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

static int multipath_end_request(struct bio *bio, unsigned int bytes_done,
				 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 (bio->bi_size)
		return 1;

	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);
	return 0;
}

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)) {
			request_queue_t *r_queue = bdev_get_queue(rdev->bdev);

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

			if (r_queue->unplug_fn)
				r_queue->unplug_fn(r_queue);

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

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


static int multipath_make_request (request_queue_t *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, bio->bi_size, -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, bio->bi_size, -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_issue_flush(request_queue_t *q, struct gendisk *disk,
				 sector_t *error_sector)
{
	mddev_t *mddev = q->queuedata;
	multipath_conf_t *conf = mddev_to_conf(mddev);
	int i, ret = 0;

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

			if (!r_queue->issue_flush_fn)
				ret = -EOPNOTSUPP;
			else {
				atomic_inc(&rdev->nr_pending);
				rcu_read_unlock();
				ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
							      error_sector);
				rdev_dec_pending(rdev, mddev);
				rcu_read_lock();
			}
		}
	}
	rcu_read_unlock();
	return ret;
}
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)) {
			request_queue_t *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	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 found = 0;
	int path;
	struct multipath_info *p;

	print_multipath_conf(conf);

	for (path=0; path<mddev->raid_disks; 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);
			found = 1;
		}

	print_multipath_conf(conf);
	return found;
}

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()]
	 */

	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;
	ITERATE_RDEV(mddev,rdev,tmp) {
		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_size = mddev->size;

