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/module.h>
  22 #include <linux/raid/raid0.h>
  23
  24 #define MAJOR_NR MD_MAJOR
  25 #define MD_DRIVER
  26 #define MD_PERSONALITY
  27
  28 static void raid0_unplug(struct request_queue *q)
  29 {
  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;
  33         int i;
  34
  35         for (i=0; i<mddev->raid_disks; i++) {
  36                 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
  37
  38                 blk_unplug(r_queue);
  39         }
  40 }
  41
  42 static int raid0_congested(void *data, int bits)
  43 {
  44         mddev_t *mddev = data;
  45         raid0_conf_t *conf = mddev_to_conf(mddev);
  46         mdk_rdev_t **devlist = conf->strip_zone[0].dev;
  47         int i, ret = 0;
  48
  49         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
  50                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  51
  52                 ret |= bdi_congested(&q->backing_dev_info, bits);
  53         }
  54         return ret;
  55 }
  56
  57
  58 static int create_strip_zones (mddev_t *mddev)
  59 {
  60         int i, c, j;
  61         sector_t current_offset, curr_zone_offset;
  62         sector_t min_spacing;
  63         raid0_conf_t *conf = mddev_to_conf(mddev);
  64         mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
  65         struct list_head *tmp1, *tmp2;
  66         struct strip_zone *zone;
  67         int cnt;
  68         char b[BDEVNAME_SIZE];
  69
  70         /*
  71          * The number of 'same size groups'
  72          */
  73         conf->nr_strip_zones = 0;
  74
  75         rdev_for_each(rdev1, tmp1, mddev) {
  76                 printk("raid0: looking at %s\n",
  77                         bdevname(rdev1->bdev,b));
  78                 c = 0;
  79                 rdev_for_each(rdev2, tmp2, mddev) {
  80                         printk("raid0:   comparing %s(%llu)",
  81                                bdevname(rdev1->bdev,b),
  82                                (unsigned long long)rdev1->size);
  83                         printk(" with %s(%llu)\n",
  84                                bdevname(rdev2->bdev,b),
  85                                (unsigned long long)rdev2->size);
  86                         if (rdev2 == rdev1) {
  87                                 printk("raid0:   END\n");
  88                                 break;
  89                         }
  90                         if (rdev2->size == rdev1->size)
  91                         {
  92                                 /*
  93                                  * Not unique, don't count it as a new
  94                                  * group
  95                                  */
  96                                 printk("raid0:   EQUAL\n");
  97                                 c = 1;
  98                                 break;
  99                         }
 100                         printk("raid0:   NOT EQUAL\n");
 101                 }
 102                 if (!c) {
 103                         printk("raid0:   ==> UNIQUE\n");
 104                         conf->nr_strip_zones++;
 105                         printk("raid0: %d zones\n", conf->nr_strip_zones);
 106                 }
 107         }
 108         printk("raid0: FINAL %d zones\n", conf->nr_strip_zones);
 109
 110         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 111                                 conf->nr_strip_zones, GFP_KERNEL);
 112         if (!conf->strip_zone)
 113                 return 1;
 114         conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
 115                                 conf->nr_strip_zones*mddev->raid_disks,
 116                                 GFP_KERNEL);
 117         if (!conf->devlist)
 118                 return 1;
 119
 120         /* The first zone must contain all devices, so here we check that
 121          * there is a proper alignment of slots to devices and find them all
 122          */
 123         zone = &conf->strip_zone[0];
 124         cnt = 0;
 125         smallest = NULL;
 126         zone->dev = conf->devlist;
 127         rdev_for_each(rdev1, tmp1, mddev) {
 128                 int j = rdev1->raid_disk;
 129
 130                 if (j < 0 || j >= mddev->raid_disks) {
 131                         printk("raid0: bad disk number %d - aborting!\n", j);
 132                         goto abort;
 133                 }
 134                 if (zone->dev[j]) {
 135                         printk("raid0: multiple devices for %d - aborting!\n",
 136                                 j);
 137                         goto abort;
 138                 }
 139                 zone->dev[j] = rdev1;
 140
 141                 blk_queue_stack_limits(mddev->queue,
 142                                        rdev1->bdev->bd_disk->queue);
 143                 /* as we don't honour merge_bvec_fn, we must never risk
 144                  * violating it, so limit ->max_sector to one PAGE, as
 145                  * a one page request is never in violation.
