[PATCHv8] nvme: fix refcounting imbalance when all paths are down

[Christoph Hellwig]

> Changes to v7:
> - Move the list_del() call outside of nvme_mpath_check_last_path()
> Changes to v6:
> - Move the list_del() into nvme_mpath_check_last_path()
> - Drop the tests for GENHD_FL_UP
> Changes to v5:
> - Synchronize between nvme_init_ns_head() and nvme_mpath_check_last_path()
> - Check for removed gendisk in nvme_ns_head_submit_bio()
> Changes to v4:
> - Call del_gendisk() in nvme_mpath_check_last_path() to avoid deadlock
> Changes to v3:
> - Simplify if() clause to detect duplicate namespaces
> Changes to v2:
> - Drop memcpy() statement
> Changes to v1:
> - Always check NSIDs after reattach

The changelog belongs under the ---.

The patch also does not actually apply due to changes in 5.14-rc, so
please respin it.

> -	nvme_mpath_check_last_path(ns);
> +	/* Synchronize with nvme_init_ns_head() */
> +	mutex_lock(&ns->head->subsys->lock);
> +	if (list_empty(&ns->head->list)) {
> +		list_del_init(&ns->head->entry);
> +		last_path = true;
> +	}
> +	mutex_unlock(&ns->head->subsys->lock);
> +	if (last_path)
> +		nvme_mpath_check_last_path(ns->head);

And while we're at it - nvme_mpath_check_last_path is a little misnamed
now as we know this is the last path when we call it now.  Maybe this
function should inherit the nvme_mpath_remove_disk name and ther other
half should become nvme_mpath_cleanup_disk?

> +void nvme_mpath_remove_disk(struct nvme_ns_head *head)
> +{
> +	if (!head->disk)
> +		return;
>  	blk_set_queue_dying(head->disk->queue);
>  	/* make sure all pending bios are cleaned up */
>  	kblockd_schedule_work(&head->requeue_work);
>  	flush_work(&head->requeue_work);
>  	blk_cleanup_queue(head->disk->queue);
> -	if (!test_bit(NVME_NSHEAD_DISK_LIVE, &head->flags)) {
> +	if (!test_and_clear_bit(NVME_NSHEAD_DISK_LIVE, &head->flags)) {

This is where the conflict happens.  But given that with the above
changes we can't find the ns_head any more once nvme_ns_remove
decided it actually is the last path I don't think we even need to
clear NVME_NSHEAD_DISK_LIVE any more.  This was only needed when we
tried to reuse the ns_head.