1 // SPDX-License-Identifier: GPL-2.0 2 3 #include <linux/err.h> 4 #include <linux/slab.h> 5 #include <linux/spinlock.h> 6 #include "messages.h" 7 #include "ctree.h" 8 #include "volumes.h" 9 #include "extent_map.h" 10 #include "compression.h" 11 #include "btrfs_inode.h" 12 13 14 static struct kmem_cache *extent_map_cache; 15 16 int __init extent_map_init(void) 17 { 18 extent_map_cache = kmem_cache_create("btrfs_extent_map", 19 sizeof(struct extent_map), 0, 20 SLAB_MEM_SPREAD, NULL); 21 if (!extent_map_cache) 22 return -ENOMEM; 23 return 0; 24 } 25 26 void __cold extent_map_exit(void) 27 { 28 kmem_cache_destroy(extent_map_cache); 29 } 30 31 /* 32 * Initialize the extent tree @tree. Should be called for each new inode or 33 * other user of the extent_map interface. 34 */ 35 void extent_map_tree_init(struct extent_map_tree *tree) 36 { 37 tree->map = RB_ROOT_CACHED; 38 INIT_LIST_HEAD(&tree->modified_extents); 39 rwlock_init(&tree->lock); 40 } 41 42 /* 43 * Allocate a new extent_map structure. The new structure is returned with a 44 * reference count of one and needs to be freed using free_extent_map() 45 */ 46 struct extent_map *alloc_extent_map(void) 47 { 48 struct extent_map *em; 49 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS); 50 if (!em) 51 return NULL; 52 RB_CLEAR_NODE(&em->rb_node); 53 em->compress_type = BTRFS_COMPRESS_NONE; 54 refcount_set(&em->refs, 1); 55 INIT_LIST_HEAD(&em->list); 56 return em; 57 } 58 59 /* 60 * Drop the reference out on @em by one and free the structure if the reference 61 * count hits zero. 62 */ 63 void free_extent_map(struct extent_map *em) 64 { 65 if (!em) 66 return; 67 if (refcount_dec_and_test(&em->refs)) { 68 WARN_ON(extent_map_in_tree(em)); 69 WARN_ON(!list_empty(&em->list)); 70 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags)) 71 kfree(em->map_lookup); 72 kmem_cache_free(extent_map_cache, em); 73 } 74 } 75 76 /* Do the math around the end of an extent, handling wrapping. */ 77 static u64 range_end(u64 start, u64 len) 78 { 79 if (start + len < start) 80 return (u64)-1; 81 return start + len; 82 } 83 84 static int tree_insert(struct rb_root_cached *root, struct extent_map *em) 85 { 86 struct rb_node **p = &root->rb_root.rb_node; 87 struct rb_node *parent = NULL; 88 struct extent_map *entry = NULL; 89 struct rb_node *orig_parent = NULL; 90 u64 end = range_end(em->start, em->len); 91 bool leftmost = true; 92 93 while (*p) { 94 parent = *p; 95 entry = rb_entry(parent, struct extent_map, rb_node); 96 97 if (em->start < entry->start) { 98 p = &(*p)->rb_left; 99 } else if (em->start >= extent_map_end(entry)) { 100 p = &(*p)->rb_right; 101 leftmost = false; 102 } else { 103 return -EEXIST; 104 } 105 } 106 107 orig_parent = parent; 108 while (parent && em->start >= extent_map_end(entry)) { 109 parent = rb_next(parent); 110 entry = rb_entry(parent, struct extent_map, rb_node); 111 } 112 if (parent) 113 if (end > entry->start && em->start < extent_map_end(entry)) 114 return -EEXIST; 115 116 parent = orig_parent; 117 entry = rb_entry(parent, struct extent_map, rb_node); 118 while (parent && em->start < entry->start) { 119 parent = rb_prev(parent); 120 entry = rb_entry(parent, struct extent_map, rb_node); 121 } 122 if (parent) 123 if (end > entry->start && em->start < extent_map_end(entry)) 124 return -EEXIST; 125 126 rb_link_node(&em->rb_node, orig_parent, p); 127 rb_insert_color_cached(&em->rb_node, root, leftmost); 128 return 0; 129 } 130 131 /* 132 * Search through the tree for an extent_map with a given offset. If it can't 133 * be found, try to find some