Loading...
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
11
12static struct kmem_cache *extent_map_cache;
13
14int __init extent_map_init(void)
15{
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)
20 return -ENOMEM;
21 return 0;
22}
23
24void __cold extent_map_exit(void)
25{
26 kmem_cache_destroy(extent_map_cache);
27}
28
29/**
30 * extent_map_tree_init - initialize extent map tree
31 * @tree: tree to initialize
32 *
33 * Initialize the extent tree @tree. Should be called for each new inode
34 * or other user of the extent_map interface.
35 */
36void extent_map_tree_init(struct extent_map_tree *tree)
37{
38 tree->map = RB_ROOT_CACHED;
39 INIT_LIST_HEAD(&tree->modified_extents);
40 rwlock_init(&tree->lock);
41}
42
43/**
44 * alloc_extent_map - allocate new extent map structure
45 *
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()
49 */
50struct extent_map *alloc_extent_map(void)
51{
52 struct extent_map *em;
53 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
54 if (!em)
55 return NULL;
56 RB_CLEAR_NODE(&em->rb_node);
57 em->flags = 0;
58 em->compress_type = BTRFS_COMPRESS_NONE;
59 em->generation = 0;
60 refcount_set(&em->refs, 1);
61 INIT_LIST_HEAD(&em->list);
62 return em;
63}
64
65/**
66 * free_extent_map - drop reference count of an extent_map
67 * @em: extent map being released
68 *
69 * Drops the reference out on @em by one and free the structure
70 * if the reference count hits zero.
71 */
72void free_extent_map(struct extent_map *em)
73{
74 if (!em)
75 return;
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);
83 }
84}
85
86/* simple helper to do math around the end of an extent, handling wrap */
87static u64 range_end(u64 start, u64 len)
88{
89 if (start + len < start)
90 return (u64)-1;
91 return start + len;
92}
93
94static int tree_insert(struct rb_root_cached *root, struct extent_map *em)
95{
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;
102
103 while (*p) {
104 parent = *p;
105 entry = rb_entry(parent, struct extent_map, rb_node);
106
107 if (em->start < entry->start) {
108 p = &(*p)->rb_left;
109 } else if (em->start >= extent_map_end(entry)) {
110 p = &(*p)->rb_right;
111 leftmost = false;
112 } else {
113 return -EEXIST;
114 }
115 }
116
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);
121 }
122 if (parent)
123 if (end > entry->start && em->start < extent_map_end(entry))
124 return -EEXIST;
125
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);
131 }
132 if (parent)
133 if (end > entry->start && em->start < extent_map_end(entry))
134 return -EEXIST;
135
136 rb_link_node(&em->rb_node, orig_parent, p);
137 rb_insert_color_cached(&em->rb_node, root, leftmost);
138 return 0;
139}
140
141/*
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
144 */
145static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
146 struct rb_node **prev_ret,
147 struct rb_node **next_ret)
148{
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;
154
155 while (n) {
156 entry = rb_entry(n, struct extent_map, rb_node);
157 prev = n;
158 prev_entry = entry;
159
160 if (offset < entry->start)
161 n = n->rb_left;
162 else if (offset >= extent_map_end(entry))
163 n = n->rb_right;
164 else
165 return n;
166 }
167
168 if (prev_ret) {
169 orig_prev = prev;
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);
173 }
174 *prev_ret = prev;
175 prev = orig_prev;
176 }
177
178 if (next_ret) {
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 *next_ret = prev;
185 }
186 return NULL;
187}
188
189/* check to see if two extent_map structs are adjacent and safe to merge */
190static int mergable_maps(struct extent_map *prev, struct extent_map *next)
191{
192 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
193 return 0;
194
195 /*
196 * don't merge compressed extents, we need to know their
197 * actual size
198 */
199 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
200 return 0;
201
202 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
203 test_bit(EXTENT_FLAG_LOGGING, &next->flags))
204 return 0;
205
206 /*
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
209 * bad.
