[client] Move the PCP implementation to the go-nat fork (#7282)

.github/workflows/no-new-replace.yml

@@ -0,0 +1,78 @@
+name: No New Replace Directives
+
+on:
+  pull_request:
+    paths:
+      - "go.mod"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.actor_id }}
+  cancel-in-progress: true
+
+jobs:
+  check-replace-directives:
+    name: check-replace-directives
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        with:
+          persist-credentials: false
+          fetch-depth: 0
+
+      - name: Install Go
+        uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 # v6.5.0
+        with:
+          go-version-file: go.mod
+
+      - name: Compare replace directives against the base branch
+        env:
+          BASE_SHA: ${{ github.event.pull_request.base.sha }}
+        run: |
+          set -euo pipefail
+
+          # A replace directive only applies when this module is the main
+          # module. Anything importing netbird as a library, the embedded
+          # clients among them, resolves the replaced path upstream instead and
+          # fails to build against whatever the replacement provides. Requiring
+          # a fork under its own module path avoids that; a replace does not.
+          #
+          # go.mod is parsed rather than diffed so that reordering, comments and
+          # single-line versus block syntax do not register as changes.
+          #
+          # Versions are part of the key because a replace can be scoped to one
+          # version of a module. Keyed on paths alone, retargeting such a
+          # directive at a different version would read as unchanged.
+          list_replaces() {
+            go mod edit -json "$1" \
+              | jq -r '
+                  def ref: .Path + (if (.Version // "") == "" then "" else " " + .Version end);
+                  (.Replace // [])[] | "\(.Old | ref) => \(.New | ref)"
+                ' \
+              | sort
+          }
+
+          git show "${BASE_SHA}:go.mod" > /tmp/base-go.mod
+          list_replaces /tmp/base-go.mod > /tmp/base-replaces
+          list_replaces go.mod > /tmp/head-replaces
+
+          added=$(comm -13 /tmp/base-replaces /tmp/head-replaces)
+          if [ -n "$added" ]; then
+            echo "::error::This PR adds a replace directive to go.mod:"
+            echo "$added" | sed 's/^/  /'
+            echo ""
+            echo "A replace directive applies only to the main module, so it does not"
+            echo "reach anything that imports netbird as a library. Require the module"
+            echo "under a path you control instead, as done for github.com/netbirdio/go-nat."
+            exit 1
+          fi
+
+          removed=$(comm -23 /tmp/base-replaces /tmp/head-replaces)
+          if [ -n "$removed" ]; then
+            echo "This PR removes replace directives:"
+            echo "$removed" | sed 's/^/  /'
+          fi
+          echo "No new replace directives."

client/embed/embed.go

@@ -91,6 +91,13 @@ type Options struct {
 	// when the embedded client must never act as a stepping stone into
 	// the host's local network (e.g. the proxy's overlay peer).
 	BlockLANAccess bool
+	// LazyConnectionEnabled is a tri-state local override for lazy connections,
+	// mirroring the NB_LAZY_CONN env var. Nil defers to the management feature
+	// flag; a set value overrides it in both directions. A short-lived client
+	// that reaches only a few known peers can set this to false, so its peers
+	// connect eagerly and the first request does not wait for the connection to
+	// be established.
+	LazyConnectionEnabled *bool
 	// WireguardPort is the port for the tunnel interface. Use 0 for a random port.
 	WireguardPort *int
 	// MTU is the MTU for the tunnel interface.
@@ -220,6 +227,15 @@ func New(opts Options) (*Client, error) {
 		config.PrivateKey = opts.PrivateKey
 	}
 
+	if opts.LazyConnectionEnabled != nil {
+		// Runtime-only override, read back through lazyconn.ParseState; a set value
+		// wins over the management feature flag in both directions.
+		config.LazyConnection = "off"
+		if *opts.LazyConnectionEnabled {
+			config.LazyConnection = "on"
+		}
+	}
+
 	if opts.Performance.PreallocatedBuffersPerPool != nil {
 		wgdevice.SetPreallocatedBuffersPerPool(*opts.Performance.PreallocatedBuffersPerPool)
 	}

client/internal/peer/worker_ice.go

@@ -389,6 +389,17 @@ func (w *WorkerICE) injectPortForwardedCandidate(srflxCandidate ice.Candidate) {
 		return
 	}
 
+	// A forwarded candidate only makes sense for an IPv4 mapping, which
+	// translates a port on the gateway's address. An IPv6 pinhole translates
+	// nothing: it unblocks the address ICE already gathers as a host candidate,
+	// so there is no second address to advertise. Injecting one here would also
+	// paste an IPv6 address onto whichever server-reflexive candidate arrived
+	// first, which is usually IPv4.
+	if mapping.ExternalIP != nil && mapping.ExternalIP.To4() == nil {
+		w.log.Debugf("skipping port-forwarded candidate: %s mapping is IPv6-only", mapping.NATType)
+		return
+	}
+
 	w.muxAgent.Lock()
 	if w.portForwardAttempted {
 		w.muxAgent.Unlock()

client/internal/portforward/manager.go

@@ -10,10 +10,8 @@ import (
 	"sync"
 	"time"
 
-	"github.com/libp2p/go-nat"
+	"github.com/netbirdio/go-nat"
 	log "github.com/sirupsen/logrus"
-
-	"github.com/netbirdio/netbird/client/internal/portforward/pcp"
 )
 
 const (
@@ -168,6 +166,11 @@ func (m *Manager) setup(ctx context.Context) (nat.NAT, *Mapping, error) {
 	if err != nil {
 		return nil, nil, fmt.Errorf("create port mapping: %w", err)
 	}
+
+	// Only meaningful once a mapping has been attempted: that is what opens the
+	// pinhole and records its outcome.
+	logIPv6Pinhole(gateway)
+
 	return gateway, mapping, nil
 }
 
@@ -265,7 +268,9 @@ func (m *Manager) checkHealthAndRecreate(ctx context.Context, gateway nat.NAT) b
 		return false
 	}
 
-	pcpNAT, ok := gateway.(*pcp.NAT)
+	// Assert on the interface, not on a concrete type: a dual-stack gateway is
+	// a wrapper around the IPv4 NAT, so a type assertion misses it.
+	checker, ok := gateway.(nat.HealthChecker)
 	if !ok {
 		return false
 	}
@@ -273,7 +278,7 @@ func (m *Manager) checkHealthAndRecreate(ctx context.Context, gateway nat.NAT) b
 	ctx, cancel := context.WithTimeout(ctx, 10*time.Second)
 	defer cancel()
 
-	epoch, serverRestarted, err := pcpNAT.CheckServerHealth(ctx)
+	epoch, serverRestarted, err := checker.CheckServerHealth(ctx)
 	if err != nil {
 		log.Debugf("PCP health check failed: %v", err)
 		return false
@@ -340,3 +345,18 @@ func (m *Manager) startTearDown(ctx context.Context) {
 func isPermanentLeaseRequired(err error) bool {
 	return err != nil && upnpErrPermanentLeaseOnly.MatchString(err.Error())
 }
+
+// logIPv6Pinhole reports the outcome of the IPv6 pinhole. Pinholes are best
+// effort and never fail a mapping on their own, so this is the only way to see
+// whether one was actually opened.
+func logIPv6Pinhole(gateway nat.NAT) {
+	reporter, ok := gateway.(nat.IPv6PinholeReporter)
+	if !ok {
+		return
+	}
+	if err := reporter.IPv6PinholeError(); err != nil {
+		log.Warnf("IPv6 pinhole: %v", err)
+		return
+	}
+	log.Infof("IPv6 pinhole open")
+}

