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 <linux/slab.h>
  24 #include "md.h"
  25 #include "raid0.h"
  26
  27 static void raid0_unplug(struct request_queue *q)
  28 {
  29         mddev_t *mddev = q->queuedata;
  30         raid0_conf_t *conf = mddev->private;
  31         mdk_rdev_t **devlist = conf->devlist;
  32         int i;
  33
  34         for (i=0; i<mddev->raid_disks; i++) {
  35                 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
  36
  37                 blk_unplug(r_queue);
  38         }
  39 }
  40
  41 static int raid0_congested(void *data, int bits)
  42 {
  43         mddev_t *mddev = data;
  44         raid0_conf_t *conf = mddev->private;
  45         mdk_rdev_t **devlist = conf->devlist;
  46         int i, ret = 0;
  47
  48         if (mddev_congested(mddev, bits))
  49                 return 1;
  50
  51         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
  52                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  53
  54                 ret |= bdi_congested(&q->backing_dev_info, bits);
  55         }
  56         return ret;
  57 }
  58
  59 /*
  60  * inform the user of the raid configuration
  61 */
  62 static void dump_zones(mddev_t *mddev)
  63 {
  64         int j, k, h;
  65         sector_t zone_size = 0;
  66         sector_t zone_start = 0;
  67         char b[BDEVNAME_SIZE];
  68         raid0_conf_t *conf = mddev->private;
  69         printk(KERN_INFO "******* %s configuration *********\n",
  70                 mdname(mddev));
  71         h = 0;
  72         for (j = 0; j < conf->nr_strip_zones; j++) {
  73                 printk(KERN_INFO "zone%d=[", j);
  74                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
  75                         printk("%s/",
  76                         bdevname(conf->devlist[j*mddev->raid_disks
  77                                                 + k]->bdev, b));
  78                 printk("]\n");
  79
  80                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
  81                 printk(KERN_INFO "        zone offset=%llukb "
  82                                 "device offset=%llukb size=%llukb\n",
  83                         (unsigned long long)zone_start>>1,
  84                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
  85                         (unsigned long long)zone_size>>1);
  86                 zone_start = conf->strip_zone[j].zone_end;
  87         }
  88         printk(KERN_INFO "**********************************\n\n");
  89 }
  90
  91 static int create_strip_zones(mddev_t *mddev)
  92 {
  93         int i, c, err;
  94         sector_t curr_zone_end, sectors;
  95         mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev, **dev;
  96         struct strip_zone *zone;
  97         int cnt;
  98         char b[BDEVNAME_SIZE];
  99         raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
 100
 101         if (!conf)
 102                 return -ENOMEM;
 103         list_for_each_entry(rdev1, &mddev->disks, same_set) {
 104                 printk(KERN_INFO "raid0: looking at %s\n",
 105                         bdevname(rdev1->bdev,b));
 106                 c = 0;
 107
 108                 /* round size to chunk_size */
 109                 sectors = rdev1->sectors;
 110                 sector_div(sectors, mddev->chunk_sectors);
 111                 rdev1->sectors = sectors * mddev->chunk_sectors;
 112
 113                 list_for_each_entry(rdev2, &mddev->disks, same_set) {
 114                         printk(KERN_INFO "raid0:   comparing %s(%llu)",
 115                                bdevname(rdev1->bdev,b),
 116                                (unsigned long long)rdev1->sectors);
 117                         printk(KERN_INFO " with %s(%llu)\n",
 118                                bdevname(rdev2->bdev,b),
 119                                (unsigned long long)rdev2->sectors);
 120                         if (rdev2 == rdev1) {
 121                                 printk(KERN_INFO "raid0:   END\n");
 122                                 break;
 123                         }
 124                         if (rdev2->sectors == rdev1->sectors) {
 125                                 /*
 126                                  * Not unique, don't count it as a new
 127                                  * group
 128                                  */
 129                                 printk(KERN_INFO "raid0:   EQUAL\n");
 130                                 c = 1;
 131                                 break;
 132                         }
 133                         printk(KERN_INFO "raid0:   NOT EQUAL\n");
 134                 }
