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 int err; 381 382 md_init_stacking_limits(&lim); 383 lim.max_hw_sectors = mddev->chunk_sectors; 384 lim.max_write_zeroes_sectors = mddev->chunk_sectors; 385 lim.max_hw_wzeroes_unmap_sectors = mddev->chunk_sectors; 386 lim.io_min = mddev->chunk_sectors << 9; 387 lim.io_opt = lim.io_min * mddev->raid_disks; 388 lim.chunk_sectors = mddev->chunk_sectors; 389 lim.features |= BLK_FEAT_ATOMIC_WRITES; 390 err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY); 391 if (err) 392 return err; 393 return queue_limits_set(mddev->gendisk->queue, &lim); 394 } 395 396 static int raid0_run(struct mddev *mddev) 397 { 398 struct r0conf *conf; 399 int ret; 400 401 if (mddev->chunk_sectors == 0) { 402 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev)); 403 return -EINVAL; 404 } 405 if (md_check_no_bitmap(mddev)) 406 return -EINVAL; 407 408 /* if private is not null, we are here after takeover */ 409 if (mddev->private == NULL) { 410 ret = create_strip_zones(mddev, &conf); 411 if (ret < 0) 412 return ret; 413 mddev->private = conf; 414 } 415 conf = mddev->private; 416 if (!mddev_is_dm(mddev)) { 417 ret = raid0_set_limits(mddev); 418 if (ret) 419 return ret; 420 } 421 422 /* calculate array device size */ 423 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); 424 425 pr_debug("md/raid0:%s: md_size is %llu sectors.\n", 426 mdname(mddev), 427 (unsigned long long)mddev->array_sectors); 428 429 dump_zones(mddev); 430 431 return md_integrity_register(mddev); 432 } 433 434 /* 435 * Convert disk_index to the disk order in which it is read/written. 436 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we 437 * write the disks starting at disk 3, then the read/write order would 438 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift() 439 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map 440 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in 441 * that 'output' space to understand the read/write disk ordering. 442 */ 443 static int map_disk_shift(int disk_index, int num_disks, int disk_shift) 444 { 445 return ((disk_index + num_disks - disk_shift) % num_disks); 446 } 447 448 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio) 449 { 450 struct r0conf *conf = mddev->private; 451 struct strip_zone *zone; 452 sector_t start = bio->bi_iter.bi_sector; 453 sector_t end; 454 unsigned int stripe_size; 455 sector_t first_stripe_index, last_stripe_index; 456 sector_t start_disk_offset; 457 unsigned int start_disk_index; 458 sector_t end_disk_offset; 459 unsigned int end_disk_index; 460 unsigned int disk; 461 sector_t orig_start, orig_end; 462 463 orig_start = start; 464 zone = find_zone(conf, &start); 465 466 if (bio_end_sector(bio) > zone->zone_end) { 467 bio = bio_submit_split_bioset(bio, 468 zone->zone_end - bio->bi_iter.bi_sector, 469 &mddev->bio_set); 470 if (!bio) 471 return; 472 473 end = zone->zone_end; 474 } else { 475 end = bio_end_sector(bio); 476 } 477 478 orig_end = end; 479 if (zone != conf->strip_zone) 480 end = end - zone[-1].zone_end; 481 482 /* Now start and end is the offset in zone */ 483 stripe_size = zone->nb_dev * mddev->chunk_sectors; 484 485 first_stripe_index = start; 486 sector_div(first_stripe_index, stripe_size); 487 last_stripe_index = end; 488 sector_div(last_stripe_index, stripe_size); 489 490 /* In the first zone the original and alternate layouts are the same */ 491 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) { 492 sector_div(orig_start, mddev->chunk_sectors); 493 start_disk_index = sector_div(orig_start, zone->nb_dev); 494 start_disk_index = map_disk_shift(start_disk_index, 495 zone->nb_dev, 496 zone->disk_shift); 497 sector_div(orig_end, mddev->chunk_sectors); 498 end_disk_index = sector_div(orig_end, zone->nb_dev); 499 end_disk_index = map_disk_shift(end_disk_index, 500 zone->nb_dev, zone->disk_shift); 501 } else { 502 start_disk_index = (int)(start - first_stripe_index * stripe_size) / 503 mddev->chunk_sectors; 504 end_disk_index = (int)(end - last_stripe_index * stripe_size) / 505 mddev->chunk_sectors; 506 } 507 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) % 508 mddev->chunk_sectors) + 509 first_stripe_index * mddev->chunk_sectors; 510 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) % 511 mddev->chunk_sectors) + 512 last_stripe_index * mddev->chunk_sectors; 513 514 for (disk = 0; disk < zone->nb_dev; disk++) { 