Merge remote-tracking branch 'origin/main' into reduce-embed-wg-pool

# Conflicts:
#	client/embed/embed.go
#	proxy/cmd/proxy/cmd/debug.go
#	proxy/internal/debug/handler.go

client/internal/connect.go

@@ -333,6 +333,10 @@ func (c *ConnectClient) run(mobileDependency MobileDependency, runningChan chan
 		c.statusRecorder.MarkSignalConnected()
 
 		relayURLs, token := parseRelayInfo(loginResp)
+		if override, ok := peer.OverrideRelayURLs(); ok {
+			log.Infof("overriding relay URLs from %s: %v", peer.EnvKeyNBHomeRelayServers, override)
+			relayURLs = override
+		}
 		peerConfig := loginResp.GetPeerConfig()
 
 		engineConfig, err := createEngineConfig(myPrivateKey, c.config, peerConfig, logPath)

client/internal/debug/debug.go

@@ -61,6 +61,7 @@ allocs.prof: Allocations profiling information.
 threadcreate.prof: Thread creation profiling information.
 cpu.prof: CPU profiling information.
 stack_trace.txt: Complete stack traces of all goroutines at the time of bundle creation.
+capture.pcap: Packet capture in pcap format. Only present when capture was running during bundle collection. Omitted from anonymized bundles because it contains raw decrypted packet data.
 
 
 Anonymization Process
@@ -234,6 +235,7 @@ type BundleGenerator struct {
 	logPath        string
 	tempDir        string
 	cpuProfile     []byte
+	capturePath    string
 	refreshStatus  func() // Optional callback to refresh status before bundle generation
 	clientMetrics  MetricsExporter
 
@@ -257,7 +259,8 @@ type GeneratorDependencies struct {
 	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
+	CapturePath    string
+	RefreshStatus  func()
 	ClientMetrics  MetricsExporter
 }
 
@@ -277,6 +280,7 @@ func NewBundleGenerator(deps GeneratorDependencies, cfg BundleConfig) *BundleGen
 		logPath:        deps.LogPath,
 		tempDir:        deps.TempDir,
 		cpuProfile:     deps.CPUProfile,
+		capturePath:    deps.CapturePath,
 		refreshStatus:  deps.RefreshStatus,
 		clientMetrics:  deps.ClientMetrics,
 
@@ -346,6 +350,10 @@ func (g *BundleGenerator) createArchive() error {
 		log.Errorf("failed to add CPU profile to debug bundle: %v", err)
 	}
 
+	if err := g.addCaptureFile(); err != nil {
+		log.Errorf("failed to add capture file to debug bundle: %v", err)
+	}
+
 	if err := g.addStackTrace(); err != nil {
 		log.Errorf("failed to add stack trace to debug bundle: %v", err)
 	}
@@ -669,6 +677,29 @@ func (g *BundleGenerator) addCPUProfile() error {
 	return nil
 }
 
+func (g *BundleGenerator) addCaptureFile() error {
+	if g.capturePath == "" {
+		return nil
+	}
+
+	if g.anonymize {
+		log.Info("skipping capture file in anonymized bundle (contains raw packet data)")
+		return nil
+	}
+
+	f, err := os.Open(g.capturePath)
+	if err != nil {
+		return fmt.Errorf("open capture file: %w", err)
+	}
+	defer f.Close()
+
+	if err := g.addFileToZip(f, "capture.pcap"); err != nil {
+		return fmt.Errorf("add capture file to zip: %w", err)
+	}
+
+	return nil
+}
+
 func (g *BundleGenerator) addStackTrace() error {
 	buf := make([]byte, 5242880) // 5 MB buffer
 	n := runtime.Stack(buf, true)

client/internal/debug/upload_test.go

@@ -3,10 +3,12 @@ package debug
 import (
 	"context"
 	"errors"
+	"net"
 	"net/http"
 	"os"
 	"path/filepath"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/require"
 
@@ -19,8 +21,10 @@ func TestUpload(t *testing.T) {
 		t.Skip("Skipping upload test on docker ci")
 	}
 	testDir := t.TempDir()
-	testURL := "http://localhost:8080"
+	addr := reserveLoopbackPort(t)
+	testURL := "http://" + addr
 	t.Setenv("SERVER_URL", testURL)
+	t.Setenv("SERVER_ADDRESS", addr)
 	t.Setenv("STORE_DIR", testDir)
 	srv := server.NewServer()
 	go func() {
@@ -33,6 +37,7 @@ func TestUpload(t *testing.T) {
 			t.Errorf("Failed to stop server: %v", err)
 		}
 	})
+	waitForServer(t, addr)
 
 	file := filepath.Join(t.TempDir(), "tmpfile")
 	fileContent := []byte("test file content")
@@ -47,3 +52,30 @@ func TestUpload(t *testing.T) {
 	require.NoError(t, err)
 	require.Equal(t, fileContent, createdFileContent)
 }
+
+// reserveLoopbackPort binds an ephemeral port on loopback to learn a free
+// address, then releases it so the server under test can rebind. The close/
+// rebind window is racy in theory; on loopback with a kernel-assigned port
+// it's essentially never contended in practice.
+func reserveLoopbackPort(t *testing.T) string {
+	t.Helper()
+	l, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+	addr := l.Addr().String()
+	require.NoError(t, l.Close())
+	return addr
+}
+
+func waitForServer(t *testing.T, addr string) {
+	t.Helper()
+	deadline := time.Now().Add(5 * time.Second)
+	for time.Now().Before(deadline) {
+		c, err := net.DialTimeout("tcp", addr, 100*time.Millisecond)
+		if err == nil {
+			_ = c.Close()
+			return
+		}
+		time.Sleep(20 * time.Millisecond)
+	}
+	t.Fatalf("server did not start listening on %s in time", addr)
+}

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 +2,83 @@ package mgmt
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"net"
-	"net/netip"
 	"net/url"
+	"os"
+	"slices"
 	"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
+
+	cacheTTL time.Duration
 }
 
 // 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),
+		cacheTTL:   resolveCacheTTL(),
 	}
 }
 