neighboring extents 134 */ 135 static struct rb_node *__tree_search(struct rb_root *root, u64 offset, 136 struct rb_node **prev_or_next_ret) 137 { 138 struct rb_node *n = root->rb_node; 139 struct rb_node *prev = NULL; 140 struct rb_node *orig_prev = NULL; 141 struct extent_map *entry; 142 struct extent_map *prev_entry = NULL; 143 144 ASSERT(prev_or_next_ret); 145 146 while (n) { 147 entry = rb_entry(n, struct extent_map, rb_node); 148 prev = n; 149 prev_entry = entry; 150 151 if (offset < entry->start) 152 n = n->rb_left; 153 else if (offset >= extent_map_end(entry)) 154 n = n->rb_right; 155 else 156 return n; 157 } 158 159 orig_prev = prev; 160 while (prev && offset >= extent_map_end(prev_entry)) { 161 prev = rb_next(prev); 162 prev_entry = rb_entry(prev, struct extent_map, rb_node); 163 } 164 165 /* 166 * Previous extent map found, return as in this case the caller does not 167 * care about the next one. 168 */ 169 if (prev) { 170 *prev_or_next_ret = prev; 171 return NULL; 172 } 173 174 prev = orig_prev; 175 prev_entry = rb_entry(prev, struct extent_map, rb_node); 176 while (prev && offset < prev_entry->start) { 177 prev = rb_prev(prev); 178 prev_entry = rb_entry(prev, struct extent_map, rb_node); 179 } 180 *prev_or_next_ret = prev; 181 182 return NULL; 183 } 184 185 /* Check to see if two extent_map structs are adjacent and safe to merge. */ 186 static int mergable_maps(struct extent_map *prev, struct extent_map *next) 187 { 188 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags)) 189 return 0; 190 191 /* 192 * don't merge compressed extents, we need to know their 193 * actual size 194 */ 195 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags)) 196 return 0; 197 198 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) || 199 test_bit(EXTENT_FLAG_LOGGING, &next->flags)) 200 return 0; 201 202 /* 203 * We don't want to merge stuff that hasn't been written to the log yet 204 * since it may not reflect exactly what is on disk, and that would be 205 * bad. 206 */ 207 if (!list_empty(&prev->list) || !list_empty(&next->list)) 208 return 0; 209 210 ASSERT(next->block_start != EXTENT_MAP_DELALLOC && 211 prev->block_start != EXTENT_MAP_DELALLOC); 212 213 if (prev->map_lookup || next->map_lookup) 214 ASSERT(test_bit(EXTENT_FLAG_FS_MAPPING, &prev->flags) && 215 test_bit(EXTENT_FLAG_FS_MAPPING, &next->flags)); 216 217 if (extent_map_end(prev) == next->start && 218 prev->flags == next->flags && 219 prev->map_lookup == next->map_lookup && 220 ((next->block_start == EXTENT_MAP_HOLE && 221 prev->block_start == EXTENT_MAP_HOLE) || 222 (next->block_start == EXTENT_MAP_INLINE && 223 prev->block_start == EXTENT_MAP_INLINE) || 224 (next->block_start < EXTENT_MAP_LAST_BYTE - 1 && 225 next->block_start == extent_map_block_end(prev)))) { 226 return 1; 227 } 228 return 0; 229 } 230 231 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em) 232 { 233 struct extent_map *merge = NULL; 234 struct rb_node *rb; 235 236 /* 237 * We can't modify an extent map that is in the tree and that is being 238 * used by another task, as it can cause that other task to see it in 239 * inconsistent state during the merging. We always have 1 reference for 240 * the tree and 1 for this task (which is unpinning the extent map or 241 * clearing the logging flag), so anything > 2 means it's being used by 242 * other tasks too. 243 */ 244 if (refcount_read(&em->refs) > 2) 245 return; 246 247 if (em->start != 0) { 248 rb = rb_prev(&em->rb_node); 249 if (rb) 250 merge = rb_entry(rb, struct extent_map, rb_node); 251 if (rb && mergable_maps(merge, em)) { 252 em->start = merge->start; 253 em->orig_start = merge->orig_start; 254 em->len += merge->len; 255 em->block_len += merge->block_len; 256 em->block_start = merge->block_start; 257 em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start; 258 em->mod_start = merge->mod_start; 259 em->generation = max(em->generation, merge->generation); 260 set_bit(EXTENT_FLAG_MERGED, &em->flags); 261 262 rb_erase_cached(&merge->rb_node, &tree->map); 263 RB_CLEAR_NODE(&merge->rb_node); 264 free_extent_map(merge); 265 } 266 } 267 268 rb = rb_next(&em->rb_node); 269 if (rb) 270 merge = rb_entry(rb, struct extent_map, rb_node); 271 if (rb && mergable_maps(em, merge)) { 272 em->len += merge->len; 273 em->block_len += merge->block_len; 274 rb_erase_cached(&merge->rb_node, &tree->map); 275 RB_CLEAR_NODE(&merge->rb_node); 276 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start; 277 em->generation = max(em->generation, merge->generation); 278 set_bit(EXTENT_FLAG_MERGED, &em->flags); 279 free_extent_map(merge); 280 } 281 } 282 283 /* 284 * Unpin an extent from the cache. 285 * 286 * @tree: tree to unpin the extent in 287 * @start: logical offset in the file 288 * @len: length of the extent 289 * @gen: generation that this extent has been modified in 290 * 291 * Called after an extent has been written to disk properly. Set the generation 292 * to the generation that actually added the file item to the inode so we know 293 * we need to sync this extent when we call fsync(). 294 */ 295 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len, 296 u64 gen) 297 { 298 int ret = 0; 299 struct extent_map *em; 300 bool prealloc = false; 301 302 write_lock(&tree->lock); 303 em = lookup_extent_mapping(tree, start, len); 304 305 WARN_ON(!em || em->start != start); 306 307 if (!em) 308 goto out; 309 310 em->generation = gen; 311 clear_bit(EXTENT_FLAG_PINNED, &em->flags); 312 em->mod_start = em->start; 313 em->mod_len = em->len; 314 315 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) { 316 prealloc = true; 317 clear_bit(EXTENT_FLAG_FILLING, &em->flags); 318 } 319 320 try_merge_map(tree, em); 321 322 if (prealloc) { 323 em->mod_start = em->start; 324 em->mod_len = em->len; 325 } 326 327 free_extent_map(em); 328 out: 329 write_unlock(&tree->lock); 330 return ret; 331 332 } 333 334 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em) 335 { 336 lockdep_assert_held_write(&tree->lock); 337 338 clear_bit(EXTENT_FLAG_LOGGING, &em->flags); 339 if (extent_map_in_tree(em)) 340 try_merge_map(tree, em); 341 } 342 343 static inline void setup_extent_mapping(struct extent_map_tree *tree, 344 struct extent_map *em, 345 int modified) 346 { 347 refcount_inc(&em->refs); 348 em->mod_start = em->start; 349 em->mod_len = em->len; 350 351 if (modified) 352 list_move(&em->list, &tree->modified_extents); 353 else 354 try_merge_map(tree, em); 355 } 356 357 static void extent_map_device_set_bits(struct extent_map *em, unsigned bits) 358 { 359 struct map_lookup *map = em->map_lookup; 360 u64 stripe_size = em->orig_block_len; 361 int i; 362 363 for (i = 0; i < map->num_stripes; i++) { 364 struct btrfs_io_stripe *stripe = &map->stripes[i]; 365 struct btrfs_device *device = stripe->dev; 366 367 set_extent_bit(&device->alloc_state, stripe->physical, 368 stripe->physical + stripe_size - 1, bits, NULL, 369 GFP_NOWAIT); 370 } 371 } 372 373 static void extent_map_device_clear_bits(struct extent_map *em, unsigned bits) 374 { 375 struct map_lookup *map = em->map_lookup; 376 u64 stripe_size = em->orig_block_len; 377 int i; 378 379 for (i = 0; i < map->num_stripes; i++) { 380 struct btrfs_io_stripe *stripe = &map->stripes[i]; 381 struct btrfs_device *device = stripe->dev; 382 383 __clear_extent_bit(&device->alloc_state, stripe->physical, 384 stripe->physical + stripe_size - 1, bits, 385 NULL, GFP_NOWAIT, NULL); 386 } 387 } 388 389 /* 390 * Add new extent map to the extent tree 391 * 392 * @tree: tree to insert new map in 393 * @em: map to insert 394 * @modified: indicate whether the given @em should be added to the 395 * modified list, which indicates the extent needs to be logged 396 * 397 * Insert @em into @tree or perform a simple forward/backward merge with 398 * existing mappings. The extent_map struct passed in will be inserted 399 * into the tree directly, with an additional reference taken, or a 400 * reference dropped if the merge attempt was successful. 401 */ 402 int add_extent_mapping(struct extent_map_tree *tree, 403 struct extent_map *em, int modified) 404 { 405 int ret = 0; 406 407 lockdep_assert_held_write(&tree->lock); 408 409 ret = tree_insert(&tree->map, em); 410 if (ret) 411 goto out; 412 413 setup_extent_mapping(tree, em, modified); 414 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags)) { 415 extent_map_device_set_bits(em, CHUNK_ALLOCATED); 416 extent_map_device_clear_bits(em, CHUNK_TRIMMED); 417 } 418 out: 419 return ret; 420 } 421 422 static struct extent_map * 423 __lookup_extent_mapping(struct extent_map_tree *tree, 424 u64 start, u64 len, int strict) 425 { 426 struct extent_map *em; 427 struct rb_node *rb_node; 428 struct rb_node *prev_or_next = NULL; 429 u64 end = range_end(start, len); 430 431 rb_node = __tree_search(&tree->map.rb_root, start, &prev_or_next); 432 if (!rb_node) { 433 if (prev_or_next) 434 rb_node = prev_or_next; 435 else 436 return NULL; 437 } 438 439 em = rb_entry(rb_node, struct extent_map, rb_node); 440 441 if (strict && !(end > em->start && start < extent_map_end(em))) 442 return NULL; 443 444 refcount_inc(&em->refs); 445 return em; 446 } 447 448 /* 449 * Lookup extent_map that intersects @start + @len range. 450 * 451 * @tree: tree to lookup in 452 * @start: byte offset to start the search 453 * @len: length of the lookup range 454 * 455 * Find and return the first extent_map struct in @tree that intersects the 456 * [start, len] range. There may be additional objects in the tree that 457 * intersect, so check the object returned carefully to make sure that no 458 * additional lookups are needed. 459 */ 460 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree, 461 u64 start, u64 len) 462 { 463 return __lookup_extent_mapping(tree, start, len, 1); 464 } 465 466 /* 467 * Find a nearby extent map intersecting @start + @len (not an exact search). 468 * 469 * @tree: tree to lookup in 470 * @start: byte offset to start the search 471 * @len: length of the lookup range 472 * 473 * Find and return the first extent_map struct in @tree that intersects the 474 * [start, len] range. 475 * 476 * If one can't be found, any nearby extent may be returned 477 */ 478 struct extent_map *search_extent_mapping(struct extent_map_tree *tree, 479 u64 start, u64 len) 480 { 481 return __lookup_extent_mapping(tree, start, len, 0); 482 } 483 484 /* 485 * Remove an extent_map from the extent tree. 486 * 487 * @tree: extent tree to remove from 488 * @em: extent map being removed 489 * 490 * Remove @em from @tree. No reference counts are dropped, and no checks 491 * are done to see if the range is in use. 492 */ 493 void remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em) 494 { 495 