Re: [PATCH V2 1/2] nvmet: make ctrl-id configurable

[Christoph Hellwig <hch@lst.de>]

On Thu, Nov 21, 2019 at 11:41:53PM -0800, Chaitanya Kulkarni wrote:
> This patch adds a new target subsys attribute which allows user to
> optionally specify target controller IDs which then used in the
> nvmet_execute_identify_ctrl() to fill up the nvme_id_ctrl structure.
> 
> For example, when using a cluster setup with two nodes, with a dual
> ported NVMe drive and exporting the drive from both the nodes,
> The connection to the host fails due to the same controller ID and
> results in the following error message:-
> 
> "nvme nvmeX: Duplicate cntlid XXX with nvmeX, rejecting"
> 
> With this patch now user can partition the controller IDs for each
> subsystem by setting up the cntlid_min and cntlid_max. These values
> will be used at the time of the controller ID creation. By partitioning
> the ctrl-ids for each subsystem results in the unique ctrl-id space
> which avoids the collision.
> 
> When new attribute is not specified target will fall back to original
> cntlid calculation method.
> 
> Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
> ---
> Changes from V1:-
> 1. Add cntlid max and min configfs attributes.
> 2. Use simple if .. else statements instead of ternary operators.
> ---
>  drivers/nvme/target/configfs.c | 65 ++++++++++++++++++++++++++++++++++
>  drivers/nvme/target/core.c     |  8 +++--
>  drivers/nvme/target/nvmet.h    |  2 ++
>  3 files changed, 73 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/nvme/target/configfs.c b/drivers/nvme/target/configfs.c
> index 98613a45bd3b..5316cdb1b271 100644
> --- a/drivers/nvme/target/configfs.c
> +++ b/drivers/nvme/target/configfs.c
> @@ -862,10 +862,75 @@ static ssize_t nvmet_subsys_attr_serial_store(struct config_item *item,
>  }
>  CONFIGFS_ATTR(nvmet_subsys_, attr_serial);
>  
> +static bool nvmet_subsys_cntlid_store(struct nvmet_subsys *s, const char *page,
> +				      bool min)
> +{
> +	bool ret = true;
> +	u16 cid;
> +
> +	down_write(&nvmet_config_sem);
> +	if (sscanf(page, "%hu\n", &cid) != 1) {
> +		ret = false;
> +		goto out;
> +	}

No need to take the lock over the sscant.  That also means you can
directly return false here and don't need the ret variable or the
out label.  Also the rest of the function is so simple that I'd rather
duplicate it over the slightly confusing min parameter.

> +static ssize_t nvmet_subsys_cntlid_show(struct nvmet_subsys *s, char *page,
> +					bool min)
> +{
> +	if (min)
> +		return snprintf(page, PAGE_SIZE, "%u\n", s->cntlid_min);
> +
> +	return snprintf(page, PAGE_SIZE, "%u\n",  s->cntlid_max);

Just opencode the snprints in the two callers, which is going to
be much simpler.

> index 28438b833c1b..990ad4c7bdfd 100644
> --- a/drivers/nvme/target/core.c
> +++ b/drivers/nvme/target/core.c
> @@ -1267,8 +1267,11 @@ u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
>  	if (!ctrl->sqs)
>  		goto out_free_cqs;
>  
> +	if (subsys->cntlid_min > subsys->cntlid_max)
> +		goto out_free_cqs;

Shouldn't this check go into the store functions?

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme