Merge branch 'main' into poc/certificate-posture

.github/workflows/release.yml

@@ -287,10 +287,15 @@ jobs:
           image_refs=()
 
           tag_and_push() {
-            local src="$1" img_name tag dst
+            local src="$1" img_name tag dst variant=""
             img_name="${src%%:*}"
+            # Client variants share a repository, so keep their tag suffixes.
+            case "$src" in
+              *-rootless-ubi-amd64) variant="-rootless-ubi" ;;
+              *-rootless-amd64) variant="-rootless" ;;
+            esac
             for tag in $(resolve_tags); do
-              dst="${img_name}:${tag}"
+              dst="${img_name}:${tag}${variant}"
               echo "Tagging ${src} -> ${dst}"
               docker tag "$src" "$dst"
               docker push "$dst"

.goreleaser.yaml

@@ -289,6 +289,43 @@ dockers_v2:
        "org.opencontainers.image.revision": "{{.FullCommit}}"
        "org.opencontainers.image.source": "{{.GitURL}}"
        "maintainer": "dev@netbird.io"
+   - id: netbird-rootless-ubi
+     disable: "{{ .Env.SKIP_DOCKER_PUSH }}"
+     ids:
+       - netbird
+     images:
+       - netbirdio/netbird
+       - ghcr.io/netbirdio/netbird
+     tags:
+       - "{{ .Version }}-rootless-ubi"
+       - "{{ if eq .Env.SKIP_PUBLISH \"false\" }}rootless-ubi-latest{{ end }}"
+     dockerfile: client/Dockerfile-rootless.ubi
+     extra_files:
+       - client/netbird-entrypoint.sh
+     platforms:
+       - linux/amd64
+       - linux/arm64
+     build_args:
+       VERSION: "{{ .Version }}"
+       RELEASE: "{{ .Timestamp }}"
+     hooks:
+       pre:
+         - cmd: 'sh client/collect-licenses.sh "{{ .ContextDir }}/licenses" amd64 arm64'
+           env:
+             - GOOS=linux
+             - CGO_ENABLED=0
+     labels:
+       "org.opencontainers.image.created": "{{.Date}}"
+       "org.opencontainers.image.version": "{{.Version}}"
+       "org.opencontainers.image.revision": "{{.FullCommit}}"
+       "org.opencontainers.image.source": "{{.GitURL}}"
+     annotations:
+       "org.opencontainers.image.created": "{{.Date}}"
+       "org.opencontainers.image.title": "{{.ProjectName}}"
+       "org.opencontainers.image.version": "{{.Version}}"
+       "org.opencontainers.image.revision": "{{.FullCommit}}"
+       "org.opencontainers.image.source": "{{.GitURL}}"
+       "maintainer": "dev@netbird.io"
    - id: relay
      disable: "{{ .Env.SKIP_DOCKER_PUSH }}"
      ids:

client/Dockerfile-rootless.ubi

@@ -0,0 +1,45 @@
+FROM registry.access.redhat.com/ubi9/ubi-minimal@sha256:7fbeae18dc9476399f565e68255f602a3374ea8614ba3d14843565131a13ff93
+
+ARG TARGETPLATFORM
+ARG NETBIRD_BINARY=$TARGETPLATFORM/netbird
+ARG VERSION=dev
+ARG RELEASE=1
+
+LABEL name="netbird-rootless" \
+      maintainer="NetBird <dev@netbird.io>" \
+      vendor="NetBird GmbH" \
+      version="${VERSION}" \
+      release="${RELEASE}" \
+      summary="NetBird Rootless Client" \
+      description="NetBird connects devices through an encrypted overlay using userspace networking without a TUN device or network administration capabilities."
+
+RUN microdnf install -y bash ca-certificates && microdnf clean all
+
+COPY --chmod=0555 client/netbird-entrypoint.sh /usr/local/bin/netbird-entrypoint.sh
+COPY --chmod=0555 ${NETBIRD_BINARY} /usr/local/bin/netbird
+COPY licenses/ /licenses/
+# Only application storage is group-writable for arbitrary non-root UIDs.
+# Runtime-created credentials keep the client's restrictive file modes.
+RUN mkdir -p /var/lib/netbird && \
+    chown 1000:0 /var/lib/netbird && \
+    chmod 0770 /var/lib/netbird && \
+    chmod -R a+rX /licenses
+
+WORKDIR /var/lib/netbird
+USER 1000:0
+
+ENV \
+    HOME="/var/lib/netbird" \
+    NETBIRD_BIN="/usr/local/bin/netbird" \
+    NB_USE_NETSTACK_MODE="true" \
+    NB_ENABLE_NETSTACK_LOCAL_FORWARDING="true" \
+    NB_CONFIG="/var/lib/netbird/config.json" \
+    NB_STATE_DIR="/var/lib/netbird" \
+    NB_DAEMON_ADDR="unix:///var/lib/netbird/netbird.sock" \
+    NB_LOG_FILE="console,/var/lib/netbird/client.log" \
+    NB_DISABLE_DNS="true" \
+    NB_ENABLE_CAPTURE="false" \
+    NB_ENTRYPOINT_SERVICE_TIMEOUT="30"
+
+STOPSIGNAL SIGTERM
+ENTRYPOINT ["/usr/local/bin/netbird-entrypoint.sh"]

client/android/client.go

@@ -9,6 +9,7 @@ import (
 	"slices"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"golang.org/x/exp/maps"
@@ -90,6 +91,14 @@ type Client struct {
 	connectClient *internal.ConnectClient
 	config        *profilemanager.Config
 	cacheDir      string
+
+	// mdmSource holds the per-Client MDM policy source and its change
+	// detector as one unit. Set by SetMDMPolicyFetcher (called from the
+	// Kotlin side). Each Run passes the loader to the resolved Config so
+	// applyMDMPolicy picks up the active overlay. Nil means "MDM
+	// enforcement off for this Client".
+	mdmSource atomic.Pointer[mdmSource]
+
 	// Identifies the running profile for the SSO login hint; see profile_state.go.
 	cfgPath string
 
@@ -178,6 +187,7 @@ func (c *Client) Run(platformFiles PlatformFiles, urlOpener URLOpener, isAndroid
 	if err != nil {
 		return err
 	}
+	c.applyMDMOverlay(cfg)
 	c.recorder.UpdateManagementAddress(cfg.ManagementURL.String())
 	c.recorder.UpdateRosenpass(cfg.RosenpassEnabled, cfg.RosenpassPermissive)
 
@@ -229,6 +239,7 @@ func (c *Client) RunWithoutLogin(platformFiles PlatformFiles, dns *DNSList, dnsR
 	if err != nil {
 		return err
 	}
+	c.applyMDMOverlay(cfg)
 	c.recorder.UpdateManagementAddress(cfg.ManagementURL.String())
 	c.recorder.UpdateRosenpass(cfg.RosenpassEnabled, cfg.RosenpassPermissive)
 
@@ -327,6 +338,7 @@ func (c *Client) DebugBundle(platformFiles PlatformFiles, anonymize bool, anonym
 		if err != nil {
 			return "", fmt.Errorf("load config: %w", err)
 		}
+		c.applyMDMOverlay(cfg)
 		cacheDir = platformFiles.CacheDir()
 	}
 

client/android/client_mdm.go

@@ -0,0 +1,52 @@
+//go:build android
+
+package android
+
+import (
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
+)
+
+type mdmSource struct {
+	loader   *mdm.Loader
+	detector *mdm.ChangeDetector
+}
+
+// SetMDMPolicyFetcher registers the native-provided MDM policy fetcher on
+// this Client; passing nil disables MDM enforcement.
+func (c *Client) SetMDMPolicyFetcher(p PolicyFetcher) {
+	loader := loaderFor(p)
+	c.mdmSource.Store(&mdmSource{loader: loader, detector: mdm.NewChangeDetector(loader)})
+}
+
+// HasMDMPolicyChanged re-reads the managed configuration and reports whether
+// it changed since the last observation; call it from the native OS-change
+// notification and restart the engine only on true.
+func (c *Client) HasMDMPolicyChanged() bool {
+	src := c.mdmSource.Load()
+	if src == nil {
+		return false
+	}
+	return src.detector.Changed()
+}
+
+// GetRestrictionsJSON returns the UI enforcement snapshot derived from the
+// active MDM policy, in the JSON shape shared with the desktop frontend.
+func (c *Client) GetRestrictionsJSON() (string, error) {
+	return mdm.BuildRestrictions(c.mdmLoader().Load()).JSON()
+}
+
+func (c *Client) applyMDMOverlay(cfg *profilemanager.Config) {
+	loader := c.mdmLoader()
+	if cfg == nil || loader == nil {
+		return
+	}
+	cfg.ApplyMDMPolicy(loader.Load())
+}
+
+func (c *Client) mdmLoader() *mdm.Loader {
+	if src := c.mdmSource.Load(); src != nil {
+		return src.loader
+	}
+	return nil
+}

client/android/login.go

@@ -8,6 +8,7 @@ import (
 
 	"github.com/netbirdio/netbird/client/internal/auth"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 	"github.com/netbirdio/netbird/client/mobile"
 	"github.com/netbirdio/netbird/client/system"
 )
@@ -46,16 +47,24 @@ type Auth struct {
 // an earlier call is orphaned on the server. It also breaks a client that enrols and then runs from
 // the persisted config, because the identity it registered is not the one it runs with — the
 // management stream rejects it with "no peer auth method provided".
-func NewAuth(cfgPath string, mgmURL string) (*Auth, error) {
-	inputCfg := profilemanager.ConfigInput{
-		ConfigPath:    cfgPath,
-		ManagementURL: mgmURL,
+//
+// Auth is constructed under the active MDM policy: the policy is overlaid on
+// the resolved config so the login runs against the enforced values, while
+// the persisted config keeps the caller-supplied ones; a caller-supplied
+// management URL is ignored while MDM manages that key. A nil fetcher
+// disables MDM enforcement.
+func NewAuth(cfgPath string, mgmURL string, fetcher PolicyFetcher) (*Auth, error) {
+	policy := loaderFor(fetcher).Load()
+	inputCfg := profilemanager.ConfigInput{ConfigPath: cfgPath}
+	if _, managed := policy.GetString(mdm.KeyManagementURL); !managed {
+		inputCfg.ManagementURL = mgmURL
 	}
 
 	cfg, err := profilemanager.UpdateOrCreateConfig(inputCfg)
 	if err != nil {
 		return nil, err
 	}
+	cfg.ApplyMDMPolicy(policy)
 
 	return &Auth{
 		ctx:     context.Background(),
@@ -75,9 +84,7 @@ func NewAuthWithConfig(ctx context.Context, config *profilemanager.Config, cfgPa
 	}
 }
 
-// SaveConfigIfSSOSupported test the connectivity with the management server by retrieving the server device flow info.
-// If it returns a flow info than save the configuration and return true. If it gets a codes.NotFound, it means that SSO
-// is not supported and returns false without saving the configuration. For other errors return false.
+// SaveConfigIfSSOSupported reports whether the management server supports SSO; the config is already persisted by NewAuth.
 func (a *Auth) SaveConfigIfSSOSupported(listener SSOListener) {
 	go func() {
 		sso, err := a.saveConfigIfSSOSupported()
@@ -101,15 +108,10 @@ func (a *Auth) saveConfigIfSSOSupported() (bool, error) {
 		return false, fmt.Errorf("failed to check SSO support: %v", err)
 	}
 
-	if !supportsSSO {
-		return false, nil
-	}
-
-	err = profilemanager.WriteOutConfig(a.cfgPath, a.config)
-	return true, err
+	return supportsSSO, nil
 }
 
-// LoginWithSetupKeyAndSaveConfig test the connectivity with the management server with the setup key.
+// LoginWithSetupKeyAndSaveConfig registers the peer with the setup key; the config is already persisted by NewAuth.
 func (a *Auth) LoginWithSetupKeyAndSaveConfig(resultListener ErrListener, setupKey string, deviceName string) {
 	go func() {
 		err := a.loginWithSetupKeyAndSaveConfig(setupKey, deviceName)
@@ -134,8 +136,7 @@ func (a *Auth) loginWithSetupKeyAndSaveConfig(setupKey string, deviceName string
 	if err != nil {
 		return fmt.Errorf("login failed: %v", err)
 	}
-
-	return profilemanager.WriteOutConfig(a.cfgPath, a.config)
+	return nil
 }
 
 // Login try register the client on the server

client/android/login_test.go

@@ -16,7 +16,7 @@ import (
 func TestNewAuth_ReusesPersistedIdentity(t *testing.T) {
 	cfgPath := filepath.Join(t.TempDir(), "config.json")
 
-	first, err := NewAuth(cfgPath, "https://api.example.com:443")
+	first, err := NewAuth(cfgPath, "https://api.example.com:443", nil)
 	if err != nil {
 		t.Fatalf("first NewAuth: %v", err)
 	}
@@ -24,7 +24,7 @@ func TestNewAuth_ReusesPersistedIdentity(t *testing.T) {
 		t.Fatal("first NewAuth produced no private key")
 	}
 
-	second, err := NewAuth(cfgPath, "https://api.example.com:443")
+	second, err := NewAuth(cfgPath, "https://api.example.com:443", nil)
 	if err != nil {
 		t.Fatalf("second NewAuth: %v", err)
 	}
@@ -38,7 +38,7 @@ func TestNewAuth_ReusesPersistedIdentity(t *testing.T) {
 func TestNewAuth_CreatesConfigWhenAbsent(t *testing.T) {
 	cfgPath := filepath.Join(t.TempDir(), "config.json")
 
-	auth, err := NewAuth(cfgPath, "https://api.example.com:443")
+	auth, err := NewAuth(cfgPath, "https://api.example.com:443", nil)
 	if err != nil {
 		t.Fatalf("NewAuth: %v", err)
 	}

client/android/mdm.go

@@ -0,0 +1,19 @@
+package android
+
+import (
+	"github.com/netbirdio/netbird/client/mdm"
+)
+
+// PolicyFetcher is implemented by the native layer to return the current
+// managed configuration as a JSON-encoded object string; "" means no MDM
+// source is present.
+type PolicyFetcher interface {
+	FetchJSON() string
+}
+
+func loaderFor(p PolicyFetcher) *mdm.Loader {
+	if p == nil {
+		return mdm.NewJSONLoader(nil)
+	}
+	return mdm.NewJSONLoader(p.FetchJSON)
+}

client/android/preferences.go

@@ -1,12 +1,16 @@
 package android
 
 import (
+	"sync/atomic"
+
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 )
 
 // Preferences exports a subset of the internal config for gomobile
 type Preferences struct {
 	configInput profilemanager.ConfigInput
+	mdmLoader   atomic.Pointer[mdm.Loader]
 }
 
 // NewPreferences creates a new Preferences instance
@@ -14,11 +18,30 @@ func NewPreferences(configPath string) *Preferences {
 	ci := profilemanager.ConfigInput{
 		ConfigPath: configPath,
 	}
-	return &Preferences{ci}
+	return &Preferences{configInput: ci}
+}
+
+// SetMDMPolicyFetcher registers the native-provided MDM policy fetcher on
+// this Preferences instance; passing nil disables MDM enforcement.
+func (p *Preferences) SetMDMPolicyFetcher(f PolicyFetcher) {
+	p.mdmLoader.Store(loaderFor(f))
+}
+
+// GetRestrictionsJSON returns the UI enforcement snapshot derived from the
+// active MDM policy, in the JSON shape shared with the desktop frontend.
+func (p *Preferences) GetRestrictionsJSON() (string, error) {
+	return mdm.BuildRestrictions(p.policy()).JSON()
+}
+
+func (p *Preferences) policy() *mdm.Policy {
+	return p.mdmLoader.Load().Load()
 }
 
 // GetManagementURL reads URL from config file
 func (p *Preferences) GetManagementURL() (string, error) {
+	if v, ok := p.policy().GetString(mdm.KeyManagementURL); ok {
+		return mdm.CanonicalURL(v), nil
+	}
 	if p.configInput.ManagementURL != "" {
 		return p.configInput.ManagementURL, nil
 	}
@@ -27,7 +50,7 @@ func (p *Preferences) GetManagementURL() (string, error) {
 	if err != nil {
 		return "", err
 	}
-	return cfg.ManagementURL.String(), err
+	return cfg.ManagementURL.String(), nil
 }
 
 // SetManagementURL stores the given URL and waits for commit
@@ -53,17 +76,21 @@ func (p *Preferences) SetAdminURL(url string) {
 	p.configInput.AdminURL = url
 }
 
-// GetPreSharedKey reads pre-shared key from config file
-func (p *Preferences) GetPreSharedKey() (string, error) {
+// HasPreSharedKey reports whether a pre-shared key is staged, persisted, or
+// enforced by MDM; the key itself is never handed to the native layer.
+func (p *Preferences) HasPreSharedKey() (bool, error) {
+	if _, ok := p.policy().GetString(mdm.KeyPreSharedKey); ok {
+		return true, nil
+	}
 	if p.configInput.PreSharedKey != nil {
-		return *p.configInput.PreSharedKey, nil
+		return *p.configInput.PreSharedKey != "", nil
 	}
 
 	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
 	if err != nil {
-		return "", err
+		return false, err
 	}
-	return cfg.PreSharedKey, err
+	return cfg.PreSharedKey != "", nil
 }
 
 // SetPreSharedKey stores the given key and waits for commit
@@ -78,6 +105,9 @@ func (p *Preferences) SetRosenpassEnabled(enabled bool) {
 
 // GetRosenpassEnabled reads Rosenpass enabled status from config file
 func (p *Preferences) GetRosenpassEnabled() (bool, error) {
+	if v, ok := p.policy().GetBool(mdm.KeyRosenpassEnabled); ok {
+		return v, nil
+	}
 	if p.configInput.RosenpassEnabled != nil {
 		return *p.configInput.RosenpassEnabled, nil
 	}
@@ -96,6 +126,9 @@ func (p *Preferences) SetRosenpassPermissive(permissive bool) {
 
 // GetRosenpassPermissive reads Rosenpass permissive setting from config file
 func (p *Preferences) GetRosenpassPermissive() (bool, error) {
+	if v, ok := p.policy().GetBool(mdm.KeyRosenpassPermissive); ok {
+		return v, nil
+	}
 	if p.configInput.RosenpassPermissive != nil {
 		return *p.configInput.RosenpassPermissive, nil
 	}
@@ -109,6 +142,9 @@ func (p *Preferences) GetRosenpassPermissive() (bool, error) {
 
 // GetDisableClientRoutes reads disable client routes setting from config file
 func (p *Preferences) GetDisableClientRoutes() (bool, error) {
+	if v, ok := p.policy().GetBool(mdm.KeyDisableClientRoutes); ok {
+		return v, nil
+	}
 	if p.configInput.DisableClientRoutes != nil {
 		return *p.configInput.DisableClientRoutes, nil
 	}
@@ -127,6 +163,9 @@ func (p *Preferences) SetDisableClientRoutes(disable bool) {
 
 // GetDisableServerRoutes reads disable server routes setting from config file
 func (p *Preferences) GetDisableServerRoutes() (bool, error) {
+	if v, ok := p.policy().GetBool(mdm.KeyDisableServerRoutes); ok {
+		return v, nil
+	}
 	if p.configInput.DisableServerRoutes != nil {
 		return *p.configInput.DisableServerRoutes, nil
 	}
@@ -181,6 +220,9 @@ func (p *Preferences) SetDisableFirewall(disable bool) {
 
 // GetServerSSHAllowed reads server SSH allowed setting from config file
 func (p *Preferences) GetServerSSHAllowed() (bool, error) {
+	if v, ok := p.policy().GetBool(mdm.KeyAllowServerSSH); ok {
+		return v, nil
+	}
 	if p.configInput.ServerSSHAllowed != nil {
 		return *p.configInput.ServerSSHAllowed, nil
 	}
@@ -291,6 +333,9 @@ func (p *Preferences) SetEnableSSHRemotePortForwarding(enabled bool) {
 
 // GetBlockInbound reads block inbound setting from config file
 func (p *Preferences) GetBlockInbound() (bool, error) {
+	if v, ok := p.policy().GetBool(mdm.KeyBlockInbound); ok {
+		return v, nil
+	}
 	if p.configInput.BlockInbound != nil {
 		return *p.configInput.BlockInbound, nil
 	}
@@ -327,7 +372,8 @@ func (p *Preferences) SetDisableIPv6(disable bool) {
 
 // GetRemoteJobsAllowed reads the remote jobs opt-in from config file
 func (p *Preferences) GetRemoteJobsAllowed() (bool, error) {
-	if p.configInput.RemoteJobsAllowed != nil {
+	policy := p.policy()
+	if !policy.HasKey(mdm.KeyRemoteJobsAllowed) && p.configInput.RemoteJobsAllowed != nil {
 		return *p.configInput.RemoteJobsAllowed, nil
 	}
 
@@ -335,10 +381,11 @@ func (p *Preferences) GetRemoteJobsAllowed() (bool, error) {
 	if err != nil {
 		return false, err
 	}
+	cfg.ApplyMDMPolicy(policy)
 	if cfg.RemoteJobsAllowed == nil {
 		return false, nil
 	}
-	return *cfg.RemoteJobsAllowed, err
+	return *cfg.RemoteJobsAllowed, nil
 }
 
 // SetRemoteJobsAllowed stores the given value and waits for commit
@@ -348,6 +395,9 @@ func (p *Preferences) SetRemoteJobsAllowed(allowed bool) {
 
 // Commit writes out the changes to the config file
 func (p *Preferences) Commit() error {
+	if err := profilemanager.CheckMDMConflicts(p.configInput, p.policy()); err != nil {
+		return err
+	}
 	_, err := profilemanager.UpdateOrCreateConfig(p.configInput)
 	return err
 }

client/android/preferences_test.go

@@ -28,14 +28,13 @@ func TestPreferences_DefaultValues(t *testing.T) {
 		t.Errorf("invalid default management url: %s", defaultVar)
 	}
 
-	var preSharedKey string
-	preSharedKey, err = p.GetPreSharedKey()
+	hasPSK, err := p.HasPreSharedKey()
 	if err != nil {
-		t.Fatalf("failed to read default preshared key: %s", err)
+		t.Fatalf("failed to read default preshared key presence: %s", err)
 	}
 
-	if preSharedKey != "" {
-		t.Errorf("invalid preshared key: %s", preSharedKey)
+	if hasPSK {
+		t.Errorf("unexpected preshared key presence on fresh config")
 	}
 }
 
@@ -65,13 +64,13 @@ func TestPreferences_ReadUncommitedValues(t *testing.T) {
 	}
 
 	p.SetPreSharedKey(exampleString)
-	resp, err = p.GetPreSharedKey()
+	hasPSK, err := p.HasPreSharedKey()
 	if err != nil {
-		t.Fatalf("failed to read preshared key: %s", err)
+		t.Fatalf("failed to read preshared key presence: %s", err)
 	}
 
-	if resp != exampleString {
-		t.Errorf("unexpected preshared key: %s", resp)
+	if !hasPSK {
+		t.Errorf("expected preshared key presence after staging one")
 	}
 }
 
@@ -109,12 +108,12 @@ func TestPreferences_Commit(t *testing.T) {
 		t.Errorf("unexpected management url: %s", resp)
 	}
 
-	resp, err = p.GetPreSharedKey()
+	hasPSK, err := p.HasPreSharedKey()
 	if err != nil {
-		t.Fatalf("failed to read preshared key: %s", err)
+		t.Fatalf("failed to read preshared key presence: %s", err)
 	}
 
-	if resp != examplePresharedKey {
-		t.Errorf("unexpected preshared key: %s", resp)
+	if !hasPSK {
+		t.Errorf("expected preshared key presence after commit")
 	}
 }

client/android/profile_manager.go

@@ -54,6 +54,12 @@ func NewProfileManager(configDir string) *ProfileManager {
 	return &ProfileManager{impl: mobile.NewProfileManager(configDir, androidUsername)}
 }
 
+// SetMDMPolicyFetcher registers the native-provided MDM policy fetcher on
+// this ProfileManager; passing nil disables MDM enforcement.
+func (pm *ProfileManager) SetMDMPolicyFetcher(f PolicyFetcher) {
+	pm.impl.SetMDMLoader(loaderFor(f))
+}
+
 // ListProfiles returns all available profiles, including the default profile,
 // with their active status set.
 func (pm *ProfileManager) ListProfiles() (*ProfileArray, error) {

client/cmd/login.go

@@ -15,6 +15,7 @@ import (
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/auth"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 	nbnet "github.com/netbirdio/netbird/client/net"
 	"github.com/netbirdio/netbird/client/proto"
 	"github.com/netbirdio/netbird/client/server"
@@ -330,6 +331,11 @@ func doForegroundLogin(ctx context.Context, cmd *cobra.Command, setupKey string,
 	if err != nil {
 		return fmt.Errorf("read config file %s: %v", configFilePath, err)
 	}
+	// CLI standalone login: profilemanager no longer auto-applies MDM,
+	// so layer in the OS-native policy here. Desktop builds construct
+	// a Loader with no fetcher — the build-tagged loadPlatform reads
+	// the registry/plist directly.
+	config.ApplyMDMPolicy(mdm.NewLoader(nil).Load())
 
 	// Mirror runInForegroundMode: recover residual state (DNS, firewall,
 	// ssh config, legacy routing) from a previous unclean shutdown and

client/cmd/up.go

@@ -21,6 +21,7 @@ import (
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/internal/peer"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 	nbnet "github.com/netbirdio/netbird/client/net"
 	"github.com/netbirdio/netbird/client/proto"
 	"github.com/netbirdio/netbird/client/server"
@@ -234,6 +235,10 @@ func runInForegroundMode(ctx context.Context, cmd *cobra.Command, activeProf *pr
 	if err != nil {
 		return fmt.Errorf("get config file: %v", err)
 	}
+	// CLI foreground path runs without the daemon Server: layer in the
+	// active MDM policy explicitly so a forced ManagementURL / PSK /
+	// other managed key actually takes effect on this run.
+	config.ApplyMDMPolicy(mdm.NewLoader(nil).Load())
 
 	_, _ = profilemanager.UpdateOldManagementURL(ctx, config, configFilePath)
 

client/collect-licenses.sh

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

client/embed/embed.go

@@ -21,6 +21,7 @@ import (
 	"github.com/netbirdio/netbird/client/internal/auth"
 	"github.com/netbirdio/netbird/client/internal/peer"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 	nbssh "github.com/netbirdio/netbird/client/ssh"
 	"github.com/netbirdio/netbird/client/system"
 	"github.com/netbirdio/netbird/shared/management/domain"
@@ -229,6 +230,10 @@ func New(opts Options) (*Client, error) {
 	if err != nil {
 		return nil, fmt.Errorf("create config: %w", err)
 	}
+	// Embedded path runs without the daemon Server: apply the active
+	// MDM policy explicitly so a forced ManagementURL / PSK / other
+	// managed key takes effect on this embedded engine instance.
+	config.ApplyMDMPolicy(mdm.NewLoader(nil).Load())
 
 	if opts.PrivateKey != "" {
 		config.PrivateKey = opts.PrivateKey

client/internal/dns/service_listener.go

@@ -6,6 +6,7 @@ import (
 	"net"
 	"net/netip"
 	"runtime"
+	"slices"
 	"strconv"
 	"sync"
 	"time"
@@ -17,17 +18,20 @@ import (
 	nberrors "github.com/netbirdio/netbird/client/errors"
 
 	firewall "github.com/netbirdio/netbird/client/firewall/manager"
-	"github.com/netbirdio/netbird/client/internal/ebpf"
-	ebpfMgr "github.com/netbirdio/netbird/client/internal/ebpf/manager"
 )
 
 const (
 	customPort = 5053
+	// randomPortAttempts bounds the search for a port free on both protocols.
+	randomPortAttempts = 5
 )
 
 var (
 	defaultIP = netip.MustParseAddr("127.0.0.1")
 	customIP  = netip.MustParseAddr("127.0.0.153")
+
+	// dnatProtocols are the protocols the port 53 redirect covers.
+	dnatProtocols = []firewall.Protocol{firewall.ProtocolUDP, firewall.ProtocolTCP}
 )
 
 type serviceViaListener struct {
@@ -40,9 +44,20 @@ type serviceViaListener struct {
 	listenPort        uint16
 	listenerIsRunning bool
 	listenerFlagLock  sync.Mutex
-	ebpfService       ebpfMgr.Manager
 	firewall          Firewall
-	tcpDNATConfigured bool
+	// dnatRules holds the port 53 redirects that are installed and not yet
+	// removed, so a removal that fails can be retried.
+	dnatRules []dnatRule
+}
+
+// dnatRule is a port 53 redirect as it was installed. The target is kept with
+// the rule because the listener can come back on a different address or port,
+// and a retried removal has to name the address and port the rule was added
+// with, not the ones in use now.
+type dnatRule struct {
+	protocol firewall.Protocol
+	ip       netip.Addr
+	port     uint16
 }
 
 func newServiceViaListener(wgIface WGIface, customAddr *netip.AddrPort, fw Firewall) *serviceViaListener {
@@ -112,34 +127,93 @@ func (s *serviceViaListener) Listen() error {
 		}
 	}()
 
-	// When eBPF redirects UDP port 53 to our listen port, TCP still needs
-	// a DNAT rule because eBPF only handles UDP.
-	if s.ebpfService != nil && s.firewall != nil && s.listenPort != DefaultPort {
-		if err := s.firewall.AddOutputDNAT(s.listenIP, firewall.ProtocolTCP, DefaultPort, s.listenPort); err != nil {
-			log.Warnf("failed to add DNS TCP DNAT rule, TCP DNS on port 53 will not work: %v", err)
-		} else {
-			s.tcpDNATConfigured = true
-			log.Infof("added DNS TCP DNAT rule: %s:%d -> %s:%d", s.listenIP, DefaultPort, s.listenIP, s.listenPort)
-		}
+	if s.listenPort != DefaultPort {
+		s.setupDNAT()
 	}
 
 	return nil
 }
 
+// setupDNAT redirects port 53 to the port the DNS server actually listens on.
+// Both protocols must be redirected or none: RuntimePort reports port 53 only
+// while the full redirect is in place, so a half-configured redirect would
+// advertise a resolver that answers over one protocol.
+func (s *serviceViaListener) setupDNAT() {
+	if s.firewall == nil {
+		log.Errorf("no firewall manager available to redirect DNS port %d to %d, "+
+			"clients pointed at %s will not reach the resolver", DefaultPort, s.listenPort, s.listenIP)
+		return
+	}
+
+	// Clear whatever an earlier removal left behind first. Those rules can point
+	// at an address or port this listener no longer uses, and they are matched
+	// before anything added now, so adding a redirect on top of one would keep
+	// sending port 53 traffic to the previous listener while reporting the
+	// redirect as complete. The rules stay recorded for a later attempt.
+	if err := s.removeDNAT(); err != nil {
+		log.Errorf("failed to remove stale DNS DNAT rules, leaving port %d redirected to the previous listener: %v",
+			DefaultPort, err)
+		return
+	}
+
+	for _, proto := range dnatProtocols {
+		if err := s.firewall.AddOutputDNAT(s.listenIP, proto, DefaultPort, s.listenPort); err != nil {
+			log.Errorf("failed to add DNS %s DNAT rule, DNS on port %d will not work: %v",
+				proto, DefaultPort, err)
+			if err := s.removeDNAT(); err != nil {
+				log.Warnf("failed to roll back DNS DNAT rules, retrying on stop: %v", err)
+			}
+			return
+		}
+		s.dnatRules = append(s.dnatRules, dnatRule{protocol: proto, ip: s.listenIP, port: s.listenPort})
+	}
+
+	log.Infof("added DNS DNAT rules: %s:%d -> %s:%d (UDP + TCP)", s.listenIP, DefaultPort, s.listenIP, s.listenPort)
+}
+
+// removeDNAT removes every installed port 53 redirect. A rule whose removal
+// fails stays recorded so a later setup or Stop retries it, rather than leaving
+// port 53 pointing at a resolver that is no longer listening.
+func (s *serviceViaListener) removeDNAT() error {
+	if s.firewall == nil {
+		return nil
+	}
+
+	var merr *multierror.Error
+	var remaining []dnatRule
+	for _, rule := range s.dnatRules {
+		if err := s.firewall.RemoveOutputDNAT(rule.ip, rule.protocol, DefaultPort, rule.port); err != nil {
+			merr = multierror.Append(merr, fmt.Errorf("remove DNS %s DNAT rule for %s:%d: %w",
+				rule.protocol, rule.ip, rule.port, err))
+			remaining = append(remaining, rule)
+		}
+	}
+	s.dnatRules = remaining
+
+	return nberrors.FormatErrorOrNil(merr)
+}
+
 func (s *serviceViaListener) Stop() error {
 	s.listenerFlagLock.Lock()
 	defer s.listenerFlagLock.Unlock()
 
+	var merr *multierror.Error
+
+	// Redirects are removed even when the listener is already stopped, so that
+	// a removal which failed earlier is retried instead of leaving port 53
+	// pointing at a resolver that no longer listens.
+	if err := s.removeDNAT(); err != nil {
+		merr = multierror.Append(merr, err)
+	}
+
 	if !s.listenerIsRunning {
-		return nil
+		return nberrors.FormatErrorOrNil(merr)
 	}
 	s.listenerIsRunning = false
 
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancel()
 
-	var merr *multierror.Error
-
 	if err := s.server.ShutdownContext(ctx); err != nil {
 		merr = multierror.Append(merr, fmt.Errorf("stop DNS UDP server: %w", err))
 	}
@@ -148,19 +222,6 @@ func (s *serviceViaListener) Stop() error {
 		merr = multierror.Append(merr, fmt.Errorf("stop DNS TCP server: %w", err))
 	}
 
-	if s.tcpDNATConfigured && s.firewall != nil {
-		if err := s.firewall.RemoveOutputDNAT(s.listenIP, firewall.ProtocolTCP, DefaultPort, s.listenPort); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("remove DNS TCP DNAT rule: %w", err))
-		}
-		s.tcpDNATConfigured = false
-	}
-
-	if s.ebpfService != nil {
-		if err := s.ebpfService.FreeDNSFwd(); err != nil {
-			merr = multierror.Append(merr, fmt.Errorf("stop traffic forwarder: %w", err))
-		}
-	}
-
 	return nberrors.FormatErrorOrNil(merr)
 }
 
@@ -177,11 +238,23 @@ func (s *serviceViaListener) RuntimePort() int {
 	s.listenerFlagLock.Lock()
 	defer s.listenerFlagLock.Unlock()
 
-	if s.ebpfService != nil {
+	if s.redirectInstalled() {
 		return DefaultPort
-	} else {
-		return int(s.listenPort)
 	}
+	return int(s.listenPort)
+}
+
+// redirectInstalled reports whether every protocol is redirected from port 53
+// to the address and port the listener currently serves. Rules left over from
+// an earlier listener do not count.
+func (s *serviceViaListener) redirectInstalled() bool {
+	for _, proto := range dnatProtocols {
+		current := dnatRule{protocol: proto, ip: s.listenIP, port: s.listenPort}
+		if !slices.Contains(s.dnatRules, current) {
+			return false
+		}
+	}
+	return true
 }
 
 func (s *serviceViaListener) RuntimeIP() netip.Addr {
@@ -190,30 +263,29 @@ func (s *serviceViaListener) RuntimeIP() netip.Addr {
 
 // evalListenAddress figures out the listen address for the DNS server.
 // IPv4-only: all peers have a v4 overlay address, and DNS config points to v4.
-// First checks port 53 on WG interface or lo, then tries eBPF on a random port,
-// then falls back to port 5053.
+// Prefers port 53 on the overlay interface or lo, so no redirect is needed at
+// all; when it is taken it falls back to port 5053 and then to a random free
+// port, both of which need the port 53 redirect set up by setupDNAT.
 func (s *serviceViaListener) evalListenAddress() (netip.Addr, uint16, error) {
 	if s.customAddr != nil {
 		return s.customAddr.Addr(), s.customAddr.Port(), nil
 	}
 
-	ip, ok := s.testFreePort(DefaultPort)
-	if ok {
+	if ip, ok := s.testFreePort(DefaultPort); ok {
 		return ip, DefaultPort, nil
 	}
 
-	ebpfSrv, port, ok := s.tryToUseeBPF()
-	if ok {
-		s.ebpfService = ebpfSrv
-		return s.wgInterface.Address().IP, port, nil
-	}
-
-	ip, ok = s.testFreePort(customPort)
-	if ok {
+	if ip, ok := s.testFreePort(customPort); ok {
 		return ip, customPort, nil
 	}
 
-	return netip.Addr{}, 0, fmt.Errorf("failed to find a free port for DNS server")
+	ip := s.wgInterface.Address().IP
+	port, err := s.randomFreePort(ip)
+	if err != nil {
+		return netip.Addr{}, 0, fmt.Errorf("find a free port for DNS server: %w", err)
+	}
+
+	return ip, port, nil
 }
 
 func (s *serviceViaListener) testFreePort(port int) (netip.Addr, bool) {
@@ -260,48 +332,25 @@ func (s *serviceViaListener) tryToBind(ip netip.Addr, port int) bool {
 	return true
 }
 
-// tryToUseeBPF decides whether to apply eBPF program to capture DNS traffic on port 53.
-// This is needed because on some operating systems if we start a DNS server not on a default port 53,
-// the domain name  resolution won't work. So, in case we are running on Linux and picked a free
-// port we should fall back to the eBPF solution that will capture traffic on port 53 and forward
-// it to a local DNS server running on the chosen port.
-func (s *serviceViaListener) tryToUseeBPF() (ebpfMgr.Manager, uint16, bool) {
-	if runtime.GOOS != "linux" {
-		return nil, 0, false
+// randomFreePort returns a port that is free on ip for both UDP and TCP, since
+// the DNS server binds both. The probe listeners are closed again, so the port
+// is only likely, not guaranteed, to still be free when the server binds it.
+func (s *serviceViaListener) randomFreePort(ip netip.Addr) (uint16, error) {
+	for range randomPortAttempts {
+		probeListener, err := net.ListenUDP("udp4", &net.UDPAddr{})
+		if err != nil {
+			return 0, fmt.Errorf("bind random port: %w", err)
+		}
+
+		port := uint16(probeListener.LocalAddr().(*net.UDPAddr).Port)
+		if err := probeListener.Close(); err != nil {
+			return 0, fmt.Errorf("free up probed port: %w", err)
+		}
+
+		if s.tryToBind(ip, int(port)) {
+			return port, nil
+		}
 	}
 