lockdep_assert_held_write(&tree->lock); 496 497 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags)); 498 rb_erase_cached(&em->rb_node, &tree->map); 499 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags)) 500 list_del_init(&em->list); 501 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags)) 502 extent_map_device_clear_bits(em, CHUNK_ALLOCATED); 503 RB_CLEAR_NODE(&em->rb_node); 504 } 505 506 void replace_extent_mapping(struct extent_map_tree *tree, 507 struct extent_map *cur, 508 struct extent_map *new, 509 int modified) 510 { 511 lockdep_assert_held_write(&tree->lock); 512 513 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags)); 514 ASSERT(extent_map_in_tree(cur)); 515 if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags)) 516 list_del_init(&cur->list); 517 rb_replace_node_cached(&cur->rb_node, &new->rb_node, &tree->map); 518 RB_CLEAR_NODE(&cur->rb_node); 519 520 setup_extent_mapping(tree, new, modified); 521 } 522 523 static struct extent_map *next_extent_map(const struct extent_map *em) 524 { 525 struct rb_node *next; 526 527 next = rb_next(&em->rb_node); 528 if (!next) 529 return NULL; 530 return container_of(next, struct extent_map, rb_node); 531 } 532 533 static struct extent_map *prev_extent_map(struct extent_map *em) 534 { 535 struct rb_node *prev; 536 537 prev = rb_prev(&em->rb_node); 538 if (!prev) 539 return NULL; 540 return container_of(prev, struct extent_map, rb_node); 541 } 542 543 /* 544 * Helper for btrfs_get_extent. Given an existing extent in the tree, 545 * the existing extent is the nearest extent to map_start, 546 * and an extent that you want to insert, deal with overlap and insert 547 * the best fitted new extent into the tree. 548 */ 549 static noinline int merge_extent_mapping(struct extent_map_tree *em_tree, 550 struct extent_map *existing, 551 struct extent_map *em, 552 u64 map_start) 553 { 554 struct extent_map *prev; 555 struct extent_map *next; 556 u64 start; 557 u64 end; 558 u64 start_diff; 559 560 BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); 561 562 if (existing->start > map_start) { 563 next = existing; 564 prev = prev_extent_map(next); 565 } else { 566 prev = existing; 567 next = next_extent_map(prev); 568 } 569 570 start = prev ? extent_map_end(prev) : em->start; 571 start = max_t(u64, start, em->start); 572 end = next ? next->start : extent_map_end(em); 573 end = min_t(u64, end, extent_map_end(em)); 574 start_diff = start - em->start; 575 em->start = start; 576 em->len = end - start; 577 if (em->block_start < EXTENT_MAP_LAST_BYTE && 578 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { 579 em->block_start += start_diff; 580 em->block_len = em->len; 581 } 582 return add_extent_mapping(em_tree, em, 0); 583 } 584 585 /* 586 * Add extent mapping into em_tree. 587 * 588 * @fs_info: the filesystem 589 * @em_tree: extent tree into which we want to insert the extent mapping 590 * @em_in: extent we are inserting 591 * @start: start of the logical range btrfs_get_extent() is requesting 592 * @len: length of the logical range btrfs_get_extent() is requesting 593 * 594 * Note that @em_in's range may be different from [start, start+len), 595 * but they must be overlapped. 596 * 597 * Insert @em_in into @em_tree. In case there is an overlapping range, handle 598 * the -EEXIST by either: 599 * a) Returning the existing extent in @em_in if @start is within the 600 * existing em. 601 * b) Merge the existing extent with @em_in passed in. 602 * 603 * Return 0 on success, otherwise -EEXIST. 