[v7,13/13] dm: add non power of 2 zoned target

Commit Message

Pankaj Raghav June 15, 2022, 10:19 a.m. UTC

Only power of 2(po2) zoned devices were supported in linux but now non
power of 2(npo2) zoned device support has been added to the block layer.

Filesystems such as F2FS and btrfs have support for zoned devices with
po2 zone size assumption. Before adding native support for npo2 zoned
devices, it was suggested to create a dm target for npo2 zoned device to
appear as po2 device so that file systems can initially work without any
explicit changes by using this target.

The design of this target is very simple: introduce gaps between the zone
capacity and the po2 zone size of the underlying device. All IOs will be
remapped from target to the actual device location. For devices that use
zone append, the bi_sector is remapped from device to target's layout.

The read IOs that fall in the "emulated" gap area will return 0 and all
the other IOs in that area will result in an error. If an read IO span
across the zone capacity boundary, then the IOs are split between the
boundary. All other IO operations that span across a zone capacity
boundary will result in an error.

The target can be easily updated as follows:
dmsetup create <label> --table '0 <size_sects> zoned-npo2 /dev/nvme<id>'

Signed-off-by: Pankaj Raghav <[e-mail]>
Suggested-by: Johannes Thumshirn <[e-mail]>
Suggested-by: Damien Le Moal <[e-mail]>
Suggested-by: Hannes Reinecke <[e-mail]>
---
 drivers/md/Kconfig                |   9 +
 drivers/md/Makefile               |   2 +
 drivers/md/dm-zone.c              |   9 +
 drivers/md/dm-zoned-npo2-target.c | 268 ++++++++++++++++++++++++++++++
 4 files changed, 288 insertions(+)
 create mode 100644 drivers/md/dm-zoned-npo2-target.c

Comments

Damien Le Moal June 15, 2022, 11:49 a.m. UTC | #1

On 6/15/22 19:19, Pankaj Raghav wrote:
> Only power of 2(po2) zoned devices were supported in linux but now non
> power of 2(npo2) zoned device support has been added to the block layer.
> 
> Filesystems such as F2FS and btrfs have support for zoned devices with
> po2 zone size assumption. Before adding native support for npo2 zoned
> devices, it was suggested to create a dm target for npo2 zoned device to
> appear as po2 device so that file systems can initially work without any
> explicit changes by using this target.
> 
> The design of this target is very simple: introduce gaps between the zone
> capacity and the po2 zone size of the underlying device. All IOs will be
> remapped from target to the actual device location. For devices that use
> zone append, the bi_sector is remapped from device to target's layout.

Nothing special for zone append in this respect. All IOs are remapped
likewise, right ?

> 
> The read IOs that fall in the "emulated" gap area will return 0 and all
> the other IOs in that area will result in an error. If an read IO span
> across the zone capacity boundary, then the IOs are split between the
> boundary. All other IO operations that span across a zone capacity
> boundary will result in an error.
> 
> The target can be easily updated as follows:

Updated ? you mean created, no ?

> dmsetup create <label> --table '0 <size_sects> zoned-npo2 /dev/nvme<id>'
> 
> Signed-off-by: Pankaj Raghav <[e-mail]>
> Suggested-by: Johannes Thumshirn <[e-mail]>
> Suggested-by: Damien Le Moal <[e-mail]>
> Suggested-by: Hannes Reinecke <[e-mail]>
> ---
>  drivers/md/Kconfig                |   9 +
>  drivers/md/Makefile               |   2 +
>  drivers/md/dm-zone.c              |   9 +
>  drivers/md/dm-zoned-npo2-target.c | 268 ++++++++++++++++++++++++++++++
>  4 files changed, 288 insertions(+)
>  create mode 100644 drivers/md/dm-zoned-npo2-target.c
> 
> diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
> index 998a5cfdb..773314536 100644
> --- a/drivers/md/Kconfig
> +++ b/drivers/md/Kconfig
> @@ -518,6 +518,15 @@ config DM_FLAKEY
>  	help
>  	 A target that intermittently fails I/O for debugging purposes.
>  
> +config DM_ZONED_NPO2
> +	tristate "Zoned non power of 2 target"
> +	depends on BLK_DEV_DM
> +	depends on BLK_DEV_ZONED
> +	help
> +	A target that converts a zoned device with non power of 2 zone size to
> +	be power of 2. This is done by introducing gaps in between the zone
> +	capacity and the power of 2 zone size.
> +
>  config DM_VERITY
>  	tristate "Verity target support"
>  	depends on BLK_DEV_DM
> diff --git a/drivers/md/Makefile b/drivers/md/Makefile
> index 0454b0885..2863a94a7 100644
> --- a/drivers/md/Makefile
> +++ b/drivers/md/Makefile
> @@ -26,6 +26,7 @@ dm-era-y	+= dm-era-target.o
>  dm-clone-y	+= dm-clone-target.o dm-clone-metadata.o
>  dm-verity-y	+= dm-verity-target.o
>  dm-zoned-y	+= dm-zoned-target.o dm-zoned-metadata.o dm-zoned-reclaim.o
> +dm-zoned-npo2-y       += dm-zoned-npo2-target.o

This naming is in my opinion very bad as it seems related to the dm-zoned
target. e.g. dm-po2z, dm-zp2, etc.

>  
>  md-mod-y	+= md.o md-bitmap.o
>  raid456-y	+= raid5.o raid5-cache.o raid5-ppl.o
> @@ -60,6 +61,7 @@ obj-$(CONFIG_DM_CRYPT)		+= dm-crypt.o
>  obj-$(CONFIG_DM_DELAY)		+= dm-delay.o
>  obj-$(CONFIG_DM_DUST)		+= dm-dust.o
>  obj-$(CONFIG_DM_FLAKEY)		+= dm-flakey.o
> +obj-$(CONFIG_DM_ZONED_NPO2)	+= dm-zoned-npo2.o
>  obj-$(CONFIG_DM_MULTIPATH)	+= dm-multipath.o dm-round-robin.o
>  obj-$(CONFIG_DM_MULTIPATH_QL)	+= dm-queue-length.o
>  obj-$(CONFIG_DM_MULTIPATH_ST)	+= dm-service-time.o
> diff --git a/drivers/md/dm-zone.c b/drivers/md/dm-zone.c
> index af36d33f9..5efb31ba0 100644
> --- a/drivers/md/dm-zone.c
> +++ b/drivers/md/dm-zone.c
> @@ -210,6 +210,11 @@ static int dm_zone_revalidate_cb(struct blk_zone *zone, unsigned int idx,
>  		}
>  		md->zwp_offset[idx] = dm_get_zone_wp_offset(zone);
>  
> +		if (q->limits.chunk_sectors != zone->len) {

Why is this if needed ?

> +			blk_queue_chunk_sectors(q, zone->len);
> +			q->nr_zones = blkdev_nr_zones(md->disk);
> +		}
> +
>  		break;
>  	default:
>  		DMERR("Invalid zone type 0x%x at sectors %llu",
> @@ -307,6 +312,9 @@ int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
>  	if (dm_table_supports_zone_append(t)) {
>  		clear_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
>  		dm_cleanup_zoned_dev(md);
> +

no need for the blank line.

> +		if (!is_power_of_2(blk_queue_zone_sectors(q)))
> +			goto revalidate_zones;
>  		return 0;
>  	}

Why do you need to change dm_set_zones_restrictions() at all ?

>  
> @@ -318,6 +326,7 @@ int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
>  	if (!get_capacity(md->disk))
>  		return 0;
>  
> +revalidate_zones:
>  	return dm_revalidate_zones(md, t);
>  }
>  
> diff --git a/drivers/md/dm-zoned-npo2-target.c b/drivers/md/dm-zoned-npo2-target.c
> new file mode 100644
> index 000000000..c1373d3ea
> --- /dev/null
> +++ b/drivers/md/dm-zoned-npo2-target.c
> @@ -0,0 +1,268 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (C) 2022 Samsung Electronics Co., Ltd.
> + */
> +
> +#include <linux/device-mapper.h>
> +
> +#define DM_MSG_PREFIX "zoned-npo2"
> +
> +struct dmz_npo2_target {
> +	struct dm_dev *dev;
> +	sector_t zsze;
> +	sector_t zsze_po2;
> +	sector_t zsze_diff;

zsze ? is that zone size ? Spell this out please. This is not nvme.

> +	u32 nr_zones;
> +};
> +
> +enum dmz_npo2_io_cond {
> +	DMZ_NPO2_IO_INSIDE_ZONE,
> +	DMZ_NPO2_IO_ACROSS_ZONE,
> +	DMZ_NPO2_IO_OUTSIDE_ZONE,
> +};
> +
> +static inline u32 npo2_zone_no(struct dmz_npo2_target *dmh, sector_t sect)
> +{
> +	return div64_u64(sect, dmh->zsze);
> +}
> +
> +static inline u32 po2_zone_no(struct dmz_npo2_target *dmh, sector_t sect)
> +{
> +	return sect >> ilog2(dmh->zsze_po2);
> +}
> +
> +static inline sector_t target_to_device_sect(struct dmz_npo2_target *dmh,
> +					     sector_t sect)
> +{
> +	u32 zone_idx = po2_zone_no(dmh, sect);
> +
> +	sect -= (zone_idx * dmh->zsze_diff);

	return sect - (po2_zone_no(dmh, sect) * dmh->zsze_diff);
> +
> +	return sect;
> +}
> +
> +static inline sector_t device_to_target_sect(struct dmz_npo2_target *dmh,
> +					     sector_t sect)
> +{
> +	u32 zone_idx = npo2_zone_no(dmh, sect);
> +
> +	sect += (zone_idx * dmh->zsze_diff);

see above. Simplify.

> +
> +	return sect;
> +}
> +
> +/*
> + * <dev-path>

What is this above line intended meaning ?

> + * This target works on the complete zoned device. Partial mapping is not
> + * supported
> + */
> +static int dmz_npo2_ctr(struct dm_target *ti, unsigned int argc, char **argv)
> +{
> +	struct dmz_npo2_target *dmh = NULL;
> +	int ret = 0;
> +	sector_t zsze;
> +	sector_t disk_size;
> +
> +	if (argc < 1)
> +		return -EINVAL;
> +
> +	dmh = kmalloc(sizeof(*dmh), GFP_KERNEL);
> +	if (!dmh)
> +		return -ENOMEM;
> +
> +	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
> +			    &dmh->dev);

No error check ?

> +
> +	zsze = blk_queue_zone_sectors(bdev_get_queue(dmh->dev->bdev));
> +
> +	disk_size = get_capacity(dmh->dev->bdev->bd_disk);

s/disk_size/dev_capacity

> +
> +	if (ti->len != disk_size || ti->begin) {
> +		DMERR("%pg Partial mapping of the target not supported",

missing a verb ("is not...")

