[linux.git] / drivers / md / dm-emc.c

/*
 * Copyright (C) 2004 SUSE LINUX Products GmbH. All rights reserved.
 * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
 *
 * This file is released under the GPL.
 *
 * Multipath support for EMC CLARiiON AX/CX-series hardware.
 */

#include "dm.h"
#include "dm-hw-handler.h"
#include <scsi/scsi.h>
#include <scsi/scsi_cmnd.h>

#define DM_MSG_PREFIX "multipath emc"

struct emc_handler {
	spinlock_t lock;

	/* Whether we should send the short trespass command (FC-series)
	 * or the long version (default for AX/CX CLARiiON arrays). */
	unsigned short_trespass;
	/* Whether or not to honor SCSI reservations when initiating a
	 * switch-over. Default: Don't. */
	unsigned hr;

	unsigned char sense[SCSI_SENSE_BUFFERSIZE];
};

#define TRESPASS_PAGE 0x22
#define EMC_FAILOVER_TIMEOUT (60 * HZ)

/* Code borrowed from dm-lsi-rdac by Mike Christie */

static inline void free_bio(struct bio *bio)
{
	__free_page(bio->bi_io_vec[0].bv_page);
	bio_put(bio);
}

static void emc_endio(struct bio *bio, int error)
{
	struct dm_path *path = bio->bi_private;

	/* We also need to look at the sense keys here whether or not to
	 * switch to the next PG etc.
	 *
	 * For now simple logic: either it works or it doesn't.
	 */
	if (error)
		dm_pg_init_complete(path, MP_FAIL_PATH);
	else
		dm_pg_init_complete(path, 0);

	/* request is freed in block layer */
	free_bio(bio);
}

static struct bio *get_failover_bio(struct dm_path *path, unsigned data_size)
{
	struct bio *bio;
	struct page *page;

	bio = bio_alloc(GFP_ATOMIC, 1);
	if (!bio) {
		DMERR("get_failover_bio: bio_alloc() failed.");
		return NULL;
	}

	bio->bi_rw |= (1 << BIO_RW);
	bio->bi_bdev = path->dev->bdev;
	bio->bi_sector = 0;
	bio->bi_private = path;
	bio->bi_end_io = emc_endio;

	page = alloc_page(GFP_ATOMIC);
	if (!page) {
		DMERR("get_failover_bio: alloc_page() failed.");
		bio_put(bio);
		return NULL;
	}

	if (bio_add_page(bio, page, data_size, 0) != data_size) {
		DMERR("get_failover_bio: bio_add_page() failed.");
		__free_page(page);
		bio_put(bio);
		return NULL;
	}

	return bio;
}

static struct request *get_failover_req(struct emc_handler *h,
					struct bio *bio, struct dm_path *path)
{
	struct request *rq;
	struct block_device *bdev = bio->bi_bdev;
	struct request_queue *q = bdev_get_queue(bdev);

	/* FIXME: Figure out why it fails with GFP_ATOMIC. */
	rq = blk_get_request(q, WRITE, __GFP_WAIT);
	if (!rq) {
		DMERR("get_failover_req: blk_get_request failed");
		return NULL;
	}

	blk_rq_append_bio(q, rq, bio);

	rq->sense = h->sense;
	memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
	rq->sense_len = 0;

	rq->timeout = EMC_FAILOVER_TIMEOUT;
	rq->cmd_type = REQ_TYPE_BLOCK_PC;
	rq->cmd_flags |= REQ_FAILFAST | REQ_NOMERGE;

