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
  35 static int raid0_congested(struct mddev *mddev, int bits)
  36 {
  37         struct r0conf *conf = mddev->private;
  38         struct md_rdev **devlist = conf->devlist;
  39         int raid_disks = conf->strip_zone[0].nb_dev;
  40         int i, ret = 0;
  41
  42         for (i = 0; i < raid_disks && !ret ; i++) {
  43                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  44
  45                 ret |= bdi_congested(q->backing_dev_info, bits);
  46         }
  47         return ret;
  48 }
  49
  50 /*
  51  * inform the user of the raid configuration
  52 */
  53 static void dump_zones(struct mddev *mddev)
  54 {
  55         int j, k;
  56         sector_t zone_size = 0;
  57         sector_t zone_start = 0;
  58         char b[BDEVNAME_SIZE];
  59         struct r0conf *conf = mddev->private;
  60         int raid_disks = conf->strip_zone[0].nb_dev;
  61         pr_debug("md: RAID0 configuration for %s - %d zone%s\n",
  62                  mdname(mddev),
  63                  conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
  64         for (j = 0; j < conf->nr_strip_zones; j++) {
  65                 char line[200];
  66                 int len = 0;
  67
  68                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
  69                         len += snprintf(line+len, 200-len, "%s%s", k?"/":"",
  70                                         bdevname(conf->devlist[j*raid_disks
  71                                                                + k]->bdev, b));
  72                 pr_debug("md: zone%d=[%s]\n", j, line);
  73
  74                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
  75                 pr_debug("      zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n",
  76                         (unsigned long long)zone_start>>1,
  77                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
  78                         (unsigned long long)zone_size>>1);
  79                 zone_start = conf->strip_zone[j].zone_end;
  80         }
  81 }
  82
  83 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
  84 {
  85         int i, c, err;
  86         sector_t curr_zone_end, sectors;
  87         struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
  88         struct strip_zone *zone;
  89         int cnt;
  90         char b[BDEVNAME_SIZE];
  91         char b2[BDEVNAME_SIZE];
  92         struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
  93         unsigned short blksize = 512;
  94
  95         *private_conf = ERR_PTR(-ENOMEM);
  96         if (!conf)
  97                 return -ENOMEM;
  98         rdev_for_each(rdev1, mddev) {
  99                 pr_debug("md/raid0:%s: looking at %s\n",
 100                          mdname(mddev),
 101                          bdevname(rdev1->bdev, b));
 102                 c = 0;
 103
 104                 /* round size to chunk_size */
 105                 sectors = rdev1->sectors;
 106                 sector_div(sectors, mddev->chunk_sectors);
 107                 rdev1->sectors = sectors * mddev->chunk_sectors;
 108
 109                 blksize = max(blksize, queue_logical_block_size(
 110                                       rdev1->bdev->bd_disk->queue));
 111
 112                 rdev_for_each(rdev2, mddev) {
 113                         pr_debug("md/raid0:%s:   comparing %s(%llu)"
 114                                  " with %s(%llu)\n",
 115                                  mdname(mddev),
 116                                  bdevname(rdev1->bdev,b),
 117                                  (unsigned long long)rdev1->sectors,
 118                                  bdevname(rdev2->bdev,b2),
 119                                  (unsigned long long)rdev2->sectors);
 120                         if (rdev2 == rdev1) {
 121                                 pr_debug("md/raid0:%s:   END\n",
 122                                          mdname(mddev));
 123                                 break;
 124                         }
 125                         if (rdev2->sectors == rdev1->sectors) {
 126                                 /*
 127                                  * Not unique, don't count it as a new
 128                                  * group
 129                                  */
 130                                 pr_debug("md/raid0:%s:   EQUAL\n",
 131                                          mdname(mddev));
 132                                 c = 1;
 133                                 break;
 134                         }
 135                         pr_debug("md/raid0:%s:   NOT EQUAL\n",
 136                                  mdname(mddev));
 137                 }
 138                 if (!c) {
