Merge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux/fpc-iii.git] / fs / btrfs / extent_map.c
blob996ad56b57db64bbc0516f4674d8ad3fd2d43976
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"
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)
17 return -ENOMEM;
18 return 0;
21 void extent_map_exit(void)
23 if (extent_map_cache)
24 kmem_cache_destroy(extent_map_cache);
27 /**
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)
36 tree->map = RB_ROOT;
37 INIT_LIST_HEAD(&tree->modified_extents);
38 rwlock_init(&tree->lock);
41 /**
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);
52 if (!em)
53 return NULL;
54 em->in_tree = 0;
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;
63 /**
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)
72 if (!em)
73 return;
74 WARN_ON(atomic_read(&em->refs) == 0);
75 if (atomic_dec_and_test(&em->refs)) {
76 WARN_ON(em->in_tree);
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)
86 return (u64)-1;
87 return start + len;
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);
98 while (*p) {
99 parent = *p;
100 entry = rb_entry(parent, struct extent_map, rb_node);
102 WARN_ON(!entry->in_tree);
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;
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);
117 if (parent)
118 if (end > entry->start && em->start < extent_map_end(entry))
119 return -EEXIST;
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);
127 if (parent)
128 if (end > entry->start && em->start < extent_map_end(entry))
129 return -EEXIST;
131 em->in_tree = 1;
132 rb_link_node(&em->rb_node, orig_parent, p);
133 rb_insert_color(&em->rb_node, root);
134 return 0;
138 * search through the tree for an extent_map with a given offset. If
139 * it can't be found, try to find some neighboring extents
141 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
142 struct rb_node **prev_ret,
143 struct rb_node **next_ret)
145 struct rb_node *n = root->rb_node;
146 struct rb_node *prev = NULL;
147 struct rb_node *orig_prev = NULL;
148 struct extent_map *entry;
149 struct extent_map *prev_entry = NULL;
151 while (n) {
152 entry = rb_entry(n, struct extent_map, rb_node);
153 prev = n;
154 prev_entry = entry;
156 WARN_ON(!entry->in_tree);
158 if (offset < entry->start)
159 n = n->rb_left;
160 else if (offset >= extent_map_end(entry))
161 n = n->rb_right;
162 else
163 return n;
166 if (prev_ret) {
167 orig_prev = prev;
168 while (prev && offset >= extent_map_end(prev_entry)) {
169 prev = rb_next(prev);
170 prev_entry = rb_entry(prev, struct extent_map, rb_node);
172 *prev_ret = prev;
173 prev = orig_prev;
176 if (next_ret) {
177 prev_entry = rb_entry(prev, struct extent_map, rb_node);
178 while (prev && offset < prev_entry->start) {
179 prev = rb_prev(prev);
180 prev_entry = rb_entry(prev, struct extent_map, rb_node);
182 *next_ret = prev;
184 return NULL;
187 /* check to see if two extent_map structs are adjacent and safe to merge */
188 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
190 if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
191 return 0;
194 * don't merge compressed extents, we need to know their
195 * actual size
197 if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
198 return 0;
200 if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
201 test_bit(EXTENT_FLAG_LOGGING, &next->flags))
202 return 0;
205 * We don't want to merge stuff that hasn't been written to the log yet
206 * since it may not reflect exactly what is on disk, and that would be
207 * bad.
209 if (!list_empty(&prev->list) || !list_empty(&next->list))
210 return 0;
212 if (extent_map_end(prev) == next->start &&
213 prev->flags == next->flags &&
214 prev->bdev == next->bdev &&
215 ((next->block_start == EXTENT_MAP_HOLE &&
216 prev->block_start == EXTENT_MAP_HOLE) ||
217 (next->block_start == EXTENT_MAP_INLINE &&
218 prev->block_start == EXTENT_MAP_INLINE) ||
219 (next->block_start == EXTENT_MAP_DELALLOC &&
220 prev->block_start == EXTENT_MAP_DELALLOC) ||
221 (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
222 next->block_start == extent_map_block_end(prev)))) {
223 return 1;
225 return 0;
228 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
230 struct extent_map *merge = NULL;
231 struct rb_node *rb;
233 if (em->start != 0) {
234 rb = rb_prev(&em->rb_node);
235 if (rb)
236 merge = rb_entry(rb, struct extent_map, rb_node);
237 if (rb && mergable_maps(merge, em)) {
238 em->start = merge->start;
239 em->orig_start = merge->orig_start;
240 em->len += merge->len;
241 em->block_len += merge->block_len;
242 em->block_start = merge->block_start;
243 merge->in_tree = 0;
244 em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
245 em->mod_start = merge->mod_start;
246 em->generation = max(em->generation, merge->generation);
248 rb_erase(&merge->rb_node, &tree->map);
249 free_extent_map(merge);
253 rb = rb_next(&em->rb_node);
254 if (rb)
255 merge = rb_entry(rb, struct extent_map, rb_node);
256 if (rb && mergable_maps(em, merge)) {
257 em->len += merge->len;
258 em->block_len += merge->block_len;
259 rb_erase(&merge->rb_node, &tree->map);
260 merge->in_tree = 0;
261 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
262 em->generation = max(em->generation, merge->generation);
263 free_extent_map(merge);
268 * unpin_extent_cache - unpin an extent from the cache
269 * @tree: tree to unpin the extent in
270 * @start: logical offset in the file
271 * @len: length of the extent
272 * @gen: generation that this extent has been modified in
274 * Called after an extent has been written to disk properly. Set the generation
275 * to the generation that actually added the file item to the inode so we know
276 * we need to sync this extent when we call fsync().
