util/workgraph: add package for concurrent execution of DAGs

This package is intended to be used for cleaning up the mess of
concurrent things happening in the netcheck package.

Updates #10972

Signed-off-by: Andrew Dunham <andrew@du.nham.ca>
Change-Id: I609b61a7838b84a74b74bdef66d1d4c4014705e1

.github/workflows/ssh-integrationtest.yml

@@ -0,0 +1,23 @@
+# Run the ssh integration tests with `make sshintegrationtest`.
+# These tests can also be running locally.
+name: "ssh-integrationtest"
+
+concurrency:
+  group: ${{ github.workflow }}-$${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+on:
+  pull_request:
+    paths:
+      - "ssh/**"
+      - "tempfork/gliderlabs/ssh/**"
+      - ".github/workflows/ssh-integrationtest"
+jobs:
+  ssh-integrationtest:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v4
+      - name: Run SSH integration tests
+        run: |
+          make sshintegrationtest

.github/workflows/test.yml

@@ -194,7 +194,7 @@ jobs:
     - name: chown
       run: chown -R $(id -u):$(id -g) $PWD
     - name: privileged tests
-      run: ./tool/go test ./util/linuxfw
+      run: ./tool/go test ./util/linuxfw ./derp/xdp
 
   vm:
     runs-on: ["self-hosted", "linux", "vm"]
@@ -480,7 +480,7 @@ jobs:
       uses: actions/checkout@v4
     - name: check that 'go generate' is clean
       run: |
-        pkgs=$(./tool/go list ./... | grep -Ev 'dnsfallback|k8s-operator')
+        pkgs=$(./tool/go list ./... | grep -Ev 'dnsfallback|k8s-operator|xdp')
         ./tool/go generate $pkgs
         echo
         echo

Makefile

@@ -110,15 +110,12 @@ publishdevnameserver: ## Build and publish k8s-nameserver image to location spec
 
 .PHONY: sshintegrationtest
 sshintegrationtest: ## Run the SSH integration tests in various Docker containers
-	@GOOS=linux GOARCH=amd64 go test -tags integrationtest -c ./ssh/tailssh -o ssh/tailssh/testcontainers/tailssh.test && \
-	GOOS=linux GOARCH=amd64 go build -o ssh/tailssh/testcontainers/tailscaled ./cmd/tailscaled && \
+	@GOOS=linux GOARCH=amd64 ./tool/go test -tags integrationtest -c ./ssh/tailssh -o ssh/tailssh/testcontainers/tailssh.test && \
+	GOOS=linux GOARCH=amd64 ./tool/go build -o ssh/tailssh/testcontainers/tailscaled ./cmd/tailscaled && \
 	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
+1.69.0

appc/appconnector.go

@@ -15,6 +15,7 @@ import (
 	"slices"
 	"strings"
 	"sync"
+	"time"
 
 	xmaps "golang.org/x/exp/maps"
 	"golang.org/x/net/dns/dnsmessage"
@@ -26,6 +27,46 @@ import (
 	"tailscale.com/util/slicesx"
 )
 
+// rateLogger responds to calls to update by adding a count for the current period and
+// calling the callback if any previous period has finished since update was last called
+type rateLogger struct {
+	interval    time.Duration
+	start       time.Time
+	periodStart time.Time
+	periodCount int64
+	now         func() time.Time
+	callback    func(int64, time.Time, int64)
+}
+
+func (rl *rateLogger) currentIntervalStart(now time.Time) time.Time {
+	millisSince := now.Sub(rl.start).Milliseconds() % rl.interval.Milliseconds()
+	return now.Add(-(time.Duration(millisSince)) * time.Millisecond)
+}
+
+func (rl *rateLogger) update(numRoutes int64) {
+	now := rl.now()
+	periodEnd := rl.periodStart.Add(rl.interval)
+	if periodEnd.Before(now) {
+		if rl.periodCount != 0 {
+			rl.callback(rl.periodCount, rl.periodStart, numRoutes)
+		}
+		rl.periodCount = 0
+		rl.periodStart = rl.currentIntervalStart(now)
+	}
+	rl.periodCount++
+}
+
+func newRateLogger(now func() time.Time, interval time.Duration, callback func(int64, time.Time, int64)) *rateLogger {
+	nowTime := now()
+	return &rateLogger{
+		callback:    callback,
+		now:         now,
+		interval:    interval,
+		start:       nowTime,
+		periodStart: nowTime,
+	}
+}
+
 // RouteAdvertiser is an interface that allows the AppConnector to advertise
 // newly discovered routes that need to be served through the AppConnector.
 type RouteAdvertiser interface {
@@ -81,6 +122,9 @@ type AppConnector struct {
 
 	// queue provides ordering for update operations
 	queue execqueue.ExecQueue
+
+	writeRateMinute *rateLogger
+	writeRateDay    *rateLogger
 }
 
 // NewAppConnector creates a new AppConnector.
@@ -95,6 +139,12 @@ func NewAppConnector(logf logger.Logf, routeAdvertiser RouteAdvertiser, routeInf
 		ac.wildcards = routeInfo.Wildcards
 		ac.controlRoutes = routeInfo.Control
 	}
+	ac.writeRateMinute = newRateLogger(time.Now, time.Minute, func(c int64, s time.Time, l int64) {
+		ac.logf("routeInfo write rate: %d in minute starting at %v (%d routes)", c, s, l)
+	})
+	ac.writeRateDay = newRateLogger(time.Now, 24*time.Hour, func(c int64, s time.Time, l int64) {
+		ac.logf("routeInfo write rate: %d in 24 hours starting at %v (%d routes)", c, s, l)
+	})
 	return ac
 }
 
@@ -109,6 +159,15 @@ func (e *AppConnector) storeRoutesLocked() error {
 	if !e.ShouldStoreRoutes() {
 		return nil
 	}
+
+	// log write rate and write size
+	numRoutes := int64(len(e.controlRoutes))
+	for _, rs := range e.domains {
+		numRoutes += int64(len(rs))
+	}
+	e.writeRateMinute.update(numRoutes)
+	e.writeRateDay.update(numRoutes)
+
 	return e.storeRoutesFunc(&RouteInfo{
 		Control:   e.controlRoutes,
 		Domains:   e.domains,

appc/appconnector_test.go

@@ -9,10 +9,12 @@ import (
 	"reflect"
 	"slices"
 	"testing"
+	"time"
 
 	xmaps "golang.org/x/exp/maps"
 	"golang.org/x/net/dns/dnsmessage"
 	"tailscale.com/appc/appctest"
+	"tailscale.com/tstest"
 	"tailscale.com/util/mak"
 	"tailscale.com/util/must"
 )
@@ -520,3 +522,50 @@ func TestRoutesWithout(t *testing.T) {
 	assert("a has fewer", routesWithout(prefixes("1.1.1.1/32", "1.1.1.2/32"), prefixes("1.1.1.1/32", "1.1.1.2/32", "1.1.1.3/32", "1.1.1.4/32")), []netip.Prefix{})
 	assert("a has more", routesWithout(prefixes("1.1.1.1/32", "1.1.1.2/32", "1.1.1.3/32", "1.1.1.4/32"), prefixes("1.1.1.1/32", "1.1.1.3/32")), prefixes("1.1.1.2/32", "1.1.1.4/32"))
 }
+
+func TestRateLogger(t *testing.T) {
+	clock := tstest.Clock{}
+	wasCalled := false
+	rl := newRateLogger(func() time.Time { return clock.Now() }, 1*time.Second, func(count int64, _ time.Time, _ int64) {
+		if count != 3 {
+			t.Fatalf("count for prev period: got %d, want 3", count)
+		}
+		wasCalled = true
+	})
+
+	for i := 0; i < 3; i++ {
+		clock.Advance(1 * time.Millisecond)
+		rl.update(0)
+		if wasCalled {
+			t.Fatalf("wasCalled: got true, want false")
+		}
+	}
+
+	clock.Advance(1 * time.Second)
+	rl.update(0)
+	if !wasCalled {
+		t.Fatalf("wasCalled: got false, want true")
+	}
+
+	wasCalled = false
+	rl = newRateLogger(func() time.Time { return clock.Now() }, 1*time.Hour, func(count int64, _ time.Time, _ int64) {
+		if count != 3 {
+			t.Fatalf("count for prev period: got %d, want 3", count)
+		}
+		wasCalled = true
+	})
+
+	for i := 0; i < 3; i++ {
+		clock.Advance(1 * time.Minute)
+		rl.update(0)
+		if wasCalled {
+			t.Fatalf("wasCalled: got true, want false")
+		}
+	}
+
+	clock.Advance(1 * time.Hour)
+	rl.update(0)
+	if !wasCalled {
+		t.Fatalf("wasCalled: got false, want true")
+	}
+}

client/tailscale/devices.go

@@ -10,6 +10,7 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"log"
 	"net/http"
 	"net/url"
 
@@ -39,6 +40,7 @@ type Device struct {
 	// It's currently just 1 element, the 100.x.y.z Tailscale IP.
 	Addresses []string `json:"addresses"`
 	DeviceID  string   `json:"id"`
+	NodeID    string   `json:"nodeId"`
 	User      string   `json:"user"`
 	Name      string   `json:"name"`
 	Hostname  string   `json:"hostname"`
@@ -213,6 +215,9 @@ func (c *Client) DeleteDevice(ctx context.Context, deviceID string) (err error)
 	if err != nil {
 		return err
 	}
+
+	log.Printf("RESP: %di, path: %s", resp.StatusCode, path)
+
 	// If status code was not successful, return the error.
 	// TODO: Change the check for the StatusCode to include other 2XX success codes.
 	if resp.StatusCode != http.StatusOK {

client/tailscale/localclient.go

@@ -253,11 +253,16 @@ func (lc *LocalClient) sendWithHeaders(
 	}
 	if res.StatusCode != wantStatus {
 		err = fmt.Errorf("%v: %s", res.Status, bytes.TrimSpace(slurp))
-		return nil, nil, bestError(err, slurp)
+		return nil, nil, httpStatusError{bestError(err, slurp), res.StatusCode}
 	}
 	return slurp, res.Header, nil
 }
 
+type httpStatusError struct {
+	error
+	HTTPStatus int
+}
+
 func (lc *LocalClient) get200(ctx context.Context, path string) ([]byte, error) {
 	return lc.send(ctx, "GET", path, 200, nil)
 }
@@ -278,9 +283,50 @@ func decodeJSON[T any](b []byte) (ret T, err error) {
 }
 
 // WhoIs returns the owner of the remoteAddr, which must be an IP or IP:port.
+//
+// If not found, the error is ErrPeerNotFound.
+//
+// For connections proxied by tailscaled, this looks up the owner of the given
+// address as TCP first, falling back to UDP; if you want to only check a
+// specific address family, use WhoIsProto.
 func (lc *LocalClient) WhoIs(ctx context.Context, remoteAddr string) (*apitype.WhoIsResponse, error) {
 	body, err := lc.get200(ctx, "/localapi/v0/whois?addr="+url.QueryEscape(remoteAddr))
 	if err != nil {
+		if hs, ok := err.(httpStatusError); ok && hs.HTTPStatus == http.StatusNotFound {
+			return nil, ErrPeerNotFound
+		}
+		return nil, err
+	}
+	return decodeJSON[*apitype.WhoIsResponse](body)
+}
+
+// ErrPeerNotFound is returned by WhoIs and WhoIsNodeKey when a peer is not found.
+var ErrPeerNotFound = errors.New("peer not found")
+
+// WhoIsNodeKey returns the owner of the given wireguard public key.
+//
+// If not found, the error is ErrPeerNotFound.
+func (lc *LocalClient) WhoIsNodeKey(ctx context.Context, key key.NodePublic) (*apitype.WhoIsResponse, error) {
+	body, err := lc.get200(ctx, "/localapi/v0/whois?addr="+url.QueryEscape(key.String()))
+	if err != nil {
+		if hs, ok := err.(httpStatusError); ok && hs.HTTPStatus == http.StatusNotFound {
+			return nil, ErrPeerNotFound
+		}
+		return nil, err
+	}
+	return decodeJSON[*apitype.WhoIsResponse](body)
+}
+
+// WhoIsProto returns the owner of the remoteAddr, which must be an IP or
+// IP:port, for the given protocol (tcp or udp).
+//
+// If not found, the error is ErrPeerNotFound.
+func (lc *LocalClient) WhoIsProto(ctx context.Context, proto, remoteAddr string) (*apitype.WhoIsResponse, error) {
+	body, err := lc.get200(ctx, "/localapi/v0/whois?proto="+url.QueryEscape(proto)+"&addr="+url.QueryEscape(remoteAddr))
+	if err != nil {
+		if hs, ok := err.(httpStatusError); ok && hs.HTTPStatus == http.StatusNotFound {
+			return nil, ErrPeerNotFound
+		}
 		return nil, err
 	}
 	return decodeJSON[*apitype.WhoIsResponse](body)
@@ -699,6 +745,27 @@ func (lc *LocalClient) CheckUDPGROForwarding(ctx context.Context) error {
 	return nil
 }
 
+// SetUDPGROForwarding enables UDP GRO forwarding for the main interface of this
+// node. This can be done to improve performance of tailnet nodes acting as exit
+// nodes or subnet routers.
+// See https://tailscale.com/kb/1320/performance-best-practices#linux-optimizations-for-subnet-routers-and-exit-nodes
+func (lc *LocalClient) SetUDPGROForwarding(ctx context.Context) error {
+	body, err := lc.get200(ctx, "/localapi/v0/set-udp-gro-forwarding")
+	if err != nil {
+		return err
+	}
+	var jres struct {
+		Warning string
+	}
+	if err := json.Unmarshal(body, &jres); err != nil {
+		return fmt.Errorf("invalid JSON from set-udp-gro-forwarding: %w", err)
+	}
+	if jres.Warning != "" {
+		return errors.New(jres.Warning)
+	}
+	return nil
+}
+
 // CheckPrefs validates the provided preferences, without making any changes.
 //
 // The CLI uses this before a Start call to fail fast if the preferences won't
@@ -778,6 +845,17 @@ 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 +868,11 @@ 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

@@ -6,9 +6,14 @@
 package tailscale
 
 import (
+	"context"
+	"net"
+	"net/http"
+	"net/http/httptest"
 	"testing"
 
 	"tailscale.com/tstest/deptest"
+	"tailscale.com/types/key"
 )
 
 func TestGetServeConfigFromJSON(t *testing.T) {
@@ -30,11 +35,38 @@ func TestGetServeConfigFromJSON(t *testing.T) {
 	}
 }
 
+func TestWhoIsPeerNotFound(t *testing.T) {
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(404)
+	}))
+	defer ts.Close()
+
+	lc := &LocalClient{
+		Dial: func(ctx context.Context, network, addr string) (net.Conn, error) {
+			var std net.Dialer
+			return std.DialContext(ctx, network, ts.Listener.Addr().(*net.TCPAddr).String())
+		},
+	}
+	var k key.NodePublic
+	if err := k.UnmarshalText([]byte("nodekey:5c8f86d5fc70d924e55f02446165a5dae8f822994ad26bcf4b08fd841f9bf261")); err != nil {
+		t.Fatal(err)
+	}
+	res, err := lc.WhoIsNodeKey(context.Background(), k)
+	if err != ErrPeerNotFound {
+		t.Errorf("got (%v, %v), want ErrPeerNotFound", res, err)
+	}
+	res, err = lc.WhoIs(context.Background(), "1.2.3.4:5678")
+	if err != ErrPeerNotFound {
+		t.Errorf("got (%v, %v), want ErrPeerNotFound", res, err)
+	}
+}
+
 func TestDeps(t *testing.T) {
 	deptest.DepChecker{
 		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

@@ -29,7 +29,6 @@ import (
 
 	"github.com/google/uuid"
 	"tailscale.com/clientupdate/distsign"
-	"tailscale.com/hostinfo"
 	"tailscale.com/types/logger"
 	"tailscale.com/util/cmpver"
 	"tailscale.com/util/winutil"
@@ -38,11 +37,18 @@ import (
 )
 
 const (
-	CurrentTrack  = ""
 	StableTrack   = "stable"
 	UnstableTrack = "unstable"
 )
 
+var CurrentTrack = func() string {
+	if version.IsUnstableBuild() {
+		return UnstableTrack
+	} else {
+		return StableTrack
+	}
+}()
+
 func versionToTrack(v string) (string, error) {
 	_, rest, ok := strings.Cut(v, ".")
 	if !ok {
@@ -107,7 +113,7 @@ func (args Arguments) validate() error {
 		return fmt.Errorf("only one of Version(%q) or Track(%q) can be set", args.Version, args.Track)
 	}
 	switch args.Track {
-	case StableTrack, UnstableTrack, CurrentTrack:
+	case StableTrack, UnstableTrack, "":
 		// All valid values.
 	default:
 		return fmt.Errorf("unsupported track %q", args.Track)
@@ -120,11 +126,17 @@ type Updater struct {
 	// Update is a platform-specific method that updates the installation. May be
 	// nil (not all platforms support updates from within Tailscale).
 	Update func() error
+
+	// currentVersion is the short form of the current client version as
+	// returned by version.Short(), typically "x.y.z". Used for tests to
+	// override the actual current version.
+	currentVersion string
 }
 
 func NewUpdater(args Arguments) (*Updater, error) {
 	up := Updater{
-		Arguments: args,
+		Arguments:      args,
+		currentVersion: version.Short(),
 	}
 	if up.Stdout == nil {
 		up.Stdout = os.Stdout
@@ -140,18 +152,15 @@ func NewUpdater(args Arguments) (*Updater, error) {
 	if args.ForAutoUpdate && !canAutoUpdate {
 		return nil, errors.ErrUnsupported
 	}
-	if up.Track == CurrentTrack {
-		switch {
-		case up.Version != "":
+	if up.Track == "" {
+		if up.Version != "" {
 			var err error
 			up.Track, err = versionToTrack(args.Version)
 			if err != nil {
 				return nil, err
 			}
-		case version.IsUnstableBuild():
-			up.Track = UnstableTrack
-		default:
-			up.Track = StableTrack
+		} else {
+			up.Track = CurrentTrack
 		}
 	}
 	if up.Arguments.PkgsAddr == "" {
@@ -163,10 +172,9 @@ func NewUpdater(args Arguments) (*Updater, error) {
 type updateFunction func() error
 
 func (up *Updater) getUpdateFunction() (fn updateFunction, canAutoUpdate bool) {
-	canAutoUpdate = !hostinfo.New().Container.EqualBool(true) // EqualBool(false) would return false if the value is not set.
 	switch runtime.GOOS {
 	case "windows":
-		return up.updateWindows, canAutoUpdate
+		return up.updateWindows, true
 	case "linux":
 		switch distro.Get() {
 		case distro.NixOS:
@@ -180,20 +188,20 @@ func (up *Updater) getUpdateFunction() (fn updateFunction, canAutoUpdate bool) {
 			// auto-update mechanism.
 			return up.updateSynology, false
 		case distro.Debian: // includes Ubuntu
-			return up.updateDebLike, canAutoUpdate
+			return up.updateDebLike, true
 		case distro.Arch:
 			if up.archPackageInstalled() {
 				// Arch update func just prints a message about how to update,
 				// it doesn't support auto-updates.
 				return up.updateArchLike, false
 			}
-			return up.updateLinuxBinary, canAutoUpdate
+			return up.updateLinuxBinary, true
 		case distro.Alpine:
-			return up.updateAlpineLike, canAutoUpdate
+			return up.updateAlpineLike, true
 		case distro.Unraid:
-			return up.updateUnraid, canAutoUpdate
+			return up.updateUnraid, true
 		case distro.QNAP:
-			return up.updateQNAP, canAutoUpdate
+			return up.updateQNAP, true
 		}
 		switch {
 		case haveExecutable("pacman"):
@@ -202,21 +210,21 @@ func (up *Updater) getUpdateFunction() (fn updateFunction, canAutoUpdate bool) {
 				// it doesn't support auto-updates.
 				return up.updateArchLike, false
 			}
-			return up.updateLinuxBinary, canAutoUpdate
+			return up.updateLinuxBinary, true
 		case haveExecutable("apt-get"): // TODO(awly): add support for "apt"
 			// The distro.Debian switch case above should catch most apt-based
 			// systems, but add this fallback just in case.
-			return up.updateDebLike, canAutoUpdate
+			return up.updateDebLike, true
 		case haveExecutable("dnf"):
-			return up.updateFedoraLike("dnf"), canAutoUpdate
+			return up.updateFedoraLike("dnf"), true
 		case haveExecutable("yum"):
-			return up.updateFedoraLike("yum"), canAutoUpdate
+			return up.updateFedoraLike("yum"), true
 		case haveExecutable("apk"):
-			return up.updateAlpineLike, canAutoUpdate
+			return up.updateAlpineLike, true
 		}
 		// If nothing matched, fall back to tarball updates.
 		if up.Update == nil {
-			return up.updateLinuxBinary, canAutoUpdate
+			return up.updateLinuxBinary, true
 		}
 	case "darwin":
 		switch {
@@ -232,7 +240,7 @@ func (up *Updater) getUpdateFunction() (fn updateFunction, canAutoUpdate bool) {
 			return nil, false
 		}
 	case "freebsd":
-		return up.updateFreeBSD, canAutoUpdate
+		return up.updateFreeBSD, true
 	}
 	return nil, false
 }
@@ -261,13 +269,16 @@ func Update(args Arguments) error {
 }
 
 func (up *Updater) confirm(ver string) bool {
-	switch cmpver.Compare(version.Short(), ver) {
-	case 0:
-		up.Logf("already running %v version %v; no update needed", up.Track, ver)
-		return false
-	case 1:
-		up.Logf("installed %v version %v is newer than the latest available version %v; no update needed", up.Track, version.Short(), ver)
-		return false
+	// Only check version when we're not switching tracks.
+	if up.Track == "" || up.Track == CurrentTrack {
+		switch c := cmpver.Compare(up.currentVersion, ver); {
+		case c == 0:
+			up.Logf("already running %v version %v; no update needed", up.Track, ver)
+			return false
+		case c > 0:
+			up.Logf("installed %v version %v is newer than the latest available version %v; no update needed", up.Track, up.currentVersion, ver)
+			return false
+		}
 	}
 	if up.Confirm != nil {
 		return up.Confirm(ver)
@@ -653,6 +664,9 @@ 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
 	}
 
@@ -680,7 +694,7 @@ func parseAlpinePackageVersion(out []byte) (string, error) {
 			return "", fmt.Errorf("malformed info line: %q", line)
 		}
 		ver := parts[1]
-		if cmpver.Compare(ver, maxVer) == 1 {
+		if cmpver.Compare(ver, maxVer) > 0 {
 			maxVer = ver
 		}
 	}
@@ -690,6 +704,37 @@ 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")
 }
@@ -848,7 +893,7 @@ func (up *Updater) installMSI(msi string) error {
 			break
 		}
 		up.Logf("Install attempt failed: %v", err)
-		uninstallVersion := version.Short()
+		uninstallVersion := up.currentVersion
 		if v := os.Getenv("TS_DEBUG_UNINSTALL_VERSION"); v != "" {
 			uninstallVersion = v
 		}
@@ -1299,12 +1344,8 @@ func requestedTailscaleVersion(ver, track string) (string, error) {
 // LatestTailscaleVersion returns the latest released version for the given
 // track from pkgs.tailscale.com.
 func LatestTailscaleVersion(track string) (string, error) {
-	if track == CurrentTrack {
-		if version.IsUnstableBuild() {
-			track = UnstableTrack
-		} else {
-			track = StableTrack
-		}
+	if track == "" {
+		track = CurrentTrack
 	}
 
 	latest, err := latestPackages(track)

clientupdate/clientupdate_test.go

@@ -846,3 +846,107 @@ func TestParseUnraidPluginVersion(t *testing.T) {
 		})
 	}
 }
+
+func TestConfirm(t *testing.T) {
+	curTrack := CurrentTrack
+	defer func() { CurrentTrack = curTrack }()
+
+	tests := []struct {
+		desc      string
+		fromTrack string
+		toTrack   string
+		fromVer   string
+		toVer     string
+		confirm   func(string) bool
+		want      bool
+	}{
+		{
+			desc:      "on latest stable",
+			fromTrack: StableTrack,
+			toTrack:   StableTrack,
+			fromVer:   "1.66.0",
+			toVer:     "1.66.0",
+			want:      false,
+		},
+		{
+			desc:      "stable upgrade",
+			fromTrack: StableTrack,
+			toTrack:   StableTrack,
+			fromVer:   "1.66.0",
+			toVer:     "1.68.0",
+			want:      true,
+		},
+		{
+			desc:      "unstable upgrade",
+			fromTrack: UnstableTrack,
+			toTrack:   UnstableTrack,
+			fromVer:   "1.67.1",
+			toVer:     "1.67.2",
+			want:      true,
+		},
+		{
+			desc:      "from stable to unstable",
+			fromTrack: StableTrack,
+			toTrack:   UnstableTrack,
+			fromVer:   "1.66.0",
+			toVer:     "1.67.1",
+			want:      true,
+		},
+		{
+			desc:      "from unstable to stable",
+			fromTrack: UnstableTrack,
+			toTrack:   StableTrack,
+			fromVer:   "1.67.1",
+			toVer:     "1.66.0",
+			want:      true,
+		},
+		{
+			desc:      "confirm callback rejects",
+			fromTrack: StableTrack,
+			toTrack:   StableTrack,
+			fromVer:   "1.66.0",
+			toVer:     "1.66.1",
+			confirm: func(string) bool {
+				return false
+			},
+			want: false,
+		},
+		{
+			desc:      "confirm callback allows",
+			fromTrack: StableTrack,
+			toTrack:   StableTrack,
+			fromVer:   "1.66.0",
+			toVer:     "1.66.1",
+			confirm: func(string) bool {
+				return true
+			},
+			want: true,
+		},
+		{
+			desc:      "downgrade",
+			fromTrack: StableTrack,
+			toTrack:   StableTrack,
+			fromVer:   "1.66.1",
+			toVer:     "1.66.0",
+			want:      false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.desc, func(t *testing.T) {
+			CurrentTrack = tt.fromTrack
+			up := Updater{
+				currentVersion: tt.fromVer,
+				Arguments: Arguments{
+					Track:   tt.toTrack,
+					Confirm: tt.confirm,
+					Logf:    t.Logf,
+				},
+			}
+
+			if got := up.confirm(tt.toVer); got != tt.want {
+				t.Errorf("got %v, want %v", got, tt.want)
+			}
+		})
+	}
+}

cmd/containerboot/kube.go

@@ -19,22 +19,20 @@ import (
 	"tailscale.com/tailcfg"
 )
 
-// storeDeviceInfo writes deviceID into the "device_id" data field of the kube
-// secret secretName.
-func storeDeviceInfo(ctx context.Context, secretName string, deviceID tailcfg.StableNodeID, fqdn string, addresses []netip.Prefix) error {
-	// First check if the secret exists at all. Even if running on
-	// kubernetes, we do not necessarily store state in a k8s secret.
-	if _, err := kc.GetSecret(ctx, secretName); err != nil {
-		if s, ok := err.(*kube.Status); ok {
-			if s.Code >= 400 && s.Code <= 499 {
-				// Assume the secret doesn't exist, or we don't have
-				// permission to access it.
-				return nil
-			}
-		}
-		return err
+// storeDeviceID writes deviceID to 'device_id' data field of the named
+// Kubernetes Secret.
+func storeDeviceID(ctx context.Context, secretName string, deviceID tailcfg.StableNodeID) error {
+	s := &kube.Secret{
+		Data: map[string][]byte{
+			"device_id": []byte(deviceID),
+		},
 	}
+	return kc.StrategicMergePatchSecret(ctx, secretName, s, "tailscale-container")
+}
 
+// storeDeviceEndpoints writes device's tailnet IPs and MagicDNS name to fields
+// 'device_ips', 'device_fqdn' of the named Kubernetes Secret.
+func storeDeviceEndpoints(ctx context.Context, secretName string, fqdn string, addresses []netip.Prefix) error {
 	var ips []string
 	for _, addr := range addresses {
 		ips = append(ips, addr.Addr().String())
@@ -44,14 +42,13 @@ func storeDeviceInfo(ctx context.Context, secretName string, deviceID tailcfg.St
 		return err
 	}
 
-	m := &kube.Secret{
+	s := &kube.Secret{
 		Data: map[string][]byte{
-			"device_id":   []byte(deviceID),
 			"device_fqdn": []byte(fqdn),
 			"device_ips":  deviceIPs,
 		},
 	}
-	return kc.StrategicMergePatchSecret(ctx, secretName, m, "tailscale-container")
+	return kc.StrategicMergePatchSecret(ctx, secretName, s, "tailscale-container")
 }
 
 // deleteAuthKey deletes the 'authkey' field of the given kube
@@ -138,9 +135,9 @@ 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

@@ -52,13 +52,20 @@
 //     ${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`.
 //     The config file contents are currently read once on container start.
 //     NB: This env var is currently experimental and the logic will likely change!
+//     TS_EXPERIMENTAL_ENABLE_FORWARDING_OPTIMIZATIONS: set to true to
+//     autoconfigure the default network interface for optimal performance for
+//     Tailscale subnet router/exit node.
+//     https://tailscale.com/kb/1320/performance-best-practices#linux-optimizations-for-subnet-routers-and-exit-nodes
+//     NB: This env var is currently experimental and the logic will likely change!
 //   - EXPERIMENTAL_ALLOW_PROXYING_CLUSTER_TRAFFIC_VIA_INGRESS: if set to true
 //     and if this containerboot instance is an L7 ingress proxy (created by
 //     the Kubernetes operator), set up rules to allow proxying cluster traffic,
@@ -92,6 +99,7 @@ import (
 	"os"
 	"os/exec"
 	"os/signal"
+	"path"
 	"path/filepath"
 	"reflect"
 	"slices"
@@ -107,6 +115,7 @@ 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"
@@ -145,9 +154,10 @@ 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", ""),
+		EnableForwardingOptimizations:         defaultBool("TS_EXPERIMENTAL_ENABLE_FORWARDING_OPTIMIZATIONS", false),
 	}
 
 	if err := cfg.validate(); err != nil {
@@ -195,6 +205,12 @@ func main() {
 	}
 	defer killTailscaled()
 
+	if cfg.EnableForwardingOptimizations {
+		if err := client.SetUDPGROForwarding(bootCtx); err != nil {
+			log.Printf("[unexpected] error enabling UDP GRO forwarding: %v", err)
+		}
+	}
+
 	w, err := client.WatchIPNBus(bootCtx, ipn.NotifyInitialNetMap|ipn.NotifyInitialPrefs|ipn.NotifyInitialState)
 	if err != nil {
 		log.Fatalf("failed to watch tailscaled for updates: %v", err)
@@ -305,7 +321,7 @@ authLoop:
 		}
 	}
 
-	if cfg.InKubernetes && cfg.KubeSecret != "" && cfg.KubernetesCanPatch && isTwoStepConfigAuthOnce(cfg) {
+	if hasKubeStateStore(cfg) && isTwoStepConfigAuthOnce(cfg) {
 		// We were told to only auth once, so any secret-bound
 		// authkey is no longer needed. We don't strictly need to
 		// wipe it, but it's good hygiene.
@@ -321,11 +337,10 @@ authLoop:
 	}
 
 	var (
-		wantProxy         = cfg.ProxyTargetIP != "" || cfg.ProxyTargetDNSName != "" || cfg.TailnetTargetIP != "" || cfg.TailnetTargetFQDN != "" || cfg.AllowProxyingClusterTrafficViaIngress
-		wantDeviceInfo    = cfg.InKubernetes && cfg.KubeSecret != "" && cfg.KubernetesCanPatch
-		startupTasksDone  = false
-		currentIPs        deephash.Sum // tailscale IPs assigned to device
-		currentDeviceInfo deephash.Sum // device ID and fqdn
+		startupTasksDone       = false
+		currentIPs             deephash.Sum // tailscale IPs assigned to device
+		currentDeviceID        deephash.Sum // device ID
+		currentDeviceEndpoints deephash.Sum // device FQDN and IPs
 
 		currentEgressIPs deephash.Sum
 
@@ -339,7 +354,7 @@ authLoop:
 		go watchServeConfigChanges(ctx, cfg.ServeConfigPath, certDomainChanged, certDomain, client)
 	}
 	var nfr linuxfw.NetfilterRunner
-	if wantProxy {
+	if isL3Proxy(cfg) {
 		nfr, err = newNetfilterRunner(log.Printf)
 		if err != nil {
 			log.Fatalf("error creating new netfilter runner: %v", err)
@@ -424,6 +439,20 @@ runLoop:
 				newCurrentIPs := deephash.Hash(&addrs)
 				ipsHaveChanged := newCurrentIPs != currentIPs
 
+				// Store device ID in a Kubernetes Secret before
+				// setting up any routing rules. This ensures
+				// that, for containerboot instances that are
+				// Kubernetes operator proxies, the operator is
+				// able to retrieve the device ID from the
+				// Kubernetes Secret to clean up tailnet nodes
+				// for proxies whose route setup continuously
+				// fails.
+				deviceID := n.NetMap.SelfNode.StableID()
+				if hasKubeStateStore(cfg) && deephash.Update(&currentDeviceID, &deviceID) {
+					if err := storeDeviceID(ctx, cfg.KubeSecret, n.NetMap.SelfNode.StableID()); err != nil {
+						log.Fatalf("storing device ID in Kubernetes Secret: %v", err)
+					}
+				}
 				if cfg.TailnetTargetFQDN != "" {
 					var (
 						egressAddrs          []netip.Prefix
@@ -517,15 +546,36 @@ runLoop:
 				}
 				currentIPs = newCurrentIPs
 
-				deviceInfo := []any{n.NetMap.SelfNode.StableID(), n.NetMap.SelfNode.Name()}
-				if cfg.InKubernetes && cfg.KubernetesCanPatch && cfg.KubeSecret != "" && deephash.Update(&currentDeviceInfo, &deviceInfo) {
-					if err := storeDeviceInfo(ctx, cfg.KubeSecret, n.NetMap.SelfNode.StableID(), n.NetMap.SelfNode.Name(), n.NetMap.SelfNode.Addresses().AsSlice()); err != nil {
-						log.Fatalf("storing device ID in kube secret: %v", err)
+				// Only store device FQDN and IP addresses to
+				// Kubernetes Secret when any required proxy
+				// route setup has succeeded. IPs and FQDN are
+				// read from the Secret by the Tailscale
+				// Kubernetes operator and, for some proxy
+				// types, such as Tailscale Ingress, advertized
+				// on the Ingress status. Writing them to the
+				// Secret only after the proxy routing has been
+				// set up ensures that the operator does not
+				// advertize endpoints of broken proxies.
+				// TODO (irbekrm): instead of using the IP and FQDN, have some other mechanism for the proxy signal that it is 'Ready'.
+				deviceEndpoints := []any{n.NetMap.SelfNode.Name(), n.NetMap.SelfNode.Addresses()}
+				if hasKubeStateStore(cfg) && deephash.Update(&currentDeviceEndpoints, &deviceEndpoints) {
+					if err := storeDeviceEndpoints(ctx, cfg.KubeSecret, n.NetMap.SelfNode.Name(), n.NetMap.SelfNode.Addresses().AsSlice()); err != nil {
+						log.Fatalf("storing device IPs and FQDN in Kubernetes Secret: %v", err)
 					}
 				}
 			}
 			if !startupTasksDone {
-				if (!wantProxy || currentIPs != deephash.Sum{}) && (!wantDeviceInfo || currentDeviceInfo != deephash.Sum{}) {
+				// For containerboot instances that act as TCP
+				// proxies (proxying traffic to an endpoint
+				// passed via one of the env vars that
+				// containerbot reads) and store state in a
+				// Kubernetes Secret, we consider startup tasks
+				// done at the point when device info has been
+				// successfully stored to state Secret.
+				// For all other containerboot instances, if we
+				// just get to this point the startup tasks can
+				// be considered done.
+				if !isL3Proxy(cfg) || !hasKubeStateStore(cfg) || (currentDeviceEndpoints != deephash.Sum{} && currentDeviceID != deephash.Sum{}) {
 					// This log message is used in tests to detect when all
 					// post-auth configuration is done.
 					log.Println("Startup complete, waiting for shutdown signal")
@@ -957,16 +1007,23 @@ 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)
 	}
@@ -1069,22 +1126,23 @@ type settings struct {
 	// TailnetTargetFQDN is an MagicDNS name to which all incoming
 	// non-Tailscale traffic should be proxied. This must be a full Tailnet
 	// node FQDN.
-	TailnetTargetFQDN        string
-	ServeConfigPath          string
-	DaemonExtraArgs          string
-	ExtraArgs                string
-	InKubernetes             bool
-	UserspaceMode            bool
-	StateDir                 string
-	AcceptDNS                *bool
-	KubeSecret               string
-	SOCKSProxyAddr           string
-	HTTPProxyAddr            string
-	Socket                   string
-	AuthOnce                 bool
-	Root                     string
-	KubernetesCanPatch       bool
-	TailscaledConfigFilePath string
+	TailnetTargetFQDN             string
+	ServeConfigPath               string
+	DaemonExtraArgs               string
+	ExtraArgs                     string
+	InKubernetes                  bool
+	UserspaceMode                 bool
+	StateDir                      string
+	AcceptDNS                     *bool
+	KubeSecret                    string
+	SOCKSProxyAddr                string
+	HTTPProxyAddr                 string
+	Socket                        string
+	AuthOnce                      bool
+	Root                          string
+	KubernetesCanPatch            bool
+	TailscaledConfigFilePath      string
+	EnableForwardingOptimizations bool
 	// If set to true and, if this containerboot instance is a Kubernetes
 	// ingress proxy, set up rules to forward incoming cluster traffic to be
 	// forwarded to the ingress target in cluster.
@@ -1097,6 +1155,13 @@ 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 +1185,7 @@ 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")
@@ -1131,6 +1196,9 @@ func (s *settings) validate() error {
 	if s.AllowProxyingClusterTrafficViaIngress && s.PodIP == "" {
 		return errors.New("EXPERIMENTAL_ALLOW_PROXYING_CLUSTER_TRAFFIC_VIA_INGRESS is set but POD_IP is not set")
 	}
+	if s.EnableForwardingOptimizations && s.UserspaceMode {
+		return errors.New("TS_EXPERIMENTAL_ENABLE_FORWARDING_OPTIMIZATIONS is not supported in userspace mode")
+	}
 	return nil
 }
 
@@ -1252,3 +1320,55 @@ func isTwoStepConfigAlwaysAuth(cfg *settings) bool {
 func isOneStepConfig(cfg *settings) bool {
 	return cfg.TailscaledConfigFilePath != ""
 }
+
+// isL3Proxy returns true if the Tailscale node needs to be configured to act
+// as an L3 proxy, proxying to an endpoint provided via one of the config env
+// vars.
+func isL3Proxy(cfg *settings) bool {
+	return cfg.ProxyTargetIP != "" || cfg.ProxyTargetDNSName != "" || cfg.TailnetTargetIP != "" || cfg.TailnetTargetFQDN != "" || cfg.AllowProxyingClusterTrafficViaIngress
+}
+
+// hasKubeStateStore returns true if the state must be stored in a Kubernetes
+// Secret.
+func hasKubeStateStore(cfg *settings) bool {
+	return cfg.InKubernetes && cfg.KubernetesCanPatch && cfg.KubeSecret != ""
+}
+
+// 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/README.md

@@ -0,0 +1,74 @@
+# DERP
+
+This is the code for the [Tailscale DERP server](https://tailscale.com/kb/1232/derp-servers).
+
+In general, you should not need to nor want to run this code. The overwhelming majority of Tailscale users (both individuals and companies) do not.
+
+In the happy path, Tailscale establishes direct connections between peers and
+data plane traffic flows directly between them, without using DERP for more than
+acting as a low bandwidth side channel to bootstrap the NAT traversal. If you
+find yourself wanting DERP for more bandwidth, the real problem is usually the
+network configuration of your Tailscale node(s), making sure that Tailscale can
+get direction connections via some mechanism.
+
+But if you've decided or been advised to run your own `derper`, then read on.
+
+## Caveats
+
+* Node sharing and other cross-Tailnet features don't work when using custom
+  DERP servers.
+
+* DERP servers only see encrypted WireGuard packets and thus are not useful for
+  network-level debugging.
+
+* The Tailscale control plane does certain geo-level steering features and
+  optimizations that are not available when using custom DERP servers.
+
+## Guide to running `cmd/derper`
+
+* You must build and update the `cmd/derper` binary yourself. There are no
+  packages. Use `go install tailscale.com/cmd/derper@latest` with the latest
+  version of Go.
+
+* The DERP protocol does a protocol switch inside TLS from HTTP to a custom
+  bidirectional binary protocol. It is thus incompatible with many HTTP proxies.
+  Do not put `derper` behind another HTTP proxy.
+
+* The `tailscaled` client does its own selection of the fastest/nearest DERP
+  server based on latency measurements. Do not put `derper` behind a global load
+  balancer.
+
+* DERP servers should ideally have both a static IPv4 and static IPv6 address.
+Both of those should be listed in the DERP map so the client doesn't need to
+rely on its DNS which might be broken and dependent on DERP to get back up.
+
+* A DERP server should not share an IP address with any other DERP server.
+
+* Avoid having multiple DERP nodes in a region. If you must, they all need to be
+  meshed with each other and monitored. Having two one-node "regions" in the
+  same datacenter is usually easier and more reliable than meshing, at the cost
+  of more required connections from clients in some cases. If your clients
+  aren't mobile (battery constrained), one node regions are definitely
+  preferred. If you really need multiple nodes in a region for HA reasons, two
+  is sufficient.
+
+* Monitor your DERP servers with [`cmd/derpprobe`](../derpprobe/).
+
+* If using `--verify-clients`, a `tailscaled` must be running alongside the
+  `derper`.
+
+* If using `--verify-clients`, a `tailscaled` must also be running alongside
+  your `derpprobe`, and `derpprobe` needs to use `--derp-map=local`.
+
+* The firewall on the `derper` should permit TCP ports 80 and 443 and UDP port
+  3478.
+
+* Only LetsEncrypt certs are rotated automatically. Other cert updates require a
+  restart.
+
+* Don't use a firewall in front of `derper` that suppresses `RST`s upon
+  receiving traffic to a dead or unknown connection.
+
+* Don't rate-limit UDP STUN packets.
+
+* Don't rate-limit outbound TCP traffic (only inbound).

