1 /* 2 * Copyright (C) 2012 Fusion-io All rights reserved. 3 * Copyright (C) 2012 Intel Corp. All rights reserved. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public 7 * License v2 as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 12 * General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public 15 * License along with this program; if not, write to the 16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 17 * Boston, MA 021110-1307, USA. 18 */ 19 #include <linux/sched.h> 20 #include <linux/wait.h> 21 #include <linux/bio.h> 22 #include <linux/slab.h> 23 #include <linux/buffer_head.h> 24 #include <linux/blkdev.h> 25 #include <linux/random.h> 26 #include <linux/iocontext.h> 27 #include <linux/capability.h> 28 #include <linux/ratelimit.h> 29 #include <linux/kthread.h> 30 #include <linux/raid/pq.h> 31 #include <linux/hash.h> 32 #include <linux/list_sort.h> 33 #include <linux/raid/xor.h> 34 #include <linux/vmalloc.h> 35 #include <asm/div64.h> 36 #include "compat.h" 37 #include "ctree.h" 38 #include "extent_map.h" 39 #include "disk-io.h" 40 #include "transaction.h" 41 #include "print-tree.h" 42 #include "volumes.h" 43 #include "raid56.h" 44 #include "async-thread.h" 45 #include "check-integrity.h" 46 #include "rcu-string.h" 47 48 /* set when additional merges to this rbio are not allowed */ 49 #define RBIO_RMW_LOCKED_BIT 1 50 51 /* 52 * set when this rbio is sitting in the hash, but it is just a cache 53 * of past RMW 54 */ 55 #define RBIO_CACHE_BIT 2 56 57 /* 58 * set when it is safe to trust the stripe_pages for caching 59 */ 60 #define RBIO_CACHE_READY_BIT 3 61 62 63 #define RBIO_CACHE_SIZE 1024 64 65 struct btrfs_raid_bio { 66 struct btrfs_fs_info *fs_info; 67 struct btrfs_bio *bbio; 68 69 /* 70 * logical block numbers for the start of each stripe 71 * The last one or two are p/q. These are sorted, 72 * so raid_map[0] is the start of our full stripe 73 */ 74 u64 *raid_map; 75 76 /* while we're doing rmw on a stripe 77 * we put it into a hash table so we can 78 * lock the stripe and merge more rbios 79 * into it. 80 */ 81 struct list_head hash_list; 82 83 /* 84 * LRU list for the stripe cache 85 */ 86 struct list_head stripe_cache; 87 88 /* 89 * for scheduling work in the helper threads 90 */ 91 struct btrfs_work work; 92 93 /* 94 * bio list and bio_list_lock are used 95 * to add more bios into the stripe 96 * in hopes of avoiding the full rmw 97 */ 98 struct bio_list bio_list; 99 spinlock_t bio_list_lock; 100 101 /* also protected by the bio_list_lock, the 102 * plug list is used by the plugging code 103 * to collect partial bios while plugged. The 104 * stripe locking code also uses it to hand off 105 * the stripe lock to the next pending IO 106 */ 107 struct list_head plug_list; 108 109 /* 110 * flags that tell us if it is safe to 111 * merge with this bio 112 */ 113 unsigned long flags; 114 115 /* size of each individual stripe on disk */ 116 int stripe_len; 117 118 /* number of data stripes (no p/q) */ 119 int nr_data; 120 121 /* 122 * set if we're doing a parity rebuild 123 * for a read from higher up, which is handled 124 * differently from a parity rebuild as part of 125 * rmw 126 */ 127 int read_rebuild; 128 129 /* first bad stripe */ 130 int faila; 131 132 /* second bad stripe (for raid6 use) */ 133 int failb; 134 135 /* 136 * number of pages needed to represent the full 137 * stripe 138 */ 139 int nr_pages; 140 141 /* 142 * size of all the bios in the bio_list. This 143 * helps us decide if the rbio maps to a full 144 * stripe or not 145 */ 146 int bio_list_bytes; 147 148 atomic_t refs; 149 150 /* 151 * these are two arrays of pointers. We allocate the 152 * rbio big enough to hold them both and setup their 153 * locations when the rbio is allocated 154 */ 155 156 /* pointers to pages that we allocated for 157 * reading/writing stripes directly from the disk (including P/Q) 158 */ 159 struct page **stripe_pages; 160 161 /* 162 * pointers to the pages in the bio_list. Stored 163 * here for faster lookup 164 */ 165 struct page **bio_pages; 166 }; 167 168 static int __raid56_parity_recover(struct btrfs_raid_bio *rbio); 169 static noinline void finish_rmw(struct btrfs_raid_bio *rbio); 170 static void rmw_work(struct btrfs_work *work); 171 static void read_rebuild_work(struct btrfs_work *work); 172 static void async_rmw_stripe(struct btrfs_raid_bio *rbio); 173 static void async_read_rebuild(struct btrfs_raid_bio *rbio); 174 static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio); 175 static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed); 176 static void __free_raid_bio(struct btrfs_raid_bio *rbio); 177 static void index_rbio_pages(struct btrfs_raid_bio *rbio); 178 static int alloc_rbio_pages(struct btrfs_raid_bio *rbio); 179 180 /* 181 * the stripe hash table is used for locking, and to collect 182 * bios in hopes of making a full stripe 183 */ 184 int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info) 185 { 186 struct btrfs_stripe_hash_table *table; 187 struct btrfs_stripe_hash_table *x; 188 struct btrfs_stripe_hash *cur; 189 struct btrfs_stripe_hash *h; 190 int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS; 191 int i; 192 int table_size; 193 194 if (info->stripe_hash_table) 195 return 0; 196 197 /* 198 * The table is large, starting with order 4 and can go as high as 199 * order 7 in case lock debugging is turned on. 200 * 201 * Try harder to allocate and fallback to vmalloc to lower the chance 202 * of a failing mount. 203 */ 204 table_size = sizeof(*table) + sizeof(*h) * num_entries; 205 table = kzalloc(table_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); 206 if (!table) { 207 table = vzalloc(table_size); 208 if (!table) 209 return -ENOMEM; 210 } 211 212 spin_lock_init(&table->cache_lock); 213 INIT_LIST_HEAD(&table->stripe_cache); 214 215 h = table->table; 216 217 for (i = 0; i < num_entries; i++) { 218 cur = h + i; 219 INIT_LIST_HEAD(&cur->hash_list); 220 spin_lock_init(&cur->lock); 221 init_waitqueue_head(&cur->wait); 222 } 223 224 x = cmpxchg(&info->stripe_hash_table, NULL, table); 225 if (x) { 226 if (is_vmalloc_addr(x)) 227 vfree(x); 228 else 229 kfree(x); 230 } 231 return 0; 232 } 233 234 /* 235 * caching an rbio means to copy anything from the 236 * bio_pages array into the stripe_pages array. We 237 * use the page uptodate bit in the stripe cache array 238 * to indicate if it has valid data 239 * 240 * once the caching is done, we set the cache ready 241 * bit. 242 */ 243 static void cache_rbio_pages(struct btrfs_raid_bio *rbio) 244 { 245 int i; 246 char *s; 247 char *d; 248 int ret; 249 250 ret = alloc_rbio_pages(rbio); 251 if (ret) 252 return; 253 254 for (i = 0; i < rbio->nr_pages; i++) { 255 if (!rbio->bio_pages[i]) 256 continue; 257 258 s = kmap(rbio->bio_pages[i]); 259 d = kmap(rbio->stripe_pages[i]); 260 261 memcpy(d, s, PAGE_CACHE_SIZE); 262 263 kunmap(rbio->bio_pages[i]); 264 kunmap(rbio->stripe_pages[i]); 265 SetPageUptodate(rbio->stripe_pages[i]); 266 } 267 set_bit(RBIO_CACHE_READY_BIT, &rbio->flags); 268 } 269 270 /* 271 * we hash on the first logical address of the stripe 272 */ 273 static int rbio_bucket(struct btrfs_raid_bio *rbio) 274 { 275 u64 num = rbio->raid_map[0]; 276 277 /* 278 * we shift down quite a bit. We're using byte 279 * addressing, and most of the lower bits are zeros. 