[deliverable/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/module.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/raid/multipath.h>
#include <linux/buffer_head.h>
#include <asm/atomic.h>

#define MAJOR_NR MD_MAJOR
#define MD_DRIVER
#define MD_PERSONALITY

#define MAX_WORK_PER_DISK 128

#define	NR_RESERVED_BUFS	32


static int multipath_map (multipath_conf_t *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++) {
		mdk_rdev_t *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;
	mddev_t *mddev = mp_bh->mddev;
	multipath_conf_t *conf = mddev_to_conf(mddev);

	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;
	multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);

	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 = (struct multipath_bh *)(bio->bi_private);
	multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
	mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;

	if (uptodate)
		multipath_end_bh_io(mp_bh, 0);
	else if (!bio_rw_ahead(bio)) {
		/*
		 * 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_sector);
		multipath_reschedule_retry(mp_bh);
	} else
		multipath_end_bh_io(mp_bh, error);
	rdev_dec_pending(rdev, conf->mddev);
}

static void unplug_slaves(mddev_t *mddev)
{
	multipath_conf_t *conf = mddev_to_conf(mddev);
	int i;

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

			atomic_inc(&rdev->nr_pending);
			rcu_read_unlock();

			blk_unplug(r_queue);

			rdev_dec_pending(rdev, mddev);
			rcu_read_lock();
		}
	}
	rcu_read_unlock();
}

static void multipath_unplug(struct request_queue *q)
{
	unplug_slaves(q->queuedata);
}


static int multipath_make_request (struct request_queue *q, struct bio * bio)
{
	mddev_t *mddev = q->queuedata;
	multipath_conf_t *conf = mddev_to_conf(mddev);
	struct multipath_bh * mp_bh;
	struct multipath_info *multipath;
	const int rw = bio_data_dir(bio);

	if (unlikely(bio_barrier(bio))) {
		bio_endio(bio, -EOPNOTSUPP);
		return 0;
	}

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

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

	disk_stat_inc(mddev->gendisk, ios[rw]);
	disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));

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

	mp_bh->bio = *bio;
	mp_bh->bio.bi_sector += multipath->rdev->data_offset;
	mp_bh->bio.bi_bdev = multipath->rdev->bdev;
	mp_bh->bio.bi_rw |= (1 << BIO_RW_FAILFAST);
	mp_bh->bio.bi_end_io = multipath_end_request;
	mp_bh->bio.bi_private = mp_bh;
	generic_make_request(&mp_bh->bio);
	return 0;
}

static void multipath_status (struct seq_file *seq, mddev_t *mddev)
{
	multipath_conf_t *conf = mddev_to_conf(mddev);
	int i;
	
	seq_printf (seq, " [%d/%d] [", conf->raid_disks,
						 conf->working_disks);
	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(void *data, int bits)
{
	mddev_t *mddev = data;
	multipath_conf_t *conf = mddev_to_conf(mddev);
	int i, ret = 0;

	rcu_read_lock();
	for (i = 0; i < mddev->raid_disks ; i++) {
		mdk_rdev_t *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 (mddev_t *mddev, mdk_rdev_t *rdev)
{
	multipath_conf_t *conf = mddev_to_conf(mddev);

	if (conf->working_disks <= 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 */
	} else {
		/*
		 * Mark disk as unusable
		 */
		if (!test_bit(Faulty, &rdev->flags)) {
			char b[BDEVNAME_SIZE];
			clear_bit(In_sync, &rdev->flags);
			set_bit(Faulty, &rdev->flags);
			set_bit(MD_CHANGE_DEVS, &mddev->flags);
			conf->working_disks--;
			mddev->degraded++;
			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->working_disks);
		}
	}
}

static void print_multipath_conf (multipath_conf_t *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->working_disks,
			 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(mddev_t *mddev, mdk_rdev_t *rdev)
{
	multipath_conf_t *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;
			blk_queue_stack_limits(mddev->queue, q);