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

@@ -4,15 +4,19 @@
 package main
 
 import (
+	"bytes"
 	"encoding/json"
+	"io"
 	"net"
 	"net/http"
 	"net/http/httptest"
+	"net/netip"
 	"net/url"
 	"reflect"
 	"testing"
 
 	"tailscale.com/tstest"
+	"tailscale.com/tstest/nettest"
 )
 
 func BenchmarkHandleBootstrapDNS(b *testing.B) {
@@ -37,7 +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()
@@ -47,14 +51,17 @@ 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) {
+	nettest.SkipIfNoNetwork(t)
+
 	const published = "login.tailscale.com"
 	const unpublished = "log.tailscale.io"
 
@@ -104,15 +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) {
@@ -122,8 +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)
@@ -138,7 +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)
 		}
@@ -163,3 +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

@@ -46,7 +46,7 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
    L    github.com/vishvananda/netns                                 from github.com/tailscale/netlink+
         github.com/x448/float16                                      from github.com/fxamacker/cbor/v2
      💣 go4.org/mem                                                  from tailscale.com/client/tailscale+
-        go4.org/netipx                                               from tailscale.com/net/tsaddr+
+        go4.org/netipx                                               from tailscale.com/net/tsaddr
    W 💣 golang.zx2c4.com/wireguard/windows/tunnel/winipcfg           from tailscale.com/net/netmon+
         google.golang.org/protobuf/encoding/protodelim               from github.com/prometheus/common/expfmt
         google.golang.org/protobuf/encoding/prototext                from github.com/prometheus/common/expfmt+
@@ -95,14 +95,12 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         tailscale.com/ipn/ipnstate                                   from tailscale.com/client/tailscale+
         tailscale.com/metrics                                        from tailscale.com/cmd/derper+
         tailscale.com/net/dnscache                                   from tailscale.com/derp/derphttp
-        tailscale.com/net/flowtrack                                  from tailscale.com/net/packet+
         tailscale.com/net/ktimeout                                   from tailscale.com/cmd/derper
         tailscale.com/net/netaddr                                    from tailscale.com/ipn+
         tailscale.com/net/netknob                                    from tailscale.com/net/netns
      💣 tailscale.com/net/netmon                                     from tailscale.com/derp/derphttp+
         tailscale.com/net/netns                                      from tailscale.com/derp/derphttp
         tailscale.com/net/netutil                                    from tailscale.com/client/tailscale
-        tailscale.com/net/packet                                     from tailscale.com/wgengine/filter
         tailscale.com/net/sockstats                                  from tailscale.com/derp/derphttp
         tailscale.com/net/stun                                       from tailscale.com/net/stunserver
         tailscale.com/net/stunserver                                 from tailscale.com/cmd/derper
@@ -119,13 +117,13 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
    W    tailscale.com/tsconst                                        from tailscale.com/net/netmon
         tailscale.com/tstime                                         from tailscale.com/derp+
         tailscale.com/tstime/mono                                    from tailscale.com/tstime/rate
-        tailscale.com/tstime/rate                                    from tailscale.com/derp+
+        tailscale.com/tstime/rate                                    from tailscale.com/derp
         tailscale.com/tsweb                                          from tailscale.com/cmd/derper
         tailscale.com/tsweb/promvarz                                 from tailscale.com/tsweb
         tailscale.com/tsweb/varz                                     from tailscale.com/tsweb+
         tailscale.com/types/dnstype                                  from tailscale.com/tailcfg
         tailscale.com/types/empty                                    from tailscale.com/ipn
-        tailscale.com/types/ipproto                                  from tailscale.com/net/flowtrack+
+        tailscale.com/types/ipproto                                  from tailscale.com/tailcfg+
         tailscale.com/types/key                                      from tailscale.com/client/tailscale+
         tailscale.com/types/lazy                                     from tailscale.com/version+
         tailscale.com/types/logger                                   from tailscale.com/cmd/derper+
@@ -157,10 +155,10 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         tailscale.com/util/syspolicy                                 from tailscale.com/ipn
         tailscale.com/util/vizerror                                  from tailscale.com/tailcfg+
    W 💣 tailscale.com/util/winutil                                   from tailscale.com/hostinfo+
-   W 💣 tailscale.com/util/winutil/winenv                            from tailscale.com/hostinfo
+   W 💣 tailscale.com/util/winutil/winenv                            from tailscale.com/hostinfo+
         tailscale.com/version                                        from tailscale.com/derp+
         tailscale.com/version/distro                                 from tailscale.com/envknob+
-        tailscale.com/wgengine/filter                                from tailscale.com/types/netmap
+        tailscale.com/wgengine/filter/filtertype                     from tailscale.com/types/netmap
         golang.org/x/crypto/acme                                     from golang.org/x/crypto/acme/autocert
         golang.org/x/crypto/acme/autocert                            from tailscale.com/cmd/derper
         golang.org/x/crypto/argon2                                   from tailscale.com/tka
@@ -175,6 +173,7 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         golang.org/x/crypto/nacl/box                                 from tailscale.com/types/key
         golang.org/x/crypto/nacl/secretbox                           from golang.org/x/crypto/nacl/box
         golang.org/x/crypto/salsa20/salsa                            from golang.org/x/crypto/nacl/box+
+   W    golang.org/x/exp/constraints                                 from tailscale.com/util/winutil
    L    golang.org/x/net/bpf                                         from github.com/mdlayher/netlink+
         golang.org/x/net/dns/dnsmessage                              from net+
         golang.org/x/net/http/httpguts                               from net/http
@@ -235,7 +234,7 @@ 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 +252,7 @@ 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 +276,7 @@ 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 +284,6 @@ 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")
@@ -191,7 +191,12 @@ func main() {
 			http.Error(w, "derp server disabled", http.StatusNotFound)
 		}))
 	}
-	mux.HandleFunc("/derp/probe", probeHandler)
+
+	// These two endpoints are the same. Different versions of the clients
+	// have assumes different paths over time so we support both.
+	mux.HandleFunc("/derp/probe", derphttp.ProbeHandler)
+	mux.HandleFunc("/derp/latency-check", derphttp.ProbeHandler)
+
 	go refreshBootstrapDNSLoop()
 	mux.HandleFunc("/bootstrap-dns", tsweb.BrowserHeaderHandlerFunc(handleBootstrapDNS))
 	mux.Handle("/", http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
@@ -370,17 +375,6 @@ func isChallengeChar(c rune) bool {
 		c == '.' || c == '-' || c == '_'
 }
 
-// probeHandler is the endpoint that js/wasm clients hit to measure
-// DERP latency, since they can't do UDP STUN queries.
-func probeHandler(w http.ResponseWriter, r *http.Request) {
-	switch r.Method {
-	case "HEAD", "GET":
-		w.Header().Set("Access-Control-Allow-Origin", "*")
-	default:
-		http.Error(w, "bogus probe method", http.StatusMethodNotAllowed)
-	}
-}
-
 var validProdHostname = regexp.MustCompile(`^derp([^.]*)\.tailscale\.com\.?$`)
 
 func prodAutocertHostPolicy(_ context.Context, host string) error {

cmd/derper/derper_test.go

@@ -99,10 +99,13 @@ 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",
 			"gvisor.dev/gvisor/pkg/tcpip/header": "https://github.com/tailscale/tailscale/issues/9756",
+			"tailscale.com/net/packet":           "not needed in derper",
+			"github.com/gaissmai/bart":           "not needed in derper",
 		},
 	}.Check(t)
 }

cmd/derpprobe/derpprobe.go

@@ -20,7 +20,7 @@ import (
 )
 
 var (
-	derpMapURL   = flag.String("derp-map", "https://login.tailscale.com/derpmap/default", "URL to DERP map (https:// or file://)")
+	derpMapURL   = flag.String("derp-map", "https://login.tailscale.com/derpmap/default", "URL to DERP map (https:// or file://) or 'local' to use the local tailscaled's DERP map")
 	versionFlag  = flag.Bool("version", false, "print version and exit")
 	listen       = flag.String("listen", ":8030", "HTTP listen address")
 	probeOnce    = flag.Bool("once", false, "probe once and print results, then exit; ignores the listen flag")

cmd/k8s-nameserver/main.go

@@ -153,11 +153,36 @@ func (n *nameserver) handleFunc() func(w dns.ResponseWriter, r *dns.Msg) {
 				m.Answer = append(m.Answer, rr)
 			}
 		case dns.TypeAAAA:
-			// TODO (irbekrm): implement IPv6 support.
-			// Kubernetes distributions that I am most familiar with
-			// default to IPv4 for Pod CIDR ranges and often many cases don't
-			// support IPv6 at all, so this should not be crucial for now.
-			fallthrough
+			// TODO (irbekrm): add IPv6 support.
+			// The nameserver currently does not support IPv6
+			// (records are not being created for IPv6 Pod addresses).
+			// However, we can expect that some callers will
+			// nevertheless send AAAA queries.
+			// We have to return NOERROR if a query is received for
+			// an AAAA record for a DNS name that we have an A
+			// record for- else the caller might not follow with an
+			// A record query.
+			// https://github.com/tailscale/tailscale/issues/12321
+			// https://datatracker.ietf.org/doc/html/rfc4074
+			q := r.Question[0].Name
+			fqdn, err := dnsname.ToFQDN(q)
+			if err != nil {
+				m = r.SetRcodeFormatError(r)
+				return
+			}
+			// The only supported use of this nameserver is as a
+			// single source of truth for MagicDNS names by
+			// non-tailnet Kubernetes workloads.
+			m.Authoritative = true
+			ips := n.lookupIP4(fqdn)
+			if len(ips) == 0 {
+				// As we are the authoritative nameserver for MagicDNS
+				// names, if we do not have a record for this MagicDNS
+				// name, it does not exist.
+				m = m.SetRcode(r, dns.RcodeNameError)
+				return
+			}
+			m.SetRcode(r, dns.RcodeSuccess)
 		default:
 			log.Printf("[unexpected] nameserver received a query for an unsupported record type: %s", r.Question[0].String())
 			m.SetRcode(r, dns.RcodeNotImplemented)

cmd/k8s-nameserver/main_test.go

@@ -79,7 +79,7 @@ func TestNameserver(t *testing.T) {
 				}},
 		},
 		{
-			name: "AAAA record query",
+			name: "AAAA record query, A record exists",
 			ip4:  map[dnsname.FQDN][]net.IP{dnsname.FQDN("foo.bar.com."): {{1, 2, 3, 4}}},
 			query: &dns.Msg{
 				Question: []dns.Question{{Name: "foo.bar.com", Qtype: dns.TypeAAAA}},
@@ -88,26 +88,28 @@ func TestNameserver(t *testing.T) {
 			wantResp: &dns.Msg{
 				Question: []dns.Question{{Name: "foo.bar.com", Qtype: dns.TypeAAAA}},
 				MsgHdr: dns.MsgHdr{
-					Id:       1,
-					Rcode:    dns.RcodeNotImplemented,
-					Response: true,
-					Opcode:   dns.OpcodeQuery,
+					Id:            1,
+					Rcode:         dns.RcodeSuccess,
+					Response:      true,
+					Opcode:        dns.OpcodeQuery,
+					Authoritative: true,
 				}},
 		},
 		{
-			name: "AAAA record query",
+			name: "AAAA record query, A record does not exist",
 			ip4:  map[dnsname.FQDN][]net.IP{dnsname.FQDN("foo.bar.com."): {{1, 2, 3, 4}}},
 			query: &dns.Msg{
-				Question: []dns.Question{{Name: "foo.bar.com", Qtype: dns.TypeAAAA}},
+				Question: []dns.Question{{Name: "baz.bar.com", Qtype: dns.TypeAAAA}},
 				MsgHdr:   dns.MsgHdr{Id: 1},
 			},
 			wantResp: &dns.Msg{
-				Question: []dns.Question{{Name: "foo.bar.com", Qtype: dns.TypeAAAA}},
+				Question: []dns.Question{{Name: "baz.bar.com", Qtype: dns.TypeAAAA}},
 				MsgHdr: dns.MsgHdr{
-					Id:       1,
-					Rcode:    dns.RcodeNotImplemented,
-					Response: true,
-					Opcode:   dns.OpcodeQuery,
+					Id:            1,
+					Rcode:         dns.RcodeNameError,
+					Response:      true,
+					Opcode:        dns.OpcodeQuery,
+					Authoritative: true,
 				}},
 		},
 		{

cmd/k8s-operator/connector.go

@@ -33,11 +33,8 @@ import (
 
 const (
 	reasonConnectorCreationFailed = "ConnectorCreationFailed"
-
-	reasonConnectorCreated           = "ConnectorCreated"
-	reasonConnectorCleanupFailed     = "ConnectorCleanupFailed"
-	reasonConnectorCleanupInProgress = "ConnectorCleanupInProgress"
-	reasonConnectorInvalid           = "ConnectorInvalid"
+	reasonConnectorCreated        = "ConnectorCreated"
+	reasonConnectorInvalid        = "ConnectorInvalid"
 
 	messageConnectorCreationFailed = "Failed creating Connector: %v"
 	messageConnectorInvalid        = "Connector is invalid: %v"
@@ -108,7 +105,7 @@ func (a *ConnectorReconciler) Reconcile(ctx context.Context, req reconcile.Reque
 	}
 
 	oldCnStatus := cn.Status.DeepCopy()
-	setStatus := func(cn *tsapi.Connector, conditionType tsapi.ConnectorConditionType, status metav1.ConditionStatus, reason, message string) (reconcile.Result, error) {
+	setStatus := func(cn *tsapi.Connector, _ tsapi.ConditionType, status metav1.ConditionStatus, reason, message string) (reconcile.Result, error) {
 		tsoperator.SetConnectorCondition(cn, tsapi.ConnectorReady, status, reason, message, cn.Generation, a.clock, logger)
 		if !apiequality.Semantic.DeepEqual(oldCnStatus, cn.Status) {
 			// An error encountered here should get returned by the Reconcile function.
@@ -184,7 +181,7 @@ func (a *ConnectorReconciler) maybeProvisionConnector(ctx context.Context, logge
 		Connector: &connector{
 			isExitNode: cn.Spec.ExitNode,
 		},
-		ProxyClass: proxyClass,
+		ProxyClassName: proxyClass,
 	}
 
 	if cn.Spec.SubnetRouter != nil && len(cn.Spec.SubnetRouter.AdvertiseRoutes) > 0 {
@@ -211,7 +208,27 @@ func (a *ConnectorReconciler) maybeProvisionConnector(ctx context.Context, logge
 	gaugeConnectorResources.Set(int64(connectors.Len()))
 
 	_, err := a.ssr.Provision(ctx, logger, sts)
-	return err
+	if err != nil {
+		return err
+	}
+
+	_, tsHost, ips, err := a.ssr.DeviceInfo(ctx, crl)
+	if err != nil {
+		return err
+	}
+
+	if tsHost == "" {
+		logger.Debugf("no Tailscale hostname known yet, waiting for connector pod to finish auth")
+		// No hostname yet. Wait for the connector pod to auth.
+		cn.Status.TailnetIPs = nil
+		cn.Status.Hostname = ""
+		return nil
+	}
+
+	cn.Status.TailnetIPs = ips
+	cn.Status.Hostname = tsHost
+
+	return nil
 }
 
 func (a *ConnectorReconciler) maybeCleanupConnector(ctx context.Context, logger *zap.SugaredLogger, cn *tsapi.Connector) (bool, error) {

cmd/k8s-operator/connector_test.go

@@ -17,6 +17,7 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client/fake"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
 	"tailscale.com/tstest"
+	"tailscale.com/util/mak"
 )
 
 func TestConnector(t *testing.T) {
@@ -29,7 +30,7 @@ func TestConnector(t *testing.T) {
 		},
 		TypeMeta: metav1.TypeMeta{
 			Kind:       tsapi.ConnectorKind,
-			APIVersion: "tailscale.io/v1alpha1",
+			APIVersion: "tailscale.com/v1alpha1",
 		},
 		Spec: tsapi.ConnectorSpec{
 			SubnetRouter: &tsapi.SubnetRouter{
@@ -74,9 +75,26 @@ func TestConnector(t *testing.T) {
 		isExitNode:   true,
 		subnetRoutes: "10.40.0.0/14",
 	}
-	expectEqual(t, fc, expectedSecret(t, opts), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, opts), removeHashAnnotation)
 
+	// Connector status should get updated with the IP/hostname info when available.
+	const hostname = "foo.tailnetxyz.ts.net"
+	mustUpdate(t, fc, "operator-ns", opts.secretName, func(secret *corev1.Secret) {
+		mak.Set(&secret.Data, "device_id", []byte("1234"))
+		mak.Set(&secret.Data, "device_fqdn", []byte(hostname))
+		mak.Set(&secret.Data, "device_ips", []byte(`["127.0.0.1", "::1"]`))
+	})
+	expectReconciled(t, cr, "", "test")
+	cn.Finalizers = append(cn.Finalizers, "tailscale.com/finalizer")
+	cn.Status.IsExitNode = cn.Spec.ExitNode
+	cn.Status.SubnetRoutes = cn.Spec.SubnetRouter.AdvertiseRoutes.Stringify()
+	cn.Status.Hostname = hostname
+	cn.Status.TailnetIPs = []string{"127.0.0.1", "::1"}
+	expectEqual(t, fc, cn, func(o *tsapi.Connector) {
+		o.Status.Conditions = nil
+	})
+
 	// Add another route to be advertised.
 	mustUpdate[tsapi.Connector](t, fc, "", "test", func(conn *tsapi.Connector) {
 		conn.Spec.SubnetRouter.AdvertiseRoutes = []tsapi.Route{"10.40.0.0/14", "10.44.0.0/20"}
@@ -152,7 +170,7 @@ func TestConnector(t *testing.T) {
 		subnetRoutes: "10.40.0.0/14",
 		hostname:     "test-connector",
 	}
-	expectEqual(t, fc, expectedSecret(t, opts), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, opts), removeHashAnnotation)
 
 	// Add an exit node.
@@ -237,7 +255,7 @@ func TestConnectorWithProxyClass(t *testing.T) {
 		isExitNode:   true,
 		subnetRoutes: "10.40.0.0/14",
 	}
-	expectEqual(t, fc, expectedSecret(t, opts), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, opts), removeHashAnnotation)
 
 	// 2. Update Connector to specify a ProxyClass. ProxyClass is not yet
@@ -254,9 +272,9 @@ func TestConnectorWithProxyClass(t *testing.T) {
 	// its resources.
 	mustUpdateStatus(t, fc, "", "custom-metadata", func(pc *tsapi.ProxyClass) {
 		pc.Status = tsapi.ProxyClassStatus{
-			Conditions: []tsapi.ConnectorCondition{{
+			Conditions: []metav1.Condition{{
 				Status:             metav1.ConditionTrue,
-				Type:               tsapi.ProxyClassready,
+				Type:               string(tsapi.ProxyClassready),
 				ObservedGeneration: pc.Generation,
 			}}}
 	})

cmd/k8s-operator/deploy/chart/templates/deployment.yaml

@@ -49,7 +49,7 @@ spec:
             {{- toYaml . | nindent 12 }}
           {{- end }}
           {{- $operatorTag:= printf ":%s" ( .Values.operatorConfig.image.tag | default .Chart.AppVersion )}}
-          image: {{ .Values.operatorConfig.image.repo }}{{- if .Values.operatorConfig.image.digest -}}{{ printf "@%s" .Values.operatorConfig.image.digest}}{{- else -}}{{ printf "%s" $operatorTag }}{{- end }}
+          image: {{ coalesce .Values.operatorConfig.image.repo .Values.operatorConfig.image.repository }}{{- if .Values.operatorConfig.image.digest -}}{{ printf "@%s" .Values.operatorConfig.image.digest}}{{- else -}}{{ printf "%s" $operatorTag }}{{- end }}
           imagePullPolicy: {{ .Values.operatorConfig.image.pullPolicy }}
           env:
             - name: OPERATOR_INITIAL_TAGS
@@ -70,7 +70,7 @@ spec:
               value: /oauth/client_secret
             {{- $proxyTag := printf ":%s" ( .Values.proxyConfig.image.tag | default .Chart.AppVersion )}}
             - name: PROXY_IMAGE
-              value: {{ .Values.proxyConfig.image.repo }}{{- if .Values.proxyConfig.image.digest -}}{{ printf "@%s" .Values.proxyConfig.image.digest}}{{- else -}}{{ printf "%s" $proxyTag }}{{- end }}
+              value: {{ coalesce .Values.proxyConfig.image.repo .Values.proxyConfig.image.repository }}{{- if .Values.proxyConfig.image.digest -}}{{ printf "@%s" .Values.proxyConfig.image.digest}}{{- else -}}{{ printf "%s" $proxyTag }}{{- end }}
             - name: PROXY_TAGS
               value: {{ .Values.proxyConfig.defaultTags }}
             - name: APISERVER_PROXY

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

@@ -23,7 +23,8 @@ operatorConfig:
     - "tag:k8s-operator"
 
   image:
-    repo: tailscale/k8s-operator
+    # Repository defaults to DockerHub, but images are also synced to ghcr.io/tailscale/k8s-operator.
+    repository: tailscale/k8s-operator
     # Digest will be prioritized over tag. If neither are set appVersion will be
     # used.
     tag: ""
@@ -51,9 +52,14 @@ 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
+    # Repository defaults to DockerHub, but images are also synced to ghcr.io/tailscale/tailscale.
+    repository: tailscale/tailscale
     # Digest will be prioritized over tag. If neither are set appVersion will be
     # used.
     tag: ""

cmd/k8s-operator/deploy/crds/tailscale.com_connectors.yaml

@@ -2,7 +2,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    controller-gen.kubebuilder.io/version: v0.13.0
+    controller-gen.kubebuilder.io/version: v0.15.1-0.20240618033008-7824932b0cab
   name: connectors.tailscale.com
 spec:
   group: tailscale.com
@@ -31,48 +31,95 @@ spec:
       name: v1alpha1
       schema:
         openAPIV3Schema:
-          description: 'Connector defines a Tailscale node that will be deployed in the cluster. The node can be configured to act as a Tailscale subnet router and/or a Tailscale exit node. Connector is a cluster-scoped resource. More info: https://tailscale.com/kb/1236/kubernetes-operator#deploying-exit-nodes-and-subnet-routers-on-kubernetes-using-connector-custom-resource'
+          description: |-
+            Connector defines a Tailscale node that will be deployed in the cluster. The
+            node can be configured to act as a Tailscale subnet router and/or a Tailscale
+            exit node.
+            Connector is a cluster-scoped resource.
+            More info:
+            https://tailscale.com/kb/1236/kubernetes-operator#deploying-exit-nodes-and-subnet-routers-on-kubernetes-using-connector-custom-resource
           type: object
           required:
             - spec
           properties:
             apiVersion:
-              description: 'APIVersion defines the versioned schema of this representation of an object. Servers should convert recognized schemas to the latest internal value, and may reject unrecognized values. More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources'
+              description: |-
+                APIVersion defines the versioned schema of this representation of an object.
+                Servers should convert recognized schemas to the latest internal value, and
+                may reject unrecognized values.
+                More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources
               type: string
             kind:
-              description: 'Kind is a string value representing the REST resource this object represents. Servers may infer this from the endpoint the client submits requests to. Cannot be updated. In CamelCase. More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds'
+              description: |-
+                Kind is a string value representing the REST resource this object represents.
+                Servers may infer this from the endpoint the client submits requests to.
+                Cannot be updated.
+                In CamelCase.
+                More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds
               type: string
             metadata:
               type: object
             spec:
-              description: 'ConnectorSpec describes the desired Tailscale component. More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#spec-and-status'
+              description: |-
+                ConnectorSpec describes the desired Tailscale component.
+                More info:
+                https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#spec-and-status
               type: object
               properties:
                 exitNode:
-                  description: ExitNode defines whether the Connector node should act as a Tailscale exit node. Defaults to false. https://tailscale.com/kb/1103/exit-nodes
+                  description: |-
+                    ExitNode defines whether the Connector node should act as a
+                    Tailscale exit node. Defaults to false.
+                    https://tailscale.com/kb/1103/exit-nodes
                   type: boolean
                 hostname:
-                  description: Hostname is the tailnet hostname that should be assigned to the Connector node. If unset, hostname defaults to <connector name>-connector. Hostname can contain lower case letters, numbers and dashes, it must not start or end with a dash and must be between 2 and 63 characters long.
+                  description: |-
+                    Hostname is the tailnet hostname that should be assigned to the
+                    Connector node. If unset, hostname defaults to <connector
+                    name>-connector. Hostname can contain lower case letters, numbers and
+                    dashes, it must not start or end with a dash and must be between 2
+                    and 63 characters long.
                   type: string
                   pattern: ^[a-z0-9][a-z0-9-]{0,61}[a-z0-9]$
                 proxyClass:
-                  description: ProxyClass is the name of the ProxyClass custom resource that contains configuration options that should be applied to the resources created for this Connector. If unset, the operator will create resources with the default configuration.
+                  description: |-
+                    ProxyClass is the name of the ProxyClass custom resource that
+                    contains configuration options that should be applied to the
+                    resources created for this Connector. If unset, the operator will
+                    create resources with the default configuration.
                   type: string
                 subnetRouter:
-                  description: SubnetRouter defines subnet routes that the Connector node should expose to tailnet. If unset, none are exposed. https://tailscale.com/kb/1019/subnets/
+                  description: |-
+                    SubnetRouter defines subnet routes that the Connector node should
+                    expose to tailnet. If unset, none are exposed.
+                    https://tailscale.com/kb/1019/subnets/
                   type: object
                   required:
                     - advertiseRoutes
                   properties:
                     advertiseRoutes:
-                      description: AdvertiseRoutes refer to CIDRs that the subnet router should make available. Route values must be strings that represent a valid IPv4 or IPv6 CIDR range. Values can be Tailscale 4via6 subnet routes. https://tailscale.com/kb/1201/4via6-subnets/
+                      description: |-
+                        AdvertiseRoutes refer to CIDRs that the subnet router should make
+                        available. Route values must be strings that represent a valid IPv4
+                        or IPv6 CIDR range. Values can be Tailscale 4via6 subnet routes.
+                        https://tailscale.com/kb/1201/4via6-subnets/
                       type: array
                       minItems: 1
                       items:
                         type: string
                         format: cidr
                 tags:
-                  description: Tags that the Tailscale node will be tagged with. Defaults to [tag:k8s]. To autoapprove the subnet routes or exit node defined by a Connector, you can configure Tailscale ACLs to give these tags the necessary permissions. See https://tailscale.com/kb/1018/acls/#auto-approvers-for-routes-and-exit-nodes. If you specify custom tags here, you must also make the operator an owner of these tags. See  https://tailscale.com/kb/1236/kubernetes-operator/#setting-up-the-kubernetes-operator. Tags cannot be changed once a Connector node has been created. Tag values must be in form ^tag:[a-zA-Z][a-zA-Z0-9-]*$.
+                  description: |-
+                    Tags that the Tailscale node will be tagged with.
+                    Defaults to [tag:k8s].
+                    To autoapprove the subnet routes or exit node defined by a Connector,
+                    you can configure Tailscale ACLs to give these tags the necessary
+                    permissions.
+                    See https://tailscale.com/kb/1018/acls/#auto-approvers-for-routes-and-exit-nodes.
+                    If you specify custom tags here, you must also make the operator an owner of these tags.
+                    See  https://tailscale.com/kb/1236/kubernetes-operator/#setting-up-the-kubernetes-operator.
+                    Tags cannot be changed once a Connector node has been created.
+                    Tag values must be in form ^tag:[a-zA-Z][a-zA-Z0-9-]*$.
                   type: array
                   items:
                     type: string
@@ -81,48 +128,93 @@ spec:
                 - rule: has(self.subnetRouter) || self.exitNode == true
                   message: A Connector needs to be either an exit node or a subnet router, or both.
             status:
-              description: ConnectorStatus describes the status of the Connector. This is set and managed by the Tailscale operator.
+              description: |-
+                ConnectorStatus describes the status of the Connector. This is set
+                and managed by the Tailscale operator.
               type: object
               properties:
                 conditions:
-                  description: List of status conditions to indicate the status of the Connector. Known condition types are `ConnectorReady`.
+                  description: |-
+                    List of status conditions to indicate the status of the Connector.
+                    Known condition types are `ConnectorReady`.
                   type: array
                   items:
-                    description: ConnectorCondition contains condition information for a Connector.
+                    description: Condition contains details for one aspect of the current state of this API Resource.
                     type: object
                     required:
+                      - lastTransitionTime
+                      - message
+                      - reason
                       - status
                       - type
                     properties:
                       lastTransitionTime:
-                        description: LastTransitionTime is the timestamp corresponding to the last status change of this condition.
+                        description: |-
+                          lastTransitionTime is the last time the condition transitioned from one status to another.
+                          This should be when the underlying condition changed.  If that is not known, then using the time when the API field changed is acceptable.
                         type: string
                         format: date-time
                       message:
-                        description: Message is a human readable description of the details of the last transition, complementing reason.
+                        description: |-
+                          message is a human readable message indicating details about the transition.
+                          This may be an empty string.
                         type: string
+                        maxLength: 32768
                       observedGeneration:
-                        description: If set, this represents the .metadata.generation that the condition was set based upon. For instance, if .metadata.generation is currently 12, but the .status.condition[x].observedGeneration is 9, the condition is out of date with respect to the current state of the Connector.
+                        description: |-
+                          observedGeneration represents the .metadata.generation that the condition was set based upon.
+                          For instance, if .metadata.generation is currently 12, but the .status.conditions[x].observedGeneration is 9, the condition is out of date
+                          with respect to the current state of the instance.
                         type: integer
                         format: int64
+                        minimum: 0
                       reason:
-                        description: Reason is a brief machine readable explanation for the condition's last transition.
+                        description: |-
+                          reason contains a programmatic identifier indicating the reason for the condition's last transition.
+                          Producers of specific condition types may define expected values and meanings for this field,
+                          and whether the values are considered a guaranteed API.
+                          The value should be a CamelCase string.
+                          This field may not be empty.
                         type: string
+                        maxLength: 1024
+                        minLength: 1
+                        pattern: ^[A-Za-z]([A-Za-z0-9_,:]*[A-Za-z0-9_])?$
                       status:
-                        description: Status of the condition, one of ('True', 'False', 'Unknown').
+                        description: status of the condition, one of True, False, Unknown.
                         type: string
+                        enum:
+                          - "True"
+                          - "False"
+                          - Unknown
                       type:
-                        description: Type of the condition, known values are (`SubnetRouterReady`).
+                        description: type of condition in CamelCase or in foo.example.com/CamelCase.
                         type: string
+                        maxLength: 316
+                        pattern: ^([a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*/)?(([A-Za-z0-9][-A-Za-z0-9_.]*)?[A-Za-z0-9])$
                   x-kubernetes-list-map-keys:
                     - type
                   x-kubernetes-list-type: map
+                hostname:
+                  description: |-
+                    Hostname is the fully qualified domain name of the Connector node.
+                    If MagicDNS is enabled in your tailnet, it is the MagicDNS name of the
+                    node.
+                  type: string
                 isExitNode:
                   description: IsExitNode is set to true if the Connector acts as an exit node.
                   type: boolean
                 subnetRoutes:
-                  description: SubnetRoutes are the routes currently exposed to tailnet via this Connector instance.
+                  description: |-
+                    SubnetRoutes are the routes currently exposed to tailnet via this
+                    Connector instance.
                   type: string
+                tailnetIPs:
+                  description: |-
+                    TailnetIPs is the set of tailnet IP addresses (both IPv4 and IPv6)
+                    assigned to the Connector node.
+                  type: array
+                  items:
+                    type: string
       served: true
       storage: true
       subresources:

