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(request_queue_t *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 request_queue_t *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_issue_flush(request_queue_t *q, struct gendisk *disk,
44 sector_t *error_sector)
46 mddev_t *mddev = q->queuedata;
47 raid0_conf_t *conf = mddev_to_conf(mddev);
48 mdk_rdev_t **devlist = conf->strip_zone[0].dev;
51 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
52 struct block_device *bdev = devlist[i]->bdev;
53 request_queue_t *r_queue = bdev_get_queue(bdev);
55 if (!r_queue->issue_flush_fn)
58 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector);
64 static int create_strip_zones (mddev_t *mddev)
67 sector_t current_offset, curr_zone_offset;
69 raid0_conf_t *conf = mddev_to_conf(mddev);
70 mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
71 struct list_head *tmp1, *tmp2;
72 struct strip_zone *zone;
74 char b[BDEVNAME_SIZE];
77 * The number of 'same size groups'
79 conf->nr_strip_zones = 0;
81 ITERATE_RDEV(mddev,rdev1,tmp1) {
82 printk("raid0: looking at %s\n",
83 bdevname(rdev1->bdev,b));
85 ITERATE_RDEV(mddev,rdev2,tmp2) {
86 printk("raid0: comparing %s(%llu)",
87 bdevname(rdev1->bdev,b),
88 (unsigned long long)rdev1->size);
89 printk(" with %s(%llu)\n",
90 bdevname(rdev2->bdev,b),
91 (unsigned long long)rdev2->size);
93 printk("raid0: END\n");
96 if (rdev2->size == rdev1->size)
99 * Not unique, don't count it as a new
102 printk("raid0: EQUAL\n");
106 printk("raid0: NOT EQUAL\n");
109 printk("raid0: ==> UNIQUE\n");
110 conf->nr_strip_zones++;
111 printk("raid0: %d zones\n", conf->nr_strip_zones);
114 printk("raid0: FINAL %d zones\n", conf->nr_strip_zones);
116 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
117 conf->nr_strip_zones, GFP_KERNEL);
118 if (!conf->strip_zone)
120 conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
121 conf->nr_strip_zones*mddev->raid_disks,
126 /* The first zone must contain all devices, so here we check that
127 * there is a proper alignment of slots to devices and find them all
129 zone = &conf->strip_zone[0];
132 zone->dev = conf->devlist;
133 ITERATE_RDEV(mddev, rdev1, tmp1) {
134 int j = rdev1->raid_disk;
136 if (j < 0 || j >= mddev->raid_disks) {
137 printk("raid0: bad disk number %d - aborting!\n", j);
141 printk("raid0: multiple devices for %d - aborting!\n",
145 zone->dev[j] = rdev1;
147 blk_queue_stack_limits(mddev->queue,
148 rdev1->bdev->bd_disk->queue);
149 /* as we don't honour merge_bvec_fn, we must never risk
150 * violating it, so limit ->max_sector to one PAGE, as
151 * a one page request is never in violation.
154 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
155 mddev->queue->max_sectors > (PAGE_SIZE>>9))
156 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
158 if (!smallest || (rdev1->size <smallest->size))
162 if (cnt != mddev->raid_disks) {
163 printk("raid0: too few disks (%d of %d) - aborting!\n",
164 cnt, mddev->raid_disks);
168 zone->size = smallest->size * cnt;
169 zone->zone_offset = 0;
171 current_offset = smallest->size;
172 curr_zone_offset = zone->size;
174 /* now do the other zones */
175 for (i = 1; i < conf->nr_strip_zones; i++)
177 zone = conf->strip_zone + i;
178 zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
180 printk("raid0: zone %d\n", i);
181 zone->dev_offset = current_offset;
185 for (j=0; j<cnt; j++) {
186 char b[BDEVNAME_SIZE];
187 rdev = conf->strip_zone[0].dev[j];
188 printk("raid0: checking %s ...", bdevname(rdev->bdev,b));
189 if (rdev->size > current_offset)
191 printk(" contained as device %d\n", c);
194 if (!smallest || (rdev->size <smallest->size)) {
196 printk(" (%llu) is smallest!.\n",
197 (unsigned long long)rdev->size);
204 zone->size = (smallest->size - current_offset) * c;
205 printk("raid0: zone->nb_dev: %d, size: %llu\n",
206 zone->nb_dev, (unsigned long long)zone->size);
208 zone->zone_offset = curr_zone_offset;
209 curr_zone_offset += zone->size;
211 current_offset = smallest->size;
212 printk("raid0: current zone offset: %llu\n",
213 (unsigned long long)current_offset);
216 /* Now find appropriate hash spacing.
