[net-next-2.6.git] / drivers / md / dm-stripe.c

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

#include <linux/device-mapper.h>

#include <linux/module.h>
#include <linux/init.h>
#include <linux/blkdev.h>
#include <linux/bio.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;

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

	/* stripe chunk size */
	uint32_t chunk_shift;
	sector_t chunk_mask;

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

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

	struct stripe stripe[0];
};

static struct workqueue_struct *kstriped;

/*
 * 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, kstriped_ws);

	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;

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

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

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

	return 0;
}

/*
 * Construct a striped mapping.
 * <number of stripes> <chunk size (2^^n)> [<dev_path> <offset>]+
 */
static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv)
{
	struct stripe_c *sc;
	sector_t width;
	uint32_t stripes;
	uint32_t chunk_size;
	char *end;
	int r;
	unsigned int i;

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

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

	chunk_size = simple_strtoul(argv[1], &end, 10);
	if (*end) {
		ti->error = "Invalid chunk_size";
		return -EINVAL;
	}

	/*
	 * chunk_size is a power of two
	 */
	if (!is_power_of_2(chunk_size) ||
	    (chunk_size < (PAGE_SIZE >> SECTOR_SHIFT))) {
		ti->error = "Invalid chunk size";
		return -EINVAL;
	}

	if (ti->len & (chunk_size - 1)) {
		ti->error = "Target length not divisible by "
		    "chunk size";
		return -EINVAL;
	}

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

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

	sc->stripes = stripes;
	sc->stripe_width = width;
	ti->split_io = chunk_size;
	ti->num_flush_requests = stripes;

	sc->chunk_mask = ((sector_t) chunk_size) - 1;
	for (sc->chunk_shift = 0; chunk_size; sc->chunk_shift++)
		chunk_size >>= 1;
	sc->chunk_shift--;

	/*
	 * 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_workqueue(kstriped);
	kfree(sc);
}

static int stripe_map(struct dm_target *ti, struct bio *bio,
		      union map_info *map_context)
{
	struct stripe_c *sc = (struct stripe_c *) ti->private;
	sector_t offset, chunk;
	uint32_t stripe;

	if (unlikely(bio_empty_barrier(bio))) {
		BUG_ON(map_context->flush_request >= sc->stripes);
		bio->bi_bdev = sc->stripe[map_context->flush_request].dev->bdev;
		return DM_MAPIO_REMAPPED;
	}

	offset = bio->bi_sector - ti->begin;
	chunk = offset >> sc->chunk_shift;
	stripe = sector_div(chunk, sc->stripes);

	bio->bi_bdev = sc->stripe[stripe].dev->bdev;
	bio->bi_sector = sc->stripe[stripe].physical_start +
	    (chunk << sc->chunk_shift) + (offset & sc->chunk_mask);
	return DM_MAPIO_REMAPPED;
}

/*
 * 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 int stripe_status(struct dm_target *ti,
			 status_type_t type, char *result, unsigned int 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_mask + 1);
		for (i = 0; i < sc->stripes; i++)
			DMEMIT(" %s %llu", sc->stripe[i].dev->name,
			    (unsigned long long)sc->stripe[i].physical_start);
		break;
	}
	return 0;
}

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

	if (!error)
		return 0; /* I/O complete */

	if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
		return error;

	if (error == -EOPNOTSUPP)
		return error;

	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)
				queue_work(kstriped, &sc->kstriped_ws);
		}

	return error;
}

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 struct target_type stripe_target = {
	.name   = "striped",
	.version = {1, 2, 0},
	.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,
};

int __init dm_stripe_init(void)
{
	int r;

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

	kstriped = create_singlethread_workqueue("kstriped");
	if (!kstriped) {
		DMERR("failed to create workqueue kstriped");
		dm_unregister_target(&stripe_target);
		return -ENOMEM;
	}

	return r;
}

void dm_stripe_exit(void)
{
	dm_unregister_target(&stripe_target);
	destroy_workqueue(kstriped);