-	port, err := s.generateFreePort() //nolint:staticcheck,unused
-	if err != nil {
-		log.Warnf("failed to generate a free port for eBPF DNS forwarder server: %s", err)
-		return nil, 0, false
-	}
-
-	ebpfSrv := ebpf.GetEbpfManagerInstance()
-	err = ebpfSrv.LoadDNSFwd(s.wgInterface.Address().IP, int(port))
-	if err != nil {
-		log.Warnf("failed to load DNS forwarder eBPF program, error: %s", err)
-		return nil, 0, false
-	}
-
-	return ebpfSrv, port, true
-}
-
-func (s *serviceViaListener) generateFreePort() (uint16, error) {
-	ok := s.tryToBind(s.wgInterface.Address().IP, customPort)
-	if ok {
-		return customPort, nil
-	}
-
-	probeListener, err := net.ListenUDP("udp4", &net.UDPAddr{})
-	if err != nil {
-		log.Debugf("failed to bind random port for DNS: %s", err)
-		return 0, err
-	}
-
-	port := uint16(probeListener.LocalAddr().(*net.UDPAddr).Port)
-	if err = probeListener.Close(); err != nil {
-		log.Debugf("failed to free up DNS port: %s", err)
-		return 0, err
-	}
-	return port, nil
+	return 0, fmt.Errorf("no port free for UDP and TCP on %s after %d attempts", ip, randomPortAttempts)
 }

client/internal/dns/service_listener_test.go

@@ -1,6 +1,7 @@
 package dns
 
 import (
+	"errors"
 	"fmt"
 	"net"
 	"net/netip"
@@ -10,6 +11,8 @@ import (
 	"github.com/miekg/dns"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+
+	firewall "github.com/netbirdio/netbird/client/firewall/manager"
 )
 
 func TestServiceViaListener_TCPAndUDP(t *testing.T) {
@@ -84,3 +87,133 @@ func TestServiceViaListener_TCPAndUDP(t *testing.T) {
 	require.NotEmpty(t, tcpResp.Answer)
 	assert.Contains(t, tcpResp.Answer[0].String(), "192.0.2.1", "TCP response should contain expected IP")
 }
+
+type dnatCall struct {
+	rule  dnatRule
+	added bool
+}
+
+// fakeFirewall records DNAT calls and fails the ones named in addErrs/removeErrs.
+type fakeFirewall struct {
+	calls      []dnatCall
+	addErrs    map[firewall.Protocol]error
+	removeErrs map[firewall.Protocol]error
+}
+
+func (f *fakeFirewall) AddOutputDNAT(ip netip.Addr, protocol firewall.Protocol, _, translatedPort uint16) error {
+	if err := f.addErrs[protocol]; err != nil {
+		return err
+	}
+	f.calls = append(f.calls, dnatCall{rule: dnatRule{protocol: protocol, ip: ip, port: translatedPort}, added: true})
+	return nil
+}
+
+func (f *fakeFirewall) RemoveOutputDNAT(ip netip.Addr, protocol firewall.Protocol, _, translatedPort uint16) error {
+	if err := f.removeErrs[protocol]; err != nil {
+		return err
+	}
+	f.calls = append(f.calls, dnatCall{rule: dnatRule{protocol: protocol, ip: ip, port: translatedPort}})
+	return nil
+}
+
+func newDNATTestService(fw Firewall) *serviceViaListener {
+	return &serviceViaListener{
+		listenIP:   netip.MustParseAddr("100.64.0.1"),
+		listenPort: customPort,
+		firewall:   fw,
+	}
+}
+
+func TestSetupDNAT_BothProtocols(t *testing.T) {
+	svc := newDNATTestService(&fakeFirewall{})
+
+	svc.setupDNAT()
+
+	assert.Len(t, svc.dnatRules, len(dnatProtocols))
+	assert.Equal(t, DefaultPort, svc.RuntimePort(), "port 53 is advertised once both redirects are installed")
+}
+
+func TestSetupDNAT_RollsBackPartialRedirect(t *testing.T) {
+	fw := &fakeFirewall{addErrs: map[firewall.Protocol]error{firewall.ProtocolTCP: errors.New("nftables busy")}}
+	svc := newDNATTestService(fw)
+
+	svc.setupDNAT()
+
+	assert.Empty(t, svc.dnatRules, "the UDP redirect installed before the failure must be rolled back")
+	assert.Equal(t, int(svc.listenPort), svc.RuntimePort(), "an incomplete redirect must not advertise port 53")
+	udp := dnatRule{protocol: firewall.ProtocolUDP, ip: svc.listenIP, port: svc.listenPort}
+	assert.Contains(t, fw.calls, dnatCall{rule: udp}, "UDP removal should have been attempted")
+}
+
+// A rollback that fails must keep the rule recorded, so port 53 is not left
+// redirected to a resolver that no longer listens.
+func TestStop_RetriesFailedDNATRemoval(t *testing.T) {
+	fw := &fakeFirewall{
+		addErrs:    map[firewall.Protocol]error{firewall.ProtocolTCP: errors.New("nftables busy")},
+		removeErrs: map[firewall.Protocol]error{firewall.ProtocolUDP: errors.New("nftables busy")},
+	}
+	svc := newDNATTestService(fw)
+
+	svc.setupDNAT()
+	udp := dnatRule{protocol: firewall.ProtocolUDP, ip: svc.listenIP, port: svc.listenPort}
+	require.Equal(t, []dnatRule{udp}, svc.dnatRules, "a failed rollback keeps the rule for a later retry")
+
+	require.Error(t, svc.Stop(), "the failing removal should be reported")
+	require.Equal(t, []dnatRule{udp}, svc.dnatRules)
+
+	delete(fw.removeErrs, firewall.ProtocolUDP)
+	require.NoError(t, svc.Stop(), "a later stop retries the removal")
+	assert.Empty(t, svc.dnatRules)
+}
+
+// A stale rule that cannot be removed is matched before anything added now, so
+// no new redirect may be installed on top of it and port 53 must not be
+// advertised as reaching this listener.
+func TestSetupDNAT_AbortsWhileStaleRuleRemains(t *testing.T) {
+	fw := &fakeFirewall{removeErrs: map[firewall.Protocol]error{firewall.ProtocolUDP: errors.New("nftables busy")}}
+	svc := newDNATTestService(fw)
+	stalePort := svc.listenPort
+
+	svc.setupDNAT()
+	require.Error(t, svc.Stop())
+	staleUDP := dnatRule{protocol: firewall.ProtocolUDP, ip: svc.listenIP, port: stalePort}
+	require.Equal(t, []dnatRule{staleUDP}, svc.dnatRules)
+
+	svc.listenPort = stalePort + 1
+	fw.calls = nil
+
+	svc.setupDNAT()
+
+	assert.Equal(t, []dnatRule{staleUDP}, svc.dnatRules, "the stale rule stays recorded for a later attempt")
+	for _, call := range fw.calls {
+		assert.False(t, call.added, "no redirect may be installed while a stale one is still in place")
+	}
+	assert.Equal(t, int(svc.listenPort), svc.RuntimePort(), "port 53 must not be advertised")
+}
+
+// A rule left behind by a failed removal must be removed with the address and
+// port it was installed with, even when the listener has since moved to another
+// port, and it must not count towards the redirect the new listener advertises.
+func TestSetupDNAT_ClearsStaleRuleAfterPortChange(t *testing.T) {
+	fw := &fakeFirewall{removeErrs: map[firewall.Protocol]error{firewall.ProtocolUDP: errors.New("nftables busy")}}
+	svc := newDNATTestService(fw)
+	stalePort := svc.listenPort
+
+	svc.setupDNAT()
+	require.Error(t, svc.Stop())
+	staleUDP := dnatRule{protocol: firewall.ProtocolUDP, ip: svc.listenIP, port: stalePort}
+	require.Equal(t, []dnatRule{staleUDP}, svc.dnatRules)
+
+	delete(fw.removeErrs, firewall.ProtocolUDP)
+	svc.listenPort = stalePort + 1
+	fw.calls = nil
+
+	svc.setupDNAT()
+
+	assert.Contains(t, fw.calls, dnatCall{rule: staleUDP}, "the stale rule must be removed with its original port")
+	assert.Len(t, svc.dnatRules, len(dnatProtocols))
+	assert.Equal(t, DefaultPort, svc.RuntimePort(), "the new listener is fully redirected")
+	for _, rule := range svc.dnatRules {
+		assert.Equal(t, svc.listenPort, rule.port, "only rules for the current listener remain")
+	}
+}

client/internal/ebpf/ebpf/bpf_bpfeb.go

@@ -1,5 +1,5 @@
 // Code generated by bpf2go; DO NOT EDIT.
-//go:build arm64be || armbe || mips || mips64 || mips64p32 || ppc64 || s390 || s390x || sparc || sparc64
+//go:build mips || mips64 || ppc64 || s390x
 
 package ebpf
 
@@ -47,9 +47,10 @@ func loadBpfObjects(obj interface{}, opts *ebpf.CollectionOptions) error {
 type bpfSpecs struct {
 	bpfProgramSpecs
 	bpfMapSpecs
+	bpfVariableSpecs
 }
 
-// bpfSpecs contains programs before they are loaded into the kernel.
+// bpfProgramSpecs contains programs before they are loaded into the kernel.
 //
 // It can be passed ebpf.CollectionSpec.Assign.
 type bpfProgramSpecs struct {
@@ -61,17 +62,28 @@ type bpfProgramSpecs struct {
 // It can be passed ebpf.CollectionSpec.Assign.
 type bpfMapSpecs struct {
 	NbFeatures           *ebpf.MapSpec `ebpf:"nb_features"`
-	NbMapDnsIp           *ebpf.MapSpec `ebpf:"nb_map_dns_ip"`
-	NbMapDnsPort         *ebpf.MapSpec `ebpf:"nb_map_dns_port"`
 	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 {
@@ -86,20 +98,28 @@ func (o *bpfObjects) Close() error {
 // It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
 type bpfMaps struct {
 	NbFeatures           *ebpf.Map `ebpf:"nb_features"`
-	NbMapDnsIp           *ebpf.Map `ebpf:"nb_map_dns_ip"`
-	NbMapDnsPort         *ebpf.Map `ebpf:"nb_map_dns_port"`
 	NbWgProxySettingsMap *ebpf.Map `ebpf:"nb_wg_proxy_settings_map"`
 }
 
 func (m *bpfMaps) Close() error {
 	return _BpfClose(
 		m.NbFeatures,
-		m.NbMapDnsIp,
-		m.NbMapDnsPort,
 		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.

client/internal/ebpf/ebpf/bpf_bpfel.go

@@ -1,5 +1,5 @@
 // Code generated by bpf2go; DO NOT EDIT.
-//go:build 386 || amd64 || amd64p32 || arm || arm64 || loong64 || mips64le || mips64p32le || mipsle || ppc64le || riscv64
+//go:build 386 || amd64 || arm || arm64 || loong64 || mips64le || mipsle || ppc64le || riscv64 || wasm
 
 package ebpf
 
@@ -47,9 +47,10 @@ func loadBpfObjects(obj interface{}, opts *ebpf.CollectionOptions) error {
 type bpfSpecs struct {
 	bpfProgramSpecs
 	bpfMapSpecs
+	bpfVariableSpecs
 }
 
-// bpfSpecs contains programs before they are loaded into the kernel.
+// bpfProgramSpecs contains programs before they are loaded into the kernel.
 //
 // It can be passed ebpf.CollectionSpec.Assign.
 type bpfProgramSpecs struct {
@@ -61,17 +62,28 @@ type bpfProgramSpecs struct {
 // It can be passed ebpf.CollectionSpec.Assign.
 type bpfMapSpecs struct {
 	NbFeatures           *ebpf.MapSpec `ebpf:"nb_features"`
-	NbMapDnsIp           *ebpf.MapSpec `ebpf:"nb_map_dns_ip"`
-	NbMapDnsPort         *ebpf.MapSpec `ebpf:"nb_map_dns_port"`
 	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 {
@@ -86,20 +98,28 @@ func (o *bpfObjects) Close() error {
 // It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
 type bpfMaps struct {
 	NbFeatures           *ebpf.Map `ebpf:"nb_features"`
-	NbMapDnsIp           *ebpf.Map `ebpf:"nb_map_dns_ip"`
-	NbMapDnsPort         *ebpf.Map `ebpf:"nb_map_dns_port"`
 	NbWgProxySettingsMap *ebpf.Map `ebpf:"nb_wg_proxy_settings_map"`
 }
 
 func (m *bpfMaps) Close() error {
 	return _BpfClose(
 		m.NbFeatures,
-		m.NbMapDnsIp,
-		m.NbMapDnsPort,
 		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.

client/internal/ebpf/ebpf/dns_fwd_linux.go

@@ -1,52 +0,0 @@
-package ebpf
-
-import (
-	"encoding/binary"
-	"fmt"
-	"net/netip"
-
-	log "github.com/sirupsen/logrus"
-)
-
-const (
-	mapKeyDNSIP   uint32 = 0
-	mapKeyDNSPort uint32 = 1
-)
-
-func (tf *GeneralManager) LoadDNSFwd(ip netip.Addr, dnsPort int) error {
-	log.Debugf("load eBPF DNS forwarder, watching addr: %s:53, redirect to port: %d", ip, dnsPort)
-	tf.lock.Lock()
-	defer tf.lock.Unlock()
-
-	err := tf.loadXdp()
-	if err != nil {
-		return err
-	}
-
-	if !ip.Is4() {
-		return fmt.Errorf("eBPF DNS forwarder only supports IPv4, got %s", ip)
-	}
-	ip4 := ip.As4()
-	err = tf.bpfObjs.NbMapDnsIp.Put(mapKeyDNSIP, binary.BigEndian.Uint32(ip4[:]))
-	if err != nil {
-		return err
-	}
-
-	err = tf.bpfObjs.NbMapDnsPort.Put(mapKeyDNSPort, uint16(dnsPort))
-	if err != nil {
-		return err
-	}
-
-	tf.setFeatureFlag(featureFlagDnsForwarder)
-	err = tf.bpfObjs.NbFeatures.Put(mapKeyFeatures, tf.featureFlags)
-	if err != nil {
-		return err
-	}
-	return nil
-}
-
-func (tf *GeneralManager) FreeDNSFwd() error {
-	log.Debugf("free ebpf DNS forwarder")
-	return tf.unsetFeatureFlag(featureFlagDnsForwarder)
-}
-

client/internal/ebpf/ebpf/manager_linux.go

@@ -15,8 +15,7 @@ import (
 const (
 	mapKeyFeatures uint32 = 0
 
-	featureFlagWGProxy      = 0b00000001
-	featureFlagDnsForwarder = 0b00000010
+	featureFlagWGProxy = 0b00000001
 )
 
 var (
@@ -28,9 +27,9 @@ var (
 
 // 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., dns_fwd.c and wg_proxy.c).
+// 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
+//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

client/internal/ebpf/ebpf/manager_linux_test.go

@@ -7,33 +7,24 @@ import (
 func TestManager_setFeatureFlag(t *testing.T) {
 	mgr := GeneralManager{}
 	mgr.setFeatureFlag(featureFlagWGProxy)
-	if mgr.featureFlags != 1 {
+	if mgr.featureFlags != featureFlagWGProxy {
 		t.Errorf("invalid feature state")
 	}
 
-	mgr.setFeatureFlag(featureFlagDnsForwarder)
-	if mgr.featureFlags != 3 {
-		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)
-	mgr.setFeatureFlag(featureFlagDnsForwarder)
 
 	err := mgr.unsetFeatureFlag(featureFlagWGProxy)
 	if err != nil {
 		t.Errorf("unexpected error: %s", err)
 	}
-	if mgr.featureFlags != 2 {
-		t.Errorf("invalid feature state, expected: %d, got: %d", 2, mgr.featureFlags)
-	}
-
-	err = mgr.unsetFeatureFlag(featureFlagDnsForwarder)
-	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

@@ -0,0 +1,16 @@
+// 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/dns_fwd.c

@@ -1,67 +0,0 @@
-const __u32 map_key_dns_ip = 0;
-const __u32 map_key_dns_port = 1;
-
-struct bpf_map_def SEC("maps") nb_map_dns_ip = {
-	.type = BPF_MAP_TYPE_ARRAY,
-	.key_size = sizeof(__u32),
-	.value_size = sizeof(__u32),
-	.max_entries = 10,
-};
-
-struct bpf_map_def SEC("maps") nb_map_dns_port = {
-	.type = BPF_MAP_TYPE_ARRAY,
-	.key_size = sizeof(__u32),
-	.value_size = sizeof(__u16),
-	.max_entries = 10,
-};
-
-__be32 dns_ip = 0;
-__be16 dns_port = 0;
-
-// 13568 is 53 in big endian
-__be16 GENERAL_DNS_PORT = 13568;
-
-bool read_settings() {
-    __u16 *port_value;
-    __u32 *ip_value;
-
-    // read dns ip
-    ip_value = bpf_map_lookup_elem(&nb_map_dns_ip, &map_key_dns_ip);
-    if(!ip_value) {
-        return false;
-    }
-    dns_ip = htonl(*ip_value);
-
-    // read dns port
-    port_value = bpf_map_lookup_elem(&nb_map_dns_port, &map_key_dns_port);
-    if (!port_value) {
-        return false;
-    }
-    dns_port = htons(*port_value);
-    return true;
-}
-
-int xdp_dns_fwd(struct iphdr  *ip, struct udphdr *udp) {
-    if (dns_port == 0) {
-        if(!read_settings()){
-            return XDP_PASS;
-        }
-        // bpf_printk("dns port: %d", ntohs(dns_port));
-        // bpf_printk("dns ip: %d", ntohl(dns_ip));
-    }
-
-    if (udp->dest == GENERAL_DNS_PORT && ip->daddr == dns_ip) {
-        udp->dest = dns_port;
-        // Clear the now-stale checksum; zero means "not computed" for IPv4.
-        udp->check = 0;
-        return XDP_PASS;
-    }
-
-    if (udp->source == dns_port && ip->saddr == dns_ip) {
-        udp->source = GENERAL_DNS_PORT;
-        udp->check = 0;
-        return XDP_PASS;
-    }
-
-    return XDP_PASS;
-}

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

@@ -5,11 +5,9 @@
 #include <netinet/in.h>
 #include <linux/bpf.h>
 #include <bpf/bpf_helpers.h>
-#include "dns_fwd.c"
 #include "wg_proxy.c"
 
 const __u16 flag_feature_wg_proxy = 0b01;
-const __u16 flag_feature_dns_fwd = 0b10;
 
 const __u32 map_key_features = 0;
 struct bpf_map_def SEC("maps") nb_features = {
@@ -48,10 +46,6 @@ int nb_xdp_prog(struct xdp_md *ctx) {
        return XDP_PASS;
     }
 
-    if (*features & flag_feature_dns_fwd) {
-        xdp_dns_fwd(ip, udp);
-    }
-
     if (*features & flag_feature_wg_proxy) {
         xdp_wg_proxy(ip, udp);
     }

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

@@ -1,8 +1,18 @@
-# DNS forwarder
+# XDP programs
 
-The agent attach the XDP program to the lo device. We can not use fake address in eBPF because the
-traffic does not appear in the eBPF program. The program capture the traffic on wg_ip:53 and 
-overwrite in it the destination port to 5053.
+`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
 

client/internal/ebpf/manager/manager.go

@@ -1,11 +1,7 @@
 package manager
 
-import "net/netip"
-
-// Manager is used to load multiple eBPF programs. E.g., current DNS programs and WireGuard proxy
+// Manager is used to load multiple eBPF programs. E.g., the WireGuard proxy
 type Manager interface {
-	LoadDNSFwd(ip netip.Addr, dnsPort int) error
-	FreeDNSFwd() error
 	LoadWgProxy(proxyPort, wgPort int) error
 	FreeWGProxy() error
 }

client/internal/engine.go

@@ -14,12 +14,14 @@ import (
 	"sort"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/hashicorp/go-multierror"
 	"github.com/pion/ice/v4"
 	"github.com/pion/stun/v3"
 	log "github.com/sirupsen/logrus"
+	wgdevice "golang.zx2c4.com/wireguard/device"
 	"golang.zx2c4.com/wireguard/tun/netstack"
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 
@@ -237,6 +239,12 @@ type Engine struct {
 
 	wgInterface WGIface
 
+	// wgDevice is a lock-free handle on the WireGuard device behind
+	// wgInterface. Reaching the device through wgInterface requires
+	// syncMsgMux, which handleSync holds while it adds and removes peers;
+	// SetPerformance must stay reachable exactly when that work is stuck.
+	wgDevice atomic.Pointer[wgdevice.Device]
+
 	udpMux *udpmux.UniversalUDPMuxDefault
 
 	// networkSerial is the latest CurrentSerial (state ID) of the network sent by the Management service
@@ -652,6 +660,7 @@ func (e *Engine) Start(netbirdConfig *mgmProto.NetbirdConfig, mgmtURL *url.URL)
 		log.Errorf("failed to pull up wgInterface [%s]: %s", e.wgInterface.Name(), err.Error())
 		return fmt.Errorf("up wg interface: %w", err)
 	}
+	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
@@ -2155,6 +2164,10 @@ func (e *Engine) close() {
 	log.Debugf("removing Netbird interface %s", e.config.WgIfaceName)
 
 	if e.wgInterface != nil {
+		// Drop the handle before the close starts: a retune that loads it
+		// afterwards would touch a device on its way out and report success
+		// for an engine that is already gone.
+		e.wgDevice.Store(nil)
 		if err := e.wgInterface.Close(); err != nil {
 			log.Errorf("failed closing Netbird interface %s %v", e.config.WgIfaceName, err)
 		}
@@ -2314,15 +2327,16 @@ type Performance struct {
 }
 
 // SetPerformance applies the given tuning to this engine's live Device.
+//
+// It deliberately does not take syncMsgMux. Raising the buffer pool cap is the
+// recovery path for a device whose pool is exhausted, and an exhausted pool
+// blocks peer removal inside handleSync, which holds syncMsgMux for as long as
+// it stays blocked. Taking the lock here would make the retune unreachable in
+// the one situation that needs it.
 func (e *Engine) SetPerformance(t Performance) error {
-	e.syncMsgMux.Lock()
-	defer e.syncMsgMux.Unlock()
-	if e.wgInterface == nil {
-		return fmt.Errorf("wg interface not initialized")
-	}
-	dev := e.wgInterface.GetWGDevice()
+	dev := e.wgDevice.Load()
 	if dev == nil {
-		return fmt.Errorf("wg device not initialized")
+		return errors.New("wg device not initialized")
 	}
 	if t.PreallocatedBuffersPerPool != nil {
 		dev.SetPreallocatedBuffersPerPool(*t.PreallocatedBuffersPerPool)

client/internal/peer/handshaker.go

@@ -116,7 +116,7 @@ func (h *Handshaker) Listen(ctx context.Context) {
 	for {
 		select {
 		case remoteOfferAnswer := <-h.remoteOffersCh:
-			h.log.Infof("received offer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials())
+			h.log.Infof("received offer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t, relay server: %s, relay IP: %s", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials(), remoteOfferAnswer.RelaySrvAddress, remoteOfferAnswer.RelaySrvIP)
 
 			// Record signaling received for reconnection attempts
 			if h.metricsStages != nil {
@@ -138,7 +138,7 @@ func (h *Handshaker) Listen(ctx context.Context) {
 				continue
 			}
 		case remoteOfferAnswer := <-h.remoteAnswerCh:
-			h.log.Infof("received answer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials())
+			h.log.Infof("received answer, running version %s, remote WireGuard listen port %d, session id: %s, remote ICE supported: %t, relay server: %s, relay IP: %s", remoteOfferAnswer.Version, remoteOfferAnswer.WgListenPort, remoteOfferAnswer.SessionIDString(), remoteOfferAnswer.hasICECredentials(), remoteOfferAnswer.RelaySrvAddress, remoteOfferAnswer.RelaySrvIP)
 
 			// Record signaling received for reconnection attempts
 			if h.metricsStages != nil {
@@ -209,14 +209,14 @@ func (h *Handshaker) sendOffer() error {
 	}
 
 	offer := h.buildOfferAnswer()
-	h.log.Debugf("sending offer with serial: %s", offer.SessionIDString())
+	h.log.Debugf("sending offer with serial: %s, relay server: %s, relay IP: %s", offer.SessionIDString(), offer.RelaySrvAddress, offer.RelaySrvIP)
 
 	return h.signaler.SignalOffer(offer, h.config.Key)
 }
 
 func (h *Handshaker) sendAnswer() error {
 	answer := h.buildOfferAnswer()
-	h.log.Debugf("sending answer with serial: %s", answer.SessionIDString())
+	h.log.Debugf("sending answer with serial: %s, relay server: %s, relay IP: %s", answer.SessionIDString(), answer.RelaySrvAddress, answer.RelaySrvIP)
 
 	return h.signaler.SignalAnswer(answer, h.config.Key)
 }

client/internal/profilemanager/config.go

@@ -58,10 +58,6 @@ var DefaultInterfaceBlacklist = []string{
 	"Tailscale", "tailscale", "docker", "veth", "br-", "lo",
 }
 
-// loadMDMPolicy is the package-level indirection used by apply() to read the
-// active MDM policy. Tests override this to inject a fake policy.
-var loadMDMPolicy = mdm.LoadPolicy
-
 // ConfigInput carries configuration changes to the client
 type ConfigInput struct {
 	ManagementURL                 string
@@ -202,14 +198,26 @@ type Config struct {
 
 	MTU uint16
 
-	// policy is the MDM policy that produced the currently-set values for
-	// any MDM-enforced fields. Set by applyMDMPolicy at the tail of apply()
-	// and reset on every apply() invocation. Never persisted to disk.
-	// Callers query enforcement state via Policy() and the mdm.Policy API
-	// (HasKey, ManagedKeys, IsEmpty).
+	// 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
+	// state via Policy() and the mdm.Policy API (HasKey, ManagedKeys,
+	// IsEmpty).
 	policy *mdm.Policy `json:"-"`
 }
 
+// ApplyMDMPolicy overlays the supplied MDM Policy on top of the current
+// Config values and records it as Policy(). The overlay is not reversible:
+// an empty Policy only clears the enforcement metadata, so resolve the base
+// Config again (from disk or JSON) before applying a changed policy, the way
+// the lifecycle owners do on every load.
+func (config *Config) ApplyMDMPolicy(policy *mdm.Policy) {
+	if config == nil {
+		return
+	}
+	config.applyMDMPolicy(policy)
+}
+
 // Policy returns the MDM policy applied to this Config. Returns a non-nil
 // empty Policy when MDM enforcement is inactive; callers can always invoke
 // HasKey / ManagedKeys / IsEmpty without a nil check.
@@ -712,9 +720,11 @@ func (config *Config) apply(input ConfigInput) (updated bool, err error) {
 		updated = true
 	}
 
-	// MDM is the last override layer: any key present in the policy
-	// supersedes defaults, on-disk config, env vars and CLI input.
-	config.applyMDMPolicy(loadMDMPolicy())
+	// Initialise the MDM overlay to "no enforcement" so Config.Policy()
+	// never returns a stale or nil policy on a freshly applied Config.
+	// Lifecycle owners that want to enforce a real MDM policy invoke
+	// Config.ApplyMDMPolicy(loader.Load()) after this returns.
+	config.applyMDMPolicy(mdm.NewPolicy(nil))
 
 	return updated, nil
 }

client/internal/profilemanager/config_mdm.go

@@ -0,0 +1,52 @@
+package profilemanager
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/netbirdio/netbird/client/mdm"
+)
+
+// ErrMDMManagedFields marks a config change rejected because it diverges from
+// MDM-enforced values.
+var ErrMDMManagedFields = errors.New("fields managed by MDM cannot be modified")
+
+// MDMConflicts returns the names of MDM-managed keys whose requested value in
+// the ConfigInput differs from the policy-enforced value; a field set to the
+// enforced value is a no-op echo, not a conflict.
+func MDMConflicts(input ConfigInput, policy *mdm.Policy) []string {
+	pskGot := input.PreSharedKey
+	if isPreSharedKeyHidden(pskGot) {
+		pskGot = nil
+	}
+	var port *int64
+	if input.WireguardPort != nil {
+		v := int64(*input.WireguardPort)
+		port = &v
+	}
+	return mdm.ResolveConflicts(policy, []mdm.ConflictCheck{
+		mdm.ConflictURL(mdm.KeyManagementURL, input.ManagementURL),
+		mdm.ConflictStringPtr(mdm.KeyPreSharedKey, pskGot),
+		mdm.ConflictBool(mdm.KeyRosenpassEnabled, input.RosenpassEnabled),
+		mdm.ConflictBool(mdm.KeyRosenpassPermissive, input.RosenpassPermissive),
+		mdm.ConflictBool(mdm.KeyDisableAutoConnect, input.DisableAutoConnect),
+		mdm.ConflictBool(mdm.KeyAllowServerSSH, input.ServerSSHAllowed),
+		mdm.ConflictBool(mdm.KeyRemoteJobsAllowed, input.RemoteJobsAllowed),
+		mdm.ConflictBool(mdm.KeyDisableClientRoutes, input.DisableClientRoutes),
+		mdm.ConflictBool(mdm.KeyDisableServerRoutes, input.DisableServerRoutes),
+		mdm.ConflictBool(mdm.KeyBlockInbound, input.BlockInbound),
+		mdm.ConflictInt64(mdm.KeyWireguardPort, port),
+		mdm.ConflictBool(mdm.KeyEnableLocalMetrics, input.LocalMetricsEnabled),
+		mdm.ConflictStringPtr(mdm.KeyLocalMetricsAddress, input.LocalMetricsAddress),
+	})
+}
+
+// CheckMDMConflicts returns an ErrMDMManagedFields-wrapped error naming the
+// conflicting keys, or nil when the input does not fight the policy.
+func CheckMDMConflicts(input ConfigInput, policy *mdm.Policy) error {
+	conflicts := MDMConflicts(input, policy)
+	if len(conflicts) == 0 {
+		return nil
+	}
+	return fmt.Errorf("%w: %v", ErrMDMManagedFields, conflicts)
+}

client/internal/profilemanager/config_mdm_test.go

@@ -10,24 +10,58 @@ import (
 	"github.com/netbirdio/netbird/client/mdm"
 )
 
-// withMDMPolicy temporarily overrides the package-level loadMDMPolicy hook so
-// apply() observes the supplied Policy. The original loader is restored at
-// test cleanup.
-func withMDMPolicy(t *testing.T, policy *mdm.Policy) {
+// fakeFetcher implements mdm.PolicyFetcher returning a pre-set policy
+// map. Test helper used to construct a Loader without touching the OS
+// or any package-level state.
+type fakeFetcher struct{ values map[string]any }
+
+func (f *fakeFetcher) Fetch() map[string]any { return f.values }
+
+// loaderFor builds an mdm.Loader whose loadPlatform returns the
+// supplied Policy's underlying values.
+func loaderFor(policy *mdm.Policy) *mdm.Loader {
+	if policy == nil || policy.IsEmpty() {
+		return mdm.NewLoader(&fakeFetcher{values: nil})
+	}
+	values := make(map[string]any)
+	for _, k := range policy.ManagedKeys() {
+		if v, ok := policy.GetString(k); ok {
+			values[k] = v
+			continue
+		}
+		if v, ok := policy.GetInt(k); ok {
+			values[k] = v
+			continue
+		}
+		if v, ok := policy.GetBool(k); ok {
+			values[k] = v
+			continue
+		}
+		if v, ok := policy.GetStringSlice(k); ok {
+			values[k] = v
+		}
+	}
+	return mdm.NewLoader(&fakeFetcher{values: values})
+}
+
+// configWithMDM is the test convenience that builds a Config via
+// UpdateOrCreateConfig and overlays the supplied MDM policy on top —
+// mirrors the production pattern (Server.getConfig / Client.applyMDMOverlay)
+// where the Loader lives outside Config and the apply step is driven
+// by the lifecycle owner.
+func configWithMDM(t *testing.T, input ConfigInput, policy *mdm.Policy) *Config {
 	t.Helper()
-	prev := loadMDMPolicy
-	loadMDMPolicy = func() *mdm.Policy { return policy }
-	t.Cleanup(func() { loadMDMPolicy = prev })
+	cfg, err := UpdateOrCreateConfig(input)
+	require.NoError(t, err)
+	require.NotNil(t, cfg)
+	cfg.ApplyMDMPolicy(loaderFor(policy).Load())
+	return cfg
 }
 
 func TestApply_MDMEmpty_NoEnforcement(t *testing.T) {
-	withMDMPolicy(t, mdm.NewPolicy(nil))
-
-	cfg, err := UpdateOrCreateConfig(ConfigInput{
+	cfg := configWithMDM(t, ConfigInput{
 		ConfigPath: filepath.Join(t.TempDir(), "config.json"),
-	})
-	require.NoError(t, err)
-	require.NotNil(t, cfg)
+	}, mdm.NewPolicy(nil))
 
 	assert.True(t, cfg.Policy().IsEmpty(), "no MDM source ⇒ empty Policy")
 	assert.False(t, cfg.Policy().HasKey(mdm.KeyManagementURL))
@@ -39,18 +73,15 @@ func TestApply_MDMEmpty_NoEnforcement(t *testing.T) {
 
 func TestApply_MDMOnly_OverridesDefaults(t *testing.T) {
 	const mdmURL = "https://corp.mdm.example.com:443"
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+
+	cfg := configWithMDM(t, ConfigInput{
+		ConfigPath: filepath.Join(t.TempDir(), "config.json"),
+	}, mdm.NewPolicy(map[string]any{
 		mdm.KeyManagementURL:       mdmURL,
 		mdm.KeyDisableClientRoutes: true,
 		mdm.KeyBlockInbound:        true,
 	}))
 
-	cfg, err := UpdateOrCreateConfig(ConfigInput{
-		ConfigPath: filepath.Join(t.TempDir(), "config.json"),
-	})
-	require.NoError(t, err)
-	require.NotNil(t, cfg)
-
 	assert.Equal(t, mdmURL, cfg.ManagementURL.String())
 	assert.True(t, cfg.DisableClientRoutes)
 	assert.True(t, cfg.BlockInbound)
@@ -65,16 +96,12 @@ func TestApply_MDMBeatsCLIInput(t *testing.T) {
 	const mdmURL = "https://mdm.example.com:443"
 	const cliURL = "https://cli.example.com:443"
 
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
-		mdm.KeyManagementURL: mdmURL,
-	}))
-
-	cfg, err := UpdateOrCreateConfig(ConfigInput{
+	cfg := configWithMDM(t, ConfigInput{
 		ConfigPath:    filepath.Join(t.TempDir(), "config.json"),
 		ManagementURL: cliURL,
-	})
-	require.NoError(t, err)
-	require.NotNil(t, cfg)
+	}, mdm.NewPolicy(map[string]any{
+		mdm.KeyManagementURL: mdmURL,
+	}))
 
 	// MDM wins over CLI-supplied management URL.
 	assert.Equal(t, mdmURL, cfg.ManagementURL.String())
@@ -82,16 +109,12 @@ func TestApply_MDMBeatsCLIInput(t *testing.T) {
 }
 
 func TestApply_MDMInvalidURL_KeepsPreviousValue(t *testing.T) {
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	cfg := configWithMDM(t, ConfigInput{
+		ConfigPath: filepath.Join(t.TempDir(), "config.json"),
+	}, mdm.NewPolicy(map[string]any{
 		mdm.KeyManagementURL: "not-a-url",
 	}))
 
-	cfg, err := UpdateOrCreateConfig(ConfigInput{
-		ConfigPath: filepath.Join(t.TempDir(), "config.json"),
-	})
-	require.NoError(t, err)
-	require.NotNil(t, cfg)
-
 	// Invalid MDM URL is logged and skipped: default URL stays in place
 	// to keep the client functional.
 	assert.Equal(t, DefaultManagementURL, cfg.ManagementURL.String())
@@ -106,24 +129,20 @@ func TestApply_MDMBoolKeysOverrideOnDiskValue(t *testing.T) {
 	tmp := filepath.Join(t.TempDir(), "config.json")
 
 	// Seed without MDM.
-	withMDMPolicy(t, mdm.NewPolicy(nil))
-	_, err := UpdateOrCreateConfig(ConfigInput{
+	configWithMDM(t, ConfigInput{
 		ConfigPath:          tmp,
 		DisableClientRoutes: boolPtr(false),
 		RosenpassEnabled:    boolPtr(false),
-	})
-	require.NoError(t, err)
+	}, mdm.NewPolicy(nil))
 
 	// Now enable MDM enforcement for these keys.
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	cfg := configWithMDM(t, ConfigInput{
+		ConfigPath: tmp,
+	}, mdm.NewPolicy(map[string]any{
 		mdm.KeyDisableClientRoutes: true,
 		mdm.KeyRosenpassEnabled:    true,
 	}))
 
-	cfg, err := UpdateOrCreateConfig(ConfigInput{ConfigPath: tmp})
-	require.NoError(t, err)
-	require.NotNil(t, cfg)
-
 	assert.True(t, cfg.DisableClientRoutes, "MDM override should flip on-disk false to true")
 	assert.True(t, cfg.RosenpassEnabled)
 	assert.True(t, cfg.Policy().HasKey(mdm.KeyDisableClientRoutes))
@@ -134,22 +153,19 @@ func TestApply_MDMLocalMetrics(t *testing.T) {
 	tmp := filepath.Join(t.TempDir(), "config.json")
 
 	// Seed without MDM.
-	withMDMPolicy(t, mdm.NewPolicy(nil))
-	_, err := UpdateOrCreateConfig(ConfigInput{
+	configWithMDM(t, ConfigInput{
 		ConfigPath:          tmp,
 		LocalMetricsEnabled: boolPtr(false),
-	})
-	require.NoError(t, err)
+	}, mdm.NewPolicy(nil))
 
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	// Now enable MDM enforcement for these keys.
+	cfg := configWithMDM(t, ConfigInput{
+		ConfigPath: tmp,
+	}, mdm.NewPolicy(map[string]any{
 		mdm.KeyEnableLocalMetrics:  true,
 		mdm.KeyLocalMetricsAddress: "127.0.0.1:9292",
 	}))
 
-	cfg, err := UpdateOrCreateConfig(ConfigInput{ConfigPath: tmp})
-	require.NoError(t, err)
-	require.NotNil(t, cfg)
-
 	assert.True(t, cfg.LocalMetricsEnabled, "MDM override should flip on-disk false to true")
 	assert.Equal(t, "127.0.0.1:9292", cfg.LocalMetricsAddress)
 	assert.True(t, cfg.Policy().HasKey(mdm.KeyEnableLocalMetrics))
@@ -171,16 +187,12 @@ func TestApply_MDMLazyConnection(t *testing.T) {
 	}
 	for _, c := range cases {
 		t.Run(c.name, func(t *testing.T) {
-			withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+			cfg := configWithMDM(t, ConfigInput{
+				ConfigPath: filepath.Join(t.TempDir(), "config.json"),
+			}, mdm.NewPolicy(map[string]any{
 				mdm.KeyLazyConnection: c.raw,
 			}))
 
