open-nomad/nomad/consul.go
Seth Hoenig 87be8c4c4b consul: correctly check consul acl token namespace when using consul oss
This PR fixes the Nomad Object Namespace <-> Consul ACL Token relationship
check when using Consul OSS (or Consul ENT without namespace support).

Nomad v1.1.0 introduced a regression where Nomad would fail the validation
when submitting Connect jobs and allow_unauthenticated set to true, with
Consul OSS - because it would do the namespace check against the Consul ACL
token assuming the "default" namespace, which does not work because Consul OSS
does not have namespaces.

Instead of making the bad assumption, expand the namespace check to handle
each special case explicitly.

Fixes #10718
2021-06-08 13:55:57 -05:00

637 lines
20 KiB
Go

package nomad
import (
"context"
"fmt"
"strings"
"sync"
"time"
"github.com/armon/go-metrics"
"github.com/hashicorp/consul/api"
"github.com/hashicorp/go-hclog"
"github.com/hashicorp/nomad/command/agent/consul"
"github.com/hashicorp/nomad/helper"
"github.com/hashicorp/nomad/nomad/structs"
"github.com/pkg/errors"
"golang.org/x/sync/errgroup"
"golang.org/x/time/rate"
)
const (
// siTokenDescriptionFmt is the format for the .Description field of
// service identity tokens generated on behalf of Nomad.
siTokenDescriptionFmt = "_nomad_si [%s] [%s] [%s]"
// siTokenRequestRateLimit is the maximum number of requests per second Nomad
// will make against Consul for requesting SI tokens.
siTokenRequestRateLimit rate.Limit = 500
// siTokenMaxParallelRevokes is the maximum number of parallel SI token
// revocation requests Nomad will make against Consul.
siTokenMaxParallelRevokes = 64
// siTokenRevocationInterval is the interval at which SI tokens that failed
// initial revocation are retried.
siTokenRevocationInterval = 5 * time.Minute
)
const (
// configEntriesRequestRateLimit is the maximum number of requests per second
// Nomad will make against Consul for operations on global Configuration Entry
// objects.
configEntriesRequestRateLimit rate.Limit = 10
)
const (
// ConsulPolicyWrite is the literal text of the policy field of a Consul Policy
// Rule that we check when validating an Operator Consul token against the
// necessary permissions for creating a Service Identity token for a given
// service.
//
// The rule may be:
// - service.<exact>
// - service."*" (wildcard)
// - service_prefix.<matching> (including empty string)
//
// e.g.
// service "web" { policy = "write" }
// service_prefix "" { policy = "write" }
ConsulPolicyWrite = "write"
// ConsulPolicyRead is the literal text of the policy field of a Consul Policy
// Rule that we check when validating a job-submitter Consul token against the
// necessary permissions for reading the key-value store.
//
// The only acceptable rule is
// - service_prefix "" { policy = "read|write" }
ConsulPolicyRead = "read"
)
type ServiceIdentityRequest struct {
ConsulNamespace string
TaskKind structs.TaskKind
TaskName string
ClusterID string
AllocID string
}
func (sir ServiceIdentityRequest) Validate() error {
switch {
case sir.ClusterID == "":
return errors.New("cluster id not set")
case sir.AllocID == "":
return errors.New("alloc id not set")
case sir.TaskName == "":
return errors.New("task name not set")
case sir.TaskKind == "":
return errors.New("task kind not set")
default:
return nil
}
}
func (sir ServiceIdentityRequest) Description() string {
return fmt.Sprintf(siTokenDescriptionFmt, sir.ClusterID, sir.AllocID, sir.TaskName)
}
// ConsulACLsAPI is an abstraction over the consul/api.ACL API used by Nomad
// Server.
//
// ACL requirements
// - acl:write (transitive through ACLsAPI)
type ConsulACLsAPI interface {
// CheckPermissions checks that the given Consul token has the necessary ACL
// permissions for each way that Consul is used as indicated by usage,
// returning an error if not.
CheckPermissions(ctx context.Context, namespace string, usage *structs.ConsulUsage, secretID string) error
// Create instructs Consul to create a Service Identity token.
CreateToken(context.Context, ServiceIdentityRequest) (*structs.SIToken, error)
// RevokeTokens instructs Consul to revoke the given token accessors.
