[linux.git] / drivers / md / multipath.c

/*
 * multipath.c : Multiple Devices driver for Linux
 *
 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
 *
 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
 *
 * MULTIPATH management functions.
 *
 * derived from raid1.c.
 *
 * 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.
 */

#include <linux/blkdev.h>
#include <linux/module.h>
#include <linux/raid/md_u.h>
#include <linux/seq_file.h>
#include <linux/slab.h>
#include "md.h"
#include "multipath.h"

#define MAX_WORK_PER_DISK 128

#define	NR_RESERVED_BUFS	32

static int multipath_map (struct mpconf *conf)
{
	int i, disks = conf->raid_disks;

	/*
	 * Later we do read balancing on the read side
	 * now we use the first available disk.
	 */

	rcu_read_lock();
	for (i = 0; i < disks; i++) {
		struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
		if (rdev && test_bit(In_sync, &rdev->flags)) {
			atomic_inc(&rdev->nr_pending);
			rcu_read_unlock();
			return i;
		}
	}
	rcu_read_unlock();

	printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
	return (-1);
}

static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
{
	unsigned long flags;
	struct mddev *mddev = mp_bh->mddev;
	struct mpconf *conf = mddev->private;

	spin_lock_irqsave(&conf->device_lock, flags);
	list_add(&mp_bh->retry_list, &conf->retry_list);
	spin_unlock_irqrestore(&conf->device_lock, flags);
	md_wakeup_thread(mddev->thread);
}

/*
 * multipath_end_bh_io() is called when we have finished servicing a multipathed
 * operation and are ready to return a success/failure code to the buffer
 * cache layer.
 */
static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
{
	struct bio *bio = mp_bh->master_bio;
	struct mpconf *conf = mp_bh->mddev->private;

	bio_endio(bio, err);
	mempool_free(mp_bh, conf->pool);
}

static void multipath_end_request(struct bio *bio, int error)
{
	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	struct multipath_bh *mp_bh = bio->bi_private;
	struct mpconf *conf = mp_bh->mddev->private;
	struct md_rdev *rdev = conf->multipaths[mp_bh->path].rdev;

	if (uptodate)
		multipath_end_bh_io(mp_bh, 0);
	else if (!(bio->bi_rw & REQ_RAHEAD)) {
		/*
		 * oops, IO error:
		 */
		char b[BDEVNAME_SIZE];
		md_error (mp_bh->mddev, rdev);
		printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
		       bdevname(rdev->bdev,b),
		       (unsigned long long)bio->bi_iter.bi_sector);
		multipath_reschedule_retry(mp_bh);
	} else
		multipath_end_bh_io(mp_bh, error);
	rdev_dec_pending(rdev, conf->mddev);
}

static void multipath_make_request(struct mddev *mddev, struct bio * bio)
{
	struct mpconf *conf = mddev->private;
	struct multipath_bh * mp_bh;
	struct multipath_info *multipath;

	if (unlikely(bio->bi_rw & REQ_FLUSH)) {
		md_flush_request(mddev, bio);
		return;
	}

	mp_bh = mempool_alloc(conf->pool, GFP_NOIO);

	mp_bh->master_bio = bio;
	mp_bh->mddev = mddev;

	mp_bh->path = multipath_map(conf);
	if (mp_bh->path < 0) {
		bio_endio(bio, -EIO);
		mempool_free(mp_bh, conf->pool);
		return;
	}
	multipath = conf->multipaths + mp_bh->path;

	mp_bh->bio = *bio;
	mp_bh->bio.bi_iter.bi_sector += multipath->rdev->data_offset;
	mp_bh->bio.bi_bdev = multipath->rdev->bdev;
	mp_bh->bio.bi_rw |= REQ_FAILFAST_TRANSPORT;
	mp_bh->bio.bi_end_io = multipath_end_request;
	mp_bh->bio.bi_private = mp_bh;
	generic_make_request(&mp_bh->bio);
	return;
}

static void multipath_status (struct seq_file *seq, struct mddev *mddev)
{
	struct mpconf *conf = mddev->private;
	int i;

	seq_printf (seq, " [%d/%d] [", conf->raid_disks,
		    conf->raid_disks - mddev->degraded);
	for (i = 0; i < conf->raid_disks; i++)
		seq_printf (seq, "%s",
			       conf->multipaths[i].rdev &&
			       test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
	seq_printf (seq, "]");
}

static int multipath_congested(struct mddev *mddev, int bits)
{
	struct mpconf *conf = mddev->private;
	int i, ret = 0;

	rcu_read_lock();
	for (i = 0; i < mddev->raid_disks ; i++) {
		struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
		if (rdev && !test_bit(Faulty, &rdev->flags)) {
			struct request_queue *q = bdev_get_queue(rdev->bdev);

			ret |= bdi_congested(&q->backing_dev_info, bits);
			/* Just like multipath_map, we just check the
			 * first available device
			 */
			break;
		}
	}
	rcu_read_unlock();
	return ret;
}

