Update client/internal/dns/handler_chain.go

Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>

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)
-	return c.connectClient.RunOnAndroid(c.tunAdapter, c.iFaceDiscover, c.networkChangeListener, slices.Clone(dns.items), dnsReadyListener, stateFile)
+	connectClient := internal.NewConnectClient(ctx, cfg, c.recorder)
+	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)
-	return c.connectClient.RunOnAndroid(c.tunAdapter, c.iFaceDiscover, c.networkChangeListener, slices.Clone(dns.items), dnsReadyListener, stateFile)
+	connectClient := internal.NewConnectClient(ctx, cfg, c.recorder)
+	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/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.addLogfile(); err != nil {
-			log.Errorf("failed to add log file to debug bundle: %v", err)
-			if err := g.trySystemdLogFallback(); err != nil {
-				log.Errorf("failed to add systemd logs as fallback: %v", err)
-			}
-		}
-	} else if err := g.trySystemdLogFallback(); err != nil {
-		log.Errorf("failed to add systemd logs: %v", err)
+	if err := g.addPlatformLog(); err != nil {
+		log.Errorf("failed to add logs to debug bundle: %v", err)
 	}
 
 	if err := g.addUpdateLogs(); err != nil {

client/internal/debug/debug_android.go

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

client/internal/debug/debug_nonandroid.go

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

client/internal/dns/handler_chain.go

@@ -1,7 +1,10 @@
 package dns
 
 import (
+	"context"
 	"fmt"
+	"math"
+	"net"
 	"slices"
 	"strconv"
 	"strings"
@@ -192,6 +195,12 @@ func (c *HandlerChain) logHandlers() {
 }
 
 func (c *HandlerChain) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
+	c.dispatch(w, r, math.MaxInt)
+}
+
+// dispatch routes a DNS request through the chain, skipping handlers with
+// priority > maxPriority. Shared by ServeDNS and ResolveInternal.
+func (c *HandlerChain) dispatch(w dns.ResponseWriter, r *dns.Msg, maxPriority int) {
 	if len(r.Question) == 0 {
 		return
 	}
@@ -216,6 +225,9 @@ func (c *HandlerChain) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 
 	// Try handlers in priority order
 	for _, entry := range handlers {
+		if entry.Priority > maxPriority {
+			continue
+		}
 		if !c.isHandlerMatch(qname, entry) {
 			continue
 		}
@@ -273,6 +285,55 @@ func (c *HandlerChain) logResponse(logger *log.Entry, cw *ResponseWriterChain, q
 		cw.response.Len(), meta, time.Since(startTime))
 }
 
+// ResolveInternal runs an in-process DNS query against the chain, skipping any
+// handler with priority > maxPriority. Used by internal callers (e.g. the mgmt
+// cache refresher) that must bypass themselves to avoid loops. Honors ctx
+// cancellation; on ctx.Done the dispatch goroutine is left to drain on its own
+// (bounded by the invoked handler's internal timeout).
+func (c *HandlerChain) ResolveInternal(ctx context.Context, r *dns.Msg, maxPriority int) (*dns.Msg, error) {
+	if len(r.Question) == 0 {
+		return nil, fmt.Errorf("empty question")
+	}
+
+	base := &internalResponseWriter{}
+	done := make(chan struct{})
+	go func() {
+		c.dispatch(base, r, maxPriority)
+		close(done)
+	}()
+
+	select {
+	case <-done:
+	case <-ctx.Done():
+		// Prefer a completed response if dispatch finished concurrently with cancellation.
+		select {
+		case <-done:
+		default:
+			return nil, fmt.Errorf("resolve %s: %w", strings.ToLower(r.Question[0].Name), ctx.Err())
+		}
+	}
+
+	if base.response == nil || base.response.Rcode == dns.RcodeRefused {
+		return nil, fmt.Errorf("no handler resolved %s at priority ≤ %d",
+			strings.ToLower(r.Question[0].Name), maxPriority)
+	}
+	return base.response, nil
+}
+
+// HasRootHandlerAtOrBelow reports whether any "." handler is registered at
+// priority ≤ maxPriority.
+func (c *HandlerChain) HasRootHandlerAtOrBelow(maxPriority int) bool {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+
+	for _, h := range c.handlers {
+		if h.Pattern == "." && h.Priority <= maxPriority {
+			return true
+		}
+	}
+	return false
+}
+
 func (c *HandlerChain) isHandlerMatch(qname string, entry HandlerEntry) bool {
 	switch {
 	case entry.Pattern == ".":
@@ -291,3 +352,36 @@ func (c *HandlerChain) isHandlerMatch(qname string, entry HandlerEntry) bool {
 		}
 	}
 }
+
+// internalResponseWriter captures a dns.Msg for in-process chain queries.
+type internalResponseWriter struct {
+	response *dns.Msg
+}
+
+func (w *internalResponseWriter) WriteMsg(m *dns.Msg) error { w.response = m; return nil }
+func (w *internalResponseWriter) LocalAddr() net.Addr       { return nil }
+func (w *internalResponseWriter) RemoteAddr() net.Addr      { return nil }
+
+// Write unpacks raw DNS bytes so handlers that call Write instead of WriteMsg
+// still surface their answer to ResolveInternal.
+func (w *internalResponseWriter) Write(p []byte) (int, error) {
+	msg := new(dns.Msg)
+	if err := msg.Unpack(p); err != nil {
+		return 0, err
+	}
+	w.response = msg
+	return len(p), nil
+}
+
+func (w *internalResponseWriter) Close() error      { return nil }
+func (w *internalResponseWriter) TsigStatus() error { return nil }
+
+// TsigTimersOnly is part of dns.ResponseWriter.
+func (w *internalResponseWriter) TsigTimersOnly(bool) {
+	// no-op: in-process queries carry no TSIG state.
+}
+
+// Hijack is part of dns.ResponseWriter.
+func (w *internalResponseWriter) Hijack() {
+	// no-op: in-process queries have no underlying connection to hand off.
+}

client/internal/dns/handler_chain_test.go

@@ -1,11 +1,15 @@
 package dns_test
 
 import (
+	"context"
+	"net"
 	"testing"
+	"time"
 