RevokeTokens(context.Context, []*structs.SITokenAccessor, bool) bool
// MarkForRevocation marks the tokens for background revocation
MarkForRevocation([]*structs.SITokenAccessor)
// Stop is used to stop background token revocations. Intended to be used
// on Nomad Server shutdown.
Stop()
}
// PurgeSITokenAccessorFunc is called to remove SI Token accessors from the
// system (i.e. raft). If the function returns an error, the token will still
// be tracked and revocation attempts will retry in the background until there
// is a success.
type PurgeSITokenAccessorFunc func([]*structs.SITokenAccessor) error
type SITokenStats struct {
TrackedForRevoke int
}
type consulACLsAPI struct {
// aclClient is the API subset of the real consul client we need for
// managing Service Identity tokens
aclClient consul.ACLsAPI
// limiter is used to rate limit requests to consul
limiter *rate.Limiter
bgRevokeLock sync.Mutex
// Track accessors that must have their revocation retried in the background.
bgRetryRevocation []*structs.SITokenAccessor
// Track whether the background revocations have been stopped, to avoid
// creating tokens we would no longer be able to revoke. Expected to be used
// on a Server shutdown.
bgRevokeStopped bool
// purgeFunc is the Nomad Server function that removes the reference to the
// SI token accessor from the persistent raft store
purgeFunc PurgeSITokenAccessorFunc
// stopC is used to signal the client is shutting down and token revocation
// background goroutine should stop
stopC chan struct{}
// logger is used to log messages
logger hclog.Logger
}
func NewConsulACLsAPI(aclClient consul.ACLsAPI, logger hclog.Logger, purgeFunc PurgeSITokenAccessorFunc) *consulACLsAPI {
if purgeFunc == nil {
purgeFunc = func([]*structs.SITokenAccessor) error { return nil }
}
c := &consulACLsAPI{
aclClient: aclClient,
limiter: rate.NewLimiter(siTokenRequestRateLimit, int(siTokenRequestRateLimit)),
stopC: make(chan struct{}),
purgeFunc: purgeFunc,
logger: logger.Named("consul_acl"),
}
go c.bgRetryRevokeDaemon()
return c
}
// Stop stops background token revocations from happening. Once stopped, tokens
// may no longer be created.
func (c *consulACLsAPI) Stop() {
c.bgRevokeLock.Lock()
defer c.bgRevokeLock.Unlock()
c.stopC <- struct{}{}
c.bgRevokeStopped = true
}
func (c *consulACLsAPI) readToken(ctx context.Context, secretID string) (*api.ACLToken, error) {
defer metrics.MeasureSince([]string{"nomad", "consul", "read_token"}, time.Now())
if id := strings.TrimSpace(secretID); !helper.IsUUID(id) {
return nil, errors.New("missing consul token")
}
// Ensure we are under our rate limit.
if err := c.limiter.Wait(ctx); err != nil {
return nil, errors.Wrap(err, "unable to read consul token")
}
consulToken, _, err := c.aclClient.TokenReadSelf(&api.QueryOptions{
AllowStale: false,
Token: secretID,
})
if err != nil {
return nil, errors.Wrap(err, "unable to read consul token")
}
return consulToken, nil
}
func (c *consulACLsAPI) CheckPermissions(ctx context.Context, namespace string, usage *structs.ConsulUsage, secretID string) error {
// consul not used, nothing to check
if !usage.Used() {
return nil
}
// lookup the token from consul
token, readErr := c.readToken(ctx, secretID)
if readErr != nil {
return readErr
}
// if the token is a global-management token, it has unrestricted privileges
if c.isManagementToken(token) {
return nil
}
// if the token cannot possibly be used to act on objects in the desired
// namespace, reject it immediately
if err := namespaceCheck(namespace, token); err != nil {
return err
}
// verify token has keystore read permission, if using template
if usage.KV {
allowable, err := c.canReadKeystore(namespace, token)
if err != nil {
return err
} else if !allowable {
return errors.New("insufficient Consul ACL permissions to use template")
}
}
// verify token has service write permission for group+task services
for _, service := range usage.Services {
allowable, err := c.canWriteService(namespace, service, token)
if err != nil {
return err