210 */
211 if (!list_empty(&prev->list) || !list_empty(&next->list))
212 return 0;
213
214 ASSERT(next->block_start != EXTENT_MAP_DELALLOC &&
215 prev->block_start != EXTENT_MAP_DELALLOC);
216
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)))) {
226 return 1;
227 }
228 return 0;
229}
230
231static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
232{
233 struct extent_map *merge = NULL;
234 struct rb_node *rb;
235
236 if (em->start != 0) {
237 rb = rb_prev(&em->rb_node);
238 if (rb)
239 merge = rb_entry(rb, struct extent_map, rb_node);
240 if (rb && mergable_maps(merge, em)) {
241 em->start = merge->start;
242 em->orig_start = merge->orig_start;
243 em->len += merge->len;
244 em->block_len += merge->block_len;
245 em->block_start = merge->block_start;
246 em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
247 em->mod_start = merge->mod_start;
248 em->generation = max(em->generation, merge->generation);
249
250 rb_erase_cached(&merge->rb_node, &tree->map);
251 RB_CLEAR_NODE(&merge->rb_node);
252 free_extent_map(merge);
253 }
254 }
255
256 rb = rb_next(&em->rb_node);
257 if (rb)
258 merge = rb_entry(rb, struct extent_map, rb_node);
259 if (rb && mergable_maps(em, merge)) {
260 em->len += merge->len;
261 em->block_len += merge->block_len;
262 rb_erase_cached(&merge->rb_node, &tree->map);
263 RB_CLEAR_NODE(&merge->rb_node);
264 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
265 em->generation = max(em->generation, merge->generation);
266 free_extent_map(merge);
267 }
268}
269
270/**
271 * unpin_extent_cache - unpin an extent from the cache
272 * @tree: tree to unpin the extent in
273 * @start: logical offset in the file
274 * @len: length of the extent
275 * @gen: generation that this extent has been modified in
276 *
277 * Called after an extent has been written to disk properly. Set the generation
278 * to the generation that actually added the file item to the inode so we know
279 * we need to sync this extent when we call fsync().
280 */
281int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
282 u64 gen)
283{
284 int ret = 0;
285 struct extent_map *em;
286 bool prealloc = false;
287
288 write_lock(&tree->lock);
289 em = lookup_extent_mapping(tree, start, len);
290
291 WARN_ON(!em || em->start != start);
292
293 if (!em)
294 goto out;
295
296 em->generation = gen;
297 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
298 em->mod_start = em->start;
299 em->mod_len = em->len;
300
301 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
302 prealloc = true;
303 clear_bit(EXTENT_FLAG_FILLING, &em->flags);
304 }
305
306 try_merge_map(tree, em);
307
308 if (prealloc) {
309 em->mod_start = em->start;
310 em->mod_len = em->len;
311 }
312
313 free_extent_map(em);
314out:
315 write_unlock(&tree->lock);
316 return ret;
317
318}
319
320void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
321{
322 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
323 if (extent_map_in_tree(em))
324 try_merge_map(tree, em);
325}
326
327static inline void setup_extent_mapping(struct extent_map_tree *tree,
328 struct extent_map *em,
329 int modified)
330{
331 refcount_inc(&em->refs);
332 em->mod_start = em->start;
333 em->mod_len = em->len;
334
335 if (modified)
336 list_move(&em->list, &tree->modified_extents);
337 else
338 try_merge_map(tree, em);
339}
340
341static void extent_map_device_set_bits(struct extent_map *em, unsigned bits)
342{
343 struct map_lookup *map = em->map_lookup;
344 u64 stripe_size = em->orig_block_len;
345 int i;
346
347 for (i = 0; i < map->num_stripes; i++) {
348 struct btrfs_bio_stripe *stripe = &map->stripes[i];
349 struct btrfs_device *device = stripe->dev;
350
351 set_extent_bits_nowait(&device->alloc_state, stripe->physical,
352 stripe->physical + stripe_size - 1, bits);
353 }
354}
355
356static void extent_map_device_clear_bits(struct extent_map *em, unsigned bits)
357{
358 struct map_lookup *map = em->map_lookup;
359 u64 stripe_size = em->orig_block_len;
360 int i;
361
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;
365
366 __clear_extent_bit(&device->alloc_state, stripe->physical,
367 stripe->physical + stripe_size - 1, bits,
368 0, 0, NULL, GFP_NOWAIT, NULL);
369 }
370}
371
372/**
373 * add_extent_mapping - add new extent map to the extent tree
374 * @tree: tree to insert new map in
375 * @em: map to insert
376 *
377 * Insert @em into @tree or perform a simple forward/backward merge with
378 * existing mappings. The extent_map struct passed in will be inserted
379 * into the tree directly, with an additional reference taken, or a
380 * reference dropped if the merge attempt was successful.
