Merge remote-tracking branch 'origin/main' into netmap_progressive_alignment

[changing behavior!] This makes the debug bundle contain the SUPERSET of the map with nm updates and config updates

client/firewall/uspfilter/filter.go

@@ -121,7 +121,6 @@ type Manager struct {
 	udpTracker     *conntrack.UDPTracker
 	icmpTracker    *conntrack.ICMPTracker
 	tcpTracker     *conntrack.TCPTracker
-	fragments      *fragmentTracker
 	forwarder      atomic.Pointer[forwarder.Forwarder]
 	pendingCapture atomic.Pointer[forwarder.PacketCapture]
 	logger         *nblog.Logger
@@ -184,41 +183,6 @@ func (d *decoder) decodePacket(data []byte) error {
 	}
 }
 
-// decodeTransport decodes the transport header of a first fragment (which
-// gopacket leaves undecoded) into the decoder and appends its layer type to
-// decoded, so the ACL pipeline can evaluate it like a normal packet. It returns
-// false if the protocol is unsupported or the header is truncated.
-func (d *decoder) decodeTransport(proto layers.IPProtocol, payload []byte) bool {
-	var l4 gopacket.DecodingLayer
-	var layerType gopacket.LayerType
-	var minLen int
-	switch proto {
-	case layers.IPProtocolTCP:
-		l4, layerType, minLen = &d.tcp, layers.LayerTypeTCP, 20
-	case layers.IPProtocolUDP:
-		l4, layerType, minLen = &d.udp, layers.LayerTypeUDP, 8
-	case layers.IPProtocolICMPv4:
-		l4, layerType, minLen = &d.icmp4, layers.LayerTypeICMPv4, 8
-	case layers.IPProtocolICMPv6:
-		l4, layerType, minLen = &d.icmp6, layers.LayerTypeICMPv6, 8
-	default:
-		return false
-	}
-
-	// Reject a fragment too small to hold the full transport header before
-	// decoding: it can't be ACL-evaluated (tiny-fragment attack), and skipping
-	// the decode avoids gopacket allocating an error on the drop path.
-	if len(payload) < minLen {
-		return false
-	}
-
-	if err := l4.DecodeFromBytes(payload, gopacket.NilDecodeFeedback); err != nil {
-		return false
-	}
-	d.decoded = append(d.decoded, layerType)
-	return true
-}
-
 // Create userspace firewall manager constructor
 func Create(iface common.IFaceMapper, disableServerRoutes bool, flowLogger nftypes.FlowLogger, mtu uint16) (*Manager, error) {
 	return create(iface, nil, disableServerRoutes, flowLogger, mtu)
@@ -322,8 +286,6 @@ func create(iface common.IFaceMapper, nativeFirewall firewall.Manager, disableSe
 	if err := m.localipmanager.UpdateLocalIPs(iface); err != nil {
 		return nil, fmt.Errorf("update local IPs: %w", err)
 	}
-	m.fragments = newFragmentTracker(m.logger)
-
 	if disableConntrack {
 		log.Info("conntrack is disabled")
 	} else {
@@ -337,7 +299,6 @@ func create(iface common.IFaceMapper, nativeFirewall firewall.Manager, disableSe
 		}
 	}
 	if err := iface.SetFilter(m); err != nil {
-		m.fragments.Close()
 		return nil, fmt.Errorf("set filter: %w", err)
 	}
 	return m, nil
@@ -733,10 +694,6 @@ func (m *Manager) resetState() {
 		m.tcpTracker.Close()
 	}
 
-	if m.fragments != nil {
-		m.fragments.Close()
-	}
-
 	if fwder := m.forwarder.Load(); fwder != nil {
 		fwder.SetCapture(nil)
 		fwder.Stop()
@@ -1089,20 +1046,19 @@ func (m *Manager) filterInbound(packetData []byte, size int) bool {
 		return true
 	}
 
-	// gopacket does not decode the transport header of any IP fragment, so
-	// fragments take a dedicated path: the first fragment's header is decoded
-	// and ACL-evaluated here, and the remaining fragments inherit its verdict.
+	// TODO: pass fragments of routed packets to forwarder
 	if fragment {
-		return m.filterInboundFragment(d, srcIP, dstIP, size)
+		if m.logger.Enabled(nblog.LevelTrace) {
+			if d.decoded[0] == layers.LayerTypeIPv4 {
+				m.logger.Trace4("packet is a fragment: src=%v dst=%v id=%v flags=%v",
+					srcIP, dstIP, d.ip4.Id, d.ip4.Flags)
+			} else {
+				m.logger.Trace2("packet is an IPv6 fragment: src=%v dst=%v", srcIP, dstIP)
+			}
+		}
+		return false
 	}
 
-	return m.filterInboundDecoded(d, srcIP, dstIP, packetData, size)
-}
-
-// filterInboundDecoded runs the ACL, DNAT and conntrack pipeline on a fully
-// decoded (non-fragment) inbound packet. It returns true if the packet should
-// be dropped.
-func (m *Manager) filterInboundDecoded(d *decoder, srcIP, dstIP netip.Addr, packetData []byte, size int) bool {
 	// TODO: optimize port DNAT by caching matched rules in conntrack
 	if translated := m.translateInboundPortDNAT(packetData, d, srcIP, dstIP); translated {
 		// Re-decode after port DNAT translation to update port information
@@ -1133,226 +1089,33 @@ func (m *Manager) filterInboundDecoded(d *decoder, srcIP, dstIP netip.Addr, pack
 	return m.handleRoutedTraffic(d, srcIP, dstIP, packetData, size)
 }
 
-// fragmentMeta holds the reassembly identity and layout of an IP fragment,
-// extracted uniformly for IPv4 and IPv6.
-type fragmentMeta struct {
-	key fragmentKey
-	// offset is the fragment offset in 8-byte units (zero for the first
-	// fragment).
-	offset uint16
-	// moreFragments is the More Fragments bit. A first fragment with it unset is
-	// an IPv6 atomic fragment (a complete datagram, RFC 6946): it has no trailing
-	// fragments to inherit a verdict, so it must not be recorded.
-	moreFragments bool
-	proto         layers.IPProtocol
-	// l4payload is the fragmentable payload of this fragment. For the first
-	// fragment it starts with the transport header.
-	l4payload []byte
-	// headerEndOctets is the first fragment's payload length in 8-byte units:
-	// the smallest offset a trailing fragment may start at without overlapping
-	// the inspected transport header.
-	headerEndOctets uint16
-}
-
-// fragmentMetadata extracts the fragment identity and layout from a decoded IP
-// fragment. It returns false for fragments it can't interpret (e.g. an IPv6
-// fragment header shorter than 8 bytes), which are then dropped.
-func fragmentMetadata(d *decoder, srcIP, dstIP netip.Addr) (fragmentMeta, bool) {
-	switch d.decoded[0] {
-	case layers.LayerTypeIPv4:
-		payload := d.ip4.Payload
-		return fragmentMeta{
-			key:             fragmentKey{srcIP: srcIP, dstIP: dstIP, id: uint32(d.ip4.Id), proto: uint8(d.ip4.Protocol)},
-			offset:          d.ip4.FragOffset,
-			moreFragments:   d.ip4.Flags&layers.IPv4MoreFragments != 0,
-			proto:           d.ip4.Protocol,
-			l4payload:       payload,
-			headerEndOctets: octets(len(payload)),
-		}, true
-
-	case layers.LayerTypeIPv6:
-		// IPv6 fragment extension header: 8 bytes, followed by the fragmentable
-		// payload. Layout: next header (1), reserved (1), offset+flags (2), id (4).
-		payload := d.ip6.Payload
-		if len(payload) < 8 {
-			return fragmentMeta{}, false
-		}
-		nextHeader := layers.IPProtocol(payload[0])
-		offsetFlags := binary.BigEndian.Uint16(payload[2:4])
-		id := binary.BigEndian.Uint32(payload[4:8])
-		l4 := payload[8:]
-		return fragmentMeta{
-			key:             fragmentKey{srcIP: srcIP, dstIP: dstIP, id: id, proto: uint8(nextHeader)},
-			offset:          offsetFlags >> 3,
-			moreFragments:   offsetFlags&1 != 0,
-			proto:           nextHeader,
-			l4payload:       l4,
-			headerEndOctets: octets(len(l4)),
-		}, true
-
-	default:
-		return fragmentMeta{}, false
-	}
-}
-
-// octets rounds a byte length up to whole 8-byte units, the granularity of the
-// IP fragment offset field.
-func octets(nbytes int) uint16 {
-	return uint16((nbytes + 7) / 8)
-}
-
-// filterInboundFragment decides the fate of an IP fragment. gopacket stops
-// decoding at the network layer for every fragment, so the first fragment's
-// transport header is decoded and ACL-evaluated here and its verdict recorded;
-// the remaining (headerless) fragments inherit that verdict. Anything that
-// cannot be tied to an allowed, non-overlapping first fragment is dropped.
-func (m *Manager) filterInboundFragment(d *decoder, srcIP, dstIP netip.Addr, size int) bool {
-	meta, ok := fragmentMetadata(d, srcIP, dstIP)
-	if !ok {
-		if m.logger.Enabled(nblog.LevelTrace) {
-			m.logger.Trace2("dropping unsupported fragment: src=%v dst=%v", srcIP, dstIP)
-		}
-		return true
-	}
-
-	if meta.offset != 0 {
-		return m.filterTrailingFragment(meta, srcIP, dstIP)
-	}
-
-	// A new first fragment supersedes any recorded verdict for this datagram, so
-	// a re-sent or overlapping offset-zero fragment can't inherit the old one.
-	m.fragments.poison(meta.key)
-
-	// First fragment: decode its transport header so the ACL can evaluate it. A
-	// decode failure means the fragment is too small to hold the full transport
-	// header (RFC 1858 §3 tiny-fragment attack); it can't be evaluated, so drop it.
-	if !d.decodeTransport(meta.proto, meta.l4payload) {
-		if m.logger.Enabled(nblog.LevelTrace) {
-			m.logger.Trace3("dropping first fragment without full L4 header: src=%v dst=%v id=%v",
-				srcIP, dstIP, meta.key.id)
-		}
-		return true
-	}
-
-	return m.filterFirstFragment(d, meta, srcIP, dstIP, size)
-}
-
-// filterTrailingFragment applies a recorded first-fragment verdict to a
-// non-first fragment.
-func (m *Manager) filterTrailingFragment(meta fragmentMeta, srcIP, dstIP netip.Addr) bool {
-	switch m.fragments.verdict(meta.key, meta.offset) {
-	case fragmentAllow:
-		return false
-	case fragmentOverlap:
-		if m.logger.Enabled(nblog.LevelTrace) {
-			m.logger.Trace3("dropping overlapping fragment rewriting inspected header: src=%v dst=%v id=%v",
-				srcIP, dstIP, meta.key.id)
-		}
-		return true
-	default:
-		if m.logger.Enabled(nblog.LevelTrace) {
-			m.logger.Trace3("dropping fragment with no allowed first fragment: src=%v dst=%v id=%v",
-				srcIP, dstIP, meta.key.id)
-		}
-		return true
-	}
-}
-
-// filterFirstFragment runs the verdict part of the inbound pipeline on a first
-// fragment with its transport header decoded. It mirrors filterInboundDecoded
-// but skips DNAT (port rewriting on fragments is unsupported) and forwarder
-// injection (fragments are left to the stack to reassemble, not forwarded).
-// Allowed fragments have their verdict recorded so the datagram's trailing
-// fragments inherit it.
-func (m *Manager) filterFirstFragment(d *decoder, meta fragmentMeta, srcIP, dstIP netip.Addr, size int) bool {
-	if m.stateful && m.isValidTrackedConnection(d, srcIP, dstIP, size) {
-		m.recordFirstFragment(meta)
-		return false
-	}
-
-	if m.localipmanager.IsLocalIP(dstIP) {
-		ruleID, blocked := m.peerACLsBlock(srcIP, d, nil)
-		if blocked {
-			m.storeDropFlow("Dropping local first fragment (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
-				d, srcIP, dstIP, ruleID, size)
-			return true
-		}
-		m.trackInbound(d, srcIP, dstIP, ruleID, size)
-		m.recordFirstFragment(meta)
-		return false
-	}
-
-	if !m.routingEnabled.Load() {
-		if m.logger.Enabled(nblog.LevelTrace) {
-			m.logger.Trace2("Dropping routed fragment (routing disabled): src=%s dst=%s", srcIP, dstIP)
-		}
-		return true
-	}
-	if m.nativeRouter.Load() {
-		m.trackInbound(d, srcIP, dstIP, nil, size)
-		m.recordFirstFragment(meta)
-		return false
-	}
-
-	// TODO: pass fragments of routed packets to the forwarder; until then
-	// allowed routed fragments go to the native stack.
-	srcPort, dstPort := getPortsFromPacket(d)
-	ruleID, pass := m.routeACLsPass(srcIP, dstIP, d.decoded[1], srcPort, dstPort)
-	if !pass {
-		m.storeDropFlow("Dropping routed first fragment (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
-			d, srcIP, dstIP, ruleID, size)
-		return true
-	}
-
-	m.recordFirstFragment(meta)
-	return false
-}
-
-// recordFirstFragment caches an allowed first fragment's verdict for its
-// trailing fragments to inherit. Atomic fragments (no More Fragments bit) are
-// complete datagrams with no trailing fragments, so they are not cached and
-// cannot exhaust the verdict table.
-func (m *Manager) recordFirstFragment(meta fragmentMeta) {
-	if !meta.moreFragments {
-		return
-	}
-	m.fragments.recordAllowed(meta.key, meta.headerEndOctets)
-}
-
-// storeDropFlow logs and records a netflow drop event for an inbound packet
-// denied by the ACLs. msg is the trace format taking rule id, protocol, source
-// and destination.
-func (m *Manager) storeDropFlow(msg string, d *decoder, srcIP, dstIP netip.Addr, ruleID []byte, size int) {
-	pnum := getProtocolFromPacket(d)
-	srcPort, dstPort := getPortsFromPacket(d)
-
-	if m.logger.Enabled(nblog.LevelTrace) {
-		m.logger.Trace6(msg, ruleID, pnum, srcIP, srcPort, dstIP, dstPort)
-	}
-
-	m.flowLogger.StoreEvent(nftypes.EventFields{
-		FlowID:     uuid.New(),
-		Type:       nftypes.TypeDrop,
-		RuleID:     ruleID,
-		Direction:  nftypes.Ingress,
-		Protocol:   pnum,
-		SourceIP:   srcIP,
-		DestIP:     dstIP,
-		SourcePort: srcPort,
-		DestPort:   dstPort,
-		// TODO: icmp type/code
-		RxPackets: 1,
-		RxBytes:   uint64(size),
-	})
-}
-
 // handleLocalTraffic handles local traffic.
 // If it returns true, the packet should be dropped.
 func (m *Manager) handleLocalTraffic(d *decoder, srcIP, dstIP netip.Addr, packetData []byte, size int) bool {
 	ruleID, blocked := m.peerACLsBlock(srcIP, d, packetData)
 	if blocked {
-		m.storeDropFlow("Dropping local packet (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
-			d, srcIP, dstIP, ruleID, size)
+		pnum := getProtocolFromPacket(d)
+		srcPort, dstPort := getPortsFromPacket(d)
+
+		if m.logger.Enabled(nblog.LevelTrace) {
+			m.logger.Trace6("Dropping local packet (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
+				ruleID, pnum, srcIP, srcPort, dstIP, dstPort)
+		}
+
+		m.flowLogger.StoreEvent(nftypes.EventFields{
+			FlowID:     uuid.New(),
+			Type:       nftypes.TypeDrop,
+			RuleID:     ruleID,
+			Direction:  nftypes.Ingress,
+			Protocol:   pnum,
+			SourceIP:   srcIP,
+			DestIP:     dstIP,
+			SourcePort: srcPort,
+			DestPort:   dstPort,
+			// TODO: icmp type/code
+			RxPackets: 1,
+			RxBytes:   uint64(size),
+		})
 		return true
 	}
 
@@ -1405,8 +1168,27 @@ func (m *Manager) handleRoutedTraffic(d *decoder, srcIP, dstIP netip.Addr, packe
 
 	ruleID, pass := m.routeACLsPass(srcIP, dstIP, protoLayer, srcPort, dstPort)
 	if !pass {
-		m.storeDropFlow("Dropping routed packet (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
-			d, srcIP, dstIP, ruleID, size)
+		proto := getProtocolFromPacket(d)
+
+		if m.logger.Enabled(nblog.LevelTrace) {
+			m.logger.Trace6("Dropping routed packet (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
+				ruleID, proto, srcIP, srcPort, dstIP, dstPort)
+		}
+
+		m.flowLogger.StoreEvent(nftypes.EventFields{
+			FlowID:     uuid.New(),
+			Type:       nftypes.TypeDrop,
+			RuleID:     ruleID,
+			Direction:  nftypes.Ingress,
+			Protocol:   proto,
+			SourceIP:   srcIP,
+			DestIP:     dstIP,
+			SourcePort: srcPort,
+			DestPort:   dstPort,
+			// TODO: icmp type/code
+			RxPackets: 1,
+			RxBytes:   uint64(size),
+		})
 		return true
 	}
 

client/firewall/uspfilter/fragment.go

@@ -1,204 +0,0 @@
-package uspfilter
-
-import (
-	"context"
-	"net/netip"
-	"os"
-	"strconv"
-	"sync"
-	"time"
-
-	nblog "github.com/netbirdio/netbird/client/firewall/uspfilter/log"
-)
-
-const (
-	// defaultFragmentTimeout bounds how long a first-fragment verdict is kept
-	// while the remaining fragments arrive. It mirrors the Linux IP reassembly
-	// timeout (net.ipv4.ipfrag_time).
-	defaultFragmentTimeout = 30 * time.Second
-	// fragmentCleanupInterval is how often expired verdicts are purged.
-	fragmentCleanupInterval = 10 * time.Second
-	// defaultMaxFragmentEntries caps the number of concurrently tracked
-	// fragmented datagrams. The table stays bounded because each datagram is a
-	// single small entry regardless of how many fragments it is split into, and
-	// the 13-bit IPv4 fragment-offset field limits any datagram to 64 KiB.
-	defaultMaxFragmentEntries = 16384
-
-	// EnvFragmentMaxEntries overrides defaultMaxFragmentEntries.
-	EnvFragmentMaxEntries = "NB_FRAGMENT_MAX_ENTRIES"
-)
-
-// fragmentVerdict is the decision for a trailing (headerless) fragment.
-type fragmentVerdict int
-
-const (
-	// fragmentDeny drops the fragment: no allowed first fragment is on record.
-	fragmentDeny fragmentVerdict = iota
-	// fragmentAllow passes the fragment: it belongs to an allowed datagram and
-	// does not overlap the already-inspected transport header.
-	fragmentAllow
-	// fragmentOverlap drops the fragment and poisons its datagram: it overlaps
-	// the transport header the ACL inspected (RFC 1858 §4, RFC 3128; RFC 5722
-	// requires discarding the whole datagram on overlap for IPv6).
-	fragmentOverlap
-)
-
-// fragmentKey identifies a fragmented datagram. It matches the RFC 791 / RFC
-// 8200 reassembly key: source, destination, protocol and identification. The id
-// is 32-bit to hold both the IPv4 (16-bit) and IPv6 (32-bit) identification.
-type fragmentKey struct {
-	srcIP netip.Addr
-	dstIP netip.Addr
-	id    uint32
-	proto uint8
-}
-
-// fragmentEntry records the verdict of an allowed first fragment.
-type fragmentEntry struct {
-	// headerEndOctets is the offset, in 8-byte units, at which the first
-	// fragment's payload ended. A trailing fragment starting before this
-	// overlaps bytes the ACL already inspected and is rejected.
-	headerEndOctets uint16
-	// recordedAt is when the first fragment was accepted. The verdict expires a
-	// fixed timeout later and is not refreshed, mirroring the kernel reassembly
-	// timer so a trailing-fragment flood can't keep a datagram alive.
-	recordedAt time.Time
-}
-
-// fragmentTracker records the ACL verdict of a datagram's first fragment so the
-// remaining fragments, which carry no L4 header, can inherit the decision
-// without reassembling the datagram. Only allowed first fragments are stored;
-// anything that cannot be tied to an allowed, non-overlapping first fragment is
-// dropped (fail closed).
-type fragmentTracker struct {
-	logger  *nblog.Logger
-	mutex   sync.Mutex
-	entries map[fragmentKey]fragmentEntry
-	timeout time.Duration
-	// maxEntries caps the table; atCapacity dedups the capacity warning until
-	// the table drains below the cap again.
-	maxEntries    int
-	atCapacity    bool
-	cleanupTicker *time.Ticker
-	cancel        context.CancelFunc
-}
-
-func newFragmentTracker(logger *nblog.Logger) *fragmentTracker {
-	ctx, cancel := context.WithCancel(context.Background())
-	t := &fragmentTracker{
-		logger:        logger,
-		entries:       make(map[fragmentKey]fragmentEntry),
-		timeout:       defaultFragmentTimeout,
-		maxEntries:    fragmentMaxEntries(logger),
-		cleanupTicker: time.NewTicker(fragmentCleanupInterval),
-		cancel:        cancel,
-	}
-	go t.cleanupRoutine(ctx)
-	return t
-}
-
-func fragmentMaxEntries(logger *nblog.Logger) int {
-	v := os.Getenv(EnvFragmentMaxEntries)
-	if v == "" {
-		return defaultMaxFragmentEntries
-	}
-	n, err := strconv.Atoi(v)
-	if err != nil || n <= 0 {
-		logger.Warn2("invalid %s=%q, using default", EnvFragmentMaxEntries, v)
-		return defaultMaxFragmentEntries
-	}
-	return n
-}
-
-// recordAllowed stores the verdict of an allowed first fragment. headerEndOctets
-// is the first fragment's payload length in 8-byte units. When the table is full
-// the record is dropped, which fails closed: the datagram's trailing fragments
-// will be denied.
-func (t *fragmentTracker) recordAllowed(key fragmentKey, headerEndOctets uint16) {
-	t.mutex.Lock()
-	defer t.mutex.Unlock()
-
-	if t.entries == nil {
-		return
-	}
-	if _, ok := t.entries[key]; !ok && len(t.entries) >= t.maxEntries {
-		if !t.atCapacity {
-			t.atCapacity = true
-			t.logger.Warn2("fragment verdict table at capacity (%d/%d): trailing fragments of new datagrams will be dropped",
-				len(t.entries), t.maxEntries)
-		}
-		return
-	}
-	t.entries[key] = fragmentEntry{
-		headerEndOctets: headerEndOctets,
-		recordedAt:      time.Now(),
-	}
-}
-
-// poison drops any recorded verdict for a datagram, so its later fragments are
-// denied until a new allowed first fragment is recorded. Called on every
-// offset-zero fragment to defeat offset-zero overlap rewrites (RFC 3128).
-func (t *fragmentTracker) poison(key fragmentKey) {
-	t.mutex.Lock()
-	defer t.mutex.Unlock()
-	delete(t.entries, key)
-}
-
-// verdict decides the fate of a trailing fragment at fragOffsetOctets (the IPv4
-// fragment offset, in 8-byte units). A fragment overlapping the inspected
-// header poisons the datagram: the entry is removed so all further fragments of
-// that datagram are denied too.
-func (t *fragmentTracker) verdict(key fragmentKey, fragOffsetOctets uint16) fragmentVerdict {
-	t.mutex.Lock()
-	defer t.mutex.Unlock()
-
-	entry, ok := t.entries[key]
-	if !ok {
-		return fragmentDeny
-	}
-	if time.Since(entry.recordedAt) > t.timeout {
-		delete(t.entries, key)
-		return fragmentDeny
-	}
-	if fragOffsetOctets < entry.headerEndOctets {
-		delete(t.entries, key)
-		return fragmentOverlap
-	}
-	return fragmentAllow
-}
-
-func (t *fragmentTracker) cleanupRoutine(ctx context.Context) {
-	defer t.cleanupTicker.Stop()
-	for {
-		select {
-		case <-t.cleanupTicker.C:
-			t.cleanup()
-		case <-ctx.Done():
-			return
-		}
-	}
-}
-
-func (t *fragmentTracker) cleanup() {
-	t.mutex.Lock()
-	defer t.mutex.Unlock()
-
-	for key, entry := range t.entries {
-		if time.Since(entry.recordedAt) > t.timeout {
-			delete(t.entries, key)
-		}
-	}
-
-	if len(t.entries) < t.maxEntries {
-		t.atCapacity = false
-	}
-}
-
-// Close stops the cleanup routine and releases resources.
-func (t *fragmentTracker) Close() {
-	t.cancel()
-
-	t.mutex.Lock()
-	t.entries = nil
-	t.mutex.Unlock()
-}

client/firewall/uspfilter/fragment_bench_test.go

@@ -1,115 +0,0 @@
-package uspfilter
-
-import (
-	"encoding/binary"
-	"testing"
-)
-
-// benchFilterInbound drives filterInbound over a fixed packet in a tight loop.
-// Packets are built once, outside the timed region, so the benchmark measures
-// only pipeline cost, which is what an attacker can amplify.
-func benchFilterInbound(b *testing.B, pkt []byte) {
-	b.Helper()
-	b.ReportAllocs()
-	b.SetBytes(int64(len(pkt)))
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		m := benchManager
-		m.filterInbound(pkt, len(pkt))
-	}
-}
-
-// benchManager is a package-level manager reused across fragment benchmarks so
-// setup cost stays out of the timed region.
-var benchManager *Manager
-
-func setupBenchManager(b *testing.B) *Manager {
-	b.Helper()
-	m := newFragmentTestManager(b)
-	allowUDP(b, m, 8080)
-	// Disable conntrack so the allowed-first-fragment path measures transport
-	// decode + ACL every iteration instead of matching the connection tracked
-	// on the first iteration.
-	m.stateful = false
-	benchManager = m
-	return m
-}
-
-// BenchmarkInbound_NormalPacket is the baseline: a full, non-fragmented UDP
-// packet that passes the ACL. Fragment paths should stay comparable to this.
-func BenchmarkInbound_NormalPacket(b *testing.B) {
-	setupBenchManager(b)
-	pkt := normalUDPPacket(b, 8080, 32)
-	benchFilterInbound(b, pkt)
-}
-
-// BenchmarkInbound_FirstFragmentAllowed measures the first-fragment path:
-// transport decode + ACL evaluation + verdict record.
-func BenchmarkInbound_FirstFragmentAllowed(b *testing.B) {
-	setupBenchManager(b)
-	pkt := firstFragmentUDP(b, 0x2000, 8080, 32)
-	benchFilterInbound(b, pkt)
-}
-
-// BenchmarkInbound_TrailingFragmentAllowed measures the common trailing-fragment
-// path: a single map lookup after the first fragment is on record.
-func BenchmarkInbound_TrailingFragmentAllowed(b *testing.B) {
-	m := setupBenchManager(b)
-	first := firstFragmentUDP(b, 0x3000, 8080, 32)
-	m.filterInbound(first, len(first))
-	pkt := trailingFragment(b, 0x3000, 5, false, 24)
-	benchFilterInbound(b, pkt)
-}
-
-// BenchmarkInbound_TrailingFragmentNoFirst is the primary DoS vector: an
-// attacker floods trailing fragments with no first fragment on record. Each is
-// a map miss and must be cheap.
-func BenchmarkInbound_TrailingFragmentNoFirst(b *testing.B) {
-	setupBenchManager(b)
-	pkt := trailingFragment(b, 0x4000, 185, false, 40)
-	benchFilterInbound(b, pkt)
-}
-
-// BenchmarkInbound_TinyFirstFragment measures the tiny-fragment drop path: a
-// first fragment too small to decode a transport header.
-func BenchmarkInbound_TinyFirstFragment(b *testing.B) {
-	setupBenchManager(b)
-	pkt := trailingFragment(b, 0x5000, 0, true, 4)
-	benchFilterInbound(b, pkt)
-}
-
-// BenchmarkInbound_TrailingFragmentDistinctIDs is the worst case for the
-// verdict table: an attacker varies the datagram id on every packet so no first
-// fragment ever matches. Verdict lookups always miss and nothing is recorded,
-// so the table cannot grow. Each iteration rewrites the id field in place.
-func BenchmarkInbound_TrailingFragmentDistinctIDs(b *testing.B) {
-	setupBenchManager(b)
-	pkt := trailingFragment(b, 0x6000, 185, false, 40)
-	m := benchManager
-
-	b.ReportAllocs()
-	b.SetBytes(int64(len(pkt)))
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		// IPv4 identification field is at bytes 4:6.
-		binary.BigEndian.PutUint16(pkt[4:6], uint16(i))
-		m.filterInbound(pkt, len(pkt))
-	}
-}
-
-// BenchmarkInbound_FirstFragmentDistinctIDs measures sustained first-fragment
-// pressure with distinct ids: transport decode + ACL + verdict insert until the
-// table caps, exercising the map growth and capacity guard.
-func BenchmarkInbound_FirstFragmentDistinctIDs(b *testing.B) {
-	setupBenchManager(b)
-	pkt := firstFragmentUDP(b, 0x7000, 8080, 32)
-	m := benchManager
-
-	b.ReportAllocs()
-	b.SetBytes(int64(len(pkt)))
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		binary.BigEndian.PutUint16(pkt[4:6], uint16(i))
-		m.filterInbound(pkt, len(pkt))
-	}
-}

