Re: [Qemu-devel] [PATCH v2 2/2] block/raw-posix: use seek_hole ahead of fiemap

[Max Reitz <mreitz@redhat.com>]

On 26.09.2014 01:14, Tony Breeds wrote:
> try_fiemap() uses FIEMAP_FLAG_SYNC which has a significant performance
> impact.
>
> Prefer seek_hole() over fiemap() to avoid this impact where possible.
> seek_hole is more widely used and, arguably, has potential to be
> optimised in the kernel.
>
> Reported-By: Michael Steffens <michael_steffens@posteo.de>
> Signed-off-by: Tony Breeds <tony@bakeyournoodle.com>
> Cc: Kevin Wolf <kwolf@redhat.com>
> Cc: Markus Armbruster <armbru@redhat.com>
> Cc: Stefan Hajnoczi <stefanha@redhat.com>
> Cc: Max Reitz <mreitz@redhat.com>
> Cc: Pádraig Brady <pbrady@redhat.com>
> Cc: Eric Blake <eblake@redhat.com>
> ---
> Changes since v1:
>   - split in to 2 patches
>   - tried to make the commit messages better

I used this order in at least one of the non-final versions of 4f11aa8a, 
so I'm fine with it. The reason it got rejected was (as far as I 
remember) that the FIEMAP ioctl() returns ENOTSUP if not supported, 
whereas lseek() with SEEK_HOLE/SEEK_DATA returns some failsafe value, 
probably reporting no holes at all. By reversing the order of 
try_fiemap() and try_seek_hole(), try_fiemap() therefore may very well 
become dead code. I don't object, however.

Reviewed-by: Max Reitz <mreitz@redhat.com>