client/internal/portforward/pcp/client.go

@@ -1,408 +0,0 @@
-package pcp
-
-import (
-	"context"
-	"crypto/rand"
-	"errors"
-	"fmt"
-	"net"
-	"net/netip"
-	"sync"
-	"time"
-
-	log "github.com/sirupsen/logrus"
-)
-
-const (
-	defaultTimeout     = 3 * time.Second
-	responseBufferSize = 128
-
-	// RFC 6887 Section 8.1.1 retry timing
-	initialRetryDelay = 3 * time.Second
-	maxRetryDelay     = 1024 * time.Second
-	maxRetries        = 4 // 3s + 6s + 12s + 24s = 45s total worst case
-)
-
-// Client is a PCP protocol client.
-// All methods are safe for concurrent use.
-type Client struct {
-	gateway netip.Addr
-	timeout time.Duration
-
-	mu sync.Mutex
-	// localIP caches the resolved local IP address.
-	localIP netip.Addr
-	// lastEpoch is the last observed server epoch value.
-	lastEpoch uint32
-	// epochTime tracks when lastEpoch was received for state loss detection.
-	epochTime time.Time
-	// externalIP caches the external IP from the last successful MAP response.
-	externalIP netip.Addr
-	// epochStateLost is set when epoch indicates server restart.
-	epochStateLost bool
-}
-
-// NewClient creates a new PCP client for the gateway at the given IP.
-func NewClient(gateway net.IP) *Client {
-	addr, ok := netip.AddrFromSlice(gateway)
-	if !ok {
-		log.Debugf("invalid gateway IP: %v", gateway)
-	}
-	return &Client{
-		gateway: addr.Unmap(),
-		timeout: defaultTimeout,
-	}
-}
-
-// NewClientWithTimeout creates a new PCP client with a custom timeout.
-func NewClientWithTimeout(gateway net.IP, timeout time.Duration) *Client {
-	addr, ok := netip.AddrFromSlice(gateway)
-	if !ok {
-		log.Debugf("invalid gateway IP: %v", gateway)
-	}
-	return &Client{
-		gateway: addr.Unmap(),
-		timeout: timeout,
-	}
-}
-
-// SetLocalIP sets the local IP address to use in PCP requests.
-func (c *Client) SetLocalIP(ip net.IP) {
-	addr, ok := netip.AddrFromSlice(ip)
-	if !ok {
-		log.Debugf("invalid local IP: %v", ip)
-	}
-	c.mu.Lock()
-	c.localIP = addr.Unmap()
-	c.mu.Unlock()
-}
-
-// Gateway returns the gateway IP address.
-func (c *Client) Gateway() net.IP {
-	return c.gateway.AsSlice()
-}
-
-// Announce sends a PCP ANNOUNCE request to discover PCP support.
-// Returns the server's epoch time on success.
-func (c *Client) Announce(ctx context.Context) (epoch uint32, err error) {
-	localIP, err := c.getLocalIP()
-	if err != nil {
-		return 0, fmt.Errorf("get local IP: %w", err)
-	}
-
-	req := buildAnnounceRequest(localIP)
-	resp, err := c.sendRequest(ctx, req)
-	if err != nil {
-		return 0, fmt.Errorf("send announce: %w", err)
-	}
-
-	parsed, err := parseResponse(resp)
-	if err != nil {
-		return 0, fmt.Errorf("parse announce response: %w", err)
-	}
-
-	if parsed.ResultCode != ResultSuccess {
-		return 0, fmt.Errorf("PCP ANNOUNCE failed: %s", ResultCodeString(parsed.ResultCode))
-	}
-
-	c.mu.Lock()
-	if c.updateEpochLocked(parsed.Epoch) {
-		log.Warnf("PCP server epoch indicates state loss - mappings may need refresh")
-	}
-	c.mu.Unlock()
-	return parsed.Epoch, nil
-}
-
-// AddPortMapping requests a port mapping from the PCP server.
-func (c *Client) AddPortMapping(ctx context.Context, protocol string, internalPort int, lifetime time.Duration) (*MapResponse, error) {
-	return c.addPortMappingWithHint(ctx, protocol, internalPort, internalPort, netip.Addr{}, lifetime)
-}
-
-// AddPortMappingWithHint requests a port mapping with suggested external port and IP.
-// Use lifetime <= 0 to delete a mapping.
-func (c *Client) AddPortMappingWithHint(ctx context.Context, protocol string, internalPort, suggestedExtPort int, suggestedExtIP net.IP, lifetime time.Duration) (*MapResponse, error) {
-	var extIP netip.Addr
-	if suggestedExtIP != nil {
-		var ok bool
-		extIP, ok = netip.AddrFromSlice(suggestedExtIP)
-		if !ok {
-			log.Debugf("invalid suggested external IP: %v", suggestedExtIP)
-		}
-		extIP = extIP.Unmap()
-	}
-	return c.addPortMappingWithHint(ctx, protocol, internalPort, suggestedExtPort, extIP, lifetime)
-}
-
-func (c *Client) addPortMappingWithHint(ctx context.Context, protocol string, internalPort, suggestedExtPort int, suggestedExtIP netip.Addr, lifetime time.Duration) (*MapResponse, error) {
-	localIP, err := c.getLocalIP()
-	if err != nil {
-		return nil, fmt.Errorf("get local IP: %w", err)
-	}
-
-	proto, err := protocolNumber(protocol)
-	if err != nil {
-		return nil, fmt.Errorf("parse protocol: %w", err)
-	}
-
-	var nonce [12]byte
-	if _, err := rand.Read(nonce[:]); err != nil {
-		return nil, fmt.Errorf("generate nonce: %w", err)
-	}
-
-	// Convert lifetime to seconds. Lifetime 0 means delete, so only apply
-	// default for positive durations that round to 0 seconds.
-	var lifetimeSec uint32
-	if lifetime > 0 {
-		lifetimeSec = uint32(lifetime.Seconds())
-		if lifetimeSec == 0 {
-			lifetimeSec = DefaultLifetime
-		}
-	}
-
-	req := buildMapRequest(localIP, nonce, proto, uint16(internalPort), uint16(suggestedExtPort), suggestedExtIP, lifetimeSec)
-
-	resp, err := c.sendRequest(ctx, req)
-	if err != nil {
-		return nil, fmt.Errorf("send map request: %w", err)
-	}
-
-	mapResp, err := parseMapResponse(resp)
-	if err != nil {
-		return nil, fmt.Errorf("parse map response: %w", err)
-	}
-
-	if mapResp.Nonce != nonce {
-		return nil, fmt.Errorf("nonce mismatch in response")
-	}
-
-	if mapResp.Protocol != proto {
-		return nil, fmt.Errorf("protocol mismatch: requested %d, got %d", proto, mapResp.Protocol)
-	}
-	if mapResp.InternalPort != uint16(internalPort) {
-		return nil, fmt.Errorf("internal port mismatch: requested %d, got %d", internalPort, mapResp.InternalPort)
-	}
-
-	if mapResp.ResultCode != ResultSuccess {
-		return nil, &Error{
-			Code:    mapResp.ResultCode,
-			Message: ResultCodeString(mapResp.ResultCode),
-		}
-	}
-
-	c.mu.Lock()
-	if c.updateEpochLocked(mapResp.Epoch) {
-		log.Warnf("PCP server epoch indicates state loss - mappings may need refresh")
-	}
-	c.cacheExternalIPLocked(mapResp.ExternalIP)
-	c.mu.Unlock()
-	return mapResp, nil
-}
-
-// DeletePortMapping removes a port mapping by requesting zero lifetime.
-func (c *Client) DeletePortMapping(ctx context.Context, protocol string, internalPort int) error {
-	if _, err := c.addPortMappingWithHint(ctx, protocol, internalPort, 0, netip.Addr{}, 0); err != nil {
-		var pcpErr *Error
-		if errors.As(err, &pcpErr) && pcpErr.Code == ResultNotAuthorized {
-			return nil
-		}
-		return fmt.Errorf("delete mapping: %w", err)
-	}
-	return nil
-}
-
-// GetExternalAddress returns the external IP address.
-// First checks for a cached value from previous MAP responses.
-// If not cached, creates a short-lived mapping to discover the external IP.
-func (c *Client) GetExternalAddress(ctx context.Context) (net.IP, error) {
-	c.mu.Lock()
-	if c.externalIP.IsValid() {
-		ip := c.externalIP.AsSlice()
-		c.mu.Unlock()
-		return ip, nil
-	}
-	c.mu.Unlock()
-
-	// Use an ephemeral port in the dynamic range (49152-65535).
-	// Port 0 is not valid with UDP/TCP protocols per RFC 6887.
-	ephemeralPort := 49152 + int(uint16(time.Now().UnixNano()))%(65535-49152)
-
-	// Use minimal lifetime (1 second) for discovery.
-	resp, err := c.AddPortMapping(ctx, "udp", ephemeralPort, time.Second)
-	if err != nil {
-		return nil, fmt.Errorf("create temporary mapping: %w", err)
-	}
-
-	if err := c.DeletePortMapping(ctx, "udp", ephemeralPort); err != nil {
-		log.Debugf("cleanup temporary PCP mapping: %v", err)
-	}
-
-	return resp.ExternalIP.AsSlice(), nil
-}
-
-// LastEpoch returns the last observed server epoch value.
-// A decrease in epoch indicates the server may have restarted and mappings may be lost.
-func (c *Client) LastEpoch() uint32 {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-	return c.lastEpoch
-}
-
-// EpochStateLost returns true if epoch state loss was detected and clears the flag.
-func (c *Client) EpochStateLost() bool {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-	lost := c.epochStateLost
-	c.epochStateLost = false
-	return lost
-}
-
-// updateEpoch updates the epoch tracking and detects potential state loss.
-// Returns true if state loss was detected (server likely restarted).
-// Caller must hold c.mu.
-func (c *Client) updateEpochLocked(newEpoch uint32) bool {
-	now := time.Now()
-	stateLost := false
-
-	// RFC 6887 Section 8.5: Detect invalid epoch indicating server state loss.
-	// client_delta = time since last response
-	// server_delta = epoch change since last response
-	// Invalid if: client_delta+2 < server_delta - server_delta/16
-	//         OR: server_delta+2 < client_delta - client_delta/16
-	// The +2 handles quantization, /16 (6.25%) handles clock drift.
-	if !c.epochTime.IsZero() && c.lastEpoch > 0 {
-		clientDelta := uint32(now.Sub(c.epochTime).Seconds())
-		serverDelta := newEpoch - c.lastEpoch
-
-		// Check for epoch going backwards or jumping unexpectedly.
-		// Subtraction is safe: serverDelta/16 is always <= serverDelta.
-		if clientDelta+2 < serverDelta-(serverDelta/16) ||
-			serverDelta+2 < clientDelta-(clientDelta/16) {
-			stateLost = true
-			c.epochStateLost = true
-		}
-	}
-
-	c.lastEpoch = newEpoch
-	c.epochTime = now
-	return stateLost
-}
-
-// cacheExternalIP stores the external IP from a successful MAP response.
-// Caller must hold c.mu.
-func (c *Client) cacheExternalIPLocked(ip netip.Addr) {
-	if ip.IsValid() && !ip.IsUnspecified() {
-		c.externalIP = ip
-	}
-}
-
-// sendRequest sends a PCP request with retries per RFC 6887 Section 8.1.1.
-func (c *Client) sendRequest(ctx context.Context, req []byte) ([]byte, error) {
-	addr := &net.UDPAddr{IP: c.gateway.AsSlice(), Port: Port}
-
-	var lastErr error
-	delay := initialRetryDelay
-
-	for range maxRetries {
-		resp, err := c.sendOnce(ctx, addr, req)
-		if err == nil {
-			return resp, nil
-		}
-		lastErr = err
-
-		if ctx.Err() != nil {
-			return nil, ctx.Err()
-		}
-
-		// RFC 6887 Section 8.1.1: RT = (1 + RAND) * MIN(2 * RTprev, MRT)
-		// RAND is random between -0.1 and +0.1
-		select {
-		case <-ctx.Done():
-			return nil, ctx.Err()
-		case <-time.After(retryDelayWithJitter(delay)):
-		}
-		delay = min(delay*2, maxRetryDelay)
-	}
-
-	return nil, fmt.Errorf("PCP request failed after %d retries: %w", maxRetries, lastErr)
-}
-
-// retryDelayWithJitter applies RFC 6887 jitter: multiply by (1 + RAND) where RAND is [-0.1, +0.1].
-func retryDelayWithJitter(d time.Duration) time.Duration {
-	var b [1]byte
-	_, _ = rand.Read(b[:])
-	// Convert byte to range [-0.1, +0.1]: (b/255 * 0.2) - 0.1
-	jitter := (float64(b[0])/255.0)*0.2 - 0.1
-	return time.Duration(float64(d) * (1 + jitter))
-}
-
-func (c *Client) sendOnce(ctx context.Context, addr *net.UDPAddr, req []byte) ([]byte, error) {
-	// Use ListenUDP instead of DialUDP to validate response source address per RFC 6887 §8.3.
-	conn, err := net.ListenUDP("udp", nil)
-	if err != nil {
-		return nil, fmt.Errorf("listen: %w", err)
-	}
-	defer func() {
-		if err := conn.Close(); err != nil {
-			log.Debugf("close UDP connection: %v", err)
-		}
-	}()
-
-	timeout := c.timeout
-	if deadline, ok := ctx.Deadline(); ok {
-		if remaining := time.Until(deadline); remaining < timeout {
-			timeout = remaining
-		}
-	}
-
-	if err := conn.SetDeadline(time.Now().Add(timeout)); err != nil {
-		return nil, fmt.Errorf("set deadline: %w", err)
-	}
-
-	if _, err := conn.WriteToUDP(req, addr); err != nil {
-		return nil, fmt.Errorf("write: %w", err)
-	}
-
-	resp := make([]byte, responseBufferSize)
-	n, from, err := conn.ReadFromUDP(resp)
-	if err != nil {
-		return nil, fmt.Errorf("read: %w", err)
-	}
-
-	// RFC 6887 §8.3: Validate response came from expected PCP server.
-	if !from.IP.Equal(addr.IP) {
-		return nil, fmt.Errorf("response from unexpected source %s (expected %s)", from.IP, addr.IP)
-	}
-
-	return resp[:n], nil
-}
-
-func (c *Client) getLocalIP() (netip.Addr, error) {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-
-	if !c.localIP.IsValid() {
-		return netip.Addr{}, fmt.Errorf("local IP not set for gateway %s", c.gateway)
-	}
-	return c.localIP, nil
-}
-
-func protocolNumber(protocol string) (uint8, error) {
-	switch protocol {
-	case "udp", "UDP":
-		return ProtoUDP, nil
-	case "tcp", "TCP":
-		return ProtoTCP, nil
-	default:
-		return 0, fmt.Errorf("unsupported protocol: %s", protocol)
-	}
-}
-
-// Error represents a PCP error response.
-type Error struct {
-	Code    uint8
-	Message string
-}
-
-func (e *Error) Error() string {
-	return fmt.Sprintf("PCP error: %s (%d)", e.Message, e.Code)
-}

client/internal/portforward/pcp/client_test.go

@@ -1,187 +0,0 @@
-package pcp
-
-import (
-	"context"
-	"net"
-	"net/netip"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/require"
-)
-
-func TestAddrConversion(t *testing.T) {
-	tests := []struct {
-		name string
-		addr netip.Addr
-	}{
-		{"IPv4", netip.MustParseAddr("192.168.1.100")},
-		{"IPv4 loopback", netip.MustParseAddr("127.0.0.1")},
-		{"IPv6", netip.MustParseAddr("2001:db8::1")},
-		{"IPv6 loopback", netip.MustParseAddr("::1")},
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			b16 := addrTo16(tt.addr)
-
-			recovered := addrFrom16(b16)
-			assert.Equal(t, tt.addr, recovered, "address should round-trip")
-		})
-	}
-}
-
-func TestBuildAnnounceRequest(t *testing.T) {
-	clientIP := netip.MustParseAddr("192.168.1.100")
-	req := buildAnnounceRequest(clientIP)
-
-	require.Len(t, req, headerSize)
-	assert.Equal(t, byte(Version), req[0], "version")
-	assert.Equal(t, byte(OpAnnounce), req[1], "opcode")
-
-	// Check client IP is properly encoded as IPv4-mapped IPv6
-	assert.Equal(t, byte(0xff), req[18], "IPv4-mapped prefix byte 10")
-	assert.Equal(t, byte(0xff), req[19], "IPv4-mapped prefix byte 11")
-	assert.Equal(t, byte(192), req[20], "IP octet 1")
-	assert.Equal(t, byte(168), req[21], "IP octet 2")
-	assert.Equal(t, byte(1), req[22], "IP octet 3")
-	assert.Equal(t, byte(100), req[23], "IP octet 4")
-}
-
-func TestBuildMapRequest(t *testing.T) {
-	clientIP := netip.MustParseAddr("192.168.1.100")
-	nonce := [12]byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12}
-	req := buildMapRequest(clientIP, nonce, ProtoUDP, 51820, 51820, netip.Addr{}, 3600)
-
-	require.Len(t, req, mapRequestSize)
-	assert.Equal(t, byte(Version), req[0], "version")
-	assert.Equal(t, byte(OpMap), req[1], "opcode")
-
-	// Lifetime at bytes 4-7
-	assert.Equal(t, uint32(3600), (uint32(req[4])<<24)|(uint32(req[5])<<16)|(uint32(req[6])<<8)|uint32(req[7]), "lifetime")
-
-	// Nonce at bytes 24-35
-	assert.Equal(t, nonce[:], req[24:36], "nonce")
-
-	// Protocol at byte 36
-	assert.Equal(t, byte(ProtoUDP), req[36], "protocol")
-
-	// Internal port at bytes 40-41
-	assert.Equal(t, uint16(51820), (uint16(req[40])<<8)|uint16(req[41]), "internal port")
-
-	// External port at bytes 42-43
-	assert.Equal(t, uint16(51820), (uint16(req[42])<<8)|uint16(req[43]), "external port")
-}
-
-func TestParseResponse(t *testing.T) {
-	// Construct a valid ANNOUNCE response
-	resp := make([]byte, headerSize)
-	resp[0] = Version
-	resp[1] = OpAnnounce | OpReply
-	// Result code = 0 (success)
-	// Lifetime = 0
-	// Epoch = 12345
-	resp[8] = 0
-	resp[9] = 0
-	resp[10] = 0x30
-	resp[11] = 0x39
-
-	parsed, err := parseResponse(resp)
-	require.NoError(t, err)
-	assert.Equal(t, uint8(Version), parsed.Version)
-	assert.Equal(t, uint8(OpAnnounce|OpReply), parsed.Opcode)
-	assert.Equal(t, uint8(ResultSuccess), parsed.ResultCode)
-	assert.Equal(t, uint32(12345), parsed.Epoch)
-}
-
-func TestParseResponseErrors(t *testing.T) {
-	t.Run("too short", func(t *testing.T) {
-		_, err := parseResponse([]byte{1, 2, 3})
-		assert.Error(t, err)
-	})
-
-	t.Run("wrong version", func(t *testing.T) {
-		resp := make([]byte, headerSize)
-		resp[0] = 1 // Wrong version
-		resp[1] = OpReply
-		_, err := parseResponse(resp)
-		assert.Error(t, err)
-	})
-
-	t.Run("missing reply bit", func(t *testing.T) {
-		resp := make([]byte, headerSize)
-		resp[0] = Version
-		resp[1] = OpAnnounce // Missing OpReply bit
-		_, err := parseResponse(resp)
-		assert.Error(t, err)
-	})
-}
-
-func TestResultCodeString(t *testing.T) {
-	assert.Equal(t, "SUCCESS", ResultCodeString(ResultSuccess))
-	assert.Equal(t, "NOT_AUTHORIZED", ResultCodeString(ResultNotAuthorized))
-	assert.Equal(t, "ADDRESS_MISMATCH", ResultCodeString(ResultAddressMismatch))
-	assert.Contains(t, ResultCodeString(255), "UNKNOWN")
-}
-
-func TestProtocolNumber(t *testing.T) {
-	proto, err := protocolNumber("udp")
-	require.NoError(t, err)
-	assert.Equal(t, uint8(ProtoUDP), proto)
-
-	proto, err = protocolNumber("tcp")
-	require.NoError(t, err)
-	assert.Equal(t, uint8(ProtoTCP), proto)
-
-	proto, err = protocolNumber("UDP")
-	require.NoError(t, err)
-	assert.Equal(t, uint8(ProtoUDP), proto)
-
-	_, err = protocolNumber("icmp")
-	assert.Error(t, err)
-}
-
-func TestClientCreation(t *testing.T) {
-	gateway := netip.MustParseAddr("192.168.1.1").AsSlice()
-
-	client := NewClient(gateway)
-	assert.Equal(t, net.IP(gateway), client.Gateway())
-	assert.Equal(t, defaultTimeout, client.timeout)
-
-	clientWithTimeout := NewClientWithTimeout(gateway, 5*time.Second)
-	assert.Equal(t, 5*time.Second, clientWithTimeout.timeout)
-}
-
-func TestNATType(t *testing.T) {
-	n := NewNAT(netip.MustParseAddr("192.168.1.1").AsSlice(), netip.MustParseAddr("192.168.1.100").AsSlice())
-	assert.Equal(t, "PCP", n.Type())
-}
-
-// Integration test - skipped unless PCP_TEST_GATEWAY env is set
-func TestClientIntegration(t *testing.T) {
-	t.Skip("Integration test - run manually with PCP_TEST_GATEWAY=<gateway-ip>")
-
-	gateway := netip.MustParseAddr("10.0.1.1").AsSlice()   // Change to your test gateway
-	localIP := netip.MustParseAddr("10.0.1.100").AsSlice() // Change to your local IP
-
-	client := NewClient(gateway)
-	client.SetLocalIP(localIP)
-	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
-	defer cancel()
-
-	// Test ANNOUNCE
-	epoch, err := client.Announce(ctx)
-	require.NoError(t, err)
-	t.Logf("Server epoch: %d", epoch)
-
-	// Test MAP
-	resp, err := client.AddPortMapping(ctx, "udp", 51820, 1*time.Hour)
-	require.NoError(t, err)
-	t.Logf("Mapping: internal=%d external=%d externalIP=%s",
-		resp.InternalPort, resp.ExternalPort, resp.ExternalIP)
-
-	// Cleanup
-	err = client.DeletePortMapping(ctx, "udp", 51820)
-	require.NoError(t, err)
-}