		/* as we don't honour merge_bvec_fn, we must never risk
		 * violating it, so limit ->max_sector to one PAGE, as
		 * a one page request is never in violation.
		 * (Note: it is very unlikely that a device with
		 * merge_bvec_fn will be involved in multipath.)
		 */
			if (q->merge_bvec_fn &&
			    mddev->queue->max_sectors > (PAGE_SIZE>>9))
				blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);

			conf->working_disks++;
			mddev->degraded--;
			rdev->raid_disk = path;
			set_bit(In_sync, &rdev->flags);
			rcu_assign_pointer(p->rdev, rdev);
			err = 0;
			break;
		}

	print_multipath_conf(conf);

	return err;
}

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

	print_multipath_conf(conf);

	rdev = p->rdev;
	if (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;
		}
	}
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 (mddev_t *mddev)
{
	struct multipath_bh *mp_bh;
	struct bio *bio;
	unsigned long flags;
	multipath_conf_t *conf = mddev_to_conf(mddev);
	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_sector = mp_bh->master_bio->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_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_sector);
			*bio = *(mp_bh->master_bio);
			bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset;
			bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
			bio->bi_rw |= (1 << BIO_RW_FAILFAST);
			bio->bi_end_io = multipath_end_request;
			bio->bi_private = mp_bh;
			generic_make_request(bio);
		}
	}
	spin_unlock_irqrestore(&conf->device_lock, flags);
}

static int multipath_run (mddev_t *mddev)
{
	multipath_conf_t *conf;
	int disk_idx;
	struct multipath_info *disk;
	mdk_rdev_t *rdev;
	struct list_head *tmp;

	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_stop()]
	 */
	mddev->queue->queue_lock = &mddev->queue->__queue_lock;

	conf = kzalloc(sizeof(multipath_conf_t), 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;
	}

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

		disk = conf->multipaths + disk_idx;
		disk->rdev = rdev;

		blk_queue_stack_limits(mddev->queue,
				       rdev->bdev->bd_disk->queue);
		/* 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 &&
		    mddev->queue->max_sectors > (PAGE_SIZE>>9))
			blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);

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

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

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

	conf->pool = mempool_create_kzalloc_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, "%s_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->working_disks, mddev->raid_disks);
	/*
	 * Ok, everything is just fine now
	 */
	mddev->array_sectors = mddev->size * 2;

	mddev->queue->unplug_fn = multipath_unplug;
	mddev->queue->backing_dev_info.congested_fn = multipath_congested;
	mddev->queue->backing_dev_info.congested_data = mddev;

	return 0;

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


static int multipath_stop (mddev_t *mddev)
{
	multipath_conf_t *conf = mddev_to_conf(mddev);

	md_unregister_thread(mddev->thread);
	mddev->thread = NULL;
	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
	mempool_destroy(conf->pool);
	kfree(conf->multipaths);
	kfree(conf);
	mddev->private = NULL;
	return 0;
}

static struct mdk_personality multipath_personality =
{
	.name		= "multipath",
	.level		= LEVEL_MULTIPATH,
	.owner		= THIS_MODULE,
	.make_request	= multipath_make_request,
	.run		= multipath_run,
	.stop		= multipath_stop,
	.status		= multipath_status,
	.error_handler	= multipath_error,
	.hot_add_disk	= multipath_add_disk,
	.hot_remove_disk= multipath_remove_disk,
};

