Re: [PATCH v2] block: ublk: enable zoned storage support
From: Niklas Cassel <hidden>
Date: 2023-02-27 10:20:18
Also in:
lkml
On Fri, Feb 24, 2023 at 09:05:01PM +0100, Andreas Hindborg wrote:
quoted hunk ↗ jump to hunk
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 This allows implementation of zoned storage devices in user space. An example user space implementation based on ubdsrv is available [1]. [1] https://github.com/metaspace/ubdsrv/commit/14a2b708f74f70cfecb076d92e680dc718cc1f6d Signed-off-by: Andreas Hindborg <redacted> --- Changes since v1: - Fixed conditional compilation bug - Refactored to collect conditional code additions together - Fixed style errors - Zero stack allocated value used for zone report Reported-by: Niklas Cassel <redacted> Reported-by: kernel test robot <redacted> Link: https://lore.kernel.org/oe-kbuild-all/202302250222.XOrw9j6z-lkp@intel.com/ (local) v1: https://lore.kernel.org/linux-block/20230224125950.214779-1-nmi@metaspace.dk/ (local) drivers/block/ublk_drv.c | 150 ++++++++++++++++++++++++++++++++-- include/uapi/linux/ublk_cmd.h | 18 ++++ 2 files changed, 162 insertions(+), 6 deletions(-)diff --git a/drivers/block/ublk_drv.c b/drivers/block/ublk_drv.c index 6368b56eacf1..37e516903867 100644 --- a/drivers/block/ublk_drv.c +++ b/drivers/block/ublk_drv.c@@ -20,6 +20,7 @@ #include <linux/major.h> #include <linux/wait.h> #include <linux/blkdev.h> +#include <linux/blkzoned.h> #include <linux/init.h> #include <linux/swap.h> #include <linux/slab.h>@@ -51,10 +52,12 @@ | UBLK_F_URING_CMD_COMP_IN_TASK \ | UBLK_F_NEED_GET_DATA \ | UBLK_F_USER_RECOVERY \ - | UBLK_F_USER_RECOVERY_REISSUE) + | UBLK_F_USER_RECOVERY_REISSUE \ + | UBLK_F_ZONED) /* All UBLK_PARAM_TYPE_* should be included here */ -#define UBLK_PARAM_TYPE_ALL (UBLK_PARAM_TYPE_BASIC | UBLK_PARAM_TYPE_DISCARD) +#define UBLK_PARAM_TYPE_ALL (UBLK_PARAM_TYPE_BASIC | UBLK_PARAM_TYPE_DISCARD \ + | UBLK_PARAM_TYPE_ZONED) struct ublk_rq_data { struct llist_node node;@@ -187,6 +190,98 @@ static DEFINE_MUTEX(ublk_ctl_mutex); static struct miscdevice ublk_misc; +#ifdef CONFIG_BLK_DEV_ZONED +static void ublk_set_nr_zones(struct ublk_device *ub) +{ + const struct ublk_param_basic *p = &ub->params.basic; + + if (ub->dev_info.flags & UBLK_F_ZONED && p->chunk_sectors) + ub->ub_disk->nr_zones = p->dev_sectors / p->chunk_sectors; +} + +static void ublk_dev_param_zoned_apply(struct ublk_device *ub) +{ + const struct ublk_param_zoned *p = &ub->params.zoned; + + if (ub->dev_info.flags & UBLK_F_ZONED) { + disk_set_max_active_zones(ub->ub_disk, p->max_active_zones); + disk_set_max_open_zones(ub->ub_disk, p->max_open_zones); + } +} + +static int ublk_revalidate_disk_zones(struct gendisk *disk) +{ + return blk_revalidate_disk_zones(disk, 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; + unsigned int zone_size; + unsigned int first_zone; + int ret = 0; + + ub = disk->private_data; + + if (!(ub->dev_info.flags & UBLK_F_ZONED)) + return -EINVAL; + + zone_size = disk->queue->limits.chunk_sectors; + first_zone = sector >> ilog2(zone_size); + nr_zones = min(ub->ub_disk->nr_zones - first_zone, nr_zones); + + for (unsigned int i = 0; i < nr_zones; i++) { + struct request *req; + blk_status_t status; + struct blk_zone info = {0}; + + req = blk_mq_alloc_request(disk->queue, REQ_OP_DRV_IN, 0); + + if (IS_ERR(req)) { + ret = PTR_ERR(req); + goto out; + } + + req->__sector = sector; + + ret = blk_rq_map_kern(disk->queue, req, &info, sizeof(info), + GFP_KERNEL); + + if (ret) + goto out; + + status = blk_execute_rq(req, 0); + ret = blk_status_to_errno(status); + if (ret) + goto out; + + blk_mq_free_request(req); + + ret = cb(&info, i, data); + if (ret) + goto out; + + /* A zero length zone means don't ask for more zones */ + if (!info.len) { + nr_zones = i; + break; + } + + sector += zone_size; + } + ret = nr_zones; + + out: + return ret; +} +#else +void ublk_set_nr_zones(struct ublk_device *ub); +void ublk_dev_param_zoned_apply(struct ublk_device *ub); +int ublk_revalidate_disk_zones(struct gendisk *disk);
These are declarations, shouldn't they be dummy definitions instead?
e.g.
static int ublk_revalidate_disk_zones(struct gendisk *disk) { return -EOPNOTSUPP; };
It would be nice if they could be avoided altogether.
Looking how e.g. null-blk and btrfs has solved this:
https://github.com/torvalds/linux/blob/v6.2/fs/btrfs/Makefile#L39
https://github.com/torvalds/linux/blob/v6.2/drivers/block/null_blk/Makefile#L11
They have put the zoned stuff in a separate C file that is only compiled
when CONFIG_BLK_DEV_ZONED is set.
I'm not sure if a similar design is desired for ublk or not.
However, if a similar design pattern was used, it could probably avoid
some of these unpleasant dummy definitions altogether.
Kind regards,
Niklas
+#endif +