client/internal/portforward/pcp/nat.go

@@ -1,222 +0,0 @@
-package pcp
-
-import (
-	"context"
-	"fmt"
-	"net"
-	"net/netip"
-	"runtime"
-	"sync"
-	"time"
-
-	log "github.com/sirupsen/logrus"
-
-	"github.com/libp2p/go-nat"
-	"github.com/libp2p/go-netroute"
-)
-
-var _ nat.NAT = (*NAT)(nil)
-
-// NAT implements the go-nat NAT interface using PCP.
-// Supports dual-stack (IPv4 and IPv6) when available.
-// All methods are safe for concurrent use.
-//
-// TODO: IPv6 pinholes use the local IPv6 address. If the address changes
-// (e.g., due to SLAAC rotation or network change), the pinhole becomes stale
-// and needs to be recreated with the new address.
-type NAT struct {
-	client *Client
-
-	mu sync.RWMutex
-	// client6 is the IPv6 PCP client, nil if IPv6 is unavailable.
-	client6 *Client
-	// localIP6 caches the local IPv6 address used for PCP requests.
-	localIP6 netip.Addr
-}
-
-// NewNAT creates a new NAT instance backed by PCP.
-func NewNAT(gateway, localIP net.IP) *NAT {
-	client := NewClient(gateway)
-	client.SetLocalIP(localIP)
-	return &NAT{
-		client: client,
-	}
-}
-
-// Type returns "PCP" as the NAT type.
-func (n *NAT) Type() string {
-	return "PCP"
-}
-
-// GetDeviceAddress returns the gateway IP address.
-func (n *NAT) GetDeviceAddress() (net.IP, error) {
-	return n.client.Gateway(), nil
-}
-
-// GetExternalAddress returns the external IP address.
-func (n *NAT) GetExternalAddress() (net.IP, error) {
-	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
-	defer cancel()
-	return n.client.GetExternalAddress(ctx)
-}
-
-// GetInternalAddress returns the local IP address used to communicate with the gateway.
-func (n *NAT) GetInternalAddress() (net.IP, error) {
-	addr, err := n.client.getLocalIP()
-	if err != nil {
-		return nil, err
-	}
-	return addr.AsSlice(), nil
-}
-
-// AddPortMapping creates a port mapping on both IPv4 and IPv6 (if available).
-func (n *NAT) AddPortMapping(ctx context.Context, protocol string, internalPort int, _ string, timeout time.Duration) (int, error) {
-	resp, err := n.client.AddPortMapping(ctx, protocol, internalPort, timeout)
-	if err != nil {
-		return 0, fmt.Errorf("add mapping: %w", err)
-	}
-
-	n.mu.RLock()
-	client6 := n.client6
-	localIP6 := n.localIP6
-	n.mu.RUnlock()
-
-	if client6 == nil {
-		return int(resp.ExternalPort), nil
-	}
-
-	if _, err := client6.AddPortMapping(ctx, protocol, internalPort, timeout); err != nil {
-		log.Warnf("IPv6 PCP mapping failed (continuing with IPv4): %v", err)
-		return int(resp.ExternalPort), nil
-	}
-
-	log.Infof("created IPv6 PCP pinhole: %s:%d", localIP6, internalPort)
-	return int(resp.ExternalPort), nil
-}
-
-// DeletePortMapping removes a port mapping from both IPv4 and IPv6.
-func (n *NAT) DeletePortMapping(ctx context.Context, protocol string, internalPort int) error {
-	err := n.client.DeletePortMapping(ctx, protocol, internalPort)
-
-	n.mu.RLock()
-	client6 := n.client6
-	n.mu.RUnlock()
-
-	if client6 != nil {
-		if err6 := client6.DeletePortMapping(ctx, protocol, internalPort); err6 != nil {
-			log.Warnf("IPv6 PCP delete mapping failed: %v", err6)
-		}
-	}
-
-	if err != nil {
-		return fmt.Errorf("delete mapping: %w", err)
-	}
-	return nil
-}
-
-// CheckServerHealth sends an ANNOUNCE to verify the server is still responsive.
-// Returns the current epoch and whether the server may have restarted (epoch state loss detected).
-func (n *NAT) CheckServerHealth(ctx context.Context) (epoch uint32, serverRestarted bool, err error) {
-	epoch, err = n.client.Announce(ctx)
-	if err != nil {
-		return 0, false, fmt.Errorf("announce: %w", err)
-	}
-	return epoch, n.client.EpochStateLost(), nil
-}
-
-// DiscoverPCP attempts to discover a PCP-capable gateway.
-// Returns a NAT interface if PCP is supported, or an error otherwise.
-// Discovers both IPv4 and IPv6 gateways when available.
-func DiscoverPCP(ctx context.Context) (nat.NAT, error) {
-	gateway, localIP, err := getDefaultGateway()
-	if err != nil {
-		return nil, fmt.Errorf("get default gateway: %w", err)
-	}
-
-	client := NewClient(gateway)
-	client.SetLocalIP(localIP)
-	if _, err := client.Announce(ctx); err != nil {
-		return nil, fmt.Errorf("PCP announce: %w", err)
-	}
-
-	result := &NAT{client: client}
-	discoverIPv6(ctx, result)
-
-	return result, nil
-}
-
-func discoverIPv6(ctx context.Context, result *NAT) {
-	gateway6, localIP6, err := getDefaultGateway6()
-	if err != nil {
-		log.Debugf("IPv6 gateway discovery failed: %v", err)
-		return
-	}
-
-	client6 := NewClient(gateway6)
-	client6.SetLocalIP(localIP6)
-	if _, err := client6.Announce(ctx); err != nil {
-		log.Debugf("PCP IPv6 announce failed: %v", err)
-		return
-	}
-
-	addr, ok := netip.AddrFromSlice(localIP6)
-	if !ok {
-		log.Debugf("invalid IPv6 local IP: %v", localIP6)
-		return
-	}
-	result.mu.Lock()
-	result.client6 = client6
-	result.localIP6 = addr
-	result.mu.Unlock()
-	log.Debugf("PCP IPv6 gateway discovered: %s (local: %s)", gateway6, localIP6)
-}
-
-// getDefaultGateway returns the default IPv4 gateway and local IP using the system routing table.
-func getDefaultGateway() (gateway net.IP, localIP net.IP, err error) {
-	router, err := netroute.New()
-	if err != nil {
-		return nil, nil, err
-	}
-
-	dst := net.IPv4zero
-	if runtime.GOOS == "linux" || runtime.GOOS == "android" {
-		// go-netroute v0.4.0 rejects unspecified destinations client-side on Linux/Android.
-		// TODO: on android/ios, use platform APIs (ConnectivityManager.getLinkProperties /
-		// NWPathMonitor) when netlink-based lookup is restricted or unavailable.
-		dst = net.IPv4(0, 0, 0, 1)
-	}
-	_, gateway, localIP, err = router.Route(dst)
-	if err != nil {
-		return nil, nil, err
-	}
-
-	if gateway == nil {
-		return nil, nil, nat.ErrNoNATFound
-	}
-
-	return gateway, localIP, nil
-}
-
-// getDefaultGateway6 returns the default IPv6 gateway IP address using the system routing table.
-func getDefaultGateway6() (gateway net.IP, localIP net.IP, err error) {
-	router, err := netroute.New()
-	if err != nil {
-		return nil, nil, err
-	}
-
-	dst := net.IPv6zero
-	if runtime.GOOS == "linux" || runtime.GOOS == "android" {
-		// ::2
-		dst = net.IP{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2}
-	}
-	_, gateway, localIP, err = router.Route(dst)
-	if err != nil {
-		return nil, nil, err
-	}
-
-	if gateway == nil {
-		return nil, nil, nat.ErrNoNATFound
-	}
-
-	return gateway, localIP, nil
-}

client/internal/portforward/pcp/protocol.go

@@ -1,225 +0,0 @@
-// Package pcp implements the Port Control Protocol (RFC 6887).
-//
-// # Implemented Features
-//
-//   - ANNOUNCE opcode: Discovers PCP server support
-//   - MAP opcode: Creates/deletes port mappings (IPv4 NAT) and firewall pinholes (IPv6)
-//   - Dual-stack: Simultaneous IPv4 and IPv6 support via separate clients
-//   - Nonce validation: Prevents response spoofing
-//   - Epoch tracking: Detects server restarts per Section 8.5
-//   - RFC-compliant retry timing: 3s initial, exponential backoff to 1024s max (Section 8.1.1)
-//
-// # Not Implemented
-//
-//   - PEER opcode: For outbound peer connections (not needed for inbound NAT traversal)
-//   - THIRD_PARTY option: For managing mappings on behalf of other devices
-//   - PREFER_FAILURE option: Requires exact external port or fail (IPv4 NAT only, not needed for IPv6 pinholing)
-//   - FILTER option: To restrict remote peer addresses
-//
-// These optional features are omitted because the primary use case is simple
-// port forwarding for WireGuard, which only requires MAP with default behavior.
-package pcp
-
-import (
-	"encoding/binary"
-	"fmt"
-	"net/netip"
-)
-
-const (
-	// Version is the PCP protocol version (RFC 6887).
-	Version = 2
-
-	// Port is the standard PCP server port.
-	Port = 5351
-
-	// DefaultLifetime is the default requested mapping lifetime in seconds.
-	DefaultLifetime = 7200 // 2 hours
-
-	// Header sizes
-	headerSize     = 24
-	mapPayloadSize = 36
-	mapRequestSize = headerSize + mapPayloadSize // 60 bytes
-)
-
-// Opcodes
-const (
-	OpAnnounce = 0
-	OpMap      = 1
-	OpPeer     = 2
-	OpReply    = 0x80 // OR'd with opcode in responses
-)
-
-// Protocol numbers for MAP requests
-const (
-	ProtoUDP = 17
-	ProtoTCP = 6
-)
-
-// Result codes (RFC 6887 Section 7.4)
-const (
-	ResultSuccess              = 0
-	ResultUnsuppVersion        = 1
-	ResultNotAuthorized        = 2
-	ResultMalformedRequest     = 3
-	ResultUnsuppOpcode         = 4
-	ResultUnsuppOption         = 5
-	ResultMalformedOption      = 6
-	ResultNetworkFailure       = 7
-	ResultNoResources          = 8
-	ResultUnsuppProtocol       = 9
-	ResultUserExQuota          = 10
-	ResultCannotProvideExt     = 11
-	ResultAddressMismatch      = 12
-	ResultExcessiveRemotePeers = 13
-)
-
-// ResultCodeString returns a human-readable string for a result code.
-func ResultCodeString(code uint8) string {
-	switch code {
-	case ResultSuccess:
-		return "SUCCESS"
-	case ResultUnsuppVersion:
-		return "UNSUPP_VERSION"
-	case ResultNotAuthorized:
-		return "NOT_AUTHORIZED"
-	case ResultMalformedRequest:
-		return "MALFORMED_REQUEST"
-	case ResultUnsuppOpcode:
-		return "UNSUPP_OPCODE"
-	case ResultUnsuppOption:
-		return "UNSUPP_OPTION"
-	case ResultMalformedOption:
-		return "MALFORMED_OPTION"
-	case ResultNetworkFailure:
-		return "NETWORK_FAILURE"
-	case ResultNoResources:
-		return "NO_RESOURCES"
-	case ResultUnsuppProtocol:
-		return "UNSUPP_PROTOCOL"
-	case ResultUserExQuota:
-		return "USER_EX_QUOTA"
-	case ResultCannotProvideExt:
-		return "CANNOT_PROVIDE_EXTERNAL"
-	case ResultAddressMismatch:
-		return "ADDRESS_MISMATCH"
-	case ResultExcessiveRemotePeers:
-		return "EXCESSIVE_REMOTE_PEERS"
-	default:
-		return fmt.Sprintf("UNKNOWN(%d)", code)
-	}
-}
-
-// Response represents a parsed PCP response header.
-type Response struct {
-	Version    uint8
-	Opcode     uint8
-	ResultCode uint8
-	Lifetime   uint32
-	Epoch      uint32
-}
-
-// MapResponse contains the full response to a MAP request.
-type MapResponse struct {
-	Response
-	Nonce        [12]byte
-	Protocol     uint8
-	InternalPort uint16
-	ExternalPort uint16
-	ExternalIP   netip.Addr
-}
-
-// addrTo16 converts an address to its 16-byte IPv4-mapped IPv6 representation.
-func addrTo16(addr netip.Addr) [16]byte {
-	if addr.Is4() {
-		return netip.AddrFrom4(addr.As4()).As16()
-	}
-	return addr.As16()
-}
-
-// addrFrom16 extracts an address from a 16-byte representation, unmapping IPv4.
-func addrFrom16(b [16]byte) netip.Addr {
-	return netip.AddrFrom16(b).Unmap()
-}
-
-// buildAnnounceRequest creates a PCP ANNOUNCE request packet.
-func buildAnnounceRequest(clientIP netip.Addr) []byte {
-	req := make([]byte, headerSize)
-	req[0] = Version
-	req[1] = OpAnnounce
-	mapped := addrTo16(clientIP)
-	copy(req[8:24], mapped[:])
-	return req
-}
-
-// buildMapRequest creates a PCP MAP request packet.
-func buildMapRequest(clientIP netip.Addr, nonce [12]byte, protocol uint8, internalPort, suggestedExtPort uint16, suggestedExtIP netip.Addr, lifetime uint32) []byte {
-	req := make([]byte, mapRequestSize)
-
-	// Header
-	req[0] = Version
-	req[1] = OpMap
-	binary.BigEndian.PutUint32(req[4:8], lifetime)
-	mapped := addrTo16(clientIP)
-	copy(req[8:24], mapped[:])
-
-	// MAP payload
-	copy(req[24:36], nonce[:])
-	req[36] = protocol
-	binary.BigEndian.PutUint16(req[40:42], internalPort)
-	binary.BigEndian.PutUint16(req[42:44], suggestedExtPort)
-	if suggestedExtIP.IsValid() {
-		extMapped := addrTo16(suggestedExtIP)
-		copy(req[44:60], extMapped[:])
-	}
-
-	return req
-}
-
-// parseResponse parses the common PCP response header.
-func parseResponse(data []byte) (*Response, error) {
-	if len(data) < headerSize {
-		return nil, fmt.Errorf("response too short: %d bytes", len(data))
-	}
-
-	resp := &Response{
-		Version:    data[0],
-		Opcode:     data[1],
-		ResultCode: data[3], // Byte 2 is reserved, byte 3 is result code (RFC 6887 §7.2)
-		Lifetime:   binary.BigEndian.Uint32(data[4:8]),
-		Epoch:      binary.BigEndian.Uint32(data[8:12]),
-	}
-
-	if resp.Version != Version {
-		return nil, fmt.Errorf("unsupported PCP version: %d", resp.Version)
-	}
-
-	if resp.Opcode&OpReply == 0 {
-		return nil, fmt.Errorf("response missing reply bit: opcode=0x%02x", resp.Opcode)
-	}
-
-	return resp, nil
-}
-
-// parseMapResponse parses a complete MAP response.
-func parseMapResponse(data []byte) (*MapResponse, error) {
-	if len(data) < mapRequestSize {
-		return nil, fmt.Errorf("MAP response too short: %d bytes", len(data))
-	}
-
-	resp, err := parseResponse(data)
-	if err != nil {
-		return nil, fmt.Errorf("parse header: %w", err)
-	}
-
-	mapResp := &MapResponse{
-		Response:     *resp,
-		Protocol:     data[36],
-		InternalPort: binary.BigEndian.Uint16(data[40:42]),
-		ExternalPort: binary.BigEndian.Uint16(data[42:44]),
-		ExternalIP:   addrFrom16([16]byte(data[44:60])),
-	}
-	copy(mapResp.Nonce[:], data[24:36])
-
-	return mapResp, nil
-}

client/internal/portforward/pinhole_test.go

@@ -0,0 +1,116 @@
+//go:build !js
+
+package portforward
+
+import (
+	"context"
+	"errors"
+	"strings"
+	"testing"
+
+	"github.com/netbirdio/go-nat"
+	log "github.com/sirupsen/logrus"
+	"github.com/sirupsen/logrus/hooks/test"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// mockPinholeNAT is a gateway that also reports an IPv6 pinhole outcome, the
+// shape a dual-stack gateway has.
+type mockPinholeNAT struct {
+	*mockNAT
+	pinholeErr error
+}
+
+func (m *mockPinholeNAT) IPv6PinholeError() error {
+	return m.pinholeErr
+}
+
+func TestSetupLogsPinholeOutcome(t *testing.T) {
+	pinholeErr := errors.New("pcp ipv6: NOT_AUTHORIZED")
+
+	tests := []struct {
+		name       string
+		pinholeErr error
+		mappingErr error
+		wantLevel  log.Level
+		wantText   string
+	}{
+		{
+			name:      "an open pinhole is reported",
+			wantLevel: log.InfoLevel,
+			wantText:  "IPv6 pinhole open",
+		},
+		{
+			name: "a failed pinhole is reported without failing the mapping",
+			// The IPv4 mapping is what the caller asked for, so the pinhole
+			// failure surfaces only in the log.
+			pinholeErr: pinholeErr,
+			wantLevel:  log.WarnLevel,
+			wantText:   pinholeErr.Error(),
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			gateway := &mockPinholeNAT{mockNAT: newMockNAT(), pinholeErr: tt.pinholeErr}
+			hook := stubGatewayDiscovery(t, gateway)
+
+			m := NewManager()
+			m.wgPort = 51820
+
+			_, mapping, err := m.setup(context.Background())
+
+			require.NoError(t, err)
+			require.NotNil(t, mapping)
+
+			entry := findEntry(hook, tt.wantText)
+			require.NotNil(t, entry, "no log entry mentioning %q", tt.wantText)
+			assert.Equal(t, tt.wantLevel, entry.Level)
+		})
+	}
+
+	t.Run("a failed mapping reports no pinhole outcome", func(t *testing.T) {
+		// Nothing opened the pinhole, so whatever it currently reports says
+		// nothing about this attempt.
+		gateway := &mockPinholeNAT{mockNAT: newMockNAT()}
+		gateway.addMappingErr = errors.New("gateway refused")
+		hook := stubGatewayDiscovery(t, gateway)
+
+		m := NewManager()
+		m.wgPort = 51820
+
+		_, _, err := m.setup(context.Background())
+
+		require.Error(t, err)
+		assert.Nil(t, findEntry(hook, "IPv6 pinhole"))
+	})
+}
+
+// stubGatewayDiscovery makes discovery return gateway and captures log output.
+func stubGatewayDiscovery(t *testing.T, gateway nat.NAT) *test.Hook {
+	t.Helper()
+
+	orig := discoverGateway
+	discoverGateway = func(context.Context) (nat.NAT, error) { return gateway, nil }
+	t.Cleanup(func() { discoverGateway = orig })
+
+	hook := test.NewGlobal()
+	origLevel := log.GetLevel()
+	log.SetLevel(log.DebugLevel)
+	t.Cleanup(func() {
+		hook.Reset()
+		log.SetLevel(origLevel)
+	})
+
+	return hook
+}
+
+func findEntry(hook *test.Hook, substr string) *log.Entry {
+	for _, entry := range hook.AllEntries() {
+		if strings.Contains(entry.Message, substr) {
+			return entry
+		}
+	}
+	return nil
+}

client/internal/portforward/state.go

@@ -4,27 +4,94 @@ package portforward
 
 import (
 	"context"
+	"errors"
 	"fmt"
+	"time"
 
-	"github.com/libp2p/go-nat"
+	"github.com/netbirdio/go-nat"
+	"github.com/netbirdio/go-nat/pcp"
 	log "github.com/sirupsen/logrus"
-
-	"github.com/netbirdio/netbird/client/internal/portforward/pcp"
 )
 
 // discoverGateway is the function used for NAT gateway discovery.
 // It can be replaced in tests to avoid real network operations.
-// Tries PCP first, then falls back to NAT-PMP/UPnP.
 var discoverGateway = defaultDiscoverGateway
 
-func defaultDiscoverGateway(ctx context.Context) (nat.NAT, error) {
-	pcpGateway, err := pcp.DiscoverPCP(ctx)
-	if err == nil {
-		return pcpGateway, nil
-	}
-	log.Debugf("PCP discovery failed: %v, trying NAT-PMP/UPnP", err)
+// pinholeDiscoveryTimeout is the slice of the discovery budget held back for
+// the IPv6 pinhole probe.
+//
+// Sizing it is coarser than it looks: PCP retransmits on a 3s socket timeout
+// and a 3s first backoff, so a second attempt needs about 9s. Anything from
+// roughly 1s to 8s therefore buys exactly one attempt, and this only sets how
+// long that attempt waits. A PCP server sits on the local link and answers in
+// milliseconds, so 3s is margin rather than need, and the rest is left to
+// gateway discovery, whose multicast SSDP search alone takes 5s. A probe lost
+// to a dropped packet is retried by the next discovery round.
+//
+// It is a variable so tests can shorten it.
+var pinholeDiscoveryTimeout = 3 * time.Second
 
-	return nat.DiscoverGateway(ctx)
+// Discovery entry points, as variables so tests can drive the fallback without
+// touching the network.
+var (
+	discoverNATGateway = nat.DiscoverGateway
+
+	discoverPCPPinhole = func(ctx context.Context) (nat.NAT, error) {
+		pinhole, err := pcp.DiscoverPCP(ctx)
+		if err != nil {
+			return nil, err
+		}
+		return pinhole, nil
+	}
+)
+
+// defaultDiscoverGateway finds a gateway that can make the WireGuard port
+// reachable. DiscoverGateway prefers PCP for IPv4, races UPnP and NAT-PMP
+// behind it, and attaches an IPv6 pinhole independently of which IPv4 protocol
+// wins.
+//
+// It reports no gateway on a network offering only IPv6, having no IPv4 mapping
+// to attach a pinhole to. Such a network still needs one: there is no
+// translation to traverse, but the router drops inbound IPv6 until something
+// opens it. Fall back to PCP alone, which yields a gateway holding just the
+// pinhole.
+func defaultDiscoverGateway(ctx context.Context) (nat.NAT, error) {
+	gatewayCtx, cancel := reserveForPinhole(ctx)
+	defer cancel()
+
+	gateway, err := discoverNATGateway(gatewayCtx)
+	if err == nil {
+		return gateway, nil
+	}
+	if !errors.Is(err, nat.ErrNoNATFound) {
+		return nil, err
+	}
+
+	pinhole, pinholeErr := discoverPCPPinhole(ctx)
+	if pinholeErr != nil {
+		log.Debugf("no IPv6 pinhole after %v: %v", err, pinholeErr)
+		return nil, err
+	}
+
+	log.Infof("no IPv4 gateway, continuing with an IPv6 pinhole only")
+	return pinhole, nil
+}
+
+// reserveForPinhole shortens ctx so that a pinhole probe still has time to run
+// afterwards. Finding nothing takes gateway discovery everything it is given,
+// so on the unshortened context the probe would start already expired. A budget
+// too small to divide is left to gateway discovery, which is the likelier win.
+func reserveForPinhole(ctx context.Context) (context.Context, context.CancelFunc) {
+	deadline, ok := ctx.Deadline()
+	if !ok {
+		return context.WithCancel(ctx)
+	}
+
+	remaining := time.Until(deadline)
+	if remaining <= pinholeDiscoveryTimeout {
+		return context.WithCancel(ctx)
+	}
+	return context.WithTimeout(ctx, remaining-pinholeDiscoveryTimeout)
 }
 