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_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
	22	#include <linux/module.h>
	23	#include <linux/slab.h>
	24	#include <linux/spinlock.h>
	25	#include <linux/raid/multipath.h>
	26	#include <linux/buffer_head.h>
	27	#include <asm/atomic.h>
	28
	29	#define MAJOR_NR MD_MAJOR
	30	#define MD_DRIVER
	31	#define MD_PERSONALITY
	32
	33	#define MAX_WORK_PER_DISK 128
	34
	35	#define NR_RESERVED_BUFS 32
	36
	37
1da177e4 LT	38	static int multipath_map (multipath_conf_t *conf)
	39	{
	40	int i, disks = conf->raid_disks;
	41
	42	/*
	43	* Later we do read balancing on the read side
	44	* now we use the first available disk.
	45	*/
	46
	47	rcu_read_lock();
	48	for (i = 0; i < disks; i++) {
d6065f7b	49	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
b2d444d7	50	if (rdev && test_bit(In_sync, &rdev->flags)) {
1da177e4 LT	51	atomic_inc(&rdev->nr_pending);
	52	rcu_read_unlock();
	53	return i;
	54	}
	55	}
	56	rcu_read_unlock();
	57
	58	printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
	59	return (-1);
	60	}
	61
	62	static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
	63	{
	64	unsigned long flags;
	65	mddev_t *mddev = mp_bh->mddev;
	66	multipath_conf_t *conf = mddev_to_conf(mddev);
	67
	68	spin_lock_irqsave(&conf->device_lock, flags);
	69	list_add(&mp_bh->retry_list, &conf->retry_list);
	70	spin_unlock_irqrestore(&conf->device_lock, flags);
	71	md_wakeup_thread(mddev->thread);
	72	}
	73
	74
	75	/*
	76	* multipath_end_bh_io() is called when we have finished servicing a multipathed
	77	* operation and are ready to return a success/failure code to the buffer
	78	* cache layer.
	79	*/
	80	static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
	81	{
	82	struct bio *bio = mp_bh->master_bio;
	83	multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
	84
6712ecf8	85	bio_endio(bio, err);
1da177e4 LT	86	mempool_free(mp_bh, conf->pool);
	87	}
	88
6712ecf8	89	static void multipath_end_request(struct bio *bio, int error)
1da177e4 LT	90	{
	91	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	92	struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private);
	93	multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
	94	mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;
	95
1da177e4 LT	96	if (uptodate)
	97	multipath_end_bh_io(mp_bh, 0);
	98	else if (!bio_rw_ahead(bio)) {
	99	/*
	100	* oops, IO error:
	101	*/
	102	char b[BDEVNAME_SIZE];
	103	md_error (mp_bh->mddev, rdev);
	104	printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
	105	bdevname(rdev->bdev,b),
	106	(unsigned long long)bio->bi_sector);
	107	multipath_reschedule_retry(mp_bh);
	108	} else
	109	multipath_end_bh_io(mp_bh, error);
	110	rdev_dec_pending(rdev, conf->mddev);
1da177e4 LT	111	}
	112
	113	static void unplug_slaves(mddev_t *mddev)
	114	{
	115	multipath_conf_t *conf = mddev_to_conf(mddev);
	116	int i;
	117
	118	rcu_read_lock();
	119	for (i=0; i<mddev->raid_disks; i++) {
d6065f7b	120	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
b2d444d7 N	121	if (rdev && !test_bit(Faulty, &rdev->flags)
b2d444d7 N	122	&& atomic_read(&rdev->nr_pending)) {
165125e1	123	struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
1da177e4 LT	124
	125	atomic_inc(&rdev->nr_pending);
	126	rcu_read_unlock();
	127
2ad8b1ef	128	blk_unplug(r_queue);
1da177e4 LT	129
	130	rdev_dec_pending(rdev, mddev);