-			cfg, err := UpdateOrCreateConfig(ConfigInput{
-				ConfigPath: filepath.Join(t.TempDir(), "config.json"),
-			})
-			require.NoError(t, err)
-			require.NotNil(t, cfg)
-
 			assert.Equal(t, c.want, cfg.LazyConnection)
 			assert.True(t, cfg.Policy().HasKey(mdm.KeyLazyConnection))
 		})
@@ -188,22 +200,83 @@ func TestApply_MDMLazyConnection(t *testing.T) {
 }
 
 func TestApply_MDMPreSharedKeyRedactionSentinelRejected(t *testing.T) {
-	const maskSentinel = "**********"
+	const maskSentinel = mdm.PreSharedKeyRedactedSentinel
 
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	cfg := configWithMDM(t, ConfigInput{
+		ConfigPath: filepath.Join(t.TempDir(), "config.json"),
+	}, mdm.NewPolicy(map[string]any{
 		mdm.KeyPreSharedKey: maskSentinel,
 	}))
 
-	cfg, err := UpdateOrCreateConfig(ConfigInput{
-		ConfigPath: filepath.Join(t.TempDir(), "config.json"),
-	})
-	require.NoError(t, err)
-	require.NotNil(t, cfg)
-
 	// Mask sentinel must not be persisted as the actual PSK.
 	assert.NotEqual(t, maskSentinel, cfg.PreSharedKey)
 	// Key still marked managed so user writes are still rejected.
 	assert.True(t, cfg.Policy().HasKey(mdm.KeyPreSharedKey))
 }
 
+func TestMDMConflicts_PreSharedKey(t *testing.T) {
+	policy := mdm.NewPolicy(map[string]any{
+		mdm.KeyPreSharedKey: "mdm-enforced-psk",
+	})
+	empty := ""
+	sentinel := mdm.PreSharedKeyRedactedSentinel
+	same := "mdm-enforced-psk"
+	other := "user-psk"
+
+	tests := []struct {
+		name string
+		psk  *string
+		want []string
+	}{
+		{name: "unset", psk: nil, want: nil},
+		{name: "explicit empty", psk: &empty, want: []string{mdm.KeyPreSharedKey}},
+		{name: "sentinel echo", psk: &sentinel, want: nil},
+		{name: "same value", psk: &same, want: nil},
+		{name: "divergent", psk: &other, want: []string{mdm.KeyPreSharedKey}},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			assert.Equal(t, tc.want, MDMConflicts(ConfigInput{PreSharedKey: tc.psk}, policy))
+		})
+	}
+}
+
+func TestMDMConflicts_RemoteJobsAndLocalMetrics(t *testing.T) {
+	policy := mdm.NewPolicy(map[string]any{
+		mdm.KeyRemoteJobsAllowed:   false,
+		mdm.KeyEnableLocalMetrics:  true,
+		mdm.KeyLocalMetricsAddress: "127.0.0.1:9999",
+	})
+	sameAddr := "127.0.0.1:9999"
+	otherAddr := "0.0.0.0:9999"
+	emptyAddr := ""
+
+	tests := []struct {
+		name  string
+		input ConfigInput
+		want  []string
+	}{
+		{name: "unset", input: ConfigInput{}, want: nil},
+		{name: "echo", input: ConfigInput{
+			RemoteJobsAllowed:   boolPtr(false),
+			LocalMetricsEnabled: boolPtr(true),
+			LocalMetricsAddress: &sameAddr,
+		}, want: nil},
+		{name: "remote jobs divergent", input: ConfigInput{RemoteJobsAllowed: boolPtr(true)}, want: []string{mdm.KeyRemoteJobsAllowed}},
+		{name: "metrics disabled", input: ConfigInput{LocalMetricsEnabled: boolPtr(false)}, want: []string{mdm.KeyEnableLocalMetrics}},
+		{name: "metrics address divergent", input: ConfigInput{LocalMetricsAddress: &otherAddr}, want: []string{mdm.KeyLocalMetricsAddress}},
+		{name: "metrics address explicit empty", input: ConfigInput{LocalMetricsAddress: &emptyAddr}, want: []string{mdm.KeyLocalMetricsAddress}},
+		{name: "all divergent", input: ConfigInput{
+			RemoteJobsAllowed:   boolPtr(true),
+			LocalMetricsEnabled: boolPtr(false),
+			LocalMetricsAddress: &otherAddr,
+		}, want: []string{mdm.KeyRemoteJobsAllowed, mdm.KeyEnableLocalMetrics, mdm.KeyLocalMetricsAddress}},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			assert.Equal(t, tc.want, MDMConflicts(tc.input, policy))
+		})
+	}
+}
+
 func boolPtr(b bool) *bool { return &b }

client/internal/profilemanager/invoking_user.go

@@ -6,6 +6,7 @@ import (
 	"os/user"
 	"path/filepath"
 	"runtime"
+	"strconv"
 
 	log "github.com/sirupsen/logrus"
 )
@@ -13,17 +14,21 @@ import (
 const envSudoUser = "SUDO_USER"
 
 var (
-	geteuid    = os.Geteuid
-	lookupUser = user.Lookup
+	currentUser = user.Current
+	getegid     = os.Getegid
+	geteuid     = os.Geteuid
+	lookupUser  = user.Lookup
 )
 
 // InvokingUser returns the user a CLI invocation acts for. Under sudo that is
 // the user who ran sudo, not root: privileged flags force commands through
 // sudo, and resolving profiles as root would silently switch the daemon to
-// root's (default) profile instead of the invoking user's. Privilege decisions
-// are not made here — those stay on the kernel credentials of the daemon
-// connection, which SUDO_USER (a plain environment variable) can never
-// influence; a forged value only selects a profile root could select anyway.
+// root's (default) profile instead of the invoking user's. An unmapped positive
+// process UID uses its numeric kernel identity; root, sudo lookup failures, and
+// unavailable platform identities still fail closed. Privilege decisions stay
+// on the kernel credentials of the daemon connection, which SUDO_USER (a plain
+// environment variable) can never influence; a forged value only selects a
+// profile root could select anyway.
 func InvokingUser() (*user.User, error) {
 	if u, ok := sudoInvokingUser(); ok {
 		return u, nil
@@ -35,7 +40,23 @@ func InvokingUser() (*user.User, error) {
 	if sudoActive() {
 		return nil, fmt.Errorf("resolve sudo invoking user %q: refusing to fall back to root", os.Getenv(envSudoUser))
 	}
-	return user.Current()
+	u, err := currentUser()
+	if err == nil {
+		return u, nil
+	}
+
+	uid := geteuid()
+	if uid <= 0 {
+		return nil, err
+	}
+
+	log.Debugf("current user lookup for UID %d: %v; using numeric UID", uid, err)
+	uidString := strconv.Itoa(uid)
+	return &user.User{
+		Username: uidString,
+		Uid:      uidString,
+		Gid:      strconv.Itoa(getegid()),
+	}, nil
 }
 
 // IsPlainRoot reports that the process runs as root with no usable sudo

client/internal/profilemanager/invoking_user_test.go

@@ -2,6 +2,7 @@ package profilemanager
 
 import (
 	"errors"
+	"fmt"
 	"io/fs"
 	"os"
 	"os/user"
@@ -21,7 +22,51 @@ func TestInvokingUserFallsBackToProcessUser(t *testing.T) {
 
 	current, err := user.Current()
 	require.NoError(t, err)
-	assert.Equal(t, current.Username, got.Username)
+	assert.Equal(t, current.Username, got.Username, "invoking user should match the process user without sudo")
+}
+
+func TestInvokingUserFailsClosedWithoutPositiveUID(t *testing.T) {
+	for _, uid := range []int{0, -1} {
+		t.Run(fmt.Sprintf("UID%d", uid), func(t *testing.T) {
+			t.Setenv(envSudoUser, "")
+			lookupErr := errors.New("current user unavailable")
+			fakeUnmappedUser(t, uid, 0, lookupErr)
+
+			got, err := InvokingUser()
+			require.ErrorIs(t, err, lookupErr)
+			assert.Nil(t, got, "root or unavailable UID must not become a synthetic identity")
+		})
+	}
+}
+
+func TestProfileFilePathUsesNumericIdentityForUnmappedNonRoot(t *testing.T) {
+	t.Setenv(envSudoUser, "")
+	fakeUnmappedUser(t, 1001230000, 0, errors.New("user: unknown userid 1001230000"))
+
+	profilesRoot := t.TempDir()
+	origDir := DefaultConfigPathDir
+	origOverride := ConfigDirOverride
+	DefaultConfigPathDir = profilesRoot
+	ConfigDirOverride = ""
+	t.Cleanup(func() {
+		DefaultConfigPathDir = origDir
+		ConfigDirOverride = origOverride
+	})
+
+	profileID := ID("0123456789abcdef0123456789abcdef")
+	got, err := (&Profile{ID: profileID}).FilePath()
+	require.NoError(t, err)
+	assert.Equal(t,
+		filepath.Join(profilesRoot, "1001230000", profileID.String()+".json"),
+		got,
+		"profile path should use the numeric UID namespace",
+	)
+
+	entries, err := os.ReadDir(profilesRoot)
+	require.NoError(t, err)
+	require.Len(t, entries, 1, "only the numeric UID directory should be created")
+	assert.Equal(t, "1001230000", entries[0].Name(), "profile namespace should be numeric")
+	assert.True(t, entries[0].IsDir(), "profile namespace should be a directory")
 }
 
 func TestSudoInvokingUserInactiveWithoutSudoContext(t *testing.T) {
@@ -60,6 +105,13 @@ func TestInvokingUserFailsClosedWhenSudoLookupFails(t *testing.T) {
 	fakeSudo(t, filepath.Join("/home", "misha"))
 	lookupUser = func(string) (*user.User, error) { return nil, errors.New("nss unavailable") }
 
+	origCurrentUser := currentUser
+	currentUser = func() (*user.User, error) {
+		t.Fatal("currentUser must not be called after a sudo lookup failure")
+		return nil, errors.New("currentUser called unexpectedly")
+	}
+	t.Cleanup(func() { currentUser = origCurrentUser })
+
 	got, err := InvokingUser()
 	require.Error(t, err)
 	assert.Nil(t, got, "must not resolve to the root process user")
@@ -215,6 +267,22 @@ func fakeSudo(t *testing.T, home string) {
 	})
 }
 
+func fakeUnmappedUser(t *testing.T, uid, gid int, lookupErr error) {
+	t.Helper()
+
+	origCurrentUser := currentUser
+	origEuid := geteuid
+	origEgid := getegid
+	currentUser = func() (*user.User, error) { return nil, lookupErr }
+	geteuid = func() int { return uid }
+	getegid = func() int { return gid }
+	t.Cleanup(func() {
+		currentUser = origCurrentUser
+		geteuid = origEuid
+		getegid = origEgid
+	})
+}
+
 func assertNoEntries(t *testing.T, root string) {
 	t.Helper()
 	err := filepath.WalkDir(root, func(path string, _ fs.DirEntry, err error) error {

client/ios/NetBirdSDK/client.go

@@ -88,9 +88,15 @@ type Client struct {
 	// netMgr outlives engine restarts: it mirrors the OS connectivity, not
 	// the engine lifecycle. Run injects its state and sweeper into each new
 	// ConnectClient.
-	netMgr *netevents.Manager
-	// preloadedConfig holds config loaded from JSON (used on tvOS where file writes are blocked)
-	preloadedConfig *profilemanager.Config
+	netMgr              *netevents.Manager
+	preloadedConfigJSON atomic.Pointer[string]
+
+	// mdmSource holds the per-Client MDM policy source and its change
+	// detector as one unit. Set by SetMDMPolicyFetcher (called from the
+	// Swift side at extension init). Each Run passes the loader to the
+	// resolved Config so applyMDMPolicy picks up the active overlay. Nil
+	// means "MDM enforcement off for this Client".
+	mdmSource atomic.Pointer[mdmSource]
 
 	// stateMu guards the run lifecycle as one unit: the cancel installed by
 	// the current run, the channel it closes on exit, and the state it
@@ -122,44 +128,44 @@ func NewClient(cfgFile, stateFile, cacheDir, logFilePath, deviceName string, osV
 	}
 }
 
-// SetConfigFromJSON loads config from a JSON string into memory.
-// This is used on tvOS where file writes to App Group containers are blocked.
-// When set, IsLoginRequired() and Run() will use this preloaded config instead of reading from file.
+// SetConfigFromJSON stores the JSON config that later loads resolve instead of the config file (tvOS).
 func (c *Client) SetConfigFromJSON(jsonStr string) error {
-	cfg, err := profilemanager.ConfigFromJSON(jsonStr)
-	if err != nil {
+	if _, err := profilemanager.ConfigFromJSON(jsonStr); err != nil {
 		log.Errorf("SetConfigFromJSON: failed to parse config JSON: %v", err)
 		return err
 	}
-	c.preloadedConfig = cfg
+	c.preloadedConfigJSON.Store(&jsonStr)
 	log.Infof("SetConfigFromJSON: config loaded successfully from JSON")
 	return nil
 }
 
+func (c *Client) loadConfig(input profilemanager.ConfigInput) (*profilemanager.Config, error) {
+	var cfg *profilemanager.Config
+	var err error
+	if preloaded := c.preloadedConfigJSON.Load(); preloaded != nil {
+		cfg, err = profilemanager.ConfigFromJSON(*preloaded)
+	} else {
+		cfg, err = profilemanager.DirectUpdateOrCreateConfig(input)
+	}
+	if err != nil {
+		return nil, err
+	}
+	c.applyMDMOverlay(cfg)
+	return cfg, nil
+}
+
 // Run start the internal client. It is a blocker function
 func (c *Client) Run(fd int32, interfaceName string, envList *EnvList) error {
 	exportEnvList(envList)
 	log.Infof("Starting NetBird client")
 	log.Debugf("Tunnel uses interface: %s", interfaceName)
 
-	var cfg *profilemanager.Config
-	var err error
-
-	// Use preloaded config if available (tvOS where file writes are blocked)
-	if c.preloadedConfig != nil {
-		log.Infof("Run: using preloaded config from memory")
-		cfg = c.preloadedConfig
-	} else {
-		log.Infof("Run: loading config from file")
-		// Use DirectUpdateOrCreateConfig to avoid atomic file operations (temp file + rename)
-		// which are blocked by the tvOS sandbox in App Group containers
-		cfg, err = profilemanager.DirectUpdateOrCreateConfig(profilemanager.ConfigInput{
-			ConfigPath:    c.cfgFile,
-			StateFilePath: c.stateFile,
-		})
-		if err != nil {
-			return err
-		}
+	cfg, err := c.loadConfig(profilemanager.ConfigInput{
+		ConfigPath:    c.cfgFile,
+		StateFilePath: c.stateFile,
+	})
+	if err != nil {
+		return err
 	}
 	c.recorder.UpdateManagementAddress(cfg.ManagementURL.String())
 	c.recorder.UpdateRosenpass(cfg.RosenpassEnabled, cfg.RosenpassPermissive)
@@ -274,19 +280,13 @@ func (c *Client) DebugBundle(anonymize bool, anonymizeLevel string) (string, err
 
 	// If the engine hasn't been started, load config so we can reach management.
 	if cfg == nil {
-		if c.preloadedConfig != nil {
-			cfg = c.preloadedConfig
-		} else {
-			var err error
-			// Use DirectUpdateOrCreateConfig to avoid atomic file operations
-			// (temp file + rename) blocked by the tvOS sandbox.
-			cfg, err = profilemanager.DirectUpdateOrCreateConfig(profilemanager.ConfigInput{
-				ConfigPath:    c.cfgFile,
-				StateFilePath: c.stateFile,
-			})
-			if err != nil {
-				return "", fmt.Errorf("load config: %w", err)
-			}
+		var err error
+		cfg, err = c.loadConfig(profilemanager.ConfigInput{
+			ConfigPath:    c.cfgFile,
+			StateFilePath: c.stateFile,
+		})
+		if err != nil {
+			return "", fmt.Errorf("load config: %w", err)
 		}
 	}
 
@@ -421,29 +421,9 @@ func (c *Client) IsLoginRequired() bool {
 	ctx, cancel := context.WithCancel(ctxWithValues)
 	defer cancel()
 
-	var cfg *profilemanager.Config
-	var err error
-
-	// Use preloaded config if available (tvOS where file writes are blocked)
-	if c.preloadedConfig != nil {
-		log.Infof("IsLoginRequired: using preloaded config from memory")
-		cfg = c.preloadedConfig
-	} else {
-		log.Infof("IsLoginRequired: loading config from file")
-		// Use DirectUpdateOrCreateConfig to avoid atomic file operations (temp file + rename)
-		// which are blocked by the tvOS sandbox in App Group containers
-		cfg, err = profilemanager.DirectUpdateOrCreateConfig(profilemanager.ConfigInput{
-			ConfigPath: c.cfgFile,
-		})
-		if err != nil {
-			log.Errorf("IsLoginRequired: failed to load config: %v", err)
-			// If we can't load config, assume login is required
-			return true
-		}
-	}
-
-	if cfg == nil {
-		log.Errorf("IsLoginRequired: config is nil")
+	cfg, err := c.loadConfig(profilemanager.ConfigInput{ConfigPath: c.cfgFile})
+	if err != nil {
+		log.Errorf("IsLoginRequired: failed to load config: %v", err)
 		return true
 	}
 
@@ -493,6 +473,7 @@ func (c *Client) LoginForMobile() string {
 		log.Errorf("LoginForMobile: failed to load config: %v", err)
 		return fmt.Sprintf("failed to load config: %v", err)
 	}
+	c.applyMDMOverlay(cfg)
 
 	oAuthFlow, err := auth.NewOAuthFlow(ctx, cfg, false, false, "")
 	if err != nil {

client/ios/NetBirdSDK/login.go

@@ -11,6 +11,7 @@ import (
 
 	"github.com/netbirdio/netbird/client/internal/auth"
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 	"github.com/netbirdio/netbird/client/mobile"
 	"github.com/netbirdio/netbird/client/system"
 )
@@ -39,14 +40,22 @@ type Auth struct {
 	ctx     context.Context
 	cancel  context.CancelFunc
 	config  *profilemanager.Config
+	base    *profilemanager.Config
+	policy  *mdm.Policy
 	cfgPath string
 }
 
-// NewAuth instantiate Auth struct and validate the management URL
-func NewAuth(cfgPath string, mgmURL string) (*Auth, error) {
-	inputCfg := profilemanager.ConfigInput{
-		ConfigPath:    cfgPath,
-		ManagementURL: mgmURL,
+// NewAuth instantiate Auth struct and validate the management URL.
+// Auth is constructed under the active MDM policy: the policy is overlaid on
+// the resolved config so the login runs against the enforced values, while
+// the persisted config keeps the caller-supplied ones; a caller-supplied
+// management URL is ignored while MDM manages that key. A nil fetcher
+// disables MDM enforcement.
+func NewAuth(cfgPath string, mgmURL string, fetcher PolicyFetcher) (*Auth, error) {
+	policy := loaderFor(fetcher).Load()
+	inputCfg := profilemanager.ConfigInput{ConfigPath: cfgPath}
+	if _, managed := policy.GetString(mdm.KeyManagementURL); !managed {
+		inputCfg.ManagementURL = mgmURL
 	}
 
 	// Load the existing config when a config file is already present so an
@@ -67,6 +76,10 @@ func NewAuth(cfgPath string, mgmURL string) (*Auth, error) {
 	if err != nil {
 		return nil, err
 	}
+	a := &Auth{policy: policy, cfgPath: cfgPath}
+	if err := a.setBaseConfig(cfg); err != nil {
+		return nil, err
+	}
 
 	// Use a cancellable context so Stop() can abort an in-progress interactive
 	// login. The PKCE flow's WaitToken blocks (and keeps its loopback HTTP server
@@ -76,14 +89,8 @@ func NewAuth(cfgPath string, mgmURL string) (*Auth, error) {
 	// process (decoupled from the network extension), so without this the server
 	// lingers after the user dismisses the browser and the next connect stalls
 	// trying to bind the same port.
-	ctx, cancel := context.WithCancel(context.Background())
-
-	return &Auth{
-		ctx:     ctx,
-		cancel:  cancel,
-		config:  cfg,
-		cfgPath: cfgPath,
-	}, nil
+	a.ctx, a.cancel = context.WithCancel(context.Background())
+	return a, nil
 }
 
 // NewAuthWithConfig instantiate Auth based on existing config
@@ -106,9 +113,7 @@ func (a *Auth) Stop() {
 	}
 }
 
-// SaveConfigIfSSOSupported test the connectivity with the management server by retrieving the server device flow info.
-// If it returns a flow info than save the configuration and return true. If it gets a codes.NotFound, it means that SSO
-// is not supported and returns false without saving the configuration. For other errors return false.
+// SaveConfigIfSSOSupported reports whether the management server supports SSO; the config is already persisted by NewAuth.
 func (a *Auth) SaveConfigIfSSOSupported(listener SSOListener) {
 	if listener == nil {
 		log.Errorf("SaveConfigIfSSOSupported: listener is nil")
@@ -136,17 +141,10 @@ func (a *Auth) saveConfigIfSSOSupported() (bool, error) {
 		return false, fmt.Errorf("failed to check SSO support: %v", err)
 	}
 
-	if !supportsSSO {
-		return false, nil
-	}
-
-	// Use DirectWriteOutConfig to avoid atomic file operations (temp file + rename)
-	// which are blocked by the tvOS sandbox in App Group containers
-	err = profilemanager.DirectWriteOutConfig(a.cfgPath, a.config)
-	return true, err
+	return supportsSSO, nil
 }
 
-// LoginWithSetupKeyAndSaveConfig test the connectivity with the management server with the setup key.
+// LoginWithSetupKeyAndSaveConfig registers the peer with the setup key; the config is already persisted by NewAuth.
 func (a *Auth) LoginWithSetupKeyAndSaveConfig(resultListener ErrListener, setupKey string, deviceName string) {
 	if resultListener == nil {
 		log.Errorf("LoginWithSetupKeyAndSaveConfig: resultListener is nil")
@@ -175,10 +173,7 @@ func (a *Auth) loginWithSetupKeyAndSaveConfig(setupKey string, deviceName string
 	if err != nil {
 		return fmt.Errorf("login failed: %v", err)
 	}
-
-	// Use DirectWriteOutConfig to avoid atomic file operations (temp file + rename)
-	// which are blocked by the tvOS sandbox in App Group containers
-	return profilemanager.DirectWriteOutConfig(a.cfgPath, a.config)
+	return nil
 }
 
 // LoginSync performs a synchronous login check without UI interaction
@@ -312,19 +307,6 @@ func (a *Auth) login(urlOpener URLOpener, forceDeviceAuth bool, deviceName strin
 		}
 	}
 
-	// Save the config before notifying success to ensure persistence completes
-	// before the callback potentially triggers teardown on the Swift side.
-	// Note: This differs from Android which doesn't save config after login.
-	// On iOS/tvOS, we save here because:
-	// 1. The config may have been modified during login (e.g., new tokens)
-	// 2. On tvOS, the Network Extension context may be the only place with
-	//    write permissions to the App Group container
-	if a.cfgPath != "" {
-		if err := profilemanager.DirectWriteOutConfig(a.cfgPath, a.config); err != nil {
-			log.Warnf("failed to save config after login: %v", err)
-		}
-	}
-
 	// Notify caller of successful login synchronously before returning
 	urlOpener.OnLoginSuccess()
 
@@ -375,23 +357,44 @@ func (a *Auth) foregroundGetTokenInfo(authClient *auth.Auth, urlOpener URLOpener
 	return &tokenInfo, nil
 }
 
-// GetConfigJSON returns the current config as a JSON string.
-// This can be used by the caller to persist the config via alternative storage
-// mechanisms (e.g., UserDefaults on tvOS where file writes are blocked).
+// GetConfigJSON returns the config without the MDM overlay as JSON, for persisting it outside the config file (tvOS).
 func (a *Auth) GetConfigJSON() (string, error) {
-	if a.config == nil {
+	cfg := a.base
+	if cfg == nil {
+		cfg = a.config
+	}
+	if cfg == nil {
 		return "", fmt.Errorf("no config available")
 	}
-	return profilemanager.ConfigToJSON(a.config)
+	return profilemanager.ConfigToJSON(cfg)
 }
 
-// SetConfigFromJSON loads config from a JSON string.
-// This can be used to restore config from alternative storage mechanisms.
+// SetConfigFromJSON replaces the config from JSON; the MDM overlay is applied on top for the login.
 func (a *Auth) SetConfigFromJSON(jsonStr string) error {
 	cfg, err := profilemanager.ConfigFromJSON(jsonStr)
 	if err != nil {
 		return err
 	}
-	a.config = cfg
+	return a.setBaseConfig(cfg)
+}
+
+func (a *Auth) setBaseConfig(base *profilemanager.Config) error {
+	overlaid, err := copyConfig(base)
+	if err != nil {
+		return err
+	}
+	if a.policy != nil {
+		overlaid.ApplyMDMPolicy(a.policy)
+	}
+	a.base = base
+	a.config = overlaid
 	return nil
 }
+
+func copyConfig(cfg *profilemanager.Config) (*profilemanager.Config, error) {
+	raw, err := profilemanager.ConfigToJSON(cfg)
+	if err != nil {
+		return nil, err
+	}
+	return profilemanager.ConfigFromJSON(raw)
+}

client/ios/NetBirdSDK/mdm.go

@@ -0,0 +1,66 @@
+//go:build ios
+
+package NetBirdSDK
+
+import (
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
+)
+
+// PolicyFetcher is implemented by the native layer to return the current
+// managed configuration as a JSON-encoded object string; "" means no MDM
+// source is present.
+type PolicyFetcher interface {
+	FetchJSON() string
+}
+
+type mdmSource struct {
+	loader   *mdm.Loader
+	detector *mdm.ChangeDetector
+}
+
+// SetMDMPolicyFetcher registers the native-provided MDM policy fetcher on
+// this Client; passing nil disables MDM enforcement.
+func (c *Client) SetMDMPolicyFetcher(p PolicyFetcher) {
+	loader := loaderFor(p)
+	c.mdmSource.Store(&mdmSource{loader: loader, detector: mdm.NewChangeDetector(loader)})
+}
+
+// HasMDMPolicyChanged re-reads the managed configuration and reports whether
+// it changed since the last observation; call it from the native OS-change
+// notification and restart the engine only on true.
+func (c *Client) HasMDMPolicyChanged() bool {
+	src := c.mdmSource.Load()
+	if src == nil {
+		return false
+	}
+	return src.detector.Changed()
+}
+
+// GetRestrictionsJSON returns the UI enforcement snapshot derived from the
+// active MDM policy, in the JSON shape shared with the desktop frontend.
+func (c *Client) GetRestrictionsJSON() (string, error) {
+	return mdm.BuildRestrictions(c.mdmLoader().Load()).JSON()
+}
+
+func (c *Client) applyMDMOverlay(cfg *profilemanager.Config) {
+	loader := c.mdmLoader()
+	if cfg == nil || loader == nil {
+		return
+	}
+	cfg.ApplyMDMPolicy(loader.Load())
+}
+
+func (c *Client) mdmLoader() *mdm.Loader {
+	if src := c.mdmSource.Load(); src != nil {
+		return src.loader
+	}
+	return nil
+}
+
+func loaderFor(p PolicyFetcher) *mdm.Loader {
+	if p == nil {
+		return mdm.NewJSONLoader(nil)
+	}
+	return mdm.NewJSONLoader(p.FetchJSON)
+}

client/ios/NetBirdSDK/preferences.go

@@ -3,12 +3,16 @@
 package NetBirdSDK
 
 import (
+	"sync/atomic"
+
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 )
 
 // Preferences export a subset of the internal config for gomobile
 type Preferences struct {
 	configInput profilemanager.ConfigInput
+	mdmLoader   atomic.Pointer[mdm.Loader]
 }
 
 // NewPreferences create new Preferences instance
@@ -17,11 +21,30 @@ func NewPreferences(configPath string, stateFilePath string) *Preferences {
 		ConfigPath:    configPath,
 		StateFilePath: stateFilePath,
 	}
-	return &Preferences{ci}
+	return &Preferences{configInput: ci}
+}
+
+// SetMDMPolicyFetcher registers the native-provided MDM policy fetcher on
+// this Preferences instance; passing nil disables MDM enforcement.
+func (p *Preferences) SetMDMPolicyFetcher(f PolicyFetcher) {
+	p.mdmLoader.Store(loaderFor(f))
+}
+
+// GetRestrictionsJSON returns the UI enforcement snapshot derived from the
+// active MDM policy, in the JSON shape shared with the desktop frontend.
+func (p *Preferences) GetRestrictionsJSON() (string, error) {
+	return mdm.BuildRestrictions(p.policy()).JSON()
+}
+
+func (p *Preferences) policy() *mdm.Policy {
+	return p.mdmLoader.Load().Load()
 }
 
 // GetManagementURL read url from config file
 func (p *Preferences) GetManagementURL() (string, error) {
+	if v, ok := p.policy().GetString(mdm.KeyManagementURL); ok {
+		return mdm.CanonicalURL(v), nil
+	}
 	if p.configInput.ManagementURL != "" {
 		return p.configInput.ManagementURL, nil
 	}
@@ -30,7 +53,7 @@ func (p *Preferences) GetManagementURL() (string, error) {
 	if err != nil {
 		return "", err
 	}
-	return cfg.ManagementURL.String(), err
+	return cfg.ManagementURL.String(), nil
 }
 
 // SetManagementURL store the given url and wait for commit
@@ -56,17 +79,21 @@ func (p *Preferences) SetAdminURL(url string) {
 	p.configInput.AdminURL = url
 }
 
-// GetPreSharedKey read preshared key from config file
-func (p *Preferences) GetPreSharedKey() (string, error) {
+// HasPreSharedKey reports whether a pre-shared key is staged, persisted, or
+// enforced by MDM; the key itself is never handed to the native layer.
+func (p *Preferences) HasPreSharedKey() (bool, error) {
+	if _, ok := p.policy().GetString(mdm.KeyPreSharedKey); ok {
+		return true, nil
+	}
 	if p.configInput.PreSharedKey != nil {
-		return *p.configInput.PreSharedKey, nil
+		return *p.configInput.PreSharedKey != "", nil
 	}
 
 	cfg, err := profilemanager.ReadConfig(p.configInput.ConfigPath)
 	if err != nil {
-		return "", err
+		return false, err
 	}
-	return cfg.PreSharedKey, err
+	return cfg.PreSharedKey != "", nil
 }
 
 // SetPreSharedKey store the given key and wait for commit
@@ -81,6 +108,9 @@ func (p *Preferences) SetRosenpassEnabled(enabled bool) {
 
 // GetRosenpassEnabled read rosenpass enabled from config file
 func (p *Preferences) GetRosenpassEnabled() (bool, error) {
+	if v, ok := p.policy().GetBool(mdm.KeyRosenpassEnabled); ok {
+		return v, nil
+	}
 	if p.configInput.RosenpassEnabled != nil {
 		return *p.configInput.RosenpassEnabled, nil
 	}
@@ -99,6 +129,9 @@ func (p *Preferences) SetRosenpassPermissive(permissive bool) {
 
 // GetRosenpassPermissive read rosenpass permissive from config file
 func (p *Preferences) GetRosenpassPermissive() (bool, error) {
+	if v, ok := p.policy().GetBool(mdm.KeyRosenpassPermissive); ok {
+		return v, nil
+	}
 	if p.configInput.RosenpassPermissive != nil {
 		return *p.configInput.RosenpassPermissive, nil
 	}
@@ -130,7 +163,8 @@ func (p *Preferences) SetDisableIPv6(disable bool) {
 
 // GetRemoteJobsAllowed reads the remote jobs opt-in from config file
 func (p *Preferences) GetRemoteJobsAllowed() (bool, error) {
-	if p.configInput.RemoteJobsAllowed != nil {
+	policy := p.policy()
+	if !policy.HasKey(mdm.KeyRemoteJobsAllowed) && p.configInput.RemoteJobsAllowed != nil {
 		return *p.configInput.RemoteJobsAllowed, nil
 	}
 
@@ -138,10 +172,11 @@ func (p *Preferences) GetRemoteJobsAllowed() (bool, error) {
 	if err != nil {
 		return false, err
 	}
+	cfg.ApplyMDMPolicy(policy)
 	if cfg.RemoteJobsAllowed == nil {
 		return false, nil
 	}
-	return *cfg.RemoteJobsAllowed, err
+	return *cfg.RemoteJobsAllowed, nil
 }
 
 // SetRemoteJobsAllowed stores the given value and waits for commit
@@ -151,6 +186,9 @@ func (p *Preferences) SetRemoteJobsAllowed(allowed bool) {
 
 // Commit write out the changes into config file
 func (p *Preferences) Commit() error {
+	if err := profilemanager.CheckMDMConflicts(p.configInput, p.policy()); err != nil {
+		return err
+	}
 	// Use DirectUpdateOrCreateConfig to avoid atomic file operations (temp file + rename)
 	// which are blocked by the tvOS sandbox in App Group containers
 	_, err := profilemanager.DirectUpdateOrCreateConfig(p.configInput)

client/ios/NetBirdSDK/preferences_test.go

@@ -31,14 +31,13 @@ func TestPreferences_DefaultValues(t *testing.T) {
 		t.Errorf("invalid default management url: %s", defaultVar)
 	}
 
-	var preSharedKey string
-	preSharedKey, err = p.GetPreSharedKey()
+	hasPSK, err := p.HasPreSharedKey()
 	if err != nil {
-		t.Fatalf("failed to read default preshared key: %s", err)
+		t.Fatalf("failed to read default preshared key presence: %s", err)
 	}
 
-	if preSharedKey != "" {
-		t.Errorf("invalid preshared key: %s", preSharedKey)
+	if hasPSK {
+		t.Errorf("unexpected preshared key presence on fresh config")
 	}
 }
 
@@ -69,13 +68,13 @@ func TestPreferences_ReadUncommitedValues(t *testing.T) {
 	}
 
 	p.SetPreSharedKey(exampleString)
-	resp, err = p.GetPreSharedKey()
+	hasPSK, err := p.HasPreSharedKey()
 	if err != nil {
-		t.Fatalf("failed to read preshared key: %s", err)
+		t.Fatalf("failed to read preshared key presence: %s", err)
 	}
 
-	if resp != exampleString {
-		t.Errorf("unexpected preshared key: %s", resp)
+	if !hasPSK {
+		t.Errorf("expected preshared key presence after staging one")
 	}
 }
 
@@ -114,12 +113,12 @@ func TestPreferences_Commit(t *testing.T) {
 		t.Errorf("unexpected management url: %s", resp)
 	}
 
-	resp, err = p.GetPreSharedKey()
+	hasPSK, err := p.HasPreSharedKey()
 	if err != nil {
-		t.Fatalf("failed to read preshared key: %s", err)
+		t.Fatalf("failed to read preshared key presence: %s", err)
 	}
 
-	if resp != examplePresharedKey {
-		t.Errorf("unexpected preshared key: %s", resp)
+	if !hasPSK {
+		t.Errorf("expected preshared key presence after commit")
 	}
 }

client/ios/NetBirdSDK/profile_manager.go

@@ -52,6 +52,12 @@ func NewProfileManager(configDir string) *ProfileManager {
 	return &ProfileManager{impl: mobile.NewProfileManager(configDir, iosUsername)}
 }
 
