fix(iam): persist and propagate scoped STS revocation

Enumerate durable STS identities for parent/type revocation, retain the operation cutoff, and notify sibling caches for each committed deletion. Repair missed notifications during periodic object/etcd IAM refresh. Carry immutable STS tombstones and expiry through the existing IAM revision protocol and deletion healing. Delayed creation cannot undo revocation, while unrelated and newly issued keys stay valid. Keep partial failures visible while propagating the successfully committed deletions. Cover uncached credentials, warm sibling caches, dropped notifications, object/etcd backends, reordered replay, concurrent issuance and partial persistence failure. No new dependencies or authentication-path I/O. Signed-off-by: Feng Ruohang <rh@vonng.com>
2026-10-18 23:46:11 +03:00 · 2026-10-18 22:39:12 +08:00
parent 1748f62c7e
commit 0cd388c666
5 changed files with 365 additions and 26 deletions
@@ -29,6 +29,7 @@ const (
 	iamRevisionPeerPath  = "/v3/site-replication/peer/iam-revisions"
 	iamUserBoundaryType  = "silo-user-revocation"
 	iamGroupBoundaryType = "silo-group-revocation"
+	iamSTSDeleteType     = "silo-sts-revocation"
 	maxIAMRevisionBatch  = 128
 )

@@ -49,6 +50,7 @@ type iamGroupBoundary struct {
 // before applying any change that would lose revocation or member metadata.
 type iamReplicationItem struct {
 	madmin.SRIAMItem
+	STSExpiresAt    time.Time            `json:"stsExpiresAt,omitempty"`
 	GroupGrants     map[string]time.Time `json:"groupGrants,omitempty"`
 	GroupSnapshot   bool                 `json:"groupSnapshot,omitempty"`
 	UserRevocation  *iamUserBoundary     `json:"userRevocation,omitempty"`
@@ -155,6 +157,13 @@ type (

 func (c *SiteReplicationSys) replicationItem(ctx context.Context, item madmin.SRIAMItem) (iamReplicationItem, error) {
 	out := iamReplicationItem{SRIAMItem: item}
+	if item.Type == iamSTSDeleteType {
+		r, err := loadIAMRevision(ctx, globalIAMSys.store, getUserIdentityPath(item.Name, stsUser))
+		if err != nil {
+			return out, err
+		}
+		out.STSExpiresAt = r.ExpiresAt
+	}
 	if item.Type == madmin.SRIAMItemSvcAcc && item.SvcAccChange != nil {
 		var key string
 		if item.SvcAccChange.Create != nil {
@@ -242,6 +251,15 @@ func applyIAMReplicationItem(ctx context.Context, item iamReplicationItem) error
 		ctx = context.WithValue(ctx, iamRecordBoundaryKey{}, item.RevokedBefore)
 	}
 	switch item.Type {
+	case iamSTSDeleteType:
+		if item.Name == "" || strings.ContainsAny(item.Name, "/\\") || item.UpdatedAt.IsZero() {
+			return errSRInvalidRequest(errInvalidArgument)
+		}
+		if err := globalIAMSys.store.revokeTempUser(withIAMReplicationTime(ctx, item.UpdatedAt), item.Name, item.STSExpiresAt); err != nil {
+			return iamReplicationError(err)
+		}
+		globalIAMSys.notifyForUser(ctx, item.Name, true)
+		return nil
 	case iamUserBoundaryType:
 		if item.UserRevocation == nil || item.UserRevocation.User == "" || item.UserRevocation.Before.IsZero() {
 			return errSRInvalidRequest(errInvalidArgument)
@@ -348,6 +348,13 @@ func saveIAMRevision(ctx context.Context, store IAMStorageAPI, path string, item
 		// STS access keys identify immutable tokens, not reusable user names.
 		return errIAMStaleUpdate
 	}
+	if strings.HasPrefix(path, iamConfigSTSPrefix) && deleted {
+		// STS keys are immutable. A delayed revocation must also defeat a
+		// replayed creation that was recorded with a newer event timestamp.
