1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 raid0.c : Multiple Devices driver for Linux 4 Copyright (C) 1994-96 Marc ZYNGIER 5 <zyngier@ufr-info-p7.ibp.fr> or 6 <maz@gloups.fdn.fr> 7 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat 8 9 RAID-0 management functions. 10 11 */ 12 13 #include <linux/blkdev.h> 14 #include <linux/seq_file.h> 15 #include <linux/module.h> 16 #include <linux/slab.h> 17 #include <trace/events/block.h> 18 #include "md.h" 19 #include "raid0.h" 20 #include "raid5.h" 21 22 static int default_layout = 0; 23 module_param(default_layout, int, 0644); 24 25 #define UNSUPPORTED_MDDEV_FLAGS \ 26 ((1L << MD_HAS_JOURNAL) | \ 27 (1L << MD_JOURNAL_CLEAN) | \ 28 (1L << MD_FAILFAST_SUPPORTED) |\ 29 (1L << MD_HAS_PPL) | \ 30 (1L << MD_HAS_MULTIPLE_PPLS)) 31 32 /* 33 * inform the user of the raid configuration 34 */ 35 static void dump_zones(struct mddev *mddev) 36 { 37 int j, k; 38 sector_t zone_size = 0; 39 sector_t zone_start = 0; 40 struct r0conf *conf = mddev->private; 41 int raid_disks = conf->strip_zone[0].nb_dev; 42 pr_debug("md: RAID0 configuration for %s - %d zone%s\n", 43 mdname(mddev), 44 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s"); 45 for (j = 0; j < conf->nr_strip_zones; j++) { 46 char line[200]; 47 int len = 0; 48 49 for (k = 0; k < conf->strip_zone[j].nb_dev; k++) 50 len += scnprintf(line+len, 200-len, "%s%pg", k?"/":"", 51 conf->devlist[j * raid_disks + k]->bdev); 52 pr_debug("md: zone%d=[%s]\n", j, line); 53 54 zone_size = conf->strip_zone[j].zone_end - zone_start; 55 pr_debug(" zone-offset=%10lluKB, device-offset=%10lluKB, size=%10lluKB\n", 56 (unsigned long long)zone_start>>1, 57 (unsigned long long)conf->strip_zone[j].dev_start>>1, 58 (unsigned long long)zone_size>>1); 59 zone_start = conf->strip_zone[j].zone_end; 60 } 61 } 62 63 static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf) 64 { 65 int i, c, err; 66 sector_t curr_zone_end, sectors; 67 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev; 68 struct strip_zone *zone; 69 int cnt; 70 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL); 71 unsigned blksize = 512; 72 73 *private_conf = ERR_PTR(-ENOMEM); 74 if (!conf) 75 return -ENOMEM; 76 rdev_for_each(rdev1, mddev) { 77 pr_debug("md/raid0:%s: looking at %pg\n", 78 mdname(mddev), 79 rdev1->bdev); 80 c = 0; 81 82 /* round size to chunk_size */ 83 sectors = rdev1->sectors; 84 sector_div(sectors, mddev->chunk_sectors); 85 rdev1->sectors = sectors * mddev->chunk_sectors; 86 87 blksize = max(blksize, queue_logical_block_size( 88 rdev1->bdev->bd_disk->queue)); 89 90 rdev_for_each(rdev2, mddev) { 91 pr_debug("md/raid0:%s: comparing %pg(%llu)" 92 " with %pg(%llu)\n", 93 mdname(mddev), 94 rdev1->bdev, 95 (unsigned long long)rdev1->sectors, 96 rdev2->bdev, 97 (unsigned long long)rdev2->sectors); 98 if (rdev2 == rdev1) { 99 pr_debug("md/raid0:%s: END\n", 100 mdname(mddev)); 101 break; 102 } 103 if (rdev2->sectors == rdev1->sectors) { 104 /* 105 * Not unique, don't count it as a new 106 * group 107 */ 108 pr_debug("md/raid0:%s: EQUAL\n", 109 mdname(mddev)); 110 c = 1; 111 break; 112 } 113 pr_debug("md/raid0:%s: NOT EQUAL\n", 114 mdname(mddev)); 115 } 116 if (!c) { 117 pr_debug("md/raid0:%s: ==> UNIQUE\n", 118 mdname(mddev)); 119 conf->nr_strip_zones++; 120 pr_debug("md/raid0:%s: %d zones\n", 121 mdname(mddev), conf->nr_strip_zones); 122 } 123 } 124 pr_debug("md/raid0:%s: FINAL %d zones\n", 125 mdname(mddev), conf->nr_strip_zones); 126 127 /* 128 * now since we have the hard sector sizes, we can make sure 129 * chunk size is a multiple of that sector size 130 */ 131 if ((mddev->chunk_sectors << 9) % blksize) { 132 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n", 133 mdname(mddev), 134 mddev->chunk_sectors << 9, blksize); 135 err = -EINVAL; 136 goto abort; 137 } 138 139 err = -ENOMEM; 140 conf->strip_zone = kcalloc(conf->nr_strip_zones, 141 sizeof(struct strip_zone), 142 GFP_KERNEL); 143 if (!conf->strip_zone) 144 goto abort; 145 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *), 146 conf->nr_strip_zones, 147 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 rdev_for_each(rdev1, mddev) { 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) { 178 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n", 179 mdname(mddev)); 180 goto abort; 181 } 182 if (j >= mddev->raid_disks) { 183 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n", 184 mdname(mddev), j); 185 goto abort; 186 } 187 if (dev[j]) { 188 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n", 189 mdname(mddev), j); 190 goto abort; 191 } 192 dev[j] = rdev1; 193 194 if (!smallest || (rdev1->sectors < smallest->sectors)) 195 smallest = rdev1; 196 cnt++; 197 } 198 if (cnt != mddev->raid_disks) { 199 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n", 200 mdname(mddev), cnt, mddev->raid_disks); 201 goto abort; 202 } 203 zone->nb_dev = cnt; 204 zone->zone_end = smallest->sectors * cnt; 205 206 curr_zone_end = zone->zone_end; 207 208 /* now do the other zones */ 209 for (i = 1; i < conf->nr_strip_zones; i++) 210 { 211 int j; 212 213 zone = conf->strip_zone + i; 214 dev = conf->devlist + i * mddev->raid_disks; 215 216 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i); 217 zone->dev_start = smallest->sectors; 218 smallest = NULL; 219 c = 0; 220 221 for (j=0; j<cnt; j++) { 222 rdev = conf->devlist[j]; 223 if (rdev->sectors <= zone->dev_start) { 224 pr_debug("md/raid0:%s: checking %pg ... nope\n", 225 mdname(mddev), 226 rdev->bdev); 227 continue; 228 } 229 pr_debug("md/raid0:%s: checking %pg ..." 230 " contained as device %d\n", 231 mdname(mddev), 232 rdev->bdev, c); 233 dev[c] = rdev; 234 c++; 235 if (!smallest || rdev->sectors < smallest->sectors) { 236 smallest = rdev; 237 pr_debug("md/raid0:%s: (%llu) is smallest!.\n", 238 mdname(mddev), 239 (unsigned long long)rdev->sectors); 240 } 241 } 242 243 zone->nb_dev = c; 244 sectors = (smallest->sectors - zone->dev_start) * c; 245 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n", 246 mdname(mddev), 247 zone->nb_dev, (unsigned long long)sectors); 248 249 curr_zone_end += sectors; 250 zone->zone_end = curr_zone_end; 251 252 pr_debug("md/raid0:%s: current zone start: %llu\n", 253 mdname(mddev), 254 (unsigned long long)smallest->sectors); 255 } 256 257 if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) { 258 conf->layout = RAID0_ORIG_LAYOUT; 259 } else if (mddev->layout == RAID0_ORIG_LAYOUT || 260 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) { 261 conf->layout = mddev->layout; 262 } else if (default_layout == RAID0_ORIG_LAYOUT || 263 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) { 264 conf->layout = default_layout; 265 } else { 266 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n", 267 mdname(mddev)); 268 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n"); 269 err = -EOPNOTSUPP; 270 goto abort; 271 } 272 273 if (conf->layout == RAID0_ORIG_LAYOUT) { 274 for (i = 1; i < conf->nr_strip_zones; i++) { 275 sector_t first_sector = conf->strip_zone[i-1].zone_end; 276 277 sector_div(first_sector, mddev->chunk_sectors); 278 zone = conf->strip_zone + i; 279 /* disk_shift is first disk index used in the zone */ 280 zone->disk_shift = sector_div(first_sector, 281 zone->nb_dev); 282 } 283 } 284 285 pr_debug("md/raid0:%s: done.