Re: [PATCH v4 3/5] nvmet: return DHCHAP status codes from nvmet_setup_auth()

[Christoph Hellwig <hch@lst.de>]

On Thu, Apr 04, 2024 at 05:44:58PM +0200, Daniel Wagner wrote:
> From: Hannes Reinecke <hare@kernel.org>
> 
> A failure in nvmet_setup_auth() does not mean that the NVMe
> authentication command failed, so we should rather return a
> protocol error with a 'failure1' response than an NVMe status.

Nit: try to use up the 73 characters available for commit logs, this
looks weirdly, um condensed.

> @@ -131,7 +131,6 @@ int nvmet_setup_auth(struct nvmet_ctrl *ctrl)
>  	int ret = 0;
>  	struct nvmet_host_link *p;
>  	struct nvmet_host *host = NULL;
> -	const char *hash_name;
>  
>  	down_read(&nvmet_config_sem);
>  	if (nvmet_is_disc_subsys(ctrl->subsys))
> @@ -149,13 +148,16 @@ int nvmet_setup_auth(struct nvmet_ctrl *ctrl)
>  	}
>  	if (!host) {
>  		pr_debug("host %s not found\n", ctrl->hostnqn);
> -		ret = -EPERM;
> +		ret = NVME_AUTH_DHCHAP_FAILURE_FAILED;
>  		goto out_unlock;

This is now returning returning random on the wire fields that aren't
even the NVMe status codes from a function otherwise returning Linux
errno values.  I can't see how this works or is maintainable long term.