+		if origin, replicated := iamReplicationTime(ctx); replicated && previous.timestamp().After(origin) {
+			ctx = withIAMReplicationTime(ctx, previous.timestamp())
+		}
+	}
 	if origin, replicated := iamReplicationTime(ctx); replicated {
 		*at = origin
 		if previous.timestamp().After(origin) || (previous.Deleted && !deleted && !origin.After(previous.timestamp())) {
@@ -394,11 +401,18 @@ func saveIAMRevision(ctx context.Context, store IAMStorageAPI, path string, item
 			// Retain only the parent name for root-account exclusion during heal.
 			v.Credentials = auth.Credentials{ParentUser: previous.Credentials.ParentUser}
 			v.RevokedBefore = *at
-			if strings.HasPrefix(path, iamConfigSTSPrefix) && !previous.Credentials.Expiration.IsZero() && !previous.Credentials.Expiration.Equal(timeSentinel) {
-				// The signed STS token cannot authorize beyond this time, even
-				// if an offline site replays it with a newer event timestamp.
-				v.ExpiresAt = previous.Credentials.Expiration.Add(globalMaxSkewTime)
-				opts = []options{{ttl: max(1, int64(time.Until(v.ExpiresAt).Seconds())+1)}}
+			if strings.HasPrefix(path, iamConfigSTSPrefix) {
+				// Carry expiry when a revocation arrives before the token itself.
+				// It is safe to forget this key only after its signed lifetime.
+				if expiration := previous.Credentials.Expiration; !expiration.IsZero() && !expiration.Equal(timeSentinel) && expiration.Add(globalMaxSkewTime).After(v.ExpiresAt) {
+					v.ExpiresAt = expiration.Add(globalMaxSkewTime)
+				}
+				if previous.ExpiresAt.After(v.ExpiresAt) {
+					v.ExpiresAt = previous.ExpiresAt
+				}
+				if !v.ExpiresAt.IsZero() {
+					opts = []options{{ttl: max(1, int64(time.Until(v.ExpiresAt).Seconds())+1)}}
+				}
 			}
 		} else {
 			if v.Credentials.SessionToken != "" && v.Credentials.Claims == nil {
@@ -489,6 +503,12 @@ func iamDeletionItem(path string, r iamRevision) (item madmin.SRIAMItem, ok bool
 		name := strings.TrimSuffix(strings.TrimPrefix(path, iamConfigUsersPrefix), "/"+iamIdentityFile)
 		item.Type = madmin.SRIAMItemIAMUser
 		item.IAMUser = &madmin.SRIAMUser{AccessKey: name, IsDeleteReq: true}
+	case strings.HasPrefix(path, iamConfigSTSPrefix):
+		if !r.ExpiresAt.IsZero() && UTCNow().After(r.ExpiresAt) {
+			return item, false
+		}
+		item.Type = iamSTSDeleteType
+		item.Name = strings.TrimSuffix(strings.TrimPrefix(path, iamConfigSTSPrefix), "/"+iamIdentityFile)
 	case strings.HasPrefix(path, iamConfigServiceAccountsPrefix):
 		name := strings.TrimSuffix(strings.TrimPrefix(path, iamConfigServiceAccountsPrefix), "/"+iamIdentityFile)
 		if name == siteReplicatorSvcAcc || r.Credentials.ParentUser == globalActiveCred.AccessKey {
@@ -518,8 +538,7 @@ func iamDeletionItem(path string, r iamRevision) (item madmin.SRIAMItem, ok bool
 		item.Type = madmin.SRIAMItemPolicyMapping
 		item.PolicyMapping = &madmin.SRPolicyMapping{UserOrGroup: strings.TrimSuffix(name, ".json"), UserType: int(typ), IsGroup: prefix == "groups"}
 	default:
-		// Expired STS credentials are not replayed. Parent revocations and
-		// their retained timestamp reject delayed copies of derived tokens.
