[linux-2.6-block.git] / drivers / md / raid0.c

/*
   raid0.c : Multiple Devices driver for Linux
             Copyright (C) 1994-96 Marc ZYNGIER
	     <zyngier@ufr-info-p7.ibp.fr> or
	     <maz@gloups.fdn.fr>
             Copyright (C) 1999, 2000 Ingo Molnar, Red Hat


   RAID-0 management functions.

   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/seq_file.h>
#include "md.h"
#include "raid0.h"

static void raid0_unplug(struct request_queue *q)
{
	mddev_t *mddev = q->queuedata;
	raid0_conf_t *conf = mddev_to_conf(mddev);
	mdk_rdev_t **devlist = conf->strip_zone[0].dev;
	int i;

	for (i=0; i<mddev->raid_disks; i++) {
		struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);

		blk_unplug(r_queue);
	}
}

static int raid0_congested(void *data, int bits)
{
	mddev_t *mddev = data;
	raid0_conf_t *conf = mddev_to_conf(mddev);
	mdk_rdev_t **devlist = conf->strip_zone[0].dev;
	int i, ret = 0;

	for (i = 0; i < mddev->raid_disks && !ret ; i++) {
		struct request_queue *q = bdev_get_queue(devlist[i]->bdev);

		ret |= bdi_congested(&q->backing_dev_info, bits);
	}
	return ret;
}


static int create_strip_zones (mddev_t *mddev)
{
	int i, c, j;
	sector_t current_start, curr_zone_start;
	sector_t min_spacing;
	raid0_conf_t *conf = mddev_to_conf(mddev);
	mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
	struct strip_zone *zone;
	int cnt;
	char b[BDEVNAME_SIZE];
 
	/*
	 * The number of 'same size groups'
	 */
	conf->nr_strip_zones = 0;
 
	list_for_each_entry(rdev1, &mddev->disks, same_set) {
		printk(KERN_INFO "raid0: looking at %s\n",
			bdevname(rdev1->bdev,b));
		c = 0;
		list_for_each_entry(rdev2, &mddev->disks, same_set) {
			printk(KERN_INFO "raid0:   comparing %s(%llu)",
			       bdevname(rdev1->bdev,b),
			       (unsigned long long)rdev1->sectors);
			printk(KERN_INFO " with %s(%llu)\n",
			       bdevname(rdev2->bdev,b),
			       (unsigned long long)rdev2->sectors);
			if (rdev2 == rdev1) {
				printk(KERN_INFO "raid0:   END\n");
				break;
			}
			if (rdev2->sectors == rdev1->sectors) {
				/*
				 * Not unique, don't count it as a new
				 * group
				 */
				printk(KERN_INFO "raid0:   EQUAL\n");
				c = 1;
				break;
			}
			printk(KERN_INFO "raid0:   NOT EQUAL\n");
		}
		if (!c) {
			printk(KERN_INFO "raid0:   ==> UNIQUE\n");
			conf->nr_strip_zones++;
			printk(KERN_INFO "raid0: %d zones\n",
				conf->nr_strip_zones);
		}
	}
	printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);

	conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
				conf->nr_strip_zones, GFP_KERNEL);
	if (!conf->strip_zone)
		return 1;
	conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
				conf->nr_strip_zones*mddev->raid_disks,
				GFP_KERNEL);
	if (!conf->devlist)
		return 1;

	/* The first zone must contain all devices, so here we check that
	 * there is a proper alignment of slots to devices and find them all
	 */
	zone = &conf->strip_zone[0];
	cnt = 0;
	smallest = NULL;
	zone->dev = conf->devlist;
	list_for_each_entry(rdev1, &mddev->disks, same_set) {
		int j = rdev1->raid_disk;

		if (j < 0 || j >= mddev->raid_disks) {
			printk(KERN_ERR "raid0: bad disk number %d - "
				"aborting!\n", j);
			goto abort;
		}
		if (zone->dev[j]) {
			printk(KERN_ERR "raid0: multiple devices for %d - "
				"aborting!\n", j);
			goto abort;
		}
		zone->dev[j] = rdev1;

		blk_queue_stack_limits(mddev->queue,
				       rdev1->bdev->bd_disk->queue);
		/* 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.
		 */

		if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
		    mddev->queue->max_sectors > (PAGE_SIZE>>9))
			blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);

		if (!smallest || (rdev1->sectors < smallest->sectors))
			smallest = rdev1;
		cnt++;
	}
	if (cnt != mddev->raid_disks) {
		printk(KERN_ERR "raid0: too few disks (%d of %d) - "
			"aborting!\n", cnt, mddev->raid_disks);
		goto abort;
	}
	zone->nb_dev = cnt;
	zone->sectors = smallest->sectors * cnt;
	zone->zone_start = 0;

	current_start = smallest->sectors;
	curr_zone_start = zone->sectors;

