1 // SPDX-License-Identifier: GPL-2.0
4 #include <linux/slab.h>
5 #include <linux/spinlock.h>
8 #include "extent_map.h"
9 #include "compression.h"
12 static struct kmem_cache *extent_map_cache;
14 int __init extent_map_init(void)
16 extent_map_cache = kmem_cache_create("btrfs_extent_map",
17 sizeof(struct extent_map), 0,
18 SLAB_MEM_SPREAD, NULL);
19 if (!extent_map_cache)
24 void __cold extent_map_exit(void)
26 kmem_cache_destroy(extent_map_cache);
30 * extent_map_tree_init - initialize extent map tree
31 * @tree: tree to initialize
33 * Initialize the extent tree @tree. Should be called for each new inode
34 * or other user of the extent_map interface.
36 void extent_map_tree_init(struct extent_map_tree *tree)
38 tree->map = RB_ROOT_CACHED;
39 INIT_LIST_HEAD(&tree->modified_extents);
40 rwlock_init(&tree->lock);
44 * alloc_extent_map - allocate new extent map structure
46 * Allocate a new extent_map structure. The new structure is
47 * returned with a reference count of one and needs to be
48 * freed using free_extent_map()
50 struct extent_map *alloc_extent_map(void)
52 struct extent_map *em;
53 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
56 RB_CLEAR_NODE(&em->rb_node);
58 em->compress_type = BTRFS_COMPRESS_NONE;
60 refcount_set(&em->refs, 1);
61 INIT_LIST_HEAD(&em->list);
66 * free_extent_map - drop reference count of an extent_map
67 * @em: extent map being released
69 * Drops the reference out on @em by one and free the structure
70 * if the reference count hits zero.
72 void free_extent_map(struct extent_map *em)
76 WARN_ON(refcount_read(&em->refs) == 0);
77 if (refcount_dec_and_test(&em->refs)) {
78 WARN_ON(extent_map_in_tree(em));
79 WARN_ON(!list_empty(&em->list));
80 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
81 kfree(em->map_lookup);
82 kmem_cache_free(extent_map_cache, em);
86 /* simple helper to do math around the end of an extent, handling wrap */
87 static u64 range_end(u64 start, u64 len)
89 if (start + len < start)
94 static int tree_insert(struct rb_root_cached *root, struct extent_map *em)
96 struct rb_node **p = &root->rb_root.rb_node;
97 struct rb_node *parent = NULL;
98 struct extent_map *entry = NULL;
99 struct rb_node *orig_parent = NULL;
100 u64 end = range_end(em->start, em->len);
101 bool leftmost = true;
105 entry = rb_entry(parent, struct extent_map, rb_node);
107 if (em->start < entry->start) {
109 } else if (em->start >= extent_map_end(entry)) {
117 orig_parent = parent;
118 while (parent && em->start >= extent_map_end(entry)) {
119 parent = rb_next(parent);
120 entry = rb_entry(parent, struct extent_map, rb_node);
123 if (end > entry->start && em->start < extent_map_end(entry))
126 parent = orig_parent;
127 entry = rb_entry(parent, struct extent_map, rb_node);
128 while (parent && em->start < entry->start) {
129 parent = rb_prev(parent);
130 entry = rb_entry(parent, struct extent_map, rb_node);
133 if (end > entry->start && em->start < extent_map_end(entry))
136 rb_link_node(&em->rb_node, orig_parent, p);
137 rb_insert_color_cached(&em->rb_node, root, leftmost);
142 * search through the tree for an extent_map with a given offset. If
143 * it can't be found, try to find some neighboring extents
145 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
146 struct rb_node **prev_ret,
147 struct rb_node **next_ret)
149 struct rb_node *n = root->rb_node;
150 struct rb_node *prev = NULL;
151 struct rb_node *orig_prev = NULL;
152 struct extent_map *entry;
153 struct extent_map *prev_entry = NULL;
156 entry = rb_entry(n, struct extent_map, rb_node);
160 if (offset < entry->start)
162 else if (offset >= extent_map_end(entry))
170 while (prev && offset >= extent_map_end(prev_entry)) {
171 prev = rb_next(prev);
172 prev_entry = rb_entry(prev, struct extent_map, rb_node);
179 prev_entry = rb_entry(prev, struct extent_map, rb_node);
180 while (prev && offset < prev_entry->start) {
181 prev = rb_prev(prev);
182 prev_entry = rb_entry(prev, struct extent_map, rb_node);
189 /* check to see if two extent_map structs are adjacent and safe to merge */
190 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
192 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
196 * don't merge compressed extents, we need to know their
199 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
202 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
203 test_bit(EXTENT_FLAG_LOGGING, &next->flags))
207 * We don't want to merge stuff that hasn't been written to the log yet
208 * since it may not reflect exactly what is on disk, and that would be
211 if (!list_empty(&prev->list) || !list_empty(&next->list))
214 ASSERT(next->block_start != EXTENT_MAP_DELALLOC &&
215 prev->block_start != EXTENT_MAP_DELALLOC);
217 if (extent_map_end(prev) == next->start &&
218 prev->flags == next->flags &&
219 prev->bdev == next->bdev &&
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)))) {
231 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
233 struct extent_map *merge = NULL;
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
244 if (refcount_read(&em->refs) > 2)
247 if (em->start != 0) {
248 rb = rb_prev(&em->rb_node);
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);
261 rb_erase_cached(&merge->rb_node, &tree->map);
262 RB_CLEAR_NODE(&merge->rb_node);
263 free_extent_map(merge);
267 rb = rb_next(&em->rb_node);
269 merge = rb_entry(rb, struct extent_map, rb_node);
270 if (rb && mergable_maps(em, merge)) {
271 em->len += merge->len;
272 em->block_len += merge->block_len;
273 rb_erase_cached(&merge->rb_node, &tree->map);
274 RB_CLEAR_NODE(&merge->rb_node);
275 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
276 em->generation = max(em->generation, merge->generation);
277 free_extent_map(merge);
282 * unpin_extent_cache - unpin an extent from the cache
283 * @tree: tree to unpin the extent in
284 * @start: logical offset in the file
285 * @len: length of the extent
286 * @gen: generation that this extent has been modified in
288 * Called after an extent has been written to disk properly. Set the generation
289 * to the generation that actually added the file item to the inode so we know
290 * we need to sync this extent when we call fsync().