client/firewall/uspfilter/fragment_test.go

@@ -1,554 +0,0 @@
-package uspfilter
-
-import (
-	"encoding/binary"
-	"net"
-	"net/netip"
-	"testing"
-	"time"
-
-	"github.com/google/gopacket"
-	"github.com/google/gopacket/layers"
-	"github.com/stretchr/testify/require"
-
-	fw "github.com/netbirdio/netbird/client/firewall/manager"
-	nbiface "github.com/netbirdio/netbird/client/iface"
-	"github.com/netbirdio/netbird/client/iface/device"
-	"github.com/netbirdio/netbird/client/iface/wgaddr"
-)
-
-const (
-	fragTestSrc   = "100.10.0.1"
-	fragTestDst   = "100.10.0.100"
-	fragTestSrcV6 = "fd00::1"
-	fragTestDstV6 = "fd00::100"
-)
-
-func newFragmentTestManager(tb testing.TB) *Manager {
-	tb.Helper()
-
-	ifaceMock := &IFaceMock{
-		SetFilterFunc: func(device.PacketFilter) error { return nil },
-		AddressFunc: func() wgaddr.Address {
-			return wgaddr.Address{
-				IP:      netip.MustParseAddr(fragTestDst),
-				Network: netip.MustParsePrefix("100.10.0.0/16"),
-				IPv6:    netip.MustParseAddr(fragTestDstV6),
-				IPv6Net: netip.MustParsePrefix("fd00::/64"),
-			}
-		},
-	}
-
-	m, err := Create(ifaceMock, false, flowLogger, nbiface.DefaultMTU)
-	require.NoError(tb, err)
-	require.NoError(tb, m.UpdateLocalIPs())
-	tb.Cleanup(func() { require.NoError(tb, m.Close(nil)) })
-	return m
-}
-
-// firstFragmentUDPTo builds the first fragment of a fragmented UDP datagram to
-// the given destination: it carries the full UDP header plus payloadLen bytes
-// of data, with the More Fragments flag set and offset zero.
-func firstFragmentUDPTo(tb testing.TB, dst string, id uint16, dstPort uint16, payloadLen int) []byte {
-	tb.Helper()
-
-	ip := &layers.IPv4{
-		Version:  4,
-		TTL:      64,
-		Id:       id,
-		Protocol: layers.IPProtocolUDP,
-		SrcIP:    net.ParseIP(fragTestSrc),
-		DstIP:    net.ParseIP(dst),
-		Flags:    layers.IPv4MoreFragments,
-	}
-	udp := &layers.UDP{SrcPort: 40000, DstPort: layers.UDPPort(dstPort)}
-	require.NoError(tb, udp.SetNetworkLayerForChecksum(ip))
-
-	buf := gopacket.NewSerializeBuffer()
-	opts := gopacket.SerializeOptions{ComputeChecksums: true, FixLengths: true}
-	require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, udp, gopacket.Payload(make([]byte, payloadLen))))
-	return buf.Bytes()
-}
-
-func firstFragmentUDP(tb testing.TB, id uint16, dstPort uint16, payloadLen int) []byte {
-	tb.Helper()
-	return firstFragmentUDPTo(tb, fragTestDst, id, dstPort, payloadLen)
-}
-
-// firstFragmentTCP builds the first fragment of a fragmented TCP datagram: the
-// full 20-byte TCP header plus 12 bytes of data, with the More Fragments flag
-// set and offset zero.
-func firstFragmentTCP(tb testing.TB, id uint16, dstPort uint16) []byte {
-	tb.Helper()
-
-	ip := &layers.IPv4{
-		Version:  4,
-		TTL:      64,
-		Id:       id,
-		Protocol: layers.IPProtocolTCP,
-		SrcIP:    net.ParseIP(fragTestSrc),
-		DstIP:    net.ParseIP(fragTestDst),
-		Flags:    layers.IPv4MoreFragments,
-	}
-	tcp := &layers.TCP{SrcPort: 40000, DstPort: layers.TCPPort(dstPort), SYN: true, Window: 64240}
-	require.NoError(tb, tcp.SetNetworkLayerForChecksum(ip))
-
-	buf := gopacket.NewSerializeBuffer()
-	opts := gopacket.SerializeOptions{ComputeChecksums: true, FixLengths: true}
-	require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, tcp, gopacket.Payload(make([]byte, 12))))
-	return buf.Bytes()
-}
-
-// trailingFragmentTo builds a non-first fragment to the given destination: an
-// IPv4 header at the given fragment offset (in 8-byte units) carrying raw
-// payload and no L4 header.
-func trailingFragmentTo(tb testing.TB, dst string, proto layers.IPProtocol, id uint16, fragOffsetOctets uint16, moreFragments bool, payloadLen int) []byte {
-	tb.Helper()
-
-	ip := &layers.IPv4{
-		Version:    4,
-		TTL:        64,
-		Id:         id,
-		Protocol:   proto,
-		SrcIP:      net.ParseIP(fragTestSrc),
-		DstIP:      net.ParseIP(dst),
-		FragOffset: fragOffsetOctets,
-	}
-	if moreFragments {
-		ip.Flags = layers.IPv4MoreFragments
-	}
-
-	buf := gopacket.NewSerializeBuffer()
-	opts := gopacket.SerializeOptions{FixLengths: true}
-	require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, gopacket.Payload(make([]byte, payloadLen))))
-	return buf.Bytes()
-}
-
-func trailingFragment(tb testing.TB, id uint16, fragOffsetOctets uint16, moreFragments bool, payloadLen int) []byte {
-	tb.Helper()
-	return trailingFragmentTo(tb, fragTestDst, layers.IPProtocolUDP, id, fragOffsetOctets, moreFragments, payloadLen)
-}
-
-// outboundUDPPacket builds a complete outbound UDP packet from the local
-// address, used to establish conntrack state for reply-direction tests.
-func outboundUDPPacket(tb testing.TB, srcPort, dstPort uint16) []byte {
-	tb.Helper()
-
-	ip := &layers.IPv4{
-		Version:  4,
-		TTL:      64,
-		Id:       1,
-		Protocol: layers.IPProtocolUDP,
-		SrcIP:    net.ParseIP(fragTestDst),
-		DstIP:    net.ParseIP(fragTestSrc),
-	}
-	udp := &layers.UDP{SrcPort: layers.UDPPort(srcPort), DstPort: layers.UDPPort(dstPort)}
-	require.NoError(tb, udp.SetNetworkLayerForChecksum(ip))
-
-	buf := gopacket.NewSerializeBuffer()
-	opts := gopacket.SerializeOptions{ComputeChecksums: true, FixLengths: true}
-	require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, udp, gopacket.Payload(make([]byte, 16))))
-	return buf.Bytes()
-}
-
-// normalUDPPacket builds a complete, non-fragmented UDP packet for baseline
-// comparisons against the fragment paths.
-func normalUDPPacket(tb testing.TB, dstPort uint16, payloadLen int) []byte {
-	tb.Helper()
-
-	ip := &layers.IPv4{
-		Version:  4,
-		TTL:      64,
-		Id:       1,
-		Protocol: layers.IPProtocolUDP,
-		SrcIP:    net.ParseIP(fragTestSrc),
-		DstIP:    net.ParseIP(fragTestDst),
-	}
-	udp := &layers.UDP{SrcPort: 40000, DstPort: layers.UDPPort(dstPort)}
-	require.NoError(tb, udp.SetNetworkLayerForChecksum(ip))
-
-	buf := gopacket.NewSerializeBuffer()
-	opts := gopacket.SerializeOptions{ComputeChecksums: true, FixLengths: true}
-	require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, udp, gopacket.Payload(make([]byte, payloadLen))))
-	return buf.Bytes()
-}
-
-func allowUDP(tb testing.TB, m *Manager, dstPort uint16) {
-	tb.Helper()
-	_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrc), fw.ProtocolUDP, nil,
-		&fw.Port{Values: []uint16{dstPort}}, fw.ActionAccept, "")
-	require.NoError(tb, err)
-}
-
-// TestFragment_TrailingWithoutFirstDropped is the core bypass repro: a trailing
-// fragment with no allowed first fragment on record must be dropped. Before the
-// fix, filterInbound returned false (allow) for any fragment.
-func TestFragment_TrailingWithoutFirstDropped(t *testing.T) {
-	m := newFragmentTestManager(t)
-
-	frag := trailingFragment(t, 0x1234, 185, false, 40)
-	require.True(t, m.filterInbound(frag, len(frag)),
-		"trailing fragment without an allowed first fragment must be dropped")
-}
-
-// TestFragment_AllowedFirstPassesTrailing verifies that once a first fragment
-// passes the ACL, its trailing fragments inherit the allow verdict.
-func TestFragment_AllowedFirstPassesTrailing(t *testing.T) {
-	m := newFragmentTestManager(t)
-	allowUDP(t, m, 8080)
-
-	// First fragment: UDP header (8) + 32 payload = 40 octets -> headerEnd = 5.
-	first := firstFragmentUDP(t, 0x2222, 8080, 32)
-	require.False(t, m.filterInbound(first, len(first)),
-		"allowed first fragment should pass and be recorded")
-
-	trailing := trailingFragment(t, 0x2222, 5, false, 24)
-	require.False(t, m.filterInbound(trailing, len(trailing)),
-		"trailing fragment of an allowed datagram should pass")
-}
-
-// TestFragment_DeniedFirstDropsTrailing verifies that a first fragment blocked
-// by the ACL leaves no verdict, so its trailing fragments are dropped.
-func TestFragment_DeniedFirstDropsTrailing(t *testing.T) {
-	m := newFragmentTestManager(t)
-	// No accept rule: local traffic defaults to deny.
-
-	first := firstFragmentUDP(t, 0x3333, 9999, 32)
-	require.True(t, m.filterInbound(first, len(first)),
-		"first fragment to a blocked port should be dropped by the ACL")
-
-	trailing := trailingFragment(t, 0x3333, 5, false, 24)
-	require.True(t, m.filterInbound(trailing, len(trailing)),
-		"trailing fragment of a denied datagram must be dropped")
-}
-
-// TestFragment_OverlappingHeaderDropped covers the RFC 1858 §4 / RFC 3128
-// overlapping-fragment rewrite: a trailing fragment starting inside the range
-// the ACL already inspected is dropped and poisons the datagram. TCP is used so
-// the overlap lands on real header bytes (the flags at byte 13).
-func TestFragment_OverlappingHeaderDropped(t *testing.T) {
-	m := newFragmentTestManager(t)
-	_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrc), fw.ProtocolTCP, nil,
-		&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
-	require.NoError(t, err)
-
-	// First fragment: TCP header (20) + 12 data = 32 bytes -> headerEnd = 4 octets.
-	first := firstFragmentTCP(t, 0x4444, 8080)
-	require.False(t, m.filterInbound(first, len(first)))
-
-	// Overlapping fragment at offset 1 (byte 8) falls inside the inspected TCP
-	// header, so it could rewrite the flags or port on reassembly.
-	overlap := trailingFragmentTo(t, fragTestDst, layers.IPProtocolTCP, 0x4444, 1, true, 32)
-	require.True(t, m.filterInbound(overlap, len(overlap)),
-		"fragment overlapping the inspected header must be dropped")
-
-	// The datagram is now poisoned: a later, non-overlapping fragment is also
-	// dropped because the verdict was removed.
-	later := trailingFragmentTo(t, fragTestDst, layers.IPProtocolTCP, 0x4444, 4, false, 24)
-	require.True(t, m.filterInbound(later, len(later)),
-		"fragments after an overlap must be dropped (datagram poisoned)")
-}
-
-// TestFragment_OffsetZeroOverlapPoisons covers the RFC 3128 offset-zero rewrite:
-// an allowed first fragment followed by a denied offset-zero fragment for the
-// same datagram must not leave the earlier allow verdict in place.
-func TestFragment_OffsetZeroOverlapPoisons(t *testing.T) {
-	m := newFragmentTestManager(t)
-	allowUDP(t, m, 8080)
-
-	allowed := firstFragmentUDP(t, 0x5A5A, 8080, 32)
-	require.False(t, m.filterInbound(allowed, len(allowed)),
-		"allowed first fragment should pass and be recorded")
-
-	// A second offset-zero fragment to a denied port supersedes the datagram's
-	// verdict; it is dropped and must not leave the allow in place.
-	denied := firstFragmentUDP(t, 0x5A5A, 9999, 32)
-	require.True(t, m.filterInbound(denied, len(denied)),
-		"denied offset-zero fragment must be dropped")
-
-	trailing := trailingFragment(t, 0x5A5A, 5, false, 24)
-	require.True(t, m.filterInbound(trailing, len(trailing)),
-		"trailing fragment must be denied after the datagram was poisoned")
-}
-
-// TestFragment_TinyFirstDropped covers the tiny-fragment attack: a first
-// fragment too small to contain the full transport header can't be
-// ACL-evaluated and must be dropped.
-func TestFragment_TinyFirstDropped(t *testing.T) {
-	m := newFragmentTestManager(t)
-	allowUDP(t, m, 8080)
-
-	// IPv4 header + 4 raw bytes, MF set, offset 0: too small for the 8-byte UDP
-	// header, so it decodes to L3 only.
-	tiny := trailingFragment(t, 0x5555, 0, true, 4)
-	require.True(t, m.filterInbound(tiny, len(tiny)),
-		"tiny first fragment without a full L4 header must be dropped")
-}
-
-// TestFragment_TCPFirstFragment verifies the TCP arm of the transport decode: a
-// first fragment carrying the full 20-byte TCP header is ACL-evaluated and its
-// trailing fragments inherit the verdict.
-func TestFragment_TCPFirstFragment(t *testing.T) {
-	m := newFragmentTestManager(t)
-	_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrc), fw.ProtocolTCP, nil,
-		&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
-	require.NoError(t, err)
-
-	// TCP header (20) + 12 data = 32 bytes -> headerEnd = 4 octets.
-	first := firstFragmentTCP(t, 0x6666, 8080)
-	require.False(t, m.filterInbound(first, len(first)),
-		"allowed TCP first fragment should pass and be recorded")
-
-	trailing := trailingFragmentTo(t, fragTestDst, layers.IPProtocolTCP, 0x6666, 4, false, 24)
-	require.False(t, m.filterInbound(trailing, len(trailing)),
-		"trailing fragment of an allowed TCP datagram should pass")
-}
-
-// TestFragment_TCPTinyFirstDropped verifies the TCP minimum header length: 12
-// bytes would satisfy a UDP header but falls short of the 20-byte TCP header.
-func TestFragment_TCPTinyFirstDropped(t *testing.T) {
-	m := newFragmentTestManager(t)
-	_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrc), fw.ProtocolTCP, nil,
-		&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
-	require.NoError(t, err)
-
-	tiny := trailingFragmentTo(t, fragTestDst, layers.IPProtocolTCP, 0x7777, 0, true, 12)
-	require.True(t, m.filterInbound(tiny, len(tiny)),
-		"first fragment shorter than the TCP header must be dropped")
-}
-
-// TestFragment_ConntrackAllowsFirstFragment verifies the conntrack branch: reply
-// fragments of an outbound-established UDP flow pass without any inbound rule.
-func TestFragment_ConntrackAllowsFirstFragment(t *testing.T) {
-	m := newFragmentTestManager(t)
-
-	out := outboundUDPPacket(t, 12345, 40000)
-	require.False(t, m.filterOutbound(out, len(out)))
-
-	first := firstFragmentUDP(t, 0x8888, 12345, 32)
-	require.False(t, m.filterInbound(first, len(first)),
-		"reply first fragment should pass via conntrack")
-
-	trailing := trailingFragment(t, 0x8888, 5, false, 24)
-	require.False(t, m.filterInbound(trailing, len(trailing)),
-		"trailing fragment of a tracked flow should pass")
-}
-
-// TestFragment_RoutingDisabledDropsFragment verifies routed first fragments are
-// dropped when routing is disabled.
-func TestFragment_RoutingDisabledDropsFragment(t *testing.T) {
-	m := newFragmentTestManager(t)
-	m.routingEnabled.Store(false)
-
-	first := firstFragmentUDPTo(t, "198.51.100.10", 0x9999, 8080, 32)
-	require.True(t, m.filterInbound(first, len(first)),
-		"routed first fragment must be dropped when routing is disabled")
-}
-
-// TestFragment_RouteACL verifies the route-ACL branch: fragments to a non-local
-// destination follow the route rules, allowed datagrams pass their trailing
-// fragments and denied ones don't.
-func TestFragment_RouteACL(t *testing.T) {
-	m := newFragmentTestManager(t)
-	m.routingEnabled.Store(true)
-	m.nativeRouter.Store(false)
-
-	_, err := m.AddRouteFiltering(
-		[]byte("rt-1"),
-		[]netip.Prefix{netip.MustParsePrefix("100.10.0.0/16")},
-		fw.Network{Prefix: netip.MustParsePrefix("198.51.100.0/24")},
-		fw.ProtocolUDP,
-		nil,
-		&fw.Port{Values: []uint16{8080}},
-		fw.ActionAccept,
-	)
-	require.NoError(t, err)
-
-	first := firstFragmentUDPTo(t, "198.51.100.10", 0xAAAA, 8080, 32)
-	require.False(t, m.filterInbound(first, len(first)),
-		"route-ACL-allowed first fragment should pass")
-	trailing := trailingFragmentTo(t, "198.51.100.10", layers.IPProtocolUDP, 0xAAAA, 5, false, 24)
-	require.False(t, m.filterInbound(trailing, len(trailing)),
-		"trailing fragment of an allowed routed datagram should pass")
-
-	denied := firstFragmentUDPTo(t, "198.51.100.10", 0xBBBB, 9999, 32)
-	require.True(t, m.filterInbound(denied, len(denied)),
-		"route-ACL-denied first fragment must be dropped")
-	deniedTrailing := trailingFragmentTo(t, "198.51.100.10", layers.IPProtocolUDP, 0xBBBB, 5, false, 24)
-	require.True(t, m.filterInbound(deniedTrailing, len(deniedTrailing)),
-		"trailing fragment of a denied routed datagram must be dropped")
-}
-
-// TestFragment_ExpiredVerdictDropsTrailing verifies a verdict older than the
-// tracker timeout no longer admits trailing fragments.
-func TestFragment_ExpiredVerdictDropsTrailing(t *testing.T) {
-	m := newFragmentTestManager(t)
-	allowUDP(t, m, 8080)
-
-	first := firstFragmentUDP(t, 0xCCCC, 8080, 32)
-	require.False(t, m.filterInbound(first, len(first)))
-
-	m.fragments.mutex.Lock()
-	for key, entry := range m.fragments.entries {
-		entry.recordedAt = time.Now().Add(-defaultFragmentTimeout - time.Second)
-		m.fragments.entries[key] = entry
-	}
-	m.fragments.mutex.Unlock()
-
-	trailing := trailingFragment(t, 0xCCCC, 5, false, 24)
-	require.True(t, m.filterInbound(trailing, len(trailing)),
-		"trailing fragment after verdict expiry must be dropped")
-}
-
-// TestFragment_CapacityFailsClosed verifies the table cap: at capacity, new
-// datagram verdicts are not recorded (their trailing fragments are dropped)
-// while already-recorded datagrams keep working.
-func TestFragment_CapacityFailsClosed(t *testing.T) {
-	m := newFragmentTestManager(t)
-	allowUDP(t, m, 8080)
-
-	m.fragments.mutex.Lock()
-	m.fragments.maxEntries = 1
-	m.fragments.mutex.Unlock()
-
-	first1 := firstFragmentUDP(t, 0x0101, 8080, 32)
-	require.False(t, m.filterInbound(first1, len(first1)))
-
-	first2 := firstFragmentUDP(t, 0x0202, 8080, 32)
-	require.False(t, m.filterInbound(first2, len(first2)),
-		"first fragment itself still passes at capacity")
-
-	trailing2 := trailingFragment(t, 0x0202, 5, false, 24)
-	require.True(t, m.filterInbound(trailing2, len(trailing2)),
-		"trailing fragment of an unrecorded datagram must be dropped at capacity")
-
-	trailing1 := trailingFragment(t, 0x0101, 5, false, 24)
-	require.False(t, m.filterInbound(trailing1, len(trailing1)),
-		"already-recorded datagram should keep passing at capacity")
-}
-
-// v6FragmentHeader builds the 8-byte IPv6 fragment extension header for the
-// given inner protocol, offset (8-byte units), More Fragments bit and id.
-func v6FragmentHeader(proto layers.IPProtocol, offsetOctets uint16, moreFragments bool, id uint32) []byte {
-	offsetFlags := offsetOctets << 3
-	if moreFragments {
-		offsetFlags |= 1
-	}
-	hdr := make([]byte, 8)
-	hdr[0] = uint8(proto)
-	binary.BigEndian.PutUint16(hdr[2:4], offsetFlags)
-	binary.BigEndian.PutUint32(hdr[4:8], id)
-	return hdr
-}
-
-func v6UDPHeader(dstPort uint16, dataLen int) []byte {
-	hdr := make([]byte, 8)
-	binary.BigEndian.PutUint16(hdr[0:2], 40000)
-	binary.BigEndian.PutUint16(hdr[2:4], dstPort)
-	binary.BigEndian.PutUint16(hdr[4:6], uint16(8+dataLen))
-	return hdr
-}
-
-// firstFragmentUDPv6 builds the first fragment of a fragmented IPv6 UDP
-// datagram: fragment header (offset 0, More Fragments set) + full UDP header +
-// data.
-func firstFragmentUDPv6(tb testing.TB, id uint32, dstPort uint16, dataLen int) []byte {
-	tb.Helper()
-	return fragmentUDPv6(tb, id, dstPort, dataLen, true)
-}
-
-// fragmentUDPv6 builds an offset-zero IPv6 UDP fragment. With moreFragments
-// false it is an atomic fragment (a complete datagram, RFC 6946).
-func fragmentUDPv6(tb testing.TB, id uint32, dstPort uint16, dataLen int, moreFragments bool) []byte {
-	tb.Helper()
-
-	ip := &layers.IPv6{
-		Version:    6,
-		NextHeader: layers.IPProtocolIPv6Fragment,
-		HopLimit:   64,
-		SrcIP:      net.ParseIP(fragTestSrcV6),
-		DstIP:      net.ParseIP(fragTestDstV6),
-	}
-	payload := append(v6FragmentHeader(layers.IPProtocolUDP, 0, moreFragments, id), v6UDPHeader(dstPort, dataLen)...)
-	payload = append(payload, make([]byte, dataLen)...)
-
-	buf := gopacket.NewSerializeBuffer()
-	require.NoError(tb, gopacket.SerializeLayers(buf, gopacket.SerializeOptions{FixLengths: true}, ip, gopacket.Payload(payload)))
-	return buf.Bytes()
-}
-
-// trailingFragmentV6 builds a non-first IPv6 fragment: fragment header at the
-// given offset carrying raw data and no transport header.
-func trailingFragmentV6(tb testing.TB, id uint32, offsetOctets uint16, moreFragments bool, dataLen int) []byte {
-	tb.Helper()
-
-	ip := &layers.IPv6{
-		Version:    6,
-		NextHeader: layers.IPProtocolIPv6Fragment,
-		HopLimit:   64,
-		SrcIP:      net.ParseIP(fragTestSrcV6),
-		DstIP:      net.ParseIP(fragTestDstV6),
-	}
-	payload := append(v6FragmentHeader(layers.IPProtocolUDP, offsetOctets, moreFragments, id), make([]byte, dataLen)...)
-
-	buf := gopacket.NewSerializeBuffer()
-	require.NoError(tb, gopacket.SerializeLayers(buf, gopacket.SerializeOptions{FixLengths: true}, ip, gopacket.Payload(payload)))
-	return buf.Bytes()
-}
-
-// TestFragmentV6_TrailingWithoutFirstDropped verifies the IPv6 bypass is closed:
-// a trailing fragment with no allowed first fragment is dropped.
-func TestFragmentV6_TrailingWithoutFirstDropped(t *testing.T) {
-	m := newFragmentTestManager(t)
-
-	frag := trailingFragmentV6(t, 0xAABBCCDD, 100, false, 40)
-	require.True(t, m.filterInbound(frag, len(frag)),
-		"IPv6 trailing fragment without an allowed first fragment must be dropped")
-}
-
-// TestFragmentV6_AllowedFirstPassesTrailing verifies IPv6 fragments are
-// evaluated like IPv4: an allowed first fragment lets its trailing fragments
-// through.
-func TestFragmentV6_AllowedFirstPassesTrailing(t *testing.T) {
-	m := newFragmentTestManager(t)
-	_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrcV6), fw.ProtocolUDP, nil,
-		&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
-	require.NoError(t, err)
-
-	// First fragment: UDP header (8) + 32 data = 40 octets -> headerEnd = 5.
-	first := firstFragmentUDPv6(t, 0xAABBCCDD, 8080, 32)
-	require.False(t, m.filterInbound(first, len(first)),
-		"allowed IPv6 first fragment should pass and be recorded")
-
-	trailing := trailingFragmentV6(t, 0xAABBCCDD, 5, false, 24)
-	require.False(t, m.filterInbound(trailing, len(trailing)),
-		"trailing fragment of an allowed IPv6 datagram should pass")
-}
-
-// TestFragmentV6_AtomicNotCached verifies an IPv6 atomic fragment (fragment
-// header with offset 0 and no More Fragments, a complete datagram per RFC 6946)
-// is evaluated but not recorded, so a flood of allowed atomic fragments can't
-// exhaust the verdict table.
-func TestFragmentV6_AtomicNotCached(t *testing.T) {
-	m := newFragmentTestManager(t)
-	_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrcV6), fw.ProtocolUDP, nil,
-		&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
-	require.NoError(t, err)
-
-	atomic := fragmentUDPv6(t, 0xA70301C, 8080, 16, false)
-	require.False(t, m.filterInbound(atomic, len(atomic)),
-		"allowed IPv6 atomic fragment should pass")
-
-	m.fragments.mutex.Lock()
-	n := len(m.fragments.entries)
-	m.fragments.mutex.Unlock()
-	require.Zero(t, n, "atomic fragment must not create a verdict entry")
-
-	// A genuine fragmented datagram (More Fragments set) is still recorded.
-	first := fragmentUDPv6(t, 0xBEEF, 8080, 32, true)
-	require.False(t, m.filterInbound(first, len(first)))
-	m.fragments.mutex.Lock()
-	n = len(m.fragments.entries)
-	m.fragments.mutex.Unlock()
-	require.Equal(t, 1, n, "genuine first fragment must record a verdict")
-}

client/internal/engine.go

@@ -220,6 +220,12 @@ type Engine struct {
 	// networkSerial is the latest CurrentSerial (state ID) of the network sent by the Management service
 	networkSerial uint64
 
+	// forwardingRules holds the ingress forward rules applied for the current target.
+	// Wholesale sections (incl. forward rules) run only on the first pass of a target;
+	// it is stashed here so the final, peer-converged pass can build the lazy-connection
+	// exclude list without recomputing them on every bounded peer pass.
+	forwardingRules []firewallManager.ForwardRule
+
 	networkMonitor *networkmonitor.NetworkMonitor
 
 	sshServer sshServer
@@ -522,7 +528,7 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, mgmtURL *url.URL)
 		} else {
 			log.Infof("running rosenpass in strict mode")
 		}
-		e.rpManager, err = rosenpass.NewManager(e.config.PreSharedKey, e.config.WgIfaceName, publicKey)
+		e.rpManager, err = rosenpass.NewManager(e.config.PreSharedKey, e.config.WgIfaceName)
 		if err != nil {
 			return fmt.Errorf("create rosenpass manager: %w", err)
 		}
@@ -774,7 +780,15 @@ func (e *Engine) blockLanAccess() {
 
