linux/drivers/md/raid0.c
<<
>>
Prefs
   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
  29static 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*/
  51static 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
  81static 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        bool discard_supported = false;
  92
  93        if (!conf)
  94                return -ENOMEM;
  95        rdev_for_each(rdev1, mddev) {
  96                pr_debug("md/raid0:%s: looking at %s\n",
  97                         mdname(mddev),
  98                         bdevname(rdev1->bdev, b));
  99                c = 0;
 100
 101                /* round size to chunk_size */
 102                sectors = rdev1->sectors;
 103                sector_div(sectors, mddev->chunk_sectors);
 104                rdev1->sectors = sectors * mddev->chunk_sectors;
 105
 106                rdev_for_each(rdev2, mddev) {
 107                        pr_debug("md/raid0:%s:   comparing %s(%llu)"
 108                                 " with %s(%llu)\n",
 109                                 mdname(mddev),
 110                                 bdevname(rdev1->bdev,b),
 111                                 (unsigned long long)rdev1->sectors,
 112                                 bdevname(rdev2->bdev,b2),
 113                                 (unsigned long long)rdev2->sectors);
 114                        if (rdev2 == rdev1) {
 115                                pr_debug("md/raid0:%s:   END\n",
 116                                         mdname(mddev));
 117                                break;
 118                        }
 119                        if (rdev2->sectors == rdev1->sectors) {
 120                                /*
 121                                 * Not unique, don't count it as a new
 122                                 * group
 123                                 */
 124                                pr_debug("md/raid0:%s:   EQUAL\n",
 125                                         mdname(mddev));
 126                                c = 1;
 127                                break;
 128                        }
 129                        pr_debug("md/raid0:%s:   NOT EQUAL\n",
 130                                 mdname(mddev));
 131                }
 132                if (!c) {
 133                        pr_debug("md/raid0:%s:   ==> UNIQUE\n",
 134                                 mdname(mddev));
 135                        conf->nr_strip_zones++;
 136                        pr_debug("md/raid0:%s: %d zones\n",
 137                                 mdname(mddev), conf->nr_strip_zones);
 138                }
 139        }
 140        pr_debug("md/raid0:%s: FINAL %d zones\n",
 141                 mdname(mddev), conf->nr_strip_zones);
 142        err = -ENOMEM;
 143        conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
 144                                conf->nr_strip_zones, GFP_KERNEL);
 145        if (!conf->strip_zone)
 146                goto abort;
 147        conf->devlist = kzalloc(sizeof(struct md_rdev*)*
 148                                conf->nr_strip_zones*mddev->raid_disks,
 149                                GFP_KERNEL);
 150        if (!conf->devlist)
 151                goto abort;
 152
 153        /* The first zone must contain all devices, so here we check that
 154         * there is a proper alignment of slots to devices and find them all
 155         */
 156        zone = &conf->strip_zone[0];
 157        cnt = 0;
 158        smallest = NULL;
 159        dev = conf->devlist;
 160        err = -EINVAL;
 161        rdev_for_each(rdev1, mddev) {
 162                int j = rdev1->raid_disk;
 163
 164                if (mddev->level == 10) {
 165                        /* taking over a raid10-n2 array */
 166                        j /= 2;
 167                        rdev1->new_raid_disk = j;
 168                }
 169
 170                if (mddev->level == 1) {
 171                        /* taiking over a raid1 array-
 172                         * we have only one active disk
 173                         */
 174                        j = 0;
 175                        rdev1->new_raid_disk = j;
 176                }
 177
 178                if (j < 0 || j >= mddev->raid_disks) {
 179                        printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
 180                               "aborting!\n", mdname(mddev), j);
 181                        goto abort;
 182                }
 183                if (dev[j]) {
 184                        printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
 185                               "aborting!\n", mdname(mddev), j);
 186                        goto abort;
 187                }
 188                dev[j] = rdev1;
 189
 190                disk_stack_limits(mddev->gendisk, rdev1->bdev,
 191                                  rdev1->data_offset << 9);
 192
 193                if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
 194                        conf->has_merge_bvec = 1;