	return;
}
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
586e80e6	7	#include <linux/device-mapper.h>
1da177e4 LT	8
	9	#include <linux/module.h>
	10	#include <linux/init.h>
	11	#include <linux/blkdev.h>
	12	#include <linux/bio.h>
	13	#include <linux/slab.h>
6f3c3f0a	14	#include <linux/log2.h>
1da177e4	15
72d94861	16	#define DM_MSG_PREFIX "striped"
a25eb944	17	#define DM_IO_ERROR_THRESHOLD 15
72d94861	18
1da177e4 LT	19	struct stripe {
	20	struct dm_dev *dev;
	21	sector_t physical_start;
a25eb944 BW	22
a25eb944 BW	23	atomic_t error_count;
1da177e4 LT	24	};
	25
	26	struct stripe_c {
	27	uint32_t stripes;
	28
	29	/* The size of this target / num. stripes */
	30	sector_t stripe_width;
	31
	32	/* stripe chunk size */
	33	uint32_t chunk_shift;
	34	sector_t chunk_mask;
	35
a25eb944 BW	36	/* Needed for handling events */
	37	struct dm_target *ti;
	38
	39	/* Work struct used for triggering events*/
	40	struct work_struct kstriped_ws;
	41
1da177e4 LT	42	struct stripe stripe[0];
	43	};
	44
a25eb944 BW	45	static struct workqueue_struct *kstriped;
	46
	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	{
	53	struct stripe_c *sc = container_of(work, struct stripe_c, kstriped_ws);
	54
	55	dm_table_event(sc->ti->table);
	56
	57	}
	58
1da177e4 LT	59	static inline struct stripe_c *alloc_context(unsigned int stripes)
	60	{
	61	size_t len;
	62
d63a5ce3 MP	63	if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe),
d63a5ce3 MP	64	stripes))
1da177e4 LT	65	return NULL;
	66
	67	len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes);
	68
	69	return kmalloc(len, GFP_KERNEL);
	70	}
	71
	72	/*
	73	* Parse a single <dev> <sector> pair
	74	*/
	75	static int get_stripe(struct dm_target ti, struct stripe_c sc,
	76	unsigned int stripe, char **argv)
	77	{
4ee218cd	78	unsigned long long start;
1da177e4	79
4ee218cd	80	if (sscanf(argv[1], "%llu", &start) != 1)
1da177e4 LT	81	return -EINVAL;
	82
	83	if (dm_get_device(ti, argv[0], start, sc->stripe_width,
	84	dm_table_get_mode(ti->table),
	85	&sc->stripe[stripe].dev))
	86	return -ENXIO;
	87
	88	sc->stripe[stripe].physical_start = start;
a25eb944	89
1da177e4 LT	90	return 0;
	91	}
	92
	93	/*
	94	* Construct a striped mapping.
	95	* <number of stripes> <chunk size (2^^n)> [<dev_path> <offset>]+
	96	*/
	97	static int stripe_ctr(struct dm_target ti, unsigned int argc, char *argv)
	98	{
	99	struct stripe_c *sc;
	100	sector_t width;
	101	uint32_t stripes;
	102	uint32_t chunk_size;
	103	char *end;
	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
	112	stripes = simple_strtoul(argv[0], &end, 10);
	113	if (*end) {
72d94861	114	ti->error = "Invalid stripe count";
1da177e4 LT	115	return -EINVAL;
	116	}
	117
	118	chunk_size = simple_strtoul(argv[1], &end, 10);
	119	if (*end) {
72d94861	120	ti->error = "Invalid chunk_size";
1da177e4 LT	121	return -EINVAL;
	122	}
	123
	124	/*
	125	* chunk_size is a power of two
	126	*/
6f3c3f0a	127	if (!is_power_of_2(chunk_size) \|\|
1da177e4	128	(chunk_size < (PAGE_SIZE >> SECTOR_SHIFT))) {
72d94861	129	ti->error = "Invalid chunk size";
1da177e4 LT	130	return -EINVAL;
	131	}
	132
a22c96c7	133	if (ti->len & (chunk_size - 1)) {
72d94861	134	ti->error = "Target length not divisible by "
8ba32fde KC	135	"chunk size";
	136	return -EINVAL;
	137	}
	138
1da177e4 LT	139	width = ti->len;
1da177e4 LT	140	if (sector_div(width, stripes)) {
