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/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)
+			}
+		})
+	}
+}

client/internal/message_convert.go

@@ -1,43 +0,0 @@
-package internal
-
-import (
-	"errors"
-	"fmt"
-	"net"
-	"net/netip"
-
-	firewallManager "github.com/netbirdio/netbird/client/firewall/manager"
-	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
-)
-
-func convertPortInfo(portInfo *mgmProto.PortInfo) (*firewallManager.Port, error) {
-	if portInfo == nil {
-		return nil, errors.New("portInfo cannot be nil")
-	}
-
-	if portInfo.GetPort() != 0 {
-		return firewallManager.NewPort(int(portInfo.GetPort()))
-	}
-
-	if portInfo.GetRange() != nil {
-		return firewallManager.NewPort(int(portInfo.GetRange().Start), int(portInfo.GetRange().End))
-	}
-
-	return nil, fmt.Errorf("invalid portInfo: %v", portInfo)
-}
-
-func convertToIP(rawIP []byte) (netip.Addr, error) {
-	if rawIP == nil {
-		return netip.Addr{}, errors.New("input bytes cannot be nil")
-	}
-
-	if len(rawIP) != net.IPv4len && len(rawIP) != net.IPv6len {
-		return netip.Addr{}, fmt.Errorf("invalid IP length: %d", len(rawIP))
-	}
-
-	if len(rawIP) == net.IPv4len {
-		return netip.AddrFrom4([4]byte(rawIP)), nil
-	}
-
-	return netip.AddrFrom16([16]byte(rawIP)), nil
-}

client/internal/peer/conn.go

@@ -828,7 +828,8 @@ func (conn *Conn) evalStatus() ConnStatus {
 //
 // The result is a tri-state:
 //   - ConnStatusConnected:          all available transports are up
-//   - ConnStatusPartiallyConnected: relay is up but ICE is still pending/reconnecting
+//   - ConnStatusPartiallyConnected: one transport carries the traffic and the other does
+//     not: relay up with ICE down, or ICE up with the shared relay transport down
 //   - ConnStatusDisconnected:       no working transport
 func (conn *Conn) isConnectedOnAllWay() (status guard.ConnStatus) {
 	defer func() {
@@ -845,13 +846,14 @@ func (conn *Conn) isConnectedOnAllWay() (status guard.ConnStatus) {
 	}
 
 	return evalConnStatus(connStatusInputs{
-		forceRelay:          IsForceRelayed(),
-		peerUsesRelay:       conn.workerRelay.IsRelayConnectionSupportedWithPeer(),
-		relayConnected:      conn.statusRelay.Get() == worker.StatusConnected,
-		remoteSupportsICE:   conn.handshaker.RemoteICESupported(),
-		iceWorkerCreated:    iceWorkerCreated,
-		iceStatusConnecting: conn.statusICE.Get() != worker.StatusDisconnected,
-		iceInProgress:       iceInProgress,
+		forceRelay:              IsForceRelayed(),
+		peerUsesRelay:           conn.workerRelay.IsRelayConnectionSupportedWithPeer(),
+		relayConnected:          conn.statusRelay.Get() == worker.StatusConnected,
+		relayTransportConnected: conn.workerRelay.IsTransportConnected(),
+		remoteSupportsICE:       conn.handshaker.RemoteICESupported(),
+		iceWorkerCreated:        iceWorkerCreated,
+		iceStatusConnected:      conn.statusICE.Get() == worker.StatusConnected,
+		iceInProgress:           iceInProgress,
 	})
 }
 
@@ -1060,19 +1062,21 @@ func evalConnStatus(in connStatusInputs) guard.ConnStatus {
 		return boolToConnStatus(relayUsedAndUp)
 	}
 
-	// ICE counts as "up" when the status is anything other than Disconnected, OR
-	// when a negotiation is currently in progress (so we don't spam offers while one is in flight).
-	iceUp := in.iceStatusConnecting || in.iceInProgress
+	// ICE counts as "running" when either connected or attempting to connect.
+	iceRunning := in.iceStatusConnected || in.iceInProgress
 
 	// Relay side is acceptable if the peer doesn't rely on relay, or relay is connected.
 	relayOK := !in.peerUsesRelay || in.relayConnected
 
 	switch {
-	case iceUp && relayOK:
+	case iceRunning && relayOK:
 		return guard.ConnStatusConnected
 	case relayUsedAndUp:
 		// Relay is up but ICE is down — partially connected.
 		return guard.ConnStatusPartiallyConnected
+	case in.iceStatusConnected && !in.relayTransportConnected:
+		// ICE is up and the shared relay transport is down — offers cannot restore it.
+		return guard.ConnStatusPartiallyConnected
 	default:
 		return guard.ConnStatusDisconnected
 	}

client/internal/peer/conn_status.go

@@ -17,13 +17,14 @@ const (
 // tri-state connection classification. Extracted so the decision logic can be unit-tested
 // without constructing full Worker/Handshaker objects.
 type connStatusInputs struct {
-	forceRelay          bool // NB_FORCE_RELAY or JS/WASM
-	peerUsesRelay       bool // remote peer advertises relay support AND local has relay
-	relayConnected      bool // statusRelay reports Connected (independent of whether peer uses relay)
-	remoteSupportsICE   bool // remote peer sent ICE credentials
-	iceWorkerCreated    bool // local WorkerICE exists (false in force-relay mode)
-	iceStatusConnecting bool // statusICE is anything other than Disconnected
-	iceInProgress       bool // a negotiation is currently in flight
+	forceRelay              bool // NB_FORCE_RELAY or JS/WASM
+	peerUsesRelay           bool // remote peer advertises relay support AND local has relay
+	relayConnected          bool // statusRelay reports Connected (independent of whether peer uses relay)
+	relayTransportConnected bool // the relay transport shared by all peers on that server is up
+	remoteSupportsICE       bool // remote peer sent ICE credentials
+	iceWorkerCreated        bool // local WorkerICE exists (false in force-relay mode)
+	iceStatusConnected      bool // statusICE reports Connected
+	iceInProgress           bool // a negotiation is currently in flight
 }
 
 // ConnStatus describe the status of a peer's connection

client/internal/peer/conn_status_eval_test.go

@@ -30,6 +30,21 @@ func TestEvalConnStatus_ForceRelay(t *testing.T) {
 			},
 			want: guard.ConnStatusDisconnected,
 		},
+		{
+			name: "force relay, relay up but the shared transport reports down",
+			in: connStatusInputs{
+				forceRelay:              true,
+				peerUsesRelay:           true,
+				relayConnected:          true,
+				relayTransportConnected: false,
+				// The ICE inputs are set so that the force-relay return is the only branch
+				// that can produce Connected here: without it the peer would fall through to
+				// relayUsedAndUp and report PartiallyConnected.
+				remoteSupportsICE: true,
+				iceWorkerCreated:  true,
+			},
+			want: guard.ConnStatusConnected,
+		},
 		{
 			name: "force relay, peer does NOT use relay - disconnected forever",
 			in: connStatusInputs{
@@ -123,24 +138,28 @@ func TestEvalConnStatus_FullyAvailable(t *testing.T) {
 			mutator: func(in *connStatusInputs) {
 				in.peerUsesRelay = true
 				in.relayConnected = true
-				in.iceStatusConnecting = true
+				in.relayTransportConnected = true
+				in.iceStatusConnected = true
 			},
 			want: guard.ConnStatusConnected,
 		},
 		{
-			name: "ICE connected, peer does NOT use relay",
+			name: "ICE connected, peer does NOT use relay, shared transport down",
 			mutator: func(in *connStatusInputs) {
 				in.peerUsesRelay = false
 				in.relayConnected = false
-				in.iceStatusConnecting = true
+				in.relayTransportConnected = false
+				in.iceStatusConnected = true
 			},
+			// A peer that does not rely on relay is unaffected by the shared transport:
+			// relayOK is true, so the first arm matches before the transport is considered.
 			want: guard.ConnStatusConnected,
 		},
 		{
 			name: "ICE InProgress only, peer does NOT use relay",
 			mutator: func(in *connStatusInputs) {
 				in.peerUsesRelay = false
-				in.iceStatusConnecting = false
+				in.iceStatusConnected = false
 				in.iceInProgress = true
 			},
 			want: guard.ConnStatusConnected,
@@ -150,7 +169,8 @@ func TestEvalConnStatus_FullyAvailable(t *testing.T) {
 			mutator: func(in *connStatusInputs) {
 				in.peerUsesRelay = true
 				in.relayConnected = true
-				in.iceStatusConnecting = false
+				in.relayTransportConnected = true
+				in.iceStatusConnected = false
 				in.iceInProgress = false
 			},
 			want: guard.ConnStatusPartiallyConnected,
@@ -160,21 +180,60 @@ func TestEvalConnStatus_FullyAvailable(t *testing.T) {
 			mutator: func(in *connStatusInputs) {
 				in.peerUsesRelay = false
 				in.relayConnected = false
-				in.iceStatusConnecting = false
+				in.iceStatusConnected = false
 				in.iceInProgress = false
 			},
 			want: guard.ConnStatusDisconnected,
 		},
 		{
-			name: "ICE up, peer uses relay but relay down -> partial (relay required, ICE ignored)",
+			name: "ICE connected, relay down for this peer but the shared transport is up -> disconnected",
 			mutator: func(in *connStatusInputs) {
 				in.peerUsesRelay = true
 				in.relayConnected = false
-				in.iceStatusConnecting = true
+				in.relayTransportConnected = true
+				in.iceStatusConnected = true
+			},
+			// The transport is fine, so the peer itself is unreachable over relay: it may have
+			// moved to another server, and only an offer carries its new relay address.
+			want: guard.ConnStatusDisconnected,
+		},
+		{
+			name: "ICE connected, the shared relay transport is down -> partial",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = true
+				in.relayConnected = false
+				in.relayTransportConnected = false
+				in.iceStatusConnected = true
+			},
+			// ICE carries the traffic and the relay transport is restored by the relay client's
+			// own guard, not by offers, so this must not trigger the aggressive retry.
+			want: guard.ConnStatusPartiallyConnected,
+		},
+		{
+			name: "ICE only negotiating while the shared relay transport is down -> disconnected",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = true
+				in.relayConnected = false
+				in.relayTransportConnected = false
+				in.iceStatusConnected = false
+				in.iceInProgress = true
+			},
+			// A negotiation in flight is not a working transport, so this peer has no path at
+			// all and must keep the aggressive retry. Calling it partially connected spends the
+			// ICE retry budget and parks the guard on the hourly ticker, and nothing wakes it
+			// when the negotiation then fails: onICEStateDisconnected is only reached once ICE
+			// has reached Connected (worker_ice.go onConnectionStateChange).
+			want: guard.ConnStatusDisconnected,
+		},
+		{
+			name: "ICE down and the shared relay transport is down -> disconnected",
+			mutator: func(in *connStatusInputs) {
+				in.peerUsesRelay = true
+				in.relayConnected = false
+				in.relayTransportConnected = false
+				in.iceStatusConnected = false
+				in.iceInProgress = false
 			},
-			// relayOK = false (peer uses relay but it's down), iceUp = true
-			// first switch arm fails (relayOK false), relayUsedAndUp = false (relay down),
-			// falls into default: Disconnected.
 			want: guard.ConnStatusDisconnected,
 		},
 		{
@@ -182,7 +241,7 @@ func TestEvalConnStatus_FullyAvailable(t *testing.T) {
 			mutator: func(in *connStatusInputs) {
 				in.peerUsesRelay = false
 				in.relayConnected = true // not actually used since peer doesn't rely on it
-				in.iceStatusConnecting = false
+				in.iceStatusConnected = false
 				in.iceInProgress = false
 			},
 			want: guard.ConnStatusDisconnected,

client/internal/peer/conn_test.go

@@ -42,7 +42,7 @@ func TestNewConn_interfaceFilter(t *testing.T) {
 	ignore := []string{iface.WgInterfaceDefault, "tun0", "zt", "ZeroTier", "utun", "wg", "ts",
 		"Tailscale", "tailscale"}
 
-	filter := stdnet.InterfaceFilter(ignore)
+	filter := stdnet.InterfaceFilter(ignore, nil)
 
 	for _, s := range ignore {
 		assert.Equal(t, filter(s), false)

client/internal/peer/guard/guard.go

@@ -14,7 +14,8 @@ type ConnStatus int
 const (
 	// ConnStatusDisconnected means neither ICE nor Relay is connected.
 	ConnStatusDisconnected ConnStatus = iota
-	// ConnStatusPartiallyConnected means Relay is connected but ICE is not.
+	// ConnStatusPartiallyConnected means one transport is usable and the other is not:
+	// relay connected with ICE down, or ICE connected with the shared relay transport down.
 	ConnStatusPartiallyConnected
 	// ConnStatusConnected means all required connections are established.
 	ConnStatusConnected
@@ -87,8 +88,9 @@ func (g *Guard) SetICEConnDisconnected() {
 //   - Connected: no action, the peer is fully reachable.
 //   - Disconnected (neither ICE nor Relay): retries aggressively with exponential backoff (800ms doubling
 //     up to timeout), never gives up. This ensures rapid recovery when the peer has no connectivity at all.
-//   - PartiallyConnected (Relay up, ICE not): retries up to 3 times with exponential backoff, then switches
-//     to one attempt per hour. This limits signaling traffic when relay already provides connectivity.
+//   - PartiallyConnected (one transport usable, the other not): retries up to 3 times
+//     with exponential backoff, then switches to one attempt per hour. This limits
+//     signaling traffic while the peer still has a working path.
 //
 // External events (relay/ICE disconnect, signal/relay reconnect, candidate changes) reset the retry
 // counter and backoff ticker, giving ICE a fresh chance after network conditions change.

client/internal/peer/ice/agent.go

@@ -39,10 +39,7 @@ func NewAgent(ctx context.Context, iFaceDiscover stdnet.ExternalIFaceDiscover, c
 	iceFailedTimeout := iceFailedTimeout()
 	iceRelayAcceptanceMinWait := iceRelayAcceptanceMinWait()
 
-	transportNet, err := newStdNet(ctx, iFaceDiscover, config.InterfaceBlackList)
-	if err != nil {
-		log.Errorf("failed to create pion's stdnet: %s", err)
-	}
+	transportNet := newStdNet(ctx, iFaceDiscover, config.InterfaceBlackList, config.WGDetector)
 
 	fac := logging.NewDefaultLoggerFactory()
 
@@ -53,7 +50,7 @@ func NewAgent(ctx context.Context, iFaceDiscover stdnet.ExternalIFaceDiscover, c
 		NetworkTypes:           []ice.NetworkType{ice.NetworkTypeUDP4, ice.NetworkTypeUDP6},
 		Urls:                   config.StunTurn.Load(),
 		CandidateTypes:         candidateTypes,
-		InterfaceFilter:        stdnet.InterfaceFilter(config.InterfaceBlackList),
+		InterfaceFilter:        stdnet.InterfaceFilter(config.InterfaceBlackList, config.WGDetector),
 		UDPMux:                 config.UDPMux,
 		UDPMuxSrflx:            config.UDPMuxSrflx,
 		NAT1To1IPs:             config.NATExternalIPs,

client/internal/peer/ice/config.go

@@ -2,6 +2,8 @@ package ice
 
 import (
 	"github.com/pion/ice/v4"
+
+	"github.com/netbirdio/netbird/client/internal/stdnet"
 )
 
 type Config struct {
@@ -17,4 +19,8 @@ type Config struct {
 	UDPMuxSrflx ice.UniversalUDPMux
 
 	NATExternalIPs []string
+
+	// WGDetector is shared by every agent so that the WireGuard check the interface
+	// filter performs is not repeated for each of them.
+	WGDetector *stdnet.WGDetector
 }

client/internal/peer/ice/stdnet.go

@@ -8,6 +8,6 @@ import (
 	"github.com/netbirdio/netbird/client/internal/stdnet"
 )
 
-func newStdNet(ctx context.Context, _ stdnet.ExternalIFaceDiscover, ifaceBlacklist []string) (*stdnet.Net, error) {
-	return stdnet.NewNet(ctx, ifaceBlacklist)
+func newStdNet(ctx context.Context, _ stdnet.ExternalIFaceDiscover, ifaceBlacklist []string, detector *stdnet.WGDetector) *stdnet.Net {
+	return stdnet.NewNet(ctx, ifaceBlacklist, detector)
 }

client/internal/peer/ice/stdnet_android.go

@@ -6,6 +6,6 @@ import (
 	"github.com/netbirdio/netbird/client/internal/stdnet"
 )
 
-func newStdNet(ctx context.Context, iFaceDiscover stdnet.ExternalIFaceDiscover, ifaceBlacklist []string) (*stdnet.Net, error) {
-	return stdnet.NewNetWithDiscover(ctx, iFaceDiscover, ifaceBlacklist)
+func newStdNet(ctx context.Context, iFaceDiscover stdnet.ExternalIFaceDiscover, ifaceBlacklist []string, detector *stdnet.WGDetector) *stdnet.Net {
+	return stdnet.NewNetWithDiscover(ctx, iFaceDiscover, ifaceBlacklist, detector)
 }

client/internal/peer/notifier.go

@@ -12,6 +12,8 @@ type notifier struct {
 	serverStateLock    sync.Mutex
 	listenersLock      sync.Mutex
 	listener           Listener
+	peerListWake       chan struct{}
+	peerListStop       chan struct{}
 	currentClientState bool
 	lastNotification   ClientState
 	lastNumberOfPeers  int
@@ -62,7 +64,6 @@ func (n *notifier) setNetworkAvailable(available bool) {
 func (n *notifier) setListener(listener Listener) {
 	n.serverStateLock.Lock()
 	lastNotification := n.effectiveState(n.lastNotification)
-	numOfPeers := n.lastNumberOfPeers
 	fqdnAddress := n.lastFqdnAddress
 	address := n.lastIPAddress
 	n.serverStateLock.Unlock()
@@ -70,17 +71,19 @@ func (n *notifier) setListener(listener Listener) {
 	n.listenersLock.Lock()
 	defer n.listenersLock.Unlock()
 
+	n.stopPeerListDelivererLocked()
 	n.listener = listener
 
 	listener.OnAddressChanged(fqdnAddress, address)
 	notifyListener(listener, lastNotification)
-	// run on go routine to avoid on Java layer to call go functions on same thread
-	go listener.OnPeersListChanged(numOfPeers)
+	n.startPeerListDelivererLocked(listener)
+	n.wakePeerListDelivererLocked()
 }
 
 func (n *notifier) removeListener() {
 	n.listenersLock.Lock()
 	defer n.listenersLock.Unlock()
+	n.stopPeerListDelivererLocked()
 	n.listener = nil
 }
 
@@ -178,15 +181,56 @@ func (n *notifier) peerListChanged(numOfPeers int) {
 	n.serverStateLock.Unlock()
 
 	n.listenersLock.Lock()
-	listener := n.listener
-	n.listenersLock.Unlock()
+	defer n.listenersLock.Unlock()
+	n.wakePeerListDelivererLocked()
+}
 
-	if listener == nil {
+func (n *notifier) startPeerListDelivererLocked(listener Listener) {
+	wake := make(chan struct{}, 1)
+	stop := make(chan struct{})
+	n.peerListWake = wake
+	n.peerListStop = stop
+	go n.deliverPeerListChanges(listener, wake, stop)
+}
+
+func (n *notifier) stopPeerListDelivererLocked() {
+	if n.peerListStop == nil {
 		return
 	}
+	close(n.peerListStop)
+	n.peerListStop = nil
+	n.peerListWake = nil
+}
 