	/* now do the other zones */
	for (i = 1; i < conf->nr_strip_zones; i++)
	{
		zone = conf->strip_zone + i;
		zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;

		printk(KERN_INFO "raid0: zone %d\n", i);
		zone->dev_start = current_start;
		smallest = NULL;
		c = 0;

		for (j=0; j<cnt; j++) {
			char b[BDEVNAME_SIZE];
			rdev = conf->strip_zone[0].dev[j];
			printk(KERN_INFO "raid0: checking %s ...",
				bdevname(rdev->bdev, b));
			if (rdev->sectors <= current_start) {
				printk(KERN_INFO " nope.\n");
				continue;
			}
			printk(KERN_INFO " contained as device %d\n", c);
			zone->dev[c] = rdev;
			c++;
			if (!smallest || rdev->sectors < smallest->sectors) {
				smallest = rdev;
				printk(KERN_INFO "  (%llu) is smallest!.\n",
					(unsigned long long)rdev->sectors);
			}
		}

		zone->nb_dev = c;
		zone->sectors = (smallest->sectors - current_start) * c;
		printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
			zone->nb_dev, (unsigned long long)zone->sectors);

		zone->zone_start = curr_zone_start;
		curr_zone_start += zone->sectors;

		current_start = smallest->sectors;
		printk(KERN_INFO "raid0: current zone start: %llu\n",
			(unsigned long long)current_start);
	}

	/* Now find appropriate hash spacing.
	 * We want a number which causes most hash entries to cover
	 * at most two strips, but the hash table must be at most
	 * 1 PAGE.  We choose the smallest strip, or contiguous collection
	 * of strips, that has big enough size.  We never consider the last
	 * strip though as it's size has no bearing on the efficacy of the hash
	 * table.
	 */
	conf->spacing = curr_zone_start;
	min_spacing = curr_zone_start;
	sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
	for (i=0; i < conf->nr_strip_zones-1; i++) {
		sector_t s = 0;
		for (j = i; j < conf->nr_strip_zones - 1 &&
				s < min_spacing; j++)
			s += conf->strip_zone[j].sectors;
		if (s >= min_spacing && s < conf->spacing)
			conf->spacing = s;
	}

	mddev->queue->unplug_fn = raid0_unplug;

	mddev->queue->backing_dev_info.congested_fn = raid0_congested;
	mddev->queue->backing_dev_info.congested_data = mddev;

	printk(KERN_INFO "raid0: done.\n");
	return 0;
 abort:
	return 1;
}

/**
 *	raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 *	@q: request queue
 *	@bvm: properties of new bio
 *	@biovec: the request that could be merged to it.
 *
 *	Return amount of bytes we can accept at this offset
 */
static int raid0_mergeable_bvec(struct request_queue *q,
				struct bvec_merge_data *bvm,
				struct bio_vec *biovec)
{
	mddev_t *mddev = q->queuedata;
	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
	int max;
	unsigned int chunk_sectors = mddev->chunk_size >> 9;
	unsigned int bio_sectors = bvm->bi_size >> 9;

	max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
	if (max < 0) max = 0; /* bio_add cannot handle a negative return */
	if (max <= biovec->bv_len && bio_sectors == 0)
		return biovec->bv_len;
	else 
		return max;
}

static int raid0_run (mddev_t *mddev)
{
	unsigned  cur=0, i=0, nb_zone;
	s64 sectors;
	raid0_conf_t *conf;
	mdk_rdev_t *rdev;

	if (mddev->chunk_size == 0) {
		printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
		return -EINVAL;
	}
	printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
	       mdname(mddev),
	       mddev->chunk_size >> 9,
	       (mddev->chunk_size>>1)-1);
	blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
	blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
	mddev->queue->queue_lock = &mddev->queue->__queue_lock;

	conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
	if (!conf)
		goto out;
	mddev->private = (void *)conf;
 
	conf->strip_zone = NULL;
	conf->devlist = NULL;
	if (create_strip_zones (mddev)) 
		goto out_free_conf;

	/* calculate array device size */
	mddev->array_sectors = 0;
	list_for_each_entry(rdev, &mddev->disks, same_set)
		mddev->array_sectors += rdev->sectors;

	printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
		(unsigned long long)mddev->array_sectors);
	printk(KERN_INFO "raid0 : conf->spacing is %llu sectors.\n",
		(unsigned long long)conf->spacing);
	{
		sector_t s = mddev->array_sectors;
		sector_t space = conf->spacing;
		int round;
		conf->sector_shift = 0;
		if (sizeof(sector_t) > sizeof(u32)) {
			/*shift down space and s so that sector_div will work */
			while (space > (sector_t) (~(u32)0)) {
				s >>= 1;
				space >>= 1;
				s += 1; /* force round-up */
				conf->sector_shift++;
			}
		}
		round = sector_div(s, (u32)space) ? 1 : 0;
		nb_zone = s + round;
	}
	printk(KERN_INFO "raid0 : nb_zone is %d.\n", nb_zone);

