNotEnabled) {
+			return nil
+		}
+		return fmt.Errorf("firewall-cmd remove-interface: %w: %s", err, msg)
+	}
+	return nil
+}
+
+func dbusErrContains(err error, code string) bool {
+	if err == nil {
+		return false
+	}
+	var de dbus.Error
+	if errors.As(err, &de) {
+		for _, b := range de.Body {
+			if s, ok := b.(string); ok && strings.Contains(s, code) {
+				return true
+			}
+		}
+	}
+	return strings.Contains(err.Error(), code)
+}

client/firewall/firewalld/firewalld_linux_test.go

@@ -0,0 +1,49 @@
+//go:build linux
+
+package firewalld
+
+import (
+	"errors"
+	"testing"
+
+	"github.com/godbus/dbus/v5"
+)
+
+func TestDBusErrContains(t *testing.T) {
+	tests := []struct {
+		name string
+		err  error
+		code string
+		want bool
+	}{
+		{"nil error", nil, errZoneAlreadySet, false},
+		{"plain error match", errors.New("ZONE_ALREADY_SET: wt0"), errZoneAlreadySet, true},
+		{"plain error miss", errors.New("something else"), errZoneAlreadySet, false},
+		{
+			"dbus.Error body match",
+			dbus.Error{Name: "org.fedoraproject.FirewallD1.Exception", Body: []any{"ZONE_ALREADY_SET: wt0"}},
+			errZoneAlreadySet,
+			true,
+		},
+		{
+			"dbus.Error body miss",
+			dbus.Error{Name: "org.fedoraproject.FirewallD1.Exception", Body: []any{"INVALID_INTERFACE"}},
+			errAlreadyEnabled,
+			false,
+		},
+		{
+			"dbus.Error non-string body falls back to Error()",
+			dbus.Error{Name: "x", Body: []any{123}},
+			"x",
+			true,
+		},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := dbusErrContains(tc.err, tc.code)
+			if got != tc.want {
+				t.Fatalf("dbusErrContains(%v, %q) = %v; want %v", tc.err, tc.code, got, tc.want)
+			}
+		})
+	}
+}

client/firewall/firewalld/firewalld_other.go

@@ -0,0 +1,25 @@
+//go:build !linux
+
+package firewalld
+
+import "context"
+
+// SetParentContext is a no-op on non-Linux platforms because firewalld only
+// runs on Linux.
+func SetParentContext(context.Context) {
+	// intentionally empty: firewalld is a Linux-only daemon
+}
+
+// TrustInterface is a no-op on non-Linux platforms because firewalld only
+// runs on Linux.
+func TrustInterface(string) error {
+	// intentionally empty: firewalld is a Linux-only daemon
+	return nil
+}
+
+// UntrustInterface is a no-op on non-Linux platforms because firewalld only
+// runs on Linux.
+func UntrustInterface(string) error {
+	// intentionally empty: firewalld is a Linux-only daemon
+	return nil
+}

client/firewall/iptables/manager_linux.go

@@ -12,6 +12,7 @@ import (
 	log "github.com/sirupsen/logrus"
 
 	nberrors "github.com/netbirdio/netbird/client/errors"
+	"github.com/netbirdio/netbird/client/firewall/firewalld"
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
 	"github.com/netbirdio/netbird/client/iface/wgaddr"
 	"github.com/netbirdio/netbird/client/internal/statemanager"
@@ -86,6 +87,12 @@ func (m *Manager) Init(stateManager *statemanager.Manager) error {
 		log.Warnf("raw table not available, notrack rules will be disabled: %v", err)
 	}
 
+	// Trust after all fatal init steps so a later failure doesn't leave the
+	// interface in firewalld's trusted zone without a corresponding Close.
+	if err := firewalld.TrustInterface(m.wgIface.Name()); err != nil {
+		log.Warnf("failed to trust interface in firewalld: %v", err)
+	}
+
 	// persist early to ensure cleanup of chains
 	go func() {
 		if err := stateManager.PersistState(context.Background()); err != nil {
@@ -191,6 +198,12 @@ func (m *Manager) Close(stateManager *statemanager.Manager) error {
 		merr = multierror.Append(merr, fmt.Errorf("reset router: %w", err))
 	}
 
+	// Appending to merr intentionally blocks DeleteState below so ShutdownState
+	// stays persisted and the crash-recovery path retries firewalld cleanup.
+	if err := firewalld.UntrustInterface(m.wgIface.Name()); err != nil {
+		merr = multierror.Append(merr, err)
+	}
+
 	// attempt to delete state only if all other operations succeeded
 	if merr == nil {
 		if err := stateManager.DeleteState(&ShutdownState{}); err != nil {
@@ -217,6 +230,11 @@ func (m *Manager) AllowNetbird() error {
 	if err != nil {
 		return fmt.Errorf("allow netbird interface traffic: %w", err)
 	}
+
+	if err := firewalld.TrustInterface(m.wgIface.Name()); err != nil {
+		log.Warnf("failed to trust interface in firewalld: %v", err)
+	}
+
 	return nil
 }
 

client/firewall/nftables/manager_linux.go

@@ -14,6 +14,7 @@ import (
 	log "github.com/sirupsen/logrus"
 	"golang.org/x/sys/unix"
 
+	"github.com/netbirdio/netbird/client/firewall/firewalld"
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
 	"github.com/netbirdio/netbird/client/iface/wgaddr"
 	"github.com/netbirdio/netbird/client/internal/statemanager"
@@ -217,6 +218,10 @@ func (m *Manager) AllowNetbird() error {
 		return fmt.Errorf("flush allow input netbird rules: %w", err)
 	}
 
+	if err := firewalld.TrustInterface(m.wgIface.Name()); err != nil {
+		log.Warnf("failed to trust interface in firewalld: %v", err)
+	}
+
 	return nil
 }
 

client/firewall/nftables/router_linux.go

@@ -19,6 +19,7 @@ import (
 	"golang.org/x/sys/unix"
 
 	nberrors "github.com/netbirdio/netbird/client/errors"
+	"github.com/netbirdio/netbird/client/firewall/firewalld"
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
 	nbid "github.com/netbirdio/netbird/client/internal/acl/id"
 	"github.com/netbirdio/netbird/client/internal/routemanager/ipfwdstate"
@@ -40,6 +41,8 @@ const (
 	chainNameForward       = "FORWARD"
 	chainNameMangleForward = "netbird-mangle-forward"
 
+	firewalldTableName = "firewalld"
+
 	userDataAcceptForwardRuleIif = "frwacceptiif"
 	userDataAcceptForwardRuleOif = "frwacceptoif"
 	userDataAcceptInputRule      = "inputaccept"
@@ -133,6 +136,10 @@ func (r *router) Reset() error {
 		merr = multierror.Append(merr, fmt.Errorf("remove accept filter rules: %w", err))
 	}
 
+	if err := firewalld.UntrustInterface(r.wgIface.Name()); err != nil {
+		merr = multierror.Append(merr, err)
+	}
+
 	if err := r.removeNatPreroutingRules(); err != nil {
 		merr = multierror.Append(merr, fmt.Errorf("remove filter prerouting rules: %w", err))
 	}
@@ -280,6 +287,10 @@ func (r *router) createContainers() error {
 		log.Errorf("failed to add accept rules for the forward chain: %s", err)
 	}
 