-	// run on go routine to avoid on Java layer to call go functions on same thread
-	go listener.OnPeersListChanged(numOfPeers)
+func (n *notifier) wakePeerListDelivererLocked() {
+	if n.peerListWake == nil {
+		return
+	}
+	select {
+	case n.peerListWake <- struct{}{}:
+	default:
+	}
+}
+
+func (n *notifier) deliverPeerListChanges(listener Listener, wake <-chan struct{}, stop <-chan struct{}) {
+	for {
+		select {
+		case <-stop:
+			return
+		case <-wake:
+		}
+		select {
+		case <-stop:
+			return
+		default:
+		}
+
+		n.serverStateLock.Lock()
+		numOfPeers := n.lastNumberOfPeers
+		n.serverStateLock.Unlock()
+
+		listener.OnPeersListChanged(numOfPeers)
+	}
 }
 
 func (n *notifier) localAddressChanged(fqdn, address string) {

client/internal/peer/notifier_test.go

@@ -2,7 +2,9 @@ package peer
 
 import (
 	"sync"
+	"sync/atomic"
 	"testing"
+	"time"
 )
 
 type mocListener struct {
@@ -115,3 +117,156 @@ func Test_notifier_RemoveListener(t *testing.T) {
 		t.Errorf("invalid state: %d", listener.peers)
 	}
 }
+
+type coalescingListener struct {
+	final       int
+	calls       atomic.Int32
+	inFlight    atomic.Int32
+	maxInFlight atomic.Int32
+	last        atomic.Int32
+	done        chan struct{}
+	entered     chan struct{}
+	release     chan struct{}
+	once        sync.Once
+}
+
+func (l *coalescingListener) OnStateChanged(ClientState)      {}
+func (l *coalescingListener) OnConnected()                    {}
+func (l *coalescingListener) OnDisconnected()                 {}
+func (l *coalescingListener) OnConnecting()                   {}
+func (l *coalescingListener) OnDisconnecting()                {}
+func (l *coalescingListener) OnAddressChanged(string, string) {}
+
+func (l *coalescingListener) OnPeersListChanged(size int) {
+	current := l.inFlight.Add(1)
+	for {
+		seen := l.maxInFlight.Load()
+		if current <= seen || l.maxInFlight.CompareAndSwap(seen, current) {
+			break
+		}
+	}
+	if l.calls.Add(1) == 1 && l.entered != nil {
+		close(l.entered)
+	}
+	if l.release != nil {
+		<-l.release
+	}
+	time.Sleep(time.Millisecond)
+	l.last.Store(int32(size))
+	l.inFlight.Add(-1)
+	if size == l.final {
+		l.once.Do(func() { close(l.done) })
+	}
+}
+
+func Test_notifier_PeerListChangedCoalesces(t *testing.T) {
+	const events = 1000
+	listener := &coalescingListener{final: events, done: make(chan struct{})}
+	n := newNotifier()
+	n.setListener(listener)
+
+	for i := 1; i <= events; i++ {
+		n.peerListChanged(i)
+	}
+
+	select {
+	case <-listener.done:
+	case <-time.After(5 * time.Second):
+		t.Fatalf("last peer count not delivered, last seen: %d", listener.last.Load())
+	}
+
+	if got := listener.maxInFlight.Load(); got != 1 {
+		t.Errorf("concurrent deliveries: %d, expected 1", got)
+	}
+	if got := listener.calls.Load(); got >= events {
+		t.Errorf("deliveries not coalesced: %d calls for %d events", got, events)
+	}
+}
+
+func Test_notifier_SetListenerStopsPreviousDeliverer(t *testing.T) {
+	old := &coalescingListener{final: -1}
+	replacement := &coalescingListener{final: 7, done: make(chan struct{})}
+	n := newNotifier()
+	n.setListener(old)
+	oldStop := n.peerListStop
+
+	n.peerListChanged(7)
+	n.setListener(replacement)
+
+	select {
+	case <-oldStop:
+	default:
+		t.Fatal("old deliverer not stopped on listener replacement")
+	}
+	waitFor(t, replacement.done, "replacement listener not notified")
+}
+
+func Test_notifier_RemoveListenerStopsDeliverer(t *testing.T) {
+	n := newNotifier()
+	n.setListener(&coalescingListener{final: -1})
+	stop := n.peerListStop
+
+	n.removeListener()
+
+	select {
+	case <-stop:
+	default:
+		t.Fatal("deliverer not stopped on listener removal")
+	}
+}
+
+func Test_notifier_DelivererExitsAfterInFlightCallback(t *testing.T) {
+	listener := &coalescingListener{
+		final:   -1,
+		entered: make(chan struct{}),
+		release: make(chan struct{}),
+	}
+	n := newNotifier()
+	wake := make(chan struct{}, 1)
+	stop := make(chan struct{})
+	exited := make(chan struct{})
+	go func() {
+		n.deliverPeerListChanges(listener, wake, stop)
+		close(exited)
+	}()
+
+	wake <- struct{}{}
+	waitFor(t, listener.entered, "listener not called")
+
+	n.peerListChanged(7)
+	wake <- struct{}{}
+	close(stop)
+	close(listener.release)
+
+	waitFor(t, exited, "deliverer did not exit after stop")
+	if got := listener.calls.Load(); got != 1 {
+		t.Errorf("deliverer ran %d callbacks after stop, expected only the in-flight one", got)
+	}
+	if got := listener.last.Load(); got == 7 {
+		t.Errorf("deliverer delivered the peer count queued after stop")
+	}
+}
+
+func Test_notifier_DelivererPrefersStopOverPendingWake(t *testing.T) {
+	listener := &coalescingListener{final: -1}
+	n := newNotifier()
+	wake := make(chan struct{}, 1)
+	stop := make(chan struct{})
+
+	wake <- struct{}{}
+	close(stop)
+	n.deliverPeerListChanges(listener, wake, stop)
+
+	if got := listener.calls.Load(); got != 0 {
+		t.Errorf("deliverer ran %d callbacks with stop closed, expected 0", got)
+	}
+}
+
+func waitFor(t *testing.T, ch <-chan struct{}, msg string) {
+	t.Helper()
+	select {
+	case <-ch:
+	case <-time.After(5 * time.Second):
+		t.Fatal(msg)
+	}
+}

client/internal/peer/status.go

@@ -18,9 +18,7 @@ import (
 	"google.golang.org/protobuf/types/known/durationpb"
 	"google.golang.org/protobuf/types/known/timestamppb"
 
-	firewall "github.com/netbirdio/netbird/client/firewall/manager"
 	"github.com/netbirdio/netbird/client/iface/configurer"
-	"github.com/netbirdio/netbird/client/internal/ingressgw"
 	"github.com/netbirdio/netbird/client/internal/relay"
 	"github.com/netbirdio/netbird/client/proto"
 	"github.com/netbirdio/netbird/route"
@@ -161,7 +159,6 @@ type FullStatus struct {
 	RosenpassState        RosenpassState
 	Relays                []relay.ProbeResult
 	NSGroupStates         []NSGroupState
-	NumOfForwardingRules  int
 	LazyConnectionEnabled bool
 	Events                []*proto.SystemEvent
 }
@@ -196,6 +193,7 @@ type Status struct {
 	muxRelays           sync.RWMutex
 	peers               map[string]State
 	ipToKey             map[string]string
+	activeRoutePeers    map[route.HAUniqueID]string
 	changeNotify        map[string]map[string]*StatusChangeSubscription // map[peerID]map[subscriptionID]*StatusChangeSubscription
 	signalState         bool
 	signalError         error
@@ -246,8 +244,6 @@ type Status struct {
 	// read it without taking mux.
 	networksRevision atomic.Uint64
 
-	ingressGwMgr *ingressgw.Manager
-
 	routeIDLookup routeIDLookup
 	wgIface       WGIfaceStatus
 }
@@ -257,6 +253,7 @@ func NewRecorder(mgmAddress string) *Status {
 	return &Status{
 		peers:                 make(map[string]State),
 		ipToKey:               make(map[string]string),
+		activeRoutePeers:      make(map[route.HAUniqueID]string),
 		changeNotify:          make(map[string]map[string]*StatusChangeSubscription),
 		eventStreams:          make(map[string]chan *proto.SystemEvent),
 		eventQueue:            NewEventQueue(eventQueueSize),
@@ -274,12 +271,6 @@ func (d *Status) SetRelayMgr(manager *relayClient.Manager) {
 	d.relayMgr = manager
 }
 
-func (d *Status) SetIngressGwMgr(ingressGwMgr *ingressgw.Manager) {
-	d.mux.Lock()
-	defer d.mux.Unlock()
-	d.ingressGwMgr = ingressGwMgr
-}
-
 // ReplaceOfflinePeers replaces
 func (d *Status) ReplaceOfflinePeers(replacement []State) {
 	d.mux.Lock()
@@ -330,18 +321,6 @@ func (d *Status) GetPeer(peerPubKey string) (State, error) {
 	return state, nil
 }
 
-func (d *Status) PeerByIP(ip string) (string, bool) {
-	d.mux.RLock()
-	defer d.mux.RUnlock()
-
-	for _, state := range d.peers {
-		if state.IP == ip {
-			return state.FQDN, true
-		}
-	}
-	return "", false
-}
-
 // PeerStateByIP returns the full peer State for the given tunnel IP.
 // Matches against either the IPv4 (State.IP) or IPv6 (State.IPv6) tunnel
 // address so dual-stack peers are reachable on either family. Only
@@ -481,6 +460,24 @@ func (d *Status) RemovePeerStateRoute(peer string, route string) error {
 	return nil
 }
 
+func (d *Status) AddActiveRoutePeer(haID route.HAUniqueID, peer string) {
+	d.mux.Lock()
+	defer d.mux.Unlock()
+	d.activeRoutePeers[haID] = peer
+}
+
+func (d *Status) RemoveActiveRoutePeer(haID route.HAUniqueID) {
+	d.mux.Lock()
+	defer d.mux.Unlock()
+	delete(d.activeRoutePeers, haID)
+}
+
+func (d *Status) GetActiveRoutePeers() map[route.HAUniqueID]string {
+	d.mux.RLock()
+	defer d.mux.RUnlock()
+	return maps.Clone(d.activeRoutePeers)
+}
+
 // CheckRoutes checks if the source and destination addresses are within the same route
 // and returns the resource ID of the route that contains the addresses
 func (d *Status) CheckRoutes(ip netip.Addr) ([]byte, bool) {
@@ -1143,16 +1140,6 @@ func (d *Status) GetRelayStates() []relay.ProbeResult {
 	return relayStates
 }
 
-func (d *Status) ForwardingRules() []firewall.ForwardRule {
-	d.mux.RLock()
-	defer d.mux.RUnlock()
-	if d.ingressGwMgr == nil {
-		return nil
-	}
-
-	return d.ingressGwMgr.Rules()
-}
-
 func (d *Status) GetDNSStates() []NSGroupState {
 	d.mux.RLock()
 	defer d.mux.RUnlock()
@@ -1187,7 +1174,6 @@ func (d *Status) GetFullStatus() FullStatus {
 		Relays:                d.GetRelayStates(),
 		RosenpassState:        d.GetRosenpassState(),
 		NSGroupStates:         d.GetDNSStates(),
-		NumOfForwardingRules:  len(d.ForwardingRules()),
 		LazyConnectionEnabled: d.GetLazyConnection(),
 	}
 
@@ -1559,7 +1545,6 @@ func (fs FullStatus) ToProto() *proto.FullStatus {
 	pbFullStatus.LocalPeerState.WgPort = int32(fs.LocalPeerState.WgPort)
 	pbFullStatus.LocalPeerState.RosenpassPermissive = fs.RosenpassState.Permissive
 	pbFullStatus.LocalPeerState.RosenpassEnabled = fs.RosenpassState.Enabled
-	pbFullStatus.NumberOfForwardingRules = int32(fs.NumOfForwardingRules)
 	pbFullStatus.LazyConnectionEnabled = fs.LazyConnectionEnabled
 
 	pbFullStatus.LocalPeerState.Networks = maps.Keys(fs.LocalPeerState.Routes)

client/internal/peer/status_test.go

@@ -9,6 +9,8 @@ import (
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/route"
 )
 
 func TestAddPeer(t *testing.T) {
@@ -372,3 +374,24 @@ func TestMarkServerStateDoesNotNotifyWhenUnchanged(t *testing.T) {
 	status.MarkManagementDisconnected(err)
 	assert.False(t, notified(ch), "redundant disconnect should not notify")
 }
+
+func TestActiveRoutePeers(t *testing.T) {
+	status := NewRecorder("https://mgm")
+	netA := route.HAUniqueID("net-a-10.0.0.0/24")
+	netB := route.HAUniqueID("net-b-10.0.0.0/24")
+
+	status.AddActiveRoutePeer(netA, "peerA")
+	status.AddActiveRoutePeer(netB, "peerB")
+
+	active := status.GetActiveRoutePeers()
+	assert.Equal(t, "peerA", active[netA])
+	assert.Equal(t, "peerB", active[netB])
+
+	status.RemoveActiveRoutePeer(netA)
+	delete(active, netB)
+
+	active = status.GetActiveRoutePeers()
+	_, ok := active[netA]
+	assert.False(t, ok)
+	assert.Equal(t, "peerB", active[netB])
+}

client/internal/peer/worker_ice.go

@@ -121,11 +121,8 @@ func (w *WorkerICE) OnNewOffer(remoteOfferAnswer *OfferAnswer) {
 			}
 		}
 
-		sessionID, err := NewICESessionID()
-		if err != nil {
-			w.log.Errorf("failed to create new session ID: %s", err)
-		}
-		w.sessionID = sessionID
+		// Keep the ID already advertised to the remote. Answers do not get a
+		// reply, so changing it here makes the next offer restart both sides.
 		w.abandonNegotiation()
 	}
 
@@ -205,6 +202,9 @@ func (w *WorkerICE) Close() {
 	w.muxAgent.Lock()
 	defer w.muxAgent.Unlock()
 
+	if w.agent != nil || w.agentConnecting {
+		w.renewSessionID()
+	}
 	if w.agent != nil {
 		w.agentDialerCancel()
 		if err := w.agent.Close(); err != nil {
@@ -366,16 +366,23 @@ func (w *WorkerICE) closeAgent(agent *icemaker.ThreadSafeAgent, cancel context.C
 	// Only the owner of the current session may reset its state: a stale dial
 	// goroutine waking after a newer attempt must not clobber it.
 	if w.agent == agent {
-		sessionID, err := NewICESessionID()
-		if err != nil {
-			w.log.Errorf("failed to create new session ID: %s", err)
-		}
-		w.sessionID = sessionID
+		w.renewSessionID()
 		w.abandonNegotiation()
 	}
 	return sessionChanged
 }
 
+// renewSessionID starts a new local session, so the remote treats our next offer
+// or answer as a restart. Caller holds muxAgent.
+func (w *WorkerICE) renewSessionID() {
+	sessionID, err := NewICESessionID()
+	if err != nil {
+		w.log.Errorf("failed to create new session ID: %s", err)
+		return
+	}
+	w.sessionID = sessionID
+}
+
 // abandonNegotiation drops all recorded ICE session state so the worker treats the
 // next offer as a fresh start instead of a duplicate of a dead negotiation. The
 // agent and agentConnecting flags must change together: leaving one stale wedges

client/internal/peer/worker_ice_session_test.go

@@ -0,0 +1,375 @@
+package peer
+
+import (
+	"context"
+	"fmt"
+	"net"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	icemaker "github.com/netbirdio/netbird/client/internal/peer/ice"
+)
+
+func TestWorkerICE_RemoteRestartPreservesAdvertisedSession(t *testing.T) {
+	w := newTestWorkerICE(t)
+	t.Cleanup(w.Close)
+	w.dialFunc = parkDial
+	advertised := w.SessionID()
+	remoteSession := ICESessionID("remote-first")
+	offer := OfferAnswer{
+		IceCredentials: IceCredentials{UFrag: "remoteufrag", Pwd: "remote-password-long-enough"},
+		SessionID:      &remoteSession,
+	}
+	w.OnNewOffer(&offer)
+	require.True(t, w.InProgress(), "the first remote session must start ICE")
+	w.muxAgent.Lock()
+	firstAgent := w.agent
+	w.muxAgent.Unlock()
+
+	// The same callback handles answers. A changed remote ID must not create
+	// an unannounced local ID that makes the remote restart on our next offer.
+	secondSession := ICESessionID("remote-restarted")
+	answer := offer
+	answer.SessionID = &secondSession
+	w.OnNewOffer(&answer)
+	assert.Equal(t, advertised, w.SessionID(), "following a remote restart must keep our advertised ID")
+	w.muxAgent.Lock()
+	secondAgent := w.agent
+	w.muxAgent.Unlock()
+	assert.NotSame(t, firstAgent, secondAgent, "the changed remote session must still rebuild ICE")
+
+	w.OnNewOffer(&answer)
+	w.muxAgent.Lock()
+	defer w.muxAgent.Unlock()
+	assert.Same(t, secondAgent, w.agent, "a repeated answer must keep the replacement agent")
+}
+
+func TestWorkerICE_LocalCloseChangesAdvertisedSession(t *testing.T) {
+	w := newTestWorkerICE(t)
+	dialStarted := make(chan struct{})
+	dialDone := make(chan struct{})
+	w.dialFunc = func(ctx context.Context, _ *icemaker.ThreadSafeAgent, _ *OfferAnswer) (net.Conn, error) {
+		close(dialStarted)
+		defer close(dialDone)
+		<-ctx.Done()
+		return nil, ctx.Err()
+	}
+	session := ICESessionID("remote-session")
+	w.OnNewOffer(&OfferAnswer{
+		IceCredentials: IceCredentials{UFrag: "remoteufrag", Pwd: "remote-password-long-enough"},
+		SessionID:      &session,
+	})
+	<-dialStarted
+	advertised := w.SessionID()
+	w.Close()
+	assert.NotEqual(t, advertised, w.SessionID(), "a local teardown must tell the remote to restart")
+	closedSession := w.SessionID()
+
+	// The abandoned dial goroutine cleans up after Close returned.
+	<-dialDone
+	assert.Never(t, func() bool { return w.SessionID() != closedSession }, 200*time.Millisecond, 10*time.Millisecond,
+		"the late cleanup of a closed negotiation must not restart again")
+	w.Close()
+	assert.Equal(t, closedSession, w.SessionID(), "closing an idle worker must not restart again")
+}
+
+// parkDial stands in for the ICE dial. It never connects and returns once the
+// negotiation is abandoned, so a test decides when a negotiation fails.
+func parkDial(ctx context.Context, _ *icemaker.ThreadSafeAgent, _ *OfferAnswer) (net.Conn, error) {
+	<-ctx.Done()
+	return nil, ctx.Err()
+}
+
+func newTestSessionID(t *testing.T) ICESessionID {
+	t.Helper()
+	sid, err := NewICESessionID()
+	require.NoError(t, err)
+	return sid
+}
+
+// handshakeSide is one end of a simulated signaling exchange.
+type handshakeSide interface {
+	// message builds the offer or answer the side would send now.
+	message() OfferAnswer
+	// receive hands a remote offer or answer to the side's ICE logic.
+	receive(msg OfferAnswer)
+	// teardowns counts negotiations the side tore down to follow a remote restart.
+	teardowns() int
+	// failAgent ends the side's current negotiation as an ICE failure does.
+	failAgent()
+}
+
+// workerSide drives a real WorkerICE.
+type workerSide struct {
+	t        *testing.T
+	w        *WorkerICE
+	replaced int
+}
+
+func newWorkerSide(t *testing.T) *workerSide {
+	t.Helper()
+	w := newTestWorkerICE(t)
+	w.dialFunc = parkDial
+	t.Cleanup(w.Close)
+	return &workerSide{t: t, w: w}
+}
+
+func (s *workerSide) message() OfferAnswer {
+	sid := s.w.SessionID()
+	ufrag, pwd := s.w.GetLocalUserCredentials()
+	return OfferAnswer{IceCredentials: IceCredentials{UFrag: ufrag, Pwd: pwd}, SessionID: &sid}
+}
+
+func (s *workerSide) receive(msg OfferAnswer) {
+	before := s.agent()
+	s.w.OnNewOffer(&msg)
+	if after := s.agent(); before != nil && after != before {
+		s.replaced++
+	}
+}
+
+func (s *workerSide) teardowns() int { return s.replaced }
+
+func (s *workerSide) agent() *icemaker.ThreadSafeAgent {
+	s.w.muxAgent.Lock()
+	defer s.w.muxAgent.Unlock()
+	return s.w.agent
+}
+
+// failAgent runs the cleanup the dial goroutine or the Failed state callback
+// performs when the current negotiation dies.
+func (s *workerSide) failAgent() {
+	s.t.Helper()
+	s.w.muxAgent.Lock()
+	agent, cancel := s.w.agent, s.w.agentDialerCancel
+	s.w.muxAgent.Unlock()
+	require.NotNil(s.t, agent, "failing requires a running negotiation")
+	s.w.closeAgent(agent, cancel)
+}
+
+// legacySide models a remote peer running a release from before this change:
+// when it follows a remote restart it also picks a new session ID of its own,
+// which it announces only with its next offer or answer.
+type legacySide struct {
+	t         *testing.T
+	sessionID ICESessionID
+	remoteID  ICESessionID
+	hasAgent  bool
+	replaced  int
+}
+
+func newLegacySide(t *testing.T) *legacySide {
+	return &legacySide{t: t, sessionID: newTestSessionID(t)}
+}
+
+func (s *legacySide) message() OfferAnswer {
+	sid := s.sessionID
+	return OfferAnswer{
+		IceCredentials: IceCredentials{UFrag: "legacyufrag", Pwd: "legacy-password-long-enough"},
+		SessionID:      &sid,
+	}
+}
+
+func (s *legacySide) receive(msg OfferAnswer) {
+	if msg.SessionID == nil {
+		s.hasAgent = true
+		return
+	}
+	if s.hasAgent {
+		if *msg.SessionID == s.remoteID {
+			return
+		}
+		s.replaced++
+		s.sessionID = newTestSessionID(s.t)
+	}
+	s.hasAgent = true
+	s.remoteID = *msg.SessionID
+}
+
+func (s *legacySide) teardowns() int { return s.replaced }
+
+func (s *legacySide) failAgent() {
+	s.hasAgent = false
+	s.remoteID = ""
+	s.sessionID = newTestSessionID(s.t)
+}
+
+// exchange runs one guard-driven round in the order Handshaker.Listen uses: the
+// answerer handles the offer and answers with the session ID it holds
+// afterwards, and the offerer handles the answer without replying.
+func exchange(offerer, answerer handshakeSide) {
+	answerer.receive(offerer.message())
+	offerer.receive(answerer.message())
+}
+
+// offerPattern decides which side's guard sends the offer in a round.
+type offerPattern struct {
+	name   string
+	picker func(round int, local, remote handshakeSide) (offerer, answerer handshakeSide)
+}
+
+var offerPatterns = []offerPattern{
+	{
+		// A routing peer whose relay is down keeps offering on its own.
+		name: "local peer offers",
+		picker: func(_ int, local, remote handshakeSide) (handshakeSide, handshakeSide) {
+			return local, remote
+		},
+	},
+	{
+		name: "both peers offer",
+		picker: func(round int, local, remote handshakeSide) (handshakeSide, handshakeSide) {
+			if round%2 == 0 {
+				return local, remote
+			}
+			return remote, local
+		},
+	},
+}
+
+// assertSettles runs guard rounds and requires the pair to stop restarting
+// each other: at most maxTeardowns in total, and none once half the rounds ran.
+func assertSettles(t *testing.T, pattern offerPattern, local, remote handshakeSide, maxTeardowns int) {
+	t.Helper()
+	const rounds = 10
+
+	total := func() int { return local.teardowns() + remote.teardowns() }
+	start := total()
+	var halfway int
+	for round := range rounds {
+		if round == rounds/2 {
+			halfway = total()
+		}
+		offerer, answerer := pattern.picker(round, local, remote)
+		exchange(offerer, answerer)
+	}
+
+	assert.LessOrEqual(t, total()-start, maxTeardowns, "the peers must not keep restarting each other")
+	assert.Equal(t, halfway, total(), "the negotiation must be stable in the later rounds")
+}
+
+// establish runs the first offer and answer, so both sides negotiate.
+func establish(t *testing.T, local, remote handshakeSide) {
+	t.Helper()
+	exchange(local, remote)
+	require.Zero(t, local.teardowns()+remote.teardowns(), "the first exchange must not restart anything")
+}
+
+func TestICESession_SettlesAfterAgentFailure(t *testing.T) {
+	sides := []struct {
+		name   string
+		remote func(t *testing.T) handshakeSide
+	}{
+		{name: "current remote", remote: func(t *testing.T) handshakeSide { return newWorkerSide(t) }},
+		{name: "legacy remote", remote: func(t *testing.T) handshakeSide { return newLegacySide(t) }},
+	}
+	failures := []struct {
+		name string
+		fail func(local, remote handshakeSide)
+	}{
+		{name: "remote agent fails", fail: func(_, remote handshakeSide) { remote.failAgent() }},
+		{name: "local agent fails", fail: func(local, _ handshakeSide) { local.failAgent() }},
+		{name: "both agents fail", fail: func(local, remote handshakeSide) {
+			local.failAgent()
+			remote.failAgent()
+		}},
+	}
+
+	for _, side := range sides {
+		for _, failure := range failures {
+			for _, pattern := range offerPatterns {
+				t.Run(fmt.Sprintf("%s/%s/%s", side.name, failure.name, pattern.name), func(t *testing.T) {
+					local := newWorkerSide(t)
+					remote := side.remote(t)
+					establish(t, local, remote)
+
+					failure.fail(local, remote)
+					assertSettles(t, pattern, local, remote, 2)
+				})
+			}
+		}
+	}
+}
+
+// TestICESession_LocalCloseRestartsRemote covers an explicit teardown, as on a
+// WireGuard handshake timeout. The remote must start over as well, or it keeps
+// answering from the negotiation this side just abandoned.
+func TestICESession_LocalCloseRestartsRemote(t *testing.T) {
+	for _, pattern := range offerPatterns {
+		t.Run(pattern.name, func(t *testing.T) {
+			local := newWorkerSide(t)
+			remote := newWorkerSide(t)
+			establish(t, local, remote)
+
+			local.w.Close()
+			assertSettles(t, pattern, local, remote, 1)
+			assert.Equal(t, 1, remote.teardowns(), "the remote must restart its negotiation exactly once")
+		})
+	}
+}
+
+func TestICESession_DuplicateMessagesKeepNegotiation(t *testing.T) {
+	local := newWorkerSide(t)
+	remote := newWorkerSide(t)
+
+	offer := local.message()
+	remote.receive(offer)
+	answer := remote.message()
+	local.receive(answer)
+
+	// Signaling may deliver the same message again, and a peer answers every
+	// offer, including repeats of one it already handled.
+	remote.receive(offer)
+	local.receive(answer)
+	local.receive(remote.message())
+
+	assert.Zero(t, local.teardowns(), "a repeated answer must not restart the negotiation")
+	assert.Zero(t, remote.teardowns(), "a repeated offer must not restart the negotiation")
+}
+
+// TestICESession_RemoteWithoutSessionIDKeepsNegotiation covers remote peers
+// too old to send session IDs: once negotiating, their messages cannot tell a
+// restart from a repeat, so they must not tear anything down.
+func TestICESession_RemoteWithoutSessionIDKeepsNegotiation(t *testing.T) {
+	local := newWorkerSide(t)
+	unversioned := OfferAnswer{IceCredentials: IceCredentials{UFrag: "oldufrag", Pwd: "old-password-long-enough"}}
+
+	local.receive(unversioned)
+	require.NotNil(t, local.agent(), "a message without a session ID must still start ICE")
+	advertised := local.w.SessionID()
+
+	for range 3 {
+		local.receive(unversioned)
+	}
+	assert.Zero(t, local.teardowns(), "messages without a session ID must not restart the negotiation")
+	assert.Equal(t, advertised, local.w.SessionID(), "the advertised session must not change")
+}
+
+// TestWorkerICE_StaleCleanupKeepsAdvertisedSession covers the cleanup of a
+// replaced negotiation finishing late, from its dial goroutine or its Closed
+// state callback. It must neither pick a new session ID, an unannounced local
+// restart, nor disturb the negotiation that replaced it.
+func TestWorkerICE_StaleCleanupKeepsAdvertisedSession(t *testing.T) {
+	local := newWorkerSide(t)
+	remote := newWorkerSide(t)
+	establish(t, local, remote)
+
+	local.w.muxAgent.Lock()
+	oldAgent, oldCancel := local.w.agent, local.w.agentDialerCancel
+	local.w.muxAgent.Unlock()
+
+	remote.failAgent()
+	exchange(local, remote)
+	require.Equal(t, 1, local.teardowns(), "the local side must follow the remote restart")
+	advertised := local.w.SessionID()
+	current := local.agent()
+
+	local.w.closeAgent(oldAgent, oldCancel)
+
+	assert.Equal(t, advertised, local.w.SessionID(), "a stale cleanup must not change the advertised session")
+	assert.Same(t, current, local.agent(), "a stale cleanup must keep the current negotiation")
+	assertSettles(t, offerPatterns[1], local, remote, 0)
+}

