SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 "md.h"
  24 #include "raid0.h"
  25 #include "raid5.h"
  26
  27 static void raid0_unplug(struct request_queue *q)
  28 {
  29         mddev_t *mddev = q->queuedata;
  30         raid0_conf_t *conf = mddev->private;
  31         mdk_rdev_t **devlist = conf->devlist;
  32         int raid_disks = conf->strip_zone[0].nb_dev;
  33         int i;
  34
  35         for (i=0; i < raid_disks; i++) {
  36                 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
  37
  38                 blk_unplug(r_queue);
  39         }
  40 }
  41
  42 static int raid0_congested(void *data, int bits)
  43 {
  44         mddev_t *mddev = data;
  45         raid0_conf_t *conf = mddev->private;
  46         mdk_rdev_t **devlist = conf->devlist;
  47         int raid_disks = conf->strip_zone[0].nb_dev;
  48         int i, ret = 0;
  49
  50         if (mddev_congested(mddev, bits))
  51                 return 1;
  52
  53         for (i = 0; i < raid_disks && !ret ; i++) {
  54                 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
  55
  56                 ret |= bdi_congested(&q->backing_dev_info, bits);
  57         }
  58         return ret;
  59 }
  60
  61 /*
  62  * inform the user of the raid configuration
  63 */
  64 static void dump_zones(mddev_t *mddev)
  65 {
  66         int j, k, h;
  67         sector_t zone_size = 0;
  68         sector_t zone_start = 0;
  69         char b[BDEVNAME_SIZE];
  70         raid0_conf_t *conf = mddev->private;
  71         int raid_disks = conf->strip_zone[0].nb_dev;
  72         printk(KERN_INFO "******* %s configuration *********\n",
  73                 mdname(mddev));
  74         h = 0;
  75         for (j = 0; j < conf->nr_strip_zones; j++) {
  76                 printk(KERN_INFO "zone%d=[", j);
  77                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
  78                         printk("%s/",
  79                         bdevname(conf->devlist[j*raid_disks
  80                                                 + k]->bdev, b));
  81                 printk("]\n");
  82
  83                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
  84                 printk(KERN_INFO "        zone offset=%llukb "
  85                                 "device offset=%llukb size=%llukb\n",
  86                         (unsigned long long)zone_start>>1,
  87                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
  88                         (unsigned long long)zone_size>>1);
  89                 zone_start = conf->strip_zone[j].zone_end;
  90         }
  91         printk(KERN_INFO "**********************************\n\n");
  92 }
  93
  94 static int create_strip_zones(mddev_t *mddev, raid0_conf_t **private_conf)
  95 {
  96         int i, c, err;
  97         sector_t curr_zone_end, sectors;
  98         mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev, **dev;
  99         struct strip_zone *zone;
 100         int cnt;
 101         char b[BDEVNAME_SIZE];
 102         raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
 103
 104         if (!conf)
 105                 return -ENOMEM;
 106         list_for_each_entry(rdev1, &mddev->disks, same_set) {
 107                 printk(KERN_INFO "raid0: looking at %s\n",
 108                         bdevname(rdev1->bdev,b));
 109                 c = 0;
 110
 111                 /* round size to chunk_size */
 112                 sectors = rdev1->sectors;
 113                 sector_div(sectors, mddev->chunk_sectors);
 114                 rdev1->sectors = sectors * mddev->chunk_sectors;
 115
 116                 list_for_each_entry(rdev2, &mddev->disks, same_set) {
 117                         printk(KERN_INFO "raid0:   comparing %s(%llu)",
 118                                bdevname(rdev1->bdev,b),
 119                                (unsigned long long)rdev1->sectors);
 120                         printk(KERN_INFO " with %s(%llu)\n",
 121                                bdevname(rdev2->bdev,b),
 122                                (unsigned long long)rdev2->sectors);
 123                         if (rdev2 == rdev1) {
 124                                 printk(KERN_INFO "raid0:   END\n");
 125                                 break;
 126                         }
 127                         if (rdev2->sectors == rdev1->sectors) {
 128                                 /*
 129                                  * Not unique, don't count it as a new
 130                                  * group
 131                                  */
 132                                 printk(KERN_INFO "raid0:   EQUAL\n");
 133                                 c = 1;
 134                                 break;
 135                         }
 136                         printk(KERN_INFO "raid0:   NOT EQUAL\n");