280 * This tends to upset hash_64, and it consistently 281 * returns just one or two different values. 282 * 283 * shifting off the lower bits fixes things. 284 */ 285 return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS); 286 } 287 288 /* 289 * stealing an rbio means taking all the uptodate pages from the stripe 290 * array in the source rbio and putting them into the destination rbio 291 */ 292 static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest) 293 { 294 int i; 295 struct page *s; 296 struct page *d; 297 298 if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags)) 299 return; 300 301 for (i = 0; i < dest->nr_pages; i++) { 302 s = src->stripe_pages[i]; 303 if (!s || !PageUptodate(s)) { 304 continue; 305 } 306 307 d = dest->stripe_pages[i]; 308 if (d) 309 __free_page(d); 310 311 dest->stripe_pages[i] = s; 312 src->stripe_pages[i] = NULL; 313 } 314 } 315 316 /* 317 * merging means we take the bio_list from the victim and 318 * splice it into the destination. The victim should 319 * be discarded afterwards. 320 * 321 * must be called with dest->rbio_list_lock held 322 */ 323 static void merge_rbio(struct btrfs_raid_bio *dest, 324 struct btrfs_raid_bio *victim) 325 { 326 bio_list_merge(&dest->bio_list, &victim->bio_list); 327 dest->bio_list_bytes += victim->bio_list_bytes; 328 bio_list_init(&victim->bio_list); 329 } 330 331 /* 332 * used to prune items that are in the cache. The caller 333 * must hold the hash table lock. 334 */ 335 static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio) 336 { 337 int bucket = rbio_bucket(rbio); 338 struct btrfs_stripe_hash_table *table; 339 struct btrfs_stripe_hash *h; 340 int freeit = 0; 341 342 /* 343 * check the bit again under the hash table lock. 344 */ 345 if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) 346 return; 347 348 table = rbio->fs_info->stripe_hash_table; 349 h = table->table + bucket; 350 351 /* hold the lock for the bucket because we may be 352 * removing it from the hash table 353 */ 354 spin_lock(&h->lock); 355 356 /* 357 * hold the lock for the bio list because we need 358 * to make sure the bio list is empty 359 */ 360 spin_lock(&rbio->bio_list_lock); 361 362 if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) { 363 list_del_init(&rbio->stripe_cache); 364 table->cache_size -= 1; 365 freeit = 1; 366 367 /* if the bio list isn't empty, this rbio is 368 * still involved in an IO. We take it out 369 * of the cache list, and drop the ref that 370 * was held for the list. 371 * 372 * If the bio_list was empty, we also remove 373 * the rbio from the hash_table, and drop 374 * the corresponding ref 375 */ 376 if (bio_list_empty(&rbio->bio_list)) { 377 if (!list_empty(&rbio->hash_list)) { 378 list_del_init(&rbio->hash_list); 379 atomic_dec(&rbio->refs); 380 BUG_ON(!list_empty(&rbio->plug_list)); 381 } 382 } 383 } 384 385 spin_unlock(&rbio->bio_list_lock); 386 spin_unlock(&h->lock); 387 388 if (freeit) 389 __free_raid_bio(rbio); 390 } 391 392 /* 393 * prune a given rbio from the cache 394 */ 395 static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio) 396 { 397 struct btrfs_stripe_hash_table *table; 398 unsigned long flags; 399 400 if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) 401 return; 402 403 table = rbio->fs_info->stripe_hash_table; 404 405 spin_lock_irqsave(&table->cache_lock, flags); 406 __remove_rbio_from_cache(rbio); 407 spin_unlock_irqrestore(&table->cache_lock, flags); 408 } 409 410 /* 411 * remove everything in the cache 412 */ 413 static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info) 414 { 415 struct btrfs_stripe_hash_table *table; 416 unsigned long flags; 417 struct btrfs_raid_bio *rbio; 418 419 table = info->stripe_hash_table; 420 421 spin_lock_irqsave(&table->cache_lock, flags); 422 while (!list_empty(&table->stripe_cache)) { 423 rbio = list_entry(table->stripe_cache.next, 424 struct btrfs_raid_bio, 425 stripe_cache); 426 __remove_rbio_from_cache(rbio); 427 } 428 spin_unlock_irqrestore(&table->cache_lock, flags); 429 } 430 431 /* 432 * remove all cached entries and free the hash table 433 * used by unmount 434 */ 435 void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info) 436 { 437 if (!info->stripe_hash_table) 438 return; 439 btrfs_clear_rbio_cache(info); 440 if (is_vmalloc_addr(info->stripe_hash_table)) 441 vfree(info->stripe_hash_table); 442 else 443 kfree(info->stripe_hash_table); 444 info->stripe_hash_table = NULL; 445 } 446 447 /* 448 * insert an rbio into the stripe cache. It 449 * must have already been prepared by calling 450 * cache_rbio_pages 451 * 452 * If this rbio was already cached, it gets 453 * moved to the front of the lru. 454 * 455 * If the size of the rbio cache is too big, we 456 * prune an item. 457 */ 458 static void cache_rbio(struct btrfs_raid_bio *rbio) 459 { 460 struct btrfs_stripe_hash_table *table; 461 unsigned long flags; 462 463 if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags)) 464 return; 465 466 table = rbio->fs_info->stripe_hash_table; 467 468 spin_lock_irqsave(&table->cache_lock, flags); 469 spin_lock(&rbio->bio_list_lock); 470 471 /* bump our ref if we were not in the list before */ 472 if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags)) 473 atomic_inc(&rbio->refs); 474 475 if (!list_empty(&rbio->stripe_cache)){ 476 list_move(&rbio->stripe_cache, &table->stripe_cache); 477 } else { 478 list_add(&rbio->stripe_cache, &table->stripe_cache); 479 table->cache_size += 1; 480 } 481 482 spin_unlock(&rbio->bio_list_lock); 483 484 if (table->cache_size > RBIO_CACHE_SIZE) { 485 struct btrfs_raid_bio *found; 486 487 found = list_entry(table->stripe_cache.prev, 488 struct btrfs_raid_bio, 489 stripe_cache); 490 491 if (found != rbio) 492 __remove_rbio_from_cache(found); 493 } 494 495 spin_unlock_irqrestore(&table->cache_lock, flags); 496 return; 497 } 498 499 /* 500 * helper function to run the xor_blocks api. It is only 501 * able to do MAX_XOR_BLOCKS at a time, so we need to 502 * loop through. 503 */ 504 static void run_xor(void **pages, int src_cnt, ssize_t len) 505 { 506 int src_off = 0; 507 int xor_src_cnt = 0; 508 void *dest = pages[src_cnt]; 509 510 while(src_cnt > 0) { 511 xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); 512 xor_blocks(xor_src_cnt, len, dest, pages + src_off); 513 514 src_cnt -= xor_src_cnt; 515 src_off += xor_src_cnt; 516 } 517 } 518 519 /* 520 * returns true if the bio list inside this rbio 521 * covers an entire stripe (no rmw required). 522 * Must be called with the bio list lock held, or 523 * at a time when you know it is impossible to add 524 * new bios into the list 525 */ 526 static int __rbio_is_full(struct btrfs_raid_bio *rbio) 527 { 528 unsigned long size = rbio->bio_list_bytes; 529 int ret = 1; 530 531 if (size != rbio->nr_data * rbio->stripe_len) 532 ret = 0; 533 534 BUG_ON(size > rbio->nr_data * rbio->stripe_len); 535 return ret; 536 } 537 538 static int rbio_is_full(struct btrfs_raid_bio *rbio) 539 { 540 unsigned long flags; 541 int ret; 542 543 spin_lock_irqsave(&rbio->bio_list_lock, flags); 544 ret = __rbio_is_full(rbio); 545 spin_unlock_irqrestore(&rbio->bio_list_lock, flags); 546 return ret; 547 } 548 549 /* 550 * returns 1 if it is safe to merge two rbios together. 