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