/*
 * Careful, this can execute in IRQ contexts as well!
 */
static void multipath_error (struct mddev *mddev, struct md_rdev *rdev)
{
	struct mpconf *conf = mddev->private;
	char b[BDEVNAME_SIZE];

	if (conf->raid_disks - mddev->degraded <= 1) {
		/*
		 * Uh oh, we can do nothing if this is our last path, but
		 * first check if this is a queued request for a device
		 * which has just failed.
		 */
		printk(KERN_ALERT
		       "multipath: only one IO path left and IO error.\n");
		/* leave it active... it's all we have */
		return;
	}
	/*
	 * Mark disk as unusable
	 */
	if (test_and_clear_bit(In_sync, &rdev->flags)) {
		unsigned long flags;
		spin_lock_irqsave(&conf->device_lock, flags);
		mddev->degraded++;
		spin_unlock_irqrestore(&conf->device_lock, flags);
	}
	set_bit(Faulty, &rdev->flags);
	set_bit(MD_CHANGE_DEVS, &mddev->flags);
	printk(KERN_ALERT "multipath: IO failure on %s,"
	       " disabling IO path.\n"
	       "multipath: Operation continuing"
	       " on %d IO paths.\n",
	       bdevname(rdev->bdev, b),
	       conf->raid_disks - mddev->degraded);
}

static void print_multipath_conf (struct mpconf *conf)
{
	int i;
	struct multipath_info *tmp;

	printk("MULTIPATH conf printout:\n");
	if (!conf) {
		printk("(conf==NULL)\n");
		return;
	}
	printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
			 conf->raid_disks);

	for (i = 0; i < conf->raid_disks; i++) {
		char b[BDEVNAME_SIZE];
		tmp = conf->multipaths + i;
		if (tmp->rdev)
			printk(" disk%d, o:%d, dev:%s\n",
				i,!test_bit(Faulty, &tmp->rdev->flags),
			       bdevname(tmp->rdev->bdev,b));
	}
}

static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
{
	struct mpconf *conf = mddev->private;
	struct request_queue *q;
	int err = -EEXIST;
	int path;
	struct multipath_info *p;
	int first = 0;
	int last = mddev->raid_disks - 1;

	if (rdev->raid_disk >= 0)
		first = last = rdev->raid_disk;

	print_multipath_conf(conf);

	for (path = first; path <= last; path++)
		if ((p=conf->multipaths+path)->rdev == NULL) {
			q = rdev->bdev->bd_disk->queue;
			disk_stack_limits(mddev->gendisk, rdev->bdev,
					  rdev->data_offset << 9);

		/* as we don't honour merge_bvec_fn, we must never risk
		 * violating it, so limit ->max_segments to one, lying
		 * within a single page.
		 * (Note: it is very unlikely that a device with
		 * merge_bvec_fn will be involved in multipath.)
		 */
			if (q->merge_bvec_fn) {
				blk_queue_max_segments(mddev->queue, 1);
				blk_queue_segment_boundary(mddev->queue,
							   PAGE_CACHE_SIZE - 1);
			}

			spin_lock_irq(&conf->device_lock);
			mddev->degraded--;
			rdev->raid_disk = path;
			set_bit(In_sync, &rdev->flags);
			spin_unlock_irq(&conf->device_lock);
			rcu_assign_pointer(p->rdev, rdev);
			err = 0;
			md_integrity_add_rdev(rdev, mddev);
			break;
		}

	print_multipath_conf(conf);

	return err;
}

static int multipath_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
{
	struct mpconf *conf = mddev->private;
	int err = 0;
	int number = rdev->raid_disk;
	struct multipath_info *p = conf->multipaths + number;

	print_multipath_conf(conf);

	if (rdev == p->rdev) {
		if (test_bit(In_sync, &rdev->flags) ||
		    atomic_read(&rdev->nr_pending)) {
			printk(KERN_ERR "hot-remove-disk, slot %d is identified"
			       " but is still operational!\n", number);
			err = -EBUSY;
			goto abort;
		}
		p->rdev = NULL;
		synchronize_rcu();
		if (atomic_read(&rdev->nr_pending)) {
			/* lost the race, try later */
			err = -EBUSY;
			p->rdev = rdev;
			goto abort;
		}
		err = md_integrity_register(mddev);
	}
abort:

	print_multipath_conf(conf);
	return err;
}

/*
 * This is a kernel thread which:
 *
 *	1.	Retries failed read operations on working multipaths.
 *	2.	Updates the raid superblock when problems encounter.
 *	3.	Performs writes following reads for array syncronising.
 */

static void multipathd(struct md_thread *thread)
{
	struct mddev *mddev = thread->mddev;
	struct multipath_bh *mp_bh;
	struct bio *bio;
	unsigned long flags;
	struct mpconf *conf = mddev->private;
	struct list_head *head = &conf->retry_list;

