更新 cmd/derper/cert.go

将与域名验证相关的内容删除或注释

ALPINE.txt

@@ -1 +1 @@
-3.18
+3.19

Dockerfile

@@ -62,8 +62,10 @@ RUN GOARCH=$TARGETARCH go install -ldflags="\
       -X tailscale.com/version.gitCommitStamp=$VERSION_GIT_HASH" \
       -v ./cmd/tailscale ./cmd/tailscaled ./cmd/containerboot
 
-FROM alpine:3.18
+FROM alpine:3.19
 RUN apk add --no-cache ca-certificates iptables iproute2 ip6tables
+RUN rm /sbin/iptables && ln -s /sbin/iptables-legacy /sbin/iptables
+RUN rm /sbin/ip6tables && ln -s /sbin/ip6tables-legacy /sbin/ip6tables
 
 COPY --from=build-env /go/bin/* /usr/local/bin/
 # For compat with the previous run.sh, although ideally you should be

Dockerfile.base

@@ -1,5 +1,12 @@
 # Copyright (c) Tailscale Inc & AUTHORS
 # SPDX-License-Identifier: BSD-3-Clause
 
-FROM alpine:3.18
-RUN apk add --no-cache ca-certificates iptables iproute2 ip6tables iputils
+FROM alpine:3.19
+RUN apk add --no-cache ca-certificates iptables iptables-legacy iproute2 ip6tables iputils
+# Alpine 3.19 replaces legacy iptables with nftables based implementation.  We
+# can't be certain that all hosts that run Tailscale containers currently
+# suppport nftables, so link back to legacy for backwards compatibility reasons.
+# TODO(irbekrm): add some way how to determine if we still run on nodes that
+# don't support nftables, so that we can eventually remove these symlinks.
+RUN rm /sbin/iptables && ln -s /sbin/iptables-legacy /sbin/iptables
+RUN rm /sbin/ip6tables && ln -s /sbin/ip6tables-legacy /sbin/ip6tables

VERSION.txt

@@ -1 +1 @@
-1.81.0
+1.83.0

build_docker.sh

@@ -16,7 +16,7 @@ eval "$(./build_dist.sh shellvars)"
 
 DEFAULT_TARGET="client"
 DEFAULT_TAGS="v${VERSION_SHORT},v${VERSION_MINOR}"
-DEFAULT_BASE="tailscale/alpine-base:3.18"
+DEFAULT_BASE="tailscale/alpine-base:3.19"
 # Set a few pre-defined OCI annotations. The source annotation is used by tools such as Renovate that scan the linked
 # Github repo to find release notes for any new image tags. Note that for official Tailscale images the default
 # annotations defined here will be overriden by release scripts that call this script.

client/tailscale/devices.go

@@ -79,6 +79,13 @@ type Device struct {
 	// Tailscale have attempted to collect this from the device but it has not
 	// opted in, PostureIdentity will have Disabled=true.
 	PostureIdentity *DevicePostureIdentity `json:"postureIdentity"`
+
+	// TailnetLockKey is the tailnet lock public key of the node as a hex string.
+	TailnetLockKey string `json:"tailnetLockKey,omitempty"`
+
+	// TailnetLockErr indicates an issue with the tailnet lock node-key signature
+	// on this device. This field is only populated when tailnet lock is enabled.
+	TailnetLockErr string `json:"tailnetLockError,omitempty"`
 }
 
 type DevicePostureIdentity struct {

client/web/web.go

@@ -335,7 +335,8 @@ func (s *Server) requireTailscaleIP(w http.ResponseWriter, r *http.Request) (han
 		ipv6ServiceHost = "[" + tsaddr.TailscaleServiceIPv6String + "]"
 	)
 	// allow requests on quad-100 (or ipv6 equivalent)
-	if r.Host == ipv4ServiceHost || r.Host == ipv6ServiceHost {
+	host := strings.TrimSuffix(r.Host, ":80")
+	if host == ipv4ServiceHost || host == ipv6ServiceHost {
 		return false
 	}
 

client/web/web_test.go

@@ -1177,6 +1177,16 @@ func TestRequireTailscaleIP(t *testing.T) {
 			target:      "http://[fd7a:115c:a1e0::53]/",
 			wantHandled: false,
 		},
+		{
+			name:        "quad-100:80",
+			target:      "http://100.100.100.100:80/",
+			wantHandled: false,
+		},
+		{
+			name:        "ipv6-service-addr:80",
+			target:      "http://[fd7a:115c:a1e0::53]:80/",
+			wantHandled: false,
+		},
 	}
 
 	for _, tt := range tests {

clientupdate/clientupdate.go

@@ -28,6 +28,7 @@ import (
 	"strings"
 
 	"tailscale.com/hostinfo"
+	"tailscale.com/types/lazy"
 	"tailscale.com/types/logger"
 	"tailscale.com/util/cmpver"
 	"tailscale.com/version"
@@ -249,9 +250,13 @@ func (up *Updater) getUpdateFunction() (fn updateFunction, canAutoUpdate bool) {
 	return nil, false
 }
 
+var canAutoUpdateCache lazy.SyncValue[bool]
+
 // CanAutoUpdate reports whether auto-updating via the clientupdate package
 // is supported for the current os/distro.
-func CanAutoUpdate() bool {
+func CanAutoUpdate() bool { return canAutoUpdateCache.Get(canAutoUpdateUncached) }
+
+func canAutoUpdateUncached() bool {
 	if version.IsMacSysExt() {
 		// Macsys uses Sparkle for auto-updates, which doesn't have an update
 		// function in this package.

cmd/containerboot/certs.go

@@ -60,6 +60,9 @@ func (cm *certManager) ensureCertLoops(ctx context.Context, sc *ipn.ServeConfig)
 		if _, exists := cm.certLoops[domain]; !exists {
 			cancelCtx, cancel := context.WithCancel(ctx)
 			mak.Set(&cm.certLoops, domain, cancel)
+			// Note that most of the issuance anyway happens
+			// serially because the cert client has a shared lock
+			// that's held during any issuance.
 			cm.tracker.Go(func() { cm.runCertLoop(cancelCtx, domain) })
 		}
 	}
@@ -116,7 +119,13 @@ func (cm *certManager) runCertLoop(ctx context.Context, domain string) {
 			// issuance endpoint that explicitly only triggers
 			// issuance and stores certs in the relevant store, but
 			// does not return certs to the caller?
-			_, _, err := cm.lc.CertPair(ctx, domain)
+
+			// An issuance holds a shared lock, so we need to avoid
+			// a situation where other services cannot issue certs
+			// because a single one is holding the lock.
+			ctxT, cancel := context.WithTimeout(ctx, time.Second*300)
+			defer cancel()
+			_, _, err := cm.lc.CertPair(ctxT, domain)
 			if err != nil {
 				log.Printf("error refreshing certificate for %s: %v", domain, err)
 			}

cmd/derper/cert.go

@@ -128,16 +128,17 @@ func (m *manualCertManager) TLSConfig() *tls.Config {
 }
 
 func (m *manualCertManager) getCertificate(hi *tls.ClientHelloInfo) (*tls.Certificate, error) {
-	if hi.ServerName != m.hostname && !m.noHostname {
-		return nil, fmt.Errorf("cert mismatch with hostname: %q", hi.ServerName)
-	}
+	// if hi.ServerName != m.hostname && !m.noHostname {
+	// 	return nil, fmt.Errorf("cert mismatch with hostname: %q", hi.ServerName)
+	// }
 
 	// Return a shallow copy of the cert so the caller can append to its
 	// Certificate field.
-	certCopy := new(tls.Certificate)
-	*certCopy = *m.cert
-	certCopy.Certificate = certCopy.Certificate[:len(certCopy.Certificate):len(certCopy.Certificate)]
-	return certCopy, nil
+	// certCopy := new(tls.Certificate)
+	// *certCopy = *m.cert
+	// certCopy.Certificate = certCopy.Certificate[:len(certCopy.Certificate):len(certCopy.Certificate)]
+	// return certCopy, nil
+	return m.cert, nil
 }
 
 func (m *manualCertManager) HTTPHandler(fallback http.Handler) http.Handler {

cmd/derper/depaware.txt

@@ -96,6 +96,7 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         tailscale.com/disco                                          from tailscale.com/derp
         tailscale.com/drive                                          from tailscale.com/client/local+
         tailscale.com/envknob                                        from tailscale.com/client/local+
+        tailscale.com/feature                                        from tailscale.com/tsweb
         tailscale.com/health                                         from tailscale.com/net/tlsdial+
         tailscale.com/hostinfo                                       from tailscale.com/net/netmon+
         tailscale.com/ipn                                            from tailscale.com/client/local
@@ -128,8 +129,8 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         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/tsweb                                          from tailscale.com/cmd/derper
-        tailscale.com/tsweb/promvarz                                 from tailscale.com/tsweb
+        tailscale.com/tsweb                                          from tailscale.com/cmd/derper+
+        tailscale.com/tsweb/promvarz                                 from tailscale.com/cmd/derper
         tailscale.com/tsweb/varz                                     from tailscale.com/tsweb+
         tailscale.com/types/dnstype                                  from tailscale.com/tailcfg+
         tailscale.com/types/empty                                    from tailscale.com/ipn
@@ -309,7 +310,7 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         html                                                         from net/http/pprof+
         html/template                                                from tailscale.com/cmd/derper
         internal/abi                                                 from crypto/x509/internal/macos+
-        internal/asan                                                from syscall+
+        internal/asan                                                from internal/runtime/maps+
         internal/bisect                                              from internal/godebug
         internal/bytealg                                             from bytes+
         internal/byteorder                                           from crypto/cipher+
@@ -319,12 +320,12 @@ tailscale.com/cmd/derper dependencies: (generated by github.com/tailscale/depawa
         internal/filepathlite                                        from os+
         internal/fmtsort                                             from fmt+
         internal/goarch                                              from crypto/internal/fips140deps/cpu+
-        internal/godebug                                             from crypto/tls+
+        internal/godebug                                             from crypto/internal/fips140deps/godebug+
         internal/godebugs                                            from internal/godebug+
-        internal/goexperiment                                        from runtime+
+        internal/goexperiment                                        from hash/maphash+
         internal/goos                                                from crypto/x509+
         internal/itoa                                                from internal/poll+
-        internal/msan                                                from syscall+
+        internal/msan                                                from internal/runtime/maps+
         internal/nettrace                                            from net+
         internal/oserror                                             from io/fs+
         internal/poll                                                from net+

cmd/derper/derper.go

@@ -49,6 +49,9 @@ import (
 	"tailscale.com/types/key"
 	"tailscale.com/types/logger"
 	"tailscale.com/version"
+
+	// Support for prometheus varz in tsweb
+	_ "tailscale.com/tsweb/promvarz"
 )
 
 var (

cmd/derpprobe/derpprobe.go

@@ -15,6 +15,9 @@ import (
 	"tailscale.com/prober"
 	"tailscale.com/tsweb"
 	"tailscale.com/version"
+
+	// Support for prometheus varz in tsweb
+	_ "tailscale.com/tsweb/promvarz"
 )
 
 var (

cmd/k8s-operator/depaware.txt

@@ -1151,7 +1151,7 @@ tailscale.com/cmd/k8s-operator dependencies: (generated by github.com/tailscale/
         html                                                         from html/template+
         html/template                                                from github.com/gorilla/csrf
         internal/abi                                                 from crypto/x509/internal/macos+
-        internal/asan                                                from syscall+
+        internal/asan                                                from internal/runtime/maps+
         internal/bisect                                              from internal/godebug
         internal/bytealg                                             from bytes+
         internal/byteorder                                           from crypto/cipher+
@@ -1163,11 +1163,11 @@ tailscale.com/cmd/k8s-operator dependencies: (generated by github.com/tailscale/
         internal/goarch                                              from crypto/internal/fips140deps/cpu+
         internal/godebug                                             from archive/tar+
         internal/godebugs                                            from internal/godebug+
-        internal/goexperiment                                        from runtime+
+        internal/goexperiment                                        from hash/maphash+
         internal/goos                                                from crypto/x509+
         internal/itoa                                                from internal/poll+
         internal/lazyregexp                                          from go/doc
-        internal/msan                                                from syscall+
+        internal/msan                                                from internal/runtime/maps+
         internal/nettrace                                            from net+
         internal/oserror                                             from io/fs+
         internal/poll                                                from net+

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

@@ -75,7 +75,7 @@ rules:
   verbs: ["get", "list", "watch", "create", "update", "deletecollection"]
 - apiGroups: ["rbac.authorization.k8s.io"]
   resources: ["roles", "rolebindings"]
-  verbs: ["get", "create", "patch", "update", "list", "watch"]
+  verbs: ["get", "create", "patch", "update", "list", "watch", "deletecollection"]
 - apiGroups: ["monitoring.coreos.com"]
   resources: ["servicemonitors"]
   verbs: ["get", "list", "update", "create", "delete"]

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

@@ -2215,6 +2215,22 @@ spec:
                         https://tailscale.com/kb/1019/subnets#use-your-subnet-routes-from-other-devices
                         Defaults to false.
                       type: boolean
+                useLetsEncryptStagingEnvironment:
+                  description: |-
+                    Set UseLetsEncryptStagingEnvironment to true to issue TLS
+                    certificates for any HTTPS endpoints exposed to the tailnet from
+                    LetsEncrypt's staging environment.
+                    https://letsencrypt.org/docs/staging-environment/
+                    This setting only affects Tailscale Ingress resources.
+                    By default Ingress TLS certificates are issued from LetsEncrypt's
+                    production environment.
+                    Changing this setting true -> false, will result in any
+                    existing certs being re-issued from the production environment.
+                    Changing this setting false (default) -> true, when certs have already
+                    been provisioned from production environment will NOT result in certs
+                    being re-issued from the staging environment before they need to be
+                    renewed.
+                  type: boolean
             status:
               description: |-
                 Status of the ProxyClass. This is set and managed automatically.

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

@@ -103,7 +103,7 @@ spec:
                     pattern: ^tag:[a-zA-Z][a-zA-Z0-9-]*$
                 type:
                   description: |-
-                    Type of the ProxyGroup proxies. Supported types are egress and ingress.
+                    Type of the ProxyGroup proxies. Currently the only supported type is egress.
                     Type is immutable once a ProxyGroup is created.
                   type: string
                   enum:

cmd/k8s-operator/deploy/manifests/operator.yaml

@@ -2685,6 +2685,22 @@ spec:
                                             Defaults to false.
                                         type: boolean
                                 type: object
+                            useLetsEncryptStagingEnvironment:
+                                description: |-
+                                    Set UseLetsEncryptStagingEnvironment to true to issue TLS
+                                    certificates for any HTTPS endpoints exposed to the tailnet from
+                                    LetsEncrypt's staging environment.
+                                    https://letsencrypt.org/docs/staging-environment/
+                                    This setting only affects Tailscale Ingress resources.
+                                    By default Ingress TLS certificates are issued from LetsEncrypt's
+                                    production environment.
+                                    Changing this setting true -> false, will result in any
+                                    existing certs being re-issued from the production environment.
+                                    Changing this setting false (default) -> true, when certs have already
+                                    been provisioned from production environment will NOT result in certs
+                                    being re-issued from the staging environment before they need to be
+                                    renewed.
+                                type: boolean
                         type: object
                     status:
                         description: |-
@@ -2860,7 +2876,7 @@ spec:
                                 type: array
                             type:
                                 description: |-
-                                    Type of the ProxyGroup proxies. Supported types are egress and ingress.
+                                    Type of the ProxyGroup proxies. Currently the only supported type is egress.
                                     Type is immutable once a ProxyGroup is created.
                                 enum:
                                     - egress
@@ -4898,6 +4914,7 @@ rules:
         - update
         - list
         - watch
+        - deletecollection
     - apiGroups:
         - monitoring.coreos.com
       resources:

cmd/k8s-operator/dnsrecords_test.go

@@ -22,6 +22,7 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client/fake"
 	operatorutils "tailscale.com/k8s-operator"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
+	"tailscale.com/kube/kubetypes"
 	"tailscale.com/tstest"
 	"tailscale.com/types/ptr"
 )
@@ -163,10 +164,10 @@ func headlessSvcForParent(o client.Object, typ string) *corev1.Service {
 			Name:      o.GetName(),
 			Namespace: "tailscale",
 			Labels: map[string]string{
-				LabelManaged:         "true",
-				LabelParentName:      o.GetName(),
-				LabelParentNamespace: o.GetNamespace(),
-				LabelParentType:      typ,
+				kubetypes.LabelManaged: "true",
+				LabelParentName:        o.GetName(),
+				LabelParentNamespace:   o.GetNamespace(),
+				LabelParentType:        typ,
 			},
 		},
 		Spec: corev1.ServiceSpec{

cmd/k8s-operator/egress-pod-readiness.go

@@ -112,9 +112,9 @@ func (er *egressPodsReconciler) Reconcile(ctx context.Context, req reconcile.Req
 	}
 	// Get all ClusterIP Services for all egress targets exposed to cluster via this ProxyGroup.
 	lbls := map[string]string{
-		LabelManaged:    "true",
-		labelProxyGroup: proxyGroupName,
-		labelSvcType:    typeEgress,
+		kubetypes.LabelManaged: "true",
+		labelProxyGroup:        proxyGroupName,
+		labelSvcType:           typeEgress,
 	}
 	svcs := &corev1.ServiceList{}
 	if err := er.List(ctx, svcs, client.InNamespace(er.tsNamespace), client.MatchingLabels(lbls)); err != nil {

cmd/k8s-operator/egress-pod-readiness_test.go

@@ -450,9 +450,9 @@ func newSvc(name string, port int32) (*corev1.Service, string) {
 			Namespace: "operator-ns",
 			Name:      name,
 			Labels: map[string]string{
-				LabelManaged:    "true",
-				labelProxyGroup: "dev",
-				labelSvcType:    typeEgress,
+				kubetypes.LabelManaged: "true",
+				labelProxyGroup:        "dev",
+				labelSvcType:           typeEgress,
 			},
 		},
 		Spec: corev1.ServiceSpec{},

cmd/k8s-operator/egress-services.go

@@ -680,12 +680,12 @@ func egressSvcsConfigs(ctx context.Context, cl client.Client, proxyGroupName, ts
 // should probably validate and truncate (?) the names is they are too long.
 func egressSvcChildResourceLabels(svc *corev1.Service) map[string]string {
 	return map[string]string{
-		LabelManaged:         "true",
-		LabelParentType:      "svc",
-		LabelParentName:      svc.Name,
-		LabelParentNamespace: svc.Namespace,
-		labelProxyGroup:      svc.Annotations[AnnotationProxyGroup],
-		labelSvcType:         typeEgress,
+		kubetypes.LabelManaged: "true",
+		LabelParentType:        "svc",
+		LabelParentName:        svc.Name,
+		LabelParentNamespace:   svc.Namespace,
+		labelProxyGroup:        svc.Annotations[AnnotationProxyGroup],
+		labelSvcType:           typeEgress,
 	}
 }
 

cmd/k8s-operator/ingress-for-pg.go

@@ -22,6 +22,7 @@ import (
 	"go.uber.org/zap"
 	corev1 "k8s.io/api/core/v1"
 	networkingv1 "k8s.io/api/networking/v1"
+	rbacv1 "k8s.io/api/rbac/v1"
 	apiequality "k8s.io/apimachinery/pkg/api/equality"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -48,10 +49,11 @@ const (
 	// FinalizerNamePG is the finalizer used by the IngressPGReconciler
 	FinalizerNamePG = "tailscale.com/ingress-pg-finalizer"
 
-	indexIngressProxyGroup = ".metadata.annotations.ingress-proxy-group"
 	// annotationHTTPEndpoint can be used to configure the Ingress to expose an HTTP endpoint to tailnet (as
 	// well as the default HTTPS endpoint).
 	annotationHTTPEndpoint = "tailscale.com/http-endpoint"
+
+	labelDomain = "tailscale.com/domain"
 )
 
 var gaugePGIngressResources = clientmetric.NewGauge(kubetypes.MetricIngressPGResourceCount)
@@ -154,13 +156,13 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 	pg := &tsapi.ProxyGroup{}
 	if err := r.Get(ctx, client.ObjectKey{Name: pgName}, pg); err != nil {
 		if apierrors.IsNotFound(err) {
-			logger.Infof("ProxyGroup %q does not exist", pgName)
+			logger.Infof("ProxyGroup does not exist")
 			return false, nil
 		}
 		return false, fmt.Errorf("getting ProxyGroup %q: %w", pgName, err)
 	}
 	if !tsoperator.ProxyGroupIsReady(pg) {
-		logger.Infof("ProxyGroup %q is not (yet) ready", pgName)
+		logger.Infof("ProxyGroup is not (yet) ready")
 		return false, nil
 	}
 
@@ -175,8 +177,6 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 		r.recorder.Event(ing, corev1.EventTypeWarning, "HTTPSNotEnabled", "HTTPS is not enabled on the tailnet; ingress may not work")
 	}
 
-	logger = logger.With("proxy-group", pg.Name)
-
 	if !slices.Contains(ing.Finalizers, FinalizerNamePG) {
 		// This log line is printed exactly once during initial provisioning,
 		// because once the finalizer is in place this block gets skipped. So,
@@ -229,12 +229,11 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 			return false, fmt.Errorf("error getting VIPService %q: %w", hostname, err)
 		}
 	}
-	// Generate the VIPService comment for new or existing VIPService. This
-	// checks and ensures that VIPService's owner references are updated for
-	// this Ingress and errors if that is not possible (i.e. because it
-	// appears that the VIPService has been created by a non-operator
-	// actor).
-	svcComment, err := r.ownerRefsComment(existingVIPSvc)
+	// Generate the VIPService owner annotation for new or existing VIPService.
+	// This checks and ensures that VIPService's owner references are updated
+	// for this Ingress and errors if that is not possible (i.e. because it
+	// appears that the VIPService has been created by a non-operator actor).
+	updatedAnnotations, err := r.ownerAnnotations(existingVIPSvc)
 	if err != nil {
 		const instr = "To proceed, you can either manually delete the existing VIPService or choose a different MagicDNS name at `.spec.tls.hosts[0] in the Ingress definition"
 		msg := fmt.Sprintf("error ensuring ownership of VIPService %s: %v. %s", hostname, err, instr)
@@ -242,8 +241,12 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 		r.recorder.Event(ing, corev1.EventTypeWarning, "InvalidVIPService", msg)
 		return false, nil
 	}
+	// 3. Ensure that TLS Secret and RBAC exists
+	if err := r.ensureCertResources(ctx, pgName, dnsName, ing); err != nil {
+		return false, fmt.Errorf("error ensuring cert resources: %w", err)
+	}
 
-	// 3. Ensure that the serve config for the ProxyGroup contains the VIPService.
+	// 4. Ensure that the serve config for the ProxyGroup contains the VIPService.
 	cm, cfg, err := r.proxyGroupServeConfig(ctx, pgName)
 	if err != nil {
 		return false, fmt.Errorf("error getting Ingress serve config: %w", err)
@@ -310,11 +313,13 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 		vipPorts = append(vipPorts, "80")
 	}
 
+	const managedVIPServiceComment = "This VIPService is managed by the Tailscale Kubernetes Operator, do not modify"
 	vipSvc := &tailscale.VIPService{
-		Name:    serviceName,
-		Tags:    tags,
-		Ports:   vipPorts,
-		Comment: svcComment,
+		Name:        serviceName,
+		Tags:        tags,
+		Ports:       vipPorts,
+		Comment:     managedVIPServiceComment,
+		Annotations: updatedAnnotations,
 	}
 	if existingVIPSvc != nil {
 		vipSvc.Addrs = existingVIPSvc.Addrs
@@ -325,8 +330,8 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 	if existingVIPSvc == nil ||
 		!reflect.DeepEqual(vipSvc.Tags, existingVIPSvc.Tags) ||
 		!reflect.DeepEqual(vipSvc.Ports, existingVIPSvc.Ports) ||
-		!strings.EqualFold(vipSvc.Comment, existingVIPSvc.Comment) {
-		logger.Infof("Ensuring VIPService %q exists and is up to date", hostname)
+		!ownersAreSetAndEqual(vipSvc, existingVIPSvc) {
+		logger.Infof("Ensuring VIPService exists and is up to date")
 		if err := r.tsClient.CreateOrUpdateVIPService(ctx, vipSvc); err != nil {
 			return false, fmt.Errorf("error creating VIPService: %w", err)
 		}
@@ -334,35 +339,67 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 
 	// 5. Update tailscaled's AdvertiseServices config, which should add the VIPService
 	// IPs to the ProxyGroup Pods' AllowedIPs in the next netmap update if approved.
-	if err = r.maybeUpdateAdvertiseServicesConfig(ctx, pg.Name, serviceName, true, logger); err != nil {
+	mode := serviceAdvertisementHTTPS
+	if isHTTPEndpointEnabled(ing) {
+		mode = serviceAdvertisementHTTPAndHTTPS
+	}
+	if err = r.maybeUpdateAdvertiseServicesConfig(ctx, pg.Name, serviceName, mode, logger); err != nil {
 		return false, fmt.Errorf("failed to update tailscaled config: %w", err)
 	}
 
-	// TODO(irbekrm): check that the replicas are ready to route traffic for the VIPService before updating Ingress
-	// status.
-	// 6. Update Ingress status
+	// 6. Update Ingress status if ProxyGroup Pods are ready.
+	count, err := r.numberPodsAdvertising(ctx, pg.Name, serviceName)
+	if err != nil {
+		return false, fmt.Errorf("failed to check if any Pods are configured: %w", err)
+	}
+
 	oldStatus := ing.Status.DeepCopy()
-	ports := []networkingv1.IngressPortStatus{
-		{
-			Protocol: "TCP",
-			Port:     443,
-		},
+
+	switch count {
+	case 0:
+		ing.Status.LoadBalancer.Ingress = nil
+	default:
+		var ports []networkingv1.IngressPortStatus
+		hasCerts, err := r.hasCerts(ctx, serviceName)
+		if err != nil {
+			return false, fmt.Errorf("error checking TLS credentials provisioned for Ingress: %w", err)
+		}
+		// If TLS certs have not been issued (yet), do not set port 443.
+		if hasCerts {
+			ports = append(ports, networkingv1.IngressPortStatus{
+				Protocol: "TCP",
+				Port:     443,
+			})
+		}
+		if isHTTPEndpointEnabled(ing) {
+			ports = append(ports, networkingv1.IngressPortStatus{
+				Protocol: "TCP",
+				Port:     80,
+			})
+		}
+		// Set Ingress status hostname only if either port 443 or 80 is advertised.
+		var hostname string
+		if len(ports) != 0 {
+			hostname = dnsName
+		}
+		ing.Status.LoadBalancer.Ingress = []networkingv1.IngressLoadBalancerIngress{
+			{
+				Hostname: hostname,
+				Ports:    ports,
+			},
+		}
 	}
-	if isHTTPEndpointEnabled(ing) {
-		ports = append(ports, networkingv1.IngressPortStatus{
-			Protocol: "TCP",
-			Port:     80,
-		})
-	}
-	ing.Status.LoadBalancer.Ingress = []networkingv1.IngressLoadBalancerIngress{
-		{
-			Hostname: dnsName,
-			Ports:    ports,
-		},
-	}
-	if apiequality.Semantic.DeepEqual(oldStatus, ing.Status) {
+	if apiequality.Semantic.DeepEqual(oldStatus, &ing.Status) {
 		return svcsChanged, nil
 	}
+
+	const prefix = "Updating Ingress status"
+	if count == 0 {
+		logger.Infof("%s. No Pods are advertising VIPService yet", prefix)
+	} else {
+		logger.Infof("%s. %d Pod(s) advertising VIPService", prefix, count)
+	}
+
 	if err := r.Status().Update(ctx, ing); err != nil {
 		return false, fmt.Errorf("failed to update Ingress status: %w", err)
 	}
@@ -402,24 +439,24 @@ func (r *HAIngressReconciler) maybeCleanupProxyGroup(ctx context.Context, proxyG
 			logger.Infof("VIPService %q is not owned by any Ingress, cleaning up", vipServiceName)
 
 			// Delete the VIPService from control if necessary.
-			svc, _ := r.tsClient.GetVIPService(ctx, vipServiceName)
-			if svc != nil && isVIPServiceForAnyIngress(svc) {
-				logger.Infof("cleaning up orphaned VIPService %q", vipServiceName)
-				svcsChanged, err = r.cleanupVIPService(ctx, vipServiceName, logger)
-				if err != nil {
-					errResp := &tailscale.ErrResponse{}
-					if !errors.As(err, &errResp) || errResp.Status != http.StatusNotFound {
-						return false, fmt.Errorf("deleting VIPService %q: %w", vipServiceName, err)
-					}
-				}
+			svcsChanged, err = r.cleanupVIPService(ctx, vipServiceName, logger)
+			if err != nil {
+				return false, fmt.Errorf("deleting VIPService %q: %w", vipServiceName, err)
 			}
 
 			// Make sure the VIPService is not advertised in tailscaled or serve config.
-			if err = r.maybeUpdateAdvertiseServicesConfig(ctx, proxyGroupName, vipServiceName, false, logger); err != nil {
+			if err = r.maybeUpdateAdvertiseServicesConfig(ctx, proxyGroupName, vipServiceName, serviceAdvertisementOff, logger); err != nil {
 				return false, fmt.Errorf("failed to update tailscaled config services: %w", err)
 			}
-			delete(cfg.Services, vipServiceName)
-			serveConfigChanged = true
+			_, ok := cfg.Services[vipServiceName]
+			if ok {
+				logger.Infof("Removing VIPService %q from serve config", vipServiceName)
+				delete(cfg.Services, vipServiceName)
+				serveConfigChanged = true
+			}
+			if err := r.cleanupCertResources(ctx, proxyGroupName, vipServiceName); err != nil {
+				return false, fmt.Errorf("failed to clean up cert resources: %w", err)
+			}
 		}
 	}
 
@@ -480,16 +517,22 @@ func (r *HAIngressReconciler) maybeCleanup(ctx context.Context, hostname string,
 	if err != nil {
 		return false, fmt.Errorf("error deleting VIPService: %w", err)
 	}
+
+	// 3. Clean up any cluster resources
+	if err := r.cleanupCertResources(ctx, pg, serviceName); err != nil {
+		return false, fmt.Errorf("failed to clean up cert resources: %w", err)
+	}
+
 	if cfg == nil || cfg.Services == nil { // user probably deleted the ProxyGroup
 		return svcChanged, nil
 	}
 
-	// 3. Unadvertise the VIPService in tailscaled config.
-	if err = r.maybeUpdateAdvertiseServicesConfig(ctx, pg, serviceName, false, logger); err != nil {
+	// 4. Unadvertise the VIPService in tailscaled config.
+	if err = r.maybeUpdateAdvertiseServicesConfig(ctx, pg, serviceName, serviceAdvertisementOff, logger); err != nil {
 		return false, fmt.Errorf("failed to update tailscaled config services: %w", err)
 	}
 
-	// 4. Remove the VIPService from the serve config for the ProxyGroup.
+	// 5. Remove the VIPService from the serve config for the ProxyGroup.
 	logger.Infof("Removing VIPService %q from serve config for ProxyGroup %q", hostname, pg)
 	delete(cfg.Services, serviceName)
 	cfgBytes, err := json.Marshal(cfg)
@@ -570,13 +613,6 @@ func (r *HAIngressReconciler) shouldExpose(ing *networkingv1.Ingress) bool {
 	return isTSIngress && pgAnnot != ""
 }
 