515 sector_t dev_start, dev_end; 516 struct md_rdev *rdev; 517 int compare_disk; 518 519 compare_disk = map_disk_shift(disk, zone->nb_dev, 520 zone->disk_shift); 521 522 if (compare_disk < start_disk_index) 523 dev_start = (first_stripe_index + 1) * 524 mddev->chunk_sectors; 525 else if (compare_disk > start_disk_index) 526 dev_start = first_stripe_index * mddev->chunk_sectors; 527 else 528 dev_start = start_disk_offset; 529 530 if (compare_disk < end_disk_index) 531 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; 532 else if (compare_disk > end_disk_index) 533 dev_end = last_stripe_index * mddev->chunk_sectors; 534 else 535 dev_end = end_disk_offset; 536 537 if (dev_end <= dev_start) 538 continue; 539 540 rdev = conf->devlist[(zone - conf->strip_zone) * 541 conf->strip_zone[0].nb_dev + disk]; 542 md_submit_discard_bio(mddev, rdev, bio, 543 dev_start + zone->dev_start + rdev->data_offset, 544 dev_end - dev_start); 545 } 546 bio_endio(bio); 547 } 548 549 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio) 550 { 551 struct r0conf *conf = mddev->private; 552 struct strip_zone *zone; 553 struct md_rdev *tmp_dev; 554 sector_t bio_sector = bio->bi_iter.bi_sector; 555 sector_t sector = bio_sector; 556 557 md_account_bio(mddev, &bio); 558 559 zone = find_zone(mddev->private, §or); 560 switch (conf->layout) { 561 case RAID0_ORIG_LAYOUT: 562 tmp_dev = map_sector(mddev, zone, bio_sector, §or); 563 break; 564 case RAID0_ALT_MULTIZONE_LAYOUT: 565 tmp_dev = map_sector(mddev, zone, sector, §or); 566 break; 567 default: 568 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev)); 569 bio_io_error(bio); 570 return; 571 } 572 573 if (unlikely(is_rdev_broken(tmp_dev))) { 574 bio_io_error(bio); 575 md_error(mddev, tmp_dev); 576 return; 577 } 578 579 bio_set_dev(bio, tmp_dev->bdev); 580 bio->bi_iter.bi_sector = sector + zone->dev_start + 581 tmp_dev->data_offset; 582 mddev_trace_remap(mddev, bio, bio_sector); 583 mddev_check_write_zeroes(mddev, bio); 584 submit_bio_noacct(bio); 585 } 586 587 static bool raid0_make_request(struct mddev *mddev, struct bio *bio) 588 { 589 sector_t sector; 590 unsigned chunk_sects; 591 unsigned sectors; 592 593 if (unlikely(bio->bi_opf & REQ_PREFLUSH) 594 && md_flush_request(mddev, bio)) 595 return true; 596 597 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) { 598 raid0_handle_discard(mddev, bio); 599 return true; 600 } 601 602 sector = bio->bi_iter.bi_sector; 603 chunk_sects = mddev->chunk_sectors; 604 605 sectors = chunk_sects - 606 (likely(is_power_of_2(chunk_sects)) 607 ? (sector & (chunk_sects-1)) 608 : sector_div(sector, chunk_sects)); 609 610 if (sectors < bio_sectors(bio)) { 611 bio = bio_submit_split_bioset(bio, sectors, 612 &mddev->bio_set); 613 if (!bio) 614 return true; 615 } 616 617 raid0_map_submit_bio(mddev, bio); 618 return true; 619 } 620 621 static void raid0_status(struct seq_file *seq, struct mddev *mddev) 622 { 623 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); 624 return; 625 } 626 627 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev) 628 { 629 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) { 630 char *md_name = mdname(mddev); 631 632 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n", 633 md_name, rdev->bdev); 634 } 635 } 636 637 static void *raid0_takeover_raid45(struct mddev *mddev) 638 { 639 struct md_rdev *rdev; 640 struct r0conf *priv_conf; 641 642 if (mddev->degraded != 1) { 643 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", 644 mdname(mddev), 645 mddev->degraded); 646 return ERR_PTR(-EINVAL); 647 } 648 649 rdev_for_each(rdev, mddev) { 650 /* check slot number for a disk */ 651 if (rdev->raid_disk == mddev->raid_disks-1) { 652 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n", 653 mdname(mddev)); 654 return ERR_PTR(-EINVAL); 655 } 656 rdev->sectors = mddev->dev_sectors; 657 } 658 659 /* Set new parameters */ 660 mddev->new_level = 0; 661 mddev->new_layout = 0; 662 mddev->new_chunk_sectors = mddev->chunk_sectors; 663 mddev->raid_disks--; 664 mddev->delta_disks = -1; 665 /* make sure it will be not marked as dirty */ 666 mddev->resync_offset = MaxSector; 667 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 668 669 create_strip_zones(mddev, &priv_conf); 670 671 return priv_conf; 672 } 673 674 static void *raid0_takeover_raid10(struct mddev *mddev) 675 { 676 struct r0conf *priv_conf; 677 678 /* Check layout: 679 * - far_copies