[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 recurrence 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/module.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);
}

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

static int check_mode(struct faulty_conf *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(struct faulty_conf *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(struct faulty_conf *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 void make_request(struct mddev *mddev, struct bio *bio)
{
	struct faulty_conf *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;
		}

		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;
		bio = b;
	} else
		bio->bi_bdev = conf->rdev->bdev;

	generic_make_request(bio);
}

static void status(struct seq_file *seq, struct mddev *mddev)
{
	struct faulty_conf *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(struct mddev *mddev)
{
	int mode = mddev->new_layout & ModeMask;
	int count = mddev->new_layout >> ModeShift;
	struct faulty_conf *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(struct mddev *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(struct mddev *mddev)
{
	struct md_rdev *rdev;
	int i;
	struct faulty_conf *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(struct mddev *mddev)
{
	struct faulty_conf *conf = mddev->private;

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

static struct md_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.
25985edc	33	* A mode can be one-shot or recurrent with the recurrence being
1da177e4 LT	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	65	#include <linux/blkdev.h>
056075c7	66	#include <linux/module.h>
bff61975	67	#include <linux/raid/md_u.h>
5a0e3ad6	68	#include <linux/slab.h>
43b2e5d8	69	#include "md.h"
bff61975	70	#include <linux/seq_file.h>
1da177e4 LT	71
1da177e4 LT	72
6712ecf8	73	static void faulty_fail(struct bio *bio, int error)
1da177e4 LT	74	{
	75	struct bio *b = bio->bi_private;
	76
	77	b->bi_size = bio->bi_size;
	78	b->bi_sector = bio->bi_sector;
	79
6712ecf8	80	bio_put(bio);
1da177e4	81
6712ecf8	82	bio_io_error(b);
1da177e4 LT	83	}
1da177e4 LT	84
8f1ae43d	85	struct faulty_conf {
1da177e4 LT	86	int period[Modes];
	87	atomic_t counters[Modes];
	88	sector_t faults[MaxFault];
	89	int modes[MaxFault];
	90	int nfaults;
3cb03002	91	struct md_rdev *rdev;
8f1ae43d	92	};
1da177e4	93
8f1ae43d	94	static int check_mode(struct faulty_conf *conf, int mode)
1da177e4 LT	95	{
	96	if (conf->period[mode] == 0 &&
	97	atomic_read(&conf->counters[mode]) <= 0)
	98	return 0; /* no failure, no decrement */
	99
	100
	101	if (atomic_dec_and_test(&conf->counters[mode])) {
	102	if (conf->period[mode])
	103	atomic_set(&conf->counters[mode], conf->period[mode]);
	104	return 1;
	105	}
	106	return 0;
	107	}
	108
8f1ae43d	109	static int check_sector(struct faulty_conf *conf, sector_t start, sector_t end, int dir)
1da177e4 LT	110	{
	111	/* If we find a ReadFixable sector, we fix it ... */
	112	int i;
	113	for (i=0; i<conf->nfaults; i++)
	114	if (conf->faults[i] >= start &&
	115	conf->faults[i] < end) {
	116	/* found it ... */
	117	switch (conf->modes[i] * 2 + dir) {
	118	case WritePersistent*2+WRITE: return 1;
	119	case ReadPersistent*2+READ: return 1;
	120	case ReadFixable*2+READ: return 1;
	121	case ReadFixable*2+WRITE:
	122	conf->modes[i] = NoPersist;
	123	return 0;
	124	case AllPersist*2+READ:
	125	case AllPersist*2+WRITE: return 1;
	126	default:
	127	return 0;
	128	}
	129	}
	130	return 0;
