[Ocfs2-devel] [PATCH] ocfs2/dlm: do not get resource spinlock if lockres is new

[Joseph Qi]

There is a deadlock case which reported by Guozhonghua
(guozhonghua at h3c.com):
https://oss.oracle.com/pipermail/ocfs2-devel/2014-September/010079.html

This case is caused by &res->spinlock and &dlm->master_lock
misordering in different threads. It was introduced by commit
8d400b81cc83 (ocfs2/dlm: Clean up refmap helpers).
Since lockres is new, it doesn't not require the &res->spinlock. So
remove it.

Signed-off-by: Joseph Qi <joseph.qi at huawei.com>
Reviewed-by: joyce.xue <xuejiufei at huawei.com>
---
 fs/ocfs2/dlm/dlmmaster.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)

diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index 3ec906e..e3cfa02 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -655,12 +655,9 @@ void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm,
 	clear_bit(bit, res->refmap);
 }

-
-void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
+static void __dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
 				   struct dlm_lock_resource *res)
 {
-	assert_spin_locked(&res->spinlock);
-
 	res->inflight_locks++;

 	mlog(0, "%s: res %.*s, inflight++: now %u, %ps()\n", dlm->name,
@@ -668,6 +665,13 @@ void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
 	     __builtin_return_address(0));
 }

+void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
+				   struct dlm_lock_resource *res)
+{
+	assert_spin_locked(&res->spinlock);
+	__dlm_lockres_grab_inflight_ref(dlm, res);
+}
+
 void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
 				   struct dlm_lock_resource *res)
 {
@@ -894,10 +898,8 @@ lookup:
 	/* finally add the lockres to its hash bucket */
 	__dlm_insert_lockres(dlm, res);

-	/* Grab inflight ref to pin the resource */
-	spin_lock(&res->spinlock);
-	dlm_lockres_grab_inflight_ref(dlm, res);
-	spin_unlock(&res->spinlock);
+	/* since this lockres is new it doesn't not require the spinlock */
+	__dlm_lockres_grab_inflight_ref(dlm, res);

 	/* get an extra ref on the mle in case this is a BLOCK
 	 * if so, the creator of the BLOCK may try to put the last
-- 
1.8.4.3