lkml.org 
[lkml]   [2013]   [Nov]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH 12/26] staging/lustre/ldlm: Fix flock detection for different mounts
Date
From: Andriy Skulysh <Andriy_Skulysh@xyratex.com>

Deadlock can happen when 2 processes take concurrent locks
on files situated in different mountpoints.

Modify flock detection algorithm to distinguish process by
pair PID+NID instead of PID+export.
It is done by searching for a blocking owner in all OBD's
exports with the same NID.

Lustre-change: http://review.whamcloud.com/3276
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-1601
Signed-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: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: Peng Tao <bergwolf@gmail.com>
Signed-off-by: Andreas Dilger <andreas.dilger@intel.com>
---
drivers/staging/lustre/lustre/ldlm/ldlm_flock.c | 45 +++++++++++++++++++++--
1 file changed, 41 insertions(+), 4 deletions(-)

diff --git a/drivers/staging/lustre/lustre/ldlm/ldlm_flock.c b/drivers/staging/lustre/lustre/ldlm/ldlm_flock.c
index 37ebd2a..396e58b 100644
--- a/drivers/staging/lustre/lustre/ldlm/ldlm_flock.c
+++ b/drivers/staging/lustre/lustre/ldlm/ldlm_flock.c
@@ -161,6 +161,31 @@ ldlm_flock_destroy(struct ldlm_lock *lock, ldlm_mode_t mode, __u64 flags)
* one client holds a lock on something and want a lock on something
* else and at the same time another client has the opposite situation).
*/
+
+struct ldlm_flock_lookup_cb_data {
+ __u64 *bl_owner;
+ struct ldlm_lock *lock;
+ struct obd_export *exp;
+};
+
+static int ldlm_flock_lookup_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd,
+ struct hlist_node *hnode, void *data)
+{
+ struct ldlm_flock_lookup_cb_data *cb_data = data;
+ struct obd_export *exp = cfs_hash_object(hs, hnode);
+ struct ldlm_lock *lock;
+
+ lock = cfs_hash_lookup(exp->exp_flock_hash, cb_data->bl_owner);
+ if (lock == NULL)
+ return 0;
+
+ /* Stop on first found lock. Same process can't sleep twice */
+ cb_data->lock = lock;
+ cb_data->exp = class_export_get(exp);
+
+ return 1;
+}
+
static int
ldlm_flock_deadlock(struct ldlm_lock *req, struct ldlm_lock *bl_lock)
{
@@ -175,16 +200,26 @@ ldlm_flock_deadlock(struct ldlm_lock *req, struct ldlm_lock *bl_lock)

class_export_get(bl_exp);
while (1) {
+ struct ldlm_flock_lookup_cb_data cb_data = {
+ .bl_owner = &bl_owner,
+ .lock = NULL,
+ .exp = NULL };
struct obd_export *bl_exp_new;
struct ldlm_lock *lock = NULL;
struct ldlm_flock *flock;

- if (bl_exp->exp_flock_hash != NULL)
- lock = cfs_hash_lookup(bl_exp->exp_flock_hash,
- &bl_owner);
+ if (bl_exp->exp_flock_hash != NULL) {
+ cfs_hash_for_each_key(bl_exp->exp_obd->obd_nid_hash,
+ &bl_exp->exp_connection->c_peer.nid,
+ ldlm_flock_lookup_cb, &cb_data);
+ lock = cb_data.lock;
+ }
if (lock == NULL)
break;

+ class_export_put(bl_exp);
+ bl_exp = cb_data.exp;
+
LASSERT(req != lock);
flock = &lock->l_policy_data.l_flock;
LASSERT(flock->owner == bl_owner);
@@ -198,7 +233,9 @@ ldlm_flock_deadlock(struct ldlm_lock *req, struct ldlm_lock *bl_lock)
if (bl_exp->exp_failed)
break;

- if (bl_owner == req_owner && bl_exp == req_exp) {
+ if (bl_owner == req_owner &&
+ (bl_exp->exp_connection->c_peer.nid ==
+ req_exp->exp_connection->c_peer.nid)) {
class_export_put(bl_exp);
return 1;
}
--
1.7.9.5


\
 
 \ /
  Last update: 2013-11-14 20:41    [W:0.216 / U:0.568 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site