@@ -44,7 +87,19 @@ 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 +115,14 @@ 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) > m.cacheTTL
+		inBackoff := !cached.lastFailedRefresh.IsZero() && time.Since(cached.lastFailedRefresh) < refreshBackoff
+		shouldRefresh = stale && !inBackoff
+	}
 	m.mutex.RUnlock()
 
 	if !found {
@@ -68,12 +130,23 @@ 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, cached)
+	}
+
 	resp := &dns.Msg{}
 	resp.SetReply(r)
 	resp.Authoritative = false
 	resp.RecursionAvailable = true
-
-	resp.Answer = append(resp.Answer, records...)
+	resp.Answer = cloneRecordsWithTTL(cached.records, m.responseTTL(cached.cachedAt))
 
 	log.Debugf("serving %d cached records for domain=%s", len(resp.Answer), question.Name)
 
@@ -98,101 +171,260 @@ 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)
+	if len(aRecords) == 0 && len(aaaaRecords) == 0 {
+		if err := errors.Join(errA, errAAAA); err != nil {
+			return fmt.Errorf("resolve %s: no A/AAAA records: %w", d.SafeString(), err)
 		}
+		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. expected is the
+// cachedRecord pointer observed by the caller; the refresh only mutates the
+// cache if that pointer is still the one stored, so a stale in-flight refresh
+// can't clobber a newer entry written by AddDomain or a competing refresh.
+func (m *Resolver) scheduleRefresh(question dns.Question, expected *cachedRecord) {
+	key := question.Name + "|" + dns.TypeToString[question.Qtype]
+	_ = m.refreshGroup.DoChan(key, func() (any, error) {
+		return nil, m.refreshQuestion(question, expected)
+	})
+}
+
+// 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.
+// expected pins the cache entry observed at schedule time; mutations only apply
+// if m.records[question] still points at it.
+func (m *Resolver) refreshQuestion(question dns.Question, expected *cachedRecord) error {
+	ctx, cancel := context.WithTimeout(context.Background(), dnsTimeout)
+	defer cancel()
+
+	d, err := domain.FromString(strings.TrimSuffix(question.Name, "."))
+	if err != nil {
+		m.markRefreshFailed(question, expected)
+		return fmt.Errorf("parse domain: %w", err)
+	}
+
+	records, err := m.lookupRecords(ctx, d, question)
+	if err != nil {
+		fails := m.markRefreshFailed(question, expected)
+		logf := log.Warnf
+		if fails == 0 || 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()
+		if m.records[question] == expected {
+			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
+		}
+		m.mutex.Unlock()
+		log.Debugf("skipping refresh evict for domain=%s type=%s: entry changed during refresh",
+			d.SafeString(), dns.TypeToString[question.Qtype])
+		return nil
 	}
-	return resp.ips, nil
+
+	now := time.Now()
+	m.mutex.Lock()
+	if m.records[question] != expected {
+		m.mutex.Unlock()
+		log.Debugf("skipping refresh write for domain=%s type=%s: entry changed 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, expected *cachedRecord) int {
+	m.mutex.Lock()
+	defer m.mutex.Unlock()
+	c, ok := m.records[question]
+	if !ok || c != expected {
+		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 = m.lookupViaChain(ctx, chain, maxPriority, dnsName, dns.TypeA)
+		aaaaRecords, errAAAA = m.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 = m.osLookup(ctx, d, dnsName, dns.TypeA)
+	aaaaRecords, errAAAA = m.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 m.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 m.osLookup(ctx, d, q.Name, q.Qtype)
+}
+
+// lookupViaChain resolves via the handler chain and rewrites each RR to use
+// dnsName as owner and m.cacheTTL as TTL, so CNAME-backed domains don't cache
+// target-owned records or upstream TTLs. NODATA returns (nil, nil).
+func (m *Resolver) 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(m.cacheTTL.Seconds())
+	owners := cnameOwners(dnsName, resp.Answer)
+	var filtered []dns.RR
+	for _, rr := range resp.Answer {
+		h := rr.Header()
+		if h.Class != dns.ClassINET || h.Rrtype != qtype {
+			continue
+		}
+		if !owners[strings.ToLower(dns.Fqdn(h.Name))] {
+			continue
+		}
+		if cp := cloneIPRecord(rr, dnsName, ttl); cp != nil {
+			filtered = append(filtered, cp)
+		}
+	}
+	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 (m *Resolver) 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(m.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])
+}
+
+// responseTTL returns the remaining cache lifetime in seconds (rounded up),
+// so downstream resolvers don't cache an answer for longer than we will.
+func (m *Resolver) responseTTL(cachedAt time.Time) uint32 {
+	remaining := m.cacheTTL - time.Since(cachedAt)
+	if remaining <= 0 {
+		return 0
+	}
+	return uint32((remaining + time.Second - 1) / time.Second)
 }
 
 // PopulateFromConfig extracts and caches domains from the client configuration.
@@ -224,19 +456,12 @@ 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)
+	qA := dns.Question{Name: dnsName, Qtype: dns.TypeA, Qclass: dns.ClassINET}
+	qAAAA := dns.Question{Name: dnsName, Qtype: dns.TypeAAAA, Qclass: dns.ClassINET}
+	delete(m.records, qA)
+	delete(m.records, qAAAA)
+	delete(m.refreshing, qA)
+	delete(m.refreshing, qAAAA)
 
 	log.Debugf("removed domain=%s from cache", d.SafeString())
 	return nil
@@ -394,3 +619,73 @@ func (m *Resolver) extractDomainsFromServerDomains(serverDomains dnsconfig.Serve
 
 	return domains
 }
+
+// cloneIPRecord returns a deep copy of rr retargeted to owner with ttl. Non
+// A/AAAA records return nil.
+func cloneIPRecord(rr dns.RR, owner string, ttl uint32) dns.RR {
+	switch r := rr.(type) {
+	case *dns.A:
+		cp := *r
+		cp.Hdr.Name = owner
+		cp.Hdr.Ttl = ttl
+		cp.A = slices.Clone(r.A)
+		return &cp
+	case *dns.AAAA:
+		cp := *r
+		cp.Hdr.Name = owner
+		cp.Hdr.Ttl = ttl
+		cp.AAAA = slices.Clone(r.AAAA)
+		return &cp
+	}
+	return nil
+}
+
+// cloneRecordsWithTTL clones A/AAAA records preserving their owner and
+// stamping ttl so the response shares no memory with the cached slice.
+func cloneRecordsWithTTL(records []dns.RR, ttl uint32) []dns.RR {
+	out := make([]dns.RR, 0, len(records))
+	for _, rr := range records {
+		if cp := cloneIPRecord(rr, rr.Header().Name, ttl); cp != nil {
+			out = append(out, cp)
+		}
+	}
+	return out
+}
+
+// cnameOwners returns dnsName plus every target reachable by following CNAMEs
+// in answer, iterating until fixed point so out-of-order chains resolve.
+func cnameOwners(dnsName string, answer []dns.RR) map[string]bool {
+	owners := map[string]bool{dnsName: true}
+	for {
+		added := false
+		for _, rr := range answer {
+			cname, ok := rr.(*dns.CNAME)
+			if !ok {
+				continue
+			}
+			name := strings.ToLower(dns.Fqdn(cname.Hdr.Name))
+			if !owners[name] {
+				continue
+			}
+			target := strings.ToLower(dns.Fqdn(cname.Target))
+			if !owners[target] {
+				owners[target] = true
+				added = true
+			}
+		}
+		if !added {
+			return owners
+		}
+	}
+}
+
+// resolveCacheTTL reads the cache TTL override env var; invalid or empty
+// values fall back to defaultTTL. Called once per Resolver from NewResolver.
+func resolveCacheTTL() time.Duration {
+	if v := os.Getenv(envMgmtCacheTTL); v != "" {
+		if d, err := time.ParseDuration(v); err == nil && d > 0 {
+			return d
+		}
+	}
+	return defaultTTL
+}