> +		      dmh->dev->bdev);
> +		return -EINVAL;
> +	}
> +
> +	if (is_power_of_2(zsze)) {
> +		DMERR("%pg zone size is power of 2", dmh->dev->bdev);

Hmmm... You would end up with no remapping needed so it would still
work... Why error this ? A warning would work too.

> +		return -EINVAL;
> +	}
> +
> +	dmh->zsze = zsze;
> +	dmh->zsze_po2 = 1 << get_count_order_long(zsze);
> +	dmh->zsze_diff = dmh->zsze_po2 - dmh->zsze;
> +
> +	ti->private = dmh;
> +	ti->num_flush_bios = 1;
> +	ti->num_discard_bios = 1;
> +	ti->num_secure_erase_bios = 1;
> +	ti->num_write_zeroes_bios = 1;

Why all these ? I know dm-linear do that but I do not see why they would
be necessary for a single device target.

> +
> +	dmh->nr_zones = npo2_zone_no(dmh, ti->len);
> +	ti->len = dmh->zsze_po2 * dmh->nr_zones;
> +
> +	return 0;
> +}
> +
> +static int dmz_npo2_report_zones_cb(struct blk_zone *zone, unsigned int idx,
> +				    void *data)
> +{
> +	struct dm_report_zones_args *args = data;
> +	struct dmz_npo2_target *dmh = args->tgt->private;
> +
> +	zone->start = device_to_target_sect(dmh, zone->start);
> +	zone->wp = device_to_target_sect(dmh, zone->wp);
> +	zone->len = dmh->zsze_po2;
> +	args->next_sector = zone->start + zone->len;
> +
> +	return args->orig_cb(zone, args->zone_idx++, args->orig_data);
> +}
> +
> +static int dmz_npo2_report_zones(struct dm_target *ti,
> +				 struct dm_report_zones_args *args,
> +				 unsigned int nr_zones)
> +{
> +	struct dmz_npo2_target *dmh = ti->private;
> +	int ret = 0;

no need for the = 0. No need for ret at all in fact.

> +	sector_t sect = po2_zone_no(dmh, args->next_sector) * dmh->zsze;
> +
> +	ret = blkdev_report_zones(dmh->dev->bdev, sect, nr_zones,
> +				  dmz_npo2_report_zones_cb, args);
> +	if (ret < 0)
> +		DMERR("report zones error");

Not useful. just "return blkdev_report_zones();"

> +
> +	return ret;
> +}
> +
> +static int check_zone_boundary_violation(struct dmz_npo2_target *dmh,
> +					 sector_t sect, sector_t size)
> +{
> +	u32 zone_idx = po2_zone_no(dmh, sect);
> +	sector_t relative_sect = 0;

No need for "= 0".

> +
> +	sect = target_to_device_sect(dmh, sect);
> +	relative_sect = sect - (zone_idx * dmh->zsze);

ofst_in_zone ? or sect_osft ?

> +
> +	if ((relative_sect + size) <= dmh->zsze)

no need for the inner brackets.

> +		return DMZ_NPO2_IO_INSIDE_ZONE;
> +	else if (relative_sect >= dmh->zsze)

no need for the else. And this is super confusing. This case correspond to
the BIO going beyond the zone capacity in the target address space,
meaning it is still WITHIN the target zone. But you call that "outside"
because it is for the device zone. Super confusing. It took me a lot of
rereading to finally get it.

> +		return DMZ_NPO2_IO_OUTSIDE_ZONE;
> +
> +	return DMZ_NPO2_IO_ACROSS_ZONE;

So you BIO is eeither fully contained within the zone or it is not. So why
not just return a bool ?

> +}
> +
> +static void split_io_across_zone_boundary(struct dmz_npo2_target *dmh,
> +					  struct bio *bio)
> +{
> +	sector_t sect = bio->bi_iter.bi_sector;
> +	sector_t sects_from_zone_start;
> +
> +	sect = target_to_device_sect(dmh, sect);

	sect = target_to_device_sect(dmh, bio->bi_iter.bi_sector);

is more readable.

> +	div64_u64_rem(sect, dmh->zsze, &sects_from_zone_start);
> +	dm_accept_partial_bio(bio, dmh->zsze - sects_from_zone_start);

So if this is a read BIO starting exactly at the target zone capacity
(sects_from_zone_start == zsze), then you accept 0 sectors ? What am I
missing here ?

> +	bio->bi_iter.bi_sector = sect;
> +}
> +
> +static int handle_zone_boundary_violation(struct dmz_npo2_target *dmh,
> +					  struct bio *bio,
> +					  enum dmz_npo2_io_cond cond)
> +{
> +	/* Read should return zeroed page */
> +	if (bio_op(bio) == REQ_OP_READ) {
> +		if (cond == DMZ_NPO2_IO_ACROSS_ZONE) {
> +			split_io_across_zone_boundary(dmh, bio);
> +			return DM_MAPIO_REMAPPED;
> +		}
> +		zero_fill_bio(bio);
> +		bio_endio(bio);
> +		return DM_MAPIO_SUBMITTED;
> +	}
> +	return DM_MAPIO_KILL;
> +}
> +
> +static int dmz_npo2_end_io(struct dm_target *ti, struct bio *bio,
> +			   blk_status_t *error)
> +{
> +	struct dmz_npo2_target *dmh = ti->private;
> +
> +	if (bio->bi_status == BLK_STS_OK && bio_op(bio) == REQ_OP_ZONE_APPEND)
> +		bio->bi_iter.bi_sector =
> +			device_to_target_sect(dmh, bio->bi_iter.bi_sector);
> +
> +	return DM_ENDIO_DONE;
> +}
> +
> +static int dmz_npo2_map(struct dm_target *ti, struct bio *bio)
> +{
> +	struct dmz_npo2_target *dmh = ti->private;
> +	enum dmz_npo2_io_cond cond;
> +
> +	bio_set_dev(bio, dmh->dev->bdev);
> +	if (bio_sectors(bio) || op_is_zone_mgmt(bio_op(bio))) {
> +		cond = check_zone_boundary_violation(dmh, bio->bi_iter.bi_sector,
> +						     bio->bi_iter.bi_size >> SECTOR_SHIFT);

Why check this for zone management BIOs ? These have length = 0, always.

> +
> +		/*
> +		 * If the starting sector is in the emulated area then fill
> +		 * all the bio with zeros. If bio is across boundaries,
> +		 * split the bio across boundaries and fill zeros only for the
> +		 * bio that is outside the zone capacity
> +		 */
> +		switch (cond) {
> +		case DMZ_NPO2_IO_INSIDE_ZONE:
> +			bio->bi_iter.bi_sector = target_to_device_sect(dmh,
> +								       bio->bi_iter.bi_sector);
> +			break;
> +		case DMZ_NPO2_IO_ACROSS_ZONE:
> +		case DMZ_NPO2_IO_OUTSIDE_ZONE:
> +			return handle_zone_boundary_violation(dmh, bio, cond);
> +		}
> +	}
> +	return DM_MAPIO_REMAPPED;

This entire function is very hard to read because everything is hidden in
helpers that are not super useful in my opinion. I would prefer seeing
cases for:
* zone management BIOs
* Reads and writes
* Everything else

where tests against the bio sector and length are visible, so one can
understand what is going on. If you need helpers, have handle_zone_mgmt(),
handle_read() etc. Something clear.

> +}
> +
> +static int dmz_npo2_iterate_devices(struct dm_target *ti,
> +				    iterate_devices_callout_fn fn, void *data)
> +{
> +	struct dmz_npo2_target *dmh = ti->private;
> +	sector_t len = 0;
> +
> +	len = dmh->nr_zones * dmh->zsze;

Move this to the declaration instead of setting len to 0 for nothing.

> +	return fn(ti, dmh->dev, 0, len, data);
> +}
> +
> +static struct target_type dmz_npo2_target = {
> +	.name = "zoned-npo2",
> +	.version = { 1, 0, 0 },
> +	.features = DM_TARGET_ZONED_HM,
> +	.map = dmz_npo2_map,
> +	.end_io = dmz_npo2_end_io,
> +	.report_zones = dmz_npo2_report_zones,
> +	.iterate_devices = dmz_npo2_iterate_devices,
> +	.module = THIS_MODULE,
> +	.ctr = dmz_npo2_ctr,
> +};
> +
> +static int __init dmz_npo2_init(void)
> +{
> +	int r = dm_register_target(&dmz_npo2_target);
> +
> +	if (r < 0)
> +		DMERR("register failed %d", r);
> +
> +	return r;
> +}
> +
> +static void __exit dmz_npo2_exit(void)
> +{
> +	dm_unregister_target(&dmz_npo2_target);
> +}
> +
> +/* Module hooks */
> +module_init(dmz_npo2_init);
> +module_exit(dmz_npo2_exit);
> +
> +MODULE_DESCRIPTION(DM_NAME " non power 2 zoned target");
> +MODULE_AUTHOR("Pankaj Raghav <[e-mail]>");
> +MODULE_LICENSE("GPL");
> +

kernel test robot June 15, 2022, 2:19 p.m. UTC | #2

Hi Pankaj,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on axboe-block/for-next]
[also build test WARNING on device-mapper-dm/for-next linus/master v5.19-rc2 next-20220615]
[cannot apply to song-md/md-next]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/intel-lab-lkp/linux/commits/Pankaj-Raghav/block-make-blkdev_nr_zones-and-blk_queue_zone_no-generic-for-npo2-zsze/20220615-191927
base:   https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux-block.git for-next
config: s390-allmodconfig (https://download.01.org/0day-ci/archive/20220615/[e-mail]/config)
compiler: s390-linux-gcc (GCC) 11.3.0
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # https://github.com/intel-lab-lkp/linux/commit/add4ab54d5b34d4a2f91f241007f23a56c164fb3
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Pankaj-Raghav/block-make-blkdev_nr_zones-and-blk_queue_zone_no-generic-for-npo2-zsze/20220615-191927
        git checkout add4ab54d5b34d4a2f91f241007f23a56c164fb3
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.3.0 make.cross W=1 O=build_dir ARCH=s390 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
Reported-by: kernel test robot <[e-mail]>

