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

# Conflicts:
#	client/proto/daemon.pb.go
#	management/internals/controllers/network_map/controller/controller.go
#	management/internals/controllers/network_map/nmaptest/runner.go
#	management/internals/shared/grpc/server.go
#	management/server/account.go
#	shared/management/proto/management.pb.go
#	upload-server/server/server.go

client/android/client.go

@@ -103,8 +103,7 @@ type Client struct {
 
 	stateChangeMu    sync.Mutex
 	stateChangeSubID string
-	eventSub         *peer.EventSubscription
-	// Closed to stop the watch goroutines from delivering buffered items to a
+	// Closed to stop the watch goroutine from delivering buffered ticks to a
 	// listener that has been removed or replaced. See stopStateChangeWatchLocked.
 	stateChangeDone chan struct{}
 
@@ -212,6 +211,7 @@ func (c *Client) Run(platformFiles PlatformFiles, urlOpener URLOpener, isAndroid
 	connectClient := internal.NewConnectClient(ctx, cfg, c.recorder,
 		internal.WithNetEvents(c.netMgr))
 	c.setState(cfg, cacheDir, cfgFile, connectClient)
+	connectClient.SetSyncResponsePersistence(true)
 	// This path runs the interactive SSO flow, so reaching here means the peer
 	// is authenticated again — release the latch Status() reports from. Clear
 	// only once the fresh connect client is installed: until then Status()
@@ -255,6 +255,7 @@ func (c *Client) RunWithoutLogin(platformFiles PlatformFiles, dns *DNSList, dnsR
 	connectClient := internal.NewConnectClient(ctx, cfg, c.recorder,
 		internal.WithNetEvents(c.netMgr))
 	c.setState(cfg, cacheDir, cfgFile, connectClient)
+	connectClient.SetSyncResponsePersistence(true)
 	return connectClient.RunOnAndroid(c.tunAdapter, c.iFaceDiscover, c.networkChangeListener, slices.Clone(dns.items), dnsReadyListener, stateFile, cacheDir)
 }
 
@@ -326,6 +327,19 @@ func (c *Client) NotifyNetworkChange() {
 // or "strict"; strict also anonymizes internal IP ranges, peer names, and
 // WireGuard public keys, and implies anonymize.
 func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool, anonymizeLevel string) (string, error) {
+	return c.debugBundle(platformFiles, anonymize, anonymizeLevel, true)
+}
+
+// DebugBundleFile generates a debug bundle and returns the path of the zip in
+// the cache directory instead of uploading it, so the app can hand the file to
+// the user for inspection. The caller owns the file and removes it once done;
+// the stale-bundle cleanup of later runs removes it only after a day.
+// anonymize and anonymizeLevel behave as in DebugBundle.
+func (c *Client) DebugBundleFile(platformFiles PlatformFiles, anonymize bool, anonymizeLevel string) (string, error) {
+	return c.debugBundle(platformFiles, anonymize, anonymizeLevel, false)
+}
+
+func (c *Client) debugBundle(platformFiles PlatformFiles, anonymize bool, anonymizeLevel string, upload bool) (string, error) {
 	cfg, cacheDir, cc := c.stateSnapshot()
 
 	// If the engine hasn't been started, load config from disk
@@ -341,6 +355,11 @@ func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool, anonym
 		cacheDir = platformFiles.CacheDir()
 	}
 
+	// Clear what an interrupted earlier run may have left in the cache before
+	// adding to it. Remote debug jobs write to the same directory, so anything
+	// younger than an hour is treated as possibly still in use.
+	debug.RemoveStaleBundles(cacheDir, time.Hour)
+
 	deps := debug.GeneratorDependencies{
 		InternalConfig: cfg,
 		StatusRecorder: c.recorder,
@@ -388,6 +407,9 @@ func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool, anonym
 	if err != nil {
 		return "", fmt.Errorf("generate debug bundle: %w", err)
 	}
+	if !upload {
+		return debug.ExportBundle(path)
+	}
 	defer func() {
 		if err := os.Remove(path); err != nil {
 			log.Errorf("failed to remove debug bundle file: %v", err)
@@ -484,6 +506,7 @@ func (c *Client) Networks() *NetworkArray {
 	routesMap := routeManager.GetClientRoutesWithNetID()
 	v6Merged := route.V6ExitMergeSet(routesMap)
 	resolvedDomains := c.recorder.GetResolvedDomainsStates()
+	activeRoutePeers := c.recorder.GetActiveRoutePeers()
 
 	networkArray := &NetworkArray{
 		items: make([]Network, 0),
@@ -497,7 +520,7 @@ func (c *Client) Networks() *NetworkArray {
 			continue
 		}
 
-		network := c.buildNetwork(id, routes, routeSelector.IsSelected(id), resolvedDomains, v6Merged)
+		network := c.buildNetwork(id, routes, routeSelector.IsSelected(id), resolvedDomains, v6Merged, activeRoutePeers)
 		if network == nil {
 			continue
 		}
@@ -506,14 +529,14 @@ func (c *Client) Networks() *NetworkArray {
 	return networkArray
 }
 
-func (c *Client) buildNetwork(id route.NetID, routes []*route.Route, selected bool, resolvedDomains map[domain.Domain]peer.ResolvedDomainInfo, v6Merged map[route.NetID]struct{}) *Network {
+func (c *Client) buildNetwork(id route.NetID, routes []*route.Route, selected bool, resolvedDomains map[domain.Domain]peer.ResolvedDomainInfo, v6Merged map[route.NetID]struct{}, activeRoutePeers map[route.HAUniqueID]string) *Network {
 	r := routes[0]
 	netStr := r.Network.String()
 	if r.IsDynamic() {
 		netStr = r.Domains.SafeString()
 	}
 
-	routePeer, err := c.findBestRoutePeer(routes)
+	routePeer, err := c.findBestRoutePeer(routes, activeRoutePeers)
 	if err != nil {
 		log.Errorf("could not get peer info for route %s: %v", id, err)
 		return nil
@@ -537,12 +560,9 @@ func (c *Client) buildNetwork(id route.NetID, routes []*route.Route, selected bo
 
 // findBestRoutePeer returns the peer actively routing traffic for the given
 // HA route group. Falls back to the first connected peer, then the first peer.
-func (c *Client) findBestRoutePeer(routes []*route.Route) (peer.State, error) {
-	netStr := routes[0].Network.String()
-
-	fullStatus := c.recorder.GetFullStatus()
-	for _, p := range fullStatus.Peers {
-		if _, ok := p.GetRoutes()[netStr]; ok {
+func (c *Client) findBestRoutePeer(routes []*route.Route, activeRoutePeers map[route.HAUniqueID]string) (peer.State, error) {
+	if peerKey, ok := activeRoutePeers[routes[0].GetHAUniqueID()]; ok {
+		if p, err := c.recorder.GetPeer(peerKey); err == nil {
 			return p, nil
 		}
 	}

client/android/login.go

@@ -194,12 +194,49 @@ func (a *Auth) login(urlOpener URLOpener, isAndroidTV bool) error {
 }
 
 func (a *Auth) foregroundGetTokenInfo(authClient *auth.Auth, urlOpener URLOpener, isAndroidTV bool) (*auth.TokenInfo, error) {
-	oAuthFlow, err := authClient.GetOAuthFlow(a.ctx, isAndroidTV, profileLoginHint(a.cfgPath))
+	return a.foregroundGetTokenInfoFlow(authClient, urlOpener, isAndroidTV, false)
+}
+
+// foregroundGetTokenInfoFlow runs the interactive flow. sessionExtend tells the
+// server the token will renew this peer's session rather than log a peer in, so
+// it can rule out a silent authorization the IdP could answer from an unrelated
+// account. See PKCEAuthorizationFlowRequest.
+func (a *Auth) foregroundGetTokenInfoFlow(authClient *auth.Auth, urlOpener URLOpener, isAndroidTV bool, sessionExtend bool) (*auth.TokenInfo, error) {
+	hint := profileLoginHint(a.cfgPath)
+
+	oAuthFlow, err := authClient.GetOAuthFlow(a.ctx, isAndroidTV, sessionExtend, hint)
 	if err != nil {
 		return nil, fmt.Errorf("failed to get OAuth flow: %v", err)
 	}
 
-	return runOAuthFlow(a.ctx, oAuthFlow, urlOpener, nil)
+	tokenInfo, err := runOAuthFlow(a.ctx, oAuthFlow, urlOpener, nil)
+	if err != nil {
+		return nil, err
+	}
+
+	if tokenInfo.MatchesAccount(hint) {
+		return tokenInfo, nil
+	}
+
+	// The IdP answered from a session belonging to another account. Retrying is
+	// what makes this recoverable: on a peer already registered the server would
+	// reject the token, and on a fresh one it would silently register the peer
+	// under the wrong account and bind the profile to it.
+	log.Infof("login returned an account other than the one this profile is bound to, retrying with an account prompt")
+	retryFlow := auth.RetryFlowForAccount(oAuthFlow)
+	if retryFlow == nil {
+		return tokenInfo, nil
+	}
+
+	retryToken, err := runOAuthFlow(a.ctx, retryFlow, urlOpener, nil)
+	if err != nil {
+		return nil, err
+	}
+	if !retryToken.MatchesAccount(hint) {
+		log.Warnf("login still returned a different account after the prompt, continuing with it")
+	}
+
+	return retryToken, nil
 }
 
 // profileLoginHint returns the stored account email for the profile at cfgPath.

client/android/preferences.go

@@ -46,7 +46,7 @@ func (p *Preferences) GetManagementURL() (string, error) {
 		return p.configInput.ManagementURL, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return "", err
 	}
@@ -64,7 +64,7 @@ func (p *Preferences) GetAdminURL() (string, error) {
 		return p.configInput.AdminURL, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return "", err
 	}
@@ -86,7 +86,7 @@ func (p *Preferences) HasPreSharedKey() (bool, error) {
 		return *p.configInput.PreSharedKey != "", nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -112,7 +112,7 @@ func (p *Preferences) GetRosenpassEnabled() (bool, error) {
 		return *p.configInput.RosenpassEnabled, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -133,7 +133,7 @@ func (p *Preferences) GetRosenpassPermissive() (bool, error) {
 		return *p.configInput.RosenpassPermissive, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -149,7 +149,7 @@ func (p *Preferences) GetDisableClientRoutes() (bool, error) {
 		return *p.configInput.DisableClientRoutes, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -170,7 +170,7 @@ func (p *Preferences) GetDisableServerRoutes() (bool, error) {
 		return *p.configInput.DisableServerRoutes, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -188,7 +188,7 @@ func (p *Preferences) GetDisableDNS() (bool, error) {
 		return *p.configInput.DisableDNS, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -206,7 +206,7 @@ func (p *Preferences) GetDisableFirewall() (bool, error) {
 		return *p.configInput.DisableFirewall, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -227,7 +227,7 @@ func (p *Preferences) GetServerSSHAllowed() (bool, error) {
 		return *p.configInput.ServerSSHAllowed, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -249,7 +249,7 @@ func (p *Preferences) GetEnableSSHRoot() (bool, error) {
 		return *p.configInput.EnableSSHRoot, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -271,7 +271,7 @@ func (p *Preferences) GetEnableSSHSFTP() (bool, error) {
 		return *p.configInput.EnableSSHSFTP, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -293,7 +293,7 @@ func (p *Preferences) GetEnableSSHLocalPortForwarding() (bool, error) {
 		return *p.configInput.EnableSSHLocalPortForwarding, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -315,7 +315,7 @@ func (p *Preferences) GetEnableSSHRemotePortForwarding() (bool, error) {
 		return *p.configInput.EnableSSHRemotePortForwarding, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -340,7 +340,7 @@ func (p *Preferences) GetBlockInbound() (bool, error) {
 		return *p.configInput.BlockInbound, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -358,7 +358,7 @@ func (p *Preferences) GetDisableIPv6() (bool, error) {
 		return *p.configInput.DisableIPv6, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}
@@ -377,7 +377,7 @@ func (p *Preferences) GetRemoteJobsAllowed() (bool, error) {
 		return *p.configInput.RemoteJobsAllowed, nil
 	}
 
-	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
+	cfg, err := profilemanager.ReadConfigOrDefault(p.configInput.ConfigPath)
 	if err != nil {
 		return false, err
 	}

client/android/session.go

@@ -6,13 +6,8 @@ import (
 	"context"
 	"fmt"
 
-	log "github.com/sirupsen/logrus"
-
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/auth"
-	"github.com/netbirdio/netbird/client/internal/auth/sessionwatch"
-	"github.com/netbirdio/netbird/client/internal/peer"
-	cProto "github.com/netbirdio/netbird/client/proto"
 )
 
 // StateChangeListener receives client state notifications.
@@ -21,16 +16,11 @@ import (
 // changed: connection state, the run-loop status label (e.g. NeedsLogin) or
 // the session deadline. It mirrors the daemon's SubscribeStatus stream
 // trigger — on each signal the consumer pulls the fresh values via
-// Status() / SessionExpiresAtUnix().
-//
-// OnSessionExpiring forwards the engine's session-expiry warnings, fired at
-// sessionwatch.WarningLead before the deadline and again at FinalWarningLead
-// (finalWarning true). The second one is suppressed when the user dismissed
-// the first via DismissSessionWarning. The daemon turns the same events into
-// its tray notification.
+// Status() / SessionExpiresAtUnix(). The engine arms no expiry-warning
+// timers on Android; the app schedules the warnings from the deadline it
+// reads here.
 type StateChangeListener interface {
 	OnStateChanged()
-	OnSessionExpiring(expiresAtUnix int64, leadMinutes int64, finalWarning bool)
 }
 
 // Status returns the connect run-loop's status label — the same value the
@@ -110,11 +100,11 @@ func (c *Client) SetStateChangeListener(listener StateChangeListener) {
 		return
 	}
 
-	// Both subscriptions are buffered (one pending tick, ten pending events),
-	// so unsubscribing is not enough to stop callbacks: the loops would drain
-	// what is already queued and deliver it to a listener the caller has
-	// already removed or replaced. Gate every callback on this registration's
-	// own signal, which is closed before unsubscribing.
+	// The subscription is buffered (one pending tick), so unsubscribing is
+	// not enough to stop callbacks: the loop would drain what is already
+	// queued and deliver it to a listener the caller has already removed or
+	// replaced. Gate every callback on this registration's own signal, which
+	// is closed before unsubscribing.
 	done := make(chan struct{})
 	c.stateChangeDone = done
 
@@ -133,9 +123,6 @@ func (c *Client) SetStateChangeListener(listener StateChangeListener) {
 			listener.OnStateChanged()
 		}
 	}()
-
-	c.eventSub = c.recorder.SubscribeToEvents()
-	go watchSessionWarnings(c.eventSub, listener, done)
 }
 
 // RemoveStateChangeListener unregisters the state notification listener.
@@ -145,21 +132,6 @@ func (c *Client) RemoveStateChangeListener() {
 	c.stopStateChangeWatchLocked()
 }
 
-// DismissSessionWarning records the user's "Dismiss" on the first expiry
-// warning and suppresses the final one for the current deadline. A refreshed
-// deadline re-arms both. No-op while the engine is not running.
-func (c *Client) DismissSessionWarning() {
-	cc := c.getConnectClient()
-	if cc == nil {
-		return
-	}
-	engine := cc.Engine()
-	if engine == nil {
-		return
-	}
-	engine.DismissSessionWarning()
-}
-
 // ExtendAuthSession runs the interactive SSO flow to obtain a fresh JWT and
 // asks the management server to extend the session deadline. The tunnel is
 // untouched: no resync, no reconnect. Async; the result arrives on the
@@ -201,8 +173,8 @@ func (c *Client) CancelExtendAuthSession() {
 }
 
 func (c *Client) stopStateChangeWatchLocked() {
-	// Signal first, unsubscribe second: closing the channels only stops new
-	// items, and the loops would still hand whatever is buffered to a listener
+	// Signal first, unsubscribe second: closing the channel only stops new
+	// items, and the loop would still hand whatever is buffered to a listener
 	// that is no longer registered.
 	if c.stateChangeDone != nil {
 		close(c.stateChangeDone)
@@ -212,49 +184,6 @@ func (c *Client) stopStateChangeWatchLocked() {
 		c.recorder.UnsubscribeFromStateChanges(c.stateChangeSubID)
 		c.stateChangeSubID = ""
 	}
-	if c.eventSub != nil {
-		// Closes the channel, which ends watchSessionWarnings.
-		c.recorder.UnsubscribeFromEvents(c.eventSub)
-		c.eventSub = nil
-	}
-}
-
-// watchSessionWarnings forwards the engine's session-expiry warnings to the
-// listener. The event stream also carries unrelated traffic — network-map
-// updates on every sync, DNS and route errors — so everything but an
-// AUTHENTICATION event carrying the session-warning marker is dropped. Exits
-// when the subscription is closed by UnsubscribeFromEvents, or earlier when
-// done is closed — the stream buffers up to ten events, and a deregistered
-// listener must not receive the ones already queued.
-func watchSessionWarnings(sub *peer.EventSubscription, listener StateChangeListener, done <-chan struct{}) {
-	for ev := range sub.Events() {
-		select {
-		case <-done:
-			return
-		default:
-		}
-		if ev.GetCategory() != cProto.SystemEvent_AUTHENTICATION {
-			continue
-		}
-		meta := ev.GetMetadata()
-		if meta[sessionwatch.MetaSessionWarning] != "true" {
-			// Other AUTHENTICATION events exist (e.g. a deadline rejected as
-			// out of range); they carry no warning marker.
-			continue
-		}
-		deadline, err := sessionwatch.ParseExpiresAt(meta[sessionwatch.MetaSessionExpiresAt])
-		if err != nil {
-			log.Warnf("session warning event with unparsable deadline: %v", err)
-			continue
-		}
-		lead, err := sessionwatch.ParseLeadMinutes(meta[sessionwatch.MetaSessionLeadMinutes])
-		if err != nil {
-			// Informational only — the deadline above is what drives the UI.
-			lead = 0
-		}
-		listener.OnSessionExpiring(deadline.Unix(), int64(lead),
-			meta[sessionwatch.MetaSessionFinal] == "true")
-	}
 }
 
 func (c *Client) beginExtend() (context.Context, error) {
@@ -293,11 +222,13 @@ func (c *Client) extendAuthSession(ctx context.Context, urlOpener URLOpener, isA
 	}
 	defer authClient.Close()
 
-	// Passing the config path makes the flow pick up the login_hint: an extend
-	// renews the session of the account already signed in, so it must not stop to
-	// offer a choice.
+	// Passing the config path makes the flow pick up the login_hint. That alone
+	// cannot keep the IdP on this profile's account though — a hint is only a
+	// suggestion, and a silent authorization is answered from whatever session the
+	// IdP already has, which need not be this peer's when several accounts are
+	// signed in. Marking the flow as an extend lets the server rule that out.
 	a := NewAuthWithConfig(ctx, cfg, cfgPath)
-	tokenInfo, err := a.foregroundGetTokenInfo(authClient, urlOpener, isAndroidTV)
+	tokenInfo, err := a.foregroundGetTokenInfoFlow(authClient, urlOpener, isAndroidTV, true)
 	if err != nil {
 		return fmt.Errorf("interactive sso login failed: %v", err)
 	}

client/android/ssh_client.go

@@ -469,7 +469,7 @@ func (s *SSHClient) requestJWTToken(cfg *profilemanager.Config, cfgPath string)
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
 	defer cancel()
 
-	flow, err := auth.NewOAuthFlow(ctx, cfg, false, true, profileLoginHint(cfgPath))
+	flow, err := auth.NewOAuthFlow(ctx, cfg, false, true, profileLoginHint(cfgPath), false)
 	if err != nil {
 		return "", fmt.Errorf("create oauth flow: %w", err)
 	}

client/cmd/debug.go

@@ -23,7 +23,10 @@ import (
 	"github.com/netbirdio/netbird/version"
 )
 
-const errCloseConnection = "Failed to close connection: %v"
+const (
+	errCloseConnection = "Failed to close connection: %v"
+	noUpDownFlag       = "no-updown"
+)
 
 // uploadBundleURLUsage documents that an empty flag is not "no upload" but
 // "wherever this deployment says": the daemon takes the destination from the
@@ -267,13 +270,14 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 	}
 
 	stateWasDown := stat.Status != string(internal.StatusConnected) && stat.Status != string(internal.StatusConnecting)
+	noUpDown, _ := cmd.Flags().GetBool(noUpDownFlag)
 
 	initialLogLevel, err := client.GetLogLevel(cmd.Context(), &proto.GetLogLevelRequest{})
 	if err != nil {
 		return fmt.Errorf("failed to get log level: %v", status.Convert(err).Message())
 	}
 
-	if stateWasDown {
+	if stateWasDown && !noUpDown {
 		if _, err := client.Up(cmd.Context(), &proto.UpRequest{}); err != nil {
 			cmd.PrintErrf("Failed to bring service up: %v\n", status.Convert(err).Message())
 		} else {
@@ -294,34 +298,20 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 	}
 
 	needsRestoreUp := false
-	if _, err := client.Down(cmd.Context(), &proto.DownRequest{}); err != nil {
-		cmd.PrintErrf("Failed to bring service down: %v\n", status.Convert(err).Message())
+	if noUpDown {
+		enableSyncResponsePersistence(cmd, client)
 	} else {
-		needsRestoreUp = !stateWasDown
-		cmd.Println("netbird down")
+		needsRestoreUp = restartDaemon(cmd, client, stateWasDown)
 	}
 
-	time.Sleep(1 * time.Second)
-
-	// Enable sync response persistence before bringing the service up
-	if _, err := client.SetSyncResponsePersistence(cmd.Context(), &proto.SetSyncResponsePersistenceRequest{
-		Enabled: true,
-	}); err != nil {
-		cmd.PrintErrf("Failed to enable sync response persistence: %v\n", status.Convert(err).Message())
-	}
-
-	if _, err := client.Up(cmd.Context(), &proto.UpRequest{}); err != nil {
-		cmd.PrintErrf("Failed to bring service up: %v\n", status.Convert(err).Message())
-	} else {
-		needsRestoreUp = false
-		cmd.Println("netbird up")
-	}
-
-	time.Sleep(3 * time.Second)
-
 	cpuProfilingStarted := false
 	if _, err := client.StartCPUProfile(cmd.Context(), &proto.StartCPUProfileRequest{}); err != nil {
-		cmd.PrintErrf("Failed to start CPU profiling: %v\n", err)
+		if msg := status.Convert(err).Message(); strings.Contains(msg, "already in progress") {
+			cmd.PrintErrln("CPU profiling is already running (started with `netbird debug cpu start`). " +
+				"It is left running and is included in a bundle created after `netbird debug cpu stop`.")
+		} else {
+			cmd.PrintErrf("Failed to start CPU profiling: %v\n", msg)
+		}
 	} else {
 		cpuProfilingStarted = true
 		defer func() {
@@ -412,7 +402,7 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 		}
 	}
 
-	if stateWasDown {
+	if stateWasDown && !noUpDown {
 		if _, err := client.Down(cmd.Context(), &proto.DownRequest{}); err != nil {
 			cmd.PrintErrf("Failed to restore service down state: %v\n", status.Convert(err).Message())
 		} else {
@@ -469,6 +459,47 @@ func setSyncResponsePersistence(cmd *cobra.Command, args []string) error {
 	return nil
 }
 
+// enableSyncResponsePersistence asks the daemon to keep the latest sync
+// response so the bundle carries the network map. With a running daemon only
+// syncs received after the call are kept.
+func enableSyncResponsePersistence(cmd *cobra.Command, client proto.DaemonServiceClient) {
+	if _, err := client.SetSyncResponsePersistence(cmd.Context(), &proto.SetSyncResponsePersistenceRequest{
+		Enabled: true,
+	}); err != nil {
+		cmd.PrintErrf("Failed to enable sync response persistence: %v\n", status.Convert(err).Message())
+	}
+}
+
+// restartDaemon cycles the daemon down and up with sync response persistence
+// enabled so the bundle carries the network map. It reports whether the
+// daemon was left down although it was running before, so the caller can
+// bring it back up.
+func restartDaemon(cmd *cobra.Command, client proto.DaemonServiceClient, stateWasDown bool) bool {
+	needsRestoreUp := false
+	if _, err := client.Down(cmd.Context(), &proto.DownRequest{}); err != nil {
+		cmd.PrintErrf("Failed to bring service down: %v\n", status.Convert(err).Message())
+	} else {
+		needsRestoreUp = !stateWasDown
+		cmd.Println("netbird down")
+	}
+
+	time.Sleep(1 * time.Second)
+
+	// Enable sync response persistence before bringing the service up
+	enableSyncResponsePersistence(cmd, client)
+
+	if _, err := client.Up(cmd.Context(), &proto.UpRequest{}); err != nil {
+		cmd.PrintErrf("Failed to bring service up: %v\n", status.Convert(err).Message())
+	} else {
+		needsRestoreUp = false
+		cmd.Println("netbird up")
+	}
+
+	time.Sleep(3 * time.Second)
+
+	return needsRestoreUp
+}
+
 func waitForDurationOrCancel(ctx context.Context, duration time.Duration, cmd *cobra.Command) error {
 	ticker := time.NewTicker(1 * time.Second)
 	defer ticker.Stop()
@@ -572,4 +603,5 @@ func init() {
 	forCmd.Flags().StringVar(&uploadBundleURLFlag, "upload-bundle-url", "", uploadBundleURLUsage)
 	forCmd.Flags().BoolVar(&uploadBundleInsecureFlag, "upload-bundle-insecure", false, "Allow uploading to an http or untrusted-TLS upload server (self-hosted); requires root")
 	forCmd.Flags().Bool("capture", false, "Capture packets during the debug duration and include in bundle")
+	forCmd.Flags().Bool(noUpDownFlag, false, "Keep the daemon running instead of bringing it down and up before collecting. The bundle only includes the network map if a sync arrives during the run")
 }

client/cmd/debug_cpu.go

@@ -0,0 +1,83 @@
+package cmd
+
+import (
+	"fmt"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/spf13/cobra"
+	"google.golang.org/grpc/status"
+
+	"github.com/netbirdio/netbird/client/proto"
+)
+
+var debugCPUCmd = &cobra.Command{
+	Use:   "cpu",
+	Short: "Profile the daemon's CPU usage",
+	Long: `Starts and stops CPU profiling in the running daemon without restarting it.
+The profile is included in the next debug bundle as cpu.prof.
+
+Profiling is not time limited: it keeps running, and keeps costing CPU, until
+"netbird debug cpu stop" is run.`,
+}
+
+var debugCPUStartCmd = &cobra.Command{
+	Use:     "start",
+	Short:   "Start CPU profiling in the daemon",
+	Example: "  netbird debug cpu start",
+	Args:    cobra.NoArgs,
+	RunE:    debugCPUStart,
+}
+
+var debugCPUStopCmd = &cobra.Command{
+	Use:   "stop",
+	Short: "Stop CPU profiling in the daemon",
+	Long: `Stops CPU profiling. The captured profile stays in the daemon until the next
+debug bundle is created, which includes it as cpu.prof.`,
+	Example: "  netbird debug cpu stop && netbird debug bundle",
+	Args:    cobra.NoArgs,
+	RunE:    debugCPUStop,
+}
+
+func debugCPUStart(cmd *cobra.Command, _ []string) error {
+	conn, err := getClient(cmd)
+	if err != nil {
+		return err
+	}
+	defer func() {
+		if err := conn.Close(); err != nil {
+			log.Errorf(errCloseConnection, err)
+		}
+	}()
+
+	if _, err := proto.NewDaemonServiceClient(conn).StartCPUProfile(cmd.Context(), &proto.StartCPUProfileRequest{}); err != nil {
+		return fmt.Errorf("start CPU profiling: %v", status.Convert(err).Message())
+	}
+
+	cmd.Println("CPU profiling started and runs until stopped. Run `netbird debug cpu stop` and then `netbird debug bundle` to collect it.")
+	return nil
+}
+
+func debugCPUStop(cmd *cobra.Command, _ []string) error {
+	conn, err := getClient(cmd)
+	if err != nil {
+		return err
+	}
+	defer func() {
+		if err := conn.Close(); err != nil {
+			log.Errorf(errCloseConnection, err)
+		}
+	}()
+
+	if _, err := proto.NewDaemonServiceClient(conn).StopCPUProfile(cmd.Context(), &proto.StopCPUProfileRequest{}); err != nil {
+		return fmt.Errorf("stop CPU profiling: %v", status.Convert(err).Message())
+	}
+
+	cmd.Println("CPU profiling stopped. Run `netbird debug bundle` to include cpu.prof.")
+	return nil
+}
+
+func init() {
+	debugCPUCmd.AddCommand(debugCPUStartCmd)
+	debugCPUCmd.AddCommand(debugCPUStopCmd)
+	debugCmd.AddCommand(debugCPUCmd)
+}

client/cmd/debug_cpu_test.go

@@ -0,0 +1,164 @@
+package cmd
+
+import (
+	"bytes"
+	"context"
+	"os/user"
+	"strings"
+	"testing"
+
+	"github.com/spf13/cobra"
+	"github.com/spf13/pflag"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal"
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
+)
+
+// startDebugTestDaemon starts an in-process daemon with an isolated profile
+// directory and returns the address the CLI should dial.
+func startDebugTestDaemon(t *testing.T) string {
+	t.Helper()
+
+	tempDir := t.TempDir()
+	origDefaultProfileDir := profilemanager.DefaultConfigPathDir
+	origActiveProfileStatePath := profilemanager.ActiveProfileStatePath
+	origConfigDirOverride := profilemanager.ConfigDirOverride
+	origDaemonAddr := daemonAddr
+	t.Cleanup(func() {
+		profilemanager.DefaultConfigPathDir = origDefaultProfileDir
+		profilemanager.ActiveProfileStatePath = origActiveProfileStatePath
+		profilemanager.ConfigDirOverride = origConfigDirOverride
+		daemonAddr = origDaemonAddr
+	})
+
+	profilemanager.DefaultConfigPathDir = tempDir
+	profilemanager.ActiveProfileStatePath = tempDir + "/active_profile.json"
+	profilemanager.ConfigDirOverride = tempDir
+
+	currUser, err := user.Current()
+	require.NoError(t, err)
+	sm := profilemanager.ServiceManager{}
+	created, err := sm.AddProfile("test1", currUser.Username)
+	require.NoError(t, err)
+	require.NoError(t, sm.SetActiveProfileState(&profilemanager.ActiveProfileState{
+		ID:       created.ID,
+		Username: currUser.Username,
+	}))
+
+	ctx, cancel := context.WithCancel(internal.CtxInitState(context.Background()))
+	srv, lis := startClientDaemon(t, ctx, "", tempDir+"/config.json")
+	t.Cleanup(func() {
+		cancel()
+		srv.Stop()
+	})
+
+	return "tcp://" + lis.Addr().String()
+}
+
+// runDebugCmd runs `netbird debug <args>` against the daemon at addr and
+// returns everything the command printed.
+func runDebugCmd(addr string, args ...string) (string, error) {
+	daemonAddr = addr
+	var out bytes.Buffer
+	rootCmd.SetOut(&out)
+	rootCmd.SetErr(&out)
+	rootCmd.SetArgs(append(append([]string{"debug"}, args...), "--daemon-addr", addr, "--log-file", ""))
+	err := rootCmd.Execute()
+	rootCmd.SetOut(nil)
+	rootCmd.SetErr(nil)
+	rootCmd.SetArgs(nil)
+	resetFlags(rootCmd)
+	return out.String(), err
+}
+
+// resetFlags puts every flag of the command and its subcommands back to its
+// default so a value parsed in one run does not leak into the next in-process
+// execution.
+func resetFlags(cmd *cobra.Command) {
+	reset := func(f *pflag.Flag) {
+		// Set appends to a slice flag and would parse the "[a,b]" default
+		// text as elements, so slices are replaced instead.
+		if sv, ok := f.Value.(pflag.SliceValue); ok {
+			var def []string
+			if trimmed := strings.Trim(f.DefValue, "[]"); trimmed != "" {
+				def = strings.Split(trimmed, ",")
+			}
+			_ = sv.Replace(def)
+		} else {
+			_ = f.Value.Set(f.DefValue)
+		}
+		f.Changed = false
+	}
+	cmd.Flags().VisitAll(reset)
+	cmd.PersistentFlags().VisitAll(reset)
+	// Commands pin their writers to the buffer of the run that first used
+	// them, so a later run would print into the old buffer.
+	cmd.SetOut(nil)
+	cmd.SetErr(nil)
+	for _, sub := range cmd.Commands() {
+		resetFlags(sub)
+	}
+}
+
+// TestResetFlagsSliceDefault guards against Set("[]") on slice flags, which
+// stores a literal "[]" element instead of the empty default.
+func TestResetFlagsSliceDefault(t *testing.T) {
+	cmd := &cobra.Command{Use: "x"}
+	var env, withDefault []string
+	cmd.Flags().StringSliceVar(&env, "env", nil, "")
+	cmd.Flags().StringSliceVar(&withDefault, "names", []string{"a", "b"}, "")
+	require.NoError(t, cmd.Flags().Parse([]string{"--env", "K=V", "--names", "c"}))
+
+	resetFlags(cmd)
+
+	assert.Empty(t, env, "slice flag with no default must reset to empty")
+	assert.Equal(t, []string{"a", "b"}, withDefault, "slice flag must reset to its default")
+}
+
+func TestDebugCPUStartStop(t *testing.T) {
+	addr := startDebugTestDaemon(t)
+
+	run := func(args ...string) error {
+		_, err := runDebugCmd(addr, append([]string{"cpu"}, args...)...)
+		return err
+	}
+
+	require.Error(t, run("stop"), "stop without a running profile must fail")
+	require.NoError(t, run("start"))
+	assert.Error(t, run("start"), "second start must be rejected while profiling")
+	require.NoError(t, run("stop"))
+	assert.Error(t, run("stop"), "second stop must be rejected")
+	assert.NoError(t, run("start"), "profiling can be started again after a stop")
+	assert.NoError(t, run("stop"))
+}
+
+// TestDebugForKeepsRunningCPUProfile covers `debug for` started while a
+// profile from `debug cpu start` is running: it must say so, leave the
+// profile alone, and still create the bundle.
+func TestDebugForKeepsRunningCPUProfile(t *testing.T) {
+	addr := startDebugTestDaemon(t)
+
+	_, err := runDebugCmd(addr, "cpu", "start")
+	require.NoError(t, err)
+
+	out, err := runDebugCmd(addr, "for", "1s", "-S=false", "--no-updown")
+	require.NoError(t, err, "output: %s", out)
+	assert.Contains(t, out, "CPU profiling is already running", "the conflict must be explained")
+	assert.NotContains(t, out, "rpc error", "the raw RPC error must not reach the user")
+	assert.Contains(t, out, "Local file:", "the bundle must still be created")
+
+	_, err = runDebugCmd(addr, "cpu", "stop")
+	assert.NoError(t, err, "the profile started by the user must still be running")
+}
+
+func TestDebugForNoUpDown(t *testing.T) {
+	addr := startDebugTestDaemon(t)
+
+	out, err := runDebugCmd(addr, "for", "1s", "-S=false", "--no-updown")
+	require.NoError(t, err, "output: %s", out)
+	assert.NotContains(t, out, "netbird down", "--no-updown must not bring the daemon down")
+	assert.NotContains(t, out, "netbird up", "--no-updown must not bring the daemon up")
+	assert.Contains(t, out, "Local file:", "the bundle must still be created")
+}

client/cmd/forwarding_rules.go

@@ -1,98 +0,0 @@
-package cmd
-
-import (
-	"fmt"
-	"sort"
-
-	"github.com/spf13/cobra"
-	"google.golang.org/grpc/status"
-
-	"github.com/netbirdio/netbird/client/proto"
-)
-
-var forwardingRulesCmd = &cobra.Command{
-	Use:   "forwarding",
-	Short: "List forwarding rules",
-	Long:  `Commands to list forwarding rules.`,
-}
-
-var forwardingRulesListCmd = &cobra.Command{
-	Use:     "list",
-	Aliases: []string{"ls"},
-	Short:   "List forwarding rules",
-	Example: "  netbird forwarding list",
-	Long:    "Commands to list forwarding rules.",
-	RunE:    listForwardingRules,
-}
-
-func listForwardingRules(cmd *cobra.Command, _ []string) error {
-	conn, err := getClient(cmd)
-	if err != nil {
-		return err
-	}
-	defer conn.Close()
-
-	client := proto.NewDaemonServiceClient(conn)
-	resp, err := client.ForwardingRules(cmd.Context(), &proto.EmptyRequest{})
-	if err != nil {
-		return fmt.Errorf("failed to list network: %v", status.Convert(err).Message())
-	}
-
-	if len(resp.GetRules()) == 0 {
-		cmd.Println("No forwarding rules available.")
-		return nil
-	}
-
-	printForwardingRules(cmd, resp.GetRules())
-	return nil
-}
-
-func printForwardingRules(cmd *cobra.Command, rules []*proto.ForwardingRule) {
-	cmd.Println("Available forwarding rules:")
-
-	// Sort rules by translated address
-	sort.Slice(rules, func(i, j int) bool {
-		if rules[i].GetTranslatedAddress() != rules[j].GetTranslatedAddress() {
-			return rules[i].GetTranslatedAddress() < rules[j].GetTranslatedAddress()
-		}
-		if rules[i].GetProtocol() != rules[j].GetProtocol() {
-			return rules[i].GetProtocol() < rules[j].GetProtocol()
-		}
-
-		return getFirstPort(rules[i].GetDestinationPort()) < getFirstPort(rules[j].GetDestinationPort())
-	})
-
-	var lastIP string
-	for _, rule := range rules {
-		dPort := portToString(rule.GetDestinationPort())
-		tPort := portToString(rule.GetTranslatedPort())
-		if lastIP != rule.GetTranslatedAddress() {
-			lastIP = rule.GetTranslatedAddress()
-			cmd.Printf("\nTranslated peer: %s\n", rule.GetTranslatedHostname())
-		}
-
-		cmd.Printf("  Local %s/%s to %s:%s\n", rule.GetProtocol(), dPort, rule.GetTranslatedAddress(), tPort)
-	}
-}
-
-func getFirstPort(portInfo *proto.PortInfo) int {
-	switch v := portInfo.PortSelection.(type) {
-	case *proto.PortInfo_Port:
-		return int(v.Port)
-	case *proto.PortInfo_Range_:
-		return int(v.Range.GetStart())
-	default:
-		return 0
-	}
-}
-
-func portToString(translatedPort *proto.PortInfo) string {
-	switch v := translatedPort.PortSelection.(type) {
-	case *proto.PortInfo_Port:
-		return fmt.Sprintf("%d", v.Port)
-	case *proto.PortInfo_Range_:
-		return fmt.Sprintf("%d-%d", v.Range.GetStart(), v.Range.GetEnd())
-	default:
-		return "No port specified"
-	}
-}

client/cmd/login.go

@@ -9,8 +9,6 @@ import (
 	log "github.com/sirupsen/logrus"
 	"github.com/spf13/cobra"
 	"golang.org/x/term"
-	"google.golang.org/grpc/codes"
-	gstatus "google.golang.org/grpc/status"
 
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/auth"
@@ -145,10 +143,7 @@ func doDaemonLogin(ctx context.Context, cmd *cobra.Command, providedSetupKey str
 	err = WithBackOff(func() error {
 		var backOffErr error
 		loginResp, backOffErr = client.Login(ctx, &loginRequest)
-		if s, ok := gstatus.FromError(backOffErr); ok && (s.Code() == codes.InvalidArgument ||
-			s.Code() == codes.PermissionDenied ||
-			s.Code() == codes.NotFound ||
-			s.Code() == codes.Unimplemented) {
+		if terminalLoginError(backOffErr) {
 			loginErr = backOffErr
 			return nil
 		}
@@ -327,10 +322,28 @@ func doForegroundLogin(ctx context.Context, cmd *cobra.Command, setupKey string,
 
 	}
 
-	config, err := profilemanager.ReadConfig(configFilePath)
+	config, err := profilemanager.ReadConfigOrDefault(configFilePath)
 	if err != nil {
 		return fmt.Errorf("read config file %s: %v", configFilePath, err)
 	}
+	// Reading a config does not provision one: this login is about to dial
+	// management with the profile's identity, so mint the keys if the profile
+	// has none yet and put them on disk — a key that stayed in memory would
+	// come back different on the next run and register a second peer.
+	//
+	// Before the MDM overlay below, on purpose: the file must keep the
+	// profile's own values. The overlay is runtime-only and re-derived on
+	// every load, so persisting it would turn an enforced management URL or
+	// pre-shared key into one the user appears to own once the policy is
+	// withdrawn.
+	if generated, err := config.EnsureIdentity(); err != nil {
+		return fmt.Errorf("ensure profile identity: %v", err)
+	} else if generated {
+		if err := profilemanager.WriteOutConfig(configFilePath, config); err != nil {
+			return fmt.Errorf("write out config file %s: %v", configFilePath, err)
+		}
+	}
+
 	// CLI standalone login: profilemanager no longer auto-applies MDM,
 	// so layer in the OS-native policy here. Desktop builds construct
 	// a Loader with no fetcher — the build-tagged loadPlatform reads
@@ -412,11 +425,44 @@ func foregroundGetTokenInfo(ctx context.Context, cmd *cobra.Command, config *pro
 		hint = profileState.Email
 	}
 
-	oAuthFlow, err := auth.NewOAuthFlow(ctx, config, util.HasGraphicalSession(), false, hint)
+	oAuthFlow, err := auth.NewOAuthFlow(ctx, config, util.HasGraphicalSession(), false, hint, false)
 	if err != nil {
 		return nil, err
 	}
 
+	tokenInfo, err := runInteractiveFlow(cmd, oAuthFlow)
+	if err != nil {
+		return nil, err
+	}
+
+	if tokenInfo.MatchesAccount(hint) {
+		return tokenInfo, nil
+	}
+
+	// The IdP answered from a session belonging to another account. Retrying is
+	// what makes this recoverable: on a peer already registered the server would
+	// reject the token, and on a fresh one it would silently register the peer
+	// under the wrong account and bind the profile to it.
+	cmd.Println("The login returned a different account than this profile uses. Asking to sign in again.")
+	retryFlow := auth.RetryFlowForAccount(oAuthFlow)
+	if retryFlow == nil {
+		return tokenInfo, nil
+	}
+
+	retryToken, err := runInteractiveFlow(cmd, retryFlow)
+	if err != nil {
+		return nil, err
+	}
+	if !retryToken.MatchesAccount(hint) {
+		log.Warnf("login still returned a different account after the prompt, continuing with it")
+	}
+
+	return retryToken, nil
+}
+
+// runInteractiveFlow requests the authorization info, shows the URL to the user
+// and blocks until the token comes back.
+func runInteractiveFlow(cmd *cobra.Command, oAuthFlow auth.OAuthFlow) (*auth.TokenInfo, error) {
 	flowInfo, err := oAuthFlow.RequestAuthInfo(context.TODO())
 	if err != nil {
 		return nil, fmt.Errorf("getting a request OAuth flow info failed: %v", err)

client/cmd/root.go

@@ -20,6 +20,8 @@ import (
 	"github.com/spf13/cobra"
 	"github.com/spf13/pflag"
 	"google.golang.org/grpc"
+	"google.golang.org/grpc/codes"
+	gstatus "google.golang.org/grpc/status"
 
 	"github.com/netbirdio/netbird/client/anonymize"
 	daddr "github.com/netbirdio/netbird/client/internal/daemonaddr"
@@ -175,7 +177,6 @@ func init() {
 	rootCmd.AddCommand(versionCmd)
 	rootCmd.AddCommand(sshCmd)
 	rootCmd.AddCommand(networksCMD)
-	rootCmd.AddCommand(forwardingRulesCmd)
 	rootCmd.AddCommand(debugCmd)
 	rootCmd.AddCommand(profileCmd)
 	rootCmd.AddCommand(exposeCmd)
@@ -183,8 +184,6 @@ func init() {
 	networksCMD.AddCommand(routesListCmd)
 	networksCMD.AddCommand(routesSelectCmd, routesDeselectCmd)
 
-	forwardingRulesCmd.AddCommand(forwardingRulesListCmd)
-
 	debugCmd.AddCommand(debugBundleCmd)
 	debugCmd.AddCommand(logCmd)
 	logCmd.AddCommand(logLevelCmd)
@@ -285,6 +284,43 @@ func DialClientGRPCServer(ctx context.Context, addr string) (*grpc.ClientConn, e
 	return grpc.DialContext(ctx, target, opts...)
 }
 