+// SetMDMPolicyFetcher registers the native-provided MDM policy fetcher on
+// this ProfileManager; passing nil disables MDM enforcement.
+func (pm *ProfileManager) SetMDMPolicyFetcher(f PolicyFetcher) {
+	pm.impl.SetMDMLoader(loaderFor(f))
+}
+
 // ListProfiles returns all available profiles, including the default profile,
 // with their active status set.
 func (pm *ProfileManager) ListProfiles() (*ProfileArray, error) {

client/mdm/changedetector.go

@@ -0,0 +1,34 @@
+package mdm
+
+import "sync"
+
+// ChangeDetector tracks the last observed policy of a Loader so an
+// OS-notification-driven caller can ask whether the managed configuration
+// actually changed before restarting anything.
+type ChangeDetector struct {
+	mu     sync.Mutex
+	loader *Loader
+	prev   *Policy
+}
+
+// NewChangeDetector constructs a ChangeDetector seeded with the loader's
+// current policy, so only a later change reports as changed.
+func NewChangeDetector(loader *Loader) *ChangeDetector {
+	return &ChangeDetector{
+		loader: loader,
+		prev:   loader.Load(),
+	}
+}
+
+// Changed re-reads the policy, logs the per-key diff, and reports whether it
+// diverged from the last observation; the new snapshot becomes the baseline.
+func (d *ChangeDetector) Changed() bool {
+	d.mu.Lock()
+	defer d.mu.Unlock()
+	curr := d.loader.Load()
+	if !policyChanged(d.prev, curr) {
+		return false
+	}
+	d.prev = curr
+	return true
+}

client/mdm/conflicts.go

@@ -0,0 +1,116 @@
+package mdm
+
+import (
+	"net/url"
+
+	"github.com/netbirdio/netbird/util"
+)
+
+// PreSharedKeyRedactedSentinel is the redaction mask returned in place of a
+// real pre-shared key; an incoming value equal to it is a round-trip echo,
+// never an override.
+const PreSharedKeyRedactedSentinel = "**********"
+
+// ConflictCheck is a value-aware comparison between a single requested field
+// and the corresponding MDM-enforced value.
+type ConflictCheck struct {
+	Key   string
+	Check func(*Policy) bool
+}
+
+// ConflictBool builds a ConflictCheck for a boolean MDM key.
+func ConflictBool(key string, p *bool) ConflictCheck {
+	return ConflictCheck{
+		Key: key,
+		Check: func(pol *Policy) bool {
+			if p == nil {
+				return true
+			}
+			want, ok := pol.GetBool(key)
+			return ok && want == *p
+		},
+	}
+}
+
+// ConflictStringPtr builds a ConflictCheck for an optional string MDM key,
+// where an explicit empty value is still a request to change the setting. A
+// nil p means "field not set" (no override requested).
+func ConflictStringPtr(key string, p *string) ConflictCheck {
+	return ConflictCheck{
+		Key: key,
+		Check: func(pol *Policy) bool {
+			if p == nil {
+				return true
+			}
+			want, ok := pol.GetString(key)
+			return ok && want == *p
+		},
+	}
+}
+
+// ConflictURL builds a ConflictCheck for a URL-typed MDM key. The two sides are
+// compared as the endpoints they address, not as strings: see
+// util.SameServiceURL.
+func ConflictURL(key, got string) ConflictCheck {
+	return ConflictCheck{
+		Key: key,
+		Check: func(pol *Policy) bool {
+			if got == "" {
+				return true
+			}
+			want, ok := pol.GetString(key)
+			return ok && util.SameServiceURLStrings(want, got)
+		},
+	}
+}
+
+// ConflictInt64 builds a ConflictCheck for an integer MDM key.
+func ConflictInt64(key string, p *int64) ConflictCheck {
+	return ConflictCheck{
+		Key: key,
+		Check: func(pol *Policy) bool {
+			if p == nil {
+				return true
+			}
+			want, ok := pol.GetInt(key)
+			return ok && want == *p
+		},
+	}
+}
+
+// ResolveConflicts returns the names of keys whose requested value diverges
+// from the policy-enforced value; keys the policy does not manage are skipped,
+// a managed key without a Check counts as a conflict.
+func ResolveConflicts(policy *Policy, checks []ConflictCheck) []string {
+	if policy.IsEmpty() {
+		return nil
+	}
+	var conflicts []string
+	for _, c := range checks {
+		if !policy.HasKey(c.Key) {
+			continue
+		}
+		if c.Check == nil || !c.Check(policy) {
+			conflicts = append(conflicts, c.Key)
+		}
+	}
+	return conflicts
+}
+
+// CanonicalURL normalizes a service URL by appending the scheme default port
+// when none is present; unparseable input is returned unchanged.
+func CanonicalURL(s string) string {
+	u, err := url.ParseRequestURI(s)
+	if err != nil {
+		return s
+	}
+	if u.Port() == "" {
+		switch u.Scheme {
+		case "https":
+			u.Host += ":443"
+		case "http":
+			u.Host += ":80"
+		}
+	}
+	return u.String()
+}

client/mdm/conflicts_test.go

@@ -0,0 +1,40 @@
+package mdm
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// The same spellings, through the conflict check that decides whether a request
+// is refused. An enforced URL restated in another spelling addresses the very
+// server the policy names, so it must not be reported as a conflict.
+func TestConflictURLComparesEndpoints(t *testing.T) {
+	policy := NewPolicy(map[string]any{KeyManagementURL: "https://mgmt.example.com"})
+	require.True(t, policy.HasKey(KeyManagementURL))
+
+	for _, restated := range []string{
+		"https://mgmt.example.com",
+		"https://mgmt.example.com:443",
+		"https://mgmt.example.com/",
+		"https://MGMT.example.com",
+		"https://mgmt.example.com:0443",
+	} {
+		conflicts := ResolveConflicts(policy, []ConflictCheck{ConflictURL(KeyManagementURL, restated)})
+		assert.Empty(t, conflicts, "%q is the enforced endpoint written differently", restated)
+	}
+
+	for _, diverging := range []string{
+		"https://other.example.com",
+		"http://mgmt.example.com",
+		"https://mgmt.example.com:8443",
+		"https://mgmt.example.com/other",
+	} {
+		conflicts := ResolveConflicts(policy, []ConflictCheck{ConflictURL(KeyManagementURL, diverging)})
+		assert.Equal(t, []string{KeyManagementURL}, conflicts, "%q addresses another endpoint", diverging)
+	}
+
+	// An unset field is not a request to change anything.
+	assert.Empty(t, ResolveConflicts(policy, []ConflictCheck{ConflictURL(KeyManagementURL, "")}))
+}

client/mdm/jsonloader.go

@@ -0,0 +1,34 @@
+package mdm
+
+import (
+	"encoding/json"
+
+	log "github.com/sirupsen/logrus"
+)
+
+type jsonPolicyFetcher struct {
+	fetch func() string
+}
+
+// NewJSONLoader constructs a Loader whose policy source is a JSON-encoded
+// object string, as produced by the mobile native layers; a nil fetch
+// disables MDM enforcement.
+func NewJSONLoader(fetch func() string) *Loader {
+	if fetch == nil {
+		return NewLoader(nil)
+	}
+	return NewLoader(&jsonPolicyFetcher{fetch: fetch})
+}
+
+func (f *jsonPolicyFetcher) Fetch() map[string]any {
+	raw := f.fetch()
+	if raw == "" {
+		return nil
+	}
+	var out map[string]any
+	if err := json.Unmarshal([]byte(raw), &out); err != nil {
+		log.Warnf("MDM mobile fetcher: invalid JSON payload from native: %v", err)
+		return nil
+	}
+	return out
+}

client/mdm/policy.go

@@ -119,16 +119,46 @@ func NewPolicy(values map[string]any) *Policy {
 	return &Policy{values: values}
 }
 
-// LoadPolicy reads the platform-native MDM configuration. Returns an
-// empty (but non-nil) Policy when no source is present, the source is
-// empty, or the platform is unsupported.
+// PolicyFetcher supplies the managed configuration to a Loader. Mobile
+// platforms (Android / iOS) implement it to push the OS-managed values
+// into the Go runtime. On every platform a non-nil fetcher takes
+// precedence over the native source, which is the test seam for the
+// registry / plist loaders; a nil fetcher leaves the native source in
+// charge, or disables MDM enforcement where there is none.
+type PolicyFetcher interface {
+	Fetch() map[string]any
+}
+
+// Loader is the DI-friendly entry point for reading the active MDM
+// policy. Construct one at the daemon's lifecycle owner (Server on
+// desktop, gomobile-exposed bridge on mobile) and pass it to anything
+// that needs to read MDM state (the reload ticker, profilemanager's
+// Config). Each callsite has the Loader handed in instead of looking
+// up package-level state.
+type Loader struct {
+	fetcher PolicyFetcher
+}
+
+// NewLoader constructs a Loader. A non-nil fetcher takes precedence over
+// the platform-native source; production desktop callers pass nil so the
+// registry / plist stays authoritative.
+func NewLoader(f PolicyFetcher) *Loader {
+	return &Loader{fetcher: f}
+}
+
+// Load reads the platform-native MDM configuration and returns a
+// Policy. Returns an empty (but non-nil) Policy when no source is
+// present, the source is empty, or the platform is unsupported.
 //
 // Diagnostic logging differentiates the three states:
 //   - source absent / unsupported platform: trace log only
 //   - source present, zero keys:             info "MDM enrolled (no managed keys)"
 //   - source present, N keys:                info "MDM enrolled with N managed keys: [...]"
-func LoadPolicy() *Policy {
-	values, err := loadPlatformPolicy()
+func (l *Loader) Load() *Policy {
+	if l == nil {
+		return &Policy{values: map[string]any{}}
+	}
+	values, err := l.loadPlatform()
 	if err != nil {
 		log.Tracef("MDM policy load: %v", err)
 		return &Policy{values: map[string]any{}}
@@ -205,6 +235,8 @@ func (p *Policy) GetBool(key string) (bool, bool) {
 		return t != 0, true
 	case int64:
 		return t != 0, true
+	case float64:
+		return t != 0, true
 	}
 	return false, false
 }
@@ -270,7 +302,7 @@ func (p *Policy) GetStringSlice(key string) ([]string, bool) {
 }
 
 // sortedKeys returns the keys of m as a deterministic, lexicographically
-// sorted slice. Used internally by Policy.ManagedKeys and LoadPolicy's
+// sorted slice. Used internally by Policy.ManagedKeys and Loader.Load's
 // diagnostic log line so callers see a stable key order across runs
 // regardless of Go's randomised map iteration.
 func sortedKeys(m map[string]any) []string {

client/mdm/policy_darwin.go

@@ -25,8 +25,9 @@ import (
 // writable plist, as a defense against tampered installs.
 const policyPlistPath = "/Library/Managed Preferences/io.netbird.client.plist"
 
-// loadPlatformPolicy reads the MDM-managed configuration from the macOS
-// managed-preferences plist at policyPlistPath. Returns:
+// loadPlatform reads the MDM-managed configuration from the macOS
+// managed-preferences plist at policyPlistPath, unless a fetcher was
+// injected, in which case its values are returned instead. Returns:
 //   - (nil, nil)  when the plist is absent (device not MDM-enrolled for
 //     NetBird, or admin has not yet pushed a payload)
 //   - (map, nil)  with N entries when N managed values are present
@@ -39,13 +40,19 @@ const policyPlistPath = "/Library/Managed Preferences/io.netbird.client.plist"
 // skipped so a stray entry in the payload does not block startup.
 // Native plist value types map naturally onto the Policy accessor
 // expectations (GetString / GetBool / GetInt / GetStringSlice).
-func loadPlatformPolicy() (map[string]any, error) {
+func (l *Loader) loadPlatform() (map[string]any, error) {
+	// Honour the injected fetcher when present so tests (and any
+	// future non-macOS MDM channel) can short-circuit the plist read
+	// with a scripted policy.
+	if l != nil && l.fetcher != nil {
+		return l.fetcher.Fetch(), nil
+	}
 	f, err := os.Open(policyPlistPath)
 	if err != nil {
 		if errors.Is(err, fs.ErrNotExist) {
 			// Not enrolled for NetBird. Caller treats nil as
 			// "no MDM source present".
-			//nolint:nilnil // (nil, nil) is the documented platform-absent sentinel; see LoadPolicy.
+			//nolint:nilnil // (nil, nil) is the documented platform-absent sentinel; see Loader.Load.
 			return nil, nil
 		}
 		return nil, fmt.Errorf("open %s: %w", policyPlistPath, err)

client/mdm/policy_mobile.go

@@ -2,13 +2,14 @@
 
 package mdm
 
-// loadPlatformPolicy is unused on mobile: the native layer (Swift on iOS,
-// Kotlin/Java on Android) reads the OS managed-config store and pushes the
-// resulting dictionary in-process via a gomobile entry point that lands in
-// Phase 5 / Phase 6. The stub keeps the package compilable for mobile
-// builds and returns (nil, nil) — the platform-absent sentinel that
-// LoadPolicy in policy.go treats as "no MDM source present".
-func loadPlatformPolicy() (map[string]any, error) {
-	//nolint:nilnil // (nil, nil) is the documented platform-absent sentinel; see LoadPolicy.
-	return nil, nil
+// loadPlatform reads the OS-managed configuration via the native
+// PolicyFetcher injected at Loader construction. Returns
+// (nil, nil) — the platform-absent sentinel that Loader.Load treats as
+// "no MDM source present" — when no fetcher was provided.
+func (l *Loader) loadPlatform() (map[string]any, error) {
+	if l == nil || l.fetcher == nil {
+		//nolint:nilnil // (nil, nil) is the documented platform-absent sentinel; see Loader.Load.
+		return nil, nil
+	}
+	return l.fetcher.Fetch(), nil
 }

client/mdm/policy_other.go

@@ -2,13 +2,17 @@
 
 package mdm
 
-// loadPlatformPolicy returns no policy on platforms without an MDM channel
-// (Linux, FreeBSD). MDM enforcement is off and the client behaves as if
-// the feature did not exist. Returns (nil, nil) — the platform-absent
-// sentinel the caller (LoadPolicy in policy.go) treats as "no MDM
-// source present"; an error here would just translate to the same
-// outcome with an extra log line.
-func loadPlatformPolicy() (map[string]any, error) {
-	//nolint:nilnil // (nil, nil) is the documented platform-absent sentinel; see LoadPolicy.
+// loadPlatform reads the MDM policy on platforms without a native MDM
+// channel (Linux, FreeBSD). When no fetcher was injected the policy is
+// (nil, nil) — the platform-absent sentinel that Loader.Load treats as
+// "MDM enforcement disabled". A non-nil fetcher takes precedence: it
+// is the test-seam used by unit tests to inject a scripted policy
+// without touching the OS, and the same hook supports any future
+// non-mobile OS that grows an out-of-band MDM channel.
+func (l *Loader) loadPlatform() (map[string]any, error) {
+	if l != nil && l.fetcher != nil {
+		return l.fetcher.Fetch(), nil
+	}
+	//nolint:nilnil // (nil, nil) is the documented platform-absent sentinel; see Loader.Load.
 	return nil, nil
 }

client/mdm/policy_test.go

@@ -1,6 +1,7 @@
 package mdm
 
 import (
+	"runtime"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -95,7 +96,8 @@ func TestPolicy_GetBool(t *testing.T) {
 		{"int64 nonzero", int64(2), true, true},
 		{"int64 zero", int64(0), false, true},
 		{"string garbage", "maybe", false, false},
-		{"float unsupported", 1.0, false, false},
+		{"float nonzero", 1.0, true, true},
+		{"float zero", 0.0, false, true},
 	}
 	for _, c := range cases {
 		t.Run(c.name, func(t *testing.T) {
@@ -155,10 +157,29 @@ func TestPolicy_GetStringSlice(t *testing.T) {
 	})
 }
 
-func TestLoadPolicy_PlatformStubReturnsEmpty(t *testing.T) {
-	// loadPlatformPolicy is a stub on every OS for Phase 1. LoadPolicy must
-	// degrade gracefully and never return nil.
-	p := LoadPolicy()
+// encoding/json decodes every JSON number into float64, so the mobile
+// loaders never see int.
+func TestJSONLoader_BoolFromNumber(t *testing.T) {
+	p := NewJSONLoader(func() string { return `{"blockInbound":1,"disableProfiles":0}` }).Load()
+
+	got, ok := p.GetBool(KeyBlockInbound)
+	assert.True(t, ok)
+	assert.True(t, got)
+
+	got, ok = p.GetBool(KeyDisableProfiles)
+	assert.True(t, ok)
+	assert.False(t, got)
+}
+
+func TestLoader_NilFetcherReturnsEmpty(t *testing.T) {
+	// Loader.Load with no fetcher (desktop construction) must degrade
+	// gracefully and never return nil; on linux loadPlatform is a stub
+	// returning (nil, nil), and Load is expected to translate that
+	// into a non-nil empty Policy.
+	if runtime.GOOS == "windows" || runtime.GOOS == "darwin" {
+		t.Skip("a nil fetcher reads the OS-managed policy on this platform")
+	}
+	p := NewLoader(nil).Load()
 	require.NotNil(t, p)
 	assert.True(t, p.IsEmpty())
 	assert.Empty(t, p.ManagedKeys())

client/mdm/policy_windows.go

@@ -61,8 +61,9 @@ func readRegistryValue(k registry.Key, name, canonical string, out map[string]an
 	}
 }
 
-// loadPlatformPolicy reads the MDM-managed configuration from the
-// Windows registry under HKLM\Software\Policies\NetBird. Returns:
+// loadPlatform reads the MDM-managed configuration from the Windows
+// registry under HKLM\Software\Policies\NetBird, unless a fetcher was
+// injected, in which case its values are returned instead. Returns:
 //   - (nil, nil)  when the key is absent (device not MDM-enrolled for NetBird)
 //   - (map, nil)  with N entries when N managed values are set (N may be 0)
 //   - (nil, err)  on open / enumerate registry errors
@@ -70,12 +71,18 @@ func readRegistryValue(k registry.Key, name, canonical string, out map[string]an
 // Per-value type coercion + skip-on-error is delegated to
 // readRegistryValue. Unknown value names are logged and skipped so a
 // malformed deployment does not block startup.
-func loadPlatformPolicy() (map[string]any, error) {
+func (l *Loader) loadPlatform() (map[string]any, error) {
+	// Honour the injected fetcher when present so tests (and any
+	// future non-Windows MDM channel) can short-circuit the registry
+	// read with a scripted policy.
+	if l != nil && l.fetcher != nil {
+		return l.fetcher.Fetch(), nil
+	}
 	k, err := registry.OpenKey(registry.LOCAL_MACHINE, policyRegistryPath, registry.QUERY_VALUE)
 	if err != nil {
 		if errors.Is(err, registry.ErrNotExist) {
 			// Not enrolled. Caller treats nil as "no MDM source present".
-			//nolint:nilnil // (nil, nil) is the documented platform-absent sentinel; see LoadPolicy.
+			//nolint:nilnil // (nil, nil) is the documented platform-absent sentinel; see Loader.Load.
 			return nil, nil
 		}
 		return nil, fmt.Errorf("open %s: %w", policyRegistryPath, err)

client/mdm/restrictions.go

@@ -0,0 +1,91 @@
+package mdm
+
+import "encoding/json"
+
+// Fields carries the per-key MDM enforcement state for a UI: value-typed
+// fields hold the enforced value (nil pointer = not managed), boolean
+// fields report that the key is managed.
+type Fields struct {
+	ManagementURL            string `json:"managementURL"`
+	PreSharedKey             bool   `json:"preSharedKey"`
+	WireguardPort            bool   `json:"wireguardPort"`
+	RosenpassEnabled         bool   `json:"rosenpassEnabled"`
+	RosenpassPermissive      bool   `json:"rosenpassPermissive"`
+	DisableClientRoutes      bool   `json:"disableClientRoutes"`
+	DisableServerRoutes      bool   `json:"disableServerRoutes"`
+	AllowServerSSH           *bool  `json:"allowServerSSH"`
+	DisableAutoConnect       bool   `json:"disableAutoConnect"`
+	DisableAutostart         bool   `json:"disableAutostart"`
+	BlockInbound             bool   `json:"blockInbound"`
+	DisableMetricsCollection bool   `json:"disableMetricsCollection"`
+	SplitTunnelMode          bool   `json:"splitTunnelMode"`
+	SplitTunnelApps          bool   `json:"splitTunnelApps"`
+	RemoteJobsAllowed        bool   `json:"allowRemoteJobs"`
+	DisableAdvancedView      *bool  `json:"disableAdvancedView"`
+}
+
+// Features carries the feature gates a UI must honor.
+type Features struct {
+	DisableProfiles       bool `json:"disableProfiles"`
+	DisableNetworks       bool `json:"disableNetworks"`
+	DisableUpdateSettings bool `json:"disableUpdateSettings"`
+}
+
+// Restrictions is the UI-facing enforcement snapshot; the JSON shape is
+// shared by the desktop frontend and the mobile bridges.
+type Restrictions struct {
+	MDM      Fields   `json:"mdm"`
+	Features Features `json:"features"`
+}
+
+// BuildRestrictions derives the UI enforcement snapshot from the active
+// policy.
+func BuildRestrictions(policy *Policy) Restrictions {
+	var r Restrictions
+	if policy.IsEmpty() {
+		return r
+	}
+
+	if v, ok := policy.GetString(KeyManagementURL); ok {
+		r.MDM.ManagementURL = CanonicalURL(v)
+	}
+	r.MDM.PreSharedKey = policy.HasKey(KeyPreSharedKey)
+	r.MDM.WireguardPort = policy.HasKey(KeyWireguardPort)
+	r.MDM.RosenpassEnabled = policy.HasKey(KeyRosenpassEnabled)
+	r.MDM.RosenpassPermissive = policy.HasKey(KeyRosenpassPermissive)
+	r.MDM.DisableClientRoutes = policy.HasKey(KeyDisableClientRoutes)
+	r.MDM.DisableServerRoutes = policy.HasKey(KeyDisableServerRoutes)
+	r.MDM.DisableAutoConnect = policy.HasKey(KeyDisableAutoConnect)
+	r.MDM.DisableAutostart = policy.HasKey(KeyDisableAutostart)
+	r.MDM.BlockInbound = policy.HasKey(KeyBlockInbound)
+	r.MDM.DisableMetricsCollection = policy.HasKey(KeyDisableMetricsCollection)
+	r.MDM.SplitTunnelMode = policy.HasKey(KeySplitTunnelMode)
+	r.MDM.SplitTunnelApps = policy.HasKey(KeySplitTunnelApps)
+	r.MDM.RemoteJobsAllowed = policy.HasKey(KeyRemoteJobsAllowed)
+	if v, ok := policy.GetBool(KeyAllowServerSSH); ok {
+		r.MDM.AllowServerSSH = &v
+	}
+	if v, ok := policy.GetBool(KeyDisableAdvancedView); ok {
+		r.MDM.DisableAdvancedView = &v
+	}
+
+	if v, ok := policy.GetBool(KeyDisableProfiles); ok {
+		r.Features.DisableProfiles = v
+	}
+	if v, ok := policy.GetBool(KeyDisableNetworks); ok {
+		r.Features.DisableNetworks = v
+	}
+	if v, ok := policy.GetBool(KeyDisableUpdateSettings); ok {
+		r.Features.DisableUpdateSettings = v
+	}
+	return r
+}
+
+// JSON renders the snapshot in the shared UI JSON shape.
+func (r Restrictions) JSON() (string, error) {
+	b, err := json.Marshal(r)
+	if err != nil {
+		return "", err
+	}
+	return string(b), nil
+}

client/mdm/ticker.go

@@ -15,33 +15,33 @@ import (
 // instead, hence anticipating the ticker mechanism entirely.
 const DefaultReloadInterval = 1 * time.Minute
 
-// policyLoader is the indirection through which the ticker reads the
-// OS-native policy, both for the initial observation and on every tick.
-// Production points it at LoadPolicy; tests in this package override it to
-// feed a scripted sequence of policies without touching the real OS store.
-var policyLoader = LoadPolicy
-
-// Ticker periodically re-reads the OS-native MDM policy via LoadPolicy and
-// invokes the onChange callback (supplied to Run) whenever the observed
-// Policy diverges from the last observation (added / removed / changed
-// keys). Launch with Run from a goroutine; cancel the supplied context
-// to stop.
+// Ticker periodically re-reads the OS-native MDM policy via the
+// injected Loader and invokes the onChange callback (supplied to Run)
+// whenever the observed Policy diverges from the last observation
+// (added / removed / changed keys). Launch with Run from a goroutine;
+// cancel the supplied context to stop.
 type Ticker struct {
 	interval time.Duration
+	loader   *Loader
 	prev     *Policy
 }
 
 // NewTicker constructs a Ticker that will re-read the OS-native policy
-// every reloadInterval once Run is called.
-// The initial snapshot is populated by calling policyLoader at
+// every reloadInterval once Run is called. The Loader is injected so
+// the ticker doesn't depend on any package-level state — production
+// passes the daemon-owned Loader, tests pass a fake Loader (built with
+// a fake PolicyFetcher).
+//
+// The initial snapshot is populated by calling loader.Load() at
 // construction time so the first tick only fires
 // onChange when the policy actually changed since boot — without
 // this baseline the first tick would report every currently-managed
 // key as "added" and trigger a spurious engine restart.
-func NewTicker(reloadInterval time.Duration) *Ticker {
+func NewTicker(reloadInterval time.Duration, loader *Loader) *Ticker {
 	return &Ticker{
 		interval: reloadInterval,
-		prev:     policyLoader(),
+		loader:   loader,
+		prev:     loader.Load(),
 	}
 }
 
@@ -58,13 +58,10 @@ func (t *Ticker) Run(ctx context.Context, onChange func(prev, curr *Policy) erro
 			log.Info("MDM policy reload ticker stopped")
 			return
 		case <-tk.C:
-			curr := policyLoader()
-			if policiesEqual(t.prev, curr) {
+			curr := t.loader.Load()
+			if !policyChanged(t.prev, curr) {
 				continue
 			}
-			added, removed, changed := diffPolicies(t.prev, curr)
-			log.Infof("MDM policy changed: added=%v removed=%v changed=%v",
-				added, removed, changed)
 			prev := t.prev
 			if err := onChange(prev, curr); err != nil {
 				log.Errorf("MDM policy change handler failed (retrying in 1 minute): %v", err)
@@ -127,3 +124,12 @@ func mapOf(p *Policy) map[string]any {
 	}
 	return out
 }
+
+func policyChanged(prev, curr *Policy) bool {
+	if policiesEqual(prev, curr) {
+		return false
+	}
+	added, removed, changed := diffPolicies(prev, curr)
+	log.Infof("MDM policy changed: added=%v removed=%v changed=%v", added, removed, changed)
+	return true
+}

client/mdm/ticker_test.go

@@ -13,28 +13,40 @@ import (
 // testReloadInterval for speeding up the ticker cadence under `go test`
 const testReloadInterval = 1 * time.Second
 
-// withPolicyLoader overrides the package-level policyLoader for the duration
-// of the test so the ticker observes a scripted policy instead of the real
-// OS-native store. The original loader is restored on cleanup.
-func withPolicyLoader(t *testing.T, fn func() *Policy) {
-	t.Helper()
-	prev := policyLoader
-	policyLoader = fn
-	t.Cleanup(func() { policyLoader = prev })
+// fakePolicyFetcher implements PolicyFetcher returning a scripted
+// policy map. Goroutine-safe so the test can mutate the script while
+// the ticker is observing it.
+type fakePolicyFetcher struct {
+	mu     sync.Mutex
+	values map[string]any
+}
+
+func (f *fakePolicyFetcher) Fetch() map[string]any {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if f.values == nil {
+		return nil
+	}
+	out := make(map[string]any, len(f.values))
+	for k, v := range f.values {
+		out[k] = v
+	}
+	return out
+}
+
+func (f *fakePolicyFetcher) set(values map[string]any) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.values = values
 }
 
 func TestTicker_FiresOnChangeWithDelta(t *testing.T) {
-	var mu sync.Mutex
-	current := NewPolicy(nil) // initial observation: empty (no enforcement)
-	withPolicyLoader(t, func() *Policy {
-		mu.Lock()
-		defer mu.Unlock()
-		return current
-	})
+	fetcher := &fakePolicyFetcher{} // initial observation: empty (no enforcement)
+	loader := NewLoader(fetcher)
 
 	type change struct{ prev, curr *Policy }
 	changes := make(chan change, 1)
-	tk := NewTicker(testReloadInterval)
+	tk := NewTicker(testReloadInterval, loader)
 	require.Equal(t, testReloadInterval, tk.interval)
 
 	ctx, cancel := context.WithCancel(context.Background())
@@ -49,15 +61,13 @@ func TestTicker_FiresOnChangeWithDelta(t *testing.T) {
 		})
 		close(done)
 	}()
-	// Stop Run and wait for it to exit before returning, so the policyLoader
-	// restore in t.Cleanup can't race the ticker goroutine still reading it.
+	// Stop Run and wait for it to exit before returning, so the test
+	// goroutine doesn't race the still-running ticker.
 	defer func() { cancel(); <-done }()
 
-	// Flip the OS-observed policy from empty to one managed key. The next
-	// tick must detect the diff and invoke onChange.
-	mu.Lock()
-	current = NewPolicy(map[string]any{KeyManagementURL: "https://mdm.example.com:443"})
-	mu.Unlock()
+	// Flip the OS-observed policy from empty to one managed key. The
+	// next tick must detect the diff and invoke onChange.
+	fetcher.set(map[string]any{KeyManagementURL: "https://mdm.example.com:443"})
 
 	select {
 	case c := <-changes:
@@ -69,12 +79,11 @@ func TestTicker_FiresOnChangeWithDelta(t *testing.T) {
 }
 
 func TestTicker_NoCallbackWhenPolicyUnchanged(t *testing.T) {
-	withPolicyLoader(t, func() *Policy {
-		return NewPolicy(map[string]any{KeyBlockInbound: true})
-	})
+	fetcher := &fakePolicyFetcher{values: map[string]any{KeyBlockInbound: true}}
+	loader := NewLoader(fetcher)
 
 	fired := make(chan struct{}, 1)
-	tk := NewTicker(testReloadInterval)
+	tk := NewTicker(testReloadInterval, loader)
 
 	ctx, cancel := context.WithCancel(context.Background())
 	done := make(chan struct{})
@@ -90,8 +99,8 @@ func TestTicker_NoCallbackWhenPolicyUnchanged(t *testing.T) {
 	}()
 	defer func() { cancel(); <-done }()
 
-	// Over ~2 ticks at the 1s test cadence the policy never changes, so the
-	// diff guard must suppress the callback entirely.
+	// Over ~2 ticks at the 1s test cadence the policy never changes,
+	// so the diff guard must suppress the callback entirely.
 	select {
 	case <-fired:
 		t.Fatal("onChange fired despite an unchanged policy")

client/mobile/profile_manager.go

@@ -4,6 +4,7 @@
 package mobile
 
 import (
+	"errors"
 	"fmt"
 	"os"
 	"path/filepath"
@@ -11,6 +12,7 @@ import (
 	log "github.com/sirupsen/logrus"
 
 	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
 )
 
 const (
@@ -22,6 +24,9 @@ const (
 	profilesSubdir = "profiles"
 )
 
+// ErrProfilesDisabled marks a profile mutation rejected by MDM policy.
+var ErrProfilesDisabled = errors.New("profile management is disabled by MDM policy")
+
 /*
 
 <configDir>/                                    ← app-writable config root
@@ -55,6 +60,7 @@ type ProfileManager struct {
 	configDir  string
 	username   string
 	serviceMgr *profilemanager.ServiceManager
+	mdmLoader  *mdm.Loader
 }
 
 // NewProfileManager creates a profile manager rooted at configDir, the
@@ -127,6 +133,9 @@ func (pm *ProfileManager) GetActiveProfile() (*Profile, error) {
 // SwitchProfile records the given profile ID as the active profile. The caller
 // must stop the VPN tunnel before switching.
 func (pm *ProfileManager) SwitchProfile(id string) error {
+	if err := pm.checkProfilesAllowed(); err != nil {
+		return err
+	}
 	if err := pm.serviceMgr.SetActiveProfileState(&profilemanager.ActiveProfileState{
 		ID:       profilemanager.ID(id),
 		Username: pm.username,
@@ -141,6 +150,9 @@ func (pm *ProfileManager) SwitchProfile(id string) error {
 // AddProfile creates a new profile with the given display name and a
 // generated ID. It returns the created profile so the caller learns the ID.
 func (pm *ProfileManager) AddProfile(displayName string) (*Profile, error) {
+	if err := pm.checkProfilesAllowed(); err != nil {
+		return nil, err
+	}
 	profile, err := pm.serviceMgr.AddProfile(displayName, pm.username)
 	if err != nil {
 		return nil, fmt.Errorf("add profile: %w", err)
@@ -153,6 +165,9 @@ func (pm *ProfileManager) AddProfile(displayName string) (*Profile, error) {
 // RenameProfile changes the display name of the profile identified by id. The
 // on-disk filename (the ID) is left unchanged.
 func (pm *ProfileManager) RenameProfile(id string, newName string) error {
+	if err := pm.checkProfilesAllowed(); err != nil {
+		return err
+	}
 	if err := pm.serviceMgr.RenameProfile(profilemanager.ID(id), pm.username, newName); err != nil {
 		return fmt.Errorf("rename profile: %w", err)
 	}
@@ -165,6 +180,9 @@ func (pm *ProfileManager) RenameProfile(id string, newName string) error {
 // private key and SSH key from the config, forcing a re-login. The management
 // URL and other settings are preserved.
 func (pm *ProfileManager) LogoutProfile(id string) error {
+	if err := pm.checkProfileLogoutAllowed(id); err != nil {
+		return err
+	}
 	configPath, err := pm.getProfileConfigPath(id)
 	if err != nil {
 		return err
@@ -196,6 +214,9 @@ func (pm *ProfileManager) LogoutProfile(id string) error {
 // RemoveProfile deletes a profile. The default profile and the active profile
 // cannot be removed.
 func (pm *ProfileManager) RemoveProfile(id string) error {
+	if err := pm.checkProfilesAllowed(); err != nil {
+		return err
+	}
 	configPath, err := pm.getProfileConfigPath(id)
 	if err != nil {
 		return err
@@ -267,6 +288,27 @@ func (pm *ProfileManager) GetActiveStateFilePath() (string, error) {
 	return pm.GetStateFilePath(activeProfile.ID)
 }
 
+// SetMDMLoader registers the MDM policy source consulted before profile
+// mutations; a nil loader disables enforcement.
+func (pm *ProfileManager) SetMDMLoader(loader *mdm.Loader) {
+	pm.mdmLoader = loader
+}
+
+func (pm *ProfileManager) checkProfilesAllowed() error {
+	if v, ok := pm.mdmLoader.Load().GetBool(mdm.KeyDisableProfiles); ok && v {
+		return ErrProfilesDisabled
+	}
+	return nil
+}
+
+func (pm *ProfileManager) checkProfileLogoutAllowed(id string) error {
+	active, err := pm.serviceMgr.GetActiveProfileState()
+	if err == nil && active.ID.String() == id {
+		return nil
+	}
+	return pm.checkProfilesAllowed()
+}
+
 // profileEmail returns the account email recorded for a profile. Display-only,
 // so an unresolvable path degrades to "" rather than an error.
 func (pm *ProfileManager) profileEmail(id string) string {

client/mobile/profile_manager_mdm_test.go

@@ -0,0 +1,83 @@
+package mobile
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/profilemanager"
+	"github.com/netbirdio/netbird/client/mdm"
+)
+
+type fakeFetcher struct{ values map[string]any }
+
+func (f *fakeFetcher) Fetch() map[string]any { return f.values }
+
+func newTestProfileManager(t *testing.T) *ProfileManager {
+	t.Helper()
+	origDir := profilemanager.DefaultConfigPathDir
+	origPath := profilemanager.DefaultConfigPath
+	origActive := profilemanager.ActiveProfileStatePath
+	t.Cleanup(func() {
+		profilemanager.DefaultConfigPathDir = origDir
+		profilemanager.DefaultConfigPath = origPath
+		profilemanager.ActiveProfileStatePath = origActive
+	})
+
+	configDir := t.TempDir()
+	pm := NewProfileManager(configDir, "mobile")
+	_, err := profilemanager.UpdateOrCreateConfig(profilemanager.ConfigInput{
+		ConfigPath: filepath.Join(configDir, defaultConfigFilename),
+	})
+	require.NoError(t, err)
+	return pm
+}
+
+func privateKeyOf(t *testing.T, pm *ProfileManager, id string) string {
+	t.Helper()
+	path, err := pm.getProfileConfigPath(id)
+	require.NoError(t, err)
+	raw, err := os.ReadFile(path)
+	require.NoError(t, err)
+	var cfg struct{ PrivateKey string }
+	require.NoError(t, json.Unmarshal(raw, &cfg))
+	return cfg.PrivateKey
+}
+
+func TestLogoutProfile_DisableProfiles(t *testing.T) {
+	pm := newTestProfileManager(t)
+	other, err := pm.AddProfile("work")
+	require.NoError(t, err)
+	require.NoError(t, pm.SwitchProfile(profilemanager.DefaultProfileName))
+	require.NotEmpty(t, privateKeyOf(t, pm, profilemanager.DefaultProfileName))
+	require.NotEmpty(t, privateKeyOf(t, pm, other.ID))
+
+	pm.SetMDMLoader(mdm.NewLoader(&fakeFetcher{values: map[string]any{
+		mdm.KeyDisableProfiles: true,
+	}}))
+
+	err = pm.LogoutProfile(other.ID)
+	assert.ErrorIs(t, err, ErrProfilesDisabled)
+	assert.NotEmpty(t, privateKeyOf(t, pm, other.ID))
+
+	require.NoError(t, pm.LogoutProfile(profilemanager.DefaultProfileName))
+	assert.Empty(t, privateKeyOf(t, pm, profilemanager.DefaultProfileName))
+}
+
+func TestLogoutProfile_ProfilesAllowed(t *testing.T) {
+	pm := newTestProfileManager(t)
+	other, err := pm.AddProfile("work")
+	require.NoError(t, err)
+	require.NoError(t, pm.SwitchProfile(profilemanager.DefaultProfileName))
+
+	pm.SetMDMLoader(mdm.NewLoader(&fakeFetcher{values: map[string]any{
+		mdm.KeyDisableProfiles: false,
+	}}))
+
+	require.NoError(t, pm.LogoutProfile(other.ID))
+	assert.Empty(t, privateKeyOf(t, pm, other.ID))
+}

client/server/mdm.go

@@ -3,7 +3,6 @@ package server
 import (
 	"context"
 	"fmt"
-	"net/url"
 	"time"
 
 	log "github.com/sirupsen/logrus"
@@ -14,28 +13,6 @@ import (
 	"github.com/netbirdio/netbird/client/proto"
 )
 
-// preSharedKeyRedactedSentinel is the value GetConfig returns in place
-// of an actual PSK, so a UI that round-trips the field back to the
-// daemon (via SetConfig / Login) can be distinguished from a deliberate
-// override. Any incoming PSK that equals this sentinel is treated as
-// a no-op echo, never as a conflict with the policy.
-const preSharedKeyRedactedSentinel = "**********"
-
-// loadMDMPolicy is the indirection used by server handlers to read the
-// active MDM policy. Tests override this to inject a fake policy.
-var loadMDMPolicy = mdm.LoadPolicy
-
-// conflictCheck is a value-aware comparison between a single field in
-// the incoming request and the corresponding MDM-enforced value. It
-// runs only when the field was actually set in the request (presence
-// already filtered upstream); ok=true reports the policy value, ok=false
-// means the policy is silent on the key — both are treated as conflicts
-// to be safe (an MDM key declared as managed must hold a value).
-type conflictCheck struct {
-	key   string
-	check func(*mdm.Policy) (match bool)
-}
-
 // onMDMPolicyChange is invoked by the MDM reload ticker every time the
 // OS-native managed-config store reports a diff vs the last observation.
 //
@@ -168,126 +145,6 @@ func (s *Server) restartEngineForMDMLocked() error {
 	return nil
 }
 