+		// Other config paths do not represent replicated IAM deletions.
 		return item, false
 	}
 	return item, true
@@ -527,6 +546,8 @@ func iamDeletionItem(path string, r iamRevision) (item madmin.SRIAMItem, ok bool

 func iamDeletionPath(item madmin.SRIAMItem) string {
 	switch item.Type {
+	case iamSTSDeleteType:
+		return getUserIdentityPath(item.Name, stsUser)
 	case madmin.SRIAMItemIAMUser:
 		if item.IAMUser != nil && item.IAMUser.IsDeleteReq {
 			return getUserIdentityPath(item.IAMUser.AccessKey, regUser)
@@ -582,6 +603,9 @@ func (c *SiteReplicationSys) healIAMDeletions(ctx context.Context) (err error) {
 			continue
 		}
 		out := iamReplicationItem{SRIAMItem: item}
+		if item.Type == iamSTSDeleteType {
+			out.STSExpiresAt = r.ExpiresAt
+		}
 		if item.Type == madmin.SRIAMItemIAMUser && !r.Deleted {
 			out.Type, out.IAMUser = iamUserBoundaryType, nil
 			out.UserRevocation = &iamUserBoundary{User: item.IAMUser.AccessKey, Before: r.RevokedBefore}
@@ -593,7 +617,7 @@ func (c *SiteReplicationSys) healIAMDeletions(ctx context.Context) (err error) {
 		byType[item.Type] = append(byType[item.Type], out)
 	}
 	var items []iamReplicationItem
-	for _, typ := range []string{madmin.SRIAMItemPolicyMapping, madmin.SRIAMItemIAMUser, madmin.SRIAMItemSvcAcc, madmin.SRIAMItemGroupInfo, madmin.SRIAMItemPolicy} {
+	for _, typ := range []string{madmin.SRIAMItemPolicyMapping, madmin.SRIAMItemIAMUser, madmin.SRIAMItemSvcAcc, iamSTSDeleteType, madmin.SRIAMItemGroupInfo, madmin.SRIAMItemPolicy} {
 		items = append(items, byType[typ]...)
 	}
 	if len(items) == 0 {
@@ -711,6 +711,11 @@ func (store *IAMStoreSys) LoadIAMCache(ctx context.Context, firstTime bool) erro
 		if err := store.loadUsers(ctx, svcUser, newCache.iamUsersMap); err != nil {
 			return err
 		}
+		// Rebuild STS revisions as well: a missed etcd watch event must not
+		// leave a cached token valid or hide its tombstone after a restart.
+		if err := store.loadUsers(ctx, stsUser, newCache.iamSTSAccountsMap); err != nil {
+			return err
+		}

 		newCache.buildUserGroupMemberships()
 	}
@@ -754,6 +759,14 @@ func (store *IAMStoreSys) LoadIAMCache(ctx context.Context, firstTime bool) erro
 		cache.updatedAt = time.Now()
 	}

+	// Revision loading is independent of cache replacement. Even if ongoing
+	// IAM writes defer the full swap, a missed STS notification must converge.
+	for key := range cache.iamSTSAccountsMap {
+		if store.revisionIndex().get(getUserIdentityPath(key, stsUser)).Deleted {
+			delete(cache.iamSTSAccountsMap, key)
+		}
+	}
+
 	return nil
 }

@@ -2184,45 +2197,68 @@ func (store *IAMStoreSys) SetTempUser(ctx context.Context, accessKey string, cre

 // RevokeTokens - revokes all temporary credentials, or those with matching type,
 // associated with the parent user.
-func (store *IAMStoreSys) RevokeTokens(ctx context.Context, parentUser string, tokenRevokeType string) error {
+func (store *IAMStoreSys) RevokeTokens(ctx context.Context, parentUser string, tokenRevokeType string) ([]string, error) {
 	if parentUser == "" {
-		return errInvalidArgument
+		return nil, errInvalidArgument
 	}
-
+	before := UTCNow()
 	cache := store.lock()
 	defer store.unlock()
-
 	secret, err := getTokenSigningKey()
 	if err != nil {
-		return err
+		return nil, err
 	}
-
-	var revoked bool
-	for _, ui := range cache.iamSTSAccountsMap {
-		if ui.Credentials.ParentUser != parentUser {
+	// Another node may have issued a token whose notification was lost.
+	// Enumerate durable identities, retaining cached entries for retrying a
+	// partially propagated revocation. Do not hold up cached authentication.
+	accounts := maps.Clone(cache.iamSTSAccountsMap)
+	if err := store.withIAMStorage(ctx, func(ctx context.Context) error {
+		return store.loadUsers(ctx, stsUser, accounts)
+	}); err != nil {
+		return nil, err
+	}
+	var revoked []string
+	var failures []error
+	for key, ui := range accounts {
+		if ui.Credentials.ParentUser != parentUser || ui.UpdatedAt.After(before) || ui.Credentials.IsExpired() {
 			continue
 		}
 		if tokenRevokeType != "" {
 			claims, err := getClaimsFromTokenWithSecret(ui.Credentials.SessionToken, secret)
 			if err != nil {
-				continue // skip if token is invalid
+				continue
 			}
-			// skip if token type is given and does not match
 			if v, _ := claims.Lookup(tokenRevokeTypeClaim); v != tokenRevokeType {
 				continue
 			}
 		}
-		if err := store.deleteUserIdentity(ctx, ui.Credentials.AccessKey, stsUser); err != nil {
-			return err
+		if err := store.deleteUserIdentity(ctx, key, stsUser); err != nil {
+			failures = append(failures, err)
+			continue
 		}
-		delete(cache.iamSTSAccountsMap, ui.Credentials.AccessKey)
-		revoked = true
+		delete(cache.iamSTSAccountsMap, key)
+		revoked = append(revoked, key)
 	}
-
-	if revoked {
+	if len(revoked) != 0 {
 		cache.updatedAt = time.Now()
 	}
+	return revoked, errors.Join(failures...)
+}

+// revokeTempUser applies an immutable token's replicated tombstone without
+// touching the parent, its policies, or other (including newer) tokens.
+func (store *IAMStoreSys) revokeTempUser(ctx context.Context, key string, expiresAt time.Time) error {
+	cache := store.lock()
+	defer store.unlock()
+	u := UserIdentity{Version: 1, Deleted: true, ExpiresAt: expiresAt}
+	if err := store.saveUserIdentity(ctx, key, stsUser, &u); err != nil {
+		return err
+	}
+	delete(cache.iamSTSAccountsMap, key)
+	cache.updatedAt = time.Now()
+	if store.group != nil {
+		store.group.Forget(key)
+	}
 	return nil
 }

@@ -0,0 +1,254 @@
+// Copyright (c) 2026 Feng Ruohang
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+package cmd
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/minio/madmin-go/v3"
+	"github.com/minio/minio/internal/auth"
+	"github.com/minio/minio/internal/grid"
+	xnet "github.com/pgsty/silo-pkg/v3/net"
+)
+
+func TestIAMSTSRevocationDurableScope(t *testing.T) {
+	for _, backend := range []string{"object", "etcd"} {
+		for _, notify := range []bool{false, true} {
+			t.Run(fmt.Sprintf("%s/notify=%t", backend, notify), func(t *testing.T) {
+				ctx, sys, obj := prepareIAMRevisionFixture(t, backend)
+				for _, parent := range []string{"sts-parent", "other-parent"} {
+					_, err := sys.CreateUser(ctx, parent, madmin.AddOrUpdateUserReq{SecretKey: "synthetic-parent-password", Status: madmin.AccountEnabled})
+					mustIAM(t, err)
+				}
+				sibling := &IAMStoreSys{IAMStorageAPI: newIAMObjectStore(obj, MinIOUsersSysType)}
+				if backend == "etcd" {
+					sibling.IAMStorageAPI = newIAMEtcdStore(sys.store.IAMStorageAPI.(*IAMEtcdStore).client, MinIOUsersSysType)
+				}
+				mustIAM(t, sibling.LoadIAMCache(ctx, false))
+				secret, err := getTokenSigningKey()
+				mustIAM(t, err)
