drivers/md/raid0.c

   1 /*
   2    raid0.c : Multiple Devices driver for Linux
   3              Copyright (C) 1994-96 Marc ZYNGIER
   4              <zyngier@ufr-info-p7.ibp.fr> or
   5              <maz@gloups.fdn.fr>
   6              Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
   7
   8
   9    RAID-0 management functions.
  10
  11    This program is free software; you can redistribute it and/or modify
  12    it under the terms of the GNU General Public License as published by
  13    the Free Software Foundation; either version 2, or (at your option)
  14    any later version.
  15
  16    You should have received a copy of the GNU General Public License
  17    (for example /usr/src/linux/COPYING); if not, write to the Free
  18    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  19 */
  20
  21 #include <linux/blkdev.h>
  22 #include <linux/seq_file.h>
  23 #include <linux/module.h>
  24 #include <linux/slab.h>
  25 #include "md.h"
  26 #include "raid0.h"
  27 #include "raid5.h"
  28
  29 static int raid0_congested(void *data, int bits)
  30 {
  31         struct mddev *mddev = data;
  32         struct r0conf *conf = mddev->private;
  33         struct md_rdev **devlist = conf->devlist;
  34         int raid_disks = conf->strip_zone[0].nb_dev;
  35         int i, ret = 0;
  36
  37         if (mddev_congested(mddev, bits))
  38                 return 1;
  39
  40         for (i = 0; i < raid_disks && !ret ; i++) {
  41                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  42
  43                 ret |= bdi_congested(&q->backing_dev_info, bits);
  44         }
  45         return ret;
  46 }
  47
  48 /*
  49  * inform the user of the raid configuration
  50 */
  51 static void dump_zones(struct mddev *mddev)
  52 {
  53         int j, k;
  54         sector_t zone_size = 0;
  55         sector_t zone_start = 0;
  56         char b[BDEVNAME_SIZE];
  57         struct r0conf *conf = mddev->private;
  58         int raid_disks = conf->strip_zone[0].nb_dev;
  59         printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
  60                mdname(mddev),
  61                conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
  62         for (j = 0; j < conf->nr_strip_zones; j++) {
  63                 printk(KERN_INFO "md: zone%d=[", j);
  64                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
  65                         printk(KERN_CONT "%s%s", k?"/":"",
  66                         bdevname(conf->devlist[j*raid_disks
  67                                                 + k]->bdev, b));
  68                 printk(KERN_CONT "]\n");
  69
  70                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
  71                 printk(KERN_INFO "      zone-offset=%10lluKB, "
  72                                 "device-offset=%10lluKB, size=%10lluKB\n",
  73                         (unsigned long long)zone_start>>1,
  74                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
  75                         (unsigned long long)zone_size>>1);
  76                 zone_start = conf->strip_zone[j].zone_end;
  77         }
  78         printk(KERN_INFO "\n");
  79 }
  80
  81 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
  82 {
  83         int i, c, err;
  84         sector_t curr_zone_end, sectors;
  85         struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
  86         struct strip_zone *zone;
  87         int cnt;
  88         char b[BDEVNAME_SIZE];
  89         char b2[BDEVNAME_SIZE];
  90         struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
  91
  92         if (!conf)
  93                 return -ENOMEM;
  94         list_for_each_entry(rdev1, &mddev->disks, same_set) {
  95                 pr_debug("md/raid0:%s: looking at %s\n",
  96                          mdname(mddev),
  97                          bdevname(rdev1->bdev, b));
  98                 c = 0;
  99
 100                 /* round size to chunk_size */
 101                 sectors = rdev1->sectors;
 102                 sector_div(sectors, mddev->chunk_sectors);
 103                 rdev1->sectors = sectors * mddev->chunk_sectors;
 104
 105                 list_for_each_entry(rdev2, &mddev->disks, same_set) {
 106                         pr_debug("md/raid0:%s:   comparing %s(%llu)"
 107                                  " with %s(%llu)\n",
 108                                  mdname(mddev),
 109                                  bdevname(rdev1->bdev,b),
 110                                  (unsigned long long)rdev1->sectors,
 111                                  bdevname(rdev2->bdev,b2),
 112                                  (unsigned long long)rdev2->sectors);
 113                         if (rdev2 == rdev1) {
 114                                 pr_debug("md/raid0:%s:   END\n",
 115                                          mdname(mddev));
 116                                 break;
 117                         }
 118                         if (rdev2->sectors == rdev1->sectors) {
 119                                 /*
 120                                  * Not unique, don't count it as a new
 121                                  * group
 122                                  */
