Merge branch 'main' into windows-dns-firewall

* Made the docker check first for getting-started.sh, better atomic support for install.sh

* Check for docker socket perms

* Added fallback for systems without rpm-ostree or bootc.

* macOS fix for docker socket check

* Change error message for docker group.

No longer using a blanket recommendation for the docker group.

.github/workflows/golang-test-darwin.yml

@@ -45,4 +45,11 @@ jobs:
         run: git --no-pager diff --exit-code
 
       - name: Test
-        run: NETBIRD_STORE_ENGINE=${{ matrix.store }} CI=true go test -tags=devcert -exec 'sudo --preserve-env=CI,NETBIRD_STORE_ENGINE' -timeout 5m -p 1 $(go list ./... | grep -v -e /management -e /signal -e /relay -e /proxy -e /combined)
+        run: NETBIRD_STORE_ENGINE=${{ matrix.store }} CI=true go test -coverprofile=coverage.txt -tags=devcert -exec 'sudo --preserve-env=CI,NETBIRD_STORE_ENGINE' -timeout 5m -p 1 $(go list ./... | grep -v -e /management -e /signal -e /relay -e /proxy -e /combined)
+
+      - name: Upload coverage reports to Codecov
+        uses: codecov/codecov-action@e79a6962e0d4c0c17b229090214935d2e33f8354 #v6.0.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          slug: netbirdio/netbird
+          flags: unit,client

.github/workflows/golang-test-linux.yml

@@ -158,7 +158,16 @@ jobs:
         run: git --no-pager diff --exit-code
 
       - name: Test
-        run: CGO_ENABLED=1 GOARCH=${{ matrix.arch }} CI=true go test -tags devcert -exec 'sudo' -timeout 10m -p 1 $(go list ./... | grep -v -e /management -e /signal -e /relay -e /proxy -e /combined)
+        run: CGO_ENABLED=1 GOARCH=${{ matrix.arch }} CI=true go test -coverprofile=coverage.txt -tags devcert -exec 'sudo' -timeout 10m -p 1 $(go list ./... | grep -v -e /management -e /signal -e /relay -e /proxy -e /combined)
+
+      - name: Upload coverage reports to Codecov
+        if: matrix.arch == 'amd64'
+        uses: codecov/codecov-action@e79a6962e0d4c0c17b229090214935d2e33f8354 #v6.0.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          slug: netbirdio/netbird
+          flags: unit,client
+
 
   test_client_on_docker:
     name: "Client (Docker) / Unit"
@@ -276,9 +285,17 @@ jobs:
         run: |
           CGO_ENABLED=1 GOARCH=${{ matrix.arch }} \
           go test ${{ matrix.raceFlag }} \
-            -exec 'sudo' \
+            -exec 'sudo' -coverprofile=coverage.txt \
             -timeout 10m -p 1 ./relay/... ./shared/relay/...
 
+      - name: Upload coverage reports to Codecov
+        if: matrix.arch == 'amd64'
+        uses: codecov/codecov-action@e79a6962e0d4c0c17b229090214935d2e33f8354 #v6.0.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          slug: netbirdio/netbird
+          flags: unit,relay
+
   test_proxy:
     name: "Proxy / Unit"
     needs: [build-cache]
@@ -326,7 +343,15 @@ jobs:
       - name: Test
         run: |
           CGO_ENABLED=1 GOARCH=${{ matrix.arch }} \
-          go test -timeout 10m -p 1 ./proxy/...
+          go test -timeout 10m -p 1 -coverprofile=coverage.txt ./proxy/...
+
+      - name: Upload coverage reports to Codecov
+        if: matrix.arch == 'amd64'
+        uses: codecov/codecov-action@e79a6962e0d4c0c17b229090214935d2e33f8354 #v6.0.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          slug: netbirdio/netbird
+          flags: unit,proxy
 
   test_signal:
     name: "Signal / Unit"
@@ -377,9 +402,17 @@ jobs:
         run: |
           CGO_ENABLED=1 GOARCH=${{ matrix.arch }} \
           go test \
-            -exec 'sudo' \
+            -exec 'sudo' -coverprofile=coverage.txt \
             -timeout 10m ./signal/... ./shared/signal/...
 
+      - name: Upload coverage reports to Codecov
+        if: matrix.arch == 'amd64'
+        uses: codecov/codecov-action@e79a6962e0d4c0c17b229090214935d2e33f8354 #v6.0.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          slug: netbirdio/netbird
+          flags: unit,signal
+
   test_management:
     name: "Management / Unit"
     needs: [build-cache]
@@ -445,10 +478,18 @@ jobs:
           CGO_ENABLED=1 GOARCH=${{ matrix.arch }} \
           NETBIRD_STORE_ENGINE=${{ matrix.store }} \
             CI=true \
-          go test -tags=devcert \
+          go test -tags=devcert -coverprofile=coverage.txt \
             -exec "sudo --preserve-env=CI,NETBIRD_STORE_ENGINE" \
             -timeout 20m ./management/... ./shared/management/...
 
+      - name: Upload coverage reports to Codecov
+        if: matrix.arch == 'amd64'
+        uses: codecov/codecov-action@e79a6962e0d4c0c17b229090214935d2e33f8354 #v6.0.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          slug: netbirdio/netbird
+          flags: unit,management
+
   benchmark:
     name: "Management / Benchmark"
     needs: [build-cache]
@@ -687,6 +728,14 @@ jobs:
           CGO_ENABLED=1 GOARCH=${{ matrix.arch }} \
           NETBIRD_STORE_ENGINE=${{ matrix.store }} \
           CI=true \
-          go test -tags=integration \
+          go test -tags=integration -coverprofile=coverage.txt \
           -exec 'sudo --preserve-env=CI,NETBIRD_STORE_ENGINE' \
           -timeout 20m ./management/server/http/...
+
+      - name: Upload coverage reports to Codecov
+        if: matrix.arch == 'amd64'
+        uses: codecov/codecov-action@e79a6962e0d4c0c17b229090214935d2e33f8354 #v6.0.1
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }}
+          slug: netbirdio/netbird
+          flags: integration,management

.github/workflows/proto-version-check.yml

@@ -20,15 +20,30 @@ jobs:
               per_page: 100,
             });
 
-            const modifiedPbFiles = files.filter(
-              f => f.filename.endsWith('.pb.go') && f.status === 'modified'
-            );
-            if (modifiedPbFiles.length === 0) {
-              console.log('No modified .pb.go files to check');
+            // Cover renamed .pb.go files in addition to plain edits.
+            // Renamed entries land under the new path with previous_filename
+            // pointing at the base-side name, so we read the base content
+            // from the old path when present.
+            const changedPbFiles = files
+              .filter(f => (f.status === 'modified' || f.status === 'renamed')
+                && f.filename.endsWith('.pb.go'))
+              .map(f => ({
+                headPath: f.filename,
+                basePath: f.previous_filename || f.filename,
+              }));
+            if (changedPbFiles.length === 0) {
+              console.log('No modified or renamed .pb.go files to check');
               return;
             }
 
-            const versionPattern = /^\s*\/\/\s+protoc(?:-gen-go)?\s+v[\d.]+/;
+            // Matches the generator version headers protoc writes at the top
+            // of generated files:
+            //   //   protoc           v3.21.12
+            //   //   protoc-gen-go    v1.26.0
+            //   // - protoc-gen-go-grpc v1.6.1   (grpc files prefix with "- ")
+            // The optional "- " prefix and the optional -gen-go / -gen-go-grpc
+            // suffixes keep the *_grpc.pb.go headers in scope.
+            const versionPattern = /^\s*\/\/\s+(?:-\s+)?protoc(?:-gen-go(?:-grpc)?)?\s+v[\d.]+/;
             const baseSha = context.payload.pull_request.base.sha;
             const headSha = context.payload.pull_request.head.sha;
 
@@ -55,20 +70,22 @@ jobs:
             }
 
             const violations = [];
-            for (const file of modifiedPbFiles) {
+            for (const file of changedPbFiles) {
               const [base, head] = await Promise.all([
-                getVersionHeader(file.filename, baseSha),
-                getVersionHeader(file.filename, headSha),
+                getVersionHeader(file.basePath, baseSha),
+                getVersionHeader(file.headPath, headSha),
               ]);
               if (!base.ok || !head.ok) {
                 core.warning(
-                  `Skipping ${file.filename}: base=${base.ok ? 'ok' : base.reason}, head=${head.ok ? 'ok' : head.reason}`
+                  `Skipping ${file.headPath}: base=${base.ok ? 'ok' : base.reason}, head=${head.ok ? 'ok' : head.reason}`
                 );
                 continue;
               }
               if (base.lines.join('\n') !== head.lines.join('\n')) {
                 violations.push({
-                  file: file.filename,
+                  file: file.basePath === file.headPath
+                    ? file.headPath
+                    : `${file.basePath} → ${file.headPath}`,
                   base: base.lines,
                   head: head.lines,
                 });

.github/workflows/release.yml

@@ -29,10 +29,10 @@ jobs:
           persist-credentials: false
 
       - name: Generate FreeBSD port diff
-        run: bash release_files/freebsd-port-diff.sh
+        run: bash -x release_files/freebsd-port-diff.sh
 
       - name: Generate FreeBSD port issue body
-        run: bash release_files/freebsd-port-issue-body.sh
+        run: bash -x release_files/freebsd-port-issue-body.sh
 
       - name: Check if diff was generated
         id: check_diff

.github/workflows/wasm-build-validation.yml

@@ -65,7 +65,7 @@ jobs:
 
           echo "Size: ${SIZE} bytes (${SIZE_MB} MB)"
 
-          if [ ${SIZE} -gt 58720256 ]; then
-            echo "Wasm binary size (${SIZE_MB}MB) exceeds 56MB limit!"
+          if [ ${SIZE} -gt 62914560 ]; then
+            echo "Wasm binary size (${SIZE_MB}MB) exceeds 60MB limit!"
             exit 1
           fi

.golangci.yaml

@@ -92,6 +92,9 @@ linters:
       - linters:
           - unused
         path: client/firewall/iptables/rule\.go
+      - linters:
+          - unused
+        path: client/internal/dns/dnsfw/(types|syscall|zsyscall)_windows.*\.go
       - linters:
           - gosec
           - mirror

client/cmd/debug.go

@@ -19,6 +19,7 @@ import (
 	"github.com/netbirdio/netbird/client/server"
 	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
 	"github.com/netbirdio/netbird/upload-server/types"
+	"github.com/netbirdio/netbird/version"
 )
 
 const errCloseConnection = "Failed to close connection: %v"
@@ -100,6 +101,7 @@ func debugBundle(cmd *cobra.Command, _ []string) error {
 		Anonymize:    anonymizeFlag,
 		SystemInfo:   systemInfoFlag,
 		LogFileCount: logFileCount,
+		CliVersion:   version.NetbirdVersion(),
 	}
 	if uploadBundleFlag {
 		request.UploadURL = uploadBundleURLFlag
@@ -298,6 +300,7 @@ func runForDuration(cmd *cobra.Command, args []string) error {
 		Anonymize:    anonymizeFlag,
 		SystemInfo:   systemInfoFlag,
 		LogFileCount: logFileCount,
+		CliVersion:   version.NetbirdVersion(),
 	}
 	if uploadBundleFlag {
 		request.UploadURL = uploadBundleURLFlag
@@ -432,6 +435,7 @@ func generateDebugBundle(config *profilemanager.Config, recorder *peer.Status, c
 			SyncResponse:   syncResponse,
 			LogPath:        logFilePath,
 			CPUProfile:     nil,
+			DaemonVersion:  version.NetbirdVersion(), // acting as daemon
 		},
 		debug.BundleConfig{
 			IncludeSystemInfo: true,

client/cmd/service_controller.go

@@ -102,7 +102,7 @@ func (p *program) Stop(srv service.Service) error {
 }
 
 // Common setup for service control commands
-func setupServiceControlCommand(cmd *cobra.Command, ctx context.Context, cancel context.CancelFunc) (service.Service, error) {
+func setupServiceControlCommand(cmd *cobra.Command, ctx context.Context, cancel context.CancelFunc, consoleLog bool) (service.Service, error) {
 	// rootCmd env vars are already applied by PersistentPreRunE.
 	SetFlagsFromEnvVars(serviceCmd)
 
@@ -112,8 +112,14 @@ func setupServiceControlCommand(cmd *cobra.Command, ctx context.Context, cancel
 		return nil, err
 	}
 
-	if err := util.InitLog(logLevel, logFiles...); err != nil {
-		return nil, fmt.Errorf("init log: %w", err)
+	if consoleLog {
+		if err := util.InitLog(logLevel, util.LogConsole); err != nil {
+			return nil, fmt.Errorf("init log: %w", err)
+		}
+	} else {
+		if err := util.InitLog(logLevel, logFiles...); err != nil {
+			return nil, fmt.Errorf("init log: %w", err)
+		}
 	}
 
 	cfg, err := newSVCConfig()
@@ -138,7 +144,7 @@ var runCmd = &cobra.Command{
 		SetupCloseHandler(ctx, cancel)
 		SetupDebugHandler(ctx, nil, nil, nil, util.FindFirstLogPath(logFiles))
 
-		s, err := setupServiceControlCommand(cmd, ctx, cancel)
+		s, err := setupServiceControlCommand(cmd, ctx, cancel, false)
 		if err != nil {
 			return err
 		}
@@ -152,7 +158,7 @@ var startCmd = &cobra.Command{
 	Short: "starts NetBird service",
 	RunE: func(cmd *cobra.Command, args []string) error {
 		ctx, cancel := context.WithCancel(cmd.Context())
-		s, err := setupServiceControlCommand(cmd, ctx, cancel)
+		s, err := setupServiceControlCommand(cmd, ctx, cancel, false)
 		if err != nil {
 			return err
 		}
@@ -170,7 +176,7 @@ var stopCmd = &cobra.Command{
 	Short: "stops NetBird service",
 	RunE: func(cmd *cobra.Command, args []string) error {
 		ctx, cancel := context.WithCancel(cmd.Context())
-		s, err := setupServiceControlCommand(cmd, ctx, cancel)
+		s, err := setupServiceControlCommand(cmd, ctx, cancel, false)
 		if err != nil {
 			return err
 		}
@@ -188,7 +194,7 @@ var restartCmd = &cobra.Command{
 	Short: "restarts NetBird service",
 	RunE: func(cmd *cobra.Command, args []string) error {
 		ctx, cancel := context.WithCancel(cmd.Context())
-		s, err := setupServiceControlCommand(cmd, ctx, cancel)
+		s, err := setupServiceControlCommand(cmd, ctx, cancel, false)
 		if err != nil {
 			return err
 		}
@@ -206,7 +212,7 @@ var svcStatusCmd = &cobra.Command{
 	Short: "shows NetBird service status",
 	RunE: func(cmd *cobra.Command, args []string) error {
 		ctx, cancel := context.WithCancel(cmd.Context())
-		s, err := setupServiceControlCommand(cmd, ctx, cancel)
+		s, err := setupServiceControlCommand(cmd, ctx, cancel, true)
 		if err != nil {
 			return err
 		}

client/cmd/version.go

@@ -12,7 +12,13 @@ var (
 		Short: "Print the NetBird's client application version",
 		Run: func(cmd *cobra.Command, args []string) {
 			cmd.SetOut(cmd.OutOrStdout())
-			cmd.Println(version.NetbirdVersion())
+			out := version.NetbirdVersion()
+			if version.IsDevelopmentVersion(out) {
+				if commit := version.NetbirdCommit(); commit != "" {
+					out += "-" + commit
+				}
+			}
+			cmd.Println(out)
 		},
 	}
 )

client/firewall/iptables/acl_linux.go

@@ -3,6 +3,7 @@ package iptables
 import (
 	"errors"
 	"fmt"
+	"maps"
 	"net"
 	"slices"
 
@@ -421,12 +422,17 @@ func (m *aclManager) updateState() {
 	currentState.Lock()
 	defer currentState.Unlock()
 
+	// Clone the maps so the persisted state holds a private snapshot. The
+	// live maps keep being mutated by subsequent rule operations while the
+	// state manager marshals the state from its periodic-save goroutine.
+	// Sharing them by reference races the two and aborts the process with a
+	// concurrent map iteration and write.
 	if m.v6 {
-		currentState.ACLEntries6 = m.entries
-		currentState.ACLIPsetStore6 = m.ipsetStore
+		currentState.ACLEntries6 = maps.Clone(m.entries)
+		currentState.ACLIPsetStore6 = m.ipsetStore.clone()
 	} else {
-		currentState.ACLEntries = m.entries
-		currentState.ACLIPsetStore = m.ipsetStore
+		currentState.ACLEntries = maps.Clone(m.entries)
+		currentState.ACLIPsetStore = m.ipsetStore.clone()
 	}
 
 	if err := m.stateManager.UpdateState(currentState); err != nil {

client/firewall/iptables/router_linux.go

@@ -4,6 +4,7 @@ package iptables
 
 import (
 	"fmt"
+	"maps"
 	"net/netip"
 	"strconv"
 	"strings"
@@ -749,11 +750,17 @@ func (r *router) updateState() {
 	currentState.Lock()
 	defer currentState.Unlock()
 
+	// Clone the rule map so the persisted state holds a private snapshot. The
+	// live map keeps being mutated by subsequent rule operations while the
+	// state manager marshals the state from its periodic-save goroutine.
+	// Sharing it by reference races the two and aborts the process with a
+	// concurrent map iteration and write. The ipset counter guards itself
+	// during marshaling, so it can be shared directly.
 	if r.v6 {
-		currentState.RouteRules6 = r.rules
+		currentState.RouteRules6 = maps.Clone(r.rules)
 		currentState.RouteIPsetCounter6 = r.ipsetCounter
 	} else {
-		currentState.RouteRules = r.rules
+		currentState.RouteRules = maps.Clone(r.rules)
 		currentState.RouteIPsetCounter = r.ipsetCounter
 	}
 

client/firewall/iptables/rulestore_linux.go

@@ -1,6 +1,9 @@
 package iptables
 
-import "encoding/json"
+import (
+	"encoding/json"
+	"maps"
+)
 
 type ipList struct {
 	ips map[string]struct{}
@@ -19,6 +22,14 @@ func (s *ipList) addIP(ip string) {
 	s.ips[ip] = struct{}{}
 }
 
+// clone returns a deep copy of the ipList with its own ips map.
+func (s *ipList) clone() *ipList {
+	if s == nil {
+		return nil
+	}
+	return &ipList{ips: maps.Clone(s.ips)}
+}
+
 // MarshalJSON implements json.Marshaler
 func (s *ipList) MarshalJSON() ([]byte, error) {
 	return json.Marshal(struct {
@@ -55,6 +66,19 @@ func newIpsetStore() *ipsetStore {
 	}
 }
 
+// clone returns a deep copy of the ipsetStore with its own ipsets map and
+// independent ipList entries.
+func (s *ipsetStore) clone() *ipsetStore {
+	if s == nil {
+		return nil
+	}
+	cloned := &ipsetStore{ipsets: make(map[string]*ipList, len(s.ipsets))}
+	for name, list := range s.ipsets {
+		cloned.ipsets[name] = list.clone()
+	}
+	return cloned
+}
+
 func (s *ipsetStore) ipset(ipsetName string) (*ipList, bool) {
 	r, ok := s.ipsets[ipsetName]
 	return r, ok

client/firewall/uspfilter/forwarder/icmp.go

@@ -362,6 +362,10 @@ func (f *Forwarder) injectICMPv6Reply(id stack.TransportEndpointID, icmpPayload
 		return 0
 	}
 
+	if pc := f.endpoint.capture.Load(); pc != nil {
+		(*pc).Offer(fullPacket, true)
+	}
+
 	return len(fullPacket)
 }
 

client/internal/connect.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"net"
 	"net/netip"
+	"path/filepath"
 	"runtime"
 	"runtime/debug"
 	"strings"
@@ -346,6 +347,11 @@ func (c *ConnectClient) run(mobileDependency MobileDependency, runningChan chan
 			return wrapErr(err)
 		}
 		engineConfig.TempDir = mobileDependency.TempDir
+		// Leave StateDir empty when there is no state path so a disk-backed
+		// syncstore falls back to os.TempDir() instead of filepath.Dir("") == ".".
+		if path != "" {
+			engineConfig.StateDir = filepath.Dir(path)
+		}
 
 		relayManager := relayClient.NewManager(engineCtx, relayURLs, myPrivateKey.PublicKey().String(), engineConfig.MTU)
 		c.statusRecorder.SetRelayMgr(relayManager)

client/internal/debug/debug.go

@@ -254,6 +254,8 @@ type BundleGenerator struct {
 	capturePath    string
 	refreshStatus  func() // Optional callback to refresh status before bundle generation
 	clientMetrics  MetricsExporter
+	daemonVersion  string
+	cliVersion     string
 
 	anonymize         bool
 	includeSystemInfo bool
@@ -278,6 +280,8 @@ type GeneratorDependencies struct {
 	CapturePath    string
 	RefreshStatus  func()
 	ClientMetrics  MetricsExporter
+	DaemonVersion  string
+	CliVersion     string
 }
 
 func NewBundleGenerator(deps GeneratorDependencies, cfg BundleConfig) *BundleGenerator {
@@ -299,6 +303,8 @@ func NewBundleGenerator(deps GeneratorDependencies, cfg BundleConfig) *BundleGen
 		capturePath:    deps.CapturePath,
 		refreshStatus:  deps.RefreshStatus,
 		clientMetrics:  deps.ClientMetrics,
+		daemonVersion:  deps.DaemonVersion,
+		cliVersion:     deps.CliVersion,
 
 		anonymize:         cfg.Anonymize,
 		includeSystemInfo: cfg.IncludeSystemInfo,
@@ -459,9 +465,11 @@ func (g *BundleGenerator) addStatus() error {
 		protoFullStatus := nbstatus.ToProtoFullStatus(fullStatus)
 		protoFullStatus.Events = g.statusRecorder.GetEventHistory()
 		overview := nbstatus.ConvertToStatusOutputOverview(protoFullStatus, nbstatus.ConvertOptions{
-			Anonymize:   g.anonymize,
-			ProfileName: profName,
+			Anonymize:     g.anonymize,
+			ProfileName:   profName,
+			DaemonVersion: g.daemonVersion,
 		})
+		overview.CliVersion = g.cliVersion
 		statusOutput := overview.FullDetailSummary()
 
 		statusReader := strings.NewReader(statusOutput)
@@ -798,6 +806,8 @@ func (g *BundleGenerator) addSyncResponse() error {
 		AllowPartial:    true,
 	}
 
+	g.maskSecrets()
+
 	jsonBytes, err := options.Marshal(g.syncResponse)
 	if err != nil {
 		return fmt.Errorf("generate json: %w", err)
@@ -810,6 +820,27 @@ func (g *BundleGenerator) addSyncResponse() error {
 	return nil
 }
 
+func (g *BundleGenerator) maskSecrets() {
+	if g.syncResponse == nil || g.syncResponse.NetbirdConfig == nil {
+		return
+	}
+
+	if g.syncResponse.NetbirdConfig.Flow != nil {
+		g.syncResponse.NetbirdConfig.Flow.TokenPayload = maskedValue
+
+	}
+
+	if g.syncResponse.NetbirdConfig.Relay != nil {
+		g.syncResponse.NetbirdConfig.Relay.TokenPayload = maskedValue
+	}
+
+	for i := range g.syncResponse.NetbirdConfig.Turns {
+		if g.syncResponse.NetbirdConfig.Turns[i] != nil {
+			g.syncResponse.NetbirdConfig.Turns[i].Password = maskedValue
+		}
+	}
+}
+
 func (g *BundleGenerator) addStateFile() error {
 	sm := profilemanager.NewServiceManager("")
 	path := sm.GetStatePath()
@@ -1039,7 +1070,8 @@ func (g *BundleGenerator) addRotatedLogFiles(logDir string) {
 		return
 	}
 
-	pattern := filepath.Join(logDir, "client-*.log.gz")
+	// This regex will match both logs rotated by us and logrotate on linux
+	pattern := filepath.Join(logDir, "client*.log.*")
 	files, err := filepath.Glob(pattern)
 	if err != nil {
 		log.Warnf("failed to glob rotated logs: %v", err)
@@ -1072,7 +1104,12 @@ func (g *BundleGenerator) addRotatedLogFiles(logDir string) {
 
 	for i := 0; i < maxFiles; i++ {
 		name := filepath.Base(files[i])
-		if err := g.addSingleLogFileGz(files[i], name); err != nil {
+		if strings.HasSuffix(name, ".gz") {
+			err = g.addSingleLogFileGz(files[i], name)
+		} else {
+			err = g.addSingleLogfile(files[i], name)
+		}
+		if err != nil {
 			log.Warnf("failed to add rotated log %s: %v", name, err)
 		}
 	}

client/internal/debug/debug_logfiles_test.go

@@ -0,0 +1,103 @@
+package debug
+
+import (
+	"archive/zip"
+	"bytes"
+	"compress/gzip"
+	"io"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+)
+
+// TestAddRotatedLogFiles_PicksUpAllVariants asserts that the rotated-log
+// glob picks up logs rotated by timberjack (gzipped) and by logrotate (plain
+// and gzipped), and skips unrelated files.
+func TestAddRotatedLogFiles_PicksUpAllVariants(t *testing.T) {
+	dir := t.TempDir()
+
+	writeFile(t, filepath.Join(dir, "client.log"), "active log\n")
+	writeFile(t, filepath.Join(dir, "other.log"), "unrelated\n")
+
+	timberjackRotated := "client-2026-05-21T10-30-45.000.log.gz"
+	writeGzFile(t, filepath.Join(dir, timberjackRotated), "timberjack rotated content\n")
+
+	logrotatePlain := "client.log.1"
+	writeFile(t, filepath.Join(dir, logrotatePlain), "logrotate plain content\n")
+
+	logrotateGz := "client.log.2.gz"
+	writeGzFile(t, filepath.Join(dir, logrotateGz), "logrotate gz content\n")
+
+	names := runAddRotatedLogFiles(t, dir, 10)
+
+	require.Contains(t, names, timberjackRotated, "timberjack rotated file should be in bundle")
+	require.Contains(t, names, logrotatePlain, "logrotate plain rotated file should be in bundle")
+	require.Contains(t, names, logrotateGz, "logrotate gzipped rotated file should be in bundle")
+	require.NotContains(t, names, "client.log", "active log should not be added by addRotatedLogFiles")
+	require.NotContains(t, names, "other.log", "unrelated files should not be in bundle")
+}
+
+// TestAddRotatedLogFiles_RespectsLogFileCount asserts that only the newest
+// logFileCount rotated files are bundled, ordered by mtime.
+func TestAddRotatedLogFiles_RespectsLogFileCount(t *testing.T) {
+	dir := t.TempDir()
+
+	oldest := filepath.Join(dir, "client.log.3")
+	middle := filepath.Join(dir, "client.log.2")
+	newest := filepath.Join(dir, "client.log.1")
+	writeFile(t, oldest, "old\n")
+	writeFile(t, middle, "mid\n")
+	writeFile(t, newest, "new\n")
+
+	now := time.Now()
+	require.NoError(t, os.Chtimes(oldest, now.Add(-2*time.Hour), now.Add(-2*time.Hour)))
+	require.NoError(t, os.Chtimes(middle, now.Add(-1*time.Hour), now.Add(-1*time.Hour)))
+	require.NoError(t, os.Chtimes(newest, now, now))
+
+	names := runAddRotatedLogFiles(t, dir, 2)
+
+	require.Contains(t, names, "client.log.1")
+	require.Contains(t, names, "client.log.2")
+	require.NotContains(t, names, "client.log.3", "oldest file should be dropped when logFileCount=2")
+}
+
+// runAddRotatedLogFiles calls addRotatedLogFiles against a fresh in-memory
+// zip writer and returns the set of entry names that ended up in the archive.
+func runAddRotatedLogFiles(t *testing.T, dir string, logFileCount uint32) map[string]struct{} {
+	t.Helper()
+
+	var buf bytes.Buffer
+	g := &BundleGenerator{
+		archive:      zip.NewWriter(&buf),
+		logFileCount: logFileCount,
+	}
+	g.addRotatedLogFiles(dir)
+	require.NoError(t, g.archive.Close())
+
+	zr, err := zip.NewReader(bytes.NewReader(buf.Bytes()), int64(buf.Len()))
+	require.NoError(t, err)
+
+	names := make(map[string]struct{}, len(zr.File))
+	for _, f := range zr.File {
+		names[f.Name] = struct{}{}
+	}
+	return names
+}
+
+func writeFile(t *testing.T, path, content string) {
+	t.Helper()
+	require.NoError(t, os.WriteFile(path, []byte(content), 0o644))
+}
+
+func writeGzFile(t *testing.T, path, content string) {
+	t.Helper()
+	var buf bytes.Buffer
+	gw := gzip.NewWriter(&buf)
+	_, err := io.WriteString(gw, content)
+	require.NoError(t, err)
+	require.NoError(t, gw.Close())
+	require.NoError(t, os.WriteFile(path, buf.Bytes(), 0o644))
+}

client/internal/dns/dnsfw/config.go

@@ -0,0 +1,63 @@
+package dnsfw
+
+import (
+	"os"
+	"strconv"
+	"strings"
+
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	// EnvDisable disables the DNS firewall entirely when set to a truthy value.
+	EnvDisable = "NB_DISABLE_DNS_FIREWALL"
+	// EnvPorts overrides the comma-separated list of remote ports to block.
+	// Empty disables the firewall.
+	EnvPorts = "NB_DNS_FIREWALL_PORTS"
+	// EnvStrict enables strict mode: permit DNS only to the virtual DNS IP
+	// and the netbird daemon. Default mode also permits anything on the
+	// netbird tunnel interface, which is safer if NRPT is silently ignored
+	// by Windows but lets apps reach custom DNS servers via the tunnel.
+	EnvStrict = "NB_DNS_FIREWALL_STRICT"
+)
+
+// defaultBlockedPorts are the well-known DNS ports we block for non-netbird
+// processes: 53 (plain DNS) and 853 (DNS-over-TLS).
+var defaultBlockedPorts = []uint16{53, 853}
+
+// blockedPorts returns the effective port list, honoring env overrides.
+// A nil return means the firewall should not be installed.
+func blockedPorts() []uint16 {
+	if disabled, _ := strconv.ParseBool(os.Getenv(EnvDisable)); disabled {
+		log.Infof("dns firewall disabled via %s", EnvDisable)
+		return nil
+	}
+
+	override, ok := os.LookupEnv(EnvPorts)
+	if !ok {
+		return defaultBlockedPorts
+	}
+
+	var ports []uint16
+	for _, raw := range strings.Split(override, ",") {
+		raw = strings.TrimSpace(raw)
+		if raw == "" {
+			continue
+		}
+		port, err := strconv.ParseUint(raw, 10, 16)
+		if err != nil {
+			log.Warnf("dns firewall: ignoring invalid port %q in %s: %v", raw, EnvPorts, err)
+			continue
+		}
+		if port == 0 {
+			log.Warnf("dns firewall: ignoring port 0 in %s", EnvPorts)
+			continue
+		}
+		ports = append(ports, uint16(port))
+	}
+	if len(ports) == 0 {
+		log.Infof("dns firewall disabled: %s yielded no valid ports", EnvPorts)
+		return nil
+	}
+	return ports
+}

client/internal/dns/dnsfw/config_test.go

@@ -0,0 +1,39 @@
+package dnsfw
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestBlockedPorts(t *testing.T) {
+	tests := []struct {
+		name     string
+		disable  string
+		ports    string
+		setPorts bool
+		want     []uint16
+	}{
+		{name: "default", want: defaultBlockedPorts},
+		{name: "disabled", disable: "true", want: nil},
+		{name: "disabled false keeps default", disable: "false", want: defaultBlockedPorts},
+		{name: "override single port", ports: "53", setPorts: true, want: []uint16{53}},
+		{name: "override multi", ports: "53, 853 ,5353", setPorts: true, want: []uint16{53, 853, 5353}},
+		{name: "override empty disables", ports: "", setPorts: true, want: nil},
+		{name: "override invalid skipped", ports: "53,not-a-port,853", setPorts: true, want: []uint16{53, 853}},
+		{name: "override zero skipped", ports: "53,0,853", setPorts: true, want: []uint16{53, 853}},
+		{name: "override only invalid disables", ports: "abc", setPorts: true, want: nil},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv(EnvDisable, tc.disable)
+			if tc.setPorts {
+				t.Setenv(EnvPorts, tc.ports)
+			}
+			got := blockedPorts()
+			if !reflect.DeepEqual(got, tc.want) {
+				t.Fatalf("blockedPorts() = %v, want %v", got, tc.want)
+			}
+		})
+	}
+}

client/internal/dns/dnsfw/dnsfw.go

@@ -0,0 +1,16 @@
+// Package dnsfw blocks DNS traffic from non-netbird processes when netbird is
+// managing the host's DNS, so that resolvers running on apps or libraries
+// outside netbird cannot bypass the configured DNS path.
+//
+// Implementation is Windows-only (uses WFP). On other platforms New returns
+// a no-op manager.
+package dnsfw
+
+import "net/netip"
+
+// Manager controls the per-tunnel DNS firewall. Both methods must be safe
+// to call multiple times.
+type Manager interface {
+	Enable(ifaceGUID string, virtualDNSIP netip.Addr) error
+	Disable() error
+}

client/internal/dns/dnsfw/dnsfw_other.go

@@ -0,0 +1,15 @@
+//go:build !windows
+
+package dnsfw
+
+import "net/netip"
+
+type noopManager struct{}
+
+func (noopManager) Enable(string, netip.Addr) error { return nil }
+func (noopManager) Disable() error                  { return nil }
+
+// New returns a no-op manager on non-Windows platforms.
+func New() Manager {
+	return noopManager{}
+}

