drivers/md/raid0.c

   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
  21 #include <linux/raid/raid0.h>
  22
  23 static void raid0_unplug(struct request_queue *q)
  24 {
  25         mddev_t *mddev = q->queuedata;
  26         raid0_conf_t *conf = mddev_to_conf(mddev);
  27         mdk_rdev_t **devlist = conf->strip_zone[0].dev;
  28         int i;
  29
  30         for (i=0; i<mddev->raid_disks; i++) {
  31                 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
  32
  33                 blk_unplug(r_queue);
  34         }
  35 }
  36
  37 static int raid0_congested(void *data, int bits)
  38 {
  39         mddev_t *mddev = data;
  40         raid0_conf_t *conf = mddev_to_conf(mddev);
  41         mdk_rdev_t **devlist = conf->strip_zone[0].dev;
  42         int i, ret = 0;
  43
  44         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
  45                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  46
  47                 ret |= bdi_congested(&q->backing_dev_info, bits);
  48         }
  49         return ret;
  50 }
  51
  52
  53 static int create_strip_zones (mddev_t *mddev)
  54 {
  55         int i, c, j;
  56         sector_t current_offset, curr_zone_offset;
  57         sector_t min_spacing;
  58         raid0_conf_t *conf = mddev_to_conf(mddev);
  59         mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
  60         struct list_head *tmp1, *tmp2;
  61         struct strip_zone *zone;
  62         int cnt;
  63         char b[BDEVNAME_SIZE];
  64
  65         /*
  66          * The number of 'same size groups'
  67          */
  68         conf->nr_strip_zones = 0;
  69
  70         rdev_for_each(rdev1, tmp1, mddev) {
  71                 printk("raid0: looking at %s\n",
  72                         bdevname(rdev1->bdev,b));
  73                 c = 0;
  74                 rdev_for_each(rdev2, tmp2, mddev) {
  75                         printk("raid0:   comparing %s(%llu)",
  76                                bdevname(rdev1->bdev,b),
  77                                (unsigned long long)rdev1->size);
  78                         printk(" with %s(%llu)\n",
  79                                bdevname(rdev2->bdev,b),
  80                                (unsigned long long)rdev2->size);
  81                         if (rdev2 == rdev1) {
  82                                 printk("raid0:   END\n");
  83                                 break;
  84                         }
  85                         if (rdev2->size == rdev1->size)
  86                         {
  87                                 /*
  88                                  * Not unique, don't count it as a new
  89                                  * group
  90                                  */
  91                                 printk("raid0:   EQUAL\n");
  92                                 c = 1;
  93                                 break;
  94                         }
  95                         printk("raid0:   NOT EQUAL\n");
  96                 }
  97                 if (!c) {
  98                         printk("raid0:   ==> UNIQUE\n");
  99                         conf->nr_strip_zones++;
 100                         printk("raid0: %d zones\n", conf->nr_strip_zones);
 101                 }
 102         }
 103         printk("raid0: FINAL %d zones\n", conf->nr_strip_zones);
 104
 105         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 106                                 conf->nr_strip_zones, GFP_KERNEL);
 107         if (!conf->strip_zone)
 108                 return 1;
 109         conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
 110                                 conf->nr_strip_zones*mddev->raid_disks,
 111                                 GFP_KERNEL);
 112         if (!conf->devlist)
 113                 return 1;
 114
 115         /* The first zone must contain all devices, so here we check that
 116          * there is a proper alignment of slots to devices and find them all
 117          */
 118         zone = &conf->strip_zone[0];
 119         cnt = 0;
 120         smallest = NULL;
 121         zone->dev = conf->devlist;
 122         rdev_for_each(rdev1, tmp1, mddev) {
 123                 int j = rdev1->raid_disk;
 124
 125                 if (j < 0 || j >= mddev->raid_disks) {
 126                         printk("raid0: bad disk number %d - aborting!\n", j);
 127                         goto abort;
 128                 }
 129                 if (zone->dev[j]) {
 130                         printk("raid0: multiple devices for %d - aborting!\n",
 131                                 j);
 132                         goto abort;
 133                 }
 134                 zone->dev[j] = rdev1;
 135
 136                 blk_queue_stack_limits(mddev->queue,
 137                                        rdev1->bdev->bd_disk->queue);
 138                 /* as we don't honour merge_bvec_fn, we must never risk
 139                  * violating it, so limit ->max_sector to one PAGE, as
 140                  * a one page request is never in violation.
