[dm-devel] dm-verity OOPSes on kernel 4.12rc

[Gilad Ben-Yossef]

On Thu, May 18, 2017 at 10:18 AM, Milan Broz <gmazyland at gmail.com> wrote:
> On 05/18/2017 08:55 AM, Gilad Ben-Yossef wrote:
>> On Thu, May 18, 2017 at 9:37 AM, Milan Broz <gmazyland at gmail.com> wrote:
>>> Hi Gilad,
>>>
>>> seems this OOPs is caused by async crypto hash changes in 4.12 for dm-verity.
>>>
>>
>> Oy, that is not good.
>>
>>> Could you please check if it is some known problem?
>>>
>>
>> I am not aware of any problem but of course I will try to reproduce an debug.
>>
>>> Fedora rawhide x86_64 (with 4.12rc patches) crashes always,
>>> running verity-compat-test from cryptsetup testsuite is enough to trigger this.
>>>
>>> I am not able to reproduce it on other distro but I guess it is just
>>> some kernel debugging switch that is enabled in Fedora by default.
>>>
>>
>> hmm... looking at the trace - there seems to be buffer io errors *before* the
>> BUG assert.  That is surprising for a KVM run.
>>
>> Also the code path triggering this I believe is when we read hash blocks.
>> Possibly this is triggered by an IO request failing and than we continue
>> with trying to hash the block. This is just a guess... I will take a look.
>
> Thanks!
>
> FYI, I am able to reproduce it on 32bit kernel as well,
> git from Linus' tree updated today.

OK, I know what the problem is. It's triggered by having no salt value.

Testing fix...

Thanks all,
Gilad


-- 
Gilad Ben-Yossef
Chief Coffee Drinker

"If you take a class in large-scale robotics, can you end up in a
situation where the homework eats your dog?"
 -- Jean-Baptiste Queru