[linux-2.6-block.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/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 (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->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;
	multipath_conf_t *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 = (struct multipath_bh *)(bio->bi_private);
	multipath_conf_t *conf = mp_bh->mddev->private;
	mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;

	if (uptodate)
		multipath_end_bh_io(mp_bh, 0);
	else if (!bio_rw_flagged(bio, BIO_RW_AHEAD)) {
		/*
		 * 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->private;
	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->private;
	struct multipath_bh * mp_bh;
	struct multipath_info *multipath;
	const int rw = bio_data_dir(bio);
	int cpu;

	if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
		md_barrier_request(mddev, bio);
		return 0;
	}

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

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

	cpu = part_stat_lock();
	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
	part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
		      bio_sectors(bio));
	part_stat_unlock();

	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_TRANSPORT);
	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->private;
	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->private;
	int i, ret = 0;

	if (mddev_congested(mddev, bits))
		return 1;

	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->private;

	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;
			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);
			}

			conf->working_disks++;
			mddev->degraded--;
			rdev->raid_disk = path;
			set_bit(In_sync, &rdev->flags);
			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(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;
			goto abort;
		}
		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 (mddev_t *mddev)
{
	struct multipath_bh *mp_bh;
	struct bio *bio;
	unsigned long flags;
	multipath_conf_t *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_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_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(mddev_t *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 (mddev_t *mddev)
{
	multipath_conf_t *conf;
	int disk_idx;
	struct multipath_info *disk;
	mdk_rdev_t *rdev;

	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_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;
	list_for_each_entry(rdev, &mddev->disks, same_set) {
		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))
			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_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, NULL);
		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
	 */
	md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));

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

	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,
	.size		= multipath_size,
};

