ipn/ipnlocal: empty allowed exit nodes syspolicy should be treated as allow all

Updates tailscale/corp#19681

If the syspolicy returns an empty list of allowed exit nodes,
this should be treated as "allow all" rather than "allow none"

Signed-off-by: Jonathan Nobels <jonathan@tailscale.com>

Makefile

@@ -115,10 +115,7 @@ sshintegrationtest: ## Run the SSH integration tests in various Docker container
 	echo "Testing on ubuntu:focal" && docker build --build-arg="BASE=ubuntu:focal" -t ssh-ubuntu-focal ssh/tailssh/testcontainers && \
 	echo "Testing on ubuntu:jammy" && docker build --build-arg="BASE=ubuntu:jammy" -t ssh-ubuntu-jammy ssh/tailssh/testcontainers && \
 	echo "Testing on ubuntu:mantic" && docker build --build-arg="BASE=ubuntu:mantic" -t ssh-ubuntu-mantic ssh/tailssh/testcontainers && \
-	echo "Testing on ubuntu:noble" && docker build --build-arg="BASE=ubuntu:noble" -t ssh-ubuntu-noble ssh/tailssh/testcontainers && \
-	echo "Testing on fedora:38" && docker build --build-arg="BASE=dokken/fedora-38" -t ssh-fedora-38 ssh/tailssh/testcontainers && \
-	echo "Testing on fedora:39" && docker build --build-arg="BASE=dokken/fedora-39" -t ssh-fedora-39 ssh/tailssh/testcontainers && \
-	echo "Testing on fedora:40" && docker build --build-arg="BASE=dokken/fedora-40" -t ssh-fedora-40 ssh/tailssh/testcontainers
+	echo "Testing on ubuntu:noble" && docker build --build-arg="BASE=ubuntu:noble" -t ssh-ubuntu-noble ssh/tailssh/testcontainers
 
 help: ## Show this help
 	@echo "\nSpecify a command. The choices are:\n"

VERSION.txt

@@ -1 +1 @@
-1.65.0
+1.67.0

client/tailscale/localclient.go

@@ -778,6 +778,17 @@ func (lc *LocalClient) SetDNS(ctx context.Context, name, value string) error {
 //
 // The ctx is only used for the duration of the call, not the lifetime of the net.Conn.
 func (lc *LocalClient) DialTCP(ctx context.Context, host string, port uint16) (net.Conn, error) {
+	return lc.UserDial(ctx, "tcp", host, port)
+}
+
+// UserDial connects to the host's port via Tailscale for the given network.
+//
+// The host may be a base DNS name (resolved from the netmap inside tailscaled),
+// a FQDN, or an IP address.
+//
+// The ctx is only used for the duration of the call, not the lifetime of the
+// net.Conn.
+func (lc *LocalClient) UserDial(ctx context.Context, network, host string, port uint16) (net.Conn, error) {
 	connCh := make(chan net.Conn, 1)
 	trace := httptrace.ClientTrace{
 		GotConn: func(info httptrace.GotConnInfo) {
@@ -790,10 +801,11 @@ func (lc *LocalClient) DialTCP(ctx context.Context, host string, port uint16) (n
 		return nil, err
 	}
 	req.Header = http.Header{
-		"Upgrade":    []string{"ts-dial"},
-		"Connection": []string{"upgrade"},
-		"Dial-Host":  []string{host},
-		"Dial-Port":  []string{fmt.Sprint(port)},
+		"Upgrade":      []string{"ts-dial"},
+		"Connection":   []string{"upgrade"},
+		"Dial-Host":    []string{host},
+		"Dial-Port":    []string{fmt.Sprint(port)},
+		"Dial-Network": []string{network},
 	}
 	res, err := lc.DoLocalRequest(req)
 	if err != nil {

client/tailscale/localclient_test.go

@@ -35,6 +35,7 @@ func TestDeps(t *testing.T) {
 		BadDeps: map[string]string{
 			// Make sure we don't again accidentally bring in a dependency on
 			// drive or its transitive dependencies
+			"testing":                        "do not use testing package in production code",
 			"tailscale.com/drive/driveimpl":  "https://github.com/tailscale/tailscale/pull/10631",
 			"github.com/studio-b12/gowebdav": "https://github.com/tailscale/tailscale/pull/10631",
 		},

client/web/web.go

@@ -1150,7 +1150,15 @@ func (s *Server) tailscaleUp(ctx context.Context, st *ipnstate.Status, opt tails
 		if !isRunning {
 			ipnOptions := ipn.Options{AuthKey: opt.AuthKey}
 			if opt.ControlURL != "" {
-				ipnOptions.UpdatePrefs = &ipn.Prefs{ControlURL: opt.ControlURL}
+				_, err := s.lc.EditPrefs(ctx, &ipn.MaskedPrefs{
+					Prefs: ipn.Prefs{
+						ControlURL: opt.ControlURL,
+					},
+					ControlURLSet: true,
+				})
+				if err != nil {
+					s.logf("edit prefs: %v", err)
+				}
 			}
 			if err := s.lc.Start(ctx, ipnOptions); err != nil {
 				s.logf("start: %v", err)

clientupdate/clientupdate.go

@@ -653,6 +653,9 @@ func (up *Updater) updateAlpineLike() (err error) {
 		return fmt.Errorf(`failed to parse latest version from "apk info tailscale": %w`, err)
 	}
 	if !up.confirm(ver) {
+		if err := checkOutdatedAlpineRepo(up.Logf, ver, up.Track); err != nil {
+			up.Logf("failed to check whether Alpine release is outdated: %v", err)
+		}
 		return nil
 	}
 
@@ -690,6 +693,37 @@ func parseAlpinePackageVersion(out []byte) (string, error) {
 	return "", errors.New("tailscale version not found in output")
 }
 
+var apkRepoVersionRE = regexp.MustCompile(`v[0-9]+\.[0-9]+`)
+
+func checkOutdatedAlpineRepo(logf logger.Logf, apkVer, track string) error {
+	latest, err := LatestTailscaleVersion(track)
+	if err != nil {
+		return err
+	}
+	if latest == apkVer {
+		// Actually on latest release.
+		return nil
+	}
+	f, err := os.Open("/etc/apk/repositories")
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+	// Read the first repo line. Typically, there are multiple repos that all
+	// contain the same version in the path, like:
+	//   https://dl-cdn.alpinelinux.org/alpine/v3.20/main
+	//   https://dl-cdn.alpinelinux.org/alpine/v3.20/community
+	s := bufio.NewScanner(f)
+	if !s.Scan() {
+		return s.Err()
+	}
+	alpineVer := apkRepoVersionRE.FindString(s.Text())
+	if alpineVer != "" {
+		logf("The latest Tailscale release for Linux is %q, but your apk repository only provides %q.\nYour Alpine version is %q, you may need to upgrade the system to get the latest Tailscale version: https://wiki.alpinelinux.org/wiki/Upgrading_Alpine", latest, apkVer, alpineVer)
+	}
+	return nil
+}
+
 func (up *Updater) updateMacSys() error {
 	return errors.New("NOTREACHED: On MacSys builds, `tailscale update` is handled in Swift to launch the GUI updater")
 }

cmd/containerboot/kube.go

@@ -138,9 +138,9 @@ func initKubeClient(root string) {
 	if err != nil {
 		log.Fatalf("Error creating kube client: %v", err)
 	}
-	if root != "/" {
-		// If we are running in a test, we need to set the URL to the
-		// httptest server.
+	if (root != "/") || os.Getenv("TS_KUBERNETES_READ_API_SERVER_ADDRESS_FROM_ENV") == "true" {
+		// Derive the API server address from the environment variables
+		// Used to set http server in tests, or optionally enabled by flag
 		kc.SetURL(fmt.Sprintf("https://%s:%s", os.Getenv("KUBERNETES_SERVICE_HOST"), os.Getenv("KUBERNETES_SERVICE_PORT_HTTPS")))
 	}
 }

cmd/containerboot/main.go

@@ -52,8 +52,10 @@
 //     ${TS_CERT_DOMAIN}, it will be replaced with the value of the available FQDN.
 //     It cannot be used in conjunction with TS_DEST_IP. The file is watched for changes,
 //     and will be re-applied when it changes.
-//   - EXPERIMENTAL_TS_CONFIGFILE_PATH: if specified, a path to tailscaled
-//     config. If this is set, TS_HOSTNAME, TS_EXTRA_ARGS, TS_AUTHKEY,
+//   - TS_EXPERIMENTAL_VERSIONED_CONFIG_DIR: if specified, a path to a
+//     directory that containers tailscaled config in file. The config file needs to be
+//     named cap-<current-tailscaled-cap>.hujson. If this is set, TS_HOSTNAME,
+//     TS_EXTRA_ARGS, TS_AUTHKEY,
 //     TS_ROUTES, TS_ACCEPT_DNS env vars must not be set. If this is set,
 //     containerboot only runs `tailscaled --config <path-to-this-configfile>`
 //     and not `tailscale up` or `tailscale set`.
@@ -92,6 +94,7 @@ import (
 	"os"
 	"os/exec"
 	"os/signal"
+	"path"
 	"path/filepath"
 	"reflect"
 	"slices"
@@ -107,6 +110,7 @@ import (
 	"tailscale.com/client/tailscale"
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/conffile"
+	kubeutils "tailscale.com/k8s-operator"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/logger"
 	"tailscale.com/types/ptr"
@@ -145,7 +149,7 @@ func main() {
 		Socket:                                defaultEnv("TS_SOCKET", "/tmp/tailscaled.sock"),
 		AuthOnce:                              defaultBool("TS_AUTH_ONCE", false),
 		Root:                                  defaultEnv("TS_TEST_ONLY_ROOT", "/"),
-		TailscaledConfigFilePath:              defaultEnv("EXPERIMENTAL_TS_CONFIGFILE_PATH", ""),
+		TailscaledConfigFilePath:              tailscaledConfigFilePath(),
 		AllowProxyingClusterTrafficViaIngress: defaultBool("EXPERIMENTAL_ALLOW_PROXYING_CLUSTER_TRAFFIC_VIA_INGRESS", false),
 		PodIP:                                 defaultEnv("POD_IP", ""),
 	}
@@ -957,16 +961,23 @@ func installIngressForwardingRule(ctx context.Context, dstStr string, tsIPs []ne
 		return err
 	}
 	var local netip.Addr
+	proxyHasIPv4Address := false
 	for _, pfx := range tsIPs {
 		if !pfx.IsSingleIP() {
 			continue
 		}
+		if pfx.Addr().Is4() {
+			proxyHasIPv4Address = true
+		}
 		if pfx.Addr().Is4() != dst.Is4() {
 			continue
 		}
 		local = pfx.Addr()
 		break
 	}
+	if proxyHasIPv4Address && dst.Is6() {
+		log.Printf("Warning: proxy backend ClusterIP is an IPv6 address and the proxy has a IPv4 tailnet address. You might need to disable IPv4 address allocation for the proxy for forwarding to work. See https://github.com/tailscale/tailscale/issues/12156")
+	}
 	if !local.IsValid() {
 		return fmt.Errorf("no tailscale IP matching family of %s found in %v", dstStr, tsIPs)
 	}
@@ -1097,6 +1108,13 @@ type settings struct {
 
 func (s *settings) validate() error {
 	if s.TailscaledConfigFilePath != "" {
+		dir, file := path.Split(s.TailscaledConfigFilePath)
+		if _, err := os.Stat(dir); err != nil {
+			return fmt.Errorf("error validating whether directory with tailscaled config file %s exists: %w", dir, err)
+		}
+		if _, err := os.Stat(s.TailscaledConfigFilePath); err != nil {
+			return fmt.Errorf("error validating whether tailscaled config directory %q contains tailscaled config for current capability version %q: %w. If this is a Tailscale Kubernetes operator proxy, please ensure that the version of the operator is not older than the version of the proxy", dir, file, err)
+		}
 		if _, err := conffile.Load(s.TailscaledConfigFilePath); err != nil {
 			return fmt.Errorf("error validating tailscaled configfile contents: %w", err)
 		}
@@ -1120,7 +1138,7 @@ func (s *settings) validate() error {
 		return errors.New("Both TS_TAILNET_TARGET_IP and TS_TAILNET_FQDN cannot be set")
 	}
 	if s.TailscaledConfigFilePath != "" && (s.AcceptDNS != nil || s.AuthKey != "" || s.Routes != nil || s.ExtraArgs != "" || s.Hostname != "") {
-		return errors.New("EXPERIMENTAL_TS_CONFIGFILE_PATH cannot be set in combination with TS_HOSTNAME, TS_EXTRA_ARGS, TS_AUTHKEY, TS_ROUTES, TS_ACCEPT_DNS.")
+		return errors.New("TS_EXPERIMENTAL_VERSIONED_CONFIG_DIR cannot be set in combination with TS_HOSTNAME, TS_EXTRA_ARGS, TS_AUTHKEY, TS_ROUTES, TS_ACCEPT_DNS.")
 	}
 	if s.AllowProxyingClusterTrafficViaIngress && s.UserspaceMode {
 		return errors.New("EXPERIMENTAL_ALLOW_PROXYING_CLUSTER_TRAFFIC_VIA_INGRESS is not supported in userspace mode")
@@ -1252,3 +1270,42 @@ func isTwoStepConfigAlwaysAuth(cfg *settings) bool {
 func isOneStepConfig(cfg *settings) bool {
 	return cfg.TailscaledConfigFilePath != ""
 }
+
+// tailscaledConfigFilePath returns the path to the tailscaled config file that
+// should be used for the current capability version. It is determined by the
+// TS_EXPERIMENTAL_VERSIONED_CONFIG_DIR environment variable and looks for a
+// file named cap-<capability_version>.hujson in the directory. It searches for
+// the highest capability version that is less than or equal to the current
+// capability version.
+func tailscaledConfigFilePath() string {
+	dir := os.Getenv("TS_EXPERIMENTAL_VERSIONED_CONFIG_DIR")
+	if dir == "" {
+		return ""
+	}
+	fe, err := os.ReadDir(dir)
+	if err != nil {
+		log.Fatalf("error reading tailscaled config directory %q: %v", dir, err)
+	}
+	maxCompatVer := tailcfg.CapabilityVersion(-1)
+	for _, e := range fe {
+		// We don't check if type if file as in most cases this will
+		// come from a mounted kube Secret, where the directory contents
+		// will be various symlinks.
+		if e.Type().IsDir() {
+			continue
+		}
+		cv, err := kubeutils.CapVerFromFileName(e.Name())
+		if err != nil {
+			log.Printf("skipping file %q in tailscaled config directory %q: %v", e.Name(), dir, err)
+			continue
+		}
+		if cv > maxCompatVer && cv <= tailcfg.CurrentCapabilityVersion {
+			maxCompatVer = cv
+		}
+	}
+	if maxCompatVer == -1 {
+		log.Fatalf("no tailscaled config file found in %q for current capability version %q", dir, tailcfg.CurrentCapabilityVersion)
+	}
+	log.Printf("Using tailscaled config file %q for capability version %q", maxCompatVer, tailcfg.CurrentCapabilityVersion)
+	return path.Join(dir, kubeutils.TailscaledConfigFileNameForCap(maxCompatVer))
+}

cmd/containerboot/main_test.go

@@ -65,7 +65,7 @@ func TestContainerBoot(t *testing.T) {
 		"dev/net",
 		"proc/sys/net/ipv4",
 		"proc/sys/net/ipv6/conf/all",
-		"etc",
+		"etc/tailscaled",
 	}
 	for _, path := range dirs {
 		if err := os.MkdirAll(filepath.Join(d, path), 0700); err != nil {
@@ -80,7 +80,7 @@ func TestContainerBoot(t *testing.T) {
 		"dev/net/tun":                           []byte(""),
 		"proc/sys/net/ipv4/ip_forward":          []byte("0"),
 		"proc/sys/net/ipv6/conf/all/forwarding": []byte("0"),
-		"etc/tailscaled":                        tailscaledConfBytes,
+		"etc/tailscaled/cap-95.hujson":          tailscaledConfBytes,
 	}
 	resetFiles := func() {
 		for path, content := range files {
@@ -638,14 +638,14 @@ func TestContainerBoot(t *testing.T) {
 			},
 		},
 		{
-			Name: "experimental tailscaled configfile",
+			Name: "experimental tailscaled config path",
 			Env: map[string]string{
-				"EXPERIMENTAL_TS_CONFIGFILE_PATH": filepath.Join(d, "etc/tailscaled"),
+				"TS_EXPERIMENTAL_VERSIONED_CONFIG_DIR": filepath.Join(d, "etc/tailscaled/"),
 			},
 			Phases: []phase{
 				{
 					WantCmds: []string{
-						"/usr/bin/tailscaled --socket=/tmp/tailscaled.sock --state=mem: --statedir=/tmp --tun=userspace-networking --config=/etc/tailscaled",
+						"/usr/bin/tailscaled --socket=/tmp/tailscaled.sock --state=mem: --statedir=/tmp --tun=userspace-networking --config=/etc/tailscaled/cap-95.hujson",
 					},
 				}, {
 					Notify: runningNotify,

cmd/derper/bootstrap_dns.go

@@ -5,35 +5,45 @@ package main
 
 import (
 	"context"
+	"encoding/binary"
 	"encoding/json"
 	"expvar"
 	"log"
+	"math/rand/v2"
 	"net"
 	"net/http"
+	"net/netip"
+	"strconv"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"tailscale.com/syncs"
+	"tailscale.com/util/mak"
 	"tailscale.com/util/slicesx"
 )
 
 const refreshTimeout = time.Minute
 
-type dnsEntryMap map[string][]net.IP
+type dnsEntryMap struct {
+	IPs     map[string][]net.IP
+	Percent map[string]float64 // "foo.com" => 0.5 for 50%
+}
 
 var (
-	dnsCache            syncs.AtomicValue[dnsEntryMap]
+	dnsCache            atomic.Pointer[dnsEntryMap]
 	dnsCacheBytes       syncs.AtomicValue[[]byte] // of JSON
-	unpublishedDNSCache syncs.AtomicValue[dnsEntryMap]
+	unpublishedDNSCache atomic.Pointer[dnsEntryMap]
 	bootstrapLookupMap  syncs.Map[string, bool]
 )
 
 var (
-	bootstrapDNSRequests = expvar.NewInt("counter_bootstrap_dns_requests")
-	publishedDNSHits     = expvar.NewInt("counter_bootstrap_dns_published_hits")
-	publishedDNSMisses   = expvar.NewInt("counter_bootstrap_dns_published_misses")
-	unpublishedDNSHits   = expvar.NewInt("counter_bootstrap_dns_unpublished_hits")
-	unpublishedDNSMisses = expvar.NewInt("counter_bootstrap_dns_unpublished_misses")
+	bootstrapDNSRequests        = expvar.NewInt("counter_bootstrap_dns_requests")
+	publishedDNSHits            = expvar.NewInt("counter_bootstrap_dns_published_hits")
+	publishedDNSMisses          = expvar.NewInt("counter_bootstrap_dns_published_misses")
+	unpublishedDNSHits          = expvar.NewInt("counter_bootstrap_dns_unpublished_hits")
+	unpublishedDNSMisses        = expvar.NewInt("counter_bootstrap_dns_unpublished_misses")
+	unpublishedDNSPercentMisses = expvar.NewInt("counter_bootstrap_dns_unpublished_percent_misses")
 )
 
 func init() {
@@ -59,15 +69,13 @@ func refreshBootstrapDNS() {
 	}
 	ctx, cancel := context.WithTimeout(context.Background(), refreshTimeout)
 	defer cancel()
-	dnsEntries := resolveList(ctx, strings.Split(*bootstrapDNS, ","))
+	dnsEntries := resolveList(ctx, *bootstrapDNS)
 	// Randomize the order of the IPs for each name to avoid the client biasing
 	// to IPv6
-	for k := range dnsEntries {
-		ips := dnsEntries[k]
-		slicesx.Shuffle(ips)
-		dnsEntries[k] = ips
+	for _, vv := range dnsEntries.IPs {
+		slicesx.Shuffle(vv)
 	}
-	j, err := json.MarshalIndent(dnsEntries, "", "\t")
+	j, err := json.MarshalIndent(dnsEntries.IPs, "", "\t")
 	if err != nil {
 		// leave the old values in place
 		return
@@ -81,27 +89,50 @@ func refreshUnpublishedDNS() {
 	if *unpublishedDNS == "" {
 		return
 	}
-
 	ctx, cancel := context.WithTimeout(context.Background(), refreshTimeout)
 	defer cancel()
-
-	dnsEntries := resolveList(ctx, strings.Split(*unpublishedDNS, ","))
+	dnsEntries := resolveList(ctx, *unpublishedDNS)
 	unpublishedDNSCache.Store(dnsEntries)
 }
 
-func resolveList(ctx context.Context, names []string) dnsEntryMap {
-	dnsEntries := make(dnsEntryMap)
+// resolveList takes a comma-separated list of DNS names to resolve.
+//
+// If an entry contains a slash, it's two DNS names: the first is the one to
+// resolve and the second is that of a TXT recording containing the rollout
+// percentage in range "0".."100". If the TXT record doesn't exist or is
+// malformed, the percentage is 0. If the TXT record is not provided (there's no
+// slash), then the percentage is 100.
+func resolveList(ctx context.Context, list string) *dnsEntryMap {
+	ents := strings.Split(list, ",")
+
+	ret := &dnsEntryMap{}
 
 	var r net.Resolver
-	for _, name := range names {
+	for _, ent := range ents {
+		name, txtName, _ := strings.Cut(ent, "/")
 		addrs, err := r.LookupIP(ctx, "ip", name)
 		if err != nil {
 			log.Printf("bootstrap DNS lookup %q: %v", name, err)
 			continue
 		}
-		dnsEntries[name] = addrs
+		mak.Set(&ret.IPs, name, addrs)
+
+		if txtName == "" {
+			mak.Set(&ret.Percent, name, 1.0)
+			continue
+		}
+		vals, err := r.LookupTXT(ctx, txtName)
+		if err != nil {
+			log.Printf("bootstrap DNS lookup %q: %v", txtName, err)
+			continue
+		}
+		for _, v := range vals {
+			if v, err := strconv.Atoi(v); err == nil && v >= 0 && v <= 100 {
+				mak.Set(&ret.Percent, name, float64(v)/100)
+			}
+		}
 	}
-	return dnsEntries
+	return ret
 }
 
 func handleBootstrapDNS(w http.ResponseWriter, r *http.Request) {
@@ -115,22 +146,36 @@ func handleBootstrapDNS(w http.ResponseWriter, r *http.Request) {
 	// Try answering a query from our hidden map first
 	if q := r.URL.Query().Get("q"); q != "" {
 		bootstrapLookupMap.Store(q, true)
-		if ips, ok := unpublishedDNSCache.Load()[q]; ok && len(ips) > 0 {
+		if bootstrapLookupMap.Len() > 500 { // defensive
+			bootstrapLookupMap.Clear()
+		}
+		if m := unpublishedDNSCache.Load(); m != nil && len(m.IPs[q]) > 0 {
 			unpublishedDNSHits.Add(1)
 
-			// Only return the specific query, not everything.
-			m := dnsEntryMap{q: ips}
-			j, err := json.MarshalIndent(m, "", "\t")
-			if err == nil {
-				w.Write(j)
-				return
+			percent := m.Percent[q]
+			if remoteAddrMatchesPercent(r.RemoteAddr, percent) {
+				// Only return the specific query, not everything.
+				m := map[string][]net.IP{q: m.IPs[q]}
+				j, err := json.MarshalIndent(m, "", "\t")
+				if err == nil {
+					w.Write(j)
+					return
+				}
+			} else {
+				unpublishedDNSPercentMisses.Add(1)
 			}
 		}
 
 		// If we have a "q" query for a name in the published cache
 		// list, then track whether that's a hit/miss.
-		if m, ok := dnsCache.Load()[q]; ok {
-			if len(m) > 0 {
+		m := dnsCache.Load()
+		var inPub bool
+		var ips []net.IP
+		if m != nil {
+			ips, inPub = m.IPs[q]
+		}
+		if inPub {
+			if len(ips) > 0 {
 				publishedDNSHits.Add(1)
 			} else {
 				publishedDNSMisses.Add(1)
@@ -146,3 +191,29 @@ func handleBootstrapDNS(w http.ResponseWriter, r *http.Request) {
 	j := dnsCacheBytes.Load()
 	w.Write(j)
 }
+
+// percent is [0.0, 1.0].
+func remoteAddrMatchesPercent(remoteAddr string, percent float64) bool {
+	if percent == 0 {
+		return false
+	}
+	if percent == 1 {
+		return true
+	}
+	reqIPStr, _, err := net.SplitHostPort(remoteAddr)
+	if err != nil {
+		return false
+	}
+	reqIP, err := netip.ParseAddr(reqIPStr)
+	if err != nil {
+		return false
+	}
+	if reqIP.IsLoopback() {
+		// For local testing.
+		return rand.Float64() < 0.5
+	}
+	reqIP16 := reqIP.As16()
+	rndSrc := rand.NewPCG(binary.LittleEndian.Uint64(reqIP16[:8]), binary.LittleEndian.Uint64(reqIP16[8:]))
+	rnd := rand.New(rndSrc)
+	return percent > rnd.Float64()
+}

cmd/derper/bootstrap_dns_test.go

@@ -4,10 +4,13 @@
 package main
 
 import (
+	"bytes"
 	"encoding/json"
+	"io"
 	"net"
 	"net/http"
 	"net/http/httptest"
+	"net/netip"
 	"net/url"
 	"reflect"
 	"testing"
@@ -38,7 +41,7 @@ func (b *bitbucketResponseWriter) Write(p []byte) (int, error) { return len(p),
 
 func (b *bitbucketResponseWriter) WriteHeader(statusCode int) {}
 
-func getBootstrapDNS(t *testing.T, q string) dnsEntryMap {
+func getBootstrapDNS(t *testing.T, q string) map[string][]net.IP {
 	t.Helper()
 	req, _ := http.NewRequest("GET", "https://localhost/bootstrap-dns?q="+url.QueryEscape(q), nil)
 	w := httptest.NewRecorder()
@@ -48,11 +51,12 @@ func getBootstrapDNS(t *testing.T, q string) dnsEntryMap {
 	if res.StatusCode != 200 {
 		t.Fatalf("got status=%d; want %d", res.StatusCode, 200)
 	}
-	var ips dnsEntryMap
-	if err := json.NewDecoder(res.Body).Decode(&ips); err != nil {
-		t.Fatalf("error decoding response body: %v", err)
+	var m map[string][]net.IP
+	var buf bytes.Buffer
+	if err := json.NewDecoder(io.TeeReader(res.Body, &buf)).Decode(&m); err != nil {
+		t.Fatalf("error decoding response body %q: %v", buf.Bytes(), err)
 	}
-	return ips
+	return m
 }
 
 func TestUnpublishedDNS(t *testing.T) {
@@ -107,15 +111,21 @@ func resetMetrics() {
 // Verify that we don't count an empty list in the unpublishedDNSCache as a
 // cache hit in our metrics.
 func TestUnpublishedDNSEmptyList(t *testing.T) {
-	pub := dnsEntryMap{
-		"tailscale.com": {net.IPv4(10, 10, 10, 10)},
+	pub := &dnsEntryMap{
+		IPs: map[string][]net.IP{"tailscale.com": {net.IPv4(10, 10, 10, 10)}},
 	}
 	dnsCache.Store(pub)
 	dnsCacheBytes.Store([]byte(`{"tailscale.com":["10.10.10.10"]}`))
 
-	unpublishedDNSCache.Store(dnsEntryMap{
-		"log.tailscale.io":           {},
-		"controlplane.tailscale.com": {net.IPv4(1, 2, 3, 4)},
+	unpublishedDNSCache.Store(&dnsEntryMap{
+		IPs: map[string][]net.IP{
+			"log.tailscale.io":           {},
+			"controlplane.tailscale.com": {net.IPv4(1, 2, 3, 4)},
+		},
+		Percent: map[string]float64{
+			"log.tailscale.io":           1.0,
+			"controlplane.tailscale.com": 1.0,
+		},
 	})
 
 	t.Run("CacheMiss", func(t *testing.T) {
@@ -125,8 +135,8 @@ func TestUnpublishedDNSEmptyList(t *testing.T) {
 			ips := getBootstrapDNS(t, q)
 
 			// Expected our public map to be returned on a cache miss
-			if !reflect.DeepEqual(ips, pub) {
-				t.Errorf("got ips=%+v; want %+v", ips, pub)
+			if !reflect.DeepEqual(ips, pub.IPs) {
+				t.Errorf("got ips=%+v; want %+v", ips, pub.IPs)
 			}
 			if v := unpublishedDNSHits.Value(); v != 0 {
 				t.Errorf("got hits=%d; want 0", v)
@@ -141,7 +151,7 @@ func TestUnpublishedDNSEmptyList(t *testing.T) {
 	t.Run("CacheHit", func(t *testing.T) {
 		resetMetrics()
 		ips := getBootstrapDNS(t, "controlplane.tailscale.com")
-		want := dnsEntryMap{"controlplane.tailscale.com": {net.IPv4(1, 2, 3, 4)}}
+		want := map[string][]net.IP{"controlplane.tailscale.com": {net.IPv4(1, 2, 3, 4)}}
 		if !reflect.DeepEqual(ips, want) {
 			t.Errorf("got ips=%+v; want %+v", ips, want)
 		}
@@ -166,3 +176,54 @@ func TestLookupMetric(t *testing.T) {
 		t.Errorf("bootstrapLookupMap.Len() want=5, got %v", bootstrapLookupMap.Len())
 	}
 }
+
+func TestRemoteAddrMatchesPercent(t *testing.T) {
+	tests := []struct {
+		remoteAddr string
+		percent    float64
+		want       bool
+	}{
+		// 0% and 100%.
+		{"10.0.0.1:1234", 0.0, false},
+		{"10.0.0.1:1234", 1.0, true},
+
+		// Invalid IP.
+		{"", 1.0, true},
+		{"", 0.0, false},
+		{"", 0.5, false},
+
+		// Small manual sample at 50%. The func uses a deterministic PRNG seed.
+		{"1.2.3.4:567", 0.5, true},
+		{"1.2.3.5:567", 0.5, true},
+		{"1.2.3.6:567", 0.5, false},
+		{"1.2.3.7:567", 0.5, true},
+		{"1.2.3.8:567", 0.5, false},
+		{"1.2.3.9:567", 0.5, true},
+		{"1.2.3.10:567", 0.5, true},
+	}
+	for _, tt := range tests {
+		got := remoteAddrMatchesPercent(tt.remoteAddr, tt.percent)
+		if got != tt.want {
+			t.Errorf("remoteAddrMatchesPercent(%q, %v) = %v; want %v", tt.remoteAddr, tt.percent, got, tt.want)
+		}
+	}
+
+	var match, all int
+	const wantPercent = 0.5
+	for a := range 256 {
+		for b := range 256 {
+			all++
+			if remoteAddrMatchesPercent(
+				netip.AddrPortFrom(netip.AddrFrom4([4]byte{1, 2, byte(a), byte(b)}), 12345).String(),
+				wantPercent) {
+				match++
+			}
+		}
+	}
+	gotPercent := float64(match) / float64(all)
+	const tolerance = 0.005
+	t.Logf("got percent %v (goal %v)", gotPercent, wantPercent)
+	if gotPercent < wantPercent-tolerance || gotPercent > wantPercent+tolerance {
+		t.Errorf("got %v; want %v ± %v", gotPercent, wantPercent, tolerance)
+	}
+}

cmd/derper/depaware.txt

@@ -235,7 +235,7 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         encoding/pem                                                 from crypto/tls+
         errors                                                       from bufio+
         expvar                                                       from github.com/prometheus/client_golang/prometheus+
-        flag                                                         from tailscale.com/cmd/derper+
+        flag                                                         from tailscale.com/cmd/derper
         fmt                                                          from compress/flate+
         go/token                                                     from google.golang.org/protobuf/internal/strs
         hash                                                         from crypto+
@@ -253,7 +253,7 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         math/big                                                     from crypto/dsa+
         math/bits                                                    from compress/flate+
         math/rand                                                    from github.com/mdlayher/netlink+
-        math/rand/v2                                                 from tailscale.com/util/fastuuid
+        math/rand/v2                                                 from tailscale.com/util/fastuuid+
         mime                                                         from github.com/prometheus/common/expfmt+
         mime/multipart                                               from net/http
         mime/quotedprintable                                         from mime/multipart
@@ -277,7 +277,7 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         runtime/debug                                                from github.com/prometheus/client_golang/prometheus+
         runtime/metrics                                              from github.com/prometheus/client_golang/prometheus+
         runtime/pprof                                                from net/http/pprof
-        runtime/trace                                                from net/http/pprof+
+        runtime/trace                                                from net/http/pprof
         slices                                                       from tailscale.com/ipn/ipnstate+
         sort                                                         from compress/flate+
         strconv                                                      from compress/flate+
@@ -285,7 +285,6 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         sync                                                         from compress/flate+
         sync/atomic                                                  from context+
         syscall                                                      from crypto/rand+
-        testing                                                      from tailscale.com/util/syspolicy
         text/tabwriter                                               from runtime/pprof
         time                                                         from compress/gzip+
         unicode                                                      from bytes+

cmd/derper/derper.go

@@ -55,7 +55,7 @@ var (
 	meshPSKFile     = flag.String("mesh-psk-file", defaultMeshPSKFile(), "if non-empty, path to file containing the mesh pre-shared key file. It should contain some hex string; whitespace is trimmed.")
 	meshWith        = flag.String("mesh-with", "", "optional comma-separated list of hostnames to mesh with; the server's own hostname can be in the list")
 	bootstrapDNS    = flag.String("bootstrap-dns-names", "", "optional comma-separated list of hostnames to make available at /bootstrap-dns")
-	unpublishedDNS  = flag.String("unpublished-bootstrap-dns-names", "", "optional comma-separated list of hostnames to make available at /bootstrap-dns and not publish in the list")
+	unpublishedDNS  = flag.String("unpublished-bootstrap-dns-names", "", "optional comma-separated list of hostnames to make available at /bootstrap-dns and not publish in the list. If an entry contains a slash, the second part names a DNS record to poll for its TXT record with a `0` to `100` value for rollout percentage.")
 	verifyClients   = flag.Bool("verify-clients", false, "verify clients to this DERP server through a local tailscaled instance.")
 	verifyClientURL = flag.String("verify-client-url", "", "if non-empty, an admission controller URL for permitting client connections; see tailcfg.DERPAdmitClientRequest")
 	verifyFailOpen  = flag.Bool("verify-client-url-fail-open", true, "whether we fail open if --verify-client-url is unreachable")

cmd/derper/derper_test.go

@@ -99,6 +99,7 @@ func TestNoContent(t *testing.T) {
 func TestDeps(t *testing.T) {
 	deptest.DepChecker{
 		BadDeps: map[string]string{
+			"testing":                            "do not use testing package in production code",
 			"gvisor.dev/gvisor/pkg/buffer":       "https://github.com/tailscale/tailscale/issues/9756",
 			"gvisor.dev/gvisor/pkg/cpuid":        "https://github.com/tailscale/tailscale/issues/9756",
 			"gvisor.dev/gvisor/pkg/tcpip":        "https://github.com/tailscale/tailscale/issues/9756",

cmd/k8s-operator/deploy/chart/values.yaml

@@ -51,6 +51,10 @@ operatorConfig:
 # proxies created by the operator.
 # https://tailscale.com/kb/1236/kubernetes-operator/#cluster-ingress
 # https://tailscale.com/kb/1236/kubernetes-operator/#cluster-egress
+# Note that this section contains only a few global configuration options and
+# will not be updated with more configuration options in the future.
+# If you need more configuration options, take a look at ProxyClass:
+# https://tailscale.com/kb/1236/kubernetes-operator#cluster-resource-customization-using-proxyclass-custom-resource
 proxyConfig:
   image:
     repo: tailscale/tailscale

cmd/k8s-operator/nameserver.go

@@ -41,6 +41,12 @@ const (
 
 	messageNameserverCreationFailed  = "Failed creating nameserver resources: %v"
 	messageMultipleDNSConfigsPresent = "Multiple DNSConfig resources found in cluster. Please ensure no more than one is present."
+
+	defaultNameserverImageRepo = "tailscale/k8s-nameserver"
+	// TODO (irbekrm): once we start publishing nameserver images for stable
+	// track, replace 'unstable' here with the version of this operator
+	// instance.
+	defaultNameserverImageTag = "unstable"
 )
 
 // NameserverReconciler knows how to create nameserver resources in cluster in
@@ -163,11 +169,13 @@ func (a *NameserverReconciler) maybeProvision(ctx context.Context, tsDNSCfg *tsa
 		ownerRefs: []metav1.OwnerReference{*metav1.NewControllerRef(tsDNSCfg, tsapi.SchemeGroupVersion.WithKind("DNSConfig"))},
 		namespace: a.tsNamespace,
 		labels:    labels,
+		imageRepo: defaultNameserverImageRepo,
+		imageTag:  defaultNameserverImageTag,
 	}
-	if tsDNSCfg.Spec.Nameserver.Image.Repo != "" {
+	if tsDNSCfg.Spec.Nameserver.Image != nil && tsDNSCfg.Spec.Nameserver.Image.Repo != "" {
 		dCfg.imageRepo = tsDNSCfg.Spec.Nameserver.Image.Repo
 	}
-	if tsDNSCfg.Spec.Nameserver.Image.Tag != "" {
+	if tsDNSCfg.Spec.Nameserver.Image != nil && tsDNSCfg.Spec.Nameserver.Image.Tag != "" {
 		dCfg.imageTag = tsDNSCfg.Spec.Nameserver.Image.Tag
 	}
 	for _, deployable := range []deployable{saDeployable, deployDeployable, svcDeployable, cmDeployable} {

cmd/k8s-operator/nameserver_test.go

@@ -115,4 +115,13 @@ func TestNameserverReconciler(t *testing.T) {
 		Data:     map[string]string{"records.json": string(bs)},
 	}
 	expectEqual(t, fc, wantCm, nil)
+
+	// Verify that if dnsconfig.spec.nameserver.image.{repo,tag} are unset,
+	// the nameserver image defaults to tailscale/k8s-nameserver:unstable.
+	mustUpdate(t, fc, "", "test", func(dnsCfg *tsapi.DNSConfig) {
+		dnsCfg.Spec.Nameserver.Image = nil
+	})
+	expectReconciled(t, nr, "", "test")
+	wantsDeploy.Spec.Template.Spec.Containers[0].Image = "tailscale/k8s-nameserver:unstable"
+	expectEqual(t, fc, wantsDeploy, nil)
 }

cmd/k8s-operator/operator.go

@@ -45,12 +45,12 @@ import (
 	"tailscale.com/version"
 )
 
-// Generate static manifests for deploying Tailscale operator on Kubernetes from the operator's Helm chart.
-//go:generate go run tailscale.com/cmd/k8s-operator/generate staticmanifests
-
 // Generate Connector and ProxyClass CustomResourceDefinition yamls from their Go types.
 //go:generate go run sigs.k8s.io/controller-tools/cmd/controller-gen crd schemapatch:manifests=./deploy/crds output:dir=./deploy/crds paths=../../k8s-operator/apis/...
 