All warnings (new ones prefixed by >>):

   drivers/md/dm-zoned-npo2-target.c: In function 'dmz_npo2_ctr':
>> drivers/md/dm-zoned-npo2-target.c:62:13: warning: variable 'ret' set but not used [-Wunused-but-set-variable]
      62 |         int ret = 0;
         |             ^~~


vim +/ret +62 drivers/md/dm-zoned-npo2-target.c

    53	
    54	/*
    55	 * <dev-path>
    56	 * This target works on the complete zoned device. Partial mapping is not
    57	 * supported
    58	 */
    59	static int dmz_npo2_ctr(struct dm_target *ti, unsigned int argc, char **argv)
    60	{
    61		struct dmz_npo2_target *dmh = NULL;
  > 62		int ret = 0;
    63		sector_t zsze;
    64		sector_t disk_size;
    65	
    66		if (argc < 1)
    67			return -EINVAL;
    68	
    69		dmh = kmalloc(sizeof(*dmh), GFP_KERNEL);
    70		if (!dmh)
    71			return -ENOMEM;
    72	
    73		ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
    74				    &dmh->dev);
    75	
    76		zsze = blk_queue_zone_sectors(bdev_get_queue(dmh->dev->bdev));
    77	
    78		disk_size = get_capacity(dmh->dev->bdev->bd_disk);
    79	
    80		if (ti->len != disk_size || ti->begin) {
    81			DMERR("%pg Partial mapping of the target not supported",
    82			      dmh->dev->bdev);
    83			return -EINVAL;
    84		}
    85	
    86		if (is_power_of_2(zsze)) {
    87			DMERR("%pg zone size is power of 2", dmh->dev->bdev);
    88			return -EINVAL;
    89		}
    90	
    91		dmh->zsze = zsze;
    92		dmh->zsze_po2 = 1 << get_count_order_long(zsze);
    93		dmh->zsze_diff = dmh->zsze_po2 - dmh->zsze;
    94	
    95		ti->private = dmh;
    96		ti->num_flush_bios = 1;
    97		ti->num_discard_bios = 1;
    98		ti->num_secure_erase_bios = 1;
    99		ti->num_write_zeroes_bios = 1;
   100	
   101		dmh->nr_zones = npo2_zone_no(dmh, ti->len);
   102		ti->len = dmh->zsze_po2 * dmh->nr_zones;
   103	
   104		return 0;
   105	}
   106

Randy Dunlap June 15, 2022, 7:54 p.m. UTC | #3

Hi--

On 6/15/22 03:19, Pankaj Raghav wrote:

> ---
>  drivers/md/Kconfig                |   9 +
>  drivers/md/Makefile               |   2 +
>  drivers/md/dm-zone.c              |   9 +
>  drivers/md/dm-zoned-npo2-target.c | 268 ++++++++++++++++++++++++++++++
>  4 files changed, 288 insertions(+)
>  create mode 100644 drivers/md/dm-zoned-npo2-target.c
> 
> diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
> index 998a5cfdb..773314536 100644
> --- a/drivers/md/Kconfig
> +++ b/drivers/md/Kconfig
> @@ -518,6 +518,15 @@ config DM_FLAKEY
>  	help
>  	 A target that intermittently fails I/O for debugging purposes.
>  
> +config DM_ZONED_NPO2
> +	tristate "Zoned non power of 2 target"

	         "Zoned non-power-of-2 target"

> +	depends on BLK_DEV_DM
> +	depends on BLK_DEV_ZONED
> +	help
> +	A target that converts a zoned device with non power of 2 zone size to

	                                           non-power-of-2 zone size to

> +	be power of 2. This is done by introducing gaps in between the zone
> +	capacity and the power of 2 zone size.

All help text should be indented with one tab and 2 spaces
according to Documentation/process/coding-style.rst.

> +
>  config DM_VERITY
>  	tristate "Verity target support"
>  	depends on BLK_DEV_DM

[snip]

> +
> +MODULE_DESCRIPTION(DM_NAME " non power 2 zoned target");

                                non-power-of-2

> +MODULE_AUTHOR("Pankaj Raghav <[e-mail]>");
> +MODULE_LICENSE("GPL");
> +

Pankaj Raghav June 16, 2022, 10:28 a.m. UTC | #4

On 2022-06-15 21:54, Randy Dunlap wrote:
> Hi--
> 
> On 6/15/22 03:19, Pankaj Raghav wrote:
> 
>> ---
>>  drivers/md/Kconfig                |   9 +
>>  drivers/md/Makefile               |   2 +
>>  drivers/md/dm-zone.c              |   9 +
>>  drivers/md/dm-zoned-npo2-target.c | 268 ++++++++++++++++++++++++++++++
>>  4 files changed, 288 insertions(+)
>>  create mode 100644 drivers/md/dm-zoned-npo2-target.c
>>
>> diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
>> index 998a5cfdb..773314536 100644
>> --- a/drivers/md/Kconfig
>> +++ b/drivers/md/Kconfig
>> @@ -518,6 +518,15 @@ config DM_FLAKEY
>>  	help
>>  	 A target that intermittently fails I/O for debugging purposes.
>>  
>> +config DM_ZONED_NPO2
>> +	tristate "Zoned non power of 2 target"
> 
> 	         "Zoned non-power-of-2 target"
> 
>> +	depends on BLK_DEV_DM
>> +	depends on BLK_DEV_ZONED
>> +	help
>> +	A target that converts a zoned device with non power of 2 zone size to
> 
> 	                                           non-power-of-2 zone size to
> 
>> +	be power of 2. This is done by introducing gaps in between the zone
>> +	capacity and the power of 2 zone size.
> 
> All help text should be indented with one tab and 2 spaces
> according to Documentation/process/coding-style.rst.
> 
>> +
>>  config DM_VERITY
>>  	tristate "Verity target support"
>>  	depends on BLK_DEV_DM
> 
> [snip]
> 
>> +
>> +MODULE_DESCRIPTION(DM_NAME " non power 2 zoned target");
> 
>                                 non-power-of-2
> 
>> +MODULE_AUTHOR("Pankaj Raghav <[e-mail]>");
>> +MODULE_LICENSE("GPL");
>> +
> 
Thanks Randy. Fixed.

Pankaj Raghav June 16, 2022, 4:12 p.m. UTC | #5

Hi Damien,
On 2022-06-15 13:49, Damien Le Moal wrote:
> On 6/15/22 19:19, Pankaj Raghav wrote:
>> Only power of 2(po2) zoned devices were supported in linux but now non
>> power of 2(npo2) zoned device support has been added to the block layer.
>>
>> Filesystems such as F2FS and btrfs have support for zoned devices with
>> po2 zone size assumption. Before adding native support for npo2 zoned
>> devices, it was suggested to create a dm target for npo2 zoned device to
>> appear as po2 device so that file systems can initially work without any
>> explicit changes by using this target.
>>
>> The design of this target is very simple: introduce gaps between the zone
>> capacity and the po2 zone size of the underlying device. All IOs will be
>> remapped from target to the actual device location. For devices that use
>> zone append, the bi_sector is remapped from device to target's layout.
> 
> Nothing special for zone append in this respect. All IOs are remapped
> likewise, right ?
> 
This is what is being done: when we submit, we adjust the sector value
from target to device, and the actual sector value from bio gets updated
in the endio function where we transform from device -> target for zone
appends.
>>
>> The read IOs that fall in the "emulated" gap area will return 0 and all
>> the other IOs in that area will result in an error. If an read IO span
>> across the zone capacity boundary, then the IOs are split between the
>> boundary. All other IO operations that span across a zone capacity
>> boundary will result in an error.
>>
>> The target can be easily updated as follows:
> 
> Updated ? you mean created, no ?
> 
Yeah. I will fix it.
>> dmsetup create <label> --table '0 <size_sects> zoned-npo2 /dev/nvme<id>'
>>
>> Signed-off-by: Pankaj Raghav <[e-mail]>
>> Suggested-by: Johannes Thumshirn <[e-mail]>
>> Suggested-by: Damien Le Moal <[e-mail]>
>> Suggested-by: Hannes Reinecke <[e-mail]>
>> ---
>>  drivers/md/Kconfig                |   9 +
>>  drivers/md/Makefile               |   2 +
>>  drivers/md/dm-zone.c              |   9 +
>>  drivers/md/dm-zoned-npo2-target.c | 268 ++++++++++++++++++++++++++++++
>>  4 files changed, 288 insertions(+)
>>  create mode 100644 drivers/md/dm-zoned-npo2-target.c
>>
>> diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
>> index 998a5cfdb..773314536 100644
>> --- a/drivers/md/Kconfig
>> +++ b/drivers/md/Kconfig
>> @@ -518,6 +518,15 @@ config DM_FLAKEY
>>  	help
>>  	 A target that intermittently fails I/O for debugging purposes.
>>  
>> +config DM_ZONED_NPO2
>> +	tristate "Zoned non power of 2 target"
>> +	depends on BLK_DEV_DM
>> +	depends on BLK_DEV_ZONED
>> +	help
>> +	A target that converts a zoned device with non power of 2 zone size to
>> +	be power of 2. This is done by introducing gaps in between the zone
>> +	capacity and the power of 2 zone size.
>> +
>>  config DM_VERITY
>>  	tristate "Verity target support"
>>  	depends on BLK_DEV_DM
>> diff --git a/drivers/md/Makefile b/drivers/md/Makefile
>> index 0454b0885..2863a94a7 100644
>> --- a/drivers/md/Makefile
>> +++ b/drivers/md/Makefile
>> @@ -26,6 +26,7 @@ dm-era-y	+= dm-era-target.o
>>  dm-clone-y	+= dm-clone-target.o dm-clone-metadata.o
>>  dm-verity-y	+= dm-verity-target.o
>>  dm-zoned-y	+= dm-zoned-target.o dm-zoned-metadata.o dm-zoned-reclaim.o
>> +dm-zoned-npo2-y       += dm-zoned-npo2-target.o
> 
> This naming is in my opinion very bad as it seems related to the dm-zoned
> target. e.g. dm-po2z, dm-zp2, etc.
> 
Probably dm-po2z sounds good. I will go for it if others don't have any
objection.
>>  
>>  md-mod-y	+= md.o md-bitmap.o
>>  raid456-y	+= raid5.o raid5-cache.o raid5-ppl.o
>> @@ -60,6 +61,7 @@ obj-$(CONFIG_DM_CRYPT)		+= dm-crypt.o
>>  obj-$(CONFIG_DM_DELAY)		+= dm-delay.o
>>  obj-$(CONFIG_DM_DUST)		+= dm-dust.o
>>  obj-$(CONFIG_DM_FLAKEY)		+= dm-flakey.o
>> +obj-$(CONFIG_DM_ZONED_NPO2)	+= dm-zoned-npo2.o
>>  obj-$(CONFIG_DM_MULTIPATH)	+= dm-multipath.o dm-round-robin.o
>>  obj-$(CONFIG_DM_MULTIPATH_QL)	+= dm-queue-length.o
>>  obj-$(CONFIG_DM_MULTIPATH_ST)	+= dm-service-time.o
>> diff --git a/drivers/md/dm-zone.c b/drivers/md/dm-zone.c
>> index af36d33f9..5efb31ba0 100644
>> --- a/drivers/md/dm-zone.c
>> +++ b/drivers/md/dm-zone.c
>> @@ -210,6 +210,11 @@ static int dm_zone_revalidate_cb(struct blk_zone *zone, unsigned int idx,
>>  		}
>>  		md->zwp_offset[idx] = dm_get_zone_wp_offset(zone);
>>  
>> +		if (q->limits.chunk_sectors != zone->len) {
> 
> Why is this if needed ?
> 
Explanation below.
>> +			blk_queue_chunk_sectors(q, zone->len);
>> +			q->nr_zones = blkdev_nr_zones(md->disk);
>> +		}
>> +
>>  		break;
>>  	default:
>>  		DMERR("Invalid zone type 0x%x at sectors %llu",
>> @@ -307,6 +312,9 @@ int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
>>  	if (dm_table_supports_zone_append(t)) {
>>  		clear_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
>>  		dm_cleanup_zoned_dev(md);
>> +
> 
> no need for the blank line.
> 
>> +		if (!is_power_of_2(blk_queue_zone_sectors(q)))
>> +			goto revalidate_zones;
>>  		return 0;
>>  	}
> 
> Why do you need to change dm_set_zones_restrictions() at all ?
> 
When the device mapper is created, the q->limits gets inherited from the
underlying device. The chunk sectors of the target and the device will
be the same but we want the chunk sector of the target to be different
(rounded to po2) compared to the underlying device's chunk sector. This
needs to be done only for the dm-po2z target and not for other targets
that uses npo2 zoned devices (like dm-linear). So to perform this
operation in a target independent way in dm-zone.c, I chose to always
revalidate npo2 zoned device and update the chunk sector and nr_zones in
dm_zone_revalidate_cb based on the zone information from the target.
This allows to set the limits correctly for dm-po2z target.
>>  
<snip>
return -EINVAL;
>> +	}
>> +
>> +	if (is_power_of_2(zsze)) {
>> +		DMERR("%pg zone size is power of 2", dmh->dev->bdev);
> 
> Hmmm... You would end up with no remapping needed so it would still
> work... Why error this ? A warning would work too.
> 
You mean a DMWARN and not return -EINVAL? I mean there is no usecase for
po2 device to use this target so why allow it in the first place?
>> +		return -EINVAL;
>> +	}
>> +
>> +	dmh->zsze = zsze;
>> +	dmh->zsze_po2 = 1 << get_count_order_long(zsze);
>> +	dmh->zsze_diff = dmh->zsze_po2 - dmh->zsze;
>> +
>> +	ti->private = dmh;
>> +	ti->num_flush_bios = 1;
>> +	ti->num_discard_bios = 1;
>> +	ti->num_secure_erase_bios = 1;
>> +	ti->num_write_zeroes_bios = 1;
> 
> Why all these ? I know dm-linear do that but I do not see why they would
> be necessary for a single device target.
> 
Good point. I will remove them

