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         char b[BDEVNAME_SIZE];
  41         struct r0conf *conf = mddev->private;
  42         int raid_disks = conf->strip_zone[0].nb_dev;
  43         pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
  44                  mdname(mddev),
  45                  conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
  46         for (j = 0; j < conf->nr_strip_zones; j++) {
  47                 char line[200];
  48                 int len = 0;
  49
  50                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
  51                         len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
  52                                         bdevname(conf->devlist[j*raid_disks
  53                                                                + k]->bdev, b));
  54                 pr_debug("md: zone%d=[%s]\n", j, line);
  55
  56                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
  57                 pr_debug("      zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
  58                         (unsigned long long)zone_start>>1,
  59                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
  60                         (unsigned long long)zone_size>>1);
  61                 zone_start = conf->strip_zone[j].zone_end;
  62         }
  63 }
  64
  65 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
  66 {
  67         int i, c, err;
  68         sector_t curr_zone_end, sectors;
  69         struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
  70         struct strip_zone *zone;
  71         int cnt;
  72         char b[BDEVNAME_SIZE];
  73         char b2[BDEVNAME_SIZE];
  74         struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
  75         unsigned blksize = 512;
  76
  77         *private_conf = ERR_PTR(-ENOMEM);
  78         if (!conf)
  79                 return -ENOMEM;
  80         rdev_for_each(rdev1, mddev) {
  81                 pr_debug("md/raid0:%s: looking at %s\n",
  82                          mdname(mddev),
  83                          bdevname(rdev1->bdev, b));
  84                 c = 0;
  85
  86                 /* round size to chunk_size */
  87                 sectors = rdev1->sectors;
  88                 sector_div(sectors, mddev->chunk_sectors);
  89                 rdev1->sectors = sectors * mddev->chunk_sectors;
  90
  91                 blksize = max(blksize, queue_logical_block_size(
  92                                       rdev1->bdev->bd_disk->queue));
  93
  94                 rdev_for_each(rdev2, mddev) {
  95                         pr_debug("md/raid0:%s:   comparing %s(%llu)"
  96                                  " with %s(%llu)\n",
  97                                  mdname(mddev),
  98                                  bdevname(rdev1->bdev,b),
  99                                  (unsigned long long)rdev1->sectors,
 100                                  bdevname(rdev2->bdev,b2),
 101                                  (unsigned long long)rdev2->sectors);
 102                         if (rdev2 == rdev1) {
 103                                 pr_debug("md/raid0:%s:   END\n",
 104                                          mdname(mddev));
 105                                 break;
 106                         }
 107                         if (rdev2->sectors == rdev1->sectors) {
 108                                 /*
 109                                  * Not unique, don't count it as a new
 110                                  * group
 111                                  */
 112                                 pr_debug("md/raid0:%s:   EQUAL\n",
 113                                          mdname(mddev));
 114                                 c = 1;
 115                                 break;
 116                         }
 117                         pr_debug("md/raid0:%s:   NOT EQUAL\n",
 118                                  mdname(mddev));
 119                 }
 120                 if (!c) {
 121                         pr_debug("md/raid0:%s:   ==> UNIQUE\n",
 122                                  mdname(mddev));
 123                         conf->nr_strip_zones++;
 124                         pr_debug("md/raid0:%s: %d zones\n",
 125                                  mdname(mddev), conf->nr_strip_zones);
 126                 }
 127         }
 128         pr_debug("md/raid0:%s: FINAL %d zones\n",
 129                  mdname(mddev), conf->nr_strip_zones);
 130
 131         if (conf->nr_strip_zones == 1) {
 132                 conf->layout = RAID0_ORIG_LAYOUT;
 133         } else if (mddev->layout == RAID0_ORIG_LAYOUT ||
 134                    mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) {
 135                 conf->layout = mddev->layout;
 136         } else if (default_layout == RAID0_ORIG_LAYOUT ||
 137                    default_layout == RAID0_ALT_MULTIZONE_LAYOUT) {
