[linux-block.git] / drivers / md / dm-stripe.c

/*
 * Copyright (C) 2001-2003 Sistina Software (UK) Limited.
 *
 * This file is released under the GPL.
 */

#include "dm.h"
#include <linux/device-mapper.h>

#include <linux/module.h>
#include <linux/init.h>
#include <linux/blkdev.h>
#include <linux/bio.h>
#include <linux/dax.h>
#include <linux/slab.h>
#include <linux/log2.h>

#define DM_MSG_PREFIX "striped"
#define DM_IO_ERROR_THRESHOLD 15

struct stripe {
	struct dm_dev *dev;
	sector_t physical_start;

	atomic_t error_count;
};

struct stripe_c {
	uint32_t stripes;
	int stripes_shift;

	/* The size of this target / num. stripes */
	sector_t stripe_width;

	uint32_t chunk_size;
	int chunk_size_shift;

	/* Needed for handling events */
	struct dm_target *ti;

	/* Work struct used for triggering events*/
	struct work_struct trigger_event;

	struct stripe stripe[0];
};

/*
 * An event is triggered whenever a drive
 * drops out of a stripe volume.
 */
static void trigger_event(struct work_struct *work)
{
	struct stripe_c *sc = container_of(work, struct stripe_c,
					   trigger_event);
	dm_table_event(sc->ti->table);
}

static inline struct stripe_c *alloc_context(unsigned int stripes)
{
	size_t len;

	if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe),
			     stripes))
		return NULL;

	len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes);

	return kmalloc(len, GFP_KERNEL);
}

/*
 * Parse a single <dev> <sector> pair
 */
static int get_stripe(struct dm_target *ti, struct stripe_c *sc,
		      unsigned int stripe, char **argv)
{
	unsigned long long start;
	char dummy;
	int ret;

	if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1)
		return -EINVAL;

	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
			    &sc->stripe[stripe].dev);
	if (ret)
		return ret;

	sc->stripe[stripe].physical_start = start;

	return 0;
}

/*
 * Construct a striped mapping.
 * <number of stripes> <chunk size> [<dev_path> <offset>]+
 */
static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv)
{
	struct stripe_c *sc;
	sector_t width, tmp_len;
	uint32_t stripes;
	uint32_t chunk_size;
	int r;
	unsigned int i;

	if (argc < 2) {
		ti->error = "Not enough arguments";
		return -EINVAL;
	}

	if (kstrtouint(argv[0], 10, &stripes) || !stripes) {
		ti->error = "Invalid stripe count";
		return -EINVAL;
	}

	if (kstrtouint(argv[1], 10, &chunk_size) || !chunk_size) {
		ti->error = "Invalid chunk_size";
		return -EINVAL;
	}

	width = ti->len;
	if (sector_div(width, stripes)) {
		ti->error = "Target length not divisible by "
		    "number of stripes";
		return -EINVAL;
	}

	tmp_len = width;
	if (sector_div(tmp_len, chunk_size)) {
		ti->error = "Target length not divisible by "
		    "chunk size";
		return -EINVAL;
	}

	/*
	 * Do we have enough arguments for that many stripes ?
	 */
	if (argc != (2 + 2 * stripes)) {
		ti->error = "Not enough destinations "
			"specified";
		return -EINVAL;
	}

	sc = alloc_context(stripes);
	if (!sc) {
		ti->error = "Memory allocation for striped context "
		    "failed";
		return -ENOMEM;
	}

	INIT_WORK(&sc->trigger_event, trigger_event);

	/* Set pointer to dm target; used in trigger_event */
	sc->ti = ti;
	sc->stripes = stripes;
	sc->stripe_width = width;

	if (stripes & (stripes - 1))
		sc->stripes_shift = -1;
	else
		sc->stripes_shift = __ffs(stripes);

	r = dm_set_target_max_io_len(ti, chunk_size);
	if (r) {
		kfree(sc);
		return r;
	}

	ti->num_flush_bios = stripes;
	ti->num_discard_bios = stripes;
	ti->num_write_same_bios = stripes;
	ti->num_write_zeroes_bios = stripes;

	sc->chunk_size = chunk_size;
	if (chunk_size & (chunk_size - 1))
		sc->chunk_size_shift = -1;
	else
		sc->chunk_size_shift = __ffs(chunk_size);

