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    RAID-0 management functions.
   9
  10    This program is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    You should have received a copy of the GNU General Public License
  16    (for example /usr/src/linux/COPYING); if not, write to the Free
  17    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  18 */
  19
  20 #include <linux/blkdev.h>
  21 #include <linux/seq_file.h>
  22 #include <linux/module.h>
  23 #include <linux/slab.h>
  24 #include <trace/events/block.h>
  25 #include "md.h"
  26 #include "raid0.h"
  27 #include "raid5.h"
  28
  29 #define UNSUPPORTED_MDDEV_FLAGS         \
  30         ((1L << MD_HAS_JOURNAL) |       \
  31          (1L << MD_JOURNAL_CLEAN) |     \
  32          (1L << MD_FAILFAST_SUPPORTED) |\
  33          (1L << MD_HAS_PPL) |           \
  34          (1L << MD_HAS_MULTIPLE_PPLS))
  35
  36 static int raid0_congested(struct mddev *mddev, int bits)
  37 {
  38         struct r0conf *conf = mddev->private;
  39         struct md_rdev **devlist = conf->devlist;
  40         int raid_disks = conf->strip_zone[0].nb_dev;
  41         int i, ret = 0;
  42
  43         for (i = 0; i < raid_disks && !ret ; i++) {
  44                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  45
  46                 ret |= bdi_congested(q->backing_dev_info, bits);
  47         }
  48         return ret;
  49 }
  50
  51 /*
  52  * inform the user of the raid configuration
  53 */
  54 static void dump_zones(struct mddev *mddev)
  55 {
  56         int j, k;
  57         sector_t zone_size = 0;
  58         sector_t zone_start = 0;
  59         char b[BDEVNAME_SIZE];
  60         struct r0conf *conf = mddev->private;
  61         int raid_disks = conf->strip_zone[0].nb_dev;
  62         pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
  63                  mdname(mddev),
  64                  conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
  65         for (j = 0; j < conf->nr_strip_zones; j++) {
  66                 char line[200];
  67                 int len = 0;
  68
  69                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
  70                         len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
  71                                         bdevname(conf->devlist[j*raid_disks
  72                                                                + k]->bdev, b));
  73                 pr_debug("md: zone%d=[%s]\n", j, line);
  74
  75                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
  76                 pr_debug("      zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
  77                         (unsigned long long)zone_start>>1,
  78                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
  79                         (unsigned long long)zone_size>>1);
  80                 zone_start = conf->strip_zone[j].zone_end;
  81         }
  82 }
  83
  84 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
  85 {
  86         int i, c, err;
  87         sector_t curr_zone_end, sectors;
  88         struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
  89         struct strip_zone *zone;
  90         int cnt;
  91         char b[BDEVNAME_SIZE];
  92         char b2[BDEVNAME_SIZE];
  93         struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
  94         unsigned short blksize = 512;
  95
  96         *private_conf = ERR_PTR(-ENOMEM);
  97         if (!conf)
  98                 return -ENOMEM;
  99         rdev_for_each(rdev1, mddev) {
 100                 pr_debug("md/raid0:%s: looking at %s\n",
 101                          mdname(mddev),
 102                          bdevname(rdev1->bdev, b));
 103                 c = 0;
 104
 105                 /* round size to chunk_size */
 106                 sectors = rdev1->sectors;
 107                 sector_div(sectors, mddev->chunk_sectors);
 108                 rdev1->sectors = sectors * mddev->chunk_sectors;
 109
 110                 blksize = max(blksize, queue_logical_block_size(
 111                                       rdev1->bdev->bd_disk->queue));
 112
 113                 rdev_for_each(rdev2, mddev) {
 114                         pr_debug("md/raid0:%s:   comparing %s(%llu)"
 115                                  " with %s(%llu)\n",
 116                                  mdname(mddev),
 117                                  bdevname(rdev1->bdev,b),
 118                                  (unsigned long long)rdev1->sectors,
 119                                  bdevname(rdev2->bdev,b2),
 120                                  (unsigned long long)rdev2->sectors);
 121                         if (rdev2 == rdev1) {
 122                                 pr_debug("md/raid0:%s:   END\n",
 123                                          mdname(mddev));
 124                                 break;
 125                         }
 126                         if (rdev2->sectors == rdev1->sectors) {
 127                                 /*
 128                                  * Not unique, don't count it as a new
 129                                  * group
 130                                  */
 131                                 pr_debug("md/raid0:%s:   EQUAL\n",
