projects/linux·blob master · refs

1404 lines · 39.8 KiB

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 "extent_map.h"
9#include "compression.h"
10#include "btrfs_inode.h"
11#include "disk-io.h"
12
13
14static struct kmem_cache *extent_map_cache;
15
16int __init btrfs_extent_map_init(void)
17{
18 extent_map_cache = kmem_cache_create("btrfs_extent_map",
19 sizeof(struct extent_map), 0, 0, NULL);
20 if (!extent_map_cache)
21 return -ENOMEM;
22 return 0;
23}
24
25void __cold btrfs_extent_map_exit(void)
26{
27 kmem_cache_destroy(extent_map_cache);
28}
29
30/*
31 * Initialize the extent tree @tree. Should be called for each new inode or
32 * other user of the extent_map interface.
33 */
34void btrfs_extent_map_tree_init(struct extent_map_tree *tree)
35{
36 tree->root = RB_ROOT;
37 INIT_LIST_HEAD(&tree->modified_extents);
38 rwlock_init(&tree->lock);
39}
40
41/*
42 * Allocate a new extent_map structure. The new structure is returned with a
43 * reference count of one and needs to be freed using free_extent_map()
44 */
45struct extent_map *btrfs_alloc_extent_map(void)
46{
47 struct extent_map *em;
48 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
49 if (!em)
50 return NULL;
51 RB_CLEAR_NODE(&em->rb_node);
52 refcount_set(&em->refs, 1);
53 INIT_LIST_HEAD(&em->list);
54 return em;
55}
56
57/*
58 * Drop the reference out on @em by one and free the structure if the reference
59 * count hits zero.
60 */
61void btrfs_free_extent_map(struct extent_map *em)
62{
63 if (!em)
64 return;
65 if (refcount_dec_and_test(&em->refs)) {
66 WARN_ON(btrfs_extent_map_in_tree(em));
67 WARN_ON(!list_empty(&em->list));
68 kmem_cache_free(extent_map_cache, em);
69 }
70}
71
72/* Do the math around the end of an extent, handling wrapping. */
73static u64 range_end(u64 start, u64 len)
74{
75 if (start + len < start)
76 return (u64)-1;
77 return start + len;
78}
79
80static void remove_em(struct btrfs_inode *inode, struct extent_map *em)
81{
82 struct btrfs_fs_info *fs_info = inode->root->fs_info;
83
84 rb_erase(&em->rb_node, &inode->extent_tree.root);
85 RB_CLEAR_NODE(&em->rb_node);
86
87 if (!btrfs_is_testing(fs_info) && btrfs_is_fstree(btrfs_root_id(inode->root)))
88 percpu_counter_dec(&fs_info->evictable_extent_maps);
89}
90
91static int tree_insert(struct rb_root *root, struct extent_map *em)
92{
93 struct rb_node **p = &root->rb_node;
94 struct rb_node *parent = NULL;
95 struct extent_map *entry = NULL;
96 struct rb_node *orig_parent = NULL;
97 u64 end = range_end(em->start, em->len);
98
99 while (*p) {
100 parent = *p;
101 entry = rb_entry(parent, struct extent_map, rb_node);
102
103 if (em->start < entry->start)
104 p = &(*p)->rb_left;
105 else if (em->start >= btrfs_extent_map_end(entry))
106 p = &(*p)->rb_right;
107 else
108 return -EEXIST;
109 }
110
111 orig_parent = parent;
112 while (parent && em->start >= btrfs_extent_map_end(entry)) {
113 parent = rb_next(parent);
114 entry = rb_entry(parent, struct extent_map, rb_node);
115 }
116 if (parent)
117 if (end > entry->start && em->start < btrfs_extent_map_end(entry))
118 return -EEXIST;
119
120 parent = orig_parent;
121 entry = rb_entry(parent, struct extent_map, rb_node);
122 while (parent && em->start < entry->start) {
123 parent = rb_prev(parent);
124 entry = rb_entry(parent, struct extent_map, rb_node);
125 }
126 if (parent)
127 if (end > entry->start && em->start < btrfs_extent_map_end(entry))
128 return -EEXIST;
129
130 rb_link_node(&em->rb_node, orig_parent, p);
131 rb_insert_color(&em->rb_node, root);
132 return 0;
133}
134
135/*
136 * Search through the tree for an extent_map with a given offset. If it can't
137 * be found, try to find some neighboring extents
138 */
139static struct rb_node *tree_search(struct rb_root *root, u64 offset,
140 struct rb_node **prev_or_next_ret)
141{
142 struct rb_node *n = root->rb_node;
143 struct rb_node *prev = NULL;
144 struct rb_node *orig_prev = NULL;
145 struct extent_map *entry;
146 struct extent_map *prev_entry = NULL;
147
148 ASSERT(prev_or_next_ret);
149
150 while (n) {
151 entry = rb_entry(n, struct extent_map, rb_node);
152 prev = n;
153 prev_entry = entry;
154
155 if (offset < entry->start)
156 n = n->rb_left;
157 else if (offset >= btrfs_extent_map_end(entry))
158 n = n->rb_right;
159 else
160 return n;
161 }
162
163 orig_prev = prev;
164 while (prev && offset >= btrfs_extent_map_end(prev_entry)) {
165 prev = rb_next(prev);
166 prev_entry = rb_entry(prev, struct extent_map, rb_node);
167 }
168
169 /*
170 * Previous extent map found, return as in this case the caller does not
171 * care about the next one.
172 */
173 if (prev) {
174 *prev_or_next_ret = prev;
175 return NULL;
176 }
177
178 prev = orig_prev;
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);
183 }
184 *prev_or_next_ret = prev;
185
186 return NULL;
187}
188
189static inline u64 extent_map_block_len(const struct extent_map *em)
190{
191 if (btrfs_extent_map_is_compressed(em))
192 return em->disk_num_bytes;
193 return em->len;
194}
195
196static inline u64 extent_map_block_end(const struct extent_map *em)
197{
198 const u64 block_start = btrfs_extent_map_block_start(em);
199 const u64 block_end = block_start + extent_map_block_len(em);
200
201 if (block_end < block_start)
202 return (u64)-1;
203
204 return block_end;
205}
206
207static bool can_merge_extent_map(const struct extent_map *em)
208{
209 if (em->flags & EXTENT_FLAG_PINNED)
210 return false;
211
212 /* Don't merge compressed extents, we need to know their actual size. */
213 if (btrfs_extent_map_is_compressed(em))
214 return false;
215
216 if (em->flags & EXTENT_FLAG_LOGGING)
217 return false;
218
219 /*
220 * We don't want to merge stuff that hasn't been written to the log yet
221 * since it may not reflect exactly what is on disk, and that would be
222 * bad.
