Re: ZFS panic in -CURRENT

(follow up below)

On 04/01/2014 06:57, R. Tyler Croy wrote:
> On Tue, 01 Apr 2014 09:41:45 +0300
> Andriy Gapon <avg_at_FreeBSD.org> wrote:
>
>> on 01/04/2014 02:22 R. Tyler Croy said the following:
>>> Bumping this with more details
>>>
>>> On Fri, 28 Mar 2014 09:53:32 -0700
>>> R Tyler Croy <tyler_at_monkeypox.org> wrote:
>>>
>>>> Apologies for the rough format here, I had to take a picture of
>>>> this failure because I didn't know what else to do.
>>>>
>>>> <http://www.flickr.com/photos/agentdero/13469355463/>
>>>>
>>>> I'm building off of the GitHub freebsd.git mirror here, and the
>>>> latest commit in the tree is neel_at_'s "Add an ioctl to suspend.."
>>>>
>>>> My dmesg/pciconf are here:
>>>> https://gist.github.com/rtyler/1faa854dff7c4396d9e8
>>> As linked before, the dmesg and `pciconf -lv` output can be found
>>> here: <https://gist.github.com/rtyler/1faa854dff7c4396d9e8>
>>>
>>> Also in addition to the photo from before of the panic, here's
>>> another reproduction photo:
>>> <https://www.flickr.com/photos/agentdero/13472248423/>
>> Are you or have you even been running with any ZFS-related kernel
>> patches?
> Negative, I've never run any specific ZFS patches on this machine (or
> any machine for that matter!)
>
> One other unique clue might be that I'm running with an encrypted
> zpool, other than that, nothing fancy here.


I've upgraded my machine to r264387 and I still experience the issue, 
here's the latest pretty picture of my panicked laptop :) 
<https://secure.flickr.com/photos/agentdero/13880032704/>

The issue still seems to stem from a failed assertion in 
zap_leaf_lookup_closest() 
(http://svnweb.freebsd.org/base/head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zap_leaf.c?revision=249195&view=markup#l446) 
but I'm not sure which assertion might be failing.

This is somewhat problematic because I cannot perform *any* FS 
operations with the tainted directory tree, not even a `du -hcs *` to 
find out how much space I can never access again :P


I can reproduce this consistently, if anybody has the time to get onto 
IRC (rtyler on Freenode and EFNet) and debug this, I can certainly act 
as remote hands with kdb to help ascertain more information about the panic.


Cheers