 // modifyPeers updates peers that have been modified (e.g. IP address has been changed).
 // It closes the existing connection, removes it from the peerConns map, and creates a new one.
-func (e *Engine) modifyPeers(peersUpdate []*mgmProto.RemotePeerConfig) error {
+// maxPeersPerSyncPass is the default per-pass cap on how many peers each of
+// removePeers/modifyPeers/addNewPeers applies, so syncMsgMux is held only for a
+// batch at a time and other subsystems can interleave between passes. It is
+// passed in (not read globally) so tests can exercise the multi-pass path.
+const maxPeersPerSyncPass = 300
+
+// modifyPeers re-applies up to maxBatch changed peers per call. It returns true
+// when more changed peers remained than the cap, so the caller re-runs.
+func (e *Engine) modifyPeers(peersUpdate []*mgmProto.RemotePeerConfig, maxBatch int) (bool, error) {
 
 	// first, check if peers have been modified
 	var modified []*mgmProto.RemotePeerConfig
@@ -804,26 +818,32 @@ func (e *Engine) modifyPeers(peersUpdate []*mgmProto.RemotePeerConfig) error {
 		}
 	}
 
+	more := false
+	if len(modified) > maxBatch {
+		modified = modified[:maxBatch]
+		more = true
+	}
+
 	// second, close all modified connections and remove them from the state map
 	for _, p := range modified {
-		err := e.removePeer(p.GetWgPubKey())
-		if err != nil {
-			return err
+		if err := e.removePeer(p.GetWgPubKey()); err != nil {
+			return false, err
 		}
 	}
 	// third, add the peer connections again
 	for _, p := range modified {
-		err := e.addNewPeer(p)
-		if err != nil {
-			return err
+		if err := e.addNewPeer(p); err != nil {
+			return false, err
 		}
 	}
-	return nil
+	return more, nil
 }
 
 // removePeers finds and removes peers that do not exist anymore in the network map received from the Management Service.
 // It also removes peers that have been modified (e.g. change of IP address). They will be added again in addPeers method.
-func (e *Engine) removePeers(peersUpdate []*mgmProto.RemotePeerConfig) error {
+// removePeers removes up to maxBatch peers per call. It returns true when more
+// peers remained to remove than the cap, so the caller re-runs.
+func (e *Engine) removePeers(peersUpdate []*mgmProto.RemotePeerConfig, maxBatch int) (bool, error) {
 	newPeers := make([]string, 0, len(peersUpdate))
 	for _, p := range peersUpdate {
 		newPeers = append(newPeers, p.GetWgPubKey())
@@ -831,14 +851,19 @@ func (e *Engine) removePeers(peersUpdate []*mgmProto.RemotePeerConfig) error {
 
 	toRemove := util.SliceDiff(e.peerStore.PeersPubKey(), newPeers)
 
+	more := false
+	if len(toRemove) > maxBatch {
+		toRemove = toRemove[:maxBatch]
+		more = true
+	}
+
 	for _, p := range toRemove {
-		err := e.removePeer(p)
-		if err != nil {
-			return err
+		if err := e.removePeer(p); err != nil {
+			return false, err
 		}
 		log.Infof("removed peer %s", p)
 	}
-	return nil
+	return more, nil
 }
 
 func (e *Engine) removeAllPeers() error {
@@ -917,19 +942,17 @@ func (e *Engine) phase(name string) func() {
 	}
 }
 
-func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
-	started := time.Now()
-	defer func() {
-		duration := time.Since(started)
-		log.Infof("sync finished in %s", duration)
-		e.clientMetrics.RecordSyncDuration(e.ctx, duration)
-	}()
+// applySyncPass applies one bounded pass of the sync update under syncMsgMux and
+// returns true if more peers remained than the per-pass cap. It is driven by the
+// mapStateManager, which re-invokes it (releasing the lock between passes) until
+// the update is fully applied.
+func (e *Engine) applySyncPass(update *mgmProto.SyncResponse, firstPass bool) (bool, error) {
 	e.syncMsgMux.Lock()
 	defer e.syncMsgMux.Unlock()
 
 	// Check context INSIDE lock to ensure atomicity with shutdown
 	if e.ctx.Err() != nil {
-		return e.ctx.Err()
+		return false, e.ctx.Err()
 	}
 
 	if update.NetworkMap != nil && update.NetworkMap.PeerConfig != nil {
@@ -940,7 +963,7 @@ func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
 	err := e.updateNetbirdConfig(update.GetNetbirdConfig())
 	done()
 	if err != nil {
-		return err
+		return false, err
 	}
 
 	// Posture checks are bound to the network map presence:
@@ -950,28 +973,25 @@ func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
 	//                                        leave the previously applied checks untouched
 	nm := update.GetNetworkMap()
 	if nm == nil {
-		return nil
+		return false, nil
 	}
 
 	done = e.phase("checks")
 	err = e.updateChecksIfNew(update.Checks)
 	done()
 	if err != nil {
-		return err
+		return false, err
 	}
 
-	done = e.phase("persist")
-	e.persistSyncResponse(update)
-	done()
-
 	// only apply new changes and ignore old ones
-	if err := e.updateNetworkMap(nm); err != nil {
-		return err
+	more, err := e.updateNetworkMap(nm, maxPeersPerSyncPass, firstPass)
+	if err != nil {
+		return false, err
 	}
 
 	e.statusRecorder.PublishEvent(cProto.SystemEvent_INFO, cProto.SystemEvent_SYSTEM, "Network map updated", "", nil)
 
-	return nil
+	return more, nil
 }
 
 // updateNetbirdConfig applies the management-provided NetBird configuration:
@@ -1019,6 +1039,13 @@ func (e *Engine) updateNetbirdConfig(wCfg *mgmProto.NetbirdConfig) error {
 // (not syncMsgMux) is held for the whole Set so the store cannot be cleared (disabled /
 // engine close) mid-call and have this write resurrect a file that was just removed.
 func (e *Engine) persistSyncResponse(update *mgmProto.SyncResponse) {
+	// Only persist updates that carry a network map. Config-only updates (e.g. relay
+	// token rotation, STUN/TURN) have a nil NetworkMap; persisting them would overwrite
+	// the last full map on disk and break restore-on-restart.
+	if update.GetNetworkMap() == nil {
+		return
+	}
+
 	e.syncRespMux.RLock()
 	defer e.syncRespMux.RUnlock()
 
@@ -1306,7 +1333,24 @@ func (e *Engine) receiveManagementEvents() {
 		}
 		e.applyInfoFlags(info)
 
-		err := e.mgmClient.Sync(e.ctx, info, e.handleSync)
+		// The map-state manager converges the latest update in the background in
+		// bounded passes; the stream callback only hands it the newest target.
+		persist := func(u *mgmProto.SyncResponse) {
+			done := e.phase("persist")
+			e.persistSyncResponse(u)
+			done()
+		}
+		manager := newMapStateManager(e.applySyncPass, persist, func(d time.Duration) {
+			log.Infof("sync finished in %s", d)
+			e.clientMetrics.RecordSyncDuration(e.ctx, d)
+		})
+		e.shutdownWg.Add(1)
+		go func() {
+			defer e.shutdownWg.Done()
+			manager.run(e.ctx)
+		}()
+
+		err := e.mgmClient.Sync(e.ctx, info, manager.SetTarget)
 		if err != nil {
 			// happens if management is unavailable for a long time.
 			// We want to cancel the operation of the whole client
@@ -1357,21 +1401,107 @@ func (e *Engine) updateTURNs(turns []*mgmProto.ProtectedHostConfig) error {
 	return nil
 }
 
-func (e *Engine) updateNetworkMap(networkMap *mgmProto.NetworkMap) error {
+// updateNetworkMap applies the wholesale parts (config, routes, ACL, DNS) in full
+// and up to maxBatch peers per phase. It returns true when more peers remained
+// than the cap, so the caller re-runs until convergence.
+func (e *Engine) updateNetworkMap(networkMap *mgmProto.NetworkMap, maxBatch int, firstPass bool) (bool, error) {
 	// intentionally leave it before checking serial because for now it can happen that peer IP changed but serial didn't
 	if networkMap.GetPeerConfig() != nil {
 		err := e.updateConfig(networkMap.GetPeerConfig())
 		if err != nil {
-			return err
+			return false, err
 		}
 	}
 
 	serial := networkMap.GetSerial()
 	if e.networkSerial > serial {
 		log.Debugf("received outdated NetworkMap with serial %d, ignoring", serial)
-		return nil
+		return false, nil
 	}
 
+	// Wholesale sections (firewall/ACL, DNS, routes, forward rules) are applied
+	// up-front and only once per target: they are cheap, local, idempotent and must
+	// be in place before peers come up (fail-closed). On the bounded re-runs that only
+	// drain the remaining peer batches they are skipped — the applied forward rules are
+	// reused from e.forwardingRules for the lazy-exclude finalize.
+	if firstPass {
+		e.applyWholesale(networkMap, serial)
+	}
+
+	log.Debugf("got peers update from Management Service, total peers to connect to = %d", len(networkMap.GetRemotePeers()))
+
+	doneOffline := e.phase("offline_peers")
+	e.updateOfflinePeers(networkMap.GetOfflinePeers())
+	doneOffline()
+
+	// Filter out own peer from the remote peers list
+	localPubKey := e.config.WgPrivateKey.PublicKey().String()
+	remotePeers := make([]*mgmProto.RemotePeerConfig, 0, len(networkMap.GetRemotePeers()))
+	for _, p := range networkMap.GetRemotePeers() {
+		if p.GetWgPubKey() != localPubKey {
+			remotePeers = append(remotePeers, p)
+		}
+	}
+
+	// No special case for cleanup: when management signals RemotePeersIsEmpty (e.g. our
+	// peer was deleted), remotePeers is already empty, so the bounded diff below removes
+	// every peer in batches — same path as a normal update, no unbounded removeAllPeers
+	// held under syncMsgMux in one shot.
+	doneRemoved := e.phase("removed_peers")
+	removeMore, err := e.removePeers(remotePeers, maxBatch)
+	doneRemoved()
+	if err != nil {
+		return false, err
+	}
+
+	doneModified := e.phase("modified_peers")
+	modifyMore, err := e.modifyPeers(remotePeers, maxBatch)
+	doneModified()
+	if err != nil {
+		return false, err
+	}
+
+	doneAdded := e.phase("added_peers")
+	addMore, err := e.addNewPeers(remotePeers, maxBatch)
+	doneAdded()
+	if err != nil {
+		return false, err
+	}
+
+	// needMore signals the caller to re-run when a peer phase hit its per-pass cap.
+	needMore := removeMore || modifyMore || addMore
+
+	e.statusRecorder.FinishPeerListModifications()
+
+	e.updatePeerSSHHostKeys(remotePeers)
+
+	if err := e.updateSSHClientConfig(remotePeers); err != nil {
+		log.Warnf("failed to update SSH client config: %v", err)
+	}
+
+	e.updateSSHServerAuth(networkMap.GetSshAuth())
+
+	// Set the exclude list only once peers have fully converged (this pass added
+	// the last batch). It needs all target peers present in the store, and
+	// ExcludePeer has replace-semantics — a partial set mid-convergence would be wrong.
+	if !needMore {
+		doneLazy := e.phase("lazy_exclude")
+		excludedLazyPeers := e.toExcludedLazyPeers(e.forwardingRules, remotePeers)
+		e.connMgr.SetExcludeList(e.ctx, excludedLazyPeers)
+		doneLazy()
+	}
+
+	e.networkSerial = serial
+
+	return needMore, nil
+}
+
+// applyWholesale applies the cheap, local, idempotent map sections — lazy feature
+// flag, firewall/legacy management, DNS, routes, ACL filtering, DNS forwarder and
+// ingress forward rules — that must be in place before peers come up. It runs once
+// per target (first pass only); the resulting forward rules are stashed in
+// e.forwardingRules for the lazy-exclude finalize on the peer-converged pass.
+func (e *Engine) applyWholesale(networkMap *mgmProto.NetworkMap, serial uint64) {
 	if err := e.connMgr.UpdatedRemoteFeatureFlag(e.ctx, networkMap.GetPeerConfig().GetLazyConnectionEnabled()); err != nil {
 		log.Errorf("failed to update lazy connection feature flag: %v", err)
 	}
@@ -1444,84 +1574,7 @@ func (e *Engine) updateNetworkMap(networkMap *mgmProto.NetworkMap) error {
 		log.Errorf("failed to update forward rules, err: %v", err)
 	}
 	done()
-
-	log.Debugf("got peers update from Management Service, total peers to connect to = %d", len(networkMap.GetRemotePeers()))
-
-	done = e.phase("offline_peers")
-	e.updateOfflinePeers(networkMap.GetOfflinePeers())
-	done()
-
-	remotePeers, err := e.reconcilePeers(networkMap)
-	if err != nil {
-		return err
-	}
-
-	// must set the exclude list after the peers are added. Without it the manager can not figure out the peers parameters from the store
-	done = e.phase("lazy_exclude")
-	excludedLazyPeers := e.toExcludedLazyPeers(forwardingRules, remotePeers)
-	e.connMgr.SetExcludeList(e.ctx, excludedLazyPeers)
-	done()
-
-	e.networkSerial = serial
-
-	return nil
-}
-
-// reconcilePeers applies the remote peer list from the network map (removing,
-// modifying and adding peers, then updating SSH config) and returns the remote
-// peers with our own peer filtered out, for use by later sync steps.
-func (e *Engine) reconcilePeers(networkMap *mgmProto.NetworkMap) ([]*mgmProto.RemotePeerConfig, error) {
-	// Filter out own peer from the remote peers list
-	localPubKey := e.config.WgPrivateKey.PublicKey().String()
-	remotePeers := make([]*mgmProto.RemotePeerConfig, 0, len(networkMap.GetRemotePeers()))
-	for _, p := range networkMap.GetRemotePeers() {
-		if p.GetWgPubKey() != localPubKey {
-			remotePeers = append(remotePeers, p)
-		}
-	}
-
-	// cleanup request, most likely our peer has been deleted
-	if networkMap.GetRemotePeersIsEmpty() {
-		err := e.removeAllPeers()
-		e.statusRecorder.FinishPeerListModifications()
-		if err != nil {
-			return nil, err
-		}
-		return remotePeers, nil
-	}
-
-	done := e.phase("removed_peers")
-	err := e.removePeers(remotePeers)
-	done()
-	if err != nil {
-		return nil, err
-	}
-
-	done = e.phase("modified_peers")
-	err = e.modifyPeers(remotePeers)
-	done()
-	if err != nil {
-		return nil, err
-	}
-
-	done = e.phase("added_peers")
-	err = e.addNewPeers(remotePeers)
-	done()
-	if err != nil {
-		return nil, err
-	}
-
-	e.statusRecorder.FinishPeerListModifications()
-
-	e.updatePeerSSHHostKeys(remotePeers)
-
-	if err := e.updateSSHClientConfig(remotePeers); err != nil {
-		log.Warnf("failed to update SSH client config: %v", err)
-	}
-
-	e.updateSSHServerAuth(networkMap.GetSshAuth())
-
-	return remotePeers, nil
+	e.forwardingRules = forwardingRules
 }
 
 func toDNSFeatureFlag(networkMap *mgmProto.NetworkMap) bool {
@@ -1701,14 +1754,23 @@ func addrToString(addr netip.Addr) string {
 }
 
 // addNewPeers adds peers that were not know before but arrived from the Management service with the update
-func (e *Engine) addNewPeers(peersUpdate []*mgmProto.RemotePeerConfig) error {
+// addNewPeers adds up to maxBatch not-yet-present peers per call. It returns true
+// when more new peers remained than the cap, so the caller re-runs.
+func (e *Engine) addNewPeers(peersUpdate []*mgmProto.RemotePeerConfig, maxBatch int) (bool, error) {
+	added := 0
 	for _, p := range peersUpdate {
-		err := e.addNewPeer(p)
-		if err != nil {
-			return err
+		if _, ok := e.peerStore.PeerConn(p.GetWgPubKey()); ok {
+			continue // already present (cheap skip), does not count toward the cap
 		}
+		if added >= maxBatch {
+			return true, nil // at least one more new peer remains
+		}
+		if err := e.addNewPeer(p); err != nil {
+			return false, err
+		}
+		added++
 	}
-	return nil
+	return false, nil
 }
 
 // addNewPeer add peer if connection doesn't exist

client/internal/engine_privileged_test.go

@@ -124,7 +124,7 @@ func TestEngine_SSH(t *testing.T) {
 		RemotePeersIsEmpty: false,
 	}
 
-	err = engine.updateNetworkMap(networkMap)
+	_, err = engine.updateNetworkMap(networkMap, maxPeersPerSyncPass, true)
 	require.NoError(t, err)
 
 	assert.Nil(t, engine.sshServer)
@@ -146,7 +146,7 @@ func TestEngine_SSH(t *testing.T) {
 		RemotePeersIsEmpty: false,
 	}
 
-	err = engine.updateNetworkMap(networkMap)
+	_, err = engine.updateNetworkMap(networkMap, maxPeersPerSyncPass, true)
 	require.NoError(t, err)
 
 	time.Sleep(250 * time.Millisecond)
@@ -159,7 +159,7 @@ func TestEngine_SSH(t *testing.T) {
 		RemotePeersIsEmpty: false,
 	}
 
-	err = engine.updateNetworkMap(networkMap)
+	_, err = engine.updateNetworkMap(networkMap, maxPeersPerSyncPass, true)
 	require.NoError(t, err)
 
 	// time.Sleep(250 * time.Millisecond)
@@ -174,7 +174,7 @@ func TestEngine_SSH(t *testing.T) {
 		RemotePeersIsEmpty: false,
 	}
 
-	err = engine.updateNetworkMap(networkMap)
+	_, err = engine.updateNetworkMap(networkMap, maxPeersPerSyncPass, true)
 	require.NoError(t, err)
 
 	assert.Nil(t, engine.sshServer)

client/internal/engine_test.go

@@ -437,7 +437,7 @@ func TestEngine_UpdateNetworkMap(t *testing.T) {
 
 	for _, c := range []testCase{case1, case2, case3, case4, case5, case6} {
 		t.Run(c.name, func(t *testing.T) {
-			err = engine.updateNetworkMap(c.networkMap)
+			_, err = engine.updateNetworkMap(c.networkMap, maxPeersPerSyncPass, true)
 			if err != nil {
 				t.Fatal(err)
 				return
@@ -464,6 +464,47 @@ func TestEngine_UpdateNetworkMap(t *testing.T) {
 			}
 		})
 	}
+
+	// chunked apply: with a per-pass cap smaller than the number of peers, a
+	// single updateNetworkMap applies one batch and reports more==true; the
+	// caller re-runs until convergence. (engine currently holds 0 peers.)
+	t.Run("chunked add converges over multiple passes", func(t *testing.T) {
+		nm := &mgmtProto.NetworkMap{
+			Serial:      6,
+			RemotePeers: []*mgmtProto.RemotePeerConfig{peer1, peer2, peer3},
+		}
+
+		more, err := engine.updateNetworkMap(nm, 1, true)
+		require.NoError(t, err)
+		require.True(t, more, "pass 1 should signal more")
+		require.Len(t, engine.peerStore.PeersPubKey(), 1)
+
+		more, err = engine.updateNetworkMap(nm, 1, false)
+		require.NoError(t, err)
+		require.True(t, more, "pass 2 should signal more")
+		require.Len(t, engine.peerStore.PeersPubKey(), 2)
+
+		more, err = engine.updateNetworkMap(nm, 1, false)
+		require.NoError(t, err)
+		require.False(t, more, "pass 3 should converge")
+		require.Len(t, engine.peerStore.PeersPubKey(), 3)
+	})
+
+	t.Run("chunked remove converges over multiple passes", func(t *testing.T) {
+		nm := &mgmtProto.NetworkMap{
+			Serial:      7,
+			RemotePeers: []*mgmtProto.RemotePeerConfig{peer1}, // remove peer2, peer3
+		}
+
+		more, err := engine.updateNetworkMap(nm, 1, true)
+		require.NoError(t, err)
+		require.True(t, more, "pass 1 should signal more (2 to remove, cap 1)")
+
+		more, err = engine.updateNetworkMap(nm, 1, false)
+		require.NoError(t, err)
+		require.False(t, more, "pass 2 should converge")
+		require.Len(t, engine.peerStore.PeersPubKey(), 1)
+	})
 }
 
 func TestEngine_UpdateNetworkMapWithRoutes(t *testing.T) {
@@ -634,7 +675,7 @@ func TestEngine_UpdateNetworkMapWithRoutes(t *testing.T) {
 				}
 			}()
 
-			err = engine.updateNetworkMap(testCase.networkMap)
+			_, err = engine.updateNetworkMap(testCase.networkMap, maxPeersPerSyncPass, true)
 			assert.NoError(t, err, "shouldn't return error")
 			assert.Equal(t, testCase.expectedSerial, input.inputSerial, "serial should match")
 			assert.Len(t, input.clientRoutes, testCase.expectedLen, "clientRoutes len should match")
@@ -838,7 +879,7 @@ func TestEngine_UpdateNetworkMapWithDNSUpdate(t *testing.T) {
 				}
 			}()
 
-			err = engine.updateNetworkMap(testCase.networkMap)
+			_, err = engine.updateNetworkMap(testCase.networkMap, maxPeersPerSyncPass, true)
 			assert.NoError(t, err, "shouldn't return error")
 			assert.Equal(t, testCase.expectedSerial, input.inputSerial, "serial should match")
 			assert.Len(t, input.inputNSGroups, testCase.expectedZonesLen, "zones len should match")

client/internal/mapsync.go

@@ -0,0 +1,214 @@
+package internal
+
+import (
+	"context"
+	"sync"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// mapStateManager is the single read/write point between the management stream
+// (writes) and the convergence loop (reads/applies).
+//
+// The stream calls SetTarget with the latest full SyncResponse — the complete
+// desired state. A single background goroutine (run) applies it to the engine in
+// bounded passes via apply() until converged, releasing syncMsgMux between passes
+// so other subsystems interleave. If a newer update arrives mid-flight, the loop
+// coalesces: it keeps converging toward the latest target and the intermediate one
+// is SKIPPED — never applied on its own (logged, no onConverged).
+//
+// Convergence is a single comparison: appliedGen == targetGen. targetGen
+// increments on every SetTarget (an internal generation counter, so it also covers
+// config-only updates that carry no network-map serial).
+//
+// onConverged fires once for each — and only each — map that is actually processed
+// (i.e. converged as the target). Skipped/superseded maps and dropped-on-error maps
+// do NOT fire it. So "sync finished in X" / RecordSyncDuration always corresponds
+// to a real, completed alignment.
+type mapStateManager struct {
+	// apply performs one bounded apply pass and reports whether more passes are needed.
+	// firstPass is true on the first pass of a given target, so the caller can run
+	// wholesale (firewall/routes/DNS/forward-rules) once per target and skip it on the
+	// re-runs that only drain the bounded peer batches. The manager owns this signal
+	// because it owns the convergence boundary; the engine need not track serials for it.
+	apply func(update *mgmProto.SyncResponse, firstPass bool) (bool, error)
+	// onConverged is called once per processed map, with the elapsed time since that
+	// map was received (for the sync-duration metric / "sync finished" log).
+	onConverged func(time.Duration)
+	// persist snapshots an update to disk for restore-on-restart. Called once per
+	// update received from management (in SetTarget), including ones later coalesced
+	// or skipped from apply, so the on-disk state mirrors what management last sent.
+	// The impl skips config-only updates (nil NetworkMap). May be nil.
+	persist func(*mgmProto.SyncResponse)
+
+	mu          sync.Mutex
+	target      *mgmProto.SyncResponse
+	targetGen   uint64
+	appliedGen  uint64
+	targetSetAt time.Time
+
+	wake chan struct{}
+}
+
+func newMapStateManager(apply func(update *mgmProto.SyncResponse, firstPass bool) (bool, error), persist func(*mgmProto.SyncResponse), onConverged func(time.Duration)) *mapStateManager {
+	return &mapStateManager{
+		apply:       apply,
+		persist:     persist,
+		onConverged: onConverged,
+		wake:        make(chan struct{}, 1),
+	}
+}
+
+// SetTarget records the latest update as the desired state and wakes the loop.
+// It returns immediately; convergence happens in the background. Serial-based
+// staleness of the network map is still enforced inside apply (updateNetworkMap).
+func (m *mapStateManager) SetTarget(update *mgmProto.SyncResponse) error {
+	m.mu.Lock()
+	// A target that has not settled yet (targetGen > appliedGen) is being superseded
+	// before it converged: we coalesce to the latest map and never apply this one on
+	// its own. It is SKIPPED — logged here, and it will not fire onConverged.
+	if m.target != nil && m.targetGen > m.appliedGen {
+		log.Debugf("sync map (gen %d) superseded before convergence, skipping", m.targetGen)
+	}
+	m.target = m.mergeTarget(m.target, update)
+	// Bump an internal generation counter, NOT the map serial: config-only updates
+	// (relay token rotation, STUN/TURN) arrive with NetworkMap == nil and carry no
+	// serial, yet must still be applied. Every SetTarget is therefore a distinct
+	// target regardless of payload. Map-serial staleness is enforced separately
+	// inside apply (updateNetworkMap).
+	m.targetGen++
+	m.targetSetAt = time.Now()
+	m.mu.Unlock()
+
+	select {
+	case m.wake <- struct{}{}:
+	default:
+	}
+
+	// Persist every update received from management — once per update (not per apply
+	// pass), and including ones that get coalesced/skipped from apply, so the on-disk
+	// state always reflects the latest map management sent. Done after waking the loop
+	// so convergence can start in parallel with the disk write. The persist impl skips
+	// config-only updates (nil NetworkMap).
+	if m.persist != nil {
+		m.persist(update)
+	}
+	return nil
+}
+
+// mergeTarget combines the currently pending target with a freshly received update
+// and returns the new desired state. It is called under m.mu from SetTarget and is
+// the single seam where the replace-vs-squash decision lives.
+//
+// Today management always sends a FULL map (the complete desired state), so the
+// update simply replaces whatever was pending — prev is ignored. When management
+// starts sending incremental/delta updates, squash `update` onto `prev` here; the
+// rest of the manager (generation tracking, convergence, signaling) is unaffected
+// because it already treats target as "the complete desired state, whatever it is".
+func (m *mapStateManager) mergeTarget(prev, update *mgmProto.SyncResponse) *mgmProto.SyncResponse {
+	// Nothing pending to preserve (no prev, or prev already fully applied): plain replace.
+	if prev == nil || update == nil || m.targetGen == m.appliedGen {
+		return update
+	}
+
+	// prev still has unapplied state (targetGen > appliedGen). In the sync protocol a
+	// nil component means "no change", so if `update` omits a component that prev
+	// carried, carry prev's forward — otherwise coalescing an update that superseded a
+	// not-yet-applied one would silently drop the map or config it uniquely brought.
+	// A present component in `update` is newer and wins. Management may send map-only
+	// updates (nil config) and config-only updates (nil map); both are handled here.
+	// A nil component in `update` means "no change", so fill it in from prev — otherwise
+	// coalescing an update that superseded a not-yet-applied one would drop the map or
+	// config it uniquely carried. A present component in `update` is newer and wins.
+	// We mutate `update` in place: it is a fresh per-message allocation from the sync
+	// stream (see receiveUpdatesEvents — not reused), and persisting this squashed target
+	// is correct, since it is the current full (superset) desired state.
+	if update.GetNetworkMap() == nil && prev.GetNetworkMap() != nil {
+		update.NetworkMap = prev.GetNetworkMap()
+		update.Checks = prev.Checks // checks travel with the map
+	}
+	if update.GetNetbirdConfig() == nil && prev.GetNetbirdConfig() != nil {
+		update.NetbirdConfig = prev.GetNetbirdConfig()
+	}
+	return update
+}
+
+// run drives convergence until ctx is done. It is meant to run in its own goroutine.
+func (m *mapStateManager) run(ctx context.Context) {
+	// passGen is the generation of the most recent apply() call (0 = none). A pass is
+	// the first for its target when its generation differs from the previous one —
+	// true on a fresh target and on a coalesced switch to a newer target mid-flight.
+	var passGen uint64
+	for {
+		m.mu.Lock()
+		target, tg, ag := m.target, m.targetGen, m.appliedGen
+		m.mu.Unlock()
+
+		// Fully converged (or nothing yet): block until a new target arrives.
+		if target == nil || ag == tg {
+			select {
+			case <-ctx.Done():
+				return
+			case <-m.wake:
+				continue
+			}
+		}
+
+		firstPass := tg != passGen
+		passGen = tg
+		more, err := m.apply(target, firstPass)
+		if err != nil {
+			if ctx.Err() != nil {
+				return
+			}
+			// Log and DROP this target — do not retry it. A deterministic failure
+			// (e.g. a malformed peer in the map) would otherwise spin every pass
+			// making no progress. Management is the source of truth and re-delivers
+			// the full map on the next sync, so dropping is safe; peers already
+			// applied this convergence stay (idempotent diffs) and the remainder is
+			// reconciled by the next target. Mirrors the legacy handleSync path,
+			// where the apply error was logged by the gRPC client and the update
+			// dropped. No onConverged: this target did not converge.
+			log.Errorf("apply sync pass, dropping update: %v", err)
+			m.settle(tg, false)
+			continue
+		}
+
+		if more {
+			// keep converging the current target; syncMsgMux was released by apply
+			// between passes so other subsystems interleave.
+			continue
+		}
+
+		// This pass converged. Mark applied and signal this one map.
+		m.settle(tg, true)
+		// if a newer target arrived mid-pass, settle is a no-op (targetGen != tg) and
+		// ag<tg next iteration -> apply it; this generation was skipped (logged in
+		// SetTarget) and is not signaled.
+	}
+}
+
+// settle marks generation tg as processed so the loop goes idle instead of
+// re-applying the same target. It is a no-op when a newer target arrived during the
+// pass (targetGen != tg), leaving appliedGen behind so that target re-applies — the
+// just-finished generation was already counted as skipped.
+//
+// When signal is true (the pass converged) it fires onConverged once for this map;
+// when false (the target was dropped on error) it does not — the map did not converge.
+func (m *mapStateManager) settle(tg uint64, signal bool) {
+	m.mu.Lock()
+	if m.targetGen != tg {
+		m.mu.Unlock()
+		return
+	}
+	m.appliedGen = tg
+	setAt := m.targetSetAt
+	m.mu.Unlock()
+
+	if signal && m.onConverged != nil {
+		m.onConverged(time.Since(setAt))
+	}
+}