-// conflictBool builds a conflictCheck for a boolean MDM key. If p is nil
-// the field is treated as matching (no override requested); otherwise the
-// check returns true only when the policy contains the key and its
-// boolean value equals *p.
-func conflictBool(key string, p *bool) conflictCheck {
-	return conflictCheck{
-		key: key,
-		check: func(pol *mdm.Policy) bool {
-			if p == nil {
-				return true // absent → match by definition
-			}
-			want, ok := pol.GetBool(key)
-			return ok && want == *p
-		},
-	}
-}
-
-func canonicalURL(s string) string {
-	u, err := url.ParseRequestURI(s)
-	if err != nil {
-		return s
-	}
-	if u.Port() == "" {
-		switch u.Scheme {
-		case "https":
-			u.Host += ":443"
-		case "http":
-			u.Host += ":80"
-		}
-	}
-	return u.String()
-}
-
-// conflictURL is conflictString for URL-typed keys: both sides are
-// normalized via canonicalURL before comparison.
-func conflictURL(key, got string) conflictCheck {
-	return conflictCheck{
-		key: key,
-		check: func(pol *mdm.Policy) bool {
-			if got == "" {
-				return true
-			}
-			want, ok := pol.GetString(key)
-			return ok && canonicalURL(want) == canonicalURL(got)
-		},
-	}
-}
-
-// conflictString builds a conflictCheck for a string MDM key. An empty
-// `got` is treated as "field not set" (no override requested); otherwise
-// the check returns true only when the policy contains the key and its
-// value equals got.
-func conflictString(key, got string) conflictCheck {
-	return conflictCheck{
-		key: key,
-		check: func(pol *mdm.Policy) bool {
-			if got == "" {
-				return true
-			}
-			want, ok := pol.GetString(key)
-			return ok && want == got
-		},
-	}
-}
-
-// conflictStringPtr is conflictString for optional proto fields, where an
-// explicit empty value is still a request to change the setting. If p is
-// nil the field is treated as matching (no override requested); otherwise
-// the check returns true only when the policy contains the key and its
-// value equals *p.
-func conflictStringPtr(key string, p *string) conflictCheck {
-	return conflictCheck{
-		key: key,
-		check: func(pol *mdm.Policy) bool {
-			if p == nil {
-				return true
-			}
-			want, ok := pol.GetString(key)
-			return ok && want == *p
-		},
-	}
-}
-
-// conflictInt64 builds a conflictCheck for an integer MDM key. If p is
-// nil the field is treated as matching; otherwise the check returns
-// true only when the policy contains the key and its int value equals *p.
-func conflictInt64(key string, p *int64) conflictCheck {
-	return conflictCheck{
-		key: key,
-		check: func(pol *mdm.Policy) bool {
-			if p == nil {
-				return true
-			}
-			want, ok := pol.GetInt(key)
-			return ok && want == *p
-		},
-	}
-}
-
-// resolveConflicts walks the per-field checks against the active MDM
-// policy and returns the names of keys whose requested value diverges
-// from the policy-enforced value. Keys not present in the policy are
-// skipped silently (the gate fires only for keys the admin has
-// actually pushed). Returns nil for an empty policy.
-func resolveConflicts(policy *mdm.Policy, checks []conflictCheck) []string {
-	if policy.IsEmpty() {
-		return nil
-	}
-	var conflicts []string
-	for _, c := range checks {
-		if !policy.HasKey(c.key) {
-			continue
-		}
-		if !c.check(policy) {
-			conflicts = append(conflicts, c.key)
-		}
-	}
-	return conflicts
-}
-
 // mdmManagedFieldConflicts returns the names of MDM-managed keys whose
 // requested value in the SetConfigRequest differs from the MDM-enforced
 // value. A field set to the same value the policy already enforces is
@@ -301,27 +158,25 @@ func mdmManagedFieldConflicts(msg *proto.SetConfigRequest, policy *mdm.Policy) [
 		return nil
 	}
 
-	// PSK round-trip echo: collapse the sentinel to empty so the
-	// shared check treats it as "field not set".
-	pskGot := ""
-	if msg.OptionalPreSharedKey != nil && *msg.OptionalPreSharedKey != preSharedKeyRedactedSentinel {
-		pskGot = *msg.OptionalPreSharedKey
+	pskGot := msg.OptionalPreSharedKey
+	if pskGot != nil && *pskGot == mdm.PreSharedKeyRedactedSentinel {
+		pskGot = nil
 	}
 
-	return resolveConflicts(policy, []conflictCheck{
-		conflictURL(mdm.KeyManagementURL, msg.ManagementUrl),
-		conflictString(mdm.KeyPreSharedKey, pskGot),
-		conflictBool(mdm.KeyRosenpassEnabled, msg.RosenpassEnabled),
-		conflictBool(mdm.KeyRosenpassPermissive, msg.RosenpassPermissive),
-		conflictBool(mdm.KeyDisableAutoConnect, msg.DisableAutoConnect),
-		conflictBool(mdm.KeyAllowServerSSH, msg.ServerSSHAllowed),
-		conflictBool(mdm.KeyRemoteJobsAllowed, msg.RemoteJobsAllowed),
-		conflictBool(mdm.KeyDisableClientRoutes, msg.DisableClientRoutes),
-		conflictBool(mdm.KeyDisableServerRoutes, msg.DisableServerRoutes),
-		conflictBool(mdm.KeyBlockInbound, msg.BlockInbound),
-		conflictInt64(mdm.KeyWireguardPort, msg.WireguardPort),
-		conflictBool(mdm.KeyEnableLocalMetrics, msg.EnableLocalMetrics),
-		conflictStringPtr(mdm.KeyLocalMetricsAddress, msg.LocalMetricsAddress),
+	return mdm.ResolveConflicts(policy, []mdm.ConflictCheck{
+		mdm.ConflictURL(mdm.KeyManagementURL, msg.ManagementUrl),
+		mdm.ConflictStringPtr(mdm.KeyPreSharedKey, pskGot),
+		mdm.ConflictBool(mdm.KeyRosenpassEnabled, msg.RosenpassEnabled),
+		mdm.ConflictBool(mdm.KeyRosenpassPermissive, msg.RosenpassPermissive),
+		mdm.ConflictBool(mdm.KeyDisableAutoConnect, msg.DisableAutoConnect),
+		mdm.ConflictBool(mdm.KeyAllowServerSSH, msg.ServerSSHAllowed),
+		mdm.ConflictBool(mdm.KeyRemoteJobsAllowed, msg.RemoteJobsAllowed),
+		mdm.ConflictBool(mdm.KeyDisableClientRoutes, msg.DisableClientRoutes),
+		mdm.ConflictBool(mdm.KeyDisableServerRoutes, msg.DisableServerRoutes),
+		mdm.ConflictBool(mdm.KeyBlockInbound, msg.BlockInbound),
+		mdm.ConflictInt64(mdm.KeyWireguardPort, msg.WireguardPort),
+		mdm.ConflictBool(mdm.KeyEnableLocalMetrics, msg.EnableLocalMetrics),
+		mdm.ConflictStringPtr(mdm.KeyLocalMetricsAddress, msg.LocalMetricsAddress),
 	})
 }
 
@@ -424,34 +279,28 @@ func loginRequestMDMConflicts(msg *proto.LoginRequest, policy *mdm.Policy) []str
 		return nil
 	}
 
-	// Collapse the two PSK fields + the redaction sentinel down to a
-	// single "got" string the shared check can compare against the
-	// policy: OptionalPreSharedKey wins if set; PreSharedKey (deprecated)
-	// is the fallback; sentinel echo is treated as "field not set".
-	pskGot := ""
-	if msg.OptionalPreSharedKey != nil {
-		pskGot = *msg.OptionalPreSharedKey
-	} else if msg.PreSharedKey != "" { //nolint:staticcheck // SA1019: legacy proto field still accepted by Login
-		pskGot = msg.PreSharedKey //nolint:staticcheck // SA1019
+	pskGot := msg.OptionalPreSharedKey
+	if pskGot == nil && msg.PreSharedKey != "" { //nolint:staticcheck // SA1019: legacy proto field still accepted by Login
+		pskGot = &msg.PreSharedKey //nolint:staticcheck // SA1019
 	}
-	if pskGot == preSharedKeyRedactedSentinel {
-		pskGot = ""
+	if pskGot != nil && *pskGot == mdm.PreSharedKeyRedactedSentinel {
+		pskGot = nil
 	}
 
-	return resolveConflicts(policy, []conflictCheck{
-		conflictURL(mdm.KeyManagementURL, msg.ManagementUrl),
-		conflictString(mdm.KeyPreSharedKey, pskGot),
-		conflictBool(mdm.KeyRosenpassEnabled, msg.RosenpassEnabled),
-		conflictBool(mdm.KeyRosenpassPermissive, msg.RosenpassPermissive),
-		conflictBool(mdm.KeyDisableAutoConnect, msg.DisableAutoConnect),
-		conflictBool(mdm.KeyAllowServerSSH, msg.ServerSSHAllowed),
-		conflictBool(mdm.KeyRemoteJobsAllowed, msg.RemoteJobsAllowed),
-		conflictBool(mdm.KeyDisableClientRoutes, msg.DisableClientRoutes),
-		conflictBool(mdm.KeyDisableServerRoutes, msg.DisableServerRoutes),
-		conflictBool(mdm.KeyBlockInbound, msg.BlockInbound),
-		conflictInt64(mdm.KeyWireguardPort, msg.WireguardPort),
-		conflictBool(mdm.KeyEnableLocalMetrics, msg.EnableLocalMetrics),
-		conflictStringPtr(mdm.KeyLocalMetricsAddress, msg.LocalMetricsAddress),
+	return mdm.ResolveConflicts(policy, []mdm.ConflictCheck{
+		mdm.ConflictURL(mdm.KeyManagementURL, msg.ManagementUrl),
+		mdm.ConflictStringPtr(mdm.KeyPreSharedKey, pskGot),
+		mdm.ConflictBool(mdm.KeyRosenpassEnabled, msg.RosenpassEnabled),
+		mdm.ConflictBool(mdm.KeyRosenpassPermissive, msg.RosenpassPermissive),
+		mdm.ConflictBool(mdm.KeyDisableAutoConnect, msg.DisableAutoConnect),
+		mdm.ConflictBool(mdm.KeyAllowServerSSH, msg.ServerSSHAllowed),
+		mdm.ConflictBool(mdm.KeyRemoteJobsAllowed, msg.RemoteJobsAllowed),
+		mdm.ConflictBool(mdm.KeyDisableClientRoutes, msg.DisableClientRoutes),
+		mdm.ConflictBool(mdm.KeyDisableServerRoutes, msg.DisableServerRoutes),
+		mdm.ConflictBool(mdm.KeyBlockInbound, msg.BlockInbound),
+		mdm.ConflictInt64(mdm.KeyWireguardPort, msg.WireguardPort),
+		mdm.ConflictBool(mdm.KeyEnableLocalMetrics, msg.EnableLocalMetrics),
+		mdm.ConflictStringPtr(mdm.KeyLocalMetricsAddress, msg.LocalMetricsAddress),
 	})
 }
 

client/server/server.go

@@ -138,6 +138,15 @@ type Server struct {
 	// stopped by the rootCtx cancellation.
 	mdmTicker *mdm.Ticker
 
+	// mdmLoader is the daemon-owned source of the active MDM policy.
+	// Constructed once during Server.Start (with a nil PolicyFetcher on
+	// desktop — the build-tagged Loader.loadPlatform reads the OS
+	// registry / plist directly) and injected into every consumer:
+	// mdmTicker for its periodic reload, the SetConfig / Login MDM
+	// gates for conflict detection, and every Config produced via
+	// getConfig() so its apply() picks up the same overlay.
+	mdmLoader *mdm.Loader
+
 	updateManager *updater.Manager
 
 	jwtCache *jwtCache
@@ -246,8 +255,14 @@ func (s *Server) Start() error {
 	// Runs re-resolves Config (re-running profilemanager.Config.apply which
 	// applies the freshly-read MDM policy as the last layer) and brings
 	// the engine back with the new values.
+	if s.mdmLoader == nil {
+		// Desktop builds pass a nil PolicyFetcher: the Loader's
+		// build-tagged loadPlatform reads the OS source directly
+		// (registry on Windows, plist on macOS, no-op elsewhere).
+		s.mdmLoader = mdm.NewLoader(nil)
+	}
 	if s.mdmTicker == nil {
-		s.mdmTicker = mdm.NewTicker(mdm.DefaultReloadInterval)
+		s.mdmTicker = mdm.NewTicker(mdm.DefaultReloadInterval, s.mdmLoader)
 		go s.mdmTicker.Run(s.rootCtx, s.onMDMPolicyChange)
 	}
 
@@ -493,7 +508,7 @@ func (s *Server) SetConfig(callerCtx context.Context, msg *proto.SetConfigReques
 	// by the active MDM policy. The error carries an MDMManagedFields-
 	// Violation detail listing the offending key names. Non-conflicting
 	// fields in the same request are not applied either.
-	policy := loadMDMPolicy()
+	policy := s.mdmLoader.Load()
 	if err := rejectMDMManagedFieldConflicts(mdmManagedFieldConflicts(msg, policy)); err != nil {
 		return nil, err
 	}
@@ -636,7 +651,7 @@ func (s *Server) Login(callerCtx context.Context, msg *proto.LoginRequest) (*pro
 		if s.checkUpdateSettingsDisabled() {
 			return nil, gstatus.Errorf(codes.Unavailable, errUpdateSettingsDisabled)
 		}
-		policy := loadMDMPolicy()
+		policy := s.mdmLoader.Load()
 		if err := rejectMDMManagedFieldConflicts(loginRequestMDMConflicts(msg, policy)); err != nil {
 			return nil, err
 		}
@@ -1487,6 +1502,12 @@ func (s *Server) getConfig(activeProf *profilemanager.ActiveProfileState) (*prof
 		return nil, false, fmt.Errorf("failed to get config: %w", err)
 	}
 
+	// Apply the daemon-owned MDM policy on top of the just-resolved
+	// Config. profilemanager's apply() initialises the policy to
+	// empty — the Loader lives outside Config, so this overlay step
+	// is driven externally here.
+	config.ApplyMDMPolicy(s.mdmLoader.Load())
+
 	return config, configExisted, nil
 }
 
@@ -1543,6 +1564,9 @@ func (s *Server) logoutFromProfile(ctx context.Context, profile *profilemanager.
 	if err != nil {
 		return fmt.Errorf("profile '%s' not found", profile.ID)
 	}
+	// Honour any MDM-enforced ManagementURL when issuing the logout
+	// RPC: the user-stored value may have been overridden by policy.
+	config.ApplyMDMPolicy(s.mdmLoader.Load())
 
 	return s.sendLogoutRequestWithConfig(ctx, config)
 }
@@ -2177,6 +2201,11 @@ func (s *Server) GetConfig(ctx context.Context, req *proto.GetConfigRequest) (*p
 		log.Errorf("failed to get active profile config: %v", err)
 		return nil, fmt.Errorf("failed to get active profile config: %w", err)
 	}
+	// Overlay the active MDM policy so the response's MDMManagedFields
+	// list reflects what the GUI / CLI must render as read-only.
+	// profilemanager.GetConfig itself returns a Config without the
+	// overlay (Loader lives outside profilemanager).
+	cfg.ApplyMDMPolicy(s.mdmLoader.Load())
 	managementURL := cfg.ManagementURL
 	adminURL := cfg.AdminURL
 

client/server/setconfig_mdm_test.go

@@ -16,14 +16,40 @@ import (
 	"github.com/netbirdio/netbird/client/proto"
 )
 
-// withMDMPolicy temporarily overrides the server-package loadMDMPolicy hook
-// so SetConfig observes the supplied Policy. Restores the original loader
-// at test cleanup.
-func withMDMPolicy(t *testing.T, policy *mdm.Policy) {
+// fakeMDMFetcher implements mdm.PolicyFetcher returning a pre-set
+// policy map. Tests build one per Server instance to inject a
+// scripted MDM overlay via a Loader rather than via package-level state.
+type fakeMDMFetcher struct{ values map[string]any }
+
+func (f *fakeMDMFetcher) Fetch() map[string]any { return f.values }
+
+// withMDMPolicy installs an mdm.Loader on the given Server whose
+// loadPlatform returns the supplied Policy's underlying values. Use
+// after setupServerWithProfile to inject the scripted policy the
+// SetConfig / Login MDM gates will observe.
+func withMDMPolicy(t *testing.T, s *Server, policy *mdm.Policy) {
 	t.Helper()
-	prev := loadMDMPolicy
-	loadMDMPolicy = func() *mdm.Policy { return policy }
-	t.Cleanup(func() { loadMDMPolicy = prev })
+	values := map[string]any{}
+	if policy != nil {
+		for _, k := range policy.ManagedKeys() {
+			if v, ok := policy.GetString(k); ok {
+				values[k] = v
+				continue
+			}
+			if v, ok := policy.GetInt(k); ok {
+				values[k] = v
+				continue
+			}
+			if v, ok := policy.GetBool(k); ok {
+				values[k] = v
+				continue
+			}
+			if v, ok := policy.GetStringSlice(k); ok {
+				values[k] = v
+			}
+		}
+	}
+	s.mdmLoader = mdm.NewLoader(&fakeMDMFetcher{values: values})
 }
 
 // setupServerWithProfile mirrors the boilerplate of TestSetConfig_AllFieldsSaved:
@@ -93,12 +119,11 @@ func extractViolation(t *testing.T, err error) *proto.MDMManagedFieldsViolation
 }
 
 func TestSetConfig_MDMReject_SingleField(t *testing.T) {
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	s, ctx, profName, username, _ := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
 		mdm.KeyManagementURL: "https://mdm.example.com:443",
 	}))
 
-	s, ctx, profName, username, _ := setupServerWithProfile(t)
-
 	_, err := s.SetConfig(ctx, &proto.SetConfigRequest{
 		ProfileName:   profName,
 		Username:      username,
@@ -110,14 +135,13 @@ func TestSetConfig_MDMReject_SingleField(t *testing.T) {
 }
 
 func TestSetConfig_MDMReject_MultipleFields(t *testing.T) {
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	s, ctx, profName, username, _ := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
 		mdm.KeyManagementURL:    "https://mdm.example.com:443",
 		mdm.KeyBlockInbound:     true,
 		mdm.KeyRosenpassEnabled: true,
 	}))
 
-	s, ctx, profName, username, _ := setupServerWithProfile(t)
-
 	blockInbound := false
 	rosenpassEnabled := false
 	_, err := s.SetConfig(ctx, &proto.SetConfigRequest{
@@ -137,13 +161,12 @@ func TestSetConfig_MDMReject_MultipleFields(t *testing.T) {
 }
 
 func TestSetConfig_MDMReject_LocalMetrics(t *testing.T) {
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	s, ctx, profName, username, _ := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
 		mdm.KeyEnableLocalMetrics:  true,
 		mdm.KeyLocalMetricsAddress: "127.0.0.1:9191",
 	}))
 
-	s, ctx, profName, username, _ := setupServerWithProfile(t)
-
 	enabled := false
 	addr := "0.0.0.0:9999"
 	_, err := s.SetConfig(ctx, &proto.SetConfigRequest{
@@ -164,12 +187,11 @@ func TestSetConfig_MDMReject_LocalMetrics(t *testing.T) {
 // (the manager falls back to the default), so presence must be honored
 // rather than collapsed to "field not set".
 func TestSetConfig_MDMReject_LocalMetricsEmptyAddress(t *testing.T) {
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	s, ctx, profName, username, _ := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
 		mdm.KeyLocalMetricsAddress: "127.0.0.1:9999",
 	}))
 
-	s, ctx, profName, username, _ := setupServerWithProfile(t)
-
 	addr := ""
 	_, err := s.SetConfig(ctx, &proto.SetConfigRequest{
 		ProfileName:         profName,
@@ -181,17 +203,80 @@ func TestSetConfig_MDMReject_LocalMetricsEmptyAddress(t *testing.T) {
 	assert.ElementsMatch(t, []string{mdm.KeyLocalMetricsAddress}, v.GetFields())
 }
 
+func TestSetConfig_MDMReject_EmptyPreSharedKey(t *testing.T) {
+	s, ctx, profName, username, _ := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
+		mdm.KeyPreSharedKey: "mdm-enforced-psk",
+	}))
+
+	psk := ""
+	_, err := s.SetConfig(ctx, &proto.SetConfigRequest{
+		ProfileName:          profName,
+		Username:             username,
+		OptionalPreSharedKey: &psk,
+	})
+
+	v := extractViolation(t, err)
+	assert.ElementsMatch(t, []string{mdm.KeyPreSharedKey}, v.GetFields())
+}
+
+func TestSetConfig_MDMAllow_PreSharedKeySentinelEcho(t *testing.T) {
+	s, ctx, profName, username, _ := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
+		mdm.KeyPreSharedKey: "mdm-enforced-psk",
+	}))
+
+	psk := mdm.PreSharedKeyRedactedSentinel
+	resp, err := s.SetConfig(ctx, &proto.SetConfigRequest{
+		ProfileName:          profName,
+		Username:             username,
+		OptionalPreSharedKey: &psk,
+	})
+
+	require.NoError(t, err)
+	require.NotNil(t, resp)
+}
+
+func TestLoginRequestMDMConflicts_PreSharedKey(t *testing.T) {
+	policy := mdm.NewPolicy(map[string]any{
+		mdm.KeyPreSharedKey: "mdm-enforced-psk",
+	})
+	empty := ""
+	sentinel := mdm.PreSharedKeyRedactedSentinel
+	same := "mdm-enforced-psk"
+	other := "user-psk"
+
+	tests := []struct {
+		name string
+		msg  *proto.LoginRequest
+		want []string
+	}{
+		{name: "unset", msg: &proto.LoginRequest{}, want: nil},
+		{name: "optional empty", msg: &proto.LoginRequest{OptionalPreSharedKey: &empty}, want: []string{mdm.KeyPreSharedKey}},
+		{name: "optional sentinel echo", msg: &proto.LoginRequest{OptionalPreSharedKey: &sentinel}, want: nil},
+		{name: "optional same value", msg: &proto.LoginRequest{OptionalPreSharedKey: &same}, want: nil},
+		{name: "optional divergent", msg: &proto.LoginRequest{OptionalPreSharedKey: &other}, want: []string{mdm.KeyPreSharedKey}},
+		{name: "legacy empty is unset", msg: &proto.LoginRequest{PreSharedKey: ""}, want: nil},                         //nolint:staticcheck // SA1019: legacy proto field still accepted by Login
+		{name: "legacy sentinel echo", msg: &proto.LoginRequest{PreSharedKey: sentinel}, want: nil},                    //nolint:staticcheck // SA1019: legacy proto field still accepted by Login
+		{name: "legacy divergent", msg: &proto.LoginRequest{PreSharedKey: other}, want: []string{mdm.KeyPreSharedKey}}, //nolint:staticcheck // SA1019: legacy proto field still accepted by Login
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			assert.Equal(t, tc.want, loginRequestMDMConflicts(tc.msg, policy))
+		})
+	}
+}
+
 func TestSetConfig_MDMReject_AllOrNothing(t *testing.T) {
 	// MDM enforces ManagementURL only; user request touches both the
 	// enforced field AND a non-enforced field (RosenpassEnabled).
 	// The whole request must be rejected — non-conflicting fields are not
 	// applied either.
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	s, ctx, profName, username, cfgPath := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
 		mdm.KeyManagementURL: "https://mdm.example.com:443",
 	}))
 
-	s, ctx, profName, username, cfgPath := setupServerWithProfile(t)
-
 	rosenpassEnabled := true
 	_, err := s.SetConfig(ctx, &proto.SetConfigRequest{
 		ProfileName:      profName,
@@ -213,12 +298,11 @@ func TestSetConfig_MDMReject_AllOrNothing(t *testing.T) {
 func TestSetConfig_MDMAllow_NonManagedFields(t *testing.T) {
 	// MDM enforces ManagementURL but the user only writes RosenpassEnabled.
 	// Request must succeed.
-	withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+	s, ctx, profName, username, _ := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
 		mdm.KeyManagementURL: "https://mdm.example.com:443",
 	}))
 
-	s, ctx, profName, username, _ := setupServerWithProfile(t)
-
 	rosenpassEnabled := true
 	resp, err := s.SetConfig(ctx, &proto.SetConfigRequest{
 		ProfileName:      profName,
@@ -247,12 +331,11 @@ func TestSetConfig_MDMAllow_ManagementURLPortNormalized(t *testing.T) {
 
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
-			withMDMPolicy(t, mdm.NewPolicy(map[string]any{
+			s, ctx, profName, username, _ := setupServerWithProfile(t)
+			withMDMPolicy(t, s, mdm.NewPolicy(map[string]any{
 				mdm.KeyManagementURL: tc.mdmURL,
 			}))
 
-			s, ctx, profName, username, _ := setupServerWithProfile(t)
-
 			rosenpassEnabled := true
 			resp, err := s.SetConfig(ctx, &proto.SetConfigRequest{
 				ProfileName:      profName,
@@ -269,9 +352,8 @@ func TestSetConfig_MDMAllow_ManagementURLPortNormalized(t *testing.T) {
 
 func TestSetConfig_MDMEmpty_NoEnforcement(t *testing.T) {
 	// No MDM policy active: any field can be written.
-	withMDMPolicy(t, mdm.NewPolicy(nil))
-
 	s, ctx, profName, username, _ := setupServerWithProfile(t)
+	withMDMPolicy(t, s, mdm.NewPolicy(nil))
 
 	resp, err := s.SetConfig(ctx, &proto.SetConfigRequest{
 		ProfileName:   profName,

client/ssh/handshake.go

@@ -5,7 +5,6 @@ import (
 	"fmt"
 	"io"
 	"net"
-	"time"
 
 	log "github.com/sirupsen/logrus"
 	"golang.org/x/crypto/ssh"
@@ -13,26 +12,23 @@ import (
 
 // Handshake runs the SSH client handshake on an already dialed conn and
 // returns the resulting client. Dialing bounds only the TCP establishment;
-// without a deadline on the socket a peer that accepts and then goes silent
-// blocks the handshake forever, so the context deadline is applied to conn
-// for the duration of the handshake. conn is closed on any error.
+// a peer that accepts and then goes silent would block the handshake forever,
+// so conn is closed as soon as ctx is done, which unblocks the handshake and
+// surfaces the context error. conn is closed on any error.
 func Handshake(ctx context.Context, conn net.Conn, addr string, config *ssh.ClientConfig) (*ssh.Client, error) {
-	if deadline, ok := ctx.Deadline(); ok {
-		if err := conn.SetDeadline(deadline); err != nil {
-			closeHandshake(conn, "conn after deadline error")
-			return nil, fmt.Errorf("set handshake deadline: %w", err)
-		}
-	}
+	stop := context.AfterFunc(ctx, func() { closeHandshake(conn, "conn on context done") })
 
 	sshConn, chans, reqs, err := ssh.NewClientConn(conn, addr, config)
 	if err != nil {
-		closeHandshake(conn, "conn after handshake error")
-		return nil, fmt.Errorf("ssh handshake: %w", err)
+		if stop() {
+			closeHandshake(conn, "conn after handshake error")
+		}
+		return nil, handshakeError(ctx, err)
 	}
 
-	if err := conn.SetDeadline(time.Time{}); err != nil {
-		closeHandshake(sshConn, "ssh conn after deadline clear error")
-		return nil, fmt.Errorf("clear handshake deadline: %w", err)
+	if !stop() {
+		closeHandshake(sshConn, "ssh conn after context done")
+		return nil, fmt.Errorf("ssh handshake: %w", ctx.Err())
 	}
 
 	return ssh.NewClient(sshConn, chans, reqs), nil
@@ -43,3 +39,10 @@ func closeHandshake(c io.Closer, label string) {
 		log.Debugf("ssh: close %s: %v", label, err)
 	}
 }
+
+func handshakeError(ctx context.Context, err error) error {
+	if ctxErr := ctx.Err(); ctxErr != nil {
+		return fmt.Errorf("ssh handshake: %w: %w", ctxErr, err)
+	}
+	return fmt.Errorf("ssh handshake: %w", err)
+}

client/ssh/handshake_test.go

@@ -0,0 +1,90 @@
+package ssh
+
+import (
+	"context"
+	"errors"
+	"net"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+	"golang.org/x/crypto/ssh"
+)
+
+func TestHandshake_ContextDeadlineWrapped(t *testing.T) {
+	conn := dialSilentServer(t)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 50*time.Millisecond)
+	defer cancel()
+
+	_, err := Handshake(ctx, conn, conn.RemoteAddr().String(), testClientConfig())
+	require.Error(t, err)
+	require.True(t, errors.Is(err, context.DeadlineExceeded), "expected context.DeadlineExceeded, got: %v", err)
+}
+
+func TestHandshake_ContextCancelUnblocks(t *testing.T) {
+	conn := dialSilentServer(t)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	time.AfterFunc(50*time.Millisecond, cancel)
+
+	errCh := make(chan error, 1)
+	go func() {
+		_, err := Handshake(ctx, conn, conn.RemoteAddr().String(), testClientConfig())
+		errCh <- err
+	}()
+
+	select {
+	case err := <-errCh:
+		require.Error(t, err)
+		require.True(t, errors.Is(err, context.Canceled), "expected context.Canceled, got: %v", err)
+	case <-time.After(5 * time.Second):
+		t.Fatal("handshake did not return after context cancellation")
+	}
+}
+
+func TestHandshake_NonContextErrorNotWrapped(t *testing.T) {
+	conn := dialSilentServer(t)
+	require.NoError(t, conn.Close())
+
+	_, err := Handshake(context.Background(), conn, conn.RemoteAddr().String(), testClientConfig())
+	require.Error(t, err)
+	require.False(t, errors.Is(err, context.Canceled))
+	require.False(t, errors.Is(err, context.DeadlineExceeded))
+}
+
+func testClientConfig() *ssh.ClientConfig {
+	return &ssh.ClientConfig{
+		User:            "test",
+		HostKeyCallback: ssh.InsecureIgnoreHostKey(),
+	}
+}
+
+// dialSilentServer returns a client conn to a server that accepts and never
+// sends anything, so the SSH handshake blocks until the context is done.
+func dialSilentServer(t *testing.T) net.Conn {
+	t.Helper()
+
+	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = listener.Close() })
+
+	done := make(chan struct{})
+	t.Cleanup(func() { close(done) })
+
+	go func() {
+		c, err := listener.Accept()
+		if err != nil {
+			return
+		}
+		defer func() { _ = c.Close() }()
+		<-done
+	}()
+
+	conn, err := net.Dial("tcp", listener.Addr().String())
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = conn.Close() })
+
+	return conn
+}

client/ui/autostart_default.go

@@ -72,7 +72,7 @@ func netbirdFootprintExists() bool {
 // retrying autostart entry writes on every launch. A user's later disable in
 // Settings is never overridden: the marker guarantees at-most-once, ever.
 func applyAutostartDefault(ctx context.Context, autostart *services.Autostart, prefs *preferences.Store, prefsFileExisted bool) {
-	mdmDisabled := autostartDisabledByMDM(mdm.LoadPolicy())
+	mdmDisabled := autostartDisabledByMDM(mdm.NewLoader(nil).Load())
 
 	if mdmDisabled {
 		if enabled, err := autostart.IsEnabled(ctx); err != nil {

client/ui/services/settings.go

@@ -11,37 +11,18 @@ import (
 
 	"github.com/netbirdio/netbird/client/internal/daemonaddr"
 	"github.com/netbirdio/netbird/client/internal/ipcauth"
+	"github.com/netbirdio/netbird/client/mdm"
 	"github.com/netbirdio/netbird/client/proto"
 )
 
-type MDMFields struct {
-	ManagementURL            string `json:"managementURL"`
-	PreSharedKey             bool   `json:"preSharedKey"`
-	WireguardPort            bool   `json:"wireguardPort"`
-	RosenpassEnabled         bool   `json:"rosenpassEnabled"`
-	RosenpassPermissive      bool   `json:"rosenpassPermissive"`
-	DisableClientRoutes      bool   `json:"disableClientRoutes"`
-	DisableServerRoutes      bool   `json:"disableServerRoutes"`
-	AllowServerSSH           *bool  `json:"allowServerSSH"`
-	DisableAutoConnect       bool   `json:"disableAutoConnect"`
-	DisableAutostart         bool   `json:"disableAutostart"`
-	BlockInbound             bool   `json:"blockInbound"`
-	DisableMetricsCollection bool   `json:"disableMetricsCollection"`
-	SplitTunnelMode          bool   `json:"splitTunnelMode"`
-	SplitTunnelApps          bool   `json:"splitTunnelApps"`
-	DisableAdvancedView      bool   `json:"disableAdvancedView"`
-}
+// MDMFields is the shared per-key MDM enforcement snapshot; see mdm.Fields.
+type MDMFields = mdm.Fields
 
-type Features struct {
-	DisableProfiles       bool `json:"disableProfiles"`
-	DisableNetworks       bool `json:"disableNetworks"`
-	DisableUpdateSettings bool `json:"disableUpdateSettings"`
-}
+// Features is the shared feature-gate snapshot; see mdm.Features.
+type Features = mdm.Features
 
-type Restrictions struct {
-	MDM      MDMFields `json:"mdm"`
-	Features Features  `json:"features"`
-}
+// Restrictions is the shared UI enforcement snapshot; see mdm.Restrictions.
+type Restrictions = mdm.Restrictions
 
 // Privilege tells the frontend whether this process may perform the changes the
 // daemon restricts to root/administrator, whether it can ask the operating
@@ -383,7 +364,7 @@ func (s *Settings) GetRestrictions(ctx context.Context) (Restrictions, error) {
 		},
 	}
 	applyMDMRestrictions(&r.MDM, cfgResp)
-	r.MDM.DisableAdvancedView = featResp.GetDisableAdvancedView()
+	r.MDM.DisableAdvancedView = featResp.DisableAdvancedView
 	return r, nil
 }
 
@@ -411,9 +392,6 @@ func applyMDMRestrictions(mdm *MDMFields, cfgResp *proto.GetConfigResponse) {
 		if v.Field(i).Kind() != reflect.Bool {
 			continue
 		}
-		if t.Field(i).Name == "DisableAdvancedView" {
-			continue
-		}
 		if _, ok := set[t.Field(i).Tag.Get("json")]; ok {
 			v.Field(i).SetBool(true)
 		}

combined/cmd/root.go

@@ -205,7 +205,7 @@ func createAllServers(ctx context.Context, cfg *CombinedConfig) (*serverInstance
 		metricsServer: metricsServer,
 	}
 
-	_, tlsSupport, err := handleTLSConfig(cfg)
+	tlsConfig, tlsSupport, err := handleTLSConfig(cfg)
 	if err != nil {
 		return nil, fmt.Errorf("failed to setup TLS config: %w", err)
 	}
@@ -214,7 +214,7 @@ func createAllServers(ctx context.Context, cfg *CombinedConfig) (*serverInstance
 		return nil, err
 	}
 
-	if err := servers.createManagementServer(ctx, cfg); err != nil {
+	if err := servers.createManagementServer(ctx, cfg, tlsConfig); err != nil {
 		return nil, err
 	}
 
@@ -264,7 +264,7 @@ func (s *serverInstances) createRelayServer(cfg *CombinedConfig, tlsSupport bool
 	return nil
 }
 
-func (s *serverInstances) createManagementServer(ctx context.Context, cfg *CombinedConfig) error {
+func (s *serverInstances) createManagementServer(ctx context.Context, cfg *CombinedConfig, tlsConfig *tls.Config) error {
 	if !cfg.Management.Enabled {
 		return nil
 	}
@@ -297,7 +297,7 @@ func (s *serverInstances) createManagementServer(ctx context.Context, cfg *Combi
 
 	LogConfigInfo(mgmtConfig)
 
-	s.mgmtSrv, err = createManagementServer(cfg, mgmtConfig)
+	s.mgmtSrv, err = createManagementServer(cfg, mgmtConfig, tlsConfig)
 	if err != nil {
 		cleanupSTUNListeners(s.stunListeners)
 		return fmt.Errorf("failed to create management server: %w", err)
@@ -513,7 +513,7 @@ func handleTLSConfig(cfg *CombinedConfig) (*tls.Config, bool, error) {
 	return nil, false, nil
 }
 
-func createManagementServer(cfg *CombinedConfig, mgmtConfig *nbconfig.Config) (mgmtServer.Server, error) {
+func createManagementServer(cfg *CombinedConfig, mgmtConfig *nbconfig.Config, tlsConfig *tls.Config) (mgmtServer.Server, error) {
 	mgmt := cfg.Management
 
 	// Extract port from listen address
@@ -542,6 +542,7 @@ func createManagementServer(cfg *CombinedConfig, mgmtConfig *nbconfig.Config) (m
 			AutoResolveDomains:      true,
 			MgmtPort:                mgmtPort,
 			MgmtMetricsPort:         cfg.Server.MetricsPort,
+			TLSConfig:               tlsConfig,
 			DisableMetrics:          mgmt.DisableAnonymousMetrics,
 			DisableGeoliteUpdate:    mgmt.DisableGeoliteUpdate,
 			// Always enable user deletion from IDP in combined server (embedded IdP is always enabled)

infrastructure_files/getting-started-enterprise.sh

@@ -808,8 +808,9 @@ server:
 
   # Trust X-Forwarded-* only from the Traefik container's static address. Both
   # keys must stay in step with the ipv4_address pinned in docker-compose.yml:
-  # trustedPeers decides whether forwarded headers are read at all, and leaving
-  # it unset falls back to 0.0.0.0/0.
+  # trustedPeers decides whether forwarded headers are read at all. Leaving it
+  # unset trusts nothing and records Traefik's own address as every peer's
+  # connection IP.
   reverseProxy:
     trustedPeers:
       - "${TRAEFIK_IP}/32"

infrastructure_files/getting-started.sh

@@ -153,6 +153,7 @@ check_domain_resolves() {
 #   NETBIRD_TRAEFIK_CERTRESOLVER      external-Traefik cert resolver (type 1)
 #   NETBIRD_BIND_LOCALHOST_ONLY       true/false (default true, types 2-5)
 #   NETBIRD_EXTERNAL_PROXY_NETWORK    docker network to join (types 2-4)
+#   NETBIRD_TRUSTED_PEERS             reverse proxy address management sees (default: built-in Traefik's IP, empty for types 1-5)
 #   NETBIRD_NON_INTERACTIVE           true forces unattended mode even with a TTY
 
 # tty_available succeeds only when we may prompt: never when the operator has
@@ -459,6 +460,8 @@ initialize_default_values() {
   MANAGEMENT_HOST_PORT="8081"  # Combined server port (management + signal + relay)
   BIND_LOCALHOST_ONLY="true"
   EXTERNAL_PROXY_NETWORK=""
+  TRUSTED_PEERS=""             # Address the reverse proxy connects to management from
+
 
   # Traefik static IP within the internal bridge network
   TRAEFIK_IP="172.30.0.10"
@@ -519,6 +522,7 @@ apply_agent_network_preset() {
   REVERSE_PROXY_TYPE="0"
   ENABLE_PROXY="true"
   ENABLE_CROWDSEC="false"
+  TRUSTED_PEERS="${NETBIRD_TRUSTED_PEERS:-$TRAEFIK_IP/32}"
 
   TRAEFIK_ACME_EMAIL=$(resolve NETBIRD_LETSENCRYPT_EMAIL required read_traefik_acme_email)
 
@@ -573,6 +577,21 @@ configure_reverse_proxy() {
     4) EXTERNAL_PROXY_NETWORK=$(resolve NETBIRD_EXTERNAL_PROXY_NETWORK "" read_proxy_docker_network "Caddy") ;;
     *) ;; # No network prompt for other options
   esac
+
+  # Only the bundled Traefik has an address we know at render time. External proxies
+  # must supply the address their proxy reaches management from.
+  if [[ "$REVERSE_PROXY_TYPE" == "0" ]]; then
+    TRUSTED_PEERS="${NETBIRD_TRUSTED_PEERS:-$TRAEFIK_IP/32}"
+  else
+    TRUSTED_PEERS="${NETBIRD_TRUSTED_PEERS:-}"
+    if [[ -z "$TRUSTED_PEERS" ]]; then
+      echo "" > /dev/stderr
+      echo "Note: reverseProxy.trustedPeers is unset, so NetBird will use the address your" > /dev/stderr
+      echo "proxy connects from as each peer's connection IP. To record real client IPs," > /dev/stderr
+      echo "set NETBIRD_TRUSTED_PEERS to your proxy's address (e.g. 172.20.0.5/32) and re-run." > /dev/stderr
+      echo "" > /dev/stderr
+    fi
+  fi
   return 0
 }
 