381 */
382int add_extent_mapping(struct extent_map_tree *tree,
383 struct extent_map *em, int modified)
384{
385 int ret = 0;
386
387 lockdep_assert_held_write(&tree->lock);
388
389 ret = tree_insert(&tree->map, em);
390 if (ret)
391 goto out;
392
393 setup_extent_mapping(tree, em, modified);
394 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags)) {
395 extent_map_device_set_bits(em, CHUNK_ALLOCATED);
396 extent_map_device_clear_bits(em, CHUNK_TRIMMED);
397 }
398out:
399 return ret;
400}
401
402static struct extent_map *
403__lookup_extent_mapping(struct extent_map_tree *tree,
404 u64 start, u64 len, int strict)
405{
406 struct extent_map *em;
407 struct rb_node *rb_node;
408 struct rb_node *prev = NULL;
409 struct rb_node *next = NULL;
410 u64 end = range_end(start, len);
411
412 rb_node = __tree_search(&tree->map.rb_root, start, &prev, &next);
413 if (!rb_node) {
414 if (prev)
415 rb_node = prev;
416 else if (next)
417 rb_node = next;
418 else
419 return NULL;
420 }
421
422 em = rb_entry(rb_node, struct extent_map, rb_node);
423
424 if (strict && !(end > em->start && start < extent_map_end(em)))
425 return NULL;
426
427 refcount_inc(&em->refs);
428 return em;
429}
430
431/**
432 * lookup_extent_mapping - lookup extent_map
433 * @tree: tree to lookup in
434 * @start: byte offset to start the search
435 * @len: length of the lookup range
436 *
437 * Find and return the first extent_map struct in @tree that intersects the
438 * [start, len] range. There may be additional objects in the tree that
439 * intersect, so check the object returned carefully to make sure that no
440 * additional lookups are needed.
441 */
442struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
443 u64 start, u64 len)
444{
445 return __lookup_extent_mapping(tree, start, len, 1);
446}
447
448/**
449 * search_extent_mapping - find a nearby extent map
450 * @tree: tree to lookup in
451 * @start: byte offset to start the search
452 * @len: length of the lookup range
453 *
454 * Find and return the first extent_map struct in @tree that intersects the
455 * [start, len] range.