client/internal/mapsync_test.go

@@ -0,0 +1,281 @@
+package internal
+
+import (
+	"context"
+	"errors"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+
+	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// mergeTarget fills components missing from the incoming update with the pending
+// (not-yet-applied) prev's, in place, so a coalesced/superseded update does not drop
+// the map or config it uniquely carried.
+func TestMapStateManager_MergeTargetPreservesPendingState(t *testing.T) {
+	m := newMapStateManager(nil, nil, nil)
+
+	// config-only update while a full map is still converging (targetGen > appliedGen):
+	// the pending map (+ checks) is filled into the update in place
+	m.targetGen, m.appliedGen = 5, 4
+	prev := &mgmProto.SyncResponse{NetworkMap: &mgmProto.NetworkMap{Serial: 5}}
+	update := &mgmProto.SyncResponse{NetbirdConfig: &mgmProto.NetbirdConfig{}}
+	merged := m.mergeTarget(prev, update)
+	require.Same(t, update, merged, "merges in place, returns the update")
+	require.EqualValues(t, 5, merged.GetNetworkMap().GetSerial(), "pending map preserved")
+	require.NotNil(t, merged.GetNetbirdConfig(), "new config kept")
+
+	// symmetric: map-only update while a config-only update is pending -> keep the config
+	m.targetGen, m.appliedGen = 5, 4
+	prev = &mgmProto.SyncResponse{NetbirdConfig: &mgmProto.NetbirdConfig{}}
+	update = &mgmProto.SyncResponse{NetworkMap: &mgmProto.NetworkMap{Serial: 7}}
+	merged = m.mergeTarget(prev, update)
+	require.EqualValues(t, 7, merged.GetNetworkMap().GetSerial(), "new map kept")
+	require.NotNil(t, merged.GetNetbirdConfig(), "pending config preserved")
+
+	// prev already applied (targetGen == appliedGen): plain replace, no fill-in
+	m.targetGen, m.appliedGen = 5, 5
+	prev = &mgmProto.SyncResponse{NetworkMap: &mgmProto.NetworkMap{Serial: 5}}
+	update = &mgmProto.SyncResponse{NetbirdConfig: &mgmProto.NetbirdConfig{}}
+	merged = m.mergeTarget(prev, update)
+	require.Same(t, update, merged)
+	require.Nil(t, merged.GetNetworkMap(), "no map grafted when prev already applied")
+
+	// nothing to carry (update has a map, prev has no config): plain replace
+	m.targetGen, m.appliedGen = 5, 4
+	prev = &mgmProto.SyncResponse{NetworkMap: &mgmProto.NetworkMap{Serial: 5}}
+	update = &mgmProto.SyncResponse{NetworkMap: &mgmProto.NetworkMap{Serial: 6}}
+	require.Same(t, update, m.mergeTarget(prev, update))
+}
+
+// converges over the bounded passes (apply returns more until the 3rd pass),
+// fires onConverged exactly once, then blocks (no further apply) until a new target.
+func TestMapStateManager_ConvergesThenStops(t *testing.T) {
+	var passes int32
+	var firstPasses int32
+	converged := make(chan struct{}, 1)
+
+	apply := func(_ *mgmProto.SyncResponse, firstPass bool) (bool, error) {
+		n := atomic.AddInt32(&passes, 1)
+		if firstPass {
+			atomic.AddInt32(&firstPasses, 1)
+		}
+		return n < 3, nil // more on pass 1 and 2, converge on pass 3
+	}
+	m := newMapStateManager(apply, nil, func(time.Duration) { converged <- struct{}{} })
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go m.run(ctx)
+
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+
+	select {
+	case <-converged:
+	case <-time.After(2 * time.Second):
+		t.Fatal("manager did not converge")
+	}
+	require.EqualValues(t, 3, atomic.LoadInt32(&passes))
+	require.EqualValues(t, 1, atomic.LoadInt32(&firstPasses), "firstPass true only on pass 1, false on re-runs of the same target")
+
+	// once converged the loop blocks: no further apply calls
+	time.Sleep(100 * time.Millisecond)
+	require.EqualValues(t, 3, atomic.LoadInt32(&passes), "apply must not run after convergence")
+}
+
+// persist runs once per received update (not per apply pass), regardless of how many
+// bounded passes that target takes to converge.
+func TestMapStateManager_PersistsOncePerUpdate(t *testing.T) {
+	var passes, persists int32
+	converged := make(chan struct{}, 1)
+	apply := func(_ *mgmProto.SyncResponse, _ bool) (bool, error) {
+		n := atomic.AddInt32(&passes, 1)
+		return n < 3, nil // 3 passes for one target
+	}
+	persist := func(*mgmProto.SyncResponse) { atomic.AddInt32(&persists, 1) }
+	m := newMapStateManager(apply, persist, func(time.Duration) { converged <- struct{}{} })
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go m.run(ctx)
+
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+	select {
+	case <-converged:
+	case <-time.After(2 * time.Second):
+		t.Fatal("did not converge")
+	}
+	require.EqualValues(t, 3, atomic.LoadInt32(&passes))
+	require.EqualValues(t, 1, atomic.LoadInt32(&persists), "persist once per update, not per pass")
+}
+
+// every update received from management is persisted — even one that is coalesced /
+// skipped from apply before it ever converges.
+func TestMapStateManager_PersistsEveryUpdateIncludingSkipped(t *testing.T) {
+	release := make(chan struct{})
+	var persists int32
+	apply := func(_ *mgmProto.SyncResponse, _ bool) (bool, error) {
+		<-release // hold the first apply so the second update coalesces/skips
+		return false, nil
+	}
+	persist := func(*mgmProto.SyncResponse) { atomic.AddInt32(&persists, 1) }
+	m := newMapStateManager(apply, persist, nil)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go m.run(ctx)
+
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{})) // map1 -> apply blocks
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{})) // map2 supersedes map1 (skipped from apply)
+	close(release)
+
+	// both updates persisted even though map1 is skipped from apply
+	require.Eventually(t, func() bool { return atomic.LoadInt32(&persists) == 2 }, 2*time.Second, 10*time.Millisecond)
+}
+
+// each map that is actually processed (converged before the next arrives) fires
+// onConverged exactly once — mirroring the legacy per-message handleSync timing.
+func TestMapStateManager_SignalsEachProcessedMap(t *testing.T) {
+	converged := make(chan struct{}, 8)
+	apply := func(_ *mgmProto.SyncResponse, _ bool) (bool, error) {
+		return false, nil // converge in one pass
+	}
+	m := newMapStateManager(apply, nil, func(time.Duration) { converged <- struct{}{} })
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go m.run(ctx)
+
+	const maps = 3
+	for i := 0; i < maps; i++ {
+		require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+		select { // wait for this map to converge before sending the next (no coalescing)
+		case <-converged:
+		case <-time.After(2 * time.Second):
+			t.Fatalf("map %d not signaled", i)
+		}
+	}
+
+	// no extra signals once the stream goes quiet
+	select {
+	case <-converged:
+		t.Fatal("unexpected extra onConverged")
+	case <-time.After(100 * time.Millisecond):
+	}
+}
+
+// a map superseded before it converges is skipped: only the latest (processed) map
+// fires onConverged, not the skipped one.
+func TestMapStateManager_SkippedMapNotSignaled(t *testing.T) {
+	release := make(chan struct{})
+	var applies, converged atomic.Int32
+	apply := func(_ *mgmProto.SyncResponse, _ bool) (bool, error) {
+		applies.Add(1)
+		<-release // hold the first apply in-flight so we can queue a newer target
+		return false, nil
+	}
+	m := newMapStateManager(apply, nil, func(time.Duration) { converged.Add(1) })
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go m.run(ctx)
+
+	// map1 is picked up; its apply blocks on release
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+	require.Eventually(t, func() bool { return applies.Load() >= 1 }, 2*time.Second, 5*time.Millisecond)
+
+	// map2 supersedes map1 before it settled -> map1 is skipped
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+	close(release) // let both applies proceed
+
+	// only the processed (latest) map signals; the skipped one does not
+	require.Eventually(t, func() bool { return converged.Load() == 1 }, 2*time.Second, 10*time.Millisecond)
+	time.Sleep(150 * time.Millisecond)
+	require.EqualValues(t, 1, converged.Load(), "skipped map must not fire onConverged")
+	require.EqualValues(t, 2, applies.Load(), "both targets entered apply (map1 once, map2 once)")
+}
+
+// an apply error drops the target: no retry of the same target, no onConverged,
+// the loop goes idle — and a fresh target is still applied afterwards.
+func TestMapStateManager_DropsTargetOnError(t *testing.T) {
+	applied := make(chan struct{}, 8)
+	var failNext atomic.Bool
+	failNext.Store(true)
+
+	apply := func(_ *mgmProto.SyncResponse, _ bool) (bool, error) {
+		applied <- struct{}{}
+		if failNext.Load() {
+			return false, errors.New("boom")
+		}
+		return false, nil // converge in one pass
+	}
+	var converged atomic.Int32
+	m := newMapStateManager(apply, nil, func(time.Duration) { converged.Add(1) })
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go m.run(ctx)
+
+	// first target errors -> applied once, then dropped (no retry, no onConverged)
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+	select {
+	case <-applied:
+	case <-time.After(2 * time.Second):
+		t.Fatal("errored target not applied")
+	}
+	select {
+	case <-applied:
+		t.Fatal("errored target must not be retried")
+	case <-time.After(150 * time.Millisecond):
+	}
+	require.EqualValues(t, 0, converged.Load(), "onConverged must not fire on error")
+
+	// a new target is still processed normally and converges
+	failNext.Store(false)
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+	select {
+	case <-applied:
+	case <-time.After(2 * time.Second):
+		t.Fatal("new target after error not applied")
+	}
+	require.Eventually(t, func() bool { return converged.Load() == 1 }, 2*time.Second, 10*time.Millisecond)
+}
+
+// a new target after convergence triggers a fresh apply; an idle (converged)
+// manager does not apply on its own.
+func TestMapStateManager_ReappliesOnNewTarget(t *testing.T) {
+	applied := make(chan struct{}, 8)
+	apply := func(_ *mgmProto.SyncResponse, _ bool) (bool, error) {
+		applied <- struct{}{}
+		return false, nil // converge in one pass
+	}
+	m := newMapStateManager(apply, nil, nil)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	go m.run(ctx)
+
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+	select {
+	case <-applied:
+	case <-time.After(2 * time.Second):
+		t.Fatal("first target not applied")
+	}
+
+	// converged → must stay idle (no spurious apply)
+	select {
+	case <-applied:
+		t.Fatal("unexpected apply while idle/converged")
+	case <-time.After(150 * time.Millisecond):
+	}
+
+	require.NoError(t, m.SetTarget(&mgmProto.SyncResponse{}))
+	select {
+	case <-applied:
+	case <-time.After(2 * time.Second):
+		t.Fatal("new target not applied")
+	}
+}

client/internal/peer/conn.go

@@ -30,11 +30,6 @@ import (
 	relayClient "github.com/netbirdio/netbird/shared/relay/client"
 )
 
-// wgTimeoutEscalationThreshold is the number of consecutive WireGuard
-// handshake timeouts after which the rosenpass state for the peer is
-// considered desynced and gets reset.
-const wgTimeoutEscalationThreshold = 3
-
 // MetricsRecorder is an interface for recording peer connection metrics
 type MetricsRecorder interface {
 	RecordConnectionStages(
@@ -123,9 +118,6 @@ type Conn struct {
 	wgWatcher       *WGWatcher
 	wgWatcherWg     sync.WaitGroup
 	wgWatcherCancel context.CancelFunc
-	// wgTimeouts counts consecutive WireGuard handshake timeouts without a
-	// successful handshake in between. Guarded by mu.
-	wgTimeouts int
 
 	// used to store the remote Rosenpass key for Relayed connection in case of connection update from ice
 	rosenpassRemoteKey []byte
@@ -691,29 +683,6 @@ func (conn *Conn) onWGDisconnected() {
 	default:
 		conn.Log.Debugf("No active connection to close on WG timeout")
 	}
-
-	conn.escalateWGTimeoutLocked()
-}
-
-// escalateWGTimeoutLocked resets the peer's rosenpass state after repeated
-// handshake timeouts. With rosenpass enabled, persistent timeouts mean the
-// preshared keys have desynced; the renewal exchange runs over the dead
-// tunnel and cannot resync them. Reporting the peer disconnected drops its
-// rosenpass state, so the next connection configuration programs the
-// rendezvous key and the tunnel can bootstrap again. Callers must hold mu.
-func (conn *Conn) escalateWGTimeoutLocked() {
-	if conn.config.RosenpassConfig.PubKey == nil {
-		return
-	}
-
-	conn.wgTimeouts++
-	if conn.wgTimeouts < wgTimeoutEscalationThreshold || conn.onDisconnected == nil {
-		return
-	}
-	conn.wgTimeouts = 0
-
-	conn.Log.Warnf("%d consecutive WireGuard handshake timeouts, resetting rosenpass state for peer", wgTimeoutEscalationThreshold)
-	conn.onDisconnected(conn.config.WgConfig.RemoteKey)
 }
 
 func (conn *Conn) updateRelayStatus(relayServerAddr string, rosenpassPubKey []byte, updateTime time.Time) {
@@ -843,7 +812,7 @@ func (conn *Conn) enableWgWatcherIfNeeded(enabledTime time.Time) {
 	conn.wgWatcherWg.Add(1)
 	go func() {
 		defer conn.wgWatcherWg.Done()
-		conn.wgWatcher.EnableWgWatcher(wgWatcherCtx, enabledTime, conn.onWGDisconnected, conn.onWGHandshakeSuccess, conn.onWGCheckSuccess)
+		conn.wgWatcher.EnableWgWatcher(wgWatcherCtx, enabledTime, conn.onWGDisconnected, conn.onWGHandshakeSuccess)
 	}()
 }
 
@@ -923,15 +892,6 @@ func (conn *Conn) onWGHandshakeSuccess(when time.Time) {
 	conn.recordConnectionMetrics()
 }
 
-// onWGCheckSuccess is called for every watcher check that observed a fresh
-// handshake, including handshakes of connections that were already up when
-// the watcher started.
-func (conn *Conn) onWGCheckSuccess() {
-	conn.mu.Lock()
-	conn.wgTimeouts = 0
-	conn.mu.Unlock()
-}
-
 // recordConnectionMetrics records connection stage timestamps as metrics
 func (conn *Conn) recordConnectionMetrics() {
 	if conn.metricsRecorder == nil {

client/internal/peer/conn_test.go

@@ -7,7 +7,6 @@ import (
 	"testing"
 	"time"
 
-	log "github.com/sirupsen/logrus"
 	"github.com/stretchr/testify/assert"
 
 	"github.com/netbirdio/netbird/client/iface"
@@ -305,84 +304,3 @@ func TestConn_presharedKey_RosenpassManaged(t *testing.T) {
 		t.Fatalf("expected non-nil presharedKey before Rosenpass manages PSK")
 	}
 }
-
-func newWGTimeoutTestConn(rosenpassEnabled bool, disconnected *[]string) *Conn {
-	cfg := ConnConfig{
-		Key:      "LLHf3Ma6z6mdLbriAJbqhX7+nM/B71lgw2+91q3LfhU=",
-		LocalKey: "RRHf3Ma6z6mdLbriAJbqhX7+nM/B71lgw2+91q3LfhU=",
-		WgConfig: WgConfig{RemoteKey: "LLHf3Ma6z6mdLbriAJbqhX7+nM/B71lgw2+91q3LfhU="},
-	}
-	if rosenpassEnabled {
-		cfg.RosenpassConfig = RosenpassConfig{PubKey: []byte("dummykey")}
-	}
-
-	conn := &Conn{
-		ctx:           context.Background(),
-		config:        cfg,
-		Log:           log.WithField("peer", cfg.Key),
-		metricsStages: &MetricsStages{},
-	}
-	conn.SetOnDisconnected(func(remotePeer string) {
-		*disconnected = append(*disconnected, remotePeer)
-	})
-	return conn
-}
-
-// TestConn_onWGDisconnected_EscalatesToRosenpassReset: repeated handshake
-// timeouts with rosenpass enabled mean the preshared keys have desynced. The
-// renewal exchange runs over the dead tunnel and cannot resync them, so after
-// wgTimeoutEscalationThreshold consecutive timeouts the conn must report the
-// peer disconnected, dropping its rosenpass state so the next configuration
-// programs the rendezvous key.
-func TestConn_onWGDisconnected_EscalatesToRosenpassReset(t *testing.T) {
-	var disconnected []string
-	conn := newWGTimeoutTestConn(true, &disconnected)
-
-	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
-		conn.onWGDisconnected()
-	}
-	assert.Empty(t, disconnected, "escalation must not fire below the threshold")
-
-	conn.onWGDisconnected()
-	assert.Equal(t, []string{conn.config.WgConfig.RemoteKey}, disconnected,
-		"reaching the threshold must report the peer disconnected once")
-
-	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
-		conn.onWGDisconnected()
-	}
-	assert.Len(t, disconnected, 1, "escalation must restart counting after firing")
-
-	conn.onWGDisconnected()
-	assert.Len(t, disconnected, 2, "continued timeouts must escalate again")
-}
-
-// TestConn_onWGDisconnected_CheckSuccessResetsEscalation: a successful
-// handshake between timeouts means the tunnel recovered; the counter must
-// start over.
-func TestConn_onWGDisconnected_CheckSuccessResetsEscalation(t *testing.T) {
-	var disconnected []string
-	conn := newWGTimeoutTestConn(true, &disconnected)
-
-	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
-		conn.onWGDisconnected()
-	}
-	conn.onWGCheckSuccess()
-
-	for i := 0; i < wgTimeoutEscalationThreshold-1; i++ {
-		conn.onWGDisconnected()
-	}
-	assert.Empty(t, disconnected, "handshake success must reset the timeout count")
-}
-
-// TestConn_onWGDisconnected_NoEscalationWithoutRosenpass: without rosenpass
-// there is no per-peer key state to reset; repeated timeouts must not report
-// disconnects.
-func TestConn_onWGDisconnected_NoEscalationWithoutRosenpass(t *testing.T) {
-	var disconnected []string
-	conn := newWGTimeoutTestConn(false, &disconnected)
-
-	for i := 0; i < wgTimeoutEscalationThreshold*3; i++ {
-		conn.onWGDisconnected()
-	}
-	assert.Empty(t, disconnected, "escalation must be limited to rosenpass connections")
-}

client/internal/peer/guard/guard.go

@@ -85,11 +85,7 @@ func (g *Guard) reconnectLoopWithRetry(ctx context.Context, callback func()) {
 	defer g.srWatcher.RemoveListener(srReconnectedChan)
 
 	ticker := g.initialTicker(ctx)
-	defer func() {
-		// If backoff.Ticker.send is blocked, context.Done will not close the Ticker goroutine.
-		// We have to explicitly call Stop, even if we use backoff.WithContext.
-		ticker.Stop()
-	}()
+	defer ticker.Stop()
 
 	tickerChannel := ticker.C
 

client/internal/peer/guard/guard_leak_test.go

@@ -1,92 +0,0 @@
-package guard
-
-import (
-	"context"
-	"runtime"
-	"strings"
-	"sync"
-	"testing"
-	"time"
-
-	log "github.com/sirupsen/logrus"
-
-	"github.com/netbirdio/netbird/client/internal/peer/ice"
-)
-
-func newTestGuard(status connStatusFunc) *Guard {
-	srw := NewSRWatcher(nil, nil, nil, ice.Config{})
-	return NewGuard(log.WithField("test", "guard"), status, 50*time.Millisecond, srw)
-}
-
-// countBackoffTickerGoroutines returns how many goroutines are currently sitting
-// in backoff/v4.(*Ticker).run (a ticker goroutine that has not exited).
-func countBackoffTickerGoroutines() int {
-	buf := make([]byte, 1<<25) // 32MB
-	n := runtime.Stack(buf, true)
-	return strings.Count(string(buf[:n]), "backoff/v4.(*Ticker).run")
-}
-
-// TestGuard_ReconnectTicker_NoGoroutineLeakOnShutdown reproduces a observed
-// leak: after a shutdown burst, ticker run/send goroutines stay parked
-// forever even though every reconnect loop has exited.
-func TestGuard_ReconnectTicker_NoGoroutineLeakOnShutdown(t *testing.T) {
-	before := countBackoffTickerGoroutines()
-
-	const peers = 6000
-	cancels := make([]context.CancelFunc, 0, peers)
-	var wg sync.WaitGroup
-
-	// A status check slower than the tick cadence. This models the real
-	// isConnectedOnAllWay/callback doing work: while the loop is busy in the
-	// handler, the ticker fires the next tick and parks in send(), because
-	// send() never selects on ctx.
-	slowStatus := func() ConnStatus {
-		time.Sleep(70 * time.Millisecond)
-		return ConnStatusConnected
-	}
-
-	for range peers {
-		g := newTestGuard(slowStatus)
-		ctx, cancel := context.WithCancel(context.Background())
-		cancels = append(cancels, cancel)
-		wg.Add(1)
-		go func() {
-			defer wg.Done()
-			g.Start(ctx, func() {})
-		}()
-		// Force the live ticker to be a newReconnectTicker.
-		g.SetRelayedConnDisconnected()
-	}
-
-	// Let the replacement tickers get past their 800ms initial interval, so
-	// many are parked in send() waiting on the (slow) consumer when we tear
-	// everything down.
-	time.Sleep(1500 * time.Millisecond)
-
-	// Shutdown burst: cancel every peer at once, like engine teardown.
-	for _, c := range cancels {
-		c()
-	}
-
-	// Every reconnect loop must return
-	waitCh := make(chan struct{})
-	go func() { wg.Wait(); close(waitCh) }()
-	select {
-	case <-waitCh:
-	case <-time.After(30 * time.Second):
-		t.Fatal("not all reconnect loops returned after ctx cancel")
-	}
-
-	// Give any correctly-stopped ticker goroutines time to unwind.
-	for range 50 {
-		runtime.Gosched()
-		time.Sleep(10 * time.Millisecond)
-	}
-
-	leaked := countBackoffTickerGoroutines() - before
-	t.Logf("backoff Ticker.run goroutines still parked after teardown of %d peers: %d", peers, leaked)
-	if leaked > 0 {
-		t.Errorf("LEAK: %d backoff ticker goroutines parked after all reconnect loops exited "+
-			"(defer ticker.Stop() stops the initial ticker, not the live replacement)", leaked)
-	}
-}

client/internal/peer/wg_watcher.go

@@ -71,11 +71,9 @@ func (w *WGWatcher) PrepareInitialHandshake() (ok bool) {
 
 // EnableWgWatcher runs the WireGuard watcher loop using the handshake baseline captured by
 // PrepareInitialHandshake. The watcher runs until ctx is cancelled. Caller is responsible
-// for context lifecycle management. onHandshakeSuccessFn is called only for the first
-// handshake observed by this run, onCheckSuccessFn for every check that observed a fresh
-// handshake, including the first.
-func (w *WGWatcher) EnableWgWatcher(ctx context.Context, enabledTime time.Time, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time), onCheckSuccessFn func()) {
-	w.periodicHandshakeCheck(ctx, onDisconnectedFn, onHandshakeSuccessFn, onCheckSuccessFn, enabledTime, w.initialHandshake)
+// for context lifecycle management.
+func (w *WGWatcher) EnableWgWatcher(ctx context.Context, enabledTime time.Time, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time)) {
+	w.periodicHandshakeCheck(ctx, onDisconnectedFn, onHandshakeSuccessFn, enabledTime, w.initialHandshake)
 
 	w.muEnabled.Lock()
 	w.enabled = false
@@ -92,7 +90,7 @@ func (w *WGWatcher) Reset() {
 }
 
 // wgStateCheck help to check the state of the WireGuard handshake and relay connection
-func (w *WGWatcher) periodicHandshakeCheck(ctx context.Context, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time), onCheckSuccessFn func(), enabledTime time.Time, initialHandshake time.Time) {
+func (w *WGWatcher) periodicHandshakeCheck(ctx context.Context, onDisconnectedFn func(), onHandshakeSuccessFn func(when time.Time), enabledTime time.Time, initialHandshake time.Time) {
 	w.log.Infof("WireGuard watcher started")
 
 	timer := time.NewTimer(wgHandshakeOvertime)
@@ -119,10 +117,6 @@ func (w *WGWatcher) periodicHandshakeCheck(ctx context.Context, onDisconnectedFn
 				}
 			}
 