<snip>
>> +		return DMZ_NPO2_IO_INSIDE_ZONE;
>> +	else if (relative_sect >= dmh->zsze)
> 
> no need for the else. And this is super confusing. This case correspond to
> the BIO going beyond the zone capacity in the target address space,
> meaning it is still WITHIN the target zone. But you call that "outside"
> because it is for the device zone. Super confusing. It took me a lot of
> rereading to finally get it.
> 
Probably my naming choice was not correct here for the enum. It should
be s/IO_INSIDE_ZONE/IO_INSIDE_ZONE_CAP, etc to be clear. I mainly wanted
to handle the case where a read across zone capacity should return
something valid instead of just an error as we emulate the LBAs from
zone cap to zone size.

DMZ_NPO2_IO_INSIDE_ZONE_CAP:
             zcap   zsize
---------------|-----|
      <------>
        bio
Normal scenario we just send what is there in the device.


DMZ_NPO2_IO_OUTSIDE_ZONE_CAP:
             zcap       zsize
---------------|---------|
                 <---->
                   bio

Read should return zero filled bio and other operation will return an
error because we are touching the emulated area.

DMZ_NPO2_IO_ACROSS_ZONE_CAP:
             zcap   zsize
---------------|-----|
           <------>
              bio
For reads, the bio should be split across zone cap and the bio on the
left hand side returns what is there in the device and the split bio on
the right hand side should just return zeroes. All other requests will
return an error.

>> +		return DMZ_NPO2_IO_OUTSIDE_ZONE;
>> +
>> +	return DMZ_NPO2_IO_ACROSS_ZONE;
> 
> So you BIO is eeither fully contained within the zone or it is not. So why
> not just return a bool ?
> 
As I explained above, I was considering the boundary as zone cap inside
a zone. The bio can be within zone cap, across zone cap into the
emulated zone size and outside zone capacity.

I didn't take into account the read across zone. I will make sure it is
correctly handled in the next revision.

             zcap  zsize          zcap
---------------|-----|--------------|
                  <------>
                     bio
>> +}
>> +
>> +static void split_io_across_zone_boundary(struct dmz_npo2_target *dmh,
>> +					  struct bio *bio)
>> +{
>> +	sector_t sect = bio->bi_iter.bi_sector;
>> +	sector_t sects_from_zone_start;
>> +
>> +	sect = target_to_device_sect(dmh, sect);
> 
> 	sect = target_to_device_sect(dmh, bio->bi_iter.bi_sector);
> 
> is more readable.
> 
>> +	div64_u64_rem(sect, dmh->zsze, &sects_from_zone_start);
>> +	dm_accept_partial_bio(bio, dmh->zsze - sects_from_zone_start);
> 
> So if this is a read BIO starting exactly at the target zone capacity
> (sects_from_zone_start == zsze), then you accept 0 sectors ? What am I
> missing here ?
> 
Your condition will not even touch this function. This function comes
into play only when the bio runs across the zone capacity as I mentioned
before.
>> +	bio->bi_iter.bi_sector = sect;
>> +}
>> +
>> +static int handle_zone_boundary_violation(struct dmz_npo2_target *dmh,
>> +					  struct bio *bio,
>> +					  enum dmz_npo2_io_cond cond)
>> +{
>> +	/* Read should return zeroed page */
>> +	if (bio_op(bio) == REQ_OP_READ) {
>> +		if (cond == DMZ_NPO2_IO_ACROSS_ZONE) {
>> +			split_io_across_zone_boundary(dmh, bio);
>> +			return DM_MAPIO_REMAPPED;
>> +		}
>> +		zero_fill_bio(bio);
>> +		bio_endio(bio);
>> +		return DM_MAPIO_SUBMITTED;
>> +	}
>> +	return DM_MAPIO_KILL;
>> +}
>> +
>> +static int dmz_npo2_end_io(struct dm_target *ti, struct bio *bio,
>> +			   blk_status_t *error)
>> +{
>> +	struct dmz_npo2_target *dmh = ti->private;
>> +
>> +	if (bio->bi_status == BLK_STS_OK && bio_op(bio) == REQ_OP_ZONE_APPEND)
>> +		bio->bi_iter.bi_sector =
>> +			device_to_target_sect(dmh, bio->bi_iter.bi_sector);
>> +
>> +	return DM_ENDIO_DONE;
>> +}
>> +
>> +static int dmz_npo2_map(struct dm_target *ti, struct bio *bio)
>> +{
>> +	struct dmz_npo2_target *dmh = ti->private;
>> +	enum dmz_npo2_io_cond cond;
>> +
>> +	bio_set_dev(bio, dmh->dev->bdev);
>> +	if (bio_sectors(bio) || op_is_zone_mgmt(bio_op(bio))) {
>> +		cond = check_zone_boundary_violation(dmh, bio->bi_iter.bi_sector,
>> +						     bio->bi_iter.bi_size >> SECTOR_SHIFT);
> 
> Why check this for zone management BIOs ? These have length = 0, always.
> 
>> +
>> +		/*
>> +		 * If the starting sector is in the emulated area then fill
>> +		 * all the bio with zeros. If bio is across boundaries,
>> +		 * split the bio across boundaries and fill zeros only for the
>> +		 * bio that is outside the zone capacity
>> +		 */
>> +		switch (cond) {
>> +		case DMZ_NPO2_IO_INSIDE_ZONE:
>> +			bio->bi_iter.bi_sector = target_to_device_sect(dmh,
>> +								       bio->bi_iter.bi_sector);
>> +			break;
>> +		case DMZ_NPO2_IO_ACROSS_ZONE:
>> +		case DMZ_NPO2_IO_OUTSIDE_ZONE:
>> +			return handle_zone_boundary_violation(dmh, bio, cond);
>> +		}
>> +	}
>> +	return DM_MAPIO_REMAPPED;
> 
> This entire function is very hard to read because everything is hidden in
> helpers that are not super useful in my opinion. I would prefer seeing
> cases for:
> * zone management BIOs
> * Reads and writes
> * Everything else
> 
> where tests against the bio sector and length are visible, so one can
> understand what is going on. If you need helpers, have handle_zone_mgmt(),
> handle_read() etc. Something clear.
> 
Got it. I see the confusion here. I will rearrange it properly in the
next revision. Thanks for this comment.
>> +}
>> +
>> +static int dmz_npo2_iterate_devices(struct dm_target *ti,
>> +				    iterate_devices_callout_fn fn, void *data)
>> +{
>> +	struct dmz_npo2_target *dmh = ti->private;
>> +	sector_t len = 0;
>> +
>> +	len = dmh->nr_zones * dmh->zsze;
> 
> Move this to the declaration instead of setting len to 0 for nothing.
> 
Ok.
>> +	return fn(ti, dmh->dev, 0, len, data);

Damien Le Moal June 16, 2022, 11:49 p.m. UTC | #6

On 6/17/22 01:12, Pankaj Raghav wrote:
> Hi Damien,
> On 2022-06-15 13:49, Damien Le Moal wrote:
>> On 6/15/22 19:19, Pankaj Raghav wrote:
>>> Only power of 2(po2) zoned devices were supported in linux but now non
>>> power of 2(npo2) zoned device support has been added to the block layer.
>>>
>>> Filesystems such as F2FS and btrfs have support for zoned devices with
>>> po2 zone size assumption. Before adding native support for npo2 zoned
>>> devices, it was suggested to create a dm target for npo2 zoned device to
>>> appear as po2 device so that file systems can initially work without any
>>> explicit changes by using this target.
>>>
>>> The design of this target is very simple: introduce gaps between the zone
>>> capacity and the po2 zone size of the underlying device. All IOs will be
>>> remapped from target to the actual device location. For devices that use
>>> zone append, the bi_sector is remapped from device to target's layout.
>>
>> Nothing special for zone append in this respect. All IOs are remapped
>> likewise, right ?
>>
> This is what is being done: when we submit, we adjust the sector value
> from target to device, and the actual sector value from bio gets updated
> in the endio function where we transform from device -> target for zone
> appends.