client/internal/dns/dnsfw/dnsfw_windows.go

@@ -0,0 +1,144 @@
+//go:build windows
+
+package dnsfw
+
+import (
+	"fmt"
+	"net/netip"
+	"os"
+	"strconv"
+	"sync"
+	"unsafe"
+
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/sys/windows"
+)
+
+var (
+	modIphlpapi                    = windows.NewLazyDLL("iphlpapi.dll")
+	procConvertInterfaceGuidToLuid = modIphlpapi.NewProc("ConvertInterfaceGuidToLuid")
+)
+
+type windowsManager struct {
+	mu sync.Mutex
+	// session is the WFP engine handle. Zero when disabled.
+	session uintptr
+}
+
+// Enable installs the dns firewall. Strict mode propagates failures;
+// non-strict mode logs and returns nil so partial protection is preserved.
+func (m *windowsManager) Enable(ifaceGUID string, virtualDNSIP netip.Addr) error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	ports := blockedPorts()
+	if len(ports) == 0 {
+		return nil
+	}
+
+	if m.session != 0 {
+		if err := m.disableLocked(); err != nil {
+			return fmt.Errorf("reset existing dns firewall session: %w", err)
+		}
+	}
+
+	strict := strictMode()
+
+	luid, err := luidFromGUID(ifaceGUID)
+	if err != nil {
+		return m.failOrLog(strict, fmt.Errorf("resolve tun luid from guid %s: %w", ifaceGUID, err))
+	}
+
+	exe, err := os.Executable()
+	if err != nil {
+		return m.failOrLog(strict, fmt.Errorf("resolve daemon executable path: %w", err))
+	}
+
+	cfg := installConfig{
+		tunLUID:      luid,
+		daemonExe:    exe,
+		blockedPorts: ports,
+		strict:       strict,
+		virtualDNSIP: virtualDNSIP,
+	}
+	// session==0 signals a hard failure; non-zero with non-nil err is a partial install.
+	session, installErr := installFilters(cfg)
+	if session == 0 {
+		return m.failOrLog(strict, fmt.Errorf("install dns firewall filters: %w", installErr))
+	}
+
+	if installErr != nil && strict {
+		_ = closeSession(session)
+		return fmt.Errorf("strict dns firewall: partial install: %w", installErr)
+	}
+
+	m.session = session
+	log.Infof("dns firewall installed: iface=%s daemon=%s ports=%v strict=%v virtual_dns=%s",
+		ifaceGUID, exe, ports, strict, virtualDNSIP)
+	if installErr != nil {
+		log.Warnf("dns firewall partially installed (some filters failed): %v", installErr)
+	}
+	return nil
+}
+
+func (m *windowsManager) Disable() error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	return m.disableLocked()
+}
+
+func (m *windowsManager) disableLocked() error {
+	if m.session == 0 {
+		return nil
+	}
+	session := m.session
+	m.session = 0
+	if err := closeSession(session); err != nil {
+		return fmt.Errorf("close wfp session: %w", err)
+	}
+	log.Info("dns firewall removed")
+	return nil
+}
+
+// failOrLog returns err unchanged in strict mode. In non-strict mode the
+// error is logged and nil is returned.
+func (m *windowsManager) failOrLog(strict bool, err error) error {
+	if strict {
+		return err
+	}
+	log.Errorf("dns firewall: %v", err)
+	return nil
+}
+
+// New returns a Windows DNS firewall manager backed by WFP.
+func New() Manager {
+	return &windowsManager{}
+}
+
+// strictMode reports whether strict mode is enabled via env.
+func strictMode() bool {
+	v, _ := strconv.ParseBool(os.Getenv(EnvStrict))
+	return v
+}
+
+// luidFromGUID converts a Windows interface GUID string to its LUID.
+func luidFromGUID(ifaceGUID string) (luid uint64, err error) {
+	defer func() {
+		if r := recover(); r != nil {
+			err = fmt.Errorf("panic in luidFromGUID: %v", r)
+		}
+	}()
+
+	guid, err := windows.GUIDFromString(ifaceGUID)
+	if err != nil {
+		return 0, fmt.Errorf("parse guid: %w", err)
+	}
+	rc, _, _ := procConvertInterfaceGuidToLuid.Call(
+		uintptr(unsafe.Pointer(&guid)),
+		uintptr(unsafe.Pointer(&luid)),
+	)
+	if rc != 0 {
+		return 0, fmt.Errorf("ConvertInterfaceGuidToLuid returned %d", rc)
+	}
+	return luid, nil
+}

client/internal/dns/dnsfw/dnsfw_windows_test.go

@@ -0,0 +1,72 @@
+//go:build windows
+
+package dnsfw
+
+import (
+	"net/netip"
+	"os"
+	"testing"
+)
+
+func TestStrictMode(t *testing.T) {
+	tests := []struct {
+		name string
+		val  string
+		set  bool
+		want bool
+	}{
+		{name: "unset", want: false},
+		{name: "true", val: "true", set: true, want: true},
+		{name: "1", val: "1", set: true, want: true},
+		{name: "false", val: "false", set: true, want: false},
+		{name: "invalid is false", val: "garbage", set: true, want: false},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv(EnvStrict, tc.val)
+			if !tc.set {
+				os.Unsetenv(EnvStrict)
+			}
+			if got := strictMode(); got != tc.want {
+				t.Fatalf("strictMode() = %v, want %v", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestWindowsManagerDisableIdempotent(t *testing.T) {
+	m := &windowsManager{}
+	if err := m.Disable(); err != nil {
+		t.Fatalf("first Disable on fresh manager: %v", err)
+	}
+	if err := m.Disable(); err != nil {
+		t.Fatalf("second Disable on fresh manager: %v", err)
+	}
+	if m.session != 0 {
+		t.Fatalf("session should remain zero, got %d", m.session)
+	}
+}
+
+func TestWindowsManagerEnableNoOpWhenDisabledByEnv(t *testing.T) {
+	t.Setenv(EnvDisable, "true")
+
+	m := &windowsManager{}
+	if err := m.Enable("00000000-0000-0000-0000-000000000000", netip.Addr{}); err != nil {
+		t.Fatalf("Enable should be a no-op when firewall disabled by env: %v", err)
+	}
+	if m.session != 0 {
+		t.Fatalf("session must remain zero when env disables firewall, got %d", m.session)
+	}
+}
+
+func TestWindowsManagerEnableNoOpWhenPortsEmpty(t *testing.T) {
+	t.Setenv(EnvPorts, "")
+
+	m := &windowsManager{}
+	if err := m.Enable("00000000-0000-0000-0000-000000000000", netip.Addr{}); err != nil {
+		t.Fatalf("Enable should be a no-op when ports list is empty: %v", err)
+	}
+	if m.session != 0 {
+		t.Fatalf("session must remain zero when ports list is empty, got %d", m.session)
+	}
+}

client/internal/dns/dnsfw/helpers_windows.go

@@ -0,0 +1,53 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (C) 2019-2021 WireGuard LLC. All Rights Reserved.
+ *
+ * Adapted from wireguard-windows tunnel/firewall/helpers.go.
+ */
+
+package dnsfw
+
+import (
+	"errors"
+	"fmt"
+	"runtime"
+	"syscall"
+
+	"golang.org/x/sys/windows"
+)
+
+func createWtFwpmDisplayData0(name, description string) (*wtFwpmDisplayData0, error) {
+	namePtr, err := windows.UTF16PtrFromString(name)
+	if err != nil {
+		return nil, wrapErr(err)
+	}
+
+	descriptionPtr, err := windows.UTF16PtrFromString(description)
+	if err != nil {
+		return nil, wrapErr(err)
+	}
+
+	return &wtFwpmDisplayData0{
+		name:        namePtr,
+		description: descriptionPtr,
+	}, nil
+}
+
+func filterWeight(weight uint8) wtFwpValue0 {
+	return wtFwpValue0{
+		_type: cFWP_UINT8,
+		value: uintptr(weight),
+	}
+}
+
+func wrapErr(err error) error {
+	var errno syscall.Errno
+	if !errors.As(err, &errno) {
+		return err
+	}
+	_, file, line, ok := runtime.Caller(1)
+	if !ok {
+		return fmt.Errorf("wfp error at unknown location: %w", err)
+	}
+	return fmt.Errorf("wfp error at %s:%d: %w", file, line, err)
+}

client/internal/dns/dnsfw/rules_windows.go

@@ -0,0 +1,249 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (C) 2019-2021 WireGuard LLC. All Rights Reserved.
+ * Copyright (C) 2026 NetBird GmbH. All Rights Reserved.
+ *
+ * Filter installers adapted from wireguard-windows tunnel/firewall/rules.go.
+ * The block-DNS approach (port 53 + UDP/TCP) matches what wireguard-windows
+ * uses for its kill-switch DNS leak protection. We extend it with a
+ * configurable port set so we also cover :853 (DoT) and any future ports.
+ */
+
+package dnsfw
+
+import (
+	"encoding/binary"
+	"fmt"
+	"net/netip"
+	"unsafe"
+
+	"github.com/hashicorp/go-multierror"
+	"golang.org/x/sys/windows"
+
+	nberrors "github.com/netbirdio/netbird/client/errors"
+)
+
+// Filters install at outbound ALE_AUTH_CONNECT layers only; inbound replies
+// follow the authorized outbound flow.
+
+// permitTunInterface installs a permit filter for any traffic whose local
+// interface is the netbird tunnel.
+func permitTunInterface(session uintptr, base *baseObjects, weight uint8, ifLUID uint64) error {
+	cond := wtFwpmFilterCondition0{
+		fieldKey:  cFWPM_CONDITION_IP_LOCAL_INTERFACE,
+		matchType: cFWP_MATCH_EQUAL,
+		conditionValue: wtFwpConditionValue0{
+			_type: cFWP_UINT64,
+			value: uintptr(unsafe.Pointer(&ifLUID)),
+		},
+	}
+
+	filter := wtFwpmFilter0{
+		providerKey:         &base.provider,
+		subLayerKey:         base.filters,
+		weight:              filterWeight(weight),
+		numFilterConditions: 1,
+		filterCondition:     (*wtFwpmFilterCondition0)(unsafe.Pointer(&cond)),
+		action:              wtFwpmAction0{_type: cFWP_ACTION_PERMIT},
+	}
+
+	return addOutboundFilters(session, &filter, "Permit netbird tunnel")
+}
+
+// permitDaemonByAppID installs a permit filter matching the netbird daemon
+// executable by App-ID. App-ID alone is sufficient because netbird.exe is a
+// dedicated binary.
+func permitDaemonByAppID(session uintptr, base *baseObjects, daemonExe string, weight uint8) error {
+	appID, err := daemonAppID(daemonExe)
+	if err != nil {
+		return err
+	}
+	defer fwpmFreeMemory0(unsafe.Pointer(&appID))
+
+	cond := wtFwpmFilterCondition0{
+		fieldKey:  cFWPM_CONDITION_ALE_APP_ID,
+		matchType: cFWP_MATCH_EQUAL,
+		conditionValue: wtFwpConditionValue0{
+			_type: cFWP_BYTE_BLOB_TYPE,
+			value: uintptr(unsafe.Pointer(appID)),
+		},
+	}
+
+	filter := wtFwpmFilter0{
+		providerKey:         &base.provider,
+		subLayerKey:         base.filters,
+		weight:              filterWeight(weight),
+		numFilterConditions: 1,
+		filterCondition:     (*wtFwpmFilterCondition0)(unsafe.Pointer(&cond)),
+		action:              wtFwpmAction0{_type: cFWP_ACTION_PERMIT},
+	}
+
+	return addOutboundFilters(session, &filter, "Permit netbird daemon")
+}
+
+// permitVirtualDNSIP installs a permit filter for DNS-port traffic destined
+// for the in-tunnel virtual DNS IP. Used in strict mode in lieu of
+// permitTunInterface.
+func permitVirtualDNSIP(session uintptr, base *baseObjects, ip netip.Addr, ports []uint16, weight uint8) error {
+	var merr *multierror.Error
+	for _, port := range ports {
+		if err := permitDNSToHost(session, base, ip, port, weight); err != nil {
+			merr = multierror.Append(merr, fmt.Errorf("permit %s:%d: %w", ip, port, err))
+		}
+	}
+	return nberrors.FormatErrorOrNil(merr)
+}
+
+func permitDNSToHost(session uintptr, base *baseObjects, ip netip.Addr, port uint16, weight uint8) error {
+	if !ip.IsValid() {
+		return fmt.Errorf("invalid address")
+	}
+
+	var addrCond wtFwpmFilterCondition0
+	var layer windows.GUID
+	// v6 backing must outlive fwpmFilterAdd0; keep it on this stack frame.
+	var v6 wtFwpByteArray16
+
+	if ip.Is4() {
+		v4 := ip.As4()
+		addrCond = wtFwpmFilterCondition0{
+			fieldKey:  cFWPM_CONDITION_IP_REMOTE_ADDRESS,
+			matchType: cFWP_MATCH_EQUAL,
+			conditionValue: wtFwpConditionValue0{
+				_type: cFWP_UINT32,
+				value: uintptr(binary.BigEndian.Uint32(v4[:])),
+			},
+		}
+		layer = cFWPM_LAYER_ALE_AUTH_CONNECT_V4
+	} else {
+		v6 = wtFwpByteArray16{byteArray16: ip.As16()}
+		addrCond = wtFwpmFilterCondition0{
+			fieldKey:  cFWPM_CONDITION_IP_REMOTE_ADDRESS,
+			matchType: cFWP_MATCH_EQUAL,
+			conditionValue: wtFwpConditionValue0{
+				_type: cFWP_BYTE_ARRAY16_TYPE,
+				value: uintptr(unsafe.Pointer(&v6)),
+			},
+		}
+		layer = cFWPM_LAYER_ALE_AUTH_CONNECT_V6
+	}
+
+	conditions := [2]wtFwpmFilterCondition0{
+		addrCond,
+		{
+			fieldKey:  cFWPM_CONDITION_IP_REMOTE_PORT,
+			matchType: cFWP_MATCH_EQUAL,
+			conditionValue: wtFwpConditionValue0{
+				_type: cFWP_UINT16,
+				value: uintptr(port),
+			},
+		},
+	}
+	filter := wtFwpmFilter0{
+		providerKey:         &base.provider,
+		subLayerKey:         base.filters,
+		weight:              filterWeight(weight),
+		numFilterConditions: uint32(len(conditions)),
+		filterCondition:     (*wtFwpmFilterCondition0)(unsafe.Pointer(&conditions[0])),
+		action:              wtFwpmAction0{_type: cFWP_ACTION_PERMIT},
+	}
+
+	display, err := createWtFwpmDisplayData0(fmt.Sprintf("Permit DNS to %s:%d", ip, port), "")
+	if err != nil {
+		return wrapErr(err)
+	}
+	filter.displayData = *display
+	filter.layerKey = layer
+
+	var filterID uint64
+	if err := fwpmFilterAdd0(session, &filter, 0, &filterID); err != nil {
+		return wrapErr(err)
+	}
+	_ = v6
+	return nil
+}
+
+// blockDNSPorts installs a deny filter for outbound traffic to each of the
+// given remote ports over UDP or TCP. Per-port and per-layer failures are
+// accumulated; partial coverage is preferred over zero coverage.
+func blockDNSPorts(session uintptr, base *baseObjects, ports []uint16, weight uint8) error {
+	var merr *multierror.Error
+	for _, port := range ports {
+		if err := blockDNSPort(session, base, port, weight); err != nil {
+			merr = multierror.Append(merr, fmt.Errorf("block port %d: %w", port, err))
+		}
+	}
+	return nberrors.FormatErrorOrNil(merr)
+}
+
+func blockDNSPort(session uintptr, base *baseObjects, port uint16, weight uint8) error {
+	conditions := [3]wtFwpmFilterCondition0{
+		{
+			fieldKey:  cFWPM_CONDITION_IP_REMOTE_PORT,
+			matchType: cFWP_MATCH_EQUAL,
+			conditionValue: wtFwpConditionValue0{
+				_type: cFWP_UINT16,
+				value: uintptr(port),
+			},
+		},
+		{
+			fieldKey:  cFWPM_CONDITION_IP_PROTOCOL,
+			matchType: cFWP_MATCH_EQUAL,
+			conditionValue: wtFwpConditionValue0{
+				_type: cFWP_UINT8,
+				value: uintptr(cIPPROTO_UDP),
+			},
+		},
+		// Repeat the IP_PROTOCOL condition for logical OR with TCP.
+		{
+			fieldKey:  cFWPM_CONDITION_IP_PROTOCOL,
+			matchType: cFWP_MATCH_EQUAL,
+			conditionValue: wtFwpConditionValue0{
+				_type: cFWP_UINT8,
+				value: uintptr(cIPPROTO_TCP),
+			},
+		},
+	}
+
+	filter := wtFwpmFilter0{
+		providerKey:         &base.provider,
+		subLayerKey:         base.filters,
+		weight:              filterWeight(weight),
+		numFilterConditions: uint32(len(conditions)),
+		filterCondition:     (*wtFwpmFilterCondition0)(unsafe.Pointer(&conditions[0])),
+		action:              wtFwpmAction0{_type: cFWP_ACTION_BLOCK},
+	}
+
+	return addOutboundFilters(session, &filter, fmt.Sprintf("Block DNS port %d", port))
+}
+
+// addOutboundFilters installs the same filter on the v4 and v6 outbound ALE
+// connect layers. v4 and v6 are installed independently: failure on one
+// layer does not abort the other, and the accumulated errors are returned.
+// Partial coverage is preferred over zero coverage.
+func addOutboundFilters(session uintptr, filter *wtFwpmFilter0, name string) error {
+	layers := [...]struct {
+		layer windows.GUID
+		label string
+	}{
+		{cFWPM_LAYER_ALE_AUTH_CONNECT_V4, name + " (IPv4)"},
+		{cFWPM_LAYER_ALE_AUTH_CONNECT_V6, name + " (IPv6)"},
+	}
+
+	var merr *multierror.Error
+	for _, l := range layers {
+		display, err := createWtFwpmDisplayData0(l.label, "")
+		if err != nil {
+			merr = multierror.Append(merr, fmt.Errorf("%s: %w", l.label, wrapErr(err)))
+			continue
+		}
+		filter.displayData = *display
+		filter.layerKey = l.layer
+
+		var filterID uint64
+		if err := fwpmFilterAdd0(session, filter, 0, &filterID); err != nil {
+			merr = multierror.Append(merr, fmt.Errorf("%s: %w", l.label, wrapErr(err)))
+		}
+	}
+	return nberrors.FormatErrorOrNil(merr)
+}

client/internal/dns/dnsfw/session_windows.go

@@ -0,0 +1,177 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (C) 2019-2021 WireGuard LLC. All Rights Reserved.
+ * Copyright (C) 2026 NetBird GmbH. All Rights Reserved.
+ *
+ * Session lifecycle and the high-level Install/Close entry points adapted
+ * from wireguard-windows tunnel/firewall.
+ */
+
+package dnsfw
+
+import (
+	"errors"
+	"fmt"
+	"net/netip"
+	"unsafe"
+
+	"github.com/hashicorp/go-multierror"
+	"golang.org/x/sys/windows"
+
+	nberrors "github.com/netbirdio/netbird/client/errors"
+)
+
+// installConfig is the input to installFilters.
+type installConfig struct {
+	tunLUID      uint64
+	daemonExe    string
+	blockedPorts []uint16
+	// strict, when true, narrows the carve-out from "anything on tun" to
+	// "DNS only to virtualDNSIP". virtualDNSIP must be valid in this case.
+	strict       bool
+	virtualDNSIP netip.Addr
+}
+
+// baseObjects holds the GUIDs of the WFP provider and sublayer registered
+// for our session. Both are randomly generated per session.
+type baseObjects struct {
+	provider windows.GUID
+	filters  windows.GUID
+}
+
+// installFilters opens a dynamic WFP session and installs the netbird DNS
+// firewall filters. Returns a zero session on hard failure (session create,
+// base objects); a non-zero session with a non-nil error is a partial install
+// (some per-filter installs failed) and is safe to close.
+func installFilters(cfg installConfig) (session uintptr, err error) {
+	defer func() {
+		if r := recover(); r != nil {
+			// Dynamic session: kernel will clean up on process exit even
+			// if we leave the handle dangling here.
+			err = fmt.Errorf("panic in installFilters: %v", r)
+		}
+	}()
+
+	if len(cfg.blockedPorts) == 0 {
+		return 0, errors.New("dns firewall: no blocked ports configured")
+	}
+	if cfg.strict && !cfg.virtualDNSIP.IsValid() {
+		return 0, errors.New("dns firewall: strict mode requires a valid virtual DNS IP")
+	}
+
+	session, err = createSession()
+	if err != nil {
+		return 0, err
+	}
+
+	base, err := registerBaseObjects(session)
+	if err != nil {
+		_ = fwpmEngineClose0(session)
+		return 0, fmt.Errorf("register base objects: %w", err)
+	}
+
+	var merr *multierror.Error
+	if cfg.strict {
+		if err := permitVirtualDNSIP(session, base, cfg.virtualDNSIP, cfg.blockedPorts, 15); err != nil {
+			merr = multierror.Append(merr, fmt.Errorf("permit virtual dns: %w", err))
+		}
+	} else {
+		if err := permitTunInterface(session, base, 15, cfg.tunLUID); err != nil {
+			merr = multierror.Append(merr, fmt.Errorf("permit tun interface: %w", err))
+		}
+	}
+	if err := permitDaemonByAppID(session, base, cfg.daemonExe, 14); err != nil {
+		merr = multierror.Append(merr, fmt.Errorf("permit netbird daemon: %w", err))
+	}
+	if err := blockDNSPorts(session, base, cfg.blockedPorts, 10); err != nil {
+		merr = multierror.Append(merr, fmt.Errorf("block dns ports: %w", err))
+	}
+
+	return session, nberrors.FormatErrorOrNil(merr)
+}
+
+// closeSession tears down a WFP session previously opened by installFilters.
+// All filters owned by the session are removed.
+func closeSession(session uintptr) (err error) {
+	defer func() {
+		if r := recover(); r != nil {
+			err = fmt.Errorf("panic in closeSession: %v", r)
+		}
+	}()
+
+	if session == 0 {
+		return nil
+	}
+	if err := fwpmEngineClose0(session); err != nil {
+		return wrapErr(err)
+	}
+	return nil
+}
+
+func createSession() (uintptr, error) {
+	displayData, err := createWtFwpmDisplayData0("NetBird DNS firewall", "NetBird DNS firewall dynamic session")
+	if err != nil {
+		return 0, wrapErr(err)
+	}
+	session := wtFwpmSession0{
+		displayData:          *displayData,
+		flags:                cFWPM_SESSION_FLAG_DYNAMIC,
+		txnWaitTimeoutInMSec: windows.INFINITE,
+	}
+	var handle uintptr
+	if err := fwpmEngineOpen0(nil, cRPC_C_AUTHN_WINNT, nil, &session, unsafe.Pointer(&handle)); err != nil {
+		return 0, wrapErr(err)
+	}
+	return handle, nil
+}
+
+func registerBaseObjects(session uintptr) (*baseObjects, error) {
+	bo := &baseObjects{}
+	var err error
+	if bo.provider, err = windows.GenerateGUID(); err != nil {
+		return nil, wrapErr(err)
+	}
+	if bo.filters, err = windows.GenerateGUID(); err != nil {
+		return nil, wrapErr(err)
+	}
+
+	displayData, err := createWtFwpmDisplayData0("NetBird DNS firewall", "NetBird DNS firewall provider")
+	if err != nil {
+		return nil, wrapErr(err)
+	}
+	provider := wtFwpmProvider0{
+		providerKey: bo.provider,
+		displayData: *displayData,
+	}
+	if err := fwpmProviderAdd0(session, &provider, 0); err != nil {
+		return nil, wrapErr(err)
+	}
+
+	subDisplay, err := createWtFwpmDisplayData0("NetBird DNS firewall filters", "Permit and block filters")
+	if err != nil {
+		return nil, wrapErr(err)
+	}
+	sublayer := wtFwpmSublayer0{
+		subLayerKey: bo.filters,
+		displayData: *subDisplay,
+		providerKey: &bo.provider,
+		weight:      ^uint16(0),
+	}
+	if err := fwpmSubLayerAdd0(session, &sublayer, 0); err != nil {
+		return nil, wrapErr(err)
+	}
+	return bo, nil
+}
+
+// daemonAppID returns the WFP App-ID byte blob for the given executable path.
+func daemonAppID(path string) (*wtFwpByteBlob, error) {
+	pathPtr, err := windows.UTF16PtrFromString(path)
+	if err != nil {
+		return nil, wrapErr(err)
+	}
+	var appID *wtFwpByteBlob
+	if err := fwpmGetAppIdFromFileName0(pathPtr, unsafe.Pointer(&appID)); err != nil {
+		return nil, wrapErr(err)
+	}
+	return appID, nil
+}

client/internal/dns/dnsfw/syscall_windows.go

@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (C) 2019-2021 WireGuard LLC. All Rights Reserved.
+ *
+ * Adapted from wireguard-windows tunnel/firewall/syscall_windows.go.
+ */
+
+package dnsfw
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmengineopen0
+//sys	fwpmEngineOpen0(serverName *uint16, authnService wtRpcCAuthN, authIdentity *uintptr, session *wtFwpmSession0, engineHandle unsafe.Pointer) (err error) [failretval!=0] = fwpuclnt.FwpmEngineOpen0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmengineclose0
+//sys	fwpmEngineClose0(engineHandle uintptr) (err error) [failretval!=0] = fwpuclnt.FwpmEngineClose0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmsublayeradd0
+//sys	fwpmSubLayerAdd0(engineHandle uintptr, subLayer *wtFwpmSublayer0, sd uintptr) (err error) [failretval!=0] = fwpuclnt.FwpmSubLayerAdd0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmgetappidfromfilename0
+//sys	fwpmGetAppIdFromFileName0(fileName *uint16, appID unsafe.Pointer) (err error) [failretval!=0] = fwpuclnt.FwpmGetAppIdFromFileName0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmfreememory0
+//sys	fwpmFreeMemory0(p unsafe.Pointer) = fwpuclnt.FwpmFreeMemory0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmfilteradd0
+//sys	fwpmFilterAdd0(engineHandle uintptr, filter *wtFwpmFilter0, sd uintptr, id *uint64) (err error) [failretval!=0] = fwpuclnt.FwpmFilterAdd0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/Fwpmu/nf-fwpmu-fwpmtransactionbegin0
+//sys	fwpmTransactionBegin0(engineHandle uintptr, flags uint32) (err error) [failretval!=0] = fwpuclnt.FwpmTransactionBegin0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmtransactioncommit0
+//sys	fwpmTransactionCommit0(engineHandle uintptr) (err error) [failretval!=0] = fwpuclnt.FwpmTransactionCommit0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmtransactionabort0
+//sys	fwpmTransactionAbort0(engineHandle uintptr) (err error) [failretval!=0] = fwpuclnt.FwpmTransactionAbort0
+
+// https://docs.microsoft.com/en-us/windows/desktop/api/fwpmu/nf-fwpmu-fwpmprovideradd0
+//sys	fwpmProviderAdd0(engineHandle uintptr, provider *wtFwpmProvider0, sd uintptr) (err error) [failretval!=0] = fwpuclnt.FwpmProviderAdd0

