[dm-devel] [PATCH] libmultipath: make vector_foreach_slot_backwards work as expected

[Benjamin Marzinski]

On Mon, May 27, 2019 at 12:03:59PM +0200, Martin Wilck wrote:
> On Fri, 2019-05-24 at 17:41 -0500, Benjamin Marzinski wrote:
> > All of the code that uses vector_foreach_slot_backwards() treats "i"
> > as
> > the index of the entry "p", but the way it was coded, that wasn't the
> > case. "i" was the number of the entry counting from 1, not 0.
> > 
> > Signed-off-by: Benjamin Marzinski <bmarzins at redhat.com>
> > ---
> >  libmultipath/vector.h | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> 
> Reviewed-by: Martin Wilck <mwilck at suse.com>
> 
> Let's add that there was only one caller that actually used
> "i"("multipath -W"). So the harm done by this bug was not as bad as one
> might think.
>  

Actually, it caused _find_controllers() to delete the wrong slot from
map->pgvec, which caused a crash on a future _find_controllers() call
because we don't check for a NULL path between

path = nvme_pg_to_path(pg);

and

path->seen = false;

Since there should never be a NULL path, I don't think we need to add
that check, but this happens in more than multipath -W. We use "i" in
_cleanup_foreign(), cleanup_nvme_map(), _delete_all(), and
_find_controllers(). 

-Ben