+	if err := firewalld.TrustInterface(r.wgIface.Name()); err != nil {
+		log.Warnf("failed to trust interface in firewalld: %v", err)
+	}
+
 	if err := r.refreshRulesMap(); err != nil {
 		log.Errorf("failed to refresh rules: %s", err)
 	}
@@ -1319,6 +1330,13 @@ func (r *router) isExternalChain(chain *nftables.Chain) bool {
 		return false
 	}
 
+	// Skip firewalld-owned chains. Firewalld creates its chains with the
+	// NFT_CHAIN_OWNER flag, so inserting rules into them returns EPERM.
+	// We delegate acceptance to firewalld by trusting the interface instead.
+	if chain.Table.Name == firewalldTableName {
+		return false
+	}
+
 	// Skip all iptables-managed tables in the ip family
 	if chain.Table.Family == nftables.TableFamilyIPv4 && isIptablesTable(chain.Table.Name) {
 		return false

client/firewall/uspfilter/allow_netbird.go

@@ -3,6 +3,9 @@
 package uspfilter
 
 import (
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/client/firewall/firewalld"
 	"github.com/netbirdio/netbird/client/internal/statemanager"
 )
 
@@ -16,6 +19,9 @@ func (m *Manager) Close(stateManager *statemanager.Manager) error {
 	if m.nativeFirewall != nil {
 		return m.nativeFirewall.Close(stateManager)
 	}
+	if err := firewalld.UntrustInterface(m.wgIface.Name()); err != nil {
+		log.Warnf("failed to untrust interface in firewalld: %v", err)
+	}
 	return nil
 }
 
@@ -24,5 +30,8 @@ func (m *Manager) AllowNetbird() error {
 	if m.nativeFirewall != nil {
 		return m.nativeFirewall.AllowNetbird()
 	}
+	if err := firewalld.TrustInterface(m.wgIface.Name()); err != nil {
+		log.Warnf("failed to trust interface in firewalld: %v", err)
+	}
 	return nil
 }

client/firewall/uspfilter/common/iface.go

@@ -9,6 +9,7 @@ import (
 
 // IFaceMapper defines subset methods of interface required for manager
 type IFaceMapper interface {
+	Name() string
 	SetFilter(device.PacketFilter) error
 	Address() wgaddr.Address
 	GetWGDevice() *wgdevice.Device

client/firewall/uspfilter/filter_test.go

@@ -31,12 +31,20 @@ var logger = log.NewFromLogrus(logrus.StandardLogger())
 var flowLogger = netflow.NewManager(nil, []byte{}, nil).GetLogger()
 
 type IFaceMock struct {
+	NameFunc        func() string
 	SetFilterFunc   func(device.PacketFilter) error
 	AddressFunc     func() wgaddr.Address
 	GetWGDeviceFunc func() *wgdevice.Device
 	GetDeviceFunc   func() *device.FilteredDevice
 }
 
+func (i *IFaceMock) Name() string {
+	if i.NameFunc == nil {
+		return "wgtest"
+	}
+	return i.NameFunc()
+}
+
 func (i *IFaceMock) GetWGDevice() *wgdevice.Device {
 	if i.GetWGDeviceFunc == nil {
 		return nil

client/iface/iface.go

@@ -217,7 +217,6 @@ func (w *WGIface) RemoveAllowedIP(peerKey string, allowedIP netip.Prefix) error
 // Close closes the tunnel interface
 func (w *WGIface) Close() error {
 	w.mu.Lock()
-	defer w.mu.Unlock()
 
 	var result *multierror.Error
 
@@ -225,7 +224,15 @@ func (w *WGIface) Close() error {
 		result = multierror.Append(result, fmt.Errorf("failed to free WireGuard proxy: %w", err))
 	}
 
-	if err := w.tun.Close(); err != nil {
+	// Release w.mu before calling w.tun.Close(): the underlying
+	// wireguard-go device.Close() waits for its send/receive goroutines
+	// to drain. Some of those goroutines re-enter WGIface methods that
+	// take w.mu (e.g. the packet filter DNS hook calls GetDevice()), so
+	// holding the mutex here would deadlock the shutdown path.
+	tun := w.tun
+	w.mu.Unlock()
+
+	if err := tun.Close(); err != nil {
 		result = multierror.Append(result, fmt.Errorf("failed to close wireguard interface %s: %w", w.Name(), err))
 	}
 

client/iface/iface_close_test.go

@@ -0,0 +1,113 @@
+//go:build !android
+
+package iface
+
+import (
+	"errors"
+	"sync"
+	"testing"
+	"time"
+
+	wgdevice "golang.zx2c4.com/wireguard/device"
+	"golang.zx2c4.com/wireguard/tun/netstack"
+
+	"github.com/netbirdio/netbird/client/iface/device"
+	"github.com/netbirdio/netbird/client/iface/udpmux"
+	"github.com/netbirdio/netbird/client/iface/wgaddr"
+	"github.com/netbirdio/netbird/client/iface/wgproxy"
+)
+
+// fakeTunDevice implements WGTunDevice and lets the test control when
+// Close() returns. It mimics the wireguard-go shutdown path, which blocks
+// until its goroutines drain. Some of those goroutines (e.g. the packet
+// filter DNS hook in client/internal/dns) call back into WGIface, so if
+// WGIface.Close() held w.mu across tun.Close() the shutdown would
+// deadlock.
+type fakeTunDevice struct {
+	closeStarted chan struct{}
+	unblockClose chan struct{}
+}
+
+func (f *fakeTunDevice) Create() (device.WGConfigurer, error) {
+	return nil, errors.New("not implemented")
+}
+func (f *fakeTunDevice) Up() (*udpmux.UniversalUDPMuxDefault, error) {
+	return nil, errors.New("not implemented")
+}
+func (f *fakeTunDevice) UpdateAddr(wgaddr.Address) error      { return nil }
+func (f *fakeTunDevice) WgAddress() wgaddr.Address            { return wgaddr.Address{} }
+func (f *fakeTunDevice) MTU() uint16                          { return DefaultMTU }
+func (f *fakeTunDevice) DeviceName() string                   { return "nb-close-test" }
+func (f *fakeTunDevice) FilteredDevice() *device.FilteredDevice { return nil }
+func (f *fakeTunDevice) Device() *wgdevice.Device             { return nil }
+func (f *fakeTunDevice) GetNet() *netstack.Net                { return nil }
+func (f *fakeTunDevice) GetICEBind() device.EndpointManager   { return nil }
+
+func (f *fakeTunDevice) Close() error {
+	close(f.closeStarted)
+	<-f.unblockClose
+	return nil
+}
+
+type fakeProxyFactory struct{}
+
+func (fakeProxyFactory) GetProxy() wgproxy.Proxy { return nil }
+func (fakeProxyFactory) GetProxyPort() uint16    { return 0 }
+func (fakeProxyFactory) Free() error             { return nil }
+
+// TestWGIface_CloseReleasesMutexBeforeTunClose guards against a deadlock
+// that surfaces as a macOS test-timeout in
+// TestDNSPermanent_updateUpstream: WGIface.Close() used to hold w.mu
+// while waiting for the wireguard-go device goroutines to finish, and
+// one of those goroutines (the DNS filter hook) calls back into
+// WGIface.GetDevice() which needs the same mutex. The fix is to drop
+// the lock before tun.Close() returns control.
+func TestWGIface_CloseReleasesMutexBeforeTunClose(t *testing.T) {
+	tun := &fakeTunDevice{
+		closeStarted: make(chan struct{}),
+		unblockClose: make(chan struct{}),
+	}
+	w := &WGIface{
+		tun:            tun,
+		wgProxyFactory: fakeProxyFactory{},
+	}
+
+	closeDone := make(chan error, 1)
+	go func() {
+		closeDone <- w.Close()
+	}()
+
+	select {
+	case <-tun.closeStarted:
+	case <-time.After(2 * time.Second):
+		close(tun.unblockClose)
+		t.Fatal("tun.Close() was never invoked")
+	}
+
+	// Simulate the WireGuard read goroutine calling back into WGIface
+	// via the packet filter's DNS hook. If Close() still held w.mu
+	// during tun.Close(), this would block until the test timeout.
+	getDeviceDone := make(chan struct{})
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		_ = w.GetDevice()
+		close(getDeviceDone)
+	}()
+
+	select {
+	case <-getDeviceDone:
+	case <-time.After(2 * time.Second):
+		close(tun.unblockClose)
+		wg.Wait()
+		t.Fatal("GetDevice() deadlocked while WGIface.Close was closing the tun")
+	}
+
+	close(tun.unblockClose)
+	select {
+	case <-closeDone:
+	case <-time.After(2 * time.Second):
+		t.Fatal("WGIface.Close() never returned after the tun was unblocked")
+	}
+}

