[PATCH 4/4] nvme: delete discovery controller after 2 minutes

[james.smart@broadcom.com (James Smart)]

On 8/21/2018 6:43 AM, Hannes Reinecke wrote:
> If the CLI crashes before the 'disconnect' command is issued or when
> the 'async_connect' option is used the controller is never removed.
> This patch cleans up stale discovery controllers after 2 minutes,
>
> Signed-off-by: Hannes Reinecke <hare at suse.com>
> ---
>   drivers/nvme/host/core.c    | 5 +++++
>   drivers/nvme/host/fabrics.c | 2 +-
>   drivers/nvme/host/nvme.h    | 1 +
>   3 files changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index 358be6d217d9..b3738b327731 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -866,6 +866,11 @@ static void nvme_keep_alive_work(struct work_struct *work)
>   	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
>   			struct nvme_ctrl, ka_work);
>   
> +	if (ctrl->opts->discovery_nqn) {
> +		nvme_delete_ctrl(ctrl);
> +		return;
> +	}
> +
>   	if (nvme_keep_alive(ctrl)) {
>   		/* allocation failure, reset the controller */
>   		dev_err(ctrl->device, "keep-alive failed\n");
> diff --git a/drivers/nvme/host/fabrics.c b/drivers/nvme/host/fabrics.c
> index e484205b4cad..b98662760051 100644
> --- a/drivers/nvme/host/fabrics.c
> +++ b/drivers/nvme/host/fabrics.c
> @@ -827,7 +827,7 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
>   	}
>   
>   	if (opts->discovery_nqn) {
> -		opts->kato = 0;
> +		opts->kato = NVME_DISCOVERY_TIMEOUT;
>   		opts->nr_io_queues = 0;
>   		opts->duplicate_connect = true;
>   	}
> diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
> index 8a4ed46b986b..551a6b1dbc8c 100644
> --- a/drivers/nvme/host/nvme.h
> +++ b/drivers/nvme/host/nvme.h
> @@ -32,6 +32,7 @@ extern unsigned int admin_timeout;
>   
>   #define NVME_DEFAULT_KATO	5
>   #define NVME_KATO_GRACE		10
> +#define NVME_DISCOVERY_TIMEOUT	120
>   
>   extern struct workqueue_struct *nvme_wq;
>   extern struct workqueue_struct *nvme_reset_wq;

this doesn't necessarily track to the new TP that adds kato support to 
discovery controllers, nor the fabric spec update that has the host 
tracking kato and deleting the controller (whether discovery controller 
or not) (the actual kato as set on the controller with the grace period).

I would rather have this be a generic timer on a host that tracks to the 
kato timeout and deletes the controller (doesn't matter if discovery or 
not) if kato times out.??? If the controller is an older discovery 
controller that doesn't support kato - then the 1st kato timeout should 
fail and remove the controller.? If it's newer and supports kato, it 
would be assumed the controller stays live for an extended period- just 
like a storage controller.

-- james