[PATCHv6] nvme: allow to re-attach namespaces after all paths are down

[Keith Busch]

On Wed, Jun 09, 2021 at 05:01:18PM +0200, Hannes Reinecke wrote:
> -void nvme_mpath_remove_disk(struct nvme_ns_head *head)
> +void nvme_mpath_check_last_path(struct nvme_ns_head *head)
>  {
> +	bool last_path = false;
>  	if (!head->disk)
>  		return;
> -	if (head->disk->flags & GENHD_FL_UP) {
> -		nvme_cdev_del(&head->cdev, &head->cdev_device);
> -		del_gendisk(head->disk);
> +
> +	/* Synchronize with nvme_init_ns_head() */
> +	mutex_lock(&head->subsys->lock);
> +	if (list_empty(&head->list))
> +		last_path = true;
> +	mutex_unlock(&head->subsys->lock);
> +	if (last_path) {
> +		kblockd_schedule_work(&head->requeue_work);
> +		if (head->disk->flags & GENHD_FL_UP) {
> +			nvme_cdev_del(&head->cdev, &head->cdev_device);
> +			del_gendisk(head->disk);
> +		}
>  	}
> +}

This is being called when a path is removed. If it is the last path, the
head's block device is being deleted. If a live head reference is held,
and the a path reconnects, the code will find the previous head, but the
disk is gone and it's not coming back, so the new connection can't be
used.