223 */
224 if (!list_empty(&em->list))
225 return false;
226
227 return true;
228}
229
230/* Check to see if two extent_map structs are adjacent and safe to merge. */
231static bool mergeable_maps(const struct extent_map *prev, const struct extent_map *next)
232{
233 if (btrfs_extent_map_end(prev) != next->start)
234 return false;
235
236 /*
237 * The merged flag is not an on-disk flag, it just indicates we had the
238 * extent maps of 2 (or more) adjacent extents merged, so factor it out.
239 */
240 if ((prev->flags & ~EXTENT_FLAG_MERGED) !=
241 (next->flags & ~EXTENT_FLAG_MERGED))
242 return false;
243
244 if (next->disk_bytenr < EXTENT_MAP_LAST_BYTE - 1)
245 return btrfs_extent_map_block_start(next) == extent_map_block_end(prev);
246
247 /* HOLES and INLINE extents. */
248 return next->disk_bytenr == prev->disk_bytenr;
249}
250
251/*
252 * Handle the on-disk data extents merge for @prev and @next.
253 *
254 * @prev: left extent to merge
255 * @next: right extent to merge
256 * @merged: the extent we will not discard after the merge; updated with new values
257 *
258 * After this, one of the two extents is the new merged extent and the other is
259 * removed from the tree and likely freed. Note that @merged is one of @prev/@next
260 * so there is const/non-const aliasing occurring here.
261 *
262 * Only touches disk_bytenr/disk_num_bytes/offset/ram_bytes.
263 * For now only uncompressed regular extent can be merged.
264 */
265static void merge_ondisk_extents(const struct extent_map *prev, const struct extent_map *next,
266 struct extent_map *merged)
267{
268 u64 new_disk_bytenr;
269 u64 new_disk_num_bytes;
270 u64 new_offset;
271
272 /* @prev and @next should not be compressed. */
273 ASSERT(!btrfs_extent_map_is_compressed(prev));
274 ASSERT(!btrfs_extent_map_is_compressed(next));
275
276 /*
277 * There are two different cases where @prev and @next can be merged.
278 *
279 * 1) They are referring to the same data extent:
280 *
281 * |<----- data extent A ----->|
282 * |<- prev ->|<- next ->|
283 *
284 * 2) They are referring to different data extents but still adjacent:
285 *
286 * |<-- data extent A -->|<-- data extent B -->|
287 * |<- prev ->|<- next ->|
288 *
289 * The calculation here always merges the data extents first, then updates
290 * @offset using the new data extents.
291 *
292 * For case 1), the merged data extent would be the same.
293 * For case 2), we just merge the two data extents into one.
294 */
295 new_disk_bytenr = min(prev->disk_bytenr, next->disk_bytenr);
296 new_disk_num_bytes = max(prev->disk_bytenr + prev->disk_num_bytes,
297 next->disk_bytenr + next->disk_num_bytes) -
298 new_disk_bytenr;
299 new_offset = prev->disk_bytenr + prev->offset - new_disk_bytenr;
300
301 merged->disk_bytenr = new_disk_bytenr;
302 merged->disk_num_bytes = new_disk_num_bytes;
303 merged->ram_bytes = new_disk_num_bytes;
304 merged->offset = new_offset;
305}
306
307static void dump_extent_map(struct btrfs_fs_info *fs_info, const char *prefix,
308 struct extent_map *em)
309{
310 if (!IS_ENABLED(CONFIG_BTRFS_DEBUG))
311 return;
312 btrfs_crit(fs_info,
313"%s, start=%llu len=%llu disk_bytenr=%llu disk_num_bytes=%llu ram_bytes=%llu offset=%llu flags=0x%x",
314 prefix, em->start, em->len, em->disk_bytenr, em->disk_num_bytes,
315 em->ram_bytes, em->offset, em->flags);
316 ASSERT(0);
317}
318
319/* Internal sanity checks for btrfs debug builds. */
320static void validate_extent_map(struct btrfs_fs_info *fs_info, struct extent_map *em)
321{
322 const u32 blocksize = fs_info->sectorsize;
323
324 if (!IS_ENABLED(CONFIG_BTRFS_DEBUG))
325 return;
326
327 if (!IS_ALIGNED(em->start, blocksize) ||
328 !IS_ALIGNED(em->len, blocksize))
329 dump_extent_map(fs_info, "unaligned start offset or length members", em);
330
331 if (em->disk_bytenr < EXTENT_MAP_LAST_BYTE) {
332 if (em->disk_num_bytes == 0)
333 dump_extent_map(fs_info, "zero disk_num_bytes", em);
334 if (em->offset + em->len > em->ram_bytes)
335 dump_extent_map(fs_info, "ram_bytes too small", em);
336 if (em->offset + em->len > em->disk_num_bytes &&
337 !btrfs_extent_map_is_compressed(em))
338 dump_extent_map(fs_info, "disk_num_bytes too small", em);
339 if (!btrfs_extent_map_is_compressed(em) &&
340 em->ram_bytes != em->disk_num_bytes)
341 dump_extent_map(fs_info,
342 "ram_bytes mismatch with disk_num_bytes for non-compressed em",
343 em);
344 if (!IS_ALIGNED(em->disk_bytenr, blocksize) ||
345 !IS_ALIGNED(em->disk_num_bytes, blocksize) ||
346 !IS_ALIGNED(em->offset, blocksize) ||
347 !IS_ALIGNED(em->ram_bytes, blocksize))
348 dump_extent_map(fs_info, "unaligned members", em);
349 } else if (em->offset) {
350 dump_extent_map(fs_info, "non-zero offset for hole/inline", em);
351 }
352}
353
354static void try_merge_map(struct btrfs_inode *inode, struct extent_map *em)
355{
356 struct btrfs_fs_info *fs_info = inode->root->fs_info;
357 struct extent_map *merge = NULL;
358 struct rb_node *rb;
359
360 /*
361 * We can't modify an extent map that is in the tree and that is being
362 * used by another task, as it can cause that other task to see it in
363 * inconsistent state during the merging. We always have 1 reference for
364 * the tree and 1 for this task (which is unpinning the extent map or
365 * clearing the logging flag), so anything > 2 means it's being used by
366 * other tasks too.
