open-nomad/nomad/vault.go

1465 lines
44 KiB
Go

package nomad
import (
"context"
"errors"
"fmt"
"math/rand"
"strconv"
"strings"
"sync"
"sync/atomic"
"time"
"github.com/hashicorp/nomad/helper"
tomb "gopkg.in/tomb.v2"
metrics "github.com/armon/go-metrics"
log "github.com/hashicorp/go-hclog"
multierror "github.com/hashicorp/go-multierror"
"github.com/hashicorp/nomad/nomad/structs"
"github.com/hashicorp/nomad/nomad/structs/config"
vapi "github.com/hashicorp/vault/api"
"golang.org/x/sync/errgroup"
"golang.org/x/time/rate"
)
const (
// vaultTokenCreateTTL is the duration the wrapped token for the client is
// valid for. The units are in seconds.
vaultTokenCreateTTL = "60s"
// minimumTokenTTL is the minimum Token TTL allowed for child tokens.
minimumTokenTTL = 5 * time.Minute
// defaultTokenTTL is the default Token TTL used when the passed token is a
// root token such that child tokens aren't being created against a role
// that has defined a TTL
defaultTokenTTL = "72h"
// requestRateLimit is the maximum number of requests per second Nomad will
// make against Vault
requestRateLimit rate.Limit = 500.0
// maxParallelRevokes is the maximum number of parallel Vault
// token revocation requests
maxParallelRevokes = 64
// vaultRevocationIntv is the interval at which Vault tokens that failed
// initial revocation are retried
vaultRevocationIntv = 5 * time.Minute
// vaultCapabilitiesLookupPath is the path to lookup the capabilities of
// ones token.
vaultCapabilitiesLookupPath = "sys/capabilities-self"
// vaultTokenRenewPath is the path used to renew our token
vaultTokenRenewPath = "auth/token/renew-self"
// vaultTokenLookupPath is the path used to lookup a token
vaultTokenLookupPath = "auth/token/lookup"
// vaultTokenRevokePath is the path used to revoke a token
vaultTokenRevokePath = "auth/token/revoke-accessor"
// vaultRoleLookupPath is the path to lookup a role
vaultRoleLookupPath = "auth/token/roles/%s"
// vaultRoleCreatePath is the path to create a token from a role
vaultTokenRoleCreatePath = "auth/token/create/%s"
)
var (
// vaultCapabilitiesCapability is the expected capability of Nomad's Vault
// token on the the path. The token must have at least one of the
// capabilities.
vaultCapabilitiesCapability = []string{"update", "root"}
// vaultTokenRenewCapability is the expected capability Nomad's
// Vault token should have on the path. The token must have at least one of
// the capabilities.
vaultTokenRenewCapability = []string{"update", "root"}
// vaultTokenLookupCapability is the expected capability Nomad's
// Vault token should have on the path. The token must have at least one of
// the capabilities.
vaultTokenLookupCapability = []string{"update", "root"}
// vaultTokenRevokeCapability is the expected capability Nomad's
// Vault token should have on the path. The token must have at least one of
// the capabilities.
vaultTokenRevokeCapability = []string{"update", "root"}
// vaultRoleLookupCapability is the the expected capability Nomad's Vault
// token should have on the path. The token must have at least one of the
// capabilities.
vaultRoleLookupCapability = []string{"read", "root"}
// vaultTokenRoleCreateCapability is the the expected capability Nomad's Vault
// token should have on the path. The token must have at least one of the
// capabilities.
vaultTokenRoleCreateCapability = []string{"update", "root"}
)
// VaultClient is the Servers interface for interfacing with Vault
type VaultClient interface {
// SetActive activates or de-activates the Vault client. When active, token
// creation/lookup/revocation operation are allowed.
SetActive(active bool)
// SetConfig updates the config used by the Vault client
SetConfig(config *config.VaultConfig) error
// GetConfig returns a copy of the config used by the Vault client, for
// testing
GetConfig() *config.VaultConfig
// CreateToken takes an allocation and task and returns an appropriate Vault
// Secret
CreateToken(ctx context.Context, a *structs.Allocation, task string) (*vapi.Secret, error)
// LookupToken takes a token string and returns its capabilities.
LookupToken(ctx context.Context, token string) (*vapi.Secret, error)
// RevokeTokens takes a set of tokens accessor and revokes the tokens
RevokeTokens(ctx context.Context, accessors []*structs.VaultAccessor, committed bool) error
// MarkForRevocation revokes the tokens in background
MarkForRevocation(accessors []*structs.VaultAccessor) error
// Stop is used to stop token renewal
Stop()
// Running returns whether the Vault client is running
Running() bool
// Stats returns the Vault clients statistics
Stats() map[string]string
// EmitStats emits that clients statistics at the given period until stopCh
// is called.
EmitStats(period time.Duration, stopCh <-chan struct{})
}
// VaultStats returns all the stats about Vault tokens created and managed by
// Nomad.
type VaultStats struct {
// TrackedForRevoke is the count of tokens that are being tracked to be
// revoked since they could not be immediately revoked.