 	"github.com/miekg/dns"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/mock"
+	"github.com/stretchr/testify/require"
 
 	nbdns "github.com/netbirdio/netbird/client/internal/dns"
 	"github.com/netbirdio/netbird/client/internal/dns/test"
@@ -1042,3 +1046,163 @@ func TestHandlerChain_AddRemoveRoundtrip(t *testing.T) {
 		})
 	}
 }
+
+// answeringHandler writes a fixed A record to ack the query. Used to verify
+// which handler ResolveInternal dispatches to.
+type answeringHandler struct {
+	name string
+	ip   string
+}
+
+func (h *answeringHandler) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
+	resp := &dns.Msg{}
+	resp.SetReply(r)
+	resp.Answer = []dns.RR{&dns.A{
+		Hdr: dns.RR_Header{Name: r.Question[0].Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+		A:   net.ParseIP(h.ip).To4(),
+	}}
+	_ = w.WriteMsg(resp)
+}
+
+func (h *answeringHandler) String() string { return h.name }
+
+func TestHandlerChain_ResolveInternal_SkipsAboveMaxPriority(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+
+	high := &answeringHandler{name: "high", ip: "10.0.0.1"}
+	low := &answeringHandler{name: "low", ip: "10.0.0.2"}
+
+	chain.AddHandler("example.com.", high, nbdns.PriorityMgmtCache)
+	chain.AddHandler("example.com.", low, nbdns.PriorityUpstream)
+
+	r := new(dns.Msg)
+	r.SetQuestion("example.com.", dns.TypeA)
+
+	resp, err := chain.ResolveInternal(context.Background(), r, nbdns.PriorityUpstream)
+	assert.NoError(t, err)
+	assert.NotNil(t, resp)
+	assert.Equal(t, 1, len(resp.Answer))
+	a, ok := resp.Answer[0].(*dns.A)
+	assert.True(t, ok)
+	assert.Equal(t, "10.0.0.2", a.A.String(), "should skip mgmtCache handler and resolve via upstream")
+}
+
+func TestHandlerChain_ResolveInternal_ErrorWhenNoMatch(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+	high := &answeringHandler{name: "high", ip: "10.0.0.1"}
+	chain.AddHandler("example.com.", high, nbdns.PriorityMgmtCache)
+
+	r := new(dns.Msg)
+	r.SetQuestion("example.com.", dns.TypeA)
+
+	_, err := chain.ResolveInternal(context.Background(), r, nbdns.PriorityUpstream)
+	assert.Error(t, err, "no handler at or below maxPriority should error")
+}
+
+// rawWriteHandler packs a response and calls ResponseWriter.Write directly
+// (instead of WriteMsg), exercising the internalResponseWriter.Write path.
+type rawWriteHandler struct {
+	ip string
+}
+
+func (h *rawWriteHandler) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
+	resp := &dns.Msg{}
+	resp.SetReply(r)
+	resp.Answer = []dns.RR{&dns.A{
+		Hdr: dns.RR_Header{Name: r.Question[0].Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+		A:   net.ParseIP(h.ip).To4(),
+	}}
+	packed, err := resp.Pack()
+	if err != nil {
+		return
+	}
+	_, _ = w.Write(packed)
+}
+
+func TestHandlerChain_ResolveInternal_CapturesRawWrite(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+	chain.AddHandler("example.com.", &rawWriteHandler{ip: "10.0.0.3"}, nbdns.PriorityUpstream)
+
+	r := new(dns.Msg)
+	r.SetQuestion("example.com.", dns.TypeA)
+
+	resp, err := chain.ResolveInternal(context.Background(), r, nbdns.PriorityUpstream)
+	assert.NoError(t, err)
+	require.NotNil(t, resp)
+	require.Len(t, resp.Answer, 1)
+	a, ok := resp.Answer[0].(*dns.A)
+	require.True(t, ok)
+	assert.Equal(t, "10.0.0.3", a.A.String(), "handlers calling Write(packed) must still surface their answer")
+}
+
+func TestHandlerChain_ResolveInternal_EmptyQuestion(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+	_, err := chain.ResolveInternal(context.Background(), new(dns.Msg), nbdns.PriorityUpstream)
+	assert.Error(t, err)
+}
+
+// hangingHandler blocks indefinitely until closed, simulating a wedged upstream.
+type hangingHandler struct {
+	block chan struct{}
+}
+
+func (h *hangingHandler) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
+	<-h.block
+	resp := &dns.Msg{}
+	resp.SetReply(r)
+	_ = w.WriteMsg(resp)
+}
+
+func (h *hangingHandler) String() string { return "hangingHandler" }
+
+func TestHandlerChain_ResolveInternal_HonorsContextTimeout(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+	h := &hangingHandler{block: make(chan struct{})}
+	defer close(h.block)
+
+	chain.AddHandler("example.com.", h, nbdns.PriorityUpstream)
+
+	r := new(dns.Msg)
+	r.SetQuestion("example.com.", dns.TypeA)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 100*time.Millisecond)
+	defer cancel()
+
+	start := time.Now()
+	_, err := chain.ResolveInternal(ctx, r, nbdns.PriorityUpstream)
+	elapsed := time.Since(start)
+
+	assert.Error(t, err)
+	assert.ErrorIs(t, err, context.DeadlineExceeded)
+	assert.Less(t, elapsed, 500*time.Millisecond, "ResolveInternal must return shortly after ctx deadline")
+}
+
+func TestHandlerChain_HasRootHandlerAtOrBelow(t *testing.T) {
+	chain := nbdns.NewHandlerChain()
+	h := &answeringHandler{name: "h", ip: "10.0.0.1"}
+
+	assert.False(t, chain.HasRootHandlerAtOrBelow(nbdns.PriorityUpstream), "empty chain")
+
+	chain.AddHandler("example.com.", h, nbdns.PriorityUpstream)
+	assert.False(t, chain.HasRootHandlerAtOrBelow(nbdns.PriorityUpstream), "non-root handler does not count")
+
+	chain.AddHandler(".", h, nbdns.PriorityMgmtCache)
+	assert.False(t, chain.HasRootHandlerAtOrBelow(nbdns.PriorityUpstream), "root handler above threshold excluded")
+
+	chain.AddHandler(".", h, nbdns.PriorityDefault)
+	assert.True(t, chain.HasRootHandlerAtOrBelow(nbdns.PriorityUpstream), "root handler at PriorityDefault included")
+
+	chain.RemoveHandler(".", nbdns.PriorityDefault)
+	assert.False(t, chain.HasRootHandlerAtOrBelow(nbdns.PriorityUpstream))
+
+	// Primary nsgroup case: root handler lands at PriorityUpstream.
+	chain.AddHandler(".", h, nbdns.PriorityUpstream)
+	assert.True(t, chain.HasRootHandlerAtOrBelow(nbdns.PriorityUpstream), "root at PriorityUpstream included")
+	chain.RemoveHandler(".", nbdns.PriorityUpstream)
+
+	// Fallback case: original /etc/resolv.conf entries land at PriorityFallback.
+	chain.AddHandler(".", h, nbdns.PriorityFallback)
+	assert.True(t, chain.HasRootHandlerAtOrBelow(nbdns.PriorityUpstream), "root at PriorityFallback included")
+	chain.RemoveHandler(".", nbdns.PriorityFallback)
+	assert.False(t, chain.HasRootHandlerAtOrBelow(nbdns.PriorityUpstream))
+}

