2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
9 RAID-0 management functions.
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)
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.
21 #include <linux/module.h>
22 #include <linux/raid/raid0.h>
24 #define MAJOR_NR MD_MAJOR
26 #define MD_PERSONALITY
28 static void raid0_unplug(struct request_queue *q)
30 mddev_t *mddev = q->queuedata;
31 raid0_conf_t *conf = mddev_to_conf(mddev);
32 mdk_rdev_t **devlist = conf->strip_zone[0].dev;
35 for (i=0; i<mddev->raid_disks; i++) {
36 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
38 if (r_queue->unplug_fn)
39 r_queue->unplug_fn(r_queue);
43 static int raid0_congested(void *data, int bits)
45 mddev_t *mddev = data;
46 raid0_conf_t *conf = mddev_to_conf(mddev);
47 mdk_rdev_t **devlist = conf->strip_zone[0].dev;
50 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
51 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
53 ret |= bdi_congested(&q->backing_dev_info, bits);
59 static int create_strip_zones (mddev_t *mddev)
62 sector_t current_offset, curr_zone_offset;
64 raid0_conf_t *conf = mddev_to_conf(mddev);
65 mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
66 struct list_head *tmp1, *tmp2;
67 struct strip_zone *zone;
69 char b[BDEVNAME_SIZE];
72 * The number of 'same size groups'
74 conf->nr_strip_zones = 0;
76 ITERATE_RDEV(mddev,rdev1,tmp1) {
77 printk("raid0: looking at %s\n",
78 bdevname(rdev1->bdev,b));
80 ITERATE_RDEV(mddev,rdev2,tmp2) {
81 printk("raid0: comparing %s(%llu)",
82 bdevname(rdev1->bdev,b),
83 (unsigned long long)rdev1->size);
84 printk(" with %s(%llu)\n",
85 bdevname(rdev2->bdev,b),
86 (unsigned long long)rdev2->size);
88 printk("raid0: END\n");
91 if (rdev2->size == rdev1->size)
94 * Not unique, don't count it as a new
97 printk("raid0: EQUAL\n");
101 printk("raid0: NOT EQUAL\n");
104 printk("raid0: ==> UNIQUE\n");
105 conf->nr_strip_zones++;
106 printk("raid0: %d zones\n", conf->nr_strip_zones);
109 printk("raid0: FINAL %d zones\n", conf->nr_strip_zones);
111 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
112 conf->nr_strip_zones, GFP_KERNEL);
113 if (!conf->strip_zone)
115 conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
116 conf->nr_strip_zones*mddev->raid_disks,
121 /* The first zone must contain all devices, so here we check that
122 * there is a proper alignment of slots to devices and find them all
124 zone = &conf->strip_zone[0];
127 zone->dev = conf->devlist;
128 ITERATE_RDEV(mddev, rdev1, tmp1) {
129 int j = rdev1->raid_disk;
131 if (j < 0 || j >= mddev->raid_disks) {
132 printk("raid0: bad disk number %d - aborting!\n", j);
136 printk("raid0: multiple devices for %d - aborting!\n",
140 zone->dev[j] = rdev1;
142 blk_queue_stack_limits(mddev->queue,
143 rdev1->bdev->bd_disk->queue);
144 /* as we don't honour merge_bvec_fn, we must never risk
145 * violating it, so limit ->max_sector to one PAGE, as
146 * a one page request is never in violation.
149 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
150 mddev->queue->max_sectors > (PAGE_SIZE>>9))
151 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
153 if (!smallest || (rdev1->size <smallest->size))
157 if (cnt != mddev->raid_disks) {
158 printk("raid0: too few disks (%d of %d) - aborting!\n",
159 cnt, mddev->raid_disks);
163 zone->size = smallest->size * cnt;
164 zone->zone_offset = 0;
166 current_offset = smallest->size;
167 curr_zone_offset = zone->size;
169 /* now do the other zones */
170 for (i = 1; i < conf->nr_strip_zones; i++)
172 zone = conf->strip_zone + i;
173 zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
175 printk("raid0: zone %d\n", i);
176 zone->dev_offset = current_offset;
180 for (j=0; j<cnt; j++) {
181 char b[BDEVNAME_SIZE];
182 rdev = conf->strip_zone[0].dev[j];
183 printk("raid0: checking %s ...", bdevname(rdev->bdev,b));
184 if (rdev->size > current_offset)
186 printk(" contained as device %d\n", c);
189 if (!smallest || (rdev->size <smallest->size)) {
191 printk(" (%llu) is smallest!.\n",
192 (unsigned long long)rdev->size);
199 zone->size = (smallest->size - current_offset) * c;
200 printk("raid0: zone->nb_dev: %d, size: %llu\n",
201 zone->nb_dev, (unsigned long long)zone->size);
203 zone->zone_offset = curr_zone_offset;
204 curr_zone_offset += zone->size;
206 current_offset = smallest->size;
207 printk("raid0: current zone offset: %llu\n",
208 (unsigned long long)current_offset);
211 /* Now find appropriate hash spacing.