 // State is persisted only for crash recovery cleanup

client/internal/portforward/state_test.go

@@ -0,0 +1,140 @@
+//go:build !js
+
+package portforward
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/netbirdio/go-nat"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// stubDiscovery replaces both discovery entry points for the duration of a
+// test. gatewayDelay simulates gateway discovery spending everything it is
+// given before reporting that it found nothing.
+func stubDiscovery(t *testing.T, gateway nat.NAT, gatewayErr error, gatewayDelay time.Duration, pinhole nat.NAT, pinholeErr error) {
+	t.Helper()
+
+	origGateway, origPinhole := discoverNATGateway, discoverPCPPinhole
+	discoverNATGateway = func(ctx context.Context) (nat.NAT, error) {
+		if gatewayDelay > 0 {
+			select {
+			case <-time.After(gatewayDelay):
+			case <-ctx.Done():
+			}
+		}
+		return gateway, gatewayErr
+	}
+	discoverPCPPinhole = func(ctx context.Context) (nat.NAT, error) {
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		return pinhole, pinholeErr
+	}
+
+	t.Cleanup(func() { discoverNATGateway, discoverPCPPinhole = origGateway, origPinhole })
+}
+
+func TestDefaultDiscoverGateway(t *testing.T) {
+	ipv4Gateway := &mockNAT{natType: "PCP+PCPv6"}
+	ipv6Pinhole := &mockNAT{natType: "PCP"}
+	otherErr := errors.New("routing table unavailable")
+
+	t.Run("an IPv4 gateway is used as is", func(t *testing.T) {
+		stubDiscovery(t, ipv4Gateway, nil, 0, ipv6Pinhole, nil)
+
+		got, err := defaultDiscoverGateway(context.Background())
+
+		require.NoError(t, err)
+		assert.Same(t, ipv4Gateway, got)
+	})
+
+	t.Run("no IPv4 gateway still opens an IPv6 pinhole", func(t *testing.T) {
+		stubDiscovery(t, nil, nat.ErrNoNATFound, 0, ipv6Pinhole, nil)
+
+		got, err := defaultDiscoverGateway(context.Background())
+
+		require.NoError(t, err)
+		assert.Same(t, ipv6Pinhole, got)
+	})
+
+	t.Run("no gateway and no pinhole reports the original failure", func(t *testing.T) {
+		stubDiscovery(t, nil, nat.ErrNoNATFound, 0, nil, errors.New("no IPv6 route"))
+
+		got, err := defaultDiscoverGateway(context.Background())
+
+		assert.Nil(t, got)
+		assert.ErrorIs(t, err, nat.ErrNoNATFound, "the pinhole failure must not mask why no gateway was found")
+	})
+
+	t.Run("a failure other than no-gateway is reported as is", func(t *testing.T) {
+		stubDiscovery(t, nil, otherErr, 0, ipv6Pinhole, nil)
+
+		got, err := defaultDiscoverGateway(context.Background())
+
+		assert.Nil(t, got)
+		assert.ErrorIs(t, err, otherErr)
+	})
+
+	t.Run("the pinhole survives gateway discovery using its whole budget", func(t *testing.T) {
+		// On one shared context the probe would start already expired, which is
+		// how this failed against a real gateway.
+		reserve := 50 * time.Millisecond
+		origReserve := pinholeDiscoveryTimeout
+		pinholeDiscoveryTimeout = reserve
+		t.Cleanup(func() { pinholeDiscoveryTimeout = origReserve })
+
+		budget := 4 * reserve
+		ctx, cancel := context.WithTimeout(context.Background(), budget)
+		defer cancel()
+
+		stubDiscovery(t, nil, nat.ErrNoNATFound, budget, ipv6Pinhole, nil)
+
+		got, err := defaultDiscoverGateway(ctx)
+
+		require.NoError(t, err)
+		assert.Same(t, ipv6Pinhole, got)
+	})
+}
+
+func TestReserveForPinhole(t *testing.T) {
+	origReserve := pinholeDiscoveryTimeout
+	pinholeDiscoveryTimeout = time.Second
+	t.Cleanup(func() { pinholeDiscoveryTimeout = origReserve })
+
+	t.Run("a budget is divided", func(t *testing.T) {
+		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+		defer cancel()
+
+		gatewayCtx, cancelGateway := reserveForPinhole(ctx)
+		defer cancelGateway()
+
+		deadline, ok := gatewayCtx.Deadline()
+		require.True(t, ok)
+		assert.InDelta(t, 9*time.Second, time.Until(deadline), float64(500*time.Millisecond))
+	})
+
+	t.Run("a budget too small to divide is left whole", func(t *testing.T) {
+		ctx, cancel := context.WithTimeout(context.Background(), 500*time.Millisecond)
+		defer cancel()
+
+		gatewayCtx, cancelGateway := reserveForPinhole(ctx)
+		defer cancelGateway()
+
+		deadline, ok := gatewayCtx.Deadline()
+		require.True(t, ok)
+		assert.InDelta(t, 500*time.Millisecond, time.Until(deadline), float64(100*time.Millisecond))
+	})
+
+	t.Run("no deadline stays unbounded", func(t *testing.T) {
+		gatewayCtx, cancelGateway := reserveForPinhole(context.Background())
+		defer cancelGateway()
+
+		_, ok := gatewayCtx.Deadline()
+		assert.False(t, ok)
+	})
+}

client/system/info_js.go

@@ -15,7 +15,7 @@ func UpdateStaticInfoAsync() {
 }
 
 // GetInfo retrieves system information for WASM environment
-func GetInfo(_ context.Context) *Info {
+func GetInfo(ctx context.Context) *Info {
 	info := &Info{
 		GoOS:           runtime.GOOS,
 		Kernel:         runtime.GOARCH,
@@ -30,6 +30,13 @@ func GetInfo(_ context.Context) *Info {
 	collectBrowserInfo(info)
 	collectLocationInfo(info)
 	collectSystemInfo(info)
+
+	// A caller-provided device name wins, as on the other platforms. A peer
+	// registered over an API keeps reporting the name it was registered with,
+	// so its meta does not change on the first sync.
+	if name := extractDeviceName(ctx, info.Hostname); name != "" {
+		info.Hostname = name
+	}
 	return info
 }
 

client/system/info_js_test.go

@@ -0,0 +1,27 @@
+//go:build js
+
+package system
+
+import (
+	"context"
+	"testing"
+)
+
+// TestGetInfoHonorsDeviceName covers a caller-provided device name reaching the
+// reported hostname, so a peer registered over an API keeps reporting the name
+// it was registered with instead of renaming itself on its first sync.
+func TestGetInfoHonorsDeviceName(t *testing.T) {
+	ctx := context.WithValue(context.Background(), DeviceNameCtxKey, "session-name")
+	if got := GetInfo(ctx).Hostname; got != "session-name" {
+		t.Errorf("hostname should carry the caller's device name, got %q", got)
+	}
+}
+
+// TestGetInfoWithoutDeviceNameKeepsFallback covers the embed layer's habit of
+// always setting the context value: an empty name must not blank the hostname.
+func TestGetInfoWithoutDeviceNameKeepsFallback(t *testing.T) {
+	ctx := context.WithValue(context.Background(), DeviceNameCtxKey, "")
+	if got := GetInfo(ctx).Hostname; got == "" {
+		t.Error("an empty device name must not blank the hostname")
+	}
+}

client/system/process_test.go

@@ -1,3 +1,5 @@
+//go:build windows || (linux && !android) || (darwin && !ios) || freebsd
+
 package system
 
 import (

client/ui/i18n/locales/de/common.json

@@ -764,7 +764,19 @@
         "message": "Sensible Informationen anonymisieren"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "Versteckt öffentliche IP-Adressen und nicht-NetBird-Domains in Logs."
+        "message": "Verbirgt IP-Adressen, Domains und andere sensible Werte."
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "Der Standardmodus lässt interne IPv4-Adressen und Peer-Namen für den Support lesbar. Der strikte Modus anonymisiert zusätzlich private (RFC 1918), CGNAT- und Link-Local-IP-Adressen, Peer-Namen und öffentliche WireGuard-Schlüssel. Wiederkehrende Werte erhalten denselben Platzhalter, sodass Peers unterscheidbar bleiben. Verwenden Sie den strikten Modus, wenn Sie das Debug-Paket außerhalb Ihrer Organisation weitergeben."
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "Keine"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "Standard"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "Strikt"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "Systeminformationen einschließen"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "Vorgang fehlgeschlagen."
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "Erfordert {actor}. Führen Sie stattdessen dies aus:"
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "Sie können dies deaktivieren, aber zum erneuten Aktivieren sind {actor} erforderlich:"
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "Sie können dies aktivieren, aber zum erneuten Deaktivieren sind {actor} erforderlich:"
     }
 }

client/ui/i18n/locales/es/common.json

@@ -764,7 +764,19 @@
         "message": "Anonimizar información sensible"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "Oculta las direcciones IP públicas y los dominios ajenos a NetBird de los registros."
+        "message": "Oculta direcciones IP, dominios y otros valores sensibles."
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "El modo predeterminado mantiene legibles las direcciones IPv4 internas y los nombres de los peers para el soporte. El modo estricto anonimiza además las direcciones IP privadas (RFC 1918), CGNAT y de enlace local, los nombres de los peers y las claves públicas de WireGuard. Los valores recurrentes se asignan al mismo marcador de posición, por lo que los peers siguen siendo distinguibles. Use el modo estricto cuando comparta el paquete de diagnóstico fuera de su organización."
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "Ninguno"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "Predeterminado"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "Estricto"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "Incluir información del sistema"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "La operación falló."
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "Requiere {actor}. Ejecute esto en su lugar:"
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "Puede desactivarlo, pero volver a activarlo requiere {actor}:"
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "Puede activarlo, pero volver a desactivarlo requiere {actor}:"
     }
 }

client/ui/i18n/locales/fr/common.json

@@ -764,7 +764,19 @@
         "message": "Anonymiser les informations sensibles"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "Masque les adresses IP publiques et les domaines non-NetBird dans les journaux."
+        "message": "Masque les adresses IP, les domaines et d'autres valeurs sensibles."
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "Le mode par défaut garde les adresses IPv4 internes et les noms des pairs lisibles pour le support. Le mode strict anonymise en plus les adresses IP privées (RFC 1918), CGNAT et de lien local, les noms des pairs et les clés publiques WireGuard. Les valeurs récurrentes reçoivent le même espace réservé, les pairs restent donc distinguables. Utilisez le mode strict lorsque vous partagez le lot de diagnostic en dehors de votre organisation."
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "Aucune"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "Par défaut"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "Strict"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "Inclure les informations système"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "L’opération a échoué."
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "Nécessite {actor}. Exécutez plutôt ceci :"
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "Vous pouvez le désactiver, mais le réactiver nécessite {actor} :"
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "Vous pouvez l’activer, mais le désactiver de nouveau nécessite {actor} :"
     }
 }

client/ui/i18n/locales/hu/common.json

@@ -764,7 +764,19 @@
         "message": "Érzékeny információk anonimizálása"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "Elrejti a nyilvános IP-címeket és a nem-NetBird tartományokat a naplókban."
+        "message": "Elrejti az IP-címeket, a tartományokat és más érzékeny értékeket."
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "Az Alapértelmezett szint a belső IPv4-címeket és a peer-neveket olvashatóan hagyja a támogatás számára. A Szigorú ezen felül anonimizálja a privát (RFC 1918), CGNAT és link-local IP-címeket, a peer-neveket és a WireGuard nyilvános kulcsokat. Az ismétlődő értékek ugyanazt a helyettesítőt kapják, így a peerek megkülönböztethetők maradnak. Használja a Szigorú szintet, ha a hibakeresési csomagot a szervezetén kívül osztja meg."
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "Nincs"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "Alapértelmezett"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "Szigorú"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "Rendszerinformációk beillesztése"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "A művelet meghiúsult."
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "{actor} szükséges hozzá. Futtassa inkább ezt:"
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "Kikapcsolhatja, de a visszakapcsolásához {actor} szükséges:"
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "Bekapcsolhatja, de az ismételt kikapcsolásához {actor} szükséges:"
     }
 }

client/ui/i18n/locales/it/common.json

@@ -764,7 +764,19 @@
         "message": "Anonimizza informazioni sensibili"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "Nasconde gli indirizzi IP pubblici e i domini non NetBird dai log."
+        "message": "Nasconde indirizzi IP, domini e altri valori sensibili."
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "La modalità predefinita mantiene leggibili gli indirizzi IPv4 interni e i nomi dei peer per il supporto. La modalità rigorosa anonimizza inoltre gli indirizzi IP privati (RFC 1918), CGNAT e link-local, i nomi dei peer e le chiavi pubbliche WireGuard. I valori ricorrenti vengono associati allo stesso segnaposto, quindi i peer restano distinguibili. Usa la modalità rigorosa quando condividi il pacchetto di debug al di fuori della tua organizzazione."
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "Nessuna"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "Predefinito"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "Rigoroso"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "Includi informazioni di sistema"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "Operazione non riuscita."
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "Richiede {actor}. Esegua invece questo:"
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "Può disabilitarlo, ma riabilitarlo richiede {actor}:"
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "Può abilitarlo, ma disabilitarlo di nuovo richiede {actor}:"
     }
 }