TrackedForRevoke int
// TokenTTL is the time-to-live duration for the current token
TokenTTL time.Duration
// TokenExpiry is the recorded expiry time of the current token
TokenExpiry time.Time
// LastRenewalTime is the time since the token was last renewed
LastRenewalTime time.Time
TimeFromLastRenewal time.Duration
// NextRenewalTime is the time the token will attempt to renew
NextRenewalTime time.Time
TimeToNextRenewal time.Duration
}
// PurgeVaultAccessorFn is called to remove VaultAccessors from the system. If
// the function returns an error, the token will still be tracked and revocation
// will retry till there is a success
type PurgeVaultAccessorFn func(accessors []*structs.VaultAccessor) error
// vaultClient is the Servers implementation of the VaultClient interface. The
// client renews the PeriodicToken given in the Vault configuration and provides
// the Server with the ability to create child tokens and lookup the permissions
// of tokens.
type vaultClient struct {
// limiter is used to rate limit requests to Vault
limiter *rate.Limiter
// client is the Vault API client used for Namespace-relative integrations
// with the Vault API (anything except `/v1/sys`). If this server is not
// configured to reference a Vault namespace, this will point to the same
// client as clientSys
client *vapi.Client
// clientSys is the Vault API client used for non-Namespace-relative integrations
// with the Vault API (anything involving `/v1/sys`). This client is never configured
// with a Vault namespace, because these endpoints may return errors if a namespace
// header is provided
clientSys *vapi.Client
// auth is the Vault token auth API client
auth *vapi.TokenAuth
// config is the user passed Vault config
config *config.VaultConfig
// connEstablished marks whether we have an established connection to Vault.
connEstablished bool
// connEstablishedErr marks an error that can occur when establishing a
// connection
connEstablishedErr error
// token is the raw token used by the client
token string
// tokenData is the data of the passed Vault token
tokenData *structs.VaultTokenData
// revoking tracks the VaultAccessors that must be revoked
revoking map[*structs.VaultAccessor]time.Time
purgeFn PurgeVaultAccessorFn
revLock sync.Mutex
// active indicates whether the vaultClient is active. It should be
// accessed using a helper and updated atomically
active int32
// running indicates whether the vault client is started.
running bool
// renewLoopActive indicates whether the renewal goroutine is running
// It should be accessed and updated atomically
// used for testing purposes only
renewLoopActive int32
// childTTL is the TTL for child tokens.
childTTL string
// currentExpiration is the time the current token lease expires
currentExpiration time.Time
currentExpirationLock sync.Mutex
lastRenewalTime time.Time
nextRenewalTime time.Time
renewalTimeLock sync.Mutex
tomb *tomb.Tomb
logger log.Logger
// l is used to lock the configuration aspects of the client such that
// multiple callers can't cause conflicting config updates
l sync.Mutex
// setConfigLock serializes access to the SetConfig method
setConfigLock sync.Mutex
// consts as struct fields for overriding in tests
maxRevokeBatchSize int
revocationIntv time.Duration
entHandler taskClientHandler
}
type taskClientHandler interface {
clientForTask(v *vaultClient, namespace string) (*vapi.Client, error)
}
// NewVaultClient returns a Vault client from the given config. If the client
// couldn't be made an error is returned.
func NewVaultClient(c *config.VaultConfig, logger log.Logger, purgeFn PurgeVaultAccessorFn, delegate taskClientHandler) (*vaultClient, error) {
if c == nil {
return nil, fmt.Errorf("must pass valid VaultConfig")
}
if logger == nil {
return nil, fmt.Errorf("must pass valid logger")
}
if purgeFn == nil {
purgeFn = func(accessors []*structs.VaultAccessor) error { return nil }
}
if delegate == nil {
delegate = &VaultNoopDelegate{}
}
v := &vaultClient{
config: c,
logger: logger.Named("vault"),
limiter: rate.NewLimiter(requestRateLimit, int(requestRateLimit)),
revoking: make(map[*structs.VaultAccessor]time.Time),
purgeFn: purgeFn,
tomb: &tomb.Tomb{},
maxRevokeBatchSize: maxVaultRevokeBatchSize,
revocationIntv: vaultRevocationIntv,
entHandler: delegate,
}
if v.config.IsEnabled() {
if err := v.buildClient(); err != nil {
return nil, err
}
// Launch the required goroutines
v.tomb.Go(wrapNilError(v.establishConnection))
v.tomb.Go(wrapNilError(v.revokeDaemon))
v.running = true
}
return v, nil
}
func (v *vaultClient) Stop() {
v.l.Lock()
running := v.running
v.running = false
v.l.Unlock()
if running {
v.tomb.Kill(nil)
v.tomb.Wait()
v.flush()
}
}
func (v *vaultClient) Running() bool {
v.l.Lock()
defer v.l.Unlock()
return v.running
}
// SetActive activates or de-activates the Vault client. When active, token
// creation/lookup/revocation operation are allowed. All queued revocations are
// cancelled if set un-active as it is assumed another instances is taking over
func (v *vaultClient) SetActive(active bool) {
if active {
atomic.StoreInt32(&v.active, 1)
} else {
atomic.StoreInt32(&v.active, 0)
}
// Clear out the revoking tokens
v.revLock.Lock()
v.revoking = make(map[*structs.VaultAccessor]time.Time)
v.revLock.Unlock()
}
// flush is used to reset the state of the vault client
func (v *vaultClient) flush() {
v.l.Lock()
defer v.l.Unlock()
v.revLock.Lock()
defer v.revLock.Unlock()
v.client = nil
v.clientSys = nil
v.auth = nil
v.connEstablished = false
v.connEstablishedErr = nil
v.token = ""
v.tokenData = nil
v.revoking = make(map[*structs.VaultAccessor]time.Time)
v.childTTL = ""
v.tomb = &tomb.Tomb{}
}
// GetConfig returns a copy of this vault client's configuration, for testing.
func (v *vaultClient) GetConfig() *config.VaultConfig {
v.setConfigLock.Lock()
defer v.setConfigLock.Unlock()
return v.config.Copy()
}
// SetConfig is used to update the Vault config being used. A temporary outage
// may occur after calling as it re-establishes a connection to Vault
func (v *vaultClient) SetConfig(config *config.VaultConfig) error {
if config == nil {
return fmt.Errorf("must pass valid VaultConfig")
}
v.setConfigLock.Lock()
defer v.setConfigLock.Unlock()
v.l.Lock()
defer v.l.Unlock()
// If reloading the same config, no-op
if v.config.Equals(config) {
return nil
}
// Kill any background routines
if v.running {
// Kill any background routine
v.tomb.Kill(nil)
// Locking around tomb.Wait can deadlock with
// establishConnection exiting, so we must unlock here.