	/*
	 * Get the stripe destinations.
	 */
	for (i = 0; i < stripes; i++) {
		argv += 2;

		r = get_stripe(ti, sc, i, argv);
		if (r < 0) {
			ti->error = "Couldn't parse stripe destination";
			while (i--)
				dm_put_device(ti, sc->stripe[i].dev);
			kfree(sc);
			return r;
		}
		atomic_set(&(sc->stripe[i].error_count), 0);
	}

	ti->private = sc;

	return 0;
}

static void stripe_dtr(struct dm_target *ti)
{
	unsigned int i;
	struct stripe_c *sc = (struct stripe_c *) ti->private;

	for (i = 0; i < sc->stripes; i++)
		dm_put_device(ti, sc->stripe[i].dev);

	flush_work(&sc->trigger_event);
	kfree(sc);
}

static void stripe_map_sector(struct stripe_c *sc, sector_t sector,
			      uint32_t *stripe, sector_t *result)
{
	sector_t chunk = dm_target_offset(sc->ti, sector);
	sector_t chunk_offset;

	if (sc->chunk_size_shift < 0)
		chunk_offset = sector_div(chunk, sc->chunk_size);
	else {
		chunk_offset = chunk & (sc->chunk_size - 1);
		chunk >>= sc->chunk_size_shift;
	}

	if (sc->stripes_shift < 0)
		*stripe = sector_div(chunk, sc->stripes);
	else {
		*stripe = chunk & (sc->stripes - 1);
		chunk >>= sc->stripes_shift;
	}

	if (sc->chunk_size_shift < 0)
		chunk *= sc->chunk_size;
	else
		chunk <<= sc->chunk_size_shift;

	*result = chunk + chunk_offset;
}

static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector,
				    uint32_t target_stripe, sector_t *result)
{
	uint32_t stripe;

	stripe_map_sector(sc, sector, &stripe, result);
	if (stripe == target_stripe)
		return;

	/* round down */
	sector = *result;
	if (sc->chunk_size_shift < 0)
		*result -= sector_div(sector, sc->chunk_size);
	else
		*result = sector & ~(sector_t)(sc->chunk_size - 1);

	if (target_stripe < stripe)
		*result += sc->chunk_size;		/* next chunk */
}

static int stripe_map_range(struct stripe_c *sc, struct bio *bio,
			    uint32_t target_stripe)
{
	sector_t begin, end;

	stripe_map_range_sector(sc, bio->bi_iter.bi_sector,
				target_stripe, &begin);
	stripe_map_range_sector(sc, bio_end_sector(bio),
				target_stripe, &end);
	if (begin < end) {
		bio->bi_bdev = sc->stripe[target_stripe].dev->bdev;
		bio->bi_iter.bi_sector = begin +
			sc->stripe[target_stripe].physical_start;
		bio->bi_iter.bi_size = to_bytes(end - begin);
		return DM_MAPIO_REMAPPED;
	} else {
		/* The range doesn't map to the target stripe */
		bio_endio(bio);
		return DM_MAPIO_SUBMITTED;
	}
}

static int stripe_map(struct dm_target *ti, struct bio *bio)
{
	struct stripe_c *sc = ti->private;
	uint32_t stripe;
	unsigned target_bio_nr;

	if (bio->bi_opf & REQ_PREFLUSH) {
		target_bio_nr = dm_bio_get_target_bio_nr(bio);
		BUG_ON(target_bio_nr >= sc->stripes);
		bio->bi_bdev = sc->stripe[target_bio_nr].dev->bdev;
		return DM_MAPIO_REMAPPED;
	}
	if (unlikely(bio_op(bio) == REQ_OP_DISCARD) ||
	    unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES) ||
	    unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) {
		target_bio_nr = dm_bio_get_target_bio_nr(bio);
		BUG_ON(target_bio_nr >= sc->stripes);
		return stripe_map_range(sc, bio, target_bio_nr);
	}

	stripe_map_sector(sc, bio->bi_iter.bi_sector,
			  &stripe, &bio->bi_iter.bi_sector);

	bio->bi_iter.bi_sector += sc->stripe[stripe].physical_start;
	bio->bi_bdev = sc->stripe[stripe].dev->bdev;

	return DM_MAPIO_REMAPPED;
}

static long stripe_dax_direct_access(struct dm_target *ti, pgoff_t pgoff,
		long nr_pages, void **kaddr, pfn_t *pfn)
{
	sector_t dev_sector, sector = pgoff * PAGE_SECTORS;
	struct stripe_c *sc = ti->private;
	struct dax_device *dax_dev;
	struct block_device *bdev;
	uint32_t stripe;
	long ret;

	stripe_map_sector(sc, sector, &stripe, &dev_sector);
	dev_sector += sc->stripe[stripe].physical_start;
	dax_dev = sc->stripe[stripe].dev->dax_dev;
	bdev = sc->stripe[stripe].dev->bdev;

	ret = bdev_dax_pgoff(bdev, dev_sector, nr_pages * PAGE_SIZE, &pgoff);
	if (ret)
		return ret;
	return dax_direct_access(dax_dev, pgoff, nr_pages, kaddr, pfn);
}