 137                 }
 138                 if (!c) {
 139                         printk(KERN_INFO "raid0:   ==> UNIQUE\n");
 140                         conf->nr_strip_zones++;
 141                         printk(KERN_INFO "raid0: %d zones\n",
 142                                 conf->nr_strip_zones);
 143                 }
 144         }
 145         printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
 146         err = -ENOMEM;
 147         conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 148                                 conf->nr_strip_zones, GFP_KERNEL);
 149         if (!conf->strip_zone)
 150                 goto abort;
 151         conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
 152                                 conf->nr_strip_zones*mddev->raid_disks,
 153                                 GFP_KERNEL);
 154         if (!conf->devlist)
 155                 goto abort;
 156
 157         /* The first zone must contain all devices, so here we check that
 158          * there is a proper alignment of slots to devices and find them all
 159          */
 160         zone = &conf->strip_zone[0];
 161         cnt = 0;
 162         smallest = NULL;
 163         dev = conf->devlist;
 164         err = -EINVAL;
 165         list_for_each_entry(rdev1, &mddev->disks, same_set) {
 166                 int j = rdev1->raid_disk;
 167
 168                 if (mddev->level == 10)
 169                         /* taking over a raid10-n2 array */
 170                         j /= 2;
 171
 172                 if (j < 0 || j >= mddev->raid_disks) {
 173                         printk(KERN_ERR "raid0: bad disk number %d - "
 174                                 "aborting!\n", j);
 175                         goto abort;
 176                 }
 177                 if (dev[j]) {
 178                         printk(KERN_ERR "raid0: multiple devices for %d - "
 179                                 "aborting!\n", j);
 180                         goto abort;
 181                 }
 182                 dev[j] = rdev1;
 183
 184                 disk_stack_limits(mddev->gendisk, rdev1->bdev,
 185                                   rdev1->data_offset << 9);
 186                 /* as we don't honour merge_bvec_fn, we must never risk
 187                  * violating it, so limit ->max_segments to 1, lying within
 188                  * a single page.
 189                  */
 190
 191                 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) {
 192                         blk_queue_max_segments(mddev->queue, 1);
 193                         blk_queue_segment_boundary(mddev->queue,
 194                                                    PAGE_CACHE_SIZE - 1);
 195                 }
 196                 if (!smallest || (rdev1->sectors < smallest->sectors))
 197                         smallest = rdev1;
 198                 cnt++;
 199         }
 200         if (cnt != mddev->raid_disks) {
 201                 printk(KERN_ERR "raid0: too few disks (%d of %d) - "
 202                         "aborting!\n", cnt, mddev->raid_disks);
 203                 goto abort;
 204         }
 205         zone->nb_dev = cnt;
 206         zone->zone_end = smallest->sectors * cnt;
 207
 208         curr_zone_end = zone->zone_end;
 209
 210         /* now do the other zones */
 211         for (i = 1; i < conf->nr_strip_zones; i++)
 212         {
 213                 int j;
 214
 215                 zone = conf->strip_zone + i;
 216                 dev = conf->devlist + i * mddev->raid_disks;
 217
 218                 printk(KERN_INFO "raid0: zone %d\n", i);
 219                 zone->dev_start = smallest->sectors;
 220                 smallest = NULL;
 221                 c = 0;
 222
 223                 for (j=0; j<cnt; j++) {
 224                         rdev = conf->devlist[j];
 225                         printk(KERN_INFO "raid0: checking %s ...",
 226                                 bdevname(rdev->bdev, b));
 227                         if (rdev->sectors <= zone->dev_start) {
 228                                 printk(KERN_INFO " nope.\n");
 229                                 continue;
 230                         }
 231                         printk(KERN_INFO " contained as device %d\n", c);
 232                         dev[c] = rdev;
 233                         c++;
 234                         if (!smallest || rdev->sectors < smallest->sectors) {
 235                                 smallest = rdev;
 236                                 printk(KERN_INFO "  (%llu) is smallest!.\n",
 237                                         (unsigned long long)rdev->sectors);
 238                         }
 239                 }
 240
 241                 zone->nb_dev = c;
 242                 sectors = (smallest->sectors - zone->dev_start) * c;