+				issue := func(store *IAMStoreSys, parent, typ string) auth.Credentials {
+					cred, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: parent, tokenRevokeTypeClaim: typ}, secret)
+					mustIAM(t, err)
+					cred.ParentUser = parent
+					_, err = store.SetTempUser(ctx, cred.AccessKey, cred, "")
+					mustIAM(t, err)
+					return cred
+				}
+				first := issue(sys.store, "sts-parent", "selected")
+				uncached := issue(sibling, "sts-parent", "selected")
+				otherType := issue(sys.store, "sts-parent", "other-type")
+				otherUser := issue(sys.store, "other-parent", "selected")
+				if _, ok := sys.store.GetUser(uncached.AccessKey); ok {
+					t.Fatal("fixture cached the remote token")
+				}
+				for _, c := range []auth.Credentials{first, otherType, otherUser} {
+					mustIAM(t, sibling.UserNotificationHandler(ctx, c.AccessKey, stsUser))
+				}
+				prior := globalNotificationSys
+				globalNotificationSys = &NotificationSys{}
+				t.Cleanup(func() { globalNotificationSys = prior })
+				if notify {
+					tg, err := grid.SetupTestGrid(2)
+					mustIAM(t, err)
+					t.Cleanup(tg.Cleanup)
+					mustIAM(t, loadUserRPC.Register(tg.Managers[1], func(r *grid.MSS) (grid.NoPayload, *grid.RemoteErr) {
+						if err := sibling.UserNotificationHandler(ctx, r.Get(peerRESTUser), stsUser); err != nil {
+							return grid.NoPayload{}, grid.NewRemoteErr(err)
+						}
+						return grid.NoPayload{}, nil
+					}))
+					host, err := xnet.ParseHost(strings.TrimPrefix(tg.Hosts[1], "http://"))
+					mustIAM(t, err)
+					globalNotificationSys = &NotificationSys{peerClients: []*peerRESTClient{{host: host, gridConn: func() *grid.Connection { return tg.Managers[0].Connection(tg.Hosts[1]) }}}}
+				}
+				mustIAM(t, sys.RevokeTokens(ctx, "sts-parent", "selected"))
+				for _, c := range []auth.Credentials{first, uncached} {
+					r, err := loadIAMRevision(ctx, sys.store, getUserIdentityPath(c.AccessKey, stsUser))
+					mustIAM(t, err)
+					if !r.Deleted {
+						t.Errorf("durable token was omitted: %s", c.AccessKey)
+					}
+					if _, ok := iamDeletionItem(getUserIdentityPath(c.AccessKey, stsUser), r); !ok {
+						t.Error("STS revocation cannot be replayed")
+					}
+					if notify && backend == "object" {
+						if _, ok := sibling.GetUser(c.AccessKey); ok {
+							t.Error("notified sibling accepted revoked token")
+						}
+					}
+				}
+				fresh := issue(sibling, "sts-parent", "selected")
+				// A missed notification must converge through the normal cache refresh.
+				mustIAM(t, sibling.LoadIAMCache(ctx, false))
+				for _, c := range []auth.Credentials{first, uncached} {
+					if _, ok := sibling.GetUser(c.AccessKey); ok {
+						t.Error("refresh retained revoked token")
+					}
+				}
+				for _, c := range []auth.Credentials{otherType, otherUser, fresh} {
+					mustIAM(t, sibling.UserNotificationHandler(ctx, c.AccessKey, stsUser))
+					if _, ok := sibling.GetUser(c.AccessKey); !ok {
+						t.Error("revocation removed an out-of-scope or new token")
+					}
+				}
+				if _, ok := sibling.GetUser("sts-parent"); !ok {
+					t.Error("revocation removed parent")
+				}
+			})
+		}
+	}
+}
+
+// Deletion is keyed by the immutable STS access key, so delayed delivery cannot
+// affect a newly issued token and a delayed creation cannot undo revocation.
