VERSION.txt: this is v1.38.3

Signed-off-by: Shayne Sweeney <shayne@tailscale.com>

VERSION.txt

@@ -1 +1 @@
-1.37.0
+1.38.3

cmd/tailscale/cli/funnel.go

@@ -29,7 +29,7 @@ var funnelCmd = newFunnelCommand(&serveEnv{lc: &localClient})
 func newFunnelCommand(e *serveEnv) *ffcli.Command {
 	return &ffcli.Command{
 		Name:      "funnel",
-		ShortHelp: "[ALPHA] turn Tailscale Funnel on or off",
+		ShortHelp: "[BETA] turn Tailscale Funnel on or off",
 		ShortUsage: strings.TrimSpace(`
 funnel <serve-port> {on|off}
   funnel status [--json]

cmd/tailscale/cli/network-lock.go

@@ -40,9 +40,17 @@ var netlockCmd = &ffcli.Command{
 		nlDisablementKDFCmd,
 		nlLogCmd,
 		nlLocalDisableCmd,
-		nlTskeyWrapCmd,
 	},
-	Exec: runNetworkLockStatus,
+	Exec: runNetworkLockNoSubcommand,
+}
+
+func runNetworkLockNoSubcommand(ctx context.Context, args []string) error {
+	// Detect & handle the deprecated command 'lock tskey-wrap'.
+	if len(args) >= 2 && args[0] == "tskey-wrap" {
+		return runTskeyWrapCmd(ctx, args[1:])
+	}
+
+	return runNetworkLockStatus(ctx, args)
 }
 
 var nlInitArgs struct {
@@ -427,13 +435,19 @@ func runNetworkLockModify(ctx context.Context, addArgs, removeArgs []string) err
 
 var nlSignCmd = &ffcli.Command{
 	Name:       "sign",
-	ShortUsage: "sign <node-key> [<rotation-key>]",
-	ShortHelp:  "Signs a node key and transmits the signature to the coordination server",
-	LongHelp:   "Signs a node key and transmits the signature to the coordination server",
-	Exec:       runNetworkLockSign,
+	ShortUsage: "sign <node-key> [<rotation-key>] or sign <auth-key>",
+	ShortHelp:  "Signs a node or pre-approved auth key",
+	LongHelp: `Either:
+  - signs a node key and transmits the signature to the coordination server, or
+  - signs a pre-approved auth key, printing it in a form that can be used to bring up nodes under tailnet lock`,
+	Exec: runNetworkLockSign,
 }
 
 func runNetworkLockSign(ctx context.Context, args []string) error {
+	if len(args) > 0 && strings.HasPrefix(args[0], "tskey-auth-") {
+		return runTskeyWrapCmd(ctx, args)
+	}
+
 	var (
 		nodeKey     key.NodePublic
 		rotationKey key.NLPublic
@@ -636,14 +650,6 @@ func runNetworkLockLog(ctx context.Context, args []string) error {
 	return nil
 }
 
-var nlTskeyWrapCmd = &ffcli.Command{
-	Name:       "tskey-wrap",
-	ShortUsage: "tskey-wrap <tailscale pre-auth key>",
-	ShortHelp:  "Modifies a pre-auth key from the admin panel to work with tailnet lock",
-	LongHelp:   "Modifies a pre-auth key from the admin panel to work with tailnet lock",
-	Exec:       runTskeyWrapCmd,
-}
-
 func runTskeyWrapCmd(ctx context.Context, args []string) error {
 	if len(args) != 1 {
 		return errors.New("usage: lock tskey-wrap <tailscale pre-auth key>")
@@ -657,21 +663,25 @@ func runTskeyWrapCmd(ctx context.Context, args []string) error {
 		return fixTailscaledConnectError(err)
 	}
 
+	return wrapAuthKey(ctx, args[0], st)
+}
+
+func wrapAuthKey(ctx context.Context, keyStr string, status *ipnstate.Status) error {
 	// Generate a separate tailnet-lock key just for the credential signature.
 	// We use the free-form meta strings to mark a little bit of metadata about this
 	// key.
 	priv := key.NewNLPrivate()
 	m := map[string]string{
 		"purpose":            "pre-auth key",
-		"wrapper_stableid":   string(st.Self.ID),
+		"wrapper_stableid":   string(status.Self.ID),
 		"wrapper_createtime": fmt.Sprint(time.Now().Unix()),
 	}
-	if strings.HasPrefix(args[0], "tskey-auth-") && strings.Index(args[0][len("tskey-auth-"):], "-") > 0 {
+	if strings.HasPrefix(keyStr, "tskey-auth-") && strings.Index(keyStr[len("tskey-auth-"):], "-") > 0 {
 		// We don't want to accidentally embed the nonce part of the authkey in
 		// the event the format changes. As such, we make sure its in the format we
 		// expect (tskey-auth-<stableID, inc CNTRL suffix>-nonce) before we parse
 		// out and embed the stableID.
-		s := strings.TrimPrefix(args[0], "tskey-auth-")
+		s := strings.TrimPrefix(keyStr, "tskey-auth-")
 		m["authkey_stableid"] = s[:strings.Index(s, "-")]
 	}
 	k := tka.Key{
@@ -681,7 +691,7 @@ func runTskeyWrapCmd(ctx context.Context, args []string) error {
 		Meta:   m,
 	}
 