client/internal/dns/mgmt/mgmt_refresh_test.go

@@ -0,0 +1,408 @@
+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_CacheTTLGatesRefresh(t *testing.T) {
+	// Same cached entry age, different cacheTTL values: the shorter TTL must
+	// trigger a background refresh, the longer one must not. Proves that the
+	// per-Resolver cacheTTL field actually drives the stale decision.
+	cachedAt := time.Now().Add(-100 * time.Millisecond)
+
+	newRec := func() *cachedRecord {
+		return &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: cachedAt,
+		}
+	}
+	q := dns.Question{Name: "mgmt.example.com.", Qtype: dns.TypeA, Qclass: dns.ClassINET}
+
+	t.Run("short TTL treats entry as stale and refreshes", func(t *testing.T) {
+		r := NewResolver()
+		r.cacheTTL = 10 * time.Millisecond
+		chain := newFakeChain()
+		chain.setAnswer(q.Name, dns.TypeA, "10.0.0.2")
+		r.SetChainResolver(chain, 50)
+		r.records[q] = newRec()
+
+		resp := queryA(t, r, q.Name)
+		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(q.Name, dns.TypeA) >= 1
+		})
+	})
+
+	t.Run("long TTL keeps entry fresh and skips refresh", func(t *testing.T) {
+		r := NewResolver()
+		r.cacheTTL = time.Hour
+		chain := newFakeChain()
+		chain.setAnswer(q.Name, dns.TypeA, "10.0.0.2")
+		r.SetChainResolver(chain, 50)
+		r.records[q] = newRec()
+
+		resp := queryA(t, r, q.Name)
+		assert.Equal(t, "10.0.0.1", firstA(t, resp))
+
+		time.Sleep(50 * time.Millisecond)
+		assert.Equal(t, 0, chain.callCount(q.Name, dns.TypeA), "fresh entry must not trigger refresh")
+	})
+}
+
+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/mgmt/mgmt_test.go

@@ -6,6 +6,7 @@ import (
 	"net/url"
 	"strings"
 	"testing"
+	"time"
 
 	"github.com/miekg/dns"
 	"github.com/stretchr/testify/assert"
@@ -23,6 +24,60 @@ func TestResolver_NewResolver(t *testing.T) {
 	assert.False(t, resolver.MatchSubdomains())
 }
 
+func TestResolveCacheTTL(t *testing.T) {
+	tests := []struct {
+		name  string
+		value string
+		want  time.Duration
+	}{
+		{"unset falls back to default", "", defaultTTL},
+		{"valid duration", "45s", 45 * time.Second},
+		{"valid minutes", "2m", 2 * time.Minute},
+		{"malformed falls back to default", "not-a-duration", defaultTTL},
+		{"zero falls back to default", "0s", defaultTTL},
+		{"negative falls back to default", "-5s", defaultTTL},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv(envMgmtCacheTTL, tc.value)
+			got := resolveCacheTTL()
+			assert.Equal(t, tc.want, got, "parsed TTL should match")
+		})
+	}
+}
+
+func TestNewResolver_CacheTTLFromEnv(t *testing.T) {
+	t.Setenv(envMgmtCacheTTL, "7s")
+	r := NewResolver()
+	assert.Equal(t, 7*time.Second, r.cacheTTL, "NewResolver should evaluate cacheTTL once from env")
+}
+
+func TestResolver_ResponseTTL(t *testing.T) {
+	now := time.Now()
+	tests := []struct {
+		name     string
+		cacheTTL time.Duration
+		cachedAt time.Time
+		wantMin  uint32
+		wantMax  uint32
+	}{
+		{"fresh entry returns full TTL", 60 * time.Second, now, 59, 60},
+		{"half-aged entry returns half TTL", 60 * time.Second, now.Add(-30 * time.Second), 29, 31},
+		{"expired entry returns zero", 60 * time.Second, now.Add(-61 * time.Second), 0, 0},
+		{"exactly expired returns zero", 10 * time.Second, now.Add(-10 * time.Second), 0, 0},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			r := &Resolver{cacheTTL: tc.cacheTTL}
+			got := r.responseTTL(tc.cachedAt)
+			assert.GreaterOrEqual(t, got, tc.wantMin, "remaining TTL should be >= wantMin")
+			assert.LessOrEqual(t, got, tc.wantMax, "remaining TTL should be <= wantMax")
+		})
+	}
+}
+
 func TestResolver_ExtractDomainFromURL(t *testing.T) {
 	tests := []struct {
 		name        string

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

@@ -28,6 +28,7 @@ import (
 	"github.com/netbirdio/netbird/client/firewall"
 	"github.com/netbirdio/netbird/client/firewall/firewalld"
 	firewallManager "github.com/netbirdio/netbird/client/firewall/manager"
+	"github.com/netbirdio/netbird/client/firewall/uspfilter/forwarder"
 	"github.com/netbirdio/netbird/client/iface"
 	"github.com/netbirdio/netbird/client/iface/device"
 	nbnetstack "github.com/netbirdio/netbird/client/iface/netstack"
@@ -68,6 +69,7 @@ import (
 	signal "github.com/netbirdio/netbird/shared/signal/client"
 	sProto "github.com/netbirdio/netbird/shared/signal/proto"
 	"github.com/netbirdio/netbird/util"
+	"github.com/netbirdio/netbird/util/capture"
 )
 