	return rq;
}

static struct request *emc_trespass_get(struct emc_handler *h,
					struct dm_path *path)
{
	struct bio *bio;
	struct request *rq;
	unsigned char *page22;
	unsigned char long_trespass_pg[] = {
		0, 0, 0, 0,
		TRESPASS_PAGE,        /* Page code */
		0x09,                 /* Page length - 2 */
		h->hr ? 0x01 : 0x81,  /* Trespass code + Honor reservation bit */
		0xff, 0xff,           /* Trespass target */
		0, 0, 0, 0, 0, 0      /* Reserved bytes / unknown */
		};
	unsigned char short_trespass_pg[] = {
		0, 0, 0, 0,
		TRESPASS_PAGE,        /* Page code */
		0x02,                 /* Page length - 2 */
		h->hr ? 0x01 : 0x81,  /* Trespass code + Honor reservation bit */
		0xff,                 /* Trespass target */
		};
	unsigned data_size = h->short_trespass ? sizeof(short_trespass_pg) :
				sizeof(long_trespass_pg);

	/* get bio backing */
	if (data_size > PAGE_SIZE)
		/* this should never happen */
		return NULL;

	bio = get_failover_bio(path, data_size);
	if (!bio) {
		DMERR("emc_trespass_get: no bio");
		return NULL;
	}

	page22 = (unsigned char *)bio_data(bio);
	memset(page22, 0, data_size);

	memcpy(page22, h->short_trespass ?
		short_trespass_pg : long_trespass_pg, data_size);

	/* get request for block layer packet command */
	rq = get_failover_req(h, bio, path);
	if (!rq) {
		DMERR("emc_trespass_get: no rq");
		free_bio(bio);
		return NULL;
	}

	/* Prepare the command. */
	rq->cmd[0] = MODE_SELECT;
	rq->cmd[1] = 0x10;
	rq->cmd[4] = data_size;
	rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);

	return rq;
}

static void emc_pg_init(struct hw_handler *hwh, unsigned bypassed,
			struct dm_path *path)
{
	struct request *rq;
	struct request_queue *q = bdev_get_queue(path->dev->bdev);

	/*
	 * We can either blindly init the pg (then look at the sense),
	 * or we can send some commands to get the state here (then
	 * possibly send the fo cmnd), or we can also have the
	 * initial state passed into us and then get an update here.
	 */
	if (!q) {
		DMINFO("emc_pg_init: no queue");
		goto fail_path;
	}

	/* FIXME: The request should be pre-allocated. */
	rq = emc_trespass_get(hwh->context, path);
	if (!rq) {
		DMERR("emc_pg_init: no rq");
		goto fail_path;
	}

	DMINFO("emc_pg_init: sending switch-over command");
	elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 1);
	return;

fail_path:
	dm_pg_init_complete(path, MP_FAIL_PATH);
}

static struct emc_handler *alloc_emc_handler(void)
{
	struct emc_handler *h = kzalloc(sizeof(*h), GFP_KERNEL);

	if (h)
		spin_lock_init(&h->lock);

	return h;
}

static int emc_create(struct hw_handler *hwh, unsigned argc, char **argv)
{
	struct emc_handler *h;
	unsigned hr, short_trespass;

	if (argc == 0) {
		/* No arguments: use defaults */
		hr = 0;
		short_trespass = 0;
	} else if (argc != 2) {
		DMWARN("incorrect number of arguments");
		return -EINVAL;
	} else {
		if ((sscanf(argv[0], "%u", &short_trespass) != 1)
			|| (short_trespass > 1)) {
			DMWARN("invalid trespass mode selected");
			return -EINVAL;
		}

		if ((sscanf(argv[1], "%u", &hr) != 1)
			|| (hr > 1)) {
			DMWARN("invalid honor reservation flag selected");
			return -EINVAL;
		}
	}

	h = alloc_emc_handler();
	if (!h)
		return -ENOMEM;

	hwh->context = h;

	if ((h->short_trespass = short_trespass))
		DMWARN("short trespass command will be send");
	else
		DMWARN("long trespass command will be send");

	if ((h->hr = hr))
		DMWARN("honor reservation bit will be set");
	else
		DMWARN("honor reservation bit will not be set (default)");

	return 0;
}

static void emc_destroy(struct hw_handler *hwh)
{
	struct emc_handler *h = (struct emc_handler *) hwh->context;