	printk(KERN_INFO "raid0 : Allocating %zu bytes for hash.\n",
				nb_zone*sizeof(struct strip_zone*));
	conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
	if (!conf->hash_table)
		goto out_free_conf;
	sectors = conf->strip_zone[cur].sectors;

	conf->hash_table[0] = conf->strip_zone + cur;
	for (i=1; i< nb_zone; i++) {
		while (sectors <= conf->spacing) {
			cur++;
			sectors += conf->strip_zone[cur].sectors;
		}
		sectors -= conf->spacing;
		conf->hash_table[i] = conf->strip_zone + cur;
	}
	if (conf->sector_shift) {
		conf->spacing >>= conf->sector_shift;
		/* round spacing up so when we divide by it, we
		 * err on the side of too-low, which is safest
		 */
		conf->spacing++;
	}

	/* calculate the max read-ahead size.
	 * For read-ahead of large files to be effective, we need to
	 * readahead at least twice a whole stripe. i.e. number of devices
	 * multiplied by chunk size times 2.
	 * If an individual device has an ra_pages greater than the
	 * chunk size, then we will not drive that device as hard as it
	 * wants.  We consider this a configuration error: a larger
	 * chunksize should be used in that case.
	 */
	{
		int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
		if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
			mddev->queue->backing_dev_info.ra_pages = 2* stripe;
	}


	blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
	return 0;

out_free_conf:
	kfree(conf->strip_zone);
	kfree(conf->devlist);
	kfree(conf);
	mddev->private = NULL;
out:
	return -ENOMEM;
}

static int raid0_stop (mddev_t *mddev)
{
	raid0_conf_t *conf = mddev_to_conf(mddev);

	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
	kfree(conf->hash_table);
	conf->hash_table = NULL;
	kfree(conf->strip_zone);
	conf->strip_zone = NULL;
	kfree(conf);
	mddev->private = NULL;

	return 0;
}

static int raid0_make_request (struct request_queue *q, struct bio *bio)
{
	mddev_t *mddev = q->queuedata;
	unsigned int sect_in_chunk, chunksect_bits, chunk_sects;
	raid0_conf_t *conf = mddev_to_conf(mddev);
	struct strip_zone *zone;
	mdk_rdev_t *tmp_dev;
	sector_t chunk;
	sector_t sector, rsect;
	const int rw = bio_data_dir(bio);
	int cpu;

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

	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();

	chunk_sects = mddev->chunk_size >> 9;
	chunksect_bits = ffz(~chunk_sects);
	sector = bio->bi_sector;

	if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
		struct bio_pair *bp;
		/* Sanity check -- queue functions should prevent this happening */
		if (bio->bi_vcnt != 1 ||
		    bio->bi_idx != 0)
			goto bad_map;
		/* This is a one page bio that upper layers
		 * refuse to split for us, so we need to split it.
		 */
		bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
		if (raid0_make_request(q, &bp->bio1))
			generic_make_request(&bp->bio1);
		if (raid0_make_request(q, &bp->bio2))
			generic_make_request(&bp->bio2);

		bio_pair_release(bp);
		return 0;
	}
 

	{
		sector_t x = sector >> conf->sector_shift;
		sector_div(x, (u32)conf->spacing);
		zone = conf->hash_table[x];
	}

	while (sector >= zone->zone_start + zone->sectors)
		zone++;

	sect_in_chunk = bio->bi_sector & (chunk_sects - 1);


	{
		sector_t x = (sector - zone->zone_start) >> chunksect_bits;

		sector_div(x, zone->nb_dev);
		chunk = x;

		x = sector >> chunksect_bits;
		tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
	}
	rsect = (chunk << chunksect_bits) + zone->dev_start + sect_in_chunk;
 
	bio->bi_bdev = tmp_dev->bdev;
	bio->bi_sector = rsect + tmp_dev->data_offset;

	/*
	 * Let the main block layer submit the IO and resolve recursion:
	 */
	return 1;

bad_map:
	printk("raid0_make_request bug: can't convert block across chunks"
		" or bigger than %dk %llu %d\n", chunk_sects / 2,
		(unsigned long long)bio->bi_sector, bio->bi_size >> 10);

	bio_io_error(bio);
	return 0;
}

static void raid0_status (struct seq_file *seq, mddev_t *mddev)
{
#undef MD_DEBUG
#ifdef MD_DEBUG
	int j, k, h;
	char b[BDEVNAME_SIZE];
	raid0_conf_t *conf = mddev_to_conf(mddev);

	h = 0;
	for (j = 0; j < conf->nr_strip_zones; j++) {
		seq_printf(seq, "      z%d", j);
		if (conf->hash_table[h] == conf->strip_zone+j)
			seq_printf(seq, "(h%d)", h++);
		seq_printf(seq, "=[");
		for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
			seq_printf(seq, "%s/", bdevname(
				conf->strip_zone[j].dev[k]->bdev,b));