-	wrapped, err := localClient.NetworkLockWrapPreauthKey(ctx, args[0], priv)
+	wrapped, err := localClient.NetworkLockWrapPreauthKey(ctx, keyStr, priv)
 	if err != nil {
 		return fmt.Errorf("wrapping failed: %w", err)
 	}

cmd/tailscale/cli/serve.go

@@ -33,7 +33,7 @@ var serveCmd = newServeCommand(&serveEnv{lc: &localClient})
 func newServeCommand(e *serveEnv) *ffcli.Command {
 	return &ffcli.Command{
 		Name:      "serve",
-		ShortHelp: "[ALPHA] Serve from your Tailscale node",
+		ShortHelp: "[BETA] Serve from your Tailscale node",
 		ShortUsage: strings.TrimSpace(`
 serve https:<port> <mount-point> <source> [off]
   serve tcp:<port> tcp://localhost:<local-port> [off]
@@ -41,7 +41,7 @@ serve https:<port> <mount-point> <source> [off]
   serve status [--json]
 `),
 		LongHelp: strings.TrimSpace(`
-*** ALPHA; all of this is subject to change ***
+*** BETA; all of this is subject to change ***
 
 The 'tailscale serve' set of commands allows you to serve
 content and local servers from your Tailscale node to

cmd/tailscaled/depaware.txt

@@ -212,7 +212,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/ipn/ipnstate                                   from tailscale.com/control/controlclient+
         tailscale.com/ipn/localapi                                   from tailscale.com/ipn/ipnserver
         tailscale.com/ipn/policy                                     from tailscale.com/ipn/ipnlocal
-        tailscale.com/ipn/store                                      from tailscale.com/cmd/tailscaled
+        tailscale.com/ipn/store                                      from tailscale.com/cmd/tailscaled+
    L    tailscale.com/ipn/store/awsstore                             from tailscale.com/ipn/store
    L    tailscale.com/ipn/store/kubestore                            from tailscale.com/ipn/store
         tailscale.com/ipn/store/mem                                  from tailscale.com/ipn/store+

ipn/ipnlocal/cert.go

@@ -31,10 +31,13 @@ import (
 	"time"
 
 	"golang.org/x/crypto/acme"
+	"tailscale.com/atomicfile"
 	"tailscale.com/envknob"
 	"tailscale.com/hostinfo"
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnstate"
+	"tailscale.com/ipn/store"
+	"tailscale.com/ipn/store/mem"
 	"tailscale.com/types/logger"
 	"tailscale.com/version"
 	"tailscale.com/version/distro"
@@ -82,11 +85,6 @@ func (b *LocalBackend) GetCertPEM(ctx context.Context, domain string) (*TLSCertK
 		return nil, errors.New("invalid domain")
 	}
 	logf := logger.WithPrefix(b.logf, fmt.Sprintf("cert(%q): ", domain))
-	dir, err := b.certDir()
-	if err != nil {
-		logf("failed to get certDir: %v", err)
-		return nil, err
-	}
 	now := time.Now()
 	traceACME := func(v any) {
 		if !acmeDebug() {
@@ -96,17 +94,22 @@ func (b *LocalBackend) GetCertPEM(ctx context.Context, domain string) (*TLSCertK
 		log.Printf("acme %T: %s", v, j)
 	}
 
-	if pair, err := b.getCertPEMCached(dir, domain, now); err == nil {
+	cs, err := b.getCertStore()
+	if err != nil {
+		return nil, err
+	}
+
+	if pair, err := getCertPEMCached(cs, domain, now); err == nil {
 		future := now.AddDate(0, 0, 14)
-		if b.shouldStartDomainRenewal(dir, domain, future) {
+		if b.shouldStartDomainRenewal(cs, domain, future) {
 			logf("starting async renewal")
 			// Start renewal in the background.
-			go b.getCertPEM(context.Background(), logf, traceACME, dir, domain, future)
+			go b.getCertPEM(context.Background(), cs, logf, traceACME, domain, future)
 		}
 		return pair, nil
 	}
 
-	pair, err := b.getCertPEM(ctx, logf, traceACME, dir, domain, now)
+	pair, err := b.getCertPEM(ctx, cs, logf, traceACME, domain, now)
 	if err != nil {
 		logf("getCertPEM: %v", err)
 		return nil, err
@@ -114,7 +117,7 @@ func (b *LocalBackend) GetCertPEM(ctx context.Context, domain string) (*TLSCertK
 	return pair, nil
 }
 
-func (b *LocalBackend) shouldStartDomainRenewal(dir, domain string, future time.Time) bool {
+func (b *LocalBackend) shouldStartDomainRenewal(cs certStore, domain string, future time.Time) bool {
 	renewMu.Lock()
 	defer renewMu.Unlock()
 	now := time.Now()
@@ -124,7 +127,7 @@ func (b *LocalBackend) shouldStartDomainRenewal(dir, domain string, future time.
 		return false
 	}
 	lastRenewCheck[domain] = now
-	_, err := b.getCertPEMCached(dir, domain, future)
+	_, err := getCertPEMCached(cs, domain, future)
 	return errors.Is(err, errCertExpired)
 }
 
@@ -140,15 +143,32 @@ type certStore interface {
 	WriteCert(domain string, cert []byte) error
 	// WriteKey writes the key for domain.
 	WriteKey(domain string, key []byte) error
+	// ACMEKey returns the value previously stored via WriteACMEKey.
+	// It is a PEM encoded ECDSA key.
+	ACMEKey() ([]byte, error)
+	// WriteACMEKey stores the provided PEM encoded ECDSA key.
+	WriteACMEKey([]byte) error
 }
 
 var errCertExpired = errors.New("cert expired")
 
