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.io_min = mddev->chunk_sectors << 9; 386 lim.io_opt = lim.io_min * mddev->raid_disks; 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 private is not null, we are here after takeover */ 407 if (mddev->private == NULL) { 408 ret = create_strip_zones(mddev, &conf); 409 if (ret < 0) 410 return ret; 411 mddev->private = conf; 412 } 413 conf = mddev->private; 414 if (!mddev_is_dm(mddev)) { 415 ret = raid0_set_limits(mddev); 416 if (ret) 417 return ret; 418 } 419 420 /* calculate array device size */ 421 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); 422 423 pr_debug("md/raid0:%s: md_size is %llu sectors.\n", 424 mdname(mddev), 425 (unsigned long long)mddev->array_sectors); 426 427 dump_zones(mddev); 428 429 return md_integrity_register(mddev); 430 } 431 432 /* 433 * Convert disk_index to the disk order in which it is read/written. 434 * For example, if we have 4 disks, they are numbered 0,1,2,3. If we 435 * write the disks starting at disk 3, then the read/write order would 436 * be disk 3, then 0, then 1, and then disk 2 and we want map_disk_shift() 437 * to map the disks as follows 0,1,2,3 => 1,2,3,0. So disk 0 would map 438 * to 1, 1 to 2, 2 to 3, and 3 to 0. That way we can compare disks in 439 * that 'output' space to understand the read/write disk ordering. 440 */ 441 static int map_disk_shift(int disk_index, int num_disks, int disk_shift) 442 { 443 return ((disk_index + num_disks - disk_shift) % num_disks); 444 } 445 446 static void raid0_handle_discard(struct mddev *mddev, struct bio *bio) 447 { 448 struct r0conf *conf = mddev->private; 449 struct strip_zone *zone; 450 sector_t start = bio->bi_iter.bi_sector; 451 sector_t end; 452 unsigned int stripe_size; 453 sector_t first_stripe_index, last_stripe_index; 454 sector_t start_disk_offset; 455 unsigned int start_disk_index; 456 sector_t end_disk_offset; 457 unsigned int end_disk_index; 458 unsigned int disk; 459 sector_t orig_start, orig_end; 460 461 orig_start = start; 462 zone = find_zone(conf, &start); 463 464 if (bio_end_sector(bio) > zone->zone_end) { 465 struct bio *split = bio_split(bio, 466 zone->zone_end - bio->bi_iter.bi_sector, GFP_NOIO, 467 &mddev->bio_set); 468 469 if (IS_ERR(split)) { 470 bio->bi_status = errno_to_blk_status(PTR_ERR(split)); 471 bio_endio(bio); 472 return; 473 } 474 bio_chain(split, bio); 475 submit_bio_noacct(bio); 476 bio = split; 477 end = zone->zone_end; 478 } else 479 end = bio_end_sector(bio); 480 481 orig_end = end; 482 if (zone != conf->strip_zone) 483 end = end - zone[-1].zone_end; 484 485 /* Now start and end is the offset in zone */ 486 stripe_size = zone->nb_dev * mddev->chunk_sectors; 487 488 first_stripe_index = start; 489 sector_div(first_stripe_index, stripe_size); 490 last_stripe_index = end; 491 sector_div(last_stripe_index, stripe_size); 492 493 /* In the first zone the original and alternate layouts are the same */ 494 if ((conf->layout == RAID0_ORIG_LAYOUT) && (zone != conf->strip_zone)) { 495 sector_div(orig_start, mddev->chunk_sectors); 496 start_disk_index = sector_div(orig_start, zone->nb_dev); 497 start_disk_index = map_disk_shift(start_disk_index, 498 zone->nb_dev, 499 zone->disk_shift); 500 sector_div(orig_end, mddev->chunk_sectors); 