1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/slab.h>
4 #include <linux/spinlock.h>
6 #include "extent_map.h"
7 #include "compression.h"
10 static struct kmem_cache *extent_map_cache;
12 int __init extent_map_init(void)
14 extent_map_cache = kmem_cache_create("btrfs_extent_map",
15 sizeof(struct extent_map), 0,
16 SLAB_MEM_SPREAD, NULL);
17 if (!extent_map_cache)
22 void __cold extent_map_exit(void)
24 kmem_cache_destroy(extent_map_cache);
28 * extent_map_tree_init - initialize extent map tree
29 * @tree: tree to initialize
31 * Initialize the extent tree @tree. Should be called for each new inode
32 * or other user of the extent_map interface.
34 void extent_map_tree_init(struct extent_map_tree *tree)
37 INIT_LIST_HEAD(&tree->modified_extents);
38 rwlock_init(&tree->lock);
42 * alloc_extent_map - allocate new extent map structure
44 * Allocate a new extent_map structure. The new structure is
45 * returned with a reference count of one and needs to be
46 * freed using free_extent_map()
48 struct extent_map *alloc_extent_map(void)
50 struct extent_map *em;
51 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
54 RB_CLEAR_NODE(&em->rb_node);
56 em->compress_type = BTRFS_COMPRESS_NONE;
58 refcount_set(&em->refs, 1);
59 INIT_LIST_HEAD(&em->list);
64 * free_extent_map - drop reference count of an extent_map
65 * @em: extent map being released
67 * Drops the reference out on @em by one and free the structure
68 * if the reference count hits zero.
70 void free_extent_map(struct extent_map *em)
74 WARN_ON(refcount_read(&em->refs) == 0);
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);
84 /* simple helper to do math around the end of an extent, handling wrap */
85 static u64 range_end(u64 start, u64 len)
87 if (start + len < start)
92 static int tree_insert(struct rb_root *root, struct extent_map *em)
94 struct rb_node **p = &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);
102 entry = rb_entry(parent, struct extent_map, rb_node);
104 if (em->start < entry->start)
106 else if (em->start >= extent_map_end(entry))
112 orig_parent = parent;
113 while (parent && em->start >= extent_map_end(entry)) {
114 parent = rb_next(parent);
115 entry = rb_entry(parent, struct extent_map, rb_node);
118 if (end > entry->start && em->start < extent_map_end(entry))
121 parent = orig_parent;
122 entry = rb_entry(parent, struct extent_map, rb_node);
123 while (parent && em->start < entry->start) {
124 parent = rb_prev(parent);
125 entry = rb_entry(parent, struct extent_map, rb_node);
128 if (end > entry->start && em->start < extent_map_end(entry))
131 rb_link_node(&em->rb_node, orig_parent, p);
132 rb_insert_color(&em->rb_node, root);
137 * search through the tree for an extent_map with a given offset. If
138 * it can't be found, try to find some neighboring extents
140 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
141 struct rb_node **prev_ret,
142 struct rb_node **next_ret)
144 struct rb_node *n = root->rb_node;
145 struct rb_node *prev = NULL;
146 struct rb_node *orig_prev = NULL;
147 struct extent_map *entry;
148 struct extent_map *prev_entry = NULL;
151 entry = rb_entry(n, struct extent_map, rb_node);
155 if (offset < entry->start)
157 else if (offset >= extent_map_end(entry))
165 while (prev && offset >= extent_map_end(prev_entry)) {
166 prev = rb_next(prev);
167 prev_entry = rb_entry(prev, struct extent_map, rb_node);
174 prev_entry = rb_entry(prev, struct extent_map, rb_node);
175 while (prev && offset < prev_entry->start) {
176 prev = rb_prev(prev);
177 prev_entry = rb_entry(prev, struct extent_map, rb_node);
184 /* check to see if two extent_map structs are adjacent and safe to merge */
185 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
187 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
191 * don't merge compressed extents, we need to know their
194 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
197 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
198 test_bit(EXTENT_FLAG_LOGGING, &next->flags))
202 * We don't want to merge stuff that hasn't been written to the log yet
203 * since it may not reflect exactly what is on disk, and that would be
206 if (!list_empty(&prev->list) || !list_empty(&next->list))
209 if (extent_map_end(prev) == next->start &&
210 prev->flags == next->flags &&
211 prev->bdev == next->bdev &&
212 ((next->block_start == EXTENT_MAP_HOLE &&
213 prev->block_start == EXTENT_MAP_HOLE) ||