367 */
368 if (refcount_read(&em->refs) > 2)
369 return;
370
371 if (!can_merge_extent_map(em))
372 return;
373
374 if (em->start != 0) {
375 rb = rb_prev(&em->rb_node);
376 merge = rb_entry_safe(rb, struct extent_map, rb_node);
377
378 if (rb && can_merge_extent_map(merge) && mergeable_maps(merge, em)) {
379 em->start = merge->start;
380 em->len += merge->len;
381 em->generation = max(em->generation, merge->generation);
382
383 if (em->disk_bytenr < EXTENT_MAP_LAST_BYTE)
384 merge_ondisk_extents(merge, em, em);
385 em->flags |= EXTENT_FLAG_MERGED;
386
387 validate_extent_map(fs_info, em);
388 remove_em(inode, merge);
389 btrfs_free_extent_map(merge);
390 }
391 }
392
393 rb = rb_next(&em->rb_node);
394 merge = rb_entry_safe(rb, struct extent_map, rb_node);
395
396 if (rb && can_merge_extent_map(merge) && mergeable_maps(em, merge)) {
397 em->len += merge->len;
398 if (em->disk_bytenr < EXTENT_MAP_LAST_BYTE)
399 merge_ondisk_extents(em, merge, em);
400 validate_extent_map(fs_info, em);
401 em->generation = max(em->generation, merge->generation);
402 em->flags |= EXTENT_FLAG_MERGED;
403 remove_em(inode, merge);
404 btrfs_free_extent_map(merge);
405 }
406}
407
408/*
409 * Unpin an extent from the cache.
410 *
411 * @inode: the inode from which we are unpinning an extent range
412 * @start: logical offset in the file
413 * @len: length of the extent
414 * @gen: generation that this extent has been modified in
415 *
416 * Called after an extent has been written to disk properly. Set the generation
417 * to the generation that actually added the file item to the inode so we know
418 * we need to sync this extent when we call fsync().
419 *
420 * Returns: 0 on success
421 * -ENOENT when the extent is not found in the tree
422 * -EUCLEAN if the found extent does not match the expected start
423 */
424int btrfs_unpin_extent_cache(struct btrfs_inode *inode, u64 start, u64 len, u64 gen)
425{
426 struct btrfs_fs_info *fs_info = inode->root->fs_info;
427 struct extent_map_tree *tree = &inode->extent_tree;
428 int ret = 0;
429 struct extent_map *em;
430
431 write_lock(&tree->lock);
432 em = btrfs_lookup_extent_mapping(tree, start, len);
433
434 if (WARN_ON(!em)) {
435 btrfs_warn(fs_info,
436"no extent map found for inode %llu (root %lld) when unpinning extent range [%llu, %llu), generation %llu",
437 btrfs_ino(inode), btrfs_root_id(inode->root),
438 start, start + len, gen);
439 ret = -ENOENT;
440 goto out;
441 }
442
443 if (WARN_ON(em->start != start)) {
444 btrfs_warn(fs_info,
445"found extent map for inode %llu (root %lld) with unexpected start offset %llu when unpinning extent range [%llu, %llu), generation %llu",
446 btrfs_ino(inode), btrfs_root_id(inode->root),
447 em->start, start, start + len, gen);
448 ret = -EUCLEAN;
449 goto out;
450 }
451
452 em->generation = gen;
453 em->flags &= ~EXTENT_FLAG_PINNED;
454
455 try_merge_map(inode, em);
456
457out:
458 write_unlock(&tree->lock);
459 btrfs_free_extent_map(em);
460 return ret;
461
462}
463
464void btrfs_clear_em_logging(struct btrfs_inode *inode, struct extent_map *em)
465{
466 lockdep_assert_held_write(&inode->extent_tree.lock);
467
468 em->flags &= ~EXTENT_FLAG_LOGGING;
469 if (btrfs_extent_map_in_tree(em))
470 try_merge_map(inode, em);
471}
472
473static inline void setup_extent_mapping(struct btrfs_inode *inode,
474 struct extent_map *em,
475 bool modified)
476{
477 refcount_inc(&em->refs);
478
479 ASSERT(list_empty(&em->list));
480
481 if (modified)
482 list_add(&em->list, &inode->extent_tree.modified_extents);
483 else
484 try_merge_map(inode, em);
485}
486
487/*
488 * Add a new extent map to an inode's extent map tree.
489 *
490 * @inode: the target inode
491 * @em: map to insert
492 * @modified: indicate whether the given @em should be added to the
493 * modified list, which indicates the extent needs to be logged
494 *
495 * Insert @em into the @inode's extent map tree or perform a simple
496 * forward/backward merge with existing mappings. The extent_map struct passed
497 * in will be inserted into the tree directly, with an additional reference
498 * taken, or a reference dropped if the merge attempt was successful.
499 */
500static int add_extent_mapping(struct btrfs_inode *inode,
501 struct extent_map *em, bool modified)
502{
503 struct extent_map_tree *tree = &inode->extent_tree;
504 struct btrfs_root *root = inode->root;
505 struct btrfs_fs_info *fs_info = root->fs_info;
506 int ret;
507
508 lockdep_assert_held_write(&tree->lock);
509
510 validate_extent_map(fs_info, em);
511 ret = tree_insert(&tree->root, em);
512 if (ret)
513 return ret;
514
515 setup_extent_mapping(inode, em, modified);
516
517 if (!btrfs_is_testing(fs_info) && btrfs_is_fstree(btrfs_root_id(root)))
518 percpu_counter_inc(&fs_info->evictable_extent_maps);
519
520 return 0;
521}
522
523static struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
524 u64 start, u64 len, bool strict)
525{
526 struct extent_map *em;
527 struct rb_node *rb_node;
528 struct rb_node *prev_or_next = NULL;
529 u64 end = range_end(start, len);
530
531 rb_node = tree_search(&tree->root, start, &prev_or_next);
532 if (!rb_node) {
533 if (prev_or_next)
534 rb_node = prev_or_next;
535 else
536 return NULL;
537 }
538
539 em = rb_entry(rb_node, struct extent_map, rb_node);
540
541 if (strict && !(end > em->start && start < btrfs_extent_map_end(em)))
542 return NULL;
543
544 refcount_inc(&em->refs);
545 return em;
546}
547
548/*
549 * Lookup extent_map that intersects @start + @len range.