+// Generate static manifests for deploying Tailscale operator on Kubernetes from the operator's Helm chart.
+//go:generate go run tailscale.com/cmd/k8s-operator/generate staticmanifests
+
 // Generate CRD docs from the yamls
 //go:generate go run fybrik.io/crdoc --resources=./deploy/crds --output=../../k8s-operator/api.md
 

cmd/k8s-operator/operator_test.go

@@ -1182,7 +1182,7 @@ func TestTailscaledConfigfileHash(t *testing.T) {
 		parentType:      "svc",
 		hostname:        "default-test",
 		clusterTargetIP: "10.20.30.40",
-		confFileHash:    "705e5ffd0bd5326237efdf542c850a65a54101284d5daa30775420fcc64d89c1",
+		confFileHash:    "e09bededa0379920141cbd0b0dbdf9b8b66545877f9e8397423f5ce3e1ba439e",
 	}
 	expectEqual(t, fc, expectedSTS(t, fc, o), nil)
 
@@ -1192,7 +1192,7 @@ func TestTailscaledConfigfileHash(t *testing.T) {
 		mak.Set(&svc.Annotations, AnnotationHostname, "another-test")
 	})
 	o.hostname = "another-test"
-	o.confFileHash = "1a087f887825d2b75d3673c7c2b0131f8ec1f0b1cb761d33e236dd28350dfe23"
+	o.confFileHash = "5d754cf55463135ee34aa9821f2fd8483b53eb0570c3740c84a086304f427684"
 	expectReconciled(t, sr, "default", "test")
 	expectEqual(t, fc, expectedSTS(t, fc, o), nil)
 }

cmd/k8s-operator/sts.go

@@ -29,6 +29,7 @@ import (
 	"sigs.k8s.io/yaml"
 	"tailscale.com/client/tailscale"
 	"tailscale.com/ipn"
+	kubeutils "tailscale.com/k8s-operator"
 	tsoperator "tailscale.com/k8s-operator"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
 	"tailscale.com/net/netutil"
@@ -92,10 +93,6 @@ const (
 	podAnnotationLastSetTailnetTargetFQDN = "tailscale.com/operator-last-set-ts-tailnet-target-fqdn"
 	// podAnnotationLastSetConfigFileHash is sha256 hash of the current tailscaled configuration contents.
 	podAnnotationLastSetConfigFileHash = "tailscale.com/operator-last-set-config-file-hash"
-
-	// tailscaledConfigKey is the name of the key in proxy Secret Data that
-	// holds the tailscaled config contents.
-	tailscaledConfigKey = "tailscaled"
 )
 
 var (
@@ -174,11 +171,11 @@ func (a *tailscaleSTSReconciler) Provision(ctx context.Context, logger *zap.Suga
 		return nil, fmt.Errorf("failed to reconcile headless service: %w", err)
 	}
 
-	secretName, tsConfigHash, err := a.createOrGetSecret(ctx, logger, sts, hsvc)
+	secretName, tsConfigHash, configs, err := a.createOrGetSecret(ctx, logger, sts, hsvc)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create or get API key secret: %w", err)
 	}
-	_, err = a.reconcileSTS(ctx, logger, sts, hsvc, secretName, tsConfigHash)
+	_, err = a.reconcileSTS(ctx, logger, sts, hsvc, secretName, tsConfigHash, configs)
 	if err != nil {
 		return nil, fmt.Errorf("failed to reconcile statefulset: %w", err)
 	}
@@ -291,7 +288,7 @@ func (a *tailscaleSTSReconciler) reconcileHeadlessService(ctx context.Context, l
 	return createOrUpdate(ctx, a.Client, a.operatorNamespace, hsvc, func(svc *corev1.Service) { svc.Spec = hsvc.Spec })
 }
 
-func (a *tailscaleSTSReconciler) createOrGetSecret(ctx context.Context, logger *zap.SugaredLogger, stsC *tailscaleSTSConfig, hsvc *corev1.Service) (string, string, error) {
+func (a *tailscaleSTSReconciler) createOrGetSecret(ctx context.Context, logger *zap.SugaredLogger, stsC *tailscaleSTSConfig, hsvc *corev1.Service) (secretName, hash string, configs tailscaleConfigs, _ error) {
 	secret := &corev1.Secret{
 		ObjectMeta: metav1.ObjectMeta{
 			// Hardcode a -0 suffix so that in future, if we support
@@ -307,25 +304,23 @@ func (a *tailscaleSTSReconciler) createOrGetSecret(ctx context.Context, logger *
 		logger.Debugf("secret %s/%s already exists", secret.GetNamespace(), secret.GetName())
 		orig = secret.DeepCopy()
 	} else if !apierrors.IsNotFound(err) {
-		return "", "", err
+		return "", "", nil, err
 	}
 
-	var (
-		authKey, hash string
-	)
+	var authKey string
 	if orig == nil {
 		// Initially it contains only tailscaled config, but when the
 		// proxy starts, it will also store there the state, certs and
 		// ACME account key.
 		sts, err := getSingleObject[appsv1.StatefulSet](ctx, a.Client, a.operatorNamespace, stsC.ChildResourceLabels)
 		if err != nil {
-			return "", "", err
+			return "", "", nil, err
 		}
 		if sts != nil {
 			// StatefulSet exists, so we have already created the secret.
 			// If the secret is missing, they should delete the StatefulSet.
 			logger.Errorf("Tailscale proxy secret doesn't exist, but the corresponding StatefulSet %s/%s already does. Something is wrong, please delete the StatefulSet.", sts.GetNamespace(), sts.GetName())
-			return "", "", nil
+			return "", "", nil, nil
 		}
 		// Create API Key secret which is going to be used by the statefulset
 		// to authenticate with Tailscale.
@@ -336,45 +331,58 @@ func (a *tailscaleSTSReconciler) createOrGetSecret(ctx context.Context, logger *
 		}
 		authKey, err = a.newAuthKey(ctx, tags)
 		if err != nil {
-			return "", "", err
+			return "", "", nil, err
 		}
 	}
-	confFileBytes, h, err := tailscaledConfig(stsC, authKey, orig)
+	configs, err := tailscaledConfig(stsC, authKey, orig)
 	if err != nil {
-		return "", "", fmt.Errorf("error creating tailscaled config: %w", err)
+		return "", "", nil, fmt.Errorf("error creating tailscaled config: %w", err)
+	}
+	hash, err = tailscaledConfigHash(configs)
+	if err != nil {
+		return "", "", nil, fmt.Errorf("error calculating hash of tailscaled configs: %w", err)
+	}
+
+	latest := tailcfg.CapabilityVersion(-1)
+	var latestConfig ipn.ConfigVAlpha
+	for key, val := range configs {
+		fn := kubeutils.TailscaledConfigFileNameForCap(key)
+		b, err := json.Marshal(val)
+		if err != nil {
+			return "", "", nil, fmt.Errorf("error marshalling tailscaled config: %w", err)
+		}
+		mak.Set(&secret.StringData, fn, string(b))
+		if key > latest {
+			latest = key
+			latestConfig = val
+		}
 	}
-	hash = h
-	mak.Set(&secret.StringData, tailscaledConfigKey, string(confFileBytes))
 
 	if stsC.ServeConfig != nil {
 		j, err := json.Marshal(stsC.ServeConfig)
 		if err != nil {
-			return "", "", err
+			return "", "", nil, err
 		}
 		mak.Set(&secret.StringData, "serve-config", string(j))
 	}
 
 	if orig != nil {
-		logger.Debugf("patching the existing proxy Secret with tailscaled config %s", sanitizeConfigBytes(secret.Data[tailscaledConfigKey]))
+		logger.Debugf("patching the existing proxy Secret with tailscaled config %s", sanitizeConfigBytes(latestConfig))
 		if err := a.Patch(ctx, secret, client.MergeFrom(orig)); err != nil {
-			return "", "", err
+			return "", "", nil, err
 		}
 	} else {
-		logger.Debugf("creating a new Secret for the proxy with tailscaled config %s", sanitizeConfigBytes([]byte(secret.StringData[tailscaledConfigKey])))
+		logger.Debugf("creating a new Secret for the proxy with tailscaled config %s", sanitizeConfigBytes(latestConfig))
 		if err := a.Create(ctx, secret); err != nil {
-			return "", "", err
+			return "", "", nil, err
 		}
 	}
-	return secret.Name, hash, nil
+	return secret.Name, hash, configs, nil
 }
 
 // sanitizeConfigBytes returns ipn.ConfigVAlpha in string form with redacted
 // auth key.
-func sanitizeConfigBytes(bs []byte) string {
-	c := &ipn.ConfigVAlpha{}
-	if err := json.Unmarshal(bs, c); err != nil {
-		return "invalid config"
-	}
+func sanitizeConfigBytes(c ipn.ConfigVAlpha) string {
 	if c.AuthKey != nil {
 		c.AuthKey = ptr.To("**redacted**")
 	}
@@ -437,7 +445,7 @@ var proxyYaml []byte
 //go:embed deploy/manifests/userspace-proxy.yaml
 var userspaceProxyYaml []byte
 
-func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.SugaredLogger, sts *tailscaleSTSConfig, headlessSvc *corev1.Service, proxySecret, tsConfigHash string) (*appsv1.StatefulSet, error) {
+func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.SugaredLogger, sts *tailscaleSTSConfig, headlessSvc *corev1.Service, proxySecret, tsConfigHash string, configs map[tailcfg.CapabilityVersion]ipn.ConfigVAlpha) (*appsv1.StatefulSet, error) {
 	ss := new(appsv1.StatefulSet)
 	if sts.ServeConfig != nil && sts.ForwardClusterTrafficViaL7IngressProxy != true { // If forwarding cluster traffic via is required we need non-userspace + NET_ADMIN + forwarding
 		if err := yaml.Unmarshal(userspaceProxyYaml, &ss); err != nil {
@@ -493,9 +501,15 @@ func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.S
 			Value: proxySecret,
 		},
 		corev1.EnvVar{
+			// Old tailscaled config key is still used for backwards compatibility.
 			Name:  "EXPERIMENTAL_TS_CONFIGFILE_PATH",
 			Value: "/etc/tsconfig/tailscaled",
 		},
+		corev1.EnvVar{
+			// New style is in the form of cap-<capability-version>.hujson.
+			Name:  "TS_EXPERIMENTAL_VERSIONED_CONFIG_DIR",
+			Value: "/etc/tsconfig",
+		},
 	)
 	if sts.ForwardClusterTrafficViaL7IngressProxy {
 		container.Env = append(container.Env, corev1.EnvVar{
@@ -505,18 +519,16 @@ func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.S
 	}
 	// Configure containeboot to run tailscaled with a configfile read from the state Secret.
 	mak.Set(&ss.Spec.Template.Annotations, podAnnotationLastSetConfigFileHash, tsConfigHash)
-	pod.Spec.Volumes = append(ss.Spec.Template.Spec.Volumes, corev1.Volume{
+
+	configVolume := corev1.Volume{
 		Name: "tailscaledconfig",
 		VolumeSource: corev1.VolumeSource{
 			Secret: &corev1.SecretVolumeSource{
 				SecretName: proxySecret,
-				Items: []corev1.KeyToPath{{
-					Key:  tailscaledConfigKey,
-					Path: tailscaledConfigKey,
-				}},
 			},
 		},
-	})
+	}
+	pod.Spec.Volumes = append(ss.Spec.Template.Spec.Volumes, configVolume)
 	container.VolumeMounts = append(container.VolumeMounts, corev1.VolumeMount{
 		Name:      "tailscaledconfig",
 		ReadOnly:  true,
@@ -571,10 +583,7 @@ func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.S
 			VolumeSource: corev1.VolumeSource{
 				Secret: &corev1.SecretVolumeSource{
 					SecretName: proxySecret,
-					Items: []corev1.KeyToPath{{
-						Key:  "serve-config",
-						Path: "serve-config",
-					}},
+					Items:      []corev1.KeyToPath{{Key: "serve-config", Path: "serve-config"}},
 				},
 			},
 		})
@@ -716,42 +725,82 @@ func enableMetrics(ss *appsv1.StatefulSet, pc *tsapi.ProxyClass) {
 	}
 }
 
+func readAuthKey(secret *corev1.Secret, key string) (*string, error) {
+	origConf := &ipn.ConfigVAlpha{}
+	if err := json.Unmarshal([]byte(secret.Data[key]), origConf); err != nil {
+		return nil, fmt.Errorf("error unmarshaling previous tailscaled config in %q: %w", key, err)
+	}
+	return origConf.AuthKey, nil
+}
+
 // tailscaledConfig takes a proxy config, a newly generated auth key if
 // generated and a Secret with the previous proxy state and auth key and
-// produces returns tailscaled configuration and a hash of that configuration.
-func tailscaledConfig(stsC *tailscaleSTSConfig, newAuthkey string, oldSecret *corev1.Secret) ([]byte, string, error) {
-	conf := ipn.ConfigVAlpha{
-		Version:      "alpha0",
-		AcceptDNS:    "false",
-		AcceptRoutes: "false", // AcceptRoutes defaults to true
-		Locked:       "false",
-		Hostname:     &stsC.Hostname,
+// returns tailscaled configuration and a hash of that configuration.
+//
+// As of 2024-05-09 it also returns legacy tailscaled config without the
+// later added NoStatefulFilter field to support proxies older than cap95.
+// TODO (irbekrm): remove the legacy config once we no longer need to support
+// versions older than cap94,
+// https://tailscale.com/kb/1236/kubernetes-operator#operator-and-proxies
+func tailscaledConfig(stsC *tailscaleSTSConfig, newAuthkey string, oldSecret *corev1.Secret) (tailscaleConfigs, error) {
+	conf := &ipn.ConfigVAlpha{
+		Version:             "alpha0",
+		AcceptDNS:           "false",
+		AcceptRoutes:        "false", // AcceptRoutes defaults to true
+		Locked:              "false",
+		Hostname:            &stsC.Hostname,
+		NoStatefulFiltering: "false",
+	}
+
+	// For egress proxies only, we need to ensure that stateful filtering is
+	// not in place so that traffic from cluster can be forwarded via
+	// Tailscale IPs.
+	if stsC.TailnetTargetFQDN != "" || stsC.TailnetTargetIP != "" {
+		conf.NoStatefulFiltering = "true"
 	}
 	if stsC.Connector != nil {
 		routes, err := netutil.CalcAdvertiseRoutes(stsC.Connector.routes, stsC.Connector.isExitNode)
 		if err != nil {
-			return nil, "", fmt.Errorf("error calculating routes: %w", err)
+			return nil, fmt.Errorf("error calculating routes: %w", err)
 		}
 		conf.AdvertiseRoutes = routes
 	}
 	if newAuthkey != "" {
 		conf.AuthKey = &newAuthkey
-	} else if oldSecret != nil && len(oldSecret.Data[tailscaledConfigKey]) > 0 { // write to StringData, read from Data as StringData is write-only
-		origConf := &ipn.ConfigVAlpha{}
-		if err := json.Unmarshal([]byte(oldSecret.Data[tailscaledConfigKey]), origConf); err != nil {
-			return nil, "", fmt.Errorf("error unmarshaling previous tailscaled config: %w", err)
+	} else if oldSecret != nil {
+		var err error
+		latest := tailcfg.CapabilityVersion(-1)
+		latestStr := ""
+		for k, data := range oldSecret.Data {
+			// write to StringData, read from Data as StringData is write-only
+			if len(data) == 0 {
+				continue
+			}
+			v, err := kubeutils.CapVerFromFileName(k)
+			if err != nil {
+				continue
+			}
+			if v > latest {
+				latestStr = k
+				latest = v
+			}
+		}
+		// Allow for configs that don't contain an auth key. Perhaps
+		// users have some mechanisms to delete them. Auth key is
+		// normally not needed after the initial login.
+		if latestStr != "" {
+			conf.AuthKey, err = readAuthKey(oldSecret, latestStr)
+			if err != nil {
+				return nil, err
+			}
 		}
-		conf.AuthKey = origConf.AuthKey
 	}
-	confFileBytes, err := json.Marshal(conf)
-	if err != nil {
-		return nil, "", fmt.Errorf("error marshaling tailscaled config : %w", err)
-	}
-	hash, err := hashBytes(confFileBytes)
-	if err != nil {
-		return nil, "", fmt.Errorf("error calculating config hash: %w", err)
-	}
-	return confFileBytes, hash, nil
+	capVerConfigs := make(map[tailcfg.CapabilityVersion]ipn.ConfigVAlpha)
+	capVerConfigs[95] = *conf
+	// legacy config should not contain NoStatefulFiltering field.
+	conf.NoStatefulFiltering.Clear()
+	capVerConfigs[94] = *conf
+	return capVerConfigs, nil
 }
 
 // ptrObject is a type constraint for pointer types that implement
@@ -761,7 +810,9 @@ type ptrObject[T any] interface {
 	*T
 }
 
-// hashBytes produces a hash for the provided bytes that is the same across
+type tailscaleConfigs map[tailcfg.CapabilityVersion]ipn.ConfigVAlpha
+
+// hashBytes produces a hash for the provided tailscaled config that is the same across
 // different invocations of this code. We do not use the
 // tailscale.com/deephash.Hash here because that produces a different hash for
 // the same value in different tailscale builds. The hash we are producing here
@@ -770,10 +821,13 @@ type ptrObject[T any] interface {
 // thing that changed is operator version (the hash is also exposed to users via
 // an annotation and might be confusing if it changes without the config having
 // changed).
-func hashBytes(b []byte) (string, error) {
-	h := sha256.New()
-	_, err := h.Write(b)
+func tailscaledConfigHash(c tailscaleConfigs) (string, error) {
+	b, err := json.Marshal(c)
 	if err != nil {
+		return "", fmt.Errorf("error marshalling tailscaled configs: %w", err)
+	}
+	h := sha256.New()
+	if _, err = h.Write(b); err != nil {
 		return "", fmt.Errorf("error calculating hash: %w", err)
 	}
 	return fmt.Sprintf("%x", h.Sum(nil)), nil

cmd/k8s-operator/svc.go

@@ -161,7 +161,7 @@ func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 	}
 	if violations := validateService(svc); len(violations) > 0 {
 		msg := fmt.Sprintf("unable to provision proxy resources: invalid Service: %s", strings.Join(violations, ", "))
-		a.recorder.Event(svc, corev1.EventTypeWarning, "INVALIDSERVCICE", msg)
+		a.recorder.Event(svc, corev1.EventTypeWarning, "INVALIDSERVICE", msg)
 		a.logger.Error(msg)
 		return nil
 	}

cmd/k8s-operator/testutils_test.go

@@ -67,6 +67,7 @@ func expectedSTS(t *testing.T, cl client.Client, opts configOpts) *appsv1.Statef
 			{Name: "POD_IP", ValueFrom: &corev1.EnvVarSource{FieldRef: &corev1.ObjectFieldSelector{APIVersion: "", FieldPath: "status.podIP"}, ResourceFieldRef: nil, ConfigMapKeyRef: nil, SecretKeyRef: nil}},
 			{Name: "TS_KUBE_SECRET", Value: opts.secretName},
 			{Name: "EXPERIMENTAL_TS_CONFIGFILE_PATH", Value: "/etc/tsconfig/tailscaled"},
+			{Name: "TS_EXPERIMENTAL_VERSIONED_CONFIG_DIR", Value: "/etc/tsconfig"},
 		},
 		SecurityContext: &corev1.SecurityContext{
 			Capabilities: &corev1.Capabilities{
@@ -89,12 +90,6 @@ func expectedSTS(t *testing.T, cl client.Client, opts configOpts) *appsv1.Statef
 			VolumeSource: corev1.VolumeSource{
 				Secret: &corev1.SecretVolumeSource{
 					SecretName: opts.secretName,
-					Items: []corev1.KeyToPath{
-						{
-							Key:  "tailscaled",
-							Path: "tailscaled",
-						},
-					},
 				},
 			},
 		},
@@ -144,9 +139,7 @@ func expectedSTS(t *testing.T, cl client.Client, opts configOpts) *appsv1.Statef
 			Name:  "TS_SERVE_CONFIG",
 			Value: "/etc/tailscaled/serve-config",
 		})
-		volumes = append(volumes, corev1.Volume{
-			Name: "serve-config", VolumeSource: corev1.VolumeSource{Secret: &corev1.SecretVolumeSource{SecretName: opts.secretName, Items: []corev1.KeyToPath{{Path: "serve-config", Key: "serve-config"}}}},
-		})
+		volumes = append(volumes, corev1.Volume{Name: "serve-config", VolumeSource: corev1.VolumeSource{Secret: &corev1.SecretVolumeSource{SecretName: opts.secretName, Items: []corev1.KeyToPath{{Key: "serve-config", Path: "serve-config"}}}}})
 		tsContainer.VolumeMounts = append(tsContainer.VolumeMounts, corev1.VolumeMount{Name: "serve-config", ReadOnly: true, MountPath: "/etc/tailscaled"})
 	}
 	ss := &appsv1.StatefulSet{
@@ -229,6 +222,7 @@ func expectedSTSUserspace(t *testing.T, cl client.Client, opts configOpts) *apps
 			{Name: "POD_IP", ValueFrom: &corev1.EnvVarSource{FieldRef: &corev1.ObjectFieldSelector{APIVersion: "", FieldPath: "status.podIP"}, ResourceFieldRef: nil, ConfigMapKeyRef: nil, SecretKeyRef: nil}},
 			{Name: "TS_KUBE_SECRET", Value: opts.secretName},
 			{Name: "EXPERIMENTAL_TS_CONFIGFILE_PATH", Value: "/etc/tsconfig/tailscaled"},
+			{Name: "TS_EXPERIMENTAL_VERSIONED_CONFIG_DIR", Value: "/etc/tsconfig"},
 			{Name: "TS_SERVE_CONFIG", Value: "/etc/tailscaled/serve-config"},
 		},
 		ImagePullPolicy: "Always",
@@ -243,20 +237,12 @@ func expectedSTSUserspace(t *testing.T, cl client.Client, opts configOpts) *apps
 			VolumeSource: corev1.VolumeSource{
 				Secret: &corev1.SecretVolumeSource{
 					SecretName: opts.secretName,
-					Items: []corev1.KeyToPath{
-						{
-							Key:  "tailscaled",
-							Path: "tailscaled",
-						},
-					},
 				},
 			},
 		},
 		{Name: "serve-config",
 			VolumeSource: corev1.VolumeSource{
-				Secret: &corev1.SecretVolumeSource{SecretName: opts.secretName,
-					Items: []corev1.KeyToPath{{Key: "serve-config", Path: "serve-config"}}}},
-		},
+				Secret: &corev1.SecretVolumeSource{SecretName: opts.secretName, Items: []corev1.KeyToPath{{Key: "serve-config", Path: "serve-config"}}}}},
 	}
 	ss := &appsv1.StatefulSet{
 		TypeMeta: metav1.TypeMeta{
@@ -388,7 +374,17 @@ func expectedSecret(t *testing.T, opts configOpts) *corev1.Secret {
 	if err != nil {
 		t.Fatalf("error marshalling tailscaled config")
 	}
+	if opts.tailnetTargetFQDN != "" || opts.tailnetTargetIP != "" {
+		conf.NoStatefulFiltering = "true"
+	} else {
+		conf.NoStatefulFiltering = "false"
+	}
+	bn, err := json.Marshal(conf)
+	if err != nil {
+		t.Fatalf("error marshalling tailscaled config")
+	}
 	mak.Set(&s.StringData, "tailscaled", string(b))
+	mak.Set(&s.StringData, "cap-95.hujson", string(bn))
 	labels := map[string]string{
 		"tailscale.com/managed":              "true",
 		"tailscale.com/parent-resource":      "test",
@@ -463,7 +459,7 @@ func mustUpdateStatus[T any, O ptrObject[T]](t *testing.T, client client.Client,
 // they are not present in the passed object and use the modify func to remove
 // them from the cluster object. If no such modifications are needed, you can
 // pass nil in place of the modify function.
-func expectEqual[T any, O ptrObject[T]](t *testing.T, client client.Client, want O, modify func(O)) {
+func expectEqual[T any, O ptrObject[T]](t *testing.T, client client.Client, want O, modifier func(O)) {
 	t.Helper()
 	got := O(new(T))
 	if err := client.Get(context.Background(), types.NamespacedName{
@@ -477,8 +473,8 @@ func expectEqual[T any, O ptrObject[T]](t *testing.T, client client.Client, want
 	// so just remove it from both got and want.
 	got.SetResourceVersion("")
 	want.SetResourceVersion("")
-	if modify != nil {
-		modify(got)
+	if modifier != nil {
+		modifier(got)
 	}
 	if diff := cmp.Diff(got, want); diff != "" {
 		t.Fatalf("unexpected object (-got +want):\n%s", diff)

cmd/pgproxy/pgproxy.go

@@ -28,7 +28,6 @@ import (
 	"tailscale.com/metrics"
 	"tailscale.com/tsnet"
 	"tailscale.com/tsweb"
-	"tailscale.com/types/logger"
 )
 
 var (
@@ -58,8 +57,6 @@ func main() {
 	ts := &tsnet.Server{
 		Dir:      *tailscaleDir,
 		Hostname: *hostname,
-		// Make the stdout logs a clean audit log of connections.
-		Logf: logger.Discard,
 	}
 
 	if os.Getenv("TS_AUTHKEY") == "" {

cmd/sniproxy/sniproxy_test.go

@@ -8,6 +8,7 @@ import (
 	"encoding/json"
 	"flag"
 	"fmt"
+	"log"
 	"net"
 	"net/http/httptest"
 	"net/netip"
@@ -99,8 +100,8 @@ func startNode(t *testing.T, ctx context.Context, controlURL, hostname string) (
 		Store:      new(mem.Store),
 		Ephemeral:  true,
 	}
-	if !*verboseNodes {
-		s.Logf = logger.Discard
+	if *verboseNodes {
+		s.Logf = log.Printf
 	}
 	t.Cleanup(func() { s.Close() })
 

cmd/stunc/stunc.go

@@ -20,7 +20,7 @@ func main() {
 	}
 	host := os.Args[1]
 
-	uaddr, err := net.ResolveUDPAddr("udp", host+":3478")
+	uaddr, err := net.ResolveUDPAddr("udp", net.JoinHostPort(host, "3478"))
 	if err != nil {
 		log.Fatal(err)
 	}

cmd/tailscale/cli/cli_test.go

@@ -24,6 +24,7 @@ import (
 	"tailscale.com/tka"
 	"tailscale.com/tstest"
 	"tailscale.com/types/logger"
+	"tailscale.com/types/opt"
 	"tailscale.com/types/persist"
 	"tailscale.com/types/preftype"
 	"tailscale.com/version/distro"
@@ -176,9 +177,10 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "bare_up_means_up",
 			flags: []string{},
 			curPrefs: &ipn.Prefs{
-				ControlURL:  ipn.DefaultControlURL,
-				WantRunning: false,
-				Hostname:    "foo",
+				ControlURL:          ipn.DefaultControlURL,
+				WantRunning:         false,
+				Hostname:            "foo",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "",
 		},
@@ -186,12 +188,12 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "losing_hostname",
 			flags: []string{"--accept-dns"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				WantRunning:      false,
-				Hostname:         "foo",
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
-				AllowSingleHosts: true,
+				ControlURL:          ipn.DefaultControlURL,
+				WantRunning:         false,
+				Hostname:            "foo",
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --accept-dns --hostname=foo",
 		},
@@ -199,11 +201,11 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "hostname_changing_explicitly",
 			flags: []string{"--hostname=bar"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
-				AllowSingleHosts: true,
-				Hostname:         "foo",
+				ControlURL:          ipn.DefaultControlURL,
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				Hostname:            "foo",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "",
 		},
@@ -211,11 +213,11 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "hostname_changing_empty_explicitly",
 			flags: []string{"--hostname="},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
-				AllowSingleHosts: true,
-				Hostname:         "foo",
+				ControlURL:          ipn.DefaultControlURL,
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				Hostname:            "foo",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "",
 		},
@@ -231,11 +233,11 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "implicit_operator_change",
 			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				OperatorUser:     "alice",
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          ipn.DefaultControlURL,
+				OperatorUser:        "alice",
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			curUser: "eve",
 			want:    accidentalUpPrefix + " --hostname=foo --operator=alice",
@@ -244,11 +246,11 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "implicit_operator_matches_shell_user",
 			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
-				OperatorUser:     "alice",
+				ControlURL:          ipn.DefaultControlURL,
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				OperatorUser:        "alice",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			curUser: "alice",
 			want:    "",
@@ -257,15 +259,15 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "error_advertised_routes_exit_node_removed",
 			flags: []string{"--advertise-routes=10.0.42.0/24"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("10.0.42.0/24"),
 					netip.MustParsePrefix("0.0.0.0/0"),
 					netip.MustParsePrefix("::/0"),
 				},
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --advertise-routes=10.0.42.0/24 --advertise-exit-node",
 		},
@@ -273,15 +275,15 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "advertised_routes_exit_node_removed_explicit",
 			flags: []string{"--advertise-routes=10.0.42.0/24", "--advertise-exit-node=false"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("10.0.42.0/24"),
 					netip.MustParsePrefix("0.0.0.0/0"),
 					netip.MustParsePrefix("::/0"),
 				},
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "",
 		},
@@ -289,15 +291,15 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "advertised_routes_includes_the_0_routes", // but no --advertise-exit-node
 			flags: []string{"--advertise-routes=11.1.43.0/24,0.0.0.0/0,::/0"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("10.0.42.0/24"),
 					netip.MustParsePrefix("0.0.0.0/0"),
 					netip.MustParsePrefix("::/0"),
 				},
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "",
 		},
@@ -305,10 +307,10 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "advertise_exit_node", // Issue 1859
 			flags: []string{"--advertise-exit-node"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          ipn.DefaultControlURL,
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "",
 		},
@@ -316,14 +318,14 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "advertise_exit_node_over_existing_routes",
 			flags: []string{"--advertise-exit-node"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("1.2.0.0/16"),
 				},
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --advertise-exit-node --advertise-routes=1.2.0.0/16",
 		},
@@ -331,15 +333,15 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "advertise_exit_node_over_existing_routes_and_exit_node",
 			flags: []string{"--advertise-exit-node"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("0.0.0.0/0"),
 					netip.MustParsePrefix("::/0"),
 					netip.MustParsePrefix("1.2.0.0/16"),
 				},
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --advertise-exit-node --advertise-routes=1.2.0.0/16",
 		},
@@ -347,12 +349,12 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "exit_node_clearing", // Issue 1777
 			flags: []string{"--exit-node="},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 
-				ExitNodeID: "fooID",
+				ExitNodeID:          "fooID",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "",
 		},
@@ -360,59 +362,59 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "remove_all_implicit",
 			flags: []string{"--force-reauth"},
 			curPrefs: &ipn.Prefs{
-				WantRunning:      true,
-				ControlURL:       ipn.DefaultControlURL,
-				RouteAll:         true,
-				AllowSingleHosts: false,
-				ExitNodeIP:       netip.MustParseAddr("100.64.5.6"),
-				CorpDNS:          false,
-				ShieldsUp:        true,
-				AdvertiseTags:    []string{"tag:foo", "tag:bar"},
-				Hostname:         "myhostname",
-				ForceDaemon:      true,
+				WantRunning:   true,
+				ControlURL:    ipn.DefaultControlURL,
+				RouteAll:      true,
+				ExitNodeIP:    netip.MustParseAddr("100.64.5.6"),
+				CorpDNS:       false,
+				ShieldsUp:     true,
+				AdvertiseTags: []string{"tag:foo", "tag:bar"},
+				Hostname:      "myhostname",
+				ForceDaemon:   true,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("10.0.0.0/16"),
 					netip.MustParsePrefix("0.0.0.0/0"),
 					netip.MustParsePrefix("::/0"),
 				},
-				NetfilterMode: preftype.NetfilterNoDivert,
-				OperatorUser:  "alice",
+				NetfilterMode:       preftype.NetfilterNoDivert,
+				OperatorUser:        "alice",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			curUser: "eve",
-			want:    accidentalUpPrefix + " --force-reauth --accept-dns=false --accept-routes --advertise-exit-node --advertise-routes=10.0.0.0/16 --advertise-tags=tag:foo,tag:bar --exit-node=100.64.5.6 --host-routes=false --hostname=myhostname --netfilter-mode=nodivert --operator=alice --shields-up",
+			want:    accidentalUpPrefix + " --force-reauth --accept-dns=false --accept-routes --advertise-exit-node --advertise-routes=10.0.0.0/16 --advertise-tags=tag:foo,tag:bar --exit-node=100.64.5.6 --hostname=myhostname --netfilter-mode=nodivert --operator=alice --shields-up",
 		},
 		{
 			name:  "remove_all_implicit_except_hostname",
 			flags: []string{"--hostname=newhostname"},
 			curPrefs: &ipn.Prefs{
-				WantRunning:      true,
-				ControlURL:       ipn.DefaultControlURL,
-				RouteAll:         true,
-				AllowSingleHosts: false,
-				ExitNodeIP:       netip.MustParseAddr("100.64.5.6"),
-				CorpDNS:          false,
-				ShieldsUp:        true,
-				AdvertiseTags:    []string{"tag:foo", "tag:bar"},
-				Hostname:         "myhostname",
-				ForceDaemon:      true,
+				WantRunning:   true,
+				ControlURL:    ipn.DefaultControlURL,
+				RouteAll:      true,
+				ExitNodeIP:    netip.MustParseAddr("100.64.5.6"),
+				CorpDNS:       false,
+				ShieldsUp:     true,
+				AdvertiseTags: []string{"tag:foo", "tag:bar"},
+				Hostname:      "myhostname",
+				ForceDaemon:   true,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("10.0.0.0/16"),
 				},
-				NetfilterMode: preftype.NetfilterNoDivert,
-				OperatorUser:  "alice",
+				NetfilterMode:       preftype.NetfilterNoDivert,
+				OperatorUser:        "alice",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			curUser: "eve",
-			want:    accidentalUpPrefix + " --hostname=newhostname --accept-dns=false --accept-routes --advertise-routes=10.0.0.0/16 --advertise-tags=tag:foo,tag:bar --exit-node=100.64.5.6 --host-routes=false --netfilter-mode=nodivert --operator=alice --shields-up",
+			want:    accidentalUpPrefix + " --hostname=newhostname --accept-dns=false --accept-routes --advertise-routes=10.0.0.0/16 --advertise-tags=tag:foo,tag:bar --exit-node=100.64.5.6 --netfilter-mode=nodivert --operator=alice --shields-up",
 		},
 		{
 			name:  "loggedout_is_implicit",
 			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				LoggedOut:        true,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          ipn.DefaultControlURL,
+				LoggedOut:           true,
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "", // not an error. LoggedOut is implicit.
 		},
@@ -422,10 +424,9 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "make_windows_exit_node",
 			flags: []string{"--advertise-exit-node"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				RouteAll:         true,
+				ControlURL: ipn.DefaultControlURL,
+				CorpDNS:    true,
+				RouteAll:   true,
 
 				// And assume this no-op accidental pre-1.8 value:
 				NoSNAT: true,
@@ -437,8 +438,7 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "ignore_netfilter_change_non_linux",
 			flags: []string{"--accept-dns"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
+				ControlURL: ipn.DefaultControlURL,
 
 				NetfilterMode: preftype.NetfilterNoDivert, // we never had this bug, but pretend it got set non-zero on Windows somehow
 			},
@@ -449,15 +449,15 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "operator_losing_routes_step1", // https://twitter.com/EXPbits/status/1390418145047887877
 			flags: []string{"--operator=expbits"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("0.0.0.0/0"),
 					netip.MustParsePrefix("::/0"),
 					netip.MustParsePrefix("1.2.0.0/16"),
 				},
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --operator=expbits --advertise-exit-node --advertise-routes=1.2.0.0/16",
 		},
@@ -465,15 +465,15 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "operator_losing_routes_step2", // https://twitter.com/EXPbits/status/1390418145047887877
 			flags: []string{"--operator=expbits", "--advertise-routes=1.2.0.0/16"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("0.0.0.0/0"),
 					netip.MustParsePrefix("::/0"),
 					netip.MustParsePrefix("1.2.0.0/16"),
 				},
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --advertise-routes=1.2.0.0/16 --operator=expbits --advertise-exit-node",
 		},
@@ -481,13 +481,13 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "errors_preserve_explicit_flags",
 			flags: []string{"--reset", "--force-reauth=false", "--authkey=secretrand"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				WantRunning:      false,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
-				AllowSingleHosts: true,
+				ControlURL:    ipn.DefaultControlURL,
+				WantRunning:   false,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 
-				Hostname: "foo",
+				Hostname:            "foo",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --auth-key=secretrand --force-reauth=false --reset --hostname=foo",
 		},
@@ -495,12 +495,12 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "error_exit_node_omit_with_ip_pref",
 			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 
-				ExitNodeIP: netip.MustParseAddr("100.64.5.4"),
+				ExitNodeIP:          netip.MustParseAddr("100.64.5.4"),
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --hostname=foo --exit-node=100.64.5.4",
 		},
@@ -509,12 +509,12 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			flags:         []string{"--hostname=foo"},
 			curExitNodeIP: netip.MustParseAddr("100.64.5.7"),
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 
-				ExitNodeID: "some_stable_id",
+				ExitNodeID:          "some_stable_id",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --hostname=foo --exit-node=100.64.5.7",
 		},
@@ -523,13 +523,13 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			flags:         []string{"--hostname=foo"},
 			curExitNodeIP: netip.MustParseAddr("100.2.3.4"),
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 
 				ExitNodeAllowLANAccess: true,
 				ExitNodeID:             "some_stable_id",
+				NoStatefulFiltering:    opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --hostname=foo --exit-node-allow-lan-access --exit-node=100.2.3.4",
 		},
@@ -537,10 +537,10 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "ignore_login_server_synonym",
 			flags: []string{"--login-server=https://controlplane.tailscale.com"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: "", // not an error
 		},
@@ -548,10 +548,10 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "ignore_login_server_synonym_on_other_change",
 			flags: []string{"--netfilter-mode=off"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				AllowSingleHosts: true,
-				CorpDNS:          false,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				CorpDNS:             false,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			want: accidentalUpPrefix + " --netfilter-mode=off --accept-dns=false",
 		},
@@ -561,11 +561,11 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "synology_permit_omit_accept_routes",
 			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				CorpDNS:          true,
-				AllowSingleHosts: true,
-				RouteAll:         true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				CorpDNS:             true,
+				RouteAll:            true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			goos:   "linux",
 			distro: distro.Synology,
@@ -577,11 +577,11 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "not_synology_dont_permit_omit_accept_routes",
 			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				CorpDNS:          true,
-				AllowSingleHosts: true,
-				RouteAll:         true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				CorpDNS:             true,
+				RouteAll:            true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			goos:   "linux",
 			distro: "", // not Synology
@@ -591,11 +591,11 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			name:  "profile_name_ignored_in_up",
 			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				CorpDNS:          true,
-				AllowSingleHosts: true,
-				NetfilterMode:    preftype.NetfilterOn,
-				ProfileName:      "foo",
+				ControlURL:          "https://login.tailscale.com",
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				ProfileName:         "foo",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			goos: "linux",
 			want: "",
@@ -655,12 +655,12 @@ func TestPrefsFromUpArgs(t *testing.T) {
 			goos: "linux",
 			args: upArgsFromOSArgs("linux"),
 			want: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				WantRunning:      true,
-				NoSNAT:           false,
-				NetfilterMode:    preftype.NetfilterOn,
-				CorpDNS:          true,
-				AllowSingleHosts: true,
+				ControlURL:          ipn.DefaultControlURL,
+				WantRunning:         true,
+				NoSNAT:              false,
+				NoStatefulFiltering: "true",
+				NetfilterMode:       preftype.NetfilterOn,
+				CorpDNS:             true,
 				AutoUpdate: ipn.AutoUpdatePrefs{
 					Check: true,
 				},
@@ -671,12 +671,13 @@ func TestPrefsFromUpArgs(t *testing.T) {
 			goos: "windows",
 			args: upArgsFromOSArgs("windows"),
 			want: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				WantRunning:      true,
-				CorpDNS:          true,
-				AllowSingleHosts: true,
-				RouteAll:         true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          ipn.DefaultControlURL,
+				WantRunning:         true,
+				CorpDNS:             true,
+				RouteAll:            true,
+				NoSNAT:              false,
+				NoStatefulFiltering: "true",
+				NetfilterMode:       preftype.NetfilterOn,
 				AutoUpdate: ipn.AutoUpdatePrefs{
 					Check: true,
 				},
@@ -686,15 +687,15 @@ func TestPrefsFromUpArgs(t *testing.T) {
 			name: "advertise_default_route",
 			args: upArgsFromOSArgs("linux", "--advertise-exit-node"),
 			want: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				WantRunning:      true,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
+				ControlURL:  ipn.DefaultControlURL,
+				WantRunning: true,
+				CorpDNS:     true,
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("0.0.0.0/0"),
 					netip.MustParsePrefix("::/0"),
 				},
-				NetfilterMode: preftype.NetfilterOn,
+				NoStatefulFiltering: "true",
+				NetfilterMode:       preftype.NetfilterOn,
 				AutoUpdate: ipn.AutoUpdatePrefs{
 					Check: true,
 				},
@@ -781,9 +782,10 @@ func TestPrefsFromUpArgs(t *testing.T) {
 			},
 			wantWarn: "netfilter=nodivert; add iptables calls to ts-* chains manually.",
 			want: &ipn.Prefs{
-				WantRunning:   true,
-				NetfilterMode: preftype.NetfilterNoDivert,
-				NoSNAT:        true,
+				WantRunning:         true,
+				NetfilterMode:       preftype.NetfilterNoDivert,
+				NoSNAT:              true,
+				NoStatefulFiltering: "true",
 				AutoUpdate: ipn.AutoUpdatePrefs{
 					Check: true,
 				},
@@ -797,9 +799,10 @@ func TestPrefsFromUpArgs(t *testing.T) {
 			},
 			wantWarn: "netfilter=off; configure iptables yourself.",
 			want: &ipn.Prefs{
-				WantRunning:   true,
-				NetfilterMode: preftype.NetfilterOff,
-				NoSNAT:        true,
+				WantRunning:         true,
+				NetfilterMode:       preftype.NetfilterOff,
+				NoSNAT:              true,
+				NoStatefulFiltering: "true",
 				AutoUpdate: ipn.AutoUpdatePrefs{
 					Check: true,
 				},
@@ -813,8 +816,9 @@ func TestPrefsFromUpArgs(t *testing.T) {
 				netfilterMode:   "off",
 			},
 			want: &ipn.Prefs{
-				WantRunning: true,
-				NoSNAT:      true,
+				WantRunning:         true,
+				NoSNAT:              true,
+				NoStatefulFiltering: "true",
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("fd7a:115c:a1e0:b1a::bb:10.0.0.0/112"),
 				},
@@ -831,8 +835,9 @@ func TestPrefsFromUpArgs(t *testing.T) {
 				netfilterMode:   "off",
 			},
 			want: &ipn.Prefs{
-				WantRunning: true,
-				NoSNAT:      true,
+				WantRunning:         true,
+				NoSNAT:              true,
+				NoStatefulFiltering: "true",
 				AdvertiseRoutes: []netip.Prefix{
 					netip.MustParsePrefix("fd7a:115c:a1e0:b1a::aabb:10.0.0.0/112"),
 				},
@@ -914,6 +919,9 @@ func TestPrefFlagMapping(t *testing.T) {
 			continue
 		}
 		switch prefName {
+		case "AllowSingleHosts":
+			// Fake pref for downgrade compat. See #12058.
+			continue
 		case "WantRunning", "Persist", "LoggedOut":
 			// All explicitly handled (ignored) by checkForAccidentalSettingReverts.
 			continue
@@ -1021,7 +1029,6 @@ func TestUpdatePrefs(t *testing.T) {
 			wantJustEditMP: &ipn.MaskedPrefs{
 				AdvertiseRoutesSet:        true,
 				AdvertiseTagsSet:          true,
-				AllowSingleHostsSet:       true,
 				AppConnectorSet:           true,
 				ControlURLSet:             true,
 				CorpDNSSet:                true,
@@ -1031,6 +1038,7 @@ func TestUpdatePrefs(t *testing.T) {
 				HostnameSet:               true,
 				NetfilterModeSet:          true,
 				NoSNATSet:                 true,
+				NoStatefulFilteringSet:    true,
 				OperatorUserSet:           true,
 				RouteAllSet:               true,
 				RunSSHSet:                 true,
@@ -1053,11 +1061,11 @@ func TestUpdatePrefs(t *testing.T) {
 			name:  "change_login_server",
 			flags: []string{"--login-server=https://localhost:1000"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				Persist:          &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				Persist:             &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			env:            upCheckEnv{backendState: "Running"},
 			wantSimpleUp:   true,
@@ -1068,11 +1076,11 @@ func TestUpdatePrefs(t *testing.T) {
 			name:  "change_tags",
 			flags: []string{"--advertise-tags=tag:foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				Persist:          &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				Persist:             &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			env: upCheckEnv{backendState: "Running"},
 		},
@@ -1081,11 +1089,11 @@ func TestUpdatePrefs(t *testing.T) {
 			name:  "explicit_empty_operator",
 			flags: []string{"--operator="},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				CorpDNS:          true,
-				AllowSingleHosts: true,
-				NetfilterMode:    preftype.NetfilterOn,
-				OperatorUser:     "somebody",
+				ControlURL:          "https://login.tailscale.com",
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				OperatorUser:        "somebody",
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			env: upCheckEnv{user: "somebody", backendState: "Running"},
 			wantJustEditMP: &ipn.MaskedPrefs{
@@ -1102,11 +1110,11 @@ func TestUpdatePrefs(t *testing.T) {
 			name:  "enable_ssh",
 			flags: []string{"--ssh"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				Persist:          &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				Persist:             &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: &ipn.MaskedPrefs{
 				RunSSHSet:      true,
@@ -1123,12 +1131,12 @@ func TestUpdatePrefs(t *testing.T) {
 			name:  "disable_ssh",
 			flags: []string{"--ssh=false"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				Persist:          &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				RunSSH:           true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				Persist:             &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
+				CorpDNS:             true,
+				RunSSH:              true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: &ipn.MaskedPrefs{
 				RunSSHSet:      true,
@@ -1148,12 +1156,12 @@ func TestUpdatePrefs(t *testing.T) {
 			flags:            []string{"--ssh=false"},
 			sshOverTailscale: true,
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				Persist:          &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
-				RunSSH:           true,
+				ControlURL:          "https://login.tailscale.com",
+				Persist:             &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				RunSSH:              true,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: &ipn.MaskedPrefs{
 				RunSSHSet:      true,
@@ -1172,11 +1180,11 @@ func TestUpdatePrefs(t *testing.T) {
 			flags:            []string{"--ssh=true"},
 			sshOverTailscale: true,
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				Persist:          &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				Persist:             &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: &ipn.MaskedPrefs{
 				RunSSHSet:      true,
@@ -1195,11 +1203,11 @@ func TestUpdatePrefs(t *testing.T) {
 			flags:            []string{"--ssh=true", "--accept-risk=lose-ssh"},
 			sshOverTailscale: true,
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				Persist:          &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				Persist:             &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: &ipn.MaskedPrefs{
 				RunSSHSet:      true,
@@ -1217,12 +1225,12 @@ func TestUpdatePrefs(t *testing.T) {
 			flags:            []string{"--ssh=false", "--accept-risk=lose-ssh"},
 			sshOverTailscale: true,
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				Persist:          &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				RunSSH:           true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				Persist:             &persist.Persist{UserProfile: tailcfg.UserProfile{LoginName: "crawshaw.github"}},
+				CorpDNS:             true,
+				RunSSH:              true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: &ipn.MaskedPrefs{
 				RunSSHSet:      true,
@@ -1240,10 +1248,10 @@ func TestUpdatePrefs(t *testing.T) {
 			flags:            []string{"--force-reauth"},
 			sshOverTailscale: true,
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			env:          upCheckEnv{backendState: "Running"},
 			wantErrSubtr: "aborted, no changes made",
@@ -1253,10 +1261,10 @@ func TestUpdatePrefs(t *testing.T) {
 			flags:            []string{"--force-reauth", "--accept-risk=lose-ssh"},
 			sshOverTailscale: true,
 			curPrefs: &ipn.Prefs{
-				ControlURL:       "https://login.tailscale.com",
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          "https://login.tailscale.com",
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: nil,
 			env:            upCheckEnv{backendState: "Running"},
@@ -1265,10 +1273,10 @@ func TestUpdatePrefs(t *testing.T) {
 			name:  "advertise_connector",
 			flags: []string{"--advertise-connector"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:          ipn.DefaultControlURL,
+				CorpDNS:             true,
+				NetfilterMode:       preftype.NetfilterOn,
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: &ipn.MaskedPrefs{
 				AppConnectorSet: true,
@@ -1285,13 +1293,13 @@ func TestUpdatePrefs(t *testing.T) {
 			name:  "no_advertise_connector",
 			flags: []string{"--advertise-connector=false"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:       ipn.DefaultControlURL,
-				AllowSingleHosts: true,
-				CorpDNS:          true,
-				NetfilterMode:    preftype.NetfilterOn,
+				ControlURL:    ipn.DefaultControlURL,
+				CorpDNS:       true,
+				NetfilterMode: preftype.NetfilterOn,
 				AppConnector: ipn.AppConnectorPrefs{
 					Advertise: true,
 				},
+				NoStatefulFiltering: opt.NewBool(true),
 			},
 			wantJustEditMP: &ipn.MaskedPrefs{
 				AppConnectorSet: true,