I know all that. This was a remark intended at pointing out that this
commit message statement does not have any value, it does not help in
understanding any peculiarity of this target driver (if any). It seems
targeted at zone append only. Reword this to explain the remapping for all
IOs, and that zone management request and report zones also need remapping.

>>>
>>> The read IOs that fall in the "emulated" gap area will return 0 and all
>>> the other IOs in that area will result in an error. If an read IO span
>>> across the zone capacity boundary, then the IOs are split between the
>>> boundary. All other IO operations that span across a zone capacity
>>> boundary will result in an error.
>>>
>>> The target can be easily updated as follows:
>>
>> Updated ? you mean created, no ?
>>
> Yeah. I will fix it.
>>> dmsetup create <label> --table '0 <size_sects> zoned-npo2 /dev/nvme<id>'
>>>
>>> Signed-off-by: Pankaj Raghav <[e-mail]>
>>> Suggested-by: Johannes Thumshirn <[e-mail]>
>>> Suggested-by: Damien Le Moal <[e-mail]>
>>> Suggested-by: Hannes Reinecke <[e-mail]>
>>> ---
>>>  drivers/md/Kconfig                |   9 +
>>>  drivers/md/Makefile               |   2 +
>>>  drivers/md/dm-zone.c              |   9 +
>>>  drivers/md/dm-zoned-npo2-target.c | 268 ++++++++++++++++++++++++++++++
>>>  4 files changed, 288 insertions(+)
>>>  create mode 100644 drivers/md/dm-zoned-npo2-target.c
>>>
>>> diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
>>> index 998a5cfdb..773314536 100644
>>> --- a/drivers/md/Kconfig
>>> +++ b/drivers/md/Kconfig
>>> @@ -518,6 +518,15 @@ config DM_FLAKEY
>>>  	help
>>>  	 A target that intermittently fails I/O for debugging purposes.
>>>  
>>> +config DM_ZONED_NPO2
>>> +	tristate "Zoned non power of 2 target"
>>> +	depends on BLK_DEV_DM
>>> +	depends on BLK_DEV_ZONED
>>> +	help
>>> +	A target that converts a zoned device with non power of 2 zone size to
>>> +	be power of 2. This is done by introducing gaps in between the zone
>>> +	capacity and the power of 2 zone size.
>>> +
>>>  config DM_VERITY
>>>  	tristate "Verity target support"
>>>  	depends on BLK_DEV_DM
>>> diff --git a/drivers/md/Makefile b/drivers/md/Makefile
>>> index 0454b0885..2863a94a7 100644
>>> --- a/drivers/md/Makefile
>>> +++ b/drivers/md/Makefile
>>> @@ -26,6 +26,7 @@ dm-era-y	+= dm-era-target.o
>>>  dm-clone-y	+= dm-clone-target.o dm-clone-metadata.o
>>>  dm-verity-y	+= dm-verity-target.o
>>>  dm-zoned-y	+= dm-zoned-target.o dm-zoned-metadata.o dm-zoned-reclaim.o
>>> +dm-zoned-npo2-y       += dm-zoned-npo2-target.o
>>
>> This naming is in my opinion very bad as it seems related to the dm-zoned
>> target. e.g. dm-po2z, dm-zp2, etc.
>>
> Probably dm-po2z sounds good. I will go for it if others don't have any
> objection.
>>>  
>>>  md-mod-y	+= md.o md-bitmap.o
>>>  raid456-y	+= raid5.o raid5-cache.o raid5-ppl.o
>>> @@ -60,6 +61,7 @@ obj-$(CONFIG_DM_CRYPT)		+= dm-crypt.o
>>>  obj-$(CONFIG_DM_DELAY)		+= dm-delay.o
>>>  obj-$(CONFIG_DM_DUST)		+= dm-dust.o
>>>  obj-$(CONFIG_DM_FLAKEY)		+= dm-flakey.o
>>> +obj-$(CONFIG_DM_ZONED_NPO2)	+= dm-zoned-npo2.o
>>>  obj-$(CONFIG_DM_MULTIPATH)	+= dm-multipath.o dm-round-robin.o
>>>  obj-$(CONFIG_DM_MULTIPATH_QL)	+= dm-queue-length.o
>>>  obj-$(CONFIG_DM_MULTIPATH_ST)	+= dm-service-time.o
>>> diff --git a/drivers/md/dm-zone.c b/drivers/md/dm-zone.c
>>> index af36d33f9..5efb31ba0 100644
>>> --- a/drivers/md/dm-zone.c
>>> +++ b/drivers/md/dm-zone.c
>>> @@ -210,6 +210,11 @@ static int dm_zone_revalidate_cb(struct blk_zone *zone, unsigned int idx,
>>>  		}
>>>  		md->zwp_offset[idx] = dm_get_zone_wp_offset(zone);
>>>  
>>> +		if (q->limits.chunk_sectors != zone->len) {
>>
>> Why is this if needed ?
>>
> Explanation below.
>>> +			blk_queue_chunk_sectors(q, zone->len);
>>> +			q->nr_zones = blkdev_nr_zones(md->disk);
>>> +		}
>>> +
>>>  		break;
>>>  	default:
>>>  		DMERR("Invalid zone type 0x%x at sectors %llu",
>>> @@ -307,6 +312,9 @@ int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
>>>  	if (dm_table_supports_zone_append(t)) {
>>>  		clear_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
>>>  		dm_cleanup_zoned_dev(md);
>>> +
>>
>> no need for the blank line.
>>
>>> +		if (!is_power_of_2(blk_queue_zone_sectors(q)))
>>> +			goto revalidate_zones;
>>>  		return 0;
>>>  	}
>>
>> Why do you need to change dm_set_zones_restrictions() at all ?
>>
> When the device mapper is created, the q->limits gets inherited from the
> underlying device. The chunk sectors of the target and the device will
> be the same but we want the chunk sector of the target to be different
> (rounded to po2) compared to the underlying device's chunk sector. This
> needs to be done only for the dm-po2z target and not for other targets
> that uses npo2 zoned devices (like dm-linear). So to perform this
> operation in a target independent way in dm-zone.c, I chose to always
> revalidate npo2 zoned device and update the chunk sector and nr_zones in
> dm_zone_revalidate_cb based on the zone information from the target.
> This allows to set the limits correctly for dm-po2z target.

But DM revalidate will be called for the target AFTER it is setup (after
its gendisk is added). So how can DM revalidate see the incorrect zone
size ? If that is the case, then the target constructor is broken or
missing something. DM revalidate zone is generic and only allocates the
zone bitmaps for the target device. There should be not need at all to
touch that function.

>>>  
> <snip>
> return -EINVAL;
>>> +	}
>>> +
>>> +	if (is_power_of_2(zsze)) {
>>> +		DMERR("%pg zone size is power of 2", dmh->dev->bdev);
>>
>> Hmmm... You would end up with no remapping needed so it would still
>> work... Why error this ? A warning would work too.
>>
> You mean a DMWARN and not return -EINVAL? I mean there is no usecase for
> po2 device to use this target so why allow it in the first place?

Why disallow it ? It will work. Either way is OK but I wanted to point out
that both path are I think equally valid.

>>> +		return -EINVAL;
>>> +	}
>>> +
>>> +	dmh->zsze = zsze;
>>> +	dmh->zsze_po2 = 1 << get_count_order_long(zsze);
>>> +	dmh->zsze_diff = dmh->zsze_po2 - dmh->zsze;
>>> +
>>> +	ti->private = dmh;
>>> +	ti->num_flush_bios = 1;
>>> +	ti->num_discard_bios = 1;
>>> +	ti->num_secure_erase_bios = 1;
>>> +	ti->num_write_zeroes_bios = 1;
>>
>> Why all these ? I know dm-linear do that but I do not see why they would
>> be necessary for a single device target.
>>
> Good point. I will remove them
> 
> <snip>
>>> +		return DMZ_NPO2_IO_INSIDE_ZONE;
>>> +	else if (relative_sect >= dmh->zsze)
>>
>> no need for the else. And this is super confusing. This case correspond to
>> the BIO going beyond the zone capacity in the target address space,
>> meaning it is still WITHIN the target zone. But you call that "outside"
>> because it is for the device zone. Super confusing. It took me a lot of
>> rereading to finally get it.
>>
> Probably my naming choice was not correct here for the enum. It should
> be s/IO_INSIDE_ZONE/IO_INSIDE_ZONE_CAP, etc to be clear. I mainly wanted
> to handle the case where a read across zone capacity should return
> something valid instead of just an error as we emulate the LBAs from
> zone cap to zone size.

Your target information fields are also badly named and make things hard
to understand.

> 
> DMZ_NPO2_IO_INSIDE_ZONE_CAP:
>              zcap   zsize
> ---------------|-----|
>       <------>
>         bio
> Normal scenario we just send what is there in the device.
> 
> 
> DMZ_NPO2_IO_OUTSIDE_ZONE_CAP:
>              zcap       zsize
> ---------------|---------|
>                  <---->
>                    bio
> 
> Read should return zero filled bio and other operation will return an
> error because we are touching the emulated area.
> 
> DMZ_NPO2_IO_ACROSS_ZONE_CAP:
>              zcap   zsize
> ---------------|-----|
>            <------>
>               bio
> For reads, the bio should be split across zone cap and the bio on the
> left hand side returns what is there in the device and the split bio on
> the right hand side should just return zeroes. All other requests will
> return an error.
> 
>>> +		return DMZ_NPO2_IO_OUTSIDE_ZONE;
>>> +
>>> +	return DMZ_NPO2_IO_ACROSS_ZONE;
>>
>> So you BIO is eeither fully contained within the zone or it is not. So why
>> not just return a bool ?
>>
> As I explained above, I was considering the boundary as zone cap inside
> a zone. The bio can be within zone cap, across zone cap into the
> emulated zone size and outside zone capacity.
> 
> I didn't take into account the read across zone. I will make sure it is
> correctly handled in the next revision.

Please test properly. This should have been caught in testing before sending.

