1 /* 2 * raid1.c : Multiple Devices driver for Linux 3 * 4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat 5 * 6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman 7 * 8 * RAID-1 management functions. 9 * 10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 11 * 12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> 13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> 14 * 15 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support 16 * bitmapped intelligence in resync: 17 * 18 * - bitmap marked during normal i/o 19 * - bitmap used to skip nondirty blocks during sync 20 * 21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: 22 * - persistent bitmap code 23 * 24 * This program is free software; you can redistribute it and/or modify 25 * it under the terms of the GNU General Public License as published by 26 * the Free Software Foundation; either version 2, or (at your option) 27 * any later version. 28 * 29 * You should have received a copy of the GNU General Public License 30 * (for example /usr/src/linux/COPYING); if not, write to the Free 31 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 32 */ 33 34 #include <linux/slab.h> 35 #include <linux/delay.h> 36 #include <linux/blkdev.h> 37 #include <linux/module.h> 38 #include <linux/seq_file.h> 39 #include <linux/ratelimit.h> 40 #include <trace/events/block.h> 41 #include "md.h" 42 #include "raid1.h" 43 #include "bitmap.h" 44 45 #define UNSUPPORTED_MDDEV_FLAGS \ 46 ((1L << MD_HAS_JOURNAL) | \ 47 (1L << MD_JOURNAL_CLEAN)) 48 49 /* 50 * Number of guaranteed r1bios in case of extreme VM load: 51 */ 52 #define NR_RAID1_BIOS 256 53 54 /* when we get a read error on a read-only array, we redirect to another 55 * device without failing the first device, or trying to over-write to 56 * correct the read error. To keep track of bad blocks on a per-bio 57 * level, we store IO_BLOCKED in the appropriate 'bios' pointer 58 */ 59 #define IO_BLOCKED ((struct bio *)1) 60 /* When we successfully write to a known bad-block, we need to remove the 61 * bad-block marking which must be done from process context. So we record 62 * the success by setting devs[n].bio to IO_MADE_GOOD 63 */ 64 #define IO_MADE_GOOD ((struct bio *)2) 65 66 #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) 67 68 /* When there are this many requests queue to be written by 69 * the raid1 thread, we become 'congested' to provide back-pressure 70 * for writeback. 71 */ 72 static int max_queued_requests = 1024; 73 74 static void allow_barrier(struct r1conf *conf, sector_t sector_nr); 75 static void lower_barrier(struct r1conf *conf, sector_t sector_nr); 76 77 #define raid1_log(md, fmt, args...) \ 78 do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0) 79 80 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) 81 { 82 struct pool_info *pi = data; 83 int size = offsetof(struct r1bio, bios[pi->raid_disks]); 84 85 /* allocate a r1bio with room for raid_disks entries in the bios array */ 86 return kzalloc(size, gfp_flags); 87 } 88 89 static void r1bio_pool_free(void *r1_bio, void *data) 90 { 91 kfree(r1_bio); 92 } 93 94 #define RESYNC_BLOCK_SIZE (64*1024) 95 #define RESYNC_DEPTH 32 96 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) 97 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) 98 #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) 99 #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) 100 #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) 101 #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) 102 103 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) 104 { 105 struct pool_info *pi = data; 106 struct r1bio *r1_bio; 107 struct bio *bio; 108 int need_pages; 109 int i, j; 110 111 r1_bio = r1bio_pool_alloc(gfp_flags, pi); 112 if (!r1_bio) 113 return NULL; 114 115 /* 116 * Allocate bios : 1 for reading, n-1 for writing 117 */ 118 for (j = pi->raid_disks ; j-- ; ) { 119 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); 120 if (!bio) 121 goto out_free_bio; 122 r1_bio->bios[j] = bio; 123 } 124 /* 125 * Allocate RESYNC_PAGES data pages and attach them to 126 * the first bio. 127 * If this is a user-requested check/repair, allocate 128 * RESYNC_PAGES for each bio. 129 */ 130 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) 131 need_pages = pi->raid_disks; 132 else 133 need_pages = 1; 134 for (j = 0; j < need_pages; j++) { 135 bio = r1_bio->bios[j]; 136 bio->bi_vcnt = RESYNC_PAGES; 137 138 if (bio_alloc_pages(bio, gfp_flags)) 139 goto out_free_pages; 140 } 141 /* If not user-requests, copy the page pointers to all bios */ 142 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { 143 for (i=0; i<RESYNC_PAGES ; i++) 144 for (j=1; j<pi->raid_disks; j++) 145 r1_bio->bios[j]->bi_io_vec[i].bv_page = 146 r1_bio->bios[0]->bi_io_vec[i].bv_page; 147 } 148 149 r1_bio->master_bio = NULL; 150 151 return r1_bio; 152 153 out_free_pages: 154 while (--j >= 0) 155 bio_free_pages(r1_bio->bios[j]); 156 157 out_free_bio: 158 while (++j < pi->raid_disks) 159 bio_put(r1_bio->bios[j]); 160 r1bio_pool_free(r1_bio, data); 161 return NULL; 162 } 163 164 static void r1buf_pool_free(void *__r1_bio, void *data) 165 { 166 struct pool_info *pi = data; 167 int i,j; 168 struct r1bio *r1bio = __r1_bio; 169 170 for (i = 0; i < RESYNC_PAGES; i++) 171 for (j = pi->raid_disks; j-- ;) { 172 if (j == 0 || 173 r1bio->bios[j]->bi_io_vec[i].bv_page != 174 r1bio->bios[0]->bi_io_vec[i].bv_page) 175 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); 176 } 177 for (i=0 ; i < pi->raid_disks; i++) 178 bio_put(r1bio->bios[i]); 179 180 r1bio_pool_free(r1bio, data); 181 } 182 183 static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) 184 { 185 int i; 186 187 for (i = 0; i < conf->raid_disks * 2; i++) { 188 struct bio **bio = r1_bio->bios + i; 189 if (!BIO_SPECIAL(*bio)) 190 bio_put(*bio); 191 *bio = NULL; 192 } 193 } 194 195 static void free_r1bio(struct r1bio *r1_bio) 196 { 197 struct r1conf *conf = r1_bio->mddev->private; 198 199 put_all_bios(conf, r1_bio); 200 mempool_free(r1_bio, conf->r1bio_pool); 201 } 202 203 static void put_buf(struct r1bio *r1_bio) 204 { 205 struct r1conf *conf = r1_bio->mddev->private; 206 sector_t sect = r1_bio->sector; 207 int i; 208 209 for (i = 0; i < conf->raid_disks * 2; i++) { 210 struct bio *bio = r1_bio->bios[i]; 211 if (bio->bi_end_io) 212 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); 213 } 214 215 mempool_free(r1_bio, conf->r1buf_pool); 216 217 lower_barrier(conf, sect); 218 } 219 220 static void reschedule_retry(struct r1bio *r1_bio) 221 { 222 unsigned long flags; 223 struct mddev *mddev = r1_bio->mddev; 224 struct r1conf *conf = mddev->private; 225 int idx; 226 227 idx = sector_to_idx(r1_bio->sector); 228 spin_lock_irqsave(&conf->device_lock, flags); 229 list_add(&r1_bio->retry_list, &conf->retry_list); 230 atomic_inc(&conf->nr_queued[idx]); 231 spin_unlock_irqrestore(&conf->device_lock, flags); 232 233 wake_up(&conf->wait_barrier); 234 md_wakeup_thread(mddev->thread); 235 } 236 237 /* 238 * raid_end_bio_io() is called when we have finished servicing a mirrored 239 * operation and are ready to return a success/failure code to the buffer 240 * cache layer. 241 */ 242 static void call_bio_endio(struct r1bio *r1_bio) 243 { 244 struct bio *bio = r1_bio->master_bio; 245 int done; 246 struct r1conf *conf = r1_bio->mddev->private; 247 sector_t bi_sector = bio->bi_iter.bi_sector; 248 249 if (bio->bi_phys_segments) { 250 unsigned long flags; 251 spin_lock_irqsave(&conf->device_lock, flags); 252 bio->bi_phys_segments--; 253 done = (bio->bi_phys_segments == 0); 254 spin_unlock_irqrestore(&conf->device_lock, flags); 255 /* 256 * make_request() might be waiting for 257 * bi_phys_segments to decrease 258 */ 259 wake_up(&conf->wait_barrier); 260 } else 261 done = 1; 262 263 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) 264 bio->bi_error = -EIO; 265 266 if (done) { 267 bio_endio(bio); 268 /* 269 * Wake up any possible resync thread that waits for the device 270 * to go idle. 271 */ 272 allow_barrier(conf, bi_sector); 273 } 274 } 275 276 static void raid_end_bio_io(struct r1bio *r1_bio) 277 { 278 struct bio *bio = r1_bio->master_bio; 279 280 /* if nobody has done the final endio yet, do it now */ 281 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { 282 pr_debug("raid1: sync end %s on sectors %llu-%llu\n", 283 (bio_data_dir(bio) == WRITE) ? "write" : "read", 284 (unsigned long long) bio->bi_iter.bi_sector, 285 (unsigned long long) bio_end_sector(bio) - 1); 286 287 call_bio_endio(r1_bio); 288 } 289 free_r1bio(r1_bio); 290 } 291 292 /* 293 * Update disk head position estimator based on IRQ completion info. 294 */ 295 static inline void update_head_pos(int disk, struct r1bio *r1_bio) 296 { 297 struct r1conf *conf = r1_bio->mddev->private; 298 299 conf->mirrors[disk].head_position = 300 r1_bio->sector + (r1_bio->sectors); 301 } 302 303 /* 304 * Find the disk number which triggered given bio 305 */ 306 static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) 307 { 308 int mirror; 309 struct r1conf *conf = r1_bio->mddev->private; 310 int raid_disks = conf->raid_disks; 311 312 for (mirror = 0; mirror < raid_disks * 2; mirror++) 313 if (r1_bio->bios[mirror] == bio) 314 break; 315 316 BUG_ON(mirror == raid_disks * 2); 317 update_head_pos(mirror, r1_bio); 318 319 return mirror; 320 } 321 322 static void raid1_end_read_request(struct bio *bio) 323 { 324 int uptodate = !bio->bi_error; 325 struct r1bio *r1_bio = bio->bi_private; 326 struct r1conf *conf = r1_bio->mddev->private; 327 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; 328 329 /* 330 * this branch is our 'one mirror IO has finished' event handler: 331 */ 332 update_head_pos(r1_bio->read_disk, r1_bio); 333 334 if (uptodate) 335 set_bit(R1BIO_Uptodate, &r1_bio->state); 336 else if (test_bit(FailFast, &rdev->flags) && 337 test_bit(R1BIO_FailFast, &r1_bio->state)) 338 /* This was a fail-fast read so we definitely 339 * want to retry */ 340 ; 341 else { 342 /* If all other devices have failed, we want to return 343 * the error upwards rather than fail the last device. 344 * Here we redefine "uptodate" to mean "Don't want to retry" 345 */ 346 unsigned long flags; 347 spin_lock_irqsave(&conf->device_lock, flags); 348 if (r1_bio->mddev->degraded == conf->raid_disks || 349 (r1_bio->mddev->degraded == conf->raid_disks-1 && 350 test_bit(In_sync, &rdev->flags))) 351 uptodate = 1; 352 spin_unlock_irqrestore(&conf->device_lock, flags); 353 } 354 355 if (uptodate) { 356 raid_end_bio_io(r1_bio); 357 rdev_dec_pending(rdev, conf->mddev); 358 } else { 359 /* 360 * oops, read error: 361 */ 362 char b[BDEVNAME_SIZE]; 363 pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", 364 mdname(conf->mddev), 365 bdevname(rdev->bdev, b), 366 (unsigned long long)r1_bio->sector); 367 set_bit(R1BIO_ReadError, &r1_bio->state); 368 reschedule_retry(r1_bio); 369 /* don't drop the reference on read_disk yet */ 370 } 371 } 372 373 static void close_write(struct r1bio *r1_bio) 374 { 375 /* it really is the end of this request */ 376 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { 377 /* free extra copy of the data pages */ 378 int i = r1_bio->behind_page_count; 379 while (i--) 380 safe_put_page(r1_bio->behind_bvecs[i].bv_page); 381 kfree(r1_bio->behind_bvecs); 382 r1_bio->behind_bvecs = NULL; 383 } 384 /* clear the bitmap if all writes complete successfully */ 385 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, 386 r1_bio->sectors, 387 !test_bit(R1BIO_Degraded, &r1_bio->state), 388 test_bit(R1BIO_BehindIO, &r1_bio->state)); 389 md_write_end(r1_bio->mddev); 390 } 391 392 static void r1_bio_write_done(struct r1bio *r1_bio) 393 { 394 if (!atomic_dec_and_test(&r1_bio->remaining)) 395 return; 396 397 if (test_bit(R1BIO_WriteError, &r1_bio->state)) 398 reschedule_retry(r1_bio); 399 else { 400 close_write(r1_bio); 401 if (test_bit(R1BIO_MadeGood, &r1_bio->state)) 402 reschedule_retry(r1_bio); 403 else 404 raid_end_bio_io(r1_bio); 405 } 406 } 407 408 static void raid1_end_write_request(struct bio *bio) 409 { 410 struct r1bio *r1_bio = bio->bi_private; 411 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); 412 struct r1conf *conf = r1_bio->mddev->private; 413 struct bio *to_put = NULL; 414 int mirror = find_bio_disk(r1_bio, bio); 415 struct md_rdev *rdev = conf->mirrors[mirror].rdev; 416 bool discard_error; 417 418 discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD; 419 420 /* 421 * 'one mirror IO has finished' event handler: 422 */ 423 if (bio->bi_error && !discard_error) { 424 set_bit(WriteErrorSeen, &rdev->flags); 425 if (!test_and_set_bit(WantReplacement, &rdev->flags)) 426 set_bit(MD_RECOVERY_NEEDED, & 427 conf->mddev->recovery); 428 429 if (test_bit(FailFast, &rdev->flags) && 430 (bio->bi_opf & MD_FAILFAST) && 431 /* We never try FailFast to WriteMostly devices */ 432 !test_bit(WriteMostly, &rdev->flags)) { 433 md_error(r1_bio->mddev, rdev); 434 if (!test_bit(Faulty, &rdev->flags)) 435 /* This is the only remaining device, 436 * We need to retry the write without 437 * FailFast 438 */ 439 set_bit(R1BIO_WriteError, &r1_bio->state); 440 else { 441 /* Finished with this branch */ 442 r1_bio->bios[mirror] = NULL; 443 to_put = bio; 444 } 445 } else 446 set_bit(R1BIO_WriteError, &r1_bio->state); 447 } else { 448 /* 449 * Set R1BIO_Uptodate in our master bio, so that we 450 * will return a good error code for to the higher 451 * levels even if IO on some other mirrored buffer 452 * fails. 