client/internal/dns/dnsfw/types_windows.go

@@ -0,0 +1,414 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (C) 2019-2021 WireGuard LLC. All Rights Reserved.
+ *
+ * Adapted from wireguard-windows tunnel/firewall/types_windows.go.
+ */
+
+package dnsfw
+
+import "golang.org/x/sys/windows"
+
+const (
+	anysizeArray = 1 // ANYSIZE_ARRAY defined in winnt.h
+
+	wtFwpBitmapArray64_Size = 8
+
+	wtFwpByteArray16_Size = 16
+
+	wtFwpByteArray6_Size = 6
+
+	wtFwpmAction0_Size              = 20
+	wtFwpmAction0_filterType_Offset = 4
+
+	wtFwpV4AddrAndMask_Size        = 8
+	wtFwpV4AddrAndMask_mask_Offset = 4
+
+	wtFwpV6AddrAndMask_Size                = 17
+	wtFwpV6AddrAndMask_prefixLength_Offset = 16
+)
+
+type wtFwpActionFlag uint32
+
+const (
+	cFWP_ACTION_FLAG_TERMINATING     wtFwpActionFlag = 0x00001000
+	cFWP_ACTION_FLAG_NON_TERMINATING wtFwpActionFlag = 0x00002000
+	cFWP_ACTION_FLAG_CALLOUT         wtFwpActionFlag = 0x00004000
+)
+
+// FWP_ACTION_TYPE defined in fwptypes.h
+type wtFwpActionType uint32
+
+const (
+	cFWP_ACTION_BLOCK               wtFwpActionType = wtFwpActionType(0x00000001 | cFWP_ACTION_FLAG_TERMINATING)
+	cFWP_ACTION_PERMIT              wtFwpActionType = wtFwpActionType(0x00000002 | cFWP_ACTION_FLAG_TERMINATING)
+	cFWP_ACTION_CALLOUT_TERMINATING wtFwpActionType = wtFwpActionType(0x00000003 | cFWP_ACTION_FLAG_CALLOUT | cFWP_ACTION_FLAG_TERMINATING)
+	cFWP_ACTION_CALLOUT_INSPECTION  wtFwpActionType = wtFwpActionType(0x00000004 | cFWP_ACTION_FLAG_CALLOUT | cFWP_ACTION_FLAG_NON_TERMINATING)
+	cFWP_ACTION_CALLOUT_UNKNOWN     wtFwpActionType = wtFwpActionType(0x00000005 | cFWP_ACTION_FLAG_CALLOUT)
+	cFWP_ACTION_CONTINUE            wtFwpActionType = wtFwpActionType(0x00000006 | cFWP_ACTION_FLAG_NON_TERMINATING)
+	cFWP_ACTION_NONE                wtFwpActionType = 0x00000007
+	cFWP_ACTION_NONE_NO_MATCH       wtFwpActionType = 0x00000008
+	cFWP_ACTION_BITMAP_INDEX_SET    wtFwpActionType = 0x00000009
+)
+
+// FWP_BYTE_BLOB defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ns-fwptypes-fwp_byte_blob_)
+type wtFwpByteBlob struct {
+	size uint32
+	data *uint8
+}
+
+// FWP_MATCH_TYPE defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ne-fwptypes-fwp_match_type_)
+type wtFwpMatchType uint32
+
+const (
+	cFWP_MATCH_EQUAL                  wtFwpMatchType = 0
+	cFWP_MATCH_GREATER                wtFwpMatchType = cFWP_MATCH_EQUAL + 1
+	cFWP_MATCH_LESS                   wtFwpMatchType = cFWP_MATCH_GREATER + 1
+	cFWP_MATCH_GREATER_OR_EQUAL       wtFwpMatchType = cFWP_MATCH_LESS + 1
+	cFWP_MATCH_LESS_OR_EQUAL          wtFwpMatchType = cFWP_MATCH_GREATER_OR_EQUAL + 1
+	cFWP_MATCH_RANGE                  wtFwpMatchType = cFWP_MATCH_LESS_OR_EQUAL + 1
+	cFWP_MATCH_FLAGS_ALL_SET          wtFwpMatchType = cFWP_MATCH_RANGE + 1
+	cFWP_MATCH_FLAGS_ANY_SET          wtFwpMatchType = cFWP_MATCH_FLAGS_ALL_SET + 1
+	cFWP_MATCH_FLAGS_NONE_SET         wtFwpMatchType = cFWP_MATCH_FLAGS_ANY_SET + 1
+	cFWP_MATCH_EQUAL_CASE_INSENSITIVE wtFwpMatchType = cFWP_MATCH_FLAGS_NONE_SET + 1
+	cFWP_MATCH_NOT_EQUAL              wtFwpMatchType = cFWP_MATCH_EQUAL_CASE_INSENSITIVE + 1
+	cFWP_MATCH_PREFIX                 wtFwpMatchType = cFWP_MATCH_NOT_EQUAL + 1
+	cFWP_MATCH_NOT_PREFIX             wtFwpMatchType = cFWP_MATCH_PREFIX + 1
+	cFWP_MATCH_TYPE_MAX               wtFwpMatchType = cFWP_MATCH_NOT_PREFIX + 1
+)
+
+// FWPM_ACTION0 defined in fwpmtypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwpmtypes/ns-fwpmtypes-fwpm_action0_)
+type wtFwpmAction0 struct {
+	_type      wtFwpActionType
+	filterType windows.GUID // Windows type: GUID
+}
+
+// Defined in fwpmu.h. 4cd62a49-59c3-4969-b7f3-bda5d32890a4
+var cFWPM_CONDITION_IP_LOCAL_INTERFACE = windows.GUID{
+	Data1: 0x4cd62a49,
+	Data2: 0x59c3,
+	Data3: 0x4969,
+	Data4: [8]byte{0xb7, 0xf3, 0xbd, 0xa5, 0xd3, 0x28, 0x90, 0xa4},
+}
+
+// Defined in fwpmu.h. b235ae9a-1d64-49b8-a44c-5ff3d9095045
+var cFWPM_CONDITION_IP_REMOTE_ADDRESS = windows.GUID{
+	Data1: 0xb235ae9a,
+	Data2: 0x1d64,
+	Data3: 0x49b8,
+	Data4: [8]byte{0xa4, 0x4c, 0x5f, 0xf3, 0xd9, 0x09, 0x50, 0x45},
+}
+
+// Defined in fwpmu.h. 3971ef2b-623e-4f9a-8cb1-6e79b806b9a7
+var cFWPM_CONDITION_IP_PROTOCOL = windows.GUID{
+	Data1: 0x3971ef2b,
+	Data2: 0x623e,
+	Data3: 0x4f9a,
+	Data4: [8]byte{0x8c, 0xb1, 0x6e, 0x79, 0xb8, 0x06, 0xb9, 0xa7},
+}
+
+// Defined in fwpmu.h. 0c1ba1af-5765-453f-af22-a8f791ac775b
+var cFWPM_CONDITION_IP_LOCAL_PORT = windows.GUID{
+	Data1: 0x0c1ba1af,
+	Data2: 0x5765,
+	Data3: 0x453f,
+	Data4: [8]byte{0xaf, 0x22, 0xa8, 0xf7, 0x91, 0xac, 0x77, 0x5b},
+}
+
+// Defined in fwpmu.h. c35a604d-d22b-4e1a-91b4-68f674ee674b
+var cFWPM_CONDITION_IP_REMOTE_PORT = windows.GUID{
+	Data1: 0xc35a604d,
+	Data2: 0xd22b,
+	Data3: 0x4e1a,
+	Data4: [8]byte{0x91, 0xb4, 0x68, 0xf6, 0x74, 0xee, 0x67, 0x4b},
+}
+
+// Defined in fwpmu.h. d78e1e87-8644-4ea5-9437-d809ecefc971
+var cFWPM_CONDITION_ALE_APP_ID = windows.GUID{
+	Data1: 0xd78e1e87,
+	Data2: 0x8644,
+	Data3: 0x4ea5,
+	Data4: [8]byte{0x94, 0x37, 0xd8, 0x09, 0xec, 0xef, 0xc9, 0x71},
+}
+
+// af043a0a-b34d-4f86-979c-c90371af6e66
+var cFWPM_CONDITION_ALE_USER_ID = windows.GUID{
+	Data1: 0xaf043a0a,
+	Data2: 0xb34d,
+	Data3: 0x4f86,
+	Data4: [8]byte{0x97, 0x9c, 0xc9, 0x03, 0x71, 0xaf, 0x6e, 0x66},
+}
+
+// d9ee00de-c1ef-4617-bfe3-ffd8f5a08957
+var cFWPM_CONDITION_IP_LOCAL_ADDRESS = windows.GUID{
+	Data1: 0xd9ee00de,
+	Data2: 0xc1ef,
+	Data3: 0x4617,
+	Data4: [8]byte{0xbf, 0xe3, 0xff, 0xd8, 0xf5, 0xa0, 0x89, 0x57},
+}
+
+var (
+	cFWPM_CONDITION_ICMP_TYPE = cFWPM_CONDITION_IP_LOCAL_PORT
+	cFWPM_CONDITION_ICMP_CODE = cFWPM_CONDITION_IP_REMOTE_PORT
+)
+
+// 7bc43cbf-37ba-45f1-b74a-82ff518eeb10
+var cFWPM_CONDITION_L2_FLAGS = windows.GUID{
+	Data1: 0x7bc43cbf,
+	Data2: 0x37ba,
+	Data3: 0x45f1,
+	Data4: [8]byte{0xb7, 0x4a, 0x82, 0xff, 0x51, 0x8e, 0xeb, 0x10},
+}
+
+type wtFwpmL2Flags uint32
+
+const cFWP_CONDITION_L2_IS_VM2VM wtFwpmL2Flags = 0x00000010
+
+var cFWPM_CONDITION_FLAGS = windows.GUID{
+	Data1: 0x632ce23b,
+	Data2: 0x5167,
+	Data3: 0x435c,
+	Data4: [8]byte{0x86, 0xd7, 0xe9, 0x03, 0x68, 0x4a, 0xa8, 0x0c},
+}
+
+type wtFwpmFlags uint32
+
+const cFWP_CONDITION_FLAG_IS_LOOPBACK wtFwpmFlags = 0x00000001
+
+// Defined in fwpmtypes.h
+type wtFwpmFilterFlags uint32
+
+const (
+	cFWPM_FILTER_FLAG_NONE                                wtFwpmFilterFlags = 0x00000000
+	cFWPM_FILTER_FLAG_PERSISTENT                          wtFwpmFilterFlags = 0x00000001
+	cFWPM_FILTER_FLAG_BOOTTIME                            wtFwpmFilterFlags = 0x00000002
+	cFWPM_FILTER_FLAG_HAS_PROVIDER_CONTEXT                wtFwpmFilterFlags = 0x00000004
+	cFWPM_FILTER_FLAG_CLEAR_ACTION_RIGHT                  wtFwpmFilterFlags = 0x00000008
+	cFWPM_FILTER_FLAG_PERMIT_IF_CALLOUT_UNREGISTERED      wtFwpmFilterFlags = 0x00000010
+	cFWPM_FILTER_FLAG_DISABLED                            wtFwpmFilterFlags = 0x00000020
+	cFWPM_FILTER_FLAG_INDEXED                             wtFwpmFilterFlags = 0x00000040
+	cFWPM_FILTER_FLAG_HAS_SECURITY_REALM_PROVIDER_CONTEXT wtFwpmFilterFlags = 0x00000080
+	cFWPM_FILTER_FLAG_SYSTEMOS_ONLY                       wtFwpmFilterFlags = 0x00000100
+	cFWPM_FILTER_FLAG_GAMEOS_ONLY                         wtFwpmFilterFlags = 0x00000200
+	cFWPM_FILTER_FLAG_SILENT_MODE                         wtFwpmFilterFlags = 0x00000400
+	cFWPM_FILTER_FLAG_IPSEC_NO_ACQUIRE_INITIATE           wtFwpmFilterFlags = 0x00000800
+)
+
+// FWPM_LAYER_ALE_AUTH_CONNECT_V4 (c38d57d1-05a7-4c33-904f-7fbceee60e82) defined in fwpmu.h
+var cFWPM_LAYER_ALE_AUTH_CONNECT_V4 = windows.GUID{
+	Data1: 0xc38d57d1,
+	Data2: 0x05a7,
+	Data3: 0x4c33,
+	Data4: [8]byte{0x90, 0x4f, 0x7f, 0xbc, 0xee, 0xe6, 0x0e, 0x82},
+}
+
+// e1cd9fe7-f4b5-4273-96c0-592e487b8650
+var cFWPM_LAYER_ALE_AUTH_RECV_ACCEPT_V4 = windows.GUID{
+	Data1: 0xe1cd9fe7,
+	Data2: 0xf4b5,
+	Data3: 0x4273,
+	Data4: [8]byte{0x96, 0xc0, 0x59, 0x2e, 0x48, 0x7b, 0x86, 0x50},
+}
+
+// FWPM_LAYER_ALE_AUTH_CONNECT_V6 (4a72393b-319f-44bc-84c3-ba54dcb3b6b4) defined in fwpmu.h
+var cFWPM_LAYER_ALE_AUTH_CONNECT_V6 = windows.GUID{
+	Data1: 0x4a72393b,
+	Data2: 0x319f,
+	Data3: 0x44bc,
+	Data4: [8]byte{0x84, 0xc3, 0xba, 0x54, 0xdc, 0xb3, 0xb6, 0xb4},
+}
+
+// a3b42c97-9f04-4672-b87e-cee9c483257f
+var cFWPM_LAYER_ALE_AUTH_RECV_ACCEPT_V6 = windows.GUID{
+	Data1: 0xa3b42c97,
+	Data2: 0x9f04,
+	Data3: 0x4672,
+	Data4: [8]byte{0xb8, 0x7e, 0xce, 0xe9, 0xc4, 0x83, 0x25, 0x7f},
+}
+
+// 94c44912-9d6f-4ebf-b995-05ab8a088d1b
+var cFWPM_LAYER_OUTBOUND_MAC_FRAME_NATIVE = windows.GUID{
+	Data1: 0x94c44912,
+	Data2: 0x9d6f,
+	Data3: 0x4ebf,
+	Data4: [8]byte{0xb9, 0x95, 0x05, 0xab, 0x8a, 0x08, 0x8d, 0x1b},
+}
+
+// d4220bd3-62ce-4f08-ae88-b56e8526df50
+var cFWPM_LAYER_INBOUND_MAC_FRAME_NATIVE = windows.GUID{
+	Data1: 0xd4220bd3,
+	Data2: 0x62ce,
+	Data3: 0x4f08,
+	Data4: [8]byte{0xae, 0x88, 0xb5, 0x6e, 0x85, 0x26, 0xdf, 0x50},
+}
+
+// FWP_BITMAP_ARRAY64 defined in fwtypes.h
+type wtFwpBitmapArray64 struct {
+	bitmapArray64 [8]uint8 // Windows type: [8]UINT8
+}
+
+// FWP_BYTE_ARRAY6 defined in fwtypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ns-fwptypes-fwp_byte_array6_)
+type wtFwpByteArray6 struct {
+	byteArray6 [6]uint8 // Windows type: [6]UINT8
+}
+
+// FWP_BYTE_ARRAY16 defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ns-fwptypes-fwp_byte_array16_)
+type wtFwpByteArray16 struct {
+	byteArray16 [16]uint8 // Windows type [16]UINT8
+}
+
+// FWP_CONDITION_VALUE0 defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ns-fwptypes-fwp_condition_value0).
+type wtFwpConditionValue0 wtFwpValue0
+
+// FWP_DATA_TYPE defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ne-fwptypes-fwp_data_type_)
+type wtFwpDataType uint
+
+const (
+	cFWP_EMPTY                         wtFwpDataType = 0
+	cFWP_UINT8                         wtFwpDataType = cFWP_EMPTY + 1
+	cFWP_UINT16                        wtFwpDataType = cFWP_UINT8 + 1
+	cFWP_UINT32                        wtFwpDataType = cFWP_UINT16 + 1
+	cFWP_UINT64                        wtFwpDataType = cFWP_UINT32 + 1
+	cFWP_INT8                          wtFwpDataType = cFWP_UINT64 + 1
+	cFWP_INT16                         wtFwpDataType = cFWP_INT8 + 1
+	cFWP_INT32                         wtFwpDataType = cFWP_INT16 + 1
+	cFWP_INT64                         wtFwpDataType = cFWP_INT32 + 1
+	cFWP_FLOAT                         wtFwpDataType = cFWP_INT64 + 1
+	cFWP_DOUBLE                        wtFwpDataType = cFWP_FLOAT + 1
+	cFWP_BYTE_ARRAY16_TYPE             wtFwpDataType = cFWP_DOUBLE + 1
+	cFWP_BYTE_BLOB_TYPE                wtFwpDataType = cFWP_BYTE_ARRAY16_TYPE + 1
+	cFWP_SID                           wtFwpDataType = cFWP_BYTE_BLOB_TYPE + 1
+	cFWP_SECURITY_DESCRIPTOR_TYPE      wtFwpDataType = cFWP_SID + 1
+	cFWP_TOKEN_INFORMATION_TYPE        wtFwpDataType = cFWP_SECURITY_DESCRIPTOR_TYPE + 1
+	cFWP_TOKEN_ACCESS_INFORMATION_TYPE wtFwpDataType = cFWP_TOKEN_INFORMATION_TYPE + 1
+	cFWP_UNICODE_STRING_TYPE           wtFwpDataType = cFWP_TOKEN_ACCESS_INFORMATION_TYPE + 1
+	cFWP_BYTE_ARRAY6_TYPE              wtFwpDataType = cFWP_UNICODE_STRING_TYPE + 1
+	cFWP_BITMAP_INDEX_TYPE             wtFwpDataType = cFWP_BYTE_ARRAY6_TYPE + 1
+	cFWP_BITMAP_ARRAY64_TYPE           wtFwpDataType = cFWP_BITMAP_INDEX_TYPE + 1
+	cFWP_SINGLE_DATA_TYPE_MAX          wtFwpDataType = 0xff
+	cFWP_V4_ADDR_MASK                  wtFwpDataType = cFWP_SINGLE_DATA_TYPE_MAX + 1
+	cFWP_V6_ADDR_MASK                  wtFwpDataType = cFWP_V4_ADDR_MASK + 1
+	cFWP_RANGE_TYPE                    wtFwpDataType = cFWP_V6_ADDR_MASK + 1
+	cFWP_DATA_TYPE_MAX                 wtFwpDataType = cFWP_RANGE_TYPE + 1
+)
+
+// FWP_V4_ADDR_AND_MASK defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ns-fwptypes-fwp_v4_addr_and_mask).
+type wtFwpV4AddrAndMask struct {
+	addr uint32
+	mask uint32
+}
+
+// FWP_V6_ADDR_AND_MASK defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ns-fwptypes-fwp_v6_addr_and_mask).
+type wtFwpV6AddrAndMask struct {
+	addr         [16]uint8
+	prefixLength uint8
+}
+
+// FWP_VALUE0 defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ns-fwptypes-fwp_value0_)
+type wtFwpValue0 struct {
+	_type wtFwpDataType
+	value uintptr
+}
+
+// FWPM_DISPLAY_DATA0 defined in fwptypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwptypes/ns-fwptypes-fwpm_display_data0).
+type wtFwpmDisplayData0 struct {
+	name        *uint16 // Windows type: *wchar_t
+	description *uint16 // Windows type: *wchar_t
+}
+
+// FWPM_FILTER_CONDITION0 defined in fwpmtypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwpmtypes/ns-fwpmtypes-fwpm_filter_condition0).
+type wtFwpmFilterCondition0 struct {
+	fieldKey       windows.GUID // Windows type: GUID
+	matchType      wtFwpMatchType
+	conditionValue wtFwpConditionValue0
+}
+
+// FWPM_PROVIDER0 defined in fwpmtypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwpmtypes/ns-fwpmtypes-fwpm_provider0_)
+type wtFwpProvider0 struct {
+	providerKey  windows.GUID // Windows type: GUID
+	displayData  wtFwpmDisplayData0
+	flags        uint32
+	providerData wtFwpByteBlob
+	serviceName  *uint16 // Windows type: *wchar_t
+}
+
+type wtFwpmSessionFlagsValue uint32
+
+const (
+	cFWPM_SESSION_FLAG_DYNAMIC wtFwpmSessionFlagsValue = 0x00000001 // FWPM_SESSION_FLAG_DYNAMIC defined in fwpmtypes.h
+)
+
+// FWPM_SESSION0 defined in fwpmtypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwpmtypes/ns-fwpmtypes-fwpm_session0).
+type wtFwpmSession0 struct {
+	sessionKey           windows.GUID // Windows type: GUID
+	displayData          wtFwpmDisplayData0
+	flags                wtFwpmSessionFlagsValue // Windows type UINT32
+	txnWaitTimeoutInMSec uint32
+	processId            uint32 // Windows type: DWORD
+	sid                  *windows.SID
+	username             *uint16 // Windows type: *wchar_t
+	kernelMode           uint8   // Windows type: BOOL
+}
+
+type wtFwpmSublayerFlags uint32
+
+const (
+	cFWPM_SUBLAYER_FLAG_PERSISTENT wtFwpmSublayerFlags = 0x00000001 // FWPM_SUBLAYER_FLAG_PERSISTENT defined in fwpmtypes.h
+)
+
+// FWPM_SUBLAYER0 defined in fwpmtypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwpmtypes/ns-fwpmtypes-fwpm_sublayer0_)
+type wtFwpmSublayer0 struct {
+	subLayerKey  windows.GUID // Windows type: GUID
+	displayData  wtFwpmDisplayData0
+	flags        wtFwpmSublayerFlags
+	providerKey  *windows.GUID // Windows type: *GUID
+	providerData wtFwpByteBlob
+	weight       uint16
+}
+
+// Defined in rpcdce.h
+type wtRpcCAuthN uint32
+
+const (
+	cRPC_C_AUTHN_NONE    wtRpcCAuthN = 0
+	cRPC_C_AUTHN_WINNT   wtRpcCAuthN = 10
+	cRPC_C_AUTHN_DEFAULT wtRpcCAuthN = 0xFFFFFFFF
+)
+
+// FWPM_PROVIDER0 defined in fwpmtypes.h
+// (https://docs.microsoft.com/sv-se/windows/desktop/api/fwpmtypes/ns-fwpmtypes-fwpm_provider0).
+type wtFwpmProvider0 struct {
+	providerKey  windows.GUID
+	displayData  wtFwpmDisplayData0
+	flags        uint32
+	providerData wtFwpByteBlob
+	serviceName  *uint16
+}
+
+type wtIPProto uint32
+
+const (
+	cIPPROTO_ICMP   wtIPProto = 1
+	cIPPROTO_ICMPV6 wtIPProto = 58
+	cIPPROTO_TCP    wtIPProto = 6
+	cIPPROTO_UDP    wtIPProto = 17
+)
+
+const (
+	cFWP_ACTRL_MATCH_FILTER = 1
+)

client/internal/dns/dnsfw/types_windows_32.go

@@ -0,0 +1,92 @@
+//go:build windows && (386 || arm)
+
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (C) 2019-2021 WireGuard LLC. All Rights Reserved.
+ *
+ * Adapted from wireguard-windows tunnel/firewall/types_windows_32.go.
+ */
+
+package dnsfw
+
+import "golang.org/x/sys/windows"
+
+const (
+	wtFwpByteBlob_Size        = 8
+	wtFwpByteBlob_data_Offset = 4
+
+	wtFwpConditionValue0_Size         = 8
+	wtFwpConditionValue0_uint8_Offset = 4
+
+	wtFwpmDisplayData0_Size               = 8
+	wtFwpmDisplayData0_description_Offset = 4
+
+	wtFwpmFilter0_Size                       = 152
+	wtFwpmFilter0_displayData_Offset         = 16
+	wtFwpmFilter0_flags_Offset               = 24
+	wtFwpmFilter0_providerKey_Offset         = 28
+	wtFwpmFilter0_providerData_Offset        = 32
+	wtFwpmFilter0_layerKey_Offset            = 40
+	wtFwpmFilter0_subLayerKey_Offset         = 56
+	wtFwpmFilter0_weight_Offset              = 72
+	wtFwpmFilter0_numFilterConditions_Offset = 80
+	wtFwpmFilter0_filterCondition_Offset     = 84
+	wtFwpmFilter0_action_Offset              = 88
+	wtFwpmFilter0_providerContextKey_Offset  = 112
+	wtFwpmFilter0_reserved_Offset            = 128
+	wtFwpmFilter0_filterID_Offset            = 136
+	wtFwpmFilter0_effectiveWeight_Offset     = 144
+
+	wtFwpmFilterCondition0_Size                  = 28
+	wtFwpmFilterCondition0_matchType_Offset      = 16
+	wtFwpmFilterCondition0_conditionValue_Offset = 20
+
+	wtFwpmSession0_Size                        = 48
+	wtFwpmSession0_displayData_Offset          = 16
+	wtFwpmSession0_flags_Offset                = 24
+	wtFwpmSession0_txnWaitTimeoutInMSec_Offset = 28
+	wtFwpmSession0_processId_Offset            = 32
+	wtFwpmSession0_sid_Offset                  = 36
+	wtFwpmSession0_username_Offset             = 40
+	wtFwpmSession0_kernelMode_Offset           = 44
+
+	wtFwpmSublayer0_Size                = 44
+	wtFwpmSublayer0_displayData_Offset  = 16
+	wtFwpmSublayer0_flags_Offset        = 24
+	wtFwpmSublayer0_providerKey_Offset  = 28
+	wtFwpmSublayer0_providerData_Offset = 32
+	wtFwpmSublayer0_weight_Offset       = 40
+
+	wtFwpProvider0_Size                = 40
+	wtFwpProvider0_displayData_Offset  = 16
+	wtFwpProvider0_flags_Offset        = 24
+	wtFwpProvider0_providerData_Offset = 28
+	wtFwpProvider0_serviceName_Offset  = 36
+
+	wtFwpTokenInformation_Size = 16
+
+	wtFwpValue0_Size         = 8
+	wtFwpValue0_value_Offset = 4
+)
+
+// FWPM_FILTER0 defined in fwpmtypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwpmtypes/ns-fwpmtypes-fwpm_filter0).
+type wtFwpmFilter0 struct {
+	filterKey           windows.GUID // Windows type: GUID
+	displayData         wtFwpmDisplayData0
+	flags               wtFwpmFilterFlags
+	providerKey         *windows.GUID // Windows type: *GUID
+	providerData        wtFwpByteBlob
+	layerKey            windows.GUID // Windows type: GUID
+	subLayerKey         windows.GUID // Windows type: GUID
+	weight              wtFwpValue0
+	numFilterConditions uint32
+	filterCondition     *wtFwpmFilterCondition0
+	action              wtFwpmAction0
+	offset1             [4]byte       // Layout correction field
+	providerContextKey  windows.GUID  // Windows type: GUID
+	reserved            *windows.GUID // Windows type: *GUID
+	offset2             [4]byte       // Layout correction field
+	filterID            uint64
+	effectiveWeight     wtFwpValue0
+}

client/internal/dns/dnsfw/types_windows_64.go

@@ -0,0 +1,89 @@
+//go:build windows && (amd64 || arm64)
+
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (C) 2019-2021 WireGuard LLC. All Rights Reserved.
+ *
+ * Adapted from wireguard-windows tunnel/firewall/types_windows_64.go.
+ */
+
+package dnsfw
+
+import "golang.org/x/sys/windows"
+
+const (
+	wtFwpByteBlob_Size        = 16
+	wtFwpByteBlob_data_Offset = 8
+
+	wtFwpConditionValue0_Size         = 16
+	wtFwpConditionValue0_uint8_Offset = 8
+
+	wtFwpmDisplayData0_Size               = 16
+	wtFwpmDisplayData0_description_Offset = 8
+
+	wtFwpmFilter0_Size                       = 200
+	wtFwpmFilter0_displayData_Offset         = 16
+	wtFwpmFilter0_flags_Offset               = 32
+	wtFwpmFilter0_providerKey_Offset         = 40
+	wtFwpmFilter0_providerData_Offset        = 48
+	wtFwpmFilter0_layerKey_Offset            = 64
+	wtFwpmFilter0_subLayerKey_Offset         = 80
+	wtFwpmFilter0_weight_Offset              = 96
+	wtFwpmFilter0_numFilterConditions_Offset = 112
+	wtFwpmFilter0_filterCondition_Offset     = 120
+	wtFwpmFilter0_action_Offset              = 128
+	wtFwpmFilter0_providerContextKey_Offset  = 152
+	wtFwpmFilter0_reserved_Offset            = 168
+	wtFwpmFilter0_filterID_Offset            = 176
+	wtFwpmFilter0_effectiveWeight_Offset     = 184
+
+	wtFwpmFilterCondition0_Size                  = 40
+	wtFwpmFilterCondition0_matchType_Offset      = 16
+	wtFwpmFilterCondition0_conditionValue_Offset = 24
+
+	wtFwpmSession0_Size                        = 72
+	wtFwpmSession0_displayData_Offset          = 16
+	wtFwpmSession0_flags_Offset                = 32
+	wtFwpmSession0_txnWaitTimeoutInMSec_Offset = 36
+	wtFwpmSession0_processId_Offset            = 40
+	wtFwpmSession0_sid_Offset                  = 48
+	wtFwpmSession0_username_Offset             = 56
+	wtFwpmSession0_kernelMode_Offset           = 64
+
+	wtFwpmSublayer0_Size                = 72
+	wtFwpmSublayer0_displayData_Offset  = 16
+	wtFwpmSublayer0_flags_Offset        = 32
+	wtFwpmSublayer0_providerKey_Offset  = 40
+	wtFwpmSublayer0_providerData_Offset = 48
+	wtFwpmSublayer0_weight_Offset       = 64
+
+	wtFwpProvider0_Size                = 64
+	wtFwpProvider0_displayData_Offset  = 16
+	wtFwpProvider0_flags_Offset        = 32
+	wtFwpProvider0_providerData_Offset = 40
+	wtFwpProvider0_serviceName_Offset  = 56
+
+	wtFwpValue0_Size         = 16
+	wtFwpValue0_value_Offset = 8
+)
+
+// FWPM_FILTER0 defined in fwpmtypes.h
+// (https://docs.microsoft.com/en-us/windows/desktop/api/fwpmtypes/ns-fwpmtypes-fwpm_filter0).
+type wtFwpmFilter0 struct {
+	filterKey           windows.GUID // Windows type: GUID
+	displayData         wtFwpmDisplayData0
+	flags               wtFwpmFilterFlags // Windows type: UINT32
+	providerKey         *windows.GUID     // Windows type: *GUID
+	providerData        wtFwpByteBlob
+	layerKey            windows.GUID // Windows type: GUID
+	subLayerKey         windows.GUID // Windows type: GUID
+	weight              wtFwpValue0
+	numFilterConditions uint32
+	filterCondition     *wtFwpmFilterCondition0
+	action              wtFwpmAction0
+	offset1             [4]byte       // Layout correction field
+	providerContextKey  windows.GUID  // Windows type: GUID
+	reserved            *windows.GUID // Windows type: *GUID
+	filterID            uint64
+	effectiveWeight     wtFwpValue0
+}

client/internal/dns/dnsfw/zsyscall_windows.go

@@ -0,0 +1,130 @@
+// Code generated by 'go generate'; DO NOT EDIT.
+
+package dnsfw
+
+import (
+	"syscall"
+	"unsafe"
+
+	"golang.org/x/sys/windows"
+)
+
+var _ unsafe.Pointer
+
+// Do the interface allocations only once for common
+// Errno values.
+const (
+	errnoERROR_IO_PENDING = 997
+)
+
+var (
+	errERROR_IO_PENDING error = syscall.Errno(errnoERROR_IO_PENDING)
+	errERROR_EINVAL     error = syscall.EINVAL
+)
+
+// errnoErr returns common boxed Errno values, to prevent
+// allocations at runtime.
+func errnoErr(e syscall.Errno) error {
+	switch e {
+	case 0:
+		return errERROR_EINVAL
+	case errnoERROR_IO_PENDING:
+		return errERROR_IO_PENDING
+	}
+	// TODO: add more here, after collecting data on the common
+	// error values see on Windows. (perhaps when running
+	// all.bat?)
+	return e
+}
+
+var (
+	modfwpuclnt = windows.NewLazySystemDLL("fwpuclnt.dll")
+
+	procFwpmEngineClose0          = modfwpuclnt.NewProc("FwpmEngineClose0")
+	procFwpmEngineOpen0           = modfwpuclnt.NewProc("FwpmEngineOpen0")
+	procFwpmFilterAdd0            = modfwpuclnt.NewProc("FwpmFilterAdd0")
+	procFwpmFreeMemory0           = modfwpuclnt.NewProc("FwpmFreeMemory0")
+	procFwpmGetAppIdFromFileName0 = modfwpuclnt.NewProc("FwpmGetAppIdFromFileName0")
+	procFwpmProviderAdd0          = modfwpuclnt.NewProc("FwpmProviderAdd0")
+	procFwpmSubLayerAdd0          = modfwpuclnt.NewProc("FwpmSubLayerAdd0")
+	procFwpmTransactionAbort0     = modfwpuclnt.NewProc("FwpmTransactionAbort0")
+	procFwpmTransactionBegin0     = modfwpuclnt.NewProc("FwpmTransactionBegin0")
+	procFwpmTransactionCommit0    = modfwpuclnt.NewProc("FwpmTransactionCommit0")
+)
+
+func fwpmEngineClose0(engineHandle uintptr) (err error) {
+	r1, _, e1 := syscall.Syscall(procFwpmEngineClose0.Addr(), 1, uintptr(engineHandle), 0, 0)
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}
+
+func fwpmEngineOpen0(serverName *uint16, authnService wtRpcCAuthN, authIdentity *uintptr, session *wtFwpmSession0, engineHandle unsafe.Pointer) (err error) {
+	r1, _, e1 := syscall.Syscall6(procFwpmEngineOpen0.Addr(), 5, uintptr(unsafe.Pointer(serverName)), uintptr(authnService), uintptr(unsafe.Pointer(authIdentity)), uintptr(unsafe.Pointer(session)), uintptr(engineHandle), 0)
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}
+
+func fwpmFilterAdd0(engineHandle uintptr, filter *wtFwpmFilter0, sd uintptr, id *uint64) (err error) {
+	r1, _, e1 := syscall.Syscall6(procFwpmFilterAdd0.Addr(), 4, uintptr(engineHandle), uintptr(unsafe.Pointer(filter)), uintptr(sd), uintptr(unsafe.Pointer(id)), 0, 0)
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}
+
+func fwpmFreeMemory0(p unsafe.Pointer) {
+	syscall.Syscall(procFwpmFreeMemory0.Addr(), 1, uintptr(p), 0, 0)
+	return
+}
+
+func fwpmGetAppIdFromFileName0(fileName *uint16, appID unsafe.Pointer) (err error) {
+	r1, _, e1 := syscall.Syscall(procFwpmGetAppIdFromFileName0.Addr(), 2, uintptr(unsafe.Pointer(fileName)), uintptr(appID), 0)
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}
+
+func fwpmProviderAdd0(engineHandle uintptr, provider *wtFwpmProvider0, sd uintptr) (err error) {
+	r1, _, e1 := syscall.Syscall(procFwpmProviderAdd0.Addr(), 3, uintptr(engineHandle), uintptr(unsafe.Pointer(provider)), uintptr(sd))
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}
+
+func fwpmSubLayerAdd0(engineHandle uintptr, subLayer *wtFwpmSublayer0, sd uintptr) (err error) {
+	r1, _, e1 := syscall.Syscall(procFwpmSubLayerAdd0.Addr(), 3, uintptr(engineHandle), uintptr(unsafe.Pointer(subLayer)), uintptr(sd))
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}
+
+func fwpmTransactionAbort0(engineHandle uintptr) (err error) {
+	r1, _, e1 := syscall.Syscall(procFwpmTransactionAbort0.Addr(), 1, uintptr(engineHandle), 0, 0)
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}
+
+func fwpmTransactionBegin0(engineHandle uintptr, flags uint32) (err error) {
+	r1, _, e1 := syscall.Syscall(procFwpmTransactionBegin0.Addr(), 2, uintptr(engineHandle), uintptr(flags), 0)
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}
+
+func fwpmTransactionCommit0(engineHandle uintptr) (err error) {
+	r1, _, e1 := syscall.Syscall(procFwpmTransactionCommit0.Addr(), 1, uintptr(engineHandle), 0, 0)
+	if r1 != 0 {
+		err = errnoErr(e1)
+	}
+	return
+}

client/internal/dns/host_windows.go

@@ -17,6 +17,7 @@ import (
 	"golang.org/x/sys/windows/registry"
 
 	nberrors "github.com/netbirdio/netbird/client/errors"
+	"github.com/netbirdio/netbird/client/internal/dns/dnsfw"
 	"github.com/netbirdio/netbird/client/internal/statemanager"
 	"github.com/netbirdio/netbird/client/internal/winregistry"
 )
@@ -74,6 +75,7 @@ type registryConfigurator struct {
 	routingAll      bool
 	gpo             bool
 	nrptEntryCount  int
+	dnsFirewall     dnsfw.Manager
 	origNameservers []netip.Addr
 }
 
@@ -94,8 +96,9 @@ func newHostManager(wgInterface WGIface) (*registryConfigurator, error) {
 	}
 
 	configurator := &registryConfigurator{
-		guid: guid,
-		gpo:  useGPO,
+		guid:        guid,
+		gpo:         useGPO,
+		dnsFirewall: dnsfw.New(),
 	}
 
 	origNameservers, err := configurator.captureOriginalNameservers()
@@ -276,16 +279,8 @@ func (r *registryConfigurator) disableWINSForInterface() error {
 }
 
 func (r *registryConfigurator) applyDNSConfig(config HostDNSConfig, stateManager *statemanager.Manager) error {
-	if config.RouteAll {
-		if err := r.addDNSSetupForAll(config.ServerIP); err != nil {
-			return fmt.Errorf("add dns setup: %w", err)
-		}
-	} else if r.routingAll {
-		if err := r.deleteInterfaceRegistryKeyProperty(interfaceConfigNameServerKey); err != nil {
-			return fmt.Errorf("delete interface registry key property: %w", err)
-		}
-		r.routingAll = false
-		log.Infof("removed %s as main DNS forwarder for this peer", config.ServerIP)
+	if err := r.applyRouteAll(config); err != nil {
+		return err
 	}
 
 	r.updateState(stateManager)
@@ -327,6 +322,35 @@ func (r *registryConfigurator) applyDNSConfig(config HostDNSConfig, stateManager
 	return nil
 }
 
+func (r *registryConfigurator) applyRouteAll(config HostDNSConfig) error {
+	if config.RouteAll {
+		if err := r.dnsFirewall.Enable(r.guid, config.ServerIP); err != nil {
+			return fmt.Errorf("dns firewall: %w", err)
+		}
+		if err := r.addDNSSetupForAll(config.ServerIP); err != nil {
+			merr := multierror.Append(nil, fmt.Errorf("add dns setup: %w", err))
+			if dErr := r.dnsFirewall.Disable(); dErr != nil {
+				merr = multierror.Append(merr, fmt.Errorf("rollback dns firewall: %w", dErr))
+			}
+			return nberrors.FormatErrorOrNil(merr)
+		}
+		return nil
+	}
+
+	if err := r.dnsFirewall.Disable(); err != nil {
+		log.Errorf("disable dns firewall: %v", err)
+	}
+	if !r.routingAll {
+		return nil
+	}
+	if err := r.deleteInterfaceRegistryKeyProperty(interfaceConfigNameServerKey); err != nil {
+		return fmt.Errorf("delete interface registry key property: %w", err)
+	}
+	r.routingAll = false
+	log.Infof("removed %s as main DNS forwarder for this peer", config.ServerIP)
+	return nil
+}
+
 func (r *registryConfigurator) updateState(stateManager *statemanager.Manager) {
 	if err := stateManager.UpdateState(&ShutdownState{
 		Guid:           r.guid,
@@ -513,6 +537,10 @@ func (r *registryConfigurator) restoreHostDNS() error {
 		return fmt.Errorf("remove interface registry key: %w", err)
 	}
 
+	if err := r.dnsFirewall.Disable(); err != nil {
+		log.Errorf("disable dns firewall: %v", err)
+	}
+
 	go r.flushDNSCache()
 
 	return nil

client/internal/dns/host_windows_test.go

@@ -8,6 +8,8 @@ import (
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"golang.org/x/sys/windows/registry"
+
+	"github.com/netbirdio/netbird/client/internal/dns/dnsfw"
 )
 
 // TestNRPTEntriesCleanupOnConfigChange tests that old NRPT entries are properly cleaned up
@@ -34,8 +36,9 @@ func TestNRPTEntriesCleanupOnConfigChange(t *testing.T) {
 	}()
 
 	cfg := &registryConfigurator{
-		guid: testGUID,
-		gpo:  false,
+		guid:        testGUID,
+		gpo:         false,
+		dnsFirewall: dnsfw.New(),
 	}
 
 	// Create 125 domains which will result in 3 NRPT rules (50+50+25)
@@ -134,8 +137,9 @@ func TestNRPTDomainBatching(t *testing.T) {
 	}()
 
 	cfg := &registryConfigurator{
-		guid: testGUID,
-		gpo:  false,
+		guid:        testGUID,
+		gpo:         false,
+		dnsFirewall: dnsfw.New(),
 	}
 
 	testCases := []struct {

client/internal/dns/server.go

@@ -777,13 +777,24 @@ func (s *DefaultServer) applyHostConfig() {
 // context is released rather than leaked until GC.
 func (s *DefaultServer) registerFallback() {
 	originalNameservers := s.hostManager.getOriginalNameservers()
-	if len(originalNameservers) == 0 {
+
+	serverIP := s.service.RuntimeIP()
+	var servers []netip.AddrPort
+	for _, ns := range originalNameservers {
+		if ns == serverIP {
+			log.Debugf("skipping original nameserver %s as it is the same as the server IP %s", ns, serverIP)
+			continue
+		}
+		servers = append(servers, netip.AddrPortFrom(ns, DefaultPort))
+	}
+
+	if len(servers) == 0 {
 		log.Debugf("no fallback upstreams to register; clearing PriorityFallback handler")
 		s.clearFallback()
 		return
 	}
 
-	log.Infof("registering original nameservers %v as upstream handlers with priority %d", originalNameservers, PriorityFallback)
+	log.Infof("registering original nameservers %v as upstream handlers with priority %d", servers, PriorityFallback)
 
 	handler, err := newUpstreamResolver(
 		s.ctx,
@@ -797,11 +808,6 @@ func (s *DefaultServer) registerFallback() {
 		return
 	}
 	handler.selectedRoutes = s.selectedRoutes
-
-	var servers []netip.AddrPort
-	for _, ns := range originalNameservers {
-		servers = append(servers, netip.AddrPortFrom(ns, DefaultPort))
-	}
 	handler.addRace(servers)
 
 	prev := s.fallbackHandler

client/internal/engine.go

@@ -22,7 +22,6 @@ import (
 	log "github.com/sirupsen/logrus"
 	"golang.zx2c4.com/wireguard/tun/netstack"
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
-	"google.golang.org/protobuf/proto"
 
 	nberrors "github.com/netbirdio/netbird/client/errors"
 	"github.com/netbirdio/netbird/client/firewall"
@@ -56,6 +55,7 @@ import (
 	"github.com/netbirdio/netbird/client/internal/routemanager"
 	"github.com/netbirdio/netbird/client/internal/routemanager/systemops"
 	"github.com/netbirdio/netbird/client/internal/statemanager"
+	"github.com/netbirdio/netbird/client/internal/syncstore"
 	"github.com/netbirdio/netbird/client/internal/updater"
 	"github.com/netbirdio/netbird/client/jobexec"
 	cProto "github.com/netbirdio/netbird/client/proto"
@@ -72,6 +72,7 @@ import (
 	sProto "github.com/netbirdio/netbird/shared/signal/proto"
 	"github.com/netbirdio/netbird/util"
 	"github.com/netbirdio/netbird/util/capture"
+	"github.com/netbirdio/netbird/version"
 )
 
 // PeerConnectionTimeoutMax is a timeout of an initial connection attempt to a remote peer.
@@ -148,6 +149,10 @@ type EngineConfig struct {
 
 	LogPath string
 	TempDir string
+
+	// StateDir is the directory holding the state file. The sync response
+	// (network map) is serialized here on platforms that persist it to disk.
+	StateDir string
 }
 
 // EngineServices holds the external service dependencies required by the Engine.
@@ -226,10 +231,15 @@ type Engine struct {
 
 	afpacketCapture *capture.AFPacketCapture
 
-	// Sync response persistence (protected by syncRespMux)
-	syncRespMux         sync.RWMutex
-	persistSyncResponse bool
-	latestSyncResponse  *mgmProto.SyncResponse
+	// Sync response persistence (protected by syncRespMux).
+	// syncStore is nil unless persistence has been enabled; its presence is
+	// what marks persistence as active. The backend (disk or memory) is
+	// selected per-platform; see the syncstore package. syncStoreDir is where
+	// a disk-backed store serializes to.
+	syncRespMux  sync.RWMutex
+	syncStore    syncstore.Store
+	syncStoreDir string
+
 	flowManager         nftypes.FlowManager
 