client/internal/dns/mgmt/mgmt.go

@@ -2,40 +2,79 @@ package mgmt
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"net"
-	"net/netip"
 	"net/url"
+	"os"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/miekg/dns"
 	log "github.com/sirupsen/logrus"
+	"golang.org/x/sync/singleflight"
 
 	dnsconfig "github.com/netbirdio/netbird/client/internal/dns/config"
+	"github.com/netbirdio/netbird/client/internal/dns/resutil"
 	"github.com/netbirdio/netbird/shared/management/domain"
 )
 
-const dnsTimeout = 5 * time.Second
+const (
+	dnsTimeout     = 5 * time.Second
+	defaultTTL     = 300 * time.Second
+	refreshBackoff = 30 * time.Second
 
-// Resolver caches critical NetBird infrastructure domains
+	// envMgmtCacheTTL overrides defaultTTL for integration/dev testing.
+	envMgmtCacheTTL = "NB_MGMT_CACHE_TTL"
+)
+
+// ChainResolver lets the cache refresh stale entries through the DNS handler
+// chain instead of net.DefaultResolver, avoiding loopback when NetBird is the
+// system resolver.
+type ChainResolver interface {
+	ResolveInternal(ctx context.Context, msg *dns.Msg, maxPriority int) (*dns.Msg, error)
+	HasRootHandlerAtOrBelow(maxPriority int) bool
+}
+
+// cachedRecord holds DNS records plus timestamps used for TTL refresh.
+// records and cachedAt are set at construction and treated as immutable;
+// lastFailedRefresh and consecFailures are mutable and must be accessed under
+// Resolver.mutex.
+type cachedRecord struct {
+	records           []dns.RR
+	cachedAt          time.Time
+	lastFailedRefresh time.Time
+	consecFailures    int
+}
+
+// Resolver caches critical NetBird infrastructure domains.
+// records, refreshing, mgmtDomain and serverDomains are all guarded by mutex.
 type Resolver struct {
-	records       map[dns.Question][]dns.RR
+	records       map[dns.Question]*cachedRecord
 	mgmtDomain    *domain.Domain
 	serverDomains *dnsconfig.ServerDomains
 	mutex         sync.RWMutex
-}
 
-type ipsResponse struct {
-	ips []netip.Addr
-	err error
+	chain            ChainResolver
+	chainMaxPriority int
+	refreshGroup     singleflight.Group
+
+	// refreshing tracks questions whose refresh is running via the OS
+	// fallback path. A ServeDNS hit for a question in this map indicates
+	// the OS resolver routed the recursive query back to us (loop). Only
+	// the OS path arms this so chain-path refreshes don't produce false
+	// positives. The atomic bool is CAS-flipped once per refresh to
+	// throttle the warning log.
+	refreshing map[dns.Question]*atomic.Bool
 }
 
 // NewResolver creates a new management domains cache resolver.
 func NewResolver() *Resolver {
 	return &Resolver{
-		records: make(map[dns.Question][]dns.RR),
+		records:    make(map[dns.Question]*cachedRecord),
+		refreshing: make(map[dns.Question]*atomic.Bool),
 	}
 }
 
@@ -44,7 +83,19 @@ func (m *Resolver) String() string {
 	return "MgmtCacheResolver"
 }
 
-// ServeDNS implements dns.Handler interface.
+// SetChainResolver wires the handler chain used to refresh stale cache entries.
+// maxPriority caps which handlers may answer refresh queries (typically
+// PriorityUpstream, so upstream/default/fallback handlers are consulted and
+// mgmt/route/local handlers are skipped).
+func (m *Resolver) SetChainResolver(chain ChainResolver, maxPriority int) {
+	m.mutex.Lock()
+	m.chain = chain
+	m.chainMaxPriority = maxPriority
+	m.mutex.Unlock()
+}
+
+// ServeDNS serves cached A/AAAA records. Stale entries are returned
+// immediately and refreshed asynchronously (stale-while-revalidate).
 func (m *Resolver) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 	if len(r.Question) == 0 {
 		m.continueToNext(w, r)
@@ -60,7 +111,14 @@ func (m *Resolver) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 	}
 
 	m.mutex.RLock()
-	records, found := m.records[question]
+	cached, found := m.records[question]
+	inflight := m.refreshing[question]
+	var shouldRefresh bool
+	if found {
+		stale := time.Since(cached.cachedAt) > cacheTTL()
+		inBackoff := !cached.lastFailedRefresh.IsZero() && time.Since(cached.lastFailedRefresh) < refreshBackoff
+		shouldRefresh = stale && !inBackoff
+	}
 	m.mutex.RUnlock()
 
 	if !found {
@@ -68,12 +126,23 @@ func (m *Resolver) ServeDNS(w dns.ResponseWriter, r *dns.Msg) {
 		return
 	}
 
+	if inflight != nil && inflight.CompareAndSwap(false, true) {
+		log.Warnf("mgmt cache: possible resolver loop for domain=%s: served stale while an OS-fallback refresh was inflight (if NetBird is the system resolver, the OS-path predicate is wrong)",
+			question.Name)
+	}
+
+	// Skip scheduling a refresh goroutine if one is already inflight for
+	// this question; singleflight would dedup anyway but skipping avoids
+	// a parked goroutine per stale hit under bursty load.
+	if shouldRefresh && inflight == nil {
+		m.scheduleRefresh(question)
+	}
+
 	resp := &dns.Msg{}
 	resp.SetReply(r)
 	resp.Authoritative = false
 	resp.RecursionAvailable = true
-
-	resp.Answer = append(resp.Answer, records...)
+	resp.Answer = append(resp.Answer, cached.records...)
 
 	log.Debugf("serving %d cached records for domain=%s", len(resp.Answer), question.Name)
 
@@ -98,101 +167,177 @@ func (m *Resolver) continueToNext(w dns.ResponseWriter, r *dns.Msg) {
 	}
 }
 