 // PeerConnectionTimeoutMax is a timeout of an initial connection attempt to a remote peer.
@@ -218,6 +220,8 @@ type Engine struct {
 	portForwardManager *portforward.Manager
 	srWatcher          *guard.SRWatcher
 
+	afpacketCapture *capture.AFPacketCapture
+
 	// Sync response persistence (protected by syncRespMux)
 	syncRespMux         sync.RWMutex
 	persistSyncResponse bool
@@ -944,7 +948,12 @@ func (e *Engine) handleRelayUpdate(update *mgmProto.RelayConfig) error {
 			return fmt.Errorf("update relay token: %w", err)
 		}
 
-		e.relayManager.UpdateServerURLs(update.Urls)
+		urls := update.Urls
+		if override, ok := peer.OverrideRelayURLs(); ok {
+			log.Infof("overriding relay URLs from %s: %v", peer.EnvKeyNBHomeRelayServers, override)
+			urls = override
+		}
+		e.relayManager.UpdateServerURLs(urls)
 
 		// Just in case the agent started with an MGM server where the relay was disabled but was later enabled.
 		// We can ignore all errors because the guard will manage the reconnection retries.
@@ -1698,6 +1707,11 @@ func (e *Engine) parseNATExternalIPMappings() []string {
 }
 
 func (e *Engine) close() {
+	if e.afpacketCapture != nil {
+		e.afpacketCapture.Stop()
+		e.afpacketCapture = nil
+	}
+
 	log.Debugf("removing Netbird interface %s", e.config.WgIfaceName)
 
 	if e.wgInterface != nil {
@@ -2186,6 +2200,62 @@ func (e *Engine) Address() (netip.Addr, error) {
 	return e.wgInterface.Address().IP, nil
 }
 
+// SetCapture sets or clears packet capture on the WireGuard device.
+// On userspace WireGuard, it taps the FilteredDevice directly.
+// On kernel WireGuard (Linux), it falls back to AF_PACKET raw socket capture.
+// Pass nil to disable capture.
+func (e *Engine) SetCapture(pc device.PacketCapture) error {
+	e.syncMsgMux.Lock()
+	defer e.syncMsgMux.Unlock()
+
+	intf := e.wgInterface
+	if intf == nil {
+		return errors.New("wireguard interface not initialized")
+	}
+
+	if e.afpacketCapture != nil {
+		e.afpacketCapture.Stop()
+		e.afpacketCapture = nil
+	}
+
+	dev := intf.GetDevice()
+	if dev != nil {
+		dev.SetCapture(pc)
+		e.setForwarderCapture(pc)
+		return nil
+	}
+
+	// Kernel mode: no FilteredDevice. Use AF_PACKET on Linux.
+	if pc == nil {
+		return nil
+	}
+	sess, ok := pc.(*capture.Session)
+	if !ok {
+		return errors.New("filtered device not available and AF_PACKET requires *capture.Session")
+	}
+
+	afc := capture.NewAFPacketCapture(intf.Name(), sess)
+	if err := afc.Start(); err != nil {
+		return fmt.Errorf("start AF_PACKET capture on %s: %w", intf.Name(), err)
+	}
+	e.afpacketCapture = afc
+	return nil
+}
+
+// setForwarderCapture propagates capture to the USP filter's forwarder endpoint.
+// This captures outbound response packets that bypass the FilteredDevice in netstack mode.
+func (e *Engine) setForwarderCapture(pc device.PacketCapture) {
+	if e.firewall == nil {
+		return
+	}
+	type forwarderCapturer interface {
+		SetPacketCapture(pc forwarder.PacketCapture)
+	}
+	if fc, ok := e.firewall.(forwarderCapturer); ok {
+		fc.SetPacketCapture(pc)
+	}
+}
+
 func (e *Engine) updateForwardRules(rules []*mgmProto.ForwardingRule) ([]firewallManager.ForwardRule, error) {
 	if e.firewall == nil {
 		log.Warn("firewall is disabled, not updating forwarding rules")

client/internal/engine_test.go

@@ -1671,7 +1671,7 @@ func startManagement(t *testing.T, dataDir, testFile string) (*grpc.Server, stri
 	if err != nil {
 		return nil, "", err
 	}
-	mgmtServer, err := nbgrpc.NewServer(config, accountManager, settingsMockManager, jobManager, secretsManager, nil, nil, &server.MockIntegratedValidator{}, networkMapController, nil)
+	mgmtServer, err := nbgrpc.NewServer(config, accountManager, settingsMockManager, jobManager, secretsManager, nil, nil, &server.MockIntegratedValidator{}, networkMapController, nil, nil)
 	if err != nil {
 		return nil, "", err
 	}

client/internal/lazyconn/activity/listener_bind_test.go

@@ -3,7 +3,6 @@ package activity
 import (
 	"net"
 	"net/netip"
-	"runtime"
 	"testing"
 	"time"
 
@@ -18,10 +17,6 @@ import (
 	peerid "github.com/netbirdio/netbird/client/internal/peer/id"
 )
 
-func isBindListenerPlatform() bool {
-	return runtime.GOOS == "windows" || runtime.GOOS == "js"
-}
-
 // mockEndpointManager implements device.EndpointManager for testing
 type mockEndpointManager struct {
 	endpoints map[netip.Addr]net.Conn
@@ -181,10 +176,6 @@ func TestBindListener_Close(t *testing.T) {
 }
 
 func TestManager_BindMode(t *testing.T) {
-	if !isBindListenerPlatform() {
-		t.Skip("BindListener only used on Windows/JS platforms")
-	}
-
 	mockEndpointMgr := newMockEndpointManager()
 	mockIface := &MockWGIfaceBind{endpointMgr: mockEndpointMgr}
 
@@ -226,10 +217,6 @@ func TestManager_BindMode(t *testing.T) {
 }
 
 func TestManager_BindMode_MultiplePeers(t *testing.T) {
-	if !isBindListenerPlatform() {
-		t.Skip("BindListener only used on Windows/JS platforms")
-	}
-
 	mockEndpointMgr := newMockEndpointManager()
 	mockIface := &MockWGIfaceBind{endpointMgr: mockEndpointMgr}
 

client/internal/lazyconn/activity/manager.go

@@ -4,14 +4,12 @@ import (
 	"errors"
 	"net"
 	"net/netip"
-	"runtime"
 	"sync"
 	"time"
 
 	log "github.com/sirupsen/logrus"
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 
-	"github.com/netbirdio/netbird/client/iface/netstack"
 	"github.com/netbirdio/netbird/client/iface/wgaddr"
 	"github.com/netbirdio/netbird/client/internal/lazyconn"
 	peerid "github.com/netbirdio/netbird/client/internal/peer/id"
@@ -75,16 +73,6 @@ func (m *Manager) createListener(peerCfg lazyconn.PeerConfig) (listener, error)
 		return NewUDPListener(m.wgIface, peerCfg)
 	}
 
