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/blkdev.h>
  22 #include <linux/raid/md_k.h>
  23 #include <linux/seq_file.h>
  24 #include "raid0.h"
  25
  26 static void raid0_unplug(struct request_queue *q)
  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++) {
  34                 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
  35
  36                 blk_unplug(r_queue);
  37         }
  38 }
  39
  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++) {
  48                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  49
  50                 ret |= bdi_congested(&q->backing_dev_info, bits);
  51         }
  52         return ret;
  53 }
  54
  55
  56 static int create_strip_zones (mddev_t *mddev)
  57 {
  58         int i, c, j;
  59         sector_t current_start, curr_zone_start;
  60         sector_t min_spacing;
  61         raid0_conf_t *conf = mddev_to_conf(mddev);
  62         mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
  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
  72         list_for_each_entry(rdev1, &mddev->disks, same_set) {
  73                 printk(KERN_INFO "raid0: looking at %s\n",
  74                         bdevname(rdev1->bdev,b));
  75                 c = 0;
  76                 list_for_each_entry(rdev2, &mddev->disks, same_set) {
  77                         printk(KERN_INFO "raid0:   comparing %s(%llu)",
  78                                bdevname(rdev1->bdev,b),
  79                                (unsigned long long)rdev1->size);
  80                         printk(KERN_INFO " with %s(%llu)\n",
  81                                bdevname(rdev2->bdev,b),
  82                                (unsigned long long)rdev2->size);
  83                         if (rdev2 == rdev1) {
  84                                 printk(KERN_INFO "raid0:   END\n");
  85                                 break;
  86                         }
  87                         if (rdev2->size == rdev1->size)
  88                         {
  89                                 /*
  90                                  * Not unique, don't count it as a new
  91                                  * group
  92                                  */
  93                                 printk(KERN_INFO "raid0:   EQUAL\n");
  94                                 c = 1;
  95                                 break;
  96                         }
  97                         printk(KERN_INFO "raid0:   NOT EQUAL\n");
  98                 }
  99                 if (!c) {
 100                         printk(KERN_INFO "raid0:   ==> UNIQUE\n");
 101                         conf->nr_strip_zones++;
 102                         printk(KERN_INFO "raid0: %d zones\n",
 103                                 conf->nr_strip_zones);
 104                 }
 105         }
 106         printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
 107
 108         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 109                                 conf->nr_strip_zones, GFP_KERNEL);
 110         if (!conf->strip_zone)
 111                 return 1;
 112         conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
 113                                 conf->nr_strip_zones*mddev->raid_disks,
 114                                 GFP_KERNEL);
 115         if (!conf->devlist)
 116                 return 1;
 117
 118         /* The first zone must contain all devices, so here we check that
 119          * there is a proper alignment of slots to devices and find them all
 120          */
 121         zone = &conf->strip_zone[0];
 122         cnt = 0;
 123         smallest = NULL;
 124         zone->dev = conf->devlist;
 125         list_for_each_entry(rdev1, &mddev->disks, same_set) {
 126                 int j = rdev1->raid_disk;
 127
 128                 if (j < 0 || j >= mddev->raid_disks) {
 129                         printk(KERN_ERR "raid0: bad disk number %d - "
 130                                 "aborting!\n", j);
 131                         goto abort;
 132                 }
 133                 if (zone->dev[j]) {
 134                         printk(KERN_ERR "raid0: multiple devices for %d - "
 135                                 "aborting!\n", j);
 136                         goto abort;
 137                 }
 138                 zone->dev[j] = rdev1;
 139
 140                 blk_queue_stack_limits(mddev->queue,
 141                                        rdev1->bdev->bd_disk->queue);
 142                 /* as we don't honour merge_bvec_fn, we must never risk
 143                  * violating it, so limit ->max_sector to one PAGE, as
 144                  * a one page request is never in violation.