		seq_printf(seq, "] zs=%d ds=%d s=%d\n",
				conf->strip_zone[j].zone_start,
				conf->strip_zone[j].dev_start,
				conf->strip_zone[j].sectors);
	}
#endif
	seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
	return;
}

static struct mdk_personality raid0_personality=
{
	.name		= "raid0",
	.level		= 0,
	.owner		= THIS_MODULE,
	.make_request	= raid0_make_request,
	.run		= raid0_run,
	.stop		= raid0_stop,
	.status		= raid0_status,
};

static int __init raid0_init (void)
{
	return register_md_personality (&raid0_personality);
}

static void raid0_exit (void)
{
	unregister_md_personality (&raid0_personality);
}

module_init(raid0_init);
module_exit(raid0_exit);
MODULE_LICENSE("GPL");
MODULE_ALIAS("md-personality-2"); /* RAID0 */
MODULE_ALIAS("md-raid0");
MODULE_ALIAS("md-level-0");
Commit	Line	Data
1da177e4 LT	1	/*
	2	raid0.c : Multiple Devices driver for Linux
	3	Copyright (C) 1994-96 Marc ZYNGIER
	4	<zyngier@ufr-info-p7.ibp.fr> or
	5	<maz@gloups.fdn.fr>
	6	Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
	7
	8
	9	RAID-0 management functions.
	10
	11	This program is free software; you can redistribute it and/or modify
	12	it under the terms of the GNU General Public License as published by
	13	the Free Software Foundation; either version 2, or (at your option)
	14	any later version.
	15
	16	You should have received a copy of the GNU General Public License
	17	(for example /usr/src/linux/COPYING); if not, write to the Free
	18	Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
	19	*/
	20
bff61975	21	#include <linux/blkdev.h>
bff61975	22	#include <linux/seq_file.h>
43b2e5d8	23	#include "md.h"
ef740c37	24	#include "raid0.h"
1da177e4	25
165125e1	26	static void raid0_unplug(struct request_queue *q)
1da177e4 LT	27	{
	28	mddev_t *mddev = q->queuedata;
	29	raid0_conf_t *conf = mddev_to_conf(mddev);
	30	mdk_rdev_t **devlist = conf->strip_zone[0].dev;
	31	int i;
	32
	33	for (i=0; i<mddev->raid_disks; i++) {
165125e1	34	struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
1da177e4	35
2ad8b1ef	36	blk_unplug(r_queue);
1da177e4 LT	37	}
	38	}
	39
26be34dc N	40	static int raid0_congested(void *data, int bits)
	41	{
	42	mddev_t *mddev = data;
	43	raid0_conf_t *conf = mddev_to_conf(mddev);
	44	mdk_rdev_t **devlist = conf->strip_zone[0].dev;
	45	int i, ret = 0;
	46
	47	for (i = 0; i < mddev->raid_disks && !ret ; i++) {
165125e1	48	struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
26be34dc N	49
	50	ret \|= bdi_congested(&q->backing_dev_info, bits);
	51	}
	52	return ret;
	53	}
	54
1da177e4 LT	55
	56	static int create_strip_zones (mddev_t *mddev)
	57	{
	58	int i, c, j;
6b8796cc	59	sector_t current_start, curr_zone_start;
1da177e4 LT	60	sector_t min_spacing;
	61	raid0_conf_t *conf = mddev_to_conf(mddev);
	62	mdk_rdev_t smallest, rdev1, rdev2, rdev;
1da177e4 LT	63	struct strip_zone *zone;
	64	int cnt;
	65	char b[BDEVNAME_SIZE];
	66
	67	/*
	68	* The number of 'same size groups'
	69	*/
	70	conf->nr_strip_zones = 0;
	71
159ec1fc	72	list_for_each_entry(rdev1, &mddev->disks, same_set) {
0825b87a	73	printk(KERN_INFO "raid0: looking at %s\n",
1da177e4 LT	74	bdevname(rdev1->bdev,b));
1da177e4 LT	75	c = 0;
159ec1fc	76	list_for_each_entry(rdev2, &mddev->disks, same_set) {
0825b87a	77	printk(KERN_INFO "raid0: comparing %s(%llu)",
1da177e4	78	bdevname(rdev1->bdev,b),
dd8ac336	79	(unsigned long long)rdev1->sectors);
0825b87a	80	printk(KERN_INFO " with %s(%llu)\n",
1da177e4	81	bdevname(rdev2->bdev,b),
dd8ac336	82	(unsigned long long)rdev2->sectors);
1da177e4	83	if (rdev2 == rdev1) {
0825b87a	84	printk(KERN_INFO "raid0: END\n");
1da177e4 LT	85	break;
1da177e4 LT	86	}
dd8ac336	87	if (rdev2->sectors == rdev1->sectors) {
1da177e4 LT	88	/*
	89	* Not unique, don't count it as a new
	90	* group
	91	*/
0825b87a	92	printk(KERN_INFO "raid0: EQUAL\n");
1da177e4 LT	93	c = 1;
	94	break;
