drivers/md/raid0.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3    raid0.c : Multiple Devices driver for Linux
   4              Copyright (C) 1994-96 Marc ZYNGIER
   5              <zyngier@ufr-info-p7.ibp.fr> or
   6              <maz@gloups.fdn.fr>
   7              Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
   8
   9    RAID-0 management functions.
  10
  11 */
  12
  13 #include <linux/blkdev.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/module.h>
  16 #include <linux/slab.h>
  17 #include <trace/events/block.h>
  18 #include "md.h"
  19 #include "raid0.h"
  20 #include "raid5.h"
  21
  22 static int default_layout = 0;
  23 module_param(default_layout, int, 0644);
  24
  25 #define UNSUPPORTED_MDDEV_FLAGS         \
  26         ((1L << MD_HAS_JOURNAL) |       \
  27          (1L << MD_JOURNAL_CLEAN) |     \
  28          (1L << MD_FAILFAST_SUPPORTED) |\
  29          (1L << MD_HAS_PPL) |           \
  30          (1L << MD_HAS_MULTIPLE_PPLS))
  31
  32 /*
  33  * inform the user of the raid configuration
  34 */
  35 static void dump_zones(struct mddev *mddev)
  36 {
  37         int j, k;
  38         sector_t zone_size = 0;
  39         sector_t zone_start = 0;
  40         struct r0conf *conf = mddev->private;
  41         int raid_disks = conf->strip_zone[0].nb_dev;
  42         pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
  43                  mdname(mddev),
  44                  conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
  45         for (j = 0; j < conf->nr_strip_zones; j++) {
  46                 char line[200];
  47                 int len = 0;
  48
  49                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
  50                         len += scnprintf(line+len, 200-len, "%s%pg", k?"/":"",
  51                                 conf->devlist[j * raid_disks + k]->bdev);
  52                 pr_debug("md: zone%d=[%s]\n", j, line);
  53
  54                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
  55                 pr_debug("      zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
  56                         (unsigned long long)zone_start>>1,
  57                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
  58                         (unsigned long long)zone_size>>1);
  59                 zone_start = conf->strip_zone[j].zone_end;
  60         }
  61 }
  62
  63 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
  64 {
  65         int i, c, err;
  66         sector_t curr_zone_end, sectors;
  67         struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
  68         struct strip_zone *zone;
  69         int cnt;
  70         struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
  71         unsigned blksize = 512;
  72
  73         *private_conf = ERR_PTR(-ENOMEM);
  74         if (!conf)
  75                 return -ENOMEM;
  76         rdev_for_each(rdev1, mddev) {
  77                 pr_debug("md/raid0:%s: looking at %pg\n",
  78                          mdname(mddev),
  79                          rdev1->bdev);
  80                 c = 0;
  81
  82                 /* round size to chunk_size */
  83                 sectors = rdev1->sectors;
  84                 sector_div(sectors, mddev->chunk_sectors);
  85                 rdev1->sectors = sectors * mddev->chunk_sectors;
  86
  87                 blksize = max(blksize, queue_logical_block_size(
  88                                       rdev1->bdev->bd_disk->queue));
  89
  90                 rdev_for_each(rdev2, mddev) {
  91                         pr_debug("md/raid0:%s:   comparing %pg(%llu)"
  92                                  " with %pg(%llu)\n",
  93                                  mdname(mddev),
  94                                  rdev1->bdev,
  95                                  (unsigned long long)rdev1->sectors,
  96                                  rdev2->bdev,
  97                                  (unsigned long long)rdev2->sectors);
  98                         if (rdev2 == rdev1) {
  99                                 pr_debug("md/raid0:%s:   END\n",
 100                                          mdname(mddev));
 101                                 break;
 102                         }
 103                         if (rdev2->sectors == rdev1->sectors) {
 104                                 /*
 105                                  * Not unique, don't count it as a new
 106                                  * group
 107                                  */
 108                                 pr_debug("md/raid0:%s:   EQUAL\n",
 109                                          mdname(mddev));
 110                                 c = 1;
 111                                 break;
 112                         }
 113                         pr_debug("md/raid0:%s:   NOT EQUAL\n",
 114                                  mdname(mddev));
 115                 }
