wgengine/magicsock: fix typo in comment

Change-Id: I53fdc085ecb2c9123b21429b3d2ae80a06f22df2
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

ALPINE.txt

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

Dockerfile

@@ -62,10 +62,8 @@ RUN GOARCH=$TARGETARCH go install -ldflags="\
       -X tailscale.com/version.gitCommitStamp=$VERSION_GIT_HASH" \
       -v ./cmd/tailscale ./cmd/tailscaled ./cmd/containerboot
 
-FROM alpine:3.19
+FROM alpine:3.18
 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

@@ -1,12 +1,5 @@
 # Copyright (c) Tailscale Inc & AUTHORS
 # SPDX-License-Identifier: BSD-3-Clause
 
-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
+FROM alpine:3.18
+RUN apk add --no-cache ca-certificates iptables iproute2 ip6tables iputils

Makefile

@@ -110,6 +110,15 @@ publishdevnameserver: ## Build and publish k8s-nameserver image to location spec
 	@test "${REPO}" != "ghcr.io/tailscale/k8s-nameserver" || (echo "REPO=... must not be ghcr.io/tailscale/k8s-nameserver" && exit 1)
 	TAGS="${TAGS}" REPOS=${REPO} PLATFORM=${PLATFORM} PUSH=true TARGET=k8s-nameserver ./build_docker.sh
 
+plan93:
+	GOOS=plan9 GOARCH=386 ${HOME}/hack/go/bin/go build -o ${HOME}/hack/rsc-plan9/386/bin/tailscaled ./cmd/tailscaled
+	GOOS=plan9 GOARCH=386 ${HOME}/hack/go/bin/go build -o ${HOME}/hack/rsc-plan9/386/bin/tailscale ./cmd/tailscale
+
+plan9a:
+	GOOS=plan9 GOARCH=amd64 ${HOME}/hack/go/bin/go build -o ${HOME}/hack/rsc-plan9/amd64/bin/tailscaled ./cmd/tailscaled
+	GOOS=plan9 GOARCH=amd64 ${HOME}/hack/go/bin/go build -o ${HOME}/hack/rsc-plan9/amd64/bin/tailscale ./cmd/tailscale
+
+
 .PHONY: sshintegrationtest
 sshintegrationtest: ## Run the SSH integration tests in various Docker containers
 	@GOOS=linux GOARCH=amd64 ./tool/go test -tags integrationtest -c ./ssh/tailssh -o ssh/tailssh/testcontainers/tailssh.test && \

VERSION.txt

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

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:3.19"
+DEFAULT_BASE="tailscale/alpine-base:3.18"
 # 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.

cmd/containerboot/certs.go

@@ -60,9 +60,6 @@ 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) })
 		}
 	}
@@ -119,13 +116,7 @@ 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?
-
-			// 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)
+			_, _, err := cm.lc.CertPair(ctx, domain)
 			if err != nil {
 				log.Printf("error refreshing certificate for %s: %v", domain, err)
 			}

cmd/derper/cert.go

@@ -128,17 +128,16 @@ 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
-	return m.cert, nil
+	certCopy := new(tls.Certificate)
+	*certCopy = *m.cert
+	certCopy.Certificate = certCopy.Certificate[:len(certCopy.Certificate):len(certCopy.Certificate)]
+	return certCopy, nil
 }
 
 func (m *manualCertManager) HTTPHandler(fallback http.Handler) http.Handler {

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. Currently the only supported type is egress.
+                    Type of the ProxyGroup proxies. Supported types are egress and ingress.
                     Type is immutable once a ProxyGroup is created.
                   type: string
                   enum:

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

@@ -2876,7 +2876,7 @@ spec:
                                 type: array
                             type:
                                 description: |-
-                                    Type of the ProxyGroup proxies. Currently the only supported type is egress.
+                                    Type of the ProxyGroup proxies. Supported types are egress and ingress.
                                     Type is immutable once a ProxyGroup is created.
                                 enum:
                                     - egress

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

@@ -49,11 +49,10 @@ 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)
@@ -242,7 +241,7 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 		return false, nil
 	}
 	// 3. Ensure that TLS Secret and RBAC exists
-	if err := r.ensureCertResources(ctx, pgName, dnsName, ing); err != nil {
+	if err := r.ensureCertResources(ctx, pgName, dnsName); err != nil {
 		return false, fmt.Errorf("error ensuring cert resources: %w", err)
 	}
 
@@ -339,11 +338,7 @@ 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.
-	mode := serviceAdvertisementHTTPS
-	if isHTTPEndpointEnabled(ing) {
-		mode = serviceAdvertisementHTTPAndHTTPS
-	}
-	if err = r.maybeUpdateAdvertiseServicesConfig(ctx, pg.Name, serviceName, mode, logger); err != nil {
+	if err = r.maybeUpdateAdvertiseServicesConfig(ctx, pg.Name, serviceName, true, logger); err != nil {
 		return false, fmt.Errorf("failed to update tailscaled config: %w", err)
 	}
 
@@ -359,17 +354,11 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 	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{
+		ports := []networkingv1.IngressPortStatus{
+			{
 				Protocol: "TCP",
 				Port:     443,
-			})
+			},
 		}
 		if isHTTPEndpointEnabled(ing) {
 			ports = append(ports, networkingv1.IngressPortStatus{
@@ -377,14 +366,9 @@ func (r *HAIngressReconciler) maybeProvision(ctx context.Context, hostname strin
 				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,
+				Hostname: dnsName,
 				Ports:    ports,
 			},
 		}
@@ -445,7 +429,7 @@ func (r *HAIngressReconciler) maybeCleanupProxyGroup(ctx context.Context, proxyG
 			}
 
 			// Make sure the VIPService is not advertised in tailscaled or serve config.
-			if err = r.maybeUpdateAdvertiseServicesConfig(ctx, proxyGroupName, vipServiceName, serviceAdvertisementOff, logger); err != nil {
+			if err = r.maybeUpdateAdvertiseServicesConfig(ctx, proxyGroupName, vipServiceName, false, logger); err != nil {
 				return false, fmt.Errorf("failed to update tailscaled config services: %w", err)
 			}
 			_, ok := cfg.Services[vipServiceName]
@@ -528,7 +512,7 @@ func (r *HAIngressReconciler) maybeCleanup(ctx context.Context, hostname string,
 	}
 
 	// 4. Unadvertise the VIPService in tailscaled config.
-	if err = r.maybeUpdateAdvertiseServicesConfig(ctx, pg, serviceName, serviceAdvertisementOff, logger); err != nil {
+	if err = r.maybeUpdateAdvertiseServicesConfig(ctx, pg, serviceName, false, logger); err != nil {
 		return false, fmt.Errorf("failed to update tailscaled config services: %w", err)
 	}
 
@@ -725,16 +709,8 @@ func isHTTPEndpointEnabled(ing *networkingv1.Ingress) bool {
 	return ing.Annotations[annotationHTTPEndpoint] == "enabled"
 }
 
-// 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) {
+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)
 
 	// Get all config Secrets for this ProxyGroup.
 	secrets := &corev1.SecretList{}
@@ -742,21 +718,6 @@ 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 {
@@ -909,8 +870,8 @@ func ownersAreSetAndEqual(a, b *tailscale.VIPService) bool {
 // (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)
+func (r *HAIngressReconciler) ensureCertResources(ctx context.Context, pgName, domain string) error {
+	secret := certSecret(pgName, r.tsNamespace, domain)
 	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)
 	}
@@ -1005,14 +966,9 @@ func certSecretRoleBinding(pgName, namespace, domain string) *rbacv1.RoleBinding
 
 // 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 {
+func certSecret(pgName, namespace, domain string) *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",
@@ -1033,9 +989,9 @@ func certSecret(pgName, namespace, domain string, ing *networkingv1.Ingress) *co
 
 func certResourceLabels(pgName, domain string) map[string]string {
 	return map[string]string{
-		kubetypes.LabelManaged: "true",
-		labelProxyGroup:        pgName,
-		labelDomain:            domain,
+		kubetypes.LabelManaged:      "true",
+		"tailscale.com/proxy-group": pgName,
+		"tailscale.com/domain":      domain,
 	}
 }
 
@@ -1048,28 +1004,3 @@ func (r *HAIngressReconciler) dnsNameForService(ctx context.Context, svc tailcfg
 	}
 	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

@@ -31,7 +31,6 @@ 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"
 )
@@ -60,7 +59,7 @@ func TestIngressPGReconciler(t *testing.T) {
 				},
 			},
 			TLS: []networkingv1.IngressTLS{
-				{Hosts: []string{"my-svc"}},
+				{Hosts: []string{"my-svc.tailnetxyz.ts.net"}},
 			},
 		},
 	}
@@ -68,14 +67,12 @@ 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.
+	// Verify cert resources were created for the first Ingress
+	expectEqual(t, fc, certSecret("test-pg", "operator-ns", "my-svc.ts.net"))
 	expectEqual(t, fc, certSecretRole("test-pg", "operator-ns", "my-svc.ts.net"))
 	expectEqual(t, fc, certSecretRoleBinding("test-pg", "operator-ns", "my-svc.ts.net"))
 
@@ -130,13 +127,11 @@ 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.
+	// Verify cert resources were created for the second Ingress
+	expectEqual(t, fc, certSecret("test-pg", "operator-ns", "my-other-svc.ts.net"))
 	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"))
 
@@ -236,7 +231,7 @@ func TestIngressPGReconciler_UpdateIngressHostname(t *testing.T) {
 				},
 			},
 			TLS: []networkingv1.IngressTLS{
-				{Hosts: []string{"my-svc"}},
+				{Hosts: []string{"my-svc.tailnetxyz.ts.net"}},
 			},
 		},
 	}
@@ -244,19 +239,15 @@ func TestIngressPGReconciler_UpdateIngressHostname(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"})
 
 	// 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"
+		ing.Spec.TLS[0].Hosts[0] = "updated-svc.tailnetxyz.ts.net"
 	})
 	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"})
@@ -477,8 +468,6 @@ 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)
 
@@ -622,7 +611,6 @@ 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)
@@ -816,28 +804,3 @@ func TestIngressPGReconciler_MultiCluster(t *testing.T) {
 		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/operator.go

@@ -9,6 +9,7 @@ package main
 
 import (
 	"context"
+	"fmt"
 	"net/http"
 	"os"
 	"regexp"
@@ -39,6 +40,7 @@ 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"
@@ -331,6 +333,40 @@ 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(&corev1.Secret{}, handler.EnqueueRequestsFromMapFunc(ingressesFromPGStateSecret(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.
@@ -1003,6 +1039,45 @@ func reconcileRequestsForPG(pg string, cl client.Client, ns string) []reconcile.
 	return reqs
 }
 
