cmd/{k8s-operator,containerboot},kube/egressservices: fix Pod IP check for dual stack clusters (#13721)

Currently egress Services for ProxyGroup only work for Pods and Services
with IPv4 addresses. Ensure that it works on dual stack clusters by reading
proxy Pod's IP from the .status.podIPs list that always contains both
IPv4 and IPv6 address (if the Pod has them) rather than .status.podIP that
could contain IPv6 only for a dual stack cluster.

Updates tailscale/tailscale#13406

Signed-off-by: Irbe Krumina <irbe@tailscale.com>

cmd/containerboot/main.go

@@ -132,36 +132,9 @@ func newNetfilterRunner(logf logger.Logf) (linuxfw.NetfilterRunner, error) {
 func main() {
 	log.SetPrefix("boot: ")
 	tailscale.I_Acknowledge_This_API_Is_Unstable = true
-	cfg := &settings{
-		AuthKey:                               defaultEnvs([]string{"TS_AUTHKEY", "TS_AUTH_KEY"}, ""),
-		Hostname:                              defaultEnv("TS_HOSTNAME", ""),
-		Routes:                                defaultEnvStringPointer("TS_ROUTES"),
-		ServeConfigPath:                       defaultEnv("TS_SERVE_CONFIG", ""),
-		ProxyTargetIP:                         defaultEnv("TS_DEST_IP", ""),
-		ProxyTargetDNSName:                    defaultEnv("TS_EXPERIMENTAL_DEST_DNS_NAME", ""),
-		TailnetTargetIP:                       defaultEnv("TS_TAILNET_TARGET_IP", ""),
-		TailnetTargetFQDN:                     defaultEnv("TS_TAILNET_TARGET_FQDN", ""),
-		DaemonExtraArgs:                       defaultEnv("TS_TAILSCALED_EXTRA_ARGS", ""),
-		ExtraArgs:                             defaultEnv("TS_EXTRA_ARGS", ""),
-		InKubernetes:                          os.Getenv("KUBERNETES_SERVICE_HOST") != "",
-		UserspaceMode:                         defaultBool("TS_USERSPACE", true),
-		StateDir:                              defaultEnv("TS_STATE_DIR", ""),
-		AcceptDNS:                             defaultEnvBoolPointer("TS_ACCEPT_DNS"),
-		KubeSecret:                            defaultEnv("TS_KUBE_SECRET", "tailscale"),
-		SOCKSProxyAddr:                        defaultEnv("TS_SOCKS5_SERVER", ""),
-		HTTPProxyAddr:                         defaultEnv("TS_OUTBOUND_HTTP_PROXY_LISTEN", ""),
-		Socket:                                defaultEnv("TS_SOCKET", "/tmp/tailscaled.sock"),
-		AuthOnce:                              defaultBool("TS_AUTH_ONCE", false),
-		Root:                                  defaultEnv("TS_TEST_ONLY_ROOT", "/"),
-		TailscaledConfigFilePath:              tailscaledConfigFilePath(),
-		AllowProxyingClusterTrafficViaIngress: defaultBool("EXPERIMENTAL_ALLOW_PROXYING_CLUSTER_TRAFFIC_VIA_INGRESS", false),
-		PodIP:                                 defaultEnv("POD_IP", ""),
-		EnableForwardingOptimizations:         defaultBool("TS_EXPERIMENTAL_ENABLE_FORWARDING_OPTIMIZATIONS", false),
-		HealthCheckAddrPort:                   defaultEnv("TS_HEALTHCHECK_ADDR_PORT", ""),
-		EgressSvcsCfgPath:                     defaultEnv("TS_EGRESS_SERVICES_CONFIG_PATH", ""),
-	}
 
-	if err := cfg.validate(); err != nil {
+	cfg, err := configFromEnv()
+	if err != nil {
 		log.Fatalf("invalid configuration: %v", err)
 	}
 
@@ -612,7 +585,7 @@ func main() {
 							kc:           kc,
 							stateSecret:  cfg.KubeSecret,
 							netmapChan:   egressSvcsNotify,
-							podIP:        cfg.PodIP,
+							podIPv4:      cfg.PodIPv4,
 							tailnetAddrs: addrs,
 						}
 						go func() {

cmd/containerboot/services.go

@@ -46,7 +46,7 @@ type egressProxy struct {
 
 	netmapChan chan ipn.Notify // chan to receive netmap updates on
 
-	podIP string // never empty string
+	podIPv4 string // never empty string, currently only IPv4 is supported
 