cmd/tailscale/cli/ffcomplete/complete.go

@@ -97,7 +97,7 @@ func Inject(root *ffcli.Command, hide func(*ffcli.Command), usageFunc func(*ffcl
 		&ffcli.Command{
 			Name:       "completion",
 			ShortUsage: root.Name + " completion <shell> [--flags] [--descs]",
-			ShortHelp:  "Shell tab-completion scripts.",
+			ShortHelp:  "Shell tab-completion scripts",
 			LongHelp:   fmt.Sprintf(cobra.UsageTemplate, root.Name),
 
 			// Print help if run without args.

cmd/tailscale/cli/netcheck.go

@@ -127,13 +127,13 @@ func printReport(dm *tailcfg.DERPMap, report *netcheck.Report) error {
 
 	printf("\nReport:\n")
 	printf("\t* UDP: %v\n", report.UDP)
-	if report.GlobalV4 != "" {
-		printf("\t* IPv4: yes, %v\n", report.GlobalV4)
+	if report.GlobalV4.IsValid() {
+		printf("\t* IPv4: yes, %s\n", report.GlobalV4)
 	} else {
 		printf("\t* IPv4: (no addr found)\n")
 	}
-	if report.GlobalV6 != "" {
-		printf("\t* IPv6: yes, %v\n", report.GlobalV6)
+	if report.GlobalV6.IsValid() {
+		printf("\t* IPv6: yes, %s\n", report.GlobalV6)
 	} else if report.IPv6 {
 		printf("\t* IPv6: (no addr found)\n")
 	} else if report.OSHasIPv6 {
@@ -142,7 +142,6 @@ func printReport(dm *tailcfg.DERPMap, report *netcheck.Report) error {
 		printf("\t* IPv6: no, unavailable in OS\n")
 	}
 	printf("\t* MappingVariesByDestIP: %v\n", report.MappingVariesByDestIP)
-	printf("\t* HairPinning: %v\n", report.HairPinning)
 	printf("\t* PortMapping: %v\n", portMapping(report))
 	if report.CaptivePortal != "" {
 		printf("\t* CaptivePortal: %v\n", report.CaptivePortal)

cmd/tailscale/cli/network-lock.go

@@ -222,7 +222,8 @@ func runNetworkLockStatus(ctx context.Context, args []string) error {
 
 	if st.Enabled && st.NodeKey != nil && !st.PublicKey.IsZero() {
 		if st.NodeKeySigned {
-			fmt.Println("This node is accessible under tailnet lock.")
+			fmt.Println("This node is accessible under tailnet lock. Node signature:")
+			fmt.Println(st.NodeKeySignature.String())
 		} else {
 			fmt.Println("This node is LOCKED OUT by tailnet-lock, and action is required to establish connectivity.")
 			fmt.Printf("Run the following command on a node with a trusted key:\n\ttailscale lock sign %v %s\n", st.NodeKey, st.PublicKey.CLIString())

cmd/tailscale/cli/set.go

@@ -58,6 +58,9 @@ type setArgsT struct {
 	updateCheck            bool
 	updateApply            bool
 	postureChecking        bool
+	snat                   bool
+	statefulFiltering      bool
+	netfilterMode          string
 }
 
 func newSetFlagSet(goos string, setArgs *setArgsT) *flag.FlagSet {
@@ -98,6 +101,10 @@ func newSetFlagSet(goos string, setArgs *setArgsT) *flag.FlagSet {
 		setf.StringVar(&setArgs.opUser, "operator", "", "Unix username to allow to operate on tailscaled without sudo")
 	}
 	switch goos {
+	case "linux":
+		setf.BoolVar(&setArgs.snat, "snat-subnet-routes", true, "source NAT traffic to local routes advertised with --advertise-routes")
+		setf.BoolVar(&setArgs.statefulFiltering, "stateful-filtering", false, "apply stateful filtering to forwarded packets (subnet routers, exit nodes, etc.)")
+		setf.StringVar(&setArgs.netfilterMode, "netfilter-mode", defaultNetfilterMode(), "netfilter mode (one of on, nodivert, off)")
 	case "windows":
 		setf.BoolVar(&setArgs.forceDaemon, "unattended", false, "run in \"Unattended Mode\" where Tailscale keeps running even after the current GUI user logs out (Windows-only)")
 	}
@@ -121,6 +128,9 @@ func runSet(ctx context.Context, args []string) (retErr error) {
 		return err
 	}
 
+	// Note that even though we set the values here regardless of whether the
+	// user passed the flag, the value is only used if the user passed the flag.
+	// See updateMaskedPrefsFromUpOrSetFlag.
 	maskedPrefs := &ipn.MaskedPrefs{
 		Prefs: ipn.Prefs{
 			ProfileName:            setArgs.profileName,
@@ -132,6 +142,7 @@ func runSet(ctx context.Context, args []string) (retErr error) {
 			RunWebClient:           setArgs.runWebClient,
 			Hostname:               setArgs.hostname,
 			OperatorUser:           setArgs.opUser,
+			NoSNAT:                 !setArgs.snat,
 			ForceDaemon:            setArgs.forceDaemon,
 			AutoUpdate: ipn.AutoUpdatePrefs{
 				Check: setArgs.updateCheck,
@@ -140,10 +151,22 @@ func runSet(ctx context.Context, args []string) (retErr error) {
 			AppConnector: ipn.AppConnectorPrefs{
 				Advertise: setArgs.advertiseConnector,
 			},
-			PostureChecking: setArgs.postureChecking,
+			PostureChecking:     setArgs.postureChecking,
+			NoStatefulFiltering: opt.NewBool(!setArgs.statefulFiltering),
 		},
 	}
 
+	if effectiveGOOS() == "linux" {
+		nfMode, warning, err := netfilterModeFromFlag(setArgs.netfilterMode)
+		if err != nil {
+			return err
+		}
+		if warning != "" {
+			warnf(warning)
+		}
+		maskedPrefs.Prefs.NetfilterMode = nfMode
+	}
+
 	if setArgs.exitNodeIP != "" {
 		if err := maskedPrefs.Prefs.SetExitNodeIP(setArgs.exitNodeIP, st); err != nil {
 			var e ipn.ExitNodeLocalIPError

cmd/tailscale/cli/up.go

@@ -20,7 +20,6 @@ import (
 	"sort"
 	"strconv"
 	"strings"
-	"sync"
 	"syscall"
 	"time"
 
@@ -105,7 +104,7 @@ func newUpFlagSet(goos string, upArgs *upArgsT, cmd string) *flag.FlagSet {
 	upf.StringVar(&upArgs.server, "login-server", ipn.DefaultControlURL, "base URL of control server")
 	upf.BoolVar(&upArgs.acceptRoutes, "accept-routes", acceptRouteDefault(goos), "accept routes advertised by other Tailscale nodes")
 	upf.BoolVar(&upArgs.acceptDNS, "accept-dns", true, "accept DNS configuration from the admin panel")
-	upf.BoolVar(&upArgs.singleRoutes, "host-routes", true, hidden+"install host routes to other Tailscale nodes")
+	upf.Var(notFalseVar{}, "host-routes", hidden+"install host routes to other Tailscale nodes (must be true as of Tailscale 1.67+)")
 	upf.StringVar(&upArgs.exitNodeIP, "exit-node", "", "Tailscale exit node (IP or base name) for internet traffic, or empty string to not use an exit node")
 	upf.BoolVar(&upArgs.exitNodeAllowLANAccess, "exit-node-allow-lan-access", false, "Allow direct access to the local network when routing traffic via an exit node")
 	upf.BoolVar(&upArgs.shieldsUp, "shields-up", false, "don't allow incoming connections")
@@ -122,6 +121,7 @@ func newUpFlagSet(goos string, upArgs *upArgsT, cmd string) *flag.FlagSet {
 	switch goos {
 	case "linux":
 		upf.BoolVar(&upArgs.snat, "snat-subnet-routes", true, "source NAT traffic to local routes advertised with --advertise-routes")
+		upf.BoolVar(&upArgs.statefulFiltering, "stateful-filtering", false, "apply stateful filtering to forwarded packets (subnet routers, exit nodes, etc.)")
 		upf.StringVar(&upArgs.netfilterMode, "netfilter-mode", defaultNetfilterMode(), "netfilter mode (one of on, nodivert, off)")
 	case "windows":
 		upf.BoolVar(&upArgs.forceDaemon, "unattended", false, "run in \"Unattended Mode\" where Tailscale keeps running even after the current GUI user logs out (Windows-only)")
@@ -143,6 +143,18 @@ func newUpFlagSet(goos string, upArgs *upArgsT, cmd string) *flag.FlagSet {
 	return upf
 }
 
+// notFalseVar is is a flag.Value that can only be "true", if set.
+type notFalseVar struct{}
+
+func (notFalseVar) IsBoolFlag() bool { return true }
+func (notFalseVar) Set(v string) error {
+	if v != "true" {
+		return fmt.Errorf("unsupported value; only 'true' is allowed")
+	}
+	return nil
+}
+func (notFalseVar) String() string { return "true" }
+
 func defaultNetfilterMode() string {
 	if distro.Get() == distro.Synology {
 		return "off"
@@ -156,7 +168,6 @@ type upArgsT struct {
 	server                 string
 	acceptRoutes           bool
 	acceptDNS              bool
-	singleRoutes           bool
 	exitNodeIP             string
 	exitNodeAllowLANAccess bool
 	shieldsUp              bool
@@ -169,6 +180,7 @@ type upArgsT struct {
 	advertiseTags          string
 	advertiseConnector     bool
 	snat                   bool
+	statefulFiltering      bool
 	netfilterMode          string
 	authKeyOrFile          string // "secret" or "file:/path/to/secret"
 	hostname               string
@@ -277,7 +289,6 @@ func prefsFromUpArgs(upArgs upArgsT, warnf logger.Logf, st *ipnstate.Status, goo
 
 	prefs.ExitNodeAllowLANAccess = upArgs.exitNodeAllowLANAccess
 	prefs.CorpDNS = upArgs.acceptDNS
-	prefs.AllowSingleHosts = upArgs.singleRoutes
 	prefs.ShieldsUp = upArgs.shieldsUp
 	prefs.RunSSH = upArgs.runSSH
 	prefs.RunWebClient = upArgs.runWebClient
@@ -292,24 +303,44 @@ func prefsFromUpArgs(upArgs upArgsT, warnf logger.Logf, st *ipnstate.Status, goo
 	if goos == "linux" {
 		prefs.NoSNAT = !upArgs.snat
 
-		switch upArgs.netfilterMode {
-		case "on":
-			prefs.NetfilterMode = preftype.NetfilterOn
-		case "nodivert":
-			prefs.NetfilterMode = preftype.NetfilterNoDivert
-			warnf("netfilter=nodivert; add iptables calls to ts-* chains manually.")
-		case "off":
-			prefs.NetfilterMode = preftype.NetfilterOff
-			if defaultNetfilterMode() != "off" {
-				warnf("netfilter=off; configure iptables yourself.")
-			}
-		default:
-			return nil, fmt.Errorf("invalid value --netfilter-mode=%q", upArgs.netfilterMode)
+		// Backfills for NoStatefulFiltering occur when loading a profile; just set it explicitly here.
+		prefs.NoStatefulFiltering.Set(!upArgs.statefulFiltering)
+		v, warning, err := netfilterModeFromFlag(upArgs.netfilterMode)
+		if err != nil {
+			return nil, err
+		}
+		prefs.NetfilterMode = v
+		if warning != "" {
+			warnf(warning)
 		}
 	}
 	return prefs, nil
 }
 
+// netfilterModeFromFlag returns the preftype.NetfilterMode for the provided
+// flag value. It returns a warning if there is something the user should know
+// about the value.
+func netfilterModeFromFlag(v string) (_ preftype.NetfilterMode, warning string, _ error) {
+	switch v {
+	case "on", "nodivert", "off":
+	default:
+		return preftype.NetfilterOn, "", fmt.Errorf("invalid value --netfilter-mode=%q", v)
+	}
+	m, err := preftype.ParseNetfilterMode(v)
+	if err != nil {
+		return preftype.NetfilterOn, "", err
+	}
+	switch m {
+	case preftype.NetfilterNoDivert:
+		warning = "netfilter=nodivert; add iptables calls to ts-* chains manually."
+	case preftype.NetfilterOff:
+		if defaultNetfilterMode() != "off" {
+			warning = "netfilter=off; configure iptables yourself."
+		}
+	}
+	return m, warning, nil
+}
+
 // updatePrefs returns how to edit preferences based on the
 // flag-provided 'prefs' and the currently active 'curPrefs'.
 //
@@ -406,6 +437,11 @@ func runUp(ctx context.Context, cmd string, args []string, upArgs upArgsT) (retE
 	// printAuthURL reports whether we should print out the
 	// provided auth URL from an IPN notify.
 	printAuthURL := func(url string) bool {
+		if url == "" {
+			// Probably unnecessary but we used to have a bug where tailscaled
+			// could send an empty URL over the IPN bus. ~Harmless to keep.
+			return false
+		}
 		if upArgs.authKeyOrFile != "" {
 			// Issue 1755: when using an authkey, don't
 			// show an authURL that might still be pending
@@ -477,11 +513,6 @@ func runUp(ctx context.Context, cmd string, args []string, upArgs upArgsT) (retE
 
 	watchCtx, cancelWatch := context.WithCancel(ctx)
 	defer cancelWatch()
-	watcher, err := localClient.WatchIPNBus(watchCtx, 0)
-	if err != nil {
-		return err
-	}
-	defer watcher.Close()
 
 	go func() {
 		interrupt := make(chan os.Signal, 1)
@@ -494,29 +525,62 @@ func runUp(ctx context.Context, cmd string, args []string, upArgs upArgsT) (retE
 	}()
 
 	running := make(chan bool, 1) // gets value once in state ipn.Running
-	pumpErr := make(chan error, 1)
+	watchErr := make(chan error, 1)
 
-	// localAPIMu should be held while doing mutable LocalAPI calls
-	// to the backend. In particular, it prevents StartLoginInteractive from
-	// being called from the watcher goroutine while the Start call from
-	// the other goroutine is in progress.
-	// See https://github.com/tailscale/tailscale/issues/7036#issuecomment-2053771466
-	// TODO(bradfitz): simplify this once #11649 is cleaned up and Start is
-	// hopefully removed.
-	var localAPIMu sync.Mutex
+	// Special case: bare "tailscale up" means to just start
+	// running, if there's ever been a login.
+	if simpleUp {
+		_, err := localClient.EditPrefs(ctx, &ipn.MaskedPrefs{
+			Prefs: ipn.Prefs{
+				WantRunning: true,
+			},
+			WantRunningSet: true,
+		})
+		if err != nil {
+			return err
+		}
+	} else {
+		if err := localClient.CheckPrefs(ctx, prefs); err != nil {
+			return err
+		}
 
-	startLoginInteractive := sync.OnceFunc(func() {
-		localAPIMu.Lock()
-		defer localAPIMu.Unlock()
-		localClient.StartLoginInteractive(ctx)
-	})
+		authKey, err := upArgs.getAuthKey()
+		if err != nil {
+			return err
+		}
+		authKey, err = resolveAuthKey(ctx, authKey, upArgs.advertiseTags)
+		if err != nil {
+			return err
+		}
+		err = localClient.Start(ctx, ipn.Options{
+			AuthKey:     authKey,
+			UpdatePrefs: prefs,
+		})
+		if err != nil {
+			return err
+		}
+		if upArgs.forceReauth || !st.HaveNodeKey {
+			err := localClient.StartLoginInteractive(ctx)
+			if err != nil {
+				return err
+			}
+		}
+	}
+
+	watcher, err := localClient.WatchIPNBus(watchCtx, ipn.NotifyInitialState)
+	if err != nil {
+		return err
+	}
+	defer watcher.Close()
 
 	go func() {
 		var printed bool // whether we've yet printed anything to stdout or stderr
+		var lastURLPrinted string
+
 		for {
 			n, err := watcher.Next()
 			if err != nil {
-				pumpErr <- err
+				watchErr <- err
 				return
 			}
 			if n.ErrMessage != nil {
@@ -525,8 +589,6 @@ func runUp(ctx context.Context, cmd string, args []string, upArgs upArgsT) (retE
 			}
 			if s := n.State; s != nil {
 				switch *s {
-				case ipn.NeedsLogin:
-					startLoginInteractive()
 				case ipn.NeedsMachineAuth:
 					printed = true
 					if env.upArgs.json {
@@ -549,12 +611,17 @@ func runUp(ctx context.Context, cmd string, args []string, upArgs upArgsT) (retE
 					cancelWatch()
 				}
 			}
-			if url := n.BrowseToURL; url != nil && printAuthURL(*url) {
+			if url := n.BrowseToURL; url != nil {
+				authURL := *url
+				if !printAuthURL(authURL) || authURL == lastURLPrinted {
+					continue
+				}
 				printed = true
+				lastURLPrinted = authURL
 				if upArgs.json {
-					js := &upOutputJSON{AuthURL: *url, BackendState: st.BackendState}
+					js := &upOutputJSON{AuthURL: authURL, BackendState: st.BackendState}
 
-					q, err := qrcode.New(*url, qrcode.Medium)
+					q, err := qrcode.New(authURL, qrcode.Medium)
 					if err == nil {
 						png, err := q.PNG(128)
 						if err == nil {
@@ -569,9 +636,9 @@ func runUp(ctx context.Context, cmd string, args []string, upArgs upArgsT) (retE
 						outln(string(data))
 					}
 				} else {
-					fmt.Fprintf(Stderr, "\nTo authenticate, visit:\n\n\t%s\n\n", *url)
+					fmt.Fprintf(Stderr, "\nTo authenticate, visit:\n\n\t%s\n\n", authURL)
 					if upArgs.qr {
-						q, err := qrcode.New(*url, qrcode.Medium)
+						q, err := qrcode.New(authURL, qrcode.Medium)
 						if err != nil {
 							log.Printf("QR code error: %v", err)
 						} else {
@@ -583,47 +650,6 @@ func runUp(ctx context.Context, cmd string, args []string, upArgs upArgsT) (retE
 		}
 	}()
 
-	// Special case: bare "tailscale up" means to just start
-	// running, if there's ever been a login.
-	if simpleUp {
-		localAPIMu.Lock()
-		_, err := localClient.EditPrefs(ctx, &ipn.MaskedPrefs{
-			Prefs: ipn.Prefs{
-				WantRunning: true,
-			},
-			WantRunningSet: true,
-		})
-		localAPIMu.Unlock()
-		if err != nil {
-			return err
-		}
-	} else {
-		if err := localClient.CheckPrefs(ctx, prefs); err != nil {
-			return err
-		}
-
-		authKey, err := upArgs.getAuthKey()
-		if err != nil {
-			return err
-		}
-		authKey, err = resolveAuthKey(ctx, authKey, upArgs.advertiseTags)
-		if err != nil {
-			return err
-		}
-		localAPIMu.Lock()
-		err = localClient.Start(ctx, ipn.Options{
-			AuthKey:     authKey,
-			UpdatePrefs: prefs,
-		})
-		localAPIMu.Unlock()
-		if err != nil {
-			return err
-		}
-		if upArgs.forceReauth {
-			startLoginInteractive()
-		}
-	}
-
 	// This whole 'up' mechanism is too complicated and results in
 	// hairy stuff like this select. We're ultimately waiting for
 	// 'running' to be done, but even in the case where
@@ -647,7 +673,7 @@ func runUp(ctx context.Context, cmd string, args []string, upArgs upArgsT) (retE
 		default:
 		}
 		return watchCtx.Err()
-	case err := <-pumpErr:
+	case err := <-watchErr:
 		select {
 		case <-running:
 			return nil
@@ -724,12 +750,12 @@ func init() {
 	addPrefFlagMapping("accept-dns", "CorpDNS")
 	addPrefFlagMapping("accept-routes", "RouteAll")
 	addPrefFlagMapping("advertise-tags", "AdvertiseTags")
-	addPrefFlagMapping("host-routes", "AllowSingleHosts")
 	addPrefFlagMapping("hostname", "Hostname")
 	addPrefFlagMapping("login-server", "ControlURL")
 	addPrefFlagMapping("netfilter-mode", "NetfilterMode")
 	addPrefFlagMapping("shields-up", "ShieldsUp")
 	addPrefFlagMapping("snat-subnet-routes", "NoSNAT")
+	addPrefFlagMapping("stateful-filtering", "NoStatefulFiltering")
 	addPrefFlagMapping("exit-node-allow-lan-access", "ExitNodeAllowLANAccess")
 	addPrefFlagMapping("unattended", "ForceDaemon")
 	addPrefFlagMapping("operator", "OperatorUser")
@@ -762,7 +788,7 @@ func addPrefFlagMapping(flagName string, prefNames ...string) {
 // correspond to an ipn.Pref.
 func preflessFlag(flagName string) bool {
 	switch flagName {
-	case "auth-key", "force-reauth", "reset", "qr", "json", "timeout", "accept-risk":
+	case "auth-key", "force-reauth", "reset", "qr", "json", "timeout", "accept-risk", "host-routes":
 		return true
 	}
 	return false
@@ -859,11 +885,26 @@ func checkForAccidentalSettingReverts(newPrefs, curPrefs *ipn.Prefs, env upCheck
 			// Issue 6811. Ignore on Synology.
 			continue
 		}
+		if flagName == "stateful-filtering" && valCur == true && valNew == false && env.goos == "linux" {
+			// See https://github.com/tailscale/tailscale/issues/12307
+			// Stateful filtering was on by default in tailscale 1.66.0-1.66.3, then off in 1.66.4.
+			// This broke Tailscale installations in containerized
+			// environments that use the default containerboot
+			// configuration that configures tailscale using
+			// 'tailscale up' command, which requires that all
+			// previously set flags are explicitly provided on
+			// subsequent restarts.
+			continue
+		}
 		missing = append(missing, fmtFlagValueArg(flagName, valCur))
 	}
 	if len(missing) == 0 {
 		return nil
 	}
+
+	// Some previously provided flags are missing. This run of 'tailscale
+	// up' will error out.
+
 	sort.Strings(missing)
 
 	// Compute the stringification of the explicitly provided args in flagSet
@@ -914,7 +955,7 @@ func applyImplicitPrefs(prefs, oldPrefs *ipn.Prefs, env upCheckEnv) {
 
 func flagAppliesToOS(flag, goos string) bool {
 	switch flag {
-	case "netfilter-mode", "snat-subnet-routes":
+	case "netfilter-mode", "snat-subnet-routes", "stateful-filtering":
 		return goos == "linux"
 	case "unattended":
 		return goos == "windows"
@@ -958,8 +999,6 @@ func prefsToFlags(env upCheckEnv, prefs *ipn.Prefs) (flagVal map[string]any) {
 			set(prefs.ControlURL)
 		case "accept-routes":
 			set(prefs.RouteAll)
-		case "host-routes":
-			set(prefs.AllowSingleHosts)
 		case "accept-dns":
 			set(prefs.CorpDNS)
 		case "shields-up":
@@ -989,6 +1028,16 @@ func prefsToFlags(env upCheckEnv, prefs *ipn.Prefs) (flagVal map[string]any) {
 			set(prefs.AppConnector.Advertise)
 		case "snat-subnet-routes":
 			set(!prefs.NoSNAT)
+		case "stateful-filtering":
+			// We only set the stateful-filtering flag to false if
+			// the pref (negated!) is explicitly set to true; unset
+			// or false is treated as enabled.
+			val, ok := prefs.NoStatefulFiltering.Get()
+			if ok && val {
+				set(false)
+			} else {
+				set(true)
+			}
 		case "netfilter-mode":
 			set(prefs.NetfilterMode.String())
 		case "unattended":

cmd/tailscale/depaware.txt

@@ -299,7 +299,6 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         regexp                                                       from github.com/coreos/go-iptables/iptables+
         regexp/syntax                                                from regexp
         runtime/debug                                                from nhooyr.io/websocket/internal/xsync+
-        runtime/trace                                                from testing
         slices                                                       from tailscale.com/client/web+
         sort                                                         from archive/tar+
         strconv                                                      from archive/tar+
@@ -307,7 +306,6 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         sync                                                         from archive/tar+
         sync/atomic                                                  from context+
         syscall                                                      from archive/tar+
-        testing                                                      from tailscale.com/util/syspolicy
         text/tabwriter                                               from github.com/peterbourgon/ff/v3/ffcli+
         text/template                                                from html/template
         text/template/parse                                          from html/template+

cmd/tailscale/tailscale_test.go

@@ -12,6 +12,7 @@ import (
 func TestDeps(t *testing.T) {
 	deptest.DepChecker{
 		BadDeps: map[string]string{
+			"testing":                            "do not use testing package in production code",
 			"gvisor.dev/gvisor/pkg/buffer":       "https://github.com/tailscale/tailscale/issues/9756",
 			"gvisor.dev/gvisor/pkg/cpuid":        "https://github.com/tailscale/tailscale/issues/9756",
 			"gvisor.dev/gvisor/pkg/tcpip":        "https://github.com/tailscale/tailscale/issues/9756",

cmd/tailscaled/depaware.txt

@@ -89,7 +89,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
   LW 💣 github.com/digitalocean/go-smbios/smbios                     from tailscale.com/posture
      💣 github.com/djherbis/times                                    from tailscale.com/drive/driveimpl
         github.com/fxamacker/cbor/v2                                 from tailscale.com/tka
-        github.com/gaissmai/bart                                     from tailscale.com/net/tstun
+        github.com/gaissmai/bart                                     from tailscale.com/net/tstun+
         github.com/go-json-experiment/json/internal                  from github.com/go-json-experiment/json/internal/jsonflags+
         github.com/go-json-experiment/json/internal/jsonflags        from github.com/go-json-experiment/json/internal/jsonopts+
         github.com/go-json-experiment/json/internal/jsonopts         from github.com/go-json-experiment/json/jsontext
@@ -144,6 +144,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
    L    github.com/pierrec/lz4/v4/internal/xxh32                     from github.com/pierrec/lz4/v4/internal/lz4stream
   LD    github.com/pkg/sftp                                          from tailscale.com/ssh/tailssh
   LD    github.com/pkg/sftp/internal/encoding/ssh/filexfer           from github.com/pkg/sftp
+   D    github.com/prometheus-community/pro-bing                     from tailscale.com/wgengine/netstack
    L 💣 github.com/safchain/ethtool                                  from tailscale.com/net/netkernelconf+
    W 💣 github.com/tailscale/certstore                               from tailscale.com/control/controlclient
    W 💣 github.com/tailscale/go-winio                                from tailscale.com/safesocket
@@ -319,6 +320,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
      💣 tailscale.com/net/tshttpproxy                                from tailscale.com/clientupdate/distsign+
         tailscale.com/net/tstun                                      from tailscale.com/cmd/tailscaled+
         tailscale.com/net/wsconn                                     from tailscale.com/control/controlhttp+
+        tailscale.com/omit                                           from tailscale.com/ipn/conffile
         tailscale.com/paths                                          from tailscale.com/client/tailscale+
      💣 tailscale.com/portlist                                       from tailscale.com/ipn/ipnlocal
         tailscale.com/posture                                        from tailscale.com/ipn/ipnlocal
@@ -439,7 +441,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         golang.org/x/net/http2                                       from golang.org/x/net/http2/h2c+
         golang.org/x/net/http2/h2c                                   from tailscale.com/ipn/ipnlocal
         golang.org/x/net/http2/hpack                                 from golang.org/x/net/http2+
-        golang.org/x/net/icmp                                        from tailscale.com/net/ping
+        golang.org/x/net/icmp                                        from tailscale.com/net/ping+
         golang.org/x/net/idna                                        from golang.org/x/net/http/httpguts+
         golang.org/x/net/ipv4                                        from github.com/miekg/dns+
         golang.org/x/net/ipv6                                        from github.com/miekg/dns+
@@ -552,7 +554,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         regexp/syntax                                                from regexp
         runtime/debug                                                from github.com/aws/aws-sdk-go-v2/internal/sync/singleflight+
         runtime/pprof                                                from net/http/pprof+
-        runtime/trace                                                from net/http/pprof+
+        runtime/trace                                                from net/http/pprof
         slices                                                       from tailscale.com/appc+
         sort                                                         from archive/tar+
         strconv                                                      from archive/tar+
@@ -560,7 +562,6 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         sync                                                         from archive/tar+
         sync/atomic                                                  from context+
         syscall                                                      from archive/tar+
-        testing                                                      from tailscale.com/util/syspolicy
         text/tabwriter                                               from runtime/pprof
         text/template                                                from html/template
         text/template/parse                                          from html/template+

cmd/tailscaled/tailscaled.go

@@ -35,6 +35,7 @@ import (
 	"tailscale.com/control/controlclient"
 	"tailscale.com/drive/driveimpl"
 	"tailscale.com/envknob"
+	"tailscale.com/ipn"
 	"tailscale.com/ipn/conffile"
 	"tailscale.com/ipn/ipnlocal"
 	"tailscale.com/ipn/ipnserver"
@@ -117,7 +118,7 @@ var args struct {
 	tunname string
 
 	cleanUp        bool
-	confFile       string
+	confFile       string // empty, file path, or "vm:user-data"
 	debug          string
 	port           uint16
 	statepath      string
@@ -168,7 +169,7 @@ func main() {
 	flag.StringVar(&args.birdSocketPath, "bird-socket", "", "path of the bird unix socket")
 	flag.BoolVar(&printVersion, "version", false, "print version information and exit")
 	flag.BoolVar(&args.disableLogs, "no-logs-no-support", false, "disable log uploads; this also disables any technical support")
-	flag.StringVar(&args.confFile, "config", "", "path to config file")
+	flag.StringVar(&args.confFile, "config", "", "path to config file, or 'vm:user-data' to use the VM's user-data (EC2)")
 
 	if len(os.Args) > 0 && filepath.Base(os.Args[0]) == "tailscale" && beCLI != nil {
 		beCLI()
@@ -480,6 +481,15 @@ func startIPNServer(ctx context.Context, logf logger.Logf, logID logid.PublicID,
 		lb, err := getLocalBackend(ctx, logf, logID, sys)
 		if err == nil {
 			logf("got LocalBackend in %v", time.Since(t0).Round(time.Millisecond))
+			if lb.Prefs().Valid() {
+				if err := lb.Start(ipn.Options{}); err != nil {
+					logf("LocalBackend.Start: %v", err)
+					lb.Shutdown()
+					lbErr.Store(err)
+					cancel()
+					return
+				}
+			}
 			srv.SetLocalBackend(lb)
 			close(wgEngineCreated)
 			return
@@ -538,14 +548,25 @@ func getLocalBackend(ctx context.Context, logf logger.Logf, logID logid.PublicID
 			return ok
 		}
 		dialer.NetstackDialTCP = func(ctx context.Context, dst netip.AddrPort) (net.Conn, error) {
-			// Note: don't just return ns.DialContextTCP or we'll
-			// return an interface containing a nil pointer.
+			// Note: don't just return ns.DialContextTCP or we'll return
+			// *gonet.TCPConn(nil) instead of a nil interface which trips up
+			// callers.
 			tcpConn, err := ns.DialContextTCP(ctx, dst)
 			if err != nil {
 				return nil, err
 			}
 			return tcpConn, nil
 		}
+		dialer.NetstackDialUDP = func(ctx context.Context, dst netip.AddrPort) (net.Conn, error) {
+			// Note: don't just return ns.DialContextUDP or we'll return
+			// *gonet.UDPConn(nil) instead of a nil interface which trips up
+			// callers.
+			udpConn, err := ns.DialContextUDP(ctx, dst)
+			if err != nil {
+				return nil, err
+			}
+			return udpConn, nil
+		}
 	}
 	if socksListener != nil || httpProxyListener != nil {
 		var addrs []string

cmd/tailscaled/tailscaled_test.go

@@ -20,6 +20,7 @@ func TestDeps(t *testing.T) {
 		GOOS:   "darwin",
 		GOARCH: "arm64",
 		BadDeps: map[string]string{
+			"testing":                        "do not use testing package in production code",
 			"gvisor.dev/gvisor/pkg/hostarch": "will crash on non-4K page sizes; see https://github.com/tailscale/tailscale/issues/8658",
 		},
 	}.Check(t)
@@ -28,6 +29,7 @@ func TestDeps(t *testing.T) {
 		GOOS:   "linux",
 		GOARCH: "arm64",
 		BadDeps: map[string]string{
+			"testing":                        "do not use testing package in production code",
 			"gvisor.dev/gvisor/pkg/hostarch": "will crash on non-4K page sizes; see https://github.com/tailscale/tailscale/issues/8658",
 		},
 	}.Check(t)

cmd/tsconnect/wasm/wasm_js.go

@@ -298,11 +298,10 @@ func (i *jsIPN) run(jsCallbacks js.Value) {
 	go func() {
 		err := i.lb.Start(ipn.Options{
 			UpdatePrefs: &ipn.Prefs{
-				ControlURL:       i.controlURL,
-				RouteAll:         false,
-				AllowSingleHosts: true,
-				WantRunning:      true,
-				Hostname:         i.hostname,
+				ControlURL:  i.controlURL,
+				RouteAll:    false,
+				WantRunning: true,
+				Hostname:    i.hostname,
 			},
 			AuthKey: i.authKey,
 		})

cmd/tsidp/tsidp.go

@@ -40,7 +40,6 @@ import (
 	"tailscale.com/tsnet"
 	"tailscale.com/types/key"
 	"tailscale.com/types/lazy"
-	"tailscale.com/types/logger"
 	"tailscale.com/types/views"
 	"tailscale.com/util/mak"
 	"tailscale.com/util/must"
@@ -95,8 +94,8 @@ func main() {
 		ts := &tsnet.Server{
 			Hostname: "idp",
 		}
-		if !*flagVerbose {
-			ts.Logf = logger.Discard
+		if *flagVerbose {
+			ts.Logf = log.Printf
 		}
 		st, err = ts.Up(ctx)
 		if err != nil {

control/controlclient/auto.go

@@ -26,9 +26,8 @@ import (
 
 type LoginGoal struct {
 	_     structs.Incomparable
-	token *tailcfg.Oauth2Token // oauth token to use when logging in
-	flags LoginFlags           // flags to use when logging in
-	url   string               // auth url that needs to be visited
+	flags LoginFlags // flags to use when logging in
+	url   string     // auth url that needs to be visited
 }
 
 var _ Client = (*Auto)(nil)
@@ -338,7 +337,7 @@ func (c *Auto) authRoutine() {
 			url, err = c.direct.WaitLoginURL(ctx, goal.url)
 			f = "WaitLoginURL"
 		} else {
-			url, err = c.direct.TryLogin(ctx, goal.token, goal.flags)
+			url, err = c.direct.TryLogin(ctx, goal.flags)
 			f = "TryLogin"
 		}
 		if err != nil {
@@ -348,9 +347,14 @@ func (c *Auto) authRoutine() {
 			continue
 		}
 		if url != "" {
-			// goal.url ought to be empty here.
-			// However, not all control servers get this right,
-			// and logging about it here just generates noise.
+			// goal.url ought to be empty here. However, not all control servers
+			// get this right, and logging about it here just generates noise.
+			//
+			// TODO(bradfitz): I don't follow that comment. Our own testcontrol
+			// used by tstest/integration hits this path, in fact.
+			if c.direct.panicOnUse {
+				panic("tainted client")
+			}
 			c.mu.Lock()
 			c.urlToVisit = url
 			c.loginGoal = &LoginGoal{
@@ -607,17 +611,19 @@ func (c *Auto) sendStatus(who string, err error, url string, nm *netmap.NetworkM
 	})
 }
 