 139                         pr_debug("md/raid0:%s:   ==> UNIQUE\n",
 140                                  mdname(mddev));
 141                         conf->nr_strip_zones++;
 142                         pr_debug("md/raid0:%s: %d zones\n",
 143                                  mdname(mddev), conf->nr_strip_zones);
 144                 }
 145         }
 146         pr_debug("md/raid0:%s: FINAL %d zones\n",
 147                  mdname(mddev), conf->nr_strip_zones);
 148         /*
 149          * now since we have the hard sector sizes, we can make sure
 150          * chunk size is a multiple of that sector size
 151          */
 152         if ((mddev->chunk_sectors << 9) % blksize) {
 153                 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n",
 154                         mdname(mddev),
 155                         mddev->chunk_sectors << 9, blksize);
 156                 err = -EINVAL;
 157                 goto abort;
 158         }
 159
 160         err = -ENOMEM;
 161         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 162                                 conf->nr_strip_zones, GFP_KERNEL);
 163         if (!conf->strip_zone)
 164                 goto abort;
 165         conf->devlist = kzalloc(sizeof(struct md_rdev*)*
 166                                 conf->nr_strip_zones*mddev->raid_disks,
 167                                 GFP_KERNEL);
 168         if (!conf->devlist)
 169                 goto abort;
 170
 171         /* The first zone must contain all devices, so here we check that
 172          * there is a proper alignment of slots to devices and find them all
 173          */
 174         zone = &conf->strip_zone[0];
 175         cnt = 0;
 176         smallest = NULL;
 177         dev = conf->devlist;
 178         err = -EINVAL;
 179         rdev_for_each(rdev1, mddev) {
 180                 int j = rdev1->raid_disk;
 181
 182                 if (mddev->level == 10) {
 183                         /* taking over a raid10-n2 array */
 184                         j /= 2;
 185                         rdev1->new_raid_disk = j;
 186                 }
 187
 188                 if (mddev->level == 1) {
 189                         /* taiking over a raid1 array-
 190                          * we have only one active disk
 191                          */
 192                         j = 0;
 193                         rdev1->new_raid_disk = j;
 194                 }
 195
 196                 if (j < 0) {
 197                         pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n",
 198                                 mdname(mddev));
 199                         goto abort;
 200                 }
 201                 if (j >= mddev->raid_disks) {
 202                         pr_warn("md/raid0:%s: bad disk number %d - aborting!\n",
 203                                 mdname(mddev), j);
 204                         goto abort;
 205                 }
 206                 if (dev[j]) {
 207                         pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n",
 208                                 mdname(mddev), j);
 209                         goto abort;
 210                 }
 211                 dev[j] = rdev1;
 212
 213                 if (!smallest || (rdev1->sectors < smallest->sectors))
 214                         smallest = rdev1;
 215                 cnt++;
 216         }
 217         if (cnt != mddev->raid_disks) {
 218                 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n",
 219                         mdname(mddev), cnt, mddev->raid_disks);
 220                 goto abort;
 221         }
 222         zone->nb_dev = cnt;
 223         zone->zone_end = smallest->sectors * cnt;
 224
 225         curr_zone_end = zone->zone_end;
 226
 227         /* now do the other zones */
 228         for (i = 1; i < conf->nr_strip_zones; i++)
 229         {
 230                 int j;
 231
 232                 zone = conf->strip_zone + i;
 233                 dev = conf->devlist + i * mddev->raid_disks;
 234
 235                 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
 236                 zone->dev_start = smallest->sectors;
 237                 smallest = NULL;
 238                 c = 0;
 239
 240                 for (j=0; j<cnt; j++) {
 241                         rdev = conf->devlist[j];
 242                         if (rdev->sectors <= zone->dev_start) {
 243                                 pr_debug("md/raid0:%s: checking %s ... nope\n",
 244                                          mdname(mddev),
 245                                          bdevname(rdev->bdev, b));
 246                                 continue;
 247                         }
 248                         pr_debug("md/raid0:%s: checking %s ..."