 123                                 pr_debug("md/raid0:%s:   EQUAL\n",
 124                                          mdname(mddev));
 125                                 c = 1;
 126                                 break;
 127                         }
 128                         pr_debug("md/raid0:%s:   NOT EQUAL\n",
 129                                  mdname(mddev));
 130                 }
 131                 if (!c) {
 132                         pr_debug("md/raid0:%s:   ==> UNIQUE\n",
 133                                  mdname(mddev));
 134                         conf->nr_strip_zones++;
 135                         pr_debug("md/raid0:%s: %d zones\n",
 136                                  mdname(mddev), conf->nr_strip_zones);
 137                 }
 138         }
 139         pr_debug("md/raid0:%s: FINAL %d zones\n",
 140                  mdname(mddev), conf->nr_strip_zones);
 141         err = -ENOMEM;
 142         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 143                                 conf->nr_strip_zones, GFP_KERNEL);
 144         if (!conf->strip_zone)
 145                 goto abort;
 146         conf->devlist = kzalloc(sizeof(struct md_rdev*)*
 147                                 conf->nr_strip_zones*mddev->raid_disks,
 148                                 GFP_KERNEL);
 149         if (!conf->devlist)
 150                 goto abort;
 151
 152         /* The first zone must contain all devices, so here we check that
 153          * there is a proper alignment of slots to devices and find them all
 154          */
 155         zone = &conf->strip_zone[0];
 156         cnt = 0;
 157         smallest = NULL;
 158         dev = conf->devlist;
 159         err = -EINVAL;
 160         list_for_each_entry(rdev1, &mddev->disks, same_set) {
 161                 int j = rdev1->raid_disk;
 162
 163                 if (mddev->level == 10) {
 164                         /* taking over a raid10-n2 array */
 165                         j /= 2;
 166                         rdev1->new_raid_disk = j;
 167                 }
 168
 169                 if (mddev->level == 1) {
 170                         /* taiking over a raid1 array-
 171                          * we have only one active disk
 172                          */
 173                         j = 0;
 174                         rdev1->new_raid_disk = j;
 175                 }
 176
 177                 if (j < 0 || j >= mddev->raid_disks) {
 178                         printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
 179                                "aborting!\n", mdname(mddev), j);
 180                         goto abort;
 181                 }
 182                 if (dev[j]) {
 183                         printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
 184                                "aborting!\n", mdname(mddev), j);
 185                         goto abort;
 186                 }
 187                 dev[j] = rdev1;
 188
 189                 disk_stack_limits(mddev->gendisk, rdev1->bdev,
 190                                   rdev1->data_offset << 9);
 191                 /* as we don't honour merge_bvec_fn, we must never risk
 192                  * violating it, so limit ->max_segments to 1, lying within
 193                  * a single page.
 194                  */
 195
 196                 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) {
 197                         blk_queue_max_segments(mddev->queue, 1);
 198                         blk_queue_segment_boundary(mddev->queue,
 199                                                    PAGE_CACHE_SIZE - 1);
 200                 }
 201                 if (!smallest || (rdev1->sectors < smallest->sectors))
 202                         smallest = rdev1;
 203                 cnt++;
 204         }
 205         if (cnt != mddev->raid_disks) {
 206                 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
 207                        "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
 208                 goto abort;
 209         }
 210         zone->nb_dev = cnt;
 211         zone->zone_end = smallest->sectors * cnt;
 212
 213         curr_zone_end = zone->zone_end;
 214
 215         /* now do the other zones */
 216         for (i = 1; i < conf->nr_strip_zones; i++)
 217         {
 218                 int j;
 219
 220                 zone = conf->strip_zone + i;
 221                 dev = conf->devlist + i * mddev->raid_disks;
 222
 223                 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
 224                 zone->dev_start = smallest->sectors;
 225                 smallest = NULL;
 226                 c = 0;
 227
 228                 for (j=0; j<cnt; j++) {
 229                         rdev = conf->devlist[j];
 230                         if (rdev->sectors <= zone->dev_start) {
 231                                 pr_debug("md/raid0:%s: checking %s ... nope\n",
 232                                          mdname(mddev),
 233                                          bdevname(rdev->bdev, b));
 234                                 continue;
 235                         }
 236                         pr_debug("md/raid0:%s: checking %s ..."