client/iface/udpmux/universal.go

@@ -171,7 +171,7 @@ func (u *UDPConn) performFilterCheck(addr net.Addr) error {
 	}
 
 	if u.address.Network.Contains(a) {
-		log.Warnf("Address %s is part of the NetBird network %s, refusing to write", addr, u.address)
+		log.Warnf("address %s is part of the NetBird network %s, refusing to write", addr, u.address)
 		return fmt.Errorf("address %s is part of the NetBird network %s, refusing to write", addr, u.address)
 	}
 
@@ -181,7 +181,7 @@ func (u *UDPConn) performFilterCheck(addr net.Addr) error {
 		u.addrCache.Store(addr.String(), isRouted)
 		if isRouted {
 			// Extra log, as the error only shows up with ICE logging enabled
-			log.Infof("Address %s is part of routed network %s, refusing to write", addr, prefix)
+			log.Infof("address %s is part of routed network %s, refusing to write", addr, prefix)
 			return fmt.Errorf("address %s is part of routed network %s, refusing to write", addr, prefix)
 		}
 	}

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.addPlatformLog(); err != nil {
-		if err := g.addLogfile(); err != nil {
+		log.Errorf("failed to add logs to debug bundle: %v", err)
-			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.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/dns/file_parser_unix.go

@@ -13,6 +13,7 @@ import (
 
 const (
 	defaultResolvConfPath = "/etc/resolv.conf"
+	nsswitchConfPath      = "/etc/nsswitch.conf"
 )
 
 type resolvConf struct {

client/internal/dns/host_unix.go

@@ -46,12 +46,12 @@ type restoreHostManager interface {
 }
 
 func newHostManager(wgInterface string) (hostManager, error) {
-	osManager, err := getOSDNSManagerType()
+	osManager, reason, err := getOSDNSManagerType()
 	if err != nil {
 		return nil, fmt.Errorf("get os dns manager type: %w", err)
 	}
 
-	log.Infof("System DNS manager discovered: %s", osManager)
+	log.Infof("System DNS manager discovered: %s (%s)", osManager, reason)
 	mgr, err := newHostManagerFromType(wgInterface, osManager)
 	// need to explicitly return nil mgr on error to avoid returning a non-nil interface containing a nil value
 	if err != nil {
@@ -74,17 +74,49 @@ func newHostManagerFromType(wgInterface string, osManager osManagerType) (restor
 	}
 }
 
-func getOSDNSManagerType() (osManagerType, error) {
+func getOSDNSManagerType() (osManagerType, string, error) {
+	resolved := isSystemdResolvedRunning()
+	nss := isLibnssResolveUsed()
+	stub := checkStub()
+
+	// Prefer systemd-resolved whenever it owns libc resolution, regardless of
+	// who wrote /etc/resolv.conf. File-mode rewrites do not affect lookups
+	// that go through nss-resolve, and in foreign mode they can loop back
+	// through resolved as an upstream.
+	if resolved && (nss || stub) {
+		return systemdManager, fmt.Sprintf("systemd-resolved active (nss-resolve=%t, stub=%t)", nss, stub), nil
+	}
+
+	mgr, reason, rejected, err := scanResolvConfHeader()
+	if err != nil {
+		return 0, "", err
+	}
+	if reason != "" {
+		return mgr, reason, nil
+	}
+
+	fallback := fmt.Sprintf("no manager matched (resolved=%t, nss-resolve=%t, stub=%t)", resolved, nss, stub)
+	if len(rejected) > 0 {
+		fallback += "; rejected: " + strings.Join(rejected, ", ")
+	}
+	return fileManager, fallback, nil
+}
+
+// scanResolvConfHeader walks /etc/resolv.conf header comments and returns the
+// matching manager. If reason is empty the caller should pick file mode and
+// use rejected for diagnostics.
+func scanResolvConfHeader() (osManagerType, string, []string, error) {
 	file, err := os.Open(defaultResolvConfPath)
 	if err != nil {
-		return 0, fmt.Errorf("unable to open %s for checking owner, got error: %w", defaultResolvConfPath, err)
+		return 0, "", nil, fmt.Errorf("unable to open %s for checking owner, got error: %w", defaultResolvConfPath, err)
 	}
 	defer func() {
-		if err := file.Close(); err != nil {
+		if cerr := file.Close(); cerr != nil {
-			log.Errorf("close file %s: %s", defaultResolvConfPath, err)
+			log.Errorf("close file %s: %s", defaultResolvConfPath, cerr)
 		}
 	}()
 
+	var rejected []string
 	scanner := bufio.NewScanner(file)
 	for scanner.Scan() {
 		text := scanner.Text()
@@ -92,41 +124,48 @@ func getOSDNSManagerType() (osManagerType, error) {
 			continue
 		}
 		if text[0] != '#' {
-			return fileManager, nil
+			break
 		}
-		if strings.Contains(text, fileGeneratedResolvConfContentHeader) {
+		if mgr, reason, rej := matchResolvConfHeader(text); reason != "" {
-			return netbirdManager, nil
+			return mgr, reason, nil, nil
-		}
+		} else if rej != "" {
-		if strings.Contains(text, "NetworkManager") && isDbusListenerRunning(networkManagerDest, networkManagerDbusObjectNode) && isNetworkManagerSupported() {
+			rejected = append(rejected, rej)
-			return networkManager, nil
-		}
-		if strings.Contains(text, "systemd-resolved") && isSystemdResolvedRunning() {
-			if checkStub() {
-				return systemdManager, nil
-			} else {
-				return fileManager, nil
-			}
-		}
-		if strings.Contains(text, "resolvconf") {
-			if isSystemdResolveConfMode() {
-				return systemdManager, nil
-			}
-
-			return resolvConfManager, nil
 		}
 	}
 	if err := scanner.Err(); err != nil && err != io.EOF {
-		return 0, fmt.Errorf("scan: %w", err)
+		return 0, "", nil, fmt.Errorf("scan: %w", err)
 	}
-
+	return 0, "", rejected, nil
-	return fileManager, nil
 }
 