	kfree(h);
	hwh->context = NULL;
}

static unsigned emc_error(struct hw_handler *hwh, struct bio *bio)
{
	/* FIXME: Patch from axboe still missing */
#if 0
	int sense;

	if (bio->bi_error & BIO_SENSE) {
		sense = bio->bi_error & 0xffffff; /* sense key / asc / ascq */

		if (sense == 0x020403) {
			/* LUN Not Ready - Manual Intervention Required
			 * indicates this is a passive path.
			 *
			 * FIXME: However, if this is seen and EVPD C0
			 * indicates that this is due to a NDU in
			 * progress, we should set FAIL_PATH too.
			 * This indicates we might have to do a SCSI
			 * inquiry in the end_io path. Ugh. */
			return MP_BYPASS_PG | MP_RETRY_IO;
		} else if (sense == 0x052501) {
			/* An array based copy is in progress. Do not
			 * fail the path, do not bypass to another PG,
			 * do not retry. Fail the IO immediately.
			 * (Actually this is the same conclusion as in
			 * the default handler, but lets make sure.) */
			return 0;
		} else if (sense == 0x062900) {
			/* Unit Attention Code. This is the first IO
			 * to the new path, so just retry. */
			return MP_RETRY_IO;
		}
	}
#endif

	/* Try default handler */
	return dm_scsi_err_handler(hwh, bio);
}

static struct hw_handler_type emc_hwh = {
	.name = "emc",
	.module = THIS_MODULE,
	.create = emc_create,
	.destroy = emc_destroy,
	.pg_init = emc_pg_init,
	.error = emc_error,
};

static int __init dm_emc_init(void)
{
	int r = dm_register_hw_handler(&emc_hwh);

	if (r < 0)
		DMERR("register failed %d", r);

	DMINFO("version 0.0.3 loaded");

	return r;
}

static void __exit dm_emc_exit(void)
{
	int r = dm_unregister_hw_handler(&emc_hwh);

	if (r < 0)
		DMERR("unregister failed %d", r);
}

module_init(dm_emc_init);
module_exit(dm_emc_exit);