456 *
457 * If one can't be found, any nearby extent may be returned
458 */
459struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
460 u64 start, u64 len)
461{
462 return __lookup_extent_mapping(tree, start, len, 0);
463}
464
465/**
466 * remove_extent_mapping - removes an extent_map from the extent tree
467 * @tree: extent tree to remove from
468 * @em: extent map being removed
469 *
470 * Removes @em from @tree. No reference counts are dropped, and no checks
471 * are done to see if the range is in use
472 */
473void remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
474{
475 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
476 rb_erase_cached(&em->rb_node, &tree->map);
477 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
478 list_del_init(&em->list);
479 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
480 extent_map_device_clear_bits(em, CHUNK_ALLOCATED);
481 RB_CLEAR_NODE(&em->rb_node);
482}
483
484void replace_extent_mapping(struct extent_map_tree *tree,
485 struct extent_map *cur,
486 struct extent_map *new,
487 int modified)
488{
489 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
490 ASSERT(extent_map_in_tree(cur));
491 if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
492 list_del_init(&cur->list);
493 rb_replace_node_cached(&cur->rb_node, &new->rb_node, &tree->map);
494 RB_CLEAR_NODE(&cur->rb_node);
495
496 setup_extent_mapping(tree, new, modified);
497}
498
499static struct extent_map *next_extent_map(struct extent_map *em)
500{
501 struct rb_node *next;
502
503 next = rb_next(&em->rb_node);
504 if (!next)
505 return NULL;
506 return container_of(next, struct extent_map, rb_node);
507}
508
509static struct extent_map *prev_extent_map(struct extent_map *em)
510{
511 struct rb_node *prev;
512
513 prev = rb_prev(&em->rb_node);
514 if (!prev)
515 return NULL;
516 return container_of(prev, struct extent_map, rb_node);
517}
518
519/*
520 * Helper for btrfs_get_extent. Given an existing extent in the tree,
521 * the existing extent is the nearest extent to map_start,
522 * and an extent that you want to insert, deal with overlap and insert
523 * the best fitted new extent into the tree.
524 */
525static noinline int merge_extent_mapping(struct extent_map_tree *em_tree,
526 struct extent_map *existing,
527 struct extent_map *em,
528 u64 map_start)
529{
530 struct extent_map *prev;
531 struct extent_map *next;
532 u64 start;
533 u64 end;
534 u64 start_diff;
535
536 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
537
538 if (existing->start > map_start) {
539 next = existing;
540 prev = prev_extent_map(next);
541 } else {
542 prev = existing;
543 next = next_extent_map(prev);
544 }
545
546 start = prev ? extent_map_end(prev) : em->start;
547 start = max_t(u64, start, em->start);
548 end = next ? next->start : extent_map_end(em);
549 end = min_t(u64, end, extent_map_end(em));
550 start_diff = start - em->start;
551 em->start = start;
552 em->len = end - start;
553 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
554 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
555 em->block_start += start_diff;
556 em->block_len = em->len;
557 }
558 return add_extent_mapping(em_tree, em, 0);
559}
560
561/**
562 * btrfs_add_extent_mapping - add extent mapping into em_tree
563 * @fs_info - used for tracepoint
564 * @em_tree - the extent tree into which we want to insert the extent mapping
565 * @em_in - extent we are inserting
566 * @start - start of the logical range btrfs_get_extent() is requesting
567 * @len - length of the logical range btrfs_get_extent() is requesting
568 *
569 * Note that @em_in's range may be different from [start, start+len),
570 * but they must be overlapped.
571 *
572 * Insert @em_in into @em_tree. In case there is an overlapping range, handle
573 * the -EEXIST by either:
574 * a) Returning the existing extent in @em_in if @start is within the
575 * existing em.
576 * b) Merge the existing extent with @em_in passed in.
577 *
578 * Return 0 on success, otherwise -EEXIST.
579 *
580 */
581int btrfs_add_extent_mapping(struct btrfs_fs_info *fs_info,
582 struct extent_map_tree *em_tree,
583 struct extent_map **em_in, u64 start, u64 len)
584{
585 int ret;
586 struct extent_map *em = *em_in;
587
588 ret = add_extent_mapping(em_tree, em, 0);
589 /* it is possible that someone inserted the extent into the tree
590 * while we had the lock dropped. It is also possible that
591 * an overlapping map exists in the tree
592 */
593 if (ret == -EEXIST) {
594 struct extent_map *existing;
595
596 ret = 0;
597
598 existing = search_extent_mapping(em_tree, start, len);
599
600 trace_btrfs_handle_em_exist(fs_info, existing, em, start, len);
601
602 /*
603 * existing will always be non-NULL, since there must be
604 * extent causing the -EEXIST.
