[Cluster-devel] [PATCH v2 0/6] SEEK_HOLE / SEEK_DATA via iomap

[Andreas Gruenbacher]

On Mon, Jun 26, 2017 at 11:50 AM, Christoph Hellwig <hch at infradead.org> wrote:
> On Fri, Jun 23, 2017 at 03:34:38PM +0200, Andreas Gruenbacher wrote:
>> These patches, on top of the page_cache_seek_hole_data patches posted earlier
>> today, convert xfs to implement SEEK_HOLE / SEEK_DATA via iomap, and implement
>> SEEK_HOLE / SEEK_DATA via iomap in gfs2.
>
> Please send those as one series - in fact to me the previous
> series doesn't even make sense - just introduce page_cache_seek_hole_data
> for the iomap code only as everyone should be using that.
>
>> ext4 doesn't implement IOMAP_REPORT, so it can't be switched to use the iomap
>> based SEEK_HOLE / SEEK_DATA or fiemap code so far.
>
> Just treat it as a no-op, it only really matters for the reflink case.

I'm actually not sure what you're talking about. The
page_cache_seek_hole_data patches fix a bug in the lseek
implementations of xfs and ext4. We want those bugs to be fixed on
both filesystems. On ext4, we cannot switch to iomap_seek_hole_data
because ext4 doesn't implement IOMAP_REPORT. We also cannot fall back
to a dummy IOMAP_REPORT function that reports the whole file as data
because that would be a regression -- after all, ext4 supports lseek
SEEK_HOLE / SEEK_DATA already.

We could squash the two xfs patches together, but __xfs_seek_hole_data
can only be removed once your patch for cleaning up the quota code is
merged, so the combined patch would convert __xfs_seek_hole_data to
use page_cache_seek_hole_data and xfs_seek_hole_data to use
iomap_seek_hole_data at the same time, which is a bit messy.

Thanks,
Andreas