Tom Proctor 138a83efe1
cmd/containerboot: wait for consistent state on shutdown (#14263)
tailscaled's ipn package writes a collection of keys to state after
authenticating to control, but one at a time. If containerboot happens
to send a SIGTERM signal to tailscaled in the middle of writing those
keys, it may shut down with an inconsistent state Secret and never
recover. While we can't durably fix this with our current single-use
auth keys (no atomic operation to auth + write state), we can reduce
the window for this race condition by checking for partial state
before sending SIGTERM to tailscaled. Best effort only.

Updates #14080

Change-Id: I0532d51b6f0b7d391e538468bd6a0a80dbe1d9f7
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>
2025-01-30 13:51:10 +00:00

195 lines
6.2 KiB
Go

// Copyright (c) Tailscale Inc & AUTHORS
// SPDX-License-Identifier: BSD-3-Clause
//go:build linux
package main
import (
"context"
"encoding/json"
"errors"
"fmt"
"log"
"net/http"
"net/netip"
"os"
"strings"
"time"
"tailscale.com/ipn"
"tailscale.com/kube/kubeapi"
"tailscale.com/kube/kubeclient"
"tailscale.com/kube/kubetypes"
"tailscale.com/logtail/backoff"
"tailscale.com/tailcfg"
"tailscale.com/types/logger"
)
// kubeClient is a wrapper around Tailscale's internal kube client that knows how to talk to the kube API server. We use
// this rather than any of the upstream Kubernetes client libaries to avoid extra imports.
type kubeClient struct {
kubeclient.Client
stateSecret string
canPatch bool // whether the client has permissions to patch Kubernetes Secrets
}
func newKubeClient(root string, stateSecret string) (*kubeClient, error) {
if root != "/" {
// If we are running in a test, we need to set the root path to the fake
// service account directory.
kubeclient.SetRootPathForTesting(root)
}
var err error
kc, err := kubeclient.New("tailscale-container")
if err != nil {
return nil, fmt.Errorf("Error creating kube client: %w", err)
}
if (root != "/") || os.Getenv("TS_KUBERNETES_READ_API_SERVER_ADDRESS_FROM_ENV") == "true" {
// Derive the API server address from the environment variables
// Used to set http server in tests, or optionally enabled by flag
kc.SetURL(fmt.Sprintf("https://%s:%s", os.Getenv("KUBERNETES_SERVICE_HOST"), os.Getenv("KUBERNETES_SERVICE_PORT_HTTPS")))
}
return &kubeClient{Client: kc, stateSecret: stateSecret}, nil
}
// storeDeviceID writes deviceID to 'device_id' data field of the client's state Secret.
func (kc *kubeClient) storeDeviceID(ctx context.Context, deviceID tailcfg.StableNodeID) error {
s := &kubeapi.Secret{
Data: map[string][]byte{
kubetypes.KeyDeviceID: []byte(deviceID),
},
}
return kc.StrategicMergePatchSecret(ctx, kc.stateSecret, s, "tailscale-container")
}
// storeDeviceEndpoints writes device's tailnet IPs and MagicDNS name to fields 'device_ips', 'device_fqdn' of client's
// state Secret.
func (kc *kubeClient) storeDeviceEndpoints(ctx context.Context, fqdn string, addresses []netip.Prefix) error {
var ips []string
for _, addr := range addresses {
ips = append(ips, addr.Addr().String())
}
deviceIPs, err := json.Marshal(ips)
if err != nil {
return err
}
s := &kubeapi.Secret{
Data: map[string][]byte{
kubetypes.KeyDeviceFQDN: []byte(fqdn),
kubetypes.KeyDeviceIPs: deviceIPs,
},
}
return kc.StrategicMergePatchSecret(ctx, kc.stateSecret, s, "tailscale-container")
}
// storeHTTPSEndpoint writes an HTTPS endpoint exposed by this device via 'tailscale serve' to the client's state
