extent_map.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414
  1. #include <linux/err.h>
  2. #include <linux/slab.h>
  3. #include <linux/module.h>
  4. #include <linux/spinlock.h>
  5. #include <linux/hardirq.h>
  6. #include "ctree.h"
  7. #include "extent_map.h"
  8. static struct kmem_cache *extent_map_cache;
  9. int __init extent_map_init(void)
  10. {
  11. extent_map_cache = kmem_cache_create("btrfs_extent_map",
  12. sizeof(struct extent_map), 0,
  13. SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
  14. if (!extent_map_cache)
  15. return -ENOMEM;
  16. return 0;
  17. }
  18. void extent_map_exit(void)
  19. {
  20. if (extent_map_cache)
  21. kmem_cache_destroy(extent_map_cache);
  22. }
  23. /**
  24. * extent_map_tree_init - initialize extent map tree
  25. * @tree: tree to initialize
  26. *
  27. * Initialize the extent tree @tree. Should be called for each new inode
  28. * or other user of the extent_map interface.
  29. */
  30. void extent_map_tree_init(struct extent_map_tree *tree)
  31. {
  32. tree->map = RB_ROOT;
  33. INIT_LIST_HEAD(&tree->modified_extents);
  34. rwlock_init(&tree->lock);
  35. }
  36. /**
  37. * alloc_extent_map - allocate new extent map structure
  38. *
  39. * Allocate a new extent_map structure. The new structure is
  40. * returned with a reference count of one and needs to be
  41. * freed using free_extent_map()
  42. */
  43. struct extent_map *alloc_extent_map(void)
  44. {
  45. struct extent_map *em;
  46. em = kmem_cache_alloc(extent_map_cache, GFP_NOFS);
  47. if (!em)
  48. return NULL;
  49. em->in_tree = 0;
  50. em->flags = 0;
  51. em->compress_type = BTRFS_COMPRESS_NONE;
  52. em->generation = 0;
  53. atomic_set(&em->refs, 1);
  54. INIT_LIST_HEAD(&em->list);
  55. return em;
  56. }
  57. /**
  58. * free_extent_map - drop reference count of an extent_map
  59. * @em: extent map beeing releasead
  60. *
  61. * Drops the reference out on @em by one and free the structure
  62. * if the reference count hits zero.
  63. */
  64. void free_extent_map(struct extent_map *em)
  65. {
  66. if (!em)
  67. return;
  68. WARN_ON(atomic_read(&em->refs) == 0);
  69. if (atomic_dec_and_test(&em->refs)) {
  70. WARN_ON(em->in_tree);
  71. WARN_ON(!list_empty(&em->list));
  72. kmem_cache_free(extent_map_cache, em);
  73. }
  74. }
  75. static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
  76. struct rb_node *node)
  77. {
  78. struct rb_node **p = &root->rb_node;
  79. struct rb_node *parent = NULL;
  80. struct extent_map *entry;
  81. while (*p) {
  82. parent = *p;
  83. entry = rb_entry(parent, struct extent_map, rb_node);
  84. WARN_ON(!entry->in_tree);
  85. if (offset < entry->start)
  86. p = &(*p)->rb_left;
  87. else if (offset >= extent_map_end(entry))
  88. p = &(*p)->rb_right;
  89. else
  90. return parent;
  91. }
  92. entry = rb_entry(node, struct extent_map, rb_node);
  93. entry->in_tree = 1;
  94. rb_link_node(node, parent, p);
  95. rb_insert_color(node, root);
  96. return NULL;
  97. }
  98. /*
  99. * search through the tree for an extent_map with a given offset. If
  100. * it can't be found, try to find some neighboring extents
  101. */
  102. static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
  103. struct rb_node **prev_ret,
  104. struct rb_node **next_ret)
  105. {
  106. struct rb_node *n = root->rb_node;
  107. struct rb_node *prev = NULL;
  108. struct rb_node *orig_prev = NULL;
  109. struct extent_map *entry;
  110. struct extent_map *prev_entry = NULL;
  111. while (n) {
  112. entry = rb_entry(n, struct extent_map, rb_node);
  113. prev = n;
  114. prev_entry = entry;
  115. WARN_ON(!entry->in_tree);
  116. if (offset < entry->start)