MODULE_DESCRIPTION(DM_NAME " EMC CX/AX/FC-family multipath");
MODULE_AUTHOR("Lars Marowsky-Bree <[email protected]>");
MODULE_LICENSE("GPL");
Commit	Line	Data
1da177e4 LT	1	/*
	2	* Copyright (C) 2004 SUSE LINUX Products GmbH. All rights reserved.
	3	* Copyright (C) 2004 Red Hat, Inc. All rights reserved.
	4	*
	5	* This file is released under the GPL.
	6	*
	7	* Multipath support for EMC CLARiiON AX/CX-series hardware.
	8	*/
	9
	10	#include "dm.h"
	11	#include "dm-hw-handler.h"
	12	#include <scsi/scsi.h>
	13	#include <scsi/scsi_cmnd.h>
	14
72d94861 AK	15	#define DM_MSG_PREFIX "multipath emc"
72d94861 AK	16
1da177e4 LT	17	struct emc_handler {
	18	spinlock_t lock;
	19
	20	/* Whether we should send the short trespass command (FC-series)
	21	* or the long version (default for AX/CX CLARiiON arrays). */
	22	unsigned short_trespass;
	23	/* Whether or not to honor SCSI reservations when initiating a
	24	* switch-over. Default: Don't. */
	25	unsigned hr;
	26
	27	unsigned char sense[SCSI_SENSE_BUFFERSIZE];
	28	};
	29
	30	#define TRESPASS_PAGE 0x22
	31	#define EMC_FAILOVER_TIMEOUT (60 * HZ)
	32
	33	/* Code borrowed from dm-lsi-rdac by Mike Christie */
	34
	35	static inline void free_bio(struct bio *bio)
	36	{
	37	__free_page(bio->bi_io_vec[0].bv_page);
	38	bio_put(bio);
	39	}
	40
6712ecf8	41	static void emc_endio(struct bio *bio, int error)
1da177e4	42	{
c922d5f7	43	struct dm_path *path = bio->bi_private;
1da177e4	44
1da177e4 LT	45	/* We also need to look at the sense keys here whether or not to
	46	* switch to the next PG etc.
	47	*
	48	* For now simple logic: either it works or it doesn't.
	49	*/
	50	if (error)
	51	dm_pg_init_complete(path, MP_FAIL_PATH);
	52	else
	53	dm_pg_init_complete(path, 0);
	54
	55	/* request is freed in block layer */
	56	free_bio(bio);
1da177e4 LT	57	}
1da177e4 LT	58
c922d5f7	59	static struct bio get_failover_bio(struct dm_path path, unsigned data_size)
1da177e4 LT	60	{
	61	struct bio *bio;
	62	struct page *page;
	63
	64	bio = bio_alloc(GFP_ATOMIC, 1);
	65	if (!bio) {
72d94861	66	DMERR("get_failover_bio: bio_alloc() failed.");
1da177e4 LT	67	return NULL;
	68	}
	69
	70	bio->bi_rw \|= (1 << BIO_RW);
	71	bio->bi_bdev = path->dev->bdev;
	72	bio->bi_sector = 0;
	73	bio->bi_private = path;
	74	bio->bi_end_io = emc_endio;
	75
	76	page = alloc_page(GFP_ATOMIC);
	77	if (!page) {
72d94861	78	DMERR("get_failover_bio: alloc_page() failed.");
1da177e4 LT	79	bio_put(bio);
	80	return NULL;
	81	}
	82
	83	if (bio_add_page(bio, page, data_size, 0) != data_size) {
d336416f	84	DMERR("get_failover_bio: bio_add_page() failed.");
1da177e4 LT	85	__free_page(page);
	86	bio_put(bio);
	87	return NULL;
	88	}
	89
	90	return bio;
	91	}
	92
	93	static struct request get_failover_req(struct emc_handler h,
c922d5f7	94	struct bio bio, struct dm_path path)
1da177e4 LT	95	{
	96	struct request *rq;
	97	struct block_device *bdev = bio->bi_bdev;
	98	struct request_queue *q = bdev_get_queue(bdev);
	99
	100	/* FIXME: Figure out why it fails with GFP_ATOMIC. */
	101	rq = blk_get_request(q, WRITE, __GFP_WAIT);
	102	if (!rq) {
72d94861	103	DMERR("get_failover_req: blk_get_request failed");
1da177e4 LT	104	return NULL;
	105	}
	106
66846572	107	blk_rq_append_bio(q, rq, bio);
1da177e4 LT	108
	109	rq->sense = h->sense;
	110	memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
	111	rq->sense_len = 0;
	112
1da177e4	113	rq->timeout = EMC_FAILOVER_TIMEOUT;