551 * The merging is safe if the two rbios correspond to 552 * the same stripe and if they are both going in the same 553 * direction (read vs write), and if neither one is 554 * locked for final IO 555 * 556 * The caller is responsible for locking such that 557 * rmw_locked is safe to test 558 */ 559 static int rbio_can_merge(struct btrfs_raid_bio *last, 560 struct btrfs_raid_bio *cur) 561 { 562 if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) || 563 test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) 564 return 0; 565 566 /* 567 * we can't merge with cached rbios, since the 568 * idea is that when we merge the destination 569 * rbio is going to run our IO for us. We can 570 * steal from cached rbio's though, other functions 571 * handle that. 572 */ 573 if (test_bit(RBIO_CACHE_BIT, &last->flags) || 574 test_bit(RBIO_CACHE_BIT, &cur->flags)) 575 return 0; 576 577 if (last->raid_map[0] != 578 cur->raid_map[0]) 579 return 0; 580 581 /* reads can't merge with writes */ 582 if (last->read_rebuild != 583 cur->read_rebuild) { 584 return 0; 585 } 586 587 return 1; 588 } 589 590 /* 591 * helper to index into the pstripe 592 */ 593 static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index) 594 { 595 index += (rbio->nr_data * rbio->stripe_len) >> PAGE_CACHE_SHIFT; 596 return rbio->stripe_pages[index]; 597 } 598 599 /* 600 * helper to index into the qstripe, returns null 601 * if there is no qstripe 602 */ 603 static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index) 604 { 605 if (rbio->nr_data + 1 == rbio->bbio->num_stripes) 606 return NULL; 607 608 index += ((rbio->nr_data + 1) * rbio->stripe_len) >> 609 PAGE_CACHE_SHIFT; 610 return rbio->stripe_pages[index]; 611 } 612 613 /* 614 * The first stripe in the table for a logical address 615 * has the lock. rbios are added in one of three ways: 616 * 617 * 1) Nobody has the stripe locked yet. The rbio is given 618 * the lock and 0 is returned. The caller must start the IO 619 * themselves. 620 * 621 * 2) Someone has the stripe locked, but we're able to merge 622 * with the lock owner. The rbio is freed and the IO will 623 * start automatically along with the existing rbio. 1 is returned. 624 * 625 * 3) Someone has the stripe locked, but we're not able to merge. 626 * The rbio is added to the lock owner's plug list, or merged into 627 * an rbio already on the plug list. When the lock owner unlocks, 628 * the next rbio on the list is run and the IO is started automatically. 629 * 1 is returned 630 * 631 * If we return 0, the caller still owns the rbio and must continue with 632 * IO submission. If we return 1, the caller must assume the rbio has 633 * already been freed. 634 */ 635 static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio) 636 { 637 int bucket = rbio_bucket(rbio); 638 struct btrfs_stripe_hash *h = rbio->fs_info->stripe_hash_table->table + bucket; 639 struct btrfs_raid_bio *cur; 640 struct btrfs_raid_bio *pending; 641 unsigned long flags; 642 DEFINE_WAIT(wait); 643 struct btrfs_raid_bio *freeit = NULL; 644 struct btrfs_raid_bio *cache_drop = NULL; 645 int ret = 0; 646 int walk = 0; 647 648 spin_lock_irqsave(&h->lock, flags); 649 list_for_each_entry(cur, &h->hash_list, hash_list) { 650 walk++; 651 if (cur->raid_map[0] == rbio->raid_map[0]) { 652 spin_lock(&cur->bio_list_lock); 653 654 /* can we steal this cached rbio's pages? */ 655 if (bio_list_empty(&cur->bio_list) && 656 list_empty(&cur->plug_list) && 657 test_bit(RBIO_CACHE_BIT, &cur->flags) && 658 !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) { 659 list_del_init(&cur->hash_list); 660 atomic_dec(&cur->refs); 661 662 steal_rbio(cur, rbio); 663 cache_drop = cur; 664 spin_unlock(&cur->bio_list_lock); 665 666 goto lockit; 667 } 668 669 /* can we merge into the lock owner? */ 670 if (rbio_can_merge(cur, rbio)) { 671 merge_rbio(cur, rbio); 672 spin_unlock(&cur->bio_list_lock); 673 freeit = rbio; 674 ret = 1; 675 goto out; 676 } 677 678 679 /* 680 * we couldn't merge with the running 681 * rbio, see if we can merge with the 682 * pending ones. We don't have to 683 * check for rmw_locked because there 684 * is no way they are inside finish_rmw 685 * right now 686 */ 687 list_for_each_entry(pending, &cur->plug_list, 688 plug_list) { 689 if (rbio_can_merge(pending, rbio)) { 690 merge_rbio(pending, rbio); 691 spin_unlock(&cur->bio_list_lock); 692 freeit = rbio; 693 ret = 1; 694 goto out; 695 } 696 } 697 698 /* no merging, put us on the tail of the plug list, 699 * our rbio will be started with the currently 700 * running rbio unlocks 701 */ 702 list_add_tail(&rbio->plug_list, &cur->plug_list); 703 spin_unlock(&cur->bio_list_lock); 704 ret = 1; 705 goto out; 706 } 707 } 708 lockit: 709 atomic_inc(&rbio->refs); 710 list_add(&rbio->hash_list, &h->hash_list); 711 out: 712 spin_unlock_irqrestore(&h->lock, flags); 713 if (cache_drop) 714 remove_rbio_from_cache(cache_drop); 715 if (freeit) 716 __free_raid_bio(freeit); 717 return ret; 718 } 719 720 /* 721 * called as rmw or parity rebuild is completed. If the plug list has more 722 * rbios waiting for this stripe, the next one on the list will be started 723 */ 724 static noinline void unlock_stripe(struct btrfs_raid_bio *rbio) 725 { 726 int bucket; 727 struct btrfs_stripe_hash *h; 728 unsigned long flags; 729 int keep_cache = 0; 730 731 bucket = rbio_bucket(rbio); 732 h = rbio->fs_info->stripe_hash_table->table + bucket; 733 734 if (list_empty(&rbio->plug_list)) 735 cache_rbio(rbio); 736 737 spin_lock_irqsave(&h->lock, flags); 738 spin_lock(&rbio->bio_list_lock); 739 740 if (!list_empty(&rbio->hash_list)) { 741 /* 742 * if we're still cached and there is no other IO 743 * to perform, just leave this rbio here for others 744 * to steal from later 745 */ 746 if (list_empty(&rbio->plug_list) && 747 test_bit(RBIO_CACHE_BIT, &rbio->flags)) { 748 keep_cache = 1; 749 clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); 750 BUG_ON(!bio_list_empty(&rbio->bio_list)); 751 goto done; 752 } 753 754 list_del_init(&rbio->hash_list); 755 atomic_dec(&rbio->refs); 756 757 /* 758 * we use the plug list to hold all the rbios 759 * waiting for the chance to lock this stripe. 760 * hand the lock over to one of them. 761 */ 762 if (!list_empty(&rbio->plug_list)) { 763 struct btrfs_raid_bio *next; 764 struct list_head *head = rbio->plug_list.next; 765 766 next = list_entry(head, struct btrfs_raid_bio, 767 plug_list); 768 769 list_del_init(&rbio->plug_list); 770 771 list_add(&next->hash_list, &h->hash_list); 772 atomic_inc(&next->refs); 773 spin_unlock(&rbio->bio_list_lock); 774 spin_unlock_irqrestore(&h->lock, flags); 775 776 if (next->read_rebuild) 777 async_read_rebuild(next); 778 else { 779 steal_rbio(rbio, next); 780 async_rmw_stripe(next); 781 } 782 783 goto done_nolock; 784 } else if (waitqueue_active(&h->wait)) { 785 spin_unlock(&rbio->bio_list_lock); 786 spin_unlock_irqrestore(&h->lock, flags); 787 wake_up(&h->wait); 788 goto done_nolock; 789 } 790 } 791 done: 792 spin_unlock(&rbio->bio_list_lock); 793 spin_unlock_irqrestore(&h->lock, flags); 794 795 done_nolock: 796 if (!keep_cache) 797 remove_rbio_from_cache(rbio); 798 } 799 800 static void __free_raid_bio(struct btrfs_raid_bio *rbio) 801 { 802 int i; 803 804 WARN_ON(atomic_read(&rbio->refs) < 0); 805 if (!atomic_dec_and_test(&rbio->refs)) 806 return; 807 808 WARN_ON(!list_empty(&rbio->stripe_cache)); 809 WARN_ON(!list_empty(&rbio->hash_list)); 810 WARN_ON(!bio_list_empty(&rbio->bio_list)); 811 812 for (i = 0; i < rbio->nr_pages; i++) { 813 if (rbio->stripe_pages[i]) { 814 __free_page(rbio->stripe_pages[i]); 815 rbio->stripe_pages[i] = NULL; 816 } 817 } 818 kfree(rbio->raid_map); 819 kfree(rbio->bbio); 820 kfree(rbio); 821 } 822 823 static void free_raid_bio(struct btrfs_raid_bio *rbio) 824 { 825 unlock_stripe(rbio); 826 __free_raid_bio(rbio); 827 } 828 829 /* 830 * this frees the rbio and runs through all the bios in the 831 * bio_list and calls end_io on them 832 */ 833 static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, int err, int uptodate) 834 { 835 struct bio *cur = bio_list_get(&rbio->bio_list); 836 struct bio *next; 837 free_raid_bio(rbio); 838 839 while (cur) { 840 next = cur->bi_next; 841 cur->bi_next = NULL; 842 if (uptodate) 843 set_bit(BIO_UPTODATE, &cur->bi_flags); 844 bio_endio(cur, err); 845 cur = next; 846 } 847 } 848 849 /* 850 * end io function used by finish_rmw. When we finally 851 * get here, we've written a full stripe 852 */ 853 static void raid_write_end_io(struct bio *bio, int err) 854 { 855 struct btrfs_raid_bio *rbio = bio->bi_private; 856 857 if (err) 858 fail_bio_stripe(rbio, bio); 859 860 bio_put(bio); 861 862 if (!atomic_dec_and_test(&rbio->bbio->stripes_pending)) 863 return; 864 865 err = 0; 866 867 /* OK, we have read all the stripes we need to. */ 868 if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors) 869 err = -EIO; 870 871 rbio_orig_end_io(rbio, err, 0); 872 return; 873 } 874 875 /* 876 * the read/modify/write code wants to use the original bio for 877 * any pages it included, and then use the rbio for everything 878 * else. This function decides if a given index (stripe number) 879 * and page number in that stripe fall inside the original bio 880 * or the rbio. 881 * 882 * if you set bio_list_only, you'll get a NULL back for any ranges 883 * that are outside the bio_list 884 * 885 * This doesn't take any refs on anything, you get a bare page pointer 886 * and the caller must bump refs as required. 887 * 888 * You must call index_rbio_pages once before you can trust 889 * the answers from this function. 890 */ 891 static struct page *page_in_rbio(struct btrfs_raid_bio *rbio, 892 int index, int pagenr, int bio_list_only) 893 { 894 int chunk_page; 895 struct page *p = NULL; 896 897 chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr; 898 899 spin_lock_irq(&rbio->bio_list_lock); 900 p = rbio->bio_pages[chunk_page]; 901 spin_unlock_irq(&rbio->bio_list_lock); 902 903 if (p || bio_list_only) 904 return p; 905 906 return rbio->stripe_pages[chunk_page]; 907 } 908 909 /* 910 * number of pages we need for the entire stripe across all the 911 * drives 912 */ 913 static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes) 914 { 915 unsigned long nr = stripe_len * nr_stripes; 916 return (nr + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 917 } 918 919 /* 920 * allocation and initial setup for the btrfs_raid_bio. Not 921 * this does not allocate any pages for rbio->pages. 922 */ 923 static struct btrfs_raid_bio *alloc_rbio(struct btrfs_root *root, 924 struct btrfs_bio *bbio, u64 *raid_map, 925 u64 stripe_len) 926 { 927 struct btrfs_raid_bio *rbio; 928 int nr_data = 0; 929 int num_pages = rbio_nr_pages(stripe_len, bbio->num_stripes); 930 void *p; 931 932 rbio = kzalloc(sizeof(*rbio) + num_pages * sizeof(struct page *) * 2, 933 GFP_NOFS); 934 if (!rbio) { 935 kfree(raid_map); 936 kfree(bbio); 937 return ERR_PTR(-ENOMEM); 938 } 939 940 bio_list_init(&rbio->bio_list); 941 INIT_LIST_HEAD(&rbio->plug_list); 942 spin_lock_init(&rbio->bio_list_lock); 943 INIT_LIST_HEAD(&rbio->stripe_cache); 944 INIT_LIST_HEAD(&rbio->hash_list); 945 rbio->bbio = bbio; 946 rbio->raid_map = raid_map; 947 rbio->fs_info = root->fs_info; 948 rbio->stripe_len = stripe_len; 949 rbio->nr_pages = num_pages; 950 rbio->faila = -1; 951 rbio->failb = -1; 952 atomic_set(&rbio->refs, 1); 953 954 /* 955 * the stripe_pages and bio_pages array point to the extra 956 * memory we allocated past the end of the rbio 957 */ 958 p = rbio + 1; 959 rbio->stripe_pages = p; 960 rbio->bio_pages = p + sizeof(struct page *) * num_pages; 961 962 if (raid_map[bbio->num_stripes - 1] == RAID6_Q_STRIPE) 963 nr_data = bbio->num_stripes - 2; 964 else 965 nr_data = bbio->num_stripes - 1; 966 967 rbio->nr_data = nr_data; 968 return rbio; 969 } 970 971 /* allocate pages for all the stripes in the bio, including parity */ 972 static int alloc_rbio_pages(struct btrfs_raid_bio *rbio) 973 { 974 int i; 975 struct page *page; 976 977 for (i = 0; i < rbio->nr_pages; i++) { 978 if (rbio->stripe_pages[i]) 979 continue; 980 page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 981 if (!page) 982 return -ENOMEM; 983 rbio->stripe_pages[i] = page; 984 ClearPageUptodate(page); 985 } 986 return 0; 987 } 988 989 /* allocate pages for just the p/q stripes */ 990 static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio) 991 { 992 int i; 993 struct page *page; 994 995 i = (rbio->nr_data * rbio->stripe_len) >> PAGE_CACHE_SHIFT; 996 997 for (; i < rbio->nr_pages; i++) { 998 if (rbio->stripe_pages[i]) 999 continue; 1000 page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 1001 if (!page) 1002 return -ENOMEM; 1003 rbio->stripe_pages[i] = page; 1004 } 1005 return 0; 1006 } 1007 1008 /* 1009 * add a single page from a specific stripe into our list of bios for IO 1010 * this will try to merge into existing bios if possible, and returns 1011 * zero if all went well. 1012 */ 1013 static int rbio_add_io_page(struct btrfs_raid_bio *rbio, 1014 struct bio_list *bio_list, 1015 struct page *page, 1016 int stripe_nr, 1017 unsigned long page_index, 1018 unsigned long bio_max_len) 1019 { 1020 struct bio *last = bio_list->tail; 1021 u64 last_end = 0; 1022 int ret; 1023 struct bio *bio; 1024 struct btrfs_bio_stripe *stripe; 1025 u64 disk_start; 1026 1027 stripe = &rbio->bbio->stripes[stripe_nr]; 1028 disk_start = stripe->physical + (page_index << PAGE_CACHE_SHIFT); 1029 1030 /* if the device is missing, just fail this stripe */ 1031 if (!stripe->dev->bdev) 1032 return fail_rbio_index(rbio, stripe_nr); 1033 1034 /* see if we can add this page onto our existing bio */ 1035 if (last) { 1036 last_end = (u64)last->bi_sector << 9; 1037 last_end += last->bi_size; 1038 1039 /* 1040 * we can't merge these if they are from different 1041 * devices or if they are not contiguous 1042 */ 1043 if (last_end == disk_start && stripe->dev->bdev && 1044 test_bit(BIO_UPTODATE, &last->bi_flags) && 1045 last->bi_bdev == stripe->dev->bdev) { 1046 ret = bio_add_page(last, page, PAGE_CACHE_SIZE, 0); 1047 if (ret == PAGE_CACHE_SIZE) 1048 return 0; 1049 } 1050 } 1051 1052 /* put a new bio on the list */ 1053 bio = btrfs_io_bio_alloc(GFP_NOFS, bio_max_len >> PAGE_SHIFT?:1); 1054 if (!bio) 1055 return -ENOMEM; 1056 1057 bio->bi_size = 0; 1058 bio->bi_bdev = stripe->dev->bdev; 1059 bio->bi_sector = disk_start >> 9; 1060 set_bit(BIO_UPTODATE, &bio->bi_flags); 1061 1062 bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); 1063 bio_list_add(bio_list, bio); 1064 return 0; 1065 } 1066 1067 /* 1068 * while we're doing the read/modify/write cycle, we could 1069 * have errors in reading pages off the disk. This checks 1070 * for errors and if we're not able to read the page it'll 1071 * trigger parity reconstruction. The rmw will be finished 1072 * after we've reconstructed the failed stripes 1073 */ 1074 static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio) 1075 { 1076 if (rbio->faila >= 0 || rbio->failb >= 0) { 1077 BUG_ON(rbio->faila == rbio->bbio->num_stripes - 1); 1078 __raid56_parity_recover(rbio); 1079 } else { 1080 finish_rmw(rbio); 1081 } 1082 } 1083 1084 /* 1085 * these are just the pages from the rbio array, not from anything 1086 * the FS sent down to us 1087 */ 1088 static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe, int page) 1089 { 1090 int index; 1091 index = stripe * (rbio->stripe_len >> PAGE_CACHE_SHIFT); 1092 index += page; 1093 return rbio->stripe_pages[index]; 1094 } 1095 1096 /* 1097 * helper function to walk our bio list and populate the bio_pages array with 1098 * the result. This seems expensive, but it is faster than constantly 1099 * searching through the bio list as we setup the IO in finish_rmw or stripe 1100 * reconstruction. 1101 * 1102 * This must be called before you trust the answers from page_in_rbio 1103 */ 1104 static void index_rbio_pages(struct btrfs_raid_bio *rbio) 1105 { 1106 struct bio *bio; 1107 u64 start; 1108 unsigned long stripe_offset; 1109 unsigned long page_index; 1110 struct page *p; 1111 int i; 1112 1113 spin_lock_irq(&rbio->bio_list_lock); 1114 bio_list_for_each(bio, &rbio->bio_list) { 1115 start = (u64)bio->bi_sector << 9; 1116 stripe_offset = start - rbio->raid_map[0]; 1117 page_index = stripe_offset >> PAGE_CACHE_SHIFT; 1118 1119 for (i = 0; i < bio->bi_vcnt; i++) { 1120 p = bio->bi_io_vec[i].bv_page; 1121 rbio->bio_pages[page_index + i] = p; 1122 } 1123 } 1124 spin_unlock_irq(&rbio->bio_list_lock); 1125 } 1126 1127 /* 1128 * this is called from one of two situations. We either 1129 * have a full stripe from the higher layers, or we've read all 1130 * the missing bits off disk. 1131 * 1132 * This will calculate the parity and then send down any 1133 * changed blocks. 1134 */ 1135 static noinline void finish_rmw(struct btrfs_raid_bio *rbio) 1136 { 1137 struct btrfs_bio *bbio = rbio->bbio; 1138 void *pointers[bbio->num_stripes]; 1139 int stripe_len = rbio->stripe_len; 1140 int nr_data = rbio->nr_data; 1141 int stripe; 1142 int pagenr; 1143 int p_stripe = -1; 1144 int q_stripe = -1; 1145 struct bio_list bio_list; 1146 struct bio *bio; 1147 int pages_per_stripe = stripe_len >> PAGE_CACHE_SHIFT; 1148 int ret; 1149 1150 bio_list_init(&bio_list); 1151 1152 if (bbio->num_stripes - rbio->nr_data == 1) { 1153 p_stripe = bbio->num_stripes - 1; 1154 } else if (bbio->num_stripes - rbio->nr_data == 2) { 1155 p_stripe = bbio->num_stripes - 2; 1156 q_stripe = bbio->num_stripes - 1; 1157 } else { 1158 BUG(); 1159 } 1160 1161 /* at this point we either have a full stripe, 1162 * or we've read the full stripe from the drive. 1163 * recalculate the parity and write the new results. 1164 * 1165 * We're not allowed to add any new bios to the 1166 * bio list here, anyone else that wants to 1167 * change this stripe needs to do their own rmw. 1168 */ 1169 spin_lock_irq(&rbio->bio_list_lock); 1170 set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); 1171 spin_unlock_irq(&rbio->bio_list_lock); 1172 1173 atomic_set(&rbio->bbio->error, 0); 1174 1175 /* 1176 * now that we've set rmw_locked, run through the 1177 * bio list one last time and map the page pointers 1178 * 1179 * We don't cache full rbios because we're assuming 1180 * the higher layers are unlikely to use this area of 1181 * the disk again soon. If they do use it again, 1182 * hopefully they will send another full bio. 1183 */ 1184 index_rbio_pages(rbio); 1185 if (!rbio_is_full(rbio)) 1186 cache_rbio_pages(rbio); 1187 else 1188 clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); 1189 1190 for (pagenr = 0; pagenr < pages_per_stripe; pagenr++) { 1191 struct page *p; 1192 /* first collect one page from each data stripe */ 1193 for (stripe = 0; stripe < nr_data; stripe++) { 1194 p = page_in_rbio(rbio, stripe, pagenr, 0); 1195 pointers[stripe] = kmap(p); 1196 } 1197 1198 /* then add the parity stripe */ 1199 p = rbio_pstripe_page(rbio, pagenr); 1200 SetPageUptodate(p); 1201 pointers[stripe++] = kmap(p); 1202 1203 if (q_stripe != -1) { 1204 1205 /* 1206 * raid6, add the qstripe and call the 1207 * library function to fill in our p/q 1208 */ 1209 p = rbio_qstripe_page(rbio, pagenr); 1210 SetPageUptodate(p); 1211 pointers[stripe++] = kmap(p); 1212 1213 raid6_call.gen_syndrome(bbio->num_stripes, PAGE_SIZE, 1214 pointers); 1215 } else { 1216 /* raid5 */ 1217 memcpy(pointers[nr_data], pointers[0], PAGE_SIZE); 1218 run_xor(pointers + 1, nr_data - 1, PAGE_CACHE_SIZE); 1219 } 1220 1221 1222 for (stripe = 0; stripe < bbio->num_stripes; stripe++) 1223 kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); 1224 } 1225 1226 /* 1227 * time to start writing. Make bios for everything from the 1228 * higher layers (the bio_list in our rbio) and our p/q. Ignore 1229 * everything else. 1230 */ 1231 for (stripe = 0; stripe < bbio->num_stripes; stripe++) { 1232 for (pagenr = 0; pagenr < pages_per_stripe; pagenr++) { 1233 struct page *page; 1234 if (stripe < rbio->nr_data) { 1235 page = page_in_rbio(rbio, stripe, pagenr, 1); 1236 if (!page) 1237 continue; 1238 } else { 1239 page = rbio_stripe_page(rbio, stripe, pagenr); 1240 } 1241 1242 ret = rbio_add_io_page(rbio, &bio_list, 1243 page, stripe, pagenr, rbio->stripe_len); 1244 if (ret) 1245 goto cleanup; 1246 } 1247 } 1248 1249 atomic_set(&bbio->stripes_pending, bio_list_size(&bio_list)); 1250 BUG_ON(atomic_read(&bbio->stripes_pending) == 0); 1251 1252 while (1) { 1253 bio = bio_list_pop(&bio_list); 1254 if (!bio) 1255 break; 1256 1257 bio->bi_private = rbio; 1258 bio->bi_end_io = raid_write_end_io; 1259 BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags)); 1260 submit_bio(WRITE, bio); 1261 } 1262 return; 1263 1264 cleanup: 1265 rbio_orig_end_io(rbio, -EIO, 0); 1266 } 1267 1268 /* 1269 * helper to find the stripe number for a given bio. Used to figure out which 1270 * stripe has failed. This expects the bio to correspond to a physical disk, 1271 * so it looks up based on physical sector numbers. 1272 */ 1273 static int find_bio_stripe(struct btrfs_raid_bio *rbio, 1274 struct bio *bio) 1275 { 1276 u64 physical = bio->bi_sector; 1277 u64 stripe_start; 1278 int i; 1279 struct btrfs_bio_stripe *stripe; 1280 1281 physical <<= 9; 1282 1283 for (i = 0; i < rbio->bbio->num_stripes; i++) { 1284 stripe = &rbio->bbio->stripes[i]; 1285 stripe_start = stripe->physical; 1286 if (physical >= stripe_start && 1287 physical < stripe_start + rbio->stripe_len) { 1288 return i; 1289 } 1290 } 1291 return -1; 1292 } 1293 1294 /* 1295 * helper to find the stripe number for a given 1296 * bio (before mapping). Used to figure out which stripe has 1297 * failed. This looks up based on logical block numbers. 1298 */ 1299 static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio, 1300 struct bio *bio) 1301 { 1302 u64 logical = bio->bi_sector; 1303 u64 stripe_start; 1304 int i; 1305 1306 logical <<= 9; 1307 1308 for (i = 0; i < rbio->nr_data; i++) { 1309 stripe_start = rbio->raid_map[i]; 1310 if (logical >= stripe_start && 1311 logical < stripe_start + rbio->stripe_len) { 1312 return i; 1313 } 1314 } 1315 return -1; 1316 } 1317 1318 /* 1319 * returns -EIO if we had too many failures 1320 */ 1321 static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed) 1322 { 1323 unsigned long flags; 1324 int ret = 0; 1325 1326 spin_lock_irqsave(&rbio->bio_list_lock, flags); 1327 1328 /* we already know this stripe is bad, move on */ 1329 if (rbio->faila == failed || rbio->failb == failed) 1330 goto out; 1331 1332 if (rbio->faila == -1) { 1333 /* first failure on this rbio */ 1334 rbio->faila = failed; 1335 atomic_inc(&rbio->bbio->error); 1336 } else if (rbio->failb == -1) { 1337 /* second failure on this rbio */ 1338 rbio->failb = failed; 1339 atomic_inc(&rbio->bbio->error); 1340 } else { 1341 ret = -EIO; 1342 } 1343 out: 1344 spin_unlock_irqrestore(&rbio->bio_list_lock, flags); 1345 1346 return ret; 1347 } 1348 1349 /* 1350 * helper to fail a stripe based on a physical disk 1351 * bio. 1352 */ 1353 static int fail_bio_stripe(struct btrfs_raid_bio *rbio, 1354 struct bio *bio) 1355 { 1356 int failed = find_bio_stripe(rbio, bio); 1357 1358 if (failed < 0) 1359 return -EIO; 1360 1361 return fail_rbio_index(rbio, failed); 1362 } 1363 1364 /* 1365 * this sets each page in the bio uptodate. It should only be used on private 1366 * rbio pages, nothing that comes in from the higher layers 1367 */ 1368 static void set_bio_pages_uptodate(struct bio *bio) 1369 { 1370 int i; 1371 struct page *p; 1372 1373 for (i = 0; i < bio->bi_vcnt; i++) { 1374 p = bio->bi_io_vec[i].bv_page; 1375 SetPageUptodate(p); 1376 } 1377 } 1378 1379 /* 1380 * end io for the read phase of the rmw cycle. All the bios here are physical 1381 * stripe bios we've read from the disk so we can recalculate the parity of the 1382 * stripe. 1383 * 1384 * This will usually kick off finish_rmw once all the bios are read in, but it 1385 * may trigger parity reconstruction if we had any errors along the way 1386 */ 1387 static void raid_rmw_end_io(struct bio *bio, int err) 1388 { 1389 struct btrfs_raid_bio *rbio = bio->bi_private; 1390 1391 if (err) 1392 fail_bio_stripe(rbio, bio); 1393 else 1394 set_bio_pages_uptodate(bio); 1395 1396 bio_put(bio); 1397 1398 if (!atomic_dec_and_test(&rbio->bbio->stripes_pending)) 1399 return; 1400 1401 err = 0; 1402 if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors) 1403 goto cleanup; 1404 1405 /* 1406 * this will normally call finish_rmw to start our write 1407 * but if there are any failed stripes we'll reconstruct 1408 * from parity first 1409 */ 1410 validate_rbio_for_rmw(rbio); 1411 return; 1412 1413 cleanup: 1414 1415 rbio_orig_end_io(rbio, -EIO, 0); 1416 } 1417 1418 static void async_rmw_stripe(struct btrfs_raid_bio *rbio) 1419 { 1420 rbio->work.flags = 0; 1421 rbio->work.func = rmw_work; 1422 1423 btrfs_queue_worker(&rbio->fs_info->rmw_workers, 1424 &rbio->work); 1425 } 1426 1427 static void async_read_rebuild(struct btrfs_raid_bio *rbio) 1428 { 1429 rbio->work.flags = 0; 1430 rbio->work.func = read_rebuild_work; 1431 1432 btrfs_queue_worker(&rbio->fs_info->rmw_workers, 1433 &rbio->work); 1434 } 1435 1436 /* 1437 * the stripe must be locked by the caller. It will 1438 * unlock after all the writes are done 1439 */ 1440 static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio) 1441 { 1442 int bios_to_read = 0; 1443 struct btrfs_bio *bbio = rbio->bbio; 1444 struct bio_list bio_list; 1445 int ret; 1446 int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1447 int pagenr; 1448 int stripe; 1449 struct bio *bio; 1450 1451 bio_list_init(&bio_list); 1452 1453 ret = alloc_rbio_pages(rbio); 1454 if (ret) 1455 goto cleanup; 1456 1457 index_rbio_pages(rbio); 1458 1459 atomic_set(&rbio->bbio->error, 0); 1460 /* 1461 * build a list of bios to read all the missing parts of this 1462 * stripe 1463 */ 1464 for (stripe = 0; stripe < rbio->nr_data; stripe++) { 1465 for (pagenr = 0; pagenr < nr_pages; pagenr++) { 1466 struct page *page; 1467 /* 1468 * we want to find all the pages missing from 1469 * the rbio and read them from the disk. If 1470 * page_in_rbio finds a page in the bio list 1471 * we don't need to read it off the stripe. 1472 */ 1473 page = page_in_rbio(rbio, stripe, pagenr, 1); 1474 if (page) 1475 continue; 1476 1477 page = rbio_stripe_page(rbio, stripe, pagenr); 1478 /* 1479 * the bio cache may have handed us an uptodate 1480 * page. If so, be happy and use it 1481 */ 1482 if (PageUptodate(page)) 1483 continue; 1484 1485 ret = rbio_add_io_page(rbio, &bio_list, page, 1486 stripe, pagenr, rbio->stripe_len); 1487 if (ret) 1488 goto cleanup; 1489 } 1490 } 1491 1492 bios_to_read = bio_list_size(&bio_list); 1493 if (!bios_to_read) { 1494 /* 1495 * this can happen if others have merged with 1496 * us, it means there is nothing left to read. 1497 * But if there are missing devices it may not be 1498 * safe to do the full stripe write yet. 1499 */ 1500 goto finish; 1501 } 1502 1503 /* 1504 * the bbio may be freed once we submit the last bio. Make sure 1505 * not to touch it after that 1506 */ 1507 atomic_set(&bbio->stripes_pending, bios_to_read); 1508 while (1) { 1509 bio = bio_list_pop(&bio_list); 1510 if (!bio) 1511 break; 1512 1513 bio->bi_private = rbio; 1514 bio->bi_end_io = raid_rmw_end_io; 1515 1516 btrfs_bio_wq_end_io(rbio->fs_info, bio, 1517 BTRFS_WQ_ENDIO_RAID56); 1518 1519 BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags)); 1520 submit_bio(READ, bio); 1521 } 1522 /* the actual write will happen once the reads are done */ 1523 return 0; 1524 1525 cleanup: 1526 rbio_orig_end_io(rbio, -EIO, 0); 1527 return -EIO; 1528 1529 finish: 1530 validate_rbio_for_rmw(rbio); 1531 return 0; 1532 } 1533 1534 /* 1535 * if the upper layers pass in a full stripe, we thank them by only allocating 1536 * enough pages to hold the parity, and sending it all down quickly. 1537 */ 1538 static int full_stripe_write(struct btrfs_raid_bio *rbio) 1539 { 1540 int ret; 1541 1542 ret = alloc_rbio_parity_pages(rbio); 1543 if (ret) 1544 return ret; 1545 1546 ret = lock_stripe_add(rbio); 1547 if (ret == 0) 1548 finish_rmw(rbio); 1549 return 0; 1550 } 1551 1552 /* 1553 * partial stripe writes get handed over to async helpers. 1554 * We're really hoping to merge a few more writes into this 1555 * rbio before calculating new parity 1556 */ 1557 static int partial_stripe_write(struct btrfs_raid_bio *rbio) 1558 { 1559 int ret; 1560 1561 ret = lock_stripe_add(rbio); 1562 if (ret == 0) 1563 async_rmw_stripe(rbio); 1564 return 0; 1565 } 1566 1567 /* 1568 * sometimes while we were reading from the drive to 1569 * recalculate parity, enough new bios come into create 1570 * a full stripe. So we do a check here to see if we can 1571 * go directly to finish_rmw 1572 */ 1573 static int __raid56_parity_write(struct btrfs_raid_bio *rbio) 1574 { 1575 /* head off into rmw land if we don't have a full stripe */ 1576 if (!rbio_is_full(rbio)) 1577 return partial_stripe_write(rbio); 1578 return full_stripe_write(rbio); 1579 } 1580 1581 /* 1582 * We use plugging call backs to collect full stripes. 1583 * Any time we get a partial stripe write while plugged 1584 * we collect it into a list. When the unplug comes down, 1585 * we sort the list by logical block number and merge 1586 * everything we can into the same rbios 1587 */ 1588 struct btrfs_plug_cb { 1589 struct blk_plug_cb cb; 1590 struct btrfs_fs_info *info; 1591 struct list_head rbio_list; 1592 struct btrfs_work work; 1593 }; 1594 1595 /* 1596 * rbios on the plug list are sorted for easier merging. 1597 */ 1598 static int plug_cmp(void *priv, struct list_head *a, struct list_head *b) 1599 { 1600 struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio, 1601 plug_list); 1602 struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio, 1603 plug_list); 1604 u64 a_sector = ra->bio_list.head->bi_sector; 1605 u64 b_sector = rb->bio_list.head->bi_sector; 1606 1607 if (a_sector < b_sector) 1608 return -1; 1609 if (a_sector > b_sector) 1610 return 1; 1611 return 0; 1612 } 1613 1614 static void run_plug(struct btrfs_plug_cb *plug) 1615 { 1616 struct btrfs_raid_bio *cur; 1617 struct btrfs_raid_bio *last = NULL; 1618 1619 /* 1620 * sort our plug list then try to merge 1621 * everything we can in hopes of creating full 1622 * stripes. 1623 */ 1624 list_sort(NULL, &plug->rbio_list, plug_cmp); 1625 while (!list_empty(&plug->rbio_list)) { 1626 cur = list_entry(plug->rbio_list.next, 1627 struct btrfs_raid_bio, plug_list); 1628 list_del_init(&cur->plug_list); 1629 1630 if (rbio_is_full(cur)) { 1631 /* we have a full stripe, send it down */ 1632 full_stripe_write(cur); 1633 continue; 1634 } 1635 if (last) { 1636 if (rbio_can_merge(last, cur)) { 1637 merge_rbio(last, cur); 1638 __free_raid_bio(cur); 1639 continue; 1640 1641 } 1642 __raid56_parity_write(last); 1643 } 1644 last = cur; 1645 } 1646 if (last) { 1647 __raid56_parity_write(last); 1648 } 1649 kfree(plug); 1650 } 1651 1652 /* 1653 * if the unplug comes from schedule, we have to push the 1654 * work off to a helper thread 1655 */ 1656 static void unplug_work(struct btrfs_work *work) 1657 { 1658 struct btrfs_plug_cb *plug; 1659 plug = container_of(work, struct btrfs_plug_cb, work); 1660 run_plug(plug); 1661 } 1662 1663 static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule) 1664 { 1665 struct btrfs_plug_cb *plug; 1666 plug = container_of(cb, struct btrfs_plug_cb, cb); 1667 1668 if (from_schedule) { 1669 plug->work.flags = 0; 1670 plug->work.func = unplug_work; 1671 btrfs_queue_worker(&plug->info->rmw_workers, 1672 &plug->work); 1673 return; 1674 } 1675 run_plug(plug); 1676 } 1677 1678 /* 1679 * our main entry point for writes from the rest of the FS. 1680 */ 1681 int raid56_parity_write(struct btrfs_root *root, struct bio *bio, 1682 struct btrfs_bio *bbio, u64 *raid_map, 1683 u64 stripe_len) 1684 { 1685 struct btrfs_raid_bio *rbio; 1686 struct btrfs_plug_cb *plug = NULL; 1687 struct blk_plug_cb *cb; 1688 1689 rbio = alloc_rbio(root, bbio, raid_map, stripe_len); 1690 if (IS_ERR(rbio)) { 1691 kfree(raid_map); 1692 kfree(bbio); 1693 return PTR_ERR(rbio); 1694 } 1695 bio_list_add(&rbio->bio_list, bio); 1696 rbio->bio_list_bytes = bio->bi_size; 1697 1698 /* 1699 * don't plug on full rbios, just get them out the door 1700 * as quickly as we can 1701 */ 1702 if (rbio_is_full(rbio)) 1703 return full_stripe_write(rbio); 1704 1705 cb = blk_check_plugged(btrfs_raid_unplug, root->fs_info, 1706 sizeof(*plug)); 1707 if (cb) { 1708 plug = container_of(cb, struct btrfs_plug_cb, cb); 1709 if (!plug->info) { 1710 plug->info = root->fs_info; 1711 INIT_LIST_HEAD(&plug->rbio_list); 1712 } 1713 list_add_tail(&rbio->plug_list, &plug->rbio_list); 1714 } else { 1715 return __raid56_parity_write(rbio); 1716 } 1717 return 0; 1718 } 1719 1720 /* 1721 * all parity reconstruction happens here. We've read in everything 1722 * we can find from the drives and this does the heavy lifting of 1723 * sorting the good from the bad. 1724 */ 1725 static void __raid_recover_end_io(struct btrfs_raid_bio *rbio) 1726 { 1727 int pagenr, stripe; 1728 void **pointers; 1729 int faila = -1, failb = -1; 1730 int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1731 struct page *page; 1732 int err; 1733 int i; 1734 1735 pointers = kzalloc(rbio->bbio->num_stripes * sizeof(void *), 1736 GFP_NOFS); 1737 if (!pointers) { 1738 err = -ENOMEM; 1739 goto cleanup_io; 1740 } 1741 1742 faila = rbio->faila; 1743 failb = rbio->failb; 1744 1745 if (rbio->read_rebuild) { 1746 spin_lock_irq(&rbio->bio_list_lock); 1747 set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); 1748 spin_unlock_irq(&rbio->bio_list_lock); 1749 } 1750 1751 index_rbio_pages(rbio); 1752 1753 for (pagenr = 0; pagenr < nr_pages; pagenr++) { 1754 /* setup our array of pointers with pages 1755 * from each stripe 1756 */ 1757 for (stripe = 0; stripe < rbio->bbio->num_stripes; stripe++) { 1758 /* 1759 * if we're rebuilding a read, we have to use 1760 * pages from the bio list 1761 */ 1762 if (rbio->read_rebuild && 1763 (stripe == faila || stripe == failb)) { 1764 page = page_in_rbio(rbio, stripe, pagenr, 0); 1765 } else { 1766 page = rbio_stripe_page(rbio, stripe, pagenr); 1767 } 1768 pointers[stripe] = kmap(page); 1769 } 1770 1771 /* all raid6 handling here */ 1772 if (rbio->raid_map[rbio->bbio->num_stripes - 1] == 1773 RAID6_Q_STRIPE) { 1774 1775 /* 1776 * single failure, rebuild from parity raid5 1777 * style 1778 */ 1779 if (failb < 0) { 1780 if (faila == rbio->nr_data) { 1781 /* 1782 * Just the P stripe has failed, without 1783 * a bad data or Q stripe. 1784 * TODO, we should redo the xor here. 1785 */ 1786 err = -EIO; 1787 goto cleanup; 1788 } 1789 /* 1790 * a single failure in raid6 is rebuilt 1791 * in the pstripe code below 1792 */ 1793 goto pstripe; 1794 } 1795 1796 /* make sure our ps and qs are in order */ 1797 if (faila > failb) { 1798 int tmp = failb; 1799 failb = faila; 1800 faila = tmp; 1801 } 1802 1803 /* if the q stripe is failed, do a pstripe reconstruction 1804 * from the xors. 1805 * If both the q stripe and the P stripe are failed, we're 1806 * here due to a crc mismatch and we can't give them the 1807 * data they want 1808 */ 1809 if (rbio->raid_map[failb] == RAID6_Q_STRIPE) { 1810 if (rbio->raid_map[faila] == RAID5_P_STRIPE) { 1811 err = -EIO; 1812 goto cleanup; 1813 } 1814 /* 1815 * otherwise we have one bad data stripe and 1816 * a good P stripe. raid5! 