client/ui/i18n/locales/ja/common.json

@@ -764,7 +764,19 @@
         "message": "機密情報を匿名化"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "ログからパブリック IP アドレスと NetBird 以外のドメインを隠します。"
+        "message": "IP アドレス、ドメイン、その他の機密性の高い値を隠します。"
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "「デフォルト」では、サポートのために内部 IPv4 アドレスとピア名は読める状態のまま残ります。「厳格」では、さらにプライベート (RFC 1918)、CGNAT、リンクローカルの IP アドレス、ピア名、WireGuard 公開鍵も匿名化されます。繰り返し現れる値は同じプレースホルダーに置き換えられるため、ピアは区別できます。デバッグバンドルを組織外に共有する場合は「厳格」を使用してください。"
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "なし"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "デフォルト"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "厳格"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "システム情報を含める"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "操作に失敗しました。"
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "{actor}が必要です。代わりに次のコマンドを実行してください:"
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "無効にはできますが、再度有効にするには{actor}が必要です:"
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "有効にはできますが、再度無効にするには{actor}が必要です:"
     }
 }

client/ui/i18n/locales/pt/common.json

@@ -764,7 +764,19 @@
         "message": "Anonimizar informações sensíveis"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "Oculta endereços IP públicos e domínios que não são do NetBird nos logs."
+        "message": "Oculta endereços IP, domínios e outros valores sensíveis."
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "O modo padrão mantém os endereços IPv4 internos e os nomes dos peers legíveis para o suporte. O modo estrito anonimiza também os endereços IP privados (RFC 1918), CGNAT e link-local, os nomes dos peers e as chaves públicas do WireGuard. Valores recorrentes recebem o mesmo marcador, então os peers continuam distinguíveis. Use o modo estrito ao compartilhar o pacote de depuração fora da sua organização."
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "Nenhum"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "Padrão"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "Estrito"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "Incluir informações do sistema"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "A operação falhou."
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "Requer {actor}. Execute isto em vez disso:"
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "Você pode desativar isto, mas ativar novamente requer {actor}:"
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "Você pode ativar isto, mas desativar novamente requer {actor}:"
     }
 }

client/ui/i18n/locales/ru/common.json

@@ -764,7 +764,19 @@
         "message": "Анонимизировать конфиденциальную информацию"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "Скрывает публичные IP-адреса и сторонние (не относящиеся к NetBird) домены в журналах."
+        "message": "Скрывает IP-адреса, домены и другие конфиденциальные значения."
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "Режим «По умолчанию» оставляет внутренние IPv4-адреса и имена пиров читаемыми для поддержки. Режим «Строгий» дополнительно анонимизирует частные (RFC 1918), CGNAT и link-local IP-адреса, имена пиров и публичные ключи WireGuard. Повторяющиеся значения заменяются одним и тем же заполнителем, поэтому пиры остаются различимыми. Используйте режим «Строгий», когда передаёте отладочный пакет за пределы вашей организации."
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "Нет"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "По умолчанию"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "Строгий"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "Включить сведения о системе"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "Не удалось выполнить операцию."
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "Требуются {actor}. Выполните вместо этого:"
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "Отключить можно, но чтобы включить снова, нужны {actor}:"
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "Включить можно, но чтобы отключить снова, нужны {actor}:"
     }
 }

client/ui/i18n/locales/zh-CN/common.json

@@ -764,7 +764,19 @@
         "message": "匿名化敏感信息"
     },
     "settings.troubleshooting.anonymize.help": {
-        "message": "从日志中隐藏公共 IP 地址和非 NetBird 域名。"
+        "message": "隐藏 IP 地址、域名和其他敏感值。"
+    },
+    "settings.troubleshooting.anonymize.info": {
+        "message": "默认级别保留内部 IPv4 地址和对等节点名称，便于支持人员阅读。严格级别还会匿名化私有 (RFC 1918)、CGNAT 和链路本地 IP 地址、对等节点名称以及 WireGuard 公钥。相同的值会映射到相同的占位符，因此对等节点仍可区分。向组织外部分享调试包时请使用严格级别。"
+    },
+    "settings.troubleshooting.anonymize.none": {
+        "message": "无"
+    },
+    "settings.troubleshooting.anonymize.default": {
+        "message": "默认"
+    },
+    "settings.troubleshooting.anonymize.strict": {
+        "message": "严格"
     },
     "settings.troubleshooting.systemInfo.label": {
         "message": "包含系统信息"
@@ -1338,5 +1350,14 @@
     },
     "error.unknown": {
         "message": "操作失败。"
+    },
+    "settings.ssh.privilege.hint": {
+        "message": "需要{actor}。请改为运行："
+    },
+    "settings.ssh.privilege.oneWay": {
+        "message": "您可以关闭此项，但重新开启需要{actor}："
+    },
+    "settings.ssh.privilege.oneWayInverted": {
+        "message": "您可以开启此项，但再次关闭需要{actor}："
     }
 }

client/wasm/cmd/main.go

@@ -56,8 +56,7 @@ func startClient(ctx context.Context, nbClient *netbird.Client) error {
 // parseClientOptions extracts NetBird options from JavaScript object
 func parseClientOptions(jsOptions js.Value) (netbird.Options, error) {
 	options := netbird.Options{
-		DeviceName: "dashboard-client",
-		LogLevel:   defaultLogLevel,
+		LogLevel: defaultLogLevel,
 	}
 
 	if jwtToken := jsOptions.Get("jwtToken"); !jwtToken.IsNull() && !jwtToken.IsUndefined() {
@@ -87,13 +86,41 @@ func parseClientOptions(jsOptions js.Value) (netbird.Options, error) {
 		options.DeviceName = deviceName.String()
 	}
 
-	if disableIPv6 := jsOptions.Get("disableIPv6"); !disableIPv6.IsNull() && !disableIPv6.IsUndefined() {
-		options.DisableIPv6 = disableIPv6.Bool()
+	disableIPv6, err := boolOption(jsOptions, "disableIPv6")
+	if err != nil {
+		return options, err
+	}
+	if disableIPv6 != nil {
+		options.DisableIPv6 = *disableIPv6
 	}
 
+	// The caller decides whether this client uses lazy connections; left unset it
+	// defers to the management feature flag. A short-lived, interactive caller
+	// turns it off so its sessions reach the few peers their grant covers eagerly,
+	// instead of the first request waiting for the connection to be established.
+	lazyConnectionEnabled, err := boolOption(jsOptions, "lazyConnectionEnabled")
+	if err != nil {
+		return options, err
+	}
+	options.LazyConnectionEnabled = lazyConnectionEnabled
+
 	return options, nil
 }
 
+// boolOption reads a boolean option, returning nil when the caller left it out.
+// js.Value.Bool panics on any other type, so a wrong type is reported instead.
+func boolOption(jsOptions js.Value, name string) (*bool, error) {
+	v := jsOptions.Get(name)
+	if v.IsNull() || v.IsUndefined() {
+		return nil, nil
+	}
+	if v.Type() != js.TypeBoolean {
+		return nil, fmt.Errorf("option %s must be a boolean, got %s", name, v.Type())
+	}
+	b := v.Bool()
+	return &b, nil
+}
+
 // createStartMethod creates the start method for the client
 func createStartMethod(client *netbird.Client) js.Func {
 	return js.FuncOf(func(this js.Value, args []js.Value) any {

client/wasm/cmd/main_test.go

@@ -0,0 +1,64 @@
+//go:build js
+
+package main
+
+import (
+	"syscall/js"
+	"testing"
+)
+
+// TestParseClientOptionsBooleans covers the boolean options against the value
+// kinds a JS caller can pass: js.Value.Bool panics on anything but a boolean,
+// so a wrong type has to be rejected before it reaches the client.
+func TestParseClientOptionsBooleans(t *testing.T) {
+	t.Run("unset leaves the lazy override empty", func(t *testing.T) {
+		options, err := parseClientOptions(js.Global().Get("Object").New())
+		if err != nil {
+			t.Fatalf("parse options: %v", err)
+		}
+		if options.LazyConnectionEnabled != nil {
+			t.Errorf("lazy override should stay unset, got %v", *options.LazyConnectionEnabled)
+		}
+		if options.DisableIPv6 {
+			t.Error("disableIPv6 should default to false")
+		}
+	})
+
+	t.Run("null defers to the management flag", func(t *testing.T) {
+		jsOptions := js.Global().Get("Object").New()
+		jsOptions.Set("lazyConnectionEnabled", js.Null())
+		options, err := parseClientOptions(jsOptions)
+		if err != nil {
+			t.Fatalf("parse options: %v", err)
+		}
+		if options.LazyConnectionEnabled != nil {
+			t.Errorf("lazy override should stay unset, got %v", *options.LazyConnectionEnabled)
+		}
+	})
+
+	t.Run("booleans are carried through", func(t *testing.T) {
+		jsOptions := js.Global().Get("Object").New()
+		jsOptions.Set("lazyConnectionEnabled", false)
+		jsOptions.Set("disableIPv6", true)
+		options, err := parseClientOptions(jsOptions)
+		if err != nil {
+			t.Fatalf("parse options: %v", err)
+		}
+		if options.LazyConnectionEnabled == nil || *options.LazyConnectionEnabled {
+			t.Errorf("lazy override should be false, got %v", options.LazyConnectionEnabled)
+		}
+		if !options.DisableIPv6 {
+			t.Error("disableIPv6 should be true")
+		}
+	})
+
+	t.Run("a non-boolean is rejected", func(t *testing.T) {
+		for _, value := range []any{"true", 1, js.Global().Get("Object").New()} {
+			jsOptions := js.Global().Get("Object").New()
+			jsOptions.Set("lazyConnectionEnabled", value)
+			if _, err := parseClientOptions(jsOptions); err == nil {
+				t.Errorf("value %v should be rejected", value)
+			}
+		}
+	})
+}

go.mod

@@ -73,7 +73,6 @@ require (
 	github.com/hashicorp/go-version v1.7.0
 	github.com/jackc/pgx/v5 v5.5.5
 	github.com/libdns/route53 v1.5.0
-	github.com/libp2p/go-nat v0.2.0
 	github.com/libp2p/go-netroute v0.4.0
 	github.com/lrh3321/ipset-go v0.0.0-20250619021614-54a0a98ace81
 	github.com/mdlayher/socket v0.5.1
@@ -81,6 +80,7 @@ require (
 	github.com/miekg/dns v1.1.72
 	github.com/mitchellh/hashstructure/v2 v2.0.2
 	github.com/moby/moby/api v1.54.1
+	github.com/netbirdio/go-nat v0.0.0-20260821095157-6b2c8c5c74e8
 	github.com/netbirdio/management-integrations/integrations v0.0.0-20260416123949-2355d972be42
 	github.com/netbirdio/signal-dispatcher/dispatcher v0.0.0-20250805121659-6b4ac470ca45
 	github.com/oapi-codegen/runtime v1.1.2

go.sum

@@ -407,8 +407,6 @@ github.com/libdns/libdns v0.2.2 h1:O6ws7bAfRPaBsgAYt8MDe2HcNBGC29hkZ9MX2eUSX3s=
 github.com/libdns/libdns v0.2.2/go.mod h1:4Bj9+5CQiNMVGf87wjX4CY3HQJypUHRuLvlsfsZqLWQ=
 github.com/libdns/route53 v1.5.0 h1:2SKdpPFl/qgWsXQvsLNJJAoX7rSxlk7zgoL4jnWdXVA=
 github.com/libdns/route53 v1.5.0/go.mod h1:joT4hKmaTNKHEwb7GmZ65eoDz1whTu7KKYPS8ZqIh6Q=
-github.com/libp2p/go-nat v0.2.0 h1:Tyz+bUFAYqGyJ/ppPPymMGbIgNRH+WqC5QrT5fKrrGk=
-github.com/libp2p/go-nat v0.2.0/go.mod h1:3MJr+GRpRkyT65EpVPBstXLvOlAPzUVlG6Pwg9ohLJk=
 github.com/libp2p/go-netroute v0.4.0 h1:sZZx9hyANYUx9PZyqcgE/E1GUG3iEtTZHUEvdtXT7/Q=
 github.com/libp2p/go-netroute v0.4.0/go.mod h1:Nkd5ShYgSMS5MUKy/MU2T57xFoOKvvLR92Lic48LEyA=
 github.com/lrh3321/ipset-go v0.0.0-20250619021614-54a0a98ace81 h1:J56rFEfUTFT9j9CiRXhi1r8lUJ4W5idG3CiaBZGojNU=
@@ -480,6 +478,8 @@ github.com/netbirdio/dex/api/v2 v2.0.0-20260512110716-8d70ad8647c1 h1:neE7z+FPUk
 github.com/netbirdio/dex/api/v2 v2.0.0-20260512110716-8d70ad8647c1/go.mod h1:awuTyT29CYALpEyET0S307EgNlPWrc7fFKRAyhsO45M=
 github.com/netbirdio/easyjson v0.9.0 h1:6Nw2lghSVuy8RSkAYDhDv1thBVEmfVbKZnV7T7Z6Aus=
 github.com/netbirdio/easyjson v0.9.0/go.mod h1:1+xMtQp2MRNVL/V1bOzuP3aP8VNwRW55fQUto+XFtTU=
+github.com/netbirdio/go-nat v0.0.0-20260821095157-6b2c8c5c74e8 h1:pBxXEsxcsO3qVUND//5j1kelYlO57x5IrRviNF0+0iA=
+github.com/netbirdio/go-nat v0.0.0-20260821095157-6b2c8c5c74e8/go.mod h1:mFViabv4PpnoDw9w7W21a7xux6APA4q7KQZRsv4BCl8=
 github.com/netbirdio/ice/v4 v4.0.0-20250908184934-6202be846b51 h1:Ov4qdafATOgGMB1wbSuh+0aAHcwz9hdvB6VZjh1mVMI=
 github.com/netbirdio/ice/v4 v4.0.0-20250908184934-6202be846b51/go.mod h1:ZSIbPdBn5hePO8CpF1PekH2SfpTxg1PDhEwtbqZS7R8=
 github.com/netbirdio/management-integrations/integrations v0.0.0-20260416123949-2355d972be42 h1:F3zS5fT9xzD1OFLfcdAE+3FfyiwjGukF1hvj0jErgs8=

