[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 "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_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 &&
			    queue_max_sectors(q) > (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;
			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 &&
		    queue_max_sectors(mddev->queue) > (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_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>
43b2e5d8	25	#include "md.h"
ef740c37	26	#include "multipath.h"
1da177e4 LT	27
	28	#define MAX_WORK_PER_DISK 128
	29
	30	#define NR_RESERVED_BUFS 32
	31
	32
1da177e4 LT	33	static int multipath_map (multipath_conf_t *conf)
	34	{
	35	int i, disks = conf->raid_disks;
	36
	37	/*
	38	* Later we do read balancing on the read side
	39	* now we use the first available disk.
	40	*/
	41
	42	rcu_read_lock();
	43	for (i = 0; i < disks; i++) {
d6065f7b	44	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
b2d444d7	45	if (rdev && test_bit(In_sync, &rdev->flags)) {
1da177e4 LT	46	atomic_inc(&rdev->nr_pending);
	47	rcu_read_unlock();
	48	return i;
	49	}
	50	}
	51	rcu_read_unlock();
	52
	53	printk(KERN_ERR "multipath_map(): no more operational IO paths?\n");
	54	return (-1);
	55	}
	56
	57	static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
	58	{
	59	unsigned long flags;
	60	mddev_t *mddev = mp_bh->mddev;
070ec55d	61	multipath_conf_t *conf = mddev->private;
1da177e4 LT	62
	63	spin_lock_irqsave(&conf->device_lock, flags);
	64	list_add(&mp_bh->retry_list, &conf->retry_list);
	65	spin_unlock_irqrestore(&conf->device_lock, flags);
	66	md_wakeup_thread(mddev->thread);
	67	}
	68
	69
	70	/*
	71	* multipath_end_bh_io() is called when we have finished servicing a multipathed
	72	* operation and are ready to return a success/failure code to the buffer
	73	* cache layer.
	74	*/
	75	static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err)
	76	{
	77	struct bio *bio = mp_bh->master_bio;
070ec55d	78	multipath_conf_t *conf = mp_bh->mddev->private;
1da177e4	79
6712ecf8	80	bio_endio(bio, err);
1da177e4 LT	81	mempool_free(mp_bh, conf->pool);
	82	}
	83
6712ecf8	84	static void multipath_end_request(struct bio *bio, int error)
1da177e4 LT	85	{
	86	int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
	87	struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private);
070ec55d	88	multipath_conf_t *conf = mp_bh->mddev->private;
1da177e4 LT	89	mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev;
1da177e4 LT	90
1da177e4 LT	91	if (uptodate)
1da177e4 LT	92	multipath_end_bh_io(mp_bh, 0);
1f98a13f	93	else if (!bio_rw_flagged(bio, BIO_RW_AHEAD)) {
1da177e4 LT	94	/*
	95	* oops, IO error:
	96	*/
	97	char b[BDEVNAME_SIZE];
	98	md_error (mp_bh->mddev, rdev);
	99	printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n",
	100	bdevname(rdev->bdev,b),
	101	(unsigned long long)bio->bi_sector);
	102	multipath_reschedule_retry(mp_bh);
	103	} else
	104	multipath_end_bh_io(mp_bh, error);
	105	rdev_dec_pending(rdev, conf->mddev);
1da177e4 LT	106	}
	107
	108	static void unplug_slaves(mddev_t *mddev)
	109	{
070ec55d	110	multipath_conf_t *conf = mddev->private;
1da177e4 LT	111	int i;
	112
	113	rcu_read_lock();
	114	for (i=0; i<mddev->raid_disks; i++) {
d6065f7b	115	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
b2d444d7 N	116	if (rdev && !test_bit(Faulty, &rdev->flags)
b2d444d7 N	117	&& atomic_read(&rdev->nr_pending)) {
165125e1	118	struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
1da177e4 LT	119