	131	}
	132
8f1ae43d	133	static void add_sector(struct faulty_conf *conf, sector_t start, int mode)
1da177e4 LT	134	{
	135	int i;
	136	int n = conf->nfaults;
	137	for (i=0; i<conf->nfaults; i++)
	138	if (conf->faults[i] == start) {
	139	switch(mode) {
	140	case NoPersist: conf->modes[i] = mode; return;
	141	case WritePersistent:
	142	if (conf->modes[i] == ReadPersistent \|\|
	143	conf->modes[i] == ReadFixable)
	144	conf->modes[i] = AllPersist;
	145	else
	146	conf->modes[i] = WritePersistent;
	147	return;
	148	case ReadPersistent:
	149	if (conf->modes[i] == WritePersistent)
	150	conf->modes[i] = AllPersist;
	151	else
	152	conf->modes[i] = ReadPersistent;
	153	return;
	154	case ReadFixable:
	155	if (conf->modes[i] == WritePersistent \|\|
	156	conf->modes[i] == ReadPersistent)
	157	conf->modes[i] = AllPersist;
	158	else
	159	conf->modes[i] = ReadFixable;
	160	return;
	161	}
	162	} else if (conf->modes[i] == NoPersist)
	163	n = i;
	164
	165	if (n >= MaxFault)
	166	return;
	167	conf->faults[n] = start;
	168	conf->modes[n] = mode;
	169	if (conf->nfaults == n)
	170	conf->nfaults = n+1;
	171	}
	172
b4fdcb02	173	static void make_request(struct mddev mddev, struct bio bio)
1da177e4	174	{
8f1ae43d	175	struct faulty_conf *conf = mddev->private;
1da177e4 LT	176	int failit = 0;
1da177e4 LT	177
802ba064	178	if (bio_data_dir(bio) == WRITE) {
1da177e4 LT	179	/* write request */
	180	if (atomic_read(&conf->counters[WriteAll])) {
	181	/* special case - don't decrement, don't generic_make_request,
	182	* just fail immediately
	183	*/
6712ecf8	184	bio_endio(bio, -EIO);
5a7bbad2	185	return;
1da177e4 LT	186	}
	187
	188	if (check_sector(conf, bio->bi_sector, bio->bi_sector+(bio->bi_size>>9),
	189	WRITE))
	190	failit = 1;
	191	if (check_mode(conf, WritePersistent)) {
	192	add_sector(conf, bio->bi_sector, WritePersistent);
	193	failit = 1;
	194	}
	195	if (check_mode(conf, WriteTransient))
	196	failit = 1;
	197	} else {
	198	/* read request */
	199	if (check_sector(conf, bio->bi_sector, bio->bi_sector + (bio->bi_size>>9),
	200	READ))
	201	failit = 1;
	202	if (check_mode(conf, ReadTransient))
	203	failit = 1;
	204	if (check_mode(conf, ReadPersistent)) {
	205	add_sector(conf, bio->bi_sector, ReadPersistent);
	206	failit = 1;
	207	}
	208	if (check_mode(conf, ReadFixable)) {
	209	add_sector(conf, bio->bi_sector, ReadFixable);
	210	failit = 1;
	211	}
	212	}
	213	if (failit) {
a167f663	214	struct bio *b = bio_clone_mddev(bio, GFP_NOIO, mddev);
5a7bbad2	215
1da177e4 LT	216	b->bi_bdev = conf->rdev->bdev;
	217	b->bi_private = bio;
	218	b->bi_end_io = faulty_fail;
5a7bbad2 CH	219	bio = b;
5a7bbad2 CH	220	} else
1da177e4	221	bio->bi_bdev = conf->rdev->bdev;
5a7bbad2 CH	222
5a7bbad2 CH	223	generic_make_request(bio);
1da177e4 LT	224	}
1da177e4 LT	225
fd01b88c	226	static void status(struct seq_file seq, struct mddev mddev)
1da177e4	227	{
8f1ae43d	228	struct faulty_conf *conf = mddev->private;
1da177e4 LT	229	int n;
	230
	231	if ((n=atomic_read(&conf->counters[WriteTransient])) != 0)
	232	seq_printf(seq, " WriteTransient=%d(%d)",
	233	n, conf->period[WriteTransient]);
	234
	235	if ((n=atomic_read(&conf->counters[ReadTransient])) != 0)
	236	seq_printf(seq, " ReadTransient=%d(%d)",
	237	n, conf->period[ReadTransient]);
	238
	239	if ((n=atomic_read(&conf->counters[WritePersistent])) != 0)
	240	seq_printf(seq, " WritePersistent=%d(%d)",
	241	n, conf->period[WritePersistent]);
	242
	243	if ((n=atomic_read(&conf->counters[ReadPersistent])) != 0)
	244	seq_printf(seq, " ReadPersistent=%d(%d)",