infrastructure_files/migrate-to-enterprise.sh

@@ -15,6 +15,12 @@ set -o pipefail
 #   2. Postgres migration — add Postgres, migrate SQLite data via migrate-store.
 #   3. Traffic flow       — add NATS + flow-enricher + flow-receiver.
 #
+# Step 2 is skipped when the deployment already runs on Postgres
+# (server.store.engine: postgres in config.yaml). Nothing is provisioned or
+# migrated in that case and the store config is left exactly as the operator
+# wrote it — the enterprise image reads the same Postgres the community image
+# did. Such a deployment gets the image swap, and can still opt into step 3.
+#
 # If any step fails once the stack has been touched, the script rolls itself
 # back automatically: generated files are removed, the Postgres volume this run
 # created is dropped, and the original deployment is started again.
@@ -38,6 +44,18 @@ ENV_BACKUP=""
 PG_VOLUME_NAME=""
 BACKUP_DIR=""
 
+# Store state. STORE_ENGINE is what the deployment runs on today; when it is
+# already postgres, MIGRATE_POSTGRES stays "no" and nothing is provisioned.
+# POSTGRES_SERVICE is empty when Postgres lives outside this compose project.
+STORE_ENGINE=""
+EXISTING_POSTGRES="no"
+POSTGRES_DSN=""
+POSTGRES_SERVICE=""
+POSTGRES_DEPENDS_CONDITION="service_healthy"
+# Whether this run needs to generate config.yaml.enterprise at all. A pure
+# image swap does not.
+ENTERPRISE_CONFIG="no"
+
 NETBIRD_EULA_URL="https://netbird.io/self-hosted-EULA"
 
 check_docker_compose() {
@@ -192,6 +210,85 @@ detect_exposed_address() {
   yq eval '.server.exposedAddress // ""' "$CONFIG_YAML_HOST"
 }
 
+# The engine is a config.yaml-only setting — there is no env override for it
+# (combined/cmd/root.go reads it from YAML and derives the env vars), so
+# config.yaml is authoritative. Absent means the sqlite default.
+detect_store_engine() {
+  local engine
+  engine=$(yq eval '.server.store.engine // ""' "$CONFIG_YAML_HOST")
+  if [[ -z "$engine" ]] || [[ "$engine" == "null" ]]; then
+    engine="sqlite"
+  fi
+  echo "$engine" | tr '[:upper:]' '[:lower:]'
+}
+
+detect_store_dsn() {
+  yq eval '.server.store.dsn // ""' "$CONFIG_YAML_HOST"
+}
+
+# config.yaml is where a combined deployment carries its DSN; this only covers
+# hand-rolled installs that keep it in the environment instead.
+detect_store_dsn_from_compose() {
+  # `compose config` re-escapes a literal $ as $$ on the way out, so undo that
+  # to get the value the container actually receives.
+  $DOCKER_COMPOSE_COMMAND config 2>/dev/null | yq eval "
+    .services[\"$COMBINED_SERVICE\"].environment.NB_STORE_ENGINE_POSTGRES_DSN //
+    .services[\"$COMBINED_SERVICE\"].environment.NETBIRD_STORE_ENGINE_POSTGRES_DSN // \"\"
+  " - 2>/dev/null | sed 's/\$\$/$/g'
+}
+
+# Reads either DSN form: "host=db ..." or "postgres://user:pass@db:5432/name".
+dsn_host() {
+  local dsn="$1"
+  case "$dsn" in
+    *://*) printf '%s' "$dsn" | sed -n 's,^[a-zA-Z+]*://\([^/?]*\).*,\1,p' | sed -e 's,.*@,,' -e 's,:.*,,' ;;
+    *) printf '%s' "$dsn" | sed -n 's/.*[[:space:]]*host=\([^[:space:]]*\).*/\1/p' ;;
+  esac
+}
+
+# flow-enricher is its own container, so a loopback host or a socket path would
+# reach the enricher rather than Postgres. Only flag hosts we can positively
+# identify — an unparseable DSN must not leave the operator with no way forward.
+dsn_host_reachable() {
+  local dsn="$1"
+  case "$(dsn_host "$dsn")" in
+    localhost | 127.* | ::1 | 0.0.0.0 | /*) return 1 ;;
+    *) return 0 ;;
+  esac
+}
+
+# Names the compose service running this deployment's Postgres, for depends_on.
+# Empty means external — the DSN host matched no service. A DSN with no readable
+# host falls back to matching on image.
+detect_postgres_service() {
+  local host
+  host=$(dsn_host "$POSTGRES_DSN")
+  if [[ -n "$host" ]]; then
+    if [[ "$(host="$host" yq eval '.services | has(env(host))' "$COMPOSE_FILE" 2>/dev/null)" == "true" ]]; then
+      echo "$host"
+    fi
+    return
+  fi
+  yq eval '.services | to_entries | map(select(.value.image // "" | test("(^|/)(postgres|postgis|pgvector|timescaledb)(:|@|$)"))) | .[0].key // ""' "$COMPOSE_FILE"
+}
+
+# depends_on: service_healthy is only legal if the service defines a healthcheck.
+detect_postgres_depends_condition() {
+  local tag
+  tag=$(yq eval ".services[\"$POSTGRES_SERVICE\"].healthcheck | tag" "$COMPOSE_FILE" 2>/dev/null)
+  if [[ "$tag" == "!!map" ]]; then
+    echo "service_healthy"
+  else
+    echo "service_started"
+  fi
+}
+
+env_value() {
+  local value="$1"
+  value=$(printf '%s' "$value" | sed -e 's/\\/\\\\/g' -e 's/"/\\"/g' -e 's/\$/$$/g')
+  printf '"%s"' "$value"
+}
+
 detect_compose_network() {
   local tag
   tag=$(yq eval ".services[\"$COMBINED_SERVICE\"].networks | tag" "$COMPOSE_FILE" 2>/dev/null)
@@ -228,16 +325,30 @@ services:
       NETBIRD_LICENSE_SERVER_BASE_URL: \${NETBIRD_LICENSE_SERVER_BASE_URL}
 EOF
 
+  # An existing Postgres is already wired up by the operator's own compose file,
+  # so only a Postgres this run creates needs a depends_on.
   if [[ "$MIGRATE_POSTGRES" == "yes" ]]; then
     cat <<EOF
     depends_on:
-      postgres:
-        condition: service_healthy
+      ${POSTGRES_SERVICE}:
+        condition: ${POSTGRES_DEPENDS_CONDITION}
+EOF
+  fi
+
+  # The server is only pointed at a different config file when this run
+  # generates one. A pure image swap leaves it on its original config.yaml.
+  if [[ "$ENTERPRISE_CONFIG" == "yes" ]]; then
+    cat <<EOF
     volumes:
       - ./${ENTERPRISE_CONFIG_FILE}:/etc/netbird/config.yaml.enterprise:ro
     command: ["--config", "/etc/netbird/config.yaml.enterprise"]
+EOF
+  fi
 
-  postgres:
+  if [[ "$MIGRATE_POSTGRES" == "yes" ]]; then
+    cat <<EOF
+
+  ${POSTGRES_SERVICE}:
     image: postgres:17
     container_name: netbird-postgres
     restart: unless-stopped
@@ -257,6 +368,14 @@ EOF
   fi
 
   if [[ "$ENABLE_FLOW" == "yes" ]]; then
+    # Nothing to wait on when Postgres is managed outside this compose project.
+    local enricher_depends=""
+    if [[ -n "$POSTGRES_SERVICE" ]]; then
+      enricher_depends="
+      ${POSTGRES_SERVICE}:
+        condition: ${POSTGRES_DEPENDS_CONDITION}"
+    fi
+
     cat <<EOF
 
   nats:
@@ -273,9 +392,7 @@ EOF
     container_name: netbird-flow-enricher
     restart: unless-stopped
     networks: [${COMPOSE_NETWORK}]
-    depends_on:
-      postgres:
-        condition: service_healthy
+    depends_on:${enricher_depends}
       nats:
         condition: service_started
     environment:
@@ -283,10 +400,10 @@ EOF
       NETBIRD_LICENSE_SERVER_BASE_URL: \${NETBIRD_LICENSE_SERVER_BASE_URL}
       NB_DATADIR: /var/lib/netbird
       NB_MANAGEMENT_STORE_ENGINE: postgres
-      NB_MANAGEMENT_POSTGRES_DSN: "host=postgres user=netbird password=\${POSTGRES_PASSWORD} dbname=netbird port=5432 sslmode=disable"
-      NB_STORE_ENGINE_POSTGRES_DSN: "host=postgres user=netbird password=\${POSTGRES_PASSWORD} dbname=netbird port=5432 sslmode=disable"
+      NB_MANAGEMENT_POSTGRES_DSN: \${NB_ENTERPRISE_POSTGRES_DSN}
+      NB_STORE_ENGINE_POSTGRES_DSN: \${NB_ENTERPRISE_POSTGRES_DSN}
       NB_TRAFFIC_EVENT_STORE_ENGINE: postgres
-      NB_TRAFFIC_EVENT_POSTGRES_DSN: "host=postgres user=netbird password=\${POSTGRES_PASSWORD} dbname=netbird port=5432 sslmode=disable"
+      NB_TRAFFIC_EVENT_POSTGRES_DSN: \${NB_ENTERPRISE_POSTGRES_DSN}
       NB_MANAGEMENT_STORE_KEY: \${NETBIRD_ENCRYPTION_KEY}
       NB_FLOW_ADAPTER_TYPE: nats
       NB_FLOW_NATS_ENDPOINTS: nats://nats:4222
@@ -343,27 +460,41 @@ EOF
   fi
 }
 
-# Build config.yaml.enterprise by yq-editing the operator's existing
-# config.yaml. We don't touch the original file.
+# Build config.yaml.enterprise from the operator's existing config.yaml. We
+# don't touch the original file. Values go through strenv() so a DSN carrying
+# quotes, backslashes or $ cannot break out of the expression.
 render_enterprise_config() {
-  local pg_dsn="host=postgres user=netbird password=${POSTGRES_PASSWORD} dbname=netbird port=5432 sslmode=disable"
+  {
+    echo "# Generated by migrate-to-enterprise.sh from ${CONFIG_YAML_HOST}."
+    echo "# The enterprise server is started with --config pointing at this file,"
+    echo "# so later edits to ${CONFIG_YAML_HOST} have no effect until copied here."
+    cat "$CONFIG_YAML_HOST"
+  } > "$ENTERPRISE_CONFIG_FILE"
 
-  yq eval "
-    .server.store.engine = \"postgres\" |
-    .server.store.dsn = \"$pg_dsn\" |
-    .server.activityStore.engine = \"postgres\" |
-    .server.activityStore.dsn = \"$pg_dsn\" |
-    .server.authStore.engine = \"postgres\" |
-    .server.authStore.dsn = \"$pg_dsn\"
-  " "$CONFIG_YAML_HOST" > "$ENTERPRISE_CONFIG_FILE"
+  if [[ "$MIGRATE_POSTGRES" == "yes" ]]; then
+    # Fresh Postgres: point every store section at it. migrate-store carries the
+    # SQLite contents across.
+    POSTGRES_DSN="$POSTGRES_DSN" yq eval -i '
+      .server.store.engine = "postgres" |
+      .server.store.dsn = strenv(POSTGRES_DSN) |
+      .server.activityStore.engine = "postgres" |
+      .server.activityStore.dsn = strenv(POSTGRES_DSN) |
+      .server.authStore.engine = "postgres" |
+      .server.authStore.dsn = strenv(POSTGRES_DSN)
+    ' "$ENTERPRISE_CONFIG_FILE"
+  fi
+  # Otherwise the store config is the operator's and stays untouched.
+  # activityStore and authStore do not inherit from server.store — each falls
+  # back to its own SQLite file under dataDir — so repointing them at Postgres
+  # here would silently strand the existing audit log and the embedded IdP's
+  # users, with no migrate-store run to carry them over.
 
