Re: [PATCH] pstore: Solve lockdep warning by moving inode locks

[Kees Cook]

On Fri, Apr 28, 2017 at 2:00 AM, Lofstedt, Marta
<marta.lofstedt@xxxxxxxxx> wrote:
>
>
>> -----Original Message-----
>> From: Chris Wilson [mailto:chris@xxxxxxxxxxxxxxxxxx]
>> Sent: Friday, April 28, 2017 10:53 AM
>> To: Kees Cook <keescook@xxxxxxxxxxxx>
>> Cc: linux-kernel@xxxxxxxxxxxxxxx; Anton Vorontsov <anton@xxxxxxxxxx>;
>> Colin Cross <ccross@xxxxxxxxxxx>; Luck, Tony <tony.luck@xxxxxxxxx>;
>> Lofstedt, Marta <marta.lofstedt@xxxxxxxxx>; Namhyung Kim
>> <namhyung@xxxxxxxxxx>
>> Subject: Re: [PATCH] pstore: Solve lockdep warning by moving inode locks
>>
>> On Thu, Apr 27, 2017 at 04:20:37PM -0700, Kees Cook wrote:
>> > Lockdep complains about a possible deadlock between mount and unlink
>> > (which is technically impossible), but fixing this improves possible
>> > future multiple-backend support, and keeps locking in the right order.
>>
>> I have merged your for-next/pstore branch (which included this patch, so I
>> hope I chose correctly ;) into our CI. That should exercise it on the machines
>> that we originally found the lockdep splat.

That's the branch, yes, thanks!

>>
>> Thanks,
>> -Chris
>>
>
> Chris, I tested this on drm-tip after you merged for-next/pstore.
> EFI_VARS_PSTORE is enabled.
> I deliberately cause kernel panic and reboot, but unfortunately that kernel doesn't reboot properly. On display I see a bunch of:
> "Cleaning orphaned inode ...", but then kernel boot  is stuck.
> I run this on a BDW NUCi5, which I have been using successfully with pstore-efi for weeks.
> Fortunately I had some other pstore enabled kernels, so if I clean out /sys/fs/pstore/* with one of them I can boot above kernel again.

Hrmm... if you isolate this down to a different pstore issue, please
let me know. I haven't seen filesystem corruption in my tests yet. :P

Thanks for testing!

-Kees

-- 
Kees Cook
Pixel Security