-	// BindListener is used on Windows, JS, and netstack platforms:
-	// - JS: Cannot listen to UDP sockets
-	// - Windows: IP_UNICAST_IF socket option forces packets out the interface the default
-	//   gateway points to, preventing them from reaching the loopback interface.
-	// - Netstack: Allows multiple instances on the same host without port conflicts.
-	// BindListener bypasses these issues by passing data directly through the bind.
-	if runtime.GOOS != "windows" && runtime.GOOS != "js" && !netstack.IsEnabled() {
-		return NewUDPListener(m.wgIface, peerCfg)
-	}
-
 	provider, ok := m.wgIface.(bindProvider)
 	if !ok {
 		return nil, errors.New("interface claims userspace bind but doesn't implement bindProvider")

client/internal/lazyconn/manager/manager.go

@@ -6,7 +6,6 @@ import (
 	"time"
 
 	log "github.com/sirupsen/logrus"
-	"golang.org/x/exp/maps"
 
 	"github.com/netbirdio/netbird/client/internal/lazyconn"
 	"github.com/netbirdio/netbird/client/internal/lazyconn/activity"
@@ -91,8 +90,8 @@ func (m *Manager) UpdateRouteHAMap(haMap route.HAMap) {
 	m.routesMu.Lock()
 	defer m.routesMu.Unlock()
 
-	maps.Clear(m.peerToHAGroups)
-	maps.Clear(m.haGroupToPeers)
+	clear(m.peerToHAGroups)
+	clear(m.haGroupToPeers)
 
 	for haUniqueID, routes := range haMap {
 		var peers []string

client/internal/netflow/store/memory.go

@@ -3,8 +3,6 @@ package store
 import (
 	"sync"
 
-	"golang.org/x/exp/maps"
-
 	"github.com/google/uuid"
 
 	"github.com/netbirdio/netbird/client/internal/netflow/types"
@@ -30,7 +28,7 @@ func (m *Memory) StoreEvent(event *types.Event) {
 func (m *Memory) Close() {
 	m.mux.Lock()
 	defer m.mux.Unlock()
-	maps.Clear(m.events)
+	clear(m.events)
 }
 
 func (m *Memory) GetEvents() []*types.Event {

client/internal/peer/env.go

@@ -7,7 +7,8 @@ import (
 )
 
 const (
-	EnvKeyNBForceRelay = "NB_FORCE_RELAY"
+	EnvKeyNBForceRelay       = "NB_FORCE_RELAY"
+	EnvKeyNBHomeRelayServers = "NB_HOME_RELAY_SERVERS"
 )
 
 func IsForceRelayed() bool {
@@ -16,3 +17,28 @@ func IsForceRelayed() bool {
 	}
 	return strings.EqualFold(os.Getenv(EnvKeyNBForceRelay), "true")
 }
+
+// OverrideRelayURLs returns the relay server URL list set in
+// NB_HOME_RELAY_SERVERS (comma-separated) and a boolean indicating whether
+// the override is active. When the env var is unset, the boolean is false
+// and the caller should keep the list received from the management server.
+// Intended for lab/debug scenarios where a peer must pin to a specific home
+// relay regardless of what management offers.
+func OverrideRelayURLs() ([]string, bool) {
+	raw := os.Getenv(EnvKeyNBHomeRelayServers)
+	if raw == "" {
+		return nil, false
+	}
+	parts := strings.Split(raw, ",")
+	urls := make([]string, 0, len(parts))
+	for _, p := range parts {
+		p = strings.TrimSpace(p)
+		if p != "" {
+			urls = append(urls, p)
+		}
+	}
+	if len(urls) == 0 {
+		return nil, false
+	}
+	return urls, true
+}

client/internal/routemanager/systemops/systemops_darwin.go

@@ -89,8 +89,16 @@ func (r *SysOps) installScopedDefaultFor(unspec netip.Addr) (bool, error) {
 		return false, fmt.Errorf("unusable default nexthop for %s (no interface)", unspec)
 	}
 
+	reused := false
 	if err := r.addScopedDefault(unspec, nexthop); err != nil {
-		return false, fmt.Errorf("add scoped default on %s: %w", nexthop.Intf.Name, err)
+		if !errors.Is(err, unix.EEXIST) {
+			return false, fmt.Errorf("add scoped default on %s: %w", nexthop.Intf.Name, err)
+		}
+		// macOS installs its own RTF_IFSCOPE defaults for primary service
+		// selection on multi-NIC setups, so a route on this ifindex can
+		// already exist before we try. Binding to it via IP[V6]_BOUND_IF
+		// still produces the scoped lookup we need.
+		reused = true
 	}
 
 	af := unix.AF_INET
@@ -102,7 +110,11 @@ func (r *SysOps) installScopedDefaultFor(unspec netip.Addr) (bool, error) {
 	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))
+	verb := "installed"
+	if reused {
+		verb = "reused existing"
+	}
+	log.Infof("%s scoped default route via %s on %s for %s", verb, via, nexthop.Intf.Name, afOf(unspec))
 	return true, nil
 }
 

client/internal/routeselector/routeselector.go

@@ -7,7 +7,6 @@ import (
 	"sync"
 
 	"github.com/hashicorp/go-multierror"
-	"golang.org/x/exp/maps"
 
 	"github.com/netbirdio/netbird/client/errors"
 	"github.com/netbirdio/netbird/route"
@@ -44,8 +43,8 @@ func (rs *RouteSelector) SelectRoutes(routes []route.NetID, appendRoute bool, al
 		if rs.selectedRoutes == nil {
 			rs.selectedRoutes = map[route.NetID]struct{}{}
 		}
-		maps.Clear(rs.deselectedRoutes)
-		maps.Clear(rs.selectedRoutes)
+		clear(rs.deselectedRoutes)
+		clear(rs.selectedRoutes)
 		for _, r := range allRoutes {
 			rs.deselectedRoutes[r] = struct{}{}
 		}
@@ -78,8 +77,8 @@ func (rs *RouteSelector) SelectAllRoutes() {
 	if rs.selectedRoutes == nil {
 		rs.selectedRoutes = map[route.NetID]struct{}{}
 	}
-	maps.Clear(rs.deselectedRoutes)
-	maps.Clear(rs.selectedRoutes)
+	clear(rs.deselectedRoutes)
+	clear(rs.selectedRoutes)
 }
 