-// AddDomain manually adds a domain to cache by resolving it.
+// AddDomain resolves a domain and stores its A/AAAA records in the cache.
+// A family that resolves NODATA (nil err, zero records) evicts any stale
+// entry for that qtype.
 func (m *Resolver) AddDomain(ctx context.Context, d domain.Domain) error {
 	dnsName := strings.ToLower(dns.Fqdn(d.PunycodeString()))
 
 	ctx, cancel := context.WithTimeout(ctx, dnsTimeout)
 	defer cancel()
 
-	ips, err := lookupIPWithExtraTimeout(ctx, d)
-	if err != nil {
-		return err
+	aRecords, aaaaRecords, errA, errAAAA := m.lookupBoth(ctx, d, dnsName)
+
+	if errA != nil && errAAAA != nil {
+		return fmt.Errorf("resolve %s: %w", d.SafeString(), errors.Join(errA, errAAAA))
 	}
 
-	var aRecords, aaaaRecords []dns.RR
-	for _, ip := range ips {
-		if ip.Is4() {
-			rr := &dns.A{
-				Hdr: dns.RR_Header{
-					Name:   dnsName,
-					Rrtype: dns.TypeA,
-					Class:  dns.ClassINET,
-					Ttl:    300,
-				},
-				A: ip.AsSlice(),
-			}
-			aRecords = append(aRecords, rr)
-		} else if ip.Is6() {
-			rr := &dns.AAAA{
-				Hdr: dns.RR_Header{
-					Name:   dnsName,
-					Rrtype: dns.TypeAAAA,
-					Class:  dns.ClassINET,
-					Ttl:    300,
-				},
-				AAAA: ip.AsSlice(),
-			}
-			aaaaRecords = append(aaaaRecords, rr)
-		}
+	// Dual NODATA: don't wipe existing entries, let the caller retry.
+	if errA == nil && errAAAA == nil && len(aRecords) == 0 && len(aaaaRecords) == 0 {
+		return fmt.Errorf("resolve %s: no A/AAAA records", d.SafeString())
 	}
 
+	now := time.Now()
 	m.mutex.Lock()
+	defer m.mutex.Unlock()
 
-	if len(aRecords) > 0 {
-		aQuestion := dns.Question{
-			Name:   dnsName,
-			Qtype:  dns.TypeA,
-			Qclass: dns.ClassINET,
-		}
-		m.records[aQuestion] = aRecords
-	}
+	m.applyFamilyRecords(dnsName, dns.TypeA, aRecords, errA, now)
+	m.applyFamilyRecords(dnsName, dns.TypeAAAA, aaaaRecords, errAAAA, now)
 
-	if len(aaaaRecords) > 0 {
-		aaaaQuestion := dns.Question{
-			Name:   dnsName,
-			Qtype:  dns.TypeAAAA,
-			Qclass: dns.ClassINET,
-		}
-		m.records[aaaaQuestion] = aaaaRecords
-	}
-
-	m.mutex.Unlock()
-
-	log.Debugf("added domain=%s with %d A records and %d AAAA records",
+	log.Debugf("added/updated domain=%s with %d A records and %d AAAA records",
 		d.SafeString(), len(aRecords), len(aaaaRecords))
 
 	return nil
 }
 
-func lookupIPWithExtraTimeout(ctx context.Context, d domain.Domain) ([]netip.Addr, error) {
-	log.Infof("looking up IP for mgmt domain=%s", d.SafeString())
-	defer log.Infof("done looking up IP for mgmt domain=%s", d.SafeString())
-	resultChan := make(chan *ipsResponse, 1)
+// applyFamilyRecords writes records, evicts on NODATA, leaves the cache
+// untouched on error. Caller holds m.mutex.
+func (m *Resolver) applyFamilyRecords(dnsName string, qtype uint16, records []dns.RR, err error, now time.Time) {
+	q := dns.Question{Name: dnsName, Qtype: qtype, Qclass: dns.ClassINET}
+	switch {
+	case len(records) > 0:
+		m.records[q] = &cachedRecord{records: records, cachedAt: now}
+	case err == nil:
+		delete(m.records, q)
+	}
+}
 
-	go func() {
-		ips, err := net.DefaultResolver.LookupNetIP(ctx, "ip", d.PunycodeString())
-		resultChan <- &ipsResponse{
-			err: err,
-			ips: ips,
+// scheduleRefresh kicks off an async refresh. DoChan spawns one goroutine per
+// unique in-flight key; bursty stale hits share its channel.
+func (m *Resolver) scheduleRefresh(question dns.Question) {
+	key := question.Name + "|" + dns.TypeToString[question.Qtype]
+	_ = m.refreshGroup.DoChan(key, func() (any, error) {
+		return nil, m.refreshQuestion(question)
+	})
+}
+
+// refreshQuestion replaces the cached records on success, or marks the entry
+// failed (arming the backoff) on failure. While this runs, ServeDNS can detect
+// a resolver loop by spotting a query for this same question arriving on us.
+func (m *Resolver) refreshQuestion(question dns.Question) error {
+	ctx, cancel := context.WithTimeout(context.Background(), dnsTimeout)
+	defer cancel()
+
+	d, err := domain.FromString(strings.TrimSuffix(question.Name, "."))
+	if err != nil {
+		m.markRefreshFailed(question)
+		return fmt.Errorf("parse domain: %w", err)
+	}
+
+	records, err := m.lookupRecords(ctx, d, question)
+	if err != nil {
+		fails := m.markRefreshFailed(question)
+		logf := log.Warnf
+		if fails > 1 {
+			logf = log.Debugf
 		}
-	}()
-
-	var resp *ipsResponse
-
-	select {
-	case <-time.After(dnsTimeout + time.Millisecond*500):
-		log.Warnf("timed out waiting for IP for mgmt domain=%s", d.SafeString())
-		return nil, fmt.Errorf("timed out waiting for ips to be available for domain %s", d.SafeString())
-	case <-ctx.Done():
-		return nil, ctx.Err()
-	case resp = <-resultChan:
+		logf("refresh mgmt cache domain=%s type=%s: %v (consecutive failures=%d)",
+			d.SafeString(), dns.TypeToString[question.Qtype], err, fails)
+		return err
 	}
 
