sr: Avoid recursion when loading site replicator credentials (#20262)

If the site replication is enabled and the code tries to extract jwt
claims while the site replication service account credentials are still
not loaded yet, the code will enter an infinite loop, causing in a
high CPU usage.

Another possibility of the infinite loop is having some service accounts
created by an old deployment version where the service account JWT was
signed by the root credentials, but not anymore.

This commit will remove the possibility of the infinite loop in the code
and add root credential fallback to extract claims from old service
accounts.

cmd/iam-etcd-store.go

@@ -249,6 +249,18 @@ func (ies *IAMEtcdStore) addUser(ctx context.Context, user string, userType IAMU
 	return nil
 }
 
+func (ies *IAMEtcdStore) loadSecretKey(ctx context.Context, user string, userType IAMUserType) (string, error) {
+	var u UserIdentity
+	err := ies.loadIAMConfig(ctx, &u, getUserIdentityPath(user, userType))
+	if err != nil {
+		if errors.Is(err, errConfigNotFound) {
+			return "", errNoSuchUser
+		}
+		return "", err
+	}
+	return u.Credentials.SecretKey, nil
+}
+
 func (ies *IAMEtcdStore) loadUser(ctx context.Context, user string, userType IAMUserType, m map[string]UserIdentity) error {
 	var u UserIdentity
 	err := ies.loadIAMConfig(ctx, &u, getUserIdentityPath(user, userType))

cmd/iam-object-store.go

@@ -225,6 +225,18 @@ func (iamOS *IAMObjectStore) loadPolicyDocs(ctx context.Context, m map[string]Po
 	return nil
 }
 
+func (iamOS *IAMObjectStore) loadSecretKey(ctx context.Context, user string, userType IAMUserType) (string, error) {
+	var u UserIdentity
+	err := iamOS.loadIAMConfig(ctx, &u, getUserIdentityPath(user, userType))
+	if err != nil {
+		if errors.Is(err, errConfigNotFound) {
+			return "", errNoSuchUser
+		}
+		return "", err
+	}
+	return u.Credentials.SecretKey, nil
+}
+
 func (iamOS *IAMObjectStore) loadUser(ctx context.Context, user string, userType IAMUserType, m map[string]UserIdentity) error {
 	var u UserIdentity
 	err := iamOS.loadIAMConfig(ctx, &u, getUserIdentityPath(user, userType))

cmd/iam-store.go

@@ -535,6 +535,7 @@ type IAMStorageAPI interface {
 	loadPolicyDocWithRetry(ctx context.Context, policy string, m map[string]PolicyDoc, retries int) error
 	loadPolicyDocs(ctx context.Context, m map[string]PolicyDoc) error
 	loadUser(ctx context.Context, user string, userType IAMUserType, m map[string]UserIdentity) error
+	loadSecretKey(ctx context.Context, user string, userType IAMUserType) (string, error)
 	loadUsers(ctx context.Context, userType IAMUserType, m map[string]UserIdentity) error
 	loadGroup(ctx context.Context, group string, m map[string]GroupInfo) error
 	loadGroups(ctx context.Context, m map[string]GroupInfo) error
@@ -2820,20 +2821,29 @@ func (store *IAMStoreSys) LoadUser(ctx context.Context, accessKey string) error
 	return err
 }
 
-func extractJWTClaims(u UserIdentity) (*jwt.MapClaims, error) {
+func extractJWTClaims(u UserIdentity) (jwtClaims *jwt.MapClaims, err error) {
-	jwtClaims, err := auth.ExtractClaims(u.Credentials.SessionToken, u.Credentials.SecretKey)
+	keys := make([]string, 0, 3)
-	if err != nil {
+	// Append credentials secret key itself
-		secretKey, err := getTokenSigningKey()
+	keys = append(keys, u.Credentials.SecretKey)
-		if err != nil {
+	// Use site-replication credentials if found
-			return nil, err
+	if globalSiteReplicationSys.isEnabled() {
+		siteReplSecretKey, e := globalSiteReplicatorCred.Get(GlobalContext)
+		if e != nil {
+			return nil, e
 		}
-		// Session tokens for STS creds will be generated with root secret or site-replicator-0 secret
+		keys = append(keys, siteReplSecretKey)
-		jwtClaims, err = auth.ExtractClaims(u.Credentials.SessionToken, secretKey)
+	}
-		if err != nil {
+	// Use root credentials for credentials created with older deployments
-			return nil, err
+	keys = append(keys, globalActiveCred.SecretKey)
+
+	// Iterate over all keys and return with the first successful claim extraction
+	for _, key := range keys {
+		jwtClaims, err = auth.ExtractClaims(u.Credentials.SessionToken, key)
+		if err == nil {
+			break
 		}
 	}
-	return jwtClaims, nil
+	return
 }
 
