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 "ctree.h" 7 #include "volumes.h" 8 #include "extent_map.h" 9 #include "compression.h" 10 #include "btrfs_inode.h" 11 12 13 static struct kmem_cache *extent_map_cache; 14 15 int __init extent_map_init(void) 16 { 17 extent_map_cache = kmem_cache_create("btrfs_extent_map", 18 sizeof(struct extent_map), 0, 19 SLAB_MEM_SPREAD, NULL); 20 if (!extent_map_cache) 21 return -ENOMEM; 22 return 0; 23 } 24 25 void __cold extent_map_exit(void) 26 { 27 kmem_cache_destroy(extent_map_cache); 28 } 29 30 /** 31 * extent_map_tree_init - initialize extent map tree 32 * @tree: tree to initialize 33 * 34 * Initialize the extent tree @tree. Should be called for each new inode 35 * or other user of the extent_map interface. 36 */ 37 void extent_map_tree_init(struct extent_map_tree *tree) 38 { 39 tree->map = RB_ROOT_CACHED; 40 INIT_LIST_HEAD(&tree->modified_extents); 41 rwlock_init(&tree->lock); 42 } 43 44 /** 45 * alloc_extent_map - allocate new extent map structure 46 * 47 * Allocate a new extent_map structure. The new structure is 48 * returned with a reference count of one and needs to be 49 * freed using free_extent_map() 50 */ 51 struct extent_map *alloc_extent_map(void) 52 { 53 struct extent_map *em; 54 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS); 55 if (!em) 56 return NULL; 57 RB_CLEAR_NODE(&em->rb_node); 58 em->compress_type = BTRFS_COMPRESS_NONE; 59 refcount_set(&em->refs, 1); 60 INIT_LIST_HEAD(&em->list); 61 return em; 62 } 63 64 /** 65 * free_extent_map - drop reference count of an extent_map 66 * @em: extent map being released 67 * 68 * Drops the reference out on @em by one and free the structure 69 * if the reference count hits zero. 70 */ 71 void free_extent_map(struct extent_map *em) 72 { 73 if (!em) 74 return; 75 if (refcount_dec_and_test(&em->refs)) { 76 WARN_ON(extent_map_in_tree(em)); 77 WARN_ON(!list_empty(&em->list)); 78 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags)) 79 kfree(em->map_lookup); 80 kmem_cache_free(extent_map_cache, em); 81 } 82 } 83 84 /* simple helper to do math around the end of an extent, handling wrap */ 85 static u64 range_end(u64 start, u64 len) 86 { 87 if (start + len < start) 88 return (u64)-1; 89 return start + len; 90 } 91 92 static int tree_insert(struct rb_root_cached *root, struct extent_map *em) 93 { 94 struct rb_node **p = &root->rb_root.rb_node; 95 struct rb_node *parent = NULL; 96 struct extent_map *entry = NULL; 97 struct rb_node *orig_parent = NULL; 98 u64 end = range_end(em->start, em->len); 99 bool leftmost = true; 100 101 while (*p) { 102 parent = *p; 103 entry = rb_entry(parent, struct extent_map, rb_node); 104 105 if (em->start < entry->start) { 106 p = &(*p)->rb_left; 107 } else if (em->start >= extent_map_end(entry)) { 108 p = &(*p)->rb_right; 109 leftmost = false; 110 } else { 111 return -EEXIST; 112 } 113 } 114 115 orig_parent = parent; 116 while (parent && em->start >= extent_map_end(entry)) { 117 parent = rb_next(parent); 118 entry = rb_entry(parent, struct extent_map, rb_node); 119 } 120 if (parent) 121 if (end > entry->start && em->start < extent_map_end(entry)) 122 return -EEXIST; 123 124 parent = orig_parent; 125 entry = rb_entry(parent, struct extent_map, rb_node); 126 while (parent && em->start < entry->start) { 127 parent = rb_prev(parent); 128 entry = rb_entry(parent, struct extent_map, rb_node); 129 } 130 if (parent) 131 if (end > entry->start && em->start < extent_map_end(entry)) 132 return -EEXIST; 133 134 rb_link_node(&em->rb_node, orig_parent, p); 135 rb_insert_color_cached(&em->rb_node, root, leftmost); 136 return 0; 137 } 138 139 /* 