   if [[ "$ENABLE_FLOW" == "yes" ]]; then
-    local flow_addr="${NETBIRD_DOMAIN}"
-    yq eval -i "
+    NETBIRD_DOMAIN="$NETBIRD_DOMAIN" yq eval -i '
       .server.trafficFlow.enabled = true |
-      .server.trafficFlow.address = \"$flow_addr\" |
-      .server.trafficFlow.interval = \"60s\"
-    " "$ENTERPRISE_CONFIG_FILE"
+      .server.trafficFlow.address = strenv(NETBIRD_DOMAIN) |
+      .server.trafficFlow.interval = "60s"
+    ' "$ENTERPRISE_CONFIG_FILE"
   fi
 }
 
@@ -630,6 +761,91 @@ on_exit() {
 # Main
 # ---------------------------------------------------------------------------
 
+# Already on Postgres: there is nothing to provision and nothing to migrate.
+# The enterprise image reads the very same store config the community image
+# did, so step 2 collapses to a no-op and the run is a plain image swap.
+configure_existing_postgres() {
+  EXISTING_POSTGRES="yes"
+  MIGRATE_POSTGRES="no"
+
+  # DSN first — detect_postgres_service prefers the host it names.
+  POSTGRES_DSN=$(detect_store_dsn)
+  if [[ -z "$POSTGRES_DSN" ]] || [[ "$POSTGRES_DSN" == "null" ]]; then
+    POSTGRES_DSN=$(detect_store_dsn_from_compose)
+  fi
+  if [[ "$POSTGRES_DSN" == "null" ]]; then
+    POSTGRES_DSN=""
+  fi
+
+  POSTGRES_SERVICE=$(detect_postgres_service)
+  if [[ -n "$POSTGRES_SERVICE" ]]; then
+    POSTGRES_DEPENDS_CONDITION=$(detect_postgres_depends_condition)
+  fi
+
+  echo "Step 2: Postgres migration not needed — this deployment already runs on"
+  echo "        Postgres. Its store configuration is reused as-is and left"
+  echo "        untouched; no database is created and no data is moved."
+  if [[ -n "$POSTGRES_SERVICE" ]]; then
+    echo "        Postgres service: $POSTGRES_SERVICE (in $COMPOSE_FILE)"
+  else
+    echo "        Postgres service: managed outside $COMPOSE_FILE"
+  fi
+}
+
+configure_sqlite_store() {
+  MIGRATE_POSTGRES=$(read_yes_no "Step 2: Migrate storage from SQLite to Postgres? (recommended)" "n")
+  [[ "$MIGRATE_POSTGRES" == "yes" ]] || return 0
+
+  # The override would otherwise merge into a service of the same name and
+  # quietly rewrite its image and credentials.
+  local existing
+  existing=$(yq eval '.services | has("postgres")' "$COMPOSE_FILE")
+  if [[ "$existing" == "true" ]]; then
+    echo "" > /dev/stderr
+    echo "$COMPOSE_FILE already defines a service named 'postgres', but config.yaml" > /dev/stderr
+    echo "still has server.store.engine: sqlite. This script would add its own" > /dev/stderr
+    echo "'postgres' service and Compose would merge the two." > /dev/stderr
+    echo "" > /dev/stderr
+    echo "Point server.store.engine at that Postgres yourself, or rename the service," > /dev/stderr
+    echo "then re-run." > /dev/stderr
+    exit 1
+  fi
+
+  echo ""
+  echo "  ⚠  Data will be migrated from SQLite to Postgres. The SQLite store"
+  echo "     will be backed up automatically. To fully revert later, restore"
+  echo "     that backup and delete docker-compose.override.yml +"
+  echo "     config.yaml.enterprise."
+  local confirm
+  confirm=$(read_yes_no "  Continue?" "y")
+  if [[ "$confirm" != "yes" ]]; then
+    MIGRATE_POSTGRES="no"
+    echo "  Skipping Postgres migration."
+    return 0
+  fi
+
+  POSTGRES_PASSWORD=$(rand_password)
+  POSTGRES_SERVICE="postgres"
+  POSTGRES_DEPENDS_CONDITION="service_healthy"
+  POSTGRES_DSN="host=postgres user=netbird password=${POSTGRES_PASSWORD} dbname=netbird port=5432 sslmode=disable"
+}
+
+# mysql, or something this script has never seen. Swapping the images is still
+# valid; touching the store is not.
+configure_unsupported_store() {
+  MIGRATE_POSTGRES="no"
+  echo "  ⚠  server.store.engine is '$STORE_ENGINE'. This script only migrates"
+  echo "     SQLite to Postgres, and traffic flow requires Postgres, so both are"
+  echo "     unavailable here. The store configuration will be left untouched."
+  echo ""
+  local proceed
+  proceed=$(read_yes_no "Step 2 skipped. Continue with the image swap only?" "n")
+  if [[ "$proceed" != "yes" ]]; then
+    echo "Aborted."
+    exit 0
+  fi
+}
+
 init_migration() {
   DOCKER_COMPOSE_COMMAND=$(check_docker_compose)
   check_yq
@@ -679,12 +895,15 @@ init_migration() {
     exit 1
   fi
 
+  STORE_ENGINE=$(detect_store_engine)
+
   echo "Detected existing deployment:"
   echo "  Combined service: $COMBINED_SERVICE"
   echo "  Dashboard:        $DASHBOARD_SERVICE"
   echo "  config.yaml:      $CONFIG_YAML_HOST"
   echo "  Data volume:      $DATA_VOLUME"
   echo "  Network:          $COMPOSE_NETWORK"
+  echo "  Store engine:     $STORE_ENGINE"
   echo ""
 
   require_eula_acceptance
@@ -703,28 +922,17 @@ init_migration() {
   echo "Step 1: Image swap (community → Enterprise). License key required."
   NB_LICENSE_KEY=$(read_secret "  License key")
 
-  # Step 2 — optional
+  # Step 2 — what this does depends on what the deployment already stores in.
   echo ""
-  MIGRATE_POSTGRES=$(read_yes_no "Step 2: Migrate storage from SQLite to Postgres? (recommended)" "n")
-  if [[ "$MIGRATE_POSTGRES" == "yes" ]]; then
-    echo ""
-    echo "  ⚠  Data will be migrated from SQLite to Postgres. The SQLite store"
-    echo "     will be backed up automatically. To fully revert later, restore"
-    echo "     that backup and delete docker-compose.override.yml +"
-    echo "     config.yaml.enterprise."
-    local confirm
-    confirm=$(read_yes_no "  Continue?" "y")
-    if [[ "$confirm" != "yes" ]]; then
-      MIGRATE_POSTGRES="no"
-      echo "  Skipping Postgres migration."
-    else
-      POSTGRES_PASSWORD=$(rand_password)
-    fi
-  fi
+  case "$STORE_ENGINE" in
+    postgres) configure_existing_postgres ;;
+    sqlite) configure_sqlite_store ;;
+    *) configure_unsupported_store ;;
+  esac
 
   # Step 3 — optional, only if Postgres is on (flow requires Postgres)
   echo ""
-  if [[ "$MIGRATE_POSTGRES" == "yes" ]]; then
+  if [[ "$MIGRATE_POSTGRES" == "yes" ]] || [[ "$EXISTING_POSTGRES" == "yes" ]]; then
     ENABLE_FLOW=$(read_yes_no "Step 3: Enable traffic flow? (requires Postgres)" "n")
     if [[ "$ENABLE_FLOW" == "yes" ]]; then
       # Auth secret MUST match server.authSecret from config.yaml
@@ -748,12 +956,46 @@ init_migration() {
         echo "Could not read server.store.encryptionKey from $CONFIG_YAML_HOST." > /dev/stderr
         exit 1
       fi
+
+      # flow-enricher talks to Postgres directly, so this is the one place an
+      # existing deployment's DSN is actually needed — and the one place a host
+      # that only works from inside the server container shows up.
+      while :; do
+        local dsn_problem=""
+        if [[ -z "$POSTGRES_DSN" ]]; then
+          dsn_problem="No DSN could be read from $CONFIG_YAML_HOST or from the $COMBINED_SERVICE environment."
+        elif ! dsn_host_reachable "$POSTGRES_DSN"; then
+          dsn_problem="Its host '$(dsn_host "$POSTGRES_DSN")' only resolves inside the server container."
+        fi
+        [[ -n "$dsn_problem" ]] || break
+
+        echo ""
+        echo "  The flow enricher reaches Postgres from a container of its own."
+        echo "  $dsn_problem"
+        echo "  Enter a DSN reachable from other containers, or press Ctrl-C to abort."
+        POSTGRES_DSN=$(read_required "  Postgres DSN (host=… user=… password=… dbname=… port=5432 sslmode=disable)")
+      done
+
+      # Only where the operator owns Postgres: a DSN entered above may name a
+      # different host. The sqlite path creates its own service, nothing to find.
+      if [[ "$EXISTING_POSTGRES" == "yes" ]]; then
+        POSTGRES_SERVICE=$(detect_postgres_service)
+        if [[ -n "$POSTGRES_SERVICE" ]]; then
+          POSTGRES_DEPENDS_CONDITION=$(detect_postgres_depends_condition)
+        fi
+      fi
     fi
   else
     ENABLE_FLOW="no"
     echo "Step 3 (traffic flow) skipped — requires Postgres."
   fi
 
+  # config.yaml.enterprise only exists to hold changes; without any there is
+  # nothing to generate and the server keeps running on its own config.yaml.
+  if [[ "$MIGRATE_POSTGRES" == "yes" ]] || [[ "$ENABLE_FLOW" == "yes" ]]; then
+    ENTERPRISE_CONFIG="yes"
+  fi
+
   check_data_directory
   check_stale_postgres_volume
 }
@@ -771,7 +1013,7 @@ apply_changes() {
     sed -i.bak '/NETBIRD_LICENSE_SERVER_BASE_URL/d' "$OVERRIDE_FILE" && rm -f "$OVERRIDE_FILE.bak"
   fi
 
-  if [[ "$MIGRATE_POSTGRES" == "yes" ]]; then
+  if [[ "$ENTERPRISE_CONFIG" == "yes" ]]; then
     echo "Writing $ENTERPRISE_CONFIG_FILE ..."
     install -m 600 /dev/null "$ENTERPRISE_CONFIG_FILE"
     render_enterprise_config
@@ -807,6 +1049,9 @@ apply_changes() {
       echo "POSTGRES_PASSWORD=${POSTGRES_PASSWORD}"
     fi
     if [[ "$ENABLE_FLOW" == "yes" ]]; then
+      # Own variable name rather than NB_STORE_ENGINE_POSTGRES_DSN so that a
+      # deployment already setting that one keeps its own value.
+      echo "NB_ENTERPRISE_POSTGRES_DSN=$(env_value "$POSTGRES_DSN")"
       echo "NB_FLOW_AUTH_SECRET=${NB_FLOW_AUTH_SECRET}"
       echo "NETBIRD_ENCRYPTION_KEY=${NETBIRD_ENCRYPTION_KEY}"
     fi
@@ -868,14 +1113,19 @@ print_summary() {
   echo " Summary"
   echo "──────────────────────────────────────────────────────────────────────"
   echo "  Images:           swapped to enterprise"
-  [[ "$MIGRATE_POSTGRES" == "yes" ]] && echo "  Storage:          Postgres (data migrated from SQLite)"
-  [[ "$MIGRATE_POSTGRES" != "yes" ]] && echo "  Storage:          SQLite (unchanged)"
+  if [[ "$MIGRATE_POSTGRES" == "yes" ]]; then
+    echo "  Storage:          Postgres (data migrated from SQLite)"
+  elif [[ "$EXISTING_POSTGRES" == "yes" ]]; then
+    echo "  Storage:          Postgres (pre-existing, configuration unchanged)"
+  else
+    echo "  Storage:          $STORE_ENGINE (unchanged)"
+  fi
   [[ "$ENABLE_FLOW" == "yes" ]] && echo "  Traffic flow:     enabled"
   [[ "$ENABLE_FLOW" != "yes" ]] && echo "  Traffic flow:     disabled"
   echo ""
   echo "  Generated files (next to your docker-compose.yml):"
   echo "    $OVERRIDE_FILE"
-  [[ "$MIGRATE_POSTGRES" == "yes" ]] && echo "    $ENTERPRISE_CONFIG_FILE"
+  [[ "$ENTERPRISE_CONFIG" == "yes" ]] && echo "    $ENTERPRISE_CONFIG_FILE"
   echo "    .env  (license key + secrets, mode 600)"
   [[ "$ENV_EXISTED" == "yes" ]] && [[ -f "$ENV_BACKUP" ]] && echo "    $ENV_BACKUP  (.env as it was before this run)"
   [[ "$MIGRATE_POSTGRES" == "yes" ]] && echo "    backups/sqlite-pre-enterprise-*/  (SQLite backup)"
@@ -899,7 +1149,11 @@ print_summary() {
   else
     echo "  $DOCKER_COMPOSE_COMMAND down"
   fi
-  echo "  rm -f $OVERRIDE_FILE $ENTERPRISE_CONFIG_FILE"
+  if [[ "$ENTERPRISE_CONFIG" == "yes" ]]; then
+    echo "  rm -f $OVERRIDE_FILE $ENTERPRISE_CONFIG_FILE"
+  else
+    echo "  rm -f $OVERRIDE_FILE"
+  fi
   if [[ "$ENV_EXISTED" == "yes" ]] && [[ -f "$ENV_BACKUP" ]]; then
     echo "  mv $ENV_BACKUP .env   # restores .env as it was before this run"
   elif [[ "$ENV_EXISTED" == "no" ]]; then