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 int blksize = queue_logical_block_size(mddev->gendisk->queue); 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 rdev_for_each(rdev2, mddev) { 88 pr_debug("md/raid0:%s: comparing %pg(%llu)" 89 " with %pg(%llu)\n", 90 mdname(mddev), 91 rdev1->bdev, 92 (unsigned long long)rdev1->sectors, 93 rdev2->bdev, 94 (unsigned long long)rdev2->sectors); 95 if (rdev2 == rdev1) { 96 pr_debug("md/raid0:%s: END\n", 97 mdname(mddev)); 98 break; 99 } 100 if (rdev2->sectors == rdev1->sectors) { 101 /* 102 * Not unique, don't count it as a new 103 * group 104 */ 105 pr_debug("md/raid0:%s: EQUAL\n", 106 mdname(mddev)); 107 c = 1; 108 break; 109 } 110 pr_debug("md/raid0:%s: NOT EQUAL\n", 111 mdname(mddev)); 112 } 113 if (!c) { 114 pr_debug("md/raid0:%s: ==> UNIQUE\n", 115 mdname(mddev)); 116 conf->nr_strip_zones++; 117 pr_debug("md/raid0:%s: %d zones\n", 118 mdname(mddev), conf->nr_strip_zones); 119 } 120 } 121 pr_debug("md/raid0:%s: FINAL %d zones\n", 122 mdname(mddev), conf->nr_strip_zones); 123 124 /* 125 * now since we have the hard sector sizes, we can make sure 126 * chunk size is a multiple of that sector size 127 */ 128 if ((mddev->chunk_sectors << 9) % blksize) { 129 pr_warn("md/raid0:%s: chunk_size of %d not multiple of block size %d\n", 130 mdname(mddev), 131 mddev->chunk_sectors << 9, blksize); 132 err = -EINVAL; 133 goto abort; 134 } 135 136 err = -ENOMEM; 137 conf->strip_zone = kcalloc(conf->nr_strip_zones, 138 sizeof(struct strip_zone), 139 GFP_KERNEL); 140 if (!conf->strip_zone) 141 goto abort; 142 conf->devlist = kzalloc(array3_size(sizeof(struct md_rdev *), 143 conf->nr_strip_zones, 144 mddev->raid_disks), 145 GFP_KERNEL); 146 if (!conf->devlist) 147 goto abort; 148 149 /* The first zone must contain all devices, so here we check that 150 * there is a proper alignment of slots to devices and find them all 151 */ 152 zone = &conf->strip_zone[0]; 153 cnt = 0; 154 smallest = NULL; 155 dev = conf->devlist; 156 err = -EINVAL; 157 rdev_for_each(rdev1, mddev) { 158 int j = rdev1->raid_disk; 159 160 if (mddev->level == 10) { 161 /* taking over a raid10-n2 array */ 162 j /= 2; 163 rdev1->new_raid_disk = j; 164 } 165 166 if (mddev->level == 1) { 167 /* taiking over a raid1 array- 168 * we have only one active disk 169 */ 170 j = 0; 171 rdev1->new_raid_disk = j; 172 } 173 174 if (j < 0) { 175 pr_warn("md/raid0:%s: remove inactive devices before converting to RAID0\n", 176 mdname(mddev)); 177 goto abort; 178 } 179 if (j >= mddev->raid_disks) { 180 pr_warn("md/raid0:%s: bad disk number %d - aborting!\n", 181 mdname(mddev), j); 182 goto abort; 183 } 184 if (dev[j]) { 185 pr_warn("md/raid0:%s: multiple devices for %d - aborting!\n", 186 mdname(mddev), j); 187 goto abort; 188 } 189 dev[j] = rdev1; 190 191 if (!smallest || (rdev1->sectors < smallest->sectors)) 192 smallest = rdev1; 193 cnt++; 194 } 195 if (cnt != mddev->raid_disks) { 196 pr_warn("md/raid0:%s: too few disks (%d of %d) - aborting!\n", 197 mdname(mddev), cnt, mddev->raid_disks); 198 goto abort; 199 } 200 zone->nb_dev = cnt; 201 zone->zone_end = smallest->sectors * cnt; 202 203 curr_zone_end = zone->zone_end; 204 205 /* now do the other zones */ 206 for (i = 1; i < conf->nr_strip_zones; i++) 207 { 208 int j; 209 210 zone = conf->strip_zone + i; 211 dev = conf->devlist + i * mddev->raid_disks; 212 213 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i); 214 zone->dev_start = smallest->sectors; 215 smallest = NULL; 216 c = 0; 217 218 for (j=0; j<cnt; j++) { 219 rdev = conf->devlist[j]; 220 if (rdev->sectors <= zone->dev_start) { 221 pr_debug("md/raid0:%s: checking %pg ... nope\n", 222 mdname(mddev), 223 rdev->bdev); 224 continue; 225 } 226 pr_debug("md/raid0:%s: checking %pg ..." 