-			if onCheckSuccessFn != nil && ctx.Err() == nil {
-				onCheckSuccessFn()
-			}
-
 			lastHandshake = *handshake
 
 			resetTime := time.Until(handshake.Add(checkPeriod))

client/internal/peer/wg_watcher_test.go

@@ -24,72 +24,6 @@ func (m *MocWgIface) disconnect() {
 	m.stop = true
 }
 
-type mockHandshakeStats struct {
-	mu        sync.Mutex
-	handshake time.Time
-}
-
-func (m *mockHandshakeStats) GetStats() (map[string]configurer.WGStats, error) {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-	return map[string]configurer.WGStats{"": {LastHandshake: m.handshake}}, nil
-}
-
-func (m *mockHandshakeStats) advance() {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-	m.handshake = time.Now()
-}
-
-// TestWGWatcher_CheckSuccessCallback: onCheckSuccessFn must fire for a fresh
-// handshake even when the watcher started with an existing handshake baseline,
-// the case where onHandshakeSuccessFn stays silent.
-func TestWGWatcher_CheckSuccessCallback(t *testing.T) {
-	// checkPeriod bounds how stale a handshake may be before the watcher treats it
-	// as a suspended-machine timeout. The first check fires after wgHandshakeOvertime,
-	// so keep checkPeriod well above any scheduling jitter to avoid a false timeout
-	// converting the expected success into a disconnect on a loaded runner.
-	checkPeriod = 1 * time.Minute
-	wgHandshakeOvertime = 1 * time.Second
-
-	mlog := log.WithField("peer", "tet")
-	// Use an old baseline so advance() yields a strictly newer handshake even on
-	// platforms with coarse clock resolution (Windows), where two time.Now() calls
-	// microseconds apart can return the same instant and read as a timed-out handshake.
-	stats := &mockHandshakeStats{handshake: time.Now().Add(-time.Hour)}
-	watcher := NewWGWatcher(mlog, stats, "", newStateDump("peer", mlog, &Status{}))
-
-	ctx, cancel := context.WithCancel(context.Background())
-	defer cancel()
-
-	require.True(t, watcher.PrepareInitialHandshake())
-
-	firstHandshake := make(chan struct{}, 1)
-	checkSuccess := make(chan struct{}, 1)
-	go watcher.EnableWgWatcher(ctx, time.Now(), func() {}, func(when time.Time) {
-		firstHandshake <- struct{}{}
-	}, func() {
-		select {
-		case checkSuccess <- struct{}{}:
-		default:
-		}
-	})
-
-	stats.advance()
-
-	select {
-	case <-checkSuccess:
-	case <-time.After(10 * time.Second):
-		t.Errorf("timeout waiting for check success callback")
-	}
-
-	select {
-	case <-firstHandshake:
-		t.Errorf("first-handshake callback must not fire for a non-zero baseline")
-	default:
-	}
-}
-
 func TestWGWatcher_EnableWgWatcher(t *testing.T) {
 	checkPeriod = 5 * time.Second
 	wgHandshakeOvertime = 1 * time.Second
@@ -110,7 +44,7 @@ func TestWGWatcher_EnableWgWatcher(t *testing.T) {
 		onDisconnected <- struct{}{}
 	}, func(when time.Time) {
 		mlog.Infof("onHandshakeSuccess: %v", when)
-	}, nil)
+	})
 
 	// wait for initial reading
 	time.Sleep(2 * time.Second)
@@ -139,7 +73,7 @@ func TestWGWatcher_ReEnable(t *testing.T) {
 	wg.Add(1)
 	go func() {
 		defer wg.Done()
-		watcher.EnableWgWatcher(ctx, time.Now(), func() {}, func(when time.Time) {}, nil)
+		watcher.EnableWgWatcher(ctx, time.Now(), func() {}, func(when time.Time) {})
 	}()
 	cancel()
 
@@ -155,7 +89,7 @@ func TestWGWatcher_ReEnable(t *testing.T) {
 	onDisconnected := make(chan struct{}, 1)
 	go watcher.EnableWgWatcher(ctx, time.Now(), func() {
 		onDisconnected <- struct{}{}
-	}, func(when time.Time) {}, nil)
+	}, func(when time.Time) {})
 
 	time.Sleep(2 * time.Second)
 	mocWgIface.disconnect()

client/internal/rosenpass/manager.go

@@ -39,7 +39,6 @@ type rpServer interface {
 
 type Manager struct {
 	ifaceName    string
-	localWgKey   wgtypes.Key
 	spk          []byte
 	ssk          []byte
 	rpKeyHash    string
@@ -52,9 +51,8 @@ type Manager struct {
 	wgIface      PresharedKeySetter
 }
 
-// NewManager creates a new Rosenpass manager. localWgKey is the local
-// WireGuard public key, used to derive the per-peer rendezvous key.
-func NewManager(preSharedKey *wgtypes.Key, wgIfaceName string, localWgKey wgtypes.Key) (*Manager, error) {
+// NewManager creates a new Rosenpass manager
+func NewManager(preSharedKey *wgtypes.Key, wgIfaceName string) (*Manager, error) {
 	public, secret, err := rp.GenerateKeyPair()
 	if err != nil {
 		return nil, err
@@ -64,7 +62,6 @@ func NewManager(preSharedKey *wgtypes.Key, wgIfaceName string, localWgKey wgtype
 	log.Tracef("generated new rosenpass key pair with public key %s", rpKeyHash)
 	return &Manager{
 		ifaceName:    wgIfaceName,
-		localWgKey:   localWgKey,
 		rpKeyHash:    rpKeyHash,
 		spk:          public,
 		ssk:          secret,
@@ -76,7 +73,7 @@ func NewManager(preSharedKey *wgtypes.Key, wgIfaceName string, localWgKey wgtype
 		// nil receiver in addPeer -> m.rpWgHandler.AddPeer. generateConfig will
 		// replace it with a fresh handler on each Run() to clear stale peer
 		// state from previous engine sessions.
-		rpWgHandler: NewNetbirdHandler((*[32]byte)(preSharedKey), localWgKey),
+		rpWgHandler: NewNetbirdHandler(),
 		lock:        sync.Mutex{},
 	}, nil
 }
@@ -164,7 +161,7 @@ func (m *Manager) generateConfig() (rp.Config, error) {
 	cfg.Peers = []rp.PeerConfig{}
 
 	m.lock.Lock()
-	m.rpWgHandler = NewNetbirdHandler(m.preSharedKey, m.localWgKey)
+	m.rpWgHandler = NewNetbirdHandler()
 	if m.wgIface != nil {
 		m.rpWgHandler.SetInterface(m.wgIface)
 	}

client/internal/rosenpass/manager_test.go

@@ -85,7 +85,7 @@ func newTestManager(spkFirstByte byte, mock *mockServer) *Manager {
 		ssk:         make([]byte, 32),
 		rpKeyHash:   "test-hash",
 		rpPeerIDs:   make(map[string]*rp.PeerID),
-		rpWgHandler: NewNetbirdHandler(nil, wgtypes.Key{0x01}),
+		rpWgHandler: NewNetbirdHandler(),
 		server:      mock,
 	}
 }
@@ -255,7 +255,7 @@ func TestAddPeer_NilServer_ReturnsErrorNoCrash(t *testing.T) {
 // issue #4341 cannot occur in the window between NewManager and Run().
 func TestNewManager_PreInitializesHandler(t *testing.T) {
 	psk := wgtypes.Key{}
-	m, err := NewManager(&psk, "wt0", wgtypes.Key{0x01})
+	m, err := NewManager(&psk, "wt0")
 	require.NoError(t, err)
 	require.NotNil(t, m.rpWgHandler, "rpWgHandler must be initialized in NewManager")
 }
@@ -329,10 +329,10 @@ func TestIsPresharedKeyInitialized_AddedButNotHandshaken_ReturnsFalse(t *testing
 	require.False(t, m.IsPresharedKeyInitialized(wgKey))
 }
 
-// --- NetbirdHandler.applyKey ----------------------------------------------
+// --- NetbirdHandler.outputKey ----------------------------------------------
 
-func TestHandler_ApplyKey_FirstCallUsesUpdateOnlyFalse(t *testing.T) {
-	h := NewNetbirdHandler(nil, wgtypes.Key{0x01})
+func TestHandler_OutputKey_FirstCallUsesUpdateOnlyFalse(t *testing.T) {
+	h := NewNetbirdHandler()
 	iface := &mockIface{}
 	h.SetInterface(iface)
 
@@ -348,8 +348,8 @@ func TestHandler_ApplyKey_FirstCallUsesUpdateOnlyFalse(t *testing.T) {
 	require.Equal(t, wgKey.String(), iface.calls[0].peerKey)
 }
 
-func TestHandler_ApplyKey_SubsequentCallsUseUpdateOnlyTrue(t *testing.T) {
-	h := NewNetbirdHandler(nil, wgtypes.Key{0x01})
+func TestHandler_OutputKey_SubsequentCallsUseUpdateOnlyTrue(t *testing.T) {
+	h := NewNetbirdHandler()
 	iface := &mockIface{}
 	h.SetInterface(iface)
 
@@ -364,8 +364,8 @@ func TestHandler_ApplyKey_SubsequentCallsUseUpdateOnlyTrue(t *testing.T) {
 	require.True(t, iface.calls[1].updateOnly, "subsequent rotations must use updateOnly=true")
 }
 
-func TestHandler_ApplyKey_NilInterface_NoCrashNoCall(t *testing.T) {
-	h := NewNetbirdHandler(nil, wgtypes.Key{0x01})
+func TestHandler_OutputKey_NilInterface_NoCrashNoCall(t *testing.T) {
+	h := NewNetbirdHandler()
 	// no SetInterface — iface remains nil
 	pid := rp.PeerID{0x03}
 	h.AddPeer(pid, "wt0", rp.Key(wgtypes.Key{}))
@@ -374,8 +374,8 @@ func TestHandler_ApplyKey_NilInterface_NoCrashNoCall(t *testing.T) {
 	h.HandshakeCompleted(pid, rp.Key{})
 }
 
-func TestHandler_ApplyKey_UnknownPeer_NoCall(t *testing.T) {
-	h := NewNetbirdHandler(nil, wgtypes.Key{0x01})
+func TestHandler_OutputKey_UnknownPeer_NoCall(t *testing.T) {
+	h := NewNetbirdHandler()
 	iface := &mockIface{}
 	h.SetInterface(iface)
 
@@ -384,7 +384,7 @@ func TestHandler_ApplyKey_UnknownPeer_NoCall(t *testing.T) {
 }
 
 func TestHandler_RemovePeer_ClearsInitializedState(t *testing.T) {
-	h := NewNetbirdHandler(nil, wgtypes.Key{0x01})
+	h := NewNetbirdHandler()
 	iface := &mockIface{}
 	h.SetInterface(iface)
 
@@ -398,7 +398,7 @@ func TestHandler_RemovePeer_ClearsInitializedState(t *testing.T) {
 }
 
 func TestHandler_SetInterfaceAfterAddPeer_StillReceivesKey(t *testing.T) {
-	h := NewNetbirdHandler(nil, wgtypes.Key{0x01})
+	h := NewNetbirdHandler()
 	pid := rp.PeerID{0x05}
 	wgKey := wgtypes.Key{0xEE}
 	h.AddPeer(pid, "wt0", rp.Key(wgKey))

client/internal/rosenpass/netbird_handler.go

@@ -18,34 +18,19 @@ type PresharedKeySetter interface {
 type wireGuardPeer struct {
 	Interface string
 	PublicKey rp.Key
-	// initialized is true once a completed exchange has set a
-	// Rosenpass-managed PSK for this peer.
-	initialized bool
-	// chainKey is the key output by the last completed exchange, advanced by
-	// one ratchet step on expiry. Nil until the first exchange completes and
-	// after the peer has fallen back to the rendezvous key.
-	chainKey *wgtypes.Key
-	// expiries counts failed renewals since the last completed exchange.
-	expiries int
 }
 
 type NetbirdHandler struct {
-	mu    sync.Mutex
-	iface PresharedKeySetter
-	// preSharedKey is the account-level preshared key, used as the rendezvous
-	// key when set. Nil means the deterministic seed key is used instead.
-	preSharedKey *[32]byte
-	// localWgKey is the local WireGuard public key, one of the two inputs to
-	// the deterministic seed key.
-	localWgKey wgtypes.Key
-	peers      map[rp.PeerID]*wireGuardPeer
+	mu               sync.Mutex
+	iface            PresharedKeySetter
+	peers            map[rp.PeerID]wireGuardPeer
+	initializedPeers map[rp.PeerID]bool
 }
 
-func NewNetbirdHandler(preSharedKey *[32]byte, localWgKey wgtypes.Key) *NetbirdHandler {
+func NewNetbirdHandler() *NetbirdHandler {
 	return &NetbirdHandler{
-		preSharedKey: preSharedKey,
-		localWgKey:   localWgKey,
-		peers:        map[rp.PeerID]*wireGuardPeer{},
+		peers:            map[rp.PeerID]wireGuardPeer{},
+		initializedPeers: map[rp.PeerID]bool{},
 	}
 }
 
@@ -57,16 +42,10 @@ func (h *NetbirdHandler) SetInterface(iface PresharedKeySetter) {
 	h.iface = iface
 }
 
-// AddPeer registers a peer with the handler. Re-adding a known peer (every
-// reconnection does) keeps its key recovery state.
 func (h *NetbirdHandler) AddPeer(pid rp.PeerID, intf string, pk rp.Key) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
-	if existing, ok := h.peers[pid]; ok && existing.PublicKey == pk {
-		existing.Interface = intf
-		return
-	}
-	h.peers[pid] = &wireGuardPeer{
+	h.peers[pid] = wireGuardPeer{
 		Interface: intf,
 		PublicKey: pk,
 	}
@@ -76,6 +55,7 @@ func (h *NetbirdHandler) RemovePeer(pid rp.PeerID) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 	delete(h.peers, pid)
+	delete(h.initializedPeers, pid)
 }
 
 // IsPeerInitialized returns true if Rosenpass has completed a handshake
@@ -83,120 +63,50 @@ func (h *NetbirdHandler) RemovePeer(pid rp.PeerID) {
 func (h *NetbirdHandler) IsPeerInitialized(pid rp.PeerID) bool {
 	h.mu.Lock()
 	defer h.mu.Unlock()
-	peer, ok := h.peers[pid]
-	return ok && peer.initialized
+	return h.initializedPeers[pid]
 }
 
-// HandshakeCompleted programs the freshly exchanged output key and resets the
-// peer's key recovery state.
 func (h *NetbirdHandler) HandshakeCompleted(pid rp.PeerID, key rp.Key) {
-	psk := wgtypes.Key(key)
-
-	h.mu.Lock()
-	defer h.mu.Unlock()
-
-	peer, ok := h.peers[pid]
-	if !ok {
-		return
-	}
-	if peer.expiries > 0 {
-		log.Infof("rosenpass exchange completed for peer %s after %d expired renewals", wgtypes.Key(peer.PublicKey), peer.expiries)
-	}
-	// chainKey tracks the shared exchange output regardless of the local write
-	// outcome, so both ends still converge on the next expiry.
-	peer.chainKey = &psk
-	peer.expiries = 0
-	if !h.applyKeyLocked(pid, psk, peer.initialized) {
-		return
-	}
-	peer.initialized = true
+	h.outputKey(rp.KeyOutputReasonStale, pid, key)
 }
 
-// HandshakeExpired replaces the expired key. The renewal exchange runs over
-// the tunnel keyed by the PSK itself, so the replacement must be derivable on
-// both ends without communication: the first expiry ratchets the last shared
-// key forward, repeated expiries (and expiries without a completed exchange)
-// fall back to the rendezvous key and drop the peer out of the initialized
-// state so connection reconfigurations reprogram the rendezvous key as well.
 func (h *NetbirdHandler) HandshakeExpired(pid rp.PeerID) {
+	key, _ := rp.GeneratePresharedKey()
+	h.outputKey(rp.KeyOutputReasonStale, pid, key)
+}
+
+func (h *NetbirdHandler) outputKey(_ rp.KeyOutputReason, pid rp.PeerID, psk rp.Key) {
 	h.mu.Lock()
-	defer h.mu.Unlock()
+	iface := h.iface
+	wg, ok := h.peers[pid]
+	isInitialized := h.initializedPeers[pid]
+	h.mu.Unlock()
 
-	peer, ok := h.peers[pid]
-	if !ok {
-		return
-	}
-
-	peer.expiries++
-
-	var psk wgtypes.Key
-	if peer.chainKey != nil && peer.expiries == 1 {
-		log.Infof("rosenpass key for peer %s expired without renewal, advancing to ratcheted key", wgtypes.Key(peer.PublicKey))
-		psk = RatchetKey(*peer.chainKey)
-		peer.chainKey = &psk
-	} else {
-		rendezvous, err := h.rendezvousKey(peer)
-		if err != nil {
-			// Fail closed: without a rendezvous key the expired key must
-			// still be rotated out, even if the replacement is unusable.
-			log.Errorf("failed to derive rendezvous key, replacing expired key with a random one: %v", err)
-			h.applyRandomKeyLocked(pid)
-			return
-		}
-		log.Warnf("rosenpass key for peer %s expired %d times without renewal, falling back to the rendezvous key", wgtypes.Key(peer.PublicKey), peer.expiries)
-		psk = rendezvous
-		peer.chainKey = nil
-		peer.initialized = false
-	}
-
-	h.applyKeyLocked(pid, psk, true)
-}
-
-// rendezvousKey returns the key both ends converge on without communication:
-// the account-level preshared key when configured, the deterministic seed key
-// otherwise. It mirrors the key that peer connections program when Rosenpass
-// does not manage the peer yet.
-func (h *NetbirdHandler) rendezvousKey(peer *wireGuardPeer) (wgtypes.Key, error) {
-	if h.preSharedKey != nil {
-		return *h.preSharedKey, nil
-	}
-
-	seed, err := DeterministicSeedKey(h.localWgKey.String(), wgtypes.Key(peer.PublicKey).String())
-	if err != nil {
-		return wgtypes.Key{}, err
-	}
-	return *seed, nil
-}
-
-// applyKeyLocked writes the preshared key for the peer to the WireGuard
-// interface and reports whether the write succeeded. Callers must hold h.mu
-// for the whole state-mutation-plus-write so that a concurrent completion and
-// expiry cannot reorder their writes relative to the in-memory chain key.
-func (h *NetbirdHandler) applyKeyLocked(pid rp.PeerID, psk wgtypes.Key, updateOnly bool) bool {
-	peer, ok := h.peers[pid]
-	if !ok {
-		return false
-	}
-
-	if h.iface == nil {
+	if iface == nil {
 		log.Warn("rosenpass: interface not set, cannot update preshared key")
-		return false
-	}
-
-	peerKey := wgtypes.Key(peer.PublicKey).String()
-	if err := h.iface.SetPresharedKey(peerKey, psk, updateOnly); err != nil {
-		log.Errorf("Failed to apply rosenpass key: %v", err)
-		return false
-	}
-
-	return true
-}
-
-func (h *NetbirdHandler) applyRandomKeyLocked(pid rp.PeerID) {
-	key, err := rp.GeneratePresharedKey()
-	if err != nil {
-		log.Errorf("failed to generate random preshared key: %v", err)
 		return
 	}
-	h.applyKeyLocked(pid, wgtypes.Key(key), true)
+
+	if !ok {
+		return
+	}
+
+	peerKey := wgtypes.Key(wg.PublicKey).String()
+	pskKey := wgtypes.Key(psk)
+
+	// Use updateOnly=true for later rotations (peer already has Rosenpass PSK)
+	// Use updateOnly=false for first rotation (peer has original/empty PSK)
+	if err := iface.SetPresharedKey(peerKey, pskKey, isInitialized); err != nil {
+		log.Errorf("Failed to apply rosenpass key: %v", err)
+		return
+	}
+
+	// Mark peer as isInitialized after the successful first rotation
+	if !isInitialized {
+		h.mu.Lock()
+		if _, exists := h.peers[pid]; exists {
+			h.initializedPeers[pid] = true
+		}
+		h.mu.Unlock()
+	}
 }

client/internal/rosenpass/netbird_handler_test.go

@@ -1,250 +0,0 @@
-package rosenpass
-
-import (
-	"testing"
-
-	rp "cunicu.li/go-rosenpass"
-	"github.com/stretchr/testify/require"
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-)
-
-// handlerTestLink wires two NetbirdHandlers as the two ends of a single
-// tunnel: handler A manages the rosenpass peer B and vice versa, the way two
-// NetBird clients see each other.
-type handlerTestLink struct {
-	handlerA, handlerB *NetbirdHandler
-	ifaceA, ifaceB     *mockIface
-	pidA, pidB         rp.PeerID
-	wgKeyA, wgKeyB     wgtypes.Key
-}
-
-func newHandlerTestLink(t *testing.T, preSharedKey *[32]byte) *handlerTestLink {
-	t.Helper()
-
-	link := &handlerTestLink{
-		ifaceA: &mockIface{},
-		ifaceB: &mockIface{},
-	}
-	link.pidA[0] = 0xaa
-	link.pidB[0] = 0xbb
-	link.wgKeyA[31] = 1
-	link.wgKeyB[31] = 2
-
-	link.handlerA = NewNetbirdHandler(preSharedKey, link.wgKeyA)
-	link.handlerB = NewNetbirdHandler(preSharedKey, link.wgKeyB)
-
-	link.handlerA.SetInterface(link.ifaceA)
-	link.handlerB.SetInterface(link.ifaceB)
-
-	link.handlerA.AddPeer(link.pidB, "wt0", rp.Key(link.wgKeyB))
-	link.handlerB.AddPeer(link.pidA, "wt0", rp.Key(link.wgKeyA))
-
-	return link
-}
-
-// complete simulates a completed rosenpass exchange: both ends derive the
-// same output key.
-func (l *handlerTestLink) complete(osk rp.Key) {
-	l.handlerA.HandshakeCompleted(l.pidB, osk)
-	l.handlerB.HandshakeCompleted(l.pidA, osk)
-}
-
-// expire simulates a failed key renewal on both ends.
-func (l *handlerTestLink) expire() {
-	l.handlerA.HandshakeExpired(l.pidB)
-	l.handlerB.HandshakeExpired(l.pidA)
-}
-
-func lastPSK(t *testing.T, m *mockIface) wgtypes.Key {
-	t.Helper()
-	m.mu.Lock()
-	defer m.mu.Unlock()
-	require.NotEmpty(t, m.calls, "expected at least one SetPresharedKey call")
-	return m.calls[len(m.calls)-1].psk
-}
-
-func TestHandshakeCompleted_SetsKeyAndInitializes(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	var osk rp.Key
-	osk[0] = 0x42
-	link.complete(osk)
-
-	require.Equal(t, wgtypes.Key(osk), lastPSK(t, link.ifaceA), "completed exchange must program the osk")
-	require.False(t, link.ifaceA.calls[0].updateOnly, "first rotation must not be update-only")
-	require.True(t, link.handlerA.IsPeerInitialized(link.pidB), "peer must be initialized after first completed exchange")
-
-	link.complete(osk)
-	require.True(t, link.ifaceA.calls[1].updateOnly, "later rotations must be update-only")
-}
-
-// TestHandshakeExpired_BothSidesConverge encodes the core recovery invariant:
-// rosenpass renewals run over the tunnel that the PSK itself keys, so when a
-// renewal fails on both ends, both ends must fall back to the same key or the
-// tunnel can never handshake again.
-func TestHandshakeExpired_BothSidesConverge(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	var osk rp.Key
-	osk[0] = 0x42
-	link.complete(osk)
-
-	link.expire()
-	keyA := lastPSK(t, link.ifaceA)
-	keyB := lastPSK(t, link.ifaceB)
-	require.NotEqual(t, wgtypes.Key(osk), keyA, "expired key must be rotated out")
-	require.Equal(t, keyA, keyB, "both ends must converge on the same key after expiry")
-
-	link.expire()
-	require.Equal(t, lastPSK(t, link.ifaceA), lastPSK(t, link.ifaceB),
-		"both ends must still converge after repeated expiries")
-}
-
-// TestHandshakeExpired_ExpiryWithoutCompletionConverges covers the bootstrap
-// case: the initial exchange never completed (the tunnel ran on the rendezvous
-// key), so an expiry must not replace the working key with an unrecoverable
-// one on either end.
-func TestHandshakeExpired_ExpiryWithoutCompletionConverges(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	link.expire()
-	require.Equal(t, lastPSK(t, link.ifaceA), lastPSK(t, link.ifaceB),
-		"both ends must converge when the exchange never completed")
-}
-
-// TestHandshakeExpired_RepeatedExpiryClearsInitialized: once renewals keep
-// failing, the peer must drop out of the initialized state so the next
-// connection reconfiguration reprograms the rendezvous key instead of
-// preserving a poisoned rosenpass-managed key.
-func TestHandshakeExpired_RepeatedExpiryClearsInitialized(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	var osk rp.Key
-	osk[0] = 0x42
-	link.complete(osk)
-
-	link.expire()
-	link.expire()
-
-	require.False(t, link.handlerA.IsPeerInitialized(link.pidB),
-		"repeated expiries must clear the initialized state")
-	require.False(t, link.handlerB.IsPeerInitialized(link.pidA),
-		"repeated expiries must clear the initialized state")
-}
-
-// TestHandshakeCompleted_AfterExpiryRecovers: a completed exchange after a
-// desync must fully reset the recovery state.
-func TestHandshakeCompleted_AfterExpiryRecovers(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	var osk1, osk2 rp.Key
-	osk1[0] = 1
-	osk2[0] = 2
-
-	link.complete(osk1)
-	link.expire()
-	link.expire()
-
-	link.complete(osk2)
-	require.Equal(t, wgtypes.Key(osk2), lastPSK(t, link.ifaceA), "new exchange must program the fresh osk")
-	require.True(t, link.handlerA.IsPeerInitialized(link.pidB), "peer must be initialized again after recovery")
-
-	link.expire()
-	require.Equal(t, lastPSK(t, link.ifaceA), lastPSK(t, link.ifaceB),
-		"recovered link must converge again on the next expiry")
-	require.NotEqual(t, wgtypes.Key(osk2), lastPSK(t, link.ifaceA), "expired key must be rotated out")
-}
-
-// TestHandshakeExpired_FirstExpiryRatchetsLastKey: the first expiry must
-// derive the replacement from the last shared key, so an attacker who only
-// blocks the renewal exchange gains nothing over the previous key.
-func TestHandshakeExpired_FirstExpiryRatchetsLastKey(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	var osk rp.Key
-	osk[0] = 0x42
-	link.complete(osk)
-
-	link.expire()
-	require.Equal(t, RatchetKey(wgtypes.Key(osk)), lastPSK(t, link.ifaceA),
-		"first expiry must program the ratcheted key")
-	require.True(t, link.handlerA.IsPeerInitialized(link.pidB),
-		"ratchet step must keep the peer initialized so reconfigurations preserve the key")
-}
-
-// TestHandshakeExpired_RepeatedExpiryFallsBackToSeed: once the ratchet key
-// also fails, both ends must land on the same key that peer connections
-// program for uninitialized peers, so a reconnect completes the recovery.
-func TestHandshakeExpired_RepeatedExpiryFallsBackToSeed(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	var osk rp.Key
-	osk[0] = 0x42
-	link.complete(osk)
-
-	link.expire()
-	link.expire()
-
-	seed, err := DeterministicSeedKey(link.wgKeyA.String(), link.wgKeyB.String())
-	require.NoError(t, err)
-	require.Equal(t, *seed, lastPSK(t, link.ifaceA), "repeated expiry must fall back to the seed key")
-	require.Equal(t, *seed, lastPSK(t, link.ifaceB), "repeated expiry must fall back to the seed key")
-}
-
-// TestHandshakeExpired_ConfiguredPSKUsedAsRendezvous: with an account-level
-// preshared key configured, the fallback must be that key, matching what peer
-// connections program for uninitialized peers.
-func TestHandshakeExpired_ConfiguredPSKUsedAsRendezvous(t *testing.T) {
-	psk := &[32]byte{0x77}
-	link := newHandlerTestLink(t, psk)
-
-	var osk rp.Key
-	osk[0] = 0x42
-	link.complete(osk)
-
-	link.expire()
-	link.expire()
-
-	require.Equal(t, wgtypes.Key(*psk), lastPSK(t, link.ifaceA),
-		"fallback must be the configured preshared key")
-	require.Equal(t, wgtypes.Key(*psk), lastPSK(t, link.ifaceB),
-		"fallback must be the configured preshared key on both ends")
-}
-
-// TestHandshakeExpired_ExpiryWritesAreUpdateOnly: expiry replacements must
-// never create a WireGuard peer that connection management has removed.
-func TestHandshakeExpired_ExpiryWritesAreUpdateOnly(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	var osk rp.Key
-	osk[0] = 0x42
-	link.complete(osk)
-
-	link.expire()
-	link.expire()
-
-	for _, call := range link.ifaceA.calls[1:] {
-		require.True(t, call.updateOnly, "expiry writes must be update-only")
-	}
-}
-
-// TestAddPeer_ReAddKeepsRecoveryState: reconnections re-add the peer on every
-// OnConnected; that must not reset the expiry chain state.
-func TestAddPeer_ReAddKeepsRecoveryState(t *testing.T) {
-	link := newHandlerTestLink(t, nil)
-
-	var osk rp.Key
-	osk[0] = 0x42
-	link.complete(osk)
-	link.expire()
-
-	link.handlerA.AddPeer(link.pidB, "wt0", rp.Key(link.wgKeyB))
-	require.True(t, link.handlerA.IsPeerInitialized(link.pidB),
-		"re-adding a known peer must keep its state")
-
-	link.expire()
-	seed, err := DeterministicSeedKey(link.wgKeyA.String(), link.wgKeyB.String())
-	require.NoError(t, err)
-	require.Equal(t, *seed, lastPSK(t, link.ifaceA),
-		"second expiry after re-add must continue to the seed fallback")
-}

client/internal/rosenpass/seed.go

@@ -1,28 +1,11 @@
 package rosenpass
 
 import (
-	"crypto/sha256"
 	"fmt"
 