	95	}
0825b87a	96	printk(KERN_INFO "raid0: NOT EQUAL\n");
1da177e4 LT	97	}
1da177e4 LT	98	if (!c) {
0825b87a	99	printk(KERN_INFO "raid0: ==> UNIQUE\n");
1da177e4	100	conf->nr_strip_zones++;
0825b87a AN	101	printk(KERN_INFO "raid0: %d zones\n",
0825b87a AN	102	conf->nr_strip_zones);
1da177e4 LT	103	}
1da177e4 LT	104	}
0825b87a	105	printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
1da177e4	106
9ffae0cf	107	conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
1da177e4 LT	108	conf->nr_strip_zones, GFP_KERNEL);
	109	if (!conf->strip_zone)
	110	return 1;
9ffae0cf	111	conf->devlist = kzalloc(sizeof(mdk_rdev_t)
1da177e4 LT	112	conf->nr_strip_zones*mddev->raid_disks,
	113	GFP_KERNEL);
	114	if (!conf->devlist)
	115	return 1;
	116
1da177e4 LT	117	/* The first zone must contain all devices, so here we check that
	118	* there is a proper alignment of slots to devices and find them all
	119	*/
	120	zone = &conf->strip_zone[0];
	121	cnt = 0;
	122	smallest = NULL;
	123	zone->dev = conf->devlist;
159ec1fc	124	list_for_each_entry(rdev1, &mddev->disks, same_set) {
1da177e4 LT	125	int j = rdev1->raid_disk;
	126
	127	if (j < 0 \|\| j >= mddev->raid_disks) {
0825b87a AN	128	printk(KERN_ERR "raid0: bad disk number %d - "
0825b87a AN	129	"aborting!\n", j);
1da177e4 LT	130	goto abort;
	131	}
	132	if (zone->dev[j]) {
0825b87a AN	133	printk(KERN_ERR "raid0: multiple devices for %d - "
0825b87a AN	134	"aborting!\n", j);
1da177e4 LT	135	goto abort;
	136	}
	137	zone->dev[j] = rdev1;
	138
	139	blk_queue_stack_limits(mddev->queue,
	140	rdev1->bdev->bd_disk->queue);
	141	/* as we don't honour merge_bvec_fn, we must never risk
	142	* violating it, so limit ->max_sector to one PAGE, as
	143	* a one page request is never in violation.
	144	*/
	145
	146	if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
	147	mddev->queue->max_sectors > (PAGE_SIZE>>9))
	148	blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
	149
dd8ac336	150	if (!smallest \|\| (rdev1->sectors < smallest->sectors))
1da177e4 LT	151	smallest = rdev1;
	152	cnt++;
	153	}
	154	if (cnt != mddev->raid_disks) {
0825b87a AN	155	printk(KERN_ERR "raid0: too few disks (%d of %d) - "
0825b87a AN	156	"aborting!\n", cnt, mddev->raid_disks);
1da177e4 LT	157	goto abort;
	158	}
	159	zone->nb_dev = cnt;
dd8ac336	160	zone->sectors = smallest->sectors * cnt;
6199d3db	161	zone->zone_start = 0;
1da177e4	162
dd8ac336	163	current_start = smallest->sectors;
83838ed8	164	curr_zone_start = zone->sectors;
1da177e4 LT	165
	166	/* now do the other zones */
	167	for (i = 1; i < conf->nr_strip_zones; i++)
	168	{
	169	zone = conf->strip_zone + i;
	170	zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
	171
0825b87a	172	printk(KERN_INFO "raid0: zone %d\n", i);
6b8796cc	173	zone->dev_start = current_start;
1da177e4 LT	174	smallest = NULL;
	175	c = 0;
	176
	177	for (j=0; j<cnt; j++) {
	178	char b[BDEVNAME_SIZE];
	179	rdev = conf->strip_zone[0].dev[j];
0825b87a AN	180	printk(KERN_INFO "raid0: checking %s ...",
0825b87a AN	181	bdevname(rdev->bdev, b));
dd8ac336	182	if (rdev->sectors <= current_start) {
0825b87a	183	printk(KERN_INFO " nope.\n");
dd8ac336 AN	184	continue;
	185	}
	186	printk(KERN_INFO " contained as device %d\n", c);
	187	zone->dev[c] = rdev;
	188	c++;
	189	if (!smallest \|\| rdev->sectors < smallest->sectors) {
	190	smallest = rdev;
	191	printk(KERN_INFO " (%llu) is smallest!.\n",
	192	(unsigned long long)rdev->sectors);
	193	}
1da177e4 LT	194	}
	195
	196	zone->nb_dev = c;
dd8ac336	197	zone->sectors = (smallest->sectors - current_start) * c;
83838ed8 AN	198	printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
83838ed8 AN	199	zone->nb_dev, (unsigned long long)zone->sectors);
1da177e4	200
6b8796cc	201	zone->zone_start = curr_zone_start;
83838ed8	202	curr_zone_start += zone->sectors;
1da177e4	203