 	// tailnetFQDNs is the egress service FQDN to tailnet IP mappings that
 	// were last used to configure firewall rules for this proxy.
@@ -361,7 +361,7 @@ func (ep *egressProxy) getStatus(ctx context.Context) (*egressservices.Status, e
 	if err := json.Unmarshal([]byte(raw), status); err != nil {
 		return nil, fmt.Errorf("error unmarshalling previous config: %w", err)
 	}
-	if reflect.DeepEqual(status.PodIP, ep.podIP) {
+	if reflect.DeepEqual(status.PodIPv4, ep.podIPv4) {
 		return status, nil
 	}
 	return nil, nil
@@ -374,7 +374,7 @@ func (ep *egressProxy) setStatus(ctx context.Context, status *egressservices.Sta
 	if status == nil {
 		status = &egressservices.Status{}
 	}
-	status.PodIP = ep.podIP
+	status.PodIPv4 = ep.podIPv4
 	secret, err := ep.kc.GetSecret(ctx, ep.stateSecret)
 	if err != nil {
 		return fmt.Errorf("error retrieving state Secret: %w", err)
@@ -565,7 +565,7 @@ func servicesStatusIsEqual(st, st1 *egressservices.Status) bool {
 	if st == nil || st1 == nil {
 		return false
 	}
-	st.PodIP = ""
+	st.PodIPv4 = ""
-	st1.PodIP = ""
+	st1.PodIPv4 = ""
 	return reflect.DeepEqual(*st, *st1)
 }

cmd/containerboot/settings.go

@@ -14,6 +14,7 @@
 	"os"
 	"path"
 	"strconv"
+	"strings"
 
 	"tailscale.com/ipn/conffile"
 	"tailscale.com/kube/kubeclient"
@@ -62,11 +63,67 @@ type settings struct {
 	// PodIP is the IP of the Pod if running in Kubernetes. This is used
 	// when setting up rules to proxy cluster traffic to cluster ingress
 	// target.
+	// Deprecated: use PodIPv4, PodIPv6 instead to support dual stack clusters
 	PodIP               string
+	PodIPv4             string
+	PodIPv6             string
 	HealthCheckAddrPort string
 	EgressSvcsCfgPath   string
 }
 
+func configFromEnv() (*settings, error) {
+	cfg := &settings{
+		AuthKey:                               defaultEnvs([]string{"TS_AUTHKEY", "TS_AUTH_KEY"}, ""),
+		Hostname:                              defaultEnv("TS_HOSTNAME", ""),
+		Routes:                                defaultEnvStringPointer("TS_ROUTES"),
+		ServeConfigPath:                       defaultEnv("TS_SERVE_CONFIG", ""),
+		ProxyTargetIP:                         defaultEnv("TS_DEST_IP", ""),
+		ProxyTargetDNSName:                    defaultEnv("TS_EXPERIMENTAL_DEST_DNS_NAME", ""),
+		TailnetTargetIP:                       defaultEnv("TS_TAILNET_TARGET_IP", ""),
+		TailnetTargetFQDN:                     defaultEnv("TS_TAILNET_TARGET_FQDN", ""),
+		DaemonExtraArgs:                       defaultEnv("TS_TAILSCALED_EXTRA_ARGS", ""),
+		ExtraArgs:                             defaultEnv("TS_EXTRA_ARGS", ""),
+		InKubernetes:                          os.Getenv("KUBERNETES_SERVICE_HOST") != "",
+		UserspaceMode:                         defaultBool("TS_USERSPACE", true),
+		StateDir:                              defaultEnv("TS_STATE_DIR", ""),
+		AcceptDNS:                             defaultEnvBoolPointer("TS_ACCEPT_DNS"),
+		KubeSecret:                            defaultEnv("TS_KUBE_SECRET", "tailscale"),
+		SOCKSProxyAddr:                        defaultEnv("TS_SOCKS5_SERVER", ""),
+		HTTPProxyAddr:                         defaultEnv("TS_OUTBOUND_HTTP_PROXY_LISTEN", ""),
+		Socket:                                defaultEnv("TS_SOCKET", "/tmp/tailscaled.sock"),
+		AuthOnce:                              defaultBool("TS_AUTH_ONCE", false),
+		Root:                                  defaultEnv("TS_TEST_ONLY_ROOT", "/"),
+		TailscaledConfigFilePath:              tailscaledConfigFilePath(),
+		AllowProxyingClusterTrafficViaIngress: defaultBool("EXPERIMENTAL_ALLOW_PROXYING_CLUSTER_TRAFFIC_VIA_INGRESS", false),
+		PodIP:                                 defaultEnv("POD_IP", ""),
+		EnableForwardingOptimizations:         defaultBool("TS_EXPERIMENTAL_ENABLE_FORWARDING_OPTIMIZATIONS", false),
+		HealthCheckAddrPort:                   defaultEnv("TS_HEALTHCHECK_ADDR_PORT", ""),
+		EgressSvcsCfgPath:                     defaultEnv("TS_EGRESS_SERVICES_CONFIG_PATH", ""),
+	}
+	podIPs, ok := os.LookupEnv("POD_IPS")
+	if ok {
+		ips := strings.Split(podIPs, ",")
+		if len(ips) > 2 {
+			return nil, fmt.Errorf("POD_IPs can contain at most 2 IPs, got %d (%v)", len(ips), ips)
+		}
+		for _, ip := range ips {
+			parsed, err := netip.ParseAddr(ip)
+			if err != nil {
+				return nil, fmt.Errorf("error parsing IP address %s: %w", ip, err)
+			}
+			if parsed.Is4() {
+				cfg.PodIPv4 = parsed.String()
+				continue
+			}
+			cfg.PodIPv6 = parsed.String()
+		}
+	}
+	if err := cfg.validate(); err != nil {
+		return nil, fmt.Errorf("invalid configuration: %v", err)
+	}
+	return cfg, nil
+}
+
 func (s *settings) validate() error {
 	if s.TailscaledConfigFilePath != "" {
 		dir, file := path.Split(s.TailscaledConfigFilePath)

