[deliverable/linux.git] / drivers / md / faulty.c

/*
 * faulty.c : Multiple Devices driver for Linux
 *
 * Copyright (C) 2004 Neil Brown
 *
 * fautly-device-simulator personality for md
 *
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2, or (at your option)
 * any later version.
 *
 * You should have received a copy of the GNU General Public License
 * (for example /usr/src/linux/COPYING); if not, write to the Free
 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */


/*
 * The "faulty" personality causes some requests to fail.
 *
 * Possible failure modes are:
 *   reads fail "randomly" but succeed on retry
 *   writes fail "randomly" but succeed on retry
 *   reads for some address fail and then persist until a write
 *   reads for some address fail and then persist irrespective of write
 *   writes for some address fail and persist
 *   all writes fail
 *
 * Different modes can be active at a time, but only
 * one can be set at array creation.  Others can be added later.
 * A mode can be one-shot or recurrent with the recurrance being
 * once in every N requests.
 * The bottom 5 bits of the "layout" indicate the mode.  The
 * remainder indicate a period, or 0 for one-shot.
 *
 * There is an implementation limit on the number of concurrently
 * persisting-faulty blocks. When a new fault is requested that would
 * exceed the limit, it is ignored.
 * All current faults can be clear using a layout of "0".
 *
 * Requests are always sent to the device.  If they are to fail,
 * we clone the bio and insert a new b_end_io into the chain.
 */

#define	WriteTransient	0
#define	ReadTransient	1
#define	WritePersistent	2
#define	ReadPersistent	3
#define	WriteAll	4 /* doesn't go to device */
#define	ReadFixable	5
#define	Modes	6

#define	ClearErrors	31
#define	ClearFaults	30

#define AllPersist	100 /* internal use only */
#define	NoPersist	101

#define	ModeMask	0x1f
#define	ModeShift	5

#define MaxFault	50
#include <linux/blkdev.h>
#include <linux/raid/md_u.h>
#include <linux/slab.h>
#include "md.h"
#include <linux/seq_file.h>


static void faulty_fail(struct bio *bio, int error)
{
	struct bio *b = bio->bi_private;

	b->bi_size = bio->bi_size;
	b->bi_sector = bio->bi_sector;

	bio_put(bio);

	bio_io_error(b);
}

typedef struct faulty_conf {
	int period[Modes];
	atomic_t counters[Modes];
	sector_t faults[MaxFault];
	int	modes[MaxFault];
	int nfaults;
	mdk_rdev_t *rdev;
} conf_t;

static int check_mode(conf_t *conf, int mode)
{
	if (conf->period[mode] == 0 &&
	    atomic_read(&conf->counters[mode]) <= 0)
		return 0; /* no failure, no decrement */


	if (atomic_dec_and_test(&conf->counters[mode])) {
		if (conf->period[mode])
			atomic_set(&conf->counters[mode], conf->period[mode]);
		return 1;
	}
	return 0;
}

static int check_sector(conf_t *conf, sector_t start, sector_t end, int dir)
{
	/* If we find a ReadFixable sector, we fix it ... */
	int i;
	for (i=0; i<conf->nfaults; i++)
		if (conf->faults[i] >= start &&
		    conf->faults[i] < end) {
			/* found it ... */
			switch (conf->modes[i] * 2 + dir) {
			case WritePersistent*2+WRITE: return 1;
			case ReadPersistent*2+READ: return 1;
			case ReadFixable*2+READ: return 1;
			case ReadFixable*2+WRITE:
				conf->modes[i] = NoPersist;
				return 0;
			case AllPersist*2+READ:
			case AllPersist*2+WRITE: return 1;
			default:
				return 0;
			}
		}
	return 0;
}