-// checkStub checks if the stub resolver is disabled in systemd-resolved. If it is disabled, we fall back to file manager.
+// matchResolvConfHeader inspects a single comment line. Returns either a
+// definitive (manager, reason) or a non-empty rejected diagnostic.
+func matchResolvConfHeader(text string) (osManagerType, string, string) {
+	if strings.Contains(text, fileGeneratedResolvConfContentHeader) {
+		return netbirdManager, "netbird-managed resolv.conf header detected", ""
+	}
+	if strings.Contains(text, "NetworkManager") {
+		if isDbusListenerRunning(networkManagerDest, networkManagerDbusObjectNode) && isNetworkManagerSupported() {
+			return networkManager, "NetworkManager header + supported version on dbus", ""
+		}
+		return 0, "", "NetworkManager header (no dbus or unsupported version)"
+	}
+	if strings.Contains(text, "resolvconf") {
+		if isSystemdResolveConfMode() {
+			return systemdManager, "resolvconf header in systemd-resolved compatibility mode", ""
+		}
+		return resolvConfManager, "resolvconf header detected", ""
+	}
+	return 0, "", ""
+}
+
+// checkStub reports whether systemd-resolved's stub (127.0.0.53) is listed
+// in /etc/resolv.conf. On parse failure we assume it is, to avoid dropping
+// into file mode while resolved is active.
 func checkStub() bool {
 	rConf, err := parseDefaultResolvConf()
 	if err != nil {
-		log.Warnf("failed to parse resolv conf: %s", err)
+		log.Warnf("failed to parse resolv conf, assuming stub is active: %s", err)
 		return true
 	}
 
@@ -139,3 +178,36 @@ func checkStub() bool {
 
 	return false
 }
+
+// isLibnssResolveUsed reports whether nss-resolve is listed before dns on
+// the hosts: line of /etc/nsswitch.conf. When it is, libc lookups are
+// delegated to systemd-resolved regardless of /etc/resolv.conf.
+func isLibnssResolveUsed() bool {
+	bs, err := os.ReadFile(nsswitchConfPath)
+	if err != nil {
+		log.Debugf("read %s: %v", nsswitchConfPath, err)
+		return false
+	}
+	return parseNsswitchResolveAhead(bs)
+}
+
+func parseNsswitchResolveAhead(data []byte) bool {
+	for _, line := range strings.Split(string(data), "\n") {
+		if i := strings.IndexByte(line, '#'); i >= 0 {
+			line = line[:i]
+		}
+		fields := strings.Fields(line)
+		if len(fields) < 2 || fields[0] != "hosts:" {
+			continue
+		}
+		for _, module := range fields[1:] {
+			switch module {
+			case "dns":
+				return false
+			case "resolve":
+				return true
+			}
+		}
+	}
+	return false
+}

client/internal/dns/host_unix_test.go

@@ -0,0 +1,76 @@
+//go:build (linux && !android) || freebsd
+
+package dns
+
+import "testing"
+
+func TestParseNsswitchResolveAhead(t *testing.T) {
+	tests := []struct {
+		name string
+		in   string
+		want bool
+	}{
+		{
+			name: "resolve before dns with action token",
+			in:   "hosts: mymachines resolve [!UNAVAIL=return] files myhostname dns\n",
+			want: true,
+		},
+		{
+			name: "dns before resolve",
+			in:   "hosts: files mdns4_minimal [NOTFOUND=return] dns resolve\n",
+			want: false,
+		},
+		{
+			name: "debian default with only dns",
+			in:   "hosts: files mdns4_minimal [NOTFOUND=return] dns mymachines\n",
+			want: false,
+		},
+		{
+			name: "neither resolve nor dns",
+			in:   "hosts: files myhostname\n",
+			want: false,
+		},
+		{
+			name: "no hosts line",
+			in:   "passwd: files systemd\ngroup: files systemd\n",
+			want: false,
+		},
+		{
+			name: "empty",
+			in:   "",
+			want: false,
+		},
+		{
+			name: "comments and blank lines ignored",
+			in:   "# comment\n\n# another\nhosts: resolve dns\n",
+			want: true,
+		},
+		{
+			name: "trailing inline comment",
+			in:   "hosts: resolve [!UNAVAIL=return] dns # fallback\n",
+			want: true,
+		},
+		{
+			name: "hosts token must be the first field",
+			in:   "  hosts: resolve dns\n",
+			want: true,
+		},
+		{
+			name: "other db line mentioning resolve is ignored",
+			in:   "networks: resolve\nhosts: dns\n",
+			want: false,
+		},
+		{
+			name: "only resolve, no dns",
+			in:   "hosts: files resolve\n",
+			want: true,
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := parseNsswitchResolveAhead([]byte(tt.in)); got != tt.want {
+				t.Errorf("parseNsswitchResolveAhead() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}

client/internal/engine.go

@@ -26,6 +26,7 @@ import (
 
 	nberrors "github.com/netbirdio/netbird/client/errors"
 	"github.com/netbirdio/netbird/client/firewall"
+	"github.com/netbirdio/netbird/client/firewall/firewalld"
 	firewallManager "github.com/netbirdio/netbird/client/firewall/manager"
 	"github.com/netbirdio/netbird/client/iface"
 	"github.com/netbirdio/netbird/client/iface/device"
@@ -140,6 +141,7 @@ type EngineConfig struct {
 	ProfileConfig *profilemanager.Config
 
 	LogPath string
+	TempDir string
 }
 
 // EngineServices holds the external service dependencies required by the Engine.
@@ -569,7 +571,7 @@ 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()
@@ -603,6 +605,8 @@ func (e *Engine) createFirewall() error {
 		return nil
 	}
 
+	firewalld.SetParentContext(e.ctx)
+
 	var err error
 	e.firewall, err = firewall.NewFirewall(e.wgInterface, e.stateManager, e.flowManager.GetLogger(), e.config.DisableServerRoutes, e.config.MTU)
 	if err != nil {
@@ -1095,6 +1099,7 @@ 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()
+	forceRelay := IsForceRelayed()
-	workerICE, err := NewWorkerICE(conn.ctx, conn.Log, conn.config, conn, conn.signaler, conn.iFaceDiscover, conn.statusRecorder, relayIsSupportedLocally)
+	if !forceRelay {
-	if err != nil {
+		relayIsSupportedLocally := conn.workerRelay.RelayIsSupportedLocally()
-		return err
+		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) {
+// isConnectedOnAllWay evaluates the overall connection status based on ICE and Relay transports.
-	// would be better to protect this with a mutex, but it could cause deadlock with Close function
+//
-
+// 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
+	iceWorkerCreated := conn.workerICE != nil
-	if runtime.GOOS == "js" {
+
-		return conn.statusRelay.Get() == worker.StatusConnected
+	var iceInProgress bool
+	if iceWorkerCreated {
+		iceInProgress = conn.workerICE.InProgress()
 	}
 
-	// For non-JS platforms: check ICE connection status
+	return evalConnStatus(connStatusInputs{
-	if conn.statusICE.Get() == worker.StatusDisconnected && !conn.workerICE.InProgress() {
+		forceRelay:          IsForceRelayed(),
-		return false
+		peerUsesRelay:       conn.workerRelay.IsRelayConnectionSupportedWithPeer(),
-	}
+		relayConnected:      conn.statusRelay.Get() == worker.StatusConnected,
-
+		remoteSupportsICE:   conn.handshaker.RemoteICESupported(),
-	// If relay is supported with peer, it must also be connected
+		iceWorkerCreated:    iceWorkerCreated,
-	if conn.workerRelay.IsRelayConnectionSupportedWithPeer() {
+		iceStatusConnecting: conn.statusICE.Get() != worker.StatusDisconnected,
-		if conn.statusRelay.Get() == worker.StatusDisconnected {
+		iceInProgress:       iceInProgress,
-			return false
+	})
-		}
-	}
-
-	return true
 }
 