605 */
606 if (start >= existing->start &&
607 start < extent_map_end(existing)) {
608 free_extent_map(em);
609 *em_in = existing;
610 ret = 0;
611 } else {
612 u64 orig_start = em->start;
613 u64 orig_len = em->len;
614
615 /*
616 * The existing extent map is the one nearest to
617 * the [start, start + len) range which overlaps
618 */
619 ret = merge_extent_mapping(em_tree, existing,
620 em, start);
621 if (ret) {
622 free_extent_map(em);
623 *em_in = NULL;
624 WARN_ONCE(ret,
625"unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n",
626 ret, existing->start, existing->len,
627 orig_start, orig_len);
628 }
629 free_extent_map(existing);
630 }
631 }
632
633 ASSERT(ret == 0 || ret == -EEXIST);
634 return ret;
635}
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
11
12static struct kmem_cache *extent_map_cache;
13
14int __init extent_map_init(void)
15{
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)
20 return -ENOMEM;
21 return 0;
22}
23
24void __cold extent_map_exit(void)
25{
26 kmem_cache_destroy(extent_map_cache);
27}
28
29/**
30 * extent_map_tree_init - initialize extent map tree
31 * @tree: tree to initialize
32 *
33 * Initialize the extent tree @tree. Should be called for each new inode
34 * or other user of the extent_map interface.
35 */
36void extent_map_tree_init(struct extent_map_tree *tree)
37{
38 tree->map = RB_ROOT_CACHED;
39 INIT_LIST_HEAD(&tree->modified_extents);
40 rwlock_init(&tree->lock);
41}
42
43/**
44 * alloc_extent_map - allocate new extent map structure
45 *
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()
49 */
50struct extent_map *alloc_extent_map(void)
51{
52 struct extent_map *em;
53 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
54 if (!em)
55 return NULL;
56 RB_CLEAR_NODE(&em->rb_node);
57 em->flags = 0;
58 em->compress_type = BTRFS_COMPRESS_NONE;
59 em->generation = 0;
60 refcount_set(&em->refs, 1);
61 INIT_LIST_HEAD(&em->list);
62 return em;
63}
64
65/**
66 * free_extent_map - drop reference count of an extent_map
67 * @em: extent map being released
68 *
69 * Drops the reference out on @em by one and free the structure
70 * if the reference count hits zero.
71 */
72void free_extent_map(struct extent_map *em)
73{
74 if (!em)
75 return;
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);
83 }
84}
85
86/* simple helper to do math around the end of an extent, handling wrap */
87static u64 range_end(u64 start, u64 len)
88{
89 if (start + len < start)
90 return (u64)-1;
91 return start + len;
92}
93
94static int tree_insert(struct rb_root_cached *root, struct extent_map *em)
95{
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;
102
103 while (*p) {
104 parent = *p;
105 entry = rb_entry(parent, struct extent_map, rb_node);
106
107 if (em->start < entry->start) {
108 p = &(*p)->rb_left;
109 } else if (em->start >= extent_map_end(entry)) {
110 p = &(*p)->rb_right;
111 leftmost = false;
112 } else {
113 return -EEXIST;
114 }
115 }
116
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);
121 }
122 if (parent)
123 if (end > entry->start && em->start < extent_map_end(entry))
124 return -EEXIST;
125
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);
131 }
132 if (parent)
133 if (end > entry->start && em->start < extent_map_end(entry))
134 return -EEXIST;
135
136 rb_link_node(&em->rb_node, orig_parent, p);
137 rb_insert_color_cached(&em->rb_node, root, leftmost);
138 return 0;
139}
140
141/*
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
144 */
145static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
146 struct rb_node **prev_ret,
147 struct rb_node **next_ret)
148{
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;
154
155 while (n) {
156 entry = rb_entry(n, struct extent_map, rb_node);
157 prev = n;
158 prev_entry = entry;
159
160 if (offset < entry->start)
161 n = n->rb_left;
162 else if (offset >= extent_map_end(entry))
163 n = n->rb_right;
164 else
165 return n;
166 }
167
168 if (prev_ret) {
169 orig_prev = prev;
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);
173 }
174 *prev_ret = prev;
175 prev = orig_prev;
176 }
177
178 if (next_ret) {
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 *next_ret = prev;
185 }
186 return NULL;
187}
188
189/* check to see if two extent_map structs are adjacent and safe to merge */
190static int mergable_maps(struct extent_map *prev, struct extent_map *next)
191{
192 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
193 return 0;
194
195 /*
196 * don't merge compressed extents, we need to know their
197 * actual size
198 */
199 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
200 return 0;
201
202 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
203 test_bit(EXTENT_FLAG_LOGGING, &next->flags))
204 return 0;
205
206 /*
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
209 * bad.