	md_check_recovery(mddev);
	for (;;) {
		char b[BDEVNAME_SIZE];
		spin_lock_irqsave(&conf->device_lock, flags);
		if (list_empty(head))
			break;
		mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
		list_del(head->prev);
		spin_unlock_irqrestore(&conf->device_lock, flags);

		bio = &mp_bh->bio;
		bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector;

		if ((mp_bh->path = multipath_map (conf))<0) {
			printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
				" error for block %llu\n",
				bdevname(bio->bi_bdev,b),
				(unsigned long long)bio->bi_iter.bi_sector);
			multipath_end_bh_io(mp_bh, -EIO);
		} else {
			printk(KERN_ERR "multipath: %s: redirecting sector %llu"
				" to another IO path\n",
				bdevname(bio->bi_bdev,b),
				(unsigned long long)bio->bi_iter.bi_sector);
			*bio = *(mp_bh->master_bio);
			bio->bi_iter.bi_sector +=
				conf->multipaths[mp_bh->path].rdev->data_offset;
			bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
			bio->bi_rw |= REQ_FAILFAST_TRANSPORT;
			bio->bi_end_io = multipath_end_request;
			bio->bi_private = mp_bh;
			generic_make_request(bio);
		}
	}
	spin_unlock_irqrestore(&conf->device_lock, flags);
}

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

	return mddev->dev_sectors;
}

static int multipath_run (struct mddev *mddev)
{
	struct mpconf *conf;
	int disk_idx;
	struct multipath_info *disk;
	struct md_rdev *rdev;
	int working_disks;

	if (md_check_no_bitmap(mddev))
		return -EINVAL;

	if (mddev->level != LEVEL_MULTIPATH) {
		printk("multipath: %s: raid level not set to multipath IO (%d)\n",
		       mdname(mddev), mddev->level);
		goto out;
	}
	/*
	 * copy the already verified devices into our private MULTIPATH
	 * bookkeeping area. [whatever we allocate in multipath_run(),
	 * should be freed in multipath_free()]
	 */

	conf = kzalloc(sizeof(struct mpconf), GFP_KERNEL);
	mddev->private = conf;
	if (!conf) {
		printk(KERN_ERR
			"multipath: couldn't allocate memory for %s\n",
			mdname(mddev));
		goto out;
	}

	conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
				   GFP_KERNEL);
	if (!conf->multipaths) {
		printk(KERN_ERR
			"multipath: couldn't allocate memory for %s\n",
			mdname(mddev));
		goto out_free_conf;
	}

	working_disks = 0;
	rdev_for_each(rdev, mddev) {
		disk_idx = rdev->raid_disk;
		if (disk_idx < 0 ||
		    disk_idx >= mddev->raid_disks)
			continue;

		disk = conf->multipaths + disk_idx;
		disk->rdev = rdev;
		disk_stack_limits(mddev->gendisk, rdev->bdev,
				  rdev->data_offset << 9);

		/* as we don't honour merge_bvec_fn, we must never risk
		 * violating it, not that we ever expect a device with
		 * a merge_bvec_fn to be involved in multipath */
		if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
			blk_queue_max_segments(mddev->queue, 1);
			blk_queue_segment_boundary(mddev->queue,
						   PAGE_CACHE_SIZE - 1);
		}

		if (!test_bit(Faulty, &rdev->flags))
			working_disks++;
	}

	conf->raid_disks = mddev->raid_disks;
	conf->mddev = mddev;
	spin_lock_init(&conf->device_lock);
	INIT_LIST_HEAD(&conf->retry_list);

	if (!working_disks) {
		printk(KERN_ERR "multipath: no operational IO paths for %s\n",
			mdname(mddev));
		goto out_free_conf;
	}
	mddev->degraded = conf->raid_disks - working_disks;

	conf->pool = mempool_create_kmalloc_pool(NR_RESERVED_BUFS,
						 sizeof(struct multipath_bh));
	if (conf->pool == NULL) {
		printk(KERN_ERR
			"multipath: couldn't allocate memory for %s\n",
			mdname(mddev));
		goto out_free_conf;
	}

	{
		mddev->thread = md_register_thread(multipathd, mddev,
						   "multipath");
		if (!mddev->thread) {
			printk(KERN_ERR "multipath: couldn't allocate thread"
				" for %s\n", mdname(mddev));
			goto out_free_conf;
		}
	}

	printk(KERN_INFO
		"multipath: array %s active with %d out of %d IO paths\n",
		mdname(mddev), conf->raid_disks - mddev->degraded,
	       mddev->raid_disks);
	/*
	 * Ok, everything is just fine now
	 */
	md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));

	if (md_integrity_register(mddev))
		goto out_free_conf;

	return 0;

out_free_conf:
	if (conf->pool)
		mempool_destroy(conf->pool);
	kfree(conf->multipaths);
	kfree(conf);
	mddev->private = NULL;
out:
	return -EIO;
}