 141                  */
 142
 143                 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
 144                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
 145                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 146
 147                 if (!smallest || (rdev1->size <smallest->size))
 148                         smallest = rdev1;
 149                 cnt++;
 150         }
 151         if (cnt != mddev->raid_disks) {
 152                 printk("raid0: too few disks (%d of %d) - aborting!\n",
 153                         cnt, mddev->raid_disks);
 154                 goto abort;
 155         }
 156         zone->nb_dev = cnt;
 157         zone->size = smallest->size * cnt;
 158         zone->zone_offset = 0;
 159
 160         current_offset = smallest->size;
 161         curr_zone_offset = zone->size;
 162
 163         /* now do the other zones */
 164         for (i = 1; i < conf->nr_strip_zones; i++)
 165         {
 166                 zone = conf->strip_zone + i;
 167                 zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
 168
 169                 printk("raid0: zone %d\n", i);
 170                 zone->dev_offset = current_offset;
 171                 smallest = NULL;
 172                 c = 0;
 173
 174                 for (j=0; j<cnt; j++) {
 175                         char b[BDEVNAME_SIZE];
 176                         rdev = conf->strip_zone[0].dev[j];
 177                         printk("raid0: checking %s ...", bdevname(rdev->bdev,b));
 178                         if (rdev->size > current_offset)
 179                         {
 180                                 printk(" contained as device %d\n", c);
 181                                 zone->dev[c] = rdev;
 182                                 c++;
 183                                 if (!smallest || (rdev->size <smallest->size)) {
 184                                         smallest = rdev;
 185                                         printk("  (%llu) is smallest!.\n",
 186                                                 (unsigned long long)rdev->size);
 187                                 }
 188                         } else
 189                                 printk(" nope.\n");
 190                 }
 191
 192                 zone->nb_dev = c;
 193                 zone->size = (smallest->size - current_offset) * c;
 194                 printk("raid0: zone->nb_dev: %d, size: %llu\n",
 195                         zone->nb_dev, (unsigned long long)zone->size);
 196
 197                 zone->zone_offset = curr_zone_offset;
 198                 curr_zone_offset += zone->size;
 199
 200                 current_offset = smallest->size;
 201                 printk("raid0: current zone offset: %llu\n",
 202                         (unsigned long long)current_offset);
 203         }
 204
 205         /* Now find appropriate hash spacing.
 206          * We want a number which causes most hash entries to cover
 207          * at most two strips, but the hash table must be at most
 208          * 1 PAGE.  We choose the smallest strip, or contiguous collection
 209          * of strips, that has big enough size.  We never consider the last
 210          * strip though as it's size has no bearing on the efficacy of the hash
 211          * table.
 212          */
 213         conf->hash_spacing = curr_zone_offset;
 214         min_spacing = curr_zone_offset;
 215         sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
 216         for (i=0; i < conf->nr_strip_zones-1; i++) {
 217                 sector_t sz = 0;
 218                 for (j=i; j<conf->nr_strip_zones-1 &&
 219                              sz < min_spacing ; j++)
 220                         sz += conf->strip_zone[j].size;
 221                 if (sz >= min_spacing && sz < conf->hash_spacing)
 222                         conf->hash_spacing = sz;
 223         }
 224
 225         mddev->queue->unplug_fn = raid0_unplug;
 226
 227         mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 228         mddev->queue->backing_dev_info.congested_data = mddev;
 229
 230         printk("raid0: done.\n");
 231         return 0;
 232  abort:
 233         return 1;
 234 }
 235
 236 /**
 237  *      raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 238  *      @q: request queue
 239  *      @bvm: properties of new bio
 240  *      @biovec: the request that could be merged to it.