227 " contained as device %d\n", 228 mdname(mddev), 229 rdev->bdev, c); 230 dev[c] = rdev; 231 c++; 232 if (!smallest || rdev->sectors < smallest->sectors) { 233 smallest = rdev; 234 pr_debug("md/raid0:%s: (%llu) is smallest!.\n", 235 mdname(mddev), 236 (unsigned long long)rdev->sectors); 237 } 238 } 239 240 zone->nb_dev = c; 241 sectors = (smallest->sectors - zone->dev_start) * c; 242 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n", 243 mdname(mddev), 244 zone->nb_dev, (unsigned long long)sectors); 245 246 curr_zone_end += sectors; 247 zone->zone_end = curr_zone_end; 248 249 pr_debug("md/raid0:%s: current zone start: %llu\n", 250 mdname(mddev), 251 (unsigned long long)smallest->sectors); 252 } 253 254 if (conf->nr_strip_zones == 1 || conf->strip_zone[1].nb_dev == 1) { 255 conf->layout = RAID0_ORIG_LAYOUT; 256 } else if (mddev->layout == RAID0_ORIG_LAYOUT || 257 mddev->layout == RAID0_ALT_MULTIZONE_LAYOUT) { 258 conf->layout = mddev->layout; 259 } else if (default_layout == RAID0_ORIG_LAYOUT || 260 default_layout == RAID0_ALT_MULTIZONE_LAYOUT) { 261 conf->layout = default_layout; 262 } else { 263 pr_err("md/raid0:%s: cannot assemble multi-zone RAID0 with default_layout setting\n", 264 mdname(mddev)); 265 pr_err("md/raid0: please set raid0.default_layout to 1 or 2\n"); 266 err = -EOPNOTSUPP; 267 goto abort; 268 } 269 270 if (conf->layout == RAID0_ORIG_LAYOUT) { 271 for (i = 1; i < conf->nr_strip_zones; i++) { 272 sector_t first_sector = conf->strip_zone[i-1].zone_end; 273 274 sector_div(first_sector, mddev->chunk_sectors); 275 zone = conf->strip_zone + i; 276 /* disk_shift is first disk index used in the zone */ 277 zone->disk_shift = sector_div(first_sector, 278 zone->nb_dev); 279 } 280 } 281 282 pr_debug("md/raid0:%s: done.\n", mdname(mddev)); 283 *private_conf = conf; 284 285 return 0; 286 abort: 287 kfree(conf->strip_zone); 288 kfree(conf->devlist); 289 kfree(conf); 290 *private_conf = ERR_PTR(err); 291 return err; 292 } 293 294 /* Find the zone which holds a particular offset 295 * Update *sectorp to be an offset in that zone 296 */ 297 static struct strip_zone *find_zone(struct r0conf *conf, 298 sector_t *sectorp) 299 { 300 int i; 301 struct strip_zone *z = conf->strip_zone; 302 sector_t sector = *sectorp; 303 304 for (i = 0; i < conf->nr_strip_zones; i++) 305 if (sector < z[i].zone_end) { 306 if (i) 307 *sectorp = sector - z[i-1].zone_end; 308 return z + i; 309 } 310 BUG(); 311 } 312 313 /* 314 * remaps the bio to the target device. we separate two flows. 315 * power 2 flow and a general flow for the sake of performance 316 */ 317 static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone, 318 sector_t sector, sector_t *sector_offset) 319 { 320 unsigned int sect_in_chunk; 321 sector_t chunk; 322 struct r0conf *conf = mddev->private; 323 int raid_disks = conf->strip_zone[0].nb_dev; 324 unsigned int chunk_sects = mddev->chunk_sectors; 325 326 if (is_power_of_2(chunk_sects)) { 327 int chunksect_bits = ffz(~chunk_sects); 328 /* find the sector offset inside the chunk */ 329 sect_in_chunk = sector & (chunk_sects - 1); 330 sector >>= chunksect_bits; 331 /* chunk in zone */ 332 chunk = *sector_offset; 333 /* quotient is the chunk in real device*/ 334 sector_div(chunk, zone->nb_dev << chunksect_bits); 335 } else{ 336 sect_in_chunk = sector_div(sector, chunk_sects); 337 chunk = *sector_offset; 338 