 249                                  " contained as device %d\n",
 250                                  mdname(mddev),
 251                                  bdevname(rdev->bdev, b), c);
 252                         dev[c] = rdev;
 253                         c++;
 254                         if (!smallest || rdev->sectors < smallest->sectors) {
 255                                 smallest = rdev;
 256                                 pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
 257                                          mdname(mddev),
 258                                          (unsigned long long)rdev->sectors);
 259                         }
 260                 }
 261
 262                 zone->nb_dev = c;
 263                 sectors = (smallest->sectors - zone->dev_start) * c;
 264                 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
 265                          mdname(mddev),
 266                          zone->nb_dev, (unsigned long long)sectors);
 267
 268                 curr_zone_end += sectors;
 269                 zone->zone_end = curr_zone_end;
 270
 271                 pr_debug("md/raid0:%s: current zone start: %llu\n",
 272                          mdname(mddev),
 273                          (unsigned long long)smallest->sectors);
 274         }
 275
 276         pr_debug("md/raid0:%s: done.\n", mdname(mddev));
 277         *private_conf = conf;
 278
 279         return 0;
 280 abort:
 281         kfree(conf->strip_zone);
 282         kfree(conf->devlist);
 283         kfree(conf);
 284         *private_conf = ERR_PTR(err);
 285         return err;
 286 }
 287
 288 /* Find the zone which holds a particular offset
 289  * Update *sectorp to be an offset in that zone
 290  */
 291 static struct strip_zone *find_zone(struct r0conf *conf,
 292                                     sector_t *sectorp)
 293 {
 294         int i;
 295         struct strip_zone *z = conf->strip_zone;
 296         sector_t sector = *sectorp;
 297
 298         for (i = 0; i < conf->nr_strip_zones; i++)
 299                 if (sector < z[i].zone_end) {
 300                         if (i)
 301                                 *sectorp = sector - z[i-1].zone_end;
 302                         return z + i;
 303                 }
 304         BUG();
 305 }
 306
 307 /*
 308  * remaps the bio to the target device. we separate two flows.
 309  * power 2 flow and a general flow for the sake of performance
 310 */
 311 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
 312                                 sector_t sector, sector_t *sector_offset)
 313 {
 314         unsigned int sect_in_chunk;
 315         sector_t chunk;
 316         struct r0conf *conf = mddev->private;
 317         int raid_disks = conf->strip_zone[0].nb_dev;
 318         unsigned int chunk_sects = mddev->chunk_sectors;
 319
 320         if (is_power_of_2(chunk_sects)) {
 321                 int chunksect_bits = ffz(~chunk_sects);
 322                 /* find the sector offset inside the chunk */
 323                 sect_in_chunk  = sector & (chunk_sects - 1);
 324                 sector >>= chunksect_bits;
 325                 /* chunk in zone */
 326                 chunk = *sector_offset;
 327                 /* quotient is the chunk in real device*/
 328                 sector_div(chunk, zone->nb_dev << chunksect_bits);
 329         } else{
 330                 sect_in_chunk = sector_div(sector, chunk_sects);
 331                 chunk = *sector_offset;
 332                 sector_div(chunk, chunk_sects * zone->nb_dev);
 333         }
 334         /*
 335         *  position the bio over the real device
 336         *  real sector = chunk in device + starting of zone
 337         *       + the position in the chunk
 338         */
 339         *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
 340         return conf->devlist[(zone - conf->strip_zone)*raid_disks
 341                              + sector_div(sector, zone->nb_dev)];
 342 }
 343
 344 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
 345 {
 346         sector_t array_sectors = 0;
 347         struct md_rdev *rdev;
 348
 349         WARN_ONCE(sectors || raid_disks,
 350                   "%s does not support generic reshape\n", __func__);
 351
 352         rdev_for_each(rdev, mddev)
 353                 array_sectors += (rdev->sectors &
 354                                   ~(sector_t)(mddev->chunk_sectors-1));
 355
 356         return array_sectors;
 357 }
 358
 359 static void raid0_free(struct mddev *mddev, void *priv);
 360
 361 static int raid0_run(struct mddev *mddev)
 362 {
 363         struct r0conf *conf;
 364         int ret;
 365
 366         if (mddev->chunk_sectors == 0) {
 367                 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev));
 368                 return -EINVAL;
 369         }
 370         if (md_check_no_bitmap(mddev))
 371                 return -EINVAL;
 372
 373         /* if private is not null, we are here after takeover */
 374         if (mddev->private == NULL) {
 375                 ret = create_strip_zones(mddev, &conf);
 376                 if (ret < 0)
 377                         return ret;
 378                 mddev->private = conf;
 379         }
 380         conf = mddev->private;
 381         if (mddev->queue) {
 382                 struct md_rdev *rdev;
 383                 bool discard_supported = false;
 384
 385                 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
 386                 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
 387                 blk_queue_max_write_zeroes_sectors(mddev->queue, mddev->chunk_sectors);
 388                 blk_queue_max_discard_sectors(mddev->queue, UINT_MAX);
 389
 390                 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
 391                 blk_queue_io_opt(mddev->queue,
 392                                  (mddev->chunk_sectors << 9) * mddev->raid_disks);
 393
 394                 rdev_for_each(rdev, mddev) {
 395                         disk_stack_limits(mddev->gendisk, rdev->bdev,
 396                                           rdev->data_offset << 9);
 397                         if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
 398                                 discard_supported = true;
 399                 }
 400                 if (!discard_supported)
 401                         queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
 402                 else
 403                         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
 404         }
 405
 406         /* calculate array device size */
 407         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 408
 409         pr_debug("md/raid0:%s: md_size is %llu sectors.\n",
 410                  mdname(mddev),
 411                  (unsigned long long)mddev->array_sectors);
 412
 413         if (mddev->queue) {
 414                 /* calculate the max read-ahead size.