+// terminalLoginError reports whether a Login failure is final, so the backoff
+// cycle stops and the caller is told what the daemon said instead of "login
+// backoff cycle failed" thirty seconds later. Retrying cannot change any of
+// these answers: the request is malformed, the caller is not allowed, the
+// target does not exist, a precondition on the daemon refuses it (the
+// update-settings kill switch, an MDM-managed field), or the method is not
+// implemented.
+//
+// Both `netbird up` and `netbird login` run Login through the backoff, and
+// they each carried their own copy of this list — which is how one of them
+// ended up retrying a refusal the other treated as final.
+func terminalLoginError(err error) bool {
+	// A successful Login reaches here with a nil error, and that is not a
+	// terminal failure. Handled explicitly rather than left to
+	// gstatus.FromError, which answers (nil, true) for a nil error and leans on
+	// Status.Code tolerating a nil receiver to come back as codes.OK.
+	if err == nil {
+		return false
+	}
+
+	s, ok := gstatus.FromError(err)
+	if !ok {
+		return false
+	}
+
+	switch s.Code() {
+	case codes.InvalidArgument,
+		codes.PermissionDenied,
+		codes.NotFound,
+		codes.FailedPrecondition,
+		codes.Unimplemented:
+		return true
+	default:
+		return false
+	}
+}
+
 // WithBackOff execute function in backoff cycle.
 func WithBackOff(bf func() error) error {
 	return backoff.RetryNotify(bf, CLIBackOffSettings, func(err error, duration time.Duration) {

client/cmd/testutil_test.go

@@ -6,9 +6,9 @@ import (
 	"testing"
 	"time"
 
-	"go.uber.org/mock/gomock"
 	"github.com/stretchr/testify/require"
 	"go.opentelemetry.io/otel"
+	"go.uber.org/mock/gomock"
 	"google.golang.org/grpc"
 
 	"github.com/netbirdio/netbird/management/server/integrations/integrated_validator/validator"
@@ -28,7 +28,6 @@ import (
 	mgmt "github.com/netbirdio/netbird/management/server"
 	"github.com/netbirdio/netbird/management/server/activity"
 	"github.com/netbirdio/netbird/management/server/groups"
-	"github.com/netbirdio/netbird/management/server/integrations/port_forwarding"
 	"github.com/netbirdio/netbird/management/server/permissions"
 	"github.com/netbirdio/netbird/management/server/settings"
 	"github.com/netbirdio/netbird/management/server/store"
@@ -124,9 +123,9 @@ func startManagement(t *testing.T, config *config.Config, testFile string) (*grp
 
 	updateManager := update_channel.NewPeersUpdateManager(metrics)
 	requestBuffer := mgmt.NewAccountRequestBuffer(ctx, store)
-	networkMapController := controller.NewController(ctx, store, metrics, updateManager, requestBuffer, mgmt.MockIntegratedValidator{}, settingsMockManager, "netbird.cloud", port_forwarding.NewControllerMock(), manager.NewEphemeralManager(store, peersmanager), config, nil)
+	networkMapController := controller.NewController(ctx, store, metrics, updateManager, requestBuffer, mgmt.MockIntegratedValidator{}, settingsMockManager, "netbird.cloud", manager.NewEphemeralManager(store, peersmanager), config, nil)
 
-	accountManager, err := mgmt.BuildManager(ctx, config, store, networkMapController, jobManager, nil, "", eventStore, nil, false, iv, metrics, port_forwarding.NewControllerMock(), settingsMockManager, permissionsManagerMock, false, cacheStore)
+	accountManager, err := mgmt.BuildManager(ctx, config, store, networkMapController, jobManager, nil, "", eventStore, nil, false, iv, metrics, settingsMockManager, permissionsManagerMock, false, cacheStore)
 	if err != nil {
 		t.Fatal(err)
 	}

client/cmd/up.go

@@ -357,9 +357,17 @@ func runInDaemonMode(ctx context.Context, cmd *cobra.Command, pm *profilemanager
 	// set the new config
 	req := setupSetConfigReq(customDNSAddressConverted, cmd, activeProf.ID.String(), username.Username)
 	if _, err := client.SetConfig(ctx, req); err != nil {
-		if st, ok := gstatus.FromError(err); ok && st.Code() == codes.Unavailable {
-			log.Warnf("setConfig method is not available in the daemon: %s", st.Message())
-		} else {
+		switch reason, refused := refusedSettingsUpdate(err); {
+		case refused:
+			// Failing here is the point: carrying on would connect while
+			// silently dropping the settings the caller asked for, since
+			// nothing further down the line applies them.
+			return fmt.Errorf("the daemon refused the settings update: %s", reason)
+		case gstatus.Code(err) == codes.Unavailable:
+			// The daemon cannot serve the method at all, which is what this
+			// code means; an older daemon without it lands here.
+			log.Warnf("the daemon did not apply the settings update: %s", gstatus.Convert(err).Message())
+		default:
 			return daemonCallError("call service setConfig method", err)
 		}
 	}
@@ -400,10 +408,7 @@ func doDaemonUp(ctx context.Context, cmd *cobra.Command, client proto.DaemonServ
 	err = WithBackOff(func() error {
 		var backOffErr error
 		loginResp, backOffErr = client.Login(ctx, loginRequest)
-		if s, ok := gstatus.FromError(backOffErr); ok && (s.Code() == codes.InvalidArgument ||
-			s.Code() == codes.PermissionDenied ||
-			s.Code() == codes.NotFound ||
-			s.Code() == codes.Unimplemented) {
+		if terminalLoginError(backOffErr) {
 			loginErr = backOffErr
 			return nil
 		}
@@ -472,6 +477,22 @@ func setSSHSetConfigFields(req *proto.SetConfigRequest, cmd *cobra.Command) {
 	}
 }
 
+// refusedSettingsUpdate reports whether err is the daemon refusing the settings
+// a request carried — the update-settings kill switch, or a field an MDM policy
+// manages — and returns the reason it gave.
+//
+// The distinction that matters is against codes.Unavailable, which means the
+// daemon cannot serve the call: that one is worth a warning, because an older
+// daemon without the method lands there and the rest of `netbird up` still
+// works. A refusal is not, because the settings would be silently dropped.
+func refusedSettingsUpdate(err error) (string, bool) {
+	st, ok := gstatus.FromError(err)
+	if !ok || st.Code() != codes.FailedPrecondition {
+		return "", false
+	}
+	return st.Message(), true
+}
+
 func setupSetConfigReq(customDNSAddressConverted []byte, cmd *cobra.Command, profileName, username string) *proto.SetConfigRequest {
 	var req proto.SetConfigRequest
 	req.ProfileName = profileName

client/cmd/up_setconfig_refusal_test.go

@@ -0,0 +1,85 @@
+package cmd
+
+import (
+	"errors"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+	"google.golang.org/grpc/codes"
+	gstatus "google.golang.org/grpc/status"
+)
+
+// A refused settings update has to fail `netbird up`, or a caller that asked
+// for a setting the daemon will not apply connects as if it had been applied.
+// The daemon being unable to serve the call is the case that stays a warning.
+func TestRefusedSettingsUpdate(t *testing.T) {
+	tests := []struct {
+		name        string
+		err         error
+		wantRefused bool
+	}{
+		{
+			name:        "the kill switch refused the change",
+			err:         gstatus.Errorf(codes.FailedPrecondition, "update settings are disabled, you cannot use this feature without update settings enabled"),
+			wantRefused: true,
+		},
+		{
+			name:        "an MDM policy manages the field",
+			err:         gstatus.Errorf(codes.FailedPrecondition, "fields managed by MDM policy: managementURL"),
+			wantRefused: true,
+		},
+		{
+			name:        "the daemon cannot serve the call",
+			err:         gstatus.Errorf(codes.Unavailable, "connection refused"),
+			wantRefused: false,
+		},
+		{
+			name:        "any other RPC failure",
+			err:         gstatus.Errorf(codes.Internal, "boom"),
+			wantRefused: false,
+		},
+		{
+			name:        "not a status error at all",
+			err:         errors.New("boom"),
+			wantRefused: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			reason, refused := refusedSettingsUpdate(tt.err)
+			require.Equal(t, tt.wantRefused, refused)
+			if tt.wantRefused {
+				require.Equal(t, gstatus.Convert(tt.err).Message(), reason, "the daemon's reason must reach the caller")
+			}
+		})
+	}
+}
+
+// Both `netbird up` and `netbird login` drive Login through the backoff cycle,
+// and a final answer has to stop it: retrying a refusal only replaces the
+// daemon's reason with "login backoff cycle failed" thirty seconds later.
+func TestTerminalLoginError(t *testing.T) {
+	tests := []struct {
+		name         string
+		err          error
+		wantTerminal bool
+	}{
+		{name: "settings refused by the kill switch", err: gstatus.Errorf(codes.FailedPrecondition, "update settings are disabled"), wantTerminal: true},
+		{name: "field managed by MDM", err: gstatus.Errorf(codes.FailedPrecondition, "fields managed by MDM policy: managementURL"), wantTerminal: true},
+		{name: "caller not allowed", err: gstatus.Errorf(codes.PermissionDenied, "nope"), wantTerminal: true},
+		{name: "malformed request", err: gstatus.Errorf(codes.InvalidArgument, "nope"), wantTerminal: true},
+		{name: "profile not found", err: gstatus.Errorf(codes.NotFound, "nope"), wantTerminal: true},
+		{name: "method missing on an older daemon", err: gstatus.Errorf(codes.Unimplemented, "nope"), wantTerminal: true},
+		{name: "daemon unreachable, worth retrying", err: gstatus.Errorf(codes.Unavailable, "connection refused"), wantTerminal: false},
+		{name: "transient internal failure", err: gstatus.Errorf(codes.Internal, "boom"), wantTerminal: false},
+		{name: "not a status error", err: errors.New("boom"), wantTerminal: false},
+		{name: "no error at all, the login succeeded", err: nil, wantTerminal: false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			require.Equal(t, tt.wantTerminal, terminalLoginError(tt.err))
+		})
+	}
+}

client/collect-licenses.sh

@@ -1,77 +0,0 @@
-#!/bin/sh
-set -eu
-
-if [ "$#" -lt 2 ]; then
-	printf '%s\n' "usage: $0 OUTPUT_DIRECTORY GOARCH..." >&2
-	exit 2
-fi
-
-repo_root=$(CDPATH= cd -- "$(dirname "$0")/.." && pwd)
-output_name=$(basename "$1")
-if [ -z "$output_name" ] || [ "$output_name" = "." ] ||
-	[ "$output_name" = ".." ] || [ "$output_name" = "/" ]; then
-	printf '%s\n' "OUTPUT_DIRECTORY must name a directory" >&2
-	exit 2
-fi
-output_parent=$(CDPATH= cd -- "$(dirname "$1")" && pwd)
-output="$output_parent/$output_name"
-shift
-modules=$(mktemp "${TMPDIR:-/tmp}/netbird-client-licenses.modules.XXXXXX")
-sorted_modules=$(mktemp "${TMPDIR:-/tmp}/netbird-client-licenses.sorted.XXXXXX")
-trap 'rm -f "$modules" "$sorted_modules"' EXIT HUP INT TERM
-
-if [ -e "$output" ] || [ -L "$output" ]; then
-	printf 'output directory already exists: %s\n' "$output" >&2
-	exit 1
-fi
-mkdir "$output"
-mkdir "$output/third_party"
-
-cp "$repo_root/LICENSE" "$output/BSD-3-Clause.txt"
-
-cd "$repo_root"
-for arch in "$@"; do
-	GOOS=${GOOS:-linux} GOARCH="$arch" CGO_ENABLED=${CGO_ENABLED:-0} \
-		go list -deps -f '{{with .Module}}{{if .Replace}}{{.Replace.Path}}{{"\t"}}{{.Replace.Version}}{{"\t"}}{{.Replace.Dir}}{{else}}{{.Path}}{{"\t"}}{{.Version}}{{"\t"}}{{.Dir}}{{end}}{{end}}' -tags load_wgnt_from_rsrc ./client >>"$modules"
-done
-LC_ALL=C sort -u "$modules" >"$sorted_modules"
-
-goroot=$(go env GOROOT)
-for term in LICENSE PATENTS; do
-	if [ ! -f "$goroot/$term" ]; then
-		printf 'missing Go standard-library term: %s\n' "$goroot/$term" >&2
-		exit 1
-	fi
-	cp "$goroot/$term" "$output/Go-$term"
-done
-
-while IFS='	' read -r module version module_dir; do
-	[ -n "$module" ] || continue
-	[ "$module" = "github.com/netbirdio/netbird" ] && continue
-
-	if [ -z "$version" ] || [ ! -d "$module_dir" ]; then
-		printf 'cannot collect terms for module %s at version %s\n' "$module" "$version" >&2
-		exit 1
-	fi
-
-	destination="$output/third_party/$module/$version"
-	mkdir -p "$destination"
-	printf 'module: %s\nversion: %s\n' "$module" "$version" >"$destination/MODULE"
-
-	found=false
-	for term in \
-		"$module_dir"/LICENSE* "$module_dir"/License* "$module_dir"/license* \
-		"$module_dir"/LICENCE* "$module_dir"/Licence* "$module_dir"/licence* \
-		"$module_dir"/COPYING* "$module_dir"/Copying* "$module_dir"/copying* \
-		"$module_dir"/NOTICE* "$module_dir"/Notice* "$module_dir"/notice* \
-		"$module_dir"/PATENTS* "$module_dir"/Patents* "$module_dir"/patents*; do
-		[ -f "$term" ] || continue
-		cp "$term" "$destination/"
-		found=true
-	done
-
-	if [ "$found" = false ]; then
-		printf 'no root license terms found for module %s at %s\n' "$module" "$module_dir" >&2
-		exit 1
-	fi
-done <"$sorted_modules"

client/embed/embed_test.go

@@ -6,8 +6,8 @@ import (
 	"testing"
 	"time"
 
-	"go.uber.org/mock/gomock"
 	"github.com/stretchr/testify/require"
+	"go.uber.org/mock/gomock"
 	"google.golang.org/grpc"
 
 	"github.com/netbirdio/netbird/management/internals/controllers/network_map/controller"
@@ -21,7 +21,6 @@ import (
 	nbcache "github.com/netbirdio/netbird/management/server/cache"
 	"github.com/netbirdio/netbird/management/server/groups"
 	"github.com/netbirdio/netbird/management/server/integrations/integrated_validator/validator"
-	"github.com/netbirdio/netbird/management/server/integrations/port_forwarding"
 	"github.com/netbirdio/netbird/management/server/job"
 	"github.com/netbirdio/netbird/management/server/permissions"
 	"github.com/netbirdio/netbird/management/server/settings"
@@ -146,8 +145,8 @@ func startManagement(t *testing.T, signalAddr string) string {
 
 	updateManager := update_channel.NewPeersUpdateManager(metrics)
 	requestBuffer := mgmt.NewAccountRequestBuffer(context.Background(), testStore)
-	networkMapController := controller.NewController(context.Background(), testStore, metrics, updateManager, requestBuffer, mgmt.MockIntegratedValidator{}, settingsMockManager, "netbird.selfhosted", port_forwarding.NewControllerMock(), manager.NewEphemeralManager(testStore, peersManager), cfg, nil)
-	accountManager, err := mgmt.BuildManager(context.Background(), cfg, testStore, networkMapController, jobManager, nil, "", eventStore, nil, false, iv, metrics, port_forwarding.NewControllerMock(), settingsMockManager, permissionsManager, false, cacheStore)
+	networkMapController := controller.NewController(context.Background(), testStore, metrics, updateManager, requestBuffer, mgmt.MockIntegratedValidator{}, settingsMockManager, "netbird.selfhosted", manager.NewEphemeralManager(testStore, peersManager), cfg, nil)
+	accountManager, err := mgmt.BuildManager(context.Background(), cfg, testStore, networkMapController, jobManager, nil, "", eventStore, nil, false, iv, metrics, settingsMockManager, permissionsManager, false, cacheStore)
 	require.NoError(t, err)
 
 	secretsManager, err := nbgrpc.NewTimeBasedAuthSecretsManager(updateManager, cfg.TURNConfig, cfg.Relay, settingsMockManager, groupsManager)

client/firewall/iptables/dnat_linux.go

@@ -8,177 +8,11 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/hashicorp/go-multierror"
 	log "github.com/sirupsen/logrus"
 
-	nberrors "github.com/netbirdio/netbird/client/errors"
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
 )
 
-func (r *family) AddDNATRule(rule firewall.ForwardRule) (firewall.Rule, error) {
-	ruleID := rule.ID()
-	if _, exists := r.rules[ruleID+dnatSuffix]; exists {
-		return rule, nil
-	}
-
-	toDestination := rule.TranslatedAddress.String()
-	switch {
-	case len(rule.TranslatedPort.Values) == 0:
-		// no translated port, use original port
-	case len(rule.TranslatedPort.Values) == 1:
-		toDestination += fmt.Sprintf(":%d", rule.TranslatedPort.Values[0])
-	case rule.TranslatedPort.IsRange && len(rule.TranslatedPort.Values) == 2:
-		// need the "/originalport" suffix to avoid dnat port randomization
-		toDestination += fmt.Sprintf(":%d-%d/%d", rule.TranslatedPort.Values[0], rule.TranslatedPort.Values[1], rule.DestinationPort.Values[0])
-	default:
-		return nil, fmt.Errorf("invalid translated port: %v", rule.TranslatedPort)
-	}
-
-	proto := strings.ToLower(string(rule.Protocol))
-
-	rules := make(map[firewall.RuleID]ruleInfo, 3)
-
-	// DNAT rule
-	dnatRule := []string{
-		"!", "-i", r.wgIface.Name(),
-		"-p", proto,
-		"-j", "DNAT",
-		"--to-destination", toDestination,
-	}
-	dnatRule = append(dnatRule, applyPort("--dport", &rule.DestinationPort)...)
-	rules[ruleID+dnatSuffix] = ruleInfo{
-		table: tableNat,
-		chain: chainRTRdr,
-		rule:  dnatRule,
-	}
-
-	// SNAT rule
-	snatRule := []string{
-		"-o", r.wgIface.Name(),
-		"-p", proto,
-		"-d", rule.TranslatedAddress.String(),
-		"-j", "MASQUERADE",
-	}
-	snatRule = append(snatRule, applyPort("--dport", &rule.TranslatedPort)...)
-	rules[ruleID+snatSuffix] = ruleInfo{
-		table: tableNat,
-		chain: chainRTNAT,
-		rule:  snatRule,
-	}
-
-	// Forward filtering rule, if fwd policy is DROP
-	forwardRule := []string{
-		"-o", r.wgIface.Name(),
-		"-p", proto,
-		"-d", rule.TranslatedAddress.String(),
-		"-j", "ACCEPT",
-	}
-	forwardRule = append(forwardRule, applyPort("--dport", &rule.TranslatedPort)...)
-	rules[ruleID+fwdSuffix] = ruleInfo{
-		table: tableFilter,
-		chain: chainRTFwdOut,
-		rule:  forwardRule,
-	}
-
-	for key, ruleInfo := range rules {
-		if err := r.iptablesClient.Append(ruleInfo.table, ruleInfo.chain, ruleInfo.rule...); err != nil {
-			r.cleanupFailedDNATAdd(rules)
-			return nil, fmt.Errorf("add rule %s: %w", key, err)
-		}
-		r.rules[key] = ruleInfo.rule
-	}
-
-	if err := r.ipFwdState.RequestForwarding(r.v6); err != nil {
-		r.cleanupFailedDNATAdd(rules)
-		return nil, fmt.Errorf("enable forwarding: %w", err)
-	}
-
-	r.updateState()
-	return rule, nil
-}
-
-// cleanupFailedDNATAdd removes the bookkeeping written by a partially applied
-// AddDNATRule before rolling back the kernel rules, so no entries remain that
-// never got a forwarding refcount. rollbackRules re-adds entries it failed to
-// remove from the kernel.
-func (r *family) cleanupFailedDNATAdd(rules map[firewall.RuleID]ruleInfo) {
-	for key := range rules {
-		delete(r.rules, key)
-	}
-	if err := r.rollbackRules(rules); err != nil {
-		log.Errorf("rollback failed: %v", err)
-	}
-}
-
-func (r *family) rollbackRules(rules map[firewall.RuleID]ruleInfo) error {
-	var merr *multierror.Error
-	for key, ruleInfo := range rules {
-		if err := r.iptablesClient.DeleteIfExists(ruleInfo.table, ruleInfo.chain, ruleInfo.rule...); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("rollback rule %s: %w", key, err))
-			// On rollback error, add to rules map for next cleanup
-			r.rules[key] = ruleInfo.rule
-		}
-	}
-	if merr != nil {
-		r.updateState()
-	}
-	return nberrors.FormatErrorOrNil(merr)
-}
-
-func (r *family) DeleteDNATRule(rule firewall.Rule) error {
-	ruleID := rule.ID()
-
-	_, hadDNAT := r.rules[ruleID+dnatSuffix]
-	_, hadSNAT := r.rules[ruleID+snatSuffix]
-	_, hadFWD := r.rules[ruleID+fwdSuffix]
-	if !hadDNAT && !hadSNAT && !hadFWD {
-		return nil
-	}
-
-	var merr *multierror.Error
-	if dnatRule, exists := r.rules[ruleID+dnatSuffix]; exists {
-		if err := r.iptablesClient.Delete(tableNat, chainRTRdr, dnatRule...); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("delete DNAT rule: %w", err))
-		} else {
-			delete(r.rules, ruleID+dnatSuffix)
-		}
-	}
-
-	if snatRule, exists := r.rules[ruleID+snatSuffix]; exists {
-		if err := r.iptablesClient.Delete(tableNat, chainRTNAT, snatRule...); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("delete SNAT rule: %w", err))
-		} else {
-			delete(r.rules, ruleID+snatSuffix)
-		}
-	}
-
-	if fwdRule, exists := r.rules[ruleID+fwdSuffix]; exists {
-		if err := r.iptablesClient.Delete(tableFilter, chainRTFwdOut, fwdRule...); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("delete forward rule: %w", err))
-		} else {
-			delete(r.rules, ruleID+fwdSuffix)
-		}
-	}
-
-	// Release the refcount only once all rules are gone from the kernel. On
-	// partial failure the failed entries stay in r.rules so a retry can remove
-	// them and release then.
-	if merr == nil {
-		r.releaseForwarding()
-	}
-
-	r.updateState()
-
-	return nberrors.FormatErrorOrNil(merr)
-}
-
-// releaseForwarding drops one IP forwarding reference, logging any error.
-func (r *family) releaseForwarding() {
-	if err := r.ipFwdState.ReleaseForwarding(r.v6); err != nil {
-		log.Errorf("release IP forwarding: %v", err)
-	}
-}
-
 func (r *family) AddInboundDNAT(localAddr netip.Addr, protocol firewall.Protocol, originalPort, translatedPort uint16) error {
 	ruleID := firewall.RuleID(fmt.Sprintf("inbound-dnat-%s-%s-%d-%d", localAddr.String(), protocol, originalPort, translatedPort))
 

client/firewall/iptables/dnat_refcount_linux_test.go

@@ -1,240 +0,0 @@
-//go:build privileged
-
-package iptables
-
-import (
-	"net/netip"
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-
-	fw "github.com/netbirdio/netbird/client/firewall/manager"
-	"github.com/netbirdio/netbird/client/iface"
-	"github.com/netbirdio/netbird/client/iface/wgaddr"
-)
-
-func iptRefcountIfaceV4() *iFaceMock {
-	return &iFaceMock{
-		NameFunc: func() string { return "wt-refcount" },
-		AddressFunc: func() wgaddr.Address {
-			return wgaddr.Address{
-				IP:      netip.MustParseAddr("10.20.0.1"),
-				Network: netip.MustParsePrefix("10.20.0.0/24"),
-			}
-		},
-	}
-}
-
-func iptRefcountIfaceDual() *iFaceMock {
-	return &iFaceMock{
-		NameFunc: func() string { return "wt-refcount" },
-		AddressFunc: func() wgaddr.Address {
-			return wgaddr.Address{
-				IP:      netip.MustParseAddr("10.20.0.1"),
-				Network: netip.MustParsePrefix("10.20.0.0/24"),
-				IPv6:    netip.MustParseAddr("fd00::1"),
-				IPv6Net: netip.MustParsePrefix("fd00::/64"),
-			}
-		},
-	}
-}
-
-func newIptRefcountManager(t *testing.T, dual bool) *Manager {
-	t.Helper()
-	var ifMock *iFaceMock
-	if dual {
-		ifMock = iptRefcountIfaceDual()
-	} else {
-		ifMock = iptRefcountIfaceV4()
-	}
-	m, err := Create(ifMock, iface.DefaultMTU)
-	require.NoError(t, err, "create manager")
-	require.NoError(t, m.Init(nil), "init manager")
-	t.Cleanup(func() {
-		require.NoError(t, m.Close(nil), "close manager")
-	})
-	return m
-}
-
-func iptDnatV4(port uint16) fw.ForwardRule {
-	return fw.ForwardRule{
-		Protocol:          fw.ProtocolTCP,
-		DestinationPort:   fw.Port{Values: []uint16{port}},
-		TranslatedAddress: netip.MustParseAddr("10.20.0.2"),
-		TranslatedPort:    fw.Port{Values: []uint16{80}},
-	}
-}
-
-func iptDnatV6(port uint16) fw.ForwardRule {
-	return fw.ForwardRule{
-		Protocol:          fw.ProtocolTCP,
-		DestinationPort:   fw.Port{Values: []uint16{port}},
-		TranslatedAddress: netip.MustParseAddr("fd00::2"),
-		TranslatedPort:    fw.Port{Values: []uint16{80}},
-	}
-}
-
-// TestIptablesRouting_RepeatedEnableSingleReference verifies that EnableRouting
-// (called on every network-map update) holds at most one reference per family
-// and a single DisableRouting drops both back to zero.
-func TestIptablesRouting_RepeatedEnableSingleReference(t *testing.T) {
-	m := newIptRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	require.NoError(t, m.EnableRouting(), "first enable")
-	require.NoError(t, m.EnableRouting(), "second enable")
-	require.NoError(t, m.EnableRouting(), "third enable")
-	v4, v6 := state.Counts()
-	assert.Equal(t, 1, v4, "repeated enable holds a single v4 reference")
-	assert.Equal(t, 1, v6, "repeated enable holds a single v6 reference")
-
-	require.NoError(t, m.DisableRouting(), "disable")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4, "single disable releases the v4 reference")
-	assert.Equal(t, 0, v6, "single disable releases the v6 reference")
-}
-
-// TestIptablesRouting_DisableKeepsDNATReference verifies that an unpaired
-// DisableRouting does not release references held by active DNAT rules.
-func TestIptablesRouting_DisableKeepsDNATReference(t *testing.T) {
-	m := newIptRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	r1, err := m.AddDNATRule(iptDnatV6(9095))
-	require.NoError(t, err, "add v6 dnat")
-
-	require.NoError(t, m.DisableRouting(), "unpaired disable")
-	_, v6 := state.Counts()
-	assert.Equal(t, 1, v6, "DNAT-held reference survives unpaired DisableRouting")
-
-	require.NoError(t, m.DeleteDNATRule(r1), "delete v6 dnat")
-	_, v6 = state.Counts()
-	assert.Equal(t, 0, v6, "delete releases the DNAT reference")
-}
-
-// TestIptablesDNAT_RefcountBalancedV4 covers a Balanced Add/Delete pair on v4.
-func TestIptablesDNAT_RefcountBalancedV4(t *testing.T) {
-	m := newIptRefcountManager(t, false)
-	state := m.family4.ipFwdState
-
-	r1, err := m.AddDNATRule(iptDnatV4(7081))
-	require.NoError(t, err, "add v4 dnat 1")
-	v4, v6 := state.Counts()
-	assert.Equal(t, 1, v4, "v4 refcount after first add")
-	assert.Equal(t, 0, v6, "v6 refcount unchanged")
-
-	r2, err := m.AddDNATRule(iptDnatV4(7082))
-	require.NoError(t, err, "add v4 dnat 2")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 2, v4, "v4 refcount after second add")
-	assert.Equal(t, 0, v6, "v6 refcount unchanged")
-
-	require.NoError(t, m.DeleteDNATRule(r1))
-	v4, v6 = state.Counts()
-	assert.Equal(t, 1, v4, "v4 refcount after first delete")
-	assert.Equal(t, 0, v6, "v6 refcount unchanged")
-
-	require.NoError(t, m.DeleteDNATRule(r2))
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4, "v4 refcount after second delete")
-	assert.Equal(t, 0, v6, "v6 refcount unchanged")
-}
-
-// TestIptablesDNAT_RefcountBalancedV6 checks the v6 path increments v6 only and
-// decrements back to zero.
-func TestIptablesDNAT_RefcountBalancedV6(t *testing.T) {
-	m := newIptRefcountManager(t, true)
-	require.NotNil(t, m.family6, "v6 family")
-	require.Same(t, m.family4.ipFwdState, m.family6.ipFwdState, "shared state")
-	state := m.family4.ipFwdState
-
-	r1, err := m.AddDNATRule(iptDnatV6(9081))
-	require.NoError(t, err, "add v6 dnat 1")
-	v4, v6 := state.Counts()
-	assert.Equal(t, 0, v4)
-	assert.Equal(t, 1, v6, "v6 refcount after first add")
-
-	r2, err := m.AddDNATRule(iptDnatV6(9082))
-	require.NoError(t, err, "add v6 dnat 2")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4, "v4 refcount unchanged")
-	assert.Equal(t, 2, v6, "v6 refcount after second add")
-
-	require.NoError(t, m.DeleteDNATRule(r1))
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4, "v4 refcount unchanged")
-	assert.Equal(t, 1, v6, "v6 refcount after first delete")
-
-	require.NoError(t, m.DeleteDNATRule(r2))
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4)
-	assert.Equal(t, 0, v6, "v6 refcount after second delete")
-}
-
-// TestIptablesDNAT_DuplicateAddNoLeak verifies the duplicate-rule path returns
-// without bumping the refcount.
-func TestIptablesDNAT_DuplicateAddNoLeak(t *testing.T) {
-	m := newIptRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	rule := iptDnatV4(7083)
-	r1, err := m.AddDNATRule(rule)
-	require.NoError(t, err)
-	v4, _ := state.Counts()
-	assert.Equal(t, 1, v4)
-
-	_, err = m.AddDNATRule(rule)
-	require.NoError(t, err, "duplicate add")
-	v4, _ = state.Counts()
-	assert.Equal(t, 1, v4, "duplicate add must not increment")
-
-	require.NoError(t, m.DeleteDNATRule(r1))
-	v4, _ = state.Counts()
-	assert.Equal(t, 0, v4, "single delete must drop to zero")
-}
-
-// TestIptablesDNAT_DeleteMissingNoUnderflow verifies Delete on an unknown rule
-// neither errors nor releases the refcount.
-func TestIptablesDNAT_DeleteMissingNoUnderflow(t *testing.T) {
-	m := newIptRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	phantom := iptDnatV4(7099)
-	require.NoError(t, m.DeleteDNATRule(&phantom), "delete missing v4")
-	v4, v6 := state.Counts()
-	assert.Equal(t, 0, v4)
-	assert.Equal(t, 0, v6)
-
-	phantom6 := iptDnatV6(9099)
-	require.NoError(t, m.DeleteDNATRule(&phantom6), "delete missing v6")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4)
-	assert.Equal(t, 0, v6)
-
-	r1, err := m.AddDNATRule(iptDnatV4(7100))
-	require.NoError(t, err)
-	v4, _ = state.Counts()
-	assert.Equal(t, 1, v4, "real add still increments after phantom delete")
-	require.NoError(t, m.DeleteDNATRule(r1))
-}
-
-// TestIptablesDNAT_DoubleDeleteNoUnderflow verifies a second Delete on the same
-// rule is a no-op.
-func TestIptablesDNAT_DoubleDeleteNoUnderflow(t *testing.T) {
-	m := newIptRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	r1, err := m.AddDNATRule(iptDnatV6(9083))
-	require.NoError(t, err)
-	_, v6 := state.Counts()
-	assert.Equal(t, 1, v6)
-
-	require.NoError(t, m.DeleteDNATRule(r1), "first delete")
-	_, v6 = state.Counts()
-	assert.Equal(t, 0, v6)
-
-	require.NoError(t, m.DeleteDNATRule(r1), "second delete must be no-op")
-	_, v6 = state.Counts()
-	assert.Equal(t, 0, v6, "double delete must not underflow")
-}

client/firewall/iptables/family_linux.go

@@ -24,6 +24,7 @@ const (
 	tableFilter = "filter"
 	tableNat    = "nat"
 	tableMangle = "mangle"
+	tableRaw    = "raw"
 
 	// chainACLInput is the peer ACL chain that holds installed
 	// peer-filtering rules.
@@ -34,6 +35,7 @@ const (
 	mangleForwardKey chainKey = "MANGLE-FORWARD"
 
 	chainInput       = "INPUT"
+	chainOutput      = "OUTPUT"
 	chainPostrouting = "POSTROUTING"
 	chainPrerouting  = "PREROUTING"
 	chainForward     = "FORWARD"
@@ -54,10 +56,6 @@ const (
 	markManglePost = "mark-mangle-post"
 	matchSet       = "--match-set"
 
-	dnatSuffix firewall.RuleID = "_dnat"
-	snatSuffix firewall.RuleID = "_snat"
-	fwdSuffix  firewall.RuleID = "_fwd"
-
 	// ipv4TCPHeaderSize is the minimum IPv4 (20) + TCP (20) header size for MSS calculation.
 	ipv4TCPHeaderSize = 40
 	// ipv6TCPHeaderSize is the minimum IPv6 (40) + TCP (20) header size for MSS calculation.

client/firewall/iptables/filter_linux.go

@@ -81,15 +81,6 @@ func (r *family) hasRule(id nbid.RuleID) bool {
 	return ok
 }
 
-// hasDNATRule reports whether this family owns the DNAT rule set for
-// the given user id. DNAT rules live in r.rules under the well-known
-// "<id>_dnat" key; the lookup here is used by Manager.DeleteDNATRule
-// to pick the right family.
-func (r *family) hasDNATRule(id firewall.RuleID) bool {
-	_, ok := r.rules[id+dnatSuffix]
-	return ok
-}
-
 // DeleteFilterRule removes a previously installed filter rule. The
 // rule's stored chain/table identify where to delete from; source set
 // references are recovered from the spec via findSets and dropped

client/firewall/iptables/manager_linux.go

@@ -25,9 +25,8 @@ type Manager struct {
 
 	wgIface iFaceMapper
 
-	ipv4Client   *iptables.IPTables
-	family4      *family
-	rawSupported bool
+	ipv4Client *iptables.IPTables
+	family4    *family
 
 	// IPv6 counterparts, nil when no v6 overlay
 	ipv6Client *iptables.IPTables
@@ -108,10 +107,6 @@ func (m *Manager) Init(stateManager *statemanager.Manager) error {
 		return err
 	}
 
-	if err := m.initNoTrackChain(); err != nil {
-		log.Warnf("raw table not available, notrack rules will be disabled: %v", err)
-	}
-
 	// Trust after all fatal init steps so a later failure doesn't leave the
 	// interface in firewalld's trusted zone without a corresponding Close.
 	if err := firewalld.TrustInterface(m.wgIface.Name()); err != nil {
@@ -285,10 +280,6 @@ func (m *Manager) Close(stateManager *statemanager.Manager) error {
 
 	var merr *multierror.Error
 
-	if err := m.cleanupNoTrackChain(); err != nil {
-		merr = multierror.Append(merr, fmt.Errorf("cleanup notrack chain: %w", err))
-	}
-
 	if m.hasIPv6() {
 		if err := m.family6.Reset(); err != nil {
 			merr = multierror.Append(merr, fmt.Errorf("reset v6 family: %w", err))
@@ -332,31 +323,6 @@ func (m *Manager) DisableRouting() error {
 	return m.family4.ipFwdState.ReleaseRouting()
 }
 
-// AddDNATRule adds a DNAT rule
-func (m *Manager) AddDNATRule(rule firewall.ForwardRule) (firewall.Rule, error) {
-	m.mutex.Lock()
-	defer m.mutex.Unlock()
-
-	if rule.TranslatedAddress.Is6() {
-		if !m.hasIPv6() {
-			return nil, fmt.Errorf("add DNAT rule: %w", firewall.ErrIPv6NotInitialized)
-		}
-		return m.family6.AddDNATRule(rule)
-	}
-	return m.family4.AddDNATRule(rule)
-}
-
-// DeleteDNATRule deletes a DNAT rule
-func (m *Manager) DeleteDNATRule(rule firewall.Rule) error {
-	m.mutex.Lock()
-	defer m.mutex.Unlock()
-
-	if m.hasIPv6() && !m.family4.hasDNATRule(rule.ID()) {
-		return m.family6.DeleteDNATRule(rule)
-	}
-	return m.family4.DeleteDNATRule(rule)
-}
-
 // UpdateSet updates the set with the given prefixes
 func (m *Manager) UpdateSet(set firewall.Set, prefixes []netip.Prefix) error {
 	m.mutex.Lock()
@@ -440,134 +406,6 @@ func (m *Manager) RemoveOutputDNAT(localAddr netip.Addr, protocol firewall.Proto
 	return m.family4.RemoveOutputDNAT(localAddr, protocol, originalPort, translatedPort)
 }
 
-const (
-	chainNameRaw = "NETBIRD-RAW"
-	chainOutput  = "OUTPUT"
-	tableRaw     = "raw"
-)
-
-// SetupEBPFProxyNoTrack creates notrack rules for eBPF proxy loopback traffic.
-// This prevents conntrack from tracking WireGuard proxy traffic on loopback, which
-// can interfere with MASQUERADE rules (e.g., from container runtimes like Podman/netavark).
-//
-// Traffic flows that need NOTRACK:
-//
-//  1. Egress: WireGuard -> fake endpoint (before eBPF rewrite)
-//     src=127.0.0.1:wgPort -> dst=127.0.0.1:fakePort
-//     Matched by: sport=wgPort
-//
-//  2. Egress: Proxy -> WireGuard (via raw socket)
-//     src=127.0.0.1:fakePort -> dst=127.0.0.1:wgPort
-//     Matched by: dport=wgPort
-//
-//  3. Ingress: Packets to WireGuard
-//     dst=127.0.0.1:wgPort
-//     Matched by: dport=wgPort
-//
-//  4. Ingress: Packets to proxy (after eBPF rewrite)
-//     dst=127.0.0.1:proxyPort
-//     Matched by: dport=proxyPort
-//
-// Rules are cleaned up when the firewall manager is closed.
-func (m *Manager) SetupEBPFProxyNoTrack(proxyPort, wgPort uint16) error {
-	m.mutex.Lock()
-	defer m.mutex.Unlock()
-
-	if !m.rawSupported {
-		return fmt.Errorf("raw table not available")
-	}
-
-	wgPortStr := fmt.Sprintf("%d", wgPort)
-	proxyPortStr := fmt.Sprintf("%d", proxyPort)
-
-	// Egress rules: match outgoing loopback UDP packets
-	outputRuleSport := []string{"-o", "lo", "-s", "127.0.0.1", "-d", "127.0.0.1", "-p", "udp", "--sport", wgPortStr, "-j", "NOTRACK"}
-	if err := m.ipv4Client.AppendUnique(tableRaw, chainNameRaw, outputRuleSport...); err != nil {
-		return fmt.Errorf("add output sport notrack rule: %w", err)
-	}
-
-	outputRuleDport := []string{"-o", "lo", "-s", "127.0.0.1", "-d", "127.0.0.1", "-p", "udp", "--dport", wgPortStr, "-j", "NOTRACK"}
-	if err := m.ipv4Client.AppendUnique(tableRaw, chainNameRaw, outputRuleDport...); err != nil {
-		return fmt.Errorf("add output dport notrack rule: %w", err)
-	}
-
-	// Ingress rules: match incoming loopback UDP packets
-	preroutingRuleWg := []string{"-i", "lo", "-s", "127.0.0.1", "-d", "127.0.0.1", "-p", "udp", "--dport", wgPortStr, "-j", "NOTRACK"}
-	if err := m.ipv4Client.AppendUnique(tableRaw, chainNameRaw, preroutingRuleWg...); err != nil {
-		return fmt.Errorf("add prerouting wg notrack rule: %w", err)
-	}
-
-	preroutingRuleProxy := []string{"-i", "lo", "-s", "127.0.0.1", "-d", "127.0.0.1", "-p", "udp", "--dport", proxyPortStr, "-j", "NOTRACK"}
-	if err := m.ipv4Client.AppendUnique(tableRaw, chainNameRaw, preroutingRuleProxy...); err != nil {
-		return fmt.Errorf("add prerouting proxy notrack rule: %w", err)
-	}
-
-	log.Debugf("set up ebpf proxy notrack rules for ports %d,%d", proxyPort, wgPort)
-	return nil
-}
-
-func (m *Manager) initNoTrackChain() error {
-	if err := m.cleanupNoTrackChain(); err != nil {
-		log.Debugf("cleanup notrack chain: %v", err)
-	}
-
-	if err := m.ipv4Client.NewChain(tableRaw, chainNameRaw); err != nil {
-		return fmt.Errorf("create chain: %w", err)
-	}
-
-	jumpRule := []string{"-j", chainNameRaw}
-
-	if err := m.ipv4Client.InsertUnique(tableRaw, chainOutput, 1, jumpRule...); err != nil {
-		if delErr := m.ipv4Client.DeleteChain(tableRaw, chainNameRaw); delErr != nil {
-			log.Debugf("delete orphan chain: %v", delErr)
-		}
-		return fmt.Errorf("add output jump rule: %w", err)
-	}
-
-	if err := m.ipv4Client.InsertUnique(tableRaw, chainPrerouting, 1, jumpRule...); err != nil {
-		if delErr := m.ipv4Client.DeleteIfExists(tableRaw, chainOutput, jumpRule...); delErr != nil {
-			log.Debugf("delete output jump rule: %v", delErr)
-		}
-		if delErr := m.ipv4Client.DeleteChain(tableRaw, chainNameRaw); delErr != nil {
-			log.Debugf("delete orphan chain: %v", delErr)
-		}
-		return fmt.Errorf("add prerouting jump rule: %w", err)
-	}
-
-	m.rawSupported = true
-	return nil
-}
-
-func (m *Manager) cleanupNoTrackChain() error {
-	exists, err := m.ipv4Client.ChainExists(tableRaw, chainNameRaw)
-	if err != nil {
-		if !m.rawSupported {
-			return nil
-		}
-		return fmt.Errorf("check chain exists: %w", err)
-	}
-	if !exists {
-		return nil
-	}
-
-	jumpRule := []string{"-j", chainNameRaw}
-
-	if err := m.ipv4Client.DeleteIfExists(tableRaw, chainOutput, jumpRule...); err != nil {
-		return fmt.Errorf("remove output jump rule: %w", err)
-	}
-
-	if err := m.ipv4Client.DeleteIfExists(tableRaw, chainPrerouting, jumpRule...); err != nil {
-		return fmt.Errorf("remove prerouting jump rule: %w", err)
-	}
-
-	if err := m.ipv4Client.ClearAndDeleteChain(tableRaw, chainNameRaw); err != nil {
-		return fmt.Errorf("clear and delete chain: %w", err)
-	}
-
-	m.rawSupported = false
-	return nil
-}
-
 func getConntrackEstablished() []string {
 	return []string{"-m", "conntrack", "--ctstate", "RELATED,ESTABLISHED", "-j", "ACCEPT"}
 }

client/firewall/iptables/manager_linux_test.go

@@ -497,16 +497,6 @@ func TestIptablesCloseRemovesAllState(t *testing.T) {
 	require.NoError(t, manager.AddNatRule(pair), "add nat rule")
 	require.NoError(t, manager.EnableRouting(), "enable routing")
 
-	// A DNAT redirect, which also holds a forwarding reference.
-	dnat := fw.ForwardRule{
-		Protocol:          fw.ProtocolTCP,
-		DestinationPort:   fw.Port{Values: []uint16{8080}},
-		TranslatedAddress: netip.MustParseAddr("10.20.0.44"),
-		TranslatedPort:    fw.Port{Values: []uint16{80}},
-	}
-	_, err = manager.AddDNATRule(dnat)
-	require.NoError(t, err, "add dnat rule")
-
 	require.NotEqual(t, before, snapshotIptables(t, ipv4Client), "the manager must have installed state")
 
 	// Everything above stays in place, so Close is what has to remove it.

client/firewall/manager/firewall.go

@@ -172,12 +172,6 @@ type Manager interface {
 
 	DisableRouting() error
 
-	// AddDNATRule adds outbound DNAT rule for forwarding external traffic to the NetBird network.
-	AddDNATRule(ForwardRule) (Rule, error)
-
-	// DeleteDNATRule deletes the outbound DNAT rule.
-	DeleteDNATRule(Rule) error
-
 	// UpdateSet updates the set with the given prefixes
 	UpdateSet(hash Set, prefixes []netip.Prefix) error
 