212 * We want a number which causes most hash entries to cover
213 * at most two strips, but the hash table must be at most
214 * 1 PAGE. We choose the smallest strip, or contiguous collection
215 * of strips, that has big enough size. We never consider the last
216 * strip though as it's size has no bearing on the efficacy of the hash
219 conf->hash_spacing = curr_zone_offset;
220 min_spacing = curr_zone_offset;
221 sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
222 for (i=0; i < conf->nr_strip_zones-1; i++) {
224 for (j=i; j<conf->nr_strip_zones-1 &&
225 sz < min_spacing ; j++)
226 sz += conf->strip_zone[j].size;
227 if (sz >= min_spacing && sz < conf->hash_spacing)
228 conf->hash_spacing = sz;
231 mddev->queue->unplug_fn = raid0_unplug;
233 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
234 mddev->queue->backing_dev_info.congested_data = mddev;
236 printk("raid0: done.\n");
243 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
245 * @bio: the buffer head that's been built up so far
246 * @biovec: the request that could be merged to it.
248 * Return amount of bytes we can accept at this offset
250 static int raid0_mergeable_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *biovec)
252 mddev_t *mddev = q->queuedata;
253 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
255 unsigned int chunk_sectors = mddev->chunk_size >> 9;
256 unsigned int bio_sectors = bio->bi_size >> 9;
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;
266 static int raid0_run (mddev_t *mddev)
268 unsigned cur=0, i=0, nb_zone;
272 struct list_head *tmp;
274 if (mddev->chunk_size == 0) {
275 printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
278 printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
280 mddev->chunk_size >> 9,
281 (mddev->chunk_size>>1)-1);
282 blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
283 blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
285 conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
288 mddev->private = (void *)conf;
290 conf->strip_zone = NULL;
291 conf->devlist = NULL;
292 if (create_strip_zones (mddev))
295 /* calculate array device size */
296 mddev->array_size = 0;
297 ITERATE_RDEV(mddev,rdev,tmp)
298 mddev->array_size += rdev->size;
300 printk("raid0 : md_size is %llu blocks.\n",
301 (unsigned long long)mddev->array_size);
302 printk("raid0 : conf->hash_spacing is %llu blocks.\n",
303 (unsigned long long)conf->hash_spacing);
305 sector_t s = mddev->array_size;
306 sector_t space = conf->hash_spacing;
309 if (sizeof(sector_t) > sizeof(u32)) {
310 /*shift down space and s so that sector_div will work */
311 while (space > (sector_t) (~(u32)0)) {
314 s += 1; /* force round-up */
318 round = sector_div(s, (u32)space) ? 1 : 0;
321 printk("raid0 : nb_zone is %d.\n", nb_zone);
323 printk("raid0 : Allocating %Zd bytes for hash.\n",
324 nb_zone*sizeof(struct strip_zone*));
325 conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
326 if (!conf->hash_table)
328 size = conf->strip_zone[cur].size;
330 conf->hash_table[0] = conf->strip_zone + cur;
331 for (i=1; i< nb_zone; i++) {
332 while (size <= conf->hash_spacing) {
334 size += conf->strip_zone[cur].size;
336 size -= conf->hash_spacing;
337 conf->hash_table[i] = conf->strip_zone + cur;
339 if (conf->preshift) {
340 conf->hash_spacing >>= conf->preshift;
341 /* round hash_spacing up so when we divide by it, we
342 * err on the side of too-low, which is safest
344 conf->hash_spacing++;
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.