v.l.Unlock()
v.tomb.Wait()
v.l.Lock()
// Stop accepting any new requests
v.connEstablished = false
v.tomb = &tomb.Tomb{}
v.running = false
}
// Store the new config
v.config = config
// Check if we should relaunch
if v.config.IsEnabled() {
// Rebuild the client
if err := v.buildClient(); err != nil {
return err
}
// Launch the required goroutines
v.tomb.Go(wrapNilError(v.establishConnection))
v.tomb.Go(wrapNilError(v.revokeDaemon))
v.running = true
}
return nil
}
// buildClient is used to build a Vault client based on the stored Vault config
func (v *vaultClient) buildClient() error {
// Validate we have the required fields.
if v.config.Token == "" {
return errors.New("Vault token must be set")
} else if v.config.Addr == "" {
return errors.New("Vault address must be set")
}
// Parse the TTL if it is set
if v.config.TaskTokenTTL != "" {
d, err := time.ParseDuration(v.config.TaskTokenTTL)
if err != nil {
return fmt.Errorf("failed to parse TaskTokenTTL %q: %v", v.config.TaskTokenTTL, err)
}
if d.Nanoseconds() < minimumTokenTTL.Nanoseconds() {
return fmt.Errorf("ChildTokenTTL is less than minimum allowed of %v", minimumTokenTTL)
}
v.childTTL = v.config.TaskTokenTTL
} else {
// Default the TaskTokenTTL
v.childTTL = defaultTokenTTL
}
// Get the Vault API configuration
apiConf, err := v.config.ApiConfig()
if err != nil {
return fmt.Errorf("Failed to create Vault API config: %v", err)
}
// Create the Vault API client
client, err := vapi.NewClient(apiConf)
if err != nil {
v.logger.Error("failed to create Vault client and not retrying", "error", err)
return err
}
// Store the client, create/assign the /sys client
v.client = client
if v.config.Namespace != "" {
v.logger.Debug("configuring Vault namespace", "namespace", v.config.Namespace)
v.clientSys, err = vapi.NewClient(apiConf)
if err != nil {
v.logger.Error("failed to create Vault sys client and not retrying", "error", err)
return err
}
client.SetNamespace(v.config.Namespace)
} else {
v.clientSys = client
}
// Set the token
v.token = v.config.Token
client.SetToken(v.token)
v.auth = client.Auth().Token()
return nil
}
// establishConnection is used to make first contact with Vault. This should be
// called in a go-routine since the connection is retried until the Vault Client
// is stopped or the connection is successfully made at which point the renew
// loop is started.
func (v *vaultClient) establishConnection() {
// Create the retry timer and set initial duration to zero so it fires
// immediately
retryTimer := time.NewTimer(0)
initStatus := false
OUTER:
for {
select {
case <-v.tomb.Dying():
return
case <-retryTimer.C:
// Retry validating the token till success
if err := v.parseSelfToken(); err != nil {
// if parsing token fails, try to distinguish legitimate token error from transient Vault initialization/connection issue
if !initStatus {
if _, err := v.clientSys.Sys().Health(); err != nil {
v.logger.Warn("failed to contact Vault API", "retry", v.config.ConnectionRetryIntv, "error", err)
retryTimer.Reset(v.config.ConnectionRetryIntv)
continue OUTER
}
initStatus = true
}
v.logger.Error("failed to validate self token/role", "retry", v.config.ConnectionRetryIntv, "error", err)
retryTimer.Reset(v.config.ConnectionRetryIntv)
v.l.Lock()
v.connEstablished = true
v.connEstablishedErr = fmt.Errorf("failed to establish connection to Vault: %v", err)
v.l.Unlock()
continue OUTER
}
break OUTER
}
}
// Set the wrapping function such that token creation is wrapped now
// that we know our role
v.client.SetWrappingLookupFunc(v.getWrappingFn())
// If we are given a non-root token, start renewing it
if v.tokenData.Root() && v.tokenData.CreationTTL == 0 {
v.logger.Debug("not renewing token as it is root")
} else {
v.logger.Debug("starting renewal loop", "creation_ttl", time.Duration(v.tokenData.CreationTTL)*time.Second)
v.tomb.Go(wrapNilError(v.renewalLoop))
}
v.l.Lock()
v.connEstablished = true
v.connEstablishedErr = nil
v.l.Unlock()
}
func (v *vaultClient) isRenewLoopActive() bool {
return atomic.LoadInt32(&v.renewLoopActive) == 1
}
// renewalLoop runs the renew loop. This should only be called if we are given a
// non-root token.