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 N	22	#include <linux/blkdev.h>
bff61975 N	23	#include <linux/raid/md_u.h>
bff61975	24	#include <linux/seq_file.h>
5a0e3ad6	25	#include <linux/slab.h>
43b2e5d8	26	#include "md.h"
ef740c37	27	#include "multipath.h"
1da177e4 LT	28
	29	#define MAX_WORK_PER_DISK 128
	30
	31	#define NR_RESERVED_BUFS 32
	32
	33
1da177e4 LT	34	static int multipath_map (multipath_conf_t *conf)
	35	{
	36	int i, disks = conf->raid_disks;
	37
	38	/*
	39	* Later we do read balancing on the read side
	40	* now we use the first available disk.
	41	*/
	42
	43	rcu_read_lock();
	44	for (i = 0; i < disks; i++) {
d6065f7b	45	mdk_rdev_t *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;
	61	mddev_t *mddev = mp_bh->mddev;
070ec55d	62	multipath_conf_t *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
	70
	71	/*
	72	* multipath_end_bh_io() is called when we have finished servicing a multipathed
	73	* operation and are ready to return a success/failure code to the buffer
	74	* cache layer.
	75	*/
	76	static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
	77	{
	78	struct bio *bio = mp_bh->master_bio;
070ec55d	79	multipath_conf_t *conf = mp_bh->mddev->private;
1da177e4	80
6712ecf8	81	bio_endio(bio, err);
1da177e4 LT	82	mempool_free(mp_bh, conf->pool);
	83	}
	84
6712ecf8	85	static void multipath_end_request(struct bio *bio, int error)
1da177e4 LT	86	{
	87	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	88	struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private);
070ec55d	89	multipath_conf_t *conf = mp_bh->mddev->private;
1da177e4 LT	90	mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;
1da177e4 LT	91
1da177e4 LT	92	if (uptodate)
1da177e4 LT	93	multipath_end_bh_io(mp_bh, 0);
1f98a13f	94	else if (!bio_rw_flagged(bio, BIO_RW_AHEAD)) {
1da177e4 LT	95	/*
	96	* oops, IO error:
	97	*/
	98	char b[BDEVNAME_SIZE];
	99	md_error (mp_bh->mddev, rdev);
	100	printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
	101	bdevname(rdev->bdev,b),
	102	(unsigned long long)bio->bi_sector);
	103	multipath_reschedule_retry(mp_bh);
	104	} else
	105	multipath_end_bh_io(mp_bh, error);
	106	rdev_dec_pending(rdev, conf->mddev);
1da177e4 LT	107	}
	108
	109	static void unplug_slaves(mddev_t *mddev)
	110	{
070ec55d	111	multipath_conf_t *conf = mddev->private;
1da177e4 LT	112	int i;
	113
	114	rcu_read_lock();
	115	for (i=0; i<mddev->raid_disks; i++) {
d6065f7b	116	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
b2d444d7 N	117	if (rdev && !test_bit(Faulty, &rdev->flags)
b2d444d7 N	118	&& atomic_read(&rdev->nr_pending)) {
165125e1	119	struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
1da177e4 LT	120
	121	atomic_inc(&rdev->nr_pending);
	122	rcu_read_unlock();
	123
2ad8b1ef	124	blk_unplug(r_queue);
1da177e4 LT	125
	126	rdev_dec_pending(rdev, mddev);
	127	rcu_read_lock();
	128	}
	129	}
	130	rcu_read_unlock();
	131	}
	132
165125e1	133	static void multipath_unplug(struct request_queue *q)
1da177e4 LT	134	{
	135	unplug_slaves(q->queuedata);
	136	}
	137
	138
165125e1	139	static int multipath_make_request (struct request_queue q, struct bio bio)
1da177e4 LT	140	{
1da177e4 LT	141	mddev_t *mddev = q->queuedata;
070ec55d	142	multipath_conf_t *conf = mddev->private;
1da177e4 LT	143	struct multipath_bh * mp_bh;
1da177e4 LT	144	struct multipath_info *multipath;
a362357b	145	const int rw = bio_data_dir(bio);
c9959059	146	int cpu;
1da177e4	147
1f98a13f	148	if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
a2826aa9	149	md_barrier_request(mddev, bio);
e5dcdd80 N	150	return 0;
	151	}
	152
1da177e4 LT	153	mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
	154
	155	mp_bh->master_bio = bio;
	156	mp_bh->mddev = mddev;
	157
074a7aca TH	158	cpu = part_stat_lock();
	159	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
	160	part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
	161	bio_sectors(bio));
	162	part_stat_unlock();
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;
6000a368	175	mp_bh->bio.bi_rw \|= (1 << BIO_RW_FAILFAST_TRANSPORT);
1da177e4 LT	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	{
070ec55d	184	multipath_conf_t *conf = mddev->private;