	131	rcu_read_lock();
	132	}
	133	}
	134	rcu_read_unlock();
	135	}
	136
165125e1	137	static void multipath_unplug(struct request_queue *q)
1da177e4 LT	138	{
	139	unplug_slaves(q->queuedata);
	140	}
	141
	142
165125e1	143	static int multipath_make_request (struct request_queue q, struct bio bio)
1da177e4 LT	144	{
	145	mddev_t *mddev = q->queuedata;
	146	multipath_conf_t *conf = mddev_to_conf(mddev);
	147	struct multipath_bh * mp_bh;
	148	struct multipath_info *multipath;
a362357b	149	const int rw = bio_data_dir(bio);
1da177e4	150
e5dcdd80	151	if (unlikely(bio_barrier(bio))) {
6712ecf8	152	bio_endio(bio, -EOPNOTSUPP);
e5dcdd80 N	153	return 0;
	154	}
	155
1da177e4 LT	156	mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
	157
	158	mp_bh->master_bio = bio;
	159	mp_bh->mddev = mddev;
	160
a362357b JA	161	disk_stat_inc(mddev->gendisk, ios[rw]);
a362357b JA	162	disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
1da177e4 LT	163
	164	mp_bh->path = multipath_map(conf);
	165	if (mp_bh->path < 0) {
6712ecf8	166	bio_endio(bio, -EIO);
1da177e4 LT	167	mempool_free(mp_bh, conf->pool);
	168	return 0;
	169	}
	170	multipath = conf->multipaths + mp_bh->path;
	171
	172	mp_bh->bio = *bio;
	173	mp_bh->bio.bi_sector += multipath->rdev->data_offset;
	174	mp_bh->bio.bi_bdev = multipath->rdev->bdev;
	175	mp_bh->bio.bi_rw \|= (1 << BIO_RW_FAILFAST);
	176	mp_bh->bio.bi_end_io = multipath_end_request;
	177	mp_bh->bio.bi_private = mp_bh;
	178	generic_make_request(&mp_bh->bio);
	179	return 0;
	180	}
	181
	182	static void multipath_status (struct seq_file seq, mddev_t mddev)
	183	{
	184	multipath_conf_t *conf = mddev_to_conf(mddev);
	185	int i;
	186
	187	seq_printf (seq, " [%d/%d] [", conf->raid_disks,
	188	conf->working_disks);
	189	for (i = 0; i < conf->raid_disks; i++)
	190	seq_printf (seq, "%s",
	191	conf->multipaths[i].rdev &&
b2d444d7	192	test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
1da177e4 LT	193	seq_printf (seq, "]");
	194	}
	195
0d129228 N	196	static int multipath_congested(void *data, int bits)
	197	{
	198	mddev_t *mddev = data;
	199	multipath_conf_t *conf = mddev_to_conf(mddev);
	200	int i, ret = 0;
	201
	202	rcu_read_lock();
	203	for (i = 0; i < mddev->raid_disks ; i++) {
	204	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
	205	if (rdev && !test_bit(Faulty, &rdev->flags)) {
165125e1	206	struct request_queue *q = bdev_get_queue(rdev->bdev);
0d129228 N	207
	208	ret \|= bdi_congested(&q->backing_dev_info, bits);
	209	/* Just like multipath_map, we just check the
	210	* first available device
	211	*/
	212	break;
	213	}
	214	}
	215	rcu_read_unlock();
	216	return ret;
	217	}
1da177e4 LT	218
	219	/*
	220	* Careful, this can execute in IRQ contexts as well!
	221	*/
	222	static void multipath_error (mddev_t mddev, mdk_rdev_t rdev)
	223	{
	224	multipath_conf_t *conf = mddev_to_conf(mddev);
	225
	226	if (conf->working_disks <= 1) {
	227	/*
	228	* Uh oh, we can do nothing if this is our last path, but
	229	* first check if this is a queued request for a device
	230	* which has just failed.
	231	*/
	232	printk(KERN_ALERT
	233	"multipath: only one IO path left and IO error.\n");
	234	/* leave it active... it's all we have */
	235	} else {
	236	/*
	237	* Mark disk as unusable
	238	*/
b2d444d7	239	if (!test_bit(Faulty, &rdev->flags)) {
1da177e4	240	char b[BDEVNAME_SIZE];
b2d444d7 N	241	clear_bit(In_sync, &rdev->flags);