 241  *
 242  *      Return amount of bytes we can accept at this offset
 243  */
 244 static int raid0_mergeable_bvec(struct request_queue *q,
 245                                 struct bvec_merge_data *bvm,
 246                                 struct bio_vec *biovec)
 247 {
 248         mddev_t *mddev = q->queuedata;
 249         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 250         int max;
 251         unsigned int chunk_sectors = mddev->chunk_size >> 9;
 252         unsigned int bio_sectors = bvm->bi_size >> 9;
 253
 254         max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
 255         if (max < 0) max = 0; /* bio_add cannot handle a negative return */
 256         if (max <= biovec->bv_len && bio_sectors == 0)
 257                 return biovec->bv_len;
 258         else
 259                 return max;
 260 }
 261
 262 static int raid0_run (mddev_t *mddev)
 263 {
 264         unsigned  cur=0, i=0, nb_zone;
 265         s64 size;
 266         raid0_conf_t *conf;
 267         mdk_rdev_t *rdev;
 268         struct list_head *tmp;
 269
 270         if (mddev->chunk_size == 0) {
 271                 printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
 272                 return -EINVAL;
 273         }
 274         printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
 275                mdname(mddev),
 276                mddev->chunk_size >> 9,
 277                (mddev->chunk_size>>1)-1);
 278         blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
 279         blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
 280         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 281
 282         conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
 283         if (!conf)
 284                 goto out;
 285         mddev->private = (void *)conf;
 286
 287         conf->strip_zone = NULL;
 288         conf->devlist = NULL;
 289         if (create_strip_zones (mddev))
 290                 goto out_free_conf;
 291
 292         /* calculate array device size */
 293         mddev->array_sectors = 0;
 294         rdev_for_each(rdev, tmp, mddev)
 295                 mddev->array_sectors += rdev->size * 2;
 296
 297         printk("raid0 : md_size is %llu blocks.\n",
 298                 (unsigned long long)mddev->array_sectors / 2);
 299         printk("raid0 : conf->hash_spacing is %llu blocks.\n",
 300                 (unsigned long long)conf->hash_spacing);
 301         {
 302                 sector_t s = mddev->array_sectors / 2;
 303                 sector_t space = conf->hash_spacing;
 304                 int round;
 305                 conf->preshift = 0;
 306                 if (sizeof(sector_t) > sizeof(u32)) {
 307                         /*shift down space and s so that sector_div will work */
 308                         while (space > (sector_t) (~(u32)0)) {
 309                                 s >>= 1;
 310                                 space >>= 1;
 311                                 s += 1; /* force round-up */
 312                                 conf->preshift++;
 313                         }
 314                 }
 315                 round = sector_div(s, (u32)space) ? 1 : 0;
 316                 nb_zone = s + round;
 317         }
 318         printk("raid0 : nb_zone is %d.\n", nb_zone);
 319
 320         printk("raid0 : Allocating %Zd bytes for hash.\n",
 321                                 nb_zone*sizeof(struct strip_zone*));
 322         conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
 323         if (!conf->hash_table)
 324                 goto out_free_conf;
 325         size = conf->strip_zone[cur].size;
 326
 327         conf->hash_table[0] = conf->strip_zone + cur;
 328         for (i=1; i< nb_zone; i++) {
 329                 while (size <= conf->hash_spacing) {
 330                         cur++;
 331                         size += conf->strip_zone[cur].size;
 332                 }
 333                 size -= conf->hash_spacing;
 334                 conf->hash_table[i] = conf->strip_zone + cur;
 335         }
 336         if (conf->preshift) {
 337                 conf->hash_spacing >>= conf->preshift;
 338                 /* round hash_spacing up so when we divide by it, we
 339                  * err on the side of too-low, which is safest
 340                  */
 341                 conf->hash_spacing++;
 342         }
 343
 344         /* calculate the max read-ahead size.
 345          * For read-ahead of large files to be effective, we need to
 346          * readahead at least twice a whole stripe. i.e. number of devices
 347          * multiplied by chunk size times 2.
 348          * If an individual device has an ra_pages greater than the
 349          * chunk size, then we will not drive that device as hard as it
 350          * wants.  We consider this a configuration error: a larger
 351          * chunksize should be used in that case.