140 * search through the tree for an extent_map with a given offset. If 141 * it can't be found, try to find some neighboring extents 142 */ 143 static struct rb_node *__tree_search(struct rb_root *root, u64 offset, 144 struct rb_node **prev_ret, 145 struct rb_node **next_ret) 146 { 147 struct rb_node *n = root->rb_node; 148 struct rb_node *prev = NULL; 149 struct rb_node *orig_prev = NULL; 150 struct extent_map *entry; 151 struct extent_map *prev_entry = NULL; 152 153 while (n) { 154 entry = rb_entry(n, struct extent_map, rb_node); 155 prev = n; 156 prev_entry = entry; 157 158 if (offset < entry->start) 159 n = n->rb_left; 160 else if (offset >= extent_map_end(entry)) 161 n = n->rb_right; 162 else 163 return n; 164 } 165 166 if (prev_ret) { 167 orig_prev = prev; 168 while (prev && offset >= extent_map_end(prev_entry)) { 169 prev = rb_next(prev); 170 prev_entry = rb_entry(prev, struct extent_map, rb_node); 171 } 172 *prev_ret = prev; 173 prev = orig_prev; 174 } 175 176 if (next_ret) { 177 prev_entry = rb_entry(prev, struct extent_map, rb_node); 178 while (prev && offset < prev_entry->start) { 179 prev = rb_prev(prev); 180 prev_entry = rb_entry(prev, struct extent_map, rb_node); 181 } 182 *next_ret = prev; 183 } 184 return NULL; 185 } 186 187 /* check to see if two extent_map structs are adjacent and safe to merge */ 188 static int mergable_maps(struct extent_map *prev, struct extent_map *next) 189 { 190 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags)) 191 return 0; 192 193 /* 194 * don't merge compressed extents, we need to know their 195 * actual size 196 */ 197 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags)) 198 return 0; 199 200 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) || 201 test_bit(EXTENT_FLAG_LOGGING, &next->flags)) 202 return 0; 203 204 /* 205 * We don't want to merge stuff that hasn't been written to the log yet 206 * since it may not reflect exactly what is on disk, and that would be 207 * bad. 208 */ 209 if (!list_empty(&prev->list) || !list_empty(&next->list)) 210 return 0; 211 212 ASSERT(next->block_start != EXTENT_MAP_DELALLOC && 213 prev->block_start != EXTENT_MAP_DELALLOC); 214 215 if (prev->map_lookup || next->map_lookup) 216 ASSERT(test_bit(EXTENT_FLAG_FS_MAPPING, &prev->flags) && 217 test_bit(EXTENT_FLAG_FS_MAPPING, &next->flags)); 218 219 if (extent_map_end(prev) == next->start && 220 prev->flags == next->flags && 221 prev->map_lookup == next->map_lookup && 222 ((next->block_start == EXTENT_MAP_HOLE && 223 prev->block_start == EXTENT_MAP_HOLE) || 224 (next->block_start == EXTENT_MAP_INLINE && 225 prev->block_start == EXTENT_MAP_INLINE) || 226 (next->block_start < EXTENT_MAP_LAST_BYTE - 1 && 227 next->block_start == extent_map_block_end(prev)))) { 228 return 1; 229 } 230 return 0; 231 } 232 233 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em) 234 { 235 struct extent_map *merge = NULL; 236 struct rb_node *rb; 237 238 /* 239 * We can't modify an extent map that is in the tree and that is being 240 * used by another task, as it can cause that other task to see it in 241 * inconsistent state during the merging. We always have 1 reference for 242 * the tree and 1 for this task (which is unpinning the extent map or 243 * clearing the logging flag), so anything > 2 means it's being used by 244 * other tasks too. 245 */ 246 if (refcount_read(&em->refs) > 2) 247 return; 248 249 if (em->start != 0) { 250 rb = rb_prev(&em->rb_node); 251 if (rb) 252 merge = rb_entry(rb, struct extent_map, rb_node); 253 if (rb && mergable_maps(merge, em)) { 254 em->start = merge->start; 255 em->orig_start = merge->orig_start; 256 