292 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
296 struct extent_map *em;
297 bool prealloc = false;
299 write_lock(&tree->lock);
300 em = lookup_extent_mapping(tree, start, len);
302 WARN_ON(!em || em->start != start);
307 em->generation = gen;
308 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
309 em->mod_start = em->start;
310 em->mod_len = em->len;
312 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
314 clear_bit(EXTENT_FLAG_FILLING, &em->flags);
317 try_merge_map(tree, em);
320 em->mod_start = em->start;
321 em->mod_len = em->len;
326 write_unlock(&tree->lock);
331 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
333 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
334 if (extent_map_in_tree(em))
335 try_merge_map(tree, em);
338 static inline void setup_extent_mapping(struct extent_map_tree *tree,
339 struct extent_map *em,
342 refcount_inc(&em->refs);
343 em->mod_start = em->start;
344 em->mod_len = em->len;
347 list_move(&em->list, &tree->modified_extents);
349 try_merge_map(tree, em);
352 static void extent_map_device_set_bits(struct extent_map *em, unsigned bits)
354 struct map_lookup *map = em->map_lookup;
355 u64 stripe_size = em->orig_block_len;
358 for (i = 0; i < map->num_stripes; i++) {
359 struct btrfs_bio_stripe *stripe = &map->stripes[i];
360 struct btrfs_device *device = stripe->dev;
362 set_extent_bits_nowait(&device->alloc_state, stripe->physical,
363 stripe->physical + stripe_size - 1, bits);
367 static void extent_map_device_clear_bits(struct extent_map *em, unsigned bits)
369 struct map_lookup *map = em->map_lookup;
370 u64 stripe_size = em->orig_block_len;
373 for (i = 0; i < map->num_stripes; i++) {
374 struct btrfs_bio_stripe *stripe = &map->stripes[i];
375 struct btrfs_device *device = stripe->dev;
377 __clear_extent_bit(&device->alloc_state, stripe->physical,
378 stripe->physical + stripe_size - 1, bits,
379 0, 0, NULL, GFP_NOWAIT, NULL);
384 * add_extent_mapping - add new extent map to the extent tree
385 * @tree: tree to insert new map in
388 * Insert @em into @tree or perform a simple forward/backward merge with
389 * existing mappings. The extent_map struct passed in will be inserted
390 * into the tree directly, with an additional reference taken, or a
391 * reference dropped if the merge attempt was successful.
393 int add_extent_mapping(struct extent_map_tree *tree,
394 struct extent_map *em, int modified)
398 lockdep_assert_held_write(&tree->lock);
400 ret = tree_insert(&tree->map, em);
404 setup_extent_mapping(tree, em, modified);
405 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags)) {
406 extent_map_device_set_bits(em, CHUNK_ALLOCATED);
407 extent_map_device_clear_bits(em, CHUNK_TRIMMED);
413 static struct extent_map *
414 __lookup_extent_mapping(struct extent_map_tree *tree,
415 u64 start, u64 len, int strict)
417 struct extent_map *em;
418 struct rb_node *rb_node;
419 struct rb_node *prev = NULL;
420 struct rb_node *next = NULL;
421 u64 end = range_end(start, len);
423 rb_node = __tree_search(&tree->map.rb_root, start, &prev, &next);
433 em = rb_entry(rb_node, struct extent_map, rb_node);
435 if (strict && !(end > em->start && start < extent_map_end(em)))
438 refcount_inc(&em->refs);
443 * lookup_extent_mapping - lookup extent_map
444 * @tree: tree to lookup in
445 * @start: byte offset to start the search
446 * @len: length of the lookup range
448 * Find and return the first extent_map struct in @tree that intersects the
449 * [start, len] range. There may be additional objects in the tree that
450 * intersect, so check the object returned carefully to make sure that no
451 * additional lookups are needed.
