Re: [LSF TOPIC] statx extensions for subvol/snapshot filesystems & more

From: Miklos Szeredi <miklos@szeredi.hu>
To: Josef Bacik <josef@toxicpanda.com>
Cc: Kent Overstreet <kent.overstreet@linux.dev>,
	linux-bcachefs@vger.kernel.org,  linux-btrfs@vger.kernel.org,
	linux-fsdevel@vger.kernel.org,  Jan Kara <jack@suse.cz>,
	linux-kernel@vger.kernel.org,  lsf-pc@lists.linux-foundation.org
Subject: Re: [LSF TOPIC] statx extensions for subvol/snapshot filesystems & more
Date: Mon, 26 Feb 2024 09:14:51 +0100	[thread overview]
Message-ID: <CAJfpeguEbd1h96OVhDAPEwoWGrF0Nk7q0GD9W6FhGp+eVgVRCQ@mail.gmail.com> (raw)
In-Reply-To: <20240222154802.GA1219527@perftesting>

On Thu, 22 Feb 2024 at 16:48, Josef Bacik <josef@toxicpanda.com> wrote:

> Right, nobody is arguing that.  Our plan is to
>
> 1) Introduce some sort of statx mechanism to expose this information.
> 2) Introduce an incompat fs feature flag to give unique inode numbers for people
>    that want them, and there stop doing the st_dev thing we currently do.

I don't get it.   What does the filesystem (the actual bits on disk)
have anything to do with how st_dev is exposed to userspace
applications?

This is not a filesystem feature, this is an interface feature.  And I
even doubt that salvaging st_dev is worth it.  Userspace should just
be converted to use something else.  In other words st_ino *and*
st_dev are legacy and we need to find superior alternatives.

Seems like there's an agreement about file handle being able to replace st_ino.

I'm not quite sure fsid or uuid can replace st_dev, but that's up for
discussion.

Thanks,
Miklos