em->len += merge->len; 257 em->block_len += merge->block_len; 258 em->block_start = merge->block_start; 259 em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start; 260 em->mod_start = merge->mod_start; 261 em->generation = max(em->generation, merge->generation); 262 set_bit(EXTENT_FLAG_MERGED, &em->flags); 263 264 rb_erase_cached(&merge->rb_node, &tree->map); 265 RB_CLEAR_NODE(&merge->rb_node); 266 free_extent_map(merge); 267 } 268 } 269 270 rb = rb_next(&em->rb_node); 271 if (rb) 272 merge = rb_entry(rb, struct extent_map, rb_node); 273 if (rb && mergable_maps(em, merge)) { 274 em->len += merge->len; 275 em->block_len += merge->block_len; 276 rb_erase_cached(&merge->rb_node, &tree->map); 277 RB_CLEAR_NODE(&merge->rb_node); 278 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start; 279 em->generation = max(em->generation, merge->generation); 280 set_bit(EXTENT_FLAG_MERGED, &em->flags); 281 free_extent_map(merge); 282 } 283 } 284 285 /** 286 * unpin_extent_cache - unpin an extent from the cache 287 * @tree: tree to unpin the extent in 288 * @start: logical offset in the file 289 * @len: length of the extent 290 * @gen: generation that this extent has been modified in 291 * 292 * Called after an extent has been written to disk properly. Set the generation 293 * to the generation that actually added the file item to the inode so we know 294 * we need to sync this extent when we call fsync(). 295 */ 296 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len, 297 u64 gen) 298 { 299 int ret = 0; 300 struct extent_map *em; 301 bool prealloc = false; 302 303 write_lock(&tree->lock); 304 em = lookup_extent_mapping(tree, start, len); 305 306 WARN_ON(!em || em->start != start); 307 308 if (!em) 309 goto out; 310 311 em->generation = gen; 312 clear_bit(EXTENT_FLAG_PINNED, &em->flags); 313 em->mod_start = em->start; 314 em->mod_len = em->len; 315 316 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) { 317 prealloc = true; 318 clear_bit(EXTENT_FLAG_FILLING, &em->flags); 319 } 320 321 try_merge_map(tree, em); 322 323 if (prealloc) { 324 em->mod_start = em->start; 325 em->mod_len = em->len; 326 } 327 328 free_extent_map(em); 329 out: 330 write_unlock(&tree->lock); 331 return ret; 332 333 } 334 335 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em) 336 { 337 lockdep_assert_held_write(&tree->lock); 338 339 clear_bit(EXTENT_FLAG_LOGGING, &em->flags); 340 if (extent_map_in_tree(em)) 341 try_merge_map(tree, em); 342 } 343 344 static inline void setup_extent_mapping(struct extent_map_tree *tree, 345 struct extent_map *em, 346 int modified) 347 { 348 refcount_inc(&em->refs); 349 em->mod_start = em->start; 350 em->mod_len = em->len; 351 352 if (modified) 353 list_move(&em->list, &tree->modified_extents); 354 else 355 try_merge_map(tree, em); 356 } 357 358 static void extent_map_device_set_bits(struct extent_map *em, unsigned bits) 359 { 360 struct map_lookup *map = em->map_lookup; 361 u64 stripe_size = em->orig_block_len; 362 int i; 363 364 for (i = 0; i < map->num_stripes; i++) { 365 struct btrfs_io_stripe *stripe = &map->stripes[i]; 366 struct btrfs_device *device = stripe->dev; 367 368 set_extent_bits_nowait(&device->alloc_state, stripe->physical, 369 stripe->physical + stripe_size - 1, bits); 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 = NULL; 429 struct rb_node *next = NULL; 430 u64 end = range_end(start, len); 431 432 rb_node = __tree_search(&tree->map.rb_root, start, &prev, &next); 433 if (!rb_node) { 434 if (prev) 435 rb_node = prev; 436 else if (next) 437 rb_node = next; 438 else 439 return NULL; 440 } 441 442 em = rb_entry(rb_node, struct extent_map, rb_node); 443 444 if (strict && !