453 * 454 * The 'master' represents the composite IO operation 455 * to user-side. So if something waits for IO, then it 456 * will wait for the 'master' bio. 457 */ 458 sector_t first_bad; 459 int bad_sectors; 460 461 r1_bio->bios[mirror] = NULL; 462 to_put = bio; 463 /* 464 * Do not set R1BIO_Uptodate if the current device is 465 * rebuilding or Faulty. This is because we cannot use 466 * such device for properly reading the data back (we could 467 * potentially use it, if the current write would have felt 468 * before rdev->recovery_offset, but for simplicity we don't 469 * check this here. 470 */ 471 if (test_bit(In_sync, &rdev->flags) && 472 !test_bit(Faulty, &rdev->flags)) 473 set_bit(R1BIO_Uptodate, &r1_bio->state); 474 475 /* Maybe we can clear some bad blocks. */ 476 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, 477 &first_bad, &bad_sectors) && !discard_error) { 478 r1_bio->bios[mirror] = IO_MADE_GOOD; 479 set_bit(R1BIO_MadeGood, &r1_bio->state); 480 } 481 } 482 483 if (behind) { 484 if (test_bit(WriteMostly, &rdev->flags)) 485 atomic_dec(&r1_bio->behind_remaining); 486 487 /* 488 * In behind mode, we ACK the master bio once the I/O 489 * has safely reached all non-writemostly 490 * disks. Setting the Returned bit ensures that this 491 * gets done only once -- we don't ever want to return 492 * -EIO here, instead we'll wait 493 */ 494 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && 495 test_bit(R1BIO_Uptodate, &r1_bio->state)) { 496 /* Maybe we can return now */ 497 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { 498 struct bio *mbio = r1_bio->master_bio; 499 pr_debug("raid1: behind end write sectors" 500 " %llu-%llu\n", 501 (unsigned long long) mbio->bi_iter.bi_sector, 502 (unsigned long long) bio_end_sector(mbio) - 1); 503 call_bio_endio(r1_bio); 504 } 505 } 506 } 507 if (r1_bio->bios[mirror] == NULL) 508 rdev_dec_pending(rdev, conf->mddev); 509 510 /* 511 * Let's see if all mirrored write operations have finished 512 * already. 513 */ 514 r1_bio_write_done(r1_bio); 515 516 if (to_put) 517 bio_put(to_put); 518 } 519 520 static sector_t align_to_barrier_unit_end(sector_t start_sector, 521 sector_t sectors) 522 { 523 sector_t len; 524 525 WARN_ON(sectors == 0); 526 /* 527 * len is the number of sectors from start_sector to end of the 528 * barrier unit which start_sector belongs to. 529 */ 530 len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - 531 start_sector; 532 533 if (len > sectors) 534 len = sectors; 535 536 return len; 537 } 538 539 /* 540 * This routine returns the disk from which the requested read should 541 * be done. There is a per-array 'next expected sequential IO' sector 542 * number - if this matches on the next IO then we use the last disk. 543 * There is also a per-disk 'last know head position' sector that is 544 * maintained from IRQ contexts, both the normal and the resync IO 545 * completion handlers update this position correctly. If there is no 546 * perfect sequential match then we pick the disk whose head is closest. 547 * 548 * If there are 2 mirrors in the same 2 devices, performance degrades 549 * because position is mirror, not device based. 550 * 551 * The rdev for the device selected will have nr_pending incremented. 552 */ 553 static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) 554 { 555 const sector_t this_sector = r1_bio->sector; 556 int sectors; 557 int best_good_sectors; 558 int best_disk, best_dist_disk, best_pending_disk; 559 int has_nonrot_disk; 560 int disk; 561 sector_t best_dist; 562 unsigned int min_pending; 563 struct md_rdev *rdev; 564 int choose_first; 565 int choose_next_idle; 566 567 rcu_read_lock(); 568 /* 569 * Check if we can balance. We can balance on the whole 570 * device if no resync is going on, or below the resync window. 571 * We take the first readable disk when above the resync window. 572 */ 573 retry: 574 sectors = r1_bio->sectors; 575 best_disk = -1; 576 best_dist_disk = -1; 577 best_dist = MaxSector; 578 best_pending_disk = -1; 579 min_pending = UINT_MAX; 580 best_good_sectors = 0; 581 has_nonrot_disk = 0; 582 choose_next_idle = 0; 583 clear_bit(R1BIO_FailFast, &r1_bio->state); 584 585 if ((conf->mddev->recovery_cp < this_sector + sectors) || 586 (mddev_is_clustered(conf->mddev) && 587 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, 588 this_sector + sectors))) 589 choose_first = 1; 590 else 591 choose_first = 0; 592 593 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { 594 sector_t dist; 595 sector_t first_bad; 596 int bad_sectors; 597 unsigned int pending; 598 bool nonrot; 599 600 rdev = rcu_dereference(conf->mirrors[disk].rdev); 601 if (r1_bio->bios[disk] == IO_BLOCKED 602 || rdev == NULL 603 || test_bit(Faulty, &rdev->flags)) 604 continue; 605 if (!test_bit(In_sync, &rdev->flags) && 606 rdev->recovery_offset < this_sector + sectors) 607 continue; 608 if (test_bit(WriteMostly, &rdev->flags)) { 609 /* Don't balance among write-mostly, just 610 * use the first as a last resort */ 611 if (best_dist_disk < 0) { 612 if (is_badblock(rdev, this_sector, sectors, 613 &first_bad, &bad_sectors)) { 614 if (first_bad <= this_sector) 615 /* Cannot use this */ 616 continue; 617 best_good_sectors = first_bad - this_sector; 618 } else 619 best_good_sectors = sectors; 620 best_dist_disk = disk; 621 best_pending_disk = disk; 622 } 623 continue; 624 } 625 /* This is a reasonable device to use. It might 626 * even be best. 627 */ 628 if (is_badblock(rdev, this_sector, sectors, 629 &first_bad, &bad_sectors)) { 630 if (best_dist < MaxSector) 631 /* already have a better device */ 632 continue; 633 if (first_bad <= this_sector) { 634 /* cannot read here. If this is the 'primary' 635 * device, then we must not read beyond 636 * bad_sectors from another device.. 637 */ 638 bad_sectors -= (this_sector - first_bad); 639 if (choose_first && sectors > bad_sectors) 640 sectors = bad_sectors; 641 if (best_good_sectors > sectors) 642 best_good_sectors = sectors; 643 644 } else { 645 sector_t good_sectors = first_bad - this_sector; 646 if (good_sectors > best_good_sectors) { 647 best_good_sectors = good_sectors; 648 best_disk = disk; 649 } 650 if (choose_first) 651 break; 652 } 653 continue; 654 } else 655 best_good_sectors = sectors; 656 657 if (best_disk >= 0) 658 /* At least two disks to choose from so failfast is OK */ 659 set_bit(R1BIO_FailFast, &r1_bio->state); 660 661 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); 662 has_nonrot_disk |= nonrot; 663 pending = atomic_read(&rdev->nr_pending); 664 dist = abs(this_sector - conf->mirrors[disk].head_position); 665 if (choose_first) { 666 best_disk = disk; 667 break; 668 } 669 /* Don't change to another disk for sequential reads */ 670 if (conf->mirrors[disk].next_seq_sect == this_sector 671 || dist == 0) { 672 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; 673 struct raid1_info *mirror = &conf->mirrors[disk]; 674 675 best_disk = disk; 676 /* 677 * If buffered sequential IO size exceeds optimal 678 * iosize, check if there is idle disk. If yes, choose 679 * the idle disk. read_balance could already choose an 680 * idle disk before noticing it's a sequential IO in 681 * this disk. This doesn't matter because this disk 682 * will idle, next time it will be utilized after the 683 * first disk has IO size exceeds optimal iosize. In 684 * this way, iosize of the first disk will be optimal 685 * iosize at least. iosize of the second disk might be 686 * small, but not a big deal since when the second disk 687 * starts IO, the first disk is likely still busy. 688 */ 689 if (nonrot && opt_iosize > 0 && 690 mirror->seq_start != MaxSector && 691 mirror->next_seq_sect > opt_iosize && 692 mirror->next_seq_sect - opt_iosize >= 693 mirror->seq_start) { 694 choose_next_idle = 1; 695 continue; 696 } 697 break; 698 } 699 700 if (choose_next_idle) 701 continue; 702 703 if (min_pending > pending) { 704 min_pending = pending; 705 best_pending_disk = disk; 706 } 707 708 if (dist < best_dist) { 709 best_dist = dist; 710 best_dist_disk = disk; 711 } 712 } 713 714 /* 715 * If all disks are rotational, choose the closest disk. If any disk is 716 * non-rotational, choose the disk with less pending request even the 717 * disk is rotational, which might/might not be optimal for raids with 718 * mixed ratation/non-rotational disks depending on workload. 719 */ 720 if (best_disk == -1) { 721 if (has_nonrot_disk || min_pending == 0) 722 best_disk = best_pending_disk; 723 else 724 best_disk = best_dist_disk; 725 } 726 727 if (best_disk >= 0) { 728 rdev = rcu_dereference(conf->mirrors[best_disk].rdev); 729 if (!rdev) 730 goto retry; 731 atomic_inc(&rdev->nr_pending); 732 sectors = best_good_sectors; 733 734 if (conf->mirrors[best_disk].next_seq_sect != this_sector) 735 conf->mirrors[best_disk].seq_start = this_sector; 736 737 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; 738 } 739 rcu_read_unlock(); 740 *max_sectors = sectors; 741 742 return best_disk; 743 } 744 745 static int raid1_congested(struct mddev *mddev, int bits) 746 { 747 struct r1conf *conf = mddev->private; 748 int i, ret = 0; 749 750 if ((bits & (1 << WB_async_congested)) && 751 conf->pending_count >= max_queued_requests) 752 return 1; 753 754 rcu_read_lock(); 755 for (i = 0; i < conf->raid_disks * 2; i++) { 756 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); 757 if (rdev && !test_bit(Faulty, &rdev->flags)) { 758 struct request_queue *q = bdev_get_queue(rdev->bdev); 759 760 BUG_ON(!q); 761 762 /* Note the '|| 1' - when read_balance prefers 763 * non-congested targets, it can be removed 764 */ 765 if ((bits & (1 << WB_async_congested)) || 1) 766 ret |= bdi_congested(&q->backing_dev_info, bits); 767 else 768 ret &= bdi_congested(&q->backing_dev_info, bits); 769 } 770 } 771 rcu_read_unlock(); 772 return ret; 773 } 774 775 static void flush_pending_writes(struct r1conf *conf) 776 { 777 /* Any writes that have been queued but are awaiting 778 * bitmap updates get flushed here. 779 */ 780 spin_lock_irq(&conf->device_lock); 781 782 if (conf->pending_bio_list.head) { 783 struct bio *bio; 784 bio = bio_list_get(&conf->pending_bio_list); 785 conf->pending_count = 0; 786 spin_unlock_irq(&conf->device_lock); 787 /* flush any pending bitmap writes to 788 * disk before proceeding w/ I/O */ 789 bitmap_unplug(conf->mddev->bitmap); 790 wake_up(&conf->wait_barrier); 791 792 while (bio) { /* submit pending writes */ 793 struct bio *next = bio->bi_next; 794 struct md_rdev *rdev = (void*)bio->bi_bdev; 795 bio->bi_next = NULL; 796 bio->bi_bdev = rdev->bdev; 797 if (test_bit(Faulty, &rdev->flags)) { 798 bio->bi_error = -EIO; 799 bio_endio(bio); 800 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && 801 !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) 802 /* Just ignore it */ 803 bio_endio(bio); 804 else 805 generic_make_request(bio); 806 bio = next; 807 } 808 } else 809 spin_unlock_irq(&conf->device_lock); 810 } 811 812 /* Barriers.... 813 * Sometimes we need to suspend IO while we do something else, 814 * either some resync/recovery, or reconfigure the array. 815 * To do this we raise a 'barrier'. 816 * The 'barrier' is a counter that can be raised multiple times 817 * to count how many activities are happening which preclude 818 * normal IO. 819 * We can only raise the barrier if there is no pending IO. 820 * i.e. if nr_pending == 0. 821 * We choose only to raise the barrier if no-one is waiting for the 822 * barrier to go down. This means that as soon as an IO request 823 * is ready, no other operations which require a barrier will start 824 * until the IO request has had a chance. 825 * 826 * So: regular IO calls 'wait_barrier'. When that returns there 827 * is no backgroup IO happening, It must arrange to call 828 * allow_barrier when it has finished its IO. 829 * backgroup IO calls must call raise_barrier. Once that returns 830 * there is no normal IO happeing. It must arrange to call 831 * lower_barrier when the particular background IO completes. 832 */ 833 static void raise_barrier(struct r1conf *conf, sector_t sector_nr) 834 { 835 int idx = sector_to_idx(sector_nr); 836 837 spin_lock_irq(&conf->resync_lock); 838 839 /* Wait until no block IO is waiting */ 840 wait_event_lock_irq(conf->wait_barrier, 841 !atomic_read(&conf->nr_waiting[idx]), 842 conf->resync_lock); 843 844 /* block any new IO from starting */ 845 atomic_inc(&conf->barrier[idx]); 846 /* 847 * In raise_barrier() we firstly increase conf->barrier[idx] then 848 * check conf->nr_pending[idx]. In _wait_barrier() we firstly 849 * increase conf->nr_pending[idx] then check conf->barrier[idx]. 850 * A memory barrier here to make sure conf->nr_pending[idx] won't 851 * be fetched before conf->barrier[idx] is increased. Otherwise 852 * there will be a race between raise_barrier() and _wait_barrier(). 