604 * 605 */ 606 int btrfs_add_extent_mapping(struct btrfs_fs_info *fs_info, 607 struct extent_map_tree *em_tree, 608 struct extent_map **em_in, u64 start, u64 len) 609 { 610 int ret; 611 struct extent_map *em = *em_in; 612 613 /* 614 * Tree-checker should have rejected any inline extent with non-zero 615 * file offset. Here just do a sanity check. 616 */ 617 if (em->block_start == EXTENT_MAP_INLINE) 618 ASSERT(em->start == 0); 619 620 ret = add_extent_mapping(em_tree, em, 0); 621 /* it is possible that someone inserted the extent into the tree 622 * while we had the lock dropped. It is also possible that 623 * an overlapping map exists in the tree 624 */ 625 if (ret == -EEXIST) { 626 struct extent_map *existing; 627 628 ret = 0; 629 630 existing = search_extent_mapping(em_tree, start, len); 631 632 trace_btrfs_handle_em_exist(fs_info, existing, em, start, len); 633 634 /* 635 * existing will always be non-NULL, since there must be 636 * extent causing the -EEXIST. 637 */ 638 if (start >= existing->start && 639 start < extent_map_end(existing)) { 640 free_extent_map(em); 641 *em_in = existing; 642 ret = 0; 643 } else { 644 u64 orig_start = em->start; 645 u64 orig_len = em->len; 646 647 /* 648 * The existing extent map is the one nearest to 649 * the [start, start + len) range which overlaps 650 */ 651 ret = merge_extent_mapping(em_tree, existing, 652 em, start); 653 if (ret) { 654 free_extent_map(em); 655 *em_in = NULL; 656 WARN_ONCE(ret, 657 "unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n", 658 ret, existing->start, existing->len, 659 orig_start, orig_len); 660 } 661 free_extent_map(existing); 662 } 663 } 664 665 ASSERT(ret == 0 || ret == -EEXIST); 666 return ret; 667 } 668 669 /* 670 * Drop all extent maps from a tree in the fastest possible way, rescheduling 671 * if needed. This avoids searching the tree, from the root down to the first 672 * extent map, before each deletion. 673 */ 674 static void drop_all_extent_maps_fast(struct extent_map_tree *tree) 675 { 676 write_lock(&tree->lock); 677 while (!RB_EMPTY_ROOT(&tree->map.rb_root)) { 678 struct extent_map *em; 679 struct rb_node *node; 680 681 node = rb_first_cached(&tree->map); 682 em = rb_entry(node, struct extent_map, rb_node); 683 clear_bit(EXTENT_FLAG_PINNED, &em->flags); 684 clear_bit(EXTENT_FLAG_LOGGING, &em->flags); 685 remove_extent_mapping(tree, em); 686 free_extent_map(em); 687 cond_resched_rwlock_write(&tree->lock); 688 } 689 write_unlock(&tree->lock); 690 } 691 692 /* 693 * Drop all extent maps in a given range. 694 * 695 * @inode: The target inode. 696 * @start: Start offset of the range. 697 * @end: End offset of the range (inclusive value). 698 * @skip_pinned: Indicate if pinned extent maps should be ignored or not. 699 * 700 * This drops all the extent maps that intersect the given range [@start, @end]. 701 * Extent maps that partially overlap the range and extend behind or beyond it, 702 * are split. 703 * The caller should have locked an appropriate file range in the inode's io 704 * tree before calling this function. 705 */ 706 void btrfs_drop_extent_map_range(struct btrfs_inode *inode, u64 start, u64 end, 707 bool skip_pinned) 708 { 709 struct extent_map *split; 710 struct extent_map *split2; 711 struct extent_map *em; 712 struct extent_map_tree *em_tree = &inode->extent_tree; 713 u64 len = end - start + 1; 714 715 WARN_ON(end < start); 716 if (end == (u64)-1) { 717 if (start == 0 && !skip_pinned) { 718 drop_all_extent_maps_fast(em_tree); 719 return; 720 } 721 len = (u64)-1; 722 } else { 723 /* Make end offset exclusive for use in the loop below. */ 724 end++; 725 } 726 727 /* 728 * It's ok if we fail to allocate the extent maps, see the comment near 729 * the bottom of the loop below. We only need two spare extent maps in 730 * the worst case, where the first extent map that intersects our range 731 * starts before the range and the last extent map that intersects our 732 * range ends after our range (and they might be the same extent map), 733 * because we need to split those two extent maps at the boundaries. 