72d94861	141	ti->error = "Target length not divisible by "
1da177e4 LT	142	"number of stripes";
	143	return -EINVAL;
	144	}
	145
	146	/*
	147	* Do we have enough arguments for that many stripes ?
	148	*/
	149	if (argc != (2 + 2 * stripes)) {
72d94861	150	ti->error = "Not enough destinations "
1da177e4 LT	151	"specified";
	152	return -EINVAL;
	153	}
	154
	155	sc = alloc_context(stripes);
	156	if (!sc) {
72d94861	157	ti->error = "Memory allocation for striped context "
1da177e4 LT	158	"failed";
	159	return -ENOMEM;
	160	}
	161
a25eb944 BW	162	INIT_WORK(&sc->kstriped_ws, trigger_event);
	163
	164	/* Set pointer to dm target; used in trigger_event */
	165	sc->ti = ti;
	166
1da177e4 LT	167	sc->stripes = stripes;
	168	sc->stripe_width = width;
	169	ti->split_io = chunk_size;
374bf7e7	170	ti->num_flush_requests = stripes;
1da177e4 LT	171
	172	sc->chunk_mask = ((sector_t) chunk_size) - 1;
	173	for (sc->chunk_shift = 0; chunk_size; sc->chunk_shift++)
	174	chunk_size >>= 1;
	175	sc->chunk_shift--;
	176
	177	/*
	178	* Get the stripe destinations.
	179	*/
	180	for (i = 0; i < stripes; i++) {
	181	argv += 2;
	182
	183	r = get_stripe(ti, sc, i, argv);
	184	if (r < 0) {
72d94861	185	ti->error = "Couldn't parse stripe destination";
1da177e4 LT	186	while (i--)
	187	dm_put_device(ti, sc->stripe[i].dev);
	188	kfree(sc);
	189	return r;
	190	}
a25eb944	191	atomic_set(&(sc->stripe[i].error_count), 0);
1da177e4 LT	192	}
	193
	194	ti->private = sc;
a25eb944	195
1da177e4 LT	196	return 0;
	197	}
	198
	199	static void stripe_dtr(struct dm_target *ti)
	200	{
	201	unsigned int i;
	202	struct stripe_c sc = (struct stripe_c ) ti->private;
	203
	204	for (i = 0; i < sc->stripes; i++)
	205	dm_put_device(ti, sc->stripe[i].dev);
	206
a25eb944	207	flush_workqueue(kstriped);
1da177e4 LT	208	kfree(sc);
	209	}
	210
	211	static int stripe_map(struct dm_target ti, struct bio bio,
	212	union map_info *map_context)
	213	{
	214	struct stripe_c sc = (struct stripe_c ) ti->private;
374bf7e7 MP	215	sector_t offset, chunk;
374bf7e7 MP	216	uint32_t stripe;
1da177e4	217
374bf7e7 MP	218	if (unlikely(bio_empty_barrier(bio))) {
	219	BUG_ON(map_context->flush_request >= sc->stripes);
	220	bio->bi_bdev = sc->stripe[map_context->flush_request].dev->bdev;
	221	return DM_MAPIO_REMAPPED;
	222	}
	223
	224	offset = bio->bi_sector - ti->begin;
	225	chunk = offset >> sc->chunk_shift;
	226	stripe = sector_div(chunk, sc->stripes);
1da177e4 LT	227
	228	bio->bi_bdev = sc->stripe[stripe].dev->bdev;
	229	bio->bi_sector = sc->stripe[stripe].physical_start +
	230	(chunk << sc->chunk_shift) + (offset & sc->chunk_mask);
d2a7ad29	231	return DM_MAPIO_REMAPPED;
1da177e4 LT	232	}
1da177e4 LT	233
4f7f5c67 BW	234	/*
	235	* Stripe status:
	236	*
	237	* INFO
	238	* #stripes [stripe_name <stripe_name>] [group word count]
	239	* [error count 'A\|D' <error count 'A\|D'>]
	240	*
	241	* TABLE
	242	* #stripes [stripe chunk size]
	243	* [stripe_name physical_start <stripe_name physical_start>]
	244	*
	245	*/
	246
1da177e4 LT	247	static int stripe_status(struct dm_target *ti,
	248	status_type_t type, char *result, unsigned int maxlen)
	249	{
	250	struct stripe_c sc = (struct stripe_c ) ti->private;
4f7f5c67	251	char buffer[sc->stripes + 1];
1da177e4 LT	252	unsigned int sz = 0;
	253	unsigned int i;
	254
	255	switch (type) {
	256	case STATUSTYPE_INFO:
4f7f5c67 BW	257	DMEMIT("%d ", sc->stripes);
