nfsd4: release lockowners on last unlock in 4.1 case
J. Bruce Fields [Sun, 7 Apr 2013 17:28:16 +0000 (13:28 -0400)]
In the 4.1 case we're supposed to release lockowners as soon as they're
no longer used.

It would probably be more efficient to reference count them, but that's
slightly fiddly due to the need to have callbacks from locks.c to take
into account lock merging and splitting.

For most cases just scanning the inode's lock list on unlock for
matching locks will be sufficient.

Signed-off-by: J. Bruce Fields <bfields@redhat.com>

fs/nfsd/nfs4state.c

index 16db25d..ff1577d 100644 (file)
@@ -4387,6 +4387,7 @@ __be32
 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
            struct nfsd4_locku *locku)
 {
+       struct nfs4_lockowner *lo;
        struct nfs4_ol_stateid *stp;
        struct file *filp = NULL;
        struct file_lock *file_lock = NULL;
@@ -4419,9 +4420,10 @@ nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                status = nfserr_jukebox;
                goto out;
        }
+       lo = lockowner(stp->st_stateowner);
        locks_init_lock(file_lock);
        file_lock->fl_type = F_UNLCK;
-       file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
+       file_lock->fl_owner = (fl_owner_t)lo;
        file_lock->fl_pid = current->tgid;
        file_lock->fl_file = filp;
        file_lock->fl_flags = FL_POSIX;
@@ -4440,6 +4442,11 @@ nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        update_stateid(&stp->st_stid.sc_stateid);
        memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
 
+       if (nfsd4_has_session(cstate) && !check_for_locks(stp->st_file, lo)) {
+               WARN_ON_ONCE(cstate->replay_owner);
+               release_lockowner(lo);
+       }
+
 out:
        nfsd4_bump_seqid(cstate, status);
        if (!cstate->replay_owner)