+func TestIAMSTSRevocationReorderedReplay(t *testing.T) {
+	for _, backend := range []string{"object", "etcd"} {
+		for _, missing := range []bool{false, true} {
+			t.Run(fmt.Sprintf("%s/missing=%t", backend, missing), func(t *testing.T) {
+				ctx, sys, _ := prepareIAMRevisionFixture(t, backend)
+				const parent = "sts-replay-parent"
+				_, err := sys.CreateUser(ctx, parent, madmin.AddOrUpdateUserReq{SecretKey: "synthetic-parent-password", Status: madmin.AccountEnabled})
+				mustIAM(t, err)
+				secret, err := getTokenSigningKey()
+				mustIAM(t, err)
+				cred, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: parent}, secret)
+				mustIAM(t, err)
+				cred.ParentUser = parent
+				created := UTCNow().Add(-time.Minute)
+				peer := madmin.SRSTSCredential{AccessKey: cred.AccessKey, SecretKey: cred.SecretKey, SessionToken: cred.SessionToken, ParentUser: parent}
+				if !missing {
+					mustIAM(t, globalSiteReplicationSys.PeerSTSAccHandler(ctx, &peer, created))
+				}
+				fresh, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: parent}, secret)
+				mustIAM(t, err)
+				fresh.ParentUser = parent
+				_, err = sys.SetTempUser(ctx, fresh.AccessKey, fresh, "")
+				mustIAM(t, err)
+				// Older than the live record on this peer: deletion must still win.
+				item := iamReplicationItem{SRIAMItem: madmin.SRIAMItem{Type: iamSTSDeleteType, Name: cred.AccessKey, UpdatedAt: created.Add(-time.Second)}, STSExpiresAt: cred.Expiration.Add(globalMaxSkewTime)}
+				for range 2 {
+					mustIAM(t, applyIAMReplicationItem(ctx, item))
+					mustIAM(t, globalSiteReplicationSys.PeerSTSAccHandler(ctx, &peer, UTCNow()))
+					mustIAM(t, sys.store.LoadIAMCache(ctx, false))
+					if _, ok := sys.GetUser(ctx, cred.AccessKey); ok {
+						t.Fatal("delayed STS creation undid revocation")
+					}
+					if _, ok := sys.GetUser(ctx, fresh.AccessKey); !ok {
+						t.Fatal("delayed revocation removed new token")
+					}
+				}
+				r, err := loadIAMRevision(ctx, sys.store, getUserIdentityPath(cred.AccessKey, stsUser))
+				mustIAM(t, err)
+				if !r.Deleted || !r.ExpiresAt.Equal(item.STSExpiresAt) || r.Credentials.SessionToken != "" || r.Credentials.SecretKey != "" {
+					t.Fatal("revocation lost expiry or retained secret material")
+				}
+				for _, status := range []madmin.AccountStatus{madmin.AccountDisabled, madmin.AccountEnabled} {
+					_, err = sys.SetUserStatus(ctx, parent, status)
+					mustIAM(t, err)
+				}
+				if _, ok := sys.GetUser(ctx, cred.AccessKey); ok {
+					t.Fatal("parent re-enable revived revoked token")
+				}
+			})
+		}
+	}
+}
+
+// Pause the durable scan while another node persists a new credential. The
+// revocation's original cutoff must survive the scan and leave that token live.
+type iamSTSScanGateStore struct {
+	IAMStorageAPI
+	entered chan struct{}
+	resume  chan struct{}
+}
+
+func (s *iamSTSScanGateStore) loadUsers(ctx context.Context, typ IAMUserType, users map[string]UserIdentity) error {
+	close(s.entered)
+	select {
+	case <-s.resume:
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+	return s.IAMStorageAPI.loadUsers(ctx, typ, users)
+}
+
+func TestIAMSTSRevocationConcurrentIssuance(t *testing.T) {
+	ctx, sys, obj := prepareIAMRevisionFixture(t)
+	_, err := sys.CreateUser(ctx, "sts-race-parent", madmin.AddOrUpdateUserReq{SecretKey: "synthetic-parent-password", Status: madmin.AccountEnabled})
+	mustIAM(t, err)
