Hi
Please, apply this before 2.6.32, there is a possibility for regression
(locking in snapshot_status was introduced in 2.6.32-rc).
Mikulas
---
Lock only when querying status, not table
LVM sometimes suspends the underlying device and asks for a table.
To prevent a possible deadlock (LVM suspends the snapshot store, snapshot
takes the lock and waits for I/O, LVM asks for table before unsuspend,
deadlock), lock only when querying the status. The lock is not needed for
table anyway, because the table doesn't change during snapshot operation.
Signed-off-by: Mikulas Patocka <mpatocka@xxxxxxxxxx>
---
drivers/md/dm-snap.c | 10 ++++++----
1 file changed, 6 insertions(+), 4 deletions(-)
Index: linux-2.6.32-rc8-devel/drivers/md/dm-snap.c
===================================================================
--- linux-2.6.32-rc8-devel.orig/drivers/md/dm-snap.c 2009-11-30 17:56:06.000000000 +0100
+++ linux-2.6.32-rc8-devel/drivers/md/dm-snap.c 2009-11-30 18:00:44.000000000 +0100
@@ -1407,10 +1407,11 @@ static int snapshot_status(struct dm_tar
unsigned sz = 0;
struct dm_snapshot *snap = ti->private;
- down_write(&snap->lock);
-
switch (type) {
case STATUSTYPE_INFO:
+
+ down_write(&snap->lock);
+
if (!snap->valid)
DMEMIT("Invalid");
else {
@@ -1429,6 +1430,9 @@ static int snapshot_status(struct dm_tar
else
DMEMIT("Unknown");
}
+
+ up_write(&snap->lock);
+
break;
case STATUSTYPE_TABLE:
@@ -1443,8 +1447,6 @@ static int snapshot_status(struct dm_tar
break;
}
- up_write(&snap->lock);
-
return 0;
}
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
