Loading...
1#include <linux/err.h>
2#include <linux/slab.h>
3#include <linux/spinlock.h>
4#include <linux/hardirq.h>
5#include "ctree.h"
6#include "extent_map.h"
7#include "compression.h"
8
9
10static struct kmem_cache *extent_map_cache;
11
12int __init extent_map_init(void)
13{
14 extent_map_cache = kmem_cache_create("btrfs_extent_map",
15 sizeof(struct extent_map), 0,
16 SLAB_MEM_SPREAD, NULL);
17 if (!extent_map_cache)
18 return -ENOMEM;
19 return 0;
20}
21
22void extent_map_exit(void)
23{
24 kmem_cache_destroy(extent_map_cache);
25}
26
27/**
28 * extent_map_tree_init - initialize extent map tree
29 * @tree: tree to initialize
30 *
31 * Initialize the extent tree @tree. Should be called for each new inode
32 * or other user of the extent_map interface.
33 */
34void extent_map_tree_init(struct extent_map_tree *tree)
35{
36 tree->map = RB_ROOT;
37 INIT_LIST_HEAD(&tree->modified_extents);
38 rwlock_init(&tree->lock);
39}
40
41/**
42 * alloc_extent_map - allocate new extent map structure
43 *
44 * Allocate a new extent_map structure. The new structure is
45 * returned with a reference count of one and needs to be
46 * freed using free_extent_map()
47 */
48struct extent_map *alloc_extent_map(void)
49{
50 struct extent_map *em;
51 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
52 if (!em)
53 return NULL;
54 RB_CLEAR_NODE(&em->rb_node);
55 em->flags = 0;
56 em->compress_type = BTRFS_COMPRESS_NONE;
57 em->generation = 0;
58 atomic_set(&em->refs, 1);
59 INIT_LIST_HEAD(&em->list);
60 return em;
61}
62
63/**
64 * free_extent_map - drop reference count of an extent_map
65 * @em: extent map being released
66 *
67 * Drops the reference out on @em by one and free the structure
68 * if the reference count hits zero.
69 */
70void free_extent_map(struct extent_map *em)
71{
72 if (!em)
73 return;
74 WARN_ON(atomic_read(&em->refs) == 0);
75 if (atomic_dec_and_test(&em->refs)) {
76 WARN_ON(extent_map_in_tree(em));
77 WARN_ON(!list_empty(&em->list));
78 if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
79 kfree(em->map_lookup);
80 kmem_cache_free(extent_map_cache, em);
81 }
82}
83
84/* simple helper to do math around the end of an extent, handling wrap */
85static u64 range_end(u64 start, u64 len)
86{
87 if (start + len < start)
88 return (u64)-1;
89 return start + len;
90}
91
92static int tree_insert(struct rb_root *root, struct extent_map *em)
93{
94 struct rb_node **p = &root->rb_node;
95 struct rb_node *parent = NULL;
96 struct extent_map *entry = NULL;
97 struct rb_node *orig_parent = NULL;
98 u64 end = range_end(em->start, em->len);
99
100 while (*p) {
101 parent = *p;
102 entry = rb_entry(parent, struct extent_map, rb_node);
103
104 if (em->start < entry->start)
105 p = &(*p)->rb_left;
106 else if (em->start >= extent_map_end(entry))
107 p = &(*p)->rb_right;
108 else
109 return -EEXIST;
110 }
111
112 orig_parent = parent;
113 while (parent && em->start >= extent_map_end(entry)) {
114 parent = rb_next(parent);
115 entry = rb_entry(parent, struct extent_map, rb_node);
116 }
117 if (parent)
118 if (end > entry->start && em->start < extent_map_end(entry))
119 return -EEXIST;
120
121 parent = orig_parent;
122 entry = rb_entry(parent, struct extent_map, rb_node);
123 while (parent && em->start < entry->start) {
124 parent = rb_prev(parent);
125 entry = rb_entry(parent, struct extent_map, rb_node);
126 }
127 if (parent)
128 if (end > entry->start && em->start < extent_map_end(entry))
129 return -EEXIST;
130
131 rb_link_node(&em->rb_node, orig_parent, p);
132 rb_insert_color(&em->rb_node, root);
133 return 0;
134}
135
136/*
137 * search through the tree for an extent_map with a given offset. If
138 * it can't be found, try to find some neighboring extents
139 */
140static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
141 struct rb_node **prev_ret,
142 struct rb_node **next_ret)
143{
