[Linux-cachefs] kernel BUG at fs/fscache/cookie.c:547

[David Howells]

Shantanu Goel <sgoel01 at yahoo.com> wrote:

> We have fscache deployed on several hosts and recently hit the bug below 
> on a couple of hosts.  We searched the mailing lists to see if it had 
> been reported but did not find anything similar.  We would appreciate 
> any insight you might have into this problem.
> We are running the 3.9.11 + fscache-20130702 branch from dhowells git tree as referenced here
> https://www.redhat.com/archives/linux-cachefs/2013-July/msg00002.html.
> ...
> Sep  6 17:22:37 kernel BUG at fs/fscache/cookie.c:547!
> ...
> Sep  6 17:22:37 RIP: 0010:[<ffffffffa041387e>]  [<ffffffffa041387e>]
> __fscache_cookie_put+0x4e/0x60 [fscache]
> ...
> Sep  6 17:22:37 Call Trace:
> Sep  6 17:22:37  [<ffffffffa0413dd9>] __fscache_relinquish_cookie+0x109/0x260 [fscache]
> Sep  6 17:22:37  [<ffffffffa043c461>] nfs_fscache_release_super_cookie+0x31/0xb0 [nfs]

My suspicion is that you need to apply a fixup that got applied to
kernel/wait.c to fix wait_on_atomic_t():

	commit 42577ca8c3616baaafdd8f167b2e1fb959026081
	Author: David Howells <dhowells at redhat.com>
	Date:   Tue Jul 23 16:49:24 2013 +0100
	Fix __wait_on_atomic_t() to call the action func if the counter != 0

Alternatively, this commit and the commit set that you applied are all in
linux-3.11 now.

David