static void multipath_free(struct mddev *mddev, void *priv)
{
	struct mpconf *conf = priv;

	mempool_destroy(conf->pool);
	kfree(conf->multipaths);
	kfree(conf);
}

static struct md_personality multipath_personality =
{
	.name		= "multipath",
	.level		= LEVEL_MULTIPATH,
	.owner		= THIS_MODULE,
	.make_request	= multipath_make_request,
	.run		= multipath_run,
	.free		= multipath_free,
	.status		= multipath_status,
	.error_handler	= multipath_error,
	.hot_add_disk	= multipath_add_disk,
	.hot_remove_disk= multipath_remove_disk,
	.size		= multipath_size,
	.congested	= multipath_congested,
};

static int __init multipath_init (void)
{
	return register_md_personality (&multipath_personality);
}

static void __exit multipath_exit (void)
{
	unregister_md_personality (&multipath_personality);
}

module_init(multipath_init);
module_exit(multipath_exit);
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("simple multi-path personality for MD");
MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
MODULE_ALIAS("md-multipath");
MODULE_ALIAS("md-level--4");
Commit	Line	Data
1da177e4 LT	1	/*
	2	* multipath.c : Multiple Devices driver for Linux
	3	*
	4	* Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
	5	*
	6	* Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
	7	*
	8	* MULTIPATH management functions.
	9	*
	10	* derived from raid1.c.
	11	*
	12	* This program is free software; you can redistribute it and/or modify
	13	* it under the terms of the GNU General Public License as published by
	14	* the Free Software Foundation; either version 2, or (at your option)
	15	* any later version.
	16	*
	17	* You should have received a copy of the GNU General Public License
	18	* (for example /usr/src/linux/COPYING); if not, write to the Free
	19	* Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
	20	*/
	21
bff61975	22	#include <linux/blkdev.h>
056075c7	23	#include <linux/module.h>
bff61975	24	#include <linux/raid/md_u.h>
bff61975	25	#include <linux/seq_file.h>
5a0e3ad6	26	#include <linux/slab.h>
43b2e5d8	27	#include "md.h"
ef740c37	28	#include "multipath.h"
1da177e4 LT	29
	30	#define MAX_WORK_PER_DISK 128
	31
	32	#define NR_RESERVED_BUFS 32
	33
69724e28	34	static int multipath_map (struct mpconf *conf)
1da177e4 LT	35	{
	36	int i, disks = conf->raid_disks;
	37
	38	/*
f72ffdd6	39	* Later we do read balancing on the read side
1da177e4 LT	40	* now we use the first available disk.
	41	*/
	42
	43	rcu_read_lock();
	44	for (i = 0; i < disks; i++) {
3cb03002	45	struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
b2d444d7	46	if (rdev && test_bit(In_sync, &rdev->flags)) {
1da177e4 LT	47	atomic_inc(&rdev->nr_pending);
	48	rcu_read_unlock();
	49	return i;
	50	}
	51	}
	52	rcu_read_unlock();
	53
	54	printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
	55	return (-1);
	56	}
	57
	58	static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
	59	{
	60	unsigned long flags;
fd01b88c	61	struct mddev *mddev = mp_bh->mddev;
69724e28	62	struct mpconf *conf = mddev->private;
1da177e4 LT	63
	64	spin_lock_irqsave(&conf->device_lock, flags);
	65	list_add(&mp_bh->retry_list, &conf->retry_list);
	66	spin_unlock_irqrestore(&conf->device_lock, flags);
	67	md_wakeup_thread(mddev->thread);
	68	}
	69
1da177e4 LT	70	/*
	71	* multipath_end_bh_io() is called when we have finished servicing a multipathed
	72	* operation and are ready to return a success/failure code to the buffer
	73	* cache layer.
	74	*/
	75	static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
	76	{
	77	struct bio *bio = mp_bh->master_bio;
69724e28	78	struct mpconf *conf = mp_bh->mddev->private;
1da177e4	79
6712ecf8	80	bio_endio(bio, err);
1da177e4 LT	81	mempool_free(mp_bh, conf->pool);
	82	}
	83
6712ecf8	84	static void multipath_end_request(struct bio *bio, int error)
1da177e4 LT	85	{
1da177e4 LT	86	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