-func (c *Auto) Login(t *tailcfg.Oauth2Token, flags LoginFlags) {
-	c.logf("client.Login(%v, %v)", t != nil, flags)
+func (c *Auto) Login(flags LoginFlags) {
+	c.logf("client.Login(%v)", flags)
 
 	c.mu.Lock()
 	defer c.mu.Unlock()
 	if c.closed {
 		return
 	}
+	if c.direct != nil && c.direct.panicOnUse {
+		panic("tainted client")
+	}
 	c.wantLoggedIn = true
 	c.loginGoal = &LoginGoal{
-		token: t,
 		flags: flags,
 	}
 	c.cancelMapCtxLocked()
@@ -632,6 +638,9 @@ func (c *Auto) Logout(ctx context.Context) error {
 	c.wantLoggedIn = false
 	c.loginGoal = nil
 	closed := c.closed
+	if c.direct != nil && c.direct.panicOnUse {
+		panic("tainted client")
+	}
 	c.mu.Unlock()
 
 	if closed {
@@ -668,37 +677,35 @@ func (c *Auto) UpdateEndpoints(endpoints []tailcfg.Endpoint) {
 }
 
 func (c *Auto) Shutdown() {
-	c.logf("client.Shutdown()")
-
 	c.mu.Lock()
-	closed := c.closed
-	direct := c.direct
-	if !closed {
-		c.closed = true
-		c.observerQueue.Shutdown()
-		c.cancelAuthCtxLocked()
-		c.cancelMapCtxLocked()
-		for _, w := range c.unpauseWaiters {
-			w <- false
-		}
-		c.unpauseWaiters = nil
+	if c.closed {
+		c.mu.Unlock()
+		return
 	}
+	c.logf("client.Shutdown ...")
+
+	direct := c.direct
+	c.closed = true
+	c.observerQueue.Shutdown()
+	c.cancelAuthCtxLocked()
+	c.cancelMapCtxLocked()
+	for _, w := range c.unpauseWaiters {
+		w <- false
+	}
+	c.unpauseWaiters = nil
 	c.mu.Unlock()
 
-	c.logf("client.Shutdown")
-	if !closed {
-		c.unregisterHealthWatch()
-		<-c.authDone
-		<-c.mapDone
-		<-c.updateDone
-		if direct != nil {
-			direct.Close()
-		}
-		ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-		defer cancel()
-		c.observerQueue.Wait(ctx)
-		c.logf("Client.Shutdown done.")
+	c.unregisterHealthWatch()
+	<-c.authDone
+	<-c.mapDone
+	<-c.updateDone
+	if direct != nil {
+		direct.Close()
 	}
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	c.observerQueue.Wait(ctx)
+	c.logf("Client.Shutdown done.")
 }
 
 // NodePublicKey returns the node public key currently in use. This is

control/controlclient/client.go

@@ -43,8 +43,9 @@ type Client interface {
 	// Login begins an interactive or non-interactive login process.
 	// Client will eventually call the Status callback with either a
 	// LoginFinished flag (on success) or an auth URL (if further
-	// interaction is needed).
-	Login(*tailcfg.Oauth2Token, LoginFlags)
+	// interaction is needed). It merely sets the process in motion,
+	// and doesn't wait for it to complete.
+	Login(LoginFlags)
 	// Logout starts a synchronous logout process. It doesn't return
 	// until the logout operation has been completed.
 	Logout(context.Context) error

control/controlclient/direct.go

@@ -80,6 +80,7 @@ type Direct struct {
 	onClientVersion            func(*tailcfg.ClientVersion) // or nil
 	onControlTime              func(time.Time)              // or nil
 	onTailnetDefaultAutoUpdate func(bool)                   // or nil
+	panicOnUse                 bool                         // if true, panic if client is used (for testing)
 
 	dialPlan ControlDialPlanner // can be nil
 
@@ -310,6 +311,9 @@ func NewDirect(opts Options) (*Direct, error) {
 		}
 		c.serverNoiseKey = key.NewMachine().Public() // prevent early error before hitting test client
 	}
+	if strings.Contains(opts.ServerURL, "controlplane.tailscale.com") && envknob.Bool("TS_PANIC_IF_HIT_MAIN_CONTROL") {
+		c.panicOnUse = true
+	}
 	return c, nil
 }
 
@@ -397,9 +401,12 @@ func (c *Direct) TryLogout(ctx context.Context) error {
 	return err
 }
 
-func (c *Direct) TryLogin(ctx context.Context, t *tailcfg.Oauth2Token, flags LoginFlags) (url string, err error) {
-	c.logf("[v1] direct.TryLogin(token=%v, flags=%v)", t != nil, flags)
-	return c.doLoginOrRegen(ctx, loginOpt{Token: t, Flags: flags})
+func (c *Direct) TryLogin(ctx context.Context, flags LoginFlags) (url string, err error) {
+	if strings.Contains(c.serverURL, "controlplane.tailscale.com") && envknob.Bool("TS_PANIC_IF_HIT_MAIN_CONTROL") {
+		panic(fmt.Sprintf("[unexpected] controlclient: TryLogin called on %s; tainted=%v", c.serverURL, c.panicOnUse))
+	}
+	c.logf("[v1] direct.TryLogin(flags=%v)", flags)
+	return c.doLoginOrRegen(ctx, loginOpt{Flags: flags})
 }
 
 // WaitLoginURL sits in a long poll waiting for the user to authenticate at url.
@@ -434,7 +441,6 @@ func (c *Direct) SetExpirySooner(ctx context.Context, expiry time.Time) error {
 }
 
 type loginOpt struct {
-	Token  *tailcfg.Oauth2Token
 	Flags  LoginFlags
 	Regen  bool // generate a new nodekey, can be overridden in doLogin
 	URL    string
@@ -459,6 +465,9 @@ func (c *Direct) hostInfoLocked() *tailcfg.Hostinfo {
 }
 
 func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, newURL string, nks tkatype.MarshaledSignature, err error) {
+	if c.panicOnUse {
+		panic("tainted client")
+	}
 	c.mu.Lock()
 	persist := c.persist.AsStruct()
 	tryingNewKey := c.tryingNewKey
@@ -549,7 +558,7 @@ func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, new
 
 	var nodeKeySignature tkatype.MarshaledSignature
 	if !oldNodeKey.IsZero() && opt.OldNodeKeySignature != nil {
-		if nodeKeySignature, err = resignNKS(persist.NetworkLockKey, tryingNewKey.Public(), opt.OldNodeKeySignature); err != nil {
+		if nodeKeySignature, err = tka.ResignNKS(persist.NetworkLockKey, tryingNewKey.Public(), opt.OldNodeKeySignature); err != nil {
 			c.logf("Failed re-signing node-key signature: %v", err)
 		}
 	} else if isWrapped {
@@ -600,10 +609,9 @@ func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, new
 	c.logf("RegisterReq: onode=%v node=%v fup=%v nks=%v",
 		request.OldNodeKey.ShortString(),
 		request.NodeKey.ShortString(), opt.URL != "", len(nodeKeySignature) > 0)
-	if opt.Token != nil || authKey != "" {
+	if authKey != "" {
 		request.Auth = &tailcfg.RegisterResponseAuth{
-			Oauth2Token: opt.Token,
-			AuthKey:     authKey,
+			AuthKey: authKey,
 		}
 	}
 	err = signRegisterRequest(&request, c.serverURL, c.serverLegacyKey, machinePrivKey.Public())
@@ -721,45 +729,6 @@ func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, new
 	return false, resp.AuthURL, nil, nil
 }
 
-// resignNKS re-signs a node-key signature for a new node-key.
-//
-// This only matters on network-locked tailnets, because node-key signatures are
-// how other nodes know that a node-key is authentic. When the node-key is
-// rotated then the existing signature becomes invalid, so this function is
-// responsible for generating a new wrapping signature to certify the new node-key.
-//
-// The signature itself is a SigRotation signature, which embeds the old signature
-// and certifies the new node-key as a replacement for the old by signing the new
-// signature with RotationPubkey (which is the node's own network-lock key).
-func resignNKS(priv key.NLPrivate, nodeKey key.NodePublic, oldNKS tkatype.MarshaledSignature) (tkatype.MarshaledSignature, error) {
-	var oldSig tka.NodeKeySignature
-	if err := oldSig.Unserialize(oldNKS); err != nil {
-		return nil, fmt.Errorf("decoding NKS: %w", err)
-	}
-
-	nk, err := nodeKey.MarshalBinary()
-	if err != nil {
-		return nil, fmt.Errorf("marshalling node-key: %w", err)
-	}
-
-	if bytes.Equal(nk, oldSig.Pubkey) {
-		// The old signature is valid for the node-key we are using, so just
-		// use it verbatim.
-		return oldNKS, nil
-	}
-
-	newSig := tka.NodeKeySignature{
-		SigKind: tka.SigRotation,
-		Pubkey:  nk,
-		Nested:  &oldSig,
-	}
-	if newSig.Signature, err = priv.SignNKS(newSig.SigHash()); err != nil {
-		return nil, fmt.Errorf("signing NKS: %w", err)
-	}
-
-	return newSig.Serialize(), nil
-}
-
 // newEndpoints acquires c.mu and sets the local port and endpoints and reports
 // whether they've changed.
 //
@@ -832,6 +801,9 @@ const watchdogTimeout = 120 * time.Second
 //
 // If nu is nil, OmitPeers will be set to true.
 func (c *Direct) sendMapRequest(ctx context.Context, isStreaming bool, nu NetmapUpdater) error {
+	if c.panicOnUse {
+		panic("tainted client")
+	}
 	if isStreaming && nu == nil {
 		panic("cb must be non-nil if isStreaming is true")
 	}
@@ -1528,6 +1500,9 @@ func (c *Direct) SetDNS(ctx context.Context, req *tailcfg.SetDNSRequest) (err er
 }
 
 func (c *Direct) DoNoiseRequest(req *http.Request) (*http.Response, error) {
+	if c.panicOnUse {
+		panic("tainted client")
+	}
 	nc, err := c.getNoiseClient()
 	if err != nil {
 		return nil, err
@@ -1624,6 +1599,9 @@ func (c *Direct) ReportHealthChange(sys health.Subsystem, sysErr error) {
 	if !ok {
 		return
 	}
+	if c.panicOnUse {
+		panic("tainted client")
+	}
 	req := &tailcfg.HealthChangeRequest{
 		Subsys:  string(sys),
 		NodeKey: nodeKey,

control/controlclient/map.go

@@ -736,6 +736,10 @@ func peerChangeDiff(was tailcfg.NodeView, n *tailcfg.Node) (_ *tailcfg.PeerChang
 			if was.IsWireGuardOnly() != n.IsWireGuardOnly {
 				return nil, false
 			}
+		case "IsJailed":
+			if was.IsJailed() != n.IsJailed {
+				return nil, false
+			}
 		case "Expired":
 			if was.Expired() != n.Expired {
 				return nil, false

control/controlknobs/controlknobs.go

@@ -76,6 +76,11 @@ type Knobs struct {
 	// AppCStoreRoutes is whether the node should store RouteInfo to StateStore
 	// if it's an app connector.
 	AppCStoreRoutes atomic.Bool
+
+	// UserDialUseRoutes is whether tsdial.Dialer.UserDial should use routes to determine
+	// how to dial the destination address. When true, it also makes the DNS forwarder
+	// use UserDial instead of SystemDial when dialing resolvers.
+	UserDialUseRoutes atomic.Bool
 }
 
 // UpdateFromNodeAttributes updates k (if non-nil) based on the provided self
@@ -101,6 +106,7 @@ func (k *Knobs) UpdateFromNodeAttributes(capMap tailcfg.NodeCapMap) {
 		seamlessKeyRenewal            = has(tailcfg.NodeAttrSeamlessKeyRenewal)
 		probeUDPLifetime              = has(tailcfg.NodeAttrProbeUDPLifetime)
 		appCStoreRoutes               = has(tailcfg.NodeAttrStoreAppCRoutes)
+		userDialUseRoutes             = has(tailcfg.NodeAttrUserDialUseRoutes)
 	)
 
 	if has(tailcfg.NodeAttrOneCGNATEnable) {
@@ -124,6 +130,7 @@ func (k *Knobs) UpdateFromNodeAttributes(capMap tailcfg.NodeCapMap) {
 	k.SeamlessKeyRenewal.Store(seamlessKeyRenewal)
 	k.ProbeUDPLifetime.Store(probeUDPLifetime)
 	k.AppCStoreRoutes.Store(appCStoreRoutes)
+	k.UserDialUseRoutes.Store(userDialUseRoutes)
 }
 
 // AsDebugJSON returns k as something that can be marshalled with json.Marshal
@@ -148,5 +155,6 @@ func (k *Knobs) AsDebugJSON() map[string]any {
 		"SeamlessKeyRenewal":            k.SeamlessKeyRenewal.Load(),
 		"ProbeUDPLifetime":              k.ProbeUDPLifetime.Load(),
 		"AppCStoreRoutes":               k.AppCStoreRoutes.Load(),
+		"UserDialUseRoutes":             k.UserDialUseRoutes.Load(),
 	}
 }

derp/derp_server.go

@@ -329,20 +329,36 @@ func NewServer(privateKey key.NodePrivate, logf logger.Logf) *Server {
 	s.initMetacert()
 	s.packetsRecvDisco = s.packetsRecvByKind.Get("disco")
 	s.packetsRecvOther = s.packetsRecvByKind.Get("other")
-	s.packetsDroppedReasonCounters = []*expvar.Int{
-		s.packetsDroppedReason.Get("unknown_dest"),
-		s.packetsDroppedReason.Get("unknown_dest_on_fwd"),
-		s.packetsDroppedReason.Get("gone_disconnected"),
-		s.packetsDroppedReason.Get("gone_not_here"),
-		s.packetsDroppedReason.Get("queue_head"),
-		s.packetsDroppedReason.Get("queue_tail"),
-		s.packetsDroppedReason.Get("write_error"),
-	}
+
+	s.packetsDroppedReasonCounters = s.genPacketsDroppedReasonCounters()
+
 	s.packetsDroppedTypeDisco = s.packetsDroppedType.Get("disco")
 	s.packetsDroppedTypeOther = s.packetsDroppedType.Get("other")
 	return s
 }
 
+func (s *Server) genPacketsDroppedReasonCounters() []*expvar.Int {
+	getMetric := s.packetsDroppedReason.Get
+	ret := []*expvar.Int{
+		dropReasonUnknownDest:      getMetric("unknown_dest"),
+		dropReasonUnknownDestOnFwd: getMetric("unknown_dest_on_fwd"),
+		dropReasonGoneDisconnected: getMetric("gone_disconnected"),
+		dropReasonQueueHead:        getMetric("queue_head"),
+		dropReasonQueueTail:        getMetric("queue_tail"),
+		dropReasonWriteError:       getMetric("write_error"),
+		dropReasonDupClient:        getMetric("dup_client"),
+	}
+	if len(ret) != int(numDropReasons) {
+		panic("dropReason metrics out of sync")
+	}
+	for i := range numDropReasons {
+		if ret[i] == nil {
+			panic("dropReason metrics out of sync")
+		}
+	}
+	return ret
+}
+
 // SetMesh sets the pre-shared key that regional DERP servers used to mesh
 // amongst themselves.
 //
@@ -776,7 +792,6 @@ func (c *sclient) run(ctx context.Context) error {
 	var grp errgroup.Group
 	sendCtx, cancelSender := context.WithCancel(ctx)
 	grp.Go(func() error { return c.sendLoop(sendCtx) })
-	grp.Go(func() error { return c.statsLoop(sendCtx) })
 	defer func() {
 		cancelSender()
 		if err := grp.Wait(); err != nil && !c.s.isClosed() {
@@ -788,6 +803,8 @@ func (c *sclient) run(ctx context.Context) error {
 		}
 	}()
 
+	c.startStatsLoop(sendCtx)
+
 	for {
 		ft, fl, err := readFrameHeader(c.br)
 		c.debugLogf("read frame type %d len %d err %v", ft, fl, err)
@@ -1046,6 +1063,7 @@ const (
 	dropReasonQueueTail                          // destination queue is full, dropped packet at queue tail
 	dropReasonWriteError                         // OS write() failed
 	dropReasonDupClient                          // the public key is connected 2+ times (active/active, fighting)
+	numDropReasons                               // unused; keep last
 )
 
 func (s *Server) recordDrop(packetBytes []byte, srcKey, dstKey key.NodePublic, reason dropReason) {

derp/derp_server_default.go

@@ -7,6 +7,7 @@ package derp
 
 import "context"
 
-func (c *sclient) statsLoop(ctx context.Context) error {
-	return nil
+func (c *sclient) startStatsLoop(ctx context.Context) {
+	// Nothing to do
+	return
 }

derp/derp_server_linux.go

@@ -12,40 +12,43 @@ import (
 	"tailscale.com/net/tcpinfo"
 )
 
-func (c *sclient) statsLoop(ctx context.Context) error {
+func (c *sclient) startStatsLoop(ctx context.Context) {
 	// Get the RTT initially to verify it's supported.
 	conn := c.tcpConn()
 	if conn == nil {
 		c.s.tcpRtt.Add("non-tcp", 1)
-		return nil
+		return
 	}
 	if _, err := tcpinfo.RTT(conn); err != nil {
 		c.logf("error fetching initial RTT: %v", err)
 		c.s.tcpRtt.Add("error", 1)
-		return nil
+		return
 	}
 
 	const statsInterval = 10 * time.Second
 
-	ticker, tickerChannel := c.s.clock.NewTicker(statsInterval)
-	defer ticker.Stop()
-
-statsLoop:
-	for {
-		select {
-		case <-tickerChannel:
-			rtt, err := tcpinfo.RTT(conn)
-			if err != nil {
-				continue statsLoop
-			}
-
-			// TODO(andrew): more metrics?
-			c.s.tcpRtt.Add(durationToLabel(rtt), 1)
-
-		case <-ctx.Done():
-			return ctx.Err()
+	// Don't launch a goroutine; use a timer instead.
+	var gatherStats func()
+	gatherStats = func() {
+		// Do nothing if the context is finished.
+		if ctx.Err() != nil {
+			return
 		}
+
+		// Reschedule ourselves when this stats gathering is finished.
+		defer c.s.clock.AfterFunc(statsInterval, gatherStats)
+
+		// Gather TCP RTT information.
+		rtt, err := tcpinfo.RTT(conn)
+		if err == nil {
+			c.s.tcpRtt.Add(durationToLabel(rtt), 1)
+		}
+
+		// TODO(andrew): more metrics?
 	}
+
+	// Kick off the initial timer.
+	c.s.clock.AfterFunc(statsInterval, gatherStats)
 }
 
 // tcpConn attempts to get the underlying *net.TCPConn from this client's

derp/dropreason_string.go

@@ -18,11 +18,12 @@ func _() {
 	_ = x[dropReasonQueueTail-4]
 	_ = x[dropReasonWriteError-5]
 	_ = x[dropReasonDupClient-6]
+	_ = x[numDropReasons-7]
 }
 
-const _dropReason_name = "UnknownDestUnknownDestOnFwdGoneDisconnectedQueueHeadQueueTailWriteErrorDupClient"
+const _dropReason_name = "UnknownDestUnknownDestOnFwdGoneDisconnectedQueueHeadQueueTailWriteErrorDupClientnumDropReasons"
 
-var _dropReason_index = [...]uint8{0, 11, 27, 43, 52, 61, 71, 80}
+var _dropReason_index = [...]uint8{0, 11, 27, 43, 52, 61, 71, 80, 94}
 
 func (i dropReason) String() string {
 	if i < 0 || i >= dropReason(len(_dropReason_index)-1) {

docs/k8s/subnet.yaml

@@ -29,5 +29,6 @@ spec:
     - name: TS_ROUTES
       value: "{{TS_ROUTES}}"
     securityContext:
-      runAsUser: 1000
-      runAsGroup: 1000
+      capabilities:
+        add:
+        - NET_ADMIN

drive/driveimpl/compositedav/compositedav.go

@@ -81,24 +81,39 @@ type Handler struct {
 	staticRoot string
 }
 
+var cacheInvalidatingMethods = map[string]bool{
+	"PUT":       true,
+	"POST":      true,
+	"COPY":      true,
+	"MKCOL":     true,
+	"MOVE":      true,
+	"PROPPATCH": true,
+	"DELETE":    true,
+}
+
 // ServeHTTP implements http.Handler.
 func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
-	if r.Method == "PROPFIND" {
-		h.handlePROPFIND(w, r)
+	pathComponents := shared.CleanAndSplit(r.URL.Path)
+	mpl := h.maxPathLength(r)
+
+	switch r.Method {
+	case "PROPFIND":
+		h.handlePROPFIND(w, r, pathComponents, mpl)
+		return
+	case "LOCK":
+		h.handleLOCK(w, r, pathComponents, mpl)
 		return
 	}
 
-	if r.Method != "GET" {
-		// If the user is performing a modification (e.g. PUT, MKDIR, etc),
+	_, shouldInvalidate := cacheInvalidatingMethods[r.Method]
+	if shouldInvalidate {
+		// If the user is performing a modification (e.g. PUT, MKDIR, etc.),
 		// we need to invalidate the StatCache to make sure we're not knowingly
 		// showing stale stats.
-		// TODO(oxtoacart): maybe be more selective about invalidating cache
+		// TODO(oxtoacart): maybe only invalidate specific paths
 		h.StatCache.invalidate()
 	}
 
-	mpl := h.maxPathLength(r)
-	pathComponents := shared.CleanAndSplit(r.URL.Path)
-
 	if len(pathComponents) >= mpl {
 		h.delegate(mpl, pathComponents[mpl-1:], w, r)
 		return
@@ -130,6 +145,8 @@ func (h *Handler) handle(w http.ResponseWriter, r *http.Request) {
 
 // delegate sends the request to the Child WebDAV server.
 func (h *Handler) delegate(mpl int, pathComponents []string, w http.ResponseWriter, r *http.Request) {
+	rewriteIfHeader(r, pathComponents, mpl)
+
 	dest := r.Header.Get("Destination")
 	if dest != "" {
 		// Rewrite destination header

drive/driveimpl/compositedav/propfind.go

@@ -1,77 +0,0 @@
-// Copyright (c) Tailscale Inc & AUTHORS
-// SPDX-License-Identifier: BSD-3-Clause
-
-package compositedav
-
-import (
-	"bytes"
-	"fmt"
-	"math"
-	"net/http"
-	"regexp"
-
-	"tailscale.com/drive/driveimpl/shared"
-)
-
-var (
-	hrefRegex = regexp.MustCompile(`(?s)<D:href>/?([^<]*)/?</D:href>`)
-)
-
-func (h *Handler) handlePROPFIND(w http.ResponseWriter, r *http.Request) {
-	pathComponents := shared.CleanAndSplit(r.URL.Path)
-	mpl := h.maxPathLength(r)
-	if !shared.IsRoot(r.URL.Path) && len(pathComponents)+getDepth(r) > mpl {
-		// Delegate to a Child.
-		depth := getDepth(r)
-
-		status, result := h.StatCache.getOr(r.URL.Path, depth, func() (int, []byte) {
-			// Use a buffering ResponseWriter so that we can manipulate the result.
-			// The only thing we use from the original ResponseWriter is Header().
-			bw := &bufferingResponseWriter{ResponseWriter: w}
-
-			mpl := h.maxPathLength(r)
-			h.delegate(mpl, pathComponents[mpl-1:], bw, r)
-
-			// Fixup paths to add the requested path as a prefix.
-			pathPrefix := shared.Join(pathComponents[0:mpl]...)
-			b := hrefRegex.ReplaceAll(bw.buf.Bytes(), []byte(fmt.Sprintf("<D:href>%s/$1</D:href>", pathPrefix)))
-
-			return bw.status, b
-		})
-
-		w.Header().Del("Content-Length")
-		w.WriteHeader(status)
-		if result != nil {
-			w.Write(result)
-		}
-		return
-	}
-
-	h.handle(w, r)
-}
-
-func getDepth(r *http.Request) int {
-	switch r.Header.Get("Depth") {
-	case "0":
-		return 0
-	case "1":
-		return 1
-	case "infinity":
-		return math.MaxInt
-	}
-	return 0
-}
-
-type bufferingResponseWriter struct {
-	http.ResponseWriter
-	status int
-	buf    bytes.Buffer
-}
-
-func (bw *bufferingResponseWriter) WriteHeader(statusCode int) {
-	bw.status = statusCode
-}
-
-func (bw *bufferingResponseWriter) Write(p []byte) (int, error) {
-	return bw.buf.Write(p)
-}

drive/driveimpl/compositedav/rewriting.go

@@ -0,0 +1,122 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package compositedav
+
+import (
+	"bytes"
+	"fmt"
+	"math"
+	"net/http"
+	"regexp"
+	"strings"
+
+	"tailscale.com/drive/driveimpl/shared"
+)
+
+var (
+	responseHrefRegex = regexp.MustCompile(`(?s)(<D:(response|lockroot)>)<D:href>/?([^<]*)/?</D:href>`)
+	ifHrefRegex       = regexp.MustCompile(`^<(https?://[^/]+)?([^>]+)>`)
+)
+
+func (h *Handler) handlePROPFIND(w http.ResponseWriter, r *http.Request, pathComponents []string, mpl int) {
+	if shouldDelegateToChild(r, pathComponents, mpl) {
+		// Delegate to a Child.
+		depth := getDepth(r)
+
+		status, result := h.StatCache.getOr(r.URL.Path, depth, func() (int, []byte) {
+			return h.delegateRewriting(w, r, pathComponents, mpl)
+		})
+
+		respondRewritten(w, status, result)
+		return
+	}
+
+	h.handle(w, r)
+}
+
+func (h *Handler) handleLOCK(w http.ResponseWriter, r *http.Request, pathComponents []string, mpl int) {
+	if shouldDelegateToChild(r, pathComponents, mpl) {
+		// Delegate to a Child.
+		status, result := h.delegateRewriting(w, r, pathComponents, mpl)
+		respondRewritten(w, status, result)
+		return
+	}
+
+	http.Error(w, "locking of top level directories is not allowed", http.StatusMethodNotAllowed)
+}
+
+// shouldDelegateToChild decides whether a request should be delegated to a
+// child filesystem, as opposed to being handled by this filesystem. It checks
+// the depth of the requested path, and if it's deeper than the portion of the
+// tree that's handled by the parent, returns true.
+func shouldDelegateToChild(r *http.Request, pathComponents []string, mpl int) bool {
+	return !shared.IsRoot(r.URL.Path) && len(pathComponents)+getDepth(r) > mpl
+}
+
+func (h *Handler) delegateRewriting(w http.ResponseWriter, r *http.Request, pathComponents []string, mpl int) (int, []byte) {
+	// Use a buffering ResponseWriter so that we can manipulate the result.
+	// The only thing we use from the original ResponseWriter is Header().
+	bw := &bufferingResponseWriter{ResponseWriter: w}
+
+	h.delegate(mpl, pathComponents[mpl-1:], bw, r)
+
+	// Fixup paths to add the requested path as a prefix, escaped for inclusion in XML.
+	pp := shared.EscapeForXML(shared.Join(pathComponents[0:mpl]...))
+	b := responseHrefRegex.ReplaceAll(bw.buf.Bytes(), []byte(fmt.Sprintf("$1<D:href>%s/$3</D:href>", pp)))
+	return bw.status, b
+}
+
+func respondRewritten(w http.ResponseWriter, status int, result []byte) {
+	w.Header().Del("Content-Length")
+	w.WriteHeader(status)
+	if result != nil {
+		w.Write(result)
+	}
+}
+
+func getDepth(r *http.Request) int {
+	switch r.Header.Get("Depth") {
+	case "0":
+		return 0
+	case "1":
+		return 1
+	case "infinity":
+		return math.MaxInt16 // a really large number, but not infinity (avoids wrapping when we do arithmetic with this)
+	}
+	return 0
+}
+
+type bufferingResponseWriter struct {
+	http.ResponseWriter
+	status int
+	buf    bytes.Buffer
+}
+
+func (bw *bufferingResponseWriter) WriteHeader(statusCode int) {
+	bw.status = statusCode
+}
+
+func (bw *bufferingResponseWriter) Write(p []byte) (int, error) {
+	return bw.buf.Write(p)
+}
+
+// rewriteIfHeader rewrites URLs in the If header by removing the host and the
+// portion of the path that corresponds to this composite filesystem. This way,
+// when we delegate requests to child filesystems, the If header will reference
+// a path that makes sense on those filesystems.
+//
+// See http://www.webdav.org/specs/rfc4918.html#HEADER_If
+func rewriteIfHeader(r *http.Request, pathComponents []string, mpl int) {
+	ih := r.Header.Get("If")
+	if ih == "" {
+		return
+	}
+	matches := ifHrefRegex.FindStringSubmatch(ih)
+	if len(matches) == 3 {
+		pp := shared.JoinEscaped(pathComponents[0:mpl]...)
+		p := strings.Replace(shared.JoinEscaped(pathComponents...), pp, "", 1)
+		nih := ifHrefRegex.ReplaceAllString(ih, fmt.Sprintf("<%s>", p))
+		r.Header.Set("If", nih)
+	}
+}

drive/driveimpl/compositedav/stat_cache.go

@@ -4,11 +4,19 @@
 package compositedav
 
 import (
+	"bytes"
+	"encoding/xml"
+	"log"
 	"net/http"
 	"sync"
 	"time"
 
 	"github.com/jellydator/ttlcache/v3"
+	"tailscale.com/drive/driveimpl/shared"
+)
+
+var (
+	notFound = newCacheEntry(http.StatusNotFound, nil)
 )
 
 // StatCache provides a cache for directory listings and file metadata.
@@ -18,12 +26,38 @@ import (
 // This is similar to the DirectoryCacheLifetime setting of Windows' built-in
 // SMB client, see
 // https://learn.microsoft.com/en-us/previous-versions/windows/it-pro/windows-7/ff686200(v=ws.10)
+//
+// StatCache is built specifically to cache the results of PROPFIND requests,
+// which come back as MultiStatus XML responses. Typical clients will issue two
+// kinds of PROPFIND:
+//
+// The first kind of PROPFIND is a directory listing performed to depth 1. At
+// this depth, the resulting XML will contain stats for the requested folder as
+// well as for all children of that folder.
+//
+// The second kind of PROPFIND is a file listing performed to depth 0. At this
+// depth, the resulting XML will contain stats only for the requested file.
+//
+// In order to avoid round-trips, when a PROPFIND at depth 0 is attempted, and
+// the requested file is not in the cache, StatCache will check to see if the
+// parent folder of that file is cached. If so, StatCache infers the correct
+// MultiStatus for the file according to the following logic:
+//
+//  1. If the parent folder is NotFound (404), treat the file itself as NotFound
+//  2. If the parent folder's XML doesn't contain the file, treat it as
+//     NotFound.
+//  3. If the parent folder's XML contains the file, build a MultiStatus for the
+//     file based on the parent's XML.
+//
+// To avoid inconsistencies from the perspective of the client, any operations
+// that modify the filesystem (e.g. PUT, MKDIR, etc.) should call invalidate()
+// to invalidate the cache.
 type StatCache struct {
 	TTL time.Duration
 
 	// mu guards the below values.
 	mu                   sync.Mutex
-	cachesByDepthAndPath map[int]*ttlcache.Cache[string, []byte]
+	cachesByDepthAndPath map[int]*ttlcache.Cache[string, *cacheEntry]
 }
 
 // getOr checks the cache for the named value at the given depth. If a cached
@@ -32,25 +66,57 @@ type StatCache struct {
 // status and value. If the function returned http.StatusMultiStatus, getOr
 // caches the resulting value at the given name and depth before returning.
 func (c *StatCache) getOr(name string, depth int, or func() (int, []byte)) (int, []byte) {
-	cached := c.get(name, depth)
-	if cached != nil {
-		return http.StatusMultiStatus, cached
+	ce := c.get(name, depth)
+	if ce == nil {
+		// Not cached, fetch value.
+		status, raw := or()
+		ce = newCacheEntry(status, raw)
+		if status == http.StatusMultiStatus || status == http.StatusNotFound {
+			// Got a legit status, cache value
+			c.set(name, depth, ce)
+		}
 	}
-	status, next := or()
-	if c != nil && status == http.StatusMultiStatus && next != nil {
-		c.set(name, depth, next)
-	}
-	return status, next
+	return ce.Status, ce.Raw
 }
 
-func (c *StatCache) get(name string, depth int) []byte {
+// get retrieves the entry for the named file at the given depth. If no entry
+// is found, and depth == 0, get will check to see if the parent path of name
+// is present in the cache at depth 1. If so, it will infer that the child does
+// not exist and return notFound (404).
+func (c *StatCache) get(name string, depth int) *cacheEntry {
 	if c == nil {
 		return nil
 	}
 
+	name = shared.Normalize(name)
+
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
+	ce := c.tryGetLocked(name, depth)
+	if ce != nil {
+		// Cache hit.
+		return ce
+	}
+
+	if depth > 0 {
+		// Cache miss.
+		return nil
+	}
+
+	// At depth 0, if child's parent is in the cache, and the child isn't
+	// cached, we can infer that the child is notFound.
+	p := c.tryGetLocked(shared.Parent(name), 1)
+	if p != nil {
+		return notFound
+	}
+
+	// No parent in cache, cache miss.
+	return nil
+}
+
+// tryGetLocked requires that c.mu be held.
+func (c *StatCache) tryGetLocked(name string, depth int) *cacheEntry {
 	if c.cachesByDepthAndPath == nil {
 		return nil
 	}
@@ -65,28 +131,80 @@ func (c *StatCache) get(name string, depth int) []byte {
 	return item.Value()
 }
 
-func (c *StatCache) set(name string, depth int, value []byte) {
+// set stores the given cacheEntry in the cache at the given name and depth. If
+// the depth is 1, set also populates depth 0 entries in the cache for the bare
+// name. If status is StatusMultiStatus, set will parse the PROPFIND result and
+// store depth 0 entries for all children. If parsing the result fails, nothing
+// is cached.
+func (c *StatCache) set(name string, depth int, ce *cacheEntry) {
 	if c == nil {
 		return
 	}
 
+	name = shared.Normalize(name)
+
+	var self *cacheEntry
+	var children map[string]*cacheEntry
+	if depth == 1 {
+		switch ce.Status {
+		case http.StatusNotFound:
+			// Record notFound as the self entry.
+			self = ce
+		case http.StatusMultiStatus:
+			// Parse the raw MultiStatus and extract specific responses
+			// corresponding to the self entry (e.g. the directory, but at depth 0)
+			// and children (e.g. files within the directory) so that subsequent
+			// requests for these can be satisfied from the cache.
+			var ms multiStatus
+			err := xml.Unmarshal(ce.Raw, &ms)
+			if err != nil {
+				// unparseable MultiStatus response, don't cache
+				log.Printf("statcache.set error: %s", err)
+				return
+			}
+			children = make(map[string]*cacheEntry, len(ms.Responses)-1)
+			for i := 0; i < len(ms.Responses); i++ {
+				response := ms.Responses[i]
+				name := shared.Normalize(response.Href)
+				raw := marshalMultiStatus(response)
+				entry := newCacheEntry(ce.Status, raw)
+				if i == 0 {
+					self = entry
+				} else {
+					children[name] = entry
+				}
+			}
+		}
+	}
+
 	c.mu.Lock()
 	defer c.mu.Unlock()
+	c.setLocked(name, depth, ce)
+	if self != nil {
+		c.setLocked(name, 0, self)
+	}
+	for childName, child := range children {
+		c.setLocked(childName, 0, child)
+	}
+}
 
+// setLocked requires that c.mu be held.
+func (c *StatCache) setLocked(name string, depth int, ce *cacheEntry) {
 	if c.cachesByDepthAndPath == nil {
-		c.cachesByDepthAndPath = make(map[int]*ttlcache.Cache[string, []byte])
+		c.cachesByDepthAndPath = make(map[int]*ttlcache.Cache[string, *cacheEntry])
 	}
 	cache := c.cachesByDepthAndPath[depth]
 	if cache == nil {
 		cache = ttlcache.New(
-			ttlcache.WithTTL[string, []byte](c.TTL),
+			ttlcache.WithTTL[string, *cacheEntry](c.TTL),
 		)
 		go cache.Start()
 		c.cachesByDepthAndPath[depth] = cache
 	}
-	cache.Set(name, value, ttlcache.DefaultTTL)
+	cache.Set(name, ce, ttlcache.DefaultTTL)
 }
 
+// invalidate invalidates the entire cache.
 func (c *StatCache) invalidate() {
 	if c == nil {
 		return
@@ -108,3 +226,54 @@ func (c *StatCache) stop() {
 		cache.Stop()
 	}
 }
+
+type cacheEntry struct {
+	Status int
+	Raw    []byte
+}
+
+func newCacheEntry(status int, raw []byte) *cacheEntry {
+	return &cacheEntry{Status: status, Raw: raw}
+}
+
+type propStat struct {
+	InnerXML []byte `xml:",innerxml"`
+}
+
+type response struct {
+	XMLName   xml.Name    `xml:"response"`
+	Href      string      `xml:"href"`
+	PropStats []*propStat `xml:"propstat"`
+}
+
+type multiStatus struct {
+	XMLName   xml.Name    `xml:"multistatus"`
+	Responses []*response `xml:"response"`
+}
+
+// marshalMultiStatus performs custom marshalling of a MultiStatus to preserve
+// the original formatting, namespacing, etc. Doing this with Go's XML encoder
+// is somewhere between difficult and impossible, which is why we use this more
+// manual approach.
+func marshalMultiStatus(response *response) []byte {
+	// TODO(percy): maybe pool these buffers
+	var buf bytes.Buffer
+	buf.WriteString(multistatusTemplateStart)
+	buf.WriteString(response.Href)
+	buf.WriteString(hrefEnd)
+	for _, propStat := range response.PropStats {
+		buf.WriteString(propstatStart)
+		buf.Write(propStat.InnerXML)
+		buf.WriteString(propstatEnd)
+	}
+	buf.WriteString(multistatusTemplateEnd)
+	return buf.Bytes()
+}
+
+const (
+	multistatusTemplateStart = `<?xml version="1.0" encoding="UTF-8"?><D:multistatus xmlns:D="DAV:"><D:response><D:href>`
+	hrefEnd                  = `</D:href>`
+	propstatStart            = `<D:propstat>`
+	propstatEnd              = `</D:propstat>`
+	multistatusTemplateEnd   = `</D:response></D:multistatus>`
+)

drive/driveimpl/compositedav/stat_cache_test.go

@@ -4,17 +4,65 @@
 package compositedav
 
 import (
-	"bytes"
+	"fmt"
+	"log"
+	"net/http"
+	"path"
+	"strings"
 	"testing"
 	"time"
 
+	"github.com/google/go-cmp/cmp"
 	"tailscale.com/tstest"
 )
 