550 *
551 * @tree: tree to lookup in
552 * @start: byte offset to start the search
553 * @len: length of the lookup range
554 *
555 * Find and return the first extent_map struct in @tree that intersects the
556 * [start, len] range. There may be additional objects in the tree that
557 * intersect, so check the object returned carefully to make sure that no
558 * additional lookups are needed.
559 */
560struct extent_map *btrfs_lookup_extent_mapping(struct extent_map_tree *tree,
561 u64 start, u64 len)
562{
563 return lookup_extent_mapping(tree, start, len, true);
564}
565
566/*
567 * Find a nearby extent map intersecting @start + @len (not an exact search).
568 *
569 * @tree: tree to lookup in
570 * @start: byte offset to start the search
571 * @len: length of the lookup range
572 *
573 * Find and return the first extent_map struct in @tree that intersects the
574 * [start, len] range.
575 *
576 * If one can't be found, any nearby extent may be returned
577 */
578struct extent_map *btrfs_search_extent_mapping(struct extent_map_tree *tree,
579 u64 start, u64 len)
580{
581 return lookup_extent_mapping(tree, start, len, false);
582}
583
584/*
585 * Remove an extent_map from its inode's extent tree.
586 *
587 * @inode: the inode the extent map belongs to
588 * @em: extent map being removed
589 *
590 * Remove @em from the extent tree of @inode. No reference counts are dropped,
591 * and no checks are done to see if the range is in use.
592 */
593void btrfs_remove_extent_mapping(struct btrfs_inode *inode, struct extent_map *em)
594{
595 struct extent_map_tree *tree = &inode->extent_tree;
596
597 lockdep_assert_held_write(&tree->lock);
598
599 WARN_ON(em->flags & EXTENT_FLAG_PINNED);
600 if (!(em->flags & EXTENT_FLAG_LOGGING))
601 list_del_init(&em->list);
602
603 remove_em(inode, em);
604}
605
606static void replace_extent_mapping(struct btrfs_inode *inode,
607 struct extent_map *cur,
608 struct extent_map *new,
609 bool modified)
610{
611 struct btrfs_fs_info *fs_info = inode->root->fs_info;
612 struct extent_map_tree *tree = &inode->extent_tree;
613
614 lockdep_assert_held_write(&tree->lock);
615
616 validate_extent_map(fs_info, new);
617
618 WARN_ON(cur->flags & EXTENT_FLAG_PINNED);
619 ASSERT(btrfs_extent_map_in_tree(cur));
620 if (!(cur->flags & EXTENT_FLAG_LOGGING))
621 list_del_init(&cur->list);
622 rb_replace_node(&cur->rb_node, &new->rb_node, &tree->root);
623 RB_CLEAR_NODE(&cur->rb_node);
624
625 setup_extent_mapping(inode, new, modified);
626}
627
628static struct extent_map *next_extent_map(const struct extent_map *em)
629{
630 struct rb_node *next;
631
632 next = rb_next(&em->rb_node);
633 if (!next)
634 return NULL;
635 return container_of(next, struct extent_map, rb_node);
636}
637
638static struct extent_map *prev_extent_map(struct extent_map *em)
639{
640 struct rb_node *prev;
641
642 prev = rb_prev(&em->rb_node);
643 if (!prev)
644 return NULL;
645 return container_of(prev, struct extent_map, rb_node);
646}
647
648/*
649 * Helper for btrfs_get_extent. Given an existing extent in the tree,
650 * the existing extent is the nearest extent to map_start,
651 * and an extent that you want to insert, deal with overlap and insert
652 * the best fitted new extent into the tree.
653 */
654static noinline int merge_extent_mapping(struct btrfs_inode *inode,
655 struct extent_map *existing,
656 struct extent_map *em,
657 u64 map_start)
658{
659 struct extent_map *prev;
660 struct extent_map *next;
661 u64 start;
662 u64 end;
663 u64 start_diff;
664
665 if (map_start < em->start || map_start >= btrfs_extent_map_end(em))
666 return -EINVAL;
667
668 if (existing->start > map_start) {
669 next = existing;
670 prev = prev_extent_map(next);
671 } else {
672 prev = existing;
673 next = next_extent_map(prev);
674 }
675
676 start = prev ? btrfs_extent_map_end(prev) : em->start;
677 start = max_t(u64, start, em->start);
678 end = next ? next->start : btrfs_extent_map_end(em);
679 end = min_t(u64, end, btrfs_extent_map_end(em));
680 start_diff = start - em->start;
681 em->start = start;
682 em->len = end - start;
683 if (em->disk_bytenr < EXTENT_MAP_LAST_BYTE)
684 em->offset += start_diff;
685 return add_extent_mapping(inode, em, false);
686}
687
688/*
689 * Add extent mapping into an inode's extent map tree.
690 *
691 * @inode: target inode
692 * @em_in: extent we are inserting
693 * @start: start of the logical range btrfs_get_extent() is requesting
694 * @len: length of the logical range btrfs_get_extent() is requesting
695 *
696 * Note that @em_in's range may be different from [start, start+len),
697 * but they must be overlapped.
698 *
699 * Insert @em_in into the inode's extent map tree. In case there is an
700 * overlapping range, handle the -EEXIST by either:
701 * a) Returning the existing extent in @em_in if @start is within the
702 * existing em.
703 * b) Merge the existing extent with @em_in passed in.
704 *
705 * Return 0 on success, otherwise -EEXIST.
706 *
707 */
708int btrfs_add_extent_mapping(struct btrfs_inode *inode,
709 struct extent_map **em_in, u64 start, u64 len)
710{
711 int ret;
712 struct extent_map *em = *em_in;
713 struct btrfs_fs_info *fs_info = inode->root->fs_info;
714
715 /*
716 * Tree-checker should have rejected any inline extent with non-zero
717 * file offset. Here just do a sanity check.