 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.
+// reconnectLoopWithRetry periodically checks the connection status and sends offers to re-establish connectivity.
-// Try to send offer while the P2P is not established or while the Relay is not connected if is it supported
+//
+// 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:
+		case <-tickerChannel:
-			if t.IsZero() {
+			switch g.isConnectedOnAllWay() {
-				g.log.Infof("retry timed out, stop periodic offer sending")
+			case ConnStatusConnected:
-				// after backoff timeout the ticker.C will be closed. We need to a dummy channel to avoid loop
+				// all good, nothing to do
-				tickerChannel = make(<-chan time.Time)
+			case ConnStatusDisconnected:
-				continue
+				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())
+	if !disableICEMonitor {
-	go iceMonitor.Start(ctx, w.onICEChanged)
+		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()
+	var sessionIDBytes []byte
-	if err != nil {
+	if offerAnswer.SessionID != nil {
-		log.Warnf("failed to get session ID bytes: %v", err)
+		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()
+	msg, err := r.buildRouteMessage(action, prefix, nexthop)
-	expBackOff.InitialInterval = 50 * time.Millisecond
+	if err != nil {
-	expBackOff.MaxInterval = 500 * time.Millisecond
+		return fmt.Errorf("build route message: %w", err)
-	expBackOff.MaxElapsedTime = 1 * time.Second
+	}
 
-	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 {
+// writeRouteMessage sends a route message over AF_ROUTE and waits for the
-	operation := func() error {
+// kernel's matching reply, retrying transient failures until budget elapses.
-		fd, err := unix.Socket(syscall.AF_ROUTE, syscall.SOCK_RAW, syscall.AF_UNSPEC)
+// Callers do not need to manage sockets or seq numbers themselves.
-		if err != nil {
+func (r *SysOps) writeRouteMessage(msg *route.RouteMessage, budget time.Duration) error {
-			return fmt.Errorf("open routing socket: %w", err)
+	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
 			}
-		}()
+			return backoff.Permanent(fmt.Errorf("read: %w", err))
-
-		msg, err := r.buildRouteMessage(action, prefix, nexthop)
-		if err != nil {
-			return backoff.Permanent(fmt.Errorf("build route message: %w", err))
 		}
-
+		if n < int(unsafe.Sizeof(unix.RtMsghdr{})) {
-		msgBytes, err := msg.Marshal()
+			continue
-		if err != nil {
-			return backoff.Permanent(fmt.Errorf("marshal route message: %w", err))
 		}
-
+		hdr := (*unix.RtMsghdr)(unsafe.Pointer(&resp[0]))
-		if _, err = unix.Write(fd, msgBytes); err != nil {
+		// Darwin reflects the sender's pid on replies; matching (Type, Seq, Pid)
-			if errors.Is(err, unix.ENOBUFS) || errors.Is(err, unix.EAGAIN) {
+		// uniquely identifies our own reply among broadcast traffic.
-				return fmt.Errorf("write: %w", err)
+		if int(hdr.Type) != wantType || int(hdr.Seq) != wantSeq || hdr.Pid != pid {
-			}
+			continue
-			return backoff.Permanent(fmt.Errorf("write: %w", err))
 		}
-
+		if hdr.Errno != 0 {
-		respBuf := make([]byte, 2048)
+			return backoff.Permanent(fmt.Errorf("kernel: %w", syscall.Errno(hdr.Errno)))
-		n, err := unix.Read(fd, respBuf)
-		if err != nil {
-			return backoff.Permanent(fmt.Errorf("read route response: %w", err))
 		}
-
-		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() {

client/net/dialer_init_darwin.go

@@ -0,0 +1,5 @@
+package net
+
+func (d *Dialer) init() {
+	d.Dialer.Control = applyBoundIfToSocket
+}

client/net/dialer_init_generic.go

@@ -1,4 +1,4 @@
-//go:build !linux && !windows
+//go:build !linux && !windows && !darwin
 
 package net
 

client/net/env_android.go

@@ -1,24 +0,0 @@
-//go:build android
-
-package net
-
-// Init initializes the network environment for Android
-func Init() {
-	// No initialization needed on Android
-}
-
-// AdvancedRouting reports whether routing loops can be avoided without using exclusion routes.
-// Always returns true on Android since we cannot handle routes dynamically.
-func AdvancedRouting() bool {
-	return true
-}
-
-// SetVPNInterfaceName is a no-op on Android
-func SetVPNInterfaceName(name string) {
-	// No-op on Android - not needed for Android VPN service
-}
-
-// GetVPNInterfaceName returns empty string on Android
-func GetVPNInterfaceName() string {
-	return ""
-}

client/net/env_bound_iface.go

@@ -1,4 +1,4 @@
-//go:build windows
+//go:build (darwin && !ios) || windows
 
 package net
 
@@ -24,17 +24,22 @@ func Init() {
 }
 
 func checkAdvancedRoutingSupport() bool {
-	var err error
+	legacyRouting := false
-	var legacyRouting bool
 	if val := os.Getenv(envUseLegacyRouting); val != "" {
-		legacyRouting, err = strconv.ParseBool(val)
+		parsed, err := strconv.ParseBool(val)
 		if err != nil {
-			log.Warnf("failed to parse %s: %v", envUseLegacyRouting, err)
+			log.Warnf("ignoring unparsable %s=%q: %v", envUseLegacyRouting, val, err)
+		} else {
+			legacyRouting = parsed
 		}
 	}
 
-	if legacyRouting || netstack.IsEnabled() {
+	if legacyRouting {
-		log.Info("advanced routing has been requested to be disabled")
+		log.Infof("advanced routing disabled: legacy routing requested via %s", envUseLegacyRouting)
+		return false
+	}
+	if netstack.IsEnabled() {
+		log.Info("advanced routing disabled: netstack mode is enabled")
 		return false
 	}
 

client/net/env_generic.go

@@ -1,4 +1,4 @@
-//go:build !linux && !windows && !android
+//go:build !linux && !windows && !darwin
 
 package net
 

client/net/env_mobile.go

@@ -0,0 +1,25 @@
+//go:build ios || android
+
+package net
+
+// Init initializes the network environment for mobile platforms.
+func Init() {
+	// no-op on mobile: routing scope is owned by the VPN extension.
+}
+
+// AdvancedRouting reports whether routing loops can be avoided without using exclusion routes.
+// Always returns true on mobile since routes cannot be handled dynamically and the VPN extension
+// owns the routing scope.
+func AdvancedRouting() bool {
+	return true
+}
+
+// SetVPNInterfaceName is a no-op on mobile.
+func SetVPNInterfaceName(string) {
+	// no-op on mobile: the VPN extension manages the interface.
+}
+
+// GetVPNInterfaceName returns an empty string on mobile.
+func GetVPNInterfaceName() string {
+	return ""
+}

client/net/listener_init_darwin.go

@@ -0,0 +1,5 @@
+package net
+
+func (l *ListenerConfig) init() {
+	l.ListenConfig.Control = applyBoundIfToSocket
+}

client/net/listener_init_generic.go

@@ -1,4 +1,4 @@
-//go:build !linux && !windows
+//go:build !linux && !windows && !darwin
 
