questions about qemu io_uring/dio

questions about qemu io_uring/dio

[Ian Kumlien]

Hi,

From what i understand, DIO is supposed to be supported...

But when i switched qemu to use io_uring it seems like the filesystems
on the VM:s get silently corrupted... (they also run btrfs, and it's
thus detected)

The system is mounted as nodatacow, I can't set +C with chattr so I
assume that the images are actually noncow

Any ideas?

Re: questions about qemu io_uring/dio

[Ian Kumlien]

Ok, so how about warning about incompatible mount options and not just
swallowing them saying nothing?

It's kinda assumed that it's working IF there is no error messages... ;)

nodatacow doesn't work on subvolumes, important information to know! =)

On Wed, Oct 28, 2020 at 12:25 PM Ian Kumlien <ian.kumlien@gmail.com> wrote:
>
> Hi,
>
> From what i understand, DIO is supposed to be supported...
>
> But when i switched qemu to use io_uring it seems like the filesystems
> on the VM:s get silently corrupted... (they also run btrfs, and it's
> thus detected)
>
> The system is mounted as nodatacow, I can't set +C with chattr so I
> assume that the images are actually noncow
>
> Any ideas?

Re: questions about qemu io_uring/dio

[Chris Murphy]

On Wed, Oct 28, 2020 at 4:17 PM Ian Kumlien <ian.kumlien@gmail.com> wrote:
>
> Hi,
>
> From what i understand, DIO is supposed to be supported...
>
> But when i switched qemu to use io_uring it seems like the filesystems
> on the VM:s get silently corrupted... (they also run btrfs, and it's
> thus detected)
>
> The system is mounted as nodatacow, I can't set +C with chattr so I
> assume that the images are actually noncow
>
> Any ideas?

Could be related.
https://bugzilla.kernel.org/show_bug.cgi?id=208875

I'm not sure if the fix made it into 5.9 but pretty sure it's in 5.10
but just haven't yet had a chance to test it.

-- 
Chris Murphy

Re: questions about qemu io_uring/dio

[Ian Kumlien]

On Thu, Oct 29, 2020 at 5:34 AM Chris Murphy <lists@colorremedies.com> wrote:
>
> On Wed, Oct 28, 2020 at 4:17 PM Ian Kumlien <ian.kumlien@gmail.com> wrote:
> >
> > Hi,
> >
> > From what i understand, DIO is supposed to be supported...
> >
> > But when i switched qemu to use io_uring it seems like the filesystems
> > on the VM:s get silently corrupted... (they also run btrfs, and it's
> > thus detected)
> >
> > The system is mounted as nodatacow, I can't set +C with chattr so I
> > assume that the images are actually noncow
> >
> > Any ideas?
>
> Could be related.
> https://bugzilla.kernel.org/show_bug.cgi?id=208875
>
> I'm not sure if the fix made it into 5.9 but pretty sure it's in 5.10
> but just haven't yet had a chance to test it.

Yeah, i found that bit when googling actually, it will be in 5.10 apparently

But the subvolume mismatch vs root volume is another thing - you can't switch
compression or cow settings and imho, this should be in dmesg as a warning

Started to look at it but i don't really know enough =)

> --
> Chris Murphy