7b92813c	87	struct multipath_bh *mp_bh = bio->bi_private;
69724e28	88	struct mpconf *conf = mp_bh->mddev->private;
3cb03002	89	struct md_rdev *rdev = conf->multipaths[mp_bh->path].rdev;
1da177e4	90
1da177e4 LT	91	if (uptodate)
1da177e4 LT	92	multipath_end_bh_io(mp_bh, 0);
7b6d91da	93	else if (!(bio->bi_rw & REQ_RAHEAD)) {
1da177e4 LT	94	/*
	95	* oops, IO error:
	96	*/
	97	char b[BDEVNAME_SIZE];
	98	md_error (mp_bh->mddev, rdev);
f72ffdd6 N	99	printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
f72ffdd6 N	100	bdevname(rdev->bdev,b),
4f024f37	101	(unsigned long long)bio->bi_iter.bi_sector);
1da177e4 LT	102	multipath_reschedule_retry(mp_bh);
	103	} else
	104	multipath_end_bh_io(mp_bh, error);
	105	rdev_dec_pending(rdev, conf->mddev);
1da177e4 LT	106	}
1da177e4 LT	107
b4fdcb02	108	static void multipath_make_request(struct mddev mddev, struct bio bio)
1da177e4	109	{
69724e28	110	struct mpconf *conf = mddev->private;
1da177e4 LT	111	struct multipath_bh * mp_bh;
	112	struct multipath_info *multipath;
	113
e9c7469b TH	114	if (unlikely(bio->bi_rw & REQ_FLUSH)) {
e9c7469b TH	115	md_flush_request(mddev, bio);
5a7bbad2	116	return;
e5dcdd80 N	117	}
e5dcdd80 N	118
1da177e4 LT	119	mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
	120
	121	mp_bh->master_bio = bio;
	122	mp_bh->mddev = mddev;
	123
1da177e4 LT	124	mp_bh->path = multipath_map(conf);
1da177e4 LT	125	if (mp_bh->path < 0) {
6712ecf8	126	bio_endio(bio, -EIO);
1da177e4	127	mempool_free(mp_bh, conf->pool);
5a7bbad2	128	return;
1da177e4 LT	129	}
	130	multipath = conf->multipaths + mp_bh->path;
	131
	132	mp_bh->bio = *bio;
4f024f37	133	mp_bh->bio.bi_iter.bi_sector += multipath->rdev->data_offset;
1da177e4	134	mp_bh->bio.bi_bdev = multipath->rdev->bdev;
7b6d91da	135	mp_bh->bio.bi_rw \|= REQ_FAILFAST_TRANSPORT;
1da177e4 LT	136	mp_bh->bio.bi_end_io = multipath_end_request;
	137	mp_bh->bio.bi_private = mp_bh;
	138	generic_make_request(&mp_bh->bio);
5a7bbad2	139	return;
1da177e4 LT	140	}
1da177e4 LT	141
fd01b88c	142	static void multipath_status (struct seq_file seq, struct mddev mddev)
1da177e4	143	{
69724e28	144	struct mpconf *conf = mddev->private;
1da177e4	145	int i;
f72ffdd6	146
1da177e4	147	seq_printf (seq, " [%d/%d] [", conf->raid_disks,
92f861a7	148	conf->raid_disks - mddev->degraded);
1da177e4 LT	149	for (i = 0; i < conf->raid_disks; i++)
1da177e4 LT	150	seq_printf (seq, "%s",
f72ffdd6	151	conf->multipaths[i].rdev &&
b2d444d7	152	test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
1da177e4 LT	153	seq_printf (seq, "]");
	154	}
	155
5c675f83	156	static int multipath_congested(struct mddev *mddev, int bits)
0d129228	157	{
69724e28	158	struct mpconf *conf = mddev->private;
0d129228 N	159	int i, ret = 0;
	160
	161	rcu_read_lock();
	162	for (i = 0; i < mddev->raid_disks ; i++) {
3cb03002	163	struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
0d129228	164	if (rdev && !test_bit(Faulty, &rdev->flags)) {
165125e1	165	struct request_queue *q = bdev_get_queue(rdev->bdev);
0d129228 N	166
	167	ret \|= bdi_congested(&q->backing_dev_info, bits);
	168	/* Just like multipath_map, we just check the
	169	* first available device
	170	*/
	171	break;
	172	}
	173	}
	174	rcu_read_unlock();
	175	return ret;
	176	}
1da177e4 LT	177
	178	/*
	179	* Careful, this can execute in IRQ contexts as well!
	180	*/
fd01b88c	181	static void multipath_error (struct mddev mddev, struct md_rdev rdev)
1da177e4	182	{
69724e28	183	struct mpconf *conf = mddev->private;
6f8d0c77	184	char b[BDEVNAME_SIZE];
1da177e4	185
92f861a7	186	if (conf->raid_disks - mddev->degraded <= 1) {
1da177e4 LT	187	/*
	188	* Uh oh, we can do nothing if this is our last path, but
	189	* first check if this is a queued request for a device
	190	* which has just failed.
	191	*/
f72ffdd6	192	printk(KERN_ALERT
6f8d0c77	193	"multipath: only one IO path left and IO error.\n");
1da177e4	194	/* leave it active... it's all we have */