 116                 if (!c) {
 117                         pr_debug("md/raid0:%s:   ==> UNIQUE\n",
 118                                  mdname(mddev));
 119                         conf->nr_strip_zones++;
 120                         pr_debug("md/raid0:%s: %d zones\n",
 121                                  mdname(mddev), conf->nr_strip_zones);
 122                 }
 123         }
 124         pr_debug("md/raid0:%s: FINAL %d zones\n",
 125                  mdname(mddev), conf->nr_strip_zones);
 126
 127         /*
 128          * now since we have the hard sector sizes, we can make sure
 129          * chunk size is a multiple of that sector size
 130          */
 131         if ((mddev->chunk_sectors << 9) % blksize) {
 132                 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
 133                         mdname(mddev),
 134                         mddev->chunk_sectors << 9, blksize);
 135                 err = -EINVAL;
 136                 goto abort;
 137         }
 138
 139         err = -ENOMEM;
 140         conf->strip_zone = kcalloc(conf->nr_strip_zones,
 141                                    sizeof(struct strip_zone),
 142                                    GFP_KERNEL);
 143         if (!conf->strip_zone)
 144                 goto abort;
 145         conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
 146                                             conf->nr_strip_zones,
 147                                             mddev->raid_disks),
 148                                 GFP_KERNEL);
 149         if (!conf->devlist)
 150                 goto abort;
 151
 152         /* The first zone must contain all devices, so here we check that
 153          * there is a proper alignment of slots to devices and find them all
 154          */
 155         zone = &conf->strip_zone[0];
 156         cnt = 0;
 157         smallest = NULL;
 158         dev = conf->devlist;
 159         err = -EINVAL;
 160         rdev_for_each(rdev1, mddev) {
 161                 int j = rdev1->raid_disk;
 162
 163                 if (mddev->level == 10) {
 164                         /* taking over a raid10-n2 array */
 165                         j /= 2;
 166                         rdev1->new_raid_disk = j;
 167                 }
 168
 169                 if (mddev->level == 1) {
 170                         /* taiking over a raid1 array-
 171                          * we have only one active disk
 172                          */
 173                         j = 0;
 174                         rdev1->new_raid_disk = j;
 175                 }
 176
 177                 if (j < 0) {
 178                         pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
 179                                 mdname(mddev));
 180                         goto abort;
 181                 }
 182                 if (j >= mddev->raid_disks) {
 183                         pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
 184                                 mdname(mddev), j);
 185                         goto abort;
 186                 }
 187                 if (dev[j]) {
 188                         pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
 189                                 mdname(mddev), j);
 190                         goto abort;
 191                 }
 192                 dev[j] = rdev1;
 193
 194                 if (!smallest || (rdev1->sectors < smallest->sectors))
 195                         smallest = rdev1;
 196                 cnt++;
 197         }
 198         if (cnt != mddev->raid_disks) {
 199                 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
 200                         mdname(mddev), cnt, mddev->raid_disks);
 201                 goto abort;
 202         }
 203         zone->nb_dev = cnt;
 204         zone->zone_end = smallest->sectors * cnt;
 205
 206         curr_zone_end = zone->zone_end;
 207
 208         /* now do the other zones */
 209         for (i = 1; i < conf->nr_strip_zones; i++)
 210         {
 211                 int j;
 212
 213                 zone = conf->strip_zone + i;
 214                 dev = conf->devlist + i * mddev->raid_disks;
 215
 216                 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
 217                 zone->dev_start = smallest->sectors;
 218                 smallest = NULL;
 219                 c = 0;
 220
 221                 for (j=0; j<cnt; j++) {
 222                         rdev = conf->devlist[j];
 223                         if (rdev->sectors <= zone->dev_start) {
 224                                 pr_debug("md/raid0:%s: checking %pg ... nope\n",
 225                                          mdname(mddev),
 226                                          rdev->bdev);
 227                                 continue;
 228                         }
 229                         pr_debug("md/raid0:%s: checking %pg ..."