client/internal/peer/worker_relay.go

@@ -101,6 +101,10 @@ func (w *WorkerRelay) RelayIsSupportedLocally() bool {
 	return w.relayManager.HasRelayAddress()
 }
 
+func (w *WorkerRelay) IsTransportConnected() bool {
+	return w.relayManager.Ready()
+}
+
 func (w *WorkerRelay) CloseConn() {
 	w.relayLock.Lock()
 	conn := w.relayedConn

client/internal/profilemanager/config.go

@@ -10,7 +10,6 @@ import (
 	"os"
 	"os/user"
 	"path/filepath"
-	"reflect"
 	"runtime"
 	"slices"
 	"strings"
@@ -198,6 +197,11 @@ type Config struct {
 
 	MTU uint16
 
+	// probing marks a config that exists only to be compared against and then
+	// thrown away, so apply() can skip the work that feeds no verdict.
+	// Unexported, so it never reaches the JSON.
+	probing bool
+
 	// policy is the MDM policy that produced the currently-set values
 	// for any MDM-enforced fields. Set by ApplyMDMPolicy on every
 	// invocation. Never persisted to disk. Callers query enforcement
@@ -300,9 +304,11 @@ func fileExists(path string) (bool, error) {
 	return false, err
 }
 
-// createNewConfig creates a new config generating a new Wireguard key and saving to file
-func createNewConfig(input ConfigInput) (*Config, error) {
-	config := &Config{
+// newConfigSkeleton returns the field values a brand-new profile config starts
+// from, before apply() fills in the rest. Shared with the dry-run baseline so
+// the two cannot disagree about what "a new config" means.
+func newConfigSkeleton() *Config {
+	return &Config{
 		// defaults to false only for new (post 0.26) configurations
 		ServerSSHAllowed: util.False(),
 		// Remote jobs are an explicit opt-in and default off, including for
@@ -310,6 +316,91 @@ func createNewConfig(input ConfigInput) (*Config, error) {
 		RemoteJobsAllowed: util.False(),
 		WgPort:            iface.DefaultWgPort,
 	}
+}
+
+// resolveUnsetDefaults is the single place where an optional field that carries
+// no value gets one, and the only place that states what each of those defaults
+// is. apply() runs it before it compares anything, and that ordering is the
+// point: with the values named, every comparison below it diffs values instead
+// of presence.
+//
+// Presence-based comparison is what broke `netbird up` for a client configured
+// through the environment. These fields mean "the effective default" when they
+// hold nothing — every consumer already reads a nil as the value resolved here,
+// the SSH toggles in engine_ssh.go and the network monitor in
+// createEngineConfig — so naming them changes nothing about what runs. But
+// while they stayed nil, an input restating the default read as a change, and
+// since the CLI sends every flag whose value came from an environment variable
+// on each `netbird up`, a client with NB_ENABLE_SSH_ROOT=false restated it
+// every time and the update-settings gate refused it.
+//
+// Filling a field in is not a settings change, so a caller measuring change
+// must not read the returned bool as one: see WouldChange, which runs a pass
+// for this and discards its verdict.
+//
+// ServerSSHAllowed is the one field whose default depends on the config's age.
+// A brand-new profile gets false from newConfigSkeleton, which runs before
+// this, so what is resolved here is only the legacy case: a config written by a
+// version that had no such field keeps SSH on, for backwards compatibility.
+func (config *Config) resolveUnsetDefaults() (updated bool) {
+	// Fields that default to false on every platform.
+	for _, field := range []**bool{
+		&config.EnableSSHRoot,
+		&config.EnableSSHSFTP,
+		&config.EnableSSHLocalPortForwarding,
+		&config.EnableSSHRemotePortForwarding,
+		&config.DisableSSHAuth,
+		// Remote jobs are an explicit opt-in: unlike SSH, a pre-existing config
+		// with no value defaults to disabled rather than being turned on.
+		&config.RemoteJobsAllowed,
+	} {
+		if *field == nil {
+			*field = util.False()
+			updated = true
+		}
+	}
+
+	if config.DisableNotifications == nil {
+		log.Infof("setting notifications to disabled by default")
+		config.DisableNotifications = util.True()
+		updated = true
+	}
+
+	if config.SSHJWTCacheTTL == nil {
+		// A zero TTL disables the JWT cache, which is what no value meant.
+		config.SSHJWTCacheTTL = new(int)
+		updated = true
+	}
+
+	if config.NetworkMonitor == nil {
+		// network monitoring is on by default on windows and darwin clients
+		enabled := runtime.GOOS == "windows" || runtime.GOOS == "darwin"
+		config.NetworkMonitor = &enabled
+		updated = true
+	}
+
+	if config.ServerSSHAllowed == nil {
+		if runtime.GOOS == "android" {
+			// default to disabled SSH on Android for security
+			log.Infof("setting SSH server to false by default on Android")
+			config.ServerSSHAllowed = util.False()
+		} else {
+			// enables SSH for configs from old versions to preserve backwards compatibility
+			log.Infof("falling back to enabled SSH server for pre-existing configuration")
+			config.ServerSSHAllowed = util.True()
+		}
+		updated = true
+	}
+
+	return updated
+}
+
+// createNewConfig resolves a new config in memory, with no identity: whoever
+// needs the peer's keys calls EnsureIdentity and persists the result, so a read
+// that lands on a missing file cannot hand back a config carrying keys that
+// nothing will ever write down.
+func createNewConfig(input ConfigInput) (*Config, error) {
+	config := newConfigSkeleton()
 
 	if _, err := config.apply(input); err != nil {
 		return nil, err
@@ -318,6 +409,52 @@ func createNewConfig(input ConfigInput) (*Config, error) {
 	return config, nil
 }
 
+// createProvisionedConfig is createNewConfig plus the peer's identity, for the
+// callers that go on to persist the config or to connect with it.
+func createProvisionedConfig(input ConfigInput) (*Config, error) {
+	config, err := createNewConfig(input)
+	if err != nil {
+		return nil, err
+	}
+
+	if _, err := config.EnsureIdentity(); err != nil {
+		return nil, err
+	}
+
+	return config, nil
+}
+
+// EnsureIdentity generates the keys that identify this peer if the config does
+// not carry them yet, reporting whether it had to generate any.
+//
+// It is deliberately not part of apply(). Everything apply() fills in is a
+// default it can recompute on the next read, but a generated key is not: it
+// has to be persisted, or the peer comes back with a different WireGuard
+// identity and re-registers. Having apply() generate keys is what forced every
+// read of a config to write it back — so identity provisioning is its own step
+// now, and the callers that perform it write the result out explicitly.
+func (config *Config) EnsureIdentity() (bool, error) {
+	generated := false
+
+	if config.PrivateKey == "" {
+		log.Infof("generated new Wireguard key")
+		config.PrivateKey = generateKey()
+		generated = true
+	}
+
+	if config.SSHKey == "" {
+		log.Infof("generated new SSH key")
+		pem, err := ssh.GeneratePrivateKey(ssh.ED25519)
+		if err != nil {
+			return generated, err
+		}
+		config.SSHKey = string(pem)
+		generated = true
+	}
+
+	return generated, nil
+}
+
 func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 	if config.Name != "" {
 		sanitized, err := sanitizeDisplayName(config.Name)
@@ -329,6 +466,13 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 			updated = true
 		}
 	}
+
+	// Every optional field gets its value here, before anything below compares
+	// one. See resolveUnsetDefaults for why that ordering is the point.
+	if config.resolveUnsetDefaults() {
+		updated = true
+	}
+
 	if config.ManagementURL == nil {
 		log.Infof("using default Management URL %s", DefaultManagementURL)
 		config.ManagementURL, err = parseURL("Management URL", DefaultManagementURL)
@@ -336,20 +480,21 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 			return false, err
 		}
 	}
-	if input.ManagementURL != "" && input.ManagementURL != config.ManagementURL.String() {
-		log.Infof("new Management URL provided, updated to %#v (old value %#v)",
-			input.ManagementURL, config.ManagementURL.String())
+	// The comparison is on the endpoint the URL addresses, not on its
+	// spelling: the same endpoint can be written several ways (an implicit
+	// :443, a trailing slash, a different host case), and treating an
+	// equivalent URL as new would rewrite the config and report a settings
+	// change where the configuration does not actually change.
+	if input.ManagementURL != "" {
 		URL, err := parseURL("Management URL", input.ManagementURL)
 		if err != nil {
 			return false, err
 		}
-		config.ManagementURL = URL
-		updated = true
-	} else if config.ManagementURL == nil {
-		log.Infof("using default Management URL %s", DefaultManagementURL)
-		config.ManagementURL, err = parseURL("Management URL", DefaultManagementURL)
-		if err != nil {
-			return false, err
+		if !SameServiceURL(URL, config.ManagementURL) {
+			log.Infof("new Management URL provided, updated to %#v (old value %#v)",
+				URL.String(), config.ManagementURL.String())
+			config.ManagementURL = URL
+			updated = true
 		}
 	}
 
@@ -360,31 +505,20 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 			return false, err
 		}
 	}
-	if input.AdminURL != "" && input.AdminURL != config.AdminURL.String() {
-		log.Infof("new Admin Panel URL provided, updated to %#v (old value %#v)",
-			input.AdminURL, config.AdminURL.String())
+	// The admin panel is opened, not dialed, so unlike the Management URL its
+	// path is part of what identifies it: a panel served under /netbird is not
+	// the one served at the root.
+	if input.AdminURL != "" {
 		newURL, err := parseURL("Admin Panel URL", input.AdminURL)
 		if err != nil {
 			return updated, err
 		}
-		config.AdminURL = newURL
-		updated = true
-	}
-
-	if config.PrivateKey == "" {
-		log.Infof("generated new Wireguard key")
-		config.PrivateKey = generateKey()
-		updated = true
-	}
-
-	if config.SSHKey == "" {
-		log.Infof("generated new SSH key")
-		pem, err := ssh.GeneratePrivateKey(ssh.ED25519)
-		if err != nil {
-			return false, err
+		if !SameServiceURLIncludingPath(newURL, config.AdminURL) {
+			log.Infof("new Admin Panel URL provided, updated to %#v (old value %#v)",
+				newURL.String(), config.AdminURL.String())
+			config.AdminURL = newURL
+			updated = true
 		}
-		config.SSHKey = string(pem)
-		updated = true
 	}
 
 	if input.WireguardPort != nil && *input.WireguardPort != config.WgPort {
@@ -405,7 +539,14 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.NATExternalIPs != nil && !reflect.DeepEqual(config.NATExternalIPs, input.NATExternalIPs) {
+	// slices.Equal, not reflect.DeepEqual, and for the same reason the DNS
+	// labels below use it: DeepEqual calls a nil slice and an empty one
+	// different, while both mean "no NAT mappings". A profile stores the
+	// absent list as JSON null and reads it back nil, and `netbird up` sends
+	// CleanNATExternalIPs — an empty list — whenever NB_EXTERNAL_IP_MAP is set
+	// to nothing, so the two met on every start and the gate read a no-op as a
+	// settings change.
+	if input.NATExternalIPs != nil && !slices.Equal(config.NATExternalIPs, input.NATExternalIPs) {
 		log.Infof("updating NAT External IP [ %s ] (old value: [ %s ])",
 			strings.Join(input.NATExternalIPs, " "),
 			strings.Join(config.NATExternalIPs, " "))
@@ -443,21 +584,12 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.NetworkMonitor != nil && (config.NetworkMonitor == nil || *input.NetworkMonitor != *config.NetworkMonitor) {
+	if input.NetworkMonitor != nil && *input.NetworkMonitor != *config.NetworkMonitor {
 		log.Infof("switching Network Monitor to %t", *input.NetworkMonitor)
 		config.NetworkMonitor = input.NetworkMonitor
 		updated = true
 	}
 
-	if config.NetworkMonitor == nil {
-		// enable network monitoring by default on windows and darwin clients
-		if runtime.GOOS == "windows" || runtime.GOOS == "darwin" {
-			enabled := true
-			config.NetworkMonitor = &enabled
-			updated = true
-		}
-	}
-
 	if input.CustomDNSAddress != nil && string(input.CustomDNSAddress) != config.CustomDNSAddress {
 		log.Infof("updating custom DNS address %#v (old value %#v)",
 			string(input.CustomDNSAddress), config.CustomDNSAddress)
@@ -490,7 +622,7 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.ServerSSHAllowed != nil && (config.ServerSSHAllowed == nil || *input.ServerSSHAllowed != *config.ServerSSHAllowed) {
+	if input.ServerSSHAllowed != nil && *input.ServerSSHAllowed != *config.ServerSSHAllowed {
 		if *input.ServerSSHAllowed {
 			log.Infof("enabling SSH server")
 		} else {
@@ -498,20 +630,9 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		}
 		config.ServerSSHAllowed = input.ServerSSHAllowed
 		updated = true
-	} else if config.ServerSSHAllowed == nil {
-		if runtime.GOOS == "android" {
-			// default to disabled SSH on Android for security
-			log.Infof("setting SSH server to false by default on Android")
-			config.ServerSSHAllowed = util.False()
-		} else {
-			// enables SSH for configs from old versions to preserve backwards compatibility
-			log.Infof("falling back to enabled SSH server for pre-existing configuration")
-			config.ServerSSHAllowed = util.True()
-		}
-		updated = true
 	}
 
-	if input.RemoteJobsAllowed != nil && (config.RemoteJobsAllowed == nil || *input.RemoteJobsAllowed != *config.RemoteJobsAllowed) {
+	if input.RemoteJobsAllowed != nil && *input.RemoteJobsAllowed != *config.RemoteJobsAllowed {
 		if *input.RemoteJobsAllowed {
 			log.Infof("enabling remote jobs")
 		} else {
@@ -519,14 +640,9 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		}
 		config.RemoteJobsAllowed = input.RemoteJobsAllowed
 		updated = true
-	} else if config.RemoteJobsAllowed == nil {
-		// Remote jobs are an explicit opt-in: unlike SSH, a pre-existing config
-		// with no value defaults to disabled rather than being turned on.
-		config.RemoteJobsAllowed = util.False()
-		updated = true
 	}
 