 146                  */
 147
 148                 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
 149                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
 150                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 151
 152                 if (!smallest || (rdev1->size <smallest->size))
 153                         smallest = rdev1;
 154                 cnt++;
 155         }
 156         if (cnt != mddev->raid_disks) {
 157                 printk("raid0: too few disks (%d of %d) - aborting!\n",
 158                         cnt, mddev->raid_disks);
 159                 goto abort;
 160         }
 161         zone->nb_dev = cnt;
 162         zone->size = smallest->size * cnt;
 163         zone->zone_offset = 0;
 164
 165         current_offset = smallest->size;
 166         curr_zone_offset = zone->size;
 167
 168         /* now do the other zones */
 169         for (i = 1; i < conf->nr_strip_zones; i++)
 170         {
 171                 zone = conf->strip_zone + i;
 172                 zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
 173
 174                 printk("raid0: zone %d\n", i);
 175                 zone->dev_offset = current_offset;
 176                 smallest = NULL;
 177                 c = 0;
 178
 179                 for (j=0; j<cnt; j++) {
 180                         char b[BDEVNAME_SIZE];
 181                         rdev = conf->strip_zone[0].dev[j];
 182                         printk("raid0: checking %s ...", bdevname(rdev->bdev,b));
 183                         if (rdev->size > current_offset)
 184                         {
 185                                 printk(" contained as device %d\n", c);
 186                                 zone->dev[c] = rdev;
 187                                 c++;
 188                                 if (!smallest || (rdev->size <smallest->size)) {
 189                                         smallest = rdev;
 190                                         printk("  (%llu) is smallest!.\n",
 191                                                 (unsigned long long)rdev->size);
 192                                 }
 193                         } else
 194                                 printk(" nope.\n");
 195                 }
 196
 197                 zone->nb_dev = c;
 198                 zone->size = (smallest->size - current_offset) * c;
 199                 printk("raid0: zone->nb_dev: %d, size: %llu\n",
 200                         zone->nb_dev, (unsigned long long)zone->size);
 201
 202                 zone->zone_offset = curr_zone_offset;
 203                 curr_zone_offset += zone->size;
 204
 205                 current_offset = smallest->size;
 206                 printk("raid0: current zone offset: %llu\n",
 207                         (unsigned long long)current_offset);
 208         }
 209
 210         /* Now find appropriate hash spacing.
 211          * We want a number which causes most hash entries to cover
 212          * at most two strips, but the hash table must be at most
 213          * 1 PAGE.  We choose the smallest strip, or contiguous collection
 214          * of strips, that has big enough size.  We never consider the last
 215          * strip though as it's size has no bearing on the efficacy of the hash
 216          * table.
 217          */
 218         conf->hash_spacing = curr_zone_offset;
 219         min_spacing = curr_zone_offset;
 220         sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
 221         for (i=0; i < conf->nr_strip_zones-1; i++) {
 222                 sector_t sz = 0;
 223                 for (j=i; j<conf->nr_strip_zones-1 &&
 224                              sz < min_spacing ; j++)
 225                         sz += conf->strip_zone[j].size;
 226                 if (sz >= min_spacing && sz < conf->hash_spacing)
 227                         conf->hash_spacing = sz;
 228         }
 229
 230         mddev->queue->unplug_fn = raid0_unplug;
 231
 232         mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 233         mddev->queue->backing_dev_info.congested_data = mddev;
 234
 235         printk("raid0: done.\n");
 236         return 0;
 237  abort:
 238         return 1;
 239 }
 240
 241 /**
 242  *      raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 243  *      @q: request queue
 244  *      @bio: the buffer head that's been built up so far
 245  *      @biovec: the request that could be merged to it.