357 int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
358 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
359 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
363 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
367 kfree(conf->strip_zone);
368 kfree(conf->devlist);
370 mddev->private = NULL;
375 static int raid0_stop (mddev_t *mddev)
377 raid0_conf_t *conf = mddev_to_conf(mddev);
379 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
380 kfree(conf->hash_table);
381 conf->hash_table = NULL;
382 kfree(conf->strip_zone);
383 conf->strip_zone = NULL;
385 mddev->private = NULL;
390 static int raid0_make_request (struct request_queue *q, struct bio *bio)
392 mddev_t *mddev = q->queuedata;
393 unsigned int sect_in_chunk, chunksize_bits, chunk_size, chunk_sects;
394 raid0_conf_t *conf = mddev_to_conf(mddev);
395 struct strip_zone *zone;
398 sector_t block, rsect;
399 const int rw = bio_data_dir(bio);
401 if (unlikely(bio_barrier(bio))) {
402 bio_endio(bio, -EOPNOTSUPP);
406 disk_stat_inc(mddev->gendisk, ios[rw]);
407 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
409 chunk_size = mddev->chunk_size >> 10;
410 chunk_sects = mddev->chunk_size >> 9;
411 chunksize_bits = ffz(~chunk_size);
412 block = bio->bi_sector >> 1;
415 if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
417 /* Sanity check -- queue functions should prevent this happening */
418 if (bio->bi_vcnt != 1 ||
421 /* This is a one page bio that upper layers
422 * refuse to split for us, so we need to split it.
424 bp = bio_split(bio, bio_split_pool, chunk_sects - (bio->bi_sector & (chunk_sects - 1)) );
425 if (raid0_make_request(q, &bp->bio1))
426 generic_make_request(&bp->bio1);
427 if (raid0_make_request(q, &bp->bio2))
428 generic_make_request(&bp->bio2);
430 bio_pair_release(bp);
436 sector_t x = block >> conf->preshift;
437 sector_div(x, (u32)conf->hash_spacing);
438 zone = conf->hash_table[x];
441 while (block >= (zone->zone_offset + zone->size))
444 sect_in_chunk = bio->bi_sector & ((chunk_size<<1) -1);
448 sector_t x = (block - zone->zone_offset) >> chunksize_bits;
450 sector_div(x, zone->nb_dev);
453 x = block >> chunksize_bits;
454 tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
456 rsect = (((chunk << chunksize_bits) + zone->dev_offset)<<1)
459 bio->bi_bdev = tmp_dev->bdev;
460 bio->bi_sector = rsect + tmp_dev->data_offset;
463 * Let the main block layer submit the IO and resolve recursion:
468 printk("raid0_make_request bug: can't convert block across chunks"
469 " or bigger than %dk %llu %d\n", chunk_size,
470 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
476 static void raid0_status (struct seq_file *seq, mddev_t *mddev)
481 char b[BDEVNAME_SIZE];
482 raid0_conf_t *conf = mddev_to_conf(mddev);
485 for (j = 0; j < conf->nr_strip_zones; j++) {
486 seq_printf(seq, " z%d", j);
487 if (conf->hash_table[h] == conf->strip_zone+j)
488 seq_printf(seq, "(h%d)", h++);
489 seq_printf(seq, "=[");
490 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
491 seq_printf(seq, "%s/", bdevname(
492 conf->strip_zone[j].dev[k]->bdev,b));
494 seq_printf(seq, "] zo=%d do=%d s=%d\n",
495 conf->strip_zone[j].zone_offset,
496 conf->strip_zone[j].dev_offset,
497 conf->strip_zone[j].size);
500 seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
504 static struct mdk_personality raid0_personality=
508 .owner = THIS_MODULE,
509 .make_request = raid0_make_request,
512 .status = raid0_status,
515 static int __init raid0_init (void)
517 return register_md_personality (&raid0_personality);
520 static void raid0_exit (void)
522 unregister_md_personality (&raid0_personality);
525 module_init(raid0_init);
526 module_exit(raid0_exit);
527 MODULE_LICENSE("GPL");
528 MODULE_ALIAS("md-personality-2"); /* RAID0 */
529 MODULE_ALIAS("md-raid0");
530 MODULE_ALIAS("md-level-0");