 138                 conf->layout = default_layout;
 139         } else {
 140                 conf->layout = RAID0_ALT_MULTIZONE_LAYOUT;
 141                 pr_warn("md/raid0:%s: !!! DEFAULTING TO ALTERNATE LAYOUT !!!\n",
 142                        mdname(mddev));
 143                 pr_warn("md/raid0: Please set raid0.default_layout to 1 or 2\n");
 144                 pr_warn("md/raid0: Read the following page for more information:\n");
 145                 pr_warn("md/raid0: https://wiki.ubuntu.com/Kernel/Raid0LayoutMigration\n");
 146         }
 147         /*
 148          * now since we have the hard sector sizes, we can make sure
 149          * chunk size is a multiple of that sector size
 150          */
 151         if ((mddev->chunk_sectors << 9) % blksize) {
 152                 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
 153                         mdname(mddev),
 154                         mddev->chunk_sectors << 9, blksize);
 155                 err = -EINVAL;
 156                 goto abort;
 157         }
 158
 159         err = -ENOMEM;
 160         conf->strip_zone = kcalloc(conf->nr_strip_zones,
 161                                    sizeof(struct strip_zone),
 162                                    GFP_KERNEL);
 163         if (!conf->strip_zone)
 164                 goto abort;
 165         conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
 166                                             conf->nr_strip_zones,
 167                                             mddev->raid_disks),
 168                                 GFP_KERNEL);
 169         if (!conf->devlist)
 170                 goto abort;
 171
 172         /* The first zone must contain all devices, so here we check that
 173          * there is a proper alignment of slots to devices and find them all
 174          */
 175         zone = &conf->strip_zone[0];
 176         cnt = 0;
 177         smallest = NULL;
 178         dev = conf->devlist;
 179         err = -EINVAL;
 180         rdev_for_each(rdev1, mddev) {
 181                 int j = rdev1->raid_disk;
 182
 183                 if (mddev->level == 10) {
 184                         /* taking over a raid10-n2 array */
 185                         j /= 2;
 186                         rdev1->new_raid_disk = j;
 187                 }
 188
 189                 if (mddev->level == 1) {
 190                         /* taiking over a raid1 array-
 191                          * we have only one active disk
 192                          */
 193                         j = 0;
 194                         rdev1->new_raid_disk = j;
 195                 }
 196
 197                 if (j < 0) {
 198                         pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
 199                                 mdname(mddev));
 200                         goto abort;
 201                 }
 202                 if (j >= mddev->raid_disks) {
 203                         pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
 204                                 mdname(mddev), j);
 205                         goto abort;
 206                 }
 207                 if (dev[j]) {
 208                         pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
 209                                 mdname(mddev), j);
 210                         goto abort;
 211                 }
 212                 dev[j] = rdev1;
 213
 214                 if (!smallest || (rdev1->sectors < smallest->sectors))
 215                         smallest = rdev1;
 216                 cnt++;
 217         }
 218         if (cnt != mddev->raid_disks) {
 219                 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
 220                         mdname(mddev), cnt, mddev->raid_disks);
 221                 goto abort;
 222         }
 223         zone->nb_dev = cnt;
 224         zone->zone_end = smallest->sectors * cnt;
 225
 226         curr_zone_end = zone->zone_end;
 227
 228         /* now do the other zones */
 229         for (i = 1; i < conf->nr_strip_zones; i++)
 230         {
 231                 int j;
 232
 233                 zone = conf->strip_zone + i;
 234                 dev = conf->devlist + i * mddev->raid_disks;
 235
 236                 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
 237                 zone->dev_start = smallest->sectors;
 238                 smallest = NULL;
 239                 c = 0;
 240
 241                 for (j=0; j<cnt; j++) {
 242                         rdev = conf->devlist[j];
 243                         if (rdev->sectors <= zone->dev_start) {
 244                                 pr_debug("md/raid0:%s: checking %s ... nope\n",
 245                                          mdname(mddev),
 246                                          bdevname(rdev->bdev, b));
 247                                 continue;
 248                         }
 249                         pr_debug("md/raid0:%s: checking %s ..."