  117. n = n->rb_left;
  118. else if (offset >= extent_map_end(entry))
  119. n = n->rb_right;
  120. else
  121. return n;
  122. }
  123. if (prev_ret) {
  124. orig_prev = prev;
  125. while (prev && offset >= extent_map_end(prev_entry)) {
  126. prev = rb_next(prev);
  127. prev_entry = rb_entry(prev, struct extent_map, rb_node);
  128. }
  129. *prev_ret = prev;
  130. prev = orig_prev;
  131. }
  132. if (next_ret) {
  133. prev_entry = rb_entry(prev, struct extent_map, rb_node);
  134. while (prev && offset < prev_entry->start) {
  135. prev = rb_prev(prev);
  136. prev_entry = rb_entry(prev, struct extent_map, rb_node);
  137. }
  138. *next_ret = prev;
  139. }
  140. return NULL;
  141. }
  142. /* check to see if two extent_map structs are adjacent and safe to merge */
  143. static int mergable_maps(struct extent_map *prev, struct extent_map *next)
  144. {
  145. if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
  146. return 0;
  147. /*
  148. * don't merge compressed extents, we need to know their
  149. * actual size
  150. */
  151. if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
  152. return 0;
  153. if (extent_map_end(prev) == next->start &&
  154. prev->flags == next->flags &&
  155. prev->bdev == next->bdev &&
  156. ((next->block_start == EXTENT_MAP_HOLE &&
  157. prev->block_start == EXTENT_MAP_HOLE) ||
  158. (next->block_start == EXTENT_MAP_INLINE &&
  159. prev->block_start == EXTENT_MAP_INLINE) ||
  160. (next->block_start == EXTENT_MAP_DELALLOC &&
  161. prev->block_start == EXTENT_MAP_DELALLOC) ||
  162. (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
  163. next->block_start == extent_map_block_end(prev)))) {
  164. return 1;
  165. }
  166. return 0;
  167. }
  168. static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
  169. {
  170. struct extent_map *merge = NULL;
  171. struct rb_node *rb;
  172. if (em->start != 0) {
  173. rb = rb_prev(&em->rb_node);
  174. if (rb)
  175. merge = rb_entry(rb, struct extent_map, rb_node);
  176. if (rb && mergable_maps(merge, em)) {
  177. em->start = merge->start;
  178. em->len += merge->len;
  179. em->block_len += merge->block_len;
  180. em->block_start = merge->block_start;
  181. merge->in_tree = 0;
  182. if (merge->generation > em->generation) {
  183. em->mod_start = em->start;
  184. em->mod_len = em->len;
  185. em->generation = merge->generation;
  186. list_move(&em->list, &tree->modified_extents);
  187. }
  188. list_del_init(&merge->list);
  189. rb_erase(&merge->rb_node, &tree->map);
  190. free_extent_map(merge);
  191. }
  192. }
  193. rb = rb_next(&em->rb_node);
  194. if (rb)
  195. merge = rb_entry(rb, struct extent_map, rb_node);
  196. if (rb && mergable_maps(em, merge)) {
  197. em->len += merge->len;
  198. em->block_len += merge->len;
  199. rb_erase(&merge->rb_node, &tree->map);
  200. merge->in_tree = 0;
  201. if (merge->generation > em->generation) {
  202. em->mod_len = em->len;
  203. em->generation = merge->generation;
  204. list_move(&em->list, &tree->modified_extents);
  205. }
  206. list_del_init(&merge->list);
  207. free_extent_map(merge);
  208. }
  209. }
  210. /**
  211. * unpint_extent_cache - unpin an extent from the cache
  212. * @tree: tree to unpin the extent in
  213. * @start: logical offset in the file
  214. * @len: length of the extent
  215. * @gen: generation that this extent has been modified in
  216. * @prealloc: if this is set we need to clear the prealloc flag
  217. *
  218. * Called after an extent has been written to disk properly. Set the generation
  219. * to the generation that actually added the file item to the inode so we know
  220. * we need to sync this extent when we call fsync().