cmd/k8s-operator/egress-eps.go

@@ -9,6 +9,7 @@
 	"context"
 	"encoding/json"
 	"fmt"
+	"net/netip"
 	"reflect"
 	"strings"
 
@@ -132,6 +133,19 @@ func (er *egressEpsReconciler) Reconcile(ctx context.Context, req reconcile.Requ
 	return res, nil
 }
 
+func podIPv4(pod *corev1.Pod) (string, error) {
+	for _, ip := range pod.Status.PodIPs {
+		parsed, err := netip.ParseAddr(ip.IP)
+		if err != nil {
+			return "", fmt.Errorf("error parsing IP address %s: %w", ip, err)
+		}
+		if parsed.Is4() {
+			return parsed.String(), nil
+		}
+	}
+	return "", nil
+}
+
 // podIsReadyToRouteTraffic returns true if it appears that the proxy Pod has configured firewall rules to be able to
 // route traffic to the given tailnet service. It retrieves the proxy's state Secret and compares the tailnet service
 // status written there to the desired service configuration.
@@ -142,14 +156,21 @@ func (er *egressEpsReconciler) podIsReadyToRouteTraffic(ctx context.Context, pod
 		l.Debugf("proxy Pod is being deleted, ignore")
 		return false, nil
 	}
-	podIP := pod.Status.PodIP
+	podIP, err := podIPv4(&pod)
+	if err != nil {
+		return false, fmt.Errorf("error determining Pod IP address: %v", err)
+	}
+	if podIP == "" {
+		l.Infof("[unexpected] Pod does not have an IPv4 address, and IPv6 is not currently supported")
+		return false, nil
+	}
 	stateS := &corev1.Secret{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      pod.Name,
 			Namespace: pod.Namespace,
 		},
 	}
-	err := er.Get(ctx, client.ObjectKeyFromObject(stateS), stateS)
+	err = er.Get(ctx, client.ObjectKeyFromObject(stateS), stateS)
 	if apierrors.IsNotFound(err) {
 		l.Debugf("proxy does not have a state Secret, waiting...")
 		return false, nil
@@ -166,8 +187,8 @@ func (er *egressEpsReconciler) podIsReadyToRouteTraffic(ctx context.Context, pod
 	if err := json.Unmarshal(svcStatusBS, svcStatus); err != nil {
 		return false, fmt.Errorf("error unmarshalling egress service status: %w", err)
 	}
-	if !strings.EqualFold(podIP, svcStatus.PodIP) {
+	if !strings.EqualFold(podIP, svcStatus.PodIPv4) {
-		l.Infof("proxy's egress service status is for Pod IP %s, current proxy's Pod IP %s, waiting for the proxy to reconfigure...", svcStatus.PodIP, podIP)
+		l.Infof("proxy's egress service status is for Pod IP %s, current proxy's Pod IP %s, waiting for the proxy to reconfigure...", svcStatus.PodIPv4, podIP)
 		return false, nil
 	}
 	st, ok := (*svcStatus).Services[tailnetSvcName]