-var (
-	val  = []byte("1")
-	file = "file"
-)
+var parentPath = "/parent"
+
+var childPath = "/parent/child.txt"
+
+var parentResponse = `<D:response>
+<D:href>/parent/</D:href>
+<D:propstat>
+<D:prop>
+<D:getlastmodified>Mon, 29 Apr 2024 19:52:23 GMT</D:getlastmodified>
+<D:creationdate>Fri, 19 Apr 2024 04:13:34 GMT</D:creationdate>
+<D:resourcetype>
+<D:collection xmlns:D="DAV:" />
+</D:resourcetype>
+</D:prop>
+<D:status>HTTP/1.1 200 OK</D:status>
+</D:propstat>
+</D:response>`
+
+var childResponse = `
+<D:response>
+<D:href>/parent/child.txt</D:href>
+<D:propstat>
+<D:prop>
+<D:getlastmodified>Mon, 29 Apr 2024 19:52:23 GMT</D:getlastmodified>
+<D:creationdate>Fri, 19 Apr 2024 04:13:34 GMT</D:creationdate>
+<D:resourcetype>
+<D:collection xmlns:D="DAV:" />
+</D:resourcetype>
+</D:prop>
+<D:status>HTTP/1.1 200 OK</D:status>
+</D:propstat>
+</D:response>`
+
+var fullParent = []byte(
+	strings.ReplaceAll(
+		fmt.Sprintf(`<?xml version="1.0" encoding="UTF-8"?><D:multistatus xmlns:D="DAV:">%s%s</D:multistatus>`, parentResponse, childResponse),
+		"\n", ""))
+
+var partialParent = []byte(
+	strings.ReplaceAll(
+		fmt.Sprintf(`<?xml version="1.0" encoding="UTF-8"?><D:multistatus xmlns:D="DAV:">%s</D:multistatus>`, parentResponse),
+		"\n", ""))
+
+var fullChild = []byte(
+	strings.ReplaceAll(
+		fmt.Sprintf(`<?xml version="1.0" encoding="UTF-8"?><D:multistatus xmlns:D="DAV:">%s</D:multistatus>`, childResponse),
+		"\n", ""))
 
 func TestStatCacheNoTimeout(t *testing.T) {
 	// Make sure we don't leak goroutines
@@ -24,22 +72,23 @@ func TestStatCacheNoTimeout(t *testing.T) {
 	defer c.stop()
 
 	// check get before set
-	fetched := c.get(file, 1)
+	fetched := c.get(childPath, 0)
 	if fetched != nil {
-		t.Errorf("got %q, want nil", fetched)
+		t.Errorf("got %v, want nil", fetched)
 	}
 
 	// set new stat
-	c.set(file, 1, val)
-	fetched = c.get(file, 1)
-	if !bytes.Equal(fetched, val) {
-		t.Errorf("got %q, want %q", fetched, val)
+	ce := newCacheEntry(http.StatusMultiStatus, fullChild)
+	c.set(childPath, 0, ce)
+	fetched = c.get(childPath, 0)
+	if diff := cmp.Diff(fetched, ce); diff != "" {
+		t.Errorf("should have gotten cached value; (-got+want):%v", diff)
 	}
 
 	// fetch stat again, should still be cached
-	fetched = c.get(file, 1)
-	if !bytes.Equal(fetched, val) {
-		t.Errorf("got %q, want %q", fetched, val)
+	fetched = c.get(childPath, 0)
+	if diff := cmp.Diff(fetched, ce); diff != "" {
+		t.Errorf("should still have gotten cached value; (-got+want):%v", diff)
 	}
 }
 
@@ -51,25 +100,114 @@ func TestStatCacheTimeout(t *testing.T) {
 	defer c.stop()
 
 	// set new stat
-	c.set(file, 1, val)
-	fetched := c.get(file, 1)
-	if !bytes.Equal(fetched, val) {
-		t.Errorf("got %q, want %q", fetched, val)
+	ce := newCacheEntry(http.StatusMultiStatus, fullChild)
+	c.set(childPath, 0, ce)
+	fetched := c.get(childPath, 0)
+	if diff := cmp.Diff(fetched, ce); diff != "" {
+		t.Errorf("should have gotten cached value; (-got+want):%v", diff)
 	}
 
 	// wait for cache to expire and refetch stat, should be empty now
 	time.Sleep(c.TTL * 2)
 
-	fetched = c.get(file, 1)
+	fetched = c.get(childPath, 0)
 	if fetched != nil {
-		t.Errorf("invalidate should have cleared cached value")
+		t.Errorf("cached value should have expired")
 	}
 
-	c.set(file, 1, val)
+	c.set(childPath, 0, ce)
 	// invalidate the cache and make sure nothing is returned
 	c.invalidate()
-	fetched = c.get(file, 1)
+	fetched = c.get(childPath, 0)
 	if fetched != nil {
 		t.Errorf("invalidate should have cleared cached value")
 	}
 }
+
+func TestParentChildRelationship(t *testing.T) {
+	// Make sure we don't leak goroutines
+	tstest.ResourceCheck(t)
+
+	c := &StatCache{TTL: 24 * time.Hour} // don't expire
+	defer c.stop()
+
+	missingParentPath := "/missingparent"
+	unparseableParentPath := "/unparseable"
+
+	c.set(parentPath, 1, newCacheEntry(http.StatusMultiStatus, fullParent))
+	c.set(missingParentPath, 1, newCacheEntry(http.StatusNotFound, nil))
+	c.set(unparseableParentPath, 1, newCacheEntry(http.StatusMultiStatus, []byte("<this will not parse")))
+
+	tests := []struct {
+		path  string
+		depth int
+		want  *cacheEntry
+	}{
+		{
+			path:  parentPath,
+			depth: 1,
+			want:  newCacheEntry(http.StatusMultiStatus, fullParent),
+		},
+		{
+			path:  parentPath,
+			depth: 0,
+			want:  newCacheEntry(http.StatusMultiStatus, partialParent),
+		},
+		{
+			path:  childPath,
+			depth: 0,
+			want:  newCacheEntry(http.StatusMultiStatus, fullChild),
+		},
+		{
+			path:  path.Join(parentPath, "nonexistent.txt"),
+			depth: 0,
+			want:  notFound,
+		},
+		{
+			path:  missingParentPath,
+			depth: 1,
+			want:  notFound,
+		},
+		{
+			path:  missingParentPath,
+			depth: 0,
+			want:  notFound,
+		},
+		{
+			path:  path.Join(missingParentPath, "filename.txt"),
+			depth: 0,
+			want:  notFound,
+		},
+		{
+			path:  unparseableParentPath,
+			depth: 1,
+			want:  nil,
+		},
+		{
+			path:  unparseableParentPath,
+			depth: 0,
+			want:  nil,
+		},
+		{
+			path:  path.Join(unparseableParentPath, "filename.txt"),
+			depth: 0,
+			want:  nil,
+		},
+		{
+			path:  "/unknown",
+			depth: 1,
+			want:  nil,
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(fmt.Sprintf("%d%s", test.depth, test.path), func(t *testing.T) {
+			got := c.get(test.path, test.depth)
+			if diff := cmp.Diff(got, test.want); diff != "" {
+				t.Errorf("unexpected cached value; (-got+want):%v", diff)
+				log.Printf("want\n%s", test.want.Raw)
+				log.Printf("got\n%s", got.Raw)
+			}
+		})
+	}
+}

drive/driveimpl/drive_test.go

@@ -14,6 +14,8 @@ import (
 	"os"
 	"path"
 	"path/filepath"
+	"regexp"
+	"runtime"
 	"slices"
 	"strings"
 	"sync"
@@ -30,14 +32,29 @@ import (
 const (
 	domain = `test$%domain.com`
 
-	remote1 = `rem ote$%1`
-	remote2 = `_rem ote$%2`
-	share11 = `sha re$%11`
-	share12 = `_sha re$%12`
-	file111 = `fi le$%111.txt`
+	remote1 = `rem ote$%<>1`
+	remote2 = `_rem ote$%<>2`
+	share11 = `sha re$%<>11`
+	share12 = `_sha re$%<>12`
 	file112 = `file112.txt`
 )
 
+var (
+	file111 = `fi le$%<>111.txt`
+)
+
+func init() {
+	if runtime.GOOS == "windows" {
+		// file with less than and greater than doesn't work on Windows
+		file111 = `fi le$%111.txt`
+	}
+}
+
+var (
+	lockRootRegex  = regexp.MustCompile(`<D:lockroot><D:href>/?([^<]*)/?</D:href>`)
+	lockTokenRegex = regexp.MustCompile(`<D:locktoken><D:href>([0-9]+)/?</D:href>`)
+)
+
 func init() {
 	// set AllowShareAs() to false so that we don't try to use sub-processes
 	// for access files on disk.
@@ -145,6 +162,206 @@ func TestSecretTokenAuth(t *testing.T) {
 	}
 }
 
+func TestLOCK(t *testing.T) {
+	s := newSystem(t)
+
+	s.addRemote(remote1)
+	s.addShare(remote1, share11, drive.PermissionReadWrite)
+	s.writeFile("writing file to read/write remote should succeed", remote1, share11, file111, "hello world", true)
+
+	client := &http.Client{
+		Transport: &http.Transport{DisableKeepAlives: true},
+	}
+
+	u := fmt.Sprintf("http://%s/%s/%s/%s/%s",
+		s.local.l.Addr(),
+		url.PathEscape(domain),
+		url.PathEscape(remote1),
+		url.PathEscape(share11),
+		url.PathEscape(file111))
+
+	// First acquire a lock with a short timeout
+	req, err := http.NewRequest("LOCK", u, strings.NewReader(lockBody))
+	if err != nil {
+		t.Fatal(err)
+	}
+	req.Header.Set("Depth", "infinity")
+	req.Header.Set("Timeout", "Second-1")
+	resp, err := client.Do(req)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != 200 {
+		t.Fatalf("expected LOCK to succeed, but got status %d", resp.StatusCode)
+	}
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	submatches := lockRootRegex.FindStringSubmatch(string(body))
+	if len(submatches) != 2 {
+		t.Fatal("failed to find lockroot")
+	}
+	want := shared.EscapeForXML(pathTo(remote1, share11, file111))
+	got := submatches[1]
+	if got != want {
+		t.Fatalf("want lockroot %q, got %q", want, got)
+	}
+
+	submatches = lockTokenRegex.FindStringSubmatch(string(body))
+	if len(submatches) != 2 {
+		t.Fatal("failed to find locktoken")
+	}
+	lockToken := submatches[1]
+	ifHeader := fmt.Sprintf("<%s> (<%s>)", u, lockToken)
+
+	// Then refresh the lock with a longer timeout
+	req, err = http.NewRequest("LOCK", u, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	req.Header.Set("Depth", "infinity")
+	req.Header.Set("Timeout", "Second-600")
+	req.Header.Set("If", ifHeader)
+	resp, err = client.Do(req)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != 200 {
+		t.Fatalf("expected LOCK refresh to succeed, but got status %d", resp.StatusCode)
+	}
+	body, err = io.ReadAll(resp.Body)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	submatches = lockRootRegex.FindStringSubmatch(string(body))
+	if len(submatches) != 2 {
+		t.Fatal("failed to find lockroot after refresh")
+	}
+	want = shared.EscapeForXML(pathTo(remote1, share11, file111))
+	got = submatches[1]
+	if got != want {
+		t.Fatalf("want lockroot after refresh %q, got %q", want, got)
+	}
+
+	submatches = lockTokenRegex.FindStringSubmatch(string(body))
+	if len(submatches) != 2 {
+		t.Fatal("failed to find locktoken after refresh")
+	}
+	if submatches[1] != lockToken {
+		t.Fatalf("on refresh, lock token changed from %q to %q", lockToken, submatches[1])
+	}
+
+	// Then wait past the original timeout, then try to delete without the lock
+	// (should fail)
+	time.Sleep(1 * time.Second)
+	req, err = http.NewRequest("DELETE", u, nil)
+	if err != nil {
+		log.Fatal(err)
+	}
+	resp, err = client.Do(req)
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != 423 {
+		t.Fatalf("deleting without lock token should fail with 423, but got %d", resp.StatusCode)
+	}
+
+	// Then delete with the lock (should succeed)
+	req, err = http.NewRequest("DELETE", u, nil)
+	if err != nil {
+		log.Fatal(err)
+	}
+	req.Header.Set("If", ifHeader)
+	resp, err = client.Do(req)
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != 204 {
+		t.Fatalf("deleting with lock token should have succeeded with 204, but got %d", resp.StatusCode)
+	}
+}
+
+func TestUNLOCK(t *testing.T) {
+	s := newSystem(t)
+
+	s.addRemote(remote1)
+	s.addShare(remote1, share11, drive.PermissionReadWrite)
+	s.writeFile("writing file to read/write remote should succeed", remote1, share11, file111, "hello world", true)
+
+	client := &http.Client{
+		Transport: &http.Transport{DisableKeepAlives: true},
+	}
+
+	u := fmt.Sprintf("http://%s/%s/%s/%s/%s",
+		s.local.l.Addr(),
+		url.PathEscape(domain),
+		url.PathEscape(remote1),
+		url.PathEscape(share11),
+		url.PathEscape(file111))
+
+	// Acquire a lock
+	req, err := http.NewRequest("LOCK", u, strings.NewReader(lockBody))
+	if err != nil {
+		t.Fatal(err)
+	}
+	req.Header.Set("Depth", "infinity")
+	req.Header.Set("Timeout", "Second-600")
+	resp, err := client.Do(req)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != 200 {
+		t.Fatalf("expected LOCK to succeed, but got status %d", resp.StatusCode)
+	}
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	submatches := lockTokenRegex.FindStringSubmatch(string(body))
+	if len(submatches) != 2 {
+		t.Fatal("failed to find locktoken")
+	}
+	lockToken := submatches[1]
+
+	// Release the lock
+	req, err = http.NewRequest("UNLOCK", u, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	req.Header.Set("Lock-Token", fmt.Sprintf("<%s>", lockToken))
+	resp, err = client.Do(req)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != 204 {
+		t.Fatalf("expected UNLOCK to succeed with a 204, but got status %d", resp.StatusCode)
+	}
+
+	// Then delete without the lock (should succeed)
+	req, err = http.NewRequest("DELETE", u, nil)
+	if err != nil {
+		log.Fatal(err)
+	}
+	resp, err = client.Do(req)
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != 204 {
+		t.Fatalf("deleting without lock should have succeeded with 204, but got %d", resp.StatusCode)
+	}
+}
+
 type local struct {
 	l  net.Listener
 	fs *FileSystemForLocal
@@ -486,3 +703,9 @@ func (a *noopAuthenticator) Clone() gowebdav.Authenticator {
 func (a *noopAuthenticator) Close() error {
 	return nil
 }
+
+const lockBody = `<?xml version="1.0" encoding="utf-8" ?> 
+<D:lockinfo xmlns:D='DAV:'> 
+  <D:lockscope><D:exclusive/></D:lockscope> 
+  <D:locktype><D:write/></D:locktype> 
+</D:lockinfo>`

drive/driveimpl/fileserver.go

@@ -151,6 +151,9 @@ func (s *FileServer) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		w.WriteHeader(http.StatusNotFound)
 		return
 	}
+	// WebDAV's locking code compares the lock resources with the request's
+	// host header, set this to empty to avoid mismatches.
+	r.Host = ""
 	h.ServeHTTP(w, r)
 }
 

drive/driveimpl/shared/pathutil.go

@@ -26,6 +26,17 @@ func CleanAndSplit(p string) []string {
 	return strings.Split(strings.Trim(path.Clean(p), sepStringAndDot), sepString)
 }
 
+// Normalize normalizes the given path (e.g. dropping trailing slashes).
+func Normalize(p string) string {
+	return Join(CleanAndSplit(p)...)
+}
+
+// Parent extracts the parent of the given path.
+func Parent(p string) string {
+	parts := CleanAndSplit(p)
+	return Join(parts[:len(parts)-1]...)
+}
+
 // Join behaves like path.Join() but also includes a leading slash.
 func Join(parts ...string) string {
 	fullParts := make([]string, 0, len(parts))

drive/driveimpl/shared/xml.go

@@ -0,0 +1,16 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package shared
+
+import (
+	"bytes"
+	"encoding/xml"
+)
+
+// EscapeForXML escapes the given string for use in XML text.
+func EscapeForXML(s string) string {
+	result := bytes.NewBuffer(nil)
+	xml.Escape(result, []byte(s))
+	return result.String()
+}

go.mod

@@ -14,6 +14,7 @@ require (
 	github.com/aws/aws-sdk-go-v2/feature/s3/manager v1.11.64
 	github.com/aws/aws-sdk-go-v2/service/s3 v1.33.0
 	github.com/aws/aws-sdk-go-v2/service/ssm v1.44.7
+	github.com/bramvdbogaerde/go-scp v1.4.0
 	github.com/coreos/go-iptables v0.7.1-0.20240112124308-65c67c9f46e6
 	github.com/coreos/go-systemd v0.0.0-20191104093116-d3cd4ed1dbcf
 	github.com/creack/pty v1.1.21
@@ -37,7 +38,7 @@ require (
 	github.com/google/go-cmp v0.6.0
 	github.com/google/go-containerregistry v0.18.0
 	github.com/google/nftables v0.2.1-0.20240414091927-5e242ec57806
-	github.com/google/uuid v1.5.0
+	github.com/google/uuid v1.6.0
 	github.com/goreleaser/nfpm/v2 v2.33.1
 	github.com/hdevalence/ed25519consensus v0.2.0
 	github.com/iancoleman/strcase v0.3.0
@@ -60,6 +61,7 @@ require (
 	github.com/peterbourgon/ff/v3 v3.4.0
 	github.com/pkg/errors v0.9.1
 	github.com/pkg/sftp v1.13.6
+	github.com/prometheus-community/pro-bing v0.4.0
 	github.com/prometheus/client_golang v1.18.0
 	github.com/prometheus/common v0.46.0
 	github.com/safchain/ethtool v0.3.0

go.sum

@@ -177,6 +177,8 @@ github.com/blizzy78/varnamelen v0.8.0 h1:oqSblyuQvFsW1hbBHh1zfwrKe3kcSj0rnXkKzsQ
 github.com/blizzy78/varnamelen v0.8.0/go.mod h1:V9TzQZ4fLJ1DSrjVDfl89H7aMnTvKkApdHeyESmyR7k=
 github.com/bombsimon/wsl/v3 v3.4.0 h1:RkSxjT3tmlptwfgEgTgU+KYKLI35p/tviNXNXiL2aNU=
 github.com/bombsimon/wsl/v3 v3.4.0/go.mod h1:KkIB+TXkqy6MvK9BDZVbZxKNYsE1/oLRJbIFtf14qqo=
+github.com/bramvdbogaerde/go-scp v1.4.0 h1:jKMwpwCbcX1KyvDbm/PDJuXcMuNVlLGi0Q0reuzjyKY=
+github.com/bramvdbogaerde/go-scp v1.4.0/go.mod h1:on2aH5AxaFb2G0N5Vsdy6B0Ml7k9HuHSwfo1y0QzAbQ=
 github.com/breml/bidichk v0.2.4 h1:i3yedFWWQ7YzjdZJHnPo9d/xURinSq3OM+gyM43K4/8=
 github.com/breml/bidichk v0.2.4/go.mod h1:7Zk0kRFt1LIZxtQdl9W9JwGAcLTTkOs+tN7wuEYGJ3s=
 github.com/breml/errchkjson v0.3.1 h1:hlIeXuspTyt8Y/UmP5qy1JocGNR00KQHgfaNtRAjoxQ=
@@ -468,8 +470,8 @@ github.com/google/rpmpack v0.5.0 h1:L16KZ3QvkFGpYhmp23iQip+mx1X39foEsqszjMNBm8A=
 github.com/google/rpmpack v0.5.0/go.mod h1:uqVAUVQLq8UY2hCDfmJ/+rtO3aw7qyhc90rCVEabEfI=
 github.com/google/uuid v1.1.1/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/google/uuid v1.1.2/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
-github.com/google/uuid v1.5.0 h1:1p67kYwdtXjb0gL0BPiP1Av9wiZPo5A8z2cWkTZ+eyU=
-github.com/google/uuid v1.5.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
+github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
+github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/googleapis/gax-go/v2 v2.0.4/go.mod h1:0Wqv26UfaUD9n4G6kQubkQ+KchISgw+vpHVxEJEs9eg=
 github.com/googleapis/gax-go/v2 v2.0.5/go.mod h1:DWXyrwAJ9X0FpwwEdw+IPEYBICEFu5mhpdKc/us6bOk=
 github.com/googleapis/google-cloud-go-testing v0.0.0-20200911160855-bcd43fbb19e8/go.mod h1:dvDLG8qkwmyD9a/MJJN3XJcT3xFxOKAvTZGvuZmac9g=
@@ -731,6 +733,8 @@ github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZb
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/polyfloyd/go-errorlint v1.4.1 h1:r8ru5FhXSn34YU1GJDOuoJv2LdsQkPmK325EOpPMJlM=
 github.com/polyfloyd/go-errorlint v1.4.1/go.mod h1:k6fU/+fQe38ednoZS51T7gSIGQW1y94d6TkSr35OzH8=
+github.com/prometheus-community/pro-bing v0.4.0 h1:YMbv+i08gQz97OZZBwLyvmmQEEzyfyrrjEaAchdy3R4=
+github.com/prometheus-community/pro-bing v0.4.0/go.mod h1:b7wRYZtCcPmt4Sz319BykUU241rWLe1VFXyiyWK/dH4=
 github.com/prometheus/client_golang v0.9.1/go.mod h1:7SWBe2y4D6OKWSNQJUaRYU/AaXPKyh/dDVn+NZz0KFw=
 github.com/prometheus/client_golang v1.0.0/go.mod h1:db9x61etRT2tGnBNRi70OPL5FsnadC4Ky3P0J6CfImo=
 github.com/prometheus/client_golang v1.7.1/go.mod h1:PY5Wy2awLA44sXw4AOSfFBetzPP4j5+D6mVACh+pe2M=

go.toolchain.rev

@@ -1 +1 @@
-48d71857bf5352daaa10b61dd3e9b1c0dd51e27a
+467a489ae3c080d80f4cfdd05f2aa08cb44c9d6a

ipn/conf.go

@@ -32,7 +32,8 @@ type ConfigVAlpha struct {
 	AdvertiseRoutes []netip.Prefix `json:",omitempty"`
 	DisableSNAT     opt.Bool       `json:",omitempty"`
 
-	NetfilterMode *string `json:",omitempty"` // "on", "off", "nodivert"
+	NetfilterMode       *string  `json:",omitempty"` // "on", "off", "nodivert"
+	NoStatefulFiltering opt.Bool `json:",omitempty"`
 
 	PostureChecking opt.Bool         `json:",omitempty"`
 	RunSSHServer    opt.Bool         `json:",omitempty"` // Tailscale SSH
@@ -50,6 +51,7 @@ func (c *ConfigVAlpha) ToPrefs() (MaskedPrefs, error) {
 	if c == nil {
 		return mp, nil
 	}
+
 	mp.WantRunning = !c.Enabled.EqualBool(false)
 	mp.WantRunningSet = mp.WantRunning || c.Enabled != ""
 	if c.ServerURL != nil {
@@ -98,6 +100,11 @@ func (c *ConfigVAlpha) ToPrefs() (MaskedPrefs, error) {
 		mp.NoSNAT = c.DisableSNAT.EqualBool(true)
 		mp.NoSNAT = true
 	}
+	if c.NoStatefulFiltering != "" {
+		mp.NoStatefulFiltering = c.NoStatefulFiltering
+		mp.NoStatefulFilteringSet = true
+	}
+
 	if c.NetfilterMode != nil {
 		m, err := preftype.ParseNetfilterMode(*c.NetfilterMode)
 		if err != nil {

ipn/conffile/cloudconf.go

@@ -0,0 +1,59 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package conffile
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
+	"tailscale.com/omit"
+)
+
+func getEC2MetadataToken() (string, error) {
+	if omit.AWS {
+		return "", omit.Err
+	}
+	req, _ := http.NewRequest("PUT", "http://169.254.169.254/latest/api/token", nil)
+	req.Header.Add("X-aws-ec2-metadata-token-ttl-seconds", "300")
+	res, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return "", fmt.Errorf("failed to get metadata token: %w", err)
+	}
+	defer res.Body.Close()
+	if res.StatusCode != 200 {
+		return "", fmt.Errorf("failed to get metadata token: %v", res.Status)
+	}
+	all, err := io.ReadAll(res.Body)
+	if err != nil {
+		return "", fmt.Errorf("failed to read metadata token: %w", err)
+	}
+	return strings.TrimSpace(string(all)), nil
+}
+
+func readVMUserData() ([]byte, error) {
+	// TODO(bradfitz): support GCP, Azure, Proxmox/cloud-init
+	// (NoCloud/ConfigDrive ISO), etc.
+
+	if omit.AWS {
+		return nil, omit.Err
+	}
+	token, tokErr := getEC2MetadataToken()
+	req, _ := http.NewRequest("GET", "http://169.254.169.254/latest/user-data", nil)
+	req.Header.Add("X-aws-ec2-metadata-token", token)
+	res, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer res.Body.Close()
+	if res.StatusCode != 200 {
+		if tokErr != nil {
+			return nil, fmt.Errorf("failed to get VM user data: %v; also failed to get metadata token: %v", res.Status, tokErr)
+		}
+		return nil, errors.New(res.Status)
+	}
+	return io.ReadAll(res.Body)
+}

ipn/conffile/conffile.go

@@ -17,7 +17,7 @@ import (
 
 // Config describes a config file.
 type Config struct {
-	Path    string // disk path of HuJSON
+	Path    string // disk path of HuJSON, or VMUserDataPath
 	Raw     []byte // raw bytes from disk, in HuJSON form
 	Std     []byte // standardized JSON form
 	Version string // "alpha0" for now
@@ -35,13 +35,22 @@ func (c *Config) WantRunning() bool {
 	return c != nil && !c.Parsed.Enabled.EqualBool(false)
 }
 
+// VMUserDataPath is a sentinel value for Load to use to get the data
+// from the VM's metadata service's user-data field.
+const VMUserDataPath = "vm:user-data"
+
 // Load reads and parses the config file at the provided path on disk.
 func Load(path string) (*Config, error) {
 	var c Config
 	c.Path = path
-
 	var err error
-	c.Raw, err = os.ReadFile(path)
+
+	switch path {
+	case VMUserDataPath:
+		c.Raw, err = readVMUserData()
+	default:
+		c.Raw, err = os.ReadFile(path)
+	}
 	if err != nil {
 		return nil, err
 	}

ipn/ipn_clone.go

@@ -11,6 +11,7 @@ import (
 
 	"tailscale.com/drive"
 	"tailscale.com/tailcfg"
+	"tailscale.com/types/opt"
 	"tailscale.com/types/persist"
 	"tailscale.com/types/preftype"
 )
@@ -39,7 +40,6 @@ func (src *Prefs) Clone() *Prefs {
 var _PrefsCloneNeedsRegeneration = Prefs(struct {
 	ControlURL             string
 	RouteAll               bool
-	AllowSingleHosts       bool
 	ExitNodeID             tailcfg.StableNodeID
 	ExitNodeIP             netip.Addr
 	InternalExitNodePrior  tailcfg.StableNodeID
@@ -57,6 +57,7 @@ var _PrefsCloneNeedsRegeneration = Prefs(struct {
 	Egg                    bool
 	AdvertiseRoutes        []netip.Prefix
 	NoSNAT                 bool
+	NoStatefulFiltering    opt.Bool
 	NetfilterMode          preftype.NetfilterMode
 	OperatorUser           string
 	ProfileName            string
@@ -65,6 +66,7 @@ var _PrefsCloneNeedsRegeneration = Prefs(struct {
 	PostureChecking        bool
 	NetfilterKind          string
 	DriveShares            []*drive.Share
+	AllowSingleHosts       marshalAsTrueInJSON
 	Persist                *persist.Persist
 }{})
 

ipn/ipn_test.go

@@ -12,6 +12,7 @@ import (
 func TestDeps(t *testing.T) {
 	deptest.DepChecker{
 		BadDeps: map[string]string{
+			"testing":                            "do not use testing package in production code",
 			"gvisor.dev/gvisor/pkg/buffer":       "https://github.com/tailscale/tailscale/issues/9756",
 			"gvisor.dev/gvisor/pkg/cpuid":        "https://github.com/tailscale/tailscale/issues/9756",
 			"gvisor.dev/gvisor/pkg/tcpip":        "https://github.com/tailscale/tailscale/issues/9756",

ipn/ipn_view.go

@@ -12,6 +12,7 @@ import (
 
 	"tailscale.com/drive"
 	"tailscale.com/tailcfg"
+	"tailscale.com/types/opt"
 	"tailscale.com/types/persist"
 	"tailscale.com/types/preftype"
 	"tailscale.com/types/views"
@@ -66,7 +67,6 @@ func (v *PrefsView) UnmarshalJSON(b []byte) error {
 
 func (v PrefsView) ControlURL() string                          { return v.ж.ControlURL }
 func (v PrefsView) RouteAll() bool                              { return v.ж.RouteAll }
-func (v PrefsView) AllowSingleHosts() bool                      { return v.ж.AllowSingleHosts }
 func (v PrefsView) ExitNodeID() tailcfg.StableNodeID            { return v.ж.ExitNodeID }
 func (v PrefsView) ExitNodeIP() netip.Addr                      { return v.ж.ExitNodeIP }
 func (v PrefsView) InternalExitNodePrior() tailcfg.StableNodeID { return v.ж.InternalExitNodePrior }
@@ -86,6 +86,7 @@ func (v PrefsView) AdvertiseRoutes() views.Slice[netip.Prefix] {
 	return views.SliceOf(v.ж.AdvertiseRoutes)
 }
 func (v PrefsView) NoSNAT() bool                          { return v.ж.NoSNAT }
+func (v PrefsView) NoStatefulFiltering() opt.Bool         { return v.ж.NoStatefulFiltering }
 func (v PrefsView) NetfilterMode() preftype.NetfilterMode { return v.ж.NetfilterMode }
 func (v PrefsView) OperatorUser() string                  { return v.ж.OperatorUser }
 func (v PrefsView) ProfileName() string                   { return v.ж.ProfileName }
@@ -96,13 +97,13 @@ func (v PrefsView) NetfilterKind() string                 { return v.ж.Netfilte
 func (v PrefsView) DriveShares() views.SliceView[*drive.Share, drive.ShareView] {
 	return views.SliceOfViews[*drive.Share, drive.ShareView](v.ж.DriveShares)
 }
-func (v PrefsView) Persist() persist.PersistView { return v.ж.Persist.View() }
+func (v PrefsView) AllowSingleHosts() marshalAsTrueInJSON { return v.ж.AllowSingleHosts }
+func (v PrefsView) Persist() persist.PersistView          { return v.ж.Persist.View() }
 
 // A compilation failure here means this code must be regenerated, with the command at the top of this file.
 var _PrefsViewNeedsRegeneration = Prefs(struct {
 	ControlURL             string
 	RouteAll               bool
-	AllowSingleHosts       bool
 	ExitNodeID             tailcfg.StableNodeID
 	ExitNodeIP             netip.Addr
 	InternalExitNodePrior  tailcfg.StableNodeID
@@ -120,6 +121,7 @@ var _PrefsViewNeedsRegeneration = Prefs(struct {
 	Egg                    bool
 	AdvertiseRoutes        []netip.Prefix
 	NoSNAT                 bool
+	NoStatefulFiltering    opt.Bool
 	NetfilterMode          preftype.NetfilterMode
 	OperatorUser           string
 	ProfileName            string
@@ -128,6 +130,7 @@ var _PrefsViewNeedsRegeneration = Prefs(struct {
 	PostureChecking        bool
 	NetfilterKind          string
 	DriveShares            []*drive.Share
+	AllowSingleHosts       marshalAsTrueInJSON
 	Persist                *persist.Persist
 }{})
 

ipn/ipnlocal/c2n.go

@@ -478,17 +478,44 @@ func findCmdTailscale() (string, error) {
 }
 
 func tailscaleUpdateCmd(cmdTS string) *exec.Cmd {
+	defaultCmd := exec.Command(cmdTS, "update", "--yes")
 	if runtime.GOOS != "linux" {
-		return exec.Command(cmdTS, "update", "--yes")
+		return defaultCmd
 	}
 	if _, err := exec.LookPath("systemd-run"); err != nil {
-		return exec.Command(cmdTS, "update", "--yes")
+		return defaultCmd
 	}
+
 	// When systemd-run is available, use it to run the update command. This
 	// creates a new temporary unit separate from the tailscaled unit. When
 	// tailscaled is restarted during the update, systemd won't kill this
 	// temporary update unit, which could cause unexpected breakage.
-	return exec.Command("systemd-run", "--wait", "--pipe", "--collect", cmdTS, "update", "--yes")
+	//
+	// We want to use the --wait flag for systemd-run, to block the update
+	// command until completion and collect output. But this flag was added in
+	// systemd 232, so we need to check the version first.
+	//
+	// The output will look like:
+	//
+	//   systemd 255 (255.7-1-arch)
+	//   +PAM +AUDIT ... other feature flags ...
+	systemdVerOut, err := exec.Command("systemd-run", "--version").Output()
+	if err != nil {
+		return defaultCmd
+	}
+	parts := strings.Fields(string(systemdVerOut))
+	if len(parts) < 2 || parts[0] != "systemd" {
+		return defaultCmd
+	}
+	systemdVer, err := strconv.Atoi(parts[1])
+	if err != nil {
+		return defaultCmd
+	}
+	if systemdVer < 232 {
+		return exec.Command("systemd-run", "--pipe", "--collect", cmdTS, "update", "--yes")
+	} else {
+		return exec.Command("systemd-run", "--wait", "--pipe", "--collect", cmdTS, "update", "--yes")
+	}
 }
 
 func regularFileExists(path string) bool {

ipn/ipnlocal/local.go

@@ -230,7 +230,8 @@ type LocalBackend struct {
 	ccGen          clientGen          // function for producing controlclient; lazily populated
 	sshServer      SSHServer          // or nil, initialized lazily.
 	appConnector   *appc.AppConnector // or nil, initialized when configured.
-	notify         func(ipn.Notify)
+	// notifyCancel cancels notifications to the current SetNotifyCallback.
+	notifyCancel   context.CancelFunc
 	cc             controlclient.Client
 	ccAuto         *controlclient.Auto // if cc is of type *controlclient.Auto
 	machinePrivKey key.MachinePrivate
@@ -258,10 +259,8 @@ type LocalBackend struct {
 	endpoints        []tailcfg.Endpoint
 	blocked          bool
 	keyExpired       bool
-	authURL          string    // cleared on Notify
-	authURLSticky    string    // not cleared on Notify
+	authURL          string    // non-empty if not Running
 	authURLTime      time.Time // when the authURL was received from the control server
-	interact         bool
 	egg              bool
 	prevIfState      *netmon.State
 	peerAPIServer    *peerAPIServer // or nil
@@ -368,7 +367,7 @@ func NewLocalBackend(logf logger.Logf, logID logid.PublicID, sys *tsd.System, lo
 	if dialer.NetMon() == nil {
 		return nil, errors.New("dialer to NewLocalBackend must have a NetMon")
 	}
-	_ = sys.MagicSock.Get() // or panic
+	mConn := sys.MagicSock.Get()
 
 	goos := envknob.GOOS()
 	if loginFlags&controlclient.LocalBackendStartKeyOSNeutral != 0 {
@@ -402,7 +401,6 @@ func NewLocalBackend(logf logger.Logf, logID logid.PublicID, sys *tsd.System, lo
 	osshare.SetFileSharingEnabled(false, logf)
 
 	ctx, cancel := context.WithCancel(context.Background())
-	portpoll := new(portlist.Poller)
 	clock := tstime.StdClock{}
 
 	b := &LocalBackend{
@@ -420,7 +418,7 @@ func NewLocalBackend(logf logger.Logf, logID logid.PublicID, sys *tsd.System, lo
 		pm:                  pm,
 		backendLogID:        logID,
 		state:               ipn.NoState,
-		portpoll:            portpoll,
+		portpoll:            new(portlist.Poller),
 		em:                  newExpiryManager(logf),
 		gotPortPollRes:      make(chan struct{}),
 		loginFlags:          loginFlags,
@@ -428,6 +426,7 @@ func NewLocalBackend(logf logger.Logf, logID logid.PublicID, sys *tsd.System, lo
 		selfUpdateProgress:  make([]ipnstate.UpdateProgress, 0),
 		lastSelfUpdateState: ipnstate.UpdateFinished,
 	}
+	mConn.SetNetInfoCallback(b.setNetInfo)
 
 	netMon := sys.NetMon.Get()
 	b.sockstatLogger, err = sockstatlog.NewLogger(logpolicy.LogsDir(logf), logf, logID, netMon, sys.HealthTracker())
@@ -440,7 +439,9 @@ func NewLocalBackend(logf logger.Logf, logID logid.PublicID, sys *tsd.System, lo
 	}
 
 	// Default filter blocks everything and logs nothing, until Start() is called.
-	b.setFilter(filter.NewAllowNone(logf, &netipx.IPSet{}))
+	noneFilter := filter.NewAllowNone(logf, &netipx.IPSet{})
+	b.setFilter(noneFilter)
+	b.e.SetJailedFilter(noneFilter)
 
 	b.setTCPPortsIntercepted(nil)
 
@@ -708,6 +709,9 @@ func (b *LocalBackend) Shutdown() {
 		b.debugSink.Close()
 		b.debugSink = nil
 	}
+	if b.notifyCancel != nil {
+		b.notifyCancel()
+	}
 	b.mu.Unlock()
 	b.webClientShutdown()
 
@@ -779,11 +783,12 @@ func (b *LocalBackend) UpdateStatus(sb *ipnstate.StatusBuilder) {
 		s.Version = version.Long()
 		s.TUN = !b.sys.IsNetstack()
 		s.BackendState = b.state.String()
-		s.AuthURL = b.authURLSticky
+		s.AuthURL = b.authURL
 		if prefs := b.pm.CurrentPrefs(); prefs.Valid() && prefs.AutoUpdate().Check {
 			s.ClientVersion = b.lastClientVersion
 		}
 		s.Health = b.health.AppendWarnings(s.Health)
+		s.HaveNodeKey = b.hasNodeKeyLocked()
 
 		// TODO(bradfitz): move this health check into a health.Warnable
 		// and remove from here.
@@ -1132,7 +1137,6 @@ func (b *LocalBackend) SetControlClientStatus(c controlclient.Client, st control
 	prefsChanged := false
 	prefs := b.pm.CurrentPrefs().AsStruct()
 	netMap := b.netMap
-	interact := b.interact
 
 	if prefs.ControlURL == "" {
 		// Once we get a message from the control plane, set
@@ -1151,7 +1155,6 @@ func (b *LocalBackend) SetControlClientStatus(c controlclient.Client, st control
 	}
 	if st.URL != "" {
 		b.authURL = st.URL
-		b.authURLSticky = st.URL
 		b.authURLTime = b.clock.Now()
 	}
 	if (wasBlocked || b.seamlessRenewalEnabled()) && st.LoginFinished() {
@@ -1269,9 +1272,7 @@ func (b *LocalBackend) SetControlClientStatus(c controlclient.Client, st control
 	}
 	if st.URL != "" {
 		b.logf("Received auth URL: %.20v...", st.URL)
-		if interact {
-			b.popBrowserAuthNow()
-		}
+		b.popBrowserAuthNow()
 	}
 	b.stateMachine()
 	// This is currently (2020-07-28) necessary; conditionally disabling it is fragile!
@@ -1555,10 +1556,26 @@ func endpointsEqual(x, y []tailcfg.Endpoint) bool {
 	return true
 }
 
+// SetNotifyCallback sets the function to call when the backend has something to
+// notify the frontend about. Only one callback can be set at a time, so calling
+// this function will replace the previous callback.
 func (b *LocalBackend) SetNotifyCallback(notify func(ipn.Notify)) {
+	ctx, cancel := context.WithCancel(b.ctx)
 	b.mu.Lock()
-	defer b.mu.Unlock()
-	b.notify = notify
+	prevCancel := b.notifyCancel
+	b.notifyCancel = cancel
+	b.mu.Unlock()
+	if prevCancel != nil {
+		prevCancel()
+	}
+
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go b.WatchNotifications(ctx, 0, wg.Done, func(n *ipn.Notify) bool {
+		notify(*n)
+		return true
+	})
+	wg.Wait()
 }
 
 // SetHTTPTestClient sets an alternate HTTP client to use with
@@ -1590,6 +1607,14 @@ func (b *LocalBackend) NodeViewByIDForTest(id tailcfg.NodeID) (_ tailcfg.NodeVie
 	return n, ok
 }
 
+// DisablePortMapperForTest disables the portmapper for tests.
+// It must be called before Start.
+func (b *LocalBackend) DisablePortMapperForTest() {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	b.portpoll = nil
+}
+
 // PeersForTest returns all the current peers, sorted by Node.ID,
 // for integration tests in another repo.
 func (b *LocalBackend) PeersForTest() []tailcfg.NodeView {
@@ -1602,9 +1627,7 @@ func (b *LocalBackend) PeersForTest() []tailcfg.NodeView {
 	return ret
 }
 