 195
 196                if (!smallest || (rdev1->sectors < smallest->sectors))
 197                        smallest = rdev1;
 198                cnt++;
 199
 200                if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
 201                        discard_supported = true;
 202        }
 203        if (cnt != mddev->raid_disks) {
 204                printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
 205                       "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
 206                goto abort;
 207        }
 208        zone->nb_dev = cnt;
 209        zone->zone_end = smallest->sectors * cnt;
 210
 211        curr_zone_end = zone->zone_end;
 212
 213        /* now do the other zones */
 214        for (i = 1; i < conf->nr_strip_zones; i++)
 215        {
 216                int j;
 217
 218                zone = conf->strip_zone + i;
 219                dev = conf->devlist + i * mddev->raid_disks;
 220
 221                pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
 222                zone->dev_start = smallest->sectors;
 223                smallest = NULL;
 224                c = 0;
 225
 226                for (j=0; j<cnt; j++) {
 227                        rdev = conf->devlist[j];
 228                        if (rdev->sectors <= zone->dev_start) {
 229                                pr_debug("md/raid0:%s: checking %s ... nope\n",
 230                                         mdname(mddev),
 231                                         bdevname(rdev->bdev, b));
 232                                continue;
 233                        }
 234                        pr_debug("md/raid0:%s: checking %s ..."
 235                                 " contained as device %d\n",
 236                                 mdname(mddev),
 237                                 bdevname(rdev->bdev, b), c);
 238                        dev[c] = rdev;
 239                        c++;
 240                        if (!smallest || rdev->sectors < smallest->sectors) {
 241                                smallest = rdev;
 242                                pr_debug("md/raid0:%s:  (%llu) is smallest!.\n",
 243                                         mdname(mddev),
 244                                         (unsigned long long)rdev->sectors);
 245                        }
 246                }
 247
 248                zone->nb_dev = c;
 249                sectors = (smallest->sectors - zone->dev_start) * c;
 250                pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
 251                         mdname(mddev),
 252                         zone->nb_dev, (unsigned long long)sectors);
 253
 254                curr_zone_end += sectors;
 255                zone->zone_end = curr_zone_end;
 256
 257                pr_debug("md/raid0:%s: current zone start: %llu\n",
 258                         mdname(mddev),
 259                         (unsigned long long)smallest->sectors);
 260        }
 261        mddev->queue->backing_dev_info.congested_fn = raid0_congested;
 262        mddev->queue->backing_dev_info.congested_data = mddev;
 263
 264        /*
 265         * now since we have the hard sector sizes, we can make sure
 266         * chunk size is a multiple of that sector size
 267         */
 268        if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
 269                printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
 270                       mdname(mddev),
 271                       mddev->chunk_sectors << 9);
 272                goto abort;
 273        }
 274
 275        blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
 276        blk_queue_io_opt(mddev->queue,
 277                         (mddev->chunk_sectors << 9) * mddev->raid_disks);
 278
 279        if (!discard_supported)
 280                queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
 281        else
 282                queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
 283
 284        pr_debug("md/raid0:%s: done.\n", mdname(mddev));
 285        *private_conf = conf;
 286
 287        return 0;
 288abort:
 289        kfree(conf->strip_zone);
 290        kfree(conf->devlist);
 291        kfree(conf);
 292        *private_conf = ERR_PTR(err);
 293        return err;
 294}
 295
 296/* Find the zone which holds a particular offset
 297 * Update *sectorp to be an offset in that zone
 298 */
 299static struct strip_zone *find_zone(struct r0conf *conf,
 300                                    sector_t *sectorp)
 301{
 302        int i;
 303        struct strip_zone *z = conf->strip_zone;
 304        sector_t sector = *sectorp;
 305
 306        for (i = 0; i < conf->nr_strip_zones; i++)
 307                if (sector < z[i].zone_end) {
 308                        if (i)
 309                                *sectorp = sector - z[i-1].zone_end;
 310                        return z + i;
 311                }
 312        BUG();
 313}
 314
 315/*
 316 * remaps the bio to the target device. we separate two flows.
 317 * power 2 flow and a general flow for the sake of perfromance
 318*/
 319static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
 320                                sector_t sector, sector_t *sector_offset)
 321{
 322        unsigned int sect_in_chunk;
 323        sector_t chunk;
 324        struct r0conf *conf = mddev->private;
 325        int raid_disks = conf->strip_zone[0].nb_dev;
 326        unsigned int chunk_sects = mddev->chunk_sectors;
 327
 328        if (is_power_of_2(chunk_sects)) {
 329                int chunksect_bits = ffz(~chunk_sects);
 330                /* find the sector offset inside the chunk */
 331                sect_in_chunk  = sector & (chunk_sects - 1);
 332                sector >>= chunksect_bits;
 333                /* chunk in zone */
 334                chunk = *sector_offset;
 335                /* quotient is the chunk in real device*/
 336                sector_div(chunk, zone->nb_dev << chunksect_bits);
 337        } else{
 338                sect_in_chunk = sector_div(sector, chunk_sects);
 339                chunk = *sector_offset;
 340                sector_div(chunk, chunk_sects * zone->nb_dev);
 341        }
 342        /*
 343        *  position the bio over the real device
 344        *  real sector = chunk in device + starting of zone
 345        *       + the position in the chunk
 346        */
 347        *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
 348        return conf->devlist[(zone - conf->strip_zone)*raid_disks
 349                             + sector_div(sector, zone->nb_dev)];
 350}
 351
 352/**
 353 *      raid0_mergeable_bvec -- tell bio layer if two requests can be merged
 354 *      @q: request queue
 355 *      @bvm: properties of new bio
 356 *      @biovec: the request that could be merged to it.
 357 *
 358 *      Return amount of bytes we can accept at this offset
 359 */
 360static int raid0_mergeable_bvec(struct request_queue *q,
 361                                struct bvec_merge_data *bvm,
 362                                struct bio_vec *biovec)
 363{
 364        struct mddev *mddev = q->queuedata;
 365        struct r0conf *conf = mddev->private;
 366        sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
 367        sector_t sector_offset = sector;
 368        int max;
 369        unsigned int chunk_sectors = mddev->chunk_sectors;
 370        unsigned int bio_sectors = bvm->bi_size >> 9;
 371        struct strip_zone *zone;
 372        struct md_rdev *rdev;
 373        struct request_queue *subq;
 374
 375        if (is_power_of_2(chunk_sectors))
 376                max =  (chunk_sectors - ((sector & (chunk_sectors-1))
 377                                                + bio_sectors)) << 9;
 378        else
 379                max =  (chunk_sectors - (sector_div(sector, chunk_sectors)
 380                                                + bio_sectors)) << 9;
 381        if (max < 0)
 382                max = 0; /* bio_add cannot handle a negative return */
 383        if (max <= biovec->bv_len && bio_sectors == 0)
 384                return biovec->bv_len;
 385        if (max < biovec->bv_len)
 386                /* too small already, no need to check further */
 387                return max;
 388        if (!conf->has_merge_bvec)
 389                return max;
 390
 391        /* May need to check subordinate device */
 392        sector = sector_offset;
 393        zone = find_zone(mddev->private, &sector_offset);
 394        rdev = map_sector(mddev, zone, sector, &sector_offset);
 395        subq = bdev_get_queue(rdev->bdev);
 396        if (subq->merge_bvec_fn) {
 397                bvm->bi_bdev = rdev->bdev;
 398                bvm->bi_sector = sector_offset + zone->dev_start +
 399                        rdev->data_offset;
 400                return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
 401        } else
 402                return max;
 403}
 404
 405static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
 406{
 407        sector_t array_sectors = 0;
 408        struct md_rdev *rdev;
 409
 410        WARN_ONCE(sectors || raid_disks,
 411                  "%s does not support generic reshape\n", __func__);
 412
 413        rdev_for_each(rdev, mddev)
 414                array_sectors += (rdev->sectors &
 415                                  ~(sector_t)(mddev->chunk_sectors-1));
 416
 417        return array_sectors;
 418}
 419
 420static int raid0_stop(struct mddev *mddev);
 421
 422static int raid0_run(struct mddev *mddev)
 423{
 424        struct r0conf *conf;
 425        int ret;
 426
 427        if (mddev->chunk_sectors == 0) {
 428                printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
 429                       mdname(mddev));
 430                return -EINVAL;
 431        }
 432        if (md_check_no_bitmap(mddev))
 433                return -EINVAL;
 434        blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
 435        blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
 436        blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
 437
 438        /* if private is not null, we are here after takeover */
 439        if (mddev->private == NULL) {
 440                ret = create_strip_zones(mddev, &conf);
 441                if (ret < 0)
 442                        return ret;
 443                mddev->private = conf;
 444        }
 445        conf = mddev->private;
 446
 447        /* calculate array device size */
 448        md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
 449
 450        printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
 451               mdname(mddev),
 452               (unsigned long long)mddev->array_sectors);
 453        /* calculate the max read-ahead size.
 454         * For read-ahead of large files to be effective, we need to
 455         * readahead at least twice a whole stripe. i.e. number of devices
 456         * multiplied by chunk size times 2.
 457         * If an individual device has an ra_pages greater than the
 458         * chunk size, then we will not drive that device as hard as it
 459         * wants.  We consider this a configuration error: a larger
 460         * chunksize should be used in that case.
 461         */
 462        {
 463                int stripe = mddev->raid_disks *
 464                        (mddev->chunk_sectors << 9) / PAGE_SIZE;
 465                if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
 466                        mddev->queue->backing_dev_info.ra_pages = 2* stripe;
 467        }
 468
 469        blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
 470        dump_zones(mddev);
 471
 472        ret = md_integrity_register(mddev);
 473        if (ret)
 474                raid0_stop(mddev);
 475
 476        return ret;
 477}
 478
 479static int raid0_stop(struct mddev *mddev)
 480{
 481        struct r0conf *conf = mddev->private;
 482
 483        blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
 484        kfree(conf->strip_zone);
 485        kfree(conf->devlist);
 486        kfree(conf);
 487        mddev->private = NULL;
 488        return 0;
 489}
 490
 491/*
 492 * Is io distribute over 1 or more chunks ?
 493*/
 494static inline int is_io_in_chunk_boundary(struct mddev *mddev,
 495                        unsigned int chunk_sects, struct bio *bio)
 496{
 497        if (likely(is_power_of_2(chunk_sects))) {
 498                return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
 499                                        + (bio->bi_size >> 9));
 500        } else{
 501                sector_t sector = bio->bi_sector;
 502                return chunk_sects >= (sector_div(sector, chunk_sects)
 503                                                + (bio->bi_size >> 9));
 504        }
 505}
 506
 507static void raid0_make_request(struct mddev *mddev, struct bio *bio)
 508{
 509        unsigned int chunk_sects;
 510        sector_t sector_offset;
 511        struct strip_zone *zone;
 512        struct md_rdev *tmp_dev;
 513
 514        if (unlikely(bio->bi_rw & REQ_FLUSH)) {
 515                md_flush_request(mddev, bio);
 516                return;
 517        }
 518
 519        chunk_sects = mddev->chunk_sectors;
 520        if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
 521                sector_t sector = bio->bi_sector;
 522                struct bio_pair *bp;
 523                /* Sanity check -- queue functions should prevent this happening */
 524                if ((bio->bi_vcnt != 1 && bio->bi_vcnt != 0) ||
 525                    bio->bi_idx != 0)
 526                        goto bad_map;
 527                /* This is a one page bio that upper layers
 528                 * refuse to split for us, so we need to split it.
 529                 */
 530                if (likely(is_power_of_2(chunk_sects)))
 531                        bp = bio_split(bio, chunk_sects - (sector &
 532                                                           (chunk_sects-1)));
 533                else
 534                        bp = bio_split(bio, chunk_sects -
 535                                       sector_div(sector, chunk_sects));
 536                raid0_make_request(mddev, &bp->bio1);
 537                raid0_make_request(mddev, &bp->bio2);
 538                bio_pair_release(bp);
 539                return;
 540        }
 541
 542        sector_offset = bio->bi_sector;
 543        zone = find_zone(mddev->private, &sector_offset);
 544        tmp_dev = map_sector(mddev, zone, bio->bi_sector,
 545                             &sector_offset);
 546        bio->bi_bdev = tmp_dev->bdev;
 547        bio->bi_sector = sector_offset + zone->dev_start +
 548                tmp_dev->data_offset;
 549
 550        if (unlikely((bio->bi_rw & REQ_DISCARD) &&
 551                     !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) {
 552                /* Just ignore it */
 553                bio_endio(bio, 0);
 554                return;
 555        }
 556
 557        generic_make_request(bio);
 558        return;
 559
 560bad_map:
 561        printk("md/raid0:%s: make_request bug: can't convert block across chunks"
 562               " or bigger than %dk %llu %d\n",
 563               mdname(mddev), chunk_sects / 2,
 564               (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
 565
 566        bio_io_error(bio);
 567        return;
 568}
 569
 570static void raid0_status(struct seq_file *seq, struct mddev *mddev)
 571{
 572        seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
 573        return;
 574}
 575
 576static void *raid0_takeover_raid45(struct mddev *mddev)
 577{
 578        struct md_rdev *rdev;
 579        struct r0conf *priv_conf;
 580
 581        if (mddev->degraded != 1) {
 582                printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
 583                       mdname(mddev),
 584                       mddev->degraded);
 585                return ERR_PTR(-EINVAL);
 586        }
 587
 588        rdev_for_each(rdev, mddev) {
 589                /* check slot number for a disk */
 590                if (rdev->raid_disk == mddev->raid_disks-1) {
 591                        printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
 592                               mdname(mddev));
 593                        return ERR_PTR(-EINVAL);
 594                }
 595        }
 596
 597        /* Set new parameters */
 598        mddev->new_level = 0;
 599        mddev->new_layout = 0;
 600        mddev->new_chunk_sectors = mddev->chunk_sectors;
 601        mddev->raid_disks--;
 602        mddev->delta_disks = -1;
 603        /* make sure it will be not marked as dirty */
 604        mddev->recovery_cp = MaxSector;
 605
 606        create_strip_zones(mddev, &priv_conf);
 607        return priv_conf;
 608}
 609
 610static void *raid0_takeover_raid10(struct mddev *mddev)
 611{
 612        struct r0conf *priv_conf;
 613
 614        /* Check layout:
 615         *  - far_copies must be 1
 616         *  - near_copies must be 2
 617         *  - disks number must be even
 618         *  - all mirrors must be already degraded
 619         */
 620        if (mddev->layout != ((1 << 8) + 2)) {
 621                printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
 622                       mdname(mddev),
 623                       mddev->layout);
 624                return ERR_PTR(-EINVAL);
 625        }
 626        if (mddev->raid_disks & 1) {
 627                printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
 628                       mdname(mddev));
 629                return ERR_PTR(-EINVAL);
 630        }
 631        if (mddev->degraded != (mddev->raid_disks>>1)) {
 632                printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
 633                       mdname(mddev));
 634                return ERR_PTR(-EINVAL);
 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->delta_disks = - mddev->raid_disks / 2;
 642        mddev->raid_disks += mddev->delta_disks;
 643        mddev->degraded = 0;
 644        /* make sure it will be not marked as dirty */
 645        mddev->recovery_cp = MaxSector;
 646
 647        create_strip_zones(mddev, &priv_conf);
 648        return priv_conf;
 649}
 650
 651static void *raid0_takeover_raid1(struct mddev *mddev)
 652{
 653        struct r0conf *priv_conf;
 654        int chunksect;
 655
 656        /* Check layout:
 657         *  - (N - 1) mirror drives must be already faulty
 658         */
 659        if ((mddev->raid_disks - 1) != mddev->degraded) {
 660                printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
 661                       mdname(mddev));
 662                return ERR_PTR(-EINVAL);
 663        }
 664
 665        /*
 666         * a raid1 doesn't have the notion of chunk size, so
 667         * figure out the largest suitable size we can use.
 668         */
 669        chunksect = 64 * 2; /* 64K by default */
 670
 671        /* The array must be an exact multiple of chunksize */
 672        while (chunksect && (mddev->array_sectors & (chunksect - 1)))
 673                chunksect >>= 1;
 674
 675        if ((chunksect << 9) < PAGE_SIZE)
 676                /* array size does not allow a suitable chunk size */
 677                return ERR_PTR(-EINVAL);
 678
 679        /* Set new parameters */
 680        mddev->new_level = 0;
 681        mddev->new_layout = 0;
 682        mddev->new_chunk_sectors = chunksect;
 683        mddev->chunk_sectors = chunksect;
 684        mddev->delta_disks = 1 - mddev->raid_disks;
 685        mddev->raid_disks = 1;
 686        /* make sure it will be not marked as dirty */
 687        mddev->recovery_cp = MaxSector;
 688
 689        create_strip_zones(mddev, &priv_conf);
 690        return priv_conf;
 691}
 692
 693static void *raid0_takeover(struct mddev *mddev)
 694{
 695        /* raid0 can take over:
 696         *  raid4 - if all data disks are active.
 697         *  raid5 - providing it is Raid4 layout and one disk is faulty
 698         *  raid10 - assuming we have all necessary active disks
 699         *  raid1 - with (N -1) mirror drives faulty
 700         */
 701        if (mddev->level == 4)
 702                return raid0_takeover_raid45(mddev);
 703
 704        if (mddev->level == 5) {
 705                if (mddev->layout == ALGORITHM_PARITY_N)
 706                        return raid0_takeover_raid45(mddev);
 707
 708                printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
 709                       mdname(mddev), ALGORITHM_PARITY_N);
 710        }
 711
 712        if (mddev->level == 10)
 713                return raid0_takeover_raid10(mddev);
 714
 715        if (mddev->level == 1)
 716                return raid0_takeover_raid1(mddev);
 717
 718        printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
 719                mddev->level);
 720
 721        return ERR_PTR(-EINVAL);
 722}
 723
 724static void raid0_quiesce(struct mddev *mddev, int state)
 725{
 726}
 727
 728static struct md_personality raid0_personality=
 729{
 730        .name           = "raid0",
 731        .level          = 0,
 732        .owner          = THIS_MODULE,
 733        .make_request   = raid0_make_request,
 734        .run            = raid0_run,
 735        .stop           = raid0_stop,
 736        .status         = raid0_status,
 737        .size           = raid0_size,
 738        .takeover       = raid0_takeover,
 739        .quiesce        = raid0_quiesce,
 740};
 741
 742static int __init raid0_init (void)
 743{
 744        return register_md_personality (&raid0_personality);
 745}
 746
 747static void raid0_exit (void)
 748{
 749        unregister_md_personality (&raid0_personality);
 750}
 751
 752module_init(raid0_init);
 753module_exit(raid0_exit);
 754MODULE_LICENSE("GPL");
 755MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
 756MODULE_ALIAS("md-personality-2"); /* RAID0 */
 757MODULE_ALIAS("md-raid0");
 758MODULE_ALIAS("md-level-0");
 759
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.