+func ingressesFromPGStateSecret(cl client.Client, logger *zap.SugaredLogger) handler.MapFunc {
+	return func(ctx context.Context, o client.Object) []reconcile.Request {
+		secret, ok := o.(*corev1.Secret)
+		if !ok {
+			logger.Infof("[unexpected] ProxyGroup handler triggered for an object that is not a ProxyGroup")
+			return nil
+		}
+		if secret.ObjectMeta.Labels[kubetypes.LabelManaged] != "true" {
+			return nil
+		}
+		if secret.ObjectMeta.Labels[LabelParentType] != "proxygroup" {
+			return nil
+		}
+		if secret.ObjectMeta.Labels[kubetypes.LabelSecretType] != "state" {
+			return nil
+		}
+		pgName, ok := secret.ObjectMeta.Labels[LabelParentName]
+		if !ok {
+			return nil
+		}
+
+		ingList := &networkingv1.IngressList{}
+		if err := cl.List(ctx, ingList, client.MatchingFields{indexIngressProxyGroup: pgName}); err != nil {
+			logger.Infof("error listing Ingresses, skipping a reconcile for event on Secret %s: %v", secret.Name, err)
+			return nil
+		}
+		reqs := make([]reconcile.Request, 0)
+		for _, ing := range ingList.Items {
+			reqs = append(reqs, reconcile.Request{
+				NamespacedName: types.NamespacedName{
+					Namespace: ing.Namespace,
+					Name:      ing.Name,
+				},
+			})
+		}
+		return reqs
+	}
+}
+
 // egressSvcsFromEgressProxyGroup is an event handler for egress ProxyGroups. It returns reconcile requests for all
 // user-created ExternalName Services that should be exposed on this ProxyGroup.
 func egressSvcsFromEgressProxyGroup(cl client.Client, logger *zap.SugaredLogger) handler.MapFunc {
@@ -1033,6 +1108,36 @@ 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 {
@@ -1153,7 +1258,63 @@ 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/natc/natc.go

@@ -94,24 +94,18 @@ func main() {
 		}
 		ignoreDstTable.Insert(pfx, true)
 	}
-	var (
-		v4Prefixes    []netip.Prefix
-		numV4DNSAddrs int
-	)
+	var v4Prefixes []netip.Prefix
 	for _, s := range strings.Split(*v4PfxStr, ",") {
 		p := netip.MustParsePrefix(strings.TrimSpace(s))
 		if p.Masked() != p {
 			log.Fatalf("v4 prefix %v is not a masked prefix", p)
 		}
 		v4Prefixes = append(v4Prefixes, p)
-		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,
 	}
@@ -159,13 +153,12 @@ func main() {
 	}
 
 	c := &connector{
-		ts:            ts,
-		lc:            lc,
-		dnsAddr:       dnsAddr,
-		v4Ranges:      v4Prefixes,
-		numV4DNSAddrs: numV4DNSAddrs,
-		v6ULA:         ula(uint16(*siteID)),
-		ignoreDsts:    ignoreDstTable,
+		ts:         ts,
+		lc:         lc,
+		dnsAddr:    dnsAddr,
+		v4Ranges:   v4Prefixes,
+		v6ULA:      ula(uint16(*siteID)),
+		ignoreDsts: ignoreDstTable,
 	}
 	c.run(ctx)
 }
@@ -184,11 +177,6 @@ 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
 
@@ -514,7 +502,6 @@ 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
@@ -560,25 +547,17 @@ 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
-	// 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")
-			}
+	for _, r := range ps.c.v4Ranges {
+		ip := randV4(r)
+		for r.Contains(ip) {
 			if !ps.isIPUsedLocked(ip) && ip != ps.c.dnsAddr {
 				return ip
 			}
+			ip = ip.Next()
 		}
 	}
+	return netip.Addr{}
 }
 
 // randV4 returns a random IPv4 address within the given prefix.
@@ -604,7 +583,6 @@ 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

@@ -1,429 +0,0 @@
-// 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/tailscaled/ssh.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build (linux || darwin || freebsd || openbsd) && !ts_omit_ssh
+//go:build (linux || darwin || freebsd || openbsd || plan9) && !ts_omit_ssh
 
 package main
 

cmd/tailscaled/tailscaled.go

@@ -82,7 +82,9 @@ func defaultTunName() string {
 		// "utun" is recognized by wireguard-go/tun/tun_darwin.go
 		// as a magic value that uses/creates any free number.
 		return "utun"
-	case "plan9", "aix", "solaris", "illumos":
+	case "plan9":
+		return "auto"
+	case "aix", "solaris", "illumos":
 		return "userspace-networking"
 	case "linux":
 		switch distro.Get() {
@@ -180,6 +182,10 @@ func main() {
 		return
 	}
 
+	if runtime.GOOS == "plan9" && os.Getenv("_NETSHELL_CHILD_") != "" {
+		os.Args = []string{"tailscaled", "be-child", "plan9-netshell"}
+	}
+
 	if len(os.Args) > 1 {
 		sub := os.Args[1]
 		if fp, ok := subCommands[sub]; ok {
@@ -230,7 +236,18 @@ func main() {
 	// Only apply a default statepath when neither have been provided, so that a
 	// user may specify only --statedir if they wish.
 	if args.statepath == "" && args.statedir == "" {
-		args.statepath = paths.DefaultTailscaledStateFile()
+		if runtime.GOOS == "plan9" {
+			home, err := os.UserHomeDir()
+			if err != nil {
+				log.Fatalf("failed to get home directory: %v", err)
+			}
+			args.statedir = filepath.Join(home, "tailscale-state")
+			if err := os.MkdirAll(args.statedir, 0700); err != nil {
+				log.Fatalf("failed to create state directory: %v", err)
+			}
+		} else {
+			args.statepath = paths.DefaultTailscaledStateFile()
+		}
 	}
 
 	if args.disableLogs {
@@ -731,6 +748,12 @@ func tryEngine(logf logger.Logf, sys *tsd.System, name string) (onlyNetstack boo
 			return false, err
 		}
 
+		if runtime.GOOS == "plan9" {
+			// TODO(bradfitz): why don't we do this on all platforms?
+			// We should. Doing it just on plan9 for now conservatively.
+			sys.NetMon.Get().SetTailscaleInterfaceName(devName)
+		}
+
 		r, err := router.New(logf, dev, sys.NetMon.Get(), sys.HealthTracker())
 		if err != nil {
 			dev.Close()

control/controlhttp/client.go

@@ -96,9 +96,6 @@ 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")
@@ -249,6 +246,11 @@ func (a *Dialer) dial(ctx context.Context) (*ClientConn, error) {
 		results[i].conn = nil // so we don't close it in the defer
 		return conn, nil
 	}
+	if ctx.Err() != nil {
+		a.logf("controlhttp: context aborted dialing")
+		return nil, ctx.Err()
+	}
+
 	merr := multierr.New(errs...)
 
 	// If we get here, then we didn't get anywhere with our dial plan; fall back to just using DNS.
@@ -281,9 +283,7 @@ 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.
-		if d.logPort80Failure.CompareAndSwap(true, false) {
-			d.logf("controlhttp: forcing port 443 dial due to recent noise dial")
-		}
+		d.logf("controlhttp: forcing port 443 dial due to recent noise dial")
 		return true
 	}
 

control/controlhttp/constants.go

@@ -6,7 +6,6 @@ package controlhttp
 import (
 	"net/http"
 	"net/url"
-	"sync/atomic"
 	"time"
 
 	"tailscale.com/health"
@@ -91,11 +90,6 @@ 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/featureknob/featureknob.go

@@ -40,7 +40,7 @@ func CanRunTailscaleSSH() error {
 		if version.IsSandboxedMacOS() {
 			return errors.New("The Tailscale SSH server does not run in sandboxed Tailscale GUI builds.")
 		}
-	case "freebsd", "openbsd":
+	case "freebsd", "openbsd", "plan9":
 	default:
 		return errors.New("The Tailscale SSH server is not supported on " + runtime.GOOS)
 	}

go.mod

@@ -3,6 +3,7 @@ module tailscale.com
 go 1.24.0
 
 require (
+	9fans.net/go v0.0.8-0.20250307142834-96bdba94b63f
 	filippo.io/mkcert v1.4.4
 	fyne.io/systray v1.11.0
 	github.com/akutz/memconn v0.1.0
@@ -36,6 +37,7 @@ require (
 	github.com/go-json-experiment/json v0.0.0-20250223041408-d3c622f1b874
 	github.com/go-logr/zapr v1.3.0
 	github.com/go-ole/go-ole v1.3.0
+	github.com/go4org/plan9netshell v0.0.0-20250324183649-788daa080737
 	github.com/godbus/dbus/v5 v5.1.1-0.20230522191255-76236955d466
 	github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da
 	github.com/golang/snappy v0.0.4
@@ -84,12 +86,12 @@ require (
 	github.com/tailscale/setec v0.0.0-20250205144240-8898a29c3fbb
 	github.com/tailscale/web-client-prebuilt v0.0.0-20250124233751-d4cd19a26976
 	github.com/tailscale/wf v0.0.0-20240214030419-6fbb0a674ee6
-	github.com/tailscale/wireguard-go v0.0.0-20250107165329-0b8b35511f19
+	github.com/tailscale/wireguard-go v0.0.0-20250304000100-91a0587fb251
 	github.com/tailscale/xnet v0.0.0-20240729143630-8497ac4dab2e
 	github.com/tc-hib/winres v0.2.1
 	github.com/tcnksm/go-httpstat v0.2.0
 	github.com/toqueteos/webbrowser v1.2.0
-	github.com/u-root/u-root v0.12.0
+	github.com/u-root/u-root v0.14.0
 	github.com/vishvananda/netns v0.0.4
 	go.uber.org/zap v1.27.0
 	go4.org/mem v0.0.0-20240501181205-ae6ca9944745