sector_div(chunk, chunk_sects * zone->nb_dev); 339 } 340 /* 341 * position the bio over the real device 342 * real sector = chunk in device + starting of zone 343 * + the position in the chunk 344 */ 345 *sector_offset = (chunk * chunk_sects) + sect_in_chunk; 346 return conf->devlist[(zone - conf->strip_zone)*raid_disks 347 + sector_div(sector, zone->nb_dev)]; 348 } 349 350 static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks) 351 { 352 sector_t array_sectors = 0; 353 struct md_rdev *rdev; 354 355 WARN_ONCE(sectors || raid_disks, 356 "%s does not support generic reshape\n", __func__); 357 358 rdev_for_each(rdev, mddev) 359 array_sectors += (rdev->sectors & 360 ~(sector_t)(mddev->chunk_sectors-1)); 361 362 return array_sectors; 363 } 364 365 static void raid0_free(struct mddev *mddev, void *priv) 366 { 367 struct r0conf *conf = priv; 368 369 kfree(conf->strip_zone); 370 kfree(conf->devlist); 371 kfree(conf); 372 } 373 374 static int raid0_set_limits(struct mddev *mddev) 375 { 376 struct queue_limits lim; 377 int err; 378 379 md_init_stacking_limits(&lim); 380 lim.max_hw_sectors = mddev->chunk_sectors; 381 lim.max_write_zeroes_sectors = mddev->chunk_sectors; 382 lim.max_hw_wzeroes_unmap_sectors = mddev->chunk_sectors; 383 lim.logical_block_size = mddev->logical_block_size; 384 lim.io_min = mddev->chunk_sectors << 9; 385 lim.io_opt = lim.io_min * mddev->raid_disks; 386 lim.chunk_sectors = mddev->chunk_sectors; 387 lim.features |= BLK_FEAT_ATOMIC_WRITES; 388 err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY); 389 if (err) 390 return err; 391 return queue_limits_set(mddev->gendisk->queue, &lim); 392 } 393 394 static int raid0_run(struct mddev *mddev) 395 { 396 struct r0conf *conf; 397 int ret; 398 399 if (mddev->chunk_sectors == 0) { 400 pr_warn("md/raid0:%s: chunk size must be set.\n", mdname(mddev)); 401 return -EINVAL; 402 } 403 if (md_check_no_bitmap(mddev)) 404 return -EINVAL; 405 406 if (!mddev_is_dm(mddev)) { 407 ret = raid0_set_limits(mddev); 408 if (ret) 409 return ret; 410 } 411 412 /* if private is not null, we are here after takeover */ 413 if (mddev->private == NULL) { 414 ret = create_strip_zones(mddev, &conf); 415 if (ret < 0) 416 return ret; 417 mddev->private = conf; 418 } 419 conf = mddev->private; 420 421 /* calculate array device size */ 422 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); 423 424 pr_debug("md/raid0:%s: md_size is %llu sectors.\n", 425 mdname(mddev), 426 (unsigned long long)mddev->array_sectors); 427 428 dump_zones(mddev); 429 430 return md_integrity_register(mddev); 431 } 432 433 /* 434 * Convert disk_index to the disk order in which it is read/written. 435 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we 436 * write the disks starting at disk 3, then the read/write order would 437 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift() 438 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map 439 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in 440 * that 'output' space to understand the read/write disk ordering. 