4aff5e23 JA	114	rq->cmd_type = REQ_TYPE_BLOCK_PC;
4aff5e23 JA	115	rq->cmd_flags \|= REQ_FAILFAST \| REQ_NOMERGE;
1da177e4 LT	116
	117	return rq;
	118	}
	119
	120	static struct request emc_trespass_get(struct emc_handler h,
c922d5f7	121	struct dm_path *path)
1da177e4 LT	122	{
	123	struct bio *bio;
	124	struct request *rq;
	125	unsigned char *page22;
	126	unsigned char long_trespass_pg[] = {
	127	0, 0, 0, 0,
	128	TRESPASS_PAGE, /* Page code */
	129	0x09, /* Page length - 2 */
	130	h->hr ? 0x01 : 0x81, /* Trespass code + Honor reservation bit */
	131	0xff, 0xff, /* Trespass target */
	132	0, 0, 0, 0, 0, 0 /* Reserved bytes / unknown */
	133	};
	134	unsigned char short_trespass_pg[] = {
	135	0, 0, 0, 0,
	136	TRESPASS_PAGE, /* Page code */
	137	0x02, /* Page length - 2 */
	138	h->hr ? 0x01 : 0x81, /* Trespass code + Honor reservation bit */
	139	0xff, /* Trespass target */
	140	};
	141	unsigned data_size = h->short_trespass ? sizeof(short_trespass_pg) :
	142	sizeof(long_trespass_pg);
	143
	144	/* get bio backing */
	145	if (data_size > PAGE_SIZE)
	146	/* this should never happen */
	147	return NULL;
	148
	149	bio = get_failover_bio(path, data_size);
	150	if (!bio) {
72d94861	151	DMERR("emc_trespass_get: no bio");
1da177e4 LT	152	return NULL;
	153	}
	154
	155	page22 = (unsigned char *)bio_data(bio);
	156	memset(page22, 0, data_size);
	157
	158	memcpy(page22, h->short_trespass ?
	159	short_trespass_pg : long_trespass_pg, data_size);
	160
	161	/* get request for block layer packet command */
	162	rq = get_failover_req(h, bio, path);
	163	if (!rq) {
72d94861	164	DMERR("emc_trespass_get: no rq");
1da177e4 LT	165	free_bio(bio);
	166	return NULL;
	167	}
	168
	169	/* Prepare the command. */
	170	rq->cmd[0] = MODE_SELECT;
	171	rq->cmd[1] = 0x10;
	172	rq->cmd[4] = data_size;
	173	rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
	174
	175	return rq;
	176	}
	177
	178	static void emc_pg_init(struct hw_handler *hwh, unsigned bypassed,
c922d5f7	179	struct dm_path *path)
1da177e4 LT	180	{
	181	struct request *rq;
	182	struct request_queue *q = bdev_get_queue(path->dev->bdev);
	183
	184	/*
	185	* We can either blindly init the pg (then look at the sense),
	186	* or we can send some commands to get the state here (then
	187	* possibly send the fo cmnd), or we can also have the
	188	* initial state passed into us and then get an update here.
	189	*/
	190	if (!q) {
72d94861	191	DMINFO("emc_pg_init: no queue");
1da177e4 LT	192	goto fail_path;
	193	}
	194
	195	/* FIXME: The request should be pre-allocated. */
	196	rq = emc_trespass_get(hwh->context, path);
	197	if (!rq) {
72d94861	198	DMERR("emc_pg_init: no rq");
1da177e4 LT	199	goto fail_path;
	200	}
	201
72d94861	202	DMINFO("emc_pg_init: sending switch-over command");
1da177e4 LT	203	elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 1);
	204	return;
	205
	206	fail_path:
	207	dm_pg_init_complete(path, MP_FAIL_PATH);
	208	}
	209
	210	static struct emc_handler *alloc_emc_handler(void)
	211	{
094262db	212	struct emc_handler h = kzalloc(sizeof(h), GFP_KERNEL);
1da177e4	213
094262db	214	if (h)
1da177e4 LT	215	spin_lock_init(&h->lock);
	216
	217	return h;
	218	}
	219
	220	static int emc_create(struct hw_handler hwh, unsigned argc, char *argv)
	221	{
	222	struct emc_handler *h;
	223	unsigned hr, short_trespass;
	224
	225	if (argc == 0) {
	226	/* No arguments: use defaults */
	227	hr = 0;