6f8d0c77 N	195	return;
	196	}
	197	/*
	198	* Mark disk as unusable
	199	*/
	200	if (test_and_clear_bit(In_sync, &rdev->flags)) {
	201	unsigned long flags;
	202	spin_lock_irqsave(&conf->device_lock, flags);
	203	mddev->degraded++;
	204	spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4	205	}
6f8d0c77 N	206	set_bit(Faulty, &rdev->flags);
	207	set_bit(MD_CHANGE_DEVS, &mddev->flags);
	208	printk(KERN_ALERT "multipath: IO failure on %s,"
	209	" disabling IO path.\n"
	210	"multipath: Operation continuing"
	211	" on %d IO paths.\n",
	212	bdevname(rdev->bdev, b),
	213	conf->raid_disks - mddev->degraded);
1da177e4 LT	214	}
1da177e4 LT	215
69724e28	216	static void print_multipath_conf (struct mpconf *conf)
1da177e4 LT	217	{
	218	int i;
	219	struct multipath_info *tmp;
	220
	221	printk("MULTIPATH conf printout:\n");
	222	if (!conf) {
	223	printk("(conf==NULL)\n");
	224	return;
	225	}
92f861a7	226	printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1da177e4 LT	227	conf->raid_disks);
	228
	229	for (i = 0; i < conf->raid_disks; i++) {
	230	char b[BDEVNAME_SIZE];
	231	tmp = conf->multipaths + i;
	232	if (tmp->rdev)
	233	printk(" disk%d, o:%d, dev:%s\n",
b2d444d7	234	i,!test_bit(Faulty, &tmp->rdev->flags),
1da177e4 LT	235	bdevname(tmp->rdev->bdev,b));
	236	}
	237	}
	238
fd01b88c	239	static int multipath_add_disk(struct mddev mddev, struct md_rdev rdev)
1da177e4	240	{
69724e28	241	struct mpconf *conf = mddev->private;
352d768b	242	struct request_queue *q;
199050ea	243	int err = -EEXIST;
1da177e4 LT	244	int path;
1da177e4 LT	245	struct multipath_info *p;
6c2fce2e NB	246	int first = 0;
	247	int last = mddev->raid_disks - 1;
	248
	249	if (rdev->raid_disk >= 0)
	250	first = last = rdev->raid_disk;
1da177e4 LT	251
	252	print_multipath_conf(conf);
	253
6c2fce2e	254	for (path = first; path <= last; path++)
1da177e4	255	if ((p=conf->multipaths+path)->rdev == NULL) {
352d768b	256	q = rdev->bdev->bd_disk->queue;
8f6c2e4b MP	257	disk_stack_limits(mddev->gendisk, rdev->bdev,
8f6c2e4b MP	258	rdev->data_offset << 9);
1da177e4 LT	259
1da177e4 LT	260	/* as we don't honour merge_bvec_fn, we must never risk
627a2d3c N	261	* violating it, so limit ->max_segments to one, lying
627a2d3c N	262	* within a single page.
1da177e4 LT	263	* (Note: it is very unlikely that a device with
	264	* merge_bvec_fn will be involved in multipath.)
	265	*/
627a2d3c N	266	if (q->merge_bvec_fn) {
	267	blk_queue_max_segments(mddev->queue, 1);
	268	blk_queue_segment_boundary(mddev->queue,
	269	PAGE_CACHE_SIZE - 1);
	270	}
1da177e4	271
6f8d0c77	272	spin_lock_irq(&conf->device_lock);
750a8f3e	273	mddev->degraded--;
1da177e4	274	rdev->raid_disk = path;
b2d444d7	275	set_bit(In_sync, &rdev->flags);
6f8d0c77	276	spin_unlock_irq(&conf->device_lock);
d6065f7b	277	rcu_assign_pointer(p->rdev, rdev);
199050ea	278	err = 0;
ac5e7113	279	md_integrity_add_rdev(rdev, mddev);
199050ea	280	break;
1da177e4 LT	281	}
	282
	283	print_multipath_conf(conf);
199050ea NB	284
199050ea NB	285	return err;
1da177e4 LT	286	}
1da177e4 LT	287
b8321b68	288	static int multipath_remove_disk(struct mddev mddev, struct md_rdev rdev)
1da177e4	289	{
69724e28	290	struct mpconf *conf = mddev->private;
1da177e4	291	int err = 0;
b8321b68	292	int number = rdev->raid_disk;
1da177e4 LT	293	struct multipath_info *p = conf->multipaths + number;
	294
	295	print_multipath_conf(conf);
	296
b8321b68	297	if (rdev == p->rdev) {
b2d444d7	298	if (test_bit(In_sync, &rdev->flags) \|\|
1da177e4	299	atomic_read(&rdev->nr_pending)) {
dfc70645 N	300	printk(KERN_ERR "hot-remove-disk, slot %d is identified"
dfc70645 N	301	" but is still operational!\n", number);
1da177e4 LT	302	err = -EBUSY;
	303	goto abort;
	304	}
	305	p->rdev = NULL;
fbd568a3	306	synchronize_rcu();
1da177e4 LT	307	if (atomic_read(&rdev->nr_pending)) {
	308	/* lost the race, try later */
	309	err = -EBUSY;
	310	p->rdev = rdev;