144 struct rb_node *n = root->rb_node;
145 struct rb_node *prev = NULL;
146 struct rb_node *orig_prev = NULL;
147 struct extent_map *entry;
148 struct extent_map *prev_entry = NULL;
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 >= extent_map_end(entry))
158 n = n->rb_right;
159 else
160 return n;
161 }
162
163 if (prev_ret) {
164 orig_prev = prev;
165 while (prev && offset >= extent_map_end(prev_entry)) {
166 prev = rb_next(prev);
167 prev_entry = rb_entry(prev, struct extent_map, rb_node);
168 }
169 *prev_ret = prev;
170 prev = orig_prev;
171 }
172
173 if (next_ret) {
174 prev_entry = rb_entry(prev, struct extent_map, rb_node);
175 while (prev && offset < prev_entry->start) {
176 prev = rb_prev(prev);
177 prev_entry = rb_entry(prev, struct extent_map, rb_node);
178 }
179 *next_ret = prev;
180 }
181 return NULL;
182}
183
184/* check to see if two extent_map structs are adjacent and safe to merge */
185static int mergable_maps(struct extent_map *prev, struct extent_map *next)
186{
187 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
188 return 0;
189
190 /*
191 * don't merge compressed extents, we need to know their
192 * actual size
193 */
194 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
195 return 0;
196
197 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
198 test_bit(EXTENT_FLAG_LOGGING, &next->flags))
199 return 0;
200
201 /*
202 * We don't want to merge stuff that hasn't been written to the log yet
203 * since it may not reflect exactly what is on disk, and that would be
204 * bad.
205 */
206 if (!list_empty(&prev->list) || !list_empty(&next->list))
207 return 0;
208
209 if (extent_map_end(prev) == next->start &&
210 prev->flags == next->flags &&
211 prev->bdev == next->bdev &&
212 ((next->block_start == EXTENT_MAP_HOLE &&
213 prev->block_start == EXTENT_MAP_HOLE) ||
214 (next->block_start == EXTENT_MAP_INLINE &&
215 prev->block_start == EXTENT_MAP_INLINE) ||
216 (next->block_start == EXTENT_MAP_DELALLOC &&
217 prev->block_start == EXTENT_MAP_DELALLOC) ||
218 (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
219 next->block_start == extent_map_block_end(prev)))) {
220 return 1;
221 }
222 return 0;
223}
224
225static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
226{
227 struct extent_map *merge = NULL;
228 struct rb_node *rb;
229
230 if (em->start != 0) {
231 rb = rb_prev(&em->rb_node);
232 if (rb)
233 merge = rb_entry(rb, struct extent_map, rb_node);
234 if (rb && mergable_maps(merge, em)) {
235 em->start = merge->start;
236 em->orig_start = merge->orig_start;
237 em->len += merge->len;
238 em->block_len += merge->block_len;
239 em->block_start = merge->block_start;
240 em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
241 em->mod_start = merge->mod_start;
242 em->generation = max(em->generation, merge->generation);
243
244 rb_erase(&merge->rb_node, &tree->map);
245 RB_CLEAR_NODE(&merge->rb_node);
246 free_extent_map(merge);
247 }
248 }
249
250 rb = rb_next(&em->rb_node);
251 if (rb)
252 merge = rb_entry(rb, struct extent_map, rb_node);
253 if (rb && mergable_maps(em, merge)) {
254 em->len += merge->len;
255 em->block_len += merge->block_len;
256 rb_erase(&merge->rb_node, &tree->map);
257 RB_CLEAR_NODE(&merge->rb_node);
258 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
259 em->generation = max(em->generation, merge->generation);
260 free_extent_map(merge);
261 }
262}
263
264/**
265 * unpin_extent_cache - unpin an extent from the cache
266 * @tree: tree to unpin the extent in
267 * @start: logical offset in the file
268 * @len: length of the extent
269 * @gen: generation that this extent has been modified in
270 *
271 * Called after an extent has been written to disk properly. Set the generation
272 * to the generation that actually added the file item to the inode so we know
273 * we need to sync this extent when we call fsync().