cmd/k8s-operator/deploy/crds/tailscale.com_dnsconfigs.yaml

@@ -2,7 +2,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    controller-gen.kubebuilder.io/version: v0.13.0
+    controller-gen.kubebuilder.io/version: v0.15.1-0.20240618033008-7824932b0cab
   name: dnsconfigs.tailscale.com
 spec:
   group: tailscale.com
@@ -23,72 +23,157 @@ spec:
       name: v1alpha1
       schema:
         openAPIV3Schema:
+          description: |-
+            DNSConfig can be deployed to cluster to make a subset of Tailscale MagicDNS
+            names resolvable by cluster workloads. Use this if: A) you need to refer to
+            tailnet services, exposed to cluster via Tailscale Kubernetes operator egress
+            proxies by the MagicDNS names of those tailnet services (usually because the
+            services run over HTTPS)
+            B) you have exposed a cluster workload to the tailnet using Tailscale Ingress
+            and you also want to refer to the workload from within the cluster over the
+            Ingress's MagicDNS name (usually because you have some callback component
+            that needs to use the same URL as that used by a non-cluster client on
+            tailnet).
+            When a DNSConfig is applied to a cluster, Tailscale Kubernetes operator will
+            deploy a nameserver for ts.net DNS names and automatically populate it with records
+            for any Tailscale egress or Ingress proxies deployed to that cluster.
+            Currently you must manually update your cluster DNS configuration to add the
+            IP address of the deployed nameserver as a ts.net stub nameserver.
+            Instructions for how to do it:
+            https://kubernetes.io/docs/tasks/administer-cluster/dns-custom-nameservers/#configuration-of-stub-domain-and-upstream-nameserver-using-coredns (for CoreDNS),
+            https://cloud.google.com/kubernetes-engine/docs/how-to/kube-dns (for kube-dns).
+            Tailscale Kubernetes operator will write the address of a Service fronting
+            the nameserver to dsnconfig.status.nameserver.ip.
+            DNSConfig is a singleton - you must not create more than one.
+            NB: if you want cluster workloads to be able to refer to Tailscale Ingress
+            using its MagicDNS name, you must also annotate the Ingress resource with
+            tailscale.com/experimental-forward-cluster-traffic-via-ingress annotation to
+            ensure that the proxy created for the Ingress listens on its Pod IP address.
+            NB: Clusters where Pods get assigned IPv6 addresses only are currently not supported.
           type: object
           required:
             - spec
           properties:
             apiVersion:
-              description: 'APIVersion defines the versioned schema of this representation of an object. Servers should convert recognized schemas to the latest internal value, and may reject unrecognized values. More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources'
+              description: |-
+                APIVersion defines the versioned schema of this representation of an object.
+                Servers should convert recognized schemas to the latest internal value, and
+                may reject unrecognized values.
+                More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources
               type: string
             kind:
-              description: 'Kind is a string value representing the REST resource this object represents. Servers may infer this from the endpoint the client submits requests to. Cannot be updated. In CamelCase. More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds'
+              description: |-
+                Kind is a string value representing the REST resource this object represents.
+                Servers may infer this from the endpoint the client submits requests to.
+                Cannot be updated.
+                In CamelCase.
+                More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds
               type: string
             metadata:
               type: object
             spec:
+              description: |-
+                Spec describes the desired DNS configuration.
+                More info:
+                https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#spec-and-status
               type: object
               required:
                 - nameserver
               properties:
                 nameserver:
+                  description: |-
+                    Configuration for a nameserver that can resolve ts.net DNS names
+                    associated with in-cluster proxies for Tailscale egress Services and
+                    Tailscale Ingresses. The operator will always deploy this nameserver
+                    when a DNSConfig is applied.
                   type: object
                   properties:
                     image:
+                      description: Nameserver image.
                       type: object
                       properties:
                         repo:
+                          description: Repo defaults to tailscale/k8s-nameserver.
                           type: string
                         tag:
+                          description: Tag defaults to operator's own tag.
                           type: string
             status:
+              description: |-
+                Status describes the status of the DNSConfig. This is set
+                and managed by the Tailscale operator.
               type: object
               properties:
                 conditions:
                   type: array
                   items:
-                    description: ConnectorCondition contains condition information for a Connector.
+                    description: Condition contains details for one aspect of the current state of this API Resource.
                     type: object
                     required:
+                      - lastTransitionTime
+                      - message
+                      - reason
                       - status
                       - type
                     properties:
                       lastTransitionTime:
-                        description: LastTransitionTime is the timestamp corresponding to the last status change of this condition.
+                        description: |-
+                          lastTransitionTime is the last time the condition transitioned from one status to another.
+                          This should be when the underlying condition changed.  If that is not known, then using the time when the API field changed is acceptable.
                         type: string
                         format: date-time
                       message:
-                        description: Message is a human readable description of the details of the last transition, complementing reason.
+                        description: |-
+                          message is a human readable message indicating details about the transition.
+                          This may be an empty string.
                         type: string
+                        maxLength: 32768
                       observedGeneration:
-                        description: If set, this represents the .metadata.generation that the condition was set based upon. For instance, if .metadata.generation is currently 12, but the .status.condition[x].observedGeneration is 9, the condition is out of date with respect to the current state of the Connector.
+                        description: |-
+                          observedGeneration represents the .metadata.generation that the condition was set based upon.
+                          For instance, if .metadata.generation is currently 12, but the .status.conditions[x].observedGeneration is 9, the condition is out of date
+                          with respect to the current state of the instance.
                         type: integer
                         format: int64
+                        minimum: 0
                       reason:
-                        description: Reason is a brief machine readable explanation for the condition's last transition.
+                        description: |-
+                          reason contains a programmatic identifier indicating the reason for the condition's last transition.
+                          Producers of specific condition types may define expected values and meanings for this field,
+                          and whether the values are considered a guaranteed API.
+                          The value should be a CamelCase string.
+                          This field may not be empty.
                         type: string
+                        maxLength: 1024
+                        minLength: 1
+                        pattern: ^[A-Za-z]([A-Za-z0-9_,:]*[A-Za-z0-9_])?$
                       status:
-                        description: Status of the condition, one of ('True', 'False', 'Unknown').
+                        description: status of the condition, one of True, False, Unknown.
                         type: string
+                        enum:
+                          - "True"
+                          - "False"
+                          - Unknown
                       type:
-                        description: Type of the condition, known values are (`SubnetRouterReady`).
+                        description: type of condition in CamelCase or in foo.example.com/CamelCase.
                         type: string
+                        maxLength: 316
+                        pattern: ^([a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*/)?(([A-Za-z0-9][-A-Za-z0-9_.]*)?[A-Za-z0-9])$
                   x-kubernetes-list-map-keys:
                     - type
                   x-kubernetes-list-type: map
                 nameserver:
+                  description: Nameserver describes the status of nameserver cluster resources.
                   type: object
                   properties:
                     ip:
+                      description: |-
+                        IP is the ClusterIP of the Service fronting the deployed ts.net nameserver.
+                        Currently you must manually update your cluster DNS config to add
+                        this address as a stub nameserver for ts.net for cluster workloads to be
+                        able to resolve MagicDNS names associated with egress or Ingress
+                        proxies.
+                        The IP address will change if you delete and recreate the DNSConfig.
                       type: string
       served: true
       storage: true

cmd/k8s-operator/deploy/examples/dnsconfig.yaml

@@ -3,7 +3,4 @@ kind: DNSConfig
 metadata:
   name: ts-dns
 spec:
-  nameserver:
-    image:
-      repo: tailscale/k8s-nameserver
-      tag: unstable-v1.65
+  nameserver: {}

cmd/k8s-operator/deploy/examples/proxyclass.yaml

@@ -15,3 +15,9 @@ spec:
         kubernetes.io/os: "linux"
       imagePullSecrets:
       - name: "foo"
+      tailscaleContainer:
+        image: "ghcr.io/tailscale/tailscale:v1.64"
+        imagePullPolicy: IfNotPresent
+      tailscaleInitContainer:
+        image: "ghcr.io/tailscale/tailscale:v1.64"
+        imagePullPolicy: IfNotPresent

cmd/k8s-operator/generate/main.go

@@ -38,10 +38,10 @@ func main() {
 	}
 	repoRoot := "../../"
 	switch os.Args[1] {
-	case "helmcrd": // insert CRD to Helm templates behind a installCRDs=true conditional check
-		log.Print("Adding Connector CRD to Helm templates")
+	case "helmcrd": // insert CRDs to Helm templates behind a installCRDs=true conditional check
+		log.Print("Adding CRDs to Helm templates")
 		if err := generate("./"); err != nil {
-			log.Fatalf("error adding Connector CRD to Helm templates: %v", err)
+			log.Fatalf("error adding CRDs to Helm templates: %v", err)
 		}
 		return
 	case "staticmanifests": // generate static manifests from Helm templates (including the CRD)

cmd/k8s-operator/ingress.go

@@ -264,7 +264,7 @@ func (a *IngressReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 		ServeConfig:         sc,
 		Tags:                tags,
 		ChildResourceLabels: crl,
-		ProxyClass:          proxyClass,
+		ProxyClassName:      proxyClass,
 	}
 
 	if val := ing.GetAnnotations()[AnnotationExperimentalForwardClusterTrafficViaL7IngresProxy]; val == "true" {

cmd/k8s-operator/ingress_test.go

@@ -100,7 +100,7 @@ func TestTailscaleIngress(t *testing.T) {
 	}
 	opts.serveConfig = serveConfig
 
-	expectEqual(t, fc, expectedSecret(t, opts), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "ingress"), nil)
 	expectEqual(t, fc, expectedSTSUserspace(t, fc, opts), removeHashAnnotation)
 
@@ -231,7 +231,7 @@ func TestTailscaleIngressWithProxyClass(t *testing.T) {
 	}
 	opts.serveConfig = serveConfig
 
-	expectEqual(t, fc, expectedSecret(t, opts), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "ingress"), nil)
 	expectEqual(t, fc, expectedSTSUserspace(t, fc, opts), removeHashAnnotation)
 
@@ -248,9 +248,9 @@ func TestTailscaleIngressWithProxyClass(t *testing.T) {
 	// created proxy resources.
 	mustUpdateStatus(t, fc, "", "custom-metadata", func(pc *tsapi.ProxyClass) {
 		pc.Status = tsapi.ProxyClassStatus{
-			Conditions: []tsapi.ConnectorCondition{{
+			Conditions: []metav1.Condition{{
 				Status:             metav1.ConditionTrue,
-				Type:               tsapi.ProxyClassready,
+				Type:               string(tsapi.ProxyClassready),
 				ObservedGeneration: pc.Generation,
 			}}}
 	})

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
@@ -95,7 +101,7 @@ func (a *NameserverReconciler) Reconcile(ctx context.Context, req reconcile.Requ
 	}
 
 	oldCnStatus := dnsCfg.Status.DeepCopy()
-	setStatus := func(dnsCfg *tsapi.DNSConfig, conditionType tsapi.ConnectorConditionType, status metav1.ConditionStatus, reason, message string) (reconcile.Result, error) {
+	setStatus := func(dnsCfg *tsapi.DNSConfig, conditionType tsapi.ConditionType, status metav1.ConditionStatus, reason, message string) (reconcile.Result, error) {
 		tsoperator.SetDNSConfigCondition(dnsCfg, tsapi.NameserverReady, status, reason, message, dnsCfg.Generation, a.clock, logger)
 		if !apiequality.Semantic.DeepEqual(oldCnStatus, dnsCfg.Status) {
 			// An error encountered here should get returned by the Reconcile function.
@@ -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

@@ -81,12 +81,12 @@ func TestNameserverReconciler(t *testing.T) {
 		IP: "1.2.3.4",
 	}
 	dnsCfg.Finalizers = []string{FinalizerName}
-	dnsCfg.Status.Conditions = append(dnsCfg.Status.Conditions, tsapi.ConnectorCondition{
-		Type:               tsapi.NameserverReady,
+	dnsCfg.Status.Conditions = append(dnsCfg.Status.Conditions, metav1.Condition{
+		Type:               string(tsapi.NameserverReady),
 		Status:             metav1.ConditionTrue,
 		Reason:             reasonNameserverCreated,
 		Message:            reasonNameserverCreated,
-		LastTransitionTime: &metav1.Time{Time: cl.Now().Truncate(time.Second)},
+		LastTransitionTime: metav1.Time{Time: cl.Now().Truncate(time.Second)},
 	})
 	expectEqual(t, fc, dnsCfg, nil)
 
@@ -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
 
@@ -60,12 +60,13 @@ func main() {
 	tailscale.I_Acknowledge_This_API_Is_Unstable = true
 
 	var (
-		tsNamespace       = defaultEnv("OPERATOR_NAMESPACE", "")
-		tslogging         = defaultEnv("OPERATOR_LOGGING", "info")
-		image             = defaultEnv("PROXY_IMAGE", "tailscale/tailscale:latest")
-		priorityClassName = defaultEnv("PROXY_PRIORITY_CLASS_NAME", "")
-		tags              = defaultEnv("PROXY_TAGS", "tag:k8s")
-		tsFirewallMode    = defaultEnv("PROXY_FIREWALL_MODE", "")
+		tsNamespace           = defaultEnv("OPERATOR_NAMESPACE", "")
+		tslogging             = defaultEnv("OPERATOR_LOGGING", "info")
+		image                 = defaultEnv("PROXY_IMAGE", "tailscale/tailscale:latest")
+		priorityClassName     = defaultEnv("PROXY_PRIORITY_CLASS_NAME", "")
+		tags                  = defaultEnv("PROXY_TAGS", "tag:k8s")
+		tsFirewallMode        = defaultEnv("PROXY_FIREWALL_MODE", "")
+		isDefaultLoadBalancer = defaultBool("OPERATOR_DEFAULT_LOAD_BALANCER", false)
 	)
 
 	var opts []kzap.Opts
@@ -94,9 +95,19 @@ func main() {
 	defer s.Close()
 	restConfig := config.GetConfigOrDie()
 	maybeLaunchAPIServerProxy(zlog, restConfig, s, mode)
-	// TODO (irbekrm): gather the reconciler options into an opts struct
-	// rather than passing a million of them in one by one.
-	runReconcilers(zlog, s, tsNamespace, restConfig, tsClient, image, priorityClassName, tags, tsFirewallMode)
+	rOpts := reconcilerOpts{
+		log:                           zlog,
+		tsServer:                      s,
+		tsClient:                      tsClient,
+		tailscaleNamespace:            tsNamespace,
+		restConfig:                    restConfig,
+		proxyImage:                    image,
+		proxyPriorityClassName:        priorityClassName,
+		proxyActAsDefaultLoadBalancer: isDefaultLoadBalancer,
+		proxyTags:                     tags,
+		proxyFirewallMode:             tsFirewallMode,
+	}
+	runReconcilers(rOpts)
 }
 
 // initTSNet initializes the tsnet.Server and logs in to Tailscale. It uses the
@@ -204,11 +215,8 @@ waitOnline:
 
 // runReconcilers starts the controller-runtime manager and registers the
 // ServiceReconciler. It blocks forever.
-func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string, restConfig *rest.Config, tsClient *tailscale.Client, image, priorityClassName, tags, tsFirewallMode string) {
-	var (
-		isDefaultLoadBalancer = defaultBool("OPERATOR_DEFAULT_LOAD_BALANCER", false)
-	)
-	startlog := zlog.Named("startReconcilers")
+func runReconcilers(opts reconcilerOpts) {
+	startlog := opts.log.Named("startReconcilers")
 	// For secrets and statefulsets, we only get permission to touch the objects
 	// in the controller's own namespace. This cannot be expressed by
 	// .Watches(...) below, instead you have to add a per-type field selector to
@@ -216,7 +224,7 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 	// implicitly filter what parts of the world the builder code gets to see at
 	// all.
 	nsFilter := cache.ByObject{
-		Field: client.InNamespace(tsNamespace).AsSelector(),
+		Field: client.InNamespace(opts.tailscaleNamespace).AsSelector(),
 	}
 	mgrOpts := manager.Options{
 		// TODO (irbekrm): stricter filtering what we watch/cache/call
@@ -234,7 +242,7 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 		},
 		Scheme: tsapi.GlobalScheme,
 	}
-	mgr, err := manager.New(restConfig, mgrOpts)
+	mgr, err := manager.New(opts.restConfig, mgrOpts)
 	if err != nil {
 		startlog.Fatalf("could not create manager: %v", err)
 	}
@@ -248,13 +256,13 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 	eventRecorder := mgr.GetEventRecorderFor("tailscale-operator")
 	ssr := &tailscaleSTSReconciler{
 		Client:                 mgr.GetClient(),
-		tsnetServer:            s,
-		tsClient:               tsClient,
-		defaultTags:            strings.Split(tags, ","),
-		operatorNamespace:      tsNamespace,
-		proxyImage:             image,
-		proxyPriorityClassName: priorityClassName,
-		tsFirewallMode:         tsFirewallMode,
+		tsnetServer:            opts.tsServer,
+		tsClient:               opts.tsClient,
+		defaultTags:            strings.Split(opts.proxyTags, ","),
+		operatorNamespace:      opts.tailscaleNamespace,
+		proxyImage:             opts.proxyImage,
+		proxyPriorityClassName: opts.proxyPriorityClassName,
+		tsFirewallMode:         opts.proxyFirewallMode,
 	}
 	err = builder.
 		ControllerManagedBy(mgr).
@@ -266,10 +274,11 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 		Complete(&ServiceReconciler{
 			ssr:                   ssr,
 			Client:                mgr.GetClient(),
-			logger:                zlog.Named("service-reconciler"),
-			isDefaultLoadBalancer: isDefaultLoadBalancer,
+			logger:                opts.log.Named("service-reconciler"),
+			isDefaultLoadBalancer: opts.proxyActAsDefaultLoadBalancer,
 			recorder:              eventRecorder,
-			tsNamespace:           tsNamespace,
+			tsNamespace:           opts.tailscaleNamespace,
+			clock:                 tstime.DefaultClock{},
 		})
 	if err != nil {
 		startlog.Fatalf("could not create service reconciler: %v", err)
@@ -291,7 +300,7 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 			ssr:      ssr,
 			recorder: eventRecorder,
 			Client:   mgr.GetClient(),
-			logger:   zlog.Named("ingress-reconciler"),
+			logger:   opts.log.Named("ingress-reconciler"),
 		})
 	if err != nil {
 		startlog.Fatalf("could not create ingress reconciler: %v", err)
@@ -310,14 +319,14 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 			ssr:      ssr,
 			recorder: eventRecorder,
 			Client:   mgr.GetClient(),
-			logger:   zlog.Named("connector-reconciler"),
+			logger:   opts.log.Named("connector-reconciler"),
 			clock:    tstime.DefaultClock{},
 		})
 	if err != nil {
 		startlog.Fatalf("could not create connector reconciler: %v", err)
 	}
 	// TODO (irbekrm): switch to metadata-only watches for resources whose
-	// spec we don't need to inspect to reduce memory consumption
+	// spec we don't need to inspect to reduce memory consumption.
 	// https://github.com/kubernetes-sigs/controller-runtime/issues/1159
 	nameserverFilter := handler.EnqueueRequestsFromMapFunc(managedResourceHandlerForType("nameserver"))
 	err = builder.ControllerManagedBy(mgr).
@@ -328,11 +337,10 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 		Watches(&corev1.ServiceAccount{}, nameserverFilter).
 		Complete(&NameserverReconciler{
 			recorder:    eventRecorder,
-			tsNamespace: tsNamespace,
-
-			Client: mgr.GetClient(),
-			logger: zlog.Named("nameserver-reconciler"),
-			clock:  tstime.DefaultClock{},
+			tsNamespace: opts.tailscaleNamespace,
+			Client:      mgr.GetClient(),
+			logger:      opts.log.Named("nameserver-reconciler"),
+			clock:       tstime.DefaultClock{},
 		})
 	if err != nil {
 		startlog.Fatalf("could not create nameserver reconciler: %v", err)
@@ -342,7 +350,7 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 		Complete(&ProxyClassReconciler{
 			Client:   mgr.GetClient(),
 			recorder: eventRecorder,
-			logger:   zlog.Named("proxyclass-reconciler"),
+			logger:   opts.log.Named("proxyclass-reconciler"),
 			clock:    tstime.DefaultClock{},
 		})
 	if err != nil {
@@ -355,12 +363,12 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 	dnsRREpsOpts := handler.EnqueueRequestsFromMapFunc(dnsRecordsReconcilerEndpointSliceHandler)
 	// On DNSConfig changes, reconcile all headless Services for
 	// ingress/egress proxies in operator namespace.
-	dnsRRDNSConfigOpts := handler.EnqueueRequestsFromMapFunc(enqueueAllIngressEgressProxySvcsInNS(tsNamespace, mgr.GetClient(), logger))
+	dnsRRDNSConfigOpts := handler.EnqueueRequestsFromMapFunc(enqueueAllIngressEgressProxySvcsInNS(opts.tailscaleNamespace, mgr.GetClient(), logger))
 	// On Service events, if it is an ingress/egress proxy headless Service, reconcile it.
 	dnsRRServiceOpts := handler.EnqueueRequestsFromMapFunc(dnsRecordsReconcilerServiceHandler)
 	// On Ingress events, if it is a tailscale Ingress or if tailscale is the default ingress controller, reconcile the proxy
 	// headless Service.
-	dnsRRIngressOpts := handler.EnqueueRequestsFromMapFunc(dnsRecordsReconcilerIngressHandler(tsNamespace, isDefaultLoadBalancer, mgr.GetClient(), logger))
+	dnsRRIngressOpts := handler.EnqueueRequestsFromMapFunc(dnsRecordsReconcilerIngressHandler(opts.tailscaleNamespace, opts.proxyActAsDefaultLoadBalancer, mgr.GetClient(), logger))
 	err = builder.ControllerManagedBy(mgr).
 		Named("dns-records-reconciler").
 		Watches(&corev1.Service{}, dnsRRServiceOpts).
@@ -369,9 +377,9 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 		Watches(&tsapi.DNSConfig{}, dnsRRDNSConfigOpts).
 		Complete(&dnsRecordsReconciler{
 			Client:                mgr.GetClient(),
-			tsNamespace:           tsNamespace,
-			logger:                zlog.Named("dns-records-reconciler"),
-			isDefaultLoadBalancer: isDefaultLoadBalancer,
+			tsNamespace:           opts.tailscaleNamespace,
+			logger:                opts.log.Named("dns-records-reconciler"),
+			isDefaultLoadBalancer: opts.proxyActAsDefaultLoadBalancer,
 		})
 	if err != nil {
 		startlog.Fatalf("could not create DNS records reconciler: %v", err)
@@ -382,6 +390,42 @@ func runReconcilers(zlog *zap.SugaredLogger, s *tsnet.Server, tsNamespace string
 	}
 }
 
+type reconcilerOpts struct {
+	log                *zap.SugaredLogger
+	tsServer           *tsnet.Server
+	tsClient           *tailscale.Client
+	tailscaleNamespace string       // namespace in which operator resources will be deployed
+	restConfig         *rest.Config // config for connecting to the kube API server
+	proxyImage         string       // <proxy-image-repo>:<proxy-image-tag>
+	// proxyPriorityClassName isPriorityClass to be set for proxy Pods. This
+	// is a legacy mechanism for cluster resource configuration options -
+	// going forward use ProxyClass.
+	// https://kubernetes.io/docs/concepts/scheduling-eviction/pod-priority-preemption/#priorityclass
+	proxyPriorityClassName string
+	// proxyTags are ACL tags to tag proxy auth keys. Multiple tags should
+	// be provided as a string with comma-separated tag values. Proxy tags
+	// default to tag:k8s.
+	// https://tailscale.com/kb/1085/auth-keys
+	proxyTags string
+	// proxyActAsDefaultLoadBalancer determines whether this operator
+	// instance should act as the default ingress controller when looking at
+	// Ingress resources with unset ingress.spec.ingressClassName.
+	// TODO (irbekrm): this setting does not respect the default
+	// IngressClass.
+	// https://kubernetes.io/docs/concepts/services-networking/ingress/#default-ingress-class
+	// We should fix that and preferably integrate with that mechanism as
+	// well - perhaps make the operator itself create the default
+	// IngressClass if this is set to true.
+	proxyActAsDefaultLoadBalancer bool
+	// proxyFirewallMode determines whether non-userspace proxies should use
+	// iptables or nftables for firewall configuration. Accepted values are
+	// iptables, nftables and auto. If set to auto, proxy will automatically
+	// determine which mode is supported for a given host (prefer nftables).
+	// Auto is usually the best choice, unless you want to explicitly set
+	// specific mode for debugging purposes.
+	proxyFirewallMode string
+}
+
 // enqueueAllIngressEgressProxySvcsinNS returns a reconcile request for each
 // ingress/egress proxy headless Service found in the provided namespace.
 func enqueueAllIngressEgressProxySvcsInNS(ns string, cl client.Client, logger *zap.SugaredLogger) handler.MapFunc {

cmd/k8s-operator/operator_test.go

@@ -9,6 +9,7 @@ import (
 	"context"
 	"fmt"
 	"testing"
+	"time"
 
 	"github.com/google/go-cmp/cmp"
 	"go.uber.org/zap"
@@ -17,10 +18,13 @@ import (
 	networkingv1 "k8s.io/api/networking/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
+	"k8s.io/client-go/tools/record"
 	"sigs.k8s.io/controller-runtime/pkg/client/fake"
 	"sigs.k8s.io/controller-runtime/pkg/reconcile"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
 	"tailscale.com/net/dns/resolvconffile"
+	"tailscale.com/tstest"
+	"tailscale.com/tstime"
 	"tailscale.com/types/ptr"
 	"tailscale.com/util/dnsname"
 	"tailscale.com/util/mak"
@@ -33,6 +37,7 @@ func TestLoadBalancerClass(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -42,11 +47,13 @@ func TestLoadBalancerClass(t *testing.T) {
 			operatorNamespace: "operator-ns",
 			proxyImage:        "tailscale/tailscale",
 		},
-		logger: zl.Sugar(),
+		logger:   zl.Sugar(),
+		clock:    clock,
+		recorder: record.NewFakeRecorder(100),
 	}
 
-	// Create a service that we should manage, and check that the initial round
-	// of objects looks right.
+	// Create a service that we should manage, but start with a miconfiguration
+	// in the annotations.
 	mustCreate(t, fc, &corev1.Service{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      "test",
@@ -55,6 +62,9 @@ func TestLoadBalancerClass(t *testing.T) {
 			// doesn't. So, set it explicitly because other code later depends
 			// on it being set.
 			UID: types.UID("1234-UID"),
+			Annotations: map[string]string{
+				AnnotationTailnetTargetFQDN: "invalid.example.com",
+			},
 		},
 		Spec: corev1.ServiceSpec{
 			ClusterIP:         "10.20.30.40",
@@ -65,6 +75,46 @@ func TestLoadBalancerClass(t *testing.T) {
 
 	expectReconciled(t, sr, "default", "test")
 
+	// The expected value of .status.conditions[0].LastTransitionTime until the
+	// proxy becomes ready.
+	t0 := conditionTime(clock)
+
+	// Should have an error about invalid config.
+	want := &corev1.Service{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "test",
+			Namespace: "default",
+			UID:       types.UID("1234-UID"),
+			Annotations: map[string]string{
+				AnnotationTailnetTargetFQDN: "invalid.example.com",
+			},
+		},
+		Spec: corev1.ServiceSpec{
+			ClusterIP:         "10.20.30.40",
+			Type:              corev1.ServiceTypeLoadBalancer,
+			LoadBalancerClass: ptr.To("tailscale"),
+		},
+		Status: corev1.ServiceStatus{
+			Conditions: []metav1.Condition{{
+				Type:               string(tsapi.ProxyReady),
+				Status:             metav1.ConditionFalse,
+				LastTransitionTime: t0,
+				Reason:             reasonProxyInvalid,
+				Message:            `unable to provision proxy resources: invalid Service: invalid value of annotation tailscale.com/tailnet-fqdn: "invalid.example.com" does not appear to be a valid MagicDNS name`,
+			}},
+		},
+	}
+	expectEqual(t, fc, want, nil)
+
+	// Delete the misconfiguration so the proxy starts getting created on the
+	// next reconcile.
+	mustUpdate(t, fc, "default", "test", func(s *corev1.Service) {
+		s.ObjectMeta.Annotations = nil
+	})
+
+	clock.Advance(time.Second)
+	expectReconciled(t, sr, "default", "test")
+
 	fullName, shortName := findGenName(t, fc, "default", "test", "svc")
 	opts := configOpts{
 		stsName:         shortName,
@@ -75,10 +125,23 @@ func TestLoadBalancerClass(t *testing.T) {
 		clusterTargetIP: "10.20.30.40",
 	}
 
-	expectEqual(t, fc, expectedSecret(t, opts), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, opts), removeHashAnnotation)
 
+	want.Annotations = nil
+	want.ObjectMeta.Finalizers = []string{"tailscale.com/finalizer"}
+	want.Status = corev1.ServiceStatus{
+		Conditions: []metav1.Condition{{
+			Type:               string(tsapi.ProxyReady),
+			Status:             metav1.ConditionFalse,
+			LastTransitionTime: t0, // Status is still false, no update to transition time
+			Reason:             reasonProxyPending,
+			Message:            "no Tailscale hostname known yet, waiting for proxy pod to finish auth",
+		}},
+	}
+	expectEqual(t, fc, want, nil)
+
 	// Normally the Tailscale proxy pod would come up here and write its info
 	// into the secret. Simulate that, then verify reconcile again and verify
 	// that we get to the end.
@@ -90,33 +153,16 @@ func TestLoadBalancerClass(t *testing.T) {
 		s.Data["device_fqdn"] = []byte("tailscale.device.name.")
 		s.Data["device_ips"] = []byte(`["100.99.98.97", "2c0a:8083:94d4:2012:3165:34a5:3616:5fdf"]`)
 	})
+	clock.Advance(time.Second)
 	expectReconciled(t, sr, "default", "test")
-	want := &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
-		ObjectMeta: metav1.ObjectMeta{
-			Name:       "test",
-			Namespace:  "default",
-			Finalizers: []string{"tailscale.com/finalizer"},
-			UID:        types.UID("1234-UID"),
-		},
-		Spec: corev1.ServiceSpec{
-			ClusterIP:         "10.20.30.40",
-			Type:              corev1.ServiceTypeLoadBalancer,
-			LoadBalancerClass: ptr.To("tailscale"),
-		},
-		Status: corev1.ServiceStatus{
-			LoadBalancer: corev1.LoadBalancerStatus{
-				Ingress: []corev1.LoadBalancerIngress{
-					{
-						Hostname: "tailscale.device.name",
-					},
-					{
-						IP: "100.99.98.97",
-					},
-				},
+	want.Status.Conditions = proxyCreatedCondition(clock)
+	want.Status.LoadBalancer = corev1.LoadBalancerStatus{
+		Ingress: []corev1.LoadBalancerIngress{
+			{
+				Hostname: "tailscale.device.name",
+			},
+			{
+				IP: "100.99.98.97",
 			},
 		},
 	}
@@ -144,11 +190,9 @@ func TestLoadBalancerClass(t *testing.T) {
 	expectMissing[appsv1.StatefulSet](t, fc, "operator-ns", shortName)
 	expectMissing[corev1.Service](t, fc, "operator-ns", shortName)
 	expectMissing[corev1.Secret](t, fc, "operator-ns", fullName)
+
+	// Note that the Tailscale-specific condition status should be gone now.
 	want = &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      "test",
 			Namespace: "default",
@@ -170,6 +214,7 @@ func TestTailnetTargetFQDNAnnotation(t *testing.T) {
 		t.Fatal(err)
 	}
 	tailnetTargetFQDN := "foo.bar.ts.net."
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -180,6 +225,7 @@ func TestTailnetTargetFQDNAnnotation(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// Create a service that we should manage, and check that the initial round
@@ -216,14 +262,10 @@ func TestTailnetTargetFQDNAnnotation(t *testing.T) {
 		hostname:          "default-test",
 	}
 
-	expectEqual(t, fc, expectedSecret(t, o), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, o), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 	want := &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:       "test",
 			Namespace:  "default",
@@ -238,9 +280,12 @@ func TestTailnetTargetFQDNAnnotation(t *testing.T) {
 			Type:         corev1.ServiceTypeExternalName,
 			Selector:     nil,
 		},
+		Status: corev1.ServiceStatus{
+			Conditions: proxyCreatedCondition(clock),
+		},
 	}
 	expectEqual(t, fc, want, nil)
-	expectEqual(t, fc, expectedSecret(t, o), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, o), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 
@@ -280,6 +325,7 @@ func TestTailnetTargetIPAnnotation(t *testing.T) {
 		t.Fatal(err)
 	}
 	tailnetTargetIP := "100.66.66.66"
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -290,6 +336,7 @@ func TestTailnetTargetIPAnnotation(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// Create a service that we should manage, and check that the initial round
@@ -326,14 +373,10 @@ func TestTailnetTargetIPAnnotation(t *testing.T) {
 		hostname:        "default-test",
 	}
 
-	expectEqual(t, fc, expectedSecret(t, o), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, o), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 	want := &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:       "test",
 			Namespace:  "default",
@@ -348,9 +391,12 @@ func TestTailnetTargetIPAnnotation(t *testing.T) {
 			Type:         corev1.ServiceTypeExternalName,
 			Selector:     nil,
 		},
+		Status: corev1.ServiceStatus{
+			Conditions: proxyCreatedCondition(clock),
+		},
 	}
 	expectEqual(t, fc, want, nil)
-	expectEqual(t, fc, expectedSecret(t, o), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, o), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 
@@ -389,6 +435,7 @@ func TestAnnotations(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -399,6 +446,7 @@ func TestAnnotations(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// Create a service that we should manage, and check that the initial round
@@ -433,14 +481,10 @@ func TestAnnotations(t *testing.T) {
 		clusterTargetIP: "10.20.30.40",
 	}
 
-	expectEqual(t, fc, expectedSecret(t, o), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, o), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 	want := &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:       "test",
 			Namespace:  "default",
@@ -454,6 +498,9 @@ func TestAnnotations(t *testing.T) {
 			ClusterIP: "10.20.30.40",
 			Type:      corev1.ServiceTypeClusterIP,
 		},
+		Status: corev1.ServiceStatus{
+			Conditions: proxyCreatedCondition(clock),
+		},
 	}
 	expectEqual(t, fc, want, nil)
 
@@ -473,10 +520,6 @@ func TestAnnotations(t *testing.T) {
 	expectMissing[corev1.Service](t, fc, "operator-ns", shortName)
 	expectMissing[corev1.Secret](t, fc, "operator-ns", fullName)
 	want = &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      "test",
 			Namespace: "default",
@@ -497,6 +540,7 @@ func TestAnnotationIntoLB(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -507,6 +551,7 @@ func TestAnnotationIntoLB(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// Create a service that we should manage, and check that the initial round
@@ -541,7 +586,7 @@ func TestAnnotationIntoLB(t *testing.T) {
 		clusterTargetIP: "10.20.30.40",
 	}
 
-	expectEqual(t, fc, expectedSecret(t, o), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, o), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 
@@ -558,10 +603,6 @@ func TestAnnotationIntoLB(t *testing.T) {
 	})
 	expectReconciled(t, sr, "default", "test")
 	want := &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:       "test",
 			Namespace:  "default",
@@ -575,6 +616,9 @@ func TestAnnotationIntoLB(t *testing.T) {
 			ClusterIP: "10.20.30.40",
 			Type:      corev1.ServiceTypeClusterIP,
 		},
+		Status: corev1.ServiceStatus{
+			Conditions: proxyCreatedCondition(clock),
+		},
 	}
 	expectEqual(t, fc, want, nil)
 
@@ -592,10 +636,6 @@ func TestAnnotationIntoLB(t *testing.T) {
 	// ... but the service should have a LoadBalancer status.
 
 	want = &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:       "test",
 			Namespace:  "default",
@@ -618,6 +658,7 @@ func TestAnnotationIntoLB(t *testing.T) {
 					},
 				},
 			},
+			Conditions: proxyCreatedCondition(clock),
 		},
 	}
 	expectEqual(t, fc, want, nil)
@@ -630,6 +671,7 @@ func TestLBIntoAnnotation(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -640,6 +682,7 @@ func TestLBIntoAnnotation(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// Create a service that we should manage, and check that the initial round
@@ -672,7 +715,7 @@ func TestLBIntoAnnotation(t *testing.T) {
 		clusterTargetIP: "10.20.30.40",
 	}
 
-	expectEqual(t, fc, expectedSecret(t, o), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, o), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 
@@ -689,10 +732,6 @@ func TestLBIntoAnnotation(t *testing.T) {
 	})
 	expectReconciled(t, sr, "default", "test")
 	want := &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:       "test",
 			Namespace:  "default",
@@ -715,6 +754,7 @@ func TestLBIntoAnnotation(t *testing.T) {
 					},
 				},
 			},
+			Conditions: proxyCreatedCondition(clock),
 		},
 	}
 	expectEqual(t, fc, want, nil)
@@ -740,10 +780,6 @@ func TestLBIntoAnnotation(t *testing.T) {
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 
 	want = &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:       "test",
 			Namespace:  "default",
@@ -757,6 +793,9 @@ func TestLBIntoAnnotation(t *testing.T) {
 			ClusterIP: "10.20.30.40",
 			Type:      corev1.ServiceTypeClusterIP,
 		},
+		Status: corev1.ServiceStatus{
+			Conditions: proxyCreatedCondition(clock),
+		},
 	}
 	expectEqual(t, fc, want, nil)
 }
@@ -768,6 +807,7 @@ func TestCustomHostname(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -778,6 +818,7 @@ func TestCustomHostname(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// Create a service that we should manage, and check that the initial round
@@ -813,14 +854,10 @@ func TestCustomHostname(t *testing.T) {
 		clusterTargetIP: "10.20.30.40",
 	}
 
-	expectEqual(t, fc, expectedSecret(t, o), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, o), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, o), removeHashAnnotation)
 	want := &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:       "test",
 			Namespace:  "default",
@@ -835,6 +872,9 @@ func TestCustomHostname(t *testing.T) {
 			ClusterIP: "10.20.30.40",
 			Type:      corev1.ServiceTypeClusterIP,
 		},
+		Status: corev1.ServiceStatus{
+			Conditions: proxyCreatedCondition(clock),
+		},
 	}
 	expectEqual(t, fc, want, nil)
 
