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/seq_file.h>
  23 #include "md.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->sectors);
  80                         printk(KERN_INFO " with %s(%llu)\n",
  81                                bdevname(rdev2->bdev,b),
  82                                (unsigned long long)rdev2->sectors);
  83                         if (rdev2 == rdev1) {
  84                                 printk(KERN_INFO "raid0:   END\n");
  85                                 break;
  86                         }
  87                         if (rdev2->sectors == rdev1->sectors) {
  88                                 /*
  89                                  * Not unique, don't count it as a new
  90                                  * group
  91                                  */
  92                                 printk(KERN_INFO "raid0:   EQUAL\n");
  93                                 c = 1;
  94                                 break;
  95                         }
  96                         printk(KERN_INFO "raid0:   NOT EQUAL\n");
  97                 }
  98                 if (!c) {
  99                         printk(KERN_INFO "raid0:   ==> UNIQUE\n");
 100                         conf->nr_strip_zones++;
 101                         printk(KERN_INFO "raid0: %d zones\n",
 102                                 conf->nr_strip_zones);
 103                 }
 104         }
 105         printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
 106
 107         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 108                                 conf->nr_strip_zones, GFP_KERNEL);
 109         if (!conf->strip_zone)
 110                 return 1;
 111         conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
 112                                 conf->nr_strip_zones*mddev->raid_disks,
 113                                 GFP_KERNEL);
 114         if (!conf->devlist)
 115                 return 1;
 116
 117         /* The first zone must contain all devices, so here we check that
 118          * there is a proper alignment of slots to devices and find them all
 119          */
 120         zone = &conf->strip_zone[0];
 121         cnt = 0;
 122         smallest = NULL;
 123         zone->dev = conf->devlist;
 124         list_for_each_entry(rdev1, &mddev->disks, same_set) {
 125                 int j = rdev1->raid_disk;
 126
 127                 if (j < 0 || j >= mddev->raid_disks) {
 128                         printk(KERN_ERR "raid0: bad disk number %d - "
 129                                 "aborting!\n", j);
 130                         goto abort;
 131                 }
 132                 if (zone->dev[j]) {
 133                         printk(KERN_ERR "raid0: multiple devices for %d - "
 134                                 "aborting!\n", j);
 135                         goto abort;
 136                 }
 137                 zone->dev[j] = rdev1;
 138
 139                 blk_queue_stack_limits(mddev->queue,
 140                                        rdev1->bdev->bd_disk->queue);
 141                 /* as we don't honour merge_bvec_fn, we must never risk
 142                  * violating it, so limit ->max_sector to one PAGE, as
 143                  * a one page request is never in violation.
 144                  */
 145
 146                 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
 147                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
 148                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
 149
 150                 if (!smallest || (rdev1->sectors < smallest->sectors))
 151                         smallest = rdev1;
 152                 cnt++;
 153         }
 154         if (cnt != mddev->raid_disks) {
 155                 printk(KERN_ERR "raid0: too few disks (%d of %d) - "
 156                         "aborting!\n", cnt, mddev->raid_disks);
 157                 goto abort;
 158         }
 159         zone->nb_dev = cnt;
 160         zone->sectors = smallest->sectors * cnt;
 161         zone->zone_start = 0;
 162
 163         current_start = smallest->sectors;
 164         curr_zone_start = zone->sectors;
 165
 166         /* now do the other zones */
 167         for (i = 1; i < conf->nr_strip_zones; i++)
 168         {
 169                 zone = conf->strip_zone + i;
 170                 zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks;
 171
 172                 printk(KERN_INFO "raid0: zone %d\n", i);
 173                 zone->dev_start = current_start;
 174                 smallest = NULL;
 175                 c = 0;
 176
 177                 for (j=0; j<cnt; j++) {
 178                         char b[BDEVNAME_SIZE];
 179                         rdev = conf->strip_zone[0].dev[j];
 180                         printk(KERN_INFO "raid0: checking %s ...",
 181                                 bdevname(rdev->bdev, b));
 182                         if (rdev->sectors <= current_start) {
 183                                 printk(KERN_INFO " nope.\n");
 184                                 continue;
 185                         }
 186                         printk(KERN_INFO " contained as device %d\n", c);
 187                         zone->dev[c] = rdev;
 188                         c++;
 189                         if (!smallest || rdev->sectors < smallest->sectors) {
 190                                 smallest = rdev;
 191                                 printk(KERN_INFO "  (%llu) is smallest!.\n",
 192                                         (unsigned long long)rdev->sectors);
 193                         }
 194                 }
 195
 196                 zone->nb_dev = c;
 197                 zone->sectors = (smallest->sectors - current_start) * c;
 198                 printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
 199                         zone->nb_dev, (unsigned long long)zone->sectors);
 200
 201                 zone->zone_start = curr_zone_start;
 202                 curr_zone_start += zone->sectors;
 203
 204                 current_start = smallest->sectors;
 205                 printk(KERN_INFO "raid0: current zone start: %llu\n",
 206                         (unsigned long long)current_start);
 207         }
 208
 209         /* Now find appropriate hash spacing.