} else if !allowable {
return errors.Errorf("insufficient Consul ACL permissions to write service %q", service)
}
}
// verify token has service identity permission for connect services
for _, kind := range usage.Kinds {
service := kind.Value()
allowable, err := c.canWriteService(namespace, service, token)
if err != nil {
return err
} else if !allowable {
return errors.Errorf("insufficient Consul ACL permissions to write Connect service %q", service)
}
}
return nil
}
func (c *consulACLsAPI) CreateToken(ctx context.Context, sir ServiceIdentityRequest) (*structs.SIToken, error) {
defer metrics.MeasureSince([]string{"nomad", "consul", "create_token"}, time.Now())
// make sure the background token revocations have not been stopped
c.bgRevokeLock.Lock()
stopped := c.bgRevokeStopped
c.bgRevokeLock.Unlock()
if stopped {
return nil, errors.New("client stopped and may no longer create tokens")
}
// Check the metadata for the token we want
if err := sir.Validate(); err != nil {
return nil, err
}
// the SI token created must be for the service, not the sidecar of the service
// https://www.consul.io/docs/acl/acl-system.html#acl-service-identities
service := sir.TaskKind.Value()
partial := &api.ACLToken{
Description: sir.Description(),
ServiceIdentities: []*api.ACLServiceIdentity{{ServiceName: service}},
Namespace: sir.ConsulNamespace,
}
// Ensure we are under our rate limit.
if err := c.limiter.Wait(ctx); err != nil {
return nil, err
}
token, _, err := c.aclClient.TokenCreate(partial, nil)
if err != nil {
return nil, err
}
return &structs.SIToken{
ConsulNamespace: token.Namespace,
AccessorID: token.AccessorID,
SecretID: token.SecretID,
TaskName: sir.TaskName,
}, nil
}
// RevokeTokens revokes the passed set of SI token accessors. If committed is set,
// the client's purge function is called (which purges the tokens from the Server's
// persistent store). If there is an error purging either because of Consul failures
// or because of the purge function, the revocation is retried in the background.
//
// The revocation of an SI token accessor is idempotent.
//
// A return value of true indicates one or more accessors were stored for
// a revocation retry attempt in the background (intended for tests).
func (c *consulACLsAPI) RevokeTokens(ctx context.Context, accessors []*structs.SITokenAccessor, committed bool) bool {
defer metrics.MeasureSince([]string{"nomad", "consul", "revoke_tokens"}, time.Now())
nTokens := float32(len(accessors))
if err := c.parallelRevoke(ctx, accessors); err != nil {
// If these tokens were uncommitted into raft, it is a best effort to
// revoke them now. If this immediate revocation does not work, Nomad loses
// track of them and will need to do a brute reconciliation later. This
// should happen rarely, and will be implemented soon.
if !committed {
metrics.IncrCounter([]string{"nomad", "consul", "undistributed_si_tokens_abandoned"}, nTokens)
}
c.logger.Warn("failed to revoke tokens, will reattempt later", "error", err)
c.storeForRevocation(accessors)
return true
}
if !committed {
// Un-committed tokens were revoked without incident (nothing to purge)
metrics.IncrCounter([]string{"nomad", "consul", "undistributed_si_tokens_revoked"}, nTokens)
return false
}
// Committed tokens were revoked without incident, now purge them
if err := c.purgeFunc(accessors); err != nil {
c.logger.Error("failed to purge SI token accessors", "error", err)
c.storeForRevocation(accessors)
return true
}
// Track that the SI tokens were revoked and purged successfully
metrics.IncrCounter([]string{"nomad", "consul", "distributed_si_tokens_revoked"}, nTokens)
return false
}
func (c *consulACLsAPI) MarkForRevocation(accessors []*structs.SITokenAccessor) {
c.storeForRevocation(accessors)
}
func (c *consulACLsAPI) storeForRevocation(accessors []*structs.SITokenAccessor) {
c.bgRevokeLock.Lock()
defer c.bgRevokeLock.Unlock()
// copy / append the set of accessors we must track for revocation in the
// background
c.bgRetryRevocation = append(c.bgRetryRevocation, accessors...)