@@ -192,10 +186,6 @@ type Manager interface {
 
 	// RemoveOutputDNAT removes an OUTPUT chain DNAT rule.
 	RemoveOutputDNAT(localAddr netip.Addr, protocol Protocol, originalPort, translatedPort uint16) error
-
-	// SetupEBPFProxyNoTrack creates static notrack rules for eBPF proxy loopback traffic.
-	// This prevents conntrack from interfering with WireGuard proxy communication.
-	SetupEBPFProxyNoTrack(proxyPort, wgPort uint16) error
 }
 
 // GenKey builds the rule id for this pair from the given format.

client/firewall/manager/forward_rule.go

@@ -1,27 +0,0 @@
-package manager
-
-import (
-	"fmt"
-	"net/netip"
-)
-
-// ForwardRule todo figure out better place to this to avoid circular imports
-type ForwardRule struct {
-	Protocol          Protocol
-	DestinationPort   Port
-	TranslatedAddress netip.Addr
-	TranslatedPort    Port
-}
-
-func (r ForwardRule) ID() RuleID {
-	id := fmt.Sprintf("%s;%s;%s;%s",
-		r.Protocol,
-		r.DestinationPort.String(),
-		r.TranslatedAddress.String(),
-		r.TranslatedPort.String())
-	return RuleID(id)
-}
-
-func (r ForwardRule) String() string {
-	return fmt.Sprintf("protocol: %s, destinationPort: %s, translatedAddress: %s, translatedPort: %s", r.Protocol, r.DestinationPort.String(), r.TranslatedAddress.String(), r.TranslatedPort.String())
-}

client/firewall/nftables/dnat_linux.go

@@ -9,332 +9,11 @@ import (
 	"github.com/google/nftables"
 	"github.com/google/nftables/binaryutil"
 	"github.com/google/nftables/expr"
-	"github.com/google/nftables/xt"
-	"github.com/hashicorp/go-multierror"
 	log "github.com/sirupsen/logrus"
 
-	nberrors "github.com/netbirdio/netbird/client/errors"
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
 )
 
-func (r *family) AddDNATRule(rule firewall.ForwardRule) (firewall.Rule, error) {
-	ruleID := rule.ID()
-	if _, exists := r.rules[ruleID+dnatSuffix]; exists {
-		return rule, nil
-	}
-
-	protoNum, err := r.af.protoNum(rule.Protocol)
-	if err != nil {
-		return nil, fmt.Errorf("convert protocol to number: %w", err)
-	}
-
-	// Request forwarding before queueing rules: addDnatRedirect/addDnatMasq
-	// buffer netlink messages on r.conn that the next caller's Flush would
-	// commit if we returned without flushing them ourselves.
-	if err := r.ipFwdState.RequestForwarding(r.isV6()); err != nil {
-		return nil, fmt.Errorf("enable forwarding: %w", err)
-	}
-
-	if err := r.addDnatRedirect(rule, protoNum, ruleID); err != nil {
-		r.releaseForwarding()
-		return nil, err
-	}
-
-	if err := r.addDnatMasq(rule, protoNum, ruleID); err != nil {
-		r.releaseForwarding()
-		delete(r.rules, ruleID+dnatSuffix)
-		return nil, err
-	}
-
-	// Unlike iptables, there's no point in adding "out" rules in the forward chain here as our policy is ACCEPT.
-	// To overcome DROP policies in other chains, we'd have to add rules to the chains there.
-	// We also cannot just add "oif <iface> accept" there and filter in our own table as we don't know what is supposed to be allowed.
-	// TODO: find chains with drop policies and add rules there
-
-	if err := r.conn.Flush(); err != nil {
-		r.releaseForwarding()
-		delete(r.rules, ruleID+dnatSuffix)
-		delete(r.rules, ruleID+snatSuffix)
-		return nil, fmt.Errorf("flush rules: %w", err)
-	}
-
-	return &rule, nil
-}
-
-func (r *family) addDnatRedirect(rule firewall.ForwardRule, protoNum uint8, ruleID firewall.RuleID) error {
-	dnatExprs := []expr.Any{
-		&expr.Meta{Key: expr.MetaKeyIIFNAME, Register: 1},
-		&expr.Cmp{
-			Op:       expr.CmpOpNeq,
-			Register: 1,
-			Data:     ifname(r.wgIface.Name()),
-		},
-		&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     []byte{protoNum},
-		},
-		&expr.Payload{
-			DestRegister: 1,
-			Base:         expr.PayloadBaseTransportHeader,
-			Offset:       2,
-			Len:          2,
-		},
-	}
-	portExprs, err := r.applyPort(&rule.DestinationPort, false)
-	if err != nil {
-		return fmt.Errorf("apply destination port: %w", err)
-	}
-	dnatExprs = append(dnatExprs, portExprs...)
-
-	// shifted translated port is not supported in nftables, so we hand this over to xtables
-	if rule.TranslatedPort.IsRange && len(rule.TranslatedPort.Values) == 2 {
-		if rule.TranslatedPort.Values[0] != rule.DestinationPort.Values[0] ||
-			rule.TranslatedPort.Values[1] != rule.DestinationPort.Values[1] {
-			return r.addXTablesRedirect(dnatExprs, ruleID, rule)
-		}
-	}
-
-	additionalExprs, regProtoMin, regProtoMax, err := r.handleTranslatedPort(rule)
-	if err != nil {
-		return err
-	}
-	dnatExprs = append(dnatExprs, additionalExprs...)
-
-	dnatExprs = append(dnatExprs,
-		&expr.NAT{
-			Type:        expr.NATTypeDestNAT,
-			Family:      uint32(r.af.tableFamily),
-			RegAddrMin:  1,
-			RegProtoMin: regProtoMin,
-			RegProtoMax: regProtoMax,
-		},
-	)
-
-	dnatRule := &nftables.Rule{
-		Table:    r.workTable,
-		Chain:    r.chains[chainNameRoutingRdr],
-		Exprs:    dnatExprs,
-		UserData: []byte(ruleID + dnatSuffix),
-	}
-	r.conn.AddRule(dnatRule)
-	r.rules[ruleID+dnatSuffix] = dnatRule
-
-	return nil
-}
-
-func (r *family) handleTranslatedPort(rule firewall.ForwardRule) ([]expr.Any, uint32, uint32, error) {
-	switch {
-	case rule.TranslatedPort.IsRange && len(rule.TranslatedPort.Values) == 2:
-		return r.handlePortRange(rule)
-	case len(rule.TranslatedPort.Values) == 0:
-		return r.handleAddressOnly(rule)
-	case len(rule.TranslatedPort.Values) == 1:
-		return r.handleSinglePort(rule)
-	default:
-		return nil, 0, 0, fmt.Errorf("invalid translated port: %v", rule.TranslatedPort)
-	}
-}
-
-func (r *family) handlePortRange(rule firewall.ForwardRule) ([]expr.Any, uint32, uint32, error) {
-	exprs := []expr.Any{
-		&expr.Immediate{
-			Register: 1,
-			Data:     rule.TranslatedAddress.AsSlice(),
-		},
-		&expr.Immediate{
-			Register: 2,
-			Data:     binaryutil.BigEndian.PutUint16(rule.TranslatedPort.Values[0]),
-		},
-		&expr.Immediate{
-			Register: 3,
-			Data:     binaryutil.BigEndian.PutUint16(rule.TranslatedPort.Values[1]),
-		},
-	}
-	return exprs, 2, 3, nil
-}
-
-func (r *family) handleAddressOnly(rule firewall.ForwardRule) ([]expr.Any, uint32, uint32, error) {
-	exprs := []expr.Any{
-		&expr.Immediate{
-			Register: 1,
-			Data:     rule.TranslatedAddress.AsSlice(),
-		},
-	}
-	return exprs, 0, 0, nil
-}
-
-func (r *family) handleSinglePort(rule firewall.ForwardRule) ([]expr.Any, uint32, uint32, error) {
-	exprs := []expr.Any{
-		&expr.Immediate{
-			Register: 1,
-			Data:     rule.TranslatedAddress.AsSlice(),
-		},
-		&expr.Immediate{
-			Register: 2,
-			Data:     binaryutil.BigEndian.PutUint16(rule.TranslatedPort.Values[0]),
-		},
-	}
-	return exprs, 2, 0, nil
-}
-
-func (r *family) addXTablesRedirect(dnatExprs []expr.Any, ruleID firewall.RuleID, rule firewall.ForwardRule) error {
-	dnatExprs = append(dnatExprs,
-		&expr.Counter{},
-		&expr.Target{
-			Name: "DNAT",
-			Rev:  2,
-			Info: &xt.NatRange2{
-				NatRange: xt.NatRange{
-					Flags:   uint(xt.NatRangeMapIPs | xt.NatRangeProtoSpecified | xt.NatRangeProtoOffset),
-					MinIP:   rule.TranslatedAddress.AsSlice(),
-					MaxIP:   rule.TranslatedAddress.AsSlice(),
-					MinPort: rule.TranslatedPort.Values[0],
-					MaxPort: rule.TranslatedPort.Values[1],
-				},
-				BasePort: rule.DestinationPort.Values[0],
-			},
-		},
-	)
-
-	natTable := &nftables.Table{
-		Name:   tableNat,
-		Family: r.af.tableFamily,
-	}
-	dnatRule := &nftables.Rule{
-		Table: natTable,
-		Chain: &nftables.Chain{
-			Name:     chainNameNatPrerouting,
-			Table:    natTable,
-			Type:     nftables.ChainTypeNAT,
-			Hooknum:  nftables.ChainHookPrerouting,
-			Priority: nftables.ChainPriorityNATDest,
-		},
-		Exprs:    dnatExprs,
-		UserData: []byte(ruleID + dnatSuffix),
-	}
-	r.conn.AddRule(dnatRule)
-	r.rules[ruleID+dnatSuffix] = dnatRule
-
-	return nil
-}
-
-func (r *family) addDnatMasq(rule firewall.ForwardRule, protoNum uint8, ruleID firewall.RuleID) error {
-	portExprs, err := r.applyPort(&rule.TranslatedPort, false)
-	if err != nil {
-		return fmt.Errorf("apply translated port: %w", err)
-	}
-
-	masqExprs := []expr.Any{
-		&expr.Meta{Key: expr.MetaKeyOIFNAME, Register: 1},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     ifname(r.wgIface.Name()),
-		},
-		&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     []byte{protoNum},
-		},
-		&expr.Payload{
-			DestRegister: 1,
-			Base:         expr.PayloadBaseNetworkHeader,
-			Offset:       r.af.dstAddrOffset,
-			Len:          r.af.addrLen,
-		},
-		&expr.Cmp{
-			Op:       expr.CmpOpEq,
-			Register: 1,
-			Data:     rule.TranslatedAddress.AsSlice(),
-		},
-	}
-
-	masqExprs = append(masqExprs, portExprs...)
-	masqExprs = append(masqExprs, &expr.Masq{})
-
-	masqRule := &nftables.Rule{
-		Table:    r.workTable,
-		Chain:    r.chains[chainNameRoutingNat],
-		Exprs:    masqExprs,
-		UserData: []byte(ruleID + snatSuffix),
-	}
-	r.conn.AddRule(masqRule)
-	r.rules[ruleID+snatSuffix] = masqRule
-
-	return nil
-}
-
-func (r *family) DeleteDNATRule(rule firewall.Rule) error {
-	ruleID := rule.ID()
-
-	if err := r.refreshRulesMap(); err != nil {
-		return fmt.Errorf(refreshRulesMapError, err)
-	}
-
-	var merr *multierror.Error
-	var needsFlush bool
-	var found bool
-
-	if dnatRule, exists := r.rules[ruleID+dnatSuffix]; exists {
-		found = true
-		if dnatRule.Handle == 0 {
-			log.Warnf("dnat rule %s has no handle, removing stale entry", ruleID+dnatSuffix)
-			delete(r.rules, ruleID+dnatSuffix)
-		} else if err := r.conn.DelRule(dnatRule); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("delete dnat rule: %w", err))
-		} else {
-			needsFlush = true
-		}
-	}
-
-	if masqRule, exists := r.rules[ruleID+snatSuffix]; exists {
-		found = true
-		if masqRule.Handle == 0 {
-			log.Warnf("snat rule %s has no handle, removing stale entry", ruleID+snatSuffix)
-			delete(r.rules, ruleID+snatSuffix)
-		} else if err := r.conn.DelRule(masqRule); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("delete snat rule: %w", err))
-		} else {
-			needsFlush = true
-		}
-	}
-
-	if needsFlush {
-		if err := r.conn.Flush(); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf(flushError, err))
-		}
-	}
-
-	if merr != nil {
-		return nberrors.FormatErrorOrNil(merr)
-	}
-
-	delete(r.rules, ruleID+dnatSuffix)
-	delete(r.rules, ruleID+snatSuffix)
-
-	// Release once, only if the rule was present and removed.
-	if found {
-		r.releaseForwarding()
-	}
-
-	return nil
-}
-
-// releaseForwarding drops one IP forwarding reference, logging any error.
-func (r *family) releaseForwarding() {
-	if err := r.ipFwdState.ReleaseForwarding(r.isV6()); err != nil {
-		log.Errorf("release IP forwarding: %v", err)
-	}
-}
-
-// isV6 reports whether this family handles the IPv6 table.
-func (r *family) isV6() bool {
-	return r.af.tableFamily == nftables.TableFamilyIPv6
-}
-
 func (r *family) AddInboundDNAT(localAddr netip.Addr, protocol firewall.Protocol, originalPort, translatedPort uint16) error {
 	ruleID := firewall.RuleID(fmt.Sprintf("inbound-dnat-%s-%s-%d-%d", localAddr.String(), protocol, originalPort, translatedPort))
 

client/firewall/nftables/dnat_refcount_linux_test.go

@@ -1,249 +0,0 @@
-//go:build privileged
-
-package nftables
-
-import (
-	"net/netip"
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-
-	fw "github.com/netbirdio/netbird/client/firewall/manager"
-	"github.com/netbirdio/netbird/client/iface"
-	"github.com/netbirdio/netbird/client/iface/wgaddr"
-)
-
-func nftRefcountIfaceV4() *iFaceMock {
-	return &iFaceMock{
-		NameFunc: func() string { return "wt-refcount" },
-		AddressFunc: func() wgaddr.Address {
-			return wgaddr.Address{
-				IP:      netip.MustParseAddr("100.96.0.1"),
-				Network: netip.MustParsePrefix("100.96.0.0/16"),
-			}
-		},
-	}
-}
-
-func nftRefcountIfaceDual() *iFaceMock {
-	return &iFaceMock{
-		NameFunc: func() string { return "wt-refcount" },
-		AddressFunc: func() wgaddr.Address {
-			return wgaddr.Address{
-				IP:      netip.MustParseAddr("100.96.0.1"),
-				Network: netip.MustParsePrefix("100.96.0.0/16"),
-				IPv6:    netip.MustParseAddr("fd00::1"),
-				IPv6Net: netip.MustParsePrefix("fd00::/64"),
-			}
-		},
-	}
-}
-
-func newNftRefcountManager(t *testing.T, dual bool) *Manager {
-	t.Helper()
-	if check() != NFTABLES {
-		t.Skip("nftables not supported on this system")
-	}
-	var ifMock *iFaceMock
-	if dual {
-		ifMock = nftRefcountIfaceDual()
-	} else {
-		ifMock = nftRefcountIfaceV4()
-	}
-	m, err := Create(ifMock, iface.DefaultMTU)
-	require.NoError(t, err, "create manager")
-	require.NoError(t, m.Init(nil), "init manager")
-	t.Cleanup(func() {
-		require.NoError(t, m.Close(nil), "close manager")
-	})
-	return m
-}
-
-func dnatV4(port uint16) fw.ForwardRule {
-	return fw.ForwardRule{
-		Protocol:          fw.ProtocolTCP,
-		DestinationPort:   fw.Port{Values: []uint16{port}},
-		TranslatedAddress: netip.MustParseAddr("100.96.0.2"),
-		TranslatedPort:    fw.Port{Values: []uint16{80}},
-	}
-}
-
-func dnatV6(port uint16) fw.ForwardRule {
-	return fw.ForwardRule{
-		Protocol:          fw.ProtocolTCP,
-		DestinationPort:   fw.Port{Values: []uint16{port}},
-		TranslatedAddress: netip.MustParseAddr("fd00::2"),
-		TranslatedPort:    fw.Port{Values: []uint16{80}},
-	}
-}
-
-// TestNftablesDNAT_RefcountBalancedV4 verifies that Add/Delete pairs leave the
-// v4 refcount at zero.
-func TestNftablesDNAT_RefcountBalancedV4(t *testing.T) {
-	m := newNftRefcountManager(t, false)
-	state := m.family4.ipFwdState
-
-	r1, err := m.AddDNATRule(dnatV4(8081))
-	require.NoError(t, err, "add v4 dnat 1")
-	v4, v6 := state.Counts()
-	assert.Equal(t, 1, v4, "v4 refcount after first add")
-	assert.Equal(t, 0, v6, "v6 refcount unchanged")
-
-	r2, err := m.AddDNATRule(dnatV4(8082))
-	require.NoError(t, err, "add v4 dnat 2")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 2, v4, "v4 refcount after second add")
-	assert.Equal(t, 0, v6, "v6 refcount unchanged")
-
-	require.NoError(t, m.DeleteDNATRule(r1), "delete v4 dnat 1")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 1, v4, "v4 refcount after first delete")
-	assert.Equal(t, 0, v6, "v6 refcount unchanged")
-
-	require.NoError(t, m.DeleteDNATRule(r2), "delete v4 dnat 2")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4, "v4 refcount after second delete")
-	assert.Equal(t, 0, v6, "v6 refcount unchanged")
-}
-
-// TestNftablesDNAT_RefcountBalancedV6 verifies the v6 path increments v6 only
-// and decrements back to zero on Delete.
-func TestNftablesDNAT_RefcountBalancedV6(t *testing.T) {
-	m := newNftRefcountManager(t, true)
-	require.NotNil(t, m.family6, "v6 family")
-	require.Same(t, m.family4.ipFwdState, m.family6.ipFwdState, "shared state")
-	state := m.family4.ipFwdState
-
-	r1, err := m.AddDNATRule(dnatV6(9091))
-	require.NoError(t, err, "add v6 dnat 1")
-	v4, v6 := state.Counts()
-	assert.Equal(t, 0, v4, "v4 refcount unchanged")
-	assert.Equal(t, 1, v6, "v6 refcount after first add")
-
-	r2, err := m.AddDNATRule(dnatV6(9092))
-	require.NoError(t, err, "add v6 dnat 2")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4)
-	assert.Equal(t, 2, v6, "v6 refcount after second add")
-
-	require.NoError(t, m.DeleteDNATRule(r1), "delete v6 dnat 1")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4, "v4 refcount unchanged")
-	assert.Equal(t, 1, v6, "v6 refcount after first delete")
-
-	require.NoError(t, m.DeleteDNATRule(r2), "delete v6 dnat 2")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4)
-	assert.Equal(t, 0, v6, "v6 refcount after second delete")
-}
-
-// TestNftablesDNAT_DuplicateAddNoLeak verifies that a duplicate Add (same
-// ForwardRule) does not double-increment the refcount.
-func TestNftablesDNAT_DuplicateAddNoLeak(t *testing.T) {
-	m := newNftRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	rule := dnatV4(8083)
-	r1, err := m.AddDNATRule(rule)
-	require.NoError(t, err, "add v4 dnat")
-	v4, _ := state.Counts()
-	assert.Equal(t, 1, v4)
-
-	// duplicate add: same rule ID, must be a no-op for the refcount.
-	_, err = m.AddDNATRule(rule)
-	require.NoError(t, err, "duplicate add")
-	v4, _ = state.Counts()
-	assert.Equal(t, 1, v4, "duplicate add must not increment")
-
-	require.NoError(t, m.DeleteDNATRule(r1), "delete v4 dnat")
-	v4, _ = state.Counts()
-	assert.Equal(t, 0, v4, "single delete must drop to zero")
-}
-
-// TestNftablesDNAT_DeleteMissingNoUnderflow verifies deleting a rule that was
-// never added does not underflow the refcount.
-func TestNftablesDNAT_DeleteMissingNoUnderflow(t *testing.T) {
-	m := newNftRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	// Construct a Rule reference for something never added. The router stores
-	// rules by ID(), and DeleteDNATRule looks them up in r.rules; a missing
-	// entry must be a no-op rather than calling Release.
-	phantom := dnatV4(8099)
-	require.NoError(t, m.DeleteDNATRule(&phantom), "delete missing v4 dnat")
-	v4, v6 := state.Counts()
-	assert.Equal(t, 0, v4, "v4 refcount unaffected by missing delete")
-	assert.Equal(t, 0, v6, "v6 refcount unaffected")
-
-	phantom6 := dnatV6(9099)
-	require.NoError(t, m.DeleteDNATRule(&phantom6), "delete missing v6 dnat")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4)
-	assert.Equal(t, 0, v6, "v6 refcount unaffected by missing delete")
-
-	// And after a phantom delete, a real add still results in count=1.
-	r1, err := m.AddDNATRule(dnatV4(8100))
-	require.NoError(t, err, "add v4 dnat after phantom delete")
-	v4, _ = state.Counts()
-	assert.Equal(t, 1, v4, "real add still increments after phantom delete")
-	require.NoError(t, m.DeleteDNATRule(r1))
-}
-
-// TestNftablesRouting_RepeatedEnableSingleReference verifies that EnableRouting
-// (called on every network-map update) holds at most one reference per family
-// and a single DisableRouting drops both back to zero.
-func TestNftablesRouting_RepeatedEnableSingleReference(t *testing.T) {
-	m := newNftRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	require.NoError(t, m.EnableRouting(), "first enable")
-	require.NoError(t, m.EnableRouting(), "second enable")
-	require.NoError(t, m.EnableRouting(), "third enable")
-	v4, v6 := state.Counts()
-	assert.Equal(t, 1, v4, "repeated enable holds a single v4 reference")
-	assert.Equal(t, 1, v6, "repeated enable holds a single v6 reference")
-
-	require.NoError(t, m.DisableRouting(), "disable")
-	v4, v6 = state.Counts()
-	assert.Equal(t, 0, v4, "single disable releases the v4 reference")
-	assert.Equal(t, 0, v6, "single disable releases the v6 reference")
-}
-
-// TestNftablesRouting_DisableKeepsDNATReference verifies that an unpaired
-// DisableRouting does not release references held by active DNAT rules.
-func TestNftablesRouting_DisableKeepsDNATReference(t *testing.T) {
-	m := newNftRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	r1, err := m.AddDNATRule(dnatV6(9095))
-	require.NoError(t, err, "add v6 dnat")
-
-	require.NoError(t, m.DisableRouting(), "unpaired disable")
-	_, v6 := state.Counts()
-	assert.Equal(t, 1, v6, "DNAT-held reference survives unpaired DisableRouting")
-
-	require.NoError(t, m.DeleteDNATRule(r1), "delete v6 dnat")
-	_, v6 = state.Counts()
-	assert.Equal(t, 0, v6, "delete releases the DNAT reference")
-}
-
-// TestNftablesDNAT_DoubleDeleteNoUnderflow verifies that deleting the same rule
-// twice does not underflow the refcount (the second delete is a no-op).
-func TestNftablesDNAT_DoubleDeleteNoUnderflow(t *testing.T) {
-	m := newNftRefcountManager(t, true)
-	state := m.family4.ipFwdState
-
-	r1, err := m.AddDNATRule(dnatV6(9093))
-	require.NoError(t, err)
-	_, v6 := state.Counts()
-	assert.Equal(t, 1, v6)
-
-	require.NoError(t, m.DeleteDNATRule(r1), "first delete")
-	_, v6 = state.Counts()
-	assert.Equal(t, 0, v6)
-
-	require.NoError(t, m.DeleteDNATRule(r1), "second delete must be no-op")
-	_, v6 = state.Counts()
-	assert.Equal(t, 0, v6, "double delete must not underflow")
-}

client/firewall/nftables/family_linux.go

@@ -24,7 +24,6 @@ const (
 	tableRaw      = "raw"
 	tableSecurity = "security"
 
-	chainNameNatPrerouting = "PREROUTING"
 	chainNameRoutingFw     = "netbird-rt-fwd"
 	chainNameRoutingNat    = "netbird-rt-postrouting"
 	chainNameRoutingRdr    = "netbird-rt-redirect"
@@ -47,9 +46,6 @@ const (
 	userDataAcceptForwardRuleOif = "frwacceptoif"
 	userDataAcceptInputRule      = "inputaccept"
 
-	dnatSuffix firewall.RuleID = "_dnat"
-	snatSuffix firewall.RuleID = "_snat"
-
 	// ipv4TCPHeaderSize is the minimum IPv4 (20) + TCP (20) header size for MSS calculation.
 	ipv4TCPHeaderSize = 40
 	// ipv6TCPHeaderSize is the minimum IPv6 (40) + TCP (20) header size for MSS calculation.
@@ -167,10 +163,6 @@ func (r *family) Reset() error {
 		merr = multierror.Append(merr, err)
 	}
 
-	if err := r.removeNatPreroutingRules(); err != nil {
-		merr = multierror.Append(merr, fmt.Errorf("remove filter prerouting rules: %w", err))
-	}
-
 	return nberrors.FormatErrorOrNil(merr)
 }
 

client/firewall/nftables/filter_linux.go

@@ -197,11 +197,6 @@ func (r *family) hasRule(id firewall.RuleID) bool {
 	return ok
 }
 
-func (r *family) hasDNATRule(id firewall.RuleID) bool {
-	_, ok := r.rules[id+dnatSuffix]
-	return ok
-}
-
 // DeleteFilterRule removes a previously installed filter rule. Source
 // set references are recovered from the stored rule's expressions via
 // findSets and dropped from the shared refcounter.

client/firewall/nftables/manager_linux.go

@@ -12,7 +12,6 @@ import (
 	"github.com/google/nftables/expr"
 	"github.com/hashicorp/go-multierror"
 	log "github.com/sirupsen/logrus"
-	"golang.org/x/sys/unix"
 
 	nberrors "github.com/netbirdio/netbird/client/errors"
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
@@ -55,9 +54,6 @@ type Manager struct {
 	// IPv6 counterpart, nil when no v6 overlay.
 	family6 *family
 
-	notrackOutputChain     *nftables.Chain
-	notrackPreroutingChain *nftables.Chain
-
 	extMonitor *externalChainMonitor
 }
 
@@ -170,10 +166,6 @@ func (m *Manager) initFirewall() (err error) {
 		}
 	}
 
-	if err := m.initNoTrackChains(workTable); err != nil {
-		log.Warnf("raw priority chains not available, notrack rules will be disabled: %v", err)
-	}
-
 	return nil
 }
 
@@ -260,7 +252,7 @@ func (m *Manager) DeleteFilterRule(rule firewall.Rule) error {
 	m.mutex.Lock()
 	defer m.mutex.Unlock()
 
-	fam, err := m.familyForRuleID(rule.ID(), (*family).hasRule, false)
+	fam, err := m.familyForRuleID(rule.ID(), (*family).hasRule)
 	if err != nil {
 		return err
 	}
@@ -268,11 +260,8 @@ func (m *Manager) DeleteFilterRule(rule firewall.Rule) error {
 }
 
 // familyForRuleID picks the family holding the rule with the given id, using
-// the supplied lookup. With refresh set, a miss in both cached maps reloads
-// the NAT/DNAT rule maps from the kernel once and re-checks before falling
-// back to the v4 family. Filter rules are tracked only in memory and have no
-// kernel-backed reload, so their callers pass refresh as false.
-func (m *Manager) familyForRuleID(id firewall.RuleID, has func(*family, firewall.RuleID) bool, refresh bool) (*family, error) {
+// the supplied lookup, and falls back to the v4 family on a miss.
+func (m *Manager) familyForRuleID(id firewall.RuleID, has func(*family, firewall.RuleID) bool) (*family, error) {
 	if has(m.family4, id) {
 		return m.family4, nil
 	}
@@ -282,18 +271,6 @@ func (m *Manager) familyForRuleID(id firewall.RuleID, has func(*family, firewall
 	if has(m.family6, id) {
 		return m.family6, nil
 	}
-	if !refresh {
-		return m.family4, nil
-	}
-	if err := m.family4.refreshRulesMap(); err != nil {
-		return nil, fmt.Errorf("refresh v4 rules: %w", err)
-	}
-	if err := m.family6.refreshRulesMap(); err != nil {
-		return nil, fmt.Errorf("refresh v6 rules: %w", err)
-	}
-	if has(m.family6, id) && !has(m.family4, id) {
-		return m.family6, nil
-	}
 	return m.family4, nil
 }
 
@@ -455,39 +432,9 @@ func (m *Manager) Flush() error {
 		}
 	}
 
-	if err := m.refreshNoTrackChains(); err != nil {
-		log.Errorf("failed to refresh notrack chains: %v", err)
-	}
-
 	return nil
 }
 
-// AddDNATRule adds a DNAT rule
-func (m *Manager) AddDNATRule(rule firewall.ForwardRule) (firewall.Rule, error) {
-	m.mutex.Lock()
-	defer m.mutex.Unlock()
-
-	if rule.TranslatedAddress.Is6() {
-		if !m.hasIPv6() {
-			return nil, fmt.Errorf("add DNAT rule: %w", firewall.ErrIPv6NotInitialized)
-		}
-		return m.family6.AddDNATRule(rule)
-	}
-	return m.family4.AddDNATRule(rule)
-}
-
-// DeleteDNATRule deletes a DNAT rule
-func (m *Manager) DeleteDNATRule(rule firewall.Rule) error {
-	m.mutex.Lock()
-	defer m.mutex.Unlock()
-
-	r, err := m.familyForRuleID(rule.ID(), (*family).hasDNATRule, true)
-	if err != nil {
-		return err
-	}
-	return r.DeleteDNATRule(rule)
-}
-
 // UpdateSet updates the set with the given prefixes
 func (m *Manager) UpdateSet(set firewall.Set, prefixes []netip.Prefix) error {
 	m.mutex.Lock()
@@ -571,176 +518,6 @@ func (m *Manager) RemoveOutputDNAT(localAddr netip.Addr, protocol firewall.Proto
 	return m.family4.RemoveOutputDNAT(localAddr, protocol, originalPort, translatedPort)
 }
 
-const (
-	chainNameRawOutput     = "netbird-raw-out"
-	chainNameRawPrerouting = "netbird-raw-pre"
-)
-
-// SetupEBPFProxyNoTrack creates notrack rules for eBPF proxy loopback traffic.
-// This prevents conntrack from tracking WireGuard proxy traffic on loopback, which
-// can interfere with MASQUERADE rules (e.g., from container runtimes like Podman/netavark).
-//
-// Traffic flows that need NOTRACK:
-//
-//  1. Egress: WireGuard -> fake endpoint (before eBPF rewrite)
-//     src=127.0.0.1:wgPort -> dst=127.0.0.1:fakePort
-//     Matched by: sport=wgPort
-//
-//  2. Egress: Proxy -> WireGuard (via raw socket)
-//     src=127.0.0.1:fakePort -> dst=127.0.0.1:wgPort
-//     Matched by: dport=wgPort
-//
-//  3. Ingress: Packets to WireGuard
-//     dst=127.0.0.1:wgPort
-//     Matched by: dport=wgPort
-//
-//  4. Ingress: Packets to proxy (after eBPF rewrite)
-//     dst=127.0.0.1:proxyPort
-//     Matched by: dport=proxyPort
-//
-// Rules are cleaned up when the firewall manager is closed.
-func (m *Manager) SetupEBPFProxyNoTrack(proxyPort, wgPort uint16) error {
-	m.mutex.Lock()
-	defer m.mutex.Unlock()
-
-	if m.notrackOutputChain == nil || m.notrackPreroutingChain == nil {
-		return fmt.Errorf("notrack chains not initialized")
-	}
-
-	proxyPortBytes := binaryutil.BigEndian.PutUint16(proxyPort)
-	wgPortBytes := binaryutil.BigEndian.PutUint16(wgPort)
-	loopback := []byte{127, 0, 0, 1}
-
-	// Egress rules: match outgoing loopback UDP packets
-	m.rConn.AddRule(&nftables.Rule{
-		Table: m.notrackOutputChain.Table,
-		Chain: m.notrackOutputChain,
-		Exprs: []expr.Any{
-			&expr.Meta{Key: expr.MetaKeyOIFNAME, Register: 1},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: ifname("lo")},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 12, Len: 4}, // saddr
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 16, Len: 4}, // daddr
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
-			&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_UDP}},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 0, Len: 2},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: wgPortBytes}, // sport=wgPort
-			&expr.Counter{},
-			&expr.Notrack{},
-		},
-	})
-	m.rConn.AddRule(&nftables.Rule{
-		Table: m.notrackOutputChain.Table,
-		Chain: m.notrackOutputChain,
-		Exprs: []expr.Any{
-			&expr.Meta{Key: expr.MetaKeyOIFNAME, Register: 1},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: ifname("lo")},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 12, Len: 4}, // saddr
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 16, Len: 4}, // daddr
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
-			&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_UDP}},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 2, Len: 2},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: wgPortBytes}, // dport=wgPort
-			&expr.Counter{},
-			&expr.Notrack{},
-		},
-	})
-
-	// Ingress rules: match incoming loopback UDP packets
-	m.rConn.AddRule(&nftables.Rule{
-		Table: m.notrackPreroutingChain.Table,
-		Chain: m.notrackPreroutingChain,
-		Exprs: []expr.Any{
-			&expr.Meta{Key: expr.MetaKeyIIFNAME, Register: 1},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: ifname("lo")},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 12, Len: 4}, // saddr
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 16, Len: 4}, // daddr
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
-			&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_UDP}},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 2, Len: 2},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: wgPortBytes}, // dport=wgPort
-			&expr.Counter{},
-			&expr.Notrack{},
-		},
-	})
-	m.rConn.AddRule(&nftables.Rule{
-		Table: m.notrackPreroutingChain.Table,
-		Chain: m.notrackPreroutingChain,
-		Exprs: []expr.Any{
-			&expr.Meta{Key: expr.MetaKeyIIFNAME, Register: 1},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: ifname("lo")},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 12, Len: 4}, // saddr
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseNetworkHeader, Offset: 16, Len: 4}, // daddr
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: loopback},
-			&expr.Meta{Key: expr.MetaKeyL4PROTO, Register: 1},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: []byte{unix.IPPROTO_UDP}},
-			&expr.Payload{DestRegister: 1, Base: expr.PayloadBaseTransportHeader, Offset: 2, Len: 2},
-			&expr.Cmp{Op: expr.CmpOpEq, Register: 1, Data: proxyPortBytes}, // dport=proxyPort
-			&expr.Counter{},
-			&expr.Notrack{},
-		},
-	})
-
-	if err := m.rConn.Flush(); err != nil {
-		return fmt.Errorf("flush notrack rules: %w", err)
-	}
-
-	log.Debugf("set up ebpf proxy notrack rules for ports %d,%d", proxyPort, wgPort)
-	return nil
-}
-
-func (m *Manager) initNoTrackChains(table *nftables.Table) error {
-	m.notrackOutputChain = m.rConn.AddChain(&nftables.Chain{
-		Name:     chainNameRawOutput,
-		Table:    table,
-		Type:     nftables.ChainTypeFilter,
-		Hooknum:  nftables.ChainHookOutput,
-		Priority: nftables.ChainPriorityRaw,
-	})
-
-	m.notrackPreroutingChain = m.rConn.AddChain(&nftables.Chain{
-		Name:     chainNameRawPrerouting,
-		Table:    table,
-		Type:     nftables.ChainTypeFilter,
-		Hooknum:  nftables.ChainHookPrerouting,
-		Priority: nftables.ChainPriorityRaw,
-	})
-
-	if err := m.rConn.Flush(); err != nil {
-		return fmt.Errorf("flush chain creation: %w", err)
-	}
-
-	return nil
-}
-
-func (m *Manager) refreshNoTrackChains() error {
-	chains, err := m.rConn.ListChainsOfTableFamily(nftables.TableFamilyIPv4)
-	if err != nil {
-		return fmt.Errorf("list chains: %w", err)
-	}
-
-	tableName := getTableName()
-	for _, c := range chains {
-		if c.Table.Name != tableName {
-			continue
-		}
-		switch c.Name {
-		case chainNameRawOutput:
-			m.notrackOutputChain = c
-		case chainNameRawPrerouting:
-			m.notrackPreroutingChain = c
-		}
-	}
-
-	return nil
-}
-
 func (m *Manager) createWorkTable() (*nftables.Table, error) {
 	return m.createWorkTableFamily(nftables.TableFamilyIPv4)
 }

client/firewall/nftables/manager_linux_test.go

@@ -378,18 +378,6 @@ func TestNftablesManagerCompatibilityWithIptables(t *testing.T) {
 	err = manager.AddNatRule(pair)
 	require.NoError(t, err, "failed to add NAT rule")
 
-	dnatRule, err := manager.AddDNATRule(fw.ForwardRule{
-		Protocol:          fw.ProtocolTCP,
-		DestinationPort:   fw.Port{Values: []uint16{8080}},
-		TranslatedAddress: netip.MustParseAddr("100.96.0.2"),
-		TranslatedPort:    fw.Port{Values: []uint16{80}},
-	})
-	require.NoError(t, err, "failed to add DNAT rule")
-
-	t.Cleanup(func() {
-		require.NoError(t, manager.DeleteDNATRule(dnatRule), "failed to delete DNAT rule")
-	})
-
 	stdout, stderr = runIptablesSave(t)
 	verifyIptablesOutput(t, stdout, stderr)
 }
@@ -453,18 +441,6 @@ func TestNftablesManagerIPv6CompatibilityWithIp6tables(t *testing.T) {
 	})
 	require.NoError(t, err, "add v6 NAT rule")
 
-	dnatRule, err := manager.AddDNATRule(fw.ForwardRule{
-		Protocol:          fw.ProtocolTCP,
-		DestinationPort:   fw.Port{Values: []uint16{8080}},
-		TranslatedAddress: netip.MustParseAddr("fd00::2"),
-		TranslatedPort:    fw.Port{Values: []uint16{80}},
-	})
-	require.NoError(t, err, "add v6 DNAT rule")
-
-	t.Cleanup(func() {
-		require.NoError(t, manager.DeleteDNATRule(dnatRule), "delete v6 DNAT rule")
-	})
-
 	stdout, stderr := runIptablesSave(t)
 	verifyIptablesOutput(t, stdout, stderr)
 

client/firewall/nftables/routing_linux.go

@@ -192,7 +192,7 @@ func (r *family) addPostroutingRules() {
 			Data:     binaryutil.NativeEndian.PutUint32(nbnet.PreroutingFwmarkMasquerade),
 		},
 
-		// We need to exclude the loopback interface as this changes the ebpf proxy port
+		// We need to exclude the loopback interface as this changes the wg proxy port
 		&expr.Meta{
 			Key:      expr.MetaKeyOIFNAME,
 			Register: 1,
@@ -459,41 +459,6 @@ func (r *family) RemoveAllLegacyRouteRules() error {
 	return nberrors.FormatErrorOrNil(merr)
 }
 
-func (r *family) removeNatPreroutingRules() error {
-	table := &nftables.Table{
-		Name:   tableNat,
-		Family: r.af.tableFamily,
-	}
-	chain := &nftables.Chain{
-		Name:     chainNameNatPrerouting,
-		Table:    table,
-		Hooknum:  nftables.ChainHookPrerouting,
-		Priority: nftables.ChainPriorityNATDest,
-		Type:     nftables.ChainTypeNAT,
-	}
-	rules, err := r.conn.GetRules(table, chain)
-	if err != nil {
-		return fmt.Errorf("get rules from nat table: %w", err)
-	}
-
-	var merr *multierror.Error
-
-	// Delete rules that have our UserData suffix
-	for _, rule := range rules {
-		if len(rule.UserData) == 0 || !strings.HasSuffix(string(rule.UserData), string(dnatSuffix)) {
-			continue
-		}
-		if err := r.conn.DelRule(rule); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("delete rule %s: %w", rule.UserData, err))
-		}
-	}
-
-	if err := r.conn.Flush(); err != nil {
-		merr = multierror.Append(merr, fmt.Errorf(flushError, err))
-	}
-	return nberrors.FormatErrorOrNil(merr)
-}
-
 func (r *family) RemoveNatRule(pair firewall.RouterPair) error {
 	if err := r.refreshRulesMap(); err != nil {
 		return fmt.Errorf(refreshRulesMapError, err)

client/firewall/uspfilter/filter.go

@@ -879,12 +879,6 @@ func (m *Manager) resetState() {
 	}
 }
 
-// SetupEBPFProxyNoTrack is not supported by the userspace firewall: eBPF isn't
-// used in userspace mode, so this should never be called.
-func (m *Manager) SetupEBPFProxyNoTrack(uint16, uint16) error {
-	return errNotSupported
-}
-
 // UpdateSet updates the rule destinations associated with the given set
 // by merging the existing prefixes with the new ones, then deduplicating.
 func (m *Manager) UpdateSet(set firewall.Set, prefixes []netip.Prefix) error {

client/firewall/uspfilter/nat.go

@@ -486,16 +486,6 @@ func incrementalUpdate(oldChecksum uint16, oldBytes, newBytes []byte) uint16 {
 	return ^uint16(sum)
 }
 