 352          */
 353         {
 354                 int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
 355                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 356                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 357         }
 358
 359
 360         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 361         return 0;
 362
 363 out_free_conf:
 364         kfree(conf->strip_zone);
 365         kfree(conf->devlist);
 366         kfree(conf);
 367         mddev->private = NULL;
 368 out:
 369         return -ENOMEM;
 370 }
 371
 372 static int raid0_stop (mddev_t *mddev)
 373 {
 374         raid0_conf_t *conf = mddev_to_conf(mddev);
 375
 376         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 377         kfree(conf->hash_table);
 378         conf->hash_table = NULL;
 379         kfree(conf->strip_zone);
 380         conf->strip_zone = NULL;
 381         kfree(conf);
 382         mddev->private = NULL;
 383
 384         return 0;
 385 }
 386
 387 static int raid0_make_request (struct request_queue *q, struct bio *bio)
 388 {
 389         mddev_t *mddev = q->queuedata;
 390         unsigned int sect_in_chunk, chunksize_bits,  chunk_size, chunk_sects;
 391         raid0_conf_t *conf = mddev_to_conf(mddev);
 392         struct strip_zone *zone;
 393         mdk_rdev_t *tmp_dev;
 394         sector_t chunk;
 395         sector_t block, rsect;
 396         const int rw = bio_data_dir(bio);
 397         int cpu;
 398
 399         if (unlikely(bio_barrier(bio))) {
 400                 bio_endio(bio, -EOPNOTSUPP);
 401                 return 0;
 402         }
 403
 404         cpu = part_stat_lock();
 405         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 406         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 407                       bio_sectors(bio));
 408         part_stat_unlock();
 409
 410         chunk_size = mddev->chunk_size >> 10;
 411         chunk_sects = mddev->chunk_size >> 9;
 412         chunksize_bits = ffz(~chunk_size);
 413         block = bio->bi_sector >> 1;
 414
 415
 416         if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
 417                 struct bio_pair *bp;
 418                 /* Sanity check -- queue functions should prevent this happening */
 419                 if (bio->bi_vcnt != 1 ||
 420                     bio->bi_idx != 0)
 421                         goto bad_map;
 422                 /* This is a one page bio that upper layers
 423                  * refuse to split for us, so we need to split it.
 424                  */
 425                 bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
 426                 if (raid0_make_request(q, &bp->bio1))
 427                         generic_make_request(&bp->bio1);
 428                 if (raid0_make_request(q, &bp->bio2))
 429                         generic_make_request(&bp->bio2);
 430
 431                 bio_pair_release(bp);
 432                 return 0;
 433         }
 434
 435
 436         {
 437                 sector_t x = block >> conf->preshift;
 438                 sector_div(x, (u32)conf->hash_spacing);
 439                 zone = conf->hash_table[x];
 440         }
 441
 442         while (block >= (zone->zone_offset + zone->size))
 443                 zone++;
 444
 445         sect_in_chunk = bio->bi_sector & ((chunk_size<<1) -1);
 446
 447
 448         {
 449                 sector_t x =  (block - zone->zone_offset) >> chunksize_bits;
 450
 451                 sector_div(x, zone->nb_dev);
 452                 chunk = x;
 453
 454                 x = block >> chunksize_bits;
 455                 tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
 456         }
 457         rsect = (((chunk << chunksize_bits) + zone->dev_offset)<<1)
 458                 + sect_in_chunk;
 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"
 470                 " or bigger than %dk %llu %d\n", chunk_size,
 471                 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 472
 473         bio_io_error(bio);
 474         return 0;
 475 }
 476
 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);
 484
 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)
 489                         seq_printf(seq, "(h%d)", h++);
 490                 seq_printf(seq, "=[");
 491                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 492                         seq_printf(seq, "%s/", bdevname(
 493                                 conf->strip_zone[j].dev[k]->bdev,b));
 494
 495                 seq_printf(seq, "] zo=%d do=%d s=%d\n",
 496                                 conf->strip_zone[j].zone_offset,
 497                                 conf->strip_zone[j].dev_offset,
 498                                 conf->strip_zone[j].size);
 499         }
 500 #endif
 501         seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
 502         return;
 503 }
 504
 505 static struct mdk_personality raid0_personality=
 506 {
 507         .name           = "raid0",
 508         .level          = 0,
 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 {
 518         return register_md_personality (&raid0_personality);
 519 }
 520
 521 static void raid0_exit (void)
 522 {
 523         unregister_md_personality (&raid0_personality);
 524 }
 525
 526 module_init(raid0_init);
 527 module_exit(raid0_exit);
 528 MODULE_LICENSE("GPL");
 529 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 530 MODULE_ALIAS("md-raid0");
 531 MODULE_ALIAS("md-level-0");