453 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
456 return __lookup_extent_mapping(tree, start, len, 1);
460 * search_extent_mapping - find a nearby extent map
461 * @tree: tree to lookup in
462 * @start: byte offset to start the search
463 * @len: length of the lookup range
465 * Find and return the first extent_map struct in @tree that intersects the
466 * [start, len] range.
468 * If one can't be found, any nearby extent may be returned
470 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
473 return __lookup_extent_mapping(tree, start, len, 0);
477 * remove_extent_mapping - removes an extent_map from the extent tree
478 * @tree: extent tree to remove from
479 * @em: extent map being removed
481 * Removes @em from @tree. No reference counts are dropped, and no checks
482 * are done to see if the range is in use
484 void remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
486 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
487 rb_erase_cached(&em->rb_node, &tree->map);
488 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
489 list_del_init(&em->list);
490 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
491 extent_map_device_clear_bits(em, CHUNK_ALLOCATED);
492 RB_CLEAR_NODE(&em->rb_node);
495 void replace_extent_mapping(struct extent_map_tree *tree,
496 struct extent_map *cur,
497 struct extent_map *new,
500 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
501 ASSERT(extent_map_in_tree(cur));
502 if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
503 list_del_init(&cur->list);
504 rb_replace_node_cached(&cur->rb_node, &new->rb_node, &tree->map);
505 RB_CLEAR_NODE(&cur->rb_node);
507 setup_extent_mapping(tree, new, modified);
510 static struct extent_map *next_extent_map(struct extent_map *em)
512 struct rb_node *next;
514 next = rb_next(&em->rb_node);
517 return container_of(next, struct extent_map, rb_node);
520 static struct extent_map *prev_extent_map(struct extent_map *em)
522 struct rb_node *prev;
524 prev = rb_prev(&em->rb_node);
527 return container_of(prev, struct extent_map, rb_node);
531 * Helper for btrfs_get_extent. Given an existing extent in the tree,
532 * the existing extent is the nearest extent to map_start,
533 * and an extent that you want to insert, deal with overlap and insert
534 * the best fitted new extent into the tree.
536 static noinline int merge_extent_mapping(struct extent_map_tree *em_tree,
537 struct extent_map *existing,
538 struct extent_map *em,
541 struct extent_map *prev;
542 struct extent_map *next;
547 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
549 if (existing->start > map_start) {
551 prev = prev_extent_map(next);
554 next = next_extent_map(prev);
557 start = prev ? extent_map_end(prev) : em->start;
558 start = max_t(u64, start, em->start);
559 end = next ? next->start : extent_map_end(em);
560 end = min_t(u64, end, extent_map_end(em));
561 start_diff = start - em->start;
563 em->len = end - start;
564 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
565 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
566 em->block_start += start_diff;
567 em->block_len = em->len;
569 return add_extent_mapping(em_tree, em, 0);
573 * btrfs_add_extent_mapping - add extent mapping into em_tree
574 * @fs_info - used for tracepoint
575 * @em_tree - the extent tree into which we want to insert the extent mapping
576 * @em_in - extent we are inserting
577 * @start - start of the logical range btrfs_get_extent() is requesting
578 * @len - length of the logical range btrfs_get_extent() is requesting
580 * Note that @em_in's range may be different from [start, start+len),
581 * but they must be overlapped.
583 * Insert @em_in into @em_tree. In case there is an overlapping range, handle
584 * the -EEXIST by either:
585 * a) Returning the existing extent in @em_in if @start is within the
587 * b) Merge the existing extent with @em_in passed in.
589 * Return 0 on success, otherwise -EEXIST.
592 int btrfs_add_extent_mapping(struct btrfs_fs_info *fs_info,
593 struct extent_map_tree *em_tree,
594 struct extent_map **em_in, u64 start, u64 len)
597 struct extent_map *em = *em_in;
599 ret = add_extent_mapping(em_tree, em, 0);
600 /* it is possible that someone inserted the extent into the tree
601 * while we had the lock dropped. It is also possible that
602 * an overlapping map exists in the tree
604 if (ret == -EEXIST) {
605 struct extent_map *existing;
609 existing = search_extent_mapping(em_tree, start, len);
611 trace_btrfs_handle_em_exist(fs_info, existing, em, start, len);
614 * existing will always be non-NULL, since there must be
615 * extent causing the -EEXIST.
617 if (start >= existing->start &&
618 start < extent_map_end(existing)) {
623 u64 orig_start = em->start;
624 u64 orig_len = em->len;
627 * The existing extent map is the one nearest to
628 * the [start, start + len) range which overlaps
630 ret = merge_extent_mapping(em_tree, existing,
636 "unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n",
637 ret, existing->start, existing->len,
638 orig_start, orig_len);
640 free_extent_map(existing);
644 ASSERT(ret == 0 || ret == -EEXIST);