-// AddDNATRule adds outbound DNAT rule for forwarding external traffic to NetBird network.
-func (m *Manager) AddDNATRule(firewall.ForwardRule) (firewall.Rule, error) {
-	return nil, errNotSupported
-}
-
-// DeleteDNATRule deletes outbound DNAT rule.
-func (m *Manager) DeleteDNATRule(firewall.Rule) error {
-	return errNotSupported
-}
-
 // addPortRedirection adds a port redirection rule.
 func (m *Manager) addPortRedirection(targetIP netip.Addr, protocol gopacket.LayerType, originalPort, translatedPort uint16) error {
 	m.portDNATMutex.Lock()

client/iface/configurer/allowedips.go

@@ -0,0 +1,226 @@
+package configurer
+
+import (
+	"net"
+	"net/netip"
+	"slices"
+	"sync"
+
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+)
+
+// allowedIPStore mirrors the allowed IPs configured on each peer of a device.
+//
+// A configurer is the only writer of its device's peer set, so the mirror is authoritative
+// by construction. It spares the paths that have to rewrite one peer's allowed IPs a full
+// device dump just to recover prefixes the process already configured itself.
+//
+// An allowed IP belongs to exactly one peer: configuring a prefix on a peer takes it away
+// from whichever peer held it before, and the configurer leaves that handover to the device
+// rather than removing the prefix from the previous holder itself. The store tracks the
+// owner of each prefix and performs the same handover, so rewriting one peer's list never
+// takes a prefix back from the peer that owns it now.
+//
+// Its own lock guards the map alone, not the device write it accompanies. Consistency
+// between the two rests on the caller serializing every configurer call, which WGIface
+// does with its mutex; two unserialized writers would interleave a device write with the
+// record of a different one.
+//
+// An operator reconfiguring the device out of band, through `wg set` or the UAPI socket,
+// is the one way the mirror can still go stale. A peer missing from it falls back to the
+// device, which reseats that peer's prefixes and their ownership; a peer that is present
+// does not, so one recorded from empty while the device already held prefixes keeps only
+// what was recorded, and the next endpoint removal drops the rest.
+type allowedIPStore struct {
+	mu     sync.RWMutex
+	peers  map[wgtypes.Key][]netip.Prefix
+	owners map[netip.Prefix]wgtypes.Key
+}
+
+func newAllowedIPStore() *allowedIPStore {
+	return &allowedIPStore{
+		peers:  make(map[wgtypes.Key][]netip.Prefix),
+		owners: make(map[netip.Prefix]wgtypes.Key),
+	}
+}
+
+// get returns the prefixes recorded for a peer, and whether the peer is known at all.
+// The caller receives a copy and may retain or modify it freely.
+func (s *allowedIPStore) get(key wgtypes.Key) ([]netip.Prefix, bool) {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	prefixes, ok := s.peers[key]
+	if !ok {
+		return nil, false
+	}
+	return slices.Clone(prefixes), true
+}
+
+// set replaces the prefixes recorded for a peer.
+func (s *allowedIPStore) set(key wgtypes.Key, prefixes []netip.Prefix) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	k := key
+	s.releaseLocked(k)
+
+	normalized := normalizePrefixes(prefixes)
+	for _, prefix := range normalized {
+		s.claimLocked(k, prefix)
+	}
+	s.peers[k] = normalized
+}
+
+// add records prefixes on a peer without dropping the ones already there, matching the
+// union semantics of a peer update that does not replace its allowed IPs. It records the
+// peer if it is not known yet, so it belongs to the operations that create a peer on the
+// device rather than to the update-only ones.
+func (s *allowedIPStore) add(key wgtypes.Key, prefixes []netip.Prefix) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	s.mergeLocked(key, prefixes)
+}
+
+// addExisting is add for an update-only device operation. Such an operation is a silent
+// no-op when the peer is absent, so recording a peer here would leave the store claiming
+// prefixes the device never took, and the peer would then be recreated by the next endpoint
+// removal, stealing those allowed IPs from the peer that legitimately holds them.
+func (s *allowedIPStore) addExisting(key wgtypes.Key, prefixes []netip.Prefix) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	k := key
+	if _, ok := s.peers[k]; !ok {
+		return
+	}
+	s.mergeLocked(k, prefixes)
+}
+
+// ensure records a peer with no prefixes unless it is already known. A device operation
+// that is not update-only creates the peer when it is absent, so it has to be recorded even
+// when it configures nothing else; otherwise the peer exists on the device while the store
+// treats it as unknown, and a prefix later handed over to it is not accounted for.
+func (s *allowedIPStore) ensure(key wgtypes.Key) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	k := key
+	if _, ok := s.peers[k]; !ok {
+		s.peers[k] = nil
+	}
+}
+
+// forget drops every prefix recorded for a peer.
+func (s *allowedIPStore) forget(key wgtypes.Key) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	k := key
+	s.releaseLocked(k)
+	delete(s.peers, k)
+}
+
+// reset drops every peer, mirroring a device reconfiguration that replaces the peer set.
+func (s *allowedIPStore) reset() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	s.peers = make(map[wgtypes.Key][]netip.Prefix)
+	s.owners = make(map[netip.Prefix]wgtypes.Key)
+}
+
+// mergeLocked unions normalized prefixes into a peer and transfers their ownership.
+// The caller must hold s.mu for writing.
+func (s *allowedIPStore) mergeLocked(k wgtypes.Key, prefixes []netip.Prefix) {
+	merged := s.peers[k]
+	for _, prefix := range prefixes {
+		prefix = normalizePrefix(prefix)
+		s.claimLocked(k, prefix)
+		if !slices.Contains(merged, prefix) {
+			merged = append(merged, prefix)
+		}
+	}
+	s.peers[k] = merged
+}
+
+// claimLocked hands a prefix over to a peer, taking it from its previous owner the way the
+// device does when the same prefix is configured on a second peer.
+func (s *allowedIPStore) claimLocked(k wgtypes.Key, prefix netip.Prefix) {
+	if owner, ok := s.owners[prefix]; ok && owner != k {
+		s.peers[owner] = slices.DeleteFunc(s.peers[owner], func(p netip.Prefix) bool {
+			return p == prefix
+		})
+	}
+	s.owners[prefix] = k
+}
+
+// releaseLocked drops a peer's claim on every prefix it currently holds.
+func (s *allowedIPStore) releaseLocked(k wgtypes.Key) {
+	for _, prefix := range s.peers[k] {
+		if s.owners[prefix] == k {
+			delete(s.owners, prefix)
+		}
+	}
+}
+
+// normalizePrefix puts a prefix into the form the store recognises it by. It clears the
+// host bits, which a device does on its own, so a caller passing 10.20.0.1/16 still matches
+// the 10.20.0.0/16 read back from the device; and it unmaps a v4-mapped prefix so that it
+// compares equal to, and marshals like, the plain v4 prefix for the same network.
+//
+// Masking comes first because it also decides the address family: only a prefix at least 96
+// bits long keeps the mapped marker through the mask, so a shorter prefix inside the mapped
+// range is a genuine v6 prefix and unmapping it would yield an invalid v4 prefix.
+func normalizePrefix(prefix netip.Prefix) netip.Prefix {
+	masked := prefix.Masked()
+
+	addr := masked.Addr()
+	if !addr.Is4In6() {
+		return masked
+	}
+	return netip.PrefixFrom(addr.Unmap(), masked.Bits()-96)
+}
+
+// normalizePrefixes returns a normalized copy without changing the caller's slice.
+func normalizePrefixes(prefixes []netip.Prefix) []netip.Prefix {
+	normalized := make([]netip.Prefix, len(prefixes))
+	for i, prefix := range prefixes {
+		normalized[i] = normalizePrefix(prefix)
+	}
+	return normalized
+}
+
+// ipNetsToPrefixes converts addresses read back from a device. Unmap keeps a v4-mapped v6
+// address comparable to the plain v4 prefix the configurer was given.
+func ipNetsToPrefixes(ipNets []net.IPNet) []netip.Prefix {
+	prefixes := make([]netip.Prefix, 0, len(ipNets))
+	for _, ipNet := range ipNets {
+		addr, ok := netip.AddrFromSlice(ipNet.IP)
+		if !ok {
+			continue
+		}
+
+		ones, maskBits := ipNet.Mask.Size()
+		// A device may report a v4 prefix as a v4-mapped address. Align the address form with
+		// the mask rather than unmapping on sight: a 32 bit mask always describes v4, while a
+		// 128 bit mask describes v4 only when it covers the mapped prefix, so a genuine v6
+		// prefix inside the mapped range stays v6 instead of being dropped as invalid.
+		if addr.Is4In6() {
+			switch {
+			case maskBits == 32:
+				addr = addr.Unmap()
+			case maskBits == 128 && ones >= 96:
+				addr, ones = addr.Unmap(), ones-96
+			}
+		}
+
+		prefix := netip.PrefixFrom(addr, ones)
+		if !prefix.IsValid() {
+			continue
+		}
+		prefixes = append(prefixes, prefix.Masked())
+	}
+	return prefixes
+}

client/iface/configurer/allowedips_test.go

@@ -0,0 +1,263 @@
+package configurer
+
+import (
+	"net"
+	"net/netip"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+)
+
+// The store keys on the parsed key, so the tests use two distinct ones rather than names.
+var (
+	testPeer  = wgtypes.Key{1}
+	otherPeer = wgtypes.Key{2}
+)
+
+func TestAllowedIPStoreUnknownPeer(t *testing.T) {
+	s := newAllowedIPStore()
+
+	prefixes, ok := s.get(testPeer)
+	assert.False(t, ok, "an unconfigured peer must be reported as unknown, not as one without prefixes")
+	assert.Nil(t, prefixes, "an unknown peer has no prefixes")
+}
+
+func TestAllowedIPStoreAddUnions(t *testing.T) {
+	s := newAllowedIPStore()
+	overlay := netip.MustParsePrefix("100.64.0.1/32")
+	routed := netip.MustParsePrefix("10.20.0.0/16")
+
+	s.set(testPeer, []netip.Prefix{overlay})
+	// A peer update does not replace allowed IPs, and a repeated prefix must not be doubled.
+	s.add(testPeer, []netip.Prefix{overlay, routed})
+
+	prefixes, ok := s.get(testPeer)
+	require.True(t, ok, "peer must be known after set")
+	assert.Equal(t, []netip.Prefix{overlay, routed}, prefixes, "add must union rather than replace")
+}
+
+func TestAllowedIPStoreGetReturnsCopy(t *testing.T) {
+	s := newAllowedIPStore()
+	overlay := netip.MustParsePrefix("100.64.0.1/32")
+	s.set(testPeer, []netip.Prefix{overlay})
+
+	prefixes, ok := s.get(testPeer)
+	require.True(t, ok, "peer must be known after set")
+	prefixes[0] = netip.MustParsePrefix("0.0.0.0/0")
+
+	stored, _ := s.get(testPeer)
+	assert.Equal(t, []netip.Prefix{overlay}, stored, "a caller mutating the returned slice must not corrupt the store")
+}
+
+func TestAllowedIPStoreForgetAndReset(t *testing.T) {
+	s := newAllowedIPStore()
+	s.set(testPeer, []netip.Prefix{netip.MustParsePrefix("100.64.0.1/32")})
+	s.set(otherPeer, []netip.Prefix{netip.MustParsePrefix("100.64.0.2/32")})
+
+	s.forget(testPeer)
+	_, ok := s.get(testPeer)
+	assert.False(t, ok, "a forgotten peer must be unknown")
+	_, ok = s.get(otherPeer)
+	assert.True(t, ok, "forgetting one peer must not touch the others")
+
+	s.reset()
+	_, ok = s.get(otherPeer)
+	assert.False(t, ok, "reset must drop every peer")
+}
+
+func TestIPNetsToPrefixes(t *testing.T) {
+	tests := []struct {
+		name  string
+		ipNet net.IPNet
+		want  string
+	}{
+		{
+			name:  "v4",
+			ipNet: net.IPNet{IP: net.IP{10, 20, 0, 0}, Mask: net.CIDRMask(16, 32)},
+			want:  "10.20.0.0/16",
+		},
+		{
+			name:  "v4 mapped under a 128 bit mask",
+			ipNet: net.IPNet{IP: net.ParseIP("10.20.0.0"), Mask: net.CIDRMask(112, 128)},
+			want:  "10.20.0.0/16",
+		},
+		{
+			name:  "v6",
+			ipNet: net.IPNet{IP: net.ParseIP("fd00::"), Mask: net.CIDRMask(64, 128)},
+			want:  "fd00::/64",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := ipNetsToPrefixes([]net.IPNet{tc.ipNet})
+			require.Len(t, got, 1, "the address must be converted, not dropped")
+			assert.Equal(t, tc.want, got[0].String(), "converted prefix")
+		})
+	}
+}
+
+func TestIPNetsToPrefixesRoundTrip(t *testing.T) {
+	prefixes := []netip.Prefix{
+		netip.MustParsePrefix("100.64.0.1/32"),
+		netip.MustParsePrefix("10.20.0.0/16"),
+		netip.MustParsePrefix("fd00::/64"),
+	}
+
+	assert.Equal(t, prefixes, ipNetsToPrefixes(prefixesToIPNets(prefixes)),
+		"prefixes handed to a device must come back unchanged")
+}
+
+func TestAllowedIPStoreNormalizesMappedPrefixes(t *testing.T) {
+	s := newAllowedIPStore()
+	v4 := netip.MustParsePrefix("10.20.0.0/16")
+	mapped := netip.PrefixFrom(netip.AddrFrom16(v4.Addr().As16()), 112)
+
+	s.set(testPeer, []netip.Prefix{mapped})
+	// A v4 rule only matches a v4-mapped address once it has been unmapped, so the store must
+	// hold the plain form and recognise the two spellings as the same prefix.
+	s.add(testPeer, []netip.Prefix{v4})
+
+	prefixes, ok := s.get(testPeer)
+	require.True(t, ok, "peer must be known after set")
+	assert.Equal(t, []netip.Prefix{v4}, prefixes, "a mapped prefix must be stored unmapped and not duplicated")
+}
+
+func TestNormalizePrefix(t *testing.T) {
+	v4 := netip.MustParsePrefix("10.20.0.0/16")
+	v6 := netip.MustParsePrefix("fd00::/64")
+
+	assert.Equal(t, v4, normalizePrefix(v4), "a plain v4 prefix is unchanged")
+	assert.Equal(t, v6, normalizePrefix(v6), "a real v6 prefix is unchanged")
+	assert.Equal(t, v4, normalizePrefix(netip.PrefixFrom(netip.AddrFrom16(v4.Addr().As16()), 112)),
+		"a mapped prefix under a 128 bit mask becomes plain v4")
+	// A prefix shorter than /96 inside the mapped range is a genuine v6 prefix. Unmapping it
+	// would pair a v4 address with a v6 sized mask, which is invalid, and the store would then
+	// record a zero prefix that can never recreate the allowed IP.
+	for _, tc := range []string{"::ffff:0:0/64", "::ffff:1.2.3.4/80", "::ffff:1.2.3.4/95"} {
+		got := normalizePrefix(netip.MustParsePrefix(tc))
+		assert.True(t, got.IsValid(), "%s must normalize to a valid prefix", tc)
+		assert.False(t, got.Addr().Is4(), "%s must stay v6", tc)
+	}
+}
+
+func TestAllowedIPStoreAddExistingDoesNotCreate(t *testing.T) {
+	s := newAllowedIPStore()
+	routed := netip.MustParsePrefix("10.20.0.0/16")
+
+	// An update-only device operation on an absent peer is a silent no-op, so nothing may be
+	// recorded for a peer the store does not already know.
+	s.addExisting(testPeer, []netip.Prefix{routed})
+	_, ok := s.get(testPeer)
+	assert.False(t, ok, "addExisting must not record an unknown peer")
+
+	overlay := netip.MustParsePrefix("100.64.0.1/32")
+	s.set(testPeer, []netip.Prefix{overlay})
+	s.addExisting(testPeer, []netip.Prefix{routed})
+
+	prefixes, _ := s.get(testPeer)
+	assert.Equal(t, []netip.Prefix{overlay, routed}, prefixes, "addExisting must union onto a known peer")
+}
+
+func TestAllowedIPStoreHandsPrefixOverToTheNewOwner(t *testing.T) {
+	s := newAllowedIPStore()
+	routed := netip.MustParsePrefix("10.20.0.0/16")
+	other := otherPeer
+
+	s.set(testPeer, []netip.Prefix{netip.MustParsePrefix("100.64.0.1/32"), routed})
+	s.set(other, []netip.Prefix{netip.MustParsePrefix("100.64.0.2/32")})
+
+	// The device takes an allowed IP away from its previous holder when it is configured on
+	// another peer, so the store must do the same rather than list it under both.
+	s.addExisting(other, []netip.Prefix{routed})
+
+	previous, _ := s.get(testPeer)
+	assert.NotContains(t, previous, routed, "the previous owner must lose the prefix")
+	current, _ := s.get(other)
+	assert.Contains(t, current, routed, "the new owner must hold the prefix")
+}
+
+func TestAllowedIPStoreForgetReleasesOwnership(t *testing.T) {
+	s := newAllowedIPStore()
+	routed := netip.MustParsePrefix("10.20.0.0/16")
+
+	s.set(testPeer, []netip.Prefix{routed})
+	s.forget(testPeer)
+	s.set(otherPeer, []netip.Prefix{routed})
+
+	// A forgotten peer must not be resurrected as a key in the peer map by a later claim.
+	_, ok := s.get(testPeer)
+	assert.False(t, ok, "the forgotten peer must stay unknown")
+	current, _ := s.get(otherPeer)
+	assert.Equal(t, []netip.Prefix{routed}, current, "the new owner must hold the prefix")
+}
+
+func TestNormalizePrefixClearsHostBits(t *testing.T) {
+	// A device stores a prefix masked, so a caller passing host bits must still match what a
+	// device fallback seeded, otherwise that prefix could never be removed by value.
+	assert.Equal(t, netip.MustParsePrefix("10.20.0.0/16"),
+		normalizePrefix(netip.MustParsePrefix("10.20.0.1/16")), "host bits must be cleared")
+	assert.Equal(t, netip.MustParsePrefix("fd00::/64"),
+		normalizePrefix(netip.MustParsePrefix("fd00::1/64")), "host bits must be cleared for v6")
+}
+
+func TestIPNetsToPrefixesKeepsV6InTheMappedRange(t *testing.T) {
+	// ::ffff:0:0/64 reads as v4-mapped but is a genuine v6 prefix: unmapping it would leave a
+	// v4 address under a 64 bit mask, which is invalid, and the allowed IP would be dropped.
+	got := ipNetsToPrefixes([]net.IPNet{{
+		IP:   net.ParseIP("::ffff:0:0"),
+		Mask: net.CIDRMask(64, 128),
+	}})
+
+	require.Len(t, got, 1, "the prefix must be converted, not dropped")
+	assert.False(t, got[0].Addr().Is4(), "a v6 prefix in the mapped range must not become v4")
+	assert.Equal(t, 64, got[0].Bits(), "the prefix length must survive the conversion")
+}
+
+func TestPrefixesToIPNetsNormalizes(t *testing.T) {
+	// net.IPNet prints a v4-mapped address as v4 but takes the length from its 16 byte
+	// mask, so an unnormalized ::ffff:10.1.2.3/64 reaches a userspace device as 10.1.2.3/0,
+	// an allowed IP that matches every v4 address.
+	tests := []struct {
+		name  string
+		given string
+		want  string
+	}{
+		{name: "mapped below /96", given: "::ffff:10.1.2.3/64", want: "::/64"},
+		{name: "mapped at /112", given: "::ffff:10.1.2.3/112", want: "10.1.0.0/16"},
+		{name: "host bits are cleared", given: "10.20.0.1/16", want: "10.20.0.0/16"},
+		{name: "v6 is untouched", given: "fd00::1/64", want: "fd00::/64"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := prefixesToIPNets([]netip.Prefix{netip.MustParsePrefix(tc.given)})
+			require.Len(t, got, 1, "the prefix must be converted, not dropped")
+			assert.Equal(t, tc.want, got[0].String(), "what the device is given")
+			assert.NotEqual(t, 0, mustOnes(t, got[0]), "a device must never be given a zero length allowed IP")
+		})
+	}
+}
+
+func mustOnes(t *testing.T, ipNet net.IPNet) int {
+	t.Helper()
+
+	ones, _ := ipNet.Mask.Size()
+	return ones
+}
+
+// TestPrefixesToIPNetsAgreesWithTheStore pins the property the store depends on: what a
+// device is given and what is recorded for it are the same prefix.
+func TestPrefixesToIPNetsAgreesWithTheStore(t *testing.T) {
+	for _, given := range []string{"::ffff:10.1.2.3/64", "::ffff:10.1.2.3/112", "10.20.0.1/16", "fd00::1/64"} {
+		prefix := netip.MustParsePrefix(given)
+
+		toDevice := prefixesToIPNets([]netip.Prefix{prefix})
+		recorded := normalizePrefix(prefix)
+
+		assert.Equal(t, recorded.String(), toDevice[0].String(),
+			"%s must reach the device in the form the store records", given)
+	}
+}

client/iface/configurer/common.go

@@ -19,12 +19,18 @@ func buildPresharedKeyConfig(peerKey wgtypes.Key, psk wgtypes.Key, updateOnly bo
 	}
 }
 
+// prefixesToIPNets converts prefixes on their way to a device. It is the only place that
+// conversion happens, so it also normalizes: the device is then given the same form the
+// store records, and a v4-mapped prefix cannot reach net.IPNet, which prints such an
+// address as v4 while taking the length from its 16 byte mask and so turns
+// ::ffff:10.1.2.3/64 into 10.1.2.3/0 — an allowed IP matching every v4 address.
 func prefixesToIPNets(prefixes []netip.Prefix) []net.IPNet {
 	ipNets := make([]net.IPNet, len(prefixes))
 	for i, prefix := range prefixes {
+		normalized := normalizePrefix(prefix)
 		ipNets[i] = net.IPNet{
-			IP:   prefix.Addr().AsSlice(),                             // Convert netip.Addr to net.IP
-			Mask: net.CIDRMask(prefix.Bits(), prefix.Addr().BitLen()), // Create subnet mask
+			IP:   normalized.Addr().AsSlice(),
+			Mask: net.CIDRMask(normalized.Bits(), normalized.Addr().BitLen()),
 		}
 	}
 	return ipNets

client/iface/configurer/kernel_unix.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"net"
 	"net/netip"
+	"slices"
 	"time"
 
 	log "github.com/sirupsen/logrus"
@@ -18,16 +19,22 @@ import (
 type KernelConfigurer struct {
 	deviceName string
 	statsCache *statsCache
+	allowedIPs *allowedIPStore
 }
 
+// NewKernelConfigurer creates a configurer with an empty allowed IP mirror
+// and a statistics cache for the named kernel device.
 func NewKernelConfigurer(deviceName string) *KernelConfigurer {
 	c := &KernelConfigurer{
 		deviceName: deviceName,
+		allowedIPs: newAllowedIPStore(),
 	}
 	c.statsCache = newStatsCache(statsCacheTTL, c.fetchStats)
 	return c
 }
 
+// ConfigureInterface sets the device key, port and firewall mark, replacing all peers.
+// The allowed IP mirror is reset only after the device accepts the configuration.
 func (c *KernelConfigurer) ConfigureInterface(privateKey string, port int) error {
 	log.Debugf("adding Wireguard private key")
 	key, err := wgtypes.ParseKey(privateKey)
@@ -46,6 +53,8 @@ func (c *KernelConfigurer) ConfigureInterface(privateKey string, port int) error
 	if err != nil {
 		return fmt.Errorf(`received error "%w" while configuring interface %s with port %d`, err, c.deviceName, port)
 	}
+
+	c.allowedIPs.reset()
 	return nil
 }
 
@@ -58,9 +67,20 @@ func (c *KernelConfigurer) SetPresharedKey(peerKey string, psk wgtypes.Key, upda
 	}
 
 	cfg := buildPresharedKeyConfig(parsedPeerKey, psk, updateOnly)
-	return c.configure(cfg)
+	if err := c.configure(cfg); err != nil {
+		return err
+	}
+
+	// Without updateOnly this creates the peer when it is absent, so the store has to
+	// know about it even though no allowed IP was configured.
+	if !updateOnly {
+		c.allowedIPs.ensure(parsedPeerKey)
+	}
+	return nil
 }
 
+// UpdatePeer creates or updates a peer, merging allowed IPs with its existing set.
+// Prefixes assigned to this peer are transferred from their previous owners.
 func (c *KernelConfigurer) UpdatePeer(peerKey string, allowedIps []netip.Prefix, keepAlive time.Duration, endpoint *net.UDPAddr, preSharedKey *wgtypes.Key) error {
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
@@ -83,19 +103,23 @@ func (c *KernelConfigurer) UpdatePeer(peerKey string, allowedIps []netip.Prefix,
 	if err != nil {
 		return fmt.Errorf(`received error "%w" while updating peer on interface %s with settings: allowed ips %s, endpoint %s`, err, c.deviceName, allowedIps, endpoint.String())
 	}
+
+	c.allowedIPs.add(peerKeyParsed, allowedIps)
 	return nil
 }
 
+// RemoveEndpointAddress clears the endpoint of a peer while keeping it configured.
+// Neither the netlink API nor the userspace one can clear an endpoint in place, so the peer
+// is removed and re-added with the allowed IPs it already had.
 func (c *KernelConfigurer) RemoveEndpointAddress(peerKey string) error {
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
 		return err
 	}
 
-	// Get the existing peer to preserve its allowed IPs
-	existingPeer, err := c.getPeer(c.deviceName, peerKey)
+	allowedIPs, err := c.peerAllowedIPs(peerKeyParsed)
 	if err != nil {
-		return fmt.Errorf("get peer: %w", err)
+		return err
 	}
 
 	removePeerCfg := wgtypes.PeerConfig{
@@ -104,26 +128,27 @@ func (c *KernelConfigurer) RemoveEndpointAddress(peerKey string) error {
 	}
 
 	if err := c.configure(wgtypes.Config{Peers: []wgtypes.PeerConfig{removePeerCfg}}); err != nil {
-		return fmt.Errorf(`error removing peer %s from interface %s: %w`, peerKey, c.deviceName, err)
+		return fmt.Errorf("remove peer %s from interface %s: %w", peerKey, c.deviceName, err)
 	}
 
-	//Re-add the peer without the endpoint but same AllowedIPs
 	reAddPeerCfg := wgtypes.PeerConfig{
 		PublicKey:         peerKeyParsed,
-		AllowedIPs:        existingPeer.AllowedIPs,
+		AllowedIPs:        prefixesToIPNets(allowedIPs),
 		ReplaceAllowedIPs: true,
 	}
 
 	if err := c.configure(wgtypes.Config{Peers: []wgtypes.PeerConfig{reAddPeerCfg}}); err != nil {
+		c.allowedIPs.forget(peerKeyParsed)
 		return fmt.Errorf(
-			`error re-adding peer %s to interface %s with allowed IPs %v: %w`,
-			peerKey, c.deviceName, existingPeer.AllowedIPs, err,
+			"re-add peer %s to interface %s with allowed IPs %v: %w",
+			peerKey, c.deviceName, allowedIPs, err,
 		)
 	}
 
 	return nil
 }
 
+// RemovePeer removes a peer and forgets its allowed IPs after a successful device write.
 func (c *KernelConfigurer) RemovePeer(peerKey string) error {
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
@@ -142,15 +167,13 @@ func (c *KernelConfigurer) RemovePeer(peerKey string) error {
 	if err != nil {
 		return fmt.Errorf(`received error "%w" while removing peer %s from interface %s`, err, peerKey, c.deviceName)
 	}
+
+	c.allowedIPs.forget(peerKeyParsed)
 	return nil
 }
 
+// AddAllowedIP adds a prefix to an existing peer; an absent peer is a silent no-op.
 func (c *KernelConfigurer) AddAllowedIP(peerKey string, allowedIP netip.Prefix) error {
-	ipNet := net.IPNet{
-		IP:   allowedIP.Addr().AsSlice(),
-		Mask: net.CIDRMask(allowedIP.Bits(), allowedIP.Addr().BitLen()),
-	}
-
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
 		return err
@@ -159,7 +182,7 @@ func (c *KernelConfigurer) AddAllowedIP(peerKey string, allowedIP netip.Prefix)
 		PublicKey:         peerKeyParsed,
 		UpdateOnly:        true,
 		ReplaceAllowedIPs: false,
-		AllowedIPs:        []net.IPNet{ipNet},
+		AllowedIPs:        prefixesToIPNets([]netip.Prefix{allowedIP}),
 	}
 
 	config := wgtypes.Config{
@@ -169,52 +192,69 @@ func (c *KernelConfigurer) AddAllowedIP(peerKey string, allowedIP netip.Prefix)
 	if err != nil {
 		return fmt.Errorf(`received error "%w" while adding allowed Ip to peer on interface %s with settings: allowed ips %s`, err, c.deviceName, allowedIP)
 	}
+
+	c.allowedIPs.addExisting(peerKeyParsed, []netip.Prefix{allowedIP})
 	return nil
 }
 
+// RemoveAllowedIP removes a prefix while preserving the peer's other allowed IPs.
+// A prefix not assigned to the peer is a no-op.
 func (c *KernelConfigurer) RemoveAllowedIP(peerKey string, allowedIP netip.Prefix) error {
-	ipNet := net.IPNet{
-		IP:   allowedIP.Addr().AsSlice(),
-		Mask: net.CIDRMask(allowedIP.Bits(), allowedIP.Addr().BitLen()),
-	}
-
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
 		return fmt.Errorf("parse peer key: %w", err)
 	}
 
-	existingPeer, err := c.getPeer(c.deviceName, peerKey)
+	currentAllowedIPs, err := c.peerAllowedIPs(peerKeyParsed)
 	if err != nil {
-		return fmt.Errorf("get peer: %w", err)
+		return err
 	}
 
-	newAllowedIPs := existingPeer.AllowedIPs
-
-	for i, existingAllowedIP := range existingPeer.AllowedIPs {
-		if existingAllowedIP.String() == ipNet.String() {
-			newAllowedIPs = append(existingPeer.AllowedIPs[:i], existingPeer.AllowedIPs[i+1:]...) //nolint:gocritic
-			break
-		}
+	idx := slices.Index(currentAllowedIPs, normalizePrefix(allowedIP))
+	if idx < 0 {
+		return nil
 	}
+	newAllowedIPs := slices.Delete(currentAllowedIPs, idx, idx+1)
 
 	peer := wgtypes.PeerConfig{
 		PublicKey:         peerKeyParsed,
 		UpdateOnly:        true,
 		ReplaceAllowedIPs: true,
-		AllowedIPs:        newAllowedIPs,
+		AllowedIPs:        prefixesToIPNets(newAllowedIPs),
 	}
 
 	config := wgtypes.Config{
 		Peers: []wgtypes.PeerConfig{peer},
 	}
-	err = c.configure(config)
-	if err != nil {
+	if err := c.configure(config); err != nil {
 		return fmt.Errorf("remove allowed IP %s on interface %s: %w", allowedIP, c.deviceName, err)
 	}
+
+	c.allowedIPs.set(peerKeyParsed, newAllowedIPs)
 	return nil
 }
 
-func (c *KernelConfigurer) getPeer(ifaceName, peerPubKey string) (wgtypes.Peer, error) {
+// peerAllowedIPs returns the allowed IPs configured for a peer, reading them from the device
+// only for a peer the store has not seen. Dumping the device costs a netlink round trip
+// proportional to the whole network map, and this runs on every relay and ICE transition.
+func (c *KernelConfigurer) peerAllowedIPs(peerKey wgtypes.Key) ([]netip.Prefix, error) {
+	if prefixes, ok := c.allowedIPs.get(peerKey); ok {
+		return prefixes, nil
+	}
+
+	existingPeer, err := c.getPeer(c.deviceName, peerKey)
+	if err != nil {
+		return nil, fmt.Errorf("get peer: %w", err)
+	}
+
+	prefixes := ipNetsToPrefixes(existingPeer.AllowedIPs)
+	c.allowedIPs.set(peerKey, prefixes)
+	return prefixes, nil
+}
+
+// getPeer scans the device for one peer. wgtypes.Key is an array, so the comparison is a
+// plain equality: Key.String would base64 encode into a fresh allocation for every peer.
+func (c *KernelConfigurer) getPeer(ifaceName string, peerPubKey wgtypes.Key) (wgtypes.Peer, error) {
 	wg, err := wgctrl.New()
 	if err != nil {
 		return wgtypes.Peer{}, fmt.Errorf("wgctl: %w", err)
@@ -231,7 +271,7 @@ func (c *KernelConfigurer) getPeer(ifaceName, peerPubKey string) (wgtypes.Peer,
 		return wgtypes.Peer{}, fmt.Errorf("get device %s: %w", ifaceName, err)
 	}
 	for _, peer := range wgDevice.Peers {
-		if peer.PublicKey.String() == peerPubKey {
+		if peer.PublicKey == peerPubKey {
 			return peer, nil
 		}
 	}

client/iface/configurer/usp.go

@@ -8,6 +8,7 @@ import (
 	"net/netip"
 	"os"
 	"runtime"
+	"slices"
 	"strconv"
 	"strings"
 	"time"
@@ -41,31 +42,38 @@ type WGUSPConfigurer struct {
 	deviceName       string
 	activityRecorder *bind.ActivityRecorder
 	statsCache       *statsCache
+	allowedIPs       *allowedIPStore
 
 	uapiListener net.Listener
 }
 
+// NewUSPConfigurer creates a userspace configurer and starts its UAPI listener.
 func NewUSPConfigurer(device *device.Device, deviceName string, activityRecorder *bind.ActivityRecorder) *WGUSPConfigurer {
 	wgCfg := &WGUSPConfigurer{
 		device:           device,
 		deviceName:       deviceName,
 		activityRecorder: activityRecorder,
+		allowedIPs:       newAllowedIPStore(),
 	}
 	wgCfg.statsCache = newStatsCache(statsCacheTTL, wgCfg.fetchStats)
 	wgCfg.startUAPI()
 	return wgCfg
 }
 
+// NewUSPConfigurerNoUAPI creates a userspace configurer without a UAPI listener.
 func NewUSPConfigurerNoUAPI(device *device.Device, deviceName string, activityRecorder *bind.ActivityRecorder) *WGUSPConfigurer {
 	wgCfg := &WGUSPConfigurer{
 		device:           device,
 		deviceName:       deviceName,
 		activityRecorder: activityRecorder,
+		allowedIPs:       newAllowedIPStore(),
 	}
 	wgCfg.statsCache = newStatsCache(statsCacheTTL, wgCfg.fetchStats)
 	return wgCfg
 }
 
+// ConfigureInterface sets the device key, port and firewall mark, replacing all peers.
+// The allowed IP mirror is reset only after the device accepts the configuration.
 func (c *WGUSPConfigurer) ConfigureInterface(privateKey string, port int) error {
 	log.Debugf("adding Wireguard private key")
 	key, err := wgtypes.ParseKey(privateKey)
@@ -80,7 +88,12 @@ func (c *WGUSPConfigurer) ConfigureInterface(privateKey string, port int) error
 		ListenPort:   &port,
 	}
 
-	return c.device.IpcSet(toWgUserspaceString(config))
+	if err := c.device.IpcSet(toWgUserspaceString(config)); err != nil {
+		return err
+	}
+
+	c.allowedIPs.reset()
+	return nil
 }
 
 // SetPresharedKey sets the preshared key for a peer.
@@ -92,14 +105,38 @@ func (c *WGUSPConfigurer) SetPresharedKey(peerKey string, psk wgtypes.Key, updat
 	}
 
 	cfg := buildPresharedKeyConfig(parsedPeerKey, psk, updateOnly)
-	return c.device.IpcSet(toWgUserspaceString(cfg))
+	if err := c.device.IpcSet(toWgUserspaceString(cfg)); err != nil {
+		return err
+	}
+
+	// Without updateOnly this creates the peer when it is absent, so the store has to
+	// know about it even though no allowed IP was configured.
+	if !updateOnly {
+		c.allowedIPs.ensure(parsedPeerKey)
+	}
+	return nil
 }
 
+// UpdatePeer creates or updates a peer, merging allowed IPs with its existing set.
+// It validates the endpoint before writing and records changes after a successful write.
 func (c *WGUSPConfigurer) UpdatePeer(peerKey string, allowedIps []netip.Prefix, keepAlive time.Duration, endpoint *net.UDPAddr, preSharedKey *wgtypes.Key) error {
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
 		return err
 	}
+
+	// Everything that can fail is done before the device is touched, so a failure here
+	// cannot leave the device holding a peer that the activity recorder and the allowed
+	// IP store never learned about.
+	var addrPort netip.AddrPort
+	if endpoint != nil {
+		addr, err := netip.ParseAddr(endpoint.IP.String())
+		if err != nil {
+			return fmt.Errorf("parse endpoint address: %w", err)
+		}
+		addrPort = netip.AddrPortFrom(addr.Unmap(), uint16(endpoint.Port))
+	}
+
 	peer := wgtypes.PeerConfig{
 		PublicKey:         peerKeyParsed,
 		ReplaceAllowedIPs: false,
@@ -119,47 +156,27 @@ func (c *WGUSPConfigurer) UpdatePeer(peerKey string, allowedIps []netip.Prefix,
 	}
 
 	if endpoint != nil {
-		addr, err := netip.ParseAddr(endpoint.IP.String())
-		if err != nil {
-			return fmt.Errorf("failed to parse endpoint address: %w", err)
-		}
-		addrPort := netip.AddrPortFrom(addr.Unmap(), uint16(endpoint.Port))
 		c.activityRecorder.UpsertAddress(peerKey, addrPort)
 	}
+
+	c.allowedIPs.add(peerKeyParsed, allowedIps)
 	return nil
 }
 
+// RemoveEndpointAddress clears the endpoint of a peer while keeping it configured.
+// The UAPI cannot clear an endpoint in place, so the peer is removed and re-added with the
+// allowed IPs it already had.
 func (c *WGUSPConfigurer) RemoveEndpointAddress(peerKey string) error {
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
 		return fmt.Errorf("parse peer key: %w", err)
 	}
 
-	ipcStr, err := c.device.IpcGet()
+	allowedIPs, err := c.peerAllowedIPs(peerKeyParsed)
 	if err != nil {
-		return fmt.Errorf("get IPC config: %w", err)
+		return err
 	}
 
-	// Parse current status to get allowed IPs for the peer
-	stats, err := parseStatus(c.deviceName, ipcStr)
-	if err != nil {
-		return fmt.Errorf("parse IPC config: %w", err)
-	}
-
-	var allowedIPs []net.IPNet
-	found := false
-	for _, peer := range stats.Peers {
-		if peer.PublicKey == peerKey {
-			allowedIPs = peer.AllowedIPs
-			found = true
-			break
-		}
-	}
-	if !found {
-		return fmt.Errorf("peer %s not found", peerKey)
-	}
-
-	// remove the peer from the WireGuard configuration
 	peer := wgtypes.PeerConfig{
 		PublicKey: peerKeyParsed,
 		Remove:    true,
@@ -169,14 +186,13 @@ func (c *WGUSPConfigurer) RemoveEndpointAddress(peerKey string) error {
 		Peers: []wgtypes.PeerConfig{peer},
 	}
 	if ipcErr := c.device.IpcSet(toWgUserspaceString(config)); ipcErr != nil {
-		return fmt.Errorf("failed to remove peer: %s", ipcErr)
+		return fmt.Errorf("remove peer: %w", ipcErr)
 	}
 
-	// Build the peer config
 	peer = wgtypes.PeerConfig{
 		PublicKey:         peerKeyParsed,
 		ReplaceAllowedIPs: true,
-		AllowedIPs:        allowedIPs,
+		AllowedIPs:        prefixesToIPNets(allowedIPs),
 	}
 
 	config = wgtypes.Config{
@@ -184,12 +200,15 @@ func (c *WGUSPConfigurer) RemoveEndpointAddress(peerKey string) error {
 	}
 
 	if err := c.device.IpcSet(toWgUserspaceString(config)); err != nil {
-		return fmt.Errorf("remove endpoint address: %w", err)
+		c.allowedIPs.forget(peerKeyParsed)
+		return fmt.Errorf("re-add peer without endpoint: %w", err)
 	}
 
 	return nil
 }
 
+// RemovePeer removes a peer, then clears its activity and allowed IP records.
+// A failed device write leaves both records intact.
 func (c *WGUSPConfigurer) RemovePeer(peerKey string) error {
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
@@ -204,18 +223,17 @@ func (c *WGUSPConfigurer) RemovePeer(peerKey string) error {
 	config := wgtypes.Config{
 		Peers: []wgtypes.PeerConfig{peer},
 	}
-	ipcErr := c.device.IpcSet(toWgUserspaceString(config))
-
-	c.activityRecorder.Remove(peerKey)
-	return ipcErr
-}
-
-func (c *WGUSPConfigurer) AddAllowedIP(peerKey string, allowedIP netip.Prefix) error {
-	ipNet := net.IPNet{
-		IP:   allowedIP.Addr().AsSlice(),
-		Mask: net.CIDRMask(allowedIP.Bits(), allowedIP.Addr().BitLen()),
+	if ipcErr := c.device.IpcSet(toWgUserspaceString(config)); ipcErr != nil {
+		return ipcErr
 	}
 
+	c.activityRecorder.Remove(peerKey)
+	c.allowedIPs.forget(peerKeyParsed)
+	return nil
+}
+
+// AddAllowedIP adds a prefix to an existing peer; an absent peer is a silent no-op.
+func (c *WGUSPConfigurer) AddAllowedIP(peerKey string, allowedIP netip.Prefix) error {
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
 	if err != nil {
 		return err
@@ -224,79 +242,89 @@ func (c *WGUSPConfigurer) AddAllowedIP(peerKey string, allowedIP netip.Prefix) e
 		PublicKey:         peerKeyParsed,
 		UpdateOnly:        true,
 		ReplaceAllowedIPs: false,
-		AllowedIPs:        []net.IPNet{ipNet},
+		AllowedIPs:        prefixesToIPNets([]netip.Prefix{allowedIP}),
 	}
 
 	config := wgtypes.Config{
 		Peers: []wgtypes.PeerConfig{peer},
 	}
 
-	return c.device.IpcSet(toWgUserspaceString(config))
+	if err := c.device.IpcSet(toWgUserspaceString(config)); err != nil {
+		return err
+	}
+
+	c.allowedIPs.addExisting(peerKeyParsed, []netip.Prefix{allowedIP})
+	return nil
 }
 
+// RemoveAllowedIP removes a prefix while preserving the peer's other allowed IPs.
+// It returns ErrAllowedIPNotFound if the prefix is not assigned to the peer.
 func (c *WGUSPConfigurer) RemoveAllowedIP(peerKey string, allowedIP netip.Prefix) error {
-	ipc, err := c.device.IpcGet()
-	if err != nil {
-		return err
-	}
-
 	peerKeyParsed, err := wgtypes.ParseKey(peerKey)
+	if err != nil {
+		return fmt.Errorf("parse peer key: %w", err)
+	}
+
+	currentAllowedIPs, err := c.peerAllowedIPs(peerKeyParsed)
 	if err != nil {
 		return err
 	}
-	hexKey := hex.EncodeToString(peerKeyParsed[:])
 
-	lines := strings.Split(ipc, "\n")
+	idx := slices.Index(currentAllowedIPs, normalizePrefix(allowedIP))
+	if idx < 0 {
+		return ErrAllowedIPNotFound
+	}
+	newAllowedIPs := slices.Delete(currentAllowedIPs, idx, idx+1)
 
 	peer := wgtypes.PeerConfig{
 		PublicKey:         peerKeyParsed,
 		UpdateOnly:        true,
 		ReplaceAllowedIPs: true,
-		AllowedIPs:        []net.IPNet{},
+		AllowedIPs:        prefixesToIPNets(newAllowedIPs),
 	}
 