 230                                  " contained as device %d\n",
 231                                  mdname(mddev),
 232                                  rdev->bdev, c);
 233                         dev[c] = rdev;
 234                         c++;
 235                         if (!smallest || rdev->sectors < smallest->sectors) {
 236                                 smallest = rdev;
 237                                 pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
 238                                          mdname(mddev),
 239                                          (unsigned long long)rdev->sectors);
 240                         }
 241                 }
 242
 243                 zone->nb_dev = c;
 244                 sectors = (smallest->sectors - zone->dev_start) * c;
 245                 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
 246                          mdname(mddev),
 247                          zone->nb_dev, (unsigned long long)sectors);
 248
 249                 curr_zone_end += sectors;
 250                 zone->zone_end = curr_zone_end;
 251
 252                 pr_debug("md/raid0:%s: current zone start: %llu\n",
 253                          mdname(mddev),
 254                          (unsigned long long)smallest->sectors);
 255         }
 256
 257         if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) {
 258                 conf->layout = RAID0_ORIG_LAYOUT;
 259         } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
 260                    mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
 261                 conf->layout = mddev->layout;
 262         } else if (default_layout == RAID0_ORIG_LAYOUT ||
 263                    default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
 264                 conf->layout = default_layout;
 265         } else {
 266                 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n",
 267                        mdname(mddev));
 268                 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n");
 269                 err = -EOPNOTSUPP;
 270                 goto abort;
 271         }
 272
 273         if (conf->layout == RAID0_ORIG_LAYOUT) {
 274                 for (i = 1; i < conf->nr_strip_zones; i++) {
 275                         sector_t first_sector = conf->strip_zone[i-1].zone_end;
 276
 277                         sector_div(first_sector, mddev->chunk_sectors);
 278                         zone = conf->strip_zone + i;
 279                         /* disk_shift is first disk index used in the zone */
 280                         zone->disk_shift = sector_div(first_sector,
 281                                                       zone->nb_dev);
 282                 }
 283         }
 284
 285         pr_debug("md/raid0:%s: done.\n", mdname(mddev));
 286         *private_conf = conf;
 287
 288         return 0;
 289 abort:
 290         kfree(conf->strip_zone);
 291         kfree(conf->devlist);
 292         kfree(conf);
 293         *private_conf = ERR_PTR(err);
 294         return err;
 295 }
 296
 297 /* Find the zone which holds a particular offset
 298  * Update *sectorp to be an offset in that zone
 299  */
 300 static struct strip_zone *find_zone(struct r0conf *conf,
 301                                     sector_t *sectorp)
 302 {
 303         int i;
 304         struct strip_zone *z = conf->strip_zone;
 305         sector_t sector = *sectorp;
 306
 307         for (i = 0; i < conf->nr_strip_zones; i++)
 308                 if (sector < z[i].zone_end) {
 309                         if (i)
 310                                 *sectorp = sector - z[i-1].zone_end;
 311                         return z + i;
 312                 }
 313         BUG();
 314 }
 315
 316 /*
 317  * remaps the bio to the target device. we separate two flows.
 318  * power 2 flow and a general flow for the sake of performance
 319 */
 320 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
 321                                 sector_t sector, sector_t *sector_offset)
 322 {
 323         unsigned int sect_in_chunk;
 324         sector_t chunk;
 325         struct r0conf *conf = mddev->private;
 326         int raid_disks = conf->strip_zone[0].nb_dev;
 327         unsigned int chunk_sects = mddev->chunk_sectors;
 328
 329         if (is_power_of_2(chunk_sects)) {
 330                 int chunksect_bits = ffz(~chunk_sects);
 331                 /* find the sector offset inside the chunk */
 332                 sect_in_chunk  = sector & (chunk_sects - 1);
 333                 sector >>= chunksect_bits;
 334                 /* chunk in zone */
 335                 chunk = *sector_offset;
 336                 /* quotient is the chunk in real device*/
 337                 sector_div(chunk, zone->nb_dev << chunksect_bits);
 338         } else{
 339                 sect_in_chunk = sector_div(sector, chunk_sects);
 340                 chunk = *sector_offset;
 341                 sector_div(chunk, chunk_sects * zone->nb_dev);
 342         }
 343         /*
 344         *  position the bio over the real device
 345         *  real sector = chunk in device + starting of zone
 346         *       + the position in the chunk
 347         */