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 )
 
-// ratchetLabel domain-separates the expiry ratchet from other uses of the
-// rosenpass output key.
-const ratchetLabel = "netbird-rosenpass-expiry-ratchet"
-
-// RatchetKey derives the successor preshared key from the previous Rosenpass
-// output key. When a key expires without a completed renewal, both peers
-// advance their last shared key by one ratchet step: the expired key is
-// rotated out while both ends still converge on an identical, non-public
-// replacement without communicating.
-func RatchetKey(prev wgtypes.Key) wgtypes.Key {
-	input := make([]byte, 0, len(ratchetLabel)+len(prev))
-	input = append(input, ratchetLabel...)
-	input = append(input, prev[:]...)
-	return sha256.Sum256(input)
-}
-
 // DeterministicSeedKey derives a 32-byte WireGuard preshared key from a pair
 // of peer public keys. Both peers, given the same key pair, produce the same
 // output regardless of which side runs the function: the inputs are ordered

client/internal/routemanager/exit_node_selection_test.go

@@ -1,191 +0,0 @@
-package routemanager
-
-import (
-	"net/netip"
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-
-	"github.com/netbirdio/netbird/client/internal/routeselector"
-	"github.com/netbirdio/netbird/route"
-)
-
-func newExitNodeTestManager() *DefaultManager {
-	return &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
-}
-
-func exitRoute(netID, peer string, skipAutoApply bool) *route.Route {
-	return &route.Route{
-		NetID:         route.NetID(netID),
-		Network:       netip.MustParsePrefix("0.0.0.0/0"),
-		Peer:          peer,
-		SkipAutoApply: skipAutoApply,
-	}
-}
-
-func TestPickPreferredExitNode(t *testing.T) {
-	tests := []struct {
-		name string
-		info exitNodeInfo
-		want route.NetID
-	}{
-		{
-			name: "persisted user selection wins over management",
-			info: exitNodeInfo{
-				allIDs:               []route.NetID{"a", "b", "c"},
-				userSelected:         []route.NetID{"b"},
-				selectedByManagement: []route.NetID{"a"},
-			},
-			want: "b",
-		},
-		{
-			name: "multiple user-selected self-heal to deterministic min",
-			info: exitNodeInfo{
-				allIDs:       []route.NetID{"a", "b", "c"},
-				userSelected: []route.NetID{"c", "a"},
-			},
-			want: "a",
-		},
-		{
-			name: "explicit opt-out keeps none",
-			info: exitNodeInfo{
-				allIDs:         []route.NetID{"a", "b"},
-				userDeselected: []route.NetID{"a", "b"},
-			},
-			want: "",
-		},
-		{
-			name: "fresh defaults to management auto-apply pick",
-			info: exitNodeInfo{
-				allIDs:               []route.NetID{"a", "b", "c"},
-				selectedByManagement: []route.NetID{"b"},
-			},
-			want: "b",
-		},
-		{
-			name: "no user pick and no management auto-apply selects none",
-			info: exitNodeInfo{
-				allIDs: []route.NetID{"c", "a", "b"},
-			},
-			want: "",
-		},
-		{
-			name: "user-deselect does not block a management auto-apply sibling",
-			info: exitNodeInfo{
-				allIDs:               []route.NetID{"a", "b"},
-				userDeselected:       []route.NetID{"a"},
-				selectedByManagement: []route.NetID{"b"},
-			},
-			want: "b",
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			assert.Equal(t, tt.want, pickPreferredExitNode(tt.info), "preferred exit node")
-		})
-	}
-}
-
-func TestEnforceSingleExitNode(t *testing.T) {
-	m := newExitNodeTestManager()
-	all := []route.NetID{"a", "b", "c"}
-
-	m.enforceSingleExitNode("b", all)
-	assert.False(t, m.routeSelector.IsSelected("a"), "a should be deselected")
-	assert.True(t, m.routeSelector.IsSelected("b"), "b should be the only selected exit node")
-	assert.False(t, m.routeSelector.IsSelected("c"), "c should be deselected")
-
-	// Switching the preferred node moves the single selection.
-	m.enforceSingleExitNode("c", all)
-	assert.False(t, m.routeSelector.IsSelected("a"), "a stays deselected")
-	assert.False(t, m.routeSelector.IsSelected("b"), "b should now be deselected")
-	assert.True(t, m.routeSelector.IsSelected("c"), "c should now be selected")
-
-	// Empty preferred turns every exit node off.
-	m.enforceSingleExitNode("", all)
-	for _, id := range all {
-		assert.False(t, m.routeSelector.IsSelected(id), "no exit node should be selected")
-	}
-}
-
-func TestEnforceSingleExitNode_RespectsDeselectAll(t *testing.T) {
-	m := newExitNodeTestManager()
-	m.routeSelector.DeselectAllRoutes()
-
-	m.enforceSingleExitNode("b", []route.NetID{"a", "b"})
-
-	assert.True(t, m.routeSelector.IsDeselectAll(), "global deselect-all must stay in effect")
-	assert.False(t, m.routeSelector.IsSelected("b"), "no exit node should be forced on while deselect-all is set")
-}
-
-func TestUpdateRouteSelectorFromManagement_FreshSelectsOne(t *testing.T) {
-	m := newExitNodeTestManager()
-	routes := route.HAMap{
-		"exitA|0.0.0.0/0":    {exitRoute("exitA", "p1", false)},
-		"exitB|0.0.0.0/0":    {exitRoute("exitB", "p2", false)},
-		"lan|192.168.1.0/24": {{NetID: "lan", Network: netip.MustParsePrefix("192.168.1.0/24"), Peer: "p3"}},
-		"exitC|0.0.0.0/0":    {exitRoute("exitC", "p4", false)},
-	}
-
-	m.updateRouteSelectorFromManagement(routes)
-
-	// Exactly one exit node (the deterministic first) is selected.
-	assert.True(t, m.routeSelector.IsSelected("exitA"), "exitA is the deterministic default")
-	assert.False(t, m.routeSelector.IsSelected("exitB"), "exitB must not also be selected")
-	assert.False(t, m.routeSelector.IsSelected("exitC"), "exitC must not also be selected")
-	// Non-exit routes are left at their default-on state.
-	assert.True(t, m.routeSelector.IsSelected("lan"), "non-exit route selection is untouched")
-}
-
-func TestUpdateRouteSelectorFromManagement_HonorsPersistedPick(t *testing.T) {
-	m := newExitNodeTestManager()
-	routes := route.HAMap{
-		"exitA|0.0.0.0/0": {exitRoute("exitA", "p1", false)},
-		"exitB|0.0.0.0/0": {exitRoute("exitB", "p2", false)},
-	}
-	all := []route.NetID{"exitA", "exitB"}
-
-	// Simulate the state the runtime select path leaves behind: exactly one
-	// exit node explicitly selected, its sibling deselected.
-	require.NoError(t, m.routeSelector.SelectRoutes([]route.NetID{"exitB"}, true, all))
-	require.NoError(t, m.routeSelector.DeselectRoutes([]route.NetID{"exitA"}, all))
-
-	m.updateRouteSelectorFromManagement(routes)
-
-	assert.True(t, m.routeSelector.IsSelected("exitB"), "persisted pick must stay selected")
-	assert.False(t, m.routeSelector.IsSelected("exitA"), "the other exit node stays deselected")
-}
-
-func TestUpdateRouteSelectorFromManagement_OptOutKeepsNone(t *testing.T) {
-	m := newExitNodeTestManager()
-	routes := route.HAMap{
-		"exitA|0.0.0.0/0": {exitRoute("exitA", "p1", false)},
-		"exitB|0.0.0.0/0": {exitRoute("exitB", "p2", false)},
-	}
-	all := []route.NetID{"exitA", "exitB"}
-
-	// User deselected exit nodes and selected none.
-	require.NoError(t, m.routeSelector.DeselectRoutes(all, all))
-
-	m.updateRouteSelectorFromManagement(routes)
-
-	assert.False(t, m.routeSelector.IsSelected("exitA"), "opt-out keeps exitA off")
-	assert.False(t, m.routeSelector.IsSelected("exitB"), "opt-out keeps exitB off")
-}
-
-func TestUpdateRouteSelectorFromManagement_NoAutoApplySelectsNone(t *testing.T) {
-	m := newExitNodeTestManager()
-	// SkipAutoApply=true: management offers the exit nodes but doesn't request
-	// auto-activation, so none should be selected until the user picks one.
-	routes := route.HAMap{
-		"exitA|0.0.0.0/0": {exitRoute("exitA", "p1", true)},
-		"exitB|0.0.0.0/0": {exitRoute("exitB", "p2", true)},
-	}
-
-	m.updateRouteSelectorFromManagement(routes)
-
-	assert.False(t, m.routeSelector.IsSelected("exitA"), "no auto-apply keeps exitA off")
-	assert.False(t, m.routeSelector.IsSelected("exitB"), "no auto-apply keeps exitB off")
-}

client/internal/routemanager/manager.go

@@ -701,13 +701,7 @@ func resolveURLsToIPs(urls []string) []net.IP {
 	return ips
 }
 
-// updateRouteSelectorFromManagement reconciles exit-node selection on every
-// network map: it keeps at most one exit node selected — the user's persisted
-// pick, else whatever management marks for auto-apply (SkipAutoApply=false),
-// else none. We never auto-activate an exit node the map doesn't request; it
-// stays off until the user picks it. Exit nodes are mutually exclusive, but the
-// RouteSelector stores routes with default-on semantics, so without this every
-// available exit node would report selected at once.
+// updateRouteSelectorFromManagement updates the route selector based on the isSelected status from the management server
 func (m *DefaultManager) updateRouteSelectorFromManagement(clientRoutes route.HAMap) {
 	m.mirrorV6ExitPairSelections(clientRoutes)
 
@@ -718,14 +712,13 @@ func (m *DefaultManager) updateRouteSelectorFromManagement(clientRoutes route.HA
 		return
 	}
 
-	info := m.collectExitNodeInfo(clientRoutes)
-	if len(info.allIDs) == 0 {
+	exitNodeInfo := m.collectExitNodeInfo(clientRoutes)
+	if len(exitNodeInfo.allIDs) == 0 {
 		return
 	}
 
-	preferred := pickPreferredExitNode(info)
-	m.enforceSingleExitNode(preferred, info.allIDs)
-	m.logExitNodeUpdate(info, preferred)
+	m.updateExitNodeSelections(exitNodeInfo)
+	m.logExitNodeUpdate(exitNodeInfo)
 }
 
 // mirrorV6ExitPairSelections keeps every synthesized "-v6" exit route's selection
@@ -753,10 +746,6 @@ type exitNodeInfo struct {
 	userDeselected       []route.NetID
 }
 
-// collectExitNodeInfo categorises the available exit nodes by their persisted
-// selection state. It keys on the base (v4) NetID and skips the synthesized
-// "-v6" partner, which inherits its base's selection through the RouteSelector
-// — counting it separately would double-count the pair.
 func (m *DefaultManager) collectExitNodeInfo(clientRoutes route.HAMap) exitNodeInfo {
 	var info exitNodeInfo
 
@@ -766,9 +755,6 @@ func (m *DefaultManager) collectExitNodeInfo(clientRoutes route.HAMap) exitNodeI
 		}
 
 		netID := haID.NetID()
-		if strings.HasSuffix(string(netID), route.V6ExitSuffix) {
-			continue
-		}
 		info.allIDs = append(info.allIDs, netID)
 
 		if m.routeSelector.HasUserSelectionForRoute(netID) {
@@ -805,52 +791,45 @@ func (m *DefaultManager) checkManagementSelection(routes []*route.Route, netID r
 	}
 }
 
-// pickPreferredExitNode chooses the single exit node to keep selected. In order:
-//   - a persisted user selection wins (deterministic if several survive from
-//     legacy state, so the set self-heals down to one);
-//   - otherwise activate only what management marks for auto-apply
-//     (SkipAutoApply=false); the lexicographically first if it marks several.
-//
-// Returns "" when neither holds — we never force an arbitrary exit node on. A
-// route the map doesn't auto-apply stays off until the user selects it.
-// info.userDeselected is informational only: an explicit deselect simply keeps
-// that route out of both lists above, so it can't be picked.
-func pickPreferredExitNode(info exitNodeInfo) route.NetID {
-	if len(info.userSelected) > 0 {
-		return minNetID(info.userSelected)
-	}
-	if len(info.selectedByManagement) > 0 {
-		return minNetID(info.selectedByManagement)
-	}
-	return ""
+func (m *DefaultManager) updateExitNodeSelections(info exitNodeInfo) {
+	routesToDeselect := m.getRoutesToDeselect(info.allIDs)
+	m.deselectExitNodes(routesToDeselect)
+	m.selectExitNodesByManagement(info.selectedByManagement, info.allIDs)
 }
 
-// enforceSingleExitNode makes preferred the only selected exit node: every other
-// available exit node is deselected and preferred (if any) is selected, without
-// disturbing non-exit route selections. The whole reconciliation runs under a
-// single RouteSelector lock (SetExclusiveExitNode) so a concurrent deselect-all
-// cannot interleave and get undone; a global deselect-all is left untouched so
-// the user's "all off" stays in effect.
-func (m *DefaultManager) enforceSingleExitNode(preferred route.NetID, allIDs []route.NetID) {
-	m.routeSelector.SetExclusiveExitNode(preferred, allIDs)
-}
-
-func (m *DefaultManager) logExitNodeUpdate(info exitNodeInfo, preferred route.NetID) {
-	log.Debugf("Exit node selection: %d available, preferred=%q (%d user-selected, %d user-deselected, %d management-selected)",
-		len(info.allIDs), preferred, len(info.userSelected), len(info.userDeselected), len(info.selectedByManagement))
-}
-
-// minNetID returns the lexicographically smallest NetID, for a deterministic
-// default pick that stays stable across restarts.
-func minNetID(ids []route.NetID) route.NetID {
-	if len(ids) == 0 {
-		return ""
-	}
-	best := ids[0]
-	for _, id := range ids[1:] {
-		if id < best {
-			best = id
+func (m *DefaultManager) getRoutesToDeselect(allIDs []route.NetID) []route.NetID {
+	var routesToDeselect []route.NetID
+	for _, netID := range allIDs {
+		if !m.routeSelector.HasUserSelectionForRoute(netID) {
+			routesToDeselect = append(routesToDeselect, netID)
 		}
 	}
-	return best
+	return routesToDeselect
+}
+
+func (m *DefaultManager) deselectExitNodes(routesToDeselect []route.NetID) {
+	if len(routesToDeselect) == 0 {
+		return
+	}
+
+	err := m.routeSelector.DeselectRoutes(routesToDeselect, routesToDeselect)
+	if err != nil {
+		log.Warnf("Failed to deselect exit nodes: %v", err)
+	}
+}
+
+func (m *DefaultManager) selectExitNodesByManagement(selectedByManagement []route.NetID, allIDs []route.NetID) {
+	if len(selectedByManagement) == 0 {
+		return
+	}
+
+	err := m.routeSelector.SelectRoutes(selectedByManagement, true, allIDs)
+	if err != nil {
+		log.Warnf("Failed to select exit nodes: %v", err)
+	}
+}
+
+func (m *DefaultManager) logExitNodeUpdate(info exitNodeInfo) {
+	log.Debugf("Updated route selector: %d exit nodes available, %d selected by management, %d user-selected, %d user-deselected",
+		len(info.allIDs), len(info.selectedByManagement), len(info.userSelected), len(info.userDeselected))
 }

client/internal/routeselector/routeselector.go

@@ -115,38 +115,7 @@ func (rs *RouteSelector) DeselectAllRoutes() {
 	clear(rs.selectedRoutes)
 }
 
-// SetExclusiveExitNode atomically makes preferred the only selected exit node
-// among exitIDs: every other ID in exitIDs is deselected and preferred (when
-// non-empty) is selected, all under a single lock. Holding the lock across the
-// whole reconciliation prevents a concurrent DeselectAllRoutes from interleaving
-// between the deselect and select steps and being silently undone. A global
-// deselect-all is left untouched so the user's "all off" stays in effect;
-// non-exit routes are never referenced, so their selection is preserved.
-func (rs *RouteSelector) SetExclusiveExitNode(preferred route.NetID, exitIDs []route.NetID) {
-	rs.mu.Lock()
-	defer rs.mu.Unlock()
-
-	if rs.deselectAll {
-		return
-	}
-
-	for _, id := range exitIDs {
-		if id == preferred {
-			continue
-		}
-		rs.deselectedRoutes[id] = struct{}{}
-		delete(rs.selectedRoutes, id)
-	}
-
-	if preferred != "" {
-		delete(rs.deselectedRoutes, preferred)
-		rs.selectedRoutes[preferred] = struct{}{}
-	}
-}
-
-// IsDeselectAll reports whether the global "deselect all" flag is set, i.e. the
-// user explicitly disabled every route. Callers enforcing per-route invariants
-// (e.g. single exit node) should leave the selection untouched when it is.
+// IsDeselectAll reports whether the user has explicitly deselected all routes.
 func (rs *RouteSelector) IsDeselectAll() bool {
 	rs.mu.RLock()
 	defer rs.mu.RUnlock()

client/ios/NetBirdSDK/login.go

@@ -44,25 +44,10 @@ type Auth struct {
 // NewAuth instantiate Auth struct and validate the management URL
 func NewAuth(cfgPath string, mgmURL string) (*Auth, error) {
 	inputCfg := profilemanager.ConfigInput{
-		ConfigPath:    cfgPath,
 		ManagementURL: mgmURL,
 	}
 
-	// Load the existing config when a config file is already present so an
-	// interactive re-login reuses the peer's persisted WireGuard private key
-	// (and thus its identity) instead of generating a fresh one. Generating a
-	// new key registers a brand-new peer on the management server on every
-	// re-auth (named after the fallback hostname). Only fall back to a fresh
-	// in-memory config for the first-time login when no config file exists yet.
-	// DirectUpdateOrCreateConfig uses non-atomic writes so it also works inside
-	// the tvOS App Group sandbox where atomic temp-file+rename is blocked.
-	var cfg *profilemanager.Config
-	var err error
-	if cfgPath != "" {
-		cfg, err = profilemanager.DirectUpdateOrCreateConfig(inputCfg)
-	} else {
-		cfg, err = profilemanager.CreateInMemoryConfig(inputCfg)
-	}
+	cfg, err := profilemanager.CreateInMemoryConfig(inputCfg)
 	if err != nil {
 		return nil, err
 	}

client/server/mdm.go

@@ -3,7 +3,6 @@ package server
 import (
 	"context"
 	"fmt"
-	"net/url"
 	"time"
 
 	log "github.com/sirupsen/logrus"
@@ -182,37 +181,6 @@ func conflictBool(key string, p *bool) conflictCheck {
 	}
 }
 
-func canonicalURL(s string) string {
-	u, err := url.ParseRequestURI(s)
-	if err != nil {
-		return s
-	}
-	if u.Port() == "" {
-		switch u.Scheme {
-		case "https":
-			u.Host += ":443"
-		case "http":
-			u.Host += ":80"
-		}
-	}
-	return u.String()
-}
-
-// conflictURL is conflictString for URL-typed keys: both sides are
-// normalized via canonicalURL before comparison.
-func conflictURL(key, got string) conflictCheck {
-	return conflictCheck{
-		key: key,
-		check: func(pol *mdm.Policy) bool {
-			if got == "" {
-				return true
-			}
-			want, ok := pol.GetString(key)
-			return ok && canonicalURL(want) == canonicalURL(got)
-		},
-	}
-}
-
 // conflictString builds a conflictCheck for a string MDM key. An empty
 // `got` is treated as "field not set" (no override requested); otherwise
 // the check returns true only when the policy contains the key and its
@@ -288,7 +256,7 @@ func mdmManagedFieldConflicts(msg *proto.SetConfigRequest, policy *mdm.Policy) [
 	}
 
 	return resolveConflicts(policy, []conflictCheck{
-		conflictURL(mdm.KeyManagementURL, msg.ManagementUrl),
+		conflictString(mdm.KeyManagementURL, msg.ManagementUrl),
 		conflictString(mdm.KeyPreSharedKey, pskGot),
 		conflictBool(mdm.KeyRosenpassEnabled, msg.RosenpassEnabled),
 		conflictBool(mdm.KeyRosenpassPermissive, msg.RosenpassPermissive),
@@ -409,7 +377,7 @@ func loginRequestMDMConflicts(msg *proto.LoginRequest, policy *mdm.Policy) []str
 	}
 
 	return resolveConflicts(policy, []conflictCheck{
-		conflictURL(mdm.KeyManagementURL, msg.ManagementUrl),
+		conflictString(mdm.KeyManagementURL, msg.ManagementUrl),
 		conflictString(mdm.KeyPreSharedKey, pskGot),
 		conflictBool(mdm.KeyRosenpassEnabled, msg.RosenpassEnabled),
 		conflictBool(mdm.KeyRosenpassPermissive, msg.RosenpassPermissive),

client/server/setconfig_mdm_test.go

@@ -181,43 +181,6 @@ func TestSetConfig_MDMAllow_NonManagedFields(t *testing.T) {
 	require.NotNil(t, resp)
 }
 
-// TestSetConfig_MDMAllow_ManagementURLPortNormalized covers the
-// regression from discussion #6483: MDM URL without explicit port vs
-// UI echo with the parseURL-appended default port must be treated as
-// a no-op echo, not a conflict.
-func TestSetConfig_MDMAllow_ManagementURLPortNormalized(t *testing.T) {
-	tests := []struct {
-		name      string
-		mdmURL    string
-		submitURL string
-	}{
-		{"policy_no_port_submit_with_443", "https://netbird.corp.example", "https://netbird.corp.example:443"},
-		{"policy_with_443_submit_no_port", "https://netbird.corp.example:443", "https://netbird.corp.example"},
-		{"http_policy_no_port_submit_with_80", "http://netbird.corp.example", "http://netbird.corp.example:80"},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			withMDMPolicy(t, mdm.NewPolicy(map[string]any{
-				mdm.KeyManagementURL: tc.mdmURL,
-			}))
-
-			s, ctx, profName, username, _ := setupServerWithProfile(t)
-
-			rosenpassEnabled := true
-			resp, err := s.SetConfig(ctx, &proto.SetConfigRequest{
-				ProfileName:      profName,
-				Username:         username,
-				ManagementUrl:    tc.submitURL,
-				RosenpassEnabled: &rosenpassEnabled,
-			})
-
-			require.NoError(t, err, "port-normalized URL echo must not trip MDM conflict gate")
-			require.NotNil(t, resp)
-		})
-	}
-}
-
 func TestSetConfig_MDMEmpty_NoEnforcement(t *testing.T) {
 	// No MDM policy active: any field can be written.
 	withMDMPolicy(t, mdm.NewPolicy(nil))

client/system/process.go

@@ -7,7 +7,7 @@ import (
 	"os"
 	"slices"
 
-	"github.com/shirou/gopsutil/v4/process"
+	"github.com/shirou/gopsutil/v3/process"
 )
 
 // getRunningProcesses returns a list of running process paths. The context bounds the work:

client/system/process_test.go

@@ -4,7 +4,7 @@ import (
 	"context"
 	"testing"
 
-	"github.com/shirou/gopsutil/v4/process"
+	"github.com/shirou/gopsutil/v3/process"
 )
 
 func Benchmark_getRunningProcesses(b *testing.B) {

e2e/agentnetwork/chat_test.go

@@ -91,7 +91,7 @@ func availableProviders() []providerCase {
 		if region == "" {
 			region = "us-east-1"
 		}
-		ps = append(ps, providerCase{name: "bedrock", catalogID: "bedrock_api", upstream: "https://bedrock-runtime." + region + ".amazonaws.com", apiKey: k, model: "us.anthropic.claude-haiku-4-5", kind: harness.WireBedrock})
+		ps = append(ps, providerCase{name: "bedrock", catalogID: "bedrock_api", upstream: "https://bedrock-runtime." + region + ".amazonaws.com", apiKey: k, model: "us.anthropic.claude-haiku-4-5", kind: harness.WireMessages})
 	}
 	return ps
 }
@@ -224,12 +224,9 @@ func TestProvidersMatrix(t *testing.T) {
 				var c int
 				var b string
 				var cerr error
-				switch pc.kind {
-				case harness.WireVertex:
+				if pc.kind == harness.WireVertex {
 					c, b, cerr = cl.Vertex(ctx, settings.Endpoint, proxyIP, pc.project, pc.region, pc.model, "Reply with exactly: pong", sessionID)
-				case harness.WireBedrock:
-					c, b, cerr = cl.Bedrock(ctx, settings.Endpoint, proxyIP, pc.model, "Reply with exactly: pong", sessionID)
-				default:
+				} else {
 					c, b, cerr = cl.Chat(ctx, settings.Endpoint, proxyIP, pc.kind, pc.model, "Reply with exactly: pong", sessionID)
 				}
 				if cerr == nil {