 package net
 

client/net/net_darwin.go

@@ -0,0 +1,160 @@
+package net
+
+import (
+	"fmt"
+	"net"
+	"net/netip"
+	"strconv"
+	"strings"
+	"sync"
+	"syscall"
+
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
+)
+
+// On darwin IPV6_BOUND_IF also scopes v4-mapped egress from dual-stack
+// (IPV6_V6ONLY=0) AF_INET6 sockets, so a single setsockopt on "udp6"/"tcp6"
+// covers both families. Setting IP_BOUND_IF on an AF_INET6 socket returns
+// EINVAL regardless of V6ONLY because the IPPROTO_IP ctloutput path is
+// dispatched by socket domain (AF_INET only) not by inp_vflag.
+
+// boundIface holds the physical interface chosen at routing setup time. Sockets
+// created via nbnet.NewDialer / nbnet.NewListener bind to it via IP_BOUND_IF
+// (IPv4) or IPV6_BOUND_IF (IPv6 / dual-stack) so their scoped route lookup
+// hits the RTF_IFSCOPE default installed by the routemanager, rather than
+// following the VPN's split default.
+var (
+	boundIfaceMu sync.RWMutex
+	boundIface4  *net.Interface
+	boundIface6  *net.Interface
+)
+
+// SetBoundInterface records the egress interface for an address family. Called
+// by the routemanager after a scoped default route has been installed.
+// af must be unix.AF_INET or unix.AF_INET6; other values are ignored.
+// nil iface is rejected — use ClearBoundInterfaces to clear all slots.
+func SetBoundInterface(af int, iface *net.Interface) {
+	if iface == nil {
+		log.Warnf("SetBoundInterface: nil iface for AF %d, ignored", af)
+		return
+	}
+	boundIfaceMu.Lock()
+	defer boundIfaceMu.Unlock()
+	switch af {
+	case unix.AF_INET:
+		boundIface4 = iface
+	case unix.AF_INET6:
+		boundIface6 = iface
+	default:
+		log.Warnf("SetBoundInterface: unsupported address family %d", af)
+	}
+}
+
+// ClearBoundInterfaces resets the cached egress interfaces. Called by the
+// routemanager during cleanup.
+func ClearBoundInterfaces() {
+	boundIfaceMu.Lock()
+	defer boundIfaceMu.Unlock()
+	boundIface4 = nil
+	boundIface6 = nil
+}
+
+// boundInterfaceFor returns the cached egress interface for a socket's address
+// family, falling back to the other family if the preferred slot is empty.
+// The kernel stores both IP_BOUND_IF and IPV6_BOUND_IF in inp_boundifp, so
+// either setsockopt scopes the socket; preferring same-family still matters
+// when v4 and v6 defaults egress different NICs.
+func boundInterfaceFor(network, address string) *net.Interface {
+	if iface := zoneInterface(address); iface != nil {
+		return iface
+	}
+
+	boundIfaceMu.RLock()
+	defer boundIfaceMu.RUnlock()
+
+	primary, secondary := boundIface4, boundIface6
+	if isV6Network(network) {
+		primary, secondary = boundIface6, boundIface4
+	}
+	if primary != nil {
+		return primary
+	}
+	return secondary
+}
+
+func isV6Network(network string) bool {
+	return strings.HasSuffix(network, "6")
+}
+
+// zoneInterface extracts an explicit interface from an IPv6 link-local zone (e.g. fe80::1%en0).
+func zoneInterface(address string) *net.Interface {
+	if address == "" {
+		return nil
+	}
+	addr, err := netip.ParseAddrPort(address)
+	if err != nil {
+		a, err := netip.ParseAddr(address)
+		if err != nil {
+			return nil
+		}
+		addr = netip.AddrPortFrom(a, 0)
+	}
+	zone := addr.Addr().Zone()
+	if zone == "" {
+		return nil
+	}
+	if iface, err := net.InterfaceByName(zone); err == nil {
+		return iface
+	}
+	if idx, err := strconv.Atoi(zone); err == nil {
+		if iface, err := net.InterfaceByIndex(idx); err == nil {
+			return iface
+		}
+	}
+	return nil
+}
+
+func setIPv4BoundIf(fd uintptr, iface *net.Interface) error {
+	if err := unix.SetsockoptInt(int(fd), unix.IPPROTO_IP, unix.IP_BOUND_IF, iface.Index); err != nil {
+		return fmt.Errorf("set IP_BOUND_IF: %w (interface: %s, index: %d)", err, iface.Name, iface.Index)
+	}
+	return nil
+}
+
+func setIPv6BoundIf(fd uintptr, iface *net.Interface) error {
+	if err := unix.SetsockoptInt(int(fd), unix.IPPROTO_IPV6, unix.IPV6_BOUND_IF, iface.Index); err != nil {
+		return fmt.Errorf("set IPV6_BOUND_IF: %w (interface: %s, index: %d)", err, iface.Name, iface.Index)
+	}
+	return nil
+}
+
+// applyBoundIfToSocket binds the socket to the cached physical egress interface
+// so scoped route lookup avoids the VPN utun and egresses the underlay directly.
+func applyBoundIfToSocket(network, address string, c syscall.RawConn) error {
+	if !AdvancedRouting() {
+		return nil
+	}
+
+	iface := boundInterfaceFor(network, address)
+	if iface == nil {
+		log.Debugf("no bound iface cached for %s to %s, skipping BOUND_IF", network, address)
+		return nil
+	}
+
+	isV6 := isV6Network(network)
+	var controlErr error
+	if err := c.Control(func(fd uintptr) {
+		if isV6 {
+			controlErr = setIPv6BoundIf(fd, iface)
+		} else {
+			controlErr = setIPv4BoundIf(fd, iface)
+		}
+		if controlErr == nil {
+			log.Debugf("set BOUND_IF=%d on %s for %s to %s", iface.Index, iface.Name, network, address)
+		}
+	}); err != nil {
+		return fmt.Errorf("control: %w", err)
+	}
+	return controlErr
+}

client/server/state.go

@@ -12,7 +12,6 @@ import (
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/routemanager/systemops"
 	"github.com/netbirdio/netbird/client/internal/statemanager"
-	nbnet "github.com/netbirdio/netbird/client/net"
 	"github.com/netbirdio/netbird/client/proto"
 )
 
@@ -138,10 +137,8 @@ func restoreResidualState(ctx context.Context, statePath string) error {
 	}
 
 	// clean up any remaining routes independently of the state file
-	if !nbnet.AdvancedRouting() {
+	if err := systemops.New(nil, nil).FlushMarkedRoutes(); err != nil {
-		if err := systemops.New(nil, nil).FlushMarkedRoutes(); err != nil {
+		merr = multierror.Append(merr, fmt.Errorf("flush marked routes: %w", err))
-			merr = multierror.Append(merr, fmt.Errorf("flush marked routes: %w", err))
-		}
 	}
 
 	return nberrors.FormatErrorOrNil(merr)

client/ssh/config/manager.go

@@ -187,24 +187,23 @@ func (m *Manager) buildPeerConfig(allHostPatterns []string) (string, error) {
 		return "", fmt.Errorf("get NetBird executable path: %w", err)
 	}
 