210 */
211 if (!list_empty(&prev->list) || !list_empty(&next->list))
212 return 0;
213
214 ASSERT(next->block_start != EXTENT_MAP_DELALLOC &&
215 prev->block_start != EXTENT_MAP_DELALLOC);
216
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));
220
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)))) {
230 return 1;
231 }
232 return 0;
233}
234
235static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
236{
237 struct extent_map *merge = NULL;
238 struct rb_node *rb;
239
240 /*
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
246 * other tasks too.
247 */
248 if (refcount_read(&em->refs) > 2)
249 return;
250
251 if (em->start != 0) {
252 rb = rb_prev(&em->rb_node);
253 if (rb)
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);
264
265 rb_erase_cached(&merge->rb_node, &tree->map);
266 RB_CLEAR_NODE(&merge->rb_node);
267 free_extent_map(merge);
268 }
269 }
270
271 rb = rb_next(&em->rb_node);
272 if (rb)
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);
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 */
296int 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);
329out:
330 write_unlock(&tree->lock);
331 return ret;
332
333}
334
335void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
336{
337 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
338 if (extent_map_in_tree(em))
339 try_merge_map(tree, em);
340}
341
342static inline void setup_extent_mapping(struct extent_map_tree *tree,
343 struct extent_map *em,
344 int modified)
345{
346 refcount_inc(&em->refs);
347 em->mod_start = em->start;
348 em->mod_len = em->len;
349
350 if (modified)
351 list_move(&em->list, &tree->modified_extents);
352 else
353 try_merge_map(tree, em);
354}
355
356static void extent_map_device_set_bits(struct extent_map *em, unsigned bits)
357{
358 struct map_lookup *map = em->map_lookup;
359 u64 stripe_size = em->orig_block_len;
360 int i;
361
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;
365
366 set_extent_bits_nowait(&device->alloc_state, stripe->physical,
367 stripe->physical + stripe_size - 1, bits);
368 }
369}
370
371static void extent_map_device_clear_bits(struct extent_map *em, unsigned bits)
372{
373 struct map_lookup *map = em->map_lookup;
374 u64 stripe_size = em->orig_block_len;
375 int i;
376
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;
380
381 __clear_extent_bit(&device->alloc_state, stripe->physical,
382 stripe->physical + stripe_size - 1, bits,
383 0, 0, NULL, GFP_NOWAIT, NULL);
384 }
385}
386
387/**
388 * Add new extent map to the extent tree
389 *
390 * @tree: tree to insert new map in
391 * @em: map to insert
392 * @modified: indicate whether the given @em should be added to the
393 * modified list, which indicates the extent needs to be logged
394 *
395 * Insert @em into @tree or perform a simple forward/backward merge with
396 * existing mappings. The extent_map struct passed in will be inserted
397 * into the tree directly, with an additional reference taken, or a
398 * reference dropped if the merge attempt was successful.