278 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
279 u64 gen)
281 int ret = 0;
282 struct extent_map *em;
283 bool prealloc = false;
285 write_lock(&tree->lock);
286 em = lookup_extent_mapping(tree, start, len);
288 WARN_ON(!em || em->start != start);
290 if (!em)
291 goto out;
293 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
294 list_move(&em->list, &tree->modified_extents);
295 em->generation = gen;
296 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
297 em->mod_start = em->start;
298 em->mod_len = em->len;
300 if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
301 prealloc = true;
302 clear_bit(EXTENT_FLAG_FILLING, &em->flags);
305 try_merge_map(tree, em);
307 if (prealloc) {
308 em->mod_start = em->start;
309 em->mod_len = em->len;
312 free_extent_map(em);
313 out:
314 write_unlock(&tree->lock);
315 return ret;
319 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
321 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
322 if (em->in_tree)
323 try_merge_map(tree, em);
327 * add_extent_mapping - add new extent map to the extent tree
328 * @tree: tree to insert new map in
329 * @em: map to insert
331 * Insert @em into @tree or perform a simple forward/backward merge with
332 * existing mappings. The extent_map struct passed in will be inserted
333 * into the tree directly, with an additional reference taken, or a
334 * reference dropped if the merge attempt was successful.
336 int add_extent_mapping(struct extent_map_tree *tree,
337 struct extent_map *em, int modified)
339 int ret = 0;
341 ret = tree_insert(&tree->map, em);
342 if (ret)
343 goto out;
345 atomic_inc(&em->refs);
347 em->mod_start = em->start;
348 em->mod_len = em->len;
350 if (modified)
351 list_move(&em->list, &tree->modified_extents);
352 else
353 try_merge_map(tree, em);
354 out:
355 return ret;
358 static struct extent_map *
359 __lookup_extent_mapping(struct extent_map_tree *tree,
360 u64 start, u64 len, int strict)
362 struct extent_map *em;
363 struct rb_node *rb_node;
364 struct rb_node *prev = NULL;
365 struct rb_node *next = NULL;
366 u64 end = range_end(start, len);
368 rb_node = __tree_search(&tree->map, start, &prev, &next);
369 if (!rb_node) {
370 if (prev)
371 rb_node = prev;
372 else if (next)
373 rb_node = next;
374 else
375 return NULL;
378 em = rb_entry(rb_node, struct extent_map, rb_node);
380 if (strict && !(end > em->start && start < extent_map_end(em)))
381 return NULL;
383 atomic_inc(&em->refs);
384 return em;
388 * lookup_extent_mapping - lookup extent_map
389 * @tree: tree to lookup in
390 * @start: byte offset to start the search
391 * @len: length of the lookup range
393 * Find and return the first extent_map struct in @tree that intersects the
394 * [start, len] range. There may be additional objects in the tree that
395 * intersect, so check the object returned carefully to make sure that no
396 * additional lookups are needed.
398 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
399 u64 start, u64 len)
401 return __lookup_extent_mapping(tree, start, len, 1);
405 * search_extent_mapping - find a nearby extent map
406 * @tree: tree to lookup in
407 * @start: byte offset to start the search
408 * @len: length of the lookup range
410 * Find and return the first extent_map struct in @tree that intersects the
411 * [start, len] range.
413 * If one can't be found, any nearby extent may be returned
415 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
416 u64 start, u64 len)
418 return __lookup_extent_mapping(tree, start, len, 0);
422 * remove_extent_mapping - removes an extent_map from the extent tree
423 * @tree: extent tree to remove from
424 * @em: extent map beeing removed
426 * Removes @em from @tree. No reference counts are dropped, and no checks
427 * are done to see if the range is in use
429 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
431 int ret = 0;
433 WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
434 rb_erase(&em->rb_node, &tree->map);
435 if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
436 list_del_init(&em->list);
437 em->in_tree = 0;
438 return ret;