853 */ 854 smp_mb__after_atomic(); 855 856 /* For these conditions we must wait: 857 * A: while the array is in frozen state 858 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O 859 * existing in corresponding I/O barrier bucket. 860 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches 861 * max resync count which allowed on current I/O barrier bucket. 862 */ 863 wait_event_lock_irq(conf->wait_barrier, 864 !conf->array_frozen && 865 !atomic_read(&conf->nr_pending[idx]) && 866 atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH, 867 conf->resync_lock); 868 869 atomic_inc(&conf->nr_pending[idx]); 870 spin_unlock_irq(&conf->resync_lock); 871 } 872 873 static void lower_barrier(struct r1conf *conf, sector_t sector_nr) 874 { 875 int idx = sector_to_idx(sector_nr); 876 877 BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); 878 879 atomic_dec(&conf->barrier[idx]); 880 atomic_dec(&conf->nr_pending[idx]); 881 wake_up(&conf->wait_barrier); 882 } 883 884 static void _wait_barrier(struct r1conf *conf, int idx) 885 { 886 /* 887 * We need to increase conf->nr_pending[idx] very early here, 888 * then raise_barrier() can be blocked when it waits for 889 * conf->nr_pending[idx] to be 0. Then we can avoid holding 890 * conf->resync_lock when there is no barrier raised in same 891 * barrier unit bucket. Also if the array is frozen, I/O 892 * should be blocked until array is unfrozen. 893 */ 894 atomic_inc(&conf->nr_pending[idx]); 895 /* 896 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then 897 * check conf->barrier[idx]. In raise_barrier() we firstly increase 898 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory 899 * barrier is necessary here to make sure conf->barrier[idx] won't be 900 * fetched before conf->nr_pending[idx] is increased. Otherwise there 901 * will be a race between _wait_barrier() and raise_barrier(). 902 */ 903 smp_mb__after_atomic(); 904 905 /* 906 * Don't worry about checking two atomic_t variables at same time 907 * here. If during we check conf->barrier[idx], the array is 908 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is 909 * 0, it is safe to return and make the I/O continue. Because the 910 * array is frozen, all I/O returned here will eventually complete 911 * or be queued, no race will happen. See code comment in 912 * frozen_array(). 913 */ 914 if (!READ_ONCE(conf->array_frozen) && 915 !atomic_read(&conf->barrier[idx])) 916 return; 917 918 /* 919 * After holding conf->resync_lock, conf->nr_pending[idx] 920 * should be decreased before waiting for barrier to drop. 921 * Otherwise, we may encounter a race condition because 922 * raise_barrer() might be waiting for conf->nr_pending[idx] 923 * to be 0 at same time. 924 */ 925 spin_lock_irq(&conf->resync_lock); 926 atomic_inc(&conf->nr_waiting[idx]); 927 atomic_dec(&conf->nr_pending[idx]); 928 /* 929 * In case freeze_array() is waiting for 930 * get_unqueued_pending() == extra 931 */ 932 wake_up(&conf->wait_barrier); 933 /* Wait for the barrier in same barrier unit bucket to drop. */ 934 wait_event_lock_irq(conf->wait_barrier, 935 !conf->array_frozen && 936 !atomic_read(&conf->barrier[idx]), 937 conf->resync_lock); 938 atomic_inc(&conf->nr_pending[idx]); 939 atomic_dec(&conf->nr_waiting[idx]); 940 spin_unlock_irq(&conf->resync_lock); 941 } 942 943 static void wait_read_barrier(struct r1conf *conf, sector_t sector_nr) 944 { 945 int idx = sector_to_idx(sector_nr); 946 947 /* 948 * Very similar to _wait_barrier(). The difference is, for read 949 * I/O we don't need wait for sync I/O, but if the whole array 950 * is frozen, the read I/O still has to wait until the array is 951 * unfrozen. Since there is no ordering requirement with 952 * conf->barrier[idx] here, memory barrier is unnecessary as well. 953 */ 954 atomic_inc(&conf->nr_pending[idx]); 955 956 if (!READ_ONCE(conf->array_frozen)) 957 return; 958 959 spin_lock_irq(&conf->resync_lock); 960 atomic_inc(&conf->nr_waiting[idx]); 961 atomic_dec(&conf->nr_pending[idx]); 962 /* 963 * In case freeze_array() is waiting for 964 * get_unqueued_pending() == extra 965 */ 966 wake_up(&conf->wait_barrier); 967 /* Wait for array to be unfrozen */ 968 wait_event_lock_irq(conf->wait_barrier, 969 !conf->array_frozen, 970 conf->resync_lock); 971 atomic_inc(&conf->nr_pending[idx]); 972 atomic_dec(&conf->nr_waiting[idx]); 973 spin_unlock_irq(&conf->resync_lock); 974 } 975 976 static void wait_barrier(struct r1conf *conf, sector_t sector_nr) 977 { 978 int idx = sector_to_idx(sector_nr); 979 980 _wait_barrier(conf, idx); 981 } 982 983 static void wait_all_barriers(struct r1conf *conf) 984 { 985 int idx; 986 987 for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) 988 _wait_barrier(conf, idx); 989 } 990 991 static void _allow_barrier(struct r1conf *conf, int idx) 992 { 993 atomic_dec(&conf->nr_pending[idx]); 994 wake_up(&conf->wait_barrier); 995 } 996 997 static void allow_barrier(struct r1conf *conf, sector_t sector_nr) 998 { 999 int idx = sector_to_idx(sector_nr); 1000 1001 _allow_barrier(conf, idx); 1002 } 1003 1004 static void allow_all_barriers(struct r1conf *conf) 1005 { 1006 int idx; 1007 1008 for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) 1009 _allow_barrier(conf, idx); 1010 } 1011 1012 /* conf->resync_lock should be held */ 1013 static int get_unqueued_pending(struct r1conf *conf) 1014 { 1015 int idx, ret; 1016 1017 for (ret = 0, idx = 0; idx < BARRIER_BUCKETS_NR; idx++) 1018 ret += atomic_read(&conf->nr_pending[idx]) - 1019 atomic_read(&conf->nr_queued[idx]); 1020 1021 return ret; 1022 } 1023 1024 static void freeze_array(struct r1conf *conf, int extra) 1025 { 1026 /* Stop sync I/O and normal I/O and wait for everything to 1027 * go quite. 1028 * This is called in two situations: 1029 * 1) management command handlers (reshape, remove disk, quiesce). 1030 * 2) one normal I/O request failed. 1031 1032 * After array_frozen is set to 1, new sync IO will be blocked at 1033 * raise_barrier(), and new normal I/O will blocked at _wait_barrier() 1034 * or wait_read_barrier(). The flying I/Os will either complete or be 1035 * queued. When everything goes quite, there are only queued I/Os left. 1036 1037 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the 1038 * barrier bucket index which this I/O request hits. When all sync and 1039 * normal I/O are queued, sum of all conf->nr_pending[] will match sum 1040 * of all conf->nr_queued[]. But normal I/O failure is an exception, 1041 * in handle_read_error(), we may call freeze_array() before trying to 1042 * fix the read error. In this case, the error read I/O is not queued, 1043 * so get_unqueued_pending() == 1. 1044 * 1045 * Therefore before this function returns, we need to wait until 1046 * get_unqueued_pendings(conf) gets equal to extra. For 1047 * normal I/O context, extra is 1, in rested situations extra is 0. 1048 */ 1049 spin_lock_irq(&conf->resync_lock); 1050 conf->array_frozen = 1; 1051 raid1_log(conf->mddev, "wait freeze"); 1052 wait_event_lock_irq_cmd( 1053 conf->wait_barrier, 1054 get_unqueued_pending(conf) == extra, 1055 conf->resync_lock, 1056 flush_pending_writes(conf)); 1057 spin_unlock_irq(&conf->resync_lock); 1058 } 1059 static void unfreeze_array(struct r1conf *conf) 1060 { 1061 /* reverse the effect of the freeze */ 1062 spin_lock_irq(&conf->resync_lock); 1063 conf->array_frozen = 0; 1064 spin_unlock_irq(&conf->resync_lock); 1065 wake_up(&conf->wait_barrier); 1066 } 1067 1068 /* duplicate the data pages for behind I/O 1069 */ 1070 static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) 1071 { 1072 int i; 1073 struct bio_vec *bvec; 1074 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), 1075 GFP_NOIO); 1076 if (unlikely(!bvecs)) 1077 return; 1078 1079 bio_for_each_segment_all(bvec, bio, i) { 1080 bvecs[i] = *bvec; 1081 bvecs[i].bv_page = alloc_page(GFP_NOIO); 1082 if (unlikely(!bvecs[i].bv_page)) 1083 goto do_sync_io; 1084 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, 1085 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); 1086 kunmap(bvecs[i].bv_page); 1087 kunmap(bvec->bv_page); 1088 } 1089 r1_bio->behind_bvecs = bvecs; 1090 r1_bio->behind_page_count = bio->bi_vcnt; 1091 set_bit(R1BIO_BehindIO, &r1_bio->state); 1092 return; 1093 1094 do_sync_io: 1095 for (i = 0; i < bio->bi_vcnt; i++) 1096 if (bvecs[i].bv_page) 1097 put_page(bvecs[i].bv_page); 1098 kfree(bvecs); 1099 pr_debug("%dB behind alloc failed, doing sync I/O\n", 1100 bio->bi_iter.bi_size); 1101 } 1102 1103 struct raid1_plug_cb { 1104 struct blk_plug_cb cb; 1105 struct bio_list pending; 1106 int pending_cnt; 1107 }; 1108 1109 static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) 1110 { 1111 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, 1112 cb); 1113 struct mddev *mddev = plug->cb.data; 1114 struct r1conf *conf = mddev->private; 1115 struct bio *bio; 1116 1117 if (from_schedule || current->bio_list) { 1118 spin_lock_irq(&conf->device_lock); 1119 bio_list_merge(&conf->pending_bio_list, &plug->pending); 1120 conf->pending_count += plug->pending_cnt; 1121 spin_unlock_irq(&conf->device_lock); 1122 wake_up(&conf->wait_barrier); 1123 md_wakeup_thread(mddev->thread); 1124 kfree(plug); 1125 return; 1126 } 1127 1128 /* we aren't scheduling, so we can do the write-out directly. */ 1129 bio = bio_list_get(&plug->pending); 1130 bitmap_unplug(mddev->bitmap); 1131 wake_up(&conf->wait_barrier); 1132 1133 while (bio) { /* submit pending writes */ 1134 struct bio *next = bio->bi_next; 1135 struct md_rdev *rdev = (void*)bio->bi_bdev; 1136 bio->bi_next = NULL; 1137 bio->bi_bdev = rdev->bdev; 1138 if (test_bit(Faulty, &rdev->flags)) { 1139 bio->bi_error = -EIO; 1140 bio_endio(bio); 1141 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && 1142 !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) 1143 /* Just ignore it */ 1144 bio_endio(bio); 1145 else 1146 generic_make_request(bio); 1147 bio = next; 1148 } 1149 kfree(plug); 1150 } 1151 1152 static inline struct r1bio * 1153 alloc_r1bio(struct mddev *mddev, struct bio *bio, sector_t sectors_handled) 1154 { 1155 struct r1conf *conf = mddev->private; 1156 struct r1bio *r1_bio; 1157 1158 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); 1159 1160 r1_bio->master_bio = bio; 1161 r1_bio->sectors = bio_sectors(bio) - sectors_handled; 1162 r1_bio->state = 0; 1163 r1_bio->mddev = mddev; 1164 r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; 1165 1166 return r1_bio; 1167 } 1168 1169 static void raid1_read_request(struct mddev *mddev, struct bio *bio) 1170 { 1171 struct r1conf *conf = mddev->private; 1172 struct raid1_info *mirror; 1173 struct r1bio *r1_bio; 1174 struct bio *read_bio; 1175 struct bitmap *bitmap = mddev->bitmap; 1176 const int op = bio_op(bio); 1177 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); 1178 int sectors_handled; 1179 int max_sectors; 1180 int rdisk; 1181 1182 /* 1183 * Still need barrier for READ in case that whole 1184 * array is frozen. 1185 */ 1186 wait_read_barrier(conf, bio->bi_iter.bi_sector); 1187 1188 r1_bio = alloc_r1bio(mddev, bio, 0); 1189 1190 /* 1191 * We might need to issue multiple reads to different 1192 * devices if there are bad blocks around, so we keep 1193 * track of the number of reads in bio->bi_phys_segments. 1194 * If this is 0, there is only one r1_bio and no locking 1195 * will be needed when requests complete. If it is 1196 * non-zero, then it is the number of not-completed requests. 1197 */ 1198 bio->bi_phys_segments = 0; 1199 bio_clear_flag(bio, BIO_SEG_VALID); 1200 1201 /* 1202 * make_request() can abort the operation when read-ahead is being 1203 * used and no empty request is available. 1204 */ 1205 read_again: 1206 rdisk = read_balance(conf, r1_bio, &max_sectors); 1207 1208 if (rdisk < 0) { 1209 /* couldn't find anywhere to read from */ 1210 raid_end_bio_io(r1_bio); 1211 return; 1212 } 1213 mirror = conf->mirrors + rdisk; 1214 1215 if (test_bit(WriteMostly, &mirror->rdev->flags) && 1216 bitmap) { 1217 /* 1218 * Reading from a write-mostly device must take care not to 1219 * over-take any writes that are 'behind' 1220 */ 1221 raid1_log(mddev, "wait behind writes"); 1222 wait_event(bitmap->behind_wait, 1223 atomic_read(&bitmap->behind_writes) == 0); 1224 } 1225 r1_bio->read_disk = rdisk; 1226 1227 read_bio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); 1228 bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, 1229 max_sectors); 1230 1231 r1_bio->bios[rdisk] = read_bio; 1232 1233 read_bio->bi_iter.bi_sector = r1_bio->sector + 1234 mirror->rdev->data_offset; 1235 read_bio->bi_bdev = mirror->rdev->bdev; 1236 read_bio->bi_end_io = raid1_end_read_request; 1237 bio_set_op_attrs(read_bio, op, do_sync); 1238 if (test_bit(FailFast, &mirror->rdev->flags) && 1239 test_bit(R1BIO_FailFast, &r1_bio->state)) 1240 read_bio->bi_opf |= MD_FAILFAST; 1241 read_bio->bi_private = r1_bio; 1242 1243 if (mddev->gendisk) 1244 trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), 1245 read_bio, disk_devt(mddev->gendisk), 1246 r1_bio->sector); 1247 1248 if (max_sectors < r1_bio->sectors) { 1249 /* 1250 * could not read all from this device, so we will need another 1251 * r1_bio. 1252 */ 1253 sectors_handled = (r1_bio->sector + max_sectors 1254 - bio->bi_iter.bi_sector); 1255 r1_bio->sectors = max_sectors; 1256 spin_lock_irq(&conf->device_lock); 1257 if (bio->bi_phys_segments == 0) 1258 bio->bi_phys_segments = 2; 1259 else 1260 bio->bi_phys_segments++; 1261 spin_unlock_irq(&conf->device_lock); 1262 1263 /* 1264 * Cannot call generic_make_request directly as that will be 1265 * queued in __make_request and subsequent mempool_alloc might 1266 * block waiting for it. So hand bio over to raid1d. 1267 */ 1268 reschedule_retry(r1_bio); 1269 1270 r1_bio = alloc_r1bio(mddev, bio, sectors_handled); 1271 goto read_again; 1272 } else 1273 generic_make_request(read_bio); 1274 } 1275 1276 static void raid1_write_request(struct mddev *mddev, struct bio *bio) 1277 { 1278 struct r1conf *conf = mddev->private; 1279 struct r1bio *r1_bio; 1280 int i, disks; 1281 struct bitmap *bitmap = mddev->bitmap; 1282 unsigned long flags; 1283 const int op = bio_op(bio); 1284 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); 1285 const unsigned long do_fua = (bio->bi_opf & REQ_FUA); 1286 struct md_rdev *blocked_rdev; 1287 struct blk_plug_cb *cb; 1288 struct raid1_plug_cb *plug = NULL; 1289 int first_clone; 1290 int sectors_handled; 1291 int max_sectors; 1292 1293 /* 1294 * Register the new request and wait if the reconstruction 1295 * thread has put up a bar for new requests. 