Re: Possible deadlock in jfs_imap.c?

[Dave Kleikamp <shaggy@linux.vnet.ibm.com>]

On Sat, 2009-05-23 at 17:19 +0200, Jan Kucera wrote:
> Hi,
> 
> for my master thesis, I've designed static checker for finding deadlocks in C.
> My tool found some possible deadlock in file fs/jfs/jfs_imap.c
> between JFS_IP(ipimap)->commit_mutex and  imap->im_aglock[agno].
> 
> Reason of this problem is probably missed unlocking of
> JFS_IP(ipimap)->commit_mutex in branch starting at line 2564 which
> leaves JFS_IP(ipimap)->commit_mutex locked:
> 
>     in function diNewIAG(..)
>         locking mutex_lock(&JFS_IP(ipimap)->commit_mutex) at line 2540
>         condition at line 2543 is false
>         condition at line 2564 is true
>         goto out: at 2579
>         IAGFREE_UNLOCK(imap); at line 2654
>     return from diNewIAG(..) at line 2656
> 
> Version of kernel is 2.6.28 (maybe this code is in newer versions too).
> Could someone confirm, that this is a real bug? Thank you for your time.

Yes, it looks like a real bug.  As the comment states, it's an unlikely
error path. This would explain how it went undetected this long.  Should
be fixed anyway.

Thanks,
Shaggy
-- 
David Kleikamp
IBM Linux Technology Center