fs/btrfs/extent_map.c

   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
   9 static struct kmem_cache *extent_map_cache;
  10
  11 int __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
  21 void 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  */
  34 void 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  */
  48 struct 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         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;
  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  */
  70 void 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(em->in_tree);
  77                 WARN_ON(!list_empty(&em->list));
  78                 kmem_cache_free(extent_map_cache, em);
  79         }
  80 }
  81
  82 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
  83                                    struct rb_node *node)
  84 {
  85         struct rb_node **p = &root->rb_node;
  86         struct rb_node *parent = NULL;
  87         struct extent_map *entry;
  88
  89         while (*p) {
  90                 parent = *p;
  91                 entry = rb_entry(parent, struct extent_map, rb_node);
  92
  93                 WARN_ON(!entry->in_tree);
  94
  95                 if (offset < entry->start)
  96                         p = &(*p)->rb_left;
  97                 else if (offset >= extent_map_end(entry))
  98                         p = &(*p)->rb_right;
  99                 else
 100                         return parent;
 101         }
 102
 103         entry = rb_entry(node, struct extent_map, rb_node);
 104         entry->in_tree = 1;
 105         rb_link_node(node, parent, p);
 106         rb_insert_color(node, root);
 107         return NULL;
 108 }
 109
 110 /*
 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
 113  */
 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)
 117 {
 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;
 123
 124         while (n) {
 125                 entry = rb_entry(n, struct extent_map, rb_node);
 126                 prev = n;
 127                 prev_entry = entry;
 128
 129                 WARN_ON(!entry->in_tree);
 130
 131                 if (offset < entry->start)
 132                         n = n->rb_left;
 133                 else if (offset >= extent_map_end(entry))
 134                         n = n->rb_right;
 135                 else
 136                         return n;
 137         }
 138
 139         if (prev_ret) {
 140                 orig_prev = prev;
 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);
 144                 }
 145                 *prev_ret = prev;
 146                 prev = orig_prev;
 147         }
 148
 149         if (next_ret) {
 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);
 154                 }
 155                 *next_ret = prev;
 156         }
 157         return NULL;
 158 }
 159
 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)
 162 {
 163         if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
 164                 return 0;
 165
 166         /*
 167          * don't merge compressed extents, we need to know their
 168          * actual size
 169          */
 170         if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
 171                 return 0;
 172
 173         if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
 174             test_bit(EXTENT_FLAG_LOGGING, &next->flags))
 175                 return 0;
 176
 177         if (extent_map_end(prev) == next->start &&
 178             prev->flags == next->flags &&
 179             prev->bdev == next->bdev &&
 180             ((next->block_start == EXTENT_MAP_HOLE &&
 181               prev->block_start == EXTENT_MAP_HOLE) ||
 182              (next->block_start == EXTENT_MAP_INLINE &&
 183               prev->block_start == EXTENT_MAP_INLINE) ||
 184              (next->block_start == EXTENT_MAP_DELALLOC &&
 185               prev->block_start == EXTENT_MAP_DELALLOC) ||
 186              (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
 187               next->block_start == extent_map_block_end(prev)))) {
 188                 return 1;
 189         }
 190         return 0;
 191 }
 192
 193 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
 194 {
 195         struct extent_map *merge = NULL;
 196         struct rb_node *rb;
 197
 198         if (em->start != 0) {
 199                 rb = rb_prev(&em->rb_node);
 200                 if (rb)
 201                         merge = rb_entry(rb, struct extent_map, rb_node);
 202                 if (rb && mergable_maps(merge, em)) {
 203                         em->start = merge->start;
 204                         em->orig_start = merge->orig_start;
 205                         em->len += merge->len;
 206                         em->block_len += merge->block_len;
 207                         em->block_start = merge->block_start;
 208                         merge->in_tree = 0;
 209                         em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
 210                         em->mod_start = merge->mod_start;
 211                         em->generation = max(em->generation, merge->generation);
 212                         list_move(&em->list, &tree->modified_extents);
 213
 214                         list_del_init(&merge->list);
 215                         rb_erase(&merge->rb_node, &tree->map);
 216                         free_extent_map(merge);
 217                 }
 218         }
 219
 220         rb = rb_next(&em->rb_node);
 221         if (rb)
 222                 merge = rb_entry(rb, struct extent_map, rb_node);
 223         if (rb && mergable_maps(em, merge)) {
 224                 em->len += merge->len;
 225                 em->block_len += merge->len;
 226                 rb_erase(&merge->rb_node, &tree->map);
 227                 merge->in_tree = 0;
 228                 em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
 229                 em->generation = max(em->generation, merge->generation);
 230                 list_del_init(&merge->list);
 231                 free_extent_map(merge);
 232         }
 233 }
 234
 235 /**
 236  * unpin_extent_cache - unpin an extent from the cache
 237  * @tree:       tree to unpin the extent in
 238  * @start:      logical offset in the file
 239  * @len:        length of the extent
 240  * @gen:        generation that this extent has been modified in
 241  *
 242  * Called after an extent has been written to disk properly.  Set the generation
 243  * to the generation that actually added the file item to the inode so we know
 244  * we need to sync this extent when we call fsync().