 135                 if (!c) {
 136                         printk(KERN_INFO "raid0:   ==> UNIQUE\n");
 137                         conf->nr_strip_zones++;
 138                         printk(KERN_INFO "raid0: %d zones\n",
 139                                 conf->nr_strip_zones);
 140                 }
 141         }
 142         printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
 143         err = -ENOMEM;
 144         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 145                                 conf->nr_strip_zones, GFP_KERNEL);
 146         if (!conf->strip_zone)
 147                 goto abort;
 148         conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
 149                                 conf->nr_strip_zones*mddev->raid_disks,
 150                                 GFP_KERNEL);
 151         if (!conf->devlist)
 152                 goto abort;
 153
 154         /* The first zone must contain all devices, so here we check that
 155          * there is a proper alignment of slots to devices and find them all
 156          */
 157         zone = &conf->strip_zone[0];
 158         cnt = 0;
 159         smallest = NULL;
 160         dev = conf->devlist;
 161         err = -EINVAL;
 162         list_for_each_entry(rdev1, &mddev->disks, same_set) {
 163                 int j = rdev1->raid_disk;
 164
 165                 if (j < 0 || j >= mddev->raid_disks) {
 166                         printk(KERN_ERR "raid0: bad disk number %d - "
 167                                 "aborting!\n", j);
 168                         goto abort;
 169                 }
 170                 if (dev[j]) {
 171                         printk(KERN_ERR "raid0: multiple devices for %d - "
 172                                 "aborting!\n", j);
 173                         goto abort;
 174                 }
 175                 dev[j] = rdev1;
 176
 177                 disk_stack_limits(mddev->gendisk, rdev1->bdev,
 178                                   rdev1->data_offset << 9);
 179                 /* as we don't honour merge_bvec_fn, we must never risk
 180                  * violating it, so limit ->max_segments to 1, lying within
 181                  * a single page.
 182                  */
 183
 184                 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) {
 185                         blk_queue_max_segments(mddev->queue, 1);
 186                         blk_queue_segment_boundary(mddev->queue,
 187                                                    PAGE_CACHE_SIZE - 1);
 188                 }
 189                 if (!smallest || (rdev1->sectors < smallest->sectors))
 190                         smallest = rdev1;
 191                 cnt++;
 192         }
 193         if (cnt != mddev->raid_disks) {
 194                 printk(KERN_ERR "raid0: too few disks (%d of %d) - "
 195                         "aborting!\n", cnt, mddev->raid_disks);
 196                 goto abort;
 197         }
 198         zone->nb_dev = cnt;
 199         zone->zone_end = smallest->sectors * cnt;
 200
 201         curr_zone_end = zone->zone_end;
 202
 203         /* now do the other zones */
 204         for (i = 1; i < conf->nr_strip_zones; i++)
 205         {
 206                 int j;
 207
 208                 zone = conf->strip_zone + i;
 209                 dev = conf->devlist + i * mddev->raid_disks;
 210
 211                 printk(KERN_INFO "raid0: zone %d\n", i);
 212                 zone->dev_start = smallest->sectors;
 213                 smallest = NULL;
 214                 c = 0;
 215
 216                 for (j=0; j<cnt; j++) {
 217                         rdev = conf->devlist[j];
 218                         printk(KERN_INFO "raid0: checking %s ...",
 219                                 bdevname(rdev->bdev, b));
 220                         if (rdev->sectors <= zone->dev_start) {
 221                                 printk(KERN_INFO " nope.\n");
 222                                 continue;
 223                         }
 224                         printk(KERN_INFO " contained as device %d\n", c);
 225                         dev[c] = rdev;
 226                         c++;
 227                         if (!smallest || rdev->sectors < smallest->sectors) {
 228                                 smallest = rdev;
 229                                 printk(KERN_INFO "  (%llu) is smallest!.\n",
 230                                         (unsigned long long)rdev->sectors);
 231                         }
 232                 }
 233
 234                 zone->nb_dev = c;
 235                 sectors = (smallest->sectors - zone->dev_start) * c;
 236                 printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