441 */ 442 static int map_disk_shift(int disk_index, int num_disks, int disk_shift) 443 { 444 return ((disk_index + num_disks - disk_shift) % num_disks); 445 } 446 447 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio) 448 { 449 struct r0conf *conf = mddev->private; 450 struct strip_zone *zone; 451 sector_t start = bio->bi_iter.bi_sector; 452 sector_t end; 453 unsigned int stripe_size; 454 sector_t first_stripe_index, last_stripe_index; 455 sector_t start_disk_offset; 456 unsigned int start_disk_index; 457 sector_t end_disk_offset; 458 unsigned int end_disk_index; 459 unsigned int disk; 460 sector_t orig_start, orig_end; 461 462 orig_start = start; 463 zone = find_zone(conf, &start); 464 465 if (bio_end_sector(bio) > zone->zone_end) { 466 bio = bio_submit_split_bioset(bio, 467 zone->zone_end - bio->bi_iter.bi_sector, 468 &mddev->bio_set); 469 if (!bio) 470 return; 471 472 end = zone->zone_end; 473 } else { 474 end = bio_end_sector(bio); 475 } 476 477 orig_end = end; 478 if (zone != conf->strip_zone) 479 end = end - zone[-1].zone_end; 480 481 /* Now start and end is the offset in zone */ 482 stripe_size = zone->nb_dev * mddev->chunk_sectors; 483 484 first_stripe_index = start; 485 sector_div(first_stripe_index, stripe_size); 486 last_stripe_index = end; 487 sector_div(last_stripe_index, stripe_size); 488 489 /* In the first zone the original and alternate layouts are the same */ 490 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) { 491 sector_div(orig_start, mddev->chunk_sectors); 492 start_disk_index = sector_div(orig_start, zone->nb_dev); 493 start_disk_index = map_disk_shift(start_disk_index, 494 zone->nb_dev, 495 zone->disk_shift); 496 sector_div(orig_end, mddev->chunk_sectors); 497 end_disk_index = sector_div(orig_end, zone->nb_dev); 498 end_disk_index = map_disk_shift(end_disk_index, 499 zone->nb_dev, zone->disk_shift); 500 } else { 501 start_disk_index = (int)(start - first_stripe_index * stripe_size) / 502 mddev->chunk_sectors; 503 end_disk_index = (int)(end - last_stripe_index * stripe_size) / 504 mddev->chunk_sectors; 505 } 506 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) % 507 mddev->chunk_sectors) + 508 first_stripe_index * mddev->chunk_sectors; 509 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) % 510 mddev->chunk_sectors) + 511 last_stripe_index * mddev->chunk_sectors; 512 513 for (disk = 0; disk < zone->nb_dev; disk++) { 514 sector_t dev_start, dev_end; 515 struct md_rdev *rdev; 516 int compare_disk; 517 518 compare_disk = map_disk_shift(disk, zone->nb_dev, 519 zone->disk_shift); 520 521 if (compare_disk < start_disk_index) 522 dev_start = (first_stripe_index + 1) * 523 mddev->chunk_sectors; 524 else if (compare_disk > start_disk_index) 525 dev_start = first_stripe_index * mddev->chunk_sectors; 526 else 527 dev_start = start_disk_offset; 528 529 if (compare_disk < end_disk_index) 530 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; 531 else if (compare_disk > end_disk_index) 532 dev_end = last_stripe_index * mddev->chunk_sectors; 533 else 534 dev_end = end_disk_offset; 535 536 if (dev_end <= dev_start) 537 continue; 538 539 rdev = conf->devlist[(zone - conf->strip_zone) * 540 conf->strip_zone[0].nb_dev + disk]; 541 md_submit_discard_bio(mddev, rdev, bio, 542 dev_start + zone->dev_start + rdev->data_offset, 543 dev_end - dev_start); 544 } 545 bio_endio(bio); 546 } 547 548 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio) 549 { 550 struct r0conf *conf = mddev->private; 551 struct strip_zone *zone; 552 struct md_rdev *tmp_dev; 553 sector_t bio_sector = bio->bi_iter.bi_sector; 554 sector_t sector = bio_sector; 555 556 md_account_bio(mddev, &bio); 557 558 zone = find_zone(mddev->private, §or); 559 switch (conf->layout) { 560 case RAID0_ORIG_LAYOUT: 561 tmp_dev = map_sector(mddev, zone, bio_sector, §or); 562 break; 563 case RAID0_ALT_MULTIZONE_LAYOUT: 564 tmp_dev = map_sector(mddev, zone, sector, §or); 565 break; 566 default: 567 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev)); 568 bio_io_error(bio); 569 return; 570 } 571 572 if (unlikely(is_rdev_broken(tmp_dev))) { 