@@ -1033,6 +1052,7 @@ server:
   reverseProxy:
     trustedHTTPProxies:
       - "$TRAEFIK_IP/32"
+$(render_trusted_peers)
 
   store:
     engine: "sqlite"
@@ -1041,6 +1061,12 @@ EOF
   return 0
 }
 
+render_trusted_peers() {
+  if [[ -n "$TRUSTED_PEERS" ]]; then
+    printf '    trustedPeers:\n      - "%s"' "$TRUSTED_PEERS"
+  fi
+}
+
 render_dashboard_env() {
   cat <<EOF
 # Endpoints
@@ -1453,6 +1479,9 @@ location ~ ^/(relay|ws-proxy/) {
 # Native gRPC (signal + management)
 location ~ ^/(signalexchange\.SignalExchange|management\.ManagementService)/ {
     grpc_pass grpc://${server_addr};
+    # Overwrite rather than pass through: without this the client's own
+    # x-forwarded-for metadata reaches NetBird as the connection IP.
+    grpc_set_header X-Forwarded-For \$proxy_add_x_forwarded_for;
     grpc_read_timeout 1d;
     grpc_send_timeout 1d;
     grpc_socket_keepalive on;

infrastructure_files/management.json.tmpl

@@ -34,9 +34,7 @@
     "ReverseProxy": {
         "TrustedHTTPProxies": [],
         "TrustedHTTPProxiesCount": 0,
-        "TrustedPeers": [
-             "0.0.0.0/0"
-        ]
+        "TrustedPeers": []
     },
     "DisableDefaultPolicy": $NETBIRD_MGMT_DISABLE_DEFAULT_POLICY,
     "Datadir": "",

management/cmd/root.go

@@ -54,6 +54,15 @@ func Execute() error {
 	return rootCmd.Execute()
 }
 
+// Customize hands the fully built root command to fn so an embedding binary
+// can extend or adjust the command tree — most commonly attaching its own
+// subcommands next to (or under) the built-in ones — before calling Execute.
+// The root command is constructed in this package's init, so Customize may be
+// called from the embedding binary's main at any point before Execute.
+func Customize(fn func(root *cobra.Command)) {
+	fn(rootCmd)
+}
+
 func init() {
 	mgmtCmd.Flags().IntVar(&mgmtPort, "port", 80, "server port to listen on (defaults to 443 if TLS is enabled, 80 otherwise")
 	mgmtCmd.Flags().BoolVar(&disableLegacyManagementPort, "disable-legacy-port", false, "disabling the old legacy port (33073)")

management/cmd/root_test.go

@@ -0,0 +1,42 @@
+package cmd
+
+import (
+	"testing"
+
+	"github.com/spf13/cobra"
+)
+
+// TestCustomize verifies an embedding binary can extend the command tree: a
+// top-level command attached through the hook, and a subcommand attached under
+// the built-in admin group, are both resolvable exactly as Execute would
+// resolve them.
+func TestCustomize(t *testing.T) {
+	topLevel := &cobra.Command{Use: "some-extra", RunE: func(*cobra.Command, []string) error { return nil }}
+	nested := &cobra.Command{Use: "cluster", RunE: func(*cobra.Command, []string) error { return nil }}
+
+	Customize(func(root *cobra.Command) {
+		root.AddCommand(topLevel)
+		for _, c := range root.Commands() {
+			if c.Name() == "admin" {
+				c.AddCommand(nested)
+				return
+			}
+		}
+		t.Fatal("admin command not found in the root tree")
+	})
+	t.Cleanup(func() {
+		rootCmd.RemoveCommand(topLevel)
+		for _, c := range rootCmd.Commands() {
+			if c.Name() == "admin" {
+				c.RemoveCommand(nested)
+			}
+		}
+	})
+
+	if found, _, err := rootCmd.Find([]string{"some-extra"}); err != nil || found != topLevel {
+		t.Fatalf("top-level command not resolvable: found=%v err=%v", found, err)
+	}
+	if found, _, err := rootCmd.Find([]string{"admin", "cluster"}); err != nil || found != nested {
+		t.Fatalf("nested admin subcommand not resolvable: found=%v err=%v", found, err)
+	}
+}

management/internals/modules/reverseproxy/domain/manager/domain_test.go

@@ -66,8 +66,8 @@ func TestExtractClusterFromFreeDomain(t *testing.T) {
 
 func TestExtractClusterFromCustomDomains(t *testing.T) {
 	customDomains := []*domain.Domain{
-		{Domain: "example.com", TargetCluster: "eu1.proxy.netbird.io"},
-		{Domain: "proxy.corp.io", TargetCluster: "us1.proxy.netbird.io"},
+		{Domain: "example.com", TargetCluster: "eu1.proxy.netbird.io", Validated: true},
+		{Domain: "proxy.corp.io", TargetCluster: "us1.proxy.netbird.io", Validated: true},
 	}
 
 	tests := []struct {
@@ -120,19 +120,49 @@ func TestExtractClusterFromCustomDomains(t *testing.T) {
 
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
-			cluster, ok := extractClusterFromCustomDomains(tc.domain, customDomains)
-			assert.Equal(t, tc.wantOK, ok)
-			if ok {
-				assert.Equal(t, tc.wantVal, cluster)
+			cluster, match := extractClusterFromCustomDomains(tc.domain, customDomains)
+			if !tc.wantOK {
+				assert.Equal(t, customDomainNoMatch, match, "unrelated domain should not match any custom domain")
+				return
 			}
+			assert.Equal(t, customDomainValidated, match, "validated custom domain should resolve a cluster")
+			assert.Equal(t, tc.wantVal, cluster)
 		})
 	}
 }
 
+// An unvalidated row must never yield a cluster: the account has not shown it
+// controls the name, so no service may be bound to it.
+func TestExtractClusterFromCustomDomains_UnvalidatedDomainRefused(t *testing.T) {
+	customDomains := []*domain.Domain{
+		{Domain: "example.com", TargetCluster: "eu1.proxy.netbird.io", Validated: false},
+	}
+
+	for _, serviceDomain := range []string{"example.com", "app.example.com"} {
+		t.Run(serviceDomain, func(t *testing.T) {
+			cluster, match := extractClusterFromCustomDomains(serviceDomain, customDomains)
+			assert.Equal(t, customDomainUnvalidated, match, "unvalidated row must be reported as such")
+			assert.Empty(t, cluster, "unvalidated row must not resolve a cluster")
+		})
+	}
+}
+
+// A more specific unvalidated row must not shadow a validated parent domain.
+func TestExtractClusterFromCustomDomains_ValidatedParentWinsOverUnvalidatedChild(t *testing.T) {
+	customDomains := []*domain.Domain{
+		{Domain: "example.com", TargetCluster: "cluster-generic", Validated: true},
+		{Domain: "app.example.com", TargetCluster: "cluster-app", Validated: false},
+	}
+
+	cluster, match := extractClusterFromCustomDomains("app.example.com", customDomains)
+	assert.Equal(t, customDomainValidated, match)
+	assert.Equal(t, "cluster-generic", cluster, "validated parent domain should provide the cluster")
+}
+
 func TestExtractClusterFromCustomDomains_OverlappingDomains(t *testing.T) {
 	customDomains := []*domain.Domain{
-		{Domain: "example.com", TargetCluster: "cluster-generic"},
-		{Domain: "app.example.com", TargetCluster: "cluster-app"},
+		{Domain: "example.com", TargetCluster: "cluster-generic", Validated: true},
+		{Domain: "app.example.com", TargetCluster: "cluster-app", Validated: true},
 	}
 
 	tests := []struct {
@@ -164,8 +194,8 @@ func TestExtractClusterFromCustomDomains_OverlappingDomains(t *testing.T) {
 
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
-			cluster, ok := extractClusterFromCustomDomains(tc.domain, customDomains)
-			assert.True(t, ok)
+			cluster, match := extractClusterFromCustomDomains(tc.domain, customDomains)
+			assert.Equal(t, customDomainValidated, match)
 			assert.Equal(t, tc.wantVal, cluster)
 		})
 	}

management/internals/modules/reverseproxy/domain/manager/manager.go

@@ -26,6 +26,7 @@ type store interface {
 	GetAgentNetworkSettings(ctx context.Context, lockStrength nbstore.LockingStrength, accountID string) (*agentnetworkTypes.Settings, error)
 
 	GetCustomDomain(ctx context.Context, accountID string, domainID string) (*domain.Domain, error)
+	GetCustomDomainByName(ctx context.Context, domainName string) (*domain.Domain, error)
 	ListFreeDomains(ctx context.Context, accountID string) ([]string, error)
 	ListCustomDomains(ctx context.Context, accountID string) ([]*domain.Domain, error)
 	CreateCustomDomain(ctx context.Context, accountID string, domainName string, targetCluster string, validated bool) (*domain.Domain, error)
@@ -150,6 +151,10 @@ func (m Manager) CreateDomain(ctx context.Context, accountID, userID, domainName
 		return nil, fmt.Errorf("target cluster %s is not available", targetCluster)
 	}
 
+	if err := m.checkDomainAvailable(ctx, domainName); err != nil {
+		return nil, err
+	}
+
 	// Attempt an initial validation against the specified cluster only
 	var validated bool
 	if m.validator.IsValid(ctx, domainName, []string{targetCluster}) {
@@ -166,6 +171,23 @@ func (m Manager) CreateDomain(ctx context.Context, accountID, userID, domainName
 	return d, nil
 }
 
+// checkDomainAvailable reports whether the domain is free to claim. The unique
+// index on the column is the real guard; this turns the violation into a
+// conflict the caller can act on instead of a database error, and says nothing
+// about which account holds the domain.
+func (m Manager) checkDomainAvailable(ctx context.Context, domainName string) error {
+	_, err := m.store.GetCustomDomainByName(ctx, domainName)
+	if err == nil {
+		return status.Errorf(status.AlreadyExists, "domain %s is already registered", domainName)
+	}
+
+	if sErr, ok := status.FromError(err); ok && sErr.Type() == status.NotFound {
+		return nil
+	}
+
+	return fmt.Errorf("look up domain: %w", err)
+}
+
 func (m Manager) DeleteDomain(ctx context.Context, accountID, userID, domainID string) error {
 	ok, ctx, err := m.permissionsManager.ValidateUserPermissions(ctx, accountID, userID, modules.Services, operations.Delete)
 	if err != nil {
@@ -203,7 +225,9 @@ func (m Manager) ValidateDomain(ctx context.Context, accountID, userID, domainID
 		log.WithFields(log.Fields{
 			"accountID": accountID,
 			"domainID":  domainID,
-		}).WithError(err).Error("validate domain")
+			"userID":    userID,
+		}).Error("validate domain: permission denied")
+		return
 	}
 
 	log.WithFields(log.Fields{
@@ -298,9 +322,12 @@ func (m Manager) DeriveClusterFromDomain(ctx context.Context, accountID, domain
 		return "", fmt.Errorf("list custom domains: %w", err)
 	}
 
-	targetCluster, valid := extractClusterFromCustomDomains(domain, customDomains)
-	if valid {
+	targetCluster, match := extractClusterFromCustomDomains(domain, customDomains)
+	switch match {
+	case customDomainValidated:
 		return targetCluster, nil
+	case customDomainUnvalidated:
+		return "", status.Errorf(status.PreconditionFailed, "domain %s is not validated", domain)
 	}
 
 	return "", fmt.Errorf("domain %s does not match any available proxy cluster", domain)
@@ -363,19 +390,46 @@ func (m Manager) reservedGatewayAddress(ctx context.Context, accountID string) (
 	return settings.ProxyAddress, nil
 }
 
-func extractClusterFromCustomDomains(serviceDomain string, customDomains []*domain.Domain) (string, bool) {
+// customDomainMatch describes how a service domain relates to the account's
+// custom domain rows.
+type customDomainMatch int
+
+const (
+	customDomainNoMatch customDomainMatch = iota
+	customDomainUnvalidated
+	customDomainValidated
+)
+
+// extractClusterFromCustomDomains finds the longest custom domain covering the
+// service domain and reports its target cluster. Only a validated row yields a
+// cluster: until the CNAME check has passed the account has not shown it
+// controls the name, so no traffic may be routed for it.
+func extractClusterFromCustomDomains(serviceDomain string, customDomains []*domain.Domain) (string, customDomainMatch) {
 	bestCluster := ""
 	bestLen := -1
+	matched := false
 	for _, cd := range customDomains {
 		if serviceDomain != cd.Domain && !strings.HasSuffix(serviceDomain, "."+cd.Domain) {
 			continue
 		}
+		matched = true
+		if !cd.Validated {
+			continue
+		}
 		if l := len(cd.Domain); l > bestLen {
 			bestLen = l
 			bestCluster = cd.TargetCluster
 		}
 	}
-	return bestCluster, bestLen >= 0
+
+	switch {
+	case bestLen >= 0:
+		return bestCluster, customDomainValidated
+	case matched:
+		return "", customDomainUnvalidated
+	default:
+		return "", customDomainNoMatch
+	}
 }
 
 // ExtractClusterFromFreeDomain extracts the cluster address from a free domain.

management/internals/modules/reverseproxy/domain/manager/manager_realstore_test.go

@@ -0,0 +1,326 @@
+package manager
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"go.opentelemetry.io/otel/metric/noop"
+
+	"github.com/netbirdio/netbird/management/internals/modules/reverseproxy/domain"
+	proxymanager "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/proxy/manager"
+	"github.com/netbirdio/netbird/management/server/activity"
+	"github.com/netbirdio/netbird/management/server/mock_server"
+	"github.com/netbirdio/netbird/management/server/permissions"
+	nbstore "github.com/netbirdio/netbird/management/server/store"
+	"github.com/netbirdio/netbird/management/server/types"
+	"github.com/netbirdio/netbird/shared/management/status"
+)
+
+const (
+	testCluster    = "eu.proxy.test"
+	accountA       = "account-a"
+	accountAUser   = "account-a-admin"
+	accountB       = "account-b"
+	accountBUser   = "account-b-admin"
+	accountAMember = "account-a-member"
+)
+
+// stubResolver answers CNAME lookups from a table the test controls, so a
+// domain can point at the cluster or nowhere without touching a real resolver.
+type stubResolver struct {
+	mu     sync.Mutex
+	cnames map[string]string
+}
+
+func (r *stubResolver) LookupCNAME(_ context.Context, host string) (string, error) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+
+	cname, ok := r.cnames[host]
+	if !ok {
+		return "", fmt.Errorf("lookup %s: no such host", host)
+	}
+	return cname + ".", nil
+}
+
+func (r *stubResolver) set(host, cname string) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	r.cnames[host] = cname
+}
+
+type domainTestEnv struct {
+	manager  Manager
+	store    nbstore.Store
+	resolver *stubResolver
+}
+
+// setupDomainTest builds the domain manager on a real SQLite store with two
+// accounts and one active public proxy cluster.
+func setupDomainTest(t *testing.T) *domainTestEnv {
+	t.Helper()
+
+	ctx := context.Background()
+	testStore, cleanup, err := nbstore.NewTestStoreFromSQL(ctx, "", t.TempDir())
+	require.NoError(t, err)
+	t.Cleanup(cleanup)
+
+	for accountID, userID := range map[string]string{accountA: accountAUser, accountB: accountBUser} {
+		users := map[string]*types.User{
+			userID: {
+				Id:        userID,
+				AccountID: accountID,
+				Role:      types.UserRoleAdmin,
+			},
+		}
+		if accountID == accountA {
+			// A real member of the account whose role denies Services:Create, so
+			// permission denial is exercised as ok=false rather than as a lookup
+			// error for a user who is not in the account at all.
+			users[accountAMember] = &types.User{
+				Id:        accountAMember,
+				AccountID: accountID,
+				Role:      types.UserRoleUser,
+			}
+		}
+
+		require.NoError(t, testStore.SaveAccount(ctx, &types.Account{
+			Id:        accountID,
+			CreatedBy: userID,
+			Settings:  &types.Settings{},
+			Users:     users,
+		}))
+	}
+
+	proxyMgr, err := proxymanager.NewManager(testStore, noop.NewMeterProvider().Meter(""))
+	require.NoError(t, err)
+
+	_, err = proxyMgr.Connect(ctx, "proxy-1", "session-1", testCluster, "127.0.0.1", nil, nil)
+	require.NoError(t, err)
+
+	resolver := &stubResolver{cnames: make(map[string]string)}
+
+	mgr := Manager{
+		store:              testStore,
+		proxyManager:       proxyMgr,
+		validator:          domain.Validator{Resolver: resolver},
+		permissionsManager: permissions.NewManager(testStore),
+		accountManager: &mock_server.MockAccountManager{
+			StoreEventFunc: func(context.Context, string, string, string, activity.ActivityDescriber, map[string]any) {},
+		},
+	}
+
+	return &domainTestEnv{manager: mgr, store: testStore, resolver: resolver}
+}
+
+// storedDomain reads a domain row back through the store so assertions are made
+// on what was persisted rather than on the value the manager returned.
+func storedDomain(t *testing.T, s nbstore.Store, accountID, domainName string) *domain.Domain {
+	t.Helper()
+
+	domains, err := s.ListCustomDomains(context.Background(), accountID)
+	require.NoError(t, err)
+	for _, d := range domains {
+		if d.Domain == domainName {
+			return d
+		}
+	}
+	return nil
+}
+
+// A domain whose CNAME check fails is stored unvalidated and must not resolve a
+// cluster, which is what service creation gates on.
+func TestCreateDomain_FailedLookupIsNotServable(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+
+	created, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "apps.example.com", testCluster)
+	require.NoError(t, err)
+	assert.False(t, created.Validated, "a domain whose CNAME lookup fails must not be created validated")
+
+	stored := storedDomain(t, env.store, accountA, "apps.example.com")
+	require.NotNil(t, stored, "domain row should exist")
+	assert.False(t, stored.Validated, "persisted row must be unvalidated")
+
+	cluster, err := env.manager.DeriveClusterFromDomain(ctx, accountA, "apps.example.com")
+	require.Error(t, err, "an unvalidated domain must not resolve a cluster")
+	assert.Empty(t, cluster)
+	assert.Contains(t, err.Error(), "not validated", "error should tell the caller what to fix")
+
+	sErr, ok := status.FromError(err)
+	require.True(t, ok, "error should be a typed status error")
+	assert.Equal(t, status.PreconditionFailed, sErr.Type())
+
+	_, err = env.manager.DeriveClusterFromDomain(ctx, accountA, "sub.apps.example.com")
+	assert.Error(t, err, "subdomains of an unvalidated custom domain are not servable either")
+}
+
+// A second account claiming a registered domain gets a clean conflict, not a
+// database error surfaced as a 500.
+func TestCreateDomain_DuplicateIsAConflict(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+
+	_, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "shared.example.com", testCluster)
+	require.NoError(t, err)
+
+	_, err = env.manager.CreateDomain(ctx, accountB, accountBUser, "shared.example.com", testCluster)
+	require.Error(t, err)
+
+	sErr, ok := status.FromError(err)
+	require.True(t, ok, "conflict must be a typed status error, not a raw database error")
+	assert.Equal(t, status.AlreadyExists, sErr.Type(), "conflict should map to 409, not 500")
+	assert.NotContains(t, sErr.Message, accountA, "the response must not reveal the holding account")
+
+	assert.Nil(t, storedDomain(t, env.store, accountB, "shared.example.com"), "no row should be written on conflict")
+}
+
+// The same account re-adding one of its own domains is a conflict too.
+func TestCreateDomain_SameAccountDuplicateIsAConflict(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+
+	_, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "dup.example.com", testCluster)
+	require.NoError(t, err)
+
+	_, err = env.manager.CreateDomain(ctx, accountA, accountAUser, "dup.example.com", testCluster)
+	require.Error(t, err)
+
+	sErr, ok := status.FromError(err)
+	require.True(t, ok)
+	assert.Equal(t, status.AlreadyExists, sErr.Type())
+}
+
+// The negative control: a validated domain still derives its cluster, for the
+// bare name and for subdomains, exactly as before.
+func TestCreateDomain_ValidatedDomainDerivesCluster(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+	env.resolver.set("validation.valid.example.com", testCluster)
+
+	created, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "valid.example.com", testCluster)
+	require.NoError(t, err)
+	require.True(t, created.Validated, "a matching CNAME should validate on create")
+
+	cluster, err := env.manager.DeriveClusterFromDomain(ctx, accountA, "valid.example.com")
+	require.NoError(t, err)
+	assert.Equal(t, testCluster, cluster)
+
+	cluster, err = env.manager.DeriveClusterFromDomain(ctx, accountA, "app.valid.example.com")
+	require.NoError(t, err)
+	assert.Equal(t, testCluster, cluster, "subdomains of a validated custom domain resolve too")
+}
+
+// Validating a domain flips the gate: the same lookup that failed before now
+// resolves a cluster.
+func TestValidateDomain_UnlocksClusterDerivation(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+
+	created, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "later.example.com", testCluster)
+	require.NoError(t, err)
+	require.False(t, created.Validated)
+
+	_, err = env.manager.DeriveClusterFromDomain(ctx, accountA, "later.example.com")
+	require.Error(t, err)
+
+	env.resolver.set("validation.later.example.com", testCluster)
+	env.manager.ValidateDomain(ctx, accountA, accountAUser, created.ID)
+
+	require.True(t, storedDomain(t, env.store, accountA, "later.example.com").Validated)
+
+	cluster, err := env.manager.DeriveClusterFromDomain(ctx, accountA, "later.example.com")
+	require.NoError(t, err)
+	assert.Equal(t, testCluster, cluster)
+}
+
+// Free cluster domains are unaffected by the custom domain gate.
+func TestDeriveClusterFromDomain_FreeDomainUnaffected(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+
+	cluster, err := env.manager.DeriveClusterFromDomain(ctx, accountA, "myapp.abc123."+testCluster)
+	require.NoError(t, err)
+	assert.Equal(t, testCluster, cluster)
+}
+
+// The manager pre-check exists to turn a conflict into a 409, but the unique
+// index on the column is what actually guarantees the domain is claimed once.
+//
+// Two requests can clear the pre-check concurrently and race to the insert.
+// Inserting twice through the store reaches the same code path the loser of
+// that race takes, without the nondeterminism of driving it from goroutines,
+// and the loser must still see a conflict rather than an internal error.
+func TestStore_DuplicateDomainRejectedByIndexAsConflict(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+
+	_, err := env.store.CreateCustomDomain(ctx, accountA, "indexed.example.com", testCluster, false)
+	require.NoError(t, err)
+
+	_, err = env.store.CreateCustomDomain(ctx, accountB, "indexed.example.com", testCluster, false)
+	require.Error(t, err, "the unique index must reject the same domain in a second account")
+
+	sErr, ok := status.FromError(err)
+	require.True(t, ok, "the losing insert must return a typed status error")
+	assert.Equal(t, status.AlreadyExists, sErr.Type(), "a lost race is a 409, not a 500")
+}
+
+// Validation is what decides whether a domain routes traffic, so a caller
+// without permission to it must not be able to flip the flag. The check logged
+// the denial and then carried on, which was inert while nothing read Validated
+// and is not once cluster derivation gates on it.
+func TestValidateDomain_PermissionDeniedDoesNotValidate(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+
+	created, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "guarded.example.com", testCluster)
+	require.NoError(t, err)
+	require.False(t, created.Validated)
+
+	// The CNAME is in place, so the only thing standing between this caller and
+	// a validated domain is the permission check.
+	env.resolver.set("validation.guarded.example.com", testCluster)
+
+	env.manager.ValidateDomain(ctx, accountA, accountAMember, created.ID)
+
+	stored := storedDomain(t, env.store, accountA, "guarded.example.com")
+	require.NotNil(t, stored)
+	assert.False(t, stored.Validated, "a caller without permission must not validate the domain")
+
+	_, err = env.manager.DeriveClusterFromDomain(ctx, accountA, "guarded.example.com")
+	assert.Error(t, err, "the domain must still be unservable")
+}
+
+// Validation runs asynchronously, so it can finish after the domain was
+// deleted and then write a stale row back. gorm's Save falls back to an insert
+// when an update affects no rows, which would resurrect the domain as
+// validated; UpdateCustomDomain avoids that by selecting explicit columns.
+// This pins that behaviour, since dropping the Select would reintroduce it.
+func TestUpdateCustomDomain_DoesNotResurrectDeletedDomain(t *testing.T) {
+	ctx := context.Background()
+	env := setupDomainTest(t)
+
+	created, err := env.manager.CreateDomain(ctx, accountA, accountAUser, "racy.example.com", testCluster)
+	require.NoError(t, err)
+
+	stale := storedDomain(t, env.store, accountA, "racy.example.com")
+	require.NotNil(t, stale)
+
+	require.NoError(t, env.manager.DeleteDomain(ctx, accountA, accountAUser, created.ID))
+	require.Nil(t, storedDomain(t, env.store, accountA, "racy.example.com"), "the domain should be gone")
+
+	// What an in-flight validation would write once its CNAME check succeeded.
+	// The write has to succeed for the assertion below to mean anything: a
+	// rejected write would leave the domain absent for the wrong reason.
+	stale.Validated = true
+	_, err = env.store.UpdateCustomDomain(ctx, accountA, stale)
+	require.NoError(t, err, "the update itself must succeed, so absence is not just a failed write")
+
+	assert.Nil(t, storedDomain(t, env.store, accountA, "racy.example.com"),
+		"a late validation write must not recreate a deleted domain")
+}

management/internals/modules/reverseproxy/domain/manager/manager_test.go

@@ -184,6 +184,10 @@ func (s *stubStore) GetCustomDomain(context.Context, string, string) (*domain.Do
 	panic("not used in allow-list tests")
 }
 
+func (s *stubStore) GetCustomDomainByName(context.Context, string) (*domain.Domain, error) {
+	panic("not used in allow-list tests")
+}
+
 func (s *stubStore) ListFreeDomains(context.Context, string) ([]string, error) {
 	panic("not used in allow-list tests")
 }

management/internals/modules/reverseproxy/service/manager/domain_validation_test.go

@@ -0,0 +1,127 @@
+package manager
+
+import (
+	"context"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"go.opentelemetry.io/otel/metric/noop"
+
+	domainmanager "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/domain/manager"
+	proxymanager "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/proxy/manager"
+	rpservice "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
+	"github.com/netbirdio/netbird/management/server/activity"
+	"github.com/netbirdio/netbird/management/server/mock_server"
+	"github.com/netbirdio/netbird/management/server/permissions"
+	"github.com/netbirdio/netbird/management/server/store"
+	"github.com/netbirdio/netbird/shared/management/status"
+)
+
+const validationTestCluster = "eu.proxy.test"
+
+// withRealDomainManager swaps the stub cluster deriver for the real domain
+// manager backed by the same store, so service creation is gated by the actual
+// domain rows rather than by a test double that always agrees.
+func withRealDomainManager(t *testing.T, mgr *Manager, testStore store.Store) {
+	t.Helper()
+
+	ctx := context.Background()
+	proxyMgr, err := proxymanager.NewManager(testStore, noop.NewMeterProvider().Meter(""))
+	require.NoError(t, err)
+
+	_, err = proxyMgr.Connect(ctx, "proxy-1", "session-1", validationTestCluster, "127.0.0.1", nil, nil)
+	require.NoError(t, err)
+
+	accountMgr := &mock_server.MockAccountManager{
+		StoreEventFunc: func(context.Context, string, string, string, activity.ActivityDescriber, map[string]any) {},
+	}
+	mgr.clusterDeriver = domainmanager.NewManager(testStore, proxyMgr, permissions.NewManager(testStore), accountMgr)
+}
+
+func newTestService(domain string) *rpservice.Service {
+	return &rpservice.Service{
+		Name:    "test-service",
+		Domain:  domain,
+		Enabled: true,
+		Mode:    rpservice.ModeHTTP,
+		Targets: []*rpservice.Target{{
+			Host:       "10.0.0.1",
+			Port:       8080,
+			Protocol:   "http",
+			TargetId:   testPeerID,
+			TargetType: "peer",
+			Enabled:    true,
+		}},
+	}
+}
+
+// A service must not bind to a domain the account has not validated, and
+// nothing may be persisted for the attempt.
+func TestCreateService_RefusesUnvalidatedDomain(t *testing.T) {
+	ctx := context.Background()
+	mgr, testStore := setupIntegrationTest(t)
+	withRealDomainManager(t, mgr, testStore)
+
+	_, err := testStore.CreateCustomDomain(ctx, testAccountID, "unproven.example.com", validationTestCluster, false)
+	require.NoError(t, err)
+
+	_, err = mgr.CreateService(ctx, testAccountID, testUserID, newTestService("unproven.example.com"))
+	require.Error(t, err, "an unvalidated domain must not bind a service")
+	assert.Contains(t, err.Error(), "not validated", "the API error should name the actual problem")
+
+	sErr, ok := status.FromError(err)
+	require.True(t, ok, "error should be a typed status error")
+	assert.Equal(t, status.PreconditionFailed, sErr.Type())
+
+	services, err := testStore.GetAccountServices(ctx, store.LockingStrengthNone, testAccountID)
+	require.NoError(t, err)
+	assert.Empty(t, services, "no service row should be written for a refused domain")
+}
+
+// The negative control: a validated domain still binds a service and derives
+// its cluster exactly as before.
+func TestCreateService_ValidatedDomainBindsService(t *testing.T) {
+	ctx := context.Background()
+	mgr, testStore := setupIntegrationTest(t)
+	withRealDomainManager(t, mgr, testStore)
+
+	_, err := testStore.CreateCustomDomain(ctx, testAccountID, "proven.example.com", validationTestCluster, true)
+	require.NoError(t, err)
+
+	created, err := mgr.CreateService(ctx, testAccountID, testUserID, newTestService("app.proven.example.com"))
+	require.NoError(t, err)
+	assert.Equal(t, validationTestCluster, created.ProxyCluster, "service should bind to the domain's target cluster")
+
+	services, err := testStore.GetAccountServices(ctx, store.LockingStrengthNone, testAccountID)
+	require.NoError(t, err)
+	require.Len(t, services, 1, "the service should be persisted")
+	assert.Equal(t, "app.proven.example.com", services[0].Domain)
+}
+
+// An update must not be a way around the creation gate: moving a live service
+// onto an unvalidated domain has to fail rather than silently keep the old
+// cluster and start serving the new hostname.
+func TestUpdateService_RefusesMoveToUnvalidatedDomain(t *testing.T) {
+	ctx := context.Background()
+	mgr, testStore := setupIntegrationTest(t)
+	withRealDomainManager(t, mgr, testStore)
+
+	_, err := testStore.CreateCustomDomain(ctx, testAccountID, "proven.example.com", validationTestCluster, true)
+	require.NoError(t, err)
+	_, err = testStore.CreateCustomDomain(ctx, testAccountID, "unproven.example.com", validationTestCluster, false)
+	require.NoError(t, err)
+
+	created, err := mgr.CreateService(ctx, testAccountID, testUserID, newTestService("app.proven.example.com"))
+	require.NoError(t, err)
+
+	moved := *created
+	moved.Domain = "app.unproven.example.com"
+	_, err = mgr.UpdateService(ctx, testAccountID, testUserID, &moved)
+	require.Error(t, err, "moving to an unvalidated domain must fail")
+	assert.Contains(t, err.Error(), "not validated")
+
+	stored, err := testStore.GetServiceByID(ctx, store.LockingStrengthNone, testAccountID, created.ID)
+	require.NoError(t, err)
+	assert.Equal(t, "app.proven.example.com", stored.Domain, "the service must keep its original domain")
+}

management/internals/modules/reverseproxy/service/manager/manager.go

@@ -606,16 +606,19 @@ func (m *Manager) resolveEffectiveCluster(ctx context.Context, accountID string,
 		return existing.ProxyCluster, nil
 	}
 
-	if m.clusterDeriver != nil {
-		derived, err := m.clusterDeriver.DeriveClusterFromDomain(ctx, accountID, svc.Domain)
-		if err != nil {
-			log.WithError(err).Warnf("could not derive cluster from domain %s", svc.Domain)
-		} else {
-			return derived, nil
-		}
+	if m.clusterDeriver == nil {
+		return existing.ProxyCluster, nil
 	}
 
-	return existing.ProxyCluster, nil
+	// Falling back to the old cluster here would let an update move a service
+	// onto a domain the account has not validated, bypassing the check that
+	// creation makes.
+	derived, err := m.clusterDeriver.DeriveClusterFromDomain(ctx, accountID, svc.Domain)
+	if err != nil {
+		return "", status.Errorf(status.PreconditionFailed, "could not derive cluster from domain %s: %v", svc.Domain, err)
+	}
+
+	return derived, nil
 }
 
 func (m *Manager) executeServiceUpdate(ctx context.Context, transaction store.Store, accountID string, service *service.Service, updateInfo *serviceUpdateInfo, customPorts *bool, effectiveCluster string) error {

management/internals/server/boot.go

@@ -31,6 +31,7 @@ import (
 	rpservice "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
 	networkmapdb "github.com/netbirdio/netbird/management/internals/network_map_db"
 	networkmapdbfactory "github.com/netbirdio/netbird/management/internals/network_map_db/factory"
+	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
 	nbgrpc "github.com/netbirdio/netbird/management/internals/shared/grpc"
 	"github.com/netbirdio/netbird/management/server/activity"
 	activitystore "github.com/netbirdio/netbird/management/server/activity/store"
@@ -111,7 +112,8 @@ func (s *BaseServer) NetworkMapStore() *networkmapdb.NetworkMapDBStoreImpl {
 			s.Config.StoreConfig.Engine,
 			s.Config.Datadir,
 			s.IntegratedValidator(),
-			s.SettingsManager())
+			s.SettingsManager(),
+		)
 		// networkmap db store supports postgres and sqlite backends only
 		// for other backends a fallback is used, so NotSupportedStoreEngineError
 		// is not a fatal error
@@ -180,24 +182,7 @@ func (s *BaseServer) RateLimiter() *middleware.APIRateLimiter {
 
 func (s *BaseServer) GRPCServer() *grpc.Server {
 	return Create(s, func() *grpc.Server {
-		trustedPeers := s.Config.ReverseProxy.TrustedPeers
-		defaultTrustedPeers := []netip.Prefix{netip.MustParsePrefix("0.0.0.0/0"), netip.MustParsePrefix("::/0")}
-		if len(trustedPeers) == 0 || slices.Equal[[]netip.Prefix](trustedPeers, defaultTrustedPeers) {
-			log.WithContext(context.Background()).Warn("TrustedPeers are configured to default value '0.0.0.0/0', '::/0'. This allows connection IP spoofing.")
-			trustedPeers = defaultTrustedPeers
-		}
-		trustedHTTPProxies := s.Config.ReverseProxy.TrustedHTTPProxies
-		trustedProxiesCount := s.Config.ReverseProxy.TrustedHTTPProxiesCount
-		if len(trustedHTTPProxies) > 0 && trustedProxiesCount > 0 {
-			log.WithContext(context.Background()).Warn("TrustedHTTPProxies and TrustedHTTPProxiesCount both are configured. " +
-				"This is not recommended way to extract X-Forwarded-For. Consider using one of these options.")
-		}
-		realipOpts := []realip.Option{
-			realip.WithTrustedPeers(trustedPeers),
-			realip.WithTrustedProxies(trustedHTTPProxies),
-			realip.WithTrustedProxiesCount(trustedProxiesCount),
-			realip.WithHeaders([]string{realip.XForwardedFor, realip.XRealIp}),
-		}
+		realipOpts := realIPOptions(s.Config.ReverseProxy)
 		proxyUnary, proxyStream, proxyAuthClose := nbgrpc.NewProxyAuthInterceptors(s.Store())
 		s.proxyAuthClose = proxyAuthClose
 		gRPCOpts := []grpc.ServerOption{
@@ -333,7 +318,7 @@ func (s *BaseServer) AccessLogsManager() accesslogs.Manager {
 	})
 }
 