 145                  */
 146
 147                 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
 148                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
 149                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 150
 151                 if (!smallest || (rdev1->size <smallest->size))
 152                         smallest = rdev1;
 153                 cnt++;
 154         }
 155         if (cnt != mddev->raid_disks) {
 156                 printk(KERN_ERR "raid0: too few disks (%d of %d) - "
 157                         "aborting!\n", cnt, mddev->raid_disks);
 158                 goto abort;
 159         }
 160         zone->nb_dev = cnt;
 161         zone->sectors = smallest->size * cnt * 2;
 162         zone->zone_start = 0;
 163
 164         current_start = smallest->size * 2;
 165         curr_zone_start = zone->sectors;
 166
 167         /* now do the other zones */
 168         for (i = 1; i < conf->nr_strip_zones; i++)
 169         {
 170                 zone = conf->strip_zone + i;
 171                 zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
 172
 173                 printk(KERN_INFO "raid0: zone %d\n", i);
 174                 zone->dev_start = current_start;
 175                 smallest = NULL;
 176                 c = 0;
 177
 178                 for (j=0; j<cnt; j++) {
 179                         char b[BDEVNAME_SIZE];
 180                         rdev = conf->strip_zone[0].dev[j];
 181                         printk(KERN_INFO "raid0: checking %s ...",
 182                                 bdevname(rdev->bdev, b));
 183                         if (rdev->size > current_start / 2) {
 184                                 printk(KERN_INFO " contained as device %d\n",
 185                                         c);
 186                                 zone->dev[c] = rdev;
 187                                 c++;
 188                                 if (!smallest || (rdev->size <smallest->size)) {
 189                                         smallest = rdev;
 190                                         printk(KERN_INFO "  (%llu) is smallest!.\n",
 191                                                 (unsigned long long)rdev->size);
 192                                 }
 193                         } else
 194                                 printk(KERN_INFO " nope.\n");
 195                 }
 196
 197                 zone->nb_dev = c;
 198                 zone->sectors = (smallest->size * 2 - current_start) * c;
 199                 printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
 200                         zone->nb_dev, (unsigned long long)zone->sectors);
 201
 202                 zone->zone_start = curr_zone_start;
 203                 curr_zone_start += zone->sectors;
 204
 205                 current_start = smallest->size * 2;
 206                 printk(KERN_INFO "raid0: current zone start: %llu\n",
 207                         (unsigned long long)current_start);
 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->spacing = curr_zone_start;
 219         min_spacing = curr_zone_start;
 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 s = 0;
 223                 for (j = i; j < conf->nr_strip_zones - 1 &&
 224                                 s < min_spacing; j++)
 225                         s += conf->strip_zone[j].sectors;
 226                 if (s >= min_spacing && s < conf->spacing)
 227                         conf->spacing = s;
 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(KERN_INFO "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  *      @bvm: properties of new bio
 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,
 250                                 struct bvec_merge_data *bvm,
 251                                 struct bio_vec *biovec)
 252 {
 253         mddev_t *mddev = q->queuedata;
 254         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 255         int max;
 256         unsigned int chunk_sectors = mddev->chunk_size >> 9;
 257         unsigned int bio_sectors = bvm->bi_size >> 9;
 258
 259         max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
 260         if (max < 0) max = 0; /* bio_add cannot handle a negative return */
 261         if (max <= biovec->bv_len && bio_sectors == 0)
 262                 return biovec->bv_len;
 263         else
 264                 return max;
 265 }
 266
 267 static int raid0_run (mddev_t *mddev)
 268 {
 269         unsigned  cur=0, i=0, nb_zone;
 270         s64 sectors;
 271         raid0_conf_t *conf;
 272         mdk_rdev_t *rdev;
 273
 274         if (mddev->chunk_size == 0) {
 275                 printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
 276                 return -EINVAL;
 277         }
 278         printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
 279                mdname(mddev),
 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);
 284         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 285
 286         conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
 287         if (!conf)
 288                 goto out;
 289         mddev->private = (void *)conf;
 290
 291         conf->strip_zone = NULL;
 292         conf->devlist = NULL;
 293         if (create_strip_zones (mddev))
 294                 goto out_free_conf;
 295
 296         /* calculate array device size */
