Re: dm-crypt with no_read_workqueue and no_write_workqueue + btrfs scrub = BUG()

[Ignat Korchagin <ignat@cloudflare.com>]

On Wed, Dec 23, 2020 at 9:20 PM Maciej S. Szmigiero
<mail@maciej.szmigiero.name> wrote:
>
> On 23.12.2020 22:09, Ignat Korchagin wrote:
> (..)
> > I've been looking into this for the last couple of days because of
> > other reports [1].
> > Just finished testing a possible solution. Will submit soon.
>
> Thanks for looking into it.
>
> By the way, on a bare metal I am actually hitting a different problem
> (scheduling while atomic) when scrubbing a btrfs filesystem, just as one
> of your users from that GitHub report had [1].

That is because dm-crypt calls "wait_for_completion" in rare cases
when Crypto API (cryptd) backlogs the decryption request. I've
reproduced that one as well (even with no FS).
We never hit these problems in the original testing probably due to
the fact we mostly used xtsproxy custom crypto module, which is
totally synchronous.
I did test it later with standard crypto, but did not encounter these
problems as well most likely because it is also depends which storage
driver underneath we are using: most of them do not submit read
requests to dm-crypt in irq/softirq context in the first place

> I've pasted that backtrace in my original Dec 14 message.
>
> Thanks,
> Maciej
>
> [1]: https://github.com/cloudflare/linux/issues/1#issuecomment-736734243

Regards,
Ignat