-func (b *LocalBackend) getNewControlClientFunc() clientGen {
-	b.mu.Lock()
-	defer b.mu.Unlock()
+func (b *LocalBackend) getNewControlClientFuncLocked() clientGen {
 	if b.ccGen == nil {
 		// Initialize it rather than just returning the
 		// default to make any future call to
@@ -1629,11 +1652,17 @@ func (b *LocalBackend) getNewControlClientFunc() clientGen {
 func (b *LocalBackend) Start(opts ipn.Options) error {
 	b.logf("Start")
 
-	b.mu.Lock()
+	var clientToShutdown controlclient.Client
+	defer func() {
+		if clientToShutdown != nil {
+			clientToShutdown.Shutdown()
+		}
+	}()
+	unlock := b.lockAndGetUnlock()
+	defer unlock()
 
 	if opts.UpdatePrefs != nil {
 		if err := b.checkPrefsLocked(opts.UpdatePrefs); err != nil {
-			b.mu.Unlock()
 			return err
 		}
 	}
@@ -1665,7 +1694,7 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 	// into sync with the minimal changes. But that's not how it
 	// is right now, which is a sign that the code is still too
 	// complicated.
-	prevCC := b.resetControlClientLocked()
+	clientToShutdown = b.resetControlClientLocked()
 	httpTestClient := b.httpTestClient
 
 	if b.hostinfo != nil {
@@ -1694,7 +1723,6 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 	wantRunning := prefs.WantRunning()
 	if wantRunning {
 		if err := b.initMachineKeyLocked(); err != nil {
-			b.mu.Unlock()
 			return fmt.Errorf("initMachineKeyLocked: %w", err)
 		}
 	}
@@ -1713,7 +1741,6 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 		persistv = new(persist.Persist)
 	}
 	b.updateFilterLocked(nil, ipn.PrefsView{})
-	b.mu.Unlock()
 
 	if b.portpoll != nil {
 		b.portpollOnce.Do(func() {
@@ -1745,15 +1772,11 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 		debugFlags = append([]string{"netstack"}, debugFlags...)
 	}
 
-	if prevCC != nil {
-		prevCC.Shutdown()
-	}
-
 	// TODO(apenwarr): The only way to change the ServerURL is to
 	// re-run b.Start, because this is the only place we create a
 	// new controlclient. EditPrefs allows you to overwrite ServerURL,
 	// but it won't take effect until the next Start.
-	cc, err := b.getNewControlClientFunc()(controlclient.Options{
+	cc, err := b.getNewControlClientFuncLocked()(controlclient.Options{
 		GetMachinePrivateKey:       b.createGetMachinePrivateKeyFunc(),
 		Logf:                       logger.WithPrefix(b.logf, "control: "),
 		Persist:                    *persistv,
@@ -1783,14 +1806,6 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 		return err
 	}
 
-	b.mu.Lock()
-	// Even though we reset b.cc above, we might have raced with
-	// another Start() call. If so, shut down the previous one again
-	// as we do not know if it was created with the same options.
-	prevCC = b.resetControlClientLocked()
-	if prevCC != nil {
-		defer prevCC.Shutdown() // must be called after b.mu is unlocked
-	}
 	b.setControlClientLocked(cc)
 	endpoints := b.endpoints
 
@@ -1801,33 +1816,36 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 	if b.tka != nil {
 		head, err := b.tka.authority.Head().MarshalText()
 		if err != nil {
-			b.mu.Unlock()
 			return fmt.Errorf("marshalling tka head: %w", err)
 		}
 		tkaHead = string(head)
 	}
 	confWantRunning := b.conf != nil && wantRunning
-	b.mu.Unlock()
 
 	if endpoints != nil {
 		cc.UpdateEndpoints(endpoints)
 	}
 	cc.SetTKAHead(tkaHead)
 
-	b.MagicConn().SetNetInfoCallback(b.setNetInfo)
-
 	blid := b.backendLogID.String()
 	b.logf("Backend: logs: be:%v fe:%v", blid, opts.FrontendLogID)
-	b.send(ipn.Notify{BackendLogID: &blid})
-	b.send(ipn.Notify{Prefs: &prefs})
+	b.sendLocked(ipn.Notify{
+		BackendLogID: &blid,
+		Prefs:        &prefs,
+	})
 
-	if !loggedOut && (b.hasNodeKey() || confWantRunning) {
-		// Even if !WantRunning, we should verify our key, if there
-		// is one. If you want tailscaled to be completely idle,
-		// use logout instead.
-		cc.Login(nil, controlclient.LoginDefault)
+	if !loggedOut && (b.hasNodeKeyLocked() || confWantRunning) {
+		// If we know that we're either logged in or meant to be
+		// running, tell the controlclient that it should also assume
+		// that we need to be logged in.
+		//
+		// Without this, the state machine transitions to "NeedsLogin" implying
+		// that user interaction is required, which is not the case and can
+		// regress tsnet.Server restarts.
+		cc.Login(controlclient.LoginDefault)
 	}
-	b.stateMachine()
+	b.stateMachineLockedOnEntry(unlock)
+
 	return nil
 }
 
@@ -1935,7 +1953,9 @@ func (b *LocalBackend) updateFilterLocked(netMap *netmap.NetworkMap, prefs ipn.P
 
 	if !haveNetmap {
 		b.logf("[v1] netmap packet filter: (not ready yet)")
-		b.setFilter(filter.NewAllowNone(b.logf, logNets))
+		noneFilter := filter.NewAllowNone(b.logf, logNets)
+		b.setFilter(noneFilter)
+		b.e.SetJailedFilter(noneFilter)
 		return
 	}
 
@@ -1947,6 +1967,9 @@ func (b *LocalBackend) updateFilterLocked(netMap *netmap.NetworkMap, prefs ipn.P
 		b.logf("[v1] netmap packet filter: %v filters", len(packetFilter))
 		b.setFilter(filter.New(packetFilter, localNets, logNets, oldFilter, b.logf))
 	}
+	// The filter for a jailed node is the exact same as a ShieldsUp filter.
+	oldJailedFilter := b.e.GetJailedFilter()
+	b.e.SetJailedFilter(filter.NewShieldsUpFilter(localNets, logNets, oldJailedFilter, b.logf))
 
 	if b.sshServer != nil {
 		go b.sshServer.OnPolicyChange()
@@ -2252,8 +2275,8 @@ func (b *LocalBackend) WatchNotifications(ctx context.Context, mask ipn.NotifyWa
 		if mask&ipn.NotifyInitialState != 0 {
 			ini.SessionID = sessionID
 			ini.State = ptr.To(b.state)
-			if b.state == ipn.NeedsLogin {
-				ini.BrowseToURL = ptr.To(b.authURLSticky)
+			if b.state == ipn.NeedsLogin && b.authURL != "" {
+				ini.BrowseToURL = ptr.To(b.authURL)
 			}
 		}
 		if mask&ipn.NotifyInitialPrefs != 0 {
@@ -2307,11 +2330,27 @@ func (b *LocalBackend) WatchNotifications(ctx context.Context, mask ipn.NotifyWa
 	// TODO(marwan-at-work): streaming background logs?
 	defer b.DeleteForegroundSession(sessionID)
 
+	var lastURLPop string // to dup suppress URL popups
 	for {
 		select {
 		case <-ctx.Done():
 			return
 		case n, ok := <-ch:
+			// URLs flow into Notify.BrowseToURL via two means:
+			//    1. From MapResponse.PopBrowserURL, which already says they're dup
+			//       suppressed if identical, and that's done by the controlclient,
+			//       so this added later adds nothing.
+			//
+			//    2. From the controlclient auth routes, on register. This makes sure
+			//       we don't tell clients (mac, windows, android) to pop the same URL
+			//       multiple times.
+			if n != nil && n.BrowseToURL != nil {
+				if v := *n.BrowseToURL; v == lastURLPop {
+					n.BrowseToURL = nil
+				} else {
+					lastURLPop = v
+				}
+			}
 			if !ok || !fn(n) {
 				return
 			}
@@ -2388,6 +2427,13 @@ func (b *LocalBackend) DebugPickNewDERP() error {
 //
 // b.mu must not be held.
 func (b *LocalBackend) send(n ipn.Notify) {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	b.sendLocked(n)
+}
+
+// sendLocked is like send, but assumes b.mu is already held.
+func (b *LocalBackend) sendLocked(n ipn.Notify) {
 	if n.Prefs != nil {
 		n.Prefs = ptr.To(stripKeysFromPrefs(*n.Prefs))
 	}
@@ -2395,8 +2441,6 @@ func (b *LocalBackend) send(n ipn.Notify) {
 		n.Version = version.Long()
 	}
 
-	b.mu.Lock()
-	notifyFunc := b.notify
 	apiSrv := b.peerAPIServer
 	if mayDeref(apiSrv).taildrop.HasFilesWaiting() {
 		n.FilesWaiting = &empty.Message{}
@@ -2409,12 +2453,6 @@ func (b *LocalBackend) send(n ipn.Notify) {
 			// Drop the notification if the channel is full.
 		}
 	}
-
-	b.mu.Unlock()
-
-	if notifyFunc != nil {
-		notifyFunc(n)
-	}
 }
 
 func (b *LocalBackend) sendFileNotify() {
@@ -2424,9 +2462,8 @@ func (b *LocalBackend) sendFileNotify() {
 	for _, wakeWaiter := range b.fileWaiters {
 		wakeWaiter()
 	}
-	notifyFunc := b.notify
 	apiSrv := b.peerAPIServer
-	if notifyFunc == nil || apiSrv == nil {
+	if apiSrv == nil {
 		b.mu.Unlock()
 		return
 	}
@@ -2449,16 +2486,15 @@ func (b *LocalBackend) sendFileNotify() {
 func (b *LocalBackend) popBrowserAuthNow() {
 	b.mu.Lock()
 	url := b.authURL
-	b.interact = false
-	b.authURL = "" // but NOT clearing authURLSticky
+	expired := b.keyExpired
 	b.mu.Unlock()
 
-	b.logf("popBrowserAuthNow: url=%v, key-expired=%v, seamless-key-renewal=%v", url != "", b.keyExpired, b.seamlessRenewalEnabled())
+	b.logf("popBrowserAuthNow: url=%v, key-expired=%v, seamless-key-renewal=%v", url != "", expired, b.seamlessRenewalEnabled())
 
 	// Deconfigure the local network data plane if:
 	// - seamless key renewal is not enabled;
 	// - key is expired (in which case tailnet connectivity is down anyway).
-	if !b.seamlessRenewalEnabled() || b.keyExpired {
+	if !b.seamlessRenewalEnabled() || expired {
 		b.blockEngineUpdates(true)
 		b.stopEngineAndWait()
 	}
@@ -2777,7 +2813,6 @@ func (b *LocalBackend) StartLoginInteractive(ctx context.Context) error {
 	if b.cc == nil {
 		panic("LocalBackend.assertClient: b.cc == nil")
 	}
-	b.interact = true
 	url := b.authURL
 	timeSinceAuthURLCreated := b.clock.Since(b.authURLTime)
 	cc := b.cc
@@ -2790,7 +2825,7 @@ func (b *LocalBackend) StartLoginInteractive(ctx context.Context) error {
 	if url != "" && timeSinceAuthURLCreated < ((7*24*time.Hour)-(1*time.Hour)) {
 		b.popBrowserAuthNow()
 	} else {
-		cc.Login(nil, b.loginFlags|controlclient.LoginInteractive)
+		cc.Login(b.loginFlags | controlclient.LoginInteractive)
 	}
 	return nil
 }
@@ -3304,7 +3339,7 @@ func (b *LocalBackend) setPrefsLockedOnEntry(newp *ipn.Prefs, unlock unlockOnce)
 
 	if !oldp.WantRunning() && newp.WantRunning {
 		b.logf("transitioning to running; doing Login...")
-		cc.Login(nil, controlclient.LoginDefault)
+		cc.Login(controlclient.LoginDefault)
 	}
 
 	if oldp.WantRunning() != newp.WantRunning {
@@ -3590,6 +3625,7 @@ func (b *LocalBackend) authReconfig() {
 	nm := b.netMap
 	hasPAC := b.prevIfState.HasPAC()
 	disableSubnetsIfPAC := nm.HasCap(tailcfg.NodeAttrDisableSubnetsIfPAC)
+	userDialUseRoutes := nm.HasCap(tailcfg.NodeAttrUserDialUseRoutes)
 	dohURL, dohURLOK := exitNodeCanProxyDNS(nm, b.peers, prefs.ExitNodeID())
 	dcfg := dnsConfigForNetmap(nm, b.peers, prefs, b.logf, version.OS())
 	// If the current node is an app connector, ensure the app connector machine is started
@@ -3613,9 +3649,6 @@ func (b *LocalBackend) authReconfig() {
 	if prefs.RouteAll() {
 		flags |= netmap.AllowSubnetRoutes
 	}
-	if prefs.AllowSingleHosts() {
-		flags |= netmap.AllowSingleHosts
-	}
 	if hasPAC && disableSubnetsIfPAC {
 		if flags&netmap.AllowSubnetRoutes != 0 {
 			b.logf("authReconfig: have PAC; disabling subnet routes")
@@ -3647,6 +3680,12 @@ func (b *LocalBackend) authReconfig() {
 	}
 	b.logf("[v1] authReconfig: ra=%v dns=%v 0x%02x: %v", prefs.RouteAll(), prefs.CorpDNS(), flags, err)
 
+	if userDialUseRoutes {
+		b.dialer.SetRoutes(rcfg.Routes, rcfg.LocalRoutes)
+	} else {
+		b.dialer.SetRoutes(nil, nil)
+	}
+
 	b.initPeerAPIListener()
 }
 
@@ -4146,13 +4185,22 @@ func (b *LocalBackend) routerConfig(cfg *wgcfg.Config, prefs ipn.PrefsView, oneC
 		netfilterKind = prefs.NetfilterKind()
 	}
 
+	var doStatefulFiltering bool
+	if v, ok := prefs.NoStatefulFiltering().Get(); ok && !v {
+		// The preferences explicitly "do stateful filtering" is turned
+		// off, or to expand the double negative, to do stateful
+		// filtering. Do so.
+		doStatefulFiltering = true
+	}
+
 	rs := &router.Config{
-		LocalAddrs:       unmapIPPrefixes(cfg.Addresses),
-		SubnetRoutes:     unmapIPPrefixes(prefs.AdvertiseRoutes().AsSlice()),
-		SNATSubnetRoutes: !prefs.NoSNAT(),
-		NetfilterMode:    prefs.NetfilterMode(),
-		Routes:           peerRoutes(b.logf, cfg.Peers, singleRouteThreshold),
-		NetfilterKind:    netfilterKind,
+		LocalAddrs:        unmapIPPrefixes(cfg.Addresses),
+		SubnetRoutes:      unmapIPPrefixes(prefs.AdvertiseRoutes().AsSlice()),
+		SNATSubnetRoutes:  !prefs.NoSNAT(),
+		StatefulFiltering: doStatefulFiltering,
+		NetfilterMode:     prefs.NetfilterMode(),
+		Routes:            peerRoutes(b.logf, cfg.Peers, singleRouteThreshold),
+		NetfilterKind:     netfilterKind,
 	}
 
 	if distro.Get() == distro.Synology {
@@ -4292,7 +4340,6 @@ func (b *LocalBackend) enterStateLockedOnEntry(newState ipn.State, unlock unlock
 	authURL := b.authURL
 	if newState == ipn.Running {
 		b.authURL = ""
-		b.authURLSticky = ""
 		b.authURLTime = time.Time{}
 	} else if oldState == ipn.Running {
 		// Transitioning away from running.
@@ -4347,14 +4394,6 @@ func (b *LocalBackend) enterStateLockedOnEntry(newState ipn.State, unlock unlock
 	}
 }
 
-// hasNodeKey reports whether a non-zero node key is present in the current
-// prefs.
-func (b *LocalBackend) hasNodeKey() bool {
-	b.mu.Lock()
-	defer b.mu.Unlock()
-	return b.hasNodeKeyLocked()
-}
-
 func (b *LocalBackend) hasNodeKeyLocked() bool {
 	// we can't use b.Prefs(), because it strips the keys, oops!
 	p := b.pm.CurrentPrefs()
@@ -4452,6 +4491,12 @@ func (b *LocalBackend) nextStateLocked() ipn.State {
 // Or maybe just call the state machine from fewer places.
 func (b *LocalBackend) stateMachine() {
 	unlock := b.lockAndGetUnlock()
+	b.stateMachineLockedOnEntry(unlock)
+}
+
+// stateMachineLockedOnEntry is like stateMachine but requires b.mu be held to
+// call it, but it unlocks b.mu when done (via unlock, a once func).
+func (b *LocalBackend) stateMachineLockedOnEntry(unlock unlockOnce) {
 	b.enterStateLockedOnEntry(b.nextStateLocked(), unlock)
 }
 
@@ -4553,6 +4598,8 @@ func (b *LocalBackend) resetControlClientLocked() controlclient.Client {
 		return nil
 	}
 
+	b.authURL = ""
+
 	// When we clear the control client, stop any outstanding netmap expiry
 	// timer; synthesizing a new netmap while we don't have a control
 	// client will break things.
@@ -4597,7 +4644,6 @@ func (b *LocalBackend) ResetForClientDisconnect() {
 	}
 	b.keyExpired = false
 	b.authURL = ""
-	b.authURLSticky = ""
 	b.authURLTime = time.Time{}
 	b.activeLogin = ""
 	b.resetDialPlan()
@@ -6368,7 +6414,7 @@ func (b *LocalBackend) SuggestExitNode() (response apitype.ExitNodeSuggestionRes
 	lastSuggestedExitNode := b.lastSuggestedExitNode
 	b.mu.Unlock()
 	if lastReport == nil || netMap == nil {
-		last, err := suggestLastExitNode(lastSuggestedExitNode)
+		last, err := lastSuggestedExitNode.asAPIType()
 		if err != nil {
 			return response, ErrCannotSuggestExitNode
 		}
@@ -6378,7 +6424,7 @@ func (b *LocalBackend) SuggestExitNode() (response apitype.ExitNodeSuggestionRes
 	r := rand.New(rand.NewSource(seed))
 	res, err := suggestExitNode(lastReport, netMap, r)
 	if err != nil {
-		last, err := suggestLastExitNode(lastSuggestedExitNode)
+		last, err := lastSuggestedExitNode.asAPIType()
 		if err != nil {
 			return response, ErrCannotSuggestExitNode
 		}
@@ -6391,12 +6437,13 @@ func (b *LocalBackend) SuggestExitNode() (response apitype.ExitNodeSuggestionRes
 	return res, err
 }
 
-// suggestLastExitNode formats a response with the last suggested exit node's ID and name.
+// asAPIType formats a response with the last suggested exit node's ID and name.
+// Returns error if there is no id or name.
 // Used as a fallback before returning a nil response and error.
-func suggestLastExitNode(lastSuggestedExitNode lastSuggestedExitNode) (res apitype.ExitNodeSuggestionResponse, err error) {
-	if lastSuggestedExitNode.id != "" && lastSuggestedExitNode.name != "" {
-		res.ID = lastSuggestedExitNode.id
-		res.Name = lastSuggestedExitNode.name
+func (n lastSuggestedExitNode) asAPIType() (res apitype.ExitNodeSuggestionResponse, _ error) {
+	if n.id != "" && n.name != "" {
+		res.ID = n.id
+		res.Name = n.name
 		return res, nil
 	}
 	return res, ErrUnableToSuggestLastExitNode
@@ -6406,8 +6453,17 @@ func suggestExitNode(report *netcheck.Report, netMap *netmap.NetworkMap, r *rand
 	if report.PreferredDERP == 0 {
 		return res, ErrNoPreferredDERP
 	}
+	var allowedCandidates set.Set[string]
+	if allowed, err := syspolicy.GetStringArray(syspolicy.AllowedSuggestedExitNodes, nil); err != nil {
+		return res, fmt.Errorf("unable to read %s policy: %w", syspolicy.AllowedSuggestedExitNodes, err)
+	} else if allowed != nil && len(allowed) > 0 {
+		allowedCandidates = set.SetOf(allowed)
+	}
 	candidates := make([]tailcfg.NodeView, 0, len(netMap.Peers))
 	for _, peer := range netMap.Peers {
+		if allowedCandidates != nil && !allowedCandidates.Contains(string(peer.StableID())) {
+			continue
+		}
 		if peer.CapMap().Has(tailcfg.NodeAttrSuggestExitNode) && tsaddr.ContainsExitRoutes(peer.AllowedIPs()) {
 			candidates = append(candidates, peer)
 		}

ipn/ipnlocal/local_test.go

@@ -431,7 +431,7 @@ func newTestLocalBackend(t testing.TB) *LocalBackend {
 	sys := new(tsd.System)
 	store := new(mem.Store)
 	sys.Set(store)
-	eng, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set)
+	eng, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set, sys.HealthTracker())
 	if err != nil {
 		t.Fatalf("NewFakeUserspaceEngine: %v", err)
 	}
@@ -1595,6 +1595,9 @@ type mockSyspolicyHandler struct {
 	// queried by the current test. If the policy is expected but unset, then
 	// use nil, otherwise use a string equal to the policy's desired value.
 	stringPolicies map[syspolicy.Key]*string
+	// stringArrayPolicies is the collection of policies that we expected to see
+	// queries by the current test, that return policy string arrays.
+	stringArrayPolicies map[syspolicy.Key][]string
 	// failUnknownPolicies is set if policies other than those in stringPolicies
 	// (uint64 or bool policies are not supported by mockSyspolicyHandler yet)
 	// should be considered a test failure if they are queried.
@@ -1632,6 +1635,12 @@ func (h *mockSyspolicyHandler) ReadStringArray(key string) ([]string, error) {
 	if h.failUnknownPolicies {
 		h.t.Errorf("ReadStringArray(%q) unexpectedly called", key)
 	}
+	if s, ok := h.stringArrayPolicies[syspolicy.Key(key)]; ok {
+		if s == nil {
+			return []string{}, syspolicy.ErrNoSuchKey
+		}
+		return s, nil
+	}
 	return nil, syspolicy.ErrNoSuchKey
 }
 
@@ -3438,7 +3447,7 @@ func TestMinLatencyDERPregion(t *testing.T) {
 	}
 }
 
-func TestSuggestLastExitNode(t *testing.T) {
+func TestLastSuggestedExitNodeAsAPIType(t *testing.T) {
 	tests := []struct {
 		name                      string
 		lastSuggestedExitNode     lastSuggestedExitNode
@@ -3460,7 +3469,7 @@ func TestSuggestLastExitNode(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			got, err := suggestLastExitNode(tt.lastSuggestedExitNode)
+			got, err := tt.lastSuggestedExitNode.asAPIType()
 			if got != tt.wantRes || err != tt.wantErr {
 				t.Errorf("got %v error %v, want %v error %v", got, err, tt.wantRes, tt.wantErr)
 			}
@@ -3472,8 +3481,9 @@ func TestLocalBackendSuggestExitNode(t *testing.T) {
 	tests := []struct {
 		name                      string
 		lastSuggestedExitNode     lastSuggestedExitNode
-		report                    netcheck.Report
+		report                    *netcheck.Report
 		netMap                    netmap.NetworkMap
+		allowedSuggestedExitNodes []string
 		wantID                    tailcfg.StableNodeID
 		wantName                  string
 		wantErr                   error
@@ -3482,7 +3492,7 @@ func TestLocalBackendSuggestExitNode(t *testing.T) {
 		{
 			name:                  "nil netmap, returns last suggested exit node",
 			lastSuggestedExitNode: lastSuggestedExitNode{name: "test", id: "test"},
-			report: netcheck.Report{
+			report: &netcheck.Report{
 				RegionLatency: map[int]time.Duration{
 					1: 0,
 					2: -1,
@@ -3518,7 +3528,7 @@ func TestLocalBackendSuggestExitNode(t *testing.T) {
 		{
 			name:                  "found better derp node, last suggested exit node updates",
 			lastSuggestedExitNode: lastSuggestedExitNode{name: "test", id: "test"},
-			report: netcheck.Report{
+			report: &netcheck.Report{
 				RegionLatency: map[int]time.Duration{
 					1: 10,
 					2: 10,
@@ -3574,7 +3584,7 @@ func TestLocalBackendSuggestExitNode(t *testing.T) {
 		{
 			name:                  "found better mullvad node, last suggested exit node updates",
 			lastSuggestedExitNode: lastSuggestedExitNode{name: "San Jose", id: "3"},
-			report: netcheck.Report{
+			report: &netcheck.Report{
 				RegionLatency: map[int]time.Duration{
 					1: 0,
 					2: 0,
@@ -3645,7 +3655,7 @@ func TestLocalBackendSuggestExitNode(t *testing.T) {
 		{
 			name:                  "ErrNoPreferredDERP, use last suggested exit node",
 			lastSuggestedExitNode: lastSuggestedExitNode{name: "test", id: "test"},
-			report: netcheck.Report{
+			report: &netcheck.Report{
 				RegionLatency: map[int]time.Duration{
 					1: 10,
 					2: 10,
@@ -3701,7 +3711,7 @@ func TestLocalBackendSuggestExitNode(t *testing.T) {
 		{
 			name:                  "ErrNoPreferredDERP, use last suggested exit node",
 			lastSuggestedExitNode: lastSuggestedExitNode{name: "test", id: "test"},
-			report: netcheck.Report{
+			report: &netcheck.Report{
 				RegionLatency: map[int]time.Duration{
 					1: 10,
 					2: 10,
@@ -3756,7 +3766,7 @@ func TestLocalBackendSuggestExitNode(t *testing.T) {
 		},
 		{
 			name: "unable to use last suggested exit node",
-			report: netcheck.Report{
+			report: &netcheck.Report{
 				RegionLatency: map[int]time.Duration{
 					1: 10,
 					2: 10,
@@ -3766,13 +3776,141 @@ func TestLocalBackendSuggestExitNode(t *testing.T) {
 			},
 			wantErr: ErrCannotSuggestExitNode,
 		},
+		{
+			name:                  "only pick from allowed suggested exit nodes",
+			lastSuggestedExitNode: lastSuggestedExitNode{name: "test", id: "test"},
+			report: &netcheck.Report{
+				RegionLatency: map[int]time.Duration{
+					1: 10,
+					2: 10,
+					3: 5,
+				},
+				PreferredDERP: 1,
+			},
+			netMap: netmap.NetworkMap{
+				SelfNode: (&tailcfg.Node{
+					Addresses: []netip.Prefix{
+						netip.MustParsePrefix("100.64.1.1/32"),
+						netip.MustParsePrefix("fe70::1/128"),
+					},
+				}).View(),
+				DERPMap: &tailcfg.DERPMap{
+					Regions: map[int]*tailcfg.DERPRegion{
+						1: {},
+						2: {},
+						3: {},
+					},
+				},
+				Peers: []tailcfg.NodeView{
+					(&tailcfg.Node{
+						ID:       2,
+						StableID: "test",
+						Name:     "test",
+						DERP:     "127.3.3.40:1",
+						AllowedIPs: []netip.Prefix{
+							netip.MustParsePrefix("0.0.0.0/0"), netip.MustParsePrefix("::/0"),
+						},
+						CapMap: (tailcfg.NodeCapMap)(map[tailcfg.NodeCapability][]tailcfg.RawMessage{
+							tailcfg.NodeAttrSuggestExitNode: {},
+							tailcfg.NodeAttrAutoExitNode:    {},
+						}),
+					}).View(),
+					(&tailcfg.Node{
+						ID:       3,
+						StableID: "foo",
+						Name:     "foo",
+						DERP:     "127.3.3.40:3",
+						AllowedIPs: []netip.Prefix{
+							netip.MustParsePrefix("0.0.0.0/0"), netip.MustParsePrefix("::/0"),
+						},
+						CapMap: (tailcfg.NodeCapMap)(map[tailcfg.NodeCapability][]tailcfg.RawMessage{
+							tailcfg.NodeAttrSuggestExitNode: {},
+							tailcfg.NodeAttrAutoExitNode:    {},
+						}),
+					}).View(),
+				},
+			},
+			allowedSuggestedExitNodes: []string{"test"},
+			wantID:                    "test",
+			wantName:                  "test",
+			wantLastSuggestedExitNode: lastSuggestedExitNode{name: "test", id: "test"},
+		},
+		{
+			name:                  "allowed suggested exit nodes not nil but length 0",
+			lastSuggestedExitNode: lastSuggestedExitNode{name: "test", id: "test"},
+			report: &netcheck.Report{
+				RegionLatency: map[int]time.Duration{
+					1: 10,
+					2: 10,
+					3: 5,
+				},
+				PreferredDERP: 1,
+			},
+			netMap: netmap.NetworkMap{
+				SelfNode: (&tailcfg.Node{
+					Addresses: []netip.Prefix{
+						netip.MustParsePrefix("100.64.1.1/32"),
+						netip.MustParsePrefix("fe70::1/128"),
+					},
+				}).View(),
+				DERPMap: &tailcfg.DERPMap{
+					Regions: map[int]*tailcfg.DERPRegion{
+						1: {},
+						2: {},
+						3: {},
+					},
+				},
+				Peers: []tailcfg.NodeView{
+					(&tailcfg.Node{
+						ID:       2,
+						StableID: "test",
+						Name:     "test",
+						DERP:     "127.3.3.40:1",
+						AllowedIPs: []netip.Prefix{
+							netip.MustParsePrefix("0.0.0.0/0"), netip.MustParsePrefix("::/0"),
+						},
+						CapMap: (tailcfg.NodeCapMap)(map[tailcfg.NodeCapability][]tailcfg.RawMessage{
+							tailcfg.NodeAttrSuggestExitNode: {},
+							tailcfg.NodeAttrAutoExitNode:    {},
+						}),
+					}).View(),
+					(&tailcfg.Node{
+						ID:       3,
+						StableID: "foo",
+						Name:     "foo",
+						DERP:     "127.3.3.40:3",
+						AllowedIPs: []netip.Prefix{
+							netip.MustParsePrefix("0.0.0.0/0"), netip.MustParsePrefix("::/0"),
+						},
+						CapMap: (tailcfg.NodeCapMap)(map[tailcfg.NodeCapability][]tailcfg.RawMessage{
+							tailcfg.NodeAttrSuggestExitNode: {},
+							tailcfg.NodeAttrAutoExitNode:    {},
+						}),
+					}).View(),
+				},
+			},
+			allowedSuggestedExitNodes: []string{},
+			wantID:                    "foo",
+			wantName:                  "foo",
+			wantLastSuggestedExitNode: lastSuggestedExitNode{name: "foo", id: "foo"},
+		},
 	}
 
 	for _, tt := range tests {
 		lb := newTestLocalBackend(t)
+		msh := &mockSyspolicyHandler{
+			t: t,
+			stringArrayPolicies: map[syspolicy.Key][]string{
+				syspolicy.AllowedSuggestedExitNodes: nil,
+			},
+		}
+		if len(tt.allowedSuggestedExitNodes) != 0 {
+			msh.stringArrayPolicies[syspolicy.AllowedSuggestedExitNodes] = tt.allowedSuggestedExitNodes
+		}
+		syspolicy.SetHandlerForTest(t, msh)
 		lb.lastSuggestedExitNode = tt.lastSuggestedExitNode
 		lb.netMap = &tt.netMap
-		lb.sys.MagicSock.Get().SetLastNetcheckReport(context.Background(), tt.report)
+		lb.sys.MagicSock.Get().SetLastNetcheckReportForTest(context.Background(), tt.report)
 		got, err := lb.SuggestExitNode()
 		if got.ID != tt.wantID {
 			t.Errorf("ID=%v, want=%v", got.ID, tt.wantID)

ipn/ipnlocal/loglines_test.go

@@ -50,7 +50,7 @@ func TestLocalLogLines(t *testing.T) {
 	sys := new(tsd.System)
 	store := new(mem.Store)
 	sys.Set(store)
-	e, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set)
+	e, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set, sys.HealthTracker())
 	if err != nil {
 		t.Fatal(err)
 	}

ipn/ipnlocal/network-lock.go

@@ -18,6 +18,7 @@ import (
 	"net/netip"
 	"os"
 	"path/filepath"
+	"slices"
 	"time"
 
 	"tailscale.com/health/healthmsg"
@@ -27,10 +28,12 @@ import (
 	"tailscale.com/tailcfg"
 	"tailscale.com/tka"
 	"tailscale.com/types/key"
+	"tailscale.com/types/logger"
 	"tailscale.com/types/netmap"
 	"tailscale.com/types/persist"
 	"tailscale.com/types/tkatype"
 	"tailscale.com/util/mak"
+	"tailscale.com/util/set"
 )
 
 // TODO(tom): RPC retry/backoff was broken and has been removed. Fix?
@@ -66,6 +69,7 @@ func (b *LocalBackend) tkaFilterNetmapLocked(nm *netmap.NetworkMap) {
 		return // TKA not enabled.
 	}
 
+	tracker := rotationTracker{logf: b.logf}
 	var toDelete map[int]bool // peer index => true
 	for i, p := range nm.Peers {
 		if p.UnsignedPeerAPIOnly() {
@@ -76,21 +80,32 @@ func (b *LocalBackend) tkaFilterNetmapLocked(nm *netmap.NetworkMap) {
 			b.logf("Network lock is dropping peer %v(%v) due to missing signature", p.ID(), p.StableID())
 			mak.Set(&toDelete, i, true)
 		} else {
-			if err := b.tka.authority.NodeKeyAuthorized(p.Key(), p.KeySignature().AsSlice()); err != nil {
+			details, err := b.tka.authority.NodeKeyAuthorizedWithDetails(p.Key(), p.KeySignature().AsSlice())
+			if err != nil {
 				b.logf("Network lock is dropping peer %v(%v) due to failed signature check: %v", p.ID(), p.StableID(), err)
 				mak.Set(&toDelete, i, true)
+				continue
+			}
+			if details != nil {
+				// Rotation details are returned when the node key is signed by a valid SigRotation signature.
+				tracker.addRotationDetails(p.Key(), details)
 			}
 		}
 	}
 
+	obsoleteByRotation := tracker.obsoleteKeys()
+
 	// nm.Peers is ordered, so deletion must be order-preserving.
-	if len(toDelete) > 0 {
+	if len(toDelete) > 0 || len(obsoleteByRotation) > 0 {
 		peers := make([]tailcfg.NodeView, 0, len(nm.Peers))
-		filtered := make([]ipnstate.TKAFilteredPeer, 0, len(toDelete))
+		filtered := make([]ipnstate.TKAFilteredPeer, 0, len(toDelete)+len(obsoleteByRotation))
 		for i, p := range nm.Peers {
-			if !toDelete[i] {
+			if !toDelete[i] && !obsoleteByRotation.Contains(p.Key()) {
 				peers = append(peers, p)
 			} else {
+				if obsoleteByRotation.Contains(p.Key()) {
+					b.logf("Network lock is dropping peer %v(%v) due to key rotation", p.ID(), p.StableID())
+				}
 				// Record information about the node we filtered out.
 				fp := ipnstate.TKAFilteredPeer{
 					Name:         p.Name(),
@@ -122,6 +137,84 @@ func (b *LocalBackend) tkaFilterNetmapLocked(nm *netmap.NetworkMap) {
 	}
 }
 
+// rotationTracker determines the set of node keys that are made obsolete by key
+// rotation.
+//   - for each SigRotation signature, all previous node keys referenced by the
+//     nested signatures are marked as obsolete.
+//   - if there are multiple SigRotation signatures tracing back to the same
+//     wrapping pubkey (e.g. if a node is cloned with all its keys), we keep
+//     just one of them, marking the others as obsolete.
+type rotationTracker struct {
+	// obsolete is the set of node keys that are obsolete due to key rotation.
+	// users of rotationTracker should use the obsoleteKeys method for complete results.
+	obsolete set.Set[key.NodePublic]
+
+	// byWrappingKey keeps track of rotation details per wrapping pubkey.
+	byWrappingKey map[string][]sigRotationDetails
+
+	logf logger.Logf
+}
+
+// sigRotationDetails holds information about a node key signed by a SigRotation.
+type sigRotationDetails struct {
+	np          key.NodePublic
+	numPrevKeys int
+}
+
+// addRotationDetails records the rotation signature details for a node key.
+func (r *rotationTracker) addRotationDetails(np key.NodePublic, d *tka.RotationDetails) {
+	r.obsolete.Make()
+	r.obsolete.AddSlice(d.PrevNodeKeys)
+	rd := sigRotationDetails{
+		np:          np,
+		numPrevKeys: len(d.PrevNodeKeys),
+	}
+	if r.byWrappingKey == nil {
+		r.byWrappingKey = make(map[string][]sigRotationDetails)
+	}
+	wp := string(d.WrappingPubkey)
+	r.byWrappingKey[wp] = append(r.byWrappingKey[wp], rd)
+}
+
+// obsoleteKeys returns the set of node keys that are obsolete due to key rotation.
+func (r *rotationTracker) obsoleteKeys() set.Set[key.NodePublic] {
+	for _, v := range r.byWrappingKey {
+		// If there are multiple rotation signatures with the same wrapping
+		// pubkey, we need to decide which one is the "latest", and keep it.
+		// The signature with the largest number of previous keys is likely to
+		// be the latest, unless it has been marked as obsolete (rotated out) by
+		// another signature (which might happen in the future if we start
+		// compacting long rotated signature chains).
+		slices.SortStableFunc(v, func(a, b sigRotationDetails) int {
+			// Group all obsolete keys after non-obsolete keys.
+			if ao, bo := r.obsolete.Contains(a.np), r.obsolete.Contains(b.np); ao != bo {
+				if ao {
+					return 1
+				}
+				return -1
+			}
+			// Sort by decreasing number of previous keys.
+			return b.numPrevKeys - a.numPrevKeys
+		})
+		// If there are several signatures with the same number of previous
+		// keys, we cannot determine which one is the latest, so all of them are
+		// rejected for safety.
+		if len(v) >= 2 && v[0].numPrevKeys == v[1].numPrevKeys {
+			r.logf("at least two nodes (%s and %s) have equally valid rotation signatures with the same wrapping pubkey, rejecting", v[0].np, v[1].np)
+			for _, rd := range v {
+				r.obsolete.Add(rd.np)
+			}
+		} else {
+			// The first key in v is the one with the longest chain of previous
+			// keys, so it must be the newest one. Mark all older keys as obsolete.
+			for _, rd := range v[1:] {
+				r.obsolete.Add(rd.np)
+			}
+		}
+	}
+	return r.obsolete
+}
+
 // tkaSyncIfNeeded examines TKA info reported from the control plane,
 // performing the steps necessary to synchronize local tka state.
 //
@@ -423,8 +516,12 @@ func (b *LocalBackend) NetworkLockStatus() *ipnstate.NetworkLockStatus {
 	copy(head[:], h[:])
 
 	var selfAuthorized bool
+	nodeKeySignature := &tka.NodeKeySignature{}
 	if b.netMap != nil {
 		selfAuthorized = b.tka.authority.NodeKeyAuthorized(b.netMap.SelfNode.Key(), b.netMap.SelfNode.KeySignature().AsSlice()) == nil
+		if err := nodeKeySignature.Unserialize(b.netMap.SelfNode.KeySignature().AsSlice()); err != nil {
+			b.logf("failed to decode self node key signature: %v", err)
+		}
 	}
 
 	keys := b.tka.authority.Keys()
@@ -445,14 +542,15 @@ func (b *LocalBackend) NetworkLockStatus() *ipnstate.NetworkLockStatus {
 	stateID1, _ := b.tka.authority.StateIDs()
 
 	return &ipnstate.NetworkLockStatus{
-		Enabled:       true,
-		Head:          &head,
-		PublicKey:     nlPriv.Public(),
-		NodeKey:       nodeKey,
-		NodeKeySigned: selfAuthorized,
-		TrustedKeys:   outKeys,
-		FilteredPeers: filtered,
-		StateID:       stateID1,
+		Enabled:          true,
+		Head:             &head,
+		PublicKey:        nlPriv.Public(),
+		NodeKey:          nodeKey,
+		NodeKeySigned:    selfAuthorized,
+		NodeKeySignature: nodeKeySignature,
+		TrustedKeys:      outKeys,
+		FilteredPeers:    filtered,
+		StateID:          stateID1,
 	}
 }
 

ipn/ipnlocal/network-lock_test.go

@@ -13,8 +13,11 @@ import (
 	"net/http/httptest"
 	"os"
 	"path/filepath"
+	"reflect"
 	"testing"
 