 348         *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
 349         return conf->devlist[(zone - conf->strip_zone)*raid_disks
 350                              + sector_div(sector, zone->nb_dev)];
 351 }
 352
 353 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
 354 {
 355         sector_t array_sectors = 0;
 356         struct md_rdev *rdev;
 357
 358         WARN_ONCE(sectors || raid_disks,
 359                   "%s does not support generic reshape\n", __func__);
 360
 361         rdev_for_each(rdev, mddev)
 362                 array_sectors += (rdev->sectors &
 363                                   ~(sector_t)(mddev->chunk_sectors-1));
 364
 365         return array_sectors;
 366 }
 367
 368 static void raid0_free(struct mddev *mddev, void *priv)
 369 {
 370         struct r0conf *conf = priv;
 371
 372         kfree(conf->strip_zone);
 373         kfree(conf->devlist);
 374         kfree(conf);
 375 }
 376
 377 static int raid0_set_limits(struct mddev *mddev)
 378 {
 379         struct queue_limits lim;
 380         int err;
 381
 382         md_init_stacking_limits(&lim);
 383         lim.max_hw_sectors = mddev->chunk_sectors;
 384         lim.max_write_zeroes_sectors = mddev->chunk_sectors;
 385         lim.io_min = mddev->chunk_sectors << 9;
 386         lim.io_opt = lim.io_min * mddev->raid_disks;
 387         lim.features |= BLK_FEAT_ATOMIC_WRITES_STACKED;
 388         err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
 389         if (err) {
 390                 queue_limits_cancel_update(mddev->gendisk->queue);
 391                 return err;
 392         }
 393         return queue_limits_set(mddev->gendisk->queue, &lim);
 394 }
 395
 396 static int raid0_run(struct mddev *mddev)
 397 {
 398         struct r0conf *conf;
 399         int ret;
 400
 401         if (mddev->chunk_sectors == 0) {
 402                 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
 403                 return -EINVAL;
 404         }
 405         if (md_check_no_bitmap(mddev))
 406                 return -EINVAL;
 407
 408         /* if private is not null, we are here after takeover */
 409         if (mddev->private == NULL) {
 410                 ret = create_strip_zones(mddev, &conf);
 411                 if (ret < 0)
 412                         return ret;
 413                 mddev->private = conf;
 414         }
 415         conf = mddev->private;
 416         if (!mddev_is_dm(mddev)) {
 417                 ret = raid0_set_limits(mddev);
 418                 if (ret)
 419                         return ret;
 420         }
 421
 422         /* calculate array device size */
 423         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 424
 425         pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
 426                  mdname(mddev),
 427                  (unsigned long long)mddev->array_sectors);
 428
 429         dump_zones(mddev);
 430
 431         return md_integrity_register(mddev);
 432 }
 433
 434 /*
 435  * Convert disk_index to the disk order in which it is read/written.
 436  *  For example, if we have 4 disks, they are numbered 0,1,2,3. If we
 437  *  write the disks starting at disk 3, then the read/write order would
 438  *  be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift()
 439  *  to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map
 440  *  to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in
 441  *  that 'output' space to understand the read/write disk ordering.
 442  */
 443 static int map_disk_shift(int disk_index, int num_disks, int disk_shift)
 444 {
 445         return ((disk_index + num_disks - disk_shift) % num_disks);
 446 }
 447
 448 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
 449 {
 450         struct r0conf *conf = mddev->private;
 451         struct strip_zone *zone;
 452         sector_t start = bio->bi_iter.bi_sector;
 453         sector_t end;
 454         unsigned int stripe_size;
 455         sector_t first_stripe_index, last_stripe_index;
 456         sector_t start_disk_offset;
 457         unsigned int start_disk_index;
 458         sector_t end_disk_offset;
 459         unsigned int end_disk_index;
 460         unsigned int disk;
 461         sector_t orig_start, orig_end;
 462
 463         orig_start = start;
 464         zone = find_zone(conf, &start);
 465
 466         if (bio_end_sector(bio) > zone->zone_end) {
 467                 struct bio *split = bio_split(bio,
 468                         zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
 469                         &mddev->bio_set);
 470
 471                 if (IS_ERR(split)) {
 472                         bio->bi_status = errno_to_blk_status(PTR_ERR(split));
 473                         bio_endio(bio);
 474                         return;
 475                 }
 476                 bio_chain(split, bio);
 477                 submit_bio_noacct(bio);
 478                 bio = split;
 479                 end = zone->zone_end;
 480         } else
 481                 end = bio_end_sector(bio);
 482
 483         orig_end = end;
 484         if (zone != conf->strip_zone)
 485                 end = end - zone[-1].zone_end;
 486
 487         /* Now start and end is the offset in zone */