-	foundPeer := false
-	removedAllowedIP := false
-	ip := allowedIP.String()
-
-	for _, line := range lines {
-		line = strings.TrimSpace(line)
-
-		// If we're within the details of the found peer and encounter another public key,
-		// this means we're starting another peer's details. So, reset the flag.
-		if strings.HasPrefix(line, "public_key=") && foundPeer {
-			foundPeer = false
-		}
-
-		// Identify the peer with the specific public key
-		if line == fmt.Sprintf("public_key=%s", hexKey) {
-			foundPeer = true
-		}
-
-		// If we're within the details of the found peer and find the specific allowed IP, skip this line
-		if foundPeer && line == "allowed_ip="+ip {
-			removedAllowedIP = true
-			continue
-		}
-
-		// Append the line to the output string
-		if foundPeer && strings.HasPrefix(line, "allowed_ip=") {
-			allowedIPStr := strings.TrimPrefix(line, "allowed_ip=")
-			_, ipNet, err := net.ParseCIDR(allowedIPStr)
-			if err != nil {
-				return err
-			}
-			peer.AllowedIPs = append(peer.AllowedIPs, *ipNet)
-		}
-	}
-
-	if !removedAllowedIP {
-		return ErrAllowedIPNotFound
-	}
 	config := wgtypes.Config{
 		Peers: []wgtypes.PeerConfig{peer},
 	}
-	return c.device.IpcSet(toWgUserspaceString(config))
+	if err := c.device.IpcSet(toWgUserspaceString(config)); err != nil {
+		return fmt.Errorf("remove allowed IP %s: %w", allowedIP, err)
+	}
+
+	c.allowedIPs.set(peerKeyParsed, newAllowedIPs)
+	return nil
+}
+
+// peerAllowedIPs returns the allowed IPs configured for a peer, reading them from the device
+// only for a peer the store has not seen. Reading them back means dumping and parsing the
+// whole device configuration, and this runs on every relay and ICE transition.
+func (c *WGUSPConfigurer) peerAllowedIPs(peerKey wgtypes.Key) ([]netip.Prefix, error) {
+	if prefixes, ok := c.allowedIPs.get(peerKey); ok {
+		return prefixes, nil
+	}
+
+	ipcStr, err := c.device.IpcGet()
+	if err != nil {
+		return nil, fmt.Errorf("get IPC config: %w", err)
+	}
+
+	stats, err := parseStatus(c.deviceName, ipcStr)
+	if err != nil {
+		return nil, fmt.Errorf("parse IPC config: %w", err)
+	}
+
+	// parseStatus reports keys in their textual form, so the comparison needs it once.
+	wanted := peerKey.String()
+	for _, peer := range stats.Peers {
+		if peer.PublicKey != wanted {
+			continue
+		}
+
+		prefixes := ipNetsToPrefixes(peer.AllowedIPs)
+		c.allowedIPs.set(peerKey, prefixes)
+		return prefixes, nil
+	}
+
+	return nil, ErrPeerNotFound
 }
 
 func (c *WGUSPConfigurer) FullStats() (*Stats, error) {

client/iface/configurer/usp_allowedips_test.go

@@ -0,0 +1,318 @@
+package configurer
+
+import (
+	"net"
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	wgconn "golang.zx2c4.com/wireguard/conn"
+	wgdevice "golang.zx2c4.com/wireguard/device"
+	"golang.zx2c4.com/wireguard/tun/tuntest"
+	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
+
+	"github.com/netbirdio/netbird/client/iface/bind"
+)
+
+// newTestUSPConfigurer builds a configurer over a real wireguard-go device backed by an
+// in-memory TUN. The device stays down, so no socket is opened and no privileges are needed.
+func newTestUSPConfigurer(t *testing.T) *WGUSPConfigurer {
+	t.Helper()
+
+	tun := tuntest.NewChannelTUN()
+	dev := wgdevice.NewDevice(tun.TUN(), wgconn.NewDefaultBind(), wgdevice.NewLogger(wgdevice.LogLevelSilent, ""))
+	t.Cleanup(dev.Close)
+
+	c := NewUSPConfigurerNoUAPI(dev, "wgtest0", bind.NewActivityRecorder())
+
+	key, err := wgtypes.GeneratePrivateKey()
+	require.NoError(t, err, "generate device private key")
+	require.NoError(t, c.ConfigureInterface(key.String(), 0), "configure test device")
+
+	return c
+}
+
+// seedPeers adds count peers, each with a /32 overlay address, and returns their public keys.
+func seedPeers(t *testing.T, c *WGUSPConfigurer, count int) []string {
+	t.Helper()
+
+	keys := make([]string, 0, count)
+	for i := 0; i < count; i++ {
+		priv, err := wgtypes.GeneratePrivateKey()
+		require.NoError(t, err, "generate peer private key")
+		pub := priv.PublicKey().String()
+
+		addr := netip.PrefixFrom(netip.AddrFrom4([4]byte{100, 64, byte(i >> 8), byte(i)}), 32)
+		require.NoError(t, c.UpdatePeer(pub, []netip.Prefix{addr}, 25*time.Second, nil, nil), "add peer")
+		keys = append(keys, pub)
+	}
+	return keys
+}
+
+func peerAllowedIPs(t *testing.T, c *WGUSPConfigurer, peerKey string) []string {
+	t.Helper()
+
+	stats, err := c.FullStats()
+	require.NoError(t, err, "read device stats")
+
+	for _, p := range stats.Peers {
+		if p.PublicKey != peerKey {
+			continue
+		}
+		got := make([]string, 0, len(p.AllowedIPs))
+		for _, ipNet := range p.AllowedIPs {
+			got = append(got, ipNet.String())
+		}
+		return got
+	}
+	t.Fatalf("peer %s not found on device", peerKey)
+	return nil
+}
+
+// TestRemoveEndpointAddressPreservesRoutedAllowedIPs covers the prefixes the route manager
+// attaches to a routing peer through AddAllowedIP. Those are not known to the peer.Conn that
+// triggers the endpoint removal, so dropping them here would silently blackhole every route
+// behind that peer on each relay or ICE disconnect.
+func TestRemoveEndpointAddressPreservesRoutedAllowedIPs(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+	peerKey := seedPeers(t, c, 3)[1]
+
+	routed := []netip.Prefix{
+		netip.MustParsePrefix("10.20.0.0/16"),
+		netip.MustParsePrefix("192.168.7.0/24"),
+	}
+	for _, prefix := range routed {
+		require.NoError(t, c.AddAllowedIP(peerKey, prefix), "add routed prefix")
+	}
+
+	before := peerAllowedIPs(t, c, peerKey)
+	require.Len(t, before, 3, "peer should hold its overlay address plus both routed prefixes")
+
+	require.NoError(t, c.RemoveEndpointAddress(peerKey), "remove endpoint address")
+
+	assert.ElementsMatch(t, before, peerAllowedIPs(t, c, peerKey),
+		"allowed IPs must survive the endpoint removal unchanged")
+}
+
+// TestRemoveEndpointAddressDoesNotScaleWithPeerCount is the regression guard for the actual
+// defect: clearing one peer's endpoint used to dump and parse the whole device, so its cost
+// grew with the size of the network map. On a routing peer with thousands of peers that dump
+// runs on every relay and ICE transition, under the interface lock.
+func TestRemoveEndpointAddressDoesNotScaleWithPeerCount(t *testing.T) {
+	measure := func(peerCount int) float64 {
+		c := newTestUSPConfigurer(t)
+		peerKey := seedPeers(t, c, peerCount)[peerCount/2]
+
+		return testing.AllocsPerRun(5, func() {
+			require.NoError(t, c.RemoveEndpointAddress(peerKey), "remove endpoint address")
+		})
+	}
+
+	small := measure(64)
+	large := measure(1024)
+
+	assert.Less(t, large, small*2,
+		"clearing one endpoint allocated %.0f objects with 1024 peers against %.0f with 64: the cost still scales with the peer count",
+		large, small)
+}
+
+// TestRemoveEndpointAddressFallsBackToDevice covers a peer the store never saw, which is what
+// an out-of-band reconfiguration of the device leaves behind. The device stays the source of
+// truth in that case, so the allowed IPs must still be preserved.
+func TestRemoveEndpointAddressFallsBackToDevice(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+	peerKey := seedPeers(t, c, 3)[1]
+	require.NoError(t, c.AddAllowedIP(peerKey, netip.MustParsePrefix("10.20.0.0/16")), "add routed prefix")
+
+	before := peerAllowedIPs(t, c, peerKey)
+	c.allowedIPs.reset()
+
+	require.NoError(t, c.RemoveEndpointAddress(peerKey), "remove endpoint address")
+
+	assert.ElementsMatch(t, before, peerAllowedIPs(t, c, peerKey),
+		"allowed IPs recovered from the device must be preserved")
+
+	recovered, ok := c.allowedIPs.get(mustParseKey(t, peerKey))
+	assert.True(t, ok, "the fallback must seed the store so the next call skips the device dump")
+	assert.Len(t, recovered, 2, "seeded prefixes")
+}
+
+func TestRemoveAllowedIPKeepsTheOtherPrefixes(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+	peerKey := seedPeers(t, c, 3)[0]
+	routed := netip.MustParsePrefix("10.20.0.0/16")
+	require.NoError(t, c.AddAllowedIP(peerKey, routed), "add routed prefix")
+	require.NoError(t, c.AddAllowedIP(peerKey, netip.MustParsePrefix("192.168.7.0/24")), "add routed prefix")
+
+	require.NoError(t, c.RemoveAllowedIP(peerKey, routed), "remove routed prefix")
+
+	assert.ElementsMatch(t, []string{"100.64.0.0/32", "192.168.7.0/24"}, peerAllowedIPs(t, c, peerKey),
+		"only the removed prefix should be gone")
+
+	assert.ErrorIs(t, c.RemoveAllowedIP(peerKey, routed), ErrAllowedIPNotFound,
+		"removing a prefix that is no longer configured must be reported")
+}
+
+// TestAddAllowedIPOnAbsentPeerDoesNotResurrectIt covers the lazy connection window documented
+// in #6863: AddAllowedIP is update-only, a silent no-op when the peer is absent, so it must not
+// leave the store claiming prefixes the device never took. RemoveEndpointAddress re-adds a peer
+// without update-only, so a phantom entry would create a peer the device had dropped, and a
+// created peer would steal those allowed IPs from whichever peer legitimately holds them.
+func TestAddAllowedIPOnAbsentPeerDoesNotResurrectIt(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+	seedPeers(t, c, 2)
+
+	priv, err := wgtypes.GeneratePrivateKey()
+	require.NoError(t, err, "generate peer private key")
+	absent := priv.PublicKey().String()
+
+	require.NoError(t, c.AddAllowedIP(absent, netip.MustParsePrefix("10.20.0.0/16")),
+		"update-only add on an absent peer is a silent no-op")
+
+	stats, err := c.FullStats()
+	require.NoError(t, err, "read device stats")
+	require.Len(t, stats.Peers, 2, "the absent peer must not have been created by AddAllowedIP")
+
+	assert.ErrorIs(t, c.RemoveEndpointAddress(absent), ErrPeerNotFound,
+		"clearing the endpoint of a peer the device does not have must fail")
+
+	stats, err = c.FullStats()
+	require.NoError(t, err, "read device stats")
+	assert.Len(t, stats.Peers, 2, "no peer may be created while clearing an endpoint")
+}
+
+// TestRemoveEndpointAddressDoesNotStealAPrefixFromAnotherPeer covers WireGuard's rule that an
+// allowed IP belongs to exactly one peer: configuring a prefix on a peer takes it away from
+// whichever peer held it before. UpdatePeer relies on that rule rather than removing the prefix
+// from the previous holder itself, so a prefix handed over between peers must not come back.
+func TestRemoveEndpointAddressDoesNotStealAPrefixFromAnotherPeer(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+	keys := seedPeers(t, c, 2)
+	peerA, peerB := keys[0], keys[1]
+	routed := netip.MustParsePrefix("10.20.0.0/16")
+
+	require.NoError(t, c.AddAllowedIP(peerA, routed), "give the prefix to A")
+	require.Contains(t, peerAllowedIPs(t, c, peerA), routed.String(), "A must hold the prefix")
+
+	// The route moves to B. The device takes it away from A on its own.
+	require.NoError(t, c.AddAllowedIP(peerB, routed), "hand the prefix over to B")
+	require.Contains(t, peerAllowedIPs(t, c, peerB), routed.String(), "B must hold the prefix")
+	require.NotContains(t, peerAllowedIPs(t, c, peerA), routed.String(), "the device must have taken it from A")
+
+	require.NoError(t, c.RemoveEndpointAddress(peerA), "clear A's endpoint")
+
+	assert.NotContains(t, peerAllowedIPs(t, c, peerA), routed.String(),
+		"clearing A's endpoint must not take the prefix back from B")
+	assert.Contains(t, peerAllowedIPs(t, c, peerB), routed.String(),
+		"B must still hold the prefix")
+}
+
+// TestPresharedKeyCreatedPeerTakesPartInPrefixHandover covers a peer created by a preshared
+// key write rather than by a peer update. Rosenpass applies a peer's first key without
+// updateOnly, which creates the peer on the device, so a store that ignored that operation
+// would treat the peer as unknown and would not account for a prefix later handed over to it.
+func TestPresharedKeyCreatedPeerTakesPartInPrefixHandover(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+	peerA := seedPeers(t, c, 1)[0]
+	routed := netip.MustParsePrefix("10.20.0.0/16")
+	require.NoError(t, c.AddAllowedIP(peerA, routed), "give the prefix to A")
+
+	priv, err := wgtypes.GeneratePrivateKey()
+	require.NoError(t, err, "generate peer private key")
+	peerB := priv.PublicKey().String()
+
+	psk, err := wgtypes.GenerateKey()
+	require.NoError(t, err, "generate preshared key")
+	require.NoError(t, c.SetPresharedKey(peerB, psk, false), "a first key creates the peer")
+
+	require.NoError(t, c.AddAllowedIP(peerB, routed), "hand the prefix over to B")
+	require.Contains(t, peerAllowedIPs(t, c, peerB), routed.String(), "B must hold the prefix")
+
+	require.NoError(t, c.RemoveEndpointAddress(peerA), "clear A's endpoint")
+
+	assert.NotContains(t, peerAllowedIPs(t, c, peerA), routed.String(),
+		"clearing A's endpoint must not take the prefix back from B")
+	assert.Contains(t, peerAllowedIPs(t, c, peerB), routed.String(), "B must still hold the prefix")
+}
+
+// TestUpdatePeerDoesNotWidenAMappedPrefixOnTheDevice is the end to end form of the
+// conversion: a v4-mapped prefix must not reach the device as a zero length allowed IP,
+// which would route every v4 address to that peer.
+func TestUpdatePeerDoesNotWidenAMappedPrefixOnTheDevice(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+
+	priv, err := wgtypes.GeneratePrivateKey()
+	require.NoError(t, err, "generate peer private key")
+	peerKey := priv.PublicKey().String()
+
+	mapped := netip.MustParsePrefix("::ffff:10.1.2.3/112")
+	require.NoError(t, c.UpdatePeer(peerKey, []netip.Prefix{mapped}, 25*time.Second, nil, nil), "add peer")
+
+	onDevice := peerAllowedIPs(t, c, peerKey)
+	assert.NotContains(t, onDevice, "0.0.0.0/0", "the device must not be given a catch-all allowed IP")
+	assert.Equal(t, []string{"10.1.0.0/16"}, onDevice, "the device holds the normalized prefix")
+
+	recorded, ok := c.allowedIPs.get(mustParseKey(t, peerKey))
+	require.True(t, ok, "the peer must be recorded")
+	require.Len(t, recorded, 1, "one prefix recorded")
+	assert.Equal(t, onDevice[0], recorded[0].String(), "device and store must agree")
+}
+
+// TestUpdatePeerWithAnUnusableEndpointTouchesNothing pins the ordering: the endpoint is
+// parsed before the device is configured, so a failure cannot leave the device holding a
+// peer that the store never learned about, with the prefix handover skipped along with it.
+func TestUpdatePeerWithAnUnusableEndpointTouchesNothing(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+	seedPeers(t, c, 2)
+
+	priv, err := wgtypes.GeneratePrivateKey()
+	require.NoError(t, err, "generate peer private key")
+	peerKey := priv.PublicKey().String()
+
+	// A three byte address has no textual form netip can parse back.
+	endpoint := &net.UDPAddr{IP: net.IP{1, 2, 3}, Port: 51820}
+	require.Error(t, c.UpdatePeer(peerKey, []netip.Prefix{netip.MustParsePrefix("10.30.0.0/16")},
+		25*time.Second, endpoint, nil), "an unusable endpoint must fail the update")
+
+	stats, err := c.FullStats()
+	require.NoError(t, err, "read device stats")
+	assert.Len(t, stats.Peers, 2, "the peer must not have reached the device")
+
+	_, ok := c.allowedIPs.get(mustParseKey(t, peerKey))
+	assert.False(t, ok, "the peer must not have been recorded either")
+}
+
+// TestRemovePeerKeepsTheRecordWhenTheDeviceRefuses covers a removal that never reached the
+// device. A single peer removal is one write, so a failure leaves the peer on the device
+// exactly as it was, and the record still describes it; dropping it would only force the
+// next caller to read the whole device back for an answer it already had.
+func TestRemovePeerKeepsTheRecordWhenTheDeviceRefuses(t *testing.T) {
+	c := newTestUSPConfigurer(t)
+	peerKey := seedPeers(t, c, 1)[0]
+	require.NoError(t, c.AddAllowedIP(peerKey, netip.MustParsePrefix("10.20.0.0/16")), "add routed prefix")
+
+	before, ok := c.allowedIPs.get(mustParseKey(t, peerKey))
+	require.True(t, ok, "the peer must be recorded before the removal")
+	require.Len(t, before, 2, "overlay address plus routed prefix")
+
+	// A closed device refuses every write, which is the shape of any failed removal.
+	c.device.Close()
+
+	require.Error(t, c.RemovePeer(peerKey), "the removal must report the failure")
+
+	after, ok := c.allowedIPs.get(mustParseKey(t, peerKey))
+	require.True(t, ok, "a peer still on the device must stay recorded")
+	assert.Equal(t, before, after, "the record must describe the peer the device kept")
+}
+
+// mustParseKey turns the textual key the configurer API takes into the form the store
+// keys on.
+func mustParseKey(t *testing.T, key string) wgtypes.Key {
+	t.Helper()
+
+	parsed, err := wgtypes.ParseKey(key)
+	require.NoError(t, err, "parse peer key")
+	return parsed
+}

client/iface/iface.go

@@ -51,7 +51,6 @@ func ValidateMTU(mtu uint16) error {
 
 type wgProxyFactory interface {
 	GetProxy() wgproxy.Proxy
-	GetProxyPort() uint16
 	Free() error
 }
 
@@ -81,12 +80,6 @@ func (w *WGIface) GetProxy() wgproxy.Proxy {
 	return w.wgProxyFactory.GetProxy()
 }
 
-// GetProxyPort returns the proxy port used by the WireGuard proxy.
-// Returns 0 if no proxy port is used (e.g., for userspace WireGuard).
-func (w *WGIface) GetProxyPort() uint16 {
-	return w.wgProxyFactory.GetProxyPort()
-}
-
 // GetBind returns the EndpointManager userspace bind mode.
 func (w *WGIface) GetBind() device.EndpointManager {
 	w.mu.Lock()

client/iface/iface_close_test.go

@@ -52,7 +52,6 @@ func (f *fakeTunDevice) Close() error {
 type fakeProxyFactory struct{}
 
 func (fakeProxyFactory) GetProxy() wgproxy.Proxy { return nil }
-func (fakeProxyFactory) GetProxyPort() uint16    { return 0 }
 func (fakeProxyFactory) Free() error             { return nil }
 
 // TestWGIface_CloseReleasesMutexBeforeTunClose guards against a deadlock

client/iface/iface_test.go

@@ -40,14 +40,18 @@ func init() {
 	peerPubKey = peerPrivateKey.PublicKey().String()
 }
 
+// testIFaceBlackList mirrors the prefixes profilemanager.DefaultInterfaceBlacklist
+// carries for the overlay interface. These tests create their own utun device, and
+// stdnet's filter probes with wgctrl every interface it is not told to skip, which
+// on a userspace WireGuard platform reaches the UAPI socket of this same process.
+// Declared here rather than imported because profilemanager imports this package.
+var testIFaceBlackList = []string{"wt", "utun", "tun0"}
+
 func TestWGIface_UpdateAddr(t *testing.T) {
 	ifaceName := fmt.Sprintf("utun%d", WgIntNumber+4)
 	addr := "100.64.0.1/8"
 	wgPort := 33100
-	newNet, err := stdnet.NewNet(context.Background(), nil)
-	if err != nil {
-		t.Fatal(err)
-	}
+	newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 
 	opts := WGIFaceOpts{
 		IFaceName:    ifaceName,
@@ -127,10 +131,7 @@ func getIfaceAddrs(ifaceName string) ([]net.Addr, error) {
 func Test_CreateInterface(t *testing.T) {
 	ifaceName := fmt.Sprintf("utun%d", WgIntNumber+1)
 	wgIP := "10.99.99.1/32"
-	newNet, err := stdnet.NewNet(context.Background(), nil)
-	if err != nil {
-		t.Fatal(err)
-	}
+	newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 	opts := WGIFaceOpts{
 		IFaceName:    ifaceName,
 		Address:      wgaddr.MustParseWGAddress(wgIP),
@@ -170,10 +171,7 @@ func Test_Close(t *testing.T) {
 	ifaceName := fmt.Sprintf("utun%d", WgIntNumber+2)
 	wgIP := "10.99.99.2/32"
 	wgPort := 33100
-	newNet, err := stdnet.NewNet(context.Background(), nil)
-	if err != nil {
-		t.Fatal(err)
-	}
+	newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 
 	opts := WGIFaceOpts{
 		IFaceName:    ifaceName,
@@ -215,10 +213,7 @@ func TestRecreation(t *testing.T) {
 			ifaceName := fmt.Sprintf("utun%d", WgIntNumber+2)
 			wgIP := "10.99.99.2/32"
 			wgPort := 33100
-			newNet, err := stdnet.NewNet(context.Background(), nil)
-			if err != nil {
-				t.Fatal(err)
-			}
+			newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 
 			opts := WGIFaceOpts{
 				IFaceName:    ifaceName,
@@ -288,10 +283,7 @@ func Test_ConfigureInterface(t *testing.T) {
 	ifaceName := fmt.Sprintf("utun%d", WgIntNumber+3)
 	wgIP := "10.99.99.5/30"
 	wgPort := 33100
-	newNet, err := stdnet.NewNet(context.Background(), nil)
-	if err != nil {
-		t.Fatal(err)
-	}
+	newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 	opts := WGIFaceOpts{
 		IFaceName:    ifaceName,
 		Address:      wgaddr.MustParseWGAddress(wgIP),
@@ -343,10 +335,7 @@ func Test_ConfigureInterface(t *testing.T) {
 func Test_UpdatePeer(t *testing.T) {
 	ifaceName := fmt.Sprintf("utun%d", WgIntNumber+4)
 	wgIP := "10.99.99.9/30"
-	newNet, err := stdnet.NewNet(context.Background(), nil)
-	if err != nil {
-		t.Fatal(err)
-	}
+	newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 
 	opts := WGIFaceOpts{
 		IFaceName:    ifaceName,
@@ -413,10 +402,7 @@ func Test_UpdatePeer(t *testing.T) {
 func Test_RemovePeer(t *testing.T) {
 	ifaceName := fmt.Sprintf("utun%d", WgIntNumber+4)
 	wgIP := "10.99.99.13/30"
-	newNet, err := stdnet.NewNet(context.Background(), nil)
-	if err != nil {
-		t.Fatal(err)
-	}
+	newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 
 	opts := WGIFaceOpts{
 		IFaceName:    ifaceName,
@@ -477,10 +463,7 @@ func Test_ConnectPeers(t *testing.T) {
 	peer2wgPort := 33200
 
 	keepAlive := 1 * time.Second
-	newNet, err := stdnet.NewNet(context.Background(), nil)
-	if err != nil {
-		t.Fatal(err)
-	}
+	newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 
 	guid := fmt.Sprintf("{%s}", uuid.New().String())
 	device.CustomWindowsGUIDString = strings.ToLower(guid)
@@ -516,10 +499,7 @@ func Test_ConnectPeers(t *testing.T) {
 	guid = fmt.Sprintf("{%s}", uuid.New().String())
 	device.CustomWindowsGUIDString = strings.ToLower(guid)
 
-	newNet, err = stdnet.NewNet(context.Background(), nil)
-	if err != nil {
-		t.Fatal(err)
-	}
+	newNet = stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 
 	optsPeer2 := WGIFaceOpts{
 		IFaceName:    peer2ifaceName,
@@ -568,11 +548,14 @@ func Test_ConnectPeers(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	// The peers use userspace WireGuard (stdnet transport). A tight busy-loop
-	// here starves the wireguard-go goroutines that process the handshake, so
-	// poll on a ticker instead and yield the CPU between checks. WireGuard also
-	// only retries a lost handshake initiation every REKEY_TIMEOUT (5s), which
-	// is why the overall wait can occasionally stretch to tens of seconds.
+	// On Linux with the kernel module both peers are kernel devices, elsewhere
+	// they run on wireguard-go. A tight busy-loop here would starve the
+	// wireguard-go goroutines that process the handshake, so poll on a ticker
+	// instead and yield the CPU between checks. WireGuard also only retries a
+	// lost handshake initiation every REKEY_TIMEOUT (5s), which is why the
+	// overall wait can occasionally stretch to tens of seconds. Each side sends
+	// its first initiation when its peer is configured, and the first one leaves
+	// before the other device knows the peer, so that one is always wasted.
 	timeout := 30 * time.Second
 	timeoutChannel := time.After(timeout)
 	ticker := time.NewTicker(500 * time.Millisecond)
@@ -590,13 +573,26 @@ func Test_ConnectPeers(t *testing.T) {
 
 		select {
 		case <-timeoutChannel:
-			t.Fatalf("waiting for peer handshake timeout after %s", timeout.String())
+			// The counters tell whether initiations were sent at all, whether they
+			// arrived, and whether only one direction is working.
+			t.Fatalf("waiting for peer handshake timeout after %s\n%s\n%s", timeout.String(),
+				describePeer(peer1ifaceName, peer2Key.PublicKey().String()),
+				describePeer(peer2ifaceName, peer1Key.PublicKey().String()))
 		case <-ticker.C:
 		}
 	}
 
 }
 
+func describePeer(ifaceName, peerPubKey string) string {
+	peer, err := getPeer(ifaceName, peerPubKey)
+	if err != nil {
+		return fmt.Sprintf("%s: peer %s: %v", ifaceName, peerPubKey, err)
+	}
+	return fmt.Sprintf("%s: peer %s endpoint=%v tx=%d rx=%d last_handshake=%v",
+		ifaceName, peerPubKey, peer.Endpoint, peer.TransmitBytes, peer.ReceiveBytes, peer.LastHandshakeTime)
+}
+
 func getPeer(ifaceName, peerPubKey string) (wgtypes.Peer, error) {
 	wg, err := wgctrl.New()
 	if err != nil {

client/iface/udpmux/mux.go

@@ -200,10 +200,7 @@ func (m *SingleSocketUDPMux) updateLocalAddresses() {
 		}
 		if len(networks) > 0 {
 			if m.params.Net == nil {
-				var err error
-				if m.params.Net, err = stdnet.NewNet(context.Background(), nil); err != nil {
-					m.params.Logger.Errorf("failed to get create network: %v", err)
-				}
+				m.params.Net = stdnet.NewNet(context.Background(), nil, nil)
 			}
 
 			ips, err := localInterfaces(m.params.Net, m.params.InterfaceFilter, nil, networks, true)

client/iface/wgproxy/ebpf/portlookup.go

@@ -1,32 +0,0 @@
-package ebpf
-
-import (
-	"fmt"
-	"net"
-)
-
-var (
-	portRangeStart = 3128
-	portRangeEnd   = portRangeStart + 100
-)
-
-type portLookup struct {
-}
-
-func (pl portLookup) searchFreePort() (int, error) {
-	for i := portRangeStart; i <= portRangeEnd; i++ {
-		if pl.tryToBind(i) == nil {
-			return i, nil
-		}
-	}
-	return 0, fmt.Errorf("failed to bind free port for eBPF proxy")
-}
-
-func (pl portLookup) tryToBind(port int) error {
-	l, err := net.ListenPacket("udp", fmt.Sprintf(":%d", port))
-	if err != nil {
-		return err
-	}
-	_ = l.Close()
-	return nil
-}

client/iface/wgproxy/ebpf/portlookup_test.go

@@ -1,45 +0,0 @@
-package ebpf
-
-import (
-	"fmt"
-	"net"
-	"testing"
-)
-
-func Test_portLookup_searchFreePort(t *testing.T) {
-	pl := portLookup{}
-	_, err := pl.searchFreePort()
-	if err != nil {
-		t.Fatal(err)
-	}
-}
-
-func Test_portLookup_on_allocated(t *testing.T) {
-	pl := portLookup{}
-
-	portRangeStart = 4128
-	portRangeEnd = portRangeStart + 100
-
-	allocatedPort, err := allocatePort(portRangeStart)
-	if err != nil {
-		t.Fatal(err)
-	}
-	defer allocatedPort.Close()
-
-	fp, err := pl.searchFreePort()
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	if fp != (portRangeStart + 1) {
-		t.Errorf("invalid free port, expected: %d, got: %d", portRangeStart+1, fp)
-	}
-}
-
-func allocatePort(port int) (net.PacketConn, error) {
-	c, err := net.ListenPacket("udp", fmt.Sprintf(":%d", port))
-	if err != nil {
-		return nil, err
-	}
-	return c, err
-}

client/iface/wgproxy/ebpf/proxy.go

@@ -1,243 +0,0 @@
-//go:build linux && !android
-
-package ebpf
-
-import (
-	"context"
-	"fmt"
-	"net"
-	"sync"
-
-	"github.com/hashicorp/go-multierror"
-	"github.com/pion/transport/v3"
-	log "github.com/sirupsen/logrus"
-
-	nberrors "github.com/netbirdio/netbird/client/errors"
-	"github.com/netbirdio/netbird/client/iface/bufsize"
-	"github.com/netbirdio/netbird/client/iface/wgproxy/rawsocket"
-	"github.com/netbirdio/netbird/client/internal/ebpf"
-	ebpfMgr "github.com/netbirdio/netbird/client/internal/ebpf/manager"
-	nbnet "github.com/netbirdio/netbird/client/net"
-)
-
-const (
-	loopbackAddr = "127.0.0.1"
-)
-
-// WGEBPFProxy definition for proxy with EBPF support
-type WGEBPFProxy struct {
-	localWGListenPort int
-	proxyPort         int
-	mtu               uint16
-
-	ebpfManager      ebpfMgr.Manager
-	relayedConnStore map[uint16]net.Conn
-	relayedConnMutex sync.Mutex
-
-	lastUsedPort uint16
-	rawConnIPv4  net.PacketConn
-	rawConnIPv6  net.PacketConn
-	conn         transport.UDPConn
-
-	ctx       context.Context
-	ctxCancel context.CancelFunc
-}
-
-// NewWGEBPFProxy create new WGEBPFProxy instance
-func NewWGEBPFProxy(wgPort int, mtu uint16) *WGEBPFProxy {
-	log.Debugf("instantiate ebpf proxy")
-	wgProxy := &WGEBPFProxy{
-		localWGListenPort: wgPort,
-		mtu:               mtu,
-		ebpfManager:       ebpf.GetEbpfManagerInstance(),
-		relayedConnStore:  make(map[uint16]net.Conn),
-	}
-	return wgProxy
-}
-
-// Listen load ebpf program and listen the proxy
-func (p *WGEBPFProxy) Listen() error {
-	pl := portLookup{}
-	proxyPort, err := pl.searchFreePort()
-	if err != nil {
-		return err
-	}
-	p.proxyPort = proxyPort
-
-	// Prepare IPv4 raw socket (required)
-	p.rawConnIPv4, err = rawsocket.PrepareSenderRawSocketIPv4()
-	if err != nil {
-		return err
-	}
-
-	// Prepare IPv6 raw socket (optional)
-	p.rawConnIPv6, err = rawsocket.PrepareSenderRawSocketIPv6()
-	if err != nil {
-		log.Warnf("failed to prepare IPv6 raw socket, continuing with IPv4 only: %v", err)
-	}
-
-	err = p.ebpfManager.LoadWgProxy(proxyPort, p.localWGListenPort)
-	if err != nil {
-		if closeErr := p.rawConnIPv4.Close(); closeErr != nil {
-			log.Warnf("failed to close IPv4 raw socket: %v", closeErr)
-		}
-		if p.rawConnIPv6 != nil {
-			if closeErr := p.rawConnIPv6.Close(); closeErr != nil {
-				log.Warnf("failed to close IPv6 raw socket: %v", closeErr)
-			}
-		}
-		return err
-	}
-
-	addr := net.UDPAddr{
-		Port: proxyPort,
-		IP:   net.ParseIP(loopbackAddr),
-	}
-
-	p.ctx, p.ctxCancel = context.WithCancel(context.Background())
-
-	conn, err := nbnet.ListenUDP("udp", &addr)
-	if err != nil {
-		if cErr := p.Free(); cErr != nil {
-			log.Errorf("Failed to close the wgproxy: %s", cErr)
-		}
-		return err
-	}
-	p.conn = conn
-
-	go p.proxyToRemote()
-	log.Infof("local wg proxy listening on: %d", proxyPort)
-	return nil
-}
-
-// AddRelayedConn add new relayed connection for the proxy
-func (p *WGEBPFProxy) AddRelayedConn(relayedConn net.Conn) (*net.UDPAddr, error) {
-	wgEndpointPort, err := p.storeRelayedConn(relayedConn)
-	if err != nil {
-		return nil, err
-	}
-
-	log.Infof("relayed conn added to wg proxy store: %s, endpoint port: :%d", relayedConn.RemoteAddr(), wgEndpointPort)
-
-	wgEndpoint := &net.UDPAddr{
-		IP:   net.ParseIP(loopbackAddr),
-		Port: int(wgEndpointPort),
-	}
-	return wgEndpoint, nil
-}
-
-// Free resources except the remoteConns will be keep open.
-func (p *WGEBPFProxy) Free() error {
-	log.Debugf("free up ebpf wg proxy")
-	if p.ctx != nil && p.ctx.Err() != nil {
-		//nolint
-		return nil
-	}
-
-	p.ctxCancel()
-
-	var result *multierror.Error
-	if p.conn != nil {
-		if err := p.conn.Close(); err != nil {
-			result = multierror.Append(result, err)
-		}
-	}
-
-	if err := p.ebpfManager.FreeWGProxy(); err != nil {
-		result = multierror.Append(result, err)
-	}
-
-	if p.rawConnIPv4 != nil {
-		if err := p.rawConnIPv4.Close(); err != nil {
-			result = multierror.Append(result, err)
-		}
-	}
-
-	if p.rawConnIPv6 != nil {
-		if err := p.rawConnIPv6.Close(); err != nil {
-			result = multierror.Append(result, err)
-		}
-	}
-	return nberrors.FormatErrorOrNil(result)
-}
-
-// GetProxyPort returns the proxy listening port.
-func (p *WGEBPFProxy) GetProxyPort() uint16 {
-	return uint16(p.proxyPort)
-}
-
-// proxyToRemote read messages from local WireGuard interface and forward it to remote conn
-// From this go routine has only one instance.
-func (p *WGEBPFProxy) proxyToRemote() {
-	buf := make([]byte, p.mtu+bufsize.WGBufferOverhead)
-	for p.ctx.Err() == nil {
-		if err := p.readAndForwardPacket(buf); err != nil {
-			if p.ctx.Err() != nil {
-				return
-			}
-			log.Errorf("failed to proxy packet to remote conn: %s", err)
-		}
-	}
-}
-
-func (p *WGEBPFProxy) readAndForwardPacket(buf []byte) error {
-	n, addr, err := p.conn.ReadFromUDP(buf)
-	if err != nil {
-		return fmt.Errorf("failed to read UDP packet from WG: %w", err)
-	}
-
-	p.relayedConnMutex.Lock()
-	conn, ok := p.relayedConnStore[uint16(addr.Port)]
-	p.relayedConnMutex.Unlock()
-	if !ok {
-		if p.ctx.Err() == nil {
-			log.Debugf("relayed conn not found by port because conn already has been closed: %d", addr.Port)
-		}
-		return nil
-	}
-
-	if _, err := conn.Write(buf[:n]); err != nil {
-		return fmt.Errorf("forward local WG packet (%d) to remote relayed conn: %w", addr.Port, err)
-	}
-	return nil
-}
-
-func (p *WGEBPFProxy) storeRelayedConn(relayedConn net.Conn) (uint16, error) {
-	p.relayedConnMutex.Lock()
-	defer p.relayedConnMutex.Unlock()
-
-	np, err := p.nextFreePort()
-	if err != nil {
-		return np, err
-	}
-	p.relayedConnStore[np] = relayedConn
-	return np, nil
-}
-
-func (p *WGEBPFProxy) removeRelayedConn(relayedConnID uint16) {
-	p.relayedConnMutex.Lock()
-	defer p.relayedConnMutex.Unlock()
-
-	_, ok := p.relayedConnStore[relayedConnID]
-	if ok {
-		log.Debugf("remove relayed conn from store by port: %d", relayedConnID)
-	}
-	delete(p.relayedConnStore, relayedConnID)
-}
-
-func (p *WGEBPFProxy) nextFreePort() (uint16, error) {
-	if len(p.relayedConnStore) == 65535 {
-		return 0, fmt.Errorf("reached maximum relayed connection numbers")
-	}
-generatePort:
-	if p.lastUsedPort == 65535 {
-		p.lastUsedPort = 1
-	} else {
-		p.lastUsedPort++
-	}
-
-	if _, ok := p.relayedConnStore[p.lastUsedPort]; ok {
-		goto generatePort
-	}
-	return p.lastUsedPort, nil
-}

client/iface/wgproxy/ebpf/proxy_test.go

@@ -1,56 +0,0 @@
-//go:build linux && !android
-
-package ebpf
-
-import (
-	"testing"
-)
-
-func TestWGEBPFProxy_connStore(t *testing.T) {
-	wgProxy := NewWGEBPFProxy(1, 1280)
-
-	p, _ := wgProxy.storeRelayedConn(nil)
-	if p != 1 {
-		t.Errorf("invalid initial port: %d", wgProxy.lastUsedPort)
-	}
-
-	numOfConns := 10
-	for i := 0; i < numOfConns; i++ {
-		p, _ = wgProxy.storeRelayedConn(nil)
-	}
-	if p != uint16(numOfConns)+1 {
-		t.Errorf("invalid last used port: %d, expected: %d", p, numOfConns+1)
-	}
-	if len(wgProxy.relayedConnStore) != numOfConns+1 {
-		t.Errorf("invalid store size: %d, expected: %d", len(wgProxy.relayedConnStore), numOfConns+1)
-	}
-}
-
-func TestWGEBPFProxy_portCalculation_overflow(t *testing.T) {
-	wgProxy := NewWGEBPFProxy(1, 1280)
-
-	_, _ = wgProxy.storeRelayedConn(nil)
-	wgProxy.lastUsedPort = 65535
-	p, _ := wgProxy.storeRelayedConn(nil)
-
-	if len(wgProxy.relayedConnStore) != 2 {
-		t.Errorf("invalid store size: %d, expected: %d", len(wgProxy.relayedConnStore), 2)
-	}
-
-	if p != 2 {
-		t.Errorf("invalid last used port: %d, expected: %d", p, 2)
-	}
-}
-
-func TestWGEBPFProxy_portCalculation_maxConn(t *testing.T) {
-	wgProxy := NewWGEBPFProxy(1, 1280)
-
-	for i := 0; i < 65535; i++ {
-		_, _ = wgProxy.storeRelayedConn(nil)
-	}
-
-	_, err := wgProxy.storeRelayedConn(nil)
-	if err == nil {
-		t.Errorf("invalid relayed conn store calculation")
-	}
-}

client/iface/wgproxy/factory_kernel.go

@@ -8,11 +8,13 @@ import (
 
 	log "github.com/sirupsen/logrus"
 
-	"github.com/netbirdio/netbird/client/iface/wgproxy/ebpf"
+	"github.com/netbirdio/netbird/client/iface/wgproxy/loopback"
 	udpProxy "github.com/netbirdio/netbird/client/iface/wgproxy/udp"
 )
 
 const (
+	envDisableKernelWGProxy = "NB_DISABLE_KERNEL_WG_PROXY"
+	// envDisableEBPFWGProxy is a deprecated alias for envDisableKernelWGProxy.
 	envDisableEBPFWGProxy = "NB_DISABLE_EBPF_WG_PROXY"
 )
 
@@ -20,7 +22,7 @@ type KernelFactory struct {
 	wgPort int
 	mtu    uint16
 
-	ebpfProxy *ebpf.WGEBPFProxy
+	loopbackProxy *loopback.Proxy
 }
 
 func NewKernelFactory(wgPort int, mtu uint16) *KernelFactory {
@@ -29,55 +31,56 @@ func NewKernelFactory(wgPort int, mtu uint16) *KernelFactory {
 		mtu:    mtu,
 	}
 
-	if isEBPFDisabled() {
+	if isKernelProxyDisabled() {
 		log.Infof("WireGuard Proxy Factory will produce UDP proxy")
-		log.Infof("eBPF WireGuard proxy is disabled via %s environment variable", envDisableEBPFWGProxy)
 		return f
 	}
 
-	ebpfProxy := ebpf.NewWGEBPFProxy(wgPort, mtu)
-	if err := ebpfProxy.Listen(); err != nil {
+	loopbackProxy := loopback.NewProxy(wgPort, mtu)
+	if err := loopbackProxy.Listen(); err != nil {
 		log.Infof("WireGuard Proxy Factory will produce UDP proxy")
-		log.Warnf("failed to initialize ebpf proxy, fallback to user space proxy: %s", err)
+		log.Warnf("failed to initialize loopback proxy, fallback to user space proxy: %s", err)
 		return f
 	}
-	log.Infof("WireGuard Proxy Factory will produce eBPF proxy")
-	f.ebpfProxy = ebpfProxy
+	log.Infof("WireGuard Proxy Factory will produce loopback proxy")
+	f.loopbackProxy = loopbackProxy
 	return f
 }
 
 func (w *KernelFactory) GetProxy() Proxy {
-	if w.ebpfProxy == nil {
+	if w.loopbackProxy == nil {
 		return udpProxy.NewWGUDPProxy(w.wgPort, w.mtu)
 	}
 
-	return ebpf.NewProxyWrapper(w.ebpfProxy)
-}
-
-// GetProxyPort returns the eBPF proxy port, or 0 if eBPF is not active.
-func (w *KernelFactory) GetProxyPort() uint16 {
-	if w.ebpfProxy == nil {
-		return 0
-	}
-	return w.ebpfProxy.GetProxyPort()
+	return loopback.NewProxyWrapper(w.loopbackProxy)
 }
 
 func (w *KernelFactory) Free() error {
-	if w.ebpfProxy == nil {
+	if w.loopbackProxy == nil {
 		return nil
 	}
-	return w.ebpfProxy.Free()
+	return w.loopbackProxy.Free()
 }
 
-func isEBPFDisabled() bool {
-	val := os.Getenv(envDisableEBPFWGProxy)
+func isKernelProxyDisabled() bool {
+	env := envDisableKernelWGProxy
+	val := os.Getenv(env)
+	if val == "" {
+		env = envDisableEBPFWGProxy
+		val = os.Getenv(env)
+	}
 	if val == "" {
 		return false
 	}
+
 	disabled, err := strconv.ParseBool(val)
 	if err != nil {
-		log.Warnf("failed to parse %s: %v", envDisableEBPFWGProxy, err)
+		log.Warnf("failed to parse %s: %v", env, err)
 		return false
 	}
+
+	if disabled {
+		log.Infof("kernel WireGuard proxy is disabled via %s", env)
+	}
 	return disabled
 }

client/iface/wgproxy/factory_usp.go

@@ -24,11 +24,6 @@ func (w *USPFactory) GetProxy() Proxy {
 	return proxyBind.NewProxyBind(w.bind, w.mtu)
 }
 
