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 (prev
->map_lookup
|| next
->map_lookup
)
218 ASSERT(test_bit(EXTENT_FLAG_FS_MAPPING
, &prev
->flags
) &&
219 test_bit(EXTENT_FLAG_FS_MAPPING
, &next
->flags
));
221 if (extent_map_end(prev
) == next
->start
&&
222 prev
->flags
== next
->flags
&&
223 prev
->map_lookup
== next
->map_lookup
&&
224 ((next
->block_start
== EXTENT_MAP_HOLE
&&
225 prev
->block_start
== EXTENT_MAP_HOLE
) ||
226 (next
->block_start
== EXTENT_MAP_INLINE
&&
227 prev
->block_start
== EXTENT_MAP_INLINE
) ||
228 (next
->block_start
< EXTENT_MAP_LAST_BYTE
- 1 &&
229 next
->block_start
== extent_map_block_end(prev
)))) {
235 static void try_merge_map(struct extent_map_tree
*tree
, struct extent_map
*em
)
237 struct extent_map
*merge
= NULL
;
241 * We can't modify an extent map that is in the tree and that is being
242 * used by another task, as it can cause that other task to see it in
243 * inconsistent state during the merging. We always have 1 reference for
244 * the tree and 1 for this task (which is unpinning the extent map or
245 * clearing the logging flag), so anything > 2 means it's being used by
248 if (refcount_read(&em
->refs
) > 2)
251 if (em
->start
!= 0) {
252 rb
= rb_prev(&em
->rb_node
);
254 merge
= rb_entry(rb
, struct extent_map
, rb_node
);
255 if (rb
&& mergable_maps(merge
, em
)) {
256 em
->start
= merge
->start
;
257 em
->orig_start
= merge
->orig_start
;
258 em
->len
+= merge
->len
;
259 em
->block_len
+= merge
->block_len
;
260 em
->block_start
= merge
->block_start
;
261 em
->mod_len
= (em
->mod_len
+ em
->mod_start
) - merge
->mod_start
;
262 em
->mod_start
= merge
->mod_start
;
263 em
->generation
= max(em
->generation
, merge
->generation
);
265 rb_erase_cached(&merge
->rb_node
, &tree
->map
);
266 RB_CLEAR_NODE(&merge
->rb_node
);
267 free_extent_map(merge
);
271 rb
= rb_next(&em
->rb_node
);
273 merge
= rb_entry(rb
, struct extent_map
, rb_node
);
274 if (rb
&& mergable_maps(em
, merge
)) {
275 em
->len
+= merge
->len
;
276 em
->block_len
+= merge
->block_len
;
277 rb_erase_cached(&merge
->rb_node
, &tree
->map
);
278 RB_CLEAR_NODE(&merge
->rb_node
);
279 em
->mod_len
= (merge
->mod_start
+ merge
->mod_len
) - em
->mod_start
;
280 em
->generation
= max(em
->generation
, merge
->generation
);
281 free_extent_map(merge
);
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
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().
296 int unpin_extent_cache(struct extent_map_tree
*tree
, u64 start
, u64 len
,
300 struct extent_map
*em
;
301 bool prealloc
= false;
303 write_lock(&tree
->lock
);
304 em
= lookup_extent_mapping(tree
, start
, len
);
306 WARN_ON(!em
|| em
->start
!= start
);
311 em
->generation
= gen
;
312 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
313 em
->mod_start
= em
->start
;
314 em
->mod_len
= em
->len
;
316 if (test_bit(EXTENT_FLAG_FILLING
, &em
->flags
)) {
318 clear_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
321 try_merge_map(tree
, em
);
324 em
->mod_start
= em
->start
;
325 em
->mod_len
= em
->len
;
330 write_unlock(&tree
->lock
);
335 void clear_em_logging(struct extent_map_tree
*tree
, struct extent_map
*em
)
337 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
338 if (extent_map_in_tree(em
))
339 try_merge_map(tree
, em
);
342 static inline void setup_extent_mapping(struct extent_map_tree