go.sum

@@ -2,6 +2,8 @@
 4d63.com/gocheckcompilerdirectives v1.2.1/go.mod h1:yjDJSxmDTtIHHCqX0ufRYZDL6vQtMG7tJdKVeWwsqvs=
 4d63.com/gochecknoglobals v0.2.1 h1:1eiorGsgHOFOuoOiJDy2psSrQbRdIHrlge0IJIkUgDc=
 4d63.com/gochecknoglobals v0.2.1/go.mod h1:KRE8wtJB3CXCsb1xy421JfTHIIbmT3U5ruxw2Qu8fSU=
+9fans.net/go v0.0.8-0.20250307142834-96bdba94b63f h1:1C7nZuxUMNz7eiQALRfiqNOm04+m3edWlRff/BYHf0Q=
+9fans.net/go v0.0.8-0.20250307142834-96bdba94b63f/go.mod h1:hHyrZRryGqVdqrknjq5OWDLGCTJ2NeEvtrpR96mjraM=
 cloud.google.com/go v0.26.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
 cloud.google.com/go v0.34.0/go.mod h1:aQUYkXzVsufM+DwF1aE+0xfcU+56JwCaLick0ClmMTw=
 cloud.google.com/go v0.38.0/go.mod h1:990N+gfupTy94rShfmMCWGDn0LpTmnzTp2qbd1dvSRU=
@@ -389,6 +391,8 @@ github.com/go-viper/mapstructure/v2 v2.0.0-alpha.1 h1:TQcrn6Wq+sKGkpyPvppOz99zsM
 github.com/go-viper/mapstructure/v2 v2.0.0-alpha.1/go.mod h1:oJDH3BJKyqBA2TXFhDsKDGDTlndYOZ6rGS0BRZIxGhM=
 github.com/go-xmlfmt/xmlfmt v1.1.2 h1:Nea7b4icn8s57fTx1M5AI4qQT5HEM3rVUO8MuE6g80U=
 github.com/go-xmlfmt/xmlfmt v1.1.2/go.mod h1:aUCEOzzezBEjDBbFBoSiya/gduyIiWYRP6CnSFIV8AM=
+github.com/go4org/plan9netshell v0.0.0-20250324183649-788daa080737 h1:cf60tHxREO3g1nroKr2osU3JWZsJzkfi7rEg+oAB0Lo=
+github.com/go4org/plan9netshell v0.0.0-20250324183649-788daa080737/go.mod h1:MIS0jDzbU/vuM9MC4YnBITCv+RYuTRq8dJzmCrFsK9g=
 github.com/gobuffalo/flect v1.0.3 h1:xeWBM2nui+qnVvNM4S3foBhCAL2XgPU+a7FdpelbTq4=
 github.com/gobuffalo/flect v1.0.3/go.mod h1:A5msMlrHtLqh9umBSnvabjsMrCcCpAyzglnDvkbYKHs=
 github.com/gobwas/glob v0.2.3 h1:A4xDbljILXROh+kObIiy5kIaPYD8e96x1tgBhUI5J+Y=
@@ -545,8 +549,8 @@ github.com/hexops/gotextdiff v1.0.3/go.mod h1:pSWU5MAI3yDq+fZBTazCSJysOMbxWL1BSo
 github.com/huandu/xstrings v1.3.3/go.mod h1:y5/lhBue+AyNmUVz9RLU9xbLR0o4KIIExikq4ovT0aE=
 github.com/huandu/xstrings v1.5.0 h1:2ag3IFq9ZDANvthTwTiqSSZLjDc+BedvHPAp5tJy2TI=
 github.com/huandu/xstrings v1.5.0/go.mod h1:y5/lhBue+AyNmUVz9RLU9xbLR0o4KIIExikq4ovT0aE=
-github.com/hugelgupf/vmtest v0.0.0-20240102225328-693afabdd27f h1:ov45/OzrJG8EKbGjn7jJZQJTN7Z1t73sFYNIRd64YlI=
-github.com/hugelgupf/vmtest v0.0.0-20240102225328-693afabdd27f/go.mod h1:JoDrYMZpDPYo6uH9/f6Peqms3zNNWT2XiGgioMOIGuI=
+github.com/hugelgupf/vmtest v0.0.0-20240216064925-0561770280a1 h1:jWoR2Yqg8tzM0v6LAiP7i1bikZJu3gxpgvu3g1Lw+a0=
+github.com/hugelgupf/vmtest v0.0.0-20240216064925-0561770280a1/go.mod h1:B63hDJMhTupLWCHwopAyEo7wRFowx9kOc8m8j1sfOqE=
 github.com/ianlancetaylor/demangle v0.0.0-20181102032728-5e5cf60278f6/go.mod h1:aSSvb/t6k1mPoxDqO4vJh6VOCGPwU4O0C2/Eqndh1Sc=
 github.com/illarion/gonotify/v3 v3.0.2 h1:O7S6vcopHexutmpObkeWsnzMJt/r1hONIEogeVNmJMk=
 github.com/illarion/gonotify/v3 v3.0.2/go.mod h1:HWGPdPe817GfvY3w7cx6zkbzNZfi3QjcBm/wgVvEL1U=
@@ -922,8 +926,8 @@ github.com/tailscale/web-client-prebuilt v0.0.0-20250124233751-d4cd19a26976 h1:U
 github.com/tailscale/web-client-prebuilt v0.0.0-20250124233751-d4cd19a26976/go.mod h1:agQPE6y6ldqCOui2gkIh7ZMztTkIQKH049tv8siLuNQ=
 github.com/tailscale/wf v0.0.0-20240214030419-6fbb0a674ee6 h1:l10Gi6w9jxvinoiq15g8OToDdASBni4CyJOdHY1Hr8M=
 github.com/tailscale/wf v0.0.0-20240214030419-6fbb0a674ee6/go.mod h1:ZXRML051h7o4OcI0d3AaILDIad/Xw0IkXaHM17dic1Y=
-github.com/tailscale/wireguard-go v0.0.0-20250107165329-0b8b35511f19 h1:BcEJP2ewTIK2ZCsqgl6YGpuO6+oKqqag5HHb7ehljKw=
-github.com/tailscale/wireguard-go v0.0.0-20250107165329-0b8b35511f19/go.mod h1:BOm5fXUBFM+m9woLNBoxI9TaBXXhGNP50LX/TGIvGb4=
+github.com/tailscale/wireguard-go v0.0.0-20250304000100-91a0587fb251 h1:h/41LFTrwMxB9Xvvug0kRdQCU5TlV1+pAMQw0ZtDE3U=
+github.com/tailscale/wireguard-go v0.0.0-20250304000100-91a0587fb251/go.mod h1:BOm5fXUBFM+m9woLNBoxI9TaBXXhGNP50LX/TGIvGb4=
 github.com/tailscale/xnet v0.0.0-20240729143630-8497ac4dab2e h1:zOGKqN5D5hHhiYUp091JqK7DPCqSARyUfduhGUY8Bek=
 github.com/tailscale/xnet v0.0.0-20240729143630-8497ac4dab2e/go.mod h1:orPd6JZXXRyuDusYilywte7k094d7dycXXU5YnWsrwg=
 github.com/tc-hib/winres v0.2.1 h1:YDE0FiP0VmtRaDn7+aaChp1KiF4owBiJa5l964l5ujA=
@@ -950,10 +954,10 @@ github.com/tommy-muehle/go-mnd/v2 v2.5.1 h1:NowYhSdyE/1zwK9QCLeRb6USWdoif80Ie+v+
 github.com/tommy-muehle/go-mnd/v2 v2.5.1/go.mod h1:WsUAkMJMYww6l/ufffCD3m+P7LEvr8TnZn9lwVDlgzw=
 github.com/toqueteos/webbrowser v1.2.0 h1:tVP/gpK69Fx+qMJKsLE7TD8LuGWPnEV71wBN9rrstGQ=
 github.com/toqueteos/webbrowser v1.2.0/go.mod h1:XWoZq4cyp9WeUeak7w7LXRUQf1F1ATJMir8RTqb4ayM=
-github.com/u-root/gobusybox/src v0.0.0-20231228173702-b69f654846aa h1:unMPGGK/CRzfg923allsikmvk2l7beBeFPUNC4RVX/8=
-github.com/u-root/gobusybox/src v0.0.0-20231228173702-b69f654846aa/go.mod h1:Zj4Tt22fJVn/nz/y6Ergm1SahR9dio1Zm/D2/S0TmXM=
-github.com/u-root/u-root v0.12.0 h1:K0AuBFriwr0w/PGS3HawiAw89e3+MU7ks80GpghAsNs=
-github.com/u-root/u-root v0.12.0/go.mod h1:FYjTOh4IkIZHhjsd17lb8nYW6udgXdJhG1c0r6u0arI=
+github.com/u-root/gobusybox/src v0.0.0-20240225013946-a274a8d5d83a h1:eg5FkNoQp76ZsswyGZ+TjYqA/rhKefxK8BW7XOlQsxo=
+github.com/u-root/gobusybox/src v0.0.0-20240225013946-a274a8d5d83a/go.mod h1:e/8TmrdreH0sZOw2DFKBaUV7bvDWRq6SeM9PzkuVM68=
+github.com/u-root/u-root v0.14.0 h1:Ka4T10EEML7dQ5XDvO9c3MBN8z4nuSnGjcd1jmU2ivg=
+github.com/u-root/u-root v0.14.0/go.mod h1:hAyZorapJe4qzbLWlAkmSVCJGbfoU9Pu4jpJ1WMluqE=
 github.com/u-root/uio v0.0.0-20240224005618-d2acac8f3701 h1:pyC9PaHYZFgEKFdlp3G8RaCKgVpHZnecvArXvPXcFkM=
 github.com/u-root/uio v0.0.0-20240224005618-d2acac8f3701/go.mod h1:P3a5rG4X7tI17Nn3aOIAYr5HbIMukwXG0urG0WuL8OA=
 github.com/ulikunitz/xz v0.5.11 h1:kpFauv27b6ynzBNT/Xy+1k+fK4WswhN/6PN5WhFAGw8=

ipn/ipnlocal/local.go

@@ -958,9 +958,7 @@ func (b *LocalBackend) linkChange(delta *netmon.ChangeDelta) {
 
 	if peerAPIListenAsync && b.netMap != nil && b.state == ipn.Running {
 		want := b.netMap.GetAddresses().Len()
-		have := len(b.peerAPIListeners)
-		b.logf("[v1] linkChange: have %d peerAPIListeners, want %d", have, want)
-		if have < want {
+		if len(b.peerAPIListeners) < want {
 			b.logf("linkChange: peerAPIListeners too low; trying again")
 			b.goTracker.Go(b.initPeerAPIListener)
 		}
@@ -2404,9 +2402,11 @@ 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 {
-		// Use memory store by default if no explicit store is provided.
+		b.logf("auditlog: [unexpected] no persistent audit log storage configured.  using memory store.")
 		store = auditlog.NewLogStore(&memstore.Store{})
 	}
 
@@ -5001,6 +5001,11 @@ func shouldUseOneCGNATRoute(logf logger.Logf, mon *netmon.Monitor, controlKnobs
 		}
 	}
 
+	if versionOS == "plan9" {
+		// Just temporarily during plan9 bringup to have fewer routes to debug.
+		return true
+	}
+
 	// Also prefer to do this on the Mac, so that we don't need to constantly
 	// update the network extension configuration (which is disruptive to
 	// Chrome, see https://github.com/tailscale/tailscale/issues/3102). Only
@@ -5369,7 +5374,6 @@ 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

ipn/ipnlocal/ssh.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build linux || (darwin && !ios) || freebsd || openbsd
+//go:build linux || (darwin && !ios) || freebsd || openbsd || plan9
 
 package ipnlocal
 

ipn/ipnlocal/ssh_stub.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build ios || (!linux && !darwin && !freebsd && !openbsd)
+//go:build ios || (!linux && !darwin && !freebsd && !openbsd && !plan9)
 
 package ipnlocal
 

ipn/ipnserver/server.go

@@ -331,7 +331,7 @@ func (a *actor) Permissions(operatorUID string) (read, write bool) {
 		// checks here. Note that this permission model is being changed in
 		// tailscale/corp#18342.
 		return true, true
-	case "js":
+	case "js", "plan9":
 		return true, true
 	}
 	if a.ci.IsUnixSock() {

ipn/store/kubestore/store_kube.go

@@ -143,6 +143,15 @@ func (s *Store) WriteTLSCertAndKey(domain string, cert, key []byte) (err error)
 	if err := dnsname.ValidHostname(domain); err != nil {
 		return fmt.Errorf("invalid domain name %q: %w", domain, err)
 	}
+	defer func() {
+		// 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.
+		if err == nil {
+			s.memory.WriteState(ipn.StateKey(domain+".crt"), cert)
+			s.memory.WriteState(ipn.StateKey(domain+".key"), key)
+		}
+	}()
 	secretName := s.secretName
 	data := map[string][]byte{
 		domain + ".crt": cert,
@@ -157,32 +166,19 @@ func (s *Store) WriteTLSCertAndKey(domain string, cert, key []byte) (err error)
 			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
+	return s.updateSecret(data, secretName)
 }
 