-func loadTLSConfig(certFile string, certKey string) (*tls.Config, error) {
+func loadTLSConfig(certFile, certKey string) (*tls.Config, error) {
 	// Load server's certificate and private key
 	serverCert, err := tls.LoadX509KeyPair(certFile, certKey)
 	if err != nil {
@@ -380,3 +365,34 @@ func streamInterceptor(
 	wrapped.WrappedContext = context.WithValue(ctx, nbContext.RequestIDKey, reqID)
 	return handler(srv, wrapped)
 }
+
+// realIPOptions builds the real-IP middleware options from the reverse proxy config.
+//
+// TrustedPeers controls which transport peers are allowed to supply forwarded-IP
+// headers. If empty, forwarded headers are ignored and the transport peer address
+// is used directly. Operators terminating connections at a reverse proxy should
+// configure TrustedPeers with that proxy's address or network.
+//
+// Only X-Forwarded-For is trusted. X-Real-IP contains a single client-supplied
+// address with no proxy chain to validate, and none of the reverse proxies we ship
+// use it on the gRPC path.
+func realIPOptions(cfg nbconfig.ReverseProxy) []realip.Option {
+	if idx := slices.IndexFunc(cfg.TrustedPeers, func(p netip.Prefix) bool { return p.Bits() == 0 }); idx >= 0 {
+		log.WithContext(context.Background()).Warnf("TrustedPeers contains the default route %s, which trusts "+
+			"X-Forwarded-For from every client and allows connection IP spoofing. Set TrustedPeers to the address "+
+			"of your reverse proxy, or leave it empty to use the connection's source address.", cfg.TrustedPeers[idx])
+	}
+	if cfg.TrustedHTTPProxiesCount > 0 {
+		log.WithContext(context.Background()).Warn(
+			"TrustedHTTPProxiesCount skips X-Forwarded-For entries by position before TrustedHTTPProxies filters by address. " +
+				"An incorrect count may skip the real client IP and produce an incorrect source address.",
+		)
+	}
+
+	return []realip.Option{
+		realip.WithTrustedPeers(cfg.TrustedPeers),
+		realip.WithTrustedProxies(cfg.TrustedHTTPProxies),
+		realip.WithTrustedProxiesCount(cfg.TrustedHTTPProxiesCount),
+		realip.WithHeaders([]string{realip.XForwardedFor}),
+	}
+}

management/internals/server/realip_test.go

@@ -0,0 +1,171 @@
+package server
+
+import (
+	"context"
+	"io"
+	"net"
+	"net/netip"
+	"testing"
+	"time"
+
+	"github.com/grpc-ecosystem/go-grpc-middleware/v2/interceptors/realip"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"google.golang.org/grpc"
+	"google.golang.org/grpc/credentials/insecure"
+	"google.golang.org/grpc/metadata"
+	"google.golang.org/protobuf/types/known/emptypb"
+
+	nbconfig "github.com/netbirdio/netbird/management/internals/server/config"
+)
+
+const (
+	realIPProbeMethod       = "/netbird.test.RealIPProbe/Probe"
+	realIPProbeStreamMethod = "/netbird.test.RealIPProbe/ProbeStream"
+)
+
+// realIPProbe records the real IP the middleware derived for each call.
+type realIPProbe struct {
+	got chan string
+}
+
+func (p *realIPProbe) record(ctx context.Context) {
+	addr, _ := realip.FromContext(ctx)
+	p.got <- addr.String()
+}
+
+func (p *realIPProbe) wait(t *testing.T) string {
+	t.Helper()
+
+	select {
+	case got := <-p.got:
+		return got
+	case <-time.After(5 * time.Second):
+		t.Fatal("timed out waiting for probe")
+		return ""
+	}
+}
+
+func startProbeServer(t *testing.T, cfg nbconfig.ReverseProxy) (*grpc.ClientConn, *realIPProbe) {
+	t.Helper()
+
+	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+
+	probe := &realIPProbe{got: make(chan string, 1)}
+	opts := realIPOptions(cfg)
+	srv := grpc.NewServer(
+		grpc.ChainUnaryInterceptor(realip.UnaryServerInterceptorOpts(opts...)),
+		grpc.ChainStreamInterceptor(realip.StreamServerInterceptorOpts(opts...)),
+	)
+	srv.RegisterService(&grpc.ServiceDesc{
+		ServiceName: "netbird.test.RealIPProbe",
+		HandlerType: (*any)(nil),
+		Methods: []grpc.MethodDesc{{
+			MethodName: "Probe",
+			Handler: func(_ any, ctx context.Context, dec func(any) error, interceptor grpc.UnaryServerInterceptor) (any, error) {
+				req := new(emptypb.Empty)
+				if err := dec(req); err != nil {
+					return nil, err
+				}
+				handler := func(ctx context.Context, _ any) (any, error) {
+					probe.record(ctx)
+					return &emptypb.Empty{}, nil
+				}
+				if interceptor == nil {
+					return handler(ctx, req)
+				}
+				return interceptor(ctx, req, &grpc.UnaryServerInfo{FullMethod: realIPProbeMethod}, handler)
+			},
+		}},
+		Streams: []grpc.StreamDesc{{
+			StreamName:    "ProbeStream",
+			ServerStreams: true,
+			Handler: func(_ any, stream grpc.ServerStream) error {
+				probe.record(stream.Context())
+				return nil
+			},
+		}},
+	}, probe)
+
+	go func() { _ = srv.Serve(listener) }()
+	t.Cleanup(srv.Stop)
+
+	conn, err := grpc.NewClient(listener.Addr().String(), grpc.WithTransportCredentials(insecure.NewCredentials()))
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = conn.Close() })
+
+	return conn, probe
+}
+
+func callUnary(t *testing.T, conn *grpc.ClientConn, probe *realIPProbe, kv ...string) string {
+	t.Helper()
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	ctx = metadata.AppendToOutgoingContext(ctx, kv...)
+	require.NoError(t, conn.Invoke(ctx, realIPProbeMethod, &emptypb.Empty{}, &emptypb.Empty{}))
+
+	return probe.wait(t)
+}
+
+func callStream(t *testing.T, conn *grpc.ClientConn, probe *realIPProbe, kv ...string) string {
+	t.Helper()
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	ctx = metadata.AppendToOutgoingContext(ctx, kv...)
+	desc := &grpc.StreamDesc{StreamName: "ProbeStream", ServerStreams: true}
+	stream, err := conn.NewStream(ctx, desc, realIPProbeStreamMethod)
+	require.NoError(t, err)
+	require.NoError(t, stream.CloseSend())
+	require.ErrorIs(t, stream.RecvMsg(&emptypb.Empty{}), io.EOF)
+
+	return probe.wait(t)
+}
+
+func assertRealIP(t *testing.T, cfg nbconfig.ReverseProxy, want string, kv ...string) {
+	t.Helper()
+
+	conn, probe := startProbeServer(t, cfg)
+	t.Run("unary", func(t *testing.T) {
+		assert.Equal(t, want, callUnary(t, conn, probe, kv...))
+	})
+	t.Run("stream", func(t *testing.T) {
+		assert.Equal(t, want, callStream(t, conn, probe, kv...))
+	})
+}
+
+func TestRealIPDefaultIgnoresClientForwardedHeaders(t *testing.T) {
+	assertRealIP(t, nbconfig.ReverseProxy{}, "127.0.0.1",
+		realip.XForwardedFor, "203.0.113.44",
+		realip.XRealIp, "203.0.113.44",
+	)
+}
+
+func TestRealIPUntrustedPeerIgnoresForwardedHeaders(t *testing.T) {
+	cfg := nbconfig.ReverseProxy{TrustedPeers: []netip.Prefix{netip.MustParsePrefix("10.9.8.7/32")}}
+
+	assertRealIP(t, cfg, "127.0.0.1",
+		realip.XForwardedFor, "203.0.113.44",
+		realip.XRealIp, "203.0.113.44",
+	)
+}
+
+func TestRealIPTrustedPeerHonoursForwardedHeaders(t *testing.T) {
+	cfg := nbconfig.ReverseProxy{TrustedPeers: []netip.Prefix{netip.MustParsePrefix("127.0.0.1/32")}}
+
+	assertRealIP(t, cfg, "203.0.113.44",
+		realip.XForwardedFor, "203.0.113.44",
+		realip.XRealIp, "203.0.113.44",
+	)
+}
+
+func TestRealIPIgnoresXRealIPWhenProxyCountIsSet(t *testing.T) {
+	cfg := nbconfig.ReverseProxy{
+		TrustedPeers:            []netip.Prefix{netip.MustParsePrefix("127.0.0.1/32")},
+		TrustedHTTPProxiesCount: 1,
+	}
+
+	assertRealIP(t, cfg, "127.0.0.1", realip.XRealIp, "203.0.113.44")
+}

management/internals/server/server.go

@@ -74,6 +74,7 @@ type BaseServer struct {
 	grpcExtensions []GRPCExtension
 
 	listener    net.Listener
+	tlsConfig   *tls.Config
 	certManager *autocert.Manager
 	update      *version.Update
 
@@ -94,6 +95,7 @@ type Config struct {
 	DisableGeoliteUpdate        bool
 	UserDeleteFromIDPEnabled    bool
 	AutoResolveDomains          bool
+	TLSConfig                   *tls.Config
 }
 
 // NewServer initializes and configures a new Server instance
@@ -110,6 +112,7 @@ func NewServer(cfg *Config) *BaseServer {
 		disableLegacyManagementPort: cfg.DisableLegacyManagementPort,
 		mgmtMetricsPort:             cfg.MgmtMetricsPort,
 		autoResolveDomains:          cfg.AutoResolveDomains,
+		tlsConfig:                   cfg.TLSConfig,
 	}
 	s.container[ContainerKeyBaseServer] = s
 
@@ -139,21 +142,9 @@ func (s *BaseServer) Start(ctx context.Context) error {
 	}
 	s.EphemeralManager().LoadInitialPeers(srvCtx)
 
-	var tlsConfig *tls.Config
-	tlsEnabled := false
-	if s.Config.HttpConfig.LetsEncryptDomain != "" {
-		s.certManager, err = encryption.CreateCertManager(s.Config.Datadir, s.Config.HttpConfig.LetsEncryptDomain)
-		if err != nil {
-			return fmt.Errorf("failed creating LetsEncrypt cert manager: %v", err)
-		}
-		tlsEnabled = true
-	} else if s.Config.HttpConfig.CertFile != "" && s.Config.HttpConfig.CertKey != "" {
-		tlsConfig, err = loadTLSConfig(s.Config.HttpConfig.CertFile, s.Config.HttpConfig.CertKey)
-		if err != nil {
-			log.WithContext(srvCtx).Errorf("cannot load TLS credentials: %v", err)
-			return err
-		}
-		tlsEnabled = true
+	tlsEnabled, err := s.setupTLS(srvCtx)
+	if err != nil {
+		return err
 	}
 
 	installationID, err := getInstallationID(srvCtx, s.Store())
@@ -215,8 +206,8 @@ func (s *BaseServer) Start(ctx context.Context) error {
 			log.WithContext(ctx).Infof("running HTTP server (LetsEncrypt challenge handler): %s", cml.Addr().String())
 			s.serveHTTP(ctx, cml, s.certManager.HTTPHandler(nil))
 		}
-	case tlsConfig != nil:
-		s.listener, err = tls.Listen("tcp", fmt.Sprintf(":%d", s.mgmtPort), tlsConfig)
+	case s.tlsConfig != nil:
+		s.listener, err = tls.Listen("tcp", fmt.Sprintf(":%d", s.mgmtPort), s.tlsConfig)
 		if err != nil {
 			return fmt.Errorf("failed creating TLS listener on port %d: %v", s.mgmtPort, err)
 		}
@@ -240,6 +231,31 @@ func (s *BaseServer) Start(ctx context.Context) error {
 	return nil
 }
 
+// setupTLS resolves the listener's TLS source: an injected config wins over the HttpConfig certificate settings
+func (s *BaseServer) setupTLS(ctx context.Context) (bool, error) {
+	switch {
+	case s.tlsConfig != nil:
+		return true, nil
+	case s.Config.HttpConfig.LetsEncryptDomain != "":
+		certManager, err := encryption.CreateCertManager(s.Config.Datadir, s.Config.HttpConfig.LetsEncryptDomain)
+		if err != nil {
+			return false, fmt.Errorf("failed creating LetsEncrypt cert manager: %v", err)
+		}
+		s.certManager = certManager
+		return true, nil
+	case s.Config.HttpConfig.CertFile != "" && s.Config.HttpConfig.CertKey != "":
+		tlsConfig, err := loadTLSConfig(s.Config.HttpConfig.CertFile, s.Config.HttpConfig.CertKey)
+		if err != nil {
+			log.WithContext(ctx).Errorf("cannot load TLS credentials: %v", err)
+			return false, err
+		}
+		s.tlsConfig = tlsConfig
+		return true, nil
+	default:
+		return false, nil
+	}
+}
+
 // Stop attempts a graceful shutdown, waiting up to 5 seconds for active connections to finish
 func (s *BaseServer) Stop() error {
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)

management/server/account.go

@@ -719,8 +719,10 @@ func (am *DefaultAccountManager) schedulePeerLoginExpiration(ctx context.Context
 		log.WithContext(ctx).Tracef("peer login expiration job for account %s is already scheduled", accountID)
 		return
 	}
+	// The job outlives the request that arms it, so it must not inherit the request's cancellation.
+	jobCtx := context.WithoutCancel(ctx)
 	if nextRun, ok := am.getNextPeerExpiration(ctx, accountID); ok {
-		go am.peerLoginExpiry.Schedule(ctx, nextRun, accountID, am.peerLoginExpirationJob(ctx, accountID))
+		go am.peerLoginExpiry.Schedule(jobCtx, nextRun, accountID, am.peerLoginExpirationJob(jobCtx, accountID))
 	}
 }
 
@@ -752,8 +754,9 @@ func (am *DefaultAccountManager) peerInactivityExpirationJob(ctx context.Context
 // checkAndSchedulePeerInactivityExpiration periodically checks for inactive peers to end their sessions
 func (am *DefaultAccountManager) checkAndSchedulePeerInactivityExpiration(ctx context.Context, accountID string) {
 	am.peerInactivityExpiry.Cancel(ctx, []string{accountID})
+	jobCtx := context.WithoutCancel(ctx)
 	if nextRun, ok := am.getNextInactivePeerExpiration(ctx, accountID); ok {
-		go am.peerInactivityExpiry.Schedule(ctx, nextRun, accountID, am.peerInactivityExpirationJob(ctx, accountID))
+		go am.peerInactivityExpiry.Schedule(jobCtx, nextRun, accountID, am.peerInactivityExpirationJob(jobCtx, accountID))
 	}
 }
 

management/server/account_test.go

@@ -1920,6 +1920,154 @@ func TestDefaultAccountManager_MarkPeerConnected_PeerLoginExpiration(t *testing.
 	}
 }
 
+func TestDefaultAccountManager_SchedulePeerLoginExpiration_IncludesOfflinePeers(t *testing.T) {
+	manager, updateManager, err := createManager(t)
+	require.NoError(t, err, "unable to create account manager")
+
+	accountID, err := manager.GetAccountIDByUserID(context.Background(), auth.UserAuth{UserId: userID})
+	require.NoError(t, err, "unable to create an account")
+
+	connectedKey, offlineKey := addExpiringPeers(t, manager)
+	_, err = manager.UpdateAccountSettings(context.Background(), accountID, userID, &types.Settings{
+		PeerLoginExpiration:        time.Hour,
+		PeerLoginExpirationEnabled: true,
+		Extra:                      &types.ExtraSettings{},
+	})
+	require.NoError(t, err, "expecting to update account settings successfully but got error")
+	manager.peerLoginExpiry.CancelAll(context.Background())
+
+	// The connected peer logged in just now, so a job computed from connected peers alone
+	// would be armed for an hour. The offline peer's login expires in two seconds; a
+	// reconnect of that peer must not have to wait for the connected peer's tick.
+	now := time.Now().UTC()
+	setPeerLogin(t, manager, accountID, connectedKey, true, now)
+	setPeerLogin(t, manager, accountID, offlineKey, false, now.Add(-time.Hour+2*time.Second))
+
+	offlinePeer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, offlineKey)
+	require.NoError(t, err)
+	updateManager.CreateChannel(context.Background(), offlinePeer.ID)
+
+	manager.peerLoginExpiry = NewDefaultScheduler()
+	t.Cleanup(func() { manager.peerLoginExpiry.CancelAll(context.Background()) })
+	manager.schedulePeerLoginExpiration(context.Background(), accountID)
+
+	// The flag is committed per peer before the disconnect fans out, so wait for both.
+	require.Eventually(t, func() bool {
+		peer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, offlineKey)
+		return err == nil && peer.Status.LoginExpired && !updateManager.HasChannel(offlinePeer.ID)
+	}, 10*time.Second, 100*time.Millisecond, "offline peer should be expired and disconnected at its own deadline")
+
+	connectedPeer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, connectedKey)
+	require.NoError(t, err)
+	assert.False(t, connectedPeer.Status.LoginExpired, "connected peer with a fresh login must not expire")
+}
+
+func TestDefaultAccountManager_SchedulePeerLoginExpiration_DetachesRequestContext(t *testing.T) {
+	manager, _, err := createManager(t)
+	require.NoError(t, err, "unable to create account manager")
+
+	accountID, err := manager.GetAccountIDByUserID(context.Background(), auth.UserAuth{UserId: userID})
+	require.NoError(t, err, "unable to create an account")
+	connectedKey, _ := addExpiringPeers(t, manager)
+	setPeerLogin(t, manager, accountID, connectedKey, true, time.Now().UTC())
+
+	scheduled := make(chan context.Context, 1)
+	manager.peerLoginExpiry = &MockScheduler{
+		IsSchedulerRunningFunc: func(string) bool { return false },
+		ScheduleFunc: func(ctx context.Context, _ time.Duration, _ string, _ func() (time.Duration, bool)) {
+			scheduled <- ctx
+		},
+	}
+
+	requestCtx, cancel := context.WithCancel(context.Background())
+	manager.schedulePeerLoginExpiration(requestCtx, accountID)
+	cancel()
+
+	select {
+	case jobCtx := <-scheduled:
+		assert.NoError(t, jobCtx.Err(), "the expiration job must outlive the request that armed it")
+	case <-time.After(time.Second):
+		t.Fatal("timeout while waiting for the job to be scheduled")
+	}
+}
+
+func TestDefaultAccountManager_ExpireAndUpdatePeers_SkipsPeerThatLoggedInAgain(t *testing.T) {
+	manager, updateManager, err := createManager(t)
+	require.NoError(t, err, "unable to create account manager")
+
+	accountID, err := manager.GetAccountIDByUserID(context.Background(), auth.UserAuth{UserId: userID})
+	require.NoError(t, err, "unable to create an account")
+
+	reloggedKey, staleKey := addExpiringPeers(t, manager)
+	_, err = manager.UpdateAccountSettings(context.Background(), accountID, userID, &types.Settings{
+		PeerLoginExpiration:        time.Hour,
+		PeerLoginExpirationEnabled: true,
+		Extra:                      &types.ExtraSettings{},
+	})
+	require.NoError(t, err, "expecting to update account settings successfully but got error")
+	manager.peerLoginExpiry.CancelAll(context.Background())
+
+	expiredLogin := time.Now().UTC().Add(-2 * time.Hour)
+	setPeerLogin(t, manager, accountID, reloggedKey, true, expiredLogin)
+	setPeerLogin(t, manager, accountID, staleKey, true, expiredLogin)
+
+	expiredPeers, err := manager.getExpiredPeers(context.Background(), accountID)
+	require.NoError(t, err)
+	require.Len(t, expiredPeers, 2, "both peers should be due for expiration")
+
+	// The job holds the candidate list while one peer completes a fresh login, which
+	// moves its deadline into the future and must win over the stale candidate entry.
+	setPeerLogin(t, manager, accountID, reloggedKey, true, time.Now().UTC())
+
+	reloggedPeer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, reloggedKey)
+	require.NoError(t, err)
+	stalePeer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, staleKey)
+	require.NoError(t, err)
+	updateManager.CreateChannel(context.Background(), reloggedPeer.ID)
+	updateManager.CreateChannel(context.Background(), stalePeer.ID)
+
+	err = manager.expireAndUpdatePeers(context.Background(), accountID, expiredPeers, peerExpirationSessionExpired)
+	require.NoError(t, err)
+
+	reloggedPeer, err = manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, reloggedKey)
+	require.NoError(t, err)
+	assert.False(t, reloggedPeer.Status.LoginExpired, "a peer that logged in again must not be flagged from the stale candidate list")
+	assert.True(t, reloggedPeer.Status.Connected, "the re-logged peer must keep its connected status")
+	assert.True(t, updateManager.HasChannel(reloggedPeer.ID), "the re-logged peer's update channel must stay open")
+
+	stalePeer, err = manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, staleKey)
+	require.NoError(t, err)
+	assert.True(t, stalePeer.Status.LoginExpired, "a peer that is still due must be flagged")
+	assert.False(t, updateManager.HasChannel(stalePeer.ID), "the expired peer's update channel must be closed")
+}
+
+// addExpiringPeers registers two SSO peers with login expiration enabled and returns their public keys.
+func addExpiringPeers(t *testing.T, manager *DefaultAccountManager) (string, string) {
+	t.Helper()
+	keys := make([]string, 0, 2)
+	for _, hostname := range []string{"connected-peer", "offline-peer"} {
+		key, err := wgtypes.GenerateKey()
+		require.NoError(t, err, "unable to generate WireGuard key")
+		_, _, _, _, err = manager.AddPeer(context.Background(), "", "", userID, &nbpeer.Peer{
+			Key:                    key.PublicKey().String(),
+			Meta:                   nbpeer.PeerSystemMeta{Hostname: hostname},
+			LoginExpirationEnabled: true,
+		}, false)
+		require.NoError(t, err, "unable to add peer")
+		keys = append(keys, key.PublicKey().String())
+	}
+	return keys[0], keys[1]
+}
+
+func setPeerLogin(t *testing.T, manager *DefaultAccountManager, accountID, peerKey string, connected bool, lastLogin time.Time) {
+	t.Helper()
+	peer, err := manager.Store.GetPeerByPeerPubKey(context.Background(), store.LockingStrengthNone, peerKey)
+	require.NoError(t, err)
+	peer.Status.Connected = connected
+	peer.LastLogin = &lastLogin
+	require.NoError(t, manager.Store.SavePeer(context.Background(), accountID, peer))
+}
+
 func TestDefaultAccountManager_MarkPeerDisconnected_SchedulesInactivityExpiration(t *testing.T) {
 	manager, _, err := createManager(t)
 	require.NoError(t, err, "unable to create account manager")
@@ -2702,7 +2850,7 @@ func TestAccount_GetNextPeerExpiration(t *testing.T) {
 			expectedNextExpiration: time.Duration(0),
 		},
 		{
-			name: "No connected peers, no expiration",
+			name: "Offline peer with expiration, return expiration",
 			peers: map[string]*nbpeer.Peer{
 				"peer-1": {
 					Status: &nbpeer.PeerStatus{
@@ -2721,8 +2869,33 @@ func TestAccount_GetNextPeerExpiration(t *testing.T) {
 			},
 			expiration:             time.Second,
 			expirationEnabled:      false,
-			expectedNextRun:        false,
-			expectedNextExpiration: time.Duration(0),
+			expectedNextRun:        true,
+			expectedNextExpiration: time.Second,
+		},
+		{
+			name: "Offline peer with the earliest deadline defines the next run",
+			peers: map[string]*nbpeer.Peer{
+				"peer-1": {
+					Status: &nbpeer.PeerStatus{
+						Connected: true,
+					},
+					LoginExpirationEnabled: true,
+					LastLogin:              util.ToPtr(time.Now().UTC()),
+					UserID:                 userID,
+				},
+				"peer-2": {
+					Status: &nbpeer.PeerStatus{
+						Connected: false,
+					},
+					LoginExpirationEnabled: true,
+					LastLogin:              util.ToPtr(time.Now().UTC().Add(-50 * time.Minute)),
+					UserID:                 userID,
+				},
+			},
+			expiration:             time.Hour,
+			expirationEnabled:      true,
+			expectedNextRun:        true,
+			expectedNextExpiration: 10 * time.Minute,
 		},
 		{
 			name: "Connected peers with disabled expiration, no expiration",

management/server/group.go

@@ -101,10 +101,8 @@ func (am *DefaultAccountManager) CreateGroup(ctx context.Context, accountID, use
 			return status.Errorf(status.Internal, "failed to create group: %v", err)
 		}
 
-		for _, peerID := range newGroup.Peers {
-			if err := transaction.AddPeerToGroup(ctx, accountID, peerID, newGroup.ID); err != nil {
-				return status.Errorf(status.Internal, "failed to add peer %s to group %s: %v", peerID, newGroup.ID, err)
-			}
+		if err = syncGroupMembership(ctx, transaction, accountID, newGroup.ID, newGroup.Peers, nil); err != nil {
+			return err
 		}
 
 		snap, err = affectedpeers.Load(ctx, transaction, accountID, change)
@@ -200,6 +198,9 @@ func (am *DefaultAccountManager) UpdateGroup(ctx context.Context, accountID, use
 
 // syncGroupMembership applies the peer membership delta for a group within a transaction.
 func syncGroupMembership(ctx context.Context, transaction store.Store, accountID, groupID string, peersToAdd, peersToRemove []string) error {
+	if err := validateGroupPeers(ctx, transaction, accountID, peersToAdd); err != nil {
+		return err
+	}
 	for _, peerID := range peersToAdd {
 		if err := transaction.AddPeerToGroup(ctx, accountID, peerID, groupID); err != nil {
 			return status.Errorf(status.Internal, "failed to add peer %s to group %s: %v", peerID, groupID, err)
@@ -213,6 +214,25 @@ func syncGroupMembership(ctx context.Context, transaction store.Store, accountID
 	return nil
 }
 
+func validateGroupPeers(ctx context.Context, transaction store.Store, accountID string, peerIDs []string) error {
+	if len(peerIDs) == 0 {
+		return nil
+	}
+
+	peers, err := transaction.GetPeersByIDs(ctx, store.LockingStrengthNone, accountID, peerIDs)
+	if err != nil {
+		return err
+	}
+
+	for _, peerID := range peerIDs {
+		if _, ok := peers[peerID]; !ok {
+			return status.Errorf(status.InvalidArgument, "peer with ID %s not found", peerID)
+		}
+	}
+
+	return nil
+}
+
 // CreateGroups adds new groups to the account.
 // Note: This function does not acquire the global lock.
 // It is the caller's responsibility to ensure proper locking is in place before invoking this method.
@@ -540,7 +560,7 @@ func (am *DefaultAccountManager) GroupAddPeer(ctx context.Context, accountID, gr
 	change := affectedpeers.Change{OutputPeerIDs: []string{peerID}, LinkGroups: []string{groupID}}
 
 	err := am.Store.ExecuteInTransaction(ctx, func(transaction store.Store) error {
-		if err := transaction.AddPeerToGroup(ctx, accountID, peerID, groupID); err != nil {
+		if err := syncGroupMembership(ctx, transaction, accountID, groupID, []string{peerID}, nil); err != nil {
 			return err
 		}
 

management/server/group_test.go

@@ -11,10 +11,10 @@ import (
 	"testing"
 	"time"
 
-	"go.uber.org/mock/gomock"
 	"github.com/google/uuid"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"go.uber.org/mock/gomock"
 	"golang.org/x/exp/maps"
 
 	nbdns "github.com/netbirdio/netbird/dns"
@@ -1236,3 +1236,82 @@ func Test_IncrementNetworkSerial(t *testing.T) {
 
 	assert.Equal(t, totalPeers, int(account.Network.Serial), "Expected %d serial increases in account %s, got %d", totalPeers, accountID, account.Network.Serial)
 }
+
+func TestDefaultAccountManager_GroupPeersMustBelongToAccount(t *testing.T) {
+	manager, _, account, peer1, _, _ := setupNetworkMapTest(t)
+
+	otherAccount, err := createAccount(manager, "other_account", "other_user", "")
+	require.NoError(t, err)
+
+	foreignPeer := &peer2.Peer{
+		ID:        "foreign-peer",
+		AccountID: otherAccount.Id,
+		Key:       "foreign-key",
+		DNSLabel:  "foreign-peer",
+		IP:        uint32ToIP(1),
+	}
+	require.NoError(t, manager.Store.AddPeerToAccount(context.Background(), foreignPeer))
+
+	assertRejected := func(t *testing.T, err error) {
+		t.Helper()
+		require.Error(t, err)
+		s, ok := status.FromError(err)
+		require.True(t, ok, "expected status error, got %v", err)
+		assert.Equal(t, status.InvalidArgument, s.Type(), "peer outside the account should be rejected as invalid argument")
+	}
+
+	t.Run("create rejects foreign peer", func(t *testing.T) {
+		err := manager.CreateGroup(context.Background(), account.Id, userID, &types.Group{
+			Name:   "foreign",
+			Issued: types.GroupIssuedAPI,
+			Peers:  []string{peer1.ID, foreignPeer.ID},
+		})
+		assertRejected(t, err)
+
+		_, err = manager.Store.GetGroupByName(context.Background(), store.LockingStrengthNone, account.Id, "foreign")
+		assert.Error(t, err, "rejected create must not persist the group")
+	})
+
+	t.Run("update rejects foreign and unknown peers", func(t *testing.T) {
+		group := &types.Group{ID: "own", Name: "own", Issued: types.GroupIssuedAPI, Peers: []string{peer1.ID}}
+		require.NoError(t, manager.CreateGroup(context.Background(), account.Id, userID, group))
+
+		group.Peers = []string{peer1.ID, foreignPeer.ID}
+		assertRejected(t, manager.UpdateGroup(context.Background(), account.Id, userID, group))
+
+		group.Peers = []string{peer1.ID, "does-not-exist"}
+		assertRejected(t, manager.UpdateGroup(context.Background(), account.Id, userID, group))
+
+		stored, err := manager.Store.GetGroupByID(context.Background(), store.LockingStrengthNone, account.Id, group.ID)
+		require.NoError(t, err)
+		assert.Equal(t, []string{peer1.ID}, stored.Peers, "rejected updates must not change membership")
+	})
+
+	t.Run("update tolerates and drops pre-existing dangling members", func(t *testing.T) {
+		group := &types.Group{ID: "polluted", Name: "polluted", Issued: types.GroupIssuedAPI, Peers: []string{peer1.ID}}
+		require.NoError(t, manager.CreateGroup(context.Background(), account.Id, userID, group))
+		require.NoError(t, manager.Store.AddPeerToGroup(context.Background(), account.Id, foreignPeer.ID, group.ID))
+
+		group.Peers = []string{peer1.ID, foreignPeer.ID}
+		assert.NoError(t, manager.UpdateGroup(context.Background(), account.Id, userID, group), "keeping an existing member must not be rejected")
+
+		group.Peers = []string{peer1.ID}
+		require.NoError(t, manager.UpdateGroup(context.Background(), account.Id, userID, group))
+
+		stored, err := manager.Store.GetGroupByID(context.Background(), store.LockingStrengthNone, account.Id, group.ID)
+		require.NoError(t, err)
+		assert.Equal(t, []string{peer1.ID}, stored.Peers, "dangling member should be removed once omitted")
+	})
+
+	t.Run("direct add rejects foreign and unknown peers", func(t *testing.T) {
+		group := &types.Group{ID: "direct", Name: "direct", Issued: types.GroupIssuedAPI, Peers: []string{peer1.ID}}
+		require.NoError(t, manager.CreateGroup(context.Background(), account.Id, userID, group))
+
+		assertRejected(t, manager.GroupAddPeer(context.Background(), account.Id, group.ID, foreignPeer.ID))
+		assertRejected(t, manager.GroupAddPeer(context.Background(), account.Id, group.ID, "does-not-exist"))
+
+		stored, err := manager.Store.GetGroupByID(context.Background(), store.LockingStrengthNone, account.Id, group.ID)
+		require.NoError(t, err)
+		assert.Equal(t, []string{peer1.ID}, stored.Peers, "rejected direct adds must not change membership")
+	})
+}

management/server/peer.go

@@ -1494,9 +1494,12 @@ func checkAuth(ctx context.Context, loginUserID string, peer *nbpeer.Peer) error
 
 func peerLoginExpired(ctx context.Context, peer *nbpeer.Peer, settings *types.Settings) bool {
 	expired, expiresIn := peer.LoginExpired(settings.PeerLoginExpiration)
-	expired = settings.PeerLoginExpirationEnabled && expired
-	if expired || peer.Status.LoginExpired {
-		log.WithContext(ctx).Debugf("peer's %s login expired %v ago", peer.ID, expiresIn)
+	if settings.PeerLoginExpirationEnabled && expired {
+		log.WithContext(ctx).Debugf("peer's %s login expired %v ago", peer.ID, -expiresIn)
+		return true
+	}
+	if peer.Status.LoginExpired {
+		log.WithContext(ctx).Debugf("peer's %s login is marked as expired", peer.ID)
 		return true
 	}
 	return false
@@ -1643,7 +1646,9 @@ func (am *DefaultAccountManager) UpdateAccountPeer(ctx context.Context, accountI
 
 // getNextPeerExpiration returns the minimum duration in which the next peer of the account will expire if it was found.
 // If there is no peer that expires this function returns false and a duration of 0.
-// This function only considers peers that haven't been expired yet and that are connected.
+// This function only considers peers that haven't been expired yet. Offline peers count too:
+// a running job is never re-armed on connect, so a peer that reconnects with an old login
+// must already be part of the scheduled run.
 func (am *DefaultAccountManager) getNextPeerExpiration(ctx context.Context, accountID string) (time.Duration, bool) {
 	peersWithExpiry, err := am.Store.GetAccountPeersWithExpiration(ctx, store.LockingStrengthNone, accountID)
 	if err != nil {
@@ -1663,8 +1668,7 @@ func (am *DefaultAccountManager) getNextPeerExpiration(ctx context.Context, acco
 
 	var nextExpiry *time.Duration
 	for _, peer := range peersWithExpiry {
-		// consider only connected peers because others will require login on connecting to the management server
-		if peer.Status.LoginExpired || !peer.Status.Connected {
+		if peer.Status.LoginExpired {
 			continue
 		}
 		_, duration := peer.LoginExpired(settings.PeerLoginExpiration)

management/server/scheduler.go

@@ -117,6 +117,7 @@ func (wm *DefaultScheduler) Schedule(ctx context.Context, in time.Duration, ID s
 	}
 
 	ticker := time.NewTicker(in)
+	period := in
 
 	wm.jobs[ID] = cancel
 	log.WithContext(ctx).Debugf("scheduled a job %s to run in %s. There are %d total jobs scheduled.", ID, in.String(), len(wm.jobs))
@@ -136,14 +137,18 @@ func (wm *DefaultScheduler) Schedule(ctx context.Context, in time.Duration, ID s
 				if !reschedule {
 					wm.mu.Lock()
 					defer wm.mu.Unlock()
-					delete(wm.jobs, ID)
+					// A Cancel during job() may have registered a replacement under this ID.
+					if current, ok := wm.jobs[ID]; ok && current == cancel {
+						delete(wm.jobs, ID)
+					}
 					log.WithContext(ctx).Debugf("job %s is not scheduled to run again", ID)
 					ticker.Stop()
 					return
 				}
 				// we need this comparison to avoid resetting the ticker with the same duration and missing the current elapsesed time
-				if runIn != in {
+				if runIn != period {
 					ticker.Reset(runIn)
+					period = runIn
 				}
 			case <-cancel:
 				log.WithContext(ctx).Debugf("job %s was canceled, stopping timer", ID)

management/server/scheduler_test.go

@@ -6,10 +6,12 @@ import (
 	"math/rand"
 	"runtime"
 	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func TestScheduler_Performance(t *testing.T) {
@@ -150,3 +152,90 @@ func TestScheduler_Schedule(t *testing.T) {
 	scheduler.cancel(context.Background(), jobID)
 
 }
+
+func TestScheduler_Schedule_ResetsTickerAfterReturningInitialInterval(t *testing.T) {
+	jobID := "test-scheduler-job-2"
+	scheduler := NewDefaultScheduler()
+	defer scheduler.Cancel(context.Background(), []string{jobID})
+
+	initial := 30 * time.Millisecond
+	stretched := 400 * time.Millisecond
+	runs := make(chan time.Time, 3)
+	count := 0
+	// The first run stretches the period; the second returns the initial interval again,
+	// which must shrink the period back instead of keeping the stretched one.
+	job := func() (nextRunIn time.Duration, reschedule bool) {
+		count++
+		runs <- time.Now()
+		switch count {
+		case 1:
+			return stretched, true
+		case 2:
+			return initial, true
+		default:
+			return 0, false
+		}
+	}
+	scheduler.Schedule(context.Background(), initial, jobID, job)
+
+	var stamps []time.Time
+	for len(stamps) < 3 {
+		select {
+		case ts := <-runs:
+			stamps = append(stamps, ts)
+		case <-time.After(2 * time.Second):
+			t.Fatalf("timed out after %d runs", len(stamps))
+		}
+	}
+	assert.Less(t, stamps[2].Sub(stamps[1]), stretched/2, "returning the initial interval must reset the stretched ticker")
+}
+
+func TestScheduler_Schedule_StaleCompletionKeepsReplacement(t *testing.T) {
+	jobID := "test-scheduler-job-3"
+	scheduler := NewDefaultScheduler()
+	defer scheduler.Cancel(context.Background(), []string{jobID})
+
+	started := make(chan struct{})
+	release := make(chan struct{})
+	staleJob := func() (nextRunIn time.Duration, reschedule bool) {
+		close(started)
+		<-release
+		return 0, false
+	}
+	scheduler.Schedule(context.Background(), 10*time.Millisecond, jobID, staleJob)
+	select {
+	case <-started:
+	case <-time.After(time.Second):
+		t.Fatal("timed out waiting for the first job to start")
+	}
+
+	// Cancel the job while it is still executing and register a replacement under the
+	// same ID, as the expiration paths do on a settings change.
+	scheduler.Cancel(context.Background(), []string{jobID})
+	var replacementRuns atomic.Int32
+	scheduler.Schedule(context.Background(), 20*time.Millisecond, jobID, func() (nextRunIn time.Duration, reschedule bool) {
+		replacementRuns.Add(1)
+		return 20 * time.Millisecond, true
+	})
+	require.True(t, scheduler.IsSchedulerRunning(jobID), "replacement must be registered")
+
+	// The stale job now completes without rescheduling; its cleanup must leave the
+	// replacement's entry in place.
+	close(release)
+	assert.Never(t, func() bool { return !scheduler.IsSchedulerRunning(jobID) }, 200*time.Millisecond, 10*time.Millisecond,
+		"stale completion must not drop the replacement job")
+
+	var duplicateRuns atomic.Int32
+	scheduler.Schedule(context.Background(), 10*time.Millisecond, jobID, func() (nextRunIn time.Duration, reschedule bool) {
+		duplicateRuns.Add(1)
+		return 10 * time.Millisecond, true
+	})
+	assert.Never(t, func() bool { return duplicateRuns.Load() > 0 }, 100*time.Millisecond, 10*time.Millisecond,
+		"a duplicate schedule must be refused while the replacement is registered")
+
+	scheduler.Cancel(context.Background(), []string{jobID})
+	assert.False(t, scheduler.IsSchedulerRunning(jobID), "cancel must find and remove the replacement")
+	runsAfterCancel := replacementRuns.Load()
+	assert.Never(t, func() bool { return replacementRuns.Load() > runsAfterCancel+1 }, 150*time.Millisecond, 10*time.Millisecond,
+		"the replacement must stop after cancel")
+}

management/server/store/sql_store.go

@@ -3476,7 +3476,7 @@ func (s *SqlStore) GetPeerGroups(ctx context.Context, lockStrength LockingStreng
 	var groups []*types.Group
 	query := tx.
 		Joins("JOIN group_peers ON group_peers.group_id = groups.id").
-		Where("group_peers.peer_id = ?", peerId).
+		Where("groups.account_id = ? AND group_peers.peer_id = ?", accountId, peerId).
 		Preload(clause.Associations).
 		Find(&groups)
 
@@ -5056,7 +5056,7 @@ func (s *SqlStore) GetPeersByGroupIDs(ctx context.Context, accountID string, gro
 		Select("DISTINCT peer_id").
 		Where("account_id = ? AND group_id IN ?", accountID, groupIDs)
 
-	result := s.db.Where("id IN (?)", peerIDsSubquery).Find(&peers)
+	result := s.db.Where("account_id = ? AND id IN (?)", accountID, peerIDsSubquery).Find(&peers)
 	if result.Error != nil {
 		log.WithContext(ctx).Errorf("failed to get peers by group IDs: %s", result.Error)
 		return nil, status.Errorf(status.Internal, "failed to get peers by group IDs")
@@ -5689,6 +5689,23 @@ func (s *SqlStore) ListCustomDomains(ctx context.Context, accountID string) ([]*
 	return domains, nil
 }
 