dd8ac336	204	current_start = smallest->sectors;
6b8796cc AN	205	printk(KERN_INFO "raid0: current zone start: %llu\n",
6b8796cc AN	206	(unsigned long long)current_start);
1da177e4 LT	207	}
	208
	209	/* Now find appropriate hash spacing.
	210	* We want a number which causes most hash entries to cover
	211	* at most two strips, but the hash table must be at most
	212	* 1 PAGE. We choose the smallest strip, or contiguous collection
	213	* of strips, that has big enough size. We never consider the last
	214	* strip though as it's size has no bearing on the efficacy of the hash
	215	* table.
	216	*/
ccacc7d2 AN	217	conf->spacing = curr_zone_start;
ccacc7d2 AN	218	min_spacing = curr_zone_start;
1da177e4 LT	219	sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
1da177e4 LT	220	for (i=0; i < conf->nr_strip_zones-1; i++) {
ccacc7d2 AN	221	sector_t s = 0;
	222	for (j = i; j < conf->nr_strip_zones - 1 &&
	223	s < min_spacing; j++)
	224	s += conf->strip_zone[j].sectors;
	225	if (s >= min_spacing && s < conf->spacing)
	226	conf->spacing = s;
1da177e4 LT	227	}
	228
	229	mddev->queue->unplug_fn = raid0_unplug;
	230
26be34dc N	231	mddev->queue->backing_dev_info.congested_fn = raid0_congested;
26be34dc N	232	mddev->queue->backing_dev_info.congested_data = mddev;
1da177e4	233
0825b87a	234	printk(KERN_INFO "raid0: done.\n");
1da177e4 LT	235	return 0;
	236	abort:
	237	return 1;
	238	}
	239
	240	/**
	241	* raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
	242	* @q: request queue
cc371e66	243	* @bvm: properties of new bio
1da177e4 LT	244	* @biovec: the request that could be merged to it.
	245	*
	246	* Return amount of bytes we can accept at this offset
	247	*/
cc371e66 AK	248	static int raid0_mergeable_bvec(struct request_queue *q,
	249	struct bvec_merge_data *bvm,
	250	struct bio_vec *biovec)
1da177e4 LT	251	{
1da177e4 LT	252	mddev_t *mddev = q->queuedata;
cc371e66	253	sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
1da177e4 LT	254	int max;
1da177e4 LT	255	unsigned int chunk_sectors = mddev->chunk_size >> 9;
cc371e66	256	unsigned int bio_sectors = bvm->bi_size >> 9;
1da177e4 LT	257
	258	max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
	259	if (max < 0) max = 0; /* bio_add cannot handle a negative return */
	260	if (max <= biovec->bv_len && bio_sectors == 0)
	261	return biovec->bv_len;
	262	else
	263	return max;
	264	}
	265
	266	static int raid0_run (mddev_t *mddev)
	267	{
	268	unsigned cur=0, i=0, nb_zone;
ccacc7d2	269	s64 sectors;
1da177e4 LT	270	raid0_conf_t *conf;
1da177e4 LT	271	mdk_rdev_t *rdev;
1da177e4	272
2604b703 N	273	if (mddev->chunk_size == 0) {
	274	printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
	275	return -EINVAL;
	276	}
	277	printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
1da177e4 LT	278	mdname(mddev),
	279	mddev->chunk_size >> 9,
	280	(mddev->chunk_size>>1)-1);
	281	blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
	282	blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
e7e72bf6	283	mddev->queue->queue_lock = &mddev->queue->__queue_lock;
1da177e4 LT	284
	285	conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
	286	if (!conf)
	287	goto out;
	288	mddev->private = (void *)conf;
	289
	290	conf->strip_zone = NULL;
	291	conf->devlist = NULL;
	292	if (create_strip_zones (mddev))
	293	goto out_free_conf;
	294
	295	/* calculate array device size */
f233ea5c	296	mddev->array_sectors = 0;
159ec1fc	297	list_for_each_entry(rdev, &mddev->disks, same_set)
dd8ac336	298	mddev->array_sectors += rdev->sectors;
1da177e4	299
ccacc7d2 AN	300	printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
	301	(unsigned long long)mddev->array_sectors);
	302	printk(KERN_INFO "raid0 : conf->spacing is %llu sectors.\n",
	303	(unsigned long long)conf->spacing);
1da177e4	304	{
ccacc7d2 AN	305	sector_t s = mddev->array_sectors;