-	if input.EnableSSHRoot != nil && (config.EnableSSHRoot == nil || *input.EnableSSHRoot != *config.EnableSSHRoot) {
+	if input.EnableSSHRoot != nil && *input.EnableSSHRoot != *config.EnableSSHRoot {
 		if *input.EnableSSHRoot {
 			log.Infof("enabling SSH root login")
 		} else {
@@ -536,7 +652,7 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.EnableSSHSFTP != nil && (config.EnableSSHSFTP == nil || *input.EnableSSHSFTP != *config.EnableSSHSFTP) {
+	if input.EnableSSHSFTP != nil && *input.EnableSSHSFTP != *config.EnableSSHSFTP {
 		if *input.EnableSSHSFTP {
 			log.Infof("enabling SSH SFTP subsystem")
 		} else {
@@ -546,7 +662,7 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.EnableSSHLocalPortForwarding != nil && (config.EnableSSHLocalPortForwarding == nil || *input.EnableSSHLocalPortForwarding != *config.EnableSSHLocalPortForwarding) {
+	if input.EnableSSHLocalPortForwarding != nil && *input.EnableSSHLocalPortForwarding != *config.EnableSSHLocalPortForwarding {
 		if *input.EnableSSHLocalPortForwarding {
 			log.Infof("enabling SSH local port forwarding")
 		} else {
@@ -556,7 +672,7 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.EnableSSHRemotePortForwarding != nil && (config.EnableSSHRemotePortForwarding == nil || *input.EnableSSHRemotePortForwarding != *config.EnableSSHRemotePortForwarding) {
+	if input.EnableSSHRemotePortForwarding != nil && *input.EnableSSHRemotePortForwarding != *config.EnableSSHRemotePortForwarding {
 		if *input.EnableSSHRemotePortForwarding {
 			log.Infof("enabling SSH remote port forwarding")
 		} else {
@@ -566,7 +682,7 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.DisableSSHAuth != nil && (config.DisableSSHAuth == nil || *input.DisableSSHAuth != *config.DisableSSHAuth) {
+	if input.DisableSSHAuth != nil && *input.DisableSSHAuth != *config.DisableSSHAuth {
 		if *input.DisableSSHAuth {
 			log.Infof("disabling SSH authentication")
 		} else {
@@ -576,7 +692,7 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.SSHJWTCacheTTL != nil && (config.SSHJWTCacheTTL == nil || *input.SSHJWTCacheTTL != *config.SSHJWTCacheTTL) {
+	if input.SSHJWTCacheTTL != nil && *input.SSHJWTCacheTTL != *config.SSHJWTCacheTTL {
 		log.Infof("updating SSH JWT cache TTL to %d seconds", *input.SSHJWTCacheTTL)
 		config.SSHJWTCacheTTL = input.SSHJWTCacheTTL
 		updated = true
@@ -659,13 +775,16 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if input.SyncMessageVersion != nil && *input.SyncMessageVersion != *config.SyncMessageVersion {
+	// Assigning the pointer, not writing through it: a config that carries no
+	// version yet would otherwise be a nil dereference, and a panic inside a
+	// request handler is not a way to fail.
+	if input.SyncMessageVersion != nil && (config.SyncMessageVersion == nil || *input.SyncMessageVersion != *config.SyncMessageVersion) {
 		log.Infof("setting SyncMessageVersion to %v", *input.SyncMessageVersion)
-		*config.SyncMessageVersion = *input.SyncMessageVersion
+		config.SyncMessageVersion = input.SyncMessageVersion
 		updated = true
 	}
 
-	if input.DisableNotifications != nil && (config.DisableNotifications == nil || *input.DisableNotifications != *config.DisableNotifications) {
+	if input.DisableNotifications != nil && *input.DisableNotifications != *config.DisableNotifications {
 		if *input.DisableNotifications {
 			log.Infof("disabling notifications")
 		} else {
@@ -675,24 +794,24 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	if config.DisableNotifications == nil {
-		disabled := true
-		config.DisableNotifications = &disabled
-		log.Infof("setting notifications to disabled by default")
-		updated = true
-	}
-
-	if input.ClientCertKeyPath != "" {
+	// Compared, not just assigned: restating the path a config already holds
+	// changes nothing, and reporting it as an update makes a caller that
+	// re-sends its own configuration look like one asking to change it.
+	if input.ClientCertKeyPath != "" && input.ClientCertKeyPath != config.ClientCertKeyPath {
 		config.ClientCertKeyPath = input.ClientCertKeyPath
 		updated = true
 	}
 
-	if input.ClientCertPath != "" {
+	if input.ClientCertPath != "" && input.ClientCertPath != config.ClientCertPath {
 		config.ClientCertPath = input.ClientCertPath
 		updated = true
 	}
 
-	if config.ClientCertPath != "" && config.ClientCertKeyPath != "" {
+	// Not on a probe: the loaded pair feeds the connection, never the
+	// comparison, and this would otherwise run on every gated SetConfig and
+	// Login — twice per request — including those that are refused or change
+	// nothing, logging an error per request when the files are missing.
+	if !config.probing && config.ClientCertPath != "" && config.ClientCertKeyPath != "" {
 		cert, err := tls.LoadX509KeyPair(config.ClientCertPath, config.ClientCertKeyPath)
 		if err != nil {
 			log.Error("Failed to load mTLS cert/key pair: ", err)
@@ -886,6 +1005,49 @@ func ParseServiceURL(serviceName, serviceURL string) (*url.URL, error) {
 	return parseURL(serviceName, serviceURL)
 }
 
+// SameServiceURL reports whether two service URLs address the same endpoint:
+// same scheme, same host compared case-insensitively as DNS names are, and
+// same effective port, where an absent port means the scheme's default.
+//
+// This is the one comparison every caller deciding "did this URL change?" must
+// use. A string comparison answers a different question: "https://host",
+// "https://host/" and "https://HOST:443" are one endpoint written three ways,
+// and reading them as three values makes a client that restates its own
+// management URL look like a client asking to be repointed. A nil operand
+// matches only another nil one.
+//
+// The path plays no part: a management URL is dialed, and only its host and
+// port are. util.SameServiceURL is this comparison plus the path, which is
+// what SameServiceURLIncludingPath needs and delegates to.
+func SameServiceURL(a, b *url.URL) bool {
+	if a == nil || b == nil {
+		return a == b
+	}
+
+	return strings.EqualFold(a.Scheme, b.Scheme) &&
+		strings.EqualFold(a.Hostname(), b.Hostname()) &&
+		util.ServiceURLPort(a) == util.ServiceURLPort(b)
+}
+
+// SameServiceURLIncludingPath is SameServiceURL plus everything a URL carries
+// past its endpoint: path, query, fragment and userinfo.
+//
+// Use it for a URL that gets opened rather than dialed. The admin panel can
+// live under a path, so two URLs with the same endpoint and different paths are
+// two different panels — where for a URL the client dials over gRPC only the
+// endpoint is ever used. Equivalent spellings still compare equal: a missing
+// path and "/" are the same root, and so is a trailing slash on any path.
+func SameServiceURLIncludingPath(a, b *url.URL) bool {
+	if a == nil || b == nil {
+		return a == b
+	}
+
+	return util.SameServiceURL(a, b) &&
+		a.RawQuery == b.RawQuery &&
+		a.Fragment == b.Fragment &&
+		a.User.String() == b.User.String()
+}
+
 func parseURL(serviceName, serviceURL string) (*url.URL, error) {
 	parsedMgmtURL, err := url.ParseRequestURI(serviceURL)
 	if err != nil {
@@ -930,6 +1092,84 @@ func isPreSharedKeyHidden(preSharedKey *string) bool {
 	return false
 }
 
+// WouldChange reports whether applying input would modify any field the
+// config persists, leaving the receiver untouched. It is the dry-run half of
+// UpdateConfig and reuses the very same diff logic (Config.apply), so a
+// caller asking "is this a settings change?" cannot drift from what an
+// actual update would do, nor go stale when a new field is added.
+//
+// A redacted pre-shared key is collapsed to "unset" exactly as
+// UpdateOrCreateConfig does, so a UI that round-trips the mask is not read as
+// a request for a new key.
+//
+// A nil receiver means the profile holds no config yet, so the baseline is the
+// config the daemon would create for it: input values matching those defaults
+// change nothing, anything else does.
+func (config *Config) WouldChange(input ConfigInput) (bool, error) {
+	probe := config.clone()
+	if probe == nil {
+		baseline, err := newDryRunBaseline(input.ConfigPath)
+		if err != nil {
+			return true, fmt.Errorf("build default config baseline: %w", err)
+		}
+		probe = baseline
+	}
+	probe.probing = true
+
+	// Normalize before measuring. apply() reports two different things through
+	// one bool: an input that changed a value, and a field it had to fill in
+	// because the config carried none. Only the first is a settings change, so
+	// the filling-in gets a pass of its own whose verdict is discarded, and the
+	// pass that answers the caller runs against a config with nothing left to
+	// fill in.
+	//
+	// Readers already hand out normalized configs — readConfig applies an empty
+	// input for this very reason — so this is normally a no-op. But a gate that
+	// refuses a request must not depend on where its caller got the config
+	// from, and it must not start reading "this profile predates a field" as
+	// "the caller asked for a change" the day someone adds one.
+	if _, err := probe.apply(ConfigInput{ConfigPath: input.ConfigPath}); err != nil {
+		return true, fmt.Errorf("normalize the config to diff against: %w", err)
+	}
+
+	if isPreSharedKeyHidden(input.PreSharedKey) {
+		input.PreSharedKey = nil
+	}
+
+	return probe.apply(input)
+}
+
+// newDryRunBaseline builds the config a brand-new profile would start from, for
+// a dry run to compare an input against. It is createNewConfig without the
+// identity: this config exists only to be compared against and thrown away, and
+// no ConfigInput field maps to either key.
+func newDryRunBaseline(configPath string) (*Config, error) {
+	baseline := newConfigSkeleton()
+
+	if _, err := baseline.apply(ConfigInput{ConfigPath: configPath}); err != nil {
+		return nil, err
+	}
+
+	return baseline, nil
+}
+
+// clone returns a copy of the config that apply can be run against without the
+// original observing the writes, or nil for a nil receiver. Only what apply
+// mutates in place needs detaching, which is the slices it replaces or appends
+// to: every pointer field it touches is reassigned rather than written through,
+// and ClientCertKeyPair is only overwritten.
+func (config *Config) clone() *Config {
+	if config == nil {
+		return nil
+	}
+
+	probe := *config
+	probe.IFaceBlackList = slices.Clone(config.IFaceBlackList)
+	probe.NATExternalIPs = slices.Clone(config.NATExternalIPs)
+	probe.DNSLabels = slices.Clone(config.DNSLabels)
+	return &probe
+}
+
 // UpdateConfig update existing configuration according to input configuration and return with the configuration
 func UpdateConfig(input ConfigInput) (*Config, error) {
 	configExists, err := fileExists(input.ConfigPath)
@@ -940,6 +1180,14 @@ func UpdateConfig(input ConfigInput) (*Config, error) {
 		return nil, fmt.Errorf("config file %s does not exist", input.ConfigPath)
 	}
 
+	// A UI that round-trips the mask GetConfig hands it back is asking to keep
+	// the stored key, not to set the mask as the new one. UpdateOrCreateConfig
+	// and DirectUpdateOrCreateConfig already collapse it; this one did not, so
+	// the same round-trip through SetConfig replaced the key with asterisks.
+	if isPreSharedKeyHidden(input.PreSharedKey) {
+		input.PreSharedKey = nil
+	}
+
 	return update(input)
 }
 
@@ -951,7 +1199,7 @@ func UpdateOrCreateConfig(input ConfigInput) (*Config, error) {
 	}
 	if !configExists {
 		log.Infof("generating new config %s", input.ConfigPath)
-		cfg, err := createNewConfig(input)
+		cfg, err := createProvisionedConfig(input)
 		if err != nil {
 			return nil, err
 		}
@@ -976,12 +1224,20 @@ func update(input ConfigInput) (*Config, error) {
 		return nil, err
 	}
 
+	// A write path is a provisioning point: a stored profile can legitimately
+	// carry no identity (a mobile logout clears the keys in place), and the
+	// next config write is what has to mint a new one. Reads leave that alone.
+	identityGenerated, err := config.EnsureIdentity()
+	if err != nil {
+		return nil, err
+	}
+
 	updated, err := config.apply(input)
 	if err != nil {
 		return nil, err
 	}
 
-	if updated {
+	if updated || identityGenerated {
 		if err := util.WriteJson(context.Background(), input.ConfigPath, config); err != nil {
 			return nil, err
 		}
@@ -990,8 +1246,8 @@ func update(input ConfigInput) (*Config, error) {
 	return config, nil
 }
 
-// GetConfig read config file and return with Config and if it was created. Errors out if it does not exist
-func GetConfig(configPath string) (*Config, error) {
+// GetExistingConfig reads and returns the config if it exists on disk. Fails otherwise.
+func GetExistingConfig(configPath string) (*Config, error) {
 	return readConfig(configPath, false)
 }
 
@@ -1074,17 +1330,27 @@ func UpdateOldManagementURL(ctx context.Context, config *Config, configPath stri
 	return newConfig, nil
 }
 
-// CreateInMemoryConfig generate a new config but do not write out it to the store
+// CreateInMemoryConfig generate a new config but do not write out it to the store.
+// It carries an identity: callers connect with what they get back.
 func CreateInMemoryConfig(input ConfigInput) (*Config, error) {
-	return createNewConfig(input)
+	return createProvisionedConfig(input)
 }
 
-// ReadConfig read config file and return with Config. If it is not exists create a new with default values
-func ReadConfig(configPath string) (*Config, error) {
+// ReadConfigOrDefault reads the profile config at configPath, or resolves the
+// default config in memory when the file does not exist. It never writes, and
+// never mints an identity — EnsureIdentity is where that happens, so the
+// caller that provisions is also the one that persists.
+func ReadConfigOrDefault(configPath string) (*Config, error) {
 	return readConfig(configPath, true)
 }
 
-// ReadConfig read config file and return with Config. If it is not exists create a new with default values
+// readConfig reads the profile config at configPath. createIfMissing resolves a
+// default config in memory when the file is absent, rather than erroring.
+//
+// Reads are pure. This used to write the config back whenever apply() had to
+// fill in a default the file was missing, which quietly made every reader a
+// writer: a gate deciding whether to refuse a request, a UI listing profiles,
+// a mobile getter reading a single preference.
 func readConfig(configPath string, createIfMissing bool) (*Config, error) {
 	configExists, err := fileExists(configPath)
 	if err != nil {
@@ -1102,12 +1368,8 @@ func readConfig(configPath string, createIfMissing bool) (*Config, error) {
 			return nil, err
 		}
 		// initialize through apply() without changes
-		if changed, err := config.apply(ConfigInput{}); err != nil {
+		if _, err := config.apply(ConfigInput{}); err != nil {
 			return nil, err
-		} else if changed {
-			if err = WriteOutConfig(configPath, config); err != nil {
-				return nil, err
-			}
 		}
 
 		return config, nil
@@ -1115,13 +1377,7 @@ func readConfig(configPath string, createIfMissing bool) (*Config, error) {
 		return nil, fmt.Errorf("config file %s does not exist", configPath)
 	}
 
-	cfg, err := createNewConfig(ConfigInput{ConfigPath: configPath})
-	if err != nil {
-		return nil, err
-	}
-
-	err = WriteOutConfig(configPath, cfg)
-	return cfg, err
+	return createNewConfig(ConfigInput{ConfigPath: configPath})
 }
 
 // WriteOutConfig write put the prepared config to the given path
@@ -1144,7 +1400,7 @@ func DirectUpdateOrCreateConfig(input ConfigInput) (*Config, error) {
 	}
 	if !configExists {
 		log.Infof("generating new config %s", input.ConfigPath)
-		cfg, err := createNewConfig(input)
+		cfg, err := createProvisionedConfig(input)
 		if err != nil {
 			return nil, err
 		}
@@ -1171,12 +1427,18 @@ func directUpdate(input ConfigInput) (*Config, error) {
 		return nil, err
 	}
 
+	// Same provisioning point as update(); see the note there.
+	identityGenerated, err := config.EnsureIdentity()
+	if err != nil {
+		return nil, err
+	}
+
 	updated, err := config.apply(input)
 	if err != nil {
 		return nil, err
 	}
 
-	if updated {
+	if updated || identityGenerated {
 		if err := util.DirectWriteJson(context.Background(), input.ConfigPath, config); err != nil {
 			return nil, err
 		}
@@ -1198,7 +1460,16 @@ func ConfigToJSON(config *Config) (string, error) {
 
 // ConfigFromJSON deserializes a JSON string to a Config struct.
 // This is useful for restoring config from alternative storage mechanisms.
-// After unmarshaling, defaults are applied to ensure the config is fully initialized.
+// After unmarshaling, defaults are applied to ensure the config is fully
+// initialized.
+//
+// The peer identity is deliberately none of its business, in either direction.
+// It does not generate one: a read cannot hand back keys that nothing will
+// write down (see ReadConfigOrDefault). Nor does it refuse a document that
+// carries none, because a config legitimately has no identity between a logout
+// and the next login — mobile logout clears both keys in place — and this is
+// also the deserializer the iOS SDK copies a config through. Whoever goes on
+// to connect is where an absent identity has to be answered.
 func ConfigFromJSON(jsonStr string) (*Config, error) {
 	config := &Config{}
 	err := json.Unmarshal([]byte(jsonStr), config)

client/internal/profilemanager/config_json_test.go

@@ -0,0 +1,44 @@
+package profilemanager
+
+import (
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+// The serialized form is how the tvOS SDK stores a profile and how the iOS SDK
+// copies one in memory, so it must round-trip whatever a profile legitimately
+// holds — including no identity at all, which is the state mobile logout leaves
+// behind when it clears both keys in place. Refusing that document here broke
+// logout, profile switching and the login that follows them.
+func TestConfigFromJSONRoundTripsALoggedOutProfile(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "exported.json")
+	stored, err := UpdateOrCreateConfig(ConfigInput{ConfigPath: path, ManagementURL: DefaultManagementURL})
+	require.NoError(t, err)
+	require.NotEmpty(t, stored.PrivateKey, "a provisioned config is the fixture this test starts from")
+	require.NotEmpty(t, stored.SSHKey)
+
+	exported, err := ConfigToJSON(stored)
+	require.NoError(t, err)
+
+	restored, err := ConfigFromJSON(exported)
+	require.NoError(t, err, "a config exported after a login must load")
+	require.Equal(t, stored.PrivateKey, restored.PrivateKey, "the restored peer is not the stored one")
+	require.Equal(t, stored.SSHKey, restored.SSHKey)
+
+	// What mobile logout leaves on disk.
+	loggedOut := stored.clone()
+	loggedOut.PrivateKey = ""
+	loggedOut.SSHKey = ""
+
+	document, err := ConfigToJSON(loggedOut)
+	require.NoError(t, err)
+
+	reloaded, err := ConfigFromJSON(document)
+	require.NoError(t, err, "a logged-out profile must still load")
+	require.Empty(t, reloaded.PrivateKey, "loading must not mint a key nothing will write down")
+	require.Empty(t, reloaded.SSHKey)
+	require.Equal(t, stored.ManagementURL.String(), reloaded.ManagementURL.String(),
+		"the rest of the profile survives the logout")
+}

client/internal/profilemanager/config_optional_fields_test.go

@@ -0,0 +1,131 @@
+package profilemanager
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"reflect"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+// optionalBoolFields lists the *bool fields of Config by name, derived from the
+// type so a field added later is covered without touching these tests.
+func optionalBoolFields() []string {
+	pointerToBool := reflect.TypeOf((*bool)(nil))
+
+	var fields []string
+	configType := reflect.TypeOf(Config{})
+	for i := range configType.NumField() {
+		field := configType.Field(i)
+		if field.Type == pointerToBool && field.Tag.Get("json") != "-" {
+			fields = append(fields, field.Name)
+		}
+	}
+	return fields
+}
+
+func requireNoUnsetOptionalBool(t *testing.T, config *Config, context string) {
+	t.Helper()
+
+	value := reflect.ValueOf(*config)
+	for _, name := range optionalBoolFields() {
+		require.False(t, value.FieldByName(name).IsNil(),
+			"%s left %s unset, so its readers have to invent a default and a diff of it compares presence instead of value", context, name)
+	}
+}
+
+// An optional bool must not be tristate. While one can be nil, true or false,
+// every reader has to invent the meaning of nil, and — the reason this test
+// exists — a diff of the config ends up comparing presence rather than value:
+// that is what made the update-settings gate refuse `netbird up` for a client
+// restating its own defaults. apply() is where a config becomes complete, so
+// the invariant belongs to it: no *bool may come out of apply() unset.
+func TestApplyLeavesNoOptionalBoolUnset(t *testing.T) {
+	require.NotEmpty(t, optionalBoolFields(), "the invariant is only meaningful while Config has optional bools")
+
+	t.Run("a config built from scratch", func(t *testing.T) {
+		config := newConfigSkeleton()
+		_, err := config.apply(ConfigInput{})
+		require.NoError(t, err)
+
+		requireNoUnsetOptionalBool(t, config, "apply on a new config")
+	})
+
+	t.Run("a config file that predates every optional field", func(t *testing.T) {
+		path := filepath.Join(t.TempDir(), "legacy.json")
+		require.NoError(t, os.WriteFile(path, []byte(`{"WgIface":"wt0"}`), 0o600))
+
+		config, err := GetExistingConfig(path)
+		require.NoError(t, err)
+
+		requireNoUnsetOptionalBool(t, config, "a read of a legacy config")
+	})
+
+	t.Run("a config file that stores them as null", func(t *testing.T) {
+		path := filepath.Join(t.TempDir(), "null.json")
+		_, err := UpdateOrCreateConfig(ConfigInput{ConfigPath: path})
+		require.NoError(t, err)
+		unsetOnDisk(t, path, optionalBoolFields()...)
+
+		config, err := GetExistingConfig(path)
+		require.NoError(t, err)
+
+		requireNoUnsetOptionalBool(t, config, "a read of a config storing nulls")
+	})
+}
+
+// The same invariant on disk: what a write leaves in the file is what the next
+// client to read it starts from, so no write may store a null.
+func TestNoWriteStoresAnUnsetOptionalBool(t *testing.T) {
+	requireNoNullOnDisk := func(t *testing.T, path string, context string) {
+		t.Helper()
+
+		raw, err := os.ReadFile(path)
+		require.NoError(t, err)
+
+		var stored map[string]json.RawMessage
+		require.NoError(t, json.Unmarshal(raw, &stored))
+
+		for _, name := range optionalBoolFields() {
+			value, present := stored[name]
+			require.True(t, present, "%s did not store %s at all", context, name)
+			require.NotEqual(t, "null", string(value), "%s stored %s as null", context, name)
+		}
+	}
+
+	t.Run("UpdateOrCreateConfig", func(t *testing.T) {
+		path := filepath.Join(t.TempDir(), "created.json")
+		_, err := UpdateOrCreateConfig(ConfigInput{ConfigPath: path, ManagementURL: DefaultManagementURL})
+		require.NoError(t, err)
+
+		requireNoNullOnDisk(t, path, "UpdateOrCreateConfig")
+	})
+
+	t.Run("UpdateConfig over a config storing nulls", func(t *testing.T) {
+		path := filepath.Join(t.TempDir(), "stored.json")
+		_, err := UpdateOrCreateConfig(ConfigInput{ConfigPath: path})
+		require.NoError(t, err)
+		unsetOnDisk(t, path, optionalBoolFields()...)
+
+		_, err = UpdateConfig(ConfigInput{ConfigPath: path, ManagementURL: "https://mgmt.example.com"})
+		require.NoError(t, err)
+
+		requireNoNullOnDisk(t, path, "UpdateConfig")
+	})
+
+	// Renaming used to copy the file back through a bare Unmarshal, which
+	// preserved the nulls a pre-fix client had written.
+	t.Run("RenameProfile", func(t *testing.T) {
+		withTestSM(t, func(sm *ServiceManager, username string) {
+			created, err := sm.AddProfile("work", username)
+			require.NoError(t, err)
+			unsetOnDisk(t, created.Path, optionalBoolFields()...)
+
+			require.NoError(t, sm.RenameProfile(created.ID, username, "office"))
+
+			requireNoNullOnDisk(t, created.Path, "RenameProfile")
+		})
+	})
+}

