Commit 1cd04dbe authored by Kurt Hackel's avatar Kurt Hackel Committed by Mark Fasheh

ocfs2_dlm: Flush dlm workqueue before starting to migrate

This is to prevent the condition in which a previously queued
up assert master asserts after we start the migration. Now
migration ensures the workqueue is flushed before proceeding
with migrating the lock to another node. This condition is
typically encountered during parallel umounts.
Signed-off-by: default avatarKurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: default avatarSunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: default avatarMark Fasheh <mark.fasheh@oracle.com>
parent e17e75ec
...@@ -1507,10 +1507,11 @@ int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data) ...@@ -1507,10 +1507,11 @@ int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data)
/* take care of the easy cases up front */ /* take care of the easy cases up front */
spin_lock(&res->spinlock); spin_lock(&res->spinlock);
if (res->state & DLM_LOCK_RES_RECOVERING) { if (res->state & (DLM_LOCK_RES_RECOVERING|
DLM_LOCK_RES_MIGRATING)) {
spin_unlock(&res->spinlock); spin_unlock(&res->spinlock);
mlog(0, "returning DLM_MASTER_RESP_ERROR since res is " mlog(0, "returning DLM_MASTER_RESP_ERROR since res is "
"being recovered\n"); "being recovered/migrated\n");
response = DLM_MASTER_RESP_ERROR; response = DLM_MASTER_RESP_ERROR;
if (mle) if (mle)
kmem_cache_free(dlm_mle_cache, mle); kmem_cache_free(dlm_mle_cache, mle);
...@@ -2493,6 +2494,9 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm, ...@@ -2493,6 +2494,9 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm,
* the lockres * the lockres
*/ */
/* now that remote nodes are spinning on the MIGRATING flag,
* ensure that all assert_master work is flushed. */
flush_workqueue(dlm->dlm_worker);
/* get an extra reference on the mle. /* get an extra reference on the mle.
* otherwise the assert_master from the new * otherwise the assert_master from the new
...@@ -2547,7 +2551,8 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm, ...@@ -2547,7 +2551,8 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm,
res->owner == target) res->owner == target)
break; break;
mlog(0, "timed out during migration\n"); mlog(0, "%s:%.*s: timed out during migration\n",
dlm->name, res->lockname.len, res->lockname.name);
/* avoid hang during shutdown when migrating lockres /* avoid hang during shutdown when migrating lockres
* to a node which also goes down */ * to a node which also goes down */
if (dlm_is_node_dead(dlm, target)) { if (dlm_is_node_dead(dlm, target)) {
...@@ -2555,10 +2560,7 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm, ...@@ -2555,10 +2560,7 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm,
"target %u is no longer up, restarting\n", "target %u is no longer up, restarting\n",
dlm->name, res->lockname.len, dlm->name, res->lockname.len,
res->lockname.name, target); res->lockname.name, target);
ret = -ERESTARTSYS; ret = -EINVAL;
}
}
if (ret == -ERESTARTSYS) {
/* migration failed, detach and clean up mle */ /* migration failed, detach and clean up mle */
dlm_mle_detach_hb_events(dlm, mle); dlm_mle_detach_hb_events(dlm, mle);
dlm_put_mle(mle); dlm_put_mle(mle);
...@@ -2568,7 +2570,9 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm, ...@@ -2568,7 +2570,9 @@ static int dlm_migrate_lockres(struct dlm_ctxt *dlm,
spin_unlock(&res->spinlock); spin_unlock(&res->spinlock);
goto leave; goto leave;
} }
/* TODO: if node died: stop, clean up, return error */ } else
mlog(0, "%s:%.*s: caught signal during migration\n",
dlm->name, res->lockname.len, res->lockname.name);
} }
/* all done, set the owner, clear the flag */ /* all done, set the owner, clear the flag */
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment