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