	228	short_trespass = 0;
	229	} else if (argc != 2) {
72d94861	230	DMWARN("incorrect number of arguments");
1da177e4 LT	231	return -EINVAL;
	232	} else {
	233	if ((sscanf(argv[0], "%u", &short_trespass) != 1)
	234	\|\| (short_trespass > 1)) {
72d94861	235	DMWARN("invalid trespass mode selected");
1da177e4 LT	236	return -EINVAL;
	237	}
	238
	239	if ((sscanf(argv[1], "%u", &hr) != 1)
	240	\|\| (hr > 1)) {
72d94861	241	DMWARN("invalid honor reservation flag selected");
1da177e4 LT	242	return -EINVAL;
	243	}
	244	}
	245
	246	h = alloc_emc_handler();
	247	if (!h)
	248	return -ENOMEM;
	249
1da177e4 LT	250	hwh->context = h;
	251
	252	if ((h->short_trespass = short_trespass))
72d94861	253	DMWARN("short trespass command will be send");
1da177e4	254	else
72d94861	255	DMWARN("long trespass command will be send");
1da177e4 LT	256
1da177e4 LT	257	if ((h->hr = hr))
72d94861	258	DMWARN("honor reservation bit will be set");
1da177e4	259	else
72d94861	260	DMWARN("honor reservation bit will not be set (default)");
1da177e4 LT	261
	262	return 0;
	263	}
	264
	265	static void emc_destroy(struct hw_handler *hwh)
	266	{
	267	struct emc_handler h = (struct emc_handler ) hwh->context;
	268
	269	kfree(h);
	270	hwh->context = NULL;
	271	}
	272
	273	static unsigned emc_error(struct hw_handler hwh, struct bio bio)
	274	{
	275	/* FIXME: Patch from axboe still missing */
	276	#if 0
	277	int sense;
	278
	279	if (bio->bi_error & BIO_SENSE) {
	280	sense = bio->bi_error & 0xffffff; /* sense key / asc / ascq */
	281
	282	if (sense == 0x020403) {
	283	/* LUN Not Ready - Manual Intervention Required
	284	* indicates this is a passive path.
	285	*
	286	* FIXME: However, if this is seen and EVPD C0
	287	* indicates that this is due to a NDU in
	288	* progress, we should set FAIL_PATH too.
	289	* This indicates we might have to do a SCSI
	290	* inquiry in the end_io path. Ugh. */
	291	return MP_BYPASS_PG \| MP_RETRY_IO;
	292	} else if (sense == 0x052501) {
	293	/* An array based copy is in progress. Do not
	294	* fail the path, do not bypass to another PG,
	295	* do not retry. Fail the IO immediately.
	296	* (Actually this is the same conclusion as in
	297	* the default handler, but lets make sure.) */
	298	return 0;
	299	} else if (sense == 0x062900) {
	300	/* Unit Attention Code. This is the first IO
	301	* to the new path, so just retry. */
	302	return MP_RETRY_IO;
	303	}
	304	}
	305	#endif
	306
	307	/* Try default handler */
	308	return dm_scsi_err_handler(hwh, bio);
	309	}
	310
	311	static struct hw_handler_type emc_hwh = {
	312	.name = "emc",
	313	.module = THIS_MODULE,
	314	.create = emc_create,
	315	.destroy = emc_destroy,
	316	.pg_init = emc_pg_init,
	317	.error = emc_error,
	318	};
	319
	320	static int __init dm_emc_init(void)
	321	{
	322	int r = dm_register_hw_handler(&emc_hwh);
	323
	324	if (r < 0)
72d94861	325	DMERR("register failed %d", r);
1da177e4	326
72d94861	327	DMINFO("version 0.0.3 loaded");
1da177e4 LT	328
	329	return r;
	330	}
	331
	332	static void __exit dm_emc_exit(void)
	333	{
	334	int r = dm_unregister_hw_handler(&emc_hwh);
	335
	336	if (r < 0)
72d94861	337	DMERR("unregister failed %d", r);
1da177e4 LT	338	}
	339
	340	module_init(dm_emc_init);
	341	module_exit(dm_emc_exit);
	342
	343	MODULE_DESCRIPTION(DM_NAME " EMC CX/AX/FC-family multipath");
	344	MODULE_AUTHOR("Lars Marowsky-Bree <[email protected]>");
	345	MODULE_LICENSE("GPL");