 237                         zone->nb_dev, (unsigned long long)sectors);
 238
 239                 curr_zone_end += sectors;
 240                 zone->zone_end = curr_zone_end;
 241
 242                 printk(KERN_INFO "raid0: current zone start: %llu\n",
 243                         (unsigned long long)smallest->sectors);
 244         }
 245         mddev->queue->unplug_fn = raid0_unplug;
 246         mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 247         mddev->queue->backing_dev_info.congested_data = mddev;
 248
 249         /*
 250          * now since we have the hard sector sizes, we can make sure
 251          * chunk size is a multiple of that sector size
 252          */
 253         if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
 254                 printk(KERN_ERR "%s chunk_size of %d not valid\n",
 255                        mdname(mddev),
 256                        mddev->chunk_sectors << 9);
 257                 goto abort;
 258         }
 259
 260         blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
 261         blk_queue_io_opt(mddev->queue,
 262                          (mddev->chunk_sectors << 9) * mddev->raid_disks);
 263
 264         printk(KERN_INFO "raid0: done.\n");
 265         mddev->private = conf;
 266         return 0;
 267 abort:
 268         kfree(conf->strip_zone);
 269         kfree(conf->devlist);
 270         kfree(conf);
 271         mddev->private = NULL;
 272         return err;
 273 }
 274
 275 /**
 276  *      raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 277  *      @q: request queue
 278  *      @bvm: properties of new bio
 279  *      @biovec: the request that could be merged to it.
 280  *
 281  *      Return amount of bytes we can accept at this offset
 282  */
 283 static int raid0_mergeable_bvec(struct request_queue *q,
 284                                 struct bvec_merge_data *bvm,
 285                                 struct bio_vec *biovec)
 286 {
 287         mddev_t *mddev = q->queuedata;
 288         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 289         int max;
 290         unsigned int chunk_sectors = mddev->chunk_sectors;
 291         unsigned int bio_sectors = bvm->bi_size >> 9;
 292
 293         if (is_power_of_2(chunk_sectors))
 294                 max =  (chunk_sectors - ((sector & (chunk_sectors-1))
 295                                                 + bio_sectors)) << 9;
 296         else
 297                 max =  (chunk_sectors - (sector_div(sector, chunk_sectors)
 298                                                 + bio_sectors)) << 9;
 299         if (max < 0) max = 0; /* bio_add cannot handle a negative return */
 300         if (max <= biovec->bv_len && bio_sectors == 0)
 301                 return biovec->bv_len;
 302         else
 303                 return max;
 304 }
 305
 306 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 307 {
 308         sector_t array_sectors = 0;
 309         mdk_rdev_t *rdev;
 310
 311         WARN_ONCE(sectors || raid_disks,
 312                   "%s does not support generic reshape\n", __func__);
 313
 314         list_for_each_entry(rdev, &mddev->disks, same_set)
 315                 array_sectors += rdev->sectors;
 316
 317         return array_sectors;
 318 }
 319
 320 static int raid0_run(mddev_t *mddev)
 321 {
 322         int ret;
 323
 324         if (mddev->chunk_sectors == 0) {
 325                 printk(KERN_ERR "md/raid0: chunk size must be set.\n");
 326                 return -EINVAL;
 327         }
 328         if (md_check_no_bitmap(mddev))
 329                 return -EINVAL;
 330         blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
 331         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 332
 333         ret = create_strip_zones(mddev);
 334         if (ret < 0)
 335                 return ret;
 336
 337         /* calculate array device size */
 338         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 339
 340         printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
 341                 (unsigned long long)mddev->array_sectors);
 342         /* calculate the max read-ahead size.
 343          * For read-ahead of large files to be effective, we need to
 344          * readahead at least twice a whole stripe. i.e. number of devices
 345          * multiplied by chunk size times 2.
 346          * If an individual device has an ra_pages greater than the
 347          * chunk size, then we will not drive that device as hard as it
 348          * wants.  We consider this a configuration error: a larger
 349          * chunksize should be used in that case.