 297         mddev->array_sectors = 0;
 298         list_for_each_entry(rdev, &mddev->disks, same_set)
 299                 mddev->array_sectors += rdev->size * 2;
 300
 301         printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
 302                 (unsigned long long)mddev->array_sectors);
 303         printk(KERN_INFO "raid0 : conf->spacing is %llu sectors.\n",
 304                 (unsigned long long)conf->spacing);
 305         {
 306                 sector_t s = mddev->array_sectors;
 307                 sector_t space = conf->spacing;
 308                 int round;
 309                 conf->sector_shift = 0;
 310                 if (sizeof(sector_t) > sizeof(u32)) {
 311                         /*shift down space and s so that sector_div will work */
 312                         while (space > (sector_t) (~(u32)0)) {
 313                                 s >>= 1;
 314                                 space >>= 1;
 315                                 s += 1; /* force round-up */
 316                                 conf->sector_shift++;
 317                         }
 318                 }
 319                 round = sector_div(s, (u32)space) ? 1 : 0;
 320                 nb_zone = s + round;
 321         }
 322         printk(KERN_INFO "raid0 : nb_zone is %d.\n", nb_zone);
 323
 324         printk(KERN_INFO "raid0 : Allocating %zu bytes for hash.\n",
 325                                 nb_zone*sizeof(struct strip_zone*));
 326         conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
 327         if (!conf->hash_table)
 328                 goto out_free_conf;
 329         sectors = conf->strip_zone[cur].sectors;
 330
 331         conf->hash_table[0] = conf->strip_zone + cur;
 332         for (i=1; i< nb_zone; i++) {
 333                 while (sectors <= conf->spacing) {
 334                         cur++;
 335                         sectors += conf->strip_zone[cur].sectors;
 336                 }
 337                 sectors -= conf->spacing;
 338                 conf->hash_table[i] = conf->strip_zone + cur;
 339         }
 340         if (conf->sector_shift) {
 341                 conf->spacing >>= conf->sector_shift;
 342                 /* round spacing up so when we divide by it, we
 343                  * err on the side of too-low, which is safest
 344                  */
 345                 conf->spacing++;
 346         }
 347
 348         /* calculate the max read-ahead size.
 349          * For read-ahead of large files to be effective, we need to
 350          * readahead at least twice a whole stripe. i.e. number of devices
 351          * multiplied by chunk size times 2.
 352          * If an individual device has an ra_pages greater than the
 353          * chunk size, then we will not drive that device as hard as it
 354          * wants.  We consider this a configuration error: a larger
 355          * chunksize should be used in that case.
 356          */
 357         {
 358                 int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
 359                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 360                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 361         }
 362
 363
 364         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 365         return 0;
 366
 367 out_free_conf:
 368         kfree(conf->strip_zone);
 369         kfree(conf->devlist);
 370         kfree(conf);
 371         mddev->private = NULL;
 372 out:
 373         return -ENOMEM;
 374 }
 375
 376 static int raid0_stop (mddev_t *mddev)
 377 {
 378         raid0_conf_t *conf = mddev_to_conf(mddev);
 379
 380         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 381         kfree(conf->hash_table);
 382         conf->hash_table = NULL;
 383         kfree(conf->strip_zone);
 384         conf->strip_zone = NULL;
 385         kfree(conf);
 386         mddev->private = NULL;
 387
 388         return 0;
 389 }
 390
 391 static int raid0_make_request (struct request_queue *q, struct bio *bio)
 392 {
 393         mddev_t *mddev = q->queuedata;
 394         unsigned int sect_in_chunk, chunksect_bits, chunk_sects;
 395         raid0_conf_t *conf = mddev_to_conf(mddev);
 396         struct strip_zone *zone;
 397         mdk_rdev_t *tmp_dev;
 398         sector_t chunk;
 399         sector_t sector, rsect;
 400         const int rw = bio_data_dir(bio);
 401         int cpu;
 402
 403         if (unlikely(bio_barrier(bio))) {
 404                 bio_endio(bio, -EOPNOTSUPP);
 405                 return 0;
 406         }
 407
 408         cpu = part_stat_lock();
 409         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 410         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 411                       bio_sectors(bio));
 412         part_stat_unlock();
 413
 414         chunk_sects = mddev->chunk_size >> 9;
 415         chunksect_bits = ffz(~chunk_sects);
 416         sector = bio->bi_sector;
 417
 418         if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
 419                 struct bio_pair *bp;
 420                 /* Sanity check -- queue functions should prevent this happening */
 421                 if (bio->bi_vcnt != 1 ||
 422                     bio->bi_idx != 0)
 423                         goto bad_map;
 424                 /* This is a one page bio that upper layers
 425                  * refuse to split for us, so we need to split it.