 250                                  " contained as device %d\n",
 251                                  mdname(mddev),
 252                                  bdevname(rdev->bdev, b), c);
 253                         dev[c] = rdev;
 254                         c++;
 255                         if (!smallest || rdev->sectors < smallest->sectors) {
 256                                 smallest = rdev;
 257                                 pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
 258                                          mdname(mddev),
 259                                          (unsigned long long)rdev->sectors);
 260                         }
 261                 }
 262
 263                 zone->nb_dev = c;
 264                 sectors = (smallest->sectors - zone->dev_start) * c;
 265                 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
 266                          mdname(mddev),
 267                          zone->nb_dev, (unsigned long long)sectors);
 268
 269                 curr_zone_end += sectors;
 270                 zone->zone_end = curr_zone_end;
 271
 272                 pr_debug("md/raid0:%s: current zone start: %llu\n",
 273                          mdname(mddev),
 274                          (unsigned long long)smallest->sectors);
 275         }
 276
 277         pr_debug("md/raid0:%s: done.\n", mdname(mddev));
 278         *private_conf = conf;
 279
 280         return 0;
 281 abort:
 282         kfree(conf->strip_zone);
 283         kfree(conf->devlist);
 284         kfree(conf);
 285         *private_conf = ERR_PTR(err);
 286         return err;
 287 }
 288
 289 /* Find the zone which holds a particular offset
 290  * Update *sectorp to be an offset in that zone
 291  */
 292 static struct strip_zone *find_zone(struct r0conf *conf,
 293                                     sector_t *sectorp)
 294 {
 295         int i;
 296         struct strip_zone *z = conf->strip_zone;
 297         sector_t sector = *sectorp;
 298
 299         for (i = 0; i < conf->nr_strip_zones; i++)
 300                 if (sector < z[i].zone_end) {
 301                         if (i)
 302                                 *sectorp = sector - z[i-1].zone_end;
 303                         return z + i;
 304                 }
 305         BUG();
 306 }
 307
 308 /*
 309  * remaps the bio to the target device. we separate two flows.
 310  * power 2 flow and a general flow for the sake of performance
 311 */
 312 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
 313                                 sector_t sector, sector_t *sector_offset)
 314 {
 315         unsigned int sect_in_chunk;
 316         sector_t chunk;
 317         struct r0conf *conf = mddev->private;
 318         int raid_disks = conf->strip_zone[0].nb_dev;
 319         unsigned int chunk_sects = mddev->chunk_sectors;
 320
 321         if (is_power_of_2(chunk_sects)) {
 322                 int chunksect_bits = ffz(~chunk_sects);
 323                 /* find the sector offset inside the chunk */
 324                 sect_in_chunk  = sector & (chunk_sects - 1);
 325                 sector >>= chunksect_bits;
 326                 /* chunk in zone */
 327                 chunk = *sector_offset;
 328                 /* quotient is the chunk in real device*/
 329                 sector_div(chunk, zone->nb_dev << chunksect_bits);
 330         } else{
 331                 sect_in_chunk = sector_div(sector, chunk_sects);
 332                 chunk = *sector_offset;
 333                 sector_div(chunk, chunk_sects * zone->nb_dev);
 334         }
 335         /*
 336         *  position the bio over the real device
 337         *  real sector = chunk in device + starting of zone
 338         *       + the position in the chunk
 339         */
 340         *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
 341         return conf->devlist[(zone - conf->strip_zone)*raid_disks
 342                              + sector_div(sector, zone->nb_dev)];
 343 }
 344
 345 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
 346 {
 347         sector_t array_sectors = 0;
 348         struct md_rdev *rdev;
 349
 350         WARN_ONCE(sectors || raid_disks,
 351                   "%s does not support generic reshape\n", __func__);
 352
 353         rdev_for_each(rdev, mddev)
 354                 array_sectors += (rdev->sectors &
 355                                   ~(sector_t)(mddev->chunk_sectors-1));
 356
 357         return array_sectors;
 358 }
 359
 360 static void free_conf(struct mddev *mddev, struct r0conf *conf)
 361 {
 362         kfree(conf->strip_zone);
 363         kfree(conf->devlist);
 364         kfree(conf);
 365         mddev->private = NULL;
 366 }
 367
 368 static void raid0_free(struct mddev *mddev, void *priv)
 369 {
 370         struct r0conf *conf = priv;
 371
 372         free_conf(mddev, conf);
 373         acct_bioset_exit(mddev);
 374 }
 375
 376 static int raid0_run(struct mddev *mddev)
 377 {
 378         struct r0conf *conf;
 379         int ret;
 380
 381         if (mddev->chunk_sectors == 0) {
 382                 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
 383                 return -EINVAL;
 384         }
 385         if (md_check_no_bitmap(mddev))
 386                 return -EINVAL;
 387
 388         if (acct_bioset_init(mddev)) {
 389                 pr_err("md/raid0:%s: alloc acct bioset failed.\n", mdname(mddev));
 390                 return -ENOMEM;
 391         }
 392
 393         /* if private is not null, we are here after takeover */
 394         if (mddev->private == NULL) {
 395                 ret = create_strip_zones(mddev, &conf);
 396                 if (ret < 0)
 397                         goto exit_acct_set;
 398                 mddev->private = conf;
 399         }