\n", mdname(mddev)); 286 *private_conf = conf; 287 288 return 0; 289 abort: 290 kfree(conf->strip_zone); 291 kfree(conf->devlist); 292 kfree(conf); 293 *private_conf = ERR_PTR(err); 294 return err; 295 } 296 297 /* Find the zone which holds a particular offset 298 * Update *sectorp to be an offset in that zone 299 */ 300 static struct strip_zone *find_zone(struct r0conf *conf, 301 sector_t *sectorp) 302 { 303 int i; 304 struct strip_zone *z = conf->strip_zone; 305 sector_t sector = *sectorp; 306 307 for (i = 0; i < conf->nr_strip_zones; i++) 308 if (sector < z[i].zone_end) { 309 if (i) 310 *sectorp = sector - z[i-1].zone_end; 311 return z + i; 312 } 313 BUG(); 314 } 315 316 /* 317 * remaps the bio to the target device. we separate two flows. 318 * power 2 flow and a general flow for the sake of performance 319 */ 320 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone, 321 sector_t sector, sector_t *sector_offset) 322 { 323 unsigned int sect_in_chunk; 324 sector_t chunk; 325 struct r0conf *conf = mddev->private; 326 int raid_disks = conf->strip_zone[0].nb_dev; 327 unsigned int chunk_sects = mddev->chunk_sectors; 328 329 if (is_power_of_2(chunk_sects)) { 330 int chunksect_bits = ffz(~chunk_sects); 331 /* find the sector offset inside the chunk */ 332 sect_in_chunk = sector & (chunk_sects - 1); 333 sector >>= chunksect_bits; 334 /* chunk in zone */ 335 chunk = *sector_offset; 336 /* quotient is the chunk in real device*/ 337 sector_div(chunk, zone->nb_dev << chunksect_bits); 338 } else{ 339 sect_in_chunk = sector_div(sector, chunk_sects); 340 chunk = *sector_offset; 341 sector_div(chunk, chunk_sects * zone->nb_dev); 342 } 343 /* 344 * position the bio over the real device 345 * real sector = chunk in device + starting of zone 346 * + the position in the chunk 347 */ 348 *sector_offset = (chunk * chunk_sects) + sect_in_chunk; 349 return conf->devlist[(zone - conf->strip_zone)*raid_disks 350 + sector_div(sector, zone->nb_dev)]; 351 } 352 353 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks) 354 { 355 sector_t array_sectors = 0; 356 struct md_rdev *rdev; 357 358 WARN_ONCE(sectors || raid_disks, 359 "%s does not support generic reshape\n", __func__); 360 361 rdev_for_each(rdev, mddev) 362 array_sectors += (rdev->sectors & 363 ~(sector_t)(mddev->chunk_sectors-1)); 364 365 return array_sectors; 366 } 367 368 static void raid0_free(struct mddev *mddev, void *priv) 369 { 370 struct r0conf *conf = priv; 371 372 kfree(conf->strip_zone); 373 kfree(conf->devlist); 374 kfree(conf); 375 } 376 377 static int raid0_set_limits(struct mddev *mddev) 378 { 379 struct queue_limits lim; 380 381 blk_set_stacking_limits(&lim); 382 lim.max_hw_sectors = mddev->chunk_sectors; 383 lim.max_write_zeroes_sectors = mddev->chunk_sectors; 384 lim.io_min = mddev->chunk_sectors << 9; 385 lim.io_opt = lim.io_min * mddev->raid_disks; 386 mddev_stack_rdev_limits(mddev, &lim); 387 return queue_limits_set(mddev->gendisk->queue, &lim); 388 } 389 390 static int raid0_run(struct mddev *mddev) 391 { 392 struct r0conf *conf; 393 int ret; 394 395 if (mddev->chunk_sectors == 0) { 396 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev)); 397 return -EINVAL; 398 } 399 if (md_check_no_bitmap(mddev)) 400 return -EINVAL; 401 402 /* if private is not null, we are here after takeover */ 403 if (mddev->private == NULL) { 404 ret = create_strip_zones(mddev, &conf); 405 if (ret < 0) 406 return ret; 407 mddev->private = conf; 408 } 409 conf = mddev->private; 410 if (!mddev_is_dm(mddev)) { 411 ret = raid0_set_limits(mddev); 412 if (ret) 413 return ret; 414 } 415 416 /* calculate array device size */ 417 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); 418 419 pr_debug("md/raid0:%s: md_size is %llu sectors.\n", 420 mdname(mddev), 421 (unsigned long long)mddev->array_sectors); 422 423 dump_zones(mddev); 424 425 return md_integrity_register(mddev); 426 } 427 428 /* 429 * Convert disk_index to the disk order in which it is read/written. 430 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we 431 * write the disks starting at disk 3, then the read/write order would 432 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift() 433 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map 434 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in 435 * that 'output' space to understand the read/write disk ordering. 