cmd/k8s-operator/egress-eps_test.go

@@ -98,7 +98,7 @@ func TestTailscaleEgressEndpointSlices(t *testing.T) {
 
 	t.Run("pods_are_ready_to_route_traffic", func(t *testing.T) {
 		pod, stateS := podAndSecretForProxyGroup("foo")
-		stBs := serviceStatusForPodIP(t, svc, pod.Status.PodIP, port)
+		stBs := serviceStatusForPodIP(t, svc, pod.Status.PodIPs[0].IP, port)
 		mustUpdate(t, fc, "operator-ns", stateS.Name, func(s *corev1.Secret) {
 			mak.Set(&s.Data, egressservices.KeyEgressServices, stBs)
 		})
@@ -114,6 +114,16 @@ func TestTailscaleEgressEndpointSlices(t *testing.T) {
 		})
 		expectEqual(t, fc, eps, nil)
 	})
+	t.Run("status_does_not_match_pod_ip", func(t *testing.T) {
+		_, stateS := podAndSecretForProxyGroup("foo")           // replica Pod has IP 10.0.0.1
+		stBs := serviceStatusForPodIP(t, svc, "10.0.0.2", port) // status is for a Pod with IP 10.0.0.2
+		mustUpdate(t, fc, "operator-ns", stateS.Name, func(s *corev1.Secret) {
+			mak.Set(&s.Data, egressservices.KeyEgressServices, stBs)
+		})
+		expectReconciled(t, er, "operator-ns", "foo")
+		eps.Endpoints = []discoveryv1.Endpoint{}
+		expectEqual(t, fc, eps, nil)
+	})
 }
 
 func configMapForSvc(t *testing.T, svc *corev1.Service, p uint16) *corev1.ConfigMap {
@@ -162,7 +172,7 @@ func serviceStatusForPodIP(t *testing.T, svc *corev1.Service, ip string, p uint1
 	}
 	svcName := tailnetSvcName(svc)
 	st := egressservices.Status{
-		PodIP:    ip,
+		PodIPv4:  ip,
 		Services: map[string]*egressservices.ServiceStatus{svcName: &svcSt},
 	}
 	bs, err := json.Marshal(st)
@@ -181,7 +191,9 @@ func podAndSecretForProxyGroup(pg string) (*corev1.Pod, *corev1.Secret) {
 			UID:       "foo",
 		},
 		Status: corev1.PodStatus{
-			PodIP: "10.0.0.1",
+			PodIPs: []corev1.PodIP{
+				{IP: "10.0.0.1"},
+			},
 		},
 	}
 	s := &corev1.Secret{

cmd/k8s-operator/proxygroup_specs.go

@@ -93,10 +93,11 @@ func pgStatefulSet(pg *tsapi.ProxyGroup, namespace, image, tsFirewallMode, cfgHa
 							Env: func() []corev1.EnvVar {
 								envs := []corev1.EnvVar{
 									{
-										Name: "POD_IP",
+										// TODO(irbekrm): verify that .status.podIPs are always set, else read in .status.podIP as well.
+										Name: "POD_IPS", // this will be a comma separate list i.e 10.136.0.6,2600:1900:4011:161:0:e:0:6
 										ValueFrom: &corev1.EnvVarSource{
 											FieldRef: &corev1.ObjectFieldSelector{
-												FieldPath: "status.podIP",
+												FieldPath: "status.podIPs",
 											},
 										},
 									},

kube/egressservices/egressservices.go

@@ -86,7 +86,7 @@ func (pm PortMap) MarshalText() ([]byte, error) {
 // Status represents the currently configured firewall rules for all egress
 // services for a proxy identified by the PodIP.
 type Status struct {
-	PodIP string `json:"podIP"`
+	PodIPv4 string `json:"podIPv4"`
 	// All egress service status keyed by service name.
 	Services map[string]*ServiceStatus `json:"services"`
 }