 246  *
 247  *      Return amount of bytes we can accept at this offset
 248  */
 249 static int raid0_mergeable_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *biovec)
 250 {
 251         mddev_t *mddev = q->queuedata;
 252         sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
 253         int max;
 254         unsigned int chunk_sectors = mddev->chunk_size >> 9;
 255         unsigned int bio_sectors = bio->bi_size >> 9;
 256
 257         max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
 258         if (max < 0) max = 0; /* bio_add cannot handle a negative return */
 259         if (max <= biovec->bv_len && bio_sectors == 0)
 260                 return biovec->bv_len;
 261         else
 262                 return max;
 263 }
 264
 265 static int raid0_run (mddev_t *mddev)
 266 {
 267         unsigned  cur=0, i=0, nb_zone;
 268         s64 size;
 269         raid0_conf_t *conf;
 270         mdk_rdev_t *rdev;
 271         struct list_head *tmp;
 272
 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",
 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);
 283
 284         conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
 285         if (!conf)
 286                 goto out;
 287         mddev->private = (void *)conf;
 288
 289         conf->strip_zone = NULL;
 290         conf->devlist = NULL;
 291         if (create_strip_zones (mddev))
 292                 goto out_free_conf;
 293
 294         /* calculate array device size */
 295         mddev->array_size = 0;
 296         rdev_for_each(rdev, tmp, mddev)
 297                 mddev->array_size += rdev->size;
 298
 299         printk("raid0 : md_size is %llu blocks.\n",
 300                 (unsigned long long)mddev->array_size);
 301         printk("raid0 : conf->hash_spacing is %llu blocks.\n",
 302                 (unsigned long long)conf->hash_spacing);
 303         {
 304                 sector_t s = mddev->array_size;
 305                 sector_t space = conf->hash_spacing;
 306                 int round;
 307                 conf->preshift = 0;
 308                 if (sizeof(sector_t) > sizeof(u32)) {
 309                         /*shift down space and s so that sector_div will work */
 310                         while (space > (sector_t) (~(u32)0)) {
 311                                 s >>= 1;
 312                                 space >>= 1;
 313                                 s += 1; /* force round-up */
 314                                 conf->preshift++;
 315                         }
 316                 }
 317                 round = sector_div(s, (u32)space) ? 1 : 0;
 318                 nb_zone = s + round;
 319         }
 320         printk("raid0 : nb_zone is %d.\n", nb_zone);
 321
 322         printk("raid0 : Allocating %Zd bytes for hash.\n",
 323                                 nb_zone*sizeof(struct strip_zone*));
 324         conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
 325         if (!conf->hash_table)
 326                 goto out_free_conf;
 327         size = conf->strip_zone[cur].size;
 328
 329         conf->hash_table[0] = conf->strip_zone + cur;
 330         for (i=1; i< nb_zone; i++) {
 331                 while (size <= conf->hash_spacing) {
 332                         cur++;
 333                         size += conf->strip_zone[cur].size;
 334                 }
 335                 size -= conf->hash_spacing;
 336                 conf->hash_table[i] = conf->strip_zone + cur;
 337         }
 338         if (conf->preshift) {
 339                 conf->hash_spacing >>= conf->preshift;
 340                 /* round hash_spacing up so when we divide by it, we
 341                  * err on the side of too-low, which is safest
 342                  */
 343                 conf->hash_spacing++;
 344         }
 345
 346         /* calculate the max read-ahead size.
 347          * For read-ahead of large files to be effective, we need to
 348          * readahead at least twice a whole stripe. i.e. number of devices
 349          * multiplied by chunk size times 2.
 350          * If an individual device has an ra_pages greater than the
 351          * chunk size, then we will not drive that device as hard as it
 352          * wants.  We consider this a configuration error: a larger
 353          * chunksize should be used in that case.
 354          */
 355         {
 356                 int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
 357                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 358                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 359         }
 360
 361
 362         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 363         return 0;
 364
 365 out_free_conf:
 366         kfree(conf->strip_zone);
 367         kfree(conf->devlist);
 368         kfree(conf);
 369         mddev->private = NULL;
 370 out:
 371         return -ENOMEM;
 372 }
 373
 374 static int raid0_stop (mddev_t *mddev)
 375 {
 376         raid0_conf_t *conf = mddev_to_conf(mddev);
 377
 378         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 379         kfree(conf->hash_table);
 380         conf->hash_table = NULL;
 381         kfree(conf->strip_zone);
 382         conf->strip_zone = NULL;
 383         kfree(conf);
 384         mddev->private = NULL;
 385
 386         return 0;
 387 }
 388
 389 static int raid0_make_request (struct request_queue *q, struct bio *bio)
 390 {
 391         mddev_t *mddev = q->queuedata;
 392         unsigned int sect_in_chunk, chunksize_bits,  chunk_size, chunk_sects;
 393         raid0_conf_t *conf = mddev_to_conf(mddev);
 394         struct strip_zone *zone;
 395         mdk_rdev_t *tmp_dev;
 396         sector_t chunk;
 397         sector_t block, rsect;
 398         const int rw = bio_data_dir(bio);
 399
 400         if (unlikely(bio_barrier(bio))) {
 401                 bio_endio(bio, -EOPNOTSUPP);
 402                 return 0;
 403         }
 404
 405         disk_stat_inc(mddev->gendisk, ios[rw]);
 406         disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
 407
 408         chunk_size = mddev->chunk_size >> 10;
 409         chunk_sects = mddev->chunk_size >> 9;
 410         chunksize_bits = ffz(~chunk_size);
 411         block = bio->bi_sector >> 1;
 412
 413
 414         if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
 415                 struct bio_pair *bp;
 416                 /* Sanity check -- queue functions should prevent this happening */
 417                 if (bio->bi_vcnt != 1 ||
 418                     bio->bi_idx != 0)
 419                         goto bad_map;
 420                 /* This is a one page bio that upper layers
 421                  * refuse to split for us, so we need to split it.