501 end_disk_index = sector_div(orig_end, zone->nb_dev); 502 end_disk_index = map_disk_shift(end_disk_index, 503 zone->nb_dev, zone->disk_shift); 504 } else { 505 start_disk_index = (int)(start - first_stripe_index * stripe_size) / 506 mddev->chunk_sectors; 507 end_disk_index = (int)(end - last_stripe_index * stripe_size) / 508 mddev->chunk_sectors; 509 } 510 start_disk_offset = ((int)(start - first_stripe_index * stripe_size) % 511 mddev->chunk_sectors) + 512 first_stripe_index * mddev->chunk_sectors; 513 end_disk_offset = ((int)(end - last_stripe_index * stripe_size) % 514 mddev->chunk_sectors) + 515 last_stripe_index * mddev->chunk_sectors; 516 517 for (disk = 0; disk < zone->nb_dev; disk++) { 518 sector_t dev_start, dev_end; 519 struct md_rdev *rdev; 520 int compare_disk; 521 522 compare_disk = map_disk_shift(disk, zone->nb_dev, 523 zone->disk_shift); 524 525 if (compare_disk < start_disk_index) 526 dev_start = (first_stripe_index + 1) * 527 mddev->chunk_sectors; 528 else if (compare_disk > start_disk_index) 529 dev_start = first_stripe_index * mddev->chunk_sectors; 530 else 531 dev_start = start_disk_offset; 532 533 if (compare_disk < end_disk_index) 534 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; 535 else if (compare_disk > end_disk_index) 536 dev_end = last_stripe_index * mddev->chunk_sectors; 537 else 538 dev_end = end_disk_offset; 539 540 if (dev_end <= dev_start) 541 continue; 542 543 rdev = conf->devlist[(zone - conf->strip_zone) * 544 conf->strip_zone[0].nb_dev + disk]; 545 md_submit_discard_bio(mddev, rdev, bio, 546 dev_start + zone->dev_start + rdev->data_offset, 547 dev_end - dev_start); 548 } 549 bio_endio(bio); 550 } 551 552 static void raid0_map_submit_bio(struct mddev *mddev, struct bio *bio) 553 { 554 struct r0conf *conf = mddev->private; 555 struct strip_zone *zone; 556 struct md_rdev *tmp_dev; 557 sector_t bio_sector = bio->bi_iter.bi_sector; 558 sector_t sector = bio_sector; 559 560 md_account_bio(mddev, &bio); 561 562 zone = find_zone(mddev->private, §or); 563 switch (conf->layout) { 564 case RAID0_ORIG_LAYOUT: 565 tmp_dev = map_sector(mddev, zone, bio_sector, §or); 566 break; 567 case RAID0_ALT_MULTIZONE_LAYOUT: 568 tmp_dev = map_sector(mddev, zone, sector, §or); 569 break; 570 default: 571 WARN(1, "md/raid0:%s: Invalid layout\n", mdname(mddev)); 572 bio_io_error(bio); 573 return; 574 } 575 576 if (unlikely(is_rdev_broken(tmp_dev))) { 577 bio_io_error(bio); 578 md_error(mddev, tmp_dev); 579 return; 580 } 581 582 bio_set_dev(bio, tmp_dev->bdev); 583 bio->bi_iter.bi_sector = sector + zone->dev_start + 584 tmp_dev->data_offset; 585 mddev_trace_remap(mddev, bio, bio_sector); 586 mddev_check_write_zeroes(mddev, bio); 587 submit_bio_noacct(bio); 588 } 589 590 static bool raid0_make_request(struct mddev *mddev, struct bio *bio) 591 { 592 sector_t sector; 593 unsigned chunk_sects; 594 unsigned sectors; 595 596 if (unlikely(bio->bi_opf & REQ_PREFLUSH) 597 && md_flush_request(mddev, bio)) 598 return true; 599 600 if (unlikely((bio_op(bio) == REQ_OP_DISCARD))) { 601 raid0_handle_discard(mddev, bio); 602 return true; 603 } 604 605 sector = bio->bi_iter.bi_sector; 606 chunk_sects = mddev->chunk_sectors; 607 608 sectors = chunk_sects - 609 (likely(is_power_of_2(chunk_sects)) 610 ? (sector & (chunk_sects-1)) 611 : sector_div(sector, chunk_sects)); 612 613 if (sectors < bio_sectors(bio)) { 614 struct bio *split = bio_split(bio, sectors, GFP_NOIO, 615 &mddev->bio_set); 616 617 if (IS_ERR(split)) { 618 bio->bi_status = errno_to_blk_status(PTR_ERR(split)); 619 bio_endio(bio); 620 return true; 621 } 622 bio_chain(split, bio); 623 raid0_map_submit_bio(mddev, bio); 624 bio = split; 625 } 626 627 raid0_map_submit_bio(mddev, bio); 628 return true; 629 } 630 631 static void raid0_status(struct seq_file *seq, struct mddev *mddev) 632 { 633 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); 634 return; 635 } 636 637 static void raid0_error(struct mddev *mddev, struct md_rdev *rdev) 638 { 639 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) { 640 char *md_name = mdname(mddev); 641 642 pr_crit("md/raid0%s: Disk failure on %pg detected, failing array.\n", 643 md_name, rdev->bdev); 644 } 645 } 646 647 static void *raid0_takeover_raid45(struct mddev *mddev) 648 { 649 struct md_rdev *rdev; 650 struct r0conf *priv_conf; 651 652 if (mddev->degraded != 1) { 653 pr_warn("md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", 654 mdname(mddev), 655 mddev->degraded); 656 return ERR_PTR(-EINVAL); 657 } 658 659 rdev_for_each(rdev, mddev) { 660 /* check slot number for a disk */ 661 if (rdev->raid_disk == mddev->raid_disks-1) { 662 pr_warn("md/raid0:%s: raid5 must have missing parity disk!\n", 663 mdname(mddev)); 664 return ERR_PTR(-EINVAL); 665 } 666 rdev->sectors = mddev->dev_sectors; 667 } 668 669 /* Set new parameters */ 670 mddev->new_level = 0; 671 mddev->new_layout = 0; 672 mddev->new_chunk_sectors = mddev->chunk_sectors; 673 mddev->raid_disks--; 674 mddev->delta_disks = -1; 675 /* make sure it will be not marked as dirty */ 676 mddev->recovery_cp = MaxSector; 677 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 678 679 create_strip_zones(mddev, &priv_conf); 680 681 return priv_conf; 682 } 683 684 static void *raid0_takeover_raid10(struct mddev *mddev) 685 { 686 struct r0conf *priv_conf; 687 688 /* Check layout: 689 * - far_copies must be 1 690 * - near_copies must be 2 691 * - disks number must be even 692 * - all mirrors must be already degraded 693 */ 694 if (mddev->layout != ((1 << 8) + 2)) { 695 pr_warn("md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n", 696 mdname(mddev), 697 mddev->layout); 698 return ERR_PTR(-EINVAL); 699 } 700 if (mddev->raid_disks & 1) { 701 pr_warn("md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n", 702 mdname(mddev)); 703 return ERR_PTR(-EINVAL); 704 } 705 if (mddev->degraded != (mddev->raid_disks>>1)) { 706 pr_warn("md/raid0:%s: All mirrors must be already degraded!\n", 707 mdname(mddev)); 708 return ERR_PTR(-EINVAL); 709 } 710 711 /* Set new parameters */ 712 mddev->new_level = 0; 713 mddev->new_layout = 0; 714 mddev->new_chunk_sectors = mddev->chunk_sectors; 715 mddev->delta_disks = - mddev->raid_disks / 2; 716 mddev->raid_disks += mddev->delta_disks; 717 mddev->degraded = 0; 718 /* make sure it will be not marked as dirty */ 719 mddev->recovery_cp = MaxSector; 720 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 721 722 create_strip_zones(mddev, &priv_conf); 723 return priv_conf; 724 } 725 726 static void *raid0_takeover_raid1(struct mddev *mddev) 