[PATCH 0/4] nvmet: fix configfs attr update handling for discovered subsystems

[Christoph Hellwig]

On Wed, Sep 24, 2025 at 04:13:57PM -0600, Keith Busch wrote:
> On Wed, Sep 24, 2025 at 11:26:00PM +0300, Max Gurtovoy wrote:
> > 1. Forbid changes to controller ID min/max attributes values on already
> >    discovered subsystems.
> > 2. Switch cntlid ida allocation from global to per-subsystem scope,
> >    matching the granularity of controller ID ranges.
> > 3. Forbid changes to vendor ID and subsystem vendor ID attributes values
> >    on already discovered subsystems.
> > 4. Forbid changes to max_qid attribute values on already discovered
> >    subsystems.
> 
> Is there a reason these should be changeable after they're initialized
> even prior to being discovered?

Yes, the whole idea of allowing modifications when enabled, but not
discovered seems odd.

This seems to come from:

commit 87fd4cc1c0dda038c9a3617c9d07d5159326e80f
Author: Noam Gottlieb <ngottlieb at nvidia.com>
Date:   Mon Jun 7 12:23:24 2021 +0300

    nvmet: make ver stable once connection established

I really think we should try to replace all that with an enable
check, hoping we're not going to break something that started to
rely on it.