On Fri, Aug 27, 2021 at 11:55:14AM -0700, Luis Chamberlain wrote:

quoted

I think the add_disk should just return r.  If you look at the
callers they eventualy end up in dm_table_destroy, which does
this cleanup.

I don't see it. What part of dm_table_destroy() does this?

Sorry, dm_destroy, not dm_table_destroy.  For dm_early_create it's
trivial as that calls both dm_table_destroy and dm_destroy in the error
path.  The normal table_load case is a separate ioctl, but if that
fails userspace needs to call the DM_DEV_REMOVE_CMD to cleanup
the state - similar to any other failure.