2 #include <linux/slab.h>
3 #include <linux/spinlock.h>
4 #include <linux/hardirq.h>
6 #include "extent_map.h"
9 static struct kmem_cache *extent_map_cache;
11 int __init extent_map_init(void)
13 extent_map_cache = kmem_cache_create("btrfs_extent_map",
14 sizeof(struct extent_map), 0,
15 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
16 if (!extent_map_cache)
21 void extent_map_exit(void)
24 kmem_cache_destroy(extent_map_cache);
28 * extent_map_tree_init - initialize extent map tree
29 * @tree: tree to initialize
31 * Initialize the extent tree @tree. Should be called for each new inode
32 * or other user of the extent_map interface.
34 void extent_map_tree_init(struct extent_map_tree *tree)
37 INIT_LIST_HEAD(&tree->modified_extents);
38 rwlock_init(&tree->lock);
42 * alloc_extent_map - allocate new extent map structure
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()
48 struct extent_map *alloc_extent_map(void)
50 struct extent_map *em;
51 em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
54 RB_CLEAR_NODE(&em->rb_node);
56 em->compress_type = BTRFS_COMPRESS_NONE;
58 atomic_set(&em->refs, 1);
59 INIT_LIST_HEAD(&em->list);
64 * free_extent_map - drop reference count of an extent_map
65 * @em: extent map beeing releasead
67 * Drops the reference out on @em by one and free the structure
68 * if the reference count hits zero.
70 void free_extent_map(struct extent_map *em)
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 kmem_cache_free(extent_map_cache, em);
82 /* simple helper to do math around the end of an extent, handling wrap */
83 static u64 range_end(u64 start, u64 len)
85 if (start + len < start)
90 static int tree_insert(struct rb_root *root, struct extent_map *em)
92 struct rb_node **p = &root->rb_node;
93 struct rb_node *parent = NULL;
94 struct extent_map *entry = NULL;
95 struct rb_node *orig_parent = NULL;
96 u64 end = range_end(em->start, em->len);
100 entry = rb_entry(parent, struct extent_map, rb_node);
102 if (em->start < entry->start)
104 else if (em->start >= extent_map_end(entry))
110 orig_parent = parent;
111 while (parent && em->start >= extent_map_end(entry)) {
112 parent = rb_next(parent);
113 entry = rb_entry(parent, struct extent_map, rb_node);
116 if (end > entry->start && em->start < extent_map_end(entry))
119 parent = orig_parent;
120 entry = rb_entry(parent, struct extent_map, rb_node);
121 while (parent && em->start < entry->start) {
122 parent = rb_prev(parent);
123 entry = rb_entry(parent, struct extent_map, rb_node);
126 if (end > entry->start && em->start < extent_map_end(entry))
129 rb_link_node(&em->rb_node, orig_parent, p);
130 rb_insert_color(&em->rb_node, root);
135 * search through the tree for an extent_map with a given offset. If
136 * it can't be found, try to find some neighboring extents
138 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
139 struct rb_node **prev_ret,
140 struct rb_node **next_ret)
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;
149 entry = rb_entry(n, struct extent_map, rb_node);
153 if (offset < entry->start)
155 else if (offset >= extent_map_end(entry))
163 while (prev && offset >= extent_map_end(prev_entry)) {
164 prev = rb_next(prev);
165 prev_entry = rb_entry(prev, struct extent_map, rb_node);
172 prev_entry = rb_entry(prev, struct extent_map, rb_node);
173 while (prev && offset < prev_entry->start) {
174 prev = rb_prev(prev);
175 prev_entry = rb_entry(prev, struct extent_map, rb_node);
182 /* check to see if two extent_map structs are adjacent and safe to merge */
183 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
185 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
189 * don't merge compressed extents, we need to know their
192 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
195 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
196 test_bit(EXTENT_FLAG_LOGGING, &next->flags))
200 * We don't want to merge stuff that hasn't been written to the log yet
201 * since it may not reflect exactly what is on disk, and that would be
204 if (!list_empty(&prev->list) || !list_empty(&next->list))
207 if (extent_map_end(prev) == next->start &&
208 prev->flags == next->flags &&
209 prev->bdev == next->bdev &&
210 ((next->block_start == EXTENT_MAP_HOLE &&
211 prev->block_start == EXTENT_MAP_HOLE) ||
212 (next->block_start == EXTENT_MAP_INLINE &&
213 prev->block_start == EXTENT_MAP_INLINE) ||
214 (next->block_start == EXTENT_MAP_DELALLOC &&
215 prev->block_start == EXTENT_MAP_DELALLOC) ||
216 (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
217 next->block_start == extent_map_block_end(prev)))) {