-// GetProxyPort returns 0 as userspace WireGuard doesn't use a separate proxy port.
-func (w *USPFactory) GetProxyPort() uint16 {
-	return 0
-}
-
 func (w *USPFactory) Free() error {
 	return nil
 }

client/iface/wgproxy/loopback/addr.go

@@ -0,0 +1,70 @@
+//go:build linux && !android
+
+package loopback
+
+import (
+	"fmt"
+	"net/netip"
+)
+
+// Peer endpoints live in the upper half of 127.0.0.0/8. Everything in that
+// range is delivered to the loopback device without any address or route being
+// configured, and staying out of 127.0.0.0/9 keeps well-known squatters such as
+// 127.0.0.53 (systemd-resolved) and 127.0.1.1 out of the way.
+const (
+	addrRangeBase   uint32 = 0x7f800000 // 127.128.0.0
+	addrRangeSize   uint32 = 1 << 23    // /9
+	addrRangePrefix        = "127.128.0.0/9"
+)
+
+// allocator hands out one loopback address per relayed connection. The address
+// is the peer's identity: WireGuard sends to it, and the proxy recovers which
+// peer a packet belongs to from the destination address.
+type allocator struct {
+	cursor uint32
+}
+
+// next returns the first free address at or after the cursor, wrapping once.
+// inUse reports whether an address is already handed out.
+func (a *allocator) next(inUse func(netip.Addr) bool) (netip.Addr, error) {
+	for i := uint32(0); i < addrRangeSize; i++ {
+		a.cursor = (a.cursor + 1) % addrRangeSize
+		addr := addrFromOffset(a.cursor)
+		if !addr.IsValid() {
+			continue
+		}
+		if inUse(addr) {
+			continue
+		}
+		return addr, nil
+	}
+	return netip.Addr{}, fmt.Errorf("no free endpoint address in %s", addrRangePrefix)
+}
+
+// addrFromOffset maps an offset in the range to an address, skipping the .0 and
+// .255 hosts. They are unremarkable on loopback, but tools and firewall rules
+// tend to treat them as network and broadcast addresses.
+func addrFromOffset(offset uint32) netip.Addr {
+	last := offset & 0xff
+	if last == 0 || last == 0xff {
+		return netip.Addr{}
+	}
+
+	v := addrRangeBase + offset
+	return netip.AddrFrom4([4]byte{
+		byte(v >> 24),
+		byte(v >> 16),
+		byte(v >> 8),
+		byte(v),
+	})
+}
+
+// inRange reports whether addr is one this proxy could have handed out.
+func inRange(addr netip.Addr) bool {
+	if !addr.Is4() {
+		return false
+	}
+	b := addr.As4()
+	v := uint32(b[0])<<24 | uint32(b[1])<<16 | uint32(b[2])<<8 | uint32(b[3])
+	return v >= addrRangeBase && v < addrRangeBase+addrRangeSize && b[3] != 0 && b[3] != 0xff
+}

client/iface/wgproxy/loopback/addr_test.go

@@ -0,0 +1,114 @@
+//go:build linux && !android
+
+package loopback
+
+import (
+	"net/netip"
+	"testing"
+)
+
+func TestAllocatorHandsOutDistinctAddresses(t *testing.T) {
+	var a allocator
+	taken := make(map[netip.Addr]bool)
+
+	for i := 0; i < 1000; i++ {
+		addr, err := a.next(func(candidate netip.Addr) bool { return taken[candidate] })
+		if err != nil {
+			t.Fatalf("allocate %d: %v", i, err)
+		}
+		if taken[addr] {
+			t.Fatalf("address %s handed out twice", addr)
+		}
+		if !inRange(addr) {
+			t.Fatalf("address %s outside %s", addr, addrRangePrefix)
+		}
+		taken[addr] = true
+	}
+}
+
+func TestAllocatorSkipsNetworkAndBroadcastHosts(t *testing.T) {
+	var a allocator
+	taken := make(map[netip.Addr]bool)
+
+	// enough allocations to walk past a .255/.0 boundary
+	for i := 0; i < 600; i++ {
+		addr, err := a.next(func(candidate netip.Addr) bool { return taken[candidate] })
+		if err != nil {
+			t.Fatalf("allocate %d: %v", i, err)
+		}
+		last := addr.As4()[3]
+		if last == 0 || last == 255 {
+			t.Fatalf("address %s ends in .%d", addr, last)
+		}
+		taken[addr] = true
+	}
+}
+
+func TestAllocatorReusesReleasedAddresses(t *testing.T) {
+	var a allocator
+	taken := make(map[netip.Addr]bool)
+	inUse := func(candidate netip.Addr) bool { return taken[candidate] }
+	alloc := func() netip.Addr {
+		t.Helper()
+		addr, err := a.next(inUse)
+		if err != nil {
+			t.Fatalf("allocate: %v", err)
+		}
+		taken[addr] = true
+		return addr
+	}
+
+	first := alloc()
+	second := alloc()
+	delete(taken, first)
+
+	// The cursor only moves forward, so a released address comes back after a
+	// wrap. Park the cursor near the end of the range instead of allocating
+	// 2^23 addresses: the next call takes the last usable address, and the one
+	// after that wraps past the skipped .255 and .0 hosts to the released one.
+	a.cursor = addrRangeSize - 3
+	last := alloc()
+	if want := netip.MustParseAddr("127.255.255.254"); last != want {
+		t.Fatalf("expected the last usable address %s before the wrap, got %s", want, last)
+	}
+
+	if reused := alloc(); reused != first {
+		t.Fatalf("expected the released address %s after the wrap, got %s", first, reused)
+	}
+
+	// second is still held, so the allocator must step over it.
+	if next := alloc(); next == second {
+		t.Fatalf("allocator handed out %s while it was still in use", second)
+	}
+}
+
+func TestInRange(t *testing.T) {
+	tests := []struct {
+		addr string
+		want bool
+	}{
+		{"127.128.0.1", true},
+		{"127.255.255.254", true},
+		{"127.128.0.0", false},     // network host, never handed out
+		{"127.128.5.255", false},   // broadcast host, never handed out
+		{"127.127.255.255", false}, // below the range, where 127.0.0.53 and friends live
+		{"127.0.0.1", false},
+		{"127.0.0.53", false},
+		{"127.0.1.1", false},
+		{"128.0.0.1", false},
+		{"10.0.0.1", false},
+	}
+
+	for _, tc := range tests {
+		addr := netip.MustParseAddr(tc.addr)
+		if got := inRange(addr); got != tc.want {
+			t.Errorf("inRange(%s) = %v, want %v", tc.addr, got, tc.want)
+		}
+	}
+}
+
+func TestInRangeIgnoresIPv6(t *testing.T) {
+	if inRange(netip.MustParseAddr("::1")) {
+		t.Error("inRange(::1) = true, want false")
+	}
+}

client/iface/wgproxy/loopback/proxy.go

@@ -0,0 +1,291 @@
+//go:build linux && !android
+
+package loopback
+
+import (
+	"context"
+	"fmt"
+	"net"
+	"net/netip"
+	"sync"
+	"syscall"
+
+	"github.com/hashicorp/go-multierror"
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/net/ipv4"
+	"golang.org/x/sys/unix"
+
+	nberrors "github.com/netbirdio/netbird/client/errors"
+	"github.com/netbirdio/netbird/client/iface/bufsize"
+	"github.com/netbirdio/netbird/client/iface/wgproxy/rawsocket"
+)
+
+const (
+	loopbackDevice = "lo"
+
+	portRangeStart = 3128
+	portRangeEnd   = portRangeStart + 100
+)
+
+// Proxy forwards packets between relayed connections and a local kernel
+// WireGuard instance. Every relayed peer gets its own loopback address as its
+// WireGuard endpoint, so a single socket serves all of them: the destination
+// address of an incoming packet identifies the peer.
+type Proxy struct {
+	localWGListenPort int
+	mtu               uint16
+	proxyPort         int
+
+	conn        *net.UDPConn
+	packetConn  *ipv4.PacketConn
+	loIndex     int
+	rawConnIPv4 net.PacketConn
+	rawConnIPv6 net.PacketConn
+
+	relayedConnMutex sync.Mutex
+	relayedConnStore map[netip.Addr]net.Conn
+	addrs            allocator
+
+	ctx       context.Context
+	ctxCancel context.CancelFunc
+}
+
+// NewProxy creates a proxy for the WireGuard instance listening on wgPort.
+func NewProxy(wgPort int, mtu uint16) *Proxy {
+	log.Debugf("instantiate loopback wg proxy")
+	return &Proxy{
+		localWGListenPort: wgPort,
+		mtu:               mtu,
+		relayedConnStore:  make(map[netip.Addr]net.Conn),
+	}
+}
+
+// Listen opens the shared socket and starts forwarding WireGuard packets to the
+// relayed connections.
+func (p *Proxy) Listen() error {
+	rawConnIPv4, err := rawsocket.PrepareSenderRawSocketIPv4()
+	if err != nil {
+		return fmt.Errorf("prepare IPv4 raw socket: %w", err)
+	}
+	p.rawConnIPv4 = rawConnIPv4
+
+	p.rawConnIPv6, err = rawsocket.PrepareSenderRawSocketIPv6()
+	if err != nil {
+		log.Warnf("failed to prepare IPv6 raw socket, continuing with IPv4 only: %v", err)
+	}
+
+	loopback, err := net.InterfaceByName(loopbackDevice)
+	if err != nil {
+		if freeErr := p.Free(); freeErr != nil {
+			log.Errorf("failed to free the wgproxy: %s", freeErr)
+		}
+		return fmt.Errorf("look up %s: %w", loopbackDevice, err)
+	}
+	p.loIndex = loopback.Index
+
+	if err := p.listen(); err != nil {
+		if freeErr := p.Free(); freeErr != nil {
+			log.Errorf("failed to free the wgproxy: %s", freeErr)
+		}
+		return err
+	}
+
+	p.ctx, p.ctxCancel = context.WithCancel(context.Background())
+
+	go p.proxyToRemote()
+	log.Infof("local wg proxy listening on %s:%d", addrRangePrefix, p.proxyPort)
+	return nil
+}
+
+// listen binds the shared socket on the first free port of the range. The bind
+// has to be a wildcard one to receive every peer address in the range, so it is
+// restricted to the loopback device: without that the port would be reachable
+// on every interface.
+func (p *Proxy) listen() error {
+	var lastErr error
+	for port := portRangeStart; port <= portRangeEnd; port++ {
+		err := p.listenOn(port)
+		if err == nil {
+			p.proxyPort = port
+			return nil
+		}
+		lastErr = err
+	}
+	return fmt.Errorf("bind proxy port in range %d-%d: %w", portRangeStart, portRangeEnd, lastErr)
+}
+
+func (p *Proxy) listenOn(proxyPort int) error {
+	lc := net.ListenConfig{
+		Control: func(_, _ string, c syscall.RawConn) error {
+			var sockErr error
+			if err := c.Control(func(fd uintptr) {
+				if err := unix.SetsockoptString(int(fd), unix.SOL_SOCKET, unix.SO_BINDTODEVICE, loopbackDevice); err != nil {
+					sockErr = fmt.Errorf("bind to %s: %w", loopbackDevice, err)
+					return
+				}
+			}); err != nil {
+				return fmt.Errorf("control socket: %w", err)
+			}
+			return sockErr
+		},
+	}
+
+	conn, err := lc.ListenPacket(context.Background(), "udp4", fmt.Sprintf(":%d", proxyPort))
+	if err != nil {
+		return fmt.Errorf("listen on :%d: %w", proxyPort, err)
+	}
+
+	udpConn, ok := conn.(*net.UDPConn)
+	if !ok {
+		if closeErr := conn.Close(); closeErr != nil {
+			log.Errorf("failed to close proxy conn: %s", closeErr)
+		}
+		return fmt.Errorf("unexpected conn type %T", conn)
+	}
+
+	packetConn := ipv4.NewPacketConn(udpConn)
+	// the destination address carries the peer identity, the interface index is
+	// checked on receive as a second line of defense behind SO_BINDTODEVICE
+	if err := packetConn.SetControlMessage(ipv4.FlagDst|ipv4.FlagInterface, true); err != nil {
+		if closeErr := udpConn.Close(); closeErr != nil {
+			log.Errorf("failed to close proxy conn: %s", closeErr)
+		}
+		return fmt.Errorf("request destination address: %w", err)
+	}
+
+	p.conn = udpConn
+	p.packetConn = packetConn
+	return nil
+}
+
+// AddRelayedConn assigns an endpoint address to the relayed connection and
+// returns the address WireGuard should send to, along with the key the
+// connection is stored under.
+func (p *Proxy) AddRelayedConn(relayedConn net.Conn) (*net.UDPAddr, netip.Addr, error) {
+	addr, err := p.storeRelayedConn(relayedConn)
+	if err != nil {
+		return nil, netip.Addr{}, err
+	}
+
+	log.Infof("relayed conn added to wg proxy store: %s, endpoint address: %s", relayedConn.RemoteAddr(), addr)
+
+	return &net.UDPAddr{
+		IP:   addr.AsSlice(),
+		Port: p.proxyPort,
+	}, addr, nil
+}
+
+// Free releases the proxy resources. The relayed connections are left open.
+func (p *Proxy) Free() error {
+	log.Debugf("free up loopback wg proxy")
+	if p.ctx != nil && p.ctx.Err() != nil {
+		//nolint
+		return nil
+	}
+
+	if p.ctxCancel != nil {
+		p.ctxCancel()
+	}
+
+	var result *multierror.Error
+	if p.conn != nil {
+		if err := p.conn.Close(); err != nil {
+			result = multierror.Append(result, err)
+		}
+	}
+
+	if p.rawConnIPv4 != nil {
+		if err := p.rawConnIPv4.Close(); err != nil {
+			result = multierror.Append(result, err)
+		}
+	}
+
+	if p.rawConnIPv6 != nil {
+		if err := p.rawConnIPv6.Close(); err != nil {
+			result = multierror.Append(result, err)
+		}
+	}
+	return nberrors.FormatErrorOrNil(result)
+}
+
+// proxyToRemote reads packets from the local WireGuard instance and forwards
+// them to the relayed connection the destination address belongs to.
+func (p *Proxy) proxyToRemote() {
+	buf := make([]byte, p.mtu+bufsize.WGBufferOverhead)
+	for p.ctx.Err() == nil {
+		if err := p.readAndForwardPacket(buf); err != nil {
+			if p.ctx.Err() != nil {
+				return
+			}
+			log.Errorf("failed to proxy packet to remote conn: %s", err)
+		}
+	}
+}
+
+func (p *Proxy) readAndForwardPacket(buf []byte) error {
+	n, cm, _, err := p.packetConn.ReadFrom(buf)
+	if err != nil {
+		return fmt.Errorf("read UDP packet from WG: %w", err)
+	}
+
+	if cm == nil {
+		return fmt.Errorf("no control message on packet")
+	}
+
+	if cm.IfIndex != p.loIndex {
+		log.Tracef("dropping packet received on interface %d instead of %s", cm.IfIndex, loopbackDevice)
+		return nil
+	}
+
+	dst, ok := netip.AddrFromSlice(cm.Dst.To4())
+	if !ok || !inRange(dst) {
+		log.Tracef("dropping packet for unexpected destination %s", cm.Dst)
+		return nil
+	}
+
+	p.relayedConnMutex.Lock()
+	conn, ok := p.relayedConnStore[dst]
+	p.relayedConnMutex.Unlock()
+	if !ok {
+		if p.ctx.Err() == nil {
+			log.Debugf("relayed conn not found by address because conn already has been closed: %s", dst)
+		}
+		return nil
+	}
+
+	if _, err := conn.Write(buf[:n]); err != nil {
+		return fmt.Errorf("forward local WG packet (%s) to remote relayed conn: %w", dst, err)
+	}
+	return nil
+}
+
+func (p *Proxy) storeRelayedConn(relayedConn net.Conn) (netip.Addr, error) {
+	p.relayedConnMutex.Lock()
+	defer p.relayedConnMutex.Unlock()
+
+	addr, err := p.addrs.next(func(a netip.Addr) bool {
+		_, ok := p.relayedConnStore[a]
+		return ok
+	})
+	if err != nil {
+		return netip.Addr{}, err
+	}
+
+	p.relayedConnStore[addr] = relayedConn
+	return addr, nil
+}
+
+// removeRelayedConn releases an endpoint address. It only removes the entry
+// while it still belongs to relayedConn, so a late release cannot take an
+// address away from the peer it was handed to next.
+func (p *Proxy) removeRelayedConn(addr netip.Addr, relayedConn net.Conn) {
+	p.relayedConnMutex.Lock()
+	defer p.relayedConnMutex.Unlock()
+
+	if stored, ok := p.relayedConnStore[addr]; !ok || stored != relayedConn {
+		return
+	}
+
+	log.Debugf("remove relayed conn from store by address: %s", addr)
+	delete(p.relayedConnStore, addr)
+}

client/iface/wgproxy/loopback/proxy_privileged_test.go

@@ -0,0 +1,196 @@
+//go:build linux && !android && privileged
+
+package loopback
+
+import (
+	"context"
+	"net"
+	"strconv"
+	"testing"
+	"time"
+)
+
+const testWGPort = 51862
+
+// relayEnd stands in for a relayed connection: the proxy writes what it read
+// from WireGuard into it, and the test reads it back out here.
+func relayEnd(t *testing.T) (proxySide net.Conn, testSide *net.UDPConn) {
+	t.Helper()
+
+	testSide, err := net.ListenUDP("udp", &net.UDPAddr{IP: net.ParseIP("127.0.0.1")})
+	if err != nil {
+		t.Fatalf("relay listener: %v", err)
+	}
+	t.Cleanup(func() {
+		if err := testSide.Close(); err != nil {
+			t.Logf("close relay listener: %v", err)
+		}
+	})
+
+	proxySide, err = net.Dial("udp", testSide.LocalAddr().String())
+	if err != nil {
+		t.Fatalf("relay conn: %v", err)
+	}
+	t.Cleanup(func() {
+		if err := proxySide.Close(); err != nil {
+			t.Logf("close relay conn: %v", err)
+		}
+	})
+
+	return proxySide, testSide
+}
+
+// TestProxyDemuxesByDestinationAddress is the core of the design: one socket
+// serves every peer, and the destination address decides which relayed
+// connection a WireGuard packet belongs to.
+func TestProxyDemuxesByDestinationAddress(t *testing.T) {
+	proxy := NewProxy(testWGPort, 1280)
+	if err := proxy.Listen(); err != nil {
+		t.Fatalf("listen: %v", err)
+	}
+	defer func() {
+		if err := proxy.Free(); err != nil {
+			t.Errorf("free proxy: %v", err)
+		}
+	}()
+
+	const peers = 3
+	endpoints := make([]*net.UDPAddr, 0, peers)
+	readers := make([]*net.UDPConn, 0, peers)
+	for i := 0; i < peers; i++ {
+		proxySide, testSide := relayEnd(t)
+		endpoint, _, err := proxy.AddRelayedConn(proxySide)
+		if err != nil {
+			t.Fatalf("add relayed conn %d: %v", i, err)
+		}
+		if endpoint.Port != proxy.proxyPort {
+			t.Errorf("peer %d endpoint port = %d, want the shared proxy port %d", i, endpoint.Port, proxy.proxyPort)
+		}
+		endpoints = append(endpoints, endpoint)
+		readers = append(readers, testSide)
+	}
+
+	// every peer must have its own address, otherwise they are indistinguishable
+	seen := make(map[string]bool, peers)
+	for i, endpoint := range endpoints {
+		if seen[endpoint.IP.String()] {
+			t.Fatalf("peer %d reuses endpoint address %s", i, endpoint.IP)
+		}
+		seen[endpoint.IP.String()] = true
+	}
+
+	wgSock, err := net.ListenUDP("udp", &net.UDPAddr{IP: net.ParseIP("127.0.0.1")})
+	if err != nil {
+		t.Fatalf("wg socket: %v", err)
+	}
+	defer func() {
+		if err := wgSock.Close(); err != nil {
+			t.Logf("close wg socket: %v", err)
+		}
+	}()
+
+	for i, endpoint := range endpoints {
+		payload := []byte{byte(i), 'p', 'k', 't'}
+		if _, err := wgSock.WriteTo(payload, endpoint); err != nil {
+			t.Fatalf("write to peer %d endpoint %s: %v", i, endpoint, err)
+		}
+
+		buf := make([]byte, 1500)
+		if err := readers[i].SetReadDeadline(time.Now().Add(2 * time.Second)); err != nil {
+			t.Fatalf("set read deadline: %v", err)
+		}
+		n, _, err := readers[i].ReadFrom(buf)
+		if err != nil {
+			t.Fatalf("peer %d did not receive its packet: %v", i, err)
+		}
+		if string(buf[:n]) != string(payload) {
+			t.Errorf("peer %d got %q, want %q", i, buf[:n], payload)
+		}
+
+		// no other peer may see it
+		for j, other := range readers {
+			if j == i {
+				continue
+			}
+			if err := other.SetReadDeadline(time.Now().Add(100 * time.Millisecond)); err != nil {
+				t.Fatalf("set read deadline: %v", err)
+			}
+			if _, _, err := other.ReadFrom(buf); err == nil {
+				t.Errorf("packet for peer %d also delivered to peer %d", i, j)
+			}
+		}
+	}
+}
+
+// TestProxyDropsPacketsOutsideTheRange guards the wildcard bind: anything that
+// is not addressed to a handed-out endpoint must not reach a relayed peer.
+func TestProxyDropsPacketsOutsideTheRange(t *testing.T) {
+	proxy := NewProxy(testWGPort+1, 1280)
+	if err := proxy.Listen(); err != nil {
+		t.Fatalf("listen: %v", err)
+	}
+	defer func() {
+		if err := proxy.Free(); err != nil {
+			t.Errorf("free proxy: %v", err)
+		}
+	}()
+
+	proxySide, testSide := relayEnd(t)
+	if _, _, err := proxy.AddRelayedConn(proxySide); err != nil {
+		t.Fatalf("add relayed conn: %v", err)
+	}
+
+	sender, err := net.Dial("udp", net.JoinHostPort("127.0.0.1", strconv.Itoa(proxy.proxyPort)))
+	if err != nil {
+		t.Fatalf("sender: %v", err)
+	}
+	defer func() {
+		if err := sender.Close(); err != nil {
+			t.Logf("close sender: %v", err)
+		}
+	}()
+
+	if _, err := sender.Write([]byte("stray")); err != nil {
+		t.Fatalf("write stray packet: %v", err)
+	}
+
+	buf := make([]byte, 1500)
+	if err := testSide.SetReadDeadline(time.Now().Add(500 * time.Millisecond)); err != nil {
+		t.Fatalf("set read deadline: %v", err)
+	}
+	if _, _, err := testSide.ReadFrom(buf); err == nil {
+		t.Error("packet addressed to 127.0.0.1 was forwarded to a relayed peer")
+	}
+}
+
+// A wrapper that is closed before it starts forwarding still has to give its
+// endpoint address back, otherwise the range leaks an address per attempt.
+func TestClosingBeforeWorkReleasesTheAddress(t *testing.T) {
+	proxy := NewProxy(testWGPort+2, 1280)
+	if err := proxy.Listen(); err != nil {
+		t.Fatalf("listen: %v", err)
+	}
+	defer func() {
+		if err := proxy.Free(); err != nil {
+			t.Errorf("free proxy: %v", err)
+		}
+	}()
+
+	proxySide, _ := relayEnd(t)
+	wrapper := NewProxyWrapper(proxy)
+	if err := wrapper.AddRelayedConn(context.Background(), nil, proxySide); err != nil {
+		t.Fatalf("add relayed conn: %v", err)
+	}
+
+	if got := len(proxy.relayedConnStore); got != 1 {
+		t.Fatalf("store holds %d entries after adding one conn, want 1", got)
+	}
+
+	if err := wrapper.CloseConn(); err != nil {
+		t.Fatalf("close conn: %v", err)
+	}
+
+	if got := len(proxy.relayedConnStore); got != 0 {
+		t.Errorf("store holds %d entries after close, want 0", got)
+	}
+}

client/iface/wgproxy/loopback/wrapper.go

@@ -1,6 +1,6 @@
 //go:build linux && !android
 
-package ebpf
+package loopback
 
 import (
 	"context"
@@ -8,6 +8,7 @@ import (
 	"fmt"
 	"io"
 	"net"
+	"net/netip"
 	"sync"
 
 	"github.com/google/gopacket"
@@ -95,13 +96,14 @@ func NewPacketHeaders(localWGListenPort int, endpoint *net.UDPAddr) (*PacketHead
 
 // ProxyWrapper help to keep the remoteConn instance for net.Conn.Close function call
 type ProxyWrapper struct {
-	wgeBPFProxy *WGEBPFProxy
+	proxy *Proxy
 
 	remoteConn net.Conn
 	ctx        context.Context
 	cancel     context.CancelFunc
 
 	wgRelayedEndpointAddr *net.UDPAddr
+	peerAddr              netip.Addr
 	headers               *PacketHeaders
 	headerCurrentUsed     *PacketHeaders
 	rawConn               net.PacketConn
@@ -113,36 +115,44 @@ type ProxyWrapper struct {
 	closeListener *listener.CloseListener
 }
 
-func NewProxyWrapper(proxy *WGEBPFProxy) *ProxyWrapper {
+func NewProxyWrapper(proxy *Proxy) *ProxyWrapper {
 	return &ProxyWrapper{
-		wgeBPFProxy:   proxy,
+		proxy:         proxy,
 		pausedCond:    sync.NewCond(&sync.Mutex{}),
 		closeListener: listener.NewCloseListener(),
 	}
 }
 
 func (p *ProxyWrapper) AddRelayedConn(ctx context.Context, _ *net.UDPAddr, remoteConn net.Conn) error {
-	addr, err := p.wgeBPFProxy.AddRelayedConn(remoteConn)
+	addr, peerAddr, err := p.proxy.AddRelayedConn(remoteConn)
 	if err != nil {
 		return fmt.Errorf("add relayed conn: %w", err)
 	}
 
-	headers, err := NewPacketHeaders(p.wgeBPFProxy.localWGListenPort, addr)
+	// the endpoint address is otherwise only released by the forwarding
+	// goroutine, which never starts when the setup below fails
+	release := func() { p.proxy.removeRelayedConn(peerAddr, remoteConn) }
+
+	headers, err := NewPacketHeaders(p.proxy.localWGListenPort, addr)
 	if err != nil {
+		release()
 		return fmt.Errorf("create packet sender: %w", err)
 	}
 
 	// Check if required raw connection is available
-	if !headers.isIPv4 && p.wgeBPFProxy.rawConnIPv6 == nil {
+	if !headers.isIPv4 && p.proxy.rawConnIPv6 == nil {
+		release()
 		return errIPv6ConnNotAvailable
 	}
-	if headers.isIPv4 && p.wgeBPFProxy.rawConnIPv4 == nil {
+	if headers.isIPv4 && p.proxy.rawConnIPv4 == nil {
+		release()
 		return errIPv4ConnNotAvailable
 	}
 
 	p.remoteConn = remoteConn
 	p.ctx, p.cancel = context.WithCancel(ctx)
 	p.wgRelayedEndpointAddr = addr
+	p.peerAddr = peerAddr
 	p.headers = headers
 	p.rawConn = p.selectRawConn(headers)
 	return nil
@@ -193,18 +203,18 @@ func (p *ProxyWrapper) RedirectAs(endpoint *net.UDPAddr) {
 		return
 	}
 
-	header, err := NewPacketHeaders(p.wgeBPFProxy.localWGListenPort, endpoint)
+	header, err := NewPacketHeaders(p.proxy.localWGListenPort, endpoint)
 	if err != nil {
 		log.Errorf("failed to create packet headers: %s", err)
 		return
 	}
 
 	// Check if required raw connection is available
-	if !header.isIPv4 && p.wgeBPFProxy.rawConnIPv6 == nil {
+	if !header.isIPv4 && p.proxy.rawConnIPv6 == nil {
 		log.Error(errIPv6ConnNotAvailable)
 		return
 	}
-	if header.isIPv4 && p.wgeBPFProxy.rawConnIPv4 == nil {
+	if header.isIPv4 && p.proxy.rawConnIPv4 == nil {
 		log.Error(errIPv4ConnNotAvailable)
 		return
 	}
@@ -240,6 +250,10 @@ func (p *ProxyWrapper) CloseConn() error {
 
 	p.closeListener.SetCloseListener(nil)
 
+	// releases the endpoint address for a wrapper that was never started, and
+	// is a no-op once the forwarding goroutine has released it
+	p.proxy.removeRelayedConn(p.peerAddr, p.remoteConn)
+
 	p.pausedCond.L.Lock()
 	p.paused = false
 	p.pausedCond.Signal()
@@ -252,9 +266,9 @@ func (p *ProxyWrapper) CloseConn() error {
 }
 
 func (p *ProxyWrapper) proxyToLocal(ctx context.Context) {
-	defer p.wgeBPFProxy.removeRelayedConn(uint16(p.wgRelayedEndpointAddr.Port))
+	defer p.proxy.removeRelayedConn(p.peerAddr, p.remoteConn)
 
-	buf := make([]byte, p.wgeBPFProxy.mtu+bufsize.WGBufferOverhead)
+	buf := make([]byte, p.proxy.mtu+bufsize.WGBufferOverhead)
 	for {
 		n, err := p.readFromRemote(ctx, buf)
 		if err != nil {
@@ -286,7 +300,7 @@ func (p *ProxyWrapper) readFromRemote(ctx context.Context, buf []byte) (int, err
 		}
 		p.closeListener.Notify()
 		if !errors.Is(err, io.EOF) {
-			log.Errorf("failed to read from relayed conn (endpoint: :%d): %s", p.wgRelayedEndpointAddr.Port, err)
+			log.Errorf("failed to read from relayed conn (endpoint: %s): %s", p.wgRelayedEndpointAddr, err)
 		}
 		return 0, err
 	}
@@ -314,7 +328,7 @@ func (p *ProxyWrapper) sendPkg(data []byte, header *PacketHeaders) error {
 
 func (p *ProxyWrapper) selectRawConn(header *PacketHeaders) net.PacketConn {
 	if header.isIPv4 {
-		return p.wgeBPFProxy.rawConnIPv4
+		return p.proxy.rawConnIPv4
 	}
-	return p.wgeBPFProxy.rawConnIPv6
+	return p.proxy.rawConnIPv6
 }

client/iface/wgproxy/proxy_linux_test.go

@@ -9,25 +9,25 @@ import (
 	"github.com/netbirdio/netbird/client/iface/bind"
 	"github.com/netbirdio/netbird/client/iface/wgaddr"
 	bindproxy "github.com/netbirdio/netbird/client/iface/wgproxy/bind"
-	"github.com/netbirdio/netbird/client/iface/wgproxy/ebpf"
+	"github.com/netbirdio/netbird/client/iface/wgproxy/loopback"
 	"github.com/netbirdio/netbird/client/iface/wgproxy/udp"
 )
 
 func seedProxies() ([]proxyInstance, error) {
 	pl := make([]proxyInstance, 0)
 
-	ebpfProxy := ebpf.NewWGEBPFProxy(51831, 1280)
-	if err := ebpfProxy.Listen(); err != nil {
-		return nil, fmt.Errorf("failed to initialize ebpf proxy: %s", err)
+	loopbackProxy := loopback.NewProxy(51831, 1280)
+	if err := loopbackProxy.Listen(); err != nil {
+		return nil, fmt.Errorf("failed to initialize loopback proxy: %s", err)
 	}
 
-	pEbpf := proxyInstance{
-		name:    "ebpf kernel proxy",
-		proxy:   ebpf.NewProxyWrapper(ebpfProxy),
+	pLoopback := proxyInstance{
+		name:    "loopback kernel proxy",
+		proxy:   loopback.NewProxyWrapper(loopbackProxy),
 		wgPort:  51831,
-		closeFn: ebpfProxy.Free,
+		closeFn: loopbackProxy.Free,
 	}
-	pl = append(pl, pEbpf)
+	pl = append(pl, pLoopback)
 
 	pUDP := proxyInstance{
 		name:    "udp kernel proxy",
@@ -42,18 +42,18 @@ func seedProxies() ([]proxyInstance, error) {
 func seedProxyForProxyCloseByRemoteConn() ([]proxyInstance, error) {
 	pl := make([]proxyInstance, 0)
 
-	ebpfProxy := ebpf.NewWGEBPFProxy(51831, 1280)
-	if err := ebpfProxy.Listen(); err != nil {
-		return nil, fmt.Errorf("failed to initialize ebpf proxy: %s", err)
+	loopbackProxy := loopback.NewProxy(51831, 1280)
+	if err := loopbackProxy.Listen(); err != nil {
+		return nil, fmt.Errorf("failed to initialize loopback proxy: %s", err)
 	}
 
-	pEbpf := proxyInstance{
-		name:    "ebpf kernel proxy",
-		proxy:   ebpf.NewProxyWrapper(ebpfProxy),
+	pLoopback := proxyInstance{
+		name:    "loopback kernel proxy",
+		proxy:   loopback.NewProxyWrapper(loopbackProxy),
 		wgPort:  51831,
-		closeFn: ebpfProxy.Free,
+		closeFn: loopbackProxy.Free,
 	}
-	pl = append(pl, pEbpf)
+	pl = append(pl, pLoopback)
 
 	pUDP := proxyInstance{
 		name:    "udp kernel proxy",

client/iface/wgproxy/redirect_test.go

@@ -8,7 +8,7 @@ import (
 	"testing"
 	"time"
 
-	"github.com/netbirdio/netbird/client/iface/wgproxy/ebpf"
+	"github.com/netbirdio/netbird/client/iface/wgproxy/loopback"
 	"github.com/netbirdio/netbird/client/iface/wgproxy/udp"
 )
 
@@ -198,20 +198,20 @@ func testRedirectAs(t *testing.T, proxy Proxy, wgPort int, nbAddr, p2pEndpoint *
 	}
 }
 
-// TestRedirectAs_eBPF_IPv4 tests RedirectAs with eBPF proxy using IPv4 addresses
-func TestRedirectAs_eBPF_IPv4(t *testing.T) {
+// TestRedirectAs_Loopback_IPv4 tests RedirectAs with the loopback proxy using IPv4 addresses
+func TestRedirectAs_Loopback_IPv4(t *testing.T) {
 	wgPort := 51850
-	ebpfProxy := ebpf.NewWGEBPFProxy(wgPort, 1280)
-	if err := ebpfProxy.Listen(); err != nil {
-		t.Fatalf("failed to initialize ebpf proxy: %v", err)
+	loopbackProxy := loopback.NewProxy(wgPort, 1280)
+	if err := loopbackProxy.Listen(); err != nil {
+		t.Fatalf("failed to initialize loopback proxy: %v", err)
 	}
 	defer func() {
-		if err := ebpfProxy.Free(); err != nil {
-			t.Errorf("failed to free ebpf proxy: %v", err)
+		if err := loopbackProxy.Free(); err != nil {
+			t.Errorf("failed to free loopback proxy: %v", err)
 		}
 	}()
 
-	proxy := ebpf.NewProxyWrapper(ebpfProxy)
+	proxy := loopback.NewProxyWrapper(loopbackProxy)
 
 	// NetBird UDP address of the remote peer
 	nbAddr := &net.UDPAddr{
@@ -227,20 +227,20 @@ func TestRedirectAs_eBPF_IPv4(t *testing.T) {
 	testRedirectAs(t, proxy, wgPort, nbAddr, p2pEndpoint)
 }
 
-// TestRedirectAs_eBPF_IPv6 tests RedirectAs with eBPF proxy using IPv6 addresses
-func TestRedirectAs_eBPF_IPv6(t *testing.T) {
+// TestRedirectAs_Loopback_IPv6 tests RedirectAs with the loopback proxy using IPv6 addresses
+func TestRedirectAs_Loopback_IPv6(t *testing.T) {
 	wgPort := 51851
-	ebpfProxy := ebpf.NewWGEBPFProxy(wgPort, 1280)
-	if err := ebpfProxy.Listen(); err != nil {
-		t.Fatalf("failed to initialize ebpf proxy: %v", err)
+	loopbackProxy := loopback.NewProxy(wgPort, 1280)
+	if err := loopbackProxy.Listen(); err != nil {
+		t.Fatalf("failed to initialize loopback proxy: %v", err)
 	}
 	defer func() {
-		if err := ebpfProxy.Free(); err != nil {
-			t.Errorf("failed to free ebpf proxy: %v", err)
+		if err := loopbackProxy.Free(); err != nil {
+			t.Errorf("failed to free loopback proxy: %v", err)
 		}
 	}()
 
-	proxy := ebpf.NewProxyWrapper(ebpfProxy)
+	proxy := loopback.NewProxyWrapper(loopbackProxy)
 
 	// NetBird UDP address of the remote peer
 	nbAddr := &net.UDPAddr{
@@ -259,17 +259,17 @@ func TestRedirectAs_eBPF_IPv6(t *testing.T) {
 // TestRedirectAs_Multiple_Switches tests switching between multiple endpoints
 func TestRedirectAs_Multiple_Switches(t *testing.T) {
 	wgPort := 51856
-	ebpfProxy := ebpf.NewWGEBPFProxy(wgPort, 1280)
-	if err := ebpfProxy.Listen(); err != nil {
-		t.Fatalf("failed to initialize ebpf proxy: %v", err)
+	loopbackProxy := loopback.NewProxy(wgPort, 1280)
+	if err := loopbackProxy.Listen(); err != nil {
+		t.Fatalf("failed to initialize loopback proxy: %v", err)
 	}
 	defer func() {
-		if err := ebpfProxy.Free(); err != nil {
-			t.Errorf("failed to free ebpf proxy: %v", err)
+		if err := loopbackProxy.Free(); err != nil {
+			t.Errorf("failed to free loopback proxy: %v", err)
 		}
 	}()
 
-	proxy := ebpf.NewProxyWrapper(ebpfProxy)
+	proxy := loopback.NewProxyWrapper(loopbackProxy)
 
 	ctx := context.Background()
 

client/internal/auth/account_match_test.go

@@ -0,0 +1,113 @@
+package auth
+
+import (
+	"encoding/base64"
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestTokenInfoMatchesAccount(t *testing.T) {
+	tests := []struct {
+		name  string
+		token TokenInfo
+		hint  string
+		match bool
+	}{
+		{
+			name:  "same account",
+			token: TokenInfo{EmailClaim: "user@example.com"},
+			hint:  "user@example.com",
+			match: true,
+		},
+		{
+			name:  "different account",
+			token: TokenInfo{EmailClaim: "other@example.com"},
+			hint:  "user@example.com",
+			match: false,
+		},
+		{
+			name:  "case differences are the same account",
+			token: TokenInfo{EmailClaim: "User@Example.com"},
+			hint:  "user@example.com",
+			match: true,
+		},
+		{
+			name:  "no hint leaves the choice to the IdP",
+			token: TokenInfo{EmailClaim: "other@example.com"},
+			hint:  "",
+			match: true,
+		},
+		{
+			name:  "token without an email claim is not judged",
+			token: TokenInfo{EmailClaim: ""},
+			hint:  "user@example.com",
+			match: true,
+		},
+		{
+			name:  "name fallback does not trigger matching",
+			token: TokenInfo{Email: "Some One"},
+			hint:  "user@example.com",
+			match: true,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			assert.Equal(t, tc.match, tc.token.MatchesAccount(tc.hint))
+		})
+	}
+}
+
+func TestParseEmailFromIDToken(t *testing.T) {
+	tests := []struct {
+		name          string
+		claims        map[string]interface{}
+		wantValue     string
+		wantFromEmail bool
+		wantErr       bool
+	}{
+		{
+			name:          "email claim",
+			claims:        map[string]interface{}{"email": "user@example.com", "name": "Some One"},
+			wantValue:     "user@example.com",
+			wantFromEmail: true,
+		},
+		{
+			name:      "name fallback",
+			claims:    map[string]interface{}{"name": "Some One"},
+			wantValue: "Some One",
+		},
+		{
+			name:    "neither claim present",
+			claims:  map[string]interface{}{"sub": "abc"},
+			wantErr: true,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			value, fromEmailClaim, err := parseEmailFromIDToken(idTokenWithClaims(t, tc.claims))
+			if tc.wantErr {
+				require.Error(t, err)
+				return
+			}
+			require.NoError(t, err)
+			assert.Equal(t, tc.wantValue, value)
+			assert.Equal(t, tc.wantFromEmail, fromEmailClaim)
+		})
+	}
+}
+
+func TestRetryFlowForAccountUnsupportedFlow(t *testing.T) {
+	assert.Nil(t, RetryFlowForAccount(&DeviceAuthorizationFlow{}))
+}
+
+func idTokenWithClaims(t *testing.T, claims map[string]interface{}) string {
+	t.Helper()
+	payload, err := json.Marshal(claims)
+	require.NoError(t, err)
+	return "header." + base64.RawURLEncoding.EncodeToString(payload) + ".signature"
+}

client/internal/auth/auth.go

@@ -103,7 +103,7 @@ func (a *Auth) IsSSOSupported(ctx context.Context) (bool, error) {
 
 	err := a.withRetry(ctx, func(client *mgm.GrpcClient) error {
 		// Try PKCE flow first
-		_, err := a.getPKCEFlow(client)
+		_, err := a.getPKCEFlow(client, false)
 		if err == nil {
 			supportsSSO = true
 			return nil
@@ -136,9 +136,13 @@ func (a *Auth) IsSSOSupported(ctx context.Context) (bool, error) {
 	return supportsSSO, err
 }
 
-// GetOAuthFlow returns an OAuth flow (PKCE or Device) using the existing management connection
-// This avoids creating a new connection to the management server
-func (a *Auth) GetOAuthFlow(ctx context.Context, forceDeviceAuth bool, hint string) (OAuthFlow, error) {
+// GetOAuthFlow returns an OAuth flow (PKCE or Device) using the existing management connection.
+// This avoids creating a new connection to the management server.
+//
+// sessionExtend marks the flow as renewing an existing peer's session rather than
+// logging one in; the server needs it to rule out a silent authorization that the
+// IdP could answer from another account. See PKCEAuthorizationFlowRequest.
+func (a *Auth) GetOAuthFlow(ctx context.Context, forceDeviceAuth bool, sessionExtend bool, hint string) (OAuthFlow, error) {
 	var flow OAuthFlow
 
 	err := a.withRetry(ctx, func(client *mgm.GrpcClient) error {
@@ -153,7 +157,7 @@ func (a *Auth) GetOAuthFlow(ctx context.Context, forceDeviceAuth bool, hint stri
 		}
 
 		// Try PKCE flow first
-		pkceFlow, err := a.getPKCEFlow(client)
+		pkceFlow, err := a.getPKCEFlow(client, sessionExtend)
 		if err != nil {
 			// If PKCE not supported, try Device flow
 			if s, ok := status.FromError(err); ok && (s.Code() == codes.NotFound || s.Code() == codes.Unimplemented) {
@@ -240,8 +244,8 @@ func (a *Auth) Login(ctx context.Context, setupKey string, jwtToken string) (err
 }
 
 // getPKCEFlow retrieves PKCE authorization flow configuration and creates a flow instance
-func (a *Auth) getPKCEFlow(client *mgm.GrpcClient) (*PKCEAuthorizationFlow, error) {
-	protoFlow, err := client.GetPKCEAuthorizationFlow()
+func (a *Auth) getPKCEFlow(client *mgm.GrpcClient, sessionExtend bool) (*PKCEAuthorizationFlow, error) {
+	protoFlow, err := client.GetPKCEAuthorizationFlow(sessionExtend)
 	if err != nil {
 		if s, ok := status.FromError(err); ok && s.Code() == codes.NotFound {
 			log.Warnf("server couldn't find pkce flow, contact admin: %v", err)

client/internal/auth/device_flow.go

@@ -308,10 +308,13 @@ func (d *DeviceAuthorizationFlow) WaitToken(ctx context.Context, info AuthFlowIn
 			// callers store to send back as the login_hint. Without it a client
 			// driven through the device flow — Android TV and tvOS — never binds
 			// an account to its profile and every later login goes out blind.
-			if email, err := parseEmailFromIDToken(tokenInfo.IDToken); err != nil {
+			if email, fromEmailClaim, err := parseEmailFromIDToken(tokenInfo.IDToken); err != nil {
 				log.Warnf("failed to parse email from ID token: %v", err)
 			} else {
 				tokenInfo.Email = email
+				if fromEmailClaim {
+					tokenInfo.EmailClaim = email
+				}
 			}
 
 			log.Infof("device flow: user authorization confirmed after %d polls in %s", polls, time.Since(start).Round(time.Second))

client/internal/auth/oauth.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"net/http"
 	"runtime"
+	"strings"
 
 	log "github.com/sirupsen/logrus"
 	"google.golang.org/grpc/codes"
@@ -25,6 +26,14 @@ type HTTPClient interface {
 	Do(req *http.Request) (*http.Response, error)
 }
 
+// accountPromptForcer is implemented by the PKCE flow only. The device code
+// flow has no equivalent: RFC 8628 defines no prompt parameter, and the user
+// confirms the code on a page that shows which account signs in, so a silent
+// wrong-account answer is not the failure mode there.
+type accountPromptForcer interface {
+	ForceAccountPrompt()
+}
+
 // AuthFlowInfo holds information for the OAuth 2.0  authorization flow
 type AuthFlowInfo struct { //nolint:revive
 	DeviceCode              string `json:"device_code"`
@@ -49,6 +58,23 @@ type TokenInfo struct {
 	ExpiresIn    int    `json:"expires_in"`
 	UseIDToken   bool   `json:"-"`
 	Email        string `json:"-"`
+	EmailClaim   string `json:"-"`
+}
+
+// MatchesAccount reports whether the token belongs to the account a profile is
+// bound to. A hint the IdP could not have acted on — no hint stored, or a token
+// that carried no email claim — is reported as a match: the check exists to catch a
+// login answered from the wrong account, not to block one it cannot judge.
+//
+// The comparison is case-insensitive. Local-parts are case-sensitive per RFC
+// 5321, but no IdP in practice issues two accounts differing only in case, and
+// an IdP that echoes a differently-cased address would otherwise fail every
+// login.
+func (t TokenInfo) MatchesAccount(hint string) bool {
+	if hint == "" || t.EmailClaim == "" {
+		return true
+	}
+	return strings.EqualFold(t.EmailClaim, hint)
 }
 
 // GetTokenToUse returns either the access or id token based on UseIDToken field
@@ -63,19 +89,22 @@ func shouldUseDeviceFlow(force bool, isUnixDesktopClient bool) bool {
 	return force || (runtime.GOOS == "linux" || runtime.GOOS == "freebsd") && !isUnixDesktopClient
 }
 
-// NewOAuthFlow initializes and returns the appropriate OAuth flow based on the management configuration
+// NewOAuthFlow initializes and returns the appropriate OAuth flow based on the management configuration.
 //
 // It starts by initializing the PKCE.If this process fails, it resorts to the Device Code Flow,
 // and if that also fails, the authentication process is deemed unsuccessful
 //
-// On Linux distros without desktop environment support, it only tries to initialize the Device Code Flow
-// forceDeviceCodeFlow can be used to skip PKCE and go directly to Device Code Flow (e.g., for Android TV)
-func NewOAuthFlow(ctx context.Context, config *profilemanager.Config, isUnixDesktopClient bool, forceDeviceCodeFlow bool, hint string) (OAuthFlow, error) {
+// On Linux distros without desktop environment support, it only tries to initialize the Device Code Flow.
+// forceDeviceCodeFlow can be used to skip PKCE and go directly to Device Code Flow (e.g., for Android TV).
+//
+// sessionExtend marks the flow as renewing an existing peer's session rather than
+// logging one in. See PKCEAuthorizationFlowRequest for what the server makes of it.
+func NewOAuthFlow(ctx context.Context, config *profilemanager.Config, isUnixDesktopClient bool, forceDeviceCodeFlow bool, hint string, sessionExtend bool) (OAuthFlow, error) {
 	if shouldUseDeviceFlow(forceDeviceCodeFlow, isUnixDesktopClient) {
 		return authenticateWithDeviceCodeFlow(ctx, config, hint)
 	}
 
-	pkceFlow, err := authenticateWithPKCEFlow(ctx, config, hint)
+	pkceFlow, err := authenticateWithPKCEFlow(ctx, config, hint, sessionExtend)
 	if err != nil {
 		log.Debugf("failed to initialize pkce authentication with error: %v\n", err)
 		log.Debug("falling back to device code flow")
@@ -85,14 +114,14 @@ func NewOAuthFlow(ctx context.Context, config *profilemanager.Config, isUnixDesk
 }
 
 // authenticateWithPKCEFlow initializes the Proof Key for Code Exchange flow auth flow
-func authenticateWithPKCEFlow(ctx context.Context, config *profilemanager.Config, hint string) (OAuthFlow, error) {
+func authenticateWithPKCEFlow(ctx context.Context, config *profilemanager.Config, hint string, sessionExtend bool) (OAuthFlow, error) {
 	authClient, err := NewAuth(ctx, config.PrivateKey, config.ManagementURL, config)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create auth client: %v", err)
 	}
 	defer authClient.Close()
 
-	pkceFlowInfo, err := authClient.getPKCEFlow(authClient.client)
+	pkceFlowInfo, err := authClient.getPKCEFlow(authClient.client, sessionExtend)
 	if err != nil {
 		return nil, fmt.Errorf("getting pkce authorization flow info failed with error: %v", err)
 	}
@@ -129,3 +158,21 @@ func authenticateWithDeviceCodeFlow(ctx context.Context, config *profilemanager.
 
