Re: crypto: ablk_decrypt causes BUG in scatterwalk

From: Dmitry Vyukov
Date: Tue Jan 12 2016 - 12:09:43 EST


I can _not_ reproduce it on clean upstream tree, nor on
git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6.git.
I probably messed up something with the unmerged fixes during my rebase.
Sorry.



On Tue, Jan 12, 2016 at 5:27 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
> On Tue, Jan 12, 2016 at 2:23 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>> On Tue, Jan 12, 2016 at 2:16 PM, Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote:
>>> On Tue, Jan 12, 2016 at 10:27:44AM +0100, Dmitry Vyukov wrote:
>>>> Hello,
>>>>
>>>> The following program cause BUG in scatterwalk:
>>>
>>> Weird, I can't reproduce the crash. Can you perhaps convert this to
>>> a non-racy reproduction, possibly by using strace to find the order
>>> of execution that leads to the crash?
>>
>>
>> Concurrent execution seems to be important to reproduce the crash. I
>> wasn't able to reproduce it in single-threaded mode.
>> Try to run it in a tight parallel loop for some time.
>
>
> I've extracted another reproducer:
> https://gist.githubusercontent.com/dvyukov/94aed45494edf1dff0a8/raw/a31f439fdcdd8a8905235a9bce6e5863ca8ccf68/gistfile1.txt
>
> I don't know whether there is a simpler way to run a reproducer in a
> tight parallel loop, but I use:
> $ go get golang.org/x/tools/cmd/stress
> $ GOPATH/bin/stress -p 16 ./a.out
>
> It triggers the BUG within seconds for me.