 488         stripe_size = zone->nb_dev * mddev->chunk_sectors;
 489
 490         first_stripe_index = start;
 491         sector_div(first_stripe_index, stripe_size);
 492         last_stripe_index = end;
 493         sector_div(last_stripe_index, stripe_size);
 494
 495         /* In the first zone the original and alternate layouts are the same */
 496         if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) {
 497                 sector_div(orig_start, mddev->chunk_sectors);
 498                 start_disk_index = sector_div(orig_start, zone->nb_dev);
 499                 start_disk_index = map_disk_shift(start_disk_index,
 500                                                   zone->nb_dev,
 501                                                   zone->disk_shift);
 502                 sector_div(orig_end, mddev->chunk_sectors);
 503                 end_disk_index = sector_div(orig_end, zone->nb_dev);
 504                 end_disk_index = map_disk_shift(end_disk_index,
 505                                                 zone->nb_dev, zone->disk_shift);
 506         } else {
 507                 start_disk_index = (int)(start - first_stripe_index * stripe_size) /
 508                         mddev->chunk_sectors;
 509                 end_disk_index = (int)(end - last_stripe_index * stripe_size) /
 510                         mddev->chunk_sectors;
 511         }
 512         start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
 513                 mddev->chunk_sectors) +
 514                 first_stripe_index * mddev->chunk_sectors;
 515         end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
 516                 mddev->chunk_sectors) +
 517                 last_stripe_index * mddev->chunk_sectors;
 518
 519         for (disk = 0; disk < zone->nb_dev; disk++) {
 520                 sector_t dev_start, dev_end;
 521                 struct md_rdev *rdev;
 522                 int compare_disk;
 523
 524                 compare_disk = map_disk_shift(disk, zone->nb_dev,
 525                                               zone->disk_shift);
 526
 527                 if (compare_disk < start_disk_index)
 528                         dev_start = (first_stripe_index + 1) *
 529                                 mddev->chunk_sectors;
 530                 else if (compare_disk > start_disk_index)
 531                         dev_start = first_stripe_index * mddev->chunk_sectors;
 532                 else
 533                         dev_start = start_disk_offset;
 534
 535                 if (compare_disk < end_disk_index)
 536                         dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
 537                 else if (compare_disk > end_disk_index)
 538                         dev_end = last_stripe_index * mddev->chunk_sectors;
 539                 else
 540                         dev_end = end_disk_offset;
 541
 542                 if (dev_end <= dev_start)
 543                         continue;
 544
 545                 rdev = conf->devlist[(zone - conf->strip_zone) *
 546                         conf->strip_zone[0].nb_dev + disk];
 547                 md_submit_discard_bio(mddev, rdev, bio,
 548                         dev_start + zone->dev_start + rdev->data_offset,
 549                         dev_end - dev_start);
 550         }
 551         bio_endio(bio);
 552 }
 553
 554 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio)
 555 {
 556         struct r0conf *conf = mddev->private;
 557         struct strip_zone *zone;
 558         struct md_rdev *tmp_dev;
 559         sector_t bio_sector = bio->bi_iter.bi_sector;
 560         sector_t sector = bio_sector;
 561
 562         md_account_bio(mddev, &bio);
 563
 564         zone = find_zone(mddev->private, &sector);
 565         switch (conf->layout) {
 566         case RAID0_ORIG_LAYOUT:
 567                 tmp_dev = map_sector(mddev, zone, bio_sector, &sector);
 568                 break;
 569         case RAID0_ALT_MULTIZONE_LAYOUT:
 570                 tmp_dev = map_sector(mddev, zone, sector, &sector);
 571                 break;
 572         default:
 573                 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
 574                 bio_io_error(bio);
 575                 return;
 576         }
 577
 578         if (unlikely(is_rdev_broken(tmp_dev))) {
 579                 bio_io_error(bio);
 580                 md_error(mddev, tmp_dev);
 581                 return;
 582         }
 583
 584         bio_set_dev(bio, tmp_dev->bdev);
 585         bio->bi_iter.bi_sector = sector + zone->dev_start +
 586                 tmp_dev->data_offset;
 587         mddev_trace_remap(mddev, bio, bio_sector);
 588         mddev_check_write_zeroes(mddev, bio);
 589         submit_bio_noacct(bio);
 590 }
 591
 592 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
 593 {
 594         sector_t sector;
 595         unsigned chunk_sects;
 596         unsigned sectors;
 597
 598         if (unlikely(bio->bi_opf & REQ_PREFLUSH)
 599             && md_flush_request(mddev, bio))
 600                 return true;
 601
 602         if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
 603                 raid0_handle_discard(mddev, bio);
 604                 return true;