func (v *vaultClient) renewalLoop() {
atomic.StoreInt32(&v.renewLoopActive, 1)
defer atomic.StoreInt32(&v.renewLoopActive, 0)
// Create the renewal timer and set initial duration to zero so it fires
// immediately
authRenewTimer := time.NewTimer(0)
// Backoff is to reduce the rate we try to renew with Vault under error
// situations
backoff := 0.0
for {
select {
case <-v.tomb.Dying():
return
case <-authRenewTimer.C:
// Renew the token and determine the new expiration
recoverable, err := v.renew()
v.currentExpirationLock.Lock()
currentExpiration := v.currentExpiration
v.currentExpirationLock.Unlock()
// Successfully renewed
if err == nil {
// Attempt to renew the token at half the expiration time
durationUntilRenew := time.Until(currentExpiration) / 2
v.renewalTimeLock.Lock()
now := time.Now()
v.lastRenewalTime = now
v.nextRenewalTime = now.Add(durationUntilRenew)
v.renewalTimeLock.Unlock()
v.logger.Info("successfully renewed token", "next_renewal", durationUntilRenew)
authRenewTimer.Reset(durationUntilRenew)
// Reset any backoff
backoff = 0
break
}
metrics.IncrCounter([]string{"nomad", "vault", "renew_failed"}, 1)
v.logger.Warn("got error or bad auth, so backing off", "error", err, "recoverable", recoverable)
if !recoverable {
return
}
backoff = nextBackoff(backoff, currentExpiration)
if backoff < 0 {
// We have failed to renew the token past its expiration. Stop
// renewing with Vault.
v.logger.Error("failed to renew Vault token before lease expiration. Shutting down Vault client",
"error", err)
v.l.Lock()
v.connEstablished = false
v.connEstablishedErr = err
v.l.Unlock()
return
}
durationUntilRetry := time.Duration(backoff) * time.Second
v.renewalTimeLock.Lock()
v.nextRenewalTime = time.Now().Add(durationUntilRetry)
v.renewalTimeLock.Unlock()
v.logger.Info("backing off renewal", "retry", durationUntilRetry)
authRenewTimer.Reset(durationUntilRetry)
}
}
}
// nextBackoff returns the delay for the next auto renew interval, in seconds.
// Returns negative value if past expiration
//
// It should increase the amount of backoff each time, with the following rules:
//
// - If token expired already despite earlier renewal attempts,
// back off for 1 minute + jitter
// - If we have an existing authentication that is going to expire,
//
// never back off more than half of the amount of time remaining
// until expiration (with 5s floor)
// * Never back off more than 30 seconds multiplied by a random
// value between 1 and 2
// * Use randomness so that many clients won't keep hitting Vault
// at the same time
func nextBackoff(backoff float64, expiry time.Time) float64 {
maxBackoff := time.Until(expiry) / 2
if maxBackoff < 0 {
// expiry passed
return 60 * (1.0 + rand.Float64())
}
switch {
case backoff >= 24:
backoff = 30
default:
backoff = backoff * 1.25
}
// Add randomness
backoff = backoff * (1.0 + rand.Float64())
if backoff > maxBackoff.Seconds() {
backoff = maxBackoff.Seconds()
}
if backoff < 5 {
backoff = 5
}
return backoff
}
// renew attempts to renew our Vault token. If the renewal fails, an error is
// returned. The boolean indicates whether it's safe to attempt to renew again.
// This method updates the currentExpiration time
func (v *vaultClient) renew() (bool, error) {
// Track how long the request takes
defer metrics.MeasureSince([]string{"nomad", "vault", "renew"}, time.Now())
// Attempt to renew the token
secret, err := v.auth.RenewSelf(v.tokenData.CreationTTL)
if err != nil {
// Check if there is a permission denied
recoverable := !structs.VaultUnrecoverableError.MatchString(err.Error())
return recoverable, fmt.Errorf("failed to renew the vault token: %v", err)
}
if secret == nil {
// It's possible for RenewSelf to return (nil, nil) if the
// response body from Vault is empty.
return true, fmt.Errorf("renewal failed: empty response from vault")
}
// these treated as transient errors, where can keep renewing
auth := secret.Auth
if auth == nil {
return true, fmt.Errorf("renewal successful but not auth information returned")
} else if auth.LeaseDuration == 0 {
return true, fmt.Errorf("renewal successful but no lease duration returned")
}
v.extendExpiration(auth.LeaseDuration)
v.logger.Debug("successfully renewed server token")
return true, nil
}
// getWrappingFn returns an appropriate wrapping function for Nomad Servers
func (v *vaultClient) getWrappingFn() func(operation, path string) string {
createPath := "auth/token/create"
role := v.getRole()
if role != "" {
createPath = fmt.Sprintf("auth/token/create/%s", role)
}
return func(operation, path string) string {
// Only wrap the token create operation
if operation != "POST" || path != createPath {
return ""
}
return vaultTokenCreateTTL
}
}
// parseSelfToken looks up the Vault token in Vault and parses its data storing
// it in the client. If the token is not valid for Nomads purposes an error is
// returned.
func (v *vaultClient) parseSelfToken() error {
// Try looking up the token using the self endpoint
secret, err := v.lookupSelf()
if err != nil {
return err
}
// Read and parse the fields
var data structs.VaultTokenData
if err := structs.DecodeVaultSecretData(secret, &data); err != nil {
return fmt.Errorf("failed to parse Vault token's data block: %v", err)
}
v.tokenData = &data
v.extendExpiration(data.TTL)
// The criteria that must be met for the token to be valid are as follows:
// 1) If token is non-root or is but has a creation ttl
// a) The token must be renewable
// b) Token must have a non-zero TTL
// 2) Must have update capability for "auth/token/lookup/" (used to verify incoming tokens)
// 3) Must have update capability for "/auth/token/revoke-accessor/" (used to revoke unneeded tokens)
// 4) If configured to create tokens against a role:
// a) Must have read capability for "auth/token/roles/<role_name" (Can just attempt a read)
// b) Must have update capability for path "auth/token/create/<role_name>"
// c) Role must:
// 1) Must allow tokens to be renewed
// 2) Must not have an explicit max TTL
// 3) Must have non-zero period
// 5) If not configured against a role, the token must be root
var mErr multierror.Error
role := v.getRole()
if !data.Root() {
// All non-root tokens must be renewable
if !data.Renewable {
_ = multierror.Append(&mErr, fmt.Errorf("Vault token is not renewable or root"))
}
// All non-root tokens must have a lease duration
if data.CreationTTL == 0 {
_ = multierror.Append(&mErr, fmt.Errorf("invalid lease duration of zero"))
}
// The lease duration can not be expired
if data.TTL == 0 {
_ = multierror.Append(&mErr, fmt.Errorf("token TTL is zero"))
}
// There must be a valid role since we aren't root
if role == "" {
_ = multierror.Append(&mErr, fmt.Errorf("token role name must be set when not using a root token"))
}
} else if data.CreationTTL != 0 {
// If the root token has a TTL it must be renewable
if !data.Renewable {
_ = multierror.Append(&mErr, fmt.Errorf("Vault token has a TTL but is not renewable"))
} else if data.TTL == 0 {
// If the token has a TTL make sure it has not expired
_ = multierror.Append(&mErr, fmt.Errorf("token TTL is zero"))
}
}
// Check we have the correct capabilities
if err := v.validateCapabilities(role, data.Root()); err != nil {
_ = multierror.Append(&mErr, err)
}
// If given a role validate it
if role != "" {
if err := v.validateRole(role); err != nil {
_ = multierror.Append(&mErr, err)
}
}
return mErr.ErrorOrNil()
}
// lookupSelf is a helper function that looks up latest self lease info.