 243                 printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
 244                         zone->nb_dev, (unsigned long long)sectors);
 245
 246                 curr_zone_end += sectors;
 247                 zone->zone_end = curr_zone_end;
 248
 249                 printk(KERN_INFO "raid0: current zone start: %llu\n",
 250                         (unsigned long long)smallest->sectors);
 251         }
 252         mddev->queue->unplug_fn = raid0_unplug;
 253         mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 254         mddev->queue->backing_dev_info.congested_data = mddev;
 255
 256         /*
 257          * now since we have the hard sector sizes, we can make sure
 258          * chunk size is a multiple of that sector size
 259          */
 260         if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
 261                 printk(KERN_ERR "%s chunk_size of %d not valid\n",
 262                        mdname(mddev),
 263                        mddev->chunk_sectors << 9);
 264                 goto abort;
 265         }
 266
 267         blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
 268         blk_queue_io_opt(mddev->queue,
 269                          (mddev->chunk_sectors << 9) * mddev->raid_disks);
 270
 271         printk(KERN_INFO "raid0: done.\n");
 272         *private_conf = conf;
 273
 274         return 0;
 275 abort:
 276         kfree(conf->strip_zone);
 277         kfree(conf->devlist);
 278         kfree(conf);
 279         *private_conf = NULL;
 280         return err;
 281 }
 282
 283 /**
 284  *      raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
 285  *      @q: request queue
 286  *      @bvm: properties of new bio
 287  *      @biovec: the request that could be merged to it.
 288  *
 289  *      Return amount of bytes we can accept at this offset
 290  */
 291 static int raid0_mergeable_bvec(struct request_queue *q,
 292                                 struct bvec_merge_data *bvm,
 293                                 struct bio_vec *biovec)
 294 {
 295         mddev_t *mddev = q->queuedata;
 296         sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 297         int max;
 298         unsigned int chunk_sectors = mddev->chunk_sectors;
 299         unsigned int bio_sectors = bvm->bi_size >> 9;
 300
 301         if (is_power_of_2(chunk_sectors))
 302                 max =  (chunk_sectors - ((sector & (chunk_sectors-1))
 303                                                 + bio_sectors)) << 9;
 304         else
 305                 max =  (chunk_sectors - (sector_div(sector, chunk_sectors)
 306                                                 + bio_sectors)) << 9;
 307         if (max < 0) max = 0; /* bio_add cannot handle a negative return */
 308         if (max <= biovec->bv_len && bio_sectors == 0)
 309                 return biovec->bv_len;
 310         else
 311                 return max;
 312 }
 313
 314 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 315 {
 316         sector_t array_sectors = 0;
 317         mdk_rdev_t *rdev;
 318
 319         WARN_ONCE(sectors || raid_disks,
 320                   "%s does not support generic reshape\n", __func__);
 321
 322         list_for_each_entry(rdev, &mddev->disks, same_set)
 323                 array_sectors += rdev->sectors;
 324
 325         return array_sectors;
 326 }
 327
 328 static int raid0_run(mddev_t *mddev)
 329 {
 330         raid0_conf_t *conf;
 331         int ret;
 332
 333         if (mddev->chunk_sectors == 0) {
 334                 printk(KERN_ERR "md/raid0: chunk size must be set.\n");
 335                 return -EINVAL;
 336         }
 337         if (md_check_no_bitmap(mddev))
 338                 return -EINVAL;
 339         blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
 340         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
 341
 342         /* if private is not null, we are here after takeover */
 343         if (mddev->private == NULL) {
 344                 ret = create_strip_zones(mddev, &conf);
 345                 if (ret < 0)
 346                         return ret;
 347                 mddev->private = conf;
 348         }
 349         conf = mddev->private;
 350         if (conf->scale_raid_disks) {
 351                 int i;
 352                 for (i=0; i < conf->strip_zone[0].nb_dev; i++)
 353                         conf->devlist[i]->raid_disk /= conf->scale_raid_disks;
 354                 /* FIXME update sysfs rd links */
 355         }
 356
 357         /* calculate array device size */
 358         md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 359
 360         printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
 361                 (unsigned long long)mddev->array_sectors);
 362         /* calculate the max read-ahead size.