must be 1 680 * - near_copies must be 2 681 * - disks number must be even 682 * - all mirrors must be already degraded 683 */ 684 if (mddev->layout != ((1 << 8) + 2)) { 685 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n", 686 mdname(mddev), 687 mddev->layout); 688 return ERR_PTR(-EINVAL); 689 } 690 if (mddev->raid_disks & 1) { 691 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n", 692 mdname(mddev)); 693 return ERR_PTR(-EINVAL); 694 } 695 if (mddev->degraded != (mddev->raid_disks>>1)) { 696 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n", 697 mdname(mddev)); 698 return ERR_PTR(-EINVAL); 699 } 700 701 /* Set new parameters */ 702 mddev->new_level = 0; 703 mddev->new_layout = 0; 704 mddev->new_chunk_sectors = mddev->chunk_sectors; 705 mddev->delta_disks = - mddev->raid_disks / 2; 706 mddev->raid_disks += mddev->delta_disks; 707 mddev->degraded = 0; 708 /* make sure it will be not marked as dirty */ 709 mddev->resync_offset = MaxSector; 710 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 711 712 create_strip_zones(mddev, &priv_conf); 713 return priv_conf; 714 } 715 716 static void *raid0_takeover_raid1(struct mddev *mddev) 717 { 718 struct r0conf *priv_conf; 719 int chunksect; 720 721 /* Check layout: 722 * - (N - 1) mirror drives must be already faulty 723 */ 724 if ((mddev->raid_disks - 1) != mddev->degraded) { 725 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", 726 mdname(mddev)); 727 return ERR_PTR(-EINVAL); 728 } 729 730 /* 731 * a raid1 doesn't have the notion of chunk size, so 732 * figure out the largest suitable size we can use. 733 */ 734 chunksect = 64 * 2; /* 64K by default */ 735 736 /* The array must be an exact multiple of chunksize */ 737 while (chunksect && (mddev->array_sectors & (chunksect - 1))) 738 chunksect >>= 1; 739 740 if ((chunksect << 9) < PAGE_SIZE) 741 /* array size does not allow a suitable chunk size */ 742 return ERR_PTR(-EINVAL); 743 744 /* Set new parameters */ 745 mddev->new_level = 0; 746 mddev->new_layout = 0; 747 mddev->new_chunk_sectors = chunksect; 748 mddev->chunk_sectors = chunksect; 749 mddev->delta_disks = 1 - mddev->raid_disks; 750 mddev->raid_disks = 1; 751 /* make sure it will be not marked as dirty */ 752 mddev->resync_offset = MaxSector; 753 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 754 755 create_strip_zones(mddev, &priv_conf); 756 return priv_conf; 757 } 758 759 static void *raid0_takeover(struct mddev *mddev) 760 { 761 /* raid0 can take over: 762 * raid4 - if all data disks are active. 763 * raid5 - providing it is Raid4 layout and one disk is faulty 764 * raid10 - assuming we have all necessary active disks 765 * raid1 - with (N -1) mirror drives faulty 766 */ 767 768 if (mddev->bitmap) { 769 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n", 770 mdname(mddev)); 771 return ERR_PTR(-EBUSY); 772 } 773 if (mddev->level == 4) 774 return raid0_takeover_raid45(mddev); 775 776 if (mddev->level == 5) { 777 if (mddev->layout == ALGORITHM_PARITY_N) 778 return raid0_takeover_raid45(mddev); 779 780 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", 781 mdname(mddev), ALGORITHM_PARITY_N); 782 } 783 784 if (mddev->level == 10) 785 return raid0_takeover_raid10(mddev); 786 787 if (mddev->level == 1) 788 return raid0_takeover_raid1(mddev); 789 790 pr_warn("Takeover from raid%i to raid0 not supported\n", 791 mddev->level); 792 793 return ERR_PTR(-EINVAL); 794 } 795 796 static void raid0_quiesce(struct mddev *mddev, int quiesce) 797 { 798 } 799 800 static struct md_personality raid0_personality= 801 { 802 .head = { 803 .type = MD_PERSONALITY, 804 .id = ID_RAID0, 805 .name = "raid0", 806 .owner = THIS_MODULE, 807 }, 808 809 .make_request = raid0_make_request, 810 .run = raid0_run, 811 .free = raid0_free, 812 .status = raid0_status, 813 .size = raid0_size, 814 .takeover = raid0_takeover, 815 .quiesce = raid0_quiesce, 816 .error_handler = raid0_error, 817 }; 818 819 static int __init raid0_init(void) 820 { 821 return register_md_submodule(&raid0_personality.head); 822 } 823 824 static void __exit raid0_exit(void) 825 { 826 unregister_md_submodule(&raid0_personality.head); 827 } 828 829 module_init(raid0_init); 830 module_exit(raid0_exit); 831 MODULE_LICENSE("GPL"); 832 MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); 833 MODULE_ALIAS("md-personality-2"); /* RAID0 */ 834 MODULE_ALIAS("md-raid0"); 835 MODULE_ALIAS("md-level-0"); 836