	245	n, conf->period[ReadPersistent]);
	246
	247
	248	if ((n=atomic_read(&conf->counters[ReadFixable])) != 0)
	249	seq_printf(seq, " ReadFixable=%d(%d)",
	250	n, conf->period[ReadFixable]);
	251
	252	if ((n=atomic_read(&conf->counters[WriteAll])) != 0)
	253	seq_printf(seq, " WriteAll");
	254
	255	seq_printf(seq, " nfaults=%d", conf->nfaults);
	256	}
	257
	258
fd01b88c	259	static int reshape(struct mddev *mddev)
1da177e4	260	{
597a711b N	261	int mode = mddev->new_layout & ModeMask;
597a711b N	262	int count = mddev->new_layout >> ModeShift;
8f1ae43d	263	struct faulty_conf *conf = mddev->private;
1da177e4	264
597a711b N	265	if (mddev->new_layout < 0)
597a711b N	266	return 0;
1da177e4 LT	267
	268	/* new layout */
	269	if (mode == ClearFaults)
	270	conf->nfaults = 0;
	271	else if (mode == ClearErrors) {
	272	int i;
	273	for (i=0 ; i < Modes ; i++) {
	274	conf->period[i] = 0;
	275	atomic_set(&conf->counters[i], 0);
	276	}
	277	} else if (mode < Modes) {
	278	conf->period[mode] = count;
	279	if (!count) count++;
	280	atomic_set(&conf->counters[mode], count);
	281	} else
	282	return -EINVAL;
597a711b	283	mddev->new_layout = -1;
1da177e4 LT	284	mddev->layout = -1; /* makes sure further changes come through */
	285	return 0;
	286	}
	287
fd01b88c	288	static sector_t faulty_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce DW	289	{
	290	WARN_ONCE(raid_disks,
	291	"%s does not support generic reshape\n", __func__);
	292
	293	if (sectors == 0)
	294	return mddev->dev_sectors;
	295
	296	return sectors;
	297	}
	298
fd01b88c	299	static int run(struct mddev *mddev)
1da177e4	300	{
3cb03002	301	struct md_rdev *rdev;
1da177e4	302	int i;
8f1ae43d	303	struct faulty_conf *conf;
1da177e4	304
0894cc30 AN	305	if (md_check_no_bitmap(mddev))
	306	return -EINVAL;
	307
	308	conf = kmalloc(sizeof(*conf), GFP_KERNEL);
08ff39f1 SW	309	if (!conf)
08ff39f1 SW	310	return -ENOMEM;
1da177e4 LT	311
	312	for (i=0; i<Modes; i++) {
	313	atomic_set(&conf->counters[i], 0);
	314	conf->period[i] = 0;
	315	}
	316	conf->nfaults = 0;
	317
159ec1fc	318	list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4 LT	319	conf->rdev = rdev;
1da177e4 LT	320
1f403624	321	md_set_array_sectors(mddev, faulty_size(mddev, 0, 0));
1da177e4 LT	322	mddev->private = conf;
1da177e4 LT	323
50ac168a	324	reshape(mddev);
1da177e4 LT	325
	326	return 0;
	327	}
	328
fd01b88c	329	static int stop(struct mddev *mddev)
1da177e4	330	{
8f1ae43d	331	struct faulty_conf *conf = mddev->private;
1da177e4 LT	332
	333	kfree(conf);
	334	mddev->private = NULL;
	335	return 0;
	336	}
	337
84fc4b56	338	static struct md_personality faulty_personality =
1da177e4 LT	339	{
1da177e4 LT	340	.name = "faulty",
2604b703	341	.level = LEVEL_FAULTY,
1da177e4 LT	342	.owner = THIS_MODULE,
	343	.make_request = make_request,
	344	.run = run,
	345	.stop = stop,
	346	.status = status,
50ac168a	347	.check_reshape = reshape,
80c3a6ce	348	.size = faulty_size,
1da177e4 LT	349	};
	350
	351	static int __init raid_init(void)
	352	{
2604b703	353	return register_md_personality(&faulty_personality);
1da177e4 LT	354	}
	355
	356	static void raid_exit(void)
	357	{
2604b703	358	unregister_md_personality(&faulty_personality);
1da177e4 LT	359	}
	360
	361	module_init(raid_init);
	362	module_exit(raid_exit);
	363	MODULE_LICENSE("GPL");
0efb9e61	364	MODULE_DESCRIPTION("Fault injection personality for MD");
1da177e4	365	MODULE_ALIAS("md-personality-10"); /* faulty */
d9d166c2	366	MODULE_ALIAS("md-faulty");
2604b703	367	MODULE_ALIAS("md-level--5");