 func validateSvcExpirationInUTC(expirationInUTC time.Time) error {

cmd/iam.go

@@ -211,7 +211,7 @@ func (sys *IAMSys) Load(ctx context.Context, firstTime bool) error {
 	if !globalSiteReplicatorCred.IsValid() {
 		sa, _, err := sys.getServiceAccount(ctx, siteReplicatorSvcAcc)
 		if err == nil {
-			globalSiteReplicatorCred.Set(sa.Credentials)
+			globalSiteReplicatorCred.Set(sa.Credentials.SecretKey)
 		}
 	}
 

cmd/site-replication.go

@@ -597,7 +597,7 @@ func (c *SiteReplicationSys) AddPeerClusters(ctx context.Context, psites []madmi
 	}
 
 	if !globalSiteReplicatorCred.IsValid() {
-		globalSiteReplicatorCred.Set(svcCred)
+		globalSiteReplicatorCred.Set(svcCred.SecretKey)
 	}
 	result := madmin.ReplicateAddStatus{
 		Success: true,
@@ -659,7 +659,7 @@ func (c *SiteReplicationSys) PeerJoinReq(ctx context.Context, arg madmin.SRPeerJ
 		return errSRBackendIssue(fmt.Errorf("unable to save cluster-replication state to drive on %s: %v", ourName, err))
 	}
 	if !globalSiteReplicatorCred.IsValid() {
-		globalSiteReplicatorCred.Set(sa)
+		globalSiteReplicatorCred.Set(sa.SecretKey)
 	}
 
 	return nil
@@ -6269,34 +6269,36 @@ func ilmExpiryReplicationEnabled(sites map[string]madmin.PeerInfo) bool {
 }
 
 type siteReplicatorCred struct {
-	Creds auth.Credentials
+	secretKey string
 	sync.RWMutex
 }
 
 // Get or attempt to load site replicator credentials from disk.
-func (s *siteReplicatorCred) Get(ctx context.Context) (auth.Credentials, error) {
+func (s *siteReplicatorCred) Get(ctx context.Context) (string, error) {
 	s.RLock()
-	if s.Creds.IsValid() {
+	secretKey := s.secretKey
 	s.RUnlock()
-		return s.Creds, nil
+
+	if secretKey != "" {
+		return secretKey, nil
 	}
-	s.RUnlock()
+
-	m := make(map[string]UserIdentity)
+	secretKey, err := globalIAMSys.store.loadSecretKey(ctx, siteReplicatorSvcAcc, svcUser)
-	if err := globalIAMSys.store.loadUser(ctx, siteReplicatorSvcAcc, svcUser, m); err != nil {
+	if err != nil {
-		return auth.Credentials{}, err
+		return "", err
 	}
-	s.Set(m[siteReplicatorSvcAcc].Credentials)
+	s.Set(secretKey)
-	return m[siteReplicatorSvcAcc].Credentials, nil
+	return secretKey, nil
 }
 
-func (s *siteReplicatorCred) Set(c auth.Credentials) {
+func (s *siteReplicatorCred) Set(secretKey string) {
 	s.Lock()
 	defer s.Unlock()
-	s.Creds = c
+	s.secretKey = secretKey
 }
 
 func (s *siteReplicatorCred) IsValid() bool {
 	s.RLock()
 	defer s.RUnlock()
-	return s.Creds.IsValid()
+	return s.secretKey != ""
 }

cmd/sts-handlers.go

@@ -241,11 +241,11 @@ func parseForm(r *http.Request) error {
 func getTokenSigningKey() (string, error) {
 	secret := globalActiveCred.SecretKey
 	if globalSiteReplicationSys.isEnabled() {
-		c, err := globalSiteReplicatorCred.Get(GlobalContext)
+		secretKey, err := globalSiteReplicatorCred.Get(GlobalContext)
 		if err != nil {
 			return "", err
 		}
-		return c.SecretKey, nil
+		return secretKey, nil
 	}
 	return secret, nil
 }