718 */
719 if (em->disk_bytenr == EXTENT_MAP_INLINE)
720 ASSERT(em->start == 0, "em->start=%llu", em->start);
721
722 ret = add_extent_mapping(inode, em, false);
723 /* it is possible that someone inserted the extent into the tree
724 * while we had the lock dropped. It is also possible that
725 * an overlapping map exists in the tree
726 */
727 if (ret == -EEXIST) {
728 struct extent_map *existing;
729
730 existing = btrfs_search_extent_mapping(&inode->extent_tree, start, len);
731
732 trace_btrfs_handle_em_exist(fs_info, existing, em, start, len);
733
734 /*
735 * existing will always be non-NULL, since there must be
736 * extent causing the -EEXIST.
737 */
738 if (start >= existing->start &&
739 start < btrfs_extent_map_end(existing)) {
740 btrfs_free_extent_map(em);
741 *em_in = existing;
742 ret = 0;
743 } else {
744 u64 orig_start = em->start;
745 u64 orig_len = em->len;
746
747 /*
748 * The existing extent map is the one nearest to
749 * the [start, start + len) range which overlaps
750 */
751 ret = merge_extent_mapping(inode, existing, em, start);
752 if (WARN_ON(ret)) {
753 btrfs_free_extent_map(em);
754 *em_in = NULL;
755 btrfs_warn(fs_info,
756"extent map merge error existing [%llu, %llu) with em [%llu, %llu) start %llu",
757 existing->start, btrfs_extent_map_end(existing),
758 orig_start, orig_start + orig_len, start);
759 }
760 btrfs_free_extent_map(existing);
761 }
762 }
763
764 ASSERT(ret == 0 || ret == -EEXIST, "ret=%d", ret);
765 return ret;
766}
767
768/*
769 * Drop all extent maps from a tree in the fastest possible way, rescheduling
770 * if needed. This avoids searching the tree, from the root down to the first
771 * extent map, before each deletion.
772 */
773static void drop_all_extent_maps_fast(struct btrfs_inode *inode)
774{
775 struct extent_map_tree *tree = &inode->extent_tree;
776 struct rb_node *node;
777
778 write_lock(&tree->lock);
779 node = rb_first(&tree->root);
780 while (node) {
781 struct extent_map *em;
782 struct rb_node *next = rb_next(node);
783
784 em = rb_entry(node, struct extent_map, rb_node);
785 em->flags &= ~(EXTENT_FLAG_PINNED | EXTENT_FLAG_LOGGING);
786 btrfs_remove_extent_mapping(inode, em);
787 btrfs_free_extent_map(em);
788
789 if (cond_resched_rwlock_write(&tree->lock))
790 node = rb_first(&tree->root);
791 else
792 node = next;
793 }
794 write_unlock(&tree->lock);
795}
796
797/*
798 * Drop all extent maps in a given range.
799 *
800 * @inode: The target inode.
801 * @start: Start offset of the range.
802 * @end: End offset of the range (inclusive value).
803 * @skip_pinned: Indicate if pinned extent maps should be ignored or not.
804 *
805 * This drops all the extent maps that intersect the given range [@start, @end].
806 * Extent maps that partially overlap the range and extend behind or beyond it,
807 * are split.
808 * The caller should have locked an appropriate file range in the inode's io
809 * tree before calling this function.
810 */
811void btrfs_drop_extent_map_range(struct btrfs_inode *inode, u64 start, u64 end,
812 bool skip_pinned)
813{
814 struct extent_map *split;
815 struct extent_map *split2;
816 struct extent_map *em;
817 struct extent_map_tree *em_tree = &inode->extent_tree;
818 u64 len = end - start + 1;
819
820 WARN_ON(end < start);
821 if (end == (u64)-1) {
822 if (start == 0 && !skip_pinned) {
823 drop_all_extent_maps_fast(inode);
824 return;
825 }
826 len = (u64)-1;
827 } else {
828 /* Make end offset exclusive for use in the loop below. */
829 end++;
830 }
831
832 /*
833 * It's ok if we fail to allocate the extent maps, see the comment near
834 * the bottom of the loop below. We only need two spare extent maps in
835 * the worst case, where the first extent map that intersects our range
836 * starts before the range and the last extent map that intersects our
837 * range ends after our range (and they might be the same extent map),
838 * because we need to split those two extent maps at the boundaries.
839 */
840 split = btrfs_alloc_extent_map();
841 split2 = btrfs_alloc_extent_map();
842
843 write_lock(&em_tree->lock);
844 em = btrfs_lookup_extent_mapping(em_tree, start, len);
845
846 while (em) {
847 /* extent_map_end() returns exclusive value (last byte + 1). */
848 const u64 em_end = btrfs_extent_map_end(em);
849 struct extent_map *next_em = NULL;
850 u64 gen;
851 unsigned long flags;
852 bool modified;
853
854 if (em_end < end) {
855 next_em = next_extent_map(em);
856 if (next_em) {
857 if (next_em->start < end)
858 refcount_inc(&next_em->refs);
859 else
860 next_em = NULL;
861 }
862 }
863
864 if (skip_pinned && (em->flags & EXTENT_FLAG_PINNED)) {
865 start = em_end;
866 goto next;
867 }
868
869 /*
870 * In case we split the extent map, we want to preserve the
871 * EXTENT_FLAG_LOGGING flag on our extent map, but we don't want
872 * it on the new extent maps.
873 */
874 flags = em->flags & ~EXTENT_FLAG_LOGGING;
875 em->flags &= ~EXTENT_FLAG_PINNED;
876 modified = !list_empty(&em->list);
877
878 /*
879 * The extent map does not cross our target range, so no need to
880 * split it, we can remove it directly.