	mddev->queue->unplug_fn = multipath_unplug;
	mddev->queue->issue_flush_fn = multipath_issue_flush;
	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
	85	bio_endio(bio, bio->bi_size, err);
	86	mempool_free(mp_bh, conf->pool);
	87	}
	88
75c96f85 AB	89	static int multipath_end_request(struct bio *bio, unsigned int bytes_done,
75c96f85 AB	90	int error)
1da177e4 LT	91	{
	92	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	93	struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private);
	94	multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev);
	95	mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;
	96
	97	if (bio->bi_size)
	98	return 1;
	99
	100	if (uptodate)
	101	multipath_end_bh_io(mp_bh, 0);
	102	else if (!bio_rw_ahead(bio)) {
	103	/*
	104	* oops, IO error:
	105	*/
	106	char b[BDEVNAME_SIZE];
	107	md_error (mp_bh->mddev, rdev);
	108	printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
	109	bdevname(rdev->bdev,b),
	110	(unsigned long long)bio->bi_sector);
	111	multipath_reschedule_retry(mp_bh);
	112	} else
	113	multipath_end_bh_io(mp_bh, error);
	114	rdev_dec_pending(rdev, conf->mddev);
	115	return 0;
	116	}
	117
	118	static void unplug_slaves(mddev_t *mddev)
	119	{
	120	multipath_conf_t *conf = mddev_to_conf(mddev);
	121	int i;
	122
	123	rcu_read_lock();
	124	for (i=0; i<mddev->raid_disks; i++) {
d6065f7b	125	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
b2d444d7 N	126	if (rdev && !test_bit(Faulty, &rdev->flags)
b2d444d7 N	127	&& atomic_read(&rdev->nr_pending)) {
1da177e4 LT	128	request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
	129
	130	atomic_inc(&rdev->nr_pending);
	131	rcu_read_unlock();
	132
	133	if (r_queue->unplug_fn)
	134	r_queue->unplug_fn(r_queue);
	135
	136	rdev_dec_pending(rdev, mddev);
	137	rcu_read_lock();
	138	}
	139	}
	140	rcu_read_unlock();
	141	}
	142
	143	static void multipath_unplug(request_queue_t *q)
	144	{
	145	unplug_slaves(q->queuedata);
	146	}
	147
	148
	149	static int multipath_make_request (request_queue_t q, struct bio bio)
	150	{
	151	mddev_t *mddev = q->queuedata;
	152	multipath_conf_t *conf = mddev_to_conf(mddev);
	153	struct multipath_bh * mp_bh;
	154	struct multipath_info *multipath;
a362357b	155	const int rw = bio_data_dir(bio);
1da177e4	156
e5dcdd80 N	157	if (unlikely(bio_barrier(bio))) {
	158	bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
	159	return 0;
	160	}
	161
1da177e4 LT	162	mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
	163
	164	mp_bh->master_bio = bio;
	165	mp_bh->mddev = mddev;
	166
a362357b JA	167	disk_stat_inc(mddev->gendisk, ios[rw]);
a362357b JA	168	disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
1da177e4 LT	169
	170	mp_bh->path = multipath_map(conf);
	171	if (mp_bh->path < 0) {
	172	bio_endio(bio, bio->bi_size, -EIO);
	173	mempool_free(mp_bh, conf->pool);
	174	return 0;
	175	}
	176	multipath = conf->multipaths + mp_bh->path;
	177
	178	mp_bh->bio = *bio;
	179	mp_bh->bio.bi_sector += multipath->rdev->data_offset;
	180	mp_bh->bio.bi_bdev = multipath->rdev->bdev;
	181	mp_bh->bio.bi_rw \|= (1 << BIO_RW_FAILFAST);
	182	mp_bh->bio.bi_end_io = multipath_end_request;
	183	mp_bh->bio.bi_private = mp_bh;
	184	generic_make_request(&mp_bh->bio);
	185	return 0;
	186	}
	187
	188	static void multipath_status (struct seq_file seq, mddev_t mddev)
	189	{
	190	multipath_conf_t *conf = mddev_to_conf(mddev);
	191	int i;
	192
	193	seq_printf (seq, " [%d/%d] [", conf->raid_disks,
	194	conf->working_disks);
	195	for (i = 0; i < conf->raid_disks; i++)
	196	seq_printf (seq, "%s",
	197	conf->multipaths[i].rdev &&
b2d444d7	198	test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
1da177e4 LT	199	seq_printf (seq, "]");
	200	}
	201
	202	static int multipath_issue_flush(request_queue_t q, struct gendisk disk,
	203	sector_t *error_sector)
	204	{
	205	mddev_t *mddev = q->queuedata;
	206	multipath_conf_t *conf = mddev_to_conf(mddev);
	207	int i, ret = 0;
	208