1817 */ 1818 goto pstripe; 1819 } 1820 1821 if (rbio->raid_map[failb] == RAID5_P_STRIPE) { 1822 raid6_datap_recov(rbio->bbio->num_stripes, 1823 PAGE_SIZE, faila, pointers); 1824 } else { 1825 raid6_2data_recov(rbio->bbio->num_stripes, 1826 PAGE_SIZE, faila, failb, 1827 pointers); 1828 } 1829 } else { 1830 void *p; 1831 1832 /* rebuild from P stripe here (raid5 or raid6) */ 1833 BUG_ON(failb != -1); 1834 pstripe: 1835 /* Copy parity block into failed block to start with */ 1836 memcpy(pointers[faila], 1837 pointers[rbio->nr_data], 1838 PAGE_CACHE_SIZE); 1839 1840 /* rearrange the pointer array */ 1841 p = pointers[faila]; 1842 for (stripe = faila; stripe < rbio->nr_data - 1; stripe++) 1843 pointers[stripe] = pointers[stripe + 1]; 1844 pointers[rbio->nr_data - 1] = p; 1845 1846 /* xor in the rest */ 1847 run_xor(pointers, rbio->nr_data - 1, PAGE_CACHE_SIZE); 1848 } 1849 /* if we're doing this rebuild as part of an rmw, go through 1850 * and set all of our private rbio pages in the 1851 * failed stripes as uptodate. This way finish_rmw will 1852 * know they can be trusted. If this was a read reconstruction, 1853 * other endio functions will fiddle the uptodate bits 1854 */ 1855 if (!rbio->read_rebuild) { 1856 for (i = 0; i < nr_pages; i++) { 1857 if (faila != -1) { 1858 page = rbio_stripe_page(rbio, faila, i); 1859 SetPageUptodate(page); 1860 } 1861 if (failb != -1) { 1862 page = rbio_stripe_page(rbio, failb, i); 1863 SetPageUptodate(page); 1864 } 1865 } 1866 } 1867 for (stripe = 0; stripe < rbio->bbio->num_stripes; stripe++) { 1868 /* 1869 * if we're rebuilding a read, we have to use 1870 * pages from the bio list 1871 */ 1872 if (rbio->read_rebuild && 1873 (stripe == faila || stripe == failb)) { 1874 page = page_in_rbio(rbio, stripe, pagenr, 0); 1875 } else { 1876 page = rbio_stripe_page(rbio, stripe, pagenr); 1877 } 1878 kunmap(page); 1879 } 1880 } 1881 1882 err = 0; 1883 cleanup: 1884 kfree(pointers); 1885 1886 cleanup_io: 1887 1888 if (rbio->read_rebuild) { 1889 if (err == 0) 1890 cache_rbio_pages(rbio); 1891 else 1892 clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); 1893 1894 rbio_orig_end_io(rbio, err, err == 0); 1895 } else if (err == 0) { 1896 rbio->faila = -1; 1897 rbio->failb = -1; 1898 finish_rmw(rbio); 1899 } else { 1900 rbio_orig_end_io(rbio, err, 0); 1901 } 1902 } 1903 1904 /* 1905 * This is called only for stripes we've read from disk to 1906 * reconstruct the parity. 1907 */ 1908 static void raid_recover_end_io(struct bio *bio, int err) 1909 { 1910 struct btrfs_raid_bio *rbio = bio->bi_private; 1911 1912 /* 1913 * we only read stripe pages off the disk, set them 1914 * up to date if there were no errors 1915 */ 1916 if (err) 1917 fail_bio_stripe(rbio, bio); 1918 else 1919 set_bio_pages_uptodate(bio); 1920 bio_put(bio); 1921 1922 if (!atomic_dec_and_test(&rbio->bbio->stripes_pending)) 1923 return; 1924 1925 if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors) 1926 rbio_orig_end_io(rbio, -EIO, 0); 1927 else 1928 __raid_recover_end_io(rbio); 1929 } 1930 1931 /* 1932 * reads everything we need off the disk to reconstruct 1933 * the parity. endio handlers trigger final reconstruction 1934 * when the IO is done. 1935 * 1936 * This is used both for reads from the higher layers and for 1937 * parity construction required to finish a rmw cycle. 1938 */ 1939 static int __raid56_parity_recover(struct btrfs_raid_bio *rbio) 1940 { 1941 int bios_to_read = 0; 1942 struct btrfs_bio *bbio = rbio->bbio; 1943 struct bio_list bio_list; 1944 int ret; 1945 int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1946 int pagenr; 1947 int stripe; 1948 struct bio *bio; 1949 1950 bio_list_init(&bio_list); 1951 1952 ret = alloc_rbio_pages(rbio); 1953 if (ret) 1954 goto cleanup; 1955 1956 atomic_set(&rbio->bbio->error, 0); 1957 1958 /* 1959 * read everything that hasn't failed. Thanks to the 1960 * stripe cache, it is possible that some or all of these 1961 * pages are going to be uptodate. 1962 */ 1963 for (stripe = 0; stripe < bbio->num_stripes; stripe++) { 1964 if (rbio->faila == stripe || 1965 rbio->failb == stripe) 1966 continue; 1967 1968 for (pagenr = 0; pagenr < nr_pages; pagenr++) { 1969 struct page *p; 1970 1971 /* 1972 * the rmw code may have already read this 1973 * page in 1974 */ 1975 p = rbio_stripe_page(rbio, stripe, pagenr); 1976 if (PageUptodate(p)) 1977 continue; 1978 1979 ret = rbio_add_io_page(rbio, &bio_list, 1980 rbio_stripe_page(rbio, stripe, pagenr), 1981 stripe, pagenr, rbio->stripe_len); 1982 if (ret < 0) 1983 goto cleanup; 1984 } 1985 } 1986 1987 bios_to_read = bio_list_size(&bio_list); 1988 if (!bios_to_read) { 1989 /* 1990 * we might have no bios to read just because the pages 1991 * were up to date, or we might have no bios to read because 1992 * the devices were gone. 1993 */ 1994 if (atomic_read(&rbio->bbio->error) <= rbio->bbio->max_errors) { 1995 __raid_recover_end_io(rbio); 1996 goto out; 1997 } else { 1998 goto cleanup; 1999 } 2000 } 2001 2002 /* 2003 * the bbio may be freed once we submit the last bio. Make sure 2004 * not to touch it after that 2005 */ 2006 atomic_set(&bbio->stripes_pending, bios_to_read); 2007 while (1) { 2008 bio = bio_list_pop(&bio_list); 2009 if (!bio) 2010 break; 2011 2012 bio->bi_private = rbio; 2013 bio->bi_end_io = raid_recover_end_io; 2014 2015 btrfs_bio_wq_end_io(rbio->fs_info, bio, 2016 BTRFS_WQ_ENDIO_RAID56); 2017 2018 BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags)); 2019 submit_bio(READ, bio); 2020 } 2021 out: 2022 return 0; 2023 2024 cleanup: 2025 if (rbio->read_rebuild) 2026 rbio_orig_end_io(rbio, -EIO, 0); 2027 return -EIO; 2028 } 2029 2030 /* 2031 * the main entry point for reads from the higher layers. This 2032 * is really only called when the normal read path had a failure, 2033 * so we assume the bio they send down corresponds to a failed part 2034 * of the drive. 2035 */ 2036 int raid56_parity_recover(struct btrfs_root *root, struct bio *bio, 2037 struct btrfs_bio *bbio, u64 *raid_map, 2038 u64 stripe_len, int mirror_num) 2039 { 2040 struct btrfs_raid_bio *rbio; 2041 int ret; 2042 2043 rbio = alloc_rbio(root, bbio, raid_map, stripe_len); 2044 if (IS_ERR(rbio)) { 2045 return PTR_ERR(rbio); 2046 } 2047 2048 rbio->read_rebuild = 1; 2049 bio_list_add(&rbio->bio_list, bio); 2050 rbio->bio_list_bytes = bio->bi_size; 2051 2052 rbio->faila = find_logical_bio_stripe(rbio, bio); 2053 if (rbio->faila == -1) { 2054 BUG(); 2055 kfree(rbio); 2056 return -EIO; 2057 } 2058 2059 /* 2060 * reconstruct from the q stripe if they are 2061 * asking for mirror 3 2062 */ 2063 if (mirror_num == 3) 2064 rbio->failb = bbio->num_stripes - 2; 2065 2066 ret = lock_stripe_add(rbio); 2067 2068 /* 2069 * __raid56_parity_recover will end the bio with 2070 * any errors it hits. We don't want to return 2071 * its error value up the stack because our caller 2072 * will end up calling bio_endio with any nonzero 2073 * return 2074 */ 2075 if (ret == 0) 2076 __raid56_parity_recover(rbio); 2077 /* 2078 * our rbio has been added to the list of 2079 * rbios that will be handled after the 2080 * currently lock owner is done 2081 */ 2082 return 0; 2083 2084 } 2085 2086 static void rmw_work(struct btrfs_work *work) 2087 { 2088 struct btrfs_raid_bio *rbio; 2089 2090 rbio = container_of(work, struct btrfs_raid_bio, work); 2091 raid56_rmw_stripe(rbio); 2092 } 2093 2094 static void read_rebuild_work(struct btrfs_work *work) 2095 { 2096 struct btrfs_raid_bio *rbio; 2097 2098 rbio = container_of(work, struct btrfs_raid_bio, work); 2099 __raid56_parity_recover(rbio); 2100 } 2101