274 */
275int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
276 u64 gen)
277{
278 int ret = 0;
279 struct extent_map *em;
280 bool prealloc = false;
281
282 write_lock(&tree->lock);
283 em = lookup_extent_mapping(tree, start, len);
284
285 WARN_ON(!em || em->start != start);
286
287 if (!em)
288 goto out;
289
290 em->generation = gen;
291 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
292 em->mod_start = em->start;
293 em->mod_len = em->len;
294
295 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
296 prealloc = true;
297 clear_bit(EXTENT_FLAG_FILLING, &em->flags);
298 }
299
300 try_merge_map(tree, em);
301
302 if (prealloc) {
303 em->mod_start = em->start;
304 em->mod_len = em->len;
305 }
306
307 free_extent_map(em);
308out:
309 write_unlock(&tree->lock);
310 return ret;
311
312}
313
314void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
315{
316 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
317 if (extent_map_in_tree(em))
318 try_merge_map(tree, em);
319}
320
321static inline void setup_extent_mapping(struct extent_map_tree *tree,
322 struct extent_map *em,
323 int modified)
324{
325 atomic_inc(&em->refs);
326 em->mod_start = em->start;
327 em->mod_len = em->len;
328
329 if (modified)
330 list_move(&em->list, &tree->modified_extents);
331 else
332 try_merge_map(tree, em);
333}
334
335/**
336 * add_extent_mapping - add new extent map to the extent tree
337 * @tree: tree to insert new map in
338 * @em: map to insert
339 *
340 * Insert @em into @tree or perform a simple forward/backward merge with
341 * existing mappings. The extent_map struct passed in will be inserted
342 * into the tree directly, with an additional reference taken, or a
343 * reference dropped if the merge attempt was successful.
344 */
345int add_extent_mapping(struct extent_map_tree *tree,
346 struct extent_map *em, int modified)
347{
348 int ret = 0;
349
350 ret = tree_insert(&tree->map, em);
351 if (ret)
352 goto out;
353
354 setup_extent_mapping(tree, em, modified);
355out:
356 return ret;
357}
358
359static struct extent_map *
360__lookup_extent_mapping(struct extent_map_tree *tree,
361 u64 start, u64 len, int strict)
362{
363 struct extent_map *em;
364 struct rb_node *rb_node;
365 struct rb_node *prev = NULL;
366 struct rb_node *next = NULL;
367 u64 end = range_end(start, len);
368
369 rb_node = __tree_search(&tree->map, start, &prev, &next);
370 if (!rb_node) {
371 if (prev)
372 rb_node = prev;
373 else if (next)
374 rb_node = next;
375 else
376 return NULL;
377 }
378
379 em = rb_entry(rb_node, struct extent_map, rb_node);
380
381 if (strict && !(end > em->start && start < extent_map_end(em)))
382 return NULL;
383
384 atomic_inc(&em->refs);
385 return em;
386}
387
388/**
389 * lookup_extent_mapping - lookup extent_map
390 * @tree: tree to lookup in
391 * @start: byte offset to start the search
392 * @len: length of the lookup range
393 *
394 * Find and return the first extent_map struct in @tree that intersects the
395 * [start, len] range. There may be additional objects in the tree that
396 * intersect, so check the object returned carefully to make sure that no
397 * additional lookups are needed.
398 */
399struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
400 u64 start, u64 len)
401{
402 return __lookup_extent_mapping(tree, start, len, 1);
403}
404
405/**
406 * search_extent_mapping - find a nearby extent map
407 * @tree: tree to lookup in
408 * @start: byte offset to start the search
409 * @len: length of the lookup range
410 *
411 * Find and return the first extent_map struct in @tree that intersects the
412 * [start, len] range.
413 *
414 * If one can't be found, any nearby extent may be returned
415 */
416struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
417 u64 start, u64 len)
418{
419 return __lookup_extent_mapping(tree, start, len, 0);
420}
421
422/**
423 * remove_extent_mapping - removes an extent_map from the extent tree
424 * @tree: extent tree to remove from
425 * @em: extent map being removed
426 *
427 * Removes @em from @tree. No reference counts are dropped, and no checks
428 * are done to see if the range is in use
429 */
430int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
431{
432 int ret = 0;
433
434 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
435 rb_erase(&em->rb_node, &tree->map);
436 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
437 list_del_init(&em->list);
438 RB_CLEAR_NODE(&em->rb_node);
439 return ret;
440}
441
442void replace_extent_mapping(struct extent_map_tree *tree,
443 struct extent_map *cur,
444 struct extent_map *new,
445 int modified)
446{
447 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
448 ASSERT(extent_map_in_tree(cur));
449 if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
450 list_del_init(&cur->list);
451 rb_replace_node(&cur->rb_node, &new->rb_node, &tree->map);
452 RB_CLEAR_NODE(&cur->rb_node);
453
454 setup_extent_mapping(tree, new, modified);
455}
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_extent_mapping - add new extent map to the extent tree
389 * @tree: tree to insert new map in
390 * @em: map to insert
391 *
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.
