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 8 9static struct kmem_cache *extent_map_cache; 10 11int __init extent_map_init(void) 12{ 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; 19} 20 21void extent_map_exit(void) 22{ 23 if (extent_map_cache) 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 beeing releasead 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->bdev); 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 beeing 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} 456