 // DeselectRoutes removes specific routes from the selection.
@@ -116,8 +115,8 @@ func (rs *RouteSelector) DeselectAllRoutes() {
 	if rs.selectedRoutes == nil {
 		rs.selectedRoutes = map[route.NetID]struct{}{}
 	}
-	maps.Clear(rs.deselectedRoutes)
-	maps.Clear(rs.selectedRoutes)
+	clear(rs.deselectedRoutes)
+	clear(rs.selectedRoutes)
 }
 
 // IsSelected checks if a specific route is selected.

client/internal/sleep/detector_darwin.go

@@ -2,217 +2,358 @@
 
 package sleep
 
-/*
-#cgo LDFLAGS: -framework IOKit -framework CoreFoundation
-#include <IOKit/pwr_mgt/IOPMLib.h>
-#include <IOKit/IOMessage.h>
-#include <CoreFoundation/CoreFoundation.h>
-
-extern void sleepCallbackBridge();
-extern void poweredOnCallbackBridge();
-extern void suspendedCallbackBridge();
-extern void resumedCallbackBridge();
-
-
-// C global variables for IOKit state
-static IONotificationPortRef g_notifyPortRef = NULL;
-static io_object_t g_notifierObject = 0;
-static io_object_t g_generalInterestNotifier = 0;
-static io_connect_t g_rootPort = 0;
-static CFRunLoopRef g_runLoop = NULL;
-
-static void sleepCallback(void* refCon, io_service_t service, natural_t messageType, void* messageArgument) {
-	switch (messageType) {
-		case kIOMessageSystemWillSleep:
-			sleepCallbackBridge();
-			IOAllowPowerChange(g_rootPort, (long)messageArgument);
-			break;
-        case kIOMessageSystemHasPoweredOn:
-          	poweredOnCallbackBridge();
-          	break;
-        case kIOMessageServiceIsSuspended:
-			suspendedCallbackBridge();
-			break;
-		case kIOMessageServiceIsResumed:
-			resumedCallbackBridge();
-			break;
-		default:
-			break;
-	}
-}
-
-static void registerNotifications() {
-	g_rootPort = IORegisterForSystemPower(
-		NULL,
-		&g_notifyPortRef,
-		(IOServiceInterestCallback)sleepCallback,
-		&g_notifierObject
-	);
-
-	if (g_rootPort == 0) {
-		return;
-	}
-
-	CFRunLoopAddSource(CFRunLoopGetCurrent(),
-		IONotificationPortGetRunLoopSource(g_notifyPortRef),
-		kCFRunLoopCommonModes);
-
-	g_runLoop = CFRunLoopGetCurrent();
-	CFRunLoopRun();
-}
-
-static void unregisterNotifications() {
-	CFRunLoopRemoveSource(g_runLoop,
-		IONotificationPortGetRunLoopSource(g_notifyPortRef),
-		kCFRunLoopCommonModes);
-
-	IODeregisterForSystemPower(&g_notifierObject);
-	IOServiceClose(g_rootPort);
-	IONotificationPortDestroy(g_notifyPortRef);
-	CFRunLoopStop(g_runLoop);
-
-	g_notifyPortRef = NULL;
-	g_notifierObject = 0;
-	g_rootPort = 0;
-	g_runLoop = NULL;
-}
-
-*/
-import "C"
-
 import (
-	"context"
 	"fmt"
 	"runtime"
 	"sync"
 	"time"
+	"unsafe"
 
+	"github.com/ebitengine/purego"
 	log "github.com/sirupsen/logrus"
 )
 
-var (
-	serviceRegistry   = make(map[*Detector]struct{})
-	serviceRegistryMu sync.Mutex
+// IOKit message types from IOKit/IOMessage.h.
+const (
+	kIOMessageCanSystemSleep     uintptr = 0xe0000270
+	kIOMessageSystemWillSleep    uintptr = 0xe0000280
+	kIOMessageSystemHasPoweredOn uintptr = 0xe0000300
 )
 
-//export sleepCallbackBridge
-func sleepCallbackBridge() {
-	log.Info("sleepCallbackBridge event triggered")
+var (
+	ioKit         iokitFuncs
+	cf            cfFuncs
+	cfCommonModes uintptr
 
-	serviceRegistryMu.Lock()
-	defer serviceRegistryMu.Unlock()
+	libInitOnce sync.Once
+	libInitErr  error
 
-	for svc := range serviceRegistry {
-		svc.triggerCallback(EventTypeSleep)
-	}
+	// callbackThunk is the single C-callable trampoline registered with IOKit.
+	callbackThunk uintptr
+
+	serviceRegistry   = make(map[*Detector]struct{})
+	serviceRegistryMu sync.Mutex
+	session           *runLoopSession
+
+	// lifecycleMu serializes Register/Deregister so a new registration can't
+	// start a second runloop while a previous teardown is still pending.
+	lifecycleMu sync.Mutex
+)
+
+// iokitFuncs holds IOKit symbols resolved once at init.
+type iokitFuncs struct {
+	IORegisterForSystemPower           func(refcon uintptr, portRef *uintptr, callback uintptr, notifier *uintptr) uintptr
+	IODeregisterForSystemPower         func(notifier *uintptr) int32
+	IOAllowPowerChange                 func(kernelPort uintptr, notificationID uintptr) int32
+	IOServiceClose                     func(connect uintptr) int32
+	IONotificationPortGetRunLoopSource func(port uintptr) uintptr
+	IONotificationPortDestroy          func(port uintptr)
 }
 
-//export resumedCallbackBridge
-func resumedCallbackBridge() {
-	log.Info("resumedCallbackBridge event triggered")
+// cfFuncs holds CoreFoundation symbols resolved once at init.
+type cfFuncs struct {
+	CFRunLoopGetCurrent   func() uintptr
+	CFRunLoopRun          func()
+	CFRunLoopStop         func(rl uintptr)
+	CFRunLoopAddSource    func(rl, source, mode uintptr)
+	CFRunLoopRemoveSource func(rl, source, mode uintptr)
 }
 