881 */
882 if (em->start >= start && em_end <= end)
883 goto remove_em;
884
885 gen = em->generation;
886
887 if (em->start < start) {
888 if (!split) {
889 split = split2;
890 split2 = NULL;
891 if (!split)
892 goto remove_em;
893 }
894 split->start = em->start;
895 split->len = start - em->start;
896
897 if (em->disk_bytenr < EXTENT_MAP_LAST_BYTE) {
898 split->disk_bytenr = em->disk_bytenr;
899 split->disk_num_bytes = em->disk_num_bytes;
900 split->offset = em->offset;
901 split->ram_bytes = em->ram_bytes;
902 } else {
903 split->disk_bytenr = em->disk_bytenr;
904 split->disk_num_bytes = 0;
905 split->offset = 0;
906 split->ram_bytes = split->len;
907 }
908
909 split->generation = gen;
910 split->flags = flags;
911 replace_extent_mapping(inode, em, split, modified);
912 btrfs_free_extent_map(split);
913 split = split2;
914 split2 = NULL;
915 }
916 if (em_end > end) {
917 if (!split) {
918 split = split2;
919 split2 = NULL;
920 if (!split)
921 goto remove_em;
922 }
923 split->start = end;
924 split->len = em_end - end;
925 split->disk_bytenr = em->disk_bytenr;
926 split->flags = flags;
927 split->generation = gen;
928
929 if (em->disk_bytenr < EXTENT_MAP_LAST_BYTE) {
930 split->disk_num_bytes = em->disk_num_bytes;
931 split->offset = em->offset + end - em->start;
932 split->ram_bytes = em->ram_bytes;
933 } else {
934 split->disk_num_bytes = 0;
935 split->offset = 0;
936 split->ram_bytes = split->len;
937 }
938
939 if (btrfs_extent_map_in_tree(em)) {
940 replace_extent_mapping(inode, em, split, modified);
941 } else {
942 int ret;
943
944 ret = add_extent_mapping(inode, split, modified);
945 /* Logic error, shouldn't happen. */
946 ASSERT(ret == 0, "ret=%d", ret);
947 if (WARN_ON(ret != 0) && modified)
948 btrfs_set_inode_full_sync(inode);
949 }
950 btrfs_free_extent_map(split);
951 split = NULL;
952 }
953remove_em:
954 if (btrfs_extent_map_in_tree(em)) {
955 /*
956 * If the extent map is still in the tree it means that
957 * either of the following is true:
958 *
959 * 1) It fits entirely in our range (doesn't end beyond
960 * it or starts before it);
961 *
962 * 2) It starts before our range and/or ends after our
963 * range, and we were not able to allocate the extent
964 * maps for split operations, @split and @split2.
965 *
966 * If we are at case 2) then we just remove the entire
967 * extent map - this is fine since if anyone needs it to
968 * access the subranges outside our range, will just
969 * load it again from the subvolume tree's file extent
970 * item. However if the extent map was in the list of
971 * modified extents, then we must mark the inode for a
972 * full fsync, otherwise a fast fsync will miss this
973 * extent if it's new and needs to be logged.
974 */
975 if ((em->start < start || em_end > end) && modified) {
976 ASSERT(!split);
977 btrfs_set_inode_full_sync(inode);
978 }
979 btrfs_remove_extent_mapping(inode, em);
980 }
981
982 /*
983 * Once for the tree reference (we replaced or removed the
984 * extent map from the tree).
985 */
986 btrfs_free_extent_map(em);
987next:
988 /* Once for us (for our lookup reference). */
989 btrfs_free_extent_map(em);
990
991 em = next_em;
992 }
993
994 write_unlock(&em_tree->lock);
995
996 btrfs_free_extent_map(split);
997 btrfs_free_extent_map(split2);
998}
999
1000/*
1001 * Replace a range in the inode's extent map tree with a new extent map.
1002 *
1003 * @inode: The target inode.
1004 * @new_em: The new extent map to add to the inode's extent map tree.
1005 * @modified: Indicate if the new extent map should be added to the list of
1006 * modified extents (for fast fsync tracking).
1007 *
1008 * Drops all the extent maps in the inode's extent map tree that intersect the
1009 * range of the new extent map and adds the new extent map to the tree.
1010 * The caller should have locked an appropriate file range in the inode's io
1011 * tree before calling this function.
1012 */
1013int btrfs_replace_extent_map_range(struct btrfs_inode *inode,
1014 struct extent_map *new_em,
1015 bool modified)
1016{
1017 const u64 end = new_em->start + new_em->len - 1;
1018 struct extent_map_tree *tree = &inode->extent_tree;
1019 int ret;
1020
1021 ASSERT(!btrfs_extent_map_in_tree(new_em));
1022
1023 /*
1024 * The caller has locked an appropriate file range in the inode's io
1025 * tree, but getting -EEXIST when adding the new extent map can still
1026 * happen in case there are extents that partially cover the range, and
1027 * this is due to two tasks operating on different parts of the extent.
1028 * See commit 18e83ac75bfe67 ("Btrfs: fix unexpected EEXIST from
1029 * btrfs_get_extent") for an example and details.
1030 */
1031 do {
1032 btrfs_drop_extent_map_range(inode, new_em->start, end, false);
1033 write_lock(&tree->lock);
1034 ret = add_extent_mapping(inode, new_em, modified);
1035 write_unlock(&tree->lock);
1036 } while (ret == -EEXIST);
1037
1038 return ret;
1039}
1040
1041/*
1042 * Split off the first pre bytes from the extent_map at [start, start + len],
1043 * and set the block_start for it to new_logical.
1044 *
1045 * This function is used when an ordered_extent needs to be split.