399 */
400int add_extent_mapping(struct extent_map_tree *tree,
401 struct extent_map *em, int modified)
402{
403 int ret = 0;
404
405 lockdep_assert_held_write(&tree->lock);
406
407 ret = tree_insert(&tree->map, em);
408 if (ret)
409 goto out;
410
411 setup_extent_mapping(tree, em, modified);
412 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags)) {
413 extent_map_device_set_bits(em, CHUNK_ALLOCATED);
414 extent_map_device_clear_bits(em, CHUNK_TRIMMED);
415 }
416out:
417 return ret;
418}
419
420static struct extent_map *
421__lookup_extent_mapping(struct extent_map_tree *tree,
422 u64 start, u64 len, int strict)
423{
424 struct extent_map *em;
425 struct rb_node *rb_node;
426 struct rb_node *prev = NULL;
427 struct rb_node *next = NULL;
428 u64 end = range_end(start, len);
429
430 rb_node = __tree_search(&tree->map.rb_root, start, &prev, &next);
431 if (!rb_node) {
432 if (prev)
433 rb_node = prev;
434 else if (next)
435 rb_node = next;
436 else
437 return NULL;
438 }
439
440 em = rb_entry(rb_node, struct extent_map, rb_node);
441
442 if (strict && !(end > em->start && start < extent_map_end(em)))
443 return NULL;
444
445 refcount_inc(&em->refs);
446 return em;
447}
448
449/**
450 * lookup_extent_mapping - lookup extent_map
451 * @tree: tree to lookup in
452 * @start: byte offset to start the search
453 * @len: length of the lookup range
454 *
455 * Find and return the first extent_map struct in @tree that intersects the
456 * [start, len] range. There may be additional objects in the tree that
457 * intersect, so check the object returned carefully to make sure that no
458 * additional lookups are needed.
459 */
460struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
461 u64 start, u64 len)
462{
463 return __lookup_extent_mapping(tree, start, len, 1);
464}
465
466/**
467 * search_extent_mapping - find a nearby extent map
468 * @tree: tree to lookup in
469 * @start: byte offset to start the search
470 * @len: length of the lookup range
471 *
472 * Find and return the first extent_map struct in @tree that intersects the
473 * [start, len] range.
474 *
475 * If one can't be found, any nearby extent may be returned
476 */
477struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
478 u64 start, u64 len)
479{
480 return __lookup_extent_mapping(tree, start, len, 0);
481}
482
483/**
484 * remove_extent_mapping - removes an extent_map from the extent tree
485 * @tree: extent tree to remove from
486 * @em: extent map being removed
487 *
488 * Removes @em from @tree. No reference counts are dropped, and no checks
489 * are done to see if the range is in use
490 */
491void remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
492{
493 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
494 rb_erase_cached(&em->rb_node, &tree->map);
495 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
496 list_del_init(&em->list);
497 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
498 extent_map_device_clear_bits(em, CHUNK_ALLOCATED);
499 RB_CLEAR_NODE(&em->rb_node);
500}
501
502void replace_extent_mapping(struct extent_map_tree *tree,
503 struct extent_map *cur,
504 struct extent_map *new,
505 int modified)
506{
507 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
508 ASSERT(extent_map_in_tree(cur));
509 if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
510 list_del_init(&cur->list);
511 rb_replace_node_cached(&cur->rb_node, &new->rb_node, &tree->map);
512 RB_CLEAR_NODE(&cur->rb_node);
513
514 setup_extent_mapping(tree, new, modified);
515}
516
517static struct extent_map *next_extent_map(struct extent_map *em)
518{
519 struct rb_node *next;
520
521 next = rb_next(&em->rb_node);
522 if (!next)
523 return NULL;
524 return container_of(next, struct extent_map, rb_node);
525}
526
527static struct extent_map *prev_extent_map(struct extent_map *em)
528{
529 struct rb_node *prev;
530
531 prev = rb_prev(&em->rb_node);
532 if (!prev)
533 return NULL;
534 return container_of(prev, struct extent_map, rb_node);
535}
536
537/*
538 * Helper for btrfs_get_extent. Given an existing extent in the tree,
539 * the existing extent is the nearest extent to map_start,
540 * and an extent that you want to insert, deal with overlap and insert
541 * the best fitted new extent into the tree.
