[dm-devel] md: fix missing check on list iterator

[Mike Snitzer]

On Sun, Mar 27 2022 at  1:37P -0400,
Xiaomeng Tong <xiam0nd.tong at gmail.com> wrote:

> The bug is here:
>     bypass_pg(m, pg, bypassed);
> 
> The list iterator 'pg' will point to a bogus position containing
> HEAD if the list is empty or no element is found. This case must
> be checked before any use of the iterator, otherwise it will lead
> to a invalid memory access.
> 
> To fix this bug, run bypass_pg(m, pg, bypassed); and return 0
> when found, otherwise return -EINVAL.
> 
> Cc: stable at vger.kernel.org
> Fixes: ^1da177e4c3f4 ("Linux-2.6.12-rc2")
> Signed-off-by: Xiaomeng Tong <xiam0nd.tong at gmail.com>
> ---
>  drivers/md/dm-mpath.c | 9 +++++----
>  1 file changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c
> index f4719b65e5e3..6ba8f1133564 100644
> --- a/drivers/md/dm-mpath.c
> +++ b/drivers/md/dm-mpath.c
> @@ -1496,12 +1496,13 @@ static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed)
>  	}
>  
>  	list_for_each_entry(pg, &m->priority_groups, list) {
> -		if (!--pgnum)
> -			break;
> +		if (!--pgnum) {
> +			bypass_pg(m, pg, bypassed);
> +			return 0;
> +		}
>  	}
>  
> -	bypass_pg(m, pg, bypassed);
> -	return 0;
> +	return -EINVAL;
>  }
>  
>  /*
> -- 
> 2.17.1
> 

Did you acually hit a bug (invalid memory access)?

I cannot see how given the checks prior to iterating m->priority_groups:

        if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum ||
            !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) {
                DMWARN("invalid PG number supplied to bypass_pg");
                return -EINVAL;
        }

So I have _not_ taken your "fix".