 237                                  " contained as device %d\n",
 238                                  mdname(mddev),
 239                                  bdevname(rdev->bdev, b), c);
 240                         dev[c] = rdev;
 241                         c++;
 242                         if (!smallest || rdev->sectors < smallest->sectors) {
 243                                 smallest = rdev;
 244                                 pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
 245                                          mdname(mddev),
 246                                          (unsigned long long)rdev->sectors);
 247                         }
 248                 }
 249
 250                 zone->nb_dev = c;
 251                 sectors = (smallest->sectors - zone->dev_start) * c;
 252                 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
 253                          mdname(mddev),
 254                          zone->nb_dev, (unsigned long long)sectors);
 255
 256                 curr_zone_end += sectors;
 257                 zone->zone_end = curr_zone_end;
 258
 259                 pr_debug("md/raid0:%s: current zone start: %llu\n",
 260                          mdname(mddev),
 261                          (unsigned long long)smallest->sectors);
 262         }
 263         mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 264         mddev->queue->backing_dev_info.congested_data = mddev;
 265
 266         /*
 267          * now since we have the hard sector sizes, we can make sure
 268          * chunk size is a multiple of that sector size
 269          */
 270         if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
 271                 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
 272                        mdname(mddev),
 273                        mddev->chunk_sectors << 9);
 274                 goto abort;
 275         }
 276
 277         blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
 278         blk_queue_io_opt(mddev->queue,
 279                          (mddev->chunk_sectors << 9) * mddev->raid_disks);
 280
 281         pr_debug("md/raid0:%s: done.\n", mdname(mddev));
 282         *private_conf = conf;
 283
 284         return 0;
 285 abort:
 286         kfree(conf->strip_zone);
 287         kfree(conf->devlist);
 288         kfree(conf);
 289         *private_conf = NULL;
 290         return err;
 291 }
 292
 293 /**
 294  *      raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 295  *      @q: request queue
 296  *      @bvm: properties of new bio
 297  *      @biovec: the request that could be merged to it.
 298  *
 299  *      Return amount of bytes we can accept at this offset
 300  */
 301 static int raid0_mergeable_bvec(struct request_queue *q,
 302                                 struct bvec_merge_data *bvm,
 303                                 struct bio_vec *biovec)
 304 {
 305         struct mddev *mddev = q->queuedata;
 306         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 307         int max;
 308         unsigned int chunk_sectors = mddev->chunk_sectors;
 309         unsigned int bio_sectors = bvm->bi_size >> 9;
 310
 311         if (is_power_of_2(chunk_sectors))
 312                 max =  (chunk_sectors - ((sector & (chunk_sectors-1))
 313                                                 + bio_sectors)) << 9;
 314         else
 315                 max =  (chunk_sectors - (sector_div(sector, chunk_sectors)
 316                                                 + bio_sectors)) << 9;
 317         if (max < 0) max = 0; /* bio_add cannot handle a negative return */
 318         if (max <= biovec->bv_len && bio_sectors == 0)
 319                 return biovec->bv_len;
 320         else
 321                 return max;
 322 }
 323
 324 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
 325 {
 326         sector_t array_sectors = 0;
 327         struct md_rdev *rdev;
 328
 329         WARN_ONCE(sectors || raid_disks,
 330                   "%s does not support generic reshape\n", __func__);
 331
 332         list_for_each_entry(rdev, &mddev->disks, same_set)
 333                 array_sectors += rdev->sectors;
 334
 335         return array_sectors;
 336 }
 337
 338 static int raid0_run(struct mddev *mddev)
 339 {
 340         struct r0conf *conf;
 341         int ret;
 342
 343         if (mddev->chunk_sectors == 0) {
 344                 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
 345                        mdname(mddev));
 346                 return -EINVAL;
 347         }
 348         if (md_check_no_bitmap(mddev))
 349                 return -EINVAL;
 350         blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
 351
 352         /* if private is not null, we are here after takeover */
 353         if (mddev->private == NULL) {
 354                 ret = create_strip_zones(mddev, &conf);
 355                 if (ret < 0)
 356                         return ret;
 357                 mddev->private = conf;
 358         }
 359         conf = mddev->private;
 360
 361         /* calculate array device size */
 362         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 363
 364         printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
 365                mdname(mddev),
 366                (unsigned long long)mddev->array_sectors);
 367         /* calculate the max read-ahead size.