static void add_sector(conf_t *conf, sector_t start, int mode)
{
	int i;
	int n = conf->nfaults;
	for (i=0; i<conf->nfaults; i++)
		if (conf->faults[i] == start) {
			switch(mode) {
			case NoPersist: conf->modes[i] = mode; return;
			case WritePersistent:
				if (conf->modes[i] == ReadPersistent ||
				    conf->modes[i] == ReadFixable)
					conf->modes[i] = AllPersist;
				else
					conf->modes[i] = WritePersistent;
				return;
			case ReadPersistent:
				if (conf->modes[i] == WritePersistent)
					conf->modes[i] = AllPersist;
				else
					conf->modes[i] = ReadPersistent;
				return;
			case ReadFixable:
				if (conf->modes[i] == WritePersistent ||
				    conf->modes[i] == ReadPersistent)
					conf->modes[i] = AllPersist;
				else
					conf->modes[i] = ReadFixable;
				return;
			}
		} else if (conf->modes[i] == NoPersist)
			n = i;

	if (n >= MaxFault)
		return;
	conf->faults[n] = start;
	conf->modes[n] = mode;
	if (conf->nfaults == n)
		conf->nfaults = n+1;
}

static int make_request(mddev_t *mddev, struct bio *bio)
{
	conf_t *conf = mddev->private;
	int failit = 0;

	if (bio_data_dir(bio) == WRITE) {
		/* write request */
		if (atomic_read(&conf->counters[WriteAll])) {
			/* special case - don't decrement, don't generic_make_request,
			 * just fail immediately
			 */
			bio_endio(bio, -EIO);
			return 0;
		}

		if (check_sector(conf, bio->bi_sector, bio->bi_sector+(bio->bi_size>>9),
				 WRITE))
			failit = 1;
		if (check_mode(conf, WritePersistent)) {
			add_sector(conf, bio->bi_sector, WritePersistent);
			failit = 1;
		}
		if (check_mode(conf, WriteTransient))
			failit = 1;
	} else {
		/* read request */
		if (check_sector(conf, bio->bi_sector, bio->bi_sector + (bio->bi_size>>9),
				 READ))
			failit = 1;
		if (check_mode(conf, ReadTransient))
			failit = 1;
		if (check_mode(conf, ReadPersistent)) {
			add_sector(conf, bio->bi_sector, ReadPersistent);
			failit = 1;
		}
		if (check_mode(conf, ReadFixable)) {
			add_sector(conf, bio->bi_sector, ReadFixable);
			failit = 1;
		}
	}
	if (failit) {
		struct bio *b = bio_clone_mddev(bio, GFP_NOIO, mddev);
		b->bi_bdev = conf->rdev->bdev;
		b->bi_private = bio;
		b->bi_end_io = faulty_fail;
		generic_make_request(b);
		return 0;
	} else {
		bio->bi_bdev = conf->rdev->bdev;
		return 1;
	}
}

static void status(struct seq_file *seq, mddev_t *mddev)
{
	conf_t *conf = mddev->private;
	int n;

	if ((n=atomic_read(&conf->counters[WriteTransient])) != 0)
		seq_printf(seq, " WriteTransient=%d(%d)",
			   n, conf->period[WriteTransient]);

	if ((n=atomic_read(&conf->counters[ReadTransient])) != 0)
		seq_printf(seq, " ReadTransient=%d(%d)",
			   n, conf->period[ReadTransient]);

	if ((n=atomic_read(&conf->counters[WritePersistent])) != 0)
		seq_printf(seq, " WritePersistent=%d(%d)",
			   n, conf->period[WritePersistent]);

	if ((n=atomic_read(&conf->counters[ReadPersistent])) != 0)
		seq_printf(seq, " ReadPersistent=%d(%d)",
			   n, conf->period[ReadPersistent]);


	if ((n=atomic_read(&conf->counters[ReadFixable])) != 0)
		seq_printf(seq, " ReadFixable=%d(%d)",
			   n, conf->period[ReadFixable]);

	if ((n=atomic_read(&conf->counters[WriteAll])) != 0)
		seq_printf(seq, " WriteAll");

	seq_printf(seq, " nfaults=%d", conf->nfaults);
}


static int reshape(mddev_t *mddev)
{
	int mode = mddev->new_layout & ModeMask;
	int count = mddev->new_layout >> ModeShift;
	conf_t *conf = mddev->private;

	if (mddev->new_layout < 0)
		return 0;

