[Ocfs2-devel] [PATCH] ocfs2/dlm: ignore cleaning the migration mle that is inuse

[xuejiufei <xuejiufei@huawei.com>]

We have found that migration source will trigger a BUG that the
refcount of mle is already zero before put when the target is
down during migration. The situation is as follows:

dlm_migrate_lockres
  dlm_add_migration_mle
  dlm_mark_lockres_migrating
  dlm_get_mle_inuse
  <<<<<< Now the refcount of the mle is 2.
  dlm_send_one_lockres and wait for the target to become the
  new master.
  <<<<<< o2hb detect the target down and clean the migration
  mle. Now the refcount is 1.

dlm_migrate_lockres woken, and put the mle twice when found
the target goes down which trigger the BUG with the following
message:
"ERROR: bad mle: ".

Signed-off-by: Jiufei Xue <xuejiufei@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
---
 fs/ocfs2/dlm/dlmmaster.c | 26 +++++++++++++++-----------
 1 file changed, 15 insertions(+), 11 deletions(-)

diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index 936e11b..b713140 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -2519,6 +2519,11 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm,
 	spin_lock(&dlm->master_lock);
 	ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name,
 				    namelen, target, dlm->node_num);
+	/* get an extra reference on the mle.
+	 * otherwise the assert_master from the new
+	 * master will destroy this.
+	 */
+	dlm_get_mle_inuse(mle);
 	spin_unlock(&dlm->master_lock);
 	spin_unlock(&dlm->spinlock);

@@ -2554,6 +2559,7 @@ fail:
 		if (mle_added) {
 			dlm_mle_detach_hb_events(dlm, mle);
 			dlm_put_mle(mle);
+			dlm_put_mle_inuse(mle);
 		} else if (mle) {
 			kmem_cache_free(dlm_mle_cache, mle);
 			mle = NULL;
@@ -2571,17 +2577,6 @@ fail:
 	 * ensure that all assert_master work is flushed. */
 	flush_workqueue(dlm->dlm_worker);

-	/* get an extra reference on the mle.
-	 * otherwise the assert_master from the new
-	 * master will destroy this.
-	 * also, make sure that all callers of dlm_get_mle
-	 * take both dlm->spinlock and dlm->master_lock */
-	spin_lock(&dlm->spinlock);
-	spin_lock(&dlm->master_lock);
-	dlm_get_mle_inuse(mle);
-	spin_unlock(&dlm->master_lock);
-	spin_unlock(&dlm->spinlock);
-
 	/* notify new node and send all lock state */
 	/* call send_one_lockres with migration flag.
 	 * this serves as notice to the target node that a
@@ -3312,6 +3307,15 @@ top:
 			    mle->new_master != dead_node)
 				continue;

+			if (mle->new_master == dead_node && mle->inuse) {
+				mlog(ML_NOTICE, "%s: target %u died during "
+						"migration from %u, the MLE is "
+						"still keep used, ignore it!\n",
+						dlm->name, dead_node,
+						mle->master);
+				continue;
+			}
+
 			/* If we have reached this point, this mle needs to be
 			 * removed from the list and freed. */
 			dlm_clean_migration_mle(dlm, mle);
-- 
1.8.4.3