 368          * For read-ahead of large files to be effective, we need to
 369          * readahead at least twice a whole stripe. i.e. number of devices
 370          * multiplied by chunk size times 2.
 371          * If an individual device has an ra_pages greater than the
 372          * chunk size, then we will not drive that device as hard as it
 373          * wants.  We consider this a configuration error: a larger
 374          * chunksize should be used in that case.
 375          */
 376         {
 377                 int stripe = mddev->raid_disks *
 378                         (mddev->chunk_sectors << 9) / PAGE_SIZE;
 379                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 380                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 381         }
 382
 383         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 384         dump_zones(mddev);
 385         return md_integrity_register(mddev);
 386 }
 387
 388 static int raid0_stop(struct mddev *mddev)
 389 {
 390         struct r0conf *conf = mddev->private;
 391
 392         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 393         kfree(conf->strip_zone);
 394         kfree(conf->devlist);
 395         kfree(conf);
 396         mddev->private = NULL;
 397         return 0;
 398 }
 399
 400 /* Find the zone which holds a particular offset
 401  * Update *sectorp to be an offset in that zone
 402  */
 403 static struct strip_zone *find_zone(struct r0conf *conf,
 404                                     sector_t *sectorp)
 405 {
 406         int i;
 407         struct strip_zone *z = conf->strip_zone;
 408         sector_t sector = *sectorp;
 409
 410         for (i = 0; i < conf->nr_strip_zones; i++)
 411                 if (sector < z[i].zone_end) {
 412                         if (i)
 413                                 *sectorp = sector - z[i-1].zone_end;
 414                         return z + i;
 415                 }
 416         BUG();
 417 }
 418
 419 /*
 420  * remaps the bio to the target device. we separate two flows.
 421  * power 2 flow and a general flow for the sake of perfromance
 422 */
 423 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
 424                                 sector_t sector, sector_t *sector_offset)
 425 {
 426         unsigned int sect_in_chunk;
 427         sector_t chunk;
 428         struct r0conf *conf = mddev->private;
 429         int raid_disks = conf->strip_zone[0].nb_dev;
 430         unsigned int chunk_sects = mddev->chunk_sectors;
 431
 432         if (is_power_of_2(chunk_sects)) {
 433                 int chunksect_bits = ffz(~chunk_sects);
 434                 /* find the sector offset inside the chunk */
 435                 sect_in_chunk  = sector & (chunk_sects - 1);
 436                 sector >>= chunksect_bits;
 437                 /* chunk in zone */
 438                 chunk = *sector_offset;
 439                 /* quotient is the chunk in real device*/
 440                 sector_div(chunk, zone->nb_dev << chunksect_bits);
 441         } else{
 442                 sect_in_chunk = sector_div(sector, chunk_sects);
 443                 chunk = *sector_offset;
 444                 sector_div(chunk, chunk_sects * zone->nb_dev);
 445         }
 446         /*
 447         *  position the bio over the real device
 448         *  real sector = chunk in device + starting of zone
 449         *       + the position in the chunk
 450         */
 451         *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
 452         return conf->devlist[(zone - conf->strip_zone)*raid_disks
 453                              + sector_div(sector, zone->nb_dev)];
 454 }
 455
 456 /*
 457  * Is io distribute over 1 or more chunks ?