	258	for (i = 0; i < sc->stripes; i++) {
	259	DMEMIT("%s ", sc->stripe[i].dev->name);
	260	buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ?
	261	'D' : 'A';
	262	}
	263	buffer[i] = '\0';
	264	DMEMIT("1 %s", buffer);
1da177e4 LT	265	break;
	266
	267	case STATUSTYPE_TABLE:
4ee218cd AM	268	DMEMIT("%d %llu", sc->stripes,
4ee218cd AM	269	(unsigned long long)sc->chunk_mask + 1);
1da177e4	270	for (i = 0; i < sc->stripes; i++)
4ee218cd AM	271	DMEMIT(" %s %llu", sc->stripe[i].dev->name,
4ee218cd AM	272	(unsigned long long)sc->stripe[i].physical_start);
1da177e4 LT	273	break;
	274	}
	275	return 0;
	276	}
	277
a25eb944 BW	278	static int stripe_end_io(struct dm_target ti, struct bio bio,
	279	int error, union map_info *map_context)
	280	{
	281	unsigned i;
	282	char major_minor[16];
	283	struct stripe_c *sc = ti->private;
	284
	285	if (!error)
	286	return 0; /* I/O complete */
	287
	288	if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
	289	return error;
	290
	291	if (error == -EOPNOTSUPP)
	292	return error;
	293
	294	memset(major_minor, 0, sizeof(major_minor));
	295	sprintf(major_minor, "%d:%d",
f331c029 TH	296	MAJOR(disk_devt(bio->bi_bdev->bd_disk)),
f331c029 TH	297	MINOR(disk_devt(bio->bi_bdev->bd_disk)));
a25eb944 BW	298
	299	/*
	300	* Test to see which stripe drive triggered the event
	301	* and increment error count for all stripes on that device.
	302	* If the error count for a given device exceeds the threshold
	303	* value we will no longer trigger any further events.
	304	*/
	305	for (i = 0; i < sc->stripes; i++)
	306	if (!strcmp(sc->stripe[i].dev->name, major_minor)) {
	307	atomic_inc(&(sc->stripe[i].error_count));
	308	if (atomic_read(&(sc->stripe[i].error_count)) <
	309	DM_IO_ERROR_THRESHOLD)
	310	queue_work(kstriped, &sc->kstriped_ws);
	311	}
	312
	313	return error;
	314	}
	315
af4874e0 MS	316	static int stripe_iterate_devices(struct dm_target *ti,
	317	iterate_devices_callout_fn fn, void *data)
	318	{
	319	struct stripe_c *sc = ti->private;
	320	int ret = 0;
	321	unsigned i = 0;
	322
5dea271b	323	do {
af4874e0	324	ret = fn(ti, sc->stripe[i].dev,
5dea271b MS	325	sc->stripe[i].physical_start,
	326	sc->stripe_width, data);
	327	} while (!ret && ++i < sc->stripes);
af4874e0 MS	328
	329	return ret;
	330	}
	331
1da177e4 LT	332	static struct target_type stripe_target = {
1da177e4 LT	333	.name = "striped",
af4874e0	334	.version = {1, 2, 0},
1da177e4 LT	335	.module = THIS_MODULE,
	336	.ctr = stripe_ctr,
	337	.dtr = stripe_dtr,
	338	.map = stripe_map,
a25eb944	339	.end_io = stripe_end_io,
1da177e4	340	.status = stripe_status,
af4874e0	341	.iterate_devices = stripe_iterate_devices,
1da177e4 LT	342	};
	343
	344	int __init dm_stripe_init(void)
	345	{
	346	int r;
	347
	348	r = dm_register_target(&stripe_target);
6edebdee	349	if (r < 0) {
72d94861	350	DMWARN("target registration failed");
6edebdee HM	351	return r;
6edebdee HM	352	}
1da177e4	353
a25eb944 BW	354	kstriped = create_singlethread_workqueue("kstriped");
	355	if (!kstriped) {
	356	DMERR("failed to create workqueue kstriped");
	357	dm_unregister_target(&stripe_target);
	358	return -ENOMEM;
	359	}
	360
1da177e4 LT	361	return r;
	362	}
	363
	364	void dm_stripe_exit(void)
	365	{
10d3bd09	366	dm_unregister_target(&stripe_target);
a25eb944 BW	367	destroy_workqueue(kstriped);
a25eb944 BW	368
1da177e4 LT	369	return;
1da177e4 LT	370	}