-func isVIPServiceForAnyIngress(svc *tailscale.VIPService) bool {
-	if svc == nil {
-		return false
-	}
-	return strings.HasPrefix(svc.Comment, "tailscale.com/k8s-operator:owned-by:")
-}
-
 // validateIngress validates that the Ingress is properly configured.
 // Currently validates:
 // - Any tags provided via tailscale.com/tags annotation are valid Tailscale ACL tags
@@ -650,34 +686,34 @@ func (r *HAIngressReconciler) cleanupVIPService(ctx context.Context, name tailcf
 	if svc == nil {
 		return false, nil
 	}
-	c, err := parseComment(svc)
+	o, err := parseOwnerAnnotation(svc)
 	if err != nil {
-		return false, fmt.Errorf("error parsing VIPService comment")
+		return false, fmt.Errorf("error parsing VIPService owner annotation")
 	}
-	if c == nil || len(c.OwnerRefs) == 0 {
+	if o == nil || len(o.OwnerRefs) == 0 {
 		return false, nil
 	}
 	// Comparing with the operatorID only means that we will not be able to
 	// clean up VIPServices in cases where the operator was deleted from the
 	// cluster before deleting the Ingress. Perhaps the comparison could be
 	// 'if or.OperatorID === r.operatorID || or.ingressUID == r.ingressUID'.
-	ix := slices.IndexFunc(c.OwnerRefs, func(or OwnerRef) bool {
+	ix := slices.IndexFunc(o.OwnerRefs, func(or OwnerRef) bool {
 		return or.OperatorID == r.operatorID
 	})
 	if ix == -1 {
 		return false, nil
 	}
-	if len(c.OwnerRefs) == 1 {
+	if len(o.OwnerRefs) == 1 {
 		logger.Infof("Deleting VIPService %q", name)
 		return false, r.tsClient.DeleteVIPService(ctx, name)
 	}
-	c.OwnerRefs = slices.Delete(c.OwnerRefs, ix, ix+1)
+	o.OwnerRefs = slices.Delete(o.OwnerRefs, ix, ix+1)
 	logger.Infof("Deleting VIPService %q", name)
-	json, err := json.Marshal(c)
+	json, err := json.Marshal(o)
 	if err != nil {
 		return false, fmt.Errorf("error marshalling updated VIPService owner reference: %w", err)
 	}
-	svc.Comment = string(json)
+	svc.Annotations[ownerAnnotation] = string(json)
 	return true, r.tsClient.CreateOrUpdateVIPService(ctx, svc)
 }
 
@@ -689,8 +725,16 @@ func isHTTPEndpointEnabled(ing *networkingv1.Ingress) bool {
 	return ing.Annotations[annotationHTTPEndpoint] == "enabled"
 }
 
-func (a *HAIngressReconciler) maybeUpdateAdvertiseServicesConfig(ctx context.Context, pgName string, serviceName tailcfg.ServiceName, shouldBeAdvertised bool, logger *zap.SugaredLogger) (err error) {
-	logger.Debugf("Updating ProxyGroup tailscaled configs to advertise service %q: %v", serviceName, shouldBeAdvertised)
+// serviceAdvertisementMode describes the desired state of a VIPService.
+type serviceAdvertisementMode int
+
+const (
+	serviceAdvertisementOff          serviceAdvertisementMode = iota // Should not be advertised
+	serviceAdvertisementHTTPS                                        // Port 443 should be advertised
+	serviceAdvertisementHTTPAndHTTPS                                 // Both ports 80 and 443 should be advertised
+)
+
+func (a *HAIngressReconciler) maybeUpdateAdvertiseServicesConfig(ctx context.Context, pgName string, serviceName tailcfg.ServiceName, mode serviceAdvertisementMode, logger *zap.SugaredLogger) (err error) {
 
 	// Get all config Secrets for this ProxyGroup.
 	secrets := &corev1.SecretList{}
@@ -698,6 +742,21 @@ func (a *HAIngressReconciler) maybeUpdateAdvertiseServicesConfig(ctx context.Con
 		return fmt.Errorf("failed to list config Secrets: %w", err)
 	}
 
+	// Verify that TLS cert for the VIPService has been successfully issued
+	// before attempting to advertise the service.
+	// This is so that in multi-cluster setups where some Ingresses succeed
+	// to issue certs and some do not (rate limits), clients are not pinned
+	// to a backend that is not able to serve HTTPS.
+	// The only exception is Ingresses with an HTTP endpoint enabled - if an
+	// Ingress has an HTTP endpoint enabled, it will be advertised even if the
+	// TLS cert is not yet provisioned.
+	hasCert, err := a.hasCerts(ctx, serviceName)
+	if err != nil {
+		return fmt.Errorf("error checking TLS credentials provisioned for service %q: %w", serviceName, err)
+	}
+	shouldBeAdvertised := (mode == serviceAdvertisementHTTPAndHTTPS) ||
+		(mode == serviceAdvertisementHTTPS && hasCert) // if we only expose port 443 and don't have certs (yet), do not advertise
+
 	for _, secret := range secrets.Items {
 		var updated bool
 		for fileName, confB := range secret.Data {
@@ -740,6 +799,39 @@ func (a *HAIngressReconciler) maybeUpdateAdvertiseServicesConfig(ctx context.Con
 	return nil
 }
 
+func (a *HAIngressReconciler) numberPodsAdvertising(ctx context.Context, pgName string, serviceName tailcfg.ServiceName) (int, error) {
+	// Get all state Secrets for this ProxyGroup.
+	secrets := &corev1.SecretList{}
+	if err := a.List(ctx, secrets, client.InNamespace(a.tsNamespace), client.MatchingLabels(pgSecretLabels(pgName, "state"))); err != nil {
+		return 0, fmt.Errorf("failed to list ProxyGroup %q state Secrets: %w", pgName, err)
+	}
+
+	var count int
+	for _, secret := range secrets.Items {
+		prefs, ok, err := getDevicePrefs(&secret)
+		if err != nil {
+			return 0, fmt.Errorf("error getting node metadata: %w", err)
+		}
+		if !ok {
+			continue
+		}
+		if slices.Contains(prefs.AdvertiseServices, serviceName.String()) {
+			count++
+		}
+	}
+
+	return count, nil
+}
+
+const ownerAnnotation = "tailscale.com/owner-references"
+
+// ownerAnnotationValue is the content of the VIPService.Annotation[ownerAnnotation] field.
+type ownerAnnotationValue struct {
+	// OwnerRefs is a list of owner references that identify all operator
+	// instances that manage this VIPService.
+	OwnerRefs []OwnerRef `json:"ownerRefs,omitempty"`
+}
+
 // OwnerRef is an owner reference that uniquely identifies a Tailscale
 // Kubernetes operator instance.
 type OwnerRef struct {
@@ -747,60 +839,110 @@ type OwnerRef struct {
 	OperatorID string `json:"operatorID,omitempty"`
 }
 
-// comment is the content of the VIPService.Comment field.
-type comment struct {
-	// OwnerRefs is a list of owner references that identify all operator
-	// instances that manage this VIPService.
-	OwnerRefs []OwnerRef `json:"ownerRefs,omitempty"`
-}
-
-// ownerRefsComment return VIPService Comment that includes owner reference for this
-// operator instance for the provided VIPService. If the VIPService is nil, a
-// new comment with owner ref is returned. If the VIPService is not nil, the
-// existing comment is returned with the owner reference added, if not already
-// present. If the VIPService is not nil, but does not contain a comment we
-// return an error as this likely means that the VIPService was created by
-// somthing other than a Tailscale Kubernetes operator.
-func (r *HAIngressReconciler) ownerRefsComment(svc *tailscale.VIPService) (string, error) {
+// ownerAnnotations returns the updated annotations required to ensure this
+// instance of the operator is included as an owner. If the VIPService is not
+// nil, but does not contain an owner we return an error as this likely means
+// that the VIPService was created by somthing other than a Tailscale
+// Kubernetes operator.
+func (r *HAIngressReconciler) ownerAnnotations(svc *tailscale.VIPService) (map[string]string, error) {
 	ref := OwnerRef{
 		OperatorID: r.operatorID,
 	}
 	if svc == nil {
-		c := &comment{OwnerRefs: []OwnerRef{ref}}
+		c := ownerAnnotationValue{OwnerRefs: []OwnerRef{ref}}
 		json, err := json.Marshal(c)
 		if err != nil {
-			return "", fmt.Errorf("[unexpected] unable to marshal VIPService comment contents: %w, please report this", err)
+			return nil, fmt.Errorf("[unexpected] unable to marshal VIPService owner annotation contents: %w, please report this", err)
 		}
-		return string(json), nil
+		return map[string]string{
+			ownerAnnotation: string(json),
+		}, nil
 	}
-	c, err := parseComment(svc)
+	o, err := parseOwnerAnnotation(svc)
 	if err != nil {
-		return "", fmt.Errorf("error parsing existing VIPService comment: %w", err)
+		return nil, err
 	}
-	if c == nil || len(c.OwnerRefs) == 0 {
-		return "", fmt.Errorf("VIPService %s exists, but does not contain Comment field with owner references- not proceeding as this is likely a resource created by something other than a Tailscale Kubernetes Operator", svc.Name)
+	if o == nil || len(o.OwnerRefs) == 0 {
+		return nil, fmt.Errorf("VIPService %s exists, but does not contain owner annotation with owner references; not proceeding as this is likely a resource created by something other than the Tailscale Kubernetes operator", svc.Name)
 	}
-	if slices.Contains(c.OwnerRefs, ref) { // up to date
-		return svc.Comment, nil
+	if slices.Contains(o.OwnerRefs, ref) { // up to date
+		return svc.Annotations, nil
 	}
-	c.OwnerRefs = append(c.OwnerRefs, ref)
-	json, err := json.Marshal(c)
+	o.OwnerRefs = append(o.OwnerRefs, ref)
+	json, err := json.Marshal(o)
 	if err != nil {
-		return "", fmt.Errorf("error marshalling updated owner references: %w", err)
+		return nil, fmt.Errorf("error marshalling updated owner references: %w", err)
 	}
-	return string(json), nil
+
+	newAnnots := make(map[string]string, len(svc.Annotations)+1)
+	for k, v := range svc.Annotations {
+		newAnnots[k] = v
+	}
+	newAnnots[ownerAnnotation] = string(json)
+	return newAnnots, nil
 }
 
-// parseComment returns VIPService comment or nil if none found or not matching the expected format.
-func parseComment(vipSvc *tailscale.VIPService) (*comment, error) {
-	if vipSvc.Comment == "" {
+// parseOwnerAnnotation returns nil if no valid owner found.
+func parseOwnerAnnotation(vipSvc *tailscale.VIPService) (*ownerAnnotationValue, error) {
+	if vipSvc.Annotations == nil || vipSvc.Annotations[ownerAnnotation] == "" {
 		return nil, nil
 	}
-	c := &comment{}
-	if err := json.Unmarshal([]byte(vipSvc.Comment), c); err != nil {
-		return nil, fmt.Errorf("error parsing VIPService Comment field %q: %w", vipSvc.Comment, err)
+	o := &ownerAnnotationValue{}
+	if err := json.Unmarshal([]byte(vipSvc.Annotations[ownerAnnotation]), o); err != nil {
+		return nil, fmt.Errorf("error parsing VIPService %s annotation %q: %w", ownerAnnotation, vipSvc.Annotations[ownerAnnotation], err)
 	}
-	return c, nil
+	return o, nil
+}
+
+func ownersAreSetAndEqual(a, b *tailscale.VIPService) bool {
+	return a != nil && b != nil &&
+		a.Annotations != nil && b.Annotations != nil &&
+		a.Annotations[ownerAnnotation] != "" &&
+		b.Annotations[ownerAnnotation] != "" &&
+		strings.EqualFold(a.Annotations[ownerAnnotation], b.Annotations[ownerAnnotation])
+}
+
+// ensureCertResources ensures that the TLS Secret for an HA Ingress and RBAC
+// resources that allow proxies to manage the Secret are created.
+// Note that Tailscale VIPService name validation matches Kubernetes
+// resource name validation, so we can be certain that the VIPService name
+// (domain) is a valid Kubernetes resource name.
+// https://github.com/tailscale/tailscale/blob/8b1e7f646ee4730ad06c9b70c13e7861b964949b/util/dnsname/dnsname.go#L99
+// https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#dns-subdomain-names
+func (r *HAIngressReconciler) ensureCertResources(ctx context.Context, pgName, domain string, ing *networkingv1.Ingress) error {
+	secret := certSecret(pgName, r.tsNamespace, domain, ing)
+	if _, err := createOrUpdate(ctx, r.Client, r.tsNamespace, secret, nil); err != nil {
+		return fmt.Errorf("failed to create or update Secret %s: %w", secret.Name, err)
+	}
+	role := certSecretRole(pgName, r.tsNamespace, domain)
+	if _, err := createOrUpdate(ctx, r.Client, r.tsNamespace, role, nil); err != nil {
+		return fmt.Errorf("failed to create or update Role %s: %w", role.Name, err)
+	}
+	rb := certSecretRoleBinding(pgName, r.tsNamespace, domain)
+	if _, err := createOrUpdate(ctx, r.Client, r.tsNamespace, rb, nil); err != nil {
+		return fmt.Errorf("failed to create or update RoleBinding %s: %w", rb.Name, err)
+	}
+	return nil
+}
+
+// cleanupCertResources ensures that the TLS Secret and associated RBAC
+// resources that allow proxies to read/write to the Secret are deleted.
+func (r *HAIngressReconciler) cleanupCertResources(ctx context.Context, pgName string, name tailcfg.ServiceName) error {
+	domainName, err := r.dnsNameForService(ctx, tailcfg.ServiceName(name))
+	if err != nil {
+		return fmt.Errorf("error getting DNS name for VIPService %s: %w", name, err)
+	}
+	labels := certResourceLabels(pgName, domainName)
+	if err := r.DeleteAllOf(ctx, &rbacv1.RoleBinding{}, client.InNamespace(r.tsNamespace), client.MatchingLabels(labels)); err != nil {
+		return fmt.Errorf("error deleting RoleBinding for domain name %s: %w", domainName, err)
+	}
+	if err := r.DeleteAllOf(ctx, &rbacv1.Role{}, client.InNamespace(r.tsNamespace), client.MatchingLabels(labels)); err != nil {
+		return fmt.Errorf("error deleting Role for domain name %s: %w", domainName, err)
+	}
+	if err := r.DeleteAllOf(ctx, &corev1.Secret{}, client.InNamespace(r.tsNamespace), client.MatchingLabels(labels)); err != nil {
+		return fmt.Errorf("error deleting Secret for domain name %s: %w", domainName, err)
+	}
+	return nil
 }
 
 // requeueInterval returns a time duration between 5 and 10 minutes, which is
@@ -811,3 +953,123 @@ func parseComment(vipSvc *tailscale.VIPService) (*comment, error) {
 func requeueInterval() time.Duration {
 	return time.Duration(rand.N(5)+5) * time.Minute
 }
+
+// certSecretRole creates a Role that will allow proxies to manage the TLS
+// Secret for the given domain. Domain must be a valid Kubernetes resource name.
+func certSecretRole(pgName, namespace, domain string) *rbacv1.Role {
+	return &rbacv1.Role{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      domain,
+			Namespace: namespace,
+			Labels:    certResourceLabels(pgName, domain),
+		},
+		Rules: []rbacv1.PolicyRule{
+			{
+				APIGroups:     []string{""},
+				Resources:     []string{"secrets"},
+				ResourceNames: []string{domain},
+				Verbs: []string{
+					"get",
+					"list",
+					"patch",
+					"update",
+				},
+			},
+		},
+	}
+}
+
+// certSecretRoleBinding creates a RoleBinding for Role that will allow proxies
+// to manage the TLS Secret for the given domain. Domain must be a valid
+// Kubernetes resource name.
+func certSecretRoleBinding(pgName, namespace, domain string) *rbacv1.RoleBinding {
+	return &rbacv1.RoleBinding{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      domain,
+			Namespace: namespace,
+			Labels:    certResourceLabels(pgName, domain),
+		},
+		Subjects: []rbacv1.Subject{
+			{
+				Kind:      "ServiceAccount",
+				Name:      pgName,
+				Namespace: namespace,
+			},
+		},
+		RoleRef: rbacv1.RoleRef{
+			Kind: "Role",
+			Name: domain,
+		},
+	}
+}
+
+// certSecret creates a Secret that will store the TLS certificate and private
+// key for the given domain. Domain must be a valid Kubernetes resource name.
+func certSecret(pgName, namespace, domain string, ing *networkingv1.Ingress) *corev1.Secret {
+	labels := certResourceLabels(pgName, domain)
+	labels[kubetypes.LabelSecretType] = "certs"
+	// Labels that let us identify the Ingress resource lets us reconcile
+	// the Ingress when the TLS Secret is updated (for example, when TLS
+	// certs have been provisioned).
+	labels[LabelParentName] = ing.Name
+	labels[LabelParentNamespace] = ing.Namespace
+	return &corev1.Secret{
+		TypeMeta: metav1.TypeMeta{
+			APIVersion: "v1",
+			Kind:       "Secret",
+		},
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      domain,
+			Namespace: namespace,
+			Labels:    labels,
+		},
+		Data: map[string][]byte{
+			corev1.TLSCertKey:       nil,
+			corev1.TLSPrivateKeyKey: nil,
+		},
+		Type: corev1.SecretTypeTLS,
+	}
+}
+
+func certResourceLabels(pgName, domain string) map[string]string {
+	return map[string]string{
+		kubetypes.LabelManaged: "true",
+		labelProxyGroup:        pgName,
+		labelDomain:            domain,
+	}
+}
+
+// dnsNameForService returns the DNS name for the given VIPService name.
+func (r *HAIngressReconciler) dnsNameForService(ctx context.Context, svc tailcfg.ServiceName) (string, error) {
+	s := svc.WithoutPrefix()
+	tcd, err := r.tailnetCertDomain(ctx)
+	if err != nil {
+		return "", fmt.Errorf("error determining DNS name base: %w", err)
+	}
+	return s + "." + tcd, nil
+}
+
+// hasCerts checks if the TLS Secret for the given service has non-zero cert and key data.
+func (r *HAIngressReconciler) hasCerts(ctx context.Context, svc tailcfg.ServiceName) (bool, error) {
+	domain, err := r.dnsNameForService(ctx, svc)
+	if err != nil {
+		return false, fmt.Errorf("failed to get DNS name for service: %w", err)
+	}
+	secret := &corev1.Secret{}
+	err = r.Get(ctx, client.ObjectKey{
+		Namespace: r.tsNamespace,
+		Name:      domain,
+	}, secret)
+
+	if err != nil {
+		if apierrors.IsNotFound(err) {
+			return false, nil
+		}
+		return false, fmt.Errorf("failed to get TLS Secret: %w", err)
+	}
+
+	cert := secret.Data[corev1.TLSCertKey]
+	key := secret.Data[corev1.TLSPrivateKeyKey]
+
+	return len(cert) > 0 && len(key) > 0, nil
+}

cmd/k8s-operator/ingress-for-pg_test.go

@@ -8,8 +8,10 @@ package main
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"maps"
+	"net/http"
 	"reflect"
 	"testing"
 
@@ -18,6 +20,7 @@ import (
 	"go.uber.org/zap"
 	corev1 "k8s.io/api/core/v1"
 	networkingv1 "k8s.io/api/networking/v1"
+	rbacv1 "k8s.io/api/rbac/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/client-go/tools/record"
@@ -28,6 +31,7 @@ import (
 	"tailscale.com/ipn/ipnstate"
 	tsoperator "tailscale.com/k8s-operator"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
+	"tailscale.com/kube/kubetypes"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/ptr"
 )
@@ -56,7 +60,7 @@ func TestIngressPGReconciler(t *testing.T) {
 				},
 			},
 			TLS: []networkingv1.IngressTLS{
-				{Hosts: []string{"my-svc.tailnetxyz.ts.net"}},
+				{Hosts: []string{"my-svc"}},
 			},
 		},
 	}
@@ -64,10 +68,17 @@ func TestIngressPGReconciler(t *testing.T) {
 
 	// Verify initial reconciliation
 	expectReconciled(t, ingPGR, "default", "test-ingress")
+	populateTLSSecret(context.Background(), fc, "test-pg", "my-svc.ts.net")
+	expectReconciled(t, ingPGR, "default", "test-ingress")
 	verifyServeConfig(t, fc, "svc:my-svc", false)
 	verifyVIPService(t, ft, "svc:my-svc", []string{"443"})
 	verifyTailscaledConfig(t, fc, []string{"svc:my-svc"})
 
+	// Verify that Role and RoleBinding have been created for the first Ingress.
+	// Do not verify the cert Secret as that was already verified implicitly above.
+	expectEqual(t, fc, certSecretRole("test-pg", "operator-ns", "my-svc.ts.net"))
+	expectEqual(t, fc, certSecretRoleBinding("test-pg", "operator-ns", "my-svc.ts.net"))
+
 	mustUpdate(t, fc, "default", "test-ingress", func(ing *networkingv1.Ingress) {
 		ing.Annotations["tailscale.com/tags"] = "tag:custom,tag:test"
 	})
@@ -119,9 +130,16 @@ func TestIngressPGReconciler(t *testing.T) {
 
 	// Verify second Ingress reconciliation
 	expectReconciled(t, ingPGR, "default", "my-other-ingress")
+	populateTLSSecret(context.Background(), fc, "test-pg", "my-other-svc.ts.net")
+	expectReconciled(t, ingPGR, "default", "my-other-ingress")
 	verifyServeConfig(t, fc, "svc:my-other-svc", false)
 	verifyVIPService(t, ft, "svc:my-other-svc", []string{"443"})
 
+	// Verify that Role and RoleBinding have been created for the first Ingress.
+	// Do not verify the cert Secret as that was already verified implicitly above.
+	expectEqual(t, fc, certSecretRole("test-pg", "operator-ns", "my-other-svc.ts.net"))
+	expectEqual(t, fc, certSecretRoleBinding("test-pg", "operator-ns", "my-other-svc.ts.net"))
+
 	// Verify first Ingress is still working
 	verifyServeConfig(t, fc, "svc:my-svc", false)
 	verifyVIPService(t, ft, "svc:my-svc", []string{"443"})
@@ -158,6 +176,9 @@ func TestIngressPGReconciler(t *testing.T) {
 	}
 
 	verifyTailscaledConfig(t, fc, []string{"svc:my-svc"})
+	expectMissing[corev1.Secret](t, fc, "operator-ns", "my-other-svc.ts.net")
+	expectMissing[rbacv1.Role](t, fc, "operator-ns", "my-other-svc.ts.net")
+	expectMissing[rbacv1.RoleBinding](t, fc, "operator-ns", "my-other-svc.ts.net")
 
 	// Delete the first Ingress and verify cleanup
 	if err := fc.Delete(context.Background(), ing); err != nil {
@@ -184,6 +205,70 @@ func TestIngressPGReconciler(t *testing.T) {
 		t.Error("serve config not cleaned up")
 	}
 	verifyTailscaledConfig(t, fc, nil)
+
+	// Add verification that cert resources were cleaned up
+	expectMissing[corev1.Secret](t, fc, "operator-ns", "my-svc.ts.net")
+	expectMissing[rbacv1.Role](t, fc, "operator-ns", "my-svc.ts.net")
+	expectMissing[rbacv1.RoleBinding](t, fc, "operator-ns", "my-svc.ts.net")
+}
+
+func TestIngressPGReconciler_UpdateIngressHostname(t *testing.T) {
+	ingPGR, fc, ft := setupIngressTest(t)
+
+	ing := &networkingv1.Ingress{
+		TypeMeta: metav1.TypeMeta{Kind: "Ingress", APIVersion: "networking.k8s.io/v1"},
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "test-ingress",
+			Namespace: "default",
+			UID:       types.UID("1234-UID"),
+			Annotations: map[string]string{
+				"tailscale.com/proxy-group": "test-pg",
+			},
+		},
+		Spec: networkingv1.IngressSpec{
+			IngressClassName: ptr.To("tailscale"),
+			DefaultBackend: &networkingv1.IngressBackend{
+				Service: &networkingv1.IngressServiceBackend{
+					Name: "test",
+					Port: networkingv1.ServiceBackendPort{
+						Number: 8080,
+					},
+				},
+			},
+			TLS: []networkingv1.IngressTLS{
+				{Hosts: []string{"my-svc"}},
+			},
+		},
+	}
+	mustCreate(t, fc, ing)
+
+	// Verify initial reconciliation
+	expectReconciled(t, ingPGR, "default", "test-ingress")
+	populateTLSSecret(context.Background(), fc, "test-pg", "my-svc.ts.net")
+	expectReconciled(t, ingPGR, "default", "test-ingress")
+	verifyServeConfig(t, fc, "svc:my-svc", false)
+	verifyVIPService(t, ft, "svc:my-svc", []string{"443"})
+	verifyTailscaledConfig(t, fc, []string{"svc:my-svc"})
+
+	// Update the Ingress hostname and make sure the original VIPService is deleted.
+	mustUpdate(t, fc, "default", "test-ingress", func(ing *networkingv1.Ingress) {
+		ing.Spec.TLS[0].Hosts[0] = "updated-svc"
+	})
+	expectReconciled(t, ingPGR, "default", "test-ingress")
+	populateTLSSecret(context.Background(), fc, "test-pg", "updated-svc.ts.net")
+	expectReconciled(t, ingPGR, "default", "test-ingress")
+	verifyServeConfig(t, fc, "svc:updated-svc", false)
+	verifyVIPService(t, ft, "svc:updated-svc", []string{"443"})
+	verifyTailscaledConfig(t, fc, []string{"svc:updated-svc"})
+
+	_, err := ft.GetVIPService(context.Background(), tailcfg.ServiceName("svc:my-svc"))
+	if err == nil {
+		t.Fatalf("svc:my-svc not cleaned up")
+	}
+	var errResp *tailscale.ErrResponse
+	if !errors.As(err, &errResp) || errResp.Status != http.StatusNotFound {
+		t.Fatalf("unexpected error: %v", err)
+	}
 }
 
 func TestValidateIngress(t *testing.T) {
@@ -392,6 +477,8 @@ func TestIngressPGReconciler_HTTPEndpoint(t *testing.T) {
 
 	// Verify initial reconciliation with HTTP enabled
 	expectReconciled(t, ingPGR, "default", "test-ingress")
+	populateTLSSecret(context.Background(), fc, "test-pg", "my-svc.ts.net")
+	expectReconciled(t, ingPGR, "default", "test-ingress")
 	verifyVIPService(t, ft, "svc:my-svc", []string{"80", "443"})
 	verifyServeConfig(t, fc, "svc:my-svc", true)
 
@@ -404,6 +491,31 @@ func TestIngressPGReconciler_HTTPEndpoint(t *testing.T) {
 		t.Fatal(err)
 	}
 
+	// Status will be empty until the VIPService shows up in prefs.
+	if !reflect.DeepEqual(ing.Status.LoadBalancer.Ingress, []networkingv1.IngressLoadBalancerIngress(nil)) {
+		t.Errorf("incorrect Ingress status: got %v, want empty",
+			ing.Status.LoadBalancer.Ingress)
+	}
+
+	// Add the VIPService to prefs to have the Ingress recognised as ready.
+	mustCreate(t, fc, &corev1.Secret{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "test-pg-0",
+			Namespace: "operator-ns",
+			Labels:    pgSecretLabels("test-pg", "state"),
+		},
+		Data: map[string][]byte{
+			"_current-profile": []byte("profile-foo"),
+			"profile-foo":      []byte(`{"AdvertiseServices":["svc:my-svc"],"Config":{"NodeID":"node-foo"}}`),
+		},
+	})
+
+	// Reconcile and re-fetch Ingress.
+	expectReconciled(t, ingPGR, "default", "test-ingress")
+	if err := fc.Get(context.Background(), client.ObjectKeyFromObject(ing), ing); err != nil {
+		t.Fatal(err)
+	}
+
 	wantStatus := []networkingv1.IngressPortStatus{
 		{Port: 443, Protocol: "TCP"},
 		{Port: 80, Protocol: "TCP"},
@@ -510,6 +622,7 @@ func verifyServeConfig(t *testing.T, fc client.Client, serviceName string, wantH
 }
 
 func verifyTailscaledConfig(t *testing.T, fc client.Client, expectedServices []string) {
+	t.Helper()
 	var expected string
 	if expectedServices != nil {
 		expectedServicesJSON, err := json.Marshal(expectedServices)
@@ -644,8 +757,10 @@ func TestIngressPGReconciler_MultiCluster(t *testing.T) {
 
 	// Simulate existing VIPService from another cluster
 	existingVIPSvc := &tailscale.VIPService{
-		Name:    "svc:my-svc",
-		Comment: `{"ownerrefs":[{"operatorID":"operator-2"}]}`,
+		Name: "svc:my-svc",
+		Annotations: map[string]string{
+			ownerAnnotation: `{"ownerrefs":[{"operatorID":"operator-2"}]}`,
+		},
 	}
 	ft.vipServices = map[tailcfg.ServiceName]*tailscale.VIPService{
 		"svc:my-svc": existingVIPSvc,
@@ -662,17 +777,17 @@ func TestIngressPGReconciler_MultiCluster(t *testing.T) {
 		t.Fatal("VIPService not found")
 	}
 
-	c := &comment{}
-	if err := json.Unmarshal([]byte(vipSvc.Comment), c); err != nil {
-		t.Fatalf("parsing comment: %v", err)
+	o, err := parseOwnerAnnotation(vipSvc)
+	if err != nil {
+		t.Fatalf("parsing owner annotation: %v", err)
 	}
 
 	wantOwnerRefs := []OwnerRef{
 		{OperatorID: "operator-2"},
 		{OperatorID: "operator-1"},
 	}
-	if !reflect.DeepEqual(c.OwnerRefs, wantOwnerRefs) {
-		t.Errorf("incorrect owner refs\ngot:  %+v\nwant: %+v", c.OwnerRefs, wantOwnerRefs)
+	if !reflect.DeepEqual(o.OwnerRefs, wantOwnerRefs) {
+		t.Errorf("incorrect owner refs\ngot:  %+v\nwant: %+v", o.OwnerRefs, wantOwnerRefs)
 	}
 
 	// Delete the Ingress and verify VIPService still exists with one owner ref
@@ -689,15 +804,40 @@ func TestIngressPGReconciler_MultiCluster(t *testing.T) {
 		t.Fatal("VIPService was incorrectly deleted")
 	}
 
-	c = &comment{}
-	if err := json.Unmarshal([]byte(vipSvc.Comment), c); err != nil {
-		t.Fatalf("parsing comment after deletion: %v", err)
+	o, err = parseOwnerAnnotation(vipSvc)
+	if err != nil {
+		t.Fatalf("parsing owner annotation: %v", err)
 	}
 
 	wantOwnerRefs = []OwnerRef{
 		{OperatorID: "operator-2"},
 	}
-	if !reflect.DeepEqual(c.OwnerRefs, wantOwnerRefs) {
-		t.Errorf("incorrect owner refs after deletion\ngot:  %+v\nwant: %+v", c.OwnerRefs, wantOwnerRefs)
+	if !reflect.DeepEqual(o.OwnerRefs, wantOwnerRefs) {
+		t.Errorf("incorrect owner refs after deletion\ngot:  %+v\nwant: %+v", o.OwnerRefs, wantOwnerRefs)
 	}
 }
+
+func populateTLSSecret(ctx context.Context, c client.Client, pgName, domain string) error {
+	secret := &corev1.Secret{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      domain,
+			Namespace: "operator-ns",
+			Labels: map[string]string{
+				kubetypes.LabelManaged:    "true",
+				labelProxyGroup:           pgName,
+				labelDomain:               domain,
+				kubetypes.LabelSecretType: "certs",
+			},
+		},
+		Type: corev1.SecretTypeTLS,
+		Data: map[string][]byte{
+			corev1.TLSCertKey:       []byte("fake-cert"),
+			corev1.TLSPrivateKeyKey: []byte("fake-key"),
+		},
+	}
+
+	_, err := createOrUpdate(ctx, c, "operator-ns", secret, func(s *corev1.Secret) {
+		s.Data = secret.Data
+	})
+	return err
+}

cmd/k8s-operator/ingress_test.go

@@ -6,6 +6,7 @@
 package main
 
 import (
+	"context"
 	"testing"
 
 	"go.uber.org/zap"
@@ -15,17 +16,18 @@ import (
 	apiextensionsv1 "k8s.io/apiextensions-apiserver/pkg/apis/apiextensions/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
+	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/client/fake"
 	"tailscale.com/ipn"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
 	"tailscale.com/kube/kubetypes"
+	"tailscale.com/tstest"
 	"tailscale.com/types/ptr"
 	"tailscale.com/util/mak"
 )
 
 func TestTailscaleIngress(t *testing.T) {
-	tsIngressClass := &networkingv1.IngressClass{ObjectMeta: metav1.ObjectMeta{Name: "tailscale"}, Spec: networkingv1.IngressClassSpec{Controller: "tailscale.com/ts-ingress"}}
-	fc := fake.NewFakeClient(tsIngressClass)
+	fc := fake.NewFakeClient(ingressClass())
 	ft := &fakeTSClient{}
 	fakeTsnetServer := &fakeTSNetServer{certDomains: []string{"foo.com"}}
 	zl, err := zap.NewDevelopment()
@@ -46,45 +48,8 @@ func TestTailscaleIngress(t *testing.T) {
 	}
 
 	// 1. Resources get created for regular Ingress
-	ing := &networkingv1.Ingress{
-		TypeMeta: metav1.TypeMeta{Kind: "Ingress", APIVersion: "networking.k8s.io/v1"},
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      "test",
-			Namespace: "default",
-			// The apiserver is supposed to set the UID, but the fake client
-			// doesn't. So, set it explicitly because other code later depends
-			// on it being set.
-			UID: types.UID("1234-UID"),
-		},
-		Spec: networkingv1.IngressSpec{
-			IngressClassName: ptr.To("tailscale"),
-			DefaultBackend: &networkingv1.IngressBackend{
-				Service: &networkingv1.IngressServiceBackend{
-					Name: "test",
-					Port: networkingv1.ServiceBackendPort{
-						Number: 8080,
-					},
-				},
-			},
-			TLS: []networkingv1.IngressTLS{
-				{Hosts: []string{"default-test"}},
-			},
-		},
-	}
-	mustCreate(t, fc, ing)
-	mustCreate(t, fc, &corev1.Service{
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      "test",
-			Namespace: "default",
-		},
-		Spec: corev1.ServiceSpec{
-			ClusterIP: "1.2.3.4",
-			Ports: []corev1.ServicePort{{
-				Port: 8080,
-				Name: "http"},
-			},
-		},
-	})
+	mustCreate(t, fc, ingress())
+	mustCreate(t, fc, service())
 
 	expectReconciled(t, ingR, "default", "test")
 
@@ -114,6 +79,9 @@ func TestTailscaleIngress(t *testing.T) {
 		mak.Set(&secret.Data, "device_fqdn", []byte("foo.tailnetxyz.ts.net"))
 	})
 	expectReconciled(t, ingR, "default", "test")
+
+	// Get the ingress and update it with expected changes
+	ing := ingress()
 	ing.Finalizers = append(ing.Finalizers, "tailscale.com/finalizer")
 	ing.Status.LoadBalancer = networkingv1.IngressLoadBalancerStatus{
 		Ingress: []networkingv1.IngressLoadBalancerIngress{
@@ -143,8 +111,7 @@ func TestTailscaleIngress(t *testing.T) {
 }
 
 func TestTailscaleIngressHostname(t *testing.T) {
-	tsIngressClass := &networkingv1.IngressClass{ObjectMeta: metav1.ObjectMeta{Name: "tailscale"}, Spec: networkingv1.IngressClassSpec{Controller: "tailscale.com/ts-ingress"}}
-	fc := fake.NewFakeClient(tsIngressClass)
+	fc := fake.NewFakeClient(ingressClass())
 	ft := &fakeTSClient{}
 	fakeTsnetServer := &fakeTSNetServer{certDomains: []string{"foo.com"}}
 	zl, err := zap.NewDevelopment()
@@ -165,45 +132,8 @@ func TestTailscaleIngressHostname(t *testing.T) {
 	}
 
 	// 1. Resources get created for regular Ingress
-	ing := &networkingv1.Ingress{
-		TypeMeta: metav1.TypeMeta{Kind: "Ingress", APIVersion: "networking.k8s.io/v1"},
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      "test",
-			Namespace: "default",
-			// The apiserver is supposed to set the UID, but the fake client
-			// doesn't. So, set it explicitly because other code later depends
-			// on it being set.
-			UID: types.UID("1234-UID"),
-		},
-		Spec: networkingv1.IngressSpec{
-			IngressClassName: ptr.To("tailscale"),
-			DefaultBackend: &networkingv1.IngressBackend{
-				Service: &networkingv1.IngressServiceBackend{
-					Name: "test",
-					Port: networkingv1.ServiceBackendPort{
-						Number: 8080,
-					},
-				},
-			},
-			TLS: []networkingv1.IngressTLS{
-				{Hosts: []string{"default-test"}},
-			},
-		},
-	}
-	mustCreate(t, fc, ing)
-	mustCreate(t, fc, &corev1.Service{
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      "test",
-			Namespace: "default",
-		},
-		Spec: corev1.ServiceSpec{
-			ClusterIP: "1.2.3.4",
-			Ports: []corev1.ServicePort{{
-				Port: 8080,
-				Name: "http"},
-			},
-		},
-	})
+	mustCreate(t, fc, ingress())
+	mustCreate(t, fc, service())
 
 	expectReconciled(t, ingR, "default", "test")
 
@@ -241,8 +171,10 @@ func TestTailscaleIngressHostname(t *testing.T) {
 		mak.Set(&secret.Data, "device_fqdn", []byte("foo.tailnetxyz.ts.net"))
 	})
 	expectReconciled(t, ingR, "default", "test")
-	ing.Finalizers = append(ing.Finalizers, "tailscale.com/finalizer")
 
+	// Get the ingress and update it with expected changes
+	ing := ingress()
+	ing.Finalizers = append(ing.Finalizers, "tailscale.com/finalizer")
 	expectEqual(t, fc, ing)
 
 	// 3. Ingress proxy with capability version >= 110 advertises HTTPS endpoint
@@ -299,10 +231,9 @@ func TestTailscaleIngressWithProxyClass(t *testing.T) {
 			Annotations: map[string]string{"bar.io/foo": "some-val"},
 			Pod:         &tsapi.Pod{Annotations: map[string]string{"foo.io/bar": "some-val"}}}},
 	}
-	tsIngressClass := &networkingv1.IngressClass{ObjectMeta: metav1.ObjectMeta{Name: "tailscale"}, Spec: networkingv1.IngressClassSpec{Controller: "tailscale.com/ts-ingress"}}
 	fc := fake.NewClientBuilder().
 		WithScheme(tsapi.GlobalScheme).
-		WithObjects(pc, tsIngressClass).
+		WithObjects(pc, ingressClass()).
 		WithStatusSubresource(pc).
 		Build()
 	ft := &fakeTSClient{}
@@ -326,45 +257,8 @@ func TestTailscaleIngressWithProxyClass(t *testing.T) {
 
 	// 1. Ingress is created with no ProxyClass specified, default proxy
 	// resources get configured.
-	ing := &networkingv1.Ingress{
-		TypeMeta: metav1.TypeMeta{Kind: "Ingress", APIVersion: "networking.k8s.io/v1"},
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      "test",
-			Namespace: "default",
-			// The apiserver is supposed to set the UID, but the fake client
-			// doesn't. So, set it explicitly because other code later depends
-			// on it being set.
-			UID: types.UID("1234-UID"),
-		},
-		Spec: networkingv1.IngressSpec{
-			IngressClassName: ptr.To("tailscale"),
-			DefaultBackend: &networkingv1.IngressBackend{
-				Service: &networkingv1.IngressServiceBackend{
-					Name: "test",
-					Port: networkingv1.ServiceBackendPort{
-						Number: 8080,
-					},
-				},
-			},
-			TLS: []networkingv1.IngressTLS{
-				{Hosts: []string{"default-test"}},
-			},
-		},
-	}
-	mustCreate(t, fc, ing)
-	mustCreate(t, fc, &corev1.Service{
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      "test",
-			Namespace: "default",
-		},
-		Spec: corev1.ServiceSpec{
-			ClusterIP: "1.2.3.4",
-			Ports: []corev1.ServicePort{{
-				Port: 8080,
-				Name: "http"},
-			},
-		},
-	})
+	mustCreate(t, fc, ingress())
+	mustCreate(t, fc, service())
 
 	expectReconciled(t, ingR, "default", "test")
 
@@ -432,54 +326,19 @@ func TestTailscaleIngressWithServiceMonitor(t *testing.T) {
 				ObservedGeneration: 1,
 			}}},
 	}
-	ing := &networkingv1.Ingress{
-		TypeMeta: metav1.TypeMeta{Kind: "Ingress", APIVersion: "networking.k8s.io/v1"},
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      "test",
-			Namespace: "default",
-			// The apiserver is supposed to set the UID, but the fake client
-			// doesn't. So, set it explicitly because other code later depends
-			// on it being set.
-			UID: types.UID("1234-UID"),
-			Labels: map[string]string{
-				"tailscale.com/proxy-class": "metrics",
-			},
-		},
-		Spec: networkingv1.IngressSpec{
-			IngressClassName: ptr.To("tailscale"),
-			DefaultBackend: &networkingv1.IngressBackend{
-				Service: &networkingv1.IngressServiceBackend{
-					Name: "test",
-					Port: networkingv1.ServiceBackendPort{
-						Number: 8080,
-					},
-				},
-			},
-			TLS: []networkingv1.IngressTLS{
-				{Hosts: []string{"default-test"}},
-			},
-		},
-	}
-	svc := &corev1.Service{
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      "test",
-			Namespace: "default",
-		},
-		Spec: corev1.ServiceSpec{
-			ClusterIP: "1.2.3.4",
-			Ports: []corev1.ServicePort{{
-				Port: 8080,
-				Name: "http"},
-			},
-		},
-	}
 	crd := &apiextensionsv1.CustomResourceDefinition{ObjectMeta: metav1.ObjectMeta{Name: serviceMonitorCRD}}
-	tsIngressClass := &networkingv1.IngressClass{ObjectMeta: metav1.ObjectMeta{Name: "tailscale"}, Spec: networkingv1.IngressClassSpec{Controller: "tailscale.com/ts-ingress"}}
+
+	// Create fake client with ProxyClass, IngressClass, Ingress with metrics ProxyClass, and Service
+	ing := ingress()
+	ing.Labels = map[string]string{
+		LabelProxyClass: "metrics",
+	}
 	fc := fake.NewClientBuilder().
 		WithScheme(tsapi.GlobalScheme).
-		WithObjects(pc, tsIngressClass, ing, svc).
+		WithObjects(pc, ingressClass(), ing, service()).
 		WithStatusSubresource(pc).
 		Build()
+
 	ft := &fakeTSClient{}
 	fakeTsnetServer := &fakeTSNetServer{certDomains: []string{"foo.com"}}
 	zl, err := zap.NewDevelopment()
@@ -560,3 +419,118 @@ func TestTailscaleIngressWithServiceMonitor(t *testing.T) {
 	expectMissing[corev1.Service](t, fc, "operator-ns", metricsResourceName(shortName))
 	// ServiceMonitor gets garbage collected when the Service is deleted - we cannot test that here.
 }
+
+func TestIngressLetsEncryptStaging(t *testing.T) {
+	cl := tstest.NewClock(tstest.ClockOpts{})
+	zl := zap.Must(zap.NewDevelopment())
+
+	pcLEStaging, pcLEStagingFalse, pcOther := proxyClassesForLEStagingTest()
+
+	testCases := testCasesForLEStagingTests(pcLEStaging, pcLEStagingFalse, pcOther)
+
+	for _, tt := range testCases {
+		t.Run(tt.name, func(t *testing.T) {
+			builder := fake.NewClientBuilder().
+				WithScheme(tsapi.GlobalScheme)
+
+			builder = builder.WithObjects(pcLEStaging, pcLEStagingFalse, pcOther).
+				WithStatusSubresource(pcLEStaging, pcLEStagingFalse, pcOther)
+
+			fc := builder.Build()
+
+			if tt.proxyClassPerResource != "" || tt.defaultProxyClass != "" {
+				name := tt.proxyClassPerResource
+				if name == "" {
+					name = tt.defaultProxyClass
+				}
+				setProxyClassReady(t, fc, cl, name)
+			}
+
+			mustCreate(t, fc, ingressClass())
+			mustCreate(t, fc, service())
+			ing := ingress()
+			if tt.proxyClassPerResource != "" {
+				ing.Labels = map[string]string{
+					LabelProxyClass: tt.proxyClassPerResource,
+				}
+			}
+			mustCreate(t, fc, ing)
+
+			ingR := &IngressReconciler{
+				Client: fc,
+				ssr: &tailscaleSTSReconciler{
+					Client:            fc,
+					tsClient:          &fakeTSClient{},
+					tsnetServer:       &fakeTSNetServer{certDomains: []string{"test-host"}},
+					defaultTags:       []string{"tag:test"},
+					operatorNamespace: "operator-ns",
+					proxyImage:        "tailscale/tailscale:test",
+				},
+				logger:            zl.Sugar(),
+				defaultProxyClass: tt.defaultProxyClass,
+			}
+
+			expectReconciled(t, ingR, "default", "test")
+
+			_, shortName := findGenName(t, fc, "default", "test", "ingress")
+			sts := &appsv1.StatefulSet{}
+			if err := fc.Get(context.Background(), client.ObjectKey{Namespace: "operator-ns", Name: shortName}, sts); err != nil {
+				t.Fatalf("failed to get StatefulSet: %v", err)
+			}
+
+			if tt.useLEStagingEndpoint {
+				verifyEnvVar(t, sts, "TS_DEBUG_ACME_DIRECTORY_URL", letsEncryptStagingEndpoint)
+			} else {
+				verifyEnvVarNotPresent(t, sts, "TS_DEBUG_ACME_DIRECTORY_URL")
+			}
+		})
+	}
+}
+
+func ingressClass() *networkingv1.IngressClass {
+	return &networkingv1.IngressClass{
+		ObjectMeta: metav1.ObjectMeta{Name: "tailscale"},
+		Spec:       networkingv1.IngressClassSpec{Controller: "tailscale.com/ts-ingress"},
+	}
+}
+
+func service() *corev1.Service {
+	return &corev1.Service{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "test",
+			Namespace: "default",
+		},
+		Spec: corev1.ServiceSpec{
+			ClusterIP: "1.2.3.4",
+			Ports: []corev1.ServicePort{{
+				Port: 8080,
+				Name: "http"},
+			},
+		},
+	}
+}
+
+func ingress() *networkingv1.Ingress {
+	return &networkingv1.Ingress{
+		TypeMeta: metav1.TypeMeta{Kind: "Ingress", APIVersion: "networking.k8s.io/v1"},
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "test",
+			Namespace: "default",
+			UID:       types.UID("1234-UID"),
+		},
+		Spec: networkingv1.IngressSpec{
+			IngressClassName: ptr.To("tailscale"),
+			DefaultBackend: &networkingv1.IngressBackend{
+				Service: &networkingv1.IngressServiceBackend{
+					Name: "test",
+					Port: networkingv1.ServiceBackendPort{
+						Number: 8080,
+					},
+				},
+			},
+			TLS: []networkingv1.IngressTLS{
+				{Hosts: []string{"default-test"}},
+			},
+		},
+	}
+}

cmd/k8s-operator/metrics_resources.go

@@ -19,6 +19,7 @@ import (
 	"k8s.io/apimachinery/pkg/types"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
+	"tailscale.com/kube/kubetypes"
 )
 
 const (
@@ -222,7 +223,7 @@ func metricsResourceName(stsName string) string {
 // proxy.
 func metricsResourceLabels(opts *metricsOpts) map[string]string {
 	lbls := map[string]string{
-		LabelManaged:             "true",
+		kubetypes.LabelManaged:   "true",
 		labelMetricsTarget:       opts.proxyStsName,
 		labelPromProxyType:       opts.proxyType,
 		labelPromProxyParentName: opts.proxyLabels[LabelParentName],

cmd/k8s-operator/operator.go

@@ -9,7 +9,6 @@ package main
 
 import (
 	"context"
-	"fmt"
 	"net/http"
 	"os"
 	"regexp"
@@ -40,7 +39,6 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/manager"
 	"sigs.k8s.io/controller-runtime/pkg/manager/signals"
 	"sigs.k8s.io/controller-runtime/pkg/reconcile"
-	"tailscale.com/client/local"
 	"tailscale.com/client/tailscale"
 	"tailscale.com/hostinfo"
 	"tailscale.com/ipn"
@@ -333,39 +331,6 @@ func runReconcilers(opts reconcilerOpts) {
 	if err != nil {
 		startlog.Fatalf("could not create ingress reconciler: %v", err)
 	}
-	lc, err := opts.tsServer.LocalClient()
-	if err != nil {
-		startlog.Fatalf("could not get local client: %v", err)
-	}
-	id, err := id(context.Background(), lc)
-	if err != nil {
-		startlog.Fatalf("error determining stable ID of the operator's Tailscale device: %v", err)
-	}
-	ingressProxyGroupFilter := handler.EnqueueRequestsFromMapFunc(ingressesFromIngressProxyGroup(mgr.GetClient(), opts.log))
-	err = builder.
-		ControllerManagedBy(mgr).
-		For(&networkingv1.Ingress{}).
-		Named("ingress-pg-reconciler").
-		Watches(&corev1.Service{}, handler.EnqueueRequestsFromMapFunc(serviceHandlerForIngressPG(mgr.GetClient(), startlog))).
-		Watches(&tsapi.ProxyGroup{}, ingressProxyGroupFilter).
-		Complete(&HAIngressReconciler{
-			recorder:    eventRecorder,
-			tsClient:    opts.tsClient,
-			tsnetServer: opts.tsServer,
-			defaultTags: strings.Split(opts.proxyTags, ","),
-			Client:      mgr.GetClient(),
-			logger:      opts.log.Named("ingress-pg-reconciler"),
-			lc:          lc,
-			operatorID:  id,
-			tsNamespace: opts.tailscaleNamespace,
-		})
-	if err != nil {
-		startlog.Fatalf("could not create ingress-pg-reconciler: %v", err)
-	}
-	if err := mgr.GetFieldIndexer().IndexField(context.Background(), new(networkingv1.Ingress), indexIngressProxyGroup, indexPGIngresses); err != nil {
-		startlog.Fatalf("failed setting up indexer for HA Ingresses: %v", err)
-	}
-
 	connectorFilter := handler.EnqueueRequestsFromMapFunc(managedResourceHandlerForType("connector"))
 	// If a ProxyClassChanges, enqueue all Connectors that have
 	// .spec.proxyClass set to the name of this ProxyClass.
@@ -636,8 +601,8 @@ func enqueueAllIngressEgressProxySvcsInNS(ns string, cl client.Client, logger *z
 
 		// Get all headless Services for proxies configured using Service.
 		svcProxyLabels := map[string]string{
-			LabelManaged:    "true",
-			LabelParentType: "svc",
+			kubetypes.LabelManaged: "true",
+			LabelParentType:        "svc",
 		}
 		svcHeadlessSvcList := &corev1.ServiceList{}
 		if err := cl.List(ctx, svcHeadlessSvcList, client.InNamespace(ns), client.MatchingLabels(svcProxyLabels)); err != nil {
@@ -650,8 +615,8 @@ func enqueueAllIngressEgressProxySvcsInNS(ns string, cl client.Client, logger *z
 
 		// Get all headless Services for proxies configured using Ingress.
 		ingProxyLabels := map[string]string{
-			LabelManaged:    "true",
-			LabelParentType: "ingress",
+			kubetypes.LabelManaged: "true",
+			LabelParentType:        "ingress",
 		}
 		ingHeadlessSvcList := &corev1.ServiceList{}
 		if err := cl.List(ctx, ingHeadlessSvcList, client.InNamespace(ns), client.MatchingLabels(ingProxyLabels)); err != nil {
@@ -718,7 +683,7 @@ func dnsRecordsReconcilerIngressHandler(ns string, isDefaultLoadBalancer bool, c
 
 func isManagedResource(o client.Object) bool {
 	ls := o.GetLabels()
-	return ls[LabelManaged] == "true"
+	return ls[kubetypes.LabelManaged] == "true"
 }
 
 func isManagedByType(o client.Object, typ string) bool {
@@ -955,7 +920,7 @@ func egressPodsHandler(_ context.Context, o client.Object) []reconcile.Request {
 // returns reconciler requests for all egress EndpointSlices for that ProxyGroup.
 func egressEpsFromPGPods(cl client.Client, ns string) handler.MapFunc {
 	return func(_ context.Context, o client.Object) []reconcile.Request {
-		if v, ok := o.GetLabels()[LabelManaged]; !ok || v != "true" {
+		if v, ok := o.GetLabels()[kubetypes.LabelManaged]; !ok || v != "true" {
 			return nil
 		}
 		// TODO(irbekrm): for now this is good enough as all ProxyGroups are egress. Add a type check once we
@@ -975,15 +940,13 @@ func egressEpsFromPGPods(cl client.Client, ns string) handler.MapFunc {
 // returns reconciler requests for all egress EndpointSlices for that ProxyGroup.
 func egressEpsFromPGStateSecrets(cl client.Client, ns string) handler.MapFunc {
 	return func(_ context.Context, o client.Object) []reconcile.Request {
-		if v, ok := o.GetLabels()[LabelManaged]; !ok || v != "true" {
+		if v, ok := o.GetLabels()[kubetypes.LabelManaged]; !ok || v != "true" {
 			return nil
 		}
-		// TODO(irbekrm): for now this is good enough as all ProxyGroups are egress. Add a type check once we
-		// have ingress ProxyGroups.
 		if parentType := o.GetLabels()[LabelParentType]; parentType != "proxygroup" {
 			return nil
 		}
-		if secretType := o.GetLabels()[labelSecretType]; secretType != "state" {
+		if secretType := o.GetLabels()[kubetypes.LabelSecretType]; secretType != "state" {
 			return nil
 		}
 		pg, ok := o.GetLabels()[LabelParentName]
@@ -1000,7 +963,7 @@ func egressSvcFromEps(_ context.Context, o client.Object) []reconcile.Request {
 	if typ := o.GetLabels()[labelSvcType]; typ != typeEgress {
 		return nil
 	}
-	if v, ok := o.GetLabels()[LabelManaged]; !ok || v != "true" {
+	if v, ok := o.GetLabels()[kubetypes.LabelManaged]; !ok || v != "true" {
 		return nil
 	}
 	svcName, ok := o.GetLabels()[LabelParentName]
@@ -1070,36 +1033,6 @@ func egressSvcsFromEgressProxyGroup(cl client.Client, logger *zap.SugaredLogger)
 	}
 }
 
-// ingressesFromIngressProxyGroup is an event handler for ingress ProxyGroups. It returns reconcile requests for all
-// user-created Ingresses that should be exposed on this ProxyGroup.
-func ingressesFromIngressProxyGroup(cl client.Client, logger *zap.SugaredLogger) handler.MapFunc {
-	return func(ctx context.Context, o client.Object) []reconcile.Request {
-		pg, ok := o.(*tsapi.ProxyGroup)
-		if !ok {
-			logger.Infof("[unexpected] ProxyGroup handler triggered for an object that is not a ProxyGroup")
-			return nil
-		}
-		if pg.Spec.Type != tsapi.ProxyGroupTypeIngress {
-			return nil
-		}
-		ingList := &networkingv1.IngressList{}
-		if err := cl.List(ctx, ingList, client.MatchingFields{indexIngressProxyGroup: pg.Name}); err != nil {
-			logger.Infof("error listing Ingresses: %v, skipping a reconcile for event on ProxyGroup %s", err, pg.Name)
-			return nil
-		}
-		reqs := make([]reconcile.Request, 0)
-		for _, svc := range ingList.Items {
-			reqs = append(reqs, reconcile.Request{
-				NamespacedName: types.NamespacedName{
-					Namespace: svc.Namespace,
-					Name:      svc.Name,
-				},
-			})
-		}
-		return reqs
-	}
-}
-
 // epsFromExternalNameService is an event handler for ExternalName Services that define a Tailscale egress service that
 // should be exposed on a ProxyGroup. It returns reconcile requests for EndpointSlices created for this Service.
 func epsFromExternalNameService(cl client.Client, logger *zap.SugaredLogger, ns string) handler.MapFunc {
@@ -1145,9 +1078,9 @@ func podsFromEgressEps(cl client.Client, logger *zap.SugaredLogger, ns string) h
 			return nil
 		}
 		podLabels := map[string]string{
-			LabelManaged:    "true",
-			LabelParentType: "proxygroup",
-			LabelParentName: eps.Labels[labelProxyGroup],
+			kubetypes.LabelManaged: "true",
+			LabelParentType:        "proxygroup",
+			LabelParentName:        eps.Labels[labelProxyGroup],
 		}
 		podList := &corev1.PodList{}
 		if err := cl.List(ctx, podList, client.InNamespace(ns),
@@ -1220,63 +1153,7 @@ func indexEgressServices(o client.Object) []string {
 	return []string{o.GetAnnotations()[AnnotationProxyGroup]}
 }
 
-// indexPGIngresses adds a local index to a cached Tailscale Ingresses meant to be exposed on a ProxyGroup. The index is
-// used a list filter.
-func indexPGIngresses(o client.Object) []string {
-	if !hasProxyGroupAnnotation(o) {
-		return nil
-	}
-	return []string{o.GetAnnotations()[AnnotationProxyGroup]}
-}
-
-// serviceHandlerForIngressPG returns a handler for Service events that ensures that if the Service
-// associated with an event is a backend Service for a tailscale Ingress with ProxyGroup annotation,
-// the associated Ingress gets reconciled.
-func serviceHandlerForIngressPG(cl client.Client, logger *zap.SugaredLogger) handler.MapFunc {
-	return func(ctx context.Context, o client.Object) []reconcile.Request {
-		ingList := networkingv1.IngressList{}
-		if err := cl.List(ctx, &ingList, client.InNamespace(o.GetNamespace())); err != nil {
-			logger.Debugf("error listing Ingresses: %v", err)
-			return nil
-		}
-		reqs := make([]reconcile.Request, 0)
-		for _, ing := range ingList.Items {
-			if ing.Spec.IngressClassName == nil || *ing.Spec.IngressClassName != tailscaleIngressClassName {
-				continue
-			}
-			if !hasProxyGroupAnnotation(&ing) {
-				continue
-			}
-			if ing.Spec.DefaultBackend != nil && ing.Spec.DefaultBackend.Service != nil && ing.Spec.DefaultBackend.Service.Name == o.GetName() {
-				reqs = append(reqs, reconcile.Request{NamespacedName: client.ObjectKeyFromObject(&ing)})
-			}
-			for _, rule := range ing.Spec.Rules {
-				if rule.HTTP == nil {
-					continue
-				}
-				for _, path := range rule.HTTP.Paths {
-					if path.Backend.Service != nil && path.Backend.Service.Name == o.GetName() {
-						reqs = append(reqs, reconcile.Request{NamespacedName: client.ObjectKeyFromObject(&ing)})
-					}
-				}
-			}
-		}
-		return reqs
-	}
-}
-
 func hasProxyGroupAnnotation(obj client.Object) bool {
 	ing := obj.(*networkingv1.Ingress)
 	return ing.Annotations[AnnotationProxyGroup] != ""
 }
-
-func id(ctx context.Context, lc *local.Client) (string, error) {
-	st, err := lc.StatusWithoutPeers(ctx)
-	if err != nil {
-		return "", fmt.Errorf("error getting tailscale status: %w", err)
-	}
-	if st.Self == nil {
-		return "", fmt.Errorf("unexpected: device's status does not contain node's metadata")
-	}
-	return string(st.Self.ID), nil
-}

cmd/k8s-operator/operator_test.go

@@ -1387,10 +1387,10 @@ func Test_serviceHandlerForIngress(t *testing.T) {
 			Name:      "headless-1",
 			Namespace: "tailscale",
 			Labels: map[string]string{
-				LabelManaged:         "true",
-				LabelParentName:      "ing-1",
-				LabelParentNamespace: "ns-1",
-				LabelParentType:      "ingress",
+				kubetypes.LabelManaged: "true",
+				LabelParentName:        "ing-1",
+				LabelParentNamespace:   "ns-1",
+				LabelParentType:        "ingress",
 			},
 		},
 	}

cmd/k8s-operator/proxygroup.go

@@ -302,7 +302,10 @@ func (r *ProxyGroupReconciler) maybeProvision(ctx context.Context, pg *tsapi.Pro
 	if err != nil {
 		return fmt.Errorf("error generating StatefulSet spec: %w", err)
 	}
-	ss = applyProxyClassToStatefulSet(proxyClass, ss, nil, logger)
+	cfg := &tailscaleSTSConfig{
+		proxyType: string(pg.Spec.Type),
+	}
+	ss = applyProxyClassToStatefulSet(proxyClass, ss, cfg, logger)
 	capver, err := r.capVerForPG(ctx, pg, logger)
 	if err != nil {
 		return fmt.Errorf("error getting device info: %w", err)
@@ -461,7 +464,7 @@ func (r *ProxyGroupReconciler) ensureConfigSecretsCreated(ctx context.Context, p
 
 		var existingCfgSecret *corev1.Secret // unmodified copy of secret
 		if err := r.Get(ctx, client.ObjectKeyFromObject(cfgSecret), cfgSecret); err == nil {
-			logger.Debugf("secret %s/%s already exists", cfgSecret.GetNamespace(), cfgSecret.GetName())
+			logger.Debugf("Secret %s/%s already exists", cfgSecret.GetNamespace(), cfgSecret.GetName())
 			existingCfgSecret = cfgSecret.DeepCopy()
 		} else if !apierrors.IsNotFound(err) {
 			return "", err
@@ -469,7 +472,7 @@ func (r *ProxyGroupReconciler) ensureConfigSecretsCreated(ctx context.Context, p
 
 		var authKey string
 		if existingCfgSecret == nil {
-			logger.Debugf("creating authkey for new ProxyGroup proxy")
+			logger.Debugf("Creating authkey for new ProxyGroup proxy")
 			tags := pg.Spec.Tags.Stringify()
 			if len(tags) == 0 {
 				tags = r.defaultTags
@@ -490,7 +493,7 @@ func (r *ProxyGroupReconciler) ensureConfigSecretsCreated(ctx context.Context, p
 			if err != nil {
 				return "", fmt.Errorf("error marshalling tailscaled config: %w", err)
 			}
-			mak.Set(&cfgSecret.StringData, tsoperator.TailscaledConfigFileName(cap), string(cfgJSON))
+			mak.Set(&cfgSecret.Data, tsoperator.TailscaledConfigFileName(cap), cfgJSON)
 		}
 
 		// The config sha256 sum is a value for a hash annotation used to trigger
@@ -520,12 +523,14 @@ func (r *ProxyGroupReconciler) ensureConfigSecretsCreated(ctx context.Context, p
 		}
 
 		if existingCfgSecret != nil {
-			logger.Debugf("patching the existing ProxyGroup config Secret %s", cfgSecret.Name)
-			if err := r.Patch(ctx, cfgSecret, client.MergeFrom(existingCfgSecret)); err != nil {
-				return "", err
+			if !apiequality.Semantic.DeepEqual(existingCfgSecret, cfgSecret) {
+				logger.Debugf("Updating the existing ProxyGroup config Secret %s", cfgSecret.Name)
+				if err := r.Update(ctx, cfgSecret); err != nil {
+					return "", err
+				}
 			}
 		} else {
-			logger.Debugf("creating a new config Secret %s for the ProxyGroup", cfgSecret.Name)
+			logger.Debugf("Creating a new config Secret %s for the ProxyGroup", cfgSecret.Name)
 			if err := r.Create(ctx, cfgSecret); err != nil {
 				return "", err
 			}
@@ -645,7 +650,7 @@ func (r *ProxyGroupReconciler) getNodeMetadata(ctx context.Context, pg *tsapi.Pr
 			return nil, fmt.Errorf("unexpected secret %s was labelled as owned by the ProxyGroup %s: %w", secret.Name, pg.Name, err)
 		}
 
-		id, dnsName, ok, err := getNodeMetadata(ctx, &secret)
+		prefs, ok, err := getDevicePrefs(&secret)
 		if err != nil {
 			return nil, err
 		}
@@ -656,8 +661,8 @@ func (r *ProxyGroupReconciler) getNodeMetadata(ctx context.Context, pg *tsapi.Pr
 		nm := nodeMetadata{
 			ordinal:     ordinal,
 			stateSecret: &secret,
-			tsID:        id,
-			dnsName:     dnsName,
+			tsID:        prefs.Config.NodeID,
+			dnsName:     prefs.Config.UserProfile.LoginName,
 		}
 		pod := &corev1.Pod{}
 		if err := r.Get(ctx, client.ObjectKey{Namespace: r.tsNamespace, Name: secret.Name}, pod); err != nil && !apierrors.IsNotFound(err) {

cmd/k8s-operator/proxygroup_specs.go

@@ -178,7 +178,15 @@ func pgStatefulSet(pg *tsapi.ProxyGroup, namespace, image, tsFirewallMode string
 				corev1.EnvVar{
 					Name:  "TS_SERVE_CONFIG",
 					Value: fmt.Sprintf("/etc/proxies/%s", serveConfigKey),
-				})
+				},
+				corev1.EnvVar{
+					// Run proxies in cert share mode to
+					// ensure that only one TLS cert is
+					// issued for an HA Ingress.
+					Name:  "TS_EXPERIMENTAL_CERT_SHARE",
+					Value: "true",
+				},
+			)
 		}
 		return append(c.Env, envs...)
 	}()
@@ -225,6 +233,13 @@ func pgRole(pg *tsapi.ProxyGroup, namespace string) *rbacv1.Role {
 			OwnerReferences: pgOwnerReference(pg),
 		},
 		Rules: []rbacv1.PolicyRule{
+			{
+				APIGroups: []string{""},
+				Resources: []string{"secrets"},
+				Verbs: []string{
+					"list",
+				},
+			},
 			{
 				APIGroups: []string{""},
 				Resources: []string{"secrets"},
@@ -318,9 +333,9 @@ func pgIngressCM(pg *tsapi.ProxyGroup, namespace string) *corev1.ConfigMap {
 	}
 }
 