 // 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))
@@ -190,12 +186,16 @@ func (s *Store) ReadTLSCertAndKey(domain string) (cert, key []byte, err error) {
 			return cert, key, nil
 		}
 	}
-	if s.certShareMode == "" {
+	if s.certShareMode != "ro" {
 		return nil, nil, ipn.ErrStateNotExist
 	}
+	// If we are in cert share read only mode, it is possible that a write
+	// replica just issued the TLS cert for this DNS name and it has not
+	// been loaded to store yet, so check the Secret.
 
 	ctx, cancel := context.WithTimeout(context.Background(), timeout)
 	defer cancel()
+
 	secret, err := s.client.GetSecret(ctx, domain)
 	if err != nil {
 		if kubeclient.IsNotFoundErr(err) {
@@ -212,18 +212,9 @@ func (s *Store) ReadTLSCertAndKey(domain string) (cert, key []byte, err error) {
 	}
 	// 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)
-	}
+	// key to the memory store in a single lock-protected operation.
+	s.memory.WriteState(ipn.StateKey(certKey), cert)
+	s.memory.WriteState(ipn.StateKey(keyKey), key)
 	return cert, key, nil
 }
 

ipn/store/kubestore/store_kube_test.go

@@ -201,6 +201,10 @@ func TestWriteTLSCertAndKey(t *testing.T) {
 				"tls.crt": []byte(testCert),
 				"tls.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_update_existing",
@@ -215,6 +219,10 @@ func TestWriteTLSCertAndKey(t *testing.T) {
 				"tls.crt": []byte(testCert),
 				"tls.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: "update_existing",
@@ -359,7 +367,7 @@ func TestReadTLSCertAndKey(t *testing.T) {
 		wantMemoryStore map[ipn.StateKey][]byte
 	}{
 		{
-			name: "found_in_memory",
+			name: "found",
 			memoryStore: map[ipn.StateKey][]byte{
 				"my-app.tailnetxyz.ts.net.crt": []byte(testCert),
 				"my-app.tailnetxyz.ts.net.key": []byte(testKey),
@@ -373,7 +381,7 @@ func TestReadTLSCertAndKey(t *testing.T) {
 			},
 		},
 		{
-			name:    "not_found_in_memory",
+			name:    "not_found",
 			domain:  testDomain,
 			wantErr: ipn.ErrStateNotExist,
 		},
@@ -392,17 +400,6 @@ func TestReadTLSCertAndKey(t *testing.T) {
 				"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",

k8s-operator/api.md

@@ -600,7 +600,7 @@ _Appears in:_
 
 | Field | Description | Default | Validation |
 | --- | --- | --- | --- |
-| `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 /> |
+| `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 /> |
 | `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_proxygroup.go

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

logpolicy/logpolicy.go

@@ -627,7 +627,7 @@ func (opts Options) New() *Policy {
 		conf.IncludeProcSequence = true
 	}
 
-	if envknob.NoLogsNoSupport() || testenv.InTest() {
+	if envknob.NoLogsNoSupport() || testenv.InTest() || runtime.GOOS == "plan9" {
 		opts.Logf("You have disabled logging. Tailscale will not be able to provide support.")
 		conf.HTTPC = &http.Client{Transport: noopPretendSuccessTransport{}}
 	} else {

net/dns/manager.go

@@ -284,7 +284,7 @@ func (m *Manager) compileConfig(cfg Config) (rcfg resolver.Config, ocfg OSConfig
 
 	// Deal with trivial configs first.
 	switch {
-	case !cfg.needsOSResolver():
+	case !cfg.needsOSResolver() || runtime.GOOS == "plan9":
 		// Set search domains, but nothing else. This also covers the
 		// case where cfg is entirely zero, in which case these
 		// configs clear all Tailscale DNS settings.

net/dns/manager_default.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build !linux && !freebsd && !openbsd && !windows && !darwin && !illumos && !solaris
+//go:build !linux && !freebsd && !openbsd && !windows && !darwin && !illumos && !solaris && !plan9
 
 package dns
 

net/dns/manager_plan9.go

@@ -0,0 +1,181 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+// TODO: man 6 ndb | grep -e 'suffix.*same line'
+// to detect Russ's https://9fans.topicbox.com/groups/9fans/T9c9d81b5801a0820/ndb-suffix-specific-dns-changes
+
+package dns
+
+import (
+	"bufio"
+	"bytes"
+	"fmt"
+	"io"
+	"net/netip"
+	"os"
+	"regexp"
+	"strings"
+	"unicode"
+
+	"tailscale.com/control/controlknobs"
+	"tailscale.com/health"
+	"tailscale.com/types/logger"
+	"tailscale.com/util/set"
+)
+
+func NewOSConfigurator(logf logger.Logf, ht *health.Tracker, knobs *controlknobs.Knobs, interfaceName string) (OSConfigurator, error) {
+	return &plan9DNSManager{
+		logf:  logf,
+		ht:    ht,
+		knobs: knobs,
+	}, nil
+}
+
+type plan9DNSManager struct {
+	logf  logger.Logf
+	ht    *health.Tracker
+	knobs *controlknobs.Knobs
+}
+
+// netNDBBytesWithoutTailscale returns raw (the contents of /net/ndb) with any
+// Tailscale bits removed.
+func netNDBBytesWithoutTailscale(raw []byte) ([]byte, error) {
+	var ret bytes.Buffer
+	bs := bufio.NewScanner(bytes.NewReader(raw))
+	removeLine := set.Set[string]{}
+	for bs.Scan() {
+		t := bs.Text()
+		if rest, ok := strings.CutPrefix(t, "#tailscaled-added-line:"); ok {
+			removeLine.Add(strings.TrimSpace(rest))
+			continue
+		}
+		trimmed := strings.TrimSpace(t)
+		if removeLine.Contains(trimmed) {
+			removeLine.Delete(trimmed)
+			continue
+		}
+
+		// Also remove any DNS line referencing *.ts.net. This is
+		// Tailscale-specific (and won't work with, say, Headscale), but
+		// the Headscale case will be covered by the #tailscaled-added-line
+		// logic above, assuming the user didn't delete those comments.
+		if (strings.HasPrefix(trimmed, "dns=") || strings.Contains(trimmed, "dnsdomain=")) &&
+			strings.HasSuffix(trimmed, ".ts.net") {
+			continue
+		}
+
+		ret.WriteString(t)
+		ret.WriteByte('\n')
+	}
+	return ret.Bytes(), bs.Err()
+}
+
+// setNDBSuffix adds lines to tsFree (the contents of /net/ndb already cleaned
+// of Tailscale-added lines) to add the optional DNS search domain (e.g.
+// "foo.ts.net") and DNS server to it.
+func setNDBSuffix(tsFree []byte, suffix string) []byte {
+	suffix = strings.TrimSuffix(suffix, ".")
+	if suffix == "" {
+		return tsFree
+	}
+	var buf bytes.Buffer
+	bs := bufio.NewScanner(bytes.NewReader(tsFree))
+	var added []string
+	addLine := func(s string) {
+		added = append(added, strings.TrimSpace(s))
+		buf.WriteString(s)
+	}
+	for bs.Scan() {
+		buf.Write(bs.Bytes())
+		buf.WriteByte('\n')
+
+		t := bs.Text()
+		if suffix != "" && len(added) == 0 && strings.HasPrefix(t, "\tdns=") {
+			addLine(fmt.Sprintf("\tdns=100.100.100.100 suffix=%s\n", suffix))
+			addLine(fmt.Sprintf("\tdnsdomain=%s\n", suffix))
+		}
+	}
+	bufTrim := bytes.TrimLeftFunc(buf.Bytes(), unicode.IsSpace)
+	if len(added) == 0 {
+		return bufTrim
+	}
+	var ret bytes.Buffer
+	for _, s := range added {
+		ret.WriteString("#tailscaled-added-line: ")
+		ret.WriteString(s)
+		ret.WriteString("\n")
+	}
+	ret.WriteString("\n")
+	ret.Write(bufTrim)
+	return ret.Bytes()
+}
+
+func (m *plan9DNSManager) SetDNS(c OSConfig) error {
+	ndbOnDisk, err := os.ReadFile("/net/ndb")
+	if err != nil {
+		return err
+	}
+
+	tsFree, err := netNDBBytesWithoutTailscale(ndbOnDisk)
+	if err != nil {
+		return err
+	}
+
+	var suffix string
+	if len(c.SearchDomains) > 0 {
+		suffix = string(c.SearchDomains[0])
+	}
+
+	newBuf := setNDBSuffix(tsFree, suffix)
+	if !bytes.Equal(newBuf, ndbOnDisk) {
+		if err := os.WriteFile("/net/ndb", newBuf, 0644); err != nil {
+			return fmt.Errorf("writing /net/ndb: %w", err)
+		}
+		if f, err := os.OpenFile("/net/dns", os.O_RDWR, 0); err == nil {
+			if _, err := io.WriteString(f, "refresh\n"); err != nil {
+				f.Close()
+				return fmt.Errorf("/net/dns refresh write: %w", err)
+			}
+			if err := f.Close(); err != nil {
+				return fmt.Errorf("/net/dns refresh close: %w", err)
+			}
+		}
+	}
+
+	return nil
+}
+
+func (m *plan9DNSManager) SupportsSplitDNS() bool { return false }
+
+func (m *plan9DNSManager) Close() error {
+	// TODO(bradfitz): remove the Tailscale bits from /net/ndb ideally
+	return nil
+}
+
+var dnsRegex = regexp.MustCompile(`\bdns=(\d+\.\d+\.\d+\.\d+)\b`)
+
+func (m *plan9DNSManager) GetBaseConfig() (OSConfig, error) {
+	var oc OSConfig
+	f, err := os.Open("/net/ndb")
+	if err != nil {
+		return oc, err
+	}
+	defer f.Close()
+	bs := bufio.NewScanner(f)
+	for bs.Scan() {
+		m := dnsRegex.FindSubmatch(bs.Bytes())
+		if m == nil {
+			continue
+		}
+		addr, err := netip.ParseAddr(string(m[1]))
+		if err != nil {
+			continue
+		}
+		oc.Nameservers = append(oc.Nameservers, addr)
+	}
+	if err := bs.Err(); err != nil {
+		return oc, err
+	}
+
+	return oc, nil
+}