 458 */
 459 static inline int is_io_in_chunk_boundary(struct mddev *mddev,
 460                         unsigned int chunk_sects, struct bio *bio)
 461 {
 462         if (likely(is_power_of_2(chunk_sects))) {
 463                 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
 464                                         + (bio->bi_size >> 9));
 465         } else{
 466                 sector_t sector = bio->bi_sector;
 467                 return chunk_sects >= (sector_div(sector, chunk_sects)
 468                                                 + (bio->bi_size >> 9));
 469         }
 470 }
 471
 472 static void raid0_make_request(struct mddev *mddev, struct bio *bio)
 473 {
 474         unsigned int chunk_sects;
 475         sector_t sector_offset;
 476         struct strip_zone *zone;
 477         struct md_rdev *tmp_dev;
 478
 479         if (unlikely(bio->bi_rw & REQ_FLUSH)) {
 480                 md_flush_request(mddev, bio);
 481                 return;
 482         }
 483
 484         chunk_sects = mddev->chunk_sectors;
 485         if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
 486                 sector_t sector = bio->bi_sector;
 487                 struct bio_pair *bp;
 488                 /* Sanity check -- queue functions should prevent this happening */
 489                 if (bio->bi_vcnt != 1 ||
 490                     bio->bi_idx != 0)
 491                         goto bad_map;
 492                 /* This is a one page bio that upper layers
 493                  * refuse to split for us, so we need to split it.
 494                  */
 495                 if (likely(is_power_of_2(chunk_sects)))
 496                         bp = bio_split(bio, chunk_sects - (sector &
 497                                                            (chunk_sects-1)));
 498                 else
 499                         bp = bio_split(bio, chunk_sects -
 500                                        sector_div(sector, chunk_sects));
 501                 raid0_make_request(mddev, &bp->bio1);
 502                 raid0_make_request(mddev, &bp->bio2);
 503                 bio_pair_release(bp);
 504                 return;
 505         }
 506
 507         sector_offset = bio->bi_sector;
 508         zone =  find_zone(mddev->private, &sector_offset);
 509         tmp_dev = map_sector(mddev, zone, bio->bi_sector,
 510                              &sector_offset);
 511         bio->bi_bdev = tmp_dev->bdev;
 512         bio->bi_sector = sector_offset + zone->dev_start +
 513                 tmp_dev->data_offset;
 514
 515         generic_make_request(bio);
 516         return;
 517
 518 bad_map:
 519         printk("md/raid0:%s: make_request bug: can't convert block across chunks"
 520                " or bigger than %dk %llu %d\n",
 521                mdname(mddev), chunk_sects / 2,
 522                (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 523
 524         bio_io_error(bio);
 525         return;
 526 }
 527
 528 static void raid0_status(struct seq_file *seq, struct mddev *mddev)
 529 {
 530         seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
 531         return;
 532 }
 533
 534 static void *raid0_takeover_raid45(struct mddev *mddev)
 535 {
 536         struct md_rdev *rdev;
 537         struct r0conf *priv_conf;
 538
 539         if (mddev->degraded != 1) {
 540                 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
 541                        mdname(mddev),
 542                        mddev->degraded);
 543                 return ERR_PTR(-EINVAL);
 544         }
 545
 546         list_for_each_entry(rdev, &mddev->disks, same_set) {
 547                 /* check slot number for a disk */
 548                 if (rdev->raid_disk == mddev->raid_disks-1) {
 549                         printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
 550                                mdname(mddev));
 551                         return ERR_PTR(-EINVAL);
 552                 }
 553         }
 554
 555         /* Set new parameters */
 556         mddev->new_level = 0;
 557         mddev->new_layout = 0;
 558         mddev->new_chunk_sectors = mddev->chunk_sectors;
 559         mddev->raid_disks--;
 560         mddev->delta_disks = -1;
 561         /* make sure it will be not marked as dirty */
 562         mddev->recovery_cp = MaxSector;
 563
 564         create_strip_zones(mddev, &priv_conf);
 565         return priv_conf;
 566 }
 567
 568 static void *raid0_takeover_raid10(struct mddev *mddev)
 569 {
 570         struct r0conf *priv_conf;
 571
 572         /* Check layout:
 573          *  - far_copies must be 1