 426                  */
 427                 bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
 428                 if (raid0_make_request(q, &bp->bio1))
 429                         generic_make_request(&bp->bio1);
 430                 if (raid0_make_request(q, &bp->bio2))
 431                         generic_make_request(&bp->bio2);
 432
 433                 bio_pair_release(bp);
 434                 return 0;
 435         }
 436
 437
 438         {
 439                 sector_t x = sector >> conf->sector_shift;
 440                 sector_div(x, (u32)conf->spacing);
 441                 zone = conf->hash_table[x];
 442         }
 443
 444         while (sector >= zone->zone_start + zone->sectors)
 445                 zone++;
 446
 447         sect_in_chunk = bio->bi_sector & (chunk_sects - 1);
 448
 449
 450         {
 451                 sector_t x = (sector - zone->zone_start) >> chunksect_bits;
 452
 453                 sector_div(x, zone->nb_dev);
 454                 chunk = x;
 455
 456                 x = sector >> chunksect_bits;
 457                 tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
 458         }
 459         rsect = (chunk << chunksect_bits) + zone->dev_start + sect_in_chunk;
 460
 461         bio->bi_bdev = tmp_dev->bdev;
 462         bio->bi_sector = rsect + tmp_dev->data_offset;
 463
 464         /*
 465          * Let the main block layer submit the IO and resolve recursion:
 466          */
 467         return 1;
 468
 469 bad_map:
 470         printk("raid0_make_request bug: can't convert block across chunks"
 471                 " or bigger than %dk %llu %d\n", chunk_sects / 2,
 472                 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 473
 474         bio_io_error(bio);
 475         return 0;
 476 }
 477
 478 static void raid0_status (struct seq_file *seq, mddev_t *mddev)
 479 {
 480 #undef MD_DEBUG
 481 #ifdef MD_DEBUG
 482         int j, k, h;
 483         char b[BDEVNAME_SIZE];
 484         raid0_conf_t *conf = mddev_to_conf(mddev);
 485
 486         h = 0;
 487         for (j = 0; j < conf->nr_strip_zones; j++) {
 488                 seq_printf(seq, "      z%d", j);
 489                 if (conf->hash_table[h] == conf->strip_zone+j)
 490                         seq_printf(seq, "(h%d)", h++);
 491                 seq_printf(seq, "=[");
 492                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 493                         seq_printf(seq, "%s/", bdevname(
 494                                 conf->strip_zone[j].dev[k]->bdev,b));
 495
 496                 seq_printf(seq, "] zs=%d ds=%d s=%d\n",
 497                                 conf->strip_zone[j].zone_start,
 498                                 conf->strip_zone[j].dev_start,
 499                                 conf->strip_zone[j].sectors);
 500         }
 501 #endif
 502         seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
 503         return;
 504 }
 505
 506 static struct mdk_personality raid0_personality=
 507 {
 508         .name           = "raid0",
 509         .level          = 0,
 510         .owner          = THIS_MODULE,
 511         .make_request   = raid0_make_request,
 512         .run            = raid0_run,
 513         .stop           = raid0_stop,
 514         .status         = raid0_status,
 515 };
 516
 517 static int __init raid0_init (void)
 518 {
 519         return register_md_personality (&raid0_personality);
 520 }
 521
 522 static void raid0_exit (void)
 523 {
 524         unregister_md_personality (&raid0_personality);
 525 }
 526
 527 module_init(raid0_init);
 528 module_exit(raid0_exit);
 529 MODULE_LICENSE("GPL");
 530 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 531 MODULE_ALIAS("md-raid0");
 532 MODULE_ALIAS("md-level-0");