Re: [PATCH] xfs: Use xarray to track SB UUIDs instead of plain array.

["Darrick J. Wong" <djwong@kernel.org>]

On Mon, Feb 02, 2026 at 10:37:23AM +0100, Lukas Herbolt wrote:
> On 2026-02-02 08:31, Christoph Hellwig wrote:
> > On Fri, Jan 30, 2026 at 08:55:34AM -0800, Darrick J. Wong wrote:
> > > > +		xa_erase(&xfs_uuid_table, index);
> > > > +	}
> > > 
> > > Why not store the xarray index in the xfs_mount so you can delete the
> > > entry directly without having to walk the entire array?
> > 
> > Yeah, that makes a lot of sense.
> > 
> I did not want to touch the xfs_mount but if there is no objection against,
> I will add the index there.
> 
> > > And while I'm on about it ... if you're going to change data
> > > structures,
> > > why not use rhashtable or something that can do a direct lookup?
> > 
> > rhashtables require quite a bit of boilerplate.  Probably not worth
> > if for a single lookup in a relatively small colletion once per
> > mount.  But yeah, if only we had a data structure that allows
> > directly lookups without all that boilerplate..
> 
> I do not have strong preference here.

<shrug> Since the original message said "krealloc prints out warning if
allocation is bigger than 2x PAGE_SIZE", I figured that meant you were
trying to mount more than (2 * 4096) / 16 == 512 different xfs
filesystems on the same host.

I don't have a particular problem with the array search and large memory
allocation since I never mount that many filesystems, but you would
appear to be the first user to complain about a scaling limit there...

:)

--D

> -- 
> -lhe
>