 415                  * For read-ahead of large files to be effective, we need to
 416                  * readahead at least twice a whole stripe. i.e. number of devices
 417                  * multiplied by chunk size times 2.
 418                  * If an individual device has an ra_pages greater than the
 419                  * chunk size, then we will not drive that device as hard as it
 420                  * wants.  We consider this a configuration error: a larger
 421                  * chunksize should be used in that case.
 422                  */
 423                 int stripe = mddev->raid_disks *
 424                         (mddev->chunk_sectors << 9) / PAGE_SIZE;
 425                 if (mddev->queue->backing_dev_info->ra_pages < 2* stripe)
 426                         mddev->queue->backing_dev_info->ra_pages = 2* stripe;
 427         }
 428
 429         dump_zones(mddev);
 430
 431         ret = md_integrity_register(mddev);
 432
 433         return ret;
 434 }
 435
 436 static void raid0_free(struct mddev *mddev, void *priv)
 437 {
 438         struct r0conf *conf = priv;
 439
 440         kfree(conf->strip_zone);
 441         kfree(conf->devlist);
 442         kfree(conf);
 443 }
 444
 445 /*
 446  * Is io distribute over 1 or more chunks ?
 447 */
 448 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
 449                         unsigned int chunk_sects, struct bio *bio)
 450 {
 451         if (likely(is_power_of_2(chunk_sects))) {
 452                 return chunk_sects >=
 453                         ((bio->bi_iter.bi_sector & (chunk_sects-1))
 454                                         + bio_sectors(bio));
 455         } else{
 456                 sector_t sector = bio->bi_iter.bi_sector;
 457                 return chunk_sects >= (sector_div(sector, chunk_sects)
 458                                                 + bio_sectors(bio));
 459         }
 460 }
 461
 462 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio)
 463 {
 464         struct r0conf *conf = mddev->private;
 465         struct strip_zone *zone;
 466         sector_t start = bio->bi_iter.bi_sector;
 467         sector_t end;
 468         unsigned int stripe_size;
 469         sector_t first_stripe_index, last_stripe_index;
 470         sector_t start_disk_offset;
 471         unsigned int start_disk_index;
 472         sector_t end_disk_offset;
 473         unsigned int end_disk_index;
 474         unsigned int disk;
 475
 476         zone = find_zone(conf, &start);
 477
 478         if (bio_end_sector(bio) > zone->zone_end) {
 479                 struct bio *split = bio_split(bio,
 480                         zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO,
 481                         mddev->bio_set);
 482                 bio_chain(split, bio);
 483                 generic_make_request(bio);
 484                 bio = split;
 485                 end = zone->zone_end;
 486         } else
 487                 end = bio_end_sector(bio);
 488
 489         if (zone != conf->strip_zone)
 490                 end = end - zone[-1].zone_end;
 491
 492         /* Now start and end is the offset in zone */
 493         stripe_size = zone->nb_dev * mddev->chunk_sectors;
 494
 495         first_stripe_index = start;
 496         sector_div(first_stripe_index, stripe_size);
 497         last_stripe_index = end;
 498         sector_div(last_stripe_index, stripe_size);
 499
 500         start_disk_index = (int)(start - first_stripe_index * stripe_size) /
 501                 mddev->chunk_sectors;
 502         start_disk_offset = ((int)(start - first_stripe_index * stripe_size) %
 503                 mddev->chunk_sectors) +
 504                 first_stripe_index * mddev->chunk_sectors;
 505         end_disk_index = (int)(end - last_stripe_index * stripe_size) /
 506                 mddev->chunk_sectors;
 507         end_disk_offset = ((int)(end - last_stripe_index * stripe_size) %
 508                 mddev->chunk_sectors) +
 509                 last_stripe_index * mddev->chunk_sectors;
 510
 511         for (disk = 0; disk < zone->nb_dev; disk++) {
 512                 sector_t dev_start, dev_end;
 513                 struct bio *discard_bio = NULL;
 514                 struct md_rdev *rdev;