 	// auto-update
@@ -292,6 +302,7 @@ func NewEngine(
 		jobExecutor:        jobexec.NewExecutor(),
 		clientMetrics:      services.ClientMetrics,
 		updateManager:      services.UpdateManager,
+		syncStoreDir:       config.StateDir,
 	}
 
 	log.Infof("I am: %s", config.WgPrivateKey.PublicKey().String())
@@ -913,19 +924,18 @@ func (e *Engine) handleSync(update *mgmProto.SyncResponse) error {
 	}
 
 	// Persist sync response under the dedicated lock (syncRespMux), not under syncMsgMux.
-	// Read the storage-enabled flag under the syncRespMux too.
+	// A non-nil syncStore is what marks persistence as enabled. Hold the lock for
+	// the whole Set so the store cannot be cleared (disabled / engine close)
+	// mid-call and have this write resurrect a file that was just removed.
 	e.syncRespMux.RLock()
-	enabled := e.persistSyncResponse
-	e.syncRespMux.RUnlock()
-
-	// Store sync response if persistence is enabled
-	if enabled {
-		e.syncRespMux.Lock()
-		e.latestSyncResponse = update
-		e.syncRespMux.Unlock()
-
-		log.Debugf("sync response persisted with serial %d", nm.GetSerial())
+	if e.syncStore != nil {
+		if err := e.syncStore.Set(update); err != nil {
+			log.Errorf("failed to persist sync response: %v", err)
+		} else {
+			log.Debugf("sync response persisted with serial %d", nm.GetSerial())
+		}
 	}
+	e.syncRespMux.RUnlock()
 
 	// only apply new changes and ignore old ones
 	if err := e.updateNetworkMap(nm); err != nil {
@@ -1063,6 +1073,7 @@ func (e *Engine) updateConfig(conf *mgmProto.PeerConfig) error {
 	state.PubKey = e.config.WgPrivateKey.PublicKey().String()
 	state.KernelInterface = !e.wgInterface.IsUserspaceBind()
 	state.FQDN = conf.GetFqdn()
+	state.WgPort = e.config.WgPort
 
 	e.statusRecorder.UpdateLocalPeerState(state)
 
@@ -1141,6 +1152,7 @@ func (e *Engine) handleBundle(params *mgmProto.BundleParameters) (*mgmProto.JobR
 		LogPath:        e.config.LogPath,
 		TempDir:        e.config.TempDir,
 		ClientMetrics:  e.clientMetrics,
+		DaemonVersion:  version.NetbirdVersion(),
 		RefreshStatus: func() {
 			e.RunHealthProbes(true)
 		},
@@ -1813,6 +1825,18 @@ func (e *Engine) close() {
 	if err := e.portForwardManager.GracefullyStop(ctx); err != nil {
 		log.Warnf("failed to gracefully stop port forwarding manager: %s", err)
 	}
+
+	// Drop any persisted sync response so its network map does not linger on
+	// disk after the engine stops (and cannot leak into a later run).
+	e.syncRespMux.Lock()
+	store := e.syncStore
+	e.syncStore = nil
+	e.syncRespMux.Unlock()
+	if store != nil {
+		if err := store.Clear(); err != nil {
+			log.Warnf("failed to clear persisted sync response on close: %v", err)
+		}
+	}
 }
 
 func (e *Engine) readInitialSettings() ([]*route.Route, *nbdns.Config, bool, error) {
@@ -2142,45 +2166,42 @@ func (e *Engine) stopDNSServer() {
 	e.statusRecorder.UpdateDNSStates(nsGroupStates)
 }
 
-// SetSyncResponsePersistence enables or disables sync response persistence
+// SetSyncResponsePersistence enables or disables sync response persistence.
+// The store is only instantiated while persistence is enabled; construction
+// itself drops any stale data left over from an earlier run (see syncstore).
 func (e *Engine) SetSyncResponsePersistence(enabled bool) {
 	e.syncRespMux.Lock()
 	defer e.syncRespMux.Unlock()
 
-	if enabled == e.persistSyncResponse {
+	if enabled == (e.syncStore != nil) {
 		return
 	}
-	e.persistSyncResponse = enabled
 	log.Debugf("Sync response persistence is set to %t", enabled)
 
 	if !enabled {
-		e.latestSyncResponse = nil
+		if err := e.syncStore.Clear(); err != nil {
+			log.Warnf("failed to clear persisted sync response: %v", err)
+		}
+		e.syncStore = nil
+		return
 	}
+
+	e.syncStore = syncstore.New(e.syncStoreDir)
 }
 
 // GetLatestSyncResponse returns the stored sync response if persistence is enabled
 func (e *Engine) GetLatestSyncResponse() (*mgmProto.SyncResponse, error) {
+	// Hold the lock for the whole Get so the store cannot be cleared
+	// (disabled / engine close) mid-call.
 	e.syncRespMux.RLock()
-	enabled := e.persistSyncResponse
-	latest := e.latestSyncResponse
-	e.syncRespMux.RUnlock()
+	defer e.syncRespMux.RUnlock()
 
-	if !enabled {
+	if e.syncStore == nil {
 		return nil, errors.New("sync response persistence is disabled")
 	}
 
-	if latest == nil {
-		//nolint:nilnil
-		return nil, nil
-	}
-
-	log.Debugf("Retrieving latest sync response with size %d bytes", proto.Size(latest))
-	sr, ok := proto.Clone(latest).(*mgmProto.SyncResponse)
-	if !ok {
-		return nil, fmt.Errorf("failed to clone sync response")
-	}
-
-	return sr, nil
+	//nolint:nilnil
+	return e.syncStore.Get()
 }
 
 // GetWgAddr returns the wireguard address
@@ -2216,7 +2237,7 @@ func (e *Engine) updateDNSForwarder(
 	enabled bool,
 	fwdEntries []*dnsfwd.ForwarderEntry,
 ) {
-	if e.config.DisableServerRoutes {
+	if e.config.DisableServerRoutes || e.config.BlockInbound {
 		return
 	}
 

client/internal/lazyconn/support.go

@@ -4,6 +4,8 @@ import (
 	"strings"
 
 	"github.com/hashicorp/go-version"
+
+	nbversion "github.com/netbirdio/netbird/version"
 )
 
 var (
@@ -11,7 +13,7 @@ var (
 )
 
 func IsSupported(agentVersion string) bool {
-	if agentVersion == "development" {
+	if nbversion.IsDevelopmentVersion(agentVersion) {
 		return true
 	}
 

client/internal/peer/status.go

@@ -111,6 +111,7 @@ type LocalPeerState struct {
 	PubKey          string
 	KernelInterface bool
 	FQDN            string
+	WgPort          int
 	Routes          map[string]struct{}
 }
 
@@ -310,8 +311,12 @@ func (d *Status) PeerByIP(ip string) (string, bool) {
 
 // PeerStateByIP returns the full peer State for the given tunnel IP.
 // Matches against either the IPv4 (State.IP) or IPv6 (State.IPv6) tunnel
-// address so dual-stack peers are reachable on either family. Returns the
-// zero State and false when no peer matches or the input is empty.
+// address so dual-stack peers are reachable on either family. Searches
+// both d.peers and d.offlinePeers — peers that have been moved into
+// the offline slice by ReplaceOfflinePeers are still part of the
+// account's roster and callers (DNS filter, embed.Client.IdentityForIP)
+// need to recognise them rather than treating them as unknown. Returns
+// the zero State and false when no peer matches or the input is empty.
 func (d *Status) PeerStateByIP(ip string) (State, bool) {
 	if ip == "" {
 		return State{}, false
@@ -324,6 +329,11 @@ func (d *Status) PeerStateByIP(ip string) (State, bool) {
 			return state, true
 		}
 	}
+	for _, state := range d.offlinePeers {
+		if (state.IP != "" && state.IP == ip) || (state.IPv6 != "" && state.IPv6 == ip) {
+			return state, true
+		}
+	}
 	return State{}, false
 }
 
@@ -1348,6 +1358,7 @@ func (fs FullStatus) ToProto() *proto.FullStatus {
 	pbFullStatus.LocalPeerState.PubKey = fs.LocalPeerState.PubKey
 	pbFullStatus.LocalPeerState.KernelInterface = fs.LocalPeerState.KernelInterface
 	pbFullStatus.LocalPeerState.Fqdn = fs.LocalPeerState.FQDN
+	pbFullStatus.LocalPeerState.WgPort = int32(fs.LocalPeerState.WgPort)
 	pbFullStatus.LocalPeerState.RosenpassPermissive = fs.RosenpassState.Permissive
 	pbFullStatus.LocalPeerState.RosenpassEnabled = fs.RosenpassState.Enabled
 	pbFullStatus.NumberOfForwardingRules = int32(fs.NumOfForwardingRules)

client/internal/peer/status_test.go

@@ -90,6 +90,28 @@ func TestStatus_PeerStateByIP_MatchesIPv6(t *testing.T) {
 	req.Equal("pk-1", state.PubKey, "matching state must carry the right pub key")
 }
 
+// TestStatus_PeerStateByIP_MatchesOfflinePeers covers peers that have
+// been moved into the offline slice via ReplaceOfflinePeers. Callers
+// (DNS filter, embed.Client.IdentityForIP) need to treat them as known
+// rather than unknown — otherwise authentication / DNS filtering treats
+// known-but-offline peers as foreign IPs.
+func TestStatus_PeerStateByIP_MatchesOfflinePeers(t *testing.T) {
+	status := NewRecorder("https://mgm")
+	req := require.New(t)
+
+	status.ReplaceOfflinePeers([]State{
+		{PubKey: "pk-offline", FQDN: "offline.netbird", IP: "100.64.0.20", IPv6: "fd00::20"},
+	})
+
+	state, ok := status.PeerStateByIP("100.64.0.20")
+	req.True(ok, "offline peer must resolve by IPv4 tunnel address")
+	req.Equal("pk-offline", state.PubKey, "matching state must carry the offline peer's pub key")
+
+	state, ok = status.PeerStateByIP("fd00::20")
+	req.True(ok, "offline peer must resolve by IPv6 tunnel address")
+	req.Equal("pk-offline", state.PubKey, "IPv6 match must carry the offline peer's pub key")
+}
+
 func TestStatus_UpdatePeerFQDN(t *testing.T) {
 	key := "abc"
 	fqdn := "peer-a.netbird.local"

client/internal/routemanager/manager.go

@@ -700,6 +700,13 @@ func resolveURLsToIPs(urls []string) []net.IP {
 
 // updateRouteSelectorFromManagement updates the route selector based on the isSelected status from the management server
 func (m *DefaultManager) updateRouteSelectorFromManagement(clientRoutes route.HAMap) {
+	// An explicit user "deselect all" must not be overridden by management auto-apply.
+	// Auto-applying an exit node here would call SelectRoutes, which clears the
+	// deselect-all flag and re-enables every route the user turned off.
+	if m.routeSelector.IsDeselectAll() {
+		return
+	}
+
 	exitNodeInfo := m.collectExitNodeInfo(clientRoutes)
 	if len(exitNodeInfo.allIDs) == 0 {
 		return

client/internal/routemanager/selector_management_test.go

@@ -0,0 +1,71 @@
+package routemanager
+
+import (
+	"net/netip"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/internal/routeselector"
+	"github.com/netbirdio/netbird/route"
+)
+
+func exitNodeRoutes(netID route.NetID, skipAutoApply bool) route.HAMap {
+	haID := route.HAUniqueID(string(netID) + "|0.0.0.0/0")
+	return route.HAMap{
+		haID: []*route.Route{
+			{
+				ID:            "r-" + route.ID(netID),
+				NetID:         netID,
+				Network:       netip.MustParsePrefix("0.0.0.0/0"),
+				NetworkType:   route.IPv4Network,
+				Enabled:       true,
+				SkipAutoApply: skipAutoApply,
+			},
+		},
+	}
+}
+
+func TestUpdateRouteSelectorFromManagement(t *testing.T) {
+	t.Run("management auto-apply selects exit node without user selection", func(t *testing.T) {
+		m := &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+		routes := exitNodeRoutes("exit1", false)
+
+		m.updateRouteSelectorFromManagement(routes)
+
+		require.True(t, m.routeSelector.IsSelected("exit1"), "auto-apply exit node should be selected")
+		require.Len(t, m.routeSelector.FilterSelectedExitNodes(routes), 1, "selected exit node should pass the filter")
+	})
+
+	t.Run("management SkipAutoApply leaves exit node deselected", func(t *testing.T) {
+		m := &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+		routes := exitNodeRoutes("exit1", true)
+
+		m.updateRouteSelectorFromManagement(routes)
+
+		require.False(t, m.routeSelector.IsSelected("exit1"), "SkipAutoApply exit node should not be selected")
+		require.Empty(t, m.routeSelector.FilterSelectedExitNodes(routes), "deselected exit node should be filtered out")
+	})
+
+	t.Run("user selection is not overridden by management", func(t *testing.T) {
+		m := &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+		require.NoError(t, m.routeSelector.SelectRoutes([]route.NetID{"exit1"}, true, []route.NetID{"exit1"}))
+		routes := exitNodeRoutes("exit1", true)
+
+		m.updateRouteSelectorFromManagement(routes)
+
+		require.True(t, m.routeSelector.IsSelected("exit1"), "explicit user selection must survive a management sync that wants to skip auto-apply")
+		require.Len(t, m.routeSelector.FilterSelectedExitNodes(routes), 1, "user-selected exit node should pass the filter")
+	})
+
+	t.Run("deselect-all is preserved across a management sync", func(t *testing.T) {
+		m := &DefaultManager{routeSelector: routeselector.NewRouteSelector()}
+		m.routeSelector.DeselectAllRoutes()
+		routes := exitNodeRoutes("exit1", false)
+
+		m.updateRouteSelectorFromManagement(routes)
+
+		require.True(t, m.routeSelector.IsDeselectAll(), "an explicit deselect-all must not be cleared by management auto-apply")
+		require.Empty(t, m.routeSelector.FilterSelectedExitNodes(routes), "no routes should be selected while deselect-all is set")
+	})
+}

client/internal/routeselector/routeselector.go

@@ -116,6 +116,14 @@ func (rs *RouteSelector) DeselectAllRoutes() {
 	clear(rs.selectedRoutes)
 }
 
+// IsDeselectAll reports whether the user has explicitly deselected all routes.
+func (rs *RouteSelector) IsDeselectAll() bool {
+	rs.mu.RLock()
+	defer rs.mu.RUnlock()
+
+	return rs.deselectAll
+}
+
 // IsSelected checks if a specific route is selected.
 func (rs *RouteSelector) IsSelected(routeID route.NetID) bool {
 	rs.mu.RLock()

client/internal/syncstore/disk.go

@@ -0,0 +1,99 @@
+package syncstore
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sync"
+
+	log "github.com/sirupsen/logrus"
+	"google.golang.org/protobuf/proto"
+
+	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+	"github.com/netbirdio/netbird/util"
+)
+
+// syncResponseFileName is the name of the file the sync response is serialized
+// to, placed inside the configured directory (the state directory).
+const syncResponseFileName = "networkmap.pb"
+
+// diskStore serializes the latest sync response to a file on disk instead of
+// keeping it in memory. This trades disk I/O for a much smaller memory
+// footprint, which matters on memory-constrained platforms (iOS).
+type diskStore struct {
+	mu   sync.Mutex
+	path string
+}
+
+// NewDiskStore returns a Store that serializes the sync response to a file in
+// the given directory. If dir is empty it falls back to the OS temp directory.
+//
+// Any file left over from a previous run is removed on construction so a fresh
+// store never reads stale data (e.g. another profile's network map).
+func NewDiskStore(dir string) Store {
+	if dir == "" {
+		dir = os.TempDir()
+	}
+	s := &diskStore{
+		path: filepath.Join(dir, syncResponseFileName),
+	}
+	if err := s.Clear(); err != nil {
+		log.Warnf("failed to clear stale sync response file: %v", err)
+	}
+	return s
+}
+
+func (s *diskStore) Set(resp *mgmProto.SyncResponse) error {
+	if resp == nil {
+		return s.Clear()
+	}
+
+	bs, err := proto.Marshal(resp)
+	if err != nil {
+		return fmt.Errorf("marshal sync response: %w", err)
+	}
+
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if err := util.WriteBytesWithRestrictedPermission(context.Background(), s.path, bs); err != nil {
+		return fmt.Errorf("write sync response to %s: %w", s.path, err)
+	}
+
+	log.Debugf("sync response persisted to %s (%d bytes)", s.path, len(bs))
+	return nil
+}
+
+func (s *diskStore) Get() (*mgmProto.SyncResponse, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	bs, err := os.ReadFile(s.path)
+	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			//nolint:nilnil // nil,nil means "nothing stored", per the Store contract; preserve the original behaviour
+			return nil, nil
+		}
+		return nil, fmt.Errorf("read sync response from %s: %w", s.path, err)
+	}
+
+	resp := &mgmProto.SyncResponse{}
+	if err := proto.Unmarshal(bs, resp); err != nil {
+		return nil, fmt.Errorf("unmarshal sync response: %w", err)
+	}
+
+	log.Debugf("retrieving latest sync response from %s (%d bytes)", s.path, len(bs))
+	return resp, nil
+}
+
+func (s *diskStore) Clear() error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if err := os.Remove(s.path); err != nil && !errors.Is(err, os.ErrNotExist) {
+		return fmt.Errorf("remove sync response file %s: %w", s.path, err)
+	}
+	return nil
+}

client/internal/syncstore/factory_ios.go

@@ -0,0 +1,9 @@
+//go:build ios
+
+package syncstore
+
+// New returns the platform default store. On iOS the sync response is
+// serialized to disk (in dir) to keep it out of the constrained process memory.
+func New(dir string) Store {
+	return NewDiskStore(dir)
+}

client/internal/syncstore/factory_other.go

@@ -0,0 +1,9 @@
+//go:build !ios
+
+package syncstore
+
+// New returns the platform default store. On all non-iOS platforms the sync
+// response is kept in memory; dir is unused.
+func New(_ string) Store {
+	return NewMemoryStore()
+}

client/internal/syncstore/memory.go

@@ -0,0 +1,56 @@
+package syncstore
+
+import (
+	"fmt"
+	"sync"
+
+	log "github.com/sirupsen/logrus"
+	"google.golang.org/protobuf/proto"
+
+	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// memoryStore keeps the latest sync response in memory.
+type memoryStore struct {
+	mu     sync.RWMutex
+	latest *mgmProto.SyncResponse
+}
+
+// NewMemoryStore returns a Store that keeps the sync response in memory.
+func NewMemoryStore() Store {
+	return &memoryStore{}
+}
+
+func (s *memoryStore) Set(resp *mgmProto.SyncResponse) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	s.latest = resp
+	return nil
+}
+
+func (s *memoryStore) Get() (*mgmProto.SyncResponse, error) {
+	s.mu.RLock()
+	latest := s.latest
+	s.mu.RUnlock()
+
+	if latest == nil {
+		//nolint:nilnil // nil,nil means "nothing stored", per the Store contract; preserve the original behaviour
+		return nil, nil
+	}
+
+	log.Debugf("retrieving latest sync response with size %d bytes", proto.Size(latest))
+	sr, ok := proto.Clone(latest).(*mgmProto.SyncResponse)
+	if !ok {
+		return nil, fmt.Errorf("clone sync response")
+	}
+	return sr, nil
+}
+
+func (s *memoryStore) Clear() error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	s.latest = nil
+	return nil
+}

client/internal/syncstore/syncstore.go

@@ -0,0 +1,29 @@
+// Package syncstore stores the latest Management sync response (which carries
+// the network map) for debug bundle generation.
+//
+// The storage backend is selected at build time per operating system: on iOS
+// the response is serialized to disk to keep it out of the (tightly
+// constrained) process memory, while on all other platforms it is kept in
+// memory. The backend is chosen by the New constructor; see factory_ios.go and
+// factory_other.go.
+package syncstore
+
+import (
+	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// Store persists the latest sync response and returns it on demand.
+//
+// Implementations must be safe for concurrent use.
+type Store interface {
+	// Set stores the given sync response, replacing any previously stored one.
+	Set(resp *mgmProto.SyncResponse) error
+
+	// Get returns the stored sync response, or nil if none is stored.
+	// The returned value is an independent copy that the caller may retain.
+	Get() (*mgmProto.SyncResponse, error)
+
+	// Clear removes any stored sync response. It is safe to call when nothing
+	// is stored.
+	Clear() error
+}

client/internal/updater/manager.go

@@ -19,8 +19,6 @@ import (
 
 const (
 	latestVersion = "latest"
-	// this version will be ignored
-	developmentVersion = "development"
 )
 
 var errNoUpdateState = errors.New("no update state found")
@@ -483,7 +481,7 @@ func (m *Manager) loadAndDeleteUpdateState(ctx context.Context) (*UpdateState, e
 }
 
 func (m *Manager) shouldUpdate(updateVersion *v.Version, forceUpdate bool) bool {
-	if m.currentVersion == developmentVersion {
+	if version.IsDevelopmentVersion(m.currentVersion) {
 		log.Debugf("skipping auto-update, running development version")
 		return false
 	}

client/proto/daemon.pb.go

@@ -1614,6 +1614,7 @@ type LocalPeerState struct {
 	RosenpassPermissive bool                   `protobuf:"varint,6,opt,name=rosenpassPermissive,proto3" json:"rosenpassPermissive,omitempty"`
 	Networks            []string               `protobuf:"bytes,7,rep,name=networks,proto3" json:"networks,omitempty"`
 	Ipv6                string                 `protobuf:"bytes,8,opt,name=ipv6,proto3" json:"ipv6,omitempty"`
+	WgPort              int32                  `protobuf:"varint,9,opt,name=wgPort,proto3" json:"wgPort,omitempty"`
 	unknownFields       protoimpl.UnknownFields
 	sizeCache           protoimpl.SizeCache
 }
@@ -1704,6 +1705,13 @@ func (x *LocalPeerState) GetIpv6() string {
 	return ""
 }
 
+func (x *LocalPeerState) GetWgPort() int32 {
+	if x != nil {
+		return x.WgPort
+	}
+	return 0
+}
+
 // SignalState contains the latest state of a signal connection
 type SignalState struct {
 	state         protoimpl.MessageState `protogen:"open.v1"`
@@ -2709,6 +2717,7 @@ type DebugBundleRequest struct {
 	SystemInfo    bool                   `protobuf:"varint,3,opt,name=systemInfo,proto3" json:"systemInfo,omitempty"`
 	UploadURL     string                 `protobuf:"bytes,4,opt,name=uploadURL,proto3" json:"uploadURL,omitempty"`
 	LogFileCount  uint32                 `protobuf:"varint,5,opt,name=logFileCount,proto3" json:"logFileCount,omitempty"`
+	CliVersion    string                 `protobuf:"bytes,6,opt,name=cliVersion,proto3" json:"cliVersion,omitempty"`
 	unknownFields protoimpl.UnknownFields
 	sizeCache     protoimpl.SizeCache
 }
@@ -2771,6 +2780,13 @@ func (x *DebugBundleRequest) GetLogFileCount() uint32 {
 	return 0
 }
 
+func (x *DebugBundleRequest) GetCliVersion() string {
+	if x != nil {
+		return x.CliVersion
+	}
+	return ""
+}
+
 type DebugBundleResponse struct {
 	state               protoimpl.MessageState `protogen:"open.v1"`
 	Path                string                 `protobuf:"bytes,1,opt,name=path,proto3" json:"path,omitempty"`
@@ -6389,7 +6405,7 @@ const file_daemon_proto_rawDesc = "" +
 	"\n" +
 	"sshHostKey\x18\x13 \x01(\fR\n" +
 	"sshHostKey\x12\x12\n" +
-	"\x04ipv6\x18\x14 \x01(\tR\x04ipv6\"\x84\x02\n" +
+	"\x04ipv6\x18\x14 \x01(\tR\x04ipv6\"\x9c\x02\n" +
 	"\x0eLocalPeerState\x12\x0e\n" +
 	"\x02IP\x18\x01 \x01(\tR\x02IP\x12\x16\n" +
 	"\x06pubKey\x18\x02 \x01(\tR\x06pubKey\x12(\n" +
@@ -6398,7 +6414,8 @@ const file_daemon_proto_rawDesc = "" +
 	"\x10rosenpassEnabled\x18\x05 \x01(\bR\x10rosenpassEnabled\x120\n" +
 	"\x13rosenpassPermissive\x18\x06 \x01(\bR\x13rosenpassPermissive\x12\x1a\n" +
 	"\bnetworks\x18\a \x03(\tR\bnetworks\x12\x12\n" +
-	"\x04ipv6\x18\b \x01(\tR\x04ipv6\"S\n" +
+	"\x04ipv6\x18\b \x01(\tR\x04ipv6\x12\x16\n" +
+	"\x06wgPort\x18\t \x01(\x05R\x06wgPort\"S\n" +
 	"\vSignalState\x12\x10\n" +
 	"\x03URL\x18\x01 \x01(\tR\x03URL\x12\x1c\n" +
 	"\tconnected\x18\x02 \x01(\bR\tconnected\x12\x14\n" +
@@ -6475,14 +6492,17 @@ const file_daemon_proto_rawDesc = "" +
 	"\x12translatedHostname\x18\x04 \x01(\tR\x12translatedHostname\x128\n" +
 	"\x0etranslatedPort\x18\x05 \x01(\v2\x10.daemon.PortInfoR\x0etranslatedPort\"G\n" +
 	"\x17ForwardingRulesResponse\x12,\n" +
-	"\x05rules\x18\x01 \x03(\v2\x16.daemon.ForwardingRuleR\x05rules\"\x94\x01\n" +
+	"\x05rules\x18\x01 \x03(\v2\x16.daemon.ForwardingRuleR\x05rules\"\xb4\x01\n" +
 	"\x12DebugBundleRequest\x12\x1c\n" +
 	"\tanonymize\x18\x01 \x01(\bR\tanonymize\x12\x1e\n" +
 	"\n" +
 	"systemInfo\x18\x03 \x01(\bR\n" +
 	"systemInfo\x12\x1c\n" +
 	"\tuploadURL\x18\x04 \x01(\tR\tuploadURL\x12\"\n" +
-	"\flogFileCount\x18\x05 \x01(\rR\flogFileCount\"}\n" +
+	"\flogFileCount\x18\x05 \x01(\rR\flogFileCount\x12\x1e\n" +
+	"\n" +
+	"cliVersion\x18\x06 \x01(\tR\n" +
+	"cliVersion\"}\n" +
 	"\x13DebugBundleResponse\x12\x12\n" +
 	"\x04path\x18\x01 \x01(\tR\x04path\x12 \n" +
 	"\vuploadedKey\x18\x02 \x01(\tR\vuploadedKey\x120\n" +

client/proto/daemon.proto

@@ -349,6 +349,7 @@ message LocalPeerState {
   bool rosenpassPermissive = 6;
   repeated string networks = 7;
   string ipv6 = 8;
+  int32 wgPort = 9;
 }
 
 // SignalState contains the latest state of a signal connection
@@ -471,6 +472,7 @@ message DebugBundleRequest {
   bool systemInfo = 3;
   string uploadURL = 4;
   uint32 logFileCount = 5;
+  string cliVersion = 6;
 }
 
 message DebugBundleResponse {

client/proto/generate.sh

@@ -1,17 +1,16 @@
 #!/bin/bash
 set -e
 
-if ! which realpath > /dev/null 2>&1
-then
-  echo realpath is not installed
-  echo run: brew install coreutils
-  exit 1
+if ! which realpath >/dev/null 2>&1; then
+	echo realpath is not installed
+	echo run: brew install coreutils
+	exit 1
 fi
 
 old_pwd=$(pwd)
-script_path=$(dirname $(realpath "$0"))
+script_path=$(dirname "$(realpath "$0")")
 cd "$script_path"
 go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.36.6
-go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@v1.1
+go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@v1.6.1
 protoc -I ./ ./daemon.proto --go_out=../ --go-grpc_out=../ --experimental_allow_proto3_optional
 cd "$old_pwd"

client/server/debug.go

@@ -14,6 +14,7 @@ import (
 	"github.com/netbirdio/netbird/client/internal/debug"
 	"github.com/netbirdio/netbird/client/proto"
 	mgmProto "github.com/netbirdio/netbird/shared/management/proto"
+	"github.com/netbirdio/netbird/version"
 )
 
 // DebugBundle creates a debug bundle and returns the location.
@@ -67,6 +68,8 @@ func (s *Server) DebugBundle(_ context.Context, req *proto.DebugBundleRequest) (
 			CapturePath:    capturePath,
 			RefreshStatus:  refreshStatus,
 			ClientMetrics:  clientMetrics,
+			DaemonVersion:  version.NetbirdVersion(),
+			CliVersion:     req.CliVersion,
 		},
 		debug.BundleConfig{
 			Anonymize:         req.GetAnonymize(),

client/status/status.go

@@ -143,6 +143,7 @@ type OutputOverview struct {
 	IPv6                    string                     `json:"netbirdIpv6,omitempty" yaml:"netbirdIpv6,omitempty"`
 	PubKey                  string                     `json:"publicKey" yaml:"publicKey"`
 	KernelInterface         bool                       `json:"usesKernelInterface" yaml:"usesKernelInterface"`
+	WgPort                  int                        `json:"wireguardPort" yaml:"wireguardPort"`
 	FQDN                    string                     `json:"fqdn" yaml:"fqdn"`
 	RosenpassEnabled        bool                       `json:"quantumResistance" yaml:"quantumResistance"`
 	RosenpassPermissive     bool                       `json:"quantumResistancePermissive" yaml:"quantumResistancePermissive"`
@@ -187,6 +188,7 @@ func ConvertToStatusOutputOverview(pbFullStatus *proto.FullStatus, opts ConvertO
 		IPv6:                    pbFullStatus.GetLocalPeerState().GetIpv6(),
 		PubKey:                  pbFullStatus.GetLocalPeerState().GetPubKey(),
 		KernelInterface:         pbFullStatus.GetLocalPeerState().GetKernelInterface(),
+		WgPort:                  int(pbFullStatus.GetLocalPeerState().GetWgPort()),
 		FQDN:                    pbFullStatus.GetLocalPeerState().GetFqdn(),
 		RosenpassEnabled:        pbFullStatus.GetLocalPeerState().GetRosenpassEnabled(),
 		RosenpassPermissive:     pbFullStatus.GetLocalPeerState().GetRosenpassPermissive(),
@@ -547,6 +549,21 @@ func (o *OutputOverview) GeneralSummary(showURL bool, showRelays bool, showNameS
 		goarm = fmt.Sprintf(" (ARMv%s)", os.Getenv("GOARM"))
 	}
 
+	daemonVersion := "N/A"
+	if o.DaemonVersion != "" {
+		daemonVersion = o.DaemonVersion
+	}
+
+	cliVersion := version.NetbirdVersion()
+	if o.CliVersion != "" {
+		cliVersion = o.CliVersion
+	}
+
+	wgPortString := "N/A"
+	if o.WgPort > 0 {
+		wgPortString = fmt.Sprintf("%d", o.WgPort)
+	}
+
 	summary := fmt.Sprintf(
 		"OS: %s\n"+
 			"Daemon version: %s\n"+
@@ -560,6 +577,7 @@ func (o *OutputOverview) GeneralSummary(showURL bool, showRelays bool, showNameS
 			"NetBird IP: %s\n"+
 			"%s"+
 			"Interface type: %s\n"+
+			"Wireguard port: %s\n"+
 			"Quantum resistance: %s\n"+
 			"Lazy connection: %s\n"+
 			"SSH Server: %s\n"+
@@ -567,8 +585,8 @@ func (o *OutputOverview) GeneralSummary(showURL bool, showRelays bool, showNameS
 			"%s"+
 			"Peers count: %s\n",
 		fmt.Sprintf("%s/%s%s", goos, goarch, goarm),
-		o.DaemonVersion,
-		version.NetbirdVersion(),
+		daemonVersion,
+		cliVersion,
 		o.ProfileName,
 		managementConnString,
 		signalConnString,
@@ -578,6 +596,7 @@ func (o *OutputOverview) GeneralSummary(showURL bool, showRelays bool, showNameS
 		interfaceIP,
 		ipv6Line,
 		interfaceTypeString,
+		wgPortString,
 		rosenpassEnabledStatus,
 		lazyConnectionEnabledStatus,
 		sshServerStatus,

client/status/status_test.go

@@ -94,6 +94,7 @@ var resp = &proto.StatusResponse{
 			Ipv6:            "fd00::100",
 			PubKey:          "Some-Pub-Key",
 			KernelInterface: true,
+			WgPort:          51820,
 			Fqdn:            "some-localhost.awesome-domain.com",
 			Networks: []string{
 				"10.10.0.0/24",
@@ -210,6 +211,7 @@ var overview = OutputOverview{
 	IPv6:            "fd00::100",
 	PubKey:          "Some-Pub-Key",
 	KernelInterface: true,
+	WgPort:          51820,
 	FQDN:            "some-localhost.awesome-domain.com",
 	NSServerGroups: []NsServerGroupStateOutput{
 		{
@@ -369,6 +371,7 @@ func TestParsingToJSON(t *testing.T) {
           "netbirdIpv6": "fd00::100",
           "publicKey": "Some-Pub-Key",
           "usesKernelInterface": true,
+          "wireguardPort": 51820,
           "fqdn": "some-localhost.awesome-domain.com",
           "quantumResistance": false,
           "quantumResistancePermissive": false,
@@ -487,6 +490,7 @@ netbirdIp: 192.168.178.100/16
 netbirdIpv6: fd00::100
 publicKey: Some-Pub-Key
 usesKernelInterface: true
+wireguardPort: 51820
 fqdn: some-localhost.awesome-domain.com
 quantumResistance: false
 quantumResistancePermissive: false
@@ -579,12 +583,13 @@ FQDN: some-localhost.awesome-domain.com
 NetBird IP: 192.168.178.100/16
 NetBird IPv6: fd00::100
 Interface type: Kernel
+Wireguard port: %d
 Quantum resistance: false
 Lazy connection: false
 SSH Server: Disabled
 Networks: 10.10.0.0/24
 Peers count: 2/2 Connected
-`, lastConnectionUpdate1, lastHandshake1, lastConnectionUpdate2, lastHandshake2, runtime.GOOS, runtime.GOARCH, overview.CliVersion)
+`, lastConnectionUpdate1, lastHandshake1, lastConnectionUpdate2, lastHandshake2, runtime.GOOS, runtime.GOARCH, overview.CliVersion, overview.WgPort)
 
 	assert.Equal(t, expectedDetail, detail)
 }
@@ -604,6 +609,7 @@ FQDN: some-localhost.awesome-domain.com
 NetBird IP: 192.168.178.100/16
 NetBird IPv6: fd00::100
 Interface type: Kernel
+Wireguard port: 51820
 Quantum resistance: false
 Lazy connection: false
 SSH Server: Disabled

client/ui/client_ui.go

@@ -502,7 +502,7 @@ func (s *serviceClient) getConnectionForm() *widget.Form {
 			{Text: "Pre-shared Key", Widget: s.iPreSharedKey},
 			{Text: "Quantum-Resistance", Widget: s.sRosenpassPermissive},
 			{Text: "Interface Name", Widget: s.iInterfaceName},
-			{Text: "Interface Port", Widget: s.iInterfacePort},
+			{Text: "Interface Port", Widget: s.iInterfacePort, HintText: "If set to 0, a random free port will be used"},
 			{Text: "MTU", Widget: s.iMTU},
 			{Text: "Log File", Widget: s.iLogFile},
 		},
@@ -558,8 +558,8 @@ func (s *serviceClient) parseNumericSettings() (int64, int64, error) {
 	if err != nil {
 		return 0, 0, errors.New("invalid interface port")
 	}
-	if port < 1 || port > 65535 {
-		return 0, 0, errors.New("invalid interface port: out of range 1-65535")
+	if port < 0 || port > 65535 {
+		return 0, 0, errors.New("invalid interface port: out of range 0-65535")
 	}
 
 	var mtu int64
@@ -1438,7 +1438,7 @@ func protoConfigToConfig(cfg *proto.GetConfigResponse) *profilemanager.Config {
 	}
 
 	config.WgIface = cfg.InterfaceName
-	if cfg.WireguardPort != 0 {
+	if cfg.WireguardPort >= 0 && cfg.WireguardPort <= 65535 {
 		config.WgPort = int(cfg.WireguardPort)
 	} else {
 		config.WgPort = iface.DefaultWgPort

client/ui/debug.go

@@ -21,6 +21,7 @@ import (
 	"github.com/netbirdio/netbird/client/internal"
 	"github.com/netbirdio/netbird/client/proto"
 	uptypes "github.com/netbirdio/netbird/upload-server/types"
+	"github.com/netbirdio/netbird/version"
 )
 
