Re: [RFC PATCH] getting misc stats/attributes via xattr API

[Christian Brauner <brauner@kernel.org>]

On Tue, May 10, 2022 at 03:15:05PM +0200, Miklos Szeredi wrote:
> On Tue, 10 May 2022 at 13:53, Christian Brauner <brauner@kernel.org> wrote:
> 
> > > What exactly are the attributes that systemd requires?
> >
> > We keep a repo with ideas for (kernel) extensions - we should probably
> > publish that somewhere - but the list we used for a prototype roughly
> > contains:
> >
> > * mount flags MOUNT_ATTR_RDONLY etc.
> > * time flags MOUNT_ATTR_RELATIME etc. (could probably be combined with
> >   mount flags. We missed the opportunity to make them proper enums
> >   separate from other mount flags imho.)
> > * propagation "flags" (MS_SHARED)
> > * peer group
> > * mnt_id of the mount
> > * mnt_id of the mount's parent
> > * owning userns
> 
> Sounds good thus far.   And hey, we don't even need a new syscall:
> statx(2) could handle these fine.
> 
> > There's a bit more advanced stuff systemd would really want but which I
> > think is misplaced in a mountinfo system call including:
> > * list of primary and auxiliary block device major/minor
> 
> It's when you need to return variable size arrays or list of strings
> that the statx kind of interface falls down.
> 
> For that a hierarchical namespace is a much better choice, as it can
> represent arbitrary levels of arrays, while doing that with a
> specialized syscall is going to be cumbersome.
> 
> > I just have a really hard time understanding how this belongs into the
> > (f)getxattr() system call family and why it would be a big deal to just
> > make this a separate system call.
> 
> Fragmenting syntactically equivalent interfaces is bad, unifying them

Fwiw, turning this around: unifying semantically distinct interfaces
because of syntactical similarities is bad. Moving them into a
syntactically equivalent system call that expresses the difference in
semantics in its name is good.