 515
 516                 if (disk < start_disk_index)
 517                         dev_start = (first_stripe_index + 1) *
 518                                 mddev->chunk_sectors;
 519                 else if (disk > start_disk_index)
 520                         dev_start = first_stripe_index * mddev->chunk_sectors;
 521                 else
 522                         dev_start = start_disk_offset;
 523
 524                 if (disk < end_disk_index)
 525                         dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
 526                 else if (disk > end_disk_index)
 527                         dev_end = last_stripe_index * mddev->chunk_sectors;
 528                 else
 529                         dev_end = end_disk_offset;
 530
 531                 if (dev_end <= dev_start)
 532                         continue;
 533
 534                 rdev = conf->devlist[(zone - conf->strip_zone) *
 535                         conf->strip_zone[0].nb_dev + disk];
 536                 if (__blkdev_issue_discard(rdev->bdev,
 537                         dev_start + zone->dev_start + rdev->data_offset,
 538                         dev_end - dev_start, GFP_NOIO, 0, &discard_bio) ||
 539                     !discard_bio)
 540                         continue;
 541                 bio_chain(discard_bio, bio);
 542                 if (mddev->gendisk)
 543                         trace_block_bio_remap(bdev_get_queue(rdev->bdev),
 544                                 discard_bio, disk_devt(mddev->gendisk),
 545                                 bio->bi_iter.bi_sector);
 546                 generic_make_request(discard_bio);
 547         }
 548         bio_endio(bio);
 549 }
 550
 551 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
 552 {
 553         struct strip_zone *zone;
 554         struct md_rdev *tmp_dev;
 555         sector_t bio_sector;
 556         sector_t sector;
 557         unsigned chunk_sects;
 558         unsigned sectors;
 559
 560         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
 561                 md_flush_request(mddev, bio);
 562                 return;
 563         }
 564
 565         if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) {
 566                 raid0_handle_discard(mddev, bio);
 567                 return;
 568         }
 569
 570         bio_sector = bio->bi_iter.bi_sector;
 571         sector = bio_sector;
 572         chunk_sects = mddev->chunk_sectors;
 573
 574         sectors = chunk_sects -
 575                 (likely(is_power_of_2(chunk_sects))
 576                  ? (sector & (chunk_sects-1))
 577                  : sector_div(sector, chunk_sects));
 578
 579         /* Restore due to sector_div */
 580         sector = bio_sector;
 581
 582         if (sectors < bio_sectors(bio)) {
 583                 struct bio *split = bio_split(bio, sectors, GFP_NOIO, mddev->bio_set);
 584                 bio_chain(split, bio);
 585                 generic_make_request(bio);
 586                 bio = split;
 587         }
 588
 589         zone = find_zone(mddev->private, &sector);
 590         tmp_dev = map_sector(mddev, zone, sector, &sector);
 591         bio->bi_bdev = 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(bdev_get_queue(bio->bi_bdev),
 597                                       bio, disk_devt(mddev->gendisk),
 598                                       bio_sector);
 599         mddev_check_writesame(mddev, bio);
 600         mddev_check_write_zeroes(mddev, bio);
 601         generic_make_request(bio);
 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 state)
 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         .congested      = raid0_congested,
 786 };
 787
 788 static int __init raid0_init (void)
 789 {
 790         return register_md_personality (&raid0_personality);
 791 }
 792
 793 static void raid0_exit (void)
 794 {
 795         unregister_md_personality (&raid0_personality);
 796 }
 797
 798 module_init(raid0_init);
 799 module_exit(raid0_exit);
 800 MODULE_LICENSE("GPL");
 801 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
 802 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 803 MODULE_ALIAS("md-raid0");
 804 MODULE_ALIAS("md-level-0");