GPF on access to presumably corrupted file

GPF on access to presumably corrupted file

[Russell Coker]

[-- Attachment #1: Type: text/plain, Size: 1207 bytes --]

The attached dmesg log shows the results of trying to cat a file on a BTRFS 
filesystem when running the latest Debian/Unstable kernel (upstream 3.12.8 
with some Debian patches that probably aren't relevant to BTRFS).

I've rebooted the Thinkpad in question and repeated the problem after a 
reboot.  So I am fairly sure that the problem isn't directly caused by memory 
corruption.  I presume that it's corruption on disk causing this repeatable 
problem and such corruption could be caused by a memory error (which is 
something I've had happen before on a different system).  But I don't think 
that such corruption should cause a GPF.

So while I think we should consider the possibility that the filesystem was 
corrupted due to a hardware fault (of which there are several possibilities 
when dealing with a laptop) the inability to recover seems like a bug in 
BTRFS.

When this happens every process that tries to access the file in question is 
reported as being stuck in D state.  Sometimes such processes respond to kill 
-9 (I thought that was impossible) and sometimes they remain until reboot.

-- 
My Main Blog         http://etbe.coker.com.au/
My Documents Blog    http://doc.coker.com.au/

[-- Attachment #2: dmesg.gz --]
[-- Type: application/x-gzip, Size: 19879 bytes --]

Re: GPF on access to presumably corrupted file

[Josef Bacik]

On 01/31/2014 02:24 AM, Russell Coker wrote:
> The attached dmesg log shows the results of trying to cat a file on a BTRFS
> filesystem when running the latest Debian/Unstable kernel (upstream 3.12.8
> with some Debian patches that probably aren't relevant to BTRFS).
>
> I've rebooted the Thinkpad in question and repeated the problem after a
> reboot.  So I am fairly sure that the problem isn't directly caused by memory
> corruption.  I presume that it's corruption on disk causing this repeatable
> problem and such corruption could be caused by a memory error (which is
> something I've had happen before on a different system).  But I don't think
> that such corruption should cause a GPF.
>
> So while I think we should consider the possibility that the filesystem was
> corrupted due to a hardware fault (of which there are several possibilities
> when dealing with a laptop) the inability to recover seems like a bug in
> BTRFS.
>
> When this happens every process that tries to access the file in question is
> reported as being stuck in D state.  Sometimes such processes respond to kill
> -9 (I thought that was impossible) and sometimes they remain until reboot.
>
This may be a bug we just fixed in the recent git pull, are you running 
with compression?  If you are try the recent for-linus branch from Chris 
and see if you still have the problem.  Thanks,

Josef

Re: GPF on access to presumably corrupted file

[Chris Mason]

On Fri 31 Jan 2014 09:07:15 AM EST, Josef Bacik wrote:
>
> On 01/31/2014 02:24 AM, Russell Coker wrote:
>> The attached dmesg log shows the results of trying to cat a file on a
>> BTRFS
>> filesystem when running the latest Debian/Unstable kernel (upstream
>> 3.12.8
>> with some Debian patches that probably aren't relevant to BTRFS).
>>
>> I've rebooted the Thinkpad in question and repeated the problem after a
>> reboot.  So I am fairly sure that the problem isn't directly caused
>> by memory
>> corruption.  I presume that it's corruption on disk causing this
>> repeatable
>> problem and such corruption could be caused by a memory error (which is
>> something I've had happen before on a different system).  But I don't
>> think
>> that such corruption should cause a GPF.
>>
>> So while I think we should consider the possibility that the
>> filesystem was
>> corrupted due to a hardware fault (of which there are several
>> possibilities
>> when dealing with a laptop) the inability to recover seems like a bug in
>> BTRFS.
>>
>> When this happens every process that tries to access the file in
>> question is
>> reported as being stuck in D state.  Sometimes such processes respond
>> to kill
>> -9 (I thought that was impossible) and sometimes they remain until
>> reboot.
>>
> This may be a bug we just fixed in the recent git pull, are you
> running with compression?  If you are try the recent for-linus branch
> from Chris and see if you still have the problem.  Thanks,
>

It does look like one we fixed, even without compression please give 
for-linus a shot.

-chris

Re: GPF on access to presumably corrupted file

[David Sterba]

On Fri, Jan 31, 2014 at 09:11:29AM -0500, Chris Mason wrote:
> On Fri 31 Jan 2014 09:07:15 AM EST, Josef Bacik wrote:
> >
> >On 01/31/2014 02:24 AM, Russell Coker wrote:
> >>The attached dmesg log shows the results of trying to cat a file on a
> >>BTRFS
> >>filesystem when running the latest Debian/Unstable kernel (upstream
> >>3.12.8
> >>with some Debian patches that probably aren't relevant to BTRFS).
> >>
> >>I've rebooted the Thinkpad in question and repeated the problem after a
> >>reboot.  So I am fairly sure that the problem isn't directly caused
> >>by memory
> >>corruption.  I presume that it's corruption on disk causing this
> >>repeatable
> >>problem and such corruption could be caused by a memory error (which is
> >>something I've had happen before on a different system).  But I don't
> >>think
> >>that such corruption should cause a GPF.
> >>
> >>So while I think we should consider the possibility that the
> >>filesystem was
> >>corrupted due to a hardware fault (of which there are several
> >>possibilities
> >>when dealing with a laptop) the inability to recover seems like a bug in
> >>BTRFS.
> >>
> >>When this happens every process that tries to access the file in
> >>question is
> >>reported as being stuck in D state.  Sometimes such processes respond
> >>to kill
> >>-9 (I thought that was impossible) and sometimes they remain until
> >>reboot.
> >>
> >This may be a bug we just fixed in the recent git pull, are you
> >running with compression?  If you are try the recent for-linus branch
> >from Chris and see if you still have the problem.  Thanks,
> >
> 
> It does look like one we fixed, even without compression please give
> for-linus a shot.

Yes it is the bug, https://bugzilla.kernel.org/show_bug.cgi?id=68411 .