217 * We want a number which causes most hash entries to cover
218 * at most two strips, but the hash table must be at most
219 * 1 PAGE. We choose the smallest strip, or contiguous collection
220 * of strips, that has big enough size. We never consider the last
221 * strip though as it's size has no bearing on the efficacy of the hash
224 conf->hash_spacing = curr_zone_offset;
225 min_spacing = curr_zone_offset;
226 sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
227 for (i=0; i < conf->nr_strip_zones-1; i++) {
229 for (j=i; j<conf->nr_strip_zones-1 &&
230 sz < min_spacing ; j++)
231 sz += conf->strip_zone[j].size;
232 if (sz >= min_spacing && sz < conf->hash_spacing)
233 conf->hash_spacing = sz;
236 mddev->queue->unplug_fn = raid0_unplug;
238 mddev->queue->issue_flush_fn = raid0_issue_flush;
240 printk("raid0: done.\n");
247 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
249 * @bio: the buffer head that's been built up so far
250 * @biovec: the request that could be merged to it.
252 * Return amount of bytes we can accept at this offset
254 static int raid0_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec)
256 mddev_t *mddev = q->queuedata;
257 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
259 unsigned int chunk_sectors = mddev->chunk_size >> 9;
260 unsigned int bio_sectors = bio->bi_size >> 9;
262 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
263 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
264 if (max <= biovec->bv_len && bio_sectors == 0)
265 return biovec->bv_len;
270 static int raid0_run (mddev_t *mddev)
272 unsigned cur=0, i=0, nb_zone;
276 struct list_head *tmp;
278 printk("%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);
308 sector_t s = mddev->array_size;
309 sector_t space = conf->hash_spacing;
312 if (sizeof(sector_t) > sizeof(u32)) {
313 /*shift down space and s so that sector_div will work */
314 while (space > (sector_t) (~(u32)0)) {
317 s += 1; /* force round-up */
321 round = sector_div(s, (u32)space) ? 1 : 0;
324 printk("raid0 : nb_zone is %d.\n", nb_zone);
326 printk("raid0 : Allocating %Zd bytes for hash.\n",
327 nb_zone*sizeof(struct strip_zone*));
328 conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
329 if (!conf->hash_table)
331 size = conf->strip_zone[cur].size;
333 for (i=0; i< nb_zone; i++) {
334 conf->hash_table[i] = conf->strip_zone + cur;
335 while (size <= conf->hash_spacing) {
337 size += conf->strip_zone[cur].size;
339 size -= conf->hash_spacing;
341 if (conf->preshift) {
342 conf->hash_spacing >>= conf->preshift;
343 /* round hash_spacing up so when we divide by it, we
344 * err on the side of too-low, which is safest
346 conf->hash_spacing++;
349 /* calculate the max read-ahead size.
350 * For read-ahead of large files to be effective, we need to
351 * readahead at least twice a whole stripe. i.e. number of devices
352 * multiplied by chunk size times 2.
353 * If an individual device has an ra_pages greater than the
354 * chunk size, then we will not drive that device as hard as it
355 * wants. We consider this a configuration error: a larger
356 * chunksize should be used in that case.
