Bulk API Keys update

internal/pkg/api/handleAck.go

@@ -24,6 +24,7 @@ import (
 	"github.com/elastic/fleet-server/v7/internal/pkg/logger"
 	"github.com/elastic/fleet-server/v7/internal/pkg/model"
 	"github.com/elastic/fleet-server/v7/internal/pkg/policy"
+	"github.com/elastic/fleet-server/v7/internal/pkg/smap"
 	"github.com/pkg/errors"
 
 	"github.com/julienschmidt/httprouter"
@@ -349,10 +350,44 @@ func (ack *AckT) handlePolicyChange(ctx context.Context, zlog zerolog.Logger, ag
 		return nil
 	}
 
+	if agent.DefaultAPIKeyID != "" {
+		res, err := ack.bulk.APIKeyRead(ctx, agent.DefaultAPIKeyID, true)
+		if err != nil {
+			zlog.Info().
+				Err(err).
+				Str("id", agent.DefaultAPIKeyID).
+				Msg("Failed to read API Key roles")
+		} else {
+			clean, removedCount, err := cleanRoles(res.RoleDescriptors)
+			if err != nil {
+				zlog.Info().
+					Err(err).
+					RawJSON("roles", res.RoleDescriptors).
+					Str("id", agent.DefaultAPIKeyID).
+					Msg("Failed to cleanup roles")
+			} else if removedCount > 0 {
+				if err := ack.bulk.APIKeyUpdate(ctx, agent.DefaultAPIKeyID, agent.PolicyOutputPermissionsHash, clean); err != nil {
+					zlog.Info().Err(err).RawJSON("roles", clean).Str("id", agent.DefaultAPIKeyID).Msg("Failed to refresh API Key")
+				} else {
+					zlog.Debug().
+						Str("hash.sha256", agent.PolicyOutputPermissionsHash).
+						Str("apikey", agent.DefaultAPIKeyID).
+						RawJSON("roles", clean).
+						Int("removed", removedCount).
+						Msg("Updating agent record to pick up reduced roles.")
+				}
+			}
+		}
+	}
+
 	sz := len(agent.DefaultAPIKeyHistory)
 	if sz > 0 {
 		ids := make([]string, sz)
 		for i := 0; i < sz; i++ {
+			if agent.DefaultAPIKeyHistory[i].ID == agent.DefaultAPIKeyID {
+				// already updated
+				continue
+			}
 			ids[i] = agent.DefaultAPIKeyHistory[i].ID
 		}
 		log.Info().Strs("ids", ids).Msg("Invalidate old API keys")
@@ -385,6 +420,32 @@ func (ack *AckT) handlePolicyChange(ctx context.Context, zlog zerolog.Logger, ag
 	return errors.Wrap(err, "handlePolicyChange update")
 }
 
+func cleanRoles(roles json.RawMessage) (json.RawMessage, int, error) {
+	rr := smap.Map{}
+	if err := json.Unmarshal(roles, &rr); err != nil {
+		return nil, 0, err
+	}
+
+	keys := make([]string, 0, len(rr))
+	for k := range rr {
+		if strings.HasSuffix(k, "-rdstale") {
+			keys = append(keys, k)
+		}
+
+	}
+
+	if len(keys) == 0 {
+		return roles, 0, nil
+	}
+
+	for _, k := range keys {
+		delete(rr, k)
+	}
+
+	r, err := json.Marshal(rr)
+	return r, len(keys), err
+}
+
 func (ack *AckT) handleUnenroll(ctx context.Context, zlog zerolog.Logger, agent *model.Agent) error {
 	apiKeys := _getAPIKeyIDs(agent)
 	if len(apiKeys) > 0 {

internal/pkg/api/handleEnroll.go

@@ -293,7 +293,7 @@ func invalidateAPIKey(ctx context.Context, zlog zerolog.Logger, bulker bulk.Bulk
 LOOP:
 	for {
 
-		_, err := bulker.APIKeyRead(ctx, apikeyID)
+		_, err := bulker.APIKeyRead(ctx, apikeyID, false)
 
 		switch {
 		case err == nil:

internal/pkg/apikey/apikey_integration_test.go

@@ -54,7 +54,7 @@ func TestCreateAPIKeyWithMetadata(t *testing.T) {
 	}
 
 	// Get the key and verify that metadata was saved correctly
-	aKeyMeta, err := Read(ctx, es, akey.ID)
+	aKeyMeta, err := Read(ctx, es, akey.ID, false)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -80,7 +80,7 @@ func TestCreateAPIKeyWithMetadata(t *testing.T) {
 	}
 