-func (b *LocalBackend) getCertStore(dir string) certStore {
-	if hostinfo.GetEnvType() == hostinfo.Kubernetes && dir == "/tmp" {
-		return certStateStore{StateStore: b.store}
+func (b *LocalBackend) getCertStore() (certStore, error) {
+	switch b.store.(type) {
+	case *store.FileStore:
+	case *mem.Store:
+	default:
+		if hostinfo.GetEnvType() == hostinfo.Kubernetes {
+			// We're running in Kubernetes with a custom StateStore,
+			// use that instead of the cert directory.
+			// TODO(maisem): expand this to other environments?
+			return certStateStore{StateStore: b.store}, nil
+		}
 	}
-	return certFileStore{dir: dir}
+	dir, err := b.certDir()
+	if err != nil {
+		return nil, err
+	}
+	return certFileStore{dir: dir}, nil
 }
 
 // certFileStore implements certStore by storing the cert & key files in the named directory.
@@ -160,6 +180,25 @@ type certFileStore struct {
 	testRoots *x509.CertPool
 }
 
+const acmePEMName = "acme-account.key.pem"
+
+func (f certFileStore) ACMEKey() ([]byte, error) {
+	pemName := filepath.Join(f.dir, acmePEMName)
+	v, err := os.ReadFile(pemName)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, ipn.ErrStateNotExist
+		}
+		return nil, err
+	}
+	return v, nil
+}
+
+func (f certFileStore) WriteACMEKey(b []byte) error {
+	pemName := filepath.Join(f.dir, acmePEMName)
+	return atomicfile.WriteFile(pemName, b, 0600)
+}
+
 func (f certFileStore) Read(domain string, now time.Time) (*TLSCertKeyPair, error) {
 	certPEM, err := os.ReadFile(certFile(f.dir, domain))
 	if err != nil {
@@ -182,11 +221,11 @@ func (f certFileStore) Read(domain string, now time.Time) (*TLSCertKeyPair, erro
 }
 
 func (f certFileStore) WriteCert(domain string, cert []byte) error {
-	return os.WriteFile(certFile(f.dir, domain), cert, 0644)
+	return atomicfile.WriteFile(certFile(f.dir, domain), cert, 0644)
 }
 
 func (f certFileStore) WriteKey(domain string, key []byte) error {
-	return os.WriteFile(keyFile(f.dir, domain), key, 0600)
+	return atomicfile.WriteFile(keyFile(f.dir, domain), key, 0600)
 }
 
 // certStateStore implements certStore by storing the cert & key files in an ipn.StateStore.
@@ -221,6 +260,14 @@ func (s certStateStore) WriteKey(domain string, key []byte) error {
 	return s.WriteState(ipn.StateKey(domain+".key"), key)
 }
 
+func (s certStateStore) ACMEKey() ([]byte, error) {
+	return s.ReadState(ipn.StateKey(acmePEMName))
+}
+
+func (s certStateStore) WriteACMEKey(key []byte) error {
+	return s.WriteState(ipn.StateKey(acmePEMName), key)
+}
+
 // TLSCertKeyPair is a TLS public and private key, and whether they were obtained
 // from cache or freshly obtained.
 type TLSCertKeyPair struct {
@@ -236,26 +283,26 @@ func certFile(dir, domain string) string { return filepath.Join(dir, domain+".cr
 // domain exists on disk in dir that is valid at the provided now time.
 // If the keypair is expired, it returns errCertExpired.
 // If the keypair doesn't exist, it returns ipn.ErrStateNotExist.
-func (b *LocalBackend) getCertPEMCached(dir, domain string, now time.Time) (p *TLSCertKeyPair, err error) {
+func getCertPEMCached(cs certStore, domain string, now time.Time) (p *TLSCertKeyPair, err error) {
 	if !validLookingCertDomain(domain) {
 		// Before we read files from disk using it, validate it's halfway
 		// reasonable looking.
 		return nil, fmt.Errorf("invalid domain %q", domain)
 	}
-	return b.getCertStore(dir).Read(domain, now)
+	return cs.Read(domain, now)
 }
 
-func (b *LocalBackend) getCertPEM(ctx context.Context, logf logger.Logf, traceACME func(any), dir, domain string, now time.Time) (*TLSCertKeyPair, error) {
+func (b *LocalBackend) getCertPEM(ctx context.Context, cs certStore, logf logger.Logf, traceACME func(any), domain string, now time.Time) (*TLSCertKeyPair, error) {
 	acmeMu.Lock()
 	defer acmeMu.Unlock()
 
-	if p, err := b.getCertPEMCached(dir, domain, now); err == nil {
+	if p, err := getCertPEMCached(cs, domain, now); err == nil {
 		return p, nil
 	} else if !errors.Is(err, ipn.ErrStateNotExist) && !errors.Is(err, errCertExpired) {
 		return nil, err
 	}
 
-	key, err := acmeKey(dir)
+	key, err := acmeKey(cs)
 	if err != nil {
 		return nil, fmt.Errorf("acmeKey: %w", err)
 	}
@@ -366,8 +413,7 @@ func (b *LocalBackend) getCertPEM(ctx context.Context, logf logger.Logf, traceAC
 	if err := encodeECDSAKey(&privPEM, certPrivKey); err != nil {
 		return nil, err
 	}
-	certStore := b.getCertStore(dir)
-	if err := certStore.WriteKey(domain, privPEM.Bytes()); err != nil {
+	if err := cs.WriteKey(domain, privPEM.Bytes()); err != nil {
 		return nil, err
 	}
 