	/* new layout */
	if (mode == ClearFaults)
		conf->nfaults = 0;
	else if (mode == ClearErrors) {
		int i;
		for (i=0 ; i < Modes ; i++) {
			conf->period[i] = 0;
			atomic_set(&conf->counters[i], 0);
		}
	} else if (mode < Modes) {
		conf->period[mode] = count;
		if (!count) count++;
		atomic_set(&conf->counters[mode], count);
	} else
		return -EINVAL;
	mddev->new_layout = -1;
	mddev->layout = -1; /* makes sure further changes come through */
	return 0;
}

static sector_t faulty_size(mddev_t *mddev, sector_t sectors, int raid_disks)
{
	WARN_ONCE(raid_disks,
		  "%s does not support generic reshape\n", __func__);

	if (sectors == 0)
		return mddev->dev_sectors;

	return sectors;
}

static int run(mddev_t *mddev)
{
	mdk_rdev_t *rdev;
	int i;
	conf_t *conf;

	if (md_check_no_bitmap(mddev))
		return -EINVAL;

	conf = kmalloc(sizeof(*conf), GFP_KERNEL);
	if (!conf)
		return -ENOMEM;

	for (i=0; i<Modes; i++) {
		atomic_set(&conf->counters[i], 0);
		conf->period[i] = 0;
	}
	conf->nfaults = 0;

	list_for_each_entry(rdev, &mddev->disks, same_set)
		conf->rdev = rdev;

	md_set_array_sectors(mddev, faulty_size(mddev, 0, 0));
	mddev->private = conf;

	reshape(mddev);

	return 0;
}

static int stop(mddev_t *mddev)
{
	conf_t *conf = mddev->private;

	kfree(conf);
	mddev->private = NULL;
	return 0;
}

static struct mdk_personality faulty_personality =
{
	.name		= "faulty",
	.level		= LEVEL_FAULTY,
	.owner		= THIS_MODULE,
	.make_request	= make_request,
	.run		= run,
	.stop		= stop,
	.status		= status,
	.check_reshape	= reshape,
	.size		= faulty_size,
};

static int __init raid_init(void)
{
	return register_md_personality(&faulty_personality);
}

