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