 605         }
 606
 607         sector = bio->bi_iter.bi_sector;
 608         chunk_sects = mddev->chunk_sectors;
 609
 610         sectors = chunk_sects -
 611                 (likely(is_power_of_2(chunk_sects))
 612                  ? (sector & (chunk_sects-1))
 613                  : sector_div(sector, chunk_sects));
 614
 615         if (sectors < bio_sectors(bio)) {
 616                 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
 617                                               &mddev->bio_set);
 618
 619                 if (IS_ERR(split)) {
 620                         bio->bi_status = errno_to_blk_status(PTR_ERR(split));
 621                         bio_endio(bio);
 622                         return true;
 623                 }
 624                 bio_chain(split, bio);
 625                 raid0_map_submit_bio(mddev, bio);
 626                 bio = split;
 627         }
 628
 629         raid0_map_submit_bio(mddev, bio);
 630         return true;
 631 }
 632
 633 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
 634 {
 635         seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
 636         return;
 637 }
 638
 639 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev)
 640 {
 641         if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
 642                 char *md_name = mdname(mddev);
 643
 644                 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n",
 645                         md_name, rdev->bdev);
 646         }
 647 }
 648
 649 static void *raid0_takeover_raid45(struct mddev *mddev)
 650 {
 651         struct md_rdev *rdev;
 652         struct r0conf *priv_conf;
 653
 654         if (mddev->degraded != 1) {
 655                 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
 656                         mdname(mddev),
 657                         mddev->degraded);
 658                 return ERR_PTR(-EINVAL);
 659         }
 660
 661         rdev_for_each(rdev, mddev) {
 662                 /* check slot number for a disk */
 663                 if (rdev->raid_disk == mddev->raid_disks-1) {
 664                         pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
 665                                 mdname(mddev));
 666                         return ERR_PTR(-EINVAL);
 667                 }
 668                 rdev->sectors = mddev->dev_sectors;
 669         }
 670
 671         /* Set new parameters */
 672         mddev->new_level = 0;
 673         mddev->new_layout = 0;
 674         mddev->new_chunk_sectors = mddev->chunk_sectors;
 675         mddev->raid_disks--;
 676         mddev->delta_disks = -1;
 677         /* make sure it will be not marked as dirty */
 678         mddev->recovery_cp = MaxSector;
 679         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 680
 681         create_strip_zones(mddev, &priv_conf);
 682
 683         return priv_conf;
 684 }
 685
 686 static void *raid0_takeover_raid10(struct mddev *mddev)
 687 {
 688         struct r0conf *priv_conf;
 689
 690         /* Check layout:
 691          *  - far_copies must be 1
 692          *  - near_copies must be 2
 693          *  - disks number must be even
 694          *  - all mirrors must be already degraded
 695          */
 696         if (mddev->layout != ((1 << 8) + 2)) {
 697                 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
 698                         mdname(mddev),
 699                         mddev->layout);
 700                 return ERR_PTR(-EINVAL);
 701         }
 702         if (mddev->raid_disks & 1) {
 703                 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
 704                         mdname(mddev));
 705                 return ERR_PTR(-EINVAL);
 706         }
 707         if (mddev->degraded != (mddev->raid_disks>>1)) {
 708                 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
 709                         mdname(mddev));
 710                 return ERR_PTR(-EINVAL);
 711         }
 712
 713         /* Set new parameters */
 714         mddev->new_level = 0;
 715         mddev->new_layout = 0;
 716         mddev->new_chunk_sectors = mddev->chunk_sectors;
 717         mddev->delta_disks = - mddev->raid_disks / 2;
 718         mddev->raid_disks += mddev->delta_disks;
 719         mddev->degraded = 0;
 720         /* make sure it will be not marked as dirty */
 721         mddev->recovery_cp = MaxSector;
 722         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 723
 724         create_strip_zones(mddev, &priv_conf);
 725         return priv_conf;
 726 }
 727
 728 static void *raid0_takeover_raid1(struct mddev *mddev)
 729 {
 730         struct r0conf *priv_conf;
 731         int chunksect;
 732
 733         /* Check layout:
 734          *  - (N - 1) mirror drives must be already faulty
 735          */
 736         if ((mddev->raid_disks - 1) != mddev->degraded) {
 737                 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
 738                        mdname(mddev));
 739                 return ERR_PTR(-EINVAL);
 740         }
 741
 742         /*
 743          * a raid1 doesn't have the notion of chunk size, so
 744          * figure out the largest suitable size we can use.
