nfsd: Fix race between FREE_STATEID and LOCK
authorChuck Lever <chuck.lever@oracle.com>
Thu, 11 Aug 2016 14:37:30 +0000 (10:37 -0400)
committerJ. Bruce Fields <bfields@redhat.com>
Thu, 11 Aug 2016 19:08:39 +0000 (15:08 -0400)
When running LTP's nfslock01 test, the Linux client can send a LOCK
and a FREE_STATEID request at the same time. The outcome is:

Frame 324    R OPEN stateid [2,O]

Frame 115004 C LOCK lockowner_is_new stateid [2,O] offset 672000 len 64
Frame 115008 R LOCK stateid [1,L]
Frame 115012 C WRITE stateid [0,L] offset 672000 len 64
Frame 115016 R WRITE NFS4_OK
Frame 115019 C LOCKU stateid [1,L] offset 672000 len 64
Frame 115022 R LOCKU NFS4_OK
Frame 115025 C FREE_STATEID stateid [2,L]
Frame 115026 C LOCK lockowner_is_new stateid [2,O] offset 672128 len 64
Frame 115029 R FREE_STATEID NFS4_OK
Frame 115030 R LOCK stateid [3,L]
Frame 115034 C WRITE stateid [0,L] offset 672128 len 64
Frame 115038 R WRITE NFS4ERR_BAD_STATEID

In other words, the server returns stateid L in a successful LOCK
reply, but it has already released it. Subsequent uses of stateid L
fail.

To address this, protect the generation check in nfsd4_free_stateid
with the st_mutex. This should guarantee that only one of two
outcomes occurs: either LOCK returns a fresh valid stateid, or
FREE_STATEID returns NFS4ERR_LOCKS_HELD.

Reported-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Fix-suggested-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Tested-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Cc: stable@vger.kernel.org
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
fs/nfsd/nfs4state.c

index 8410ca275db1aecf0a1f8a022b92cdc1597ff258..0edc1822f1444a0243b5aac9037f1175f968f6cb 100644 (file)
@@ -4903,6 +4903,32 @@ nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        return nfs_ok;
 }
 
+static __be32
+nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
+{
+       struct nfs4_ol_stateid *stp = openlockstateid(s);
+       __be32 ret;
+
+       mutex_lock(&stp->st_mutex);
+
+       ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
+       if (ret)
+               goto out;
+
+       ret = nfserr_locks_held;
+       if (check_for_locks(stp->st_stid.sc_file,
+                           lockowner(stp->st_stateowner)))
+               goto out;
+
+       release_lock_stateid(stp);
+       ret = nfs_ok;
+
+out:
+       mutex_unlock(&stp->st_mutex);
+       nfs4_put_stid(s);
+       return ret;
+}
+
 __be32
 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                   struct nfsd4_free_stateid *free_stateid)
@@ -4910,7 +4936,6 @@ nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        stateid_t *stateid = &free_stateid->fr_stateid;
        struct nfs4_stid *s;
        struct nfs4_delegation *dp;
-       struct nfs4_ol_stateid *stp;
        struct nfs4_client *cl = cstate->session->se_client;
        __be32 ret = nfserr_bad_stateid;
 
@@ -4929,18 +4954,9 @@ nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                ret = nfserr_locks_held;
                break;
        case NFS4_LOCK_STID:
-               ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
-               if (ret)
-                       break;
-               stp = openlockstateid(s);
-               ret = nfserr_locks_held;
-               if (check_for_locks(stp->st_stid.sc_file,
-                                   lockowner(stp->st_stateowner)))
-                       break;
-               WARN_ON(!unhash_lock_stateid(stp));
+               atomic_inc(&s->sc_count);
                spin_unlock(&cl->cl_lock);
-               nfs4_put_stid(s);
-               ret = nfs_ok;
+               ret = nfsd4_free_lock_stateid(stateid, s);
                goto out;
        case NFS4_REVOKED_DELEG_STID:
                dp = delegstateid(s);