Re: [PATCH v9 2/2] ublk: enable zoned storage support

[Niklas Cassel]

On Fri, Jul 14, 2023 at 09:25:10AM +0200, Andreas Hindborg wrote:
> From: Andreas Hindborg <a.hindborg@xxxxxxxxxxx>

Hello Andreas!

> 
> Add zoned storage support to ublk: report_zones and operations:
>  - REQ_OP_ZONE_OPEN
>  - REQ_OP_ZONE_CLOSE
>  - REQ_OP_ZONE_FINISH
>  - REQ_OP_ZONE_RESET
>  - REQ_OP_ZONE_APPEND
> 
> The zone append feature uses the `addr` field of `struct ublksrv_io_cmd` to
> communicate ALBA back to the kernel. Therefore ublk must be used with the
> user copy feature (UBLK_F_USER_COPY) for zoned storage support to be
> available. Without this feature, ublk will not allow zoned storage support.
> 
> Signed-off-by: Andreas Hindborg <a.hindborg@xxxxxxxxxxx>
> ---
>  drivers/block/ublk_drv.c      | 335 ++++++++++++++++++++++++++++++++--
>  include/uapi/linux/ublk_cmd.h |  36 +++-
>  2 files changed, 354 insertions(+), 17 deletions(-)
> 
> diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c
> index 8d271901efac..5698f4575e05 100644
> --- a/drivers/block/ublk_drv.c
> +++ b/drivers/block/ublk_drv.c
> @@ -56,16 +56,21 @@
>  		| UBLK_F_USER_RECOVERY_REISSUE \
>  		| UBLK_F_UNPRIVILEGED_DEV \
>  		| UBLK_F_CMD_IOCTL_ENCODE \
> -		| UBLK_F_USER_COPY)
> +		| UBLK_F_USER_COPY \
> +		| UBLK_F_ZONED)
>  
>  /* All UBLK_PARAM_TYPE_* should be included here */
> -#define UBLK_PARAM_TYPE_ALL (UBLK_PARAM_TYPE_BASIC | \
> -		UBLK_PARAM_TYPE_DISCARD | UBLK_PARAM_TYPE_DEVT)
> +#define UBLK_PARAM_TYPE_ALL                                \
> +	(UBLK_PARAM_TYPE_BASIC | UBLK_PARAM_TYPE_DISCARD | \
> +	 UBLK_PARAM_TYPE_DEVT | UBLK_PARAM_TYPE_ZONED)
>  
>  struct ublk_rq_data {
>  	struct llist_node node;
>  
>  	struct kref ref;
> +	__u64 sector;
> +	__u32 operation;
> +	__u32 nr_sectors;
>  };
>  
>  struct ublk_uring_cmd_pdu {
> @@ -110,6 +115,11 @@ struct ublk_uring_cmd_pdu {
>   */
>  #define UBLK_IO_FLAG_NEED_GET_DATA 0x08
>  
> +/*
> + * Set when IO is Zone Append
> + */
> +#define UBLK_IO_FLAG_ZONE_APPEND 0x10
> +
>  struct ublk_io {
>  	/* userspace buffer address from io cmd */
>  	__u64	addr;
> @@ -185,11 +195,264 @@ struct ublk_params_header {
>  	__u32	types;
>  };
>  
> +static inline unsigned int ublk_req_build_flags(struct request *req);
> +static inline struct ublksrv_io_desc *ublk_get_iod(struct ublk_queue *ubq,
> +						   int tag);
> +
>  static inline bool ublk_dev_is_user_copy(const struct ublk_device *ub)
>  {
>  	return ub->dev_info.flags & UBLK_F_USER_COPY;
>  }
>  
> +static inline bool ublk_dev_is_zoned(const struct ublk_device *ub)
> +{
> +	return ub->dev_info.flags & UBLK_F_ZONED;
> +}
> +
> +static inline bool ublk_queue_is_zoned(struct ublk_queue *ubq)
> +{
> +	return ubq->flags & UBLK_F_ZONED;
> +}
> +
> +#ifdef CONFIG_BLK_DEV_ZONED
> +
> +static int ublk_get_nr_zones(const struct ublk_device *ub)
> +{
> +	const struct ublk_param_basic *p = &ub->params.basic;
> +
> +	/* Zone size is a power of 2 */
> +	return p->dev_sectors >> ilog2(p->chunk_sectors);
> +}
> +
> +static int ublk_revalidate_disk_zones(struct ublk_device *ub)
