staging/lustre/ldlm: Fix flock deadlock detection race

Deadlock isn't detected if 2 threads are trying to
grant 2 locks which deadlock on each other.
They call ldlm_flock_deadlock() simultaneously
and deadlock ins't detected.

The soulition is to add lock to blocking list before
calling ldlm_flock_deadlock()

Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-1602
Lustre-change: http://review.whamcloud.com/3277Signed-off-by: Andriy Skulysh <Andriy_Skulysh@xyratex.com>
Reviewed-by: Vitaly Fertman <vitaly_fertman@xyratex.com>
Reviewed-by: Bruce Korb <bruce_korb@xyratex.com>
Reviewed-by: Keith Mannthey <keith.mannthey@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: Peng Tao <tao.peng@emc.com>
Signed-off-by: Andreas Dilger <andreas.dilger@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/staging/lustre/lustre/ldlm/ldlm_flock.c

@@ -198,6 +198,7 @@ ldlm_flock_deadlock(struct ldlm_lock *req, struct ldlm_lock *bl_lock)
 		if (lock == NULL)
 			break;
 
+		LASSERT(req != lock);
 		flock = &lock->l_policy_data.l_flock;
 		LASSERT(flock->owner == bl_owner);
 		bl_owner = flock->blocking_owner;
@@ -329,18 +330,21 @@ ldlm_process_flock_lock(struct ldlm_lock *req, __u64 *flags, int first_enq,
 				RETURN(LDLM_ITER_STOP);
 			}
 
-			if (ldlm_flock_deadlock(req, lock)) {
-				ldlm_flock_destroy(req, mode, *flags);
-				*err = -EDEADLK;
-				RETURN(LDLM_ITER_STOP);
-			}
-
+			/* add lock to blocking list before deadlock
+			 * check to prevent race */
 			rc = ldlm_flock_blocking_link(req, lock);
 			if (rc) {
 				ldlm_flock_destroy(req, mode, *flags);
 				*err = rc;
 				RETURN(LDLM_ITER_STOP);
 			}
+			if (ldlm_flock_deadlock(req, lock)) {
+				ldlm_flock_blocking_unlink(req);
+				ldlm_flock_destroy(req, mode, *flags);
+				*err = -EDEADLK;
+				RETURN(LDLM_ITER_STOP);
+			}
+
 			ldlm_resource_add_lock(res, &res->lr_waiting, req);
 			*flags |= LDLM_FL_BLOCK_GRANTED;
 			RETURN(LDLM_ITER_STOP);