 // Initial state for the debug collection
@@ -462,6 +463,7 @@ func (s *serviceClient) createDebugBundleFromCollection(
 	request := &proto.DebugBundleRequest{
 		Anonymize:  params.anonymize,
 		SystemInfo: params.systemInfo,
+		CliVersion: version.NetbirdVersion(),
 	}
 
 	if params.upload {
@@ -593,6 +595,7 @@ func (s *serviceClient) createDebugBundle(anonymize bool, systemInfo bool, uploa
 	request := &proto.DebugBundleRequest{
 		Anonymize:  anonymize,
 		SystemInfo: systemInfo,
+		CliVersion: version.NetbirdVersion(),
 	}
 
 	if uploadURL != "" {

formatter/hook/hook.go

@@ -99,6 +99,9 @@ func addFields(entry *logrus.Entry) {
 	if ctxAccountID, ok := entry.Context.Value(context.AccountIDKey).(string); ok {
 		entry.Data[context.AccountIDKey] = ctxAccountID
 	}
+	if ctxUserAgent, ok := entry.Context.Value(context.UserAgentKey).(string); ok {
+		entry.Data[context.UserAgentKey] = ctxUserAgent
+	}
 	if ctxInitiatorID, ok := entry.Context.Value(context.UserIDKey).(string); ok {
 		entry.Data[context.UserIDKey] = ctxInitiatorID
 	}

go.mod

@@ -24,13 +24,13 @@ require (
 	golang.zx2c4.com/wireguard/windows v0.5.3
 	google.golang.org/grpc v1.80.0
 	google.golang.org/protobuf v1.36.11
-	gopkg.in/natefinch/lumberjack.v2 v2.2.1
 )
 
 require (
 	fyne.io/fyne/v2 v2.7.0
 	fyne.io/systray v1.12.1-0.20260116214250-81f8e1a496f9
 	git.sr.ht/~jackmordaunt/go-toast/v2 v2.0.3
+	github.com/DeRuina/timberjack v1.4.2
 	github.com/awnumar/memguard v0.23.0
 	github.com/aws/aws-sdk-go-v2 v1.38.3
 	github.com/aws/aws-sdk-go-v2/config v1.31.6

go.sum

@@ -29,6 +29,8 @@ github.com/Azure/go-ntlmssp v0.1.0 h1:DjFo6YtWzNqNvQdrwEyr/e4nhU3vRiwenz5QX7sFz+
 github.com/Azure/go-ntlmssp v0.1.0/go.mod h1:NYqdhxd/8aAct/s4qSYZEerdPuH1liG2/X9DiVTbhpk=
 github.com/BurntSushi/toml v1.5.0 h1:W5quZX/G/csjUnuI8SUYlsHs9M38FC7znL0lIO+DvMg=
 github.com/BurntSushi/toml v1.5.0/go.mod h1:ukJfTF/6rtPPRCnwkur4qwRxa8vTRFBF0uk2lLoLwho=
+github.com/DeRuina/timberjack v1.4.2 h1:4bKlzhKdsR+2oNkgef9mqb4n11ICow8VK88RfzJPzN8=
+github.com/DeRuina/timberjack v1.4.2/go.mod h1:RLoeQrwrCGIEF8gO5nV5b/gMD0QIy7bzQhBUgpp1EqE=
 github.com/Masterminds/goutils v1.1.1 h1:5nUrii3FMTL5diU80unEVvNevw1nH4+ZV4DSLVJLSYI=
 github.com/Masterminds/goutils v1.1.1/go.mod h1:8cTjp+g8YejhMuvIA5y2vz3BpJxksy863GQaJW2MFNU=
 github.com/Masterminds/semver/v3 v3.3.0 h1:B8LGeaivUe71a5qox1ICM/JLl0NqZSW5CHyL+hmvYS0=
@@ -940,8 +942,6 @@ gopkg.in/go-playground/assert.v1 v1.2.1/go.mod h1:9RXL0bg/zibRAgZUYszZSwO/z8Y/a8
 gopkg.in/go-playground/validator.v9 v9.29.1/go.mod h1:+c9/zcJMFNgbLvly1L1V+PpxWdVbfP1avr/N00E2vyQ=
 gopkg.in/ini.v1 v1.67.0 h1:Dgnx+6+nfE+IfzjUEISNeydPJh9AXNNsWbGP9KzCsOA=
 gopkg.in/ini.v1 v1.67.0/go.mod h1:pNLf8WUiyNEtQjuu5G5vTm06TEv9tsIgeAvK8hOrP4k=
-gopkg.in/natefinch/lumberjack.v2 v2.2.1 h1:bBRl1b0OH9s/DuPhuXpNl+VtCaJXFZ5/uEFST95x9zc=
-gopkg.in/natefinch/lumberjack.v2 v2.2.1/go.mod h1:YD8tP3GAjkrDg1eZH7EGmyESg/lsYskCTPBJVb9jqSc=
 gopkg.in/square/go-jose.v2 v2.6.0 h1:NGk74WTnPKBNUhNzQX7PYcTLUjoq7mzKk2OKbvwk2iI=
 gopkg.in/square/go-jose.v2 v2.6.0/go.mod h1:M9dMgbHiYLoDGQrXy7OpJDJWiKiU//h+vD76mk0e1AI=
 gopkg.in/tomb.v1 v1.0.0-20141024135613-dd632973f1e7 h1:uRGJdciOHaEIrze2W8Q3AKkepLTh2hOroT7a+7czfdQ=

infrastructure_files/getting-started.sh

@@ -19,6 +19,46 @@ readonly MSG_SEPARATOR="=========================================="
 # Utility Functions
 ############################################
 
+check_docker_sock_perms() {
+  local sock="${DOCKER_HOST:-unix:///var/run/docker.sock}"
+  sock="${sock#unix://}"
+
+  if [[ ! -S "$sock" ]]; then
+    return 0
+  fi
+
+  if [[ ! -r "$sock" ]] || [[ ! -w "$sock" ]]; then
+    local group
+    if [[ "${OSTYPE}" == "darwin"* ]]; then
+      group="$(stat -f '%Sg' "$sock")"
+    else
+      group="$(stat -c '%G' "$sock")"
+    fi
+
+    echo "Cannot access Docker socket: $sock" > /dev/stderr
+    echo "" > /dev/stderr
+    echo "Socket permissions:" > /dev/stderr
+    ls -l "$sock" > /dev/stderr
+    echo "" > /dev/stderr
+
+    if [[ "$group" == "docker" ]]; then
+      echo "Your user may need to be added to the '$group' group:" > /dev/stderr
+      echo "  sudo usermod -aG $group \"$USER\"" > /dev/stderr
+      echo "Then log out and back in, or run this for the current shell:" > /dev/stderr
+      echo "  newgrp $group" > /dev/stderr
+      echo "Note: newgrp is temporary; usermod is the permanent group change." > /dev/stderr
+    else
+      echo "The Docker socket is owned by the '$group' group, which is not the standard 'docker' group." > /dev/stderr
+      echo "For safety, this script will not suggest adding your user to '$group'." > /dev/stderr
+      echo "Instead, either run this script with appropriate privileges (for example, via sudo) or follow Docker's post-install steps to configure access via the 'docker' group:" > /dev/stderr
+      echo "  https://docs.docker.com/engine/install/linux-postinstall/" > /dev/stderr
+    fi
+
+    exit 1
+  fi
+  return 0
+}
+
 check_docker_compose() {
   if command -v docker-compose &> /dev/null
   then
@@ -311,11 +351,12 @@ initialize_default_values() {
   NETBIRD_STUN_PORT=3478
 
   # Docker images
-  DASHBOARD_IMAGE="netbirdio/dashboard:latest"
+  DASHBOARD_IMAGE=${DASHBOARD_IMAGE:-"netbirdio/dashboard:latest"}
   # Combined server replaces separate signal, relay, and management containers
-  NETBIRD_SERVER_IMAGE="netbirdio/netbird-server:latest"
-  NETBIRD_PROXY_IMAGE="netbirdio/reverse-proxy:latest"
-
+  NETBIRD_SERVER_IMAGE=${NETBIRD_SERVER_IMAGE:-"netbirdio/netbird-server:latest"}
+  NETBIRD_PROXY_IMAGE=${NETBIRD_PROXY_IMAGE:-"netbirdio/reverse-proxy:latest"}
+  TRAEFIK_IMAGE=${TRAEFIK_IMAGE:-"traefik:v3.6"}
+  CROWDSEC_IMAGE=${CROWDSEC_IMAGE:-"crowdsecurity/crowdsec:v1.7.7"}
   # Reverse proxy configuration
   REVERSE_PROXY_TYPE="0"
   TRAEFIK_EXTERNAL_NETWORK=""
@@ -580,12 +621,15 @@ start_services_and_show_instructions() {
 }
 
 init_environment() {
+  # Check if docker compose is installed using check_docker_compose function
+  DOCKER_COMPOSE_COMMAND=$(check_docker_compose)
+  check_docker_sock_perms
+
   initialize_default_values
   configure_domain
   configure_reverse_proxy
 
   check_jq
-  DOCKER_COMPOSE_COMMAND=$(check_docker_compose)
 
   check_existing_installation
   generate_configuration_files
@@ -656,7 +700,7 @@ render_docker_compose_traefik_builtin() {
     if [[ "$ENABLE_CROWDSEC" == "true" ]]; then
       crowdsec_service="
   crowdsec:
-    image: crowdsecurity/crowdsec:v1.7.7
+    image: $CROWDSEC_IMAGE
     container_name: netbird-crowdsec
     restart: unless-stopped
     networks: [netbird]
@@ -687,7 +731,7 @@ render_docker_compose_traefik_builtin() {
 services:
   # Traefik reverse proxy (automatic TLS via Let's Encrypt)
   traefik:
-    image: traefik:v3.6
+    image: $TRAEFIK_IMAGE
     container_name: netbird-traefik
     restart: unless-stopped
     networks:
@@ -771,7 +815,7 @@ $traefik_dynamic_volume
     labels:
       - traefik.enable=true
       # gRPC router (needs h2c backend for HTTP/2 cleartext)
-      - traefik.http.routers.netbird-grpc.rule=Host(\`$NETBIRD_DOMAIN\`) && (PathPrefix(\`/signalexchange.SignalExchange/\`) || PathPrefix(\`/management.ManagementService/\`))
+      - traefik.http.routers.netbird-grpc.rule=Host(\`$NETBIRD_DOMAIN\`) && (PathPrefix(\`/signalexchange.SignalExchange/\`) || PathPrefix(\`/management.ManagementService/\`) || PathPrefix(\`/management.ProxyService/\`))
       - traefik.http.routers.netbird-grpc.entrypoints=websecure
       - traefik.http.routers.netbird-grpc.tls=true
       - traefik.http.routers.netbird-grpc.tls.certresolver=letsencrypt

management/internals/controllers/network_map/controller/controller.go

@@ -32,6 +32,7 @@ import (
 	"github.com/netbirdio/netbird/shared/management/proto"
 	"github.com/netbirdio/netbird/shared/management/status"
 	"github.com/netbirdio/netbird/util"
+	"github.com/netbirdio/netbird/version"
 )
 
 type Controller struct {
@@ -514,7 +515,7 @@ func computeForwarderPort(peers []*nbpeer.Peer, requiredVersion string) int64 {
 	for _, peer := range peers {
 
 		// Development version is always supported
-		if peer.Meta.WtVersion == "development" {
+		if version.IsDevelopmentVersion(peer.Meta.WtVersion) {
 			continue
 		}
 		peerVersion := semver.Canonical("v" + peer.Meta.WtVersion)

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

@@ -932,7 +932,11 @@ func (s *Service) validateL4Target(target *Target) error {
 	if target.TargetId == "" {
 		return errors.New("target_id is required for L4 services")
 	}
-	if target.TargetType != TargetTypeCluster && target.Port == 0 {
+	// Cluster targets resolve their upstream host:port from the target's
+	// own Host/Port fields just like the other L4 types — buildPathMappings
+	// emits net.JoinHostPort(target.Host, target.Port) for every L4
+	// target, so allowing port=0 here would let ":0" reach the proxy.
+	if target.Port == 0 {
 		return errors.New("target port is required for L4 services")
 	}
 	switch target.TargetType {

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

@@ -1176,7 +1176,12 @@ func TestValidate_HTTPClusterTarget_RequiresDirectUpstream(t *testing.T) {
 	assert.ErrorContains(t, rp.Validate(), "direct upstream disabled", "cluster target must reject direct_upstream=false")
 }
 
-func TestValidate_L4ClusterTarget(t *testing.T) {
+// TestValidate_L4ClusterTarget_RequiresPort confirms that an L4 cluster
+// target without an explicit port is rejected. buildPathMappings emits
+// net.JoinHostPort(target.Host, target.Port) for every L4 target — so
+// allowing port=0 would let the proxy ship ":0" upstreams. The port
+// requirement is the same as every other L4 target type.
+func TestValidate_L4ClusterTarget_RequiresPort(t *testing.T) {
 	rp := validProxy()
 	rp.Mode = ModeTCP
 	rp.ListenPort = 9000
@@ -1186,7 +1191,12 @@ func TestValidate_L4ClusterTarget(t *testing.T) {
 		Protocol:   "tcp",
 		Enabled:    true,
 	}}
-	require.NoError(t, rp.Validate(), "L4 cluster target must validate without an explicit port")
+	assert.ErrorContains(t, rp.Validate(), "port is required",
+		"L4 cluster target must require an explicit port like other L4 target types")
+
+	rp.Targets[0].Port = 5432
+	rp.Targets[0].Host = "db.lan"
+	require.NoError(t, rp.Validate(), "L4 cluster target with host:port must validate")
 }
 
 func TestService_Copy_RoundtripsPrivate(t *testing.T) {

management/internals/server/server.go

@@ -122,7 +122,7 @@ func (s *BaseServer) Start(ctx context.Context) error {
 	s.errCh = make(chan error, 4)
 
 	if s.autoResolveDomains {
-		s.resolveDomains(srvCtx)
+		s.ResolveDomains(srvCtx)
 	}
 
 	s.PeersManager()
@@ -398,10 +398,10 @@ func (s *BaseServer) serveGRPCWithHTTP(ctx context.Context, listener net.Listene
 	}()
 }
 
-// resolveDomains determines dnsDomain and mgmtSingleAccModeDomain based on store state.
+// ResolveDomains determines dnsDomain and mgmtSingleAccModeDomain based on store state.
 // Fresh installs use the default self-hosted domain, while existing installs reuse the
 // persisted account domain to keep addressing stable across config changes.
-func (s *BaseServer) resolveDomains(ctx context.Context) {
+func (s *BaseServer) ResolveDomains(ctx context.Context) {
 	st := s.Store()
 
 	setDefault := func(logMsg string, args ...any) {

management/internals/server/server_resolve_domains_test.go

@@ -22,7 +22,7 @@ func TestResolveDomains_FreshInstallUsesDefault(t *testing.T) {
 	srv := NewServer(&Config{NbConfig: &nbconfig.Config{}})
 	Inject[store.Store](srv, mockStore)
 
-	srv.resolveDomains(context.Background())
+	srv.ResolveDomains(context.Background())
 
 	require.Equal(t, DefaultSelfHostedDomain, srv.dnsDomain)
 	require.Equal(t, DefaultSelfHostedDomain, srv.mgmtSingleAccModeDomain)
@@ -40,7 +40,7 @@ func TestResolveDomains_ExistingInstallUsesPersistedDomain(t *testing.T) {
 	srv := NewServer(&Config{NbConfig: &nbconfig.Config{}})
 	Inject[store.Store](srv, mockStore)
 
-	srv.resolveDomains(context.Background())
+	srv.ResolveDomains(context.Background())
 
 	require.Equal(t, "vpn.mycompany.com", srv.dnsDomain)
 	require.Equal(t, "vpn.mycompany.com", srv.mgmtSingleAccModeDomain)
@@ -56,7 +56,7 @@ func TestResolveDomains_StoreErrorFallsBackToDefault(t *testing.T) {
 	srv := NewServer(&Config{NbConfig: &nbconfig.Config{}})
 	Inject[store.Store](srv, mockStore)
 
-	srv.resolveDomains(context.Background())
+	srv.ResolveDomains(context.Background())
 
 	require.Equal(t, DefaultSelfHostedDomain, srv.dnsDomain)
 	require.Equal(t, DefaultSelfHostedDomain, srv.mgmtSingleAccModeDomain)

management/internals/shared/grpc/proxy.go

@@ -666,8 +666,10 @@ func (s *ProxyServiceServer) sender(conn *proxyConnection, errChan chan<- error)
 		case resp := <-conn.sendChan:
 			if err := conn.sendResponse(resp); err != nil {
 				errChan <- err
+				log.WithContext(conn.ctx).Tracef("Failed to send response to proxy %s: %v", conn.proxyID, err)
 				return
 			}
+			log.WithContext(conn.ctx).Tracef("Send response to proxy %s", conn.proxyID)
 		case <-conn.ctx.Done():
 			return
 		}
@@ -978,6 +980,7 @@ func shallowCloneMapping(m *proto.ProxyMapping) *proto.ProxyMapping {
 		Mode:               m.Mode,
 		ListenPort:         m.ListenPort,
 		AccessRestrictions: m.AccessRestrictions,
+		Private:            m.Private,
 	}
 }
 

management/internals/shared/grpc/proxy_clone_test.go

@@ -0,0 +1,88 @@
+package grpc
+
+import (
+	"reflect"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// authTokenField is the only per-proxy field that shallowCloneMapping must NOT
+// copy from the source, since callers assign it individually after cloning.
+const authTokenField = "AuthToken"
+
+// TestShallowCloneMapping_ClonesAllFields populates every exported field of
+// ProxyMapping with a non-zero value and verifies the clone carries each one
+// (except AuthToken). It uses reflection so adding a new field to ProxyMapping
+// without updating shallowCloneMapping fails this test.
+func TestShallowCloneMapping_ClonesAllFields(t *testing.T) {
+	src := &proto.ProxyMapping{}
+	populated := populateExportedFields(t, reflect.ValueOf(src).Elem())
+	require.NotEmpty(t, populated, "ProxyMapping should expose fields to populate")
+
+	clone := shallowCloneMapping(src)
+	require.NotNil(t, clone, "clone must not be nil")
+
+	srcVal := reflect.ValueOf(src).Elem()
+	cloneVal := reflect.ValueOf(clone).Elem()
+
+	for _, name := range populated {
+		srcField := srcVal.FieldByName(name).Interface()
+		cloneField := cloneVal.FieldByName(name).Interface()
+
+		if name == authTokenField {
+			assert.Zero(t, cloneField, "AuthToken must not be cloned; it is set per proxy after cloning")
+			continue
+		}
+
+		assert.Equal(t, srcField, cloneField, "field %s must be carried over by shallowCloneMapping", name)
+	}
+}
+
+// populateExportedFields sets a non-zero value on every settable exported field
+// of the struct and returns their names.
+func populateExportedFields(t *testing.T, v reflect.Value) []string {
+	t.Helper()
+
+	var names []string
+	typ := v.Type()
+	for i := 0; i < v.NumField(); i++ {
+		field := v.Field(i)
+		structField := typ.Field(i)
+
+		if structField.PkgPath != "" || !field.CanSet() {
+			continue
+		}
+
+		setNonZero(t, field, structField.Name)
+		names = append(names, structField.Name)
+	}
+	return names
+}
+
+// setNonZero assigns a deterministic non-zero value based on the field kind.
+func setNonZero(t *testing.T, field reflect.Value, name string) {
+	t.Helper()
+
+	switch field.Kind() {
+	case reflect.String:
+		field.SetString("non-zero-" + name)
+	case reflect.Bool:
+		field.SetBool(true)
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		field.SetInt(7)
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64:
+		field.SetUint(7)
+	case reflect.Ptr:
+		field.Set(reflect.New(field.Type().Elem()))
+	case reflect.Slice:
+		field.Set(reflect.MakeSlice(field.Type(), 1, 1))
+	case reflect.Map:
+		field.Set(reflect.MakeMapWithSize(field.Type(), 0))
+	default:
+		t.Fatalf("unhandled field kind %s for field %s; extend setNonZero", field.Kind(), name)
+	}
+}

management/internals/shared/grpc/validate_session_test.go

@@ -102,7 +102,7 @@ func generateSessionKeyPair(t *testing.T) (string, string) {
 
 func createSessionToken(t *testing.T, privKeyB64, userID, domain string) string {
 	t.Helper()
-	token, err := sessionkey.SignToken(privKeyB64, userID, domain, auth.MethodOIDC, nil, time.Hour)
+	token, err := sessionkey.SignToken(privKeyB64, userID, "", domain, auth.MethodOIDC, nil, nil, time.Hour)
 	require.NoError(t, err)
 	return token
 }
@@ -394,6 +394,10 @@ func (m *testValidateSessionProxyManager) ClusterSupportsCrowdSec(_ context.Cont
 	return nil
 }
 
+func (m *testValidateSessionProxyManager) ClusterSupportsPrivate(_ context.Context, _ string) *bool {
+	return nil
+}
+
 type testValidateSessionUsersManager struct {
 	store store.Store
 }
@@ -401,3 +405,24 @@ type testValidateSessionUsersManager struct {
 func (m *testValidateSessionUsersManager) GetUser(ctx context.Context, userID string) (*types.User, error) {
 	return m.store.GetUserByUserID(ctx, store.LockingStrengthNone, userID)
 }
+
+func (m *testValidateSessionUsersManager) GetUserWithGroups(ctx context.Context, userID string) (*types.User, []*types.Group, error) {
+	user, err := m.store.GetUserByUserID(ctx, store.LockingStrengthNone, userID)
+	if err != nil {
+		return nil, nil, err
+	}
+	if len(user.AutoGroups) == 0 {
+		return user, nil, nil
+	}
+	groupsMap, err := m.store.GetGroupsByIDs(ctx, store.LockingStrengthNone, user.AccountID, user.AutoGroups)
+	if err != nil {
+		return nil, nil, err
+	}
+	groups := make([]*types.Group, 0, len(user.AutoGroups))
+	for _, id := range user.AutoGroups {
+		if g, ok := groupsMap[id]; ok && g != nil {
+			groups = append(groups, g)
+		}
+	}
+	return user, groups, nil
+}

management/server/context/keys.go

@@ -12,6 +12,7 @@ const (
 	RoleKey      = nbcontext.RoleKey
 	UserIDKey    = nbcontext.UserIDKey
 	PeerIDKey    = nbcontext.PeerIDKey
+	UserAgentKey = nbcontext.UserAgentKey
 )
 
 // RoleFromContext returns the role stored in ctx, or empty string and false if absent.

management/server/peer.go

@@ -30,6 +30,7 @@ import (
 	nbpeer "github.com/netbirdio/netbird/management/server/peer"
 	"github.com/netbirdio/netbird/management/server/telemetry"
 	"github.com/netbirdio/netbird/shared/management/status"
+	"github.com/netbirdio/netbird/version"
 )
 
 const remoteJobsMinVer = "0.64.0"
@@ -372,7 +373,7 @@ func (am *DefaultAccountManager) CreatePeerJob(ctx context.Context, accountID, p
 	}
 
 	meetMinVer, err := posture.MeetsMinVersion(remoteJobsMinVer, p.Meta.WtVersion)
-	if !strings.Contains(p.Meta.WtVersion, "dev") && (!meetMinVer || err != nil) {
+	if !version.IsDevelopmentVersion(p.Meta.WtVersion) && (!meetMinVer || err != nil) {
 		return status.Errorf(status.PreconditionFailed, "peer version %s does not meet the minimum required version %s for remote jobs", p.Meta.WtVersion, remoteJobsMinVer)
 	}
 

management/server/store/sql_store.go

@@ -1216,6 +1216,7 @@ func (s *SqlStore) getAccountGorm(ctx context.Context, accountID string) (*types
 		Preload("NetworkResources").
 		Preload("Onboarding").
 		Preload("Services.Targets").
+		Preload("Domains").
 		Take(&account, idQueryCondition, accountID)
 	if result.Error != nil {
 		log.WithContext(ctx).Errorf("error when getting account %s from the store: %s", accountID, result.Error)
@@ -1302,7 +1303,7 @@ func (s *SqlStore) getAccountPgx(ctx context.Context, accountID string) (*types.
 	}
 
 	var wg sync.WaitGroup
-	errChan := make(chan error, 12)
+	errChan := make(chan error, 16)
 
 	wg.Add(1)
 	go func() {
@@ -1403,6 +1404,17 @@ func (s *SqlStore) getAccountPgx(ctx context.Context, accountID string) (*types.
 		account.Services = services
 	}()
 
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		domains, err := s.ListCustomDomains(ctx, accountID)
+		if err != nil {
+			errChan <- err
+			return
+		}
+		account.Domains = domains
+	}()
+
 	wg.Add(1)
 	go func() {
 		defer wg.Done()
@@ -4734,7 +4746,13 @@ func (s *SqlStore) GetPeerByIP(ctx context.Context, lockStrength LockingStrength
 	result := tx.
 		Take(&peer, fmt.Sprintf("account_id = ? AND %s = ?", column), accountID, jsonValue)
 	if result.Error != nil {
-		// no logging here
+		// A tunnel-IP miss is an expected outcome (e.g. the proxy's
+		// ValidateTunnelPeer probing an address that isn't in the
+		// account roster); surface it as NotFound so callers can tell
+		// it apart from a real store failure.
+		if errors.Is(result.Error, gorm.ErrRecordNotFound) {
+			return nil, status.Errorf(status.NotFound, "peer with ip %s not found", ip.String())
+		}
 		return nil, status.Errorf(status.Internal, "failed to get peer from store")
 	}
 
@@ -5962,6 +5980,7 @@ func (s *SqlStore) getClusterCapability(ctx context.Context, clusterAddr, column
 	}
 
 	err := s.db.
+		WithContext(ctx).
 		Model(&proxy.Proxy{}).
 		Select("COUNT(CASE WHEN "+column+" IS NOT NULL THEN 1 END) > 0 AS has_capability, "+
 			"COALESCE(MAX(CASE WHEN "+column+" = true THEN 1 ELSE 0 END), 0) = 1 AS any_true").

management/server/store/sql_store_get_account_test.go

@@ -4,6 +4,8 @@ import (
 	"context"
 	"net"
 	"net/netip"
+	"os"
+	"runtime"
 	"testing"
 	"time"
 
@@ -21,6 +23,63 @@ import (
 	"github.com/netbirdio/netbird/route"
 )
 
+// TestGetAccount_LoadsCustomDomains verifies GetAccount populates account.Domains.
+// SynthesizePrivateServiceZones depends on this relation to anchor a custom-domain
+// private service's DNS zone; without the preload the relation is empty and the
+// service is silently skipped, so a custom domain never resolves on clients.
+func TestGetAccount_LoadsCustomDomains(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("The SQLite store is not properly supported by Windows yet")
+	}
+
+	store, cleanup, err := NewTestStoreFromSQL(context.Background(), "", t.TempDir())
+	require.NoError(t, err)
+	defer cleanup()
+
+	assertGetAccountLoadsCustomDomains(t, store)
+}
+
+func TestPostgresql_GetAccount_LoadsCustomDomains(t *testing.T) {
+	if (os.Getenv("CI") == "true" && runtime.GOOS == "darwin") || runtime.GOOS == "windows" {
+		t.Skip("skip CI tests on darwin and windows")
+	}
+
+	t.Setenv("NETBIRD_STORE_ENGINE", string(types.PostgresStoreEngine))
+	store, cleanup, err := NewTestStoreFromSQL(context.Background(), "", t.TempDir())
+	require.NoError(t, err)
+	t.Cleanup(cleanup)
+
+	assertGetAccountLoadsCustomDomains(t, store)
+}
+
+// assertGetAccountLoadsCustomDomains exercises both the gorm and pgx GetAccount
+// paths: it persists two custom domains and asserts the relation comes back
+// populated, which SynthesizePrivateServiceZones relies on.
+func assertGetAccountLoadsCustomDomains(t *testing.T, store Store) {
+	t.Helper()
+	ctx := context.Background()
+
+	accountID := "acct-custom-domains"
+	require.NoError(t, store.SaveAccount(ctx, newAccountWithId(ctx, accountID, "user-1", "")))
+
+	_, err := store.CreateCustomDomain(ctx, accountID, "example.com", "eu.proxy.netbird.io", true)
+	require.NoError(t, err, "creating the first custom domain must succeed")
+	_, err = store.CreateCustomDomain(ctx, accountID, "apps.acme.io", "us.proxy.netbird.io", false)
+	require.NoError(t, err, "creating the second custom domain must succeed")
+
+	account, err := store.GetAccount(ctx, accountID)
+	require.NoError(t, err)
+	require.Len(t, account.Domains, 2, "GetAccount must preload the account's custom domains")
+
+	byDomain := map[string]string{}
+	for _, d := range account.Domains {
+		require.NotNil(t, d)
+		byDomain[d.Domain] = d.TargetCluster
+	}
+	assert.Equal(t, "eu.proxy.netbird.io", byDomain["example.com"], "custom domain must carry its target cluster")
+	assert.Equal(t, "us.proxy.netbird.io", byDomain["apps.acme.io"], "custom domain must carry its target cluster")
+}
+
 // TestGetAccount_ComprehensiveFieldValidation validates that GetAccount properly loads
 // all fields and nested objects from the database, including deeply nested structures.
 func TestGetAccount_ComprehensiveFieldValidation(t *testing.T) {

management/server/store/sql_store_service_test.go

@@ -13,7 +13,7 @@ import (
 )
 
 func TestSqlStore_GetAccount_PrivateServiceRoundtrip(t *testing.T) {
-	if (os.Getenv("CI") == "true" && runtime.GOOS == "darwin") || runtime.GOOS == "windows" {
+	if os.Getenv("CI") == "true" && (runtime.GOOS == "darwin" || runtime.GOOS == "windows") {
 		t.Skip("skip CI tests on darwin and windows")
 	}
 

management/server/store/sql_store_test.go

@@ -491,6 +491,27 @@ func Test_GetAccount(t *testing.T) {
 	})
 }
 
+// TestSqlStore_GetPeerByIP_NotFound pins the not-found semantics the
+// proxy's ValidateTunnelPeer relies on: a tunnel-IP that isn't in the
+// account roster must surface as a NotFound error (not a generic
+// Internal) so callers can distinguish an expected miss from a real
+// store failure. A known IP still resolves.
+func TestSqlStore_GetPeerByIP_NotFound(t *testing.T) {
+	runTestForAllEngines(t, "../testdata/store.sql", func(t *testing.T, store Store) {
+		const accountID = "bf1c8084-ba50-4ce7-9439-34653001fc3b"
+
+		peer, err := store.GetPeerByIP(context.Background(), LockingStrengthNone, accountID, net.ParseIP("192.168.0.0"))
+		require.NoError(t, err, "known tunnel IP must resolve")
+		require.NotNil(t, peer)
+
+		_, err = store.GetPeerByIP(context.Background(), LockingStrengthNone, accountID, net.ParseIP("100.65.0.99"))
+		require.Error(t, err, "unknown tunnel IP must error")
+		parsedErr, ok := status.FromError(err)
+		require.True(t, ok, "error must be a status error")
+		require.Equal(t, status.NotFound, parsedErr.Type(), "tunnel-IP miss must be NotFound, not Internal")
+	})
+}
+
 func TestSqlStore_SavePeer(t *testing.T) {
 	store, cleanUp, err := NewTestStoreFromSQL(context.Background(), "../testdata/store.sql", t.TempDir())
 	t.Cleanup(cleanUp)

management/server/telemetry/http_api_metrics.go

@@ -21,6 +21,8 @@ const (
 	httpRequestCounterPrefix  = "management.http.request.counter"
 	httpResponseCounterPrefix = "management.http.response.counter"
 	httpRequestDurationPrefix = "management.http.request.duration.ms"
+
+	RequestIDHeader = "X-Request-Id"
 )
 
 // WrappedResponseWriter is a wrapper for http.ResponseWriter that allows the
@@ -172,6 +174,10 @@ func (m *HTTPMiddleware) Handler(h http.Handler) http.Handler {
 		reqID := xid.New().String()
 		//nolint
 		ctx = context.WithValue(ctx, nbContext.RequestIDKey, reqID)
+		//nolint
+		ctx = context.WithValue(ctx, nbContext.UserAgentKey, r.UserAgent())
+
+		rw.Header().Set(RequestIDHeader, reqID)
 
 		log.WithContext(ctx).Tracef("HTTP request %v: %v %v", reqID, r.Method, r.URL)
 

management/server/types/account.go

@@ -29,6 +29,7 @@ import (
 	"github.com/netbirdio/netbird/route"
 	"github.com/netbirdio/netbird/shared/management/domain"
 	"github.com/netbirdio/netbird/shared/management/status"
+	"github.com/netbirdio/netbird/version"
 )
 
 const (
@@ -272,7 +273,7 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 	}
 
 	peerGroups := a.GetPeerGroups(peerID)
-	zonesByCluster := map[string]*nbdns.CustomZone{}
+	zonesByApex := map[string]*nbdns.CustomZone{}
 
 	for _, svc := range a.Services {
 		if svc == nil || !svc.Enabled || !svc.Private {
@@ -289,19 +290,24 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 			continue
 		}
 
-		zone, exists := zonesByCluster[svc.ProxyCluster]
+		serviceDomainZone := a.privateServiceDomainZone(svc)
+		if serviceDomainZone == "" {
+			continue
+		}
+
+		zone, exists := zonesByApex[serviceDomainZone]
 		if !exists {
 			// NonAuthoritative makes this a match-only zone: queries for
 			// names without an explicit record fall through to the
 			// upstream resolver instead of returning NXDOMAIN. Without
 			// it, adding a single private service would black-hole every
-			// other name under the cluster apex.
+			// other name under the zone apex.
 			zone = &nbdns.CustomZone{
-				Domain:           dns.Fqdn(svc.ProxyCluster),
+				Domain:           dns.Fqdn(serviceDomainZone),
 				Records:          []nbdns.SimpleRecord{},
 				NonAuthoritative: true,
 			}
-			zonesByCluster[svc.ProxyCluster] = zone
+			zonesByApex[serviceDomainZone] = zone
 		}
 
 		emitted := 0
@@ -339,8 +345,8 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 		}
 	}
 