	120	atomic_inc(&rdev->nr_pending);
	121	rcu_read_unlock();
	122
2ad8b1ef	123	blk_unplug(r_queue);
1da177e4 LT	124
	125	rdev_dec_pending(rdev, mddev);
	126	rcu_read_lock();
	127	}
	128	}
	129	rcu_read_unlock();
	130	}
	131
165125e1	132	static void multipath_unplug(struct request_queue *q)
1da177e4 LT	133	{
	134	unplug_slaves(q->queuedata);
	135	}
	136
	137
165125e1	138	static int multipath_make_request (struct request_queue q, struct bio bio)
1da177e4 LT	139	{
1da177e4 LT	140	mddev_t *mddev = q->queuedata;
070ec55d	141	multipath_conf_t *conf = mddev->private;
1da177e4 LT	142	struct multipath_bh * mp_bh;
1da177e4 LT	143	struct multipath_info *multipath;
a362357b	144	const int rw = bio_data_dir(bio);
c9959059	145	int cpu;
1da177e4	146
1f98a13f	147	if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
a2826aa9	148	md_barrier_request(mddev, bio);
e5dcdd80 N	149	return 0;
	150	}
	151
1da177e4 LT	152	mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
	153
	154	mp_bh->master_bio = bio;
	155	mp_bh->mddev = mddev;
	156
074a7aca TH	157	cpu = part_stat_lock();
	158	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
	159	part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
	160	bio_sectors(bio));
	161	part_stat_unlock();
1da177e4 LT	162
	163	mp_bh->path = multipath_map(conf);
	164	if (mp_bh->path < 0) {
6712ecf8	165	bio_endio(bio, -EIO);
1da177e4 LT	166	mempool_free(mp_bh, conf->pool);
	167	return 0;
	168	}
	169	multipath = conf->multipaths + mp_bh->path;
	170
	171	mp_bh->bio = *bio;
	172	mp_bh->bio.bi_sector += multipath->rdev->data_offset;
	173	mp_bh->bio.bi_bdev = multipath->rdev->bdev;
6000a368	174	mp_bh->bio.bi_rw \|= (1 << BIO_RW_FAILFAST_TRANSPORT);
1da177e4 LT	175	mp_bh->bio.bi_end_io = multipath_end_request;
	176	mp_bh->bio.bi_private = mp_bh;
	177	generic_make_request(&mp_bh->bio);
	178	return 0;
	179	}
	180
	181	static void multipath_status (struct seq_file seq, mddev_t mddev)
	182	{
070ec55d	183	multipath_conf_t *conf = mddev->private;
1da177e4 LT	184	int i;
	185
	186	seq_printf (seq, " [%d/%d] [", conf->raid_disks,
	187	conf->working_disks);
	188	for (i = 0; i < conf->raid_disks; i++)
	189	seq_printf (seq, "%s",
	190	conf->multipaths[i].rdev &&
b2d444d7	191	test_bit(In_sync, &conf->multipaths[i].rdev->flags) ? "U" : "_");
1da177e4 LT	192	seq_printf (seq, "]");
	193	}
	194
0d129228 N	195	static int multipath_congested(void *data, int bits)
	196	{
	197	mddev_t *mddev = data;
070ec55d	198	multipath_conf_t *conf = mddev->private;
0d129228 N	199	int i, ret = 0;
0d129228 N	200
3fa841d7 N	201	if (mddev_congested(mddev, bits))
	202	return 1;
	203
0d129228 N	204	rcu_read_lock();
	205	for (i = 0; i < mddev->raid_disks ; i++) {
	206	mdk_rdev_t *rdev = rcu_dereference(conf->multipaths[i].rdev);
	207	if (rdev && !test_bit(Faulty, &rdev->flags)) {
165125e1	208	struct request_queue *q = bdev_get_queue(rdev->bdev);
0d129228 N	209
	210	ret \|= bdi_congested(&q->backing_dev_info, bits);
	211	/* Just like multipath_map, we just check the
	212	* first available device
	213	*/
	214	break;
	215	}
	216	}
	217	rcu_read_unlock();
	218	return ret;
	219	}
1da177e4 LT	220
	221	/*
	222	* Careful, this can execute in IRQ contexts as well!
	223	*/
	224	static void multipath_error (mddev_t mddev, mdk_rdev_t rdev)
	225	{
070ec55d	226	multipath_conf_t *conf = mddev->private;
1da177e4 LT	227
	228	if (conf->working_disks <= 1) {
	229	/*
	230	* Uh oh, we can do nothing if this is our last path, but
	231	* first check if this is a queued request for a device
	232	* which has just failed.