@@ -854,10 +894,6 @@ func TestCustomHostname(t *testing.T) {
 	expectMissing[corev1.Service](t, fc, "operator-ns", shortName)
 	expectMissing[corev1.Secret](t, fc, "operator-ns", fullName)
 	want = &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      "test",
 			Namespace: "default",
@@ -881,6 +917,7 @@ func TestCustomPriorityClassName(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -892,6 +929,7 @@ func TestCustomPriorityClassName(t *testing.T) {
 			proxyPriorityClassName: "custom-priority-class-name",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// Create a service that we should manage, and check that the initial round
@@ -935,10 +973,14 @@ func TestProxyClassForService(t *testing.T) {
 	// Setup
 	pc := &tsapi.ProxyClass{
 		ObjectMeta: metav1.ObjectMeta{Name: "custom-metadata"},
-		Spec: tsapi.ProxyClassSpec{StatefulSet: &tsapi.StatefulSet{
-			Labels:      map[string]string{"foo": "bar"},
-			Annotations: map[string]string{"bar.io/foo": "some-val"},
-			Pod:         &tsapi.Pod{Annotations: map[string]string{"foo.io/bar": "some-val"}}}},
+		Spec: tsapi.ProxyClassSpec{
+			TailscaleConfig: &tsapi.TailscaleConfig{
+				AcceptRoutes: true,
+			},
+			StatefulSet: &tsapi.StatefulSet{
+				Labels:      map[string]string{"foo": "bar"},
+				Annotations: map[string]string{"bar.io/foo": "some-val"},
+				Pod:         &tsapi.Pod{Annotations: map[string]string{"foo.io/bar": "some-val"}}}},
 	}
 	fc := fake.NewClientBuilder().
 		WithScheme(tsapi.GlobalScheme).
@@ -950,6 +992,7 @@ func TestProxyClassForService(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -960,6 +1003,7 @@ func TestProxyClassForService(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// 1. A new tailscale LoadBalancer Service is created without any
@@ -989,7 +1033,7 @@ func TestProxyClassForService(t *testing.T) {
 		hostname:        "default-test",
 		clusterTargetIP: "10.20.30.40",
 	}
-	expectEqual(t, fc, expectedSecret(t, opts), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, opts), removeHashAnnotation)
 
@@ -1001,21 +1045,23 @@ func TestProxyClassForService(t *testing.T) {
 	})
 	expectReconciled(t, sr, "default", "test")
 	expectEqual(t, fc, expectedSTS(t, fc, opts), removeHashAnnotation)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 
 	// 3. ProxyClass is set to Ready, the Service gets reconciled by the
 	// services-reconciler and the customization from the ProxyClass is
 	// applied to the proxy resources.
 	mustUpdateStatus(t, fc, "", "custom-metadata", func(pc *tsapi.ProxyClass) {
 		pc.Status = tsapi.ProxyClassStatus{
-			Conditions: []tsapi.ConnectorCondition{{
+			Conditions: []metav1.Condition{{
 				Status:             metav1.ConditionTrue,
-				Type:               tsapi.ProxyClassready,
+				Type:               string(tsapi.ProxyClassready),
 				ObservedGeneration: pc.Generation,
 			}}}
 	})
 	opts.proxyClass = pc.Name
 	expectReconciled(t, sr, "default", "test")
 	expectEqual(t, fc, expectedSTS(t, fc, opts), removeHashAnnotation)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), removeAuthKeyIfExistsModifier(t))
 
 	// 4. tailscale.com/proxy-class label is removed from the Service, the
 	// configuration from the ProxyClass is removed from the cluster
@@ -1035,6 +1081,7 @@ func TestDefaultLoadBalancer(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -1045,6 +1092,7 @@ func TestDefaultLoadBalancer(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger:                zl.Sugar(),
+		clock:                 clock,
 		isDefaultLoadBalancer: true,
 	}
 
@@ -1089,6 +1137,7 @@ func TestProxyFirewallMode(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -1100,6 +1149,7 @@ func TestProxyFirewallMode(t *testing.T) {
 			tsFirewallMode:    "nftables",
 		},
 		logger:                zl.Sugar(),
+		clock:                 clock,
 		isDefaultLoadBalancer: true,
 	}
 
@@ -1142,6 +1192,7 @@ func TestTailscaledConfigfileHash(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -1152,6 +1203,7 @@ func TestTailscaledConfigfileHash(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger:                zl.Sugar(),
+		clock:                 clock,
 		isDefaultLoadBalancer: true,
 	}
 
@@ -1182,7 +1234,7 @@ 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 +1244,7 @@ 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)
 }
@@ -1433,6 +1485,7 @@ func Test_externalNameService(t *testing.T) {
 
 	// 1. A External name Service that should be exposed via Tailscale gets
 	// created.
+	clock := tstest.NewClock(tstest.ClockOpts{})
 	sr := &ServiceReconciler{
 		Client: fc,
 		ssr: &tailscaleSTSReconciler{
@@ -1443,6 +1496,7 @@ func Test_externalNameService(t *testing.T) {
 			proxyImage:        "tailscale/tailscale",
 		},
 		logger: zl.Sugar(),
+		clock:  clock,
 	}
 
 	// 1. Create an ExternalName Service that we should manage, and check that the initial round
@@ -1477,7 +1531,7 @@ func Test_externalNameService(t *testing.T) {
 		clusterTargetDNS: "foo.com",
 	}
 
-	expectEqual(t, fc, expectedSecret(t, opts), nil)
+	expectEqual(t, fc, expectedSecret(t, fc, opts), nil)
 	expectEqual(t, fc, expectedHeadlessService(shortName, "svc"), nil)
 	expectEqual(t, fc, expectedSTS(t, fc, opts), removeHashAnnotation)
 
@@ -1498,3 +1552,18 @@ func toFQDN(t *testing.T, s string) dnsname.FQDN {
 	}
 	return fqdn
 }
+
+func proxyCreatedCondition(clock tstime.Clock) []metav1.Condition {
+	return []metav1.Condition{{
+		Type:               string(tsapi.ProxyReady),
+		Status:             metav1.ConditionTrue,
+		ObservedGeneration: 0,
+		LastTransitionTime: conditionTime(clock),
+		Reason:             reasonProxyCreated,
+		Message:            reasonProxyCreated,
+	}}
+}
+
+func conditionTime(clock tstime.Clock) metav1.Time {
+	return metav1.NewTime(clock.Now().Truncate(time.Second))
+}

cmd/k8s-operator/proxy.go

@@ -20,6 +20,7 @@ import (
 	"k8s.io/client-go/transport"
 	"tailscale.com/client/tailscale"
 	"tailscale.com/client/tailscale/apitype"
+	tskube "tailscale.com/kube"
 	"tailscale.com/tailcfg"
 	"tailscale.com/tsnet"
 	"tailscale.com/util/clientmetric"
@@ -207,32 +208,24 @@ func runAPIServerProxy(s *tsnet.Server, rt http.RoundTripper, log *zap.SugaredLo
 }
 
 const (
-	capabilityName    = "tailscale.com/cap/kubernetes"
-	oldCapabilityName = "https://" + capabilityName
+	// oldCapabilityName is a legacy form of
+	// tailfcg.PeerCapabilityKubernetes capability. The only capability rule
+	// that is respected for this form is group impersonation - for
+	// backwards compatibility reasons.
+	// TODO (irbekrm): determine if anyone uses this and remove if possible.
+	oldCapabilityName = "https://" + tailcfg.PeerCapabilityKubernetes
 )
 
-type capRule struct {
-	// Impersonate is a list of rules that specify how to impersonate the caller
-	// when proxying to the Kubernetes API.
-	Impersonate *impersonateRule `json:"impersonate,omitempty"`
-}
-
-// TODO(maisem): move this to some well-known location so that it can be shared
-// with control.
-type impersonateRule struct {
-	Groups []string `json:"groups,omitempty"`
-}
-
 // addImpersonationHeaders adds the appropriate headers to r to impersonate the
 // caller when proxying to the Kubernetes API. It uses the WhoIsResponse stashed
 // in the context by the apiserverProxy.
 func addImpersonationHeaders(r *http.Request, log *zap.SugaredLogger) error {
 	log = log.With("remote", r.RemoteAddr)
 	who := whoIsKey.Value(r.Context())
-	rules, err := tailcfg.UnmarshalCapJSON[capRule](who.CapMap, capabilityName)
+	rules, err := tailcfg.UnmarshalCapJSON[tskube.KubernetesCapRule](who.CapMap, tailcfg.PeerCapabilityKubernetes)
 	if len(rules) == 0 && err == nil {
 		// Try the old capability name for backwards compatibility.
-		rules, err = tailcfg.UnmarshalCapJSON[capRule](who.CapMap, oldCapabilityName)
+		rules, err = tailcfg.UnmarshalCapJSON[tskube.KubernetesCapRule](who.CapMap, oldCapabilityName)
 	}
 	if err != nil {
 		return fmt.Errorf("failed to unmarshal capability: %v", err)

cmd/k8s-operator/proxy_test.go

@@ -49,7 +49,7 @@ func TestImpersonationHeaders(t *testing.T) {
 			name:     "user-with-cap",
 			emailish: "foo@example.com",
 			capMap: tailcfg.PeerCapMap{
-				capabilityName: {
+				tailcfg.PeerCapabilityKubernetes: {
 					tailcfg.RawMessage(`{"impersonate":{"groups":["group1","group2"]}}`),
 					tailcfg.RawMessage(`{"impersonate":{"groups":["group1","group3"]}}`), // One group is duplicated.
 					tailcfg.RawMessage(`{"impersonate":{"groups":["group4"]}}`),
@@ -71,7 +71,7 @@ func TestImpersonationHeaders(t *testing.T) {
 			emailish: "tagged-device",
 			tags:     []string{"tag:foo", "tag:bar"},
 			capMap: tailcfg.PeerCapMap{
-				capabilityName: {
+				tailcfg.PeerCapabilityKubernetes: {
 					tailcfg.RawMessage(`{"impersonate":{"groups":["group1"]}}`),
 				},
 			},
@@ -80,12 +80,26 @@ func TestImpersonationHeaders(t *testing.T) {
 				"Impersonate-User":  {"node.ts.net"},
 			},
 		},
+		{
+			name:     "mix-of-caps",
+			emailish: "tagged-device",
+			tags:     []string{"tag:foo", "tag:bar"},
+			capMap: tailcfg.PeerCapMap{
+				tailcfg.PeerCapabilityKubernetes: {
+					tailcfg.RawMessage(`{"impersonate":{"groups":["group1"]},"recorder":["tag:foo"],"enforceRecorder":true}`),
+				},
+			},
+			wantHeaders: http.Header{
+				"Impersonate-Group": {"group1"},
+				"Impersonate-User":  {"node.ts.net"},
+			},
+		},
 		{
 			name:     "bad-cap",
 			emailish: "tagged-device",
 			tags:     []string{"tag:foo", "tag:bar"},
 			capMap: tailcfg.PeerCapMap{
-				capabilityName: {
+				tailcfg.PeerCapabilityKubernetes: {
 					tailcfg.RawMessage(`[]`),
 				},
 			},

cmd/k8s-operator/proxyclass.go

@@ -10,6 +10,7 @@ import (
 	"fmt"
 	"strings"
 
+	dockerref "github.com/distribution/reference"
 	"go.uber.org/zap"
 	corev1 "k8s.io/api/core/v1"
 	apiequality "k8s.io/apimachinery/pkg/api/equality"
@@ -111,6 +112,20 @@ func (a *ProxyClassReconciler) validate(pc *tsapi.ProxyClass) (violations field.
 						a.recorder.Event(pc, corev1.EventTypeWarning, reasonCustomTSEnvVar, fmt.Sprintf(messageCustomTSEnvVar, string(e.Name), "tailscale"))
 					}
 				}
+				if tc.Image != "" {
+					// Same validation as used by kubelet https://github.com/kubernetes/kubernetes/blob/release-1.30/pkg/kubelet/images/image_manager.go#L212
+					if _, err := dockerref.ParseNormalizedNamed(tc.Image); err != nil {
+						violations = append(violations, field.TypeInvalid(field.NewPath("spec", "statefulSet", "pod", "tailscaleContainer", "image"), tc.Image, err.Error()))
+					}
+				}
+			}
+			if tc := pod.TailscaleInitContainer; tc != nil {
+				if tc.Image != "" {
+					// Same validation as used by kubelet https://github.com/kubernetes/kubernetes/blob/release-1.30/pkg/kubelet/images/image_manager.go#L212
+					if _, err := dockerref.ParseNormalizedNamed(tc.Image); err != nil {
+						violations = append(violations, field.TypeInvalid(field.NewPath("spec", "statefulSet", "pod", "tailscaleInitContainer", "image"), tc.Image, err.Error()))
+					}
+				}
 			}
 		}
 	}

cmd/k8s-operator/proxyclass_test.go

@@ -36,9 +36,13 @@ func TestProxyClass(t *testing.T) {
 				Labels:      map[string]string{"foo": "bar", "xyz1234": "abc567"},
 				Annotations: map[string]string{"foo.io/bar": "{'key': 'val1232'}"},
 				Pod: &tsapi.Pod{
-					Labels:             map[string]string{"foo": "bar", "xyz1234": "abc567"},
-					Annotations:        map[string]string{"foo.io/bar": "{'key': 'val1232'}"},
-					TailscaleContainer: &tsapi.Container{Env: []tsapi.Env{{Name: "FOO", Value: "BAR"}}},
+					Labels:      map[string]string{"foo": "bar", "xyz1234": "abc567"},
+					Annotations: map[string]string{"foo.io/bar": "{'key': 'val1232'}"},
+					TailscaleContainer: &tsapi.Container{
+						Env:             []tsapi.Env{{Name: "FOO", Value: "BAR"}},
+						ImagePullPolicy: "IfNotPresent",
+						Image:           "ghcr.my-repo/tailscale:v0.01testsomething",
+					},
 				},
 			},
 		},
@@ -63,17 +67,17 @@ func TestProxyClass(t *testing.T) {
 
 	// 1. A valid ProxyClass resource gets its status updated to Ready.
 	expectReconciled(t, pcr, "", "test")
-	pc.Status.Conditions = append(pc.Status.Conditions, tsapi.ConnectorCondition{
-		Type:               tsapi.ProxyClassready,
+	pc.Status.Conditions = append(pc.Status.Conditions, metav1.Condition{
+		Type:               string(tsapi.ProxyClassready),
 		Status:             metav1.ConditionTrue,
 		Reason:             reasonProxyClassValid,
 		Message:            reasonProxyClassValid,
-		LastTransitionTime: &metav1.Time{Time: cl.Now().Truncate(time.Second)},
+		LastTransitionTime: metav1.Time{Time: cl.Now().Truncate(time.Second)},
 	})
 
 	expectEqual(t, fc, pc, nil)
 
-	// 2. An invalid ProxyClass resource gets its status updated to Invalid.
+	// 2. A ProxyClass resource with invalid labels gets its status updated to Invalid with an error message.
 	pc.Spec.StatefulSet.Labels["foo"] = "?!someVal"
 	mustUpdate(t, fc, "", "test", func(proxyClass *tsapi.ProxyClass) {
 		proxyClass.Spec.StatefulSet.Labels = pc.Spec.StatefulSet.Labels
@@ -85,9 +89,43 @@ func TestProxyClass(t *testing.T) {
 	expectedEvent := "Warning ProxyClassInvalid ProxyClass is not valid: .spec.statefulSet.labels: Invalid value: \"?!someVal\": a valid label must be an empty string or consist of alphanumeric characters, '-', '_' or '.', and must start and end with an alphanumeric character (e.g. 'MyValue',  or 'my_value',  or '12345', regex used for validation is '(([A-Za-z0-9][-A-Za-z0-9_.]*)?[A-Za-z0-9])?')"
 	expectEvents(t, fr, []string{expectedEvent})
 
-	// 2. An valid ProxyClass but with a Tailscale env vars set results in warning events.
+	// 3. A ProxyClass resource with invalid image reference gets it status updated to Invalid with an error message.
+	pc.Spec.StatefulSet.Labels = nil
+	pc.Spec.StatefulSet.Pod.TailscaleContainer.Image = "FOO bar"
 	mustUpdate(t, fc, "", "test", func(proxyClass *tsapi.ProxyClass) {
-		proxyClass.Spec.StatefulSet.Labels = nil // unset invalid labels from the previous test
+		proxyClass.Spec.StatefulSet.Labels = nil
+		proxyClass.Spec.StatefulSet.Pod.TailscaleContainer.Image = pc.Spec.StatefulSet.Pod.TailscaleContainer.Image
+	})
+	expectReconciled(t, pcr, "", "test")
+	msg = `ProxyClass is not valid: spec.statefulSet.pod.tailscaleContainer.image: Invalid value: "FOO bar": invalid reference format: repository name (library/FOO bar) must be lowercase`
+	tsoperator.SetProxyClassCondition(pc, tsapi.ProxyClassready, metav1.ConditionFalse, reasonProxyClassInvalid, msg, 0, cl, zl.Sugar())
+	expectEqual(t, fc, pc, nil)
+	expectedEvent = `Warning ProxyClassInvalid ProxyClass is not valid: spec.statefulSet.pod.tailscaleContainer.image: Invalid value: "FOO bar": invalid reference format: repository name (library/FOO bar) must be lowercase`
+	expectEvents(t, fr, []string{expectedEvent})
+
+	// 4. A ProxyClass resource with invalid init container image reference gets it status updated to Invalid with an error message.
+	pc.Spec.StatefulSet.Labels = nil
+	pc.Spec.StatefulSet.Pod.TailscaleContainer.Image = ""
+	pc.Spec.StatefulSet.Pod.TailscaleInitContainer = &tsapi.Container{
+		Image: "FOO bar",
+	}
+	mustUpdate(t, fc, "", "test", func(proxyClass *tsapi.ProxyClass) {
+		proxyClass.Spec.StatefulSet.Pod.TailscaleContainer.Image = pc.Spec.StatefulSet.Pod.TailscaleContainer.Image
+		proxyClass.Spec.StatefulSet.Pod.TailscaleInitContainer = &tsapi.Container{
+			Image: pc.Spec.StatefulSet.Pod.TailscaleInitContainer.Image,
+		}
+	})
+	expectReconciled(t, pcr, "", "test")
+	msg = `ProxyClass is not valid: spec.statefulSet.pod.tailscaleInitContainer.image: Invalid value: "FOO bar": invalid reference format: repository name (library/FOO bar) must be lowercase`
+	tsoperator.SetProxyClassCondition(pc, tsapi.ProxyClassready, metav1.ConditionFalse, reasonProxyClassInvalid, msg, 0, cl, zl.Sugar())
+	expectEqual(t, fc, pc, nil)
+	expectedEvent = `Warning ProxyClassInvalid ProxyClass is not valid: spec.statefulSet.pod.tailscaleInitContainer.image: Invalid value: "FOO bar": invalid reference format: repository name (library/FOO bar) must be lowercase`
+	expectEvents(t, fr, []string{expectedEvent})
+
+	// 5. An valid ProxyClass but with a Tailscale env vars set results in warning events.
+	pc.Spec.StatefulSet.Pod.TailscaleInitContainer.Image = "" // unset previous test
+	mustUpdate(t, fc, "", "test", func(proxyClass *tsapi.ProxyClass) {
+		proxyClass.Spec.StatefulSet.Pod.TailscaleInitContainer.Image = pc.Spec.StatefulSet.Pod.TailscaleInitContainer.Image
 		proxyClass.Spec.StatefulSet.Pod.TailscaleContainer.Env = []tsapi.Env{{Name: "TS_USERSPACE", Value: "true"}, {Name: "EXPERIMENTAL_TS_CONFIGFILE_PATH"}, {Name: "EXPERIMENTAL_ALLOW_PROXYING_CLUSTER_TRAFFIC_VIA_INGRESS"}}
 	})
 	expectedEvents := []string{"Warning CustomTSEnvVar ProxyClass overrides the default value for TS_USERSPACE env var for tailscale container. Running with custom values for Tailscale env vars is not recommended and might break in the future.",

cmd/k8s-operator/sts.go

@@ -35,7 +35,6 @@ import (
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/opt"
 	"tailscale.com/types/ptr"
-	"tailscale.com/util/dnsname"
 	"tailscale.com/util/mak"
 )
 
@@ -92,10 +91,6 @@ 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 (
@@ -128,7 +123,9 @@ type tailscaleSTSConfig struct {
 	// what this StatefulSet should be created for.
 	Connector *connector
 
-	ProxyClass string
+	ProxyClassName string // name of ProxyClass if one needs to be applied to the proxy
+
+	ProxyClass *tsapi.ProxyClass // ProxyClass that needs to be applied to the proxy (if there is one)
 }
 
 type connector struct {
@@ -174,11 +171,23 @@ 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)
+	proxyClass := new(tsapi.ProxyClass)
+	if sts.ProxyClassName != "" {
+		if err := a.Get(ctx, types.NamespacedName{Name: sts.ProxyClassName}, proxyClass); err != nil {
+			return nil, fmt.Errorf("failed to get ProxyClass: %w", err)
+		}
+		if !tsoperator.ProxyClassIsReady(proxyClass) {
+			logger.Infof("ProxyClass %s specified for the proxy, but it is not (yet) in a ready state, waiting..")
+			return nil, nil
+		}
+	}
+	sts.ProxyClass = proxyClass
+
+	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 +300,7 @@ 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 +316,23 @@ 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 +343,58 @@ 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 := tsoperator.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**")
 	}
@@ -385,8 +405,10 @@ func sanitizeConfigBytes(bs []byte) string {
 	return string(sanitizedBytes)
 }
 
-// DeviceInfo returns the device ID and hostname for the Tailscale device
-// associated with the given labels.
+// DeviceInfo returns the device ID, hostname and IPs for the Tailscale device
+// that acts as an operator proxy. It retrieves info from a Kubernetes Secret
+// labeled with the provided labels.
+// Either of device ID, hostname and IPs can be empty string if not found in the Secret.
 func (a *tailscaleSTSReconciler) DeviceInfo(ctx context.Context, childLabels map[string]string) (id tailcfg.StableNodeID, hostname string, ips []string, err error) {
 	sec, err := getSingleObject[corev1.Secret](ctx, a.Client, a.operatorNamespace, childLabels)
 	if err != nil {
@@ -403,7 +425,12 @@ func (a *tailscaleSTSReconciler) DeviceInfo(ctx context.Context, childLabels map
 	// to remove it.
 	hostname = strings.TrimSuffix(string(sec.Data["device_fqdn"]), ".")
 	if hostname == "" {
-		return "", "", nil, nil
+		// Device ID gets stored and retrieved in a different flow than
+		// FQDN and IPs. A device that acts as Kubernetes operator
+		// proxy, but whose route setup has failed might have an device
+		// ID, but no FQDN/IPs. If so, return the ID, to allow the
+		// operator to clean up such devices.
+		return id, "", nil, nil
 	}
 	if rawDeviceIPs, ok := sec.Data["device_ips"]; ok {
 		if err := json.Unmarshal(rawDeviceIPs, &ips); err != nil {
@@ -437,7 +464,7 @@ 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 {
@@ -457,16 +484,6 @@ func (a *tailscaleSTSReconciler) reconcileSTS(ctx context.Context, logger *zap.S
 	}
 	pod := &ss.Spec.Template
 	container := &pod.Spec.Containers[0]
-	proxyClass := new(tsapi.ProxyClass)
-	if sts.ProxyClass != "" {
-		if err := a.Get(ctx, types.NamespacedName{Name: sts.ProxyClass}, proxyClass); err != nil {
-			return nil, fmt.Errorf("failed to get ProxyClass: %w", err)
-		}
-		if !tsoperator.ProxyClassIsReady(proxyClass) {
-			logger.Infof("ProxyClass %s specified for the proxy, but it is not (yet) in a ready state, waiting..")
-			return nil, nil
-		}
-	}
 	container.Image = a.proxyImage
 	ss.ObjectMeta = metav1.ObjectMeta{
 		Name:      headlessSvc.Name,
@@ -493,9 +510,15 @@ 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 +528,16 @@ 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,
@@ -571,18 +592,15 @@ 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"}},
 				},
 			},
 		})
 	}
 	logger.Debugf("reconciling statefulset %s/%s", ss.GetNamespace(), ss.GetName())
-	if sts.ProxyClass != "" {
-		logger.Debugf("configuring proxy resources with ProxyClass %s", sts.ProxyClass)
-		ss = applyProxyClassToStatefulSet(proxyClass, ss, sts, logger)
+	if sts.ProxyClassName != "" {
+		logger.Debugf("configuring proxy resources with ProxyClass %s", sts.ProxyClassName)
+		ss = applyProxyClassToStatefulSet(sts.ProxyClass, ss, sts, logger)
 	}
 	updateSS := func(s *appsv1.StatefulSet) {
 		s.Spec = ss.Spec
@@ -682,6 +700,12 @@ func applyProxyClassToStatefulSet(pc *tsapi.ProxyClass, ss *appsv1.StatefulSet,
 			// in the env var list overrides an earlier one.
 			base.Env = append(base.Env, corev1.EnvVar{Name: string(e.Name), Value: e.Value})
 		}
+		if overlay.Image != "" {
+			base.Image = overlay.Image
+		}
+		if overlay.ImagePullPolicy != "" {
+			base.ImagePullPolicy = overlay.ImagePullPolicy
+		}
 		return base
 	}
 	for i, c := range ss.Spec.Template.Spec.Containers {
@@ -716,42 +740,90 @@ 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 shouldAcceptRoutes(stsC.ProxyClass) {
+		conf.AcceptRoutes = "true"
+	}
+
 	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 := tsoperator.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
+}
+
+func shouldAcceptRoutes(pc *tsapi.ProxyClass) bool {
+	return pc != nil && pc.Spec.TailscaleConfig != nil && pc.Spec.TailscaleConfig.AcceptRoutes
 }
 
 // ptrObject is a type constraint for pointer types that implement
@@ -761,7 +833,9 @@ 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 +844,13 @@ 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
@@ -869,14 +946,11 @@ func defaultEnv(envName, defVal string) string {
 	return v
 }
 
-func nameForService(svc *corev1.Service) (string, error) {
+func nameForService(svc *corev1.Service) string {
 	if h, ok := svc.Annotations[AnnotationHostname]; ok {
-		if err := dnsname.ValidLabel(h); err != nil {
-			return "", fmt.Errorf("invalid Tailscale hostname %q: %w", h, err)
-		}
-		return h, nil
+		return h
 	}
-	return svc.Namespace + "-" + svc.Name, nil
+	return svc.Namespace + "-" + svc.Name
 }
 
 func isValidFirewallMode(m string) bool {

cmd/k8s-operator/sts_test.go

@@ -81,7 +81,9 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 							Limits:   corev1.ResourceList{corev1.ResourceCPU: resource.MustParse("1000m"), corev1.ResourceMemory: resource.MustParse("128Mi")},
 							Requests: corev1.ResourceList{corev1.ResourceCPU: resource.MustParse("500m"), corev1.ResourceMemory: resource.MustParse("64Mi")},
 						},
-						Env: []tsapi.Env{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}},
+						Env:             []tsapi.Env{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}},
+						ImagePullPolicy: "IfNotPresent",
+						Image:           "ghcr.io/my-repo/tailscale:v0.01testsomething",
 					},
 					TailscaleInitContainer: &tsapi.Container{
 						SecurityContext: &corev1.SecurityContext{
@@ -92,7 +94,9 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 							Limits:   corev1.ResourceList{corev1.ResourceCPU: resource.MustParse("1000m"), corev1.ResourceMemory: resource.MustParse("128Mi")},
 							Requests: corev1.ResourceList{corev1.ResourceCPU: resource.MustParse("500m"), corev1.ResourceMemory: resource.MustParse("64Mi")},
 						},
-						Env: []tsapi.Env{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}},
+						Env:             []tsapi.Env{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}},
+						ImagePullPolicy: "IfNotPresent",
+						Image:           "ghcr.io/my-repo/tailscale:v0.01testsomething",
 					},
 				},
 			},
@@ -135,10 +139,12 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	env := []corev1.EnvVar{{Name: "TS_HOSTNAME", Value: "nginx"}}
 	userspaceProxySS.Labels = labels
 	userspaceProxySS.Annotations = annots
+	userspaceProxySS.Spec.Template.Spec.Containers[0].Image = "tailscale/tailscale:v0.0.1"
 	userspaceProxySS.Spec.Template.Spec.Containers[0].Env = env
 	nonUserspaceProxySS.ObjectMeta.Labels = labels
 	nonUserspaceProxySS.ObjectMeta.Annotations = annots
 	nonUserspaceProxySS.Spec.Template.Spec.Containers[0].Env = env
+	nonUserspaceProxySS.Spec.Template.Spec.InitContainers[0].Image = "tailscale/tailscale:v0.0.1"
 
 	// 1. Test that a ProxyClass with all fields set gets correctly applied
 	// to a Statefulset built from non-userspace proxy template.
@@ -159,6 +165,10 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.Spec.Template.Spec.InitContainers[0].Resources = proxyClassAllOpts.Spec.StatefulSet.Pod.TailscaleInitContainer.Resources
 	wantSS.Spec.Template.Spec.InitContainers[0].Env = append(wantSS.Spec.Template.Spec.InitContainers[0].Env, []corev1.EnvVar{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}}...)
 	wantSS.Spec.Template.Spec.Containers[0].Env = append(wantSS.Spec.Template.Spec.Containers[0].Env, []corev1.EnvVar{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}}...)
+	wantSS.Spec.Template.Spec.Containers[0].Image = "ghcr.io/my-repo/tailscale:v0.01testsomething"
+	wantSS.Spec.Template.Spec.Containers[0].ImagePullPolicy = "IfNotPresent"
+	wantSS.Spec.Template.Spec.InitContainers[0].Image = "ghcr.io/my-repo/tailscale:v0.01testsomething"
+	wantSS.Spec.Template.Spec.InitContainers[0].ImagePullPolicy = "IfNotPresent"
 
 	gotSS := applyProxyClassToStatefulSet(proxyClassAllOpts, nonUserspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
 	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
@@ -194,9 +204,11 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	wantSS.Spec.Template.Spec.Containers[0].SecurityContext = proxyClassAllOpts.Spec.StatefulSet.Pod.TailscaleContainer.SecurityContext
 	wantSS.Spec.Template.Spec.Containers[0].Resources = proxyClassAllOpts.Spec.StatefulSet.Pod.TailscaleContainer.Resources
 	wantSS.Spec.Template.Spec.Containers[0].Env = append(wantSS.Spec.Template.Spec.Containers[0].Env, []corev1.EnvVar{{Name: "foo", Value: "bar"}, {Name: "TS_USERSPACE", Value: "true"}, {Name: "bar"}}...)
+	wantSS.Spec.Template.Spec.Containers[0].ImagePullPolicy = "IfNotPresent"
+	wantSS.Spec.Template.Spec.Containers[0].Image = "ghcr.io/my-repo/tailscale:v0.01testsomething"
 	gotSS = applyProxyClassToStatefulSet(proxyClassAllOpts, userspaceProxySS.DeepCopy(), new(tailscaleSTSConfig), zl.Sugar())
 	if diff := cmp.Diff(gotSS, wantSS); diff != "" {
-		t.Fatalf("Unexpected result applying ProxyClass with custom labels and annotations to a StatefulSet for a userspace proxy (-got +want):\n%s", diff)
+		t.Fatalf("Unexpected result applying ProxyClass with all options to a StatefulSet for a userspace proxy (-got +want):\n%s", diff)
 	}
 
 	// 4. Test that a ProxyClass with custom labels and annotations gets correctly applied

cmd/k8s-operator/svc.go

@@ -7,6 +7,7 @@ package main
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"net/netip"
 	"slices"
@@ -15,7 +16,9 @@ import (
 
 	"go.uber.org/zap"
 	corev1 "k8s.io/api/core/v1"
+	apiequality "k8s.io/apimachinery/pkg/api/equality"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/client-go/tools/record"
 	"sigs.k8s.io/controller-runtime/pkg/client"
@@ -23,13 +26,20 @@ import (
 	tsoperator "tailscale.com/k8s-operator"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
 	"tailscale.com/net/dns/resolvconffile"
+	"tailscale.com/tstime"
 	"tailscale.com/util/clientmetric"
+	"tailscale.com/util/dnsname"
 	"tailscale.com/util/set"
 )
 
 const (
 	resolvConfPath       = "/etc/resolv.conf"
 	defaultClusterDomain = "cluster.local"
+
+	reasonProxyCreated = "ProxyCreated"
+	reasonProxyInvalid = "ProxyInvalid"
+	reasonProxyFailed  = "ProxyFailed"
+	reasonProxyPending = "ProxyPending"
 )
 
 type ServiceReconciler struct {
@@ -50,6 +60,8 @@ type ServiceReconciler struct {
 	recorder record.EventRecorder
 
 	tsNamespace string
+
+	clock tstime.Clock
 }
 
 var (
@@ -76,6 +88,12 @@ func childResourceLabels(name, ns, typ string) map[string]string {
 	}
 }
 
+func (a *ServiceReconciler) isTailscaleService(svc *corev1.Service) bool {
+	targetIP := tailnetTargetAnnotation(svc)
+	targetFQDN := svc.Annotations[AnnotationTailnetTargetFQDN]
+	return a.shouldExpose(svc) || targetIP != "" || targetFQDN != ""
+}
+
 func (a *ServiceReconciler) Reconcile(ctx context.Context, req reconcile.Request) (_ reconcile.Result, err error) {
 	logger := a.logger.With("service-ns", req.Namespace, "service-name", req.Name)
 	logger.Debugf("starting reconcile")
@@ -90,9 +108,8 @@ func (a *ServiceReconciler) Reconcile(ctx context.Context, req reconcile.Request
 	} else if err != nil {
 		return reconcile.Result{}, fmt.Errorf("failed to get svc: %w", err)
 	}
-	targetIP := tailnetTargetAnnotation(svc)
-	targetFQDN := svc.Annotations[AnnotationTailnetTargetFQDN]
-	if !svc.DeletionTimestamp.IsZero() || !a.shouldExpose(svc) && targetIP == "" && targetFQDN == "" {
+
+	if !svc.DeletionTimestamp.IsZero() || !a.isTailscaleService(svc) {
 		logger.Debugf("service is being deleted or is (no longer) referring to Tailscale ingress/egress, ensuring any created resources are cleaned up")
 		return reconcile.Result{}, a.maybeCleanup(ctx, logger, svc)
 	}
@@ -104,7 +121,14 @@ func (a *ServiceReconciler) Reconcile(ctx context.Context, req reconcile.Request
 //
 // This function is responsible for removing the finalizer from the service,
 // once all associated resources are gone.
-func (a *ServiceReconciler) maybeCleanup(ctx context.Context, logger *zap.SugaredLogger, svc *corev1.Service) error {
+func (a *ServiceReconciler) maybeCleanup(ctx context.Context, logger *zap.SugaredLogger, svc *corev1.Service) (err error) {
+	oldSvcStatus := svc.Status.DeepCopy()
+	defer func() {
+		if !apiequality.Semantic.DeepEqual(oldSvcStatus, svc.Status) {
+			// An error encountered here should get returned by the Reconcile function.
+			err = errors.Join(err, a.Client.Status().Update(ctx, svc))
+		}
+	}()
 	ix := slices.Index(svc.Finalizers, FinalizerName)
 	if ix < 0 {
 		logger.Debugf("no finalizer, nothing to do")
@@ -114,6 +138,10 @@ func (a *ServiceReconciler) maybeCleanup(ctx context.Context, logger *zap.Sugare
 		a.managedEgressProxies.Remove(svc.UID)
 		gaugeIngressProxies.Set(int64(a.managedIngressProxies.Len()))
 		gaugeEgressProxies.Set(int64(a.managedEgressProxies.Len()))
+
+		if !a.isTailscaleService(svc) {
+			tsoperator.RemoveServiceCondition(svc, tsapi.ProxyReady)
+		}
 		return nil
 	}
 
@@ -133,7 +161,7 @@ func (a *ServiceReconciler) maybeCleanup(ctx context.Context, logger *zap.Sugare
 	// exactly once at the very end of cleanup, because the final step of
 	// cleanup removes the tailscale finalizer, which will make all future
 	// reconciles exit early.
-	logger.Infof("unexposed service from tailnet")
+	logger.Infof("unexposed Service from tailnet")
 
 	a.mu.Lock()
 	defer a.mu.Unlock()
@@ -141,6 +169,10 @@ func (a *ServiceReconciler) maybeCleanup(ctx context.Context, logger *zap.Sugare
 	a.managedEgressProxies.Remove(svc.UID)
 	gaugeIngressProxies.Set(int64(a.managedIngressProxies.Len()))
 	gaugeEgressProxies.Set(int64(a.managedEgressProxies.Len()))
+
+	if !a.isTailscaleService(svc) {
+		tsoperator.RemoveServiceCondition(svc, tsapi.ProxyReady)
+	}
 	return nil
 }
 