573 bio_io_error(bio); 574 md_error(mddev, tmp_dev); 575 return; 576 } 577 578 bio_set_dev(bio, tmp_dev->bdev); 579 bio->bi_iter.bi_sector = sector + zone->dev_start + 580 tmp_dev->data_offset; 581 mddev_trace_remap(mddev, bio, bio_sector); 582 mddev_check_write_zeroes(mddev, bio); 583 submit_bio_noacct(bio); 584 } 585 586 static bool raid0_make_request(struct mddev *mddev, struct bio *bio) 587 { 588 sector_t sector; 589 unsigned chunk_sects; 590 unsigned sectors; 591 592 if (unlikely(bio->bi_opf & REQ_PREFLUSH) 593 && md_flush_request(mddev, bio)) 594 return true; 595 596 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) { 597 raid0_handle_discard(mddev, bio); 598 return true; 599 } 600 601 sector = bio->bi_iter.bi_sector; 602 chunk_sects = mddev->chunk_sectors; 603 604 sectors = chunk_sects - 605 (likely(is_power_of_2(chunk_sects)) 606 ? (sector & (chunk_sects-1)) 607 : sector_div(sector, chunk_sects)); 608 609 if (sectors < bio_sectors(bio)) { 610 bio = bio_submit_split_bioset(bio, sectors, 611 &mddev->bio_set); 612 if (!bio) 613 return true; 614 } 615 616 raid0_map_submit_bio(mddev, bio); 617 return true; 618 } 619 620 static void raid0_status(struct seq_file *seq, struct mddev *mddev) 621 { 622 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); 623 return; 624 } 625 626 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev) 627 { 628 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) { 629 char *md_name = mdname(mddev); 630 631 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n", 632 md_name, rdev->bdev); 633 } 634 } 635 636 static void *raid0_takeover_raid45(struct mddev *mddev) 637 { 638 struct md_rdev *rdev; 639 struct r0conf *priv_conf; 640 641 if (mddev->degraded != 1) { 642 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", 643 mdname(mddev), 644 mddev->degraded); 645 return ERR_PTR(-EINVAL); 646 } 647 648 rdev_for_each(rdev, mddev) { 649 /* check slot number for a disk */ 650 if (rdev->raid_disk == mddev->raid_disks-1) { 651 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n", 652 mdname(mddev)); 653 return ERR_PTR(-EINVAL); 654 } 655 rdev->sectors = mddev->dev_sectors; 656 } 657 658 /* Set new parameters */ 659 mddev->new_level = 0; 660 mddev->new_layout = 0; 661 mddev->new_chunk_sectors = mddev->chunk_sectors; 662 mddev->raid_disks--; 663 mddev->delta_disks = -1; 664 /* make sure it will be not marked as dirty */ 665 mddev->resync_offset = MaxSector; 666 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 667 668 create_strip_zones(mddev, &priv_conf); 669 670 return priv_conf; 671 } 672 673 static void *raid0_takeover_raid10(struct mddev *mddev) 674 { 675 struct r0conf *priv_conf; 676 677 /* Check layout: 678 * - far_copies must be 1 679 * - near_copies must be 2 680 * - disks number must be even 681 * - all mirrors must be already degraded 682 */ 683 if (mddev->layout != ((1 << 8) + 2)) { 684 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n", 685 mdname(mddev), 686 mddev->layout); 687 return ERR_PTR(-EINVAL); 688 } 689 if (mddev->raid_disks & 1) { 690 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n", 691 mdname(mddev)); 692 return ERR_PTR(-EINVAL); 693 } 694 if (mddev->degraded != (mddev->raid_disks>>1)) { 695 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n", 696 mdname(mddev)); 697 return ERR_PTR(-EINVAL); 698 } 699 700 /* Set new parameters */ 701 mddev->new_level = 0; 702 mddev->new_layout = 0; 703 mddev->new_chunk_sectors = mddev->chunk_sectors; 704 mddev->delta_disks = - mddev->raid_disks / 2; 705 mddev->raid_disks += mddev->delta_disks; 706 mddev->degraded = 0; 707 /* make sure it will be not marked as dirty */ 708 mddev->resync_offset = MaxSector; 709 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 710 711 create_strip_zones(mddev, &priv_conf); 712 return priv_conf; 713 } 714 715 static void *raid0_takeover_raid1(struct mddev *mddev) 716 { 717 struct r0conf *priv_conf; 718 int chunksect; 719 720 /* Check layout: 721 * - (N - 1) mirror drives must be already faulty 722 */ 723 if ((mddev->raid_disks - 1) != mddev->degraded) { 724 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", 725 mdname(mddev)); 726 return ERR_PTR(-EINVAL); 727 } 728 729 /* 730 * a raid1 doesn't have the notion of chunk size, so 731 * figure out the largest suitable size we can use. 732 */ 733 chunksect = 64 * 2; /* 64K by default */ 734 735 /* The array must be an exact multiple of chunksize */ 736 while (chunksect && (mddev->array_sectors & (chunksect - 1))) 737 chunksect >>= 1; 738 739 if ((chunksect << 9) < PAGE_SIZE) 740 /* array size does not allow a suitable chunk size */ 741 return ERR_PTR(-EINVAL); 742 743 /* Set new parameters */ 744 mddev->new_level = 0; 745 mddev->new_layout = 0; 746 mddev->new_chunk_sectors = chunksect; 747 mddev->chunk_sectors = chunksect; 748 mddev->delta_disks = 1 - mddev->raid_disks; 749 mddev->raid_disks = 1; 750 /* make sure it will be not marked as dirty */ 751 mddev->resync_offset = MaxSector; 752 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 753 754 create_strip_zones(mddev, &priv_conf); 755 return priv_conf; 756 } 757 758 static void *raid0_takeover(struct mddev *mddev) 759 { 760 /* raid0 can take over: 761 * raid4 - if all data disks are active. 762 * raid5 - providing it is Raid4 layout and one disk is faulty 763 * raid10 - assuming we have all necessary active disks 764 * raid1 - with (N -1) mirror drives faulty 765 */ 766 767 if (mddev->bitmap) { 768 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n", 769 mdname(mddev)); 770 return ERR_PTR(-EBUSY); 771 } 772 if (mddev->level == 4) 773 return raid0_takeover_raid45(mddev); 774 775 if (mddev->level == 5) { 776 if (mddev->layout == ALGORITHM_PARITY_N) 777 return raid0_takeover_raid45(mddev); 778 779 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", 780 mdname(mddev), ALGORITHM_PARITY_N); 781 } 782 783 if (mddev->level == 10) 784 return raid0_takeover_raid10(mddev); 785 786 if (mddev->level == 1) 787 return raid0_takeover_raid1(mddev); 788 789 pr_warn("Takeover from raid%i to raid0 not supported\n", 790 mddev->level); 791 792 return ERR_PTR(-EINVAL); 793 } 794 795 static void raid0_quiesce(struct mddev *mddev, int quiesce) 796 { 797 } 798 799 static struct md_personality raid0_personality= 800 { 801 .head = { 802 .type = MD_PERSONALITY, 803 .id = ID_RAID0, 804 .name = "raid0", 805 .owner = THIS_MODULE, 806 }, 807 808 .make_request = raid0_make_request, 809 .run = raid0_run, 810 .free = raid0_free, 811 .status = raid0_status, 812 .size = raid0_size, 813 .takeover = raid0_takeover, 814 .quiesce = raid0_quiesce, 815 .error_handler = raid0_error, 816 }; 817 818 static int __init raid0_init(void) 819 { 820 return register_md_submodule(&raid0_personality.head); 821 } 822 823 static void __exit raid0_exit(void) 824 { 825 unregister_md_submodule(&raid0_personality.head); 826 } 827 828 module_init(raid0_init); 829 module_exit(raid0_exit); 830 MODULE_LICENSE("GPL"); 831 MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); 832 MODULE_ALIAS("md-personality-2"); /* RAID0 */ 833 MODULE_ALIAS("md-raid0"); 834 MODULE_ALIAS("md-level-0"); 835