b2d444d7 N	242	set_bit(Faulty, &rdev->flags);
850b2b42	243	set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4	244	conf->working_disks--;
750a8f3e	245	mddev->degraded++;
1da177e4	246	printk(KERN_ALERT "multipath: IO failure on %s,"
d7a420c9 NA	247	" disabling IO path.\n"
d7a420c9 NA	248	"multipath: Operation continuing"
1da177e4 LT	249	" on %d IO paths.\n",
	250	bdevname (rdev->bdev,b),
	251	conf->working_disks);
	252	}
	253	}
	254	}
	255
	256	static void print_multipath_conf (multipath_conf_t *conf)
	257	{
	258	int i;
	259	struct multipath_info *tmp;
	260
	261	printk("MULTIPATH conf printout:\n");
	262	if (!conf) {
	263	printk("(conf==NULL)\n");
	264	return;
	265	}
	266	printk(" --- wd:%d rd:%d\n", conf->working_disks,
	267	conf->raid_disks);
	268
	269	for (i = 0; i < conf->raid_disks; i++) {
	270	char b[BDEVNAME_SIZE];
	271	tmp = conf->multipaths + i;
	272	if (tmp->rdev)
	273	printk(" disk%d, o:%d, dev:%s\n",
b2d444d7	274	i,!test_bit(Faulty, &tmp->rdev->flags),
1da177e4 LT	275	bdevname(tmp->rdev->bdev,b));
	276	}
	277	}
	278
	279
	280	static int multipath_add_disk(mddev_t mddev, mdk_rdev_t rdev)
	281	{
	282	multipath_conf_t *conf = mddev->private;
352d768b	283	struct request_queue *q;
199050ea	284	int err = -EEXIST;
1da177e4 LT	285	int path;
1da177e4 LT	286	struct multipath_info *p;
6c2fce2e NB	287	int first = 0;
	288	int last = mddev->raid_disks - 1;
	289
	290	if (rdev->raid_disk >= 0)
	291	first = last = rdev->raid_disk;
1da177e4 LT	292
	293	print_multipath_conf(conf);
	294
6c2fce2e	295	for (path = first; path <= last; path++)
1da177e4	296	if ((p=conf->multipaths+path)->rdev == NULL) {
352d768b JJ	297	q = rdev->bdev->bd_disk->queue;
352d768b JJ	298	blk_queue_stack_limits(mddev->queue, q);
1da177e4 LT	299
	300	/* as we don't honour merge_bvec_fn, we must never risk
	301	* violating it, so limit ->max_sector to one PAGE, as
	302	* a one page request is never in violation.
	303	* (Note: it is very unlikely that a device with
	304	* merge_bvec_fn will be involved in multipath.)
	305	*/
352d768b	306	if (q->merge_bvec_fn &&
1da177e4 LT	307	mddev->queue->max_sectors > (PAGE_SIZE>>9))
	308	blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
	309
	310	conf->working_disks++;
750a8f3e	311	mddev->degraded--;
1da177e4	312	rdev->raid_disk = path;
b2d444d7	313	set_bit(In_sync, &rdev->flags);
d6065f7b	314	rcu_assign_pointer(p->rdev, rdev);
199050ea NB	315	err = 0;
199050ea NB	316	break;
1da177e4 LT	317	}
	318
	319	print_multipath_conf(conf);
199050ea NB	320
199050ea NB	321	return err;
1da177e4 LT	322	}
	323
	324	static int multipath_remove_disk(mddev_t *mddev, int number)
	325	{
	326	multipath_conf_t *conf = mddev->private;
	327	int err = 0;
	328	mdk_rdev_t *rdev;
	329	struct multipath_info *p = conf->multipaths + number;
	330
	331	print_multipath_conf(conf);
	332
	333	rdev = p->rdev;
	334	if (rdev) {
b2d444d7	335	if (test_bit(In_sync, &rdev->flags) \|\|
1da177e4	336	atomic_read(&rdev->nr_pending)) {
dfc70645 N	337	printk(KERN_ERR "hot-remove-disk, slot %d is identified"
dfc70645 N	338	" but is still operational!\n", number);
1da177e4 LT	339	err = -EBUSY;
	340	goto abort;
	341	}
	342	p->rdev = NULL;
fbd568a3	343	synchronize_rcu();
1da177e4 LT	344	if (atomic_read(&rdev->nr_pending)) {
	345	/* lost the race, try later */
	346	err = -EBUSY;
	347	p->rdev = rdev;
	348	}
	349	}
	350	abort:
	351
	352	print_multipath_conf(conf);
	353	return err;