static size_t stripe_dax_copy_from_iter(struct dm_target *ti, pgoff_t pgoff,
		void *addr, size_t bytes, struct iov_iter *i)
{
	sector_t dev_sector, sector = pgoff * PAGE_SECTORS;
	struct stripe_c *sc = ti->private;
	struct dax_device *dax_dev;
	struct block_device *bdev;
	uint32_t stripe;

	stripe_map_sector(sc, sector, &stripe, &dev_sector);
	dev_sector += sc->stripe[stripe].physical_start;
	dax_dev = sc->stripe[stripe].dev->dax_dev;
	bdev = sc->stripe[stripe].dev->bdev;

	if (bdev_dax_pgoff(bdev, dev_sector, ALIGN(bytes, PAGE_SIZE), &pgoff))
		return 0;
	return dax_copy_from_iter(dax_dev, pgoff, addr, bytes, i);
}

static void stripe_dax_flush(struct dm_target *ti, pgoff_t pgoff, void *addr,
		size_t size)
{
	sector_t dev_sector, sector = pgoff * PAGE_SECTORS;
	struct stripe_c *sc = ti->private;
	struct dax_device *dax_dev;
	struct block_device *bdev;
	uint32_t stripe;

	stripe_map_sector(sc, sector, &stripe, &dev_sector);
	dev_sector += sc->stripe[stripe].physical_start;
	dax_dev = sc->stripe[stripe].dev->dax_dev;
	bdev = sc->stripe[stripe].dev->bdev;

	if (bdev_dax_pgoff(bdev, dev_sector, ALIGN(size, PAGE_SIZE), &pgoff))
		return;
	dax_flush(dax_dev, pgoff, addr, size);
}

/*
 * Stripe status:
 *
 * INFO
 * #stripes [stripe_name <stripe_name>] [group word count]
 * [error count 'A|D' <error count 'A|D'>]
 *
 * TABLE
 * #stripes [stripe chunk size]
 * [stripe_name physical_start <stripe_name physical_start>]
 *
 */

static void stripe_status(struct dm_target *ti, status_type_t type,
			  unsigned status_flags, char *result, unsigned maxlen)
{
	struct stripe_c *sc = (struct stripe_c *) ti->private;
	char buffer[sc->stripes + 1];
	unsigned int sz = 0;
	unsigned int i;

	switch (type) {
	case STATUSTYPE_INFO:
		DMEMIT("%d ", sc->stripes);
		for (i = 0; i < sc->stripes; i++)  {
			DMEMIT("%s ", sc->stripe[i].dev->name);
			buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ?
				'D' : 'A';
		}
		buffer[i] = '\0';
		DMEMIT("1 %s", buffer);
		break;

	case STATUSTYPE_TABLE:
		DMEMIT("%d %llu", sc->stripes,
			(unsigned long long)sc->chunk_size);
		for (i = 0; i < sc->stripes; i++)
			DMEMIT(" %s %llu", sc->stripe[i].dev->name,
			    (unsigned long long)sc->stripe[i].physical_start);
		break;
	}
}

static int stripe_end_io(struct dm_target *ti, struct bio *bio,
		blk_status_t *error)
{
	unsigned i;
	char major_minor[16];
	struct stripe_c *sc = ti->private;

	if (!*error)
		return DM_ENDIO_DONE; /* I/O complete */

	if (bio->bi_opf & REQ_RAHEAD)
		return DM_ENDIO_DONE;

	if (*error == BLK_STS_NOTSUPP)
		return DM_ENDIO_DONE;

	memset(major_minor, 0, sizeof(major_minor));
	sprintf(major_minor, "%d:%d",
		MAJOR(disk_devt(bio->bi_bdev->bd_disk)),
		MINOR(disk_devt(bio->bi_bdev->bd_disk)));

	/*
	 * Test to see which stripe drive triggered the event
	 * and increment error count for all stripes on that device.
	 * If the error count for a given device exceeds the threshold
	 * value we will no longer trigger any further events.
	 */
	for (i = 0; i < sc->stripes; i++)
		if (!strcmp(sc->stripe[i].dev->name, major_minor)) {
			atomic_inc(&(sc->stripe[i].error_count));
			if (atomic_read(&(sc->stripe[i].error_count)) <
			    DM_IO_ERROR_THRESHOLD)
				schedule_work(&sc->trigger_event);
		}

	return DM_ENDIO_DONE;
}

static int stripe_iterate_devices(struct dm_target *ti,
				  iterate_devices_callout_fn fn, void *data)
{
	struct stripe_c *sc = ti->private;
	int ret = 0;
	unsigned i = 0;

	do {
		ret = fn(ti, sc->stripe[i].dev,
			 sc->stripe[i].physical_start,
			 sc->stripe_width, data);
	} while (!ret && ++i < sc->stripes);