@@ -149,7 +181,15 @@ func (a *ServiceReconciler) maybeCleanup(ctx context.Context, logger *zap.Sugare
 //
 // This function adds a finalizer to svc, ensuring that we can handle orderly
 // deprovisioning later.
-func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.SugaredLogger, svc *corev1.Service) error {
+func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.SugaredLogger, svc *corev1.Service) (err error) {
+	oldSvcStatus := svc.Status.DeepCopy()
+	defer func() {
+		if !apiequality.Semantic.DeepEqual(oldSvcStatus, svc.Status) {
+			// An error encountered here should get returned by the Reconcile function.
+			err = errors.Join(err, a.Client.Status().Update(ctx, svc))
+		}
+	}()
+
 	// Run for proxy config related validations here as opposed to running
 	// them earlier. This is to prevent cleanup being blocked on a
 	// misconfigured proxy param.
@@ -157,30 +197,31 @@ func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 		msg := fmt.Sprintf("unable to provision proxy resources: invalid config: %v", err)
 		a.recorder.Event(svc, corev1.EventTypeWarning, "INVALIDCONFIG", msg)
 		a.logger.Error(msg)
+		tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyInvalid, msg, a.clock, logger)
 		return nil
 	}
 	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)
+		tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyInvalid, msg, a.clock, logger)
 		return nil
 	}
 
 	proxyClass := proxyClassForObject(svc)
 	if proxyClass != "" {
 		if ready, err := proxyClassIsReady(ctx, proxyClass, a.Client); err != nil {
-			return fmt.Errorf("error verifying ProxyClass for Service: %w", err)
+			errMsg := fmt.Errorf("error verifying ProxyClass for Service: %w", err)
+			tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyFailed, errMsg.Error(), a.clock, logger)
+			return errMsg
 		} else if !ready {
-			logger.Infof("ProxyClass %s specified for the Service, but is not (yet) Ready, waiting..", proxyClass)
+			msg := fmt.Sprintf("ProxyClass %s specified for the Service, but is not (yet) Ready, waiting..", proxyClass)
+			tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyPending, msg, a.clock, logger)
+			logger.Info(msg)
 			return nil
 		}
 	}
 
-	hostname, err := nameForService(svc)
-	if err != nil {
-		return err
-	}
-
 	if !slices.Contains(svc.Finalizers, FinalizerName) {
 		// This log line is printed exactly once during initial provisioning,
 		// because once the finalizer is in place this block gets skipped. So,
@@ -189,7 +230,9 @@ func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 		logger.Infof("exposing service over tailscale")
 		svc.Finalizers = append(svc.Finalizers, FinalizerName)
 		if err := a.Update(ctx, svc); err != nil {
-			return fmt.Errorf("failed to add finalizer: %w", err)
+			errMsg := fmt.Errorf("failed to add finalizer: %w", err)
+			tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyFailed, errMsg.Error(), a.clock, logger)
+			return errMsg
 		}
 	}
 	crl := childResourceLabels(svc.Name, svc.Namespace, "svc")
@@ -201,10 +244,10 @@ func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 	sts := &tailscaleSTSConfig{
 		ParentResourceName:  svc.Name,
 		ParentResourceUID:   string(svc.UID),
-		Hostname:            hostname,
+		Hostname:            nameForService(svc),
 		Tags:                tags,
 		ChildResourceLabels: crl,
-		ProxyClass:          proxyClass,
+		ProxyClassName:      proxyClass,
 	}
 
 	a.mu.Lock()
@@ -233,10 +276,12 @@ func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 
 	var hsvc *corev1.Service
 	if hsvc, err = a.ssr.Provision(ctx, logger, sts); err != nil {
-		return fmt.Errorf("failed to provision: %w", err)
+		errMsg := fmt.Errorf("failed to provision: %w", err)
+		tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyFailed, errMsg.Error(), a.clock, logger)
+		return errMsg
 	}
 
-	if sts.TailnetTargetIP != "" || sts.TailnetTargetFQDN != "" {
+	if sts.TailnetTargetIP != "" || sts.TailnetTargetFQDN != "" { // if an egress proxy
 		clusterDomain := retrieveClusterDomain(a.tsNamespace, logger)
 		headlessSvcName := hsvc.Name + "." + hsvc.Namespace + ".svc." + clusterDomain
 		if svc.Spec.ExternalName != headlessSvcName || svc.Spec.Type != corev1.ServiceTypeExternalName {
@@ -244,14 +289,18 @@ func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 			svc.Spec.Selector = nil
 			svc.Spec.Type = corev1.ServiceTypeExternalName
 			if err := a.Update(ctx, svc); err != nil {
-				return fmt.Errorf("failed to update service: %w", err)
+				errMsg := fmt.Errorf("failed to update service: %w", err)
+				tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyFailed, errMsg.Error(), a.clock, logger)
+				return errMsg
 			}
 		}
+		tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionTrue, reasonProxyCreated, reasonProxyCreated, a.clock, logger)
 		return nil
 	}
 
 	if !isTailscaleLoadBalancerService(svc, a.isDefaultLoadBalancer) {
 		logger.Debugf("service is not a LoadBalancer, so not updating ingress")
+		tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionTrue, reasonProxyCreated, reasonProxyCreated, a.clock, logger)
 		return nil
 	}
 
@@ -260,22 +309,23 @@ func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 		return fmt.Errorf("failed to get device ID: %w", err)
 	}
 	if tsHost == "" {
-		logger.Debugf("no Tailscale hostname known yet, waiting for proxy pod to finish auth")
+		msg := "no Tailscale hostname known yet, waiting for proxy pod to finish auth"
+		logger.Debug(msg)
 		// No hostname yet. Wait for the proxy pod to auth.
 		svc.Status.LoadBalancer.Ingress = nil
-		if err := a.Status().Update(ctx, svc); err != nil {
-			return fmt.Errorf("failed to update service status: %w", err)
-		}
+		tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyPending, msg, a.clock, logger)
 		return nil
 	}
 
-	logger.Debugf("setting ingress to %q, %s", tsHost, strings.Join(tsIPs, ", "))
+	logger.Debugf("setting Service LoadBalancer status to %q, %s", tsHost, strings.Join(tsIPs, ", "))
 	ingress := []corev1.LoadBalancerIngress{
 		{Hostname: tsHost},
 	}
 	clusterIPAddr, err := netip.ParseAddr(svc.Spec.ClusterIP)
 	if err != nil {
-		return fmt.Errorf("failed to parse cluster IP: %w", err)
+		msg := fmt.Sprintf("failed to parse cluster IP: %v", err)
+		tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionFalse, reasonProxyFailed, msg, a.clock, logger)
+		return fmt.Errorf(msg)
 	}
 	for _, ip := range tsIPs {
 		addr, err := netip.ParseAddr(ip)
@@ -287,22 +337,28 @@ func (a *ServiceReconciler) maybeProvision(ctx context.Context, logger *zap.Suga
 		}
 	}
 	svc.Status.LoadBalancer.Ingress = ingress
-	if err := a.Status().Update(ctx, svc); err != nil {
-		return fmt.Errorf("failed to update service status: %w", err)
-	}
+	tsoperator.SetServiceCondition(svc, tsapi.ProxyReady, metav1.ConditionTrue, reasonProxyCreated, reasonProxyCreated, a.clock, logger)
 	return nil
 }
 
 func validateService(svc *corev1.Service) []string {
 	violations := make([]string, 0)
 	if svc.Annotations[AnnotationTailnetTargetFQDN] != "" && svc.Annotations[AnnotationTailnetTargetIP] != "" {
-		violations = append(violations, "only one of annotations %s and %s can be set", AnnotationTailnetTargetIP, AnnotationTailnetTargetFQDN)
+		violations = append(violations, fmt.Sprintf("only one of annotations %s and %s can be set", AnnotationTailnetTargetIP, AnnotationTailnetTargetFQDN))
 	}
 	if fqdn := svc.Annotations[AnnotationTailnetTargetFQDN]; fqdn != "" {
 		if !isMagicDNSName(fqdn) {
 			violations = append(violations, fmt.Sprintf("invalid value of annotation %s: %q does not appear to be a valid MagicDNS name", AnnotationTailnetTargetFQDN, fqdn))
 		}
 	}
+	svcName := nameForService(svc)
+	if err := dnsname.ValidLabel(svcName); err != nil {
+		if _, ok := svc.Annotations[AnnotationHostname]; ok {
+			violations = append(violations, fmt.Sprintf("invalid Tailscale hostname specified %q: %s", svcName, err))
+		} else {
+			violations = append(violations, fmt.Sprintf("invalid Tailscale hostname %q, use %q annotation to override: %s", svcName, AnnotationHostname, err))
+		}
+	}
 	return violations
 }
 
@@ -334,7 +390,7 @@ func hasExposeAnnotation(svc *corev1.Service) bool {
 	return svc != nil && svc.Annotations[AnnotationExpose] == "true"
 }
 
-// hasTailnetTargetAnnotation returns the value of tailscale.com/tailnet-ip
+// tailnetTargetAnnotation returns the value of tailscale.com/tailnet-ip
 // annotation or of the deprecated tailscale.com/ts-tailnet-target-ip
 // annotation. If neither is set, it returns an empty string. If both are set,
 // it returns the value of the new annotation.

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{
@@ -318,10 +304,6 @@ func expectedSTSUserspace(t *testing.T, cl client.Client, opts configOpts) *apps
 
 func expectedHeadlessService(name string, parentType string) *corev1.Service {
 	return &corev1.Service{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Service",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:         name,
 			GenerateName: "ts-test-",
@@ -342,13 +324,9 @@ func expectedHeadlessService(name string, parentType string) *corev1.Service {
 	}
 }
 
-func expectedSecret(t *testing.T, opts configOpts) *corev1.Secret {
+func expectedSecret(t *testing.T, cl client.Client, opts configOpts) *corev1.Secret {
 	t.Helper()
 	s := &corev1.Secret{
-		TypeMeta: metav1.TypeMeta{
-			Kind:       "Secret",
-			APIVersion: "v1",
-		},
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      opts.secretName,
 			Namespace: "operator-ns",
@@ -369,6 +347,16 @@ func expectedSecret(t *testing.T, opts configOpts) *corev1.Secret {
 		AuthKey:      ptr.To("secret-authkey"),
 		AcceptRoutes: "false",
 	}
+	if opts.proxyClass != "" {
+		t.Logf("applying configuration from ProxyClass %s", opts.proxyClass)
+		proxyClass := new(tsapi.ProxyClass)
+		if err := cl.Get(context.Background(), types.NamespacedName{Name: opts.proxyClass}, proxyClass); err != nil {
+			t.Fatalf("error getting ProxyClass: %v", err)
+		}
+		if proxyClass.Spec.TailscaleConfig != nil && proxyClass.Spec.TailscaleConfig.AcceptRoutes {
+			conf.AcceptRoutes = "true"
+		}
+	}
 	var routes []netip.Prefix
 	if opts.subnetRoutes != "" || opts.isExitNode {
 		r := opts.subnetRoutes
@@ -388,7 +376,17 @@ 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",
@@ -459,11 +457,11 @@ func mustUpdateStatus[T any, O ptrObject[T]](t *testing.T, client client.Client,
 
 // expectEqual accepts a Kubernetes object and a Kubernetes client. It tests
 // whether an object with equivalent contents can be retrieved by the passed
-// client. If you want to NOT test some object fields for equality, ensure that
-// 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)) {
+// client. If you want to NOT test some object fields for equality, use the
+// modify func to ensure that they are removed from the cluster object and the
+// object passed as 'want'. 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, modifier func(O)) {
 	t.Helper()
 	got := O(new(T))
 	if err := client.Get(context.Background(), types.NamespacedName{
@@ -477,8 +475,9 @@ 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(want)
+		modifier(got)
 	}
 	if diff := cmp.Diff(got, want); diff != "" {
 		t.Fatalf("unexpected object (-got +want):\n%s", diff)
@@ -612,3 +611,33 @@ func (c *fakeTSClient) Deleted() []string {
 func removeHashAnnotation(sts *appsv1.StatefulSet) {
 	delete(sts.Spec.Template.Annotations, podAnnotationLastSetConfigFileHash)
 }
+
+func removeAuthKeyIfExistsModifier(t *testing.T) func(s *corev1.Secret) {
+	return func(secret *corev1.Secret) {
+		t.Helper()
+		if len(secret.StringData["tailscaled"]) != 0 {
+			conf := &ipn.ConfigVAlpha{}
+			if err := json.Unmarshal([]byte(secret.StringData["tailscaled"]), conf); err != nil {
+				t.Fatalf("error unmarshalling 'tailscaled' contents: %v", err)
+			}
+			conf.AuthKey = nil
+			b, err := json.Marshal(conf)
+			if err != nil {
+				t.Fatalf("error marshalling updated 'tailscaled' config: %v", err)
+			}
+			mak.Set(&secret.StringData, "tailscaled", string(b))
+		}
+		if len(secret.StringData["cap-95.hujson"]) != 0 {
+			conf := &ipn.ConfigVAlpha{}
+			if err := json.Unmarshal([]byte(secret.StringData["cap-95.hujson"]), conf); err != nil {
+				t.Fatalf("error umarshalling 'cap-95.hujson' contents: %v", err)
+			}
+			conf.AuthKey = nil
+			b, err := json.Marshal(conf)
+			if err != nil {
+				t.Fatalf("error marshalling 'cap-95.huson' contents: %v", err)
+			}
+			mak.Set(&secret.StringData, "cap-95.hujson", string(b))
+		}
+	}
+}

cmd/natc/natc.go

@@ -0,0 +1,567 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+// The natc command is a work-in-progress implementation of a NAT based
+// connector for Tailscale. It is intended to be used to route traffic to a
+// specific domain through a specific node.
+package main
+
+import (
+	"context"
+	"encoding/binary"
+	"errors"
+	"flag"
+	"fmt"
+	"log"
+	"math/rand/v2"
+	"net"
+	"net/http"
+	"net/netip"
+	"os"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/gaissmai/bart"
+	"github.com/inetaf/tcpproxy"
+	"github.com/peterbourgon/ff/v3"
+	"golang.org/x/net/dns/dnsmessage"
+	"tailscale.com/client/tailscale"
+	"tailscale.com/envknob"
+	"tailscale.com/hostinfo"
+	"tailscale.com/ipn"
+	"tailscale.com/net/netutil"
+	"tailscale.com/syncs"
+	"tailscale.com/tailcfg"
+	"tailscale.com/tsnet"
+	"tailscale.com/tsweb"
+	"tailscale.com/util/dnsname"
+	"tailscale.com/util/mak"
+)
+
+func main() {
+	hostinfo.SetApp("natc")
+	if !envknob.UseWIPCode() {
+		log.Fatal("cmd/natc is a work in progress and has not been security reviewed;\nits use requires TAILSCALE_USE_WIP_CODE=1 be set in the environment for now.")
+	}
+
+	// Parse flags
+	fs := flag.NewFlagSet("natc", flag.ExitOnError)
+	var (
+		debugPort       = fs.Int("debug-port", 8893, "Listening port for debug/metrics endpoint")
+		hostname        = fs.String("hostname", "", "Hostname to register the service under")
+		siteID          = fs.Uint("site-id", 1, "an integer site ID to use for the ULA prefix which allows for multiple proxies to act in a HA configuration")
+		v4PfxStr        = fs.String("v4-pfx", "100.64.1.0/24", "comma-separated list of IPv4 prefixes to advertise")
+		verboseTSNet    = fs.Bool("verbose-tsnet", false, "enable verbose logging in tsnet")
+		printULA        = fs.Bool("print-ula", false, "print the ULA prefix and exit")
+		ignoreDstPfxStr = fs.String("ignore-destinations", "", "comma-separated list of prefixes to ignore")
+		wgPort          = fs.Uint("wg-port", 0, "udp port for wireguard and peer to peer traffic")
+	)
+	ff.Parse(fs, os.Args[1:], ff.WithEnvVarPrefix("TS_NATC"))
+
+	if *printULA {
+		fmt.Println(ula(uint16(*siteID)))
+		return
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	if *siteID == 0 {
+		log.Fatalf("site-id must be set")
+	} else if *siteID > 0xffff {
+		log.Fatalf("site-id must be in the range [0, 65535]")
+	}
+
+	var ignoreDstTable *bart.Table[bool]
+	for _, s := range strings.Split(*ignoreDstPfxStr, ",") {
+		s := strings.TrimSpace(s)
+		if s == "" {
+			continue
+		}
+		if ignoreDstTable == nil {
+			ignoreDstTable = &bart.Table[bool]{}
+		}
+		pfx, err := netip.ParsePrefix(s)
+		if err != nil {
+			log.Fatalf("unable to parse prefix: %v", err)
+		}
+		if pfx.Masked() != pfx {
+			log.Fatalf("prefix %v is not normalized (bits are set outside the mask)", pfx)
+		}
+		ignoreDstTable.Insert(pfx, true)
+	}
+	var v4Prefixes []netip.Prefix
+	for _, s := range strings.Split(*v4PfxStr, ",") {
+		p := netip.MustParsePrefix(strings.TrimSpace(s))
+		if p.Masked() != p {
+			log.Fatalf("v4 prefix %v is not a masked prefix", p)
+		}
+		v4Prefixes = append(v4Prefixes, p)
+	}
+	if len(v4Prefixes) == 0 {
+		log.Fatalf("no v4 prefixes specified")
+	}
+	dnsAddr := v4Prefixes[0].Addr()
+	ts := &tsnet.Server{
+		Hostname: *hostname,
+	}
+	if *wgPort != 0 {
+		if *wgPort >= 1<<16 {
+			log.Fatalf("wg-port must be in the range [0, 65535]")
+		}
+		ts.Port = uint16(*wgPort)
+	}
+	defer ts.Close()
+	if *verboseTSNet {
+		ts.Logf = log.Printf
+	}
+
+	// Start special-purpose listeners: dns, http promotion, debug server
+	if *debugPort != 0 {
+		mux := http.NewServeMux()
+		tsweb.Debugger(mux)
+		dln, err := ts.Listen("tcp", fmt.Sprintf(":%d", *debugPort))
+		if err != nil {
+			log.Fatalf("failed listening on debug port: %v", err)
+		}
+		defer dln.Close()
+		go func() {
+			log.Fatalf("debug serve: %v", http.Serve(dln, mux))
+		}()
+	}
+	lc, err := ts.LocalClient()
+	if err != nil {
+		log.Fatalf("LocalClient() failed: %v", err)
+	}
+	if _, err := ts.Up(ctx); err != nil {
+		log.Fatalf("ts.Up: %v", err)
+	}
+
+	c := &connector{
+		ts:         ts,
+		lc:         lc,
+		dnsAddr:    dnsAddr,
+		v4Ranges:   v4Prefixes,
+		v6ULA:      ula(uint16(*siteID)),
+		ignoreDsts: ignoreDstTable,
+	}
+	c.run(ctx)
+}
+
+type connector struct {
+	// ts is the tsnet.Server used to host the connector.
+	ts *tsnet.Server
+	// lc is the LocalClient used to interact with the tsnet.Server hosting this
+	// connector.
+	lc *tailscale.LocalClient
+
+	// dnsAddr is the IPv4 address to listen on for DNS requests. It is used to
+	// prevent the app connector from assigning it to a domain.
+	dnsAddr netip.Addr
+
+	// v4Ranges is the list of IPv4 ranges to advertise and assign addresses from.
+	// These are masked prefixes.
+	v4Ranges []netip.Prefix
+	// v6ULA is the ULA prefix used by the app connector to assign IPv6 addresses.
+	v6ULA netip.Prefix
+
+	perPeerMap syncs.Map[tailcfg.NodeID, *perPeerState]
+
+	// ignoreDsts is initialized at start up with the contents of --ignore-destinations (if none it is nil)
+	// It is never mutated, only used for lookups.
+	// Users who want to natc a DNS wildcard but not every address record in that domain can supply the
+	// exceptions in --ignore-destinations. When we receive a dns request we will look up the fqdn
+	// and if any of the ip addresses in response to the lookup match any 'ignore destinations' prefix we will
+	// return a dns response that contains the ip addresses we discovered with the lookup (ie not the
+	// natc behavior, which would return a dummy ip address pointing at natc).
+	ignoreDsts *bart.Table[bool]
+}
+
+// v6ULA is the ULA prefix used by the app connector to assign IPv6 addresses.
+// The 8th and 9th bytes are used to encode the site ID which allows for
+// multiple proxies to act in a HA configuration.
+// mnemonic: a99c = appc
+var v6ULA = netip.MustParsePrefix("fd7a:115c:a1e0:a99c::/64")
+
+func ula(siteID uint16) netip.Prefix {
+	as16 := v6ULA.Addr().As16()
+	as16[8] = byte(siteID >> 8)
+	as16[9] = byte(siteID)
+	return netip.PrefixFrom(netip.AddrFrom16(as16), 64+16)
+}
+
+// run runs the connector.
+//
+// The passed in context is only used for the initial setup. The connector runs
+// forever.
+func (c *connector) run(ctx context.Context) {
+	if _, err := c.lc.EditPrefs(ctx, &ipn.MaskedPrefs{
+		AdvertiseRoutesSet: true,
+		Prefs: ipn.Prefs{
+			AdvertiseRoutes: append(c.v4Ranges, c.v6ULA),
+		},
+	}); err != nil {
+		log.Fatalf("failed to advertise routes: %v", err)
+	}
+	c.ts.RegisterFallbackTCPHandler(c.handleTCPFlow)
+	c.serveDNS()
+}
+
+func (c *connector) serveDNS() {
+	pc, err := c.ts.ListenPacket("udp", net.JoinHostPort(c.dnsAddr.String(), "53"))
+	if err != nil {
+		log.Fatalf("failed listening on port 53: %v", err)
+	}
+	defer pc.Close()
+	log.Printf("Listening for DNS on %s", pc.LocalAddr().String())
+	for {
+		buf := make([]byte, 1500)
+		n, addr, err := pc.ReadFrom(buf)
+		if err != nil {
+			if errors.Is(err, net.ErrClosed) {
+				return
+			}
+			log.Printf("serveDNS.ReadFrom failed: %v", err)
+			continue
+		}
+		go c.handleDNS(pc, buf[:n], addr.(*net.UDPAddr))
+	}
+}
+
+func lookupDestinationIP(domain string) ([]netip.Addr, error) {
+	netIPs, err := net.LookupIP(domain)
+	if err != nil {
+		var dnsError *net.DNSError
+		if errors.As(err, &dnsError) && dnsError.IsNotFound {
+			return nil, nil
+		} else {
+			return nil, err
+		}
+	}
+	var addrs []netip.Addr
+	for _, ip := range netIPs {
+		a, ok := netip.AddrFromSlice(ip)
+		if ok {
+			addrs = append(addrs, a)
+		}
+	}
+	return addrs, nil
+}
+
+// handleDNS handles a DNS request to the app connector.
+// It generates a response based on the request and the node that sent it.
+//
+// Each node is assigned a unique pair of IP addresses for each domain it
+// queries. This assignment is done lazily and is not persisted across restarts.
+// A per-peer assignment allows the connector to reuse a limited number of IP
+// addresses across multiple nodes and domains. It also allows for clear
+// failover behavior when an app connector is restarted.
+//
+// This assignment later allows the connector to determine where to forward
+// traffic based on the destination IP address.
+func (c *connector) handleDNS(pc net.PacketConn, buf []byte, remoteAddr *net.UDPAddr) {
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	who, err := c.lc.WhoIs(ctx, remoteAddr.String())
+	if err != nil {
+		log.Printf("HandleDNS: WhoIs failed: %v\n", err)
+		return
+	}
+
+	var msg dnsmessage.Message
+	err = msg.Unpack(buf)
+	if err != nil {
+		log.Printf("HandleDNS: dnsmessage unpack failed: %v\n ", err)
+		return
+	}
+
+	// If there are destination ips that we don't want to route, we
+	// have to do a dns lookup here to find the destination ip.
+	if c.ignoreDsts != nil {
+		if len(msg.Questions) > 0 {
+			q := msg.Questions[0]
+			switch q.Type {
+			case dnsmessage.TypeAAAA, dnsmessage.TypeA:
+				dstAddrs, err := lookupDestinationIP(q.Name.String())
+				if err != nil {
+					log.Printf("HandleDNS: lookup destination failed: %v\n ", err)
+					return
+				}
+				if c.ignoreDestination(dstAddrs) {
+					bs, err := dnsResponse(&msg, dstAddrs)
+					// TODO (fran): treat as SERVFAIL
+					if err != nil {
+						log.Printf("HandleDNS: generate ignore response failed: %v\n", err)
+						return
+					}
+					_, err = pc.WriteTo(bs, remoteAddr)
+					if err != nil {
+						log.Printf("HandleDNS: write failed: %v\n", err)
+					}
+					return
+				}
+			}
+		}
+	}
+	// None of the destination IP addresses match an ignore destination prefix, do
+	// the natc thing.
+
+	resp, err := c.generateDNSResponse(&msg, who.Node.ID)
+	// TODO (fran): treat as SERVFAIL
+	if err != nil {
+		log.Printf("HandleDNS: connector handling failed: %v\n", err)
+		return
+	}
+	// TODO (fran): treat as NXDOMAIN
+	if len(resp) == 0 {
+		return
+	}
+	// This connector handled the DNS request
+	_, err = pc.WriteTo(resp, remoteAddr)
+	if err != nil {
+		log.Printf("HandleDNS: write failed: %v\n", err)
+	}
+}
+
+// tsMBox is the mailbox used in SOA records.
+// The convention is to replace the @ symbol with a dot.
+// So in this case, the mailbox is support.tailscale.com. with the trailing dot
+// to indicate that it is a fully qualified domain name.
+var tsMBox = dnsmessage.MustNewName("support.tailscale.com.")
+
+// generateDNSResponse generates a DNS response for the given request. The from
+// argument is the NodeID of the node that sent the request.
+func (c *connector) generateDNSResponse(req *dnsmessage.Message, from tailcfg.NodeID) ([]byte, error) {
+	pm, _ := c.perPeerMap.LoadOrStore(from, &perPeerState{c: c})
+	var addrs []netip.Addr
+	if len(req.Questions) > 0 {
+		switch req.Questions[0].Type {
+		case dnsmessage.TypeAAAA, dnsmessage.TypeA:
+			var err error
+			addrs, err = pm.ipForDomain(req.Questions[0].Name.String())
+			if err != nil {
+				return nil, err
+			}
+		}
+	}
+	return dnsResponse(req, addrs)
+}
+
+// dnsResponse makes a DNS response for the natc. If the dnsmessage is requesting TypeAAAA
+// or TypeA the provided addrs of the requested type will be used.
+func dnsResponse(req *dnsmessage.Message, addrs []netip.Addr) ([]byte, error) {
+	b := dnsmessage.NewBuilder(nil,
+		dnsmessage.Header{
+			ID:            req.Header.ID,
+			Response:      true,
+			Authoritative: true,
+		})
+	b.EnableCompression()
+
+	if len(req.Questions) == 0 {
+		return b.Finish()
+	}
+	q := req.Questions[0]
+	if err := b.StartQuestions(); err != nil {
+		return nil, err
+	}
+	if err := b.Question(q); err != nil {
+		return nil, err
+	}
+	if err := b.StartAnswers(); err != nil {
+		return nil, err
+	}
+	switch q.Type {
+	case dnsmessage.TypeAAAA, dnsmessage.TypeA:
+		want6 := q.Type == dnsmessage.TypeAAAA
+		for _, ip := range addrs {
+			if want6 != ip.Is6() {
+				continue
+			}
+			if want6 {
+				if err := b.AAAAResource(
+					dnsmessage.ResourceHeader{Name: q.Name, Class: q.Class, TTL: 5},
+					dnsmessage.AAAAResource{AAAA: ip.As16()},
+				); err != nil {
+					return nil, err
+				}
+			} else {
+				if err := b.AResource(
+					dnsmessage.ResourceHeader{Name: q.Name, Class: q.Class, TTL: 5},
+					dnsmessage.AResource{A: ip.As4()},
+				); err != nil {
+					return nil, err
+				}
+			}
+		}
+	case dnsmessage.TypeSOA:
+		if err := b.SOAResource(
+			dnsmessage.ResourceHeader{Name: q.Name, Class: q.Class, TTL: 120},
+			dnsmessage.SOAResource{NS: q.Name, MBox: tsMBox, Serial: 2023030600,
+				Refresh: 120, Retry: 120, Expire: 120, MinTTL: 60},
+		); err != nil {
+			return nil, err
+		}
+	case dnsmessage.TypeNS:
+		if err := b.NSResource(
+			dnsmessage.ResourceHeader{Name: q.Name, Class: q.Class, TTL: 120},
+			dnsmessage.NSResource{NS: tsMBox},
+		); err != nil {
+			return nil, err
+		}
+	}
+	return b.Finish()
+}
+
+// handleTCPFlow handles a TCP flow from the given source to the given
+// destination. It uses the source address to determine the node that sent the
+// request and the destination address to determine the domain that the request
+// is for based on the IP address assigned to the destination in the DNS
+// response.
+func (c *connector) handleTCPFlow(src, dst netip.AddrPort) (handler func(net.Conn), intercept bool) {
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	who, err := c.lc.WhoIs(ctx, src.Addr().String())
+	cancel()
+	if err != nil {
+		log.Printf("HandleTCPFlow: WhoIs failed: %v\n", err)
+		return nil, false
+	}
+
+	from := who.Node.ID
+	ps, ok := c.perPeerMap.Load(from)
+	if !ok {
+		log.Printf("handleTCPFlow: no perPeerState for %v", from)
+		return nil, false
+	}
+	domain, ok := ps.domainForIP(dst.Addr())
+	if !ok {
+		log.Printf("handleTCPFlow: no domain for IP %v\n", dst.Addr())
+		return nil, false
+	}
+	return func(conn net.Conn) {
+		proxyTCPConn(conn, domain)
+	}, true
+}
+
+// ignoreDestination reports whether any of the provided dstAddrs match the prefixes configured
+// in --ignore-destinations
+func (c *connector) ignoreDestination(dstAddrs []netip.Addr) bool {
+	for _, a := range dstAddrs {
+		if _, ok := c.ignoreDsts.Get(a); ok {
+			return true
+		}
+	}
+	return false
+}
+
+func proxyTCPConn(c net.Conn, dest string) {
+	addrPortStr := c.LocalAddr().String()
+	_, port, err := net.SplitHostPort(addrPortStr)
+	if err != nil {
+		log.Printf("tcpRoundRobinHandler.Handle: bogus addrPort %q", addrPortStr)
+		c.Close()
+		return
+	}
+
+	p := &tcpproxy.Proxy{
+		ListenFunc: func(net, laddr string) (net.Listener, error) {
+			return netutil.NewOneConnListener(c, nil), nil
+		},
+	}
+	p.AddRoute(addrPortStr, &tcpproxy.DialProxy{
+		Addr: fmt.Sprintf("%s:%s", dest, port),
+	})
+	p.Start()
+}
+
+// perPeerState holds the state for a single peer.
+type perPeerState struct {
+	c *connector
+
+	mu           sync.Mutex
+	domainToAddr map[string][]netip.Addr
+	addrToDomain *bart.Table[string]
+}
+
+// domainForIP returns the domain name assigned to the given IP address and
+// whether it was found.
+func (ps *perPeerState) domainForIP(ip netip.Addr) (_ string, ok bool) {
+	ps.mu.Lock()
+	defer ps.mu.Unlock()
+	return ps.addrToDomain.Get(ip)
+}
+
+// ipForDomain assigns a pair of unique IP addresses for the given domain and
+// returns them. The first address is an IPv4 address and the second is an IPv6
+// address. If the domain already has assigned addresses, it returns them.
+func (ps *perPeerState) ipForDomain(domain string) ([]netip.Addr, error) {
+	fqdn, err := dnsname.ToFQDN(domain)
+	if err != nil {
+		return nil, err
+	}
+	domain = fqdn.WithoutTrailingDot()
+
+	ps.mu.Lock()
+	defer ps.mu.Unlock()
+	if addrs, ok := ps.domainToAddr[domain]; ok {
+		return addrs, nil
+	}
+	addrs := ps.assignAddrsLocked(domain)
+	return addrs, nil
+}
+
+// isIPUsedLocked reports whether the given IP address is already assigned to a
+// domain.
+// ps.mu must be held.
+func (ps *perPeerState) isIPUsedLocked(ip netip.Addr) bool {
+	_, ok := ps.addrToDomain.Get(ip)
+	return ok
+}
+
+// unusedIPv4Locked returns an unused IPv4 address from the available ranges.
+func (ps *perPeerState) unusedIPv4Locked() netip.Addr {
+	// TODO: skip ranges that have been exhausted
+	for _, r := range ps.c.v4Ranges {
+		ip := randV4(r)
+		for r.Contains(ip) {
+			if !ps.isIPUsedLocked(ip) && ip != ps.c.dnsAddr {
+				return ip
+			}
+			ip = ip.Next()
+		}
+	}
+	return netip.Addr{}
+}
+
+// randV4 returns a random IPv4 address within the given prefix.
+func randV4(maskedPfx netip.Prefix) netip.Addr {
+	bits := 32 - maskedPfx.Bits()
+	randBits := rand.Uint32N(1 << uint(bits))
+
+	ip4 := maskedPfx.Addr().As4()
+	pn := binary.BigEndian.Uint32(ip4[:])
+	binary.BigEndian.PutUint32(ip4[:], randBits|pn)
+	return netip.AddrFrom4(ip4)
+}
+
+// assignAddrsLocked assigns a pair of unique IP addresses for the given domain
+// and returns them. The first address is an IPv4 address and the second is an
+// IPv6 address. It does not check if the domain already has assigned addresses.
+// ps.mu must be held.
+func (ps *perPeerState) assignAddrsLocked(domain string) []netip.Addr {
+	if ps.addrToDomain == nil {
+		ps.addrToDomain = &bart.Table[string]{}
+	}
+	v4 := ps.unusedIPv4Locked()
+	as16 := ps.c.v6ULA.Addr().As16()
+	as4 := v4.As4()
+	copy(as16[12:], as4[:])
+	v6 := netip.AddrFrom16(as16)
+	addrs := []netip.Addr{v4, v6}
+	mak.Set(&ps.domainToAddr, domain, addrs)
+	for _, a := range addrs {
+		ps.addrToDomain.Insert(netip.PrefixFrom(a, a.BitLen()), domain)
+	}
+	return addrs
+}

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/proxy-to-grafana/proxy-to-grafana.go

@@ -46,6 +46,7 @@ var (
 	backendAddr  = flag.String("backend-addr", "", "Address of the Grafana server served over HTTP, in host:port format. Typically localhost:nnnn.")
 	tailscaleDir = flag.String("state-dir", "./", "Alternate directory to use for Tailscale state storage. If empty, a default is used.")
 	useHTTPS     = flag.Bool("use-https", false, "Serve over HTTPS via your *.ts.net subdomain if enabled in Tailscale admin.")
+	loginServer  = flag.String("login-server", "", "URL to alternative control server. If empty, the default Tailscale control is used.")
 )
 
 func main() {
@@ -57,8 +58,9 @@ func main() {
 		log.Fatal("missing --backend-addr")
 	}
 	ts := &tsnet.Server{
-		Dir:      *tailscaleDir,
-		Hostname: *hostname,
+		Dir:        *tailscaleDir,
+		Hostname:   *hostname,
+		ControlURL: *loginServer,
 	}
 
 	// TODO(bradfitz,maisem): move this to a method on tsnet.Server probably.

cmd/sniproxy/handlers.go

@@ -7,7 +7,7 @@ import (
 	"context"
 	"fmt"
 	"log"
-	"math/rand"
+	"math/rand/v2"
 	"net"
 	"net/netip"
 	"slices"
@@ -47,7 +47,7 @@ func (h *tcpRoundRobinHandler) Handle(c net.Conn) {
 		return netutil.NewOneConnListener(c, nil), nil
 	}
 
-	dest := h.To[rand.Intn(len(h.To))]
+	dest := h.To[rand.IntN(len(h.To))]
 	dial := &tcpproxy.DialProxy{
 		Addr:        fmt.Sprintf("%s:%s", dest, port),
 		DialContext: h.DialContext,

cmd/sniproxy/sniproxy_test.go

@@ -8,6 +8,7 @@ import (
 	"encoding/json"
 	"flag"
 	"fmt"
+	"log"
 	"net"
 	"net/http/httptest"
 	"net/netip"
@@ -24,6 +25,7 @@ import (
 	"tailscale.com/tsnet"
 	"tailscale.com/tstest/integration"
 	"tailscale.com/tstest/integration/testcontrol"
+	"tailscale.com/tstest/nettest"
 	"tailscale.com/types/appctype"
 	"tailscale.com/types/ipproto"
 	"tailscale.com/types/key"
@@ -98,8 +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() })
 
@@ -111,6 +113,7 @@ func startNode(t *testing.T, ctx context.Context, controlURL, hostname string) (
 }
 
 func TestSNIProxyWithNetmapConfig(t *testing.T) {
+	nettest.SkipIfNoNetwork(t)
 	c, controlURL := startControl(t)
 	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
 	defer cancel()
@@ -189,6 +192,7 @@ func TestSNIProxyWithNetmapConfig(t *testing.T) {
 }
 
 func TestSNIProxyWithFlagConfig(t *testing.T) {
+	nettest.SkipIfNoNetwork(t)
 	_, controlURL := startControl(t)
 	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
 	defer cancel()

cmd/stunc/stunc.go

@@ -8,6 +8,7 @@ import (
 	"log"
 	"net"
 	"os"
+	"strconv"
 
 	"tailscale.com/net/stun"
 )
@@ -15,12 +16,20 @@ import (
 func main() {
 	log.SetFlags(0)
 
-	if len(os.Args) != 2 {
-		log.Fatalf("usage: %s <hostname>", os.Args[0])
+	if len(os.Args) < 2 || len(os.Args) > 3 {
+		log.Fatalf("usage: %s <hostname> [port]", os.Args[0])
 	}
 	host := os.Args[1]
+	port := "3478"
+	if len(os.Args) == 3 {
+		port = os.Args[2]
+	}
+	_, err := strconv.ParseUint(port, 10, 16)
+	if err != nil {
+		log.Fatalf("invalid port: %v", err)
+	}
 