	354	}
	355
	356
	357
	358	/*
	359	* This is a kernel thread which:
	360	*
	361	* 1. Retries failed read operations on working multipaths.
	362	* 2. Updates the raid superblock when problems encounter.
	363	* 3. Performs writes following reads for array syncronising.
	364	*/
	365
	366	static void multipathd (mddev_t *mddev)
	367	{
	368	struct multipath_bh *mp_bh;
	369	struct bio *bio;
	370	unsigned long flags;
	371	multipath_conf_t *conf = mddev_to_conf(mddev);
	372	struct list_head *head = &conf->retry_list;
	373
	374	md_check_recovery(mddev);
	375	for (;;) {
	376	char b[BDEVNAME_SIZE];
	377	spin_lock_irqsave(&conf->device_lock, flags);
	378	if (list_empty(head))
	379	break;
	380	mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
	381	list_del(head->prev);
	382	spin_unlock_irqrestore(&conf->device_lock, flags);
	383
	384	bio = &mp_bh->bio;
	385	bio->bi_sector = mp_bh->master_bio->bi_sector;
	386
	387	if ((mp_bh->path = multipath_map (conf))<0) {
	388	printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
	389	" error for block %llu\n",
	390	bdevname(bio->bi_bdev,b),
	391	(unsigned long long)bio->bi_sector);
	392	multipath_end_bh_io(mp_bh, -EIO);
	393	} else {
	394	printk(KERN_ERR "multipath: %s: redirecting sector %llu"
	395	" to another IO path\n",
	396	bdevname(bio->bi_bdev,b),
	397	(unsigned long long)bio->bi_sector);
	398	bio = (mp_bh->master_bio);
	399	bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset;
	400	bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
	401	bio->bi_rw \|= (1 << BIO_RW_FAILFAST);
	402	bio->bi_end_io = multipath_end_request;
	403	bio->bi_private = mp_bh;
	404	generic_make_request(bio);
	405	}
	406	}
	407	spin_unlock_irqrestore(&conf->device_lock, flags);
408	}
409
410	static int multipath_run (mddev_t *mddev)
411	{
412	multipath_conf_t *conf;
413	int disk_idx;
414	struct multipath_info *disk;
415	mdk_rdev_t *rdev;
416	struct list_head *tmp;
417
418	if (mddev->level != LEVEL_MULTIPATH) {
419	printk("multipath: %s: raid level not set to multipath IO (%d)\n",
420	mdname(mddev), mddev->level);
421	goto out;
422	}
423	/*
424	* copy the already verified devices into our private MULTIPATH
425	* bookkeeping area. [whatever we allocate in multipath_run(),
426	* should be freed in multipath_stop()]
427	*/
e7e72bf6	428	mddev->queue->queue_lock = &mddev->queue->__queue_lock;
1da177e4	429
9ffae0cf	430	conf = kzalloc(sizeof(multipath_conf_t), GFP_KERNEL);
1da177e4 LT	431	mddev->private = conf;
	432	if (!conf) {
	433	printk(KERN_ERR
	434	"multipath: couldn't allocate memory for %s\n",
	435	mdname(mddev));
	436	goto out;
	437	}
1da177e4	438
9ffae0cf	439	conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
1da177e4 LT	440	GFP_KERNEL);
	441	if (!conf->multipaths) {
	442	printk(KERN_ERR
	443	"multipath: couldn't allocate memory for %s\n",
	444	mdname(mddev));
	445	goto out_free_conf;
	446	}
1da177e4	447
1da177e4	448	conf->working_disks = 0;
d089c6af	449	rdev_for_each(rdev, tmp, mddev) {
1da177e4 LT	450	disk_idx = rdev->raid_disk;
	451	if (disk_idx < 0 \|\|
	452	disk_idx >= mddev->raid_disks)
	453	continue;
	454
	455	disk = conf->multipaths + disk_idx;
	456	disk->rdev = rdev;
	457
	458	blk_queue_stack_limits(mddev->queue,
	459	rdev->bdev->bd_disk->queue);
	460	/* as we don't honour merge_bvec_fn, we must never risk
	461	* violating it, not that we ever expect a device with