net/dns/manager_plan9_test.go

@@ -0,0 +1,86 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build plan9
+
+package dns
+
+import "testing"
+
+func TestNetNDBBytesWithoutTailscale(t *testing.T) {
+	tests := []struct {
+		name string
+		raw  string
+		want string
+	}{
+		{
+			name: "empty",
+			raw:  "",
+			want: "",
+		},
+		{
+			name: "no-tailscale",
+			raw:  "# This is a comment\nip=10.0.2.15 ipmask=255.255.255.0 ipgw=10.0.2.2\n\tsys=gnot\n",
+			want: "# This is a comment\nip=10.0.2.15 ipmask=255.255.255.0 ipgw=10.0.2.2\n\tsys=gnot\n",
+		},
+		{
+			name: "remove-by-comments",
+			raw:  "# This is a comment\n#tailscaled-added-line: dns=100.100.100.100\nip=10.0.2.15 ipmask=255.255.255.0 ipgw=10.0.2.2\n\tdns=100.100.100.100\n\tsys=gnot\n",
+			want: "# This is a comment\nip=10.0.2.15 ipmask=255.255.255.0 ipgw=10.0.2.2\n\tsys=gnot\n",
+		},
+		{
+			name: "remove-by-ts.net",
+			raw:  "Some line\n\tdns=100.100.100.100 suffix=foo.ts.net\n\tfoo=bar\n",
+			want: "Some line\n\tfoo=bar\n",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got, err := netNDBBytesWithoutTailscale([]byte(tt.raw))
+			if err != nil {
+				t.Fatal(err)
+			}
+			if string(got) != tt.want {
+				t.Errorf("GOT:\n%s\n\nWANT:\n%s\n", string(got), tt.want)
+			}
+		})
+	}
+}
+
+func TestSetNDBSuffix(t *testing.T) {
+	tests := []struct {
+		name string
+		raw  string
+		want string
+	}{
+		{
+			name: "empty",
+			raw:  "",
+			want: "",
+		},
+		{
+			name: "set",
+			raw:  "ip=10.0.2.15 ipmask=255.255.255.0 ipgw=10.0.2.2\n\tsys=gnot\n\tdns=100.100.100.100\n\n# foo\n",
+			want: `#tailscaled-added-line: dns=100.100.100.100 suffix=foo.ts.net
+#tailscaled-added-line: dnsdomain=foo.ts.net
+
+ip=10.0.2.15 ipmask=255.255.255.0 ipgw=10.0.2.2
+	sys=gnot
+	dns=100.100.100.100
+	dns=100.100.100.100 suffix=foo.ts.net
+	dnsdomain=foo.ts.net
+
+# foo
+`,
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := setNDBSuffix([]byte(tt.raw), "foo.ts.net")
+			if string(got) != tt.want {
+				t.Errorf("wrong value\n GOT %q:\n%s\n\nWANT %q:\n%s\n", got, got, tt.want, tt.want)
+			}
+		})
+	}
+
+}

net/netcheck/netcheck.go

@@ -1188,6 +1188,10 @@ func (c *Client) measureAllICMPLatency(ctx context.Context, rs *reportState, nee
 	if len(need) == 0 {
 		return nil
 	}
+	if runtime.GOOS == "plan9" {
+		// ICMP isn't implemented.
+		return nil
+	}
 	ctx, done := context.WithTimeout(ctx, icmpProbeTimeout)
 	defer done()
 

net/netmon/netmon.go

@@ -596,7 +596,7 @@ func (m *Monitor) pollWallTime() {
 //
 // We don't do this on mobile platforms for battery reasons, and because these
 // platforms don't really sleep in the same way.
-const shouldMonitorTimeJump = runtime.GOOS != "android" && runtime.GOOS != "ios"
+const shouldMonitorTimeJump = runtime.GOOS != "android" && runtime.GOOS != "ios" && runtime.GOOS != "plan9"
 
 // checkWallTimeAdvanceLocked reports whether wall time jumped more than 150% of
 // pollWallTimeInterval, indicating we probably just came out of sleep. Once a

net/tsdial/tsdial.go

@@ -242,7 +242,7 @@ func changeAffectsConn(delta *netmon.ChangeDelta, conn net.Conn) bool {
 	// In a few cases, we don't have a new DefaultRouteInterface (e.g. on
 	// Android; see tailscale/corp#19124); if so, pessimistically assume
 	// that all connections are affected.
-	if delta.New.DefaultRouteInterface == "" {
+	if delta.New.DefaultRouteInterface == "" && runtime.GOOS != "plan9" {
 		return true
 	}
 

net/tstun/tstun_stub.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build plan9 || aix || solaris || illumos
+//go:build aix || solaris || illumos
 
 package tstun
 

net/tstun/tun.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build !wasm && !plan9 && !tamago && !aix && !solaris && !illumos
+//go:build !wasm && !tamago && !aix && !solaris && !illumos
 
 // Package tun creates a tuntap device, working around OS-specific
 // quirks if necessary.
@@ -9,6 +9,9 @@ package tstun
 
 import (
 	"errors"
+	"fmt"
+	"log"
+	"os"
 	"runtime"
 	"strings"
 	"time"
@@ -45,6 +48,9 @@ func New(logf logger.Logf, tunName string) (tun.Device, string, error) {
 		}
 		dev, err = CreateTAP.Get()(logf, tapName, bridgeName)
 	} else {
+		if runtime.GOOS == "plan9" {
+			cleanUpPlan9Interfaces()
+		}
 		dev, err = tun.CreateTUN(tunName, int(DefaultTUNMTU()))
 	}
 	if err != nil {
@@ -65,6 +71,35 @@ func New(logf logger.Logf, tunName string) (tun.Device, string, error) {
 	return dev, name, nil
 }
 
+func cleanUpPlan9Interfaces() {
+	maybeUnbind := func(n int) {
+		b, err := os.ReadFile(fmt.Sprintf("/net/ipifc/%d/status", n))
+		if err != nil {
+			return
+		}
+		status := string(b)
+		if !(strings.HasPrefix(status, "device  maxtu ") ||
+			strings.Contains(status, "fd7a:115c:a1e0:")) {
+			return
+		}
+		f, err := os.OpenFile(fmt.Sprintf("/net/ipifc/%d/ctl", n), os.O_RDWR, 0)
+		if err != nil {
+			return
+		}
+		defer f.Close()
+		if _, err := fmt.Fprintf(f, "unbind\n"); err != nil {
+			log.Printf("unbind interface %v: %v", n, err)
+			return
+		}
+		log.Printf("tun: unbound stale interface %v", n)
+	}
+
+	// A common case: after unclean shutdown, the /net/ipifc/clone file
+	for n := 2; n < 5; n++ {
+		maybeUnbind(n)
+	}
+}
+
 // tunDiagnoseFailure, if non-nil, does OS-specific diagnostics of why
 // TUN failed to work.
 var tunDiagnoseFailure func(tunName string, logf logger.Logf, err error)

net/tstun/wrap.go

@@ -928,8 +928,10 @@ func (t *Wrapper) Read(buffs [][]byte, sizes []int, offset int) (int, error) {
 	// packet from OS read and sent to WG
 	res, ok := <-t.vectorOutbound
 	if !ok {
+		t.logf("XXX Wrapper.vectorInbound done")
 		return 0, io.EOF
 	}
+	// t.logf("XXX Wrapper.vec in: err=%v, len(data)=%d, offset=%d", res.err, len(res.data), offset)
 	if res.err != nil && len(res.data) == 0 {
 		return 0, res.err
 	}
@@ -947,6 +949,7 @@ func (t *Wrapper) Read(buffs [][]byte, sizes []int, offset int) (int, error) {
 	var buffsGRO *gro.GRO
 	for _, data := range res.data {
 		p.Decode(data[res.dataOffset:])
+		// t.logf("XXX Wrapper.Read decode (off=%d): %v", res.dataOffset, p.String())
 
 		if m := t.destIPActivity.Load(); m != nil {
 			if fn := m[p.Dst.Addr()]; fn != nil {

portlist/portlist_plan9.go

@@ -0,0 +1,122 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package portlist
+
+import (
+	"bufio"
+	"bytes"
+	"os"
+	"strconv"
+	"strings"
+	"time"
+)
+
+func init() {
+	newOSImpl = newPlan9Impl
+
+	pollInterval = 5 * time.Second
+}
+
+type plan9Impl struct {
+	known map[protoPort]*portMeta // inode string => metadata
+
+	br               *bufio.Reader // reused
+	portsBuf         []Port
+	includeLocalhost bool
+}
+
+type protoPort struct {
+	proto string
+	port  uint16
+}
+
+type portMeta struct {
+	port Port
+	keep bool
+}
+
+func newPlan9Impl(includeLocalhost bool) osImpl {
+	return &plan9Impl{
+		known:            map[protoPort]*portMeta{},
+		br:               bufio.NewReader(bytes.NewReader(nil)),
+		includeLocalhost: includeLocalhost,
+	}
+}
+
+func (*plan9Impl) Close() error { return nil }
+
+func (im *plan9Impl) AppendListeningPorts(base []Port) ([]Port, error) {
+	ret := base
+
+	des, err := os.ReadDir("/proc")
+	if err != nil {
+		return nil, err
+	}
+	for _, de := range des {
+		if !de.IsDir() {
+			continue
+		}
+		pidStr := de.Name()
+		pid, err := strconv.Atoi(pidStr)
+		if err != nil {
+			continue
+		}
+		st, _ := os.ReadFile("/proc/" + pidStr + "/fd")
+		if !bytes.Contains(st, []byte("/net/tcp/clone")) {
+			continue
+		}
+		args, _ := os.ReadFile("/proc/" + pidStr + "/args")
+		procName := string(bytes.TrimSpace(args))
+		// term% cat /proc/417/fd
+		// /usr/glenda
+		//   0 r  M   35 (0000000000000001 0 00) 16384      260 /dev/cons
+		//   1 w  c    0 (000000000000000a 0 00)     0      471 /dev/null
+		//   2 w  M   35 (0000000000000001 0 00) 16384      108 /dev/cons
+		//   3 rw I    0 (000000000000002c 0 00)     0       14 /net/tcp/clone
+		for line := range bytes.Lines(st) {
+			if !bytes.Contains(line, []byte("/net/tcp/clone")) {
+				continue
+			}
+			f := strings.Fields(string(line))
+			if len(f) < 10 {
+				continue
+			}
+			if f[9] != "/net/tcp/clone" {
+				continue
+			}
+			qid, err := strconv.ParseUint(strings.TrimPrefix(f[4], "("), 16, 64)
+			if err != nil {
+				continue
+			}
+			tcpN := (qid >> 5) & (1<<12 - 1)
+			tcpNStr := strconv.FormatUint(tcpN, 10)
+			st, _ := os.ReadFile("/net/tcp/" + tcpNStr + "/status")
+			if !bytes.Contains(st, []byte("Listen ")) {
+				// Unexpected. Or a race.
+				continue
+			}
+			bl, _ := os.ReadFile("/net/tcp/" + tcpNStr + "/local")
+			i := bytes.LastIndexByte(bl, '!')
+			if i == -1 {
+				continue
+			}
+			if bytes.HasPrefix(bl, []byte("127.0.0.1!")) && !im.includeLocalhost {
+				continue
+			}
+			portStr := strings.TrimSpace(string(bl[i+1:]))
+			port, _ := strconv.Atoi(portStr)
+			if port == 0 {
+				continue
+			}
+			ret = append(ret, Port{
+				Proto:   "tcp",
+				Port:    uint16(port),
+				Process: procName,
+				Pid:     pid,
+			})
+		}
+	}
+
+	return ret, nil
+}

prober/derp.go

@@ -596,23 +596,11 @@ 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",
-		Labels: initLabels,
+		Class: "derp_udp",
 	}
 }
 

prober/prober.go

@@ -404,14 +404,10 @@ 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_plan9.go