> 
>              zcap  zsize          zcap
> ---------------|-----|--------------|
>                   <------>
>                      bio
>>> +}
>>> +
>>> +static void split_io_across_zone_boundary(struct dmz_npo2_target *dmh,
>>> +					  struct bio *bio)
>>> +{
>>> +	sector_t sect = bio->bi_iter.bi_sector;
>>> +	sector_t sects_from_zone_start;
>>> +
>>> +	sect = target_to_device_sect(dmh, sect);
>>
>> 	sect = target_to_device_sect(dmh, bio->bi_iter.bi_sector);
>>
>> is more readable.
>>
>>> +	div64_u64_rem(sect, dmh->zsze, &sects_from_zone_start);
>>> +	dm_accept_partial_bio(bio, dmh->zsze - sects_from_zone_start);
>>
>> So if this is a read BIO starting exactly at the target zone capacity
>> (sects_from_zone_start == zsze), then you accept 0 sectors ? What am I
>> missing here ?
>>
> Your condition will not even touch this function. This function comes
> into play only when the bio runs across the zone capacity as I mentioned
> before.
>>> +	bio->bi_iter.bi_sector = sect;
>>> +}
>>> +
>>> +static int handle_zone_boundary_violation(struct dmz_npo2_target *dmh,
>>> +					  struct bio *bio,
>>> +					  enum dmz_npo2_io_cond cond)
>>> +{
>>> +	/* Read should return zeroed page */
>>> +	if (bio_op(bio) == REQ_OP_READ) {
>>> +		if (cond == DMZ_NPO2_IO_ACROSS_ZONE) {
>>> +			split_io_across_zone_boundary(dmh, bio);
>>> +			return DM_MAPIO_REMAPPED;
>>> +		}
>>> +		zero_fill_bio(bio);
>>> +		bio_endio(bio);
>>> +		return DM_MAPIO_SUBMITTED;
>>> +	}
>>> +	return DM_MAPIO_KILL;
>>> +}
>>> +
>>> +static int dmz_npo2_end_io(struct dm_target *ti, struct bio *bio,
>>> +			   blk_status_t *error)
>>> +{
>>> +	struct dmz_npo2_target *dmh = ti->private;
>>> +
>>> +	if (bio->bi_status == BLK_STS_OK && bio_op(bio) == REQ_OP_ZONE_APPEND)
>>> +		bio->bi_iter.bi_sector =
>>> +			device_to_target_sect(dmh, bio->bi_iter.bi_sector);
>>> +
>>> +	return DM_ENDIO_DONE;
>>> +}
>>> +
>>> +static int dmz_npo2_map(struct dm_target *ti, struct bio *bio)
>>> +{
>>> +	struct dmz_npo2_target *dmh = ti->private;
>>> +	enum dmz_npo2_io_cond cond;
>>> +
>>> +	bio_set_dev(bio, dmh->dev->bdev);
>>> +	if (bio_sectors(bio) || op_is_zone_mgmt(bio_op(bio))) {
>>> +		cond = check_zone_boundary_violation(dmh, bio->bi_iter.bi_sector,
>>> +						     bio->bi_iter.bi_size >> SECTOR_SHIFT);
>>
>> Why check this for zone management BIOs ? These have length = 0, always.
>>
>>> +
>>> +		/*
>>> +		 * If the starting sector is in the emulated area then fill
>>> +		 * all the bio with zeros. If bio is across boundaries,
>>> +		 * split the bio across boundaries and fill zeros only for the
>>> +		 * bio that is outside the zone capacity
>>> +		 */
>>> +		switch (cond) {
>>> +		case DMZ_NPO2_IO_INSIDE_ZONE:
>>> +			bio->bi_iter.bi_sector = target_to_device_sect(dmh,
>>> +								       bio->bi_iter.bi_sector);
>>> +			break;
>>> +		case DMZ_NPO2_IO_ACROSS_ZONE:
>>> +		case DMZ_NPO2_IO_OUTSIDE_ZONE:
>>> +			return handle_zone_boundary_violation(dmh, bio, cond);
>>> +		}
>>> +	}
>>> +	return DM_MAPIO_REMAPPED;
>>
>> This entire function is very hard to read because everything is hidden in
>> helpers that are not super useful in my opinion. I would prefer seeing
>> cases for:
>> * zone management BIOs
>> * Reads and writes
>> * Everything else
>>
>> where tests against the bio sector and length are visible, so one can
>> understand what is going on. If you need helpers, have handle_zone_mgmt(),
>> handle_read() etc. Something clear.
>>
> Got it. I see the confusion here. I will rearrange it properly in the
> next revision. Thanks for this comment.
>>> +}
>>> +
>>> +static int dmz_npo2_iterate_devices(struct dm_target *ti,
>>> +				    iterate_devices_callout_fn fn, void *data)
>>> +{
>>> +	struct dmz_npo2_target *dmh = ti->private;
>>> +	sector_t len = 0;
>>> +
>>> +	len = dmh->nr_zones * dmh->zsze;
>>
>> Move this to the declaration instead of setting len to 0 for nothing.
>>
> Ok.
>>> +	return fn(ti, dmh->dev, 0, len, data);
>

Pankaj Raghav June 17, 2022, 5:45 a.m. UTC | #7

On 2022-06-17 01:49, Damien Le Moal wrote:
>>> Why do you need to change dm_set_zones_restrictions() at all ?
>>>
>> When the device mapper is created, the q->limits gets inherited from the
>> underlying device. The chunk sectors of the target and the device will
>> be the same but we want the chunk sector of the target to be different
>> (rounded to po2) compared to the underlying device's chunk sector. This
>> needs to be done only for the dm-po2z target and not for other targets
>> that uses npo2 zoned devices (like dm-linear). So to perform this
>> operation in a target independent way in dm-zone.c, I chose to always
>> revalidate npo2 zoned device and update the chunk sector and nr_zones in
>> dm_zone_revalidate_cb based on the zone information from the target.
>> This allows to set the limits correctly for dm-po2z target.
> 
> But DM revalidate will be called for the target AFTER it is setup (after
> its gendisk is added). So how can DM revalidate see the incorrect zone
> size ? If that is the case, then the target constructor is broken or
> missing something. DM revalidate zone is generic and only allocates the
> zone bitmaps for the target device. There should be not need at all to
> touch that function.
> 
I think this is a cleaner approach using features flag and io_hints
instead of messing with the revalidate zone function:

diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
index 135c0cc190fb..c97a71e0473f 100644
--- a/drivers/md/dm-table.c
+++ b/drivers/md/dm-table.c
@@ -1618,6 +1618,9 @@ static int device_not_matches_zone_sectors(struct
dm_target *ti, struct dm_dev *
 	if (!blk_queue_is_zoned(q))
 		return 0;

+	if(dm_target_supports_emulated_zone_size(ti->type))
+		return 0;
+
 	return blk_queue_zone_sectors(q) != *zone_sectors;
 }

diff --git a/drivers/md/dm-zoned-npo2-target.c
b/drivers/md/dm-zoned-npo2-target.c
index dad135964e09..b203be808f09 100644
--- a/drivers/md/dm-zoned-npo2-target.c
+++ b/drivers/md/dm-zoned-npo2-target.c
@@ -187,6 +187,12 @@ static int dmz_npo2_end_io(struct dm_target *ti,
struct bio *bio,
 	return DM_ENDIO_DONE;
 }

+static void dmz_npo2_io_hints(struct dm_target *ti, struct queue_limits
*limits)
+{
+	struct dmz_npo2_target *dmh = ti->private;
+	limits->chunk_sectors = dmh->zsze_po2;
+}
+
 static int dmz_npo2_map(struct dm_target *ti, struct bio *bio)
 {
 	struct dmz_npo2_target *dmh = ti->private;
@@ -233,12 +239,13 @@ static int dmz_npo2_iterate_devices(struct
dm_target *ti,
 static struct target_type dmz_npo2_target = {
 	.name = "zoned-npo2",
 	.version = { 1, 0, 0 },
-	.features = DM_TARGET_ZONED_HM,
+	.features = DM_TARGET_ZONED_HM | DM_TARGET_EMULATED_ZONE_SIZE,
 	.map = dmz_npo2_map,
 	.end_io = dmz_npo2_end_io,
 	.report_zones = dmz_npo2_report_zones,
 	.iterate_devices = dmz_npo2_iterate_devices,
 	.module = THIS_MODULE,
+	.io_hints = dmz_npo2_io_hints,
 	.ctr = dmz_npo2_ctr,
 };

diff --git a/include/linux/device-mapper.h b/include/linux/device-mapper.h
index c2a3758c4aaa..9f3a4d98a22a 100644
--- a/include/linux/device-mapper.h
+++ b/include/linux/device-mapper.h
@@ -283,6 +283,15 @@ struct target_type {
 #define dm_target_supports_mixed_zoned_model(type) (false)
 #endif

+#ifdef CONFIG_BLK_DEV_ZONED
+#define DM_TARGET_EMULATED_ZONE_SIZE	0x00000400
+#define dm_target_supports_emulated_zone_size(type) \
+	((type)->features & DM_TARGET_EMULATED_ZONE_SIZE)
+#else
+#define DM_TARGET_EMULATED_ZONE_SIZE	0x00000000
+#define dm_target_supports_emulated_zone_size(type) (false)
+#endif
+
 struct dm_target {
 	struct dm_table *table;
 	struct target_type *type;

Damien Le Moal June 17, 2022, 6:12 a.m. UTC | #8

On 6/17/22 14:45, Pankaj Raghav wrote:
> On 2022-06-17 01:49, Damien Le Moal wrote:
>>>> Why do you need to change dm_set_zones_restrictions() at all ?
>>>>
>>> When the device mapper is created, the q->limits gets inherited from the
>>> underlying device. The chunk sectors of the target and the device will
>>> be the same but we want the chunk sector of the target to be different
>>> (rounded to po2) compared to the underlying device's chunk sector. This
>>> needs to be done only for the dm-po2z target and not for other targets
>>> that uses npo2 zoned devices (like dm-linear). So to perform this
>>> operation in a target independent way in dm-zone.c, I chose to always
>>> revalidate npo2 zoned device and update the chunk sector and nr_zones in
>>> dm_zone_revalidate_cb based on the zone information from the target.
>>> This allows to set the limits correctly for dm-po2z target.
>>
>> But DM revalidate will be called for the target AFTER it is setup (after
>> its gendisk is added). So how can DM revalidate see the incorrect zone
>> size ? If that is the case, then the target constructor is broken or
>> missing something. DM revalidate zone is generic and only allocates the
>> zone bitmaps for the target device. There should be not need at all to
>> touch that function.
>>
> I think this is a cleaner approach using features flag and io_hints
> instead of messing with the revalidate zone function:
> 
> diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
> index 135c0cc190fb..c97a71e0473f 100644
> --- a/drivers/md/dm-table.c
> +++ b/drivers/md/dm-table.c
> @@ -1618,6 +1618,9 @@ static int device_not_matches_zone_sectors(struct
> dm_target *ti, struct dm_dev *
>  	if (!blk_queue_is_zoned(q))
>  		return 0;
> 
> +	if(dm_target_supports_emulated_zone_size(ti->type))
> +		return 0;
> +

This should be in validate_hardware_zoned_model(), not here.