 	return deviceFlowInfo, nil
 }
+
+// RetryFlowForAccount returns a flow that asks the IdP to re-authenticate, for
+// a login answered with an account other than the one hinted. Returns nil when
+// the flow cannot ask — the caller then proceeds with the token it has.
+//
+// Proceeding rather than failing is deliberate. The hint is an email that may
+// simply have changed since it was stored, and refusing the login would lock a
+// user out of their own profile over a rename. The retry gives the account a
+// chance to be corrected; the server still rejects a token that does not own
+// the peer.
+func RetryFlowForAccount(flow OAuthFlow) OAuthFlow {
+	forcer, ok := flow.(accountPromptForcer)
+	if !ok {
+		return nil
+	}
+	forcer.ForceAccountPrompt()
+	return flow
+}

client/internal/auth/pkce_flow.go

@@ -87,10 +87,11 @@ func validatePKCEConfig(config *PKCEAuthProviderConfig) error {
 // PKCEAuthorizationFlow implements the OAuthFlow interface for
 // the Authorization Code Flow with PKCE.
 type PKCEAuthorizationFlow struct {
-	providerConfig PKCEAuthProviderConfig
-	state          string
-	codeVerifier   string
-	oAuthConfig    *oauth2.Config
+	providerConfig     PKCEAuthProviderConfig
+	state              string
+	codeVerifier       string
+	oAuthConfig        *oauth2.Config
+	forceAccountPrompt bool
 }
 
 // NewPKCEAuthorizationFlow returns new PKCE authorization code flow.
@@ -153,11 +154,16 @@ func (p *PKCEAuthorizationFlow) RequestAuthInfo(ctx context.Context) (AuthFlowIn
 		oauth2.SetAuthURLParam("code_challenge", codeChallenge),
 		oauth2.SetAuthURLParam("audience", p.providerConfig.Audience),
 	}
+	forceAccountPrompt := p.forceAccountPrompt
+	p.forceAccountPrompt = false
+
 	if !p.providerConfig.DisablePromptLogin {
-		switch p.providerConfig.LoginFlag {
-		case common.LoginFlagPromptLogin:
+		switch {
+		case forceAccountPrompt:
 			params = append(params, oauth2.SetAuthURLParam("prompt", "login"))
-		case common.LoginFlagMaxAge0:
+		case p.providerConfig.LoginFlag == common.LoginFlagPromptLogin:
+			params = append(params, oauth2.SetAuthURLParam("prompt", "login"))
+		case p.providerConfig.LoginFlag == common.LoginFlagMaxAge0:
 			params = append(params, oauth2.SetAuthURLParam("max_age", "0"))
 		}
 	}
@@ -178,6 +184,20 @@ func (p *PKCEAuthorizationFlow) SetLoginHint(hint string) {
 	p.providerConfig.LoginHint = hint
 }
 
+// ForceAccountPrompt makes the next authorization request ask the IdP to
+// re-authenticate instead of answering from the session it already holds. Used
+// to retry a login that came back for an account other than the one hinted.
+//
+// The next RequestAuthInfo consumes the flag, so a flow that outlives its retry
+// goes back to the configured behaviour instead of re-authenticating forever.
+//
+// DisablePromptLogin still wins: it is set for IdPs that break on prompt=login,
+// where retrying with it would replace a wrong-account login with one that
+// cannot complete at all.
+func (p *PKCEAuthorizationFlow) ForceAccountPrompt() {
+	p.forceAccountPrompt = true
+}
+
 // WaitToken waits for the OAuth token in the PKCE Authorization Flow.
 // It starts an HTTP server to receive the OAuth token callback and waits for the token or an error.
 // Once the token is received, it is converted to TokenInfo and validated before returning.
@@ -310,49 +330,48 @@ func (p *PKCEAuthorizationFlow) parseOAuthToken(token *oauth2.Token) (TokenInfo,
 		return TokenInfo{}, fmt.Errorf("authentication failed: invalid access token - %w", err)
 	}
 
-	email, err := parseEmailFromIDToken(tokenInfo.IDToken)
+	email, fromEmailClaim, err := parseEmailFromIDToken(tokenInfo.IDToken)
 	if err != nil {
 		log.Warnf("failed to parse email from ID token: %v", err)
 	} else {
 		tokenInfo.Email = email
+		if fromEmailClaim {
+			tokenInfo.EmailClaim = email
+		}
 	}
 
 	return tokenInfo, nil
 }
 
 // parseEmailFromIDToken extracts the email (or name) claim from an ID token
-// without verifying its signature. The value is best-effort and used only as a
-// UX convenience (login hint prefill and display); it never drives an
-// authorization decision. The authoritative identity is established server-side
-// from the signature-verified token.
-func parseEmailFromIDToken(token string) (string, error) {
+// without verifying its signature. The value is best-effort: it prefills the
+// login hint and is displayed. Account matching (see MatchesAccount) only uses
+// it when it came from the email claim, which fromEmailClaim reports. It never
+// grants anything — the authoritative identity is established server-side from
+// the signature-verified token.
+func parseEmailFromIDToken(token string) (value string, fromEmailClaim bool, err error) {
 	parts := strings.Split(token, ".")
 	if len(parts) < 2 {
-		return "", fmt.Errorf("invalid token format")
+		return "", false, fmt.Errorf("invalid token format")
 	}
 
 	data, err := base64.RawURLEncoding.DecodeString(parts[1])
 	if err != nil {
-		return "", fmt.Errorf("failed to decode payload: %w", err)
+		return "", false, fmt.Errorf("failed to decode payload: %w", err)
 	}
 	var claims map[string]interface{}
 	if err := json.Unmarshal(data, &claims); err != nil {
-		return "", fmt.Errorf("json unmarshal error: %w", err)
+		return "", false, fmt.Errorf("json unmarshal error: %w", err)
 	}
 
-	var email string
-	if emailValue, ok := claims["email"].(string); ok {
-		email = emailValue
-	} else {
-		val, ok := claims["name"].(string)
-		if ok {
-			email = val
-		} else {
-			return "", fmt.Errorf("email or name field not found in token payload")
-		}
+	if email, ok := claims["email"].(string); ok {
+		return email, true, nil
+	}
+	if name, ok := claims["name"].(string); ok {
+		return name, false, nil
 	}
 
-	return email, nil
+	return "", false, fmt.Errorf("email or name field not found in token payload")
 }
 
 func createCodeChallenge(codeVerifier string) string {

client/internal/auth/pkce_flow_test.go

@@ -76,6 +76,32 @@ func TestPromptLogin(t *testing.T) {
 	}
 }
 
+func TestForceAccountPromptAppliesOnlyToTheRetry(t *testing.T) {
+	config := PKCEAuthProviderConfig{
+		ClientID:              "test-client-id",
+		Audience:              "test-audience",
+		TokenEndpoint:         "https://test-token-endpoint.com/token",
+		Scope:                 "openid email profile",
+		AuthorizationEndpoint: "https://test-auth-endpoint.com/authorize",
+		RedirectURLs:          []string{"http://127.0.0.1:33992/"},
+		UseIDToken:            true,
+		LoginFlag:             mgm.LoginFlagNone,
+	}
+	pkce, err := NewPKCEAuthorizationFlow(config)
+	require.NoError(t, err)
+
+	pkce.ForceAccountPrompt()
+
+	retry, err := pkce.RequestAuthInfo(context.Background())
+	require.NoError(t, err)
+	require.Contains(t, retry.VerificationURIComplete, "prompt=login")
+
+	next, err := pkce.RequestAuthInfo(context.Background())
+	require.NoError(t, err)
+	require.NotContains(t, next.VerificationURIComplete, "prompt=login",
+		"the forced prompt outlived the retry it was armed for")
+}
+
 func TestIsPortInExcludedRange(t *testing.T) {
 	tests := []struct {
 		name            string

client/internal/auth/sessionwatch/watcher.go

@@ -90,8 +90,9 @@ type StatusRecorder interface {
 // fallback T-FinalWarningLead dialog (suppressed when the user dismissed
 // the first one for the same deadline). Safe for concurrent use.
 type Watcher struct {
-	lead      time.Duration
-	finalLead time.Duration
+	lead         time.Duration
+	finalLead    time.Duration
+	deadlineOnly bool
 
 	mu           sync.Mutex
 	current      time.Time
@@ -102,6 +103,7 @@ type Watcher struct {
 	dismissedAt  time.Time // deadline value the user dismissed via Dismiss(); gates fireFinal
 	closed       bool
 	recorder     StatusRecorder
+	nowFn        func() time.Time
 }
 
 // New returns a watcher with the package defaults WarningLead and
@@ -122,9 +124,17 @@ func NewWithLeads(lead, final time.Duration, recorder StatusRecorder) *Watcher {
 		lead:      lead,
 		finalLead: final,
 		recorder:  recorder,
+		nowFn:     time.Now,
 	}
 }
 
+// NewDeadlineOnly returns a watcher that validates and records deadlines but arms no warning timers.
+func NewDeadlineOnly(recorder StatusRecorder) *Watcher {
+	w := New(recorder)
+	w.deadlineOnly = true
+	return w
+}
+
 // Update sets the latest deadline. Pass the zero time to clear (e.g. when
 // a Sync push from the server omits the field because login expiration
 // was disabled).
@@ -181,7 +191,7 @@ func (w *Watcher) Update(deadline time.Time) error {
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
 
-	if deadline.After(now) {
+	if deadline.After(now) && !w.deadlineOnly {
 		w.armTimerLocked(deadline)
 	}
 	recorder := w.recorder
@@ -303,6 +313,11 @@ func (w *Watcher) fire(armedFor time.Time) {
 		w.mu.Unlock()
 		return
 	}
+	now := w.nowFn()
+	if isLate(now, armedFor, max(w.finalLead, 0)) {
+		w.fireLateLocked(armedFor, now)
+		return
+	}
 	w.firedAt = armedFor
 	recorder := w.recorder
 	w.mu.Unlock()
@@ -331,6 +346,14 @@ func (w *Watcher) fireFinal(armedFor time.Time) {
 		log.Infof("auth session final-warning skipped (dismissed by user)")
 		return
 	}
+	now := w.nowFn()
+	if isLate(now, armedFor, 0) {
+		w.finalFiredAt = armedFor
+		w.mu.Unlock()
+		log.Infof("auth session final-warning skipped for deadline %s (passed %s ago)",
+			armedFor.Format(time.RFC3339), now.Round(0).Sub(armedFor).Round(time.Second))
+		return
+	}
 	w.finalFiredAt = armedFor
 	recorder := w.recorder
 	w.mu.Unlock()
@@ -341,6 +364,39 @@ func (w *Watcher) fireFinal(armedFor time.Time) {
 	publishWarning(recorder, armedFor, true)
 }
 
+// fireLateLocked handles a T-WarningLead callback that fired inside the
+// final-warning window: it sends the final warning in its place while the
+// deadline has not passed and the user has not dismissed it, so a resume
+// with time left still warns. The caller must hold w.mu; this helper
+// releases it.
+func (w *Watcher) fireLateLocked(armedFor, now time.Time) {
+	w.firedAt = armedFor
+	switch {
+	case w.dismissedAt.Equal(armedFor):
+		w.mu.Unlock()
+		log.Infof("auth session expiry soon warning skipped (dismissed by user)")
+		return
+	case w.finalFiredAt.Equal(armedFor):
+		w.mu.Unlock()
+		log.Infof("auth session expiry soon warning skipped (final warning already fired)")
+		return
+	case isLate(now, armedFor, 0):
+		w.mu.Unlock()
+		log.Infof("auth session expiry soon warning skipped for deadline %s (passed %s ago)",
+			armedFor.Format(time.RFC3339), now.Round(0).Sub(armedFor).Round(time.Second))
+		return
+	}
+	w.finalFiredAt = armedFor
+	recorder := w.recorder
+	w.mu.Unlock()
+	if recorder == nil {
+		return
+	}
+	log.Infof("auth session expiry soon warning fired inside the final-warning window, sending final warning for deadline %s",
+		armedFor.Format(time.RFC3339))
+	publishWarning(recorder, armedFor, true)
+}
+
 // armOneShotLocked schedules cb at fireAt. When fireAt is already in the
 // past it dispatches on the next scheduler tick so a state-change recorder
 // notification (invoked after w.mu is released) lands first. Caller must
@@ -380,3 +436,11 @@ func publishWarning(recorder StatusRecorder, deadline time.Time, final bool) {
 		meta,
 	)
 }
+
+// isLate reports whether the wall clock now has already reached armedFor
+// minus cutoffLead. The timers run on the monotonic clock, which can stall
+// while the host sleeps, so a timer can fire long after the window it was
+// armed for.
+func isLate(now, armedFor time.Time, cutoffLead time.Duration) bool {
+	return !now.Round(0).Before(armedFor.Add(-cutoffLead).Round(0))
+}

client/internal/auth/sessionwatch/watcher_test.go

@@ -527,3 +527,201 @@ func TestDismissBeforeUpdateIsNoop(t *testing.T) {
 	}
 	t.Fatalf("final-warning did not publish after no-op pre-Update Dismiss, events=%+v", r.snapshot())
 }
+
+func TestIsLate(t *testing.T) {
+	armedFor := time.Date(2026, 10, 1, 12, 0, 0, 0, time.UTC)
+	lead := 2 * time.Minute
+	tests := []struct {
+		name       string
+		now        time.Time
+		cutoffLead time.Duration
+		want       bool
+	}{
+		{"before cutoff", armedFor.Add(-3 * time.Minute), lead, false},
+		{"at cutoff", armedFor.Add(-lead), lead, true},
+		{"after cutoff", armedFor.Add(-time.Minute), lead, true},
+		{"zero lead before deadline", armedFor.Add(-time.Second), 0, false},
+		{"zero lead at deadline", armedFor, 0, true},
+		{"zero lead after deadline", armedFor.Add(time.Second), 0, true},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := isLate(tt.now, armedFor, tt.cutoffLead); got != tt.want {
+				t.Fatalf("isLate(%s, %s, %s) = %v, want %v", tt.now, armedFor, tt.cutoffLead, got, tt.want)
+			}
+		})
+	}
+}
+
+func TestIsLateIgnoresMonotonicReading(t *testing.T) {
+	now := time.Now()
+	wallOnly := now.Round(0)
+	if isLate(now, wallOnly.Add(time.Second), 0) {
+		t.Fatalf("now with monotonic reading must compare as wall clock before a later wall-only deadline")
+	}
+	if !isLate(now, wallOnly, 0) {
+		t.Fatalf("now with monotonic reading must compare as wall clock at an equal wall-only deadline")
+	}
+}
+
+func TestLateTimerFiring(t *testing.T) {
+	tests := []struct {
+		name       string
+		final      bool
+		beforeDl   time.Duration
+		wantWarns  int
+		wantFinals int
+	}{
+		{"warning on resume inside window", false, 3 * time.Minute, 1, 0},
+		{"warning promoted to final inside final window", false, time.Minute, 0, 1},
+		{"warning skipped past deadline", false, -time.Minute, 0, 0},
+		{"final on resume before deadline", true, time.Minute, 0, 1},
+		{"final skipped past deadline", true, -time.Minute, 0, 0},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			r := &fakeRecorder{}
+			w := New(r)
+			defer w.Close()
+
+			// The deadline is an hour out so the real timers never fire
+			// during the test; the late callback is invoked directly with an
+			// injected clock that simulates a resume near the deadline.
+			d := time.Now().Add(time.Hour).Round(0)
+			w.nowFn = func() time.Time { return d.Add(-tt.beforeDl) }
+			if err := w.Update(d); err != nil {
+				t.Fatalf("Update: %v", err)
+			}
+
+			if tt.final {
+				w.fireFinal(d)
+			} else {
+				w.fire(d)
+			}
+
+			events := r.snapshot()
+			if got := countWhere(events, event.isWarning); got != tt.wantWarns {
+				t.Fatalf("expected %d warning publishes, got %d: %+v", tt.wantWarns, got, events)
+			}
+			if got := countWhere(events, event.isFinalWarning); got != tt.wantFinals {
+				t.Fatalf("expected %d final-warning publishes, got %d: %+v", tt.wantFinals, got, events)
+			}
+		})
+	}
+}
+
+func TestPromotedFinalWarningIsNotRepeated(t *testing.T) {
+	r := &fakeRecorder{}
+	w := New(r)
+	defer w.Close()
+
+	d := time.Now().Add(time.Hour).Round(0)
+	now := d.Add(-time.Minute)
+	w.nowFn = func() time.Time { return now }
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	w.fire(d)
+	// The final timer was suspended too, so it fires even later than the
+	// warning timer, here still just before the deadline.
+	now = d.Add(-30 * time.Second)
+	w.fireFinal(d)
+
+	events := r.snapshot()
+	if got := countWhere(events, event.isFinalWarning); got != 1 {
+		t.Fatalf("expected exactly 1 final-warning publish, got %d: %+v", got, events)
+	}
+	if got := countWhere(events, event.isWarning); got != 0 {
+		t.Fatalf("expected no regular warning publish, got %d: %+v", got, events)
+	}
+}
+
+func TestPromotionRespectsDismiss(t *testing.T) {
+	r := &fakeRecorder{}
+	w := New(r)
+	defer w.Close()
+
+	d := time.Now().Add(time.Hour).Round(0)
+	w.nowFn = func() time.Time { return d.Add(-time.Minute) }
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	w.Dismiss()
+	w.fire(d)
+
+	events := r.snapshot()
+	if got := countWhere(events, func(e event) bool { return e.kind == publish }); got != 0 {
+		t.Fatalf("expected no publish after dismiss, got %d: %+v", got, events)
+	}
+}
+
+func TestPromotionSkippedWhenFinalAlreadyFired(t *testing.T) {
+	r := &fakeRecorder{}
+	w := New(r)
+	defer w.Close()
+
+	// Both timers fall in the past after a long suspend and are dispatched
+	// with a zero delay, so the final callback can run before the warning one.
+	d := time.Now().Add(time.Hour).Round(0)
+	w.nowFn = func() time.Time { return d.Add(-time.Minute) }
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	w.fireFinal(d)
+	w.fire(d)
+
+	events := r.snapshot()
+	if got := countWhere(events, event.isFinalWarning); got != 1 {
+		t.Fatalf("expected exactly 1 final-warning publish, got %d: %+v", got, events)
+	}
+	if got := countWhere(events, event.isWarning); got != 0 {
+		t.Fatalf("expected no regular warning publish, got %d: %+v", got, events)
+	}
+}
+
+func TestDeadlineOnlyRecordsDeadlineWithoutWarnings(t *testing.T) {
+	r := &fakeRecorder{}
+	w := NewDeadlineOnly(r)
+	defer w.Close()
+
+	// With the default leads this deadline would otherwise fire both
+	// timers on the next tick.
+	d := time.Now().Add(50 * time.Millisecond).Round(0)
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+	if got := r.deadline(); !got.Equal(d) {
+		t.Fatalf("expected recorder deadline %v, got %v", d, got)
+	}
+
+	time.Sleep(100 * time.Millisecond)
+
+	events := r.snapshot()
+	if got := countWhere(events, func(e event) bool { return e.kind == publish }); got != 0 {
+		t.Fatalf("expected no publish in deadline-only mode, got %d: %+v", got, events)
+	}
+	if w.timer != nil || w.finalTimer != nil {
+		t.Fatal("expected no timers armed in deadline-only mode")
+	}
+}
+
+func TestDeadlineOnlyStillRejectsOutOfRangeDeadlines(t *testing.T) {
+	r := &fakeRecorder{}
+	w := NewDeadlineOnly(r)
+	defer w.Close()
+
+	if err := w.Update(time.Now().Add(time.Hour)); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	err := w.Update(time.Now().Add(-maxPastHorizon - time.Hour))
+	if !errors.Is(err, ErrDeadlineInPast) {
+		t.Fatalf("expected ErrDeadlineInPast, got %v", err)
+	}
+	if got := r.deadline(); !got.IsZero() {
+		t.Fatalf("expected recorder cleared after rejection, got %v", got)
+	}
+}

client/internal/connect.go

@@ -292,6 +292,10 @@ func (c *ConnectClient) run(mobileDependency MobileDependency, runningChan chan
 			return nil
 		}
 
+		if c.updateManager != nil {
+			c.updateManager.ResetMode()
+		}
+
 		// suspend connection attempts while the OS reports no usable network
 		if waited, err := c.netMgr.Wait(c.ctx); err != nil {
 			return nil

client/internal/debug/debug.go

@@ -379,9 +379,38 @@ func NewBundleGenerator(deps GeneratorDependencies, cfg BundleConfig) *BundleGen
 	}
 }
 
+// bundleFilePattern names the bundle zips Generate creates in tempDir; the
+// asterisk is filled in by os.CreateTemp.
+const bundleFilePattern = "netbird.debug.*.zip"
+
+const exportedBundlePrefix = "netbird.debug-file."
+
+const exportedBundleMaxAge = 24 * time.Hour
+
+// RemoveStaleBundles deletes bundle zips that an interrupted generation or
+// upload left behind in dir. Only files older than maxAge go, so a bundle that
+// another caller is still writing or uploading in the same directory survives.
+// Exported bundles are kept for exportedBundleMaxAge instead.
+func RemoveStaleBundles(dir string, maxAge time.Duration) {
+	removeStaleFiles(dir, bundleFilePattern, maxAge)
+	removeStaleFiles(dir, exportedBundlePrefix+"*.zip", exportedBundleMaxAge)
+}
+
+// ExportBundle renames a generated bundle out of the RemoveStaleBundles pattern
+// and returns the new path. The caller owns the file from then on; an export
+// abandoned for longer than exportedBundleMaxAge is removed by RemoveStaleBundles.
+func ExportBundle(path string) (string, error) {
+	base := strings.TrimPrefix(filepath.Base(path), strings.SplitN(bundleFilePattern, "*", 2)[0])
+	exported := filepath.Join(filepath.Dir(path), exportedBundlePrefix+base)
+	if err := os.Rename(path, exported); err != nil {
+		return "", fmt.Errorf("export debug bundle: %w", err)
+	}
+	return exported, nil
+}
+
 // Generate creates a debug bundle and returns the location.
 func (g *BundleGenerator) Generate() (resp string, err error) {
-	bundlePath, err := os.CreateTemp(g.tempDir, "netbird.debug.*.zip")
+	bundlePath, err := os.CreateTemp(g.tempDir, bundleFilePattern)
 	if err != nil {
 		return "", fmt.Errorf("create zip file: %w", err)
 	}
@@ -1725,3 +1754,26 @@ func anonymizeSlice(v []any, anonymizer *anonymize.Anonymizer) []any {
 	}
 	return v
 }
+
+func removeStaleFiles(dir, pattern string, maxAge time.Duration) {
+	matches, err := filepath.Glob(filepath.Join(dir, pattern))
+	if err != nil {
+		log.Debugf("glob stale debug bundles in %s: %v", dir, err)
+		return
+	}
+
+	cutoff := time.Now().Add(-maxAge)
+	for _, path := range matches {
+		info, err := os.Stat(path)
+		if err != nil || info.ModTime().After(cutoff) {
+			continue
+		}
+		if err := os.Remove(path); err != nil {
+			if !errors.Is(err, fs.ErrNotExist) {
+				log.Warnf("remove stale debug bundle %s: %v", path, err)
+			}
+			continue
+		}
+		log.Infof("removed stale debug bundle %s", path)
+	}
+}

client/internal/debug/debug_test.go

@@ -4,6 +4,7 @@ import (
 	"archive/zip"
 	"bytes"
 	"encoding/json"
+	"fmt"
 	"net"
 	"net/netip"
 	"net/url"
@@ -845,6 +846,7 @@ func TestAddConfig_AllFieldsCovered(t *testing.T) {
 		"ClientCertKeyPair":    "non-config: parsed cert pair, not serialized",
 		"Name":                 "non-config: profile name is not needed for debug purposes",
 		"policy":               "non-config: in-memory MDM policy snapshot, surfaced via Config.Policy() / GetConfigResponse.MDMManagedFields",
+		"probing":              "non-config: marks a throwaway copy built to be diffed against; never set on a config anyone runs with",
 		"DebugBundleUploadURL": "sensitive: MDM-provided upload URL may carry credentials or query tokens; kept out of the shared bundle",
 	}
 
@@ -969,3 +971,52 @@ func renderAddConfigSpecific(g *BundleGenerator) string {
 func newAnonymizerForTest() *anonymize.Anonymizer {
 	return anonymize.NewAnonymizer(anonymize.DefaultAddresses())
 }
+
+func TestRemoveStaleBundles(t *testing.T) {
+	dir := t.TempDir()
+	stale := filepath.Join(dir, "netbird.debug.111.zip")
+	fresh := filepath.Join(dir, "netbird.debug.222.zip")
+	other := filepath.Join(dir, "netbird.debug.333.txt")
+	owned := filepath.Join(dir, "netbird.debug.444.zip")
+	abandoned := filepath.Join(dir, "netbird.debug.555.zip")
+	for _, p := range []string{stale, fresh, other, owned, abandoned} {
+		require.NoError(t, os.WriteFile(p, []byte("x"), 0o600))
+	}
+	exported, err := ExportBundle(owned)
+	require.NoError(t, err)
+	exportedAbandoned, err := ExportBundle(abandoned)
+	require.NoError(t, err)
+	old := time.Now().Add(-2 * time.Hour)
+	for _, p := range []string{stale, other, exported} {
+		require.NoError(t, os.Chtimes(p, old, old))
+	}
+	ancient := time.Now().Add(-exportedBundleMaxAge - time.Hour)
+	require.NoError(t, os.Chtimes(exportedAbandoned, ancient, ancient))
+
+	RemoveStaleBundles(dir, time.Hour)
+
+	assert.NoFileExists(t, stale, "bundle older than maxAge should be removed")
+	assert.FileExists(t, fresh, "bundle younger than maxAge must survive, it may still be uploading")
+	assert.FileExists(t, other, "files outside the bundle pattern must not be touched")
+	assert.NoFileExists(t, owned)
+	assert.FileExists(t, exported, "exported bundle is caller-owned and must survive maxAge")
+	assert.NoFileExists(t, exportedAbandoned, "exported bundle older than exportedBundleMaxAge is abandoned")
+}
+
+func TestBundleIncludesNetworkMap(t *testing.T) {
+	for _, anonymize := range []bool{false, true} {
+		t.Run(fmt.Sprintf("anonymize=%t", anonymize), func(t *testing.T) {
+			g := NewBundleGenerator(GeneratorDependencies{
+				SyncResponse: &mgmProto.SyncResponse{NetworkMap: &mgmProto.NetworkMap{Serial: 1}},
+			}, BundleConfig{Anonymize: anonymize})
+
+			require.Contains(t, bundleEntries(t, g), "network_map.json")
+		})
+	}
+}
+
+func TestBundleOmitsNetworkMapWithoutSyncResponse(t *testing.T) {
+	g := NewBundleGenerator(GeneratorDependencies{}, BundleConfig{})
+
+	require.NotContains(t, bundleEntries(t, g), "network_map.json")
+}

client/internal/dns/server_privileged_test.go

@@ -9,9 +9,9 @@ import (
 	"os"
 	"testing"
 
-	"go.uber.org/mock/gomock"
 	"github.com/miekg/dns"
 	"github.com/stretchr/testify/assert"
+	"go.uber.org/mock/gomock"
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 
 	"github.com/netbirdio/netbird/client/iface"
@@ -24,6 +24,10 @@ import (
 	nbdns "github.com/netbirdio/netbird/dns"
 )
 
+// testIFaceBlackList mirrors the overlay prefixes profilemanager.DefaultInterfaceBlacklist
+// carries. Declared here rather than imported because profilemanager imports this package.
+var testIFaceBlackList = []string{"wt", "utun", "tun0"}
+
 func TestUpdateDNSServer(t *testing.T) {
 
 	nameServers := []nbdns.NameServer{
@@ -243,10 +247,7 @@ func TestUpdateDNSServer(t *testing.T) {
 	for n, testCase := range testCases {
 		t.Run(testCase.name, func(t *testing.T) {
 			privKey, _ := wgtypes.GenerateKey()
-			newNet, err := stdnet.NewNet(context.Background(), nil)
-			if err != nil {
-				t.Fatal(err)
-			}
+			newNet := stdnet.NewNet(context.Background(), testIFaceBlackList, nil)
 
 			opts := iface.WGIFaceOpts{
 				IFaceName:    fmt.Sprintf("utun230%d", n),
@@ -348,11 +349,7 @@ func TestDNSFakeResolverHandleUpdates(t *testing.T) {
 	defer t.Setenv("NB_WG_KERNEL_DISABLED", ov)
 
 	t.Setenv("NB_WG_KERNEL_DISABLED", "true")
-	newNet, err := stdnet.NewNet(context.Background(), []string{"utun2301"})
-	if err != nil {
-		t.Errorf("create stdnet: %v", err)
-		return
-	}
+	newNet := stdnet.NewNet(context.Background(), []string{"utun2301"}, nil)
 
 	privKey, _ := wgtypes.GeneratePrivateKey()
 	opts := iface.WGIFaceOpts{

client/internal/dns/server_test.go

@@ -394,11 +394,7 @@ func createWgInterfaceWithBind(t *testing.T) (*iface.WGIface, error) {
 	defer t.Setenv("NB_WG_KERNEL_DISABLED", ov)
 
 	t.Setenv("NB_WG_KERNEL_DISABLED", "true")
-	newNet, err := stdnet.NewNet(context.Background(), []string{"utun2301"})
-	if err != nil {
-		t.Fatalf("create stdnet: %v", err)
-		return nil, err
-	}
+	newNet := stdnet.NewNet(context.Background(), []string{"utun2301"}, nil)
 
 	privKey, _ := wgtypes.GeneratePrivateKey()
 

client/internal/ebpf/ebpf/bpf_bpfeb.go

@@ -1,148 +0,0 @@
-// Code generated by bpf2go; DO NOT EDIT.
-//go:build mips || mips64 || ppc64 || s390x
-
-package ebpf
-
-import (
-	"bytes"
-	_ "embed"
-	"fmt"
-	"io"
-
-	"github.com/cilium/ebpf"
-)
-
-// loadBpf returns the embedded CollectionSpec for bpf.
-func loadBpf() (*ebpf.CollectionSpec, error) {
-	reader := bytes.NewReader(_BpfBytes)
-	spec, err := ebpf.LoadCollectionSpecFromReader(reader)
-	if err != nil {
-		return nil, fmt.Errorf("can't load bpf: %w", err)
-	}
-
-	return spec, err
-}
-
-// loadBpfObjects loads bpf and converts it into a struct.
-//
-// The following types are suitable as obj argument:
-//
-//	*bpfObjects
-//	*bpfPrograms
-//	*bpfMaps
-//
-// See ebpf.CollectionSpec.LoadAndAssign documentation for details.
-func loadBpfObjects(obj interface{}, opts *ebpf.CollectionOptions) error {
-	spec, err := loadBpf()
-	if err != nil {
-		return err
-	}
-
-	return spec.LoadAndAssign(obj, opts)
-}
-
-// bpfSpecs contains maps and programs before they are loaded into the kernel.
-//
-// It can be passed ebpf.CollectionSpec.Assign.
-type bpfSpecs struct {
-	bpfProgramSpecs
-	bpfMapSpecs
-	bpfVariableSpecs
-}
-
-// bpfProgramSpecs contains programs before they are loaded into the kernel.
-//
-// It can be passed ebpf.CollectionSpec.Assign.
-type bpfProgramSpecs struct {
-	NbXdpProg *ebpf.ProgramSpec `ebpf:"nb_xdp_prog"`
-}
-
-// bpfMapSpecs contains maps before they are loaded into the kernel.
-//
-// It can be passed ebpf.CollectionSpec.Assign.
-type bpfMapSpecs struct {
-	NbFeatures           *ebpf.MapSpec `ebpf:"nb_features"`
-	NbWgProxySettingsMap *ebpf.MapSpec `ebpf:"nb_wg_proxy_settings_map"`
-}
-
-// bpfVariableSpecs contains global variables before they are loaded into the kernel.
-//
-// It can be passed ebpf.CollectionSpec.Assign.
-type bpfVariableSpecs struct {
-	FlagFeatureWgProxy *ebpf.VariableSpec `ebpf:"flag_feature_wg_proxy"`
-	MapKeyFeatures     *ebpf.VariableSpec `ebpf:"map_key_features"`
-	MapKeyProxyPort    *ebpf.VariableSpec `ebpf:"map_key_proxy_port"`
-	MapKeyWgPort       *ebpf.VariableSpec `ebpf:"map_key_wg_port"`
-	ProxyPort          *ebpf.VariableSpec `ebpf:"proxy_port"`
-	WgPort             *ebpf.VariableSpec `ebpf:"wg_port"`
-}
-
-// bpfObjects contains all objects after they have been loaded into the kernel.
-//
-// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
-type bpfObjects struct {
-	bpfPrograms
-	bpfMaps
-	bpfVariables
-}
-
-func (o *bpfObjects) Close() error {
-	return _BpfClose(
-		&o.bpfPrograms,
-		&o.bpfMaps,
-	)
-}
-
-// bpfMaps contains all maps after they have been loaded into the kernel.
-//
-// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
-type bpfMaps struct {
-	NbFeatures           *ebpf.Map `ebpf:"nb_features"`
-	NbWgProxySettingsMap *ebpf.Map `ebpf:"nb_wg_proxy_settings_map"`
-}
-
-func (m *bpfMaps) Close() error {
-	return _BpfClose(
-		m.NbFeatures,
-		m.NbWgProxySettingsMap,
-	)
-}
-
-// bpfVariables contains all global variables after they have been loaded into the kernel.
-//
-// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
-type bpfVariables struct {
-	FlagFeatureWgProxy *ebpf.Variable `ebpf:"flag_feature_wg_proxy"`
-	MapKeyFeatures     *ebpf.Variable `ebpf:"map_key_features"`
-	MapKeyProxyPort    *ebpf.Variable `ebpf:"map_key_proxy_port"`
-	MapKeyWgPort       *ebpf.Variable `ebpf:"map_key_wg_port"`
-	ProxyPort          *ebpf.Variable `ebpf:"proxy_port"`
-	WgPort             *ebpf.Variable `ebpf:"wg_port"`
-}
-
-// bpfPrograms contains all programs after they have been loaded into the kernel.
-//
-// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
-type bpfPrograms struct {
-	NbXdpProg *ebpf.Program `ebpf:"nb_xdp_prog"`
-}
-
-func (p *bpfPrograms) Close() error {
-	return _BpfClose(
-		p.NbXdpProg,
-	)
-}
-
-func _BpfClose(closers ...io.Closer) error {
-	for _, closer := range closers {
-		if err := closer.Close(); err != nil {
-			return err
-		}
-	}
-	return nil
-}
-
-// Do not access this directly.
-//
-//go:embed bpf_bpfeb.o
-var _BpfBytes []byte

client/internal/ebpf/ebpf/bpf_bpfel.go

@@ -1,148 +0,0 @@
-// Code generated by bpf2go; DO NOT EDIT.
-//go:build 386 || amd64 || arm || arm64 || loong64 || mips64le || mipsle || ppc64le || riscv64 || wasm
-
-package ebpf
-
-import (
-	"bytes"
-	_ "embed"
-	"fmt"
-	"io"
-
-	"github.com/cilium/ebpf"
-)
-
-// loadBpf returns the embedded CollectionSpec for bpf.
-func loadBpf() (*ebpf.CollectionSpec, error) {
-	reader := bytes.NewReader(_BpfBytes)
-	spec, err := ebpf.LoadCollectionSpecFromReader(reader)
-	if err != nil {
-		return nil, fmt.Errorf("can't load bpf: %w", err)
-	}
-
-	return spec, err
-}
-
-// loadBpfObjects loads bpf and converts it into a struct.
-//
-// The following types are suitable as obj argument:
-//
-//	*bpfObjects
-//	*bpfPrograms
-//	*bpfMaps
-//
-// See ebpf.CollectionSpec.LoadAndAssign documentation for details.
-func loadBpfObjects(obj interface{}, opts *ebpf.CollectionOptions) error {
-	spec, err := loadBpf()
-	if err != nil {
-		return err
-	}
-
-	return spec.LoadAndAssign(obj, opts)
-}
-
-// bpfSpecs contains maps and programs before they are loaded into the kernel.
-//
-// It can be passed ebpf.CollectionSpec.Assign.
-type bpfSpecs struct {
-	bpfProgramSpecs
-	bpfMapSpecs
-	bpfVariableSpecs
-}
-
-// bpfProgramSpecs contains programs before they are loaded into the kernel.
-//
-// It can be passed ebpf.CollectionSpec.Assign.
-type bpfProgramSpecs struct {
-	NbXdpProg *ebpf.ProgramSpec `ebpf:"nb_xdp_prog"`
-}
-
-// bpfMapSpecs contains maps before they are loaded into the kernel.
-//
-// It can be passed ebpf.CollectionSpec.Assign.
-type bpfMapSpecs struct {
-	NbFeatures           *ebpf.MapSpec `ebpf:"nb_features"`
-	NbWgProxySettingsMap *ebpf.MapSpec `ebpf:"nb_wg_proxy_settings_map"`
-}
-
-// bpfVariableSpecs contains global variables before they are loaded into the kernel.
-//
-// It can be passed ebpf.CollectionSpec.Assign.
-type bpfVariableSpecs struct {
-	FlagFeatureWgProxy *ebpf.VariableSpec `ebpf:"flag_feature_wg_proxy"`
-	MapKeyFeatures     *ebpf.VariableSpec `ebpf:"map_key_features"`
-	MapKeyProxyPort    *ebpf.VariableSpec `ebpf:"map_key_proxy_port"`
-	MapKeyWgPort       *ebpf.VariableSpec `ebpf:"map_key_wg_port"`
-	ProxyPort          *ebpf.VariableSpec `ebpf:"proxy_port"`
-	WgPort             *ebpf.VariableSpec `ebpf:"wg_port"`
-}
-
-// bpfObjects contains all objects after they have been loaded into the kernel.
-//
-// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
-type bpfObjects struct {
-	bpfPrograms
-	bpfMaps
-	bpfVariables
-}
-
-func (o *bpfObjects) Close() error {
-	return _BpfClose(
-		&o.bpfPrograms,
-		&o.bpfMaps,
-	)
-}
-
-// bpfMaps contains all maps after they have been loaded into the kernel.
-//
-// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
-type bpfMaps struct {
-	NbFeatures           *ebpf.Map `ebpf:"nb_features"`
-	NbWgProxySettingsMap *ebpf.Map `ebpf:"nb_wg_proxy_settings_map"`
-}
-
-func (m *bpfMaps) Close() error {
-	return _BpfClose(
-		m.NbFeatures,
-		m.NbWgProxySettingsMap,
-	)
-}
-
-// bpfVariables contains all global variables after they have been loaded into the kernel.
-//
-// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
-type bpfVariables struct {
-	FlagFeatureWgProxy *ebpf.Variable `ebpf:"flag_feature_wg_proxy"`
-	MapKeyFeatures     *ebpf.Variable `ebpf:"map_key_features"`
-	MapKeyProxyPort    *ebpf.Variable `ebpf:"map_key_proxy_port"`
-	MapKeyWgPort       *ebpf.Variable `ebpf:"map_key_wg_port"`
-	ProxyPort          *ebpf.Variable `ebpf:"proxy_port"`
-	WgPort             *ebpf.Variable `ebpf:"wg_port"`
-}
-
-// bpfPrograms contains all programs after they have been loaded into the kernel.
-//
-// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
-type bpfPrograms struct {
-	NbXdpProg *ebpf.Program `ebpf:"nb_xdp_prog"`
-}
-
-func (p *bpfPrograms) Close() error {
-	return _BpfClose(
-		p.NbXdpProg,
-	)
-}
-
-func _BpfClose(closers ...io.Closer) error {
-	for _, closer := range closers {
-		if err := closer.Close(); err != nil {
-			return err
-		}
-	}
-	return nil
-}
-
-// Do not access this directly.
-//
-//go:embed bpf_bpfel.o
-var _BpfBytes []byte

client/internal/ebpf/ebpf/manager_linux.go

@@ -1,115 +0,0 @@
-package ebpf
-
-import (
-	_ "embed"
-	"net"
-	"sync"
-
-	"github.com/cilium/ebpf/link"
-	"github.com/cilium/ebpf/rlimit"
-	log "github.com/sirupsen/logrus"
-
-	"github.com/netbirdio/netbird/client/internal/ebpf/manager"
-)
-
-const (
-	mapKeyFeatures uint32 = 0
-
-	featureFlagWGProxy = 0b00000001
-)
-
-var (
-	singleton     manager.Manager
-	singletonLock = &sync.Mutex{}
-)
-
-// required packages libbpf-dev, libc6-dev-i386-amd64-cross
-
-// GeneralManager is used to load multiple eBPF programs with a custom check (if then) done in prog.c
-// The manager simply adds a feature (byte) of each program to a map that is shared between the userspace and kernel.
-// When packet arrives, the C code checks for each feature (if it is set) and executes each enabled program (e.g., wg_proxy.c).
-//
-//go:generate go run github.com/cilium/ebpf/cmd/bpf2go -cc clang-14 bpf src/prog.c -- -I /usr/x86_64-linux-gnu/include -include src/bpf_map_def.h
-type GeneralManager struct {
-	lock         sync.Mutex
-	link         link.Link
-	featureFlags uint16
-	bpfObjs      bpfObjects
-}
-
-// GetEbpfManagerInstance return a static eBpf Manager instance
-func GetEbpfManagerInstance() manager.Manager {
-	singletonLock.Lock()
-	defer singletonLock.Unlock()
-	if singleton != nil {
-		return singleton
-	}
-	singleton = &GeneralManager{}
-	return singleton
-}
-
-func (tf *GeneralManager) setFeatureFlag(feature uint16) {
-	tf.featureFlags |= feature
-}
-
-func (tf *GeneralManager) loadXdp() error {
-	if tf.link != nil {
-		return nil
-	}
-	// it required for Docker
-	err := rlimit.RemoveMemlock()
-	if err != nil {
-		return err
-	}
-
-	iFace, err := net.InterfaceByName("lo")
-	if err != nil {
-		return err
-	}
-
-	// load pre-compiled programs into the kernel.
-	err = loadBpfObjects(&tf.bpfObjs, nil)
-	if err != nil {
-		return err
-	}
-
-	tf.link, err = link.AttachXDP(link.XDPOptions{
-		Program:   tf.bpfObjs.NbXdpProg,
-		Interface: iFace.Index,
-	})
-
-	if err != nil {
-		_ = tf.bpfObjs.Close()
-		tf.link = nil
-		return err
-	}
-	return nil
-}
-
-func (tf *GeneralManager) unsetFeatureFlag(feature uint16) error {
-	tf.lock.Lock()
-	defer tf.lock.Unlock()
-	tf.featureFlags &^= feature
-
-	if tf.link == nil {
-		return nil
-	}
-
-	if tf.featureFlags == 0 {
-		return tf.close()
-	}
-
-	return tf.bpfObjs.NbFeatures.Put(mapKeyFeatures, tf.featureFlags)
-}
-
-func (tf *GeneralManager) close() error {
-	log.Debugf("detach ebpf program ")
-	err := tf.bpfObjs.Close()
-	if err != nil {
-		log.Warnf("failed to close eBpf objects: %s", err)
-	}
-
-	err = tf.link.Close()
-	tf.link = nil
-	return err
-}

client/internal/ebpf/ebpf/manager_linux_test.go

@@ -1,31 +0,0 @@
-package ebpf
-
-import (
-	"testing"
-)
-
-func TestManager_setFeatureFlag(t *testing.T) {
-	mgr := GeneralManager{}
-	mgr.setFeatureFlag(featureFlagWGProxy)
-	if mgr.featureFlags != featureFlagWGProxy {
-		t.Errorf("invalid feature state")
-	}
-
-	mgr.setFeatureFlag(featureFlagWGProxy)
-	if mgr.featureFlags != featureFlagWGProxy {
-		t.Errorf("setting a flag twice must be idempotent, got: %d", mgr.featureFlags)
-	}
-}
-
-func TestManager_unsetFeatureFlag(t *testing.T) {
-	mgr := GeneralManager{}
-	mgr.setFeatureFlag(featureFlagWGProxy)
-
-	err := mgr.unsetFeatureFlag(featureFlagWGProxy)
-	if err != nil {
-		t.Errorf("unexpected error: %s", err)
-	}
-	if mgr.featureFlags != 0 {
-		t.Errorf("invalid feature state, expected: %d, got: %d", 0, mgr.featureFlags)
-	}
-}

client/internal/ebpf/ebpf/src/bpf_map_def.h

@@ -1,16 +0,0 @@
-// libbpf 1.0 removed struct bpf_map_def, but the programs here keep the legacy
-// map definitions: they load on kernels built without BTF, which BTF-style
-// (SEC(".maps")) definitions do not. Define the struct ourselves so the
-// programs compile against current libbpf headers.
-#ifndef NB_BPF_MAP_DEF_H
-#define NB_BPF_MAP_DEF_H
-
-struct bpf_map_def {
-	unsigned int type;
-	unsigned int key_size;
-	unsigned int value_size;
-	unsigned int max_entries;
-	unsigned int map_flags;
-};
-
-#endif