-	if resp.err != nil {
-		return nil, fmt.Errorf("resolve domain %s: %w", d.SafeString(), resp.err)
+	// NOERROR/NODATA: family gone upstream, evict so we stop serving stale.
+	if len(records) == 0 {
+		m.mutex.Lock()
+		delete(m.records, question)
+		m.mutex.Unlock()
+		log.Infof("removed mgmt cache domain=%s type=%s: no records returned",
+			d.SafeString(), dns.TypeToString[question.Qtype])
+		return nil
 	}
-	return resp.ips, nil
+
+	now := time.Now()
+	m.mutex.Lock()
+	if _, stillCached := m.records[question]; !stillCached {
+		m.mutex.Unlock()
+		log.Debugf("skipping refresh write for domain=%s type=%s: entry was removed during refresh",
+			d.SafeString(), dns.TypeToString[question.Qtype])
+		return nil
+	}
+	m.records[question] = &cachedRecord{records: records, cachedAt: now}
+	m.mutex.Unlock()
+
+	log.Infof("refreshed mgmt cache domain=%s type=%s",
+		d.SafeString(), dns.TypeToString[question.Qtype])
+	return nil
+}
+
+func (m *Resolver) markRefreshing(question dns.Question) {
+	m.mutex.Lock()
+	m.refreshing[question] = &atomic.Bool{}
+	m.mutex.Unlock()
+}
+
+func (m *Resolver) clearRefreshing(question dns.Question) {
+	m.mutex.Lock()
+	delete(m.refreshing, question)
+	m.mutex.Unlock()
+}
+
+// markRefreshFailed arms the backoff and returns the new consecutive-failure
+// count so callers can downgrade subsequent failure logs to debug.
+func (m *Resolver) markRefreshFailed(question dns.Question) int {
+	m.mutex.Lock()
+	defer m.mutex.Unlock()
+	c, ok := m.records[question]
+	if !ok {
+		return 0
+	}
+	c.lastFailedRefresh = time.Now()
+	c.consecFailures++
+	return c.consecFailures
+}
+
+// lookupBoth resolves A and AAAA via chain or OS. Per-family errors let
+// callers tell records, NODATA (nil err, no records), and failure apart.
+func (m *Resolver) lookupBoth(ctx context.Context, d domain.Domain, dnsName string) (aRecords, aaaaRecords []dns.RR, errA, errAAAA error) {
+	m.mutex.RLock()
+	chain := m.chain
+	maxPriority := m.chainMaxPriority
+	m.mutex.RUnlock()
+
+	if chain != nil && chain.HasRootHandlerAtOrBelow(maxPriority) {
+		aRecords, errA = lookupViaChain(ctx, chain, maxPriority, dnsName, dns.TypeA)
+		aaaaRecords, errAAAA = lookupViaChain(ctx, chain, maxPriority, dnsName, dns.TypeAAAA)
+		return
+	}
+
+	// TODO: drop once every supported OS registers a fallback resolver. Safe
+	// today: no root handler at priority ≤ PriorityUpstream means NetBird is
+	// not the system resolver, so net.DefaultResolver will not loop back.
+	aRecords, errA = osLookup(ctx, d, dnsName, dns.TypeA)
+	aaaaRecords, errAAAA = osLookup(ctx, d, dnsName, dns.TypeAAAA)
+	return
+}
+
+// lookupRecords resolves a single record type via chain or OS. The OS branch
+// arms the loop detector for the duration of its call so that ServeDNS can
+// spot the OS resolver routing the recursive query back to us.
+func (m *Resolver) lookupRecords(ctx context.Context, d domain.Domain, q dns.Question) ([]dns.RR, error) {
+	m.mutex.RLock()
+	chain := m.chain
+	maxPriority := m.chainMaxPriority
+	m.mutex.RUnlock()
+
+	if chain != nil && chain.HasRootHandlerAtOrBelow(maxPriority) {
+		return lookupViaChain(ctx, chain, maxPriority, q.Name, q.Qtype)
+	}
+
+	// TODO: drop once every supported OS registers a fallback resolver.
+	m.markRefreshing(q)
+	defer m.clearRefreshing(q)
+
+	return osLookup(ctx, d, q.Name, q.Qtype)
 }
 
 // PopulateFromConfig extracts and caches domains from the client configuration.
@@ -224,19 +369,8 @@ func (m *Resolver) RemoveDomain(d domain.Domain) error {
 	m.mutex.Lock()
 	defer m.mutex.Unlock()
 
-	aQuestion := dns.Question{
-		Name:   dnsName,
-		Qtype:  dns.TypeA,
-		Qclass: dns.ClassINET,
-	}
-	delete(m.records, aQuestion)
-
-	aaaaQuestion := dns.Question{
-		Name:   dnsName,
-		Qtype:  dns.TypeAAAA,
-		Qclass: dns.ClassINET,
-	}
-	delete(m.records, aaaaQuestion)
+	delete(m.records, dns.Question{Name: dnsName, Qtype: dns.TypeA, Qclass: dns.ClassINET})
+	delete(m.records, dns.Question{Name: dnsName, Qtype: dns.TypeAAAA, Qclass: dns.ClassINET})
 
 	log.Debugf("removed domain=%s from cache", d.SafeString())
 	return nil
@@ -394,3 +528,82 @@ func (m *Resolver) extractDomainsFromServerDomains(serverDomains dnsconfig.Serve
 