-	out := make([]nbdns.CustomZone, 0, len(zonesByCluster))
-	for _, zone := range zonesByCluster {
+	out := make([]nbdns.CustomZone, 0, len(zonesByApex))
+	for _, zone := range zonesByApex {
 		if len(zone.Records) == 0 {
 			continue
 		}
@@ -356,6 +362,33 @@ func (a *Account) SynthesizePrivateServiceZones(peerID string) []nbdns.CustomZon
 	return out
 }
 
+// privateServiceDomainZone returns the DNS zone name for the given private service domain by
+// looking at the proxy cluster domain then the custom domains.
+func (a *Account) privateServiceDomainZone(svc *service.Service) string {
+	if domainFromSuffix(svc.Domain, svc.ProxyCluster) {
+		return svc.ProxyCluster
+	}
+
+	// Longest matching custom domain wins
+	zoneName := ""
+	for _, d := range a.Domains {
+		if d == nil || d.TargetCluster != svc.ProxyCluster {
+			continue
+		}
+		if domainFromSuffix(svc.Domain, d.Domain) && len(d.Domain) > len(zoneName) {
+			zoneName = d.Domain
+		}
+	}
+	return zoneName
+}
+
+func domainFromSuffix(domain, suffix string) bool {
+	if suffix == "" {
+		return false
+	}
+	return domain == suffix || strings.HasSuffix(domain, "."+suffix)
+}
+
 // peerInDistributionGroups reports whether any of the peer's groups
 // matches the service's bearer-auth distribution_groups.
 func peerInDistributionGroups(peerGroups LookupMap, distributionGroups []string) bool {
@@ -1804,7 +1837,7 @@ func shouldCheckRulesForNativeSSH(supportsNative bool, rule *PolicyRule, peer *n
 
 // peerSupportedFirewallFeatures checks if the peer version supports port ranges.
 func peerSupportedFirewallFeatures(peerVer string) supportedFeatures {
-	if strings.Contains(peerVer, "dev") {
+	if version.IsDevelopmentVersion(peerVer) {
 		return supportedFeatures{true, true}
 	}
 

management/server/types/account_private_zones_test.go

@@ -11,6 +11,7 @@ import (
 	"github.com/stretchr/testify/require"
 
 	nbdns "github.com/netbirdio/netbird/dns"
+	proxydomain "github.com/netbirdio/netbird/management/internals/modules/reverseproxy/domain"
 	"github.com/netbirdio/netbird/management/internals/modules/reverseproxy/service"
 	nbpeer "github.com/netbirdio/netbird/management/server/peer"
 )
@@ -234,6 +235,113 @@ func TestPrivateZone_GetPeerNetworkMap_PeerOutsideGroups_OmitsSynthZone(t *testi
 	assert.False(t, ok, "peer outside the distribution_groups must not see the synth zone")
 }
 
+func TestSynthesizePrivateServiceZones_CustomDomain_ZoneApexIsRegisteredDomain(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	// A custom-domain service: Domain is the custom FQDN, ProxyCluster
+	// is the cluster serving it, and account.Domains holds the registered
+	// custom domain. The synth zone apex must be the registered domain,
+	// not the cluster, or the client's match-only zone never intercepts
+	// the query.
+	account.Services[0].Domain = "app.example.com"
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "eu.proxy.netbird.io", Validated: true},
+	}
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	require.Len(t, zones, 1, "custom-domain service must still produce one zone")
+	zone := zones[0]
+	assert.Equal(t, "example.com.", zone.Domain, "zone apex must be the registered custom domain, not the cluster or the service FQDN")
+	assert.True(t, zone.NonAuthoritative, "synth zone must remain match-only")
+	require.Len(t, zone.Records, 1, "custom-domain service yields one A record")
+	rec := zone.Records[0]
+	assert.Equal(t, "app.example.com.", rec.Name, "record name is the custom service FQDN")
+	assert.Equal(t, "100.64.0.99", rec.RData, "record points at the embedded proxy peer's tunnel IP")
+}
+
+func TestSynthesizePrivateServiceZones_CustomAndFreeDomain_SeparateZones(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "eu.proxy.netbird.io", Validated: true},
+	}
+	account.Services = append(account.Services, &service.Service{
+		ID:           "svc-2",
+		AccountID:    "acct-1",
+		Name:         "custom",
+		Domain:       "app.example.com",
+		ProxyCluster: "eu.proxy.netbird.io",
+		Enabled:      true,
+		Private:      true,
+		Mode:         service.ModeHTTP,
+		AccessGroups: []string{"grp-admins"},
+	})
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	require.Len(t, zones, 2, "a free-domain and a custom-domain service must not collapse into one zone")
+
+	free, ok := findCustomZone(zones, "eu.proxy.netbird.io")
+	require.True(t, ok, "free-domain service keeps the shared cluster-apex zone")
+	require.Len(t, free.Records, 1, "cluster zone carries only the free-domain record")
+	assert.Equal(t, "myapp.eu.proxy.netbird.io.", free.Records[0].Name, "cluster zone record is the free-domain FQDN")
+
+	custom, ok := findCustomZone(zones, "example.com")
+	require.True(t, ok, "custom-domain service gets its own zone at the registered custom domain apex")
+	require.Len(t, custom.Records, 1, "custom zone carries only the custom-domain record")
+	assert.Equal(t, "app.example.com.", custom.Records[0].Name, "custom zone record is the custom-domain FQDN")
+}
+
+func TestSynthesizePrivateServiceZones_TwoServicesSameCustomDomain_OneZone(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "eu.proxy.netbird.io", Validated: true},
+	}
+	account.Services[0].Domain = "a.example.com"
+	account.Services = append(account.Services, &service.Service{
+		ID:           "svc-2",
+		AccountID:    "acct-1",
+		Name:         "bapp",
+		Domain:       "b.example.com",
+		ProxyCluster: "eu.proxy.netbird.io",
+		Enabled:      true,
+		Private:      true,
+		Mode:         service.ModeHTTP,
+		AccessGroups: []string{"grp-admins"},
+	})
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	require.Len(t, zones, 1, "two services under the same registered custom domain must share one zone")
+	assert.Equal(t, "example.com.", zones[0].Domain, "shared zone apex is the registered custom domain")
+	require.Len(t, zones[0].Records, 2, "both services surface as records in the shared custom-domain zone")
+	names := []string{zones[0].Records[0].Name, zones[0].Records[1].Name}
+	assert.ElementsMatch(t, []string{"a.example.com.", "b.example.com."}, names, "both custom-domain service FQDNs must surface")
+}
+
+func TestSynthesizePrivateServiceZones_CustomDomainNotRegistered_NoZone(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	// Service domain is outside the cluster and no account.Domains entry
+	// covers it: there is no apex that would intercept the query, so the
+	// service must be skipped rather than emit an unmatchable record.
+	account.Services[0].Domain = "app.example.com"
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	assert.Empty(t, zones, "a custom-domain service with no registered domain apex must not produce a zone")
+}
+
+func TestSynthesizePrivateServiceZones_CustomDomainClusterMismatch_NoZone(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	// The registered custom domain matches the service FQDN by suffix but
+	// targets a different cluster than the service's ProxyCluster. It must
+	// be ignored, leaving no apex to intercept the query — otherwise the
+	// zone would point at this cluster's proxy peers under a domain owned
+	// by a different cluster.
+	account.Services[0].Domain = "app.example.com"
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "us.proxy.netbird.io", Validated: true},
+	}
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	assert.Empty(t, zones, "a custom domain targeting a different cluster must not anchor the service zone")
+}
+
 func TestSynthesizePrivateServiceZones_TwoServicesSameCluster_OneZone(t *testing.T) {
 	account := privateZoneTestAccount(t)
 	account.Services = append(account.Services, &service.Service{
@@ -254,3 +362,72 @@ func TestSynthesizePrivateServiceZones_TwoServicesSameCluster_OneZone(t *testing
 	names := []string{zones[0].Records[0].Name, zones[0].Records[1].Name}
 	assert.ElementsMatch(t, []string{"myapp.eu.proxy.netbird.io.", "anotherapp.eu.proxy.netbird.io."}, names, "both service domains must surface")
 }
+
+func TestSynthesizePrivateServiceZones_MixedClusterCustomAndPublic(t *testing.T) {
+	account := privateZoneTestAccount(t)
+	account.Domains = []*proxydomain.Domain{
+		{Domain: "example.com", AccountID: "acct-1", TargetCluster: "eu.proxy.netbird.io", Validated: true},
+	}
+
+	privateService := func(id, domain string) *service.Service {
+		return &service.Service{
+			ID:           id,
+			AccountID:    "acct-1",
+			Name:         id,
+			Domain:       domain,
+			ProxyCluster: "eu.proxy.netbird.io",
+			Enabled:      true,
+			Private:      true,
+			Mode:         service.ModeHTTP,
+			AccessGroups: []string{"grp-admins"},
+		}
+	}
+	publicService := func(id, domain string) *service.Service {
+		s := privateService(id, domain)
+		s.Private = false
+		return s
+	}
+
+	account.Services = []*service.Service{
+		// 3 private services under the cluster suffix.
+		privateService("cluster-1", "cluster1.eu.proxy.netbird.io"),
+		privateService("cluster-2", "cluster2.eu.proxy.netbird.io"),
+		privateService("cluster-3", "cluster3.eu.proxy.netbird.io"),
+		// 4 private services under the custom domain suffix.
+		privateService("custom-1", "custom1.example.com"),
+		privateService("custom-2", "custom2.example.com"),
+		privateService("custom-3", "custom3.example.com"),
+		privateService("custom-4", "custom4.example.com"),
+		// 2 public services, one per suffix, must not surface.
+		publicService("public-cluster", "public.eu.proxy.netbird.io"),
+		publicService("public-custom", "public.example.com"),
+	}
+
+	zones := account.SynthesizePrivateServiceZones("user-peer")
+	require.Len(t, zones, 2, "one zone per apex: the cluster apex and the custom domain apex")
+
+	cluster, ok := findCustomZone(zones, "eu.proxy.netbird.io")
+	require.True(t, ok, "cluster-suffix services collapse into the cluster-apex zone")
+	clusterNames := recordNames(cluster)
+	assert.ElementsMatch(t,
+		[]string{"cluster1.eu.proxy.netbird.io.", "cluster2.eu.proxy.netbird.io.", "cluster3.eu.proxy.netbird.io."},
+		clusterNames,
+		"only the 3 private cluster services surface in the cluster zone (public one excluded)")
+
+	custom, ok := findCustomZone(zones, "example.com")
+	require.True(t, ok, "custom-suffix services collapse into the custom-domain-apex zone")
+	customNames := recordNames(custom)
+	assert.ElementsMatch(t,
+		[]string{"custom1.example.com.", "custom2.example.com.", "custom3.example.com.", "custom4.example.com."},
+		customNames,
+		"only the 4 private custom services surface in the custom zone (public one excluded)")
+}
+
+// recordNames returns the record names of a zone for order-independent assertions.
+func recordNames(zone nbdns.CustomZone) []string {
+	names := make([]string, 0, len(zone.Records))
+	for _, r := range zone.Records {
+		names = append(names, r.Name)
+	}
+	return names
+}

management/server/types/account_test.go

@@ -646,41 +646,7 @@ func Test_ExpandPortsAndRanges_SSHRuleExpansion(t *testing.T) {
 			expectedPorts: []string{"20-25", "10-100", "22022"},
 		},
 		{
-			name: "dev suffix version supports all features",
-			peer: &nbpeer.Peer{
-				ID:         "peer1",
-				SSHEnabled: true,
-				Meta: nbpeer.PeerSystemMeta{
-					WtVersion: "0.50.0-dev",
-					Flags:     nbpeer.Flags{ServerSSHAllowed: true},
-				},
-			},
-			rule: &PolicyRule{
-				Protocol: PolicyRuleProtocolTCP,
-				Ports:    []string{"22"},
-			},
-			base:          FirewallRule{PeerIP: "10.0.0.1", Direction: 0, Action: "accept", Protocol: "tcp"},
-			expectedPorts: []string{"22", "22022"},
-		},
-		{
-			name: "dev suffix version supports all features",
-			peer: &nbpeer.Peer{
-				ID:         "peer1",
-				SSHEnabled: true,
-				Meta: nbpeer.PeerSystemMeta{
-					WtVersion: "dev",
-					Flags:     nbpeer.Flags{ServerSSHAllowed: true},
-				},
-			},
-			rule: &PolicyRule{
-				Protocol: PolicyRuleProtocolTCP,
-				Ports:    []string{"22"},
-			},
-			base:          FirewallRule{PeerIP: "10.0.0.1", Direction: 0, Action: "accept", Protocol: "tcp"},
-			expectedPorts: []string{"22", "22022"},
-		},
-		{
-			name: "development suffix version supports all features",
+			name: "development version supports all features",
 			peer: &nbpeer.Peer{
 				ID:         "peer1",
 				SSHEnabled: true,

management/server/types/networkmap_components.go

@@ -557,7 +557,6 @@ func (c *NetworkMapComponents) getRoutingPeerRoutes(peerID string) (enabledRoute
 	return enabledRoutes, disabledRoutes
 }
 
-
 func (c *NetworkMapComponents) filterRoutesByGroups(routes []*route.Route, groupListMap LookupMap) []*route.Route {
 	var filteredRoutes []*route.Route
 	for _, r := range routes {
@@ -628,9 +627,14 @@ func (c *NetworkMapComponents) getDefaultPermit(r *route.Route, includeIPv6 bool
 
 	rules := []*RouteFirewallRule{&rule}
 
-	if includeIPv6 && r.IsDynamic() {
+	isDefaultV4 := r.Network.Addr().Is4() && r.Network.Bits() == 0
+	if includeIPv6 && (r.IsDynamic() || isDefaultV4) {
 		ruleV6 := rule
 		ruleV6.SourceRanges = []string{"::/0"}
+		if isDefaultV4 {
+			ruleV6.Destination = "::/0"
+			ruleV6.RouteID = r.ID + "-v6-default"
+		}
 		rules = append(rules, &ruleV6)
 	}
 

management/server/types/networkmap_components_correctness_test.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"net"
 	"net/netip"
+	"slices"
 	"testing"
 	"time"
 
@@ -1029,6 +1030,48 @@ func TestComponents_RouteDefaultPermit(t *testing.T) {
 	assert.True(t, hasDefaultPermit, "route without ACG should have default permit rule with 0.0.0.0/0 source")
 }
 
+// TestComponents_ExitNodeDefaultPermitIPv6 verifies that a default exit node route
+// (0.0.0.0/0) without AccessControlGroups also emits an IPv6 default permit rule
+// (::/0 source and destination) for peers that support IPv6, mirroring the route
+// the client installs. Without it, IPv6 traffic is routed to the exit node but
+// dropped at the forward chain.
+func TestComponents_ExitNodeDefaultPermitIPv6(t *testing.T) {
+	account, validatedPeers := scalableTestAccount(20, 2)
+
+	routingPeerID := "peer-5"
+	routingPeer := account.Peers[routingPeerID]
+	routingPeer.IPv6 = netip.MustParseAddr("fd00::5")
+	routingPeer.Meta.Capabilities = append(routingPeer.Meta.Capabilities, nbpeer.PeerCapabilityIPv6Overlay)
+
+	account.Routes["route-exit"] = &route.Route{
+		ID: "route-exit", Network: netip.MustParsePrefix("0.0.0.0/0"),
+		PeerID: routingPeerID, Peer: routingPeer.Key,
+		Enabled: true, Groups: []string{"group-all"}, PeerGroups: []string{"group-0"},
+		AccessControlGroups: []string{},
+		AccountID:           "test-account",
+	}
+
+	nm := componentsNetworkMap(account, routingPeerID, validatedPeers)
+	require.NotNil(t, nm)
+
+	hasV4 := false
+	hasV6 := false
+	for _, rfr := range nm.RoutesFirewallRules {
+		switch rfr.Destination {
+		case "0.0.0.0/0":
+			if slices.Contains(rfr.SourceRanges, "0.0.0.0/0") {
+				hasV4 = true
+			}
+		case "::/0":
+			if slices.Contains(rfr.SourceRanges, "::/0") {
+				hasV6 = true
+			}
+		}
+	}
+	assert.True(t, hasV4, "exit node route should have an IPv4 default permit rule (0.0.0.0/0)")
+	assert.True(t, hasV6, "exit node route should have an IPv6 default permit rule (::/0)")
+}
+
 // ──────────────────────────────────────────────────────────────────────────────
 // 15. MULTIPLE ROUTERS PER NETWORK
 // ──────────────────────────────────────────────────────────────────────────────

proxy/cmd/proxy/cmd/root.go

@@ -214,7 +214,10 @@ func runServer(cmd *cobra.Command, args []string) error {
 		return fmt.Errorf("invalid --trusted-proxies: %w", err)
 	}
 
-	srv := proxy.New(proxy.Config{
+	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGTERM, syscall.SIGINT)
+	defer stop()
+
+	srv := proxy.New(ctx, proxy.Config{
 		ListenAddr:               addr,
 		Logger:                   logger,
 		Version:                  Version,
@@ -246,14 +249,12 @@ func runServer(cmd *cobra.Command, args []string) error {
 		Private:                  private,
 		MaxDialTimeout:           maxDialTimeout,
 		MaxSessionIdleTimeout:    maxSessionIdleTimeout,
+		MappingBatchWatchdog:     envDurationOrDefault("NB_PROXY_MAPPING_BATCH_WATCHDOG", 0),
 		GeoDataDir:               geoDataDir,
 		CrowdSecAPIURL:           crowdsecAPIURL,
 		CrowdSecAPIKey:           crowdsecAPIKey,
 	})
 
-	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGTERM, syscall.SIGINT)
-	defer stop()
-
 	return srv.ListenAndServe(ctx, addr)
 }
 

proxy/inbound.go

@@ -5,6 +5,7 @@ import (
 	"crypto/tls"
 	"errors"
 	"fmt"
+	"io"
 	stdlog "log"
 	"net"
 	"net/http"
@@ -42,7 +43,7 @@ const privateInboundPortHTTPS = 443
 const privateInboundPortHTTP = 80
 
 // inboundManager wires per-account inbound listeners into the proxy
-// pipeline when --private-inbound is enabled. When disabled the manager
+// pipeline when --private is enabled. When disabled the manager
 // is nil and every method on *Server that touches it short-circuits.
 type inboundManager struct {
 	logger    *log.Logger
@@ -55,15 +56,18 @@ type inboundManager struct {
 }
 
 // inboundEntry owns the listeners, router and HTTP servers for a single
-// account's embedded netstack.
+// account's embedded netstack. errorLogWriters retain the logrus pipe
+// writers backing each http.Server's ErrorLog so tearDown can close
+// them — otherwise the pipe + its scanner goroutine leak per account.
 type inboundEntry struct {
-	router        *nbtcp.Router
-	tlsListener   net.Listener
-	plainListener net.Listener
-	httpsServer   *http.Server
-	httpServer    *http.Server
-	cancel        context.CancelFunc
-	wg            sync.WaitGroup
+	router          *nbtcp.Router
+	tlsListener     net.Listener
+	plainListener   net.Listener
+	httpsServer     *http.Server
+	httpServer      *http.Server
+	errorLogWriters []*io.PipeWriter
+	cancel          context.CancelFunc
+	wg              sync.WaitGroup
 }
 
 // pendingInboundRoute holds a route that arrived before the account's
@@ -147,30 +151,34 @@ func (m *inboundManager) bringUp(ctx context.Context, accountID types.AccountID,
 		return types.WithOverlayOrigin(ctx)
 	}
 
+	httpsErrLog, httpsErrW := newInboundErrorLog(m.logger, "https", accountID)
+	httpErrLog, httpErrW := newInboundErrorLog(m.logger, "http", accountID)
+
 	httpsServer := &http.Server{
 		Handler:           scopedHandler,
 		TLSConfig:         m.tlsConfig,
 		ReadHeaderTimeout: httpInboundReadHeaderTimeout,
 		IdleTimeout:       httpInboundIdleTimeout,
-		ErrorLog:          newInboundErrorLog(m.logger, "https", accountID),
+		ErrorLog:          httpsErrLog,
 		ConnContext:       markOverlayOrigin,
 	}
 	httpServer := &http.Server{
 		Handler:           scopedHandler,
 		ReadHeaderTimeout: httpInboundReadHeaderTimeout,
 		IdleTimeout:       httpInboundIdleTimeout,
-		ErrorLog:          newInboundErrorLog(m.logger, "http", accountID),
+		ErrorLog:          httpErrLog,
 		ConnContext:       markOverlayOrigin,
 	}
 
 	runCtx, cancel := context.WithCancel(ctx)
 	entry := &inboundEntry{
-		router:        router,
-		tlsListener:   tlsListener,
-		plainListener: plainListener,
-		httpsServer:   httpsServer,
-		httpServer:    httpServer,
-		cancel:        cancel,
+		router:          router,
+		tlsListener:     tlsListener,
+		plainListener:   plainListener,
+		httpsServer:     httpsServer,
+		httpServer:      httpServer,
+		errorLogWriters: []*io.PipeWriter{httpsErrW, httpErrW},
+		cancel:          cancel,
 	}
 
 	entry.wg.Add(1)
@@ -237,6 +245,14 @@ func (m *inboundManager) tearDown(accountID types.AccountID, entry *inboundEntry
 		m.logger.Debugf("close per-account plain listener: %v", err)
 	}
 	entry.wg.Wait()
+	// Close the ErrorLog pipes only after the http.Servers have fully
+	// stopped so any straggling stdlib write doesn't race with the
+	// close. Each writer also tears down the logrus scanner goroutine.
+	for _, w := range entry.errorLogWriters {
+		if err := w.Close(); err != nil {
+			m.logger.Debugf("close per-account inbound error log writer: %v", err)
+		}
+	}
 }
 
 // AddRoute records an SNI/host route on the account's per-account router.
@@ -374,7 +390,7 @@ func (m *inboundManager) ListenerInfo(accountID types.AccountID) (InboundListene
 }
 
 // Snapshot returns the inbound listener state for every account that has
-// a live listener at call time. Empty when --private-inbound is off or
+// a live listener at call time. Empty when --private is off or
 // no accounts have come up yet.
 func (m *inboundManager) Snapshot() map[types.AccountID]InboundListenerInfo {
 	if m == nil {
@@ -497,7 +513,7 @@ func accountTunnelLookup(client *embed.Client) auth.TunnelLookupFunc {
 // peerstore lookup to every request's context before delegating to next.
 // Calling on the host-level listener is a no-op because that path never
 // installs this wrapper, so the existing behaviour stays byte-for-byte
-// identical when --private-inbound is off or the request didn't arrive
+// identical when --private is off or the request didn't arrive
 // on a per-account listener.
 func withTunnelLookup(next http.Handler, lookup auth.TunnelLookupFunc) http.Handler {
 	if lookup == nil {
@@ -538,10 +554,14 @@ func (a inboundDebugAdapter) InboundListeners() map[types.AccountID]debug.Inboun
 }
 
 // newInboundErrorLog routes a per-account http.Server's stdlib error
-// stream through logrus at warn level.
-func newInboundErrorLog(logger *log.Logger, scheme string, accountID types.AccountID) *stdlog.Logger {
-	return stdlog.New(logger.WithFields(log.Fields{
+// stream through logrus at warn level. The returned PipeWriter must be
+// closed by the caller (tearDown) once the http.Server has shut down —
+// otherwise the pipe and its scanner goroutine leak per account, see
+// logrus.Entry.WriterLevel.
+func newInboundErrorLog(logger *log.Logger, scheme string, accountID types.AccountID) (*stdlog.Logger, *io.PipeWriter) {
+	w := logger.WithFields(log.Fields{
 		"inbound-http": scheme,
 		"account_id":   accountID,
-	}).WriterLevel(log.WarnLevel), "", 0)
+	}).WriterLevel(log.WarnLevel)
+	return stdlog.New(w, "", 0), w
 }

proxy/inbound_test.go

@@ -4,6 +4,7 @@ import (
 	"bufio"
 	"context"
 	"crypto/tls"
+	"io"
 	"net"
 	"net/http"
 	"net/http/httptest"
@@ -110,7 +111,7 @@ func TestServer_PrivateInbound_Enabled_WiresLifecycle(t *testing.T) {
 	// Construct a NetBird transport. We can't actually start the embedded
 	// client here (that needs a real management server), but we can
 	// confirm that the lifecycle callbacks are registered.
-	s.netbird = roundtrip.NewNetBird("test", "test", roundtrip.ClientConfig{
+	s.netbird = roundtrip.NewNetBird(t.Context(), "test", "test", roundtrip.ClientConfig{
 		MgmtAddr: "http://invalid.test",
 	}, quietLogger(), nil, fakeMgmtClient{})
 
@@ -139,7 +140,7 @@ func TestInboundManager_AddRouteAfterReady_RegistersDirectly(t *testing.T) {
 
 // TestPrivateCapability_DerivedFromPrivateOnly tests that the capability
 // bit reported upstream tracks --private exclusively. The previous
-// --private-inbound flag has been folded into --private.
+// --private flag has been folded into --private.
 func TestPrivateCapability_DerivedFromPrivateOnly(t *testing.T) {
 	tests := []struct {
 		name     string
@@ -318,7 +319,7 @@ func TestInboundManager_ListenerInfo(t *testing.T) {
 }
 
 // TestInboundManager_NilManagerSafe ensures the observability accessors
-// are safe to call when --private-inbound is off (nil manager).
+// are safe to call when --private is off (nil manager).
 func TestInboundManager_NilManagerSafe(t *testing.T) {
 	var mgr *inboundManager
 	_, ok := mgr.ListenerInfo("anything")
@@ -482,6 +483,38 @@ func selfSignedTLSConfig(t *testing.T) *tls.Config {
 	return &tls.Config{Certificates: []tls.Certificate{cert}, MinVersion: tls.VersionTLS12} //nolint:gosec
 }
 
+// TestNewInboundErrorLog_WriterIsCloseable guards the close path on the
+// logrus PipeWriter that backs each per-account http.Server's ErrorLog.
+// logrus.Entry.WriterLevel returns an *io.PipeWriter that owns a pipe +
+// scanner goroutine; the caller must Close() it on teardown or the
+// resources leak per account. The contract is verified two ways:
+//
+//   - the constructor returns a non-nil writer the caller can keep,
+//   - writing to the writer after Close() fails with io.ErrClosedPipe,
+//     which is the only externally observable sign that Close was wired.
+//
+// A leaking refactor (forgetting to thread the writer to tearDown, or
+// dropping the Close call) would still pass this test individually but
+// fail an integration goleak check; this unit test is the cheap first
+// line of defence.
+func TestNewInboundErrorLog_WriterIsCloseable(t *testing.T) {
+	logger := quietLogger()
+	stdLog, writer := newInboundErrorLog(logger, "https", types.AccountID("acct-1"))
+
+	require.NotNil(t, stdLog, "newInboundErrorLog must return a non-nil *log.Logger")
+	require.NotNil(t, writer, "newInboundErrorLog must return the underlying PipeWriter so tearDown can Close it")
+
+	// First Close succeeds.
+	require.NoError(t, writer.Close(), "PipeWriter.Close should succeed the first time")
+
+	// After Close, the writer must refuse new writes — that's the only
+	// behavioural signal that the pipe (and its scanner goroutine) has
+	// shut down.
+	_, err := writer.Write([]byte("post-close write\n"))
+	require.ErrorIs(t, err, io.ErrClosedPipe,
+		"writes after Close must surface io.ErrClosedPipe so callers know the writer is gone")
+}
+
 // testCertPEM / testKeyPEM are a minimal RSA self-signed cert for
 // 127.0.0.1 — only used by tests that need a working TLS handshake.
 var testCertPEM = []byte(`-----BEGIN CERTIFICATE-----

proxy/internal/auth/middleware.go

@@ -346,13 +346,15 @@ func (mw *Middleware) forwardWithSessionCookie(w http.ResponseWriter, r *http.Re
 // management unreachable, peer unknown, user not in group) returns false so
 // the caller falls back to the existing OIDC scheme dispatch.
 //
-// Phase 3 adds a local-first short-circuit: when the request arrived on a
-// per-account inbound listener the context carries a peerstore lookup
-// (TunnelLookupFromContext). If the lookup says the IP isn't in the account's
-// roster the proxy denies fast without calling management. If the lookup
-// confirms a known peer the RPC still runs for the user-identity tail
-// (UserID + group access), but its result is cached for tunnelCacheTTL so
-// repeat requests skip management entirely.
+// The fast-path is gated on TunnelLookupFromContext(r.Context()) being
+// present — that context value is attached only by the per-account
+// inbound (overlay) listener. The host listener never sets it, so a
+// public client whose source IP happens to fall inside an RFC1918 / ULA
+// / CGNAT range can't impersonate a mesh peer by colliding with a
+// tunnel-IP. Once we know the request arrived over WireGuard the
+// per-account peerstore lookup is consulted: a miss denies fast (no
+// management round-trip), a hit gates the cached ValidateTunnelPeer RPC
+// that mints the session JWT.
 func (mw *Middleware) forwardWithTunnelPeer(w http.ResponseWriter, r *http.Request, host string, config DomainConfig, next http.Handler) bool {
 	if mw.sessionValidator == nil {
 		return false
@@ -361,18 +363,24 @@ func (mw *Middleware) forwardWithTunnelPeer(w http.ResponseWriter, r *http.Reque
 	if !clientIP.IsValid() {
 		return false
 	}
+
+	// Anti-spoof: only honour the tunnel-peer fast-path on requests that
+	// were stamped by an overlay listener. Without that marker an
+	// attacker could send a request from a colliding RFC1918 / CGNAT
+	// source on the public listener and bypass operator auth.
+	lookup := TunnelLookupFromContext(r.Context())
+	if lookup == nil {
+		return false
+	}
 	if !isTunnelSourceIP(clientIP) {
 		return false
 	}
-
-	if lookup := TunnelLookupFromContext(r.Context()); lookup != nil {
-		if _, ok := lookup(clientIP); !ok {
-			mw.logger.WithFields(log.Fields{
-				"host":   host,
-				"remote": clientIP,
-			}).Debug("local peerstore: tunnel IP not in account roster; denying without RPC")
-			return false
-		}
+	if _, ok := lookup(clientIP); !ok {
+		mw.logger.WithFields(log.Fields{
+			"host":   host,
+			"remote": clientIP,
+		}).Debug("local peerstore: tunnel IP not in account roster; denying without RPC")
+		return false
 	}
 
 	resp, _, err := mw.tunnelCache.fetch(r.Context(), tunnelCacheKey{

proxy/internal/auth/middleware_test.go

@@ -1227,3 +1227,93 @@ func TestProtect_NonOIDCSchemes_PlainHTTP_NotBlocked(t *testing.T) {
 
 	assert.Equal(t, http.StatusUnauthorized, rec.Code, "PIN-only domain should serve the login page on plain HTTP")
 }
+
+// stubTunnelValidator records ValidateTunnelPeer calls so a test can
+// assert whether the fast-path reached management.
+type stubTunnelValidator struct {
+	called bool
+	resp   *proto.ValidateTunnelPeerResponse
+}
+
+func (s *stubTunnelValidator) ValidateSession(context.Context, *proto.ValidateSessionRequest, ...grpc.CallOption) (*proto.ValidateSessionResponse, error) {
+	return nil, errors.New("not used in this test")
+}
+
+func (s *stubTunnelValidator) ValidateTunnelPeer(context.Context, *proto.ValidateTunnelPeerRequest, ...grpc.CallOption) (*proto.ValidateTunnelPeerResponse, error) {
+	s.called = true
+	return s.resp, nil
+}
+
+// TestProtect_TunnelPeerFastPath_RequiresInboundMarker guards the
+// anti-spoof gate: a request with an RFC1918 source IP arriving on the
+// public listener (no TunnelLookupFromContext attached) must not be
+// allowed to take the tunnel-peer fast-path. Without this gate a public
+// client whose source IP happens to fall inside an RFC1918 range could
+// bypass the configured auth scheme by colliding with a known tunnel
+// IP.
+func TestProtect_TunnelPeerFastPath_RequiresInboundMarker(t *testing.T) {
+	validator := &stubTunnelValidator{
+		resp: &proto.ValidateTunnelPeerResponse{
+			Valid:        true,
+			SessionToken: "should-not-be-used",
+			UserId:       "user-1",
+		},
+	}
+	mw := NewMiddleware(log.StandardLogger(), validator, nil)
+	kp := generateTestKeyPair(t)
+
+	scheme := &stubScheme{method: auth.MethodPIN, promptID: "pin"}
+	require.NoError(t, mw.AddDomain("example.com", []Scheme{scheme}, kp.PublicKey, time.Hour, "", "", nil, false))
+
+	handler := mw.Protect(newPassthroughHandler())
+
+	// Request from an RFC1918 source IP on the public listener — no
+	// TunnelLookupFromContext attached. The fast-path must reject this
+	// and fall through to the PIN scheme (which renders 401 on plain
+	// HTTP for a non-authenticated request).
+	req := httptest.NewRequest(http.MethodGet, "http://example.com/", nil)
+	req.RemoteAddr = "100.64.0.5:5000"
+	rec := httptest.NewRecorder()
+	handler.ServeHTTP(rec, req)
+
+	assert.False(t, validator.called,
+		"ValidateTunnelPeer must not be invoked when the request lacks the inbound TunnelLookup marker")
+	assert.Equal(t, http.StatusUnauthorized, rec.Code,
+		"without the inbound marker the request must fall through to the operator auth scheme")
+}
+
+// TestProtect_TunnelPeerFastPath_TakesPathWithInboundMarker verifies
+// the positive side: a request marked as overlay-origin (carrying the
+// TunnelLookup context value) and matching a tunnel-IP range does take
+// the fast-path and reach management.
+func TestProtect_TunnelPeerFastPath_TakesPathWithInboundMarker(t *testing.T) {
+	validator := &stubTunnelValidator{
+		resp: &proto.ValidateTunnelPeerResponse{
+			Valid:        true,
+			SessionToken: "tunnel-session-token",
+			UserId:       "user-1",
+		},
+	}
+	mw := NewMiddleware(log.StandardLogger(), validator, nil)
+	kp := generateTestKeyPair(t)
+
+	scheme := &stubScheme{method: auth.MethodPIN, promptID: "pin"}
+	require.NoError(t, mw.AddDomain("example.com", []Scheme{scheme}, kp.PublicKey, time.Hour, "", "", nil, false))
+
+	handler := mw.Protect(newPassthroughHandler())
+
+	lookup := TunnelLookupFunc(func(_ netip.Addr) (PeerIdentity, bool) {
+		return PeerIdentity{}, true
+	})
+
+	req := httptest.NewRequest(http.MethodGet, "http://example.com/", nil)
+	req.RemoteAddr = "100.64.0.5:5000"
+	req = req.WithContext(WithTunnelLookup(req.Context(), lookup))
+	rec := httptest.NewRecorder()
+	handler.ServeHTTP(rec, req)
+
+	assert.True(t, validator.called,
+		"ValidateTunnelPeer must run when the request carries the inbound TunnelLookup marker")
+	assert.Equal(t, http.StatusOK, rec.Code,
+		"a successful tunnel-peer validation must forward to the next handler")
+}

proxy/internal/auth/tunnel_lookup_test.go

@@ -101,7 +101,10 @@ func TestForwardWithTunnelPeer_GroupsPropagateToCapturedData(t *testing.T) {
 
 	w, r := newTunnelRequest("100.64.0.10:55555")
 	cd := proxy.NewCapturedData("")
-	r = r.WithContext(proxy.WithCapturedData(r.Context(), cd))
+	lookup := TunnelLookupFunc(func(_ netip.Addr) (PeerIdentity, bool) {
+		return PeerIdentity{}, true
+	})
+	r = r.WithContext(proxy.WithCapturedData(WithTunnelLookup(r.Context(), lookup), cd))
 
 	called := false
 	next := http.HandlerFunc(func(http.ResponseWriter, *http.Request) { called = true })
@@ -148,9 +151,13 @@ func TestForwardWithTunnelPeer_LocalLookupKnownPeerStillRPCs(t *testing.T) {
 	assert.Equal(t, int32(1), validator.tunnelCalls.Load(), "RPC must run for the user-identity tail when local lookup confirms the peer")
 }
 