-	hostLine := strings.Join(deduplicatedPatterns, " ")
+	hostList := strings.Join(deduplicatedPatterns, ",")
-	config := fmt.Sprintf("Host %s\n", hostLine)
+	config := fmt.Sprintf("Match host \"%s\" exec \"%s ssh detect %%h %%p\"\n", hostList, execPath)
-	config += fmt.Sprintf("    Match exec \"%s ssh detect %%h %%p\"\n", execPath)
+	config += "    PreferredAuthentications password,publickey,keyboard-interactive\n"
-	config += "        PreferredAuthentications password,publickey,keyboard-interactive\n"
+	config += "    PasswordAuthentication yes\n"
-	config += "        PasswordAuthentication yes\n"
+	config += "    PubkeyAuthentication yes\n"
-	config += "        PubkeyAuthentication yes\n"
+	config += "    BatchMode no\n"
-	config += "        BatchMode no\n"
+	config += fmt.Sprintf("    ProxyCommand %s ssh proxy %%h %%p\n", execPath)
-	config += fmt.Sprintf("        ProxyCommand %s ssh proxy %%h %%p\n", execPath)
+	config += "    StrictHostKeyChecking no\n"
-	config += "        StrictHostKeyChecking no\n"
 
 	if runtime.GOOS == "windows" {
-		config += "        UserKnownHostsFile NUL\n"
+		config += "    UserKnownHostsFile NUL\n"
 	} else {
-		config += "        UserKnownHostsFile /dev/null\n"
+		config += "    UserKnownHostsFile /dev/null\n"
 	}
 
-	config += "        CheckHostIP no\n"
+	config += "    CheckHostIP no\n"
-	config += "        LogLevel ERROR\n\n"
+	config += "    LogLevel ERROR\n\n"
 
 	return config, nil
 }

client/ssh/config/manager_test.go

@@ -116,6 +116,37 @@ func TestManager_PeerLimit(t *testing.T) {
 	assert.True(t, os.IsNotExist(err), "SSH config should not be created with too many peers")
 }
 
+func TestManager_MatchHostFormat(t *testing.T) {
+	tempDir, err := os.MkdirTemp("", "netbird-ssh-config-test")
+	require.NoError(t, err)
+	defer func() { assert.NoError(t, os.RemoveAll(tempDir)) }()
+
+	manager := &Manager{
+		sshConfigDir:  filepath.Join(tempDir, "ssh_config.d"),
+		sshConfigFile: "99-netbird.conf",
+	}
+
+	peers := []PeerSSHInfo{
+		{Hostname: "peer1", IP: "100.125.1.1", FQDN: "peer1.nb.internal"},
+		{Hostname: "peer2", IP: "100.125.1.2", FQDN: "peer2.nb.internal"},
+	}
+
+	err = manager.SetupSSHClientConfig(peers)
+	require.NoError(t, err)
+
+	configPath := filepath.Join(manager.sshConfigDir, manager.sshConfigFile)
+	content, err := os.ReadFile(configPath)
+	require.NoError(t, err)
+	configStr := string(content)
+
+	// Must use "Match host" with comma-separated patterns, not a bare "Host" directive.
+	// A bare "Host" followed by "Match exec" is incorrect per ssh_config(5): the Host block
+	// ends at the next Match keyword, making it a no-op and leaving the Match exec unscoped.
+	assert.NotContains(t, configStr, "\nHost ", "should not use bare Host directive")
+	assert.Contains(t, configStr, "Match host \"100.125.1.1,peer1.nb.internal,peer1,100.125.1.2,peer2.nb.internal,peer2\"",
+		"should use Match host with comma-separated patterns")
+}
+
 func TestManager_ForcedSSHConfig(t *testing.T) {
 	// Set force environment variable
 	t.Setenv(EnvForceSSHConfig, "true")

client/system/info.go

@@ -2,7 +2,6 @@ package system
 
 import (
 	"context"
-	"net"
 	"net/netip"
 	"strings"
 
@@ -145,59 +144,6 @@ func extractDeviceName(ctx context.Context, defaultName string) string {
 	return v
 }
 
-func networkAddresses() ([]NetworkAddress, error) {
-	interfaces, err := net.Interfaces()
-	if err != nil {
-		return nil, err
-	}
-
-	var netAddresses []NetworkAddress
-	for _, iface := range interfaces {
-		if iface.Flags&net.FlagUp == 0 {
-			continue
-		}
-		if iface.HardwareAddr.String() == "" {
-			continue
-		}
-		addrs, err := iface.Addrs()
-		if err != nil {
-			continue
-		}
-
-		for _, address := range addrs {
-			ipNet, ok := address.(*net.IPNet)
-			if !ok {
-				continue
-			}
-
-			if ipNet.IP.IsLoopback() {
-				continue
-			}
-
-			netAddr := NetworkAddress{
-				NetIP: netip.MustParsePrefix(ipNet.String()),
-				Mac:   iface.HardwareAddr.String(),
-			}
-
-			if isDuplicated(netAddresses, netAddr) {
-				continue
-			}
-
-			netAddresses = append(netAddresses, netAddr)
-		}
-	}
-	return netAddresses, nil
-}
-
-func isDuplicated(addresses []NetworkAddress, addr NetworkAddress) bool {
-	for _, duplicated := range addresses {
-		if duplicated.NetIP == addr.NetIP {
-			return true
-		}
-	}
-	return false
-}
-
 // GetInfoWithChecks retrieves and parses the system information with applied checks.
 func GetInfoWithChecks(ctx context.Context, checks []*proto.Checks) (*Info, error) {
 	log.Debugf("gathering system information with checks: %d", len(checks))

client/system/info_ios.go

@@ -2,6 +2,8 @@ package system
 
 import (
 	"context"
+	"net"
+	"net/netip"
 	"runtime"
 
 	log "github.com/sirupsen/logrus"
@@ -42,6 +44,66 @@ func GetInfo(ctx context.Context) *Info {
 	return gio
 }
 
+// networkAddresses returns the list of network addresses on iOS.
+// On iOS, hardware (MAC) addresses are not available due to Apple's privacy
+// restrictions (iOS returns a fixed 02:00:00:00:00:00 placeholder), so we
+// leave Mac empty to match Android's behavior. We also skip the HardwareAddr
+// check that other platforms use and filter out link-local addresses as they
+// are not useful for posture checks.
+func networkAddresses() ([]NetworkAddress, error) {
+	interfaces, err := net.Interfaces()
+	if err != nil {
+		return nil, err
+	}
+
+	var netAddresses []NetworkAddress
+	for _, iface := range interfaces {
+		if iface.Flags&net.FlagUp == 0 {
+			continue
+		}
+		addrs, err := iface.Addrs()
+		if err != nil {
+			continue
+		}
+
+		for _, address := range addrs {
+			netAddr, ok := toNetworkAddress(address)
+			if !ok {
+				continue
+			}
+			if isDuplicated(netAddresses, netAddr) {
+				continue
+			}
+			netAddresses = append(netAddresses, netAddr)
+		}
+	}
+	return netAddresses, nil
+}
+
+func toNetworkAddress(address net.Addr) (NetworkAddress, bool) {
+	ipNet, ok := address.(*net.IPNet)
+	if !ok {
+		return NetworkAddress{}, false
+	}
+	if ipNet.IP.IsLoopback() || ipNet.IP.IsLinkLocalUnicast() || ipNet.IP.IsMulticast() {
+		return NetworkAddress{}, false
+	}
+	prefix, err := netip.ParsePrefix(ipNet.String())
+	if err != nil {
+		return NetworkAddress{}, false
+	}
+	return NetworkAddress{NetIP: prefix, Mac: ""}, true
+}
+
+func isDuplicated(addresses []NetworkAddress, addr NetworkAddress) bool {
+	for _, duplicated := range addresses {
+		if duplicated.NetIP == addr.NetIP {
+			return true
+		}
+	}
+	return false
+}
+
 // checkFileAndProcess checks if the file path exists and if a process is running at that path.
 func checkFileAndProcess(paths []string) ([]File, error) {
 	return []File{}, nil