	return ret;
}

static void stripe_io_hints(struct dm_target *ti,
			    struct queue_limits *limits)
{
	struct stripe_c *sc = ti->private;
	unsigned chunk_size = sc->chunk_size << SECTOR_SHIFT;

	blk_limits_io_min(limits, chunk_size);
	blk_limits_io_opt(limits, chunk_size * sc->stripes);
}

static struct target_type stripe_target = {
	.name   = "striped",
	.version = {1, 6, 0},
	.features = DM_TARGET_PASSES_INTEGRITY,
	.module = THIS_MODULE,
	.ctr    = stripe_ctr,
	.dtr    = stripe_dtr,
	.map    = stripe_map,
	.end_io = stripe_end_io,
	.status = stripe_status,
	.iterate_devices = stripe_iterate_devices,
	.io_hints = stripe_io_hints,
	.direct_access = stripe_dax_direct_access,
	.dax_copy_from_iter = stripe_dax_copy_from_iter,
	.dax_flush = stripe_dax_flush,
};

int __init dm_stripe_init(void)
{
	int r;

	r = dm_register_target(&stripe_target);
	if (r < 0)
		DMWARN("target registration failed");

	return r;
}

void dm_stripe_exit(void)
{
	dm_unregister_target(&stripe_target);
}
Commit	Line	Data
1da177e4 LT	1	/*
	2	* Copyright (C) 2001-2003 Sistina Software (UK) Limited.
	3	*
	4	* This file is released under the GPL.
	5	*/
	6
7fff5e8f	7	#include "dm.h"
586e80e6	8	#include <linux/device-mapper.h>
1da177e4 LT	9
	10	#include <linux/module.h>
	11	#include <linux/init.h>
	12	#include <linux/blkdev.h>
	13	#include <linux/bio.h>
817bf402	14	#include <linux/dax.h>
1da177e4	15	#include <linux/slab.h>
6f3c3f0a	16	#include <linux/log2.h>
1da177e4	17
72d94861	18	#define DM_MSG_PREFIX "striped"
a25eb944	19	#define DM_IO_ERROR_THRESHOLD 15
72d94861	20
1da177e4 LT	21	struct stripe {
	22	struct dm_dev *dev;
	23	sector_t physical_start;
a25eb944 BW	24
a25eb944 BW	25	atomic_t error_count;
1da177e4 LT	26	};
	27
	28	struct stripe_c {
	29	uint32_t stripes;
c96053b7	30	int stripes_shift;
1da177e4 LT	31
	32	/* The size of this target / num. stripes */
	33	sector_t stripe_width;
	34
eb850de6	35	uint32_t chunk_size;
33d07c0d	36	int chunk_size_shift;
1da177e4	37
a25eb944 BW	38	/* Needed for handling events */
	39	struct dm_target *ti;
	40
	41	/* Work struct used for triggering events*/
f521f074	42	struct work_struct trigger_event;
a25eb944	43
1da177e4 LT	44	struct stripe stripe[0];
	45	};
	46
a25eb944 BW	47	/*
	48	* An event is triggered whenever a drive
	49	* drops out of a stripe volume.
	50	*/
	51	static void trigger_event(struct work_struct *work)
	52	{
f521f074 TH	53	struct stripe_c *sc = container_of(work, struct stripe_c,
f521f074 TH	54	trigger_event);
a25eb944	55	dm_table_event(sc->ti->table);
a25eb944 BW	56	}
a25eb944 BW	57
1da177e4 LT	58	static inline struct stripe_c *alloc_context(unsigned int stripes)
	59	{
	60	size_t len;
	61
d63a5ce3 MP	62	if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe),
d63a5ce3 MP	63	stripes))
1da177e4 LT	64	return NULL;
	65
	66	len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes);
	67
	68	return kmalloc(len, GFP_KERNEL);
	69	}
	70
	71	/*
	72	* Parse a single <dev> <sector> pair
	73	*/
	74	static int get_stripe(struct dm_target ti, struct stripe_c sc,
	75	unsigned int stripe, char **argv)
	76	{
4ee218cd	77	unsigned long long start;
31998ef1	78	char dummy;
e80d1c80	79	int ret;
1da177e4	80
31998ef1	81	if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1)
1da177e4 LT	82	return -EINVAL;
1da177e4 LT	83
e80d1c80 VG	84	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
	85	&sc->stripe[stripe].dev);
	86	if (ret)
	87	return ret;
1da177e4 LT	88