func (v *vaultClient) lookupSelf() (*vapi.Secret, error) {
// Get the initial lease duration
auth := v.client.Auth().Token()
secret, err := auth.LookupSelf()
if err == nil && secret != nil && secret.Data != nil {
return secret, nil
}
// Try looking up our token directly, even when we get an empty response,
// in case of an unexpected event - a true failure would occur in this lookup again
secret, err = auth.Lookup(v.client.Token())
switch {
case err != nil:
return nil, fmt.Errorf("failed to lookup Vault periodic token: %v", err)
case secret == nil || secret.Data == nil:
return nil, fmt.Errorf("failed to lookup Vault periodic token: got empty response")
default:
return secret, nil
}
}
// getRole returns the role name to be used when creating tokens
func (v *vaultClient) getRole() string {
if v.config.Role != "" {
return v.config.Role
}
return v.tokenData.Role
}
// validateCapabilities checks that Nomad's Vault token has the correct
// capabilities.
func (v *vaultClient) validateCapabilities(role string, root bool) error {
// Check if the token can lookup capabilities.
var mErr multierror.Error
_, _, err := v.hasCapability(vaultCapabilitiesLookupPath, vaultCapabilitiesCapability)
if err != nil {
// Check if there is a permission denied
if structs.VaultUnrecoverableError.MatchString(err.Error()) {
// Since we can't read permissions, we just log a warning that we
// can't tell if the Vault token will work
msg := fmt.Sprintf("can not lookup token capabilities. "+
"As such certain operations may fail in the future. "+
"Please give Nomad a Vault token with one of the following "+
"capabilities %q on %q so that the required capabilities can be verified",
vaultCapabilitiesCapability, vaultCapabilitiesLookupPath)
v.logger.Warn(msg)
return nil
} else {
_ = multierror.Append(&mErr, err)
}
}
// verify is a helper function that verifies the token has one of the
// capabilities on the given path and adds an issue to the error
verify := func(path string, requiredCaps []string) {
ok, caps, err := v.hasCapability(path, requiredCaps)
if err != nil {
_ = multierror.Append(&mErr, err)
} else if !ok {
_ = multierror.Append(&mErr,
fmt.Errorf("token must have one of the following capabilities %q on %q; has %v", requiredCaps, path, caps))
}
}
// Check if we are verifying incoming tokens
if !v.config.AllowsUnauthenticated() {
verify(vaultTokenLookupPath, vaultTokenLookupCapability)
}
// Verify we can renew our selves tokens
verify(vaultTokenRenewPath, vaultTokenRenewCapability)
// Verify we can revoke tokens
verify(vaultTokenRevokePath, vaultTokenRevokeCapability)
// If we are using a role verify the capability
if role != "" {
// Verify we can read the role
verify(fmt.Sprintf(vaultRoleLookupPath, role), vaultRoleLookupCapability)
// Verify we can create from the role
verify(fmt.Sprintf(vaultTokenRoleCreatePath, role), vaultTokenRoleCreateCapability)
}
return mErr.ErrorOrNil()
}
// hasCapability takes a path and returns whether the token has at least one of
// the required capabilities on the given path. It also returns the set of
// capabilities the token does have as well as any error that occurred.
func (v *vaultClient) hasCapability(path string, required []string) (bool, []string, error) {
caps, err := v.client.Sys().CapabilitiesSelf(path)
if err != nil {
return false, nil, err
}
for _, c := range caps {
for _, r := range required {
if c == r {
return true, caps, nil
}
}
}
return false, caps, nil
}
// validateRole contacts Vault and checks that the given Vault role is valid for
// the purposes of being used by Nomad
func (v *vaultClient) validateRole(role string) error {
if role == "" {
return fmt.Errorf("Invalid empty role name")
}
// Validate the role
rsecret, err := v.client.Logical().Read(fmt.Sprintf("auth/token/roles/%s", role))
if err != nil {
return fmt.Errorf("failed to lookup role %q: %v", role, err)
}
if rsecret == nil {
return fmt.Errorf("Role %q does not exist", role)
}
// Read and parse the fields
var data structs.VaultTokenRoleData
if err := structs.DecodeVaultSecretData(rsecret, &data); err != nil {
return fmt.Errorf("failed to parse Vault role's data block: %v", err)
}
// Validate the role is acceptable
var mErr multierror.Error
if !data.Renewable {
_ = multierror.Append(&mErr, fmt.Errorf("Role must allow tokens to be renewed"))
}
if data.ExplicitMaxTtl != 0 || data.TokenExplicitMaxTtl != 0 {
_ = multierror.Append(&mErr, fmt.Errorf("Role can not use an explicit max ttl. Token must be periodic."))