 210          * We want a number which causes most hash entries to cover
 211          * at most two strips, but the hash table must be at most
 212          * 1 PAGE.  We choose the smallest strip, or contiguous collection
 213          * of strips, that has big enough size.  We never consider the last
 214          * strip though as it's size has no bearing on the efficacy of the hash
 215          * table.
 216          */
 217         conf->spacing = curr_zone_start;
 218         min_spacing = curr_zone_start;
 219         sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
 220         for (i=0; i < conf->nr_strip_zones-1; i++) {
 221                 sector_t s = 0;
 222                 for (j = i; j < conf->nr_strip_zones - 1 &&
 223                                 s < min_spacing; j++)
 224                         s += conf->strip_zone[j].sectors;
 225                 if (s >= min_spacing && s < conf->spacing)
 226                         conf->spacing = s;
 227         }
 228
 229         mddev->queue->unplug_fn = raid0_unplug;
 230
 231         mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 232         mddev->queue->backing_dev_info.congested_data = mddev;
 233
 234         printk(KERN_INFO "raid0: done.\n");
 235         return 0;
 236  abort:
 237         return 1;
 238 }
 239
 240 /**
 241  *      raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 242  *      @q: request queue
 243  *      @bvm: properties of new bio
 244  *      @biovec: the request that could be merged to it.
 245  *
 246  *      Return amount of bytes we can accept at this offset
 247  */
 248 static int raid0_mergeable_bvec(struct request_queue *q,
 249                                 struct bvec_merge_data *bvm,
 250                                 struct bio_vec *biovec)
 251 {
 252         mddev_t *mddev = q->queuedata;
 253         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 254         int max;
 255         unsigned int chunk_sectors = mddev->chunk_size >> 9;
 256         unsigned int bio_sectors = bvm->bi_size >> 9;
 257
 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;
 262         else
 263                 return max;
 264 }
 265
 266 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 267 {
 268         sector_t array_sectors = 0;
 269         mdk_rdev_t *rdev;
 270
 271         WARN_ONCE(sectors || raid_disks,
 272                   "%s does not support generic reshape\n", __func__);
 273
 274         list_for_each_entry(rdev, &mddev->disks, same_set)
 275                 array_sectors += rdev->sectors;
 276
 277         return array_sectors;
 278 }
 279
 280 static int raid0_run (mddev_t *mddev)
 281 {
 282         unsigned  cur=0, i=0, nb_zone;
 283         s64 sectors;
 284         raid0_conf_t *conf;
 285
 286         if (mddev->chunk_size == 0) {
 287                 printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
 288                 return -EINVAL;
 289         }
 290         printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
 291                mdname(mddev),
 292                mddev->chunk_size >> 9,
 293                (mddev->chunk_size>>1)-1);
 294         blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
 295         blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
 296         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 297
 298         conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
 299         if (!conf)
 300                 goto out;
 301         mddev->private = (void *)conf;
 302
 303         conf->strip_zone = NULL;
 304         conf->devlist = NULL;
 305         if (create_strip_zones (mddev))
 306                 goto out_free_conf;
 307
 308         /* calculate array device size */
 309         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 310
 311         printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
 312                 (unsigned long long)mddev->array_sectors);
 313         printk(KERN_INFO "raid0 : conf->spacing is %llu sectors.\n",
 314                 (unsigned long long)conf->spacing);
 315         {
 316                 sector_t s = raid0_size(mddev, 0, 0);
 317                 sector_t space = conf->spacing;
 318                 int round;
 319                 conf->sector_shift = 0;
 320                 if (sizeof(sector_t) > sizeof(u32)) {
 321                         /*shift down space and s so that sector_div will work */
 322                         while (space > (sector_t) (~(u32)0)) {
 323                                 s >>= 1;