 245  */
 246 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
 247                        u64 gen)
 248 {
 249         int ret = 0;
 250         struct extent_map *em;
 251         bool prealloc = false;
 252
 253         write_lock(&tree->lock);
 254         em = lookup_extent_mapping(tree, start, len);
 255
 256         WARN_ON(!em || em->start != start);
 257
 258         if (!em)
 259                 goto out;
 260
 261         if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
 262                 list_move(&em->list, &tree->modified_extents);
 263         em->generation = gen;
 264         clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 265         em->mod_start = em->start;
 266         em->mod_len = em->len;
 267
 268         if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
 269                 prealloc = true;
 270                 clear_bit(EXTENT_FLAG_FILLING, &em->flags);
 271         }
 272
 273         try_merge_map(tree, em);
 274
 275         if (prealloc) {
 276                 em->mod_start = em->start;
 277                 em->mod_len = em->len;
 278         }
 279
 280         free_extent_map(em);
 281 out:
 282         write_unlock(&tree->lock);
 283         return ret;
 284
 285 }
 286
 287 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
 288 {
 289         clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
 290         if (em->in_tree)
 291                 try_merge_map(tree, em);
 292 }
 293
 294 /**
 295  * add_extent_mapping - add new extent map to the extent tree
 296  * @tree:       tree to insert new map in
 297  * @em:         map to insert
 298  *
 299  * Insert @em into @tree or perform a simple forward/backward merge with
 300  * existing mappings.  The extent_map struct passed in will be inserted
 301  * into the tree directly, with an additional reference taken, or a
 302  * reference dropped if the merge attempt was successful.
 303  */
 304 int add_extent_mapping(struct extent_map_tree *tree,
 305                        struct extent_map *em)
 306 {
 307         int ret = 0;
 308         struct rb_node *rb;
 309         struct extent_map *exist;
 310
 311         exist = lookup_extent_mapping(tree, em->start, em->len);
 312         if (exist) {
 313                 free_extent_map(exist);
 314                 ret = -EEXIST;
 315                 goto out;
 316         }
 317         rb = tree_insert(&tree->map, em->start, &em->rb_node);
 318         if (rb) {
 319                 ret = -EEXIST;
 320                 goto out;
 321         }
 322         atomic_inc(&em->refs);
 323
 324         em->mod_start = em->start;
 325         em->mod_len = em->len;
 326
 327         try_merge_map(tree, em);
 328 out:
 329         return ret;
 330 }
 331
 332 /* simple helper to do math around the end of an extent, handling wrap */
 333 static u64 range_end(u64 start, u64 len)
 334 {
 335         if (start + len < start)
 336                 return (u64)-1;
 337         return start + len;
 338 }
 339
 340 struct extent_map *__lookup_extent_mapping(struct extent_map_tree *tree,
 341                                            u64 start, u64 len, int strict)
 342 {
 343         struct extent_map *em;
 344         struct rb_node *rb_node;
 345         struct rb_node *prev = NULL;
 346         struct rb_node *next = NULL;
 347         u64 end = range_end(start, len);
 348
 349         rb_node = __tree_search(&tree->map, start, &prev, &next);
 350         if (!rb_node) {
 351                 if (prev)
 352                         rb_node = prev;
 353                 else if (next)
 354                         rb_node = next;
 355                 else
 356                         return NULL;
 357         }
 358
 359         em = rb_entry(rb_node, struct extent_map, rb_node);
 360
 361         if (strict && !(end > em->start && start < extent_map_end(em)))
 362                 return NULL;
 363
 364         atomic_inc(&em->refs);
 365         return em;
 366 }
 367
 368 /**
 369  * lookup_extent_mapping - lookup extent_map
 370  * @tree:       tree to lookup in
 371  * @start:      byte offset to start the search
 372  * @len:        length of the lookup range
 373  *
 374  * Find and return the first extent_map struct in @tree that intersects the
 375  * [start, len] range.  There may be additional objects in the tree that
 376  * intersect, so check the object returned carefully to make sure that no
 377  * additional lookups are needed.
 378  */
 379 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
 380                                          u64 start, u64 len)
 381 {
 382         return __lookup_extent_mapping(tree, start, len, 1);
 383 }
 384
 385 /**
 386  * search_extent_mapping - find a nearby extent map
 387  * @tree:       tree to lookup in
 388  * @start:      byte offset to start the search
 389  * @len:        length of the lookup range
 390  *
 391  * Find and return the first extent_map struct in @tree that intersects the
 392  * [start, len] range.
 393  *
 394  * If one can't be found, any nearby extent may be returned
 395  */
 396 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
 397                                          u64 start, u64 len)
 398 {
 399         return __lookup_extent_mapping(tree, start, len, 0);
 400 }
 401
 402 /**
 403  * remove_extent_mapping - removes an extent_map from the extent tree
 404  * @tree:       extent tree to remove from
 405  * @em:         extent map beeing removed
 406  *
 407  * Removes @em from @tree.  No reference counts are dropped, and no checks
 408  * are done to see if the range is in use
 409  */
 410 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
 411 {
 412         int ret = 0;
 413
 414         WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
 415         rb_erase(&em->rb_node, &tree->map);
 416         if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
 417                 list_del_init(&em->list);
 418         em->in_tree = 0;
 419         return ret;
 420 }