1296 * Continue immediately if no resync is active currently. 1297 */ 1298 1299 md_write_start(mddev, bio); /* wait on superblock update early */ 1300 1301 if ((bio_end_sector(bio) > mddev->suspend_lo && 1302 bio->bi_iter.bi_sector < mddev->suspend_hi) || 1303 (mddev_is_clustered(mddev) && 1304 md_cluster_ops->area_resyncing(mddev, WRITE, 1305 bio->bi_iter.bi_sector, bio_end_sector(bio)))) { 1306 1307 /* 1308 * As the suspend_* range is controlled by userspace, we want 1309 * an interruptible wait. 1310 */ 1311 DEFINE_WAIT(w); 1312 for (;;) { 1313 flush_signals(current); 1314 prepare_to_wait(&conf->wait_barrier, 1315 &w, TASK_INTERRUPTIBLE); 1316 if (bio_end_sector(bio) <= mddev->suspend_lo || 1317 bio->bi_iter.bi_sector >= mddev->suspend_hi || 1318 (mddev_is_clustered(mddev) && 1319 !md_cluster_ops->area_resyncing(mddev, WRITE, 1320 bio->bi_iter.bi_sector, 1321 bio_end_sector(bio)))) 1322 break; 1323 schedule(); 1324 } 1325 finish_wait(&conf->wait_barrier, &w); 1326 } 1327 wait_barrier(conf, bio->bi_iter.bi_sector); 1328 1329 r1_bio = alloc_r1bio(mddev, bio, 0); 1330 1331 /* We might need to issue multiple writes to different 1332 * devices if there are bad blocks around, so we keep 1333 * track of the number of writes in bio->bi_phys_segments. 1334 * If this is 0, there is only one r1_bio and no locking 1335 * will be needed when requests complete. If it is 1336 * non-zero, then it is the number of not-completed requests. 1337 */ 1338 bio->bi_phys_segments = 0; 1339 bio_clear_flag(bio, BIO_SEG_VALID); 1340 1341 if (conf->pending_count >= max_queued_requests) { 1342 md_wakeup_thread(mddev->thread); 1343 raid1_log(mddev, "wait queued"); 1344 wait_event(conf->wait_barrier, 1345 conf->pending_count < max_queued_requests); 1346 } 1347 /* first select target devices under rcu_lock and 1348 * inc refcount on their rdev. Record them by setting 1349 * bios[x] to bio 1350 * If there are known/acknowledged bad blocks on any device on 1351 * which we have seen a write error, we want to avoid writing those 1352 * blocks. 1353 * This potentially requires several writes to write around 1354 * the bad blocks. Each set of writes gets it's own r1bio 1355 * with a set of bios attached. 1356 */ 1357 1358 disks = conf->raid_disks * 2; 1359 retry_write: 1360 blocked_rdev = NULL; 1361 rcu_read_lock(); 1362 max_sectors = r1_bio->sectors; 1363 for (i = 0; i < disks; i++) { 1364 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); 1365 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { 1366 atomic_inc(&rdev->nr_pending); 1367 blocked_rdev = rdev; 1368 break; 1369 } 1370 r1_bio->bios[i] = NULL; 1371 if (!rdev || test_bit(Faulty, &rdev->flags)) { 1372 if (i < conf->raid_disks) 1373 set_bit(R1BIO_Degraded, &r1_bio->state); 1374 continue; 1375 } 1376 1377 atomic_inc(&rdev->nr_pending); 1378 if (test_bit(WriteErrorSeen, &rdev->flags)) { 1379 sector_t first_bad; 1380 int bad_sectors; 1381 int is_bad; 1382 1383 is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, 1384 &first_bad, &bad_sectors); 1385 if (is_bad < 0) { 1386 /* mustn't write here until the bad block is 1387 * acknowledged*/ 1388 set_bit(BlockedBadBlocks, &rdev->flags); 1389 blocked_rdev = rdev; 1390 break; 1391 } 1392 if (is_bad && first_bad <= r1_bio->sector) { 1393 /* Cannot write here at all */ 1394 bad_sectors -= (r1_bio->sector - first_bad); 1395 if (bad_sectors < max_sectors) 1396 /* mustn't write more than bad_sectors 1397 * to other devices yet 1398 */ 1399 max_sectors = bad_sectors; 1400 rdev_dec_pending(rdev, mddev); 1401 /* We don't set R1BIO_Degraded as that 1402 * only applies if the disk is 1403 * missing, so it might be re-added, 1404 * and we want to know to recover this 1405 * chunk. 1406 * In this case the device is here, 1407 * and the fact that this chunk is not 1408 * in-sync is recorded in the bad 1409 * block log 1410 */ 1411 continue; 1412 } 1413 if (is_bad) { 1414 int good_sectors = first_bad - r1_bio->sector; 1415 if (good_sectors < max_sectors) 1416 max_sectors = good_sectors; 1417 } 1418 } 1419 r1_bio->bios[i] = bio; 1420 } 1421 rcu_read_unlock(); 1422 1423 if (unlikely(blocked_rdev)) { 1424 /* Wait for this device to become unblocked */ 1425 int j; 1426 1427 for (j = 0; j < i; j++) 1428 if (r1_bio->bios[j]) 1429 rdev_dec_pending(conf->mirrors[j].rdev, mddev); 1430 r1_bio->state = 0; 1431 allow_barrier(conf, bio->bi_iter.bi_sector); 1432 raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); 1433 md_wait_for_blocked_rdev(blocked_rdev, mddev); 1434 wait_barrier(conf, bio->bi_iter.bi_sector); 1435 goto retry_write; 1436 } 1437 1438 if (max_sectors < r1_bio->sectors) { 1439 /* We are splitting this write into multiple parts, so 1440 * we need to prepare for allocating another r1_bio. 1441 */ 1442 r1_bio->sectors = max_sectors; 1443 spin_lock_irq(&conf->device_lock); 1444 if (bio->bi_phys_segments == 0) 1445 bio->bi_phys_segments = 2; 1446 else 1447 bio->bi_phys_segments++; 1448 spin_unlock_irq(&conf->device_lock); 1449 } 1450 sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; 1451 1452 atomic_set(&r1_bio->remaining, 1); 1453 atomic_set(&r1_bio->behind_remaining, 0); 1454 1455 first_clone = 1; 1456 for (i = 0; i < disks; i++) { 1457 struct bio *mbio = NULL; 1458 sector_t offset; 1459 if (!r1_bio->bios[i]) 1460 continue; 1461 1462 offset = r1_bio->sector - bio->bi_iter.bi_sector; 1463 1464 if (first_clone) { 1465 /* do behind I/O ? 1466 * Not if there are too many, or cannot 1467 * allocate memory, or a reader on WriteMostly 1468 * is waiting for behind writes to flush */ 1469 if (bitmap && 1470 (atomic_read(&bitmap->behind_writes) 1471 < mddev->bitmap_info.max_write_behind) && 1472 !waitqueue_active(&bitmap->behind_wait)) { 1473 mbio = bio_clone_bioset_partial(bio, GFP_NOIO, 1474 mddev->bio_set, 1475 offset, 1476 max_sectors); 1477 alloc_behind_pages(mbio, r1_bio); 1478 } 1479 1480 bitmap_startwrite(bitmap, r1_bio->sector, 1481 r1_bio->sectors, 1482 test_bit(R1BIO_BehindIO, 1483 &r1_bio->state)); 1484 first_clone = 0; 1485 } 1486 1487 if (!mbio) { 1488 mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); 1489 bio_trim(mbio, offset, max_sectors); 1490 } 1491 1492 if (r1_bio->behind_bvecs) { 1493 struct bio_vec *bvec; 1494 int j; 1495 1496 /* 1497 * We trimmed the bio, so _all is legit 1498 */ 1499 bio_for_each_segment_all(bvec, mbio, j) 1500 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; 1501 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) 1502 atomic_inc(&r1_bio->behind_remaining); 1503 } 1504 1505 r1_bio->bios[i] = mbio; 1506 1507 mbio->bi_iter.bi_sector = (r1_bio->sector + 1508 conf->mirrors[i].rdev->data_offset); 1509 mbio->bi_bdev = conf->mirrors[i].rdev->bdev; 1510 mbio->bi_end_io = raid1_end_write_request; 1511 bio_set_op_attrs(mbio, op, do_fua | do_sync); 1512 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && 1513 !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && 1514 conf->raid_disks - mddev->degraded > 1) 1515 mbio->bi_opf |= MD_FAILFAST; 1516 mbio->bi_private = r1_bio; 1517 1518 atomic_inc(&r1_bio->remaining); 1519 1520 if (mddev->gendisk) 1521 trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), 1522 mbio, disk_devt(mddev->gendisk), 1523 r1_bio->sector); 1524 /* flush_pending_writes() needs access to the rdev so...*/ 1525 mbio->bi_bdev = (void*)conf->mirrors[i].rdev; 1526 1527 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); 1528 if (cb) 1529 plug = container_of(cb, struct raid1_plug_cb, cb); 1530 else 1531 plug = NULL; 1532 spin_lock_irqsave(&conf->device_lock, flags); 1533 if (plug) { 1534 bio_list_add(&plug->pending, mbio); 1535 plug->pending_cnt++; 1536 } else { 1537 bio_list_add(&conf->pending_bio_list, mbio); 1538 conf->pending_count++; 1539 } 1540 spin_unlock_irqrestore(&conf->device_lock, flags); 1541 if (!plug) 1542 md_wakeup_thread(mddev->thread); 1543 } 1544 /* Mustn't call r1_bio_write_done before this next test, 1545 * as it could result in the bio being freed. 1546 */ 1547 if (sectors_handled < bio_sectors(bio)) { 1548 r1_bio_write_done(r1_bio); 1549 /* We need another r1_bio. It has already been counted 1550 * in bio->bi_phys_segments 1551 */ 1552 r1_bio = alloc_r1bio(mddev, bio, sectors_handled); 1553 goto retry_write; 1554 } 1555 1556 r1_bio_write_done(r1_bio); 1557 1558 /* In case raid1d snuck in to freeze_array */ 1559 wake_up(&conf->wait_barrier); 1560 } 1561 1562 static void raid1_make_request(struct mddev *mddev, struct bio *bio) 1563 { 1564 struct bio *split; 1565 sector_t sectors; 1566 1567 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { 1568 md_flush_request(mddev, bio); 1569 return; 1570 } 1571 1572 /* if bio exceeds barrier unit boundary, split it */ 1573 do { 1574 sectors = align_to_barrier_unit_end( 1575 bio->bi_iter.bi_sector, bio_sectors(bio)); 1576 if (sectors < bio_sectors(bio)) { 1577 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set); 1578 bio_chain(split, bio); 1579 } else { 1580 split = bio; 1581 } 1582 1583 if (bio_data_dir(split) == READ) 1584 raid1_read_request(mddev, split); 1585 else 1586 raid1_write_request(mddev, split); 1587 } while (split != bio); 1588 } 1589 1590 static void raid1_status(struct seq_file *seq, struct mddev *mddev) 1591 { 1592 struct r1conf *conf = mddev->private; 1593 int i; 1594 1595 seq_printf(seq, " [%d/%d] [", conf->raid_disks, 1596 conf->raid_disks - mddev->degraded); 1597 rcu_read_lock(); 1598 for (i = 0; i < conf->raid_disks; i++) { 1599 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); 1600 seq_printf(seq, "%s", 1601 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); 1602 } 1603 rcu_read_unlock(); 1604 seq_printf(seq, "]"); 1605 } 1606 1607 static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) 1608 { 1609 char b[BDEVNAME_SIZE]; 1610 struct r1conf *conf = mddev->private; 1611 unsigned long flags; 1612 1613 /* 1614 * If it is not operational, then we have already marked it as dead 1615 * else if it is the last working disks, ignore the error, let the 1616 * next level up know. 1617 * else mark the drive as failed 1618 */ 1619 spin_lock_irqsave(&conf->device_lock, flags); 1620 if (test_bit(In_sync, &rdev->flags) 1621 && (conf->raid_disks - mddev->degraded) == 1) { 1622 /* 1623 * Don't fail the drive, act as though we were just a 1624 * normal single drive. 1625 * However don't try a recovery from this drive as 1626 * it is very likely to fail. 1627 */ 1628 conf->recovery_disabled = mddev->recovery_disabled; 1629 spin_unlock_irqrestore(&conf->device_lock, flags); 1630 return; 1631 } 1632 set_bit(Blocked, &rdev->flags); 1633 if (test_and_clear_bit(In_sync, &rdev->flags)) { 1634 mddev->degraded++; 1635 set_bit(Faulty, &rdev->flags); 1636 } else 1637 set_bit(Faulty, &rdev->flags); 1638 spin_unlock_irqrestore(&conf->device_lock, flags); 1639 /* 1640 * if recovery is running, make sure it aborts. 1641 */ 1642 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 1643 set_mask_bits(&mddev->sb_flags, 0, 1644 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); 1645 pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" 1646 "md/raid1:%s: Operation continuing on %d devices.\n", 1647 mdname(mddev), bdevname(rdev->bdev, b), 1648 mdname(mddev), conf->raid_disks - mddev->degraded); 1649 } 1650 1651 static void print_conf(struct r1conf *conf) 1652 { 1653 int i; 1654 1655 pr_debug("RAID1 conf printout:\n"); 1656 if (!conf) { 1657 pr_debug("(!conf)\n"); 1658 return; 1659 } 1660 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, 1661 conf->raid_disks); 1662 1663 rcu_read_lock(); 1664 for (i = 0; i < conf->raid_disks; i++) { 1665 char b[BDEVNAME_SIZE]; 1666 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); 1667 if (rdev) 1668 pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", 1669 i, !test_bit(In_sync, &rdev->flags), 1670 !test_bit(Faulty, &rdev->flags), 1671 bdevname(rdev->bdev,b)); 1672 } 1673 rcu_read_unlock(); 1674 } 1675 1676 static void close_sync(struct r1conf *conf) 1677 { 1678 wait_all_barriers(conf); 1679 allow_all_barriers(conf); 1680 1681 mempool_destroy(conf->r1buf_pool); 1682 conf->r1buf_pool = NULL; 1683 } 1684 1685 static int raid1_spare_active(struct mddev *mddev) 1686 { 1687 int i; 1688 struct r1conf *conf = mddev->private; 1689 int count = 0; 1690 unsigned long flags; 1691 1692 /* 1693 * Find all failed disks within the RAID1 configuration 1694 * and mark them readable. 