359 int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
360 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
361 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
365 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
369 kfree(conf->strip_zone);
370 kfree(conf->devlist);
372 mddev->private = NULL;
377 static int raid0_stop (mddev_t *mddev)
379 raid0_conf_t *conf = mddev_to_conf(mddev);
381 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
382 kfree(conf->hash_table);
383 conf->hash_table = NULL;
384 kfree(conf->strip_zone);
385 conf->strip_zone = NULL;
387 mddev->private = NULL;
392 static int raid0_make_request (request_queue_t *q, struct bio *bio)
394 mddev_t *mddev = q->queuedata;
395 unsigned int sect_in_chunk, chunksize_bits, chunk_size, chunk_sects;
396 raid0_conf_t *conf = mddev_to_conf(mddev);
397 struct strip_zone *zone;
400 sector_t block, rsect;
401 const int rw = bio_data_dir(bio);
403 if (unlikely(bio_barrier(bio))) {
404 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
408 disk_stat_inc(mddev->gendisk, ios[rw]);
409 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
411 chunk_size = mddev->chunk_size >> 10;
412 chunk_sects = mddev->chunk_size >> 9;
413 chunksize_bits = ffz(~chunk_size);
414 block = bio->bi_sector >> 1;
417 if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
419 /* Sanity check -- queue functions should prevent this happening */
420 if (bio->bi_vcnt != 1 ||
423 /* This is a one page bio that upper layers
424 * refuse to split for us, so we need to split it.
426 bp = bio_split(bio, bio_split_pool, chunk_sects - (bio->bi_sector & (chunk_sects - 1)) );
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);
432 bio_pair_release(bp);
441 sector_t x = block >> conf->preshift;
442 sector_div(x, (u32)conf->hash_spacing);
443 zone = conf->hash_table[x];
446 while (block >= (zone->zone_offset + zone->size))
449 sect_in_chunk = bio->bi_sector & ((chunk_size<<1) -1);
453 sector_t x = (block - zone->zone_offset) >> chunksize_bits;
455 sector_div(x, zone->nb_dev);
457 BUG_ON(x != (sector_t)chunk);
459 x = block >> chunksize_bits;
460 tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
462 rsect = (((chunk << chunksize_bits) + zone->dev_offset)<<1)
465 bio->bi_bdev = tmp_dev->bdev;
466 bio->bi_sector = rsect + tmp_dev->data_offset;
469 * Let the main block layer submit the IO and resolve recursion:
474 printk("raid0_make_request bug: can't convert block across chunks"
475 " or bigger than %dk %llu %d\n", chunk_size,
476 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
478 bio_io_error(bio, bio->bi_size);
482 static void raid0_status (struct seq_file *seq, mddev_t *mddev)
487 char b[BDEVNAME_SIZE];
488 raid0_conf_t *conf = mddev_to_conf(mddev);
491 for (j = 0; j < conf->nr_strip_zones; j++) {
492 seq_printf(seq, " z%d", j);
493 if (conf->hash_table[h] == conf->strip_zone+j)
494 seq_printf("(h%d)", h++);
495 seq_printf(seq, "=[");
496 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
497 seq_printf (seq, "%s/", bdevname(
498 conf->strip_zone[j].dev[k]->bdev,b));
500 seq_printf (seq, "] zo=%d do=%d s=%d\n",
501 conf->strip_zone[j].zone_offset,
502 conf->strip_zone[j].dev_offset,
503 conf->strip_zone[j].size);
506 seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
510 static mdk_personality_t raid0_personality=
513 .owner = THIS_MODULE,
514 .make_request = raid0_make_request,
517 .status = raid0_status,
520 static int __init raid0_init (void)
522 return register_md_personality (RAID0, &raid0_personality);
525 static void raid0_exit (void)
527 unregister_md_personality (RAID0);
530 module_init(raid0_init);
531 module_exit(raid0_exit);
532 MODULE_LICENSE("GPL");
533 MODULE_ALIAS("md-personality-2"); /* RAID0 */