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);
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)) {
77 WARN_ON(!list_empty(&em->list));
78 kmem_cache_free(extent_map_cache, em);
82 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
85 struct rb_node **p = &root->rb_node;
86 struct rb_node *parent = NULL;
87 struct extent_map *entry;
91 entry = rb_entry(parent, struct extent_map, rb_node);
93 WARN_ON(!entry->in_tree);
95 if (offset < entry->start)
97 else if (offset >= extent_map_end(entry))
103 entry = rb_entry(node, struct extent_map, rb_node);
105 rb_link_node(node, parent, p);
106 rb_insert_color(node, root);
111 * search through the tree for an extent_map with a given offset. If
112 * it can't be found, try to find some neighboring extents
114 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
115 struct rb_node **prev_ret,
116 struct rb_node **next_ret)
118 struct rb_node *n = root->rb_node;
119 struct rb_node *prev = NULL;
120 struct rb_node *orig_prev = NULL;
121 struct extent_map *entry;
122 struct extent_map *prev_entry = NULL;
125 entry = rb_entry(n, struct extent_map, rb_node);
129 WARN_ON(!entry->in_tree);
131 if (offset < entry->start)
133 else if (offset >= extent_map_end(entry))
141 while (prev && offset >= extent_map_end(prev_entry)) {
142 prev = rb_next(prev);
143 prev_entry = rb_entry(prev, struct extent_map, rb_node);
150 prev_entry = rb_entry(prev, struct extent_map, rb_node);
151 while (prev && offset < prev_entry->start) {
152 prev = rb_prev(prev);
153 prev_entry = rb_entry(prev, struct extent_map, rb_node);
160 /* check to see if two extent_map structs are adjacent and safe to merge */
161 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
163 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
167 * don't merge compressed extents, we need to know their
170 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
173 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
174 test_bit(EXTENT_FLAG_LOGGING, &next->flags))
178 * We don't want to merge stuff that hasn't been written to the log yet
179 * since it may not reflect exactly what is on disk, and that would be
182 if (!list_empty(&prev->list) || !list_empty(&next->list))
185 if (extent_map_end(prev) == next->start &&
186 prev->flags == next->flags &&
187 prev->bdev == next->bdev &&
188 ((next->block_start == EXTENT_MAP_HOLE &&
189 prev->block_start == EXTENT_MAP_HOLE) ||
190 (next->block_start == EXTENT_MAP_INLINE &&
191 prev->block_start == EXTENT_MAP_INLINE) ||
192 (next->block_start == EXTENT_MAP_DELALLOC &&
193 prev->block_start == EXTENT_MAP_DELALLOC) ||
194 (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
195 next->block_start == extent_map_block_end(prev)))) {
201 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
203 struct extent_map *merge = NULL;
206 if (em->start != 0) {
207 rb = rb_prev(&em->rb_node);
209 merge = rb_entry(rb, struct extent_map, rb_node);
210 if (rb && mergable_maps(merge, em)) {
211 em->start = merge->start;
212 em->orig_start = merge->orig_start;
213 em->len += merge->len;
214 em->block_len += merge->block_len;
215 em->block_start = merge->block_start;
217 em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
218 em->mod_start = merge->mod_start;
219 em->generation = max(em->generation, merge->generation);
221 rb_erase(&merge->rb_node, &tree->map);
222 free_extent_map(merge);
226 rb = rb_next(&em->rb_node);
228 merge = rb_entry(rb, struct extent_map, rb_node);
229 if (rb && mergable_maps(em, merge)) {
230 em->len += merge->len;
231 em->block_len += merge->len;
232 rb_erase(&merge->rb_node, &tree->map);
234 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
235 em->generation = max(em->generation, merge->generation);
236 free_extent_map(merge);
241 * unpin_extent_cache - unpin an extent from the cache
242 * @tree: tree to unpin the extent in
243 * @start: logical offset in the file
244 * @len: length of the extent
245 * @gen: generation that this extent has been modified in
247 * Called after an extent has been written to disk properly. Set the generation
248 * to the generation that actually added the file item to the inode so we know
249 * we need to sync this extent when we call fsync().
251 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
255 struct extent_map *em;
256 bool prealloc = false;
258 write_lock(&tree->lock);
259 em = lookup_extent_mapping(tree, start, len);
261 WARN_ON(!em || em->start != start);
266 em->generation = gen;
267 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
268 em->mod_start = em->start;
269 em->mod_len = em->len;
271 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
273 clear_bit(EXTENT_FLAG_FILLING, &em->flags);
276 try_merge_map(tree, em);
279 em->mod_start = em->start;
280 em->mod_len = em->len;
285 write_unlock(&tree->lock);
290 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
292 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
294 try_merge_map(tree, em);
298 * add_extent_mapping - add new extent map to the extent tree
299 * @tree: tree to insert new map in
302 * Insert @em into @tree or perform a simple forward/backward merge with
303 * existing mappings. The extent_map struct passed in will be inserted
304 * into the tree directly, with an additional reference taken, or a
305 * reference dropped if the merge attempt was successful.
307 int add_extent_mapping(struct extent_map_tree *tree,
308 struct extent_map *em, int modified)
312 struct extent_map *exist;
314 exist = lookup_extent_mapping(tree, em->start, em->len);
316 free_extent_map(exist);
320 rb = tree_insert(&tree->map, em->start, &em->rb_node);
325 atomic_inc(&em->refs);
327 em->mod_start = em->start;
328 em->mod_len = em->len;
331 list_move(&em->list, &tree->modified_extents);
333 try_merge_map(tree, em);
338 /* simple helper to do math around the end of an extent, handling wrap */
339 static u64 range_end(u64 start, u64 len)
341 if (start + len < start)
346 static struct extent_map *
347 __lookup_extent_mapping(struct extent_map_tree *tree,
348 u64 start, u64 len, int strict)
350 struct extent_map *em;
351 struct rb_node *rb_node;
352 struct rb_node *prev = NULL;
353 struct rb_node *next = NULL;
354 u64 end = range_end(start, len);
356 rb_node = __tree_search(&tree->map, start, &prev, &next);
366 em = rb_entry(rb_node, struct extent_map, rb_node);
368 if (strict && !(end > em->start && start < extent_map_end(em)))
371 atomic_inc(&em->refs);
376 * lookup_extent_mapping - lookup extent_map
377 * @tree: tree to lookup in
378 * @start: byte offset to start the search
379 * @len: length of the lookup range
381 * Find and return the first extent_map struct in @tree that intersects the
382 * [start, len] range. There may be additional objects in the tree that
383 * intersect, so check the object returned carefully to make sure that no
384 * additional lookups are needed.
386 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
389 return __lookup_extent_mapping(tree, start, len, 1);
393 * search_extent_mapping - find a nearby extent map
394 * @tree: tree to lookup in
395 * @start: byte offset to start the search
396 * @len: length of the lookup range
398 * Find and return the first extent_map struct in @tree that intersects the
399 * [start, len] range.
401 * If one can't be found, any nearby extent may be returned
403 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
406 return __lookup_extent_mapping(tree, start, len, 0);
410 * remove_extent_mapping - removes an extent_map from the extent tree
411 * @tree: extent tree to remove from
412 * @em: extent map beeing removed
414 * Removes @em from @tree. No reference counts are dropped, and no checks
415 * are done to see if the range is in use
417 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
421 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
422 rb_erase(&em->rb_node, &tree->map);
423 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
424 list_del_init(&em->list);