client/system/network_addr.go

@@ -0,0 +1,66 @@
+//go:build !ios
+
+package system
+
+import (
+	"net"
+	"net/netip"
+)
+
+func networkAddresses() ([]NetworkAddress, error) {
+	interfaces, err := net.Interfaces()
+	if err != nil {
+		return nil, err
+	}
+
+	var netAddresses []NetworkAddress
+	for _, iface := range interfaces {
+		if iface.Flags&net.FlagUp == 0 {
+			continue
+		}
+		if iface.HardwareAddr.String() == "" {
+			continue
+		}
+		addrs, err := iface.Addrs()
+		if err != nil {
+			continue
+		}
+
+		mac := iface.HardwareAddr.String()
+		for _, address := range addrs {
+			netAddr, ok := toNetworkAddress(address, mac)
+			if !ok {
+				continue
+			}
+			if isDuplicated(netAddresses, netAddr) {
+				continue
+			}
+			netAddresses = append(netAddresses, netAddr)
+		}
+	}
+	return netAddresses, nil
+}
+
+func toNetworkAddress(address net.Addr, mac string) (NetworkAddress, bool) {
+	ipNet, ok := address.(*net.IPNet)
+	if !ok {
+		return NetworkAddress{}, false
+	}
+	if ipNet.IP.IsLoopback() {
+		return NetworkAddress{}, false
+	}
+	prefix, err := netip.ParsePrefix(ipNet.String())
+	if err != nil {
+		return NetworkAddress{}, false
+	}
+	return NetworkAddress{NetIP: prefix, Mac: mac}, true
+}
+
+func isDuplicated(addresses []NetworkAddress, addr NetworkAddress) bool {
+	for _, duplicated := range addresses {
+		if duplicated.NetIP == addr.NetIP {
+			return true
+		}
+	}
+	return false
+}

combined/config.yaml.example

@@ -119,6 +119,8 @@ server:
 
   # Reverse proxy settings (optional)
   # reverseProxy:
-  #   trustedHTTPProxies: []
+  #   trustedHTTPProxies: []           # CIDRs of trusted reverse proxies (e.g. ["10.0.0.0/8"])
-  #   trustedHTTPProxiesCount: 0
+  #   trustedHTTPProxiesCount: 0       # Number of trusted proxies in front of the server (alternative to trustedHTTPProxies)
-  #   trustedPeers: []
+  #   trustedPeers: []                 # CIDRs of trusted peer networks (e.g. ["100.64.0.0/10"])
+  #   accessLogRetentionDays: 7        # Days to retain HTTP access logs. 0 (or unset) defaults to 7. Negative values disable cleanup (logs kept indefinitely).
+  #   accessLogCleanupIntervalHours: 24 # How often (in hours) to run the access-log cleanup job. 0 (or unset) is treated as "not set" and defaults to 24 hours; cleanup remains enabled. To disable cleanup, set accessLogRetentionDays to a negative value.

go.mod

@@ -323,3 +323,5 @@ replace github.com/pion/ice/v4 => github.com/netbirdio/ice/v4 v4.0.0-20250908184
 replace github.com/libp2p/go-netroute => github.com/netbirdio/go-netroute v0.0.0-20240611143515-f59b0e1d3944
 
 replace github.com/dexidp/dex => github.com/netbirdio/dex v0.244.0
+
+replace github.com/mailru/easyjson => github.com/netbirdio/easyjson v0.9.0

go.sum

@@ -400,8 +400,6 @@ github.com/lufia/plan9stats v0.0.0-20240513124658-fba389f38bae h1:dIZY4ULFcto4tA
 github.com/lufia/plan9stats v0.0.0-20240513124658-fba389f38bae/go.mod h1:ilwx/Dta8jXAgpFYFvSWEMwxmbWXyiUHkd5FwyKhb5k=
 github.com/magiconair/properties v1.8.10 h1:s31yESBquKXCV9a/ScB3ESkOjUYYv+X0rg8SYxI99mE=
 github.com/magiconair/properties v1.8.10/go.mod h1:Dhd985XPs7jluiymwWYZ0G4Z61jb3vdS329zhj2hYo0=
-github.com/mailru/easyjson v0.9.0 h1:PrnmzHw7262yW8sTBwxi1PdJA3Iw/EKBa8psRf7d9a4=
-github.com/mailru/easyjson v0.9.0/go.mod h1:1+xMtQp2MRNVL/V1bOzuP3aP8VNwRW55fQUto+XFtTU=
 github.com/mattermost/xml-roundtrip-validator v0.1.0 h1:RXbVD2UAl7A7nOTR4u7E3ILa4IbtvKBHw64LDsmu9hU=
 github.com/mattermost/xml-roundtrip-validator v0.1.0/go.mod h1:qccnGMcpgwcNaBnxqpJpWWUiPNr5H3O8eDgGV9gT5To=
 github.com/mattn/go-isatty v0.0.9/go.mod h1:YNRxwqDuOph6SZLI9vUUz6OYw3QyUt7WiY2yME+cCiQ=
@@ -449,6 +447,8 @@ github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822 h1:C3w9PqII01/Oq
 github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822/go.mod h1:+n7T8mK8HuQTcFwEeznm/DIxMOiR9yIdICNftLE1DvQ=
 github.com/netbirdio/dex v0.244.0 h1:1GOvi8wnXYassnKGildzNqRHq0RbcfEUw7LKYpKIN7U=
 github.com/netbirdio/dex v0.244.0/go.mod h1:STGInJhPcAflrHmDO7vyit2kSq03PdL+8zQPoGALtcU=
+github.com/netbirdio/easyjson v0.9.0 h1:6Nw2lghSVuy8RSkAYDhDv1thBVEmfVbKZnV7T7Z6Aus=
+github.com/netbirdio/easyjson v0.9.0/go.mod h1:1+xMtQp2MRNVL/V1bOzuP3aP8VNwRW55fQUto+XFtTU=
 github.com/netbirdio/go-netroute v0.0.0-20240611143515-f59b0e1d3944 h1:TDtJKmM6Sf8uYFx/dMeqNOL90KUoRscdfpFZ3Im89uk=
 github.com/netbirdio/go-netroute v0.0.0-20240611143515-f59b0e1d3944/go.mod h1:sHA6TRxjQ6RLbnI+3R4DZo2Eseg/iKiPRfNmcuNySVQ=
 github.com/netbirdio/ice/v4 v4.0.0-20250908184934-6202be846b51 h1:Ov4qdafATOgGMB1wbSuh+0aAHcwz9hdvB6VZjh1mVMI=

infrastructure_files/getting-started.sh

@@ -472,7 +472,7 @@ start_services_and_show_instructions() {
       if [[ "$ENABLE_CROWDSEC" == "true" ]]; then
         echo "Registering CrowdSec bouncer..."
         local cs_retries=0
-        while ! $DOCKER_COMPOSE_COMMAND exec -T crowdsec cscli capi status >/dev/null 2>&1; do
+        while ! $DOCKER_COMPOSE_COMMAND exec -T crowdsec cscli lapi status >/dev/null 2>&1; do
           cs_retries=$((cs_retries + 1))
           if [[ $cs_retries -ge 30 ]]; then
             echo "WARNING: CrowdSec did not become ready. Skipping CrowdSec setup." > /dev/stderr