e2e/agentnetwork/guardrail_test.go

@@ -1,168 +0,0 @@
-//go:build e2e
-
-package agentnetwork
-
-import (
-	"context"
-	"strings"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-
-	"github.com/netbirdio/netbird/e2e/harness"
-	"github.com/netbirdio/netbird/shared/management/http/api"
-)
-
-// catalogModel returns the normalized catalog id the proxy stamps for a
-// path-routed provider's configured model — the form the guardrail allowlist is
-// compared against (region prefix / @version stripped).
-func catalogModel(pc providerCase) string {
-	switch pc.kind {
-	case harness.WireBedrock:
-		return strings.TrimPrefix(pc.model, "us.")
-	case harness.WireVertex:
-		return strings.SplitN(pc.model, "@", 2)[0]
-	default:
-		return pc.model
-	}
-}
-
-// disallowedModel returns a valid-shaped model id for the provider that is NOT
-// the configured/allowed one, so the guardrail must reject it before the
-// request ever reaches the upstream.
-func disallowedModel(pc providerCase) string {
-	switch pc.kind {
-	case harness.WireBedrock:
-		return "us.anthropic.claude-opus-4-8"
-	case harness.WireVertex:
-		return "claude-opus-4-8@20250101"
-	default:
-		return "unlisted-model"
-	}
-}
-
-// sendModel drives one request for the given model through the provider's native
-// wire shape and returns the HTTP status.
-func sendModel(ctx context.Context, t *testing.T, cl *harness.Client, endpoint, proxyIP string, pc providerCase, model string) int {
-	t.Helper()
-	var code int
-	var err error
-	switch pc.kind {
-	case harness.WireBedrock:
-		code, _, err = cl.Bedrock(ctx, endpoint, proxyIP, model, "Reply with exactly: pong", "")
-	case harness.WireVertex:
-		code, _, err = cl.Vertex(ctx, endpoint, proxyIP, pc.project, pc.region, model, "Reply with exactly: pong", "")
-	default:
-		code, _, err = cl.Chat(ctx, endpoint, proxyIP, pc.kind, model, "Reply with exactly: pong", "")
-	}
-	require.NoError(t, err, "request must reach the proxy for %s", pc.name)
-	return code
-}
-
-// TestModelAllowlistEnforced provisions a Model Allowlist guardrail limiting each
-// path-routed provider (Bedrock, Vertex) to its configured model, then drives
-// requests over the tunnel: the allowed model returns 200 while a model outside
-// the allowlist is denied 403 by the guardrail before it reaches the upstream.
-// This is the coverage missing for #6751 — the model for these providers travels
-// in the URL path, and the allowlist must be enforced there.
-func TestModelAllowlistEnforced(t *testing.T) {
-	var providers []providerCase
-	for _, pc := range availableProviders() {
-		if pc.kind == harness.WireBedrock || pc.kind == harness.WireVertex {
-			providers = append(providers, pc)
-		}
-	}
-	if len(providers) == 0 {
-		t.Skip("no path-routed provider keys set (AWS_BEARER_TOKEN_BEDROCK / GOOGLE_VERTEX_*); source ~/.llm-keys")
-	}
-
-	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Minute)
-	defer cancel()
-
-	grp, err := srv.API().Groups.Create(ctx, api.PostApiGroupsJSONRequestBody{Name: "e2e-allowlist"})
-	require.NoError(t, err, "create group")
-	t.Cleanup(func() { _ = srv.API().Groups.Delete(context.Background(), grp.Id) })
-
-	ephemeral := false
-	sk, err := srv.API().SetupKeys.Create(ctx, api.PostApiSetupKeysJSONRequestBody{
-		Name:       "e2e-allowlist-client",
-		Type:       "reusable",
-		ExpiresIn:  86400,
-		UsageLimit: 0,
-		AutoGroups: []string{grp.Id},
-		Ephemeral:  &ephemeral,
-	})
-	require.NoError(t, err, "mint setup key")
-
-	// Providers with their configured (allowed) models; the first bootstraps the cluster.
-	ids := make([]string, 0, len(providers))
-	allowed := make([]string, 0, len(providers))
-	for i, pc := range providers {
-		req := providerRequest(pc)
-		if i == 0 {
-			req.BootstrapCluster = ptr(harness.AgentNetworkCluster)
-		}
-		prov, perr := srv.CreateProvider(ctx, req)
-		require.NoError(t, perr, "create provider %s", pc.name)
-		id := prov.Id
-		ids = append(ids, id)
-		allowed = append(allowed, catalogModel(pc))
-		t.Cleanup(func() { _ = srv.DeleteProvider(context.Background(), id) })
-	}
-
-	// Guardrail allowlisting exactly the configured models.
-	var gr api.AgentNetworkGuardrailRequest
-	gr.Name = "e2e-allowlist"
-	gr.Checks.ModelAllowlist.Enabled = true
-	gr.Checks.ModelAllowlist.Models = allowed
-	guard, err := srv.CreateGuardrail(ctx, gr)
-	require.NoError(t, err, "create guardrail")
-	t.Cleanup(func() { _ = srv.DeleteGuardrail(context.Background(), guard.Id) })
-
-	enabled := true
-	pol, err := srv.CreatePolicy(ctx, api.AgentNetworkPolicyRequest{
-		Name:                   "e2e-allowlist",
-		Enabled:                &enabled,
-		SourceGroups:           []string{grp.Id},
-		DestinationProviderIds: ids,
-		GuardrailIds:           &[]string{guard.Id},
-	})
-	require.NoError(t, err, "create policy")
-	t.Cleanup(func() { _ = srv.DeletePolicy(context.Background(), pol.Id) })
-
-	settings, err := srv.GetSettings(ctx)
-	require.NoError(t, err, "read settings for endpoint")
-	require.NotEmpty(t, settings.Endpoint, "agent-network endpoint must be assigned")
-
-	proxyToken, err := srv.CreateProxyTokenCLI(ctx, "e2e-proxy-allowlist")
-	require.NoError(t, err, "mint proxy token via CLI")
-	px, err := harness.StartProxy(ctx, srv, proxyToken)
-	require.NoError(t, err, "start proxy")
-	t.Cleanup(func() { _ = px.Terminate(context.Background()) })
-
-	cl, err := harness.StartClient(ctx, srv, sk.Key)
-	require.NoError(t, err, "start client")
-	t.Cleanup(func() { _ = cl.Terminate(context.Background()) })
-
-	require.NoError(t, cl.WaitConnected(ctx, 90*time.Second), "client must connect to management")
-	if err := cl.WaitProxyPeer(ctx, 180*time.Second); err != nil {
-		t.Fatalf("client did not see the proxy peer: %v\n=== proxy logs ===\n%s", err, px.Logs(context.Background()))
-	}
-	proxyIP, err := cl.ResolveProxyIP(ctx, settings.Endpoint)
-	require.NoError(t, err, "resolve agent-network endpoint to proxy IP")
-
-	for _, pc := range providers {
-		pc := pc
-		t.Run(pc.name, func(t *testing.T) {
-			// The admin's allowlisted model is served end to end.
-			assert.Equal(t, 200, sendModel(ctx, t, cl, settings.Endpoint, proxyIP, pc, pc.model),
-				"allowlisted model must be permitted for %s", pc.name)
-			// A model outside the allowlist is rejected by the guardrail (before
-			// the upstream), regardless of whether it is a real catalog model.
-			assert.Equal(t, 403, sendModel(ctx, t, cl, settings.Endpoint, proxyIP, pc, disallowedModel(pc)),
-				"model outside the allowlist must be denied for %s", pc.name)
-		})
-	}
-}

e2e/harness/agentnetwork.go

@@ -107,17 +107,6 @@ func (c *Combined) DeletePolicy(ctx context.Context, id string) error {
 	return anDelete(ctx, c, "/api/agent-network/policies/"+id)
 }
 
-// CreateGuardrail creates an agent-network guardrail (e.g. a model allowlist)
-// that can then be attached to a policy via its GuardrailIds.
-func (c *Combined) CreateGuardrail(ctx context.Context, req api.AgentNetworkGuardrailRequest) (api.AgentNetworkGuardrail, error) {
-	return anRequest[api.AgentNetworkGuardrail](ctx, c, http.MethodPost, "/api/agent-network/guardrails", req)
-}
-
-// DeleteGuardrail removes a guardrail by id.
-func (c *Combined) DeleteGuardrail(ctx context.Context, id string) error {
-	return anDelete(ctx, c, "/api/agent-network/guardrails/"+id)
-}
-
 // GetSettings returns the account's agent-network settings row. It exists only
 // after the first provider create bootstraps it.
 func (c *Combined) GetSettings(ctx context.Context) (api.AgentNetworkSettings, error) {

e2e/harness/client.go

@@ -194,11 +194,6 @@ const (
 	// WireVertex is the Anthropic-on-Vertex rawPredict shape: the client posts
 	// the full Vertex model path and the proxy mints the SA OAuth token.
 	WireVertex = "vertex"
-	// WireBedrock is the native AWS Bedrock InvokeModel shape: the model id
-	// travels in the URL path (/model/{id}/invoke), not the body, so the proxy
-	// routes by path. This is what a Bedrock SDK client sends and the shape the
-	// model-allowlist guardrail must enforce.
-	WireBedrock = "bedrock"
 )
 
 // Chat issues a chat-completion POST to the agent-network endpoint over the
@@ -231,17 +226,6 @@ func (cl *Client) Vertex(ctx context.Context, endpoint, proxyIP, project, region
 	return cl.post(ctx, endpoint, proxyIP, path, body, withSessionID(nil, sessionID))
 }
 
-// Bedrock issues a native AWS Bedrock InvokeModel POST over the tunnel. The
-// model id is carried in the request path (/model/{id}/invoke), so the proxy
-// routes by path; the body uses the bedrock anthropic_version rather than a
-// model field. A non-empty sessionID is sent as the universal x-session-id
-// header the proxy records.
-func (cl *Client) Bedrock(ctx context.Context, endpoint, proxyIP, model, prompt, sessionID string) (int, string, error) {
-	path := "/model/" + model + "/invoke"
-	body := fmt.Sprintf(`{"anthropic_version":"bedrock-2023-05-31","max_tokens":64,"messages":[{"role":"user","content":%q}]}`, prompt)
-	return cl.post(ctx, endpoint, proxyIP, path, body, withSessionID(nil, sessionID))
-}
-
 // withSessionID appends the x-session-id header when sessionID is non-empty.
 func withSessionID(headers []string, sessionID string) []string {
 	if sessionID == "" {

go.mod

@@ -2,7 +2,7 @@ module github.com/netbirdio/netbird
 
 go 1.25.5
 
-toolchain go1.25.12
+toolchain go1.25.11
 
 require (
 	cunicu.li/go-rosenpass v0.5.42
@@ -19,8 +19,8 @@ require (
 	github.com/spf13/cobra v1.10.2
 	github.com/spf13/pflag v1.0.9
 	github.com/vishvananda/netlink v1.3.1
-	golang.org/x/crypto v0.54.0
-	golang.org/x/sys v0.47.0
+	golang.org/x/crypto v0.50.0
+	golang.org/x/sys v0.43.0
 	golang.zx2c4.com/wireguard v0.0.0-20231211153847-12269c276173
 	golang.zx2c4.com/wireguard/wgctrl v0.0.0-20241231184526-a9ab2273dd10
 	golang.zx2c4.com/wireguard/windows v0.5.3
@@ -104,7 +104,6 @@ require (
 	github.com/redis/go-redis/v9 v9.7.3
 	github.com/rs/xid v1.3.0
 	github.com/shirou/gopsutil/v3 v3.24.4
-	github.com/shirou/gopsutil/v4 v4.25.8
 	github.com/skratchdot/open-golang v0.0.0-20200116055534-eef842397966
 	github.com/songgao/water v0.0.0-20200317203138-2b4b6d7c09d8
 	github.com/stretchr/testify v1.11.1
@@ -128,11 +127,11 @@ require (
 	goauthentik.io/api/v3 v3.2023051.3
 	golang.org/x/exp v0.0.0-20250620022241-b7579e27df2b
 	golang.org/x/mobile v0.0.0-20251113184115-a159579294ab
-	golang.org/x/mod v0.37.0
-	golang.org/x/net v0.56.0
+	golang.org/x/mod v0.34.0
+	golang.org/x/net v0.53.0
 	golang.org/x/oauth2 v0.36.0
-	golang.org/x/sync v0.22.0
-	golang.org/x/term v0.45.0
+	golang.org/x/sync v0.20.0
+	golang.org/x/term v0.42.0
 	golang.org/x/time v0.15.0
 	google.golang.org/api v0.276.0
 	gopkg.in/yaml.v3 v3.0.1
@@ -309,6 +308,7 @@ require (
 	github.com/russellhaering/goxmldsig v1.6.0 // indirect
 	github.com/ryanuber/go-glob v1.0.0 // indirect
 	github.com/rymdport/portal v0.4.2 // indirect
+	github.com/shirou/gopsutil/v4 v4.25.8 // indirect
 	github.com/shoenig/go-m1cpu v0.2.1 // indirect
 	github.com/shopspring/decimal v1.4.0 // indirect
 	github.com/spf13/cast v1.7.0 // indirect
@@ -332,8 +332,8 @@ require (
 	go.uber.org/multierr v1.11.0 // indirect
 	go.yaml.in/yaml/v2 v2.4.3 // indirect
 	golang.org/x/image v0.33.0 // indirect
-	golang.org/x/text v0.40.0 // indirect
-	golang.org/x/tools v0.47.0 // indirect
+	golang.org/x/text v0.36.0 // indirect
+	golang.org/x/tools v0.43.0 // indirect
 	golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2 // indirect
 	google.golang.org/genproto/googleapis/rpc v0.0.0-20260401024825-9d38bb4040a9 // indirect
 	gopkg.in/square/go-jose.v2 v2.6.0 // indirect

go.sum

@@ -781,8 +781,8 @@ golang.org/x/crypto v0.18.0/go.mod h1:R0j02AL6hcrfOiy9T4ZYp/rcWeMxM3L6QYxlOuEG1m
 golang.org/x/crypto v0.19.0/go.mod h1:Iy9bg/ha4yyC70EfRS8jz+B6ybOBKMaSxLj6P6oBDfU=
 golang.org/x/crypto v0.23.0/go.mod h1:CKFgDieR+mRhux2Lsu27y0fO304Db0wZe70UKqHu0v8=
 golang.org/x/crypto v0.31.0/go.mod h1:kDsLvtWBEx7MV9tJOj9bnXsPbxwJQ6csT/x4KIN4Ssk=
-golang.org/x/crypto v0.54.0 h1:YLIA59K4fiNzHzjnZt2tUJQjQtUWfWbeHBqKtk3eScw=
-golang.org/x/crypto v0.54.0/go.mod h1:KWL8ny2AZdGR2cWmzeHrp2azQPGogOv+HeQaVEXC2dk=
+golang.org/x/crypto v0.50.0 h1:zO47/JPrL6vsNkINmLoo/PH1gcxpls50DNogFvB5ZGI=
+golang.org/x/crypto v0.50.0/go.mod h1:3muZ7vA7PBCE6xgPX7nkzzjiUq87kRItoJQM1Yo8S+Q=
 golang.org/x/exp v0.0.0-20250620022241-b7579e27df2b h1:M2rDM6z3Fhozi9O7NWsxAkg/yqS/lQJ6PmkyIV3YP+o=
 golang.org/x/exp v0.0.0-20250620022241-b7579e27df2b/go.mod h1:3//PLf8L/X+8b4vuAfHzxeRUl04Adcb341+IGKfnqS8=
 golang.org/x/image v0.33.0 h1:LXRZRnv1+zGd5XBUVRFmYEphyyKJjQjCRiOuAP3sZfQ=
@@ -799,8 +799,8 @@ golang.org/x/mod v0.8.0/go.mod h1:iBbtSCu2XBx23ZKBPSOrRkjjQPZFPuis4dIYUhu/chs=
 golang.org/x/mod v0.12.0/go.mod h1:iBbtSCu2XBx23ZKBPSOrRkjjQPZFPuis4dIYUhu/chs=
 golang.org/x/mod v0.15.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
 golang.org/x/mod v0.17.0/go.mod h1:hTbmBsO62+eylJbnUtE2MGJUyE7QWk4xUqPFrRgJ+7c=
-golang.org/x/mod v0.37.0 h1:vF1DjpVEshcIqoEaauuHebaLk1O1forxjxBaVn884JQ=
-golang.org/x/mod v0.37.0/go.mod h1:m8S8VeM9r4dzDwjrKO0a1sZP3YjeMamRRlD+fmR2Q/0=
+golang.org/x/mod v0.34.0 h1:xIHgNUUnW6sYkcM5Jleh05DvLOtwc6RitGHbDk4akRI=
+golang.org/x/mod v0.34.0/go.mod h1:ykgH52iCZe79kzLLMhyCUzhMci+nQj+0XkbXpNYtVjY=
 golang.org/x/net v0.0.0-20180906233101-161cd47e91fd/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190603091049-60506f45cf65/go.mod h1:HSz+uSET+XFnRR8LxR5pz3Of3rY3CfYBVs4xY44aLks=
@@ -819,8 +819,8 @@ golang.org/x/net v0.15.0/go.mod h1:idbUs1IY1+zTqbi8yxTbhexhEEk5ur9LInksu6HrEpk=
 golang.org/x/net v0.20.0/go.mod h1:z8BVo6PvndSri0LbOE3hAn0apkU+1YvI6E70E9jsnvY=
 golang.org/x/net v0.21.0/go.mod h1:bIjVDfnllIU7BJ2DNgfnXvpSvtn8VRwhlsaeUTyUS44=
 golang.org/x/net v0.25.0/go.mod h1:JkAGAh7GEvH74S6FOH42FLoXpXbE/aqXSrIQjXgsiwM=
-golang.org/x/net v0.56.0 h1:Rw8j/hFzGvJUZwNBXnAtf5sVDVt+65SK2C7IxCxZt5o=
-golang.org/x/net v0.56.0/go.mod h1:D3Ku6r+V6JROoZK144D2XfMHFcMq/0zSfLelVTCFKec=
+golang.org/x/net v0.53.0 h1:d+qAbo5L0orcWAr0a9JweQpjXF19LMXJE8Ey7hwOdUA=
+golang.org/x/net v0.53.0/go.mod h1:JvMuJH7rrdiCfbeHoo3fCQU24Lf5JJwT9W3sJFulfgs=
 golang.org/x/oauth2 v0.8.0/go.mod h1:yr7u4HXZRm1R1kBWqr/xKNqewf0plRYoB7sla+BCIXE=
 golang.org/x/oauth2 v0.36.0 h1:peZ/1z27fi9hUOFCAZaHyrpWG5lwe0RJEEEeH0ThlIs=
 golang.org/x/oauth2 v0.36.0/go.mod h1:YDBUJMTkDnJS+A4BP4eZBjCqtokkg1hODuPjwiGPO7Q=
@@ -835,8 +835,8 @@ golang.org/x/sync v0.3.0/go.mod h1:FU7BRWz2tNW+3quACPkgCx/L+uEAv1htQ0V83Z9Rj+Y=
 golang.org/x/sync v0.6.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
 golang.org/x/sync v0.7.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
 golang.org/x/sync v0.10.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
-golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
-golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+golang.org/x/sync v0.20.0 h1:e0PTpb7pjO8GAtTs2dQ6jYa5BWYlMuX047Dco/pItO4=
+golang.org/x/sync v0.20.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.0.0-20180909124046-d0be0721c37e/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
@@ -872,8 +872,8 @@ golang.org/x/sys v0.17.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 golang.org/x/sys v0.19.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 golang.org/x/sys v0.20.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 golang.org/x/sys v0.28.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
-golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/sys v0.43.0 h1:Rlag2XtaFTxp19wS8MXlJwTvoh8ArU6ezoyFsMyCTNI=
+golang.org/x/sys v0.43.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
 golang.org/x/telemetry v0.0.0-20240228155512-f48c80bd79b2/go.mod h1:TeRTkGYfJXctD9OcfyVLyj2J3IxLnKwHJR8f4D8a3YE=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
 golang.org/x/term v0.0.0-20210927222741-03fcf44c2211/go.mod h1:jbD1KX2456YbFQfuXm/mYQcufACuNUgVhRMnK/tPxf8=
@@ -886,8 +886,8 @@ golang.org/x/term v0.16.0/go.mod h1:yn7UURbUtPyrVJPGPq404EukNFxcm/foM+bV/bfcDsY=
 golang.org/x/term v0.17.0/go.mod h1:lLRBjIVuehSbZlaOtGMbcMncT+aqLLLmKrsjNrUguwk=
 golang.org/x/term v0.20.0/go.mod h1:8UkIAJTvZgivsXaD6/pH6U9ecQzZ45awqEOzuCvwpFY=
 golang.org/x/term v0.27.0/go.mod h1:iMsnZpn0cago0GOrHO2+Y7u7JPn5AylBrcoWkElMTSM=
-golang.org/x/term v0.45.0 h1:NwWyBmoJCbfTHpxrWoZ9C6/VxOf7ic219I8xZZFdrf0=
-golang.org/x/term v0.45.0/go.mod h1:9aqxs0blBcrm/n0L9QW0aRVD+ktan8ssZromtqJC43w=
+golang.org/x/term v0.42.0 h1:UiKe+zDFmJobeJ5ggPwOshJIVt6/Ft0rcfrXZDLWAWY=
+golang.org/x/term v0.42.0/go.mod h1:Dq/D+snpsbazcBG5+F9Q1n2rXV8Ma+71xEjTRufARgY=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
@@ -899,8 +899,8 @@ golang.org/x/text v0.13.0/go.mod h1:TvPlkZtksWOMsz7fbANvkp4WM8x/WCo/om8BMLbz+aE=
 golang.org/x/text v0.14.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
 golang.org/x/text v0.15.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
 golang.org/x/text v0.21.0/go.mod h1:4IBbMaMmOPCJ8SecivzSH54+73PCFmPWxNTLm+vZkEQ=
-golang.org/x/text v0.40.0 h1:Ub2Z6/xjgF1WrYQz2nuITOEegKFtiIy+rieRJ5lHZKs=
-golang.org/x/text v0.40.0/go.mod h1:hpnzDAfGV753zIKo+wk3u1bVKCGPbrnF7+7LBF/UHVY=
+golang.org/x/text v0.36.0 h1:JfKh3XmcRPqZPKevfXVpI1wXPTqbkE5f7JA92a55Yxg=
+golang.org/x/text v0.36.0/go.mod h1:NIdBknypM8iqVmPiuco0Dh6P5Jcdk8lJL0CUebqK164=
 golang.org/x/time v0.15.0 h1:bbrp8t3bGUeFOx08pvsMYRTCVSMk89u4tKbNOZbp88U=
 golang.org/x/time v0.15.0/go.mod h1:Y4YMaQmXwGQZoFaVFk4YpCt4FLQMYKZe9oeV/f4MSno=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
@@ -914,8 +914,8 @@ golang.org/x/tools v0.1.12/go.mod h1:hNGJHUnrk76NpqgfD5Aqm5Crs+Hm0VOH/i9J2+nxYbc
 golang.org/x/tools v0.6.0/go.mod h1:Xwgl3UAJ/d3gWutnCtw505GrjyAbvKui8lOU390QaIU=
 golang.org/x/tools v0.13.0/go.mod h1:HvlwmtVNQAhOuCjW7xxvovg8wbNq7LwfXh/k7wXUl58=
 golang.org/x/tools v0.21.1-0.20240508182429-e35e4ccd0d2d/go.mod h1:aiJjzUbINMkxbQROHiO6hDPo2LHcIPhhQsa9DLh0yGk=
-golang.org/x/tools v0.47.0 h1:7Kn5x/d1svx/PzryTsqeoZN4TZwqeH5pGWjefhLi/1Q=
-golang.org/x/tools v0.47.0/go.mod h1:dFHnyTvFWY212G+h7ZY4Vsp/K3U4/7W9TyVaAul8uCA=
+golang.org/x/tools v0.43.0 h1:12BdW9CeB3Z+J/I/wj34VMl8X+fEXBxVR90JeMX5E7s=
+golang.org/x/tools v0.43.0/go.mod h1:uHkMso649BX2cZK6+RpuIPXS3ho2hZo4FVwfoy1vIk0=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=

management/internals/modules/peers/manager.go

@@ -224,6 +224,30 @@ func (m *managerImpl) CreateProxyPeer(ctx context.Context, accountID string, pee
 		return nil
 	}
 
+	// Dedupe stale embedded peer records for the same (account, cluster).
+	// The proxy generates a fresh WireGuard keypair on every startup
+	// (proxy/internal/roundtrip/netbird.go), so without this sweep the
+	// prior embedded peer would linger forever — holding its CGNAT IP
+	// allocation, polluting other peers' rosters, and (most visibly)
+	// leaving the synth DNS pointing at the dead address. The
+	// (account, cluster) tuple identifies "the embedded peer for this
+	// proxy instance at this cluster"; any record matching that tuple
+	// with a different pubkey is by definition stale and must go.
+	staleIDs, err := m.findStaleEmbeddedProxyPeers(ctx, accountID, cluster, peerKey)
+	if err != nil {
+		return fmt.Errorf("scan for stale embedded proxy peers: %w", err)
+	}
+	if len(staleIDs) > 0 {
+		// userID="" + checkConnected=false: the deletion is initiated
+		// by management itself on behalf of the freshly-registering
+		// proxy, not by an end user; the stale peer may still be
+		// marked Connected from its prior session, but its session is
+		// dead by definition (its key no longer exists).
+		if err := m.DeletePeers(ctx, accountID, staleIDs, "", false); err != nil {
+			return fmt.Errorf("delete stale embedded proxy peers %v: %w", staleIDs, err)
+		}
+	}
+
 	name := fmt.Sprintf("proxy-%s", xid.New().String())
 	newPeer := &peer.Peer{
 		Ephemeral: true,
@@ -249,3 +273,29 @@ func (m *managerImpl) CreateProxyPeer(ctx context.Context, accountID string, pee
 
 	return nil
 }
+
+// findStaleEmbeddedProxyPeers returns the peer IDs of embedded proxy peer
+// records in accountID that target the same cluster but carry a different
+// WireGuard pubkey than the freshly-registering one. Used by CreateProxyPeer
+// to garbage-collect stale records left behind when the proxy restarts with a
+// regenerated keypair.
+func (m *managerImpl) findStaleEmbeddedProxyPeers(ctx context.Context, accountID, cluster, newKey string) ([]string, error) {
+	account, err := m.store.GetAccount(ctx, accountID)
+	if err != nil {
+		return nil, err
+	}
+	var stale []string
+	for _, p := range account.Peers {
+		if p == nil || !p.ProxyMeta.Embedded {
+			continue
+		}
+		if p.ProxyMeta.Cluster != cluster {
+			continue
+		}
+		if p.Key == newKey {
+			continue
+		}
+		stale = append(stale, p.ID)
+	}
+	return stale, nil
+}

management/internals/server/server.go

@@ -15,7 +15,7 @@ import (
 	"go.opentelemetry.io/otel/metric"
 	"golang.org/x/crypto/acme/autocert"
 	"golang.org/x/net/http2"
-	"golang.org/x/net/http2/h2c" //nolint:staticcheck
+	"golang.org/x/net/http2/h2c"
 	"google.golang.org/grpc"
 
