Re: zfs: Fatal trap 12: page fault while in kernel mode

On Aug 5, 2009, at 09:09, Thomas Backman wrote:
>
> On Aug 5, 2009, at 08:50, Pawel Jakub Dawidek wrote:
>
>> On Fri, Jul 31, 2009 at 11:05:01AM +0200, Thomas Backman wrote:
>>> I'm able to reliably reproduce this panic, by having zfs recv  
>>> destroy
>>> a filesystem on the receiving end.
>>>
>>> 1) Use DDEBUG=1, I guess
>>> 2) Create a FS on the source pool you don't care about: zfs create  
>>> -o
>>> mountpoint=/testfs source/testfs
>>> 3) Clone a pool to another: zfs snapshot -r source_at_snap && zfs  
>>> send -R
>>> source_at_snap | zfs recv -Fvd target
>>> 4) zfs destroy -r source/testfs
>>> 4) zfs snapshot -r source_at_snap2 && zfs send -R -I snap  
>>> source_at_snap2 |
>>> zfs recv -Fvd target
>>> 5) ^ Panic while receiving the FS the destroyed one is mounted  
>>> under.
>>> In my case, this was tank/root three times out of three; I then  
>>> tried
>>> creating testfs under /tmp (tank/tmp/testfs), *mounting* it under / 
>>> usr/
>>> testfs, and it panics on receiving tank/usr:
>> [...]
>>
>> I repeated precisevly those steps and it doesn't panic for me.
>> Could you confirm that you use this patch?
>>
>> 	http://people.freebsd.org/~pjd/patches/zfs_vnops.c.2.patch
>>
>> If so, could you give me exact steps and all of them how to  
>> reproduce it?
>> Starting with pool creation.
> Yup, I'm using that patch (I diffed the diffs, heh). I'll try to  
> write a script to recreate the panic; I hope it's as easy as in real- 
> world conditions though.

Oh! I noticed that I actually finised my test case for this panic; I  
thought I stopped midway, but that was something else.
Here are all the details: http://lists.freebsd.org/pipermail/freebsd-fs/2009-July/006585.html
(If you have the libzfs_sendrecv patch, your own vnops patch and  
DDEBUG=1, there's no need to patch anything at all.)

Regards,
Thomas