@@ -390,7 +436,7 @@ func (b *LocalBackend) getCertPEM(ctx context.Context, logf logger.Logf, traceAC
 			return nil, err
 		}
 	}
-	if err := certStore.WriteCert(domain, certPEM.Bytes()); err != nil {
+	if err := cs.WriteCert(domain, certPEM.Bytes()); err != nil {
 		return nil, err
 	}
 
@@ -444,14 +490,15 @@ func parsePrivateKey(der []byte) (crypto.Signer, error) {
 	return nil, errors.New("acme/autocert: failed to parse private key")
 }
 
-func acmeKey(dir string) (crypto.Signer, error) {
-	pemName := filepath.Join(dir, "acme-account.key.pem")
-	if v, err := os.ReadFile(pemName); err == nil {
+func acmeKey(cs certStore) (crypto.Signer, error) {
+	if v, err := cs.ACMEKey(); err == nil {
 		priv, _ := pem.Decode(v)
 		if priv == nil || !strings.Contains(priv.Type, "PRIVATE") {
 			return nil, errors.New("acme/autocert: invalid account key found in cache")
 		}
 		return parsePrivateKey(priv.Bytes)
+	} else if err != nil && !errors.Is(err, ipn.ErrStateNotExist) {
+		return nil, err
 	}
 
 	privKey, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
@@ -462,7 +509,7 @@ func acmeKey(dir string) (crypto.Signer, error) {
 	if err := encodeECDSAKey(&pemBuf, privKey); err != nil {
 		return nil, err
 	}
-	if err := os.WriteFile(pemName, pemBuf.Bytes(), 0600); err != nil {
+	if err := cs.WriteACMEKey(pemBuf.Bytes()); err != nil {
 		return nil, err
 	}
 	return privKey, nil

ipn/ipnlocal/local.go

@@ -298,7 +298,7 @@ func NewLocalBackend(logf logger.Logf, logid string, store ipn.StateStore, diale
 		statsLogf:      logger.LogOnChange(logf, 5*time.Minute, time.Now),
 		e:              e,
 		pm:             pm,
-		store:          pm.Store(),
+		store:          store,
 		dialer:         dialer,
 		backendLogID:   logid,
 		state:          ipn.NoState,

ipn/ipnlocal/serve.go

@@ -439,18 +439,26 @@ func (b *LocalBackend) proxyHandlerForBackend(backend string) (*httputil.Reverse
 	if err != nil {
 		return nil, fmt.Errorf("invalid url %s: %w", targetURL, err)
 	}
-	rp := httputil.NewSingleHostReverseProxy(u)
-	rp.Transport = &http.Transport{
-		DialContext: b.dialer.SystemDial,
-		TLSClientConfig: &tls.Config{
-			InsecureSkipVerify: insecure,
+	rp := &httputil.ReverseProxy{
+		Rewrite: func(r *httputil.ProxyRequest) {
+			r.SetURL(u)
+			r.Out.Host = r.In.Host
+			if c, ok := r.Out.Context().Value(serveHTTPContextKey{}).(*serveHTTPContext); ok {
+				r.Out.Header.Set("X-Forwarded-For", c.SrcAddr.Addr().String())
+			}
+		},
+		Transport: &http.Transport{
+			DialContext: b.dialer.SystemDial,
+			TLSClientConfig: &tls.Config{
+				InsecureSkipVerify: insecure,
+			},
+			// Values for the following parameters have been copied from http.DefaultTransport.
+			ForceAttemptHTTP2:     true,
+			MaxIdleConns:          100,
+			IdleConnTimeout:       90 * time.Second,
+			TLSHandshakeTimeout:   10 * time.Second,
+			ExpectContinueTimeout: 1 * time.Second,
 		},
-		// Values for the following parameters have been copied from http.DefaultTransport.
-		ForceAttemptHTTP2:     true,
-		MaxIdleConns:          100,
-		IdleConnTimeout:       90 * time.Second,
-		TLSHandshakeTimeout:   10 * time.Second,
-		ExpectContinueTimeout: 1 * time.Second,
 	}
 	return rp, nil
 }
@@ -476,7 +484,12 @@ func (b *LocalBackend) serveWebHandler(w http.ResponseWriter, r *http.Request) {
 			http.Error(w, "unknown proxy destination", http.StatusInternalServerError)
 			return
 		}
-		p.(http.Handler).ServeHTTP(w, r)
+		h := p.(http.Handler)
+		// Trim the mount point from the URL path before proxying. (#6571)
+		if r.URL.Path != "/" {
+			h = http.StripPrefix(strings.TrimSuffix(mountPoint, "/"), h)
+		}
+		h.ServeHTTP(w, r)
 		return
 	}
 

ipn/serve.go

@@ -180,7 +180,7 @@ func (sc *ServeConfig) IsFunnelOn() bool {
 // CheckFunnelAccess checks whether Funnel access is allowed for the given node
 // and port.
 // It checks:
-//  1. an invite was used to join the Funnel alpha
+//  1. Funnel is enabled on the Tailnet
 //  2. HTTPS is enabled on the Tailnet
 //  3. the node has the "funnel" nodeAttr
 //  4. the port is allowed for Funnel
@@ -190,7 +190,7 @@ func (sc *ServeConfig) IsFunnelOn() bool {
 // Funnel.
 func CheckFunnelAccess(port uint16, nodeAttrs []string) error {
 	if slices.Contains(nodeAttrs, tailcfg.CapabilityWarnFunnelNoInvite) {
-		return errors.New("Funnel not available; an invite is required to join the alpha. See https://tailscale.com/s/no-funnel.")
+		return errors.New("Funnel not enabled; See https://tailscale.com/s/no-funnel.")
 	}
 	if slices.Contains(nodeAttrs, tailcfg.CapabilityWarnFunnelNoHTTPS) {
 		return errors.New("Funnel not available; HTTPS must be enabled. See https://tailscale.com/s/https.")