	209	rcu_read_lock();
	210	for (i=0; i<mddev->raid_disks && ret == 0; i++) {
d6065f7b	211	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
b2d444d7	212	if (rdev && !test_bit(Faulty, &rdev->flags)) {
1da177e4 LT	213	struct block_device *bdev = rdev->bdev;
	214	request_queue_t *r_queue = bdev_get_queue(bdev);
	215
	216	if (!r_queue->issue_flush_fn)
	217	ret = -EOPNOTSUPP;
	218	else {
	219	atomic_inc(&rdev->nr_pending);
	220	rcu_read_unlock();
	221	ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
	222	error_sector);
	223	rdev_dec_pending(rdev, mddev);
	224	rcu_read_lock();
	225	}
	226	}
	227	}
	228	rcu_read_unlock();
	229	return ret;
	230	}
0d129228 N	231	static int multipath_congested(void *data, int bits)
	232	{
	233	mddev_t *mddev = data;
	234	multipath_conf_t *conf = mddev_to_conf(mddev);
	235	int i, ret = 0;
	236
	237	rcu_read_lock();
	238	for (i = 0; i < mddev->raid_disks ; i++) {
	239	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
	240	if (rdev && !test_bit(Faulty, &rdev->flags)) {
	241	request_queue_t *q = bdev_get_queue(rdev->bdev);
	242
	243	ret \|= bdi_congested(&q->backing_dev_info, bits);
	244	/* Just like multipath_map, we just check the
	245	* first available device
	246	*/
	247	break;
	248	}
	249	}
	250	rcu_read_unlock();
	251	return ret;
	252	}
1da177e4 LT	253
	254	/*
	255	* Careful, this can execute in IRQ contexts as well!
	256	*/
	257	static void multipath_error (mddev_t mddev, mdk_rdev_t rdev)
	258	{
	259	multipath_conf_t *conf = mddev_to_conf(mddev);
	260
	261	if (conf->working_disks <= 1) {
	262	/*
	263	* Uh oh, we can do nothing if this is our last path, but
	264	* first check if this is a queued request for a device
	265	* which has just failed.
	266	*/
	267	printk(KERN_ALERT
	268	"multipath: only one IO path left and IO error.\n");
	269	/* leave it active... it's all we have */
	270	} else {
	271	/*
	272	* Mark disk as unusable
	273	*/
b2d444d7	274	if (!test_bit(Faulty, &rdev->flags)) {
1da177e4	275	char b[BDEVNAME_SIZE];
b2d444d7 N	276	clear_bit(In_sync, &rdev->flags);
b2d444d7 N	277	set_bit(Faulty, &rdev->flags);
850b2b42	278	set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4	279	conf->working_disks--;
750a8f3e	280	mddev->degraded++;
1da177e4 LT	281	printk(KERN_ALERT "multipath: IO failure on %s,"
	282	" disabling IO path. \n Operation continuing"
	283	" on %d IO paths.\n",
	284	bdevname (rdev->bdev,b),
	285	conf->working_disks);
	286	}
	287	}
	288	}
	289
	290	static void print_multipath_conf (multipath_conf_t *conf)
	291	{
	292	int i;
	293	struct multipath_info *tmp;
	294
	295	printk("MULTIPATH conf printout:\n");
	296	if (!conf) {
	297	printk("(conf==NULL)\n");
	298	return;
	299	}
	300	printk(" --- wd:%d rd:%d\n", conf->working_disks,
	301	conf->raid_disks);
	302
	303	for (i = 0; i < conf->raid_disks; i++) {
	304	char b[BDEVNAME_SIZE];
	305	tmp = conf->multipaths + i;
	306	if (tmp->rdev)
	307	printk(" disk%d, o:%d, dev:%s\n",
b2d444d7	308	i,!test_bit(Faulty, &tmp->rdev->flags),
1da177e4 LT	309	bdevname(tmp->rdev->bdev,b));
	310	}
	311	}
	312
	313
	314	static int multipath_add_disk(mddev_t mddev, mdk_rdev_t rdev)
	315	{
	316	multipath_conf_t *conf = mddev->private;
352d768b	317	struct request_queue *q;
1da177e4 LT	318	int found = 0;
	319	int path;
	320	struct multipath_info *p;
	321
	322	print_multipath_conf(conf);
	323
	324	for (path=0; path<mddev->raid_disks; path++)
	325	if ((p=conf->multipaths+path)->rdev == NULL) {
352d768b JJ	326	q = rdev->bdev->bd_disk->queue;
352d768b JJ	327	blk_queue_stack_limits(mddev->queue, q);
1da177e4 LT	328
	329	/* as we don't honour merge_bvec_fn, we must never risk
	330	* violating it, so limit ->max_sector to one PAGE, as
	331	* a one page request is never in violation.
	332	* (Note: it is very unlikely that a device with
	333	* merge_bvec_fn will be involved in multipath.)
	334	*/
352d768b	335	if (q->merge_bvec_fn &&
1da177e4 LT	336	mddev->queue->max_sectors > (PAGE_SIZE>>9))