ccacc7d2 AN	306	sector_t space = conf->spacing;
1da177e4	307	int round;
ccacc7d2	308	conf->sector_shift = 0;
1eb29128	309	if (sizeof(sector_t) > sizeof(u32)) {
1da177e4	310	/shift down space and s so that sector_div will work /
1eb29128	311	while (space > (sector_t) (~(u32)0)) {
1da177e4 LT	312	s >>= 1;
	313	space >>= 1;
	314	s += 1; /* force round-up */
ccacc7d2	315	conf->sector_shift++;
1da177e4 LT	316	}
1da177e4 LT	317	}
1eb29128	318	round = sector_div(s, (u32)space) ? 1 : 0;
1da177e4 LT	319	nb_zone = s + round;
1da177e4 LT	320	}
ccacc7d2	321	printk(KERN_INFO "raid0 : nb_zone is %d.\n", nb_zone);
1da177e4	322
ccacc7d2	323	printk(KERN_INFO "raid0 : Allocating %zu bytes for hash.\n",
1da177e4 LT	324	nb_zonesizeof(struct strip_zone));
	325	conf->hash_table = kmalloc (sizeof (struct strip_zone )nb_zone, GFP_KERNEL);
	326	if (!conf->hash_table)
	327	goto out_free_conf;
ccacc7d2	328	sectors = conf->strip_zone[cur].sectors;
1da177e4	329
5c4c3331 N	330	conf->hash_table[0] = conf->strip_zone + cur;
5c4c3331 N	331	for (i=1; i< nb_zone; i++) {
ccacc7d2	332	while (sectors <= conf->spacing) {
1da177e4	333	cur++;
ccacc7d2	334	sectors += conf->strip_zone[cur].sectors;
1da177e4	335	}
ccacc7d2	336	sectors -= conf->spacing;
5c4c3331	337	conf->hash_table[i] = conf->strip_zone + cur;
1da177e4	338	}
ccacc7d2 AN	339	if (conf->sector_shift) {
	340	conf->spacing >>= conf->sector_shift;
	341	/* round spacing up so when we divide by it, we
1da177e4 LT	342	* err on the side of too-low, which is safest
1da177e4 LT	343	*/
ccacc7d2	344	conf->spacing++;
1da177e4 LT	345	}
	346
	347	/* calculate the max read-ahead size.
	348	* For read-ahead of large files to be effective, we need to
	349	* readahead at least twice a whole stripe. i.e. number of devices
	350	* multiplied by chunk size times 2.
	351	* If an individual device has an ra_pages greater than the
	352	* chunk size, then we will not drive that device as hard as it
	353	* wants. We consider this a configuration error: a larger
	354	* chunksize should be used in that case.
	355	*/
	356	{
2d1f3b5d	357	int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
1da177e4 LT	358	if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
	359	mddev->queue->backing_dev_info.ra_pages = 2* stripe;
	360	}
	361
	362
	363	blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
	364	return 0;
	365
	366	out_free_conf:
990a8baf JJ	367	kfree(conf->strip_zone);
990a8baf JJ	368	kfree(conf->devlist);
1da177e4 LT	369	kfree(conf);
	370	mddev->private = NULL;
	371	out:
29fc7e3e	372	return -ENOMEM;
1da177e4 LT	373	}
	374
	375	static int raid0_stop (mddev_t *mddev)
	376	{
	377	raid0_conf_t *conf = mddev_to_conf(mddev);
	378
	379	blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
990a8baf	380	kfree(conf->hash_table);
1da177e4	381	conf->hash_table = NULL;
990a8baf	382	kfree(conf->strip_zone);
1da177e4	383	conf->strip_zone = NULL;
990a8baf	384	kfree(conf);
1da177e4 LT	385	mddev->private = NULL;
	386
	387	return 0;
	388	}
	389
165125e1	390	static int raid0_make_request (struct request_queue q, struct bio bio)
1da177e4 LT	391	{
1da177e4 LT	392	mddev_t *mddev = q->queuedata;
a4712005	393	unsigned int sect_in_chunk, chunksect_bits, chunk_sects;
1da177e4 LT	394	raid0_conf_t *conf = mddev_to_conf(mddev);
	395	struct strip_zone *zone;
	396	mdk_rdev_t *tmp_dev;
787f17fe	397	sector_t chunk;
e0f06868	398	sector_t sector, rsect;
a362357b	399	const int rw = bio_data_dir(bio);
c9959059	400	int cpu;
1da177e4	401
e5dcdd80	402	if (unlikely(bio_barrier(bio))) {
6712ecf8	403	bio_endio(bio, -EOPNOTSUPP);
e5dcdd80 N	404	return 0;
	405	}
	406
074a7aca TH	407	cpu = part_stat_lock();
	408	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
	409	part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
	410	bio_sectors(bio));
	411	part_stat_unlock();
1da177e4	412
1da177e4	413	chunk_sects = mddev->chunk_size >> 9;