 	"github.com/netbirdio/netbird/encryption"
@@ -382,7 +382,6 @@ func (s *BaseServer) serveGRPCWithHTTP(ctx context.Context, listener net.Listene
 			// the following magic is needed to support HTTP2 without TLS
 			// and still share a single port between gRPC and HTTP APIs
 			h1s := &http.Server{
-				//nolint:staticcheck // h2c also handles the HTTP/1 Upgrade mechanism, which http.Server's UnencryptedHTTP2 does not
 				Handler: h2c.NewHandler(handler, &http2.Server{}),
 			}
 			err = h1s.Serve(listener)

management/server/agentnetwork_proxypeer_restart_test.go

@@ -0,0 +1,199 @@
+package server
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/management/internals/modules/peers"
+	"github.com/netbirdio/netbird/management/internals/modules/agentnetwork"
+	agenttypes "github.com/netbirdio/netbird/management/internals/modules/agentnetwork/types"
+	nbpeer "github.com/netbirdio/netbird/management/server/peer"
+	"github.com/netbirdio/netbird/management/server/permissions"
+	"github.com/netbirdio/netbird/management/server/store"
+	"github.com/netbirdio/netbird/management/server/types"
+)
+
+// TestAgentNetwork_ProxyRestart_PropagatesNewPeerAndDropsStale is the no-mock
+// regression guard for the bug the user reported: restarting the proxy creates
+// a fresh embedded peer with a NEW WireGuard public key (the proxy generates
+// the keypair on every startup at proxy/internal/roundtrip/netbird.go:312).
+// The PRIOR embedded peer record is never deleted on management, so the
+// account accumulates a stale peer holding a stale CGNAT IP. Other peers
+// in the account either keep routing to the dead IP, or — if synth DNS
+// picks the wrong record — never see the new IP at all.
+//
+// What this test exercises (no mocks):
+//   - real SQLite test store
+//   - real DefaultAccountManager, network-map controller, peer-update channels
+//   - real peers.Manager.CreateProxyPeer path (the very method the proxy
+//     invokes over gRPC on every startup)
+//   - real agentnetwork.Manager + synth chain so the client receives a
+//     concrete DNS record that must point at the LATEST proxy peer.
+//
+// Pre-fix expected behavior (red): two embedded peers exist after the
+// "restart"; the synth DNS record points at the stale one; the client
+// receives an update reflecting the new peer but the old one lingers.
+// Post-fix expected behavior (green): exactly one embedded peer exists
+// after restart (with the new key) AND the client's network map carries
+// the synth DNS pointing at that new peer's CGNAT IP.
+func TestAgentNetwork_ProxyRestart_PropagatesNewPeerAndDropsStale(t *testing.T) {
+	am, updateManager, err := createManager(t)
+	require.NoError(t, err, "createManager must succeed")
+	ctx := context.Background()
+
+	const (
+		accountID   = "an-restart-acct"
+		adminUserID = "an-restart-admin"
+		groupAID    = "an-restart-grp-A"
+		clusterAddr = "eu.proxy.netbird.io"
+		clientKey   = "BhRPtynAAYRDy08+q4HTMsos8fs4plTP4NOSh7C1ry8="
+		// Two different proxy pubkeys — the "before" and "after" of a
+		// proxy-process restart with fresh-keypair generation.
+		proxyKey1 = "Aaaaa1aaaaYRDy08+q4HTMsos8fs4plTP4NOSh7C1ry8="
+		proxyKey2 = "Bbbbb2bbbbYRDy08+q4HTMsos8fs4plTP4NOSh7C1ry8="
+	)
+
+	// --- Account scaffold ---
+	account := newAccountWithId(ctx, accountID, adminUserID, "an-restart.test", "", "", false)
+	require.NoError(t, am.Store.SaveAccount(ctx, account))
+
+	clientPeer := &nbpeer.Peer{
+		Key:      clientKey,
+		Name:     "an-restart-client",
+		DNSLabel: "an-restart-client",
+		Meta:     nbpeer.PeerSystemMeta{Hostname: "an-restart-client", GoOS: "linux", WtVersion: "development"},
+	}
+	addedClient, _, _, _, err := am.AddPeer(ctx, "", "", adminUserID, clientPeer, false)
+	require.NoError(t, err, "AddPeer for client must succeed")
+	require.NoError(t, am.MarkPeerConnected(ctx, clientKey, accountID, time.Now().UnixNano(), &types.NetworkMap{}),
+		"MarkPeerConnected for the client peer must succeed (affected-peer fan-out skips disconnected peers)")
+
+	// Place the client in group A so the synth policy reaches it.
+	account, err = am.Store.GetAccount(ctx, accountID)
+	require.NoError(t, err)
+	account.Groups[groupAID] = &types.Group{ID: groupAID, Name: "groupA", Peers: []string{addedClient.ID}}
+	require.NoError(t, am.Store.SaveAccount(ctx, account), "SaveAccount must persist group A")
+
+	// --- Real peers + agent-network managers ---
+	permMgr := permissions.NewManager(am.Store)
+	peersMgr := peers.NewManager(am.Store, permMgr)
+	peersMgr.SetAccountManager(am)
+	peersMgr.SetNetworkMapController(am.networkMapController)
+	agentMgr := agentnetwork.NewManager(am.Store, permMgr, am, nil)
+
+	// Subscribe BEFORE any state-mutating call so we don't lose the update
+	// that contains the synth DNS record.
+	clientCh := updateManager.CreateChannel(ctx, addedClient.ID)
+	t.Cleanup(func() { updateManager.CloseChannel(ctx, addedClient.ID) })
+	drain(clientCh)
+
+	// --- First proxy startup: register peer key K1, then mark it
+	// connected. In production the proxy follows CreateProxyPeer with the
+	// regular sync stream which lands on MarkPeerConnected; the synth DNS
+	// path filters out peers that aren't Connected (types/account.go:323),
+	// so without this step no DNS record would be emitted.
+	require.NoError(t, peersMgr.CreateProxyPeer(ctx, accountID, proxyKey1, clusterAddr),
+		"first CreateProxyPeer (proxy startup) must succeed")
+
+	peer1ID, err := am.Store.GetPeerIDByKey(ctx, store.LockingStrengthNone, proxyKey1)
+	require.NoError(t, err, "proxy peer for K1 must be persisted after CreateProxyPeer")
+	require.NotEmpty(t, peer1ID)
+
+	require.NoError(t, am.MarkPeerConnected(ctx, proxyKey1, accountID, time.Now().UnixNano(), &types.NetworkMap{}),
+		"MarkPeerConnected for K1 must succeed")
+
+	account, err = am.Store.GetAccount(ctx, accountID)
+	require.NoError(t, err)
+	proxyIP1 := account.Peers[peer1ID].IP.String()
+	require.NotEmpty(t, proxyIP1, "K1 must have an assigned overlay IP")
+
+	// --- Provider + policy. CreateProvider / CreatePolicy trigger the
+	// agentnetwork reconcile which runs UpdateAccountPeers; the resulting
+	// NetworkMap delivered to the client carries the synth DNS record
+	// pointing at K1's IP. ---
+	provider, err := agentMgr.CreateProvider(ctx, adminUserID, &agenttypes.Provider{
+		AccountID:   accountID,
+		ProviderID:  "openai_api",
+		Name:        "openai-test",
+		UpstreamURL: "https://api.openai.com",
+		APIKey:      "sk-test-key",
+		Enabled:     true,
+		Models:      []agenttypes.ProviderModel{{ID: "gpt-5.4"}},
+	}, clusterAddr)
+	require.NoError(t, err, "CreateProvider must succeed")
+
+	_, err = agentMgr.CreatePolicy(ctx, adminUserID, &agenttypes.Policy{
+		AccountID:              accountID,
+		Name:                   "p1",
+		Enabled:                true,
+		SourceGroups:           []string{groupAID},
+		DestinationProviderIDs: []string{provider.ID},
+	})
+	require.NoError(t, err, "CreatePolicy must succeed")
+
+	settings, err := am.Store.GetAgentNetworkSettings(ctx, store.LockingStrengthNone, accountID)
+	require.NoError(t, err)
+	fqdn := settings.Endpoint()
+
+	rdata1 := awaitZoneRData(clientCh, clusterAddr, fqdn, true)
+	require.Equal(t, proxyIP1, rdata1,
+		"client must receive a synth DNS record pointing at K1's overlay IP after the synth path runs")
+	drain(clientCh)
+
+	// --- Proxy restart: NEW keypair K2, same account, same cluster ---
+	require.NoError(t, peersMgr.CreateProxyPeer(ctx, accountID, proxyKey2, clusterAddr),
+		"second CreateProxyPeer (proxy restart with fresh keypair) must succeed")
+
+	peer2ID, err := am.Store.GetPeerIDByKey(ctx, store.LockingStrengthNone, proxyKey2)
+	require.NoError(t, err, "proxy peer for K2 must be persisted after restart")
+	require.NotEmpty(t, peer2ID)
+
+	require.NoError(t, am.MarkPeerConnected(ctx, proxyKey2, accountID, time.Now().UnixNano(), &types.NetworkMap{}),
+		"MarkPeerConnected for K2 must succeed")
+
+	// In production the agent's sync stream pulls a fresh NetworkMap as
+	// part of its normal reconcile cadence; in this isolated test
+	// MarkPeerConnected's affected-peer fan-out can race the channel-side
+	// buffer in a way that swallows the synth-DNS-bearing update before
+	// our await reads it. Trigger an explicit account-wide fan-out so the
+	// assertion below tests what production actually delivers, not the
+	// in-test buffer race.
+	am.UpdateAccountPeers(ctx, accountID, types.UpdateReason{Resource: types.UpdateResourcePeer, Operation: types.UpdateOperationUpdate})
+
+	account, err = am.Store.GetAccount(ctx, accountID)
+	require.NoError(t, err)
+	proxyIP2 := account.Peers[peer2ID].IP.String()
+	require.NotEmpty(t, proxyIP2, "K2 must have an assigned overlay IP")
+	require.NotEqual(t, proxyIP1, proxyIP2, "K2 must get a different overlay IP than K1 (sanity)")
+
+	// CRITICAL ASSERTION 1: K1 must no longer be in the store. The SqlStore
+	// returns ("", nil) for a missing key rather than NotFound, so assert
+	// on the returned ID being empty.
+	staleID, err := am.Store.GetPeerIDByKey(ctx, store.LockingStrengthNone, proxyKey1)
+	require.NoError(t, err, "GetPeerIDByKey for a missing peer must not error")
+	assert.Empty(t, staleID,
+		"stale embedded proxy peer K1 must be removed when a new embedded peer registers for the same (account, cluster); pre-fix this assertion fails because management never cleans up the prior peer record")
+
+	// CRITICAL ASSERTION 2: exactly one embedded proxy peer remains, and it
+	// is K2.
+	account, err = am.Store.GetAccount(ctx, accountID)
+	require.NoError(t, err)
+	embeddedKeys := []string{}
+	for _, p := range account.Peers {
+		if p.ProxyMeta.Embedded {
+			embeddedKeys = append(embeddedKeys, p.Key)
+		}
+	}
+	assert.Equal(t, []string{proxyKey2}, embeddedKeys,
+		"after a proxy restart exactly one embedded proxy peer should remain — the one with the new key K2")
+
+	// CRITICAL ASSERTION 3: the synth DNS record the client receives now
+	// points at K2's IP, not K1's.
+	rdata2 := awaitZoneRData(clientCh, clusterAddr, fqdn, true)
+	assert.Equal(t, proxyIP2, rdata2,
+		"after proxy restart, the client's synth DNS record must point at the NEW embedded peer's IP, not the stale K1 IP")
+}

management/server/types/account.go

@@ -305,8 +305,7 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 			zone = &nbdns.CustomZone{
 				Domain:           dns.Fqdn(serviceDomainZone),
 				Records:          []nbdns.SimpleRecord{},
-				NonAuthoritative:     true,
-				SearchDomainDisabled: true,
+				NonAuthoritative: true,
 			}
 			zonesByApex[serviceDomainZone] = zone
 		}

proxy/internal/middleware/builtin/llm_guardrail/middleware.go

@@ -25,14 +25,6 @@ const (
 	denyCodeModel    = "llm_policy.model_blocked"
 	denyReasonModel  = "model_blocked"
 	denyMessageModel = "model is not in the policy allowlist"
-	// Deny reason used when an allowlist is configured but the request model
-	// could not be determined. URL/path-routed providers (AWS Bedrock, Google
-	// Vertex, ...) carry the model outside the JSON body, so a request shape the
-	// parser does not recognise reaches the guardrail with no model. Such a
-	// request must be denied (fail closed), never waved through.
-	denyCodeModelUnknown    = "llm_policy.model_unknown"
-	denyReasonModelUnknown  = "model_unknown"
-	denyMessageModelUnknown = "request model could not be determined for the policy allowlist"
 )
 
 // Middleware enforces the model allowlist and optionally captures the
@@ -116,37 +108,23 @@ func (m *Middleware) evaluateAllowlist(model string, modelPresent bool) *middlew
 	if len(m.cfg.ModelAllowlist) == 0 {
 		return nil
 	}
-	// Fail closed: with an allowlist configured, a request whose model the
-	// upstream parser could not extract (absent or empty) must be denied rather
-	// than allowed. This is what enforces the allowlist for URL/path-routed
-	// providers (Bedrock, Vertex, ...) whose model lives outside the JSON body.
-	if !modelPresent || normaliseModel(model) == "" {
-		return denyModel("", denyCodeModelUnknown, denyMessageModelUnknown, denyReasonModelUnknown)
+	if !modelPresent {
+		return nil
 	}
 	if m.modelInAllowlist(model) {
 		return nil
 	}
-	return denyModel(model, denyCodeModel, denyMessageModel, denyReasonModel)
-}
-
-// denyModel builds a 403 deny Output for a model-allowlist rejection. model is
-// included in the details only when non-empty.
-func denyModel(model, code, message, reason string) *middleware.Output {
-	details := map[string]string{}
-	if model != "" {
-		details["model"] = model
-	}
 	return &middleware.Output{
 		Decision:   middleware.DecisionDeny,
 		DenyStatus: 403,
 		DenyReason: &middleware.DenyReason{
-			Code:    code,
-			Message: message,
-			Details: details,
+			Code:    denyCodeModel,
+			Message: denyMessageModel,
+			Details: map[string]string{"model": model},
 		},
 		Metadata: []middleware.KV{
 			{Key: middleware.KeyLLMPolicyDecision, Value: "deny"},
-			{Key: middleware.KeyLLMPolicyReason, Value: reason},
+			{Key: middleware.KeyLLMPolicyReason, Value: denyReasonModel},
 		},
 	}
 }

proxy/internal/middleware/builtin/llm_guardrail/middleware_test.go

@@ -102,44 +102,13 @@ func TestAllowlistCaseInsensitive(t *testing.T) {
 	}
 }
 
-func TestAllowlistMissingModelKeyDenies(t *testing.T) {
-	// Fail closed: with an allowlist configured, a request whose model the
-	// parser could not extract (URL/path-routed providers such as Bedrock or
-	// Vertex whose shape wasn't recognised) must be denied, not allowed.
+func TestAllowlistMissingModelKeyAllows(t *testing.T) {
 	mw := New(Config{ModelAllowlist: []string{"gpt-4o"}})
 	out, err := mw.Invoke(context.Background(), newInput())
 	require.NoError(t, err)
-	require.NotNil(t, out)
-	assert.Equal(t, middleware.DecisionDeny, out.Decision, "absent model must be denied when an allowlist is set")
-	assert.Equal(t, 403, out.DenyStatus, "deny status must be 403")
-	require.NotNil(t, out.DenyReason, "deny reason must be populated")
-	assert.Equal(t, "llm_policy.model_unknown", out.DenyReason.Code, "deny code must be model_unknown")
+	assert.Equal(t, middleware.DecisionAllow, out.Decision, "missing model key must allow even with non-empty allowlist")
 	dec, _ := metaValue(t, out.Metadata, middleware.KeyLLMPolicyDecision)
-	assert.Equal(t, "deny", dec, "decision must be deny when model key is absent")
-	reason, _ := metaValue(t, out.Metadata, middleware.KeyLLMPolicyReason)
-	assert.Equal(t, "model_unknown", reason, "reason metadata must be model_unknown")
-}
-
-func TestAllowlistEmptyModelValueDenies(t *testing.T) {
-	// A present-but-empty model is as undeterminable as an absent one.
-	mw := New(Config{ModelAllowlist: []string{"gpt-4o"}})
-	out, err := mw.Invoke(context.Background(), newInput(
-		middleware.KV{Key: middleware.KeyLLMModel, Value: "   "},
-	))
-	require.NoError(t, err)
-	require.NotNil(t, out)
-	assert.Equal(t, middleware.DecisionDeny, out.Decision, "empty model must be denied when an allowlist is set")
-	require.NotNil(t, out.DenyReason, "deny reason must be populated")
-	assert.Equal(t, "llm_policy.model_unknown", out.DenyReason.Code, "deny code must be model_unknown")
-}
-
-func TestAllowlistEmptyListAllowsMissingModel(t *testing.T) {
-	// Without an allowlist there is nothing to enforce, so a missing model is
-	// still allowed — the fail-closed rule only applies when a list is set.
-	mw := New(Config{})
-	out, err := mw.Invoke(context.Background(), newInput())
-	require.NoError(t, err)
-	assert.Equal(t, middleware.DecisionAllow, out.Decision, "no allowlist must allow even without a model")
+	assert.Equal(t, "allow", dec, "decision must be allow when model key is absent")
 }
 
 func TestPromptCaptureDisabledEmitsNoPrompt(t *testing.T) {

proxy/internal/middleware/builtin/llm_request_parser/guardrail_allowlist_test.go

@@ -1,106 +0,0 @@
-package llm_request_parser
-
-import (
-	"context"
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-
-	"github.com/netbirdio/netbird/proxy/internal/middleware"
-	"github.com/netbirdio/netbird/proxy/internal/middleware/builtin/llm_guardrail"
-)
-
-// runParserGuardrail runs the request parser then the model-allowlist guardrail
-// in SlotOnRequest order, threading the parser's metadata into the guardrail the
-// same way the real chain does. It returns the guardrail decision so tests can
-// assert allowlist enforcement for URL/path-routed providers end to end.
-func runParserGuardrail(t *testing.T, url string, body []byte, allowlist []string) *middleware.Output {
-	t.Helper()
-	parser := newMiddleware(t)
-	parsed, err := parser.Invoke(context.Background(), &middleware.Input{
-		Slot: middleware.SlotOnRequest,
-		URL:  url,
-		Body: body,
-	})
-	require.NoError(t, err, "parser must not error")
-
-	guard := llm_guardrail.New(llm_guardrail.Config{ModelAllowlist: allowlist})
-	out, err := guard.Invoke(context.Background(), &middleware.Input{
-		Slot:     middleware.SlotOnRequest,
-		Metadata: parsed.Metadata,
-	})
-	require.NoError(t, err, "guardrail must not error")
-	require.NotNil(t, out, "guardrail must return an output")
-	return out
-}
-
-// TestModelAllowlist_URLRoutedProviders validates that the model allowlist is
-// enforced for providers whose model travels in the URL path (AWS Bedrock,
-// Google Vertex) rather than the JSON body. The "unknown action" case is the
-// regression guard for #6751: a Bedrock request shape the parser cannot map to a
-// model must fail closed under an allowlist instead of bypassing it.
-func TestModelAllowlist_URLRoutedProviders(t *testing.T) {
-	const bedrockBody = `{"anthropic_version":"bedrock-2023-05-31","messages":[{"role":"user","content":"hi"}]}`
-	const vertexBody = `{"anthropic_version":"vertex-2023-10-16","messages":[{"role":"user","content":"hi"}]}`
-
-	tests := []struct {
-		name      string
-		url       string
-		body      string
-		allowlist []string
-		decision  middleware.Decision
-		denyCode  string
-	}{
-		{
-			name:      "bedrock allowed model passes",
-			url:       "https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-haiku-4-5-v1:0/invoke",
-			body:      bedrockBody,
-			allowlist: []string{"anthropic.claude-haiku-4-5"},
-			decision:  middleware.DecisionAllow,
-		},
-		{
-			name:      "bedrock disallowed model denied",
-			url:       "https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-opus-4-8-v1:0/invoke",
-			body:      bedrockBody,
-			allowlist: []string{"anthropic.claude-haiku-4-5"},
-			decision:  middleware.DecisionDeny,
-			denyCode:  "llm_policy.model_blocked",
-		},
-		{
-			name:      "bedrock unknown action fails closed",
-			url:       "https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-opus-4-8-v1:0/some-future-action",
-			body:      bedrockBody,
-			allowlist: []string{"anthropic.claude-haiku-4-5"},
-			decision:  middleware.DecisionDeny,
-			denyCode:  "llm_policy.model_unknown",
-		},
-		{
-			name:      "vertex disallowed model denied",
-			url:       "/v1/projects/p/locations/global/publishers/anthropic/models/claude-opus-4-8@20250101:rawPredict",
-			body:      vertexBody,
-			allowlist: []string{"claude-haiku-4-5"},
-			decision:  middleware.DecisionDeny,
-			denyCode:  "llm_policy.model_blocked",
-		},
-		{
-			name:      "vertex allowed model passes",
-			url:       "/v1/projects/p/locations/global/publishers/anthropic/models/claude-haiku-4-5@20250101:rawPredict",
-			body:      vertexBody,
-			allowlist: []string{"claude-haiku-4-5"},
-			decision:  middleware.DecisionAllow,
-		},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			out := runParserGuardrail(t, tt.url, []byte(tt.body), tt.allowlist)
-			assert.Equal(t, tt.decision, out.Decision, "unexpected decision for %s", tt.name)
-			if tt.decision == middleware.DecisionDeny {
-				require.NotNil(t, out.DenyReason, "deny reason must be set for %s", tt.name)
-				assert.Equal(t, 403, out.DenyStatus, "deny status must be 403 for %s", tt.name)
-				assert.Equal(t, tt.denyCode, out.DenyReason.Code, "deny code for %s", tt.name)
-			}
-		})
-	}
-}

shared/relay/client/manager.go

@@ -30,16 +30,11 @@ type RelayTrack struct {
 	relayClient *Client
 	err         error
 	created     time.Time
-	// ready is closed once the dial started by openConnVia finishes (relayClient
-	// or err is set). Callers reusing a track wait on this instead of the track
-	// lock, so the dial never runs under rt.Lock.
-	ready chan struct{}
 }
 
 func NewRelayTrack() *RelayTrack {
 	return &RelayTrack{
 		created: time.Now(),
-		ready:   make(chan struct{}),
 	}
 }
 
@@ -331,24 +326,34 @@ func (m *Manager) openConnVia(ctx context.Context, serverAddress, peerKey string
 	// check if already has a connection to the desired relay server
 	m.relayClientsMutex.RLock()
 	rt, ok := m.relayClients[serverAddress]
-	m.relayClientsMutex.RUnlock()
 	if ok {
-		return m.openConnOnTrack(ctx, rt, peerKey)
+		rt.RLock()
+		m.relayClientsMutex.RUnlock()
+		defer rt.RUnlock()
+		if rt.err != nil {
+			return nil, rt.err
+		}
+		return rt.relayClient.OpenConn(ctx, peerKey)
 	}
+	m.relayClientsMutex.RUnlock()
 
 	// if not, establish a new connection but check it again (because changed the lock type) before starting the
 	// connection
 	m.relayClientsMutex.Lock()
 	rt, ok = m.relayClients[serverAddress]
 	if ok {
+		rt.RLock()
 		m.relayClientsMutex.Unlock()
-		return m.openConnOnTrack(ctx, rt, peerKey)
+		defer rt.RUnlock()
+		if rt.err != nil {
+			return nil, rt.err
+		}
+		return rt.relayClient.OpenConn(ctx, peerKey)
 	}
 
-	// Publish the track and release the map lock BEFORE dialing, so the dial does
-	// not run under rt.Lock (which would block RelayStates and the cleanup loop
-	// for the full dial). Concurrent callers find this track and wait on rt.ready.
+	// create a new relay client and store it in the relayClients map
 	rt = NewRelayTrack()
+	rt.Lock()
 	m.relayClients[serverAddress] = rt
 	m.relayClientsMutex.Unlock()
 
@@ -356,10 +361,8 @@ func (m *Manager) openConnVia(ctx context.Context, serverAddress, peerKey string
 	relayClient.SetTransportFallback(m.transportFallback)
 	err := relayClient.Connect(m.ctx)
 	if err != nil {
-		rt.Lock()
 		rt.err = err
 		rt.Unlock()
-		close(rt.ready)
 		m.relayClientsMutex.Lock()
 		delete(m.relayClients, serverAddress)
 		m.relayClientsMutex.Unlock()
@@ -367,34 +370,14 @@ func (m *Manager) openConnVia(ctx context.Context, serverAddress, peerKey string
 	}
 	// if connection closed then delete the relay client from the list
 	relayClient.SetOnDisconnectListener(m.onServerDisconnected)
-	rt.Lock()
 	rt.relayClient = relayClient
 	rt.Unlock()
-	close(rt.ready)
 
-	return relayClient.OpenConn(ctx, peerKey)
-}
-
-// openConnOnTrack opens a peer connection through an existing relay track,
-// waiting for the dial started by another openConnVia call to finish. It waits
-// on rt.ready rather than the track lock, so it neither holds nor contends the
-// track lock across the dial.
-func (m *Manager) openConnOnTrack(ctx context.Context, rt *RelayTrack, peerKey string) (net.Conn, error) {
-	select {
-	case <-rt.ready:
-	case <-ctx.Done():
-		return nil, ctx.Err()
+	conn, err := relayClient.OpenConn(ctx, peerKey)
+	if err != nil {
+		return nil, err
 	}
-
-	rt.RLock()
-	defer rt.RUnlock()
-	if rt.err != nil {
-		return nil, rt.err
-	}
-	if rt.relayClient == nil {
-		return nil, ErrRelayClientNotConnected
-	}
-	return rt.relayClient.OpenConn(ctx, peerKey)
+	return conn, nil
 }
 
 func (m *Manager) onServerConnected() {
@@ -493,13 +476,6 @@ func (m *Manager) cleanUpUnusedRelays() {
 			continue
 		}
 
-		// dial still in progress (openConnVia publishes the track before Connect
-		// completes and no longer holds rt.Lock during it), nothing to clean up.
-		if rt.relayClient == nil {
-			rt.Unlock()
-			continue
-		}
-
 		if time.Since(rt.created) <= m.keepUnusedServerTime {
 			rt.Unlock()
 			continue

shared/relay/client/manager_cleanup_test.go

@@ -1,60 +0,0 @@
-package client
-
-import (
-	"context"
-	"net/netip"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/require"
-)
-
-// TestCleanUpUnusedRelays_DoesNotBlockOnRealHangingDial drives a real, hanging foreign
-// relay dial and asserts cleanUpUnusedRelays does not stall behind it.
-func TestCleanUpUnusedRelays_DoesNotBlockOnRealHangingDial(t *testing.T) {
-	serverAddr := stallingRelayListener(t)
-
-	mCtx, mCancel := context.WithCancel(context.Background())
-	t.Cleanup(mCancel)
-
-	m := NewManager(mCtx, nil, "alice", 1280)
-
-	dialDone := make(chan struct{})
-	go func() {
-		defer close(dialDone)
-		_, _ = m.openConnVia(mCtx, serverAddr, "peerKey", netip.Addr{})
-	}()
-
-	// The track appears in the map once the dial is in flight.
-	require.Eventually(t, func() bool {
-		m.relayClientsMutex.RLock()
-		defer m.relayClientsMutex.RUnlock()
-		_, ok := m.relayClients[serverAddr]
-		return ok
-	}, 5*time.Second, 5*time.Millisecond, "relay dial did not start")
-
-	cleanupDone := make(chan struct{})
-	go func() {
-		defer close(cleanupDone)
-		m.cleanUpUnusedRelays()
-	}()
-
-	select {
-	case <-cleanupDone:
-	case <-time.After(2 * time.Second):
-		t.Fatal("cleanUpUnusedRelays blocked on an in-progress relay dial while holding the relay map lock")
-	}
-
-	m.relayClientsMutex.RLock()
-	_, stillTracked := m.relayClients[serverAddr]
-	m.relayClientsMutex.RUnlock()
-	require.True(t, stillTracked, "an in-progress relay dial must not be evicted by cleanup")
-
-	// Release the hanging dial so the goroutine can exit cleanly.
-	mCancel()
-	select {
-	case <-dialDone:
-	case <-time.After(5 * time.Second):
-		t.Fatal("openConnVia did not return after context cancellation")
-	}
-}

shared/relay/client/manager_relaystates_test.go

@@ -1,91 +0,0 @@
-package client
-
-import (
-	"context"
-	"net"
-	"net/netip"
-	"sync"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/require"
-)
-
-// stallingRelayListener accepts TCP connections and holds them open without ever
-// responding, so a relay handshake dialed against it blocks until its context is
-// cancelled. It returns the "rel://host:port" URL to dial.
-func stallingRelayListener(t *testing.T) string {
-	t.Helper()
-
-	ln, err := net.Listen("tcp", "127.0.0.1:0")
-	require.NoError(t, err)
-
-	var mu sync.Mutex
-	var conns []net.Conn
-	go func() {
-		for {
-			c, err := ln.Accept()
-			if err != nil {
-				return
-			}
-			mu.Lock()
-			conns = append(conns, c)
-			mu.Unlock()
-		}
-	}()
-	t.Cleanup(func() {
-		_ = ln.Close()
-		mu.Lock()
-		for _, c := range conns {
-			_ = c.Close()
-		}
-		mu.Unlock()
-	})
-
-	return "rel://" + ln.Addr().String()
-}
-
-// TestRelayStates_DoesNotBlockOnRealHangingDial is a regression test for
-// RelayStates() called by a "status -d command" hanging behind an in-progress
-// relay dial.
-func TestRelayStates_DoesNotBlockOnRealHangingDial(t *testing.T) {
-	serverAddr := stallingRelayListener(t)
-
-	mCtx, mCancel := context.WithCancel(context.Background())
-	t.Cleanup(mCancel)
-
-	m := NewManager(mCtx, nil, "alice", 1280)
-
-	dialDone := make(chan struct{})
-	go func() {
-		defer close(dialDone)
-		_, _ = m.openConnVia(mCtx, serverAddr, "peerKey", netip.Addr{})
-	}()
-
-	require.Eventually(t, func() bool {
-		m.relayClientsMutex.RLock()
-		defer m.relayClientsMutex.RUnlock()
-		_, ok := m.relayClients[serverAddr]
-		return ok
-	}, 5*time.Second, 5*time.Millisecond, "relay dial did not start")
-
-	done := make(chan []RelayConnState, 1)
-	go func() {
-		done <- m.RelayStates()
-	}()
-
-	select {
-	case states := <-done:
-		require.Empty(t, states, "a relay still being dialed carries no state and must be omitted")
-	case <-time.After(2 * time.Second):
-		t.Fatal("RelayStates blocked on a foreign relay whose Connect() is in progress")
-	}
-
-	// Release the hanging dial so the goroutine can exit cleanly.
-	mCancel()
-	select {
-	case <-dialDone:
-	case <-time.After(5 * time.Second):
-		t.Fatal("openConnVia did not return after context cancellation")
-	}
-}

signal/cmd/run.go

@@ -16,7 +16,7 @@ import (
 	"go.opentelemetry.io/otel/metric"
 	"golang.org/x/crypto/acme/autocert"
 	"golang.org/x/net/http2"
-	"golang.org/x/net/http2/h2c" //nolint:staticcheck
+	"golang.org/x/net/http2/h2c"
 
 	"github.com/netbirdio/netbird/shared/metrics"
 
@@ -281,7 +281,6 @@ func serveHTTP(httpListener net.Listener, handler http.Handler) {
 	go func() {
 		// Use h2c to support HTTP/2 without TLS (needed for gRPC)
 		h1s := &http.Server{
-			//nolint:staticcheck // h2c also handles the HTTP/1 Upgrade mechanism, which http.Server's UnencryptedHTTP2 does not
 			Handler: h2c.NewHandler(handler, &http2.Server{}),
 		}
 		err := h1s.Serve(httpListener)