 422                  */
 423                 bp = bio_split(bio, bio_split_pool, chunk_sects - (bio->bi_sector & (chunk_sects - 1)) );
 424                 if (raid0_make_request(q, &bp->bio1))
 425                         generic_make_request(&bp->bio1);
 426                 if (raid0_make_request(q, &bp->bio2))
 427                         generic_make_request(&bp->bio2);
 428
 429                 bio_pair_release(bp);
 430                 return 0;
 431         }
 432
 433
 434         {
 435                 sector_t x = block >> conf->preshift;
 436                 sector_div(x, (u32)conf->hash_spacing);
 437                 zone = conf->hash_table[x];
 438         }
 439
 440         while (block >= (zone->zone_offset + zone->size))
 441                 zone++;
 442
 443         sect_in_chunk = bio->bi_sector & ((chunk_size<<1) -1);
 444
 445
 446         {
 447                 sector_t x =  (block - zone->zone_offset) >> chunksize_bits;
 448
 449                 sector_div(x, zone->nb_dev);
 450                 chunk = x;
 451
 452                 x = block >> chunksize_bits;
 453                 tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
 454         }
 455         rsect = (((chunk << chunksize_bits) + zone->dev_offset)<<1)
 456                 + sect_in_chunk;
 457
 458         bio->bi_bdev = tmp_dev->bdev;
 459         bio->bi_sector = rsect + tmp_dev->data_offset;
 460
 461         /*
 462          * Let the main block layer submit the IO and resolve recursion:
 463          */
 464         return 1;
 465
 466 bad_map:
 467         printk("raid0_make_request bug: can't convert block across chunks"
 468                 " or bigger than %dk %llu %d\n", chunk_size,
 469                 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 470
 471         bio_io_error(bio);
 472         return 0;
 473 }
 474
 475 static void raid0_status (struct seq_file *seq, mddev_t *mddev)
 476 {
 477 #undef MD_DEBUG
 478 #ifdef MD_DEBUG
 479         int j, k, h;
 480         char b[BDEVNAME_SIZE];
 481         raid0_conf_t *conf = mddev_to_conf(mddev);
 482
 483         h = 0;
 484         for (j = 0; j < conf->nr_strip_zones; j++) {
 485                 seq_printf(seq, "      z%d", j);
 486                 if (conf->hash_table[h] == conf->strip_zone+j)
 487                         seq_printf(seq, "(h%d)", h++);
 488                 seq_printf(seq, "=[");
 489                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 490                         seq_printf(seq, "%s/", bdevname(
 491                                 conf->strip_zone[j].dev[k]->bdev,b));
 492
 493                 seq_printf(seq, "] zo=%d do=%d s=%d\n",
 494                                 conf->strip_zone[j].zone_offset,
 495                                 conf->strip_zone[j].dev_offset,
 496                                 conf->strip_zone[j].size);
 497         }
 498 #endif
 499         seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
 500         return;
 501 }
 502
 503 static struct mdk_personality raid0_personality=
 504 {
 505         .name           = "raid0",
 506         .level          = 0,
 507         .owner          = THIS_MODULE,
 508         .make_request   = raid0_make_request,
 509         .run            = raid0_run,
 510         .stop           = raid0_stop,
 511         .status         = raid0_status,
 512 };
 513
 514 static int __init raid0_init (void)
 515 {
 516         return register_md_personality (&raid0_personality);
 517 }
 518
 519 static void raid0_exit (void)
 520 {
 521         unregister_md_personality (&raid0_personality);
 522 }
 523
 524 module_init(raid0_init);
 525 module_exit(raid0_exit);
 526 MODULE_LICENSE("GPL");
 527 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 528 MODULE_ALIAS("md-raid0");
 529 MODULE_ALIAS("md-level-0");