-//export suspendedCallbackBridge
-func suspendedCallbackBridge() {
-	log.Info("suspendedCallbackBridge event triggered")
+// runLoopSession bundles the handles owned by one CFRunLoop lifetime. A nil
+// session means no runloop is active and the next Register must start one.
+type runLoopSession struct {
+	rl       uintptr
+	port     uintptr
+	notifier uintptr
+	rp       uintptr
 }
 
-//export poweredOnCallbackBridge
-func poweredOnCallbackBridge() {
-	log.Info("poweredOnCallbackBridge event triggered")
-	serviceRegistryMu.Lock()
-	defer serviceRegistryMu.Unlock()
-
-	for svc := range serviceRegistry {
-		svc.triggerCallback(EventTypeWakeUp)
-	}
+// detectorSnapshot pins a detector's callback and done channel so dispatch
+// runs with values valid at snapshot time, even if a concurrent
+// Deregister/Register rewrites the detector's fields.
+type detectorSnapshot struct {
+	detector *Detector
+	callback func(event EventType)
+	done     <-chan struct{}
 }
 
+// Detector delivers sleep and wake events to a registered callback.
 type Detector struct {
 	callback func(event EventType)
-	ctx      context.Context
-	cancel   context.CancelFunc
-}
-
-func NewDetector() (*Detector, error) {
-	return &Detector{}, nil
+	done     chan struct{}
 }
 
+// Register installs callback for power events. The first registration starts
+// the CFRunLoop on a dedicated OS-locked thread and blocks until IOKit
+// registration succeeds or fails; subsequent registrations just add to the
+// dispatch set.
 func (d *Detector) Register(callback func(event EventType)) error {
-	serviceRegistryMu.Lock()
-	defer serviceRegistryMu.Unlock()
+	lifecycleMu.Lock()
+	defer lifecycleMu.Unlock()
 
+	serviceRegistryMu.Lock()
 	if _, exists := serviceRegistry[d]; exists {
+		serviceRegistryMu.Unlock()
 		return fmt.Errorf("detector service already registered")
 	}
-
 	d.callback = callback
+	d.done = make(chan struct{})
+	serviceRegistry[d] = struct{}{}
+	needSetup := session == nil
+	serviceRegistryMu.Unlock()
 
-	d.ctx, d.cancel = context.WithCancel(context.Background())
-
-	if len(serviceRegistry) > 0 {
-		serviceRegistry[d] = struct{}{}
+	if !needSetup {
 		return nil
 	}
 
-	serviceRegistry[d] = struct{}{}
-
-	// CFRunLoop must run on a single fixed OS thread
-	go func() {
-		runtime.LockOSThread()
-		defer runtime.UnlockOSThread()
-
-		C.registerNotifications()
-	}()
+	errCh := make(chan error, 1)
+	go runRunLoop(errCh)
+	if err := <-errCh; err != nil {
+		serviceRegistryMu.Lock()
+		delete(serviceRegistry, d)
+		close(d.done)
+		d.done = nil
+		serviceRegistryMu.Unlock()
+		return err
+	}
 
 	log.Info("sleep detection service started on macOS")
 	return nil
 }
 
-// Deregister removes the detector. When the last detector is removed, IOKit registration is torn down
-// and the runloop is stopped and cleaned up.
+// Deregister removes the detector. When the last detector leaves, IOKit
+// notifications are torn down and the runloop is stopped.
 func (d *Detector) Deregister() error {
+	lifecycleMu.Lock()
+	defer lifecycleMu.Unlock()
+
 	serviceRegistryMu.Lock()
-	defer serviceRegistryMu.Unlock()
-	_, exists := serviceRegistry[d]
-	if !exists {
+	if _, exists := serviceRegistry[d]; !exists {
+		serviceRegistryMu.Unlock()
 		return nil
 	}
-
-	// cancel and remove this detector
-	d.cancel()
+	close(d.done)
 	delete(serviceRegistry, d)
 
-	// If other Detectors still exist, leave IOKit running
 	if len(serviceRegistry) > 0 {
+		serviceRegistryMu.Unlock()
 		return nil
 	}
+	sess := session
+	serviceRegistryMu.Unlock()
 
 	log.Info("sleep detection service stopping (deregister)")
 
-	// Deregister IOKit notifications, stop runloop, and free resources
-	C.unregisterNotifications()
+	if sess == nil {
+		return nil
+	}
+
+	if sess.rl != 0 && sess.port != 0 {
+		source := ioKit.IONotificationPortGetRunLoopSource(sess.port)
+		cf.CFRunLoopRemoveSource(sess.rl, source, cfCommonModes)
+	}
+	if sess.notifier != 0 {
+		n := sess.notifier
+		ioKit.IODeregisterForSystemPower(&n)
+	}
+
+	// Clear session only after IODeregisterForSystemPower returns so any
+	// in-flight powerCallback can still look up session.rp to ack sleep.
+	serviceRegistryMu.Lock()
+	session = nil
+	serviceRegistryMu.Unlock()
+
+	if sess.rp != 0 {
+		ioKit.IOServiceClose(sess.rp)
+	}
+	if sess.port != 0 {
+		ioKit.IONotificationPortDestroy(sess.port)
+	}
+	if sess.rl != 0 {
+		cf.CFRunLoopStop(sess.rl)
+	}
 
 	return nil
 }
 
-func (d *Detector) triggerCallback(event EventType) {
-	doneChan := make(chan struct{})
+func (d *Detector) triggerCallback(event EventType, cb func(event EventType), done <-chan struct{}) {
+	if cb == nil || done == nil {
+		return
+	}
 
+	select {
+	case <-done:
+		return
+	default:
+	}
+
+	doneChan := make(chan struct{})
 	timeout := time.NewTimer(500 * time.Millisecond)
 	defer timeout.Stop()
 