	233	*/
	234	printk(KERN_ALERT
	235	"multipath: only one IO path left and IO error.\n");
	236	/* leave it active... it's all we have */
	237	} else {
	238	/*
	239	* Mark disk as unusable
	240	*/
b2d444d7	241	if (!test_bit(Faulty, &rdev->flags)) {
1da177e4	242	char b[BDEVNAME_SIZE];
b2d444d7 N	243	clear_bit(In_sync, &rdev->flags);
b2d444d7 N	244	set_bit(Faulty, &rdev->flags);
850b2b42	245	set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4	246	conf->working_disks--;
750a8f3e	247	mddev->degraded++;
1da177e4	248	printk(KERN_ALERT "multipath: IO failure on %s,"
d7a420c9 NA	249	" disabling IO path.\n"
d7a420c9 NA	250	"multipath: Operation continuing"
1da177e4 LT	251	" on %d IO paths.\n",
	252	bdevname (rdev->bdev,b),
	253	conf->working_disks);
	254	}
	255	}
	256	}
	257
	258	static void print_multipath_conf (multipath_conf_t *conf)
	259	{
	260	int i;
	261	struct multipath_info *tmp;
	262
	263	printk("MULTIPATH conf printout:\n");
	264	if (!conf) {
	265	printk("(conf==NULL)\n");
	266	return;
	267	}
	268	printk(" --- wd:%d rd:%d\n", conf->working_disks,
	269	conf->raid_disks);
	270
	271	for (i = 0; i < conf->raid_disks; i++) {
	272	char b[BDEVNAME_SIZE];
	273	tmp = conf->multipaths + i;
	274	if (tmp->rdev)
	275	printk(" disk%d, o:%d, dev:%s\n",
b2d444d7	276	i,!test_bit(Faulty, &tmp->rdev->flags),
1da177e4 LT	277	bdevname(tmp->rdev->bdev,b));
	278	}
	279	}
	280
	281
	282	static int multipath_add_disk(mddev_t mddev, mdk_rdev_t rdev)
	283	{
	284	multipath_conf_t *conf = mddev->private;
352d768b	285	struct request_queue *q;
199050ea	286	int err = -EEXIST;
1da177e4 LT	287	int path;
1da177e4 LT	288	struct multipath_info *p;
6c2fce2e NB	289	int first = 0;
	290	int last = mddev->raid_disks - 1;
	291
	292	if (rdev->raid_disk >= 0)
	293	first = last = rdev->raid_disk;
1da177e4 LT	294
	295	print_multipath_conf(conf);
	296
6c2fce2e	297	for (path = first; path <= last; path++)
1da177e4	298	if ((p=conf->multipaths+path)->rdev == NULL) {
352d768b	299	q = rdev->bdev->bd_disk->queue;
8f6c2e4b MP	300	disk_stack_limits(mddev->gendisk, rdev->bdev,
8f6c2e4b MP	301	rdev->data_offset << 9);
1da177e4 LT	302
	303	/* as we don't honour merge_bvec_fn, we must never risk
	304	* violating it, so limit ->max_sector to one PAGE, as
	305	* a one page request is never in violation.
	306	* (Note: it is very unlikely that a device with
	307	* merge_bvec_fn will be involved in multipath.)
	308	*/
352d768b	309	if (q->merge_bvec_fn &&
ae03bf63	310	queue_max_sectors(q) > (PAGE_SIZE>>9))
1da177e4 LT	311	blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
	312
	313	conf->working_disks++;
750a8f3e	314	mddev->degraded--;
1da177e4	315	rdev->raid_disk = path;
b2d444d7	316	set_bit(In_sync, &rdev->flags);
d6065f7b	317	rcu_assign_pointer(p->rdev, rdev);
199050ea	318	err = 0;
ac5e7113	319	md_integrity_add_rdev(rdev, mddev);
199050ea	320	break;
1da177e4 LT	321	}
	322
	323	print_multipath_conf(conf);
199050ea NB	324
199050ea NB	325	return err;
1da177e4 LT	326	}
	327
	328	static int multipath_remove_disk(mddev_t *mddev, int number)
	329	{
	330	multipath_conf_t *conf = mddev->private;
	331	int err = 0;
	332	mdk_rdev_t *rdev;
	333	struct multipath_info *p = conf->multipaths + number;
	334
	335	print_multipath_conf(conf);
	336
	337	rdev = p->rdev;
	338	if (rdev) {
b2d444d7	339	if (test_bit(In_sync, &rdev->flags) \|\|
1da177e4	340	atomic_read(&rdev->nr_pending)) {
dfc70645 N	341	printk(KERN_ERR "hot-remove-disk, slot %d is identified"
dfc70645 N	342	" but is still operational!\n", number);
1da177e4 LT	343	err = -EBUSY;
	344	goto abort;
	345	}
	346	p->rdev = NULL;
fbd568a3	347	synchronize_rcu();
1da177e4 LT	348	if (atomic_read(&rdev->nr_pending)) {
	349	/* lost the race, try later */
	350	err = -EBUSY;
	351	p->rdev = rdev;
ac5e7113	352	goto abort;
1da177e4	353	}
ac5e7113	354	md_integrity_register(mddev);
1da177e4 LT	355	}
	356	abort:
	357
	358	print_multipath_conf(conf);
	359	return err;
	360	}
	361
	362
	363
	364	/*
	365	* This is a kernel thread which:
	366	*
	367	* 1. Retries failed read operations on working multipaths.