1695 * Called under mddev lock, so rcu protection not needed. 1696 * device_lock used to avoid races with raid1_end_read_request 1697 * which expects 'In_sync' flags and ->degraded to be consistent. 1698 */ 1699 spin_lock_irqsave(&conf->device_lock, flags); 1700 for (i = 0; i < conf->raid_disks; i++) { 1701 struct md_rdev *rdev = conf->mirrors[i].rdev; 1702 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; 1703 if (repl 1704 && !test_bit(Candidate, &repl->flags) 1705 && repl->recovery_offset == MaxSector 1706 && !test_bit(Faulty, &repl->flags) 1707 && !test_and_set_bit(In_sync, &repl->flags)) { 1708 /* replacement has just become active */ 1709 if (!rdev || 1710 !test_and_clear_bit(In_sync, &rdev->flags)) 1711 count++; 1712 if (rdev) { 1713 /* Replaced device not technically 1714 * faulty, but we need to be sure 1715 * it gets removed and never re-added 1716 */ 1717 set_bit(Faulty, &rdev->flags); 1718 sysfs_notify_dirent_safe( 1719 rdev->sysfs_state); 1720 } 1721 } 1722 if (rdev 1723 && rdev->recovery_offset == MaxSector 1724 && !test_bit(Faulty, &rdev->flags) 1725 && !test_and_set_bit(In_sync, &rdev->flags)) { 1726 count++; 1727 sysfs_notify_dirent_safe(rdev->sysfs_state); 1728 } 1729 } 1730 mddev->degraded -= count; 1731 spin_unlock_irqrestore(&conf->device_lock, flags); 1732 1733 print_conf(conf); 1734 return count; 1735 } 1736 1737 static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) 1738 { 1739 struct r1conf *conf = mddev->private; 1740 int err = -EEXIST; 1741 int mirror = 0; 1742 struct raid1_info *p; 1743 int first = 0; 1744 int last = conf->raid_disks - 1; 1745 1746 if (mddev->recovery_disabled == conf->recovery_disabled) 1747 return -EBUSY; 1748 1749 if (md_integrity_add_rdev(rdev, mddev)) 1750 return -ENXIO; 1751 1752 if (rdev->raid_disk >= 0) 1753 first = last = rdev->raid_disk; 1754 1755 /* 1756 * find the disk ... but prefer rdev->saved_raid_disk 1757 * if possible. 1758 */ 1759 if (rdev->saved_raid_disk >= 0 && 1760 rdev->saved_raid_disk >= first && 1761 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) 1762 first = last = rdev->saved_raid_disk; 1763 1764 for (mirror = first; mirror <= last; mirror++) { 1765 p = conf->mirrors+mirror; 1766 if (!p->rdev) { 1767 1768 if (mddev->gendisk) 1769 disk_stack_limits(mddev->gendisk, rdev->bdev, 1770 rdev->data_offset << 9); 1771 1772 p->head_position = 0; 1773 rdev->raid_disk = mirror; 1774 err = 0; 1775 /* As all devices are equivalent, we don't need a full recovery 1776 * if this was recently any drive of the array 1777 */ 1778 if (rdev->saved_raid_disk < 0) 1779 conf->fullsync = 1; 1780 rcu_assign_pointer(p->rdev, rdev); 1781 break; 1782 } 1783 if (test_bit(WantReplacement, &p->rdev->flags) && 1784 p[conf->raid_disks].rdev == NULL) { 1785 /* Add this device as a replacement */ 1786 clear_bit(In_sync, &rdev->flags); 1787 set_bit(Replacement, &rdev->flags); 1788 rdev->raid_disk = mirror; 1789 err = 0; 1790 conf->fullsync = 1; 1791 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); 1792 break; 1793 } 1794 } 1795 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) 1796 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); 1797 print_conf(conf); 1798 return err; 1799 } 1800 1801 static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) 1802 { 1803 struct r1conf *conf = mddev->private; 1804 int err = 0; 1805 int number = rdev->raid_disk; 1806 struct raid1_info *p = conf->mirrors + number; 1807 1808 if (rdev != p->rdev) 1809 p = conf->mirrors + conf->raid_disks + number; 1810 1811 print_conf(conf); 1812 if (rdev == p->rdev) { 1813 if (test_bit(In_sync, &rdev->flags) || 1814 atomic_read(&rdev->nr_pending)) { 1815 err = -EBUSY; 1816 goto abort; 1817 } 1818 /* Only remove non-faulty devices if recovery 1819 * is not possible. 1820 */ 1821 if (!test_bit(Faulty, &rdev->flags) && 1822 mddev->recovery_disabled != conf->recovery_disabled && 1823 mddev->degraded < conf->raid_disks) { 1824 err = -EBUSY; 1825 goto abort; 1826 } 1827 p->rdev = NULL; 1828 if (!test_bit(RemoveSynchronized, &rdev->flags)) { 1829 synchronize_rcu(); 1830 if (atomic_read(&rdev->nr_pending)) { 1831 /* lost the race, try later */ 1832 err = -EBUSY; 1833 p->rdev = rdev; 1834 goto abort; 1835 } 1836 } 1837 if (conf->mirrors[conf->raid_disks + number].rdev) { 1838 /* We just removed a device that is being replaced. 1839 * Move down the replacement. We drain all IO before 1840 * doing this to avoid confusion. 1841 */ 1842 struct md_rdev *repl = 1843 conf->mirrors[conf->raid_disks + number].rdev; 1844 freeze_array(conf, 0); 1845 clear_bit(Replacement, &repl->flags); 1846 p->rdev = repl; 1847 conf->mirrors[conf->raid_disks + number].rdev = NULL; 1848 unfreeze_array(conf); 1849 clear_bit(WantReplacement, &rdev->flags); 1850 } else 1851 clear_bit(WantReplacement, &rdev->flags); 1852 err = md_integrity_register(mddev); 1853 } 1854 abort: 1855 1856 print_conf(conf); 1857 return err; 1858 } 1859 1860 static void end_sync_read(struct bio *bio) 1861 { 1862 struct r1bio *r1_bio = bio->bi_private; 1863 1864 update_head_pos(r1_bio->read_disk, r1_bio); 1865 1866 /* 1867 * we have read a block, now it needs to be re-written, 1868 * or re-read if the read failed. 1869 * We don't do much here, just schedule handling by raid1d 1870 */ 1871 if (!bio->bi_error) 1872 set_bit(R1BIO_Uptodate, &r1_bio->state); 1873 1874 if (atomic_dec_and_test(&r1_bio->remaining)) 1875 reschedule_retry(r1_bio); 1876 } 1877 1878 static void end_sync_write(struct bio *bio) 1879 { 1880 int uptodate = !bio->bi_error; 1881 struct r1bio *r1_bio = bio->bi_private; 1882 struct mddev *mddev = r1_bio->mddev; 1883 struct r1conf *conf = mddev->private; 1884 sector_t first_bad; 1885 int bad_sectors; 1886 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; 1887 1888 if (!uptodate) { 1889 sector_t sync_blocks = 0; 1890 sector_t s = r1_bio->sector; 1891 long sectors_to_go = r1_bio->sectors; 1892 /* make sure these bits doesn't get cleared. */ 1893 do { 1894 bitmap_end_sync(mddev->bitmap, s, 1895 &sync_blocks, 1); 1896 s += sync_blocks; 1897 sectors_to_go -= sync_blocks; 1898 } while (sectors_to_go > 0); 1899 set_bit(WriteErrorSeen, &rdev->flags); 1900 if (!test_and_set_bit(WantReplacement, &rdev->flags)) 1901 set_bit(MD_RECOVERY_NEEDED, & 1902 mddev->recovery); 1903 set_bit(R1BIO_WriteError, &r1_bio->state); 1904 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, 1905 &first_bad, &bad_sectors) && 1906 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, 1907 r1_bio->sector, 1908 r1_bio->sectors, 1909 &first_bad, &bad_sectors) 1910 ) 1911 set_bit(R1BIO_MadeGood, &r1_bio->state); 1912 1913 if (atomic_dec_and_test(&r1_bio->remaining)) { 1914 int s = r1_bio->sectors; 1915 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || 1916 test_bit(R1BIO_WriteError, &r1_bio->state)) 1917 reschedule_retry(r1_bio); 1918 else { 1919 put_buf(r1_bio); 1920 md_done_sync(mddev, s, uptodate); 1921 } 1922 } 1923 } 1924 1925 static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, 1926 int sectors, struct page *page, int rw) 1927 { 1928 if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) 1929 /* success */ 1930 return 1; 1931 if (rw == WRITE) { 1932 set_bit(WriteErrorSeen, &rdev->flags); 1933 if (!test_and_set_bit(WantReplacement, 1934 &rdev->flags)) 1935 set_bit(MD_RECOVERY_NEEDED, & 1936 rdev->mddev->recovery); 1937 } 1938 /* need to record an error - either for the block or the device */ 1939 if (!rdev_set_badblocks(rdev, sector, sectors, 0)) 1940 md_error(rdev->mddev, rdev); 1941 return 0; 1942 } 1943 1944 static int fix_sync_read_error(struct r1bio *r1_bio) 1945 { 1946 /* Try some synchronous reads of other devices to get 1947 * good data, much like with normal read errors. Only 1948 * read into the pages we already have so we don't 1949 * need to re-issue the read request. 1950 * We don't need to freeze the array, because being in an 1951 * active sync request, there is no normal IO, and 1952 * no overlapping syncs. 1953 * We don't need to check is_badblock() again as we 1954 * made sure that anything with a bad block in range 1955 * will have bi_end_io clear. 1956 */ 1957 struct mddev *mddev = r1_bio->mddev; 1958 struct r1conf *conf = mddev->private; 1959 struct bio *bio = r1_bio->bios[r1_bio->read_disk]; 1960 sector_t sect = r1_bio->sector; 1961 int sectors = r1_bio->sectors; 1962 int idx = 0; 1963 struct md_rdev *rdev; 1964 1965 rdev = conf->mirrors[r1_bio->read_disk].rdev; 1966 if (test_bit(FailFast, &rdev->flags)) { 1967 /* Don't try recovering from here - just fail it 1968 * ... unless it is the last working device of course */ 1969 md_error(mddev, rdev); 1970 if (test_bit(Faulty, &rdev->flags)) 1971 /* Don't try to read from here, but make sure 1972 * put_buf does it's thing 1973 */ 1974 bio->bi_end_io = end_sync_write; 1975 } 1976 1977 while(sectors) { 1978 int s = sectors; 1979 int d = r1_bio->read_disk; 1980 int success = 0; 1981 int start; 1982 1983 if (s > (PAGE_SIZE>>9)) 1984 s = PAGE_SIZE >> 9; 1985 do { 1986 if (r1_bio->bios[d]->bi_end_io == end_sync_read) { 1987 /* No rcu protection needed here devices 1988 * can only be removed when no resync is 1989 * active, and resync is currently active 1990 */ 1991 rdev = conf->mirrors[d].rdev; 1992 if (sync_page_io(rdev, sect, s<<9, 1993 bio->bi_io_vec[idx].bv_page, 1994 REQ_OP_READ, 0, false)) { 1995 success = 1; 1996 break; 1997 } 1998 } 1999 d++; 2000 if (d == conf->raid_disks * 2) 2001 d = 0; 2002 } while (!success && d != r1_bio->read_disk); 2003 2004 if (!success) { 2005 char b[BDEVNAME_SIZE]; 2006 int abort = 0; 2007 /* Cannot read from anywhere, this block is lost. 2008 * Record a bad block on each device. If that doesn't 2009 * work just disable and interrupt the recovery. 2010 * Don't fail devices as that won't really help. 2011 */ 2012 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", 2013 mdname(mddev), 2014 bdevname(bio->bi_bdev, b), 2015 (unsigned long long)r1_bio->sector); 2016 for (d = 0; d < conf->raid_disks * 2; d++) { 2017 rdev = conf->mirrors[d].rdev; 2018 if (!rdev || test_bit(Faulty, &rdev->flags)) 2019 continue; 2020 if (!rdev_set_badblocks(rdev, sect, s, 0)) 2021 abort = 1; 2022 } 2023 if (abort) { 2024 conf->recovery_disabled = 2025 mddev->recovery_disabled; 2026 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 2027 md_done_sync(mddev, r1_bio->sectors, 0); 2028 put_buf(r1_bio); 2029 return 0; 2030 } 2031 /* Try next page */ 2032 sectors -= s; 2033 sect += s; 2034 idx++; 2035 continue; 2036 } 2037 2038 start = d; 2039 /* write it back and re-read */ 2040 while (d != r1_bio->read_disk) { 2041 if (d == 0) 2042 d = conf->raid_disks * 2; 2043 d--; 2044 if (r1_bio->bios[d]->bi_end_io != end_sync_read) 2045 continue; 2046 rdev = conf->mirrors[d].rdev; 2047 if (r1_sync_page_io(rdev, sect, s, 2048 bio->bi_io_vec[idx].bv_page, 2049 WRITE) == 0) { 2050 r1_bio->bios[d]->bi_end_io = NULL; 2051 rdev_dec_pending(rdev, mddev); 2052 } 2053 } 2054 d = start; 2055 while (d != r1_bio->read_disk) { 2056 if (d == 0) 2057 d = conf->raid_disks * 2; 2058 d--; 2059 if (r1_bio->bios[d]->bi_end_io != end_sync_read) 2060 continue; 2061 rdev = conf->mirrors[d].rdev; 2062 if (r1_sync_page_io(rdev, sect, s, 2063 bio->bi_io_vec[idx].bv_page, 2064 READ) != 0) 2065 atomic_add(s, &rdev->corrected_errors); 2066 } 2067 sectors -= s; 2068 sect += s; 2069 idx ++; 2070 } 2071 set_bit(R1BIO_Uptodate, &r1_bio->state); 2072 bio->bi_error = 0; 2073 return 1; 2074 } 2075 2076 static void process_checks(struct r1bio *r1_bio) 2077 { 2078 /* We have read all readable devices. If we haven't 2079 * got the block, then there is no hope left. 2080 * If we have, then we want to do a comparison 2081 * and skip the write if everything is the same. 2082 * If any blocks failed to read, then we need to 2083 * attempt an over-write 2084 */ 2085 struct mddev *mddev = r1_bio->mddev; 2086 struct r1conf *conf = mddev->private; 2087 int primary; 2088 int i; 2089 int vcnt; 2090 2091 /* Fix variable parts of all bios */ 2092 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); 2093 for (i = 0; i < conf->raid_disks * 2; i++) { 2094 int j; 2095 int size; 2096 int error; 2097 struct bio *b = r1_bio->bios[i]; 2098 if (b->bi_end_io != end_sync_read) 2099 continue; 2100 /* fixup the bio for reuse, but preserve errno */ 2101 error = b->bi_error; 2102 bio_reset(b); 2103 b->bi_error = error; 2104 b->bi_vcnt = vcnt; 2105 b->bi_iter.bi_size = r1_bio->sectors << 9; 2106 b->bi_iter.bi_sector = r1_bio->sector + 2107 conf->mirrors[i].rdev->data_offset; 2108 b->bi_bdev = conf->mirrors[i].rdev->bdev; 2109 