	462	* a merge_bvec_fn to be involved in multipath */
	463	if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
	464	mddev->queue->max_sectors > (PAGE_SIZE>>9))
	465	blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
	466
b2d444d7	467	if (!test_bit(Faulty, &rdev->flags))
1da177e4 LT	468	conf->working_disks++;
	469	}
	470
	471	conf->raid_disks = mddev->raid_disks;
1da177e4 LT	472	conf->mddev = mddev;
	473	spin_lock_init(&conf->device_lock);
	474	INIT_LIST_HEAD(&conf->retry_list);
	475
	476	if (!conf->working_disks) {
	477	printk(KERN_ERR "multipath: no operational IO paths for %s\n",
	478	mdname(mddev));
	479	goto out_free_conf;
	480	}
2e333e89	481	mddev->degraded = conf->raid_disks - conf->working_disks;
1da177e4	482
26b6e051 MD	483	conf->pool = mempool_create_kzalloc_pool(NR_RESERVED_BUFS,
26b6e051 MD	484	sizeof(struct multipath_bh));
1da177e4 LT	485	if (conf->pool == NULL) {
	486	printk(KERN_ERR
	487	"multipath: couldn't allocate memory for %s\n",
	488	mdname(mddev));
	489	goto out_free_conf;
	490	}
	491
	492	{
	493	mddev->thread = md_register_thread(multipathd, mddev, "%s_multipath");
	494	if (!mddev->thread) {
	495	printk(KERN_ERR "multipath: couldn't allocate thread"
	496	" for %s\n", mdname(mddev));
	497	goto out_free_conf;
	498	}
	499	}
	500
	501	printk(KERN_INFO
	502	"multipath: array %s active with %d out of %d IO paths\n",
	503	mdname(mddev), conf->working_disks, mddev->raid_disks);
	504	/*
	505	* Ok, everything is just fine now
	506	*/
f233ea5c	507	mddev->array_sectors = mddev->size * 2;
7a5febe9 N	508
7a5febe9 N	509	mddev->queue->unplug_fn = multipath_unplug;
0d129228 N	510	mddev->queue->backing_dev_info.congested_fn = multipath_congested;
0d129228 N	511	mddev->queue->backing_dev_info.congested_data = mddev;
7a5febe9	512
1da177e4 LT	513	return 0;
	514
	515	out_free_conf:
	516	if (conf->pool)
	517	mempool_destroy(conf->pool);
990a8baf	518	kfree(conf->multipaths);
1da177e4 LT	519	kfree(conf);
	520	mddev->private = NULL;
	521	out:
	522	return -EIO;
	523	}
	524
	525
	526	static int multipath_stop (mddev_t *mddev)
	527	{
	528	multipath_conf_t *conf = mddev_to_conf(mddev);
	529
	530	md_unregister_thread(mddev->thread);
	531	mddev->thread = NULL;
	532	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
	533	mempool_destroy(conf->pool);
	534	kfree(conf->multipaths);
	535	kfree(conf);
	536	mddev->private = NULL;
	537	return 0;
	538	}
	539
2604b703	540	static struct mdk_personality multipath_personality =
1da177e4 LT	541	{
1da177e4 LT	542	.name = "multipath",
2604b703	543	.level = LEVEL_MULTIPATH,
1da177e4 LT	544	.owner = THIS_MODULE,
	545	.make_request = multipath_make_request,
	546	.run = multipath_run,
	547	.stop = multipath_stop,
	548	.status = multipath_status,
	549	.error_handler = multipath_error,
	550	.hot_add_disk = multipath_add_disk,
	551	.hot_remove_disk= multipath_remove_disk,
	552	};
	553
	554	static int __init multipath_init (void)
	555	{
2604b703	556	return register_md_personality (&multipath_personality);
1da177e4 LT	557	}
	558
	559	static void __exit multipath_exit (void)
	560	{
2604b703	561	unregister_md_personality (&multipath_personality);
1da177e4 LT	562	}
	563
	564	module_init(multipath_init);
	565	module_exit(multipath_exit);
	566	MODULE_LICENSE("GPL");
	567	MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
d9d166c2	568	MODULE_ALIAS("md-multipath");
2604b703	569	MODULE_ALIAS("md-level--4");