727 { 728 struct r0conf *priv_conf; 729 int chunksect; 730 731 /* Check layout: 732 * - (N - 1) mirror drives must be already faulty 733 */ 734 if ((mddev->raid_disks - 1) != mddev->degraded) { 735 pr_err("md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", 736 mdname(mddev)); 737 return ERR_PTR(-EINVAL); 738 } 739 740 /* 741 * a raid1 doesn't have the notion of chunk size, so 742 * figure out the largest suitable size we can use. 743 */ 744 chunksect = 64 * 2; /* 64K by default */ 745 746 /* The array must be an exact multiple of chunksize */ 747 while (chunksect && (mddev->array_sectors & (chunksect - 1))) 748 chunksect >>= 1; 749 750 if ((chunksect << 9) < PAGE_SIZE) 751 /* array size does not allow a suitable chunk size */ 752 return ERR_PTR(-EINVAL); 753 754 /* Set new parameters */ 755 mddev->new_level = 0; 756 mddev->new_layout = 0; 757 mddev->new_chunk_sectors = chunksect; 758 mddev->chunk_sectors = chunksect; 759 mddev->delta_disks = 1 - mddev->raid_disks; 760 mddev->raid_disks = 1; 761 /* make sure it will be not marked as dirty */ 762 mddev->recovery_cp = MaxSector; 763 mddev_clear_unsupported_flags(mddev, UNSUPPORTED_MDDEV_FLAGS); 764 765 create_strip_zones(mddev, &priv_conf); 766 return priv_conf; 767 } 768 769 static void *raid0_takeover(struct mddev *mddev) 770 { 771 /* raid0 can take over: 772 * raid4 - if all data disks are active. 773 * raid5 - providing it is Raid4 layout and one disk is faulty 774 * raid10 - assuming we have all necessary active disks 775 * raid1 - with (N -1) mirror drives faulty 776 */ 777 778 if (mddev->bitmap) { 779 pr_warn("md/raid0: %s: cannot takeover array with bitmap\n", 780 mdname(mddev)); 781 return ERR_PTR(-EBUSY); 782 } 783 if (mddev->level == 4) 784 return raid0_takeover_raid45(mddev); 785 786 if (mddev->level == 5) { 787 if (mddev->layout == ALGORITHM_PARITY_N) 788 return raid0_takeover_raid45(mddev); 789 790 pr_warn("md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", 791 mdname(mddev), ALGORITHM_PARITY_N); 792 } 793 794 if (mddev->level == 10) 795 return raid0_takeover_raid10(mddev); 796 797 if (mddev->level == 1) 798 return raid0_takeover_raid1(mddev); 799 800 pr_warn("Takeover from raid%i to raid0 not supported\n", 801 mddev->level); 802 803 return ERR_PTR(-EINVAL); 804 } 805 806 static void raid0_quiesce(struct mddev *mddev, int quiesce) 807 { 808 } 809 810 static struct md_personality raid0_personality= 811 { 812 .head = { 813 .type = MD_PERSONALITY, 814 .id = ID_RAID0, 815 .name = "raid0", 816 .owner = THIS_MODULE, 817 }, 818 819 .make_request = raid0_make_request, 820 .run = raid0_run, 821 .free = raid0_free, 822 .status = raid0_status, 823 .size = raid0_size, 824 .takeover = raid0_takeover, 825 .quiesce = raid0_quiesce, 826 .error_handler = raid0_error, 827 }; 828 829 static int __init raid0_init(void) 830 { 831 return register_md_submodule(&raid0_personality.head); 832 } 833 834 static void __exit raid0_exit(void) 835 { 836 unregister_md_submodule(&raid0_personality.head); 837 } 838 839 module_init(raid0_init); 840 module_exit(raid0_exit); 841 MODULE_LICENSE("GPL"); 842 MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); 843 MODULE_ALIAS("md-personality-2"); /* RAID0 */ 844 MODULE_ALIAS("md-raid0"); 845 MODULE_ALIAS("md-level-0"); 846