-	cb := d.callback
-	go func(callback func(event EventType)) {
+	go func() {
+		defer close(doneChan)
+		defer func() {
+			if r := recover(); r != nil {
+				log.Errorf("panic in sleep callback: %v", r)
+			}
+		}()
 		log.Info("sleep detection event fired")
-		callback(event)
-		close(doneChan)
-	}(cb)
+		cb(event)
+	}()
 
 	select {
 	case <-doneChan:
-	case <-d.ctx.Done():
+	case <-done:
 	case <-timeout.C:
-		log.Warnf("sleep callback timed out")
+		log.Warn("sleep callback timed out")
 	}
 }
+
+// NewDetector initializes IOKit/CoreFoundation bindings and returns a Detector.
+func NewDetector() (*Detector, error) {
+	if err := initLibs(); err != nil {
+		return nil, err
+	}
+	return &Detector{}, nil
+}
+
+func initLibs() error {
+	libInitOnce.Do(func() {
+		iokit, err := purego.Dlopen("/System/Library/Frameworks/IOKit.framework/IOKit", purego.RTLD_NOW|purego.RTLD_GLOBAL)
+		if err != nil {
+			libInitErr = fmt.Errorf("dlopen IOKit: %w", err)
+			return
+		}
+		cfLib, err := purego.Dlopen("/System/Library/Frameworks/CoreFoundation.framework/CoreFoundation", purego.RTLD_NOW|purego.RTLD_GLOBAL)
+		if err != nil {
+			libInitErr = fmt.Errorf("dlopen CoreFoundation: %w", err)
+			return
+		}
+
+		purego.RegisterLibFunc(&ioKit.IORegisterForSystemPower, iokit, "IORegisterForSystemPower")
+		purego.RegisterLibFunc(&ioKit.IODeregisterForSystemPower, iokit, "IODeregisterForSystemPower")
+		purego.RegisterLibFunc(&ioKit.IOAllowPowerChange, iokit, "IOAllowPowerChange")
+		purego.RegisterLibFunc(&ioKit.IOServiceClose, iokit, "IOServiceClose")
+		purego.RegisterLibFunc(&ioKit.IONotificationPortGetRunLoopSource, iokit, "IONotificationPortGetRunLoopSource")
+		purego.RegisterLibFunc(&ioKit.IONotificationPortDestroy, iokit, "IONotificationPortDestroy")
+
+		purego.RegisterLibFunc(&cf.CFRunLoopGetCurrent, cfLib, "CFRunLoopGetCurrent")
+		purego.RegisterLibFunc(&cf.CFRunLoopRun, cfLib, "CFRunLoopRun")
+		purego.RegisterLibFunc(&cf.CFRunLoopStop, cfLib, "CFRunLoopStop")
+		purego.RegisterLibFunc(&cf.CFRunLoopAddSource, cfLib, "CFRunLoopAddSource")
+		purego.RegisterLibFunc(&cf.CFRunLoopRemoveSource, cfLib, "CFRunLoopRemoveSource")
+
+		modeAddr, err := purego.Dlsym(cfLib, "kCFRunLoopCommonModes")
+		if err != nil {
+			libInitErr = fmt.Errorf("dlsym kCFRunLoopCommonModes: %w", err)
+			return
+		}
+		// Launder the uintptr-to-pointer conversion through a Go variable so
+		// go vet's unsafeptr analyzer doesn't flag a system-library global.
+		cfCommonModes = **(**uintptr)(unsafe.Pointer(&modeAddr))
+
+		// NewCallback slots are a finite, non-reclaimable resource, so register
+		// a single thunk that dispatches to the current Detector set.
+		callbackThunk = purego.NewCallback(powerCallback)
+	})
+	return libInitErr
+}
+
+// powerCallback is the IOServiceInterestCallback trampoline, invoked on the
+// runloop thread. A Go panic crossing the purego boundary has undefined
+// behavior, so contain it here.
+func powerCallback(refcon, service, messageType, messageArgument uintptr) uintptr {
+	defer func() {
+		if r := recover(); r != nil {
+			log.Errorf("panic in sleep powerCallback: %v", r)
+		}
+	}()
+	switch messageType {
+	case kIOMessageCanSystemSleep:
+		// Not acknowledging forces a 30s IOKit timeout before idle sleep.
+		allowPowerChange(messageArgument)
+	case kIOMessageSystemWillSleep:
+		dispatchEvent(EventTypeSleep)
+		allowPowerChange(messageArgument)
+	case kIOMessageSystemHasPoweredOn:
+		dispatchEvent(EventTypeWakeUp)
+	}
+	return 0
+}
+
+func allowPowerChange(messageArgument uintptr) {
+	serviceRegistryMu.Lock()
+	var port uintptr
+	if session != nil {
+		port = session.rp
+	}
+	serviceRegistryMu.Unlock()
+	if port != 0 {
+		ioKit.IOAllowPowerChange(port, messageArgument)
+	}
+}
+
+func dispatchEvent(event EventType) {
+	serviceRegistryMu.Lock()
+	snaps := make([]detectorSnapshot, 0, len(serviceRegistry))
+	for d := range serviceRegistry {
+		snaps = append(snaps, detectorSnapshot{
+			detector: d,
+			callback: d.callback,
+			done:     d.done,
+		})
+	}
+	serviceRegistryMu.Unlock()
+
+	for _, s := range snaps {
+		s.detector.triggerCallback(event, s.callback, s.done)
+	}
+}
+
+// runRunLoop owns the OS-locked thread that CFRunLoop is pinned to. Setup
+// result is reported on errCh so Register can surface failures synchronously.
+func runRunLoop(errCh chan<- error) {
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+
+	sess, err := setupSession()
+	if err == nil {
+		serviceRegistryMu.Lock()
+		session = sess
+		serviceRegistryMu.Unlock()
+	}
+	errCh <- err
+	if err != nil {
+		return
+	}
+
+	defer func() {
+		if r := recover(); r != nil {
+			log.Errorf("panic in sleep runloop: %v", r)
+		}
+	}()
+	cf.CFRunLoopRun()
+}
+
+// setupSession performs the IOKit registration on the current thread. Panics
+// are converted to errors so runRunLoop never leaves errCh unsent.
+func setupSession() (s *runLoopSession, err error) {
+	defer func() {
+		if r := recover(); r != nil {
+			err = fmt.Errorf("panic during runloop setup: %v", r)
+		}
+	}()
+
+	var portRef, notifier uintptr
+	rp := ioKit.IORegisterForSystemPower(0, &portRef, callbackThunk, &notifier)
+	if rp == 0 {
+		return nil, fmt.Errorf("IORegisterForSystemPower returned zero")
+	}
+
+	rl := cf.CFRunLoopGetCurrent()
+	source := ioKit.IONotificationPortGetRunLoopSource(portRef)
+	cf.CFRunLoopAddSource(rl, source, cfCommonModes)
+
+	return &runLoopSession{rl: rl, port: portRef, notifier: notifier, rp: rp}, nil
+}