}
if data.Period == 0 && data.TokenPeriod == 0 {
_ = multierror.Append(&mErr, fmt.Errorf("Role must have a non-zero period to make tokens periodic."))
}
return mErr.ErrorOrNil()
}
// ConnectionEstablished returns whether a connection to Vault has been
// established and any error that potentially caused it to be false
func (v *vaultClient) ConnectionEstablished() (bool, error) {
v.l.Lock()
defer v.l.Unlock()
return v.connEstablished, v.connEstablishedErr
}
// Enabled returns whether the client is active
func (v *vaultClient) Enabled() bool {
v.l.Lock()
defer v.l.Unlock()
return v.config.IsEnabled()
}
// Active returns whether the client is active
func (v *vaultClient) Active() bool {
return atomic.LoadInt32(&v.active) == 1
}
// CreateToken takes the allocation and task and returns an appropriate Vault
// token. The call is rate limited and may be canceled with the passed policy.
// When the error is recoverable, it will be of type RecoverableError
func (v *vaultClient) CreateToken(ctx context.Context, a *structs.Allocation, task string) (*vapi.Secret, error) {
if !v.Enabled() {
return nil, fmt.Errorf("Vault integration disabled")
}
if !v.Active() {
return nil, structs.NewRecoverableError(fmt.Errorf("Vault client not active"), true)
}
// Check if we have established a connection with Vault
if established, err := v.ConnectionEstablished(); !established && err == nil {
return nil, structs.NewRecoverableError(fmt.Errorf("Connection to Vault has not been established"), true)
} else if err != nil {
return nil, err
}
// Track how long the request takes
defer metrics.MeasureSince([]string{"nomad", "vault", "create_token"}, time.Now())
// Retrieve the Vault block for the task
vaultBlocks := a.Job.Vault()
if vaultBlocks == nil {
return nil, fmt.Errorf("Job does not require Vault token")
}
tg, ok := vaultBlocks[a.TaskGroup]
if !ok {
return nil, fmt.Errorf("Task group does not require Vault token")
}
taskVault, ok := tg[task]
if !ok {
return nil, fmt.Errorf("Task does not require Vault token")
}
// Set namespace for task
namespaceForTask := v.config.Namespace
if taskVault.Namespace != "" {
namespaceForTask = taskVault.Namespace
}
// Build the creation request
req := &vapi.TokenCreateRequest{
Policies: taskVault.Policies,
Metadata: map[string]string{
"AllocationID": a.ID,
"JobID": a.JobID,
"TaskGroup": a.TaskGroup,
"Task": task,
"NodeID": a.NodeID,
"Namespace": namespaceForTask,
},
TTL: v.childTTL,
DisplayName: fmt.Sprintf("%s-%s", a.ID, task),
}
// Ensure we are under our rate limit
if err := v.limiter.Wait(ctx); err != nil {
return nil, err
}
// Make the request and switch depending on whether we are using a root
// token or a role based token
var secret *vapi.Secret
var err error
role := v.getRole()
// Fetch client for task
taskClient, err := v.entHandler.clientForTask(v, namespaceForTask)
if err != nil {
return nil, err
}
if v.tokenData.Root() && role == "" {
req.Period = v.childTTL
secret, err = taskClient.Auth().Token().Create(req)
} else {
// Make the token using the role
secret, err = taskClient.Auth().Token().CreateWithRole(req, role)
}
// Determine whether it is unrecoverable
if err != nil {
err = fmt.Errorf("failed to create an alloc vault token: %v", err)
if structs.VaultUnrecoverableError.MatchString(err.Error()) {
return secret, err
}
// The error is recoverable
return nil, structs.NewRecoverableError(err, true)
}
// Validate the response
var validationErr error
if secret == nil {
validationErr = fmt.Errorf("Vault returned nil Secret")
} else if secret.WrapInfo == nil {
validationErr = fmt.Errorf("Vault returned Secret with nil WrapInfo. Secret warnings: %v", secret.Warnings)
} else if secret.WrapInfo.WrappedAccessor == "" {
validationErr = fmt.Errorf("Vault returned WrapInfo without WrappedAccessor. Secret warnings: %v", secret.Warnings)
}
if validationErr != nil {
v.logger.Warn("failed to CreateToken", "error", validationErr)
return nil, structs.NewRecoverableError(validationErr, true)
}
// Got a valid response
return secret, nil
}
// LookupToken takes a Vault token and does a lookup against Vault. The call is
// rate limited and may be canceled with passed context.
func (v *vaultClient) LookupToken(ctx context.Context, token string) (*vapi.Secret, error) {
if !v.Enabled() {
return nil, fmt.Errorf("Vault integration disabled")
}
if !v.Active() {
return nil, fmt.Errorf("Vault client not active")
}
// Check if we have established a connection with Vault
if established, err := v.ConnectionEstablished(); !established && err == nil {
return nil, structs.NewRecoverableError(fmt.Errorf("Connection to Vault has not been established"), true)
} else if err != nil {
return nil, err
}
// Track how long the request takes
defer metrics.MeasureSince([]string{"nomad", "vault", "lookup_token"}, time.Now())
// Ensure we are under our rate limit
if err := v.limiter.Wait(ctx); err != nil {
return nil, err
}
// Lookup the token
return v.auth.Lookup(token)
}
// RevokeTokens revokes the passed set of accessors. If committed is set, the
// purge function passed to the client is called. If there is an error purging
// either because of Vault failures or because of the purge function, the
// revocation is retried until the tokens TTL.