 745          */
 746         chunksect = 64 * 2; /* 64K by default */
 747
 748         /* The array must be an exact multiple of chunksize */
 749         while (chunksect && (mddev->array_sectors & (chunksect - 1)))
 750                 chunksect >>= 1;
 751
 752         if ((chunksect << 9) < PAGE_SIZE)
 753                 /* array size does not allow a suitable chunk size */
 754                 return ERR_PTR(-EINVAL);
 755
 756         /* Set new parameters */
 757         mddev->new_level = 0;
 758         mddev->new_layout = 0;
 759         mddev->new_chunk_sectors = chunksect;
 760         mddev->chunk_sectors = chunksect;
 761         mddev->delta_disks = 1 - mddev->raid_disks;
 762         mddev->raid_disks = 1;
 763         /* make sure it will be not marked as dirty */
 764         mddev->recovery_cp = MaxSector;
 765         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 766
 767         create_strip_zones(mddev, &priv_conf);
 768         return priv_conf;
 769 }
 770
 771 static void *raid0_takeover(struct mddev *mddev)
 772 {
 773         /* raid0 can take over:
 774          *  raid4 - if all data disks are active.
 775          *  raid5 - providing it is Raid4 layout and one disk is faulty
 776          *  raid10 - assuming we have all necessary active disks
 777          *  raid1 - with (N -1) mirror drives faulty
 778          */
 779
 780         if (mddev->bitmap) {
 781                 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
 782                         mdname(mddev));
 783                 return ERR_PTR(-EBUSY);
 784         }
 785         if (mddev->level == 4)
 786                 return raid0_takeover_raid45(mddev);
 787
 788         if (mddev->level == 5) {
 789                 if (mddev->layout == ALGORITHM_PARITY_N)
 790                         return raid0_takeover_raid45(mddev);
 791
 792                 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
 793                         mdname(mddev), ALGORITHM_PARITY_N);
 794         }
 795
 796         if (mddev->level == 10)
 797                 return raid0_takeover_raid10(mddev);
 798
 799         if (mddev->level == 1)
 800                 return raid0_takeover_raid1(mddev);
 801
 802         pr_warn("Takeover from raid%i to raid0 not supported\n",
 803                 mddev->level);
 804
 805         return ERR_PTR(-EINVAL);
 806 }
 807
 808 static void raid0_quiesce(struct mddev *mddev, int quiesce)
 809 {
 810 }
 811
 812 static struct md_personality raid0_personality=
 813 {
 814         .name           = "raid0",
 815         .level          = 0,
 816         .owner          = THIS_MODULE,
 817         .make_request   = raid0_make_request,
 818         .run            = raid0_run,
 819         .free           = raid0_free,
 820         .status         = raid0_status,
 821         .size           = raid0_size,
 822         .takeover       = raid0_takeover,
 823         .quiesce        = raid0_quiesce,
 824         .error_handler  = raid0_error,
 825 };
 826
 827 static int __init raid0_init (void)
 828 {
 829         return register_md_personality (&raid0_personality);
 830 }
 831
 832 static void raid0_exit (void)
 833 {
 834         unregister_md_personality (&raid0_personality);
 835 }
 836
 837 module_init(raid0_init);
 838 module_exit(raid0_exit);
 839 MODULE_LICENSE("GPL");
 840 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
 841 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 842 MODULE_ALIAS("md-raid0");
 843 MODULE_ALIAS("md-level-0");