396 */
397int add_extent_mapping(struct extent_map_tree *tree,
398 struct extent_map *em, int modified)
399{
400 int ret = 0;
401
402 lockdep_assert_held_write(&tree->lock);
403
404 ret = tree_insert(&tree->map, em);
405 if (ret)
406 goto out;
407
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);
412 }
413out:
414 return ret;
415}
416
417static struct extent_map *
418__lookup_extent_mapping(struct extent_map_tree *tree,
419 u64 start, u64 len, int strict)
420{
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);
426
427 rb_node = __tree_search(&tree->map.rb_root, start, &prev, &next);
428 if (!rb_node) {
429 if (prev)
430 rb_node = prev;
431 else if (next)
432 rb_node = next;
433 else
434 return NULL;
435 }
436
437 em = rb_entry(rb_node, struct extent_map, rb_node);
438
439 if (strict && !(end > em->start && start < extent_map_end(em)))
440 return NULL;
441
442 refcount_inc(&em->refs);
443 return em;
444}
445
446/**
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
451 *
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.
456 */
457struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
458 u64 start, u64 len)
459{
460 return __lookup_extent_mapping(tree, start, len, 1);
461}
462
463/**
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
468 *
469 * Find and return the first extent_map struct in @tree that intersects the
470 * [start, len] range.
471 *
472 * If one can't be found, any nearby extent may be returned
473 */
474struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
475 u64 start, u64 len)
476{
477 return __lookup_extent_mapping(tree, start, len, 0);
478}
479
480/**
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
484 *
485 * Removes @em from @tree. No reference counts are dropped, and no checks
486 * are done to see if the range is in use
487 */
488void remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
489{
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);
497}
498
499void replace_extent_mapping(struct extent_map_tree *tree,
500 struct extent_map *cur,
501 struct extent_map *new,
502 int modified)
503{
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);
510
511 setup_extent_mapping(tree, new, modified);
512}
513
514static struct extent_map *next_extent_map(struct extent_map *em)
515{
516 struct rb_node *next;
517
518 next = rb_next(&em->rb_node);
519 if (!next)
520 return NULL;
521 return container_of(next, struct extent_map, rb_node);
522}
523
524static struct extent_map *prev_extent_map(struct extent_map *em)
525{
526 struct rb_node *prev;
527
528 prev = rb_prev(&em->rb_node);
529 if (!prev)
530 return NULL;
531 return container_of(prev, struct extent_map, rb_node);
532}
533
534/*
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.
539 */
540static noinline int merge_extent_mapping(struct extent_map_tree *em_tree,
541 struct extent_map *existing,
542 struct extent_map *em,
543 u64 map_start)
544{
545 struct extent_map *prev;
546 struct extent_map *next;
547 u64 start;
548 u64 end;
549 u64 start_diff;
550
551 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
552
553 if (existing->start > map_start) {
554 next = existing;
555 prev = prev_extent_map(next);
556 } else {
557 prev = existing;
558 next = next_extent_map(prev);
559 }
560
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;
566 em->start = 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;
572 }
573 return add_extent_mapping(em_tree, em, 0);
574}
575
576/**
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
583 *
584 * Note that @em_in's range may be different from [start, start+len),
585 * but they must be overlapped.
586 *
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
590 * existing em.
591 * b) Merge the existing extent with @em_in passed in.
592 *
593 * Return 0 on success, otherwise -EEXIST.
594 *
595 */
596int 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)
599{
600 int ret;
601 struct extent_map *em = *em_in;
602
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
607 */
608 if (ret == -EEXIST) {
609 struct extent_map *existing;
610
611 ret = 0;
612
613 existing = search_extent_mapping(em_tree, start, len);
614
615 trace_btrfs_handle_em_exist(fs_info, existing, em, start, len);
616
617 /*
618 * existing will always be non-NULL, since there must be
619 * extent causing the -EEXIST.
620 */
621 if (start >= existing->start &&
622 start < extent_map_end(existing)) {
623 free_extent_map(em);
624 *em_in = existing;
625 ret = 0;
626 } else {
627 u64 orig_start = em->start;
628 u64 orig_len = em->len;
629
630 /*
631 * The existing extent map is the one nearest to
632 * the [start, start + len) range which overlaps
633 */
634 ret = merge_extent_mapping(em_tree, existing,
635 em, start);
636 if (ret) {
637 free_extent_map(em);
638 *em_in = NULL;
639 WARN_ONCE(ret,
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);
643 }
644 free_extent_map(existing);
645 }
646 }
647
648 ASSERT(ret == 0 || ret == -EEXIST);
649 return ret;
650}