Received-SPF: pass (google.com: domain of linux-kernel+bounces-62531-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Date: Mon, 12 Feb 2024 18:24:25 -0500
From: Kent Overstreet <kent.overstreet@linux.dev>
To: Theodore Ts'o <tytso@mit.edu>
Cc: brauner@kernel.org, linux-fsdevel@vger.kernel.org, 
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 0/7] filesystem visibililty ioctls
Message-ID: <3yuohpx2vcnl5kjs54h5med3q2525xzkxvzxqqctrgadptytww@36kyayedfnbs>
References: <20240206201858.952303-1-kent.overstreet@linux.dev>
 <20240207174009.GF119530@mit.edu>
 <kq2mh37o6goojweon37kct4r3oitiwmrbjigurc566djrdu5hd@u56irarzd452>
 <20240212224740.GA394352@mit.edu>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240212224740.GA394352@mit.edu>

On Mon, Feb 12, 2024 at 05:47:40PM -0500, Theodore Ts'o wrote:
> On Wed, Feb 07, 2024 at 03:26:55PM -0500, Kent Overstreet wrote:
> > You've still got the ext4 version, we're not taking that away. But I
> > don't think other filesystems will want to deal with the hassle of
> > changing UUIDs at runtime, since that's effectively used for API access
> > via sysfs and debugfs.
> 
> Thanks. I misunderstood the log.  I didn't realize this was just about
> not hoisting the ioctl to the VFS level, and dropping the generic uuid
> set.
> 
> I'm not convinced that we should be using the UUID for kernel API
> access, if for no other reason that not all file systems have UUID's.
> Sure, modern file systems have UUID's, and individual file systems
> might have to have specific features that don't play well with UUID's
> changing while the file system is mounted.  But I'm hoping that we
> don't add any new interfaces that rely on using the UUID for API
> access at the VFS layer.  After all, ext2 (not just ext3 and ext4) has
> supported changing the UUID while the file system has been mounted for
> *decades*.

*nod*

The intention isn't for every filesystem to be using the UUID for API
access - there's no reason to for single device filesystems, after all.

The intent is rather - for filesystems that _do_ need the UUID as a
stable identifier, let's try to standardize how's it's exposed and
used...