func (v *vaultClient) RevokeTokens(ctx context.Context, accessors []*structs.VaultAccessor, committed bool) error {
if !v.Enabled() {
return nil
}
if !v.Active() {
return fmt.Errorf("Vault client not active")
}
// Track how long the request takes
defer metrics.MeasureSince([]string{"nomad", "vault", "revoke_tokens"}, time.Now())
// Check if we have established a connection with Vault. If not just add it
// to the queue
if established, err := v.ConnectionEstablished(); !established && err == nil {
// Only bother tracking it for later revocation if the accessor was
// committed
if committed {
v.storeForRevocation(accessors)
}
// Track that we are abandoning these accessors.
metrics.IncrCounter([]string{"nomad", "vault", "undistributed_tokens_abandoned"}, float32(len(accessors)))
return nil
}
// Attempt to revoke immediately and if it fails, add it to the revoke queue
err := v.parallelRevoke(ctx, accessors)
if err != nil {
// If it is uncommitted, it is a best effort revoke as it will shortly
// TTL within the cubbyhole and has not been leaked to any outside
// system
if !committed {
metrics.IncrCounter([]string{"nomad", "vault", "undistributed_tokens_abandoned"}, float32(len(accessors)))
return nil
}
v.logger.Warn("failed to revoke tokens. Will reattempt until TTL", "error", err)
v.storeForRevocation(accessors)
return nil
} else if !committed {
// Mark that it was revoked but there is nothing to purge so exit
metrics.IncrCounter([]string{"nomad", "vault", "undistributed_tokens_revoked"}, float32(len(accessors)))
return nil
}
if err := v.purgeFn(accessors); err != nil {
v.logger.Error("failed to purge Vault accessors", "error", err)
v.storeForRevocation(accessors)
return nil
}
// Track that it was revoked successfully
metrics.IncrCounter([]string{"nomad", "vault", "distributed_tokens_revoked"}, float32(len(accessors)))
return nil
}
func (v *vaultClient) MarkForRevocation(accessors []*structs.VaultAccessor) error {
if !v.Enabled() {
return nil
}
if !v.Active() {
return fmt.Errorf("Vault client not active")
}
v.storeForRevocation(accessors)
return nil
}
// storeForRevocation stores the passed set of accessors for revocation. It
// captures their effective TTL by storing their create TTL plus the current
// time.
func (v *vaultClient) storeForRevocation(accessors []*structs.VaultAccessor) {
v.revLock.Lock()
now := time.Now()
for _, a := range accessors {
if _, ok := v.revoking[a]; !ok {
v.revoking[a] = now.Add(time.Duration(a.CreationTTL) * time.Second)
}
}
v.revLock.Unlock()
}
// parallelRevoke revokes the passed VaultAccessors in parallel.
func (v *vaultClient) parallelRevoke(ctx context.Context, accessors []*structs.VaultAccessor) error {
if !v.Enabled() {
return fmt.Errorf("Vault integration disabled")
}
if !v.Active() {
return fmt.Errorf("Vault client not active")
}
// Check if we have established a connection with Vault
if established, err := v.ConnectionEstablished(); !established && err == nil {
return structs.NewRecoverableError(fmt.Errorf("Connection to Vault has not been established"), true)
} else if err != nil {
return err
}
g, pCtx := errgroup.WithContext(ctx)
// Cap the handlers
handlers := len(accessors)
if handlers > maxParallelRevokes {
handlers = maxParallelRevokes
}
// Revoke the Vault Token Accessors
input := make(chan *structs.VaultAccessor, handlers)
for i := 0; i < handlers; i++ {
g.Go(func() error {
for {
select {
case va, ok := <-input:
if !ok {
return nil
}
err := v.auth.RevokeAccessor(va.Accessor)
if err != nil && !strings.Contains(err.Error(), "invalid accessor") {
return fmt.Errorf("failed to revoke token (alloc: %q, node: %q, task: %q): %v", va.AllocID, va.NodeID, va.Task, err)
}
case <-pCtx.Done():
return nil
}
}
})
}
// Send the input
go func() {
defer close(input)
for _, va := range accessors {
select {
case <-pCtx.Done():
return
case input <- va:
}
}
}()
// Wait for everything to complete
return g.Wait()
}
// maxVaultRevokeBatchSize is the maximum tokens a revokeDaemon should revoke
// and purge at any given time.
//
// Limiting the revocation batch size is beneficial for few reasons:
// - A single revocation failure of any entry in batch result into retrying the whole batch;
// the larger the batch is the higher likelihood of such failure
// - Smaller batch sizes result into more co-operativeness: provides hooks for
// reconsidering token TTL and leadership steps down.
// - Batches limit the size of the Raft message purging tokens. Due to bugs
// pre-0.11.3, expired tokens were not properly purged, so users upgrading from
// older versions may have huge numbers (millions) of expired tokens to purge.
const maxVaultRevokeBatchSize = 1000
// revokeDaemon should be called in a goroutine and is used to periodically
// revoke Vault accessors that failed the original revocation
func (v *vaultClient) revokeDaemon() {
ticker := time.NewTicker(v.revocationIntv)
defer ticker.Stop()
for {
select {
case <-v.tomb.Dying():
return
case now := <-ticker.C:
if established, err := v.ConnectionEstablished(); !established || err != nil {
continue
}
v.revLock.Lock()
// Fast path
if len(v.revoking) == 0 {
v.revLock.Unlock()
continue
}
// Build the list of accessors that need to be revoked while pruning any TTL'd checks
toRevoke := len(v.revoking)
if toRevoke > v.maxRevokeBatchSize {
v.logger.Info("batching tokens to be revoked",
"to_revoke", toRevoke, "batch_size", v.maxRevokeBatchSize,
"batch_interval", v.revocationIntv)
toRevoke = v.maxRevokeBatchSize
}
revoking := make([]*structs.VaultAccessor, 0, toRevoke)
ttlExpired := []*structs.VaultAccessor{}
for va, ttl := range v.revoking {
if now.After(ttl) {
ttlExpired = append(ttlExpired, va)
} else {
revoking = append(revoking, va)
}
// Batches should consider tokens to be revoked
// as well as expired tokens to ensure the Raft
// message is reasonably sized.