436 */ 437 static int map_disk_shift(int disk_index, int num_disks, int disk_shift) 438 { 439 return ((disk_index + num_disks - disk_shift) % num_disks); 440 } 441 442 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio) 443 { 444 struct r0conf *conf = mddev->private; 445 struct strip_zone *zone; 446 sector_t start = bio->bi_iter.bi_sector; 447 sector_t end; 448 unsigned int stripe_size; 449 sector_t first_stripe_index, last_stripe_index; 450 sector_t start_disk_offset; 451 unsigned int start_disk_index; 452 sector_t end_disk_offset; 453 unsigned int end_disk_index; 454 unsigned int disk; 455 sector_t orig_start, orig_end; 456 457 orig_start = start; 458 zone = find_zone(conf, &start); 459 460 if (bio_end_sector(bio) > zone->zone_end) { 461 struct bio *split = bio_split(bio, 462 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO, 463 &mddev->bio_set); 464 bio_chain(split, bio); 465 submit_bio_noacct(bio); 466 bio = split; 467 end = zone->zone_end; 468 } else 469 end = bio_end_sector(bio); 470 471 orig_end = end; 472 if (zone != conf->strip_zone) 473 end = end - zone[-1].zone_end; 474 475 /* Now start and end is the offset in zone */ 476 stripe_size = zone->nb_dev * mddev->chunk_sectors; 477 478 first_stripe_index = start; 479 sector_div(first_stripe_index, stripe_size); 480 last_stripe_index = end; 481 sector_div(last_stripe_index, stripe_size); 482 483 /* In the first zone the original and alternate layouts are the same */ 484 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) { 485 sector_div(orig_start, mddev->chunk_sectors); 486 start_disk_index = sector_div(orig_start, zone->nb_dev); 487 start_disk_index = map_disk_shift(start_disk_index, 488 zone->nb_dev, 489 zone->disk_shift); 490 sector_div(orig_end, mddev->chunk_sectors); 491 end_disk_index = sector_div(orig_end, zone->nb_dev); 492 end_disk_index = map_disk_shift(end_disk_index, 493 zone->nb_dev, zone->disk_shift); 494 } else { 495 start_disk_index = (int)(start - first_stripe_index * stripe_size) / 496 mddev->chunk_sectors; 497 end_disk_index = (int)(end - last_stripe_index * stripe_size) / 498 mddev->chunk_sectors; 499 } 500 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) % 501 mddev->chunk_sectors) + 502 first_stripe_index * mddev->chunk_sectors; 503 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) % 504 mddev->chunk_sectors) + 505 last_stripe_index * mddev->chunk_sectors; 506 507 for (disk = 0; disk < zone->nb_dev; disk++) { 508 sector_t dev_start, dev_end; 509 struct md_rdev *rdev; 510 int compare_disk; 511 512 compare_disk = map_disk_shift(disk, zone->nb_dev, 513 zone->disk_shift); 514 515 if (compare_disk < start_disk_index) 516 dev_start = (first_stripe_index + 1) * 517 mddev->chunk_sectors; 518 else if (compare_disk > start_disk_index) 519 dev_start = first_stripe_index * mddev->chunk_sectors; 520 else 521 dev_start = start_disk_offset; 522 523 if (compare_disk < end_disk_index) 524 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; 525 else if (compare_disk > end_disk_index) 526 dev_end = last_stripe_index * mddev->chunk_sectors; 527 else 528 dev_end = end_disk_offset; 529 530 if (dev_end <= dev_start) 531 continue; 532 533 rdev = conf->devlist[(zone - conf->strip_zone) * 534 conf->strip_zone[0].nb_dev + disk]; 535 md_submit_discard_bio(mddev, rdev, bio, 536 dev_start + zone->dev_start + rdev->data_offset, 537 dev_end - dev_start); 538 } 539 bio_endio(bio); 540 } 541 542 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio) 543 { 544 struct r0conf *conf = mddev->private; 545 struct strip_zone *zone; 546 struct md_rdev *tmp_dev; 547 sector_t bio_sector = bio->bi_iter.bi_sector; 548 sector_t sector = bio_sector; 549 550 md_account_bio(mddev, &bio); 551 552 zone = find_zone(mddev->private, §or); 553 switch (conf->layout) { 554 case RAID0_ORIG_LAYOUT: 555 tmp_dev = map_sector(mddev, zone, bio_sector, §or); 556 break; 557 case RAID0_ALT_MULTIZONE_LAYOUT: 558 tmp_dev = map_sector(mddev, zone, sector, §or); 559 break; 560 default: 561 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev)); 562 bio_io_error(bio); 563 return; 564 } 565 566 if (unlikely(is_rdev_broken(tmp_dev))) { 567 bio_io_error(bio); 568 md_error(mddev, tmp_dev); 569 return; 570 } 571 572 bio_set_dev(bio, tmp_dev->bdev); 573 bio->bi_iter.bi_sector = sector + zone->dev_start + 574 tmp_dev->data_offset; 575 mddev_trace_remap(mddev, bio, bio_sector); 576 mddev_check_write_zeroes(mddev, bio); 577 submit_bio_noacct(bio); 578 } 579 580 static bool raid0_make_request(struct mddev *mddev, struct bio *bio) 581 { 582 sector_t sector; 583 unsigned chunk_sects; 584 unsigned sectors; 585 586 if (unlikely(bio->bi_opf & REQ_PREFLUSH) 587 && md_flush_request(mddev, bio)) 588 return true; 589 590 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) { 591 raid0_handle_discard(mddev, bio); 592 return true; 593 } 594 595 sector = bio->bi_iter.bi_sector; 596 chunk_sects = mddev->chunk_sectors; 597 598 sectors = chunk_sects - 599 (likely(is_power_of_2(chunk_sects)) 600 ? (sector & (chunk_sects-1)) 601 : sector_div(sector, chunk_sects)); 602 603 if (sectors < bio_sectors(bio)) { 604 struct bio *split = bio_split(bio, sectors, GFP_NOIO, 605 &mddev->bio_set); 606 bio_chain(split, bio); 607 raid0_map_submit_bio(mddev, bio); 608 bio = split; 609 } 610 611 raid0_map_submit_bio(mddev, bio); 612 return true; 613 } 614 615 static void raid0_status(struct seq_file *seq, struct mddev *mddev) 616 { 617 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); 618 return; 619 } 620 621 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev) 622 { 623 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) { 624 char *md_name = mdname(mddev); 625 626 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n", 627 md_name, rdev->bdev); 628 } 629 } 630 631 static void *raid0_takeover_raid45(struct mddev *mddev) 632 { 633 struct md_rdev *rdev; 634 struct r0conf *priv_conf; 635 636 if (mddev->degraded != 1) { 637 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", 638 mdname(mddev), 639 mddev->degraded); 640 return ERR_PTR(-EINVAL); 641 } 642 643 rdev_for_each(rdev, mddev) { 644 /* check slot number for a disk */ 645 if (rdev->raid_disk == mddev->raid_disks-1) { 646 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n", 647 mdname(mddev)); 648 return ERR_PTR(-EINVAL); 649 } 650 rdev->sectors = mddev->dev_sectors; 651 } 652 653 /* Set new parameters */ 654 mddev->new_level = 0; 655 mddev->new_layout = 0; 656 mddev->new_chunk_sectors = mddev->chunk_sectors; 657 mddev->raid_disks--; 658 mddev->delta_disks = -1; 659 /* make sure it will be not marked as dirty */ 660 mddev->recovery_cp = MaxSector; 661 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 662 663 create_strip_zones(mddev, &priv_conf); 664 665 return priv_conf; 666 } 667 668 static void *raid0_takeover_raid10(struct mddev *mddev) 669 { 670 struct r0conf *priv_conf; 671 672 /* Check layout: 673 * - far_copies must be 1 674 * - near_copies must be 2 675 * - disks number must be even 676 * - all mirrors must be already degraded 677 */ 678 if (mddev->layout != ((1 << 8) + 2)) { 679 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n", 680 mdname(mddev), 681 mddev->layout); 682 return ERR_PTR(-EINVAL); 683 } 684 if (mddev->raid_disks & 1) { 685 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n", 686 mdname(mddev)); 687 return ERR_PTR(-EINVAL); 688 } 689 if (mddev->degraded != (mddev->raid_disks>>1)) { 690 