 363          * For read-ahead of large files to be effective, we need to
 364          * readahead at least twice a whole stripe. i.e. number of devices
 365          * multiplied by chunk size times 2.
 366          * If an individual device has an ra_pages greater than the
 367          * chunk size, then we will not drive that device as hard as it
 368          * wants.  We consider this a configuration error: a larger
 369          * chunksize should be used in that case.
 370          */
 371         {
 372                 int stripe = mddev->raid_disks *
 373                         (mddev->chunk_sectors << 9) / PAGE_SIZE;
 374                 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 375                         mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 376         }
 377
 378         blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 379         dump_zones(mddev);
 380         md_integrity_register(mddev);
 381         return 0;
 382 }
 383
 384 static int raid0_stop(mddev_t *mddev)
 385 {
 386         raid0_conf_t *conf = mddev->private;
 387
 388         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 389         kfree(conf->strip_zone);
 390         kfree(conf->devlist);
 391         kfree(conf);
 392         mddev->private = NULL;
 393         return 0;
 394 }
 395
 396 /* Find the zone which holds a particular offset
 397  * Update *sectorp to be an offset in that zone
 398  */
 399 static struct strip_zone *find_zone(struct raid0_private_data *conf,
 400                                     sector_t *sectorp)
 401 {
 402         int i;
 403         struct strip_zone *z = conf->strip_zone;
 404         sector_t sector = *sectorp;
 405
 406         for (i = 0; i < conf->nr_strip_zones; i++)
 407                 if (sector < z[i].zone_end) {
 408                         if (i)
 409                                 *sectorp = sector - z[i-1].zone_end;
 410                         return z + i;
 411                 }
 412         BUG();
 413 }
 414
 415 /*
 416  * remaps the bio to the target device. we separate two flows.
 417  * power 2 flow and a general flow for the sake of perfromance
 418 */
 419 static mdk_rdev_t *map_sector(mddev_t *mddev, struct strip_zone *zone,
 420                                 sector_t sector, sector_t *sector_offset)
 421 {
 422         unsigned int sect_in_chunk;
 423         sector_t chunk;
 424         raid0_conf_t *conf = mddev->private;
 425         int raid_disks = conf->strip_zone[0].nb_dev;
 426         unsigned int chunk_sects = mddev->chunk_sectors;
 427
 428         if (is_power_of_2(chunk_sects)) {
 429                 int chunksect_bits = ffz(~chunk_sects);
 430                 /* find the sector offset inside the chunk */
 431                 sect_in_chunk  = sector & (chunk_sects - 1);
 432                 sector >>= chunksect_bits;
 433                 /* chunk in zone */
 434                 chunk = *sector_offset;
 435                 /* quotient is the chunk in real device*/
 436                 sector_div(chunk, zone->nb_dev << chunksect_bits);
 437         } else{
 438                 sect_in_chunk = sector_div(sector, chunk_sects);
 439                 chunk = *sector_offset;
 440                 sector_div(chunk, chunk_sects * zone->nb_dev);
 441         }
 442         /*
 443         *  position the bio over the real device
 444         *  real sector = chunk in device + starting of zone
 445         *       + the position in the chunk
 446         */
 447         *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
 448         return conf->devlist[(zone - conf->strip_zone)*raid_disks
 449                              + sector_div(sector, zone->nb_dev)];
 450 }
 451
 452 /*
 453  * Is io distribute over 1 or more chunks ?