client/internal/profilemanager/config_probe_test.go

@@ -0,0 +1,96 @@
+package profilemanager
+
+import (
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"crypto/x509"
+	"crypto/x509/pkix"
+	"encoding/pem"
+	"math/big"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+)
+
+// writeCertPair writes a throwaway certificate and key, so apply() has
+// something real to load rather than a missing file it would only log about.
+func writeCertPair(t *testing.T) (certPath, keyPath string) {
+	t.Helper()
+
+	key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err)
+
+	template := x509.Certificate{
+		SerialNumber: big.NewInt(1),
+		Subject:      pkix.Name{CommonName: "probe-test"},
+		NotBefore:    time.Now().Add(-time.Hour),
+		NotAfter:     time.Now().Add(time.Hour),
+	}
+	der, err := x509.CreateCertificate(rand.Reader, &template, &template, &key.PublicKey, key)
+	require.NoError(t, err)
+
+	keyDER, err := x509.MarshalECPrivateKey(key)
+	require.NoError(t, err)
+
+	dir := t.TempDir()
+	certPath = filepath.Join(dir, "client.crt")
+	keyPath = filepath.Join(dir, "client.key")
+	require.NoError(t, os.WriteFile(certPath, pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: der}), 0o600))
+	require.NoError(t, os.WriteFile(keyPath, pem.EncodeToMemory(&pem.Block{Type: "EC PRIVATE KEY", Bytes: keyDER}), 0o600))
+	return certPath, keyPath
+}
+
+// The dry run behind the update-settings gate must not read the mTLS pair off
+// disk. The loaded pair feeds the connection, never the comparison, and the
+// gate runs it on every SetConfig and Login — twice per request — including the
+// ones it refuses.
+func TestProbeDoesNotLoadTheCertificatePair(t *testing.T) {
+	certPath, keyPath := writeCertPair(t)
+
+	t.Run("a real apply loads it", func(t *testing.T) {
+		config := newConfigSkeleton()
+		config.ClientCertPath, config.ClientCertKeyPath = certPath, keyPath
+
+		_, err := config.apply(ConfigInput{})
+		require.NoError(t, err)
+		require.NotNil(t, config.ClientCertKeyPair, "the connection would have no client certificate")
+	})
+
+	t.Run("a probe does not", func(t *testing.T) {
+		config := newConfigSkeleton()
+		config.ClientCertPath, config.ClientCertKeyPath = certPath, keyPath
+		config.probing = true
+
+		_, err := config.apply(ConfigInput{})
+		require.NoError(t, err)
+		require.Nil(t, config.ClientCertKeyPair, "the dry run read the certificate off disk")
+	})
+
+	// And the verdict is the same either way, which is the only thing the gate
+	// asks of the probe.
+	t.Run("the verdict is unaffected", func(t *testing.T) {
+		path := filepath.Join(t.TempDir(), "mtls.json")
+		_, err := UpdateOrCreateConfig(ConfigInput{
+			ConfigPath:        path,
+			ManagementURL:     DefaultManagementURL,
+			ClientCertPath:    certPath,
+			ClientCertKeyPath: keyPath,
+		})
+		require.NoError(t, err)
+
+		stored, err := GetExistingConfig(path)
+		require.NoError(t, err)
+
+		changed, err := stored.WouldChange(ConfigInput{ClientCertPath: certPath, ClientCertKeyPath: keyPath})
+		require.NoError(t, err)
+		require.False(t, changed, "restating the stored certificate paths is not a change")
+
+		changed, err = stored.WouldChange(ConfigInput{ClientCertPath: filepath.Join(t.TempDir(), "other.crt")})
+		require.NoError(t, err)
+		require.True(t, changed, "a different certificate path is a change")
+	})
+}

client/internal/profilemanager/config_test.go

@@ -196,7 +196,7 @@ func TestWireguardPortZeroExplicit(t *testing.T) {
 	assert.Equal(t, 0, config.WgPort, "WgPort should be 0 when explicitly set by user")
 
 	// Verify it persists
-	readConfig, err := GetConfig(configPath)
+	readConfig, err := GetExistingConfig(configPath)
 	require.NoError(t, err)
 	assert.Equal(t, 0, readConfig.WgPort, "WgPort should remain 0 after reading from file")
 }

client/internal/profilemanager/config_would_change_test.go

@@ -0,0 +1,529 @@
+package profilemanager
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"runtime"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/iface"
+	"github.com/netbirdio/netbird/shared/management/domain"
+)
+
+func seededConfig(t *testing.T) *Config {
+	t.Helper()
+
+	path := filepath.Join(t.TempDir(), "seeded.json")
+	cfg, err := UpdateOrCreateConfig(ConfigInput{
+		ConfigPath:    path,
+		ManagementURL: "https://api.netbird.io:443",
+		PreSharedKey:  strPointer("stored-key"),
+	})
+	require.NoError(t, err)
+	return cfg
+}
+
+func strPointer(s string) *string { return &s }
+
+func intPtr(i int) *int { return &i }
+
+func TestWouldChange(t *testing.T) {
+	tests := []struct {
+		name  string
+		input ConfigInput
+		want  bool
+	}{
+		{name: "empty input", input: ConfigInput{}, want: false},
+		{name: "same management URL", input: ConfigInput{ManagementURL: "https://api.netbird.io:443"}, want: false},
+		{name: "management URL without its default port", input: ConfigInput{ManagementURL: "https://api.netbird.io"}, want: false},
+		{name: "different management URL", input: ConfigInput{ManagementURL: "https://other.example:443"}, want: true},
+		{name: "same pre-shared key", input: ConfigInput{PreSharedKey: strPointer("stored-key")}, want: false},
+		{name: "redacted pre-shared key", input: ConfigInput{PreSharedKey: strPointer("**********")}, want: false},
+		{name: "different pre-shared key", input: ConfigInput{PreSharedKey: strPointer("other-key")}, want: true},
+		{name: "new interface blacklist entry", input: ConfigInput{ExtraIFaceBlackList: []string{"nb-probe0"}}, want: true},
+		{name: "blacklist entry already present", input: ConfigInput{ExtraIFaceBlackList: []string{"lo"}}, want: false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			cfg := seededConfig(t)
+
+			changed, err := cfg.WouldChange(tt.input)
+			require.NoError(t, err)
+			require.Equal(t, tt.want, changed)
+		})
+	}
+}
+
+// The dry run must not be observable on the config it is run against: it
+// decides whether a write is allowed, it does not perform one.
+func TestWouldChangeLeavesTheConfigAlone(t *testing.T) {
+	cfg := seededConfig(t)
+	blacklist := len(cfg.IFaceBlackList)
+
+	changed, err := cfg.WouldChange(ConfigInput{
+		ManagementURL:       "https://other.example:443",
+		PreSharedKey:        strPointer("other-key"),
+		ExtraIFaceBlackList: []string{"nb-probe0"},
+		DNSLabels:           domain.FromPunycodeList([]string{"probe"}),
+		NATExternalIPs:      []string{"1.2.3.4"},
+	})
+	require.NoError(t, err)
+	require.True(t, changed)
+
+	require.Equal(t, "https://api.netbird.io:443", cfg.ManagementURL.String())
+	require.Equal(t, "stored-key", cfg.PreSharedKey)
+	require.Len(t, cfg.IFaceBlackList, blacklist)
+	require.Empty(t, cfg.DNSLabels)
+	require.Empty(t, cfg.NATExternalIPs)
+}
+
+// A nil config means the profile holds nothing yet, so the baseline is what
+// the daemon would create for it.
+func TestWouldChangeWithoutAStoredConfig(t *testing.T) {
+	var cfg *Config
+
+	changed, err := cfg.WouldChange(ConfigInput{})
+	require.NoError(t, err)
+	require.False(t, changed, "a request carrying nothing cannot change anything")
+
+	changed, err = cfg.WouldChange(ConfigInput{ManagementURL: DefaultManagementURL})
+	require.NoError(t, err)
+	require.False(t, changed, "the default management URL is what would be written anyway")
+
+	changed, err = cfg.WouldChange(ConfigInput{ManagementURL: "https://other.example:443"})
+	require.NoError(t, err)
+	require.True(t, changed)
+}
+
+func TestWouldChangeReportsAnInvalidInput(t *testing.T) {
+	cfg := seededConfig(t)
+
+	_, err := cfg.WouldChange(ConfigInput{ManagementURL: "not-a-url"})
+	require.Error(t, err)
+}
+
+// Reads must not write. A config file missing a field apply() fills in (MTU,
+// here) is what used to trigger the write-back.
+func TestReadsDoNotWriteTheConfigBack(t *testing.T) {
+	denormalized := []byte(`{"WgIface":"wt0"}`)
+
+	for name, read := range map[string]func(string) (*Config, error){
+		"GetExistingConfig":   GetExistingConfig,
+		"ReadConfigOrDefault": ReadConfigOrDefault,
+	} {
+		t.Run(name, func(t *testing.T) {
+			path := filepath.Join(t.TempDir(), "profile.json")
+			require.NoError(t, os.WriteFile(path, denormalized, 0o600))
+
+			cfg, err := read(path)
+			require.NoError(t, err)
+			require.Equal(t, uint16(iface.DefaultMTU), cfg.MTU, "the returned config is still normalized in memory")
+			require.Empty(t, cfg.PrivateKey, "a read must not mint an identity either")
+
+			after, err := os.ReadFile(path)
+			require.NoError(t, err)
+			require.Equal(t, string(denormalized), string(after), "%s rewrote the config file", name)
+		})
+	}
+}
+
+// ReadConfigOrDefault resolves a default config for a profile that has no file
+// yet, and that must not create the file either.
+func TestReadConfigDoesNotCreateTheFile(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "absent.json")
+
+	cfg, err := ReadConfigOrDefault(path)
+	require.NoError(t, err)
+	require.Equal(t, DefaultManagementURL, cfg.ManagementURL.String())
+
+	_, err = os.Stat(path)
+	require.True(t, os.IsNotExist(err), "ReadConfigOrDefault created the config file")
+}
+
+// The identity is the one thing a read cannot recompute, so it is provisioned
+// on request and its caller persists it.
+func TestEnsureIdentity(t *testing.T) {
+	cfg := newConfigSkeleton()
+
+	generated, err := cfg.EnsureIdentity()
+	require.NoError(t, err)
+	require.True(t, generated)
+	require.NotEmpty(t, cfg.PrivateKey)
+	require.NotEmpty(t, cfg.SSHKey)
+
+	key := cfg.PrivateKey
+	generated, err = cfg.EnsureIdentity()
+	require.NoError(t, err)
+	require.False(t, generated, "a config that already has an identity keeps it")
+	require.Equal(t, key, cfg.PrivateKey)
+}
+
+// One endpoint written several ways is one endpoint. A gate that compared
+// spellings refused a client restating its own management URL with a trailing
+// slash, which is a normal way to write it.
+func TestSameServiceURL(t *testing.T) {
+	tests := []struct {
+		a, b string
+		want bool
+	}{
+		{a: "https://mgmt.example.com", b: "https://mgmt.example.com:443", want: true},
+		{a: "https://mgmt.example.com", b: "https://mgmt.example.com/", want: true},
+		{a: "https://mgmt.example.com/", b: "https://mgmt.example.com:443/", want: true},
+		{a: "https://MGMT.example.com", b: "https://mgmt.example.com", want: true},
+		{a: "http://mgmt.example.com", b: "http://mgmt.example.com:80", want: true},
+		{a: "https://mgmt.example.com", b: "http://mgmt.example.com", want: false},
+		{a: "https://mgmt.example.com", b: "https://mgmt.example.com:8443", want: false},
+		{a: "https://mgmt.example.com", b: "https://other.example.com", want: false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.a+" vs "+tt.b, func(t *testing.T) {
+			a, err := ParseServiceURL("a", tt.a)
+			require.NoError(t, err)
+			b, err := ParseServiceURL("b", tt.b)
+			require.NoError(t, err)
+
+			require.Equal(t, tt.want, SameServiceURL(a, b))
+			require.Equal(t, tt.want, SameServiceURL(b, a), "the comparison must be symmetric")
+		})
+	}
+}
+
+// The same spellings, through the dry run the update-settings gate uses.
+func TestWouldChangeIgnoresURLSpelling(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "seeded.json")
+	_, err := UpdateOrCreateConfig(ConfigInput{
+		ConfigPath:    path,
+		ManagementURL: "https://mgmt.example.com",
+	})
+	require.NoError(t, err)
+
+	cfg, err := GetExistingConfig(path)
+	require.NoError(t, err)
+
+	for _, spelling := range []string{
+		"https://mgmt.example.com",
+		"https://mgmt.example.com/",
+		"https://mgmt.example.com:443",
+		"https://mgmt.example.com:443/",
+		"https://MGMT.example.com",
+	} {
+		changed, err := cfg.WouldChange(ConfigInput{ManagementURL: spelling})
+		require.NoError(t, err)
+		require.False(t, changed, "%q is the stored endpoint written differently", spelling)
+	}
+
+	changed, err := cfg.WouldChange(ConfigInput{ManagementURL: "https://mgmt.example.com:8443"})
+	require.NoError(t, err)
+	require.True(t, changed, "a different port is a different endpoint")
+}
+
+// The dry-run baseline exists to be compared against and discarded, so it must
+// not mint keys — the CLI's login backoff loop would otherwise log a fresh
+// "generated new Wireguard key" on every attempt.
+func TestDryRunBaselineDoesNotGenerateKeys(t *testing.T) {
+	baseline, err := newDryRunBaseline(filepath.Join(t.TempDir(), "absent.json"))
+	require.NoError(t, err)
+
+	require.Empty(t, baseline.PrivateKey, "generated a WireGuard key for a throwaway config")
+	require.Empty(t, baseline.SSHKey, "generated an SSH key for a throwaway config")
+
+	// Everything the comparison actually looks at is still the default config.
+	require.Equal(t, DefaultManagementURL, baseline.ManagementURL.String())
+	require.Equal(t, uint16(iface.DefaultMTU), baseline.MTU)
+	require.Equal(t, iface.DefaultWgPort, baseline.WgPort)
+}
+
+// A stored profile can carry no identity — a mobile logout clears the keys in
+// place — so the next config write has to mint one, which is what keeps the
+// following login from dialing management with an empty key.
+func TestUpdateConfigProvisionsAMissingIdentity(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "logged-out.json")
+	_, err := UpdateOrCreateConfig(ConfigInput{
+		ConfigPath:    path,
+		ManagementURL: "https://api.netbird.io:443",
+	})
+	require.NoError(t, err)
+
+	// Stand in for the logout, which zeroes the keys and writes the config out.
+	loggedOut, err := GetExistingConfig(path)
+	require.NoError(t, err)
+	loggedOut.PrivateKey = ""
+	loggedOut.SSHKey = ""
+	require.NoError(t, WriteOutConfig(path, loggedOut))
+
+	cfg, err := UpdateOrCreateConfig(ConfigInput{ConfigPath: path})
+	require.NoError(t, err)
+	require.NotEmpty(t, cfg.PrivateKey, "the write path did not provision an identity")
+	require.NotEmpty(t, cfg.SSHKey)
+
+	persisted, err := GetExistingConfig(path)
+	require.NoError(t, err)
+	require.Equal(t, cfg.PrivateKey, persisted.PrivateKey, "the provisioned identity was not persisted")
+}
+
+// A config that carries no sync message version must not make the dry run
+// panic: the gate runs inside a request handler, where failing closed is the
+// worst acceptable outcome.
+func TestWouldChangeWithoutAStoredSyncMessageVersion(t *testing.T) {
+	cfg := seededConfig(t)
+	require.Nil(t, cfg.SyncMessageVersion, "the fixture is only useful while the field starts out unset")
+
+	version := 2
+	changed, err := cfg.WouldChange(ConfigInput{SyncMessageVersion: &version})
+	require.NoError(t, err)
+	require.True(t, changed)
+	require.Nil(t, cfg.SyncMessageVersion, "the dry run set the version on the stored config")
+}
+
+// Restating the certificate paths a config already holds is not a change, for
+// the same reason restating any other value is not.
+func TestWouldChangeIgnoresRestatedCertificatePaths(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "mtls.json")
+	_, err := UpdateOrCreateConfig(ConfigInput{
+		ConfigPath:        path,
+		ManagementURL:     "https://api.netbird.io:443",
+		ClientCertPath:    "/etc/netbird/client.crt",
+		ClientCertKeyPath: "/etc/netbird/client.key",
+	})
+	require.NoError(t, err)
+
+	cfg, err := GetExistingConfig(path)
+	require.NoError(t, err)
+
+	changed, err := cfg.WouldChange(ConfigInput{
+		ClientCertPath:    "/etc/netbird/client.crt",
+		ClientCertKeyPath: "/etc/netbird/client.key",
+	})
+	require.NoError(t, err)
+	require.False(t, changed, "the stored certificate paths were restated")
+
+	changed, err = cfg.WouldChange(ConfigInput{ClientCertPath: "/etc/netbird/other.crt"})
+	require.NoError(t, err)
+	require.True(t, changed, "a different certificate path is a change")
+}
+
+// A read that lands on a missing file must not hand back keys: nothing would
+// write them down, so the caller would connect with an identity that changes on
+// the next run and registers a second peer.
+func TestReadConfigOrDefaultCarriesNoIdentity(t *testing.T) {
+	cfg, err := ReadConfigOrDefault(filepath.Join(t.TempDir(), "absent.json"))
+	require.NoError(t, err)
+
+	require.Empty(t, cfg.PrivateKey, "a read minted a WireGuard key")
+	require.Empty(t, cfg.SSHKey, "a read minted an SSH key")
+
+	// So the caller's own EnsureIdentity is the one that reports the work, and
+	// therefore the one that triggers the write.
+	generated, err := cfg.EnsureIdentity()
+	require.NoError(t, err)
+	require.True(t, generated, "the provisioning caller could not tell it had to persist the identity")
+}
+
+// CreateInMemoryConfig is the opposite contract: its callers connect with what
+// they get back, so it does carry an identity.
+func TestCreateInMemoryConfigCarriesAnIdentity(t *testing.T) {
+	cfg, err := CreateInMemoryConfig(ConfigInput{ManagementURL: "https://api.netbird.io:443"})
+	require.NoError(t, err)
+
+	require.NotEmpty(t, cfg.PrivateKey)
+	require.NotEmpty(t, cfg.SSHKey)
+}
+
+// The admin panel is opened, not dialed, so its path identifies it. Comparing
+// it as a bare endpoint left a custom panel URL unable to change.
+func TestAdminURLPathIsPartOfTheIdentity(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "panel.json")
+	_, err := UpdateOrCreateConfig(ConfigInput{
+		ConfigPath: path,
+		AdminURL:   "https://app.example.com/netbird",
+	})
+	require.NoError(t, err)
+
+	cfg, err := GetExistingConfig(path)
+	require.NoError(t, err)
+	require.Equal(t, "https://app.example.com:443/netbird", cfg.AdminURL.String())
+
+	// Equivalent spellings of the same panel are still not a change.
+	for _, same := range []string{
+		"https://app.example.com/netbird",
+		"https://app.example.com:443/netbird",
+		"https://app.example.com/netbird/",
+		"https://APP.example.com/netbird",
+	} {
+		changed, err := cfg.WouldChange(ConfigInput{AdminURL: same})
+		require.NoError(t, err)
+		require.False(t, changed, "%q is the stored panel written differently", same)
+	}
+
+	// A different path is a different panel, and it must be persisted.
+	changed, err := cfg.WouldChange(ConfigInput{AdminURL: "https://app.example.com/other"})
+	require.NoError(t, err)
+	require.True(t, changed, "a different panel path is a change")
+
+	updated, err := UpdateConfig(ConfigInput{ConfigPath: path, AdminURL: "https://app.example.com/other"})
+	require.NoError(t, err)
+	require.Equal(t, "https://app.example.com:443/other", updated.AdminURL.String(), "the new panel path was not persisted")
+}
+
+// unsetOnDisk rewrites the stored config so the named fields carry a JSON null,
+// which is how a profile written before apply() resolved them looks on disk.
+// It synthesizes that state: no write produces it any more.
+func unsetOnDisk(t *testing.T, path string, fields ...string) {
+	t.Helper()
+
+	raw, err := os.ReadFile(path)
+	require.NoError(t, err)
+
+	var stored map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(raw, &stored))
+
+	for _, field := range fields {
+		_, present := stored[field]
+		require.True(t, present, "%s is not a field of the stored config", field)
+		stored[field] = json.RawMessage("null")
+	}
+
+	rewritten, err := json.Marshal(stored)
+	require.NoError(t, err)
+	require.NoError(t, os.WriteFile(path, rewritten, 0600))
+}
+
+// Seven fields mean "the effective default" when they hold no value, and every
+// profile written before apply() resolved them holds them as null. Restating
+// that default is asking for no change — and the CLI restates it on every
+// `netbird up`, because a flag set through an environment variable is a flag
+// pflag reports as Changed. Judging those restatements as changes made the
+// update-settings gate refuse `netbird up` outright for a client configured
+// through the environment, which is the shape of a Kubernetes deployment.
+//
+// A login now writes those fields set, so the fixture puts the null state back
+// on disk with unsetOnDisk instead of getting it from a login.
+func TestWouldChangeIgnoresRestatedDefaultsOfUnsetFields(t *testing.T) {
+	networkMonitorDefault := runtime.GOOS == "windows" || runtime.GOOS == "darwin"
+
+	tests := []struct {
+		field       string
+		theDefault  ConfigInput
+		theOtherWay ConfigInput
+	}{
+		{"EnableSSHRoot",
+			ConfigInput{EnableSSHRoot: boolPtr(false)}, ConfigInput{EnableSSHRoot: boolPtr(true)}},
+		{"EnableSSHSFTP",
+			ConfigInput{EnableSSHSFTP: boolPtr(false)}, ConfigInput{EnableSSHSFTP: boolPtr(true)}},
+		{"EnableSSHLocalPortForwarding",
+			ConfigInput{EnableSSHLocalPortForwarding: boolPtr(false)}, ConfigInput{EnableSSHLocalPortForwarding: boolPtr(true)}},
+		{"EnableSSHRemotePortForwarding",
+			ConfigInput{EnableSSHRemotePortForwarding: boolPtr(false)}, ConfigInput{EnableSSHRemotePortForwarding: boolPtr(true)}},
+		{"DisableSSHAuth",
+			ConfigInput{DisableSSHAuth: boolPtr(false)}, ConfigInput{DisableSSHAuth: boolPtr(true)}},
+		{"SSHJWTCacheTTL",
+			ConfigInput{SSHJWTCacheTTL: intPtr(0)}, ConfigInput{SSHJWTCacheTTL: intPtr(300)}},
+		{"NetworkMonitor",
+			ConfigInput{NetworkMonitor: boolPtr(networkMonitorDefault)}, ConfigInput{NetworkMonitor: boolPtr(!networkMonitorDefault)}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.field, func(t *testing.T) {
+			path := filepath.Join(t.TempDir(), "unset.json")
+			_, err := UpdateOrCreateConfig(ConfigInput{
+				ConfigPath:    path,
+				ManagementURL: "https://api.netbird.io:443",
+			})
+			require.NoError(t, err)
+			unsetOnDisk(t, path, tt.field)
+
+			cfg, err := GetExistingConfig(path)
+			require.NoError(t, err)
+
+			changed, err := cfg.WouldChange(tt.theDefault)
+			require.NoError(t, err)
+			require.False(t, changed, "restating the default of an unset %s was judged a change", tt.field)
+
+			// The gate still has to refuse a request that does ask for something.
+			changed, err = cfg.WouldChange(tt.theOtherWay)
+			require.NoError(t, err)
+			require.True(t, changed, "asking for a non-default %s is a change", tt.field)
+		})
+	}
+}
+
+// The verdict must not depend on where the caller got the config from. Readers
+// normalize what they hand out, but apply() signals "I filled in a default"
+// through the same bool as "the input changed something", so a config that
+// never passed through a read would otherwise report a change for an input
+// that asks for nothing.
+func TestWouldChangeNormalizesBeforeMeasuring(t *testing.T) {
+	rawConfig := func(t *testing.T) *Config {
+		t.Helper()
+
+		cfg := &Config{WgIface: iface.WgInterfaceDefault}
+		require.Nil(t, cfg.ServerSSHAllowed, "the fixture is only useful while the config is not normalized")
+		require.Nil(t, cfg.EnableSSHRoot)
+		require.Empty(t, cfg.IFaceBlackList)
+		return cfg
+	}
+
+	changed, err := rawConfig(t).WouldChange(ConfigInput{})
+	require.NoError(t, err)
+	require.False(t, changed, "an input carrying nothing cannot change anything")
+
+	changed, err = rawConfig(t).WouldChange(ConfigInput{EnableSSHRoot: boolPtr(false)})
+	require.NoError(t, err)
+	require.False(t, changed, "the default of a field the config never held is not a change")
+
+	changed, err = rawConfig(t).WouldChange(ConfigInput{EnableSSHRoot: boolPtr(true)})
+	require.NoError(t, err)
+	require.True(t, changed, "a non-default value is still a change")
+}
+
+// A zero-padded port addresses the same port. The normalization itself belongs
+// to util.ServiceURLPort and is tested there; this asserts that the comparison
+// this package hands its callers inherits it.
+func TestServiceURLPortIsNormalizedNumerically(t *testing.T) {
+	padded, err := ParseServiceURL("padded", "https://mgmt.example.com:0443")
+	require.NoError(t, err)
+	plain, err := ParseServiceURL("plain", "https://mgmt.example.com:443")
+	require.NoError(t, err)
+
+	require.True(t, SameServiceURL(padded, plain))
+}
+
+// A list the profile does not have and a list the request empties are the same
+// thing: no NAT mappings, no DNS labels. The profile stores an absent list as
+// JSON null and reads it back as a nil slice, while `netbird up` sends the
+// emptied list — CleanNATExternalIPs / CleanDNSLabels — whenever the matching
+// environment variable is set to nothing, which a deployment template does by
+// default. Judging nil and empty as different made the gate refuse that start,
+// which is the very deadlock this branch exists to remove, on another field.
+func TestWouldChangeIgnoresAnEmptiedListThatWasAlreadyAbsent(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "lists.json")
+	_, err := UpdateOrCreateConfig(ConfigInput{ConfigPath: path, ManagementURL: DefaultManagementURL})
+	require.NoError(t, err)
+
+	stored, err := GetExistingConfig(path)
+	require.NoError(t, err)
+	require.Nil(t, stored.NATExternalIPs, "the fixture is only useful while the stored list is absent")
+	require.Nil(t, stored.DNSLabels)
+
+	changed, err := stored.WouldChange(ConfigInput{NATExternalIPs: make([]string, 0)})
+	require.NoError(t, err)
+	require.False(t, changed, "emptying a NAT list the profile never had is not a change")
+
+	changed, err = stored.WouldChange(ConfigInput{DNSLabels: domain.List{}})
+	require.NoError(t, err)
+	require.False(t, changed, "emptying a DNS label list the profile never had is not a change")
+
+	// A list that does hold something still moves when the request empties it.
+	withEntries, err := UpdateConfig(ConfigInput{ConfigPath: path, NATExternalIPs: []string{"1.2.3.4"}})
+	require.NoError(t, err)
+	require.Equal(t, []string{"1.2.3.4"}, withEntries.NATExternalIPs)
+
+	changed, err = withEntries.WouldChange(ConfigInput{NATExternalIPs: make([]string, 0)})
+	require.NoError(t, err)
+	require.True(t, changed, "clearing a NAT list that had an entry is a change")
+}

