ocfs2/dlm: do not get resource spinlock if lockres is new
PATCH 1/1 ocfs2/dlm: do not get resource spinlock if lockres is new
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 <[email protected]>
Reviewed-by joyce.xue <[email protected]>
--- fs/ocfs2/dlm/dlmmaster.c | 18 ++++++++++-------- 1 file changed, 10 insertions(+), 8 deletions(-)
fs/ocfs2/dlm/dlmmaster.c +10 -8
@@ -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
Signature · 1 line
1.8.4.3