-func pgSecretLabels(pgName, typ string) map[string]string {
+func pgSecretLabels(pgName, secretType string) map[string]string {
 	return pgLabels(pgName, map[string]string{
-		labelSecretType: typ, // "config" or "state".
+		kubetypes.LabelSecretType: secretType, // "config" or "state".
 	})
 }
 
@@ -330,7 +345,7 @@ func pgLabels(pgName string, customLabels map[string]string) map[string]string {
 		l[k] = v
 	}
 
-	l[LabelManaged] = "true"
+	l[kubetypes.LabelManaged] = "true"
 	l[LabelParentType] = "proxygroup"
 	l[LabelParentName] = pgName
 

cmd/k8s-operator/proxygroup_test.go

@@ -247,7 +247,6 @@ func TestProxyGroup(t *testing.T) {
 		// The fake client does not clean up objects whose owner has been
 		// deleted, so we can't test for the owned resources getting deleted.
 	})
-
 }
 
 func TestProxyGroupTypes(t *testing.T) {
@@ -417,6 +416,7 @@ func TestProxyGroupTypes(t *testing.T) {
 		}
 		verifyEnvVar(t, sts, "TS_INTERNAL_APP", kubetypes.AppProxyGroupIngress)
 		verifyEnvVar(t, sts, "TS_SERVE_CONFIG", "/etc/proxies/serve-config.json")
+		verifyEnvVar(t, sts, "TS_EXPERIMENTAL_CERT_SHARE", "true")
 
 		// Verify ConfigMap volume mount
 		cmName := fmt.Sprintf("%s-ingress-config", pg.Name)
@@ -475,8 +475,6 @@ func TestIngressAdvertiseServicesConfigPreserved(t *testing.T) {
 			Name:      pgConfigSecretName(pgName, 0),
 			Namespace: tsNamespace,
 		},
-		// Write directly to Data because the fake client doesn't copy the write-only
-		// StringData field across to Data for us.
 		Data: map[string][]byte{
 			tsoperator.TailscaledConfigFileName(106): existingConfigBytes,
 		},
@@ -514,10 +512,64 @@ func TestIngressAdvertiseServicesConfigPreserved(t *testing.T) {
 			Namespace:       tsNamespace,
 			ResourceVersion: "2",
 		},
-		StringData: map[string]string{
-			tsoperator.TailscaledConfigFileName(106): string(expectedConfigBytes),
+		Data: map[string][]byte{
+			tsoperator.TailscaledConfigFileName(106): expectedConfigBytes,
 		},
-	}, omitSecretData)
+	})
+}
+
+func proxyClassesForLEStagingTest() (*tsapi.ProxyClass, *tsapi.ProxyClass, *tsapi.ProxyClass) {
+	pcLEStaging := &tsapi.ProxyClass{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:       "le-staging",
+			Generation: 1,
+		},
+		Spec: tsapi.ProxyClassSpec{
+			UseLetsEncryptStagingEnvironment: true,
+		},
+	}
+
+	pcLEStagingFalse := &tsapi.ProxyClass{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:       "le-staging-false",
+			Generation: 1,
+		},
+		Spec: tsapi.ProxyClassSpec{
+			UseLetsEncryptStagingEnvironment: false,
+		},
+	}
+
+	pcOther := &tsapi.ProxyClass{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:       "other",
+			Generation: 1,
+		},
+		Spec: tsapi.ProxyClassSpec{},
+	}
+
+	return pcLEStaging, pcLEStagingFalse, pcOther
+}
+
+func setProxyClassReady(t *testing.T, fc client.Client, cl *tstest.Clock, name string) *tsapi.ProxyClass {
+	t.Helper()
+	pc := &tsapi.ProxyClass{}
+	if err := fc.Get(context.Background(), client.ObjectKey{Name: name}, pc); err != nil {
+		t.Fatal(err)
+	}
+	pc.Status = tsapi.ProxyClassStatus{
+		Conditions: []metav1.Condition{{
+			Type:               string(tsapi.ProxyClassReady),
+			Status:             metav1.ConditionTrue,
+			Reason:             reasonProxyClassValid,
+			Message:            reasonProxyClassValid,
+			LastTransitionTime: metav1.Time{Time: cl.Now().Truncate(time.Second)},
+			ObservedGeneration: pc.Generation,
+		}},
+	}
+	if err := fc.Status().Update(context.Background(), pc); err != nil {
+		t.Fatal(err)
+	}
+	return pc
 }
 
 func verifyProxyGroupCounts(t *testing.T, r *ProxyGroupReconciler, wantIngress, wantEgress int) {
@@ -543,6 +595,16 @@ func verifyEnvVar(t *testing.T, sts *appsv1.StatefulSet, name, expectedValue str
 	t.Errorf("%s environment variable not found", name)
 }
 
+func verifyEnvVarNotPresent(t *testing.T, sts *appsv1.StatefulSet, name string) {
+	t.Helper()
+	for _, env := range sts.Spec.Template.Spec.Containers[0].Env {
+		if env.Name == name {
+			t.Errorf("environment variable %s should not be present", name)
+			return
+		}
+	}
+}
+
 func expectProxyGroupResources(t *testing.T, fc client.WithWatch, pg *tsapi.ProxyGroup, shouldExist bool, cfgHash string, proxyClass *tsapi.ProxyClass) {
 	t.Helper()
 
@@ -621,10 +683,145 @@ func addNodeIDToStateSecrets(t *testing.T, fc client.WithWatch, pg *tsapi.ProxyG
 	}
 }
 
-// The operator mostly writes to StringData and reads from Data, but the fake
-// client doesn't copy StringData across to Data on write. When comparing actual
-// vs expected Secrets, use this function to only check what the operator writes
-// to StringData.
-func omitSecretData(secret *corev1.Secret) {
-	secret.Data = nil
+func TestProxyGroupLetsEncryptStaging(t *testing.T) {
+	cl := tstest.NewClock(tstest.ClockOpts{})
+	zl := zap.Must(zap.NewDevelopment())
+
+	// Set up test cases- most are shared with non-HA Ingress.
+	type proxyGroupLETestCase struct {
+		leStagingTestCase
+		pgType tsapi.ProxyGroupType
+	}
+	pcLEStaging, pcLEStagingFalse, pcOther := proxyClassesForLEStagingTest()
+	sharedTestCases := testCasesForLEStagingTests(pcLEStaging, pcLEStagingFalse, pcOther)
+	var tests []proxyGroupLETestCase
+	for _, tt := range sharedTestCases {
+		tests = append(tests, proxyGroupLETestCase{
+			leStagingTestCase: tt,
+			pgType:            tsapi.ProxyGroupTypeIngress,
+		})
+	}
+	tests = append(tests, proxyGroupLETestCase{
+		leStagingTestCase: leStagingTestCase{
+			name:                  "egress_pg_with_staging_proxyclass",
+			proxyClassPerResource: "le-staging",
+			useLEStagingEndpoint:  false,
+		},
+		pgType: tsapi.ProxyGroupTypeEgress,
+	})
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			builder := fake.NewClientBuilder().
+				WithScheme(tsapi.GlobalScheme)
+
+			// Pre-populate the fake client with ProxyClasses.
+			builder = builder.WithObjects(pcLEStaging, pcLEStagingFalse, pcOther).
+				WithStatusSubresource(pcLEStaging, pcLEStagingFalse, pcOther)
+
+			fc := builder.Build()
+
+			// If the test case needs a ProxyClass to exist, ensure it is set to Ready.
+			if tt.proxyClassPerResource != "" || tt.defaultProxyClass != "" {
+				name := tt.proxyClassPerResource
+				if name == "" {
+					name = tt.defaultProxyClass
+				}
+				setProxyClassReady(t, fc, cl, name)
+			}
+
+			// Create ProxyGroup
+			pg := &tsapi.ProxyGroup{
+				ObjectMeta: metav1.ObjectMeta{
+					Name: "test",
+				},
+				Spec: tsapi.ProxyGroupSpec{
+					Type:       tt.pgType,
+					Replicas:   ptr.To[int32](1),
+					ProxyClass: tt.proxyClassPerResource,
+				},
+			}
+			mustCreate(t, fc, pg)
+
+			reconciler := &ProxyGroupReconciler{
+				tsNamespace:       tsNamespace,
+				proxyImage:        testProxyImage,
+				defaultTags:       []string{"tag:test"},
+				defaultProxyClass: tt.defaultProxyClass,
+				Client:            fc,
+				tsClient:          &fakeTSClient{},
+				l:                 zl.Sugar(),
+				clock:             cl,
+			}
+
+			expectReconciled(t, reconciler, "", pg.Name)
+
+			// Verify that the StatefulSet created for ProxyGrup has
+			// the expected setting for the staging endpoint.
+			sts := &appsv1.StatefulSet{}
+			if err := fc.Get(context.Background(), client.ObjectKey{Namespace: tsNamespace, Name: pg.Name}, sts); err != nil {
+				t.Fatalf("failed to get StatefulSet: %v", err)
+			}
+
+			if tt.useLEStagingEndpoint {
+				verifyEnvVar(t, sts, "TS_DEBUG_ACME_DIRECTORY_URL", letsEncryptStagingEndpoint)
+			} else {
+				verifyEnvVarNotPresent(t, sts, "TS_DEBUG_ACME_DIRECTORY_URL")
+			}
+		})
+	}
+}
+
+type leStagingTestCase struct {
+	name string
+	// ProxyClass set on ProxyGroup or Ingress resource.
+	proxyClassPerResource string
+	// Default ProxyClass.
+	defaultProxyClass    string
+	useLEStagingEndpoint bool
+}
+
+// Shared test cases for LE staging endpoint configuration for ProxyGroup and
+// non-HA Ingress.
+func testCasesForLEStagingTests(pcLEStaging, pcLEStagingFalse, pcOther *tsapi.ProxyClass) []leStagingTestCase {
+	return []leStagingTestCase{
+		{
+			name:                  "with_staging_proxyclass",
+			proxyClassPerResource: "le-staging",
+			useLEStagingEndpoint:  true,
+		},
+		{
+			name:                  "with_staging_proxyclass_false",
+			proxyClassPerResource: "le-staging-false",
+			useLEStagingEndpoint:  false,
+		},
+		{
+			name:                  "with_other_proxyclass",
+			proxyClassPerResource: "other",
+			useLEStagingEndpoint:  false,
+		},
+		{
+			name:                  "no_proxyclass",
+			proxyClassPerResource: "",
+			useLEStagingEndpoint:  false,
+		},
+		{
+			name:                  "with_default_staging_proxyclass",
+			proxyClassPerResource: "",
+			defaultProxyClass:     "le-staging",
+			useLEStagingEndpoint:  true,
+		},
+		{
+			name:                  "with_default_other_proxyclass",
+			proxyClassPerResource: "",
+			defaultProxyClass:     "other",
+			useLEStagingEndpoint:  false,
+		},
+		{
+			name:                  "with_default_staging_proxyclass_false",
+			proxyClassPerResource: "",
+			defaultProxyClass:     "le-staging-false",
+			useLEStagingEndpoint:  false,
+		},
+	}
 }

cmd/k8s-operator/sts.go

@@ -44,11 +44,9 @@ const (
 	// Labels that the operator sets on StatefulSets and Pods. If you add a
 	// new label here, do also add it to tailscaleManagedLabels var to
 	// ensure that it does not get overwritten by ProxyClass configuration.
-	LabelManaged         = "tailscale.com/managed"
 	LabelParentType      = "tailscale.com/parent-resource-type"
 	LabelParentName      = "tailscale.com/parent-resource"
 	LabelParentNamespace = "tailscale.com/parent-resource-ns"
-	labelSecretType      = "tailscale.com/secret-type" // "config" or "state".
 
 	// LabelProxyClass can be set by users on tailscale Ingresses and Services that define cluster ingress or
 	// cluster egress, to specify that configuration in this ProxyClass should be applied to resources created for
@@ -104,11 +102,13 @@ const (
 
 	envVarTSLocalAddrPort = "TS_LOCAL_ADDR_PORT"
 	defaultLocalAddrPort  = 9002 // metrics and health check port
+
+	letsEncryptStagingEndpoint = "https://acme-staging-v02.api.letsencrypt.org/directory"
 )
 
 var (
 	// tailscaleManagedLabels are label keys that tailscale operator sets on StatefulSets and Pods.
-	tailscaleManagedLabels = []string{LabelManaged, LabelParentType, LabelParentName, LabelParentNamespace, "app"}
+	tailscaleManagedLabels = []string{kubetypes.LabelManaged, LabelParentType, LabelParentName, LabelParentNamespace, "app"}
 	// tailscaleManagedAnnotations are annotation keys that tailscale operator sets on StatefulSets and Pods.
 	tailscaleManagedAnnotations = []string{podAnnotationLastSetClusterIP, podAnnotationLastSetTailnetTargetIP, podAnnotationLastSetTailnetTargetFQDN, podAnnotationLastSetConfigFileHash}
 )
@@ -785,6 +785,17 @@ func applyProxyClassToStatefulSet(pc *tsapi.ProxyClass, ss *appsv1.StatefulSet,
 			enableEndpoints(ss, metricsEnabled, debugEnabled)
 		}
 	}
+	if pc.Spec.UseLetsEncryptStagingEnvironment && (stsCfg.proxyType == proxyTypeIngressResource || stsCfg.proxyType == string(tsapi.ProxyGroupTypeIngress)) {
+		for i, c := range ss.Spec.Template.Spec.Containers {
+			if c.Name == "tailscale" {
+				ss.Spec.Template.Spec.Containers[i].Env = append(ss.Spec.Template.Spec.Containers[i].Env, corev1.EnvVar{
+					Name:  "TS_DEBUG_ACME_DIRECTORY_URL",
+					Value: letsEncryptStagingEndpoint,
+				})
+				break
+			}
+		}
+	}
 
 	if pc.Spec.StatefulSet == nil {
 		return ss

cmd/k8s-operator/sts_test.go

@@ -21,6 +21,7 @@ import (
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"sigs.k8s.io/yaml"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
+	"tailscale.com/kube/kubetypes"
 	"tailscale.com/types/ptr"
 )
 
@@ -156,8 +157,8 @@ func Test_applyProxyClassToStatefulSet(t *testing.T) {
 	// Set a couple additional fields so we can test that we don't
 	// mistakenly override those.
 	labels := map[string]string{
-		LabelManaged:    "true",
-		LabelParentName: "foo",
+		kubetypes.LabelManaged: "true",
+		LabelParentName:        "foo",
 	}
 	annots := map[string]string{
 		podAnnotationLastSetClusterIP: "1.2.3.4",
@@ -303,28 +304,28 @@ func Test_mergeStatefulSetLabelsOrAnnots(t *testing.T) {
 	}{
 		{
 			name:    "no custom labels specified and none present in current labels, return current labels",
-			current: map[string]string{LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
-			want:    map[string]string{LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
+			current: map[string]string{kubetypes.LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
+			want:    map[string]string{kubetypes.LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
 			managed: tailscaleManagedLabels,
 		},
 		{
 			name:    "no custom labels specified, but some present in current labels, return tailscale managed labels only from the current labels",
-			current: map[string]string{"foo": "bar", "something.io/foo": "bar", LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
-			want:    map[string]string{LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
+			current: map[string]string{"foo": "bar", "something.io/foo": "bar", kubetypes.LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
+			want:    map[string]string{kubetypes.LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
 			managed: tailscaleManagedLabels,
 		},
 		{
 			name:    "custom labels specified, current labels only contain tailscale managed labels, return a union of both",
-			current: map[string]string{LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
+			current: map[string]string{kubetypes.LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
 			custom:  map[string]string{"foo": "bar", "something.io/foo": "bar"},
-			want:    map[string]string{"foo": "bar", "something.io/foo": "bar", LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
+			want:    map[string]string{"foo": "bar", "something.io/foo": "bar", kubetypes.LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
 			managed: tailscaleManagedLabels,
 		},
 		{
 			name:    "custom labels specified, current labels contain tailscale managed labels and custom labels, some of which re not present in the new custom labels, return a union of managed labels and the desired custom labels",
-			current: map[string]string{"foo": "bar", "bar": "baz", "app": "1234", LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
+			current: map[string]string{"foo": "bar", "bar": "baz", "app": "1234", kubetypes.LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
 			custom:  map[string]string{"foo": "bar", "something.io/foo": "bar"},
-			want:    map[string]string{"foo": "bar", "something.io/foo": "bar", "app": "1234", LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
+			want:    map[string]string{"foo": "bar", "something.io/foo": "bar", "app": "1234", kubetypes.LabelManaged: "true", LabelParentName: "foo", LabelParentType: "svc", LabelParentNamespace: "foo"},
 			managed: tailscaleManagedLabels,
 		},
 		{

cmd/k8s-operator/svc.go

@@ -84,10 +84,10 @@ func childResourceLabels(name, ns, typ string) map[string]string {
 	// proxying. Instead, we have to do our own filtering and tracking with
 	// labels.
 	return map[string]string{
-		LabelManaged:         "true",
-		LabelParentName:      name,
-		LabelParentNamespace: ns,
-		LabelParentType:      typ,
+		kubetypes.LabelManaged: "true",
+		LabelParentName:        name,
+		LabelParentNamespace:   ns,
+		LabelParentType:        typ,
 	}
 }
 

cmd/k8s-operator/testutils_test.go

@@ -32,6 +32,7 @@ import (
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnstate"
 	tsapi "tailscale.com/k8s-operator/apis/v1alpha1"
+	"tailscale.com/kube/kubetypes"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/ptr"
 	"tailscale.com/util/mak"
@@ -563,10 +564,10 @@ func expectedSecret(t *testing.T, cl client.Client, opts configOpts) *corev1.Sec
 func findGenName(t *testing.T, client client.Client, ns, name, typ string) (full, noSuffix string) {
 	t.Helper()
 	labels := map[string]string{
-		LabelManaged:         "true",
-		LabelParentName:      name,
-		LabelParentNamespace: ns,
-		LabelParentType:      typ,
+		kubetypes.LabelManaged: "true",
+		LabelParentName:        name,
+		LabelParentNamespace:   ns,
+		LabelParentType:        typ,
 	}
 	s, err := getSingleObject[corev1.Secret](context.Background(), client, "operator-ns", labels)
 	if err != nil {

cmd/k8s-operator/tsrecorder.go

@@ -230,7 +230,7 @@ func (r *RecorderReconciler) maybeProvision(ctx context.Context, tsr *tsapi.Reco
 func (r *RecorderReconciler) maybeCleanup(ctx context.Context, tsr *tsapi.Recorder) (bool, error) {
 	logger := r.logger(tsr.Name)
 
-	id, _, ok, err := r.getNodeMetadata(ctx, tsr.Name)
+	prefs, ok, err := r.getDevicePrefs(ctx, tsr.Name)
 	if err != nil {
 		return false, err
 	}
@@ -243,6 +243,7 @@ func (r *RecorderReconciler) maybeCleanup(ctx context.Context, tsr *tsapi.Record
 		return true, nil
 	}
 
+	id := string(prefs.Config.NodeID)
 	logger.Debugf("deleting device %s from control", string(id))
 	if err := r.tsClient.DeleteDevice(ctx, string(id)); err != nil {
 		errResp := &tailscale.ErrResponse{}
@@ -327,34 +328,33 @@ func (r *RecorderReconciler) getStateSecret(ctx context.Context, tsrName string)
 	return secret, nil
 }
 
-func (r *RecorderReconciler) getNodeMetadata(ctx context.Context, tsrName string) (id tailcfg.StableNodeID, dnsName string, ok bool, err error) {
+func (r *RecorderReconciler) getDevicePrefs(ctx context.Context, tsrName string) (prefs prefs, ok bool, err error) {
 	secret, err := r.getStateSecret(ctx, tsrName)
 	if err != nil || secret == nil {
-		return "", "", false, err
+		return prefs, false, err
 	}
 
-	return getNodeMetadata(ctx, secret)
+	return getDevicePrefs(secret)
 }
 
-// getNodeMetadata returns 'ok == true' iff the node ID is found. The dnsName
+// getDevicePrefs returns 'ok == true' iff the node ID is found. The dnsName
 // is expected to always be non-empty if the node ID is, but not required.
-func getNodeMetadata(ctx context.Context, secret *corev1.Secret) (id tailcfg.StableNodeID, dnsName string, ok bool, err error) {
+func getDevicePrefs(secret *corev1.Secret) (prefs prefs, ok bool, err error) {
 	// TODO(tomhjp): Should maybe use ipn to parse the following info instead.
 	currentProfile, ok := secret.Data[currentProfileKey]
 	if !ok {
-		return "", "", false, nil
+		return prefs, false, nil
 	}
 	profileBytes, ok := secret.Data[string(currentProfile)]
 	if !ok {
-		return "", "", false, nil
+		return prefs, false, nil
 	}
-	var profile profile
-	if err := json.Unmarshal(profileBytes, &profile); err != nil {
-		return "", "", false, fmt.Errorf("failed to extract node profile info from state Secret %s: %w", secret.Name, err)
+	if err := json.Unmarshal(profileBytes, &prefs); err != nil {
+		return prefs, false, fmt.Errorf("failed to extract node profile info from state Secret %s: %w", secret.Name, err)
 	}
 
-	ok = profile.Config.NodeID != ""
-	return tailcfg.StableNodeID(profile.Config.NodeID), profile.Config.UserProfile.LoginName, ok, nil
+	ok = prefs.Config.NodeID != ""
+	return prefs, ok, nil
 }
 
 func (r *RecorderReconciler) getDeviceInfo(ctx context.Context, tsrName string) (d tsapi.RecorderTailnetDevice, ok bool, err error) {
@@ -367,14 +367,14 @@ func (r *RecorderReconciler) getDeviceInfo(ctx context.Context, tsrName string)
 }
 
 func getDeviceInfo(ctx context.Context, tsClient tsClient, secret *corev1.Secret) (d tsapi.RecorderTailnetDevice, ok bool, err error) {
-	nodeID, dnsName, ok, err := getNodeMetadata(ctx, secret)
+	prefs, ok, err := getDevicePrefs(secret)
 	if !ok || err != nil {
 		return tsapi.RecorderTailnetDevice{}, false, err
 	}
 
 	// TODO(tomhjp): The profile info doesn't include addresses, which is why we
 	// need the API. Should we instead update the profile to include addresses?
-	device, err := tsClient.Device(ctx, string(nodeID), nil)
+	device, err := tsClient.Device(ctx, string(prefs.Config.NodeID), nil)
 	if err != nil {
 		return tsapi.RecorderTailnetDevice{}, false, fmt.Errorf("failed to get device info from API: %w", err)
 	}
@@ -383,20 +383,25 @@ func getDeviceInfo(ctx context.Context, tsClient tsClient, secret *corev1.Secret
 		Hostname:   device.Hostname,
 		TailnetIPs: device.Addresses,
 	}
-	if dnsName != "" {
+	if dnsName := prefs.Config.UserProfile.LoginName; dnsName != "" {
 		d.URL = fmt.Sprintf("https://%s", dnsName)
 	}
 
 	return d, true, nil
 }
 
-type profile struct {
+// [prefs] is a subset of the ipn.Prefs struct used for extracting information
+// from the state Secret of Tailscale devices.
+type prefs struct {
 	Config struct {
-		NodeID      string `json:"NodeID"`
+		NodeID      tailcfg.StableNodeID `json:"NodeID"`
 		UserProfile struct {
+			// LoginName is the MagicDNS name of the device, e.g. foo.tail-scale.ts.net.
 			LoginName string `json:"LoginName"`
 		} `json:"UserProfile"`
 	} `json:"Config"`
+
+	AdvertiseServices []string `json:"AdvertiseServices"`
 }
 
 func markedForDeletion(obj metav1.Object) bool {

cmd/natc/natc.go

@@ -94,18 +94,24 @@ func main() {
 		}
 		ignoreDstTable.Insert(pfx, true)
 	}
-	var v4Prefixes []netip.Prefix
+	var (
+		v4Prefixes    []netip.Prefix
+		numV4DNSAddrs int
+	)
 	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)
+		numIPs := 1 << (32 - p.Bits())
+		numV4DNSAddrs += numIPs
 	}
 	if len(v4Prefixes) == 0 {
 		log.Fatalf("no v4 prefixes specified")
 	}
 	dnsAddr := v4Prefixes[0].Addr()
+	numV4DNSAddrs -= 1 // Subtract the dnsAddr allocated above.
 	ts := &tsnet.Server{
 		Hostname: *hostname,
 	}
@@ -153,12 +159,13 @@ func main() {
 	}
 
 	c := &connector{
-		ts:         ts,
-		lc:         lc,
-		dnsAddr:    dnsAddr,
-		v4Ranges:   v4Prefixes,
-		v6ULA:      ula(uint16(*siteID)),
-		ignoreDsts: ignoreDstTable,
+		ts:            ts,
+		lc:            lc,
+		dnsAddr:       dnsAddr,
+		v4Ranges:      v4Prefixes,
+		numV4DNSAddrs: numV4DNSAddrs,
+		v6ULA:         ula(uint16(*siteID)),
+		ignoreDsts:    ignoreDstTable,
 	}
 	c.run(ctx)
 }
@@ -177,6 +184,11 @@ type connector struct {
 	// v4Ranges is the list of IPv4 ranges to advertise and assign addresses from.
 	// These are masked prefixes.
 	v4Ranges []netip.Prefix
+
+	// numV4DNSAddrs is the total size of the IPv4 ranges in addresses, minus the
+	// dnsAddr allocation.
+	numV4DNSAddrs int
+
 	// v6ULA is the ULA prefix used by the app connector to assign IPv6 addresses.
 	v6ULA netip.Prefix
 
@@ -502,6 +514,7 @@ type perPeerState struct {
 	mu           sync.Mutex
 	domainToAddr map[string][]netip.Addr
 	addrToDomain *bart.Table[string]
+	numV4Allocs  int
 }
 
 // domainForIP returns the domain name assigned to the given IP address and
@@ -547,17 +560,25 @@ func (ps *perPeerState) isIPUsedLocked(ip netip.Addr) bool {
 
 // unusedIPv4Locked returns an unused IPv4 address from the available ranges.
 func (ps *perPeerState) unusedIPv4Locked() netip.Addr {
+	// All addresses have been allocated.
+	if ps.numV4Allocs >= ps.c.numV4DNSAddrs {
+		return netip.Addr{}
+	}
+
 	// TODO: skip ranges that have been exhausted
-	for _, r := range ps.c.v4Ranges {
-		ip := randV4(r)
-		for r.Contains(ip) {
+	// TODO: implement a much more efficient algorithm for finding unused IPs,
+	// this is fairly crazy.
+	for {
+		for _, r := range ps.c.v4Ranges {
+			ip := randV4(r)
+			if !r.Contains(ip) {
+				panic("error: randV4 returned invalid address")
+			}
 			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.
@@ -583,6 +604,7 @@ func (ps *perPeerState) assignAddrsLocked(domain string) []netip.Addr {
 	if !v4.IsValid() {
 		return nil
 	}
+	ps.numV4Allocs++
 	as16 := ps.c.v6ULA.Addr().As16()
 	as4 := v4.As4()
 	copy(as16[12:], as4[:])