 454 */
 455 static inline int is_io_in_chunk_boundary(mddev_t *mddev,
 456                         unsigned int chunk_sects, struct bio *bio)
 457 {
 458         if (likely(is_power_of_2(chunk_sects))) {
 459                 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
 460                                         + (bio->bi_size >> 9));
 461         } else{
 462                 sector_t sector = bio->bi_sector;
 463                 return chunk_sects >= (sector_div(sector, chunk_sects)
 464                                                 + (bio->bi_size >> 9));
 465         }
 466 }
 467
 468 static int raid0_make_request(struct request_queue *q, struct bio *bio)
 469 {
 470         mddev_t *mddev = q->queuedata;
 471         unsigned int chunk_sects;
 472         sector_t sector_offset;
 473         struct strip_zone *zone;
 474         mdk_rdev_t *tmp_dev;
 475         const int rw = bio_data_dir(bio);
 476         int cpu;
 477
 478         if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
 479                 md_barrier_request(mddev, bio);
 480                 return 0;
 481         }
 482
 483         cpu = part_stat_lock();
 484         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 485         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
 486                       bio_sectors(bio));
 487         part_stat_unlock();
 488
 489         chunk_sects = mddev->chunk_sectors;
 490         if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
 491                 sector_t sector = bio->bi_sector;
 492                 struct bio_pair *bp;
 493                 /* Sanity check -- queue functions should prevent this happening */
 494                 if (bio->bi_vcnt != 1 ||
 495                     bio->bi_idx != 0)
 496                         goto bad_map;
 497                 /* This is a one page bio that upper layers
 498                  * refuse to split for us, so we need to split it.
 499                  */
 500                 if (likely(is_power_of_2(chunk_sects)))
 501                         bp = bio_split(bio, chunk_sects - (sector &
 502                                                            (chunk_sects-1)));
 503                 else
 504                         bp = bio_split(bio, chunk_sects -
 505                                        sector_div(sector, chunk_sects));
 506                 if (raid0_make_request(q, &bp->bio1))
 507                         generic_make_request(&bp->bio1);
 508                 if (raid0_make_request(q, &bp->bio2))
 509                         generic_make_request(&bp->bio2);
 510
 511                 bio_pair_release(bp);
 512                 return 0;
 513         }
 514
 515         sector_offset = bio->bi_sector;
 516         zone =  find_zone(mddev->private, &sector_offset);
 517         tmp_dev = map_sector(mddev, zone, bio->bi_sector,
 518                              &sector_offset);
 519         bio->bi_bdev = tmp_dev->bdev;
 520         bio->bi_sector = sector_offset + zone->dev_start +
 521                 tmp_dev->data_offset;
 522         /*
 523          * Let the main block layer submit the IO and resolve recursion:
 524          */
 525         return 1;
 526
 527 bad_map:
 528         printk("raid0_make_request bug: can't convert block across chunks"
 529                 " or bigger than %dk %llu %d\n", chunk_sects / 2,
 530                 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 531
 532         bio_io_error(bio);
 533         return 0;
 534 }
 535
 536 static void raid0_status(struct seq_file *seq, mddev_t *mddev)
 537 {
 538 #undef MD_DEBUG
 539 #ifdef MD_DEBUG
 540         int j, k, h;
 541         char b[BDEVNAME_SIZE];
 542         raid0_conf_t *conf = mddev->private;
 543         int raid_disks = conf->strip_zone[0].nb_dev;
 544
 545         sector_t zone_size;
 546         sector_t zone_start = 0;
 547         h = 0;
 548
 549         for (j = 0; j < conf->nr_strip_zones; j++) {
 550                 seq_printf(seq, "      z%d", j);
 551                 seq_printf(seq, "=[");
 552                 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
 553                         seq_printf(seq, "%s/", bdevname(
 554                                 conf->devlist[j*raid_disks + k]
 555                                                 ->bdev, b));
 556
 557                 zone_size  = conf->strip_zone[j].zone_end - zone_start;
 558                 seq_printf(seq, "] ze=%lld ds=%lld s=%lld\n",
 559                         (unsigned long long)zone_start>>1,
 560                         (unsigned long long)conf->strip_zone[j].dev_start>>1,
 561                         (unsigned long long)zone_size>>1);