1da177e4 LT	89	sc->stripe[stripe].physical_start = start;
a25eb944	90
1da177e4 LT	91	return 0;
	92	}
	93
	94	/*
	95	* Construct a striped mapping.
eb850de6	96	* <number of stripes> <chunk size> [<dev_path> <offset>]+
1da177e4 LT	97	*/
	98	static int stripe_ctr(struct dm_target ti, unsigned int argc, char *argv)
	99	{
	100	struct stripe_c *sc;
d793e684	101	sector_t width, tmp_len;
1da177e4 LT	102	uint32_t stripes;
1da177e4 LT	103	uint32_t chunk_size;
1da177e4 LT	104	int r;
	105	unsigned int i;
	106
	107	if (argc < 2) {
72d94861	108	ti->error = "Not enough arguments";
1da177e4 LT	109	return -EINVAL;
	110	}
	111
1a66a08a	112	if (kstrtouint(argv[0], 10, &stripes) \|\| !stripes) {
72d94861	113	ti->error = "Invalid stripe count";
1da177e4 LT	114	return -EINVAL;
	115	}
	116
8f069b41	117	if (kstrtouint(argv[1], 10, &chunk_size) \|\| !chunk_size) {
eb850de6	118	ti->error = "Invalid chunk_size";
1da177e4 LT	119	return -EINVAL;
	120	}
	121
eb850de6	122	width = ti->len;
d793e684	123	if (sector_div(width, stripes)) {
72d94861	124	ti->error = "Target length not divisible by "
d793e684	125	"number of stripes";
8ba32fde KC	126	return -EINVAL;
	127	}
	128
d793e684 MS	129	tmp_len = width;
d793e684 MS	130	if (sector_div(tmp_len, chunk_size)) {
72d94861	131	ti->error = "Target length not divisible by "
d793e684	132	"chunk size";
1da177e4 LT	133	return -EINVAL;
	134	}
	135
	136	/*
	137	* Do we have enough arguments for that many stripes ?
	138	*/
	139	if (argc != (2 + 2 * stripes)) {
72d94861	140	ti->error = "Not enough destinations "
1da177e4 LT	141	"specified";
	142	return -EINVAL;
	143	}
	144
	145	sc = alloc_context(stripes);
	146	if (!sc) {
72d94861	147	ti->error = "Memory allocation for striped context "
1da177e4 LT	148	"failed";
	149	return -ENOMEM;
	150	}
	151
f521f074	152	INIT_WORK(&sc->trigger_event, trigger_event);
a25eb944 BW	153
	154	/* Set pointer to dm target; used in trigger_event */
	155	sc->ti = ti;
1da177e4 LT	156	sc->stripes = stripes;
1da177e4 LT	157	sc->stripe_width = width;
c96053b7 MP	158
	159	if (stripes & (stripes - 1))
	160	sc->stripes_shift = -1;
1df05483 MP	161	else
1df05483 MP	162	sc->stripes_shift = __ffs(stripes);
c96053b7	163
542f9038	164	r = dm_set_target_max_io_len(ti, chunk_size);
a3f2af25 PK	165	if (r) {
a3f2af25 PK	166	kfree(sc);
542f9038	167	return r;
a3f2af25	168	}
542f9038	169
55a62eef AK	170	ti->num_flush_bios = stripes;
	171	ti->num_discard_bios = stripes;
	172	ti->num_write_same_bios = stripes;
ac62d620	173	ti->num_write_zeroes_bios = stripes;
1da177e4	174
eb850de6	175	sc->chunk_size = chunk_size;
33d07c0d MP	176	if (chunk_size & (chunk_size - 1))
	177	sc->chunk_size_shift = -1;
	178	else
	179	sc->chunk_size_shift = __ffs(chunk_size);
1da177e4 LT	180
	181	/*
	182	* Get the stripe destinations.
	183	*/
	184	for (i = 0; i < stripes; i++) {
	185	argv += 2;
	186
	187	r = get_stripe(ti, sc, i, argv);
	188	if (r < 0) {
72d94861	189	ti->error = "Couldn't parse stripe destination";
1da177e4 LT	190	while (i--)
	191	dm_put_device(ti, sc->stripe[i].dev);
	192	kfree(sc);
	193	return r;
	194	}
a25eb944	195	atomic_set(&(sc->stripe[i].error_count), 0);
1da177e4 LT	196	}
	197
	198	ti->private = sc;
a25eb944	199
1da177e4 LT	200	return 0;
	201	}
	202
	203	static void stripe_dtr(struct dm_target *ti)
	204	{
	205	unsigned int i;
	206	struct stripe_c sc = (struct stripe_c ) ti->private;
	207
	208	for (i = 0; i < sc->stripes; i++)
	209	dm_put_device(ti, sc->stripe[i].dev);
	210
43829731	211	flush_work(&sc->trigger_event);