@@ -7,119 +7,13 @@ package safesocket
 
 import (
 	"context"
-	"fmt"
 	"net"
-	"os"
-	"syscall"
-	"time"
-
-	"golang.org/x/sys/plan9"
 )
 
-// Plan 9's devsrv srv(3) is a server registry  and
-// it is conventionally bound to "/srv" in the default
-// namespace. It is "a one level directory for holding
-// already open channels to services". Post one end of
-// a pipe to "/srv/tailscale.sock" and use the other
-// end for communication with a requestor. Plan 9 pipes
-// are bidirectional.
-
-type plan9SrvAddr string
-
-func (sl plan9SrvAddr) Network() string {
-	return "/srv"
-}
-
-func (sl plan9SrvAddr) String() string {
-	return string(sl)
-}
-
-// There is no net.FileListener for Plan 9 at this time
-type plan9SrvListener struct {
-	name string
-	srvf *os.File
-	file *os.File
-}
-
-func (sl *plan9SrvListener) Accept() (net.Conn, error) {
-	// sl.file is the server end of the pipe that's
-	// connected to /srv/tailscale.sock
-	return plan9FileConn{name: sl.name, file: sl.file}, nil
-}
-
-func (sl *plan9SrvListener) Close() error {
-	sl.file.Close()
-	return sl.srvf.Close()
-}
-
-func (sl *plan9SrvListener) Addr() net.Addr {
-	return plan9SrvAddr(sl.name)
-}
-
-type plan9FileConn struct {
-	name string
-	file *os.File
-}
-
-func (fc plan9FileConn) Read(b []byte) (n int, err error) {
-	return fc.file.Read(b)
-}
-func (fc plan9FileConn) Write(b []byte) (n int, err error) {
-	return fc.file.Write(b)
-}
-func (fc plan9FileConn) Close() error {
-	return fc.file.Close()
-}
-func (fc plan9FileConn) LocalAddr() net.Addr {
-	return plan9SrvAddr(fc.name)
-}
-func (fc plan9FileConn) RemoteAddr() net.Addr {
-	return plan9SrvAddr(fc.name)
-}
-func (fc plan9FileConn) SetDeadline(t time.Time) error {
-	return syscall.EPLAN9
-}
-func (fc plan9FileConn) SetReadDeadline(t time.Time) error {
-	return syscall.EPLAN9
-}
-func (fc plan9FileConn) SetWriteDeadline(t time.Time) error {
-	return syscall.EPLAN9
-}
-
 func connect(_ context.Context, path string) (net.Conn, error) {
-	f, err := os.OpenFile(path, os.O_RDWR, 0666)
-	if err != nil {
-		return nil, err
-	}
-
-	return plan9FileConn{name: path, file: f}, nil
+	return net.Dial("tcp", "localhost:5252")
 }
 
-// Create an entry in /srv, open a pipe, write the
-// client end to the entry and return the server
-// end of the pipe to the caller. When the server
-// end of the pipe is closed, /srv name associated
-// with it will be removed (controlled by ORCLOSE flag)
 func listen(path string) (net.Listener, error) {
-	const O_RCLOSE = 64 // remove on close; should be in plan9 package
-	var pip [2]int
-
-	err := plan9.Pipe(pip[:])
-	if err != nil {
-		return nil, err
-	}
-	defer plan9.Close(pip[1])
-
-	srvfd, err := plan9.Create(path, plan9.O_WRONLY|plan9.O_CLOEXEC|O_RCLOSE, 0600)
-	if err != nil {
-		return nil, err
-	}
-	srv := os.NewFile(uintptr(srvfd), path)
-
-	_, err = fmt.Fprintf(srv, "%d", pip[1])
-	if err != nil {
-		return nil, err
-	}
-
-	return &plan9SrvListener{name: path, srvf: srv, file: os.NewFile(uintptr(pip[0]), path)}, nil
+	return net.Listen("tcp", "localhost:5252")
 }

