On Mon, Jan 30 2017, Artur Paszkiewicz wrote:
>
> +static void raid5_reset_stripe_cache(struct mddev *mddev)
> +{
> + struct r5conf *conf = mddev->private;
> +
> + mutex_lock(&conf->cache_size_mutex);
> + while (conf->max_nr_stripes &&
> + drop_one_stripe(conf))
> + ;
> + while (conf->min_nr_stripes > conf->max_nr_stripes &&
> + grow_one_stripe(conf, GFP_KERNEL))
> + ;
> + mutex_unlock(&conf->cache_size_mutex);
> +}
This is called with the raid array suspended, so any writeout to the
array will block.
A GFP_KERNEL allocation can block waiting for writeout.
So this can deadlock.
At the very least, this should use GFP_NOIO.
It would be better to do something like resize_stripes() does, or maybe
even just use resize_stripes().
It allocates the new stripes first before suspending IO. Then it
cleans out the old stripes, and inserts the pre-allocated stripes.
NeilBrown
Attachment:
signature.asc
Description: PGP signature