1da177e4 LT	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;
070ec55d	199	multipath_conf_t *conf = mddev->private;
0d129228 N	200	int i, ret = 0;
0d129228 N	201
3fa841d7 N	202	if (mddev_congested(mddev, bits))
	203	return 1;
	204
0d129228 N	205	rcu_read_lock();
	206	for (i = 0; i < mddev->raid_disks ; i++) {
	207	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
	208	if (rdev && !test_bit(Faulty, &rdev->flags)) {
165125e1	209	struct request_queue *q = bdev_get_queue(rdev->bdev);
0d129228 N	210
	211	ret \|= bdi_congested(&q->backing_dev_info, bits);
	212	/* Just like multipath_map, we just check the
	213	* first available device
	214	*/
	215	break;
	216	}
	217	}
	218	rcu_read_unlock();
	219	return ret;
	220	}
1da177e4 LT	221
	222	/*
	223	* Careful, this can execute in IRQ contexts as well!
	224	*/
	225	static void multipath_error (mddev_t mddev, mdk_rdev_t rdev)
	226	{
070ec55d	227	multipath_conf_t *conf = mddev->private;
1da177e4 LT	228
	229	if (conf->working_disks <= 1) {
	230	/*
	231	* Uh oh, we can do nothing if this is our last path, but
	232	* first check if this is a queued request for a device
	233	* which has just failed.
	234	*/
	235	printk(KERN_ALERT
	236	"multipath: only one IO path left and IO error.\n");
	237	/* leave it active... it's all we have */
	238	} else {
	239	/*
	240	* Mark disk as unusable
	241	*/
b2d444d7	242	if (!test_bit(Faulty, &rdev->flags)) {
1da177e4	243	char b[BDEVNAME_SIZE];
b2d444d7 N	244	clear_bit(In_sync, &rdev->flags);
b2d444d7 N	245	set_bit(Faulty, &rdev->flags);
850b2b42	246	set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4	247	conf->working_disks--;
750a8f3e	248	mddev->degraded++;
1da177e4	249	printk(KERN_ALERT "multipath: IO failure on %s,"
d7a420c9 NA	250	" disabling IO path.\n"
d7a420c9 NA	251	"multipath: Operation continuing"
1da177e4 LT	252	" on %d IO paths.\n",
	253	bdevname (rdev->bdev,b),
	254	conf->working_disks);
	255	}
	256	}
	257	}
	258
	259	static void print_multipath_conf (multipath_conf_t *conf)
	260	{
	261	int i;
	262	struct multipath_info *tmp;
	263
	264	printk("MULTIPATH conf printout:\n");
	265	if (!conf) {
	266	printk("(conf==NULL)\n");
	267	return;
	268	}
	269	printk(" --- wd:%d rd:%d\n", conf->working_disks,
	270	conf->raid_disks);
	271
	272	for (i = 0; i < conf->raid_disks; i++) {
	273	char b[BDEVNAME_SIZE];
	274	tmp = conf->multipaths + i;
	275	if (tmp->rdev)
	276	printk(" disk%d, o:%d, dev:%s\n",
b2d444d7	277	i,!test_bit(Faulty, &tmp->rdev->flags),
1da177e4 LT	278	bdevname(tmp->rdev->bdev,b));
	279	}
	280	}
	281
	282
	283	static int multipath_add_disk(mddev_t mddev, mdk_rdev_t rdev)
	284	{
	285	multipath_conf_t *conf = mddev->private;
352d768b	286	struct request_queue *q;
199050ea	287	int err = -EEXIST;
1da177e4 LT	288	int path;
1da177e4 LT	289	struct multipath_info *p;
6c2fce2e NB	290	int first = 0;
	291	int last = mddev->raid_disks - 1;
	292
	293	if (rdev->raid_disk >= 0)
	294	first = last = rdev->raid_disk;
1da177e4 LT	295
	296	print_multipath_conf(conf);
	297
6c2fce2e	298	for (path = first; path <= last; path++)
1da177e4	299	if ((p=conf->multipaths+path)->rdev == NULL) {
352d768b	300	q = rdev->bdev->bd_disk->queue;
8f6c2e4b MP	301	disk_stack_limits(mddev->gendisk, rdev->bdev,
8f6c2e4b MP	302	rdev->data_offset << 9);
1da177e4 LT	303
1da177e4 LT	304	/* as we don't honour merge_bvec_fn, we must never risk
627a2d3c N	305	* violating it, so limit ->max_segments to one, lying
627a2d3c N	306	* within a single page.
1da177e4 LT	307	* (Note: it is very unlikely that a device with
	308	* merge_bvec_fn will be involved in multipath.)
	309	*/
627a2d3c N	310	if (q->merge_bvec_fn) {
	311	blk_queue_max_segments(mddev->queue, 1);
	312	blk_queue_segment_boundary(mddev->queue,
	313	PAGE_CACHE_SIZE - 1);
	314	}
1da177e4 LT	315
1da177e4 LT	316	conf->working_disks++;
750a8f3e	317	mddev->degraded--;
1da177e4	318	rdev->raid_disk = path;
b2d444d7	319	set_bit(In_sync, &rdev->flags);
d6065f7b	320	rcu_assign_pointer(p->rdev, rdev);
199050ea	321	err = 0;
ac5e7113	322	md_integrity_add_rdev(rdev, mddev);
199050ea	323	break;
1da177e4 LT	324	}
	325
	326	print_multipath_conf(conf);
199050ea NB	327
199050ea NB	328	return err;
1da177e4 LT	329	}
	330