 400         conf = mddev->private;
 401         if (mddev->queue) {
 402                 struct md_rdev *rdev;
 403                 bool discard_supported = false;
 404
 405                 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
 406                 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
 407                 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
 408                 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
 409
 410                 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
 411                 blk_queue_io_opt(mddev->queue,
 412                                  (mddev->chunk_sectors << 9) * mddev->raid_disks);
 413
 414                 rdev_for_each(rdev, mddev) {
 415                         disk_stack_limits(mddev->gendisk, rdev->bdev,
 416                                           rdev->data_offset << 9);
 417                         if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
 418                                 discard_supported = true;
 419                 }
 420                 if (!discard_supported)
 421                         blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
 422                 else
 423                         blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
 424         }
 425
 426         /* calculate array device size */
 427         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 428
 429         pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
 430                  mdname(mddev),
 431                  (unsigned long long)mddev->array_sectors);
 432
 433         dump_zones(mddev);
 434
 435         ret = md_integrity_register(mddev);
 436         if (ret)
 437                 goto free;
 438
 439         return ret;
 440
 441 free:
 442         free_conf(mddev, conf);
 443 exit_acct_set:
 444         acct_bioset_exit(mddev);
 445         return ret;
 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
 462         zone = find_zone(conf, &start);
 463
 464         if (bio_end_sector(bio) > zone->zone_end) {
 465                 struct bio *split = bio_split(bio,
 466                         zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
 467                         &mddev->bio_set);
 468                 bio_chain(split, bio);
 469                 submit_bio_noacct(bio);
 470                 bio = split;
 471                 end = zone->zone_end;
 472         } else
 473                 end = bio_end_sector(bio);
 474
 475         if (zone != conf->strip_zone)
 476                 end = end - zone[-1].zone_end;
 477
 478         /* Now start and end is the offset in zone */
 479         stripe_size = zone->nb_dev * mddev->chunk_sectors;
 480
 481         first_stripe_index = start;
 482         sector_div(first_stripe_index, stripe_size);
 483         last_stripe_index = end;
 484         sector_div(last_stripe_index, stripe_size);
 485
 486         start_disk_index = (int)(start - first_stripe_index * stripe_size) /
 487                 mddev->chunk_sectors;
 488         start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
 489                 mddev->chunk_sectors) +
 490                 first_stripe_index * mddev->chunk_sectors;
 491         end_disk_index = (int)(end - last_stripe_index * stripe_size) /
 492                 mddev->chunk_sectors;
 493         end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
 494                 mddev->chunk_sectors) +
 495                 last_stripe_index * mddev->chunk_sectors;
 496
 497         for (disk = 0; disk < zone->nb_dev; disk++) {
 498                 sector_t dev_start, dev_end;
 499                 struct md_rdev *rdev;
 500
 501                 if (disk < start_disk_index)
 502                         dev_start = (first_stripe_index + 1) *
 503                                 mddev->chunk_sectors;
 504                 else if (disk > start_disk_index)
 505                         dev_start = first_stripe_index * mddev->chunk_sectors;
 506                 else
 507                         dev_start = start_disk_offset;
 508
 509                 if (disk < end_disk_index)
 510                         dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
 511                 else if (disk > end_disk_index)
 512                         dev_end = last_stripe_index * mddev->chunk_sectors;
 513                 else
 514                         dev_end = end_disk_offset;
 515
 516                 if (dev_end <= dev_start)
 517                         continue;
 518
 519                 rdev = conf->devlist[(zone - conf->strip_zone) *
 520                         conf->strip_zone[0].nb_dev + disk];
 521                 md_submit_discard_bio(mddev, rdev, bio,
 522                         dev_start + zone->dev_start + rdev->data_offset,
 523                         dev_end - dev_start);
 524         }
 525         bio_endio(bio);
 526 }
 527
 528 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
 529 {
 530         struct r0conf *conf = mddev->private;
 531         struct strip_zone *zone;
 532         struct md_rdev *tmp_dev;
 533         sector_t bio_sector;
 534         sector_t sector;
 535         sector_t orig_sector;
 536         unsigned chunk_sects;
 537         unsigned sectors;
 538
 539         if (unlikely(bio->bi_opf & REQ_PREFLUSH)
 540             && md_flush_request(mddev, bio))
 541                 return true;
 542
 543         if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
 544                 raid0_handle_discard(mddev, bio);
 545                 return true;
 546         }
 547
 548         bio_sector = bio->bi_iter.bi_sector;
 549         sector = bio_sector;
 550         chunk_sects = mddev->chunk_sectors;
 551
 552         sectors = chunk_sects -
 553                 (likely(is_power_of_2(chunk_sects))
 554                  ? (sector & (chunk_sects-1))
 555                  : sector_div(sector, chunk_sects));
 556
 557         /* Restore due to sector_div */
 558         sector = bio_sector;
 559
 560         if (sectors < bio_sectors(bio)) {
 561                 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
 562                                               &mddev->bio_set);
 563                 bio_chain(split, bio);
 564                 submit_bio_noacct(bio);
 565                 bio = split;
 566         }
 567
 568         if (bio->bi_pool != &mddev->bio_set)
 569                 md_account_bio(mddev, &bio);
 570
 571         orig_sector = sector;
 572         zone = find_zone(mddev->private, &sector);
 573         switch (conf->layout) {
 574         case RAID0_ORIG_LAYOUT:
 575                 tmp_dev = map_sector(mddev, zone, orig_sector, &sector);
 576                 break;
 577         case RAID0_ALT_MULTIZONE_LAYOUT:
 578                 tmp_dev = map_sector(mddev, zone, sector, &sector);
 579                 break;
 580         default:
 581                 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev));
 582                 bio_io_error(bio);
 583                 return true;
 584         }
 585
 586         if (unlikely(is_mddev_broken(tmp_dev, "raid0"))) {
 587                 bio_io_error(bio);
 588                 return true;
 589         }
 590
 591         bio_set_dev(bio, tmp_dev->bdev);
 592         bio->bi_iter.bi_sector = sector + zone->dev_start +
 593                 tmp_dev->data_offset;
 594
 595         if (mddev->gendisk)
 596                 trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
 597                                       bio_sector);
 598         mddev_check_writesame(mddev, bio);
 599         mddev_check_write_zeroes(mddev, bio);
 600         submit_bio_noacct(bio);
 601         return true;
 602 }
 603
 604 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
 605 {
 606         seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
 607         return;
 608 }
 609
 610 static void *raid0_takeover_raid45(struct mddev *mddev)
 611 {
 612         struct md_rdev *rdev;
 613         struct r0conf *priv_conf;
 614
 615         if (mddev->degraded != 1) {
 616                 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
 617                         mdname(mddev),
 618                         mddev->degraded);
 619                 return ERR_PTR(-EINVAL);
 620         }
 621
 622         rdev_for_each(rdev, mddev) {
 623                 /* check slot number for a disk */
 624                 if (rdev->raid_disk == mddev->raid_disks-1) {
 625                         pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
 626                                 mdname(mddev));
 627                         return ERR_PTR(-EINVAL);
 628                 }
 629                 rdev->sectors = mddev->dev_sectors;
 630         }
 631
 632         /* Set new parameters */
 633         mddev->new_level = 0;
 634         mddev->new_layout = 0;
 635         mddev->new_chunk_sectors = mddev->chunk_sectors;
 636         mddev->raid_disks--;
 637         mddev->delta_disks = -1;
 638         /* make sure it will be not marked as dirty */
 639         mddev->recovery_cp = MaxSector;
 640         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 641
 642         create_strip_zones(mddev, &priv_conf);
 643
 644         return priv_conf;
 645 }
 646
 647 static void *raid0_takeover_raid10(struct mddev *mddev)
 648 {
 649         struct r0conf *priv_conf;
 650
 651         /* Check layout:
 652          *  - far_copies must be 1
 653          *  - near_copies must be 2
 654          *  - disks number must be even
 655          *  - all mirrors must be already degraded
 656          */
 657         if (mddev->layout != ((1 << 8) + 2)) {
 658                 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
 659                         mdname(mddev),
 660                         mddev->layout);
 661                 return ERR_PTR(-EINVAL);
 662         }
 663         if (mddev->raid_disks & 1) {
 664                 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
 665                         mdname(mddev));
 666                 return ERR_PTR(-EINVAL);
 667         }
 668         if (mddev->degraded != (mddev->raid_disks>>1)) {
 669                 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
 670                         mdname(mddev));
 671                 return ERR_PTR(-EINVAL);
 672         }
 673
 674         /* Set new parameters */
 675         mddev->new_level = 0;
 676         mddev->new_layout = 0;
 677         mddev->new_chunk_sectors = mddev->chunk_sectors;
 678         mddev->delta_disks = - mddev->raid_disks / 2;
 679         mddev->raid_disks += mddev->delta_disks;
 680         mddev->degraded = 0;
 681         /* make sure it will be not marked as dirty */
 682         mddev->recovery_cp = MaxSector;
 683         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 684
 685         create_strip_zones(mddev, &priv_conf);
 686         return priv_conf;
 687 }
 688
 689 static void *raid0_takeover_raid1(struct mddev *mddev)
 690 {
 691         struct r0conf *priv_conf;
 692         int chunksect;
 693
 694         /* Check layout:
 695          *  - (N - 1) mirror drives must be already faulty
 696          */
 697         if ((mddev->raid_disks - 1) != mddev->degraded) {
 698                 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
 699                        mdname(mddev));
 700                 return ERR_PTR(-EINVAL);
 701         }
 702
 703         /*
 704          * a raid1 doesn't have the notion of chunk size, so
 705          * figure out the largest suitable size we can use.
 706          */
 707         chunksect = 64 * 2; /* 64K by default */
 708
 709         /* The array must be an exact multiple of chunksize */
 710         while (chunksect && (mddev->array_sectors & (chunksect - 1)))
 711                 chunksect >>= 1;
 712
 713         if ((chunksect << 9) < PAGE_SIZE)
 714                 /* array size does not allow a suitable chunk size */
 715                 return ERR_PTR(-EINVAL);
 716
 717         /* Set new parameters */
 718         mddev->new_level = 0;
 719         mddev->new_layout = 0;
 720         mddev->new_chunk_sectors = chunksect;
 721         mddev->chunk_sectors = chunksect;
 722         mddev->delta_disks = 1 - mddev->raid_disks;
 723         mddev->raid_disks = 1;
 724         /* make sure it will be not marked as dirty */
 725         mddev->recovery_cp = MaxSector;
 726         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 727
 728         create_strip_zones(mddev, &priv_conf);
 729         return priv_conf;
 730 }
 731
 732 static void *raid0_takeover(struct mddev *mddev)
 733 {
 734         /* raid0 can take over:
 735          *  raid4 - if all data disks are active.
 736          *  raid5 - providing it is Raid4 layout and one disk is faulty
 737          *  raid10 - assuming we have all necessary active disks
 738          *  raid1 - with (N -1) mirror drives faulty
 739          */
 740
 741         if (mddev->bitmap) {
 742                 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
 743                         mdname(mddev));
 744                 return ERR_PTR(-EBUSY);
 745         }
 746         if (mddev->level == 4)
 747                 return raid0_takeover_raid45(mddev);
 748
 749         if (mddev->level == 5) {
 750                 if (mddev->layout == ALGORITHM_PARITY_N)
 751                         return raid0_takeover_raid45(mddev);
 752
 753                 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
 754                         mdname(mddev), ALGORITHM_PARITY_N);
 755         }
 756
 757         if (mddev->level == 10)
 758                 return raid0_takeover_raid10(mddev);
 759
 760         if (mddev->level == 1)
 761                 return raid0_takeover_raid1(mddev);
 762
 763         pr_warn("Takeover from raid%i to raid0 not supported\n",
 764                 mddev->level);
 765
 766         return ERR_PTR(-EINVAL);
 767 }
 768
 769 static void raid0_quiesce(struct mddev *mddev, int quiesce)
 770 {
 771 }
 772
 773 static struct md_personality raid0_personality=
 774 {
 775         .name           = "raid0",
 776         .level          = 0,
 777         .owner          = THIS_MODULE,
 778         .make_request   = raid0_make_request,
 779         .run            = raid0_run,
 780         .free           = raid0_free,
 781         .status         = raid0_status,
 782         .size           = raid0_size,
 783         .takeover       = raid0_takeover,
 784         .quiesce        = raid0_quiesce,
 785 };
 786
 787 static int __init raid0_init (void)
 788 {
 789         return register_md_personality (&raid0_personality);
 790 }
 791
 792 static void raid0_exit (void)
 793 {
 794         unregister_md_personality (&raid0_personality);
 795 }
 796
 797 module_init(raid0_init);
 798 module_exit(raid0_exit);
 799 MODULE_LICENSE("GPL");
 800 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
 801 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 802 MODULE_ALIAS("md-raid0");
 803 MODULE_ALIAS("md-level-0");