static void raid_exit(void)
{
	unregister_md_personality(&faulty_personality);
}

module_init(raid_init);
module_exit(raid_exit);
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Fault injection personality for MD");
MODULE_ALIAS("md-personality-10"); /* faulty */
MODULE_ALIAS("md-faulty");
MODULE_ALIAS("md-level--5");
Commit	Line	Data
1da177e4 LT	1	/*
	2	* faulty.c : Multiple Devices driver for Linux
	3	*
	4	* Copyright (C) 2004 Neil Brown
	5	*
	6	* fautly-device-simulator personality for md
	7	*
	8	*
	9	* This program is free software; you can redistribute it and/or modify
	10	* it under the terms of the GNU General Public License as published by
	11	* the Free Software Foundation; either version 2, or (at your option)
	12	* any later version.
	13	*
	14	* You should have received a copy of the GNU General Public License
	15	* (for example /usr/src/linux/COPYING); if not, write to the Free
	16	* Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
	17	*/
	18
	19
	20	/*
	21	* The "faulty" personality causes some requests to fail.
	22	*
	23	* Possible failure modes are:
	24	* reads fail "randomly" but succeed on retry
	25	* writes fail "randomly" but succeed on retry
	26	* reads for some address fail and then persist until a write
	27	* reads for some address fail and then persist irrespective of write
	28	* writes for some address fail and persist
	29	* all writes fail
	30	*
	31	* Different modes can be active at a time, but only
	32	* one can be set at array creation. Others can be added later.
	33	* A mode can be one-shot or recurrent with the recurrance being
	34	* once in every N requests.
	35	* The bottom 5 bits of the "layout" indicate the mode. The
	36	* remainder indicate a period, or 0 for one-shot.
	37	*
	38	* There is an implementation limit on the number of concurrently
	39	* persisting-faulty blocks. When a new fault is requested that would
	40	* exceed the limit, it is ignored.
	41	* All current faults can be clear using a layout of "0".
	42	*
	43	* Requests are always sent to the device. If they are to fail,
	44	* we clone the bio and insert a new b_end_io into the chain.
	45	*/
	46
	47	#define WriteTransient 0
	48	#define ReadTransient 1
	49	#define WritePersistent 2
	50	#define ReadPersistent 3
	51	#define WriteAll 4 /* doesn't go to device */
	52	#define ReadFixable 5
	53	#define Modes 6
	54
	55	#define ClearErrors 31
	56	#define ClearFaults 30
	57
	58	#define AllPersist 100 /* internal use only */
	59	#define NoPersist 101
	60
	61	#define ModeMask 0x1f
	62	#define ModeShift 5
	63
	64	#define MaxFault 50
bff61975 N	65	#include <linux/blkdev.h>
bff61975 N	66	#include <linux/raid/md_u.h>
5a0e3ad6	67	#include <linux/slab.h>
43b2e5d8	68	#include "md.h"
bff61975	69	#include <linux/seq_file.h>
1da177e4 LT	70
1da177e4 LT	71
6712ecf8	72	static void faulty_fail(struct bio *bio, int error)
1da177e4 LT	73	{
	74	struct bio *b = bio->bi_private;
	75
	76	b->bi_size = bio->bi_size;
	77	b->bi_sector = bio->bi_sector;
	78
6712ecf8	79	bio_put(bio);
1da177e4	80
6712ecf8	81	bio_io_error(b);
1da177e4 LT	82	}
	83
	84	typedef struct faulty_conf {
	85	int period[Modes];
	86	atomic_t counters[Modes];
	87	sector_t faults[MaxFault];
	88	int modes[MaxFault];
	89	int nfaults;
	90	mdk_rdev_t *rdev;
	91	} conf_t;
	92
	93	static int check_mode(conf_t *conf, int mode)
	94	{
	95	if (conf->period[mode] == 0 &&
	96	atomic_read(&conf->counters[mode]) <= 0)
	97	return 0; /* no failure, no decrement */
	98
	99
	100	if (atomic_dec_and_test(&conf->counters[mode])) {
	101	if (conf->period[mode])
	102	atomic_set(&conf->counters[mode], conf->period[mode]);
	103	return 1;
	104	}
	105	return 0;
	106	}
	107