734 */ 735 split = alloc_extent_map(); 736 split2 = alloc_extent_map(); 737 738 write_lock(&em_tree->lock); 739 em = lookup_extent_mapping(em_tree, start, len); 740 741 while (em) { 742 /* extent_map_end() returns exclusive value (last byte + 1). */ 743 const u64 em_end = extent_map_end(em); 744 struct extent_map *next_em = NULL; 745 u64 gen; 746 unsigned long flags; 747 bool modified; 748 bool compressed; 749 750 if (em_end < end) { 751 next_em = next_extent_map(em); 752 if (next_em) { 753 if (next_em->start < end) 754 refcount_inc(&next_em->refs); 755 else 756 next_em = NULL; 757 } 758 } 759 760 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { 761 start = em_end; 762 if (end != (u64)-1) 763 len = start + len - em_end; 764 goto next; 765 } 766 767 flags = em->flags; 768 clear_bit(EXTENT_FLAG_PINNED, &em->flags); 769 /* 770 * In case we split the extent map, we want to preserve the 771 * EXTENT_FLAG_LOGGING flag on our extent map, but we don't want 772 * it on the new extent maps. 773 */ 774 clear_bit(EXTENT_FLAG_LOGGING, &flags); 775 modified = !list_empty(&em->list); 776 777 /* 778 * The extent map does not cross our target range, so no need to 779 * split it, we can remove it directly. 780 */ 781 if (em->start >= start && em_end <= end) 782 goto remove_em; 783 784 gen = em->generation; 785 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 786 787 if (em->start < start) { 788 if (!split) { 789 split = split2; 790 split2 = NULL; 791 if (!split) 792 goto remove_em; 793 } 794 split->start = em->start; 795 split->len = start - em->start; 796 797 if (em->block_start < EXTENT_MAP_LAST_BYTE) { 798 split->orig_start = em->orig_start; 799 split->block_start = em->block_start; 800 801 if (compressed) 802 split->block_len = em->block_len; 803 else 804 split->block_len = split->len; 805 split->orig_block_len = max(split->block_len, 806 em->orig_block_len); 807 split->ram_bytes = em->ram_bytes; 808 } else { 809 split->orig_start = split->start; 810 split->block_len = 0; 811 split->block_start = em->block_start; 812 split->orig_block_len = 0; 813 split->ram_bytes = split->len; 814 } 815 816 split->generation = gen; 817 split->flags = flags; 818 split->compress_type = em->compress_type; 819 replace_extent_mapping(em_tree, em, split, modified); 820 free_extent_map(split); 821 split = split2; 822 split2 = NULL; 823 } 824 if (em_end > end) { 825 if (!split) { 826 split = split2; 827 split2 = NULL; 828 if (!split) 829 goto remove_em; 830 } 831 split->start = start + len; 832 split->len = em_end - (start + len); 833 split->block_start = em->block_start; 834 split->flags = flags; 835 split->compress_type = em->compress_type; 836 split->generation = gen; 837 838 if (em->block_start < EXTENT_MAP_LAST_BYTE) { 839 split->orig_block_len = max(em->block_len, 840 em->orig_block_len); 841 842 split->ram_bytes = em->ram_bytes; 843 if (compressed) { 844 split->block_len = em->block_len; 845 split->orig_start = em->orig_start; 846 } else { 847 const u64 diff = start + len - em->start; 848 849 split->block_len = split->len; 850 split->block_start += diff; 851 split->orig_start = em->orig_start; 852 } 853 } else { 854 split->ram_bytes = split->len; 855 split->orig_start = split->start; 856 split->block_len = 0; 857 split->orig_block_len = 0; 858 } 859 860 if (extent_map_in_tree(em)) { 861 replace_extent_mapping(em_tree, em, split, 862 modified); 863 } else { 864 int ret; 865 866 ret = add_extent_mapping(em_tree, split, 867 modified); 868 /* Logic error, shouldn't happen. */ 869 ASSERT(ret == 0); 870 if (WARN_ON(ret != 0) && modified) 871 btrfs_set_inode_full_sync(inode); 872 } 873 free_extent_map(split); 874 split = NULL; 875 } 876 remove_em: 877 if (extent_map_in_tree(em)) { 878 /* 879 * If the extent map is still in the tree it means that 880 * either of the following is true: 881 * 882 * 1) It fits entirely in our range (doesn't end beyond 883 * it or starts before it); 884 * 885 * 2) It starts before our range and/or ends after our 886 * range, and we were not able to allocate the extent 887 * maps for split operations, @split and @split2. 888 * 889 * If we are at case 2) then we just remove the entire 890 * extent map - this is fine since if anyone needs it to 891 * access the subranges outside our range, will just 892 * load it again from the subvolume tree's file extent 893 * item. However if the extent map was in the list of 894 * modified extents, then we must mark the inode for a 895 * full fsync, otherwise a fast fsync will miss this 896 * extent if it's new and needs to be logged. 897 */ 898 if ((em->start < start || em_end > end) && modified) { 899 ASSERT(!split); 900 btrfs_set_inode_full_sync(inode); 901 } 902 remove_extent_mapping(em_tree, em); 903 } 904 905 /* 906 * Once for the tree reference (we replaced or removed the 907 * extent map from the tree). 908 */ 909 free_extent_map(em); 910 next: 911 /* Once for us (for our lookup reference). */ 912 free_extent_map(em); 913 914 em = next_em; 915 } 916 917 write_unlock(&em_tree->lock); 918 919 free_extent_map(split); 920 free_extent_map(split2); 921 } 922 923 /* 924 * Replace a range in the inode's extent map tree with a new extent map. 925 * 926 * @inode: The target inode. 927 * @new_em: The new extent map to add to the inode's extent map tree. 928 * @modified: Indicate if the new extent map should be added to the list of 929 * modified extents (for fast fsync tracking). 930 * 931 * Drops all the extent maps in the inode's extent map tree that intersect the 932 * range of the new extent map and adds the new extent map to the tree. 933 * The caller should have locked an appropriate file range in the inode's io 934 * tree before calling this function. 935 */ 936 int btrfs_replace_extent_map_range(struct btrfs_inode *inode, 937 struct extent_map *new_em, 938 bool modified) 939 { 940 const u64 end = new_em->start + new_em->len - 1; 941 struct extent_map_tree *tree = &inode->extent_tree; 942 int ret; 943 944 ASSERT(!extent_map_in_tree(new_em)); 945 946 /* 947 * The caller has locked an appropriate file range in the inode's io 948 * tree, but getting -EEXIST when adding the new extent map can still 949 * happen in case there are extents that partially cover the range, and 950 * this is due to two tasks operating on different parts of the extent. 951 * See commit 18e83ac75bfe67 ("Btrfs: fix unexpected EEXIST from 952 * btrfs_get_extent") for an example and details. 953 */ 954 do { 955 btrfs_drop_extent_map_range(inode, new_em->start, end, false); 956 write_lock(&tree->lock); 957 ret = add_extent_mapping(tree, new_em, modified); 958 write_unlock(&tree->lock); 959 } while (ret == -EEXIST); 960 961 return ret; 962 } 963