client/internal/profilemanager/service.go

@@ -313,7 +313,11 @@ func (s *ServiceManager) AddProfile(displayName, username string) (*Profile, err
 	}
 
 	profPath := filepath.Join(configDir, id.String()+".json")
-	cfg, err := createNewConfig(ConfigInput{ConfigPath: profPath})
+	// Provisioned, not bare: this config goes straight to disk, and a profile
+	// file with no identity is one whose first reader has to mint the keys and
+	// remember to write them back. Before identity generation moved out of
+	// apply() into EnsureIdentity, createNewConfig produced them here too.
+	cfg, err := createProvisionedConfig(ConfigInput{ConfigPath: profPath})
 	if err != nil {
 		return nil, fmt.Errorf("failed to create new config: %w", err)
 	}
@@ -330,6 +334,19 @@ func (s *ServiceManager) AddProfile(displayName, username string) (*Profile, err
 	}, nil
 }
 
+// RenameProfile changes a profile's display name. It rewrites the whole
+// profile file, not just the name: the config is read through the normalizing
+// reader, so apply()'s resolved values — the optional booleans, the interface
+// blacklist, the DNS route interval — are persisted along with the new name.
+//
+// That is deliberate. A write that skipped apply() is what left profiles on
+// disk carrying null where a value was meant, and made a diff of the config
+// compare presence instead of value. Two consequences worth knowing: the
+// platform-dependent defaults resolved here are the renaming host's
+// (ServerSSHAllowed and the network monitor differ per OS), and a profile
+// whose stored name does not survive sanitizeDisplayName now fails to rename
+// rather than being rewritten — though apply() rejects such a profile on every
+// other read too, so it was already unusable.
 func (s *ServiceManager) RenameProfile(id ID, username string, newName string) error {
 	displayName, err := sanitizeDisplayName(newName)
 	if err != nil {
@@ -356,17 +373,17 @@ func (s *ServiceManager) RenameProfile(id ID, username string, newName string) e
 		return ErrProfileNotFound
 	}
 
-	data, err := os.ReadFile(target.Path)
+	// Through the reader, not a bare Unmarshal: this was the one write that
+	// skipped apply(), so it copied back whatever the file held — including an
+	// optional field left unset, which every other write resolves to its
+	// default. Renaming a profile is a poor place to leave that behind.
+	cfg, err := GetExistingConfig(target.Path)
 	if err != nil {
-		return err
-	}
-	var cfg Config
-	if err := json.Unmarshal(data, &cfg); err != nil {
-		return err
+		return fmt.Errorf("read profile config: %w", err)
 	}
 	cfg.Name = displayName
 
-	if err := util.WriteJson(context.Background(), target.Path, cfg); err != nil {
+	if err := WriteOutConfig(target.Path, cfg); err != nil {
 		return fmt.Errorf("failed to write profile name: %w", err)
 	}
 	return nil

client/internal/profilemanager/service_test.go

@@ -228,3 +228,27 @@ func TestRemoveProfile_DeletesStateFile(t *testing.T) {
 		assert.True(t, errors.Is(err, os.ErrNotExist), "state file should be removed")
 	})
 }
+
+// A profile file is written here and read back by whoever connects with it, so
+// it has to carry the peer's identity. While AddProfile used the bare
+// constructor, it wrote a config with no keys: the first reader had to mint
+// them, and the paths that read without writing — a gate deciding whether to
+// refuse a request, the mobile SDKs loading a stored profile — got a config
+// that cannot connect.
+func TestAddProfileWritesAnIdentity(t *testing.T) {
+	withTestSM(t, func(sm *ServiceManager, username string) {
+		created, err := sm.AddProfile("work", username)
+		require.NoError(t, err)
+
+		stored, err := GetExistingConfig(created.Path)
+		require.NoError(t, err)
+
+		require.NotEmpty(t, stored.PrivateKey, "the profile was written without a WireGuard key")
+		require.NotEmpty(t, stored.SSHKey, "the profile was written without an SSH key")
+
+		// And the identity is the one on disk, not one minted per read.
+		reread, err := GetExistingConfig(created.Path)
+		require.NoError(t, err)
+		require.Equal(t, stored.PrivateKey, reread.PrivateKey)
+	})
+}

client/internal/relay/relay.go

@@ -201,11 +201,7 @@ func (p *StunTurnProbe) probeSTUN(ctx context.Context, uri *stun.URI) (addr stri
 		}
 	}()
 
-	net, err := stdnet.NewNet(ctx, nil)
-	if err != nil {
-		probeErr = fmt.Errorf("new net: %w", err)
-		return
-	}
+	net := stdnet.NewNet(ctx, nil, nil)
 
 	client, err := stun.DialURI(uri, &stun.DialConfig{
 		Net: net,
@@ -290,11 +286,7 @@ func (p *StunTurnProbe) probeTURN(ctx context.Context, uri *stun.URI) (addr stri
 		}
 	}()
 
-	net, err := stdnet.NewNet(ctx, nil)
-	if err != nil {
-		probeErr = fmt.Errorf("new net: %w", err)
-		return
-	}
+	net := stdnet.NewNet(ctx, nil, nil)
 	cfg := &turn.ClientConfig{
 		STUNServerAddr: turnServerAddr,
 		TURNServerAddr: turnServerAddr,

client/internal/routemanager/client/client.go

@@ -294,6 +294,7 @@ func (w *Watcher) addAllowedIPs(route *route.Route) error {
 		return fmt.Errorf("add allowed IPs for peer %s: %w", route.Peer, err)
 	}
 
+	w.statusRecorder.AddActiveRoutePeer(route.GetHAUniqueID(), route.Peer)
 	if err := w.statusRecorder.AddPeerStateRoute(route.Peer, w.handler.String(), route.GetResourceID()); err != nil {
 		log.Warnf("Failed to update peer state: %v", err)
 	}
@@ -303,6 +304,7 @@ func (w *Watcher) addAllowedIPs(route *route.Route) error {
 }
 
 func (w *Watcher) removeAllowedIPs(route *route.Route, rsn reason) error {
+	w.statusRecorder.RemoveActiveRoutePeer(route.GetHAUniqueID())
 	if err := w.statusRecorder.RemovePeerStateRoute(route.Peer, w.handler.String()); err != nil {
 		log.Warnf("Failed to update peer state: %v", err)
 	}

client/internal/routemanager/ipfwdstate/ipfwdstate.go

@@ -19,8 +19,7 @@ type IPForwardingState struct {
 
 	// routingV4/routingV6 track whether the routing path currently holds a
 	// reference, so repeated EnableRouting calls (one per network-map update)
-	// hold at most one reference per family and an unpaired DisableRouting
-	// can't release references held by DNAT rules.
+	// hold at most one reference per family.
 	routingV4 bool
 	routingV6 bool
 
@@ -95,31 +94,6 @@ func (f *IPForwardingState) ReleaseRouting() error {
 	return nil
 }
 
-// RequestForwarding enables the family's forwarding sysctl on first request.
-func (f *IPForwardingState) RequestForwarding(v6 bool) error {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-
-	if v6 {
-		return f.requestV6()
-	}
-	return f.requestV4()
-}
-
-// ReleaseForwarding decrements the family counter. The last v6 release restores
-// what enable captured. v4 stays on: net.ipv4.ip_forward is co-owned by other
-// tooling (docker, k8s, libvirt).
-func (f *IPForwardingState) ReleaseForwarding(v6 bool) error {
-	f.mu.Lock()
-	defer f.mu.Unlock()
-
-	if v6 {
-		return f.releaseV6()
-	}
-	f.releaseV4()
-	return nil
-}
-
 func (f *IPForwardingState) requestV4() error {
 	if f.v4Count == 0 {
 		if err := systemops.EnableV4IPForwarding(); err != nil {

client/internal/routemanager/ipfwdstate/ipfwdstate_privileged_linux_test.go

@@ -10,8 +10,7 @@ import (
 )
 
 // TestRequestRoutingV6ToV4Transition verifies that a v4-only routing request
-// releases a previously held routing-owned v6 reference without touching
-// references held by DNAT rules.
+// releases a previously held routing-owned v6 reference.
 func TestRequestRoutingV6ToV4Transition(t *testing.T) {
 	f := NewIPForwardingState("wt-fwd-test")
 
@@ -25,13 +24,6 @@ func TestRequestRoutingV6ToV4Transition(t *testing.T) {
 	assert.Equal(t, 1, v4, "v4 reference kept")
 	assert.Equal(t, 0, v6, "routing-owned v6 reference released")
 
-	// A DNAT-held reference survives a v4-only routing request.
-	require.NoError(t, f.RequestForwarding(true), "dnat v6 reference")
-	require.NoError(t, f.RequestRouting(false), "repeat v4-only request")
-	_, v6 = f.Counts()
-	assert.Equal(t, 1, v6, "dnat-held v6 reference survives")
-	require.NoError(t, f.ReleaseForwarding(true), "release dnat v6 reference")
-
 	require.NoError(t, f.ReleaseRouting(), "release routing")
 	v4, v6 = f.Counts()
 	assert.Equal(t, 0, v4, "all v4 references released")

client/internal/routemanager/manager_test.go

@@ -8,6 +8,7 @@ import (
 	"net/netip"
 	"testing"
 
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
 	"github.com/netbirdio/netbird/client/internal/stdnet"
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 
@@ -406,10 +407,7 @@ func TestManagerUpdateRoutes(t *testing.T) {
 	for n, testCase := range testCases {
 		t.Run(testCase.name, func(t *testing.T) {
 			peerPrivateKey, _ := wgtypes.GeneratePrivateKey()
-			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:    fmt.Sprintf("utun43%d", n),
 				Address:      wgaddr.MustParseWGAddress("100.65.65.2/24"),

client/internal/routemanager/systemops/systemops_generic_test.go

@@ -15,6 +15,7 @@ import (
 	"syscall"
 	"testing"
 
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
 	"github.com/netbirdio/netbird/client/internal/stdnet"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -436,8 +437,7 @@ func createWGInterface(t *testing.T, interfaceName, ipAddressCIDR string, listen
 	peerPrivateKey, err := wgtypes.GeneratePrivateKey()
 	require.NoError(t, err)
 
-	newNet, err := stdnet.NewNet(context.Background(), nil)
-	require.NoError(t, err)
+	newNet := stdnet.NewNet(context.Background(), profilemanager.DefaultInterfaceBlacklist, nil)
 
 	opts := iface.WGIFaceOpts{
 		IFaceName:    interfaceName,

client/internal/stdnet/filter.go

@@ -3,17 +3,13 @@ package stdnet
 import (
 	"runtime"
 	"strings"
-
-	log "github.com/sirupsen/logrus"
-	"golang.zx2c4.com/wireguard/wgctrl"
 )
 
 // InterfaceFilter is a function passed to ICE Agent to filter out not allowed interfaces
-// to avoid building tunnel over them.
-func InterfaceFilter(disallowList []string) func(string) bool {
-
+// to avoid building tunnel over them. A nil detector probes the interface on every call,
+// which is what the callers that build one filter for their whole lifetime want.
+func InterfaceFilter(disallowList []string, detector *WGDetector) func(string) bool {
 	return func(iFace string) bool {
-
 		if strings.HasPrefix(iFace, "lo") {
 			// hardcoded loopback check to support already installed agents
 			return false
@@ -24,17 +20,8 @@ func InterfaceFilter(disallowList []string) func(string) bool {
 				return false
 			}
 		}
-		// look for unlisted WireGuard interfaces
-		wg, err := wgctrl.New()
-		if err != nil {
-			log.Debugf("trying to create a wgctrl client failed with: %v", err)
-			return true
-		}
-		defer func() {
-			_ = wg.Close()
-		}()
 
-		_, err = wg.Device(iFace)
-		return err != nil
+		// look for unlisted WireGuard interfaces
+		return !detector.IsWireGuard(iFace)
 	}
 }

client/internal/stdnet/stdnet.go

@@ -45,12 +45,12 @@ type Net struct {
 }
 
 // NewNetWithDiscover creates a new StdNet instance.
-func NewNetWithDiscover(ctx context.Context, iFaceDiscover ExternalIFaceDiscover, disallowList []string) (*Net, error) {
+func NewNetWithDiscover(ctx context.Context, iFaceDiscover ExternalIFaceDiscover, disallowList []string, detector *WGDetector) *Net {
 	if ctx == nil {
 		ctx = context.Background()
 	}
 	n := &Net{
-		interfaceFilter: InterfaceFilter(disallowList),
+		interfaceFilter: InterfaceFilter(disallowList, detector),
 		ctx:             ctx,
 	}
 	// current ExternalIFaceDiscover implement in android-client https://github.dev/netbirdio/android-client
@@ -60,20 +60,19 @@ func NewNetWithDiscover(ctx context.Context, iFaceDiscover ExternalIFaceDiscover
 	} else {
 		n.iFaceDiscover = newMobileIFaceDiscover(iFaceDiscover)
 	}
-	return n, n.UpdateInterfaces()
+	return n
 }
 