1b7fdf8f	414	chunksect_bits = ffz(~chunk_sects);
e0f06868	415	sector = bio->bi_sector;
1da177e4 LT	416
	417	if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
	418	struct bio_pair *bp;
	419	/* Sanity check -- queue functions should prevent this happening */
	420	if (bio->bi_vcnt != 1 \|\|
	421	bio->bi_idx != 0)
	422	goto bad_map;
	423	/* This is a one page bio that upper layers
	424	* refuse to split for us, so we need to split it.
	425	*/
6feef531	426	bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
1da177e4 LT	427	if (raid0_make_request(q, &bp->bio1))
	428	generic_make_request(&bp->bio1);
	429	if (raid0_make_request(q, &bp->bio2))
	430	generic_make_request(&bp->bio2);
	431
	432	bio_pair_release(bp);
	433	return 0;
	434	}
	435
	436
	437	{
ccacc7d2 AN	438	sector_t x = sector >> conf->sector_shift;
ccacc7d2 AN	439	sector_div(x, (u32)conf->spacing);
1da177e4 LT	440	zone = conf->hash_table[x];
1da177e4 LT	441	}
83838ed8 AN	442
83838ed8 AN	443	while (sector >= zone->zone_start + zone->sectors)
1da177e4	444	zone++;
83838ed8	445
a4712005	446	sect_in_chunk = bio->bi_sector & (chunk_sects - 1);
1da177e4 LT	447
	448
	449	{
6199d3db	450	sector_t x = (sector - zone->zone_start) >> chunksect_bits;
1da177e4 LT	451
	452	sector_div(x, zone->nb_dev);
	453	chunk = x;
1da177e4	454
e0f06868	455	x = sector >> chunksect_bits;
1da177e4 LT	456	tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
1da177e4 LT	457	}
019c4e2f	458	rsect = (chunk << chunksect_bits) + zone->dev_start + sect_in_chunk;
1da177e4 LT	459
	460	bio->bi_bdev = tmp_dev->bdev;
	461	bio->bi_sector = rsect + tmp_dev->data_offset;
	462
	463	/*
	464	* Let the main block layer submit the IO and resolve recursion:
	465	*/
	466	return 1;
	467
	468	bad_map:
	469	printk("raid0_make_request bug: can't convert block across chunks"
a4712005	470	" or bigger than %dk %llu %d\n", chunk_sects / 2,
1da177e4 LT	471	(unsigned long long)bio->bi_sector, bio->bi_size >> 10);
1da177e4 LT	472
6712ecf8	473	bio_io_error(bio);
1da177e4 LT	474	return 0;
1da177e4 LT	475	}
8299d7f7	476
1da177e4 LT	477	static void raid0_status (struct seq_file seq, mddev_t mddev)
	478	{
	479	#undef MD_DEBUG
	480	#ifdef MD_DEBUG
	481	int j, k, h;
	482	char b[BDEVNAME_SIZE];
	483	raid0_conf_t *conf = mddev_to_conf(mddev);
8299d7f7	484
1da177e4 LT	485	h = 0;
	486	for (j = 0; j < conf->nr_strip_zones; j++) {
	487	seq_printf(seq, " z%d", j);
	488	if (conf->hash_table[h] == conf->strip_zone+j)
8299d7f7	489	seq_printf(seq, "(h%d)", h++);
1da177e4 LT	490	seq_printf(seq, "=[");
1da177e4 LT	491	for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
8299d7f7	492	seq_printf(seq, "%s/", bdevname(
1da177e4 LT	493	conf->strip_zone[j].dev[k]->bdev,b));
1da177e4 LT	494
6199d3db AN	495	seq_printf(seq, "] zs=%d ds=%d s=%d\n",
6199d3db AN	496	conf->strip_zone[j].zone_start,
019c4e2f	497	conf->strip_zone[j].dev_start,
83838ed8	498	conf->strip_zone[j].sectors);
1da177e4 LT	499	}
	500	#endif
	501	seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
	502	return;
	503	}
	504
2604b703	505	static struct mdk_personality raid0_personality=
1da177e4 LT	506	{
1da177e4 LT	507	.name = "raid0",
2604b703	508	.level = 0,
1da177e4 LT	509	.owner = THIS_MODULE,
	510	.make_request = raid0_make_request,
	511	.run = raid0_run,
	512	.stop = raid0_stop,
	513	.status = raid0_status,
	514	};
	515
	516	static int __init raid0_init (void)
	517	{
2604b703	518	return register_md_personality (&raid0_personality);
1da177e4 LT	519	}
	520
	521	static void raid0_exit (void)
	522	{
2604b703	523	unregister_md_personality (&raid0_personality);
1da177e4 LT	524	}
	525
	526	module_init(raid0_init);
	527	module_exit(raid0_exit);
	528	MODULE_LICENSE("GPL");
	529	MODULE_ALIAS("md-personality-2"); /* RAID0 */
d9d166c2	530	MODULE_ALIAS("md-raid0");
2604b703	531	MODULE_ALIAS("md-level-0");