 132                                          mdname(mddev));
 133                                 c = 1;
 134                                 break;
 135                         }
 136                         pr_debug("md/raid0:%s:   NOT EQUAL\n",
 137                                  mdname(mddev));
 138                 }
 139                 if (!c) {
 140                         pr_debug("md/raid0:%s:   ==> UNIQUE\n",
 141                                  mdname(mddev));
 142                         conf->nr_strip_zones++;
 143                         pr_debug("md/raid0:%s: %d zones\n",
 144                                  mdname(mddev), conf->nr_strip_zones);
 145                 }
 146         }
 147         pr_debug("md/raid0:%s: FINAL %d zones\n",
 148                  mdname(mddev), conf->nr_strip_zones);
 149         /*
 150          * now since we have the hard sector sizes, we can make sure
 151          * chunk size is a multiple of that sector size
 152          */
 153         if ((mddev->chunk_sectors << 9) % blksize) {
 154                 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
 155                         mdname(mddev),
 156                         mddev->chunk_sectors << 9, blksize);
 157                 err = -EINVAL;
 158                 goto abort;
 159         }
 160
 161         err = -ENOMEM;
 162         conf->strip_zone = kcalloc(conf->nr_strip_zones,
 163                                    sizeof(struct strip_zone),
 164                                    GFP_KERNEL);
 165         if (!conf->strip_zone)
 166                 goto abort;
 167         conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *),
 168                                             conf->nr_strip_zones,
 169                                             mddev->raid_disks),
 170                                 GFP_KERNEL);
 171         if (!conf->devlist)
 172                 goto abort;
 173
 174         /* The first zone must contain all devices, so here we check that
 175          * there is a proper alignment of slots to devices and find them all
 176          */
 177         zone = &conf->strip_zone[0];
 178         cnt = 0;
 179         smallest = NULL;
 180         dev = conf->devlist;
 181         err = -EINVAL;
 182         rdev_for_each(rdev1, mddev) {
 183                 int j = rdev1->raid_disk;
 184
 185                 if (mddev->level == 10) {
 186                         /* taking over a raid10-n2 array */
 187                         j /= 2;
 188                         rdev1->new_raid_disk = j;
 189                 }
 190
 191                 if (mddev->level == 1) {
 192                         /* taiking over a raid1 array-
 193                          * we have only one active disk
 194                          */
 195                         j = 0;
 196                         rdev1->new_raid_disk = j;
 197                 }
 198
 199                 if (j < 0) {
 200                         pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
 201                                 mdname(mddev));
 202                         goto abort;
 203                 }
 204                 if (j >= mddev->raid_disks) {
 205                         pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
 206                                 mdname(mddev), j);
 207                         goto abort;
 208                 }
 209                 if (dev[j]) {
 210                         pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
 211                                 mdname(mddev), j);
 212                         goto abort;
 213                 }
 214                 dev[j] = rdev1;
 215
 216                 if (!smallest || (rdev1->sectors < smallest->sectors))
 217                         smallest = rdev1;
 218                 cnt++;
 219         }
 220         if (cnt != mddev->raid_disks) {
 221                 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
 222                         mdname(mddev), cnt, mddev->raid_disks);
 223                 goto abort;
 224         }
 225         zone->nb_dev = cnt;
 226         zone->zone_end = smallest->sectors * cnt;
 227
 228         curr_zone_end = zone->zone_end;
 229
 230         /* now do the other zones */
 231         for (i = 1; i < conf->nr_strip_zones; i++)
 232         {
 233                 int j;
 234
 235                 zone = conf->strip_zone + i;
 236                 dev = conf->devlist + i * mddev->raid_disks;
 237
 238                 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
 239                 zone->dev_start = smallest->sectors;
 240                 smallest = NULL;
 241                 c = 0;
 242
 243                 for (j=0; j<cnt; j++) {
 244                         rdev = conf->devlist[j];
 245                         if (rdev->sectors <= zone->dev_start) {
 246                                 pr_debug("md/raid0:%s: checking %s ... nope\n",
 247                                          mdname(mddev),
 248                                          bdevname(rdev->bdev, b));
 249                                 continue;
 250                         }
 251                         pr_debug("md/raid0:%s: checking %s ..."