 // NewNet creates a new StdNet instance.
-func NewNet(ctx context.Context, disallowList []string) (*Net, error) {
+func NewNet(ctx context.Context, disallowList []string, detector *WGDetector) *Net {
 	if ctx == nil {
 		ctx = context.Background()
 	}
-	n := &Net{
+	return &Net{
 		iFaceDiscover:   pionDiscover{},
-		interfaceFilter: InterfaceFilter(disallowList),
+		interfaceFilter: InterfaceFilter(disallowList, detector),
 		ctx:             ctx,
 	}
-	return n, n.UpdateInterfaces()
 }
 
 // resolveAddr performs DNS resolution with context support and timeout.
@@ -122,45 +121,18 @@ func (n *Net) resolveAddr(network, address string) (netip.AddrPort, error) {
 	return netip.AddrPortFrom(addrs[0], uint16(port)), nil
 }
 
-// UpdateInterfaces updates the internal list of network interfaces
-// and associated addresses filtering them by name.
-// The interfaces are discovered by an external iFaceDiscover function or by a default discoverer if the external one
-// wasn't specified.
-func (n *Net) UpdateInterfaces() (err error) {
-	n.mu.Lock()
-	defer n.mu.Unlock()
-
-	return n.updateInterfaces()
-}
-
-func (n *Net) updateInterfaces() (err error) {
-	allIfaces, err := n.iFaceDiscover.iFaces()
-	if err != nil {
-		return err
-	}
-
-	n.interfaces = n.filterInterfaces(allIfaces)
-
-	n.lastUpdate = time.Now()
-
-	return nil
-}
-
 // Interfaces returns a slice of interfaces which are available on the
 // system
 func (n *Net) Interfaces() ([]*transport.Interface, error) {
 	n.mu.Lock()
 	defer n.mu.Unlock()
 
-	if time.Since(n.lastUpdate) < updateInterval {
-		return slices.Clone(n.interfaces), nil
+	iFaces, err := n.freshInterfacesLocked()
+	if err != nil {
+		return nil, err
 	}
 
-	if err := n.updateInterfaces(); err != nil {
-		return nil, fmt.Errorf("update interfaces: %w", err)
-	}
-
-	return slices.Clone(n.interfaces), nil
+	return slices.Clone(iFaces), nil
 }
 
 // InterfaceByIndex returns the interface specified by index.
@@ -171,7 +143,13 @@ func (n *Net) Interfaces() ([]*transport.Interface, error) {
 func (n *Net) InterfaceByIndex(index int) (*transport.Interface, error) {
 	n.mu.Lock()
 	defer n.mu.Unlock()
-	for _, ifc := range n.interfaces {
+
+	iFaces, err := n.freshInterfacesLocked()
+	if err != nil {
+		return nil, err
+	}
+
+	for _, ifc := range iFaces {
 		if ifc.Index == index {
 			return ifc, nil
 		}
@@ -184,7 +162,13 @@ func (n *Net) InterfaceByIndex(index int) (*transport.Interface, error) {
 func (n *Net) InterfaceByName(name string) (*transport.Interface, error) {
 	n.mu.Lock()
 	defer n.mu.Unlock()
-	for _, ifc := range n.interfaces {
+
+	iFaces, err := n.freshInterfacesLocked()
+	if err != nil {
+		return nil, err
+	}
+
+	for _, ifc := range iFaces {
 		if ifc.Name == name {
 			return ifc, nil
 		}
@@ -193,6 +177,31 @@ func (n *Net) InterfaceByName(name string) (*transport.Interface, error) {
 	return nil, fmt.Errorf("%w: %s", transport.ErrInterfaceNotFound, name)
 }
 
+func (n *Net) freshInterfacesLocked() ([]*transport.Interface, error) {
+	if time.Since(n.lastUpdate) < updateInterval {
+		return n.interfaces, nil
+	}
+
+	if err := n.updateInterfacesLocked(); err != nil {
+		return nil, fmt.Errorf("update interfaces: %w", err)
+	}
+
+	return n.interfaces, nil
+}
+
+func (n *Net) updateInterfacesLocked() error {
+	allIFaces, err := n.iFaceDiscover.iFaces()
+	if err != nil {
+		return err
+	}
+
+	n.interfaces = n.filterInterfaces(allIFaces)
+
+	n.lastUpdate = time.Now()
+
+	return nil
+}
+
 func (n *Net) filterInterfaces(interfaces []*transport.Interface) []*transport.Interface {
 	if n.interfaceFilter == nil {
 		return interfaces

client/internal/stdnet/stdnet_test.go

@@ -0,0 +1,136 @@
+package stdnet
+
+import (
+	"context"
+	"errors"
+	"net"
+	"testing"
+
+	"github.com/pion/transport/v3"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+type countingDiscover struct {
+	calls int
+	list  []*transport.Interface
+	err   error
+}
+
+func (d *countingDiscover) iFaces() ([]*transport.Interface, error) {
+	d.calls++
+	if d.err != nil {
+		return nil, d.err
+	}
+	return d.list, nil
+}
+
+func newTestNet(t *testing.T, d iFaceDiscover) *Net {
+	t.Helper()
+	return &Net{
+		iFaceDiscover: d,
+		ctx:           context.Background(),
+	}
+}
+
+func testIFace(index int, name string) *transport.Interface {
+	return transport.NewInterface(net.Interface{Index: index, Name: name})
+}
+
+func TestNet_InterfacesDiscoversLazilyAndCaches(t *testing.T) {
+	d := &countingDiscover{list: []*transport.Interface{testIFace(1, "eth0")}}
+	n := newTestNet(t, d)
+
+	require.Zero(t, d.calls, "construction must not discover interfaces")
+
+	iFaces, err := n.Interfaces()
+	require.NoError(t, err)
+	require.Len(t, iFaces, 1)
+	assert.Equal(t, 1, d.calls)
+
+	_, err = n.Interfaces()
+	require.NoError(t, err)
+	assert.Equal(t, 1, d.calls)
+}
+
+func TestNewNet_DoesNotDiscoverAtConstruction(t *testing.T) {
+	n := NewNet(context.Background(), nil, nil)
+	require.NotNil(t, n)
+	assert.True(t, n.lastUpdate.IsZero(), "constructor must leave the cache cold")
+}
+
+func TestNewNetWithDiscover_DoesNotDiscoverAtConstruction(t *testing.T) {
+	n := NewNetWithDiscover(context.Background(), nil, nil, nil)
+	require.NotNil(t, n)
+	assert.True(t, n.lastUpdate.IsZero(), "constructor must leave the cache cold")
+}
+
+func TestNet_InterfacesRetryAfterDiscoveryFailure(t *testing.T) {
+	discoverErr := errors.New("discover failed")
+	d := &countingDiscover{err: discoverErr}
+	n := newTestNet(t, d)
+
+	_, err := n.Interfaces()
+	require.ErrorIs(t, err, discoverErr)
+
+	d.err = nil
+	d.list = []*transport.Interface{testIFace(1, "eth0")}
+
+	iFaces, err := n.Interfaces()
+	require.NoError(t, err)
+	require.Len(t, iFaces, 1)
+	assert.Equal(t, 2, d.calls)
+}
+
+func TestNet_InterfaceByNameRefreshes(t *testing.T) {
+	d := &countingDiscover{list: []*transport.Interface{testIFace(3, "eth0")}}
+	n := newTestNet(t, d)
+
+	ifc, err := n.InterfaceByName("eth0")
+	require.NoError(t, err)
+	assert.Equal(t, "eth0", ifc.Name)
+	assert.Equal(t, 1, d.calls)
+
+	_, err = n.InterfaceByName("nope")
+	require.ErrorIs(t, err, transport.ErrInterfaceNotFound)
+}
+
+func TestNet_InterfaceByIndexRefreshes(t *testing.T) {
+	d := &countingDiscover{list: []*transport.Interface{testIFace(3, "eth0")}}
+	n := newTestNet(t, d)
+
+	ifc, err := n.InterfaceByIndex(3)
+	require.NoError(t, err)
+	assert.Equal(t, "eth0", ifc.Name)
+	assert.Equal(t, 1, d.calls)
+
+	_, err = n.InterfaceByIndex(99)
+	require.ErrorIs(t, err, transport.ErrInterfaceNotFound)
+}
+
+func TestNet_InterfaceLookupPropagatesDiscoveryError(t *testing.T) {
+	discoverErr := errors.New("discover failed")
+	n := newTestNet(t, &countingDiscover{err: discoverErr})
+
+	_, err := n.InterfaceByName("eth0")
+	require.ErrorIs(t, err, discoverErr)
+
+	_, err = n.InterfaceByIndex(1)
+	require.ErrorIs(t, err, discoverErr)
+}
+
+func TestNet_InterfacesReturnsCopy(t *testing.T) {
+	d := &countingDiscover{list: []*transport.Interface{testIFace(1, "eth0")}}
+	n := newTestNet(t, d)
+
+	iFaces, err := n.Interfaces()
+	require.NoError(t, err)
+	require.Len(t, iFaces, 1)
+
+	iFaces[0] = testIFace(2, "tampered")
+
+	iFaces, err = n.Interfaces()
+	require.NoError(t, err)
+	require.Len(t, iFaces, 1)
+	assert.Equal(t, "eth0", iFaces[0].Name)
+}

client/internal/stdnet/wgdetector.go

@@ -0,0 +1,119 @@
+package stdnet
+
+import (
+	"sync"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/sync/singleflight"
+	"golang.zx2c4.com/wireguard/wgctrl"
+)
+
+// wgDetectorTTL bounds how long a cached answer is trusted. An interface rarely
+// becomes, or stops being, a WireGuard device, and the window only has to be short
+// enough that ICE does not keep gathering candidates on one that just appeared.
+const wgDetectorTTL = 1 * time.Second
+
+type wgDetectorEntry struct {
+	isWireGuard bool
+	expireAt    time.Time
+}
+
+// WGDetector answers whether an interface is a WireGuard device, remembering the
+// answer for a short while.
+//
+// The question is asked once per interface for every ICE agent, and an agent is
+// created per peer connection attempt, so on a large network the uncached form runs
+// constantly. Answering it means opening a wgctrl client, which builds both a kernel
+// and a userspace client and resolves the netlink family, and then a round trip that
+// usually just reports the device does not exist.
+//
+// A detector is safe for concurrent use and is meant to be shared by every agent.
+type WGDetector struct {
+	ttl time.Duration
+	// probe is replaced in tests; it is the call this type exists to avoid repeating.
+	probe func(string) bool
+
+	mu    sync.RWMutex
+	cache map[string]wgDetectorEntry
+
+	sf singleflight.Group
+}
+
+// NewWGDetector returns a detector with the default time to live.
+func NewWGDetector() *WGDetector {
+	return &WGDetector{
+		ttl:   wgDetectorTTL,
+		probe: probeWireGuard,
+		cache: make(map[string]wgDetectorEntry),
+	}
+}
+
+// IsWireGuard reports whether the named interface is a WireGuard device. An interface
+// it cannot ask about is reported as not being one, which leaves it available to ICE
+// exactly as an uncached probe would.
+func (d *WGDetector) IsWireGuard(iFace string) bool {
+	if d == nil {
+		return probeWireGuard(iFace)
+	}
+
+	if isWireGuard, ok := d.cached(iFace); ok {
+		return isWireGuard
+	}
+
+	result, _, _ := d.sf.Do(iFace, func() (interface{}, error) {
+		// A caller that saw the entry expire may get here after another caller already
+		// refreshed it and left the singleflight group.
+		if isWireGuard, ok := d.cached(iFace); ok {
+			return isWireGuard, nil
+		}
+
+		isWireGuard := d.probe(iFace)
+
+		d.store(iFace, isWireGuard)
+
+		return isWireGuard, nil
+	})
+	return result.(bool)
+}
+
+func (d *WGDetector) cached(iFace string) (isWireGuard, ok bool) {
+	d.mu.RLock()
+	defer d.mu.RUnlock()
+
+	entry, found := d.cache[iFace]
+	if !found || !time.Now().Before(entry.expireAt) {
+		return false, false
+	}
+	return entry.isWireGuard, true
+}
+
+// store records an answer and drops the expired ones, so names of interfaces that came
+// and went, such as container veths, do not pile up for the life of the engine.
+func (d *WGDetector) store(iFace string, isWireGuard bool) {
+	now := time.Now()
+
+	d.mu.Lock()
+	defer d.mu.Unlock()
+
+	for name, entry := range d.cache {
+		if !now.Before(entry.expireAt) {
+			delete(d.cache, name)
+		}
+	}
+	d.cache[iFace] = wgDetectorEntry{isWireGuard: isWireGuard, expireAt: now.Add(d.ttl)}
+}
+
+func probeWireGuard(iFace string) bool {
+	wg, err := wgctrl.New()
+	if err != nil {
+		log.Debugf("trying to create a wgctrl client failed with: %v", err)
+		return false
+	}
+	defer func() {
+		_ = wg.Close()
+	}()
+
+	_, err = wg.Device(iFace)
+	return err == nil
+}

client/internal/stdnet/wgdetector_test.go

@@ -0,0 +1,152 @@
+package stdnet
+
+import (
+	"runtime"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// newCountingDetector returns a detector whose probe records how often it ran, so a test
+// can assert on the thing this type exists for rather than on its return value alone.
+func newCountingDetector(t *testing.T, ttl time.Duration, answer bool) (*WGDetector, *atomic.Int64) {
+	t.Helper()
+
+	var calls atomic.Int64
+	d := &WGDetector{
+		ttl:   ttl,
+		cache: make(map[string]wgDetectorEntry),
+		probe: func(string) bool {
+			calls.Add(1)
+			return answer
+		},
+	}
+	return d, &calls
+}
+
+func TestWGDetectorAsksOncePerInterfaceWithinTheTTL(t *testing.T) {
+	d, calls := newCountingDetector(t, time.Minute, true)
+
+	for i := 0; i < 20; i++ {
+		assert.True(t, d.IsWireGuard("wt0"), "cached answer must not change")
+	}
+	assert.Equal(t, int64(1), calls.Load(), "the interface must be probed once within the TTL")
+
+	d.IsWireGuard("eth0")
+	assert.Equal(t, int64(2), calls.Load(), "a different interface is a different question and is probed on its own")
+}
+
+func TestWGDetectorReprobesAfterTheTTL(t *testing.T) {
+	d, calls := newCountingDetector(t, time.Millisecond, true)
+
+	require.True(t, d.IsWireGuard("wt0"), "first answer")
+	require.Equal(t, int64(1), calls.Load(), "first call probes")
+
+	time.Sleep(5 * time.Millisecond)
+
+	require.True(t, d.IsWireGuard("wt0"), "answer after expiry")
+	assert.Equal(t, int64(2), calls.Load(), "an expired entry must be probed again")
+}
+
+func TestWGDetectorCollapsesConcurrentProbes(t *testing.T) {
+	var calls atomic.Int64
+	release := make(chan struct{})
+	d := &WGDetector{
+		ttl:   time.Minute,
+		cache: make(map[string]wgDetectorEntry),
+		probe: func(string) bool {
+			calls.Add(1)
+			<-release
+			return true
+		},
+	}
+
+	var wg sync.WaitGroup
+	for i := 0; i < 50; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			d.IsWireGuard("wt0")
+		}()
+	}
+
+	// The sleep only lets the callers pile up on the blocked probe so the collapse is
+	// exercised. The count does not depend on it: a caller that arrives after the probe
+	// finished finds the fresh entry, either before or inside the singleflight group.
+	time.Sleep(20 * time.Millisecond)
+	close(release)
+	wg.Wait()
+
+	assert.Equal(t, int64(1), calls.Load(), "concurrent callers must share one probe")
+}
+
+func TestWGDetectorNilProbesEveryTime(t *testing.T) {
+	var d *WGDetector
+	// A nil detector keeps the uncached behaviour, which is what the callers that build one
+	// filter for their whole lifetime rely on. It must not panic.
+	assert.NotPanics(t, func() { d.IsWireGuard("definitely-not-an-interface-0") },
+		"a nil detector must fall back to probing")
+}
+
+func TestInterfaceFilter(t *testing.T) {
+	wgDetector, calls := newCountingDetector(t, time.Minute, true)
+	plainDetector, _ := newCountingDetector(t, time.Minute, false)
+
+	t.Run("loopback is rejected without probing", func(t *testing.T) {
+		filter := InterfaceFilter(nil, wgDetector)
+		assert.False(t, filter("lo"), "loopback must never be offered to ICE")
+		assert.Equal(t, int64(0), calls.Load(), "a name settled by prefix must not reach the probe")
+	})
+
+	t.Run("a disallowed interface is rejected without probing", func(t *testing.T) {
+		if runtime.GOOS == "ios" {
+			t.Skip("the disallow list is not applied on iOS")
+		}
+		filter := InterfaceFilter([]string{"wt"}, wgDetector)
+		assert.False(t, filter("wt0"), "a blacklisted interface must be rejected")
+		assert.Equal(t, int64(0), calls.Load(), "a name settled by the disallow list must not reach the probe")
+	})
+
+	t.Run("an unlisted WireGuard interface is rejected", func(t *testing.T) {
+		filter := InterfaceFilter(nil, wgDetector)
+		assert.False(t, filter("somewg0"), "a WireGuard interface must not be used to build a tunnel")
+	})
+
+	t.Run("an ordinary interface is allowed", func(t *testing.T) {
+		filter := InterfaceFilter(nil, plainDetector)
+		assert.True(t, filter("eth0"), "a plain interface must remain available to ICE")
+	})
+}
+
+func TestInterfaceFilterSharesOneProbeAcrossFilters(t *testing.T) {
+	d, calls := newCountingDetector(t, time.Minute, false)
+
+	// Every ICE agent builds its own filter, twice, and each one is asked about every
+	// interface. Sharing the detector is what keeps that from repeating the probe.
+	for i := 0; i < 10; i++ {
+		filter := InterfaceFilter(nil, d)
+		require.True(t, filter("eth0"), "a plain interface stays allowed")
+		require.True(t, filter("eth1"), "a plain interface stays allowed")
+	}
+
+	assert.Equal(t, int64(2), calls.Load(), "one probe per interface, not per filter")
+}
+
+func TestWGDetectorDropsExpiredEntries(t *testing.T) {
+	d, _ := newCountingDetector(t, time.Millisecond, false)
+
+	for _, name := range []string{"veth1", "veth2", "veth3"} {
+		d.IsWireGuard(name)
+	}
+	time.Sleep(5 * time.Millisecond)
+	d.IsWireGuard("eth0")
+
+	d.mu.RLock()
+	defer d.mu.RUnlock()
+	assert.Len(t, d.cache, 1, "expired entries of vanished interfaces must be dropped")
+	assert.Contains(t, d.cache, "eth0", "the fresh answer must be kept")
+}

client/internal/updater/manager.go

@@ -21,8 +21,16 @@ const (
 	latestVersion = "latest"
 )
 
+const (
+	modeUndecided updateMode = iota
+	modeDownloadOnly
+	modeManaged
+)
+
 var errNoUpdateState = errors.New("no update state found")
 
+type updateMode int
+
 type UpdateState struct {
 	PreUpdateVersion string
 	TargetVersion    string
@@ -36,8 +44,9 @@ type Manager struct {
 	statusRecorder *peer.Status
 	stateManager   *statemanager.Manager
 
-	downloadOnly bool // true when no enforcement from management; notifies UI to download latest
-	forceUpdate  bool // true when management sets AlwaysUpdate; skips UI interaction and installs directly
+	mode        updateMode
+	modeGen     uint64
+	forceUpdate bool // true when management sets AlwaysUpdate; skips UI interaction and installs directly
 
 	lastTrigger    time.Time
 	mgmUpdateChan  chan struct{}
@@ -54,7 +63,7 @@ type Manager struct {
 	pendingVersion *v.Version
 
 	// updateMutex protects update, expectedVersion, updateToLatestVersion,
-	// downloadOnly, forceUpdate, pendingVersion, and lastTrigger fields
+	// mode, modeGen, forceUpdate, pendingVersion, and lastTrigger fields
 	updateMutex sync.Mutex
 
 	// installMutex and installing guard against concurrent installation attempts
@@ -76,7 +85,6 @@ func NewManager(statusRecorder *peer.Status, stateManager *statemanager.Manager)
 		updateChannel:       make(chan struct{}, 1),
 		currentVersion:      version.NetbirdVersion(),
 		update:              version.NewUpdate("nb/client"),
-		downloadOnly:        true,
 		autoUpdateSupported: isAutoUpdateSupported,
 	}
 
@@ -151,11 +159,7 @@ func (m *Manager) Start(ctx context.Context) {
 
 func (m *Manager) SetDownloadOnly() {
 	m.updateMutex.Lock()
-	m.downloadOnly = true
-	m.forceUpdate = false
-	m.expectedVersion = nil
-	m.updateToLatestVersion = false
-	m.lastTrigger = time.Time{}
+	m.setModeLocked(modeDownloadOnly)
 	m.updateMutex.Unlock()
 
 	select {
@@ -169,6 +173,7 @@ func (m *Manager) SetVersion(expectedVersion string, forceUpdate bool) {
 
 	if !m.autoUpdateSupported() {
 		log.Warnf("auto-update not supported on this platform")
+		m.SetDownloadOnly()
 		return
 	}
 