 	return domains
 }
+
+func cacheTTL() time.Duration {
+	if v := os.Getenv(envMgmtCacheTTL); v != "" {
+		if d, err := time.ParseDuration(v); err == nil && d > 0 {
+			return d
+		}
+	}
+	return defaultTTL
+}
+
+// lookupViaChain resolves via the handler chain and rewrites each RR to use
+// dnsName as owner and cacheTTL() as TTL, so CNAME-backed domains don't cache
+// target-owned records or upstream TTLs. NODATA returns (nil, nil).
+func lookupViaChain(ctx context.Context, chain ChainResolver, maxPriority int, dnsName string, qtype uint16) ([]dns.RR, error) {
+	msg := &dns.Msg{}
+	msg.SetQuestion(dnsName, qtype)
+	msg.RecursionDesired = true
+
+	resp, err := chain.ResolveInternal(ctx, msg, maxPriority)
+	if err != nil {
+		return nil, fmt.Errorf("chain resolve: %w", err)
+	}
+	if resp == nil {
+		return nil, fmt.Errorf("chain resolve returned nil response")
+	}
+	if resp.Rcode != dns.RcodeSuccess {
+		return nil, fmt.Errorf("chain resolve rcode=%s", dns.RcodeToString[resp.Rcode])
+	}
+
+	ttl := uint32(cacheTTL().Seconds())
+	var filtered []dns.RR
+	for _, rr := range resp.Answer {
+		if rr.Header().Class != dns.ClassINET {
+			continue
+		}
+		switch r := rr.(type) {
+		case *dns.A:
+			if qtype != dns.TypeA {
+				continue
+			}
+			filtered = append(filtered, &dns.A{
+				Hdr: dns.RR_Header{Name: dnsName, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: ttl},
+				A:   append(net.IP(nil), r.A...),
+			})
+		case *dns.AAAA:
+			if qtype != dns.TypeAAAA {
+				continue
+			}
+			filtered = append(filtered, &dns.AAAA{
+				Hdr:  dns.RR_Header{Name: dnsName, Rrtype: dns.TypeAAAA, Class: dns.ClassINET, Ttl: ttl},
+				AAAA: append(net.IP(nil), r.AAAA...),
+			})
+		}
+	}
+	return filtered, nil
+}
+
+// osLookup resolves a single family via net.DefaultResolver using resutil,
+// which disambiguates NODATA from NXDOMAIN and Unmaps v4-mapped-v6. NODATA
+// returns (nil, nil).
+func osLookup(ctx context.Context, d domain.Domain, dnsName string, qtype uint16) ([]dns.RR, error) {
+	network := resutil.NetworkForQtype(qtype)
+	if network == "" {
+		return nil, fmt.Errorf("unsupported qtype %s", dns.TypeToString[qtype])
+	}
+
+	log.Infof("looking up IP for mgmt domain=%s type=%s", d.SafeString(), dns.TypeToString[qtype])
+	defer log.Infof("done looking up IP for mgmt domain=%s type=%s", d.SafeString(), dns.TypeToString[qtype])
+
+	result := resutil.LookupIP(ctx, net.DefaultResolver, network, d.PunycodeString(), qtype)
+	if result.Rcode == dns.RcodeSuccess {
+		return resutil.IPsToRRs(dnsName, result.IPs, uint32(cacheTTL().Seconds())), nil
+	}
+
+	if result.Err != nil {
+		return nil, fmt.Errorf("resolve %s type=%s: %w", d.SafeString(), dns.TypeToString[qtype], result.Err)
+	}
+	return nil, fmt.Errorf("resolve %s type=%s: rcode=%s", d.SafeString(), dns.TypeToString[qtype], dns.RcodeToString[result.Rcode])
+}