	108	static int check_sector(conf_t *conf, sector_t start, sector_t end, int dir)
	109	{
	110	/* If we find a ReadFixable sector, we fix it ... */
	111	int i;
	112	for (i=0; i<conf->nfaults; i++)
	113	if (conf->faults[i] >= start &&
	114	conf->faults[i] < end) {
	115	/* found it ... */
	116	switch (conf->modes[i] * 2 + dir) {
	117	case WritePersistent*2+WRITE: return 1;
	118	case ReadPersistent*2+READ: return 1;
	119	case ReadFixable*2+READ: return 1;
	120	case ReadFixable*2+WRITE:
	121	conf->modes[i] = NoPersist;
	122	return 0;
	123	case AllPersist*2+READ:
	124	case AllPersist*2+WRITE: return 1;
	125	default:
	126	return 0;
	127	}
	128	}
	129	return 0;
	130	}
	131
	132	static void add_sector(conf_t *conf, sector_t start, int mode)
	133	{
	134	int i;
	135	int n = conf->nfaults;
	136	for (i=0; i<conf->nfaults; i++)
	137	if (conf->faults[i] == start) {
	138	switch(mode) {
	139	case NoPersist: conf->modes[i] = mode; return;
	140	case WritePersistent:
	141	if (conf->modes[i] == ReadPersistent \|\|
	142	conf->modes[i] == ReadFixable)
	143	conf->modes[i] = AllPersist;
	144	else
	145	conf->modes[i] = WritePersistent;
146	return;
147	case ReadPersistent:
148	if (conf->modes[i] == WritePersistent)
149	conf->modes[i] = AllPersist;
150	else
151	conf->modes[i] = ReadPersistent;
152	return;
153	case ReadFixable:
154	if (conf->modes[i] == WritePersistent \|\|
155	conf->modes[i] == ReadPersistent)
156	conf->modes[i] = AllPersist;
157	else
158	conf->modes[i] = ReadFixable;
159	return;
160	}
161	} else if (conf->modes[i] == NoPersist)
162	n = i;
163
164	if (n >= MaxFault)
165	return;
166	conf->faults[n] = start;
167	conf->modes[n] = mode;
168	if (conf->nfaults == n)
169	conf->nfaults = n+1;
170	}
171
21a52c6d	172	static int make_request(mddev_t mddev, struct bio bio)
1da177e4	173	{
7b92813c	174	conf_t *conf = mddev->private;
1da177e4 LT	175	int failit = 0;
1da177e4 LT	176
802ba064	177	if (bio_data_dir(bio) == WRITE) {
1da177e4 LT	178	/* write request */
	179	if (atomic_read(&conf->counters[WriteAll])) {
	180	/* special case - don't decrement, don't generic_make_request,
	181	* just fail immediately
	182	*/
6712ecf8	183	bio_endio(bio, -EIO);
1da177e4 LT	184	return 0;
	185	}
	186
	187	if (check_sector(conf, bio->bi_sector, bio->bi_sector+(bio->bi_size>>9),
	188	WRITE))
	189	failit = 1;
	190	if (check_mode(conf, WritePersistent)) {
	191	add_sector(conf, bio->bi_sector, WritePersistent);
	192	failit = 1;
	193	}
	194	if (check_mode(conf, WriteTransient))
	195	failit = 1;
	196	} else {
	197	/* read request */
	198	if (check_sector(conf, bio->bi_sector, bio->bi_sector + (bio->bi_size>>9),
	199	READ))
	200	failit = 1;
	201	if (check_mode(conf, ReadTransient))
	202	failit = 1;
	203	if (check_mode(conf, ReadPersistent)) {
	204	add_sector(conf, bio->bi_sector, ReadPersistent);
	205	failit = 1;
	206	}
	207	if (check_mode(conf, ReadFixable)) {
	208	add_sector(conf, bio->bi_sector, ReadFixable);
	209	failit = 1;
	210	}
	211	}
	212	if (failit) {
a167f663	213	struct bio *b = bio_clone_mddev(bio, GFP_NOIO, mddev);
1da177e4 LT	214	b->bi_bdev = conf->rdev->bdev;
	215	b->bi_private = bio;
	216	b->bi_end_io = faulty_fail;
	217	generic_make_request(b);
	218	return 0;
	219	} else {
	220	bio->bi_bdev = conf->rdev->bdev;
	221	return 1;
	222	}
	223	}
	224
	225	static void status(struct seq_file seq, mddev_t mddev)
	226	{
7b92813c	227	conf_t *conf = mddev->private;
1da177e4 LT	228	int n;
	229
	230	if ((n=atomic_read(&conf->counters[WriteTransient])) != 0)
	231	seq_printf(seq, " WriteTransient=%d(%d)",
	232	n, conf->period[WriteTransient]);
	233
	234	if ((n=atomic_read(&conf->counters[ReadTransient])) != 0)
	235	seq_printf(seq, " ReadTransient=%d(%d)",
	236	n, conf->period[ReadTransient]);