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n", 691 mdname(mddev)); 692 return ERR_PTR(-EINVAL); 693 } 694 695 /* Set new parameters */ 696 mddev->new_level = 0; 697 mddev->new_layout = 0; 698 mddev->new_chunk_sectors = mddev->chunk_sectors; 699 mddev->delta_disks = - mddev->raid_disks / 2; 700 mddev->raid_disks += mddev->delta_disks; 701 mddev->degraded = 0; 702 /* make sure it will be not marked as dirty */ 703 mddev->recovery_cp = MaxSector; 704 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 705 706 create_strip_zones(mddev, &priv_conf); 707 return priv_conf; 708 } 709 710 static void *raid0_takeover_raid1(struct mddev *mddev) 711 { 712 struct r0conf *priv_conf; 713 int chunksect; 714 715 /* Check layout: 716 * - (N - 1) mirror drives must be already faulty 717 */ 718 if ((mddev->raid_disks - 1) != mddev->degraded) { 719 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", 720 mdname(mddev)); 721 return ERR_PTR(-EINVAL); 722 } 723 724 /* 725 * a raid1 doesn't have the notion of chunk size, so 726 * figure out the largest suitable size we can use. 727 */ 728 chunksect = 64 * 2; /* 64K by default */ 729 730 /* The array must be an exact multiple of chunksize */ 731 while (chunksect && (mddev->array_sectors & (chunksect - 1))) 732 chunksect >>= 1; 733 734 if ((chunksect << 9) < PAGE_SIZE) 735 /* array size does not allow a suitable chunk size */ 736 return ERR_PTR(-EINVAL); 737 738 /* Set new parameters */ 739 mddev->new_level = 0; 740 mddev->new_layout = 0; 741 mddev->new_chunk_sectors = chunksect; 742 mddev->chunk_sectors = chunksect; 743 mddev->delta_disks = 1 - mddev->raid_disks; 744 mddev->raid_disks = 1; 745 /* make sure it will be not marked as dirty */ 746 mddev->recovery_cp = MaxSector; 747 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 748 749 create_strip_zones(mddev, &priv_conf); 750 return priv_conf; 751 } 752 753 static void *raid0_takeover(struct mddev *mddev) 754 { 755 /* raid0 can take over: 756 * raid4 - if all data disks are active. 757 * raid5 - providing it is Raid4 layout and one disk is faulty 758 * raid10 - assuming we have all necessary active disks 759 * raid1 - with (N -1) mirror drives faulty 760 */ 761 762 if (mddev->bitmap) { 763 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n", 764 mdname(mddev)); 765 return ERR_PTR(-EBUSY); 766 } 767 if (mddev->level == 4) 768 return raid0_takeover_raid45(mddev); 769 770 if (mddev->level == 5) { 771 if (mddev->layout == ALGORITHM_PARITY_N) 772 return raid0_takeover_raid45(mddev); 773 774 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", 775 mdname(mddev), ALGORITHM_PARITY_N); 776 } 777 778 if (mddev->level == 10) 779 return raid0_takeover_raid10(mddev); 780 781 if (mddev->level == 1) 782 return raid0_takeover_raid1(mddev); 783 784 pr_warn("Takeover from raid%i to raid0 not supported\n", 785 mddev->level); 786 787 return ERR_PTR(-EINVAL); 788 } 789 790 static void raid0_quiesce(struct mddev *mddev, int quiesce) 791 { 792 } 793 794 static struct md_personality raid0_personality= 795 { 796 .name = "raid0", 797 .level = 0, 798 .owner = THIS_MODULE, 799 .make_request = raid0_make_request, 800 .run = raid0_run, 801 .free = raid0_free, 802 .status = raid0_status, 803 .size = raid0_size, 804 .takeover = raid0_takeover, 805 .quiesce = raid0_quiesce, 806 .error_handler = raid0_error, 807 }; 808 809 static int __init raid0_init (void) 810 { 811 return register_md_personality (&raid0_personality); 812 } 813 814 static void raid0_exit (void) 815 { 816 unregister_md_personality (&raid0_personality); 817 } 818 819 module_init(raid0_init); 820 module_exit(raid0_exit); 821 MODULE_LICENSE("GPL"); 822 MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); 823 MODULE_ALIAS("md-personality-2"); /* RAID0 */ 824 MODULE_ALIAS("md-raid0"); 825 MODULE_ALIAS("md-level-0"); 826