Re: dm: add support for get_unique_id

[Mike Snitzer <snitzer@kernel.org>]

On Wed, Oct 30, 2024 at 10:19:11AM -0400, Benjamin Coddington wrote:
> On 30 Oct 2024, at 10:08, Christoph Hellwig wrote:
> 
> > On Wed, Oct 30, 2024 at 10:05:03AM -0400, Benjamin Coddington wrote:
> >> Match each other in a multipath device you mean?  No, this will just return
> >> the first one where get_unique_id returns non-zero.  Can they actually be
> >> different, and if so should we return an error?
> >
> > That's what I've been wondering.  IIRC you can in theory create a
> > kernel mpath table for any devices you want.  multipathd only creates
> > them when the ids match, but do we want to rely on that?  It might be
> > perfectly fine to say if you break you keep the pieces, but then
> > I'd expected a comment about it in the comment.
> 
> Comment is easier than comparing them all, I'm happy to add it.  Seems like
> a mpath table with different devices would make little pieces of anything
> you try to put on it, and you wouldn't even get to keep those pieces.
> 
> Maybe other dm experts can think of ways that might break things.. I'll wait
> a day or so.

It would be a concern for multipathd, you don't need to worry about this.

Documenting that dm_blk_get_unique_id returns the uuid from the
first underlying device that supports ->get_unique_id should be
sufficient.

Mike