 562                 zone_start = conf->strip_zone[j].zone_end;
 563         }
 564 #endif
 565         seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
 566         return;
 567 }
 568
 569 static void *raid0_takeover_raid5(mddev_t *mddev)
 570 {
 571         mdk_rdev_t *rdev;
 572         raid0_conf_t *priv_conf;
 573
 574         if (mddev->degraded != 1) {
 575                 printk(KERN_ERR "md: raid5 must be degraded! Degraded disks: %d\n",
 576                        mddev->degraded);
 577                 return ERR_PTR(-EINVAL);
 578         }
 579
 580         list_for_each_entry(rdev, &mddev->disks, same_set) {
 581                 /* check slot number for a disk */
 582                 if (rdev->raid_disk == mddev->raid_disks-1) {
 583                         printk(KERN_ERR "md: raid5 must have missing parity disk!\n");
 584                         return ERR_PTR(-EINVAL);
 585                 }
 586         }
 587
 588         /* Set new parameters */
 589         mddev->new_level = 0;
 590         mddev->new_chunk_sectors = mddev->chunk_sectors;
 591         mddev->raid_disks--;
 592         mddev->delta_disks = -1;
 593         /* make sure it will be not marked as dirty */
 594         mddev->recovery_cp = MaxSector;
 595
 596         create_strip_zones(mddev, &priv_conf);
 597         return priv_conf;
 598 }
 599
 600 static void *raid0_takeover_raid10(mddev_t *mddev)
 601 {
 602         raid0_conf_t *priv_conf;
 603
 604         /* Check layout:
 605          *  - far_copies must be 1
 606          *  - near_copies must be 2
 607          *  - disks number must be even
 608          *  - all mirrors must be already degraded
 609          */
 610         if (mddev->layout != ((1 << 8) + 2)) {
 611                 printk(KERN_ERR "md: Raid0 cannot takover layout: %x\n",
 612                        mddev->layout);
 613                 return ERR_PTR(-EINVAL);
 614         }
 615         if (mddev->raid_disks & 1) {
 616                 printk(KERN_ERR "md: Raid0 cannot takover Raid10 with odd disk number.\n");
 617                 return ERR_PTR(-EINVAL);
 618         }
 619         if (mddev->degraded != (mddev->raid_disks>>1)) {
 620                 printk(KERN_ERR "md: All mirrors must be already degraded!\n");
 621                 return ERR_PTR(-EINVAL);
 622         }
 623
 624         /* Set new parameters */
 625         mddev->new_level = 0;
 626         mddev->new_chunk_sectors = mddev->chunk_sectors;
 627         mddev->delta_disks = - mddev->raid_disks / 2;
 628         mddev->raid_disks += mddev->delta_disks;
 629         mddev->degraded = 0;
 630         /* make sure it will be not marked as dirty */
 631         mddev->recovery_cp = MaxSector;
 632
 633         create_strip_zones(mddev, &priv_conf);
 634         priv_conf->scale_raid_disks = 2;
 635         return priv_conf;
 636 }
 637
 638 static void *raid0_takeover(mddev_t *mddev)
 639 {
 640         /* raid0 can take over:
 641          *  raid5 - providing it is Raid4 layout and one disk is faulty
 642          *  raid10 - assuming we have all necessary active disks
 643          */
 644         if (mddev->level == 5) {
 645                 if (mddev->layout == ALGORITHM_PARITY_N)
 646                         return raid0_takeover_raid5(mddev);
 647
 648                 printk(KERN_ERR "md: Raid can only takeover Raid5 with layout: %d\n",
 649                        ALGORITHM_PARITY_N);
 650         }
 651
 652         if (mddev->level == 10)
 653                 return raid0_takeover_raid10(mddev);
 654
 655         return ERR_PTR(-EINVAL);
 656 }
 657
 658 static void raid0_quiesce(mddev_t *mddev, int state)
 659 {
 660 }
 661
 662 static struct mdk_personality raid0_personality=
 663 {
 664         .name           = "raid0",
 665         .level          = 0,
 666         .owner          = THIS_MODULE,
 667         .make_request   = raid0_make_request,
 668         .run            = raid0_run,
 669         .stop           = raid0_stop,
 670         .status         = raid0_status,
 671         .size           = raid0_size,
 672         .takeover       = raid0_takeover,
 673         .quiesce        = raid0_quiesce,
 674 };
 675
 676 static int __init raid0_init (void)
 677 {
 678         return register_md_personality (&raid0_personality);
 679 }
 680
 681 static void raid0_exit (void)
 682 {
 683         unregister_md_personality (&raid0_personality);
 684 }
 685
 686 module_init(raid0_init);
 687 module_exit(raid0_exit);
 688 MODULE_LICENSE("GPL");
 689 MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
 690 MODULE_ALIAS("md-personality-2"); /* RAID0 */
 691 MODULE_ALIAS("md-raid0");
 692 MODULE_ALIAS("md-level-0");