net/interfaces/interfaces.go

@@ -153,11 +153,9 @@ func LocalAddresses() (regular, loopback []netip.Addr, err error) {
 	if len(regular4) == 0 && len(regular6) == 0 {
 		// if we have no usable IP addresses then be willing to accept
 		// addresses we otherwise wouldn't, like:
-		//   + 169.254.x.x (AWS Lambda uses NAT with these)
+		//   + 169.254.x.x (AWS Lambda and Azure App Services use NAT with these)
 		//   + IPv6 ULA (Google Cloud Run uses these with address translation)
-		if hostinfo.GetEnvType() == hostinfo.AWSLambda {
-			regular4 = linklocal4
-		}
+		regular4 = linklocal4
 		regular6 = ula6
 	}
 	regular = append(regular4, regular6...)
@@ -645,7 +643,14 @@ func isUsableV4(ip netip.Addr) bool {
 		return false
 	}
 	if ip.IsLinkLocalUnicast() {
-		return hostinfo.GetEnvType() == hostinfo.AWSLambda
+		switch hostinfo.GetEnvType() {
+		case hostinfo.AWSLambda:
+			return true
+		case hostinfo.AzureAppService:
+			return true
+		default:
+			return false
+		}
 	}
 	return true
 }

net/sockstats/sockstats_tsgo.go

@@ -25,7 +25,7 @@ type sockStatCounters struct {
 	txBytes, rxBytes                       atomic.Uint64
 	rxBytesByInterface, txBytesByInterface map[int]*atomic.Uint64
 
-	txBytesMetric, rxBytesMetric *clientmetric.Metric
+	txBytesMetric, rxBytesMetric, txBytesCellularMetric, rxBytesCellularMetric *clientmetric.Metric
 
 	// Validate counts for TCP sockets by using the TCP_CONNECTION_INFO
 	// getsockopt. We get current counts, as well as save final values when
@@ -65,10 +65,12 @@ func withSockStats(ctx context.Context, label Label) context.Context {
 	counters, ok := sockStats.countersByLabel[label]
 	if !ok {
 		counters = &sockStatCounters{
-			rxBytesByInterface: make(map[int]*atomic.Uint64),
-			txBytesByInterface: make(map[int]*atomic.Uint64),
-			txBytesMetric:      clientmetric.NewCounter(fmt.Sprintf("sockstats_tx_bytes_%s", label)),
-			rxBytesMetric:      clientmetric.NewCounter(fmt.Sprintf("sockstats_rx_bytes_%s", label)),
+			rxBytesByInterface:    make(map[int]*atomic.Uint64),
+			txBytesByInterface:    make(map[int]*atomic.Uint64),
+			txBytesMetric:         clientmetric.NewCounter(fmt.Sprintf("sockstats_tx_bytes_%s", label)),
+			rxBytesMetric:         clientmetric.NewCounter(fmt.Sprintf("sockstats_rx_bytes_%s", label)),
+			txBytesCellularMetric: clientmetric.NewCounter(fmt.Sprintf("sockstats_tx_bytes_cellular_%s", label)),
+			rxBytesCellularMetric: clientmetric.NewCounter(fmt.Sprintf("sockstats_rx_bytes_cellular_%s", label)),
 		}
 
 		// We might be called before setLinkMonitor has been called (and we've
@@ -119,6 +121,7 @@ func withSockStats(ctx context.Context, label Label) context.Context {
 		}
 		if sockStats.currentInterfaceCellular.Load() {
 			sockStats.rxBytesCellularMetric.Add(int64(n))
+			counters.rxBytesCellularMetric.Add(int64(n))
 		}
 	}
 	didWrite := func(n int) {
@@ -132,6 +135,7 @@ func withSockStats(ctx context.Context, label Label) context.Context {
 		}
 		if sockStats.currentInterfaceCellular.Load() {
 			sockStats.txBytesCellularMetric.Add(int64(n))
+			counters.txBytesCellularMetric.Add(int64(n))
 		}
 	}
 	willOverwrite := func(trace *net.SockTrace) {

ssh/tailssh/incubator.go

@@ -235,6 +235,7 @@ func beIncubator(args []string) error {
 	if err == nil && sessionCloser != nil {
 		defer sessionCloser()
 	}
+
 	var groupIDs []int
 	for _, g := range strings.Split(ia.groups, ",") {
 		gid, err := strconv.ParseInt(g, 10, 32)
@@ -244,22 +245,10 @@ func beIncubator(args []string) error {
 		groupIDs = append(groupIDs, int(gid))
 	}
 
-	if err := setGroups(groupIDs); err != nil {
+	if err := dropPrivileges(logf, int(ia.uid), ia.gid, groupIDs); err != nil {
 		return err
 	}
-	if egid := os.Getegid(); egid != ia.gid {
-		if err := syscall.Setgid(int(ia.gid)); err != nil {
-			logf(err.Error())
-			os.Exit(1)
-		}
-	}
-	if euid != ia.uid {
-		// Switch users if required before starting the desired process.
-		if err := syscall.Setuid(int(ia.uid)); err != nil {
-			logf(err.Error())
-			os.Exit(1)
-		}
-	}
+
 	if ia.isSFTP {
 		logf("handling sftp")
 