+// GetCustomDomainByName returns the custom domain row holding the given name,
+// regardless of which account owns it.
+func (s *SqlStore) GetCustomDomainByName(ctx context.Context, domainName string) (*domain.Domain, error) {
+	customDomain := &domain.Domain{}
+	result := s.db.Take(customDomain, "domain = ?", domainName)
+	if result.Error != nil {
+		if errors.Is(result.Error, gorm.ErrRecordNotFound) {
+			return nil, status.Errorf(status.NotFound, "custom domain %s not found", domainName)
+		}
+
+		log.WithContext(ctx).Errorf("failed to get custom domain by name from store: %v", result.Error)
+		return nil, status.Errorf(status.Internal, "failed to get custom domain from store")
+	}
+
+	return customDomain, nil
+}
+
 func (s *SqlStore) CreateCustomDomain(ctx context.Context, accountID string, domainName string, targetCluster string, validated bool) (*domain.Domain, error) {
 	newDomain := &domain.Domain{
 		ID:            xid.New().String(), // Generate our own ID because gorm doesn't always configure the database to handle this for us.
@@ -5700,6 +5717,18 @@ func (s *SqlStore) CreateCustomDomain(ctx context.Context, accountID string, dom
 	}
 	result := s.db.Create(newDomain)
 	if result.Error != nil {
+		// The unique index is the last guard when two requests clear the
+		// manager's availability check at the same time. The one that loses the
+		// insert is a conflict, not an internal failure.
+		var count int64
+		if err := s.db.Model(&domain.Domain{}).Where("domain = ?", domainName).Count(&count).Error; err == nil && count > 0 {
+			// The insert error is logged even on this path: the name being taken
+			// is what the caller has to act on, but if the insert also failed for
+			// an unrelated reason the operator still needs to see it.
+			log.WithContext(ctx).Warnf("create reverse proxy custom domain %s rejected, name already registered: %v", domainName, result.Error)
+			return nil, status.Errorf(status.AlreadyExists, "domain %s is already registered", domainName)
+		}
+
 		log.WithContext(ctx).Errorf("failed to create reverse proxy custom domain to store: %v", result.Error)
 		return nil, status.Errorf(status.Internal, "failed to create reverse proxy custom domain to store")
 	}

management/server/store/sql_store_test.go

@@ -2844,6 +2844,14 @@ func TestSqlStore_GetPeerGroups(t *testing.T) {
 	groups, err = store.GetPeerGroups(context.Background(), LockingStrengthNone, accountID, peerID)
 	require.NoError(t, err)
 	assert.Len(t, groups, 2)
+
+	foreignPeerID := "foreign-peer"
+	err = store.AddPeerToGroup(context.Background(), accountID, foreignPeerID, "cfefqs706sqkneg59g4h")
+	require.NoError(t, err)
+
+	groups, err = store.GetPeerGroups(context.Background(), LockingStrengthNone, "other-account", foreignPeerID)
+	require.NoError(t, err)
+	assert.Empty(t, groups, "groups of another account must not be returned")
 }
 
 func TestSqlStore_GetAccountPeers(t *testing.T) {
@@ -4039,9 +4047,15 @@ func TestSqlStore_GetPeersByGroupIDs(t *testing.T) {
 			}
 			require.NoError(t, store.CreateGroups(ctx, accountID, groups))
 
+			otherAccount := newAccountWithId(ctx, "other-account", "other-user", "")
+			require.NoError(t, store.SaveAccount(ctx, otherAccount))
+			foreignPeer := &nbpeer.Peer{ID: "foreign-peer", AccountID: otherAccount.Id}
+			require.NoError(t, store.AddPeerToAccount(ctx, foreignPeer))
+
 			require.NoError(t, store.AddPeerToGroup(ctx, accountID, peer1, group1ID))
 			require.NoError(t, store.AddPeerToGroup(ctx, accountID, peer2, group1ID))
 			require.NoError(t, store.AddPeerToGroup(ctx, accountID, peer1, group2ID))
+			require.NoError(t, store.AddPeerToGroup(ctx, accountID, foreignPeer.ID, group1ID))
 
 			peers, err := store.GetPeersByGroupIDs(ctx, accountID, tt.groupIDs)
 			require.NoError(t, err)

management/server/store/store.go

@@ -302,6 +302,7 @@ type Store interface {
 	GetCustomDomain(ctx context.Context, accountID string, domainID string) (*domain.Domain, error)
 	ListFreeDomains(ctx context.Context, accountID string) ([]string, error)
 	ListCustomDomains(ctx context.Context, accountID string) ([]*domain.Domain, error)
+	GetCustomDomainByName(ctx context.Context, domainName string) (*domain.Domain, error)
 	CreateCustomDomain(ctx context.Context, accountID string, domainName string, targetCluster string, validated bool) (*domain.Domain, error)
 	UpdateCustomDomain(ctx context.Context, accountID string, d *domain.Domain) (*domain.Domain, error)
 	DeleteCustomDomain(ctx context.Context, accountID string, domainID string) error

management/server/store/store_mock.go

@@ -1941,6 +1941,21 @@ func (mr *MockStoreMockRecorder) GetCustomDomain(ctx, accountID, domainID any) *
 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetCustomDomain", reflect.TypeOf((*MockStore)(nil).GetCustomDomain), ctx, accountID, domainID)
 }
 
+// GetCustomDomainByName mocks base method.
+func (m *MockStore) GetCustomDomainByName(ctx context.Context, domainName string) (*domain.Domain, error) {
+	m.ctrl.T.Helper()
+	ret := m.ctrl.Call(m, "GetCustomDomainByName", ctx, domainName)
+	ret0, _ := ret[0].(*domain.Domain)
+	ret1, _ := ret[1].(error)
+	return ret0, ret1
+}
+
+// GetCustomDomainByName indicates an expected call of GetCustomDomainByName.
+func (mr *MockStoreMockRecorder) GetCustomDomainByName(ctx, domainName any) *gomock.Call {
+	mr.mock.ctrl.T.Helper()
+	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetCustomDomainByName", reflect.TypeOf((*MockStore)(nil).GetCustomDomainByName), ctx, domainName)
+}
+
 // GetCustomDomainsCounts mocks base method.
 func (m *MockStore) GetCustomDomainsCounts(ctx context.Context) (int64, int64, error) {
 	m.ctrl.T.Helper()

management/server/types/account.go

@@ -404,7 +404,7 @@ func (a *Account) GetExpiredPeers() []*nbpeer.Peer {
 
 // GetNextPeerExpiration returns the minimum duration in which the next peer of the account will expire if it was found.
 // If there is no peer that expires this function returns false and a duration of 0.
-// This function only considers peers that haven't been expired yet and that are connected.
+// This function only considers peers that haven't been expired yet, whether connected or not.
 func (a *Account) GetNextPeerExpiration() (time.Duration, bool) {
 	peersWithExpiry := a.GetPeersWithExpiration()
 	if len(peersWithExpiry) == 0 {
@@ -412,8 +412,7 @@ func (a *Account) GetNextPeerExpiration() (time.Duration, bool) {
 	}
 	var nextExpiry *time.Duration
 	for _, peer := range peersWithExpiry {
-		// consider only connected peers because others will require login on connecting to the management server
-		if peer.Status.LoginExpired || !peer.Status.Connected {
+		if peer.Status.LoginExpired {
 			continue
 		}
 		_, duration := peer.LoginExpired(a.Settings.PeerLoginExpiration)

management/server/user.go

@@ -1177,28 +1177,35 @@ func (am *DefaultAccountManager) expireAndUpdatePeers(ctx context.Context, accou
 	dnsDomain := am.networkMapController.GetDNSDomain(settings)
 
 	var peerIDs []string
-	for _, peer := range peers {
+	defer func() {
+		if len(peerIDs) == 0 {
+			return
+		}
+		// this will trigger peer disconnect from the management service
+		log.Debugf("Expiring %d peers for account %s", len(peerIDs), accountID)
+		am.networkMapController.DisconnectPeers(ctx, accountID, peerIDs)
+	}()
+	for _, candidate := range peers {
 		// nolint:staticcheck
-		ctx = context.WithValue(ctx, nbcontext.PeerIDKey, peer.Key)
+		peerCtx := context.WithValue(ctx, nbcontext.PeerIDKey, candidate.Key)
 
-		if peer.UserID == "" {
+		if candidate.UserID == "" {
 			// we do not want to expire peers that are added via setup key
 			continue
 		}
 
-		if peer.Status.LoginExpired {
+		peer, err := am.expirePeerIfStillDue(peerCtx, accountID, candidate.ID, settings, reason)
+		if err != nil {
+			return err
+		}
+		if peer == nil {
 			continue
 		}
 		peerIDs = append(peerIDs, peer.ID)
-		peer.MarkLoginExpired(true)
-
-		if err := am.Store.SavePeerStatus(ctx, accountID, peer.ID, *peer.Status); err != nil {
-			return err
-		}
 		meta := peer.EventMeta(dnsDomain)
 		meta["reason"] = string(reason)
 		am.StoreEvent(
-			ctx,
+			peerCtx,
 			peer.UserID, peer.ID, accountID,
 			activity.PeerLoginExpired, meta,
 		)
@@ -1215,15 +1222,53 @@ func (am *DefaultAccountManager) expireAndUpdatePeers(ctx context.Context, accou
 	if err != nil {
 		return fmt.Errorf("notify network map controller of peer update: %w", err)
 	}
-
-	if len(peerIDs) != 0 {
-		// this will trigger peer disconnect from the management service
-		log.Debugf("Expiring %d peers for account %s", len(peerIDs), accountID)
-		am.networkMapController.DisconnectPeers(ctx, accountID, peerIDs)
-	}
 	return nil
 }
 
+// expirePeerIfStillDue flags the peer as login-expired and returns its fresh copy, or nil
+// when it no longer qualifies. The candidate list is read without a lock, so a login that
+// landed in between would otherwise be overwritten with a stale expired status.
+func (am *DefaultAccountManager) expirePeerIfStillDue(ctx context.Context, accountID, peerID string, settings *types.Settings, reason peerExpirationReason) (*nbpeer.Peer, error) {
+	var expired *nbpeer.Peer
+	err := am.Store.ExecuteInTransaction(ctx, func(transaction store.Store) error {
+		peer, err := transaction.GetPeerByID(ctx, store.LockingStrengthUpdate, accountID, peerID)
+		if err != nil {
+			if s, ok := status.FromError(err); ok && s.Type() == status.NotFound {
+				return nil
+			}
+			return err
+		}
+		if peer.Status.LoginExpired || !peerExpirationDue(peer, settings, reason) {
+			return nil
+		}
+		peer.MarkLoginExpired(true)
+		if err := transaction.SavePeerStatus(ctx, accountID, peer.ID, *peer.Status); err != nil {
+			return err
+		}
+		expired = peer
+		return nil
+	})
+	if err != nil {
+		return nil, err
+	}
+	return expired, nil
+}
+
+// peerExpirationDue re-evaluates a time-based expiry against the peer's current state.
+// Administrative reasons expire the peer unconditionally.
+func peerExpirationDue(peer *nbpeer.Peer, settings *types.Settings, reason peerExpirationReason) bool {
+	switch reason {
+	case peerExpirationSessionExpired:
+		expired, _ := peer.LoginExpired(settings.PeerLoginExpiration)
+		return settings.PeerLoginExpirationEnabled && expired
+	case peerExpirationInactivity:
+		expired, _ := peer.SessionExpired(settings.PeerInactivityExpiration)
+		return settings.PeerInactivityExpirationEnabled && expired
+	default:
+		return true
+	}
+}
+
 func (am *DefaultAccountManager) deleteUserFromIDP(ctx context.Context, targetUserID, accountID string) error {
 	if am.userDeleteFromIDPEnabled {
 		log.WithContext(ctx).Debugf("user %s deleted from IdP", targetUserID)
@@ -1337,6 +1382,10 @@ func (am *DefaultAccountManager) deleteRegularUser(ctx context.Context, accountI
 			return fmt.Errorf("failed to get user to delete: %w", err)
 		}
 
+		if targetUser.Role == types.UserRoleOwner && targetUser.Id != initiatorUserID {
+			return status.NewOwnerDeletePermissionError()
+		}
+
 		settings, err = transaction.GetAccountSettings(ctx, store.LockingStrengthNone, accountID)
 		if err != nil {
 			return fmt.Errorf("failed to get account settings: %w", err)

management/server/user_test.go

@@ -942,6 +942,49 @@ func TestUser_DeleteUser_regularUser(t *testing.T) {
 
 }
 
+func TestUser_deleteRegularUser_RejectsOwner(t *testing.T) {
+	s, cleanup, err := store.NewTestStoreFromSQL(context.Background(), "", t.TempDir())
+	require.NoError(t, err)
+	t.Cleanup(cleanup)
+
+	account := newAccountWithId(context.Background(), mockAccountID, mockUserID, "", "", "", false)
+	account.Users[mockTargetUserId] = &types.User{
+		Id:     mockTargetUserId,
+		Issued: types.UserIssuedAPI,
+		Role:   types.UserRoleOwner,
+	}
+	require.NoError(t, s.SaveAccount(context.Background(), account))
+
+	am := DefaultAccountManager{Store: s}
+
+	_, err = am.deleteRegularUser(context.Background(), mockAccountID, mockUserID, &types.UserInfo{ID: mockTargetUserId})
+	assert.EqualError(t, err, status.NewOwnerDeletePermissionError().Error())
+}
+
+func TestUser_deleteRegularUser_InitiatorOwnerDeletesThemself(t *testing.T) {
+	s, cleanup, err := store.NewTestStoreFromSQL(context.Background(), "", t.TempDir())
+	require.NoError(t, err)
+	t.Cleanup(cleanup)
+
+	account := newAccountWithId(context.Background(), mockAccountID, mockUserID, "", "", "", false)
+	require.NoError(t, s.SaveAccount(context.Background(), account))
+
+	networkMapControllerMock := network_map.NewMockController(gomock.NewController(t))
+	networkMapControllerMock.EXPECT().OnPeersDeleted(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any()).Return(nil)
+
+	am := DefaultAccountManager{
+		Store:                s,
+		eventStore:           &activity.InMemoryEventStore{},
+		networkMapController: networkMapControllerMock,
+	}
+
+	_, err = am.deleteRegularUser(context.Background(), mockAccountID, mockUserID, &types.UserInfo{ID: mockUserID})
+	require.NoError(t, err)
+
+	_, err = s.GetUserByUserID(context.Background(), store.LockingStrengthNone, mockUserID)
+	assert.Equal(t, status.NewUserNotFoundError(mockUserID), err)
+}
+
 func TestUser_DeleteUser_RegularUsers(t *testing.T) {
 	store, cleanup, err := store.NewTestStoreFromSQL(context.Background(), "", t.TempDir())
 	if err != nil {

proxy/internal/acme/locker.go

@@ -2,12 +2,14 @@ package acme
 
 import (
 	"context"
+	"fmt"
 	"path/filepath"
 
 	log "github.com/sirupsen/logrus"
 
 	"github.com/netbirdio/netbird/proxy/internal/flock"
 	"github.com/netbirdio/netbird/proxy/internal/k8s"
+	"github.com/netbirdio/netbird/shared/management/domain"
 )
 
 // certLocker provides distributed mutual exclusion for certificate operations.
@@ -74,9 +76,15 @@ func newFlockLocker(certDir string, logger *log.Logger) *flockLocker {
 	return &flockLocker{certDir: certDir, logger: logger}
 }
 
-// Lock acquires an advisory file lock for the given domain.
-func (l *flockLocker) Lock(ctx context.Context, domain string) (func(), error) {
-	lockPath := filepath.Join(l.certDir, domain+".lock")
+// Lock acquires an advisory file lock for the given domain. The domain must
+// be a valid hostname so the lock file always resolves to a direct child of
+// certDir; anything else is rejected before touching the filesystem.
+func (l *flockLocker) Lock(ctx context.Context, name string) (func(), error) {
+	if !domain.IsValidDomainNoWildcard(name) {
+		return nil, fmt.Errorf("invalid domain %q for lock file", name)
+	}
+
+	lockPath := filepath.Join(l.certDir, name+".lock")
 	lockFile, err := flock.Lock(ctx, lockPath)
 	if err != nil {
 		return nil, err
@@ -89,7 +97,7 @@ func (l *flockLocker) Lock(ctx context.Context, domain string) (func(), error) {
 
 	return func() {
 		if err := flock.Unlock(lockFile); err != nil {
-			l.logger.Debugf("release cert lock for domain %q: %v", domain, err)
+			l.logger.Debugf("release cert lock for domain %q: %v", name, err)
 		}
 	}, nil
 }

proxy/internal/acme/locker_test.go

@@ -63,3 +63,33 @@ func TestNewCertLockerK8sFallsBackToFlock(t *testing.T) {
 	_, ok := locker.(*flockLocker)
 	assert.True(t, ok, "k8s-lease without SA should fall back to flockLocker")
 }
+
+func TestFlockLockerRejectsUnsafeDomain(t *testing.T) {
+	root := t.TempDir()
+	certDir := filepath.Join(root, "certs")
+	require.NoError(t, os.Mkdir(certDir, 0o700))
+	locker := newFlockLocker(certDir, nil)
+
+	for _, d := range []string{
+		"",
+		".",
+		"..",
+		"../escape",
+		"../../etc/cron.d/attacker",
+		"sub/dir.example.com",
+		`back\slash.example.com`,
+		"*.example.com",
+	} {
+		unlock, err := locker.Lock(context.Background(), d)
+		assert.Error(t, err, "domain %q", d)
+		assert.Nil(t, unlock, "domain %q", d)
+	}
+
+	assert.NoFileExists(t, filepath.Join(root, "escape.lock"))
+	certEntries, err := os.ReadDir(certDir)
+	require.NoError(t, err)
+	assert.Empty(t, certEntries)
+	rootEntries, err := os.ReadDir(root)
+	require.NoError(t, err)
+	assert.Len(t, rootEntries, 1)
+}

proxy/internal/debug/handler.go

@@ -105,6 +105,20 @@ type Handler struct {
 	startTime  time.Time
 	templates  *template.Template
 	templateMu sync.RWMutex
+
+	// setPerformance applies a buffer cap to one client. Held as a field so
+	// tests can drive applyBufferCap without a live embedded client.
+	setPerformance func(*nbembed.Client, uint32) error
+
+	perfMu       sync.Mutex
+	perfInflight map[types.AccountID]*perfWorker
+}
+
+// perfWorker is the single in-flight retune for one account. err is valid once
+// done is closed.
+type perfWorker struct {
+	done chan struct{}
+	err  error
 }
 
 // NewHandler creates a new debug handler.
@@ -113,10 +127,11 @@ func NewHandler(provider clientProvider, healthChecker healthChecker, logger *lo
 		logger = log.StandardLogger()
 	}
 	h := &Handler{
-		provider:  provider,
-		health:    healthChecker,
-		logger:    logger,
-		startTime: time.Now(),
+		provider:       provider,
+		health:         healthChecker,
+		logger:         logger,
+		startTime:      time.Now(),
+		setPerformance: setClientPerformance,
 	}
 	if err := h.loadTemplates(); err != nil {
 		logger.Errorf("failed to load embedded templates: %v", err)
@@ -716,15 +731,7 @@ func (h *Handler) handlePerf(w http.ResponseWriter, r *http.Request) {
 	}
 
 	capN := uint32(n)
-	applied := 0
-	failed := map[string]string{}
-	for accountID, client := range h.provider.ListClientsForStartup() {
-		if err := client.SetPerformance(nbembed.Performance{PreallocatedBuffersPerPool: &capN}); err != nil {
-			failed[string(accountID)] = err.Error()
-			continue
-		}
-		applied++
-	}
+	applied, failed, inFlight := h.applyBufferCap(capN)
 
 	resp := map[string]any{
 		"success": true,
@@ -734,9 +741,165 @@ func (h *Handler) handlePerf(w http.ResponseWriter, r *http.Request) {
 	if len(failed) > 0 {
 		resp["failed"] = failed
 	}
+	if len(inFlight) > 0 {
+		resp["in_flight"] = inFlight
+	}
 	h.writeJSON(w, resp)
 }
 
+// perfApplyTimeout bounds the whole apply, however many clients are registered.
+// A var, not a const, so tests can shorten the wait.
+var perfApplyTimeout = 5 * time.Second
+
+type perfResult struct {
+	accountID types.AccountID
+	err       error
+}
+
+// setClientPerformance is the production implementation behind Handler.setPerformance.
+func setClientPerformance(client *nbembed.Client, capN uint32) error {
+	return client.SetPerformance(nbembed.Performance{PreallocatedBuffersPerPool: &capN})
+}
+
+// collectBuffered takes every result already sitting in the channel, removing
+// those accounts from pending, and returns how many of them succeeded. It is
+// called when the deadline fires: select picks at random among ready cases, so
+// a result that landed in time would otherwise be reported as a timeout.
+func collectBuffered(results <-chan perfResult, pending map[types.AccountID]*perfWorker, failed map[string]string) int {
+	applied := 0
+	for {
+		select {
+		case res := <-results:
+			delete(pending, res.accountID)
+			if res.err != nil {
+				failed[string(res.accountID)] = res.err.Error()
+				continue
+			}
+			applied++
+		default:
+			return applied
+		}
+	}
+}
+
+// resolvePending closes out the accounts still pending when the deadline fires.
+// A worker whose done channel is closed has finished, whatever the results
+// channel has managed to deliver, so its own error is the truth; the rest are
+// genuinely still running and are reported as timed out. Returns how many of
+// them had in fact succeeded.
+func resolvePending(pending map[types.AccountID]*perfWorker, failed map[string]string) int {
+	applied := 0
+	for accountID, w := range pending {
+		select {
+		case <-w.done:
+			if w.err != nil {
+				failed[string(accountID)] = w.err.Error()
+				continue
+			}
+			applied++
+		default:
+			failed[string(accountID)] = fmt.Sprintf("timed out after %s waiting for the client", perfApplyTimeout)
+		}
+	}
+	return applied
+}
+
+// startPerfWorker returns the in-flight retune for the account, starting one if
+// there is none. The bool reports whether this call started it.
+//
+// At most one retune runs per account at a time. A client wedged inside its own
+// lock never returns, so without this a caller could add one permanently blocked
+// goroutine per request just by retrying the endpoint.
+func (h *Handler) startPerfWorker(accountID types.AccountID, client *nbembed.Client, capN uint32, results chan<- perfResult) (*perfWorker, bool) {
+	h.perfMu.Lock()
+	defer h.perfMu.Unlock()
+
+	if w, ok := h.perfInflight[accountID]; ok {
+		return w, false
+	}
+
+	w := &perfWorker{done: make(chan struct{})}
+	if h.perfInflight == nil {
+		h.perfInflight = make(map[types.AccountID]*perfWorker)
+	}
+	h.perfInflight[accountID] = w
+
+	go func() {
+		err := h.setPerformance(client, capN)
+		w.err = err
+		close(w.done)
+
+		// Publish before touching the registry: perfMu is taken once per
+		// account by every caller walking the fleet, so a finishing worker
+		// can queue behind a long apply and miss its own deadline.
+		results <- perfResult{accountID: accountID, err: err}
+
+		h.perfMu.Lock()
+		delete(h.perfInflight, accountID)
+		h.perfMu.Unlock()
+	}()
+
+	return w, true
+}
+
+// applyBufferCap sets the WireGuard buffer pool cap on every registered client
+// and reports how many took it, a per-account error for those that did not, and
+// the accounts whose earlier retune has not come back yet.
+//
+// Clients are handled concurrently and the wait is bounded: SetPerformance goes
+// through the embedded client's lock, which Start and Stop hold for as long as
+// they take - and on a wedged client Stop never returns. This endpoint is the
+// recovery path for exactly that fleet, so one stuck account must neither delay
+// the others nor accumulate goroutines across retries.
+func (h *Handler) applyBufferCap(capN uint32) (int, map[string]string, []string) {
+	clients := h.provider.ListClientsForStartup()
+	results := make(chan perfResult, len(clients))
+
+	applied := 0
+	failed := map[string]string{}
+	var inFlight []string
+	pending := make(map[types.AccountID]*perfWorker, len(clients))
+
+	for accountID, client := range clients {
+		w, started := h.startPerfWorker(accountID, client, capN, results)
+		if started {
+			pending[accountID] = w
+			continue
+		}
+		// Another request owns this account's retune. Take its result if it
+		// has already landed, otherwise report it as still running instead of
+		// waiting on it again.
+		select {
+		case <-w.done:
+			if w.err != nil {
+				failed[string(accountID)] = w.err.Error()
+				continue
+			}
+			applied++
+		default:
+			inFlight = append(inFlight, string(accountID))
+		}
+	}
+
+	deadline := time.After(perfApplyTimeout)
+	for range len(pending) {
+		select {
+		case res := <-results:
+			delete(pending, res.accountID)
+			if res.err != nil {
+				failed[string(res.accountID)] = res.err.Error()
+				continue
+			}
+			applied++
+		case <-deadline:
+			applied += collectBuffered(results, pending, failed)
+			applied += resolvePending(pending, failed)
+			return applied, failed, inFlight
+		}
+	}
+	return applied, failed, inFlight
+}
+
 // handleRuntime returns cheap runtime and process stats. Safe to hit on a
 // running proxy; does not read pprof profiles.
 func (h *Handler) handleRuntime(w http.ResponseWriter, _ *http.Request) {

proxy/internal/debug/perf_test.go

@@ -0,0 +1,158 @@
+package debug
+
+import (
+	"context"
+	"errors"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	nbembed "github.com/netbirdio/netbird/client/embed"
+	"github.com/netbirdio/netbird/proxy/internal/health"
+	"github.com/netbirdio/netbird/proxy/internal/roundtrip"
+	"github.com/netbirdio/netbird/proxy/internal/types"
+)
+
+// perfProvider serves a fixed set of accounts. The clients are nil: the tests
+// drive Handler.setPerformance, which never dereferences them.
+type perfProvider struct {
+	accounts []types.AccountID
+}
+
+func (p *perfProvider) GetClient(types.AccountID) (*nbembed.Client, bool) { return nil, false }
+
+func (p *perfProvider) ListClientsForDebug() map[types.AccountID]roundtrip.ClientDebugInfo {
+	return nil
+}
+
+func (p *perfProvider) ListClientsForStartup() map[types.AccountID]*nbembed.Client {
+	out := make(map[types.AccountID]*nbembed.Client, len(p.accounts))
+	for _, id := range p.accounts {
+		out[id] = nil
+	}
+	return out
+}
+
+type stubHealth struct{}
+
+func (stubHealth) ReadinessProbe() bool              { return true }
+func (stubHealth) StartupProbe(context.Context) bool { return true }
+func (stubHealth) CheckClientsConnected(context.Context) (bool, map[types.AccountID]health.ClientHealth) {
+	return true, nil
+}
+
+func shortenPerfTimeout(t *testing.T, d time.Duration) {
+	t.Helper()
+	prev := perfApplyTimeout
+	perfApplyTimeout = d
+	t.Cleanup(func() { perfApplyTimeout = prev })
+}
+
+// TestCollectBufferedCountsResultsReadyAtTheDeadline covers the select-ordering
+// trap: when the deadline fires, results already buffered must be counted, not
+// reported as timeouts. Driving collectBuffered directly keeps it deterministic
+// - through applyBufferCap the two select cases race by construction.
+func TestCollectBufferedCountsResultsReadyAtTheDeadline(t *testing.T) {
+	results := make(chan perfResult, 3)
+	results <- perfResult{accountID: "ok"}
+	results <- perfResult{accountID: "broken", err: errors.New("boom")}
+
+	pending := map[types.AccountID]*perfWorker{
+		"ok":     {done: make(chan struct{})},
+		"broken": {done: make(chan struct{})},
+		"wedged": {done: make(chan struct{})},
+	}
+	failed := map[string]string{}
+
+	applied := collectBuffered(results, pending, failed)
+
+	if applied != 1 {
+		t.Fatalf("applied = %d, want 1", applied)
+	}
+	if failed["broken"] != "boom" {
+		t.Fatalf("failed = %v, want the error recorded for \"broken\"", failed)
+	}
+	if _, ok := pending["wedged"]; !ok || len(pending) != 1 {
+		t.Fatalf("pending = %v, want only the account that never answered", pending)
+	}
+}
+
+// TestApplyBufferCapSingleFlightPerAccount covers the goroutine accumulation
+// reported on PR #7452: repeated calls against a client stuck in its own lock
+// must not start a second attempt for the same account.
+func TestApplyBufferCapSingleFlightPerAccount(t *testing.T) {
+	shortenPerfTimeout(t, 50*time.Millisecond)
+
+	release := make(chan struct{})
+	t.Cleanup(func() { close(release) })
+
+	var calls atomic.Int32
+	h := &Handler{
+		provider: &perfProvider{accounts: []types.AccountID{"wedged"}},
+		health:   stubHealth{},
+		setPerformance: func(_ *nbembed.Client, _ uint32) error {
+			calls.Add(1)
+			<-release
+			return nil
+		},
+	}
+
+	for i := range 5 {
+		applied, failed, inFlight := h.applyBufferCap(4096)
+		if applied != 0 {
+			t.Fatalf("call %d: applied = %d, want 0", i, applied)
+		}
+		if i == 0 {
+			if len(failed) != 1 {
+				t.Fatalf("first call: failed = %v, want the account reported as timed out", failed)
+			}
+			continue
+		}
+		if len(inFlight) != 1 {
+			t.Fatalf("call %d: inFlight = %v, want the account reported as still running", i, inFlight)
+		}
+		if len(failed) != 0 {
+			t.Fatalf("call %d: failed = %v, want empty while the retune is in flight", i, failed)
+		}
+	}
+
+	if got := calls.Load(); got != 1 {
+		t.Fatalf("setPerformance called %d times, want 1: each retry started another blocked worker", got)
+	}
+}
+
+// TestResolvePendingTrustsFinishedWorkers covers the reporting race cubic
+// flagged on PR #7452: a retune that finished just before the deadline must be
+// reported by its outcome, not as a timeout, whatever the results channel has
+// delivered so far.
+func TestResolvePendingTrustsFinishedWorkers(t *testing.T) {
+	ok := &perfWorker{done: make(chan struct{})}
+	close(ok.done)
+
+	broken := &perfWorker{done: make(chan struct{}), err: errors.New("boom")}
+	close(broken.done)
+
+	stillRunning := &perfWorker{done: make(chan struct{})}
+
+	pending := map[types.AccountID]*perfWorker{
+		"ok":      ok,
+		"broken":  broken,
+		"running": stillRunning,
+	}
+	failed := map[string]string{}
+
+	applied := resolvePending(pending, failed)
+
+	if applied != 1 {
+		t.Fatalf("applied = %d, want 1", applied)
+	}
+	if failed["broken"] != "boom" {
+		t.Fatalf("failed[broken] = %q, want the worker's own error", failed["broken"])
+	}
+	if _, ok := failed["ok"]; ok {
+		t.Fatalf("failed = %v, want no entry for the account that succeeded", failed)
+	}
+	if got := failed["running"]; got == "" || got == "boom" {
+		t.Fatalf("failed[running] = %q, want the timeout message", got)
+	}
+}

shared/relay/client/client.go

@@ -279,7 +279,7 @@ func (c *Client) Connect(ctx context.Context) error {
 	c.stateSubscription = NewPeersStateSubscription(c.log, c.relayConn, c.closeConnsByPeerID)
 
 	c.log = c.log.WithField("relay", instanceURL.String())
-	c.log.Infof("relay connection established")
+	c.log.Infof("relay connection established, server IP: %s", connectedIP(c.relayConn))
 
 	c.serviceIsRunning = true
 
@@ -364,23 +364,6 @@ func (c *Client) ServerInstanceURL() (string, error) {
 	return c.instanceURL.String(), nil
 }
 
-// ConnectedIP returns the IP address of the live relay-server connection,
-// extracted from the underlying socket's RemoteAddr. Zero value if not
-// connected or if the address is not an IP literal.
-func (c *Client) ConnectedIP() netip.Addr {
-	c.mu.Lock()
-	conn := c.relayConn
-	c.mu.Unlock()
-	if conn == nil {
-		return netip.Addr{}
-	}
-	addr := conn.RemoteAddr()
-	if addr == nil {
-		return netip.Addr{}
-	}
-	return extractIPLiteral(addr.String())
-}
-
 // SetOnDisconnectListener sets a function that will be called when the connection to the relay server is closed.
 func (c *Client) SetOnDisconnectListener(fn func(string)) {
 	c.listenerMutex.Lock()
@@ -777,6 +760,17 @@ func (c *Client) listenForStopEvents(ctx context.Context, hc *healthcheck.Receiv
 	}
 }
 
+func (c *Client) serverInstanceAddress() (string, netip.Addr, error) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	addr, err := c.ServerInstanceURL()
+	if err != nil {
+		return "", netip.Addr{}, err
+	}
+	return addr, connectedIP(c.relayConn), nil
+}
+
 func (c *Client) closeAllConns() {
 	for _, container := range c.conns {
 		container.close()
@@ -923,6 +917,17 @@ func (c *Client) handlePeersWentOfflineMsg(buf []byte) {
 	c.stateSubscription.OnPeersWentOffline(peersID)
 }
 
+func connectedIP(conn net.Conn) netip.Addr {
+	if conn == nil {
+		return netip.Addr{}
+	}
+	addr := conn.RemoteAddr()
+	if addr == nil {
+		return netip.Addr{}
+	}
+	return extractIPLiteral(addr.String())
+}
+
 // extractIPLiteral returns the IP from address forms produced by the relay
 // dialers (URL or host:port). Zero value if the host is not an IP.
 func extractIPLiteral(s string) netip.Addr {

shared/relay/client/client_serverip_test.go

@@ -8,6 +8,8 @@ import (
 	"testing"
 	"time"
 
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 	"go.opentelemetry.io/otel"
 
 	"github.com/netbirdio/netbird/client/iface"
@@ -68,18 +70,17 @@ func TestClient_ServerIPRecoversFromUnresolvableFQDN(t *testing.T) {
 		if !c.Ready() {
 			t.Fatalf("client not ready after connect")
 		}
-		if got := c.ConnectedIP(); got.String() != "127.0.0.1" {
-			t.Fatalf("ConnectedIP = %q, want 127.0.0.1", got)
-		}
+		url, ip, err := c.serverInstanceAddress()
+		require.NoError(t, err)
+		assert.Equal(t, srvCfg.ExposedAddress, url, "relay URL must come from the handshake")
+		assert.Equal(t, netip.MustParseAddr("127.0.0.1"), ip, "relay IP must come from the connection")
 	})
 }
 
-// TestClient_ConnectedIPAfterFQDNDial verifies ConnectedIP returns the
-// resolved IP after a successful FQDN-based dial. The underlying socket's
-// RemoteAddr must be exposed through the dialer wrappers; if it returns
-// the dial-time URL instead, ConnectedIP returns empty and the dial
-// IP we advertise to peers is empty too.
-func TestClient_ConnectedIPAfterFQDNDial(t *testing.T) {
+// TestClient_ServerInstanceAddressAfterFQDNDial verifies the relay address
+// includes the resolved IP after an FQDN dial. The dialer wrappers must expose
+// the socket's RemoteAddr; returning the dial-time URL would lose the IP.
+func TestClient_ServerInstanceAddressAfterFQDNDial(t *testing.T) {
 	ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
 	defer cancel()
 
@@ -111,10 +112,10 @@ func TestClient_ConnectedIPAfterFQDNDial(t *testing.T) {
 	}
 	t.Cleanup(func() { _ = c.Close() })
 
-	got := c.ConnectedIP().String()
-	if got != "127.0.0.1" && got != "::1" {
-		t.Fatalf("ConnectedIP after FQDN dial = %q, want 127.0.0.1 or ::1", got)
-	}
+	url, ip, err := c.serverInstanceAddress()
+	require.NoError(t, err)
+	assert.Equal(t, srvCfg.ExposedAddress, url, "relay URL must come from the handshake")
+	assert.Contains(t, []string{"127.0.0.1", "::1"}, ip.String(), "relay IP must resolve to localhost")
 }
 
 func TestSubstituteHost(t *testing.T) {
@@ -214,15 +215,12 @@ func TestSubstituteHost(t *testing.T) {
 	}
 }
 
-func TestClient_ConnectedIPEmptyWhenNotConnected(t *testing.T) {
-	c := NewClient("rel://example.invalid:80", hmacTokenStore, "x", iface.DefaultMTU)
-	if got := c.ConnectedIP(); got.IsValid() {
-		t.Fatalf("ConnectedIP on disconnected client = %q, want zero", got)
-	}
+func TestConnectedIPNilConnection(t *testing.T) {
+	assert.False(t, connectedIP(nil).IsValid(), "missing connection must not provide an IP")
 }
 
 // staticAddr is a net.Addr that returns a fixed string. Used to verify
-// ConnectedIP parses RemoteAddr correctly.
+// connectedIP parses RemoteAddr correctly.
 type staticAddr struct{ s string }
 
 func (a staticAddr) Network() string { return "tcp" }
@@ -235,7 +233,7 @@ type stubConn struct {
 
 func (s stubConn) RemoteAddr() net.Addr { return s.remote }
 
-func TestClient_ConnectedIPParsesRemoteAddr(t *testing.T) {
+func TestConnectedIPParsesRemoteAddr(t *testing.T) {
 	tests := []struct {
 		name string
 		s    string
@@ -252,15 +250,12 @@ func TestClient_ConnectedIPParsesRemoteAddr(t *testing.T) {
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			c := &Client{relayConn: stubConn{remote: staticAddr{s: tt.s}}}
-			got := c.ConnectedIP()
+			got := connectedIP(stubConn{remote: staticAddr{s: tt.s}})
 			var gotStr string
 			if got.IsValid() {
 				gotStr = got.String()
 			}
-			if gotStr != tt.want {
-				t.Errorf("ConnectedIP(%q) = %q, want %q", tt.s, gotStr, tt.want)
-			}
+			assert.Equal(t, tt.want, gotStr, "IP extracted from RemoteAddr %q", tt.s)
 		})
 	}
 }

shared/relay/client/manager.go

@@ -256,11 +256,7 @@ func (m *Manager) RelayInstanceAddress() (string, netip.Addr, error) {
 	if m.relayClient == nil {
 		return "", netip.Addr{}, ErrRelayClientNotConnected
 	}
-	addr, err := m.relayClient.ServerInstanceURL()
-	if err != nil {
-		return "", netip.Addr{}, err
-	}
-	return addr, m.relayClient.ConnectedIP(), nil
+	return m.relayClient.serverInstanceAddress()
 }
 
 // ServerURLs returns the addresses of the relay servers.