-// TestForwardWithTunnelPeer_NoLookupKeepsLegacyPath ensures the existing
-// behaviour stays intact on the host-level listener (no lookup attached).
-func TestForwardWithTunnelPeer_NoLookupKeepsLegacyPath(t *testing.T) {
+// TestForwardWithTunnelPeer_NoLookupRefusesFastPath guards the
+// anti-spoof gate: requests that didn't arrive on the per-account
+// inbound listener (no TunnelLookup attached) must never reach
+// management's ValidateTunnelPeer, even when the source IP looks like
+// a tunnel address. A colliding RFC1918 / CGNAT source on the public
+// listener would otherwise impersonate a mesh peer.
+func TestForwardWithTunnelPeer_NoLookupRefusesFastPath(t *testing.T) {
 	validator := &stubSessionValidator{
 		respFn: func(_ *proto.ValidateTunnelPeerRequest) *proto.ValidateTunnelPeerResponse {
 			return &proto.ValidateTunnelPeerResponse{Valid: true, SessionToken: "tok", UserId: "user-1"}
@@ -165,9 +172,9 @@ func TestForwardWithTunnelPeer_NoLookupKeepsLegacyPath(t *testing.T) {
 	config, _ := mw.getDomainConfig("svc.example")
 	handled := mw.forwardWithTunnelPeer(w, r, "svc.example", config, next)
 
-	assert.True(t, handled, "host-level path forwards on positive RPC result")
-	assert.True(t, called, "next handler runs on host-level success")
-	assert.Equal(t, int32(1), validator.tunnelCalls.Load(), "host-level path always RPCs (Phase 3 unchanged)")
+	assert.False(t, handled, "fast-path must refuse without the inbound marker")
+	assert.False(t, called, "next handler must not run")
+	assert.Equal(t, int32(0), validator.tunnelCalls.Load(), "ValidateTunnelPeer must not be invoked without the inbound marker")
 }
 
 // TestForwardWithTunnelPeer_RPCErrorFallsThrough validates that an RPC
@@ -201,8 +208,13 @@ func TestForwardWithTunnelPeer_CacheReusesPositiveResponse(t *testing.T) {
 	}
 	mw := newTunnelMiddleware(t, validator)
 
+	lookup := TunnelLookupFunc(func(_ netip.Addr) (PeerIdentity, bool) {
+		return PeerIdentity{}, true
+	})
+
 	for i := 0; i < 4; i++ {
 		w, r := newTunnelRequest("100.64.0.10:55555")
+		r = r.WithContext(WithTunnelLookup(r.Context(), lookup))
 		next := http.HandlerFunc(func(http.ResponseWriter, *http.Request) {})
 		config, _ := mw.getDomainConfig("svc.example")
 		handled := mw.forwardWithTunnelPeer(w, r, "svc.example", config, next)
@@ -226,11 +238,21 @@ func TestForwardWithTunnelPeer_RoutesAccountIDIntoCacheKey(t *testing.T) {
 	require.NoError(t, mw.AddDomain("svc-a.example", nil, "", 0, "acct-a", "svc-a", nil, false))
 	require.NoError(t, mw.AddDomain("svc-b.example", nil, "", 0, "acct-b", "svc-b", nil, false))
 
+	// The fast-path requires the inbound-listener marker on the context.
+	// The peerstore lookup itself is account-agnostic at this level
+	// (one TunnelLookupFunc per account is attached by inbound.go); a
+	// trivial "always hit" lookup is enough to exercise the cache-key
+	// branch this test covers.
+	lookup := TunnelLookupFunc(func(_ netip.Addr) (PeerIdentity, bool) {
+		return PeerIdentity{}, true
+	})
+
 	for _, host := range []string{"svc-a.example", "svc-b.example"} {
 		w := httptest.NewRecorder()
 		r := httptest.NewRequest(http.MethodGet, "https://"+host+"/", nil)
 		r.Host = host
 		r.RemoteAddr = "100.64.0.10:55555"
+		r = r.WithContext(WithTunnelLookup(r.Context(), lookup))
 		config, _ := mw.getDomainConfig(host)
 		handled := mw.forwardWithTunnelPeer(w, r, host, config, http.HandlerFunc(func(http.ResponseWriter, *http.Request) {}))
 		require.True(t, handled, "host %s should forward", host)
@@ -314,9 +336,17 @@ func TestPrivateService_ForwardsOnTunnelPeerSuccess(t *testing.T) {
 		w.WriteHeader(http.StatusOK)
 	}))
 
+	// Per-account inbound listener attaches WithTunnelLookup; without it
+	// forwardWithTunnelPeer refuses to take the fast-path. Mirror the
+	// real flow so this test exercises the post-gating success branch.
+	lookup := TunnelLookupFunc(func(_ netip.Addr) (PeerIdentity, bool) {
+		return PeerIdentity{}, true
+	})
+
 	req := httptest.NewRequest(http.MethodGet, "https://private.svc/", nil)
 	req.Host = "private.svc"
 	req.RemoteAddr = "100.64.0.10:55555"
+	req = req.WithContext(WithTunnelLookup(req.Context(), lookup))
 	w := httptest.NewRecorder()
 	handler.ServeHTTP(w, req)
 

proxy/internal/debug/handler.go

@@ -131,7 +131,7 @@ func (h *Handler) SetCertStatus(cs certStatus) {
 
 // SetInboundProvider wires per-account inbound listener observability.
 // Pass nil (or skip the call) to keep the inbound section out of debug
-// responses on proxies that don't run --private-inbound.
+// responses on proxies that don't run --private.
 func (h *Handler) SetInboundProvider(p InboundProvider) {
 	h.inbound = p
 }

proxy/internal/proxy/reverseproxy.go

@@ -66,6 +66,22 @@ func (p *ReverseProxy) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 
+	// Loop guard for private services: a peer that hosts the target
+	// dialing its own service URL would round-trip its own traffic
+	// through the proxy and back over WG to itself. Refuse the request
+	// with 421 (Misdirected Request) so the caller sees an explicit
+	// error instead of silently doubling tunnel traffic.
+	if p.isSelfTargetLoop(r, result.target.URL) {
+		if cd := CapturedDataFromContext(r.Context()); cd != nil {
+			cd.SetOrigin(OriginNoRoute)
+		}
+		requestID := getRequestID(r)
+		web.ServeErrorPage(w, r, http.StatusMisdirectedRequest, "Loop Detected",
+			"This peer is the target of the requested service. Reach the backend directly instead of dialing the public service URL from the same machine.",
+			requestID, web.ErrorStatus{Proxy: true, Destination: false})
+		return
+	}
+
 	ctx := r.Context()
 	// Set the account ID in the context for the roundtripper to use.
 	ctx = roundtrip.WithAccountID(ctx, result.accountID)
@@ -107,6 +123,32 @@ func (p *ReverseProxy) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	rp.ServeHTTP(w, r.WithContext(ctx))
 }
 
+// isSelfTargetLoop reports whether an overlay-origin request is about to
+// be forwarded back to the very peer that initiated it. The detection
+// is intentionally narrow: it only fires when the request arrived on
+// the per-account inbound (overlay) listener (so we're confident the
+// source address is the caller's tunnel IP), and only when the resolved
+// target host matches that tunnel IP. Catching this here returns 421 to
+// the caller instead of letting the proxy round-trip its own traffic
+// over WG twice.
+func (p *ReverseProxy) isSelfTargetLoop(r *http.Request, target *url.URL) bool {
+	if target == nil {
+		return false
+	}
+	if !types.IsOverlayOrigin(r.Context()) {
+		return false
+	}
+	srcIP := extractHostIP(r.RemoteAddr)
+	if !srcIP.IsValid() {
+		return false
+	}
+	targetIP, err := netip.ParseAddr(target.Hostname())
+	if err != nil {
+		return false
+	}
+	return srcIP.Unmap() == targetIP.Unmap()
+}
+
 // rewriteFunc returns a Rewrite function for httputil.ReverseProxy that rewrites
 // inbound requests to target the backend service while setting security-relevant
 // forwarding headers and stripping proxy authentication credentials.

proxy/internal/proxy/reverseproxy_test.go

@@ -20,6 +20,7 @@ import (
 
 	"github.com/netbirdio/netbird/proxy/auth"
 	"github.com/netbirdio/netbird/proxy/internal/roundtrip"
+	"github.com/netbirdio/netbird/proxy/internal/types"
 	"github.com/netbirdio/netbird/proxy/web"
 )
 
@@ -1285,6 +1286,103 @@ func TestStampNetBirdIdentity_OmitsGroupsHeaderWhenAllInvalid(t *testing.T) {
 		"X-NetBird-Groups must not be set when every group label is rejected")
 }
 
+// nopOKTransport returns 200 for every request without dialing — used
+// by the self-target-loop tests so the non-loop cases don't pay a real
+// TCP-dial timeout.
+type nopOKTransport struct{}
+
+func (nopOKTransport) RoundTrip(*http.Request) (*http.Response, error) {
+	return &http.Response{StatusCode: http.StatusOK, Body: http.NoBody, Header: http.Header{}}, nil
+}
+
+// TestServeHTTP_SelfTargetLoopReturns421 covers the loop guard for
+// private services: when a peer dials a service whose only target is
+// the peer itself, the proxy must refuse with 421 (Misdirected
+// Request) rather than round-tripping the request back over WG to
+// the same peer.
+func TestServeHTTP_SelfTargetLoopReturns421(t *testing.T) {
+	rp := NewReverseProxy(nopOKTransport{}, "auto", nil, nil)
+	rp.AddMapping(Mapping{
+		ID:        "svc-1",
+		AccountID: "acct-1",
+		Host:      "private.svc",
+		Paths: map[string]*PathTarget{
+			"/": {
+				URL: &url.URL{Scheme: "http", Host: "100.64.0.5:8080"},
+			},
+		},
+	})
+
+	req := httptest.NewRequest(http.MethodGet, "http://private.svc/", nil)
+	req.Host = "private.svc"
+	req.RemoteAddr = "100.64.0.5:55555"
+	req = req.WithContext(types.WithOverlayOrigin(req.Context()))
+	rec := httptest.NewRecorder()
+
+	rp.ServeHTTP(rec, req)
+
+	assert.Equal(t, http.StatusMisdirectedRequest, rec.Code,
+		"a peer dialing a service whose target is itself must get 421")
+}
+
+// TestServeHTTP_SelfTargetLoop_NonOverlayRequestPassesThrough verifies
+// the guard is scoped to overlay-origin requests. A public-listener
+// request that happens to share a source IP with the target host must
+// not be misinterpreted as a loop — the gating relies on the inbound
+// marker being attached only by the per-account overlay listener.
+func TestServeHTTP_SelfTargetLoop_NonOverlayRequestPassesThrough(t *testing.T) {
+	rp := NewReverseProxy(nopOKTransport{}, "auto", nil, nil)
+	rp.AddMapping(Mapping{
+		ID:        "svc-1",
+		AccountID: "acct-1",
+		Host:      "public.svc",
+		Paths: map[string]*PathTarget{
+			"/": {
+				URL: &url.URL{Scheme: "http", Host: "100.64.0.5:8080"},
+			},
+		},
+	})
+
+	req := httptest.NewRequest(http.MethodGet, "http://public.svc/", nil)
+	req.Host = "public.svc"
+	req.RemoteAddr = "100.64.0.5:55555"
+	// No WithOverlayOrigin → the guard must not fire.
+	rec := httptest.NewRecorder()
+
+	rp.ServeHTTP(rec, req)
+
+	assert.NotEqual(t, http.StatusMisdirectedRequest, rec.Code,
+		"a non-overlay request with a colliding source IP must not be flagged as a loop")
+}
+
+// TestServeHTTP_SelfTargetLoop_OverlayDifferentIPPassesThrough confirms
+// that overlay-origin requests with a source IP that does *not* match
+// the target host are forwarded normally.
+func TestServeHTTP_SelfTargetLoop_OverlayDifferentIPPassesThrough(t *testing.T) {
+	rp := NewReverseProxy(nopOKTransport{}, "auto", nil, nil)
+	rp.AddMapping(Mapping{
+		ID:        "svc-1",
+		AccountID: "acct-1",
+		Host:      "private.svc",
+		Paths: map[string]*PathTarget{
+			"/": {
+				URL: &url.URL{Scheme: "http", Host: "100.64.0.5:8080"},
+			},
+		},
+	})
+
+	req := httptest.NewRequest(http.MethodGet, "http://private.svc/", nil)
+	req.Host = "private.svc"
+	req.RemoteAddr = "100.64.0.99:55555" // different from the target
+	req = req.WithContext(types.WithOverlayOrigin(req.Context()))
+	rec := httptest.NewRecorder()
+
+	rp.ServeHTTP(rec, req)
+
+	assert.NotEqual(t, http.StatusMisdirectedRequest, rec.Code,
+		"overlay request with a non-matching source IP must not be flagged as a loop")
+}
+
 // TestStampNetBirdIdentity_CapturedDataPresentButEmpty covers requests
 // that carry CapturedData with no identity fields populated (e.g. the
 // auth middleware ran but the request didn't authenticate). Both

proxy/internal/roundtrip/netbird.go

@@ -28,6 +28,10 @@ import (
 
 const deviceNamePrefix = "ingress-proxy-"
 
+const clientStopTimeout = 30 * time.Second
+
+const createProxyPeerTimeout = 30 * time.Second
+
 // backendKey identifies a backend by its host:port from the target URL.
 type backendKey string
 
@@ -152,6 +156,7 @@ type managementClient interface {
 // backed by underlying NetBird connections.
 // Clients are keyed by AccountID, allowing multiple services to share the same connection.
 type NetBird struct {
+	ctx          context.Context
 	proxyID      string
 	proxyAddr    string
 	clientCfg    ClientConfig
@@ -161,6 +166,7 @@ type NetBird struct {
 
 	clientsMux     sync.RWMutex
 	clients        map[types.AccountID]*clientEntry
+	lifecycleMu    sync.Map
 	initLogOnce    sync.Once
 	statusNotifier statusNotifier
 	// readyHandler runs after the embedded client for an account reports
@@ -176,6 +182,10 @@ type NetBird struct {
 	// (i.e. when a new client was actually created, not when an existing one
 	// was reused). The duration covers keygen + gRPC CreateProxyPeer + embed.New.
 	OnAddPeer func(d time.Duration, err error)
+
+	// startClient runs the post-create client startup. Nil uses runClientStartup;
+	// tests override it to avoid a real embed client.Start.
+	startClient func(accountID types.AccountID, client *embed.Client)
 }
 
 // ClientDebugInfo contains debug information about a client.
@@ -199,27 +209,20 @@ type skipTLSVerifyContextKey struct{}
 func (n *NetBird) AddPeer(ctx context.Context, accountID types.AccountID, key ServiceKey, authToken string, serviceID types.ServiceID) error {
 	si := serviceInfo{serviceID: serviceID}
 
-	n.clientsMux.Lock()
+	if n.registerExistingClient(accountID, key, si) {
+		return nil
+	}
 
-	entry, exists := n.clients[accountID]
-	if exists {
-		entry.services[key] = si
-		started := entry.started
-		n.clientsMux.Unlock()
-
-		n.logger.WithFields(log.Fields{
-			"account_id":  accountID,
-			"service_key": key,
-		}).Debug("registered service with existing client")
-
-		if started && n.statusNotifier != nil {
-			if err := n.statusNotifier.NotifyStatus(ctx, accountID, serviceID, true); err != nil {
-				n.logger.WithFields(log.Fields{
-					"account_id":  accountID,
-					"service_key": key,
-				}).WithError(err).Warn("failed to notify status for existing client")
-			}
+	lifecycle := n.accountLifecycle(accountID)
+	lifecycle.Lock()
+	transferred := false
+	defer func() {
+		if !transferred {
+			lifecycle.Unlock()
 		}
+	}()
+
+	if n.registerExistingClient(accountID, key, si) {
 		return nil
 	}
 
@@ -229,10 +232,10 @@ func (n *NetBird) AddPeer(ctx context.Context, accountID types.AccountID, key Se
 		n.OnAddPeer(time.Since(createStart), err)
 	}
 	if err != nil {
-		n.clientsMux.Unlock()
 		return err
 	}
 
+	n.clientsMux.Lock()
 	n.clients[accountID] = entry
 	n.clientsMux.Unlock()
 
@@ -241,15 +244,64 @@ func (n *NetBird) AddPeer(ctx context.Context, accountID types.AccountID, key Se
 		"service_key": key,
 	}).Info("created new client for account")
 
-	// Attempt to start the client in the background; if this fails we will
-	// retry on the first request via RoundTrip.
-	go n.runClientStartup(ctx, accountID, entry.client)
+	transferred = true
+	go func() {
+		defer lifecycle.Unlock()
+		n.startClientStartup(accountID, entry.client)
+	}()
 
 	return nil
 }
 
+func (n *NetBird) startClientStartup(accountID types.AccountID, client *embed.Client) {
+	if n.startClient != nil {
+		n.startClient(accountID, client)
+		return
+	}
+	n.runClientStartup(accountID, client)
+}
+
+// registerExistingClient registers the service against an already-present
+// client for the account and returns true when it did. It notifies management
+// of the new service when the client is already started.
+func (n *NetBird) registerExistingClient(accountID types.AccountID, key ServiceKey, si serviceInfo) bool {
+	n.clientsMux.Lock()
+	entry, exists := n.clients[accountID]
+	if !exists {
+		n.clientsMux.Unlock()
+		return false
+	}
+	entry.services[key] = si
+	started := entry.started
+	n.clientsMux.Unlock()
+
+	n.logger.WithFields(log.Fields{
+		"account_id":  accountID,
+		"service_key": key,
+	}).Debug("registered service with existing client")
+
+	if started && n.statusNotifier != nil {
+		if err := n.statusNotifier.NotifyStatus(context.Background(), accountID, si.serviceID, true); err != nil {
+			n.logger.WithFields(log.Fields{
+				"account_id":  accountID,
+				"service_key": key,
+			}).WithError(err).Warn("failed to notify status for existing client")
+		}
+	}
+	return true
+}
+
+// accountLifecycle returns the per-account lifecycle mutex, serialising client
+// creation against teardown so a slow client.Stop cannot race a new
+// client.Start for the same account, without blocking clientsMux.
+func (n *NetBird) accountLifecycle(accountID types.AccountID) *sync.Mutex {
+	mu, _ := n.lifecycleMu.LoadOrStore(accountID, &sync.Mutex{})
+	return mu.(*sync.Mutex)
+}
+
 // createClientEntry generates a WireGuard keypair, authenticates with management,
-// and creates an embedded NetBird client. Must be called with clientsMux held.
+// and creates an embedded NetBird client. Must be called with the account's
+// lifecycle mutex held.
 func (n *NetBird) createClientEntry(ctx context.Context, accountID types.AccountID, key ServiceKey, authToken string, si serviceInfo) (*clientEntry, error) {
 	serviceID := si.serviceID
 	n.logger.WithFields(log.Fields{
@@ -269,7 +321,9 @@ func (n *NetBird) createClientEntry(ctx context.Context, accountID types.Account
 		"public_key": publicKey.String(),
 	}).Debug("authenticating new proxy peer with management")
 
-	resp, err := n.mgmtClient.CreateProxyPeer(ctx, &proto.CreateProxyPeerRequest{
+	createCtx, cancel := context.WithTimeout(ctx, createProxyPeerTimeout)
+	defer cancel()
+	resp, err := n.mgmtClient.CreateProxyPeer(createCtx, &proto.CreateProxyPeerRequest{
 		ServiceId:          string(serviceID),
 		AccountId:          string(accountID),
 		Token:              authToken,
@@ -307,7 +361,7 @@ func (n *NetBird) createClientEntry(ctx context.Context, accountID types.Account
 		ManagementURL: n.clientCfg.MgmtAddr,
 		PrivateKey:    privateKey.String(),
 		LogLevel:      log.WarnLevel.String(),
-		BlockInbound:   n.clientCfg.BlockInbound,
+		BlockInbound:  n.clientCfg.BlockInbound,
 		// The embedded proxy peer must never be a stepping stone into
 		// the proxy host's LAN: it only exists to reach NetBird mesh
 		// targets or, when direct_upstream is set, the host network
@@ -355,8 +409,14 @@ func (n *NetBird) createClientEntry(ctx context.Context, accountID types.Account
 	}, nil
 }
 
-// runClientStartup starts the client and notifies registered services on success.
-func (n *NetBird) runClientStartup(ctx context.Context, accountID types.AccountID, client *embed.Client) {
+// runClientStartup starts the client and notifies registered services on
+// success. This function runs in a goroutine launched from AddPeer, so it
+// must never inherit the caller's request-scoped context — a canceled
+// request must not abort the inbound listener bring-up or the management
+// status notification. The embedded client.Start gets its own bounded
+// startCtx; once Start succeeds, notifyClientReady takes over with a
+// fresh context.Background() (see that function for the contract).
+func (n *NetBird) runClientStartup(accountID types.AccountID, client *embed.Client) {
 	startCtx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
 	defer cancel()
 
@@ -369,7 +429,17 @@ func (n *NetBird) runClientStartup(ctx context.Context, accountID types.AccountI
 		return
 	}
 
-	// Mark client as started and collect services to notify outside the lock.
+	n.notifyClientReady(accountID, client)
+}
+
+// notifyClientReady marks the account's client as started, fires the
+// readyHandler hook, and notifies management of the new tunnel
+// connection for every registered service. It is split out of
+// runClientStartup so a regression test can drive the post-Start tail
+// without needing a live embedded client. The contract that the
+// hooks/notifier see context.Background() — never the AddPeer caller's
+// ctx — lives here.
+func (n *NetBird) notifyClientReady(accountID types.AccountID, client *embed.Client) {
 	n.clientsMux.Lock()
 	entry, exists := n.clients[accountID]
 	if exists {
@@ -385,7 +455,7 @@ func (n *NetBird) runClientStartup(ctx context.Context, accountID types.AccountI
 	n.clientsMux.Unlock()
 
 	if readyHandler != nil {
-		state := readyHandler(ctx, accountID, client)
+		state := readyHandler(n.ctx, accountID, client)
 		n.clientsMux.Lock()
 		if e, ok := n.clients[accountID]; ok {
 			e.inbound = state
@@ -404,7 +474,7 @@ func (n *NetBird) runClientStartup(ctx context.Context, accountID types.AccountI
 		return
 	}
 	for _, sn := range toNotify {
-		if err := n.statusNotifier.NotifyStatus(ctx, accountID, sn.serviceID, true); err != nil {
+		if err := n.statusNotifier.NotifyStatus(n.ctx, accountID, sn.serviceID, true); err != nil {
 			n.logger.WithFields(log.Fields{
 				"account_id":  accountID,
 				"service_key": sn.key,
@@ -421,6 +491,15 @@ func (n *NetBird) runClientStartup(ctx context.Context, accountID types.AccountI
 // RemovePeer unregisters a service from an account. The client is only stopped
 // when no services are using it anymore.
 func (n *NetBird) RemovePeer(ctx context.Context, accountID types.AccountID, key ServiceKey) error {
+	lifecycle := n.accountLifecycle(accountID)
+	lifecycle.Lock()
+	transferred := false
+	defer func() {
+		if !transferred {
+			lifecycle.Unlock()
+		}
+	}()
+
 	n.clientsMux.Lock()
 
 	entry, exists := n.clients[accountID]
@@ -443,17 +522,8 @@ func (n *NetBird) RemovePeer(ctx context.Context, accountID types.AccountID, key
 	delete(entry.services, key)
 
 	stopClient := len(entry.services) == 0
-	var client *embed.Client
-	var transport, insecureTransport *http.Transport
-	var inbound any
-	var stopHandler func(types.AccountID, any)
 	if stopClient {
 		n.logger.WithField("account_id", accountID).Info("stopping client, no more services")
-		client = entry.client
-		transport = entry.transport
-		insecureTransport = entry.insecureTransport
-		inbound = entry.inbound
-		stopHandler = n.stopHandler
 		delete(n.clients, accountID)
 	} else {
 		n.logger.WithFields(log.Fields{
@@ -467,19 +537,40 @@ func (n *NetBird) RemovePeer(ctx context.Context, accountID types.AccountID, key
 	n.notifyDisconnect(ctx, accountID, key, si.serviceID)
 
 	if stopClient {
-		if inbound != nil && stopHandler != nil {
-			stopHandler(accountID, inbound)
-		}
-		transport.CloseIdleConnections()
-		insecureTransport.CloseIdleConnections()
-		if err := client.Stop(ctx); err != nil {
-			n.logger.WithField("account_id", accountID).WithError(err).Warn("failed to stop netbird client")
-		}
+		transferred = true
+		go n.stopClientLocked(accountID, lifecycle, entry)
 	}
 
 	return nil
 }
 
+// stopClientLocked releases a client's resources off the caller's goroutine so a
+// slow client.Stop cannot wedge the mapping receive loop (which calls RemovePeer
+// synchronously). It unlocks lifecycle when done so a new client.Start for the
+// same account waits for this teardown.
+func (n *NetBird) stopClientLocked(accountID types.AccountID, lifecycle *sync.Mutex, entry *clientEntry) {
+	defer lifecycle.Unlock()
+
+	if entry.inbound != nil && n.stopHandler != nil {
+		n.stopHandler(accountID, entry.inbound)
+	}
+	if entry.transport != nil {
+		entry.transport.CloseIdleConnections()
+	}
+	if entry.insecureTransport != nil {
+		entry.insecureTransport.CloseIdleConnections()
+	}
+	if entry.client == nil {
+		return
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), clientStopTimeout)
+	defer cancel()
+	if err := entry.client.Stop(ctx); err != nil {
+		n.logger.WithField("account_id", accountID).WithError(err).Warn("failed to stop netbird client")
+	}
+}
+
 func (n *NetBird) notifyDisconnect(ctx context.Context, accountID types.AccountID, key ServiceKey, serviceID types.ServiceID) {
 	if n.statusNotifier == nil {
 		return
@@ -666,11 +757,12 @@ func (n *NetBird) ListClientsForStartup() map[types.AccountID]*embed.Client {
 // NewNetBird creates a new NetBird transport. Set clientCfg.WGPort to 0 for a random
 // OS-assigned port. A fixed port only works with single-account deployments;
 // multiple accounts will fail to bind the same port.
-func NewNetBird(proxyID, proxyAddr string, clientCfg ClientConfig, logger *log.Logger, notifier statusNotifier, mgmtClient managementClient) *NetBird {
+func NewNetBird(ctx context.Context, proxyID, proxyAddr string, clientCfg ClientConfig, logger *log.Logger, notifier statusNotifier, mgmtClient managementClient) *NetBird {
 	if logger == nil {
 		logger = log.StandardLogger()
 	}
 	return &NetBird{
+		ctx:            ctx,
 		proxyID:        proxyID,
 		proxyAddr:      proxyAddr,
 		clientCfg:      clientCfg,

proxy/internal/roundtrip/netbird_test.go

@@ -6,11 +6,13 @@ import (
 	"net/netip"
 	"sync"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"google.golang.org/grpc"
 
+	"github.com/netbirdio/netbird/client/embed"
 	"github.com/netbirdio/netbird/proxy/internal/types"
 	"github.com/netbirdio/netbird/shared/management/proto"
 )
@@ -21,6 +23,18 @@ func (m *mockMgmtClient) CreateProxyPeer(_ context.Context, _ *proto.CreateProxy
 	return &proto.CreateProxyPeerResponse{Success: true}, nil
 }
 
+// signalMgmtClient closes entered the first time CreateProxyPeer is called, so
+// tests can detect AddPeer reaching client creation.
+type signalMgmtClient struct {
+	entered chan struct{}
+	once    sync.Once
+}
+
+func (m *signalMgmtClient) CreateProxyPeer(_ context.Context, _ *proto.CreateProxyPeerRequest, _ ...grpc.CallOption) (*proto.CreateProxyPeerResponse, error) {
+	m.once.Do(func() { close(m.entered) })
+	return &proto.CreateProxyPeerResponse{Success: true}, nil
+}
+
 type mockStatusNotifier struct {
 	mu       sync.Mutex
 	statuses []statusCall
@@ -30,12 +44,15 @@ type statusCall struct {
 	accountID types.AccountID
 	serviceID types.ServiceID
 	connected bool
+	// ctx is captured so tests can assert the notifier received a
+	// fresh background context rather than an inherited request ctx.
+	ctx context.Context
 }
 
-func (m *mockStatusNotifier) NotifyStatus(_ context.Context, accountID types.AccountID, serviceID types.ServiceID, connected bool) error {
+func (m *mockStatusNotifier) NotifyStatus(ctx context.Context, accountID types.AccountID, serviceID types.ServiceID, connected bool) error {
 	m.mu.Lock()
 	defer m.mu.Unlock()
-	m.statuses = append(m.statuses, statusCall{accountID, serviceID, connected})
+	m.statuses = append(m.statuses, statusCall{accountID, serviceID, connected, ctx})
 	return nil
 }
 
@@ -48,11 +65,15 @@ func (m *mockStatusNotifier) calls() []statusCall {
 // mockNetBird creates a NetBird instance for testing without actually connecting.
 // It uses an invalid management URL to prevent real connections.
 func mockNetBird() *NetBird {
-	return NewNetBird("test-proxy", "invalid.test", ClientConfig{
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
 		MgmtAddr:     "http://invalid.test:9999",
 		WGPort:       0,
 		PreSharedKey: "",
 	}, nil, nil, &mockMgmtClient{})
+	// Skip the real embed client.Start, which would hang against the unreachable
+	// mgmt URL and (now that the lifecycle lock spans startup) serialise removes.
+	nb.startClient = func(types.AccountID, *embed.Client) {}
+	return nb
 }
 
 func TestNetBird_AddPeer_CreatesClientForNewAccount(t *testing.T) {
@@ -279,11 +300,12 @@ func TestNetBird_RoundTrip_RequiresExistingClient(t *testing.T) {
 
 func TestNetBird_AddPeer_ExistingStartedClient_NotifiesStatus(t *testing.T) {
 	notifier := &mockStatusNotifier{}
-	nb := NewNetBird("test-proxy", "invalid.test", ClientConfig{
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
 		MgmtAddr:     "http://invalid.test:9999",
 		WGPort:       0,
 		PreSharedKey: "",
 	}, nil, notifier, &mockMgmtClient{})
+	nb.startClient = func(types.AccountID, *embed.Client) {}
 	accountID := types.AccountID("account-1")
 
 	// Add first service — creates a new client entry.
@@ -295,8 +317,12 @@ func TestNetBird_AddPeer_ExistingStartedClient_NotifiesStatus(t *testing.T) {
 	nb.clients[accountID].started = true
 	nb.clientsMux.Unlock()
 
-	// Add second service — should notify immediately since client is already started.
-	err = nb.AddPeer(context.Background(), accountID, "domain2.test", "key-1", types.ServiceID("svc-2"))
+	// Add second service with an already-cancelled caller context —
+	// should notify immediately (client is started) AND the notification
+	// must not inherit the cancelled ctx.
+	cancelledCtx, cancel := context.WithCancel(context.Background())
+	cancel()
+	err = nb.AddPeer(cancelledCtx, accountID, "domain2.test", "key-1", types.ServiceID("svc-2"))
 	require.NoError(t, err)
 
 	calls := notifier.calls()
@@ -304,6 +330,9 @@ func TestNetBird_AddPeer_ExistingStartedClient_NotifiesStatus(t *testing.T) {
 	assert.Equal(t, accountID, calls[0].accountID)
 	assert.Equal(t, types.ServiceID("svc-2"), calls[0].serviceID)
 	assert.True(t, calls[0].connected)
+	require.NotNil(t, calls[0].ctx, "NotifyStatus must receive a context")
+	require.NoError(t, calls[0].ctx.Err(),
+		"already-started NotifyStatus must use a background ctx, not the cancelled caller ctx")
 }
 