@@ -304,6 +293,108 @@ func beIncubator(args []string) error {
 	return err
 }
 
+// TODO(andrew-d): verify that this works in more configurations before
+// enabling by default.
+const assertDropPrivileges = false
+
+// dropPrivileges contains all the logic for dropping privileges to a different
+// UID, GID, and set of supplementary groups. This function is
+// security-sensitive and ordering-dependent; please be very cautious if/when
+// refactoring.
+//
+// WARNING: if you change this function, you *MUST* run the TestDropPrivileges
+// test in this package as root on at least Linux, FreeBSD and Darwin. This can
+// be done by running:
+//
+//	go test -c ./ssh/tailssh/ && sudo ./tailssh.test -test.v -test.run TestDropPrivileges
+func dropPrivileges(logf logger.Logf, wantUid, wantGid int, supplementaryGroups []int) error {
+	fatalf := func(format string, args ...any) {
+		logf(format, args...)
+		os.Exit(1)
+	}
+
+	euid := os.Geteuid()
+	egid := os.Getegid()
+
+	if runtime.GOOS == "darwin" || runtime.GOOS == "freebsd" {
+		// On FreeBSD and Darwin, the first entry returned from the
+		// getgroups(2) syscall is the egid, and changing it with
+		// setgroups(2) changes the egid of the process. This is
+		// technically a violation of the POSIX standard; see the
+		// following article for more detail:
+		//    https://www.usenix.org/system/files/login/articles/325-tsafrir.pdf
+		//
+		// In this case, we add an entry at the beginning of the
+		// groupIDs list containing the expected gid if it's not
+		// already there, which modifies the egid and additional groups
+		// as one unit.
+		if len(supplementaryGroups) == 0 || supplementaryGroups[0] != wantGid {
+			supplementaryGroups = append([]int{wantGid}, supplementaryGroups...)
+		}
+	}
+
+	if err := setGroups(supplementaryGroups); err != nil {
+		return err
+	}
+	if egid != wantGid {
+		// On FreeBSD and Darwin, we may have already called the
+		// equivalent of setegid(wantGid) via the call to setGroups,
+		// above. However, per the manpage, setgid(getegid()) is an
+		// allowed operation regardless of privilege level.
+		//
+		// FreeBSD:
+		//	The setgid() system call is permitted if the specified ID
+		//	is equal to the real group ID or the effective group ID
+		//	of the process, or if the effective user ID is that of
+		//	the super user.
+		//
+		// Darwin:
+		//	The setgid() function is permitted if the effective
+		//	user ID is that of the super user, or if the specified
+		//	group ID is the same as the effective group ID.  If
+		//	not, but the specified group ID is the same as the real
+		//	group ID, setgid() will set the effective group ID to
+		//	the real group ID.
+		if err := syscall.Setgid(wantGid); err != nil {
+			fatalf("Setgid(%d): %v", wantGid, err)
+		}
+	}
+	if euid != wantUid {
+		// Switch users if required before starting the desired process.
+		if err := syscall.Setuid(wantUid); err != nil {
+			fatalf("Setuid(%d): %v", wantUid, err)
+		}
+	}
+
+	// If we changed either the UID or GID, defensively assert that we
+	// cannot reset the it back to our original values, and that the
+	// current egid/euid are the expected values after we change
+	// everything; if not, we exit the process.
+	if assertDropPrivileges {
+		if egid != wantGid {
+			if err := syscall.Setegid(egid); err == nil {
+				fatalf("unexpectedly able to set egid back to %d", egid)
+			}
+		}
+		if euid != wantUid {
+			if err := syscall.Seteuid(euid); err == nil {
+				fatalf("unexpectedly able to set euid back to %d", euid)
+			}
+		}
+
+		if got := os.Getegid(); got != wantGid {
+			fatalf("got egid=%d, want %d", got, wantGid)
+		}
+		if got := os.Geteuid(); got != wantUid {
+			fatalf("got euid=%d, want %d", got, wantUid)
+		}
+
+		// TODO(andrew-d): assert that our supplementary groups are correct
+	}
+
+	return nil
+}
+
 // launchProcess launches an incubator process for the provided session.
 // It is responsible for configuring the process execution environment.
 // The caller can wait for the process to exit by calling cmd.Wait().