cmd/natc/natc_test.go

@@ -0,0 +1,429 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package main
+
+import (
+	"errors"
+	"fmt"
+	"net/netip"
+	"slices"
+	"testing"
+
+	"github.com/gaissmai/bart"
+	"github.com/google/go-cmp/cmp"
+	"golang.org/x/net/dns/dnsmessage"
+	"tailscale.com/tailcfg"
+)
+
+func prefixEqual(a, b netip.Prefix) bool {
+	return a.Bits() == b.Bits() && a.Addr() == b.Addr()
+}
+
+func TestULA(t *testing.T) {
+	tests := []struct {
+		name     string
+		siteID   uint16
+		expected string
+	}{
+		{"zero", 0, "fd7a:115c:a1e0:a99c:0000::/80"},
+		{"one", 1, "fd7a:115c:a1e0:a99c:0001::/80"},
+		{"max", 65535, "fd7a:115c:a1e0:a99c:ffff::/80"},
+		{"random", 12345, "fd7a:115c:a1e0:a99c:3039::/80"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := ula(tc.siteID)
+			expected := netip.MustParsePrefix(tc.expected)
+			if !prefixEqual(got, expected) {
+				t.Errorf("ula(%d) = %s; want %s", tc.siteID, got, expected)
+			}
+		})
+	}
+}
+
+func TestRandV4(t *testing.T) {
+	pfx := netip.MustParsePrefix("100.64.1.0/24")
+
+	for i := 0; i < 512; i++ {
+		ip := randV4(pfx)
+		if !pfx.Contains(ip) {
+			t.Errorf("randV4(%s) = %s; not contained in prefix", pfx, ip)
+		}
+	}
+}
+
+func TestDNSResponse(t *testing.T) {
+	tests := []struct {
+		name        string
+		questions   []dnsmessage.Question
+		addrs       []netip.Addr
+		wantEmpty   bool
+		wantAnswers []struct {
+			name  string
+			qType dnsmessage.Type
+			addr  netip.Addr
+		}
+	}{
+		{
+			name:        "empty_request",
+			questions:   []dnsmessage.Question{},
+			addrs:       []netip.Addr{},
+			wantEmpty:   false,
+			wantAnswers: nil,
+		},
+		{
+			name: "a_record",
+			questions: []dnsmessage.Question{
+				{
+					Name:  dnsmessage.MustNewName("example.com."),
+					Type:  dnsmessage.TypeA,
+					Class: dnsmessage.ClassINET,
+				},
+			},
+			addrs: []netip.Addr{netip.MustParseAddr("100.64.1.5")},
+			wantAnswers: []struct {
+				name  string
+				qType dnsmessage.Type
+				addr  netip.Addr
+			}{
+				{
+					name:  "example.com.",
+					qType: dnsmessage.TypeA,
+					addr:  netip.MustParseAddr("100.64.1.5"),
+				},
+			},
+		},
+		{
+			name: "aaaa_record",
+			questions: []dnsmessage.Question{
+				{
+					Name:  dnsmessage.MustNewName("example.com."),
+					Type:  dnsmessage.TypeAAAA,
+					Class: dnsmessage.ClassINET,
+				},
+			},
+			addrs: []netip.Addr{netip.MustParseAddr("fd7a:115c:a1e0:a99c:0001:0505:0505:0505")},
+			wantAnswers: []struct {
+				name  string
+				qType dnsmessage.Type
+				addr  netip.Addr
+			}{
+				{
+					name:  "example.com.",
+					qType: dnsmessage.TypeAAAA,
+					addr:  netip.MustParseAddr("fd7a:115c:a1e0:a99c:0001:0505:0505:0505"),
+				},
+			},
+		},
+		{
+			name: "soa_record",
+			questions: []dnsmessage.Question{
+				{
+					Name:  dnsmessage.MustNewName("example.com."),
+					Type:  dnsmessage.TypeSOA,
+					Class: dnsmessage.ClassINET,
+				},
+			},
+			addrs:       []netip.Addr{},
+			wantAnswers: nil,
+		},
+		{
+			name: "ns_record",
+			questions: []dnsmessage.Question{
+				{
+					Name:  dnsmessage.MustNewName("example.com."),
+					Type:  dnsmessage.TypeNS,
+					Class: dnsmessage.ClassINET,
+				},
+			},
+			addrs:       []netip.Addr{},
+			wantAnswers: nil,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			req := &dnsmessage.Message{
+				Header: dnsmessage.Header{
+					ID: 1234,
+				},
+				Questions: tc.questions,
+			}
+
+			resp, err := dnsResponse(req, tc.addrs)
+			if err != nil {
+				t.Fatalf("dnsResponse() error = %v", err)
+			}
+
+			if tc.wantEmpty && len(resp) != 0 {
+				t.Errorf("dnsResponse() returned non-empty response when expected empty")
+			}
+
+			if !tc.wantEmpty && len(resp) == 0 {
+				t.Errorf("dnsResponse() returned empty response when expected non-empty")
+			}
+
+			if len(resp) > 0 {
+				var msg dnsmessage.Message
+				err = msg.Unpack(resp)
+				if err != nil {
+					t.Fatalf("Failed to unpack response: %v", err)
+				}
+
+				if !msg.Header.Response {
+					t.Errorf("Response header is not set")
+				}
+
+				if msg.Header.ID != req.Header.ID {
+					t.Errorf("Response ID = %d, want %d", msg.Header.ID, req.Header.ID)
+				}
+
+				if len(tc.wantAnswers) > 0 {
+					if len(msg.Answers) != len(tc.wantAnswers) {
+						t.Errorf("got %d answers, want %d", len(msg.Answers), len(tc.wantAnswers))
+					} else {
+						for i, want := range tc.wantAnswers {
+							ans := msg.Answers[i]
+
+							gotName := ans.Header.Name.String()
+							if gotName != want.name {
+								t.Errorf("answer[%d] name = %s, want %s", i, gotName, want.name)
+							}
+
+							if ans.Header.Type != want.qType {
+								t.Errorf("answer[%d] type = %v, want %v", i, ans.Header.Type, want.qType)
+							}
+
+							var gotIP netip.Addr
+							switch want.qType {
+							case dnsmessage.TypeA:
+								if ans.Body.(*dnsmessage.AResource) == nil {
+									t.Errorf("answer[%d] not an A record", i)
+									continue
+								}
+								resource := ans.Body.(*dnsmessage.AResource)
+								gotIP = netip.AddrFrom4([4]byte(resource.A))
+							case dnsmessage.TypeAAAA:
+								if ans.Body.(*dnsmessage.AAAAResource) == nil {
+									t.Errorf("answer[%d] not an AAAA record", i)
+									continue
+								}
+								resource := ans.Body.(*dnsmessage.AAAAResource)
+								gotIP = netip.AddrFrom16([16]byte(resource.AAAA))
+							}
+
+							if gotIP != want.addr {
+								t.Errorf("answer[%d] IP = %s, want %s", i, gotIP, want.addr)
+							}
+						}
+					}
+				}
+			}
+		})
+	}
+}
+
+func TestPerPeerState(t *testing.T) {
+	c := &connector{
+		v4Ranges:      []netip.Prefix{netip.MustParsePrefix("100.64.1.0/24")},
+		v6ULA:         netip.MustParsePrefix("fd7a:115c:a1e0:a99c:0001::/80"),
+		dnsAddr:       netip.MustParseAddr("100.64.1.0"),
+		numV4DNSAddrs: (1<<(32-24) - 1),
+	}
+
+	ps := &perPeerState{c: c}
+
+	addrs, err := ps.ipForDomain("example.com")
+	if err != nil {
+		t.Fatalf("ipForDomain() error = %v", err)
+	}
+
+	if len(addrs) != 2 {
+		t.Fatalf("ipForDomain() returned %d addresses, want 2", len(addrs))
+	}
+
+	v4 := addrs[0]
+	v6 := addrs[1]
+
+	if !v4.Is4() {
+		t.Errorf("First address is not IPv4: %s", v4)
+	}
+
+	if !v6.Is6() {
+		t.Errorf("Second address is not IPv6: %s", v6)
+	}
+
+	if !c.v4Ranges[0].Contains(v4) {
+		t.Errorf("IPv4 address %s not in range %s", v4, c.v4Ranges[0])
+	}
+
+	domain, ok := ps.domainForIP(v4)
+	if !ok {
+		t.Errorf("domainForIP(%s) not found", v4)
+	} else if domain != "example.com" {
+		t.Errorf("domainForIP(%s) = %s, want %s", v4, domain, "example.com")
+	}
+
+	domain, ok = ps.domainForIP(v6)
+	if !ok {
+		t.Errorf("domainForIP(%s) not found", v6)
+	} else if domain != "example.com" {
+		t.Errorf("domainForIP(%s) = %s, want %s", v6, domain, "example.com")
+	}
+
+	addrs2, err := ps.ipForDomain("example.com")
+	if err != nil {
+		t.Fatalf("ipForDomain() second call error = %v", err)
+	}
+
+	if !slices.Equal(addrs, addrs2) {
+		t.Errorf("ipForDomain() second call = %v, want %v", addrs2, addrs)
+	}
+}
+
+func TestIgnoreDestination(t *testing.T) {
+	ignoreDstTable := &bart.Table[bool]{}
+	ignoreDstTable.Insert(netip.MustParsePrefix("192.168.1.0/24"), true)
+	ignoreDstTable.Insert(netip.MustParsePrefix("10.0.0.0/8"), true)
+
+	c := &connector{
+		ignoreDsts: ignoreDstTable,
+	}
+
+	tests := []struct {
+		name     string
+		addrs    []netip.Addr
+		expected bool
+	}{
+		{
+			name:     "no_match",
+			addrs:    []netip.Addr{netip.MustParseAddr("8.8.8.8"), netip.MustParseAddr("1.1.1.1")},
+			expected: false,
+		},
+		{
+			name:     "one_match",
+			addrs:    []netip.Addr{netip.MustParseAddr("8.8.8.8"), netip.MustParseAddr("192.168.1.5")},
+			expected: true,
+		},
+		{
+			name:     "all_match",
+			addrs:    []netip.Addr{netip.MustParseAddr("10.0.0.1"), netip.MustParseAddr("192.168.1.5")},
+			expected: true,
+		},
+		{
+			name:     "empty_addrs",
+			addrs:    []netip.Addr{},
+			expected: false,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := c.ignoreDestination(tc.addrs)
+			if got != tc.expected {
+				t.Errorf("ignoreDestination(%v) = %v, want %v", tc.addrs, got, tc.expected)
+			}
+		})
+	}
+}
+
+func TestConnectorGenerateDNSResponse(t *testing.T) {
+	c := &connector{
+		v4Ranges:      []netip.Prefix{netip.MustParsePrefix("100.64.1.0/24")},
+		v6ULA:         netip.MustParsePrefix("fd7a:115c:a1e0:a99c:0001::/80"),
+		dnsAddr:       netip.MustParseAddr("100.64.1.0"),
+		numV4DNSAddrs: (1<<(32-24) - 1),
+	}
+
+	req := &dnsmessage.Message{
+		Header: dnsmessage.Header{ID: 1234},
+		Questions: []dnsmessage.Question{
+			{
+				Name:  dnsmessage.MustNewName("example.com."),
+				Type:  dnsmessage.TypeA,
+				Class: dnsmessage.ClassINET,
+			},
+		},
+	}
+
+	nodeID := tailcfg.NodeID(12345)
+
+	resp1, err := c.generateDNSResponse(req, nodeID)
+	if err != nil {
+		t.Fatalf("generateDNSResponse() error = %v", err)
+	}
+	if len(resp1) == 0 {
+		t.Fatalf("generateDNSResponse() returned empty response")
+	}
+
+	resp2, err := c.generateDNSResponse(req, nodeID)
+	if err != nil {
+		t.Fatalf("generateDNSResponse() second call error = %v", err)
+	}
+
+	if !cmp.Equal(resp1, resp2) {
+		t.Errorf("generateDNSResponse() responses differ between calls")
+	}
+}
+
+func TestIPPoolExhaustion(t *testing.T) {
+	smallPrefix := netip.MustParsePrefix("100.64.1.0/30") // Only 4 IPs: .0, .1, .2, .3
+	c := &connector{
+		v6ULA:         netip.MustParsePrefix("fd7a:115c:a1e0:a99c:0001::/80"),
+		v4Ranges:      []netip.Prefix{smallPrefix},
+		dnsAddr:       netip.MustParseAddr("100.64.1.0"),
+		numV4DNSAddrs: 3,
+	}
+
+	ps := &perPeerState{c: c}
+
+	assignedIPs := make(map[netip.Addr]string)
+
+	domains := []string{"a.example.com", "b.example.com", "c.example.com", "d.example.com"}
+
+	var errs []error
+
+	for i := 0; i < 5; i++ {
+		for _, domain := range domains {
+			addrs, err := ps.ipForDomain(domain)
+			if err != nil {
+				errs = append(errs, fmt.Errorf("failed to get IP for domain %q: %w", domain, err))
+				continue
+			}
+
+			for _, addr := range addrs {
+				if d, ok := assignedIPs[addr]; ok {
+					if d != domain {
+						t.Errorf("IP %s reused for domain %q, previously assigned to %q", addr, domain, d)
+					}
+				} else {
+					assignedIPs[addr] = domain
+				}
+			}
+		}
+	}
+
+	for addr, domain := range assignedIPs {
+		if addr.Is4() && !smallPrefix.Contains(addr) {
+			t.Errorf("IP %s for domain %q not in expected range %s", addr, domain, smallPrefix)
+		}
+		if addr.Is6() && !c.v6ULA.Contains(addr) {
+			t.Errorf("IP %s for domain %q not in expected range %s", addr, domain, c.v6ULA)
+		}
+		if addr == c.dnsAddr {
+			t.Errorf("IP %s for domain %q is the reserved DNS address", addr, domain)
+		}
+	}
+
+	// expect one error for each iteration with the 4th domain
+	if len(errs) != 5 {
+		t.Errorf("Expected 5 errors, got %d: %v", len(errs), errs)
+	}
+	for _, err := range errs {
+		if !errors.Is(err, ErrNoIPsAvailable) {
+			t.Errorf("generateDNSResponse() error = %v, want ErrNoIPsAvailable", err)
+		}
+	}
+}

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

@@ -19,8 +19,25 @@
 //	header_property = username
 //	auto_sign_up = true
 //	whitelist = 127.0.0.1
-//	headers = Name:X-WEBAUTH-NAME
+//	headers = Email:X-Webauth-User, Name:X-Webauth-Name, Role:X-Webauth-Role
 //	enable_login_token = true
+//
+// You can use grants in Tailscale ACL to give users different roles in Grafana.
+// For example, to give group:eng the Editor role, add the following to your ACLs:
+//
+//	 "grants": [
+//			{
+//				"src": ["group:eng"],
+//				"dst": ["tag:grafana"],
+//				"app": {
+//					"tailscale.com/cap/proxy-to-grafana": [{
+//						"role": "editor",
+//					}],
+//				},
+//			},
+//	 ],
+//
+// If multiple roles are specified, the most permissive role is used.
 package main
 
 import (
@@ -49,6 +66,57 @@ var (
 	loginServer  = flag.String("login-server", "", "URL to alternative control server. If empty, the default Tailscale control is used.")
 )
 
+// aclCap is the Tailscale ACL capability used to configure proxy-to-grafana.
+const aclCap tailcfg.PeerCapability = "tailscale.com/cap/proxy-to-grafana"
+
+// aclGrant is an access control rule that assigns Grafana permissions
+// while provisioning a user.
+type aclGrant struct {
+	// Role is one of: "viewer", "editor", "admin".
+	Role string `json:"role"`
+}
+
+// grafanaRole defines possible Grafana roles.
+type grafanaRole int
+
+const (
+	// Roles are ordered by their permissions, with the least permissive role first.
+	// If a user has multiple roles, the most permissive role is used.
+	ViewerRole grafanaRole = iota
+	EditorRole
+	AdminRole
+)
+
+// String returns the string representation of a grafanaRole.
+// It is used as a header value in the HTTP request to Grafana.
+func (r grafanaRole) String() string {
+	switch r {
+	case ViewerRole:
+		return "Viewer"
+	case EditorRole:
+		return "Editor"
+	case AdminRole:
+		return "Admin"
+	default:
+		// A safe default.
+		return "Viewer"
+	}
+}
+
+// roleFromString converts a string to a grafanaRole.
+// It is used to parse the role from the ACL grant.
+func roleFromString(s string) (grafanaRole, error) {
+	switch strings.ToLower(s) {
+	case "viewer":
+		return ViewerRole, nil
+	case "editor":
+		return EditorRole, nil
+	case "admin":
+		return AdminRole, nil
+	}
+	return ViewerRole, fmt.Errorf("unknown role: %q", s)
+}
+
 func main() {
 	flag.Parse()
 	if *hostname == "" || strings.Contains(*hostname, ".") {
@@ -134,7 +202,15 @@ func modifyRequest(req *http.Request, localClient *local.Client) {
 		return
 	}
 
-	user, err := getTailscaleUser(req.Context(), localClient, req.RemoteAddr)
+	// Delete any existing X-Webauth-* headers to prevent possible spoofing
+	// if getting Tailnet identity fails.
+	for h := range req.Header {
+		if strings.HasPrefix(h, "X-Webauth-") {
+			req.Header.Del(h)
+		}
+	}
+
+	user, role, err := getTailscaleIdentity(req.Context(), localClient, req.RemoteAddr)
 	if err != nil {
 		log.Printf("error getting Tailscale user: %v", err)
 		return
@@ -142,19 +218,33 @@ func modifyRequest(req *http.Request, localClient *local.Client) {
 
 	req.Header.Set("X-Webauth-User", user.LoginName)
 	req.Header.Set("X-Webauth-Name", user.DisplayName)
+	req.Header.Set("X-Webauth-Role", role.String())
 }
 
-func getTailscaleUser(ctx context.Context, localClient *local.Client, ipPort string) (*tailcfg.UserProfile, error) {
+func getTailscaleIdentity(ctx context.Context, localClient *local.Client, ipPort string) (*tailcfg.UserProfile, grafanaRole, error) {
 	whois, err := localClient.WhoIs(ctx, ipPort)
 	if err != nil {
-		return nil, fmt.Errorf("failed to identify remote host: %w", err)
+		return nil, ViewerRole, fmt.Errorf("failed to identify remote host: %w", err)
 	}
 	if whois.Node.IsTagged() {
-		return nil, fmt.Errorf("tagged nodes are not users")
+		return nil, ViewerRole, fmt.Errorf("tagged nodes are not users")
 	}
 	if whois.UserProfile == nil || whois.UserProfile.LoginName == "" {
-		return nil, fmt.Errorf("failed to identify remote user")
+		return nil, ViewerRole, fmt.Errorf("failed to identify remote user")
 	}
 
-	return whois.UserProfile, nil
+	role := ViewerRole
+	grants, err := tailcfg.UnmarshalCapJSON[aclGrant](whois.CapMap, aclCap)
+	if err != nil {
+		return nil, ViewerRole, fmt.Errorf("failed to unmarshal ACL grants: %w", err)
+	}
+	for _, g := range grants {
+		r, err := roleFromString(g.Role)
+		if err != nil {
+			return nil, ViewerRole, fmt.Errorf("failed to parse role: %w", err)
+		}
+		role = max(role, r)
+	}
+
+	return whois.UserProfile, role, nil
 }

cmd/stund/depaware.txt

@@ -49,6 +49,7 @@ tailscale.com/cmd/stund dependencies: (generated by github.com/tailscale/depawar
         google.golang.org/protobuf/types/known/timestamppb           from github.com/prometheus/client_golang/prometheus+
         tailscale.com                                                from tailscale.com/version
         tailscale.com/envknob                                        from tailscale.com/tsweb+
+        tailscale.com/feature                                        from tailscale.com/tsweb
         tailscale.com/kube/kubetypes                                 from tailscale.com/envknob
         tailscale.com/metrics                                        from tailscale.com/net/stunserver+
         tailscale.com/net/netaddr                                    from tailscale.com/net/tsaddr
@@ -57,8 +58,8 @@ tailscale.com/cmd/stund dependencies: (generated by github.com/tailscale/depawar
         tailscale.com/net/tsaddr                                     from tailscale.com/tsweb
         tailscale.com/syncs                                          from tailscale.com/metrics
         tailscale.com/tailcfg                                        from tailscale.com/version
-        tailscale.com/tsweb                                          from tailscale.com/cmd/stund
-        tailscale.com/tsweb/promvarz                                 from tailscale.com/tsweb
+        tailscale.com/tsweb                                          from tailscale.com/cmd/stund+
+        tailscale.com/tsweb/promvarz                                 from tailscale.com/cmd/stund
         tailscale.com/tsweb/varz                                     from tailscale.com/tsweb+
         tailscale.com/types/dnstype                                  from tailscale.com/tailcfg
         tailscale.com/types/ipproto                                  from tailscale.com/tailcfg
@@ -194,7 +195,7 @@ tailscale.com/cmd/stund dependencies: (generated by github.com/tailscale/depawar
         hash/maphash                                                 from go4.org/mem
         html                                                         from net/http/pprof+
         internal/abi                                                 from crypto/x509/internal/macos+
-        internal/asan                                                from syscall+
+        internal/asan                                                from internal/runtime/maps+
         internal/bisect                                              from internal/godebug
         internal/bytealg                                             from bytes+
         internal/byteorder                                           from crypto/cipher+
@@ -204,12 +205,12 @@ tailscale.com/cmd/stund dependencies: (generated by github.com/tailscale/depawar
         internal/filepathlite                                        from os+
         internal/fmtsort                                             from fmt
         internal/goarch                                              from crypto/internal/fips140deps/cpu+
-        internal/godebug                                             from crypto/tls+
+        internal/godebug                                             from crypto/internal/fips140deps/godebug+
         internal/godebugs                                            from internal/godebug+
-        internal/goexperiment                                        from runtime+
+        internal/goexperiment                                        from hash/maphash+
         internal/goos                                                from crypto/x509+
         internal/itoa                                                from internal/poll+
-        internal/msan                                                from syscall+
+        internal/msan                                                from internal/runtime/maps+
         internal/nettrace                                            from net+
         internal/oserror                                             from io/fs+
         internal/poll                                                from net+

cmd/stund/stund.go

@@ -15,6 +15,9 @@ import (
 
 	"tailscale.com/net/stunserver"
 	"tailscale.com/tsweb"
+
+	// Support for prometheus varz in tsweb
+	_ "tailscale.com/tsweb/promvarz"
 )
 
 var (

cmd/tailscale/cli/debug.go

@@ -136,6 +136,17 @@ func debugCmd() *ffcli.Command {
 				Exec:       runLocalCreds,
 				ShortHelp:  "Print how to access Tailscale LocalAPI",
 			},
+			{
+				Name:       "localapi",
+				ShortUsage: "tailscale debug localapi [<method>] <path> [<body| \"-\">]",
+				Exec:       runLocalAPI,
+				ShortHelp:  "Call a LocalAPI method directly",
+				FlagSet: (func() *flag.FlagSet {
+					fs := newFlagSet("localapi")
+					fs.BoolVar(&localAPIFlags.verbose, "v", false, "verbose; dump HTTP headers")
+					return fs
+				})(),
+			},
 			{
 				Name:       "restun",
 				ShortUsage: "tailscale debug restun",
@@ -451,6 +462,81 @@ func runLocalCreds(ctx context.Context, args []string) error {
 	return nil
 }
 
+func looksLikeHTTPMethod(s string) bool {
+	if len(s) > len("OPTIONS") {
+		return false
+	}
+	for _, r := range s {
+		if r < 'A' || r > 'Z' {
+			return false
+		}
+	}
+	return true
+}
+
+var localAPIFlags struct {
+	verbose bool
+}
+
+func runLocalAPI(ctx context.Context, args []string) error {
+	if len(args) == 0 {
+		return errors.New("expected at least one argument")
+	}
+	method := "GET"
+	if looksLikeHTTPMethod(args[0]) {
+		method = args[0]
+		args = args[1:]
+		if len(args) == 0 {
+			return errors.New("expected at least one argument after method")
+		}
+	}
+	path := args[0]
+	if !strings.HasPrefix(path, "/localapi/") {
+		if !strings.Contains(path, "/") {
+			path = "/localapi/v0/" + path
+		} else {
+			path = "/localapi/" + path
+		}
+	}
+
+	var body io.Reader
+	if len(args) > 1 {
+		if args[1] == "-" {
+			fmt.Fprintf(Stderr, "# reading request body from stdin...\n")
+			all, err := io.ReadAll(os.Stdin)
+			if err != nil {
+				return fmt.Errorf("reading Stdin: %q", err)
+			}
+			body = bytes.NewReader(all)
+		} else {
+			body = strings.NewReader(args[1])
+		}
+	}
+	req, err := http.NewRequest(method, "http://local-tailscaled.sock"+path, body)
+	if err != nil {
+		return err
+	}
+	fmt.Fprintf(Stderr, "# doing request %s %s\n", method, path)
+
+	res, err := localClient.DoLocalRequest(req)
+	if err != nil {
+		return err
+	}
+	is2xx := res.StatusCode >= 200 && res.StatusCode <= 299
+	if localAPIFlags.verbose {
+		res.Write(Stdout)
+	} else {
+		if !is2xx {
+			fmt.Fprintf(Stderr, "# Response status %s\n", res.Status)
+		}
+		io.Copy(Stdout, res.Body)
+	}
+	if is2xx {
+		return nil
+	}
+	return errors.New(res.Status)
+}
+
 type localClientRoundTripper struct{}
 
 func (localClientRoundTripper) RoundTrip(req *http.Request) (*http.Response, error) {

cmd/tailscale/depaware.txt

@@ -333,7 +333,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         image/color                                                  from github.com/skip2/go-qrcode+
         image/png                                                    from github.com/skip2/go-qrcode
         internal/abi                                                 from crypto/x509/internal/macos+
-        internal/asan                                                from syscall+
+        internal/asan                                                from internal/runtime/maps+
         internal/bisect                                              from internal/godebug
         internal/bytealg                                             from bytes+
         internal/byteorder                                           from crypto/cipher+
@@ -345,10 +345,10 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         internal/goarch                                              from crypto/internal/fips140deps/cpu+
         internal/godebug                                             from archive/tar+
         internal/godebugs                                            from internal/godebug+
-        internal/goexperiment                                        from runtime+
+        internal/goexperiment                                        from hash/maphash+
         internal/goos                                                from crypto/x509+
         internal/itoa                                                from internal/poll+
-        internal/msan                                                from syscall+
+        internal/msan                                                from internal/runtime/maps+
         internal/nettrace                                            from net+
         internal/oserror                                             from io/fs+
         internal/poll                                                from net+

cmd/tailscaled/depaware.txt

@@ -286,7 +286,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/ipn/store/mem                                  from tailscale.com/ipn/ipnlocal+
    L    tailscale.com/kube/kubeapi                                   from tailscale.com/ipn/store/kubestore+
    L    tailscale.com/kube/kubeclient                                from tailscale.com/ipn/store/kubestore
-        tailscale.com/kube/kubetypes                                 from tailscale.com/envknob
+        tailscale.com/kube/kubetypes                                 from tailscale.com/envknob+
         tailscale.com/licenses                                       from tailscale.com/client/web
         tailscale.com/log/filelogger                                 from tailscale.com/logpolicy
         tailscale.com/log/sockstatlog                                from tailscale.com/ipn/ipnlocal
@@ -589,7 +589,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         html                                                         from html/template+
         html/template                                                from github.com/gorilla/csrf
         internal/abi                                                 from crypto/x509/internal/macos+
-        internal/asan                                                from syscall+
+        internal/asan                                                from internal/runtime/maps+
         internal/bisect                                              from internal/godebug
         internal/bytealg                                             from bytes+
         internal/byteorder                                           from crypto/cipher+
@@ -601,10 +601,10 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         internal/goarch                                              from crypto/internal/fips140deps/cpu+
         internal/godebug                                             from archive/tar+
         internal/godebugs                                            from internal/godebug+
-        internal/goexperiment                                        from runtime+
+        internal/goexperiment                                        from hash/maphash+
         internal/goos                                                from crypto/x509+
         internal/itoa                                                from internal/poll+
-        internal/msan                                                from syscall+
+        internal/msan                                                from internal/runtime/maps+
         internal/nettrace                                            from net+
         internal/oserror                                             from io/fs+
         internal/poll                                                from net+

cmd/xdpderper/xdpderper.go

@@ -18,6 +18,9 @@ import (
 	"tailscale.com/derp/xdp"
 	"tailscale.com/net/netutil"
 	"tailscale.com/tsweb"
+
+	// Support for prometheus varz in tsweb
+	_ "tailscale.com/tsweb/promvarz"
 )
 
 var (

control/controlhttp/client.go

@@ -96,6 +96,9 @@ func (a *Dialer) httpsFallbackDelay() time.Duration {
 var _ = envknob.RegisterBool("TS_USE_CONTROL_DIAL_PLAN") // to record at init time whether it's in use
 
 func (a *Dialer) dial(ctx context.Context) (*ClientConn, error) {
+
+	a.logPort80Failure.Store(true)
+
 	// If we don't have a dial plan, just fall back to dialing the single
 	// host we know about.
 	useDialPlan := envknob.BoolDefaultTrue("TS_USE_CONTROL_DIAL_PLAN")
@@ -278,7 +281,9 @@ func (d *Dialer) forceNoise443() bool {
 		// This heuristic works around networks where port 80 is MITMed and
 		// appears to work for a bit post-Upgrade but then gets closed,
 		// such as seen in https://github.com/tailscale/tailscale/issues/13597.
-		d.logf("controlhttp: forcing port 443 dial due to recent noise dial")
+		if d.logPort80Failure.CompareAndSwap(true, false) {
+			d.logf("controlhttp: forcing port 443 dial due to recent noise dial")
+		}
 		return true
 	}
 

control/controlhttp/constants.go

@@ -6,6 +6,7 @@ package controlhttp
 import (
 	"net/http"
 	"net/url"
+	"sync/atomic"
 	"time"
 
 	"tailscale.com/health"
@@ -90,6 +91,11 @@ type Dialer struct {
 
 	proxyFunc func(*http.Request) (*url.URL, error) // or nil
 
+	// logPort80Failure is whether we should log about port 80 interceptions
+	// and forcing a port 443 dial. We do this only once per "dial" method
+	// which can result in many concurrent racing dialHost calls.
+	logPort80Failure atomic.Bool
+
 	// For tests only
 	drainFinished        chan struct{}
 	omitCertErrorLogging bool

envknob/envknob.go

@@ -429,10 +429,16 @@ func App() string {
 // is a shared cert available.
 func IsCertShareReadOnlyMode() bool {
 	m := String("TS_CERT_SHARE_MODE")
-	return m == modeRO
+	return m == "ro"
 }
 
-const modeRO = "ro"
+// IsCertShareReadWriteMode returns true if this instance is the replica
+// responsible for issuing and renewing TLS certs in an HA setup with certs
+// shared between multiple replicas.
+func IsCertShareReadWriteMode() bool {
+	m := String("TS_CERT_SHARE_MODE")
+	return m == "rw"
+}
 
 // CrashOnUnexpected reports whether the Tailscale client should panic
 // on unexpected conditions. If TS_DEBUG_CRASH_ON_UNEXPECTED is set, that's

internal/client/tailscale/vip_service.go

@@ -27,6 +27,8 @@ type VIPService struct {
 	Addrs []string `json:"addrs,omitempty"`
 	// Comment is an optional text string for display in the admin panel.
 	Comment string `json:"comment,omitempty"`
+	// Annotations are optional key-value pairs that can be used to store arbitrary metadata.
+	Annotations map[string]string `json:"annotations,omitempty"`
 	// Ports are the ports of a VIPService that will be configured via Tailscale serve config.
 	// If set, any node wishing to advertise this VIPService must have this port configured via Tailscale serve.
 	Ports []string `json:"ports,omitempty"`

ipn/conf.go

@@ -145,9 +145,15 @@ func (c *ConfigVAlpha) ToPrefs() (MaskedPrefs, error) {
 		mp.AppConnector = *c.AppConnector
 		mp.AppConnectorSet = true
 	}
+	// Configfile should be the source of truth for whether this node
+	// advertises any services.  We need to ensure that each reload updates
+	// currently advertised services as else the transition from 'some
+	// services are advertised' to 'advertised services are empty/unset in
+	// conffile' would have no effect (especially given that an empty
+	// service slice would be omitted from the JSON config).
+	mp.AdvertiseServicesSet = true
 	if c.AdvertiseServices != nil {
 		mp.AdvertiseServices = c.AdvertiseServices
-		mp.AdvertiseServicesSet = true
 	}
 	return mp, nil
 }

ipn/ipnlocal/local.go

@@ -958,7 +958,9 @@ func (b *LocalBackend) linkChange(delta *netmon.ChangeDelta) {
 
 	if peerAPIListenAsync && b.netMap != nil && b.state == ipn.Running {
 		want := b.netMap.GetAddresses().Len()
-		if len(b.peerAPIListeners) < want {
+		have := len(b.peerAPIListeners)
+		b.logf("[v1] linkChange: have %d peerAPIListeners, want %d", have, want)
+		if have < want {
 			b.logf("linkChange: peerAPIListeners too low; trying again")
 			b.goTracker.Go(b.initPeerAPIListener)
 		}
@@ -2380,12 +2382,10 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 	}
 	b.applyPrefsToHostinfoLocked(hostinfo, prefs)
 
-	b.setNetMapLocked(nil)
 	persistv := prefs.Persist().AsStruct()
 	if persistv == nil {
 		persistv = new(persist.Persist)
 	}
-	b.updateFilterLocked(nil, ipn.PrefsView{})
 
 	if b.portpoll != nil {
 		b.portpollOnce.Do(func() {
@@ -2404,11 +2404,9 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 	}
 
 	var auditLogShutdown func()
-	// Audit logging is only available if the client has set up a proper persistent
-	// store for the logs in sys.
 	store, ok := b.sys.AuditLogStore.GetOK()
 	if !ok {
-		b.logf("auditlog: [unexpected] no persistent audit log storage configured.  using memory store.")
+		// Use memory store by default if no explicit store is provided.
 		store = auditlog.NewLogStore(&memstore.Store{})
 	}
 
@@ -3481,18 +3479,20 @@ func (b *LocalBackend) onTailnetDefaultAutoUpdate(au bool) {
 		// can still manually enable auto-updates on this node.
 		return
 	}
-	b.logf("using tailnet default auto-update setting: %v", au)
-	prefsClone := prefs.AsStruct()
-	prefsClone.AutoUpdate.Apply = opt.NewBool(au)
-	_, err := b.editPrefsLockedOnEntry(&ipn.MaskedPrefs{
-		Prefs: *prefsClone,
-		AutoUpdateSet: ipn.AutoUpdatePrefsMask{
-			ApplySet: true,
-		},
-	}, unlock)
-	if err != nil {
-		b.logf("failed to apply tailnet-wide default for auto-updates (%v): %v", au, err)
-		return
+	if clientupdate.CanAutoUpdate() {
+		b.logf("using tailnet default auto-update setting: %v", au)
+		prefsClone := prefs.AsStruct()
+		prefsClone.AutoUpdate.Apply = opt.NewBool(au)
+		_, err := b.editPrefsLockedOnEntry(&ipn.MaskedPrefs{
+			Prefs: *prefsClone,
+			AutoUpdateSet: ipn.AutoUpdatePrefsMask{
+				ApplySet: true,
+			},
+		}, unlock)
+		if err != nil {
+			b.logf("failed to apply tailnet-wide default for auto-updates (%v): %v", au, err)
+			return
+		}
 	}
 }
 
@@ -4968,7 +4968,7 @@ func (b *LocalBackend) authReconfig() {
 		return
 	}
 
-	oneCGNATRoute := shouldUseOneCGNATRoute(b.logf, b.sys.ControlKnobs(), version.OS())
+	oneCGNATRoute := shouldUseOneCGNATRoute(b.logf, b.sys.NetMon.Get(), b.sys.ControlKnobs(), version.OS())
 	rcfg := b.routerConfig(cfg, prefs, oneCGNATRoute)
 
 	err = b.e.Reconfig(cfg, rcfg, dcfg)
@@ -4992,7 +4992,7 @@ func (b *LocalBackend) authReconfig() {
 //
 // The versionOS is a Tailscale-style version ("iOS", "macOS") and not
 // a runtime.GOOS.
-func shouldUseOneCGNATRoute(logf logger.Logf, controlKnobs *controlknobs.Knobs, versionOS string) bool {
+func shouldUseOneCGNATRoute(logf logger.Logf, mon *netmon.Monitor, controlKnobs *controlknobs.Knobs, versionOS string) bool {
 	if controlKnobs != nil {
 		// Explicit enabling or disabling always take precedence.
 		if v, ok := controlKnobs.OneCGNAT.Load().Get(); ok {
@@ -5007,7 +5007,7 @@ func shouldUseOneCGNATRoute(logf logger.Logf, controlKnobs *controlknobs.Knobs,
 	// use fine-grained routes if another interfaces is also using the CGNAT
 	// IP range.
 	if versionOS == "macOS" {
-		hasCGNATInterface, err := netmon.HasCGNATInterface()
+		hasCGNATInterface, err := mon.HasCGNATInterface()
 		if err != nil {
 			logf("shouldUseOneCGNATRoute: Could not determine if any interfaces use CGNAT: %v", err)
 			return false
@@ -5369,6 +5369,7 @@ func (b *LocalBackend) initPeerAPIListener() {
 			ln, err = ps.listen(a.Addr(), b.prevIfState)
 			if err != nil {
 				if peerAPIListenAsync {
+					b.logf("possibly transient peerapi listen(%q) error, will try again on linkChange: %v", a.Addr(), err)
 					// Expected. But we fix it later in linkChange
 					// ("peerAPIListeners too low").
 					continue
@@ -5920,6 +5921,9 @@ func (b *LocalBackend) requestEngineStatusAndWait() {
 	b.logf("requestEngineStatusAndWait: got status update.")
 }
 