 324                                 space >>= 1;
 325                                 s += 1; /* force round-up */
 326                                 conf->sector_shift++;
 327                         }
 328                 }
 329                 round = sector_div(s, (u32)space) ? 1 : 0;
 330                 nb_zone = s + round;
 331         }
 332         printk(KERN_INFO "raid0 : nb_zone is %d.\n", nb_zone);
 333
 334         printk(KERN_INFO "raid0 : Allocating %zu bytes for hash.\n",
 335                                 nb_zone*sizeof(struct strip_zone*));
 336         conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
 337         if (!conf->hash_table)
 338                 goto out_free_conf;
 339         sectors = conf->strip_zone[cur].sectors;
 340
 341         conf->hash_table[0] = conf->strip_zone + cur;
 342         for (i=1; i< nb_zone; i++) {
 343                 while (sectors <= conf->spacing) {
 344                         cur++;
 345                         sectors += conf->strip_zone[cur].sectors;
 346                 }
 347                 sectors -= conf->spacing;
 348                 conf->hash_table[i] = conf->strip_zone + cur;
 349         }
 350         if (conf->sector_shift) {
 351                 conf->spacing >>= conf->sector_shift;
 352                 /* round spacing up so when we divide by it, we
 353                  * err on the side of too-low, which is safest
 354                  */
 355                 conf->spacing++;
 356         }
 357
 358         /* calculate the max read-ahead size.
 359          * For read-ahead of large files to be effective, we need to
 360          * readahead at least twice a whole stripe. i.e. number of devices
 361          * multiplied by chunk size times 2.
 362          * If an individual device has an ra_pages greater than the
 363          * chunk size, then we will not drive that device as hard as it
 364          * wants.  We consider this a configuration error: a larger
 365          * chunksize should be used in that case.
 366          */
 367         {
 368                 int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
 369                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 370                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 371         }
 372
 373
 374         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 375         return 0;
 376
 377 out_free_conf:
 378         kfree(conf->strip_zone);
 379         kfree(conf->devlist);
 380         kfree(conf);
 381         mddev->private = NULL;
 382 out:
 383         return -ENOMEM;
 384 }
 385
 386 static int raid0_stop (mddev_t *mddev)
 387 {
 388         raid0_conf_t *conf = mddev_to_conf(mddev);
 389
 390         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 391         kfree(conf->hash_table);
 392         conf->hash_table = NULL;
 393         kfree(conf->strip_zone);
 394         conf->strip_zone = NULL;
 395         kfree(conf);
 396         mddev->private = NULL;
 397
 398         return 0;
 399 }
 400
 401 static int raid0_make_request (struct request_queue *q, struct bio *bio)
 402 {
 403         mddev_t *mddev = q->queuedata;
 404         unsigned int sect_in_chunk, chunksect_bits, chunk_sects;
 405         raid0_conf_t *conf = mddev_to_conf(mddev);
 406         struct strip_zone *zone;
 407         mdk_rdev_t *tmp_dev;
 408         sector_t chunk;
 409         sector_t sector, rsect;
 410         const int rw = bio_data_dir(bio);
 411         int cpu;
 412
 413         if (unlikely(bio_barrier(bio))) {
 414                 bio_endio(bio, -EOPNOTSUPP);
 415                 return 0;
 416         }
 417
 418         cpu = part_stat_lock();
 419         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 420         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 421                       bio_sectors(bio));
 422         part_stat_unlock();
 423
 424         chunk_sects = mddev->chunk_size >> 9;
 425         chunksect_bits = ffz(~chunk_sects);
 426         sector = bio->bi_sector;
 427
 428         if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
 429                 struct bio_pair *bp;
 430                 /* Sanity check -- queue functions should prevent this happening */
 431                 if (bio->bi_vcnt != 1 ||
 432                     bio->bi_idx != 0)
 433                         goto bad_map;
 434                 /* This is a one page bio that upper layers
 435                  * refuse to split for us, so we need to split it.