  221. */
  222. int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
  223. u64 gen)
  224. {
  225. int ret = 0;
  226. struct extent_map *em;
  227. bool prealloc = false;
  228. write_lock(&tree->lock);
  229. em = lookup_extent_mapping(tree, start, len);
  230. WARN_ON(!em || em->start != start);
  231. if (!em)
  232. goto out;
  233. list_move(&em->list, &tree->modified_extents);
  234. em->generation = gen;
  235. clear_bit(EXTENT_FLAG_PINNED, &em->flags);
  236. em->mod_start = em->start;
  237. em->mod_len = em->len;
  238. if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
  239. prealloc = true;
  240. clear_bit(EXTENT_FLAG_PREALLOC, &em->flags);
  241. }
  242. try_merge_map(tree, em);
  243. if (prealloc) {
  244. em->mod_start = em->start;
  245. em->mod_len = em->len;
  246. }
  247. free_extent_map(em);
  248. out:
  249. write_unlock(&tree->lock);
  250. return ret;
  251. }
  252. /**
  253. * add_extent_mapping - add new extent map to the extent tree
  254. * @tree: tree to insert new map in
  255. * @em: map to insert
  256. *
  257. * Insert @em into @tree or perform a simple forward/backward merge with
  258. * existing mappings. The extent_map struct passed in will be inserted
  259. * into the tree directly, with an additional reference taken, or a
  260. * reference dropped if the merge attempt was successful.
  261. */
  262. int add_extent_mapping(struct extent_map_tree *tree,
  263. struct extent_map *em)
  264. {
  265. int ret = 0;
  266. struct rb_node *rb;
  267. struct extent_map *exist;
  268. exist = lookup_extent_mapping(tree, em->start, em->len);
  269. if (exist) {
  270. free_extent_map(exist);
  271. ret = -EEXIST;
  272. goto out;
  273. }
  274. rb = tree_insert(&tree->map, em->start, &em->rb_node);
  275. if (rb) {
  276. ret = -EEXIST;
  277. goto out;
  278. }
  279. atomic_inc(&em->refs);
  280. em->mod_start = em->start;
  281. em->mod_len = em->len;
  282. try_merge_map(tree, em);
  283. out:
  284. return ret;
  285. }
  286. /* simple helper to do math around the end of an extent, handling wrap */
  287. static u64 range_end(u64 start, u64 len)
  288. {
  289. if (start + len < start)
  290. return (u64)-1;
  291. return start + len;
  292. }
  293. struct extent_map *__lookup_extent_mapping(struct extent_map_tree *tree,
  294. u64 start, u64 len, int strict)
  295. {
  296. struct extent_map *em;
  297. struct rb_node *rb_node;
  298. struct rb_node *prev = NULL;
  299. struct rb_node *next = NULL;
  300. u64 end = range_end(start, len);
  301. rb_node = __tree_search(&tree->map, start, &prev, &next);
  302. if (!rb_node) {
  303. if (prev)
  304. rb_node = prev;
  305. else if (next)
  306. rb_node = next;
  307. else
  308. return NULL;
  309. }
  310. em = rb_entry(rb_node, struct extent_map, rb_node);
  311. if (strict && !(end > em->start && start < extent_map_end(em)))
  312. return NULL;
  313. atomic_inc(&em->refs);
  314. return em;
  315. }
  316. /**
  317. * lookup_extent_mapping - lookup extent_map
  318. * @tree: tree to lookup in
  319. * @start: byte offset to start the search
  320. * @len: length of the lookup range
  321. *
  322. * Find and return the first extent_map struct in @tree that intersects the
  323. * [start, len] range. There may be additional objects in the tree that
  324. * intersect, so check the object returned carefully to make sure that no
  325. * additional lookups are needed.
  326. */
  327. struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
  328. u64 start, u64 len)
  329. {
  330. return __lookup_extent_mapping(tree, start, len, 1);
  331. }
  332. /**
  333. * search_extent_mapping - find a nearby extent map
  334. * @tree: tree to lookup in
  335. * @start: byte offset to start the search
  336. * @len: length of the lookup range
  337. *
  338. * Find and return the first extent_map struct in @tree that intersects the
  339. * [start, len] range.
  340. *
  341. * If one can't be found, any nearby extent may be returned
  342. */
  343. struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
  344. u64 start, u64 len)
  345. {
  346. return __lookup_extent_mapping(tree, start, len, 0);
  347. }
  348. /**
  349. * remove_extent_mapping - removes an extent_map from the extent tree
  350. * @tree: extent tree to remove from
  351. * @em: extent map beeing removed
  352. *
  353. * Removes @em from @tree. No reference counts are dropped, and no checks
  354. * are done to see if the range is in use
  355. */
  356. int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
  357. {
  358. int ret = 0;
  359. WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
  360. rb_erase(&em->rb_node, &tree->map);
  361. if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
  362. list_del_init(&em->list);
  363. em->in_tree = 0;
  364. return ret;
  365. }