> +{
> +	return blk_revalidate_disk_zones(ub->ub_disk, NULL);
> +}
> +
> +static int ublk_dev_param_zoned_validate(const struct ublk_device *ub)
> +{
> +	const struct ublk_param_zoned *p = &ub->params.zoned;
> +	int nr_zones;
> +
> +	if (!ublk_dev_is_zoned(ub))
> +		return -EINVAL;
> +
> +	if (!p->max_zone_append_sectors)
> +		return -EINVAL;
> +
> +	nr_zones = ublk_get_nr_zones(ub);
> +
> +	if (p->max_active_zones > nr_zones)
> +		return -EINVAL;
> +
> +	if (p->max_open_zones > nr_zones)
> +		return -EINVAL;
> +
> +	return 0;
> +}
> +
> +static int ublk_dev_param_zoned_apply(struct ublk_device *ub)
> +{
> +	const struct ublk_param_zoned *p = &ub->params.zoned;
> +
> +	disk_set_zoned(ub->ub_disk, BLK_ZONED_HM);
> +	blk_queue_required_elevator_features(ub->ub_disk->queue,
> +					     ELEVATOR_F_ZBD_SEQ_WRITE);
> +	disk_set_max_active_zones(ub->ub_disk, p->max_active_zones);
> +	disk_set_max_open_zones(ub->ub_disk, p->max_open_zones);
> +	blk_queue_max_zone_append_sectors(ub->ub_disk->queue, p->max_zone_append_sectors);
> +
> +	ub->ub_disk->nr_zones = ublk_get_nr_zones(ub);
> +
> +	return 0;
> +}
> +
> +/* Based on virtblk_alloc_report_buffer */
> +static void *ublk_alloc_report_buffer(struct ublk_device *ublk,
> +				      unsigned int nr_zones, size_t *buflen)
> +{
> +	struct request_queue *q = ublk->ub_disk->queue;
> +	size_t bufsize;
> +	void *buf;
> +
> +	nr_zones = min_t(unsigned int, nr_zones,
> +			 ublk->ub_disk->nr_zones);
> +
> +	bufsize = nr_zones * sizeof(struct blk_zone);
> +	bufsize =
> +		min_t(size_t, bufsize, queue_max_hw_sectors(q) << SECTOR_SHIFT);
> +
> +	while (bufsize >= sizeof(struct blk_zone)) {
> +		buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
> +		if (buf) {
> +			*buflen = bufsize;
> +			return buf;
> +		}
> +		bufsize >>= 1;
> +	}
> +
> +	*buflen = 0;
> +	return NULL;
> +}
> +
> +static int ublk_report_zones(struct gendisk *disk, sector_t sector,
> +		      unsigned int nr_zones, report_zones_cb cb, void *data)
> +{
> +	struct ublk_device *ub = disk->private_data;
> +	unsigned int zone_size_sectors = disk->queue->limits.chunk_sectors;
> +	unsigned int first_zone = sector >> ilog2(zone_size_sectors);
> +	unsigned int done_zones = 0;
> +	unsigned int max_zones_per_request;
> +	int ret;
> +	struct blk_zone *buffer;
> +	size_t buffer_length;
> +
> +	nr_zones = min_t(unsigned int, ub->ub_disk->nr_zones - first_zone,
> +			 nr_zones);
> +
> +	buffer = ublk_alloc_report_buffer(ub, nr_zones, &buffer_length);
> +	if (!buffer)
> +		return -ENOMEM;
> +
> +	max_zones_per_request = buffer_length / sizeof(struct blk_zone);
> +
> +	while (done_zones < nr_zones) {
> +		unsigned int remaining_zones = nr_zones - done_zones;
> +		unsigned int zones_in_request =
> +			min_t(unsigned int, remaining_zones, max_zones_per_request);
> +		struct request *req;
> +		struct ublk_rq_data *pdu;
> +		blk_status_t status;
> +
> +		memset(buffer, 0, buffer_length);
> +
> +		req = blk_mq_alloc_request(disk->queue, REQ_OP_DRV_IN, 0);
> +		if (IS_ERR(req)) {
> +			ret = PTR_ERR(req);
> +			goto out;
> +		}
> +