1da177e4 LT	212	kfree(sc);
	213	}
	214
65988525 MP	215	static void stripe_map_sector(struct stripe_c *sc, sector_t sector,
	216	uint32_t stripe, sector_t result)
	217	{
eb850de6	218	sector_t chunk = dm_target_offset(sc->ti, sector);
33d07c0d MP	219	sector_t chunk_offset;
	220
	221	if (sc->chunk_size_shift < 0)
	222	chunk_offset = sector_div(chunk, sc->chunk_size);
	223	else {
	224	chunk_offset = chunk & (sc->chunk_size - 1);
	225	chunk >>= sc->chunk_size_shift;
	226	}
65988525	227
c96053b7 MP	228	if (sc->stripes_shift < 0)
	229	*stripe = sector_div(chunk, sc->stripes);
	230	else {
1df05483	231	*stripe = chunk & (sc->stripes - 1);
c96053b7 MP	232	chunk >>= sc->stripes_shift;
	233	}
	234
33d07c0d MP	235	if (sc->chunk_size_shift < 0)
	236	chunk *= sc->chunk_size;
	237	else
	238	chunk <<= sc->chunk_size_shift;
	239
	240	*result = chunk + chunk_offset;
65988525 MP	241	}
65988525 MP	242
7b76ec11 MP	243	static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector,
	244	uint32_t target_stripe, sector_t *result)
	245	{
	246	uint32_t stripe;
	247
	248	stripe_map_sector(sc, sector, &stripe, result);
	249	if (stripe == target_stripe)
	250	return;
eb850de6 MS	251
	252	/* round down */
	253	sector = *result;
33d07c0d MP	254	if (sc->chunk_size_shift < 0)
	255	*result -= sector_div(sector, sc->chunk_size);
	256	else
	257	*result = sector & ~(sector_t)(sc->chunk_size - 1);
eb850de6	258
7b76ec11	259	if (target_stripe < stripe)
eb850de6	260	result += sc->chunk_size; / next chunk */
7b76ec11 MP	261	}
7b76ec11 MP	262
45e621d4 MS	263	static int stripe_map_range(struct stripe_c sc, struct bio bio,
45e621d4 MS	264	uint32_t target_stripe)
7b76ec11 MP	265	{
	266	sector_t begin, end;
	267
4f024f37 KO	268	stripe_map_range_sector(sc, bio->bi_iter.bi_sector,
4f024f37 KO	269	target_stripe, &begin);
f73a1c7d	270	stripe_map_range_sector(sc, bio_end_sector(bio),
7b76ec11 MP	271	target_stripe, &end);
	272	if (begin < end) {
	273	bio->bi_bdev = sc->stripe[target_stripe].dev->bdev;
4f024f37 KO	274	bio->bi_iter.bi_sector = begin +
	275	sc->stripe[target_stripe].physical_start;
	276	bio->bi_iter.bi_size = to_bytes(end - begin);
7b76ec11 MP	277	return DM_MAPIO_REMAPPED;
	278	} else {
	279	/* The range doesn't map to the target stripe */
4246a0b6	280	bio_endio(bio);
7b76ec11 MP	281	return DM_MAPIO_SUBMITTED;
	282	}
	283	}
	284
7de3ee57	285	static int stripe_map(struct dm_target ti, struct bio bio)
1da177e4	286	{
65988525	287	struct stripe_c *sc = ti->private;
374bf7e7	288	uint32_t stripe;
55a62eef	289	unsigned target_bio_nr;
1da177e4	290
1eff9d32	291	if (bio->bi_opf & REQ_PREFLUSH) {
55a62eef AK	292	target_bio_nr = dm_bio_get_target_bio_nr(bio);
	293	BUG_ON(target_bio_nr >= sc->stripes);
	294	bio->bi_bdev = sc->stripe[target_bio_nr].dev->bdev;
374bf7e7 MP	295	return DM_MAPIO_REMAPPED;
374bf7e7 MP	296	}
e6047149	297	if (unlikely(bio_op(bio) == REQ_OP_DISCARD) \|\|
ac62d620	298	unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES) \|\|
e6047149	299	unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) {
55a62eef AK	300	target_bio_nr = dm_bio_get_target_bio_nr(bio);
	301	BUG_ON(target_bio_nr >= sc->stripes);
	302	return stripe_map_range(sc, bio, target_bio_nr);
7b76ec11	303	}
374bf7e7	304
4f024f37 KO	305	stripe_map_sector(sc, bio->bi_iter.bi_sector,
4f024f37 KO	306	&stripe, &bio->bi_iter.bi_sector);
1da177e4	307
4f024f37	308	bio->bi_iter.bi_sector += sc->stripe[stripe].physical_start;
1da177e4	309	bio->bi_bdev = sc->stripe[stripe].dev->bdev;