*tree
,
343 struct extent_map
*em
,
346 refcount_inc(&em
->refs
);
347 em
->mod_start
= em
->start
;
348 em
->mod_len
= em
->len
;
351 list_move(&em
->list
, &tree
->modified_extents
);
353 try_merge_map(tree
, em
);
356 static void extent_map_device_set_bits(struct extent_map
*em
, unsigned bits
)
358 struct map_lookup
*map
= em
->map_lookup
;
359 u64 stripe_size
= em
->orig_block_len
;
362 for (i
= 0; i
< map
->num_stripes
; i
++) {
363 struct btrfs_bio_stripe
*stripe
= &map
->stripes
[i
];
364 struct btrfs_device
*device
= stripe
->dev
;
366 set_extent_bits_nowait(&device
->alloc_state
, stripe
->physical
,
367 stripe
->physical
+ stripe_size
- 1, bits
);
371 static void extent_map_device_clear_bits(struct extent_map
*em
, unsigned bits
)
373 struct map_lookup
*map
= em
->map_lookup
;
374 u64 stripe_size
= em
->orig_block_len
;
377 for (i
= 0; i
< map
->num_stripes
; i
++) {
378 struct btrfs_bio_stripe
*stripe
= &map
->stripes
[i
];
379 struct btrfs_device
*device
= stripe
->dev
;
381 __clear_extent_bit(&device
->alloc_state
, stripe
->physical
,
382 stripe
->physical
+ stripe_size
- 1, bits
,
383 0, 0, NULL
, GFP_NOWAIT
, NULL
);
388 * add_extent_mapping - add new extent map to the extent tree
389 * @tree: tree to insert new map in
392 * Insert @em into @tree or perform a simple forward/backward merge with
393 * existing mappings. The extent_map struct passed in will be inserted
394 * into the tree directly, with an additional reference taken, or a
395 * reference dropped if the merge attempt was successful.
397 int add_extent_mapping(struct extent_map_tree
*tree
,
398 struct extent_map
*em
, int modified
)
402 lockdep_assert_held_write(&tree
->lock
);
404 ret
= tree_insert(&tree
->map
, em
);
408 setup_extent_mapping(tree
, em
, modified
);
409 if (test_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
)) {
410 extent_map_device_set_bits(em
, CHUNK_ALLOCATED
);
411 extent_map_device_clear_bits(em
, CHUNK_TRIMMED
);
417 static struct extent_map
*
418 __lookup_extent_mapping(struct extent_map_tree
*tree
,
419 u64 start
, u64 len
, int strict
)
421 struct extent_map
*em
;
422 struct rb_node
*rb_node
;
423 struct rb_node
*prev
= NULL
;
424 struct rb_node
*next
= NULL
;
425 u64 end
= range_end(start
, len
);
427 rb_node
= __tree_search(&tree
->map
.rb_root
, start
, &prev
, &next
);
437 em
= rb_entry(rb_node
, struct extent_map
, rb_node
);
439 if (strict
&& !(end
> em
->start
&& start
< extent_map_end(em
)))
442 refcount_inc(&em
->refs
);
447 * lookup_extent_mapping - lookup extent_map
448 * @tree: tree to lookup in
449 * @start: byte offset to start the search
450 * @len: length of the lookup range
452 * Find and return the first extent_map struct in @tree that intersects the
453 * [start, len] range. There may be additional objects in the tree that
454 * intersect, so check the object returned carefully to make sure that no
455 * additional lookups are needed.
457 struct extent_map
*lookup_extent_mapping(struct extent_map_tree
*tree
,
460 return __lookup_extent_mapping(tree
, start
, len
, 1);
464 * search_extent_mapping - find a nearby extent map
465 * @tree: tree to lookup in
466 * @start: byte offset to start the search
467 * @len: length of the lookup range
469 * Find and return the first extent_map struct in @tree that intersects the
470 * [start, len] range.