	337	blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
	338
	339	conf->working_disks++;
750a8f3e	340	mddev->degraded--;
1da177e4	341	rdev->raid_disk = path;
b2d444d7	342	set_bit(In_sync, &rdev->flags);
d6065f7b	343	rcu_assign_pointer(p->rdev, rdev);
1da177e4 LT	344	found = 1;
	345	}
	346
	347	print_multipath_conf(conf);
	348	return found;
	349	}
	350
	351	static int multipath_remove_disk(mddev_t *mddev, int number)
	352	{
	353	multipath_conf_t *conf = mddev->private;
	354	int err = 0;
	355	mdk_rdev_t *rdev;
	356	struct multipath_info *p = conf->multipaths + number;
	357
	358	print_multipath_conf(conf);
	359
	360	rdev = p->rdev;
	361	if (rdev) {
b2d444d7	362	if (test_bit(In_sync, &rdev->flags) \|\|
1da177e4 LT	363	atomic_read(&rdev->nr_pending)) {
	364	printk(KERN_ERR "hot-remove-disk, slot %d is identified" " but is still operational!\n", number);
	365	err = -EBUSY;
	366	goto abort;
	367	}
	368	p->rdev = NULL;
fbd568a3	369	synchronize_rcu();
1da177e4 LT	370	if (atomic_read(&rdev->nr_pending)) {
	371	/* lost the race, try later */
	372	err = -EBUSY;
	373	p->rdev = rdev;
	374	}
	375	}
	376	abort:
	377
	378	print_multipath_conf(conf);
	379	return err;
	380	}
	381
	382
	383
	384	/*
	385	* This is a kernel thread which:
	386	*
	387	* 1. Retries failed read operations on working multipaths.
	388	* 2. Updates the raid superblock when problems encounter.
	389	* 3. Performs writes following reads for array syncronising.
	390	*/
	391
	392	static void multipathd (mddev_t *mddev)
	393	{
	394	struct multipath_bh *mp_bh;
	395	struct bio *bio;
	396	unsigned long flags;
	397	multipath_conf_t *conf = mddev_to_conf(mddev);
	398	struct list_head *head = &conf->retry_list;
	399
	400	md_check_recovery(mddev);
	401	for (;;) {
	402	char b[BDEVNAME_SIZE];
	403	spin_lock_irqsave(&conf->device_lock, flags);
	404	if (list_empty(head))
	405	break;
	406	mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
	407	list_del(head->prev);
	408	spin_unlock_irqrestore(&conf->device_lock, flags);
	409
	410	bio = &mp_bh->bio;
	411	bio->bi_sector = mp_bh->master_bio->bi_sector;
	412
	413	if ((mp_bh->path = multipath_map (conf))<0) {
	414	printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
	415	" error for block %llu\n",
	416	bdevname(bio->bi_bdev,b),
	417	(unsigned long long)bio->bi_sector);
	418	multipath_end_bh_io(mp_bh, -EIO);
	419	} else {
	420	printk(KERN_ERR "multipath: %s: redirecting sector %llu"
	421	" to another IO path\n",
	422	bdevname(bio->bi_bdev,b),
	423	(unsigned long long)bio->bi_sector);
	424	bio = (mp_bh->master_bio);
	425	bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset;
	426	bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
	427	bio->bi_rw \|= (1 << BIO_RW_FAILFAST);
	428	bio->bi_end_io = multipath_end_request;
	429	bio->bi_private = mp_bh;
	430	generic_make_request(bio);
	431	}
	432	}
	433	spin_unlock_irqrestore(&conf->device_lock, flags);
434	}
435
436	static int multipath_run (mddev_t *mddev)
437	{
438	multipath_conf_t *conf;
439	int disk_idx;
440	struct multipath_info *disk;
441	mdk_rdev_t *rdev;
442	struct list_head *tmp;
443
444	if (mddev->level != LEVEL_MULTIPATH) {
445	printk("multipath: %s: raid level not set to multipath IO (%d)\n",
446	mdname(mddev), mddev->level);
447	goto out;
448	}
449	/*
450	* copy the already verified devices into our private MULTIPATH
451	* bookkeeping area. [whatever we allocate in multipath_run(),
452	* should be freed in multipath_stop()]
453	*/
454
9ffae0cf	455	conf = kzalloc(sizeof(multipath_conf_t), GFP_KERNEL);
1da177e4 LT	456	mddev->private = conf;
	457	if (!conf) {
	458	printk(KERN_ERR
	459	"multipath: couldn't allocate memory for %s\n",
	460	mdname(mddev));
	461	goto out;
	462	}
1da177e4	463
9ffae0cf	464	conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
1da177e4 LT	465	GFP_KERNEL);
	466	if (!conf->multipaths) {
	467	printk(KERN_ERR
	468	"multipath: couldn't allocate memory for %s\n",
	469	mdname(mddev));