+// [controlclient.Auto] implements [auditlog.Transport].
+var _ auditlog.Transport = (*controlclient.Auto)(nil)
+
 // setControlClientLocked sets the control client to cc,
 // which may be nil.
 //
@@ -5927,12 +5931,12 @@ func (b *LocalBackend) requestEngineStatusAndWait() {
 func (b *LocalBackend) setControlClientLocked(cc controlclient.Client) {
 	b.cc = cc
 	b.ccAuto, _ = cc.(*controlclient.Auto)
-	if b.auditLogger != nil {
+	if t, ok := b.cc.(auditlog.Transport); ok && b.auditLogger != nil {
 		if err := b.auditLogger.SetProfileID(b.pm.CurrentProfile().ID()); err != nil {
 			b.logf("audit logger set profile ID failure: %v", err)
 		}
 
-		if err := b.auditLogger.Start(b.ccAuto); err != nil {
+		if err := b.auditLogger.Start(t); err != nil {
 			b.logf("audit logger start failure: %v", err)
 		}
 	}
@@ -7531,6 +7535,7 @@ func (b *LocalBackend) resetForProfileChangeLockedOnEntry(unlock unlockOnce) err
 		return nil
 	}
 	b.setNetMapLocked(nil) // Reset netmap.
+	b.updateFilterLocked(nil, ipn.PrefsView{})
 	// Reset the NetworkMap in the engine
 	b.e.SetNetworkMap(new(netmap.NetworkMap))
 	if prevCC := b.resetControlClientLocked(); prevCC != nil {

ipn/ipnlocal/local_test.go

@@ -1510,6 +1510,15 @@ func TestReconfigureAppConnector(t *testing.T) {
 func TestBackfillAppConnectorRoutes(t *testing.T) {
 	// Create backend with an empty app connector.
 	b := newTestBackend(t)
+	// newTestBackend creates a backend with a non-nil netmap,
+	// but this test requires a nil netmap.
+	// Otherwise, instead of backfilling, [LocalBackend.reconfigAppConnectorLocked]
+	// uses the domains and routes from netmap's [appctype.AppConnectorAttr].
+	// Additionally, a non-nil netmap makes reconfigAppConnectorLocked
+	// asynchronous, resulting in a flaky test.
+	// Therefore, we set the netmap to nil to simulate a fresh backend start
+	// or a profile switch where the netmap is not yet available.
+	b.setNetMapLocked(nil)
 	if err := b.Start(ipn.Options{}); err != nil {
 		t.Fatal(err)
 	}
@@ -4745,32 +4754,133 @@ func TestLoginNotifications(t *testing.T) {
 // TestConfigFileReload tests that the LocalBackend reloads its configuration
 // when the configuration file changes.
 func TestConfigFileReload(t *testing.T) {
-	cfg1 := `{"Hostname": "foo", "Version": "alpha0"}`
-	f := filepath.Join(t.TempDir(), "cfg")
-	must.Do(os.WriteFile(f, []byte(cfg1), 0600))
-	sys := new(tsd.System)
-	sys.InitialConfig = must.Get(conffile.Load(f))
-	lb := newTestLocalBackendWithSys(t, sys)
-	must.Do(lb.Start(ipn.Options{}))
-
-	lb.mu.Lock()
-	hn := lb.hostinfo.Hostname
-	lb.mu.Unlock()
-	if hn != "foo" {
-		t.Fatalf("got %q; want %q", hn, "foo")
+	type testCase struct {
+		name    string
+		initial *conffile.Config
+		updated *conffile.Config
+		checkFn func(*testing.T, *LocalBackend)
 	}
 
-	cfg2 := `{"Hostname": "bar", "Version": "alpha0"}`
-	must.Do(os.WriteFile(f, []byte(cfg2), 0600))
-	if !must.Get(lb.ReloadConfig()) {
-		t.Fatal("reload failed")
+	tests := []testCase{
+		{
+			name: "hostname_change",
+			initial: &conffile.Config{
+				Parsed: ipn.ConfigVAlpha{
+					Version:  "alpha0",
+					Hostname: ptr.To("initial-host"),
+				},
+			},
+			updated: &conffile.Config{
+				Parsed: ipn.ConfigVAlpha{
+					Version:  "alpha0",
+					Hostname: ptr.To("updated-host"),
+				},
+			},
+			checkFn: func(t *testing.T, b *LocalBackend) {
+				if got := b.Prefs().Hostname(); got != "updated-host" {
+					t.Errorf("hostname = %q; want updated-host", got)
+				}
+			},
+		},
+		{
+			name: "start_advertising_services",
+			initial: &conffile.Config{
+				Parsed: ipn.ConfigVAlpha{
+					Version: "alpha0",
+				},
+			},
+			updated: &conffile.Config{
+				Parsed: ipn.ConfigVAlpha{
+					Version:           "alpha0",
+					AdvertiseServices: []string{"svc:abc", "svc:def"},
+				},
+			},
+			checkFn: func(t *testing.T, b *LocalBackend) {
+				if got := b.Prefs().AdvertiseServices().AsSlice(); !reflect.DeepEqual(got, []string{"svc:abc", "svc:def"}) {
+					t.Errorf("AdvertiseServices = %v; want [svc:abc, svc:def]", got)
+				}
+			},
+		},
+		{
+			name: "change_advertised_services",
+			initial: &conffile.Config{
+				Parsed: ipn.ConfigVAlpha{
+					Version:           "alpha0",
+					AdvertiseServices: []string{"svc:abc", "svc:def"},
+				},
+			},
+			updated: &conffile.Config{
+				Parsed: ipn.ConfigVAlpha{
+					Version:           "alpha0",
+					AdvertiseServices: []string{"svc:abc", "svc:ghi"},
+				},
+			},
+			checkFn: func(t *testing.T, b *LocalBackend) {
+				if got := b.Prefs().AdvertiseServices().AsSlice(); !reflect.DeepEqual(got, []string{"svc:abc", "svc:ghi"}) {
+					t.Errorf("AdvertiseServices = %v; want [svc:abc, svc:ghi]", got)
+				}
+			},
+		},
+		{
+			name: "unset_advertised_services",
+			initial: &conffile.Config{
+				Parsed: ipn.ConfigVAlpha{
+					Version:           "alpha0",
+					AdvertiseServices: []string{"svc:abc"},
+				},
+			},
+			updated: &conffile.Config{
+				Parsed: ipn.ConfigVAlpha{
+					Version: "alpha0",
+				},
+			},
+			checkFn: func(t *testing.T, b *LocalBackend) {
+				if b.Prefs().AdvertiseServices().Len() != 0 {
+					t.Errorf("got %d AdvertiseServices wants none", b.Prefs().AdvertiseServices().Len())
+				}
+			},
+		},
 	}
 
-	lb.mu.Lock()
-	hn = lb.hostinfo.Hostname
-	lb.mu.Unlock()
-	if hn != "bar" {
-		t.Fatalf("got %q; want %q", hn, "bar")
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			dir := t.TempDir()
+			path := filepath.Join(dir, "tailscale.conf")
+
+			// Write initial config
+			initialJSON, err := json.Marshal(tc.initial.Parsed)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if err := os.WriteFile(path, initialJSON, 0644); err != nil {
+				t.Fatal(err)
+			}
+
+			// Create backend with initial config
+			tc.initial.Path = path
+			tc.initial.Raw = initialJSON
+			sys := &tsd.System{
+				InitialConfig: tc.initial,
+			}
+			b := newTestLocalBackendWithSys(t, sys)
+
+			// Update config file
+			updatedJSON, err := json.Marshal(tc.updated.Parsed)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if err := os.WriteFile(path, updatedJSON, 0644); err != nil {
+				t.Fatal(err)
+			}
+
+			// Trigger reload
+			if ok, err := b.ReloadConfig(); !ok || err != nil {
+				t.Fatalf("ReloadConfig() = %v, %v; want true, nil", ok, err)
+			}
+
+			// Check outcome
+			tc.checkFn(t, b)
+		})
 	}
 }
 

ipn/ipnlocal/peerapi.go

@@ -481,7 +481,7 @@ func (h *peerAPIHandler) handleServeInterfaces(w http.ResponseWriter, r *http.Re
 		fmt.Fprintf(w, "<h3>Could not get the default route: %s</h3>\n", html.EscapeString(err.Error()))
 	}
 
-	if hasCGNATInterface, err := netmon.HasCGNATInterface(); hasCGNATInterface {
+	if hasCGNATInterface, err := h.ps.b.sys.NetMon.Get().HasCGNATInterface(); hasCGNATInterface {
 		fmt.Fprintln(w, "<p>There is another interface using the CGNAT range.</p>")
 	} else if err != nil {
 		fmt.Fprintf(w, "<p>Could not check for CGNAT interfaces: %s</p>\n", html.EscapeString(err.Error()))

ipn/ipnlocal/state_test.go

@@ -735,12 +735,10 @@ func TestStateMachine(t *testing.T) {
 		// b.Shutdown() explicitly ourselves.
 		previousCC.assertShutdown(false)
 
-		// Note: unpause happens because ipn needs to get at least one netmap
-		//  on startup, otherwise UIs can't show the node list, login
-		//  name, etc when in state ipn.Stopped.
-		//  Arguably they shouldn't try. But they currently do.
 		nn := notifies.drain(2)
-		cc.assertCalls("New", "Login")
+		// We already have a netmap for this node,
+		// and WantRunning is false, so cc should be paused.
+		cc.assertCalls("New", "Login", "pause")
 		c.Assert(nn[0].Prefs, qt.IsNotNil)
 		c.Assert(nn[1].State, qt.IsNotNil)
 		c.Assert(nn[0].Prefs.WantRunning(), qt.IsFalse)
@@ -751,7 +749,11 @@ func TestStateMachine(t *testing.T) {
 	// When logged in but !WantRunning, ipn leaves us unpaused to retrieve
 	// the first netmap. Simulate that netmap being received, after which
 	// it should pause us, to avoid wasting CPU retrieving unnecessarily
-	// additional netmap updates.
+	// additional netmap updates. Since our LocalBackend instance already
+	// has a netmap, we will reset it to nil to simulate the first netmap
+	// retrieval.
+	b.setNetMapLocked(nil)
+	cc.assertCalls("unpause")
 	//
 	// TODO: really the various GUIs and prefs should be refactored to
 	//  not require the netmap structure at all when starting while
@@ -853,7 +855,7 @@ func TestStateMachine(t *testing.T) {
 	// The last test case is the most common one: restarting when both
 	// logged in and WantRunning.
 	t.Logf("\n\nStart5")
-	notifies.expect(1)
+	notifies.expect(2)
 	c.Assert(b.Start(ipn.Options{}), qt.IsNil)
 	{
 		// NOTE: cc.Shutdown() is correct here, since we didn't call
@@ -861,30 +863,32 @@ func TestStateMachine(t *testing.T) {
 		previousCC.assertShutdown(false)
 		cc.assertCalls("New", "Login")
 
-		nn := notifies.drain(1)
+		nn := notifies.drain(2)
 		cc.assertCalls()
 		c.Assert(nn[0].Prefs, qt.IsNotNil)
 		c.Assert(nn[0].Prefs.LoggedOut(), qt.IsFalse)
 		c.Assert(nn[0].Prefs.WantRunning(), qt.IsTrue)
-		c.Assert(b.State(), qt.Equals, ipn.NoState)
+		// We're logged in and have a valid netmap, so we should
+		// be in the Starting state.
+		c.Assert(nn[1].State, qt.IsNotNil)
+		c.Assert(*nn[1].State, qt.Equals, ipn.Starting)
+		c.Assert(b.State(), qt.Equals, ipn.Starting)
 	}
 
 	// Control server accepts our valid key from before.
 	t.Logf("\n\nLoginFinished5")
-	notifies.expect(1)
+	notifies.expect(0)
 	cc.send(nil, "", true, &netmap.NetworkMap{
 		SelfNode: (&tailcfg.Node{MachineAuthorized: true}).View(),
 	})
 	{
-		nn := notifies.drain(1)
+		notifies.drain(0)
 		cc.assertCalls()
 		// NOTE: No LoginFinished message since no interactive
 		// login was needed.
-		c.Assert(nn[0].State, qt.IsNotNil)
-		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
 		// NOTE: No prefs change this time. WantRunning stays true.
 		// We were in Starting in the first place, so that doesn't
-		// change either.
+		// change either, so we don't expect any notifications.
 		c.Assert(ipn.Starting, qt.Equals, b.State())
 	}
 	t.Logf("\n\nExpireKey")

ipn/store/kubestore/store_kube.go

@@ -13,11 +13,14 @@ import (
 	"strings"
 	"time"
 
+	"tailscale.com/envknob"
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/store/mem"
 	"tailscale.com/kube/kubeapi"
 	"tailscale.com/kube/kubeclient"
+	"tailscale.com/kube/kubetypes"
 	"tailscale.com/types/logger"
+	"tailscale.com/util/dnsname"
 	"tailscale.com/util/mak"
 )
 
@@ -32,21 +35,37 @@ const (
 	reasonTailscaleStateLoadFailed   = "TailscaleStateLoadFailed"
 	eventTypeWarning                 = "Warning"
 	eventTypeNormal                  = "Normal"
+
+	keyTLSCert = "tls.crt"
+	keyTLSKey  = "tls.key"
 )
 
 // Store is an ipn.StateStore that uses a Kubernetes Secret for persistence.
 type Store struct {
-	client     kubeclient.Client
-	canPatch   bool
-	secretName string
+	client        kubeclient.Client
+	canPatch      bool
+	secretName    string // state Secret
+	certShareMode string // 'ro', 'rw', or empty
+	podName       string
 
-	// memory holds the latest tailscale state. Writes write state to a kube Secret and memory, Reads read from
-	// memory.
+	// memory holds the latest tailscale state. Writes write state to a kube
+	// Secret and memory, Reads read from memory.
 	memory mem.Store
 }
 
-// New returns a new Store that persists to the named Secret.
-func New(_ logger.Logf, secretName string) (*Store, error) {
+// New returns a new Store that persists state to Kubernets Secret(s).
+// Tailscale state is stored in a Secret named by the secretName parameter.
+// TLS certs are stored and retrieved from state Secret or separate Secrets
+// named after TLS endpoints if running in cert share mode.
+func New(logf logger.Logf, secretName string) (*Store, error) {
+	c, err := newClient()
+	if err != nil {
+		return nil, err
+	}
+	return newWithClient(logf, c, secretName)
+}
+
+func newClient() (kubeclient.Client, error) {
 	c, err := kubeclient.New("tailscale-state-store")
 	if err != nil {
 		return nil, err
@@ -55,6 +74,10 @@ func New(_ logger.Logf, secretName string) (*Store, error) {
 		// Derive the API server address from the environment variables
 		c.SetURL(fmt.Sprintf("https://%s:%s", os.Getenv("KUBERNETES_SERVICE_HOST"), os.Getenv("KUBERNETES_SERVICE_PORT_HTTPS")))
 	}
+	return c, nil
+}
+
+func newWithClient(logf logger.Logf, c kubeclient.Client, secretName string) (*Store, error) {
 	canPatch, _, err := c.CheckSecretPermissions(context.Background(), secretName)
 	if err != nil {
 		return nil, err
@@ -63,11 +86,30 @@ func New(_ logger.Logf, secretName string) (*Store, error) {
 		client:     c,
 		canPatch:   canPatch,
 		secretName: secretName,
+		podName:    os.Getenv("POD_NAME"),
 	}
+	if envknob.IsCertShareReadWriteMode() {
+		s.certShareMode = "rw"
+	} else if envknob.IsCertShareReadOnlyMode() {
+		s.certShareMode = "ro"
+	}
+
 	// Load latest state from kube Secret if it already exists.
 	if err := s.loadState(); err != nil && err != ipn.ErrStateNotExist {
 		return nil, fmt.Errorf("error loading state from kube Secret: %w", err)
 	}
+	// If we are in cert share mode, pre-load existing shared certs.
+	if s.certShareMode == "rw" || s.certShareMode == "ro" {
+		sel := s.certSecretSelector()
+		if err := s.loadCerts(context.Background(), sel); err != nil {
+			// We will attempt to again retrieve the certs from Secrets when a request for an HTTPS endpoint
+			// is received.
+			log.Printf("[unexpected] error loading TLS certs: %v", err)
+		}
+	}
+	if s.certShareMode == "ro" {
+		go s.runCertReload(context.Background(), logf)
+	}
 	return s, nil
 }
 
@@ -84,27 +126,110 @@ func (s *Store) ReadState(id ipn.StateKey) ([]byte, error) {
 
 // WriteState implements the StateStore interface.
 func (s *Store) WriteState(id ipn.StateKey, bs []byte) (err error) {
-	return s.updateStateSecret(map[string][]byte{string(id): bs})
-}
-
-// WriteTLSCertAndKey writes a TLS cert and key to domain.crt, domain.key fields of a Tailscale Kubernetes node's state
-// Secret.
-func (s *Store) WriteTLSCertAndKey(domain string, cert, key []byte) error {
-	return s.updateStateSecret(map[string][]byte{domain + ".crt": cert, domain + ".key": key})
-}
-
-func (s *Store) updateStateSecret(data map[string][]byte) (err error) {
-	ctx, cancel := context.WithTimeout(context.Background(), timeout)
 	defer func() {
 		if err == nil {
-			for id, bs := range data {
-				// The in-memory store does not distinguish between values read from state Secret on
-				// init and values written to afterwards. Values read from the state
-				// Secret will always be sanitized, so we also need to sanitize values written to store
-				// later, so that the Read logic can just lookup keys in sanitized form.
-				s.memory.WriteState(ipn.StateKey(sanitizeKey(id)), bs)
-			}
+			s.memory.WriteState(ipn.StateKey(sanitizeKey(id)), bs)
 		}
+	}()
+	return s.updateSecret(map[string][]byte{string(id): bs}, s.secretName)
+}
+
+// WriteTLSCertAndKey writes a TLS cert and key to domain.crt, domain.key fields
+// of a Tailscale Kubernetes node's state Secret.
+func (s *Store) WriteTLSCertAndKey(domain string, cert, key []byte) (err error) {
+	if s.certShareMode == "ro" {
+		log.Printf("[unexpected] TLS cert and key write in read-only mode")
+	}
+	if err := dnsname.ValidHostname(domain); err != nil {
+		return fmt.Errorf("invalid domain name %q: %w", domain, err)
+	}
+	secretName := s.secretName
+	data := map[string][]byte{
+		domain + ".crt": cert,
+		domain + ".key": key,
+	}
+	// If we run in cert share mode, cert and key for a DNS name are written
+	// to a separate Secret.
+	if s.certShareMode == "rw" {
+		secretName = domain
+		data = map[string][]byte{
+			keyTLSCert: cert,
+			keyTLSKey:  key,
+		}
+	}
+	if err := s.updateSecret(data, secretName); err != nil {
+		return fmt.Errorf("error writing TLS cert and key to Secret: %w", err)
+	}
+	// TODO(irbekrm): certs for write replicas are currently not
+	// written to memory to avoid out of sync memory state after
+	// Ingress resources have been recreated.  This means that TLS
+	// certs for write replicas are retrieved from the Secret on
+	// each HTTPS request.  This is a temporary solution till we
+	// implement a Secret watch.
+	if s.certShareMode != "rw" {
+		s.memory.WriteState(ipn.StateKey(domain+".crt"), cert)
+		s.memory.WriteState(ipn.StateKey(domain+".key"), key)
+	}
+	return nil
+}
+
+// ReadTLSCertAndKey reads a TLS cert and key from memory or from a
+// domain-specific Secret. It first checks the in-memory store, if not found in
+// memory and running cert store in read-only mode, looks up a Secret.
+// Note that write replicas of HA Ingress always retrieve TLS certs from Secrets.
+func (s *Store) ReadTLSCertAndKey(domain string) (cert, key []byte, err error) {
+	if err := dnsname.ValidHostname(domain); err != nil {
+		return nil, nil, fmt.Errorf("invalid domain name %q: %w", domain, err)
+	}
+	certKey := domain + ".crt"
+	keyKey := domain + ".key"
+	cert, err = s.memory.ReadState(ipn.StateKey(certKey))
+	if err == nil {
+		key, err = s.memory.ReadState(ipn.StateKey(keyKey))
+		if err == nil {
+			return cert, key, nil
+		}
+	}
+	if s.certShareMode == "" {
+		return nil, nil, ipn.ErrStateNotExist
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)
+	defer cancel()
+	secret, err := s.client.GetSecret(ctx, domain)
+	if err != nil {
+		if kubeclient.IsNotFoundErr(err) {
+			// TODO(irbekrm): we should return a more specific error
+			// that wraps ipn.ErrStateNotExist here.
+			return nil, nil, ipn.ErrStateNotExist
+		}
+		return nil, nil, fmt.Errorf("getting TLS Secret %q: %w", domain, err)
+	}
+	cert = secret.Data[keyTLSCert]
+	key = secret.Data[keyTLSKey]
+	if len(cert) == 0 || len(key) == 0 {
+		return nil, nil, ipn.ErrStateNotExist
+	}
+	// TODO(irbekrm): a read between these two separate writes would
+	// get a mismatched cert and key.  Allow writing both cert and
+	// key to the memory store in a single, lock-protected operation.
+	//
+	// TODO(irbekrm): currently certs for write replicas of HA Ingress get
+	// retrieved from the cluster Secret on each HTTPS request to avoid a
+	// situation when after Ingress recreation stale certs are read from
+	// memory.
+	// Fix this by watching Secrets to ensure that memory store gets updated
+	// when Secrets are deleted.
+	if s.certShareMode == "ro" {
+		s.memory.WriteState(ipn.StateKey(certKey), cert)
+		s.memory.WriteState(ipn.StateKey(keyKey), key)
+	}
+	return cert, key, nil
+}
+
+func (s *Store) updateSecret(data map[string][]byte, secretName string) (err error) {
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)
+	defer func() {
 		if err != nil {
 			if err := s.client.Event(ctx, eventTypeWarning, reasonTailscaleStateUpdateFailed, err.Error()); err != nil {
 				log.Printf("kubestore: error creating tailscaled state update Event: %v", err)
@@ -116,17 +241,17 @@ func (s *Store) updateStateSecret(data map[string][]byte) (err error) {
 		}
 		cancel()
 	}()
-	secret, err := s.client.GetSecret(ctx, s.secretName)
+	secret, err := s.client.GetSecret(ctx, secretName)
 	if err != nil {
 		// If the Secret does not exist, create it with the required data.
-		if kubeclient.IsNotFoundErr(err) {
+		if kubeclient.IsNotFoundErr(err) && s.canCreateSecret(secretName) {
 			return s.client.CreateSecret(ctx, &kubeapi.Secret{
 				TypeMeta: kubeapi.TypeMeta{
 					APIVersion: "v1",
 					Kind:       "Secret",
 				},
 				ObjectMeta: kubeapi.ObjectMeta{
-					Name: s.secretName,
+					Name: secretName,
 				},
 				Data: func(m map[string][]byte) map[string][]byte {
 					d := make(map[string][]byte, len(m))
@@ -137,9 +262,9 @@ func (s *Store) updateStateSecret(data map[string][]byte) (err error) {
 				}(data),
 			})
 		}
-		return err
+		return fmt.Errorf("error getting Secret %s: %w", secretName, err)
 	}
-	if s.canPatch {
+	if s.canPatchSecret(secretName) {
 		var m []kubeclient.JSONPatch
 		// If the user has pre-created a Secret with no data, we need to ensure the top level /data field.
 		if len(secret.Data) == 0 {
@@ -166,8 +291,8 @@ func (s *Store) updateStateSecret(data map[string][]byte) (err error) {
 				})
 			}
 		}
-		if err := s.client.JSONPatchResource(ctx, s.secretName, kubeclient.TypeSecrets, m); err != nil {
-			return fmt.Errorf("error patching Secret %s: %w", s.secretName, err)
+		if err := s.client.JSONPatchResource(ctx, secretName, kubeclient.TypeSecrets, m); err != nil {
+			return fmt.Errorf("error patching Secret %s: %w", secretName, err)
 		}
 		return nil
 	}
@@ -176,9 +301,9 @@ func (s *Store) updateStateSecret(data map[string][]byte) (err error) {
 		mak.Set(&secret.Data, sanitizeKey(key), val)
 	}
 	if err := s.client.UpdateSecret(ctx, secret); err != nil {
-		return err
+		return fmt.Errorf("error updating Secret %s: %w", s.secretName, err)
 	}
-	return err
+	return nil
 }
 
 func (s *Store) loadState() (err error) {
@@ -202,6 +327,96 @@ func (s *Store) loadState() (err error) {
 	return nil
 }
 
+// runCertReload relists and reloads all TLS certs for endpoints shared by this
+// node from Secrets other than the state Secret to ensure that renewed certs get eventually loaded.
+// It is not critical to reload a cert immediately after
+// renewal, so a daily check is acceptable.
+// Currently (3/2025) this is only used for the shared HA Ingress certs on 'read' replicas.
+// Note that if shared certs are not found in memory on an HTTPS request, we
+// do a Secret lookup, so this mechanism does not need to ensure that newly
+// added Ingresses' certs get loaded.
+func (s *Store) runCertReload(ctx context.Context, logf logger.Logf) {
+	ticker := time.NewTicker(time.Hour * 24)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+			sel := s.certSecretSelector()
+			if err := s.loadCerts(ctx, sel); err != nil {
+				logf("[unexpected] error reloading TLS certs: %v", err)
+			}
+		}
+	}
+}
+
+// loadCerts lists all Secrets matching the provided selector and loads TLS
+// certs and keys from those.
+func (s *Store) loadCerts(ctx context.Context, sel map[string]string) error {
+	ss, err := s.client.ListSecrets(ctx, sel)
+	if err != nil {
+		return fmt.Errorf("error listing TLS Secrets: %w", err)
+	}
+	for _, secret := range ss.Items {
+		if !hasTLSData(&secret) {
+			continue
+		}
+		// Only load secrets that have valid domain names (ending in .ts.net)
+		if !strings.HasSuffix(secret.Name, ".ts.net") {
+			continue
+		}
+		s.memory.WriteState(ipn.StateKey(secret.Name)+".crt", secret.Data[keyTLSCert])
+		s.memory.WriteState(ipn.StateKey(secret.Name)+".key", secret.Data[keyTLSKey])
+	}
+	return nil
+}
+
+// canCreateSecret returns true if this node should be allowed to create the given
+// Secret in its namespace.
+func (s *Store) canCreateSecret(secret string) bool {
+	// Only allow creating the state Secret (and not TLS Secrets).
+	return secret == s.secretName
+}
+
+// canPatchSecret returns true if this node should be allowed to patch the given
+// Secret.
+func (s *Store) canPatchSecret(secret string) bool {
+	// For backwards compatibility reasons, setups where the proxies are not
+	// given PATCH permissions for state Secrets are allowed. For TLS
+	// Secrets, we should always have PATCH permissions.
+	if secret == s.secretName {
+		return s.canPatch
+	}
+	return true
+}
+
+// certSecretSelector returns a label selector that can be used to list all
+// Secrets that aren't Tailscale state Secrets and contain TLS certificates for
+// HTTPS endpoints that this node serves.
+// Currently (3/2025) this only applies to the Kubernetes Operator's ingress
+// ProxyGroup.
+func (s *Store) certSecretSelector() map[string]string {
+	if s.podName == "" {
+		return map[string]string{}
+	}
+	p := strings.LastIndex(s.podName, "-")
+	if p == -1 {
+		return map[string]string{}
+	}
+	pgName := s.podName[:p]
+	return map[string]string{
+		kubetypes.LabelSecretType:   "certs",
+		kubetypes.LabelManaged:      "true",
+		"tailscale.com/proxy-group": pgName,
+	}
+}
+
+// hasTLSData returns true if the provided Secret contains non-empty TLS cert and key.
+func hasTLSData(s *kubeapi.Secret) bool {
+	return len(s.Data[keyTLSCert]) != 0 && len(s.Data[keyTLSKey]) != 0
+}
+
 // sanitizeKey converts any value that can be converted to a string into a valid Kubernetes Secret key.
 // Valid characters are alphanumeric, -, _, and .
 // https://kubernetes.io/docs/concepts/configuration/secret/#restriction-names-data.

ipn/store/kubestore/store_kube_test.go

@@ -4,33 +4,37 @@
 package kubestore
 
 import (
+	"bytes"
 	"context"
+	"encoding/json"
+	"fmt"
 	"strings"
 	"testing"
 
 	"github.com/google/go-cmp/cmp"
+	"tailscale.com/envknob"
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/store/mem"
 	"tailscale.com/kube/kubeapi"
 	"tailscale.com/kube/kubeclient"
 )
 
-func TestUpdateStateSecret(t *testing.T) {
+func TestWriteState(t *testing.T) {
 	tests := []struct {
 		name       string
 		initial    map[string][]byte
-		updates    map[string][]byte
+		key        ipn.StateKey
+		value      []byte
 		wantData   map[string][]byte
 		allowPatch bool
 	}{
 		{
-			name: "basic_update",
+			name: "basic_write",
 			initial: map[string][]byte{
 				"existing": []byte("old"),
 			},
-			updates: map[string][]byte{
-				"foo": []byte("bar"),
-			},
+			key:   "foo",
+			value: []byte("bar"),
 			wantData: map[string][]byte{
 				"existing": []byte("old"),
 				"foo":      []byte("bar"),
@@ -42,35 +46,17 @@ func TestUpdateStateSecret(t *testing.T) {
 			initial: map[string][]byte{
 				"foo": []byte("old"),
 			},
-			updates: map[string][]byte{
-				"foo": []byte("new"),
-			},
+			key:   "foo",
+			value: []byte("new"),
 			wantData: map[string][]byte{
 				"foo": []byte("new"),
 			},
 			allowPatch: true,
 		},
 		{
-			name: "multiple_updates",
-			initial: map[string][]byte{
-				"keep": []byte("keep"),
-			},
-			updates: map[string][]byte{
-				"foo": []byte("bar"),
-				"baz": []byte("qux"),
-			},
-			wantData: map[string][]byte{
-				"keep": []byte("keep"),
-				"foo":  []byte("bar"),
-				"baz":  []byte("qux"),
-			},
-			allowPatch: true,
-		},
-		{
-			name: "create_new_secret",
-			updates: map[string][]byte{
-				"foo": []byte("bar"),
-			},
+			name:  "create_new_secret",
+			key:   "foo",
+			value: []byte("bar"),
 			wantData: map[string][]byte{
 				"foo": []byte("bar"),
 			},
@@ -81,29 +67,23 @@ func TestUpdateStateSecret(t *testing.T) {
 			initial: map[string][]byte{
 				"foo": []byte("old"),
 			},
-			updates: map[string][]byte{
-				"foo": []byte("new"),
-			},
+			key:   "foo",
+			value: []byte("new"),
 			wantData: map[string][]byte{
 				"foo": []byte("new"),
 			},
 			allowPatch: false,
 		},
 		{
-			name: "sanitize_keys",
+			name: "sanitize_key",
 			initial: map[string][]byte{
 				"clean-key": []byte("old"),
 			},
-			updates: map[string][]byte{
-				"dirty@key": []byte("new"),
-				"also/bad":  []byte("value"),
-				"good.key":  []byte("keep"),
-			},
+			key:   "dirty@key",
+			value: []byte("new"),
 			wantData: map[string][]byte{
 				"clean-key": []byte("old"),
 				"dirty_key": []byte("new"),
-				"also_bad":  []byte("value"),
-				"good.key":  []byte("keep"),
 			},
 			allowPatch: true,
 		},
@@ -152,13 +132,13 @@ func TestUpdateStateSecret(t *testing.T) {
 			s := &Store{
 				client:     client,
 				canPatch:   tt.allowPatch,
-				secretName: "test-secret",
+				secretName: "ts-state",
 				memory:     mem.Store{},
 			}
 