ssh/tailssh/incubator_plan9.go

@@ -0,0 +1,421 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+// This file contains the plan9-specific version of the incubator. Tailscaled
+// launches the incubator as the same user as it was launched as. The
+// incubator then registers a new session with the OS, sets its UID
+// and groups to the specified `--uid`, `--gid` and `--groups`, and
+// then launches the requested `--cmd`.
+
+package tailssh
+
+import (
+	"encoding/json"
+	"errors"
+	"flag"
+	"fmt"
+	"io"
+	"log"
+	"os"
+	"os/exec"
+	"runtime"
+	"strconv"
+	"strings"
+	"sync/atomic"
+
+	"github.com/go4org/plan9netshell"
+	"github.com/pkg/sftp"
+	"tailscale.com/cmd/tailscaled/childproc"
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/logger"
+)
+
+func init() {
+	childproc.Add("ssh", beIncubator)
+	childproc.Add("sftp", beSFTP)
+	childproc.Add("plan9-netshell", beNetshell)
+}
+
+// newIncubatorCommand returns a new exec.Cmd configured with
+// `tailscaled be-child ssh` as the entrypoint.
+//
+// If ss.srv.tailscaledPath is empty, this method is equivalent to
+// exec.CommandContext.
+//
+// The returned Cmd.Env is guaranteed to be nil; the caller populates it.
+func (ss *sshSession) newIncubatorCommand(logf logger.Logf) (cmd *exec.Cmd, err error) {
+	defer func() {
+		if cmd.Env != nil {
+			panic("internal error")
+		}
+	}()
+
+	var isSFTP, isShell bool
+	switch ss.Subsystem() {
+	case "sftp":
+		isSFTP = true
+	case "":
+		isShell = ss.RawCommand() == ""
+	default:
+		panic(fmt.Sprintf("unexpected subsystem: %v", ss.Subsystem()))
+	}
+
+	if ss.conn.srv.tailscaledPath == "" {
+		if isSFTP {
+			// SFTP relies on the embedded Go-based SFTP server in tailscaled,
+			// so without tailscaled, we can't serve SFTP.
+			return nil, errors.New("no tailscaled found on path, can't serve SFTP")
+		}
+
+		loginShell := ss.conn.localUser.LoginShell()
+		logf("directly running /bin/rc -c %q", ss.RawCommand())
+		return exec.CommandContext(ss.ctx, loginShell, "-c", ss.RawCommand()), nil
+	}
+
+	lu := ss.conn.localUser
+	ci := ss.conn.info
+	remoteUser := ci.uprof.LoginName
+	if ci.node.IsTagged() {
+		remoteUser = strings.Join(ci.node.Tags().AsSlice(), ",")
+	}
+
+	incubatorArgs := []string{
+		"be-child",
+		"ssh",
+		// TODO: "--uid=" + lu.Uid,
+		// TODO: "--gid=" + lu.Gid,
+		"--local-user=" + lu.Username,
+		"--home-dir=" + lu.HomeDir,
+		"--remote-user=" + remoteUser,
+		"--remote-ip=" + ci.src.Addr().String(),
+		"--has-tty=false", // updated in-place by startWithPTY
+		"--tty-name=",     // updated in-place by startWithPTY
+	}
+
+	nm := ss.conn.srv.lb.NetMap()
+	forceV1Behavior := nm.HasCap(tailcfg.NodeAttrSSHBehaviorV1) && !nm.HasCap(tailcfg.NodeAttrSSHBehaviorV2)
+	if forceV1Behavior {
+		incubatorArgs = append(incubatorArgs, "--force-v1-behavior")
+	}
+
+	if debugTest.Load() {
+		incubatorArgs = append(incubatorArgs, "--debug-test")
+	}
+
+	switch {
+	case isSFTP:
+		// Note that we include both the `--sftp` flag and a command to launch
+		// tailscaled as `be-child sftp`. If login or su is available, and
+		// we're not running with tailcfg.NodeAttrSSHBehaviorV1, this will
+		// result in serving SFTP within a login shell, with full PAM
+		// integration. Otherwise, we'll serve SFTP in the incubator process
+		// with no PAM integration.
+		incubatorArgs = append(incubatorArgs, "--sftp", fmt.Sprintf("--cmd=%s be-child sftp", ss.conn.srv.tailscaledPath))
+	case isShell:
+		incubatorArgs = append(incubatorArgs, "--shell")
+	default:
+		incubatorArgs = append(incubatorArgs, "--cmd="+ss.RawCommand())
+	}
+
+	allowSendEnv := nm.HasCap(tailcfg.NodeAttrSSHEnvironmentVariables)
+	if allowSendEnv {
+		env, err := filterEnv(ss.conn.acceptEnv, ss.Session.Environ())
+		if err != nil {
+			return nil, err
+		}
+
+		if len(env) > 0 {
+			encoded, err := json.Marshal(env)
+			if err != nil {
+				return nil, fmt.Errorf("failed to encode environment: %w", err)
+			}
+			incubatorArgs = append(incubatorArgs, fmt.Sprintf("--encoded-env=%q", encoded))
+		}
+	}
+
+	return exec.CommandContext(ss.ctx, ss.conn.srv.tailscaledPath, incubatorArgs...), nil
+}
+
+var debugTest atomic.Bool
+
+type stdRWC struct{}
+
+func (stdRWC) Read(p []byte) (n int, err error) {
+	return os.Stdin.Read(p)
+}
+
+func (stdRWC) Write(b []byte) (n int, err error) {
+	return os.Stdout.Write(b)
+}
+
+func (stdRWC) Close() error {
+	os.Exit(0)
+	return nil
+}
+
+type incubatorArgs struct {
+	localUser          string
+	homeDir            string
+	remoteUser         string
+	remoteIP           string
+	ttyName            string
+	hasTTY             bool
+	cmd                string
+	isSFTP             bool
+	isShell            bool
+	forceV1Behavior    bool
+	debugTest          bool
+	isSELinuxEnforcing bool
+	encodedEnv         string
+}
+
+func parseIncubatorArgs(args []string) (incubatorArgs, error) {
+	var ia incubatorArgs
+
+	flags := flag.NewFlagSet("", flag.ExitOnError)
+	flags.StringVar(&ia.localUser, "local-user", "", "the user to run as")
+	flags.StringVar(&ia.homeDir, "home-dir", "/", "the user's home directory")
+	flags.StringVar(&ia.remoteUser, "remote-user", "", "the remote user/tags")
+	flags.StringVar(&ia.remoteIP, "remote-ip", "", "the remote Tailscale IP")
+	flags.StringVar(&ia.ttyName, "tty-name", "", "the tty name (pts/3)")
+	flags.BoolVar(&ia.hasTTY, "has-tty", false, "is the output attached to a tty")
+	flags.StringVar(&ia.cmd, "cmd", "", "the cmd to launch, including all arguments (ignored in sftp mode)")
+	flags.BoolVar(&ia.isShell, "shell", false, "is launching a shell (with no cmds)")
+	flags.BoolVar(&ia.isSFTP, "sftp", false, "run sftp server (cmd is ignored)")
+	flags.BoolVar(&ia.forceV1Behavior, "force-v1-behavior", false, "allow falling back to the su command if login is unavailable")
+	flags.BoolVar(&ia.debugTest, "debug-test", false, "should debug in test mode")
+	flags.BoolVar(&ia.isSELinuxEnforcing, "is-selinux-enforcing", false, "whether SELinux is in enforcing mode")
+	flags.StringVar(&ia.encodedEnv, "encoded-env", "", "JSON encoded array of environment variables in '['key=value']' format")
+	flags.Parse(args)
+	return ia, nil
+}
+
+func (ia incubatorArgs) forwardedEnviron() ([]string, string, error) {
+	environ := os.Environ()
+	// pass through SSH_AUTH_SOCK environment variable to support ssh agent forwarding
+	allowListKeys := "SSH_AUTH_SOCK"
+
+	if ia.encodedEnv != "" {
+		unquoted, err := strconv.Unquote(ia.encodedEnv)
+		if err != nil {
+			return nil, "", fmt.Errorf("unable to parse encodedEnv %q: %w", ia.encodedEnv, err)
+		}
+
+		var extraEnviron []string
+
+		err = json.Unmarshal([]byte(unquoted), &extraEnviron)
+		if err != nil {
+			return nil, "", fmt.Errorf("unable to parse encodedEnv %q: %w", ia.encodedEnv, err)
+		}
+
+		environ = append(environ, extraEnviron...)
+
+		for _, v := range extraEnviron {
+			allowListKeys = fmt.Sprintf("%s,%s", allowListKeys, strings.Split(v, "=")[0])
+		}
+	}
+
+	return environ, allowListKeys, nil
+}
+
+func beNetshell(args []string) error {
+	plan9netshell.Main()
+	return nil
+}
+
+// beIncubator is the entrypoint to the `tailscaled be-child ssh` subcommand.
+// It is responsible for informing the system of a new login session for the
+// user. This is sometimes necessary for mounting home directories and
+// decrypting file systems.
+//
+// Tailscaled launches the incubator as the same user as it was launched as.
+func beIncubator(args []string) error {
+	// To defend against issues like https://golang.org/issue/1435,
+	// defensively lock our current goroutine's thread to the current
+	// system thread before we start making any UID/GID/group changes.
+	//
+	// This shouldn't matter on Linux because syscall.AllThreadsSyscall is
+	// used to invoke syscalls on all OS threads, but (as of 2023-03-23)
+	// that function is not implemented on all platforms.
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+
+	ia, err := parseIncubatorArgs(args)
+	if err != nil {
+		return err
+	}
+	if ia.isSFTP && ia.isShell {
+		return fmt.Errorf("--sftp and --shell are mutually exclusive")
+	}
+
+	if ia.isShell {
+		plan9netshell.Main()
+		return nil
+	}
+
+	dlogf := logger.Discard
+	if ia.debugTest {
+		// In testing, we don't always have syslog, so log to a temp file.
+		if logFile, err := os.OpenFile("/tmp/tailscalessh.log", os.O_APPEND|os.O_WRONLY, 0666); err == nil {
+			lf := log.New(logFile, "", 0)
+			dlogf = func(msg string, args ...any) {
+				lf.Printf(msg, args...)
+				logFile.Sync()
+			}
+			defer logFile.Close()
+		}
+	}
+
+	return handleInProcess(dlogf, ia)
+}
+
+func handleInProcess(dlogf logger.Logf, ia incubatorArgs) error {
+	if ia.isSFTP {
+		return handleSFTPInProcess(dlogf, ia)
+	}
+	return handleSSHInProcess(dlogf, ia)
+}
+
+func handleSFTPInProcess(dlogf logger.Logf, ia incubatorArgs) error {
+	dlogf("handling sftp")
+
+	return serveSFTP()
+}
+
+// beSFTP serves SFTP in-process.
+func beSFTP(args []string) error {
+	return serveSFTP()
+}
+
+func serveSFTP() error {
+	server, err := sftp.NewServer(stdRWC{})
+	if err != nil {
+		return err
+	}
+	// TODO(https://github.com/pkg/sftp/pull/554): Revert the check for io.EOF,
+	// when sftp is patched to report clean termination.
+	if err := server.Serve(); err != nil && err != io.EOF {
+		return err
+	}
+	return nil
+}
+
+// handleSSHInProcess is a last resort if we couldn't use login or su. It
+// registers a new session with the OS, sets its UID, GID and groups to the
+// specified values, and then launches the requested `--cmd` in the user's
+// login shell.
+func handleSSHInProcess(dlogf logger.Logf, ia incubatorArgs) error {
+
+	environ, _, err := ia.forwardedEnviron()
+	if err != nil {
+		return err
+	}
+
+	dlogf("running /bin/rc -c %q", ia.cmd)
+	cmd := newCommand("/bin/rc", environ, []string{"-c", ia.cmd})
+	err = cmd.Run()
+	if ee, ok := err.(*exec.ExitError); ok {
+		ps := ee.ProcessState
+		code := ps.ExitCode()
+		if code < 0 {
+			// TODO(bradfitz): do we need to also check the syscall.WaitStatus
+			// and make our process look like it also died by signal/same signal
+			// as our child process? For now we just do the exit code.
+			fmt.Fprintf(os.Stderr, "[tailscale-ssh: process died: %v]\n", ps.String())
+			code = 1 // for now. so we don't exit with negative
+		}
+		os.Exit(code)
+	}
+	return err
+}
+
+func newCommand(cmdPath string, cmdEnviron []string, cmdArgs []string) *exec.Cmd {
+	cmd := exec.Command(cmdPath, cmdArgs...)
+	cmd.Stdin = os.Stdin
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+	cmd.Env = cmdEnviron
+
+	return cmd
+}
+
+// launchProcess launches an incubator process for the provided session.
+// It is responsible for configuring the process execution environment.
+// The caller can wait for the process to exit by calling cmd.Wait().
+//
+// It sets ss.cmd, stdin, stdout, and stderr.
+func (ss *sshSession) launchProcess() error {
+	var err error
+	ss.cmd, err = ss.newIncubatorCommand(ss.logf)
+	if err != nil {
+		return err
+	}
+
+	cmd := ss.cmd
+	cmd.Dir = "/"
+	cmd.Env = append(os.Environ(), envForUser(ss.conn.localUser)...)
+	for _, kv := range ss.Environ() {
+		if acceptEnvPair(kv) {
+			cmd.Env = append(cmd.Env, kv)
+		}
+	}
+
+	ci := ss.conn.info
+	cmd.Env = append(cmd.Env,
+		fmt.Sprintf("SSH_CLIENT=%s %d %d", ci.src.Addr(), ci.src.Port(), ci.dst.Port()),
+		fmt.Sprintf("SSH_CONNECTION=%s %d %s %d", ci.src.Addr(), ci.src.Port(), ci.dst.Addr(), ci.dst.Port()),
+	)
+
+	if ss.agentListener != nil {
+		cmd.Env = append(cmd.Env, fmt.Sprintf("SSH_AUTH_SOCK=%s", ss.agentListener.Addr()))
+	}
+
+	return ss.startWithStdPipes()
+}
+
+// startWithStdPipes starts cmd with os.Pipe for Stdin, Stdout and Stderr.
+func (ss *sshSession) startWithStdPipes() (err error) {
+	var rdStdin, wrStdout, wrStderr io.ReadWriteCloser
+	defer func() {
+		if err != nil {
+			closeAll(rdStdin, ss.wrStdin, ss.rdStdout, wrStdout, ss.rdStderr, wrStderr)
+		}
+	}()
+	if ss.cmd == nil {
+		return errors.New("nil cmd")
+	}
+	if rdStdin, ss.wrStdin, err = os.Pipe(); err != nil {
+		return err
+	}
+	if ss.rdStdout, wrStdout, err = os.Pipe(); err != nil {
+		return err
+	}
+	if ss.rdStderr, wrStderr, err = os.Pipe(); err != nil {
+		return err
+	}
+	ss.cmd.Stdin = rdStdin
+	ss.cmd.Stdout = wrStdout
+	ss.cmd.Stderr = wrStderr
+	ss.childPipes = []io.Closer{rdStdin, wrStdout, wrStderr}
+	return ss.cmd.Start()
+}
+
+func envForUser(u *userMeta) []string {
+	return []string{
+		fmt.Sprintf("user=%s", u.Username),
+		fmt.Sprintf("home=%s", u.HomeDir),
+		fmt.Sprintf("path=%s", defaultPathForUser(&u.User)),
+	}
+}
+
+// acceptEnvPair reports whether the environment variable key=value pair
+// should be accepted from the client. It uses the same default as OpenSSH
+// AcceptEnv.
+func acceptEnvPair(kv string) bool {
+	k, _, ok := strings.Cut(kv, "=")
+	if !ok {
+		return false
+	}
+	_ = k
+	return true // permit anything on plan9 during bringup, for debugging at least
+}

ssh/tailssh/tailssh.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build linux || (darwin && !ios) || freebsd || openbsd
+//go:build linux || (darwin && !ios) || freebsd || openbsd || plan9
 
 // Package tailssh is an SSH server integrated into Tailscale.
 package tailssh
@@ -672,7 +672,6 @@ type sshSession struct {
 	wrStdin  io.WriteCloser
 	rdStdout io.ReadCloser
 	rdStderr io.ReadCloser // rdStderr is nil for pty sessions
-	ptyReq   *ssh.Pty      // non-nil for pty sessions
 
 	// childPipes is a list of pipes that need to be closed when the process exits.
 	// For pty sessions, this is the tty fd.
@@ -903,7 +902,7 @@ func (ss *sshSession) run() {
 		defer t.Stop()
 	}
 