(end > em->start && start < extent_map_end(em))) 445 return NULL; 446 447 refcount_inc(&em->refs); 448 return em; 449 } 450 451 /** 452 * lookup_extent_mapping - lookup extent_map 453 * @tree: tree to lookup in 454 * @start: byte offset to start the search 455 * @len: length of the lookup range 456 * 457 * Find and return the first extent_map struct in @tree that intersects the 458 * [start, len] range. There may be additional objects in the tree that 459 * intersect, so check the object returned carefully to make sure that no 460 * additional lookups are needed. 461 */ 462 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree, 463 u64 start, u64 len) 464 { 465 return __lookup_extent_mapping(tree, start, len, 1); 466 } 467 468 /** 469 * search_extent_mapping - find a nearby extent map 470 * @tree: tree to lookup in 471 * @start: byte offset to start the search 472 * @len: length of the lookup range 473 * 474 * Find and return the first extent_map struct in @tree that intersects the 475 * [start, len] range. 476 * 477 * If one can't be found, any nearby extent may be returned 478 */ 479 struct extent_map *search_extent_mapping(struct extent_map_tree *tree, 480 u64 start, u64 len) 481 { 482 return __lookup_extent_mapping(tree, start, len, 0); 483 } 484 485 /** 486 * remove_extent_mapping - removes an extent_map from the extent tree 487 * @tree: extent tree to remove from 488 * @em: extent map being removed 489 * 490 * Removes @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(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 ret = add_extent_mapping(em_tree, em, 0); 614 /* it is possible that someone inserted the extent into the tree 615 * while we had the lock dropped. It is also possible that 616 * an overlapping map exists in the tree 617 */ 618 if (ret == -EEXIST) { 619 struct extent_map *existing; 620 621 ret = 0; 622 623 existing = search_extent_mapping(em_tree, start, len); 624 625 trace_btrfs_handle_em_exist(fs_info, existing, em, start, len); 626 627 /* 628 * existing will always be non-NULL, since there must be 629 * extent causing the -EEXIST. 630 */ 631 if (start >= existing->start && 632 start < extent_map_end(existing)) { 633 free_extent_map(em); 634 *em_in = existing; 635 ret = 0; 636 } else { 637 u64 orig_start = em->start; 638 u64 orig_len = em->len; 639 640 /* 641 * The existing extent map is the one nearest to 642 * the [start, start + len) range which overlaps 643 */ 644 ret = merge_extent_mapping(em_tree, existing, 645 em, start); 646 if (ret) { 647 free_extent_map(em); 648 *em_in = NULL; 649 WARN_ONCE(ret, 650 "unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n", 651 ret, existing->start, existing->len, 652 orig_start, orig_len); 653 } 654 free_extent_map(existing); 655 } 656 } 657 658 ASSERT(ret == 0 || ret == -EEXIST); 659 return ret; 660 } 661 662 /* 663 * Drop all extent maps from a tree in the fastest possible way, rescheduling 664 * if needed. This avoids searching the tree, from the root down to the first 665 * extent map, before each deletion. 666 */ 667 static void drop_all_extent_maps_fast(struct extent_map_tree *tree) 668 { 669 write_lock(&tree->lock); 670 while (!RB_EMPTY_ROOT(&tree->map.rb_root)) { 671 struct extent_map *em; 672 struct rb_node *node; 673 674 node = rb_first_cached(&tree->map); 675 em = rb_entry(node, struct extent_map, rb_node); 676 clear_bit(EXTENT_FLAG_PINNED, &em->flags); 677 clear_bit(EXTENT_FLAG_LOGGING, &em->flags); 678 remove_extent_mapping(tree, em); 679 free_extent_map(em); 680 cond_resched_rwlock_write(&tree->lock); 681 } 682 write_unlock(&tree->lock); 683 } 684 685 /* 686 * Drop all extent maps in a given range. 