Re: [RFC v2 0/6] block/io: avoid failure caused by misaligned BLKZEROOUT ioctl

[Kevin Wolf <kwolf@redhat.com>]

Am 05.02.2026 um 13:13 hat Fiona Ebner geschrieben:
> Hi Stefan,
> 
> Am 02.02.26 um 11:15 PM schrieb Stefan Hajnoczi:
> > On Fri, Jan 09, 2026 at 01:08:27PM +0100, Fiona Ebner wrote:
> >> Previous discussion here:
> >> https://lore.kernel.org/qemu-devel/20260105143416.737482-1-f.ebner@proxmox.com/
> >>
> >> Commit 5634622bcb ("file-posix: allow BLKZEROOUT with -t writeback")
> >> enables the BLKZEROOUT ioctl when using 'writeback' cache, regressing
> >> certain 'qemu-img convert' invocations, because of a pre-existing
> >> issue. Namely, the BLKZEROOUT ioctl might fail with errno EINVAL when
> >> the request is shorter than the block size of the block device.
> >>
> >> Stefan suggested prioritizing bl.pwrite_zeroes_alignment in
> >> bdrv_co_do_zero_pwritev(). This RFC explores that approach and the
> >> issues with qcow2 I encountered, where
> >> bl.pwrite_zeroes_alignment = s->subcluster_size;
> >> I would be happy to discuss potential solutions and whether we should
> >> use this approach after all.
> > 
> > Hi Fiona,
> > I wanted to continue this discussion. My thoughts are that making
> > bdrv_co_do_zero_pwritev() use bl.pwrite_zeroes_alignment is the right
> > long-term solution to keep all the padding logic in one place.
> > 
> > On the other hand, your series shows it involves fixing a bunch of test
> > failures and that's not fun. The original bug that is being solved here
> > is my doing, so feel free to hand this over to me if you decide you
> > don't want to work on it.
> 
> in your other mail, you mentioned you'll ask Kevin for his opinion. So
> in part, I was waiting for that. But I also was side-tracked by other
> things, and it will be 1-2 more weeks until I can really focus on this
> again. If that is too long, please go ahead and pick it up.

I didn't review this thoroughly yet, but I agree that considering the
alignment from the start is the better solution and also more consistent
with what we're already doing for normal reads and writes.

We just need to make sure that we use the right alignments in the right
places, which can be a bit confusing with the fallbacks to buffered zero
writes here and there.

I assume that there is enough time left to do this before the 11.0
release and there is no need to take something like v1 as an
intermediate solution?

Kevin