1046 */
1047int btrfs_split_extent_map(struct btrfs_inode *inode, u64 start, u64 len, u64 pre,
1048 u64 new_logical)
1049{
1050 struct extent_map_tree *em_tree = &inode->extent_tree;
1051 struct extent_map *em;
1052 struct extent_map *split_pre = NULL;
1053 struct extent_map *split_mid = NULL;
1054 int ret = 0;
1055 unsigned long flags;
1056
1057 ASSERT(pre != 0);
1058 ASSERT(pre < len);
1059
1060 split_pre = btrfs_alloc_extent_map();
1061 if (!split_pre)
1062 return -ENOMEM;
1063 split_mid = btrfs_alloc_extent_map();
1064 if (!split_mid) {
1065 ret = -ENOMEM;
1066 goto out_free_pre;
1067 }
1068
1069 btrfs_lock_extent(&inode->io_tree, start, start + len - 1, NULL);
1070 write_lock(&em_tree->lock);
1071 em = btrfs_lookup_extent_mapping(em_tree, start, len);
1072 if (unlikely(!em)) {
1073 ret = -EIO;
1074 goto out_unlock;
1075 }
1076
1077 ASSERT(em->len == len);
1078 ASSERT(!btrfs_extent_map_is_compressed(em));
1079 ASSERT(em->disk_bytenr < EXTENT_MAP_LAST_BYTE);
1080 ASSERT(em->flags & EXTENT_FLAG_PINNED);
1081 ASSERT(!(em->flags & EXTENT_FLAG_LOGGING));
1082 ASSERT(!list_empty(&em->list));
1083
1084 flags = em->flags;
1085 em->flags &= ~EXTENT_FLAG_PINNED;
1086
1087 /* First, replace the em with a new extent_map starting from * em->start */
1088 split_pre->start = em->start;
1089 split_pre->len = pre;
1090 split_pre->disk_bytenr = new_logical;
1091 split_pre->disk_num_bytes = split_pre->len;
1092 split_pre->offset = 0;
1093 split_pre->ram_bytes = split_pre->len;
1094 split_pre->flags = flags;
1095 split_pre->generation = em->generation;
1096
1097 replace_extent_mapping(inode, em, split_pre, true);
1098
1099 /*
1100 * Now we only have an extent_map at:
1101 * [em->start, em->start + pre]
1102 */
1103
1104 /* Insert the middle extent_map. */
1105 split_mid->start = em->start + pre;
1106 split_mid->len = em->len - pre;
1107 split_mid->disk_bytenr = btrfs_extent_map_block_start(em) + pre;
1108 split_mid->disk_num_bytes = split_mid->len;
1109 split_mid->offset = 0;
1110 split_mid->ram_bytes = split_mid->len;
1111 split_mid->flags = flags;
1112 split_mid->generation = em->generation;
1113 add_extent_mapping(inode, split_mid, true);
1114
1115 /* Once for us */
1116 btrfs_free_extent_map(em);
1117 /* Once for the tree */
1118 btrfs_free_extent_map(em);
1119
1120out_unlock:
1121 write_unlock(&em_tree->lock);
1122 btrfs_unlock_extent(&inode->io_tree, start, start + len - 1, NULL);
1123 btrfs_free_extent_map(split_mid);
1124out_free_pre:
1125 btrfs_free_extent_map(split_pre);
1126 return ret;
1127}
1128
1129struct btrfs_em_shrink_ctx {
1130 long nr_to_scan;
1131 long scanned;
1132};
1133
1134static long btrfs_scan_inode(struct btrfs_inode *inode, struct btrfs_em_shrink_ctx *ctx)
1135{
1136 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1137 const u64 cur_fs_gen = btrfs_get_fs_generation(fs_info);
1138 struct extent_map_tree *tree = &inode->extent_tree;
1139 long nr_dropped = 0;
1140 struct rb_node *node;
1141
1142 lockdep_assert_held_write(&tree->lock);
1143
1144 /*
1145 * Take the mmap lock so that we serialize with the inode logging phase
1146 * of fsync because we may need to set the full sync flag on the inode,
1147 * in case we have to remove extent maps in the tree's list of modified
1148 * extents. If we set the full sync flag in the inode while an fsync is
1149 * in progress, we may risk missing new extents because before the flag
1150 * is set, fsync decides to only wait for writeback to complete and then
1151 * during inode logging it sees the flag set and uses the subvolume tree
1152 * to find new extents, which may not be there yet because ordered
1153 * extents haven't completed yet.
1154 *
1155 * We also do a try lock because we don't want to block for too long and
1156 * we are holding the extent map tree's lock in write mode.
1157 */
1158 if (!down_read_trylock(&inode->i_mmap_lock))
1159 return 0;
1160
1161 node = rb_first(&tree->root);
1162 while (node) {
1163 struct rb_node *next = rb_next(node);
1164 struct extent_map *em;
1165
1166 em = rb_entry(node, struct extent_map, rb_node);
1167 ctx->scanned++;
1168
1169 if (em->flags & EXTENT_FLAG_PINNED)
1170 goto next;
1171
1172 /*
1173 * If the inode is in the list of modified extents (new) and its
1174 * generation is the same (or is greater than) the current fs
1175 * generation, it means it was not yet persisted so we have to
1176 * set the full sync flag so that the next fsync will not miss
1177 * it.
1178 */
1179 if (!list_empty(&em->list) && em->generation >= cur_fs_gen)
1180 btrfs_set_inode_full_sync(inode);
1181
1182 btrfs_remove_extent_mapping(inode, em);
1183 trace_btrfs_extent_map_shrinker_remove_em(inode, em);
1184 /* Drop the reference for the tree. */
1185 btrfs_free_extent_map(em);
1186 nr_dropped++;
1187next:
1188 if (ctx->scanned >= ctx->nr_to_scan)
1189 break;
1190
1191 /*
1192 * Stop if we need to reschedule or there's contention on the
1193 * lock. This is to avoid slowing other tasks trying to take the
1194 * lock.
1195 */
1196 if (need_resched() || rwlock_needbreak(&tree->lock) ||
1197 btrfs_fs_closing(fs_info))
1198 break;
1199 node = next;
1200 }
1201 up_read(&inode->i_mmap_lock);
1202
1203 return nr_dropped;
1204}
1205
1206static struct btrfs_inode *find_first_inode_to_shrink(struct btrfs_root *root,
1207 u64 min_ino)
1208{
1209 struct btrfs_inode *inode;
1210 unsigned long from = min_ino;
1211
1212 xa_lock(&root->inodes);
1213 while (true) {
1214 struct extent_map_tree *tree;
1215
1216 inode = xa_find(&root->inodes, &from, ULONG_MAX, XA_PRESENT);
1217 if (!inode)
1218 break;
1219
1220 tree = &inode->extent_tree;
1221
1222 /*
1223 * Most inodes have no extent maps, so check without the lock.
1224 * The race is harmless: a false empty just defers the inode to
1225 * a later scan, and a false non-empty is caught under the lock.
1226 */
1227 if (data_race(RB_EMPTY_ROOT(&tree->root)))
1228 goto next;
1229
1230 /*
1231 * We want to be fast so if the lock is busy we don't want to
1232 * spend time waiting for it (some task is about to do IO for
1233 * the inode).
1234 */
1235 if (!write_trylock(&tree->lock))
1236 goto next;
1237
1238 /*
1239 * Skip inode if it doesn't have loaded extent maps, so we avoid
1240 * getting a reference and doing an iput later. This includes
1241 * cases like files that were opened for things like stat(2), or
1242 * files with all extent maps previously released through the
1243 * release folio callback (btrfs_release_folio()) or released in
1244 * a previous run, or directories which never have extent maps.