> +		pdu = blk_mq_rq_to_pdu(req);
> +		pdu->operation = UBLK_IO_OP_REPORT_ZONES;
> +		pdu->sector = sector;
> +		pdu->nr_sectors = zones_in_request * zone_size_sectors;
> +
> +		ret = blk_rq_map_kern(disk->queue, req, buffer, buffer_length,
> +					GFP_KERNEL);
> +		if (ret) {
> +			blk_mq_free_request(req);
> +			goto out;
> +		}
> +
> +		status = blk_execute_rq(req, 0);
> +		ret = blk_status_to_errno(status);
> +		blk_mq_free_request(req);
> +		if (ret)
> +			goto out;
> +
> +		for (unsigned int i = 0; i < zones_in_request; i++) {
> +			struct blk_zone *zone = buffer + i;
> +
> +			/* A zero length zone means no more zones in this response */
> +			if (!zone->len)
> +				break;
> +
> +			ret = cb(zone, i, data);
> +			if (ret)
> +				goto out;
> +
> +			done_zones++;
> +			sector += zone_size_sectors;
> +
> +		}
> +	}
> +
> +	ret = done_zones;
> +
> +out:
> +	kvfree(buffer);
> +	return ret;
> +}
> +
> +static blk_status_t ublk_setup_iod_zoned(struct ublk_queue *ubq,
> +					 struct request *req)
> +{
> +	struct ublksrv_io_desc *iod = ublk_get_iod(ubq, req->tag);
> +	struct ublk_io *io = &ubq->ios[req->tag];
> +	struct ublk_rq_data *pdu = blk_mq_rq_to_pdu(req);
> +	u32 ublk_op;
> +
> +	switch (req_op(req)) {
> +	case REQ_OP_ZONE_OPEN:
> +		ublk_op = UBLK_IO_OP_ZONE_OPEN;
> +		break;
> +	case REQ_OP_ZONE_CLOSE:
> +		ublk_op = UBLK_IO_OP_ZONE_CLOSE;
> +		break;
> +	case REQ_OP_ZONE_FINISH:
> +		ublk_op = UBLK_IO_OP_ZONE_FINISH;
> +		break;
> +	case REQ_OP_ZONE_RESET:
> +		ublk_op = UBLK_IO_OP_ZONE_RESET;
> +		break;
> +	case REQ_OP_ZONE_APPEND:
> +		ublk_op = UBLK_IO_OP_ZONE_APPEND;
> +		io->flags |= UBLK_IO_FLAG_ZONE_APPEND;
> +		break;
> +	case REQ_OP_DRV_IN:
> +		ublk_op = pdu->operation;
> +		switch (ublk_op) {
> +		case UBLK_IO_OP_REPORT_ZONES:
> +			iod->op_flags = ublk_op | ublk_req_build_flags(req);
> +			iod->nr_sectors = pdu->nr_sectors;
> +			iod->start_sector = pdu->sector;
> +			return BLK_STS_OK;
> +		default:
> +			return BLK_STS_IOERR;
> +		}
> +	case REQ_OP_ZONE_RESET_ALL:
> +	case REQ_OP_DRV_OUT:
> +		/* We do not support reset_all and drv_out */
> +		return BLK_STS_NOTSUPP;
> +	default:
> +		return BLK_STS_IOERR;
> +	}
> +
> +	iod->op_flags = ublk_op | ublk_req_build_flags(req);
> +	iod->nr_sectors = blk_rq_sectors(req);
> +	iod->start_sector = blk_rq_pos(req);
> +	iod->addr = io->addr;
> +
> +	return BLK_STS_OK;
> +}
> +
> +#else
> +
> +#define ublk_report_zones (NULL)
> +
> +static int ublk_dev_param_zoned_validate(const struct ublk_device *ub)
> +{
> +	return -EOPNOTSUPP;
> +}
> +
> +static int ublk_dev_param_zoned_apply(struct ublk_device *ub)
> +{
> +	return -EOPNOTSUPP;
> +}
> +
> +static int ublk_revalidate_disk_zones(struct ublk_device *ub)
> +{
> +	return 0;
> +}
> +
> +static blk_status_t ublk_setup_iod_zoned(struct ublk_queue *ubq,
> +					 struct request *req)
> +{
> +	return -EOPNOTSUPP;
> +}
> +
> +#endif
> +
>  static inline void __ublk_complete_rq(struct request *req);
>  static void ublk_complete_rq(struct kref *ref);
>  
> @@ -286,6 +549,9 @@ static int ublk_validate_params(const struct ublk_device *ub)