}
func (c *consulACLsAPI) parallelRevoke(ctx context.Context, accessors []*structs.SITokenAccessor) error {
g, pCtx := errgroup.WithContext(ctx)
// Cap the handlers
handlers := len(accessors)
if handlers > siTokenMaxParallelRevokes {
handlers = siTokenMaxParallelRevokes
}
// Revoke the SI Token Accessors
input := make(chan *structs.SITokenAccessor, handlers)
for i := 0; i < handlers; i++ {
g.Go(func() error {
for {
select {
case accessor, ok := <-input:
if !ok {
return nil
}
if err := c.singleRevoke(ctx, accessor); err != nil {
return errors.Wrapf(err,
"failed to revoke SI token accessor (alloc %q, node %q, task %q)",
accessor.AllocID, accessor.NodeID, accessor.TaskName,
)
}
case <-pCtx.Done():
return nil
}
}
})
}
// Send the input
go func() {
defer close(input)
for _, accessor := range accessors {
select {
case <-pCtx.Done():
return
case input <- accessor:
}
}
}()
// Wait for everything to complete
return g.Wait()
}
func (c *consulACLsAPI) singleRevoke(ctx context.Context, accessor *structs.SITokenAccessor) error {
c.logger.Trace("revoke SI token", "task", accessor.TaskName, "alloc_id", accessor.AllocID, "node_id", accessor.NodeID)
// Ensure we are under our rate limit.
if err := c.limiter.Wait(ctx); err != nil {
return err
}
// Consul will no-op the deletion of a non-existent token (no error)
_, err := c.aclClient.TokenDelete(accessor.AccessorID, &api.WriteOptions{Namespace: accessor.ConsulNamespace})
return err
}
func (c *consulACLsAPI) bgRetryRevokeDaemon() {
ticker := time.NewTicker(siTokenRevocationInterval)
defer ticker.Stop()
for {
select {
case <-c.stopC:
return
case <-ticker.C:
c.bgRetryRevoke()
}
}
}
// maxConsulRevocationBatchSize is the maximum tokens a bgRetryRevoke should revoke
// at any given time.
const maxConsulRevocationBatchSize = 1000
func (c *consulACLsAPI) bgRetryRevoke() {
c.bgRevokeLock.Lock()
defer c.bgRevokeLock.Unlock()
// fast path, nothing to do
if len(c.bgRetryRevocation) == 0 {
return
}
// unlike vault tokens, SI tokens do not have a TTL, and so we must try to
// remove all SI token accessors, every time, until they're gone
toRevoke := len(c.bgRetryRevocation)
if toRevoke > maxConsulRevocationBatchSize {
toRevoke = maxConsulRevocationBatchSize
}
toPurge := make([]*structs.SITokenAccessor, toRevoke)
copy(toPurge, c.bgRetryRevocation)
if err := c.parallelRevoke(context.Background(), toPurge); err != nil {
c.logger.Warn("background SI token revocation failed", "error", err)
return
}
// Call the revocation function
if err := c.purgeFunc(toPurge); err != nil {
// Just try again later (revocation is idempotent)
c.logger.Error("background SI token purge failed", "error", err)
return
}
// Track that the SI tokens were revoked successfully
nTokens := float32(len(toPurge))
metrics.IncrCounter([]string{"nomad", "consul", "distributed_tokens_revoked"}, nTokens)
// Reset the list of accessors to retry, since we just removed them all.
c.bgRetryRevocation = nil
}
func (c *consulACLsAPI) ListTokens() ([]string, error) {
// defer metrics.MeasureSince([]string{"nomad", "consul", "list_tokens"}, time.Now())
return nil, errors.New("not yet implemented")
}
// purgeSITokenAccessors is the Nomad Server method which will remove the set
// of SI token accessors from the persistent raft store.
func (s *Server) purgeSITokenAccessors(accessors []*structs.SITokenAccessor) error {
// Commit this update via Raft
request := structs.SITokenAccessorsRequest{Accessors: accessors}
_, _, err := s.raftApply(structs.ServiceIdentityAccessorDeregisterRequestType, request)
return err
}
// ConsulConfigsAPI is an abstraction over the consul/api.ConfigEntries API used by
// Nomad Server.
//
// Nomad will only perform write operations on Consul Ingress/Terminating Gateway
// Configuration Entries. Removing the entries is not yet safe, given that multiple
// Nomad clusters may be writing to the same config entries, which are global in
// the Consul scope. There was a Meta field introduced which Nomad can leverage
// in the future, when Consul no longer supports versions that do not contain the
// field. The Meta field would be used to track which Nomad "owns" the CE.
// https://github.com/hashicorp/nomad/issues/8971
type ConsulConfigsAPI interface {
// SetIngressCE adds the given ConfigEntry to Consul, overwriting
// the previous entry if set.
SetIngressCE(ctx context.Context, namespace, service string, entry *structs.ConsulIngressConfigEntry) error
// SetTerminatingCE adds the given ConfigEntry to Consul, overwriting
// the previous entry if set.
SetTerminatingCE(ctx context.Context, namespace, service string, entry *structs.ConsulTerminatingConfigEntry) error
// Stop is used to stop additional creations of Configuration Entries. Intended to
// be used on Nomad Server shutdown.
Stop()
}
type consulConfigsAPI struct {
// configsClient is the API subset of the real Consul client we need for
// managing Configuration Entries.
configsClient consul.ConfigAPI
// limiter is used to rate limit requests to Consul
limiter *rate.Limiter
// logger is used to log messages
logger hclog.Logger
// lock protects the stopped flag, which prevents use of the consul configs API
// client after shutdown.
lock sync.Mutex
stopped bool
}
func NewConsulConfigsAPI(configsClient consul.ConfigAPI, logger hclog.Logger) *consulConfigsAPI {
return &consulConfigsAPI{
configsClient: configsClient,
limiter: rate.NewLimiter(configEntriesRequestRateLimit, int(configEntriesRequestRateLimit)),
logger: logger,
}
}
func (c *consulConfigsAPI) Stop() {
c.lock.Lock()
defer c.lock.Unlock()
c.stopped = true
}
func (c *consulConfigsAPI) SetIngressCE(ctx context.Context, namespace, service string, entry *structs.ConsulIngressConfigEntry) error {
return c.setCE(ctx, convertIngressCE(namespace, service, entry))
}
func (c *consulConfigsAPI) SetTerminatingCE(ctx context.Context, namespace, service string, entry *structs.ConsulTerminatingConfigEntry) error {
return c.setCE(ctx, convertTerminatingCE(namespace, service, entry))
}
// setCE will set the Configuration Entry of any type Consul supports.
func (c *consulConfigsAPI) setCE(ctx context.Context, entry api.ConfigEntry) error {
defer metrics.MeasureSince([]string{"nomad", "consul", "create_config_entry"}, time.Now())
// make sure the background deletion goroutine has not been stopped
c.lock.Lock()
stopped := c.stopped
c.lock.Unlock()
if stopped {
return errors.New("client stopped and may not longer create config entries")
}
// ensure we are under our wait limit
if err := c.limiter.Wait(ctx); err != nil {
return err
}
_, _, err := c.configsClient.Set(entry, &api.WriteOptions{Namespace: entry.GetNamespace()})
return err
}
func convertIngressCE(namespace, service string, entry *structs.ConsulIngressConfigEntry) api.ConfigEntry {
var listeners []api.IngressListener = nil
for _, listener := range entry.Listeners {
var services []api.IngressService = nil
for _, s := range listener.Services {
services = append(services, api.IngressService{
Name: s.Name,
Hosts: helper.CopySliceString(s.Hosts),
})
}
listeners = append(listeners, api.IngressListener{
Port: listener.Port,
Protocol: listener.Protocol,
Services: services,
})
}
tlsEnabled := false
if entry.TLS != nil && entry.TLS.Enabled {
tlsEnabled = true
}
return &api.IngressGatewayConfigEntry{
Namespace: namespace,
Kind: api.IngressGateway,
Name: service,
TLS: api.GatewayTLSConfig{Enabled: tlsEnabled},
Listeners: listeners,
}
}
func convertTerminatingCE(namespace, service string, entry *structs.ConsulTerminatingConfigEntry) api.ConfigEntry {
var linked []api.LinkedService = nil
for _, s := range entry.Services {
linked = append(linked, api.LinkedService{
Name: s.Name,
CAFile: s.CAFile,
CertFile: s.CertFile,
KeyFile: s.KeyFile,
SNI: s.SNI,
})
}
return &api.TerminatingGatewayConfigEntry{
Namespace: namespace,
Kind: api.TerminatingGateway,
Name: service,
Services: linked,
}
}