ssh/tailssh/privs_test.go

@@ -0,0 +1,295 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build linux || darwin || freebsd || openbsd || netbsd || dragonfly
+
+package tailssh
+
+import (
+	"encoding/json"
+	"errors"
+	"os"
+	"os/exec"
+	"os/user"
+	"path/filepath"
+	"reflect"
+	"regexp"
+	"runtime"
+	"strconv"
+	"syscall"
+	"testing"
+
+	"golang.org/x/exp/slices"
+	"tailscale.com/types/logger"
+)
+
+func TestDropPrivileges(t *testing.T) {
+	type SubprocInput struct {
+		UID              int
+		GID              int
+		AdditionalGroups []int
+	}
+	type SubprocOutput struct {
+		UID              int
+		GID              int
+		EUID             int
+		EGID             int
+		AdditionalGroups []int
+	}
+
+	if v := os.Getenv("TS_TEST_DROP_PRIVILEGES_CHILD"); v != "" {
+		t.Logf("in child process")
+
+		var input SubprocInput
+		if err := json.Unmarshal([]byte(v), &input); err != nil {
+			t.Fatal(err)
+		}
+
+		// Get a handle to our provided JSON file before dropping privs.
+		f := os.NewFile(3, "out.json")
+
+		// We're in our subprocess; actually drop privileges now.
+		dropPrivileges(t.Logf, input.UID, input.GID, input.AdditionalGroups)
+
+		additional, _ := syscall.Getgroups()
+
+		// Print our IDs
+		json.NewEncoder(f).Encode(SubprocOutput{
+			UID:              os.Getuid(),
+			GID:              os.Getgid(),
+			EUID:             os.Geteuid(),
+			EGID:             os.Getegid(),
+			AdditionalGroups: additional,
+		})
+
+		// Close output file to ensure that it's flushed to disk before we exit
+		f.Close()
+
+		// Always exit the process now that we have a different
+		// UID/GID/etc.; we don't want the Go test framework to try and
+		// clean anything up, since it might no longer have access.
+		os.Exit(0)
+	}
+
+	if os.Getuid() != 0 {
+		t.Skip("test only works when run as root")
+	}
+
+	rerunSelf := func(t *testing.T, input SubprocInput) []byte {
+		fpath := filepath.Join(t.TempDir(), "out.json")
+		outf, err := os.Create(fpath)
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		inputb, err := json.Marshal(input)
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		cmd := exec.Command(os.Args[0], "-test.v", "-test.run", "^"+regexp.QuoteMeta(t.Name())+"$")
+		cmd.Env = append(os.Environ(), "TS_TEST_DROP_PRIVILEGES_CHILD="+string(inputb))
+		cmd.ExtraFiles = []*os.File{outf}
+		cmd.Stdout = logger.FuncWriter(logger.WithPrefix(t.Logf, "child: "))
+		cmd.Stderr = logger.FuncWriter(logger.WithPrefix(t.Logf, "child: "))
+		if err := cmd.Run(); err != nil {
+			t.Fatal(err)
+		}
+		outf.Close()
+
+		jj, err := os.ReadFile(fpath)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return jj
+	}
+
+	// We want to ensure we're not colliding with existing users; find some
+	// unused UIDs and GIDs for the tests we run.
+	uid1 := findUnusedUID(t)
+	gid1 := findUnusedGID(t)
+	gid2 := findUnusedGID(t, gid1)
+	gid3 := findUnusedGID(t, gid1, gid2)
+
+	// For some tests, we want a UID/GID pair with the same numerical
+	// value; this finds one.
+	uidgid1 := findUnusedUIDGID(t, uid1, gid1, gid2, gid3)
+
+	t.Logf("uid1=%d gid1=%d gid2=%d gid3=%d uidgid1=%d",
+		uid1, gid1, gid2, gid3, uidgid1)
+
+	testCases := []struct {
+		name             string
+		uid              int
+		gid              int
+		additionalGroups []int
+	}{
+		{
+			name:             "all_different_values",
+			uid:              uid1,
+			gid:              gid1,
+			additionalGroups: []int{gid2, gid3},
+		},
+		{
+			name:             "no_additional_groups",
+			uid:              uid1,
+			gid:              gid1,
+			additionalGroups: []int{},
+		},
+		// This is a regression test for the following bug, triggered
+		// on Darwin & FreeBSD:
+		//    https://github.com/tailscale/tailscale/issues/7616
+		{
+			name:             "same_values",
+			uid:              uidgid1,
+			gid:              uidgid1,
+			additionalGroups: []int{uidgid1},
+		},
+	}
+
+	for _, tt := range testCases {
+		t.Run(tt.name, func(t *testing.T) {
+			subprocOut := rerunSelf(t, SubprocInput{
+				UID:              tt.uid,
+				GID:              tt.gid,
+				AdditionalGroups: tt.additionalGroups,
+			})
+
+			var out SubprocOutput
+			if err := json.Unmarshal(subprocOut, &out); err != nil {
+				t.Logf("%s", subprocOut)
+				t.Fatal(err)
+			}
+			t.Logf("output: %+v", out)
+
+			if out.UID != tt.uid {
+				t.Errorf("got uid %d; want %d", out.UID, tt.uid)
+			}
+			if out.GID != tt.gid {
+				t.Errorf("got gid %d; want %d", out.GID, tt.gid)
+			}
+			if out.EUID != tt.uid {
+				t.Errorf("got euid %d; want %d", out.EUID, tt.uid)
+			}
+			if out.EGID != tt.gid {
+				t.Errorf("got egid %d; want %d", out.EGID, tt.gid)
+			}
+
+			// On FreeBSD and Darwin, the set of additional groups
+			// is prefixed with the egid; handle that case by
+			// modifying our expected set.
+			wantGroups := make(map[int]bool)
+			for _, id := range tt.additionalGroups {
+				wantGroups[id] = true
+			}
+			if runtime.GOOS == "darwin" || runtime.GOOS == "freebsd" {
+				wantGroups[tt.gid] = true
+			}
+
+			gotGroups := make(map[int]bool)
+			for _, id := range out.AdditionalGroups {
+				gotGroups[id] = true
+			}
+
+			if !reflect.DeepEqual(gotGroups, wantGroups) {
+				t.Errorf("got additional groups %+v; want %+v", gotGroups, wantGroups)
+			}
+		})
+	}
+}
+
+func findUnusedUID(t *testing.T, not ...int) int {
+	for i := 1000; i < 65535; i++ {
+		// Skip UIDs that might be valid
+		if maybeValidUID(i) {
+			continue
+		}
+
+		// Skip UIDs that we're avoiding
+		if slices.Contains(not, i) {
+			continue
+		}
+
+		// Not a valid UID, not one we're avoiding... all good!
+		return i
+	}
+
+	t.Fatalf("unable to find an unused UID")
+	return -1
+}
+
+func findUnusedGID(t *testing.T, not ...int) int {
+	for i := 1000; i < 65535; i++ {
+		if maybeValidGID(i) {
+			continue
+		}
+
+		// Skip GIDs that we're avoiding
+		if slices.Contains(not, i) {
+			continue
+		}
+
+		// Not a valid GID, not one we're avoiding... all good!
+		return i
+	}
+
+	t.Fatalf("unable to find an unused GID")
+	return -1
+}
+
+func findUnusedUIDGID(t *testing.T, not ...int) int {
+	for i := 1000; i < 65535; i++ {
+		if maybeValidUID(i) || maybeValidGID(i) {
+			continue
+		}
+
+		// Skip IDs that we're avoiding
+		if slices.Contains(not, i) {
+			continue
+		}
+
+		// Not a valid ID, not one we're avoiding... all good!
+		return i
+	}
+
+	t.Fatalf("unable to find an unused UID/GID pair")
+	return -1
+}
+
+func maybeValidUID(id int) bool {
+	_, err := user.LookupId(strconv.Itoa(id))
+	if err == nil {
+		return true
+	}
+
+	var u1 user.UnknownUserIdError
+	if errors.As(err, &u1) {
+		return false
+	}
+	var u2 user.UnknownUserError
+	if errors.As(err, &u2) {
+		return false
+	}
+
+	// Some other error; might be valid
+	return true
+}
+
+func maybeValidGID(id int) bool {
+	_, err := user.LookupGroupId(strconv.Itoa(id))
+	if err == nil {
+		return true
+	}
+
+	var u1 user.UnknownGroupIdError
+	if errors.As(err, &u1) {
+		return false
+	}
+	var u2 user.UnknownGroupError
+	if errors.As(err, &u2) {
+		return false
+	}
+
+	// Some other error; might be valid
+	return true
+}