if len(revoking)+len(ttlExpired) >= toRevoke {
break
}
}
if err := v.parallelRevoke(context.Background(), revoking); err != nil {
v.logger.Warn("background token revocation errored", "error", err)
v.revLock.Unlock()
continue
}
// Unlock before a potentially expensive operation
v.revLock.Unlock()
// purge all explicitly revoked as well as ttl expired tokens
// and only remove them locally on purge success
revoking = append(revoking, ttlExpired...)
// Call the passed in token revocation function
if err := v.purgeFn(revoking); err != nil {
// Can continue since revocation is idempotent
v.logger.Error("token revocation errored", "error", err)
continue
}
// Track that tokens were revoked successfully
metrics.IncrCounter([]string{"nomad", "vault", "distributed_tokens_revoked"}, float32(len(revoking)))
// Can delete from the tracked list now that we have purged
v.revLock.Lock()
for _, va := range revoking {
delete(v.revoking, va)
}
v.revLock.Unlock()
}
}
}
// purgeVaultAccessors creates a Raft transaction to remove the passed Vault
// Accessors
func (s *Server) purgeVaultAccessors(accessors []*structs.VaultAccessor) error {
// Commit this update via Raft
req := structs.VaultAccessorsRequest{Accessors: accessors}
_, _, err := s.raftApply(structs.VaultAccessorDeregisterRequestType, req)
return err
}
// wrapNilError is a helper that returns a wrapped function that returns a nil
// error
func wrapNilError(f func()) func() error {
return func() error {
f()
return nil
}
}
// setLimit is used to update the rate limit
func (v *vaultClient) setLimit(l rate.Limit) {
v.l.Lock()
defer v.l.Unlock()
v.limiter = rate.NewLimiter(l, int(l))
}
func (v *vaultClient) Stats() map[string]string {
stat := v.stats()
expireTimeStr := ""
if !stat.TokenExpiry.IsZero() {
expireTimeStr = stat.TokenExpiry.Format(time.RFC3339)
}
lastRenewTimeStr := ""
if !stat.LastRenewalTime.IsZero() {
lastRenewTimeStr = stat.LastRenewalTime.Format(time.RFC3339)
}
nextRenewTimeStr := ""
if !stat.NextRenewalTime.IsZero() {
nextRenewTimeStr = stat.NextRenewalTime.Format(time.RFC3339)
}
return map[string]string{
"tracked_for_revoked": strconv.Itoa(stat.TrackedForRevoke),
"token_ttl": stat.TokenTTL.Round(time.Second).String(),
"token_expire_time": expireTimeStr,
"token_last_renewal_time": lastRenewTimeStr,
"token_next_renewal_time": nextRenewTimeStr,
}
}
func (v *vaultClient) stats() *VaultStats {
// Allocate a new stats struct
stats := new(VaultStats)
v.revLock.Lock()
stats.TrackedForRevoke = len(v.revoking)
v.revLock.Unlock()
v.currentExpirationLock.Lock()
stats.TokenExpiry = v.currentExpiration
v.currentExpirationLock.Unlock()
v.renewalTimeLock.Lock()
stats.NextRenewalTime = v.nextRenewalTime
stats.LastRenewalTime = v.lastRenewalTime
v.renewalTimeLock.Unlock()
if !stats.TokenExpiry.IsZero() {
stats.TokenTTL = time.Until(stats.TokenExpiry)
}
if !stats.LastRenewalTime.IsZero() {
stats.TimeFromLastRenewal = time.Since(stats.LastRenewalTime)
}
if !stats.NextRenewalTime.IsZero() {
stats.TimeToNextRenewal = time.Until(stats.NextRenewalTime)
}
return stats
}
// EmitStats is used to export metrics about the blocked eval tracker while enabled
func (v *vaultClient) EmitStats(period time.Duration, stopCh <-chan struct{}) {
timer, stop := helper.NewSafeTimer(period)
defer stop()
for {
timer.Reset(period)
select {
case <-timer.C:
stats := v.stats()
metrics.SetGauge([]string{"nomad", "vault", "distributed_tokens_revoking"}, float32(stats.TrackedForRevoke))
metrics.SetGauge([]string{"nomad", "vault", "token_ttl"}, float32(stats.TokenTTL/time.Millisecond))
metrics.SetGauge([]string{"nomad", "vault", "token_last_renewal"}, float32(stats.TimeFromLastRenewal/time.Millisecond))
metrics.SetGauge([]string{"nomad", "vault", "token_next_renewal"}, float32(stats.TimeToNextRenewal/time.Millisecond))
case <-stopCh:
return
}
}
}
// extendExpiration sets the current auth token expiration record to ttLSeconds seconds from now
func (v *vaultClient) extendExpiration(ttlSeconds int) {
v.currentExpirationLock.Lock()
v.currentExpiration = time.Now().Add(time.Duration(ttlSeconds) * time.Second)
v.currentExpirationLock.Unlock()
}
// VaultVaultNoopDelegate returns the default vault api auth token handler
type VaultNoopDelegate struct{}
func (e *VaultNoopDelegate) clientForTask(v *vaultClient, namespace string) (*vapi.Client, error) {
return v.client, nil
}