b->bi_end_io = end_sync_read; 2110 b->bi_private = r1_bio; 2111 2112 size = b->bi_iter.bi_size; 2113 for (j = 0; j < vcnt ; j++) { 2114 struct bio_vec *bi; 2115 bi = &b->bi_io_vec[j]; 2116 bi->bv_offset = 0; 2117 if (size > PAGE_SIZE) 2118 bi->bv_len = PAGE_SIZE; 2119 else 2120 bi->bv_len = size; 2121 size -= PAGE_SIZE; 2122 } 2123 } 2124 for (primary = 0; primary < conf->raid_disks * 2; primary++) 2125 if (r1_bio->bios[primary]->bi_end_io == end_sync_read && 2126 !r1_bio->bios[primary]->bi_error) { 2127 r1_bio->bios[primary]->bi_end_io = NULL; 2128 rdev_dec_pending(conf->mirrors[primary].rdev, mddev); 2129 break; 2130 } 2131 r1_bio->read_disk = primary; 2132 for (i = 0; i < conf->raid_disks * 2; i++) { 2133 int j; 2134 struct bio *pbio = r1_bio->bios[primary]; 2135 struct bio *sbio = r1_bio->bios[i]; 2136 int error = sbio->bi_error; 2137 2138 if (sbio->bi_end_io != end_sync_read) 2139 continue; 2140 /* Now we can 'fixup' the error value */ 2141 sbio->bi_error = 0; 2142 2143 if (!error) { 2144 for (j = vcnt; j-- ; ) { 2145 struct page *p, *s; 2146 p = pbio->bi_io_vec[j].bv_page; 2147 s = sbio->bi_io_vec[j].bv_page; 2148 if (memcmp(page_address(p), 2149 page_address(s), 2150 sbio->bi_io_vec[j].bv_len)) 2151 break; 2152 } 2153 } else 2154 j = 0; 2155 if (j >= 0) 2156 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); 2157 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) 2158 && !error)) { 2159 /* No need to write to this device. */ 2160 sbio->bi_end_io = NULL; 2161 rdev_dec_pending(conf->mirrors[i].rdev, mddev); 2162 continue; 2163 } 2164 2165 bio_copy_data(sbio, pbio); 2166 } 2167 } 2168 2169 static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) 2170 { 2171 struct r1conf *conf = mddev->private; 2172 int i; 2173 int disks = conf->raid_disks * 2; 2174 struct bio *bio, *wbio; 2175 2176 bio = r1_bio->bios[r1_bio->read_disk]; 2177 2178 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) 2179 /* ouch - failed to read all of that. */ 2180 if (!fix_sync_read_error(r1_bio)) 2181 return; 2182 2183 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 2184 process_checks(r1_bio); 2185 2186 /* 2187 * schedule writes 2188 */ 2189 atomic_set(&r1_bio->remaining, 1); 2190 for (i = 0; i < disks ; i++) { 2191 wbio = r1_bio->bios[i]; 2192 if (wbio->bi_end_io == NULL || 2193 (wbio->bi_end_io == end_sync_read && 2194 (i == r1_bio->read_disk || 2195 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) 2196 continue; 2197 2198 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); 2199 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) 2200 wbio->bi_opf |= MD_FAILFAST; 2201 2202 wbio->bi_end_io = end_sync_write; 2203 atomic_inc(&r1_bio->remaining); 2204 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); 2205 2206 generic_make_request(wbio); 2207 } 2208 2209 if (atomic_dec_and_test(&r1_bio->remaining)) { 2210 /* if we're here, all write(s) have completed, so clean up */ 2211 int s = r1_bio->sectors; 2212 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || 2213 test_bit(R1BIO_WriteError, &r1_bio->state)) 2214 reschedule_retry(r1_bio); 2215 else { 2216 put_buf(r1_bio); 2217 md_done_sync(mddev, s, 1); 2218 } 2219 } 2220 } 2221 2222 /* 2223 * This is a kernel thread which: 2224 * 2225 * 1. Retries failed read operations on working mirrors. 2226 * 2. Updates the raid superblock when problems encounter. 2227 * 3. Performs writes following reads for array synchronising. 2228 */ 2229 2230 static void fix_read_error(struct r1conf *conf, int read_disk, 2231 sector_t sect, int sectors) 2232 { 2233 struct mddev *mddev = conf->mddev; 2234 while(sectors) { 2235 int s = sectors; 2236 int d = read_disk; 2237 int success = 0; 2238 int start; 2239 struct md_rdev *rdev; 2240 2241 if (s > (PAGE_SIZE>>9)) 2242 s = PAGE_SIZE >> 9; 2243 2244 do { 2245 sector_t first_bad; 2246 int bad_sectors; 2247 2248 rcu_read_lock(); 2249 rdev = rcu_dereference(conf->mirrors[d].rdev); 2250 if (rdev && 2251 (test_bit(In_sync, &rdev->flags) || 2252 (!test_bit(Faulty, &rdev->flags) && 2253 rdev->recovery_offset >= sect + s)) && 2254 is_badblock(rdev, sect, s, 2255 &first_bad, &bad_sectors) == 0) { 2256 atomic_inc(&rdev->nr_pending); 2257 rcu_read_unlock(); 2258 if (sync_page_io(rdev, sect, s<<9, 2259 conf->tmppage, REQ_OP_READ, 0, false)) 2260 success = 1; 2261 rdev_dec_pending(rdev, mddev); 2262 if (success) 2263 break; 2264 } else 2265 rcu_read_unlock(); 2266 d++; 2267 if (d == conf->raid_disks * 2) 2268 d = 0; 2269 } while (!success && d != read_disk); 2270 2271 if (!success) { 2272 /* Cannot read from anywhere - mark it bad */ 2273 struct md_rdev *rdev = conf->mirrors[read_disk].rdev; 2274 if (!rdev_set_badblocks(rdev, sect, s, 0)) 2275 md_error(mddev, rdev); 2276 break; 2277 } 2278 /* write it back and re-read */ 2279 start = d; 2280 while (d != read_disk) { 2281 if (d==0) 2282 d = conf->raid_disks * 2; 2283 d--; 2284 rcu_read_lock(); 2285 rdev = rcu_dereference(conf->mirrors[d].rdev); 2286 if (rdev && 2287 !test_bit(Faulty, &rdev->flags)) { 2288 atomic_inc(&rdev->nr_pending); 2289 rcu_read_unlock(); 2290 r1_sync_page_io(rdev, sect, s, 2291 conf->tmppage, WRITE); 2292 rdev_dec_pending(rdev, mddev); 2293 } else 2294 rcu_read_unlock(); 2295 } 2296 d = start; 2297 while (d != read_disk) { 2298 char b[BDEVNAME_SIZE]; 2299 if (d==0) 2300 d = conf->raid_disks * 2; 2301 d--; 2302 rcu_read_lock(); 2303 rdev = rcu_dereference(conf->mirrors[d].rdev); 2304 if (rdev && 2305 !test_bit(Faulty, &rdev->flags)) { 2306 atomic_inc(&rdev->nr_pending); 2307 rcu_read_unlock(); 2308 if (r1_sync_page_io(rdev, sect, s, 2309 conf->tmppage, READ)) { 2310 atomic_add(s, &rdev->corrected_errors); 2311 pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", 2312 mdname(mddev), s, 2313 (unsigned long long)(sect + 2314 rdev->data_offset), 2315 bdevname(rdev->bdev, b)); 2316 } 2317 rdev_dec_pending(rdev, mddev); 2318 } else 2319 rcu_read_unlock(); 2320 } 2321 sectors -= s; 2322 sect += s; 2323 } 2324 } 2325 2326 static int narrow_write_error(struct r1bio *r1_bio, int i) 2327 { 2328 struct mddev *mddev = r1_bio->mddev; 2329 struct r1conf *conf = mddev->private; 2330 struct md_rdev *rdev = conf->mirrors[i].rdev; 2331 2332 /* bio has the data to be written to device 'i' where 2333 * we just recently had a write error. 2334 * We repeatedly clone the bio and trim down to one block, 2335 * then try the write. Where the write fails we record 2336 * a bad block. 2337 * It is conceivable that the bio doesn't exactly align with 2338 * blocks. We must handle this somehow. 2339 * 2340 * We currently own a reference on the rdev. 2341 */ 2342 2343 int block_sectors; 2344 sector_t sector; 2345 int sectors; 2346 int sect_to_write = r1_bio->sectors; 2347 int ok = 1; 2348 2349 if (rdev->badblocks.shift < 0) 2350 return 0; 2351 2352 block_sectors = roundup(1 << rdev->badblocks.shift, 2353 bdev_logical_block_size(rdev->bdev) >> 9); 2354 sector = r1_bio->sector; 2355 sectors = ((sector + block_sectors) 2356 & ~(sector_t)(block_sectors - 1)) 2357 - sector; 2358 2359 while (sect_to_write) { 2360 struct bio *wbio; 2361 if (sectors > sect_to_write) 2362 sectors = sect_to_write; 2363 /* Write at 'sector' for 'sectors'*/ 2364 2365 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { 2366 unsigned vcnt = r1_bio->behind_page_count; 2367 struct bio_vec *vec = r1_bio->behind_bvecs; 2368 2369 while (!vec->bv_page) { 2370 vec++; 2371 vcnt--; 2372 } 2373 2374 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); 2375 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); 2376 2377 wbio->bi_vcnt = vcnt; 2378 } else { 2379 wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, 2380 mddev->bio_set); 2381 } 2382 2383 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); 2384 wbio->bi_iter.bi_sector = r1_bio->sector; 2385 wbio->bi_iter.bi_size = r1_bio->sectors << 9; 2386 2387 bio_trim(wbio, sector - r1_bio->sector, sectors); 2388 wbio->bi_iter.bi_sector += rdev->data_offset; 2389 wbio->bi_bdev = rdev->bdev; 2390 2391 if (submit_bio_wait(wbio) < 0) 2392 /* failure! */ 2393 ok = rdev_set_badblocks(rdev, sector, 2394 sectors, 0) 2395 && ok; 2396 2397 bio_put(wbio); 2398 sect_to_write -= sectors; 2399 sector += sectors; 2400 sectors = block_sectors; 2401 } 2402 return ok; 2403 } 2404 2405 static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) 2406 { 2407 int m; 2408 int s = r1_bio->sectors; 2409 for (m = 0; m < conf->raid_disks * 2 ; m++) { 2410 struct md_rdev *rdev = conf->mirrors[m].rdev; 2411 struct bio *bio = r1_bio->bios[m]; 2412 if (bio->bi_end_io == NULL) 2413 continue; 2414 if (!bio->bi_error && 2415 test_bit(R1BIO_MadeGood, &r1_bio->state)) { 2416 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); 2417 } 2418 if (bio->bi_error && 2419 test_bit(R1BIO_WriteError, &r1_bio->state)) { 2420 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) 2421 md_error(conf->mddev, rdev); 2422 } 2423 } 2424 put_buf(r1_bio); 2425 md_done_sync(conf->mddev, s, 1); 2426 } 2427 2428 static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) 2429 { 2430 int m, idx; 2431 bool fail = false; 2432 2433 for (m = 0; m < conf->raid_disks * 2 ; m++) 2434 if (r1_bio->bios[m] == IO_MADE_GOOD) { 2435 struct md_rdev *rdev = conf->mirrors[m].rdev; 2436 rdev_clear_badblocks(rdev, 2437 r1_bio->sector, 2438 r1_bio->sectors, 0); 2439 rdev_dec_pending(rdev, conf->mddev); 2440 } else if (r1_bio->bios[m] != NULL) { 2441 /* This drive got a write error. We need to 2442 * narrow down and record precise write 2443 * errors. 2444 */ 2445 fail = true; 2446 if (!narrow_write_error(r1_bio, m)) { 2447 md_error(conf->mddev, 2448 conf->mirrors[m].rdev); 2449 /* an I/O failed, we can't clear the bitmap */ 2450 set_bit(R1BIO_Degraded, &r1_bio->state); 2451 } 2452 rdev_dec_pending(conf->mirrors[m].rdev, 2453 conf->mddev); 2454 } 2455 if (fail) { 2456 spin_lock_irq(&conf->device_lock); 2457 list_add(&r1_bio->retry_list, &conf->bio_end_io_list); 2458 idx = sector_to_idx(r1_bio->sector); 2459 atomic_inc(&conf->nr_queued[idx]); 2460 spin_unlock_irq(&conf->device_lock); 2461 /* 2462 * In case freeze_array() is waiting for condition 2463 * get_unqueued_pending() == extra to be true. 2464 */ 2465 wake_up(&conf->wait_barrier); 2466 md_wakeup_thread(conf->mddev->thread); 2467 } else { 2468 if (test_bit(R1BIO_WriteError, &r1_bio->state)) 2469 close_write(r1_bio); 2470 raid_end_bio_io(r1_bio); 2471 } 2472 } 2473 2474 static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) 2475 { 2476 int disk; 2477 int max_sectors; 2478 struct mddev *mddev = conf->mddev; 2479 struct bio *bio; 2480 char b[BDEVNAME_SIZE]; 2481 struct md_rdev *rdev; 2482 dev_t bio_dev; 2483 sector_t bio_sector; 2484 2485 clear_bit(R1BIO_ReadError, &r1_bio->state); 2486 /* we got a read error. Maybe the drive is bad. Maybe just 2487 * the block and we can fix it. 2488 * We freeze all other IO, and try reading the block from 2489 * other devices. When we find one, we re-write 2490 * and check it that fixes the read error. 2491 * This is all done synchronously while the array is 2492 * frozen 2493 */ 2494 2495 bio = r1_bio->bios[r1_bio->read_disk]; 2496 bdevname(bio->bi_bdev, b); 2497 bio_dev = bio->bi_bdev->bd_dev; 2498 bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector; 2499 bio_put(bio); 2500 r1_bio->bios[r1_bio->read_disk] = NULL; 2501 2502 rdev = conf->mirrors[r1_bio->read_disk].rdev; 2503 if (mddev->ro == 0 2504 && !test_bit(FailFast, &rdev->flags)) { 2505 freeze_array(conf, 1); 2506 fix_read_error(conf, r1_bio->read_disk, 2507 r1_bio->sector, r1_bio->sectors); 2508 unfreeze_array(conf); 2509 } else { 2510 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; 2511 } 2512 2513 rdev_dec_pending(rdev, conf->mddev); 2514 2515 read_more: 2516 disk = read_balance(conf, r1_bio, &max_sectors); 2517 if (disk == -1) { 2518 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", 2519 mdname(mddev), b, (unsigned long long)r1_bio->sector); 2520 raid_end_bio_io(r1_bio); 2521 } else { 2522 const unsigned long do_sync 2523 = r1_bio->master_bio->bi_opf & REQ_SYNC; 2524 r1_bio->read_disk = disk; 2525 bio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, 2526 mddev->bio_set); 2527 bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, 2528 max_sectors); 2529 r1_bio->bios[r1_bio->read_disk] = bio; 2530 rdev = conf->mirrors[disk].rdev; 2531 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", 2532 mdname(mddev), 2533 (unsigned long long)r1_bio->sector, 2534 bdevname(rdev->bdev, b)); 2535 bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; 2536 bio->bi_bdev = rdev->bdev; 2537 bio->bi_end_io = raid1_end_read_request; 2538 bio_set_op_attrs(bio, REQ_OP_READ, do_sync); 2539 if (test_bit(FailFast, &rdev->flags) && 2540 test_bit(R1BIO_FailFast, &r1_bio->state)) 2541 bio->bi_opf |= MD_FAILFAST; 2542 bio->bi_private = r1_bio; 2543 if (max_sectors < r1_bio->sectors) { 2544 /* Drat - have to split this up more */ 2545 struct bio *mbio = r1_bio->master_bio; 2546 int sectors_handled = (r1_bio->sector + max_sectors 2547 - mbio->bi_iter.bi_sector); 2548 r1_bio->sectors = max_sectors; 2549 spin_lock_irq(&conf->device_lock); 2550 if (mbio->bi_phys_segments == 0) 2551 