tailcfg/tailcfg.go

@@ -1822,7 +1822,8 @@ const (
 
 	// Funnel warning capabilities used for reporting errors to the user.
 
-	// CapabilityWarnFunnelNoInvite indicates an invite has not been accepted for the Funnel alpha.
+	// CapabilityWarnFunnelNoInvite indicates whether Funnel is enabled for the tailnet.
+	// NOTE: In transition from Alpha to Beta, this capability is being reused as the enablement.
 	CapabilityWarnFunnelNoInvite = "https://tailscale.com/cap/warn-funnel-no-invite"
 
 	// CapabilityWarnFunnelNoHTTPS indicates HTTPS has not been enabled for the tailnet.

version/mkversion/mkversion.go

@@ -150,14 +150,14 @@ func InfoFrom(dir string) (VersionInfo, error) {
 	}
 
 	// Note, this mechanism doesn't correctly support go.mod replacements,
-	// or go workdirs. We only parse out the commit hash from go.mod's
+	// or go workdirs. We only parse out the commit ref from go.mod's
 	// "require" line, nothing else.
-	tailscaleHash, err := tailscaleModuleHash(modBs)
+	tailscaleRef, err := tailscaleModuleRef(modBs)
 	if err != nil {
 		return VersionInfo{}, err
 	}
 
-	v, err := infoFromCache(tailscaleHash, runner)
+	v, err := infoFromCache(tailscaleRef, runner)
 	if err != nil {
 		return VersionInfo{}, err
 	}
@@ -171,9 +171,10 @@ func InfoFrom(dir string) (VersionInfo, error) {
 	return mkOutput(v)
 }
 
-// tailscaleModuleHash returns the git hash of the 'require tailscale.com' line
-// in the given go.mod bytes.
-func tailscaleModuleHash(modBs []byte) (string, error) {
+// tailscaleModuleRef returns the git ref of the 'require tailscale.com' line
+// in the given go.mod bytes. The ref is either a short commit hash, or a git
+// tag.
+func tailscaleModuleRef(modBs []byte) (string, error) {
 	mod, err := modfile.Parse("go.mod", modBs, nil)
 	if err != nil {
 		return "", err
@@ -187,7 +188,8 @@ func tailscaleModuleHash(modBs []byte) (string, error) {
 		if i := strings.LastIndexByte(req.Mod.Version, '-'); i != -1 {
 			return req.Mod.Version[i+1:], nil
 		}
-		return "", fmt.Errorf("couldn't parse git hash from tailscale.com version %q", req.Mod.Version)
+		// If there are no dashes, the version is a tag.
+		return req.Mod.Version, nil
 	}
 	return "", fmt.Errorf("no require tailscale.com line in go.mod")
 }
@@ -310,7 +312,7 @@ type verInfo struct {
 // sentinel patch number.
 const unknownPatchVersion = 9999999
 
-func infoFromCache(shortHash string, runner dirRunner) (verInfo, error) {
+func infoFromCache(ref string, runner dirRunner) (verInfo, error) {
 	cacheDir, err := os.UserCacheDir()
 	if err != nil {
 		return verInfo{}, fmt.Errorf("Getting user cache dir: %w", err)
@@ -324,16 +326,16 @@ func infoFromCache(shortHash string, runner dirRunner) (verInfo, error) {
 		}
 	}
 
-	if !r.ok("git", "cat-file", "-e", shortHash) {
+	if !r.ok("git", "cat-file", "-e", ref) {
 		if !r.ok("git", "fetch", "origin") {
 			return verInfo{}, fmt.Errorf("updating OSS repo failed")
 		}
 	}
-	hash, err := r.output("git", "rev-parse", shortHash)
+	hash, err := r.output("git", "rev-parse", ref)
 	if err != nil {
 		return verInfo{}, err
 	}
-	date, err := r.output("git", "log", "-n1", "--format=%ct", shortHash)
+	date, err := r.output("git", "log", "-n1", "--format=%ct", ref)
 	if err != nil {
 		return verInfo{}, err
 	}