client/internal/ebpf/ebpf/src/prog.c

@@ -1,54 +0,0 @@
-#include <stdbool.h>
-#include <linux/if_ether.h> // ETH_P_IP
-#include <linux/udp.h>
-#include <linux/ip.h>
-#include <netinet/in.h>
-#include <linux/bpf.h>
-#include <bpf/bpf_helpers.h>
-#include "wg_proxy.c"
-
-const __u16 flag_feature_wg_proxy = 0b01;
-
-const __u32 map_key_features = 0;
-struct bpf_map_def SEC("maps") nb_features = {
-	.type = BPF_MAP_TYPE_ARRAY,
-	.key_size = sizeof(__u32),
-	.value_size = sizeof(__u16),
-	.max_entries = 10,
-};
-
-SEC("xdp")
-int nb_xdp_prog(struct xdp_md *ctx) {
-    __u16 *features;
-    features = bpf_map_lookup_elem(&nb_features, &map_key_features);
-    if (!features) {
-        return XDP_PASS;
-    }
-
-    void *data     = (void *)(long)ctx->data;
-    void *data_end = (void *)(long)ctx->data_end;
-    struct ethhdr *eth = data;
-    struct iphdr  *ip  = (data + sizeof(struct ethhdr));
-    struct udphdr *udp = (data + sizeof(struct ethhdr) + sizeof(struct iphdr));
-
-    // return early if not enough data
-    if (data + sizeof(struct ethhdr) + sizeof(struct iphdr) + sizeof(struct udphdr) > data_end){
-        return XDP_PASS;
-    }
-
-    // skip non IPv4 packages
-    if (eth->h_proto != htons(ETH_P_IP)) {
-       return XDP_PASS;
-    }
-
-    // skip non UPD packages
-    if (ip->protocol != IPPROTO_UDP) {
-       return XDP_PASS;
-    }
-
-    if (*features & flag_feature_wg_proxy) {
-        xdp_wg_proxy(ip, udp);
-    }
-    return XDP_PASS;
-}
-char _license[] SEC("license") = "GPL";

client/internal/ebpf/ebpf/src/readme.md

@@ -1,27 +0,0 @@
-# XDP programs
-
-`prog.c` is attached to the `lo` device and dispatches to the features enabled in the
-`nb_features` map. The only feature is the WireGuard proxy (`wg_proxy.c`): it rewrites
-loopback UDP sent from the WireGuard listen port so it reaches the userspace relay proxy
-port instead, and swaps the peer endpoint port into the source so the proxy can tell
-peers apart.
-
-Maps use the legacy `struct bpf_map_def` form, defined in `bpf_map_def.h` because libbpf
-1.0 removed it. They load on kernels built without BTF, which BTF-style (`SEC(".maps")`)
-definitions do not.
-
-Regenerate the objects with `go generate ./client/internal/ebpf/ebpf/`; it needs
-`clang-14`. Loading a regenerated object needs root, attaching it needs `bpf_link`
-(kernel >= 5.7), and only one XDP program can own `lo` at a time.
-
-# Debug
-
-The CONFIG_BPF_EVENTS kernel module is required for bpf_printk.
-Apply this code to use bpf_printk
-```
-#define bpf_printk(fmt, ...)                                                   \
-  ({                                                                           \
-    char ____fmt[] = fmt;                                                      \
-    bpf_trace_printk(____fmt, sizeof(____fmt), ##__VA_ARGS__);                 \
-  })
-```

client/internal/ebpf/ebpf/src/wg_proxy.c

@@ -1,60 +0,0 @@
-const __u32 map_key_proxy_port = 0;
-const __u32 map_key_wg_port = 1;
-
-struct bpf_map_def SEC("maps") nb_wg_proxy_settings_map = {
-	.type = BPF_MAP_TYPE_ARRAY,
-	.key_size = sizeof(__u32),
-	.value_size = sizeof(__u16),
-	.max_entries = 10,
-};
-
-__u16 proxy_port = 0;
-__u16 wg_port = 0;
-
-bool read_port_settings() {
-    __u16 *value;
-    value = bpf_map_lookup_elem(&nb_wg_proxy_settings_map, &map_key_proxy_port);
-    if (!value) {
-        return false;
-    }
-
-    proxy_port = *value;
-
-    value = bpf_map_lookup_elem(&nb_wg_proxy_settings_map, &map_key_wg_port);
-    if (!value) {
-        return false;
-    }
-    wg_port = htons(*value);
-
-    return true;
-}
-
-int xdp_wg_proxy(struct iphdr  *ip, struct udphdr *udp) {
-    if (proxy_port == 0 || wg_port == 0) {
-        if (!read_port_settings()){
-            return XDP_PASS;
-        }
-        // bpf_printk("proxy port: %d, wg port: %d", proxy_port, wg_port);
-    }
-
-    // 2130706433 = 127.0.0.1
-    if (ip->daddr != htonl(2130706433)) {
-        return XDP_PASS;
-    }
-
-    if (udp->source != wg_port){
-        return XDP_PASS;
-    }
-
-    __be16 new_src_port = udp->dest;
-    __be16 new_dst_port = htons(proxy_port);
-    udp->dest = new_dst_port;
-    udp->source = new_src_port;
-
-    // The ports are covered by the UDP checksum. This is an IPv4 loopback hop
-    // and the payload is already integrity-protected, so clear the checksum (a
-    // zero UDP checksum means "not computed" for IPv4) rather than leave a
-    // stale value the kernel would drop as UDP_CSUM.
-    udp->check = 0;
-	return XDP_PASS;
-}

client/internal/ebpf/ebpf/wg_proxy_linux.go

@@ -1,41 +0,0 @@
-package ebpf
-
-import log "github.com/sirupsen/logrus"
-
-const (
-	mapKeyProxyPort uint32 = 0
-	mapKeyWgPort    uint32 = 1
-)
-
-func (tf *GeneralManager) LoadWgProxy(proxyPort, wgPort int) error {
-	log.Debugf("load ebpf WG proxy")
-	tf.lock.Lock()
-	defer tf.lock.Unlock()
-
-	err := tf.loadXdp()
-	if err != nil {
-		return err
-	}
-
-	err = tf.bpfObjs.NbWgProxySettingsMap.Put(mapKeyProxyPort, uint16(proxyPort))
-	if err != nil {
-		return err
-	}
-
-	err = tf.bpfObjs.NbWgProxySettingsMap.Put(mapKeyWgPort, uint16(wgPort))
-	if err != nil {
-		return err
-	}
-
-	tf.setFeatureFlag(featureFlagWGProxy)
-	err = tf.bpfObjs.NbFeatures.Put(mapKeyFeatures, tf.featureFlags)
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (tf *GeneralManager) FreeWGProxy() error {
-	log.Debugf("free ebpf WG proxy")
-	return tf.unsetFeatureFlag(featureFlagWGProxy)
-}

client/internal/ebpf/instantiater_linux.go

@@ -1,15 +0,0 @@
-//go:build !android
-
-package ebpf
-
-import (
-	"github.com/netbirdio/netbird/client/internal/ebpf/ebpf"
-	"github.com/netbirdio/netbird/client/internal/ebpf/manager"
-)
-
-// GetEbpfManagerInstance is a wrapper function. This encapsulation is required because if the code import the internal
-// ebpf package the Go compiler will include the object files. But it is not supported on Android. It can cause instant
-// panic on older Android version.
-func GetEbpfManagerInstance() manager.Manager {
-	return ebpf.GetEbpfManagerInstance()
-}

client/internal/ebpf/instantiater_nonlinux.go

@@ -1,10 +0,0 @@
-//go:build !linux || android
-
-package ebpf
-
-import "github.com/netbirdio/netbird/client/internal/ebpf/manager"
-
-// GetEbpfManagerInstance return error because ebpf is not supported on all os
-func GetEbpfManagerInstance() manager.Manager {
-	panic("unsupported os")
-}

client/internal/ebpf/manager/manager.go

@@ -1,7 +0,0 @@
-package manager
-
-// Manager is used to load multiple eBPF programs. E.g., the WireGuard proxy
-type Manager interface {
-	LoadWgProxy(proxyPort, wgPort int) error
-	FreeWGProxy() error
-}

client/internal/engine.go

@@ -42,7 +42,6 @@ import (
 	dnsconfig "github.com/netbirdio/netbird/client/internal/dns/config"
 	"github.com/netbirdio/netbird/client/internal/dnsfwd"
 	"github.com/netbirdio/netbird/client/internal/expose"
-	"github.com/netbirdio/netbird/client/internal/ingressgw"
 	"github.com/netbirdio/netbird/client/internal/lazyconn"
 	"github.com/netbirdio/netbird/client/internal/metrics"
 	"github.com/netbirdio/netbird/client/internal/netflow"
@@ -58,6 +57,7 @@ import (
 	"github.com/netbirdio/netbird/client/internal/rosenpass"
 	"github.com/netbirdio/netbird/client/internal/routemanager"
 	"github.com/netbirdio/netbird/client/internal/statemanager"
+	"github.com/netbirdio/netbird/client/internal/stdnet"
 	"github.com/netbirdio/netbird/client/internal/syncstore"
 	"github.com/netbirdio/netbird/client/internal/updater"
 	"github.com/netbirdio/netbird/client/jobexec"
@@ -253,6 +253,9 @@ type Engine struct {
 
 	udpMux *udpmux.UniversalUDPMuxDefault
 
+	// wgDetector is shared by every ICE agent through the ICE config.
+	wgDetector *stdnet.WGDetector
+
 	// networkSerial is the latest CurrentSerial (state ID) of the network sent by the Management service
 	networkSerial uint64
 
@@ -269,11 +272,10 @@ type Engine struct {
 
 	statusRecorder *peer.Status
 
-	firewall          firewallManager.Manager
-	routeManager      routemanager.Manager
-	acl               acl.Manager
-	dnsForwardMgr     *dnsfwd.Manager
-	ingressGatewayMgr *ingressgw.Manager
+	firewall      firewallManager.Manager
+	routeManager  routemanager.Manager
+	acl           acl.Manager
+	dnsForwardMgr *dnsfwd.Manager
 
 	dnsServer dns.Server
 
@@ -371,6 +373,7 @@ func NewEngine(
 		mgmClient:          services.MgmClient,
 		relayManager:       services.RelayManager,
 		peerStore:          peerstore.NewConnStore(),
+		wgDetector:         stdnet.NewWGDetector(),
 		syncMsgMux:         &sync.Mutex{},
 		config:             config,
 		mobileDep:          mobileDep,
@@ -455,13 +458,6 @@ func (e *Engine) stopLocked() {
 
 	e.cleanupSSHConfig()
 
-	if e.ingressGatewayMgr != nil {
-		if err := e.ingressGatewayMgr.Close(); err != nil {
-			log.Warnf("failed to cleanup forward rules: %v", err)
-		}
-		e.ingressGatewayMgr = nil
-	}
-
 	if e.srWatcher != nil {
 		e.srWatcher.Close()
 	}
@@ -471,7 +467,7 @@ func (e *Engine) stopLocked() {
 	}
 
 	if e.updateManager != nil {
-		e.updateManager.SetDownloadOnly()
+		e.updateManager.ResetMode()
 	}
 
 	log.Info("cleaning up status recorder states")
@@ -676,10 +672,6 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, peerConfig *mgmPro
 	}
 	e.wgDevice.Store(e.wgInterface.GetWGDevice())
 
-	// Set up notrack rules immediately after proxy is listening to prevent
-	// conntrack entries from being created before the rules are in place
-	e.setupWGProxyNoTrack()
-
 	// Start after interface is up since port may have been resolved from 0 or changed if occupied
 	e.shutdownWg.Add(1)
 	go func() {
@@ -817,23 +809,6 @@ func (e *Engine) initFirewall() error {
 	return nil
 }
 
-// setupWGProxyNoTrack configures connection tracking exclusion for WireGuard proxy traffic.
-// This prevents conntrack/MASQUERADE from affecting loopback traffic between WireGuard and the eBPF proxy.
-func (e *Engine) setupWGProxyNoTrack() {
-	if e.firewall == nil {
-		return
-	}
-
-	proxyPort := e.wgInterface.GetProxyPort()
-	if proxyPort == 0 {
-		return
-	}
-
-	if err := e.firewall.SetupEBPFProxyNoTrack(proxyPort, uint16(e.config.WgPort)); err != nil {
-		log.Warnf("failed to setup ebpf proxy notrack: %v", err)
-	}
-}
-
 func (e *Engine) blockLanAccess() {
 	if e.config.BlockInbound {
 		// no need to set up extra deny rules if inbound is already blocked in general
@@ -996,11 +971,13 @@ func (e *Engine) handleAutoUpdateVersion(autoUpdateSettings *mgmProto.AutoUpdate
 	}
 
 	if autoUpdateSettings == nil {
+		log.Infof("no auto-update settings received, defaulting to download-only")
+		e.updateManager.SetDownloadOnly()
 		return
 	}
 
 	if autoUpdateSettings.Version == disableAutoUpdate {
-		log.Infof("auto-update is disabled")
+		log.Infof("auto-update is disabled, switching to download-only")
 		e.updateManager.SetDownloadOnly()
 		return
 	}
@@ -1698,13 +1675,6 @@ func (e *Engine) updateNetworkMap(networkMap *mgmProto.NetworkMap) error {
 	e.updateDNSForwarder(dnsRouteFeatureFlag, fwdEntries)
 	done()
 
-	// Ingress forward rules
-	done = e.phase("forward_rules")
-	if _, err := e.updateForwardRules(networkMap.GetForwardingRules()); err != nil {
-		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")
@@ -2249,10 +2219,7 @@ func (e *Engine) close() {
 }
 
 func (e *Engine) newWgIface() (*iface.WGIface, error) {
-	transportNet, err := e.newStdNet()
-	if err != nil {
-		log.Errorf("failed to create pion's stdnet: %s", err)
-	}
+	transportNet := e.newStdNet()
 
 	opts := iface.WGIFaceOpts{
 		IFaceName:    e.config.WgIfaceName,
@@ -2807,74 +2774,6 @@ func (e *Engine) setForwarderCapture(pc device.PacketCapture) {
 	}
 }
 
-func (e *Engine) updateForwardRules(rules []*mgmProto.ForwardingRule) ([]firewallManager.ForwardRule, error) {
-	if e.firewall == nil {
-		log.Warn("firewall is disabled, not updating forwarding rules")
-		return nil, nil
-	}
-
-	if len(rules) == 0 {
-		if e.ingressGatewayMgr == nil {
-			return nil, nil
-		}
-
-		err := e.ingressGatewayMgr.Close()
-		e.ingressGatewayMgr = nil
-		e.statusRecorder.SetIngressGwMgr(nil)
-		return nil, err
-	}
-
-	if e.ingressGatewayMgr == nil {
-		mgr := ingressgw.NewManager(e.firewall)
-		e.ingressGatewayMgr = mgr
-		e.statusRecorder.SetIngressGwMgr(mgr)
-	}
-
-	var merr *multierror.Error
-	forwardingRules := make([]firewallManager.ForwardRule, 0, len(rules))
-	for _, rule := range rules {
-		proto, err := acl.ConvertToFirewallProtocol(rule.GetProtocol())
-		if err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("failed to convert protocol '%s': %w", rule.GetProtocol(), err))
-			continue
-		}
-
-		dstPortInfo, err := convertPortInfo(rule.GetDestinationPort())
-		if err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("invalid destination port '%v': %w", rule.GetDestinationPort(), err))
-			continue
-		}
-
-		translateIP, err := convertToIP(rule.GetTranslatedAddress())
-		if err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("failed to convert translated address '%s': %w", rule.GetTranslatedAddress(), err))
-			continue
-		}
-
-		translatePort, err := convertPortInfo(rule.GetTranslatedPort())
-		if err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("invalid translate port '%v': %w", rule.GetTranslatedPort(), err))
-			continue
-		}
-
-		forwardRule := firewallManager.ForwardRule{
-			Protocol:          proto,
-			DestinationPort:   *dstPortInfo,
-			TranslatedAddress: translateIP,
-			TranslatedPort:    *translatePort,
-		}
-
-		forwardingRules = append(forwardingRules, forwardRule)
-	}
-
-	log.Infof("updating forwarding rules: %d", len(forwardingRules))
-	if err := e.ingressGatewayMgr.Update(forwardingRules); err != nil {
-		log.Errorf("failed to update forwarding rules: %v", err)
-	}
-
-	return forwardingRules, nberrors.FormatErrorOrNil(merr)
-}
-
 // toExcludedLazyPeers returns the peers that must have an always-active
 // connection: those that are not lazy by policy (the per-peer lazy state or the
 // account flag, subject to the local override).

client/internal/engine_generic.go

@@ -15,5 +15,6 @@ func (e *Engine) createICEConfig() icemaker.Config {
 		UDPMux:               e.udpMux.SingleSocketUDPMux,
 		UDPMuxSrflx:          e.udpMux,
 		NATExternalIPs:       e.parseNATExternalIPMappings(),
+		WGDetector:           e.wgDetector,
 	}
 }

client/internal/engine_js.go

@@ -13,6 +13,7 @@ func (e *Engine) createICEConfig() icemaker.Config {
 		InterfaceBlackList:   e.config.IFaceBlackList,
 		DisableIPv6Discovery: e.config.DisableIPv6Discovery,
 		NATExternalIPs:       e.parseNATExternalIPMappings(),
+		WGDetector:           e.wgDetector,
 	}
 	return cfg
 }

client/internal/engine_privileged_test.go

@@ -27,6 +27,7 @@ import (
 	"github.com/netbirdio/netbird/client/iface/wgaddr"
 	"github.com/netbirdio/netbird/client/internal/dns"
 	"github.com/netbirdio/netbird/client/internal/peer"
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
 	nbssh "github.com/netbirdio/netbird/client/ssh"
 	"github.com/netbirdio/netbird/client/system"
 	nbdns "github.com/netbirdio/netbird/dns"
@@ -41,7 +42,6 @@ import (
 	nbcache "github.com/netbirdio/netbird/management/server/cache"
 	"github.com/netbirdio/netbird/management/server/groups"
 	"github.com/netbirdio/netbird/management/server/integrations/integrated_validator/validator"
-	"github.com/netbirdio/netbird/management/server/integrations/port_forwarding"
 	"github.com/netbirdio/netbird/management/server/job"
 	"github.com/netbirdio/netbird/management/server/permissions"
 	"github.com/netbirdio/netbird/management/server/settings"
@@ -81,6 +81,7 @@ func TestEngine_SSH(t *testing.T) {
 			WgAddr:           wgaddr.MustParseWGAddress("100.64.0.1/24"),
 			WgPrivateKey:     key,
 			WgPort:           33100,
+			IFaceBlackList:   profilemanager.DefaultInterfaceBlacklist,
 			ServerSSHAllowed: true,
 			MTU:              iface.DefaultMTU,
 			SSHKey:           sshKey,
@@ -204,11 +205,12 @@ func TestEngine_Sync(t *testing.T) {
 	}
 	relayMgr := relayClient.NewManager(ctx, nil, key.PublicKey().String(), iface.DefaultMTU)
 	engine := NewEngine(ctx, cancel, &EngineConfig{
-		WgIfaceName:  "utun103",
-		WgAddr:       wgaddr.MustParseWGAddress("100.64.0.1/24"),
-		WgPrivateKey: key,
-		WgPort:       33100,
-		MTU:          iface.DefaultMTU,
+		WgIfaceName:    "utun103",
+		WgAddr:         wgaddr.MustParseWGAddress("100.64.0.1/24"),
+		WgPrivateKey:   key,
+		WgPort:         33100,
+		IFaceBlackList: profilemanager.DefaultInterfaceBlacklist,
+		MTU:            iface.DefaultMTU,
 	}, EngineServices{
 		SignalClient:   &signal.MockClient{},
 		MgmClient:      &mgmt.MockClient{SyncFunc: syncFunc},
@@ -412,11 +414,12 @@ func createEngine(ctx context.Context, cancel context.CancelFunc, setupKey strin
 
 	wgPort := 33100 + i
 	conf := &EngineConfig{
-		WgIfaceName:  ifaceName,
-		WgAddr:       wgaddr.MustParseWGAddress(resp.PeerConfig.Address),
-		WgPrivateKey: key,
-		WgPort:       wgPort,
-		MTU:          iface.DefaultMTU,
+		WgIfaceName:    ifaceName,
+		WgAddr:         wgaddr.MustParseWGAddress(resp.PeerConfig.Address),
+		WgPrivateKey:   key,
+		WgPort:         wgPort,
+		IFaceBlackList: profilemanager.DefaultInterfaceBlacklist,
+		MTU:            iface.DefaultMTU,
 	}
 
 	relayMgr := relayClient.NewManager(ctx, nil, key.PublicKey().String(), iface.DefaultMTU)
@@ -519,8 +522,8 @@ func startManagement(t *testing.T, dataDir, testFile string) (*grpc.Server, stri
 
 	updateManager := update_channel.NewPeersUpdateManager(metrics)
 	requestBuffer := server.NewAccountRequestBuffer(context.Background(), store)
-	networkMapController := controller.NewController(context.Background(), store, metrics, updateManager, requestBuffer, server.MockIntegratedValidator{}, settingsMockManager, "netbird.selfhosted", port_forwarding.NewControllerMock(), manager.NewEphemeralManager(store, peersManager), config, nil)
-	accountManager, err := server.BuildManager(context.Background(), config, store, networkMapController, jobManager, nil, "", eventStore, nil, false, ia, metrics, port_forwarding.NewControllerMock(), settingsMockManager, permissionsManager, false, cacheStore)
+	networkMapController := controller.NewController(context.Background(), store, metrics, updateManager, requestBuffer, server.MockIntegratedValidator{}, settingsMockManager, "netbird.selfhosted", manager.NewEphemeralManager(store, peersManager), config, nil)
+	accountManager, err := server.BuildManager(context.Background(), config, store, networkMapController, jobManager, nil, "", eventStore, nil, false, ia, metrics, settingsMockManager, permissionsManager, false, cacheStore)
 	if err != nil {
 		return nil, "", err
 	}

client/internal/engine_sessionwatch.go

@@ -1,4 +1,4 @@
-//go:build !js
+//go:build !js && !android
 
 package internal
 
@@ -7,10 +7,12 @@ import (
 	"github.com/netbirdio/netbird/client/internal/peer"
 )
 
-// newSessionWatcher returns the real SSO session expiry watcher for every
-// non-wasm build. The js/wasm build gets a no-op stub from
-// engine_sessionwatch_js.go so the sessionwatch package (and its timer
-// machinery) never links into the wasm binary.
+// newSessionWatcher returns the real SSO session expiry watcher. The js/wasm
+// build gets a no-op stub from engine_sessionwatch_js.go so the sessionwatch
+// package (and its timer machinery) never links into the wasm binary; the
+// android build gets a deadline-only watcher from
+// engine_sessionwatch_android.go because the app schedules the warnings
+// itself.
 func newSessionWatcher(recorder *peer.Status) sessionDeadlineWatcher {
 	return sessionwatch.New(recorder)
 }

client/internal/engine_sessionwatch_android.go

@@ -0,0 +1,12 @@
+//go:build android
+
+package internal
+
+import (
+	"github.com/netbirdio/netbird/client/internal/auth/sessionwatch"
+	"github.com/netbirdio/netbird/client/internal/peer"
+)
+
+func newSessionWatcher(recorder *peer.Status) sessionDeadlineWatcher {
+	return sessionwatch.NewDeadlineOnly(recorder)
+}

client/internal/engine_stdnet.go

@@ -6,6 +6,6 @@ import (
 	"github.com/netbirdio/netbird/client/internal/stdnet"
 )
 
-func (e *Engine) newStdNet() (*stdnet.Net, error) {
-	return stdnet.NewNet(e.clientCtx, e.config.IFaceBlackList)
+func (e *Engine) newStdNet() *stdnet.Net {
+	return stdnet.NewNet(e.clientCtx, e.config.IFaceBlackList, e.wgDetector)
 }

client/internal/engine_stdnet_android.go

@@ -2,6 +2,6 @@ package internal
 
 import "github.com/netbirdio/netbird/client/internal/stdnet"
 
-func (e *Engine) newStdNet() (*stdnet.Net, error) {
-	return stdnet.NewNetWithDiscover(e.clientCtx, e.mobileDep.IFaceDiscover, e.config.IFaceBlackList)
+func (e *Engine) newStdNet() *stdnet.Net {
+	return stdnet.NewNetWithDiscover(e.clientCtx, e.mobileDep.IFaceDiscover, e.config.IFaceBlackList, e.wgDetector)
 }

client/internal/engine_test.go

@@ -65,7 +65,6 @@ type MockWGIface struct {
 	GetStatsFunc               func() (map[string]configurer.WGStats, error)
 	GetInterfaceGUIDStringFunc func() (string, error)
 	GetProxyFunc               func() wgproxy.Proxy
-	GetProxyPortFunc           func() uint16
 	GetNetFunc                 func() *netstack.Net
 	LastActivitiesFunc         func() map[string]monotime.Time
 }
@@ -162,13 +161,6 @@ func (m *MockWGIface) GetProxy() wgproxy.Proxy {
 	return m.GetProxyFunc()
 }
 
-func (m *MockWGIface) GetProxyPort() uint16 {
-	if m.GetProxyPortFunc != nil {
-		return m.GetProxyPortFunc()
-	}
-	return 0
-}
-
 func (m *MockWGIface) GetNet() *netstack.Net {
 	return m.GetNetFunc()
 }
@@ -696,10 +688,7 @@ func TestEngine_UpdateNetworkMapWithRoutes(t *testing.T) {
 				StatusRecorder: peer.NewRecorder("https://mgm"),
 			}, MobileDependency{})
 			engine.ctx = ctx
-			newNet, err := stdnet.NewNet(context.Background(), nil)
-			if err != nil {
-				t.Fatal(err)
-			}
+			newNet := stdnet.NewNet(context.Background(), profilemanager.DefaultInterfaceBlacklist, nil)
 
 			opts := iface.WGIFaceOpts{
 				IFaceName:    wgIfaceName,
@@ -904,10 +893,7 @@ func TestEngine_UpdateNetworkMapWithDNSUpdate(t *testing.T) {
 			}, MobileDependency{})
 			engine.ctx = ctx
 
-			newNet, err := stdnet.NewNet(context.Background(), nil)
-			if err != nil {
-				t.Fatal(err)
-			}
+			newNet := stdnet.NewNet(context.Background(), profilemanager.DefaultInterfaceBlacklist, nil)
 			opts := iface.WGIFaceOpts{
 				IFaceName:    wgIfaceName,
 				Address:      wgaddr.MustParseWGAddress(wgAddr),
@@ -1507,3 +1493,21 @@ func TestOverlayAddrsFromAllowedIPs(t *testing.T) {
 		})
 	}
 }
+
+func TestEngine_SyncResponsePersistence(t *testing.T) {
+	e := &Engine{}
+
+	_, err := e.GetLatestSyncResponse()
+	require.Error(t, err, "persistence is disabled by default")
+
+	e.SetSyncResponsePersistence(true)
+	e.persistSyncResponse(&mgmtProto.SyncResponse{NetworkMap: &mgmtProto.NetworkMap{Serial: 7}})
+
+	got, err := e.GetLatestSyncResponse()
+	require.NoError(t, err)
+	assert.Equal(t, uint64(7), got.GetNetworkMap().GetSerial())
+
+	e.SetSyncResponsePersistence(false)
+	_, err = e.GetLatestSyncResponse()
+	require.Error(t, err)
+}

client/internal/iface_common.go

@@ -28,7 +28,6 @@ type wgIfaceBase interface {
 	Up() (*udpmux.UniversalUDPMuxDefault, error)
 	UpdateAddr(newAddr wgaddr.Address) error
 	GetProxy() wgproxy.Proxy
-	GetProxyPort() uint16
 	UpdatePeer(peerKey string, allowedIps []netip.Prefix, keepAlive time.Duration, endpoint *net.UDPAddr, preSharedKey *wgtypes.Key) error
 	RemoveEndpointAddress(key string) error
 	RemovePeer(peerKey string) error

client/internal/ingressgw/manager.go

@@ -1,111 +0,0 @@
-package ingressgw
-
-import (
-	"fmt"
-	"sync"
-
-	"github.com/hashicorp/go-multierror"
-	log "github.com/sirupsen/logrus"
-
-	nberrors "github.com/netbirdio/netbird/client/errors"
-	firewall "github.com/netbirdio/netbird/client/firewall/manager"
-)
-
-type DNATFirewall interface {
-	AddDNATRule(fwdRule firewall.ForwardRule) (firewall.Rule, error)
-	DeleteDNATRule(rule firewall.Rule) error
-}
-
-type RulePair struct {
-	firewall.ForwardRule
-	firewall.Rule
-}
-
-type Manager struct {
-	dnatFirewall DNATFirewall
-
-	rules   map[firewall.RuleID]RulePair
-	rulesMu sync.Mutex
-}
-
-func NewManager(dnatFirewall DNATFirewall) *Manager {
-	return &Manager{
-		dnatFirewall: dnatFirewall,
-		rules:        make(map[firewall.RuleID]RulePair),
-	}
-}
-
-func (h *Manager) Update(forwardRules []firewall.ForwardRule) error {
-	h.rulesMu.Lock()
-	defer h.rulesMu.Unlock()
-
-	var mErr *multierror.Error
-
-	toDelete := make(map[firewall.RuleID]RulePair, len(h.rules))
-	for id, r := range h.rules {
-		toDelete[id] = r
-	}
-
-	// Process new/updated rules
-	for _, fwdRule := range forwardRules {
-		id := fwdRule.ID()
-		if _, ok := h.rules[id]; ok {
-			delete(toDelete, id)
-			continue
-		}
-
-		rule, err := h.dnatFirewall.AddDNATRule(fwdRule)
-		if err != nil {
-			mErr = multierror.Append(mErr, fmt.Errorf("add forward rule '%s': %v", fwdRule.String(), err))
-			continue
-		}
-		if rule == nil {
-			mErr = multierror.Append(mErr, fmt.Errorf("add forward rule '%s': backend returned no rule", fwdRule.String()))
-			continue
-		}
-		log.Infof("forward rule has been added '%s'", fwdRule)
-		h.rules[id] = RulePair{
-			ForwardRule: fwdRule,
-			Rule:        rule,
-		}
-	}
-
-	// Remove deleted rules
-	for id, rulePair := range toDelete {
-		if err := h.dnatFirewall.DeleteDNATRule(rulePair.Rule); err != nil {
-			mErr = multierror.Append(mErr, fmt.Errorf("failed to delete forward rule '%s': %v", rulePair.ForwardRule.String(), err))
-		}
-		log.Infof("forward rule has been deleted '%s'", rulePair.ForwardRule)
-		delete(h.rules, id)
-	}
-
-	return nberrors.FormatErrorOrNil(mErr)
-}
-
-func (h *Manager) Close() error {
-	h.rulesMu.Lock()
-	defer h.rulesMu.Unlock()
-
-	log.Infof("clean up all (%d) forward rules", len(h.rules))
-	var mErr *multierror.Error
-	for _, rule := range h.rules {
-		if err := h.dnatFirewall.DeleteDNATRule(rule.Rule); err != nil {
-			mErr = multierror.Append(mErr, fmt.Errorf("failed to delete forward rule '%s': %v", rule, err))
-		}
-	}
-
-	h.rules = make(map[firewall.RuleID]RulePair)
-	return nberrors.FormatErrorOrNil(mErr)
-}
-
-func (h *Manager) Rules() []firewall.ForwardRule {
-	h.rulesMu.Lock()
-	defer h.rulesMu.Unlock()
-
-	rules := make([]firewall.ForwardRule, 0, len(h.rules))
-	for _, rulePair := range h.rules {
-		rules = append(rules, rulePair.ForwardRule)
-	}
-
-	return rules
-}

client/internal/ingressgw/manager_test.go

@@ -1,281 +0,0 @@
-package ingressgw
-
-import (
-	"fmt"
-	"net/netip"
-	"testing"
-
-	firewall "github.com/netbirdio/netbird/client/firewall/manager"
-)
-
-var (
-	_ firewall.Rule = (*MocFwRule)(nil)
-	_ DNATFirewall  = &MockDNATFirewall{}
-)
-
-type MocFwRule struct {
-	id firewall.RuleID
-}
-
-func (m *MocFwRule) ID() firewall.RuleID {
-	return m.id
-}
-
-type MockDNATFirewall struct {
-	throwError bool
-}
-
-func (m *MockDNATFirewall) AddDNATRule(fwdRule firewall.ForwardRule) (firewall.Rule, error) {
-	if m.throwError {
-		return nil, fmt.Errorf("moc error")
-	}
-
-	fwRule := &MocFwRule{
-		id: fwdRule.ID(),
-	}
-	return fwRule, nil
-}
-
-func (m *MockDNATFirewall) DeleteDNATRule(rule firewall.Rule) error {
-	if m.throwError {
-		return fmt.Errorf("moc error")
-	}
-	return nil
-}
-
-func (m *MockDNATFirewall) forceToThrowErrors() {
-	m.throwError = true
-}
-
-func TestManager_AddRule(t *testing.T) {
-	fw := &MockDNATFirewall{}
-	mgr := NewManager(fw)
-
-	port, _ := firewall.NewPort(8080)
-
-	updates := []firewall.ForwardRule{
-		{
-			Protocol:          firewall.ProtocolTCP,
-			DestinationPort:   *port,
-			TranslatedAddress: netip.MustParseAddr("172.16.254.1"),
-			TranslatedPort:    *port,
-		},
-		{
-			Protocol:          firewall.ProtocolUDP,
-			DestinationPort:   *port,
-			TranslatedAddress: netip.MustParseAddr("172.16.254.1"),
-			TranslatedPort:    *port,
-		}}
-
-	if err := mgr.Update(updates); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	rules := mgr.Rules()
-	if len(rules) != len(updates) {
-		t.Errorf("unexpected rules count: %d", len(rules))
-	}
-}
-
-func TestManager_UpdateRule(t *testing.T) {
-	fw := &MockDNATFirewall{}
-	mgr := NewManager(fw)
-
-	port, _ := firewall.NewPort(8080)
-	ruleTCP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolTCP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.1"),
-		TranslatedPort:    *port,
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleTCP}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	ruleUDP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolUDP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.2"),
-		TranslatedPort:    *port,
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleUDP}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	rules := mgr.Rules()
-	if len(rules) != 1 {
-		t.Errorf("unexpected rules count: %d", len(rules))
-	}
-
-	if rules[0].TranslatedAddress.String() != ruleUDP.TranslatedAddress.String() {
-		t.Errorf("unexpected rule: %v", rules[0])
-	}
-
-	if rules[0].TranslatedPort.String() != ruleUDP.TranslatedPort.String() {
-		t.Errorf("unexpected rule: %v", rules[0])
-	}
-
-	if rules[0].DestinationPort.String() != ruleUDP.DestinationPort.String() {
-		t.Errorf("unexpected rule: %v", rules[0])
-	}
-
-	if rules[0].Protocol != ruleUDP.Protocol {
-		t.Errorf("unexpected rule: %v", rules[0])
-	}
-}
-
-func TestManager_ExtendRules(t *testing.T) {
-	fw := &MockDNATFirewall{}
-	mgr := NewManager(fw)
-
-	port, _ := firewall.NewPort(8080)
-	ruleTCP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolTCP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.1"),
-		TranslatedPort:    *port,
-	}
-
-	ruleUDP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolUDP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.2"),
-		TranslatedPort:    *port,
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleTCP}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleTCP, ruleUDP}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	rules := mgr.Rules()
-	if len(rules) != 2 {
-		t.Errorf("unexpected rules count: %d", len(rules))
-	}
-}
-
-func TestManager_UnderlingError(t *testing.T) {
-	fw := &MockDNATFirewall{}
-	mgr := NewManager(fw)
-
-	port, _ := firewall.NewPort(8080)
-	ruleTCP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolTCP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.1"),
-		TranslatedPort:    *port,
-	}
-
-	ruleUDP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolUDP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.2"),
-		TranslatedPort:    *port,
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleTCP}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	fw.forceToThrowErrors()
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleTCP, ruleUDP}); err == nil {
-		t.Errorf("expected error")
-	}
-
-	rules := mgr.Rules()
-	if len(rules) != 1 {
-		t.Errorf("unexpected rules count: %d", len(rules))
-	}
-}
-
-func TestManager_Cleanup(t *testing.T) {
-	fw := &MockDNATFirewall{}
-	mgr := NewManager(fw)
-
-	port, _ := firewall.NewPort(8080)
-	ruleTCP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolTCP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.1"),
-		TranslatedPort:    *port,
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleTCP}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	rules := mgr.Rules()
-	if len(rules) != 0 {
-		t.Errorf("unexpected rules count: %d", len(rules))
-	}
-}
-
-func TestManager_DeleteBrokenRule(t *testing.T) {
-	fw := &MockDNATFirewall{}
-
-	// force to throw errors when Add DNAT Rule
-	fw.forceToThrowErrors()
-	mgr := NewManager(fw)
-
-	port, _ := firewall.NewPort(8080)
-	ruleTCP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolTCP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.1"),
-		TranslatedPort:    *port,
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleTCP}); err == nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	rules := mgr.Rules()
-	if len(rules) != 0 {
-		t.Errorf("unexpected rules count: %d", len(rules))
-	}
-
-	// simulate that to remove a broken rule
-	if err := mgr.Update([]firewall.ForwardRule{}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	if err := mgr.Close(); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-}
-
-func TestManager_Close(t *testing.T) {
-	fw := &MockDNATFirewall{}
-	mgr := NewManager(fw)
-
-	port, _ := firewall.NewPort(8080)
-	ruleTCP := firewall.ForwardRule{
-		Protocol:          firewall.ProtocolTCP,
-		DestinationPort:   *port,
-		TranslatedAddress: netip.MustParseAddr("172.16.254.1"),
-		TranslatedPort:    *port,
-	}
-
-	if err := mgr.Update([]firewall.ForwardRule{ruleTCP}); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	if err := mgr.Close(); err != nil {
-		t.Errorf("unexpected error: %v", err)
-	}
-
-	rules := mgr.Rules()
-	if len(rules) != 0 {
-		t.Errorf("unexpected rules count: %d", len(rules))
-	}
-}

client/internal/ipcauth/forward_test.go

@@ -38,7 +38,7 @@ func asDaemon(t *testing.T, id Identity) {
 	prevID, prevKnown, prevDelegate := selfIdentity, selfKnown, selfMayDelegate
 	t.Cleanup(func() { selfIdentity, selfKnown, selfMayDelegate = prevID, prevKnown, prevDelegate })
 	selfIdentity, selfKnown = id, true
-	selfMayDelegate = !id.IsPrivileged()
+	selfMayDelegate = mayDelegate(id)
 }
 
 func TestCallerIdentity_DirectConnections(t *testing.T) {

client/internal/ipcauth/identity.go

@@ -18,7 +18,8 @@ import (
 	"google.golang.org/grpc/peer"
 )
 
-// Well-known Windows SIDs that identify a fully privileged principal.
+// Well-known Windows SIDs. Only LocalSystem and BUILTIN\Administrators identify a
+// privileged principal; the service accounts are shared by unrelated services.
 const (
 	sidLocalSystem    = "S-1-5-18"     // NT AUTHORITY\SYSTEM
 	sidLocalService   = "S-1-5-19"     // NT AUTHORITY\LOCAL SERVICE
@@ -67,9 +68,9 @@ func (i Identity) IsWindows() bool {
 // user-to-root boundary.
 //
 // On Windows the decision comes from the caller's token rather than from
-// account names or group RIDs: an elevated token, one of the service accounts
-// the daemon itself may run as, or a token with BUILTIN\Administrators
-// enabled. A UAC-filtered administrator has that group marked deny-only, and
+// account names or group RIDs: an elevated token, the LocalSystem SID, or a
+// token with BUILTIN\Administrators enabled. LocalService and NetworkService
+// are not privileged by SID. A UAC-filtered administrator has that group marked deny-only, and
 // deny-only groups are dropped when the identity is captured, so such a
 // caller is correctly reported as unprivileged. Domain group memberships
 // (Domain Admins and friends) are deliberately not consulted: they say
@@ -83,8 +84,7 @@ func (i Identity) IsPrivileged() bool {
 		return true
 	}
 
-	switch i.SID {
-	case sidLocalSystem, sidLocalService, sidNetworkService:
+	if i.SID == sidLocalSystem {
 		return true
 	}
 

client/internal/ipcauth/identity_test.go

@@ -64,3 +64,58 @@ func TestIdentitySameUser(t *testing.T) {
 		})
 	}
 }
+
+func TestIdentityIsPrivileged(t *testing.T) {
+	tests := []struct {
+		name string
+		id   Identity
+		want bool
+	}{
+		{
+			name: "Root",
+			id:   Identity{UID: 0, GID: 0},
+			want: true,
+		},
+		{
+			name: "Non-root",
+			id:   Identity{UID: 1000, GID: 1000},
+			want: false,
+		},
+		{
+			name: "Local system windows",
+			id:   Identity{SID: sidLocalSystem},
+			want: true,
+		},
+		{
+			name: "Windows elevated",
+			id:   Identity{SID: "S-1-5-21-1927267129-3959769253-3036563910-1001", Elevated: true},
+			want: true,
+		},
+		{
+			name: "Admin group windows",
+			id:   Identity{SID: "S-1-5-21-1927267129-3959769253-3036563910-1001", Groups: []string{sidAdministrators}},
+			want: true,
+		},
+		{
+			name: "Regular user windows",
+			id:   Identity{SID: "S-1-5-21-1927267129-3959769253-3036563910-1001"},
+			want: false,
+		},
+		{
+			name: "Network service windows",
+			id:   Identity{SID: sidNetworkService},
+			want: false,
+		},
+		{
+			name: "Local service windows",
+			id:   Identity{SID: sidLocalService},
+			want: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.want, tt.id.IsPrivileged())
+		})
+	}
+}

client/internal/ipcauth/privileged.go

@@ -45,7 +45,15 @@ func init() {
 	// matching there would let a non-elevated shell of an administrator account
 	// act as an administrator, which is the boundary the token check exists to
 	// keep.
-	selfMayDelegate = !id.IsPrivileged()
+	selfMayDelegate = mayDelegate(id)
+}
+
+// mayDelegate reports whether a daemon running as id may extend its authority to
+// callers sharing its identity. The shared service accounts are excluded: their
+// SID is held by unrelated services, so matching on it would grant them the
+// daemon's authority.
+func mayDelegate(id Identity) bool {
+	return !id.IsPrivileged() && id.SID != sidLocalService && id.SID != sidNetworkService
 }
 
 // IsDaemonSelf reports whether an identity is this very process. The JSON gateway

client/internal/ipcauth/privileged_test.go

@@ -98,7 +98,7 @@ func TestIsPrivilegedCaller_SelfRule(t *testing.T) {
 			t.Cleanup(func() { selfIdentity, selfKnown, selfMayDelegate = prevID, prevKnown, prevDelegate })
 
 			selfIdentity, selfKnown = tt.self, tt.selfKnown
-			selfMayDelegate = tt.selfKnown && !tt.self.IsPrivileged()
+			selfMayDelegate = tt.selfKnown && mayDelegate(tt.self)
 
 			if got := IsPrivilegedCaller(tt.caller); got != tt.want {
 				t.Fatalf("IsPrivilegedCaller(%v) with daemon %v = %t, want %t",
@@ -132,3 +132,28 @@ func TestIsPrivilegedCaller_ThisProcess(t *testing.T) {
 		t.Errorf("an unrelated identity %v was treated as privileged", other)
 	}
 }
+
+// The shared service accounts are held by unrelated services, so a daemon running
+// as one of them must not extend its authority to every process with that SID.
+func TestMayDelegate(t *testing.T) {
+	tests := []struct {
+		name string
+		self Identity
+		want bool
+	}{
+		{name: "unprivileged unix user", self: Identity{UID: 1000}, want: true},
+		{name: "root", self: Identity{UID: 0}, want: false},
+		{name: "unprivileged windows user", self: Identity{SID: "S-1-5-21-1-2-3-1001"}, want: true},
+		{name: "elevated windows user", self: Identity{SID: "S-1-5-21-1-2-3-1001", Elevated: true}, want: false},
+		{name: "local system", self: Identity{SID: sidLocalSystem}, want: false},
+		{name: "local service", self: Identity{SID: sidLocalService}, want: false},
+		{name: "network service", self: Identity{SID: sidNetworkService}, want: false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := mayDelegate(tt.self); got != tt.want {
+				t.Errorf("mayDelegate(%+v) = %v, want %v", tt.self, got, tt.want)
+			}
+		})
+	}
+}