Re: [PATCH for-4.2 v2 0/3] qcow2: Fix data corruption on XFS

[Max Reitz]

On 01.11.19 16:25, Max Reitz wrote:
> Hi,
> 
> As I reasoned here:
> https://lists.nongnu.org/archive/html/qemu-block/2019-11/msg00027.html
> I’m no longer convinced of reverting c8bb23cbdbe.  I could see a
> significant performance improvement from it on ext4 with aio=native in a
> guest that does random 4k writes, and as such I feel it would be a
> regression to revert it for 4.2.
> 
> To work around the XFS corruption, we still need the other three patches
> from the series, of course.  We cannot restrict the workaround to just
> XFS, because maybe the file is on a remote filesystem and then we don’t
> know anything about the host configuration.
> 
> The performance impact should still be minimal because this just
> serializes post-EOF write-zeroes and data writes, and that just doesn’t
> happen very often, even with handle_alloc_space() in qcow2.
> 
> 
> I would love to have more time to make a decision, but there simply
> isn’t any.  Patches for 4.1.1 are to be collected on Monday, AFAIU.

I would have liked some reviews on this series (so I waited over the
weekend, even though I didn’t expect any), but now I’ve applied it
anyway (and sent a pull request).  I understand it was difficult last
week because of KVM Forum.

AFAIU we need it today for stable, and there didn’t seem to be any
opposition on these patches here, just on the revert of c8bb23cbdbe.

I welcome you to still review the series and then shout “STOP” at the
pull request if you find it necessary.

Max

signature.asc
Description: OpenPGP digital signature