65988525	310
d2a7ad29	311	return DM_MAPIO_REMAPPED;
1da177e4 LT	312	}
1da177e4 LT	313
817bf402 DW	314	static long stripe_dax_direct_access(struct dm_target *ti, pgoff_t pgoff,
817bf402 DW	315	long nr_pages, void *kaddr, pfn_t pfn)
beec25b4	316	{
817bf402	317	sector_t dev_sector, sector = pgoff * PAGE_SECTORS;
beec25b4	318	struct stripe_c *sc = ti->private;
817bf402	319	struct dax_device *dax_dev;
beec25b4	320	struct block_device *bdev;
817bf402	321	uint32_t stripe;
beec25b4 TK	322	long ret;
beec25b4 TK	323
817bf402 DW	324	stripe_map_sector(sc, sector, &stripe, &dev_sector);
	325	dev_sector += sc->stripe[stripe].physical_start;
	326	dax_dev = sc->stripe[stripe].dev->dax_dev;
beec25b4 TK	327	bdev = sc->stripe[stripe].dev->bdev;
beec25b4 TK	328
817bf402 DW	329	ret = bdev_dax_pgoff(bdev, dev_sector, nr_pages * PAGE_SIZE, &pgoff);
	330	if (ret)
	331	return ret;
	332	return dax_direct_access(dax_dev, pgoff, nr_pages, kaddr, pfn);
beec25b4 TK	333	}
beec25b4 TK	334
7e026c8c DW	335	static size_t stripe_dax_copy_from_iter(struct dm_target *ti, pgoff_t pgoff,
	336	void addr, size_t bytes, struct iov_iter i)
	337	{
	338	sector_t dev_sector, sector = pgoff * PAGE_SECTORS;
	339	struct stripe_c *sc = ti->private;
	340	struct dax_device *dax_dev;
	341	struct block_device *bdev;
	342	uint32_t stripe;
	343
	344	stripe_map_sector(sc, sector, &stripe, &dev_sector);
	345	dev_sector += sc->stripe[stripe].physical_start;
	346	dax_dev = sc->stripe[stripe].dev->dax_dev;
	347	bdev = sc->stripe[stripe].dev->bdev;
	348
	349	if (bdev_dax_pgoff(bdev, dev_sector, ALIGN(bytes, PAGE_SIZE), &pgoff))
	350	return 0;
	351	return dax_copy_from_iter(dax_dev, pgoff, addr, bytes, i);
	352	}
	353
abebfbe2 DW	354	static void stripe_dax_flush(struct dm_target ti, pgoff_t pgoff, void addr,
	355	size_t size)
	356	{
	357	sector_t dev_sector, sector = pgoff * PAGE_SECTORS;
	358	struct stripe_c *sc = ti->private;
	359	struct dax_device *dax_dev;
	360	struct block_device *bdev;
	361	uint32_t stripe;
	362
	363	stripe_map_sector(sc, sector, &stripe, &dev_sector);
	364	dev_sector += sc->stripe[stripe].physical_start;
	365	dax_dev = sc->stripe[stripe].dev->dax_dev;
	366	bdev = sc->stripe[stripe].dev->bdev;
	367
	368	if (bdev_dax_pgoff(bdev, dev_sector, ALIGN(size, PAGE_SIZE), &pgoff))
	369	return;
	370	dax_flush(dax_dev, pgoff, addr, size);
	371	}
	372
4f7f5c67 BW	373	/*
	374	* Stripe status:
	375	*
	376	* INFO
	377	* #stripes [stripe_name <stripe_name>] [group word count]
	378	* [error count 'A\|D' <error count 'A\|D'>]
	379	*
	380	* TABLE
	381	* #stripes [stripe chunk size]
	382	* [stripe_name physical_start <stripe_name physical_start>]
	383	*
	384	*/
	385
fd7c092e MP	386	static void stripe_status(struct dm_target *ti, status_type_t type,
fd7c092e MP	387	unsigned status_flags, char *result, unsigned maxlen)
1da177e4 LT	388	{
1da177e4 LT	389	struct stripe_c sc = (struct stripe_c ) ti->private;
4f7f5c67	390	char buffer[sc->stripes + 1];
1da177e4 LT	391	unsigned int sz = 0;
	392	unsigned int i;
	393
	394	switch (type) {
	395	case STATUSTYPE_INFO:
4f7f5c67 BW	396	DMEMIT("%d ", sc->stripes);
	397	for (i = 0; i < sc->stripes; i++) {
	398	DMEMIT("%s ", sc->stripe[i].dev->name);
	399	buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ?
	400	'D' : 'A';
	401	}
	402	buffer[i] = '\0';
	403	DMEMIT("1 %s", buffer);
1da177e4 LT	404	break;
	405
	406	case STATUSTYPE_TABLE:
4ee218cd	407	DMEMIT("%d %llu", sc->stripes,