	368	* 2. Updates the raid superblock when problems encounter.
	369	* 3. Performs writes following reads for array syncronising.
	370	*/
	371
	372	static void multipathd (mddev_t *mddev)
	373	{
	374	struct multipath_bh *mp_bh;
	375	struct bio *bio;
	376	unsigned long flags;
070ec55d	377	multipath_conf_t *conf = mddev->private;
1da177e4 LT	378	struct list_head *head = &conf->retry_list;
	379
	380	md_check_recovery(mddev);
	381	for (;;) {
	382	char b[BDEVNAME_SIZE];
	383	spin_lock_irqsave(&conf->device_lock, flags);
	384	if (list_empty(head))
	385	break;
	386	mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
	387	list_del(head->prev);
	388	spin_unlock_irqrestore(&conf->device_lock, flags);
	389
	390	bio = &mp_bh->bio;
	391	bio->bi_sector = mp_bh->master_bio->bi_sector;
	392
	393	if ((mp_bh->path = multipath_map (conf))<0) {
	394	printk(KERN_ALERT "multipath: %s: unrecoverable IO read"
	395	" error for block %llu\n",
	396	bdevname(bio->bi_bdev,b),
	397	(unsigned long long)bio->bi_sector);
	398	multipath_end_bh_io(mp_bh, -EIO);
	399	} else {
	400	printk(KERN_ERR "multipath: %s: redirecting sector %llu"
	401	" to another IO path\n",
	402	bdevname(bio->bi_bdev,b),
	403	(unsigned long long)bio->bi_sector);
	404	bio = (mp_bh->master_bio);
	405	bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset;
	406	bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev;
6000a368	407	bio->bi_rw \|= (1 << BIO_RW_FAILFAST_TRANSPORT);
1da177e4 LT	408	bio->bi_end_io = multipath_end_request;
	409	bio->bi_private = mp_bh;
	410	generic_make_request(bio);
	411	}
	412	}
	413	spin_unlock_irqrestore(&conf->device_lock, flags);
	414	}
	415
80c3a6ce DW	416	static sector_t multipath_size(mddev_t *mddev, sector_t sectors, int raid_disks)
	417	{
	418	WARN_ONCE(sectors \|\| raid_disks,
	419	"%s does not support generic reshape\n", __func__);
	420
	421	return mddev->dev_sectors;
	422	}
	423
1da177e4 LT	424	static int multipath_run (mddev_t *mddev)
	425	{
	426	multipath_conf_t *conf;
	427	int disk_idx;
	428	struct multipath_info *disk;
	429	mdk_rdev_t *rdev;
1da177e4	430
0894cc30 AN	431	if (md_check_no_bitmap(mddev))
	432	return -EINVAL;
	433
1da177e4 LT	434	if (mddev->level != LEVEL_MULTIPATH) {
	435	printk("multipath: %s: raid level not set to multipath IO (%d)\n",
	436	mdname(mddev), mddev->level);
	437	goto out;
	438	}
	439	/*
	440	* copy the already verified devices into our private MULTIPATH
	441	* bookkeeping area. [whatever we allocate in multipath_run(),
	442	* should be freed in multipath_stop()]
	443	*/
e7e72bf6	444	mddev->queue->queue_lock = &mddev->queue->__queue_lock;
1da177e4	445
9ffae0cf	446	conf = kzalloc(sizeof(multipath_conf_t), GFP_KERNEL);
1da177e4 LT	447	mddev->private = conf;
	448	if (!conf) {
	449	printk(KERN_ERR
	450	"multipath: couldn't allocate memory for %s\n",
	451	mdname(mddev));
	452	goto out;
	453	}
1da177e4	454
9ffae0cf	455	conf->multipaths = kzalloc(sizeof(struct multipath_info)*mddev->raid_disks,
1da177e4 LT	456	GFP_KERNEL);
	457	if (!conf->multipaths) {
	458	printk(KERN_ERR
	459	"multipath: couldn't allocate memory for %s\n",
	460	mdname(mddev));
	461	goto out_free_conf;
	462	}
1da177e4	463
1da177e4	464	conf->working_disks = 0;
159ec1fc	465	list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 LT	466	disk_idx = rdev->raid_disk;
	467	if (disk_idx < 0 \|\|
	468	disk_idx >= mddev->raid_disks)
	469	continue;
	470
	471	disk = conf->multipaths + disk_idx;
	472	disk->rdev = rdev;
8f6c2e4b MP	473	disk_stack_limits(mddev->gendisk, rdev->bdev,
8f6c2e4b MP	474	rdev->data_offset << 9);
1da177e4	475
1da177e4 LT	476	/* as we don't honour merge_bvec_fn, we must never risk