 350          */
 351         {
 352                 int stripe = mddev->raid_disks *
 353                         (mddev->chunk_sectors << 9) / PAGE_SIZE;
 354                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 355                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 356         }
 357
 358         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 359         dump_zones(mddev);
 360         md_integrity_register(mddev);
 361         return 0;
 362 }
 363
 364 static int raid0_stop(mddev_t *mddev)
 365 {
 366         raid0_conf_t *conf = mddev->private;
 367
 368         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 369         kfree(conf->strip_zone);
 370         kfree(conf->devlist);
 371         kfree(conf);
 372         mddev->private = NULL;
 373         return 0;
 374 }
 375
 376 /* Find the zone which holds a particular offset
 377  * Update *sectorp to be an offset in that zone
 378  */
 379 static struct strip_zone *find_zone(struct raid0_private_data *conf,
 380                                     sector_t *sectorp)
 381 {
 382         int i;
 383         struct strip_zone *z = conf->strip_zone;
 384         sector_t sector = *sectorp;
 385
 386         for (i = 0; i < conf->nr_strip_zones; i++)
 387                 if (sector < z[i].zone_end) {
 388                         if (i)
 389                                 *sectorp = sector - z[i-1].zone_end;
 390                         return z + i;
 391                 }
 392         BUG();
 393 }
 394
 395 /*
 396  * remaps the bio to the target device. we separate two flows.
 397  * power 2 flow and a general flow for the sake of perfromance
 398 */
 399 static mdk_rdev_t *map_sector(mddev_t *mddev, struct strip_zone *zone,
 400                                 sector_t sector, sector_t *sector_offset)
 401 {
 402         unsigned int sect_in_chunk;
 403         sector_t chunk;
 404         raid0_conf_t *conf = mddev->private;
 405         unsigned int chunk_sects = mddev->chunk_sectors;
 406
 407         if (is_power_of_2(chunk_sects)) {
 408                 int chunksect_bits = ffz(~chunk_sects);
 409                 /* find the sector offset inside the chunk */
 410                 sect_in_chunk  = sector & (chunk_sects - 1);
 411                 sector >>= chunksect_bits;
 412                 /* chunk in zone */
 413                 chunk = *sector_offset;
 414                 /* quotient is the chunk in real device*/
 415                 sector_div(chunk, zone->nb_dev << chunksect_bits);
 416         } else{
 417                 sect_in_chunk = sector_div(sector, chunk_sects);
 418                 chunk = *sector_offset;
 419                 sector_div(chunk, chunk_sects * zone->nb_dev);
 420         }
 421         /*
 422         *  position the bio over the real device
 423         *  real sector = chunk in device + starting of zone
 424         *       + the position in the chunk
 425         */
 426         *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
 427         return conf->devlist[(zone - conf->strip_zone)*mddev->raid_disks
 428                              + sector_div(sector, zone->nb_dev)];
 429 }
 430
 431 /*
 432  * Is io distribute over 1 or more chunks ?
 433 */
 434 static inline int is_io_in_chunk_boundary(mddev_t *mddev,
 435                         unsigned int chunk_sects, struct bio *bio)
 436 {
 437         if (likely(is_power_of_2(chunk_sects))) {
 438                 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
 439                                         + (bio->bi_size >> 9));
 440         } else{
 441                 sector_t sector = bio->bi_sector;
 442                 return chunk_sects >= (sector_div(sector, chunk_sects)
 443                                                 + (bio->bi_size >> 9));
 444         }
 445 }
 446
 447 static int raid0_make_request(struct request_queue *q, struct bio *bio)
 448 {
 449         mddev_t *mddev = q->queuedata;
 450         unsigned int chunk_sects;
 451         sector_t sector_offset;
 452         struct strip_zone *zone;
 453         mdk_rdev_t *tmp_dev;
 454         const int rw = bio_data_dir(bio);
 455         int cpu;
 456
 457         if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
 458                 md_barrier_request(mddev, bio);
 459                 return 0;
 460         }
 461
 462         cpu = part_stat_lock();
 463         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 464         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 465                       bio_sectors(bio));
 466         part_stat_unlock();
 467
 468         chunk_sects = mddev->chunk_sectors;
 469         if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
 470                 sector_t sector = bio->bi_sector;
 471                 struct bio_pair *bp;
 472                 /* Sanity check -- queue functions should prevent this happening */
 473                 if (bio->bi_vcnt != 1 ||
 474                     bio->bi_idx != 0)
 475                         goto bad_map;
 476                 /* This is a one page bio that upper layers
 477                  * refuse to split for us, so we need to split it.
