[PATCH v2] nvme: check if the namespace supports metadata in nvme_map_user_request()

[Keith Busch]

On Wed, Aug 28, 2024 at 03:31:09PM +0000, Puranjay Mohan wrote:
> Keith Busch <kbusch at kernel.org> writes:
> 
> > On Tue, Aug 27, 2024 at 01:23:27PM +0000, Puranjay Mohan wrote:
> >> @@ -119,9 +120,13 @@ static int nvme_map_user_request(struct request *req, u64 ubuffer,
> >>  	struct request_queue *q = req->q;
> >>  	struct nvme_ns *ns = q->queuedata;
> >>  	struct block_device *bdev = ns ? ns->disk->part0 : NULL;
> >> +	bool has_metadata = bdev && meta_buffer && meta_len;
> >
> > If this is an admin command, then bdev is NULL, so "has_metadata" is
> > false.
> >
> >>  	struct bio *bio = NULL;
> >>  	int ret;
> >>  
> >> +	if (has_metadata && !blk_get_integrity(bdev->bd_disk))
> >> +		return -EINVAL;
> >> +
> >
> > Since has_metadata is false, we continue on to process this admin
> > command, but ignore the user's metadata settings. Do we want to return
> > error there too?
> 
> As an admin command with metadata is an invalid configuration, we can

It's not that it's an invalid configuration. The spec defines the common
command format allowing admin commands to transfer metadata.

There's just no existing spec defined command that makes use of it.
Nothing stops a vendor specific command from using it. If someone
tried, the kernel reports success, but we didn't execute the requested
command.

> ignore the metada and go ahead with the admin command or I can add the
> following after the above check:
> 
> 	if (!bdev && (meta_buffer || meta_len))
>     	return -EINVAL;
> 
> I don't know what is the best approach here.

Yeah, or just do it in one line with the bdev case too:

	bool has_metadata = meta_buffer && meta_len;
	...

	if (has_metadata && (!bdev || !blk_get_integrity(bdev->bd_disk)))
		return -EINVAL