214 (next->block_start == EXTENT_MAP_INLINE &&
215 prev->block_start == EXTENT_MAP_INLINE) ||
216 (next->block_start == EXTENT_MAP_DELALLOC &&
217 prev->block_start == EXTENT_MAP_DELALLOC) ||
218 (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
219 next->block_start == extent_map_block_end(prev)))) {
225 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
227 struct extent_map *merge = NULL;
230 if (em->start != 0) {
231 rb = rb_prev(&em->rb_node);
233 merge = rb_entry(rb, struct extent_map, rb_node);
234 if (rb && mergable_maps(merge, em)) {
235 em->start = merge->start;
236 em->orig_start = merge->orig_start;
237 em->len += merge->len;
238 em->block_len += merge->block_len;
239 em->block_start = merge->block_start;
240 em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
241 em->mod_start = merge->mod_start;
242 em->generation = max(em->generation, merge->generation);
244 rb_erase(&merge->rb_node, &tree->map);
245 RB_CLEAR_NODE(&merge->rb_node);
246 free_extent_map(merge);
250 rb = rb_next(&em->rb_node);
252 merge = rb_entry(rb, struct extent_map, rb_node);
253 if (rb && mergable_maps(em, merge)) {
254 em->len += merge->len;
255 em->block_len += merge->block_len;
256 rb_erase(&merge->rb_node, &tree->map);
257 RB_CLEAR_NODE(&merge->rb_node);
258 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
259 em->generation = max(em->generation, merge->generation);
260 free_extent_map(merge);
265 * unpin_extent_cache - unpin an extent from the cache
266 * @tree: tree to unpin the extent in
267 * @start: logical offset in the file
268 * @len: length of the extent
269 * @gen: generation that this extent has been modified in
271 * Called after an extent has been written to disk properly. Set the generation
272 * to the generation that actually added the file item to the inode so we know
273 * we need to sync this extent when we call fsync().
275 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
279 struct extent_map *em;
280 bool prealloc = false;
282 write_lock(&tree->lock);
283 em = lookup_extent_mapping(tree, start, len);
285 WARN_ON(!em || em->start != start);
290 em->generation = gen;
291 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
292 em->mod_start = em->start;
293 em->mod_len = em->len;
295 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
297 clear_bit(EXTENT_FLAG_FILLING, &em->flags);
300 try_merge_map(tree, em);
303 em->mod_start = em->start;
304 em->mod_len = em->len;
309 write_unlock(&tree->lock);
314 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
316 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
317 if (extent_map_in_tree(em))
318 try_merge_map(tree, em);
321 static inline void setup_extent_mapping(struct extent_map_tree *tree,
322 struct extent_map *em,
325 refcount_inc(&em->refs);
326 em->mod_start = em->start;
327 em->mod_len = em->len;
330 list_move(&em->list, &tree->modified_extents);
332 try_merge_map(tree, em);
336 * add_extent_mapping - add new extent map to the extent tree
337 * @tree: tree to insert new map in
340 * Insert @em into @tree or perform a simple forward/backward merge with
341 * existing mappings. The extent_map struct passed in will be inserted
342 * into the tree directly, with an additional reference taken, or a
343 * reference dropped if the merge attempt was successful.
345 int add_extent_mapping(struct extent_map_tree *tree,
346 struct extent_map *em, int modified)
350 ret = tree_insert(&tree->map, em);
354 setup_extent_mapping(tree, em, modified);
359 static struct extent_map *
360 __lookup_extent_mapping(struct extent_map_tree *tree,
361 u64 start, u64 len, int strict)
363 struct extent_map *em;
364 struct rb_node *rb_node;
365 struct rb_node *prev = NULL;
366 struct rb_node *next = NULL;
367 u64 end = range_end(start, len);
369 rb_node = __tree_search(&tree->map, start, &prev, &next);
379 em = rb_entry(rb_node, struct extent_map, rb_node);
381 if (strict && !(end > em->start && start < extent_map_end(em)))
384 refcount_inc(&em->refs);
389 * lookup_extent_mapping - lookup extent_map
390 * @tree: tree to lookup in
391 * @start: byte offset to start the search
392 * @len: length of the lookup range
394 * Find and return the first extent_map struct in @tree that intersects the
395 * [start, len] range. There may be additional objects in the tree that
396 * intersect, so check the object returned carefully to make sure that no
397 * additional lookups are needed.
399 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
402 return __lookup_extent_mapping(tree, start, len, 1);
406 * search_extent_mapping - find a nearby extent map
407 * @tree: tree to lookup in
408 * @start: byte offset to start the search
409 * @len: length of the lookup range
411 * Find and return the first extent_map struct in @tree that intersects the
412 * [start, len] range.
414 * If one can't be found, any nearby extent may be returned
416 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
419 return __lookup_extent_mapping(tree, start, len, 0);
423 * remove_extent_mapping - removes an extent_map from the extent tree
424 * @tree: extent tree to remove from
425 * @em: extent map being removed
427 * Removes @em from @tree. No reference counts are dropped, and no checks
428 * are done to see if the range is in use
430 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
434 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
435 rb_erase(&em->rb_node, &tree->map);
436 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
437 list_del_init(&em->list);
438 RB_CLEAR_NODE(&em->rb_node);
442 void replace_extent_mapping(struct extent_map_tree *tree,
443 struct extent_map *cur,
444 struct extent_map *new,
447 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
448 ASSERT(extent_map_in_tree(cur));
449 if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
450 list_del_init(&cur->list);
451 rb_replace_node(&cur->rb_node, &new->rb_node, &tree->map);
452 RB_CLEAR_NODE(&cur->rb_node);
454 setup_extent_mapping(tree, new, modified);
457 static struct extent_map *next_extent_map(struct extent_map *em)
459 struct rb_node *next;
461 next = rb_next(&em->rb_node);
464 return container_of(next, struct extent_map, rb_node);
467 static struct extent_map *prev_extent_map(struct extent_map *em)
469 struct rb_node *prev;
471 prev = rb_prev(&em->rb_node);
474 return container_of(prev, struct extent_map, rb_node);
477 /* helper for btfs_get_extent. Given an existing extent in the tree,
478 * the existing extent is the nearest extent to map_start,
479 * and an extent that you want to insert, deal with overlap and insert
480 * the best fitted new extent into the tree.
482 static noinline int merge_extent_mapping(struct extent_map_tree *em_tree,
483 struct extent_map *existing,
484 struct extent_map *em,
487 struct extent_map *prev;
488 struct extent_map *next;
493 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
495 if (existing->start > map_start) {
497 prev = prev_extent_map(next);
500 next = next_extent_map(prev);
503 start = prev ? extent_map_end(prev) : em->start;
504 start = max_t(u64, start, em->start);
505 end = next ? next->start : extent_map_end(em);
506 end = min_t(u64, end, extent_map_end(em));
507 start_diff = start - em->start;
509 em->len = end - start;
510 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
511 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
512 em->block_start += start_diff;
513 em->block_len = em->len;
515 return add_extent_mapping(em_tree, em, 0);
519 * btrfs_add_extent_mapping - add extent mapping into em_tree
520 * @em_tree - the extent tree into which we want to insert the extent mapping
521 * @em_in - extent we are inserting
522 * @start - start of the logical range btrfs_get_extent() is requesting
523 * @len - length of the logical range btrfs_get_extent() is requesting
525 * Note that @em_in's range may be different from [start, start+len),
526 * but they must be overlapped.
528 * Insert @em_in into @em_tree. In case there is an overlapping range, handle
529 * the -EEXIST by either:
530 * a) Returning the existing extent in @em_in if @start is within the
532 * b) Merge the existing extent with @em_in passed in.
534 * Return 0 on success, otherwise -EEXIST.
537 int btrfs_add_extent_mapping(struct extent_map_tree *em_tree,
538 struct extent_map **em_in, u64 start, u64 len)
541 struct extent_map *em = *em_in;
543 ret = add_extent_mapping(em_tree, em, 0);
544 /* it is possible that someone inserted the extent into the tree
545 * while we had the lock dropped. It is also possible that
546 * an overlapping map exists in the tree
548 if (ret == -EEXIST) {
549 struct extent_map *existing;
553 existing = search_extent_mapping(em_tree, start, len);
555 trace_btrfs_handle_em_exist(existing, em, start, len);
558 * existing will always be non-NULL, since there must be
559 * extent causing the -EEXIST.
561 if (start >= existing->start &&
562 start < extent_map_end(existing)) {
567 u64 orig_start = em->start;
568 u64 orig_len = em->len;
571 * The existing extent map is the one nearest to
572 * the [start, start + len) range which overlaps
574 ret = merge_extent_mapping(em_tree, existing,
580 "unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n",
581 ret, existing->start, existing->len,
582 orig_start, orig_len);
584 free_extent_map(existing);
588 ASSERT(ret == 0 || ret == -EEXIST);