// Secret. In practice this will be the same value that gets written to 'device_fqdn', but this should only be called
// when the serve config has been successfully set up.
func (kc *kubeClient) storeHTTPSEndpoint(ctx context.Context, ep string) error {
s := &kubeapi.Secret{
Data: map[string][]byte{
kubetypes.KeyHTTPSEndpoint: []byte(ep),
},
}
return kc.StrategicMergePatchSecret(ctx, kc.stateSecret, s, "tailscale-container")
}
// deleteAuthKey deletes the 'authkey' field of the given kube
// secret. No-op if there is no authkey in the secret.
func (kc *kubeClient) deleteAuthKey(ctx context.Context) error {
// m is a JSON Patch data structure, see https://jsonpatch.com/ or RFC 6902.
m := []kubeclient.JSONPatch{
{
Op: "remove",
Path: "/data/authkey",
},
}
if err := kc.JSONPatchResource(ctx, kc.stateSecret, kubeclient.TypeSecrets, m); err != nil {
if s, ok := err.(*kubeapi.Status); ok && s.Code == http.StatusUnprocessableEntity {
// This is kubernetes-ese for "the field you asked to
// delete already doesn't exist", aka no-op.
return nil
}
return err
}
return nil
}
// storeCapVerUID stores the current capability version of tailscale and, if provided, UID of the Pod in the tailscale
// state Secret.
// These two fields are used by the Kubernetes Operator to observe the current capability version of tailscaled running in this container.
func (kc *kubeClient) storeCapVerUID(ctx context.Context, podUID string) error {
capVerS := fmt.Sprintf("%d", tailcfg.CurrentCapabilityVersion)
d := map[string][]byte{
kubetypes.KeyCapVer: []byte(capVerS),
}
if podUID != "" {
d[kubetypes.KeyPodUID] = []byte(podUID)
}
s := &kubeapi.Secret{
Data: d,
}
return kc.StrategicMergePatchSecret(ctx, kc.stateSecret, s, "tailscale-container")
}
// waitForConsistentState waits for tailscaled to finish writing state if it
// looks like it's started. It is designed to reduce the likelihood that
// tailscaled gets shut down in the window between authenticating to control
// and finishing writing state. However, it's not bullet proof because we can't
// atomically authenticate and write state.
func (kc *kubeClient) waitForConsistentState(ctx context.Context) error {
var logged bool
bo := backoff.NewBackoff("", logger.Discard, 2*time.Second)
for {
select {
case <-ctx.Done():
return ctx.Err()
default:
}
secret, err := kc.GetSecret(ctx, kc.stateSecret)
if ctx.Err() != nil || kubeclient.IsNotFoundErr(err) {
return nil
}
if err != nil {
return fmt.Errorf("getting Secret %q: %v", kc.stateSecret, err)
}
if hasConsistentState(secret.Data) {
return nil
}
if !logged {
log.Printf("Waiting for tailscaled to finish writing state to Secret %q", kc.stateSecret)
logged = true
}
bo.BackOff(ctx, errors.New("")) // Fake error to trigger actual sleep.
}
}
// hasConsistentState returns true is there is either no state or the full set
// of expected keys are present.
func hasConsistentState(d map[string][]byte) bool {
var (
_, hasCurrent = d[string(ipn.CurrentProfileStateKey)]
_, hasKnown = d[string(ipn.KnownProfilesStateKey)]
_, hasMachine = d[string(ipn.MachineKeyStateKey)]
hasProfile bool
)
for k := range d {
if strings.HasPrefix(k, "profile-") {
if hasProfile {
return false // We only expect one profile.
}
hasProfile = true
}
}
// Approximate check, we don't want to reimplement all of profileManager.
return (hasCurrent && hasKnown && hasMachine && hasProfile) ||
(!hasCurrent && !hasKnown && !hasMachine && !hasProfile)
}