[Cluster-devel] [PATCH] dlm: Avoid that dlm_release_lockspace() incorrectly returns -EBUSY

[Bart Van Assche <bvanassche@acm.org>]

When dlm_release_lockspace(ls, 1) is invoked on a busy system
immediately after the last dlm_unlock() AST has finished it can occur
that lkb_idr_is_local() is invoked for the unlocked LKB since removal
from ls_lkbidr only occurs after the AST has returned. If that happens
dlm_release_lockspace(ls, 1) will return -EBUSY instead of releasing
the lockspace. Fix this race condition by changing lkb_idr_is_local()
such that it only returns true for LKB's that have not yet been
unlocked.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Cc: David Teigland <teigland@redhat.com>
---
 fs/dlm/lockspace.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/fs/dlm/lockspace.c b/fs/dlm/lockspace.c
index 88556dc..d5abafd 100644
--- a/fs/dlm/lockspace.c
+++ b/fs/dlm/lockspace.c
@@ -706,9 +706,7 @@ static int lkb_idr_is_local(int id, void *p, void *data)
 {
 	struct dlm_lkb *lkb = p;
 
-	if (!lkb->lkb_nodeid)
-		return 1;
-	return 0;
+	return lkb->lkb_nodeid == 0 && lkb->lkb_grmode != DLM_LOCK_IV;
 }
 
 static int lkb_idr_is_any(int id, void *p, void *data)
-- 
1.8.1.4