 	// Try to get the key that doesn't exists, expect ErrApiKeyNotFound
-	_, err = Read(ctx, es, "0000000000000")
+	_, err = Read(ctx, es, "0000000000000", false)
 	if !errors.Is(err, ErrAPIKeyNotFound) {
 		t.Errorf("Unexpected error type: %v", err)
 	}

internal/pkg/apikey/get.go

@@ -15,17 +15,21 @@ import (
 
 // APIKetMetadata tracks Metadata associated with an APIKey.
 type APIKeyMetadata struct {
-	ID       string
-	Metadata Metadata
+	ID              string
+	Metadata        Metadata
+	RoleDescriptors json.RawMessage
 }
 
 // Read gathers APIKeyMetadata from Elasticsearch using the given client.
-func Read(ctx context.Context, client *elasticsearch.Client, id string) (*APIKeyMetadata, error) {
+func Read(ctx context.Context, client *elasticsearch.Client, id string, withOwner bool) (*APIKeyMetadata, error) {
 
 	opts := []func(*esapi.SecurityGetAPIKeyRequest){
 		client.Security.GetAPIKey.WithContext(ctx),
 		client.Security.GetAPIKey.WithID(id),
 	}
+	if withOwner {
+		opts = append(opts, client.Security.GetAPIKey.WithOwner(true))
+	}
 
 	res, err := client.Security.GetAPIKey(
 		opts...,
@@ -42,8 +46,9 @@ func Read(ctx context.Context, client *elasticsearch.Client, id string) (*APIKey
 	}
 
 	type APIKeyResponse struct {
-		ID       string   `json:"id"`
-		Metadata Metadata `json:"metadata"`
+		ID              string          `json:"id"`
+		Metadata        Metadata        `json:"metadata"`
+		RoleDescriptors json.RawMessage `json:"role_descriptors"`
 	}
 	type GetAPIKeyResponse struct {
 		APIKeys []APIKeyResponse `json:"api_keys"`
@@ -62,7 +67,8 @@ func Read(ctx context.Context, client *elasticsearch.Client, id string) (*APIKey
 	first := resp.APIKeys[0]
 
 	return &APIKeyMetadata{
-		ID:       first.ID,
-		Metadata: first.Metadata,
+		ID:              first.ID,
+		Metadata:        first.Metadata,
+		RoleDescriptors: first.RoleDescriptors,
 	}, nil
 }

internal/pkg/bulk/block.go

@@ -43,6 +43,7 @@ const (
 	ActionDelete
 	ActionIndex
 	ActionUpdate
+	ActionUpdateAPIKey
 	ActionRead
 	ActionSearch
 	ActionFleetSearch
@@ -53,6 +54,7 @@ var actionStrings = []string{
 	"delete",
 	"index",
 	"update",
+	"update_api_key",
 	"read",
 	"search",
 	"fleet_search",

internal/pkg/bulk/engine.go

@@ -55,9 +55,10 @@ type Bulk interface {
 
 	// APIKey operations
 	APIKeyCreate(ctx context.Context, name, ttl string, roles []byte, meta interface{}) (*APIKey, error)
-	APIKeyRead(ctx context.Context, id string) (*APIKeyMetadata, error)
+	APIKeyRead(ctx context.Context, id string, withOwner bool) (*APIKeyMetadata, error)
 	APIKeyAuth(ctx context.Context, key APIKey) (*SecurityInfo, error)
 	APIKeyInvalidate(ctx context.Context, ids ...string) error
+	APIKeyUpdate(ctx context.Context, id, outputPolicyHash string, roles []byte) error
 
 	// Accessor used to talk to elastic search direcly bypassing bulk engine
 	Client() *elasticsearch.Client
@@ -81,6 +82,7 @@ const (
 	defaultMaxPending        = 32
 	defaultBlockQueueSz      = 32 // Small capacity to allow multiOp to spin fast
 	defaultAPIKeyMaxParallel = 32
+	defaultApikeyMaxReqSize  = 100 * 1024 * 1024
 )
 
 func NewBulker(es esapi.Transport, tracer *apm.Tracer, opts ...BulkOpt) *Bulker {
@@ -136,6 +138,8 @@ func blkToQueueType(blk *bulkT) queueType {
 		} else {
 			queueIdx = kQueueRead
 		}
+	case ActionUpdateAPIKey:
+		queueIdx = kQueueAPIKeyUpdate
 	default:
 		if forceRefresh {
 			queueIdx = kQueueRefreshBulk
@@ -288,6 +292,8 @@ func (b *Bulker) flushQueue(ctx context.Context, w *semaphore.Weighted, queue qu
 			err = b.flushRead(ctx, queue)
 		case kQueueSearch, kQueueFleetSearch:
 			err = b.flushSearch(ctx, queue)
+		case kQueueAPIKeyUpdate:
+			err = b.flushUpdateAPIKey(ctx, queue)
 		default:
 			err = b.flushBulk(ctx, queue)
 		}