ac5e7113	311	goto abort;
1da177e4	312	}
a91a2785	313	err = md_integrity_register(mddev);
1da177e4 LT	314	}
	315	abort:
	316
	317	print_multipath_conf(conf);
	318	return err;
	319	}
	320
1da177e4 LT	321	/*
	322	* This is a kernel thread which:
	323	*
	324	* 1. Retries failed read operations on working multipaths.
	325	* 2. Updates the raid superblock when problems encounter.
	326	* 3. Performs writes following reads for array syncronising.
	327	*/
	328
4ed8731d	329	static void multipathd(struct md_thread *thread)
1da177e4	330	{
4ed8731d	331	struct mddev *mddev = thread->mddev;
1da177e4 LT	332	struct multipath_bh *mp_bh;
	333	struct bio *bio;
	334	unsigned long flags;
69724e28	335	struct mpconf *conf = mddev->private;
1da177e4 LT	336	struct list_head *head = &conf->retry_list;
	337
	338	md_check_recovery(mddev);
	339	for (;;) {
	340	char b[BDEVNAME_SIZE];
	341	spin_lock_irqsave(&conf->device_lock, flags);
	342	if (list_empty(head))
	343	break;
	344	mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
	345	list_del(head->prev);
	346	spin_unlock_irqrestore(&conf->device_lock, flags);
	347
	348	bio = &mp_bh->bio;
4f024f37	349	bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector;
f72ffdd6	350
1da177e4 LT	351	if ((mp_bh->path = multipath_map (conf))<0) {
	352	printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
	353	" error for block %llu\n",
	354	bdevname(bio->bi_bdev,b),
4f024f37	355	(unsigned long long)bio->bi_iter.bi_sector);
1da177e4 LT	356	multipath_end_bh_io(mp_bh, -EIO);
	357	} else {
	358	printk(KERN_ERR "multipath: %s: redirecting sector %llu"
	359	" to another IO path\n",
	360	bdevname(bio->bi_bdev,b),
4f024f37	361	(unsigned long long)bio->bi_iter.bi_sector);
1da177e4	362	bio = (mp_bh->master_bio);
4f024f37 KO	363	bio->bi_iter.bi_sector +=
4f024f37 KO	364	conf->multipaths[mp_bh->path].rdev->data_offset;
1da177e4	365	bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
7b6d91da	366	bio->bi_rw \|= REQ_FAILFAST_TRANSPORT;
1da177e4 LT	367	bio->bi_end_io = multipath_end_request;
	368	bio->bi_private = mp_bh;
	369	generic_make_request(bio);
	370	}
	371	}
	372	spin_unlock_irqrestore(&conf->device_lock, flags);
	373	}
	374
fd01b88c	375	static sector_t multipath_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce DW	376	{
	377	WARN_ONCE(sectors \|\| raid_disks,
	378	"%s does not support generic reshape\n", __func__);
	379
	380	return mddev->dev_sectors;
	381	}
	382
fd01b88c	383	static int multipath_run (struct mddev *mddev)
1da177e4	384	{
69724e28	385	struct mpconf *conf;
1da177e4 LT	386	int disk_idx;
1da177e4 LT	387	struct multipath_info *disk;
3cb03002	388	struct md_rdev *rdev;
92f861a7	389	int working_disks;
1da177e4	390
0894cc30 AN	391	if (md_check_no_bitmap(mddev))
	392	return -EINVAL;
	393
1da177e4 LT	394	if (mddev->level != LEVEL_MULTIPATH) {
	395	printk("multipath: %s: raid level not set to multipath IO (%d)\n",
	396	mdname(mddev), mddev->level);
	397	goto out;
	398	}
	399	/*
	400	* copy the already verified devices into our private MULTIPATH
	401	* bookkeeping area. [whatever we allocate in multipath_run(),
afa0f557	402	* should be freed in multipath_free()]
1da177e4 LT	403	*/
1da177e4 LT	404
69724e28	405	conf = kzalloc(sizeof(struct mpconf), GFP_KERNEL);
1da177e4 LT	406	mddev->private = conf;
1da177e4 LT	407	if (!conf) {
f72ffdd6	408	printk(KERN_ERR
1da177e4 LT	409	"multipath: couldn't allocate memory for %s\n",
	410	mdname(mddev));
	411	goto out;
	412	}
1da177e4	413
9ffae0cf	414	conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
1da177e4 LT	415	GFP_KERNEL);
1da177e4 LT	416	if (!conf->multipaths) {
f72ffdd6	417	printk(KERN_ERR
1da177e4 LT	418	"multipath: couldn't allocate memory for %s\n",
	419	mdname(mddev));
	420	goto out_free_conf;
	421	}
1da177e4	422
92f861a7	423	working_disks = 0;
dafb20fa	424	rdev_for_each(rdev, mddev) {
1da177e4 LT	425	disk_idx = rdev->raid_disk;
	426	if (disk_idx < 0 \|\|