	237
	238	if ((n=atomic_read(&conf->counters[WritePersistent])) != 0)
	239	seq_printf(seq, " WritePersistent=%d(%d)",
	240	n, conf->period[WritePersistent]);
	241
	242	if ((n=atomic_read(&conf->counters[ReadPersistent])) != 0)
	243	seq_printf(seq, " ReadPersistent=%d(%d)",
	244	n, conf->period[ReadPersistent]);
	245
	246
	247	if ((n=atomic_read(&conf->counters[ReadFixable])) != 0)
	248	seq_printf(seq, " ReadFixable=%d(%d)",
	249	n, conf->period[ReadFixable]);
	250
	251	if ((n=atomic_read(&conf->counters[WriteAll])) != 0)
	252	seq_printf(seq, " WriteAll");
	253
	254	seq_printf(seq, " nfaults=%d", conf->nfaults);
	255	}
	256
	257
50ac168a	258	static int reshape(mddev_t *mddev)
1da177e4	259	{
597a711b N	260	int mode = mddev->new_layout & ModeMask;
597a711b N	261	int count = mddev->new_layout >> ModeShift;
1da177e4 LT	262	conf_t *conf = mddev->private;
1da177e4 LT	263
597a711b N	264	if (mddev->new_layout < 0)
597a711b N	265	return 0;
1da177e4 LT	266
	267	/* new layout */
	268	if (mode == ClearFaults)
	269	conf->nfaults = 0;
	270	else if (mode == ClearErrors) {
	271	int i;
	272	for (i=0 ; i < Modes ; i++) {
	273	conf->period[i] = 0;
	274	atomic_set(&conf->counters[i], 0);
	275	}
	276	} else if (mode < Modes) {
	277	conf->period[mode] = count;
	278	if (!count) count++;
	279	atomic_set(&conf->counters[mode], count);
	280	} else
	281	return -EINVAL;
597a711b	282	mddev->new_layout = -1;
1da177e4 LT	283	mddev->layout = -1; /* makes sure further changes come through */
	284	return 0;
	285	}
	286
80c3a6ce DW	287	static sector_t faulty_size(mddev_t *mddev, sector_t sectors, int raid_disks)
	288	{
	289	WARN_ONCE(raid_disks,
	290	"%s does not support generic reshape\n", __func__);
	291
	292	if (sectors == 0)
	293	return mddev->dev_sectors;
	294
	295	return sectors;
	296	}
	297
1da177e4 LT	298	static int run(mddev_t *mddev)
	299	{
	300	mdk_rdev_t *rdev;
1da177e4	301	int i;
0894cc30	302	conf_t *conf;
1da177e4	303
0894cc30 AN	304	if (md_check_no_bitmap(mddev))
	305	return -EINVAL;
	306
	307	conf = kmalloc(sizeof(*conf), GFP_KERNEL);
08ff39f1 SW	308	if (!conf)
08ff39f1 SW	309	return -ENOMEM;
1da177e4 LT	310
	311	for (i=0; i<Modes; i++) {
	312	atomic_set(&conf->counters[i], 0);
	313	conf->period[i] = 0;
	314	}
	315	conf->nfaults = 0;
	316
159ec1fc	317	list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4 LT	318	conf->rdev = rdev;
1da177e4 LT	319
1f403624	320	md_set_array_sectors(mddev, faulty_size(mddev, 0, 0));
1da177e4 LT	321	mddev->private = conf;
1da177e4 LT	322
50ac168a	323	reshape(mddev);
1da177e4 LT	324
	325	return 0;
	326	}
	327
	328	static int stop(mddev_t *mddev)
	329	{
7b92813c	330	conf_t *conf = mddev->private;
1da177e4 LT	331
	332	kfree(conf);
	333	mddev->private = NULL;
	334	return 0;
	335	}
	336
2604b703	337	static struct mdk_personality faulty_personality =
1da177e4 LT	338	{
1da177e4 LT	339	.name = "faulty",
2604b703	340	.level = LEVEL_FAULTY,
1da177e4 LT	341	.owner = THIS_MODULE,
	342	.make_request = make_request,
	343	.run = run,
	344	.stop = stop,
	345	.status = status,
50ac168a	346	.check_reshape = reshape,
80c3a6ce	347	.size = faulty_size,
1da177e4 LT	348	};
	349
	350	static int __init raid_init(void)
	351	{
2604b703	352	return register_md_personality(&faulty_personality);
1da177e4 LT	353	}
	354
	355	static void raid_exit(void)
	356	{
2604b703	357	unregister_md_personality(&faulty_personality);
1da177e4 LT	358	}
	359
	360	module_init(raid_init);
	361	module_exit(raid_exit);
	362	MODULE_LICENSE("GPL");
0efb9e61	363	MODULE_DESCRIPTION("Fault injection personality for MD");
1da177e4	364	MODULE_ALIAS("md-personality-10"); /* faulty */
d9d166c2	365	MODULE_ALIAS("md-faulty");
2604b703	366	MODULE_ALIAS("md-level--5");