In osc_ldlm_completion_ast(), it tries to hold mutex of cl_lock
and update the lock even it's known that the lock is not granted,
this can introduce a deadlock case if the process is matching a
dlm lock with the calling stack:
osc_lock_enqueue()
-> osc_enqueue_base()
-> ldlm_lock_match()
-> osc_ldlm_completion_ast()
-> cl_lock_mutex_get()
The enqueuing lock and the lock for completion_ast() are different
lock. This is a deadlock case because we're holding a lock's mutex
and then request the other.
It's not necessary to acquire the lock in completion_ast() if it's
known that the lock has not been granted.
Signed-off-by: Jinshan Xiong <jinshan.xiong@intel.com>
Change-Id: Ia8c0a4cb1e48c9f7fd921052174ce4d8d07713b1
Reviewed-on: http://review.whamcloud.com/10581
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Bobi Jam <bobijam@gmail.com>
Reviewed-by: Patrick Farrell <paf@cray.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
int result;
int dlmrc;
- /* first, do dlm part of the work */
- dlmrc = ldlm_completion_ast_async(dlmlock, flags, data);
+ /* first, do dlm part of the work */
+ dlmrc = ldlm_completion_ast_async(dlmlock, flags, data);
+ if (flags == LDLM_FL_WAIT_NOREPROC)
+ return dlmrc;
+
/* then, notify cl_lock */
env = cl_env_nested_get(&nest);
if (!IS_ERR(env)) {