223 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
225 struct extent_map *merge = NULL;
228 if (em->start != 0) {
229 rb = rb_prev(&em->rb_node);
231 merge = rb_entry(rb, struct extent_map, rb_node);
232 if (rb && mergable_maps(merge, em)) {
233 em->start = merge->start;
234 em->orig_start = merge->orig_start;
235 em->len += merge->len;
236 em->block_len += merge->block_len;
237 em->block_start = merge->block_start;
238 em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
239 em->mod_start = merge->mod_start;
240 em->generation = max(em->generation, merge->generation);
242 rb_erase(&merge->rb_node, &tree->map);
243 RB_CLEAR_NODE(&merge->rb_node);
244 free_extent_map(merge);
248 rb = rb_next(&em->rb_node);
250 merge = rb_entry(rb, struct extent_map, rb_node);
251 if (rb && mergable_maps(em, merge)) {
252 em->len += merge->len;
253 em->block_len += merge->block_len;
254 rb_erase(&merge->rb_node, &tree->map);
255 RB_CLEAR_NODE(&merge->rb_node);
256 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
257 em->generation = max(em->generation, merge->generation);
258 free_extent_map(merge);
263 * unpin_extent_cache - unpin an extent from the cache
264 * @tree: tree to unpin the extent in
265 * @start: logical offset in the file
266 * @len: length of the extent
267 * @gen: generation that this extent has been modified in
269 * Called after an extent has been written to disk properly. Set the generation
270 * to the generation that actually added the file item to the inode so we know
271 * we need to sync this extent when we call fsync().
273 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
277 struct extent_map *em;
278 bool prealloc = false;
280 write_lock(&tree->lock);
281 em = lookup_extent_mapping(tree, start, len);
283 WARN_ON(!em || em->start != start);
288 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
289 list_move(&em->list, &tree->modified_extents);
290 em->generation = gen;
291 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
292 em->mod_start = em->start;
293 em->mod_len = em->len;
295 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
297 clear_bit(EXTENT_FLAG_FILLING, &em->flags);
300 try_merge_map(tree, em);
303 em->mod_start = em->start;
304 em->mod_len = em->len;
309 write_unlock(&tree->lock);
314 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
316 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
317 if (extent_map_in_tree(em))
318 try_merge_map(tree, em);
321 static inline void setup_extent_mapping(struct extent_map_tree *tree,
322 struct extent_map *em,
325 atomic_inc(&em->refs);
326 em->mod_start = em->start;
327 em->mod_len = em->len;
330 list_move(&em->list, &tree->modified_extents);
332 try_merge_map(tree, em);
336 * add_extent_mapping - add new extent map to the extent tree
337 * @tree: tree to insert new map in
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.
345 int add_extent_mapping(struct extent_map_tree *tree,
346 struct extent_map *em, int modified)
350 ret = tree_insert(&tree->map, em);
354 setup_extent_mapping(tree, em, modified);
359 static struct extent_map *
360 __lookup_extent_mapping(struct extent_map_tree *tree,
361 u64 start, u64 len, int strict)
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);
369 rb_node = __tree_search(&tree->map, start, &prev, &next);
379 em = rb_entry(rb_node, struct extent_map, rb_node);
381 if (strict && !(end > em->start && start < extent_map_end(em)))
384 atomic_inc(&em->refs);
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
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.
399 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
402 return __lookup_extent_mapping(tree, start, len, 1);
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
411 * Find and return the first extent_map struct in @tree that intersects the
412 * [start, len] range.
414 * If one can't be found, any nearby extent may be returned
416 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
419 return __lookup_extent_mapping(tree, start, len, 0);
423 * remove_extent_mapping - removes an extent_map from the extent tree
424 * @tree: extent tree to remove from
425 * @em: extent map beeing removed
427 * Removes @em from @tree. No reference counts are dropped, and no checks
428 * are done to see if the range is in use
430 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
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);
442 void replace_extent_mapping(struct extent_map_tree *tree,
443 struct extent_map *cur,
444 struct extent_map *new,
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);
454 setup_extent_mapping(tree, new, modified);