	477	* violating it, not that we ever expect a device with
	478	* a merge_bvec_fn to be involved in multipath */
	479	if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
ae03bf63	480	queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
1da177e4 LT	481	blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1da177e4 LT	482
b2d444d7	483	if (!test_bit(Faulty, &rdev->flags))
1da177e4 LT	484	conf->working_disks++;
	485	}
	486
	487	conf->raid_disks = mddev->raid_disks;
1da177e4 LT	488	conf->mddev = mddev;
	489	spin_lock_init(&conf->device_lock);
	490	INIT_LIST_HEAD(&conf->retry_list);
	491
	492	if (!conf->working_disks) {
	493	printk(KERN_ERR "multipath: no operational IO paths for %s\n",
	494	mdname(mddev));
	495	goto out_free_conf;
	496	}
2e333e89	497	mddev->degraded = conf->raid_disks - conf->working_disks;
1da177e4	498
bbba809e	499	conf->pool = mempool_create_kmalloc_pool(NR_RESERVED_BUFS,
26b6e051	500	sizeof(struct multipath_bh));
1da177e4 LT	501	if (conf->pool == NULL) {
	502	printk(KERN_ERR
	503	"multipath: couldn't allocate memory for %s\n",
	504	mdname(mddev));
	505	goto out_free_conf;
	506	}
	507
	508	{
0da3c619	509	mddev->thread = md_register_thread(multipathd, mddev, NULL);
1da177e4 LT	510	if (!mddev->thread) {
	511	printk(KERN_ERR "multipath: couldn't allocate thread"
	512	" for %s\n", mdname(mddev));
	513	goto out_free_conf;
	514	}
	515	}
	516
	517	printk(KERN_INFO
	518	"multipath: array %s active with %d out of %d IO paths\n",
	519	mdname(mddev), conf->working_disks, mddev->raid_disks);
	520	/*
	521	* Ok, everything is just fine now
	522	*/
1f403624	523	md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
7a5febe9 N	524
7a5febe9 N	525	mddev->queue->unplug_fn = multipath_unplug;
0d129228 N	526	mddev->queue->backing_dev_info.congested_fn = multipath_congested;
0d129228 N	527	mddev->queue->backing_dev_info.congested_data = mddev;
ac5e7113	528	md_integrity_register(mddev);
1da177e4 LT	529	return 0;
	530
	531	out_free_conf:
	532	if (conf->pool)
	533	mempool_destroy(conf->pool);
990a8baf	534	kfree(conf->multipaths);
1da177e4 LT	535	kfree(conf);
	536	mddev->private = NULL;
	537	out:
	538	return -EIO;
	539	}
	540
	541
	542	static int multipath_stop (mddev_t *mddev)
	543	{
070ec55d	544	multipath_conf_t *conf = mddev->private;
1da177e4 LT	545
	546	md_unregister_thread(mddev->thread);
	547	mddev->thread = NULL;
	548	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
	549	mempool_destroy(conf->pool);
	550	kfree(conf->multipaths);
	551	kfree(conf);
	552	mddev->private = NULL;
	553	return 0;
	554	}
	555
2604b703	556	static struct mdk_personality multipath_personality =
1da177e4 LT	557	{
1da177e4 LT	558	.name = "multipath",
2604b703	559	.level = LEVEL_MULTIPATH,
1da177e4 LT	560	.owner = THIS_MODULE,
	561	.make_request = multipath_make_request,
	562	.run = multipath_run,
	563	.stop = multipath_stop,
	564	.status = multipath_status,
	565	.error_handler = multipath_error,
	566	.hot_add_disk = multipath_add_disk,
	567	.hot_remove_disk= multipath_remove_disk,
80c3a6ce	568	.size = multipath_size,
1da177e4 LT	569	};
	570
	571	static int __init multipath_init (void)
	572	{
2604b703	573	return register_md_personality (&multipath_personality);
1da177e4 LT	574	}
	575
	576	static void __exit multipath_exit (void)
	577	{
2604b703	578	unregister_md_personality (&multipath_personality);
1da177e4 LT	579	}
	580
	581	module_init(multipath_init);
	582	module_exit(multipath_exit);
	583	MODULE_LICENSE("GPL");
0efb9e61	584	MODULE_DESCRIPTION("simple multi-path personality for MD");
1da177e4	585	MODULE_ALIAS("md-personality-7"); /* MULTIPATH */
d9d166c2	586	MODULE_ALIAS("md-multipath");
2604b703	587	MODULE_ALIAS("md-level--4");