+	go4mem "go4.org/mem"
+
 	"github.com/google/go-cmp/cmp"
 	"tailscale.com/control/controlclient"
 	"tailscale.com/health"
@@ -30,6 +33,7 @@ import (
 	"tailscale.com/types/persist"
 	"tailscale.com/types/tkatype"
 	"tailscale.com/util/must"
+	"tailscale.com/util/set"
 )
 
 type observerFunc func(controlclient.Status)
@@ -563,18 +567,32 @@ func TestTKAFilterNetmap(t *testing.T) {
 	}
 	n4Sig.Signature[3] = 42 // mess up the signature
 	n4Sig.Signature[4] = 42 // mess up the signature
-	n5GoodSig, err := signNodeKey(tailcfg.TKASignInfo{NodePublic: n5.Public()}, nlPriv)
+
+	n5nl := key.NewNLPrivate()
+	n5InitialSig, err := signNodeKey(tailcfg.TKASignInfo{NodePublic: n5.Public(), RotationPubkey: n5nl.Public().Verifier()}, nlPriv)
 	if err != nil {
 		t.Fatal(err)
 	}
 
+	resign := func(nl key.NLPrivate, currentSig tkatype.MarshaledSignature) (key.NodePrivate, tkatype.MarshaledSignature) {
+		nk := key.NewNode()
+		sig, err := tka.ResignNKS(nl, nk.Public(), currentSig)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return nk, sig
+	}
+
+	n5Rotated, n5RotatedSig := resign(n5nl, n5InitialSig.Serialize())
+
 	nm := &netmap.NetworkMap{
 		Peers: nodeViews([]*tailcfg.Node{
 			{ID: 1, Key: n1.Public(), KeySignature: n1GoodSig.Serialize()},
-			{ID: 2, Key: n2.Public(), KeySignature: nil},                   // missing sig
-			{ID: 3, Key: n3.Public(), KeySignature: n1GoodSig.Serialize()}, // someone elses sig
-			{ID: 4, Key: n4.Public(), KeySignature: n4Sig.Serialize()},     // messed-up signature
-			{ID: 5, Key: n5.Public(), KeySignature: n5GoodSig.Serialize()},
+			{ID: 2, Key: n2.Public(), KeySignature: nil},                       // missing sig
+			{ID: 3, Key: n3.Public(), KeySignature: n1GoodSig.Serialize()},     // someone elses sig
+			{ID: 4, Key: n4.Public(), KeySignature: n4Sig.Serialize()},         // messed-up signature
+			{ID: 50, Key: n5.Public(), KeySignature: n5InitialSig.Serialize()}, // rotated
+			{ID: 51, Key: n5Rotated.Public(), KeySignature: n5RotatedSig},
 		}),
 	}
 
@@ -586,12 +604,39 @@ func TestTKAFilterNetmap(t *testing.T) {
 
 	want := nodeViews([]*tailcfg.Node{
 		{ID: 1, Key: n1.Public(), KeySignature: n1GoodSig.Serialize()},
-		{ID: 5, Key: n5.Public(), KeySignature: n5GoodSig.Serialize()},
+		{ID: 51, Key: n5Rotated.Public(), KeySignature: n5RotatedSig},
 	})
 	nodePubComparer := cmp.Comparer(func(x, y key.NodePublic) bool {
 		return x.Raw32() == y.Raw32()
 	})
-	if diff := cmp.Diff(nm.Peers, want, nodePubComparer); diff != "" {
+	if diff := cmp.Diff(want, nm.Peers, nodePubComparer); diff != "" {
+		t.Errorf("filtered netmap differs (-want, +got):\n%s", diff)
+	}
+
+	// Create two more node signatures using the same wrapping key as n5.
+	// Since they have the same rotation chain, both will be filtered out.
+	n7, n7Sig := resign(n5nl, n5RotatedSig)
+	n8, n8Sig := resign(n5nl, n5RotatedSig)
+
+	nm = &netmap.NetworkMap{
+		Peers: nodeViews([]*tailcfg.Node{
+			{ID: 1, Key: n1.Public(), KeySignature: n1GoodSig.Serialize()},
+			{ID: 2, Key: n2.Public(), KeySignature: nil},                       // missing sig
+			{ID: 3, Key: n3.Public(), KeySignature: n1GoodSig.Serialize()},     // someone elses sig
+			{ID: 4, Key: n4.Public(), KeySignature: n4Sig.Serialize()},         // messed-up signature
+			{ID: 50, Key: n5.Public(), KeySignature: n5InitialSig.Serialize()}, // rotated
+			{ID: 51, Key: n5Rotated.Public(), KeySignature: n5RotatedSig},      // rotated
+			{ID: 7, Key: n7.Public(), KeySignature: n7Sig},                     // same rotation chain as n8
+			{ID: 8, Key: n8.Public(), KeySignature: n8Sig},                     // same rotation chain as n7
+		}),
+	}
+
+	b.tkaFilterNetmapLocked(nm)
+
+	want = nodeViews([]*tailcfg.Node{
+		{ID: 1, Key: n1.Public(), KeySignature: n1GoodSig.Serialize()},
+	})
+	if diff := cmp.Diff(want, nm.Peers, nodePubComparer); diff != "" {
 		t.Errorf("filtered netmap differs (-want, +got):\n%s", diff)
 	}
 }
@@ -1130,3 +1175,85 @@ func TestTKARecoverCompromisedKeyFlow(t *testing.T) {
 		t.Errorf("NetworkLockSubmitRecoveryAUM() failed: %v", err)
 	}
 }
+
+func TestRotationTracker(t *testing.T) {
+	newNK := func(idx byte) key.NodePublic {
+		// single-byte public key to make it human-readable in tests.
+		raw32 := [32]byte{idx}
+		return key.NodePublicFromRaw32(go4mem.B(raw32[:]))
+	}
+	n1, n2, n3, n4, n5 := newNK(1), newNK(2), newNK(3), newNK(4), newNK(5)
+
+	pk1, pk2, pk3 := []byte{1}, []byte{2}, []byte{3}
+	type addDetails struct {
+		np      key.NodePublic
+		details *tka.RotationDetails
+	}
+	tests := []struct {
+		name       string
+		addDetails []addDetails
+		want       set.Set[key.NodePublic]
+	}{
+		{
+			name: "empty",
+			want: nil,
+		},
+		{
+			name: "single_prev_key",
+			addDetails: []addDetails{
+				{np: n1, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n2}, WrappingPubkey: pk1}},
+			},
+			want: set.SetOf([]key.NodePublic{n2}),
+		},
+		{
+			name: "several_prev_keys",
+			addDetails: []addDetails{
+				{np: n1, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n2}, WrappingPubkey: pk1}},
+				{np: n3, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n4}, WrappingPubkey: pk2}},
+				{np: n2, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n3, n4}, WrappingPubkey: pk1}},
+			},
+			want: set.SetOf([]key.NodePublic{n2, n3, n4}),
+		},
+		{
+			name: "several_per_pubkey_latest_wins",
+			addDetails: []addDetails{
+				{np: n2, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1}, WrappingPubkey: pk3}},
+				{np: n3, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1, n2}, WrappingPubkey: pk3}},
+				{np: n4, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1, n2, n3}, WrappingPubkey: pk3}},
+				{np: n5, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n4}, WrappingPubkey: pk3}},
+			},
+			want: set.SetOf([]key.NodePublic{n1, n2, n3, n4}),
+		},
+		{
+			name: "several_per_pubkey_same_chain_length_all_rejected",
+			addDetails: []addDetails{
+				{np: n2, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1}, WrappingPubkey: pk3}},
+				{np: n3, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1, n2}, WrappingPubkey: pk3}},
+				{np: n4, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1, n2}, WrappingPubkey: pk3}},
+				{np: n5, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1, n2}, WrappingPubkey: pk3}},
+			},
+			want: set.SetOf([]key.NodePublic{n1, n2, n3, n4, n5}),
+		},
+		{
+			name: "several_per_pubkey_longest_wins",
+			addDetails: []addDetails{
+				{np: n2, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1}, WrappingPubkey: pk3}},
+				{np: n3, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1, n2}, WrappingPubkey: pk3}},
+				{np: n4, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1, n2}, WrappingPubkey: pk3}},
+				{np: n5, details: &tka.RotationDetails{PrevNodeKeys: []key.NodePublic{n1, n2, n3}, WrappingPubkey: pk3}},
+			},
+			want: set.SetOf([]key.NodePublic{n1, n2, n3, n4}),
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			r := &rotationTracker{logf: t.Logf}
+			for _, ad := range tt.addDetails {
+				r.addRotationDetails(ad.np, ad.details)
+			}
+			if got := r.obsoleteKeys(); !reflect.DeepEqual(got, tt.want) {
+				t.Errorf("rotationTracker.obsoleteKeys() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}

ipn/ipnlocal/peerapi_test.go

@@ -642,8 +642,9 @@ func TestPeerAPIReplyToDNSQueries(t *testing.T) {
 	h.isSelf = false
 	h.remoteAddr = netip.MustParseAddrPort("100.150.151.152:12345")
 
-	eng, _ := wgengine.NewFakeUserspaceEngine(logger.Discard, 0)
-	pm := must.Get(newProfileManager(new(mem.Store), t.Logf, new(health.Tracker)))
+	ht := new(health.Tracker)
+	eng, _ := wgengine.NewFakeUserspaceEngine(logger.Discard, 0, ht)
+	pm := must.Get(newProfileManager(new(mem.Store), t.Logf, ht))
 	h.ps = &peerAPIServer{
 		b: &LocalBackend{
 			e:     eng,
@@ -692,8 +693,9 @@ func TestPeerAPIPrettyReplyCNAME(t *testing.T) {
 		var h peerAPIHandler
 		h.remoteAddr = netip.MustParseAddrPort("100.150.151.152:12345")
 
-		eng, _ := wgengine.NewFakeUserspaceEngine(logger.Discard, 0)
-		pm := must.Get(newProfileManager(new(mem.Store), t.Logf, new(health.Tracker)))
+		ht := new(health.Tracker)
+		eng, _ := wgengine.NewFakeUserspaceEngine(logger.Discard, 0, ht)
+		pm := must.Get(newProfileManager(new(mem.Store), t.Logf, ht))
 		var a *appc.AppConnector
 		if shouldStore {
 			a = appc.NewAppConnector(t.Logf, &appctest.RouteCollector{}, &appc.RouteInfo{}, fakeStoreRoutes)
@@ -764,8 +766,9 @@ func TestPeerAPIReplyToDNSQueriesAreObserved(t *testing.T) {
 		h.remoteAddr = netip.MustParseAddrPort("100.150.151.152:12345")
 
 		rc := &appctest.RouteCollector{}
-		eng, _ := wgengine.NewFakeUserspaceEngine(logger.Discard, 0)
-		pm := must.Get(newProfileManager(new(mem.Store), t.Logf, new(health.Tracker)))
+		ht := new(health.Tracker)
+		eng, _ := wgengine.NewFakeUserspaceEngine(logger.Discard, 0, ht)
+		pm := must.Get(newProfileManager(new(mem.Store), t.Logf, ht))
 		var a *appc.AppConnector
 		if shouldStore {
 			a = appc.NewAppConnector(t.Logf, rc, &appc.RouteInfo{}, fakeStoreRoutes)
@@ -826,9 +829,10 @@ func TestPeerAPIReplyToDNSQueriesAreObservedWithCNAMEFlattening(t *testing.T) {
 		var h peerAPIHandler
 		h.remoteAddr = netip.MustParseAddrPort("100.150.151.152:12345")
 
+		ht := new(health.Tracker)
 		rc := &appctest.RouteCollector{}
-		eng, _ := wgengine.NewFakeUserspaceEngine(logger.Discard, 0)
-		pm := must.Get(newProfileManager(new(mem.Store), t.Logf, new(health.Tracker)))
+		eng, _ := wgengine.NewFakeUserspaceEngine(logger.Discard, 0, ht)
+		pm := must.Get(newProfileManager(new(mem.Store), t.Logf, ht))
 		var a *appc.AppConnector
 		if shouldStore {
 			a = appc.NewAppConnector(t.Logf, rc, &appc.RouteInfo{}, fakeStoreRoutes)

ipn/ipnlocal/profiles.go

@@ -353,9 +353,9 @@ func (pm *profileManager) loadSavedPrefs(key ipn.StateKey) (ipn.PrefsView, error
 	if err != nil {
 		return ipn.PrefsView{}, err
 	}
-	savedPrefs, err := ipn.PrefsFromBytes(bs)
-	if err != nil {
-		return ipn.PrefsView{}, fmt.Errorf("PrefsFromBytes: %v", err)
+	savedPrefs := ipn.NewPrefs()
+	if err := ipn.PrefsFromBytes(bs, savedPrefs); err != nil {
+		return ipn.PrefsView{}, fmt.Errorf("parsing saved prefs: %v", err)
 	}
 	pm.logf("using backend prefs for %q: %v", key, savedPrefs.Pretty())
 
@@ -371,12 +371,13 @@ func (pm *profileManager) loadSavedPrefs(key ipn.StateKey) (ipn.PrefsView, error
 	// https://github.com/tailscale/tailscale/pull/11814/commits/1613b18f8280c2bce786980532d012c9f0454fa2#diff-314ba0d799f70c8998940903efb541e511f352b39a9eeeae8d475c921d66c2ac
 	// prefs could set AutoUpdate.Apply=true via EditPrefs or tailnet
 	// auto-update defaults. After that change, such value is "invalid" and
-	// cause any EditPrefs calls to fail (other than disabling autp-updates).
+	// cause any EditPrefs calls to fail (other than disabling auto-updates).
 	//
 	// Reset AutoUpdate.Apply if we detect such invalid prefs.
 	if savedPrefs.AutoUpdate.Apply.EqualBool(true) && !clientupdate.CanAutoUpdate() {
 		savedPrefs.AutoUpdate.Apply.Clear()
 	}
+
 	return savedPrefs.View(), nil
 }
 

ipn/ipnlocal/profiles_test.go

@@ -600,3 +600,16 @@ func TestProfileManagementWindows(t *testing.T) {
 		t.Fatalf("CurrentUserID = %q; want %q", pm.CurrentUserID(), uid)
 	}
 }
+
+// TestDefaultPrefs tests that defaultPrefs is just NewPrefs with
+// LoggedOut=true (the Prefs we use before connecting to control). We shouldn't
+// be putting any defaulting there, and instead put all defaults in NewPrefs.
+func TestDefaultPrefs(t *testing.T) {
+	p1 := ipn.NewPrefs()
+	p1.LoggedOut = true
+	p1.WantRunning = false
+	p2 := defaultPrefs
+	if !p1.View().Equals(p2) {
+		t.Errorf("defaultPrefs is %s, want %s; defaultPrefs should only modify WantRunning and LoggedOut, all other defaults should be in ipn.NewPrefs.", p2.Pretty(), p1.Pretty())
+	}
+}

ipn/ipnlocal/profiles_windows.go

@@ -57,7 +57,7 @@ func (pm *profileManager) loadLegacyPrefs() (string, ipn.PrefsView, error) {
 	}
 
 	prefsPath := filepath.Join(userLegacyPrefsDir, legacyPrefsFile+legacyPrefsExt)
-	prefs, err := ipn.LoadPrefs(prefsPath)
+	prefs, err := ipn.LoadPrefsWindows(prefsPath)
 	pm.dlogf("ipn.LoadPrefs(%q) = %v, %v", prefsPath, prefs, err)
 	if errors.Is(err, fs.ErrNotExist) {
 		return "", ipn.PrefsView{}, errAlreadyMigrated

ipn/ipnlocal/serve_test.go

@@ -672,7 +672,10 @@ func newTestBackend(t *testing.T) *LocalBackend {
 	}
 
 	sys := &tsd.System{}
-	e, err := wgengine.NewUserspaceEngine(logf, wgengine.Config{SetSubsystem: sys.Set})
+	e, err := wgengine.NewUserspaceEngine(logf, wgengine.Config{
+		SetSubsystem:  sys.Set,
+		HealthTracker: sys.HealthTracker(),
+	})
 	if err != nil {
 		t.Fatal(err)
 	}

ipn/ipnlocal/state_test.go

@@ -97,7 +97,6 @@ type mockControl struct {
 	paused atomic.Bool
 
 	mu          sync.Mutex
-	machineKey  key.MachinePrivate
 	persist     *persist.Persist
 	calls       []string
 	authBlocked bool
@@ -134,12 +133,6 @@ func (cc *mockControl) populateKeys() (newKeys bool) {
 	cc.mu.Lock()
 	defer cc.mu.Unlock()
 
-	if cc.machineKey.IsZero() {
-		cc.logf("Copying machineKey.")
-		cc.machineKey, _ = cc.opts.GetMachinePrivateKey()
-		newKeys = true
-	}
-
 	if cc.persist == nil {
 		cc.persist = &persist.Persist{}
 	}
@@ -205,8 +198,8 @@ func (cc *mockControl) Shutdown() {
 // Login starts a login process. Note that in this mock, we don't automatically
 // generate notifications about the progress of the login operation. You have to
 // call send() as required by the test.
-func (cc *mockControl) Login(t *tailcfg.Oauth2Token, flags controlclient.LoginFlags) {
-	cc.logf("Login token=%v flags=%v", t, flags)
+func (cc *mockControl) Login(flags controlclient.LoginFlags) {
+	cc.logf("Login flags=%v", flags)
 	cc.called("Login")
 	newKeys := cc.populateKeys()
 
@@ -272,7 +265,7 @@ func (b *LocalBackend) nonInteractiveLoginForStateTest() {
 	cc := b.cc
 	b.mu.Unlock()
 
-	cc.Login(nil, b.loginFlags|controlclient.LoginInteractive)
+	cc.Login(b.loginFlags | controlclient.LoginInteractive)
 }
 
 // A very precise test of the sequence of function calls generated by
@@ -305,7 +298,7 @@ func TestStateMachine(t *testing.T) {
 	sys := new(tsd.System)
 	store := new(testStateStorage)
 	sys.Set(store)
-	e, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set)
+	e, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set, sys.HealthTracker())
 	if err != nil {
 		t.Fatalf("NewFakeUserspaceEngine: %v", err)
 	}
@@ -316,6 +309,7 @@ func TestStateMachine(t *testing.T) {
 	if err != nil {
 		t.Fatalf("NewLocalBackend: %v", err)
 	}
+	b.DisablePortMapperForTest()
 
 	var cc, previousCC *mockControl
 	b.SetControlClientGetterForTesting(func(opts controlclient.Options) (controlclient.Client, error) {
@@ -335,7 +329,7 @@ func TestStateMachine(t *testing.T) {
 			(n.Prefs != nil && n.Prefs.Valid()) ||
 			n.BrowseToURL != nil ||
 			n.LoginFinished != nil {
-			logf("%v\n\n", n)
+			logf("%+v\n\n", n)
 			notifies.put(n)
 		} else {
 			logf("(ignored) %v\n\n", n)
@@ -412,7 +406,7 @@ func TestStateMachine(t *testing.T) {
 	// the user needs to visit a login URL.
 	t.Logf("\n\nLogin (url response)")
 
-	notifies.expect(2)
+	notifies.expect(3)
 	b.EditPrefs(&ipn.MaskedPrefs{
 		ControlURLSet: true,
 		Prefs: ipn.Prefs{
@@ -427,12 +421,15 @@ func TestStateMachine(t *testing.T) {
 		// ...but backend eats that notification, because the user
 		// didn't explicitly request interactive login yet, and
 		// we're already in NeedsLogin state.
-		nn := notifies.drain(2)
+		nn := notifies.drain(3)
 
 		c.Assert(nn[1].Prefs, qt.IsNotNil)
 		c.Assert(nn[1].Prefs.LoggedOut(), qt.IsTrue)
 		c.Assert(nn[1].Prefs.WantRunning(), qt.IsFalse)
 		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+		c.Assert(nn[2].BrowseToURL, qt.IsNotNil)
+		c.Assert(url1, qt.Equals, *nn[2].BrowseToURL)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
 	}
 
 	// Now we'll try an interactive login.
@@ -440,13 +437,10 @@ func TestStateMachine(t *testing.T) {
 	// ask control to do anything. Instead backend will emit an event
 	// indicating that the UI should browse to the given URL.
 	t.Logf("\n\nLogin (interactive)")
-	notifies.expect(1)
+	notifies.expect(0)
 	b.StartLoginInteractive(context.Background())
 	{
-		nn := notifies.drain(1)
 		cc.assertCalls()
-		c.Assert(nn[0].BrowseToURL, qt.IsNotNil)
-		c.Assert(url1, qt.Equals, *nn[0].BrowseToURL)
 		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
 	}
 
@@ -459,9 +453,8 @@ func TestStateMachine(t *testing.T) {
 	notifies.expect(0)
 	b.StartLoginInteractive(context.Background())
 	{
-		notifies.drain(0)
 		// backend asks control for another login sequence
-		cc.assertCalls("Login")
+		cc.assertCalls()
 		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
 	}
 
@@ -830,7 +823,7 @@ func TestStateMachine(t *testing.T) {
 	// The last test case is the most common one: restarting when both
 	// logged in and WantRunning.
 	t.Logf("\n\nStart5")
-	notifies.expect(2)
+	notifies.expect(1)
 	c.Assert(b.Start(ipn.Options{}), qt.IsNil)
 	{
 		// NOTE: cc.Shutdown() is correct here, since we didn't call
@@ -838,27 +831,27 @@ func TestStateMachine(t *testing.T) {
 		previousCC.assertShutdown(false)
 		cc.assertCalls("New", "Login")
 
-		nn := notifies.drain(2)
+		nn := notifies.drain(1)
 		cc.assertCalls()
 		c.Assert(nn[0].Prefs, qt.IsNotNil)
 		c.Assert(nn[0].Prefs.LoggedOut(), qt.IsFalse)
 		c.Assert(nn[0].Prefs.WantRunning(), qt.IsTrue)
-		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+		c.Assert(b.State(), qt.Equals, ipn.NoState)
 	}
 
 	// Control server accepts our valid key from before.
 	t.Logf("\n\nLoginFinished5")
-	notifies.expect(2)
+	notifies.expect(1)
 	cc.send(nil, "", true, &netmap.NetworkMap{
 		SelfNode: (&tailcfg.Node{MachineAuthorized: true}).View(),
 	})
 	{
-		nn := notifies.drain(2)
+		nn := notifies.drain(1)
 		cc.assertCalls()
 		// NOTE: No LoginFinished message since no interactive
 		// login was needed.
-		c.Assert(nn[1].State, qt.IsNotNil)
-		c.Assert(ipn.Starting, qt.Equals, *nn[1].State)
+		c.Assert(nn[0].State, qt.IsNotNil)
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
 		// NOTE: No prefs change this time. WantRunning stays true.
 		// We were in Starting in the first place, so that doesn't
 		// change either.
@@ -909,7 +902,7 @@ func TestEditPrefsHasNoKeys(t *testing.T) {
 	logf := tstest.WhileTestRunningLogger(t)
 	sys := new(tsd.System)
 	sys.Set(new(mem.Store))
-	e, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set)
+	e, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set, sys.HealthTracker())
 	if err != nil {
 		t.Fatalf("NewFakeUserspaceEngine: %v", err)
 	}

ipn/ipnserver/server.go

@@ -46,9 +46,6 @@ type Server struct {
 	// is true, the ForceDaemon pref can override this.
 	resetOnZero bool
 
-	startBackendOnce sync.Once
-	runCalled        atomic.Bool
-
 	// mu guards the fields that follow.
 	// lock order: mu, then LocalBackend.mu
 	mu            sync.Mutex
@@ -471,16 +468,15 @@ func New(logf logger.Logf, logID logid.PublicID, netMon *netmon.Monitor) *Server
 
 // SetLocalBackend sets the server's LocalBackend.
 //
-// If b.Run has already been called, then lb.Start will be called.
-// Otherwise Start will be called once Run is called.
+// It should only call be called after calling lb.Start.
 func (s *Server) SetLocalBackend(lb *ipnlocal.LocalBackend) {
 	if lb == nil {
 		panic("nil LocalBackend")
 	}
+
 	if !s.lb.CompareAndSwap(nil, lb) {
 		panic("already set")
 	}
-	s.startBackendIfNeeded()
 
 	s.mu.Lock()
 	s.backendWaiter.wakeAll()
@@ -490,21 +486,6 @@ func (s *Server) SetLocalBackend(lb *ipnlocal.LocalBackend) {
 	// https://github.com/tailscale/tailscale/issues/6522
 }
 
-func (b *Server) startBackendIfNeeded() {
-	if !b.runCalled.Load() {
-		return
-	}
-	lb := b.lb.Load()
-	if lb == nil {
-		return
-	}
-	if lb.Prefs().Valid() {
-		b.startBackendOnce.Do(func() {
-			lb.Start(ipn.Options{})
-		})
-	}
-}
-
 // connIdentityContextKey is the http.Request.Context's context.Value key for either an
 // *ipnauth.ConnIdentity or an error.
 type connIdentityContextKey struct{}
@@ -517,7 +498,6 @@ type connIdentityContextKey struct{}
 // If the Server's LocalBackend has already been set, Run starts it.
 // Otherwise, the next call to SetLocalBackend will start it.
 func (s *Server) Run(ctx context.Context, ln net.Listener) error {
-	s.runCalled.Store(true)
 	defer func() {
 		if lb := s.lb.Load(); lb != nil {
 			lb.Shutdown()
@@ -537,7 +517,6 @@ func (s *Server) Run(ctx context.Context, ln net.Listener) error {
 		ln.Close()
 	}()
 
-	s.startBackendIfNeeded()
 	systemd.Ready()
 
 	hs := &http.Server{

ipn/ipnstate/ipnstate.go

@@ -18,6 +18,7 @@ import (
 	"time"
 
 	"tailscale.com/tailcfg"
+	"tailscale.com/tka"
 	"tailscale.com/types/key"
 	"tailscale.com/types/ptr"
 	"tailscale.com/types/views"
@@ -41,6 +42,9 @@ type Status struct {
 	//  "Starting", "Running".
 	BackendState string
 
+	// HaveNodeKey is whether the current profile has a node key configured.
+	HaveNodeKey bool `json:",omitempty"`
+
 	AuthURL      string       // current URL provided by control to authorize client
 	TailscaleIPs []netip.Addr // Tailscale IP(s) assigned to this node
 	Self         *PeerStatus
@@ -123,6 +127,9 @@ type NetworkLockStatus struct {
 	// NodeKeySigned is true if our node is authorized by network-lock.
 	NodeKeySigned bool
 
+	// NodeKeySignature is the current signature of this node's key.
+	NodeKeySignature *tka.NodeKeySignature
+
 	// TrustedKeys describes the keys currently trusted to make changes
 	// to network-lock.
 	TrustedKeys []TKAKey

ipn/localapi/localapi.go

@@ -6,6 +6,7 @@ package localapi
 
 import (
 	"bytes"
+	"cmp"
 	"context"
 	"crypto/sha256"
 	"encoding/hex"
@@ -1939,8 +1940,10 @@ func (h *Handler) serveDial(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 
+	network := cmp.Or(r.Header.Get("Dial-Network"), "tcp")
+
 	addr := net.JoinHostPort(hostStr, portStr)
-	outConn, err := h.b.Dialer().UserDial(r.Context(), "tcp", addr)
+	outConn, err := h.b.Dialer().UserDial(r.Context(), network, addr)
 	if err != nil {
 		http.Error(w, "dial failure: "+err.Error(), http.StatusBadGateway)
 		return

ipn/localapi/localapi_test.go

@@ -314,7 +314,7 @@ func newTestLocalBackend(t testing.TB) *ipnlocal.LocalBackend {
 	sys := new(tsd.System)
 	store := new(mem.Store)
 	sys.Set(store)
-	eng, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set)
+	eng, err := wgengine.NewFakeUserspaceEngine(logf, sys.Set, sys.HealthTracker())
 	if err != nil {
 		t.Fatalf("NewFakeUserspaceEngine: %v", err)
 	}

ipn/prefs.go

@@ -75,18 +75,6 @@ type Prefs struct {
 	// controlled by ExitNodeID/IP below.
 	RouteAll bool
 
-	// AllowSingleHosts specifies whether to install routes for each
-	// node IP on the tailscale network, in addition to a route for
-	// the whole network.
-	// This corresponds to the "tailscale up --host-routes" value,
-	// which defaults to true.
-	//
-	// TODO(danderson): why do we have this? It dumps a lot of stuff
-	// into the routing table, and a single network route _should_ be
-	// all that we need. But when I turn this off in my tailscaled,
-	// packets stop flowing. What's up with that?
-	AllowSingleHosts bool
-
 	// ExitNodeID and ExitNodeIP specify the node that should be used
 	// as an exit node for internet traffic. At most one of these
 	// should be non-zero.
@@ -203,6 +191,20 @@ type Prefs struct {
 	// Linux-only.
 	NoSNAT bool
 
+	// NoStatefulFiltering specifies whether to apply stateful filtering when
+	// advertising routes in AdvertiseRoutes. The default is to not apply
+	// stateful filtering.
+	//
+	// To allow inbound connections from advertised routes, both NoSNAT and
+	// NoStatefulFiltering must be true.
+	//
+	// This is an opt.Bool because it was first added after NoSNAT, with a
+	// backfill based on the value of that parameter. The backfill has been
+	// removed since then, but the field remains an opt.Bool.
+	//
+	// Linux-only.
+	NoStatefulFiltering opt.Bool `json:",omitempty"`
+
 	// NetfilterMode specifies how much to manage netfilter rules for
 	// Tailscale, if at all.
 	NetfilterMode preftype.NetfilterMode
@@ -237,6 +239,16 @@ type Prefs struct {
 	// by name.
 	DriveShares []*drive.Share
 
+	// AllowSingleHosts was a legacy field that was always true
+	// for the past 4.5 years. It controlled whether Tailscale
+	// peers got /32 or /127 routes for each other.
+	// As of 2024-05-17 we're starting to ignore it, but to let
+	// people still downgrade Tailscale versions and not break
+	// all peer-to-peer networking we still write it to disk (as JSON)
+	// so it can be loaded back by old versions.
+	// TODO(bradfitz): delete this in 2025 sometime. See #12058.
+	AllowSingleHosts marshalAsTrueInJSON
+
 	// The Persist field is named 'Config' in the file for backward
 	// compatibility with earlier versions.
 	// TODO(apenwarr): We should move this out of here, it's not a pref.
@@ -267,6 +279,13 @@ func (au1 AutoUpdatePrefs) Equals(au2 AutoUpdatePrefs) bool {
 		ok1 == ok2
 }
 
+type marshalAsTrueInJSON struct{}
+
+var trueJSON = []byte("true")
+
+func (marshalAsTrueInJSON) MarshalJSON() ([]byte, error) { return trueJSON, nil }
+func (*marshalAsTrueInJSON) UnmarshalJSON([]byte) error  { return nil }
+
 // AppConnectorPrefs are the app connector settings for the node agent.
 type AppConnectorPrefs struct {
 	// Advertise specifies whether the app connector subsystem is advertising
@@ -284,7 +303,6 @@ type MaskedPrefs struct {
 
 	ControlURLSet             bool                `json:",omitempty"`
 	RouteAllSet               bool                `json:",omitempty"`
-	AllowSingleHostsSet       bool                `json:",omitempty"`
 	ExitNodeIDSet             bool                `json:",omitempty"`
 	ExitNodeIPSet             bool                `json:",omitempty"`
 	InternalExitNodePriorSet  bool                `json:",omitempty"` // Internal; can't be set by LocalAPI clients
@@ -302,6 +320,7 @@ type MaskedPrefs struct {
 	EggSet                    bool                `json:",omitempty"`
 	AdvertiseRoutesSet        bool                `json:",omitempty"`
 	NoSNATSet                 bool                `json:",omitempty"`
+	NoStatefulFilteringSet    bool                `json:",omitempty"`
 	NetfilterModeSet          bool                `json:",omitempty"`
 	OperatorUserSet           bool                `json:",omitempty"`
 	ProfileNameSet            bool                `json:",omitempty"`
@@ -468,9 +487,6 @@ func (p *Prefs) pretty(goos string) string {
 	var sb strings.Builder
 	sb.WriteString("Prefs{")
 	fmt.Fprintf(&sb, "ra=%v ", p.RouteAll)
-	if !p.AllowSingleHosts {
-		sb.WriteString("mesh=false ")
-	}
 	fmt.Fprintf(&sb, "dns=%v want=%v ", p.CorpDNS, p.WantRunning)
 	if p.RunSSH {
 		sb.WriteString("ssh=true ")
@@ -501,6 +517,13 @@ func (p *Prefs) pretty(goos string) string {
 	if len(p.AdvertiseRoutes) > 0 || p.NoSNAT {
 		fmt.Fprintf(&sb, "snat=%v ", !p.NoSNAT)
 	}
+	if len(p.AdvertiseRoutes) > 0 || p.NoStatefulFiltering.EqualBool(true) {
+		// Only print if we're advertising any routes, or the user has
+		// turned off stateful filtering (NoStatefulFiltering=true ⇒
+		// StatefulFiltering=false).
+		bb, _ := p.NoStatefulFiltering.Get()
+		fmt.Fprintf(&sb, "statefulFiltering=%v ", !bb)
+	}
 	if len(p.AdvertiseTags) > 0 {
 		fmt.Fprintf(&sb, "tags=%s ", strings.Join(p.AdvertiseTags, ","))
 	}
@@ -556,7 +579,6 @@ func (p *Prefs) Equals(p2 *Prefs) bool {
 
 	return p.ControlURL == p2.ControlURL &&
 		p.RouteAll == p2.RouteAll &&
-		p.AllowSingleHosts == p2.AllowSingleHosts &&
 		p.ExitNodeID == p2.ExitNodeID &&
 		p.ExitNodeIP == p2.ExitNodeIP &&
 		p.InternalExitNodePrior == p2.InternalExitNodePrior &&
@@ -569,6 +591,7 @@ func (p *Prefs) Equals(p2 *Prefs) bool {
 		p.NotepadURLs == p2.NotepadURLs &&
 		p.ShieldsUp == p2.ShieldsUp &&
 		p.NoSNAT == p2.NoSNAT &&
+		p.NoStatefulFiltering == p2.NoStatefulFiltering &&
 		p.NetfilterMode == p2.NetfilterMode &&
 		p.OperatorUser == p2.OperatorUser &&
 		p.Hostname == p2.Hostname &&
@@ -638,11 +661,11 @@ func NewPrefs() *Prefs {
 		// later anyway.
 		ControlURL: "",
 
-		RouteAll:         true,
-		AllowSingleHosts: true,
-		CorpDNS:          true,
-		WantRunning:      false,
-		NetfilterMode:    preftype.NetfilterOn,
+		RouteAll:            true,
+		CorpDNS:             true,
+		WantRunning:         false,
+		NetfilterMode:       preftype.NetfilterOn,
+		NoStatefulFiltering: opt.NewBool(true),
 		AutoUpdate: AutoUpdatePrefs{
 			Check: true,
 			Apply: opt.Bool("unset"),
@@ -851,24 +874,21 @@ func (p *Prefs) ShouldWebClientBeRunning() bool {
 	return p.WantRunning && p.RunWebClient
 }
 
-// PrefsFromBytes deserializes Prefs from a JSON blob.
-func PrefsFromBytes(b []byte) (*Prefs, error) {
-	p := NewPrefs()
+// PrefsFromBytes deserializes Prefs from a JSON blob b into base. Values in
+// base are preserved, unless they are populated in the JSON blob.
+func PrefsFromBytes(b []byte, base *Prefs) error {
 	if len(b) == 0 {
-		return p, nil
+		return nil
 	}
 
-	if err := json.Unmarshal(b, p); err != nil {
-		return nil, err
-	}
-	return p, nil
+	return json.Unmarshal(b, base)
 }
 
 var jsonEscapedZero = []byte(`\u0000`)
 
-// LoadPrefs loads a legacy relaynode config file into Prefs
-// with sensible migration defaults set.
-func LoadPrefs(filename string) (*Prefs, error) {
+// LoadPrefsWindows loads a legacy relaynode config file into Prefs with
+// sensible migration defaults set. Windows-only.
+func LoadPrefsWindows(filename string) (*Prefs, error) {
 	data, err := os.ReadFile(filename)
 	if err != nil {
 		return nil, fmt.Errorf("LoadPrefs open: %w", err) // err includes path
@@ -881,8 +901,8 @@ func LoadPrefs(filename string) (*Prefs, error) {
 		// to log in again. (better than crashing)
 		return nil, os.ErrNotExist
 	}
-	p, err := PrefsFromBytes(data)
-	if err != nil {
+	p := NewPrefs()
+	if err := PrefsFromBytes(data, p); err != nil {
 		return nil, fmt.Errorf("LoadPrefs(%q) decode: %w", filename, err)
 	}
 	return p, nil

ipn/prefs_test.go

@@ -38,7 +38,6 @@ func TestPrefsEqual(t *testing.T) {
 	prefsHandles := []string{
 		"ControlURL",
 		"RouteAll",
-		"AllowSingleHosts",
 		"ExitNodeID",
 		"ExitNodeIP",
 		"InternalExitNodePrior",
@@ -56,6 +55,7 @@ func TestPrefsEqual(t *testing.T) {
 		"Egg",
 		"AdvertiseRoutes",
 		"NoSNAT",
+		"NoStatefulFiltering",
 		"NetfilterMode",
 		"OperatorUser",
 		"ProfileName",
@@ -64,6 +64,7 @@ func TestPrefsEqual(t *testing.T) {
 		"PostureChecking",
 		"NetfilterKind",
 		"DriveShares",
+		"AllowSingleHosts",
 		"Persist",
 	}
 	if have := fieldsOf(reflect.TypeFor[Prefs]()); !reflect.DeepEqual(have, prefsHandles) {
@@ -122,18 +123,6 @@ func TestPrefsEqual(t *testing.T) {
 			&Prefs{RouteAll: true},
 			true,
 		},
-
-		{
-			&Prefs{AllowSingleHosts: true},
-			&Prefs{AllowSingleHosts: false},
-			false,
-		},
-		{
-			&Prefs{AllowSingleHosts: true},
-			&Prefs{AllowSingleHosts: true},
-			true,
-		},
-
 		{
 			&Prefs{ExitNodeID: "n1234"},
 			&Prefs{},
@@ -372,9 +361,10 @@ func checkPrefs(t *testing.T, p Prefs) {
 	if p.Equals(p2) {
 		t.Fatalf("p == p2\n")
 	}
-	p2b, err = PrefsFromBytes(p2.ToBytes())
+	p2b = new(Prefs)
+	err = PrefsFromBytes(p2.ToBytes(), p2b)
 	if err != nil {
-		t.Fatalf("PrefsFromBytes(p2) failed\n")
+		t.Fatalf("PrefsFromBytes(p2) failed: bytes=%q; err=%v\n", p2.ToBytes(), err)
 	}
 	p2p := p2.Pretty()
 	p2bp := p2b.Pretty()
@@ -425,46 +415,43 @@ func TestPrefsPretty(t *testing.T) {
 		{
 			Prefs{},
 			"linux",
-			"Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=off Persist=nil}",
+			"Prefs{ra=false dns=false want=false routes=[] nf=off update=off Persist=nil}",
 		},
 		{
 			Prefs{},
 			"windows",
-			"Prefs{ra=false mesh=false dns=false want=false update=off Persist=nil}",
+			"Prefs{ra=false dns=false want=false update=off Persist=nil}",
 		},
 		{
 			Prefs{ShieldsUp: true},
 			"windows",
-			"Prefs{ra=false mesh=false dns=false want=false shields=true update=off Persist=nil}",
+			"Prefs{ra=false dns=false want=false shields=true update=off Persist=nil}",
 		},
 		{
-			Prefs{AllowSingleHosts: true},
+			Prefs{},
 			"windows",
 			"Prefs{ra=false dns=false want=false update=off Persist=nil}",
 		},
 		{
 			Prefs{
-				NotepadURLs:      true,
-				AllowSingleHosts: true,
+				NotepadURLs: true,
 			},
 			"windows",
 			"Prefs{ra=false dns=false want=false notepad=true update=off Persist=nil}",
 		},
 		{
 			Prefs{
-				AllowSingleHosts: true,
-				WantRunning:      true,
-				ForceDaemon:      true, // server mode
+				WantRunning: true,
+				ForceDaemon: true, // server mode
 			},
 			"windows",
 			"Prefs{ra=false dns=false want=true server=true update=off Persist=nil}",
 		},
 		{
 			Prefs{
-				AllowSingleHosts: true,
-				WantRunning:      true,
-				ControlURL:       "http://localhost:1234",
-				AdvertiseTags:    []string{"tag:foo", "tag:bar"},
+				WantRunning:   true,
+				ControlURL:    "http://localhost:1234",
+				AdvertiseTags: []string{"tag:foo", "tag:bar"},
 			},
 			"darwin",
 			`Prefs{ra=false dns=false want=true tags=tag:foo,tag:bar url="http://localhost:1234" update=off Persist=nil}`,
@@ -474,7 +461,7 @@ func TestPrefsPretty(t *testing.T) {
 				Persist: &persist.Persist{},
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=off Persist{lm=, o=, n= u=""}}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off update=off Persist{lm=, o=, n= u=""}}`,
 		},
 		{
 			Prefs{
@@ -483,21 +470,21 @@ func TestPrefsPretty(t *testing.T) {
 				},
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=off Persist{lm=, o=, n=[B1VKl] u=""}}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off update=off Persist{lm=, o=, n=[B1VKl] u=""}}`,
 		},
 		{
 			Prefs{
 				ExitNodeIP: netip.MustParseAddr("1.2.3.4"),
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false exit=1.2.3.4 lan=false routes=[] nf=off update=off Persist=nil}`,
+			`Prefs{ra=false dns=false want=false exit=1.2.3.4 lan=false routes=[] nf=off update=off Persist=nil}`,
 		},
 		{
 			Prefs{
 				ExitNodeID: tailcfg.StableNodeID("myNodeABC"),
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false exit=myNodeABC lan=false routes=[] nf=off update=off Persist=nil}`,
+			`Prefs{ra=false dns=false want=false exit=myNodeABC lan=false routes=[] nf=off update=off Persist=nil}`,
 		},
 		{
 			Prefs{
@@ -505,21 +492,21 @@ func TestPrefsPretty(t *testing.T) {
 				ExitNodeAllowLANAccess: true,
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false exit=myNodeABC lan=true routes=[] nf=off update=off Persist=nil}`,
+			`Prefs{ra=false dns=false want=false exit=myNodeABC lan=true routes=[] nf=off update=off Persist=nil}`,
 		},
 		{
 			Prefs{
 				ExitNodeAllowLANAccess: true,
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=off Persist=nil}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off update=off Persist=nil}`,
 		},
 		{
 			Prefs{
 				Hostname: "foo",
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off host="foo" update=off Persist=nil}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off host="foo" update=off Persist=nil}`,
 		},
 		{
 			Prefs{
@@ -529,7 +516,7 @@ func TestPrefsPretty(t *testing.T) {
 				},
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=check Persist=nil}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off update=check Persist=nil}`,
 		},
 		{
 			Prefs{
@@ -539,7 +526,7 @@ func TestPrefsPretty(t *testing.T) {
 				},
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=on Persist=nil}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off update=on Persist=nil}`,
 		},
 		{
 			Prefs{
@@ -548,7 +535,7 @@ func TestPrefsPretty(t *testing.T) {
 				},
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=off appconnector=advertise Persist=nil}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off update=off appconnector=advertise Persist=nil}`,
 		},
 		{
 			Prefs{
@@ -557,21 +544,21 @@ func TestPrefsPretty(t *testing.T) {
 				},
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=off Persist=nil}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off update=off Persist=nil}`,
 		},
 		{
 			Prefs{
 				NetfilterKind: "iptables",
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off netfilterKind=iptables update=off Persist=nil}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off netfilterKind=iptables update=off Persist=nil}`,
 		},
 		{
 			Prefs{
 				NetfilterKind: "",
 			},
 			"linux",
-			`Prefs{ra=false mesh=false dns=false want=false routes=[] nf=off update=off Persist=nil}`,
+			`Prefs{ra=false dns=false want=false routes=[] nf=off update=off Persist=nil}`,
 		},
 	}
 	for i, tt := range tests {
@@ -585,7 +572,7 @@ func TestPrefsPretty(t *testing.T) {
 func TestLoadPrefsNotExist(t *testing.T) {
 	bogusFile := fmt.Sprintf("/tmp/not-exist-%d", time.Now().UnixNano())
 