client/internal/dns/mgmt/mgmt_refresh_test.go

@@ -0,0 +1,359 @@
+package mgmt
+
+import (
+	"context"
+	"errors"
+	"net"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/miekg/dns"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/dns/test"
+	"github.com/netbirdio/netbird/shared/management/domain"
+)
+
+type fakeChain struct {
+	mu       sync.Mutex
+	calls    map[string]int
+	answers  map[string][]dns.RR
+	err      error
+	hasRoot  bool
+	onLookup func()
+}
+
+func newFakeChain() *fakeChain {
+	return &fakeChain{
+		calls:   map[string]int{},
+		answers: map[string][]dns.RR{},
+		hasRoot: true,
+	}
+}
+
+func (f *fakeChain) HasRootHandlerAtOrBelow(maxPriority int) bool {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.hasRoot
+}
+
+func (f *fakeChain) ResolveInternal(ctx context.Context, msg *dns.Msg, maxPriority int) (*dns.Msg, error) {
+	f.mu.Lock()
+	q := msg.Question[0]
+	key := q.Name + "|" + dns.TypeToString[q.Qtype]
+	f.calls[key]++
+	answers := f.answers[key]
+	err := f.err
+	onLookup := f.onLookup
+	f.mu.Unlock()
+
+	if onLookup != nil {
+		onLookup()
+	}
+	if err != nil {
+		return nil, err
+	}
+	resp := &dns.Msg{}
+	resp.SetReply(msg)
+	resp.Answer = answers
+	return resp, nil
+}
+
+func (f *fakeChain) setAnswer(name string, qtype uint16, ip string) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	key := name + "|" + dns.TypeToString[qtype]
+	hdr := dns.RR_Header{Name: name, Rrtype: qtype, Class: dns.ClassINET, Ttl: 60}
+	switch qtype {
+	case dns.TypeA:
+		f.answers[key] = []dns.RR{&dns.A{Hdr: hdr, A: net.ParseIP(ip).To4()}}
+	case dns.TypeAAAA:
+		f.answers[key] = []dns.RR{&dns.AAAA{Hdr: hdr, AAAA: net.ParseIP(ip).To16()}}
+	}
+}
+
+func (f *fakeChain) callCount(name string, qtype uint16) int {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.calls[name+"|"+dns.TypeToString[qtype]]
+}
+
+// waitFor polls the predicate until it returns true or the deadline passes.
+func waitFor(t *testing.T, d time.Duration, fn func() bool) {
+	t.Helper()
+	deadline := time.Now().Add(d)
+	for time.Now().Before(deadline) {
+		if fn() {
+			return
+		}
+		time.Sleep(5 * time.Millisecond)
+	}
+	t.Fatalf("condition not met within %s", d)
+}
+
+func queryA(t *testing.T, r *Resolver, name string) *dns.Msg {
+	t.Helper()
+	msg := new(dns.Msg)
+	msg.SetQuestion(name, dns.TypeA)
+	w := &test.MockResponseWriter{}
+	r.ServeDNS(w, msg)
+	return w.GetLastResponse()
+}
+
+func firstA(t *testing.T, resp *dns.Msg) string {
+	t.Helper()
+	require.NotNil(t, resp)
+	require.Greater(t, len(resp.Answer), 0, "expected at least one answer")
+	a, ok := resp.Answer[0].(*dns.A)
+	require.True(t, ok, "expected A record")
+	return a.A.String()
+}
+
+func TestResolver_ServeFresh_NoRefresh(t *testing.T) {
+	r := NewResolver()
+	chain := newFakeChain()
+	chain.setAnswer("mgmt.example.com.", dns.TypeA, "10.0.0.2")
+	r.SetChainResolver(chain, 50)
+
+	r.records[dns.Question{Name: "mgmt.example.com.", Qtype: dns.TypeA, Qclass: dns.ClassINET}] = &cachedRecord{
+		records: []dns.RR{&dns.A{
+			Hdr: dns.RR_Header{Name: "mgmt.example.com.", Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+			A:   net.ParseIP("10.0.0.1").To4(),
+		}},
+		cachedAt: time.Now(), // fresh
+	}
+
+	resp := queryA(t, r, "mgmt.example.com.")
+	assert.Equal(t, "10.0.0.1", firstA(t, resp))
+
+	time.Sleep(20 * time.Millisecond)
+	assert.Equal(t, 0, chain.callCount("mgmt.example.com.", dns.TypeA), "fresh entry must not trigger refresh")
+}
+
+func TestResolver_StaleTriggersAsyncRefresh(t *testing.T) {
+	r := NewResolver()
+	chain := newFakeChain()
+	chain.setAnswer("mgmt.example.com.", dns.TypeA, "10.0.0.2")
+	r.SetChainResolver(chain, 50)
+
+	q := dns.Question{Name: "mgmt.example.com.", Qtype: dns.TypeA, Qclass: dns.ClassINET}
+	r.records[q] = &cachedRecord{
+		records: []dns.RR{&dns.A{
+			Hdr: dns.RR_Header{Name: q.Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+			A:   net.ParseIP("10.0.0.1").To4(),
+		}},
+		cachedAt: time.Now().Add(-2 * defaultTTL), // stale
+	}
+
+	// First query: serves stale immediately.
+	resp := queryA(t, r, "mgmt.example.com.")
+	assert.Equal(t, "10.0.0.1", firstA(t, resp), "stale entry must be served while refresh runs")
+
+	waitFor(t, time.Second, func() bool {
+		return chain.callCount("mgmt.example.com.", dns.TypeA) >= 1
+	})
+
+	// Next query should now return the refreshed IP.
+	waitFor(t, time.Second, func() bool {
+		resp := queryA(t, r, "mgmt.example.com.")
+		return resp != nil && len(resp.Answer) > 0 && firstA(t, resp) == "10.0.0.2"
+	})
+}
+
+func TestResolver_ConcurrentStaleHitsCollapseRefresh(t *testing.T) {
+	r := NewResolver()
+	chain := newFakeChain()
+	chain.setAnswer("mgmt.example.com.", dns.TypeA, "10.0.0.2")
+
+	var inflight atomic.Int32
+	var maxInflight atomic.Int32
+	chain.onLookup = func() {
+		cur := inflight.Add(1)
+		defer inflight.Add(-1)
+		for {
+			prev := maxInflight.Load()
+			if cur <= prev || maxInflight.CompareAndSwap(prev, cur) {
+				break
+			}
+		}
+		time.Sleep(50 * time.Millisecond) // hold inflight long enough to collide
+	}
+
+	r.SetChainResolver(chain, 50)
+
+	q := dns.Question{Name: "mgmt.example.com.", Qtype: dns.TypeA, Qclass: dns.ClassINET}
+	r.records[q] = &cachedRecord{
+		records: []dns.RR{&dns.A{
+			Hdr: dns.RR_Header{Name: q.Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+			A:   net.ParseIP("10.0.0.1").To4(),
+		}},
+		cachedAt: time.Now().Add(-2 * defaultTTL),
+	}
+
+	var wg sync.WaitGroup
+	for i := 0; i < 50; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			queryA(t, r, "mgmt.example.com.")
+		}()
+	}
+	wg.Wait()
+
+	waitFor(t, 2*time.Second, func() bool {
+		return inflight.Load() == 0
+	})
+
+	calls := chain.callCount("mgmt.example.com.", dns.TypeA)
+	assert.LessOrEqual(t, calls, 2, "singleflight must collapse concurrent refreshes (got %d)", calls)
+	assert.Equal(t, int32(1), maxInflight.Load(), "only one refresh should run concurrently")
+}
+
+func TestResolver_RefreshFailureArmsBackoff(t *testing.T) {
+	r := NewResolver()
+	chain := newFakeChain()
+	chain.err = errors.New("boom")
+	r.SetChainResolver(chain, 50)
+
+	q := dns.Question{Name: "mgmt.example.com.", Qtype: dns.TypeA, Qclass: dns.ClassINET}
+	r.records[q] = &cachedRecord{
+		records: []dns.RR{&dns.A{
+			Hdr: dns.RR_Header{Name: q.Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+			A:   net.ParseIP("10.0.0.1").To4(),
+		}},
+		cachedAt: time.Now().Add(-2 * defaultTTL),
+	}
+
+	// First stale hit triggers a refresh attempt that fails.
+	resp := queryA(t, r, "mgmt.example.com.")
+	assert.Equal(t, "10.0.0.1", firstA(t, resp), "stale entry served while refresh fails")
+
+	waitFor(t, time.Second, func() bool {
+		return chain.callCount("mgmt.example.com.", dns.TypeA) == 1
+	})
+	waitFor(t, time.Second, func() bool {
+		r.mutex.RLock()
+		defer r.mutex.RUnlock()
+		c, ok := r.records[q]
+		return ok && !c.lastFailedRefresh.IsZero()
+	})
+
+	// Subsequent stale hits within backoff window should not schedule more refreshes.
+	for i := 0; i < 10; i++ {
+		queryA(t, r, "mgmt.example.com.")
+	}
+	time.Sleep(50 * time.Millisecond)
+	assert.Equal(t, 1, chain.callCount("mgmt.example.com.", dns.TypeA), "backoff must suppress further refreshes")
+}
+
+func TestResolver_NoRootHandler_SkipsChain(t *testing.T) {
+	r := NewResolver()
+	chain := newFakeChain()
+	chain.hasRoot = false
+	chain.setAnswer("mgmt.example.com.", dns.TypeA, "10.0.0.2")
+	r.SetChainResolver(chain, 50)
+
+	// With hasRoot=false the chain must not be consulted. Use a short
+	// deadline so the OS fallback returns quickly without waiting on a
+	// real network call in CI.
+	ctx, cancel := context.WithTimeout(context.Background(), 50*time.Millisecond)
+	defer cancel()
+	_, _, _, _ = r.lookupBoth(ctx, domain.Domain("mgmt.example.com"), "mgmt.example.com.")
+
+	assert.Equal(t, 0, chain.callCount("mgmt.example.com.", dns.TypeA),
+		"chain must not be used when no root handler is registered at the bound priority")
+}
+
+func TestResolver_ServeDuringRefreshSetsLoopFlag(t *testing.T) {
+	// ServeDNS being invoked for a question while a refresh for that question
+	// is inflight indicates a resolver loop (OS resolver sent the recursive
+	// query back to us). The inflightRefresh.loopLoggedOnce flag must be set.
+	r := NewResolver()
+
+	q := dns.Question{Name: "mgmt.example.com.", Qtype: dns.TypeA, Qclass: dns.ClassINET}
+	r.records[q] = &cachedRecord{
+		records: []dns.RR{&dns.A{
+			Hdr: dns.RR_Header{Name: q.Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+			A:   net.ParseIP("10.0.0.1").To4(),
+		}},
+		cachedAt: time.Now(),
+	}
+
+	// Simulate an inflight refresh.
+	r.markRefreshing(q)
+	defer r.clearRefreshing(q)
+
+	resp := queryA(t, r, "mgmt.example.com.")
+	assert.Equal(t, "10.0.0.1", firstA(t, resp), "stale entry must still be served to avoid breaking external queries")
+
+	r.mutex.RLock()
+	inflight := r.refreshing[q]
+	r.mutex.RUnlock()
+	require.NotNil(t, inflight)
+	assert.True(t, inflight.Load(), "loop flag must be set once a ServeDNS during refresh was observed")
+}
+
+func TestResolver_LoopFlagOnlyTrippedOncePerRefresh(t *testing.T) {
+	r := NewResolver()
+
+	q := dns.Question{Name: "mgmt.example.com.", Qtype: dns.TypeA, Qclass: dns.ClassINET}
+	r.records[q] = &cachedRecord{
+		records: []dns.RR{&dns.A{
+			Hdr: dns.RR_Header{Name: q.Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+			A:   net.ParseIP("10.0.0.1").To4(),
+		}},
+		cachedAt: time.Now(),
+	}
+
+	r.markRefreshing(q)
+	defer r.clearRefreshing(q)
+
+	// Multiple ServeDNS calls during the same refresh must not re-set the flag
+	// (CompareAndSwap from false -> true returns true only on the first call).
+	for range 5 {
+		queryA(t, r, "mgmt.example.com.")
+	}
+
+	r.mutex.RLock()
+	inflight := r.refreshing[q]
+	r.mutex.RUnlock()
+	assert.True(t, inflight.Load())
+}
+
+func TestResolver_NoLoopFlagWhenNotRefreshing(t *testing.T) {
+	r := NewResolver()
+
+	q := dns.Question{Name: "mgmt.example.com.", Qtype: dns.TypeA, Qclass: dns.ClassINET}
+	r.records[q] = &cachedRecord{
+		records: []dns.RR{&dns.A{
+			Hdr: dns.RR_Header{Name: q.Name, Rrtype: dns.TypeA, Class: dns.ClassINET, Ttl: 60},
+			A:   net.ParseIP("10.0.0.1").To4(),
+		}},
+		cachedAt: time.Now(),
+	}
+
+	queryA(t, r, "mgmt.example.com.")
+
+	r.mutex.RLock()
+	_, ok := r.refreshing[q]
+	r.mutex.RUnlock()
+	assert.False(t, ok, "no refresh inflight means no loop tracking")
+}
+
+func TestResolver_AddDomain_UsesChainWhenRootRegistered(t *testing.T) {
+	r := NewResolver()
+	chain := newFakeChain()
+	chain.setAnswer("mgmt.example.com.", dns.TypeA, "10.0.0.2")
+	chain.setAnswer("mgmt.example.com.", dns.TypeAAAA, "fd00::2")
+	r.SetChainResolver(chain, 50)
+
+	require.NoError(t, r.AddDomain(context.Background(), domain.Domain("mgmt.example.com")))
+
+	resp := queryA(t, r, "mgmt.example.com.")
+	assert.Equal(t, "10.0.0.2", firstA(t, resp))
+	assert.Equal(t, 1, chain.callCount("mgmt.example.com.", dns.TypeA))
+	assert.Equal(t, 1, chain.callCount("mgmt.example.com.", dns.TypeAAAA))
+}