472 * If one can't be found, any nearby extent may be returned
474 struct extent_map
*search_extent_mapping(struct extent_map_tree
*tree
,
477 return __lookup_extent_mapping(tree
, start
, len
, 0);
481 * remove_extent_mapping - removes an extent_map from the extent tree
482 * @tree: extent tree to remove from
483 * @em: extent map being removed
485 * Removes @em from @tree. No reference counts are dropped, and no checks
486 * are done to see if the range is in use
488 void remove_extent_mapping(struct extent_map_tree
*tree
, struct extent_map
*em
)
490 WARN_ON(test_bit(EXTENT_FLAG_PINNED
, &em
->flags
));
491 rb_erase_cached(&em
->rb_node
, &tree
->map
);
492 if (!test_bit(EXTENT_FLAG_LOGGING
, &em
->flags
))
493 list_del_init(&em
->list
);
494 if (test_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
))
495 extent_map_device_clear_bits(em
, CHUNK_ALLOCATED
);
496 RB_CLEAR_NODE(&em
->rb_node
);
499 void replace_extent_mapping(struct extent_map_tree
*tree
,
500 struct extent_map
*cur
,
501 struct extent_map
*new,
504 WARN_ON(test_bit(EXTENT_FLAG_PINNED
, &cur
->flags
));
505 ASSERT(extent_map_in_tree(cur
));
506 if (!test_bit(EXTENT_FLAG_LOGGING
, &cur
->flags
))
507 list_del_init(&cur
->list
);
508 rb_replace_node_cached(&cur
->rb_node
, &new->rb_node
, &tree
->map
);
509 RB_CLEAR_NODE(&cur
->rb_node
);
511 setup_extent_mapping(tree
, new, modified
);
514 static struct extent_map
*next_extent_map(struct extent_map
*em
)
516 struct rb_node
*next
;
518 next
= rb_next(&em
->rb_node
);
521 return container_of(next
, struct extent_map
, rb_node
);
524 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
526 struct rb_node
*prev
;
528 prev
= rb_prev(&em
->rb_node
);
531 return container_of(prev
, struct extent_map
, rb_node
);
535 * Helper for btrfs_get_extent. Given an existing extent in the tree,
536 * the existing extent is the nearest extent to map_start,
537 * and an extent that you want to insert, deal with overlap and insert
538 * the best fitted new extent into the tree.
540 static noinline
int merge_extent_mapping(struct extent_map_tree
*em_tree
,
541 struct extent_map
*existing
,
542 struct extent_map
*em
,
545 struct extent_map
*prev
;
546 struct extent_map
*next
;
551 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
553 if (existing
->start
> map_start
) {
555 prev
= prev_extent_map(next
);
558 next
= next_extent_map(prev
);
561 start
= prev
? extent_map_end(prev
) : em
->start
;
562 start
= max_t(u64
, start
, em
->start
);
563 end
= next
? next
->start
: extent_map_end(em
);
564 end
= min_t(u64
, end
, extent_map_end(em
));
565 start_diff
= start
- em
->start
;
567 em
->len
= end
- start
;
568 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
569 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
570 em
->block_start
+= start_diff
;
571 em
->block_len
= em
->len
;
573 return add_extent_mapping(em_tree
, em
, 0);
577 * btrfs_add_extent_mapping - add extent mapping into em_tree
578 * @fs_info - used for tracepoint
579 * @em_tree - the extent tree into which we want to insert the extent mapping
580 * @em_in - extent we are inserting
581 * @start - start of the logical range btrfs_get_extent() is requesting
582 * @len - length of the logical range btrfs_get_extent() is requesting
584 * Note that @em_in's range may be different from [start, start+len),
585 * but they must be overlapped.
587 * Insert @em_in into @em_tree. In case there is an overlapping range, handle
588 * the -EEXIST by either:
589 * a) Returning the existing extent in @em_in if @start is within the
591 * b) Merge the existing extent with @em_in passed in.
593 * Return 0 on success, otherwise -EEXIST.
596 int btrfs_add_extent_mapping(struct btrfs_fs_info
*fs_info
,
597 struct extent_map_tree
*em_tree
,
598 struct extent_map
**em_in
, u64 start
, u64 len
)
601 struct extent_map
*em
= *em_in
;
603 ret
= add_extent_mapping(em_tree
, em
, 0);
604 /* it is possible that someone inserted the extent into the tree
605 * while we had the lock dropped. It is also possible that
606 * an overlapping map exists in the tree
608 if (ret
== -EEXIST
) {
609 struct extent_map
*existing
;
613 existing
= search_extent_mapping(em_tree
, start
, len
);
615 trace_btrfs_handle_em_exist(fs_info
, existing
, em
, start
, len
);
618 * existing will always be non-NULL, since there must be
619 * extent causing the -EEXIST.
621 if (start
>= existing
->start
&&
622 start
< extent_map_end(existing
)) {
627 u64 orig_start
= em
->start
;
628 u64 orig_len
= em
->len
;
631 * The existing extent map is the one nearest to
632 * the [start, start + len) range which overlaps
634 ret
= merge_extent_mapping(em_tree
, existing
,
640 "unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n",
641 ret
, existing
->start
, existing
->len
,
642 orig_start
, orig_len
);
644 free_extent_map(existing
);
648 ASSERT(ret
== 0 || ret
== -EEXIST
);