eb850de6	408	(unsigned long long)sc->chunk_size);
1da177e4	409	for (i = 0; i < sc->stripes; i++)
4ee218cd AM	410	DMEMIT(" %s %llu", sc->stripe[i].dev->name,
4ee218cd AM	411	(unsigned long long)sc->stripe[i].physical_start);
1da177e4 LT	412	break;
1da177e4 LT	413	}
1da177e4 LT	414	}
1da177e4 LT	415
4e4cbee9 CH	416	static int stripe_end_io(struct dm_target ti, struct bio bio,
4e4cbee9 CH	417	blk_status_t *error)
a25eb944 BW	418	{
	419	unsigned i;
	420	char major_minor[16];
	421	struct stripe_c *sc = ti->private;
	422
1be56909 CH	423	if (!*error)
1be56909 CH	424	return DM_ENDIO_DONE; /* I/O complete */
a25eb944	425
9966afaf	426	if (bio->bi_opf & REQ_RAHEAD)
1be56909	427	return DM_ENDIO_DONE;
a25eb944	428
4e4cbee9	429	if (*error == BLK_STS_NOTSUPP)
1be56909	430	return DM_ENDIO_DONE;
a25eb944 BW	431
	432	memset(major_minor, 0, sizeof(major_minor));
	433	sprintf(major_minor, "%d:%d",
f331c029 TH	434	MAJOR(disk_devt(bio->bi_bdev->bd_disk)),
f331c029 TH	435	MINOR(disk_devt(bio->bi_bdev->bd_disk)));
a25eb944 BW	436
	437	/*
	438	* Test to see which stripe drive triggered the event
	439	* and increment error count for all stripes on that device.
	440	* If the error count for a given device exceeds the threshold
	441	* value we will no longer trigger any further events.
	442	*/
	443	for (i = 0; i < sc->stripes; i++)
	444	if (!strcmp(sc->stripe[i].dev->name, major_minor)) {
	445	atomic_inc(&(sc->stripe[i].error_count));
	446	if (atomic_read(&(sc->stripe[i].error_count)) <
	447	DM_IO_ERROR_THRESHOLD)
f521f074	448	schedule_work(&sc->trigger_event);
a25eb944 BW	449	}
a25eb944 BW	450
1be56909	451	return DM_ENDIO_DONE;
a25eb944 BW	452	}
a25eb944 BW	453
af4874e0 MS	454	static int stripe_iterate_devices(struct dm_target *ti,
	455	iterate_devices_callout_fn fn, void *data)
	456	{
	457	struct stripe_c *sc = ti->private;
	458	int ret = 0;
	459	unsigned i = 0;
	460
5dea271b	461	do {
af4874e0	462	ret = fn(ti, sc->stripe[i].dev,
5dea271b MS	463	sc->stripe[i].physical_start,
	464	sc->stripe_width, data);
	465	} while (!ret && ++i < sc->stripes);
af4874e0 MS	466
	467	return ret;
	468	}
	469
40bea431 MS	470	static void stripe_io_hints(struct dm_target *ti,
	471	struct queue_limits *limits)
	472	{
	473	struct stripe_c *sc = ti->private;
eb850de6	474	unsigned chunk_size = sc->chunk_size << SECTOR_SHIFT;
40bea431 MS	475
40bea431 MS	476	blk_limits_io_min(limits, chunk_size);
3c5820c7	477	blk_limits_io_opt(limits, chunk_size * sc->stripes);
40bea431 MS	478	}
40bea431 MS	479
1da177e4 LT	480	static struct target_type stripe_target = {
1da177e4 LT	481	.name = "striped",
beec25b4	482	.version = {1, 6, 0},
e2460f2a	483	.features = DM_TARGET_PASSES_INTEGRITY,
1da177e4 LT	484	.module = THIS_MODULE,
	485	.ctr = stripe_ctr,
	486	.dtr = stripe_dtr,
	487	.map = stripe_map,
a25eb944	488	.end_io = stripe_end_io,
1da177e4	489	.status = stripe_status,
af4874e0	490	.iterate_devices = stripe_iterate_devices,
40bea431	491	.io_hints = stripe_io_hints,
817bf402	492	.direct_access = stripe_dax_direct_access,
7e026c8c	493	.dax_copy_from_iter = stripe_dax_copy_from_iter,
abebfbe2	494	.dax_flush = stripe_dax_flush,
1da177e4 LT	495	};
	496
	497	int __init dm_stripe_init(void)
	498	{
	499	int r;
	500
	501	r = dm_register_target(&stripe_target);
e223e1de	502	if (r < 0)
72d94861	503	DMWARN("target registration failed");
1da177e4 LT	504
	505	return r;
	506	}
	507
	508	void dm_stripe_exit(void)
	509	{
10d3bd09	510	dm_unregister_target(&stripe_target);
1da177e4	511	}