Re: read corruption with qemu master io_uring engine / linux master / btrfs(?)

[Dominique MARTINET <dominique.martinet@atmark-techno.com>]

Thanks for the replies.

Nikolay Borisov wrote on Tue, Jun 28, 2022 at 10:03:20PM +0300:
> >    qemu-system-x86_64 -drive file=qemu/atde-test,if=none,id=hd0,format=raw,cache=none,aio=io_uring \
> >        -device virtio-blk-pci,drive=hd0 -m 8G -smp 4 -serial mon:stdio -enable-kvm
> 
> So cache=none means O_DIRECT and using io_uring. This really sounds similar
> to:
> 
> ca93e44bfb5fd7996b76f0f544999171f647f93b

That looks close, yes...

> This commit got merged into v5.17 so you shouldn't be seeing it on 5.17 and
> onwards.
> 
> <snip>
> 
> > 
> > Perhaps at this point it might be simpler to just try to take qemu out
> > of the equation and issue many parallel reads to different offsets
> > (overlapping?) of a large file in a similar way qemu io_uring engine
> > does and check their contents?
> 
> Care to run the sample program in the aforementioned commit and verify it's
> not failing

But unfortunately it seems like it is properly fixed on my machines:
---
io_uring read result for file foo:

  cqe->res == 8192 (expected 8192)
  memcmp(read_buf, write_buf) == 0 (expected 0)
---

Nikolay Borisov wrote on Tue, Jun 28, 2022 at 10:05:39PM +0300:
> Alternatively change cache=none (O_DIRECT) to cache=writeback (ordinary
> buffered writeback path) that way we'll know if it's related to the
> iomap-based O_DIRECT code in btrfs.

Good idea; I can confirm this doesn't reproduce without cache=none, so
O_DIRECT probably is another requirement here (probably because I
haven't been able to reproduce on a freshly created fs either, so not
being able to reproducing in a few tries is no guarantee...)


Jens Axboe wrote on Tue, Jun 28, 2022 at 01:12:54PM -0600:
> Not sure what's going on here, but I use qemu with io_uring many times
> each day and haven't seen anything odd. This is on ext4 and xfs however,
> I haven't used btrfs as the backing file system. I wonder if we can boil
> this down into a test case and try and figure out what is doing on here.

Yes I'd say it's fs specific, I've not been able to reproduce on ext4 or
xfs -- but then again I couldn't reproduce with btrfs on a new
filesystem so there probably are some other conditions :/

I also agree writing a simple program like the io_uring test in the
above commit that'd sort of do it like qemu and compare contents would
be ideal.
I'll have a stab at this today.

-- 
Dominique