-	if euid := os.Geteuid(); euid != 0 {
+	if euid := os.Geteuid(); euid != 0 && runtime.GOOS != "plan9" {
 		if lu.Uid != fmt.Sprint(euid) {
 			ss.logf("can't switch to user %q from process euid %v", lu.Username, euid)
 			fmt.Fprintf(ss, "can't switch user\r\n")

ssh/tailssh/user.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build linux || (darwin && !ios) || freebsd || openbsd
+//go:build linux || (darwin && !ios) || freebsd || openbsd || plan9
 
 package tailssh
 
@@ -48,6 +48,9 @@ func userLookup(username string) (*userMeta, error) {
 }
 
 func (u *userMeta) LoginShell() string {
+	if runtime.GOOS == "plan9" {
+		return "/bin/rc"
+	}
 	if u.loginShellCached != "" {
 		// This field should be populated on Linux, at least, because
 		// func userLookup on Linux uses "getent" to look up the user
@@ -85,6 +88,9 @@ func defaultPathForUser(u *user.User) string {
 	if s := defaultPathTmpl(); s != "" {
 		return expandDefaultPathTmpl(s, u)
 	}
+	if runtime.GOOS == "plan9" {
+		return "/bin"
+	}
 	isRoot := u.Uid == "0"
 	switch distro.Get() {
 	case distro.Debian:

tsnet/tsnet.go

@@ -505,11 +505,6 @@ 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
 		}

tstest/integration/testcontrol/testcontrol.go

@@ -839,17 +839,15 @@ func (s *Server) serveMap(w http.ResponseWriter, r *http.Request, mkey key.Machi
 
 	w.WriteHeader(200)
 	for {
-		// 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
-				}
+		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 {
 				continue
 			}
+			return
 		}
 
 		if s.canGenerateAutomaticMapResponseFor(req.NodeKey) {

types/logger/logger.go

@@ -14,6 +14,7 @@ import (
 	"fmt"
 	"io"
 	"log"
+	"runtime"
 	"strings"
 	"sync"
 	"time"
@@ -162,6 +163,10 @@ func RateLimitedFnWithClock(logf Logf, f time.Duration, burst int, maxCache int,
 	if envknob.String("TS_DEBUG_LOG_RATE") == "all" {
 		return logf
 	}
+	if runtime.GOOS == "plan9" {
+		// To ease bring-up.
+		return logf
+	}
 	var (
 		mu       sync.Mutex
 		msgLim   = make(map[string]*limitData) // keyed by logf format

util/osuser/group_ids.go

@@ -19,6 +19,10 @@ import (
 // an error. It will first try to use the 'id' command to get the group IDs,
 // and if that fails, it will fall back to the user.GroupIds method.
 func GetGroupIds(user *user.User) ([]string, error) {
+	if runtime.GOOS == "plan9" {
+		return nil, nil
+	}
+
 	if runtime.GOOS != "linux" {
 		return user.GroupIds()
 	}

util/osuser/user.go

@@ -54,9 +54,18 @@ func lookup(usernameOrUID string, std lookupStd, wantShell bool) (*user.User, st
 	// Skip getent entirely on Non-Unix platforms that won't ever have it.
 	// (Using HasPrefix for "wasip1", anticipating that WASI support will
 	// move beyond "preview 1" some day.)
-	if runtime.GOOS == "windows" || runtime.GOOS == "js" || runtime.GOARCH == "wasm" {
+	if runtime.GOOS == "windows" || runtime.GOOS == "js" || runtime.GOARCH == "wasm" || runtime.GOOS == "plan9" {
+		var shell string
+		if wantShell && runtime.GOOS == "plan9" {
+			shell = "/bin/rc"
+		}
+		if runtime.GOOS == "plan9" {
+			if u, err := user.Current(); err == nil {
+				return u, shell, nil
+			}
+		}
 		u, err := std(usernameOrUID)
-		return u, "", err
+		return u, shell, err
 	}
 
 	// No getent on Gokrazy. So hard-code the login shell.
@@ -78,6 +87,16 @@ func lookup(usernameOrUID string, std lookupStd, wantShell bool) (*user.User, st
 		return u, shell, nil
 	}
 
+	if runtime.GOOS == "plan9" {
+		return &user.User{
+			Uid:      "0",
+			Gid:      "0",
+			Username: "glenda",
+			Name:     "Glenda",
+			HomeDir:  "/",
+		}, "/bin/rc", nil
+	}
+
 	// Start with getent if caller wants to get the user shell.
 	if wantShell {
 		return userLookupGetent(usernameOrUID, std)

wgengine/magicsock/magicsock.go

@@ -3018,6 +3018,10 @@ func (c *Conn) DebugForcePreferDERP(n int) {
 // portableTrySetSocketBuffer sets SO_SNDBUF and SO_RECVBUF on pconn to socketBufferSize,
 // logging an error if it occurs.
 func portableTrySetSocketBuffer(pconn nettype.PacketConn, logf logger.Logf) {
+	if runtime.GOOS == "plan9" {
+		// Not supported. Don't try. Avoid logspam.
+		return
+	}
 	if c, ok := pconn.(*net.UDPConn); ok {
 		// Attempt to increase the buffer size, and allow failures.
 		if err := c.SetReadBuffer(socketBufferSize); err != nil {

wgengine/magicsock/rebinding_conn.go

@@ -7,9 +7,11 @@ import (
 	"errors"
 	"net"
 	"net/netip"
+	"runtime"
 	"sync"
 	"sync/atomic"
 	"syscall"
+	"time"
 
 	"golang.org/x/net/ipv6"
 	"tailscale.com/net/netaddr"
@@ -150,6 +152,12 @@ func (c *RebindingUDPConn) closeLocked() error {
 		return errNilPConn
 	}
 	c.port = 0
+	if runtime.GOOS == "plan9" {
+		// Work around Go bug https://github.com/golang/go/issues/72770.
+		// This does https://go-review.googlesource.com/c/go/+/656395
+		// manually until the upstream Go bug is fixed + released.
+		c.pconn.SetReadDeadline(time.Now().Add(-time.Hour))
+	}
 	return c.pconn.Close()
 }
 

wgengine/router/router_default.go

@@ -1,7 +1,7 @@
 // Copyright (c) Tailscale Inc & AUTHORS
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build !windows && !linux && !darwin && !openbsd && !freebsd
+//go:build !windows && !linux && !darwin && !openbsd && !freebsd && !plan9
 
 package router
 

wgengine/router/router_plan9.go

@@ -0,0 +1,158 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package router
+
+import (
+	"bufio"
+	"bytes"
+	"fmt"
+	"net/netip"
+	"os"
+	"strings"
+
+	"github.com/tailscale/wireguard-go/tun"
+	"tailscale.com/health"
+	"tailscale.com/net/netmon"
+	"tailscale.com/types/logger"
+)
+
+func newUserspaceRouter(logf logger.Logf, tundev tun.Device, netMon *netmon.Monitor, health *health.Tracker) (Router, error) {
+	r := &plan9Router{
+		logf:   logf,
+		tundev: tundev,
+		netMon: netMon,
+	}
+	cleanAllTailscaleRoutes(logf)
+	return r, nil
+}
+
+type plan9Router struct {
+	logf   logger.Logf
+	tundev tun.Device
+	netMon *netmon.Monitor
+	health *health.Tracker
+}
+
+func (r *plan9Router) Up() error {
+	return nil
+}
+
+func (r *plan9Router) Set(cfg *Config) error {
+	if cfg == nil {
+		cleanAllTailscaleRoutes(r.logf)
+		return nil
+	}
+
+	var self4, self6 netip.Addr
+	for _, addr := range cfg.LocalAddrs {
+		ctl := r.tundev.File()
+		maskBits := addr.Bits()
+		if addr.Addr().Is4() {
+			// The mask sizes in Plan9 are in IPv6 bits, even for IPv4.
+			maskBits += (128 - 32)
+			self4 = addr.Addr()
+		}
+		if addr.Addr().Is6() {
+			self6 = addr.Addr()
+		}
+		_, err := fmt.Fprintf(ctl, "add %s /%d\n", addr.Addr().String(), maskBits)
+		r.logf("route/plan9: add %s /%d = %v", addr.Addr().String(), maskBits, err)
+	}
+
+	ipr, err := os.OpenFile("/net/iproute", os.O_RDWR, 0)
+	if err != nil {
+		return fmt.Errorf("open /net/iproute: %w", err)
+	}
+	defer ipr.Close()
+
+	// TODO(bradfitz): read existing routes, delete ones tagged "tail"
+	// that aren't in cfg.LocalRoutes.
+
+	if _, err := fmt.Fprintf(ipr, "tag tail\n"); err != nil {
+		return fmt.Errorf("tag tail: %w", err)
+	}
+
+	for _, route := range cfg.Routes {
+		maskBits := route.Bits()
+		if route.Addr().Is4() {
+			// The mask sizes in Plan9 are in IPv6 bits, even for IPv4.
+			maskBits += (128 - 32)
+		}
+		var nextHop netip.Addr
+		if route.Addr().Is4() {
+			nextHop = self4
+		} else if route.Addr().Is6() {
+			nextHop = self6
+		}
+		if !nextHop.IsValid() {
+			r.logf("route/plan9: skipping route %s: no next hop (no self addr)", route.String())
+			continue
+		}
+		r.logf("route/plan9: plan9.router: add %s /%d %s", route.Addr(), maskBits, nextHop)
+		if _, err := fmt.Fprintf(ipr, "add %s /%d %s\n", route.Addr(), maskBits, nextHop); err != nil {
+			return fmt.Errorf("add %s: %w", route.String(), err)
+		}
+	}
+
+	if len(cfg.LocalRoutes) > 0 {
+		r.logf("route/plan9: TODO: Set LocalRoutes %v", cfg.LocalRoutes)
+	}
+	if len(cfg.SubnetRoutes) > 0 {
+		r.logf("route/plan9: TODO: Set SubnetRoutes %v", cfg.SubnetRoutes)
+	}
+
+	return nil
+}
+
+// UpdateMagicsockPort implements the Router interface. This implementation
+// does nothing and returns nil because this router does not currently need
+// to know what the magicsock UDP port is.
+func (r *plan9Router) UpdateMagicsockPort(_ uint16, _ string) error {
+	return nil
+}
+
+func (r *plan9Router) Close() error {
+	// TODO(bradfitz): unbind
+	return nil
+}
+
+func cleanUp(logf logger.Logf, _ string) {
+	cleanAllTailscaleRoutes(logf)
+}
+
+func cleanAllTailscaleRoutes(logf logger.Logf) {
+	routes, err := os.OpenFile("/net/iproute", os.O_RDWR, 0)
+	if err != nil {
+		logf("cleaning routes: %v", err)
+		return
+	}
+	defer routes.Close()
+
+	// Using io.ReadAll or os.ReadFile on /net/iproute fails; it results in a
+	// 511 byte result when the actual /net/iproute contents are over 1k.
+	// So do it in one big read instead. Who knows.
+	routeBuf := make([]byte, 1<<20)
+	n, err := routes.Read(routeBuf)
+	if err != nil {
+		logf("cleaning routes: %v", err)
+		return
+	}
+	routeBuf = routeBuf[:n]
+
+	//logf("cleaning routes: %d bytes: %q", len(routeBuf), routeBuf)
+
+	bs := bufio.NewScanner(bytes.NewReader(routeBuf))
+	for bs.Scan() {
+		f := strings.Fields(bs.Text())
+		if len(f) < 6 {
+			continue
+		}
+		tag := f[4]
+		if tag != "tail" {
+			continue
+		}
+		_, err := fmt.Fprintf(routes, "remove %s %s\n", f[0], f[1])
+		logf("router: cleaning route %s %s: %v", f[0], f[1], err)
+	}
+}

wgengine/userspace.go

@@ -569,6 +569,18 @@ func (e *userspaceEngine) handleLocalPackets(p *packet.Parsed, t *tstun.Wrapper)
 			return filter.Drop
 		}
 	}
+	if runtime.GOOS == "plan9" {
+		isLocalAddr, ok := e.isLocalAddr.LoadOk()
+		if ok {
+			if isLocalAddr(p.Dst.Addr()) {
+				e.logf("XXX plan9 inject inbound")
+				// On Plan9's "tun" equivalent, everything goes back in and out
+				// the tun, even when the kernel's replying to itself.
+				t.InjectInboundCopy(p.Buffer())
+				return filter.Drop
+			}
+		}
+	}
 
 	return filter.Accept
 }