client/internal/dns/server.go

@@ -212,6 +212,7 @@ func newDefaultServer(
 	ctx, stop := context.WithCancel(ctx)
 
 	mgmtCacheResolver := mgmt.NewResolver()
+	mgmtCacheResolver.SetChainResolver(handlerChain, PriorityUpstream)
 
 	defaultServer := &DefaultServer{
 		ctx:               ctx,

client/internal/engine.go

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

client/internal/mobile_dependency.go

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

client/internal/peer/conn.go

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

client/internal/peer/conn_status.go

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

client/internal/peer/conn_status_eval_test.go

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

client/internal/peer/env.go

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

client/internal/peer/guard/guard.go

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

client/internal/peer/guard/ice_retry_state.go

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

client/internal/peer/guard/ice_retry_state_test.go

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

client/internal/peer/guard/sr_watcher.go

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

client/internal/peer/handshaker.go

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

client/internal/peer/signaler.go

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

client/internal/routemanager/systemops/systemops_bsd_other.go

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

client/internal/routemanager/systemops/systemops_darwin.go

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

client/internal/routemanager/systemops/systemops_generic.go

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

client/internal/routemanager/systemops/systemops_js.go

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

client/internal/routemanager/systemops/systemops_linux.go

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

client/internal/routemanager/systemops/systemops_nonlinux.go

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

client/internal/routemanager/systemops/systemops_unix.go

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

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
-	var legacyRouting bool
+	legacyRouting := false
 	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() {
-		log.Info("advanced routing has been requested to be disabled")
+	if legacyRouting {
+		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 {
-			merr = multierror.Append(merr, fmt.Errorf("flush marked routes: %w", err))
-		}
+	if err := systemops.New(nil, nil).FlushMarkedRoutes(); err != nil {
+		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, " ")
-	config := fmt.Sprintf("Host %s\n", hostLine)
-	config += fmt.Sprintf("    Match exec \"%s ssh detect %%h %%p\"\n", execPath)
-	config += "        PreferredAuthentications password,publickey,keyboard-interactive\n"
-	config += "        PasswordAuthentication yes\n"
-	config += "        PubkeyAuthentication yes\n"
-	config += "        BatchMode no\n"
-	config += fmt.Sprintf("        ProxyCommand %s ssh proxy %%h %%p\n", execPath)
-	config += "        StrictHostKeyChecking no\n"
+	hostList := strings.Join(deduplicatedPatterns, ",")
+	config := fmt.Sprintf("Match host \"%s\" exec \"%s ssh detect %%h %%p\"\n", hostList, execPath)
+	config += "    PreferredAuthentications password,publickey,keyboard-interactive\n"
+	config += "    PasswordAuthentication yes\n"
+	config += "    PubkeyAuthentication yes\n"
+	config += "    BatchMode no\n"
+	config += fmt.Sprintf("    ProxyCommand %s ssh proxy %%h %%p\n", execPath)
+	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 += "        LogLevel ERROR\n\n"
+	config += "    CheckHostIP no\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: []
-  #   trustedHTTPProxiesCount: 0
-  #   trustedPeers: []
+  #   trustedHTTPProxies: []           # CIDRs of trusted reverse proxies (e.g. ["10.0.0.0/8"])
+  #   trustedHTTPProxiesCount: 0       # Number of trusted proxies in front of the server (alternative to trustedHTTPProxies)
+  #   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=