542 */
543static noinline int merge_extent_mapping(struct extent_map_tree *em_tree,
544 struct extent_map *existing,
545 struct extent_map *em,
546 u64 map_start)
547{
548 struct extent_map *prev;
549 struct extent_map *next;
550 u64 start;
551 u64 end;
552 u64 start_diff;
553
554 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
555
556 if (existing->start > map_start) {
557 next = existing;
558 prev = prev_extent_map(next);
559 } else {
560 prev = existing;
561 next = next_extent_map(prev);
562 }
563
564 start = prev ? extent_map_end(prev) : em->start;
565 start = max_t(u64, start, em->start);
566 end = next ? next->start : extent_map_end(em);
567 end = min_t(u64, end, extent_map_end(em));
568 start_diff = start - em->start;
569 em->start = start;
570 em->len = end - start;
571 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
572 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
573 em->block_start += start_diff;
574 em->block_len = em->len;
575 }
576 return add_extent_mapping(em_tree, em, 0);
577}
578
579/**
580 * Add extent mapping into em_tree
581 *
582 * @fs_info: the filesystem
583 * @em_tree: extent tree into which we want to insert the extent mapping
584 * @em_in: extent we are inserting
585 * @start: start of the logical range btrfs_get_extent() is requesting
586 * @len: length of the logical range btrfs_get_extent() is requesting
587 *
588 * Note that @em_in's range may be different from [start, start+len),
589 * but they must be overlapped.
590 *
591 * Insert @em_in into @em_tree. In case there is an overlapping range, handle
592 * the -EEXIST by either:
593 * a) Returning the existing extent in @em_in if @start is within the
594 * existing em.
595 * b) Merge the existing extent with @em_in passed in.
596 *
597 * Return 0 on success, otherwise -EEXIST.
598 *
599 */
600int btrfs_add_extent_mapping(struct btrfs_fs_info *fs_info,
601 struct extent_map_tree *em_tree,
602 struct extent_map **em_in, u64 start, u64 len)
603{
604 int ret;
605 struct extent_map *em = *em_in;
606
607 ret = add_extent_mapping(em_tree, em, 0);
608 /* it is possible that someone inserted the extent into the tree
609 * while we had the lock dropped. It is also possible that
610 * an overlapping map exists in the tree
611 */
612 if (ret == -EEXIST) {
613 struct extent_map *existing;
614
615 ret = 0;
616
617 existing = search_extent_mapping(em_tree, start, len);
618
619 trace_btrfs_handle_em_exist(fs_info, existing, em, start, len);
620
621 /*
622 * existing will always be non-NULL, since there must be
623 * extent causing the -EEXIST.
624 */
625 if (start >= existing->start &&
626 start < extent_map_end(existing)) {
627 free_extent_map(em);
628 *em_in = existing;
629 ret = 0;
630 } else {
631 u64 orig_start = em->start;
632 u64 orig_len = em->len;
633
634 /*
635 * The existing extent map is the one nearest to
636 * the [start, start + len) range which overlaps
637 */
638 ret = merge_extent_mapping(em_tree, existing,
639 em, start);
640 if (ret) {
641 free_extent_map(em);
642 *em_in = NULL;
643 WARN_ONCE(ret,
644"unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n",
645 ret, existing->start, existing->len,
646 orig_start, orig_len);
647 }
648 free_extent_map(existing);
649 }
650 }
651
652 ASSERT(ret == 0 || ret == -EEXIST);
653 return ret;
654}