-	uaddr, err := net.ResolveUDPAddr("udp", host+":3478")
+	uaddr, err := net.ResolveUDPAddr("udp", net.JoinHostPort(host, port))
 	if err != nil {
 		log.Fatal(err)
 	}

cmd/stund/depaware.txt

@@ -153,7 +153,7 @@ tailscale.com/cmd/stund dependencies: (generated by github.com/tailscale/depawar
         math/big                                                     from crypto/dsa+
         math/bits                                                    from compress/flate+
         math/rand                                                    from math/big+
-        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

cmd/stunstamp/api.go

@@ -0,0 +1,142 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package main
+
+import (
+	"compress/gzip"
+	"encoding/json"
+	"errors"
+	"net/http"
+	"net/url"
+	"strconv"
+	"strings"
+	"time"
+
+	sq "github.com/Masterminds/squirrel"
+)
+
+type api struct {
+	db  *db
+	mux *http.ServeMux
+}
+
+func newAPI(db *db) *api {
+	a := &api{
+		db: db,
+	}
+	mux := http.NewServeMux()
+	mux.HandleFunc("/query", a.query)
+	a.mux = mux
+	return a
+}
+
+type apiResult struct {
+	At         int    `json:"at"` // time.Time.Unix()
+	RegionID   int    `json:"regionID"`
+	Hostname   string `json:"hostname"`
+	Af         int    `json:"af"` // 4 or 6
+	Addr       string `json:"addr"`
+	Source     int    `json:"source"` // timestampSourceUserspace (0) or timestampSourceKernel (1)
+	StableConn bool   `json:"stableConn"`
+	DstPort    int    `json:"dstPort"`
+	RttNS      *int   `json:"rttNS"`
+}
+
+func getTimeBounds(vals url.Values) (from time.Time, to time.Time, err error) {
+	lastForm, ok := vals["last"]
+	if ok && len(lastForm) > 0 {
+		dur, err := time.ParseDuration(lastForm[0])
+		if err != nil {
+			return time.Time{}, time.Time{}, err
+		}
+		now := time.Now()
+		return now.Add(-dur), now, nil
+	}
+
+	fromForm, ok := vals["from"]
+	if ok && len(fromForm) > 0 {
+		fromUnixSec, err := strconv.Atoi(fromForm[0])
+		if err != nil {
+			return time.Time{}, time.Time{}, err
+		}
+		from = time.Unix(int64(fromUnixSec), 0)
+		toForm, ok := vals["to"]
+		if ok && len(toForm) > 0 {
+			toUnixSec, err := strconv.Atoi(toForm[0])
+			if err != nil {
+				return time.Time{}, time.Time{}, err
+			}
+			to = time.Unix(int64(toUnixSec), 0)
+		} else {
+			return time.Time{}, time.Time{}, errors.New("from specified without to")
+		}
+		return from, to, nil
+	}
+
+	// no time bounds specified, default to last 1h
+	now := time.Now()
+	return now.Add(-time.Hour), now, nil
+}
+
+func (a *api) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	a.mux.ServeHTTP(w, r)
+}
+
+func (a *api) query(w http.ResponseWriter, r *http.Request) {
+	err := r.ParseForm()
+	if err != nil {
+		http.Error(w, err.Error(), 500)
+		return
+	}
+	from, to, err := getTimeBounds(r.Form)
+	if err != nil {
+		http.Error(w, err.Error(), 500)
+		return
+	}
+
+	sb := sq.Select("at_unix", "region_id", "hostname", "af", "address", "timestamp_source", "stable_conn", "dst_port", "rtt_ns").From("rtt")
+	sb = sb.Where(sq.And{
+		sq.GtOrEq{"at_unix": from.Unix()},
+		sq.LtOrEq{"at_unix": to.Unix()},
+	})
+	query, args, err := sb.ToSql()
+	if err != nil {
+		return
+	}
+
+	rows, err := a.db.Query(query, args...)
+	if err != nil {
+		http.Error(w, err.Error(), 500)
+		return
+	}
+	results := make([]apiResult, 0)
+	for rows.Next() {
+		rtt := 0
+		result := apiResult{
+			RttNS: &rtt,
+		}
+		err = rows.Scan(&result.At, &result.RegionID, &result.Hostname, &result.Af, &result.Addr, &result.Source, &result.StableConn, &result.DstPort, &result.RttNS)
+		if err != nil {
+			http.Error(w, err.Error(), 500)
+			return
+		}
+		results = append(results, result)
+	}
+	if rows.Err() != nil {
+		http.Error(w, rows.Err().Error(), 500)
+		return
+	}
+	if strings.Contains(r.Header.Get("Accept-Encoding"), "gzip") {
+		gz := gzip.NewWriter(w)
+		defer gz.Close()
+		w.Header().Set("Content-Encoding", "gzip")
+		err = json.NewEncoder(gz).Encode(&results)
+	} else {
+		err = json.NewEncoder(w).Encode(&results)
+	}
+	if err != nil {
+		http.Error(w, err.Error(), 500)
+		return
+	}
+}

cmd/stunstamp/stunstamp.go

@@ -0,0 +1,886 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+// The stunstamp binary measures STUN round-trip latency with DERPs.
+package main
+
+import (
+	"bytes"
+	"cmp"
+	"context"
+	"encoding/json"
+	"errors"
+	"flag"
+	"fmt"
+	"io"
+	"log"
+	"math"
+	"math/rand/v2"
+	"net"
+	"net/http"
+	"net/netip"
+	"net/url"
+	"os"
+	"os/signal"
+	"slices"
+	"strconv"
+	"strings"
+	"sync"
+	"syscall"
+	"time"
+
+	"github.com/golang/snappy"
+	"github.com/prometheus/prometheus/prompb"
+	"tailscale.com/logtail/backoff"
+	"tailscale.com/net/stun"
+	"tailscale.com/tailcfg"
+)
+
+var (
+	flagDERPMap        = flag.String("derp-map", "https://login.tailscale.com/derpmap/default", "URL to DERP map")
+	flagOut            = flag.String("out", "", "output sqlite filename")
+	flagInterval       = flag.Duration("interval", time.Minute, "interval to probe at in time.ParseDuration() format")
+	flagAPI            = flag.String("api", "", "listen addr for HTTP API")
+	flagIPv6           = flag.Bool("ipv6", false, "probe IPv6 addresses")
+	flagRetention      = flag.Duration("retention", time.Hour*24*7, "sqlite retention period in time.ParseDuration() format")
+	flagRemoteWriteURL = flag.String("rw-url", "", "prometheus remote write URL")
+	flagInstance       = flag.String("instance", "", "instance label value; defaults to hostname if unspecified")
+	flagDstPorts       = flag.String("dst-ports", "", "comma-separated list of destination ports to monitor")
+)
+
+const (
+	minInterval       = time.Second
+	maxBufferDuration = time.Hour
+)
+
+func getDERPMap(ctx context.Context, url string) (*tailcfg.DERPMap, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", url, nil)
+	if err != nil {
+		return nil, err
+	}
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+	dm := tailcfg.DERPMap{}
+	err = json.NewDecoder(resp.Body).Decode(&dm)
+	if err != nil {
+		return nil, nil
+	}
+	return &dm, nil
+}
+
+type timestampSource int
+
+const (
+	timestampSourceUserspace timestampSource = iota
+	timestampSourceKernel
+)
+
+func (t timestampSource) String() string {
+	switch t {
+	case timestampSourceUserspace:
+		return "userspace"
+	case timestampSourceKernel:
+		return "kernel"
+	default:
+		return "unknown"
+	}
+}
+
+// resultKey contains the stable dimensions and their values for a given
+// timeseries, i.e. not time and not rtt/timeout.
+type resultKey struct {
+	meta            nodeMeta
+	timestampSource timestampSource
+	connStability   connStability
+	dstPort         int
+}
+
+type result struct {
+	key resultKey
+	at  time.Time
+	rtt *time.Duration // nil signifies failure, e.g. timeout
+}
+
+func measureRTT(conn io.ReadWriteCloser, dst *net.UDPAddr) (rtt time.Duration, err error) {
+	uconn, ok := conn.(*net.UDPConn)
+	if !ok {
+		return 0, fmt.Errorf("unexpected conn type: %T", conn)
+	}
+	err = uconn.SetReadDeadline(time.Now().Add(time.Second * 2))
+	if err != nil {
+		return 0, fmt.Errorf("error setting read deadline: %w", err)
+	}
+	txID := stun.NewTxID()
+	req := stun.Request(txID)
+	txAt := time.Now()
+	_, err = uconn.WriteToUDP(req, dst)
+	if err != nil {
+		return 0, fmt.Errorf("error writing to udp socket: %w", err)
+	}
+	b := make([]byte, 1460)
+	for {
+		n, err := uconn.Read(b)
+		rxAt := time.Now()
+		if err != nil {
+			return 0, fmt.Errorf("error reading from udp socket: %w", err)
+		}
+		gotTxID, _, err := stun.ParseResponse(b[:n])
+		if err != nil || gotTxID != txID {
+			continue
+		}
+		return rxAt.Sub(txAt), nil
+	}
+
+}
+
+func isTemporaryOrTimeoutErr(err error) bool {
+	if errors.Is(err, os.ErrDeadlineExceeded) || errors.Is(err, context.DeadlineExceeded) {
+		return true
+	}
+	if err, ok := err.(interface{ Temporary() bool }); ok {
+		return err.Temporary()
+	}
+	return false
+}
+
+type nodeMeta struct {
+	regionID   int
+	regionCode string
+	hostname   string
+	addr       netip.Addr
+}
+
+type measureFn func(conn io.ReadWriteCloser, dst *net.UDPAddr) (rtt time.Duration, err error)
+
+// probe measures STUN round trip time for the node described by meta over
+// conn against dstPort. It may return a nil duration and nil error if the
+// STUN request timed out. A non-nil error indicates an unrecoverable or
+// non-temporary error.
+func probe(meta nodeMeta, conn io.ReadWriteCloser, fn measureFn, dstPort int) (*time.Duration, error) {
+	ua := &net.UDPAddr{
+		IP:   net.IP(meta.addr.AsSlice()),
+		Port: dstPort,
+	}
+
+	time.Sleep(rand.N(200 * time.Millisecond)) // jitter across tx
+	rtt, err := fn(conn, ua)
+	if err != nil {
+		if isTemporaryOrTimeoutErr(err) {
+			log.Printf("temp error measuring RTT to %s(%s): %v", meta.hostname, ua.String(), err)
+			return nil, nil
+		}
+		return nil, err
+	}
+	return &rtt, nil
+}
+
+// nodeMetaFromDERPMap parses the provided DERP map in order to update nodeMeta
+// in the provided nodeMetaByAddr. It returns a slice of nodeMeta containing
+// the nodes that are no longer seen in the DERP map, but were previously held
+// in nodeMetaByAddr.
+func nodeMetaFromDERPMap(dm *tailcfg.DERPMap, nodeMetaByAddr map[netip.Addr]nodeMeta, ipv6 bool) (stale []nodeMeta, err error) {
+	// Parse the new derp map before making any state changes in nodeMetaByAddr.
+	// If parse fails we just stick with the old state.
+	updated := make(map[netip.Addr]nodeMeta)
+	for regionID, region := range dm.Regions {
+		for _, node := range region.Nodes {
+			v4, err := netip.ParseAddr(node.IPv4)
+			if err != nil || !v4.Is4() {
+				return nil, fmt.Errorf("invalid ipv4 addr for node in derp map: %v", node.Name)
+			}
+			metas := make([]nodeMeta, 0, 2)
+			metas = append(metas, nodeMeta{
+				regionID:   regionID,
+				regionCode: region.RegionCode,
+				hostname:   node.HostName,
+				addr:       v4,
+			})
+			if ipv6 {
+				v6, err := netip.ParseAddr(node.IPv6)
+				if err != nil || !v6.Is6() {
+					return nil, fmt.Errorf("invalid ipv6 addr for node in derp map: %v", node.Name)
+				}
+				metas = append(metas, metas[0])
+				metas[1].addr = v6
+			}
+			for _, meta := range metas {
+				updated[meta.addr] = meta
+			}
+		}
+	}
+
+	// Find nodeMeta that have changed
+	for addr, updatedMeta := range updated {
+		previousMeta, ok := nodeMetaByAddr[addr]
+		if ok {
+			if previousMeta == updatedMeta {
+				continue
+			}
+			stale = append(stale, previousMeta)
+			nodeMetaByAddr[addr] = updatedMeta
+		} else {
+			nodeMetaByAddr[addr] = updatedMeta
+		}
+	}
+
+	// Find nodeMeta that no longer exist
+	for addr, potentialStale := range nodeMetaByAddr {
+		_, ok := updated[addr]
+		if !ok {
+			stale = append(stale, potentialStale)
+		}
+	}
+
+	return stale, nil
+}
+
+func getStableConns(stableConns map[netip.Addr]map[int][2]io.ReadWriteCloser, addr netip.Addr, dstPort int) ([2]io.ReadWriteCloser, error) {
+	conns := [2]io.ReadWriteCloser{}
+	byDstPort, ok := stableConns[addr]
+	if ok {
+		conns, ok = byDstPort[dstPort]
+		if ok {
+			return conns, nil
+		}
+	}
+	if supportsKernelTS() {
+		kconn, err := getConnKernelTimestamp()
+		if err != nil {
+			return conns, err
+		}
+		conns[timestampSourceKernel] = kconn
+	}
+	uconn, err := net.ListenUDP("udp", &net.UDPAddr{})
+	if err != nil {
+		if supportsKernelTS() {
+			conns[timestampSourceKernel].Close()
+		}
+		return conns, err
+	}
+	conns[timestampSourceUserspace] = uconn
+	if byDstPort == nil {
+		byDstPort = make(map[int][2]io.ReadWriteCloser)
+	}
+	byDstPort[dstPort] = conns
+	stableConns[addr] = byDstPort
+	return conns, nil
+}
+
+// probeNodes measures the round-trip time for STUN binding requests against the
+// DERP nodes described by nodeMetaByAddr while using/updating stableConns for
+// UDP sockets that should be recycled across runs. It returns the results or
+// an error if one occurs.
+func probeNodes(nodeMetaByAddr map[netip.Addr]nodeMeta, stableConns map[netip.Addr]map[int][2]io.ReadWriteCloser, dstPorts []int) ([]result, error) {
+	wg := sync.WaitGroup{}
+	results := make([]result, 0)
+	resultsCh := make(chan result)
+	errCh := make(chan error)
+	doneCh := make(chan struct{})
+	numProbes := 0
+	at := time.Now()
+	addrsToProbe := make(map[netip.Addr]bool)
+
+	doProbe := func(conn io.ReadWriteCloser, meta nodeMeta, source timestampSource, dstPort int) {
+		defer wg.Done()
+		r := result{
+			key: resultKey{
+				meta:            meta,
+				timestampSource: source,
+				dstPort:         dstPort,
+			},
+			at: at,
+		}
+		if conn == nil {
+			var err error
+			if source == timestampSourceKernel {
+				conn, err = getConnKernelTimestamp()
+			} else {
+				conn, err = net.ListenUDP("udp", &net.UDPAddr{})
+			}
+			if err != nil {
+				select {
+				case <-doneCh:
+					return
+				case errCh <- err:
+					return
+				}
+			}
+			defer conn.Close()
+		} else {
+			r.key.connStability = stableConn
+		}
+		fn := measureRTT
+		if source == timestampSourceKernel {
+			fn = measureRTTKernel
+		}
+		rtt, err := probe(meta, conn, fn, dstPort)
+		if err != nil {
+			select {
+			case <-doneCh:
+				return
+			case errCh <- err:
+				return
+			}
+		}
+		r.rtt = rtt
+		select {
+		case <-doneCh:
+		case resultsCh <- r:
+		}
+	}
+
+	for _, meta := range nodeMetaByAddr {
+		addrsToProbe[meta.addr] = true
+		for _, port := range dstPorts {
+			stable, err := getStableConns(stableConns, meta.addr, port)
+			if err != nil {
+				close(doneCh)
+				wg.Wait()
+				return nil, err
+			}
+
+			wg.Add(2)
+			numProbes += 2
+			go doProbe(stable[timestampSourceUserspace], meta, timestampSourceUserspace, port)
+			go doProbe(nil, meta, timestampSourceUserspace, port)
+			if supportsKernelTS() {
+				wg.Add(2)
+				numProbes += 2
+				go doProbe(stable[timestampSourceKernel], meta, timestampSourceKernel, port)
+				go doProbe(nil, meta, timestampSourceKernel, port)
+			}
+		}
+	}
+
+	// cleanup conns we no longer need
+	for k, byDstPort := range stableConns {
+		if !addrsToProbe[k] {
+			for _, conns := range byDstPort {
+				if conns[timestampSourceKernel] != nil {
+					conns[timestampSourceKernel].Close()
+				}
+				conns[timestampSourceUserspace].Close()
+				delete(stableConns, k)
+			}
+		}
+	}
+
+	for {
+		select {
+		case err := <-errCh:
+			close(doneCh)
+			wg.Wait()
+			return nil, err
+		case result := <-resultsCh:
+			results = append(results, result)
+			if len(results) == numProbes {
+				return results, nil
+			}
+		}
+	}
+}
+
+type connStability bool
+
+const (
+	unstableConn connStability = false
+	stableConn   connStability = true
+)
+
+const (
+	rttMetricName      = "stunstamp_derp_stun_rtt_ns"
+	timeoutsMetricName = "stunstamp_derp_stun_timeouts_total"
+)
+
+func timeSeriesLabels(metricName string, meta nodeMeta, instance string, source timestampSource, stability connStability, dstPort int) []prompb.Label {
+	addressFamily := "ipv4"
+	if meta.addr.Is6() {
+		addressFamily = "ipv6"
+	}
+	labels := make([]prompb.Label, 0)
+	labels = append(labels, prompb.Label{
+		Name:  "job",
+		Value: "stunstamp-rw",
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "instance",
+		Value: instance,
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "region_id",
+		Value: fmt.Sprintf("%d", meta.regionID),
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "region_code",
+		Value: meta.regionCode,
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "address_family",
+		Value: addressFamily,
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "hostname",
+		Value: meta.hostname,
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "dst_port",
+		Value: strconv.Itoa(dstPort),
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "__name__",
+		Value: metricName,
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "timestamp_source",
+		Value: source.String(),
+	})
+	labels = append(labels, prompb.Label{
+		Name:  "stable_conn",
+		Value: fmt.Sprintf("%v", stability),
+	})
+	slices.SortFunc(labels, func(a, b prompb.Label) int {
+		// prometheus remote-write spec requires lexicographically sorted label names
+		return cmp.Compare(a.Name, b.Name)
+	})
+	return labels
+}
+
+const (
+	// https://prometheus.io/docs/concepts/remote_write_spec/#stale-markers
+	staleNaN uint64 = 0x7ff0000000000002
+)
+
+func staleMarkersFromNodeMeta(stale []nodeMeta, instance string, dstPorts []int) []prompb.TimeSeries {
+	staleMarkers := make([]prompb.TimeSeries, 0)
+	now := time.Now()
+	for _, s := range stale {
+		for _, dstPort := range dstPorts {
+			samples := []prompb.Sample{
+				{
+					Timestamp: now.UnixMilli(),
+					Value:     math.Float64frombits(staleNaN),
+				},
+			}
+			staleMarkers = append(staleMarkers, prompb.TimeSeries{
+				Labels:  timeSeriesLabels(rttMetricName, s, instance, timestampSourceUserspace, unstableConn, dstPort),
+				Samples: samples,
+			})
+			staleMarkers = append(staleMarkers, prompb.TimeSeries{
+				Labels:  timeSeriesLabels(rttMetricName, s, instance, timestampSourceUserspace, stableConn, dstPort),
+				Samples: samples,
+			})
+			staleMarkers = append(staleMarkers, prompb.TimeSeries{
+				Labels:  timeSeriesLabels(timeoutsMetricName, s, instance, timestampSourceUserspace, unstableConn, dstPort),
+				Samples: samples,
+			})
+			staleMarkers = append(staleMarkers, prompb.TimeSeries{
+				Labels:  timeSeriesLabels(timeoutsMetricName, s, instance, timestampSourceUserspace, stableConn, dstPort),
+				Samples: samples,
+			})
+			if supportsKernelTS() {
+				staleMarkers = append(staleMarkers, prompb.TimeSeries{
+					Labels:  timeSeriesLabels(rttMetricName, s, instance, timestampSourceKernel, unstableConn, dstPort),
+					Samples: samples,
+				})
+				staleMarkers = append(staleMarkers, prompb.TimeSeries{
+					Labels:  timeSeriesLabels(rttMetricName, s, instance, timestampSourceKernel, stableConn, dstPort),
+					Samples: samples,
+				})
+				staleMarkers = append(staleMarkers, prompb.TimeSeries{
+					Labels:  timeSeriesLabels(timeoutsMetricName, s, instance, timestampSourceKernel, unstableConn, dstPort),
+					Samples: samples,
+				})
+				staleMarkers = append(staleMarkers, prompb.TimeSeries{
+					Labels:  timeSeriesLabels(timeoutsMetricName, s, instance, timestampSourceKernel, stableConn, dstPort),
+					Samples: samples,
+				})
+			}
+		}
+	}
+	return staleMarkers
+}
+
+// resultsToPromTimeSeries returns a slice of prometheus TimeSeries for the
+// provided results and instance. timeouts is updated based on results, i.e.
+// all result.key's are added to timeouts if they do not exist, and removed
+// from timeouts if they are not present in results.
+func resultsToPromTimeSeries(results []result, instance string, timeouts map[resultKey]uint64) []prompb.TimeSeries {
+	all := make([]prompb.TimeSeries, 0, len(results)*2)
+	seenKeys := make(map[resultKey]bool)
+	for _, r := range results {
+		timeoutsCount := timeouts[r.key] // a non-existent key will return a zero val
+		seenKeys[r.key] = true
+		rttLabels := timeSeriesLabels(rttMetricName, r.key.meta, instance, r.key.timestampSource, r.key.connStability, r.key.dstPort)
+		rttSamples := make([]prompb.Sample, 1)
+		rttSamples[0].Timestamp = r.at.UnixMilli()
+		if r.rtt != nil {
+			rttSamples[0].Value = float64(*r.rtt)
+		} else {
+			rttSamples[0].Value = math.NaN()
+			timeoutsCount++
+		}
+		rttTS := prompb.TimeSeries{
+			Labels:  rttLabels,
+			Samples: rttSamples,
+		}
+		all = append(all, rttTS)
+		timeouts[r.key] = timeoutsCount
+		timeoutsLabels := timeSeriesLabels(timeoutsMetricName, r.key.meta, instance, r.key.timestampSource, r.key.connStability, r.key.dstPort)
+		timeoutsSamples := make([]prompb.Sample, 1)
+		timeoutsSamples[0].Timestamp = r.at.UnixMilli()
+		timeoutsSamples[0].Value = float64(timeoutsCount)
+		timeoutsTS := prompb.TimeSeries{
+			Labels:  timeoutsLabels,
+			Samples: timeoutsSamples,
+		}
+		all = append(all, timeoutsTS)
+	}
+	for k := range timeouts {
+		if !seenKeys[k] {
+			delete(timeouts, k)
+		}
+	}
+	return all
+}
+
+type remoteWriteClient struct {
+	c   *http.Client
+	url string
+}
+
+type recoverableErr struct {
+	error
+}
+
+func newRemoteWriteClient(url string) *remoteWriteClient {
+	return &remoteWriteClient{
+		c: &http.Client{
+			Timeout: time.Second * 30,
+		},
+		url: url,
+	}
+}
+
+func (r *remoteWriteClient) write(ctx context.Context, ts []prompb.TimeSeries) error {
+	wr := &prompb.WriteRequest{
+		Timeseries: ts,
+	}
+	b, err := wr.Marshal()
+	if err != nil {
+		return fmt.Errorf("unable to marshal write request: %w", err)
+	}
+	compressed := snappy.Encode(nil, b)
+	req, err := http.NewRequestWithContext(ctx, "POST", r.url, bytes.NewReader(compressed))
+	if err != nil {
+		return fmt.Errorf("unable to create write request: %w", err)
+	}
+	req.Header.Add("Content-Encoding", "snappy")
+	req.Header.Set("Content-Type", "application/x-protobuf")
+	req.Header.Set("User-Agent", "stunstamp")
+	req.Header.Set("X-Prometheus-Remote-Write-Version", "0.1.0")
+	resp, err := r.c.Do(req)
+	if err != nil {
+		return recoverableErr{fmt.Errorf("error performing write request: %w", err)}
+	}
+	if resp.StatusCode/100 != 2 {
+		err = fmt.Errorf("remote server %s returned HTTP status %d", r.url, resp.StatusCode)
+	}
+	if resp.StatusCode/100 == 5 || resp.StatusCode == http.StatusTooManyRequests {
+		return recoverableErr{err}
+	}
+	return err
+}
+
+func remoteWriteTimeSeries(client *remoteWriteClient, tsCh chan []prompb.TimeSeries) {
+	bo := backoff.NewBackoff("remote-write", log.Printf, time.Second*30)
+	// writeErr may contribute to bo's backoff schedule across tsCh read ops,
+	// i.e. if an unrecoverable error occurs for client.write(ctx, A), that
+	// should be accounted against bo prior to attempting to
+	// client.write(ctx, B).
+	var writeErr error
+	for ts := range tsCh {
+		for {
+			bo.BackOff(context.Background(), writeErr)
+			reqCtx, cancel := context.WithTimeout(context.Background(), time.Second*30)
+			writeErr = client.write(reqCtx, ts)
+			cancel()
+			var re recoverableErr
+			recoverable := errors.As(writeErr, &re)
+			if writeErr != nil {
+				log.Printf("remote write error(recoverable=%v): %v", recoverable, writeErr)
+			}
+			if !recoverable {
+				// a nil err is not recoverable
+				break
+			}
+		}
+	}
+}
+
+func main() {
+	flag.Parse()
+	if len(*flagDstPorts) == 0 {
+		log.Fatal("dst-ports flag is unset")
+	}
+	dstPortsSplit := strings.Split(*flagDstPorts, ",")
+	slices.Sort(dstPortsSplit)
+	dstPortsSplit = slices.Compact(dstPortsSplit)
+	dstPorts := make([]int, 0, len(dstPortsSplit))
+	for _, d := range dstPortsSplit {
+		i, err := strconv.ParseUint(d, 10, 16)
+		if err != nil {
+			log.Fatal("invalid dst-ports")
+		}
+		dstPorts = append(dstPorts, int(i))
+	}
+	if len(*flagDERPMap) < 1 {
+		log.Fatal("derp-map flag is unset")
+	}
+	if len(*flagOut) < 1 {
+		log.Fatal("out flag is unset")
+	}
+	if *flagInterval < minInterval || *flagInterval > maxBufferDuration {
+		log.Fatalf("interval must be >= %s and <= %s", minInterval, maxBufferDuration)
+	}
+	if *flagRetention < *flagInterval {
+		log.Fatal("retention must be >= interval")
+	}
+	if len(*flagRemoteWriteURL) < 1 {
+		log.Fatal("rw-url flag is unset")
+	}
+	_, err := url.Parse(*flagRemoteWriteURL)
+	if err != nil {
+		log.Fatalf("invalid rw-url flag value: %v", err)
+	}
+	if len(*flagInstance) < 1 {
+		hostname, err := os.Hostname()
+		if err != nil {
+			log.Fatalf("failed to get hostname: %v", err)
+		}
+		*flagInstance = hostname
+	}
+
+	sigCh := make(chan os.Signal, 1)
+	signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM)
+	dmCh := make(chan *tailcfg.DERPMap)
+
+	go func() {
+		bo := backoff.NewBackoff("derp-map", log.Printf, time.Second*30)
+		for {
+			ctx, cancel := context.WithTimeout(context.Background(), time.Second*10)
+			dm, err := getDERPMap(ctx, *flagDERPMap)
+			cancel()
+			bo.BackOff(context.Background(), err)
+			if err != nil {
+				continue
+			}
+			dmCh <- dm
+			return
+		}
+	}()
+
+	nodeMetaByAddr := make(map[netip.Addr]nodeMeta)
+	select {
+	case <-sigCh:
+		return
+	case dm := <-dmCh:
+		_, err := nodeMetaFromDERPMap(dm, nodeMetaByAddr, *flagIPv6)
+		if err != nil {
+			log.Fatalf("error parsing derp map on startup: %v", err)
+		}
+	}
+
+	db, err := newDB(*flagOut)
+	if err != nil {
+		log.Fatalf("error opening output file for writing: %v", err)
+	}
+	defer db.Close()
+
+	_, err = db.Exec("PRAGMA journal_mode=WAL")
+	if err != nil {
+		log.Fatalf("error enabling WAL mode: %v", err)
+	}
+
+	// No indices or primary key. Keep it simple for now. Reads will be full
+	// scans. We can AUTOINCREMENT rowid in the future and hold an in-memory
+	// index to at_unix if needed as reads are almost always going to be
+	// time-bound (e.g. WHERE at_unix >= ?). At the time of authorship we have
+	// ~300 data points per-interval w/o ipv6 w/kernel timestamping resulting
+	// in ~2.6m rows in 24h w/a 10s probe interval.
+	_, err = db.Exec(`
+CREATE TABLE IF NOT EXISTS rtt(at_unix INT, region_id INT, hostname TEXT, af INT, address TEXT, timestamp_source INT, stable_conn INT, dst_port INT, rtt_ns INT)
+`)
+	if err != nil {
+		log.Fatalf("error initializing db: %v", err)
+	}
+
+	wg := sync.WaitGroup{}
+	httpErrCh := make(chan error, 1)
+	var httpServer *http.Server
+	if len(*flagAPI) > 0 {
+		api := newAPI(db)
+		httpServer = &http.Server{
+			Addr:         *flagAPI,
+			Handler:      api,
+			ReadTimeout:  time.Second * 60,
+			WriteTimeout: time.Second * 60,
+		}
+		wg.Add(1)
+		go func() {
+			err := httpServer.ListenAndServe()
+			httpErrCh <- err
+			wg.Done()
+		}()
+	}
+
+	tsCh := make(chan []prompb.TimeSeries, maxBufferDuration / *flagInterval)
+	remoteWriteDoneCh := make(chan struct{})
+	rwc := newRemoteWriteClient(*flagRemoteWriteURL)
+	go func() {
+		remoteWriteTimeSeries(rwc, tsCh)
+		close(remoteWriteDoneCh)
+	}()
+
+	shutdown := func() {
+		if httpServer != nil {
+			httpServer.Close()
+		}
+		close(tsCh)
+		select {
+		case <-time.After(time.Second * 10): // give goroutine some time to flush
+		case <-remoteWriteDoneCh:
+		}
+
+		// send stale markers on shutdown
+		staleMeta := make([]nodeMeta, 0, len(nodeMetaByAddr))
+		for _, v := range nodeMetaByAddr {
+			staleMeta = append(staleMeta, v)
+		}
+		staleMarkers := staleMarkersFromNodeMeta(staleMeta, *flagInstance, dstPorts)
+		if len(staleMarkers) > 0 {
+			ctx, cancel := context.WithTimeout(context.Background(), time.Second*5)
+			rwc.write(ctx, staleMarkers)
+			cancel()
+		}
+
+		wg.Wait()
+		return
+	}
+
+	log.Println("stunstamp started")
+
+	// Re-using sockets means we get the same 5-tuple across runs. This results
+	// in a higher probability of the packets traversing the same underlay path.
+	// Comparison of stable and unstable 5-tuple results can shed light on
+	// differences between paths where hashing (multipathing/load balancing)
+	// comes into play.
+	stableConns := make(map[netip.Addr]map[int][2]io.ReadWriteCloser)
+
+	// timeouts holds counts of timeout events. Values are persisted for the
+	// lifetime of the related node in the DERP map.
+	timeouts := make(map[resultKey]uint64)
+
+	derpMapTicker := time.NewTicker(time.Minute * 5)
+	defer derpMapTicker.Stop()
+	probeTicker := time.NewTicker(*flagInterval)
+	defer probeTicker.Stop()
+	cleanupTicker := time.NewTicker(time.Hour)
+	defer cleanupTicker.Stop()
+
+	for {
+		select {
+		case <-cleanupTicker.C:
+			older := time.Now().Add(-*flagRetention)
+			log.Printf("cleaning up measurements older than %v", older)
+			_, err := db.Exec("DELETE FROM rtt WHERE at_unix < ?", older.Unix())
+			if err != nil {
+				log.Printf("error cleaning up old data: %v", err)
+				shutdown()
+				return
+			}
+		case <-probeTicker.C:
+			results, err := probeNodes(nodeMetaByAddr, stableConns, dstPorts)
+			if err != nil {
+				log.Printf("unrecoverable error while probing: %v", err)
+				shutdown()
+				return
+			}
+			ts := resultsToPromTimeSeries(results, *flagInstance, timeouts)
+			select {
+			case tsCh <- ts:
+			default:
+				select {
+				case <-tsCh:
+					log.Println("prometheus remote-write buffer full, dropped measurements")
+				default:
+					tsCh <- ts
+				}
+			}
+			tx, err := db.Begin()
+			if err != nil {
+				log.Printf("error beginning sqlite tx: %v", err)
+				shutdown()
+				return
+			}
+			for _, result := range results {
+				af := 4
+				if result.key.meta.addr.Is6() {
+					af = 6
+				}
+				_, err = tx.Exec("INSERT INTO rtt(at_unix, region_id, hostname, af, address, timestamp_source, stable_conn, dst_port, rtt_ns) VALUES(?, ?, ?, ?, ?, ?, ?, ?, ?)",
+					result.at.Unix(), result.key.meta.regionID, result.key.meta.hostname, af, result.key.meta.addr.String(), result.key.timestampSource, result.key.connStability, result.key.dstPort, result.rtt)
+				if err != nil {
+					tx.Rollback()
+					log.Printf("error adding result to tx: %v", err)
+					shutdown()
+					return
+				}
+			}
+			err = tx.Commit()
+			if err != nil {
+				log.Printf("error committing tx: %v", err)
+				shutdown()
+				return
+			}
+		case dm := <-dmCh:
+			staleMeta, err := nodeMetaFromDERPMap(dm, nodeMetaByAddr, *flagIPv6)
+			if err != nil {
+				log.Printf("error parsing DERP map, continuing with stale map: %v", err)
+				continue
+			}
+			staleMarkers := staleMarkersFromNodeMeta(staleMeta, *flagInstance, dstPorts)
+			if len(staleMarkers) < 1 {
+				continue
+			}
+			select {
+			case tsCh <- staleMarkers:
+			default:
+				select {
+				case <-tsCh:
+					log.Println("prometheus remote-write buffer full, dropped measurements")
+				default:
+					tsCh <- staleMarkers
+				}
+			}
+		case <-derpMapTicker.C:
+			go func() {
+				ctx, cancel := context.WithTimeout(context.Background(), time.Second*10)
+				defer cancel()
+				updatedDM, err := getDERPMap(ctx, *flagDERPMap)
+				if err != nil {
+					dmCh <- updatedDM
+				}
+			}()
+		case err := <-httpErrCh:
+			log.Printf("http server error: %v", err)
+			shutdown()
+			return
+		case <-sigCh:
+			shutdown()
+			return
+		}
+	}
+}

cmd/stunstamp/stunstamp_db_default.go

@@ -0,0 +1,26 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build !(windows && 386)
+
+package main
+
+import (
+	"database/sql"
+
+	_ "modernc.org/sqlite"
+)
+
+type db struct {
+	*sql.DB
+}
+
+func newDB(path string) (*db, error) {
+	d, err := sql.Open("sqlite", *flagOut)
+	if err != nil {
+		return nil, err
+	}
+	return &db{
+		DB: d,
+	}, nil
+}

cmd/stunstamp/stunstamp_db_windows_386.go

@@ -0,0 +1,17 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package main
+
+import (
+	"database/sql"
+	"errors"
+)
+
+type db struct {
+	*sql.DB
+}
+
+func newDB(path string) (*db, error) {
+	return nil, errors.New("unsupported platform")
+}

cmd/stunstamp/stunstamp_default.go

@@ -0,0 +1,25 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build !linux
+
+package main
+
+import (
+	"errors"
+	"io"
+	"net"
+	"time"
+)
+
+func getConnKernelTimestamp() (io.ReadWriteCloser, error) {
+	return nil, errors.New("unimplemented")
+}
+
+func measureRTTKernel(conn io.ReadWriteCloser, dst *net.UDPAddr) (rtt time.Duration, err error) {
+	return 0, errors.New("unimplemented")
+}
+
+func supportsKernelTS() bool {
+	return false
+}

cmd/stunstamp/stunstamp_linux.go

@@ -0,0 +1,143 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package main
+
+import (
+	"bytes"
+	"context"
+	"encoding/binary"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"time"
+
+	"github.com/mdlayher/socket"
+	"golang.org/x/sys/unix"
+	"tailscale.com/net/stun"
+)
+
+const (
+	flags = unix.SOF_TIMESTAMPING_TX_SOFTWARE | // tx timestamp generation in device driver
+		unix.SOF_TIMESTAMPING_RX_SOFTWARE | // rx timestamp generation in the kernel
+		unix.SOF_TIMESTAMPING_SOFTWARE // report software timestamps
+)
+
+func getConnKernelTimestamp() (io.ReadWriteCloser, error) {
+	sconn, err := socket.Socket(unix.AF_INET6, unix.SOCK_DGRAM, unix.IPPROTO_UDP, "udp", nil)
+	if err != nil {
+		return nil, err
+	}
+	sa := unix.SockaddrInet6{}
+	err = sconn.Bind(&sa)
+	if err != nil {
+		return nil, err
+	}
+	err = sconn.SetsockoptInt(unix.SOL_SOCKET, unix.SO_TIMESTAMPING_NEW, flags)
+	if err != nil {
+		return nil, err
+	}
+	return sconn, nil
+}
+
+func parseTimestampFromCmsgs(oob []byte) (time.Time, error) {
+	msgs, err := unix.ParseSocketControlMessage(oob)
+	if err != nil {
+		return time.Time{}, fmt.Errorf("error parsing oob as cmsgs: %w", err)
+	}
+	for _, msg := range msgs {
+		if msg.Header.Level == unix.SOL_SOCKET && msg.Header.Type == unix.SO_TIMESTAMPING_NEW && len(msg.Data) >= 16 {
+			sec := int64(binary.NativeEndian.Uint64(msg.Data[:8]))
+			ns := int64(binary.NativeEndian.Uint64(msg.Data[8:16]))
+			return time.Unix(sec, ns), nil
+		}
+	}
+	return time.Time{}, errors.New("failed to parse timestamp from cmsgs")
+}
+
+func measureRTTKernel(conn io.ReadWriteCloser, dst *net.UDPAddr) (rtt time.Duration, err error) {
+	sconn, ok := conn.(*socket.Conn)
+	if !ok {
+		return 0, fmt.Errorf("conn of unexpected type: %T", conn)
+	}
+
+	var to unix.Sockaddr
+	to4 := dst.IP.To4()
+	if to4 != nil {
+		to = &unix.SockaddrInet4{
+			Port: dst.Port,
+		}
+		copy(to.(*unix.SockaddrInet4).Addr[:], to4)
+	} else {
+		to = &unix.SockaddrInet6{
+			Port: dst.Port,
+		}
+		copy(to.(*unix.SockaddrInet6).Addr[:], dst.IP)
+	}
+
+	txID := stun.NewTxID()
+	req := stun.Request(txID)
+
+	err = sconn.Sendto(context.Background(), req, 0, to)
+	if err != nil {
+		return 0, fmt.Errorf("sendto error: %v", err) // don't wrap
+	}
+
+	txCtx, txCancel := context.WithTimeout(context.Background(), time.Second*2)
+	defer txCancel()
+
+	buf := make([]byte, 1024)
+	oob := make([]byte, 1024)
+	var txAt time.Time
+
+	for {
+		n, oobn, _, _, err := sconn.Recvmsg(txCtx, buf, oob, unix.MSG_ERRQUEUE)
+		if err != nil {
+			return 0, fmt.Errorf("recvmsg (MSG_ERRQUEUE) error: %v", err) // don't wrap
+		}
+
+		buf = buf[:n]
+		if n < len(req) || !bytes.Equal(req, buf[len(buf)-len(req):]) {
+			// Spin until we find the message we sent. We get the full packet
+			// looped including eth header so match against the tail.
+			continue
+		}
+		txAt, err = parseTimestampFromCmsgs(oob[:oobn])
+		if err != nil {
+			return 0, fmt.Errorf("failed to get tx timestamp: %v", err) // don't wrap
+		}
+		break
+	}
+
+	rxCtx, rxCancel := context.WithTimeout(context.Background(), time.Second*2)
+	defer rxCancel()
+
+	for {
+		n, oobn, _, _, err := sconn.Recvmsg(rxCtx, buf, oob, 0)
+		if err != nil {
+			return 0, fmt.Errorf("recvmsg error: %w", err) // wrap for timeout-related error unwrapping
+		}
+
+		gotTxID, _, err := stun.ParseResponse(buf[:n])
+		if err != nil || gotTxID != txID {
+			// Spin until we find the txID we sent. We may end up reading
+			// extremely late arriving responses from previous intervals. As
+			// such, we can't be certain if we're parsing the "current"
+			// response, so spin for parse errors too.
+			continue
+		}
+
+		rxAt, err := parseTimestampFromCmsgs(oob[:oobn])
+		if err != nil {
+			return 0, fmt.Errorf("failed to get rx timestamp: %v", err) // don't wrap
+		}
+
+		return rxAt.Sub(txAt), nil
+	}
+
+}
+
+func supportsKernelTS() bool {
+	return true
+}