-			err := s.updateStateSecret(tt.updates)
+			err := s.WriteState(tt.key, tt.value)
 			if err != nil {
-				t.Errorf("updateStateSecret() error = %v", err)
+				t.Errorf("WriteState() error = %v", err)
 				return
 			}
 
@@ -168,16 +148,579 @@ func TestUpdateStateSecret(t *testing.T) {
 			}
 
 			// Verify memory store was updated
-			for k, v := range tt.updates {
-				got, err := s.memory.ReadState(ipn.StateKey(sanitizeKey(k)))
+			got, err := s.memory.ReadState(ipn.StateKey(sanitizeKey(string(tt.key))))
+			if err != nil {
+				t.Errorf("reading from memory store: %v", err)
+			}
+			if !cmp.Equal(got, tt.value) {
+				t.Errorf("memory store key %q = %v, want %v", tt.key, got, tt.value)
+			}
+		})
+	}
+}
+
+func TestWriteTLSCertAndKey(t *testing.T) {
+	const (
+		testDomain = "my-app.tailnetxyz.ts.net"
+		testCert   = "fake-cert"
+		testKey    = "fake-key"
+	)
+
+	tests := []struct {
+		name            string
+		initial         map[string][]byte // pre-existing cert and key
+		certShareMode   string
+		allowPatch      bool   // whether client can patch the Secret
+		wantSecretName  string // name of the Secret where cert and key should be written
+		wantSecretData  map[string][]byte
+		wantMemoryStore map[ipn.StateKey][]byte
+	}{
+		{
+			name: "basic_write",
+			initial: map[string][]byte{
+				"existing": []byte("old"),
+			},
+			allowPatch:     true,
+			wantSecretName: "ts-state",
+			wantSecretData: map[string][]byte{
+				"existing":                     []byte("old"),
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+			wantMemoryStore: map[ipn.StateKey][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+		},
+		{
+			name:           "cert_share_mode_write",
+			certShareMode:  "rw",
+			allowPatch:     true,
+			wantSecretName: "my-app.tailnetxyz.ts.net",
+			wantSecretData: map[string][]byte{
+				"tls.crt": []byte(testCert),
+				"tls.key": []byte(testKey),
+			},
+		},
+		{
+			name: "cert_share_mode_write_update_existing",
+			initial: map[string][]byte{
+				"tls.crt": []byte("old-cert"),
+				"tls.key": []byte("old-key"),
+			},
+			certShareMode:  "rw",
+			allowPatch:     true,
+			wantSecretName: "my-app.tailnetxyz.ts.net",
+			wantSecretData: map[string][]byte{
+				"tls.crt": []byte(testCert),
+				"tls.key": []byte(testKey),
+			},
+		},
+		{
+			name: "update_existing",
+			initial: map[string][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte("old-cert"),
+				"my-app.tailnetxyz.ts.net.key": []byte("old-key"),
+			},
+			certShareMode:  "",
+			allowPatch:     true,
+			wantSecretName: "ts-state",
+			wantSecretData: map[string][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+			wantMemoryStore: map[ipn.StateKey][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+		},
+		{
+			name:           "patch_denied",
+			certShareMode:  "",
+			allowPatch:     false,
+			wantSecretName: "ts-state",
+			wantSecretData: map[string][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+			wantMemoryStore: map[ipn.StateKey][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+
+			// Set POD_NAME for testing selectors
+			envknob.Setenv("POD_NAME", "ingress-proxies-1")
+			defer envknob.Setenv("POD_NAME", "")
+
+			secret := tt.initial // track current state
+			client := &kubeclient.FakeClient{
+				GetSecretImpl: func(ctx context.Context, name string) (*kubeapi.Secret, error) {
+					if secret == nil {
+						return nil, &kubeapi.Status{Code: 404}
+					}
+					return &kubeapi.Secret{Data: secret}, nil
+				},
+				CheckSecretPermissionsImpl: func(ctx context.Context, name string) (bool, bool, error) {
+					return tt.allowPatch, true, nil
+				},
+				CreateSecretImpl: func(ctx context.Context, s *kubeapi.Secret) error {
+					if s.Name != tt.wantSecretName {
+						t.Errorf("CreateSecret called with wrong name, got %q, want %q", s.Name, tt.wantSecretName)
+					}
+					secret = s.Data
+					return nil
+				},
+				UpdateSecretImpl: func(ctx context.Context, s *kubeapi.Secret) error {
+					if s.Name != tt.wantSecretName {
+						t.Errorf("UpdateSecret called with wrong name, got %q, want %q", s.Name, tt.wantSecretName)
+					}
+					secret = s.Data
+					return nil
+				},
+				JSONPatchResourceImpl: func(ctx context.Context, name, resourceType string, patches []kubeclient.JSONPatch) error {
+					if !tt.allowPatch {
+						return &kubeapi.Status{Reason: "Forbidden"}
+					}
+					if name != tt.wantSecretName {
+						t.Errorf("JSONPatchResource called with wrong name, got %q, want %q", name, tt.wantSecretName)
+					}
+					if secret == nil {
+						secret = make(map[string][]byte)
+					}
+					for _, p := range patches {
+						if p.Op == "add" && p.Path == "/data" {
+							secret = p.Value.(map[string][]byte)
+						} else if p.Op == "add" && strings.HasPrefix(p.Path, "/data/") {
+							key := strings.TrimPrefix(p.Path, "/data/")
+							secret[key] = p.Value.([]byte)
+						}
+					}
+					return nil
+				},
+			}
+
+			s := &Store{
+				client:        client,
+				canPatch:      tt.allowPatch,
+				secretName:    tt.wantSecretName,
+				certShareMode: tt.certShareMode,
+				memory:        mem.Store{},
+			}
+
+			err := s.WriteTLSCertAndKey(testDomain, []byte(testCert), []byte(testKey))
+			if err != nil {
+				t.Errorf("WriteTLSCertAndKey() error = '%v'", err)
+				return
+			}
+
+			// Verify secret data
+			if diff := cmp.Diff(secret, tt.wantSecretData); diff != "" {
+				t.Errorf("secret data mismatch (-got +want):\n%s", diff)
+			}
+
+			// Verify memory store was updated
+			for key, want := range tt.wantMemoryStore {
+				got, err := s.memory.ReadState(key)
 				if err != nil {
 					t.Errorf("reading from memory store: %v", err)
 					continue
 				}
-				if !cmp.Equal(got, v) {
-					t.Errorf("memory store key %q = %v, want %v", k, got, v)
+				if !cmp.Equal(got, want) {
+					t.Errorf("memory store key %q = %v, want %v", key, got, want)
 				}
 			}
 		})
 	}
 }
+
+func TestReadTLSCertAndKey(t *testing.T) {
+	const (
+		testDomain = "my-app.tailnetxyz.ts.net"
+		testCert   = "fake-cert"
+		testKey    = "fake-key"
+	)
+
+	tests := []struct {
+		name          string
+		memoryStore   map[ipn.StateKey][]byte // pre-existing memory store state
+		certShareMode string
+		domain        string
+		secretData    map[string][]byte // data to return from mock GetSecret
+		secretGetErr  error             // error to return from mock GetSecret
+		wantCert      []byte
+		wantKey       []byte
+		wantErr       error
+		// what should end up in memory store after the store is created
+		wantMemoryStore map[ipn.StateKey][]byte
+	}{
+		{
+			name: "found_in_memory",
+			memoryStore: map[ipn.StateKey][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+			domain:   testDomain,
+			wantCert: []byte(testCert),
+			wantKey:  []byte(testKey),
+			wantMemoryStore: map[ipn.StateKey][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+		},
+		{
+			name:    "not_found_in_memory",
+			domain:  testDomain,
+			wantErr: ipn.ErrStateNotExist,
+		},
+		{
+			name:          "cert_share_ro_mode_found_in_secret",
+			certShareMode: "ro",
+			domain:        testDomain,
+			secretData: map[string][]byte{
+				"tls.crt": []byte(testCert),
+				"tls.key": []byte(testKey),
+			},
+			wantCert: []byte(testCert),
+			wantKey:  []byte(testKey),
+			wantMemoryStore: map[ipn.StateKey][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+		},
+		{
+			name:          "cert_share_rw_mode_found_in_secret",
+			certShareMode: "rw",
+			domain:        testDomain,
+			secretData: map[string][]byte{
+				"tls.crt": []byte(testCert),
+				"tls.key": []byte(testKey),
+			},
+			wantCert: []byte(testCert),
+			wantKey:  []byte(testKey),
+		},
+		{
+			name:          "cert_share_ro_mode_found_in_memory",
+			certShareMode: "ro",
+			memoryStore: map[ipn.StateKey][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+			domain:   testDomain,
+			wantCert: []byte(testCert),
+			wantKey:  []byte(testKey),
+			wantMemoryStore: map[ipn.StateKey][]byte{
+				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
+				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
+			},
+		},
+		{
+			name:          "cert_share_ro_mode_not_found",
+			certShareMode: "ro",
+			domain:        testDomain,
+			secretGetErr:  &kubeapi.Status{Code: 404},
+			wantErr:       ipn.ErrStateNotExist,
+		},
+		{
+			name:          "cert_share_ro_mode_empty_cert_in_secret",
+			certShareMode: "ro",
+			domain:        testDomain,
+			secretData: map[string][]byte{
+				"tls.crt": {},
+				"tls.key": []byte(testKey),
+			},
+			wantErr: ipn.ErrStateNotExist,
+		},
+		{
+			name:          "cert_share_ro_mode_kube_api_error",
+			certShareMode: "ro",
+			domain:        testDomain,
+			secretGetErr:  fmt.Errorf("api error"),
+			wantErr:       fmt.Errorf("getting TLS Secret %q: api error", sanitizeKey(testDomain)),
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+
+			client := &kubeclient.FakeClient{
+				GetSecretImpl: func(ctx context.Context, name string) (*kubeapi.Secret, error) {
+					if tt.secretGetErr != nil {
+						return nil, tt.secretGetErr
+					}
+					return &kubeapi.Secret{Data: tt.secretData}, nil
+				},
+			}
+
+			s := &Store{
+				client:        client,
+				secretName:    "ts-state",
+				certShareMode: tt.certShareMode,
+				memory:        mem.Store{},
+			}
+
+			// Initialize memory store
+			for k, v := range tt.memoryStore {
+				s.memory.WriteState(k, v)
+			}
+
+			gotCert, gotKey, err := s.ReadTLSCertAndKey(tt.domain)
+			if tt.wantErr != nil {
+				if err == nil {
+					t.Errorf("ReadTLSCertAndKey() error = nil, want error containing %v", tt.wantErr)
+					return
+				}
+				if !strings.Contains(err.Error(), tt.wantErr.Error()) {
+					t.Errorf("ReadTLSCertAndKey() error = %v, want error containing %v", err, tt.wantErr)
+				}
+				return
+			}
+			if err != nil {
+				t.Errorf("ReadTLSCertAndKey() unexpected error: %v", err)
+				return
+			}
+
+			if !bytes.Equal(gotCert, tt.wantCert) {
+				t.Errorf("ReadTLSCertAndKey() gotCert = %v, want %v", gotCert, tt.wantCert)
+			}
+			if !bytes.Equal(gotKey, tt.wantKey) {
+				t.Errorf("ReadTLSCertAndKey() gotKey = %v, want %v", gotKey, tt.wantKey)
+			}
+
+			// Verify memory store contents after operation
+			if tt.wantMemoryStore != nil {
+				for key, want := range tt.wantMemoryStore {
+					got, err := s.memory.ReadState(key)
+					if err != nil {
+						t.Errorf("reading from memory store: %v", err)
+						continue
+					}
+					if !bytes.Equal(got, want) {
+						t.Errorf("memory store key %q = %v, want %v", key, got, want)
+					}
+				}
+			}
+		})
+	}
+}
+
+func TestNewWithClient(t *testing.T) {
+	const (
+		secretName = "ts-state"
+		testCert   = "fake-cert"
+		testKey    = "fake-key"
+	)
+
+	certSecretsLabels := map[string]string{
+		"tailscale.com/secret-type": "certs",
+		"tailscale.com/managed":     "true",
+		"tailscale.com/proxy-group": "ingress-proxies",
+	}
+
+	// Helper function to create Secret objects for testing
+	makeSecret := func(name string, labels map[string]string, certSuffix string) kubeapi.Secret {
+		return kubeapi.Secret{
+			ObjectMeta: kubeapi.ObjectMeta{
+				Name:   name,
+				Labels: labels,
+			},
+			Data: map[string][]byte{
+				"tls.crt": []byte(testCert + certSuffix),
+				"tls.key": []byte(testKey + certSuffix),
+			},
+		}
+	}
+
+	tests := []struct {
+		name                    string
+		stateSecretContents     map[string][]byte // data in state Secret
+		TLSSecrets              []kubeapi.Secret  // list of TLS cert Secrets
+		certMode                string
+		secretGetErr            error // error to return from GetSecret
+		secretsListErr          error // error to return from ListSecrets
+		wantMemoryStoreContents map[ipn.StateKey][]byte
+		wantErr                 error
+	}{
+		{
+			name:                    "empty_state_secret",
+			stateSecretContents:     map[string][]byte{},
+			wantMemoryStoreContents: map[ipn.StateKey][]byte{},
+		},
+		{
+			name:                    "state_secret_not_found",
+			secretGetErr:            &kubeapi.Status{Code: 404},
+			wantMemoryStoreContents: map[ipn.StateKey][]byte{},
+		},
+		{
+			name:         "state_secret_get_error",
+			secretGetErr: fmt.Errorf("some error"),
+			wantErr:      fmt.Errorf("error loading state from kube Secret: some error"),
+		},
+		{
+			name: "load_existing_state",
+			stateSecretContents: map[string][]byte{
+				"foo": []byte("bar"),
+				"baz": []byte("qux"),
+			},
+			wantMemoryStoreContents: map[ipn.StateKey][]byte{
+				"foo": []byte("bar"),
+				"baz": []byte("qux"),
+			},
+		},
+		{
+			name:     "load_select_certs_in_read_only_mode",
+			certMode: "ro",
+			stateSecretContents: map[string][]byte{
+				"foo": []byte("bar"),
+			},
+			TLSSecrets: []kubeapi.Secret{
+				makeSecret("app1.tailnetxyz.ts.net", certSecretsLabels, "1"),
+				makeSecret("app2.tailnetxyz.ts.net", certSecretsLabels, "2"),
+				makeSecret("some-other-secret", nil, "3"),
+				makeSecret("app3.other-proxies.ts.net", map[string]string{
+					"tailscale.com/secret-type": "certs",
+					"tailscale.com/managed":     "true",
+					"tailscale.com/proxy-group": "some-other-proxygroup",
+				}, "4"),
+			},
+			wantMemoryStoreContents: map[ipn.StateKey][]byte{
+				"foo":                        []byte("bar"),
+				"app1.tailnetxyz.ts.net.crt": []byte(testCert + "1"),
+				"app1.tailnetxyz.ts.net.key": []byte(testKey + "1"),
+				"app2.tailnetxyz.ts.net.crt": []byte(testCert + "2"),
+				"app2.tailnetxyz.ts.net.key": []byte(testKey + "2"),
+			},
+		},
+		{
+			name:     "load_select_certs_in_read_write_mode",
+			certMode: "rw",
+			stateSecretContents: map[string][]byte{
+				"foo": []byte("bar"),
+			},
+			TLSSecrets: []kubeapi.Secret{
+				makeSecret("app1.tailnetxyz.ts.net", certSecretsLabels, "1"),
+				makeSecret("app2.tailnetxyz.ts.net", certSecretsLabels, "2"),
+				makeSecret("some-other-secret", nil, "3"),
+				makeSecret("app3.other-proxies.ts.net", map[string]string{
+					"tailscale.com/secret-type": "certs",
+					"tailscale.com/managed":     "true",
+					"tailscale.com/proxy-group": "some-other-proxygroup",
+				}, "4"),
+			},
+			wantMemoryStoreContents: map[ipn.StateKey][]byte{
+				"foo":                        []byte("bar"),
+				"app1.tailnetxyz.ts.net.crt": []byte(testCert + "1"),
+				"app1.tailnetxyz.ts.net.key": []byte(testKey + "1"),
+				"app2.tailnetxyz.ts.net.crt": []byte(testCert + "2"),
+				"app2.tailnetxyz.ts.net.key": []byte(testKey + "2"),
+			},
+		},
+		{
+			name:     "list_cert_secrets_fails",
+			certMode: "ro",
+			stateSecretContents: map[string][]byte{
+				"foo": []byte("bar"),
+			},
+			secretsListErr: fmt.Errorf("list error"),
+			// The error is logged but not returned, and state is still loaded
+			wantMemoryStoreContents: map[ipn.StateKey][]byte{
+				"foo": []byte("bar"),
+			},
+		},
+		{
+			name:     "cert_secrets_not_loaded_when_not_in_share_mode",
+			certMode: "",
+			stateSecretContents: map[string][]byte{
+				"foo": []byte("bar"),
+			},
+			TLSSecrets: []kubeapi.Secret{
+				makeSecret("app1.tailnetxyz.ts.net", certSecretsLabels, "1"),
+			},
+			wantMemoryStoreContents: map[ipn.StateKey][]byte{
+				"foo": []byte("bar"),
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			envknob.Setenv("TS_CERT_SHARE_MODE", tt.certMode)
+
+			t.Setenv("POD_NAME", "ingress-proxies-1")
+
+			client := &kubeclient.FakeClient{
+				GetSecretImpl: func(ctx context.Context, name string) (*kubeapi.Secret, error) {
+					if tt.secretGetErr != nil {
+						return nil, tt.secretGetErr
+					}
+					if name == secretName {
+						return &kubeapi.Secret{Data: tt.stateSecretContents}, nil
+					}
+					return nil, &kubeapi.Status{Code: 404}
+				},
+				CheckSecretPermissionsImpl: func(ctx context.Context, name string) (bool, bool, error) {
+					return true, true, nil
+				},
+				ListSecretsImpl: func(ctx context.Context, selector map[string]string) (*kubeapi.SecretList, error) {
+					if tt.secretsListErr != nil {
+						return nil, tt.secretsListErr
+					}
+					var matchingSecrets []kubeapi.Secret
+					for _, secret := range tt.TLSSecrets {
+						matches := true
+						for k, v := range selector {
+							if secret.Labels[k] != v {
+								matches = false
+								break
+							}
+						}
+						if matches {
+							matchingSecrets = append(matchingSecrets, secret)
+						}
+					}
+					return &kubeapi.SecretList{Items: matchingSecrets}, nil
+				},
+			}
+
+			s, err := newWithClient(t.Logf, client, secretName)
+			if tt.wantErr != nil {
+				if err == nil {
+					t.Errorf("NewWithClient() error = nil, want error containing %v", tt.wantErr)
+					return
+				}
+				if !strings.Contains(err.Error(), tt.wantErr.Error()) {
+					t.Errorf("NewWithClient() error = %v, want error containing %v", err, tt.wantErr)
+				}
+				return
+			}
+
+			if err != nil {
+				t.Errorf("NewWithClient() unexpected error: %v", err)
+				return
+			}
+
+			// Verify memory store contents
+			gotJSON, err := s.memory.ExportToJSON()
+			if err != nil {
+				t.Errorf("ExportToJSON failed: %v", err)
+				return
+			}
+			var got map[ipn.StateKey][]byte
+			if err := json.Unmarshal(gotJSON, &got); err != nil {
+				t.Errorf("failed to unmarshal memory store JSON: %v", err)
+				return
+			}
+			want := tt.wantMemoryStoreContents
+			if want == nil {
+				want = map[ipn.StateKey][]byte{}
+			}
+			if diff := cmp.Diff(got, want); diff != "" {
+				t.Errorf("memory store contents mismatch (-got +want):\n%s", diff)
+			}
+		})
+	}
+}

k8s-operator/api.md

@@ -517,6 +517,7 @@ _Appears in:_
 | `statefulSet` _[StatefulSet](#statefulset)_ | Configuration parameters for the proxy's StatefulSet. Tailscale<br />Kubernetes operator deploys a StatefulSet for each of the user<br />configured proxies (Tailscale Ingress, Tailscale Service, Connector). |  |  |
 | `metrics` _[Metrics](#metrics)_ | Configuration for proxy metrics. Metrics are currently not supported<br />for egress proxies and for Ingress proxies that have been configured<br />with tailscale.com/experimental-forward-cluster-traffic-via-ingress<br />annotation. Note that the metrics are currently considered unstable<br />and will likely change in breaking ways in the future - we only<br />recommend that you use those for debugging purposes. |  |  |
 | `tailscale` _[TailscaleConfig](#tailscaleconfig)_ | TailscaleConfig contains options to configure the tailscale-specific<br />parameters of proxies. |  |  |
+| `useLetsEncryptStagingEnvironment` _boolean_ | Set UseLetsEncryptStagingEnvironment to true to issue TLS<br />certificates for any HTTPS endpoints exposed to the tailnet from<br />LetsEncrypt's staging environment.<br />https://letsencrypt.org/docs/staging-environment/<br />This setting only affects Tailscale Ingress resources.<br />By default Ingress TLS certificates are issued from LetsEncrypt's<br />production environment.<br />Changing this setting true -> false, will result in any<br />existing certs being re-issued from the production environment.<br />Changing this setting false (default) -> true, when certs have already<br />been provisioned from production environment will NOT result in certs<br />being re-issued from the staging environment before they need to be<br />renewed. |  |  |
 
 
 #### ProxyClassStatus
@@ -599,7 +600,7 @@ _Appears in:_
 
 | Field | Description | Default | Validation |
 | --- | --- | --- | --- |
-| `type` _[ProxyGroupType](#proxygrouptype)_ | Type of the ProxyGroup proxies. Supported types are egress and ingress.<br />Type is immutable once a ProxyGroup is created. |  | Enum: [egress ingress] <br />Type: string <br /> |
+| `type` _[ProxyGroupType](#proxygrouptype)_ | Type of the ProxyGroup proxies. Currently the only supported type is egress.<br />Type is immutable once a ProxyGroup is created. |  | Enum: [egress ingress] <br />Type: string <br /> |
 | `tags` _[Tags](#tags)_ | Tags that the Tailscale devices will be tagged with. Defaults to [tag:k8s].<br />If you specify custom tags here, make sure you also make the operator<br />an owner of these tags.<br />See  https://tailscale.com/kb/1236/kubernetes-operator/#setting-up-the-kubernetes-operator.<br />Tags cannot be changed once a ProxyGroup device has been created.<br />Tag values must be in form ^tag:[a-zA-Z][a-zA-Z0-9-]*$. |  | Pattern: `^tag:[a-zA-Z][a-zA-Z0-9-]*$` <br />Type: string <br /> |
 | `replicas` _integer_ | Replicas specifies how many replicas to create the StatefulSet with.<br />Defaults to 2. |  | Minimum: 0 <br /> |
 | `hostnamePrefix` _[HostnamePrefix](#hostnameprefix)_ | HostnamePrefix is the hostname prefix to use for tailnet devices created<br />by the ProxyGroup. Each device will have the integer number from its<br />StatefulSet pod appended to this prefix to form the full hostname.<br />HostnamePrefix can contain lower case letters, numbers and dashes, it<br />must not start with a dash and must be between 1 and 62 characters long. |  | Pattern: `^[a-z0-9][a-z0-9-]{0,61}$` <br />Type: string <br /> |

k8s-operator/apis/v1alpha1/types_proxyclass.go

@@ -66,6 +66,21 @@ type ProxyClassSpec struct {
 	// parameters of proxies.
 	// +optional
 	TailscaleConfig *TailscaleConfig `json:"tailscale,omitempty"`
+	// Set UseLetsEncryptStagingEnvironment to true to issue TLS
+	// certificates for any HTTPS endpoints exposed to the tailnet from
+	// LetsEncrypt's staging environment.
+	// https://letsencrypt.org/docs/staging-environment/
+	// This setting only affects Tailscale Ingress resources.
+	// By default Ingress TLS certificates are issued from LetsEncrypt's
+	// production environment.
+	// Changing this setting true -> false, will result in any
+	// existing certs being re-issued from the production environment.
+	// Changing this setting false (default) -> true, when certs have already
+	// been provisioned from production environment will NOT result in certs
+	// being re-issued from the staging environment before they need to be
+	// renewed.
+	// +optional
+	UseLetsEncryptStagingEnvironment bool `json:"useLetsEncryptStagingEnvironment,omitempty"`
 }
 
 type TailscaleConfig struct {

k8s-operator/apis/v1alpha1/types_proxygroup.go

@@ -48,7 +48,7 @@ type ProxyGroupList struct {
 }
 
 type ProxyGroupSpec struct {
-	// Type of the ProxyGroup proxies. Supported types are egress and ingress.
+	// Type of the ProxyGroup proxies. Currently the only supported type is egress.
 	// Type is immutable once a ProxyGroup is created.
 	// +kubebuilder:validation:XValidation:rule="self == oldSelf",message="ProxyGroup type is immutable"
 	Type ProxyGroupType `json:"type"`

kube/kubeapi/api.go

@@ -153,6 +153,14 @@ type Secret struct {
 	Data map[string][]byte `json:"data,omitempty"`
 }
 
+// SecretList is a list of Secret objects.
+type SecretList struct {
+	TypeMeta   `json:",inline"`
+	ObjectMeta `json:"metadata"`
+
+	Items []Secret `json:"items,omitempty"`
+}
+
 // Event contains a subset of fields from corev1.Event.
 // https://github.com/kubernetes/api/blob/6cc44b8953ae704d6d9ec2adf32e7ae19199ea9f/core/v1/types.go#L7034
 // It is copied here to avoid having to import kube libraries.

kube/kubeclient/client.go

@@ -60,6 +60,7 @@ func readFile(n string) ([]byte, error) {
 // It expects to be run inside a cluster.
 type Client interface {
 	GetSecret(context.Context, string) (*kubeapi.Secret, error)
+	ListSecrets(context.Context, map[string]string) (*kubeapi.SecretList, error)
 	UpdateSecret(context.Context, *kubeapi.Secret) error
 	CreateSecret(context.Context, *kubeapi.Secret) error
 	// Event attempts to ensure an event with the specified options associated with the Pod in which we are
@@ -248,21 +249,35 @@ func (c *client) newRequest(ctx context.Context, method, url string, in any) (*h
 // GetSecret fetches the secret from the Kubernetes API.
 func (c *client) GetSecret(ctx context.Context, name string) (*kubeapi.Secret, error) {
 	s := &kubeapi.Secret{Data: make(map[string][]byte)}
-	if err := c.kubeAPIRequest(ctx, "GET", c.resourceURL(name, TypeSecrets), nil, s); err != nil {
+	if err := c.kubeAPIRequest(ctx, "GET", c.resourceURL(name, TypeSecrets, ""), nil, s); err != nil {
 		return nil, err
 	}
 	return s, nil
 }
 
+// ListSecrets fetches the secret from the Kubernetes API.
+func (c *client) ListSecrets(ctx context.Context, selector map[string]string) (*kubeapi.SecretList, error) {
+	sl := new(kubeapi.SecretList)
+	s := make([]string, 0, len(selector))
+	for key, val := range selector {
+		s = append(s, key+"="+url.QueryEscape(val))
+	}
+	ss := strings.Join(s, ",")
+	if err := c.kubeAPIRequest(ctx, "GET", c.resourceURL("", TypeSecrets, ss), nil, sl); err != nil {
+		return nil, err
+	}
+	return sl, nil
+}
+
 // CreateSecret creates a secret in the Kubernetes API.
 func (c *client) CreateSecret(ctx context.Context, s *kubeapi.Secret) error {
 	s.Namespace = c.ns
-	return c.kubeAPIRequest(ctx, "POST", c.resourceURL("", TypeSecrets), s, nil)
+	return c.kubeAPIRequest(ctx, "POST", c.resourceURL("", TypeSecrets, ""), s, nil)
 }
 
 // UpdateSecret updates a secret in the Kubernetes API.
 func (c *client) UpdateSecret(ctx context.Context, s *kubeapi.Secret) error {
-	return c.kubeAPIRequest(ctx, "PUT", c.resourceURL(s.Name, TypeSecrets), s, nil)
+	return c.kubeAPIRequest(ctx, "PUT", c.resourceURL(s.Name, TypeSecrets, ""), s, nil)
 }
 
 // JSONPatch is a JSON patch operation.
@@ -283,14 +298,14 @@ func (c *client) JSONPatchResource(ctx context.Context, name, typ string, patche
 			return fmt.Errorf("unsupported JSON patch operation: %q", p.Op)
 		}
 	}
-	return c.kubeAPIRequest(ctx, "PATCH", c.resourceURL(name, typ), patches, nil, setHeader("Content-Type", "application/json-patch+json"))
+	return c.kubeAPIRequest(ctx, "PATCH", c.resourceURL(name, typ, ""), patches, nil, setHeader("Content-Type", "application/json-patch+json"))
 }
 
 // StrategicMergePatchSecret updates a secret in the Kubernetes API using a
 // strategic merge patch.
 // If a fieldManager is provided, it will be used to track the patch.
 func (c *client) StrategicMergePatchSecret(ctx context.Context, name string, s *kubeapi.Secret, fieldManager string) error {
-	surl := c.resourceURL(name, TypeSecrets)
+	surl := c.resourceURL(name, TypeSecrets, "")
 	if fieldManager != "" {
 		uv := url.Values{
 			"fieldManager": {fieldManager},
@@ -342,7 +357,7 @@ func (c *client) Event(ctx context.Context, typ, reason, msg string) error {
 			LastTimestamp:  now,
 			Count:          1,
 		}
-		return c.kubeAPIRequest(ctx, "POST", c.resourceURL("", typeEvents), &ev, nil)
+		return c.kubeAPIRequest(ctx, "POST", c.resourceURL("", typeEvents, ""), &ev, nil)
 	}
 	// If the Event already exists, we patch its count and last timestamp. This ensures that when users run 'kubectl
 	// describe pod...', they see the event just once (but with a message of how many times it has appeared over
@@ -472,9 +487,13 @@ func (c *client) checkPermission(ctx context.Context, verb, typ, name string) (b
 // resourceURL returns a URL that can be used to interact with the given resource type and, if name is not empty string,
 // the named resource of that type.
 // Note that this only works for core/v1 resource types.
-func (c *client) resourceURL(name, typ string) string {
+func (c *client) resourceURL(name, typ, sel string) string {
 	if name == "" {
-		return fmt.Sprintf("%s/api/v1/namespaces/%s/%s", c.url, c.ns, typ)
+		url := fmt.Sprintf("%s/api/v1/namespaces/%s/%s", c.url, c.ns, typ)
+		if sel != "" {
+			url += "?labelSelector=" + sel
+		}
+		return url
 	}
 	return fmt.Sprintf("%s/api/v1/namespaces/%s/%s/%s", c.url, c.ns, typ, name)
 }
@@ -487,7 +506,7 @@ func (c *client) nameForEvent(reason string) string {
 // getEvent fetches the event from the Kubernetes API.
 func (c *client) getEvent(ctx context.Context, name string) (*kubeapi.Event, error) {
 	e := &kubeapi.Event{}
-	if err := c.kubeAPIRequest(ctx, "GET", c.resourceURL(name, typeEvents), nil, e); err != nil {
+	if err := c.kubeAPIRequest(ctx, "GET", c.resourceURL(name, typeEvents, ""), nil, e); err != nil {
 		return nil, err
 	}
 	return e, nil

kube/kubeclient/fake_client.go

@@ -18,6 +18,7 @@ type FakeClient struct {
 	CreateSecretImpl           func(context.Context, *kubeapi.Secret) error
 	UpdateSecretImpl           func(context.Context, *kubeapi.Secret) error
 	JSONPatchResourceImpl      func(context.Context, string, string, []JSONPatch) error
+	ListSecretsImpl            func(context.Context, map[string]string) (*kubeapi.SecretList, error)
 }
 
 func (fc *FakeClient) CheckSecretPermissions(ctx context.Context, name string) (bool, bool, error) {
@@ -45,3 +46,9 @@ func (fc *FakeClient) UpdateSecret(ctx context.Context, secret *kubeapi.Secret)
 func (fc *FakeClient) CreateSecret(ctx context.Context, secret *kubeapi.Secret) error {
 	return fc.CreateSecretImpl(ctx, secret)
 }
+func (fc *FakeClient) ListSecrets(ctx context.Context, selector map[string]string) (*kubeapi.SecretList, error) {
+	if fc.ListSecretsImpl != nil {
+		return fc.ListSecretsImpl(ctx, selector)
+	}
+	return nil, nil
+}

kube/kubetypes/types.go

@@ -48,4 +48,7 @@ const (
 	PodIPv4Header string = "Pod-IPv4"
 
