Hi,
Excuse me for the long Cc list.
I kept people in the thread below in Cc.
http://marc.theaimsgroup.com/?l=linux-kernel&m=116221664710826&w=2
The patch below fixes bd_claim_by_kobject to release bdev correctly
in case that bd_claim succeeds but following add_bd_holder fails.
If it happens, the caller takes it didn't bd_claim() where actually
it did. The bdev becomes no longer bd_claim-able from others.
I don't have any reproducible test case for it but it's a bug.
The bug is introduced in 2.6.18-rc1-mm2 and now in 2.6.19-rc3.
Please consider to apply.
Thanks,
--
Jun'ichi Nomura, NEC Corporation of America
The patch below fixes bd_claim_by_kobject to release bdev correctly
in case that bd_claim succeeds but following add_bd_holder fails.
fs/block_dev.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
Signed-off-by: Jun'ichi Nomura <j-nomura@xxxxxxxxxxxxx>
--- linux-2.6.orig/fs/block_dev.c 2006-10-30 10:07:16.000000000 -0500
+++ linux-2.6/fs/block_dev.c 2006-10-30 15:38:21.000000000 -0500
@@ -751,8 +751,11 @@ static int bd_claim_by_kobject(struct bl
mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION);
res = bd_claim(bdev, holder);
- if (res == 0)
+ if (res == 0) {
res = add_bd_holder(bdev, bo);
+ if (res)
+ bd_release(bdev);
+ }
if (res)
free_bd_holder(bo);
mutex_unlock(&bdev->bd_mutex);