>  
>  		if (p->max_sectors > (ub->dev_info.max_io_buf_bytes >> 9))
>  			return -EINVAL;
> +
> +		if (ublk_dev_is_zoned(ub) && !p->chunk_sectors)
> +			return -EINVAL;
>  	} else
>  		return -EINVAL;
>  
> @@ -304,6 +570,11 @@ static int ublk_validate_params(const struct ublk_device *ub)
>  	if (ub->params.types & UBLK_PARAM_TYPE_DEVT)
>  		return -EINVAL;
>  
> +	if (ub->params.types & UBLK_PARAM_TYPE_ZONED)
> +		return ublk_dev_param_zoned_validate(ub);
> +	else if (ublk_dev_is_zoned(ub))
> +		return -EINVAL;
> +
>  	return 0;
>  }
>  
> @@ -317,6 +588,9 @@ static int ublk_apply_params(struct ublk_device *ub)
>  	if (ub->params.types & UBLK_PARAM_TYPE_DISCARD)
>  		ublk_dev_param_discard_apply(ub);
>  
> +	if (ub->params.types & UBLK_PARAM_TYPE_ZONED)
> +		return ublk_dev_param_zoned_apply(ub);
> +
>  	return 0;
>  }
>  
> @@ -487,6 +761,7 @@ static const struct block_device_operations ub_fops = {
>  	.owner =	THIS_MODULE,
>  	.open =		ublk_open,
>  	.free_disk =	ublk_free_disk,
> +	.report_zones =	ublk_report_zones,
>  };
>  
>  #define UBLK_MAX_PIN_PAGES	32
> @@ -601,7 +876,8 @@ static inline bool ublk_need_map_req(const struct request *req)
>  
>  static inline bool ublk_need_unmap_req(const struct request *req)
>  {
> -	return ublk_rq_has_data(req) && req_op(req) == REQ_OP_READ;
> +	return ublk_rq_has_data(req) &&
> +	       (req_op(req) == REQ_OP_READ || req_op(req) == REQ_OP_DRV_IN);
>  }
>  
>  static int ublk_map_io(const struct ublk_queue *ubq, const struct request *req,
> @@ -685,8 +961,13 @@ static blk_status_t ublk_setup_iod(struct ublk_queue *ubq, struct request *req)
>  {
>  	struct ublksrv_io_desc *iod = ublk_get_iod(ubq, req->tag);
>  	struct ublk_io *io = &ubq->ios[req->tag];
> +	enum req_op op = req_op(req);
>  	u32 ublk_op;
>  
> +	if (!ublk_queue_is_zoned(ubq) &&
> +	    (op_is_zone_mgmt(op) || op == REQ_OP_ZONE_APPEND))
> +		return -EIO;
> +
>  	switch (req_op(req)) {
>  	case REQ_OP_READ:
>  		ublk_op = UBLK_IO_OP_READ;
> @@ -704,6 +985,8 @@ static blk_status_t ublk_setup_iod(struct ublk_queue *ubq, struct request *req)
>  		ublk_op = UBLK_IO_OP_WRITE_ZEROES;
>  		break;
>  	default:
> +		if (ublk_queue_is_zoned(ubq))
> +			return ublk_setup_iod_zoned(ubq, req);
>  		return BLK_STS_IOERR;
>  	}
>  
> @@ -756,7 +1039,8 @@ static inline void __ublk_complete_rq(struct request *req)
>  	 *
>  	 * Both the two needn't unmap.
>  	 */
> -	if (req_op(req) != REQ_OP_READ && req_op(req) != REQ_OP_WRITE)
> +	if (req_op(req) != REQ_OP_READ && req_op(req) != REQ_OP_WRITE &&
> +	    req_op(req) != REQ_OP_DRV_IN)
>  		goto exit;
>  
>  	/* for READ request, writing data in iod->addr to rq buffers */
> @@ -1120,6 +1404,11 @@ static void ublk_commit_completion(struct ublk_device *ub,
>  	/* find the io request and complete */
>  	req = blk_mq_tag_to_rq(ub->tag_set.tags[qid], tag);
>  
> +	if (io->flags & UBLK_IO_FLAG_ZONE_APPEND) {

Do we really need to introduce a completely new flag just for this?

if (req_op(req) == REQ_OP_ZONE_APPEND)

should work just as well, no?


Kind regards,
Niklas