 252                                  " contained as device %d\n",
 253                                  mdname(mddev),
 254                                  bdevname(rdev->bdev, b), c);
 255                         dev[c] = rdev;
 256                         c++;
 257                         if (!smallest || rdev->sectors < smallest->sectors) {
 258                                 smallest = rdev;
 259                                 pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
 260                                          mdname(mddev),
 261                                          (unsigned long long)rdev->sectors);
 262                         }
 263                 }
 264
 265                 zone->nb_dev = c;
 266                 sectors = (smallest->sectors - zone->dev_start) * c;
 267                 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
 268                          mdname(mddev),
 269                          zone->nb_dev, (unsigned long long)sectors);
 270
 271                 curr_zone_end += sectors;
 272                 zone->zone_end = curr_zone_end;
 273
 274                 pr_debug("md/raid0:%s: current zone start: %llu\n",
 275                          mdname(mddev),
 276                          (unsigned long long)smallest->sectors);
 277         }
 278
 279         pr_debug("md/raid0:%s: done.\n", mdname(mddev));
 280         *private_conf = conf;
 281
 282         return 0;
 283 abort:
 284         kfree(conf->strip_zone);
 285         kfree(conf->devlist);
 286         kfree(conf);
 287         *private_conf = ERR_PTR(err);
 288         return err;
 289 }
 290
 291 /* Find the zone which holds a particular offset
 292  * Update *sectorp to be an offset in that zone
 293  */
 294 static struct strip_zone *find_zone(struct r0conf *conf,
 295                                     sector_t *sectorp)
 296 {
 297         int i;
 298         struct strip_zone *z = conf->strip_zone;
 299         sector_t sector = *sectorp;
 300
 301         for (i = 0; i < conf->nr_strip_zones; i++)
 302                 if (sector < z[i].zone_end) {
 303                         if (i)
 304                                 *sectorp = sector - z[i-1].zone_end;
 305                         return z + i;
 306                 }
 307         BUG();
 308 }
 309
 310 /*
 311  * remaps the bio to the target device. we separate two flows.
 312  * power 2 flow and a general flow for the sake of performance
 313 */
 314 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
 315                                 sector_t sector, sector_t *sector_offset)
 316 {
 317         unsigned int sect_in_chunk;
 318         sector_t chunk;
 319         struct r0conf *conf = mddev->private;
 320         int raid_disks = conf->strip_zone[0].nb_dev;
 321         unsigned int chunk_sects = mddev->chunk_sectors;
 322
 323         if (is_power_of_2(chunk_sects)) {
 324                 int chunksect_bits = ffz(~chunk_sects);
 325                 /* find the sector offset inside the chunk */
 326                 sect_in_chunk  = sector & (chunk_sects - 1);
 327                 sector >>= chunksect_bits;
 328                 /* chunk in zone */
 329                 chunk = *sector_offset;
 330                 /* quotient is the chunk in real device*/
 331                 sector_div(chunk, zone->nb_dev << chunksect_bits);
 332         } else{
 333                 sect_in_chunk = sector_div(sector, chunk_sects);
 334                 chunk = *sector_offset;
 335                 sector_div(chunk, chunk_sects * zone->nb_dev);
 336         }
 337         /*
 338         *  position the bio over the real device
 339         *  real sector = chunk in device + starting of zone
 340         *       + the position in the chunk
 341         */
 342         *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
 343         return conf->devlist[(zone - conf->strip_zone)*raid_disks
 344                              + sector_div(sector, zone->nb_dev)];
 345 }
 346
 347 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
 348 {
 349         sector_t array_sectors = 0;
 350         struct md_rdev *rdev;
 351
 352         WARN_ONCE(sectors || raid_disks,
 353                   "%s does not support generic reshape\n", __func__);
 354
 355         rdev_for_each(rdev, mddev)
 356                 array_sectors += (rdev->sectors &
 357                                   ~(sector_t)(mddev->chunk_sectors-1));
 358
 359         return array_sectors;
 360 }
 361
 362 static void raid0_free(struct mddev *mddev, void *priv);
 363
 364 static int raid0_run(struct mddev *mddev)
 365 {
 366         struct r0conf *conf;
 367         int ret;
 368
 369         if (mddev->chunk_sectors == 0) {
 370                 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
 371                 return -EINVAL;
 372         }
 373         if (md_check_no_bitmap(mddev))
 374                 return -EINVAL;
 375
 376         /* if private is not null, we are here after takeover */
 377         if (mddev->private == NULL) {
 378                 ret = create_strip_zones(mddev, &conf);
 379                 if (ret < 0)
 380                         return ret;
 381                 mddev->private = conf;
 382         }
 383         conf = mddev->private;
 384         if (mddev->queue) {
 385                 struct md_rdev *rdev;
 386                 bool discard_supported = false;
 387
 388                 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
 389                 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
 390                 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
 391                 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
 392
 393                 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
 394                 blk_queue_io_opt(mddev->queue,
 395                                  (mddev->chunk_sectors << 9) * mddev->raid_disks);
 396
 397                 rdev_for_each(rdev, mddev) {
 398                         disk_stack_limits(mddev->gendisk, rdev->bdev,
 399                                           rdev->data_offset << 9);
 400                         if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
 401                                 discard_supported = true;
 402                 }
 403                 if (!discard_supported)
 404                         blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
 405                 else
 406                         blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
 407         }
 408
 409         /* calculate array device size */
 410         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 411
 412         pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
 413                  mdname(mddev),
 414                  (unsigned long long)mddev->array_sectors);
 415
 416         if (mddev->queue) {
 417                 /* calculate the max read-ahead size.
 418                  * For read-ahead of large files to be effective, we need to
 419                  * readahead at least twice a whole stripe. i.e. number of devices
 420                  * multiplied by chunk size times 2.
 421                  * If an individual device has an ra_pages greater than the
 422                  * chunk size, then we will not drive that device as hard as it
 423                  * wants.  We consider this a configuration error: a larger
 424                  * chunksize should be used in that case.
 425                  */
 426                 int stripe = mddev->raid_disks *
 427                         (mddev->chunk_sectors << 9) / PAGE_SIZE;
 428                 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
 429                         mddev->queue->backing_dev_info->ra_pages = 2* stripe;
 430         }
 431
 432         dump_zones(mddev);
 433
 434         ret = md_integrity_register(mddev);
 435
 436         return ret;
 437 }
 438
 439 static void raid0_free(struct mddev *mddev, void *priv)
 440 {
 441         struct r0conf *conf = priv;
 442
 443         kfree(conf->strip_zone);
 444         kfree(conf->devlist);
 445         kfree(conf);
 446 }
 447
 448 /*
 449  * Is io distribute over 1 or more chunks ?
 450 */
 451 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
 452                         unsigned int chunk_sects, struct bio *bio)
 453 {
 454         if (likely(is_power_of_2(chunk_sects))) {
 455                 return chunk_sects >=
 456                         ((bio->bi_iter.bi_sector & (chunk_sects-1))
 457                                         + bio_sectors(bio));
 458         } else{
 459                 sector_t sector = bio->bi_iter.bi_sector;
 460                 return chunk_sects >= (sector_div(sector, chunk_sects)
 461                                                 + bio_sectors(bio));
 462         }
 463 }
 464
 465 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
 466 {
 467         struct r0conf *conf = mddev->private;
 468         struct strip_zone *zone;
 469         sector_t start = bio->bi_iter.bi_sector;
 470         sector_t end;
 471         unsigned int stripe_size;
 472         sector_t first_stripe_index, last_stripe_index;
 473         sector_t start_disk_offset;
 474         unsigned int start_disk_index;
 475         sector_t end_disk_offset;
 476         unsigned int end_disk_index;
 477         unsigned int disk;
 478
 479         zone = find_zone(conf, &start);
 480
 481         if (bio_end_sector(bio) > zone->zone_end) {
 482                 struct bio *split = bio_split(bio,
 483                         zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
 484                         &mddev->bio_set);
 485                 bio_chain(split, bio);
 486                 generic_make_request(bio);
 487                 bio = split;
 488                 end = zone->zone_end;
 489         } else
 490                 end = bio_end_sector(bio);
 491
 492         if (zone != conf->strip_zone)
 493                 end = end - zone[-1].zone_end;
 494
 495         /* Now start and end is the offset in zone */
 496         stripe_size = zone->nb_dev * mddev->chunk_sectors;
 497
 498         first_stripe_index = start;
 499         sector_div(first_stripe_index, stripe_size);
 500         last_stripe_index = end;
 501         sector_div(last_stripe_index, stripe_size);
 502
 503         start_disk_index = (int)(start - first_stripe_index * stripe_size) /
 504                 mddev->chunk_sectors;
 505         start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
 506                 mddev->chunk_sectors) +
 507                 first_stripe_index * mddev->chunk_sectors;
 508         end_disk_index = (int)(end - last_stripe_index * stripe_size) /
 509                 mddev->chunk_sectors;
 510         end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
 511                 mddev->chunk_sectors) +
 512                 last_stripe_index * mddev->chunk_sectors;
 513
 514         for (disk = 0; disk < zone->nb_dev; disk++) {
 515                 sector_t dev_start, dev_end;
 516                 struct bio *discard_bio = NULL;
 517                 struct md_rdev *rdev;
 518
 519                 if (disk < start_disk_index)
 520                         dev_start = (first_stripe_index + 1) *
 521                                 mddev->chunk_sectors;
 522                 else if (disk > start_disk_index)
 523                         dev_start = first_stripe_index * mddev->chunk_sectors;
 524                 else
 525                         dev_start = start_disk_offset;
 526
 527                 if (disk < end_disk_index)
 528                         dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
 529                 else if (disk > end_disk_index)
 530                         dev_end = last_stripe_index * mddev->chunk_sectors;
 531                 else
 532                         dev_end = end_disk_offset;
 533
 534                 if (dev_end <= dev_start)
 535                         continue;
 536
 537                 rdev = conf->devlist[(zone - conf->strip_zone) *
 538                         conf->strip_zone[0].nb_dev + disk];
 539                 if (__blkdev_issue_discard(rdev->bdev,
 540                         dev_start + zone->dev_start + rdev->data_offset,
 541                         dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
 542                     !discard_bio)
 543                         continue;
 544                 bio_chain(discard_bio, bio);
 545                 bio_clone_blkg_association(discard_bio, bio);
 546                 if (mddev->gendisk)
 547                         trace_block_bio_remap(bdev_get_queue(rdev->bdev),
 548                                 discard_bio, disk_devt(mddev->gendisk),
 549                                 bio->bi_iter.bi_sector);
 550                 generic_make_request(discard_bio);
 551         }
 552         bio_endio(bio);
 553 }
 554
 555 static bool raid0_make_request(struct mddev *mddev, struct bio *bio)
 556 {
 557         struct strip_zone *zone;
 558         struct md_rdev *tmp_dev;
 559         sector_t bio_sector;
 560         sector_t sector;
 561         unsigned chunk_sects;
 562         unsigned sectors;
 563
 564         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
 565                 md_flush_request(mddev, bio);
 566                 return true;
 567         }
 568
 569         if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
 570                 raid0_handle_discard(mddev, bio);
 571                 return true;
 572         }
 573
 574         bio_sector = bio->bi_iter.bi_sector;
 575         sector = bio_sector;
 576         chunk_sects = mddev->chunk_sectors;
 577
 578         sectors = chunk_sects -
 579                 (likely(is_power_of_2(chunk_sects))
 580                  ? (sector & (chunk_sects-1))
 581                  : sector_div(sector, chunk_sects));
 582
 583         /* Restore due to sector_div */
 584         sector = bio_sector;
 585
 586         if (sectors < bio_sectors(bio)) {
 587                 struct bio *split = bio_split(bio, sectors, GFP_NOIO,
 588                                               &mddev->bio_set);
 589                 bio_chain(split, bio);
 590                 generic_make_request(bio);
 591                 bio = split;
 592         }
 593
 594         zone = find_zone(mddev->private, &sector);
 595         tmp_dev = map_sector(mddev, zone, sector, &sector);
 596         bio_set_dev(bio, tmp_dev->bdev);
 597         bio->bi_iter.bi_sector = sector + zone->dev_start +
 598                 tmp_dev->data_offset;
 599
 600         if (mddev->gendisk)
 601                 trace_block_bio_remap(bio->bi_disk->queue, bio,
 602                                 disk_devt(mddev->gendisk), bio_sector);
 603         mddev_check_writesame(mddev, bio);
 604         mddev_check_write_zeroes(mddev, bio);
 605         generic_make_request(bio);
 606         return true;
 607 }
 608
 609 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
 610 {
 611         seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
 612         return;
 613 }
 614
 615 static void *raid0_takeover_raid45(struct mddev *mddev)
 616 {
 617         struct md_rdev *rdev;
 618         struct r0conf *priv_conf;
 619
 620         if (mddev->degraded != 1) {
 621                 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
 622                         mdname(mddev),
 623                         mddev->degraded);
 624                 return ERR_PTR(-EINVAL);
 625         }
 626
 627         rdev_for_each(rdev, mddev) {
 628                 /* check slot number for a disk */
 629                 if (rdev->raid_disk == mddev->raid_disks-1) {
 630                         pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n",
 631                                 mdname(mddev));
 632                         return ERR_PTR(-EINVAL);
 633                 }
 634                 rdev->sectors = mddev->dev_sectors;
 635         }
 636
 637         /* Set new parameters */
 638         mddev->new_level = 0;
 639         mddev->new_layout = 0;
 640         mddev->new_chunk_sectors = mddev->chunk_sectors;
 641         mddev->raid_disks--;
 642         mddev->delta_disks = -1;
 643         /* make sure it will be not marked as dirty */
 644         mddev->recovery_cp = MaxSector;
 645         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 646
 647         create_strip_zones(mddev, &priv_conf);
 648
 649         return priv_conf;
 650 }
 651
 652 static void *raid0_takeover_raid10(struct mddev *mddev)
 653 {
 654         struct r0conf *priv_conf;
 655
 656         /* Check layout:
 657          *  - far_copies must be 1
 658          *  - near_copies must be 2
 659          *  - disks number must be even
 660          *  - all mirrors must be already degraded
 661          */
 662         if (mddev->layout != ((1 << 8) + 2)) {
 663                 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
 664                         mdname(mddev),
 665                         mddev->layout);
 666                 return ERR_PTR(-EINVAL);
 667         }
 668         if (mddev->raid_disks & 1) {
 669                 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
 670                         mdname(mddev));
 671                 return ERR_PTR(-EINVAL);
 672         }
 673         if (mddev->degraded != (mddev->raid_disks>>1)) {
 674                 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n",
 675                         mdname(mddev));
 676                 return ERR_PTR(-EINVAL);
 677         }
 678
 679         /* Set new parameters */
 680         mddev->new_level = 0;
 681         mddev->new_layout = 0;
 682         mddev->new_chunk_sectors = mddev->chunk_sectors;
 683         mddev->delta_disks = - mddev->raid_disks / 2;
 684         mddev->raid_disks += mddev->delta_disks;
 685         mddev->degraded = 0;
 686         /* make sure it will be not marked as dirty */
 687         mddev->recovery_cp = MaxSector;
 688         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 689
 690         create_strip_zones(mddev, &priv_conf);
 691         return priv_conf;
 692 }
 693
 694 static void *raid0_takeover_raid1(struct mddev *mddev)
 695 {
 696         struct r0conf *priv_conf;
 697         int chunksect;
 698
 699         /* Check layout:
 700          *  - (N - 1) mirror drives must be already faulty
 701          */
 702         if ((mddev->raid_disks - 1) != mddev->degraded) {
 703                 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
 704                        mdname(mddev));
 705                 return ERR_PTR(-EINVAL);
 706         }
 707
 708         /*
 709          * a raid1 doesn't have the notion of chunk size, so
 710          * figure out the largest suitable size we can use.
 711          */
 712         chunksect = 64 * 2; /* 64K by default */
 713
 714         /* The array must be an exact multiple of chunksize */
 715         while (chunksect && (mddev->array_sectors & (chunksect - 1)))
 716                 chunksect >>= 1;
 717
 718         if ((chunksect << 9) < PAGE_SIZE)
 719                 /* array size does not allow a suitable chunk size */
 720                 return ERR_PTR(-EINVAL);
 721
 722         /* Set new parameters */
 723         mddev->new_level = 0;
 724         mddev->new_layout = 0;
 725         mddev->new_chunk_sectors = chunksect;
 726         mddev->chunk_sectors = chunksect;
 727         mddev->delta_disks = 1 - mddev->raid_disks;
 728         mddev->raid_disks = 1;
 729         /* make sure it will be not marked as dirty */
 730         mddev->recovery_cp = MaxSector;
 731         mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS);
 732
 733         create_strip_zones(mddev, &priv_conf);
 734         return priv_conf;
 735 }
 736
 737 static void *raid0_takeover(struct mddev *mddev)
 738 {
 739         /* raid0 can take over:
 740          *  raid4 - if all data disks are active.
 741          *  raid5 - providing it is Raid4 layout and one disk is faulty
 742          *  raid10 - assuming we have all necessary active disks
 743          *  raid1 - with (N -1) mirror drives faulty
 744          */
 745
 746         if (mddev->bitmap) {
 747                 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n",
 748                         mdname(mddev));
 749                 return ERR_PTR(-EBUSY);
 750         }
 751         if (mddev->level == 4)
 752                 return raid0_takeover_raid45(mddev);
 753
 754         if (mddev->level == 5) {
 755                 if (mddev->layout == ALGORITHM_PARITY_N)
 756                         return raid0_takeover_raid45(mddev);
 757
 758                 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
 759                         mdname(mddev), ALGORITHM_PARITY_N);
 760         }
 761
 762         if (mddev->level == 10)
 763                 return raid0_takeover_raid10(mddev);
 764
 765         if (mddev->level == 1)
 766                 return raid0_takeover_raid1(mddev);
 767
 768         pr_warn("Takeover from raid%i to raid0 not supported\n",
 769                 mddev->level);
 770
 771         return ERR_PTR(-EINVAL);
 772 }
 773
 774 static void raid0_quiesce(struct mddev *mddev, int quiesce)
 775 {
 776 }
 777
 778 static struct md_personality raid0_personality=
 779 {
 780         .name           = "raid0",
 781         .level          = 0,
 782         .owner          = THIS_MODULE,
 783         .make_request   = raid0_make_request,
 784         .run            = raid0_run,
 785         .free           = raid0_free,
 786         .status         = raid0_status,
 787         .size           = raid0_size,
 788         .takeover       = raid0_takeover,
 789         .quiesce        = raid0_quiesce,
 790         .congested      = raid0_congested,
 791 };
 792
 793 static int __init raid0_init (void)
 794 {
 795         return register_md_personality (&raid0_personality);
 796 }
 797
 798 static void raid0_exit (void)
 799 {
 800         unregister_md_personality (&raid0_personality);
 801 }
 802
 803 module_init(raid0_init);
 804 module_exit(raid0_exit);
 805 MODULE_LICENSE("GPL");
 806 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
 807 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 808 MODULE_ALIAS("md-raid0");
 809 MODULE_ALIAS("md-level-0");