-	p, err := LoadPrefs(bogusFile)
+	p, err := LoadPrefsWindows(bogusFile)
 	if errors.Is(err, os.ErrNotExist) {
 		// expected.
 		return
@@ -607,7 +594,7 @@ func TestLoadPrefsFileWithZeroInIt(t *testing.T) {
 	f.Close()
 	defer os.Remove(path)
 
-	p, err := LoadPrefs(path)
+	p, err := LoadPrefsWindows(path)
 	if errors.Is(err, os.ErrNotExist) {
 		// expected.
 		return
@@ -631,8 +618,9 @@ func TestMaskedPrefsSetsInternal(t *testing.T) {
 func TestMaskedPrefsFields(t *testing.T) {
 	have := map[string]bool{}
 	for _, f := range fieldsOf(reflect.TypeFor[Prefs]()) {
-		if f == "Persist" {
-			// This one can't be edited.
+		switch f {
+		case "Persist", "AllowSingleHosts":
+			// These can't be edited.
 			continue
 		}
 		have[f] = true
@@ -751,13 +739,12 @@ func TestMaskedPrefsPretty(t *testing.T) {
 		{
 			m: &MaskedPrefs{
 				Prefs: Prefs{
-					Hostname:         "bar",
-					OperatorUser:     "galaxybrain",
-					AllowSingleHosts: true,
-					RouteAll:         false,
-					ExitNodeID:       "foo",
-					AdvertiseTags:    []string{"tag:foo", "tag:bar"},
-					NetfilterMode:    preftype.NetfilterNoDivert,
+					Hostname:      "bar",
+					OperatorUser:  "galaxybrain",
+					RouteAll:      false,
+					ExitNodeID:    "foo",
+					AdvertiseTags: []string{"tag:foo", "tag:bar"},
+					NetfilterMode: preftype.NetfilterNoDivert,
 				},
 				RouteAllSet:      true,
 				HostnameSet:      true,
@@ -1062,3 +1049,24 @@ func TestNotifyPrefsJSONRoundtrip(t *testing.T) {
 		t.Fatal("Prefs should not be valid after deserialization")
 	}
 }
+
+// Verify that our Prefs type writes out an AllowSingleHosts field so we can
+// downgrade to older versions that require it.
+func TestPrefsDowngrade(t *testing.T) {
+	var p Prefs
+	j, err := json.Marshal(p)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	type oldPrefs struct {
+		AllowSingleHosts bool
+	}
+	var op oldPrefs
+	if err := json.Unmarshal(j, &op); err != nil {
+		t.Fatal(err)
+	}
+	if !op.AllowSingleHosts {
+		t.Fatal("AllowSingleHosts should be true")
+	}
+}

ipn/serve.go

@@ -626,7 +626,7 @@ func (v ServeConfigView) HasAllowFunnel() bool {
 	}()
 }
 
-// FindFunnel reports whether target exists in in either the background AllowFunnel
+// FindFunnel reports whether target exists in either the background AllowFunnel
 // or any of the foreground configs.
 func (v ServeConfigView) HasFunnelForTarget(target HostPort) bool {
 	if v.AllowFunnel().Get(target) {

ipn/store/kubestore/store_kube.go

@@ -9,6 +9,7 @@ import (
 	"context"
 	"fmt"
 	"net"
+	"os"
 	"strings"
 	"time"
 
@@ -30,6 +31,10 @@ func New(_ logger.Logf, secretName string) (*Store, error) {
 	if err != nil {
 		return nil, err
 	}
+	if os.Getenv("TS_KUBERNETES_READ_API_SERVER_ADDRESS_FROM_ENV") == "true" {
+		// Derive the API server address from the environment variables
+		c.SetURL(fmt.Sprintf("https://%s:%s", os.Getenv("KUBERNETES_SERVICE_HOST"), os.Getenv("KUBERNETES_SERVICE_PORT_HTTPS")))
+	}
 	canPatch, _, err := c.CheckSecretPermissions(context.Background(), secretName)
 	if err != nil {
 		return nil, err

k8s-operator/tsdns.go

@@ -1,23 +0,0 @@
-// Copyright (c) Tailscale Inc & AUTHORS
-// SPDX-License-Identifier: BSD-3-Clause
-
-//go:build !plan9
-
-package kube
-
-const (
-	Alpha1Version = "v1alpha1"
-
-	DNSRecordsCMName = "dnsrecords"
-	DNSRecordsCMKey  = "records.json"
-)
-
-type Records struct {
-	// Version is the version of this Records configuration. Version is
-	// written by the operator, i.e when it first populates the Records.
-	// k8s-nameserver must verify that it knows how to parse a given
-	// version.
-	Version string `json:"version"`
-	// IP4 contains a mapping of DNS names to IPv4 address(es).
-	IP4 map[string][]string `json:"ip4"`
-}

k8s-operator/utils.go

@@ -0,0 +1,49 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build !plan9
+
+package kube
+
+import (
+	"fmt"
+
+	"tailscale.com/tailcfg"
+)
+
+const (
+	Alpha1Version = "v1alpha1"
+
+	DNSRecordsCMName = "dnsrecords"
+	DNSRecordsCMKey  = "records.json"
+)
+
+type Records struct {
+	// Version is the version of this Records configuration. Version is
+	// written by the operator, i.e when it first populates the Records.
+	// k8s-nameserver must verify that it knows how to parse a given
+	// version.
+	Version string `json:"version"`
+	// IP4 contains a mapping of DNS names to IPv4 address(es).
+	IP4 map[string][]string `json:"ip4"`
+}
+
+// TailscaledConfigFileNameForCap returns a tailscaled config file name in
+// format expected by containerboot for the given CapVer.
+func TailscaledConfigFileNameForCap(cap tailcfg.CapabilityVersion) string {
+	if cap < 95 {
+		return "tailscaled"
+	}
+	return fmt.Sprintf("cap-%v.hujson", cap)
+}
+
+// CapVerFromFileName parses the capability version from a tailscaled
+// config file name previously generated by TailscaledConfigFileNameForCap.
+func CapVerFromFileName(name string) (tailcfg.CapabilityVersion, error) {
+	if name == "tailscaled" {
+		return 0, nil
+	}
+	var cap tailcfg.CapabilityVersion
+	_, err := fmt.Sscanf(name, "cap-%d.hujson", &cap)
+	return cap, err
+}

licenses/android.md

@@ -35,7 +35,7 @@ Client][].  See also the dependencies in the [Tailscale CLI][].
  - [github.com/golang/groupcache/lru](https://pkg.go.dev/github.com/golang/groupcache/lru) ([Apache-2.0](https://github.com/golang/groupcache/blob/41bb18bfe9da/LICENSE))
  - [github.com/google/btree](https://pkg.go.dev/github.com/google/btree) ([Apache-2.0](https://github.com/google/btree/blob/v1.1.2/LICENSE))
  - [github.com/google/nftables](https://pkg.go.dev/github.com/google/nftables) ([Apache-2.0](https://github.com/google/nftables/blob/5e242ec57806/LICENSE))
- - [github.com/google/uuid](https://pkg.go.dev/github.com/google/uuid) ([BSD-3-Clause](https://github.com/google/uuid/blob/v1.5.0/LICENSE))
+ - [github.com/google/uuid](https://pkg.go.dev/github.com/google/uuid) ([BSD-3-Clause](https://github.com/google/uuid/blob/v1.6.0/LICENSE))
  - [github.com/hdevalence/ed25519consensus](https://pkg.go.dev/github.com/hdevalence/ed25519consensus) ([BSD-3-Clause](https://github.com/hdevalence/ed25519consensus/blob/v0.2.0/LICENSE))
  - [github.com/illarion/gonotify](https://pkg.go.dev/github.com/illarion/gonotify) ([MIT](https://github.com/illarion/gonotify/blob/v1.0.1/LICENSE))
  - [github.com/insomniacslk/dhcp](https://pkg.go.dev/github.com/insomniacslk/dhcp) ([BSD-3-Clause](https://github.com/insomniacslk/dhcp/blob/8c70d406f6d2/LICENSE))

licenses/apple.md

@@ -58,6 +58,7 @@ See also the dependencies in the [Tailscale CLI][].
  - [github.com/miekg/dns](https://pkg.go.dev/github.com/miekg/dns) ([BSD-3-Clause](https://github.com/miekg/dns/blob/v1.1.58/LICENSE))
  - [github.com/mitchellh/go-ps](https://pkg.go.dev/github.com/mitchellh/go-ps) ([MIT](https://github.com/mitchellh/go-ps/blob/v1.0.0/LICENSE.md))
  - [github.com/pierrec/lz4/v4](https://pkg.go.dev/github.com/pierrec/lz4/v4) ([BSD-3-Clause](https://github.com/pierrec/lz4/blob/v4.1.21/LICENSE))
+ - [github.com/prometheus-community/pro-bing](https://pkg.go.dev/github.com/prometheus-community/pro-bing) ([MIT](https://github.com/prometheus-community/pro-bing/blob/v0.4.0/LICENSE))
  - [github.com/safchain/ethtool](https://pkg.go.dev/github.com/safchain/ethtool) ([Apache-2.0](https://github.com/safchain/ethtool/blob/v0.3.0/LICENSE))
  - [github.com/tailscale/golang-x-crypto](https://pkg.go.dev/github.com/tailscale/golang-x-crypto) ([BSD-3-Clause](https://github.com/tailscale/golang-x-crypto/blob/7ce1f622c780/LICENSE))
  - [github.com/tailscale/goupnp](https://pkg.go.dev/github.com/tailscale/goupnp) ([BSD-2-Clause](https://github.com/tailscale/goupnp/blob/c64d0f06ea05/LICENSE))

licenses/tailscale.md

@@ -47,7 +47,7 @@ Some packages may only be included on certain architectures or operating systems
  - [github.com/golang/groupcache/lru](https://pkg.go.dev/github.com/golang/groupcache/lru) ([Apache-2.0](https://github.com/golang/groupcache/blob/41bb18bfe9da/LICENSE))
  - [github.com/google/btree](https://pkg.go.dev/github.com/google/btree) ([Apache-2.0](https://github.com/google/btree/blob/v1.1.2/LICENSE))
  - [github.com/google/nftables](https://pkg.go.dev/github.com/google/nftables) ([Apache-2.0](https://github.com/google/nftables/blob/5e242ec57806/LICENSE))
- - [github.com/google/uuid](https://pkg.go.dev/github.com/google/uuid) ([BSD-3-Clause](https://github.com/google/uuid/blob/v1.5.0/LICENSE))
+ - [github.com/google/uuid](https://pkg.go.dev/github.com/google/uuid) ([BSD-3-Clause](https://github.com/google/uuid/blob/v1.6.0/LICENSE))
  - [github.com/gorilla/csrf](https://pkg.go.dev/github.com/gorilla/csrf) ([BSD-3-Clause](https://github.com/gorilla/csrf/blob/v1.7.2/LICENSE))
  - [github.com/gorilla/securecookie](https://pkg.go.dev/github.com/gorilla/securecookie) ([BSD-3-Clause](https://github.com/gorilla/securecookie/blob/v1.1.2/LICENSE))
  - [github.com/hdevalence/ed25519consensus](https://pkg.go.dev/github.com/hdevalence/ed25519consensus) ([BSD-3-Clause](https://github.com/hdevalence/ed25519consensus/blob/v0.2.0/LICENSE))
@@ -73,6 +73,7 @@ Some packages may only be included on certain architectures or operating systems
  - [github.com/peterbourgon/ff/v3](https://pkg.go.dev/github.com/peterbourgon/ff/v3) ([Apache-2.0](https://github.com/peterbourgon/ff/blob/v3.4.0/LICENSE))
  - [github.com/pierrec/lz4/v4](https://pkg.go.dev/github.com/pierrec/lz4/v4) ([BSD-3-Clause](https://github.com/pierrec/lz4/blob/v4.1.21/LICENSE))
  - [github.com/pkg/sftp](https://pkg.go.dev/github.com/pkg/sftp) ([BSD-2-Clause](https://github.com/pkg/sftp/blob/v1.13.6/LICENSE))
+ - [github.com/prometheus-community/pro-bing](https://pkg.go.dev/github.com/prometheus-community/pro-bing) ([MIT](https://github.com/prometheus-community/pro-bing/blob/v0.4.0/LICENSE))
  - [github.com/safchain/ethtool](https://pkg.go.dev/github.com/safchain/ethtool) ([Apache-2.0](https://github.com/safchain/ethtool/blob/v0.3.0/LICENSE))
  - [github.com/skip2/go-qrcode](https://pkg.go.dev/github.com/skip2/go-qrcode) ([MIT](https://github.com/skip2/go-qrcode/blob/da1b6568686e/LICENSE))
  - [github.com/tailscale/certstore](https://pkg.go.dev/github.com/tailscale/certstore) ([MIT](https://github.com/tailscale/certstore/blob/d3fa0460f47e/LICENSE.md))

logtail/backoff/backoff.go

@@ -44,9 +44,8 @@ func NewBackoff(name string, logf logger.Logf, maxBackoff time.Duration) *Backof
 	}
 }
 
-// Backoff sleeps an increasing amount of time if err is non-nil.
-// and the context is not a
-// It resets the backoff schedule once err is nil.
+// BackOff sleeps an increasing amount of time if err is non-nil while the
+// context is active. It resets the backoff schedule once err is nil.
 func (b *Backoff) BackOff(ctx context.Context, err error) {
 	if err == nil {
 		// No error. Reset number of consecutive failures.

net/dns/manager.go

@@ -262,6 +262,18 @@ func (m *Manager) compileConfig(cfg Config) (rcfg resolver.Config, ocfg OSConfig
 		// config is empty, then we need to fallback to SplitDNS mode.
 		ocfg.MatchDomains = cfg.matchDomains()
 	} else {
+		// On iOS only (for now), check if all route names point to resources inside the tailnet.
+		// If so, we can set those names as MatchDomains to enable a split DNS configuration
+		// which will help preserve battery life.
+		// Because on iOS MatchDomains must equal SearchDomains, we cannot do this when
+		// we have any Routes outside the tailnet. Otherwise when app connectors are enabled,
+		// a query for 'work-laptop' might lead to search domain expansion, resolving
+		// as 'work-laptop.aws.com' for example.
+		if runtime.GOOS == "ios" && rcfg.RoutesRequireNoCustomResolvers() {
+			for r := range rcfg.Routes {
+				ocfg.MatchDomains = append(ocfg.MatchDomains, r)
+			}
+		}
 		var defaultRoutes []*dnstype.Resolver
 		for _, ip := range baseCfg.Nameservers {
 			defaultRoutes = append(defaultRoutes, &dnstype.Resolver{Addr: ip.String()})

net/dns/publicdns/publicdns.go

@@ -125,8 +125,8 @@ func DoHIPsOfBase(dohBase string) []netip.Addr {
 		return []netip.Addr{
 			controlDv4One,
 			controlDv4Two,
-			controlDv6Gen(nextDNSv6RangeA.Addr(), pathStr),
-			controlDv6Gen(nextDNSv6RangeB.Addr(), pathStr),
+			controlDv6Gen(controlDv6RangeA.Addr(), pathStr),
+			controlDv6Gen(controlDv6RangeB.Addr(), pathStr),
 		}
 	}
 	return nil

net/dns/publicdns/publicdns_test.go

@@ -121,8 +121,8 @@ func TestDoHIPsOfBase(t *testing.T) {
 			want: ips(
 				"76.76.2.22",
 				"76.76.10.22",
-				"2a07:a8c0:0:6:7b5b:5949:35ad:0",
-				"2a07:a8c1:0:6:7b5b:5949:35ad:0",
+				"2606:1a40:0:6:7b5b:5949:35ad:0",
+				"2606:1a40:1:6:7b5b:5949:35ad:0",
 			),
 		},
 		{
@@ -130,8 +130,8 @@ func TestDoHIPsOfBase(t *testing.T) {
 			want: ips(
 				"76.76.2.22",
 				"76.76.10.22",
-				"2a07:a8c0:0:ffff:ffff:ffff:ffff:0",
-				"2a07:a8c1:0:ffff:ffff:ffff:ffff:0",
+				"2606:1a40:0:ffff:ffff:ffff:ffff:0",
+				"2606:1a40:1:ffff:ffff:ffff:ffff:0",
 			),
 		},
 	}

net/dns/resolver/forwarder.go

@@ -391,20 +391,8 @@ func (f *forwarder) getKnownDoHClientForProvider(urlBase string) (c *http.Client
 	if err != nil {
 		return nil, false
 	}
-	// NOTE: use f.dialer.SystemDial so we close connections on a link
-	// change; on mobile devices when switching between WiFi and cellular,
-	// we need to ensure we don't retain a connection on the old interface
-	// or we can block DNS resolution.
-	//
-	// NOTE: if we ever support arbitrary user-defined DoH providers, this
-	// isn't sufficient; we'd need a dialer that dial a DoH server on the
-	// internet, without going through Tailscale (as SystemDial does), but
-	// also can dial a node on the tailnet (e.g. a PiHole).
-	//
-	// As of the time of writing (2024-02-11), this isn't a problem because
-	// we only support a restricted set of public DoH providers that aren't
-	// on a user's tailnet.
-	dialer := dnscache.Dialer(f.dialer.SystemDial, &dnscache.Resolver{
+
+	dialer := dnscache.Dialer(f.getDialerType(), &dnscache.Resolver{
 		SingleHost:             dohURL.Hostname(),
 		SingleHostStaticResult: allIPs,
 		Logf:                   f.logf,
@@ -699,6 +687,23 @@ func (f *forwarder) sendUDP(ctx context.Context, fq *forwardQuery, rr resolverAn
 	return out, nil
 }
 
+func (f *forwarder) getDialerType() dnscache.DialContextFunc {
+	if f.controlKnobs != nil && f.controlKnobs.UserDialUseRoutes.Load() {
+		// It is safe to use UserDial as it dials external servers without going through Tailscale
+		// and closes connections on interface change in the same way as SystemDial does,
+		// thus preventing DNS resolution issues when switching between WiFi and cellular,
+		// but can also dial an internal DNS server on the Tailnet or via a subnet router.
+		//
+		// TODO(nickkhyl): Update tsdial.Dialer to reuse the bart.Table we create in net/tstun.Wrapper
+		// to avoid having two bart tables in memory, especially on iOS. Once that's done,
+		// we can get rid of the nodeAttr/control knob and always use UserDial for DNS.
+		//
+		// See https://github.com/tailscale/tailscale/issues/12027.
+		return f.dialer.UserDial
+	}
+	return f.dialer.SystemDial
+}
+
 func (f *forwarder) sendTCP(ctx context.Context, fq *forwardQuery, rr resolverAndDelay) (ret []byte, err error) {
 	ipp, ok := rr.name.IPPort()
 	if !ok {
@@ -717,7 +722,7 @@ func (f *forwarder) sendTCP(ctx context.Context, fq *forwardQuery, rr resolverAn
 	ctx, cancel := context.WithTimeout(ctx, tcpQueryTimeout)
 	defer cancel()
 
-	conn, err := f.dialer.SystemDial(ctx, tcpFam, ipp.String())
+	conn, err := f.getDialerType()(ctx, tcpFam, ipp.String())
 	if err != nil {
 		return nil, err
 	}

net/dns/resolver/tsdns.go

@@ -175,6 +175,25 @@ func WriteRoutes(w *bufio.Writer, routes map[dnsname.FQDN][]*dnstype.Resolver) {
 	}
 }
 
+// RoutesRequireNoCustomResolvers returns true if this resolver.Config only contains routes
+// that do not specify a set of custom resolver(s), i.e. they can be resolved by the local
+// upstream DNS resolver.
+func (c *Config) RoutesRequireNoCustomResolvers() bool {
+	for route, resolvers := range c.Routes {
+		if route.WithoutTrailingDot() == "ts.net" {
+			// Ignore the "ts.net" route here. It always specifies the corp resolvers but
+			// its presence is not an issue, as ts.net will be a search domain.
+			continue
+		}
+		if len(resolvers) != 0 {
+			// Found a route with custom resolvers.
+			return false
+		}
+	}
+	// No routes other than ts.net have specified one or more resolvers.
+	return true
+}
+
 // Resolver is a DNS resolver for nodes on the Tailscale network,
 // associating them with domain names of the form <mynode>.<mydomain>.<root>.
 // If it is asked to resolve a domain that is not of that form,

net/dns/resolver/tsdns_test.go

@@ -243,6 +243,43 @@ func mustIP(str string) netip.Addr {
 	return ip
 }
 
+func TestRoutesRequireNoCustomResolvers(t *testing.T) {
+	tests := []struct {
+		name     string
+		config   Config
+		expected bool
+	}{
+		{"noRoutes", Config{Routes: map[dnsname.FQDN][]*dnstype.Resolver{}}, true},
+		{"onlyDefault", Config{Routes: map[dnsname.FQDN][]*dnstype.Resolver{
+			"ts.net.": {
+				{},
+			},
+		}}, true},
+		{"oneOther", Config{Routes: map[dnsname.FQDN][]*dnstype.Resolver{
+			"example.com.": {
+				{},
+			},
+		}}, false},
+		{"defaultAndOneOther", Config{Routes: map[dnsname.FQDN][]*dnstype.Resolver{
+			"ts.net.": {
+				{},
+			},
+			"example.com.": {
+				{},
+			},
+		}}, false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := tt.config.RoutesRequireNoCustomResolvers()
+			if result != tt.expected {
+				t.Errorf("result = %v; want %v", result, tt.expected)
+			}
+		})
+	}
+}
+
 func TestRDNSNameToIPv4(t *testing.T) {
 	tests := []struct {
 		name   string

net/netcheck/netcheck.go

@@ -13,6 +13,7 @@ import (
 	"fmt"
 	"io"
 	"log"
+	"maps"
 	"math/rand"
 	"net"
 	"net/http"
@@ -63,9 +64,6 @@ const (
 	// icmpProbeTimeout is the maximum amount of time netcheck will spend
 	// probing with ICMP packets.
 	icmpProbeTimeout = 1 * time.Second
-	// hairpinCheckTimeout is the amount of time we wait for a
-	// hairpinned packet to come back.
-	hairpinCheckTimeout = 100 * time.Millisecond
 	// defaultActiveRetransmitTime is the retransmit interval we use
 	// for STUN probes when we're in steady state (not in start-up),
 	// but don't have previous latency information for a DERP
@@ -95,11 +93,6 @@ type Report struct {
 	// STUN server you're talking to (on IPv4).
 	MappingVariesByDestIP opt.Bool
 
-	// HairPinning is whether the router supports communicating
-	// between two local devices through the NATted public IP address
-	// (on IPv4).
-	HairPinning opt.Bool
-
 	// UPnP is whether UPnP appears present on the LAN.
 	// Empty means not checked.
 	UPnP opt.Bool
@@ -115,8 +108,11 @@ type Report struct {
 	RegionV4Latency map[int]time.Duration // keyed by DERP Region ID
 	RegionV6Latency map[int]time.Duration // keyed by DERP Region ID
 
-	GlobalV4 string // ip:port of global IPv4
-	GlobalV6 string // [ip]:port of global IPv6
+	GlobalV4Counters map[netip.AddrPort]int // number of times the endpoint was observed
+	GlobalV6Counters map[netip.AddrPort]int // number of times the endpoint was observed
+
+	GlobalV4 netip.AddrPort
+	GlobalV6 netip.AddrPort
 
 	// CaptivePortal is set when we think there's a captive portal that is
 	// intercepting HTTP traffic.
@@ -125,6 +121,43 @@ type Report struct {
 	// TODO: update Clone when adding new fields
 }
 
+// GetGlobalAddrs returns the v4 and v6 global addresses observed during the
+// netcheck, which includes the best latency endpoint first, followed by any
+// other endpoints that were observed repeatedly. It excludes singular endpoints
+// that are likely only the result of a hard NAT.
+func (r *Report) GetGlobalAddrs() (v4, v6 []netip.AddrPort) {
+	// Always add the best latency entries first.
+	if r.GlobalV4.IsValid() {
+		v4 = append(v4, r.GlobalV4)
+	}
+	if r.GlobalV6.IsValid() {
+		v6 = append(v6, r.GlobalV6)
+	}
+	// Add any other entries for which we have multiple observations.
+	// This covers a case of bad NATs that start to provide new mappings for new
+	// STUN sessions mid-expiration, even while a live mapping for the best
+	// latency endpoint still exists. This has been observed on some Palo Alto
+	// Networks firewalls, wherein new traffic to the old endpoint will not
+	// succeed, but new traffic to the newly discovered endpoints does succeed.
+	for ipp, count := range r.GlobalV4Counters {
+		if ipp == r.GlobalV4 {
+			continue
+		}
+		if count > 1 {
+			v4 = append(v4, ipp)
+		}
+	}
+	for ipp, count := range r.GlobalV6Counters {
+		if ipp == r.GlobalV6 {
+			continue
+		}
+		if count > 1 {
+			v6 = append(v6, ipp)
+		}
+	}
+	return v4, v6
+}
+
 // AnyPortMappingChecked reports whether any of UPnP, PMP, or PCP are non-empty.
 func (r *Report) AnyPortMappingChecked() bool {
 	return r.UPnP != "" || r.PMP != "" || r.PCP != ""
@@ -138,6 +171,8 @@ func (r *Report) Clone() *Report {
 	r2.RegionLatency = cloneDurationMap(r2.RegionLatency)
 	r2.RegionV4Latency = cloneDurationMap(r2.RegionV4Latency)
 	r2.RegionV6Latency = cloneDurationMap(r2.RegionV6Latency)
+	r2.GlobalV4Counters = maps.Clone(r2.GlobalV4Counters)
+	r2.GlobalV6Counters = maps.Clone(r2.GlobalV6Counters)
 	return &r2
 }
 
@@ -243,23 +278,6 @@ func (c *Client) vlogf(format string, a ...any) {
 	}
 }
 
-// handleHairSTUN reports whether pkt (from src) was our magic hairpin
-// probe packet that we sent to ourselves.
-func (c *Client) handleHairSTUNLocked(pkt []byte, src netip.AddrPort) bool {
-	rs := c.curState
-	if rs == nil {
-		return false
-	}
-	if tx, err := stun.ParseBindingRequest(pkt); err == nil && tx == rs.hairTX {
-		select {
-		case rs.gotHairSTUN <- src:
-		default:
-		}
-		return true
-	}
-	return false
-}
-
 // MakeNextReportFull forces the next GetReport call to be a full
 // (non-incremental) probe of all DERP regions.
 func (c *Client) MakeNextReportFull() {
@@ -282,10 +300,6 @@ func (c *Client) ReceiveSTUNPacket(pkt []byte, src netip.AddrPort) {
 	}
 
 	c.mu.Lock()
-	if c.handleHairSTUNLocked(pkt, src) {
-		c.mu.Unlock()
-		return
-	}
 	rs := c.curState
 	c.mu.Unlock()
 
@@ -296,6 +310,8 @@ func (c *Client) ReceiveSTUNPacket(pkt []byte, src netip.AddrPort) {
 	tx, addrPort, err := stun.ParseResponse(pkt)
 	if err != nil {
 		if _, err := stun.ParseBindingRequest(pkt); err == nil {
+			// We no longer send hairpin checks, but perhaps we might catch a
+			// stray from earlier versions.
 			// This was probably our own netcheck hairpin
 			// check probe coming in late. Ignore.
 			return
@@ -450,10 +466,10 @@ func makeProbePlan(dm *tailcfg.DERPMap, ifState *netmon.State, last *Report) (pl
 			if try > 1 {
 				delay += time.Duration(try) * 50 * time.Millisecond
 			}
-			if do4 || n.IsTestNode() {
+			if n.IPv4 != "none" && (do4 || n.IsTestNode()) {
 				p4 = append(p4, probe{delay: delay, node: n.Name, proto: probeIPv4})
 			}
-			if do6 || n.IsTestNode() {
+			if n.IPv6 != "none" && (do6 || n.IsTestNode()) {
 				p6 = append(p6, probe{delay: delay, node: n.Name, proto: probeIPv6})
 			}
 		}
@@ -476,10 +492,10 @@ func makeProbePlanInitial(dm *tailcfg.DERPMap, ifState *netmon.State) (plan prob
 		for try := 0; try < 3; try++ {
 			n := reg.Nodes[try%len(reg.Nodes)]
 			delay := time.Duration(try) * defaultInitialRetransmitTime
-			if ifState.HaveV4 && nodeMight4(n) || n.IsTestNode() {
+			if n.IPv4 != "none" && ((ifState.HaveV4 && nodeMight4(n)) || n.IsTestNode()) {
 				p4 = append(p4, probe{delay: delay, node: n.Name, proto: probeIPv4})
 			}
-			if ifState.HaveV6 && nodeMight6(n) || n.IsTestNode() {
+			if n.IPv6 != "none" && ((ifState.HaveV6 && nodeMight6(n)) || n.IsTestNode()) {
 				p6 = append(p6, probe{delay: delay, node: n.Name, proto: probeIPv6})
 			}
 		}
@@ -521,20 +537,15 @@ type reportState struct {
 	c           *Client
 	start       time.Time
 	opts        *GetReportOpts
-	hairTX      stun.TxID
-	gotHairSTUN chan netip.AddrPort
-	hairTimeout chan struct{} // closed on timeout
-	pc4Hair     nettype.PacketConn
 	incremental bool // doing a lite, follow-up netcheck
 	stopProbeCh chan struct{}
 	waitPortMap sync.WaitGroup
 
-	mu            sync.Mutex
-	sentHairCheck bool
-	report        *Report                            // to be returned by GetReport
-	inFlight      map[stun.TxID]func(netip.AddrPort) // called without c.mu held
-	gotEP4        string
-	timers        []*time.Timer
+	mu       sync.Mutex
+	report   *Report                            // to be returned by GetReport
+	inFlight map[stun.TxID]func(netip.AddrPort) // called without c.mu held
+	gotEP4   netip.AddrPort
+	timers   []*time.Timer
 }
 
 func (rs *reportState) anyUDP() bool {
@@ -584,50 +595,6 @@ func (rs *reportState) probeWouldHelp(probe probe, node *tailcfg.DERPNode) bool
 	return false
 }
 
-func (rs *reportState) startHairCheckLocked(dst netip.AddrPort) {
-	if rs.sentHairCheck || rs.incremental {
-		return
-	}
-	rs.sentHairCheck = true
-	rs.pc4Hair.WriteToUDPAddrPort(stun.Request(rs.hairTX), dst)
-	rs.c.vlogf("sent haircheck to %v", dst)
-	time.AfterFunc(hairpinCheckTimeout, func() { close(rs.hairTimeout) })
-}
-
-func (rs *reportState) waitHairCheck(ctx context.Context) {
-	rs.mu.Lock()
-	defer rs.mu.Unlock()
-	ret := rs.report
-	if rs.incremental {
-		if rs.c.last != nil {
-			ret.HairPinning = rs.c.last.HairPinning
-		}
-		return
-	}
-	if !rs.sentHairCheck {
-		return
-	}
-
-	// First, check whether we have a value before we check for timeouts.
-	select {
-	case <-rs.gotHairSTUN:
-		ret.HairPinning.Set(true)
-		return
-	default:
-	}
-
-	// Now, wait for a response or a timeout.
-	select {
-	case <-rs.gotHairSTUN:
-		ret.HairPinning.Set(true)
-	case <-rs.hairTimeout:
-		rs.c.vlogf("hairCheck timeout")
-		ret.HairPinning.Set(false)
-	case <-ctx.Done():
-		rs.c.vlogf("hairCheck context timeout")
-	}
-}
-
 func (rs *reportState) stopTimers() {
 	rs.mu.Lock()
 	defer rs.mu.Unlock()
@@ -640,11 +607,6 @@ func (rs *reportState) stopTimers() {
 // is non-zero (for all but HTTPS replies), it's recorded as our UDP
 // IP:port.
 func (rs *reportState) addNodeLatency(node *tailcfg.DERPNode, ipp netip.AddrPort, d time.Duration) {
-	var ipPortStr string
-	if ipp != (netip.AddrPort{}) {
-		ipPortStr = net.JoinHostPort(ipp.Addr().String(), fmt.Sprint(ipp.Port()))
-	}
-
 	rs.mu.Lock()
 	defer rs.mu.Unlock()
 	ret := rs.report
@@ -670,18 +632,19 @@ func (rs *reportState) addNodeLatency(node *tailcfg.DERPNode, ipp netip.AddrPort
 	case ipp.Addr().Is6():
 		updateLatency(ret.RegionV6Latency, node.RegionID, d)
 		ret.IPv6 = true
-		ret.GlobalV6 = ipPortStr
+		ret.GlobalV6 = ipp
+		mak.Set(&ret.GlobalV6Counters, ipp, ret.GlobalV6Counters[ipp]+1)
 		// TODO: track MappingVariesByDestIP for IPv6
 		// too? Would be sad if so, but who knows.
 	case ipp.Addr().Is4():
 		updateLatency(ret.RegionV4Latency, node.RegionID, d)
 		ret.IPv4 = true
-		if rs.gotEP4 == "" {
-			rs.gotEP4 = ipPortStr
-			ret.GlobalV4 = ipPortStr
-			rs.startHairCheckLocked(ipp)
+		mak.Set(&ret.GlobalV4Counters, ipp, ret.GlobalV4Counters[ipp]+1)
+		if !rs.gotEP4.IsValid() {
+			rs.gotEP4 = ipp
+			ret.GlobalV4 = ipp
 		} else {
-			if rs.gotEP4 != ipPortStr {
+			if rs.gotEP4 != ipp {
 				ret.MappingVariesByDestIP.Set(true)
 			} else if ret.MappingVariesByDestIP == "" {
 				ret.MappingVariesByDestIP.Set(false)
@@ -793,9 +756,6 @@ func (c *Client) GetReport(ctx context.Context, dm *tailcfg.DERPMap, opts *GetRe
 		opts:        opts,
 		report:      newReport(),
 		inFlight:    map[stun.TxID]func(netip.AddrPort){},
-		hairTX:      stun.NewTxID(), // random payload
-		gotHairSTUN: make(chan netip.AddrPort, 1),
-		hairTimeout: make(chan struct{}),
 		stopProbeCh: make(chan struct{}, 1),
 	}
 	c.curState = rs
@@ -853,34 +813,11 @@ func (c *Client) GetReport(ctx context.Context, dm *tailcfg.DERPMap, opts *GetRe
 		v6udp.Close()
 	}
 
-	// Create a UDP4 socket used for sending to our discovered IPv4 address.
-	rs.pc4Hair, err = nettype.MakePacketListenerWithNetIP(netns.Listener(c.logf, c.NetMon)).ListenPacket(ctx, "udp4", ":0")
-	if err != nil {
-		c.logf("udp4: %v", err)
-		return nil, err
-	}
-	defer rs.pc4Hair.Close()
-
 	if !c.SkipExternalNetwork && c.PortMapper != nil {
 		rs.waitPortMap.Add(1)
 		go rs.probePortMapServices()
 	}
 
-	// At least the Apple Airport Extreme doesn't allow hairpin
-	// sends from a private socket until it's seen traffic from
-	// that src IP:port to something else out on the internet.
-	//
-	// See https://github.com/tailscale/tailscale/issues/188#issuecomment-600728643
-	//
-	// And it seems that even sending to a likely-filtered RFC 5737
-	// documentation-only IPv4 range is enough to set up the mapping.
-	// So do that for now. In the future we might want to classify networks
-	// that do and don't require this separately. But for now help it.
-	const documentationIP = "203.0.113.1"
-	rs.pc4Hair.WriteToUDPAddrPort(
-		[]byte("tailscale netcheck; see https://github.com/tailscale/tailscale/issues/188"),
-		netip.AddrPortFrom(netip.MustParseAddr(documentationIP), 12345))
-
 	plan := makeProbePlan(dm, ifState, last)
 
 	// If we're doing a full probe, also check for a captive portal. We
@@ -958,8 +895,6 @@ func (c *Client) GetReport(ctx context.Context, dm *tailcfg.DERPMap, opts *GetRe
 		captivePortalStop()
 	}
 
-	rs.waitHairCheck(ctx)
-	c.vlogf("hairCheck done")
 	if !c.SkipExternalNetwork && c.PortMapper != nil {
 		rs.waitPortMap.Wait()
 		c.vlogf("portMap done")
@@ -1328,17 +1263,16 @@ func (c *Client) logConciseReport(r *Report, dm *tailcfg.DERPMap) {
 			fmt.Fprintf(w, " v6os=%v", r.OSHasIPv6)
 		}
 		fmt.Fprintf(w, " mapvarydest=%v", r.MappingVariesByDestIP)
-		fmt.Fprintf(w, " hair=%v", r.HairPinning)
 		if r.AnyPortMappingChecked() {
 			fmt.Fprintf(w, " portmap=%v%v%v", conciseOptBool(r.UPnP, "U"), conciseOptBool(r.PMP, "M"), conciseOptBool(r.PCP, "C"))
 		} else {
 			fmt.Fprintf(w, " portmap=?")
 		}
-		if r.GlobalV4 != "" {
-			fmt.Fprintf(w, " v4a=%v", r.GlobalV4)
+		if r.GlobalV4.IsValid() {
+			fmt.Fprintf(w, " v4a=%s", r.GlobalV4)
 		}
-		if r.GlobalV6 != "" {
-			fmt.Fprintf(w, " v6a=%v", r.GlobalV6)
+		if r.GlobalV6.IsValid() {
+			fmt.Fprintf(w, " v6a=%s", r.GlobalV6)
 		}
 		if r.CaptivePortal != "" {
 			fmt.Fprintf(w, " captiveportal=%v", r.CaptivePortal)