>  	return blk_queue_zone_sectors(q) != *zone_sectors;
>  }
> 
> diff --git a/drivers/md/dm-zoned-npo2-target.c
> b/drivers/md/dm-zoned-npo2-target.c
> index dad135964e09..b203be808f09 100644
> --- a/drivers/md/dm-zoned-npo2-target.c
> +++ b/drivers/md/dm-zoned-npo2-target.c
> @@ -187,6 +187,12 @@ static int dmz_npo2_end_io(struct dm_target *ti,
> struct bio *bio,
>  	return DM_ENDIO_DONE;
>  }
> 
> +static void dmz_npo2_io_hints(struct dm_target *ti, struct queue_limits
> *limits)
> +{
> +	struct dmz_npo2_target *dmh = ti->private;
> +	limits->chunk_sectors = dmh->zsze_po2;
> +}
> +
>  static int dmz_npo2_map(struct dm_target *ti, struct bio *bio)
>  {
>  	struct dmz_npo2_target *dmh = ti->private;
> @@ -233,12 +239,13 @@ static int dmz_npo2_iterate_devices(struct
> dm_target *ti,
>  static struct target_type dmz_npo2_target = {
>  	.name = "zoned-npo2",
>  	.version = { 1, 0, 0 },
> -	.features = DM_TARGET_ZONED_HM,
> +	.features = DM_TARGET_ZONED_HM | DM_TARGET_EMULATED_ZONE_SIZE,
>  	.map = dmz_npo2_map,
>  	.end_io = dmz_npo2_end_io,
>  	.report_zones = dmz_npo2_report_zones,
>  	.iterate_devices = dmz_npo2_iterate_devices,
>  	.module = THIS_MODULE,
> +	.io_hints = dmz_npo2_io_hints,
>  	.ctr = dmz_npo2_ctr,
>  };
> 
> diff --git a/include/linux/device-mapper.h b/include/linux/device-mapper.h
> index c2a3758c4aaa..9f3a4d98a22a 100644
> --- a/include/linux/device-mapper.h
> +++ b/include/linux/device-mapper.h
> @@ -283,6 +283,15 @@ struct target_type {
>  #define dm_target_supports_mixed_zoned_model(type) (false)
>  #endif
> 
> +#ifdef CONFIG_BLK_DEV_ZONED
> +#define DM_TARGET_EMULATED_ZONE_SIZE	0x00000400

Make it general: DM_TARGET_EMULATED_ZONES

> +#define dm_target_supports_emulated_zone_size(type) \
> +	((type)->features & DM_TARGET_EMULATED_ZONE_SIZE)
> +#else
> +#define DM_TARGET_EMULATED_ZONE_SIZE	0x00000000
> +#define dm_target_supports_emulated_zone_size(type) (false)
> +#endif
> +
>  struct dm_target {
>  	struct dm_table *table;
>  	struct target_type *type;
>

Pankaj Raghav June 17, 2022, 6:40 a.m. UTC | #9

On 2022-06-17 08:12, Damien Le Moal wrote:
>> I think this is a cleaner approach using features flag and io_hints
>> instead of messing with the revalidate zone function:
>>
>> diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
>> index 135c0cc190fb..c97a71e0473f 100644
>> --- a/drivers/md/dm-table.c
>> +++ b/drivers/md/dm-table.c
>> @@ -1618,6 +1618,9 @@ static int device_not_matches_zone_sectors(struct
>> dm_target *ti, struct dm_dev *
>>  	if (!blk_queue_is_zoned(q))
>>  		return 0;
>>
>> +	if(dm_target_supports_emulated_zone_size(ti->type))
>> +		return 0;
>> +
> 
> This should be in validate_hardware_zoned_model(), not here.
> 
I am not sure about this comment. We need to peek into the individual
target from the table to check for this feature right?

if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors,
&zone_sectors)) {
	DMERR("%s: zone sectors is not consistent across all zoned devices",
        dm_device_name(table->md));
	return -EINVAL;
	}

So we call this function device_not_matches_zone_sectors() from
validate_hardware_zoned_model() for each target and we let the validate
succeed even if the target's zone size is different from the underlying
device zone size if this feature flag is set. Let me know if I am
missing something and how this can be moved to
validate_hardware_zoned_model().

Damien Le Moal June 17, 2022, 6:56 a.m. UTC | #10

On 6/17/22 15:40, Pankaj Raghav wrote:
> On 2022-06-17 08:12, Damien Le Moal wrote:
>>> I think this is a cleaner approach using features flag and io_hints
>>> instead of messing with the revalidate zone function:
>>>
>>> diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
>>> index 135c0cc190fb..c97a71e0473f 100644
>>> --- a/drivers/md/dm-table.c
>>> +++ b/drivers/md/dm-table.c
>>> @@ -1618,6 +1618,9 @@ static int device_not_matches_zone_sectors(struct
>>> dm_target *ti, struct dm_dev *
>>>  	if (!blk_queue_is_zoned(q))
>>>  		return 0;
>>>
>>> +	if(dm_target_supports_emulated_zone_size(ti->type))
>>> +		return 0;
>>> +
>>
>> This should be in validate_hardware_zoned_model(), not here.
>>
> I am not sure about this comment. We need to peek into the individual
> target from the table to check for this feature right?
> 
> if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors,
> &zone_sectors)) {
> 	DMERR("%s: zone sectors is not consistent across all zoned devices",
>         dm_device_name(table->md));
> 	return -EINVAL;
> 	}
> 
> So we call this function device_not_matches_zone_sectors() from
> validate_hardware_zoned_model() for each target and we let the validate
> succeed even if the target's zone size is different from the underlying
> device zone size if this feature flag is set. Let me know if I am
> missing something and how this can be moved to
> validate_hardware_zoned_model().

Your change does not match the function name
device_not_matches_zone_sectors(), at all. So I think this is wrong.

The fact is that zone support in DM has been built under the following
assumptions:
1) A zoned device can be used to create a *zoned* target (e.g. dm-linear,
dm-flakey, dm-crypt). For this case, the target *must* use the same zone
size as the underlying devices and all devices used for the target must
have the same zone size.
2) A zoned device can be used to create a *regular* device target (e.g.
dm-zoned). All zoned devices used for the target must have the same zone size.

This new target driver completely breaks (1) and does not fit with (2). I
suspect this is why you are seeing problems with dm_revalidate_zones() as
that one uses the underlying device instead of the target report zones.

Based on this analysis, validate_hardware_zoned_model() definitely needs
to be changed. But device_not_matches_zone_sectors() is to check the
assumptions (1) and (2) so changing it for your new case is wrong in my
opinion. You need another set of assumptions (3) (define that well please)
and modify validate_hardware_zoned_model() so that the defined constraints
are checked. Using a target flag to indicate the type of zoned target is
fine by me.

Pankaj Raghav June 17, 2022, 8:03 a.m. UTC | #11

On 2022-06-17 08:56, Damien Le Moal wrote:
>>
>> So we call this function device_not_matches_zone_sectors() from
>> validate_hardware_zoned_model() for each target and we let the validate
>> succeed even if the target's zone size is different from the underlying
>> device zone size if this feature flag is set. Let me know if I am
>> missing something and how this can be moved to
>> validate_hardware_zoned_model().
> 
> Your change does not match the function name
> device_not_matches_zone_sectors(), at all. So I think this is wrong.
> 
> The fact is that zone support in DM has been built under the following
> assumptions:
> 1) A zoned device can be used to create a *zoned* target (e.g. dm-linear,
> dm-flakey, dm-crypt). For this case, the target *must* use the same zone
> size as the underlying devices and all devices used for the target must
> have the same zone size.
> 2) A zoned device can be used to create a *regular* device target (e.g.
> dm-zoned). All zoned devices used for the target must have the same zone size.
> 
> This new target driver completely breaks (1) and does not fit with (2). I
> suspect this is why you are seeing problems with dm_revalidate_zones() as
> that one uses the underlying device instead of the target report zones.
> 
> Based on this analysis, validate_hardware_zoned_model() definitely needs
> to be changed. But device_not_matches_zone_sectors() is to check the
> assumptions (1) and (2) so changing it for your new case is wrong in my
> opinion. You need another set of assumptions (3) (define that well please)
> and modify validate_hardware_zoned_model() so that the defined constraints
> are checked. Using a target flag to indicate the type of zoned target is
> fine by me.
> 
Got it. Thanks for the explanation. Renaming
device_not_matches_zone_sectors() function to something meaningful with
my changes should address what you have pointed out to accommodate all
three types.

I see that something similar was done to dm_table_supports_zoned_model()
to accommodate type 2(dm-zoned) with different underlying zoned models
even though the initial impl. supported only type 1.

Patch

diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
index 998a5cfdb..773314536 100644
--- a/drivers/md/Kconfig
+++ b/drivers/md/Kconfig
@@ -518,6 +518,15 @@  config DM_FLAKEY
 	help
 	 A target that intermittently fails I/O for debugging purposes.
 
+config DM_ZONED_NPO2
+	tristate "Zoned non power of 2 target"
+	depends on BLK_DEV_DM
+	depends on BLK_DEV_ZONED
+	help
+	A target that converts a zoned device with non power of 2 zone size to
+	be power of 2. This is done by introducing gaps in between the zone
+	capacity and the power of 2 zone size.
+
 config DM_VERITY
 	tristate "Verity target support"
 	depends on BLK_DEV_DM
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index 0454b0885..2863a94a7 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -26,6 +26,7 @@  dm-era-y	+= dm-era-target.o
 dm-clone-y	+= dm-clone-target.o dm-clone-metadata.o
 dm-verity-y	+= dm-verity-target.o
 dm-zoned-y	+= dm-zoned-target.o dm-zoned-metadata.o dm-zoned-reclaim.o
+dm-zoned-npo2-y       += dm-zoned-npo2-target.o
 
 md-mod-y	+= md.o md-bitmap.o
 raid456-y	+= raid5.o raid5-cache.o raid5-ppl.o
@@ -60,6 +61,7 @@  obj-$(CONFIG_DM_CRYPT)		+= dm-crypt.o
 obj-$(CONFIG_DM_DELAY)		+= dm-delay.o
 obj-$(CONFIG_DM_DUST)		+= dm-dust.o
 obj-$(CONFIG_DM_FLAKEY)		+= dm-flakey.o
+obj-$(CONFIG_DM_ZONED_NPO2)	+= dm-zoned-npo2.o
 obj-$(CONFIG_DM_MULTIPATH)	+= dm-multipath.o dm-round-robin.o
 obj-$(CONFIG_DM_MULTIPATH_QL)	+= dm-queue-length.o
 obj-$(CONFIG_DM_MULTIPATH_ST)	+= dm-service-time.o
diff --git a/drivers/md/dm-zone.c b/drivers/md/dm-zone.c
index af36d33f9..5efb31ba0 100644
--- a/drivers/md/dm-zone.c
+++ b/drivers/md/dm-zone.c
@@ -210,6 +210,11 @@  static int dm_zone_revalidate_cb(struct blk_zone *zone, unsigned int idx,
 		}
 		md->zwp_offset[idx] = dm_get_zone_wp_offset(zone);
 
+		if (q->limits.chunk_sectors != zone->len) {
+			blk_queue_chunk_sectors(q, zone->len);
+			q->nr_zones = blkdev_nr_zones(md->disk);
+		}
+
 		break;
 	default:
 		DMERR("Invalid zone type 0x%x at sectors %llu",
@@ -307,6 +312,9 @@  int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
 	if (dm_table_supports_zone_append(t)) {
 		clear_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
 		dm_cleanup_zoned_dev(md);
+
+		if (!is_power_of_2(blk_queue_zone_sectors(q)))
+			goto revalidate_zones;
 		return 0;
 	}
 
@@ -318,6 +326,7 @@  int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
 	if (!get_capacity(md->disk))
 		return 0;
 
+revalidate_zones:
 	return dm_revalidate_zones(md, t);
 }
 
diff --git a/drivers/md/dm-zoned-npo2-target.c b/drivers/md/dm-zoned-npo2-target.c
new file mode 100644
index 000000000..c1373d3ea
--- /dev/null
+++ b/drivers/md/dm-zoned-npo2-target.c
@@ -0,0 +1,268 @@ 
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2022 Samsung Electronics Co., Ltd.
+ */
+
+#include <linux/device-mapper.h>
+
+#define DM_MSG_PREFIX "zoned-npo2"
+
+struct dmz_npo2_target {
+	struct dm_dev *dev;
+	sector_t zsze;
+	sector_t zsze_po2;
+	sector_t zsze_diff;
+	u32 nr_zones;
+};
+
+enum dmz_npo2_io_cond {
+	DMZ_NPO2_IO_INSIDE_ZONE,
+	DMZ_NPO2_IO_ACROSS_ZONE,
+	DMZ_NPO2_IO_OUTSIDE_ZONE,
+};
+
+static inline u32 npo2_zone_no(struct dmz_npo2_target *dmh, sector_t sect)
+{
+	return div64_u64(sect, dmh->zsze);
+}
+
+static inline u32 po2_zone_no(struct dmz_npo2_target *dmh, sector_t sect)
+{
+	return sect >> ilog2(dmh->zsze_po2);
+}
+
+static inline sector_t target_to_device_sect(struct dmz_npo2_target *dmh,
+					     sector_t sect)
+{
+	u32 zone_idx = po2_zone_no(dmh, sect);
+
+	sect -= (zone_idx * dmh->zsze_diff);
+
+	return sect;
+}
+
+static inline sector_t device_to_target_sect(struct dmz_npo2_target *dmh,
+					     sector_t sect)
+{
+	u32 zone_idx = npo2_zone_no(dmh, sect);
+
+	sect += (zone_idx * dmh->zsze_diff);
+
+	return sect;
+}
+
+/*
+ * <dev-path>
+ * This target works on the complete zoned device. Partial mapping is not
+ * supported
+ */
+static int dmz_npo2_ctr(struct dm_target *ti, unsigned int argc, char **argv)
+{
+	struct dmz_npo2_target *dmh = NULL;
+	int ret = 0;
+	sector_t zsze;
+	sector_t disk_size;
+
+	if (argc < 1)
+		return -EINVAL;
+
+	dmh = kmalloc(sizeof(*dmh), GFP_KERNEL);
+	if (!dmh)
+		return -ENOMEM;
+
+	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
+			    &dmh->dev);
+
+	zsze = blk_queue_zone_sectors(bdev_get_queue(dmh->dev->bdev));
+
+	disk_size = get_capacity(dmh->dev->bdev->bd_disk);
+
+	if (ti->len != disk_size || ti->begin) {
+		DMERR("%pg Partial mapping of the target not supported",
+		      dmh->dev->bdev);
+		return -EINVAL;
+	}
+
+	if (is_power_of_2(zsze)) {
+		DMERR("%pg zone size is power of 2", dmh->dev->bdev);
+		return -EINVAL;
+	}
+
+	dmh->zsze = zsze;
+	dmh->zsze_po2 = 1 << get_count_order_long(zsze);
+	dmh->zsze_diff = dmh->zsze_po2 - dmh->zsze;
+
+	ti->private = dmh;
+	ti->num_flush_bios = 1;
+	ti->num_discard_bios = 1;
+	ti->num_secure_erase_bios = 1;
+	ti->num_write_zeroes_bios = 1;
+
+	dmh->nr_zones = npo2_zone_no(dmh, ti->len);
+	ti->len = dmh->zsze_po2 * dmh->nr_zones;
+
+	return 0;
+}
+
+static int dmz_npo2_report_zones_cb(struct blk_zone *zone, unsigned int idx,
+				    void *data)
+{
+	struct dm_report_zones_args *args = data;
+	struct dmz_npo2_target *dmh = args->tgt->private;
+
+	zone->start = device_to_target_sect(dmh, zone->start);
+	zone->wp = device_to_target_sect(dmh, zone->wp);
+	zone->len = dmh->zsze_po2;
+	args->next_sector = zone->start + zone->len;
+
+	return args->orig_cb(zone, args->zone_idx++, args->orig_data);
+}
+
+static int dmz_npo2_report_zones(struct dm_target *ti,
+				 struct dm_report_zones_args *args,
+				 unsigned int nr_zones)
+{
+	struct dmz_npo2_target *dmh = ti->private;
+	int ret = 0;
+	sector_t sect = po2_zone_no(dmh, args->next_sector) * dmh->zsze;
+
+	ret = blkdev_report_zones(dmh->dev->bdev, sect, nr_zones,
+				  dmz_npo2_report_zones_cb, args);
+	if (ret < 0)
+		DMERR("report zones error");
+
+	return ret;
+}
+
+static int check_zone_boundary_violation(struct dmz_npo2_target *dmh,
+					 sector_t sect, sector_t size)
+{
+	u32 zone_idx = po2_zone_no(dmh, sect);
+	sector_t relative_sect = 0;
+
+	sect = target_to_device_sect(dmh, sect);
+	relative_sect = sect - (zone_idx * dmh->zsze);
+
+	if ((relative_sect + size) <= dmh->zsze)
+		return DMZ_NPO2_IO_INSIDE_ZONE;
+	else if (relative_sect >= dmh->zsze)
+		return DMZ_NPO2_IO_OUTSIDE_ZONE;
+
+	return DMZ_NPO2_IO_ACROSS_ZONE;
+}
+
+static void split_io_across_zone_boundary(struct dmz_npo2_target *dmh,
+					  struct bio *bio)
+{
+	sector_t sect = bio->bi_iter.bi_sector;
+	sector_t sects_from_zone_start;
+
+	sect = target_to_device_sect(dmh, sect);
+	div64_u64_rem(sect, dmh->zsze, &sects_from_zone_start);
+	dm_accept_partial_bio(bio, dmh->zsze - sects_from_zone_start);
+	bio->bi_iter.bi_sector = sect;
+}
+
+static int handle_zone_boundary_violation(struct dmz_npo2_target *dmh,
+					  struct bio *bio,
+					  enum dmz_npo2_io_cond cond)
+{
+	/* Read should return zeroed page */
+	if (bio_op(bio) == REQ_OP_READ) {
+		if (cond == DMZ_NPO2_IO_ACROSS_ZONE) {
+			split_io_across_zone_boundary(dmh, bio);
+			return DM_MAPIO_REMAPPED;
+		}
+		zero_fill_bio(bio);
+		bio_endio(bio);
+		return DM_MAPIO_SUBMITTED;
+	}
+	return DM_MAPIO_KILL;
+}
+
+static int dmz_npo2_end_io(struct dm_target *ti, struct bio *bio,
+			   blk_status_t *error)
+{
+	struct dmz_npo2_target *dmh = ti->private;
+
+	if (bio->bi_status == BLK_STS_OK && bio_op(bio) == REQ_OP_ZONE_APPEND)
+		bio->bi_iter.bi_sector =
+			device_to_target_sect(dmh, bio->bi_iter.bi_sector);
+
+	return DM_ENDIO_DONE;
+}
+
+static int dmz_npo2_map(struct dm_target *ti, struct bio *bio)
+{
+	struct dmz_npo2_target *dmh = ti->private;
+	enum dmz_npo2_io_cond cond;
+
+	bio_set_dev(bio, dmh->dev->bdev);
+	if (bio_sectors(bio) || op_is_zone_mgmt(bio_op(bio))) {
+		cond = check_zone_boundary_violation(dmh, bio->bi_iter.bi_sector,
+						     bio->bi_iter.bi_size >> SECTOR_SHIFT);
+
+		/*
+		 * If the starting sector is in the emulated area then fill
+		 * all the bio with zeros. If bio is across boundaries,
+		 * split the bio across boundaries and fill zeros only for the
+		 * bio that is outside the zone capacity
+		 */
+		switch (cond) {
+		case DMZ_NPO2_IO_INSIDE_ZONE:
+			bio->bi_iter.bi_sector = target_to_device_sect(dmh,
+								       bio->bi_iter.bi_sector);
+			break;
+		case DMZ_NPO2_IO_ACROSS_ZONE:
+		case DMZ_NPO2_IO_OUTSIDE_ZONE:
+			return handle_zone_boundary_violation(dmh, bio, cond);
+		}
+	}
+	return DM_MAPIO_REMAPPED;
+}
+
+static int dmz_npo2_iterate_devices(struct dm_target *ti,
+				    iterate_devices_callout_fn fn, void *data)
+{
+	struct dmz_npo2_target *dmh = ti->private;
+	sector_t len = 0;
+
+	len = dmh->nr_zones * dmh->zsze;
+	return fn(ti, dmh->dev, 0, len, data);
+}
+
+static struct target_type dmz_npo2_target = {
+	.name = "zoned-npo2",
+	.version = { 1, 0, 0 },
+	.features = DM_TARGET_ZONED_HM,
+	.map = dmz_npo2_map,
+	.end_io = dmz_npo2_end_io,
+	.report_zones = dmz_npo2_report_zones,
+	.iterate_devices = dmz_npo2_iterate_devices,
+	.module = THIS_MODULE,
+	.ctr = dmz_npo2_ctr,
+};
+
+static int __init dmz_npo2_init(void)
+{
+	int r = dm_register_target(&dmz_npo2_target);
+
+	if (r < 0)
+		DMERR("register failed %d", r);
+
+	return r;
+}
+
+static void __exit dmz_npo2_exit(void)
+{
+	dm_unregister_target(&dmz_npo2_target);
+}
+
+/* Module hooks */
+module_init(dmz_npo2_init);
+module_exit(dmz_npo2_exit);
+
+MODULE_DESCRIPTION(DM_NAME " non power 2 zoned target");
+MODULE_AUTHOR("Pankaj Raghav <[e-mail]>");
+MODULE_LICENSE("GPL");
+