mbio->bi_phys_segments = 2; 2552 else 2553 mbio->bi_phys_segments++; 2554 spin_unlock_irq(&conf->device_lock); 2555 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), 2556 bio, bio_dev, bio_sector); 2557 generic_make_request(bio); 2558 bio = NULL; 2559 2560 r1_bio = alloc_r1bio(mddev, mbio, sectors_handled); 2561 set_bit(R1BIO_ReadError, &r1_bio->state); 2562 2563 goto read_more; 2564 } else { 2565 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), 2566 bio, bio_dev, bio_sector); 2567 generic_make_request(bio); 2568 } 2569 } 2570 } 2571 2572 static void raid1d(struct md_thread *thread) 2573 { 2574 struct mddev *mddev = thread->mddev; 2575 struct r1bio *r1_bio; 2576 unsigned long flags; 2577 struct r1conf *conf = mddev->private; 2578 struct list_head *head = &conf->retry_list; 2579 struct blk_plug plug; 2580 int idx; 2581 2582 md_check_recovery(mddev); 2583 2584 if (!list_empty_careful(&conf->bio_end_io_list) && 2585 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { 2586 LIST_HEAD(tmp); 2587 spin_lock_irqsave(&conf->device_lock, flags); 2588 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) 2589 list_splice_init(&conf->bio_end_io_list, &tmp); 2590 spin_unlock_irqrestore(&conf->device_lock, flags); 2591 while (!list_empty(&tmp)) { 2592 r1_bio = list_first_entry(&tmp, struct r1bio, 2593 retry_list); 2594 list_del(&r1_bio->retry_list); 2595 idx = sector_to_idx(r1_bio->sector); 2596 atomic_dec(&conf->nr_queued[idx]); 2597 if (mddev->degraded) 2598 set_bit(R1BIO_Degraded, &r1_bio->state); 2599 if (test_bit(R1BIO_WriteError, &r1_bio->state)) 2600 close_write(r1_bio); 2601 raid_end_bio_io(r1_bio); 2602 } 2603 } 2604 2605 blk_start_plug(&plug); 2606 for (;;) { 2607 2608 flush_pending_writes(conf); 2609 2610 spin_lock_irqsave(&conf->device_lock, flags); 2611 if (list_empty(head)) { 2612 spin_unlock_irqrestore(&conf->device_lock, flags); 2613 break; 2614 } 2615 r1_bio = list_entry(head->prev, struct r1bio, retry_list); 2616 list_del(head->prev); 2617 idx = sector_to_idx(r1_bio->sector); 2618 atomic_dec(&conf->nr_queued[idx]); 2619 spin_unlock_irqrestore(&conf->device_lock, flags); 2620 2621 mddev = r1_bio->mddev; 2622 conf = mddev->private; 2623 if (test_bit(R1BIO_IsSync, &r1_bio->state)) { 2624 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || 2625 test_bit(R1BIO_WriteError, &r1_bio->state)) 2626 handle_sync_write_finished(conf, r1_bio); 2627 else 2628 sync_request_write(mddev, r1_bio); 2629 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || 2630 test_bit(R1BIO_WriteError, &r1_bio->state)) 2631 handle_write_finished(conf, r1_bio); 2632 else if (test_bit(R1BIO_ReadError, &r1_bio->state)) 2633 handle_read_error(conf, r1_bio); 2634 else 2635 /* just a partial read to be scheduled from separate 2636 * context 2637 */ 2638 generic_make_request(r1_bio->bios[r1_bio->read_disk]); 2639 2640 cond_resched(); 2641 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) 2642 md_check_recovery(mddev); 2643 } 2644 blk_finish_plug(&plug); 2645 } 2646 2647 static int init_resync(struct r1conf *conf) 2648 { 2649 int buffs; 2650 2651 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; 2652 BUG_ON(conf->r1buf_pool); 2653 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, 2654 conf->poolinfo); 2655 if (!conf->r1buf_pool) 2656 return -ENOMEM; 2657 return 0; 2658 } 2659 2660 /* 2661 * perform a "sync" on one "block" 2662 * 2663 * We need to make sure that no normal I/O request - particularly write 2664 * requests - conflict with active sync requests. 2665 * 2666 * This is achieved by tracking pending requests and a 'barrier' concept 2667 * that can be installed to exclude normal IO requests. 2668 */ 2669 2670 static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, 2671 int *skipped) 2672 { 2673 struct r1conf *conf = mddev->private; 2674 struct r1bio *r1_bio; 2675 struct bio *bio; 2676 sector_t max_sector, nr_sectors; 2677 int disk = -1; 2678 int i; 2679 int wonly = -1; 2680 int write_targets = 0, read_targets = 0; 2681 sector_t sync_blocks; 2682 int still_degraded = 0; 2683 int good_sectors = RESYNC_SECTORS; 2684 int min_bad = 0; /* number of sectors that are bad in all devices */ 2685 int idx = sector_to_idx(sector_nr); 2686 2687 if (!conf->r1buf_pool) 2688 if (init_resync(conf)) 2689 return 0; 2690 2691 max_sector = mddev->dev_sectors; 2692 if (sector_nr >= max_sector) { 2693 /* If we aborted, we need to abort the 2694 * sync on the 'current' bitmap chunk (there will 2695 * only be one in raid1 resync. 2696 * We can find the current addess in mddev->curr_resync 2697 */ 2698 if (mddev->curr_resync < max_sector) /* aborted */ 2699 bitmap_end_sync(mddev->bitmap, mddev->curr_resync, 2700 &sync_blocks, 1); 2701 else /* completed sync */ 2702 conf->fullsync = 0; 2703 2704 bitmap_close_sync(mddev->bitmap); 2705 close_sync(conf); 2706 2707 if (mddev_is_clustered(mddev)) { 2708 conf->cluster_sync_low = 0; 2709 conf->cluster_sync_high = 0; 2710 } 2711 return 0; 2712 } 2713 2714 if (mddev->bitmap == NULL && 2715 mddev->recovery_cp == MaxSector && 2716 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && 2717 conf->fullsync == 0) { 2718 *skipped = 1; 2719 return max_sector - sector_nr; 2720 } 2721 /* before building a request, check if we can skip these blocks.. 2722 * This call the bitmap_start_sync doesn't actually record anything 2723 */ 2724 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && 2725 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 2726 /* We can skip this block, and probably several more */ 2727 *skipped = 1; 2728 return sync_blocks; 2729 } 2730 2731 /* 2732 * If there is non-resync activity waiting for a turn, then let it 2733 * though before starting on this new sync request. 2734 */ 2735 if (atomic_read(&conf->nr_waiting[idx])) 2736 schedule_timeout_uninterruptible(1); 2737 2738 /* we are incrementing sector_nr below. To be safe, we check against 2739 * sector_nr + two times RESYNC_SECTORS 2740 */ 2741 2742 bitmap_cond_end_sync(mddev->bitmap, sector_nr, 2743 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); 2744 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); 2745 2746 raise_barrier(conf, sector_nr); 2747 2748 rcu_read_lock(); 2749 /* 2750 * If we get a correctably read error during resync or recovery, 2751 * we might want to read from a different device. So we 2752 * flag all drives that could conceivably be read from for READ, 2753 * and any others (which will be non-In_sync devices) for WRITE. 2754 * If a read fails, we try reading from something else for which READ 2755 * is OK. 2756 */ 2757 2758 r1_bio->mddev = mddev; 2759 r1_bio->sector = sector_nr; 2760 r1_bio->state = 0; 2761 set_bit(R1BIO_IsSync, &r1_bio->state); 2762 /* make sure good_sectors won't go across barrier unit boundary */ 2763 good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors); 2764 2765 for (i = 0; i < conf->raid_disks * 2; i++) { 2766 struct md_rdev *rdev; 2767 bio = r1_bio->bios[i]; 2768 bio_reset(bio); 2769 2770 rdev = rcu_dereference(conf->mirrors[i].rdev); 2771 if (rdev == NULL || 2772 test_bit(Faulty, &rdev->flags)) { 2773 if (i < conf->raid_disks) 2774 still_degraded = 1; 2775 } else if (!test_bit(In_sync, &rdev->flags)) { 2776 bio_set_op_attrs(bio, REQ_OP_WRITE, 0); 2777 bio->bi_end_io = end_sync_write; 2778 write_targets ++; 2779 } else { 2780 /* may need to read from here */ 2781 sector_t first_bad = MaxSector; 2782 int bad_sectors; 2783 2784 if (is_badblock(rdev, sector_nr, good_sectors, 2785 &first_bad, &bad_sectors)) { 2786 if (first_bad > sector_nr) 2787 good_sectors = first_bad - sector_nr; 2788 else { 2789 bad_sectors -= (sector_nr - first_bad); 2790 if (min_bad == 0 || 2791 min_bad > bad_sectors) 2792 min_bad = bad_sectors; 2793 } 2794 } 2795 if (sector_nr < first_bad) { 2796 if (test_bit(WriteMostly, &rdev->flags)) { 2797 if (wonly < 0) 2798 wonly = i; 2799 } else { 2800 if (disk < 0) 2801 disk = i; 2802 } 2803 bio_set_op_attrs(bio, REQ_OP_READ, 0); 2804 bio->bi_end_io = end_sync_read; 2805 read_targets++; 2806 } else if (!test_bit(WriteErrorSeen, &rdev->flags) && 2807 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && 2808 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { 2809 /* 2810 * The device is suitable for reading (InSync), 2811 * but has bad block(s) here. Let's try to correct them, 2812 * if we are doing resync or repair. Otherwise, leave 2813 * this device alone for this sync request. 2814 */ 2815 bio_set_op_attrs(bio, REQ_OP_WRITE, 0); 2816 bio->bi_end_io = end_sync_write; 2817 write_targets++; 2818 } 2819 } 2820 if (bio->bi_end_io) { 2821 atomic_inc(&rdev->nr_pending); 2822 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; 2823 bio->bi_bdev = rdev->bdev; 2824 bio->bi_private = r1_bio; 2825 if (test_bit(FailFast, &rdev->flags)) 2826 bio->bi_opf |= MD_FAILFAST; 2827 } 2828 } 2829 rcu_read_unlock(); 2830 if (disk < 0) 2831 disk = wonly; 2832 r1_bio->read_disk = disk; 2833 2834 if (read_targets == 0 && min_bad > 0) { 2835 /* These sectors are bad on all InSync devices, so we 2836 * need to mark them bad on all write targets 2837 */ 2838 int ok = 1; 2839 for (i = 0 ; i < conf->raid_disks * 2 ; i++) 2840 if (r1_bio->bios[i]->bi_end_io == end_sync_write) { 2841 struct md_rdev *rdev = conf->mirrors[i].rdev; 2842 ok = rdev_set_badblocks(rdev, sector_nr, 2843 min_bad, 0 2844 ) && ok; 2845 } 2846 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); 2847 *skipped = 1; 2848 put_buf(r1_bio); 2849 2850 if (!ok) { 2851 /* Cannot record the badblocks, so need to 2852 * abort the resync. 2853 * If there are multiple read targets, could just 2854 * fail the really bad ones ??? 2855 */ 2856 conf->recovery_disabled = mddev->recovery_disabled; 2857 set_bit(MD_RECOVERY_INTR, &mddev->recovery); 2858 return 0; 2859 } else 2860 return min_bad; 2861 2862 } 2863 if (min_bad > 0 && min_bad < good_sectors) { 2864 /* only resync enough to reach the next bad->good 2865 * transition */ 2866 good_sectors = min_bad; 2867 } 2868 2869 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) 2870 /* extra read targets are also write targets */ 2871 write_targets += read_targets-1; 2872 2873 if (write_targets == 0 || read_targets == 0) { 2874 /* There is nowhere to write, so all non-sync 2875 * drives must be failed - so we are finished 2876 */ 2877 sector_t rv; 2878 if (min_bad > 0) 2879 max_sector = sector_nr + min_bad; 2880 rv = max_sector - sector_nr; 2881 *skipped = 1; 2882 put_buf(r1_bio); 2883 return rv; 2884 } 2885 2886 if (max_sector > mddev->resync_max) 2887 max_sector = mddev->resync_max; /* Don't do IO beyond here */ 2888 if (max_sector > sector_nr + good_sectors) 2889 max_sector = sector_nr + good_sectors; 2890 nr_sectors = 0; 2891 sync_blocks = 0; 2892 do { 2893 struct page *page; 2894 int len = PAGE_SIZE; 2895 if (sector_nr + (len>>9) > max_sector) 2896 len = (max_sector - sector_nr) << 9; 2897 if (len == 0) 2898 break; 2899 if (sync_blocks == 0) { 2900 if (!bitmap_start_sync(mddev->bitmap, sector_nr, 2901 &sync_blocks, still_degraded) && 2902 !conf->fullsync && 2903 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 2904 break; 2905 if ((len >> 9) > sync_blocks) 2906 len = sync_blocks<<9; 2907 } 2908 2909 for (i = 0 ; i < conf->raid_disks * 2; i++) { 2910 bio = r1_bio->bios[i]; 2911 if (bio->bi_end_io) { 2912 page = bio->bi_io_vec[bio->bi_vcnt].bv_page; 2913 if (bio_add_page(bio, page, len, 0) == 0) { 2914 /* stop here */ 2915 bio->bi_io_vec[bio->bi_vcnt].bv_page = page; 2916 while (i > 0) { 2917 i--; 2918 bio = r1_bio->bios[i]; 2919 if (bio->bi_end_io==NULL) 2920 continue; 2921 /* remove last page from this bio */ 2922 bio->bi_vcnt--; 2923 bio->bi_iter.bi_size -= len; 2924 bio_clear_flag(bio, BIO_SEG_VALID); 2925 } 2926 goto bio_full; 2927 } 2928 } 2929 } 2930 nr_sectors += len>>9; 2931 sector_nr += len>>9; 2932 sync_blocks -= (len>>9); 2933 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); 2934 bio_full: 2935 r1_bio->sectors = nr_sectors; 2936 2937 if (mddev_is_clustered(mddev) && 2938 conf->cluster_sync_high < sector_nr + nr_sectors) { 2939 conf->cluster_sync_low = mddev->curr_resync_completed; 2940 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; 2941 /* Send resync message */ 2942 md_cluster_ops->resync_info_update(mddev, 2943 conf->cluster_sync_low, 2944 conf->cluster_sync_high); 2945 } 2946 2947 /* For a user-requested sync, we read all readable devices and do a 2948 * compare 2949 */ 2950 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 2951 atomic_set(&r1_bio->remaining, read_targets); 2952 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { 2953 bio = r1_bio->bios[i]; 2954 if (bio->bi_end_io == end_sync_read) { 2955 read_targets--; 2956 md_sync_acct(bio->bi_bdev, nr_sectors); 2957 if (read_targets == 1) 2958 bio->bi_opf &= ~MD_FAILFAST; 2959 generic_make_request(bio); 2960 } 2961 } 2962 } else { 2963 atomic_set(&r1_bio->remaining, 1); 2964 bio = r1_bio->bios[r1_bio->read_disk]; 2965 md_sync_acct(bio->bi_bdev, nr_sectors); 2966 if (read_targets == 1) 2967 bio->bi_opf &= ~MD_FAILFAST; 2968 generic_make_request(bio); 2969 2970 } 2971 return nr_sectors; 2972 } 2973 2974 static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) 