	427	disk_idx >= mddev->raid_disks)
	428	continue;
	429
	430	disk = conf->multipaths + disk_idx;
	431	disk->rdev = rdev;
8f6c2e4b MP	432	disk_stack_limits(mddev->gendisk, rdev->bdev,
8f6c2e4b MP	433	rdev->data_offset << 9);
1da177e4	434
1da177e4 LT	435	/* as we don't honour merge_bvec_fn, we must never risk
	436	* violating it, not that we ever expect a device with
	437	* a merge_bvec_fn to be involved in multipath */
627a2d3c N	438	if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
	439	blk_queue_max_segments(mddev->queue, 1);
	440	blk_queue_segment_boundary(mddev->queue,
	441	PAGE_CACHE_SIZE - 1);
	442	}
1da177e4	443
b2d444d7	444	if (!test_bit(Faulty, &rdev->flags))
92f861a7	445	working_disks++;
1da177e4 LT	446	}
	447
	448	conf->raid_disks = mddev->raid_disks;
1da177e4 LT	449	conf->mddev = mddev;
	450	spin_lock_init(&conf->device_lock);
	451	INIT_LIST_HEAD(&conf->retry_list);
	452
92f861a7	453	if (!working_disks) {
1da177e4 LT	454	printk(KERN_ERR "multipath: no operational IO paths for %s\n",
	455	mdname(mddev));
	456	goto out_free_conf;
	457	}
92f861a7	458	mddev->degraded = conf->raid_disks - working_disks;
1da177e4	459
bbba809e	460	conf->pool = mempool_create_kmalloc_pool(NR_RESERVED_BUFS,
26b6e051	461	sizeof(struct multipath_bh));
1da177e4	462	if (conf->pool == NULL) {
f72ffdd6	463	printk(KERN_ERR
1da177e4 LT	464	"multipath: couldn't allocate memory for %s\n",
	465	mdname(mddev));
	466	goto out_free_conf;
	467	}
	468
	469	{
0232605d N	470	mddev->thread = md_register_thread(multipathd, mddev,
0232605d N	471	"multipath");
1da177e4 LT	472	if (!mddev->thread) {
	473	printk(KERN_ERR "multipath: couldn't allocate thread"
	474	" for %s\n", mdname(mddev));
	475	goto out_free_conf;
	476	}
	477	}
	478
f72ffdd6	479	printk(KERN_INFO
1da177e4	480	"multipath: array %s active with %d out of %d IO paths\n",
92f861a7 N	481	mdname(mddev), conf->raid_disks - mddev->degraded,
92f861a7 N	482	mddev->raid_disks);
1da177e4 LT	483	/*
	484	* Ok, everything is just fine now
	485	*/
1f403624	486	md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
7a5febe9	487
a91a2785 MP	488	if (md_integrity_register(mddev))
	489	goto out_free_conf;
	490
1da177e4 LT	491	return 0;
	492
	493	out_free_conf:
	494	if (conf->pool)
	495	mempool_destroy(conf->pool);
990a8baf	496	kfree(conf->multipaths);
1da177e4 LT	497	kfree(conf);
	498	mddev->private = NULL;
	499	out:
	500	return -EIO;
	501	}
	502
afa0f557	503	static void multipath_free(struct mddev mddev, void priv)
1da177e4	504	{
afa0f557	505	struct mpconf *conf = priv;
1da177e4	506
1da177e4 LT	507	mempool_destroy(conf->pool);
	508	kfree(conf->multipaths);
	509	kfree(conf);
1da177e4 LT	510	}
1da177e4 LT	511
84fc4b56	512	static struct md_personality multipath_personality =
1da177e4 LT	513	{
1da177e4 LT	514	.name = "multipath",
2604b703	515	.level = LEVEL_MULTIPATH,
1da177e4 LT	516	.owner = THIS_MODULE,
	517	.make_request = multipath_make_request,
	518	.run = multipath_run,
afa0f557	519	.free = multipath_free,
1da177e4 LT	520	.status = multipath_status,
	521	.error_handler = multipath_error,
	522	.hot_add_disk = multipath_add_disk,
	523	.hot_remove_disk= multipath_remove_disk,
80c3a6ce	524	.size = multipath_size,
5c675f83	525	.congested = multipath_congested,
1da177e4 LT	526	};
	527
	528	static int __init multipath_init (void)
	529	{
2604b703	530	return register_md_personality (&multipath_personality);
1da177e4 LT	531	}
	532
	533	static void __exit multipath_exit (void)
	534	{
2604b703	535	unregister_md_personality (&multipath_personality);
1da177e4 LT	536	}
	537
	538	module_init(multipath_init);
	539	module_exit(multipath_exit);
	540	MODULE_LICENSE("GPL");
0efb9e61	541	MODULE_DESCRIPTION("simple multi-path personality for MD");
1da177e4	542	MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
d9d166c2	543	MODULE_ALIAS("md-multipath");
2604b703	544	MODULE_ALIAS("md-level--4");