	331	static int multipath_remove_disk(mddev_t *mddev, int number)
	332	{
	333	multipath_conf_t *conf = mddev->private;
	334	int err = 0;
	335	mdk_rdev_t *rdev;
	336	struct multipath_info *p = conf->multipaths + number;
	337
	338	print_multipath_conf(conf);
	339
	340	rdev = p->rdev;
	341	if (rdev) {
b2d444d7	342	if (test_bit(In_sync, &rdev->flags) \|\|
1da177e4	343	atomic_read(&rdev->nr_pending)) {
dfc70645 N	344	printk(KERN_ERR "hot-remove-disk, slot %d is identified"
dfc70645 N	345	" but is still operational!\n", number);
1da177e4 LT	346	err = -EBUSY;
	347	goto abort;
	348	}
	349	p->rdev = NULL;
fbd568a3	350	synchronize_rcu();
1da177e4 LT	351	if (atomic_read(&rdev->nr_pending)) {
	352	/* lost the race, try later */
	353	err = -EBUSY;
	354	p->rdev = rdev;
ac5e7113	355	goto abort;
1da177e4	356	}
ac5e7113	357	md_integrity_register(mddev);
1da177e4 LT	358	}
	359	abort:
	360
	361	print_multipath_conf(conf);
	362	return err;
	363	}
	364
	365
	366
	367	/*
	368	* This is a kernel thread which:
	369	*
	370	* 1. Retries failed read operations on working multipaths.
	371	* 2. Updates the raid superblock when problems encounter.
	372	* 3. Performs writes following reads for array syncronising.
	373	*/
	374
	375	static void multipathd (mddev_t *mddev)
	376	{
	377	struct multipath_bh *mp_bh;
	378	struct bio *bio;
	379	unsigned long flags;
070ec55d	380	multipath_conf_t *conf = mddev->private;
1da177e4 LT	381	struct list_head *head = &conf->retry_list;
	382
	383	md_check_recovery(mddev);
	384	for (;;) {
	385	char b[BDEVNAME_SIZE];
	386	spin_lock_irqsave(&conf->device_lock, flags);
	387	if (list_empty(head))
	388	break;
	389	mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
	390	list_del(head->prev);
	391	spin_unlock_irqrestore(&conf->device_lock, flags);
	392
	393	bio = &mp_bh->bio;
	394	bio->bi_sector = mp_bh->master_bio->bi_sector;
	395
	396	if ((mp_bh->path = multipath_map (conf))<0) {
	397	printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
	398	" error for block %llu\n",
	399	bdevname(bio->bi_bdev,b),
	400	(unsigned long long)bio->bi_sector);
	401	multipath_end_bh_io(mp_bh, -EIO);
	402	} else {
	403	printk(KERN_ERR "multipath: %s: redirecting sector %llu"
	404	" to another IO path\n",
	405	bdevname(bio->bi_bdev,b),
	406	(unsigned long long)bio->bi_sector);
	407	bio = (mp_bh->master_bio);
	408	bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset;
	409	bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
6000a368	410	bio->bi_rw \|= (1 << BIO_RW_FAILFAST_TRANSPORT);
1da177e4 LT	411	bio->bi_end_io = multipath_end_request;
	412	bio->bi_private = mp_bh;
	413	generic_make_request(bio);
	414	}
	415	}
	416	spin_unlock_irqrestore(&conf->device_lock, flags);
	417	}
	418
80c3a6ce DW	419	static sector_t multipath_size(mddev_t *mddev, sector_t sectors, int raid_disks)
	420	{
	421	WARN_ONCE(sectors \|\| raid_disks,
	422	"%s does not support generic reshape\n", __func__);
	423
	424	return mddev->dev_sectors;
	425	}
	426
1da177e4 LT	427	static int multipath_run (mddev_t *mddev)
	428	{
	429	multipath_conf_t *conf;
	430	int disk_idx;
	431	struct multipath_info *disk;
	432	mdk_rdev_t *rdev;
1da177e4	433
0894cc30 AN	434	if (md_check_no_bitmap(mddev))
	435	return -EINVAL;
	436
1da177e4 LT	437	if (mddev->level != LEVEL_MULTIPATH) {
	438	printk("multipath: %s: raid level not set to multipath IO (%d)\n",
	439	mdname(mddev), mddev->level);
	440	goto out;
	441	}
	442	/*
	443	* copy the already verified devices into our private MULTIPATH
	444	* bookkeeping area. [whatever we allocate in multipath_run(),
	445	* should be freed in multipath_stop()]
	446	*/
e7e72bf6	447	mddev->queue->queue_lock = &mddev->queue->__queue_lock;
1da177e4	448
9ffae0cf	449	conf = kzalloc(sizeof(multipath_conf_t), GFP_KERNEL);
1da177e4 LT	450	mddev->private = conf;
	451	if (!conf) {
	452	printk(KERN_ERR
	453	"multipath: couldn't allocate memory for %s\n",
	454	mdname(mddev));
	455	goto out;
	456	}
1da177e4	457
9ffae0cf	458	conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
1da177e4 LT	459	GFP_KERNEL);
	460	if (!conf->multipaths) {
	461	printk(KERN_ERR
	462	"multipath: couldn't allocate memory for %s\n",
	463	mdname(mddev));
	464	goto out_free_conf;