2975 { 2976 if (sectors) 2977 return sectors; 2978 2979 return mddev->dev_sectors; 2980 } 2981 2982 static struct r1conf *setup_conf(struct mddev *mddev) 2983 { 2984 struct r1conf *conf; 2985 int i; 2986 struct raid1_info *disk; 2987 struct md_rdev *rdev; 2988 int err = -ENOMEM; 2989 2990 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); 2991 if (!conf) 2992 goto abort; 2993 2994 conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, 2995 sizeof(atomic_t), GFP_KERNEL); 2996 if (!conf->nr_pending) 2997 goto abort; 2998 2999 conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, 3000 sizeof(atomic_t), GFP_KERNEL); 3001 if (!conf->nr_waiting) 3002 goto abort; 3003 3004 conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, 3005 sizeof(atomic_t), GFP_KERNEL); 3006 if (!conf->nr_queued) 3007 goto abort; 3008 3009 conf->barrier = kcalloc(BARRIER_BUCKETS_NR, 3010 sizeof(atomic_t), GFP_KERNEL); 3011 if (!conf->barrier) 3012 goto abort; 3013 3014 conf->mirrors = kzalloc(sizeof(struct raid1_info) 3015 * mddev->raid_disks * 2, 3016 GFP_KERNEL); 3017 if (!conf->mirrors) 3018 goto abort; 3019 3020 conf->tmppage = alloc_page(GFP_KERNEL); 3021 if (!conf->tmppage) 3022 goto abort; 3023 3024 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); 3025 if (!conf->poolinfo) 3026 goto abort; 3027 conf->poolinfo->raid_disks = mddev->raid_disks * 2; 3028 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, 3029 r1bio_pool_free, 3030 conf->poolinfo); 3031 if (!conf->r1bio_pool) 3032 goto abort; 3033 3034 conf->poolinfo->mddev = mddev; 3035 3036 err = -EINVAL; 3037 spin_lock_init(&conf->device_lock); 3038 rdev_for_each(rdev, mddev) { 3039 struct request_queue *q; 3040 int disk_idx = rdev->raid_disk; 3041 if (disk_idx >= mddev->raid_disks 3042 || disk_idx < 0) 3043 continue; 3044 if (test_bit(Replacement, &rdev->flags)) 3045 disk = conf->mirrors + mddev->raid_disks + disk_idx; 3046 else 3047 disk = conf->mirrors + disk_idx; 3048 3049 if (disk->rdev) 3050 goto abort; 3051 disk->rdev = rdev; 3052 q = bdev_get_queue(rdev->bdev); 3053 3054 disk->head_position = 0; 3055 disk->seq_start = MaxSector; 3056 } 3057 conf->raid_disks = mddev->raid_disks; 3058 conf->mddev = mddev; 3059 INIT_LIST_HEAD(&conf->retry_list); 3060 INIT_LIST_HEAD(&conf->bio_end_io_list); 3061 3062 spin_lock_init(&conf->resync_lock); 3063 init_waitqueue_head(&conf->wait_barrier); 3064 3065 bio_list_init(&conf->pending_bio_list); 3066 conf->pending_count = 0; 3067 conf->recovery_disabled = mddev->recovery_disabled - 1; 3068 3069 err = -EIO; 3070 for (i = 0; i < conf->raid_disks * 2; i++) { 3071 3072 disk = conf->mirrors + i; 3073 3074 if (i < conf->raid_disks && 3075 disk[conf->raid_disks].rdev) { 3076 /* This slot has a replacement. */ 3077 if (!disk->rdev) { 3078 /* No original, just make the replacement 3079 * a recovering spare 3080 */ 3081 disk->rdev = 3082 disk[conf->raid_disks].rdev; 3083 disk[conf->raid_disks].rdev = NULL; 3084 } else if (!test_bit(In_sync, &disk->rdev->flags)) 3085 /* Original is not in_sync - bad */ 3086 goto abort; 3087 } 3088 3089 if (!disk->rdev || 3090 !test_bit(In_sync, &disk->rdev->flags)) { 3091 disk->head_position = 0; 3092 if (disk->rdev && 3093 (disk->rdev->saved_raid_disk < 0)) 3094 conf->fullsync = 1; 3095 } 3096 } 3097 3098 err = -ENOMEM; 3099 conf->thread = md_register_thread(raid1d, mddev, "raid1"); 3100 if (!conf->thread) 3101 goto abort; 3102 3103 return conf; 3104 3105 abort: 3106 if (conf) { 3107 mempool_destroy(conf->r1bio_pool); 3108 kfree(conf->mirrors); 3109 safe_put_page(conf->tmppage); 3110 kfree(conf->poolinfo); 3111 kfree(conf->nr_pending); 3112 kfree(conf->nr_waiting); 3113 kfree(conf->nr_queued); 3114 kfree(conf->barrier); 3115 kfree(conf); 3116 } 3117 return ERR_PTR(err); 3118 } 3119 3120 static void raid1_free(struct mddev *mddev, void *priv); 3121 static int raid1_run(struct mddev *mddev) 3122 { 3123 struct r1conf *conf; 3124 int i; 3125 struct md_rdev *rdev; 3126 int ret; 3127 bool discard_supported = false; 3128 3129 if (mddev->level != 1) { 3130 pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", 3131 mdname(mddev), mddev->level); 3132 return -EIO; 3133 } 3134 if (mddev->reshape_position != MaxSector) { 3135 pr_warn("md/raid1:%s: reshape_position set but not supported\n", 3136 mdname(mddev)); 3137 return -EIO; 3138 } 3139 /* 3140 * copy the already verified devices into our private RAID1 3141 * bookkeeping area. [whatever we allocate in run(), 3142 * should be freed in raid1_free()] 3143 */ 3144 if (mddev->private == NULL) 3145 conf = setup_conf(mddev); 3146 else 3147 conf = mddev->private; 3148 3149 if (IS_ERR(conf)) 3150 return PTR_ERR(conf); 3151 3152 if (mddev->queue) 3153 blk_queue_max_write_same_sectors(mddev->queue, 0); 3154 3155 rdev_for_each(rdev, mddev) { 3156 if (!mddev->gendisk) 3157 continue; 3158 disk_stack_limits(mddev->gendisk, rdev->bdev, 3159 rdev->data_offset << 9); 3160 if (blk_queue_discard(bdev_get_queue(rdev->bdev))) 3161 discard_supported = true; 3162 } 3163 3164 mddev->degraded = 0; 3165 for (i=0; i < conf->raid_disks; i++) 3166 if (conf->mirrors[i].rdev == NULL || 3167 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || 3168 test_bit(Faulty, &conf->mirrors[i].rdev->flags)) 3169 mddev->degraded++; 3170 3171 if (conf->raid_disks - mddev->degraded == 1) 3172 mddev->recovery_cp = MaxSector; 3173 3174 if (mddev->recovery_cp != MaxSector) 3175 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", 3176 mdname(mddev)); 3177 pr_info("md/raid1:%s: active with %d out of %d mirrors\n", 3178 mdname(mddev), mddev->raid_disks - mddev->degraded, 3179 mddev->raid_disks); 3180 3181 /* 3182 * Ok, everything is just fine now 3183 */ 3184 mddev->thread = conf->thread; 3185 conf->thread = NULL; 3186 mddev->private = conf; 3187 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); 3188 3189 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); 3190 3191 if (mddev->queue) { 3192 if (discard_supported) 3193 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, 3194 mddev->queue); 3195 else 3196 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, 3197 mddev->queue); 3198 } 3199 3200 ret = md_integrity_register(mddev); 3201 if (ret) { 3202 md_unregister_thread(&mddev->thread); 3203 raid1_free(mddev, conf); 3204 } 3205 return ret; 3206 } 3207 3208 static void raid1_free(struct mddev *mddev, void *priv) 3209 { 3210 struct r1conf *conf = priv; 3211 3212 mempool_destroy(conf->r1bio_pool); 3213 kfree(conf->mirrors); 3214 safe_put_page(conf->tmppage); 3215 kfree(conf->poolinfo); 3216 kfree(conf->nr_pending); 3217 kfree(conf->nr_waiting); 3218 kfree(conf->nr_queued); 3219 kfree(conf->barrier); 3220 kfree(conf); 3221 } 3222 3223 static int raid1_resize(struct mddev *mddev, sector_t sectors) 3224 { 3225 /* no resync is happening, and there is enough space 3226 * on all devices, so we can resize. 3227 * We need to make sure resync covers any new space. 3228 * If the array is shrinking we should possibly wait until 3229 * any io in the removed space completes, but it hardly seems 3230 * worth it. 3231 */ 3232 sector_t newsize = raid1_size(mddev, sectors, 0); 3233 if (mddev->external_size && 3234 mddev->array_sectors > newsize) 3235 return -EINVAL; 3236 if (mddev->bitmap) { 3237 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); 3238 if (ret) 3239 return ret; 3240 } 3241 md_set_array_sectors(mddev, newsize); 3242 set_capacity(mddev->gendisk, mddev->array_sectors); 3243 revalidate_disk(mddev->gendisk); 3244 if (sectors > mddev->dev_sectors && 3245 mddev->recovery_cp > mddev->dev_sectors) { 3246 mddev->recovery_cp = mddev->dev_sectors; 3247 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3248 } 3249 mddev->dev_sectors = sectors; 3250 mddev->resync_max_sectors = sectors; 3251 return 0; 3252 } 3253 3254 static int raid1_reshape(struct mddev *mddev) 3255 { 3256 /* We need to: 3257 * 1/ resize the r1bio_pool 3258 * 2/ resize conf->mirrors 3259 * 3260 * We allocate a new r1bio_pool if we can. 3261 * Then raise a device barrier and wait until all IO stops. 3262 * Then resize conf->mirrors and swap in the new r1bio pool. 3263 * 3264 * At the same time, we "pack" the devices so that all the missing 3265 * devices have the higher raid_disk numbers. 3266 */ 3267 mempool_t *newpool, *oldpool; 3268 struct pool_info *newpoolinfo; 3269 struct raid1_info *newmirrors; 3270 struct r1conf *conf = mddev->private; 3271 int cnt, raid_disks; 3272 unsigned long flags; 3273 int d, d2, err; 3274 3275 /* Cannot change chunk_size, layout, or level */ 3276 if (mddev->chunk_sectors != mddev->new_chunk_sectors || 3277 mddev->layout != mddev->new_layout || 3278 mddev->level != mddev->new_level) { 3279 mddev->new_chunk_sectors = mddev->chunk_sectors; 3280 mddev->new_layout = mddev->layout; 3281 mddev->new_level = mddev->level; 3282 return -EINVAL; 3283 } 3284 3285 if (!mddev_is_clustered(mddev)) { 3286 err = md_allow_write(mddev); 3287 if (err) 3288 return err; 3289 } 3290 3291 raid_disks = mddev->raid_disks + mddev->delta_disks; 3292 3293 if (raid_disks < conf->raid_disks) { 3294 cnt=0; 3295 for (d= 0; d < conf->raid_disks; d++) 3296 if (conf->mirrors[d].rdev) 3297 cnt++; 3298 if (cnt > raid_disks) 3299 return -EBUSY; 3300 } 3301 3302 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); 3303 if (!newpoolinfo) 3304 return -ENOMEM; 3305 newpoolinfo->mddev = mddev; 3306 newpoolinfo->raid_disks = raid_disks * 2; 3307 3308 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, 3309 r1bio_pool_free, newpoolinfo); 3310 if (!newpool) { 3311 kfree(newpoolinfo); 3312 return -ENOMEM; 3313 } 3314 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, 3315 GFP_KERNEL); 3316 if (!newmirrors) { 3317 kfree(newpoolinfo); 3318 mempool_destroy(newpool); 3319 return -ENOMEM; 3320 } 3321 3322 freeze_array(conf, 0); 3323 3324 /* ok, everything is stopped */ 3325 oldpool = conf->r1bio_pool; 3326 conf->r1bio_pool = newpool; 3327 3328 for (d = d2 = 0; d < conf->raid_disks; d++) { 3329 struct md_rdev *rdev = conf->mirrors[d].rdev; 3330 if (rdev && rdev->raid_disk != d2) { 3331 sysfs_unlink_rdev(mddev, rdev); 3332 rdev->raid_disk = d2; 3333 sysfs_unlink_rdev(mddev, rdev); 3334 if (sysfs_link_rdev(mddev, rdev)) 3335 pr_warn("md/raid1:%s: cannot register rd%d\n", 3336 mdname(mddev), rdev->raid_disk); 3337 } 3338 if (rdev) 3339 newmirrors[d2++].rdev = rdev; 3340 } 3341 kfree(conf->mirrors); 3342 conf->mirrors = newmirrors; 3343 kfree(conf->poolinfo); 3344 conf->poolinfo = newpoolinfo; 3345 3346 spin_lock_irqsave(&conf->device_lock, flags); 3347 mddev->degraded += (raid_disks - conf->raid_disks); 3348 spin_unlock_irqrestore(&conf->device_lock, flags); 3349 conf->raid_disks = mddev->raid_disks = raid_disks; 3350 mddev->delta_disks = 0; 3351 3352 unfreeze_array(conf); 3353 3354 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); 3355 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 3356 md_wakeup_thread(mddev->thread); 3357 3358 mempool_destroy(oldpool); 3359 return 0; 3360 } 3361 3362 static void raid1_quiesce(struct mddev *mddev, int state) 3363 { 3364 struct r1conf *conf = mddev->private; 3365 3366 switch(state) { 3367 case 2: /* wake for suspend */ 3368 wake_up(&conf->wait_barrier); 3369 break; 3370 case 1: 3371 freeze_array(conf, 0); 3372 break; 3373 case 0: 3374 unfreeze_array(conf); 3375 break; 3376 } 3377 } 3378 3379 static void *raid1_takeover(struct mddev *mddev) 3380 { 3381 /* raid1 can take over: 3382 * raid5 with 2 devices, any layout or chunk size 3383 */ 3384 if (mddev->level == 5 && mddev->raid_disks == 2) { 3385 struct r1conf *conf; 3386 mddev->new_level = 1; 3387 mddev->new_layout = 0; 3388 mddev->new_chunk_sectors = 0; 3389 conf = setup_conf(mddev); 3390 if (!IS_ERR(conf)) { 3391 /* Array must appear to be quiesced */ 3392 conf->array_frozen = 1; 3393 mddev_clear_unsupported_flags(mddev, 3394 UNSUPPORTED_MDDEV_FLAGS); 3395 } 3396 return conf; 3397 } 3398 return ERR_PTR(-EINVAL); 3399 } 3400 3401 static struct md_personality raid1_personality = 3402 { 3403 .name = "raid1", 3404 .level = 1, 3405 .owner = THIS_MODULE, 3406 .make_request = raid1_make_request, 3407 .run = raid1_run, 3408 .free = raid1_free, 3409 .status = raid1_status, 3410 .error_handler = raid1_error, 3411 .hot_add_disk = raid1_add_disk, 3412 .hot_remove_disk= raid1_remove_disk, 3413 .spare_active = raid1_spare_active, 3414 .sync_request = raid1_sync_request, 3415 .resize = raid1_resize, 3416 .size = raid1_size, 3417 .check_reshape = raid1_reshape, 3418 .quiesce = raid1_quiesce, 3419 .takeover = raid1_takeover, 3420 .congested = raid1_congested, 3421 }; 3422 3423 static int __init raid_init(void) 3424 { 3425 return register_md_personality(&raid1_personality); 3426 } 3427 3428 static void raid_exit(void) 3429 { 3430 unregister_md_personality(&raid1_personality); 3431 } 3432 3433 module_init(raid_init); 3434 module_exit(raid_exit); 3435 MODULE_LICENSE("GPL"); 3436 MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); 3437 MODULE_ALIAS("md-personality-3"); /* RAID1 */ 3438 MODULE_ALIAS("md-raid1"); 3439 MODULE_ALIAS("md-level-1"); 3440 3441 module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); 3442