@@ -177,30 +182,32 @@ func (m *Manager) SetVersion(expectedVersion string, forceUpdate bool) {
 
 	if expectedVersion == "" {
 		log.Errorf("empty expected version provided")
-		m.expectedVersion = nil
-		m.updateToLatestVersion = false
-		m.downloadOnly = true
+		m.setModeLocked(modeDownloadOnly)
 		return
 	}
 
-	if expectedVersion == latestVersion {
-		m.updateToLatestVersion = true
-		m.expectedVersion = nil
-	} else {
-		expectedSemVer, err := v.NewVersion(expectedVersion)
+	var expectedSemVer *v.Version
+	if expectedVersion != latestVersion {
+		parsed, err := v.NewVersion(expectedVersion)
 		if err != nil {
-			log.Errorf("error parsing version: %v", err)
+			log.Errorf("error parsing version, switching to download-only: %v", err)
+			m.setModeLocked(modeDownloadOnly)
+			select {
+			case m.mgmUpdateChan <- struct{}{}:
+			default:
+			}
 			return
 		}
-		if m.expectedVersion != nil && m.expectedVersion.Equal(expectedSemVer) {
-			return
-		}
-		m.expectedVersion = expectedSemVer
-		m.updateToLatestVersion = false
+		expectedSemVer = parsed
 	}
 
-	m.lastTrigger = time.Time{}
-	m.downloadOnly = false
+	if m.sameDirectiveLocked(expectedSemVer, forceUpdate) {
+		return
+	}
+
+	m.setModeLocked(modeManaged)
+	m.expectedVersion = expectedSemVer
+	m.updateToLatestVersion = expectedSemVer == nil
 	m.forceUpdate = forceUpdate
 
 	select {
@@ -209,6 +216,13 @@ func (m *Manager) SetVersion(expectedVersion string, forceUpdate bool) {
 	}
 }
 
+func (m *Manager) ResetMode() {
+	m.updateMutex.Lock()
+	defer m.updateMutex.Unlock()
+
+	m.setModeLocked(modeUndecided)
+}
+
 // Install triggers the installation of the pending version. It is called when the user clicks the install button in the UI.
 func (m *Manager) Install(ctx context.Context) error {
 	if !m.autoUpdateSupported() {
@@ -255,12 +269,17 @@ func (m *Manager) NotifyUI() {
 		m.updateMutex.Unlock()
 		return
 	}
-	downloadOnly := m.downloadOnly
+	mode := m.mode
+	gen := m.modeGen
 	pendingVersion := m.pendingVersion
 	latestVersion := m.update.LatestVersion()
 	m.updateMutex.Unlock()
 
-	if downloadOnly {
+	if mode == modeUndecided {
+		return
+	}
+
+	if mode == modeDownloadOnly {
 		if latestVersion == nil {
 			return
 		}
@@ -268,6 +287,9 @@ func (m *Manager) NotifyUI() {
 		if err != nil || currentVersion.GreaterThanOrEqual(latestVersion) {
 			return
 		}
+		if m.modeChanged(gen) {
+			return
+		}
 		m.statusRecorder.PublishEvent(
 			cProto.SystemEvent_INFO,
 			cProto.SystemEvent_SYSTEM,
@@ -278,7 +300,7 @@ func (m *Manager) NotifyUI() {
 		return
 	}
 
-	if pendingVersion != nil {
+	if pendingVersion != nil && !m.modeChanged(gen) {
 		m.statusRecorder.PublishEvent(
 			cProto.SystemEvent_INFO,
 			cProto.SystemEvent_SYSTEM,
@@ -343,13 +365,18 @@ func (m *Manager) handleUpdate(ctx context.Context) {
 		return
 	}
 
-	downloadOnly := m.downloadOnly
+	mode := m.mode
+	gen := m.modeGen
 	forceUpdate := m.forceUpdate
 	curLatestVersion := m.update.LatestVersion()
 
 	switch {
+	case mode == modeUndecided:
+		log.Tracef("auto-update mode not decided yet")
+		m.updateMutex.Unlock()
+		return
 	// Download-only mode or resolve "latest" to actual version
-	case downloadOnly, m.updateToLatestVersion:
+	case mode == modeDownloadOnly, m.updateToLatestVersion:
 		if curLatestVersion == nil {
 			log.Tracef("latest version not fetched yet")
 			m.updateMutex.Unlock()
@@ -374,12 +401,17 @@ func (m *Manager) handleUpdate(ctx context.Context) {
 	m.lastTrigger = time.Now()
 	log.Infof("new version available: %s", updateVersion)
 
-	if !downloadOnly && !forceUpdate {
+	if mode == modeManaged && !forceUpdate {
 		m.pendingVersion = updateVersion
 	}
 	m.updateMutex.Unlock()
 
-	if downloadOnly {
+	if m.modeChanged(gen) {
+		log.Debugf("auto-update mode changed while checking %s, discarding", updateVersion)
+		return
+	}
+
+	if mode == modeDownloadOnly {
 		m.statusRecorder.PublishEvent(
 			cProto.SystemEvent_INFO,
 			cProto.SystemEvent_SYSTEM,
@@ -406,6 +438,33 @@ func (m *Manager) handleUpdate(ctx context.Context) {
 	)
 }
 
+func (m *Manager) modeChanged(gen uint64) bool {
+	m.updateMutex.Lock()
+	defer m.updateMutex.Unlock()
+
+	return m.modeGen != gen
+}
+
+func (m *Manager) sameDirectiveLocked(expectedVersion *v.Version, forceUpdate bool) bool {
+	if m.mode != modeManaged || m.forceUpdate != forceUpdate {
+		return false
+	}
+	if expectedVersion == nil {
+		return m.updateToLatestVersion
+	}
+	return m.expectedVersion != nil && m.expectedVersion.Equal(expectedVersion)
+}
+
+func (m *Manager) setModeLocked(mode updateMode) {
+	m.mode = mode
+	m.modeGen++
+	m.forceUpdate = false
+	m.expectedVersion = nil
+	m.updateToLatestVersion = false
+	m.pendingVersion = nil
+	m.lastTrigger = time.Time{}
+}
+
 func (m *Manager) install(ctx context.Context, pendingVersion *v.Version) error {
 	m.statusRecorder.PublishEvent(
 		cProto.SystemEvent_CRITICAL,

client/internal/updater/manager_linux_test.go

@@ -16,7 +16,7 @@ import (
 )
 
 // On Linux, only Mode 1 (downloadOnly) is supported.
-// SetVersion is a no-op because auto-update installation is not supported.
+// SetVersion falls back to download-only because auto-update installation is not supported.
 
 func Test_LatestVersion_Linux(t *testing.T) {
 	testMatrix := []struct {
@@ -70,7 +70,7 @@ func Test_LatestVersion_Linux(t *testing.T) {
 			t.Errorf("%s: Initial version mismatch, expected %v, got %v", c.name, c.initialLatestVersion.String(), ver)
 		}
 
-		mockUpdate.latestVersion = c.latestVersion
+		mockUpdate.setLatestVersion(c.latestVersion)
 		mockUpdate.onUpdate()
 
 		ver, enforced = waitForUpdateEvent(sub, 500*time.Millisecond)
@@ -89,22 +89,24 @@ func Test_LatestVersion_Linux(t *testing.T) {
 	}
 }
 
-func Test_SetVersion_NoOp_Linux(t *testing.T) {
-	// On Linux, SetVersion should be a no-op — no events fired
-	tmpFile := path.Join(t.TempDir(), "update-test-noop.json")
+func Test_SetVersion_FallsBackToDownloadOnly_Linux(t *testing.T) {
+	tmpFile := path.Join(t.TempDir(), "update-test-fallback.json")
 	recorder := peer.NewRecorder("")
 	sub := recorder.SubscribeToEvents()
 	defer recorder.UnsubscribeFromEvents(sub)
 
 	m := NewManager(recorder, statemanager.New(tmpFile))
-	m.update = &versionUpdateMock{latestVersion: v.Must(v.NewSemver("1.0.1"))}
+	m.update = &versionUpdateMock{latestVersion: v.Must(v.NewSemver("1.0.5"))}
 	m.currentVersion = "1.0.0"
 	m.Start(context.Background())
 	m.SetVersion("1.0.1", false)
 
-	ver, _ := waitForUpdateEvent(sub, 500*time.Millisecond)
-	if ver != "" {
-		t.Errorf("SetVersion should be a no-op on Linux, but got event with version %s", ver)
+	ver, enforced := waitForUpdateEvent(sub, 500*time.Millisecond)
+	if ver != "1.0.5" {
+		t.Fatalf("expected download-only event for fetched 1.0.5, got %q", ver)
+	}
+	if enforced {
+		t.Error("Linux fallback must never have enforced metadata")
 	}
 
 	m.Stop()

client/internal/updater/manager_mode_test.go

@@ -0,0 +1,217 @@
+package updater
+
+import (
+	"context"
+	"path"
+	"testing"
+	"time"
+
+	v "github.com/hashicorp/go-version"
+
+	"github.com/netbirdio/netbird/client/internal/peer"
+	"github.com/netbirdio/netbird/client/internal/statemanager"
+)
+
+func Test_UndecidedMode_SuppressesNotification(t *testing.T) {
+	tmpFile := path.Join(t.TempDir(), "update-test-undecided.json")
+	recorder := peer.NewRecorder("")
+	sub := recorder.SubscribeToEvents()
+	defer recorder.UnsubscribeFromEvents(sub)
+
+	mockUpdate := &versionUpdateMock{latestVersion: v.Must(v.NewSemver("1.0.1"))}
+	m := NewManager(recorder, statemanager.New(tmpFile))
+	m.update = mockUpdate
+	m.currentVersion = "1.0.0"
+	m.Start(context.Background())
+	defer m.Stop()
+
+	mockUpdate.onUpdate()
+	if ver, _ := waitForUpdateEvent(sub, 300*time.Millisecond); ver != "" {
+		t.Fatalf("undecided mode must not publish, got %q", ver)
+	}
+
+	m.NotifyUI()
+	if ver, _ := waitForUpdateEvent(sub, 300*time.Millisecond); ver != "" {
+		t.Fatalf("NotifyUI in undecided mode must not publish, got %q", ver)
+	}
+
+	m.SetDownloadOnly()
+	ver, enforced := waitForUpdateEvent(sub, 500*time.Millisecond)
+	if ver != "1.0.1" {
+		t.Fatalf("expected download-only event for 1.0.1, got %q", ver)
+	}
+	if enforced {
+		t.Error("download-only event must not carry enforced metadata")
+	}
+}
+
+func Test_ResetMode_ReturnsToUndecided(t *testing.T) {
+	tmpFile := path.Join(t.TempDir(), "update-test-reset.json")
+	recorder := peer.NewRecorder("")
+	sub := recorder.SubscribeToEvents()
+	defer recorder.UnsubscribeFromEvents(sub)
+
+	mockUpdate := &versionUpdateMock{latestVersion: v.Must(v.NewSemver("1.0.1"))}
+	m := NewManager(recorder, statemanager.New(tmpFile))
+	m.update = mockUpdate
+	m.currentVersion = "1.0.0"
+	m.autoUpdateSupported = func() bool { return true }
+	m.Start(context.Background())
+	defer m.Stop()
+
+	m.SetVersion("1.0.1", false)
+	ver, enforced := waitForUpdateEvent(sub, 500*time.Millisecond)
+	if ver != "1.0.1" || !enforced {
+		t.Fatalf("expected enforced event for 1.0.1, got %q enforced=%v", ver, enforced)
+	}
+
+	m.ResetMode()
+
+	mockUpdate.onUpdate()
+	if ver, _ := waitForUpdateEvent(sub, 300*time.Millisecond); ver != "" {
+		t.Fatalf("reset mode must not publish on fetch, got %q", ver)
+	}
+
+	m.NotifyUI()
+	if ver, _ := waitForUpdateEvent(sub, 300*time.Millisecond); ver != "" {
+		t.Fatalf("NotifyUI after reset must not publish, got %q", ver)
+	}
+
+	if err := m.Install(context.Background()); err == nil {
+		t.Fatal("Install after reset must fail without a pending version")
+	}
+
+	m.SetVersion("1.0.1", false)
+	ver, enforced = waitForUpdateEvent(sub, 500*time.Millisecond)
+	if ver != "1.0.1" || !enforced {
+		t.Fatalf("expected enforced event again after reset, got %q enforced=%v", ver, enforced)
+	}
+}
+
+func Test_SetDownloadOnly_ClearsPendingVersion(t *testing.T) {
+	tmpFile := path.Join(t.TempDir(), "update-test-pending.json")
+	recorder := peer.NewRecorder("")
+	sub := recorder.SubscribeToEvents()
+	defer recorder.UnsubscribeFromEvents(sub)
+
+	m := NewManager(recorder, statemanager.New(tmpFile))
+	m.update = &versionUpdateMock{latestVersion: v.Must(v.NewSemver("1.0.1"))}
+	m.currentVersion = "1.0.0"
+	m.autoUpdateSupported = func() bool { return true }
+	m.Start(context.Background())
+	defer m.Stop()
+
+	m.SetVersion("1.0.1", false)
+	if ver, enforced := waitForUpdateEvent(sub, 500*time.Millisecond); ver != "1.0.1" || !enforced {
+		t.Fatalf("expected enforced event for 1.0.1, got %q enforced=%v", ver, enforced)
+	}
+
+	m.SetDownloadOnly()
+	if ver, enforced := waitForUpdateEvent(sub, 500*time.Millisecond); ver != "1.0.1" || enforced {
+		t.Fatalf("expected download-only event for 1.0.1, got %q enforced=%v", ver, enforced)
+	}
+
+	if err := m.Install(context.Background()); err == nil {
+		t.Fatal("Install in download-only mode must not install the staged managed version")
+	}
+}
+
+func Test_ResetMode_SilencesStaleForceDirective(t *testing.T) {
+	tmpFile := path.Join(t.TempDir(), "update-test-force-reset.json")
+	recorder := peer.NewRecorder("")
+	sub := recorder.SubscribeToEvents()
+	defer recorder.UnsubscribeFromEvents(sub)
+
+	mockUpdate := &versionUpdateMock{}
+	m := NewManager(recorder, statemanager.New(tmpFile))
+	m.update = mockUpdate
+	m.currentVersion = "1.0.0"
+	m.autoUpdateSupported = func() bool { return true }
+	m.Start(context.Background())
+	defer m.Stop()
+
+	// Management enforces "latest" before the fetcher has reported any version,
+	// so nothing can be installed while the engine is still up.
+	m.SetVersion(latestVersion, true)
+	if event := waitForAnyEvent(sub, 300*time.Millisecond); event != nil {
+		t.Fatalf("no event expected before the latest version is known, got %v", event)
+	}
+
+	// The engine stop resets the mode. A release published afterwards must not
+	// trigger the stale forced install or any notification.
+	m.ResetMode()
+	mockUpdate.setLatestVersion(v.Must(v.NewSemver("1.0.1")))
+	mockUpdate.onUpdate()
+	if event := waitForAnyEvent(sub, 300*time.Millisecond); event != nil {
+		t.Fatalf("stale force directive must stay silent after reset, got %v", event)
+	}
+
+	m.SetVersion("1.0.1", false)
+	ver, enforced := waitForUpdateEvent(sub, 500*time.Millisecond)
+	if ver != "1.0.1" || !enforced {
+		t.Fatalf("expected enforced event after a fresh directive, got %q enforced=%v", ver, enforced)
+	}
+}
+
+func Test_SetVersion_MalformedFallsBackToDownloadOnly(t *testing.T) {
+	tmpFile := path.Join(t.TempDir(), "update-test-malformed.json")
+	recorder := peer.NewRecorder("")
+	sub := recorder.SubscribeToEvents()
+	defer recorder.UnsubscribeFromEvents(sub)
+
+	m := NewManager(recorder, statemanager.New(tmpFile))
+	m.update = &versionUpdateMock{latestVersion: v.Must(v.NewSemver("1.0.1"))}
+	m.currentVersion = "1.0.0"
+	m.autoUpdateSupported = func() bool { return true }
+	m.Start(context.Background())
+	defer m.Stop()
+
+	m.SetVersion("not-a-version", false)
+	ver, enforced := waitForUpdateEvent(sub, 500*time.Millisecond)
+	if ver != "1.0.1" {
+		t.Fatalf("expected download-only event for 1.0.1 after malformed version, got %q", ver)
+	}
+	if enforced {
+		t.Error("malformed version fallback must not carry enforced metadata")
+	}
+}
+
+func Test_SetVersion_ForceChangeAppliesWithSameVersion(t *testing.T) {
+	m := NewManager(peer.NewRecorder(""), statemanager.New(path.Join(t.TempDir(), "update-test-force-change.json")))
+	m.update = &versionUpdateMock{}
+	m.autoUpdateSupported = func() bool { return true }
+
+	m.SetVersion("1.0.1", false)
+	m.SetVersion("1.0.1", true)
+
+	m.updateMutex.Lock()
+	defer m.updateMutex.Unlock()
+	if !m.forceUpdate {
+		t.Fatal("enabling force update without a version change must take effect")
+	}
+	if m.expectedVersion == nil || m.expectedVersion.String() != "1.0.1" {
+		t.Fatalf("expected version 1.0.1 to stay set, got %v", m.expectedVersion)
+	}
+}
+
+func Test_SetVersion_RepeatedDirectiveKeepsMode(t *testing.T) {
+	m := NewManager(peer.NewRecorder(""), statemanager.New(path.Join(t.TempDir(), "update-test-repeat.json")))
+	m.update = &versionUpdateMock{}
+	m.autoUpdateSupported = func() bool { return true }
+
+	for _, expected := range []string{"1.0.1", latestVersion} {
+		m.SetVersion(expected, false)
+		m.updateMutex.Lock()
+		gen := m.modeGen
+		m.updateMutex.Unlock()
+
+		m.SetVersion(expected, false)
+		m.updateMutex.Lock()
+		repeatedGen := m.modeGen
+		m.updateMutex.Unlock()
+
+		if repeatedGen != gen {
+			t.Errorf("repeating the %q directive must not reset the mode", expected)
+		}
+	}
+}

client/internal/updater/manager_test.go

@@ -66,7 +66,7 @@ func Test_LatestVersion(t *testing.T) {
 			t.Errorf("%s: Initial update version mismatch, expected %v, got %v", c.name, c.initialLatestVersion.String(), ver)
 		}
 
-		mockUpdate.latestVersion = c.latestVersion
+		mockUpdate.setLatestVersion(c.latestVersion)
 		mockUpdate.onUpdate()
 
 		ver, _ = waitForUpdateEvent(sub, 500*time.Millisecond)

client/internal/updater/manager_test_helpers_test.go

@@ -2,21 +2,24 @@ package updater
 
 import (
 	"strconv"
+	"sync"
 	"time"
 
 	v "github.com/hashicorp/go-version"
 
 	"github.com/netbirdio/netbird/client/internal/peer"
+	cProto "github.com/netbirdio/netbird/client/proto"
 )
 
 type versionUpdateMock struct {
 	latestVersion *v.Version
 	onUpdate      func()
+	mu            sync.Mutex
 }
 
-func (m versionUpdateMock) StopWatch() {}
+func (m *versionUpdateMock) StopWatch() {}
 
-func (m versionUpdateMock) SetDaemonVersion(newVersion string) bool {
+func (m *versionUpdateMock) SetDaemonVersion(newVersion string) bool {
 	return false
 }
 
@@ -24,11 +27,19 @@ func (m *versionUpdateMock) SetOnUpdateListener(updateFn func()) {
 	m.onUpdate = updateFn
 }
 
-func (m versionUpdateMock) LatestVersion() *v.Version {
+func (m *versionUpdateMock) LatestVersion() *v.Version {
+	m.mu.Lock()
+	defer m.mu.Unlock()
 	return m.latestVersion
 }
 
-func (m versionUpdateMock) StartFetcher() {}
+func (m *versionUpdateMock) StartFetcher() {}
+
+func (m *versionUpdateMock) setLatestVersion(version *v.Version) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.latestVersion = version
+}
 
 // waitForUpdateEvent waits for a new_version_available event, returns the version string or "" on timeout.
 func waitForUpdateEvent(sub *peer.EventSubscription, timeout time.Duration) (version string, enforced bool) {
@@ -54,3 +65,20 @@ func waitForUpdateEvent(sub *peer.EventSubscription, timeout time.Duration) (ver
 		}
 	}
 }
+
+// waitForAnyEvent returns the first published event of any kind, or nil on timeout.
+// Unlike waitForUpdateEvent it also catches the install-progress events, so a test
+// can assert that a forced install never started.
+func waitForAnyEvent(sub *peer.EventSubscription, timeout time.Duration) *cProto.SystemEvent {
+	timer := time.NewTimer(timeout)
+	defer timer.Stop()
+	select {
+	case event, ok := <-sub.Events():
+		if !ok {
+			return nil
+		}
+		return event
+	case <-timer.C:
+		return nil
+	}
+}