nfsd: only set file_lock.fl_lmops in nfsd4_lockt if a stateowner is found
authorJeff Layton <jlayton@redhat.com>
Thu, 22 Jan 2009 19:16:04 +0000 (14:16 -0500)
committerJ. Bruce Fields <bfields@pig.fieldses.org>
Tue, 27 Jan 2009 22:26:59 +0000 (17:26 -0500)
nfsd4_lockt does a search for a lockstateowner when building the lock
struct to test. If one is found, it'll set fl_owner to it. Regardless of
whether that happens, it'll also set fl_lmops. Given that this lock is
basically a "lightweight" lock that's just used for checking conflicts,
setting fl_lmops is probably not appropriate for it.

This behavior exposed a bug in DLM's GETLK implementation where it
wasn't clearing out the fields in the file_lock before filling in
conflicting lock info. While we were able to fix this in DLM, it
still seems pointless and dangerous to set the fl_lmops this way
when we may have a NULL lockstateowner.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: J. Bruce Fields <bfields@pig.fieldses.org>
fs/nfsd/nfs4state.c

index 88db7d3ec120dea33140e757a5ee30601baf220e..b6f60f48e94b4038473faf04b766ce706e60a9d4 100644 (file)
@@ -2871,7 +2871,6 @@ nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
        file_lock.fl_pid = current->tgid;
        file_lock.fl_flags = FL_POSIX;
-       file_lock.fl_lmops = &nfsd_posix_mng_ops;
 
        file_lock.fl_start = lockt->lt_offset;
        file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);