	465	}
1da177e4	466
1da177e4	467	conf->working_disks = 0;
159ec1fc	468	list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 LT	469	disk_idx = rdev->raid_disk;
	470	if (disk_idx < 0 \|\|
	471	disk_idx >= mddev->raid_disks)
	472	continue;
	473
	474	disk = conf->multipaths + disk_idx;
	475	disk->rdev = rdev;
8f6c2e4b MP	476	disk_stack_limits(mddev->gendisk, rdev->bdev,
8f6c2e4b MP	477	rdev->data_offset << 9);
1da177e4	478
1da177e4 LT	479	/* as we don't honour merge_bvec_fn, we must never risk
	480	* violating it, not that we ever expect a device with
	481	* a merge_bvec_fn to be involved in multipath */
627a2d3c N	482	if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
	483	blk_queue_max_segments(mddev->queue, 1);
	484	blk_queue_segment_boundary(mddev->queue,
	485	PAGE_CACHE_SIZE - 1);
	486	}
1da177e4	487
b2d444d7	488	if (!test_bit(Faulty, &rdev->flags))
1da177e4 LT	489	conf->working_disks++;
	490	}
	491
	492	conf->raid_disks = mddev->raid_disks;
1da177e4 LT	493	conf->mddev = mddev;
	494	spin_lock_init(&conf->device_lock);
	495	INIT_LIST_HEAD(&conf->retry_list);
	496
	497	if (!conf->working_disks) {
	498	printk(KERN_ERR "multipath: no operational IO paths for %s\n",
	499	mdname(mddev));
	500	goto out_free_conf;
	501	}
2e333e89	502	mddev->degraded = conf->raid_disks - conf->working_disks;
1da177e4	503
bbba809e	504	conf->pool = mempool_create_kmalloc_pool(NR_RESERVED_BUFS,
26b6e051	505	sizeof(struct multipath_bh));
1da177e4 LT	506	if (conf->pool == NULL) {
	507	printk(KERN_ERR
	508	"multipath: couldn't allocate memory for %s\n",
	509	mdname(mddev));
	510	goto out_free_conf;
	511	}
	512
	513	{
0da3c619	514	mddev->thread = md_register_thread(multipathd, mddev, NULL);
1da177e4 LT	515	if (!mddev->thread) {
	516	printk(KERN_ERR "multipath: couldn't allocate thread"
	517	" for %s\n", mdname(mddev));
	518	goto out_free_conf;
	519	}
	520	}
	521
	522	printk(KERN_INFO
	523	"multipath: array %s active with %d out of %d IO paths\n",
	524	mdname(mddev), conf->working_disks, mddev->raid_disks);
	525	/*
	526	* Ok, everything is just fine now
	527	*/
1f403624	528	md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
7a5febe9 N	529
7a5febe9 N	530	mddev->queue->unplug_fn = multipath_unplug;
0d129228 N	531	mddev->queue->backing_dev_info.congested_fn = multipath_congested;
0d129228 N	532	mddev->queue->backing_dev_info.congested_data = mddev;
ac5e7113	533	md_integrity_register(mddev);
1da177e4 LT	534	return 0;
	535
	536	out_free_conf:
	537	if (conf->pool)
	538	mempool_destroy(conf->pool);
990a8baf	539	kfree(conf->multipaths);
1da177e4 LT	540	kfree(conf);
	541	mddev->private = NULL;
	542	out:
	543	return -EIO;
	544	}
	545
	546
	547	static int multipath_stop (mddev_t *mddev)
	548	{
070ec55d	549	multipath_conf_t *conf = mddev->private;
1da177e4 LT	550
	551	md_unregister_thread(mddev->thread);
	552	mddev->thread = NULL;
	553	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
	554	mempool_destroy(conf->pool);
	555	kfree(conf->multipaths);
	556	kfree(conf);
	557	mddev->private = NULL;
	558	return 0;
	559	}
	560
2604b703	561	static struct mdk_personality multipath_personality =
1da177e4 LT	562	{
1da177e4 LT	563	.name = "multipath",
2604b703	564	.level = LEVEL_MULTIPATH,
1da177e4 LT	565	.owner = THIS_MODULE,
	566	.make_request = multipath_make_request,
	567	.run = multipath_run,
	568	.stop = multipath_stop,
	569	.status = multipath_status,
	570	.error_handler = multipath_error,
	571	.hot_add_disk = multipath_add_disk,
	572	.hot_remove_disk= multipath_remove_disk,
80c3a6ce	573	.size = multipath_size,
1da177e4 LT	574	};
	575
	576	static int __init multipath_init (void)
	577	{
2604b703	578	return register_md_personality (&multipath_personality);
1da177e4 LT	579	}
	580
	581	static void __exit multipath_exit (void)
	582	{
2604b703	583	unregister_md_personality (&multipath_personality);
1da177e4 LT	584	}
	585
	586	module_init(multipath_init);
	587	module_exit(multipath_exit);
	588	MODULE_LICENSE("GPL");
0efb9e61	589	MODULE_DESCRIPTION("simple multi-path personality for MD");
1da177e4	590	MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
d9d166c2	591	MODULE_ALIAS("md-multipath");
2604b703	592	MODULE_ALIAS("md-level--4");