 	EgessServicesPreshutdownEP = "/internal-egress-services-preshutdown"
+
+	LabelManaged    = "tailscale.com/managed"
+	LabelSecretType = "tailscale.com/secret-type" // "config", "state" "certs"
 )

net/dns/manager.go

@@ -35,6 +35,9 @@ import (
 
 var (
 	errFullQueue = errors.New("request queue full")
+	// ErrNoDNSConfig is returned by RecompileDNSConfig when the Manager
+	// has no existing DNS configuration.
+	ErrNoDNSConfig = errors.New("no DNS configuration")
 )
 
 // maxActiveQueries returns the maximal number of DNS requests that can
@@ -91,21 +94,18 @@ func NewManager(logf logger.Logf, oscfg OSConfigurator, health *health.Tracker,
 	}
 
 	// Rate limit our attempts to correct our DNS configuration.
+	// This is done on incoming queries, we don't want to spam it.
 	limiter := rate.NewLimiter(1.0/5.0, 1)
 
 	// This will recompile the DNS config, which in turn will requery the system
 	// DNS settings. The recovery func should triggered only when we are missing
 	// upstream nameservers and require them to forward a query.
 	m.resolver.SetMissingUpstreamRecovery(func() {
-		m.mu.Lock()
-		defer m.mu.Unlock()
-		if m.config == nil {
-			return
-		}
-
 		if limiter.Allow() {
-			m.logf("DNS resolution failed due to missing upstream nameservers.  Recompiling DNS configuration.")
-			m.setLocked(*m.config)
+			m.logf("resolution failed due to missing upstream nameservers.  Recompiling DNS configuration.")
+			if err := m.RecompileDNSConfig(); err != nil {
+				m.logf("config recompilation failed: %v", err)
+			}
 		}
 	})
 
@@ -117,6 +117,26 @@ func NewManager(logf logger.Logf, oscfg OSConfigurator, health *health.Tracker,
 // Resolver returns the Manager's DNS Resolver.
 func (m *Manager) Resolver() *resolver.Resolver { return m.resolver }
 
+// RecompileDNSConfig sets the DNS config to the current value, which has
+// the side effect of re-querying the OS's interface nameservers.  This should be used
+// on platforms where the interface nameservers can change.  Darwin, for example,
+// where the nameservers aren't always available when we process a major interface
+// change event, or platforms where the nameservers may change while tunnel is up.
+//
+// This should be called if it is determined that [OSConfigurator.GetBaseConfig] may
+// give a better or different result than when [Manager.Set] was last called.  The
+// logic for making that determination is up to the caller.
+//
+// It returns [ErrNoDNSConfig] if the [Manager] has no existing DNS configuration.
+func (m *Manager) RecompileDNSConfig() error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if m.config == nil {
+		return ErrNoDNSConfig
+	}
+	return m.setLocked(*m.config)
+}
+
 func (m *Manager) Set(cfg Config) error {
 	m.mu.Lock()
 	defer m.mu.Unlock()

net/netmon/interfaces_test.go

@@ -13,7 +13,7 @@ import (
 )
 
 func TestGetState(t *testing.T) {
-	st, err := GetState()
+	st, err := getState("")
 	if err != nil {
 		t.Fatal(err)
 	}

net/netmon/netmon.go

@@ -161,7 +161,7 @@ func (m *Monitor) InterfaceState() *State {
 }
 
 func (m *Monitor) interfaceStateUncached() (*State, error) {
-	return GetState()
+	return getState(m.tsIfName)
 }
 
 // SetTailscaleInterfaceName sets the name of the Tailscale interface. For

net/netmon/state.go

@@ -461,21 +461,22 @@ func isTailscaleInterface(name string, ips []netip.Prefix) bool {
 // getPAC, if non-nil, returns the current PAC file URL.
 var getPAC func() string
 
-// GetState returns the state of all the current machine's network interfaces.
+// getState returns the state of all the current machine's network interfaces.
 //
 // It does not set the returned State.IsExpensive. The caller can populate that.
 //
-// Deprecated: use netmon.Monitor.InterfaceState instead.
-func GetState() (*State, error) {
+// optTSInterfaceName is the name of the Tailscale interface, if known.
+func getState(optTSInterfaceName string) (*State, error) {
 	s := &State{
 		InterfaceIPs: make(map[string][]netip.Prefix),
 		Interface:    make(map[string]Interface),
 	}
 	if err := ForeachInterface(func(ni Interface, pfxs []netip.Prefix) {
+		isTSInterfaceName := optTSInterfaceName != "" && ni.Name == optTSInterfaceName
 		ifUp := ni.IsUp()
 		s.Interface[ni.Name] = ni
 		s.InterfaceIPs[ni.Name] = append(s.InterfaceIPs[ni.Name], pfxs...)
-		if !ifUp || isTailscaleInterface(ni.Name, pfxs) {
+		if !ifUp || isTSInterfaceName || isTailscaleInterface(ni.Name, pfxs) {
 			return
 		}
 		for _, pfx := range pfxs {
@@ -755,11 +756,12 @@ func DefaultRoute() (DefaultRouteDetails, error) {
 
 // HasCGNATInterface reports whether there are any non-Tailscale interfaces that
 // use a CGNAT IP range.
-func HasCGNATInterface() (bool, error) {
+func (m *Monitor) HasCGNATInterface() (bool, error) {
 	hasCGNATInterface := false
 	cgnatRange := tsaddr.CGNATRange()
 	err := ForeachInterface(func(i Interface, pfxs []netip.Prefix) {
-		if hasCGNATInterface || !i.IsUp() || isTailscaleInterface(i.Name, pfxs) {
+		isTSInterfaceName := m.tsIfName != "" && i.Name == m.tsIfName
+		if hasCGNATInterface || !i.IsUp() || isTSInterfaceName || isTailscaleInterface(i.Name, pfxs) {
 			return
 		}
 		for _, pfx := range pfxs {

pkgdoc_test.go

@@ -26,6 +26,9 @@ func TestPackageDocs(t *testing.T) {
 		if err != nil {
 			return err
 		}
+		if fi.Mode().IsDir() && path == ".git" {
+			return filepath.SkipDir // No documentation lives in .git
+		}
 		if fi.Mode().IsRegular() && strings.HasSuffix(path, ".go") {
 			if strings.HasSuffix(path, "_test.go") {
 				return nil

prober/derp.go

@@ -596,11 +596,23 @@ func (d *derpProber) updateMap(ctx context.Context) error {
 }
 
 func (d *derpProber) ProbeUDP(ipaddr string, port int) ProbeClass {
+	initLabels := make(Labels)
+	ip := net.ParseIP(ipaddr)
+
+	if ip.To4() != nil {
+		initLabels["address_family"] = "ipv4"
+	} else if ip.To16() != nil { // Will return an IPv4 as 16 byte, so ensure the check for IPv4 precedes this
+		initLabels["address_family"] = "ipv6"
+	} else {
+		initLabels["address_family"] = "unknown"
+	}
+
 	return ProbeClass{
 		Probe: func(ctx context.Context) error {
 			return derpProbeUDP(ctx, ipaddr, port)
 		},
-		Class: "derp_udp",
+		Class:  "derp_udp",
+		Labels: initLabels,
 	}
 }
 

prober/prober.go

@@ -404,10 +404,14 @@ func (p *Probe) recordEndLocked(err error) {
 		p.mSeconds.WithLabelValues("ok").Add(latency.Seconds())
 		p.latencyHist.Value = latency
 		p.latencyHist = p.latencyHist.Next()
+		p.mAttempts.WithLabelValues("fail").Add(0)
+		p.mSeconds.WithLabelValues("fail").Add(0)
 	} else {
 		p.latency = 0
 		p.mAttempts.WithLabelValues("fail").Inc()
 		p.mSeconds.WithLabelValues("fail").Add(latency.Seconds())
+		p.mAttempts.WithLabelValues("ok").Add(0)
+		p.mSeconds.WithLabelValues("ok").Add(0)
 	}
 	p.successHist.Value = p.succeeded
 	p.successHist = p.successHist.Next()

safesocket/safesocket.go

@@ -61,7 +61,11 @@ func ConnectContext(ctx context.Context, path string) (net.Conn, error) {
 			if ctx.Err() != nil {
 				return nil, ctx.Err()
 			}
-			time.Sleep(250 * time.Millisecond)
+			select {
+			case <-ctx.Done():
+				return nil, ctx.Err()
+			case <-time.After(250 * time.Millisecond):
+			}
 			continue
 		}
 		return c, err

tsnet/tsnet.go

@@ -505,6 +505,11 @@ func (s *Server) start() (reterr error) {
 			// directory and hostname when they're not supplied. But we can fall
 			// back to "tsnet" as well.
 			exe = "tsnet"
+		case "ios":
+			// When compiled as a framework (via TailscaleKit in libtailscale),
+			// os.Executable() returns an error, so fall back to "tsnet" there
+			// too.
+			exe = "tsnet"
 		default:
 			return err
 		}

tsnet/tsnet_test.go

@@ -120,6 +120,7 @@ func startControl(t *testing.T) (controlURL string, control *testcontrol.Server)
 			Proxied: true,
 		},
 		MagicDNSDomain: "tail-scale.ts.net",
+		Logf:           t.Logf,
 	}
 	control.HTTPTestServer = httptest.NewUnstartedServer(control)
 	control.HTTPTestServer.Start()
@@ -221,7 +222,7 @@ func startServer(t *testing.T, ctx context.Context, controlURL, hostname string)
 		getCertForTesting: testCertRoot.getCert,
 	}
 	if *verboseNodes {
-		s.Logf = log.Printf
+		s.Logf = t.Logf
 	}
 	t.Cleanup(func() { s.Close() })
 

tstest/integration/integration_test.go

@@ -1942,6 +1942,8 @@ func (n *testNode) AwaitIP6() netip.Addr {
 
 // AwaitRunning waits for n to reach the IPN state "Running".
 func (n *testNode) AwaitRunning() {
+	t := n.env.t
+	t.Helper()
 	n.AwaitBackendState("Running")
 }
 
@@ -2015,7 +2017,7 @@ func (n *testNode) Status() (*ipnstate.Status, error) {
 	}
 	st := new(ipnstate.Status)
 	if err := json.Unmarshal(out, st); err != nil {
-		return nil, fmt.Errorf("decoding tailscale status JSON: %w", err)
+		return nil, fmt.Errorf("decoding tailscale status JSON: %w\njson:\n%s", err, out)
 	}
 	return st, nil
 }

tstest/integration/testcontrol/testcontrol.go

@@ -839,15 +839,17 @@ func (s *Server) serveMap(w http.ResponseWriter, r *http.Request, mkey key.Machi
 
 	w.WriteHeader(200)
 	for {
-		if resBytes, ok := s.takeRawMapMessage(req.NodeKey); ok {
-			if err := s.sendMapMsg(w, compress, resBytes); err != nil {
-				s.logf("sendMapMsg of raw message: %v", err)
-				return
-			}
-			if streaming {
+		// Only send raw map responses to the streaming poll, to avoid a
+		// non-streaming map request beating the streaming poll in a race and
+		// potentially dropping the map response.
+		if streaming {
+			if resBytes, ok := s.takeRawMapMessage(req.NodeKey); ok {
+				if err := s.sendMapMsg(w, compress, resBytes); err != nil {
+					s.logf("sendMapMsg of raw message: %v", err)
+					return
+				}
 				continue
 			}
-			return
 		}
 
 		if s.canGenerateAutomaticMapResponseFor(req.NodeKey) {

tstest/mts/mts.go

@@ -0,0 +1,599 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build linux || darwin
+
+// The mts ("Multiple Tailscale") command runs multiple tailscaled instances for
+// development, managing their directories and sockets, and lets you easily direct
+// tailscale CLI commands to them.
+package main
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"io"
+	"log"
+	"maps"
+	"net"
+	"net/http"
+	"net/url"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"regexp"
+	"slices"
+	"strings"
+	"sync"
+	"syscall"
+	"time"
+
+	"tailscale.com/client/local"
+	"tailscale.com/types/bools"
+	"tailscale.com/types/lazy"
+	"tailscale.com/util/mak"
+)
+
+func usage(args ...any) {
+	var format string
+	if len(args) > 0 {
+		format, args = args[0].(string), args[1:]
+	}
+	if format != "" {
+		format = strings.TrimSpace(format) + "\n\n"
+		fmt.Fprintf(os.Stderr, format, args...)
+	}
+	io.WriteString(os.Stderr, strings.TrimSpace(`
+usage:
+
+   mts server <subcommand>      # manage tailscaled instances
+   mts server run               # run the mts server (parent process of all tailscaled)
+   mts server list              # list all tailscaled and their state
+   mts server list <name>       # show details of named instance
+   mts server add <name>        # add+start new named tailscaled
+   mts server start <name>      # start a previously added tailscaled
+   mts server stop <name>       # stop & remove a named tailscaled
+   mts server rm <name>         # stop & remove a named tailscaled
+   mts server logs [-f] <name>  # get/follow tailscaled logs
+
+  mts <inst-name> [tailscale CLI args] # run Tailscale CLI against a named instance
+    e.g.
+      mts gmail1 up
+      mts github2 status --json
+	`)+"\n")
+	os.Exit(1)
+}
+
+func main() {
+	// Don't use flag.Parse here; we mostly just delegate through
+	// to the Tailscale CLI.
+
+	if len(os.Args) < 2 {
+		usage()
+	}
+	firstArg, args := os.Args[1], os.Args[2:]
+	if firstArg == "server" || firstArg == "s" {
+		if err := runMTSServer(args); err != nil {
+			log.Fatal(err)
+		}
+	} else {
+		var c Client
+		inst := firstArg
+		c.RunCommand(inst, args)
+	}
+}
+
+func runMTSServer(args []string) error {
+	if len(args) == 0 {
+		usage()
+	}
+	cmd, args := args[0], args[1:]
+	if cmd == "run" {
+		var s Server
+		return s.Run()
+	}
+
+	// Commands other than "run" all use the HTTP client to
+	// hit the mts server over its unix socket.
+	var c Client
+
+	switch cmd {
+	default:
+		usage("unknown mts server subcommand %q", cmd)
+	case "list", "ls":
+		list, err := c.List()
+		if err != nil {
+			return err
+		}
+		if len(args) == 0 {
+			names := slices.Sorted(maps.Keys(list.Instances))
+			for _, name := range names {
+				running := list.Instances[name].Running
+				fmt.Printf("%10s %s\n", bools.IfElse(running, "RUNNING", "stopped"), name)
+			}
+		} else {
+			for _, name := range args {
+				inst, ok := list.Instances[name]
+				if !ok {
+					return fmt.Errorf("no instance named %q", name)
+				}
+				je := json.NewEncoder(os.Stdout)
+				je.SetIndent("", "  ")
+				if err := je.Encode(inst); err != nil {
+					return err
+				}
+			}
+		}
+
+	case "rm":
+		if len(args) == 0 {
+			return fmt.Errorf("missing instance name(s) to remove")
+		}
+		log.SetFlags(0)
+		for _, name := range args {
+			ok, err := c.Remove(name)
+			if err != nil {
+				return err
+			}
+			if ok {
+				log.Printf("%s deleted.", name)
+			} else {
+				log.Printf("%s didn't exist.", name)
+			}
+		}
+	case "stop":
+		if len(args) == 0 {
+			return fmt.Errorf("missing instance name(s) to stop")
+		}
+		log.SetFlags(0)
+		for _, name := range args {
+			ok, err := c.Stop(name)
+			if err != nil {
+				return err
+			}
+			if ok {
+				log.Printf("%s stopped.", name)
+			} else {
+				log.Printf("%s didn't exist.", name)
+			}
+		}
+	case "start", "restart":
+		list, err := c.List()
+		if err != nil {
+			return err
+		}
+		shouldStop := cmd == "restart"
+		for _, arg := range args {
+			is, ok := list.Instances[arg]
+			if !ok {
+				return fmt.Errorf("no instance named %q", arg)
+			}
+			if is.Running {
+				if shouldStop {
+					if _, err := c.Stop(arg); err != nil {
+						return fmt.Errorf("stopping %q: %w", arg, err)
+					}
+				} else {
+					log.SetFlags(0)
+					log.Printf("%s already running.", arg)
+					continue
+				}
+			}
+			// Creating an existing one starts it up.
+			if err := c.Create(arg); err != nil {
+				return fmt.Errorf("starting %q: %w", arg, err)
+			}
+		}
+	case "add":
+		if len(args) == 0 {
+			return fmt.Errorf("missing instance name(s) to add")
+		}
+		for _, name := range args {
+			if err := c.Create(name); err != nil {
+				return fmt.Errorf("creating %q: %w", name, err)
+			}
+		}
+	case "logs":
+		fs := flag.NewFlagSet("logs", flag.ExitOnError)
+		fs.Usage = func() { usage() }
+		follow := fs.Bool("f", false, "follow logs")
+		fs.Parse(args)
+		log.Printf("Parsed; following=%v, args=%q", *follow, fs.Args())
+		if fs.NArg() != 1 {
+			usage()
+		}
+		cmd := bools.IfElse(*follow, "tail", "cat")
+		args := []string{cmd}
+		if *follow {
+			args = append(args, "-f")
+		}
+		path, err := exec.LookPath(cmd)
+		if err != nil {
+			return fmt.Errorf("looking up %q: %w", cmd, err)
+		}
+		args = append(args, instLogsFile(fs.Arg(0)))
+		log.Fatal(syscall.Exec(path, args, os.Environ()))
+	}
+	return nil
+}
+
+type Client struct {
+}
+
+func (c *Client) client() *http.Client {
+	return &http.Client{
+		Transport: &http.Transport{
+			DialContext: func(ctx context.Context, network, addr string) (net.Conn, error) {
+				return net.Dial("unix", mtsSock())
+			},
+		},
+	}
+}
+
+func getJSON[T any](res *http.Response, err error) (T, error) {
+	var ret T
+	if err != nil {
+		return ret, err
+	}
+	defer res.Body.Close()
+	if res.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(res.Body)
+		return ret, fmt.Errorf("unexpected status: %v: %s", res.Status, body)
+	}
+	if err := json.NewDecoder(res.Body).Decode(&ret); err != nil {
+		return ret, err
+	}
+	return ret, nil
+}
+
+func (c *Client) List() (listResponse, error) {
+	return getJSON[listResponse](c.client().Get("http://mts/list"))
+}
+
+func (c *Client) Remove(name string) (found bool, err error) {
+	return getJSON[bool](c.client().PostForm("http://mts/rm", url.Values{
+		"name": []string{name},
+	}))
+}
+
+func (c *Client) Stop(name string) (found bool, err error) {
+	return getJSON[bool](c.client().PostForm("http://mts/stop", url.Values{
+		"name": []string{name},
+	}))
+}
+
+func (c *Client) Create(name string) error {
+	req, err := http.NewRequest("POST", "http://mts/create/"+name, nil)
+	if err != nil {
+		return err
+	}
+	resp, err := c.client().Do(req)
+	if err != nil {
+		return err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(resp.Body)
+		return fmt.Errorf("unexpected status: %v: %s", resp.Status, body)
+	}
+	return nil
+}
+
+func (c *Client) RunCommand(name string, args []string) {
+	sock := instSock(name)
+	lc := &local.Client{
+		Socket:        sock,
+		UseSocketOnly: true,
+	}
+	probeCtx, cancel := context.WithTimeout(context.Background(), 250*time.Millisecond)
+	defer cancel()
+	if _, err := lc.StatusWithoutPeers(probeCtx); err != nil {
+		log.Fatalf("instance %q not running? start with 'mts server start %q'; got error: %v", name, name, err)
+	}
+	args = append([]string{"run", "tailscale.com/cmd/tailscale", "--socket=" + sock}, args...)
+	cmd := exec.Command("go", args...)
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+	cmd.Stdin = os.Stdin
+	err := cmd.Run()
+	if err == nil {
+		os.Exit(0)
+	}
+	if exitErr, ok := err.(*exec.ExitError); ok {
+		os.Exit(exitErr.ExitCode())
+	}
+	panic(err)
+}
+
+type Server struct {
+	lazyTailscaled lazy.GValue[string]
+
+	mu   sync.Mutex
+	cmds map[string]*exec.Cmd // running tailscaled instances
+}
+
+func (s *Server) tailscaled() string {
+	v, err := s.lazyTailscaled.GetErr(func() (string, error) {
+		out, err := exec.Command("go", "list", "-f", "{{.Target}}", "tailscale.com/cmd/tailscaled").CombinedOutput()
+		if err != nil {
+			return "", err
+		}
+		return strings.TrimSpace(string(out)), nil
+	})
+	if err != nil {
+		panic(err)
+	}
+	return v
+}
+
+func (s *Server) Run() error {
+	if err := os.MkdirAll(mtsRoot(), 0700); err != nil {
+		return err
+	}
+	sock := mtsSock()
+	os.Remove(sock)
+	log.Printf("Multi-Tailscaled Server running; listening on %q ...", sock)
+	ln, err := net.Listen("unix", sock)
+	if err != nil {
+		return err
+	}
+	return http.Serve(ln, s)
+}
+
+var validNameRx = regexp.MustCompile(`^[a-zA-Z0-9_-]+$`)
+
+func validInstanceName(name string) bool {
+	return validNameRx.MatchString(name)
+}
+
+func (s *Server) InstanceRunning(name string) bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	_, ok := s.cmds[name]
+	return ok
+}
+
+func (s *Server) Stop(name string) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if cmd, ok := s.cmds[name]; ok {
+		if err := cmd.Process.Kill(); err != nil {
+			log.Printf("error killing %q: %v", name, err)
+		}
+		delete(s.cmds, name)
+	}
+}
+
+func (s *Server) RunInstance(name string) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if _, ok := s.cmds[name]; ok {
+		return fmt.Errorf("instance %q already running", name)
+	}
+
+	if !validInstanceName(name) {
+		return fmt.Errorf("invalid instance name %q", name)
+	}
+	dir := filepath.Join(mtsRoot(), name)
+	if err := os.MkdirAll(dir, 0700); err != nil {
+		return err
+	}
+
+	env := os.Environ()
+	env = append(env, "TS_DEBUG_LOG_RATE=all")
+	if ef, err := os.Open(instEnvFile(name)); err == nil {
+		defer ef.Close()
+		sc := bufio.NewScanner(ef)
+		for sc.Scan() {
+			t := strings.TrimSpace(sc.Text())
+			if strings.HasPrefix(t, "#") || !strings.Contains(t, "=") {
+				continue
+			}
+			env = append(env, t)
+		}
+	} else if os.IsNotExist(err) {
+		// Write an example one.
+		os.WriteFile(instEnvFile(name), fmt.Appendf(nil, "# Example mts env.txt file; uncomment/add stuff you want for %q\n\n#TS_DEBUG_MAP=1\n#TS_DEBUG_REGISTER=1\n#TS_NO_LOGS_NO_SUPPORT=1\n", name), 0600)
+	}
+
+	extraArgs := []string{"--verbose=1"}
+	if af, err := os.Open(instArgsFile(name)); err == nil {
+		extraArgs = nil // clear default args
+		defer af.Close()
+		sc := bufio.NewScanner(af)
+		for sc.Scan() {
+			t := strings.TrimSpace(sc.Text())
+			if strings.HasPrefix(t, "#") || t == "" {
+				continue
+			}
+			extraArgs = append(extraArgs, t)
+		}
+	} else if os.IsNotExist(err) {
+		// Write an example one.
+		os.WriteFile(instArgsFile(name), fmt.Appendf(nil, "# Example mts args.txt file for instance %q.\n# One line per extra arg to tailscaled; no magic string quoting\n\n--verbose=1\n#--socks5-server=127.0.0.1:5000\n", name), 0600)
+	}
+
+	log.Printf("Running Tailscale daemon %q in %q", name, dir)
+
+	args := []string{
+		"--tun=userspace-networking",
+		"--statedir=" + filepath.Join(dir),
+		"--socket=" + filepath.Join(dir, "tailscaled.sock"),
+	}
+	args = append(args, extraArgs...)
+
+	cmd := exec.Command(s.tailscaled(), args...)
+	cmd.Dir = dir
+	cmd.Env = env
+
+	out, err := cmd.StdoutPipe()
+	if err != nil {
+		return err
+	}
+	cmd.Stderr = cmd.Stdout
+
+	logs := instLogsFile(name)
+	logFile, err := os.OpenFile(logs, os.O_CREATE|os.O_WRONLY|os.O_APPEND|os.O_TRUNC, 0644)
+	if err != nil {
+		return fmt.Errorf("opening logs file: %w", err)
+	}
+
+	go func() {
+		bs := bufio.NewScanner(out)
+		for bs.Scan() {
+			// TODO(bradfitz): record in memory too, serve via HTTP
+			line := strings.TrimSpace(bs.Text())
+			fmt.Fprintf(logFile, "%s\n", line)
+			fmt.Printf("tailscaled[%s]: %s\n", name, line)
+		}
+	}()
+
+	if err := cmd.Start(); err != nil {
+		return err
+	}
+	go func() {
+		err := cmd.Wait()
+		logFile.Close()
+		log.Printf("Tailscale daemon %q exited: %v", name, err)
+		s.mu.Lock()
+		defer s.mu.Unlock()
+		delete(s.cmds, name)
+	}()
+
+	mak.Set(&s.cmds, name, cmd)
+	return nil
+}
+
+type listResponse struct {
+	// Instances maps instance name to its details.
+	Instances map[string]listResponseInstance `json:"instances"`
+}
+
+type listResponseInstance struct {
+	Name    string `json:"name"`
+	Dir     string `json:"dir"`
+	Sock    string `json:"sock"`
+	Running bool   `json:"running"`
+	Env     string `json:"env"`
+	Args    string `json:"args"`
+	Logs    string `json:"logs"`
+}
+
+func writeJSON(w http.ResponseWriter, v any) {
+	w.Header().Set("Content-Type", "application/json")
+	e := json.NewEncoder(w)
+	e.SetIndent("", "  ")
+	e.Encode(v)
+}
+
+func (s *Server) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	if r.URL.Path == "/list" {
+		var res listResponse
+		for _, name := range s.InstanceNames() {
+			mak.Set(&res.Instances, name, listResponseInstance{
+				Name:    name,
+				Dir:     instDir(name),
+				Sock:    instSock(name),
+				Running: s.InstanceRunning(name),
+				Env:     instEnvFile(name),
+				Args:    instArgsFile(name),
+				Logs:    instLogsFile(name),
+			})
+		}
+		writeJSON(w, res)
+		return
+	}
+	if r.URL.Path == "/rm" || r.URL.Path == "/stop" {
+		shouldRemove := r.URL.Path == "/rm"
+		if r.Method != "POST" {
+			http.Error(w, "POST required", http.StatusMethodNotAllowed)
+			return
+		}
+		target := r.FormValue("name")
+		var ok bool
+		for _, name := range s.InstanceNames() {
+			if name != target {
+				continue
+			}
+			ok = true
+			s.Stop(name)
+			if shouldRemove {
+				if err := os.RemoveAll(instDir(name)); err != nil {
+					http.Error(w, err.Error(), http.StatusInternalServerError)
+					return
+				}
+			}
+			break
+		}
+		writeJSON(w, ok)
+		return
+	}
+	if inst, ok := strings.CutPrefix(r.URL.Path, "/create/"); ok {
+		if !s.InstanceRunning(inst) {
+			if err := s.RunInstance(inst); err != nil {
+				http.Error(w, err.Error(), http.StatusInternalServerError)
+				return
+			}
+		}
+		fmt.Fprintf(w, "OK\n")
+		return
+	}
+	if r.URL.Path == "/" {
+		fmt.Fprintf(w, "This is mts, the multi-tailscaled server.\n")
+		return
+	}
+	http.NotFound(w, r)
+}
+
+func (s *Server) InstanceNames() []string {
+	var ret []string
+	des, err := os.ReadDir(mtsRoot())
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil
+		}
+		panic(err)
+	}
+	for _, de := range des {
+		if !de.IsDir() {
+			continue
+		}
+		ret = append(ret, de.Name())
+	}
+	return ret
+}
+
+func mtsRoot() string {
+	dir, err := os.UserConfigDir()
+	if err != nil {
+		panic(err)
+	}
+	return filepath.Join(dir, "multi-tailscale-dev")
+}
+
+func instDir(name string) string {
+	return filepath.Join(mtsRoot(), name)
+}
+
+func instSock(name string) string {
+	return filepath.Join(instDir(name), "tailscaled.sock")
+}
+
+func instEnvFile(name string) string {
+	return filepath.Join(mtsRoot(), name, "env.txt")
+}
+
+func instArgsFile(name string) string {
+	return filepath.Join(mtsRoot(), name, "args.txt")
+}
+
+func instLogsFile(name string) string {
+	return filepath.Join(mtsRoot(), name, "logs.txt")
+}
+
+func mtsSock() string {
+	return filepath.Join(mtsRoot(), "mts.sock")
+}

tsweb/debug.go

@@ -14,7 +14,7 @@ import (
 	"os"
 	"runtime"
 
-	"tailscale.com/tsweb/promvarz"
+	"tailscale.com/feature"
 	"tailscale.com/tsweb/varz"
 	"tailscale.com/version"
 )
@@ -37,6 +37,11 @@ type DebugHandler struct {
 	title    string                           // title displayed on index page
 }
 
+// PrometheusHandler is an optional hook to enable native Prometheus
+// support in the debug handler. It is disabled by default. Import the
+// tailscale.com/tsweb/promvarz package to enable this feature.
+var PrometheusHandler feature.Hook[func(*DebugHandler)]
+
 // Debugger returns the DebugHandler registered on mux at /debug/,
 // creating it if necessary.
 func Debugger(mux *http.ServeMux) *DebugHandler {
@@ -53,7 +58,11 @@ func Debugger(mux *http.ServeMux) *DebugHandler {
 	ret.KVFunc("Uptime", func() any { return varz.Uptime() })
 	ret.KV("Version", version.Long())
 	ret.Handle("vars", "Metrics (Go)", expvar.Handler())
-	ret.Handle("varz", "Metrics (Prometheus)", http.HandlerFunc(promvarz.Handler))
+	if PrometheusHandler.IsSet() {
+		PrometheusHandler.Get()(ret)
+	} else {
+		ret.Handle("varz", "Metrics (Prometheus)", http.HandlerFunc(varz.Handler))
+	}
 
 	// pprof.Index serves everything that runtime/pprof.Lookup finds:
 	// goroutine, threadcreate, heap, allocs, block, mutex

tsweb/promvarz/promvarz.go

@@ -11,12 +11,21 @@ import (
 
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/common/expfmt"
+	"tailscale.com/tsweb"
 	"tailscale.com/tsweb/varz"
 )
 
-// Handler returns Prometheus metrics exported by our expvar converter
+func init() {
+	tsweb.PrometheusHandler.Set(registerVarz)
+}
+
+func registerVarz(debug *tsweb.DebugHandler) {
+	debug.Handle("varz", "Metrics (Prometheus)", http.HandlerFunc(handler))
+}
+
+// handler returns Prometheus metrics exported by our expvar converter
 // and the official Prometheus client.
-func Handler(w http.ResponseWriter, r *http.Request) {
+func handler(w http.ResponseWriter, r *http.Request) {
 	if err := gatherNativePrometheusMetrics(w); err != nil {
 		w.WriteHeader(http.StatusInternalServerError)
 		w.Write([]byte(err.Error()))

tsweb/promvarz/promvarz_test.go

@@ -23,7 +23,7 @@ func TestHandler(t *testing.T) {
 	testVar1.Set(42)
 	testVar2.Set(4242)
 
-	svr := httptest.NewServer(http.HandlerFunc(Handler))
+	svr := httptest.NewServer(http.HandlerFunc(handler))
 	defer svr.Close()
 
 	want := `

util/eventbus/debughttp.go

@@ -1,6 +1,8 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
+//go:build !ios
+
 package eventbus
 
 import (

util/eventbus/debughttp_ios.go

@@ -0,0 +1,18 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build ios
+
+package eventbus
+
+import "tailscale.com/tsweb"
+
+func registerHTTPDebugger(d *Debugger, td *tsweb.DebugHandler) {
+	// The event bus debugging UI uses html/template, which uses
+	// reflection for method lookups. This forces the compiler to
+	// retain a lot more code and information to make dynamic method
+	// dispatch work, which is unacceptable bloat for the iOS build.
+	//
+	// TODO: https://github.com/tailscale/tailscale/issues/15297 to
+	// bring the debug UI back to iOS somehow.
+}