 574          *  - near_copies must be 2
 575          *  - disks number must be even
 576          *  - all mirrors must be already degraded
 577          */
 578         if (mddev->layout != ((1 << 8) + 2)) {
 579                 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
 580                        mdname(mddev),
 581                        mddev->layout);
 582                 return ERR_PTR(-EINVAL);
 583         }
 584         if (mddev->raid_disks & 1) {
 585                 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
 586                        mdname(mddev));
 587                 return ERR_PTR(-EINVAL);
 588         }
 589         if (mddev->degraded != (mddev->raid_disks>>1)) {
 590                 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
 591                        mdname(mddev));
 592                 return ERR_PTR(-EINVAL);
 593         }
 594
 595         /* Set new parameters */
 596         mddev->new_level = 0;
 597         mddev->new_layout = 0;
 598         mddev->new_chunk_sectors = mddev->chunk_sectors;
 599         mddev->delta_disks = - mddev->raid_disks / 2;
 600         mddev->raid_disks += mddev->delta_disks;
 601         mddev->degraded = 0;
 602         /* make sure it will be not marked as dirty */
 603         mddev->recovery_cp = MaxSector;
 604
 605         create_strip_zones(mddev, &priv_conf);
 606         return priv_conf;
 607 }
 608
 609 static void *raid0_takeover_raid1(struct mddev *mddev)
 610 {
 611         struct r0conf *priv_conf;
 612
 613         /* Check layout:
 614          *  - (N - 1) mirror drives must be already faulty
 615          */
 616         if ((mddev->raid_disks - 1) != mddev->degraded) {
 617                 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
 618                        mdname(mddev));
 619                 return ERR_PTR(-EINVAL);
 620         }
 621
 622         /* Set new parameters */
 623         mddev->new_level = 0;
 624         mddev->new_layout = 0;
 625         mddev->new_chunk_sectors = 128; /* by default set chunk size to 64k */
 626         mddev->delta_disks = 1 - mddev->raid_disks;
 627         mddev->raid_disks = 1;
 628         /* make sure it will be not marked as dirty */
 629         mddev->recovery_cp = MaxSector;
 630
 631         create_strip_zones(mddev, &priv_conf);
 632         return priv_conf;
 633 }
 634
 635 static void *raid0_takeover(struct mddev *mddev)
 636 {
 637         /* raid0 can take over:
 638          *  raid4 - if all data disks are active.
 639          *  raid5 - providing it is Raid4 layout and one disk is faulty
 640          *  raid10 - assuming we have all necessary active disks
 641          *  raid1 - with (N -1) mirror drives faulty
 642          */
 643         if (mddev->level == 4)
 644                 return raid0_takeover_raid45(mddev);
 645
 646         if (mddev->level == 5) {
 647                 if (mddev->layout == ALGORITHM_PARITY_N)
 648                         return raid0_takeover_raid45(mddev);
 649
 650                 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
 651                        mdname(mddev), ALGORITHM_PARITY_N);
 652         }
 653
 654         if (mddev->level == 10)
 655                 return raid0_takeover_raid10(mddev);
 656
 657         if (mddev->level == 1)
 658                 return raid0_takeover_raid1(mddev);
 659
 660         printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
 661                 mddev->level);
 662
 663         return ERR_PTR(-EINVAL);
 664 }
 665
 666 static void raid0_quiesce(struct mddev *mddev, int state)
 667 {
 668 }
 669
 670 static struct md_personality raid0_personality=
 671 {
 672         .name           = "raid0",
 673         .level          = 0,
 674         .owner          = THIS_MODULE,
 675         .make_request   = raid0_make_request,
 676         .run            = raid0_run,
 677         .stop           = raid0_stop,
 678         .status         = raid0_status,
 679         .size           = raid0_size,
 680         .takeover       = raid0_takeover,
 681         .quiesce        = raid0_quiesce,
 682 };
 683
 684 static int __init raid0_init (void)
 685 {
 686         return register_md_personality (&raid0_personality);
 687 }
 688
 689 static void raid0_exit (void)
 690 {
 691         unregister_md_personality (&raid0_personality);
 692 }
 693
 694 module_init(raid0_init);
 695 module_exit(raid0_exit);
 696 MODULE_LICENSE("GPL");
 697 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
 698 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 699 MODULE_ALIAS("md-raid0");
 700 MODULE_ALIAS("md-level-0");