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/debug.go

@@ -28,10 +28,12 @@ import (
 
 	"github.com/peterbourgon/ff/v3/ffcli"
 	"golang.org/x/net/http/httpproxy"
+	"golang.org/x/net/http2"
 	"tailscale.com/client/tailscale"
 	"tailscale.com/client/tailscale/apitype"
 	"tailscale.com/control/controlhttp"
 	"tailscale.com/hostinfo"
+	"tailscale.com/internal/noiseconn"
 	"tailscale.com/ipn"
 	"tailscale.com/net/tsaddr"
 	"tailscale.com/net/tshttpproxy"
@@ -801,7 +803,10 @@ func runTS2021(ctx context.Context, args []string) error {
 		log.Printf("Dial(%q, %q) ...", network, address)
 		c, err := dialer.DialContext(ctx, network, address)
 		if err != nil {
-			log.Printf("Dial(%q, %q) = %v", network, address, err)
+			// skip logging context cancellation errors
+			if !errors.Is(err, context.Canceled) {
+				log.Printf("Dial(%q, %q) = %v", network, address, err)
+			}
 		} else {
 			log.Printf("Dial(%q, %q) = %v / %v", network, address, c.LocalAddr(), c.RemoteAddr())
 		}
@@ -834,6 +839,52 @@ func runTS2021(ctx context.Context, args []string) error {
 	}
 
 	log.Printf("final underlying conn: %v / %v", conn.LocalAddr(), conn.RemoteAddr())
+
+	h2Transport, err := http2.ConfigureTransports(&http.Transport{
+		IdleConnTimeout: time.Second,
+	})
+	if err != nil {
+		return fmt.Errorf("http2.ConfigureTransports: %w", err)
+	}
+
+	// Now, create a Noise conn over the existing conn.
+	nc, err := noiseconn.New(conn.Conn, h2Transport, 0, nil)
+	if err != nil {
+		return fmt.Errorf("noiseconn.New: %w", err)
+	}
+	defer nc.Close()
+
+	// Reserve a RoundTrip for the whoami request.
+	ok, _, err := nc.ReserveNewRequest(ctx)
+	if err != nil {
+		return fmt.Errorf("ReserveNewRequest: %w", err)
+	}
+	if !ok {
+		return errors.New("ReserveNewRequest failed")
+	}
+
+	// Make a /whoami request to the server to verify that we can actually
+	// communicate over the newly-established connection.
+	whoamiURL := "http://" + ts2021Args.host + "/machine/whoami"
+	req, err = http.NewRequestWithContext(ctx, "GET", whoamiURL, nil)
+	if err != nil {
+		return err
+	}
+	resp, err := nc.RoundTrip(req)
+	if err != nil {
+		return fmt.Errorf("RoundTrip whoami request: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != 200 {
+		log.Printf("whoami request returned status %v", resp.Status)
+	} else {
+		body, err := io.ReadAll(resp.Body)
+		if err != nil {
+			return fmt.Errorf("reading whoami response: %w", err)
+		}
+		log.Printf("whoami response: %q", body)
+	}
 	return nil
 }
 

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

@@ -78,9 +78,13 @@ func runNetcheck(ctx context.Context, args []string) error {
 		log.Printf("No DERP map from tailscaled; using default.")
 	}
 	if err != nil || noRegions {
-		hc := &http.Client{Transport: tlsdial.NewTransport()}
+		hc := &http.Client{
+			Transport: tlsdial.NewTransport(),
+			Timeout:   10 * time.Second,
+		}
 		dm, err = prodDERPMap(ctx, hc)
 		if err != nil {
+			log.Println("Failed to fetch a DERP map, so netcheck cannot continue. Check your Internet connection.")
 			return err
 		}
 	}
@@ -127,13 +131,13 @@ 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 +146,6 @@ 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)
@@ -210,6 +213,7 @@ func portMapping(r *netcheck.Report) string {
 }
 
 func prodDERPMap(ctx context.Context, httpc *http.Client) (*tailcfg.DERPMap, error) {
+	log.Printf("attempting to fetch a DERPMap from %s", ipn.DefaultControlURL)
 	req, err := http.NewRequestWithContext(ctx, "GET", ipn.DefaultControlURL+"/derpmap/default", nil)
 	if err != nil {
 		return nil, fmt.Errorf("create prodDERPMap request: %w", err)

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/cli/update.go

@@ -33,11 +33,13 @@ var updateCmd = &ffcli.Command{
 		//  - Alpine (and other apk-based distros)
 		//  - FreeBSD (and other pkg-based distros)
 		//  - Unraid/QNAP/Synology
+		//  - macOS
 		if distro.Get() != distro.Arch &&
 			distro.Get() != distro.Alpine &&
 			distro.Get() != distro.QNAP &&
 			distro.Get() != distro.Synology &&
-			runtime.GOOS != "freebsd" {
+			runtime.GOOS != "freebsd" &&
+			runtime.GOOS != "darwin" {
 			fs.StringVar(&updateArgs.track, "track", "", `which track to check for updates: "stable" or "unstable" (dev); empty means same as current`)
 			fs.StringVar(&updateArgs.version, "version", "", `explicit version to update/downgrade to`)
 		}

cmd/tailscale/cli/whois.go

@@ -26,12 +26,14 @@ var whoisCmd = &ffcli.Command{
 	FlagSet: func() *flag.FlagSet {
 		fs := newFlagSet("whois")
 		fs.BoolVar(&whoIsArgs.json, "json", false, "output in JSON format")
+		fs.StringVar(&whoIsArgs.proto, "proto", "", `protocol; one of "tcp" or "udp"; empty mans both `)
 		return fs
 	}(),
 }
 
 var whoIsArgs struct {
-	json bool // output in JSON format
+	json  bool   // output in JSON format
+	proto string // "tcp" or "udp"
 }
 
 func runWhoIs(ctx context.Context, args []string) error {
@@ -40,7 +42,7 @@ func runWhoIs(ctx context.Context, args []string) error {
 	} else if len(args) == 0 {
 		return errors.New("missing argument, expected one peer")
 	}
-	who, err := localClient.WhoIs(ctx, args[0])
+	who, err := localClient.WhoIsProto(ctx, whoIsArgs.proto, args[0])
 	if err != nil {
 		return err
 	}

cmd/tailscale/depaware.txt

@@ -57,7 +57,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
    L    github.com/vishvananda/netns                                 from github.com/tailscale/netlink+
         github.com/x448/float16                                      from github.com/fxamacker/cbor/v2
      💣 go4.org/mem                                                  from tailscale.com/client/tailscale+
-        go4.org/netipx                                               from tailscale.com/net/tsaddr+
+        go4.org/netipx                                               from tailscale.com/net/tsaddr
    W 💣 golang.zx2c4.com/wireguard/windows/tunnel/winipcfg           from tailscale.com/net/netmon+
         k8s.io/client-go/util/homedir                                from tailscale.com/cmd/tailscale/cli
         nhooyr.io/websocket                                          from tailscale.com/control/controlhttp+
@@ -78,7 +78,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         tailscale.com/cmd/tailscale/cli                              from tailscale.com/cmd/tailscale
         tailscale.com/cmd/tailscale/cli/ffcomplete                   from tailscale.com/cmd/tailscale/cli
         tailscale.com/cmd/tailscale/cli/ffcomplete/internal          from tailscale.com/cmd/tailscale/cli/ffcomplete
-        tailscale.com/control/controlbase                            from tailscale.com/control/controlhttp
+        tailscale.com/control/controlbase                            from tailscale.com/control/controlhttp+
         tailscale.com/control/controlhttp                            from tailscale.com/cmd/tailscale/cli
         tailscale.com/control/controlknobs                           from tailscale.com/net/portmapper
         tailscale.com/derp                                           from tailscale.com/derp/derphttp
@@ -89,6 +89,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         tailscale.com/health                                         from tailscale.com/net/tlsdial+
         tailscale.com/health/healthmsg                               from tailscale.com/cmd/tailscale/cli
         tailscale.com/hostinfo                                       from tailscale.com/client/web+
+        tailscale.com/internal/noiseconn                             from tailscale.com/cmd/tailscale/cli
         tailscale.com/ipn                                            from tailscale.com/client/tailscale+
         tailscale.com/ipn/ipnstate                                   from tailscale.com/client/tailscale+
         tailscale.com/licenses                                       from tailscale.com/client/web+
@@ -96,7 +97,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         tailscale.com/net/dns/recursive                              from tailscale.com/net/dnsfallback
         tailscale.com/net/dnscache                                   from tailscale.com/control/controlhttp+
         tailscale.com/net/dnsfallback                                from tailscale.com/control/controlhttp
-        tailscale.com/net/flowtrack                                  from tailscale.com/net/packet+
+        tailscale.com/net/flowtrack                                  from tailscale.com/net/packet
         tailscale.com/net/netaddr                                    from tailscale.com/ipn+
         tailscale.com/net/netcheck                                   from tailscale.com/cmd/tailscale/cli
         tailscale.com/net/neterror                                   from tailscale.com/net/netcheck+
@@ -104,7 +105,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
      💣 tailscale.com/net/netmon                                     from tailscale.com/cmd/tailscale/cli+
         tailscale.com/net/netns                                      from tailscale.com/derp/derphttp+
         tailscale.com/net/netutil                                    from tailscale.com/client/tailscale+
-        tailscale.com/net/packet                                     from tailscale.com/wgengine/capture+
+        tailscale.com/net/packet                                     from tailscale.com/wgengine/capture
         tailscale.com/net/ping                                       from tailscale.com/net/netcheck
         tailscale.com/net/portmapper                                 from tailscale.com/cmd/tailscale/cli+
         tailscale.com/net/sockstats                                  from tailscale.com/control/controlhttp+
@@ -164,11 +165,11 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         tailscale.com/util/vizerror                                  from tailscale.com/tailcfg+
      💣 tailscale.com/util/winutil                                   from tailscale.com/clientupdate+
    W 💣 tailscale.com/util/winutil/authenticode                      from tailscale.com/clientupdate
-   W 💣 tailscale.com/util/winutil/winenv                            from tailscale.com/hostinfo
+   W 💣 tailscale.com/util/winutil/winenv                            from tailscale.com/hostinfo+
         tailscale.com/version                                        from tailscale.com/client/web+
         tailscale.com/version/distro                                 from tailscale.com/client/web+
         tailscale.com/wgengine/capture                               from tailscale.com/cmd/tailscale/cli
-        tailscale.com/wgengine/filter                                from tailscale.com/types/netmap
+        tailscale.com/wgengine/filter/filtertype                     from tailscale.com/types/netmap
         golang.org/x/crypto/argon2                                   from tailscale.com/tka
         golang.org/x/crypto/blake2b                                  from golang.org/x/crypto/argon2+
         golang.org/x/crypto/blake2s                                  from tailscale.com/clientupdate/distsign+
@@ -182,13 +183,14 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         golang.org/x/crypto/nacl/secretbox                           from golang.org/x/crypto/nacl/box
         golang.org/x/crypto/pbkdf2                                   from software.sslmate.com/src/go-pkcs12
         golang.org/x/crypto/salsa20/salsa                            from golang.org/x/crypto/nacl/box+
-   W    golang.org/x/exp/constraints                                 from github.com/dblohm7/wingoes/pe
+   W    golang.org/x/exp/constraints                                 from github.com/dblohm7/wingoes/pe+
         golang.org/x/exp/maps                                        from tailscale.com/cmd/tailscale/cli
         golang.org/x/net/bpf                                         from github.com/mdlayher/netlink+
         golang.org/x/net/dns/dnsmessage                              from net+
         golang.org/x/net/http/httpguts                               from net/http+
         golang.org/x/net/http/httpproxy                              from net/http+
-        golang.org/x/net/http2/hpack                                 from net/http
+        golang.org/x/net/http2                                       from tailscale.com/cmd/tailscale/cli+
+        golang.org/x/net/http2/hpack                                 from net/http+
         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+
@@ -277,6 +279,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         math/big                                                     from crypto/dsa+
         math/bits                                                    from compress/flate+
         math/rand                                                    from github.com/mdlayher/netlink+
+        math/rand/v2                                                 from tailscale.com/derp+
         mime                                                         from golang.org/x/oauth2/internal+
         mime/multipart                                               from net/http
         mime/quotedprintable                                         from mime/multipart
@@ -299,7 +302,6 @@ 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 +309,6 @@ 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

@@ -12,10 +12,15 @@ 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",
 			"gvisor.dev/gvisor/pkg/tcpip/header": "https://github.com/tailscale/tailscale/issues/9756",
+			"tailscale.com/wgengine/filter":      "brings in bart, etc",
+			"github.com/bits-and-blooms/bitset":  "unneeded in CLI",
+			"github.com/gaissmai/bart":           "unneeded in CLI",
+			"tailscale.com/net/ipset":            "unneeded in CLI",
 		},
 	}.Check(t)
 }

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
@@ -244,7 +245,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/clientupdate                                   from tailscale.com/client/web+
         tailscale.com/clientupdate/distsign                          from tailscale.com/clientupdate
         tailscale.com/cmd/tailscaled/childproc                       from tailscale.com/cmd/tailscaled+
-        tailscale.com/control/controlbase                            from tailscale.com/control/controlclient+
+        tailscale.com/control/controlbase                            from tailscale.com/control/controlhttp+
         tailscale.com/control/controlclient                          from tailscale.com/cmd/tailscaled+
         tailscale.com/control/controlhttp                            from tailscale.com/control/controlclient
         tailscale.com/control/controlknobs                           from tailscale.com/control/controlclient+
@@ -264,6 +265,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/health                                         from tailscale.com/control/controlclient+
         tailscale.com/health/healthmsg                               from tailscale.com/ipn/ipnlocal
         tailscale.com/hostinfo                                       from tailscale.com/client/web+
+        tailscale.com/internal/noiseconn                             from tailscale.com/control/controlclient
         tailscale.com/ipn                                            from tailscale.com/client/tailscale+
         tailscale.com/ipn/conffile                                   from tailscale.com/cmd/tailscaled+
      💣 tailscale.com/ipn/ipnauth                                    from tailscale.com/ipn/ipnlocal+
@@ -294,6 +296,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/net/dnscache                                   from tailscale.com/control/controlclient+
         tailscale.com/net/dnsfallback                                from tailscale.com/cmd/tailscaled+
         tailscale.com/net/flowtrack                                  from tailscale.com/net/packet+
+        tailscale.com/net/ipset                                      from tailscale.com/ipn/ipnlocal+
         tailscale.com/net/netaddr                                    from tailscale.com/ipn+
         tailscale.com/net/netcheck                                   from tailscale.com/wgengine/magicsock+
         tailscale.com/net/neterror                                   from tailscale.com/net/dns/resolver+
@@ -319,6 +322,7 @@ 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
@@ -398,7 +402,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
      💣 tailscale.com/util/winutil                                   from tailscale.com/clientupdate+
    W 💣 tailscale.com/util/winutil/authenticode                      from tailscale.com/clientupdate+
    W    tailscale.com/util/winutil/policy                            from tailscale.com/ipn/ipnlocal
-   W 💣 tailscale.com/util/winutil/winenv                            from tailscale.com/hostinfo
+   W 💣 tailscale.com/util/winutil/winenv                            from tailscale.com/hostinfo+
         tailscale.com/util/zstdframe                                 from tailscale.com/control/controlclient+
         tailscale.com/version                                        from tailscale.com/client/web+
         tailscale.com/version/distro                                 from tailscale.com/client/web+
@@ -406,6 +410,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/wgengine                                       from tailscale.com/cmd/tailscaled+
         tailscale.com/wgengine/capture                               from tailscale.com/ipn/ipnlocal+
         tailscale.com/wgengine/filter                                from tailscale.com/control/controlclient+
+        tailscale.com/wgengine/filter/filtertype                     from tailscale.com/types/netmap+
      💣 tailscale.com/wgengine/magicsock                             from tailscale.com/ipn/ipnlocal+
         tailscale.com/wgengine/netlog                                from tailscale.com/wgengine
         tailscale.com/wgengine/netstack                              from tailscale.com/cmd/tailscaled
@@ -439,7 +444,7 @@ 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+
@@ -512,7 +517,6 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         hash                                                         from compress/zlib+
         hash/adler32                                                 from compress/zlib+
         hash/crc32                                                   from compress/gzip+
-        hash/fnv                                                     from tailscale.com/wgengine/magicsock
         hash/maphash                                                 from go4.org/mem
         html                                                         from html/template+
         html/template                                                from github.com/gorilla/csrf
@@ -527,7 +531,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         math/big                                                     from crypto/dsa+
         math/bits                                                    from compress/flate+
         math/rand                                                    from github.com/mdlayher/netlink+
-        math/rand/v2                                                 from tailscale.com/util/rands
+        math/rand/v2                                                 from tailscale.com/util/rands+
         mime                                                         from github.com/tailscale/xnet/webdav+
         mime/multipart                                               from net/http+
         mime/quotedprintable                                         from mime/multipart
@@ -552,7 +556,7 @@ 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 +564,6 @@ 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
@@ -677,7 +698,7 @@ func tryEngine(logf logger.Logf, sys *tsd.System, name string) (onlyNetstack boo
 			// configuration being unavailable (from the noop
 			// manager). More in Issue 4017.
 			// TODO(bradfitz): add a Synology-specific DNS manager.
-			conf.DNS, err = dns.NewOSConfigurator(logf, sys.HealthTracker(), "") // empty interface name
+			conf.DNS, err = dns.NewOSConfigurator(logf, sys.HealthTracker(), sys.ControlKnobs(), "") // empty interface name
 			if err != nil {
 				return false, fmt.Errorf("dns.NewOSConfigurator: %w", err)
 			}
@@ -705,7 +726,7 @@ func tryEngine(logf logger.Logf, sys *tsd.System, name string) (onlyNetstack boo
 			return false, fmt.Errorf("creating router: %w", err)
 		}
 
-		d, err := dns.NewOSConfigurator(logf, sys.HealthTracker(), devName)
+		d, err := dns.NewOSConfigurator(logf, sys.HealthTracker(), sys.ControlKnobs(), devName)
 		if err != nil {
 			dev.Close()
 			r.Close()

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/tailscaled/tailscaled_windows.go

@@ -435,6 +435,9 @@ func babysitProc(ctx context.Context, args []string, logf logger.Logf) {
 		startTime := time.Now()
 		log.Printf("exec: %#v %v", executable, args)
 		cmd := exec.Command(executable, args...)
+		cmd.SysProcAttr = &syscall.SysProcAttr{
+			CreationFlags: windows.DETACHED_PROCESS,
+		}
 
 		// Create a pipe object to use as the subproc's stdin.
 		// When the writer goes away, the reader gets EOF.

cmd/tsconnect/wasm/wasm_js.go

@@ -16,7 +16,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"log"
-	"math/rand"
+	"math/rand/v2"
 	"net"
 	"net/http"
 	"net/netip"
@@ -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,
 		})
@@ -605,7 +604,7 @@ func filterSlice[T any](a []T, f func(T) bool) []T {
 func generateHostname() string {
 	tails := words.Tails()
 	scales := words.Scales()
-	if rand.Int()%2 == 0 {
+	if rand.IntN(2) == 0 {
 		// JavaScript
 		tails = filterSlice(tails, func(s string) bool { return strings.HasPrefix(s, "j") })
 		scales = filterSlice(scales, func(s string) bool { return strings.HasPrefix(s, "s") })
@@ -615,8 +614,8 @@ func generateHostname() string {
 		scales = filterSlice(scales, func(s string) bool { return strings.HasPrefix(s, "a") })
 	}
 
-	tail := tails[rand.Intn(len(tails))]
-	scale := scales[rand.Intn(len(scales))]
+	tail := tails[rand.IntN(len(tails))]
+	scale := scales[rand.IntN(len(scales))]
 	return fmt.Sprintf("%s-%s", tail, scale)
 }
 

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 {

cmd/viewer/tests/tests_view.go

@@ -188,11 +188,7 @@ func (v *MapView) UnmarshalJSON(b []byte) error {
 
 func (v MapView) Int() views.Map[string, int] { return views.MapOf(v.ж.Int) }
 
-func (v MapView) SliceInt() views.MapFn[string, []int, views.Slice[int]] {
-	return views.MapFnOf(v.ж.SliceInt, func(t []int) views.Slice[int] {
-		return views.SliceOf(t)
-	})
-}
+func (v MapView) SliceInt() views.MapSlice[string, int] { return views.MapSliceOf(v.ж.SliceInt) }
 
 func (v MapView) StructPtrWithPtr() views.MapFn[string, *StructWithPtrs, StructWithPtrsView] {
 	return views.MapFnOf(v.ж.StructPtrWithPtr, func(t *StructWithPtrs) StructWithPtrsView {

cmd/viewer/viewer.go

@@ -92,6 +92,9 @@ func(v {{.ViewName}}) {{.FieldName}}() views.MapFn[{{.MapKeyType}},{{.MapValueTy
 	return {{.MapFn}}
 })}
 {{end}}
+{{define "mapSliceField"}}
+func(v {{.ViewName}}) {{.FieldName}}() views.MapSlice[{{.MapKeyType}},{{.MapValueType}}] { return views.MapSliceOf(v.ж.{{.FieldName}}) }
+{{end}}
 {{define "unsupportedField"}}func(v {{.ViewName}}) {{.FieldName}}() {{.FieldType}} {panic("unsupported")}
 {{end}}
 {{define "stringFunc"}}func(v {{.ViewName}}) String() string { return v.ж.String() }
@@ -241,9 +244,8 @@ func genView(buf *bytes.Buffer, it *codegen.ImportTracker, typ *types.Named, thi
 				case *types.Basic, *types.Named:
 					sElem := it.QualifiedName(sElem)
 					args.MapValueView = fmt.Sprintf("views.Slice[%v]", sElem)
-					args.MapValueType = "[]" + sElem
-					args.MapFn = "views.SliceOf(t)"
-					template = "mapFnField"
+					args.MapValueType = sElem
+					template = "mapSliceField"
 				case *types.Pointer:
 					ptr := x
 					pElem := ptr.Elem()

cmd/xdpderper/xdpderper.go

@@ -0,0 +1,96 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package main
+
+import (
+	"flag"
+	"io"
+	"log"
+	"net/http"
+	"os"
+	"os/signal"
+	"strings"
+	"syscall"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"tailscale.com/derp/xdp"
+	"tailscale.com/tsweb"
+)
+
+var (
+	flagDevice  = flag.String("device", "", "target device name")
+	flagPort    = flag.Int("dst-port", 0, "destination UDP port to serve")
+	flagVerbose = flag.Bool("verbose", false, "verbose output including verifier errors")
+	flagMode    = flag.String("mode", "xdp", "XDP mode; valid modes: [xdp, xdpgeneric, xdpdrv, xdpoffload]")
+	flagHTTP    = flag.String("http", ":8230", "HTTP listen address")
+)
+
+func main() {
+	flag.Parse()
+	var attachFlags xdp.XDPAttachFlags
+	switch strings.ToLower(*flagMode) {
+	case "xdp":
+		attachFlags = 0
+	case "xdpgeneric":
+		attachFlags = xdp.XDPGenericMode
+	case "xdpdrv":
+		attachFlags = xdp.XDPDriverMode
+	case "xdpoffload":
+		attachFlags = xdp.XDPOffloadMode
+	default:
+		log.Fatal("invalid mode")
+	}
+	server, err := xdp.NewSTUNServer(&xdp.STUNServerConfig{
+		DeviceName:      *flagDevice,
+		DstPort:         *flagPort,
+		AttachFlags:     attachFlags,
+		FullVerifierErr: *flagVerbose,
+	})
+	if err != nil {
+		log.Fatalf("failed to init XDP STUN server: %v", err)
+	}
+	defer server.Close()
+	err = prometheus.Register(server)
+	if err != nil {
+		log.Fatalf("failed to register XDP STUN server as a prometheus collector: %v", err)
+	}
+	log.Println("XDP STUN server started")
+
+	mux := http.NewServeMux()
+	debug := tsweb.Debugger(mux)
+	debug.KVFunc("Drop STUN", func() any {
+		return server.GetDropSTUN()
+	})
+	debug.Handle("drop-stun-on", "Drop STUN packets", http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		err := server.SetDropSTUN(true)
+		if err != nil {
+			http.Error(w, err.Error(), 500)
+		} else {
+			io.WriteString(w, "STUN packets are now being dropped.")
+		}
+	}))
+	debug.Handle("drop-stun-off", "Handle STUN packets", http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		err := server.SetDropSTUN(false)
+		if err != nil {
+			http.Error(w, err.Error(), 500)
+		} else {
+			io.WriteString(w, "STUN packets are now being handled.")
+		}
+	}))
+	errCh := make(chan error, 1)
+	go func() {
+		err := http.ListenAndServe(*flagHTTP, mux)
+		errCh <- err
+	}()
+
+	sigCh := make(chan os.Signal, 1)
+	signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM)
+	select {
+	case err := <-errCh:
+		log.Printf("HTTP serve err: %v", err)
+	case sig := <-sigCh:
+		log.Printf("received signal: %s", sig)
+	}
+
+}

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

@@ -15,6 +15,7 @@ import (
 	"fmt"
 	"io"
 	"log"
+	"net"
 	"net/http"
 	"net/http/httptest"
 	"net/netip"
@@ -25,6 +26,7 @@ import (
 	"slices"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"go4.org/mem"
@@ -62,6 +64,7 @@ import (
 // Direct is the client that connects to a tailcontrol server for a node.
 type Direct struct {
 	httpc                      *http.Client // HTTP client used to talk to tailcontrol
+	interceptedDial            *atomic.Bool // if non-nil, pointer to bool whether ScreenTime intercepted our dial
 	dialer                     *tsdial.Dialer
 	dnsCache                   *dnscache.Resolver
 	controlKnobs               *controlknobs.Knobs // always non-nil
@@ -80,6 +83,7 @@ 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
 
@@ -257,23 +261,28 @@ func NewDirect(opts Options) (*Direct, error) {
 		// etc set).
 		httpc = http.DefaultClient
 	}
+	var interceptedDial *atomic.Bool
 	if httpc == nil {
 		tr := http.DefaultTransport.(*http.Transport).Clone()
 		tr.Proxy = tshttpproxy.ProxyFromEnvironment
 		tshttpproxy.SetTransportGetProxyConnectHeader(tr)
 		tr.TLSClientConfig = tlsdial.Config(serverURL.Hostname(), opts.HealthTracker, tr.TLSClientConfig)
-		tr.DialContext = dnscache.Dialer(opts.Dialer.SystemDial, dnsCache)
-		tr.DialTLSContext = dnscache.TLSDialer(opts.Dialer.SystemDial, dnsCache, tr.TLSClientConfig)
+		var dialFunc dialFunc
+		dialFunc, interceptedDial = makeScreenTimeDetectingDialFunc(opts.Dialer.SystemDial)
+		tr.DialContext = dnscache.Dialer(dialFunc, dnsCache)
+		tr.DialTLSContext = dnscache.TLSDialer(dialFunc, dnsCache, tr.TLSClientConfig)
 		tr.ForceAttemptHTTP2 = true
 		// Disable implicit gzip compression; the various
 		// handlers (register, map, set-dns, etc) do their own
 		// zstd compression per naclbox.
 		tr.DisableCompression = true
+
 		httpc = &http.Client{Transport: tr}
 	}
 
 	c := &Direct{
 		httpc:                      httpc,
+		interceptedDial:            interceptedDial,
 		controlKnobs:               opts.ControlKnobs,
 		getMachinePrivKey:          opts.GetMachinePrivateKey,
 		serverURL:                  opts.ServerURL,
@@ -310,6 +319,9 @@ 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 +409,12 @@ 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 +449,6 @@ 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
@@ -458,7 +472,20 @@ func (c *Direct) hostInfoLocked() *tailcfg.Hostinfo {
 	return hi
 }
 
+var macOSScreenTime = health.Register(&health.Warnable{
+	Code:     "macos-screen-time-controlclient",
+	Severity: health.SeverityHigh,
+	Title:    "Tailscale blocked by Screen Time",
+	Text: func(args health.Args) string {
+		return "macOS Screen Time seems to be blocking Tailscale. Try disabling Screen Time in System Settings > Screen Time > Content & Privacy > Access to Web Content."
+	},
+	ImpactsConnectivity: true,
+})
+
 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
@@ -496,6 +523,11 @@ func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, new
 	c.logf("doLogin(regen=%v, hasUrl=%v)", regen, opt.URL != "")
 	if serverKey.IsZero() {
 		keys, err := loadServerPubKeys(ctx, c.httpc, c.serverURL)
+		if err != nil && c.interceptedDial != nil && c.interceptedDial.Load() {
+			c.health.SetUnhealthy(macOSScreenTime, nil)
+		} else {
+			c.health.SetHealthy(macOSScreenTime)
+		}
 		if err != nil {
 			return regen, opt.URL, nil, err
 		}
@@ -549,7 +581,7 @@ 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 +632,9 @@ 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 +752,6 @@ 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 +824,9 @@ 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 +1523,9 @@ 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
@@ -1606,9 +1604,9 @@ func postPingResult(start time.Time, logf logger.Logf, c *http.Client, pr *tailc
 }
 
 // ReportHealthChange reports to the control plane a change to this node's
-// health.
-func (c *Direct) ReportHealthChange(sys health.Subsystem, sysErr error) {
-	if sys == health.SysOverall {
+// health. w must be non-nil. us can be nil to indicate a healthy state for w.
+func (c *Direct) ReportHealthChange(w *health.Warnable, us *health.UnhealthyState) {
+	if w == health.NetworkStatusWarnable || w == health.IPNStateWarnable || w == health.LoginStateWarnable {
 		// We don't report these. These include things like the network is down
 		// (in which case we can't report anyway) or the user wanted things
 		// stopped, as opposed to the more unexpected failure types in the other
@@ -1624,12 +1622,16 @@ func (c *Direct) ReportHealthChange(sys health.Subsystem, sysErr error) {
 	if !ok {
 		return
 	}
+	if c.panicOnUse {
+		panic("tainted client")
+	}
+	// TODO(angott): at some point, update `Subsys` in the request to be `Warnable`
 	req := &tailcfg.HealthChangeRequest{
-		Subsys:  string(sys),
+		Subsys:  string(w.Code),
 		NodeKey: nodeKey,
 	}
-	if sysErr != nil {
-		req.Error = sysErr.Error()
+	if us != nil {
+		req.Error = us.Text
 	}
 
 	// Best effort, no logging:
@@ -1685,6 +1687,38 @@ func addLBHeader(req *http.Request, nodeKey key.NodePublic) {
 	}
 }
 
+type dialFunc = func(ctx context.Context, network, addr string) (net.Conn, error)
+
+// makeScreenTimeDetectingDialFunc returns dialFunc, optionally wrapped (on
+// Apple systems) with a func that sets the returned atomic.Bool for whether
+// Screen Time seemed to intercept the connection.
+//
+// The returned *atomic.Bool is nil on non-Apple systems.
+func makeScreenTimeDetectingDialFunc(dial dialFunc) (dialFunc, *atomic.Bool) {
+	switch runtime.GOOS {
+	case "darwin", "ios":
+		// Continue below.
+	default:
+		return dial, nil
+	}
+	ab := new(atomic.Bool)
+	return func(ctx context.Context, network, addr string) (net.Conn, error) {
+		c, err := dial(ctx, network, addr)
+		if err != nil {
+			return nil, err
+		}
+		ab.Store(isTCPLoopback(c.LocalAddr()) && isTCPLoopback(c.RemoteAddr()))
+		return c, nil
+	}, ab
+}
+
+func isTCPLoopback(a net.Addr) bool {
+	if ta, ok := a.(*net.TCPAddr); ok {
+		return ta.IP.IsLoopback()
+	}
+	return false
+}
+
 var (
 	metricMapRequestsActive = clientmetric.NewGauge("controlclient_map_requests_active")
 

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/controlclient/noise.go

@@ -6,10 +6,8 @@ package controlclient
 import (
 	"bytes"
 	"context"
-	"encoding/binary"
 	"encoding/json"
 	"errors"
-	"io"
 	"math"
 	"net/http"
 	"net/url"
@@ -17,9 +15,9 @@ import (
 	"time"
 
 	"golang.org/x/net/http2"
-	"tailscale.com/control/controlbase"
 	"tailscale.com/control/controlhttp"
 	"tailscale.com/health"
+	"tailscale.com/internal/noiseconn"
 	"tailscale.com/net/dnscache"
 	"tailscale.com/net/netmon"
 	"tailscale.com/net/tsdial"
@@ -32,113 +30,6 @@ import (
 	"tailscale.com/util/singleflight"
 )
 