 // TestNetBird_IdentityForIP_UnknownAccountReturnsFalse confirms that the
@@ -338,7 +367,7 @@ func TestClientEntry_IdentityForIP_InvalidIPReturnsFalse(t *testing.T) {
 
 func TestNetBird_RemovePeer_NotifiesDisconnection(t *testing.T) {
 	notifier := &mockStatusNotifier{}
-	nb := NewNetBird("test-proxy", "invalid.test", ClientConfig{
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
 		MgmtAddr:     "http://invalid.test:9999",
 		WGPort:       0,
 		PreSharedKey: "",
@@ -360,3 +389,164 @@ func TestNetBird_RemovePeer_NotifiesDisconnection(t *testing.T) {
 	assert.Equal(t, types.ServiceID("svc-1"), calls[0].serviceID)
 	assert.False(t, calls[0].connected)
 }
+
+// TestNetBird_RemovePeer_TeardownIsAsync proves the fix for the receive-loop
+// stall: RemovePeer must return promptly even when the client teardown blocks,
+// because teardown runs off the caller's goroutine. The receive loop calls
+// RemovePeer synchronously, so a blocking teardown inline would wedge it.
+func TestNetBird_RemovePeer_TeardownIsAsync(t *testing.T) {
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
+		MgmtAddr: "http://invalid.test:9999",
+	}, nil, &mockStatusNotifier{}, &mockMgmtClient{})
+
+	accountID := types.AccountID("acct-async-teardown")
+	key := DomainServiceKey("svc.example")
+
+	teardownEntered := make(chan struct{})
+	releaseTeardown := make(chan struct{})
+	nb.SetClientLifecycle(nil, func(types.AccountID, any) {
+		close(teardownEntered)
+		<-releaseTeardown
+	})
+
+	nb.clientsMux.Lock()
+	nb.clients[accountID] = &clientEntry{
+		services: map[ServiceKey]serviceInfo{key: {serviceID: types.ServiceID("svc-1")}},
+		started:  true,
+		inbound:  struct{}{},
+	}
+	nb.clientsMux.Unlock()
+
+	done := make(chan error, 1)
+	go func() { done <- nb.RemovePeer(context.Background(), accountID, key) }()
+
+	select {
+	case err := <-done:
+		require.NoError(t, err)
+	case <-time.After(2 * time.Second):
+		t.Fatal("RemovePeer did not return while teardown was blocked — teardown is not async")
+	}
+
+	select {
+	case <-teardownEntered:
+	case <-time.After(2 * time.Second):
+		t.Fatal("teardown never ran")
+	}
+
+	close(releaseTeardown)
+}
+
+// TestNetBird_AddPeer_WaitsForTeardown proves the lifecycle lock serialises a
+// new client bringup behind an in-flight teardown for the same account, so a
+// slow client.Stop can never race a new client.Start for that account.
+//
+// It targets the handoff race specifically: AddPeer is launched immediately
+// after RemovePeer returns, WITHOUT waiting for the teardown goroutine to start.
+// This only passes if RemovePeer acquires the lifecycle lock synchronously
+// (before returning) and hands it to the teardown goroutine — if the goroutine
+// acquired the lock itself, AddPeer could win the lock in this window and start
+// a replacement client while the old teardown is still pending.
+func TestNetBird_AddPeer_WaitsForTeardown(t *testing.T) {
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
+		MgmtAddr: "http://invalid.test:9999",
+	}, nil, &mockStatusNotifier{}, &mockMgmtClient{})
+	nb.startClient = func(types.AccountID, *embed.Client) {}
+
+	accountID := types.AccountID("acct-serialize")
+	key := DomainServiceKey("svc.example")
+
+	addEntered := make(chan struct{})
+	releaseTeardown := make(chan struct{})
+	nb.SetClientLifecycle(nil, func(types.AccountID, any) {
+		// Block teardown until released. If AddPeer ever reaches createClientEntry
+		// (signalled via the mgmt client below) while we hold the lock, the lock
+		// failed to serialise and the test fails before we release.
+		<-releaseTeardown
+	})
+
+	nb.clientsMux.Lock()
+	nb.clients[accountID] = &clientEntry{
+		services: map[ServiceKey]serviceInfo{key: {serviceID: types.ServiceID("svc-1")}},
+		started:  true,
+		inbound:  struct{}{},
+	}
+	nb.clientsMux.Unlock()
+
+	// createClientEntry calls CreateProxyPeer; closing addEntered there tells us
+	// AddPeer got past the lifecycle lock and into client creation.
+	nb.mgmtClient = &signalMgmtClient{entered: addEntered}
+
+	require.NoError(t, nb.RemovePeer(context.Background(), accountID, key))
+
+	// Launch AddPeer with NO synchronisation against the teardown goroutine.
+	addReturned := make(chan struct{})
+	go func() {
+		_ = nb.AddPeer(context.Background(), accountID, DomainServiceKey("svc2.example"), "key-2", types.ServiceID("svc-2"))
+		close(addReturned)
+	}()
+
+	select {
+	case <-addEntered:
+		t.Fatal("AddPeer entered client creation while teardown held the lifecycle lock — handoff race not closed")
+	case <-addReturned:
+		t.Fatal("AddPeer completed while teardown held the lifecycle lock — not serialised")
+	case <-time.After(300 * time.Millisecond):
+	}
+
+	close(releaseTeardown)
+	select {
+	case <-addReturned:
+	case <-time.After(2 * time.Second):
+		t.Fatal("AddPeer never completed after teardown released the lifecycle lock")
+	}
+}
+
+// TestNotifyClientReady_UsesBackgroundCtx pins the contract that the
+// post-Start hooks (readyHandler + statusNotifier.NotifyStatus) run on
+// a fresh context.Background() rather than inheriting the AddPeer
+// caller's request- or stream-scoped ctx. Without this, a cancelled
+// caller ctx could abort the inbound listener bring-up or cause the
+// management status notification to fail spuriously and leave the
+// account in a half-connected state.
+func TestNotifyClientReady_UsesBackgroundCtx(t *testing.T) {
+	notifier := &mockStatusNotifier{}
+	nb := NewNetBird(context.Background(), "test-proxy", "invalid.test", ClientConfig{
+		MgmtAddr: "http://invalid.test:9999",
+	}, nil, notifier, &mockMgmtClient{})
+
+	accountID := types.AccountID("acct-async")
+	// Pre-populate a client entry so notifyClientReady has something
+	// to mark started + something to enumerate for NotifyStatus.
+	nb.clientsMux.Lock()
+	nb.clients[accountID] = &clientEntry{
+		services: map[ServiceKey]serviceInfo{
+			DomainServiceKey("svc.example"): {serviceID: types.ServiceID("svc-1")},
+		},
+	}
+	nb.clientsMux.Unlock()
+
+	var capturedReadyCtx context.Context
+	nb.SetClientLifecycle(
+		func(ctx context.Context, _ types.AccountID, _ *embed.Client) any {
+			capturedReadyCtx = ctx
+			return nil
+		},
+		nil,
+	)
+
+	// Drive the post-Start path directly; a real client.Start would
+	// need a working management URL.
+	nb.notifyClientReady(accountID, nil)
+
+	require.NotNil(t, capturedReadyCtx, "readyHandler must have been invoked")
+	require.NoError(t, capturedReadyCtx.Err(),
+		"readyHandler must receive a background context, not an inherited cancelled one")
+	deadline, ok := capturedReadyCtx.Deadline()
+	assert.False(t, ok, "readyHandler ctx must have no deadline (background); got %v", deadline)
+
+	calls := notifier.calls()
+	require.Len(t, calls, 1, "NotifyStatus must be invoked once per registered service")
+	require.NotNil(t, calls[0].ctx, "NotifyStatus must receive a context")
+	require.NoError(t, calls[0].ctx.Err(),
+		"NotifyStatus must receive a background context, not an inherited cancelled one")
+}

proxy/internal/tcp/router_test.go

@@ -1781,11 +1781,14 @@ func TestRouter_PlainHTTP_RoutesToPlainChannel(t *testing.T) {
 		}
 	}()
 
+	tlsListener, ok := router.HTTPListener().(*chanListener)
+	require.True(t, ok, "router.HTTPListener() must be the test's chanListener; the test relies on observing its channel directly")
+
 	select {
 	case conn := <-acceptDone:
 		require.NotNil(t, conn)
 		_ = conn.Close()
-	case <-router.HTTPListener().(*chanListener).ch:
+	case <-tlsListener.ch:
 		t.Fatal("plain HTTP request leaked into TLS channel")
 	case <-time.After(3 * time.Second):
 		t.Fatal("plain HTTP connection never reached plain channel")

proxy/lifecycle.go

@@ -1,6 +1,7 @@
 package proxy
 
 import (
+	"context"
 	"net/netip"
 	"time"
 
@@ -20,14 +21,17 @@ import (
 type Config struct {
 	// ListenAddr is the TCP address the main listener binds. Required.
 	ListenAddr string
-	// ID identifies this proxy instance to management. Empty value lets
-	// New generate a timestamped default.
+	// ID identifies this proxy instance to management. Empty values are
+	// replaced with a timestamped default at Server.Start time (see
+	// initDefaults), not in New.
 	ID string
-	// Logger is the logrus logger used everywhere. Empty value falls back
-	// to log.StandardLogger().
+	// Logger is the logrus logger used everywhere. Empty values fall
+	// back to log.StandardLogger() at Server.Start time (see
+	// initDefaults), not in New.
 	Logger *log.Logger
 	// Version is the build version string reported to management. Empty
-	// becomes "dev".
+	// values are replaced with "dev" at Server.Start time (see
+	// initDefaults), not in New.
 	Version string
 	// ProxyURL is the public address operators use to reach this proxy.
 	ProxyURL string
@@ -110,6 +114,10 @@ type Config struct {
 	MaxDialTimeout time.Duration
 	// MaxSessionIdleTimeout caps the per-service session idle timeout.
 	MaxSessionIdleTimeout time.Duration
+	// MappingBatchWatchdog bounds how long a single mapping batch may spend
+	// being applied before the receive loop reconnects to resync. Zero falls
+	// back to the internal default.
+	MappingBatchWatchdog time.Duration
 
 	// GeoDataDir is the directory containing GeoLite2 MMDB files.
 	GeoDataDir string
@@ -125,8 +133,9 @@ type Config struct {
 // bound — call Start to bring the proxy up. Returning a fully-formed
 // Server keeps the standalone code path (which still constructs Server
 // directly) byte-for-byte equivalent.
-func New(cfg Config) *Server {
+func New(ctx context.Context, cfg Config) *Server {
 	return &Server{
+		ctx:                      ctx,
 		ListenAddr:               cfg.ListenAddr,
 		ID:                       cfg.ID,
 		Logger:                   cfg.Logger,
@@ -159,6 +168,7 @@ func New(cfg Config) *Server {
 		Private:                  cfg.Private,
 		MaxDialTimeout:           cfg.MaxDialTimeout,
 		MaxSessionIdleTimeout:    cfg.MaxSessionIdleTimeout,
+		MappingBatchWatchdog:     cfg.MappingBatchWatchdog,
 		GeoDataDir:               cfg.GeoDataDir,
 		CrowdSecAPIURL:           cfg.CrowdSecAPIURL,
 		CrowdSecAPIKey:           cfg.CrowdSecAPIKey,

proxy/mapping_stall_test.go

@@ -0,0 +1,282 @@
+package proxy
+
+import (
+	"context"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/stretchr/testify/assert"
+	"google.golang.org/grpc"
+	"google.golang.org/grpc/metadata"
+
+	"github.com/netbirdio/netbird/proxy/internal/roundtrip"
+	"github.com/netbirdio/netbird/proxy/internal/types"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// blockingMgmtClient implements roundtrip's managementClient interface.
+// CreateProxyPeer parks until release is closed, signalling entry on entered.
+// This reproduces the confirmed real-world stall: createClientEntry calls
+// CreateProxyPeer synchronously while holding clientsMux, and the proxy's
+// receive loop calls that path synchronously inside processMappings.
+type blockingMgmtClient struct {
+	entered chan struct{}
+	once    sync.Once
+}
+
+func (b *blockingMgmtClient) CreateProxyPeer(ctx context.Context, _ *proto.CreateProxyPeerRequest, _ ...grpc.CallOption) (*proto.CreateProxyPeerResponse, error) {
+	b.once.Do(func() { close(b.entered) })
+	// Park until the caller's context is cancelled. In production this ctx is
+	// the gRPC mapping-stream context with no per-call timeout, so a slow or
+	// unresponsive CreateProxyPeer parks the receive loop here indefinitely.
+	<-ctx.Done()
+	return nil, ctx.Err()
+}
+
+// gatedMappingStream is a mock GetMappingUpdate client stream that hands out a
+// pre-seeded list of messages, then records how many times Recv advanced. It
+// lets the test observe whether the single-threaded receive loop ever gets
+// past the first (blocking) batch to pull the second message.
+type gatedMappingStream struct {
+	grpc.ClientStream
+	messages []*proto.GetMappingUpdateResponse
+	idx      int32
+}
+
+func (g *gatedMappingStream) Recv() (*proto.GetMappingUpdateResponse, error) {
+	i := int(atomic.LoadInt32(&g.idx))
+	if i >= len(g.messages) {
+		// Block instead of returning EOF so the loop doesn't exit; we only
+		// care whether the loop ever reaches this second Recv at all.
+		select {}
+	}
+	msg := g.messages[i]
+	atomic.AddInt32(&g.idx, 1)
+	return msg, nil
+}
+
+func (g *gatedMappingStream) deliveredCount() int32 { return atomic.LoadInt32(&g.idx) }
+
+func (g *gatedMappingStream) Header() (metadata.MD, error) { return nil, nil } //nolint:nilnil
+func (g *gatedMappingStream) Trailer() metadata.MD         { return nil }
+func (g *gatedMappingStream) CloseSend() error             { return nil }
+func (g *gatedMappingStream) Context() context.Context     { return context.Background() }
+func (g *gatedMappingStream) SendMsg(any) error            { return nil }
+func (g *gatedMappingStream) RecvMsg(any) error            { return nil }
+
+// noopNotifier satisfies roundtrip's statusNotifier interface.
+type noopNotifier struct{}
+
+func (noopNotifier) NotifyStatus(context.Context, types.AccountID, types.ServiceID, bool) error {
+	return nil
+}
+
+// noopProxyClient is a proto.ProxyServiceClient that no-ops the one method the
+// teardown unwind reaches (SendStatusUpdate, via notifyError when the parked
+// AddPeer is cancelled). The embedded nil interface satisfies the rest at
+// compile time; none of those methods are called by this test.
+type noopProxyClient struct {
+	proto.ProxyServiceClient
+}
+
+func (noopProxyClient) SendStatusUpdate(context.Context, *proto.SendStatusUpdateRequest, ...grpc.CallOption) (*proto.SendStatusUpdateResponse, error) {
+	return &proto.SendStatusUpdateResponse{}, nil
+}
+
+// TestMappingStream_StallsWhenApplyBlocks proves the deadlock: the proxy's
+// mapping receive loop processes batches strictly serially, so when applying
+// one batch blocks (here: createClientEntry parked on a synchronous
+// CreateProxyPeer call, exactly as observed in production), the loop never
+// advances to Recv the next batch. Management can keep sending updates onto
+// the stream with no error and no channel overflow, yet the proxy applies
+// nothing further — it is stuck.
+func TestMappingStream_StallsWhenApplyBlocks(t *testing.T) {
+	logger := log.New()
+	logger.SetLevel(log.PanicLevel)
+
+	mgmt := &blockingMgmtClient{
+		entered: make(chan struct{}),
+	}
+
+	nb := roundtrip.NewNetBird(
+		context.Background(),
+		"proxy-test",
+		"proxy.example.com",
+		roundtrip.ClientConfig{},
+		logger,
+		noopNotifier{},
+		mgmt,
+	)
+
+	s := &Server{
+		Logger:       logger,
+		netbird:      nb,
+		mgmtClient:   noopProxyClient{},
+		routerReady:  closedChan(),
+		lastMappings: make(map[types.ServiceID]*proto.ProxyMapping),
+	}
+
+	// First batch: a CREATED mapping for a brand-new account. addMapping ->
+	// netbird.AddPeer -> createClientEntry -> CreateProxyPeer, which blocks.
+	// Empty Path keeps setupHTTPMapping a no-op (it returns early), so the
+	// ONLY blocking point is the synchronous CreateProxyPeer in AddPeer —
+	// no routers/auth need wiring. The second batch exists only to detect
+	// whether the loop ever advances past the blocked first batch.
+	stream := &gatedMappingStream{
+		messages: []*proto.GetMappingUpdateResponse{
+			{
+				Mapping: []*proto.ProxyMapping{
+					{
+						Type:      proto.ProxyMappingUpdateType_UPDATE_TYPE_CREATED,
+						Id:        "svc-1",
+						AccountId: "acct-1",
+						AuthToken: "token-1",
+					},
+				},
+			},
+			{
+				Mapping: []*proto.ProxyMapping{
+					{
+						Type:      proto.ProxyMappingUpdateType_UPDATE_TYPE_CREATED,
+						Id:        "svc-2",
+						AccountId: "acct-2",
+						AuthToken: "token-2",
+					},
+				},
+			},
+		},
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	// Unblock the parked apply on teardown via ctx (CreateProxyPeer returns
+	// ctx.Err()), so the wedged loop goroutine unwinds before embed.New —
+	// avoiding any dependency on collaborators this test deliberately leaves
+	// nil. The deadlock is fully proven before this fires.
+	t.Cleanup(cancel)
+
+	loopDone := make(chan struct{})
+	syncDone := false
+	go func() {
+		defer close(loopDone)
+		_ = s.handleMappingStream(ctx, stream, &syncDone, time.Time{})
+	}()
+
+	// The loop must reach the blocking apply for the first batch.
+	select {
+	case <-mgmt.entered:
+	case <-time.After(2 * time.Second):
+		t.Fatal("receive loop never reached CreateProxyPeer for the first batch")
+	}
+
+	// THE DEADLOCK: while the first batch is parked in CreateProxyPeer, the
+	// single-threaded loop cannot advance. The second batch is never pulled,
+	// even though it is already available on the stream. Give it ample time.
+	// deliveredCount is atomic; syncDone is intentionally not read here because
+	// the loop goroutine owns it (reading it from the test would race).
+	time.Sleep(500 * time.Millisecond)
+	assert.Equal(t, int32(1), stream.deliveredCount(),
+		"loop must NOT consume the second batch while the first is blocked in apply — proxy is stuck")
+
+	select {
+	case <-loopDone:
+		t.Fatal("receive loop returned while it should be wedged in apply")
+	default:
+		// Still wedged, as expected.
+	}
+}
+
+// TestMappingStream_StallsWhenRemoveBlocks proves the deadlock for the REMOVE
+// path observed in production: a mapping remove tears down the account's last
+// embedded client via netbird.RemovePeer -> client.Stop -> Engine.Stop, whose
+// jobExecutorWG.Wait() is unbounded. Because the receive loop is single-
+// threaded, a blocked remove wedges the loop: no further mapping updates of any
+// kind (create/modify/remove) are applied, while management keeps sending them
+// successfully (no send error, no channel-full). Matches the reported symptom:
+// the last log line is a remove that stops a client, then silence.
+func TestMappingStream_StallsWhenRemoveBlocks(t *testing.T) {
+	logger := log.New()
+	logger.SetLevel(log.PanicLevel)
+
+	enteredRemove := make(chan struct{})
+	blockRemove := make(chan struct{})
+	var once sync.Once
+
+	s := &Server{
+		Logger:       logger,
+		mgmtClient:   noopProxyClient{},
+		routerReady:  closedChan(),
+		lastMappings: make(map[types.ServiceID]*proto.ProxyMapping),
+		// Stand in for netbird.RemovePeer -> client.Stop hanging on
+		// Engine.Stop's unbounded jobExecutorWG.Wait(). Only the first remove
+		// blocks; later removes return immediately so the recovery assertion
+		// can observe the loop advancing.
+		removePeer: func(ctx context.Context, _ types.AccountID, _ roundtrip.ServiceKey) error {
+			first := false
+			once.Do(func() {
+				first = true
+				close(enteredRemove)
+			})
+			if !first {
+				return nil
+			}
+			select {
+			case <-blockRemove:
+			case <-ctx.Done():
+			}
+			return nil
+		},
+	}
+
+	// Batch 1 removes a service (blocks in teardown). Batch 2 is a later update
+	// that must never be applied while the remove is wedged.
+	stream := &gatedMappingStream{
+		messages: []*proto.GetMappingUpdateResponse{
+			{
+				Mapping: []*proto.ProxyMapping{
+					{Type: proto.ProxyMappingUpdateType_UPDATE_TYPE_REMOVED, Id: "svc-1", AccountId: "acct-1"},
+				},
+			},
+			{
+				Mapping: []*proto.ProxyMapping{
+					{Type: proto.ProxyMappingUpdateType_UPDATE_TYPE_REMOVED, Id: "svc-2", AccountId: "acct-1"},
+				},
+			},
+		},
+	}
+
+	loopDone := make(chan struct{})
+	syncDone := false
+	go func() {
+		defer close(loopDone)
+		_ = s.handleMappingStream(context.Background(), stream, &syncDone, time.Time{})
+	}()
+
+	select {
+	case <-enteredRemove:
+	case <-time.After(2 * time.Second):
+		t.Fatal("receive loop never reached the blocking remove for the first batch")
+	}
+
+	// THE DEADLOCK: the loop is parked in the blocked remove and cannot advance.
+	// syncDone is owned by the loop goroutine, so it is not read here.
+	time.Sleep(500 * time.Millisecond)
+	assert.Equal(t, int32(1), stream.deliveredCount(),
+		"loop must NOT consume the second batch while the first remove is blocked — proxy is stuck")
+
+	select {
+	case <-loopDone:
+		t.Fatal("receive loop returned while it should be wedged on the remove")
+	default:
+	}
+
+	// Unblock and confirm the wedge was solely the blocked remove: the loop
+	// then advances and consumes the next batch.
+	close(blockRemove)
+	assert.Eventually(t, func() bool {
+		return stream.deliveredCount() >= 2
+	}, 2*time.Second, 5*time.Millisecond,
+		"once the remove unblocks, the loop must advance and consume the next batch")
+}

proxy/process_mappings_bench_test.go

@@ -73,7 +73,7 @@ func benchServerWithLatency(b *testing.B, createPeerDelay, statusDelay time.Dura
 		statusUpdateDelay: statusDelay,
 	}
 
-	nb := roundtrip.NewNetBird("bench-proxy", "bench.test",
+	nb := roundtrip.NewNetBird(b.Context(), "bench-proxy", "bench.test",
 		roundtrip.ClientConfig{MgmtAddr: "http://bench.test:9999"},
 		logger, nil, mgmtClient)
 

proxy/server.go

@@ -75,6 +75,7 @@ type portRouter struct {
 }
 
 type Server struct {
+	ctx           context.Context
 	mgmtClient    proto.ProxyServiceClient
 	proxy         *proxy.ReverseProxy
 	netbird       *roundtrip.NetBird
@@ -117,6 +118,9 @@ type Server struct {
 	// The mapping worker waits on this before processing updates.
 	routerReady chan struct{}
 
+	// removePeer defaults to netbird.RemovePeer; overridable in tests.
+	removePeer func(ctx context.Context, accountID types.AccountID, key roundtrip.ServiceKey) error
+
 	// inbound, when non-nil, manages per-account inbound listeners. Set by
 	// initPrivateInbound only when Private is true so the standalone
 	// proxy keeps its zero-overhead default path.
@@ -226,6 +230,10 @@ type Server struct {
 	// Zero means no cap (the proxy honors whatever management sends).
 	// Set via NB_PROXY_MAX_SESSION_IDLE_TIMEOUT for shared deployments.
 	MaxSessionIdleTimeout time.Duration
+	// MappingBatchWatchdog bounds how long a single mapping batch may spend
+	// in processMappings before the receive loop reconnects to resync.
+	// Zero uses defaultMappingBatchWatchdog.
+	MappingBatchWatchdog time.Duration
 }
 
 // clampIdleTimeout returns d capped to MaxSessionIdleTimeout when configured.
@@ -281,7 +289,7 @@ func (s *Server) NotifyCertificateIssued(ctx context.Context, accountID types.Ac
 }
 
 // inboundListenerProto resolves the per-account inbound listener state for
-// the SendStatusUpdate payload. Returns nil when --private-inbound is off
+// the SendStatusUpdate payload. Returns nil when --private is off
 // or the account has no live listener so management treats the field as
 // absent.
 func (s *Server) inboundListenerProto(accountID types.AccountID) *proto.ProxyInboundListener {
@@ -528,10 +536,10 @@ func (s *Server) initManagementClient() error {
 }
 
 // initNetBirdClient builds the multi-tenant embedded NetBird client used
-// for outbound RoundTripping and (when --private-inbound is on) per-account
+// for outbound RoundTripping and (when --private is on) per-account
 // inbound listeners.
 func (s *Server) initNetBirdClient() {
-	s.netbird = roundtrip.NewNetBird(s.ID, s.ProxyURL, roundtrip.ClientConfig{
+	s.netbird = roundtrip.NewNetBird(s.ctx, s.ID, s.ProxyURL, roundtrip.ClientConfig{
 		MgmtAddr:     s.ManagementAddress,
 		WGPort:       s.WireguardPort,
 		PreSharedKey: s.PreSharedKey,
@@ -1171,24 +1179,30 @@ func (s *Server) newManagementMappingWorker(ctx context.Context, client proto.Pr
 			s.healthChecker.SetManagementConnected(false)
 		}
 
+		connected := false
+		onConnected := func() { connected = true }
+
 		var streamErr error
 		if syncSupported {
-			streamErr = s.trySyncMappings(ctx, client, &initialSyncDone)
+			streamErr = s.trySyncMappings(ctx, client, &initialSyncDone, onConnected)
 			if isSyncUnimplemented(streamErr) {
 				syncSupported = false
 				s.Logger.Info("management does not support SyncMappings, falling back to GetMappingUpdate")
-				streamErr = s.tryGetMappingUpdate(ctx, client, &initialSyncDone)
+				streamErr = s.tryGetMappingUpdate(ctx, client, &initialSyncDone, onConnected)
 			}
 		} else {
-			streamErr = s.tryGetMappingUpdate(ctx, client, &initialSyncDone)
+			streamErr = s.tryGetMappingUpdate(ctx, client, &initialSyncDone, onConnected)
 		}
 
 		if s.healthChecker != nil {
 			s.healthChecker.SetManagementConnected(false)
 		}
 
-		// Stream established — reset backoff so the next failure retries quickly.
-		bo.Reset()
+		// Reset backoff only when a stream actually connected, so immediate
+		// connect failures still back off instead of spinning.
+		if connected {
+			bo.Reset()
+		}
 
 		if streamErr == nil {
 			return fmt.Errorf("stream closed by server")
@@ -1220,7 +1234,7 @@ func (s *Server) proxyCapabilities() *proto.ProxyCapabilities {
 	}
 }
 
-func (s *Server) tryGetMappingUpdate(ctx context.Context, client proto.ProxyServiceClient, initialSyncDone *bool) error {
+func (s *Server) tryGetMappingUpdate(ctx context.Context, client proto.ProxyServiceClient, initialSyncDone *bool, onConnected func()) error {
 	connectTime := time.Now()
 	mappingClient, err := client.GetMappingUpdate(ctx, &proto.GetMappingUpdateRequest{
 		ProxyId:      s.ID,
@@ -1233,6 +1247,7 @@ func (s *Server) tryGetMappingUpdate(ctx context.Context, client proto.ProxyServ
 		return fmt.Errorf("create mapping stream: %w", err)
 	}
 
+	onConnected()
 	if s.healthChecker != nil {
 		s.healthChecker.SetManagementConnected(true)
 	}
@@ -1241,7 +1256,7 @@ func (s *Server) tryGetMappingUpdate(ctx context.Context, client proto.ProxyServ
 	return s.handleMappingStream(ctx, mappingClient, initialSyncDone, connectTime)
 }
 
-func (s *Server) trySyncMappings(ctx context.Context, client proto.ProxyServiceClient, initialSyncDone *bool) error {
+func (s *Server) trySyncMappings(ctx context.Context, client proto.ProxyServiceClient, initialSyncDone *bool, onConnected func()) error {
 	connectTime := time.Now()
 	stream, err := client.SyncMappings(ctx)
 	if err != nil {
@@ -1262,6 +1277,7 @@ func (s *Server) trySyncMappings(ctx context.Context, client proto.ProxyServiceC
 		return fmt.Errorf("send sync init: %w", err)
 	}
 
+	onConnected()
 	if s.healthChecker != nil {
 		s.healthChecker.SetManagementConnected(true)
 	}
@@ -1306,7 +1322,9 @@ func (s *Server) handleSyncMappingsStream(ctx context.Context, stream proto.Prox
 
 			batchStart := time.Now()
 			s.Logger.Debug("Received mapping update, starting processing")
-			s.processMappings(ctx, msg.GetMapping())
+			if err := s.processMappingsGuarded(ctx, msg.GetMapping()); err != nil {
+				return err
+			}
 			s.Logger.Debug("Processing mapping update completed")
 			tracker.recordBatch(ctx, s, msg.GetMapping(), msg.GetInitialSyncComplete(), batchStart)
 
@@ -1390,7 +1408,9 @@ func (s *Server) handleMappingStream(ctx context.Context, mappingClient proto.Pr
 
 			batchStart := time.Now()
 			s.Logger.Debug("Received mapping update, starting processing")
-			s.processMappings(ctx, msg.GetMapping())
+			if err := s.processMappingsGuarded(ctx, msg.GetMapping()); err != nil {
+				return err
+			}
 			s.Logger.Debug("Processing mapping update completed")
 			tracker.recordBatch(ctx, s, msg.GetMapping(), msg.GetInitialSyncComplete(), batchStart)
 		}
@@ -1455,6 +1475,44 @@ func redactMappingForLog(m *proto.ProxyMapping) *proto.ProxyMapping {
 	return c
 }
 
+const defaultMappingBatchWatchdog = 2 * time.Minute
+
+// mappingBatchWatchdog returns the configured batch watchdog or the default.
+func (s *Server) mappingBatchWatchdog() time.Duration {
+	if s.MappingBatchWatchdog > 0 {
+		return s.MappingBatchWatchdog
+	}
+	return defaultMappingBatchWatchdog
+}
+
+// processMappingsGuarded applies a batch under a watchdog, returning an error
+// if processing exceeds the watchdog so the caller reconnects and resyncs
+// instead of wedging silently.
+func (s *Server) processMappingsGuarded(ctx context.Context, mappings []*proto.ProxyMapping) error {
+	batchCtx, cancel := context.WithCancel(ctx)
+	defer cancel()
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		s.processMappings(batchCtx, mappings)
+	}()
+
+	watchdog := s.mappingBatchWatchdog()
+	timer := time.NewTimer(watchdog)
+	defer timer.Stop()
+
+	select {
+	case <-done:
+		return nil
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-timer.C:
+		s.Logger.Errorf("processing mapping batch exceeded %s, cancelling and reconnecting to resync", watchdog)
+		return fmt.Errorf("mapping batch processing stalled after %s", watchdog)
+	}
+}
+
 func (s *Server) processMappings(ctx context.Context, mappings []*proto.ProxyMapping) {
 	debug := s.Logger != nil && s.Logger.IsLevelEnabled(log.DebugLevel)
 	for _, mapping := range mappings {
@@ -1950,7 +2008,11 @@ func (s *Server) updateMapping(ctx context.Context, mapping *proto.ProxyMapping)
 func (s *Server) removeMapping(ctx context.Context, mapping *proto.ProxyMapping) {
 	accountID := types.AccountID(mapping.GetAccountId())
 	svcKey := s.serviceKeyForMapping(mapping)
-	if err := s.netbird.RemovePeer(ctx, accountID, svcKey); err != nil {
+	removePeer := s.removePeer
+	if removePeer == nil {
+		removePeer = s.netbird.RemovePeer
+	}
+	if err := removePeer(ctx, accountID, svcKey); err != nil {
 		s.Logger.WithFields(log.Fields{
 			"account_id": accountID,
 			"service_id": mapping.GetId(),

proxy/server_test.go

@@ -64,7 +64,7 @@ func quietLifecycleLogger() *log.Logger {
 }
 
 func TestStopBeforeStartIsNoOp(t *testing.T) {
-	srv := New(Config{Logger: quietLifecycleLogger()})
+	srv := New(t.Context(), Config{Logger: quietLifecycleLogger()})
 
 	ctx, cancel := context.WithTimeout(context.Background(), time.Second)
 	defer cancel()
@@ -77,7 +77,7 @@ func TestStopBeforeStartIsNoOp(t *testing.T) {
 }
 
 func TestStartFailsWithoutManagement(t *testing.T) {
-	srv := New(Config{
+	srv := New(t.Context(), Config{
 		Logger:            quietLifecycleLogger(),
 		ListenAddr:        "127.0.0.1:0",
 		ManagementAddress: "://broken-url",
@@ -137,7 +137,7 @@ func TestRecordRunErrPreservesFirstFailure(t *testing.T) {
 }
 
 func TestStopSkipsShutdownWhenNeverStarted(t *testing.T) {
-	srv := New(Config{Logger: quietLifecycleLogger()})
+	srv := New(t.Context(), Config{Logger: quietLifecycleLogger()})
 
 	ctx, cancel := context.WithCancel(context.Background())
 	cancel()

release_files/install.sh

@@ -417,15 +417,30 @@ if type uname >/dev/null 2>&1; then
               # Check the availability of a compatible package manager
               if check_use_bin_variable; then
                   PACKAGE_MANAGER="bin"
+              elif [ -e /run/ostree-booted ]; then
+                  if [ -x "$(command -v rpm-ostree)" ]; then
+                      PACKAGE_MANAGER="rpm-ostree"
+                      echo "The installation will be performed using rpm-ostree package manager"
+                  elif [ -x "$(command -v bootc)" ]; then
+                      echo "Detected bootc system without rpm-ostree." >&2
+                      echo "NetBird cannot be installed via package manager on this system." >&2
+                      echo "Options:" >&2
+                      echo "  1. Install via Distrobox (instructions in the installation docs)" >&2
+                      echo "  2. Rebuild your base image with rpm-ostree included" >&2
+                      echo "  3. Bake NetBird into your Containerfile" >&2
+                      exit 1
+                  else
+                      echo "Detected ostree-booted system without rpm-ostree or bootc." >&2
+                      echo "NetBird cannot be installed automatically on this atomic system." >&2
+                      echo "Please install NetBird by rebuilding your base image or use a supported package manager." >&2
+                      exit 1
+                  fi
               elif [ -x "$(command -v apt-get)" ]; then
                   PACKAGE_MANAGER="apt"
                   echo "The installation will be performed using apt package manager"
               elif [ -x "$(command -v dnf)" ]; then
                   PACKAGE_MANAGER="dnf"
                   echo "The installation will be performed using dnf package manager"
-              elif [ -x "$(command -v rpm-ostree)" ]; then
-                  PACKAGE_MANAGER="rpm-ostree"
-                  echo "The installation will be performed using rpm-ostree package manager"
               elif [ -x "$(command -v yum)" ]; then
                   PACKAGE_MANAGER="yum"
                   echo "The installation will be performed using yum package manager"

shared/context/keys.go

@@ -6,4 +6,5 @@ const (
 	RoleKey      = "role"
 	UserIDKey    = "userID"
 	PeerIDKey    = "peerID"
+	UserAgentKey = "userAgent"
 )

shared/management/client/rest/reverse_proxy_clusters.go

@@ -2,6 +2,7 @@ package rest
 
 import (
 	"context"
+	"errors"
 	"net/url"
 
 	"github.com/netbirdio/netbird/shared/management/http/api"
@@ -33,6 +34,12 @@ func (a *ReverseProxyClustersAPI) List(ctx context.Context) ([]api.ProxyCluster,
 // NetBird cannot be deleted via this endpoint; the server returns 404 / 400
 // for cluster addresses the account does not own.
 func (a *ReverseProxyClustersAPI) Delete(ctx context.Context, clusterAddress string) error {
+	// Guard against the empty input: url.PathEscape("") returns "" which
+	// would collapse the request URL onto the collection endpoint and
+	// silently delete nothing (or 405 depending on routing).
+	if clusterAddress == "" {
+		return errors.New("clusterAddress is required")
+	}
 	resp, err := a.c.NewRequest(ctx, "DELETE", "/api/reverse-proxies/clusters/"+url.PathEscape(clusterAddress), nil, nil)
 	if err != nil {
 		return err

shared/management/client/rest/reverse_proxy_clusters_test.go

@@ -88,3 +88,17 @@ func TestReverseProxyClusters_Delete_Err(t *testing.T) {
 		assert.Error(t, err)
 	})
 }
+
+// TestReverseProxyClusters_Delete_EmptyAddress guards against an empty
+// clusterAddress reaching the wire — that would collapse the URL onto
+// the collection endpoint instead of a specific cluster. The client
+// must short-circuit with a typed error before any request is issued.
+func TestReverseProxyClusters_Delete_EmptyAddress(t *testing.T) {
+	withMockClient(func(c *rest.Client, mux *http.ServeMux) {
+		mux.HandleFunc("/api/reverse-proxies/clusters/", func(http.ResponseWriter, *http.Request) {
+			t.Fatal("empty clusterAddress must be rejected client-side; no request should reach the server")
+		})
+		err := c.ReverseProxyClusters.Delete(context.Background(), "")
+		assert.Error(t, err, "empty clusterAddress must surface as an error")
+	})
+}

shared/management/client/rest/reverse_proxy_tokens.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"net/url"
 
 	"github.com/netbirdio/netbird/shared/management/http/api"
@@ -61,6 +62,12 @@ func (a *ReverseProxyTokensAPI) Create(ctx context.Context, request api.ProxyTok
 // credentials existed; the plain secret can no longer authenticate any
 // new proxy registration.
 func (a *ReverseProxyTokensAPI) Delete(ctx context.Context, tokenID string) error {
+	// Guard against the empty input: url.PathEscape("") returns "" which
+	// would collapse the request URL onto the collection endpoint and
+	// silently delete nothing (or 405 depending on routing).
+	if tokenID == "" {
+		return errors.New("tokenID is required")
+	}
 	resp, err := a.c.NewRequest(ctx, "DELETE", "/api/reverse-proxies/proxy-tokens/"+url.PathEscape(tokenID), nil, nil)
 	if err != nil {
 		return err