	470	goto out_free_conf;
	471	}
1da177e4	472
1da177e4 LT	473	conf->working_disks = 0;
	474	ITERATE_RDEV(mddev,rdev,tmp) {
	475	disk_idx = rdev->raid_disk;
	476	if (disk_idx < 0 \|\|
	477	disk_idx >= mddev->raid_disks)
	478	continue;
	479
	480	disk = conf->multipaths + disk_idx;
	481	disk->rdev = rdev;
	482
	483	blk_queue_stack_limits(mddev->queue,
	484	rdev->bdev->bd_disk->queue);
	485	/* as we don't honour merge_bvec_fn, we must never risk
	486	* violating it, not that we ever expect a device with
	487	* a merge_bvec_fn to be involved in multipath */
	488	if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
	489	mddev->queue->max_sectors > (PAGE_SIZE>>9))
	490	blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
	491
b2d444d7	492	if (!test_bit(Faulty, &rdev->flags))
1da177e4 LT	493	conf->working_disks++;
	494	}
	495
	496	conf->raid_disks = mddev->raid_disks;
1da177e4 LT	497	conf->mddev = mddev;
	498	spin_lock_init(&conf->device_lock);
	499	INIT_LIST_HEAD(&conf->retry_list);
	500
	501	if (!conf->working_disks) {
	502	printk(KERN_ERR "multipath: no operational IO paths for %s\n",
	503	mdname(mddev));
	504	goto out_free_conf;
	505	}
2e333e89	506	mddev->degraded = conf->raid_disks - conf->working_disks;
1da177e4	507
26b6e051 MD	508	conf->pool = mempool_create_kzalloc_pool(NR_RESERVED_BUFS,
26b6e051 MD	509	sizeof(struct multipath_bh));
1da177e4 LT	510	if (conf->pool == NULL) {
	511	printk(KERN_ERR
	512	"multipath: couldn't allocate memory for %s\n",
	513	mdname(mddev));
	514	goto out_free_conf;
	515	}
	516
	517	{
	518	mddev->thread = md_register_thread(multipathd, mddev, "%s_multipath");
	519	if (!mddev->thread) {
	520	printk(KERN_ERR "multipath: couldn't allocate thread"
	521	" for %s\n", mdname(mddev));
	522	goto out_free_conf;
	523	}
	524	}
	525
	526	printk(KERN_INFO
	527	"multipath: array %s active with %d out of %d IO paths\n",
	528	mdname(mddev), conf->working_disks, mddev->raid_disks);
	529	/*
	530	* Ok, everything is just fine now
	531	*/
	532	mddev->array_size = mddev->size;
7a5febe9 N	533
	534	mddev->queue->unplug_fn = multipath_unplug;
	535	mddev->queue->issue_flush_fn = multipath_issue_flush;
0d129228 N	536	mddev->queue->backing_dev_info.congested_fn = multipath_congested;
0d129228 N	537	mddev->queue->backing_dev_info.congested_data = mddev;
7a5febe9	538
1da177e4 LT	539	return 0;
	540
	541	out_free_conf:
	542	if (conf->pool)
	543	mempool_destroy(conf->pool);
990a8baf	544	kfree(conf->multipaths);
1da177e4 LT	545	kfree(conf);
	546	mddev->private = NULL;
	547	out:
	548	return -EIO;
	549	}
	550
	551
	552	static int multipath_stop (mddev_t *mddev)
	553	{
	554	multipath_conf_t *conf = mddev_to_conf(mddev);
	555
	556	md_unregister_thread(mddev->thread);
	557	mddev->thread = NULL;
	558	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
	559	mempool_destroy(conf->pool);
	560	kfree(conf->multipaths);
	561	kfree(conf);
	562	mddev->private = NULL;
	563	return 0;
	564	}
	565
2604b703	566	static struct mdk_personality multipath_personality =
1da177e4 LT	567	{
1da177e4 LT	568	.name = "multipath",
2604b703	569	.level = LEVEL_MULTIPATH,
1da177e4 LT	570	.owner = THIS_MODULE,
	571	.make_request = multipath_make_request,
	572	.run = multipath_run,
	573	.stop = multipath_stop,
	574	.status = multipath_status,
	575	.error_handler = multipath_error,
	576	.hot_add_disk = multipath_add_disk,
	577	.hot_remove_disk= multipath_remove_disk,
	578	};
	579
	580	static int __init multipath_init (void)
	581	{
2604b703	582	return register_md_personality (&multipath_personality);
1da177e4 LT	583	}
	584
	585	static void __exit multipath_exit (void)
	586	{
2604b703	587	unregister_md_personality (&multipath_personality);
1da177e4 LT	588	}
	589
	590	module_init(multipath_init);
	591	module_exit(multipath_exit);
	592	MODULE_LICENSE("GPL");
	593	MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
d9d166c2	594	MODULE_ALIAS("md-multipath");
2604b703	595	MODULE_ALIAS("md-level--4");