 436                  */
 437                 bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
 438                 if (raid0_make_request(q, &bp->bio1))
 439                         generic_make_request(&bp->bio1);
 440                 if (raid0_make_request(q, &bp->bio2))
 441                         generic_make_request(&bp->bio2);
 442
 443                 bio_pair_release(bp);
 444                 return 0;
 445         }
 446
 447
 448         {
 449                 sector_t x = sector >> conf->sector_shift;
 450                 sector_div(x, (u32)conf->spacing);
 451                 zone = conf->hash_table[x];
 452         }
 453
 454         while (sector >= zone->zone_start + zone->sectors)
 455                 zone++;
 456
 457         sect_in_chunk = bio->bi_sector & (chunk_sects - 1);
 458
 459
 460         {
 461                 sector_t x = (sector - zone->zone_start) >> chunksect_bits;
 462
 463                 sector_div(x, zone->nb_dev);
 464                 chunk = x;
 465
 466                 x = sector >> chunksect_bits;
 467                 tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
 468         }
 469         rsect = (chunk << chunksect_bits) + zone->dev_start + sect_in_chunk;
 470
 471         bio->bi_bdev = tmp_dev->bdev;
 472         bio->bi_sector = rsect + tmp_dev->data_offset;
 473
 474         /*
 475          * Let the main block layer submit the IO and resolve recursion:
 476          */
 477         return 1;
 478
 479 bad_map:
 480         printk("raid0_make_request bug: can't convert block across chunks"
 481                 " or bigger than %dk %llu %d\n", chunk_sects / 2,
 482                 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 483
 484         bio_io_error(bio);
 485         return 0;
 486 }
 487
 488 static void raid0_status (struct seq_file *seq, mddev_t *mddev)
 489 {
 490 #undef MD_DEBUG
 491 #ifdef MD_DEBUG
 492         int j, k, h;
 493         char b[BDEVNAME_SIZE];
 494         raid0_conf_t *conf = mddev_to_conf(mddev);
 495
 496         h = 0;
 497         for (j = 0; j < conf->nr_strip_zones; j++) {
 498                 seq_printf(seq, "      z%d", j);
 499                 if (conf->hash_table[h] == conf->strip_zone+j)
 500                         seq_printf(seq, "(h%d)", h++);
 501                 seq_printf(seq, "=[");
 502                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 503                         seq_printf(seq, "%s/", bdevname(
 504                                 conf->strip_zone[j].dev[k]->bdev,b));
 505
 506                 seq_printf(seq, "] zs=%d ds=%d s=%d\n",
 507                                 conf->strip_zone[j].zone_start,
 508                                 conf->strip_zone[j].dev_start,
 509                                 conf->strip_zone[j].sectors);
 510         }
 511 #endif
 512         seq_printf(seq, " %dk chunks", mddev->chunk_size/1024);
 513         return;
 514 }
 515
 516 static struct mdk_personality raid0_personality=
 517 {
 518         .name           = "raid0",
 519         .level          = 0,
 520         .owner          = THIS_MODULE,
 521         .make_request   = raid0_make_request,
 522         .run            = raid0_run,
 523         .stop           = raid0_stop,
 524         .status         = raid0_status,
 525         .size           = raid0_size,
 526 };
 527
 528 static int __init raid0_init (void)
 529 {
 530         return register_md_personality (&raid0_personality);
 531 }
 532
 533 static void raid0_exit (void)
 534 {
 535         unregister_md_personality (&raid0_personality);
 536 }
 537
 538 module_init(raid0_init);
 539 module_exit(raid0_exit);
 540 MODULE_LICENSE("GPL");
 541 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 542 MODULE_ALIAS("md-raid0");
 543 MODULE_ALIAS("md-level-0");