1245 */
1246 if (RB_EMPTY_ROOT(&tree->root)) {
1247 write_unlock(&tree->lock);
1248 goto next;
1249 }
1250
1251 if (igrab(&inode->vfs_inode))
1252 break;
1253
1254 write_unlock(&tree->lock);
1255next:
1256 from = btrfs_ino(inode) + 1;
1257 xa_unlock(&root->inodes);
1258 cond_resched();
1259 xa_lock(&root->inodes);
1260 }
1261 xa_unlock(&root->inodes);
1262
1263 return inode;
1264}
1265
1266static long btrfs_scan_root(struct btrfs_root *root, struct btrfs_em_shrink_ctx *ctx)
1267{
1268 struct btrfs_fs_info *fs_info = root->fs_info;
1269 struct btrfs_inode *inode;
1270 long nr_dropped = 0;
1271 u64 min_ino = fs_info->em_shrinker_last_ino + 1;
1272
1273 inode = find_first_inode_to_shrink(root, min_ino);
1274 while (inode) {
1275 nr_dropped += btrfs_scan_inode(inode, ctx);
1276 write_unlock(&inode->extent_tree.lock);
1277
1278 min_ino = btrfs_ino(inode) + 1;
1279 fs_info->em_shrinker_last_ino = btrfs_ino(inode);
1280 iput(&inode->vfs_inode);
1281
1282 if (ctx->scanned >= ctx->nr_to_scan || btrfs_fs_closing(fs_info))
1283 break;
1284
1285 cond_resched();
1286
1287 inode = find_first_inode_to_shrink(root, min_ino);
1288 }
1289
1290 if (inode) {
1291 /*
1292 * There are still inodes in this root or we happened to process
1293 * the last one and reached the scan limit. In either case set
1294 * the current root to this one, so we'll resume from the next
1295 * inode if there is one or we will find out this was the last
1296 * one and move to the next root.
1297 */
1298 fs_info->em_shrinker_last_root = btrfs_root_id(root);
1299 } else {
1300 /*
1301 * No more inodes in this root, set extent_map_shrinker_last_ino to 0 so
1302 * that when processing the next root we start from its first inode.
1303 */
1304 fs_info->em_shrinker_last_ino = 0;
1305 fs_info->em_shrinker_last_root = btrfs_root_id(root) + 1;
1306 }
1307
1308 return nr_dropped;
1309}
1310
1311static void btrfs_extent_map_shrinker_worker(struct work_struct *work)
1312{
1313 struct btrfs_fs_info *fs_info;
1314 struct btrfs_em_shrink_ctx ctx;
1315 u64 start_root_id;
1316 u64 next_root_id;
1317 bool cycled = false;
1318 long nr_dropped = 0;
1319
1320 fs_info = container_of(work, struct btrfs_fs_info, em_shrinker_work);
1321
1322 ctx.scanned = 0;
1323 ctx.nr_to_scan = atomic64_read(&fs_info->em_shrinker_nr_to_scan);
1324
1325 start_root_id = fs_info->em_shrinker_last_root;
1326 next_root_id = fs_info->em_shrinker_last_root;
1327
1328 if (trace_btrfs_extent_map_shrinker_scan_enter_enabled()) {
1329 s64 nr = percpu_counter_sum_positive(&fs_info->evictable_extent_maps);
1330
1331 trace_call__btrfs_extent_map_shrinker_scan_enter(fs_info, nr);
1332 }
1333
1334 while (ctx.scanned < ctx.nr_to_scan && !btrfs_fs_closing(fs_info)) {
1335 struct btrfs_root *root;
1336 unsigned long count;
1337
1338 cond_resched();
1339
1340 spin_lock(&fs_info->fs_roots_radix_lock);
1341 count = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1342 (void **)&root,
1343 (unsigned long)next_root_id, 1);
1344 if (count == 0) {
1345 spin_unlock(&fs_info->fs_roots_radix_lock);
1346 if (start_root_id > 0 && !cycled) {
1347 next_root_id = 0;
1348 fs_info->em_shrinker_last_root = 0;
1349 fs_info->em_shrinker_last_ino = 0;
1350 cycled = true;
1351 continue;
1352 }
1353 break;
1354 }
1355 next_root_id = btrfs_root_id(root) + 1;
1356 root = btrfs_grab_root(root);
1357 spin_unlock(&fs_info->fs_roots_radix_lock);
1358
1359 if (!root)
1360 continue;
1361
1362 if (btrfs_is_fstree(btrfs_root_id(root)))
1363 nr_dropped += btrfs_scan_root(root, &ctx);
1364
1365 btrfs_put_root(root);
1366 }
1367
1368 if (trace_btrfs_extent_map_shrinker_scan_exit_enabled()) {
1369 s64 nr = percpu_counter_sum_positive(&fs_info->evictable_extent_maps);
1370
1371 trace_call__btrfs_extent_map_shrinker_scan_exit(fs_info, nr_dropped, nr);
1372 }
1373
1374 atomic64_set(&fs_info->em_shrinker_nr_to_scan, 0);
1375}
1376
1377void btrfs_free_extent_maps(struct btrfs_fs_info *fs_info, long nr_to_scan)
1378{
1379 /*
1380 * Do nothing if the shrinker is already running. In case of high memory
1381 * pressure we can have a lot of tasks calling us and all passing the
1382 * same nr_to_scan value, but in reality we may need only to free
1383 * nr_to_scan extent maps (or less). In case we need to free more than
1384 * that, we will be called again by the fs shrinker, so no worries about
1385 * not doing enough work to reclaim memory from extent maps.
1386 * We can also be repeatedly called with the same nr_to_scan value
1387 * simply because the shrinker runs asynchronously and multiple calls
1388 * to this function are made before the shrinker does enough progress.
1389 *
1390 * That's why we set the atomic counter to nr_to_scan only if its
1391 * current value is zero, instead of incrementing the counter by
1392 * nr_to_scan.
1393 */
1394 if (atomic64_cmpxchg(&fs_info->em_shrinker_nr_to_scan, 0, nr_to_scan) != 0)
1395 return;
1396
1397 queue_work(system_dfl_wq, &fs_info->em_shrinker_work);
1398}
1399
1400void btrfs_init_extent_map_shrinker_work(struct btrfs_fs_info *fs_info)
1401{
1402 atomic64_set(&fs_info->em_shrinker_nr_to_scan, 0);
1403 INIT_WORK(&fs_info->em_shrinker_work, btrfs_extent_map_shrinker_worker);
1404}