Re: fallocate on XFS for swap

[Dave Chinner <david@fromorbit.com>]

[Aleski, can you word wrap your email text at 72 columns? ]

On Mon, Mar 12, 2018 at 10:01:54PM +0000, Besogonov, Aleksei wrote:
> [snip unrelated]
> 
> So I'm looking at the XFS code and it appears that the iomap is
> limited to 1024*PAGE_SIZE blocks at a time,

Take a closer look - that code is not used for reading file extents
and returning them to the caller.

> which is too small for
> most of swap use-cases. I can of course just loop through the file
> in 4Mb increments and, just like the bmap() code does today. But
> this just doesn't look right and it's not atomic. And it looks
> like iomap in ext2 doesn't have this limitation. 
> 
> The stated rationale for the XFS limit is:
> >/*
> > * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
> > * to keep the chunks of work done where somewhat symmetric with the
> > * work writeback does. This is a completely arbitrary number pulled
> > * out of thin air as a best guess for initial testing.
> > *
> > * Note that the values needs to be less than 32-bits wide until
> > * the lower level functions are updated.
> > */

Yeah, that's in the IOMAP_WRITE path used for block allocation. swap
file mapping should not be asking for IOMAP_WRITE mappings that
trigger extent allocation, so you should never hit this case.

You should probably be using the IOMAP_REPORT path (i.e. basically
very similar code to iomap_fiemap/iomap_fiemap_apply and rejecting
any file that returns an iomap that is not IOMAP_MAPPED or
IOMAP_UNWRITTEN. Also, you want to reject any file that returns
IOMAP_F_SHARED in iomap->flags, too, because swapfiles can't do COW
to break extent sharing on writes.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com