687 * 688 * @inode: The target inode. 689 * @start: Start offset of the range. 690 * @end: End offset of the range (inclusive value). 691 * @skip_pinned: Indicate if pinned extent maps should be ignored or not. 692 * 693 * This drops all the extent maps that intersect the given range [@start, @end]. 694 * Extent maps that partially overlap the range and extend behind or beyond it, 695 * are split. 696 * The caller should have locked an appropriate file range in the inode's io 697 * tree before calling this function. 698 */ 699 void btrfs_drop_extent_map_range(struct btrfs_inode *inode, u64 start, u64 end, 700 bool skip_pinned) 701 { 702 struct extent_map *split = NULL; 703 struct extent_map *split2 = NULL; 704 struct extent_map_tree *em_tree = &inode->extent_tree; 705 u64 len = end - start + 1; 706 bool testend = true; 707 708 WARN_ON(end < start); 709 if (end == (u64)-1) { 710 if (start == 0 && !skip_pinned) { 711 drop_all_extent_maps_fast(em_tree); 712 return; 713 } 714 len = (u64)-1; 715 testend = false; 716 } 717 while (1) { 718 struct extent_map *em; 719 u64 em_end; 720 u64 gen; 721 unsigned long flags; 722 bool ends_after_range = false; 723 bool no_splits = false; 724 bool modified; 725 bool compressed; 726 727 if (!split) 728 split = alloc_extent_map(); 729 if (!split2) 730 split2 = alloc_extent_map(); 731 if (!split || !split2) 732 no_splits = true; 733 734 write_lock(&em_tree->lock); 735 em = lookup_extent_mapping(em_tree, start, len); 736 if (!em) { 737 write_unlock(&em_tree->lock); 738 break; 739 } 740 em_end = extent_map_end(em); 741 if (testend && em_end > start + len) 742 ends_after_range = true; 743 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { 744 if (ends_after_range) { 745 free_extent_map(em); 746 write_unlock(&em_tree->lock); 747 break; 748 } 749 start = em_end; 750 if (testend) 751 len = start + len - em_end; 752 free_extent_map(em); 753 write_unlock(&em_tree->lock); 754 continue; 755 } 756 flags = em->flags; 757 gen = em->generation; 758 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); 759 clear_bit(EXTENT_FLAG_PINNED, &em->flags); 760 clear_bit(EXTENT_FLAG_LOGGING, &flags); 761 modified = !list_empty(&em->list); 762 if (no_splits) 763 goto next; 764 765 if (em->start < start) { 766 split->start = em->start; 767 split->len = start - em->start; 768 769 if (em->block_start < EXTENT_MAP_LAST_BYTE) { 770 split->orig_start = em->orig_start; 771 split->block_start = em->block_start; 772 773 if (compressed) 774 split->block_len = em->block_len; 775 else 776 split->block_len = split->len; 777 split->orig_block_len = max(split->block_len, 778 em->orig_block_len); 779 split->ram_bytes = em->ram_bytes; 780 } else { 781 split->orig_start = split->start; 782 split->block_len = 0; 783 split->block_start = em->block_start; 784 split->orig_block_len = 0; 785 split->ram_bytes = split->len; 786 } 787 788 split->generation = gen; 789 split->flags = flags; 790 split->compress_type = em->compress_type; 791 replace_extent_mapping(em_tree, em, split, modified); 792 free_extent_map(split); 793 split = split2; 794 split2 = NULL; 795 } 796 if (ends_after_range) { 797 split->start = start + len; 798 split->len = em_end - (start + len); 799 split->block_start = em->block_start; 800 split->flags = flags; 801 split->compress_type = em->compress_type; 802 split->generation = gen; 803 804 if (em->block_start < EXTENT_MAP_LAST_BYTE) { 805 split->orig_block_len = max(em->block_len, 806 em->orig_block_len); 807 808 split->ram_bytes = em->ram_bytes; 809 if (compressed) { 810 split->block_len = em->block_len; 811 split->orig_start = em->orig_start; 812 } else { 813 const u64 diff = start + len - em->start; 814 815 split->block_len = split->len; 816 split->block_start += diff; 817 split->orig_start = em->orig_start; 818 } 819 } else { 820 split->ram_bytes = split->len; 821 split->orig_start = split->start; 822 split->block_len = 0; 823 split->orig_block_len = 0; 824 } 825 826 if (extent_map_in_tree(em)) { 827 replace_extent_mapping(em_tree, em, split, 828 modified); 829 } else { 830 int ret; 831 832 ret = add_extent_mapping(em_tree, split, 833 modified); 834 /* Logic error, shouldn't happen. */ 835 ASSERT(ret == 0); 836 if (WARN_ON(ret != 0) && modified) 837 btrfs_set_inode_full_sync(inode); 838 } 839 free_extent_map(split); 840 split = NULL; 841 } 842 next: 843 if (extent_map_in_tree(em)) { 844 /* 845 * If the extent map is still in the tree it means that 846 * either of the following is true: 847 * 848 * 1) It fits entirely in our range (doesn't end beyond 849 * it or starts before it); 850 * 851 * 2) It starts before our range and/or ends after our 852 * range, and we were not able to allocate the extent 853 * maps for split operations, @split and @split2. 854 * 855 * If we are at case 2) then we just remove the entire 856 * extent map - this is fine since if anyone needs it to 857 * access the subranges outside our range, will just 858 * load it again from the subvolume tree's file extent 859 * item. However if the extent map was in the list of 860 * modified extents, then we must mark the inode for a 861 * full fsync, otherwise a fast fsync will miss this 862 * extent if it's new and needs to be logged. 863 */ 864 if ((em->start < start || ends_after_range) && modified) { 865 ASSERT(no_splits); 866 btrfs_set_inode_full_sync(inode); 867 } 868 remove_extent_mapping(em_tree, em); 869 } 870 write_unlock(&em_tree->lock); 871 872 /* Once for us. */ 873 free_extent_map(em); 874 /* And once for the tree. */ 875 free_extent_map(em); 876 } 877 878 free_extent_map(split); 879 free_extent_map(split2); 880 } 881 882 /* 883 * Replace a range in the inode's extent map tree with a new extent map. 884 * 885 * @inode: The target inode. 886 * @new_em: The new extent map to add to the inode's extent map tree. 887 * @modified: Indicate if the new extent map should be added to the list of 888 * modified extents (for fast fsync tracking). 889 * 890 * Drops all the extent maps in the inode's extent map tree that intersect the 891 * range of the new extent map and adds the new extent map to the tree. 892 * The caller should have locked an appropriate file range in the inode's io 893 * tree before calling this function. 894 */ 895 int btrfs_replace_extent_map_range(struct btrfs_inode *inode, 896 struct extent_map *new_em, 897 bool modified) 898 { 899 const u64 end = new_em->start + new_em->len - 1; 900 struct extent_map_tree *tree = &inode->extent_tree; 901 int ret; 902 903 ASSERT(!extent_map_in_tree(new_em)); 904 905 /* 906 * The caller has locked an appropriate file range in the inode's io 907 * tree, but getting -EEXIST when adding the new extent map can still 908 * happen in case there are extents that partially cover the range, and 909 * this is due to two tasks operating on different parts of the extent. 910 * See commit 18e83ac75bfe67 ("Btrfs: fix unexpected EEXIST from 911 * btrfs_get_extent") for an example and details. 912 */ 913 do { 914 btrfs_drop_extent_map_range(inode, new_em->start, end, false); 915 write_lock(&tree->lock); 916 ret = add_extent_mapping(tree, new_em, modified); 917 write_unlock(&tree->lock); 918 } while (ret == -EEXIST); 919 920 return ret; 921 } 922