-// noiseConn is a wrapper around controlbase.Conn.
-// It allows attaching an ID to a connection to allow
-// cleaning up references in the pool when the connection
-// is closed.
-type noiseConn struct {
-	*controlbase.Conn
-	id   int
-	pool *NoiseClient
-	h2cc *http2.ClientConn
-
-	readHeaderOnce    sync.Once     // guards init of reader field
-	reader            io.Reader     // (effectively Conn.Reader after header)
-	earlyPayloadReady chan struct{} // closed after earlyPayload is set (including set to nil)
-	earlyPayload      *tailcfg.EarlyNoise
-	earlyPayloadErr   error
-}
-
-func (c *noiseConn) RoundTrip(r *http.Request) (*http.Response, error) {
-	return c.h2cc.RoundTrip(r)
-}
-
-// getEarlyPayload waits for the early noise payload to arrive.
-// It may return (nil, nil) if the server begins HTTP/2 without one.
-func (c *noiseConn) getEarlyPayload(ctx context.Context) (*tailcfg.EarlyNoise, error) {
-	select {
-	case <-c.earlyPayloadReady:
-		return c.earlyPayload, c.earlyPayloadErr
-	case <-ctx.Done():
-		return nil, ctx.Err()
-	}
-}
-
-// The first 9 bytes from the server to client over Noise are either an HTTP/2
-// settings frame (a normal HTTP/2 setup) or, as we added later, an "early payload"
-// header that's also 9 bytes long: 5 bytes (earlyPayloadMagic) followed by 4 bytes
-// of length. Then that many bytes of JSON-encoded tailcfg.EarlyNoise.
-// The early payload is optional. Some servers may not send it.
-const (
-	hdrLen            = 9 // http2 frame header size; also size of our early payload size header
-	earlyPayloadMagic = "\xff\xff\xffTS"
-)
-
-// returnErrReader is an io.Reader that always returns an error.
-type returnErrReader struct {
-	err error // the error to return
-}
-
-func (r returnErrReader) Read([]byte) (int, error) { return 0, r.err }
-
-// Read is basically the same as controlbase.Conn.Read, but it first reads the
-// "early payload" header from the server which may or may not be present,
-// depending on the server.
-func (c *noiseConn) Read(p []byte) (n int, err error) {
-	c.readHeaderOnce.Do(c.readHeader)
-	return c.reader.Read(p)
-}
-
-// readHeader reads the optional "early payload" from the server that arrives
-// after the Noise handshake but before the HTTP/2 session begins.
-//
-// readHeader is responsible for reading the header (if present), initializing
-// c.earlyPayload, closing c.earlyPayloadReady, and initializing c.reader for
-// future reads.
-func (c *noiseConn) readHeader() {
-	defer close(c.earlyPayloadReady)
-
-	setErr := func(err error) {
-		c.reader = returnErrReader{err}
-		c.earlyPayloadErr = err
-	}
-
-	var hdr [hdrLen]byte
-	if _, err := io.ReadFull(c.Conn, hdr[:]); err != nil {
-		setErr(err)
-		return
-	}
-	if string(hdr[:len(earlyPayloadMagic)]) != earlyPayloadMagic {
-		// No early payload. We have to return the 9 bytes read we already
-		// consumed.
-		c.reader = io.MultiReader(bytes.NewReader(hdr[:]), c.Conn)
-		return
-	}
-	epLen := binary.BigEndian.Uint32(hdr[len(earlyPayloadMagic):])
-	if epLen > 10<<20 {
-		setErr(errors.New("invalid early payload length"))
-		return
-	}
-	payBuf := make([]byte, epLen)
-	if _, err := io.ReadFull(c.Conn, payBuf); err != nil {
-		setErr(err)
-		return
-	}
-	if err := json.Unmarshal(payBuf, &c.earlyPayload); err != nil {
-		setErr(err)
-		return
-	}
-	c.reader = c.Conn
-}
-
-func (c *noiseConn) Close() error {
-	if err := c.Conn.Close(); err != nil {
-		return err
-	}
-	c.pool.connClosed(c.id)
-	return nil
-}
-
 // NoiseClient provides a http.Client to connect to tailcontrol over
 // the ts2021 protocol.
 type NoiseClient struct {
@@ -158,7 +49,7 @@ type NoiseClient struct {
 
 	// sfDial ensures that two concurrent requests for a noise connection only
 	// produce one shared one between the two callers.
-	sfDial singleflight.Group[struct{}, *noiseConn]
+	sfDial singleflight.Group[struct{}, *noiseconn.Conn]
 
 	dialer       *tsdial.Dialer
 	dnsCache     *dnscache.Resolver
@@ -180,9 +71,9 @@ type NoiseClient struct {
 	// mu only protects the following variables.
 	mu       sync.Mutex
 	closed   bool
-	last     *noiseConn // or nil
+	last     *noiseconn.Conn // or nil
 	nextID   int
-	connPool map[int]*noiseConn // active connections not yet closed; see noiseConn.Close
+	connPool map[int]*noiseconn.Conn // active connections not yet closed; see noiseconn.Conn.Close
 }
 
 // NoiseOpts contains options for the NewNoiseClient function. All fields are
@@ -283,11 +174,11 @@ func (nc *NoiseClient) GetSingleUseRoundTripper(ctx context.Context) (http.Round
 		if err != nil {
 			return nil, nil, err
 		}
-		earlyPayloadMaybeNil, err := conn.getEarlyPayload(ctx)
+		ok, earlyPayloadMaybeNil, err := conn.ReserveNewRequest(ctx)
 		if err != nil {
 			return nil, nil, err
 		}
-		if conn.h2cc.ReserveNewRequest() {
+		if ok {
 			return conn, earlyPayloadMaybeNil, nil
 		}
 	}
@@ -308,14 +199,14 @@ func (e contextErr) Unwrap() error {
 	return e.err
 }
 
-// getConn returns a noiseConn that can be used to make requests to the
+// getConn returns a noiseconn.Conn that can be used to make requests to the
 // coordination server. It may return a cached connection or create a new one.
 // Dials are singleflighted, so concurrent calls to getConn may only dial once.
 // As such, context values may not be respected as there are no guarantees that
 // the context passed to getConn is the same as the context passed to dial.
-func (nc *NoiseClient) getConn(ctx context.Context) (*noiseConn, error) {
+func (nc *NoiseClient) getConn(ctx context.Context) (*noiseconn.Conn, error) {
 	nc.mu.Lock()
-	if last := nc.last; last != nil && last.canTakeNewRequest() {
+	if last := nc.last; last != nil && last.CanTakeNewRequest() {
 		nc.mu.Unlock()
 		return last, nil
 	}
@@ -327,7 +218,7 @@ func (nc *NoiseClient) getConn(ctx context.Context) (*noiseConn, error) {
 		// canceled. Instead, we have to additionally check that the context
 		// which was canceled is our context and retry if our context is still
 		// valid.
-		conn, err, _ := nc.sfDial.Do(struct{}{}, func() (*noiseConn, error) {
+		conn, err, _ := nc.sfDial.Do(struct{}{}, func() (*noiseconn.Conn, error) {
 			c, err := nc.dial(ctx)
 			if err != nil {
 				if ctx.Err() != nil {
@@ -395,7 +286,7 @@ func (nc *NoiseClient) Close() error {
 
 // dial opens a new connection to tailcontrol, fetching the server noise key
 // if not cached.
-func (nc *NoiseClient) dial(ctx context.Context) (*noiseConn, error) {
+func (nc *NoiseClient) dial(ctx context.Context) (*noiseconn.Conn, error) {
 	nc.mu.Lock()
 	connID := nc.nextID
 	nc.nextID++
@@ -465,18 +356,10 @@ func (nc *NoiseClient) dial(ctx context.Context) (*noiseConn, error) {
 		return nil, err
 	}
 
-	ncc := &noiseConn{
-		Conn:              clientConn.Conn,
-		id:                connID,
-		pool:              nc,
-		earlyPayloadReady: make(chan struct{}),
-	}
-
-	h2cc, err := nc.h2t.NewClientConn(ncc)
+	ncc, err := noiseconn.New(clientConn.Conn, nc.h2t, connID, nc.connClosed)
 	if err != nil {
 		return nil, err
 	}
-	ncc.h2cc = h2cc
 
 	nc.mu.Lock()
 	if nc.closed {
@@ -485,7 +368,7 @@ func (nc *NoiseClient) dial(ctx context.Context) (*noiseConn, error) {
 		return nil, errors.New("noise client closed")
 	}
 	defer nc.mu.Unlock()
-	mak.Set(&nc.connPool, ncc.id, ncc)
+	mak.Set(&nc.connPool, connID, ncc)
 	nc.last = ncc
 	return ncc, nil
 }
@@ -508,9 +391,5 @@ func (nc *NoiseClient) post(ctx context.Context, path string, nodeKey key.NodePu
 	if err != nil {
 		return nil, err
 	}
-	return conn.h2cc.RoundTrip(req)
-}
-
-func (c *noiseConn) canTakeNewRequest() bool {
-	return c.h2cc.CanTakeNewRequest()
+	return conn.RoundTrip(req)
 }

control/controlclient/noise_test.go

@@ -16,6 +16,7 @@ import (
 
 	"golang.org/x/net/http2"
 	"tailscale.com/control/controlhttp"
+	"tailscale.com/internal/noiseconn"
 	"tailscale.com/net/netmon"
 	"tailscale.com/net/tsdial"
 	"tailscale.com/tailcfg"
@@ -93,22 +94,21 @@ func (tt noiseClientTest) run(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	select {
-	case <-c.earlyPayloadReady:
-		gotNonNil := c.earlyPayload != nil
-		if gotNonNil != tt.sendEarlyPayload {
-			t.Errorf("sendEarlyPayload = %v but got earlyPayload = %T", tt.sendEarlyPayload, c.earlyPayload)
-		}
-		if c.earlyPayload != nil {
-			if c.earlyPayload.NodeKeyChallenge != chalPrivate.Public() {
-				t.Errorf("earlyPayload.NodeKeyChallenge = %v; want %v", c.earlyPayload.NodeKeyChallenge, chalPrivate.Public())
-			}
-		}
-
-	case <-ctx.Done():
+	payload, err := c.GetEarlyPayload(ctx)
+	if err != nil {
 		t.Fatal("timed out waiting for didReadHeaderCh")
 	}
 
+	gotNonNil := payload != nil
+	if gotNonNil != tt.sendEarlyPayload {
+		t.Errorf("sendEarlyPayload = %v but got earlyPayload = %T", tt.sendEarlyPayload, payload)
+	}
+	if payload != nil {
+		if payload.NodeKeyChallenge != chalPrivate.Public() {
+			t.Errorf("earlyPayload.NodeKeyChallenge = %v; want %v", payload.NodeKeyChallenge, chalPrivate.Public())
+		}
+	}
+
 	checkRes := func(t *testing.T, res *http.Response) {
 		t.Helper()
 		defer res.Body.Close()
@@ -184,7 +184,7 @@ func (up *Upgrader) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		// https://httpwg.org/specs/rfc7540.html#rfc.section.4.1 (Especially not
 		// an HTTP/2 settings frame, which isn't of type 'T')
 		var notH2Frame [5]byte
-		copy(notH2Frame[:], earlyPayloadMagic)
+		copy(notH2Frame[:], noiseconn.EarlyPayloadMagic)
 		var lenBuf [4]byte
 		binary.BigEndian.PutUint32(lenBuf[:], uint32(len(earlyJSON)))
 		// These writes are all buffered by caller, so fine to do them

control/controlhttp/client.go

@@ -32,12 +32,14 @@ import (
 	"net/http/httptrace"
 	"net/netip"
 	"net/url"
+	"runtime"
 	"sort"
 	"sync/atomic"
 	"time"
 
 	"tailscale.com/control/controlbase"
 	"tailscale.com/envknob"
+	"tailscale.com/health"
 	"tailscale.com/net/dnscache"
 	"tailscale.com/net/dnsfallback"
 	"tailscale.com/net/netutil"
@@ -396,12 +398,29 @@ func (a *Dialer) resolver() *dnscache.Resolver {
 	}
 }
 
+func isLoopback(a net.Addr) bool {
+	if ta, ok := a.(*net.TCPAddr); ok {
+		return ta.IP.IsLoopback()
+	}
+	return false
+}
+
+var macOSScreenTime = health.Register(&health.Warnable{
+	Code:     "macos-screen-time",
+	Severity: health.SeverityHigh,
+	Title:    "Tailscale blocked by Screen Time",
+	Text: func(args health.Args) string {
+		return "macOS Screen Time seems to be blocking Tailscale. Try disabling Screen Time in System Settings > Screen Time > Content & Privacy > Access to Web Content."
+	},
+	ImpactsConnectivity: true,
+})
+
 // tryURLUpgrade connects to u, and tries to upgrade it to a net.Conn. If addr
 // is valid, then no DNS is used and the connection will be made to the
 // provided address.
 //
 // Only the provided ctx is used, not a.ctx.
-func (a *Dialer) tryURLUpgrade(ctx context.Context, u *url.URL, addr netip.Addr, init []byte) (net.Conn, error) {
+func (a *Dialer) tryURLUpgrade(ctx context.Context, u *url.URL, addr netip.Addr, init []byte) (_ net.Conn, retErr error) {
 	var dns *dnscache.Resolver
 
 	// If we were provided an address to dial, then create a resolver that just
@@ -423,6 +442,30 @@ func (a *Dialer) tryURLUpgrade(ctx context.Context, u *url.URL, addr netip.Addr,
 		dialer = stdDialer.DialContext
 	}
 
+	// On macOS, see if Screen Time is blocking things.
+	if runtime.GOOS == "darwin" {
+		var proxydIntercepted atomic.Bool // intercepted by macOS webfilterproxyd
+		origDialer := dialer
+		dialer = func(ctx context.Context, network, address string) (net.Conn, error) {
+			c, err := origDialer(ctx, network, address)
+			if err != nil {
+				return nil, err
+			}
+			if isLoopback(c.LocalAddr()) && isLoopback(c.RemoteAddr()) {
+				proxydIntercepted.Store(true)
+			}
+			return c, nil
+		}
+		defer func() {
+			if retErr != nil && proxydIntercepted.Load() {
+				a.HealthTracker.SetUnhealthy(macOSScreenTime, nil)
+				retErr = fmt.Errorf("macOS Screen Time is blocking network access: %w", retErr)
+			} else {
+				a.HealthTracker.SetHealthy(macOSScreenTime)
+			}
+		}()
+	}
+
 	tr := http.DefaultTransport.(*http.Transport).Clone()
 	defer tr.CloseIdleConnections()
 	tr.Proxy = a.getProxyFunc()

control/controlknobs/controlknobs.go

@@ -76,6 +76,29 @@ 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
+
+	// DisableSplitDNSWhenNoCustomResolvers indicates that the node's DNS manager
+	// should not adopt a split DNS configuration even though the Config of the
+	// resolver only contains routes that do not specify custom resolver(s), hence
+	// all DNS queries can be safely sent to the upstream DNS resolver and the
+	// node's DNS forwarder doesn't need to handle all DNS traffic.
+	// This is for now (2024-06-06) an iOS-specific battery life optimization,
+	// and this knob allows us to disable the optimization remotely if needed.
+	DisableSplitDNSWhenNoCustomResolvers atomic.Bool
+
+	// DisableLocalDNSOverrideViaNRPT indicates that the node's DNS manager should not
+	// create a default (catch-all) Windows NRPT rule when "Override local DNS" is enabled.
+	// Without this rule, Windows 8.1 and newer devices issue parallel DNS requests to DNS servers
+	// associated with all network adapters, even when "Override local DNS" is enabled and/or
+	// a Mullvad exit node is being used, resulting in DNS leaks.
+	// We began creating this rule on 2024-06-14, and this knob
+	// allows us to disable the new behavior remotely if needed.
+	DisableLocalDNSOverrideViaNRPT atomic.Bool
 }
 
 // UpdateFromNodeAttributes updates k (if non-nil) based on the provided self
@@ -86,21 +109,24 @@ func (k *Knobs) UpdateFromNodeAttributes(capMap tailcfg.NodeCapMap) {
 	}
 	has := capMap.Contains
 	var (
-		keepFullWG                    = has(tailcfg.NodeAttrDebugDisableWGTrim)
-		disableDRPO                   = has(tailcfg.NodeAttrDebugDisableDRPO)
-		disableUPnP                   = has(tailcfg.NodeAttrDisableUPnP)
-		randomizeClientPort           = has(tailcfg.NodeAttrRandomizeClientPort)
-		disableDeltaUpdates           = has(tailcfg.NodeAttrDisableDeltaUpdates)
-		oneCGNAT                      opt.Bool
-		forceBackgroundSTUN           = has(tailcfg.NodeAttrDebugForceBackgroundSTUN)
-		peerMTUEnable                 = has(tailcfg.NodeAttrPeerMTUEnable)
-		dnsForwarderDisableTCPRetries = has(tailcfg.NodeAttrDNSForwarderDisableTCPRetries)
-		silentDisco                   = has(tailcfg.NodeAttrSilentDisco)
-		forceIPTables                 = has(tailcfg.NodeAttrLinuxMustUseIPTables)
-		forceNfTables                 = has(tailcfg.NodeAttrLinuxMustUseNfTables)
-		seamlessKeyRenewal            = has(tailcfg.NodeAttrSeamlessKeyRenewal)
-		probeUDPLifetime              = has(tailcfg.NodeAttrProbeUDPLifetime)
-		appCStoreRoutes               = has(tailcfg.NodeAttrStoreAppCRoutes)
+		keepFullWG                           = has(tailcfg.NodeAttrDebugDisableWGTrim)
+		disableDRPO                          = has(tailcfg.NodeAttrDebugDisableDRPO)
+		disableUPnP                          = has(tailcfg.NodeAttrDisableUPnP)
+		randomizeClientPort                  = has(tailcfg.NodeAttrRandomizeClientPort)
+		disableDeltaUpdates                  = has(tailcfg.NodeAttrDisableDeltaUpdates)
+		oneCGNAT                             opt.Bool
+		forceBackgroundSTUN                  = has(tailcfg.NodeAttrDebugForceBackgroundSTUN)
+		peerMTUEnable                        = has(tailcfg.NodeAttrPeerMTUEnable)
+		dnsForwarderDisableTCPRetries        = has(tailcfg.NodeAttrDNSForwarderDisableTCPRetries)
+		silentDisco                          = has(tailcfg.NodeAttrSilentDisco)
+		forceIPTables                        = has(tailcfg.NodeAttrLinuxMustUseIPTables)
+		forceNfTables                        = has(tailcfg.NodeAttrLinuxMustUseNfTables)
+		seamlessKeyRenewal                   = has(tailcfg.NodeAttrSeamlessKeyRenewal)
+		probeUDPLifetime                     = has(tailcfg.NodeAttrProbeUDPLifetime)
+		appCStoreRoutes                      = has(tailcfg.NodeAttrStoreAppCRoutes)
+		userDialUseRoutes                    = has(tailcfg.NodeAttrUserDialUseRoutes)
+		disableSplitDNSWhenNoCustomResolvers = has(tailcfg.NodeAttrDisableSplitDNSWhenNoCustomResolvers)
+		disableLocalDNSOverrideViaNRPT       = has(tailcfg.NodeAttrDisableLocalDNSOverrideViaNRPT)
 	)
 
 	if has(tailcfg.NodeAttrOneCGNATEnable) {
@@ -124,6 +150,9 @@ func (k *Knobs) UpdateFromNodeAttributes(capMap tailcfg.NodeCapMap) {
 	k.SeamlessKeyRenewal.Store(seamlessKeyRenewal)
 	k.ProbeUDPLifetime.Store(probeUDPLifetime)
 	k.AppCStoreRoutes.Store(appCStoreRoutes)
+	k.UserDialUseRoutes.Store(userDialUseRoutes)
+	k.DisableSplitDNSWhenNoCustomResolvers.Store(disableSplitDNSWhenNoCustomResolvers)
+	k.DisableLocalDNSOverrideViaNRPT.Store(disableLocalDNSOverrideViaNRPT)
 }
 
 // AsDebugJSON returns k as something that can be marshalled with json.Marshal
@@ -133,20 +162,23 @@ func (k *Knobs) AsDebugJSON() map[string]any {
 		return nil
 	}
 	return map[string]any{
-		"DisableUPnP":                   k.DisableUPnP.Load(),
-		"DisableDRPO":                   k.DisableDRPO.Load(),
-		"KeepFullWGConfig":              k.KeepFullWGConfig.Load(),
-		"RandomizeClientPort":           k.RandomizeClientPort.Load(),
-		"OneCGNAT":                      k.OneCGNAT.Load(),
-		"ForceBackgroundSTUN":           k.ForceBackgroundSTUN.Load(),
-		"DisableDeltaUpdates":           k.DisableDeltaUpdates.Load(),
-		"PeerMTUEnable":                 k.PeerMTUEnable.Load(),
-		"DisableDNSForwarderTCPRetries": k.DisableDNSForwarderTCPRetries.Load(),
-		"SilentDisco":                   k.SilentDisco.Load(),
-		"LinuxForceIPTables":            k.LinuxForceIPTables.Load(),
-		"LinuxForceNfTables":            k.LinuxForceNfTables.Load(),
-		"SeamlessKeyRenewal":            k.SeamlessKeyRenewal.Load(),
-		"ProbeUDPLifetime":              k.ProbeUDPLifetime.Load(),
-		"AppCStoreRoutes":               k.AppCStoreRoutes.Load(),
+		"DisableUPnP":                          k.DisableUPnP.Load(),
+		"DisableDRPO":                          k.DisableDRPO.Load(),
+		"KeepFullWGConfig":                     k.KeepFullWGConfig.Load(),
+		"RandomizeClientPort":                  k.RandomizeClientPort.Load(),
+		"OneCGNAT":                             k.OneCGNAT.Load(),
+		"ForceBackgroundSTUN":                  k.ForceBackgroundSTUN.Load(),
+		"DisableDeltaUpdates":                  k.DisableDeltaUpdates.Load(),
+		"PeerMTUEnable":                        k.PeerMTUEnable.Load(),
+		"DisableDNSForwarderTCPRetries":        k.DisableDNSForwarderTCPRetries.Load(),
+		"SilentDisco":                          k.SilentDisco.Load(),
+		"LinuxForceIPTables":                   k.LinuxForceIPTables.Load(),
+		"LinuxForceNfTables":                   k.LinuxForceNfTables.Load(),
+		"SeamlessKeyRenewal":                   k.SeamlessKeyRenewal.Load(),
+		"ProbeUDPLifetime":                     k.ProbeUDPLifetime.Load(),
+		"AppCStoreRoutes":                      k.AppCStoreRoutes.Load(),
+		"UserDialUseRoutes":                    k.UserDialUseRoutes.Load(),
+		"DisableSplitDNSWhenNoCustomResolvers": k.DisableSplitDNSWhenNoCustomResolvers.Load(),
+		"DisableLocalDNSOverrideViaNRPT":       k.DisableLocalDNSOverrideViaNRPT.Load(),
 	}
 }

derp/derp_server.go

@@ -22,7 +22,7 @@ import (
 	"log"
 	"math"
 	"math/big"
-	"math/rand"
+	"math/rand/v2"
 	"net"
 	"net/http"
 	"net/netip"
@@ -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.
 //
@@ -709,7 +725,7 @@ func (s *Server) accept(ctx context.Context, nc Conn, brw *bufio.ReadWriter, rem
 
 	clientAP, _ := netip.ParseAddrPort(remoteAddr)
 	if err := s.verifyClient(ctx, clientKey, clientInfo, clientAP.Addr()); err != nil {
-		return fmt.Errorf("client %x rejected: %v", clientKey, err)
+		return fmt.Errorf("client %v rejected: %v", clientKey, err)
 	}
 
 	// At this point we trust the client so we don't time out.
@@ -735,7 +751,7 @@ func (s *Server) accept(ctx context.Context, nc Conn, brw *bufio.ReadWriter, rem
 		discoSendQueue: make(chan pkt, perClientSendQueueDepth),
 		sendPongCh:     make(chan [8]byte, 1),
 		peerGone:       make(chan peerGoneMsg),
-		canMesh:        clientInfo.MeshKey != "" && clientInfo.MeshKey == s.meshKey,
+		canMesh:        s.isMeshPeer(clientInfo),
 		peerGoneLim:    rate.NewLimiter(rate.Every(time.Second), 3),
 	}
 
@@ -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)
@@ -922,7 +939,7 @@ func (c *sclient) handleFrameForwardPacket(ft frameType, fl uint32) error {
 
 	srcKey, dstKey, contents, err := s.recvForwardPacket(c.br, fl)
 	if err != nil {
-		return fmt.Errorf("client %x: recvForwardPacket: %v", c.key, err)
+		return fmt.Errorf("client %v: recvForwardPacket: %v", c.key, err)
 	}
 	s.packetsForwardedIn.Add(1)
 
@@ -977,7 +994,7 @@ func (c *sclient) handleFrameSendPacket(ft frameType, fl uint32) error {
 
 	dstKey, contents, err := s.recvPacket(c.br, fl)
 	if err != nil {
-		return fmt.Errorf("client %x: recvPacket: %v", c.key, err)
+		return fmt.Errorf("client %v: recvPacket: %v", c.key, err)
 	}
 
 	var fwd PacketForwarder
@@ -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) {
@@ -1133,20 +1151,32 @@ func (c *sclient) requestMeshUpdate() {
 	}
 }
 
+var localClient tailscale.LocalClient
+
+// isMeshPeer reports whether the client is a trusted mesh peer
+// node in the DERP region.
+func (s *Server) isMeshPeer(info *clientInfo) bool {
+	return info != nil && info.MeshKey != "" && info.MeshKey == s.meshKey
+}
+
 // verifyClient checks whether the client is allowed to connect to the derper,
 // depending on how & whether the server's been configured to verify.
 func (s *Server) verifyClient(ctx context.Context, clientKey key.NodePublic, info *clientInfo, clientIP netip.Addr) error {
+	if s.isMeshPeer(info) {
+		// Trusted mesh peer. No need to verify further. In fact, verifying
+		// further wouldn't work: it's not part of the tailnet so tailscaled and
+		// likely the admission control URL wouldn't know about it.
+		return nil
+	}
+
 	// tailscaled-based verification:
 	if s.verifyClientsLocalTailscaled {
-		status, err := tailscale.Status(ctx)
+		_, err := localClient.WhoIsNodeKey(ctx, clientKey)
+		if err == tailscale.ErrPeerNotFound {
+			return fmt.Errorf("peer %v not authorized (not found in local tailscaled)", clientKey)
+		}
 		if err != nil {
-			return fmt.Errorf("failed to query local tailscaled status: %w", err)
-		}
-		if clientKey == status.Self.PublicKey {
-			return nil
-		}
-		if _, exists := status.Peer[clientKey]; !exists {
-			return fmt.Errorf("client %v not in set of peers", clientKey)
+			return fmt.Errorf("failed to query local tailscaled status for %v: %w", clientKey, err)
 		}
 	}
 
@@ -1496,7 +1526,7 @@ func (c *sclient) sendLoop(ctx context.Context) error {
 		}
 	}()
 
-	jitter := time.Duration(rand.Intn(5000)) * time.Millisecond
+	jitter := rand.N(5 * time.Second)
 	keepAliveTick, keepAliveTickChannel := c.s.clock.NewTicker(keepAlive + jitter)
 	defer keepAliveTick.Stop()
 
@@ -1938,12 +1968,37 @@ func (s *Server) ConsistencyCheck() error {
 			s.curClients.Value(),
 			len(s.clients)))
 	}
+
+	if s.verifyClientsLocalTailscaled {
+		if err := s.checkVerifyClientsLocalTailscaled(); err != nil {
+			errs = append(errs, err.Error())
+		}
+	}
+
 	if len(errs) == 0 {
 		return nil
 	}
 	return errors.New(strings.Join(errs, ", "))
 }
 
+// checkVerifyClientsLocalTailscaled checks that a verifyClients call can be made successfully for the derper hosts own node key.
+func (s *Server) checkVerifyClientsLocalTailscaled() error {
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	status, err := localClient.StatusWithoutPeers(ctx)
+	if err != nil {
+		return fmt.Errorf("localClient.Status: %w", err)
+	}
+	info := &clientInfo{
+		IsProber: true,
+	}
+	clientIP := netip.IPv6Loopback()
+	if err := s.verifyClient(ctx, status.Self.PublicKey, info, clientIP); err != nil {
+		return fmt.Errorf("verifyClient for self nodekey: %w", err)
+	}
+	return nil
+}
+
 const minTimeBetweenLogs = 2 * time.Second
 
 // BytesSentRecv records the number of bytes that have been sent since the last traffic check

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/derphttp/derphttp_client.go

@@ -136,8 +136,15 @@ func NewRegionClient(privateKey key.NodePrivate, logf logger.Logf, netMon *netmo
 
 // NewNetcheckClient returns a Client that's only able to have its DialRegionTLS method called.
 // It's used by the netcheck package.
-func NewNetcheckClient(logf logger.Logf) *Client {
-	return &Client{logf: logf, clock: tstime.StdClock{}}
+func NewNetcheckClient(logf logger.Logf, netMon *netmon.Monitor) *Client {
+	if netMon == nil {
+		panic("nil netMon")
+	}
+	return &Client{
+		logf:   logf,
+		clock:  tstime.StdClock{},
+		netMon: netMon,
+	}
 }
 
 // NewClient returns a new DERP-over-HTTP client. It connects lazily.

derp/derphttp/derphttp_server.go

@@ -20,6 +20,16 @@ const fastStartHeader = "Derp-Fast-Start"
 
 func Handler(s *derp.Server) http.Handler {
 	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// These are installed both here and in cmd/derper. The check here
+		// catches both cmd/derper run with DERP disabled (STUN only mode) as
+		// well as DERP being run in tests with derphttp.Handler directly,
+		// as netcheck still assumes this replies.
+		switch r.URL.Path {
+		case "/derp/probe", "/derp/latency-check":
+			ProbeHandler(w, r)
+			return
+		}
+
 		up := strings.ToLower(r.Header.Get("Upgrade"))
 		if up != "websocket" && up != "derp" {
 			if up != "" {
@@ -58,3 +68,14 @@ func Handler(s *derp.Server) http.Handler {
 		s.Accept(r.Context(), netConn, conn, netConn.RemoteAddr().String())
 	})
 }
+
+// ProbeHandler is the endpoint that clients without UDP access (including js/wasm) hit to measure
+// DERP latency, as a replacement for UDP STUN queries.
+func ProbeHandler(w http.ResponseWriter, r *http.Request) {
+	switch r.Method {
+	case "HEAD", "GET":
+		w.Header().Set("Access-Control-Allow-Origin", "*")
+	default:
+		http.Error(w, "bogus probe method", http.StatusMethodNotAllowed)
+	}
+}

derp/derphttp/derphttp_test.go

@@ -10,6 +10,7 @@ import (
 	"fmt"
 	"net"
 	"net/http"
+	"net/http/httptest"
 	"net/netip"
 	"sync"
 	"testing"
@@ -464,3 +465,24 @@ func TestLocalAddrNoMutex(t *testing.T) {
 		t.Errorf("got error %q; want %q", got, want)
 	}
 }
+
+func TestProbe(t *testing.T) {
+	h := Handler(nil)
+
+	tests := []struct {
+		path string
+		want int
+	}{
+		{"/derp/probe", 200},
+		{"/derp/latency-check", 200},
+		{"/derp/sdf", http.StatusUpgradeRequired},
+	}
+
+	for _, tt := range tests {
+		rec := httptest.NewRecorder()
+		h.ServeHTTP(rec, httptest.NewRequest("GET", tt.path, nil))
+		if got := rec.Result().StatusCode; got != tt.want {
+			t.Errorf("for path %q got HTTP status %v; want %v", tt.path, got, tt.want)
+		}
+	}
+}

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) {

derp/xdp/bpf_bpfeb.go

@@ -0,0 +1,170 @@
+// Code generated by bpf2go; DO NOT EDIT.
+//go:build mips || mips64 || ppc64 || s390x
+
+package xdp
+
+import (
+	"bytes"
+	_ "embed"
+	"fmt"
+	"io"
+
+	"github.com/cilium/ebpf"
+)
+
+type bpfConfig struct {
+	DstPort  uint16
+	DropStun uint16
+}
+
+type bpfCounterKeyAf uint32
+
+const (
+	bpfCounterKeyAfCOUNTER_KEY_AF_UNKNOWN bpfCounterKeyAf = 0
+	bpfCounterKeyAfCOUNTER_KEY_AF_IPV4    bpfCounterKeyAf = 1
+	bpfCounterKeyAfCOUNTER_KEY_AF_IPV6    bpfCounterKeyAf = 2
+	bpfCounterKeyAfCOUNTER_KEY_AF_LEN     bpfCounterKeyAf = 3
+)
+
+type bpfCounterKeyPacketsBytesAction uint32
+
+const (
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_PACKETS_PASS_TOTAL       bpfCounterKeyPacketsBytesAction = 0
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_BYTES_PASS_TOTAL         bpfCounterKeyPacketsBytesAction = 1
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_PACKETS_ABORTED_TOTAL    bpfCounterKeyPacketsBytesAction = 2
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_BYTES_ABORTED_TOTAL      bpfCounterKeyPacketsBytesAction = 3
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_PACKETS_TX_TOTAL         bpfCounterKeyPacketsBytesAction = 4
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_BYTES_TX_TOTAL           bpfCounterKeyPacketsBytesAction = 5
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_PACKETS_DROP_TOTAL       bpfCounterKeyPacketsBytesAction = 6
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_BYTES_DROP_TOTAL         bpfCounterKeyPacketsBytesAction = 7
+	bpfCounterKeyPacketsBytesActionCOUNTER_KEY_PACKETS_BYTES_ACTION_LEN bpfCounterKeyPacketsBytesAction = 8
+)
+
+type bpfCounterKeyProgEnd uint32
+
+const (
+	bpfCounterKeyProgEndCOUNTER_KEY_END_UNSPECIFIED                bpfCounterKeyProgEnd = 0
+	bpfCounterKeyProgEndCOUNTER_KEY_END_UNEXPECTED_FIRST_STUN_ATTR bpfCounterKeyProgEnd = 1
+	bpfCounterKeyProgEndCOUNTER_KEY_END_INVALID_UDP_CSUM           bpfCounterKeyProgEnd = 2
+	bpfCounterKeyProgEndCOUNTER_KEY_END_INVALID_IP_CSUM            bpfCounterKeyProgEnd = 3
+	bpfCounterKeyProgEndCOUNTER_KEY_END_NOT_STUN_PORT              bpfCounterKeyProgEnd = 4
+	bpfCounterKeyProgEndCOUNTER_KEY_END_INVALID_SW_ATTR_VAL        bpfCounterKeyProgEnd = 5
+	bpfCounterKeyProgEndCOUNTER_KEY_END_DROP_STUN                  bpfCounterKeyProgEnd = 6
+	bpfCounterKeyProgEndCOUNTER_KEY_END_LEN                        bpfCounterKeyProgEnd = 7
+)
+
+type bpfCountersKey struct {
+	Unused  uint8
+	Af      uint8
+	Pba     uint8
+	ProgEnd uint8
+}
+
+// loadBpf returns the embedded CollectionSpec for bpf.
+func loadBpf() (*ebpf.CollectionSpec, error) {
+	reader := bytes.NewReader(_BpfBytes)
+	spec, err := ebpf.LoadCollectionSpecFromReader(reader)
+	if err != nil {
+		return nil, fmt.Errorf("can't load bpf: %w", err)
+	}
+
+	return spec, err
+}
+
+// loadBpfObjects loads bpf and converts it into a struct.
+//
+// The following types are suitable as obj argument:
+//
+//	*bpfObjects
+//	*bpfPrograms
+//	*bpfMaps
+//
+// See ebpf.CollectionSpec.LoadAndAssign documentation for details.
+func loadBpfObjects(obj interface{}, opts *ebpf.CollectionOptions) error {
+	spec, err := loadBpf()
+	if err != nil {
+		return err
+	}
+
+	return spec.LoadAndAssign(obj, opts)
+}
+
+// bpfSpecs contains maps and programs before they are loaded into the kernel.
+//
+// It can be passed ebpf.CollectionSpec.Assign.
+type bpfSpecs struct {
+	bpfProgramSpecs
+	bpfMapSpecs
+}
+
+// bpfSpecs contains programs before they are loaded into the kernel.
+//
+// It can be passed ebpf.CollectionSpec.Assign.
+type bpfProgramSpecs struct {
+	XdpProgFunc *ebpf.ProgramSpec `ebpf:"xdp_prog_func"`
+}
+
+// bpfMapSpecs contains maps before they are loaded into the kernel.
+//
+// It can be passed ebpf.CollectionSpec.Assign.
+type bpfMapSpecs struct {
+	ConfigMap   *ebpf.MapSpec `ebpf:"config_map"`
+	CountersMap *ebpf.MapSpec `ebpf:"counters_map"`
+}
+
+// bpfObjects contains all objects after they have been loaded into the kernel.
+//
+// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
+type bpfObjects struct {
+	bpfPrograms
+	bpfMaps
+}
+
+func (o *bpfObjects) Close() error {
+	return _BpfClose(
+		&o.bpfPrograms,
+		&o.bpfMaps,
+	)
+}
+
+// bpfMaps contains all maps after they have been loaded into the kernel.
+//
+// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
+type bpfMaps struct {
+	ConfigMap   *ebpf.Map `ebpf:"config_map"`
+	CountersMap *ebpf.Map `ebpf:"counters_map"`
+}
+
+func (m *bpfMaps) Close() error {
+	return _BpfClose(
+		m.ConfigMap,
+		m.CountersMap,
+	)
+}
+
+// bpfPrograms contains all programs after they have been loaded into the kernel.
+//
+// It can be passed to loadBpfObjects or ebpf.CollectionSpec.LoadAndAssign.
+type bpfPrograms struct {
+	XdpProgFunc *ebpf.Program `ebpf:"xdp_prog_func"`
+}
+
+func (p *bpfPrograms) Close() error {
+	return _BpfClose(
+		p.XdpProgFunc,
+	)
+}
+
+func _BpfClose(closers ...io.Closer) error {
+	for _, closer := range closers {
+		if err := closer.Close(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// Do not access this directly.
+//
+//go:embed bpf_bpfeb.o
+var _BpfBytes []byte