 478                  */
 479                 if (likely(is_power_of_2(chunk_sects)))
 480                         bp = bio_split(bio, chunk_sects - (sector &
 481                                                            (chunk_sects-1)));
 482                 else
 483                         bp = bio_split(bio, chunk_sects -
 484                                        sector_div(sector, chunk_sects));
 485                 if (raid0_make_request(q, &bp->bio1))
 486                         generic_make_request(&bp->bio1);
 487                 if (raid0_make_request(q, &bp->bio2))
 488                         generic_make_request(&bp->bio2);
 489
 490                 bio_pair_release(bp);
 491                 return 0;
 492         }
 493
 494         sector_offset = bio->bi_sector;
 495         zone =  find_zone(mddev->private, &sector_offset);
 496         tmp_dev = map_sector(mddev, zone, bio->bi_sector,
 497                              &sector_offset);
 498         bio->bi_bdev = tmp_dev->bdev;
 499         bio->bi_sector = sector_offset + zone->dev_start +
 500                 tmp_dev->data_offset;
 501         /*
 502          * Let the main block layer submit the IO and resolve recursion:
 503          */
 504         return 1;
 505
 506 bad_map:
 507         printk("raid0_make_request bug: can't convert block across chunks"
 508                 " or bigger than %dk %llu %d\n", chunk_sects / 2,
 509                 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 510
 511         bio_io_error(bio);
 512         return 0;
 513 }
 514
 515 static void raid0_status(struct seq_file *seq, mddev_t *mddev)
 516 {
 517 #undef MD_DEBUG
 518 #ifdef MD_DEBUG
 519         int j, k, h;
 520         char b[BDEVNAME_SIZE];
 521         raid0_conf_t *conf = mddev->private;
 522
 523         sector_t zone_size;
 524         sector_t zone_start = 0;
 525         h = 0;
 526
 527         for (j = 0; j < conf->nr_strip_zones; j++) {
 528                 seq_printf(seq, "      z%d", j);
 529                 seq_printf(seq, "=[");
 530                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 531                         seq_printf(seq, "%s/", bdevname(
 532                                 conf->devlist[j*mddev->raid_disks + k]
 533                                                 ->bdev, b));
 534
 535                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
 536                 seq_printf(seq, "] ze=%lld ds=%lld s=%lld\n",
 537                         (unsigned long long)zone_start>>1,
 538                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
 539                         (unsigned long long)zone_size>>1);
 540                 zone_start = conf->strip_zone[j].zone_end;
 541         }
 542 #endif
 543         seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
 544         return;
 545 }
 546
 547 static struct mdk_personality raid0_personality=
 548 {
 549         .name           = "raid0",
 550         .level          = 0,
 551         .owner          = THIS_MODULE,
 552         .make_request   = raid0_make_request,
 553         .run            = raid0_run,
 554         .stop           = raid0_stop,
 555         .status         = raid0_status,
 556         .size           = raid0_size,
 557 };
 558
 559 static int __init raid0_init (void)
 560 {
 561         return register_md_personality (&raid0_personality);
 562 }
 563
 564 static void raid0_exit (void)
 565 {
 566         unregister_md_personality (&raid0_personality);
 567 }
 568
 569 module_init(raid0_init);
 570 module_exit(raid0_exit);
 571 MODULE_LICENSE("GPL");
 572 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
 573 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 574 MODULE_ALIAS("md-raid0");
 575 MODULE_ALIAS("md-level-0");