[PATCH 2/2] nvme: add 'queue_if_no_path' semantics

[Sagi Grimberg]

> I think what you're showing is a problem for everyone, including
> fabrics. The namespace was attached to the previously existing
> subsystem, nvme1, but we didn't find the previously existing "head" to
> reattach the namespace.
> 
> I just threw the below together very hastily, so I'm sure there's
> something wrong with it, but just to capture what's wrong with the
> current code:
> 
> ---
> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index 3833584b5cee..0b3bbbe6d1e8 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -3897,8 +3897,6 @@ static void nvme_ns_remove(struct nvme_ns *ns)
>   
>   	mutex_lock(&ns->ctrl->subsys->lock);
>   	list_del_rcu(&ns->siblings);
> -	if (list_empty(&ns->head->list))
> -		list_del_init(&ns->head->entry);

Not removing it ever from the head->list?

>   	mutex_unlock(&ns->ctrl->subsys->lock);
>   
>   	synchronize_rcu(); /* guarantee not available in head->list */
> --

This is a problem. beacuse if you don't remove the ns from the
head->list nothing prevents it from being re-selected as the
current_path after we synchronize the srcu, although it is going away...