+	sibling := &IAMStoreSys{IAMStorageAPI: newIAMObjectStore(obj, MinIOUsersSysType)}
+	secret, err := getTokenSigningKey()
+	mustIAM(t, err)
+	issue := func() auth.Credentials {
+		c, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: "sts-race-parent"}, secret)
+		mustIAM(t, err)
+		c.ParentUser = "sts-race-parent"
+		_, err = sibling.SetTempUser(ctx, c.AccessKey, c, "")
+		mustIAM(t, err)
+		return c
+	}
+	old := issue()
+	gate := &iamSTSScanGateStore{IAMStorageAPI: sys.store.IAMStorageAPI, entered: make(chan struct{}), resume: make(chan struct{})}
+	sys.store.IAMStorageAPI = gate
+	done := make(chan error, 1)
+	go func() { done <- sys.RevokeTokens(ctx, "sts-race-parent", "") }()
+	<-gate.entered
+	fresh := issue()
+	close(gate.resume)
+	mustIAM(t, <-done)
+	mustIAM(t, sibling.UserNotificationHandler(ctx, old.AccessKey, stsUser))
+	if _, ok := sibling.GetUser(old.AccessKey); ok {
+		t.Fatal("old uncached token survived")
+	}
+	mustIAM(t, sibling.UserNotificationHandler(ctx, fresh.AccessKey, stsUser))
+	if _, ok := sibling.GetUser(fresh.AccessKey); !ok {
+		t.Fatal("new token was included in earlier revocation")
+	}
+}
+
+type iamSTSFailureStore struct {
+	IAMStorageAPI
+	failedPath string
+}
+
+func (s *iamSTSFailureStore) saveIAMConfig(ctx context.Context, item any, path string, opts ...options) error {
+	if path == s.failedPath {
+		return errIAMInjectedWrite
+	}
+	return s.IAMStorageAPI.saveIAMConfig(ctx, item, path, opts...)
+}
+
+func TestIAMSTSRevocationPartialFailure(t *testing.T) {
+	ctx, sys, _ := prepareIAMRevisionFixture(t)
+	_, err := sys.CreateUser(ctx, "sts-partial-parent", madmin.AddOrUpdateUserReq{SecretKey: "synthetic-parent-password", Status: madmin.AccountEnabled})
+	mustIAM(t, err)
+	secret, err := getTokenSigningKey()
+	mustIAM(t, err)
+	var keys []string
+	for range 2 {
+		c, err := auth.GetNewCredentialsWithMetadata(map[string]any{"exp": UTCNow().Add(time.Hour).Unix(), parentClaim: "sts-partial-parent"}, secret)
+		mustIAM(t, err)
+		c.ParentUser = "sts-partial-parent"
+		_, err = sys.SetTempUser(ctx, c.AccessKey, c, "")
+		mustIAM(t, err)
+		keys = append(keys, c.AccessKey)
+	}
+	sys.store.IAMStorageAPI = &iamSTSFailureStore{IAMStorageAPI: sys.store.IAMStorageAPI, failedPath: getUserIdentityPath(keys[0], stsUser)}
+	revoked, err := sys.store.RevokeTokens(ctx, "sts-partial-parent", "")
+	if !errors.Is(err, errIAMInjectedWrite) || len(revoked) != 1 || revoked[0] != keys[1] {
+		t.Fatalf("committed keys or failure lost: %v %v", revoked, err)
+	}
+	if _, ok := sys.store.GetUser(keys[1]); ok {
+		t.Fatal("committed revocation retained cache entry")
+	}
+	if _, ok := sys.store.GetUser(keys[0]); !ok {
+		t.Fatal("failed revocation claimed cache success")
+	}
+}
@@ -717,7 +717,14 @@ func (sys *IAMSys) RevokeTokens(ctx context.Context, accessKey, tokenRevokeType
 		return errServerNotInitialized
 	}

-	return sys.store.RevokeTokens(ctx, accessKey, tokenRevokeType)
+	keys, err := sys.store.RevokeTokens(ctx, accessKey, tokenRevokeType)
+	// Publish every committed tombstone even if another token failed to save.
+	for _, key := range keys {
+		sys.notifyForUser(ctx, key, true)
+		// Offline peers recover this same record through IAM deletion healing.
+		iamLogIf(ctx, globalSiteReplicationSys.IAMChangeHook(ctx, madmin.SRIAMItem{Type: iamSTSDeleteType, Name: key}))
+	}
+	return err
 }

 // DeleteUser - delete user (only for long-term users not STS users).