On Tue, Nov 19, 2024 at 05:39:26PM +0100, Martin Wilck wrote: > On Fri, 2024-11-15 at 18:22 -0500, Benjamin Marzinski wrote: > > if libmp_mapinfo() is run on a device that has no active table, > > it will now return with a new exit code, DMP_EMPTY, to signal this. > > Currently all code paths treat this identically to DMP_NOT_FOUND. > > > > Signed-off-by: Benjamin Marzinski <bmarzins@xxxxxxxxxx> > > I just looked through the code again. I think with this change, we need > to modify dm_flush_map__() and do_foreach_partmaps(). They should > remove / act on empty maps. What do you think? I'm not sure that we need to add extra code to handle the possiblity that an empty device could appear at any time, since like I said, this is a corner case that I've never actually seen in the wild. So if a device was previously a valid multipath device, I don't think we need to worry about the possibility that it suddenly became empty. But I can see the value in running something like # multipath -F and having it clean up any empty multipath devices. As for do_foreach_partmaps(), are you thinking about cleaning up empty partition devices or non-empty partition devices on top of empty multipath devices? Non-empty partition devices on top of empty multipath devices would imply that a multipath device was valid at one point, and then became empty, which I don't see an easy way of happening. The problem with empty partition devices is that partition devices are created by kpartx completely asynchronously to us. That empty partition device could be in the process of being created. So I'm not against checking for DMP_EMPTY in dm_flush_map__() and removing the empty device if its opencount is 0. But I'd rather not try messing with partmaps. Do you have a specific case you are worried about? > Also, we can skip the call to is_mpath_part_uuid() in > do_foreach_partmaps() after my "make MAPINFO_CHECK_UUID work with > partitions" patch. I can submit a patch myself, but as you're at > already, maybe you want to do that? Yep. I can do that. -Ben > Martin
